- 03 Apr, 2020 4 commits
-
-
Chao Yu authored
In below error path, tpages[i] could be NULL, fix to check it before releasing it. - f2fs_read_multi_pages - f2fs_alloc_dic - f2fs_free_dic Fixes: 61fbae2b ("f2fs: fix to avoid NULL pointer dereference") Signed-off-by:
Chao Yu <yuchao0@huawei.com> Signed-off-by:
Jaegeuk Kim <jaegeuk@kernel.org>
-
Chao Yu authored
fstest reports below message when compression is on: generic/424 1s ... - output mismatch --- tests/generic/424.out +++ results/generic/424.out.bad @@ -1,2 +1,26 @@ QA output created by 424 +[!] Attribute compressed should be set +Failed +stat_test failed +[!] Attribute compressed should be set +Failed +stat_test failed We missed to set STATX_ATTR_COMPRESSED on compressed inode in getattr(), fix it. Signed-off-by: -
Chao Yu authored
Just cleanup, no logic change. Signed-off-by:
-
Chao Yu authored
Add zstd compress algorithm support, use "compress_algorithm=zstd" mountoption to enable it. Signed-off-by:
-
- 31 Mar, 2020 19 commits
-
-
Chao Yu authored
Add below two callback interfaces in struct f2fs_compress_ops: int (*init_decompress_ctx)(struct decompress_io_ctx *dic); void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic); Which will be used by zstd compress algorithm later. In addition, this patch adds callback function pointer check, so that specified algorithm can avoid defining unneeded functions. Signed-off-by:
-
Chao Yu authored
Otherwise, it will cause memory leak. Signed-off-by:
-
Chao Yu authored
Use LZ4 as default compression algorithm, as compared to LZO, it shows almost the same compression ratio and much better decompression speed. Signed-off-by:
-
Chao Yu authored
{cic,dic}.ref should be initialized to number of compressed pages, let's initialize it directly rather than doing w/ f2fs_set_compressed_page(). Signed-off-by: -
Chao Yu authored
Multipage read flow should consider fsverity, so it needs to use f2fs_readpage_limit() instead of i_size_read() to check EOF condition. Signed-off-by:
-
YueHaibing authored
Fix gcc warnings: In file included from fs/f2fs/dir.c:15:0: fs/f2fs/xattr.h:157:13: warning: 'f2fs_destroy_xattr_caches' defined but not used [-Wunused-function] static void f2fs_destroy_xattr_caches(struct f2fs_sb_info *sbi) { } ^~~~~~~~~~~~~~~~~~~~~~~~~ fs/f2fs/xattr.h:156:12: warning: 'f2fs_init_xattr_caches' defined but not used [-Wunused-function] static int f2fs_init_xattr_caches(struct f2fs_sb_info *sbi) { return 0; } Reported-by:Hulk Robot <hulkci@huawei.com> Fixes: a999150f ("f2fs: use kmem_cache pool during inline xattr lookups") Signed-off-by:
YueHaibing <yuehaibing@huawei.com> Reviewed-by:
-
Chao Yu authored
On image that has verity and compression feature, if compressed pages and non-compressed pages are mixed in one bio, we may double unlock non-compressed page in below flow: - f2fs_post_read_work - f2fs_decompress_work - f2fs_decompress_bio - __read_end_io - unlock_page - fsverity_enqueue_verify_work - f2fs_verity_work - f2fs_verify_bio - unlock_page So it should skip handling non-compressed page in f2fs_decompress_work() if verity is on. Besides, add missing dec_page_count() in f2fs_verify_bio(). Signed-off-by: -
Chao Yu authored
f2fs_inode_info.flags is unsigned long variable, it has 32 bits in 32bit architecture, since we introduced FI_MMAP_FILE flag when we support data compression, we may access memory cross the border of .flags field, corrupting .i_sem field, result in below deadlock. To fix this issue, let's expand .flags as an array to grab enough space to store new flags. Call Trace: __schedule+0x8d0/0x13fc ? mark_held_locks+0xac/0x100 schedule+0xcc/0x260 rwsem_down_write_slowpath+0x3ab/0x65d down_write+0xc7/0xe0 f2fs_drop_nlink+0x3d/0x600 [f2fs] f2fs_delete_inline_entry+0x300/0x440 [f2fs] f2fs_delete_entry+0x3a1/0x7f0 [f2fs] f2fs_unlink+0x500/0x790 [f2fs] vfs_unlink+0x211/0x490 do_unlinkat+0x483/0x520 sys_unlink+0x4a/0x70 do_fast_syscall_32+0x12b/0x683 entry_SYSENTER_32+0xaa/0x102 Fixes: 4c8ff709 ("f2fs: support data compression") Tested-by:
Ondrej Jirman <megous@megous.com> Signed-off-by:
-
Chao Yu authored
If both compression and fsverity feature is on, generic/572 will report below NULL pointer dereference bug. BUG: kernel NULL pointer dereference, address: 0000000000000018 RIP: 0010:f2fs_verity_work+0x60/0x90 [f2fs] #PF: supervisor read access in kernel mode Workqueue: fsverity_read_queue f2fs_verity_work [f2fs] RIP: 0010:f2fs_verity_work+0x60/0x90 [f2fs] Call Trace: process_one_work+0x16c/0x3f0 worker_thread+0x4c/0x440 ? rescuer_thread+0x350/0x350 kthread+0xf8/0x130 ? kthread_unpark+0x70/0x70 ret_from_fork+0x35/0x40 There are two issue in f2fs_verity_work(): - it needs to traverse and verify all pages in bio. - if pages in bio belong to non-compressed cluster, accessing decompress IO context stored in page private will cause NULL pointer dereference. Fix them. Signed-off-by:
-
Chao Yu authored
In f2fs_decompress_end_io(), we should clear PG_error flag before page unlock, otherwise reread will fail due to the flag as described in commit fb7d70db ("f2fs: clear PageError on the read path"). Signed-off-by:
-
Chao Yu authored
In f2fs_tmpfile(), parent inode's encryption info is only used when inheriting encryption context to its child inode, however, we have already called fscrypt_get_encryption_info() in fscrypt_inherit_context() to get the encryption info, so just removing unneeded one in f2fs_tmpfile(). Signed-off-by:
-
Chao Yu authored
Data flush can generate heavy IO and cause long latency during flush, so it's not appropriate to trigger it in foreground operation. And also, we may face below potential deadlock during data flush: - f2fs_write_multi_pages - f2fs_write_raw_pages - f2fs_write_single_data_page - f2fs_balance_fs - f2fs_balance_fs_bg - f2fs_sync_dirty_inodes - filemap_fdatawrite -- stuck on flush same cluster Signed-off-by: -
Chao Yu authored
BUG: kernel NULL pointer dereference, address: 0000000000000000 RIP: 0010:f2fs_write_begin+0x823/0xb90 [f2fs] Call Trace: f2fs_quota_write+0x139/0x1d0 [f2fs] write_blk+0x36/0x80 [quota_tree] get_free_dqblk+0x42/0xa0 [quota_tree] do_insert_tree+0x235/0x4a0 [quota_tree] do_insert_tree+0x26e/0x4a0 [quota_tree] do_insert_tree+0x26e/0x4a0 [quota_tree] do_insert_tree+0x26e/0x4a0 [quota_tree] qtree_write_dquot+0x70/0x190 [quota_tree] v2_write_dquot+0x43/0x90 [quota_v2] dquot_acquire+0x77/0x100 f2fs_dquot_acquire+0x2f/0x60 [f2fs] dqget+0x310/0x450 dquot_transfer+0x7e/0x120 f2fs_setattr+0x11a/0x4a0 [f2fs] notify_change+0x349/0x480 chown_common+0x168/0x1c0 do_fchownat+0xbc/0xf0 __x64_sys_fchownat+0x20/0x30 do_syscall_64+0x5f/0x220 entry_SYSCALL_64_after_hwframe+0x44/0xa9 Passing fsdata parameter to .write_{begin,end} in f2fs_quota_write(), so that if quota file is compressed one, we can avoid above NULL pointer dereference when updating quota content. Signed-off-by: -
Chao Yu authored
Merge below two conditions into f2fs_may_encrypt() for cleanup - IS_ENCRYPTED() - DUMMY_ENCRYPTION_ENABLED() Check IS_ENCRYPTED(inode) condition in f2fs_init_inode_metadata() is enough since we have already set encrypt flag in f2fs_new_inode(). Signed-off-by:
-
Chao Yu authored
We should always check F2FS_I(inode)->cp_task condition in prior to other conditions in __should_serialize_io() to avoid deadloop described in commit 040d2bb3 ("f2fs: fix to avoid deadloop if data_flush is on"), however we break this rule when we support compression, fix it. Signed-off-by:
-
Chao Yu authored
In order to shrink page lock coverage. Signed-off-by:
-
Chao Yu authored
generic/232 reports below deadlock: fsstress D 0 96980 96969 0x00084000 Call Trace: schedule+0x4a/0xb0 io_schedule+0x12/0x40 __lock_page+0x127/0x1d0 pagecache_get_page+0x1d8/0x250 prepare_compress_overwrite+0xe0/0x490 [f2fs] f2fs_prepare_compress_overwrite+0x5d/0x80 [f2fs] f2fs_write_begin+0x833/0xb90 [f2fs] f2fs_quota_write+0x145/0x1e0 [f2fs] write_blk+0x36/0x80 [quota_tree] do_insert_tree+0x2ac/0x4a0 [quota_tree] do_insert_tree+0x26e/0x4a0 [quota_tree] qtree_write_dquot+0x70/0x190 [quota_tree] v2_write_dquot+0x43/0x90 [quota_v2] dquot_acquire+0x77/0x100 f2fs_dquot_acquire+0x2f/0x60 [f2fs] dqget+0x310/0x450 dquot_transfer+0xb2/0x120 f2fs_setattr+0x11a/0x4a0 [f2fs] notify_change+0x349/0x480 chown_common+0x168/0x1c0 do_fchownat+0xbc/0xf0 __x64_sys_lchown+0x21/0x30 do_syscall_64+0x5f/0x220 entry_SYSCALL_64_after_hwframe+0x44/0xa9 task PC stack pid father kworker/u256:0 D 0 103444 2 0x80084000 Workqueue: writeback wb_workfn (flush-251:1) Call Trace: schedule+0x4a/0xb0 schedule_timeout+0x15e/0x2f0 io_schedule_timeout+0x19/0x40 congestion_wait+0x7e/0x120 f2fs_write_multi_pages+0x12a/0x840 [f2fs] f2fs_write_cache_pages+0x48f/0x790 [f2fs] f2fs_write_data_pages+0x2db/0x330 [f2fs] do_writepages+0x1a/0x60 __writeback_single_inode+0x3d/0x340 writeback_sb_inodes+0x225/0x4a0 wb_writeback+0xf7/0x320 wb_workfn+0xba/0x470 process_one_work+0x16c/0x3f0 worker_thread+0x4c/0x440 kthread+0xf8/0x130 ret_from_fork+0x35/0x40 fsstress D 0 5277 5266 0x00084000 Call Trace: schedule+0x4a/0xb0 rwsem_down_write_slowpath+0x29d/0x540 block_operations+0x105/0x360 [f2fs] f2fs_write_checkpoint+0x101/0x1010 [f2fs] f2fs_sync_fs+0xa8/0x130 [f2fs] f2fs_do_sync_file+0x1ad/0x890 [f2fs] do_fsync+0x38/0x60 __x64_sys_fdatasync+0x13/0x20 do_syscall_64+0x5f/0x220 entry_SYSCALL_64_after_hwframe+0x44/0xa9 The root cause is there is potential deadlock between quota data update and writeback. Kworker Thread B Thread C - f2fs_write_cache_pages - lock whole cluster --- A - f2fs_write_multi_pages - f2fs_write_raw_pages - f2fs_write_single_data_page - f2fs_do_write_data_page - f2fs_setattr - f2fs_lock_op --- B - f2fs_write_checkpoint - block_operations - f2fs_lock_all --- B - dquot_transfer - f2fs_quota_write - f2fs_prepare_compress_overwrite - pagecache_get_page --- A - f2fs_trylock_op failed --- B - congestion_wait - goto rewrite To fix this issue, during quota file writeback, just redirty all pages left in cluster rather holding pages' lock in cluster and looping retrying lock cp_rwsem. Signed-off-by: -
DongDongJu authored
This lock can be a contention with multi 4k random read IO with single inode. example) fio --output=test --name=test --numjobs=60 --filename=/media/samsung960pro/file_test --rw=randread --bs=4k --direct=1 --time_based --runtime=7 --ioengine=libaio --iodepth=256 --group_reporting --size=10G With this commit, it remove that possible lock contention. Signed-off-by:
Dongjoo Seo <commisori28@gmail.com> Signed-off-by:
-
Chao Yu authored
- f2fs_iget - do_read_inode - set_inode_flag(, FI_COMPRESSED_FILE) - __mark_inode_dirty_flag(, true) It's unnecessary, so let's just mark compressed inode dirty while compressed inode conversion. Signed-off-by:
-
- 23 Mar, 2020 6 commits
-
-
Chao Yu authored
por_fsstress reports inconsistent status in orphan inode, the root cause of this is in f2fs_write_raw_pages() we decrease i_compr_blocks incorrectly due to wrong calculation in f2fs_compressed_blocks(). So this patch exposes below two functions based on __f2fs_cluster_blocks: - f2fs_compressed_blocks: get count of compressed blocks in compressed cluster - f2fs_cluster_blocks: get count of valid blocks (including reserved blocks) in compressed cluster. Then use f2fs_compress_blocks() to get correct compressed blocks count in f2fs_write_raw_pages(). sanity_check_inode: inode (ino=ad80) hash inconsistent i_compr_blocks:2, i_blocks:1, run fsck to fix Signed-off-by:
-
Gustavo A. R. Silva authored
The current codebase makes use of the zero-length array language extension to the C90 standard, but the preferred mechanism to declare variable-length types such as these ones is a flexible array member[1][2], introduced in C99: struct foo { int stuff; struct boo array[]; }; By making use of the mechanism above, we will get a compiler warning in case the flexible array does not occur last in the structure, which will help us prevent some kind of undefined behavior bugs from being inadvertently introduced[3] to the codebase from now on. Also, notice that, dynamic memory allocations won't be affected by this change: "Flexible array members have incomplete type, and so the sizeof operator may not be applied. As a quirk of the original implementation of zero-length arrays, sizeof evaluates to zero."[1] This issue was found with the help of Coccinelle. [1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html [2] https://github.com/KSPP/linux/issues/21 [3] commit 76497732 ("cxgb3/l2t: Fix undefined behaviour") Signed-off-by:Gustavo A. R. Silva <gustavo@embeddedor.com> Reviewed-by:
-
Chao Yu authored
Fields in struct f2fs_super_block should be updated under coverage of sb_lock, fix to adjust update_sb_metadata() for that rule. Fixes: 04f0b2ea ("f2fs: ioctl for removing a range from F2FS") Signed-off-by:
-
Sahitya Tummala authored
Add and set a new CP flag CP_RESIZEFS_FLAG during online resize FS to help fsck fix the metadata mismatch that may happen due to SPO during resize, where SB got updated but CP data couldn't be written yet. fsck errors - Info: CKPT version = 6ed7bccb Wrong user_block_count(2233856) [f2fs_do_mount:3365] Checkpoint is polluted Signed-off-by:Sahitya Tummala <stummala@codeaurora.org> Reviewed-by:
-
Sahitya Tummala authored
Even though online resize is successfully done, a SPO immediately after resize, still causes below error in the next mount. [ 11.294650] F2FS-fs (sda8): Wrong user_block_count: 2233856 [ 11.300272] F2FS-fs (sda8): Failed to get valid F2FS checkpoint This is because after FS metadata is updated in update_fs_metadata() if the SBI_IS_DIRTY is not dirty, then CP will not be done to reflect the new user_block_count. Signed-off-by:
-
Chao Yu authored
It's been observed that kzalloc() on lookup_all_xattrs() are called millions of times on Android, quickly becoming the top abuser of slub memory allocator. Use a dedicated kmem cache pool for xattr lookups to mitigate this. Signed-off-by:
Park Ju Hyung <qkrwngud825@gmail.com> Signed-off-by:
-
- 19 Mar, 2020 11 commits
-
-
Jaegeuk Kim authored
FG_GC needs to move entire section more quickly. Reviewed-by:
-
Chao Yu authored
Signed-off-by:
-
Chao Yu authored
__f2fs_bio_alloc() won't fail due to memory pool backend, remove unneeded __GFP_NOFAIL flag in __f2fs_bio_alloc(). Signed-off-by:
-
Chao Yu authored
With this newly introduced interface, user can get block number compression saved in target inode. Signed-off-by:
-
Chao Yu authored
If we are in write IO path, we need to avoid using GFP_KERNEL. Signed-off-by:
-
Chao Yu authored
In order to avoid polluting global slab cache namespace. Signed-off-by:
-
Chao Yu authored
As Geert Uytterhoeven reported: for parameter HZ/50 in congestion_wait(BLK_RW_ASYNC, HZ/50); On some platforms, HZ can be less than 50, then unexpected 0 timeout jiffies will be set in congestion_wait(). This patch introduces a macro DEFAULT_IO_TIMEOUT to wrap a determinate value with msecs_to_jiffies(20) to instead HZ/50 to avoid such issue. Quoted from Geert Uytterhoeven: "A timeout of HZ means 1 second. HZ/50 means 20 ms, but has the risk of being zero, if HZ < 50. If you want to use a timeout of 20 ms, you best use msecs_to_jiffies(20), as that takes care of the special cases, and never returns 0." Signed-off-by:
-
Jaegeuk Kim authored
This fixes skipping GC when segment is full in large section. Signed-off-by: -
Jaegeuk Kim authored
If first segment is empty and migration_granularity is 1, we can't move this at all. Reviewed-by:
-
Chao Yu authored
There are three status for background gc: on, off and sync, it's a little bit confused to use test_opt(BG_GC) and test_opt(FORCE_FG_GC) combinations to indicate status of background gc. So let's remove F2FS_MOUNT_BG_GC and F2FS_MOUNT_FORCE_FG_GC mount options, and add F2FS_OPTION().bggc_mode with below three status to clean up codes and enhance bggc mode's scalability. enum { BGGC_MODE_ON, /* background gc is on */ BGGC_MODE_OFF, /* background gc is off */ BGGC_MODE_SYNC, /* * background gc is on, migrating blocks * like foreground gc */ }; Signed-off-by: -
Chao Yu authored
This patch removes F2FS_MOUNT_ADAPTIVE and F2FS_MOUNT_LFS mount options, and add F2FS_OPTION.fs_mode with below two status to indicate filesystem mode. enum { FS_MODE_ADAPTIVE, /* use both lfs/ssr allocation */ FS_MODE_LFS, /* use lfs allocation only */ }; It can enhance code readability and fs mode's scalability. Signed-off-by:
-
