2013-04-30 13:44:33 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
|
|
|
rbd.c -- Export ceph rados objects as a Linux block device
|
|
|
|
|
|
|
|
|
|
|
|
based on drivers/block/osdblk.c:
|
|
|
|
|
|
|
|
Copyright 2009 Red Hat, Inc.
|
|
|
|
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
the Free Software Foundation.
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with this program; see the file COPYING. If not, write to
|
|
|
|
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
For usage instructions, please refer to:
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
Documentation/ABI/testing/sysfs-bus-rbd
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/ceph/libceph.h>
|
|
|
|
#include <linux/ceph/osd_client.h>
|
|
|
|
#include <linux/ceph/mon_client.h>
|
2016-08-12 22:40:02 +08:00
|
|
|
#include <linux/ceph/cls_lock_client.h>
|
2018-02-02 22:23:22 +08:00
|
|
|
#include <linux/ceph/striper.h>
|
2010-08-13 07:11:25 +08:00
|
|
|
#include <linux/ceph/decode.h>
|
2011-03-22 06:10:11 +08:00
|
|
|
#include <linux/parser.h>
|
2013-05-02 01:43:03 +08:00
|
|
|
#include <linux/bsearch.h>
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/device.h>
|
|
|
|
#include <linux/module.h>
|
2015-01-14 00:20:04 +08:00
|
|
|
#include <linux/blk-mq.h>
|
2010-08-13 07:11:25 +08:00
|
|
|
#include <linux/fs.h>
|
|
|
|
#include <linux/blkdev.h>
|
2013-05-02 01:43:03 +08:00
|
|
|
#include <linux/slab.h>
|
2013-12-13 21:28:57 +08:00
|
|
|
#include <linux/idr.h>
|
2014-08-04 22:04:39 +08:00
|
|
|
#include <linux/workqueue.h>
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
#include "rbd_types.h"
|
|
|
|
|
2012-09-07 05:00:54 +08:00
|
|
|
#define RBD_DEBUG /* Activate rbd_assert() calls */
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
/*
|
|
|
|
* Increment the given counter and return its updated value.
|
|
|
|
* If the counter is already 0 it will not be incremented.
|
|
|
|
* If the counter is already at its maximum value returns
|
|
|
|
* -EINVAL without updating it.
|
|
|
|
*/
|
|
|
|
static int atomic_inc_return_safe(atomic_t *v)
|
|
|
|
{
|
|
|
|
unsigned int counter;
|
|
|
|
|
atomics/treewide: Rename __atomic_add_unless() => atomic_fetch_add_unless()
While __atomic_add_unless() was originally intended as a building-block
for atomic_add_unless(), it's now used in a number of places around the
kernel. It's the only common atomic operation named __atomic*(), rather
than atomic_*(), and for consistency it would be better named
atomic_fetch_add_unless().
This lack of consistency is slightly confusing, and gets in the way of
scripting atomics. Given that, let's clean things up and promote it to
an official part of the atomics API, in the form of
atomic_fetch_add_unless().
This patch converts definitions and invocations over to the new name,
including the instrumented version, using the following script:
----
git grep -w __atomic_add_unless | while read line; do
sed -i '{s/\<__atomic_add_unless\>/atomic_fetch_add_unless/}' "${line%%:*}";
done
git grep -w __arch_atomic_add_unless | while read line; do
sed -i '{s/\<__arch_atomic_add_unless\>/arch_atomic_fetch_add_unless/}' "${line%%:*}";
done
----
Note that we do not have atomic{64,_long}_fetch_add_unless(), which will
be introduced by later patches.
There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Palmer Dabbelt <palmer@sifive.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/lkml/20180621121321.4761-2-mark.rutland@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
2018-06-21 20:13:04 +08:00
|
|
|
counter = (unsigned int)atomic_fetch_add_unless(v, 1, 0);
|
2013-05-09 11:50:04 +08:00
|
|
|
if (counter <= (unsigned int)INT_MAX)
|
|
|
|
return (int)counter;
|
|
|
|
|
|
|
|
atomic_dec(v);
|
|
|
|
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Decrement the counter. Return the resulting value, or -EINVAL */
|
|
|
|
static int atomic_dec_return_safe(atomic_t *v)
|
|
|
|
{
|
|
|
|
int counter;
|
|
|
|
|
|
|
|
counter = atomic_dec_return(v);
|
|
|
|
if (counter >= 0)
|
|
|
|
return counter;
|
|
|
|
|
|
|
|
atomic_inc(v);
|
|
|
|
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2012-01-30 03:57:44 +08:00
|
|
|
#define RBD_DRV_NAME "rbd"
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-12-17 01:26:32 +08:00
|
|
|
#define RBD_MINORS_PER_MAJOR 256
|
|
|
|
#define RBD_SINGLE_MAJOR_PART_SHIFT 4
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
#define RBD_MAX_PARENT_CHAIN_LEN 16
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
#define RBD_SNAP_DEV_NAME_PREFIX "snap_"
|
|
|
|
#define RBD_MAX_SNAP_NAME_LEN \
|
|
|
|
(NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1))
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
#define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
#define RBD_SNAP_HEAD_NAME "-"
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
#define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
/* This allows a single page to hold an image name sent by OSD */
|
|
|
|
#define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1)
|
2012-07-04 05:01:19 +08:00
|
|
|
#define RBD_IMAGE_ID_LEN_MAX 64
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
#define RBD_OBJ_PREFIX_LEN_MAX 64
|
2012-07-11 09:30:11 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
#define RBD_NOTIFY_TIMEOUT 5 /* seconds */
|
2016-08-12 22:11:41 +08:00
|
|
|
#define RBD_RETRY_DELAY msecs_to_jiffies(1000)
|
|
|
|
|
2012-10-10 04:50:17 +08:00
|
|
|
/* Feature bits */
|
|
|
|
|
2017-03-03 02:56:57 +08:00
|
|
|
#define RBD_FEATURE_LAYERING (1ULL<<0)
|
|
|
|
#define RBD_FEATURE_STRIPINGV2 (1ULL<<1)
|
|
|
|
#define RBD_FEATURE_EXCLUSIVE_LOCK (1ULL<<2)
|
2019-02-26 01:55:38 +08:00
|
|
|
#define RBD_FEATURE_DEEP_FLATTEN (1ULL<<5)
|
2017-03-03 02:56:57 +08:00
|
|
|
#define RBD_FEATURE_DATA_POOL (1ULL<<7)
|
2018-01-16 22:41:54 +08:00
|
|
|
#define RBD_FEATURE_OPERATIONS (1ULL<<8)
|
2017-03-03 02:56:57 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
#define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \
|
|
|
|
RBD_FEATURE_STRIPINGV2 | \
|
2017-01-26 01:16:22 +08:00
|
|
|
RBD_FEATURE_EXCLUSIVE_LOCK | \
|
2019-02-26 01:55:38 +08:00
|
|
|
RBD_FEATURE_DEEP_FLATTEN | \
|
2018-01-16 22:41:54 +08:00
|
|
|
RBD_FEATURE_DATA_POOL | \
|
|
|
|
RBD_FEATURE_OPERATIONS)
|
2012-10-10 04:50:17 +08:00
|
|
|
|
|
|
|
/* Features supported by this (client software) implementation. */
|
|
|
|
|
2012-10-26 12:34:40 +08:00
|
|
|
#define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL)
|
2012-10-10 04:50:17 +08:00
|
|
|
|
2012-02-02 22:13:30 +08:00
|
|
|
/*
|
|
|
|
* An RBD device name will be "rbd#", where the "rbd" comes from
|
|
|
|
* RBD_DRV_NAME above, and # is a unique integer identifier.
|
|
|
|
*/
|
2010-08-13 07:11:25 +08:00
|
|
|
#define DEV_NAME_LEN 32
|
|
|
|
|
|
|
|
/*
|
|
|
|
* block device image metadata (in-memory version)
|
|
|
|
*/
|
|
|
|
struct rbd_image_header {
|
2013-05-06 22:51:29 +08:00
|
|
|
/* These six fields never change for a given rbd image */
|
2012-07-10 10:04:24 +08:00
|
|
|
char *object_prefix;
|
2010-08-13 07:11:25 +08:00
|
|
|
__u8 obj_order;
|
2013-05-06 22:51:29 +08:00
|
|
|
u64 stripe_unit;
|
|
|
|
u64 stripe_count;
|
2017-01-26 01:16:22 +08:00
|
|
|
s64 data_pool_id;
|
2013-05-06 22:51:29 +08:00
|
|
|
u64 features; /* Might be changeable someday? */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-09-01 06:29:51 +08:00
|
|
|
/* The remaining fields need to be updated occasionally */
|
|
|
|
u64 image_size;
|
|
|
|
struct ceph_snap_context *snapc;
|
2013-05-06 22:51:29 +08:00
|
|
|
char *snap_names; /* format 1 only */
|
|
|
|
u64 *snap_sizes; /* format 1 only */
|
2011-03-22 06:10:11 +08:00
|
|
|
};
|
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
/*
|
|
|
|
* An rbd image specification.
|
|
|
|
*
|
|
|
|
* The tuple (pool_id, image_id, snap_id) is sufficient to uniquely
|
2012-11-01 21:39:26 +08:00
|
|
|
* identify an image. Each rbd_dev structure includes a pointer to
|
|
|
|
* an rbd_spec structure that encapsulates this identity.
|
|
|
|
*
|
|
|
|
* Each of the id's in an rbd_spec has an associated name. For a
|
|
|
|
* user-mapped image, the names are supplied and the id's associated
|
|
|
|
* with them are looked up. For a layered image, a parent image is
|
|
|
|
* defined by the tuple, and the names are looked up.
|
|
|
|
*
|
|
|
|
* An rbd_dev structure contains a parent_spec pointer which is
|
|
|
|
* non-null if the image it represents is a child in a layered
|
|
|
|
* image. This pointer will refer to the rbd_spec structure used
|
|
|
|
* by the parent rbd_dev for its own identity (i.e., the structure
|
|
|
|
* is shared between the parent and child).
|
|
|
|
*
|
|
|
|
* Since these structures are populated once, during the discovery
|
|
|
|
* phase of image construction, they are effectively immutable so
|
|
|
|
* we make no effort to synchronize access to them.
|
|
|
|
*
|
|
|
|
* Note that code herein does not assume the image name is known (it
|
|
|
|
* could be a null pointer).
|
2012-10-26 12:34:41 +08:00
|
|
|
*/
|
|
|
|
struct rbd_spec {
|
|
|
|
u64 pool_id;
|
2013-04-26 22:43:47 +08:00
|
|
|
const char *pool_name;
|
2018-07-03 21:28:43 +08:00
|
|
|
const char *pool_ns; /* NULL if default, never "" */
|
2012-10-26 12:34:41 +08:00
|
|
|
|
2013-04-26 22:43:47 +08:00
|
|
|
const char *image_id;
|
|
|
|
const char *image_name;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
|
|
|
u64 snap_id;
|
2013-04-26 22:43:47 +08:00
|
|
|
const char *snap_name;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
|
|
|
struct kref kref;
|
|
|
|
};
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
2012-01-30 03:57:44 +08:00
|
|
|
* an instance of the client. multiple devices may share an rbd client.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
|
|
|
struct rbd_client {
|
|
|
|
struct ceph_client *client;
|
|
|
|
struct kref kref;
|
|
|
|
struct list_head node;
|
|
|
|
};
|
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
struct pending_result {
|
|
|
|
int result; /* first nonzero result */
|
|
|
|
int num_pending;
|
|
|
|
};
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
struct rbd_img_request;
|
|
|
|
|
2013-01-19 02:31:10 +08:00
|
|
|
enum obj_request_type {
|
2018-01-16 19:15:02 +08:00
|
|
|
OBJ_REQUEST_NODATA = 1,
|
2018-01-20 17:30:10 +08:00
|
|
|
OBJ_REQUEST_BIO, /* pointer into provided bio (list) */
|
2018-01-20 17:30:11 +08:00
|
|
|
OBJ_REQUEST_BVECS, /* pointer into provided bio_vec array */
|
2018-02-07 02:26:35 +08:00
|
|
|
OBJ_REQUEST_OWN_BVECS, /* private bio_vec array, doesn't own pages */
|
2013-01-19 02:31:10 +08:00
|
|
|
};
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2014-03-13 11:21:35 +08:00
|
|
|
enum obj_operation_type {
|
2018-01-16 19:15:02 +08:00
|
|
|
OBJ_OP_READ = 1,
|
2014-03-13 11:21:35 +08:00
|
|
|
OBJ_OP_WRITE,
|
2014-04-01 22:22:16 +08:00
|
|
|
OBJ_OP_DISCARD,
|
2019-01-29 19:46:25 +08:00
|
|
|
OBJ_OP_ZEROOUT,
|
2014-03-13 11:21:35 +08:00
|
|
|
};
|
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
#define RBD_OBJ_FLAG_DELETION (1U << 0)
|
|
|
|
#define RBD_OBJ_FLAG_COPYUP_ENABLED (1U << 1)
|
|
|
|
|
2019-05-08 19:35:57 +08:00
|
|
|
enum rbd_obj_read_state {
|
2019-05-15 03:06:07 +08:00
|
|
|
RBD_OBJ_READ_START = 1,
|
|
|
|
RBD_OBJ_READ_OBJECT,
|
2019-05-08 19:35:57 +08:00
|
|
|
RBD_OBJ_READ_PARENT,
|
|
|
|
};
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
|
|
|
* Writes go through the following state machine to deal with
|
|
|
|
* layering:
|
|
|
|
*
|
2019-02-28 21:20:28 +08:00
|
|
|
* . . . . . RBD_OBJ_WRITE_GUARD. . . . . . . . . . . . . .
|
|
|
|
* . | .
|
|
|
|
* . v .
|
|
|
|
* . RBD_OBJ_WRITE_READ_FROM_PARENT. . . .
|
|
|
|
* . | . .
|
|
|
|
* . v v (deep-copyup .
|
|
|
|
* (image . RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC . not needed) .
|
|
|
|
* flattened) v | . .
|
|
|
|
* . v . .
|
|
|
|
* . . . .RBD_OBJ_WRITE_COPYUP_OPS. . . . . (copyup .
|
|
|
|
* | not needed) v
|
|
|
|
* v .
|
|
|
|
* done . . . . . . . . . . . . . . . . . .
|
|
|
|
* ^
|
|
|
|
* |
|
|
|
|
* RBD_OBJ_WRITE_FLAT
|
2018-01-29 21:04:08 +08:00
|
|
|
*
|
|
|
|
* Writes start in RBD_OBJ_WRITE_GUARD or _FLAT, depending on whether
|
2019-02-28 21:20:28 +08:00
|
|
|
* assert_exists guard is needed or not (in some cases it's not needed
|
|
|
|
* even if there is a parent).
|
2018-01-29 21:04:08 +08:00
|
|
|
*/
|
|
|
|
enum rbd_obj_write_state {
|
2019-05-15 03:06:07 +08:00
|
|
|
RBD_OBJ_WRITE_START = 1,
|
|
|
|
RBD_OBJ_WRITE_OBJECT,
|
2019-02-28 17:49:12 +08:00
|
|
|
RBD_OBJ_WRITE_READ_FROM_PARENT,
|
2019-02-28 21:20:28 +08:00
|
|
|
RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC,
|
2019-02-28 17:49:12 +08:00
|
|
|
RBD_OBJ_WRITE_COPYUP_OPS,
|
2013-02-12 02:33:24 +08:00
|
|
|
};
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
struct rbd_obj_request {
|
2018-02-02 22:23:22 +08:00
|
|
|
struct ceph_object_extent ex;
|
2019-05-15 02:45:38 +08:00
|
|
|
unsigned int flags; /* RBD_OBJ_FLAG_* */
|
2013-02-12 02:33:24 +08:00
|
|
|
union {
|
2019-05-08 19:35:57 +08:00
|
|
|
enum rbd_obj_read_state read_state; /* for reads */
|
2018-01-29 21:04:08 +08:00
|
|
|
enum rbd_obj_write_state write_state; /* for writes */
|
2013-02-12 02:33:24 +08:00
|
|
|
};
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_img_request *img_request;
|
2018-02-07 02:26:33 +08:00
|
|
|
struct ceph_file_extent *img_extents;
|
|
|
|
u32 num_img_extents;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2013-01-18 02:25:27 +08:00
|
|
|
union {
|
2018-01-20 17:30:10 +08:00
|
|
|
struct ceph_bio_iter bio_pos;
|
2013-01-18 02:25:27 +08:00
|
|
|
struct {
|
2018-01-20 17:30:11 +08:00
|
|
|
struct ceph_bvec_iter bvec_pos;
|
|
|
|
u32 bvec_count;
|
2018-02-07 02:26:35 +08:00
|
|
|
u32 bvec_idx;
|
2013-01-18 02:25:27 +08:00
|
|
|
};
|
|
|
|
};
|
2018-01-20 17:30:11 +08:00
|
|
|
struct bio_vec *copyup_bvecs;
|
|
|
|
u32 copyup_bvec_count;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
struct list_head osd_reqs; /* w/ r_private_item */
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
struct mutex state_mutex;
|
2012-11-22 14:00:08 +08:00
|
|
|
struct kref kref;
|
|
|
|
};
|
|
|
|
|
2013-02-08 23:55:49 +08:00
|
|
|
enum img_req_flags {
|
2013-01-25 06:13:36 +08:00
|
|
|
IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */
|
2013-01-25 06:13:36 +08:00
|
|
|
IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */
|
2013-02-08 23:55:49 +08:00
|
|
|
};
|
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
enum rbd_img_state {
|
|
|
|
RBD_IMG_START = 1,
|
|
|
|
__RBD_IMG_OBJECT_REQUESTS,
|
|
|
|
RBD_IMG_OBJECT_REQUESTS,
|
|
|
|
};
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
struct rbd_img_request {
|
|
|
|
struct rbd_device *rbd_dev;
|
2018-01-31 00:52:10 +08:00
|
|
|
enum obj_operation_type op_type;
|
2018-02-01 18:50:47 +08:00
|
|
|
enum obj_request_type data_type;
|
2013-02-08 23:55:49 +08:00
|
|
|
unsigned long flags;
|
2019-05-16 21:06:56 +08:00
|
|
|
enum rbd_img_state state;
|
2012-11-22 14:00:08 +08:00
|
|
|
union {
|
2013-01-25 06:13:36 +08:00
|
|
|
u64 snap_id; /* for reads */
|
2012-11-22 14:00:08 +08:00
|
|
|
struct ceph_snap_context *snapc; /* for writes */
|
2013-01-25 06:13:36 +08:00
|
|
|
};
|
|
|
|
union {
|
|
|
|
struct request *rq; /* block request */
|
|
|
|
struct rbd_obj_request *obj_request; /* obj req initiator */
|
2012-11-22 14:00:08 +08:00
|
|
|
};
|
|
|
|
|
2018-02-02 22:23:22 +08:00
|
|
|
struct list_head object_extents; /* obj_req.ex structs */
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
struct mutex state_mutex;
|
|
|
|
struct pending_result pending;
|
|
|
|
struct work_struct work;
|
|
|
|
int work_result;
|
2012-11-22 14:00:08 +08:00
|
|
|
struct kref kref;
|
|
|
|
};
|
|
|
|
|
|
|
|
#define for_each_obj_request(ireq, oreq) \
|
2018-02-02 22:23:22 +08:00
|
|
|
list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item)
|
2012-11-22 14:00:08 +08:00
|
|
|
#define for_each_obj_request_safe(ireq, oreq, n) \
|
2018-02-02 22:23:22 +08:00
|
|
|
list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item)
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
enum rbd_watch_state {
|
|
|
|
RBD_WATCH_STATE_UNREGISTERED,
|
|
|
|
RBD_WATCH_STATE_REGISTERED,
|
|
|
|
RBD_WATCH_STATE_ERROR,
|
|
|
|
};
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
enum rbd_lock_state {
|
|
|
|
RBD_LOCK_STATE_UNLOCKED,
|
|
|
|
RBD_LOCK_STATE_LOCKED,
|
|
|
|
RBD_LOCK_STATE_RELEASING,
|
|
|
|
};
|
|
|
|
|
|
|
|
/* WatchNotify::ClientId */
|
|
|
|
struct rbd_client_id {
|
|
|
|
u64 gid;
|
|
|
|
u64 handle;
|
|
|
|
};
|
|
|
|
|
2012-09-01 06:29:51 +08:00
|
|
|
struct rbd_mapping {
|
2012-08-31 03:42:15 +08:00
|
|
|
u64 size;
|
2012-07-14 09:35:12 +08:00
|
|
|
u64 features;
|
2012-09-01 06:29:51 +08:00
|
|
|
};
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
|
|
|
* a single device
|
|
|
|
*/
|
|
|
|
struct rbd_device {
|
2012-07-04 05:01:19 +08:00
|
|
|
int dev_id; /* blkdev unique id */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
int major; /* blkdev assigned major */
|
2013-12-13 21:28:57 +08:00
|
|
|
int minor;
|
2010-08-13 07:11:25 +08:00
|
|
|
struct gendisk *disk; /* blkdev's gendisk and rq */
|
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
u32 image_format; /* Either 1 or 2 */
|
2010-08-13 07:11:25 +08:00
|
|
|
struct rbd_client *rbd_client;
|
|
|
|
|
|
|
|
char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
|
|
|
|
|
2013-01-15 02:43:31 +08:00
|
|
|
spinlock_t lock; /* queue, flags, open_count */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
struct rbd_image_header header;
|
2013-01-15 02:43:31 +08:00
|
|
|
unsigned long flags; /* possibly lock protected */
|
2012-10-26 12:34:41 +08:00
|
|
|
struct rbd_spec *spec;
|
2015-06-22 18:24:48 +08:00
|
|
|
struct rbd_options *opts;
|
2016-08-19 00:38:45 +08:00
|
|
|
char *config_info; /* add{,_single_major} string */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2016-04-30 02:01:25 +08:00
|
|
|
struct ceph_object_id header_oid;
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_object_locator header_oloc;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
struct ceph_file_layout layout; /* used for all rbd requests */
|
2012-11-15 02:25:19 +08:00
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
struct mutex watch_mutex;
|
|
|
|
enum rbd_watch_state watch_state;
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_osd_linger_request *watch_handle;
|
2016-08-12 22:11:41 +08:00
|
|
|
u64 watch_cookie;
|
|
|
|
struct delayed_work watch_dwork;
|
2011-03-22 06:10:11 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
struct rw_semaphore lock_rwsem;
|
|
|
|
enum rbd_lock_state lock_state;
|
2017-04-13 18:17:38 +08:00
|
|
|
char lock_cookie[32];
|
2016-08-12 22:40:02 +08:00
|
|
|
struct rbd_client_id owner_cid;
|
|
|
|
struct work_struct acquired_lock_work;
|
|
|
|
struct work_struct released_lock_work;
|
|
|
|
struct delayed_work lock_dwork;
|
|
|
|
struct work_struct unlock_work;
|
|
|
|
wait_queue_head_t lock_waitq;
|
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
struct workqueue_struct *task_wq;
|
2011-03-22 06:10:11 +08:00
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
struct rbd_spec *parent_spec;
|
|
|
|
u64 parent_overlap;
|
2013-05-09 11:50:04 +08:00
|
|
|
atomic_t parent_ref;
|
2012-10-31 08:40:33 +08:00
|
|
|
struct rbd_device *parent;
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
/* Block layer tags. */
|
|
|
|
struct blk_mq_tag_set tag_set;
|
|
|
|
|
2011-11-22 09:11:12 +08:00
|
|
|
/* protects updating the header */
|
|
|
|
struct rw_semaphore header_rwsem;
|
2012-09-01 06:29:51 +08:00
|
|
|
|
|
|
|
struct rbd_mapping mapping;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
struct list_head node;
|
2010-11-20 06:51:04 +08:00
|
|
|
|
|
|
|
/* sysfs related */
|
|
|
|
struct device dev;
|
2013-01-15 02:43:31 +08:00
|
|
|
unsigned long open_count; /* protected by lock */
|
2010-11-20 06:51:04 +08:00
|
|
|
};
|
|
|
|
|
2013-01-15 02:43:31 +08:00
|
|
|
/*
|
2016-09-29 19:41:05 +08:00
|
|
|
* Flag bits for rbd_dev->flags:
|
|
|
|
* - REMOVING (which is coupled with rbd_dev->open_count) is protected
|
|
|
|
* by rbd_dev->lock
|
|
|
|
* - BLACKLISTED is protected by rbd_dev->lock_rwsem
|
2013-01-15 02:43:31 +08:00
|
|
|
*/
|
2013-01-15 02:43:31 +08:00
|
|
|
enum rbd_dev_flags {
|
|
|
|
RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */
|
2013-01-15 02:43:31 +08:00
|
|
|
RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */
|
2016-09-29 19:41:05 +08:00
|
|
|
RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */
|
2013-01-15 02:43:31 +08:00
|
|
|
};
|
|
|
|
|
2013-06-01 06:40:45 +08:00
|
|
|
static DEFINE_MUTEX(client_mutex); /* Serialize client creation */
|
2012-01-30 03:57:44 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
static LIST_HEAD(rbd_dev_list); /* devices */
|
2012-01-30 03:57:44 +08:00
|
|
|
static DEFINE_SPINLOCK(rbd_dev_list_lock);
|
|
|
|
|
2012-01-30 03:57:44 +08:00
|
|
|
static LIST_HEAD(rbd_client_list); /* clients */
|
|
|
|
static DEFINE_SPINLOCK(rbd_client_list_lock);
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-05-02 01:43:04 +08:00
|
|
|
/* Slab caches for frequently-allocated structures */
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
static struct kmem_cache *rbd_img_request_cache;
|
2013-05-02 01:43:03 +08:00
|
|
|
static struct kmem_cache *rbd_obj_request_cache;
|
2013-05-02 01:43:03 +08:00
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
static int rbd_major;
|
2013-12-13 21:28:57 +08:00
|
|
|
static DEFINE_IDA(rbd_dev_id_ida);
|
|
|
|
|
2014-10-09 21:06:01 +08:00
|
|
|
static struct workqueue_struct *rbd_wq;
|
|
|
|
|
2019-02-28 21:20:28 +08:00
|
|
|
static struct ceph_snap_context rbd_empty_snapc = {
|
|
|
|
.nref = REFCOUNT_INIT(1),
|
|
|
|
};
|
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
/*
|
2017-11-13 17:35:40 +08:00
|
|
|
* single-major requires >= 0.75 version of userspace rbd utility.
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
*/
|
2017-11-13 17:35:40 +08:00
|
|
|
static bool single_major = true;
|
2018-05-25 03:38:59 +08:00
|
|
|
module_param(single_major, bool, 0444);
|
2017-11-13 17:35:40 +08:00
|
|
|
MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)");
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count);
|
|
|
|
static ssize_t remove_store(struct bus_type *bus, const char *buf,
|
|
|
|
size_t count);
|
|
|
|
static ssize_t add_single_major_store(struct bus_type *bus, const char *buf,
|
|
|
|
size_t count);
|
|
|
|
static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf,
|
|
|
|
size_t count);
|
2015-10-12 01:38:00 +08:00
|
|
|
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth);
|
2012-01-30 03:57:44 +08:00
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
static int rbd_dev_id_to_minor(int dev_id)
|
|
|
|
{
|
2013-12-17 01:26:32 +08:00
|
|
|
return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT;
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int minor_to_rbd_dev_id(int minor)
|
|
|
|
{
|
2013-12-17 01:26:32 +08:00
|
|
|
return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED ||
|
|
|
|
rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool rbd_is_lock_owner(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
bool is_lock_owner;
|
|
|
|
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
|
|
|
is_lock_owner = __rbd_is_lock_owner(rbd_dev);
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
return is_lock_owner;
|
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t supported_features_show(struct bus_type *bus, char *buf)
|
2017-03-03 02:56:57 +08:00
|
|
|
{
|
|
|
|
return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
|
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static BUS_ATTR_WO(add);
|
|
|
|
static BUS_ATTR_WO(remove);
|
|
|
|
static BUS_ATTR_WO(add_single_major);
|
|
|
|
static BUS_ATTR_WO(remove_single_major);
|
|
|
|
static BUS_ATTR_RO(supported_features);
|
2013-08-24 05:24:28 +08:00
|
|
|
|
|
|
|
static struct attribute *rbd_bus_attrs[] = {
|
|
|
|
&bus_attr_add.attr,
|
|
|
|
&bus_attr_remove.attr,
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
&bus_attr_add_single_major.attr,
|
|
|
|
&bus_attr_remove_single_major.attr,
|
2017-03-03 02:56:57 +08:00
|
|
|
&bus_attr_supported_features.attr,
|
2013-08-24 05:24:28 +08:00
|
|
|
NULL,
|
2012-01-30 03:57:44 +08:00
|
|
|
};
|
2013-12-13 21:28:57 +08:00
|
|
|
|
|
|
|
static umode_t rbd_bus_is_visible(struct kobject *kobj,
|
|
|
|
struct attribute *attr, int index)
|
|
|
|
{
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
if (!single_major &&
|
|
|
|
(attr == &bus_attr_add_single_major.attr ||
|
|
|
|
attr == &bus_attr_remove_single_major.attr))
|
|
|
|
return 0;
|
|
|
|
|
2013-12-13 21:28:57 +08:00
|
|
|
return attr->mode;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const struct attribute_group rbd_bus_group = {
|
|
|
|
.attrs = rbd_bus_attrs,
|
|
|
|
.is_visible = rbd_bus_is_visible,
|
|
|
|
};
|
|
|
|
__ATTRIBUTE_GROUPS(rbd_bus);
|
2012-01-30 03:57:44 +08:00
|
|
|
|
|
|
|
static struct bus_type rbd_bus_type = {
|
|
|
|
.name = "rbd",
|
2013-08-24 05:24:28 +08:00
|
|
|
.bus_groups = rbd_bus_groups,
|
2012-01-30 03:57:44 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
static void rbd_root_dev_release(struct device *dev)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct device rbd_root_dev = {
|
|
|
|
.init_name = "rbd",
|
|
|
|
.release = rbd_root_dev_release,
|
|
|
|
};
|
|
|
|
|
2012-11-01 23:17:15 +08:00
|
|
|
static __printf(2, 3)
|
|
|
|
void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
|
|
|
|
{
|
|
|
|
struct va_format vaf;
|
|
|
|
va_list args;
|
|
|
|
|
|
|
|
va_start(args, fmt);
|
|
|
|
vaf.fmt = fmt;
|
|
|
|
vaf.va = &args;
|
|
|
|
|
|
|
|
if (!rbd_dev)
|
|
|
|
printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf);
|
|
|
|
else if (rbd_dev->disk)
|
|
|
|
printk(KERN_WARNING "%s: %s: %pV\n",
|
|
|
|
RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf);
|
|
|
|
else if (rbd_dev->spec && rbd_dev->spec->image_name)
|
|
|
|
printk(KERN_WARNING "%s: image %s: %pV\n",
|
|
|
|
RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf);
|
|
|
|
else if (rbd_dev->spec && rbd_dev->spec->image_id)
|
|
|
|
printk(KERN_WARNING "%s: id %s: %pV\n",
|
|
|
|
RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf);
|
|
|
|
else /* punt */
|
|
|
|
printk(KERN_WARNING "%s: rbd_dev %p: %pV\n",
|
|
|
|
RBD_DRV_NAME, rbd_dev, &vaf);
|
|
|
|
va_end(args);
|
|
|
|
}
|
|
|
|
|
2012-09-07 05:00:54 +08:00
|
|
|
#ifdef RBD_DEBUG
|
|
|
|
#define rbd_assert(expr) \
|
|
|
|
if (unlikely(!(expr))) { \
|
|
|
|
printk(KERN_ERR "\nAssertion failure in %s() " \
|
|
|
|
"at line %d:\n\n" \
|
|
|
|
"\trbd_assert(%s);\n\n", \
|
|
|
|
__func__, __LINE__, #expr); \
|
|
|
|
BUG(); \
|
|
|
|
}
|
|
|
|
#else /* !RBD_DEBUG */
|
|
|
|
# define rbd_assert(expr) ((void) 0)
|
|
|
|
#endif /* !RBD_DEBUG */
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static int rbd_dev_refresh(struct rbd_device *rbd_dev);
|
2013-05-06 22:51:30 +08:00
|
|
|
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
|
2014-07-23 21:11:19 +08:00
|
|
|
static int rbd_dev_header_info(struct rbd_device *rbd_dev);
|
2014-07-24 14:42:13 +08:00
|
|
|
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
|
2013-04-30 13:44:33 +08:00
|
|
|
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
|
|
|
|
u64 snap_id);
|
2013-04-30 13:44:33 +08:00
|
|
|
static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u8 *order, u64 *snap_size);
|
|
|
|
static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u64 *snap_features);
|
2011-03-22 06:10:11 +08:00
|
|
|
|
2019-05-11 22:21:49 +08:00
|
|
|
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result);
|
2019-05-16 21:06:56 +08:00
|
|
|
static void rbd_img_handle_request(struct rbd_img_request *img_req, int result);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Return true if nothing else is pending.
|
|
|
|
*/
|
|
|
|
static bool pending_result_dec(struct pending_result *pending, int *result)
|
|
|
|
{
|
|
|
|
rbd_assert(pending->num_pending > 0);
|
|
|
|
|
|
|
|
if (*result && !pending->result)
|
|
|
|
pending->result = *result;
|
|
|
|
if (--pending->num_pending)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
*result = pending->result;
|
|
|
|
return true;
|
|
|
|
}
|
2019-05-11 22:21:49 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
static int rbd_open(struct block_device *bdev, fmode_t mode)
|
|
|
|
{
|
2012-01-30 03:57:44 +08:00
|
|
|
struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
|
2013-01-15 02:43:31 +08:00
|
|
|
bool removing = false;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-02-06 03:23:12 +08:00
|
|
|
spin_lock_irq(&rbd_dev->lock);
|
2013-01-15 02:43:31 +08:00
|
|
|
if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags))
|
|
|
|
removing = true;
|
|
|
|
else
|
|
|
|
rbd_dev->open_count++;
|
2013-02-06 03:23:12 +08:00
|
|
|
spin_unlock_irq(&rbd_dev->lock);
|
2013-01-15 02:43:31 +08:00
|
|
|
if (removing)
|
|
|
|
return -ENOENT;
|
|
|
|
|
2012-11-16 23:29:16 +08:00
|
|
|
(void) get_device(&rbd_dev->dev);
|
2012-08-11 04:12:07 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-05-06 09:52:57 +08:00
|
|
|
static void rbd_release(struct gendisk *disk, fmode_t mode)
|
2010-11-20 06:51:04 +08:00
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = disk->private_data;
|
2013-01-15 02:43:31 +08:00
|
|
|
unsigned long open_count_before;
|
|
|
|
|
2013-02-06 03:23:12 +08:00
|
|
|
spin_lock_irq(&rbd_dev->lock);
|
2013-01-15 02:43:31 +08:00
|
|
|
open_count_before = rbd_dev->open_count--;
|
2013-02-06 03:23:12 +08:00
|
|
|
spin_unlock_irq(&rbd_dev->lock);
|
2013-01-15 02:43:31 +08:00
|
|
|
rbd_assert(open_count_before > 0);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2012-11-16 23:29:16 +08:00
|
|
|
put_device(&rbd_dev->dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2013-09-24 11:25:36 +08:00
|
|
|
static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg)
|
|
|
|
{
|
2017-10-12 18:35:19 +08:00
|
|
|
int ro;
|
2013-09-24 11:25:36 +08:00
|
|
|
|
2017-10-12 18:35:19 +08:00
|
|
|
if (get_user(ro, (int __user *)arg))
|
2013-09-24 11:25:36 +08:00
|
|
|
return -EFAULT;
|
|
|
|
|
2017-10-12 18:35:19 +08:00
|
|
|
/* Snapshots can't be marked read-write */
|
2013-09-24 11:25:36 +08:00
|
|
|
if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro)
|
|
|
|
return -EROFS;
|
|
|
|
|
2017-10-12 18:35:19 +08:00
|
|
|
/* Let blkdev_roset() handle it */
|
|
|
|
return -ENOTTY;
|
2013-09-24 11:25:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_ioctl(struct block_device *bdev, fmode_t mode,
|
|
|
|
unsigned int cmd, unsigned long arg)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = bdev->bd_disk->private_data;
|
2017-10-12 18:35:19 +08:00
|
|
|
int ret;
|
2013-09-24 11:25:36 +08:00
|
|
|
|
|
|
|
switch (cmd) {
|
|
|
|
case BLKROSET:
|
|
|
|
ret = rbd_ioctl_set_ro(rbd_dev, arg);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -ENOTTY;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef CONFIG_COMPAT
|
|
|
|
static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode,
|
|
|
|
unsigned int cmd, unsigned long arg)
|
|
|
|
{
|
|
|
|
return rbd_ioctl(bdev, mode, cmd, arg);
|
|
|
|
}
|
|
|
|
#endif /* CONFIG_COMPAT */
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
static const struct block_device_operations rbd_bd_ops = {
|
|
|
|
.owner = THIS_MODULE,
|
|
|
|
.open = rbd_open,
|
2010-11-20 06:51:04 +08:00
|
|
|
.release = rbd_release,
|
2013-09-24 11:25:36 +08:00
|
|
|
.ioctl = rbd_ioctl,
|
|
|
|
#ifdef CONFIG_COMPAT
|
|
|
|
.compat_ioctl = rbd_compat_ioctl,
|
|
|
|
#endif
|
2010-08-13 07:11:25 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
2013-05-17 04:04:20 +08:00
|
|
|
* Initialize an rbd client instance. Success or not, this function
|
2013-06-01 06:40:45 +08:00
|
|
|
* consumes ceph_opts. Caller holds client_mutex.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
2012-08-11 04:12:07 +08:00
|
|
|
static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
|
|
|
struct rbd_client *rbdc;
|
|
|
|
int ret = -ENOMEM;
|
|
|
|
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s:\n", __func__);
|
2010-08-13 07:11:25 +08:00
|
|
|
rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
|
|
|
|
if (!rbdc)
|
|
|
|
goto out_opt;
|
|
|
|
|
|
|
|
kref_init(&rbdc->kref);
|
|
|
|
INIT_LIST_HEAD(&rbdc->node);
|
|
|
|
|
2017-03-04 01:16:07 +08:00
|
|
|
rbdc->client = ceph_create_client(ceph_opts, rbdc);
|
2010-08-13 07:11:25 +08:00
|
|
|
if (IS_ERR(rbdc->client))
|
2013-05-30 00:19:00 +08:00
|
|
|
goto out_rbdc;
|
2012-07-04 05:01:18 +08:00
|
|
|
ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
ret = ceph_open_session(rbdc->client);
|
|
|
|
if (ret < 0)
|
2013-05-30 00:19:00 +08:00
|
|
|
goto out_client;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-01-30 03:57:44 +08:00
|
|
|
spin_lock(&rbd_client_list_lock);
|
2010-08-13 07:11:25 +08:00
|
|
|
list_add_tail(&rbdc->node, &rbd_client_list);
|
2012-01-30 03:57:44 +08:00
|
|
|
spin_unlock(&rbd_client_list_lock);
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: rbdc %p\n", __func__, rbdc);
|
2012-01-30 03:57:44 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
return rbdc;
|
2013-05-30 00:19:00 +08:00
|
|
|
out_client:
|
2010-08-13 07:11:25 +08:00
|
|
|
ceph_destroy_client(rbdc->client);
|
2013-05-30 00:19:00 +08:00
|
|
|
out_rbdc:
|
2010-08-13 07:11:25 +08:00
|
|
|
kfree(rbdc);
|
|
|
|
out_opt:
|
2012-07-04 05:01:18 +08:00
|
|
|
if (ceph_opts)
|
|
|
|
ceph_destroy_options(ceph_opts);
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: error %d\n", __func__, ret);
|
|
|
|
|
2010-09-26 16:59:37 +08:00
|
|
|
return ERR_PTR(ret);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc)
|
|
|
|
{
|
|
|
|
kref_get(&rbdc->kref);
|
|
|
|
|
|
|
|
return rbdc;
|
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
2012-08-11 04:12:07 +08:00
|
|
|
* Find a ceph client with specific addr and configuration. If
|
|
|
|
* found, bump its reference count.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
2012-08-11 04:12:07 +08:00
|
|
|
static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
|
|
|
struct rbd_client *client_node;
|
2012-08-11 04:12:07 +08:00
|
|
|
bool found = false;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-07-04 05:01:18 +08:00
|
|
|
if (ceph_opts->flags & CEPH_OPT_NOSHARE)
|
2010-08-13 07:11:25 +08:00
|
|
|
return NULL;
|
|
|
|
|
2012-08-11 04:12:07 +08:00
|
|
|
spin_lock(&rbd_client_list_lock);
|
|
|
|
list_for_each_entry(client_node, &rbd_client_list, node) {
|
|
|
|
if (!ceph_compare_options(ceph_opts, client_node->client)) {
|
2012-10-31 08:40:33 +08:00
|
|
|
__rbd_get_client(client_node);
|
|
|
|
|
2012-08-11 04:12:07 +08:00
|
|
|
found = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
spin_unlock(&rbd_client_list_lock);
|
|
|
|
|
|
|
|
return found ? client_node : NULL;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2011-03-22 06:10:11 +08:00
|
|
|
/*
|
2015-06-22 18:24:48 +08:00
|
|
|
* (Per device) rbd map options
|
2011-03-22 06:10:11 +08:00
|
|
|
*/
|
|
|
|
enum {
|
2015-06-23 21:21:19 +08:00
|
|
|
Opt_queue_depth,
|
2019-01-30 22:14:48 +08:00
|
|
|
Opt_alloc_size,
|
2018-03-26 22:22:55 +08:00
|
|
|
Opt_lock_timeout,
|
2011-03-22 06:10:11 +08:00
|
|
|
Opt_last_int,
|
|
|
|
/* int args above */
|
2018-07-03 21:28:43 +08:00
|
|
|
Opt_pool_ns,
|
2011-03-22 06:10:11 +08:00
|
|
|
Opt_last_string,
|
|
|
|
/* string args above */
|
2012-08-11 04:12:07 +08:00
|
|
|
Opt_read_only,
|
|
|
|
Opt_read_write,
|
2016-09-20 20:23:17 +08:00
|
|
|
Opt_lock_on_read,
|
2017-04-13 18:17:39 +08:00
|
|
|
Opt_exclusive,
|
2018-03-23 13:14:47 +08:00
|
|
|
Opt_notrim,
|
2015-06-22 18:24:48 +08:00
|
|
|
Opt_err
|
2011-03-22 06:10:11 +08:00
|
|
|
};
|
|
|
|
|
2012-07-04 05:01:18 +08:00
|
|
|
static match_table_t rbd_opts_tokens = {
|
2015-06-23 21:21:19 +08:00
|
|
|
{Opt_queue_depth, "queue_depth=%d"},
|
2019-01-30 22:14:48 +08:00
|
|
|
{Opt_alloc_size, "alloc_size=%d"},
|
2018-03-26 22:22:55 +08:00
|
|
|
{Opt_lock_timeout, "lock_timeout=%d"},
|
2011-03-22 06:10:11 +08:00
|
|
|
/* int args above */
|
2018-07-03 21:28:43 +08:00
|
|
|
{Opt_pool_ns, "_pool_ns=%s"},
|
2011-03-22 06:10:11 +08:00
|
|
|
/* string args above */
|
2012-10-23 00:31:26 +08:00
|
|
|
{Opt_read_only, "read_only"},
|
2012-08-11 04:12:07 +08:00
|
|
|
{Opt_read_only, "ro"}, /* Alternate spelling */
|
|
|
|
{Opt_read_write, "read_write"},
|
|
|
|
{Opt_read_write, "rw"}, /* Alternate spelling */
|
2016-09-20 20:23:17 +08:00
|
|
|
{Opt_lock_on_read, "lock_on_read"},
|
2017-04-13 18:17:39 +08:00
|
|
|
{Opt_exclusive, "exclusive"},
|
2018-03-23 13:14:47 +08:00
|
|
|
{Opt_notrim, "notrim"},
|
2015-06-22 18:24:48 +08:00
|
|
|
{Opt_err, NULL}
|
2011-03-22 06:10:11 +08:00
|
|
|
};
|
|
|
|
|
2013-01-21 04:44:42 +08:00
|
|
|
struct rbd_options {
|
2015-06-23 21:21:19 +08:00
|
|
|
int queue_depth;
|
2019-01-30 22:14:48 +08:00
|
|
|
int alloc_size;
|
2018-03-26 22:22:55 +08:00
|
|
|
unsigned long lock_timeout;
|
2013-01-21 04:44:42 +08:00
|
|
|
bool read_only;
|
2016-09-20 20:23:17 +08:00
|
|
|
bool lock_on_read;
|
2017-04-13 18:17:39 +08:00
|
|
|
bool exclusive;
|
2018-03-23 13:14:47 +08:00
|
|
|
bool trim;
|
2013-01-21 04:44:42 +08:00
|
|
|
};
|
|
|
|
|
2015-06-23 21:21:19 +08:00
|
|
|
#define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ
|
2019-01-30 22:14:48 +08:00
|
|
|
#define RBD_ALLOC_SIZE_DEFAULT (64 * 1024)
|
2018-03-26 22:22:55 +08:00
|
|
|
#define RBD_LOCK_TIMEOUT_DEFAULT 0 /* no timeout */
|
2013-01-21 04:44:42 +08:00
|
|
|
#define RBD_READ_ONLY_DEFAULT false
|
2016-09-20 20:23:17 +08:00
|
|
|
#define RBD_LOCK_ON_READ_DEFAULT false
|
2017-04-13 18:17:39 +08:00
|
|
|
#define RBD_EXCLUSIVE_DEFAULT false
|
2018-03-23 13:14:47 +08:00
|
|
|
#define RBD_TRIM_DEFAULT true
|
2013-01-21 04:44:42 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
struct parse_rbd_opts_ctx {
|
|
|
|
struct rbd_spec *spec;
|
|
|
|
struct rbd_options *opts;
|
|
|
|
};
|
|
|
|
|
2011-03-22 06:10:11 +08:00
|
|
|
static int parse_rbd_opts_token(char *c, void *private)
|
|
|
|
{
|
2018-07-03 21:28:43 +08:00
|
|
|
struct parse_rbd_opts_ctx *pctx = private;
|
2011-03-22 06:10:11 +08:00
|
|
|
substring_t argstr[MAX_OPT_ARGS];
|
|
|
|
int token, intval, ret;
|
|
|
|
|
2012-07-04 05:01:18 +08:00
|
|
|
token = match_token(c, rbd_opts_tokens, argstr);
|
2011-03-22 06:10:11 +08:00
|
|
|
if (token < Opt_last_int) {
|
|
|
|
ret = match_int(&argstr[0], &intval);
|
|
|
|
if (ret < 0) {
|
2018-06-27 22:38:13 +08:00
|
|
|
pr_err("bad option arg (not int) at '%s'\n", c);
|
2011-03-22 06:10:11 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
dout("got int token %d val %d\n", token, intval);
|
|
|
|
} else if (token > Opt_last_int && token < Opt_last_string) {
|
2015-06-22 18:24:48 +08:00
|
|
|
dout("got string token %d val %s\n", token, argstr[0].from);
|
2011-03-22 06:10:11 +08:00
|
|
|
} else {
|
|
|
|
dout("got token %d\n", token);
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (token) {
|
2015-06-23 21:21:19 +08:00
|
|
|
case Opt_queue_depth:
|
|
|
|
if (intval < 1) {
|
|
|
|
pr_err("queue_depth out of range\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->queue_depth = intval;
|
2015-06-23 21:21:19 +08:00
|
|
|
break;
|
2019-01-30 22:14:48 +08:00
|
|
|
case Opt_alloc_size:
|
2019-03-15 21:50:04 +08:00
|
|
|
if (intval < SECTOR_SIZE) {
|
2019-01-30 22:14:48 +08:00
|
|
|
pr_err("alloc_size out of range\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
if (!is_power_of_2(intval)) {
|
|
|
|
pr_err("alloc_size must be a power of 2\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
pctx->opts->alloc_size = intval;
|
|
|
|
break;
|
2018-03-26 22:22:55 +08:00
|
|
|
case Opt_lock_timeout:
|
|
|
|
/* 0 is "wait forever" (i.e. infinite timeout) */
|
|
|
|
if (intval < 0 || intval > INT_MAX / 1000) {
|
|
|
|
pr_err("lock_timeout out of range\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->lock_timeout = msecs_to_jiffies(intval * 1000);
|
2018-03-26 22:22:55 +08:00
|
|
|
break;
|
2018-07-03 21:28:43 +08:00
|
|
|
case Opt_pool_ns:
|
|
|
|
kfree(pctx->spec->pool_ns);
|
|
|
|
pctx->spec->pool_ns = match_strdup(argstr);
|
|
|
|
if (!pctx->spec->pool_ns)
|
|
|
|
return -ENOMEM;
|
2018-03-26 22:22:55 +08:00
|
|
|
break;
|
2012-08-11 04:12:07 +08:00
|
|
|
case Opt_read_only:
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->read_only = true;
|
2012-08-11 04:12:07 +08:00
|
|
|
break;
|
|
|
|
case Opt_read_write:
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->read_only = false;
|
2012-08-11 04:12:07 +08:00
|
|
|
break;
|
2016-09-20 20:23:17 +08:00
|
|
|
case Opt_lock_on_read:
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->lock_on_read = true;
|
2016-09-20 20:23:17 +08:00
|
|
|
break;
|
2017-04-13 18:17:39 +08:00
|
|
|
case Opt_exclusive:
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->exclusive = true;
|
2017-04-13 18:17:39 +08:00
|
|
|
break;
|
2018-03-23 13:14:47 +08:00
|
|
|
case Opt_notrim:
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx->opts->trim = false;
|
2018-03-23 13:14:47 +08:00
|
|
|
break;
|
2011-03-22 06:10:11 +08:00
|
|
|
default:
|
2015-06-22 18:24:48 +08:00
|
|
|
/* libceph prints "bad option" msg */
|
|
|
|
return -EINVAL;
|
2011-03-22 06:10:11 +08:00
|
|
|
}
|
2015-06-22 18:24:48 +08:00
|
|
|
|
2011-03-22 06:10:11 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-03-13 11:21:35 +08:00
|
|
|
static char* obj_op_name(enum obj_operation_type op_type)
|
|
|
|
{
|
|
|
|
switch (op_type) {
|
|
|
|
case OBJ_OP_READ:
|
|
|
|
return "read";
|
|
|
|
case OBJ_OP_WRITE:
|
|
|
|
return "write";
|
2014-04-01 22:22:16 +08:00
|
|
|
case OBJ_OP_DISCARD:
|
|
|
|
return "discard";
|
2019-01-29 19:46:25 +08:00
|
|
|
case OBJ_OP_ZEROOUT:
|
|
|
|
return "zeroout";
|
2014-03-13 11:21:35 +08:00
|
|
|
default:
|
|
|
|
return "???";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
|
|
|
* Destroy ceph client
|
2012-01-30 03:57:43 +08:00
|
|
|
*
|
2012-01-30 03:57:44 +08:00
|
|
|
* Caller must hold rbd_client_list_lock.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
|
|
|
static void rbd_client_release(struct kref *kref)
|
|
|
|
{
|
|
|
|
struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
|
|
|
|
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: rbdc %p\n", __func__, rbdc);
|
2012-04-05 02:35:44 +08:00
|
|
|
spin_lock(&rbd_client_list_lock);
|
2010-08-13 07:11:25 +08:00
|
|
|
list_del(&rbdc->node);
|
2012-04-05 02:35:44 +08:00
|
|
|
spin_unlock(&rbd_client_list_lock);
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
ceph_destroy_client(rbdc->client);
|
|
|
|
kfree(rbdc);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Drop reference to ceph client node. If it's not referenced anymore, release
|
|
|
|
* it.
|
|
|
|
*/
|
2012-10-26 12:34:42 +08:00
|
|
|
static void rbd_put_client(struct rbd_client *rbdc)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2012-10-26 12:34:42 +08:00
|
|
|
if (rbdc)
|
|
|
|
kref_put(&rbdc->kref, rbd_client_release);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2018-02-22 20:19:04 +08:00
|
|
|
/*
|
|
|
|
* Get a ceph client with specific addr and configuration, if one does
|
|
|
|
* not exist create it. Either way, ceph_opts is consumed by this
|
|
|
|
* function.
|
|
|
|
*/
|
|
|
|
static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts)
|
|
|
|
{
|
|
|
|
struct rbd_client *rbdc;
|
2018-02-22 20:43:24 +08:00
|
|
|
int ret;
|
2018-02-22 20:19:04 +08:00
|
|
|
|
2019-05-02 21:56:00 +08:00
|
|
|
mutex_lock(&client_mutex);
|
2018-02-22 20:19:04 +08:00
|
|
|
rbdc = rbd_client_find(ceph_opts);
|
2018-02-22 20:43:24 +08:00
|
|
|
if (rbdc) {
|
2018-02-22 20:19:04 +08:00
|
|
|
ceph_destroy_options(ceph_opts);
|
2018-02-22 20:43:24 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Using an existing client. Make sure ->pg_pools is up to
|
|
|
|
* date before we look up the pool id in do_rbd_add().
|
|
|
|
*/
|
2019-03-20 17:58:05 +08:00
|
|
|
ret = ceph_wait_for_latest_osdmap(rbdc->client,
|
|
|
|
rbdc->client->options->mount_timeout);
|
2018-02-22 20:43:24 +08:00
|
|
|
if (ret) {
|
|
|
|
rbd_warn(NULL, "failed to get latest osdmap: %d", ret);
|
|
|
|
rbd_put_client(rbdc);
|
|
|
|
rbdc = ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
} else {
|
2018-02-22 20:19:04 +08:00
|
|
|
rbdc = rbd_client_create(ceph_opts);
|
2018-02-22 20:43:24 +08:00
|
|
|
}
|
2018-02-22 20:19:04 +08:00
|
|
|
mutex_unlock(&client_mutex);
|
|
|
|
|
|
|
|
return rbdc;
|
|
|
|
}
|
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
static bool rbd_image_format_valid(u32 image_format)
|
|
|
|
{
|
|
|
|
return image_format == 1 || image_format == 2;
|
|
|
|
}
|
|
|
|
|
2012-07-25 22:32:40 +08:00
|
|
|
static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk)
|
|
|
|
{
|
2012-08-03 00:29:45 +08:00
|
|
|
size_t size;
|
|
|
|
u32 snap_count;
|
|
|
|
|
|
|
|
/* The header has to start with the magic rbd header text */
|
|
|
|
if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)))
|
|
|
|
return false;
|
|
|
|
|
2012-10-21 11:17:27 +08:00
|
|
|
/* The bio layer requires at least sector-sized I/O */
|
|
|
|
|
|
|
|
if (ondisk->options.order < SECTOR_SHIFT)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* If we use u64 in a few spots we may be able to loosen this */
|
|
|
|
|
|
|
|
if (ondisk->options.order > 8 * sizeof (int) - 1)
|
|
|
|
return false;
|
|
|
|
|
2012-08-03 00:29:45 +08:00
|
|
|
/*
|
|
|
|
* The size of a snapshot header has to fit in a size_t, and
|
|
|
|
* that limits the number of snapshots.
|
|
|
|
*/
|
|
|
|
snap_count = le32_to_cpu(ondisk->snap_count);
|
|
|
|
size = SIZE_MAX - sizeof (struct ceph_snap_context);
|
|
|
|
if (snap_count > size / sizeof (__le64))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Not only that, but the size of the entire the snapshot
|
|
|
|
* header must also be representable in a size_t.
|
|
|
|
*/
|
|
|
|
size -= snap_count * sizeof (__le64);
|
|
|
|
if ((u64) size < le64_to_cpu(ondisk->snap_names_len))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return true;
|
2012-07-25 22:32:40 +08:00
|
|
|
}
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
/*
|
|
|
|
* returns the size of an object in the image
|
|
|
|
*/
|
|
|
|
static u32 rbd_obj_bytes(struct rbd_image_header *header)
|
|
|
|
{
|
|
|
|
return 1U << header->obj_order;
|
|
|
|
}
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
static void rbd_init_layout(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
if (rbd_dev->header.stripe_unit == 0 ||
|
|
|
|
rbd_dev->header.stripe_count == 0) {
|
|
|
|
rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header);
|
|
|
|
rbd_dev->header.stripe_count = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit;
|
|
|
|
rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count;
|
|
|
|
rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header);
|
2017-01-26 01:16:22 +08:00
|
|
|
rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ?
|
|
|
|
rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id;
|
2017-01-26 01:16:22 +08:00
|
|
|
RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
|
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
2013-05-06 22:51:29 +08:00
|
|
|
* Fill an rbd image header with information from the given format 1
|
|
|
|
* on-disk header.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
2013-05-06 22:51:29 +08:00
|
|
|
static int rbd_header_from_disk(struct rbd_device *rbd_dev,
|
2012-08-03 00:29:46 +08:00
|
|
|
struct rbd_image_header_ondisk *ondisk)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2013-05-06 22:51:29 +08:00
|
|
|
struct rbd_image_header *header = &rbd_dev->header;
|
2013-05-06 22:51:29 +08:00
|
|
|
bool first_time = header->object_prefix == NULL;
|
|
|
|
struct ceph_snap_context *snapc;
|
|
|
|
char *object_prefix = NULL;
|
|
|
|
char *snap_names = NULL;
|
|
|
|
u64 *snap_sizes = NULL;
|
rbd: fixes in rbd_header_from_disk()
This fixes a few issues in rbd_header_from_disk():
- There is a check intended to catch overflow, but it's wrong in
two ways.
- First, the type we don't want to overflow is size_t, not
unsigned int, and there is now a SIZE_MAX we can use for
use with that type.
- Second, we're allocating the snapshot ids and snapshot
image sizes separately (each has type u64; on disk they
grouped together as a rbd_image_header_ondisk structure).
So we can use the size of u64 in this overflow check.
- If there are no snapshots, then there should be no snapshot
names. Enforce this, and issue a warning if we encounter a
header with no snapshots but a non-zero snap_names_len.
- When saving the snapshot names into the header, be more direct
in defining the offset in the on-disk structure from which
they're being copied by using "snap_count" rather than "i"
in the array index.
- If an error occurs, the "snapc" and "snap_names" fields are
freed at the end of the function. Make those fields be null
pointers after they're freed, to be explicit that they are
no longer valid.
- Finally, move the definition of the local variable "i" to the
innermost scope in which it's needed.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2012-07-11 09:30:10 +08:00
|
|
|
u32 snap_count;
|
2013-05-06 22:51:29 +08:00
|
|
|
int ret = -ENOMEM;
|
2012-08-24 12:22:06 +08:00
|
|
|
u32 i;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* Allocate this now to avoid having to handle failure below */
|
2012-07-20 06:12:59 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
if (first_time) {
|
2017-01-26 01:16:21 +08:00
|
|
|
object_prefix = kstrndup(ondisk->object_prefix,
|
|
|
|
sizeof(ondisk->object_prefix),
|
|
|
|
GFP_KERNEL);
|
2013-05-06 22:51:29 +08:00
|
|
|
if (!object_prefix)
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2012-02-08 02:03:36 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* Allocate the snapshot context and fill it in */
|
2012-02-08 02:03:36 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
snap_count = le32_to_cpu(ondisk->snap_count);
|
|
|
|
snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
|
|
|
|
if (!snapc)
|
|
|
|
goto out_err;
|
|
|
|
snapc->seq = le64_to_cpu(ondisk->snap_seq);
|
2010-08-13 07:11:25 +08:00
|
|
|
if (snap_count) {
|
2013-05-06 22:51:29 +08:00
|
|
|
struct rbd_image_snap_ondisk *snaps;
|
2012-08-24 12:22:06 +08:00
|
|
|
u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len);
|
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* We'll keep a copy of the snapshot names... */
|
2012-08-24 12:22:06 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
if (snap_names_len > (u64)SIZE_MAX)
|
|
|
|
goto out_2big;
|
|
|
|
snap_names = kmalloc(snap_names_len, GFP_KERNEL);
|
|
|
|
if (!snap_names)
|
2012-07-20 06:12:59 +08:00
|
|
|
goto out_err;
|
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* ...as well as the array of their sizes. */
|
2016-09-11 18:21:25 +08:00
|
|
|
snap_sizes = kmalloc_array(snap_count,
|
|
|
|
sizeof(*header->snap_sizes),
|
|
|
|
GFP_KERNEL);
|
2013-05-06 22:51:29 +08:00
|
|
|
if (!snap_sizes)
|
2012-07-20 06:12:59 +08:00
|
|
|
goto out_err;
|
2013-05-06 22:51:29 +08:00
|
|
|
|
2012-08-24 12:22:06 +08:00
|
|
|
/*
|
2013-05-06 22:51:29 +08:00
|
|
|
* Copy the names, and fill in each snapshot's id
|
|
|
|
* and size.
|
|
|
|
*
|
2013-05-06 22:51:30 +08:00
|
|
|
* Note that rbd_dev_v1_header_info() guarantees the
|
2013-05-06 22:51:29 +08:00
|
|
|
* ondisk buffer we're working with has
|
2012-08-24 12:22:06 +08:00
|
|
|
* snap_names_len bytes beyond the end of the
|
|
|
|
* snapshot id array, this memcpy() is safe.
|
|
|
|
*/
|
2013-05-06 22:51:29 +08:00
|
|
|
memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len);
|
|
|
|
snaps = ondisk->snaps;
|
|
|
|
for (i = 0; i < snap_count; i++) {
|
|
|
|
snapc->snaps[i] = le64_to_cpu(snaps[i].id);
|
|
|
|
snap_sizes[i] = le64_to_cpu(snaps[i].image_size);
|
|
|
|
}
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
2012-07-20 06:12:59 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* We won't fail any more, fill in the header */
|
2012-08-24 12:22:06 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
if (first_time) {
|
|
|
|
header->object_prefix = object_prefix;
|
|
|
|
header->obj_order = ondisk->options.order;
|
2017-01-26 01:16:22 +08:00
|
|
|
rbd_init_layout(rbd_dev);
|
2010-08-13 07:11:25 +08:00
|
|
|
} else {
|
2013-05-06 22:51:29 +08:00
|
|
|
ceph_put_snap_context(header->snapc);
|
|
|
|
kfree(header->snap_names);
|
|
|
|
kfree(header->snap_sizes);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
2012-07-10 10:04:24 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
/* The remaining fields always get updated (when we refresh) */
|
2012-08-24 12:22:06 +08:00
|
|
|
|
2012-09-01 06:29:51 +08:00
|
|
|
header->image_size = le64_to_cpu(ondisk->image_size);
|
2013-05-06 22:51:29 +08:00
|
|
|
header->snapc = snapc;
|
|
|
|
header->snap_names = snap_names;
|
|
|
|
header->snap_sizes = snap_sizes;
|
2013-04-26 22:43:47 +08:00
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
return 0;
|
2013-05-06 22:51:29 +08:00
|
|
|
out_2big:
|
|
|
|
ret = -EIO;
|
2012-07-20 06:12:59 +08:00
|
|
|
out_err:
|
2013-05-06 22:51:29 +08:00
|
|
|
kfree(snap_sizes);
|
|
|
|
kfree(snap_names);
|
|
|
|
ceph_put_snap_context(snapc);
|
|
|
|
kfree(object_prefix);
|
rbd: fixes in rbd_header_from_disk()
This fixes a few issues in rbd_header_from_disk():
- There is a check intended to catch overflow, but it's wrong in
two ways.
- First, the type we don't want to overflow is size_t, not
unsigned int, and there is now a SIZE_MAX we can use for
use with that type.
- Second, we're allocating the snapshot ids and snapshot
image sizes separately (each has type u64; on disk they
grouped together as a rbd_image_header_ondisk structure).
So we can use the size of u64 in this overflow check.
- If there are no snapshots, then there should be no snapshot
names. Enforce this, and issue a warning if we encounter a
header with no snapshots but a non-zero snap_names_len.
- When saving the snapshot names into the header, be more direct
in defining the offset in the on-disk structure from which
they're being copied by using "snap_count" rather than "i"
in the array index.
- If an error occurs, the "snapc" and "snap_names" fields are
freed at the end of the function. Make those fields be null
pointers after they're freed, to be explicit that they are
no longer valid.
- Finally, move the definition of the local variable "i" to the
innermost scope in which it's needed.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2012-07-11 09:30:10 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
return ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which)
|
|
|
|
{
|
|
|
|
const char *snap_name;
|
|
|
|
|
|
|
|
rbd_assert(which < rbd_dev->header.snapc->num_snaps);
|
|
|
|
|
|
|
|
/* Skip over names until we find the one we are looking for */
|
|
|
|
|
|
|
|
snap_name = rbd_dev->header.snap_names;
|
|
|
|
while (which--)
|
|
|
|
snap_name += strlen(snap_name) + 1;
|
|
|
|
|
|
|
|
return kstrdup(snap_name, GFP_KERNEL);
|
|
|
|
}
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
/*
|
|
|
|
* Snapshot id comparison function for use with qsort()/bsearch().
|
|
|
|
* Note that result is for snapshots in *descending* order.
|
|
|
|
*/
|
|
|
|
static int snapid_compare_reverse(const void *s1, const void *s2)
|
|
|
|
{
|
|
|
|
u64 snap_id1 = *(u64 *)s1;
|
|
|
|
u64 snap_id2 = *(u64 *)s2;
|
|
|
|
|
|
|
|
if (snap_id1 < snap_id2)
|
|
|
|
return 1;
|
|
|
|
return snap_id1 == snap_id2 ? 0 : -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Search a snapshot context to see if the given snapshot id is
|
|
|
|
* present.
|
|
|
|
*
|
|
|
|
* Returns the position of the snapshot id in the array if it's found,
|
|
|
|
* or BAD_SNAP_INDEX otherwise.
|
|
|
|
*
|
|
|
|
* Note: The snapshot array is in kept sorted (by the osd) in
|
|
|
|
* reverse order, highest snapshot id first.
|
|
|
|
*/
|
2013-04-30 13:44:33 +08:00
|
|
|
static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id)
|
|
|
|
{
|
|
|
|
struct ceph_snap_context *snapc = rbd_dev->header.snapc;
|
2013-05-02 01:43:03 +08:00
|
|
|
u64 *found;
|
2013-04-30 13:44:33 +08:00
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps,
|
|
|
|
sizeof (snap_id), snapid_compare_reverse);
|
2013-04-30 13:44:33 +08:00
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX;
|
2013-04-30 13:44:33 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
|
|
|
|
u64 snap_id)
|
2012-10-31 08:40:33 +08:00
|
|
|
{
|
2013-04-30 13:44:33 +08:00
|
|
|
u32 which;
|
2013-09-05 08:57:31 +08:00
|
|
|
const char *snap_name;
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
which = rbd_dev_snap_index(rbd_dev, snap_id);
|
|
|
|
if (which == BAD_SNAP_INDEX)
|
2013-09-05 08:57:31 +08:00
|
|
|
return ERR_PTR(-ENOENT);
|
2013-04-30 13:44:33 +08:00
|
|
|
|
2013-09-05 08:57:31 +08:00
|
|
|
snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
|
|
|
|
return snap_name ? snap_name : ERR_PTR(-ENOMEM);
|
2013-04-30 13:44:33 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
|
|
|
|
{
|
2012-10-31 08:40:33 +08:00
|
|
|
if (snap_id == CEPH_NOSNAP)
|
|
|
|
return RBD_SNAP_HEAD_NAME;
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
|
|
|
|
if (rbd_dev->image_format == 1)
|
|
|
|
return rbd_dev_v1_snap_name(rbd_dev, snap_id);
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
return rbd_dev_v2_snap_name(rbd_dev, snap_id);
|
2012-10-31 08:40:33 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u64 *snap_size)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2013-04-30 13:44:33 +08:00
|
|
|
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
|
|
|
|
if (snap_id == CEPH_NOSNAP) {
|
|
|
|
*snap_size = rbd_dev->header.image_size;
|
|
|
|
} else if (rbd_dev->image_format == 1) {
|
|
|
|
u32 which;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
which = rbd_dev_snap_index(rbd_dev, snap_id);
|
|
|
|
if (which == BAD_SNAP_INDEX)
|
|
|
|
return -ENOENT;
|
2012-07-11 09:30:11 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
*snap_size = rbd_dev->header.snap_sizes[which];
|
|
|
|
} else {
|
|
|
|
u64 size = 0;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
*snap_size = size;
|
|
|
|
}
|
|
|
|
return 0;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u64 *snap_features)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2013-04-30 13:44:33 +08:00
|
|
|
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
|
|
|
|
if (snap_id == CEPH_NOSNAP) {
|
|
|
|
*snap_features = rbd_dev->header.features;
|
|
|
|
} else if (rbd_dev->image_format == 1) {
|
|
|
|
*snap_features = 0; /* No features for format 1 */
|
2010-08-13 07:11:25 +08:00
|
|
|
} else {
|
2013-04-30 13:44:33 +08:00
|
|
|
u64 features = 0;
|
|
|
|
int ret;
|
2013-04-26 12:15:08 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
*snap_features = features;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_dev_mapping_set(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2013-05-06 20:40:30 +08:00
|
|
|
u64 snap_id = rbd_dev->spec->snap_id;
|
2013-04-30 13:44:33 +08:00
|
|
|
u64 size = 0;
|
|
|
|
u64 features = 0;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = rbd_snap_size(rbd_dev, snap_id, &size);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
ret = rbd_snap_features(rbd_dev, snap_id, &features);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
rbd_dev->mapping.size = size;
|
|
|
|
rbd_dev->mapping.features = features;
|
|
|
|
|
2013-04-26 12:15:08 +08:00
|
|
|
return 0;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-04-27 22:59:30 +08:00
|
|
|
static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
rbd_dev->mapping.size = 0;
|
|
|
|
rbd_dev->mapping.features = 0;
|
2013-04-29 12:32:34 +08:00
|
|
|
}
|
|
|
|
|
2018-01-20 17:30:10 +08:00
|
|
|
static void zero_bvec(struct bio_vec *bv)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
|
|
|
void *buf;
|
2018-01-20 17:30:10 +08:00
|
|
|
unsigned long flags;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2018-01-20 17:30:10 +08:00
|
|
|
buf = bvec_kmap_irq(bv, &flags);
|
|
|
|
memset(buf, 0, bv->bv_len);
|
|
|
|
flush_dcache_page(bv->bv_page);
|
|
|
|
bvec_kunmap_irq(buf, &flags);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2018-01-20 17:30:10 +08:00
|
|
|
static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2018-01-20 17:30:10 +08:00
|
|
|
struct ceph_bio_iter it = *bio_pos;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-20 17:30:10 +08:00
|
|
|
ceph_bio_iter_advance(&it, off);
|
|
|
|
ceph_bio_iter_advance_step(&it, bytes, ({
|
|
|
|
zero_bvec(&bv);
|
|
|
|
}));
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2018-01-20 17:30:11 +08:00
|
|
|
struct ceph_bvec_iter it = *bvec_pos;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
ceph_bvec_iter_advance(&it, off);
|
|
|
|
ceph_bvec_iter_advance_step(&it, bytes, ({
|
|
|
|
zero_bvec(&bv);
|
|
|
|
}));
|
rbd: simplify rbd_rq_fn()
When processing a request, rbd_rq_fn() makes clones of the bio's in
the request's bio chain and submits the results to osd's to be
satisfied. If a request bio straddles the boundary between objects
backing the rbd image, it must be represented by two cloned bio's,
one for the first part (at the end of one object) and one for the
second (at the beginning of the next object).
This has been handled by a function bio_chain_clone(), which
includes an interface only a mother could love, and which has
been found to have other problems.
This patch defines two new fairly generic bio functions (one which
replaces bio_chain_clone()) to help out the situation, and then
revises rbd_rq_fn() to make use of them.
First, bio_clone_range() clones a portion of a single bio, starting
at a given offset within the bio and including only as many bytes
as requested. As a convenience, a request to clone the entire bio
is passed directly to bio_clone().
Second, bio_chain_clone_range() performs a similar function,
producing a chain of cloned bio's covering a sub-range of the
source chain. No bio_pair structures are used, and if successful
the result will represent exactly the specified range.
Using bio_chain_clone_range() makes bio_rq_fn() a little easier
to understand, because it avoids the need to pass very much
state information between consecutive calls. By avoiding the need
to track a bio_pair structure, it also eliminates the problem
described here: http://tracker.newdream.net/issues/2933
Note that a block request (and therefore the complete length of
a bio chain processed in rbd_rq_fn()) is an unsigned int, while
the result of rbd_segment_length() is u64. This change makes
this range trunctation explicit, and trips a bug if the the
segment boundary is too far off.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2012-10-21 11:17:27 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2018-01-29 21:04:08 +08:00
|
|
|
* Zero a range in @obj_req data buffer defined by a bio (list) or
|
2018-02-07 02:26:35 +08:00
|
|
|
* (private) bio_vec array.
|
rbd: simplify rbd_rq_fn()
When processing a request, rbd_rq_fn() makes clones of the bio's in
the request's bio chain and submits the results to osd's to be
satisfied. If a request bio straddles the boundary between objects
backing the rbd image, it must be represented by two cloned bio's,
one for the first part (at the end of one object) and one for the
second (at the beginning of the next object).
This has been handled by a function bio_chain_clone(), which
includes an interface only a mother could love, and which has
been found to have other problems.
This patch defines two new fairly generic bio functions (one which
replaces bio_chain_clone()) to help out the situation, and then
revises rbd_rq_fn() to make use of them.
First, bio_clone_range() clones a portion of a single bio, starting
at a given offset within the bio and including only as many bytes
as requested. As a convenience, a request to clone the entire bio
is passed directly to bio_clone().
Second, bio_chain_clone_range() performs a similar function,
producing a chain of cloned bio's covering a sub-range of the
source chain. No bio_pair structures are used, and if successful
the result will represent exactly the specified range.
Using bio_chain_clone_range() makes bio_rq_fn() a little easier
to understand, because it avoids the need to pass very much
state information between consecutive calls. By avoiding the need
to track a bio_pair structure, it also eliminates the problem
described here: http://tracker.newdream.net/issues/2933
Note that a block request (and therefore the complete length of
a bio chain processed in rbd_rq_fn()) is an unsigned int, while
the result of rbd_segment_length() is u64. This change makes
this range trunctation explicit, and trips a bug if the the
segment boundary is too far off.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2012-10-21 11:17:27 +08:00
|
|
|
*
|
2018-01-29 21:04:08 +08:00
|
|
|
* @off is relative to the start of the data buffer.
|
2013-02-12 02:33:24 +08:00
|
|
|
*/
|
2018-01-29 21:04:08 +08:00
|
|
|
static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off,
|
|
|
|
u32 bytes)
|
2013-02-12 02:33:24 +08:00
|
|
|
{
|
2019-05-11 22:21:49 +08:00
|
|
|
dout("%s %p data buf %u~%u\n", __func__, obj_req, off, bytes);
|
|
|
|
|
2018-02-01 18:50:47 +08:00
|
|
|
switch (obj_req->img_request->data_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_REQUEST_BIO:
|
|
|
|
zero_bios(&obj_req->bio_pos, off, bytes);
|
|
|
|
break;
|
|
|
|
case OBJ_REQUEST_BVECS:
|
2018-02-07 02:26:35 +08:00
|
|
|
case OBJ_REQUEST_OWN_BVECS:
|
2018-01-29 21:04:08 +08:00
|
|
|
zero_bvecs(&obj_req->bvec_pos, off, bytes);
|
|
|
|
break;
|
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2013-02-12 02:33:24 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
static void rbd_obj_request_destroy(struct kref *kref);
|
|
|
|
static void rbd_obj_request_put(struct rbd_obj_request *obj_request)
|
|
|
|
{
|
|
|
|
rbd_assert(obj_request != NULL);
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: obj %p (was %d)\n", __func__, obj_request,
|
2016-11-15 00:29:48 +08:00
|
|
|
kref_read(&obj_request->kref));
|
2012-11-22 14:00:08 +08:00
|
|
|
kref_put(&obj_request->kref, rbd_obj_request_destroy);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_img_request_destroy(struct kref *kref);
|
|
|
|
static void rbd_img_request_put(struct rbd_img_request *img_request)
|
|
|
|
{
|
|
|
|
rbd_assert(img_request != NULL);
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: img %p (was %d)\n", __func__, img_request,
|
2016-11-15 00:29:48 +08:00
|
|
|
kref_read(&img_request->kref));
|
2018-02-07 02:26:35 +08:00
|
|
|
kref_put(&img_request->kref, rbd_img_request_destroy);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request,
|
|
|
|
struct rbd_obj_request *obj_request)
|
|
|
|
{
|
2013-01-26 07:08:55 +08:00
|
|
|
rbd_assert(obj_request->img_request == NULL);
|
|
|
|
|
2013-04-16 03:50:37 +08:00
|
|
|
/* Image request now owns object's original reference */
|
2012-11-22 14:00:08 +08:00
|
|
|
obj_request->img_request = img_request;
|
2018-02-01 18:50:47 +08:00
|
|
|
dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request,
|
|
|
|
struct rbd_obj_request *obj_request)
|
|
|
|
{
|
2018-02-01 18:50:47 +08:00
|
|
|
dout("%s: img %p obj %p\n", __func__, img_request, obj_request);
|
2018-02-02 22:23:22 +08:00
|
|
|
list_del(&obj_request->ex.oe_item);
|
2012-11-22 14:00:08 +08:00
|
|
|
rbd_assert(obj_request->img_request == img_request);
|
|
|
|
rbd_obj_request_put(obj_request);
|
|
|
|
}
|
|
|
|
|
2016-09-13 00:59:42 +08:00
|
|
|
static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req =
|
|
|
|
list_last_entry(&obj_request->osd_reqs, struct ceph_osd_request,
|
|
|
|
r_private_item);
|
2016-09-13 00:59:42 +08:00
|
|
|
|
2017-01-26 01:16:23 +08:00
|
|
|
dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__,
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_request, obj_request->ex.oe_objno, obj_request->ex.oe_off,
|
|
|
|
obj_request->ex.oe_len, osd_req);
|
2016-09-13 00:59:42 +08:00
|
|
|
ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2013-02-08 23:55:49 +08:00
|
|
|
/*
|
|
|
|
* The default/initial value for all image request flags is 0. Each
|
|
|
|
* is conditionally set to 1 at image request initialization time
|
|
|
|
* and currently never change thereafter.
|
|
|
|
*/
|
2013-01-25 06:13:36 +08:00
|
|
|
static void img_request_layered_set(struct rbd_img_request *img_request)
|
|
|
|
{
|
|
|
|
set_bit(IMG_REQ_LAYERED, &img_request->flags);
|
|
|
|
smp_mb();
|
|
|
|
}
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
static void img_request_layered_clear(struct rbd_img_request *img_request)
|
|
|
|
{
|
|
|
|
clear_bit(IMG_REQ_LAYERED, &img_request->flags);
|
|
|
|
smp_mb();
|
|
|
|
}
|
|
|
|
|
2013-01-25 06:13:36 +08:00
|
|
|
static bool img_request_layered_test(struct rbd_img_request *img_request)
|
|
|
|
{
|
|
|
|
smp_mb();
|
|
|
|
return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0;
|
|
|
|
}
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req)
|
2013-03-27 22:16:30 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-02 22:23:22 +08:00
|
|
|
return !obj_req->ex.oe_off &&
|
|
|
|
obj_req->ex.oe_len == rbd_dev->layout.object_size;
|
2013-03-27 22:16:30 +08:00
|
|
|
}
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-02 22:23:22 +08:00
|
|
|
return obj_req->ex.oe_off + obj_req->ex.oe_len ==
|
2018-01-29 21:04:08 +08:00
|
|
|
rbd_dev->layout.object_size;
|
2016-09-26 21:43:52 +08:00
|
|
|
}
|
|
|
|
|
2019-02-25 19:37:50 +08:00
|
|
|
/*
|
|
|
|
* Must be called after rbd_obj_calc_img_extents().
|
|
|
|
*/
|
|
|
|
static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
if (!obj_req->num_img_extents ||
|
2019-02-28 22:51:39 +08:00
|
|
|
(rbd_obj_is_entire(obj_req) &&
|
|
|
|
!obj_req->img_request->snapc->num_snaps))
|
2019-02-25 19:37:50 +08:00
|
|
|
return false;
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2018-02-07 02:26:33 +08:00
|
|
|
return ceph_file_extents_bytes(obj_req->img_extents,
|
|
|
|
obj_req->num_img_extents);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
static bool rbd_img_is_write(struct rbd_img_request *img_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2018-01-31 00:52:10 +08:00
|
|
|
switch (img_req->op_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_OP_READ:
|
|
|
|
return false;
|
|
|
|
case OBJ_OP_WRITE:
|
|
|
|
case OBJ_OP_DISCARD:
|
2019-01-29 19:46:25 +08:00
|
|
|
case OBJ_OP_ZEROOUT:
|
2018-01-29 21:04:08 +08:00
|
|
|
return true;
|
|
|
|
default:
|
2018-04-04 20:53:39 +08:00
|
|
|
BUG();
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2014-04-01 22:22:16 +08:00
|
|
|
}
|
|
|
|
|
2016-04-28 22:07:24 +08:00
|
|
|
static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_obj_request *obj_req = osd_req->r_priv;
|
2019-05-11 22:21:49 +08:00
|
|
|
int result;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req,
|
|
|
|
osd_req->r_result, obj_req);
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-11 22:21:49 +08:00
|
|
|
/*
|
|
|
|
* Writes aren't allowed to return a data payload. In some
|
|
|
|
* guarded write cases (e.g. stat + zero on an empty object)
|
|
|
|
* a stat response makes it through, but we don't care.
|
|
|
|
*/
|
|
|
|
if (osd_req->r_result > 0 && rbd_img_is_write(obj_req->img_request))
|
|
|
|
result = 0;
|
2018-01-29 21:04:08 +08:00
|
|
|
else
|
2019-05-11 22:21:49 +08:00
|
|
|
result = osd_req->r_result;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-11 22:21:49 +08:00
|
|
|
rbd_obj_handle_request(obj_req, result);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static void rbd_osd_format_read(struct ceph_osd_request *osd_req)
|
2013-04-04 10:32:51 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct rbd_obj_request *obj_request = osd_req->r_priv;
|
2013-04-04 10:32:51 +08:00
|
|
|
|
2018-01-31 00:52:10 +08:00
|
|
|
osd_req->r_flags = CEPH_OSD_FLAG_READ;
|
2016-09-15 23:56:39 +08:00
|
|
|
osd_req->r_snapid = obj_request->img_request->snap_id;
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static void rbd_osd_format_write(struct ceph_osd_request *osd_req)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct rbd_obj_request *obj_request = osd_req->r_priv;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-31 00:52:10 +08:00
|
|
|
osd_req->r_flags = CEPH_OSD_FLAG_WRITE;
|
2018-07-14 04:18:37 +08:00
|
|
|
ktime_get_real_ts64(&osd_req->r_mtime);
|
2018-02-02 22:23:22 +08:00
|
|
|
osd_req->r_data_offset = obj_request->ex.oe_off;
|
2013-04-04 10:32:51 +08:00
|
|
|
}
|
|
|
|
|
2017-01-26 01:16:23 +08:00
|
|
|
static struct ceph_osd_request *
|
2019-05-27 17:41:36 +08:00
|
|
|
__rbd_obj_add_osd_request(struct rbd_obj_request *obj_req,
|
|
|
|
struct ceph_snap_context *snapc, int num_ops)
|
2017-01-26 01:16:23 +08:00
|
|
|
{
|
2019-02-25 18:42:26 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
2017-01-26 01:16:23 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
struct ceph_osd_request *req;
|
2017-01-26 01:16:23 +08:00
|
|
|
const char *name_format = rbd_dev->image_format == 1 ?
|
|
|
|
RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT;
|
2019-05-27 17:41:36 +08:00
|
|
|
int ret;
|
2017-01-26 01:16:23 +08:00
|
|
|
|
2019-02-25 18:42:26 +08:00
|
|
|
req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO);
|
2017-01-26 01:16:23 +08:00
|
|
|
if (!req)
|
2019-05-27 17:41:36 +08:00
|
|
|
return ERR_PTR(-ENOMEM);
|
2017-01-26 01:16:23 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
list_add_tail(&req->r_private_item, &obj_req->osd_reqs);
|
2017-01-26 01:16:23 +08:00
|
|
|
req->r_callback = rbd_osd_req_callback;
|
2018-01-31 00:52:10 +08:00
|
|
|
req->r_priv = obj_req;
|
2017-01-26 01:16:23 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
/*
|
|
|
|
* Data objects may be stored in a separate pool, but always in
|
|
|
|
* the same namespace in that pool as the header in its pool.
|
|
|
|
*/
|
|
|
|
ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc);
|
2017-01-26 01:16:23 +08:00
|
|
|
req->r_base_oloc.pool = rbd_dev->layout.pool_id;
|
2018-07-03 21:28:43 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format,
|
|
|
|
rbd_dev->header.object_prefix,
|
|
|
|
obj_req->ex.oe_objno);
|
|
|
|
if (ret)
|
|
|
|
return ERR_PTR(ret);
|
2017-01-26 01:16:23 +08:00
|
|
|
|
|
|
|
return req;
|
|
|
|
}
|
|
|
|
|
2019-02-25 18:42:26 +08:00
|
|
|
static struct ceph_osd_request *
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc,
|
|
|
|
num_ops);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2018-02-01 18:50:47 +08:00
|
|
|
static struct rbd_obj_request *rbd_obj_request_create(void)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_request;
|
|
|
|
|
2015-06-24 22:24:33 +08:00
|
|
|
obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO);
|
2017-01-26 01:16:23 +08:00
|
|
|
if (!obj_request)
|
2013-05-02 01:43:03 +08:00
|
|
|
return NULL;
|
|
|
|
|
2018-02-02 22:23:22 +08:00
|
|
|
ceph_object_extent_init(&obj_request->ex);
|
2019-05-27 17:41:36 +08:00
|
|
|
INIT_LIST_HEAD(&obj_request->osd_reqs);
|
2019-05-15 03:06:07 +08:00
|
|
|
mutex_init(&obj_request->state_mutex);
|
2012-11-22 14:00:08 +08:00
|
|
|
kref_init(&obj_request->kref);
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
dout("%s %p\n", __func__, obj_request);
|
2012-11-22 14:00:08 +08:00
|
|
|
return obj_request;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_obj_request_destroy(struct kref *kref)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_request;
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2018-01-20 17:30:11 +08:00
|
|
|
u32 i;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
|
|
|
obj_request = container_of(kref, struct rbd_obj_request, kref);
|
|
|
|
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: obj %p\n", __func__, obj_request);
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
while (!list_empty(&obj_request->osd_reqs)) {
|
|
|
|
osd_req = list_first_entry(&obj_request->osd_reqs,
|
|
|
|
struct ceph_osd_request, r_private_item);
|
|
|
|
list_del_init(&osd_req->r_private_item);
|
|
|
|
ceph_osdc_put_request(osd_req);
|
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-01 18:50:47 +08:00
|
|
|
switch (obj_request->img_request->data_type) {
|
2013-01-19 02:31:10 +08:00
|
|
|
case OBJ_REQUEST_NODATA:
|
2012-11-22 14:00:08 +08:00
|
|
|
case OBJ_REQUEST_BIO:
|
2018-01-20 17:30:11 +08:00
|
|
|
case OBJ_REQUEST_BVECS:
|
2018-01-20 17:30:10 +08:00
|
|
|
break; /* Nothing to do */
|
2018-02-07 02:26:35 +08:00
|
|
|
case OBJ_REQUEST_OWN_BVECS:
|
|
|
|
kfree(obj_request->bvec_pos.bvecs);
|
2013-01-18 02:25:27 +08:00
|
|
|
break;
|
2018-01-20 17:30:11 +08:00
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
kfree(obj_request->img_extents);
|
2018-01-20 17:30:11 +08:00
|
|
|
if (obj_request->copyup_bvecs) {
|
|
|
|
for (i = 0; i < obj_request->copyup_bvec_count; i++) {
|
|
|
|
if (obj_request->copyup_bvecs[i].bv_page)
|
|
|
|
__free_page(obj_request->copyup_bvecs[i].bv_page);
|
|
|
|
}
|
|
|
|
kfree(obj_request->copyup_bvecs);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
kmem_cache_free(rbd_obj_request_cache, obj_request);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
/* It's OK to call this for a device with no parent */
|
|
|
|
|
|
|
|
static void rbd_spec_put(struct rbd_spec *spec);
|
|
|
|
static void rbd_dev_unparent(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
rbd_dev_remove_parent(rbd_dev);
|
|
|
|
rbd_spec_put(rbd_dev->parent_spec);
|
|
|
|
rbd_dev->parent_spec = NULL;
|
|
|
|
rbd_dev->parent_overlap = 0;
|
|
|
|
}
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
/*
|
|
|
|
* Parent image reference counting is used to determine when an
|
|
|
|
* image's parent fields can be safely torn down--after there are no
|
|
|
|
* more in-flight requests to the parent image. When the last
|
|
|
|
* reference is dropped, cleaning them up is safe.
|
|
|
|
*/
|
|
|
|
static void rbd_dev_parent_put(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
int counter;
|
|
|
|
|
|
|
|
if (!rbd_dev->parent_spec)
|
|
|
|
return;
|
|
|
|
|
|
|
|
counter = atomic_dec_return_safe(&rbd_dev->parent_ref);
|
|
|
|
if (counter > 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Last reference; clean up parent data structures */
|
|
|
|
|
|
|
|
if (!counter)
|
|
|
|
rbd_dev_unparent(rbd_dev);
|
|
|
|
else
|
2014-07-11 16:11:20 +08:00
|
|
|
rbd_warn(rbd_dev, "parent reference underflow");
|
2013-05-09 11:50:04 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If an image has a non-zero parent overlap, get a reference to its
|
|
|
|
* parent.
|
|
|
|
*
|
|
|
|
* Returns true if the rbd device has a parent with a non-zero
|
|
|
|
* overlap and a reference for it was successfully taken, or
|
|
|
|
* false otherwise.
|
|
|
|
*/
|
|
|
|
static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2015-01-19 23:13:43 +08:00
|
|
|
int counter = 0;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
|
|
|
if (!rbd_dev->parent_spec)
|
|
|
|
return false;
|
|
|
|
|
2015-01-19 23:13:43 +08:00
|
|
|
down_read(&rbd_dev->header_rwsem);
|
|
|
|
if (rbd_dev->parent_overlap)
|
|
|
|
counter = atomic_inc_return_safe(&rbd_dev->parent_ref);
|
|
|
|
up_read(&rbd_dev->header_rwsem);
|
2013-05-09 11:50:04 +08:00
|
|
|
|
|
|
|
if (counter < 0)
|
2014-07-11 16:11:20 +08:00
|
|
|
rbd_warn(rbd_dev, "parent reference overflow");
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2015-01-19 23:13:43 +08:00
|
|
|
return counter > 0;
|
2013-05-09 11:50:04 +08:00
|
|
|
}
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
/*
|
|
|
|
* Caller is responsible for filling in the list of object requests
|
|
|
|
* that comprises the image request, and the Linux request pointer
|
|
|
|
* (if there is one).
|
|
|
|
*/
|
2013-02-20 02:25:56 +08:00
|
|
|
static struct rbd_img_request *rbd_img_request_create(
|
|
|
|
struct rbd_device *rbd_dev,
|
2014-03-13 11:21:35 +08:00
|
|
|
enum obj_operation_type op_type,
|
2014-04-09 02:12:11 +08:00
|
|
|
struct ceph_snap_context *snapc)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
|
|
|
struct rbd_img_request *img_request;
|
|
|
|
|
2018-01-22 23:03:06 +08:00
|
|
|
img_request = kmem_cache_zalloc(rbd_img_request_cache, GFP_NOIO);
|
2012-11-22 14:00:08 +08:00
|
|
|
if (!img_request)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
img_request->rbd_dev = rbd_dev;
|
2018-01-31 00:52:10 +08:00
|
|
|
img_request->op_type = op_type;
|
|
|
|
if (!rbd_img_is_write(img_request))
|
2012-11-22 14:00:08 +08:00
|
|
|
img_request->snap_id = rbd_dev->spec->snap_id;
|
2018-01-31 00:52:10 +08:00
|
|
|
else
|
|
|
|
img_request->snapc = snapc;
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
if (rbd_dev_parent_get(rbd_dev))
|
2013-01-25 06:13:36 +08:00
|
|
|
img_request_layered_set(img_request);
|
2018-01-22 23:03:06 +08:00
|
|
|
|
2018-02-02 22:23:22 +08:00
|
|
|
INIT_LIST_HEAD(&img_request->object_extents);
|
2019-05-16 21:06:56 +08:00
|
|
|
mutex_init(&img_request->state_mutex);
|
2012-11-22 14:00:08 +08:00
|
|
|
kref_init(&img_request->kref);
|
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
dout("%s: rbd_dev %p %s -> img %p\n", __func__, rbd_dev,
|
|
|
|
obj_op_name(op_type), img_request);
|
2012-11-22 14:00:08 +08:00
|
|
|
return img_request;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_img_request_destroy(struct kref *kref)
|
|
|
|
{
|
|
|
|
struct rbd_img_request *img_request;
|
|
|
|
struct rbd_obj_request *obj_request;
|
|
|
|
struct rbd_obj_request *next_obj_request;
|
|
|
|
|
|
|
|
img_request = container_of(kref, struct rbd_img_request, kref);
|
|
|
|
|
2013-02-21 07:32:08 +08:00
|
|
|
dout("%s: img %p\n", __func__, img_request);
|
|
|
|
|
2012-11-22 14:00:08 +08:00
|
|
|
for_each_obj_request_safe(img_request, obj_request, next_obj_request)
|
|
|
|
rbd_img_obj_request_del(img_request, obj_request);
|
|
|
|
|
2013-05-09 11:50:04 +08:00
|
|
|
if (img_request_layered_test(img_request)) {
|
|
|
|
img_request_layered_clear(img_request);
|
|
|
|
rbd_dev_parent_put(img_request->rbd_dev);
|
|
|
|
}
|
|
|
|
|
2018-01-31 00:52:10 +08:00
|
|
|
if (rbd_img_is_write(img_request))
|
2013-04-30 13:44:32 +08:00
|
|
|
ceph_put_snap_context(img_request->snapc);
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
kmem_cache_free(rbd_img_request_cache, img_request);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
static void prune_extents(struct ceph_file_extent *img_extents,
|
|
|
|
u32 *num_img_extents, u64 overlap)
|
2013-05-09 11:50:04 +08:00
|
|
|
{
|
2018-02-07 02:26:33 +08:00
|
|
|
u32 cnt = *num_img_extents;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
/* drop extents completely beyond the overlap */
|
|
|
|
while (cnt && img_extents[cnt - 1].fe_off >= overlap)
|
|
|
|
cnt--;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
if (cnt) {
|
|
|
|
struct ceph_file_extent *ex = &img_extents[cnt - 1];
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
/* trim final overlapping extent */
|
|
|
|
if (ex->fe_off + ex->fe_len > overlap)
|
|
|
|
ex->fe_len = overlap - ex->fe_off;
|
|
|
|
}
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
*num_img_extents = cnt;
|
2013-05-09 11:50:04 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
/*
|
|
|
|
* Determine the byte range(s) covered by either just the object extent
|
|
|
|
* or the entire object in the parent image.
|
|
|
|
*/
|
|
|
|
static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req,
|
|
|
|
bool entire)
|
2013-05-09 11:50:04 +08:00
|
|
|
{
|
2018-02-07 02:26:33 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
|
|
|
int ret;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
if (!rbd_dev->parent_overlap)
|
|
|
|
return 0;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno,
|
|
|
|
entire ? 0 : obj_req->ex.oe_off,
|
|
|
|
entire ? rbd_dev->layout.object_size :
|
|
|
|
obj_req->ex.oe_len,
|
|
|
|
&obj_req->img_extents,
|
|
|
|
&obj_req->num_img_extents);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2013-05-09 11:50:04 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
|
|
|
|
rbd_dev->parent_overlap);
|
|
|
|
return 0;
|
2013-05-09 11:50:04 +08:00
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static void rbd_osd_setup_data(struct ceph_osd_request *osd_req, int which)
|
2013-02-08 23:55:49 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct rbd_obj_request *obj_req = osd_req->r_priv;
|
|
|
|
|
2018-02-01 18:50:47 +08:00
|
|
|
switch (obj_req->img_request->data_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_REQUEST_BIO:
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_osd_data_bio(osd_req, which,
|
2018-01-29 21:04:08 +08:00
|
|
|
&obj_req->bio_pos,
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_req->ex.oe_len);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
|
|
|
case OBJ_REQUEST_BVECS:
|
2018-02-07 02:26:35 +08:00
|
|
|
case OBJ_REQUEST_OWN_BVECS:
|
2018-01-29 21:04:08 +08:00
|
|
|
rbd_assert(obj_req->bvec_pos.iter.bi_size ==
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_req->ex.oe_len);
|
2018-02-07 02:26:35 +08:00
|
|
|
rbd_assert(obj_req->bvec_idx == obj_req->bvec_count);
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_osd_data_bvec_pos(osd_req, which,
|
2018-01-29 21:04:08 +08:00
|
|
|
&obj_req->bvec_pos);
|
|
|
|
break;
|
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2013-02-08 23:55:49 +08:00
|
|
|
}
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2013-02-08 23:55:49 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
static int rbd_obj_setup_read(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
|
|
|
|
|
|
|
osd_req = __rbd_obj_add_osd_request(obj_req, NULL, 1);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2017-06-03 15:38:04 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_init(osd_req, 0, CEPH_OSD_OP_READ,
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_setup_data(osd_req, 0);
|
2015-01-14 00:20:04 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_format_read(osd_req);
|
2019-05-15 03:06:07 +08:00
|
|
|
obj_req->read_state = RBD_OBJ_READ_START;
|
2018-01-29 21:04:08 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static int rbd_osd_setup_stat(struct ceph_osd_request *osd_req, int which)
|
2018-01-29 21:04:08 +08:00
|
|
|
{
|
|
|
|
struct page **pages;
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
|
|
|
* The response data for a STAT call consists of:
|
|
|
|
* le64 length;
|
|
|
|
* struct {
|
|
|
|
* le32 tv_sec;
|
|
|
|
* le32 tv_nsec;
|
|
|
|
* } mtime;
|
|
|
|
*/
|
|
|
|
pages = ceph_alloc_page_vector(1, GFP_NOIO);
|
|
|
|
if (IS_ERR(pages))
|
|
|
|
return PTR_ERR(pages);
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_init(osd_req, which, CEPH_OSD_OP_STAT, 0);
|
|
|
|
osd_req_op_raw_data_in_pages(osd_req, which, pages,
|
2018-01-29 21:04:08 +08:00
|
|
|
8 + sizeof(struct ceph_timespec),
|
|
|
|
0, false, true);
|
|
|
|
return 0;
|
2013-02-08 23:55:49 +08:00
|
|
|
}
|
|
|
|
|
2019-02-25 19:37:50 +08:00
|
|
|
static int count_write_ops(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
return 2; /* setallochint + write/writefull */
|
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
|
|
|
|
int which)
|
2013-04-05 14:27:12 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct rbd_obj_request *obj_req = osd_req->r_priv;
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
|
|
|
u16 opcode;
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_alloc_hint_init(osd_req, which++,
|
2018-01-29 21:04:08 +08:00
|
|
|
rbd_dev->layout.object_size,
|
|
|
|
rbd_dev->layout.object_size);
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
if (rbd_obj_is_entire(obj_req))
|
|
|
|
opcode = CEPH_OSD_OP_WRITEFULL;
|
|
|
|
else
|
|
|
|
opcode = CEPH_OSD_OP_WRITE;
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_init(osd_req, which, opcode,
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0);
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_setup_data(osd_req, which);
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_format_write(osd_req);
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
static int rbd_obj_setup_write(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2018-01-29 21:04:08 +08:00
|
|
|
unsigned int num_osd_ops, which = 0;
|
|
|
|
int ret;
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
/* reverse map the entire object onto the parent */
|
|
|
|
ret = rbd_obj_calc_img_extents(obj_req, true);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
if (rbd_obj_copyup_enabled(obj_req))
|
|
|
|
obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
|
|
|
|
|
|
|
|
num_osd_ops = count_write_ops(obj_req);
|
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)
|
|
|
|
num_osd_ops++; /* stat */
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = rbd_obj_add_osd_request(obj_req, num_osd_ops);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2013-04-05 14:27:12 +08:00
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = rbd_osd_setup_stat(osd_req, which++);
|
2018-01-29 21:04:08 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_START;
|
2019-05-27 17:41:36 +08:00
|
|
|
__rbd_osd_setup_write_ops(osd_req, which);
|
2018-01-29 21:04:08 +08:00
|
|
|
return 0;
|
2013-04-05 14:27:12 +08:00
|
|
|
}
|
|
|
|
|
2019-01-29 19:46:25 +08:00
|
|
|
static u16 truncate_or_zero_opcode(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
return rbd_obj_is_tail(obj_req) ? CEPH_OSD_OP_TRUNCATE :
|
|
|
|
CEPH_OSD_OP_ZERO;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_obj_setup_discard(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
2019-01-30 22:14:48 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2019-01-30 22:14:48 +08:00
|
|
|
u64 off = obj_req->ex.oe_off;
|
|
|
|
u64 next_off = obj_req->ex.oe_off + obj_req->ex.oe_len;
|
2019-01-29 19:46:25 +08:00
|
|
|
int ret;
|
|
|
|
|
2019-01-30 22:14:48 +08:00
|
|
|
/*
|
|
|
|
* Align the range to alloc_size boundary and punt on discards
|
|
|
|
* that are too small to free up any space.
|
|
|
|
*
|
|
|
|
* alloc_size == object_size && is_tail() is a special case for
|
|
|
|
* filestore with filestore_punch_hole = false, needed to allow
|
|
|
|
* truncate (in addition to delete).
|
|
|
|
*/
|
|
|
|
if (rbd_dev->opts->alloc_size != rbd_dev->layout.object_size ||
|
|
|
|
!rbd_obj_is_tail(obj_req)) {
|
|
|
|
off = round_up(off, rbd_dev->opts->alloc_size);
|
|
|
|
next_off = round_down(next_off, rbd_dev->opts->alloc_size);
|
|
|
|
if (off >= next_off)
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2019-01-29 19:46:25 +08:00
|
|
|
/* reverse map the entire object onto the parent */
|
|
|
|
ret = rbd_obj_calc_img_extents(obj_req, true);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents)
|
|
|
|
obj_req->flags |= RBD_OBJ_FLAG_DELETION;
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = rbd_obj_add_osd_request(obj_req, 1);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2019-01-29 19:46:25 +08:00
|
|
|
|
|
|
|
if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) {
|
2019-05-15 02:45:38 +08:00
|
|
|
rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_init(osd_req, 0, CEPH_OSD_OP_DELETE, 0);
|
2019-01-29 19:46:25 +08:00
|
|
|
} else {
|
2019-01-30 22:14:48 +08:00
|
|
|
dout("%s %p %llu~%llu -> %llu~%llu\n", __func__,
|
|
|
|
obj_req, obj_req->ex.oe_off, obj_req->ex.oe_len,
|
|
|
|
off, next_off - off);
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_init(osd_req, 0,
|
2019-01-29 19:46:25 +08:00
|
|
|
truncate_or_zero_opcode(obj_req),
|
2019-01-30 22:14:48 +08:00
|
|
|
off, next_off - off, 0, 0);
|
2019-01-29 19:46:25 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_START;
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_format_write(osd_req);
|
2019-01-29 19:46:25 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-02-25 19:37:50 +08:00
|
|
|
static int count_zeroout_ops(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
int num_osd_ops;
|
|
|
|
|
2019-02-28 22:51:39 +08:00
|
|
|
if (rbd_obj_is_entire(obj_req) && obj_req->num_img_extents &&
|
|
|
|
!rbd_obj_copyup_enabled(obj_req))
|
2019-02-25 19:37:50 +08:00
|
|
|
num_osd_ops = 2; /* create + truncate */
|
|
|
|
else
|
|
|
|
num_osd_ops = 1; /* delete/truncate/zero */
|
|
|
|
|
|
|
|
return num_osd_ops;
|
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
static void __rbd_osd_setup_zeroout_ops(struct ceph_osd_request *osd_req,
|
|
|
|
int which)
|
2018-01-29 21:04:08 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct rbd_obj_request *obj_req = osd_req->r_priv;
|
2014-04-05 08:32:15 +08:00
|
|
|
u16 opcode;
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
if (rbd_obj_is_entire(obj_req)) {
|
2018-02-07 02:26:33 +08:00
|
|
|
if (obj_req->num_img_extents) {
|
2019-05-15 02:45:38 +08:00
|
|
|
if (!(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED))
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_init(osd_req, which++,
|
2019-02-28 22:51:39 +08:00
|
|
|
CEPH_OSD_OP_CREATE, 0);
|
2014-04-05 08:32:15 +08:00
|
|
|
opcode = CEPH_OSD_OP_TRUNCATE;
|
|
|
|
} else {
|
2019-05-15 02:45:38 +08:00
|
|
|
rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION);
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_init(osd_req, which++,
|
2018-01-29 21:04:08 +08:00
|
|
|
CEPH_OSD_OP_DELETE, 0);
|
|
|
|
opcode = 0;
|
2014-04-05 08:32:15 +08:00
|
|
|
}
|
|
|
|
} else {
|
2019-01-29 19:46:25 +08:00
|
|
|
opcode = truncate_or_zero_opcode(obj_req);
|
2014-04-05 08:32:15 +08:00
|
|
|
}
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
if (opcode)
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_extent_init(osd_req, which, opcode,
|
2018-02-02 22:23:22 +08:00
|
|
|
obj_req->ex.oe_off, obj_req->ex.oe_len,
|
2018-01-29 21:04:08 +08:00
|
|
|
0, 0);
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_format_write(osd_req);
|
2014-04-05 08:32:15 +08:00
|
|
|
}
|
|
|
|
|
2019-01-29 19:46:25 +08:00
|
|
|
static int rbd_obj_setup_zeroout(struct rbd_obj_request *obj_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2018-01-29 21:04:08 +08:00
|
|
|
unsigned int num_osd_ops, which = 0;
|
|
|
|
int ret;
|
2013-02-21 07:32:08 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
/* reverse map the entire object onto the parent */
|
|
|
|
ret = rbd_obj_calc_img_extents(obj_req, true);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
if (rbd_obj_copyup_enabled(obj_req))
|
|
|
|
obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
|
|
|
|
if (!obj_req->num_img_extents) {
|
|
|
|
if (rbd_obj_is_entire(obj_req))
|
|
|
|
obj_req->flags |= RBD_OBJ_FLAG_DELETION;
|
|
|
|
}
|
|
|
|
|
|
|
|
num_osd_ops = count_zeroout_ops(obj_req);
|
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)
|
|
|
|
num_osd_ops++; /* stat */
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = rbd_obj_add_osd_request(obj_req, num_osd_ops);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = rbd_osd_setup_stat(osd_req, which++);
|
2018-01-29 21:04:08 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
2014-04-05 08:32:15 +08:00
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_START;
|
2019-05-27 17:41:36 +08:00
|
|
|
__rbd_osd_setup_zeroout_ops(osd_req, which);
|
2018-01-29 21:04:08 +08:00
|
|
|
return 0;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
|
|
|
* For each object request in @img_req, allocate an OSD request, add
|
|
|
|
* individual OSD ops and prepare them for submission. The number of
|
|
|
|
* OSD ops depends on op_type and the overlap point (if any).
|
|
|
|
*/
|
|
|
|
static int __rbd_img_fill_request(struct rbd_img_request *img_req)
|
|
|
|
{
|
2019-01-30 22:14:48 +08:00
|
|
|
struct rbd_obj_request *obj_req, *next_obj_req;
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2018-01-29 21:04:08 +08:00
|
|
|
int ret;
|
2013-04-04 10:32:51 +08:00
|
|
|
|
2019-01-30 22:14:48 +08:00
|
|
|
for_each_obj_request_safe(img_req, obj_req, next_obj_req) {
|
2018-01-31 00:52:10 +08:00
|
|
|
switch (img_req->op_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_OP_READ:
|
|
|
|
ret = rbd_obj_setup_read(obj_req);
|
|
|
|
break;
|
|
|
|
case OBJ_OP_WRITE:
|
|
|
|
ret = rbd_obj_setup_write(obj_req);
|
|
|
|
break;
|
|
|
|
case OBJ_OP_DISCARD:
|
|
|
|
ret = rbd_obj_setup_discard(obj_req);
|
|
|
|
break;
|
2019-01-29 19:46:25 +08:00
|
|
|
case OBJ_OP_ZEROOUT:
|
|
|
|
ret = rbd_obj_setup_zeroout(obj_req);
|
|
|
|
break;
|
2018-01-29 21:04:08 +08:00
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2019-01-30 22:14:48 +08:00
|
|
|
if (ret < 0)
|
2018-01-29 21:04:08 +08:00
|
|
|
return ret;
|
2019-01-30 22:14:48 +08:00
|
|
|
if (ret > 0) {
|
|
|
|
rbd_img_obj_request_del(img_req, obj_req);
|
|
|
|
continue;
|
|
|
|
}
|
2018-10-15 22:11:37 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = list_last_entry(&obj_req->osd_reqs,
|
|
|
|
struct ceph_osd_request,
|
|
|
|
r_private_item);
|
|
|
|
ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
|
2018-10-15 22:11:37 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
img_req->state = RBD_IMG_START;
|
2012-11-22 14:00:08 +08:00
|
|
|
return 0;
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
union rbd_img_fill_iter {
|
|
|
|
struct ceph_bio_iter bio_iter;
|
|
|
|
struct ceph_bvec_iter bvec_iter;
|
|
|
|
};
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
struct rbd_img_fill_ctx {
|
|
|
|
enum obj_request_type pos_type;
|
|
|
|
union rbd_img_fill_iter *pos;
|
|
|
|
union rbd_img_fill_iter iter;
|
|
|
|
ceph_object_extent_fn_t set_pos_fn;
|
2018-02-07 02:26:35 +08:00
|
|
|
ceph_object_extent_fn_t count_fn;
|
|
|
|
ceph_object_extent_fn_t copy_fn;
|
2018-02-07 02:26:34 +08:00
|
|
|
};
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static struct ceph_object_extent *alloc_object_extent(void *arg)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2018-02-07 02:26:34 +08:00
|
|
|
struct rbd_img_request *img_req = arg;
|
|
|
|
struct rbd_obj_request *obj_req;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
obj_req = rbd_obj_request_create();
|
|
|
|
if (!obj_req)
|
|
|
|
return NULL;
|
rbd: fix copyup completion race
For write/discard obj_requests that involved a copyup method call, the
opcode of the first op is CEPH_OSD_OP_CALL and the ->callback is
rbd_img_obj_copyup_callback(). The latter frees copyup pages, sets
->xferred and delegates to rbd_img_obj_callback(), the "normal" image
object callback, for reporting to block layer and putting refs.
rbd_osd_req_callback() however treats CEPH_OSD_OP_CALL as a trivial op,
which means obj_request is marked done in rbd_osd_trivial_callback(),
*before* ->callback is invoked and rbd_img_obj_copyup_callback() has
a chance to run. Marking obj_request done essentially means giving
rbd_img_obj_callback() a license to end it at any moment, so if another
obj_request from the same img_request is being completed concurrently,
rbd_img_obj_end_request() may very well be called on such prematurally
marked done request:
<obj_request-1/2 reply>
handle_reply()
rbd_osd_req_callback()
rbd_osd_trivial_callback()
rbd_obj_request_complete()
rbd_img_obj_copyup_callback()
rbd_img_obj_callback()
<obj_request-2/2 reply>
handle_reply()
rbd_osd_req_callback()
rbd_osd_trivial_callback()
for_each_obj_request(obj_request->img_request) {
rbd_img_obj_end_request(obj_request-1/2)
rbd_img_obj_end_request(obj_request-2/2) <--
}
Calling rbd_img_obj_end_request() on such a request leads to trouble,
in particular because its ->xfferred is 0. We report 0 to the block
layer with blk_update_request(), get back 1 for "this request has more
data in flight" and then trip on
rbd_assert(more ^ (which == img_request->obj_request_count));
with rhs (which == ...) being 1 because rbd_img_obj_end_request() has
been called for both requests and lhs (more) being 1 because we haven't
got a chance to set ->xfferred in rbd_img_obj_copyup_callback() yet.
To fix this, leverage that rbd wants to call class methods in only two
cases: one is a generic method call wrapper (obj_request is standalone)
and the other is a copyup (obj_request is part of an img_request). So
make a dedicated handler for CEPH_OSD_OP_CALL and directly invoke
rbd_img_obj_copyup_callback() from it if obj_request is part of an
img_request, similar to how CEPH_OSD_OP_READ handler invokes
rbd_img_obj_request_read_callback().
Since rbd_img_obj_copyup_callback() is now being called from the OSD
request callback (only), it is renamed to rbd_osd_copyup_callback().
Cc: Alex Elder <elder@linaro.org>
Cc: stable@vger.kernel.org # 3.10+, needs backporting for < 3.18
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Alex Elder <elder@linaro.org>
2015-07-16 22:36:11 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
rbd_img_obj_request_add(img_req, obj_req);
|
|
|
|
return &obj_req->ex;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
/*
|
|
|
|
* While su != os && sc == 1 is technically not fancy (it's the same
|
|
|
|
* layout as su == os && sc == 1), we can't use the nocopy path for it
|
|
|
|
* because ->set_pos_fn() should be called only once per object.
|
|
|
|
* ceph_file_to_extents() invokes action_fn once per stripe unit, so
|
|
|
|
* treat su != os && sc == 1 as fancy.
|
|
|
|
*/
|
|
|
|
static bool rbd_layout_is_fancy(struct ceph_file_layout *l)
|
|
|
|
{
|
|
|
|
return l->stripe_unit != l->object_size;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req,
|
|
|
|
struct ceph_file_extent *img_extents,
|
|
|
|
u32 num_img_extents,
|
|
|
|
struct rbd_img_fill_ctx *fctx)
|
|
|
|
{
|
|
|
|
u32 i;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
img_req->data_type = fctx->pos_type;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
|
|
|
/*
|
2018-02-07 02:26:35 +08:00
|
|
|
* Create object requests and set each object request's starting
|
|
|
|
* position in the provided bio (list) or bio_vec array.
|
2013-04-20 04:34:50 +08:00
|
|
|
*/
|
2018-02-07 02:26:35 +08:00
|
|
|
fctx->iter = *fctx->pos;
|
|
|
|
for (i = 0; i < num_img_extents; i++) {
|
|
|
|
ret = ceph_file_to_extents(&img_req->rbd_dev->layout,
|
|
|
|
img_extents[i].fe_off,
|
|
|
|
img_extents[i].fe_len,
|
|
|
|
&img_req->object_extents,
|
|
|
|
alloc_object_extent, img_req,
|
|
|
|
fctx->set_pos_fn, &fctx->iter);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
return __rbd_img_fill_request(img_req);
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
/*
|
|
|
|
* Map a list of image extents to a list of object extents, create the
|
|
|
|
* corresponding object requests (normally each to a different object,
|
|
|
|
* but not always) and add them to @img_req. For each object request,
|
2018-02-07 02:26:35 +08:00
|
|
|
* set up its data descriptor to point to the corresponding chunk(s) of
|
2018-02-07 02:26:34 +08:00
|
|
|
* @fctx->pos data buffer.
|
|
|
|
*
|
2018-02-07 02:26:35 +08:00
|
|
|
* Because ceph_file_to_extents() will merge adjacent object extents
|
|
|
|
* together, each object request's data descriptor may point to multiple
|
|
|
|
* different chunks of @fctx->pos data buffer.
|
|
|
|
*
|
2018-02-07 02:26:34 +08:00
|
|
|
* @fctx->pos data buffer is assumed to be large enough.
|
|
|
|
*/
|
|
|
|
static int rbd_img_fill_request(struct rbd_img_request *img_req,
|
|
|
|
struct ceph_file_extent *img_extents,
|
|
|
|
u32 num_img_extents,
|
|
|
|
struct rbd_img_fill_ctx *fctx)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2018-02-07 02:26:35 +08:00
|
|
|
struct rbd_device *rbd_dev = img_req->rbd_dev;
|
|
|
|
struct rbd_obj_request *obj_req;
|
2018-02-07 02:26:34 +08:00
|
|
|
u32 i;
|
|
|
|
int ret;
|
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
if (fctx->pos_type == OBJ_REQUEST_NODATA ||
|
|
|
|
!rbd_layout_is_fancy(&rbd_dev->layout))
|
|
|
|
return rbd_img_fill_request_nocopy(img_req, img_extents,
|
|
|
|
num_img_extents, fctx);
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
img_req->data_type = OBJ_REQUEST_OWN_BVECS;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2013-05-07 06:40:33 +08:00
|
|
|
/*
|
2018-02-07 02:26:35 +08:00
|
|
|
* Create object requests and determine ->bvec_count for each object
|
|
|
|
* request. Note that ->bvec_count sum over all object requests may
|
|
|
|
* be greater than the number of bio_vecs in the provided bio (list)
|
|
|
|
* or bio_vec array because when mapped, those bio_vecs can straddle
|
|
|
|
* stripe unit boundaries.
|
2013-05-07 06:40:33 +08:00
|
|
|
*/
|
2018-02-07 02:26:34 +08:00
|
|
|
fctx->iter = *fctx->pos;
|
|
|
|
for (i = 0; i < num_img_extents; i++) {
|
2018-02-07 02:26:35 +08:00
|
|
|
ret = ceph_file_to_extents(&rbd_dev->layout,
|
2018-02-07 02:26:34 +08:00
|
|
|
img_extents[i].fe_off,
|
|
|
|
img_extents[i].fe_len,
|
|
|
|
&img_req->object_extents,
|
|
|
|
alloc_object_extent, img_req,
|
2018-02-07 02:26:35 +08:00
|
|
|
fctx->count_fn, &fctx->iter);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2013-05-07 06:40:33 +08:00
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
for_each_obj_request(img_req, obj_req) {
|
|
|
|
obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count,
|
|
|
|
sizeof(*obj_req->bvec_pos.bvecs),
|
|
|
|
GFP_NOIO);
|
|
|
|
if (!obj_req->bvec_pos.bvecs)
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2013-05-09 23:08:49 +08:00
|
|
|
/*
|
2018-02-07 02:26:35 +08:00
|
|
|
* Fill in each object request's private bio_vec array, splitting and
|
|
|
|
* rearranging the provided bio_vecs in stripe unit chunks as needed.
|
2013-05-09 23:08:49 +08:00
|
|
|
*/
|
2018-02-07 02:26:35 +08:00
|
|
|
fctx->iter = *fctx->pos;
|
|
|
|
for (i = 0; i < num_img_extents; i++) {
|
|
|
|
ret = ceph_iterate_extents(&rbd_dev->layout,
|
|
|
|
img_extents[i].fe_off,
|
|
|
|
img_extents[i].fe_len,
|
|
|
|
&img_req->object_extents,
|
|
|
|
fctx->copy_fn, &fctx->iter);
|
2018-02-07 02:26:34 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
return __rbd_img_fill_request(img_req);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_img_fill_nodata(struct rbd_img_request *img_req,
|
|
|
|
u64 off, u64 len)
|
|
|
|
{
|
|
|
|
struct ceph_file_extent ex = { off, len };
|
|
|
|
union rbd_img_fill_iter dummy;
|
|
|
|
struct rbd_img_fill_ctx fctx = {
|
|
|
|
.pos_type = OBJ_REQUEST_NODATA,
|
|
|
|
.pos = &dummy,
|
|
|
|
};
|
|
|
|
|
|
|
|
return rbd_img_fill_request(img_req, &ex, 1, &fctx);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bio_iter *it = arg;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
|
|
|
|
obj_req->bio_pos = *it;
|
|
|
|
ceph_bio_iter_advance(it, bytes);
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bio_iter *it = arg;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
|
|
|
|
ceph_bio_iter_advance_step(it, bytes, ({
|
|
|
|
obj_req->bvec_count++;
|
|
|
|
}));
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bio_iter *it = arg;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes);
|
|
|
|
ceph_bio_iter_advance_step(it, bytes, ({
|
|
|
|
obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
|
|
|
|
obj_req->bvec_pos.iter.bi_size += bv.bv_len;
|
|
|
|
}));
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req,
|
|
|
|
struct ceph_file_extent *img_extents,
|
|
|
|
u32 num_img_extents,
|
|
|
|
struct ceph_bio_iter *bio_pos)
|
|
|
|
{
|
|
|
|
struct rbd_img_fill_ctx fctx = {
|
|
|
|
.pos_type = OBJ_REQUEST_BIO,
|
|
|
|
.pos = (union rbd_img_fill_iter *)bio_pos,
|
|
|
|
.set_pos_fn = set_bio_pos,
|
2018-02-07 02:26:35 +08:00
|
|
|
.count_fn = count_bio_bvecs,
|
|
|
|
.copy_fn = copy_bio_bvecs,
|
2018-02-07 02:26:34 +08:00
|
|
|
};
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
return rbd_img_fill_request(img_req, img_extents, num_img_extents,
|
|
|
|
&fctx);
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static int rbd_img_fill_from_bio(struct rbd_img_request *img_req,
|
|
|
|
u64 off, u64 len, struct bio *bio)
|
|
|
|
{
|
|
|
|
struct ceph_file_extent ex = { off, len };
|
|
|
|
struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter };
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
return __rbd_img_fill_from_bio(img_req, &ex, 1, &it);
|
|
|
|
}
|
2013-04-21 13:32:07 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bvec_iter *it = arg;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
obj_req->bvec_pos = *it;
|
|
|
|
ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes);
|
|
|
|
ceph_bvec_iter_advance(it, bytes);
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bvec_iter *it = arg;
|
2016-09-12 20:44:45 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
ceph_bvec_iter_advance_step(it, bytes, ({
|
|
|
|
obj_req->bvec_count++;
|
|
|
|
}));
|
|
|
|
}
|
2016-09-12 20:44:45 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg)
|
|
|
|
{
|
|
|
|
struct rbd_obj_request *obj_req =
|
|
|
|
container_of(ex, struct rbd_obj_request, ex);
|
|
|
|
struct ceph_bvec_iter *it = arg;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
ceph_bvec_iter_advance_step(it, bytes, ({
|
|
|
|
obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv;
|
|
|
|
obj_req->bvec_pos.iter.bi_size += bv.bv_len;
|
|
|
|
}));
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
|
|
|
|
struct ceph_file_extent *img_extents,
|
|
|
|
u32 num_img_extents,
|
|
|
|
struct ceph_bvec_iter *bvec_pos)
|
2013-02-12 02:33:24 +08:00
|
|
|
{
|
2018-02-07 02:26:34 +08:00
|
|
|
struct rbd_img_fill_ctx fctx = {
|
|
|
|
.pos_type = OBJ_REQUEST_BVECS,
|
|
|
|
.pos = (union rbd_img_fill_iter *)bvec_pos,
|
|
|
|
.set_pos_fn = set_bvec_pos,
|
2018-02-07 02:26:35 +08:00
|
|
|
.count_fn = count_bvecs,
|
|
|
|
.copy_fn = copy_bvecs,
|
2018-02-07 02:26:34 +08:00
|
|
|
};
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
return rbd_img_fill_request(img_req, img_extents, num_img_extents,
|
|
|
|
&fctx);
|
|
|
|
}
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req,
|
|
|
|
struct ceph_file_extent *img_extents,
|
|
|
|
u32 num_img_extents,
|
|
|
|
struct bio_vec *bvecs)
|
|
|
|
{
|
|
|
|
struct ceph_bvec_iter it = {
|
|
|
|
.bvecs = bvecs,
|
|
|
|
.iter = { .bi_size = ceph_file_extents_bytes(img_extents,
|
|
|
|
num_img_extents) },
|
|
|
|
};
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2018-02-07 02:26:34 +08:00
|
|
|
return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents,
|
|
|
|
&it);
|
|
|
|
}
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
static void rbd_img_handle_request_work(struct work_struct *work)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2019-05-16 21:06:56 +08:00
|
|
|
struct rbd_img_request *img_req =
|
|
|
|
container_of(work, struct rbd_img_request, work);
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
rbd_img_handle_request(img_req, img_req->work_result);
|
|
|
|
}
|
2016-09-14 02:18:01 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
static void rbd_img_schedule(struct rbd_img_request *img_req, int result)
|
|
|
|
{
|
|
|
|
INIT_WORK(&img_req->work, rbd_img_handle_request_work);
|
|
|
|
img_req->work_result = result;
|
|
|
|
queue_work(rbd_wq, &img_req->work);
|
2013-02-12 02:33:24 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
static int rbd_obj_read_object(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
rbd_obj_request_submit(obj_req);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
|
2013-02-12 02:33:24 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_img_request *img_req = obj_req->img_request;
|
|
|
|
struct rbd_img_request *child_img_req;
|
2013-02-12 02:33:24 +08:00
|
|
|
int ret;
|
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
child_img_req = rbd_img_request_create(img_req->rbd_dev->parent,
|
|
|
|
OBJ_OP_READ, NULL);
|
2018-01-29 21:04:08 +08:00
|
|
|
if (!child_img_req)
|
2016-09-15 23:53:32 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
__set_bit(IMG_REQ_CHILD, &child_img_req->flags);
|
|
|
|
child_img_req->obj_request = obj_req;
|
2017-01-26 01:16:23 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
if (!rbd_img_is_write(img_req)) {
|
2018-02-01 18:50:47 +08:00
|
|
|
switch (img_req->data_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_REQUEST_BIO:
|
2018-02-07 02:26:34 +08:00
|
|
|
ret = __rbd_img_fill_from_bio(child_img_req,
|
|
|
|
obj_req->img_extents,
|
|
|
|
obj_req->num_img_extents,
|
|
|
|
&obj_req->bio_pos);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
|
|
|
case OBJ_REQUEST_BVECS:
|
2018-02-07 02:26:35 +08:00
|
|
|
case OBJ_REQUEST_OWN_BVECS:
|
2018-02-07 02:26:34 +08:00
|
|
|
ret = __rbd_img_fill_from_bvecs(child_img_req,
|
|
|
|
obj_req->img_extents,
|
|
|
|
obj_req->num_img_extents,
|
|
|
|
&obj_req->bvec_pos);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
|
|
|
default:
|
2019-03-22 22:36:37 +08:00
|
|
|
BUG();
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
|
|
|
} else {
|
2018-02-07 02:26:34 +08:00
|
|
|
ret = rbd_img_fill_from_bvecs(child_img_req,
|
|
|
|
obj_req->img_extents,
|
|
|
|
obj_req->num_img_extents,
|
|
|
|
obj_req->copyup_bvecs);
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
|
|
|
if (ret) {
|
|
|
|
rbd_img_request_put(child_img_req);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
/* avoid parent chain recursion */
|
|
|
|
rbd_img_schedule(child_img_req, 0);
|
2018-01-29 21:04:08 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
static bool rbd_obj_advance_read(struct rbd_obj_request *obj_req, int *result)
|
2018-01-29 21:04:08 +08:00
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
|
|
|
int ret;
|
|
|
|
|
2019-05-08 19:35:57 +08:00
|
|
|
switch (obj_req->read_state) {
|
2019-05-15 03:06:07 +08:00
|
|
|
case RBD_OBJ_READ_START:
|
|
|
|
rbd_assert(!*result);
|
|
|
|
|
|
|
|
ret = rbd_obj_read_object(obj_req);
|
|
|
|
if (ret) {
|
|
|
|
*result = ret;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
obj_req->read_state = RBD_OBJ_READ_OBJECT;
|
|
|
|
return false;
|
2019-05-08 19:35:57 +08:00
|
|
|
case RBD_OBJ_READ_OBJECT:
|
|
|
|
if (*result == -ENOENT && rbd_dev->parent_overlap) {
|
|
|
|
/* reverse map this object extent onto the parent */
|
|
|
|
ret = rbd_obj_calc_img_extents(obj_req, false);
|
2018-02-07 02:26:33 +08:00
|
|
|
if (ret) {
|
2019-05-11 22:21:49 +08:00
|
|
|
*result = ret;
|
2018-02-07 02:26:33 +08:00
|
|
|
return true;
|
|
|
|
}
|
2019-05-08 19:35:57 +08:00
|
|
|
if (obj_req->num_img_extents) {
|
|
|
|
ret = rbd_obj_read_from_parent(obj_req);
|
|
|
|
if (ret) {
|
|
|
|
*result = ret;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
obj_req->read_state = RBD_OBJ_READ_PARENT;
|
|
|
|
return false;
|
|
|
|
}
|
2018-02-07 02:26:33 +08:00
|
|
|
}
|
2016-09-15 23:53:32 +08:00
|
|
|
|
2019-05-08 19:35:57 +08:00
|
|
|
/*
|
|
|
|
* -ENOENT means a hole in the image -- zero-fill the entire
|
|
|
|
* length of the request. A short read also implies zero-fill
|
|
|
|
* to the end of the request.
|
|
|
|
*/
|
|
|
|
if (*result == -ENOENT) {
|
|
|
|
rbd_obj_zero_range(obj_req, 0, obj_req->ex.oe_len);
|
|
|
|
*result = 0;
|
|
|
|
} else if (*result >= 0) {
|
|
|
|
if (*result < obj_req->ex.oe_len)
|
|
|
|
rbd_obj_zero_range(obj_req, *result,
|
|
|
|
obj_req->ex.oe_len - *result);
|
|
|
|
else
|
|
|
|
rbd_assert(*result == obj_req->ex.oe_len);
|
|
|
|
*result = 0;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
case RBD_OBJ_READ_PARENT:
|
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
BUG();
|
2016-09-15 23:53:32 +08:00
|
|
|
}
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
static int rbd_obj_write_object(struct rbd_obj_request *obj_req)
|
|
|
|
{
|
|
|
|
rbd_obj_request_submit(obj_req);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
|
|
|
* copyup_bvecs pages are never highmem pages
|
|
|
|
*/
|
|
|
|
static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes)
|
|
|
|
{
|
|
|
|
struct ceph_bvec_iter it = {
|
|
|
|
.bvecs = bvecs,
|
|
|
|
.iter = { .bi_size = bytes },
|
|
|
|
};
|
2013-02-12 02:33:24 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
ceph_bvec_iter_advance_step(&it, bytes, ({
|
|
|
|
if (memchr_inv(page_address(bv.bv_page) + bv.bv_offset, 0,
|
|
|
|
bv.bv_len))
|
|
|
|
return false;
|
|
|
|
}));
|
|
|
|
return true;
|
2013-02-12 02:33:24 +08:00
|
|
|
}
|
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
#define MODS_ONLY U32_MAX
|
|
|
|
|
2019-02-28 21:20:28 +08:00
|
|
|
static int rbd_obj_issue_copyup_empty_snapc(struct rbd_obj_request *obj_req,
|
|
|
|
u32 bytes)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2018-04-12 12:04:55 +08:00
|
|
|
int ret;
|
2014-09-12 20:02:01 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
|
2019-02-28 21:20:28 +08:00
|
|
|
rbd_assert(bytes > 0 && bytes != MODS_ONLY);
|
2014-09-12 20:02:01 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = __rbd_obj_add_osd_request(obj_req, &rbd_empty_snapc, 1);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = osd_req_op_cls_init(osd_req, 0, "rbd", "copyup");
|
2018-04-12 12:04:55 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_cls_request_data_bvecs(osd_req, 0,
|
2018-05-04 22:57:30 +08:00
|
|
|
obj_req->copyup_bvecs,
|
|
|
|
obj_req->copyup_bvec_count,
|
|
|
|
bytes);
|
2019-05-27 17:41:36 +08:00
|
|
|
rbd_osd_format_write(osd_req);
|
2018-01-29 21:04:08 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
|
2019-02-28 21:20:28 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
rbd_obj_request_submit(obj_req);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
static int rbd_obj_issue_copyup_ops(struct rbd_obj_request *obj_req, u32 bytes)
|
2013-04-20 04:34:50 +08:00
|
|
|
{
|
2019-02-25 19:37:50 +08:00
|
|
|
struct rbd_img_request *img_req = obj_req->img_request;
|
2019-05-27 17:41:36 +08:00
|
|
|
struct ceph_osd_request *osd_req;
|
2019-02-28 17:49:12 +08:00
|
|
|
unsigned int num_osd_ops = (bytes != MODS_ONLY);
|
|
|
|
unsigned int which = 0;
|
2018-04-12 12:04:55 +08:00
|
|
|
int ret;
|
2014-09-12 20:02:01 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes);
|
2014-09-12 20:02:01 +08:00
|
|
|
|
2019-02-25 19:37:50 +08:00
|
|
|
switch (img_req->op_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_OP_WRITE:
|
2019-02-25 19:37:50 +08:00
|
|
|
num_osd_ops += count_write_ops(obj_req);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
2019-02-25 19:37:50 +08:00
|
|
|
case OBJ_OP_ZEROOUT:
|
|
|
|
num_osd_ops += count_zeroout_ops(obj_req);
|
|
|
|
break;
|
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2019-02-25 19:37:50 +08:00
|
|
|
}
|
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req = rbd_obj_add_osd_request(obj_req, num_osd_ops);
|
|
|
|
if (IS_ERR(osd_req))
|
|
|
|
return PTR_ERR(osd_req);
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
if (bytes != MODS_ONLY) {
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = osd_req_op_cls_init(osd_req, which, "rbd",
|
2019-02-28 17:49:12 +08:00
|
|
|
"copyup");
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2018-04-12 12:04:55 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
osd_req_op_cls_request_data_bvecs(osd_req, which++,
|
2019-02-28 17:49:12 +08:00
|
|
|
obj_req->copyup_bvecs,
|
|
|
|
obj_req->copyup_bvec_count,
|
|
|
|
bytes);
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
|
|
|
|
2019-02-25 19:37:50 +08:00
|
|
|
switch (img_req->op_type) {
|
2018-01-29 21:04:08 +08:00
|
|
|
case OBJ_OP_WRITE:
|
2019-05-27 17:41:36 +08:00
|
|
|
__rbd_osd_setup_write_ops(osd_req, which);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
2019-01-29 19:46:25 +08:00
|
|
|
case OBJ_OP_ZEROOUT:
|
2019-05-27 17:41:36 +08:00
|
|
|
__rbd_osd_setup_zeroout_ops(osd_req, which);
|
2018-01-29 21:04:08 +08:00
|
|
|
break;
|
|
|
|
default:
|
2019-03-23 00:53:56 +08:00
|
|
|
BUG();
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2014-09-12 20:02:01 +08:00
|
|
|
|
2019-05-27 17:41:36 +08:00
|
|
|
ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO);
|
2018-10-15 22:11:37 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
rbd_obj_request_submit(obj_req);
|
|
|
|
return 0;
|
2014-09-12 20:02:01 +08:00
|
|
|
}
|
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
static int rbd_obj_issue_copyup(struct rbd_obj_request *obj_req, u32 bytes)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Only send non-zero copyup data to save some I/O and network
|
|
|
|
* bandwidth -- zero copyup data is equivalent to the object not
|
|
|
|
* existing.
|
|
|
|
*/
|
|
|
|
if (is_zero_bvecs(obj_req->copyup_bvecs, bytes)) {
|
|
|
|
dout("%s obj_req %p detected zeroes\n", __func__, obj_req);
|
|
|
|
bytes = 0;
|
|
|
|
}
|
|
|
|
|
2019-02-28 21:20:28 +08:00
|
|
|
if (obj_req->img_request->snapc->num_snaps && bytes > 0) {
|
|
|
|
/*
|
|
|
|
* Send a copyup request with an empty snapshot context to
|
|
|
|
* deep-copyup the object through all existing snapshots.
|
|
|
|
* A second request with the current snapshot context will be
|
|
|
|
* sent for the actual modification.
|
|
|
|
*/
|
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC;
|
|
|
|
return rbd_obj_issue_copyup_empty_snapc(obj_req, bytes);
|
|
|
|
}
|
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_COPYUP_OPS;
|
|
|
|
return rbd_obj_issue_copyup_ops(obj_req, bytes);
|
|
|
|
}
|
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap)
|
2014-09-12 20:02:01 +08:00
|
|
|
{
|
2018-01-20 17:30:11 +08:00
|
|
|
u32 i;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
rbd_assert(!obj_req->copyup_bvecs);
|
|
|
|
obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap);
|
|
|
|
obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count,
|
|
|
|
sizeof(*obj_req->copyup_bvecs),
|
|
|
|
GFP_NOIO);
|
|
|
|
if (!obj_req->copyup_bvecs)
|
|
|
|
return -ENOMEM;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
for (i = 0; i < obj_req->copyup_bvec_count; i++) {
|
|
|
|
unsigned int len = min(obj_overlap, (u64)PAGE_SIZE);
|
|
|
|
|
|
|
|
obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO);
|
|
|
|
if (!obj_req->copyup_bvecs[i].bv_page)
|
|
|
|
return -ENOMEM;
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
obj_req->copyup_bvecs[i].bv_offset = 0;
|
|
|
|
obj_req->copyup_bvecs[i].bv_len = len;
|
|
|
|
obj_overlap -= len;
|
|
|
|
}
|
2013-04-20 04:34:50 +08:00
|
|
|
|
2018-01-20 17:30:11 +08:00
|
|
|
rbd_assert(!obj_overlap);
|
|
|
|
return 0;
|
2013-04-20 04:34:50 +08:00
|
|
|
}
|
|
|
|
|
2019-05-15 02:45:38 +08:00
|
|
|
/*
|
|
|
|
* The target object doesn't exist. Read the data for the entire
|
|
|
|
* target object up to the overlap point (if any) from the parent,
|
|
|
|
* so we can use it for a copyup.
|
|
|
|
*/
|
2018-01-29 21:04:08 +08:00
|
|
|
static int rbd_obj_handle_write_guard(struct rbd_obj_request *obj_req)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
|
|
|
|
int ret;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
rbd_assert(obj_req->num_img_extents);
|
|
|
|
prune_extents(obj_req->img_extents, &obj_req->num_img_extents,
|
|
|
|
rbd_dev->parent_overlap);
|
|
|
|
if (!obj_req->num_img_extents) {
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
|
|
|
* The overlap has become 0 (most likely because the
|
2019-02-28 17:49:12 +08:00
|
|
|
* image has been flattened). Re-submit the original write
|
|
|
|
* request -- pass MODS_ONLY since the copyup isn't needed
|
|
|
|
* anymore.
|
2018-01-29 21:04:08 +08:00
|
|
|
*/
|
2019-02-28 17:49:12 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_COPYUP_OPS;
|
|
|
|
return rbd_obj_issue_copyup_ops(obj_req, MODS_ONLY);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:33 +08:00
|
|
|
ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req));
|
2018-01-29 21:04:08 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2019-02-28 17:49:12 +08:00
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_READ_FROM_PARENT;
|
2018-02-07 02:26:33 +08:00
|
|
|
return rbd_obj_read_from_parent(obj_req);
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
static bool rbd_obj_advance_write(struct rbd_obj_request *obj_req, int *result)
|
2013-01-25 06:13:36 +08:00
|
|
|
{
|
2018-01-29 21:04:08 +08:00
|
|
|
int ret;
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
switch (obj_req->write_state) {
|
2019-05-15 03:06:07 +08:00
|
|
|
case RBD_OBJ_WRITE_START:
|
|
|
|
rbd_assert(!*result);
|
|
|
|
|
|
|
|
ret = rbd_obj_write_object(obj_req);
|
|
|
|
if (ret) {
|
|
|
|
*result = ret;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_OBJECT;
|
|
|
|
return false;
|
2019-05-15 02:45:38 +08:00
|
|
|
case RBD_OBJ_WRITE_OBJECT:
|
2019-05-11 22:21:49 +08:00
|
|
|
if (*result == -ENOENT) {
|
2019-05-15 02:45:38 +08:00
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) {
|
|
|
|
ret = rbd_obj_handle_write_guard(obj_req);
|
|
|
|
if (ret) {
|
|
|
|
*result = ret;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
2019-05-15 02:45:38 +08:00
|
|
|
* On a non-existent object:
|
|
|
|
* delete - -ENOENT, truncate/zero - 0
|
2018-01-29 21:04:08 +08:00
|
|
|
*/
|
2019-05-15 02:45:38 +08:00
|
|
|
if (obj_req->flags & RBD_OBJ_FLAG_DELETION)
|
|
|
|
*result = 0;
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
|
|
|
/* fall through */
|
2019-02-28 17:49:12 +08:00
|
|
|
case RBD_OBJ_WRITE_COPYUP_OPS:
|
2018-01-29 21:04:08 +08:00
|
|
|
return true;
|
2019-02-28 17:49:12 +08:00
|
|
|
case RBD_OBJ_WRITE_READ_FROM_PARENT:
|
2019-05-08 19:35:57 +08:00
|
|
|
if (*result)
|
2019-02-28 17:49:12 +08:00
|
|
|
return true;
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-08 19:35:57 +08:00
|
|
|
ret = rbd_obj_issue_copyup(obj_req,
|
|
|
|
rbd_obj_img_extents_bytes(obj_req));
|
2018-01-29 21:04:08 +08:00
|
|
|
if (ret) {
|
2019-05-11 22:21:49 +08:00
|
|
|
*result = ret;
|
2018-01-29 21:04:08 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
2019-02-28 21:20:28 +08:00
|
|
|
case RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC:
|
2019-05-11 22:21:49 +08:00
|
|
|
if (*result)
|
2019-02-28 21:20:28 +08:00
|
|
|
return true;
|
|
|
|
|
|
|
|
obj_req->write_state = RBD_OBJ_WRITE_COPYUP_OPS;
|
|
|
|
ret = rbd_obj_issue_copyup_ops(obj_req, MODS_ONLY);
|
2018-01-29 21:04:08 +08:00
|
|
|
if (ret) {
|
2019-05-11 22:21:49 +08:00
|
|
|
*result = ret;
|
2018-01-29 21:04:08 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
default:
|
2018-04-04 20:53:39 +08:00
|
|
|
BUG();
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
|
|
|
}
|
2013-05-07 06:40:33 +08:00
|
|
|
|
2018-01-29 21:04:08 +08:00
|
|
|
/*
|
2019-05-15 02:45:38 +08:00
|
|
|
* Return true if @obj_req is completed.
|
2018-01-29 21:04:08 +08:00
|
|
|
*/
|
2019-05-11 22:21:49 +08:00
|
|
|
static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req,
|
|
|
|
int *result)
|
2018-01-29 21:04:08 +08:00
|
|
|
{
|
2019-05-15 02:45:38 +08:00
|
|
|
struct rbd_img_request *img_req = obj_req->img_request;
|
2019-05-16 21:06:56 +08:00
|
|
|
struct rbd_device *rbd_dev = img_req->rbd_dev;
|
2019-05-15 02:45:38 +08:00
|
|
|
bool done;
|
|
|
|
|
2019-05-15 03:06:07 +08:00
|
|
|
mutex_lock(&obj_req->state_mutex);
|
2019-05-15 02:45:38 +08:00
|
|
|
if (!rbd_img_is_write(img_req))
|
2019-05-15 03:06:07 +08:00
|
|
|
done = rbd_obj_advance_read(obj_req, result);
|
2019-05-15 02:45:38 +08:00
|
|
|
else
|
2019-05-15 03:06:07 +08:00
|
|
|
done = rbd_obj_advance_write(obj_req, result);
|
|
|
|
mutex_unlock(&obj_req->state_mutex);
|
2019-05-15 02:45:38 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
if (done && *result) {
|
|
|
|
rbd_assert(*result < 0);
|
|
|
|
rbd_warn(rbd_dev, "%s at objno %llu %llu~%llu result %d",
|
|
|
|
obj_op_name(img_req->op_type), obj_req->ex.oe_objno,
|
|
|
|
obj_req->ex.oe_off, obj_req->ex.oe_len, *result);
|
|
|
|
}
|
2019-05-15 02:45:38 +08:00
|
|
|
return done;
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2013-05-07 06:40:33 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
/*
|
|
|
|
* This is open-coded in rbd_img_handle_request() to avoid parent chain
|
|
|
|
* recursion.
|
|
|
|
*/
|
|
|
|
static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result)
|
|
|
|
{
|
|
|
|
if (__rbd_obj_handle_request(obj_req, &result))
|
|
|
|
rbd_img_handle_request(obj_req->img_request, result);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_img_object_requests(struct rbd_img_request *img_req)
|
2018-02-01 18:50:47 +08:00
|
|
|
{
|
2019-05-16 21:06:56 +08:00
|
|
|
struct rbd_obj_request *obj_req;
|
2018-02-01 18:50:47 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
|
|
|
|
|
|
|
|
for_each_obj_request(img_req, obj_req) {
|
|
|
|
int result = 0;
|
2013-04-21 13:32:07 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
if (__rbd_obj_handle_request(obj_req, &result)) {
|
|
|
|
if (result) {
|
|
|
|
img_req->pending.result = result;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
img_req->pending.num_pending++;
|
|
|
|
}
|
|
|
|
}
|
2013-01-25 06:13:36 +08:00
|
|
|
}
|
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
|
2013-01-25 06:13:36 +08:00
|
|
|
{
|
2019-05-16 21:06:56 +08:00
|
|
|
again:
|
|
|
|
switch (img_req->state) {
|
|
|
|
case RBD_IMG_START:
|
|
|
|
rbd_assert(!*result);
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
rbd_img_object_requests(img_req);
|
|
|
|
if (!img_req->pending.num_pending) {
|
|
|
|
*result = img_req->pending.result;
|
|
|
|
img_req->state = RBD_IMG_OBJECT_REQUESTS;
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
img_req->state = __RBD_IMG_OBJECT_REQUESTS;
|
|
|
|
return false;
|
|
|
|
case __RBD_IMG_OBJECT_REQUESTS:
|
|
|
|
if (!pending_result_dec(&img_req->pending, result))
|
|
|
|
return false;
|
|
|
|
/* fall through */
|
|
|
|
case RBD_IMG_OBJECT_REQUESTS:
|
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
BUG();
|
|
|
|
}
|
2018-01-29 21:04:08 +08:00
|
|
|
}
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
/*
|
|
|
|
* Return true if @img_req is completed.
|
|
|
|
*/
|
|
|
|
static bool __rbd_img_handle_request(struct rbd_img_request *img_req,
|
|
|
|
int *result)
|
2018-01-29 21:04:08 +08:00
|
|
|
{
|
2019-05-16 21:06:56 +08:00
|
|
|
struct rbd_device *rbd_dev = img_req->rbd_dev;
|
|
|
|
bool done;
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
mutex_lock(&img_req->state_mutex);
|
|
|
|
done = rbd_img_advance(img_req, result);
|
|
|
|
mutex_unlock(&img_req->state_mutex);
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
if (done && *result) {
|
|
|
|
rbd_assert(*result < 0);
|
|
|
|
rbd_warn(rbd_dev, "%s%s result %d",
|
|
|
|
test_bit(IMG_REQ_CHILD, &img_req->flags) ? "child " : "",
|
|
|
|
obj_op_name(img_req->op_type), *result);
|
2018-02-01 18:50:47 +08:00
|
|
|
}
|
2019-05-16 21:06:56 +08:00
|
|
|
return done;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_img_handle_request(struct rbd_img_request *img_req, int result)
|
|
|
|
{
|
|
|
|
again:
|
|
|
|
if (!__rbd_img_handle_request(img_req, &result))
|
|
|
|
return;
|
2013-01-25 06:13:36 +08:00
|
|
|
|
2018-02-01 18:50:47 +08:00
|
|
|
if (test_bit(IMG_REQ_CHILD, &img_req->flags)) {
|
2019-05-16 21:06:56 +08:00
|
|
|
struct rbd_obj_request *obj_req = img_req->obj_request;
|
|
|
|
|
2019-05-11 22:21:49 +08:00
|
|
|
rbd_img_request_put(img_req);
|
2019-05-16 21:06:56 +08:00
|
|
|
if (__rbd_obj_handle_request(obj_req, &result)) {
|
|
|
|
img_req = obj_req->img_request;
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
struct request *rq = img_req->rq;
|
|
|
|
|
|
|
|
rbd_img_request_put(img_req);
|
|
|
|
blk_mq_end_request(rq, errno_to_blk_status(result));
|
2018-02-01 18:50:47 +08:00
|
|
|
}
|
2013-01-25 06:13:36 +08:00
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static const struct rbd_client_id rbd_empty_cid;
|
2012-12-01 07:53:04 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static bool rbd_cid_equal(const struct rbd_client_id *lhs,
|
|
|
|
const struct rbd_client_id *rhs)
|
|
|
|
{
|
|
|
|
return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct rbd_client_id cid;
|
|
|
|
|
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
|
|
|
cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
|
|
|
|
cid.handle = rbd_dev->watch_cookie;
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
|
|
|
return cid;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
|
|
|
static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
|
|
|
|
const struct rbd_client_id *cid)
|
|
|
|
{
|
|
|
|
dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
|
|
|
|
rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
|
|
|
|
cid->gid, cid->handle);
|
|
|
|
rbd_dev->owner_cid = *cid; /* struct */
|
|
|
|
}
|
|
|
|
|
|
|
|
static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
|
|
|
|
{
|
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
|
|
|
sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
|
|
|
}
|
|
|
|
|
2017-12-13 23:43:59 +08:00
|
|
|
static void __rbd_lock(struct rbd_device *rbd_dev, const char *cookie)
|
|
|
|
{
|
|
|
|
struct rbd_client_id cid = rbd_get_cid(rbd_dev);
|
|
|
|
|
|
|
|
strcpy(rbd_dev->lock_cookie, cookie);
|
|
|
|
rbd_set_owner_cid(rbd_dev, &cid);
|
|
|
|
queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
|
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
|
|
|
static int rbd_lock(struct rbd_device *rbd_dev)
|
2012-12-01 07:53:04 +08:00
|
|
|
{
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
2016-08-12 22:40:02 +08:00
|
|
|
char cookie[32];
|
2013-05-06 20:40:30 +08:00
|
|
|
int ret;
|
2012-12-01 07:53:04 +08:00
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
WARN_ON(__rbd_is_lock_owner(rbd_dev) ||
|
|
|
|
rbd_dev->lock_cookie[0] != '\0');
|
2014-07-23 21:11:20 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
format_lock_cookie(rbd_dev, cookie);
|
|
|
|
ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
|
|
|
|
RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
|
|
|
|
RBD_LOCK_TAG, "", 0);
|
2013-05-06 20:40:30 +08:00
|
|
|
if (ret)
|
2016-08-12 22:40:02 +08:00
|
|
|
return ret;
|
2012-12-01 07:53:04 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
|
2017-12-13 23:43:59 +08:00
|
|
|
__rbd_lock(rbd_dev, cookie);
|
2016-08-12 22:40:02 +08:00
|
|
|
return 0;
|
2012-12-01 07:53:04 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
2017-04-13 18:17:38 +08:00
|
|
|
static void rbd_unlock(struct rbd_device *rbd_dev)
|
2014-06-19 15:38:14 +08:00
|
|
|
{
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
2014-06-19 15:38:14 +08:00
|
|
|
int ret;
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
WARN_ON(!__rbd_is_lock_owner(rbd_dev) ||
|
|
|
|
rbd_dev->lock_cookie[0] == '\0');
|
2014-06-19 15:38:14 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
|
2017-04-13 18:17:38 +08:00
|
|
|
RBD_LOCK_NAME, rbd_dev->lock_cookie);
|
2017-04-13 18:17:38 +08:00
|
|
|
if (ret && ret != -ENOENT)
|
|
|
|
rbd_warn(rbd_dev, "failed to unlock: %d", ret);
|
2014-06-19 15:38:14 +08:00
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
/* treat errors as the image is unlocked */
|
|
|
|
rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
|
2017-04-13 18:17:38 +08:00
|
|
|
rbd_dev->lock_cookie[0] = '\0';
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
|
|
|
|
queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
|
2014-06-19 15:38:14 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
|
|
|
|
enum rbd_notify_op notify_op,
|
|
|
|
struct page ***preply_pages,
|
|
|
|
size_t *preply_len)
|
2013-01-19 02:31:10 +08:00
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
2016-08-12 22:40:02 +08:00
|
|
|
struct rbd_client_id cid = rbd_get_cid(rbd_dev);
|
2018-03-18 00:44:01 +08:00
|
|
|
char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN];
|
|
|
|
int buf_size = sizeof(buf);
|
2016-08-12 22:40:02 +08:00
|
|
|
void *p = buf;
|
2013-01-19 02:31:10 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
|
2013-01-19 02:31:10 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
/* encode *LockPayload NotifyMessage (op + ClientId) */
|
|
|
|
ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
|
|
|
|
ceph_encode_32(&p, notify_op);
|
|
|
|
ceph_encode_64(&p, cid.gid);
|
|
|
|
ceph_encode_64(&p, cid.handle);
|
2013-01-26 07:08:55 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, buf, buf_size,
|
|
|
|
RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
|
2014-05-22 23:28:52 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
|
|
|
|
enum rbd_notify_op notify_op)
|
2014-05-22 23:28:52 +08:00
|
|
|
{
|
2016-08-12 22:40:02 +08:00
|
|
|
struct page **reply_pages;
|
|
|
|
size_t reply_len;
|
2014-05-22 23:28:52 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
|
|
|
|
ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
|
|
|
|
}
|
2014-05-22 23:28:52 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static void rbd_notify_acquired_lock(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
|
|
|
|
acquired_lock_work);
|
2014-06-20 22:29:20 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
|
2016-04-28 22:07:26 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static void rbd_notify_released_lock(struct work_struct *work)
|
2016-04-28 22:07:26 +08:00
|
|
|
{
|
2016-08-12 22:40:02 +08:00
|
|
|
struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
|
|
|
|
released_lock_work);
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
|
2013-12-17 00:02:40 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
static int rbd_request_lock(struct rbd_device *rbd_dev)
|
2013-01-19 14:30:28 +08:00
|
|
|
{
|
2016-08-12 22:40:02 +08:00
|
|
|
struct page **reply_pages;
|
|
|
|
size_t reply_len;
|
|
|
|
bool lock_owner_responded = false;
|
2013-01-19 14:30:28 +08:00
|
|
|
int ret;
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
|
|
|
|
&reply_pages, &reply_len);
|
|
|
|
if (ret && ret != -ETIMEDOUT) {
|
|
|
|
rbd_warn(rbd_dev, "failed to request lock: %d", ret);
|
2013-01-19 14:30:28 +08:00
|
|
|
goto out;
|
2016-08-12 22:40:02 +08:00
|
|
|
}
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
if (reply_len > 0 && reply_len <= PAGE_SIZE) {
|
|
|
|
void *p = page_address(reply_pages[0]);
|
|
|
|
void *const end = p + reply_len;
|
|
|
|
u32 n;
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
|
|
|
|
while (n--) {
|
|
|
|
u8 struct_v;
|
|
|
|
u32 len;
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
ceph_decode_need(&p, end, 8 + 8, e_inval);
|
|
|
|
p += 8 + 8; /* skip gid and cookie */
|
2013-04-06 03:46:02 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
ceph_decode_32_safe(&p, end, len, e_inval);
|
|
|
|
if (!len)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (lock_owner_responded) {
|
|
|
|
rbd_warn(rbd_dev,
|
|
|
|
"duplicate lock owners detected");
|
|
|
|
ret = -EIO;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
lock_owner_responded = true;
|
|
|
|
ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
|
|
|
|
&struct_v, &len);
|
|
|
|
if (ret) {
|
|
|
|
rbd_warn(rbd_dev,
|
|
|
|
"failed to decode ResponseMessage: %d",
|
|
|
|
ret);
|
|
|
|
goto e_inval;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = ceph_decode_32(&p);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!lock_owner_responded) {
|
|
|
|
rbd_warn(rbd_dev, "no lock owners detected");
|
|
|
|
ret = -ETIMEDOUT;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
e_inval:
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void wake_requests(struct rbd_device *rbd_dev, bool wake_all)
|
|
|
|
{
|
|
|
|
dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all);
|
|
|
|
|
|
|
|
cancel_delayed_work(&rbd_dev->lock_dwork);
|
|
|
|
if (wake_all)
|
|
|
|
wake_up_all(&rbd_dev->lock_waitq);
|
|
|
|
else
|
|
|
|
wake_up(&rbd_dev->lock_waitq);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int get_lock_owner_info(struct rbd_device *rbd_dev,
|
|
|
|
struct ceph_locker **lockers, u32 *num_lockers)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
u8 lock_type;
|
|
|
|
char *lock_tag;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
|
|
|
|
|
|
|
ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, RBD_LOCK_NAME,
|
|
|
|
&lock_type, &lock_tag, lockers, num_lockers);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (*num_lockers == 0) {
|
|
|
|
dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (strcmp(lock_tag, RBD_LOCK_TAG)) {
|
|
|
|
rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
|
|
|
|
lock_tag);
|
|
|
|
ret = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (lock_type == CEPH_CLS_LOCK_SHARED) {
|
|
|
|
rbd_warn(rbd_dev, "shared lock type detected");
|
|
|
|
ret = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
|
|
|
|
strlen(RBD_LOCK_COOKIE_PREFIX))) {
|
|
|
|
rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
|
|
|
|
(*lockers)[0].id.cookie);
|
|
|
|
ret = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
kfree(lock_tag);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int find_watcher(struct rbd_device *rbd_dev,
|
|
|
|
const struct ceph_locker *locker)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
struct ceph_watch_item *watchers;
|
|
|
|
u32 num_watchers;
|
|
|
|
u64 cookie;
|
|
|
|
int i;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, &watchers,
|
|
|
|
&num_watchers);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
|
|
|
|
for (i = 0; i < num_watchers; i++) {
|
|
|
|
if (!memcmp(&watchers[i].addr, &locker->info.addr,
|
|
|
|
sizeof(locker->info.addr)) &&
|
|
|
|
watchers[i].cookie == cookie) {
|
|
|
|
struct rbd_client_id cid = {
|
|
|
|
.gid = le64_to_cpu(watchers[i].name.num),
|
|
|
|
.handle = cookie,
|
|
|
|
};
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
|
|
|
|
rbd_dev, cid.gid, cid.handle);
|
|
|
|
rbd_set_owner_cid(rbd_dev, &cid);
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
|
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
kfree(watchers);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
|
|
|
static int rbd_try_lock(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct ceph_client *client = rbd_dev->rbd_client->client;
|
|
|
|
struct ceph_locker *lockers;
|
|
|
|
u32 num_lockers;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
ret = rbd_lock(rbd_dev);
|
|
|
|
if (ret != -EBUSY)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
/* determine if the current lock holder is still alive */
|
|
|
|
ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (num_lockers == 0)
|
|
|
|
goto again;
|
|
|
|
|
|
|
|
ret = find_watcher(rbd_dev, lockers);
|
|
|
|
if (ret) {
|
|
|
|
if (ret > 0)
|
|
|
|
ret = 0; /* have to request lock */
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock",
|
|
|
|
ENTITY_NAME(lockers[0].id.name));
|
|
|
|
|
|
|
|
ret = ceph_monc_blacklist_add(&client->monc,
|
|
|
|
&lockers[0].info.addr);
|
|
|
|
if (ret) {
|
|
|
|
rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d",
|
|
|
|
ENTITY_NAME(lockers[0].id.name), ret);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, RBD_LOCK_NAME,
|
|
|
|
lockers[0].id.cookie,
|
|
|
|
&lockers[0].id.name);
|
|
|
|
if (ret && ret != -ENOENT)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
again:
|
|
|
|
ceph_free_lockers(lockers, num_lockers);
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
ceph_free_lockers(lockers, num_lockers);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED
|
|
|
|
*/
|
|
|
|
static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev,
|
|
|
|
int *pret)
|
|
|
|
{
|
|
|
|
enum rbd_lock_state lock_state;
|
|
|
|
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
|
|
|
dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
|
|
|
|
rbd_dev->lock_state);
|
|
|
|
if (__rbd_is_lock_owner(rbd_dev)) {
|
|
|
|
lock_state = rbd_dev->lock_state;
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
return lock_state;
|
|
|
|
}
|
|
|
|
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
|
|
|
|
rbd_dev->lock_state);
|
|
|
|
if (!__rbd_is_lock_owner(rbd_dev)) {
|
|
|
|
*pret = rbd_try_lock(rbd_dev);
|
|
|
|
if (*pret)
|
|
|
|
rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret);
|
|
|
|
}
|
|
|
|
|
|
|
|
lock_state = rbd_dev->lock_state;
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
return lock_state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_acquire_lock(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
|
|
|
|
struct rbd_device, lock_dwork);
|
|
|
|
enum rbd_lock_state lock_state;
|
2017-07-13 15:46:35 +08:00
|
|
|
int ret = 0;
|
2016-08-12 22:40:02 +08:00
|
|
|
|
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
|
|
|
again:
|
|
|
|
lock_state = rbd_try_acquire_lock(rbd_dev, &ret);
|
|
|
|
if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) {
|
|
|
|
if (lock_state == RBD_LOCK_STATE_LOCKED)
|
|
|
|
wake_requests(rbd_dev, true);
|
|
|
|
dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__,
|
|
|
|
rbd_dev, lock_state, ret);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = rbd_request_lock(rbd_dev);
|
|
|
|
if (ret == -ETIMEDOUT) {
|
|
|
|
goto again; /* treat this as a dead client */
|
2017-04-13 18:17:39 +08:00
|
|
|
} else if (ret == -EROFS) {
|
|
|
|
rbd_warn(rbd_dev, "peer will not release lock");
|
|
|
|
/*
|
|
|
|
* If this is rbd_add_acquire_lock(), we want to fail
|
|
|
|
* immediately -- reuse BLACKLISTED flag. Otherwise we
|
|
|
|
* want to block.
|
|
|
|
*/
|
|
|
|
if (!(rbd_dev->disk->flags & GENHD_FL_UP)) {
|
|
|
|
set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
|
|
|
|
/* wake "rbd map --exclusive" process */
|
|
|
|
wake_requests(rbd_dev, false);
|
|
|
|
}
|
2016-08-12 22:40:02 +08:00
|
|
|
} else if (ret < 0) {
|
|
|
|
rbd_warn(rbd_dev, "error requesting lock: %d", ret);
|
|
|
|
mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
|
|
|
|
RBD_RETRY_DELAY);
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* lock owner acked, but resend if we don't see them
|
|
|
|
* release the lock
|
|
|
|
*/
|
|
|
|
dout("%s rbd_dev %p requeueing lock_dwork\n", __func__,
|
|
|
|
rbd_dev);
|
|
|
|
mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
|
|
|
|
msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
|
|
|
static bool rbd_release_lock(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
|
|
|
|
rbd_dev->lock_state);
|
|
|
|
if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
|
|
|
|
downgrade_write(&rbd_dev->lock_rwsem);
|
2014-07-23 21:11:20 +08:00
|
|
|
/*
|
2016-08-12 22:40:02 +08:00
|
|
|
* Ensure that all in-flight IO is flushed.
|
2014-07-23 21:11:20 +08:00
|
|
|
*
|
2016-08-12 22:40:02 +08:00
|
|
|
* FIXME: ceph_osdc_sync() flushes the entire OSD client, which
|
|
|
|
* may be shared with other devices.
|
2014-07-23 21:11:20 +08:00
|
|
|
*/
|
2016-08-12 22:40:02 +08:00
|
|
|
ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc);
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
|
|
|
|
rbd_dev->lock_state);
|
|
|
|
if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
|
|
|
|
return false;
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
rbd_unlock(rbd_dev);
|
|
|
|
/*
|
|
|
|
* Give others a chance to grab the lock - we would re-acquire
|
|
|
|
* almost immediately if we got new IO during ceph_osdc_sync()
|
|
|
|
* otherwise. We need to ack our own notifications, so this
|
|
|
|
* lock_dwork will be requeued from rbd_wait_state_locked()
|
|
|
|
* after wake_requests() in rbd_handle_released_lock().
|
|
|
|
*/
|
|
|
|
cancel_delayed_work(&rbd_dev->lock_dwork);
|
2016-08-12 22:40:02 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_release_lock_work(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
|
|
|
|
unlock_work);
|
|
|
|
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
rbd_release_lock(rbd_dev);
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
|
|
|
|
void **p)
|
|
|
|
{
|
|
|
|
struct rbd_client_id cid = { 0 };
|
|
|
|
|
|
|
|
if (struct_v >= 2) {
|
|
|
|
cid.gid = ceph_decode_64(p);
|
|
|
|
cid.handle = ceph_decode_64(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
|
|
|
|
cid.handle);
|
|
|
|
if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
|
|
|
|
/*
|
|
|
|
* we already know that the remote client is
|
|
|
|
* the owner
|
|
|
|
*/
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
rbd_set_owner_cid(rbd_dev, &cid);
|
|
|
|
downgrade_write(&rbd_dev->lock_rwsem);
|
|
|
|
} else {
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!__rbd_is_lock_owner(rbd_dev))
|
|
|
|
wake_requests(rbd_dev, false);
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v,
|
|
|
|
void **p)
|
|
|
|
{
|
|
|
|
struct rbd_client_id cid = { 0 };
|
|
|
|
|
|
|
|
if (struct_v >= 2) {
|
|
|
|
cid.gid = ceph_decode_64(p);
|
|
|
|
cid.handle = ceph_decode_64(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
|
|
|
|
cid.handle);
|
|
|
|
if (!rbd_cid_equal(&cid, &rbd_empty_cid)) {
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) {
|
|
|
|
dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n",
|
|
|
|
__func__, rbd_dev, cid.gid, cid.handle,
|
|
|
|
rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle);
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
|
|
|
|
downgrade_write(&rbd_dev->lock_rwsem);
|
|
|
|
} else {
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!__rbd_is_lock_owner(rbd_dev))
|
|
|
|
wake_requests(rbd_dev, false);
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
/*
|
|
|
|
* Returns result for ResponseMessage to be encoded (<= 0), or 1 if no
|
|
|
|
* ResponseMessage is needed.
|
|
|
|
*/
|
|
|
|
static int rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v,
|
|
|
|
void **p)
|
2016-08-12 22:40:02 +08:00
|
|
|
{
|
|
|
|
struct rbd_client_id my_cid = rbd_get_cid(rbd_dev);
|
|
|
|
struct rbd_client_id cid = { 0 };
|
2017-04-13 18:17:39 +08:00
|
|
|
int result = 1;
|
2016-08-12 22:40:02 +08:00
|
|
|
|
|
|
|
if (struct_v >= 2) {
|
|
|
|
cid.gid = ceph_decode_64(p);
|
|
|
|
cid.handle = ceph_decode_64(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid,
|
|
|
|
cid.handle);
|
|
|
|
if (rbd_cid_equal(&cid, &my_cid))
|
2017-04-13 18:17:39 +08:00
|
|
|
return result;
|
2016-08-12 22:40:02 +08:00
|
|
|
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
2017-04-13 18:17:39 +08:00
|
|
|
if (__rbd_is_lock_owner(rbd_dev)) {
|
|
|
|
if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED &&
|
|
|
|
rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid))
|
|
|
|
goto out_unlock;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* encode ResponseMessage(0) so the peer can detect
|
|
|
|
* a missing owner
|
|
|
|
*/
|
|
|
|
result = 0;
|
|
|
|
|
|
|
|
if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) {
|
2017-04-13 18:17:39 +08:00
|
|
|
if (!rbd_dev->opts->exclusive) {
|
|
|
|
dout("%s rbd_dev %p queueing unlock_work\n",
|
|
|
|
__func__, rbd_dev);
|
|
|
|
queue_work(rbd_dev->task_wq,
|
|
|
|
&rbd_dev->unlock_work);
|
|
|
|
} else {
|
|
|
|
/* refuse to release the lock */
|
|
|
|
result = -EROFS;
|
|
|
|
}
|
2016-08-12 22:40:02 +08:00
|
|
|
}
|
|
|
|
}
|
2017-04-13 18:17:39 +08:00
|
|
|
|
|
|
|
out_unlock:
|
2016-08-12 22:40:02 +08:00
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
2017-04-13 18:17:39 +08:00
|
|
|
return result;
|
2016-08-12 22:40:02 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev,
|
|
|
|
u64 notify_id, u64 cookie, s32 *result)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
2018-03-18 00:44:01 +08:00
|
|
|
char buf[4 + CEPH_ENCODING_START_BLK_LEN];
|
|
|
|
int buf_size = sizeof(buf);
|
2016-08-12 22:40:02 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (result) {
|
|
|
|
void *p = buf;
|
|
|
|
|
|
|
|
/* encode ResponseMessage */
|
|
|
|
ceph_start_encoding(&p, 1, 1,
|
|
|
|
buf_size - CEPH_ENCODING_START_BLK_LEN);
|
|
|
|
ceph_encode_32(&p, *result);
|
|
|
|
} else {
|
|
|
|
buf_size = 0;
|
|
|
|
}
|
2012-12-01 07:53:04 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, notify_id, cookie,
|
2016-08-12 22:40:02 +08:00
|
|
|
buf, buf_size);
|
2014-07-23 21:11:20 +08:00
|
|
|
if (ret)
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id,
|
|
|
|
u64 cookie)
|
|
|
|
{
|
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
|
|
|
__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev,
|
|
|
|
u64 notify_id, u64 cookie, s32 result)
|
|
|
|
{
|
|
|
|
dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result);
|
|
|
|
__rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
|
|
|
|
u64 notifier_id, void *data, size_t data_len)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = arg;
|
|
|
|
void *p = data;
|
|
|
|
void *const end = p + data_len;
|
2016-09-06 17:15:48 +08:00
|
|
|
u8 struct_v = 0;
|
2016-08-12 22:40:02 +08:00
|
|
|
u32 len;
|
|
|
|
u32 notify_op;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n",
|
|
|
|
__func__, rbd_dev, cookie, notify_id, data_len);
|
|
|
|
if (data_len) {
|
|
|
|
ret = ceph_start_decoding(&p, end, 1, "NotifyMessage",
|
|
|
|
&struct_v, &len);
|
|
|
|
if (ret) {
|
|
|
|
rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d",
|
|
|
|
ret);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
notify_op = ceph_decode_32(&p);
|
|
|
|
} else {
|
|
|
|
/* legacy notification for header updates */
|
|
|
|
notify_op = RBD_NOTIFY_OP_HEADER_UPDATE;
|
|
|
|
len = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op);
|
|
|
|
switch (notify_op) {
|
|
|
|
case RBD_NOTIFY_OP_ACQUIRED_LOCK:
|
|
|
|
rbd_handle_acquired_lock(rbd_dev, struct_v, &p);
|
|
|
|
rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
|
|
|
|
break;
|
|
|
|
case RBD_NOTIFY_OP_RELEASED_LOCK:
|
|
|
|
rbd_handle_released_lock(rbd_dev, struct_v, &p);
|
|
|
|
rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
|
|
|
|
break;
|
|
|
|
case RBD_NOTIFY_OP_REQUEST_LOCK:
|
2017-04-13 18:17:39 +08:00
|
|
|
ret = rbd_handle_request_lock(rbd_dev, struct_v, &p);
|
|
|
|
if (ret <= 0)
|
2016-08-12 22:40:02 +08:00
|
|
|
rbd_acknowledge_notify_result(rbd_dev, notify_id,
|
2017-04-13 18:17:39 +08:00
|
|
|
cookie, ret);
|
2016-08-12 22:40:02 +08:00
|
|
|
else
|
|
|
|
rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
|
|
|
|
break;
|
|
|
|
case RBD_NOTIFY_OP_HEADER_UPDATE:
|
|
|
|
ret = rbd_dev_refresh(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
rbd_warn(rbd_dev, "refresh failed: %d", ret);
|
|
|
|
|
|
|
|
rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
if (rbd_is_lock_owner(rbd_dev))
|
|
|
|
rbd_acknowledge_notify_result(rbd_dev, notify_id,
|
|
|
|
cookie, -EOPNOTSUPP);
|
|
|
|
else
|
|
|
|
rbd_acknowledge_notify(rbd_dev, notify_id, cookie);
|
|
|
|
break;
|
|
|
|
}
|
2012-12-01 07:53:04 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
static void __rbd_unregister_watch(struct rbd_device *rbd_dev);
|
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
static void rbd_watch_errcb(void *arg, u64 cookie, int err)
|
2014-06-19 15:38:14 +08:00
|
|
|
{
|
2016-05-26 07:15:02 +08:00
|
|
|
struct rbd_device *rbd_dev = arg;
|
2014-06-19 15:38:14 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
rbd_warn(rbd_dev, "encountered watch error: %d", err);
|
2014-06-19 15:38:14 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
|
|
|
if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) {
|
|
|
|
__rbd_unregister_watch(rbd_dev);
|
|
|
|
rbd_dev->watch_state = RBD_WATCH_STATE_ERROR;
|
2014-06-19 15:38:14 +08:00
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0);
|
2014-06-19 15:38:14 +08:00
|
|
|
}
|
2016-08-12 22:11:41 +08:00
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
2014-06-19 15:38:14 +08:00
|
|
|
}
|
|
|
|
|
2013-01-19 02:31:10 +08:00
|
|
|
/*
|
2016-08-12 22:11:41 +08:00
|
|
|
* watch_mutex must be locked
|
2013-01-19 02:31:10 +08:00
|
|
|
*/
|
2016-08-12 22:11:41 +08:00
|
|
|
static int __rbd_register_watch(struct rbd_device *rbd_dev)
|
2013-01-19 02:31:10 +08:00
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_osd_linger_request *handle;
|
2013-01-19 02:31:10 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
rbd_assert(!rbd_dev->watch_handle);
|
2016-08-12 22:11:41 +08:00
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
2013-01-19 02:31:10 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, rbd_watch_cb,
|
|
|
|
rbd_watch_errcb, rbd_dev);
|
|
|
|
if (IS_ERR(handle))
|
|
|
|
return PTR_ERR(handle);
|
2013-01-26 07:08:55 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
rbd_dev->watch_handle = handle;
|
2014-05-22 23:28:52 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
/*
|
|
|
|
* watch_mutex must be locked
|
|
|
|
*/
|
|
|
|
static void __rbd_unregister_watch(struct rbd_device *rbd_dev)
|
2014-05-22 23:28:52 +08:00
|
|
|
{
|
2016-05-26 07:15:02 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
int ret;
|
2014-05-22 23:28:52 +08:00
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
rbd_assert(rbd_dev->watch_handle);
|
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
2014-05-22 23:28:52 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
|
|
|
|
if (ret)
|
|
|
|
rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
|
2014-06-20 22:29:20 +08:00
|
|
|
|
2016-05-26 07:15:02 +08:00
|
|
|
rbd_dev->watch_handle = NULL;
|
2016-04-28 22:07:26 +08:00
|
|
|
}
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
static int rbd_register_watch(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
|
|
|
rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED);
|
|
|
|
ret = __rbd_register_watch(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
|
|
|
|
rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
|
|
|
|
|
|
|
|
out:
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cancel_tasks_sync(struct rbd_device *rbd_dev)
|
2016-04-28 22:07:26 +08:00
|
|
|
{
|
2016-08-12 22:11:41 +08:00
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
cancel_work_sync(&rbd_dev->acquired_lock_work);
|
|
|
|
cancel_work_sync(&rbd_dev->released_lock_work);
|
|
|
|
cancel_delayed_work_sync(&rbd_dev->lock_dwork);
|
|
|
|
cancel_work_sync(&rbd_dev->unlock_work);
|
2016-08-12 22:11:41 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_unregister_watch(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2016-08-12 22:40:02 +08:00
|
|
|
WARN_ON(waitqueue_active(&rbd_dev->lock_waitq));
|
2016-08-12 22:11:41 +08:00
|
|
|
cancel_tasks_sync(rbd_dev);
|
|
|
|
|
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
|
|
|
if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED)
|
|
|
|
__rbd_unregister_watch(rbd_dev);
|
|
|
|
rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
|
2018-06-04 18:24:37 +08:00
|
|
|
cancel_delayed_work_sync(&rbd_dev->watch_dwork);
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
|
2013-12-17 00:02:40 +08:00
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for write
|
|
|
|
*/
|
|
|
|
static void rbd_reacquire_lock(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
char cookie[32];
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
|
|
|
|
|
|
|
|
format_lock_cookie(rbd_dev, cookie);
|
|
|
|
ret = ceph_cls_set_cookie(osdc, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, RBD_LOCK_NAME,
|
|
|
|
CEPH_CLS_LOCK_EXCLUSIVE, rbd_dev->lock_cookie,
|
|
|
|
RBD_LOCK_TAG, cookie);
|
|
|
|
if (ret) {
|
|
|
|
if (ret != -EOPNOTSUPP)
|
|
|
|
rbd_warn(rbd_dev, "failed to update lock cookie: %d",
|
|
|
|
ret);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Lock cookie cannot be updated on older OSDs, so do
|
|
|
|
* a manual release and queue an acquire.
|
|
|
|
*/
|
|
|
|
if (rbd_release_lock(rbd_dev))
|
|
|
|
queue_delayed_work(rbd_dev->task_wq,
|
|
|
|
&rbd_dev->lock_dwork, 0);
|
|
|
|
} else {
|
2017-12-13 23:43:59 +08:00
|
|
|
__rbd_lock(rbd_dev, cookie);
|
2017-04-13 18:17:38 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
static void rbd_reregister_watch(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
|
|
|
|
struct rbd_device, watch_dwork);
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
dout("%s rbd_dev %p\n", __func__, rbd_dev);
|
|
|
|
|
|
|
|
mutex_lock(&rbd_dev->watch_mutex);
|
2016-09-29 19:41:05 +08:00
|
|
|
if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) {
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
2017-04-13 18:17:38 +08:00
|
|
|
return;
|
2016-09-29 19:41:05 +08:00
|
|
|
}
|
2016-08-12 22:11:41 +08:00
|
|
|
|
|
|
|
ret = __rbd_register_watch(rbd_dev);
|
|
|
|
if (ret) {
|
|
|
|
rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
|
2016-09-29 20:23:12 +08:00
|
|
|
if (ret == -EBLACKLISTED || ret == -ENOENT) {
|
2016-09-29 19:41:05 +08:00
|
|
|
set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags);
|
2017-04-13 18:17:38 +08:00
|
|
|
wake_requests(rbd_dev, true);
|
2016-09-29 19:41:05 +08:00
|
|
|
} else {
|
2016-08-12 22:11:41 +08:00
|
|
|
queue_delayed_work(rbd_dev->task_wq,
|
|
|
|
&rbd_dev->watch_dwork,
|
|
|
|
RBD_RETRY_DELAY);
|
2016-09-29 19:41:05 +08:00
|
|
|
}
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
2017-04-13 18:17:38 +08:00
|
|
|
return;
|
2016-08-12 22:11:41 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED;
|
|
|
|
rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id;
|
|
|
|
mutex_unlock(&rbd_dev->watch_mutex);
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
|
|
|
|
rbd_reacquire_lock(rbd_dev);
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
ret = rbd_dev_refresh(rbd_dev);
|
|
|
|
if (ret)
|
2018-03-19 21:33:10 +08:00
|
|
|
rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret);
|
2016-08-12 22:11:41 +08:00
|
|
|
}
|
|
|
|
|
2013-01-19 14:30:28 +08:00
|
|
|
/*
|
2013-04-26 04:09:42 +08:00
|
|
|
* Synchronous osd object method call. Returns the number of bytes
|
|
|
|
* returned in the outbound buffer, or a negative error code.
|
2013-01-19 14:30:28 +08:00
|
|
|
*/
|
|
|
|
static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
|
2017-01-26 01:16:21 +08:00
|
|
|
struct ceph_object_id *oid,
|
|
|
|
struct ceph_object_locator *oloc,
|
2013-01-19 14:30:28 +08:00
|
|
|
const char *method_name,
|
2013-04-22 01:14:45 +08:00
|
|
|
const void *outbound,
|
2013-01-19 14:30:28 +08:00
|
|
|
size_t outbound_size,
|
2013-04-22 01:14:45 +08:00
|
|
|
void *inbound,
|
2013-04-30 13:44:33 +08:00
|
|
|
size_t inbound_size)
|
2013-01-19 14:30:28 +08:00
|
|
|
{
|
2017-01-26 01:16:21 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
struct page *req_page = NULL;
|
|
|
|
struct page *reply_page;
|
2013-01-19 14:30:28 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
/*
|
2013-04-05 14:27:11 +08:00
|
|
|
* Method calls are ultimately read operations. The result
|
|
|
|
* should placed into the inbound buffer provided. They
|
|
|
|
* also supply outbound data--parameters for the object
|
|
|
|
* method. Currently if this is present it will be a
|
|
|
|
* snapshot id.
|
2013-01-19 14:30:28 +08:00
|
|
|
*/
|
2017-01-26 01:16:21 +08:00
|
|
|
if (outbound) {
|
|
|
|
if (outbound_size > PAGE_SIZE)
|
|
|
|
return -E2BIG;
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
req_page = alloc_page(GFP_KERNEL);
|
|
|
|
if (!req_page)
|
|
|
|
return -ENOMEM;
|
2013-04-06 03:46:02 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
memcpy(page_address(req_page), outbound, outbound_size);
|
2013-04-06 03:46:02 +08:00
|
|
|
}
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
reply_page = alloc_page(GFP_KERNEL);
|
|
|
|
if (!reply_page) {
|
|
|
|
if (req_page)
|
|
|
|
__free_page(req_page);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2013-04-22 01:14:45 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name,
|
|
|
|
CEPH_OSD_FLAG_READ, req_page, outbound_size,
|
|
|
|
reply_page, &inbound_size);
|
|
|
|
if (!ret) {
|
|
|
|
memcpy(inbound, page_address(reply_page), inbound_size);
|
|
|
|
ret = inbound_size;
|
|
|
|
}
|
2013-01-19 14:30:28 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
if (req_page)
|
|
|
|
__free_page(req_page);
|
|
|
|
__free_page(reply_page);
|
2013-01-19 14:30:28 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
/*
|
|
|
|
* lock_rwsem must be held for read
|
|
|
|
*/
|
2018-04-04 16:15:38 +08:00
|
|
|
static int rbd_wait_state_locked(struct rbd_device *rbd_dev, bool may_acquire)
|
2016-08-12 22:40:02 +08:00
|
|
|
{
|
|
|
|
DEFINE_WAIT(wait);
|
2018-03-26 22:22:55 +08:00
|
|
|
unsigned long timeout;
|
2018-04-04 16:15:38 +08:00
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags))
|
|
|
|
return -EBLACKLISTED;
|
|
|
|
|
|
|
|
if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!may_acquire) {
|
|
|
|
rbd_warn(rbd_dev, "exclusive lock required");
|
|
|
|
return -EROFS;
|
|
|
|
}
|
2016-08-12 22:40:02 +08:00
|
|
|
|
|
|
|
do {
|
|
|
|
/*
|
|
|
|
* Note the use of mod_delayed_work() in rbd_acquire_lock()
|
|
|
|
* and cancel_delayed_work() in wake_requests().
|
|
|
|
*/
|
|
|
|
dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev);
|
|
|
|
queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
|
|
|
|
prepare_to_wait_exclusive(&rbd_dev->lock_waitq, &wait,
|
|
|
|
TASK_UNINTERRUPTIBLE);
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
2018-03-26 22:22:55 +08:00
|
|
|
timeout = schedule_timeout(ceph_timeout_jiffies(
|
|
|
|
rbd_dev->opts->lock_timeout));
|
2016-08-12 22:40:02 +08:00
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
2018-04-04 16:15:38 +08:00
|
|
|
if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) {
|
|
|
|
ret = -EBLACKLISTED;
|
|
|
|
break;
|
|
|
|
}
|
2018-03-26 22:22:55 +08:00
|
|
|
if (!timeout) {
|
|
|
|
rbd_warn(rbd_dev, "timed out waiting for lock");
|
|
|
|
ret = -ETIMEDOUT;
|
|
|
|
break;
|
|
|
|
}
|
2018-04-04 16:15:38 +08:00
|
|
|
} while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED);
|
2016-09-29 19:41:05 +08:00
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
finish_wait(&rbd_dev->lock_waitq, &wait);
|
2018-04-04 16:15:38 +08:00
|
|
|
return ret;
|
2016-08-12 22:40:02 +08:00
|
|
|
}
|
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
static void rbd_queue_workfn(struct work_struct *work)
|
2012-11-22 14:00:08 +08:00
|
|
|
{
|
2015-01-14 00:20:04 +08:00
|
|
|
struct request *rq = blk_mq_rq_from_pdu(work);
|
|
|
|
struct rbd_device *rbd_dev = rq->q->queuedata;
|
2014-08-04 22:04:39 +08:00
|
|
|
struct rbd_img_request *img_request;
|
2014-04-09 02:12:11 +08:00
|
|
|
struct ceph_snap_context *snapc = NULL;
|
2014-08-04 22:04:39 +08:00
|
|
|
u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
|
|
|
|
u64 length = blk_rq_bytes(rq);
|
2014-03-13 11:21:35 +08:00
|
|
|
enum obj_operation_type op_type;
|
2014-04-09 02:12:11 +08:00
|
|
|
u64 mapping_size;
|
2016-09-20 20:23:17 +08:00
|
|
|
bool must_be_locked;
|
2012-11-22 14:00:08 +08:00
|
|
|
int result;
|
|
|
|
|
2017-01-31 23:57:31 +08:00
|
|
|
switch (req_op(rq)) {
|
|
|
|
case REQ_OP_DISCARD:
|
2014-04-01 22:22:16 +08:00
|
|
|
op_type = OBJ_OP_DISCARD;
|
2017-01-31 23:57:31 +08:00
|
|
|
break;
|
2019-01-29 19:46:25 +08:00
|
|
|
case REQ_OP_WRITE_ZEROES:
|
|
|
|
op_type = OBJ_OP_ZEROOUT;
|
|
|
|
break;
|
2017-01-31 23:57:31 +08:00
|
|
|
case REQ_OP_WRITE:
|
2014-03-13 11:21:35 +08:00
|
|
|
op_type = OBJ_OP_WRITE;
|
2017-01-31 23:57:31 +08:00
|
|
|
break;
|
|
|
|
case REQ_OP_READ:
|
2014-03-13 11:21:35 +08:00
|
|
|
op_type = OBJ_OP_READ;
|
2017-01-31 23:57:31 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
dout("%s: non-fs request type %d\n", __func__, req_op(rq));
|
|
|
|
result = -EIO;
|
|
|
|
goto err;
|
|
|
|
}
|
2014-03-13 11:21:35 +08:00
|
|
|
|
2014-08-04 22:04:39 +08:00
|
|
|
/* Ignore/skip any zero-length requests */
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2014-08-04 22:04:39 +08:00
|
|
|
if (!length) {
|
|
|
|
dout("%s: zero-length request\n", __func__);
|
|
|
|
result = 0;
|
|
|
|
goto err_rq;
|
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-03 23:27:03 +08:00
|
|
|
if (op_type != OBJ_OP_READ && rbd_dev->spec->snap_id != CEPH_NOSNAP) {
|
|
|
|
rbd_warn(rbd_dev, "%s on read-only snapshot",
|
|
|
|
obj_op_name(op_type));
|
|
|
|
result = -EIO;
|
|
|
|
goto err;
|
|
|
|
}
|
2013-02-21 11:59:33 +08:00
|
|
|
|
2014-08-04 22:04:39 +08:00
|
|
|
/*
|
|
|
|
* Quit early if the mapped snapshot no longer exists. It's
|
|
|
|
* still possible the snapshot will have disappeared by the
|
|
|
|
* time our request arrives at the osd, but there's no sense in
|
|
|
|
* sending it if we already know.
|
|
|
|
*/
|
|
|
|
if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) {
|
|
|
|
dout("request for non-existent snapshot");
|
|
|
|
rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP);
|
|
|
|
result = -ENXIO;
|
|
|
|
goto err_rq;
|
|
|
|
}
|
2013-02-21 11:59:33 +08:00
|
|
|
|
2014-08-04 22:04:39 +08:00
|
|
|
if (offset && length > U64_MAX - offset + 1) {
|
|
|
|
rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset,
|
|
|
|
length);
|
|
|
|
result = -EINVAL;
|
|
|
|
goto err_rq; /* Shouldn't happen */
|
|
|
|
}
|
2013-02-21 11:59:33 +08:00
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
blk_mq_start_request(rq);
|
|
|
|
|
2014-04-09 02:12:11 +08:00
|
|
|
down_read(&rbd_dev->header_rwsem);
|
|
|
|
mapping_size = rbd_dev->mapping.size;
|
2014-03-13 11:21:35 +08:00
|
|
|
if (op_type != OBJ_OP_READ) {
|
2014-04-09 02:12:11 +08:00
|
|
|
snapc = rbd_dev->header.snapc;
|
|
|
|
ceph_get_snap_context(snapc);
|
|
|
|
}
|
|
|
|
up_read(&rbd_dev->header_rwsem);
|
|
|
|
|
|
|
|
if (offset + length > mapping_size) {
|
2014-08-04 22:04:39 +08:00
|
|
|
rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset,
|
2014-04-09 02:12:11 +08:00
|
|
|
length, mapping_size);
|
2014-08-04 22:04:39 +08:00
|
|
|
result = -EIO;
|
|
|
|
goto err_rq;
|
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
must_be_locked =
|
|
|
|
(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
|
|
|
|
(op_type != OBJ_OP_READ || rbd_dev->opts->lock_on_read);
|
2016-08-12 22:40:02 +08:00
|
|
|
if (must_be_locked) {
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
2018-04-04 16:15:38 +08:00
|
|
|
result = rbd_wait_state_locked(rbd_dev,
|
|
|
|
!rbd_dev->opts->exclusive);
|
|
|
|
if (result)
|
2016-09-29 19:41:05 +08:00
|
|
|
goto err_unlock;
|
2016-08-12 22:40:02 +08:00
|
|
|
}
|
|
|
|
|
2018-02-07 02:26:35 +08:00
|
|
|
img_request = rbd_img_request_create(rbd_dev, op_type, snapc);
|
2014-08-04 22:04:39 +08:00
|
|
|
if (!img_request) {
|
|
|
|
result = -ENOMEM;
|
2016-08-12 22:40:02 +08:00
|
|
|
goto err_unlock;
|
2014-08-04 22:04:39 +08:00
|
|
|
}
|
|
|
|
img_request->rq = rq;
|
2015-11-28 02:23:24 +08:00
|
|
|
snapc = NULL; /* img_request consumes a ref */
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-01-29 19:46:25 +08:00
|
|
|
if (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_ZEROOUT)
|
2018-02-07 02:26:34 +08:00
|
|
|
result = rbd_img_fill_nodata(img_request, offset, length);
|
2014-04-01 22:22:16 +08:00
|
|
|
else
|
2018-02-07 02:26:34 +08:00
|
|
|
result = rbd_img_fill_from_bio(img_request, offset, length,
|
|
|
|
rq->bio);
|
2019-05-16 21:06:56 +08:00
|
|
|
if (result)
|
2014-08-04 22:04:39 +08:00
|
|
|
goto err_img_request;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2019-05-16 21:06:56 +08:00
|
|
|
rbd_img_handle_request(img_request, 0);
|
2016-08-12 22:40:02 +08:00
|
|
|
if (must_be_locked)
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
2014-08-04 22:04:39 +08:00
|
|
|
return;
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2014-08-04 22:04:39 +08:00
|
|
|
err_img_request:
|
|
|
|
rbd_img_request_put(img_request);
|
2016-08-12 22:40:02 +08:00
|
|
|
err_unlock:
|
|
|
|
if (must_be_locked)
|
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
2014-08-04 22:04:39 +08:00
|
|
|
err_rq:
|
|
|
|
if (result)
|
|
|
|
rbd_warn(rbd_dev, "%s %llx at %llx result %d",
|
2014-03-13 11:21:35 +08:00
|
|
|
obj_op_name(op_type), length, offset, result);
|
2014-11-02 22:20:59 +08:00
|
|
|
ceph_put_snap_context(snapc);
|
2015-01-14 00:20:04 +08:00
|
|
|
err:
|
2017-06-03 15:38:04 +08:00
|
|
|
blk_mq_end_request(rq, errno_to_blk_status(result));
|
2014-08-04 22:04:39 +08:00
|
|
|
}
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2017-06-03 15:38:05 +08:00
|
|
|
static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx,
|
2015-01-14 00:20:04 +08:00
|
|
|
const struct blk_mq_queue_data *bd)
|
2014-08-04 22:04:39 +08:00
|
|
|
{
|
2015-01-14 00:20:04 +08:00
|
|
|
struct request *rq = bd->rq;
|
|
|
|
struct work_struct *work = blk_mq_rq_to_pdu(rq);
|
2012-11-22 14:00:08 +08:00
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
queue_work(rbd_wq, work);
|
2017-06-03 15:38:05 +08:00
|
|
|
return BLK_STS_OK;
|
2012-11-22 14:00:08 +08:00
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
static void rbd_free_disk(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2017-04-13 18:17:38 +08:00
|
|
|
blk_cleanup_queue(rbd_dev->disk->queue);
|
|
|
|
blk_mq_free_tag_set(&rbd_dev->tag_set);
|
|
|
|
put_disk(rbd_dev->disk);
|
2013-04-26 12:15:08 +08:00
|
|
|
rbd_dev->disk = NULL;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-01-18 02:25:27 +08:00
|
|
|
static int rbd_obj_read_sync(struct rbd_device *rbd_dev,
|
2017-01-26 01:16:21 +08:00
|
|
|
struct ceph_object_id *oid,
|
|
|
|
struct ceph_object_locator *oloc,
|
|
|
|
void *buf, int buf_len)
|
2013-01-18 02:25:27 +08:00
|
|
|
|
|
|
|
{
|
2017-01-26 01:16:21 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
struct ceph_osd_request *req;
|
|
|
|
struct page **pages;
|
|
|
|
int num_pages = calc_pages_for(0, buf_len);
|
2013-01-18 02:25:27 +08:00
|
|
|
int ret;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL);
|
|
|
|
if (!req)
|
|
|
|
return -ENOMEM;
|
2013-01-18 02:25:27 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ceph_oid_copy(&req->r_base_oid, oid);
|
|
|
|
ceph_oloc_copy(&req->r_base_oloc, oloc);
|
|
|
|
req->r_flags = CEPH_OSD_FLAG_READ;
|
2013-04-04 10:32:51 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
|
|
|
|
if (IS_ERR(pages)) {
|
|
|
|
ret = PTR_ERR(pages);
|
|
|
|
goto out_req;
|
|
|
|
}
|
2013-02-07 03:11:38 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0);
|
|
|
|
osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false,
|
|
|
|
true);
|
|
|
|
|
2018-10-15 22:11:37 +08:00
|
|
|
ret = ceph_osdc_alloc_messages(req, GFP_KERNEL);
|
|
|
|
if (ret)
|
|
|
|
goto out_req;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ceph_osdc_start_request(osdc, req, false);
|
|
|
|
ret = ceph_osdc_wait_request(osdc, req);
|
|
|
|
if (ret >= 0)
|
|
|
|
ceph_copy_from_page_vector(pages, buf, 0, ret);
|
2013-01-18 02:25:27 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
out_req:
|
|
|
|
ceph_osdc_put_request(req);
|
2013-01-18 02:25:27 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
2013-05-06 22:51:29 +08:00
|
|
|
* Read the complete header for the given rbd device. On successful
|
|
|
|
* return, the rbd_dev->header field will contain up-to-date
|
|
|
|
* information about the image.
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
2013-05-06 22:51:30 +08:00
|
|
|
static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2012-08-03 00:29:46 +08:00
|
|
|
struct rbd_image_header_ondisk *ondisk = NULL;
|
2012-04-21 04:49:44 +08:00
|
|
|
u32 snap_count = 0;
|
2012-08-03 00:29:46 +08:00
|
|
|
u64 names_size = 0;
|
|
|
|
u32 want_count;
|
|
|
|
int ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-02-08 02:03:36 +08:00
|
|
|
/*
|
2012-08-03 00:29:46 +08:00
|
|
|
* The complete header will include an array of its 64-bit
|
|
|
|
* snapshot ids, followed by the names of those snapshots as
|
|
|
|
* a contiguous block of NUL-terminated strings. Note that
|
|
|
|
* the number of snapshots could change by the time we read
|
|
|
|
* it in, in which case we re-read it.
|
2012-02-08 02:03:36 +08:00
|
|
|
*/
|
2012-08-03 00:29:46 +08:00
|
|
|
do {
|
|
|
|
size_t size;
|
|
|
|
|
|
|
|
kfree(ondisk);
|
|
|
|
|
|
|
|
size = sizeof (*ondisk);
|
|
|
|
size += snap_count * sizeof (struct rbd_image_snap_ondisk);
|
|
|
|
size += names_size;
|
|
|
|
ondisk = kmalloc(size, GFP_KERNEL);
|
|
|
|
if (!ondisk)
|
2013-05-06 22:51:29 +08:00
|
|
|
return -ENOMEM;
|
2012-08-03 00:29:46 +08:00
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, ondisk, size);
|
2012-08-03 00:29:46 +08:00
|
|
|
if (ret < 0)
|
2013-05-06 22:51:29 +08:00
|
|
|
goto out;
|
2013-04-26 22:43:47 +08:00
|
|
|
if ((size_t)ret < size) {
|
2012-08-03 00:29:46 +08:00
|
|
|
ret = -ENXIO;
|
2012-11-01 23:17:15 +08:00
|
|
|
rbd_warn(rbd_dev, "short header read (want %zd got %d)",
|
|
|
|
size, ret);
|
2013-05-06 22:51:29 +08:00
|
|
|
goto out;
|
2012-08-03 00:29:46 +08:00
|
|
|
}
|
|
|
|
if (!rbd_dev_ondisk_valid(ondisk)) {
|
|
|
|
ret = -ENXIO;
|
2012-11-01 23:17:15 +08:00
|
|
|
rbd_warn(rbd_dev, "invalid header");
|
2013-05-06 22:51:29 +08:00
|
|
|
goto out;
|
2011-11-16 06:49:53 +08:00
|
|
|
}
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-08-03 00:29:46 +08:00
|
|
|
names_size = le64_to_cpu(ondisk->snap_names_len);
|
|
|
|
want_count = snap_count;
|
|
|
|
snap_count = le32_to_cpu(ondisk->snap_count);
|
|
|
|
} while (snap_count != want_count);
|
2012-02-08 02:03:36 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
ret = rbd_header_from_disk(rbd_dev, ondisk);
|
|
|
|
out:
|
2012-08-03 00:29:46 +08:00
|
|
|
kfree(ondisk);
|
|
|
|
|
|
|
|
return ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
/*
|
|
|
|
* Clear the rbd device's EXISTS flag if the snapshot it's mapped to
|
|
|
|
* has disappeared from the (just updated) snapshot context.
|
|
|
|
*/
|
|
|
|
static void rbd_exists_validate(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
u64 snap_id;
|
|
|
|
|
|
|
|
if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags))
|
|
|
|
return;
|
|
|
|
|
|
|
|
snap_id = rbd_dev->spec->snap_id;
|
|
|
|
if (snap_id == CEPH_NOSNAP)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX)
|
|
|
|
clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
|
|
|
|
}
|
|
|
|
|
2013-08-30 08:26:31 +08:00
|
|
|
static void rbd_dev_update_size(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
sector_t size;
|
|
|
|
|
|
|
|
/*
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
* If EXISTS is not set, rbd_dev->disk may be NULL, so don't
|
|
|
|
* try to update its size. If REMOVING is set, updating size
|
|
|
|
* is just useless work since the device can't be opened.
|
2013-08-30 08:26:31 +08:00
|
|
|
*/
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) &&
|
|
|
|
!test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) {
|
2013-08-30 08:26:31 +08:00
|
|
|
size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
|
|
|
|
dout("setting size to %llu sectors", (unsigned long long)size);
|
|
|
|
set_capacity(rbd_dev->disk, size);
|
|
|
|
revalidate_disk(rbd_dev->disk);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
|
2012-07-25 22:32:41 +08:00
|
|
|
{
|
2013-05-06 20:40:30 +08:00
|
|
|
u64 mapping_size;
|
2012-07-25 22:32:41 +08:00
|
|
|
int ret;
|
|
|
|
|
2013-06-01 06:40:45 +08:00
|
|
|
down_write(&rbd_dev->header_rwsem);
|
2013-05-30 00:18:59 +08:00
|
|
|
mapping_size = rbd_dev->mapping.size;
|
2014-07-23 21:11:19 +08:00
|
|
|
|
|
|
|
ret = rbd_dev_header_info(rbd_dev);
|
2014-07-23 21:11:20 +08:00
|
|
|
if (ret)
|
2015-01-09 01:18:22 +08:00
|
|
|
goto out;
|
2013-05-02 01:43:03 +08:00
|
|
|
|
2014-07-24 14:42:13 +08:00
|
|
|
/*
|
|
|
|
* If there is a parent, see if it has disappeared due to the
|
|
|
|
* mapped image getting flattened.
|
|
|
|
*/
|
|
|
|
if (rbd_dev->parent) {
|
|
|
|
ret = rbd_dev_v2_parent_info(rbd_dev);
|
|
|
|
if (ret)
|
2015-01-09 01:18:22 +08:00
|
|
|
goto out;
|
2014-07-24 14:42:13 +08:00
|
|
|
}
|
|
|
|
|
2014-07-23 21:11:21 +08:00
|
|
|
if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
|
2015-01-09 01:18:22 +08:00
|
|
|
rbd_dev->mapping.size = rbd_dev->header.image_size;
|
2014-07-23 21:11:21 +08:00
|
|
|
} else {
|
|
|
|
/* validate mapped snapshot's EXISTS flag */
|
|
|
|
rbd_exists_validate(rbd_dev);
|
|
|
|
}
|
2013-05-02 01:43:03 +08:00
|
|
|
|
2015-01-09 01:18:22 +08:00
|
|
|
out:
|
2013-06-01 06:40:45 +08:00
|
|
|
up_write(&rbd_dev->header_rwsem);
|
2015-01-09 01:18:22 +08:00
|
|
|
if (!ret && mapping_size != rbd_dev->mapping.size)
|
2013-08-30 08:26:31 +08:00
|
|
|
rbd_dev_update_size(rbd_dev);
|
2012-07-25 22:32:41 +08:00
|
|
|
|
2015-01-09 01:18:22 +08:00
|
|
|
return ret;
|
2012-07-25 22:32:41 +08:00
|
|
|
}
|
|
|
|
|
2017-05-02 00:19:08 +08:00
|
|
|
static int rbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
|
|
|
|
unsigned int hctx_idx, unsigned int numa_node)
|
2015-01-14 00:20:04 +08:00
|
|
|
{
|
|
|
|
struct work_struct *work = blk_mq_rq_to_pdu(rq);
|
|
|
|
|
|
|
|
INIT_WORK(work, rbd_queue_workfn);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2017-03-31 04:39:16 +08:00
|
|
|
static const struct blk_mq_ops rbd_mq_ops = {
|
2015-01-14 00:20:04 +08:00
|
|
|
.queue_rq = rbd_queue_rq,
|
|
|
|
.init_request = rbd_init_request,
|
|
|
|
};
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
static int rbd_init_disk(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct gendisk *disk;
|
|
|
|
struct request_queue *q;
|
2018-04-16 15:32:18 +08:00
|
|
|
unsigned int objset_bytes =
|
|
|
|
rbd_dev->layout.object_size * rbd_dev->layout.stripe_count;
|
2015-01-14 00:20:04 +08:00
|
|
|
int err;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
/* create gendisk info */
|
2013-12-17 01:26:32 +08:00
|
|
|
disk = alloc_disk(single_major ?
|
|
|
|
(1 << RBD_SINGLE_MAJOR_PART_SHIFT) :
|
|
|
|
RBD_MINORS_PER_MAJOR);
|
2010-08-13 07:11:25 +08:00
|
|
|
if (!disk)
|
2012-08-30 06:11:06 +08:00
|
|
|
return -ENOMEM;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-01-30 03:57:44 +08:00
|
|
|
snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d",
|
2012-07-04 05:01:19 +08:00
|
|
|
rbd_dev->dev_id);
|
2010-08-13 07:11:25 +08:00
|
|
|
disk->major = rbd_dev->major;
|
2013-12-13 21:28:57 +08:00
|
|
|
disk->first_minor = rbd_dev->minor;
|
2013-12-17 01:26:32 +08:00
|
|
|
if (single_major)
|
|
|
|
disk->flags |= GENHD_FL_EXT_DEVT;
|
2010-08-13 07:11:25 +08:00
|
|
|
disk->fops = &rbd_bd_ops;
|
|
|
|
disk->private_data = rbd_dev;
|
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set));
|
|
|
|
rbd_dev->tag_set.ops = &rbd_mq_ops;
|
2015-06-23 21:21:19 +08:00
|
|
|
rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth;
|
2015-01-14 00:20:04 +08:00
|
|
|
rbd_dev->tag_set.numa_node = NUMA_NO_NODE;
|
2019-02-15 19:13:24 +08:00
|
|
|
rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
|
2015-01-14 00:20:04 +08:00
|
|
|
rbd_dev->tag_set.nr_hw_queues = 1;
|
|
|
|
rbd_dev->tag_set.cmd_size = sizeof(struct work_struct);
|
|
|
|
|
|
|
|
err = blk_mq_alloc_tag_set(&rbd_dev->tag_set);
|
|
|
|
if (err)
|
2010-08-13 07:11:25 +08:00
|
|
|
goto out_disk;
|
2011-07-23 02:35:23 +08:00
|
|
|
|
2015-01-14 00:20:04 +08:00
|
|
|
q = blk_mq_init_queue(&rbd_dev->tag_set);
|
|
|
|
if (IS_ERR(q)) {
|
|
|
|
err = PTR_ERR(q);
|
|
|
|
goto out_tag_set;
|
|
|
|
}
|
|
|
|
|
2018-03-08 09:10:10 +08:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
|
2015-03-24 21:15:17 +08:00
|
|
|
/* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */
|
2012-02-08 02:03:37 +08:00
|
|
|
|
2018-04-16 15:32:18 +08:00
|
|
|
blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT);
|
2015-10-07 22:09:35 +08:00
|
|
|
q->limits.max_sectors = queue_max_hw_sectors(q);
|
2017-12-21 22:35:11 +08:00
|
|
|
blk_queue_max_segments(q, USHRT_MAX);
|
2018-01-13 00:22:10 +08:00
|
|
|
blk_queue_max_segment_size(q, UINT_MAX);
|
2019-03-15 21:50:04 +08:00
|
|
|
blk_queue_io_min(q, rbd_dev->opts->alloc_size);
|
|
|
|
blk_queue_io_opt(q, rbd_dev->opts->alloc_size);
|
2011-07-23 02:35:23 +08:00
|
|
|
|
2018-03-23 13:14:47 +08:00
|
|
|
if (rbd_dev->opts->trim) {
|
|
|
|
blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
|
2019-03-15 21:50:04 +08:00
|
|
|
q->limits.discard_granularity = rbd_dev->opts->alloc_size;
|
2018-03-23 13:14:47 +08:00
|
|
|
blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT);
|
|
|
|
blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT);
|
|
|
|
}
|
2014-04-01 22:22:16 +08:00
|
|
|
|
2015-10-16 02:50:46 +08:00
|
|
|
if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC))
|
2017-02-02 22:56:50 +08:00
|
|
|
q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
|
2015-10-16 02:50:46 +08:00
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
/*
|
|
|
|
* disk_release() expects a queue ref from add_disk() and will
|
|
|
|
* put it. Hold an extra ref until add_disk() is called.
|
|
|
|
*/
|
|
|
|
WARN_ON(!blk_get_queue(q));
|
2010-08-13 07:11:25 +08:00
|
|
|
disk->queue = q;
|
|
|
|
q->queuedata = rbd_dev;
|
|
|
|
|
|
|
|
rbd_dev->disk = disk;
|
|
|
|
|
|
|
|
return 0;
|
2015-01-14 00:20:04 +08:00
|
|
|
out_tag_set:
|
|
|
|
blk_mq_free_tag_set(&rbd_dev->tag_set);
|
2010-08-13 07:11:25 +08:00
|
|
|
out_disk:
|
|
|
|
put_disk(disk);
|
2015-01-14 00:20:04 +08:00
|
|
|
return err;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
/*
|
|
|
|
sysfs
|
|
|
|
*/
|
|
|
|
|
2012-02-08 02:03:37 +08:00
|
|
|
static struct rbd_device *dev_to_rbd_dev(struct device *dev)
|
|
|
|
{
|
|
|
|
return container_of(dev, struct rbd_device, dev);
|
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_size_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2011-12-06 02:35:04 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
return sprintf(buf, "%llu\n",
|
|
|
|
(unsigned long long)rbd_dev->mapping.size);
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2012-07-14 09:35:12 +08:00
|
|
|
/*
|
|
|
|
* Note this shows the features for whatever's mapped, which is not
|
|
|
|
* necessarily the base image.
|
|
|
|
*/
|
|
|
|
static ssize_t rbd_features_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
|
|
|
return sprintf(buf, "0x%016llx\n",
|
2013-04-27 04:44:36 +08:00
|
|
|
(unsigned long long)rbd_dev->mapping.features);
|
2012-07-14 09:35:12 +08:00
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_major_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
if (rbd_dev->major)
|
|
|
|
return sprintf(buf, "%d\n", rbd_dev->major);
|
|
|
|
|
|
|
|
return sprintf(buf, "(none)\n");
|
2013-12-13 21:28:57 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t rbd_minor_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2013-04-27 04:44:36 +08:00
|
|
|
|
2013-12-13 21:28:57 +08:00
|
|
|
return sprintf(buf, "%d\n", rbd_dev->minor);
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2016-08-19 00:38:43 +08:00
|
|
|
static ssize_t rbd_client_addr_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
struct ceph_entity_addr *client_addr =
|
|
|
|
ceph_client_addr(rbd_dev->rbd_client->client);
|
|
|
|
|
|
|
|
return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr,
|
|
|
|
le32_to_cpu(client_addr->nonce));
|
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_client_id_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2012-01-25 00:08:37 +08:00
|
|
|
return sprintf(buf, "client%lld\n",
|
2016-08-12 20:59:58 +08:00
|
|
|
ceph_client_gid(rbd_dev->rbd_client->client));
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2016-08-19 00:38:43 +08:00
|
|
|
static ssize_t rbd_cluster_fsid_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
|
|
|
return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid);
|
|
|
|
}
|
|
|
|
|
2016-08-19 00:38:45 +08:00
|
|
|
static ssize_t rbd_config_info_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
|
|
|
return sprintf(buf, "%s\n", rbd_dev->config_info);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_pool_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return sprintf(buf, "%s\n", rbd_dev->spec->pool_name);
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2012-07-12 23:46:35 +08:00
|
|
|
static ssize_t rbd_pool_id_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return sprintf(buf, "%llu\n",
|
2013-04-27 04:44:36 +08:00
|
|
|
(unsigned long long) rbd_dev->spec->pool_id);
|
2012-07-12 23:46:35 +08:00
|
|
|
}
|
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
static ssize_t rbd_pool_ns_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
|
|
|
return sprintf(buf, "%s\n", rbd_dev->spec->pool_ns ?: "");
|
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_name_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
if (rbd_dev->spec->image_name)
|
|
|
|
return sprintf(buf, "%s\n", rbd_dev->spec->image_name);
|
|
|
|
|
|
|
|
return sprintf(buf, "(unknown)\n");
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
static ssize_t rbd_image_id_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return sprintf(buf, "%s\n", rbd_dev->spec->image_id);
|
2012-07-11 09:30:11 +08:00
|
|
|
}
|
|
|
|
|
2012-07-14 09:35:12 +08:00
|
|
|
/*
|
|
|
|
* Shows the name of the currently-mapped snapshot (or
|
|
|
|
* RBD_SNAP_HEAD_NAME for the base image).
|
|
|
|
*/
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_snap_show(struct device *dev,
|
|
|
|
struct device_attribute *attr,
|
|
|
|
char *buf)
|
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return sprintf(buf, "%s\n", rbd_dev->spec->snap_name);
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
|
|
|
|
2016-08-19 00:38:44 +08:00
|
|
|
static ssize_t rbd_snap_id_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
|
|
|
|
return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id);
|
|
|
|
}
|
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
/*
|
2014-07-23 01:53:07 +08:00
|
|
|
* For a v2 image, shows the chain of parent images, separated by empty
|
|
|
|
* lines. For v1 images or if there is no parent, shows "(no parent
|
|
|
|
* image)".
|
2012-10-26 12:34:42 +08:00
|
|
|
*/
|
|
|
|
static ssize_t rbd_parent_show(struct device *dev,
|
2014-07-23 01:53:07 +08:00
|
|
|
struct device_attribute *attr,
|
|
|
|
char *buf)
|
2012-10-26 12:34:42 +08:00
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2014-07-23 01:53:07 +08:00
|
|
|
ssize_t count = 0;
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2014-07-23 01:53:07 +08:00
|
|
|
if (!rbd_dev->parent)
|
2012-10-26 12:34:42 +08:00
|
|
|
return sprintf(buf, "(no parent image)\n");
|
|
|
|
|
2014-07-23 01:53:07 +08:00
|
|
|
for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) {
|
|
|
|
struct rbd_spec *spec = rbd_dev->parent_spec;
|
|
|
|
|
|
|
|
count += sprintf(&buf[count], "%s"
|
|
|
|
"pool_id %llu\npool_name %s\n"
|
2018-08-22 23:26:10 +08:00
|
|
|
"pool_ns %s\n"
|
2014-07-23 01:53:07 +08:00
|
|
|
"image_id %s\nimage_name %s\n"
|
|
|
|
"snap_id %llu\nsnap_name %s\n"
|
|
|
|
"overlap %llu\n",
|
|
|
|
!count ? "" : "\n", /* first? */
|
|
|
|
spec->pool_id, spec->pool_name,
|
2018-08-22 23:26:10 +08:00
|
|
|
spec->pool_ns ?: "",
|
2014-07-23 01:53:07 +08:00
|
|
|
spec->image_id, spec->image_name ?: "(unknown)",
|
|
|
|
spec->snap_id, spec->snap_name,
|
|
|
|
rbd_dev->parent_overlap);
|
|
|
|
}
|
|
|
|
|
|
|
|
return count;
|
2012-10-26 12:34:42 +08:00
|
|
|
}
|
|
|
|
|
2010-11-20 06:51:04 +08:00
|
|
|
static ssize_t rbd_image_refresh(struct device *dev,
|
|
|
|
struct device_attribute *attr,
|
|
|
|
const char *buf,
|
|
|
|
size_t size)
|
|
|
|
{
|
2012-02-08 02:03:37 +08:00
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
2012-07-25 22:32:41 +08:00
|
|
|
int ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
ret = rbd_dev_refresh(rbd_dev);
|
2013-05-06 20:40:30 +08:00
|
|
|
if (ret)
|
2014-07-23 21:11:20 +08:00
|
|
|
return ret;
|
2012-07-25 22:32:41 +08:00
|
|
|
|
2014-07-23 21:11:20 +08:00
|
|
|
return size;
|
2010-11-20 06:51:04 +08:00
|
|
|
}
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2018-05-25 03:38:59 +08:00
|
|
|
static DEVICE_ATTR(size, 0444, rbd_size_show, NULL);
|
|
|
|
static DEVICE_ATTR(features, 0444, rbd_features_show, NULL);
|
|
|
|
static DEVICE_ATTR(major, 0444, rbd_major_show, NULL);
|
|
|
|
static DEVICE_ATTR(minor, 0444, rbd_minor_show, NULL);
|
|
|
|
static DEVICE_ATTR(client_addr, 0444, rbd_client_addr_show, NULL);
|
|
|
|
static DEVICE_ATTR(client_id, 0444, rbd_client_id_show, NULL);
|
|
|
|
static DEVICE_ATTR(cluster_fsid, 0444, rbd_cluster_fsid_show, NULL);
|
|
|
|
static DEVICE_ATTR(config_info, 0400, rbd_config_info_show, NULL);
|
|
|
|
static DEVICE_ATTR(pool, 0444, rbd_pool_show, NULL);
|
|
|
|
static DEVICE_ATTR(pool_id, 0444, rbd_pool_id_show, NULL);
|
2018-07-03 21:28:43 +08:00
|
|
|
static DEVICE_ATTR(pool_ns, 0444, rbd_pool_ns_show, NULL);
|
2018-05-25 03:38:59 +08:00
|
|
|
static DEVICE_ATTR(name, 0444, rbd_name_show, NULL);
|
|
|
|
static DEVICE_ATTR(image_id, 0444, rbd_image_id_show, NULL);
|
|
|
|
static DEVICE_ATTR(refresh, 0200, NULL, rbd_image_refresh);
|
|
|
|
static DEVICE_ATTR(current_snap, 0444, rbd_snap_show, NULL);
|
|
|
|
static DEVICE_ATTR(snap_id, 0444, rbd_snap_id_show, NULL);
|
|
|
|
static DEVICE_ATTR(parent, 0444, rbd_parent_show, NULL);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
|
|
|
static struct attribute *rbd_attrs[] = {
|
|
|
|
&dev_attr_size.attr,
|
2012-07-14 09:35:12 +08:00
|
|
|
&dev_attr_features.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_major.attr,
|
2013-12-13 21:28:57 +08:00
|
|
|
&dev_attr_minor.attr,
|
2016-08-19 00:38:43 +08:00
|
|
|
&dev_attr_client_addr.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_client_id.attr,
|
2016-08-19 00:38:43 +08:00
|
|
|
&dev_attr_cluster_fsid.attr,
|
2016-08-19 00:38:45 +08:00
|
|
|
&dev_attr_config_info.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_pool.attr,
|
2012-07-12 23:46:35 +08:00
|
|
|
&dev_attr_pool_id.attr,
|
2018-07-03 21:28:43 +08:00
|
|
|
&dev_attr_pool_ns.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_name.attr,
|
2012-07-11 09:30:11 +08:00
|
|
|
&dev_attr_image_id.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_current_snap.attr,
|
2016-08-19 00:38:44 +08:00
|
|
|
&dev_attr_snap_id.attr,
|
2012-10-26 12:34:42 +08:00
|
|
|
&dev_attr_parent.attr,
|
2010-11-20 06:51:04 +08:00
|
|
|
&dev_attr_refresh.attr,
|
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct attribute_group rbd_attr_group = {
|
|
|
|
.attrs = rbd_attrs,
|
|
|
|
};
|
|
|
|
|
|
|
|
static const struct attribute_group *rbd_attr_groups[] = {
|
|
|
|
&rbd_attr_group,
|
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
2015-10-17 02:11:25 +08:00
|
|
|
static void rbd_dev_release(struct device *dev);
|
2010-11-20 06:51:04 +08:00
|
|
|
|
2017-02-11 14:44:38 +08:00
|
|
|
static const struct device_type rbd_device_type = {
|
2010-11-20 06:51:04 +08:00
|
|
|
.name = "rbd",
|
|
|
|
.groups = rbd_attr_groups,
|
2015-10-17 02:11:25 +08:00
|
|
|
.release = rbd_dev_release,
|
2010-11-20 06:51:04 +08:00
|
|
|
};
|
|
|
|
|
2012-10-27 06:25:24 +08:00
|
|
|
static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec)
|
|
|
|
{
|
|
|
|
kref_get(&spec->kref);
|
|
|
|
|
|
|
|
return spec;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_spec_free(struct kref *kref);
|
|
|
|
static void rbd_spec_put(struct rbd_spec *spec)
|
|
|
|
{
|
|
|
|
if (spec)
|
|
|
|
kref_put(&spec->kref, rbd_spec_free);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct rbd_spec *rbd_spec_alloc(void)
|
|
|
|
{
|
|
|
|
struct rbd_spec *spec;
|
|
|
|
|
|
|
|
spec = kzalloc(sizeof (*spec), GFP_KERNEL);
|
|
|
|
if (!spec)
|
|
|
|
return NULL;
|
2014-07-23 21:11:20 +08:00
|
|
|
|
|
|
|
spec->pool_id = CEPH_NOPOOL;
|
|
|
|
spec->snap_id = CEPH_NOSNAP;
|
2012-10-27 06:25:24 +08:00
|
|
|
kref_init(&spec->kref);
|
|
|
|
|
|
|
|
return spec;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_spec_free(struct kref *kref)
|
|
|
|
{
|
|
|
|
struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref);
|
|
|
|
|
|
|
|
kfree(spec->pool_name);
|
2018-07-03 21:28:43 +08:00
|
|
|
kfree(spec->pool_ns);
|
2012-10-27 06:25:24 +08:00
|
|
|
kfree(spec->image_id);
|
|
|
|
kfree(spec->image_name);
|
|
|
|
kfree(spec->snap_name);
|
|
|
|
kfree(spec);
|
|
|
|
}
|
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
static void rbd_dev_free(struct rbd_device *rbd_dev)
|
2015-10-16 23:09:24 +08:00
|
|
|
{
|
2016-08-12 22:11:41 +08:00
|
|
|
WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED);
|
2016-08-12 22:40:02 +08:00
|
|
|
WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED);
|
2015-10-16 23:09:24 +08:00
|
|
|
|
2016-04-30 02:01:25 +08:00
|
|
|
ceph_oid_destroy(&rbd_dev->header_oid);
|
2016-08-05 22:15:38 +08:00
|
|
|
ceph_oloc_destroy(&rbd_dev->header_oloc);
|
2016-08-19 00:38:45 +08:00
|
|
|
kfree(rbd_dev->config_info);
|
2016-04-30 02:01:25 +08:00
|
|
|
|
2015-10-16 23:09:24 +08:00
|
|
|
rbd_put_client(rbd_dev->rbd_client);
|
|
|
|
rbd_spec_put(rbd_dev->spec);
|
|
|
|
kfree(rbd_dev->opts);
|
|
|
|
kfree(rbd_dev);
|
2016-08-12 21:45:52 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_dev_release(struct device *dev)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = dev_to_rbd_dev(dev);
|
|
|
|
bool need_put = !!rbd_dev->opts;
|
|
|
|
|
|
|
|
if (need_put) {
|
|
|
|
destroy_workqueue(rbd_dev->task_wq);
|
|
|
|
ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
|
|
|
|
}
|
|
|
|
|
|
|
|
rbd_dev_free(rbd_dev);
|
2015-10-16 23:09:24 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* This is racy, but way better than putting module outside of
|
|
|
|
* the release callback. The race window is pretty small, so
|
|
|
|
* doing something similar to dm (dm-builtin.c) is overkill.
|
|
|
|
*/
|
|
|
|
if (need_put)
|
|
|
|
module_put(THIS_MODULE);
|
|
|
|
}
|
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
|
|
|
|
struct rbd_spec *spec)
|
2012-10-26 12:34:42 +08:00
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev;
|
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
|
2012-10-26 12:34:42 +08:00
|
|
|
if (!rbd_dev)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
spin_lock_init(&rbd_dev->lock);
|
|
|
|
INIT_LIST_HEAD(&rbd_dev->node);
|
|
|
|
init_rwsem(&rbd_dev->header_rwsem);
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
rbd_dev->header.data_pool_id = CEPH_NOPOOL;
|
2016-04-30 02:01:25 +08:00
|
|
|
ceph_oid_init(&rbd_dev->header_oid);
|
2017-01-26 01:16:21 +08:00
|
|
|
rbd_dev->header_oloc.pool = spec->pool_id;
|
2018-07-03 21:28:43 +08:00
|
|
|
if (spec->pool_ns) {
|
|
|
|
WARN_ON(!*spec->pool_ns);
|
|
|
|
rbd_dev->header_oloc.pool_ns =
|
|
|
|
ceph_find_or_create_string(spec->pool_ns,
|
|
|
|
strlen(spec->pool_ns));
|
|
|
|
}
|
2016-04-30 02:01:25 +08:00
|
|
|
|
2016-08-12 22:11:41 +08:00
|
|
|
mutex_init(&rbd_dev->watch_mutex);
|
|
|
|
rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED;
|
|
|
|
INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch);
|
|
|
|
|
2016-08-12 22:40:02 +08:00
|
|
|
init_rwsem(&rbd_dev->lock_rwsem);
|
|
|
|
rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
|
|
|
|
INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock);
|
|
|
|
INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock);
|
|
|
|
INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock);
|
|
|
|
INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work);
|
|
|
|
init_waitqueue_head(&rbd_dev->lock_waitq);
|
|
|
|
|
2015-10-16 23:09:24 +08:00
|
|
|
rbd_dev->dev.bus = &rbd_bus_type;
|
|
|
|
rbd_dev->dev.type = &rbd_device_type;
|
|
|
|
rbd_dev->dev.parent = &rbd_root_dev;
|
|
|
|
device_initialize(&rbd_dev->dev);
|
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
rbd_dev->rbd_client = rbdc;
|
2015-06-22 18:24:48 +08:00
|
|
|
rbd_dev->spec = spec;
|
2012-11-15 02:25:19 +08:00
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
return rbd_dev;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Create a mapping rbd_dev.
|
|
|
|
*/
|
|
|
|
static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
|
|
|
|
struct rbd_spec *spec,
|
|
|
|
struct rbd_options *opts)
|
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev;
|
|
|
|
|
|
|
|
rbd_dev = __rbd_dev_create(rbdc, spec);
|
|
|
|
if (!rbd_dev)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
rbd_dev->opts = opts;
|
|
|
|
|
|
|
|
/* get an id and fill in device name */
|
|
|
|
rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0,
|
|
|
|
minor_to_rbd_dev_id(1 << MINORBITS),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (rbd_dev->dev_id < 0)
|
|
|
|
goto fail_rbd_dev;
|
|
|
|
|
|
|
|
sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id);
|
|
|
|
rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM,
|
|
|
|
rbd_dev->name);
|
|
|
|
if (!rbd_dev->task_wq)
|
|
|
|
goto fail_dev_id;
|
2015-10-16 23:09:24 +08:00
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
/* we have a ref from do_rbd_add() */
|
|
|
|
__module_get(THIS_MODULE);
|
2015-10-16 23:09:24 +08:00
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
|
2012-10-26 12:34:42 +08:00
|
|
|
return rbd_dev;
|
2016-08-12 21:45:52 +08:00
|
|
|
|
|
|
|
fail_dev_id:
|
|
|
|
ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id);
|
|
|
|
fail_rbd_dev:
|
|
|
|
rbd_dev_free(rbd_dev);
|
|
|
|
return NULL;
|
2012-10-26 12:34:42 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_dev_destroy(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2015-10-16 23:09:24 +08:00
|
|
|
if (rbd_dev)
|
|
|
|
put_device(&rbd_dev->dev);
|
2012-10-26 12:34:42 +08:00
|
|
|
}
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
/*
|
|
|
|
* Get the size and object order for an image snapshot, or if
|
|
|
|
* snap_id is CEPH_NOSNAP, gets this information for the base
|
|
|
|
* image.
|
|
|
|
*/
|
|
|
|
static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u8 *order, u64 *snap_size)
|
|
|
|
{
|
|
|
|
__le64 snapid = cpu_to_le64(snap_id);
|
|
|
|
int ret;
|
|
|
|
struct {
|
|
|
|
u8 order;
|
|
|
|
__le64 size;
|
|
|
|
} __attribute__ ((packed)) size_buf = { 0 };
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_size",
|
|
|
|
&snapid, sizeof(snapid),
|
|
|
|
&size_buf, sizeof(size_buf));
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2012-07-04 05:01:19 +08:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
2013-04-22 01:14:45 +08:00
|
|
|
if (ret < sizeof (size_buf))
|
|
|
|
return -ERANGE;
|
2012-07-04 05:01:19 +08:00
|
|
|
|
2013-08-29 08:08:10 +08:00
|
|
|
if (order) {
|
2013-04-26 04:09:41 +08:00
|
|
|
*order = size_buf.order;
|
2013-08-29 08:08:10 +08:00
|
|
|
dout(" order %u", (unsigned int)*order);
|
|
|
|
}
|
2012-07-04 05:01:19 +08:00
|
|
|
*snap_size = le64_to_cpu(size_buf.size);
|
|
|
|
|
2013-08-29 08:08:10 +08:00
|
|
|
dout(" snap_id 0x%016llx snap_size = %llu\n",
|
|
|
|
(unsigned long long)snap_id,
|
2013-04-22 01:14:45 +08:00
|
|
|
(unsigned long long)*snap_size);
|
2012-07-04 05:01:19 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
|
|
|
|
&rbd_dev->header.obj_order,
|
|
|
|
&rbd_dev->header.image_size);
|
|
|
|
}
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
void *reply_buf;
|
|
|
|
int ret;
|
|
|
|
void *p;
|
|
|
|
|
|
|
|
reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL);
|
|
|
|
if (!reply_buf)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_object_prefix",
|
|
|
|
NULL, 0, reply_buf, RBD_OBJ_PREFIX_LEN_MAX);
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2012-07-04 05:01:19 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
p = reply_buf;
|
|
|
|
rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
|
2013-04-22 01:14:45 +08:00
|
|
|
p + ret, NULL, GFP_NOIO);
|
|
|
|
ret = 0;
|
2012-07-04 05:01:19 +08:00
|
|
|
|
|
|
|
if (IS_ERR(rbd_dev->header.object_prefix)) {
|
|
|
|
ret = PTR_ERR(rbd_dev->header.object_prefix);
|
|
|
|
rbd_dev->header.object_prefix = NULL;
|
|
|
|
} else {
|
|
|
|
dout(" object_prefix = %s\n", rbd_dev->header.object_prefix);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
kfree(reply_buf);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
|
|
|
|
u64 *snap_features)
|
|
|
|
{
|
|
|
|
__le64 snapid = cpu_to_le64(snap_id);
|
|
|
|
struct {
|
|
|
|
__le64 features;
|
|
|
|
__le64 incompat;
|
2013-04-22 01:14:45 +08:00
|
|
|
} __attribute__ ((packed)) features_buf = { 0 };
|
2016-04-13 20:15:50 +08:00
|
|
|
u64 unsup;
|
2012-07-04 05:01:19 +08:00
|
|
|
int ret;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_features",
|
|
|
|
&snapid, sizeof(snapid),
|
|
|
|
&features_buf, sizeof(features_buf));
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2012-07-04 05:01:19 +08:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
2013-04-22 01:14:45 +08:00
|
|
|
if (ret < sizeof (features_buf))
|
|
|
|
return -ERANGE;
|
2012-10-10 04:50:17 +08:00
|
|
|
|
2016-04-13 20:15:50 +08:00
|
|
|
unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED;
|
|
|
|
if (unsup) {
|
|
|
|
rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx",
|
|
|
|
unsup);
|
2012-11-01 21:39:26 +08:00
|
|
|
return -ENXIO;
|
2016-04-13 20:15:50 +08:00
|
|
|
}
|
2012-10-10 04:50:17 +08:00
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
*snap_features = le64_to_cpu(features_buf.features);
|
|
|
|
|
|
|
|
dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n",
|
2013-04-22 01:14:45 +08:00
|
|
|
(unsigned long long)snap_id,
|
|
|
|
(unsigned long long)*snap_features,
|
|
|
|
(unsigned long long)le64_to_cpu(features_buf.incompat));
|
2012-07-04 05:01:19 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
|
|
|
|
&rbd_dev->header.features);
|
|
|
|
}
|
|
|
|
|
2018-08-22 23:11:27 +08:00
|
|
|
struct parent_image_info {
|
|
|
|
u64 pool_id;
|
2018-08-22 23:26:10 +08:00
|
|
|
const char *pool_ns;
|
2018-08-22 23:11:27 +08:00
|
|
|
const char *image_id;
|
|
|
|
u64 snap_id;
|
|
|
|
|
2018-08-22 23:26:10 +08:00
|
|
|
bool has_overlap;
|
2018-08-22 23:11:27 +08:00
|
|
|
u64 overlap;
|
|
|
|
};
|
|
|
|
|
2018-08-22 23:26:10 +08:00
|
|
|
/*
|
|
|
|
* The caller is responsible for @pii.
|
|
|
|
*/
|
|
|
|
static int decode_parent_image_spec(void **p, void *end,
|
|
|
|
struct parent_image_info *pii)
|
|
|
|
{
|
|
|
|
u8 struct_v;
|
|
|
|
u32 struct_len;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = ceph_start_decoding(p, end, 1, "ParentImageSpec",
|
|
|
|
&struct_v, &struct_len);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
ceph_decode_64_safe(p, end, pii->pool_id, e_inval);
|
|
|
|
pii->pool_ns = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
|
|
|
|
if (IS_ERR(pii->pool_ns)) {
|
|
|
|
ret = PTR_ERR(pii->pool_ns);
|
|
|
|
pii->pool_ns = NULL;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
pii->image_id = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL);
|
|
|
|
if (IS_ERR(pii->image_id)) {
|
|
|
|
ret = PTR_ERR(pii->image_id);
|
|
|
|
pii->image_id = NULL;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
ceph_decode_64_safe(p, end, pii->snap_id, e_inval);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
e_inval:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __get_parent_info(struct rbd_device *rbd_dev,
|
|
|
|
struct page *req_page,
|
|
|
|
struct page *reply_page,
|
|
|
|
struct parent_image_info *pii)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
size_t reply_len = PAGE_SIZE;
|
|
|
|
void *p, *end;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
|
|
|
|
"rbd", "parent_get", CEPH_OSD_FLAG_READ,
|
|
|
|
req_page, sizeof(u64), reply_page, &reply_len);
|
|
|
|
if (ret)
|
|
|
|
return ret == -EOPNOTSUPP ? 1 : ret;
|
|
|
|
|
|
|
|
p = page_address(reply_page);
|
|
|
|
end = p + reply_len;
|
|
|
|
ret = decode_parent_image_spec(&p, end, pii);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
|
|
|
|
"rbd", "parent_overlap_get", CEPH_OSD_FLAG_READ,
|
|
|
|
req_page, sizeof(u64), reply_page, &reply_len);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
p = page_address(reply_page);
|
|
|
|
end = p + reply_len;
|
|
|
|
ceph_decode_8_safe(&p, end, pii->has_overlap, e_inval);
|
|
|
|
if (pii->has_overlap)
|
|
|
|
ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
e_inval:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2018-08-22 23:11:27 +08:00
|
|
|
/*
|
|
|
|
* The caller is responsible for @pii.
|
|
|
|
*/
|
|
|
|
static int __get_parent_info_legacy(struct rbd_device *rbd_dev,
|
|
|
|
struct page *req_page,
|
|
|
|
struct page *reply_page,
|
|
|
|
struct parent_image_info *pii)
|
|
|
|
{
|
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
size_t reply_len = PAGE_SIZE;
|
|
|
|
void *p, *end;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
|
|
|
|
"rbd", "get_parent", CEPH_OSD_FLAG_READ,
|
|
|
|
req_page, sizeof(u64), reply_page, &reply_len);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
p = page_address(reply_page);
|
|
|
|
end = p + reply_len;
|
|
|
|
ceph_decode_64_safe(&p, end, pii->pool_id, e_inval);
|
|
|
|
pii->image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
|
|
|
|
if (IS_ERR(pii->image_id)) {
|
|
|
|
ret = PTR_ERR(pii->image_id);
|
|
|
|
pii->image_id = NULL;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
ceph_decode_64_safe(&p, end, pii->snap_id, e_inval);
|
2018-08-22 23:26:10 +08:00
|
|
|
pii->has_overlap = true;
|
2018-08-22 23:11:27 +08:00
|
|
|
ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
e_inval:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int get_parent_info(struct rbd_device *rbd_dev,
|
|
|
|
struct parent_image_info *pii)
|
|
|
|
{
|
|
|
|
struct page *req_page, *reply_page;
|
|
|
|
void *p;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
req_page = alloc_page(GFP_KERNEL);
|
|
|
|
if (!req_page)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
reply_page = alloc_page(GFP_KERNEL);
|
|
|
|
if (!reply_page) {
|
|
|
|
__free_page(req_page);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
p = page_address(req_page);
|
|
|
|
ceph_encode_64(&p, rbd_dev->spec->snap_id);
|
2018-08-22 23:26:10 +08:00
|
|
|
ret = __get_parent_info(rbd_dev, req_page, reply_page, pii);
|
|
|
|
if (ret > 0)
|
|
|
|
ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page,
|
|
|
|
pii);
|
2018-08-22 23:11:27 +08:00
|
|
|
|
|
|
|
__free_page(req_page);
|
|
|
|
__free_page(reply_page);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct rbd_spec *parent_spec;
|
2018-08-22 23:11:27 +08:00
|
|
|
struct parent_image_info pii = { 0 };
|
2012-10-26 12:34:42 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
parent_spec = rbd_spec_alloc();
|
|
|
|
if (!parent_spec)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2018-08-22 23:11:27 +08:00
|
|
|
ret = get_parent_info(rbd_dev, &pii);
|
|
|
|
if (ret)
|
2012-10-26 12:34:42 +08:00
|
|
|
goto out_err;
|
|
|
|
|
2018-08-22 23:26:10 +08:00
|
|
|
dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
|
|
|
|
__func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id,
|
|
|
|
pii.has_overlap, pii.overlap);
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2018-08-22 23:26:10 +08:00
|
|
|
if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) {
|
rbd: detect when clone image is flattened
A format 2 clone image can be the subject of a "flatten" operation,
during which all of its data gets "copied up" from its parent image,
leaving the image fully populated. Once this is complete, the
clone's association with the parent is abolished.
Since this can occur when a clone is mapped, we need to detect when
it has occurred and handle it accordingly. We know an image has
been flattened when we know it at one time had a parent, but we have
learned (via a "get_parent" object class method call) it no longer
has one.
There might be in-flight requests at the point we learn an image has
been flattened, so we can't simply clean up parent data structures
right away. Instead, we'll drop the initial parent reference when
the parent has disappeared (rather than when the image gets
destroyed), which will allow the last in-flight reference to clean
things up when it's complete.
We leverage the fact that a zero parent overlap renders an image
effectively unlayered. We set the overlap to 0 at the point we
detect the clone image has flattened, which allows the unlayered
behavior to take effect immediately, while keeping other parent
structures in place until in-flight requests to complete.
This and the next few patches resolve:
http://tracker.ceph.com/issues/3763
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-05-07 06:40:33 +08:00
|
|
|
/*
|
|
|
|
* Either the parent never existed, or we have
|
|
|
|
* record of it but the image got flattened so it no
|
|
|
|
* longer has a parent. When the parent of a
|
|
|
|
* layered image disappears we immediately set the
|
|
|
|
* overlap to 0. The effect of this is that all new
|
|
|
|
* requests will be treated as if the image had no
|
|
|
|
* parent.
|
2018-08-22 23:26:10 +08:00
|
|
|
*
|
|
|
|
* If !pii.has_overlap, the parent image spec is not
|
|
|
|
* applicable. It's there to avoid duplication in each
|
|
|
|
* snapshot record.
|
rbd: detect when clone image is flattened
A format 2 clone image can be the subject of a "flatten" operation,
during which all of its data gets "copied up" from its parent image,
leaving the image fully populated. Once this is complete, the
clone's association with the parent is abolished.
Since this can occur when a clone is mapped, we need to detect when
it has occurred and handle it accordingly. We know an image has
been flattened when we know it at one time had a parent, but we have
learned (via a "get_parent" object class method call) it no longer
has one.
There might be in-flight requests at the point we learn an image has
been flattened, so we can't simply clean up parent data structures
right away. Instead, we'll drop the initial parent reference when
the parent has disappeared (rather than when the image gets
destroyed), which will allow the last in-flight reference to clean
things up when it's complete.
We leverage the fact that a zero parent overlap renders an image
effectively unlayered. We set the overlap to 0 at the point we
detect the clone image has flattened, which allows the unlayered
behavior to take effect immediately, while keeping other parent
structures in place until in-flight requests to complete.
This and the next few patches resolve:
http://tracker.ceph.com/issues/3763
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-05-07 06:40:33 +08:00
|
|
|
*/
|
|
|
|
if (rbd_dev->parent_overlap) {
|
|
|
|
rbd_dev->parent_overlap = 0;
|
|
|
|
rbd_dev_parent_put(rbd_dev);
|
|
|
|
pr_info("%s: clone image has been flattened\n",
|
|
|
|
rbd_dev->disk->disk_name);
|
|
|
|
}
|
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
goto out; /* No parent? No problem. */
|
rbd: detect when clone image is flattened
A format 2 clone image can be the subject of a "flatten" operation,
during which all of its data gets "copied up" from its parent image,
leaving the image fully populated. Once this is complete, the
clone's association with the parent is abolished.
Since this can occur when a clone is mapped, we need to detect when
it has occurred and handle it accordingly. We know an image has
been flattened when we know it at one time had a parent, but we have
learned (via a "get_parent" object class method call) it no longer
has one.
There might be in-flight requests at the point we learn an image has
been flattened, so we can't simply clean up parent data structures
right away. Instead, we'll drop the initial parent reference when
the parent has disappeared (rather than when the image gets
destroyed), which will allow the last in-flight reference to clean
things up when it's complete.
We leverage the fact that a zero parent overlap renders an image
effectively unlayered. We set the overlap to 0 at the point we
detect the clone image has flattened, which allows the unlayered
behavior to take effect immediately, while keeping other parent
structures in place until in-flight requests to complete.
This and the next few patches resolve:
http://tracker.ceph.com/issues/3763
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-05-07 06:40:33 +08:00
|
|
|
}
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2012-11-15 02:25:19 +08:00
|
|
|
/* The ceph file layout needs to fit pool id in 32 bits */
|
|
|
|
|
|
|
|
ret = -EIO;
|
2018-08-22 23:11:27 +08:00
|
|
|
if (pii.pool_id > (u64)U32_MAX) {
|
2014-07-11 16:11:20 +08:00
|
|
|
rbd_warn(NULL, "parent pool id too large (%llu > %u)",
|
2018-08-22 23:11:27 +08:00
|
|
|
(unsigned long long)pii.pool_id, U32_MAX);
|
2012-10-26 12:34:42 +08:00
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
2013-05-30 00:18:59 +08:00
|
|
|
/*
|
|
|
|
* The parent won't change (except when the clone is
|
|
|
|
* flattened, already handled that). So we only need to
|
|
|
|
* record the parent spec we have not already done so.
|
|
|
|
*/
|
|
|
|
if (!rbd_dev->parent_spec) {
|
2018-08-22 23:11:27 +08:00
|
|
|
parent_spec->pool_id = pii.pool_id;
|
2018-08-22 23:26:10 +08:00
|
|
|
if (pii.pool_ns && *pii.pool_ns) {
|
|
|
|
parent_spec->pool_ns = pii.pool_ns;
|
|
|
|
pii.pool_ns = NULL;
|
|
|
|
}
|
2018-08-22 23:11:27 +08:00
|
|
|
parent_spec->image_id = pii.image_id;
|
|
|
|
pii.image_id = NULL;
|
|
|
|
parent_spec->snap_id = pii.snap_id;
|
2018-07-03 21:28:43 +08:00
|
|
|
|
2013-05-07 06:40:33 +08:00
|
|
|
rbd_dev->parent_spec = parent_spec;
|
|
|
|
parent_spec = NULL; /* rbd_dev now owns this */
|
2013-05-30 00:18:59 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2015-01-20 03:57:39 +08:00
|
|
|
* We always update the parent overlap. If it's zero we issue
|
|
|
|
* a warning, as we will proceed as if there was no parent.
|
2013-05-30 00:18:59 +08:00
|
|
|
*/
|
2018-08-22 23:11:27 +08:00
|
|
|
if (!pii.overlap) {
|
2013-05-30 00:18:59 +08:00
|
|
|
if (parent_spec) {
|
2015-01-20 03:57:39 +08:00
|
|
|
/* refresh, careful to warn just once */
|
|
|
|
if (rbd_dev->parent_overlap)
|
|
|
|
rbd_warn(rbd_dev,
|
|
|
|
"clone now standalone (overlap became 0)");
|
2013-05-30 00:18:59 +08:00
|
|
|
} else {
|
2015-01-20 03:57:39 +08:00
|
|
|
/* initial probe */
|
|
|
|
rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
|
2013-05-30 00:18:59 +08:00
|
|
|
}
|
2013-05-07 06:40:33 +08:00
|
|
|
}
|
2018-08-22 23:11:27 +08:00
|
|
|
rbd_dev->parent_overlap = pii.overlap;
|
2015-01-20 03:57:39 +08:00
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
out:
|
|
|
|
ret = 0;
|
|
|
|
out_err:
|
2018-08-22 23:26:10 +08:00
|
|
|
kfree(pii.pool_ns);
|
2018-08-22 23:11:27 +08:00
|
|
|
kfree(pii.image_id);
|
2012-10-26 12:34:42 +08:00
|
|
|
rbd_spec_put(parent_spec);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-04-22 01:14:45 +08:00
|
|
|
static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct {
|
|
|
|
__le64 stripe_unit;
|
|
|
|
__le64 stripe_count;
|
|
|
|
} __attribute__ ((packed)) striping_info_buf = { 0 };
|
|
|
|
size_t size = sizeof (striping_info_buf);
|
|
|
|
void *p;
|
|
|
|
int ret;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_stripe_unit_count",
|
|
|
|
NULL, 0, &striping_info_buf, size);
|
2013-04-22 01:14:45 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
if (ret < size)
|
|
|
|
return -ERANGE;
|
|
|
|
|
|
|
|
p = &striping_info_buf;
|
2018-02-07 19:09:12 +08:00
|
|
|
rbd_dev->header.stripe_unit = ceph_decode_64(&p);
|
|
|
|
rbd_dev->header.stripe_count = ceph_decode_64(&p);
|
2013-04-22 01:14:45 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
__le64 data_pool_id;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_data_pool",
|
|
|
|
NULL, 0, &data_pool_id, sizeof(data_pool_id));
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
if (ret < sizeof(data_pool_id))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id);
|
|
|
|
WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
static char *rbd_dev_image_name(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2017-01-26 01:16:21 +08:00
|
|
|
CEPH_DEFINE_OID_ONSTACK(oid);
|
2012-10-31 08:40:33 +08:00
|
|
|
size_t image_id_size;
|
|
|
|
char *image_id;
|
|
|
|
void *p;
|
|
|
|
void *end;
|
|
|
|
size_t size;
|
|
|
|
void *reply_buf = NULL;
|
|
|
|
size_t len = 0;
|
|
|
|
char *image_name = NULL;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
rbd_assert(!rbd_dev->spec->image_name);
|
|
|
|
|
2012-11-01 21:39:26 +08:00
|
|
|
len = strlen(rbd_dev->spec->image_id);
|
|
|
|
image_id_size = sizeof (__le32) + len;
|
2012-10-31 08:40:33 +08:00
|
|
|
image_id = kmalloc(image_id_size, GFP_KERNEL);
|
|
|
|
if (!image_id)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
p = image_id;
|
2013-04-22 01:14:45 +08:00
|
|
|
end = image_id + image_id_size;
|
2013-04-22 01:14:45 +08:00
|
|
|
ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len);
|
2012-10-31 08:40:33 +08:00
|
|
|
|
|
|
|
size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX;
|
|
|
|
reply_buf = kmalloc(size, GFP_KERNEL);
|
|
|
|
if (!reply_buf)
|
|
|
|
goto out;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ceph_oid_printf(&oid, "%s", RBD_DIRECTORY);
|
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
|
|
|
|
"dir_get_name", image_id, image_id_size,
|
|
|
|
reply_buf, size);
|
2012-10-31 08:40:33 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
p = reply_buf;
|
2013-04-26 04:09:42 +08:00
|
|
|
end = reply_buf + ret;
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL);
|
|
|
|
if (IS_ERR(image_name))
|
|
|
|
image_name = NULL;
|
|
|
|
else
|
|
|
|
dout("%s: name is %s len is %zd\n", __func__, image_name, len);
|
|
|
|
out:
|
|
|
|
kfree(reply_buf);
|
|
|
|
kfree(image_id);
|
|
|
|
|
|
|
|
return image_name;
|
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
|
|
|
|
{
|
|
|
|
struct ceph_snap_context *snapc = rbd_dev->header.snapc;
|
|
|
|
const char *snap_name;
|
|
|
|
u32 which = 0;
|
|
|
|
|
|
|
|
/* Skip over names until we find the one we are looking for */
|
|
|
|
|
|
|
|
snap_name = rbd_dev->header.snap_names;
|
|
|
|
while (which < snapc->num_snaps) {
|
|
|
|
if (!strcmp(name, snap_name))
|
|
|
|
return snapc->snaps[which];
|
|
|
|
snap_name += strlen(snap_name) + 1;
|
|
|
|
which++;
|
|
|
|
}
|
|
|
|
return CEPH_NOSNAP;
|
|
|
|
}
|
|
|
|
|
|
|
|
static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
|
|
|
|
{
|
|
|
|
struct ceph_snap_context *snapc = rbd_dev->header.snapc;
|
|
|
|
u32 which;
|
|
|
|
bool found = false;
|
|
|
|
u64 snap_id;
|
|
|
|
|
|
|
|
for (which = 0; !found && which < snapc->num_snaps; which++) {
|
|
|
|
const char *snap_name;
|
|
|
|
|
|
|
|
snap_id = snapc->snaps[which];
|
|
|
|
snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
|
2013-08-30 10:16:42 +08:00
|
|
|
if (IS_ERR(snap_name)) {
|
|
|
|
/* ignore no-longer existing snapshots */
|
|
|
|
if (PTR_ERR(snap_name) == -ENOENT)
|
|
|
|
continue;
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
2013-04-30 13:44:33 +08:00
|
|
|
found = !strcmp(name, snap_name);
|
|
|
|
kfree(snap_name);
|
|
|
|
}
|
|
|
|
return found ? snap_id : CEPH_NOSNAP;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if
|
|
|
|
* no snapshot by that name is found, or if an error occurs.
|
|
|
|
*/
|
|
|
|
static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
|
|
|
|
{
|
|
|
|
if (rbd_dev->image_format == 1)
|
|
|
|
return rbd_v1_snap_id_by_name(rbd_dev, name);
|
|
|
|
|
|
|
|
return rbd_v2_snap_id_by_name(rbd_dev, name);
|
|
|
|
}
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
/*
|
2014-07-23 21:11:20 +08:00
|
|
|
* An image being mapped will have everything but the snap id.
|
|
|
|
*/
|
|
|
|
static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct rbd_spec *spec = rbd_dev->spec;
|
|
|
|
|
|
|
|
rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name);
|
|
|
|
rbd_assert(spec->image_id && spec->image_name);
|
|
|
|
rbd_assert(spec->snap_name);
|
|
|
|
|
|
|
|
if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) {
|
|
|
|
u64 snap_id;
|
|
|
|
|
|
|
|
snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name);
|
|
|
|
if (snap_id == CEPH_NOSNAP)
|
|
|
|
return -ENOENT;
|
|
|
|
|
|
|
|
spec->snap_id = snap_id;
|
|
|
|
} else {
|
|
|
|
spec->snap_id = CEPH_NOSNAP;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A parent image will have all ids but none of the names.
|
2013-04-26 12:15:08 +08:00
|
|
|
*
|
2014-07-23 21:11:20 +08:00
|
|
|
* All names in an rbd spec are dynamically allocated. It's OK if we
|
|
|
|
* can't figure out the name for an image id.
|
2012-10-31 08:40:33 +08:00
|
|
|
*/
|
2014-07-23 21:11:20 +08:00
|
|
|
static int rbd_spec_fill_names(struct rbd_device *rbd_dev)
|
2012-10-31 08:40:33 +08:00
|
|
|
{
|
2013-04-26 22:43:48 +08:00
|
|
|
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
|
|
|
|
struct rbd_spec *spec = rbd_dev->spec;
|
|
|
|
const char *pool_name;
|
|
|
|
const char *image_name;
|
|
|
|
const char *snap_name;
|
2012-10-31 08:40:33 +08:00
|
|
|
int ret;
|
|
|
|
|
2014-07-23 21:11:20 +08:00
|
|
|
rbd_assert(spec->pool_id != CEPH_NOPOOL);
|
|
|
|
rbd_assert(spec->image_id);
|
|
|
|
rbd_assert(spec->snap_id != CEPH_NOSNAP);
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-04-26 22:43:48 +08:00
|
|
|
/* Get the pool name; we have to make our own copy of this */
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-04-26 22:43:48 +08:00
|
|
|
pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id);
|
|
|
|
if (!pool_name) {
|
|
|
|
rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id);
|
2012-11-01 23:17:15 +08:00
|
|
|
return -EIO;
|
|
|
|
}
|
2013-04-26 22:43:48 +08:00
|
|
|
pool_name = kstrdup(pool_name, GFP_KERNEL);
|
|
|
|
if (!pool_name)
|
2012-10-31 08:40:33 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
/* Fetch the image name; tolerate failure here */
|
|
|
|
|
2013-04-26 22:43:48 +08:00
|
|
|
image_name = rbd_dev_image_name(rbd_dev);
|
|
|
|
if (!image_name)
|
2012-11-01 23:17:15 +08:00
|
|
|
rbd_warn(rbd_dev, "unable to get image name");
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2014-07-23 21:11:20 +08:00
|
|
|
/* Fetch the snapshot name */
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-04-26 22:43:48 +08:00
|
|
|
snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
|
2013-09-05 08:57:31 +08:00
|
|
|
if (IS_ERR(snap_name)) {
|
|
|
|
ret = PTR_ERR(snap_name);
|
2012-10-31 08:40:33 +08:00
|
|
|
goto out_err;
|
2013-04-26 22:43:48 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
spec->pool_name = pool_name;
|
|
|
|
spec->image_name = image_name;
|
|
|
|
spec->snap_name = snap_name;
|
2012-10-31 08:40:33 +08:00
|
|
|
|
|
|
|
return 0;
|
2014-07-23 21:11:20 +08:00
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
out_err:
|
2013-04-26 22:43:48 +08:00
|
|
|
kfree(image_name);
|
|
|
|
kfree(pool_name);
|
2012-10-31 08:40:33 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
|
2012-07-04 05:01:19 +08:00
|
|
|
{
|
|
|
|
size_t size;
|
|
|
|
int ret;
|
|
|
|
void *reply_buf;
|
|
|
|
void *p;
|
|
|
|
void *end;
|
|
|
|
u64 seq;
|
|
|
|
u32 snap_count;
|
|
|
|
struct ceph_snap_context *snapc;
|
|
|
|
u32 i;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We'll need room for the seq value (maximum snapshot id),
|
|
|
|
* snapshot count, and array of that many snapshot ids.
|
|
|
|
* For now we have a fixed upper limit on the number we're
|
|
|
|
* prepared to receive.
|
|
|
|
*/
|
|
|
|
size = sizeof (__le64) + sizeof (__le32) +
|
|
|
|
RBD_MAX_SNAP_COUNT * sizeof (__le64);
|
|
|
|
reply_buf = kzalloc(size, GFP_KERNEL);
|
|
|
|
if (!reply_buf)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_snapcontext",
|
|
|
|
NULL, 0, reply_buf, size);
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2012-07-04 05:01:19 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
p = reply_buf;
|
2013-04-22 01:14:45 +08:00
|
|
|
end = reply_buf + ret;
|
|
|
|
ret = -ERANGE;
|
2012-07-04 05:01:19 +08:00
|
|
|
ceph_decode_64_safe(&p, end, seq, out);
|
|
|
|
ceph_decode_32_safe(&p, end, snap_count, out);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Make sure the reported number of snapshot ids wouldn't go
|
|
|
|
* beyond the end of our buffer. But before checking that,
|
|
|
|
* make sure the computed size of the snapshot context we
|
|
|
|
* allocate is representable in a size_t.
|
|
|
|
*/
|
|
|
|
if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context))
|
|
|
|
/ sizeof (u64)) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (!ceph_has_room(&p, end, snap_count * sizeof (__le64)))
|
|
|
|
goto out;
|
2013-04-26 22:43:47 +08:00
|
|
|
ret = 0;
|
2012-07-04 05:01:19 +08:00
|
|
|
|
2013-04-30 13:44:32 +08:00
|
|
|
snapc = ceph_create_snap_context(snap_count, GFP_KERNEL);
|
2012-07-04 05:01:19 +08:00
|
|
|
if (!snapc) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
snapc->seq = seq;
|
|
|
|
for (i = 0; i < snap_count; i++)
|
|
|
|
snapc->snaps[i] = ceph_decode_64(&p);
|
|
|
|
|
2013-05-06 21:37:00 +08:00
|
|
|
ceph_put_snap_context(rbd_dev->header.snapc);
|
2012-07-04 05:01:19 +08:00
|
|
|
rbd_dev->header.snapc = snapc;
|
|
|
|
|
|
|
|
dout(" snap context seq = %llu, snap_count = %u\n",
|
2013-04-22 01:14:45 +08:00
|
|
|
(unsigned long long)seq, (unsigned int)snap_count);
|
2012-07-04 05:01:19 +08:00
|
|
|
out:
|
|
|
|
kfree(reply_buf);
|
|
|
|
|
2013-04-22 01:14:45 +08:00
|
|
|
return ret;
|
2012-07-04 05:01:19 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
|
|
|
|
u64 snap_id)
|
2012-07-04 05:01:19 +08:00
|
|
|
{
|
|
|
|
size_t size;
|
|
|
|
void *reply_buf;
|
2013-04-30 13:44:33 +08:00
|
|
|
__le64 snapid;
|
2012-07-04 05:01:19 +08:00
|
|
|
int ret;
|
|
|
|
void *p;
|
|
|
|
void *end;
|
|
|
|
char *snap_name;
|
|
|
|
|
|
|
|
size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN;
|
|
|
|
reply_buf = kmalloc(size, GFP_KERNEL);
|
|
|
|
if (!reply_buf)
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
snapid = cpu_to_le64(snap_id);
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
|
|
|
|
&rbd_dev->header_oloc, "get_snapshot_name",
|
|
|
|
&snapid, sizeof(snapid), reply_buf, size);
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2013-04-26 04:09:42 +08:00
|
|
|
if (ret < 0) {
|
|
|
|
snap_name = ERR_PTR(ret);
|
2012-07-04 05:01:19 +08:00
|
|
|
goto out;
|
2013-04-26 04:09:42 +08:00
|
|
|
}
|
2012-07-04 05:01:19 +08:00
|
|
|
|
|
|
|
p = reply_buf;
|
2013-04-26 04:09:42 +08:00
|
|
|
end = reply_buf + ret;
|
2012-10-26 12:34:41 +08:00
|
|
|
snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL);
|
2013-04-26 04:09:42 +08:00
|
|
|
if (IS_ERR(snap_name))
|
2012-07-04 05:01:19 +08:00
|
|
|
goto out;
|
|
|
|
|
2013-04-26 04:09:42 +08:00
|
|
|
dout(" snap_id 0x%016llx snap_name = %s\n",
|
2013-04-30 13:44:33 +08:00
|
|
|
(unsigned long long)snap_id, snap_name);
|
2012-07-04 05:01:19 +08:00
|
|
|
out:
|
|
|
|
kfree(reply_buf);
|
|
|
|
|
2013-04-26 04:09:42 +08:00
|
|
|
return snap_name;
|
2012-07-04 05:01:19 +08:00
|
|
|
}
|
|
|
|
|
2013-05-06 22:51:30 +08:00
|
|
|
static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
|
2012-09-01 06:29:55 +08:00
|
|
|
{
|
2013-05-06 22:51:30 +08:00
|
|
|
bool first_time = rbd_dev->header.object_prefix == NULL;
|
2012-09-01 06:29:55 +08:00
|
|
|
int ret;
|
|
|
|
|
2013-06-13 05:43:10 +08:00
|
|
|
ret = rbd_dev_v2_image_size(rbd_dev);
|
|
|
|
if (ret)
|
2013-06-01 06:40:45 +08:00
|
|
|
return ret;
|
2013-06-13 05:43:10 +08:00
|
|
|
|
2013-05-06 22:51:30 +08:00
|
|
|
if (first_time) {
|
|
|
|
ret = rbd_dev_v2_header_onetime(rbd_dev);
|
|
|
|
if (ret)
|
2013-06-01 06:40:45 +08:00
|
|
|
return ret;
|
2013-05-06 22:51:30 +08:00
|
|
|
}
|
|
|
|
|
2013-04-30 13:44:33 +08:00
|
|
|
ret = rbd_dev_v2_snap_context(rbd_dev);
|
2015-08-31 23:22:10 +08:00
|
|
|
if (ret && first_time) {
|
|
|
|
kfree(rbd_dev->header.object_prefix);
|
|
|
|
rbd_dev->header.object_prefix = NULL;
|
|
|
|
}
|
2012-09-01 06:29:55 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2014-07-23 21:11:19 +08:00
|
|
|
static int rbd_dev_header_info(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
|
|
|
|
|
|
|
|
if (rbd_dev->image_format == 1)
|
|
|
|
return rbd_dev_v1_header_info(rbd_dev);
|
|
|
|
|
|
|
|
return rbd_dev_v2_header_info(rbd_dev);
|
|
|
|
}
|
|
|
|
|
2012-02-02 22:13:30 +08:00
|
|
|
/*
|
|
|
|
* Skips over white space at *buf, and updates *buf to point to the
|
|
|
|
* first found non-space character (if any). Returns the length of
|
2012-02-08 02:03:37 +08:00
|
|
|
* the token (string of non-white space characters) found. Note
|
|
|
|
* that *buf must be terminated with '\0'.
|
2012-02-02 22:13:30 +08:00
|
|
|
*/
|
|
|
|
static inline size_t next_token(const char **buf)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* These are the characters that produce nonzero for
|
|
|
|
* isspace() in the "C" and "POSIX" locales.
|
|
|
|
*/
|
|
|
|
const char *spaces = " \f\n\r\t\v";
|
|
|
|
|
|
|
|
*buf += strspn(*buf, spaces); /* Find start of token */
|
|
|
|
|
|
|
|
return strcspn(*buf, spaces); /* Return token length */
|
|
|
|
}
|
|
|
|
|
2012-07-10 10:04:23 +08:00
|
|
|
/*
|
|
|
|
* Finds the next token in *buf, dynamically allocates a buffer big
|
|
|
|
* enough to hold a copy of it, and copies the token into the new
|
|
|
|
* buffer. The copy is guaranteed to be terminated with '\0'. Note
|
|
|
|
* that a duplicate buffer is created even for a zero-length token.
|
|
|
|
*
|
|
|
|
* Returns a pointer to the newly-allocated duplicate, or a null
|
|
|
|
* pointer if memory for the duplicate was not available. If
|
|
|
|
* the lenp argument is a non-null pointer, the length of the token
|
|
|
|
* (not including the '\0') is returned in *lenp.
|
|
|
|
*
|
|
|
|
* If successful, the *buf pointer will be updated to point beyond
|
|
|
|
* the end of the found token.
|
|
|
|
*
|
|
|
|
* Note: uses GFP_KERNEL for allocation.
|
|
|
|
*/
|
|
|
|
static inline char *dup_token(const char **buf, size_t *lenp)
|
|
|
|
{
|
|
|
|
char *dup;
|
|
|
|
size_t len;
|
|
|
|
|
|
|
|
len = next_token(buf);
|
2012-11-01 21:39:27 +08:00
|
|
|
dup = kmemdup(*buf, len + 1, GFP_KERNEL);
|
2012-07-10 10:04:23 +08:00
|
|
|
if (!dup)
|
|
|
|
return NULL;
|
|
|
|
*(dup + len) = '\0';
|
|
|
|
*buf += len;
|
|
|
|
|
|
|
|
if (lenp)
|
|
|
|
*lenp = len;
|
|
|
|
|
|
|
|
return dup;
|
|
|
|
}
|
|
|
|
|
2012-02-02 22:13:30 +08:00
|
|
|
/*
|
2012-10-26 12:34:42 +08:00
|
|
|
* Parse the options provided for an "rbd add" (i.e., rbd image
|
|
|
|
* mapping) request. These arrive via a write to /sys/bus/rbd/add,
|
|
|
|
* and the data written is passed here via a NUL-terminated buffer.
|
|
|
|
* Returns 0 if successful or an error code otherwise.
|
2012-07-12 23:46:35 +08:00
|
|
|
*
|
2012-10-26 12:34:42 +08:00
|
|
|
* The information extracted from these options is recorded in
|
|
|
|
* the other parameters which return dynamically-allocated
|
|
|
|
* structures:
|
|
|
|
* ceph_opts
|
|
|
|
* The address of a pointer that will refer to a ceph options
|
|
|
|
* structure. Caller must release the returned pointer using
|
|
|
|
* ceph_destroy_options() when it is no longer needed.
|
|
|
|
* rbd_opts
|
|
|
|
* Address of an rbd options pointer. Fully initialized by
|
|
|
|
* this function; caller must release with kfree().
|
|
|
|
* spec
|
|
|
|
* Address of an rbd image specification pointer. Fully
|
|
|
|
* initialized by this function based on parsed options.
|
|
|
|
* Caller must release with rbd_spec_put().
|
|
|
|
*
|
|
|
|
* The options passed take this form:
|
|
|
|
* <mon_addrs> <options> <pool_name> <image_name> [<snap_id>]
|
|
|
|
* where:
|
|
|
|
* <mon_addrs>
|
|
|
|
* A comma-separated list of one or more monitor addresses.
|
|
|
|
* A monitor address is an ip address, optionally followed
|
|
|
|
* by a port number (separated by a colon).
|
|
|
|
* I.e.: ip1[:port1][,ip2[:port2]...]
|
|
|
|
* <options>
|
|
|
|
* A comma-separated list of ceph and/or rbd options.
|
|
|
|
* <pool_name>
|
|
|
|
* The name of the rados pool containing the rbd image.
|
|
|
|
* <image_name>
|
|
|
|
* The name of the image in that pool to map.
|
|
|
|
* <snap_id>
|
|
|
|
* An optional snapshot id. If provided, the mapping will
|
|
|
|
* present data from the image at the time that snapshot was
|
|
|
|
* created. The image head is used if no snapshot id is
|
|
|
|
* provided. Snapshot mappings are always read-only.
|
2012-02-02 22:13:30 +08:00
|
|
|
*/
|
2012-10-26 12:34:42 +08:00
|
|
|
static int rbd_add_parse_args(const char *buf,
|
2012-10-26 12:34:41 +08:00
|
|
|
struct ceph_options **ceph_opts,
|
2012-10-26 12:34:42 +08:00
|
|
|
struct rbd_options **opts,
|
|
|
|
struct rbd_spec **rbd_spec)
|
2012-02-02 22:13:30 +08:00
|
|
|
{
|
2012-07-12 23:46:35 +08:00
|
|
|
size_t len;
|
2012-10-26 12:34:42 +08:00
|
|
|
char *options;
|
2012-10-26 12:34:41 +08:00
|
|
|
const char *mon_addrs;
|
2013-04-26 22:43:47 +08:00
|
|
|
char *snap_name;
|
2012-10-26 12:34:41 +08:00
|
|
|
size_t mon_addrs_size;
|
2018-07-03 21:28:43 +08:00
|
|
|
struct parse_rbd_opts_ctx pctx = { 0 };
|
2012-10-26 12:34:42 +08:00
|
|
|
struct ceph_options *copts;
|
2012-10-26 12:34:41 +08:00
|
|
|
int ret;
|
2012-02-02 22:13:30 +08:00
|
|
|
|
|
|
|
/* The first four tokens are required */
|
|
|
|
|
2012-02-02 22:13:30 +08:00
|
|
|
len = next_token(&buf);
|
2012-11-01 23:17:15 +08:00
|
|
|
if (!len) {
|
|
|
|
rbd_warn(NULL, "no monitor address(es) provided");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2012-10-26 12:34:41 +08:00
|
|
|
mon_addrs = buf;
|
2012-10-26 12:34:41 +08:00
|
|
|
mon_addrs_size = len + 1;
|
2012-02-02 22:13:30 +08:00
|
|
|
buf += len;
|
2012-02-02 22:13:30 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
ret = -EINVAL;
|
2012-10-26 12:34:41 +08:00
|
|
|
options = dup_token(&buf, NULL);
|
|
|
|
if (!options)
|
2012-10-26 12:34:41 +08:00
|
|
|
return -ENOMEM;
|
2012-11-01 23:17:15 +08:00
|
|
|
if (!*options) {
|
|
|
|
rbd_warn(NULL, "no options provided");
|
|
|
|
goto out_err;
|
|
|
|
}
|
2012-02-02 22:13:30 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.spec = rbd_spec_alloc();
|
|
|
|
if (!pctx.spec)
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_mem;
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.spec->pool_name = dup_token(&buf, NULL);
|
|
|
|
if (!pctx.spec->pool_name)
|
2012-10-26 12:34:42 +08:00
|
|
|
goto out_mem;
|
2018-07-03 21:28:43 +08:00
|
|
|
if (!*pctx.spec->pool_name) {
|
2012-11-01 23:17:15 +08:00
|
|
|
rbd_warn(NULL, "no pool name provided");
|
|
|
|
goto out_err;
|
|
|
|
}
|
2012-02-02 22:13:30 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.spec->image_name = dup_token(&buf, NULL);
|
|
|
|
if (!pctx.spec->image_name)
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_mem;
|
2018-07-03 21:28:43 +08:00
|
|
|
if (!*pctx.spec->image_name) {
|
2012-11-01 23:17:15 +08:00
|
|
|
rbd_warn(NULL, "no image name provided");
|
|
|
|
goto out_err;
|
|
|
|
}
|
2012-07-04 05:01:19 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
/*
|
|
|
|
* Snapshot name is optional; default is to use "-"
|
|
|
|
* (indicating the head/no snapshot).
|
|
|
|
*/
|
2012-09-01 06:29:52 +08:00
|
|
|
len = next_token(&buf);
|
2012-07-10 10:04:24 +08:00
|
|
|
if (!len) {
|
2012-09-01 06:29:52 +08:00
|
|
|
buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */
|
|
|
|
len = sizeof (RBD_SNAP_HEAD_NAME) - 1;
|
2012-10-26 12:34:41 +08:00
|
|
|
} else if (len > RBD_MAX_SNAP_NAME_LEN) {
|
2012-10-26 12:34:41 +08:00
|
|
|
ret = -ENAMETOOLONG;
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_err;
|
2012-07-10 10:04:24 +08:00
|
|
|
}
|
2013-04-26 22:43:47 +08:00
|
|
|
snap_name = kmemdup(buf, len + 1, GFP_KERNEL);
|
|
|
|
if (!snap_name)
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_mem;
|
2013-04-26 22:43:47 +08:00
|
|
|
*(snap_name + len) = '\0';
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.spec->snap_name = snap_name;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
/* Initialize all rbd options to the defaults */
|
2012-02-02 22:13:30 +08:00
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.opts = kzalloc(sizeof(*pctx.opts), GFP_KERNEL);
|
|
|
|
if (!pctx.opts)
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_mem;
|
|
|
|
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.opts->read_only = RBD_READ_ONLY_DEFAULT;
|
|
|
|
pctx.opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT;
|
2019-01-30 22:14:48 +08:00
|
|
|
pctx.opts->alloc_size = RBD_ALLOC_SIZE_DEFAULT;
|
2018-07-03 21:28:43 +08:00
|
|
|
pctx.opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT;
|
|
|
|
pctx.opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT;
|
|
|
|
pctx.opts->exclusive = RBD_EXCLUSIVE_DEFAULT;
|
|
|
|
pctx.opts->trim = RBD_TRIM_DEFAULT;
|
2012-07-12 23:46:35 +08:00
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
copts = ceph_parse_options(options, mon_addrs,
|
2018-07-03 21:28:43 +08:00
|
|
|
mon_addrs + mon_addrs_size - 1,
|
|
|
|
parse_rbd_opts_token, &pctx);
|
2012-10-26 12:34:42 +08:00
|
|
|
if (IS_ERR(copts)) {
|
|
|
|
ret = PTR_ERR(copts);
|
2012-10-26 12:34:41 +08:00
|
|
|
goto out_err;
|
|
|
|
}
|
2012-10-26 12:34:42 +08:00
|
|
|
kfree(options);
|
|
|
|
|
|
|
|
*ceph_opts = copts;
|
2018-07-03 21:28:43 +08:00
|
|
|
*opts = pctx.opts;
|
|
|
|
*rbd_spec = pctx.spec;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return 0;
|
2012-10-26 12:34:41 +08:00
|
|
|
out_mem:
|
2012-10-26 12:34:41 +08:00
|
|
|
ret = -ENOMEM;
|
2012-07-12 23:46:35 +08:00
|
|
|
out_err:
|
2018-07-03 21:28:43 +08:00
|
|
|
kfree(pctx.opts);
|
|
|
|
rbd_spec_put(pctx.spec);
|
2012-10-26 12:34:41 +08:00
|
|
|
kfree(options);
|
2012-07-12 23:46:35 +08:00
|
|
|
|
2012-10-26 12:34:41 +08:00
|
|
|
return ret;
|
2012-02-02 22:13:30 +08:00
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
static void rbd_dev_image_unlock(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
down_write(&rbd_dev->lock_rwsem);
|
|
|
|
if (__rbd_is_lock_owner(rbd_dev))
|
|
|
|
rbd_unlock(rbd_dev);
|
|
|
|
up_write(&rbd_dev->lock_rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rbd_add_acquire_lock(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2018-04-04 16:15:38 +08:00
|
|
|
int ret;
|
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) {
|
|
|
|
rbd_warn(rbd_dev, "exclusive-lock feature is not enabled");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* FIXME: "rbd map --exclusive" should be in interruptible */
|
|
|
|
down_read(&rbd_dev->lock_rwsem);
|
2018-04-04 16:15:38 +08:00
|
|
|
ret = rbd_wait_state_locked(rbd_dev, true);
|
2017-04-13 18:17:39 +08:00
|
|
|
up_read(&rbd_dev->lock_rwsem);
|
2018-04-04 16:15:38 +08:00
|
|
|
if (ret) {
|
2017-04-13 18:17:39 +08:00
|
|
|
rbd_warn(rbd_dev, "failed to acquire exclusive lock");
|
|
|
|
return -EROFS;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
/*
|
|
|
|
* An rbd format 2 image has a unique identifier, distinct from the
|
|
|
|
* name given to it by the user. Internally, that identifier is
|
|
|
|
* what's used to specify the names of objects related to the image.
|
|
|
|
*
|
|
|
|
* A special "rbd id" object is used to map an rbd image name to its
|
|
|
|
* id. If that object doesn't exist, then there is no v2 rbd image
|
|
|
|
* with the supplied name.
|
|
|
|
*
|
|
|
|
* This function will record the given rbd_dev's image_id field if
|
|
|
|
* it can be determined, and in that case will return 0. If any
|
|
|
|
* errors occur a negative errno will be returned and the rbd_dev's
|
|
|
|
* image_id field will be unchanged (and should be NULL).
|
|
|
|
*/
|
|
|
|
static int rbd_dev_image_id(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
size_t size;
|
2017-01-26 01:16:21 +08:00
|
|
|
CEPH_DEFINE_OID_ONSTACK(oid);
|
2012-07-11 09:30:11 +08:00
|
|
|
void *response;
|
2013-04-26 12:15:08 +08:00
|
|
|
char *image_id;
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
/*
|
|
|
|
* When probing a parent image, the image id is already
|
|
|
|
* known (and the image name likely is not). There's no
|
2013-04-26 12:15:08 +08:00
|
|
|
* need to fetch the image id again in this case. We
|
|
|
|
* do still need to set the image format though.
|
2012-10-31 08:40:33 +08:00
|
|
|
*/
|
2013-04-26 12:15:08 +08:00
|
|
|
if (rbd_dev->spec->image_id) {
|
|
|
|
rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1;
|
|
|
|
|
2012-10-31 08:40:33 +08:00
|
|
|
return 0;
|
2013-04-26 12:15:08 +08:00
|
|
|
}
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
/*
|
|
|
|
* First, see if the format 2 image id file exists, and if
|
|
|
|
* so, get the image's persistent id from it.
|
|
|
|
*/
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX,
|
|
|
|
rbd_dev->spec->image_name);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
dout("rbd id object name is %s\n", oid.name);
|
2012-07-11 09:30:11 +08:00
|
|
|
|
|
|
|
/* Response will be an encoded string, which includes a length */
|
|
|
|
|
|
|
|
size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX;
|
|
|
|
response = kzalloc(size, GFP_NOIO);
|
|
|
|
if (!response) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2013-04-26 12:15:08 +08:00
|
|
|
/* If it doesn't exist we'll assume it's a format 1 image */
|
|
|
|
|
2017-01-26 01:16:21 +08:00
|
|
|
ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc,
|
|
|
|
"get_id", NULL, 0,
|
|
|
|
response, RBD_IMAGE_ID_LEN_MAX);
|
2013-01-19 14:30:28 +08:00
|
|
|
dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
|
2013-04-26 12:15:08 +08:00
|
|
|
if (ret == -ENOENT) {
|
|
|
|
image_id = kstrdup("", GFP_KERNEL);
|
|
|
|
ret = image_id ? 0 : -ENOMEM;
|
|
|
|
if (!ret)
|
|
|
|
rbd_dev->image_format = 1;
|
2014-09-11 22:49:18 +08:00
|
|
|
} else if (ret >= 0) {
|
2013-04-26 12:15:08 +08:00
|
|
|
void *p = response;
|
|
|
|
|
|
|
|
image_id = ceph_extract_encoded_string(&p, p + ret,
|
2012-11-01 21:39:26 +08:00
|
|
|
NULL, GFP_NOIO);
|
2014-04-11 16:38:12 +08:00
|
|
|
ret = PTR_ERR_OR_ZERO(image_id);
|
2013-04-26 12:15:08 +08:00
|
|
|
if (!ret)
|
|
|
|
rbd_dev->image_format = 2;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!ret) {
|
|
|
|
rbd_dev->spec->image_id = image_id;
|
|
|
|
dout("image_id is %s\n", image_id);
|
2012-07-11 09:30:11 +08:00
|
|
|
}
|
|
|
|
out:
|
|
|
|
kfree(response);
|
2017-01-26 01:16:21 +08:00
|
|
|
ceph_oid_destroy(&oid);
|
2012-07-11 09:30:11 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-05-14 09:35:37 +08:00
|
|
|
/*
|
|
|
|
* Undo whatever state changes are made by v1 or v2 header info
|
|
|
|
* call.
|
|
|
|
*/
|
2013-04-29 12:32:34 +08:00
|
|
|
static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct rbd_image_header *header;
|
|
|
|
|
2015-01-19 17:06:14 +08:00
|
|
|
rbd_dev_parent_put(rbd_dev);
|
2013-04-29 12:32:34 +08:00
|
|
|
|
|
|
|
/* Free dynamic fields from the header, then zero it out */
|
|
|
|
|
|
|
|
header = &rbd_dev->header;
|
2013-04-30 13:44:32 +08:00
|
|
|
ceph_put_snap_context(header->snapc);
|
2013-04-29 12:32:34 +08:00
|
|
|
kfree(header->snap_sizes);
|
|
|
|
kfree(header->snap_names);
|
|
|
|
kfree(header->object_prefix);
|
|
|
|
memset(header, 0, sizeof (*header));
|
|
|
|
}
|
|
|
|
|
2013-05-06 22:51:30 +08:00
|
|
|
static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
|
2012-07-11 09:30:11 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
ret = rbd_dev_v2_object_prefix(rbd_dev);
|
2013-04-22 01:14:45 +08:00
|
|
|
if (ret)
|
2012-07-04 05:01:19 +08:00
|
|
|
goto out_err;
|
|
|
|
|
2013-05-06 22:51:30 +08:00
|
|
|
/*
|
|
|
|
* Get the and check features for the image. Currently the
|
|
|
|
* features are assumed to never change.
|
|
|
|
*/
|
2012-07-04 05:01:19 +08:00
|
|
|
ret = rbd_dev_v2_features(rbd_dev);
|
2013-04-22 01:14:45 +08:00
|
|
|
if (ret)
|
2012-07-04 05:01:19 +08:00
|
|
|
goto out_err;
|
2012-07-04 05:01:19 +08:00
|
|
|
|
2013-04-22 01:14:45 +08:00
|
|
|
/* If the image supports fancy striping, get its parameters */
|
|
|
|
|
|
|
|
if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
|
|
|
|
ret = rbd_dev_v2_striping_info(rbd_dev);
|
|
|
|
if (ret < 0)
|
|
|
|
goto out_err;
|
|
|
|
}
|
2012-07-11 09:30:11 +08:00
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) {
|
|
|
|
ret = rbd_dev_v2_data_pool(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
2017-01-26 01:16:22 +08:00
|
|
|
rbd_init_layout(rbd_dev);
|
2012-09-01 06:29:55 +08:00
|
|
|
return 0;
|
2017-01-26 01:16:22 +08:00
|
|
|
|
2012-07-04 05:01:19 +08:00
|
|
|
out_err:
|
2013-05-07 06:40:33 +08:00
|
|
|
rbd_dev->header.features = 0;
|
2012-07-04 05:01:19 +08:00
|
|
|
kfree(rbd_dev->header.object_prefix);
|
|
|
|
rbd_dev->header.object_prefix = NULL;
|
2012-07-04 05:01:19 +08:00
|
|
|
return ret;
|
2012-07-11 09:30:11 +08:00
|
|
|
}
|
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
/*
|
|
|
|
* @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() ->
|
|
|
|
* rbd_dev_image_probe() recursion depth, which means it's also the
|
|
|
|
* length of the already discovered part of the parent chain.
|
|
|
|
*/
|
|
|
|
static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
|
2012-10-31 04:47:17 +08:00
|
|
|
{
|
2012-10-31 08:40:33 +08:00
|
|
|
struct rbd_device *parent = NULL;
|
2013-04-27 04:44:36 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!rbd_dev->parent_spec)
|
|
|
|
return 0;
|
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
if (++depth > RBD_MAX_PARENT_CHAIN_LEN) {
|
|
|
|
pr_info("parent chain is too long (%d)\n", depth);
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
2016-08-12 21:45:52 +08:00
|
|
|
parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
|
2015-10-12 01:38:00 +08:00
|
|
|
if (!parent) {
|
|
|
|
ret = -ENOMEM;
|
2013-04-27 04:44:36 +08:00
|
|
|
goto out_err;
|
2015-10-12 01:38:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Images related by parent/child relationships always share
|
|
|
|
* rbd_client and spec/parent_spec, so bump their refcounts.
|
|
|
|
*/
|
|
|
|
__rbd_get_client(rbd_dev->rbd_client);
|
|
|
|
rbd_spec_get(rbd_dev->parent_spec);
|
2013-04-27 04:44:36 +08:00
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
ret = rbd_dev_image_probe(parent, depth);
|
2013-04-27 04:44:36 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out_err;
|
2015-10-12 01:38:00 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
rbd_dev->parent = parent;
|
2013-05-09 11:50:04 +08:00
|
|
|
atomic_set(&rbd_dev->parent_ref, 1);
|
2013-04-27 04:44:36 +08:00
|
|
|
return 0;
|
2015-10-12 01:38:00 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
out_err:
|
2015-10-12 01:38:00 +08:00
|
|
|
rbd_dev_unparent(rbd_dev);
|
2015-11-24 03:16:45 +08:00
|
|
|
rbd_dev_destroy(parent);
|
2013-04-27 04:44:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
static void rbd_dev_device_release(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
|
|
|
|
rbd_dev_mapping_clear(rbd_dev);
|
|
|
|
rbd_free_disk(rbd_dev);
|
|
|
|
if (!single_major)
|
|
|
|
unregister_blkdev(rbd_dev->major, rbd_dev->name);
|
|
|
|
}
|
|
|
|
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
/*
|
|
|
|
* rbd_dev->header_rwsem must be locked for write and will be unlocked
|
|
|
|
* upon return.
|
|
|
|
*/
|
2013-04-29 12:32:34 +08:00
|
|
|
static int rbd_dev_device_setup(struct rbd_device *rbd_dev)
|
2013-04-27 04:44:36 +08:00
|
|
|
{
|
2012-10-31 04:47:17 +08:00
|
|
|
int ret;
|
2013-04-27 22:59:30 +08:00
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
/* Record our major and minor device numbers. */
|
2012-10-31 04:47:17 +08:00
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
if (!single_major) {
|
|
|
|
ret = register_blkdev(0, rbd_dev->name);
|
|
|
|
if (ret < 0)
|
2016-08-12 21:45:52 +08:00
|
|
|
goto err_out_unlock;
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
|
|
|
|
rbd_dev->major = ret;
|
|
|
|
rbd_dev->minor = 0;
|
|
|
|
} else {
|
|
|
|
rbd_dev->major = rbd_major;
|
|
|
|
rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id);
|
|
|
|
}
|
2012-10-31 04:47:17 +08:00
|
|
|
|
|
|
|
/* Set up the blkdev mapping. */
|
|
|
|
|
|
|
|
ret = rbd_init_disk(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
goto err_out_blkdev;
|
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
ret = rbd_dev_mapping_set(rbd_dev);
|
2012-10-31 04:47:17 +08:00
|
|
|
if (ret)
|
|
|
|
goto err_out_disk;
|
2014-08-04 22:04:39 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE);
|
2017-10-12 18:35:19 +08:00
|
|
|
set_disk_ro(rbd_dev->disk, rbd_dev->opts->read_only);
|
2013-05-06 22:51:29 +08:00
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id);
|
2013-05-06 22:51:29 +08:00
|
|
|
if (ret)
|
2014-10-09 21:06:01 +08:00
|
|
|
goto err_out_mapping;
|
2012-10-31 04:47:17 +08:00
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
up_write(&rbd_dev->header_rwsem);
|
2017-04-13 18:17:38 +08:00
|
|
|
return 0;
|
2012-10-31 08:40:33 +08:00
|
|
|
|
2013-05-06 22:51:29 +08:00
|
|
|
err_out_mapping:
|
|
|
|
rbd_dev_mapping_clear(rbd_dev);
|
2012-10-31 04:47:17 +08:00
|
|
|
err_out_disk:
|
|
|
|
rbd_free_disk(rbd_dev);
|
|
|
|
err_out_blkdev:
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
if (!single_major)
|
|
|
|
unregister_blkdev(rbd_dev->major, rbd_dev->name);
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
err_out_unlock:
|
|
|
|
up_write(&rbd_dev->header_rwsem);
|
2012-10-31 04:47:17 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-04-27 22:59:30 +08:00
|
|
|
static int rbd_dev_header_name(struct rbd_device *rbd_dev)
|
|
|
|
{
|
|
|
|
struct rbd_spec *spec = rbd_dev->spec;
|
2016-04-30 02:01:25 +08:00
|
|
|
int ret;
|
2013-04-27 22:59:30 +08:00
|
|
|
|
|
|
|
/* Record the header object name for this rbd image. */
|
|
|
|
|
|
|
|
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
|
|
|
|
if (rbd_dev->image_format == 1)
|
2016-04-30 02:01:25 +08:00
|
|
|
ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
|
|
|
|
spec->image_name, RBD_SUFFIX);
|
2013-04-27 22:59:30 +08:00
|
|
|
else
|
2016-04-30 02:01:25 +08:00
|
|
|
ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
|
|
|
|
RBD_HEADER_PREFIX, spec->image_id);
|
2013-04-27 22:59:30 +08:00
|
|
|
|
2016-04-30 02:01:25 +08:00
|
|
|
return ret;
|
2013-04-27 22:59:30 +08:00
|
|
|
}
|
|
|
|
|
2013-04-29 12:32:34 +08:00
|
|
|
static void rbd_dev_image_release(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2013-04-29 12:32:34 +08:00
|
|
|
rbd_dev_unprobe(rbd_dev);
|
2017-04-13 18:17:37 +08:00
|
|
|
if (rbd_dev->opts)
|
|
|
|
rbd_unregister_watch(rbd_dev);
|
2013-04-29 12:32:34 +08:00
|
|
|
rbd_dev->image_format = 0;
|
|
|
|
kfree(rbd_dev->spec->image_id);
|
|
|
|
rbd_dev->spec->image_id = NULL;
|
2013-04-29 12:32:34 +08:00
|
|
|
}
|
|
|
|
|
2012-07-11 09:30:11 +08:00
|
|
|
/*
|
|
|
|
* Probe for the existence of the header object for the given rbd
|
2013-05-07 06:40:33 +08:00
|
|
|
* device. If this image is the one being mapped (i.e., not a
|
|
|
|
* parent), initiate a watch on its header object before using that
|
|
|
|
* object to get detailed information about the rbd image.
|
2012-07-11 09:30:11 +08:00
|
|
|
*/
|
2015-10-12 01:38:00 +08:00
|
|
|
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
|
2012-07-11 09:30:11 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/*
|
2013-05-14 09:35:37 +08:00
|
|
|
* Get the id from the image id object. Unless there's an
|
|
|
|
* error, rbd_dev->spec->image_id will be filled in with
|
|
|
|
* a dynamically-allocated string, and rbd_dev->image_format
|
|
|
|
* will be set to either 1 or 2.
|
2012-07-11 09:30:11 +08:00
|
|
|
*/
|
|
|
|
ret = rbd_dev_image_id(rbd_dev);
|
|
|
|
if (ret)
|
2013-04-26 12:15:08 +08:00
|
|
|
return ret;
|
|
|
|
|
2013-04-27 22:59:30 +08:00
|
|
|
ret = rbd_dev_header_name(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
goto err_out_format;
|
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
if (!depth) {
|
2016-08-12 22:11:41 +08:00
|
|
|
ret = rbd_register_watch(rbd_dev);
|
2015-03-05 15:47:22 +08:00
|
|
|
if (ret) {
|
|
|
|
if (ret == -ENOENT)
|
2018-07-03 21:28:43 +08:00
|
|
|
pr_info("image %s/%s%s%s does not exist\n",
|
2015-03-05 15:47:22 +08:00
|
|
|
rbd_dev->spec->pool_name,
|
2018-07-03 21:28:43 +08:00
|
|
|
rbd_dev->spec->pool_ns ?: "",
|
|
|
|
rbd_dev->spec->pool_ns ? "/" : "",
|
2015-03-05 15:47:22 +08:00
|
|
|
rbd_dev->spec->image_name);
|
2016-04-30 02:01:25 +08:00
|
|
|
goto err_out_format;
|
2015-03-05 15:47:22 +08:00
|
|
|
}
|
2013-05-07 06:40:33 +08:00
|
|
|
}
|
2013-04-27 22:59:31 +08:00
|
|
|
|
2014-07-23 21:11:19 +08:00
|
|
|
ret = rbd_dev_header_info(rbd_dev);
|
2013-04-26 12:15:08 +08:00
|
|
|
if (ret)
|
2013-04-27 22:59:31 +08:00
|
|
|
goto err_out_watch;
|
2012-10-31 04:47:17 +08:00
|
|
|
|
2014-07-23 21:11:20 +08:00
|
|
|
/*
|
|
|
|
* If this image is the one being mapped, we have pool name and
|
|
|
|
* id, image name and id, and snap name - need to fill snap id.
|
|
|
|
* Otherwise this is a parent image, identified by pool, image
|
|
|
|
* and snap ids - need to fill in names for those ids.
|
|
|
|
*/
|
2015-10-12 01:38:00 +08:00
|
|
|
if (!depth)
|
2014-07-23 21:11:20 +08:00
|
|
|
ret = rbd_spec_fill_snap_id(rbd_dev);
|
|
|
|
else
|
|
|
|
ret = rbd_spec_fill_names(rbd_dev);
|
2015-03-05 15:47:22 +08:00
|
|
|
if (ret) {
|
|
|
|
if (ret == -ENOENT)
|
2018-07-03 21:28:43 +08:00
|
|
|
pr_info("snap %s/%s%s%s@%s does not exist\n",
|
2015-03-05 15:47:22 +08:00
|
|
|
rbd_dev->spec->pool_name,
|
2018-07-03 21:28:43 +08:00
|
|
|
rbd_dev->spec->pool_ns ?: "",
|
|
|
|
rbd_dev->spec->pool_ns ? "/" : "",
|
2015-03-05 15:47:22 +08:00
|
|
|
rbd_dev->spec->image_name,
|
|
|
|
rbd_dev->spec->snap_name);
|
2013-04-30 13:44:33 +08:00
|
|
|
goto err_out_probe;
|
2015-03-05 15:47:22 +08:00
|
|
|
}
|
2013-04-27 22:59:30 +08:00
|
|
|
|
2014-07-24 14:42:13 +08:00
|
|
|
if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
|
|
|
|
ret = rbd_dev_v2_parent_info(rbd_dev);
|
|
|
|
if (ret)
|
|
|
|
goto err_out_probe;
|
|
|
|
}
|
|
|
|
|
2015-10-12 01:38:00 +08:00
|
|
|
ret = rbd_dev_probe_parent(rbd_dev, depth);
|
2013-05-06 22:51:30 +08:00
|
|
|
if (ret)
|
|
|
|
goto err_out_probe;
|
|
|
|
|
|
|
|
dout("discovered format %u image, header name is %s\n",
|
2016-04-30 02:01:25 +08:00
|
|
|
rbd_dev->image_format, rbd_dev->header_oid.name);
|
2013-05-06 22:51:30 +08:00
|
|
|
return 0;
|
2014-07-24 14:42:13 +08:00
|
|
|
|
2013-04-29 12:32:34 +08:00
|
|
|
err_out_probe:
|
|
|
|
rbd_dev_unprobe(rbd_dev);
|
2013-04-27 22:59:31 +08:00
|
|
|
err_out_watch:
|
2015-10-12 01:38:00 +08:00
|
|
|
if (!depth)
|
2016-08-12 22:11:41 +08:00
|
|
|
rbd_unregister_watch(rbd_dev);
|
2013-04-27 22:59:30 +08:00
|
|
|
err_out_format:
|
|
|
|
rbd_dev->image_format = 0;
|
2013-04-26 12:15:08 +08:00
|
|
|
kfree(rbd_dev->spec->image_id);
|
|
|
|
rbd_dev->spec->image_id = NULL;
|
2012-07-11 09:30:11 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
static ssize_t do_rbd_add(struct bus_type *bus,
|
|
|
|
const char *buf,
|
|
|
|
size_t count)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2012-07-10 10:04:23 +08:00
|
|
|
struct rbd_device *rbd_dev = NULL;
|
2012-10-26 12:34:41 +08:00
|
|
|
struct ceph_options *ceph_opts = NULL;
|
2012-10-26 12:34:41 +08:00
|
|
|
struct rbd_options *rbd_opts = NULL;
|
2012-10-26 12:34:42 +08:00
|
|
|
struct rbd_spec *spec = NULL;
|
2012-10-26 12:34:42 +08:00
|
|
|
struct rbd_client *rbdc;
|
2015-10-15 21:38:57 +08:00
|
|
|
int rc;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
if (!try_module_get(THIS_MODULE))
|
|
|
|
return -ENODEV;
|
|
|
|
|
|
|
|
/* parse add command */
|
2012-10-26 12:34:42 +08:00
|
|
|
rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec);
|
2012-10-26 12:34:41 +08:00
|
|
|
if (rc < 0)
|
2015-10-16 23:09:24 +08:00
|
|
|
goto out;
|
2012-10-26 12:34:41 +08:00
|
|
|
|
2012-10-26 12:34:42 +08:00
|
|
|
rbdc = rbd_get_client(ceph_opts);
|
|
|
|
if (IS_ERR(rbdc)) {
|
|
|
|
rc = PTR_ERR(rbdc);
|
2012-10-26 12:34:41 +08:00
|
|
|
goto err_out_args;
|
2012-10-26 12:34:42 +08:00
|
|
|
}
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
/* pick the pool */
|
2018-02-22 20:43:24 +08:00
|
|
|
rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name);
|
2015-03-05 15:47:22 +08:00
|
|
|
if (rc < 0) {
|
|
|
|
if (rc == -ENOENT)
|
|
|
|
pr_info("pool %s does not exist\n", spec->pool_name);
|
2010-08-13 07:11:25 +08:00
|
|
|
goto err_out_client;
|
2015-03-05 15:47:22 +08:00
|
|
|
}
|
2013-04-26 22:43:47 +08:00
|
|
|
spec->pool_id = (u64)rc;
|
2012-10-26 12:34:42 +08:00
|
|
|
|
2015-06-22 18:24:48 +08:00
|
|
|
rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts);
|
2015-10-15 21:38:57 +08:00
|
|
|
if (!rbd_dev) {
|
|
|
|
rc = -ENOMEM;
|
2012-10-26 12:34:42 +08:00
|
|
|
goto err_out_client;
|
2015-10-15 21:38:57 +08:00
|
|
|
}
|
2012-10-26 12:34:42 +08:00
|
|
|
rbdc = NULL; /* rbd_dev now owns this */
|
|
|
|
spec = NULL; /* rbd_dev now owns this */
|
2015-06-22 18:24:48 +08:00
|
|
|
rbd_opts = NULL; /* rbd_dev now owns this */
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2016-08-19 00:38:45 +08:00
|
|
|
rbd_dev->config_info = kstrdup(buf, GFP_KERNEL);
|
|
|
|
if (!rbd_dev->config_info) {
|
|
|
|
rc = -ENOMEM;
|
|
|
|
goto err_out_rbd_dev;
|
|
|
|
}
|
|
|
|
|
rbd: fix rbd map vs notify races
A while ago, commit 9875201e1049 ("rbd: fix use-after free of
rbd_dev->disk") fixed rbd unmap vs notify race by introducing
an exported wrapper for flushing notifies and sticking it into
do_rbd_remove().
A similar problem exists on the rbd map path, though: the watch is
registered in rbd_dev_image_probe(), while the disk is set up quite
a few steps later, in rbd_dev_device_setup(). Nothing prevents
a notify from coming in and crashing on a NULL rbd_dev->disk:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
Call Trace:
[<ffffffffa0508344>] rbd_watch_cb+0x34/0x180 [rbd]
[<ffffffffa04bd290>] do_event_work+0x40/0xb0 [libceph]
[<ffffffff8109d5db>] process_one_work+0x17b/0x470
[<ffffffff8109e3ab>] worker_thread+0x11b/0x400
[<ffffffff8109e290>] ? rescuer_thread+0x400/0x400
[<ffffffff810a5acf>] kthread+0xcf/0xe0
[<ffffffff810b41b3>] ? finish_task_switch+0x53/0x170
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
[<ffffffff81645dd8>] ret_from_fork+0x58/0x90
[<ffffffff810a5a00>] ? kthread_create_on_node+0x140/0x140
RIP [<ffffffffa050828a>] rbd_dev_refresh+0xfa/0x180 [rbd]
If an error occurs during rbd map, we have to error out, potentially
tearing down a watch. Just like on rbd unmap, notifies have to be
flushed, otherwise rbd_watch_cb() may end up trying to read in the
image header after rbd_dev_image_release() has run:
Assertion failure in rbd_dev_header_info() at line 4722:
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
Call Trace:
[<ffffffff81cccee0>] ? rbd_parent_request_create+0x150/0x150
[<ffffffff81cd4e59>] rbd_dev_refresh+0x59/0x390
[<ffffffff81cd5229>] rbd_watch_cb+0x69/0x290
[<ffffffff81fde9bf>] do_event_work+0x10f/0x1c0
[<ffffffff81107799>] process_one_work+0x689/0x1a80
[<ffffffff811076f7>] ? process_one_work+0x5e7/0x1a80
[<ffffffff81132065>] ? finish_task_switch+0x225/0x640
[<ffffffff81107110>] ? pwq_dec_nr_in_flight+0x2b0/0x2b0
[<ffffffff81108c69>] worker_thread+0xd9/0x1320
[<ffffffff81108b90>] ? process_one_work+0x1a80/0x1a80
[<ffffffff8111b02d>] kthread+0x21d/0x2e0
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
[<ffffffff82022802>] ret_from_fork+0x22/0x40
[<ffffffff8111ae10>] ? kthread_stop+0x550/0x550
RIP [<ffffffff81ccd8f9>] rbd_dev_header_info+0xa19/0x1e30
To fix this, a) check if RBD_DEV_FLAG_EXISTS is set before calling
revalidate_disk(), b) move ceph_osdc_flush_notifies() call into
rbd_dev_header_unwatch_sync() to cover rbd map error paths and c) turn
header read-in into a critical section. The latter also happens to
take care of rbd map foo@bar vs rbd snap rm foo@bar race.
Fixes: http://tracker.ceph.com/issues/15490
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
2016-04-15 22:22:16 +08:00
|
|
|
down_write(&rbd_dev->header_rwsem);
|
2015-10-12 01:38:00 +08:00
|
|
|
rc = rbd_dev_image_probe(rbd_dev, 0);
|
2016-08-19 00:38:45 +08:00
|
|
|
if (rc < 0) {
|
|
|
|
up_write(&rbd_dev->header_rwsem);
|
2012-10-26 12:34:42 +08:00
|
|
|
goto err_out_rbd_dev;
|
2016-08-19 00:38:45 +08:00
|
|
|
}
|
2012-08-30 06:11:07 +08:00
|
|
|
|
2013-05-07 06:40:33 +08:00
|
|
|
/* If we are mapping a snapshot it must be marked read-only */
|
|
|
|
if (rbd_dev->spec->snap_id != CEPH_NOSNAP)
|
2017-10-12 18:35:19 +08:00
|
|
|
rbd_dev->opts->read_only = true;
|
2013-05-07 06:40:33 +08:00
|
|
|
|
2019-01-30 22:14:48 +08:00
|
|
|
if (rbd_dev->opts->alloc_size > rbd_dev->layout.object_size) {
|
|
|
|
rbd_warn(rbd_dev, "alloc_size adjusted to %u",
|
|
|
|
rbd_dev->layout.object_size);
|
|
|
|
rbd_dev->opts->alloc_size = rbd_dev->layout.object_size;
|
|
|
|
}
|
|
|
|
|
2013-04-29 12:32:34 +08:00
|
|
|
rc = rbd_dev_device_setup(rbd_dev);
|
2017-04-13 18:17:37 +08:00
|
|
|
if (rc)
|
2017-04-13 18:17:37 +08:00
|
|
|
goto err_out_image_probe;
|
2013-05-14 09:35:37 +08:00
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
if (rbd_dev->opts->exclusive) {
|
|
|
|
rc = rbd_add_acquire_lock(rbd_dev);
|
|
|
|
if (rc)
|
|
|
|
goto err_out_device_setup;
|
2013-05-14 09:35:37 +08:00
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
/* Everything's ready. Announce the disk to the world. */
|
|
|
|
|
|
|
|
rc = device_add(&rbd_dev->dev);
|
|
|
|
if (rc)
|
2017-04-13 18:17:39 +08:00
|
|
|
goto err_out_image_lock;
|
2017-04-13 18:17:38 +08:00
|
|
|
|
|
|
|
add_disk(rbd_dev->disk);
|
|
|
|
/* see rbd_init_disk() */
|
|
|
|
blk_put_queue(rbd_dev->disk->queue);
|
|
|
|
|
|
|
|
spin_lock(&rbd_dev_list_lock);
|
|
|
|
list_add_tail(&rbd_dev->node, &rbd_dev_list);
|
|
|
|
spin_unlock(&rbd_dev_list_lock);
|
|
|
|
|
|
|
|
pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name,
|
|
|
|
(unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT,
|
|
|
|
rbd_dev->header.features);
|
2015-10-16 23:09:24 +08:00
|
|
|
rc = count;
|
|
|
|
out:
|
|
|
|
module_put(THIS_MODULE);
|
|
|
|
return rc;
|
2013-04-29 12:32:34 +08:00
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
err_out_image_lock:
|
|
|
|
rbd_dev_image_unlock(rbd_dev);
|
2017-04-13 18:17:38 +08:00
|
|
|
err_out_device_setup:
|
|
|
|
rbd_dev_device_release(rbd_dev);
|
2017-04-13 18:17:37 +08:00
|
|
|
err_out_image_probe:
|
|
|
|
rbd_dev_image_release(rbd_dev);
|
2012-10-26 12:34:42 +08:00
|
|
|
err_out_rbd_dev:
|
|
|
|
rbd_dev_destroy(rbd_dev);
|
2012-10-26 12:34:42 +08:00
|
|
|
err_out_client:
|
2012-10-26 12:34:42 +08:00
|
|
|
rbd_put_client(rbdc);
|
2012-10-26 12:34:41 +08:00
|
|
|
err_out_args:
|
2012-10-26 12:34:42 +08:00
|
|
|
rbd_spec_put(spec);
|
2015-06-22 18:24:48 +08:00
|
|
|
kfree(rbd_opts);
|
2015-10-16 23:09:24 +08:00
|
|
|
goto out;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count)
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
{
|
|
|
|
if (single_major)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
return do_rbd_add(bus, buf, count);
|
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t add_single_major_store(struct bus_type *bus, const char *buf,
|
|
|
|
size_t count)
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
{
|
|
|
|
return do_rbd_add(bus, buf, count);
|
|
|
|
}
|
|
|
|
|
2013-04-27 04:44:36 +08:00
|
|
|
static void rbd_dev_remove_parent(struct rbd_device *rbd_dev)
|
|
|
|
{
|
2013-04-27 04:44:36 +08:00
|
|
|
while (rbd_dev->parent) {
|
2013-04-27 04:44:36 +08:00
|
|
|
struct rbd_device *first = rbd_dev;
|
|
|
|
struct rbd_device *second = first->parent;
|
|
|
|
struct rbd_device *third;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Follow to the parent with no grandparent and
|
|
|
|
* remove it.
|
|
|
|
*/
|
|
|
|
while (second && (third = second->parent)) {
|
|
|
|
first = second;
|
|
|
|
second = third;
|
|
|
|
}
|
2013-04-27 04:44:36 +08:00
|
|
|
rbd_assert(second);
|
2013-04-29 12:32:34 +08:00
|
|
|
rbd_dev_image_release(second);
|
2017-04-13 18:17:37 +08:00
|
|
|
rbd_dev_destroy(second);
|
2013-04-27 04:44:36 +08:00
|
|
|
first->parent = NULL;
|
|
|
|
first->parent_overlap = 0;
|
|
|
|
|
|
|
|
rbd_assert(first->parent_spec);
|
2013-04-27 04:44:36 +08:00
|
|
|
rbd_spec_put(first->parent_spec);
|
|
|
|
first->parent_spec = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
static ssize_t do_rbd_remove(struct bus_type *bus,
|
|
|
|
const char *buf,
|
|
|
|
size_t count)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
|
|
|
struct rbd_device *rbd_dev = NULL;
|
2013-06-01 04:17:01 +08:00
|
|
|
struct list_head *tmp;
|
|
|
|
int dev_id;
|
2016-08-19 00:38:45 +08:00
|
|
|
char opt_buf[6];
|
|
|
|
bool force = false;
|
2013-04-27 22:59:30 +08:00
|
|
|
int ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2016-08-19 00:38:45 +08:00
|
|
|
dev_id = -1;
|
|
|
|
opt_buf[0] = '\0';
|
|
|
|
sscanf(buf, "%d %5s", &dev_id, opt_buf);
|
|
|
|
if (dev_id < 0) {
|
|
|
|
pr_err("dev_id out of range\n");
|
2010-08-13 07:11:25 +08:00
|
|
|
return -EINVAL;
|
2016-08-19 00:38:45 +08:00
|
|
|
}
|
|
|
|
if (opt_buf[0] != '\0') {
|
|
|
|
if (!strcmp(opt_buf, "force")) {
|
|
|
|
force = true;
|
|
|
|
} else {
|
|
|
|
pr_err("bad remove option at '%s'\n", opt_buf);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
}
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2013-06-01 04:17:01 +08:00
|
|
|
ret = -ENOENT;
|
|
|
|
spin_lock(&rbd_dev_list_lock);
|
|
|
|
list_for_each(tmp, &rbd_dev_list) {
|
|
|
|
rbd_dev = list_entry(tmp, struct rbd_device, node);
|
|
|
|
if (rbd_dev->dev_id == dev_id) {
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
2012-11-16 23:29:16 +08:00
|
|
|
}
|
2013-06-01 04:17:01 +08:00
|
|
|
if (!ret) {
|
|
|
|
spin_lock_irq(&rbd_dev->lock);
|
2016-08-19 00:38:45 +08:00
|
|
|
if (rbd_dev->open_count && !force)
|
2013-06-01 04:17:01 +08:00
|
|
|
ret = -EBUSY;
|
rbd: don't return 0 on unmap if RBD_DEV_FLAG_REMOVING is set
There is a window between when RBD_DEV_FLAG_REMOVING is set and when
the device is removed from rbd_dev_list. During this window, we set
"already" and return 0.
Returning 0 from write(2) can confuse userspace tools because
0 indicates that nothing was written. In particular, "rbd unmap"
will retry the write multiple times a second:
10:28:05.463299 write(4, "0", 1) = 0
10:28:05.463509 write(4, "0", 1) = 0
10:28:05.463720 write(4, "0", 1) = 0
10:28:05.463942 write(4, "0", 1) = 0
10:28:05.464155 write(4, "0", 1) = 0
Cc: stable@vger.kernel.org
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Tested-by: Dongsheng Yang <dongsheng.yang@easystack.cn>
2019-01-09 02:47:38 +08:00
|
|
|
else if (test_and_set_bit(RBD_DEV_FLAG_REMOVING,
|
|
|
|
&rbd_dev->flags))
|
|
|
|
ret = -EINPROGRESS;
|
2013-06-01 04:17:01 +08:00
|
|
|
spin_unlock_irq(&rbd_dev->lock);
|
|
|
|
}
|
|
|
|
spin_unlock(&rbd_dev_list_lock);
|
rbd: don't return 0 on unmap if RBD_DEV_FLAG_REMOVING is set
There is a window between when RBD_DEV_FLAG_REMOVING is set and when
the device is removed from rbd_dev_list. During this window, we set
"already" and return 0.
Returning 0 from write(2) can confuse userspace tools because
0 indicates that nothing was written. In particular, "rbd unmap"
will retry the write multiple times a second:
10:28:05.463299 write(4, "0", 1) = 0
10:28:05.463509 write(4, "0", 1) = 0
10:28:05.463720 write(4, "0", 1) = 0
10:28:05.463942 write(4, "0", 1) = 0
10:28:05.464155 write(4, "0", 1) = 0
Cc: stable@vger.kernel.org
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Tested-by: Dongsheng Yang <dongsheng.yang@easystack.cn>
2019-01-09 02:47:38 +08:00
|
|
|
if (ret)
|
2013-06-01 04:17:01 +08:00
|
|
|
return ret;
|
2013-06-01 04:17:01 +08:00
|
|
|
|
2016-08-19 00:38:45 +08:00
|
|
|
if (force) {
|
|
|
|
/*
|
|
|
|
* Prevent new IO from being queued and wait for existing
|
|
|
|
* IO to complete/fail.
|
|
|
|
*/
|
|
|
|
blk_mq_freeze_queue(rbd_dev->disk->queue);
|
|
|
|
blk_set_queue_dying(rbd_dev->disk->queue);
|
|
|
|
}
|
|
|
|
|
2017-04-13 18:17:38 +08:00
|
|
|
del_gendisk(rbd_dev->disk);
|
|
|
|
spin_lock(&rbd_dev_list_lock);
|
|
|
|
list_del_init(&rbd_dev->node);
|
|
|
|
spin_unlock(&rbd_dev_list_lock);
|
|
|
|
device_del(&rbd_dev->dev);
|
2013-12-17 00:02:40 +08:00
|
|
|
|
2017-04-13 18:17:39 +08:00
|
|
|
rbd_dev_image_unlock(rbd_dev);
|
2015-10-16 23:09:24 +08:00
|
|
|
rbd_dev_device_release(rbd_dev);
|
2013-04-29 12:32:34 +08:00
|
|
|
rbd_dev_image_release(rbd_dev);
|
2017-04-13 18:17:37 +08:00
|
|
|
rbd_dev_destroy(rbd_dev);
|
2013-06-01 04:17:01 +08:00
|
|
|
return count;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t remove_store(struct bus_type *bus, const char *buf, size_t count)
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
{
|
|
|
|
if (single_major)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
return do_rbd_remove(bus, buf, count);
|
|
|
|
}
|
|
|
|
|
2018-12-21 15:54:38 +08:00
|
|
|
static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf,
|
|
|
|
size_t count)
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
{
|
|
|
|
return do_rbd_remove(bus, buf, count);
|
|
|
|
}
|
|
|
|
|
2010-08-13 07:11:25 +08:00
|
|
|
/*
|
|
|
|
* create control files in sysfs
|
2010-11-20 06:51:04 +08:00
|
|
|
* /sys/bus/rbd/...
|
2010-08-13 07:11:25 +08:00
|
|
|
*/
|
2018-08-12 23:06:54 +08:00
|
|
|
static int __init rbd_sysfs_init(void)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2010-11-20 06:51:04 +08:00
|
|
|
int ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-02-08 02:03:36 +08:00
|
|
|
ret = device_register(&rbd_root_dev);
|
2012-01-25 00:08:36 +08:00
|
|
|
if (ret < 0)
|
2010-11-20 06:51:04 +08:00
|
|
|
return ret;
|
2010-08-13 07:11:25 +08:00
|
|
|
|
2012-02-08 02:03:36 +08:00
|
|
|
ret = bus_register(&rbd_bus_type);
|
|
|
|
if (ret < 0)
|
|
|
|
device_unregister(&rbd_root_dev);
|
2010-08-13 07:11:25 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2018-08-12 23:06:54 +08:00
|
|
|
static void __exit rbd_sysfs_cleanup(void)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2010-11-20 06:51:04 +08:00
|
|
|
bus_unregister(&rbd_bus_type);
|
2012-02-08 02:03:36 +08:00
|
|
|
device_unregister(&rbd_root_dev);
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2018-08-12 23:06:54 +08:00
|
|
|
static int __init rbd_slab_init(void)
|
2013-05-02 01:43:03 +08:00
|
|
|
{
|
|
|
|
rbd_assert(!rbd_img_request_cache);
|
2016-03-13 15:17:32 +08:00
|
|
|
rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0);
|
2013-05-02 01:43:03 +08:00
|
|
|
if (!rbd_img_request_cache)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
rbd_assert(!rbd_obj_request_cache);
|
2016-03-13 15:17:32 +08:00
|
|
|
rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
|
2013-05-02 01:43:04 +08:00
|
|
|
if (!rbd_obj_request_cache)
|
|
|
|
goto out_err;
|
|
|
|
|
2017-01-26 01:16:23 +08:00
|
|
|
return 0;
|
2013-05-02 01:43:03 +08:00
|
|
|
|
2017-01-26 01:16:23 +08:00
|
|
|
out_err:
|
2013-05-02 01:43:03 +08:00
|
|
|
kmem_cache_destroy(rbd_img_request_cache);
|
|
|
|
rbd_img_request_cache = NULL;
|
2013-05-02 01:43:03 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rbd_slab_exit(void)
|
|
|
|
{
|
2013-05-02 01:43:03 +08:00
|
|
|
rbd_assert(rbd_obj_request_cache);
|
|
|
|
kmem_cache_destroy(rbd_obj_request_cache);
|
|
|
|
rbd_obj_request_cache = NULL;
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
rbd_assert(rbd_img_request_cache);
|
|
|
|
kmem_cache_destroy(rbd_img_request_cache);
|
|
|
|
rbd_img_request_cache = NULL;
|
|
|
|
}
|
|
|
|
|
2013-02-20 02:25:56 +08:00
|
|
|
static int __init rbd_init(void)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
|
|
|
int rc;
|
|
|
|
|
2013-01-31 01:13:33 +08:00
|
|
|
if (!libceph_compatible(NULL)) {
|
|
|
|
rbd_warn(NULL, "libceph incompatibility (quitting)");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2013-12-13 21:28:57 +08:00
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
rc = rbd_slab_init();
|
2010-08-13 07:11:25 +08:00
|
|
|
if (rc)
|
|
|
|
return rc;
|
2013-12-13 21:28:57 +08:00
|
|
|
|
2014-10-09 21:06:01 +08:00
|
|
|
/*
|
|
|
|
* The number of active work items is limited by the number of
|
2015-04-22 23:28:13 +08:00
|
|
|
* rbd devices * queue depth, so leave @max_active at default.
|
2014-10-09 21:06:01 +08:00
|
|
|
*/
|
|
|
|
rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0);
|
|
|
|
if (!rbd_wq) {
|
|
|
|
rc = -ENOMEM;
|
|
|
|
goto err_out_slab;
|
|
|
|
}
|
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
if (single_major) {
|
|
|
|
rbd_major = register_blkdev(0, RBD_DRV_NAME);
|
|
|
|
if (rbd_major < 0) {
|
|
|
|
rc = rbd_major;
|
2014-10-09 21:06:01 +08:00
|
|
|
goto err_out_wq;
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-05-02 01:43:03 +08:00
|
|
|
rc = rbd_sysfs_init();
|
|
|
|
if (rc)
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
goto err_out_blkdev;
|
|
|
|
|
|
|
|
if (single_major)
|
|
|
|
pr_info("loaded (major %d)\n", rbd_major);
|
|
|
|
else
|
|
|
|
pr_info("loaded\n");
|
2013-05-02 01:43:03 +08:00
|
|
|
|
2013-12-13 21:28:57 +08:00
|
|
|
return 0;
|
|
|
|
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
err_out_blkdev:
|
|
|
|
if (single_major)
|
|
|
|
unregister_blkdev(rbd_major, RBD_DRV_NAME);
|
2014-10-09 21:06:01 +08:00
|
|
|
err_out_wq:
|
|
|
|
destroy_workqueue(rbd_wq);
|
2013-12-13 21:28:57 +08:00
|
|
|
err_out_slab:
|
|
|
|
rbd_slab_exit();
|
2013-05-02 01:43:03 +08:00
|
|
|
return rc;
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
2013-02-20 02:25:56 +08:00
|
|
|
static void __exit rbd_exit(void)
|
2010-08-13 07:11:25 +08:00
|
|
|
{
|
2014-05-20 19:46:04 +08:00
|
|
|
ida_destroy(&rbd_dev_id_ida);
|
2010-08-13 07:11:25 +08:00
|
|
|
rbd_sysfs_cleanup();
|
rbd: add support for single-major device number allocation scheme
Currently each rbd device is allocated its own major number, which
leads to a hard limit of 230-250 images mapped at once. This commit
adds support for a new single-major device number allocation scheme,
which is hidden behind a new single_major boolean module parameter and
is disabled by default for backwards compatibility reasons. (Old
userspace cannot correctly unmap images mapped under single-major
scheme and would essentially just unmap a random image, if that.)
$ rbd showmapped
id pool image snap device
0 rbd b100 - /dev/rbd0
1 rbd b101 - /dev/rbd1
2 rbd b102 - /dev/rbd2
3 rbd b103 - /dev/rbd3
Old scheme (modprobe rbd):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:24 /dev/rbd0
brw-rw---- 1 root disk 252, 0 Dec 10 12:28 /dev/rbd1
brw-rw---- 1 root disk 252, 1 Dec 10 12:28 /dev/rbd1p1
brw-rw---- 1 root disk 252, 2 Dec 10 12:28 /dev/rbd1p2
brw-rw---- 1 root disk 252, 3 Dec 10 12:28 /dev/rbd1p3
brw-rw---- 1 root disk 251, 0 Dec 10 12:28 /dev/rbd2
brw-rw---- 1 root disk 251, 1 Dec 10 12:28 /dev/rbd2p1
brw-rw---- 1 root disk 250, 0 Dec 10 12:24 /dev/rbd3
New scheme (modprobe rbd single_major=Y):
$ ls -l /dev/rbd*
brw-rw---- 1 root disk 253, 0 Dec 10 12:30 /dev/rbd0
brw-rw---- 1 root disk 253, 256 Dec 10 12:30 /dev/rbd1
brw-rw---- 1 root disk 253, 257 Dec 10 12:30 /dev/rbd1p1
brw-rw---- 1 root disk 253, 258 Dec 10 12:30 /dev/rbd1p2
brw-rw---- 1 root disk 253, 259 Dec 10 12:30 /dev/rbd1p3
brw-rw---- 1 root disk 253, 512 Dec 10 12:30 /dev/rbd2
brw-rw---- 1 root disk 253, 513 Dec 10 12:30 /dev/rbd2p1
brw-rw---- 1 root disk 253, 768 Dec 10 12:30 /dev/rbd3
(major 253 was assigned dynamically at module load time)
The new limit is 4096 images mapped at once, and it comes from the fact
that, as before, 256 minor numbers are reserved for each mapping.
(A follow-up commit changes the number of minors reserved and the way
we deal with partitions over that number.)
If single_major is set to true, two new sysfs interfaces show up:
/sys/bus/rbd/{add,remove}_single_major. These are to be used instead
of /sys/bus/rbd/{add,remove}, which are disabled for backwards
compatibility reasons outlined above.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
2013-12-13 21:28:57 +08:00
|
|
|
if (single_major)
|
|
|
|
unregister_blkdev(rbd_major, RBD_DRV_NAME);
|
2014-10-09 21:06:01 +08:00
|
|
|
destroy_workqueue(rbd_wq);
|
2013-05-02 01:43:03 +08:00
|
|
|
rbd_slab_exit();
|
2010-08-13 07:11:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
module_init(rbd_init);
|
|
|
|
module_exit(rbd_exit);
|
|
|
|
|
2013-06-01 09:13:09 +08:00
|
|
|
MODULE_AUTHOR("Alex Elder <elder@inktank.com>");
|
2010-08-13 07:11:25 +08:00
|
|
|
MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
|
|
|
|
MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
|
|
|
|
/* following authorship retained from original osdblk.c */
|
|
|
|
MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
|
|
|
|
|
2013-12-13 21:28:56 +08:00
|
|
|
MODULE_DESCRIPTION("RADOS Block Device (RBD) driver");
|
2010-08-13 07:11:25 +08:00
|
|
|
MODULE_LICENSE("GPL");
|