We've continued mainly to fix bugs in this round, as f2fs has been shipped
in more devices. Especially, we've focused on stabilizing checkpoint=disable
feature, and provided some interfaces for QA.
Enhancement:
- expose FS_NOCOW_FL for pin_file
- run discard jobs at unmount time with timeout
- tune discarding thread to avoid idling which consumes power
- some checking codes to address vulnerabilities
- give random value to i_generation
- shutdown with more flags for QA
Bug fix:
- clean up stale objects when mount is failed along with checkpoint=disable
- fix system being stuck due to wrong count by atomic writes
- handle some corrupted disk cases
- fix a deadlock in f2fs_read_inline_dir
We've also added some minor build errors and clean-up patches.
-----BEGIN PGP SIGNATURE-----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=9BoH
-----END PGP SIGNATURE-----
Merge tag 'f2fs-for-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs
Pull f2fs updates from Jaegeuk Kim:
"We've continued mainly to fix bugs in this round, as f2fs has been
shipped in more devices. Especially, we've focused on stabilizing
checkpoint=disable feature, and provided some interfaces for QA.
Enhancements:
- expose FS_NOCOW_FL for pin_file
- run discard jobs at unmount time with timeout
- tune discarding thread to avoid idling which consumes power
- some checking codes to address vulnerabilities
- give random value to i_generation
- shutdown with more flags for QA
Bug fixes:
- clean up stale objects when mount is failed along with
checkpoint=disable
- fix system being stuck due to wrong count by atomic writes
- handle some corrupted disk cases
- fix a deadlock in f2fs_read_inline_dir
We've also added some minor build error fixes and clean-up patches"
* tag 'f2fs-for-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs: (53 commits)
f2fs: set pin_file under CAP_SYS_ADMIN
f2fs: fix to avoid deadlock in f2fs_read_inline_dir()
f2fs: fix to adapt small inline xattr space in __find_inline_xattr()
f2fs: fix to do sanity check with inode.i_inline_xattr_size
f2fs: give some messages for inline_xattr_size
f2fs: don't trigger read IO for beyond EOF page
f2fs: fix to add refcount once page is tagged PG_private
f2fs: remove wrong comment in f2fs_invalidate_page()
f2fs: fix to use kvfree instead of kzfree
f2fs: print more parameters in trace_f2fs_map_blocks
f2fs: trace f2fs_ioc_shutdown
f2fs: fix to avoid deadlock of atomic file operations
f2fs: fix to dirty inode for i_mode recovery
f2fs: give random value to i_generation
f2fs: no need to take page lock in readdir
f2fs: fix to update iostat correctly in IPU path
f2fs: fix encrypted page memory leak
f2fs: make fault injection covering __submit_flush_wait()
f2fs: fix to retry fill_super only if recovery failed
f2fs: silence VM_WARN_ON_ONCE in mempool_alloc
...
Android uses pin_file for uncrypt during OTA, and that should be managed by
CAP_SYS_ADMIN only.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Seulbae Kim reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=202637
We didn't recover permission field correctly after sudden power-cut,
the reason is in setattr we didn't add inode into global dirty list
once i_mode is changed, so latter checkpoint triggered by fsync will
not flush last i_mode into disk, result in this problem, fix it.
Reported-by: Seulbae Kim <seulbae@gatech.edu>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, we changed lock from cp_rwsem to node_change, it solved
the deadlock issue which was caused by below race condition:
Thread A Thread B
- f2fs_setattr
- f2fs_lock_op -- read_lock
- dquot_transfer
- __dquot_transfer
- dquot_acquire
- commit_dqblk
- f2fs_quota_write
- f2fs_write_begin
- f2fs_write_failed
- write_checkpoint
- block_operations
- f2fs_lock_all -- write_lock
- f2fs_truncate_blocks
- f2fs_lock_op -- read_lock
But it breaks the sematics of cp_rwsem, in other callers like:
- f2fs_file_write_iter -> f2fs_write_begin -> f2fs_write_failed
- f2fs_direct_IO -> f2fs_write_failed
We allow to truncate dnode w/o cp_rwsem held, result in incorrect sit
bitmap update, which can cause further data corruption.
So this patch reverts previous fix implementation, and try to fix
deadlock by skipping calling f2fs_truncate_blocks() in f2fs_write_failed()
only for quota file, and keep the preallocated data/node in the tail of
quota file, we can expecte that the preallocated space can be used to
store quota info latter soon.
Fixes: af033b2aa8 ("f2fs: guarantee journalled quota data by checkpoint")
Signed-off-by: Gao Xiang <gaoxiang25@huawei.com>
Signed-off-by: Sheng Yong <shengyong1@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This mode returns mount() quickly with EAGAIN. We can trigger this by
shutdown(F2FS_GOING_DOWN_NEED_FSCK).
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This commit removes the f2fs specific f2fs_encrypted_inode() and makes
use of the generic IS_ENCRYPTED() macro to check for the encryption
status of an inode.
Acked-by: Chao Yu <yuchao0@huawei.com>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Otherwise, we can get wrong counts incurring checkpoint hang.
IO_W (CP: -24, Data: 24, Flush: ( 0 0 1), Discard: ( 0 0))
Thread A Thread B
- f2fs_write_data_pages
- __write_data_page
- f2fs_submit_page_write
- inc_page_count(F2FS_WB_DATA)
type is F2FS_WB_DATA due to file is non-atomic one
- f2fs_ioc_start_atomic_write
- set_inode_flag(FI_ATOMIC_FILE)
- f2fs_write_end_io
- dec_page_count(F2FS_WB_CP_DATA)
type is F2FS_WB_DATA due to file becomes
atomic one
Cc: <stable@vger.kernel.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For all ordered cases in f2fs_wait_on_page_writeback(), we need to
check PageWriteback status, so let's clean up to relocate the check
into f2fs_wait_on_page_writeback().
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If user change inode's i_flags via ioctl, let's add it into global
dirty list, so that checkpoint can guarantee its persistence before
fsync, it can make checkpoint keeping strong consistency.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This adds an option in ioctl(F2FS_IOC_SHUTDOWN) in order to trigger fsck by
setting a NEED_FSCK flag.
Generally, shutdown is used for the test to validate filesystem consistency, and
setting NEED_FSCK flag can be used for Android to trigger fsck.f2fs at boot time
explicitly so that we could measure the elapsed time as well as force filesystem
check.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, we added a parameter @map.m_may_create to trigger OPU
allocation and call f2fs_balance_fs() correctly.
But in get_more_blocks(), @create has been overwritten by below code.
So the function f2fs_map_blocks() will not allocate new block address
but directly go out. Meanwile,there are several functions calling
f2fs_map_blocks() directly and @map.m_may_create not initialized.
CODE:
create = dio->op == REQ_OP_WRITE;
if (dio->flags & DIO_SKIP_HOLES) {
if (fs_startblk <= ((i_size_read(dio->inode) - 1) >>
i_blkbits))
create = 0;
}
This patch fixes it.
Signed-off-by: Jia Zhu <zhujia13@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In get_more_blocks(), we may override @create as below code:
create = dio->op == REQ_OP_WRITE;
if (dio->flags & DIO_SKIP_HOLES) {
if (fs_startblk <= ((i_size_read(dio->inode) - 1) >>
i_blkbits))
create = 0;
}
But in f2fs_map_blocks(), we only trigger f2fs_balance_fs() if @create
is 1, so in LFS mode, dio overwrite under LFS mode can easily run out
of free segments, result in below panic.
Call Trace:
allocate_segment_by_default+0xa8/0x270 [f2fs]
f2fs_allocate_data_block+0x1ea/0x5c0 [f2fs]
__allocate_data_block+0x306/0x480 [f2fs]
f2fs_map_blocks+0x6f6/0x920 [f2fs]
__get_data_block+0x4f/0xb0 [f2fs]
get_data_block_dio_write+0x50/0x60 [f2fs]
do_blockdev_direct_IO+0xcd5/0x21e0
__blockdev_direct_IO+0x3a/0x3c
f2fs_direct_IO+0x1ff/0x4a0 [f2fs]
generic_file_direct_write+0xd9/0x160
__generic_file_write_iter+0xbb/0x1e0
f2fs_file_write_iter+0xaf/0x220 [f2fs]
__vfs_write+0xd0/0x130
vfs_write+0xb2/0x1b0
SyS_pwrite64+0x69/0xa0
? vtime_user_exit+0x29/0x70
do_syscall_64+0x6e/0x160
entry_SYSCALL64_slow_path+0x25/0x25
RIP: new_curseg+0x36f/0x380 [f2fs] RSP: ffffac570393f7a8
So this patch introduces a parameter map.m_may_create to indicate that
f2fs_map_blocks() is called from write or read path, which can give the
right hint to let f2fs_map_blocks() trigger OPU allocation and call
f2fs_balanc_fs() correctly.
BTW, it disables physical address preallocation for direct IO in
f2fs_preallocate_blocks, which is redundant to OPU allocation of
f2fs_map_blocks.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch move dir data flush to write checkpoint process, by
doing this, it may reduce some time for dir fsync.
pre:
-f2fs_do_sync_file enter
-file_write_and_wait_range <- flush & wait
-write_checkpoint
-do_checkpoint <- wait all
-f2fs_do_sync_file exit
now:
-f2fs_do_sync_file enter
-write_checkpoint
-block_operations <- flush dir & no wait
-do_checkpoint <- wait all
-f2fs_do_sync_file exit
Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
f2fs_ioc_gc_range skips blocks_per_seg each time, however, f2fs_gc moves
blocks of section each time, so fix it from segment to section.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Introduce a wrapper __is_large_section() to clean up codes.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In F2FS_HAS_FEATURE(), we will use F2FS_SB(sb) to get sbi pointer to
access .raw_super field, to avoid unneeded pointer conversion, this
patch changes to F2FS_HAS_FEATURE() accept sbi parameter directly.
Just do cleanup, no logic change.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch does below changes to keep consistence of project quota data
in sudden power-cut case:
- update inode.i_projid and project quota atomically under lock_op() in
f2fs_ioc_setproject()
- recover inode.i_projid and project quota in recover_inode()
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For journalled quota mode, let checkpoint to flush dquot dirty data
and quota file data to guarntee persistence of all quota sysfile in
last checkpoint, by this way, we can avoid corrupting quota sysfile
when encountering SPO.
The implementation is as below:
1. add a global state SBI_QUOTA_NEED_FLUSH to indicate that there is
cached dquot metadata changes in quota subsystem, and later checkpoint
should:
a) flush dquot metadata into quota file.
b) flush quota file to storage to keep file usage be consistent.
2. add a global state SBI_QUOTA_NEED_REPAIR to indicate that quota
operation failed due to -EIO or -ENOSPC, so later,
a) checkpoint will skip syncing dquot metadata.
b) CP_QUOTA_NEED_FSCK_FLAG will be set in last cp pack to give a
hint for fsck repairing.
3. add a global state SBI_QUOTA_SKIP_FLUSH, in checkpoint, if quota
data updating is very heavy, it may cause hungtask in block_operation().
To avoid this, if our retry time exceed threshold, let's just skip
flushing and retry in next checkpoint().
Signed-off-by: Weichao Guo <guoweichao@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: avoid warnings and set fsck flag]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The REQ_TIME should be updated only in case of success cases
as followed at all other places in the file system.
Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Note that, it requires "f2fs: return correct errno in f2fs_gc".
This adds a lightweight non-persistent snapshotting scheme to f2fs.
To use, mount with the option checkpoint=disable, and to return to
normal operation, remount with checkpoint=enable. If the filesystem
is shut down before remounting with checkpoint=enable, it will revert
back to its apparent state when it was first mounted with
checkpoint=disable. This is useful for situations where you wish to be
able to roll back the state of the disk in case of some critical
failure.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
[Jaegeuk Kim: use SB_RDONLY instead of MS_RDONLY]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Normally, DIO uses in-pllace-update, but in LFS mode, f2fs doesn't
allow triggering any in-place-update writes, so we fallback direct
write to buffered write, result in bad performance of large size
write.
This patch adds to support triggering out-place-update for direct IO
to enhance its performance.
Note that it needs to exclude direct read IO during direct write,
since new data writing to new block address will no be valid until
write finished.
storage: zram
time xfs_io -f -d /mnt/f2fs/file -c "pwrite 0 1073741824" -c "fsync"
Before:
real 0m13.061s
user 0m0.327s
sys 0m12.486s
After:
real 0m6.448s
user 0m0.228s
sys 0m6.212s
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Thread A Thread B
- f2fs_vm_page_mkwrite
- f2fs_setattr
- down_write(i_mmap_sem)
- truncate_setsize
- f2fs_truncate
- up_write(i_mmap_sem)
- f2fs_reserve_block
reserve NEW_ADDR
- skip dirty page due to truncation
1. we don't need to rserve new block address for a truncated page.
2. dn.data_blkaddr is used out of node page lock coverage.
Refactor ->page_mkwrite() flow to fix above issues:
- use __do_map_lock() to avoid racing checkpoint()
- lock data page in prior to dnode page
- cover f2fs_reserve_block with i_mmap_sem lock
- wait page writeback before zeroing page
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Remove the verbose license text from f2fs files and replace them with
SPDX tags. This does not change the license of any of the code.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Currently, project quota could be changed by fssetxattr
ioctl, and existed permission check inode_owner_or_capable()
is obviously not enough, just think that common users could
change project id of file, that could make users to
break project quota easily.
This patch try to follow same regular of xfs project
quota:
"Project Quota ID state is only allowed to change from
within the init namespace. Enforce that restriction only
if we are trying to change the quota ID state.
Everything else is allowed in user namespaces."
Besides that, check and set project id'state should
be an atomic operation, protect whole operation with
inode lock.
Signed-off-by: Wang Shilong <wshilong@ddn.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
1. Create a file in an encrypted directory
2. Do GC & drop caches
3. Read stale data before its bio for metapage was not issued yet
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
https://bugzilla.kernel.org/show_bug.cgi?id=200951
These is a NULL pointer dereference issue reported in bugzilla:
Hi,
in the setup there is a SATA SSD connected to a SATA-to-USB bridge.
The disc is "Samsung SSD 850 PRO 256G" which supports TRIM.
There are four partitions:
sda1: FAT /boot
sda2: F2FS /
sda3: F2FS /home
sda4: F2FS
The bridge is ASMT1153e which uses the "uas" driver.
There is no TRIM pass-through, so, when mounting it reports:
mounting with "discard" option, but the device does not support discard
The USB host is USB3.0 and UASP capable. It is the one on RK3399.
Given this everything works fine, except there is no TRIM support.
In order to enable TRIM a new UDEV rule is added [1]:
/etc/udev/rules.d/10-sata-bridge-trim.rules:
ACTION=="add|change", ATTRS{idVendor}=="174c", ATTRS{idProduct}=="55aa", SUBSYSTEM=="scsi_disk", ATTR{provisioning_mode}="unmap"
After reboot any F2FS write hangs forever and dmesg reports:
Unable to handle kernel NULL pointer dereference
Also tested on a x86_64 system: works fine even with TRIM enabled.
same disc
same bridge
different usb host controller
different cpu architecture
not root filesystem
Regards,
Vicenç.
[1] Post #5 in https://bbs.archlinux.org/viewtopic.php?id=236280
Unable to handle kernel NULL pointer dereference at virtual address 000000000000003e
Mem abort info:
ESR = 0x96000004
Exception class = DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000004
CM = 0, WnR = 0
user pgtable: 4k pages, 48-bit VAs, pgdp = 00000000626e3122
[000000000000003e] pgd=0000000000000000
Internal error: Oops: 96000004 [#1] SMP
Modules linked in: overlay snd_soc_hdmi_codec rc_cec dw_hdmi_i2s_audio dw_hdmi_cec snd_soc_simple_card snd_soc_simple_card_utils snd_soc_rockchip_i2s rockchip_rga snd_soc_rockchip_pcm rockchipdrm videobuf2_dma_sg v4l2_mem2mem rtc_rk808 videobuf2_memops analogix_dp videobuf2_v4l2 videobuf2_common dw_hdmi dw_wdt cec rc_core videodev drm_kms_helper media drm rockchip_thermal rockchip_saradc realtek drm_panel_orientation_quirks syscopyarea sysfillrect sysimgblt fb_sys_fops dwmac_rk stmmac_platform stmmac pwm_bl squashfs loop crypto_user gpio_keys hid_kensington
CPU: 5 PID: 957 Comm: nvim Not tainted 4.19.0-rc1-1-ARCH #1
Hardware name: Sapphire-RK3399 Board (DT)
pstate: 00000005 (nzcv daif -PAN -UAO)
pc : update_sit_entry+0x304/0x4b0
lr : update_sit_entry+0x108/0x4b0
sp : ffff00000ca13bd0
x29: ffff00000ca13bd0 x28: 000000000000003e
x27: 0000000000000020 x26: 0000000000080000
x25: 0000000000000048 x24: ffff8000ebb85cf8
x23: 0000000000000253 x22: 00000000ffffffff
x21: 00000000000535f2 x20: 00000000ffffffdf
x19: ffff8000eb9e6800 x18: ffff8000eb9e6be8
x17: 0000000007ce6926 x16: 000000001c83ffa8
x15: 0000000000000000 x14: ffff8000f602df90
x13: 0000000000000006 x12: 0000000000000040
x11: 0000000000000228 x10: 0000000000000000
x9 : 0000000000000000 x8 : 0000000000000000
x7 : 00000000000535f2 x6 : ffff8000ebff3440
x5 : ffff8000ebff3440 x4 : ffff8000ebe3a6c8
x3 : 00000000ffffffff x2 : 0000000000000020
x1 : 0000000000000000 x0 : ffff8000eb9e5800
Process nvim (pid: 957, stack limit = 0x0000000063a78320)
Call trace:
update_sit_entry+0x304/0x4b0
f2fs_invalidate_blocks+0x98/0x140
truncate_node+0x90/0x400
f2fs_remove_inode_page+0xe8/0x340
f2fs_evict_inode+0x2b0/0x408
evict+0xe0/0x1e0
iput+0x160/0x260
do_unlinkat+0x214/0x298
__arm64_sys_unlinkat+0x3c/0x68
el0_svc_handler+0x94/0x118
el0_svc+0x8/0xc
Code: f9400800 b9488400 36080140 f9400f01 (387c4820)
---[ end trace a0f21a307118c477 ]---
The reason is it is possible to enable discard flag on block queue via
UDEV, but during mount, f2fs will initialize se->discard_map only if
this flag is set, once the flag is set after mount, f2fs may dereference
NULL pointer on se->discard_map.
So this patch does below changes to fix this issue:
- initialize and update se->discard_map all the time.
- don't clear DISCARD option if device has no QUEUE_FLAG_DISCARD flag
during mount.
- don't issue small discard on zoned block device.
- introduce some functions to enhance the readability.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Tested-by: Vicente Bergas <vicencb@gmail.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The f2fs_gc() called by f2fs_balance_fs() requires to be called outside of
fi->i_gc_rwsem[WRITE], since f2fs_gc() can try to grab it in a loop.
If it hits the miximum retrials in GC, let's give a chance to release
gc_mutex for a short time in order not to go into live lock in the worst
case.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When CONFIG_F2FS_FAULT_INJECTION is disabled, we get a warning about an
unused label:
fs/f2fs/segment.c: In function '__submit_discard_cmd':
fs/f2fs/segment.c:1059:1: error: label 'submit' defined but not used [-Werror=unused-label]
This could be fixed by adding another #ifdef around it, but the more
reliable way of doing this seems to be to remove the other #ifdefs
where that is easily possible.
By defining time_to_inject() as a trivial stub, most of the checks for
CONFIG_F2FS_FAULT_INJECTION can go away. This also leads to nicer
formatting of the code.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Thread A Background GC
- f2fs_setattr isize to 0
- truncate_setsize
- gc_data_segment
- f2fs_get_read_data_page page #0
- set_page_dirty
- set_cold_data
- f2fs_truncate
- f2fs_setattr isize to 4k
- read 4k <--- hit data in cached page #0
Above race condition can cause read out invalid data in a truncated
page, fix it by i_gc_rwsem[WRITE] lock.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Thread A Background GC
- f2fs_zero_range
- truncate_pagecache_range
- gc_data_segment
- get_read_data_page
- move_data_page
- set_page_dirty
- set_cold_data
- f2fs_do_zero_range
- dn->data_blkaddr = NEW_ADDR;
- f2fs_set_data_blkaddr
Actually, we don't need to set dirty & checked flag on the page, since
all valid data in the page should be zeroed by zero_range().
Use i_gc_rwsem[WRITE] to avoid such race condition.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Let's reset i_gc_failures to zero when we unset pinned state for file.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
f2fs recovery flow is relying on dnode block link list, it means fsynced
file recovery depends on previous dnode's persistence in the list, so
during fsync() we should wait on all regular inode's dnode writebacked
before issuing flush.
By this way, we can avoid dnode block list being broken by out-of-order
IO submission due to IO scheduler or driver.
Sheng Yong helps to do the test with this patch:
Target:/data (f2fs, -)
64MB / 32768KB / 4KB / 8
1 / PERSIST / Index
Base:
SEQ-RD(MB/s) SEQ-WR(MB/s) RND-RD(IOPS) RND-WR(IOPS) Insert(TPS) Update(TPS) Delete(TPS)
1 867.82 204.15 41440.03 41370.54 680.8 1025.94 1031.08
2 871.87 205.87 41370.3 40275.2 791.14 1065.84 1101.7
3 866.52 205.69 41795.67 40596.16 694.69 1037.16 1031.48
Avg 868.7366667 205.2366667 41535.33333 40747.3 722.21 1042.98 1054.753333
After:
SEQ-RD(MB/s) SEQ-WR(MB/s) RND-RD(IOPS) RND-WR(IOPS) Insert(TPS) Update(TPS) Delete(TPS)
1 798.81 202.5 41143 40613.87 602.71 838.08 913.83
2 805.79 206.47 40297.2 41291.46 604.44 840.75 924.27
3 814.83 206.17 41209.57 40453.62 602.85 834.66 927.91
Avg 806.4766667 205.0466667 40883.25667 40786.31667 603.3333333 837.83 922.0033333
Patched/Original:
0.928332713 0.999074239 0.984300676 1.000957528 0.835398753 0.803303994 0.874141189
It looks like atomic write will suffer performance regression.
I suspect that the criminal is that we forcing to wait all dnode being in
storage cache before we issue PREFLUSH+FUA.
BTW, will commit ("f2fs: don't need to wait for node writes for atomic write")
cause the problem: we will lose data of last transaction after SPO, even if
atomic write return no error:
- atomic_open();
- write() P1, P2, P3;
- atomic_commit();
- writeback data: P1, P2, P3;
- writeback node: N1, N2, N3; <--- If N1, N2 is not writebacked, N3 with fsync_mark is
writebacked, In SPOR, we won't find N3 since node chain is broken, turns out that losing
last transaction.
- preflush + fua;
- power-cut
If we don't wait dnode writeback for atomic_write:
SEQ-RD(MB/s) SEQ-WR(MB/s) RND-RD(IOPS) RND-WR(IOPS) Insert(TPS) Update(TPS) Delete(TPS)
1 779.91 206.03 41621.5 40333.16 716.9 1038.21 1034.85
2 848.51 204.35 40082.44 39486.17 791.83 1119.96 1083.77
3 772.12 206.27 41335.25 41599.65 723.29 1055.07 971.92
Avg 800.18 205.55 41013.06333 40472.99333 744.0066667 1071.08 1030.18
Patched/Original:
0.92108464 1.001526693 0.987425886 0.993268102 1.030180511 1.026942031 0.976702294
SQLite's performance recovers.
Jaegeuk:
"Practically, I don't see db corruption becase of this. We can excuse to lose
the last transaction."
Finally, we decide to keep original implementation of atomic write interface
sematics that we don't wait all dnode writeback before preflush+fua submission.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In order to prevent abusing atomic writes by abnormal users, we've added a
threshold, 20% over memory footprint, which disallows further atomic writes.
Previously, however, SQLite doesn't know the files became normal, so that
it could write stale data and commit on revoked normal database file.
Once f2fs detects such the abnormal behavior, this patch tries to avoid further
writes in write_begin().
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Romve redundant prefix 'f2fs_' in the middle of f2fs_ioc_f2fs_write_checkpoint().
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If caller of __get_meta_page() can handle error, let's propagate error
from __get_meta_page().
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces verify_blkaddr to check meta/data block address
with valid range to detect bug earlier.
In addition, once we encounter an invalid blkaddr, notice user to run
fsck to fix, and let the kernel panic.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
"ret" can be uninitialized on the success path when "in ==
F2FS_GOING_DOWN_FULLSYNC".
Fixes: 60b2b4ee2b ("f2fs: Fix deadlock in shutdown ioctl")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Once we shutdown f2fs, we have to flush stale pages in order to unmount
the system. In order to make stable, we need to stop fault injection as well.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This is a late set of changes from Deepa Dinamani doing an automated
treewide conversion of the inode and iattr structures from 'timespec'
to 'timespec64', to push the conversion from the VFS layer into the
individual file systems.
There were no conflicts between this and the contents of linux-next
until just before the merge window, when we saw multiple problems:
- A minor conflict with my own y2038 fixes, which I could address
by adding another patch on top here.
- One semantic conflict with late changes to the NFS tree. I addressed
this by merging Deepa's original branch on top of the changes that
now got merged into mainline and making sure the merge commit includes
the necessary changes as produced by coccinelle.
- A trivial conflict against the removal of staging/lustre.
- Multiple conflicts against the VFS changes in the overlayfs tree.
These are still part of linux-next, but apparently this is no longer
intended for 4.18 [1], so I am ignoring that part.
As Deepa writes:
The series aims to switch vfs timestamps to use struct timespec64.
Currently vfs uses struct timespec, which is not y2038 safe.
The series involves the following:
1. Add vfs helper functions for supporting struct timepec64 timestamps.
2. Cast prints of vfs timestamps to avoid warnings after the switch.
3. Simplify code using vfs timestamps so that the actual
replacement becomes easy.
4. Convert vfs timestamps to use struct timespec64 using a script.
This is a flag day patch.
Next steps:
1. Convert APIs that can handle timespec64, instead of converting
timestamps at the boundaries.
2. Update internal data structures to avoid timestamp conversions.
Thomas Gleixner adds:
I think there is no point to drag that out for the next merge window.
The whole thing needs to be done in one go for the core changes which
means that you're going to play that catchup game forever. Let's get
over with it towards the end of the merge window.
[1] https://www.spinics.net/lists/linux-fsdevel/msg128294.html
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=CZX2
-----END PGP SIGNATURE-----
Merge tag 'vfs-timespec64' of git://git.kernel.org/pub/scm/linux/kernel/git/arnd/playground
Pull inode timestamps conversion to timespec64 from Arnd Bergmann:
"This is a late set of changes from Deepa Dinamani doing an automated
treewide conversion of the inode and iattr structures from 'timespec'
to 'timespec64', to push the conversion from the VFS layer into the
individual file systems.
As Deepa writes:
'The series aims to switch vfs timestamps to use struct timespec64.
Currently vfs uses struct timespec, which is not y2038 safe.
The series involves the following:
1. Add vfs helper functions for supporting struct timepec64
timestamps.
2. Cast prints of vfs timestamps to avoid warnings after the switch.
3. Simplify code using vfs timestamps so that the actual replacement
becomes easy.
4. Convert vfs timestamps to use struct timespec64 using a script.
This is a flag day patch.
Next steps:
1. Convert APIs that can handle timespec64, instead of converting
timestamps at the boundaries.
2. Update internal data structures to avoid timestamp conversions'
Thomas Gleixner adds:
'I think there is no point to drag that out for the next merge
window. The whole thing needs to be done in one go for the core
changes which means that you're going to play that catchup game
forever. Let's get over with it towards the end of the merge window'"
* tag 'vfs-timespec64' of git://git.kernel.org/pub/scm/linux/kernel/git/arnd/playground:
pstore: Remove bogus format string definition
vfs: change inode times to use struct timespec64
pstore: Convert internal records to timespec64
udf: Simplify calls to udf_disk_stamp_to_time
fs: nfs: get rid of memcpys for inode times
ceph: make inode time prints to be long long
lustre: Use long long type to print inode time
fs: add timespec64_truncate()
Thread A Thread B
- f2fs_release_file
- clear_inode_flag(FI_VOLATILE_FILE)
- wb_writeback
- writeback_sb_inodes
- __writeback_single_inode
- do_writepages
- f2fs_write_data_pages
- __write_data_page
all volatile file's pages
are writebacked to storage
- set_inode_flag(FI_DROP_CACHE)
- filemap_fdatawrite
There is a hole that mm can flush all dirty pages of volatile file as
inode is not tagged with both FI_VOLATILE_FILE and FI_DROP_CACHE flags,
we should never writeback the page #0 and also it's unneeded to writeback
other pages.
This patch adjusts to relocate clear_inode_flag(FI_VOLATILE_FILE), so that
FI_VOLATILE_FILE flag can be remained before all dirty pages were dropped
to avoid issue.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Although mixed sync/async IOs can have continuous LBA, as they have
different IO priority, block IO scheduler will add them into different
queues and commit them separately, result in splited IOs which causes
wrose performance.
This patch gives high priority to synchronous IO of nodes, means that
once synchronous flow starts, it can interrupt asynchronous writeback
flow of system flusher, so more big IOs can be expected.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
clean up checkpatch error:
ERROR: space required after that ':'
ERROR: space required after that ','
Signed-off-by: youngjun yoo <youngjun.willow@gmail.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
clean up checkpatch warning:
WARNING: Missing a blank line after declarations
Signed-off-by: youngjun yoo <youngjun.willow@gmail.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
clean up checkpatch warning:
WARNING: Prefer 'unsigned int' to bare use of 'unsigned'
Signed-off-by: youngjun yoo <youngjun.willow@gmail.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Linus Torvalds