Section is minimal garbage collection unit of f2fs, in zoned block
device, or ancient block mapping flash device, in order to improve
GC efficiency, we can align GC unit to lower device erase unit,
normally, it consists of multiple of segments.
Once background or foreground GC triggers, it brings a large number
of IOs, which will impact user IO, and also occupy cpu/memory resource
intensively.
So, to reduce impact of GC on large size section, this patch supports
subsectional GC, in one cycle of GC, it only migrate partial segment{s}
in victim section. Currently, by default, we use sbi->segs_per_sec as
migration granularity.
Signed-off-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>
Pull XArray conversion from Matthew Wilcox:
"The XArray provides an improved interface to the radix tree data
structure, providing locking as part of the API, specifying GFP flags
at allocation time, eliminating preloading, less re-walking the tree,
more efficient iterations and not exposing RCU-protected pointers to
its users.
This patch set
1. Introduces the XArray implementation
2. Converts the pagecache to use it
3. Converts memremap to use it
The page cache is the most complex and important user of the radix
tree, so converting it was most important. Converting the memremap
code removes the only other user of the multiorder code, which allows
us to remove the radix tree code that supported it.
I have 40+ followup patches to convert many other users of the radix
tree over to the XArray, but I'd like to get this part in first. The
other conversions haven't been in linux-next and aren't suitable for
applying yet, but you can see them in the xarray-conv branch if you're
interested"
* 'xarray' of git://git.infradead.org/users/willy/linux-dax: (90 commits)
radix tree: Remove multiorder support
radix tree test: Convert multiorder tests to XArray
radix tree tests: Convert item_delete_rcu to XArray
radix tree tests: Convert item_kill_tree to XArray
radix tree tests: Move item_insert_order
radix tree test suite: Remove multiorder benchmarking
radix tree test suite: Remove __item_insert
memremap: Convert to XArray
xarray: Add range store functionality
xarray: Move multiorder_check to in-kernel tests
xarray: Move multiorder_shrink to kernel tests
xarray: Move multiorder account test in-kernel
radix tree test suite: Convert iteration test to XArray
radix tree test suite: Convert tag_tagged_items to XArray
radix tree: Remove radix_tree_clear_tags
radix tree: Remove radix_tree_maybe_preload_order
radix tree: Remove split/join code
radix tree: Remove radix_tree_update_node_t
page cache: Finish XArray conversion
dax: Convert page fault handlers to XArray
...
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>
Direct IO can be used in case of hardware encryption. The following
scenario results into data corruption issue in this path -
Thread A - Thread B-
-> write file#1 in direct IO
-> GC gets kicked in
-> GC submitted bio on meta mapping
for file#1, but pending completion
-> write file#1 again with new data
in direct IO
-> GC bio gets completed now
-> GC writes old data to the new
location and thus file#1 is
corrupted.
Fix this by submitting and waiting for pending io on meta mapping
for direct IO case in f2fs_map_blocks().
Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch changes codes as below:
- use f2fs_set_inode_flags() to update i_flags atomically to avoid
potential race.
- synchronize F2FS_I(inode)->i_flags to inode->i_flags in
f2fs_new_inode().
- use f2fs_set_inode_flags() to simply codes in f2fs_quota_{on,off}.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds issued read IO counts which is under block layer.
Chao modified a bit, since:
Below race can cause reversed reference on F2FS_RD_DATA, there is
the same issue in f2fs_submit_page_bio(), fix them by relocate
__submit_bio() and inc_page_count.
Thread A Thread B
- f2fs_write_begin
- f2fs_submit_page_read
- __submit_bio
- f2fs_read_end_io
- __read_end_io
- dec_page_count(, F2FS_RD_DATA)
- inc_page_count(, F2FS_RD_DATA)
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This doesn't work on stacked devices, and it doesn't work on
blk-mq devices. The request_list is only used on legacy, which
we don't have much of anymore, and soon won't have any of.
Kill the check.
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Cc: linux-f2fs-devel@lists.sourceforge.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Commit a2a12b679f ("f2fs: export SSR allocation threshold") introduced
two threshold .min_ssr_sections and .trigger_ssr_threshold, but only
.min_ssr_sections is used, so just remove redundant one for cleanup.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As rbtree supports caching leftmost node natively, update f2fs codes
to use rb_*_cached helpers to speed up leftmost node visiting.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When migrating encrypted block from background GC thread, we only add
them into f2fs inner bio cache, but forget to submit the cached bio, it
may cause potential deadlock when we are waiting page writebacked, fix
it.
Signed-off-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>
There is one case that we can leave bio in f2fs, result in hanging
page writeback waiter.
Thread A Thread B
- f2fs_write_cache_pages
- f2fs_submit_page_write
page #0 cached in bio #0 of cold log
- f2fs_submit_page_write
page #1 cached in bio #1 of warm log
- f2fs_write_cache_pages
- f2fs_submit_page_write
bio is full, submit bio #1 contain page #1
- f2fs_submit_merged_write_cond(, page #1)
fail to submit bio #0 due to page #1 is not in any cached bios.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Now we support crc32 checksum for superblock.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Junling Zheng <zhengjunling@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to account skip count of background GC, and show stat
info via 'status' debugfs entry.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch supports to account meta IO, it enables to show write IO
from f2fs more comprehensively via 'status' debugfs entry.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch avoids BUG_ON when f2fs_get_meta_page_nofail got EIO during
xfstests/generic/475.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This is related to
ee70daaba8 ("xfs: update i_size after unwritten conversion in dio completion")
If we update i_size during dio_write, dio_read can read out stale data, which
breaks xfstests/465.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This helps to control the frequency of submission of discard and
GC requests independently, based on the need.
Suggested-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to support injecting error for write IO, this can simulate
IO error like fail_make_request or dm_flakey does.
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>
In the call trace below, we might sleep in function dput().
So in order to avoid sleeping under spin_lock, we remove f2fs_mark_inode_dirty_sync
from __try_update_largest_extent && __drop_largest_extent.
BUG: sleeping function called from invalid context at fs/dcache.c:796
Call trace:
dump_backtrace+0x0/0x3f4
show_stack+0x24/0x30
dump_stack+0xe0/0x138
___might_sleep+0x2a8/0x2c8
__might_sleep+0x78/0x10c
dput+0x7c/0x750
block_dump___mark_inode_dirty+0x120/0x17c
__mark_inode_dirty+0x344/0x11f0
f2fs_mark_inode_dirty_sync+0x40/0x50
__insert_extent_tree+0x2e0/0x2f4
f2fs_update_extent_tree_range+0xcf4/0xde8
f2fs_update_extent_cache+0x114/0x12c
f2fs_update_data_blkaddr+0x40/0x50
write_data_page+0x150/0x314
do_write_data_page+0x648/0x2318
__write_data_page+0xdb4/0x1640
f2fs_write_cache_pages+0x768/0xafc
__f2fs_write_data_pages+0x590/0x1218
f2fs_write_data_pages+0x64/0x74
do_writepages+0x74/0xe4
__writeback_single_inode+0xdc/0x15f0
writeback_sb_inodes+0x574/0xc98
__writeback_inodes_wb+0x190/0x204
wb_writeback+0x730/0xf14
wb_check_old_data_flush+0x1bc/0x1c8
wb_workfn+0x554/0xf74
process_one_work+0x440/0x118c
worker_thread+0xac/0x974
kthread+0x1a0/0x1c8
ret_from_fork+0x10/0x1c
Signed-off-by: Zhikang Zhang <zhangzhikang1@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
generic/417 reported as blow:
------------[ cut here ]------------
kernel BUG at /home/yuchao/git/devf2fs/inode.c:695!
invalid opcode: 0000 [#1] PREEMPT SMP
CPU: 1 PID: 21697 Comm: umount Tainted: G W O 4.18.0-rc2+ #39
Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
EIP: f2fs_evict_inode+0x556/0x580 [f2fs]
Call Trace:
? _raw_spin_unlock+0x2c/0x50
evict+0xa8/0x170
dispose_list+0x34/0x40
evict_inodes+0x118/0x120
generic_shutdown_super+0x41/0x100
? rcu_read_lock_sched_held+0x97/0xa0
kill_block_super+0x22/0x50
kill_f2fs_super+0x6f/0x80 [f2fs]
deactivate_locked_super+0x3d/0x70
deactivate_super+0x40/0x60
cleanup_mnt+0x39/0x70
__cleanup_mnt+0x10/0x20
task_work_run+0x81/0xa0
exit_to_usermode_loop+0x59/0xa7
do_fast_syscall_32+0x1f5/0x22c
entry_SYSENTER_32+0x53/0x86
EIP: f2fs_evict_inode+0x556/0x580 [f2fs]
It can simply reproduced with scripts:
Enable quota feature during mkfs.
Testcase1:
1. mkfs.f2fs /dev/zram0
2. mount -t f2fs /dev/zram0 /mnt/f2fs
3. xfs_io -f /mnt/f2fs/file -c "pwrite 0 4k" -c "fsync"
4. godown /mnt/f2fs
5. umount /mnt/f2fs
6. mount -t f2fs -o ro /dev/zram0 /mnt/f2fs
7. umount /mnt/f2fs
Testcase2:
1. mkfs.f2fs /dev/zram0
2. mount -t f2fs /dev/zram0 /mnt/f2fs
3. touch /mnt/f2fs/file
4. create process[pid = x] do:
a) open /mnt/f2fs/file;
b) unlink /mnt/f2fs/file
5. godown -f /mnt/f2fs
6. kill process[pid = x]
7. umount /mnt/f2fs
8. mount -t f2fs -o ro /dev/zram0 /mnt/f2fs
9. umount /mnt/f2fs
The reason is: during recovery, i_{c,m}time of inode will be updated, then
the inode can be set dirty w/o being tracked in sbi->inode_list[DIRTY_META]
global list, so later write_checkpoint will not flush such dirty inode into
node page.
Once umount is called, sync_filesystem() in generic_shutdown_super() will
skip syncng dirty inodes due to sb_rdonly check, leaving dirty inodes
there.
To solve this issue, during umount, add remove SB_RDONLY flag in
sb->s_flags, to make sure sync_filesystem() will not be skipped.
Signed-off-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>
Don't limit printing log, so that we will not miss any key messages.
This reverts commit a36c106dff.
In addition, we use printk_ratelimited to avoid too many log prints.
- error injection
- discard submission failure
Signed-off-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>
In this round, we've tuned f2fs to improve general performance by serializing
block allocation and enhancing discard flows like fstrim which avoids user IO
contention. And we've added fsync_mode=nobarrier which gives an option to user
where it skips issuing cache_flush commands to underlying flash storage. And
there are many bug fixes related to fuzzed images, revoked atomic writes, quota
ops, and minor direct IO.
Enhancement:
- add fsync_mode=nobarrier which bypasses cache_flush command
- enhance the discarding flow which avoids user IOs and issues in LBA order
- readahead some encrypted blocks during GC
- enable in-memory inode checksum to verify the blocks if F2FS_CHECK_FS is set
- enhance nat_bits behavior
- set -o discard by default
- set REQ_RAHEAD to bio in ->readpages
Bug fixes:
- fix a corner case to corrupt atomic_writes revoking flow
- revisit i_gc_rwsem to fix race conditions
- fix some dio behaviors captured by xfstests
- correct handling errors given by quota-related failures
- add many sanity check flows to avoid fuzz test failures
- add more error number propagation to their callers
- fix several corner cases to continue fault injection w/ shutdown loop
-----BEGIN PGP SIGNATURE-----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=zgbL
-----END PGP SIGNATURE-----
Merge tag 'f2fs-for-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs
Pull f2fs updates from Jaegeuk Kim:
"In this round, we've tuned f2fs to improve general performance by
serializing block allocation and enhancing discard flows like fstrim
which avoids user IO contention. And we've added fsync_mode=nobarrier
which gives an option to user where it skips issuing cache_flush
commands to underlying flash storage. And there are many bug fixes
related to fuzzed images, revoked atomic writes, quota ops, and minor
direct IO.
Enhancements:
- add fsync_mode=nobarrier which bypasses cache_flush command
- enhance the discarding flow which avoids user IOs and issues in
LBA order
- readahead some encrypted blocks during GC
- enable in-memory inode checksum to verify the blocks if
F2FS_CHECK_FS is set
- enhance nat_bits behavior
- set -o discard by default
- set REQ_RAHEAD to bio in ->readpages
Bug fixes:
- fix a corner case to corrupt atomic_writes revoking flow
- revisit i_gc_rwsem to fix race conditions
- fix some dio behaviors captured by xfstests
- correct handling errors given by quota-related failures
- add many sanity check flows to avoid fuzz test failures
- add more error number propagation to their callers
- fix several corner cases to continue fault injection w/ shutdown
loop"
* tag 'f2fs-for-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs: (89 commits)
f2fs: readahead encrypted block during GC
f2fs: avoid fi->i_gc_rwsem[WRITE] lock in f2fs_gc
f2fs: fix performance issue observed with multi-thread sequential read
f2fs: fix to skip verifying block address for non-regular inode
f2fs: rework fault injection handling to avoid a warning
f2fs: support fault_type mount option
f2fs: fix to return success when trimming meta area
f2fs: fix use-after-free of dicard command entry
f2fs: support discard submission error injection
f2fs: split discard command in prior to block layer
f2fs: wake up gc thread immediately when gc_urgent is set
f2fs: fix incorrect range->len in f2fs_trim_fs()
f2fs: refresh recent accessed nat entry in lru list
f2fs: fix avoid race between truncate and background GC
f2fs: avoid race between zero_range and background GC
f2fs: fix to do sanity check with block address in main area v2
f2fs: fix to do sanity check with inline flags
f2fs: fix to reset i_gc_failures correctly
f2fs: fix invalid memory access
f2fs: fix to avoid broken of dnode block list
...
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>
This reverts the commit - "b93f771 - f2fs: remove writepages lock"
to fix the drop in sequential read throughput.
Test: ./tiotest -t 32 -d /data/tio_tmp -f 32 -b 524288 -k 1 -k 3 -L
device: UFS
Before -
read throughput: 185 MB/s
total read requests: 85177 (of these ~80000 are 4KB size requests).
total write requests: 2546 (of these ~2208 requests are written in 512KB).
After -
read throughput: 758 MB/s
total read requests: 2417 (of these ~2042 are 512KB reads).
total write requests: 2701 (of these ~2034 requests are written in 512KB).
Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
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>
Previously, once fault injection is on, by default, all kind of faults
will be injected to f2fs, if we want to trigger single or specified
combined type during the test, we need to configure sysfs entry, it will
be a little inconvenient to integrate sysfs configuring into testsuit,
such as xfstest.
So this patch introduces a new mount option 'fault_type' to assist old
option 'fault_injection', with these two mount options, we can specify
any fault rate/type at mount-time.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to support discard submission error injection for testing
error handling of __submit_discard_cmd().
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Some devices has small max_{hw,}discard_sectors, so that in
__blkdev_issue_discard(), one big size discard bio can be split
into multiple small size discard bios, result in heavy load in IO
scheduler and device, which can hang other sync IO for long time.
Now, f2fs is trying to control discard commands more elaboratively,
in order to make less conflict in between discard IO and user IO
to enhance application's performance, so in this patch, we will
split discard bio in f2fs in prior to in block layer to reduce
issuing multiple discard bios in a short time.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Introduce nat_list_lock to protect nm_i->nat_entries list, and manage
it as a LRU list, refresh location for therein recent accessed entries
in the list.
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>
Return statements in functions returning bool should use true or false
instead of an integer value.
This issue was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If config CONFIG_F2FS_FAULT_INJECTION is on, for both read or write path
we will call find_lock_page() to get the page, but for read path, it
missed to passing FGP_ACCESSED to allocator to active the page in LRU
list, result in being reclaimed in advance incorrectly, fix it.
Reported-by: Xianrong Zhou <zhouxianrong@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In order to give advise to f2fs to recognize hot/cold file, it is possible
that we can set specific bit in inode.i_advise through setxattr(), but
there are several bits which are used internally, such as encrypt_bit,
keep_size_bit, they should never be changed through setxattr().
So that this patch 1) adds FADVISE_MODIFIABLE_BITS to filter modifiable
bits user given, 2) supports to clear {hot,cold}_file bits.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Since commit 201ef5e080 ("f2fs: improve shrink performance of extent nodes"),
there is no user of EXT_TREE_VEC_SIZE, just kill it for cleanup.
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>
If we attempt to request more blocks than we have room for, we try to
instead request as much as we can, however, alloc_valid_block_count
is not decremented to match the new value, allowing it to drift higher
until the next checkpoint. This always decrements it when the requested
amount cannot be fulfilled.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For small granularity discard which size is smaller than 64KB, if we
issue those kind of discards orderly by size, their IOs will be spread
into entire logical address, so that in FTL, L2P table will be updated
randomly, result bad wear rate in the table.
In this patch, we choose to issue small discard by LBA order, by this
way, we can expect that L2P table updates from adjacent discard IOs can
be merged in the cache, so it can reduce lifetime wearing of flash.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Chao Yu