2018-04-04 01:23:33 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
2007-06-12 21:07:21 +08:00
|
|
|
/*
|
|
|
|
|
* Copyright (C) 2007 Oracle. All rights reserved.
|
|
|
|
|
*/
|
|
|
|
|
|
2022-10-19 22:50:49 +08:00
|
|
|
#include "messages.h"
|
2007-02-24 19:24:44 +08:00
|
|
|
#include "ctree.h"
|
|
|
|
|
#include "disk-io.h"
|
2007-05-03 03:53:43 +08:00
|
|
|
#include "print-tree.h"
|
2022-10-19 22:51:00 +08:00
|
|
|
#include "accessors.h"
|
2022-11-15 17:44:04 +08:00
|
|
|
#include "tree-checker.h"
|
2007-02-24 19:24:44 +08:00
|
|
|
|
2020-09-04 02:29:51 +08:00
|
|
|
struct root_name_map {
|
|
|
|
|
u64 id;
|
|
|
|
|
char name[16];
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static const struct root_name_map root_map[] = {
|
|
|
|
|
{ BTRFS_ROOT_TREE_OBJECTID, "ROOT_TREE" },
|
|
|
|
|
{ BTRFS_EXTENT_TREE_OBJECTID, "EXTENT_TREE" },
|
|
|
|
|
{ BTRFS_CHUNK_TREE_OBJECTID, "CHUNK_TREE" },
|
|
|
|
|
{ BTRFS_DEV_TREE_OBJECTID, "DEV_TREE" },
|
|
|
|
|
{ BTRFS_FS_TREE_OBJECTID, "FS_TREE" },
|
|
|
|
|
{ BTRFS_CSUM_TREE_OBJECTID, "CSUM_TREE" },
|
|
|
|
|
{ BTRFS_TREE_LOG_OBJECTID, "TREE_LOG" },
|
|
|
|
|
{ BTRFS_QUOTA_TREE_OBJECTID, "QUOTA_TREE" },
|
|
|
|
|
{ BTRFS_UUID_TREE_OBJECTID, "UUID_TREE" },
|
|
|
|
|
{ BTRFS_FREE_SPACE_TREE_OBJECTID, "FREE_SPACE_TREE" },
|
2021-12-16 04:40:07 +08:00
|
|
|
{ BTRFS_BLOCK_GROUP_TREE_OBJECTID, "BLOCK_GROUP_TREE" },
|
2020-09-04 02:29:51 +08:00
|
|
|
{ BTRFS_DATA_RELOC_TREE_OBJECTID, "DATA_RELOC_TREE" },
|
|
|
|
|
};
|
|
|
|
|
|
2020-12-17 00:18:44 +08:00
|
|
|
const char *btrfs_root_name(const struct btrfs_key *key, char *buf)
|
2020-09-04 02:29:51 +08:00
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
|
2020-12-17 00:18:44 +08:00
|
|
|
if (key->objectid == BTRFS_TREE_RELOC_OBJECTID) {
|
2020-09-04 02:29:51 +08:00
|
|
|
snprintf(buf, BTRFS_ROOT_NAME_BUF_LEN,
|
2020-12-17 00:18:44 +08:00
|
|
|
"TREE_RELOC offset=%llu", key->offset);
|
2020-09-04 02:29:51 +08:00
|
|
|
return buf;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < ARRAY_SIZE(root_map); i++) {
|
2020-12-17 00:18:44 +08:00
|
|
|
if (root_map[i].id == key->objectid)
|
2020-09-04 02:29:51 +08:00
|
|
|
return root_map[i].name;
|
|
|
|
|
}
|
|
|
|
|
|
2020-12-17 00:18:44 +08:00
|
|
|
snprintf(buf, BTRFS_ROOT_NAME_BUF_LEN, "%llu", key->objectid);
|
2020-09-04 02:29:51 +08:00
|
|
|
return buf;
|
|
|
|
|
}
|
|
|
|
|
|
2008-03-25 03:01:56 +08:00
|
|
|
static void print_chunk(struct extent_buffer *eb, struct btrfs_chunk *chunk)
|
|
|
|
|
{
|
|
|
|
|
int num_stripes = btrfs_chunk_num_stripes(eb, chunk);
|
|
|
|
|
int i;
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tchunk length %llu owner %llu type %llu num_stripes %d\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_chunk_length(eb, chunk), btrfs_chunk_owner(eb, chunk),
|
|
|
|
|
btrfs_chunk_type(eb, chunk), num_stripes);
|
2008-03-25 03:01:56 +08:00
|
|
|
for (i = 0 ; i < num_stripes ; i++) {
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\t\tstripe %d devid %llu offset %llu\n", i,
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_stripe_devid_nr(eb, chunk, i),
|
|
|
|
|
btrfs_stripe_offset_nr(eb, chunk, i));
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
static void print_dev_item(struct extent_buffer *eb,
|
|
|
|
|
struct btrfs_dev_item *dev_item)
|
|
|
|
|
{
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tdev item devid %llu total_bytes %llu bytes used %llu\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_device_id(eb, dev_item),
|
|
|
|
|
btrfs_device_total_bytes(eb, dev_item),
|
|
|
|
|
btrfs_device_bytes_used(eb, dev_item));
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
static void print_extent_data_ref(struct extent_buffer *eb,
|
|
|
|
|
struct btrfs_extent_data_ref *ref)
|
|
|
|
|
{
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_cont("extent data backref root %llu objectid %llu offset %llu count %u\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_extent_data_ref_root(eb, ref),
|
|
|
|
|
btrfs_extent_data_ref_objectid(eb, ref),
|
|
|
|
|
btrfs_extent_data_ref_offset(eb, ref),
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_extent_data_ref_count(eb, ref));
|
|
|
|
|
}
|
|
|
|
|
|
2014-07-03 01:20:48 +08:00
|
|
|
static void print_extent_item(struct extent_buffer *eb, int slot, int type)
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
{
|
|
|
|
|
struct btrfs_extent_item *ei;
|
|
|
|
|
struct btrfs_extent_inline_ref *iref;
|
|
|
|
|
struct btrfs_extent_data_ref *dref;
|
|
|
|
|
struct btrfs_shared_data_ref *sref;
|
|
|
|
|
struct btrfs_disk_key key;
|
|
|
|
|
unsigned long end;
|
|
|
|
|
unsigned long ptr;
|
2021-10-22 02:58:35 +08:00
|
|
|
u32 item_size = btrfs_item_size(eb, slot);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
u64 flags;
|
|
|
|
|
u64 offset;
|
2017-08-19 05:15:24 +08:00
|
|
|
int ref_index = 0;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
2018-06-26 22:20:59 +08:00
|
|
|
if (unlikely(item_size < sizeof(*ei))) {
|
2018-06-26 21:57:36 +08:00
|
|
|
btrfs_print_v0_err(eb->fs_info);
|
|
|
|
|
btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
|
|
|
|
|
}
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
|
|
|
|
ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
|
|
|
|
|
flags = btrfs_extent_flags(eb, ei);
|
|
|
|
|
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\textent refs %llu gen %llu flags %llu\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_extent_refs(eb, ei), btrfs_extent_generation(eb, ei),
|
|
|
|
|
flags);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
|
2014-07-03 01:20:48 +08:00
|
|
|
if ((type == BTRFS_EXTENT_ITEM_KEY) &&
|
|
|
|
|
flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
struct btrfs_tree_block_info *info;
|
|
|
|
|
info = (struct btrfs_tree_block_info *)(ei + 1);
|
|
|
|
|
btrfs_tree_block_key(eb, info, &key);
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\ttree block key (%llu %u %llu) level %d\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_disk_key_objectid(&key), key.type,
|
|
|
|
|
btrfs_disk_key_offset(&key),
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_tree_block_level(eb, info));
|
|
|
|
|
iref = (struct btrfs_extent_inline_ref *)(info + 1);
|
|
|
|
|
} else {
|
|
|
|
|
iref = (struct btrfs_extent_inline_ref *)(ei + 1);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ptr = (unsigned long)iref;
|
|
|
|
|
end = (unsigned long)ei + item_size;
|
|
|
|
|
while (ptr < end) {
|
|
|
|
|
iref = (struct btrfs_extent_inline_ref *)ptr;
|
|
|
|
|
type = btrfs_extent_inline_ref_type(eb, iref);
|
|
|
|
|
offset = btrfs_extent_inline_ref_offset(eb, iref);
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_info("\t\tref#%d: ", ref_index++);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
switch (type) {
|
|
|
|
|
case BTRFS_TREE_BLOCK_REF_KEY:
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_cont("tree block backref root %llu\n", offset);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_SHARED_BLOCK_REF_KEY:
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_cont("shared block backref parent %llu\n", offset);
|
|
|
|
|
/*
|
|
|
|
|
* offset is supposed to be a tree block which
|
|
|
|
|
* must be aligned to nodesize.
|
|
|
|
|
*/
|
btrfs: require only sector size alignment for parent eb bytenr
[BUG]
A completely sane converted fs will cause kernel warning at balance
time:
[ 1557.188633] BTRFS info (device sda7): relocating block group 8162107392 flags data
[ 1563.358078] BTRFS info (device sda7): found 11722 extents
[ 1563.358277] BTRFS info (device sda7): leaf 7989321728 gen 95 total ptrs 213 free space 3458 owner 2
[ 1563.358280] item 0 key (7984947200 169 0) itemoff 16250 itemsize 33
[ 1563.358281] extent refs 1 gen 90 flags 2
[ 1563.358282] ref#0: tree block backref root 4
[ 1563.358285] item 1 key (7985602560 169 0) itemoff 16217 itemsize 33
[ 1563.358286] extent refs 1 gen 93 flags 258
[ 1563.358287] ref#0: shared block backref parent 7985602560
[ 1563.358288] (parent 7985602560 is NOT ALIGNED to nodesize 16384)
[ 1563.358290] item 2 key (7985635328 169 0) itemoff 16184 itemsize 33
...
[ 1563.358995] BTRFS error (device sda7): eb 7989321728 invalid extent inline ref type 182
[ 1563.358996] ------------[ cut here ]------------
[ 1563.359005] WARNING: CPU: 14 PID: 2930 at 0xffffffff9f231766
Then with transaction abort, and obviously failed to balance the fs.
[CAUSE]
That mentioned inline ref type 182 is completely sane, it's
BTRFS_SHARED_BLOCK_REF_KEY, it's some extra check making kernel to
believe it's invalid.
Commit 64ecdb647ddb ("Btrfs: add one more sanity check for shared ref
type") introduced extra checks for backref type.
One of the requirement is, parent bytenr must be aligned to node size,
which is not correct.
One example is like this:
0 1G 1G+4K 2G 2G+4K
| |///////////////////|//| <- A chunk starts at 1G+4K
| | <- A tree block get reserved at bytenr 1G+4K
Then we have a valid tree block at bytenr 1G+4K, but not aligned to
nodesize (16K).
Such chunk is not ideal, but current kernel can handle it pretty well.
We may warn about such tree block in the future, but should not reject
them.
[FIX]
Change the alignment requirement from node size alignment to sector size
alignment.
Also, to make our lives a little easier, also output @iref when
btrfs_get_extent_inline_ref_type() failed, so we can locate the item
easier.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=205475
Fixes: 64ecdb647ddb ("Btrfs: add one more sanity check for shared ref type")
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
[ update comments and messages ]
Signed-off-by: David Sterba <dsterba@suse.com>
2020-08-26 17:26:43 +08:00
|
|
|
if (!IS_ALIGNED(offset, eb->fs_info->sectorsize))
|
|
|
|
|
pr_info(
|
|
|
|
|
"\t\t\t(parent %llu not aligned to sectorsize %u)\n",
|
|
|
|
|
offset, eb->fs_info->sectorsize);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_EXTENT_DATA_REF_KEY:
|
|
|
|
|
dref = (struct btrfs_extent_data_ref *)(&iref->offset);
|
|
|
|
|
print_extent_data_ref(eb, dref);
|
|
|
|
|
break;
|
|
|
|
|
case BTRFS_SHARED_DATA_REF_KEY:
|
|
|
|
|
sref = (struct btrfs_shared_data_ref *)(iref + 1);
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_cont("shared data backref parent %llu count %u\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
offset, btrfs_shared_data_ref_count(eb, sref));
|
2017-08-19 05:15:24 +08:00
|
|
|
/*
|
|
|
|
|
* offset is supposed to be a tree block which
|
|
|
|
|
* must be aligned to nodesize.
|
|
|
|
|
*/
|
|
|
|
|
if (!IS_ALIGNED(offset, eb->fs_info->nodesize))
|
btrfs: require only sector size alignment for parent eb bytenr
[BUG]
A completely sane converted fs will cause kernel warning at balance
time:
[ 1557.188633] BTRFS info (device sda7): relocating block group 8162107392 flags data
[ 1563.358078] BTRFS info (device sda7): found 11722 extents
[ 1563.358277] BTRFS info (device sda7): leaf 7989321728 gen 95 total ptrs 213 free space 3458 owner 2
[ 1563.358280] item 0 key (7984947200 169 0) itemoff 16250 itemsize 33
[ 1563.358281] extent refs 1 gen 90 flags 2
[ 1563.358282] ref#0: tree block backref root 4
[ 1563.358285] item 1 key (7985602560 169 0) itemoff 16217 itemsize 33
[ 1563.358286] extent refs 1 gen 93 flags 258
[ 1563.358287] ref#0: shared block backref parent 7985602560
[ 1563.358288] (parent 7985602560 is NOT ALIGNED to nodesize 16384)
[ 1563.358290] item 2 key (7985635328 169 0) itemoff 16184 itemsize 33
...
[ 1563.358995] BTRFS error (device sda7): eb 7989321728 invalid extent inline ref type 182
[ 1563.358996] ------------[ cut here ]------------
[ 1563.359005] WARNING: CPU: 14 PID: 2930 at 0xffffffff9f231766
Then with transaction abort, and obviously failed to balance the fs.
[CAUSE]
That mentioned inline ref type 182 is completely sane, it's
BTRFS_SHARED_BLOCK_REF_KEY, it's some extra check making kernel to
believe it's invalid.
Commit 64ecdb647ddb ("Btrfs: add one more sanity check for shared ref
type") introduced extra checks for backref type.
One of the requirement is, parent bytenr must be aligned to node size,
which is not correct.
One example is like this:
0 1G 1G+4K 2G 2G+4K
| |///////////////////|//| <- A chunk starts at 1G+4K
| | <- A tree block get reserved at bytenr 1G+4K
Then we have a valid tree block at bytenr 1G+4K, but not aligned to
nodesize (16K).
Such chunk is not ideal, but current kernel can handle it pretty well.
We may warn about such tree block in the future, but should not reject
them.
[FIX]
Change the alignment requirement from node size alignment to sector size
alignment.
Also, to make our lives a little easier, also output @iref when
btrfs_get_extent_inline_ref_type() failed, so we can locate the item
easier.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=205475
Fixes: 64ecdb647ddb ("Btrfs: add one more sanity check for shared ref type")
CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
[ update comments and messages ]
Signed-off-by: David Sterba <dsterba@suse.com>
2020-08-26 17:26:43 +08:00
|
|
|
pr_info(
|
|
|
|
|
"\t\t\t(parent %llu not aligned to sectorsize %u)\n",
|
|
|
|
|
offset, eb->fs_info->sectorsize);
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
break;
|
|
|
|
|
default:
|
2017-08-19 05:15:24 +08:00
|
|
|
pr_cont("(extent %llu has INVALID ref type %d)\n",
|
2017-08-19 05:15:21 +08:00
|
|
|
eb->start, type);
|
|
|
|
|
return;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
}
|
|
|
|
|
ptr += btrfs_extent_inline_ref_size(type);
|
|
|
|
|
}
|
|
|
|
|
WARN_ON(ptr > end);
|
|
|
|
|
}
|
|
|
|
|
|
2013-08-15 23:11:18 +08:00
|
|
|
static void print_uuid_item(struct extent_buffer *l, unsigned long offset,
|
|
|
|
|
u32 item_size)
|
|
|
|
|
{
|
|
|
|
|
if (!IS_ALIGNED(item_size, sizeof(u64))) {
|
2013-12-21 00:37:06 +08:00
|
|
|
pr_warn("BTRFS: uuid item with illegal size %lu!\n",
|
2013-08-15 23:11:18 +08:00
|
|
|
(unsigned long)item_size);
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
while (item_size) {
|
|
|
|
|
__le64 subvol_id;
|
|
|
|
|
|
|
|
|
|
read_extent_buffer(l, &subvol_id, offset, sizeof(subvol_id));
|
2020-09-15 20:18:23 +08:00
|
|
|
pr_info("\t\tsubvol_id %llu\n", le64_to_cpu(subvol_id));
|
2013-08-15 23:11:18 +08:00
|
|
|
item_size -= sizeof(u64);
|
|
|
|
|
offset += sizeof(u64);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2018-04-24 13:03:13 +08:00
|
|
|
/*
|
|
|
|
|
* Helper to output refs and locking status of extent buffer. Useful to debug
|
|
|
|
|
* race condition related problems.
|
|
|
|
|
*/
|
|
|
|
|
static void print_eb_refs_lock(struct extent_buffer *eb)
|
|
|
|
|
{
|
|
|
|
|
#ifdef CONFIG_BTRFS_DEBUG
|
btrfs: switch extent buffer tree lock to rw_semaphore
Historically we've implemented our own locking because we wanted to be
able to selectively spin or sleep based on what we were doing in the
tree. For instance, if all of our nodes were in cache then there's
rarely a reason to need to sleep waiting for node locks, as they'll
likely become available soon. At the time this code was written the
rw_semaphore didn't do adaptive spinning, and thus was orders of
magnitude slower than our home grown locking.
However now the opposite is the case. There are a few problems with how
we implement blocking locks, namely that we use a normal waitqueue and
simply wake everybody up in reverse sleep order. This leads to some
suboptimal performance behavior, and a lot of context switches in highly
contended cases. The rw_semaphores actually do this properly, and also
have adaptive spinning that works relatively well.
The locking code is also a bit of a bear to understand, and we lose the
benefit of lockdep for the most part because the blocking states of the
lock are simply ad-hoc and not mapped into lockdep.
So rework the locking code to drop all of this custom locking stuff, and
simply use a rw_semaphore for everything. This makes the locking much
simpler for everything, as we can now drop a lot of cruft and blocking
transitions. The performance numbers vary depending on the workload,
because generally speaking there doesn't tend to be a lot of contention
on the btree. However, on my test system which is an 80 core single
socket system with 256GiB of RAM and a 2TiB NVMe drive I get the
following results (with all debug options off):
dbench 200 baseline
Throughput 216.056 MB/sec 200 clients 200 procs max_latency=1471.197 ms
dbench 200 with patch
Throughput 737.188 MB/sec 200 clients 200 procs max_latency=714.346 ms
Previously we also used fs_mark to test this sort of contention, and
those results are far less impressive, mostly because there's not enough
tasks to really stress the locking
fs_mark -d /d[0-15] -S 0 -L 20 -n 100000 -s 0 -t 16
baseline
Average Files/sec: 160166.7
p50 Files/sec: 165832
p90 Files/sec: 123886
p99 Files/sec: 123495
real 3m26.527s
user 2m19.223s
sys 48m21.856s
patched
Average Files/sec: 164135.7
p50 Files/sec: 171095
p90 Files/sec: 122889
p99 Files/sec: 113819
real 3m29.660s
user 2m19.990s
sys 44m12.259s
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-08-20 23:46:09 +08:00
|
|
|
btrfs_info(eb->fs_info, "refs %u lock_owner %u current %u",
|
|
|
|
|
atomic_read(&eb->refs), eb->lock_owner, current->pid);
|
2018-04-24 13:03:13 +08:00
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2017-06-30 00:37:49 +08:00
|
|
|
void btrfs_print_leaf(struct extent_buffer *l)
|
2007-02-24 19:24:44 +08:00
|
|
|
{
|
2017-06-30 00:37:49 +08:00
|
|
|
struct btrfs_fs_info *fs_info;
|
2007-02-24 19:24:44 +08:00
|
|
|
int i;
|
2011-06-23 23:01:01 +08:00
|
|
|
u32 type, nr;
|
2007-03-14 04:47:54 +08:00
|
|
|
struct btrfs_root_item *ri;
|
2007-03-16 03:18:43 +08:00
|
|
|
struct btrfs_dir_item *di;
|
2007-03-21 03:57:25 +08:00
|
|
|
struct btrfs_inode_item *ii;
|
2007-04-27 04:46:15 +08:00
|
|
|
struct btrfs_block_group_item *bi;
|
2007-06-18 21:57:58 +08:00
|
|
|
struct btrfs_file_extent_item *fi;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
struct btrfs_extent_data_ref *dref;
|
|
|
|
|
struct btrfs_shared_data_ref *sref;
|
|
|
|
|
struct btrfs_dev_extent *dev_extent;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_key key;
|
|
|
|
|
struct btrfs_key found_key;
|
2007-03-16 03:18:43 +08:00
|
|
|
|
2011-06-23 23:01:01 +08:00
|
|
|
if (!l)
|
|
|
|
|
return;
|
|
|
|
|
|
2017-06-30 00:37:49 +08:00
|
|
|
fs_info = l->fs_info;
|
2011-06-23 23:01:01 +08:00
|
|
|
nr = btrfs_header_nritems(l);
|
|
|
|
|
|
2018-04-11 17:08:12 +08:00
|
|
|
btrfs_info(fs_info,
|
|
|
|
|
"leaf %llu gen %llu total ptrs %d free space %d owner %llu",
|
|
|
|
|
btrfs_header_bytenr(l), btrfs_header_generation(l), nr,
|
2019-03-20 21:36:46 +08:00
|
|
|
btrfs_leaf_free_space(l), btrfs_header_owner(l));
|
2018-04-24 13:03:13 +08:00
|
|
|
print_eb_refs_lock(l);
|
2007-02-24 19:24:44 +08:00
|
|
|
for (i = 0 ; i < nr ; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_key_to_cpu(l, &key, i);
|
2014-06-05 00:41:45 +08:00
|
|
|
type = key.type;
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\titem %d key (%llu %u %llu) itemoff %d itemsize %d\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
i, key.objectid, type, key.offset,
|
2021-10-22 02:58:35 +08:00
|
|
|
btrfs_item_offset(l, i), btrfs_item_size(l, i));
|
2007-03-16 00:56:47 +08:00
|
|
|
switch (type) {
|
|
|
|
|
case BTRFS_INODE_ITEM_KEY:
|
2007-03-21 03:57:25 +08:00
|
|
|
ii = btrfs_item_ptr(l, i, struct btrfs_inode_item);
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tinode generation %llu size %llu mode %o\n",
|
2009-01-06 10:25:51 +08:00
|
|
|
btrfs_inode_generation(l, ii),
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_inode_size(l, ii),
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_inode_mode(l, ii));
|
2007-03-16 00:56:47 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_DIR_ITEM_KEY:
|
2007-03-16 03:18:43 +08:00
|
|
|
di = btrfs_item_ptr(l, i, struct btrfs_dir_item);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_dir_item_key_to_cpu(l, di, &found_key);
|
2022-10-21 00:58:28 +08:00
|
|
|
pr_info("\t\tdir oid %llu flags %u\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
found_key.objectid,
|
2022-10-21 00:58:28 +08:00
|
|
|
btrfs_dir_flags(l, di));
|
2007-03-16 00:56:47 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_ROOT_ITEM_KEY:
|
|
|
|
|
ri = btrfs_item_ptr(l, i, struct btrfs_root_item);
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\troot data bytenr %llu refs %u\n",
|
2009-01-06 10:25:51 +08:00
|
|
|
btrfs_disk_root_bytenr(l, ri),
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_disk_root_refs(l, ri));
|
2007-03-16 00:56:47 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_EXTENT_ITEM_KEY:
|
2014-07-03 01:20:48 +08:00
|
|
|
case BTRFS_METADATA_ITEM_KEY:
|
|
|
|
|
print_extent_item(l, i, type);
|
2007-03-16 00:56:47 +08:00
|
|
|
break;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
case BTRFS_TREE_BLOCK_REF_KEY:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\ttree block backref\n");
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_SHARED_BLOCK_REF_KEY:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tshared block backref\n");
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_EXTENT_DATA_REF_KEY:
|
|
|
|
|
dref = btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_extent_data_ref);
|
|
|
|
|
print_extent_data_ref(l, dref);
|
|
|
|
|
break;
|
|
|
|
|
case BTRFS_SHARED_DATA_REF_KEY:
|
|
|
|
|
sref = btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_shared_data_ref);
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tshared data backref count %u\n",
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
btrfs_shared_data_ref_count(l, sref));
|
2007-12-11 22:25:06 +08:00
|
|
|
break;
|
2007-06-18 21:57:58 +08:00
|
|
|
case BTRFS_EXTENT_DATA_KEY:
|
|
|
|
|
fi = btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_file_extent_item);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_file_extent_type(l, fi) ==
|
2007-06-18 21:57:58 +08:00
|
|
|
BTRFS_FILE_EXTENT_INLINE) {
|
2018-06-06 15:41:49 +08:00
|
|
|
pr_info("\t\tinline extent data size %llu\n",
|
|
|
|
|
btrfs_file_extent_ram_bytes(l, fi));
|
2007-06-18 21:57:58 +08:00
|
|
|
break;
|
|
|
|
|
}
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\textent data disk bytenr %llu nr %llu\n",
|
2009-01-06 10:25:51 +08:00
|
|
|
btrfs_file_extent_disk_bytenr(l, fi),
|
|
|
|
|
btrfs_file_extent_disk_num_bytes(l, fi));
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\textent data offset %llu nr %llu ram %llu\n",
|
2009-01-06 10:25:51 +08:00
|
|
|
btrfs_file_extent_offset(l, fi),
|
|
|
|
|
btrfs_file_extent_num_bytes(l, fi),
|
|
|
|
|
btrfs_file_extent_ram_bytes(l, fi));
|
2007-06-18 21:57:58 +08:00
|
|
|
break;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 22:45:14 +08:00
|
|
|
case BTRFS_EXTENT_REF_V0_KEY:
|
2018-06-26 21:57:36 +08:00
|
|
|
btrfs_print_v0_err(fs_info);
|
|
|
|
|
btrfs_handle_fs_error(fs_info, -EINVAL, NULL);
|
2011-01-26 14:22:33 +08:00
|
|
|
break;
|
2007-04-27 04:46:15 +08:00
|
|
|
case BTRFS_BLOCK_GROUP_ITEM_KEY:
|
|
|
|
|
bi = btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_block_group_item);
|
2017-05-26 08:08:12 +08:00
|
|
|
pr_info(
|
|
|
|
|
"\t\tblock group used %llu chunk_objectid %llu flags %llu\n",
|
2019-10-24 00:48:20 +08:00
|
|
|
btrfs_block_group_used(l, bi),
|
|
|
|
|
btrfs_block_group_chunk_objectid(l, bi),
|
|
|
|
|
btrfs_block_group_flags(l, bi));
|
2007-03-16 00:56:47 +08:00
|
|
|
break;
|
2008-03-25 03:01:56 +08:00
|
|
|
case BTRFS_CHUNK_ITEM_KEY:
|
2009-01-06 10:25:51 +08:00
|
|
|
print_chunk(l, btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_chunk));
|
2008-03-25 03:01:56 +08:00
|
|
|
break;
|
|
|
|
|
case BTRFS_DEV_ITEM_KEY:
|
|
|
|
|
print_dev_item(l, btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_dev_item));
|
|
|
|
|
break;
|
|
|
|
|
case BTRFS_DEV_EXTENT_KEY:
|
|
|
|
|
dev_extent = btrfs_item_ptr(l, i,
|
|
|
|
|
struct btrfs_dev_extent);
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tdev extent chunk_tree %llu\n\t\tchunk objectid %llu chunk offset %llu length %llu\n",
|
2008-04-16 03:41:47 +08:00
|
|
|
btrfs_dev_extent_chunk_tree(l, dev_extent),
|
|
|
|
|
btrfs_dev_extent_chunk_objectid(l, dev_extent),
|
|
|
|
|
btrfs_dev_extent_chunk_offset(l, dev_extent),
|
|
|
|
|
btrfs_dev_extent_length(l, dev_extent));
|
2013-01-31 08:54:57 +08:00
|
|
|
break;
|
2016-01-26 01:44:13 +08:00
|
|
|
case BTRFS_PERSISTENT_ITEM_KEY:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tpersistent item objectid %llu offset %llu\n",
|
2016-01-26 01:44:13 +08:00
|
|
|
key.objectid, key.offset);
|
|
|
|
|
switch (key.objectid) {
|
|
|
|
|
case BTRFS_DEV_STATS_OBJECTID:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tdevice stats\n");
|
2016-01-26 01:44:13 +08:00
|
|
|
break;
|
|
|
|
|
default:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tunknown persistent item\n");
|
2016-01-26 01:44:13 +08:00
|
|
|
}
|
2012-05-25 22:06:10 +08:00
|
|
|
break;
|
2016-01-26 01:44:13 +08:00
|
|
|
case BTRFS_TEMPORARY_ITEM_KEY:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\ttemporary item objectid %llu offset %llu\n",
|
2016-01-26 01:44:13 +08:00
|
|
|
key.objectid, key.offset);
|
|
|
|
|
switch (key.objectid) {
|
|
|
|
|
case BTRFS_BALANCE_OBJECTID:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tbalance status\n");
|
2016-01-26 01:44:13 +08:00
|
|
|
break;
|
|
|
|
|
default:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tunknown temporary item\n");
|
2016-01-26 01:44:13 +08:00
|
|
|
}
|
|
|
|
|
break;
|
2012-11-06 00:32:20 +08:00
|
|
|
case BTRFS_DEV_REPLACE_KEY:
|
2016-09-20 22:05:01 +08:00
|
|
|
pr_info("\t\tdev replace\n");
|
2012-11-06 00:32:20 +08:00
|
|
|
break;
|
2013-08-15 23:11:18 +08:00
|
|
|
case BTRFS_UUID_KEY_SUBVOL:
|
|
|
|
|
case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
|
|
|
|
|
print_uuid_item(l, btrfs_item_ptr_offset(l, i),
|
2021-10-22 02:58:35 +08:00
|
|
|
btrfs_item_size(l, i));
|
2013-08-15 23:11:18 +08:00
|
|
|
break;
|
2019-12-19 17:25:34 +08:00
|
|
|
}
|
2007-02-24 19:24:44 +08:00
|
|
|
}
|
|
|
|
|
}
|
2007-03-23 00:13:20 +08:00
|
|
|
|
2018-04-11 17:08:12 +08:00
|
|
|
void btrfs_print_tree(struct extent_buffer *c, bool follow)
|
2007-02-24 19:24:44 +08:00
|
|
|
{
|
2017-06-30 00:37:49 +08:00
|
|
|
struct btrfs_fs_info *fs_info;
|
2008-04-16 03:41:47 +08:00
|
|
|
int i; u32 nr;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct btrfs_key key;
|
2007-10-16 04:15:53 +08:00
|
|
|
int level;
|
2007-02-24 19:24:44 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
if (!c)
|
2007-02-24 19:24:44 +08:00
|
|
|
return;
|
2017-06-30 00:37:49 +08:00
|
|
|
fs_info = c->fs_info;
|
2007-10-16 04:14:19 +08:00
|
|
|
nr = btrfs_header_nritems(c);
|
2007-10-16 04:15:53 +08:00
|
|
|
level = btrfs_header_level(c);
|
|
|
|
|
if (level == 0) {
|
2017-06-30 00:37:49 +08:00
|
|
|
btrfs_print_leaf(c);
|
2007-02-24 19:24:44 +08:00
|
|
|
return;
|
|
|
|
|
}
|
2016-06-23 06:54:23 +08:00
|
|
|
btrfs_info(fs_info,
|
2018-04-11 17:08:12 +08:00
|
|
|
"node %llu level %d gen %llu total ptrs %d free spc %u owner %llu",
|
|
|
|
|
btrfs_header_bytenr(c), level, btrfs_header_generation(c),
|
|
|
|
|
nr, (u32)BTRFS_NODEPTRS_PER_BLOCK(fs_info) - nr,
|
|
|
|
|
btrfs_header_owner(c));
|
2018-04-24 13:03:13 +08:00
|
|
|
print_eb_refs_lock(c);
|
2007-02-24 19:24:44 +08:00
|
|
|
for (i = 0; i < nr; i++) {
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_node_key_to_cpu(c, &key, i);
|
2018-04-11 17:08:12 +08:00
|
|
|
pr_info("\tkey %d (%llu %u %llu) block %llu gen %llu\n",
|
2013-08-20 19:20:07 +08:00
|
|
|
i, key.objectid, key.type, key.offset,
|
2018-04-11 17:08:12 +08:00
|
|
|
btrfs_node_blockptr(c, i),
|
|
|
|
|
btrfs_node_ptr_generation(c, i));
|
2007-02-24 19:24:44 +08:00
|
|
|
}
|
2018-04-11 17:08:12 +08:00
|
|
|
if (!follow)
|
|
|
|
|
return;
|
2007-02-24 19:24:44 +08:00
|
|
|
for (i = 0; i < nr; i++) {
|
2022-09-14 13:32:50 +08:00
|
|
|
struct btrfs_tree_parent_check check = {
|
|
|
|
|
.level = level - 1,
|
|
|
|
|
.transid = btrfs_node_ptr_generation(c, i),
|
|
|
|
|
.owner_root = btrfs_header_owner(c),
|
|
|
|
|
.has_first_key = true
|
|
|
|
|
};
|
2018-03-29 09:08:11 +08:00
|
|
|
struct extent_buffer *next;
|
|
|
|
|
|
2022-09-14 13:32:50 +08:00
|
|
|
btrfs_node_key_to_cpu(c, &check.first_key, i);
|
|
|
|
|
next = read_tree_block(fs_info, btrfs_node_blockptr(c, i), &check);
|
2022-02-22 15:41:19 +08:00
|
|
|
if (IS_ERR(next))
|
2016-09-15 08:23:39 +08:00
|
|
|
continue;
|
2022-02-22 15:41:19 +08:00
|
|
|
if (!extent_buffer_uptodate(next)) {
|
2016-09-15 08:23:39 +08:00
|
|
|
free_extent_buffer(next);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
2007-03-13 00:01:18 +08:00
|
|
|
if (btrfs_is_leaf(next) &&
|
2009-07-23 04:52:13 +08:00
|
|
|
level != 1)
|
2007-02-24 19:24:44 +08:00
|
|
|
BUG();
|
2007-10-16 04:14:19 +08:00
|
|
|
if (btrfs_header_level(next) !=
|
2009-07-23 04:52:13 +08:00
|
|
|
level - 1)
|
2007-02-24 19:24:44 +08:00
|
|
|
BUG();
|
2018-04-11 17:08:12 +08:00
|
|
|
btrfs_print_tree(next, follow);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(next);
|
2007-02-24 19:24:44 +08:00
|
|
|
}
|
|
|
|
|
}
|