- 29 Aug, 2017 6 commits
-
-
Jaegeuk Kim authored
This patch fixes to avoid needless wake ups. Reviewed-by:
Chao Yu <yuchao0@huawei.com> Signed-off-by:
Jaegeuk Kim <jaegeuk@kernel.org>
-
Jaegeuk Kim authored
If file offset is insane, we have to return error instead of kernel panic. Reported-by:
Eric Zhang <followme999@163.com> Signed-off-by:
-
Chao Yu authored
If file was not opened with atomic write mode, but user uses atomic write ioctl to fsync datas, in the flow, we should not fsync that file with atomic write mode. Fixes: 608514de ("f2fs: set fsync mark only for the last dnode") Signed-off-by:
-
Chao Yu authored
This patch fixes to clear FI_HOT_DATA correctly in below path: - error handling in f2fs_ioc_start_atomic_write - after commit atomic write in f2fs_ioc_commit_atomic_write - after drop atomic write in drop_inmem_pages Signed-off-by:
-
Chao Yu authored
In f2fs_issue_flush, due to out-of-order execution of CPU, wake_up can be called before we insert issue_list, result in long latency of wait_for_completion. Fix this by adding smp_mb() to force the order of related codes. Signed-off-by:
-
Jaegeuk Kim authored
It's time to issue all the discard commands, if user sets the idle time. Reviewed-by:
-
- 21 Aug, 2017 8 commits
-
-
Chao Yu authored
Commit d618ebaf ("f2fs: enable small discard by default") enables f2fs to issue 4K size discard in real-time discard mode. However, issuing smaller discard may cost more lifetime but releasing less free space in flash device. Since f2fs has ability of separating hot/cold data and garbage collection, we can expect that small-sized invalid region would expand soon with OPU, deletion or garbage collection on valid datas, so it's better to delay or skip issuing smaller size discards, it could help to reduce overmuch consumption of IO bandwidth and lifetime of flash storage. This patch makes f2fs selectng 64K size as its default minimal granularity, and issue discard with the size which is not smaller than minimal granularity. Also it exposes discard granularity as sysfs entry for configuration in different scenario. Jaegeuk Kim: We must issue all the accumulated discard commands when fstrim is called. So, I've added pend_list_tag[] to indicate whether we should issue the commands or not. If tag sets P_ACTIVE or P_TRIM, we have to issue them. P_TRIM is set once at a time, given fstrim trigger. In addition, issue_discard_thread is calling too much due to the number of discard commands remaining in the pending list. I added a timer to control it likewise gc_thread. Signed-off-by:
-
Yunlong Song authored
Signed-off-by:
Yunlong Song <yunlong.song@huawei.com> Reviewed-by:
-
Jaegeuk Kim authored
We need to check HOT_DATA to truncate any previous data block when doing roll-forward recovery. Cc: <stable@vger.kernel.org> Reviewed-by:
-
Chao Yu authored
This patch adds tracepoint for f2fs_gc. Signed-off-by:
-
Chao Yu authored
During atomic committing, if we encounter -ENOMEM in revoke path, it's better to give a chance to retry revoking. Signed-off-by:
-
Jaegeuk Kim authored
If we set CP_ERROR_FLAG in roll-forward error, f2fs is no longer to proceed any IOs due to f2fs_cp_error(). But, for example, if some stale data is involved on roll-forward process, we're able to get -ENOENT, getting fs stuck. If we get any error, let fill_super set SBI_NEED_FSCK and try to recover back to stable point. Cc: <stable@vger.kernel.org> Reviewed-by:
-
Qiuyang Sun authored
Currently, the two flags F2FS_GET_BLOCK_[READ|DIO] are totally equivalent and can be used interchangably in all scenarios they are involved in. Neither of the flags is referenced in f2fs_map_blocks(), making them both the default case. To remove the ambiguity, this patch merges both flags into F2FS_GET_BLOCK_DEFAULT, and introduces an enum for all distinct flags. Signed-off-by:
Qiuyang Sun <sunqiuyang@huawei.com> Reviewed-by:
-
Chao Yu authored
This patch supports to enable f2fs to accept quota information through mount option: - {usr,grp,prj}jquota=<quota file path> - jqfmt=<quota type> Then, in ->mount flow, we can recover quota file during log replaying, by this, journelled quota can be supported. Signed-off-by:
-
- 15 Aug, 2017 5 commits
-
-
Chao Yu authored
While comparing signed and unsigned variables, compiler will converts the signed value to unsigned one, due to this reason, {in,de}crease_sleep_time may return overflowed result. Signed-off-by: -
Chao Yu authored
We only need to sync quota file with appointed quota type instead of all types in f2fs_quota_{on,off}. Signed-off-by: -
Jaegeuk Kim authored
This patch adds a sysfs entry to control urgent mode for background GC. If this is set, background GC thread conducts GC with gc_urgent_sleep_time all the time. Reviewed-by:
-
Jaegeuk Kim authored
We expect cold files write data sequentially, but sometimes some of small data can be updated, which incurs fragmentation. Let's avoid that. Reviewed-by:
-
Yunlong Song authored
The current size value is not correct and will miss bitmap check. Signed-off-by:
-
- 10 Aug, 2017 3 commits
-
-
Chao Yu authored
This patch enables inner app/fs io stats and introduces below virtual fs nodes for exposing stats info: /sys/fs/f2fs/<dev>/iostat_enable /proc/fs/f2fs/<dev>/iostat_info Signed-off-by:
-
Yunlong Song authored
Signed-off-by:
-
Yunlong Song authored
When cur_valid_map passes the f2fs_test_and_set(,clear)_bit test, cur_valid_map_mir update is skipped unlikely, so fix it. The fix now changes the mirror check together with cur_valid_map all the time. Signed-off-by:
-
- 04 Aug, 2017 5 commits
-
-
Jaegeuk Kim authored
This patch reduces contention of printks. Signed-off-by: -
Jaegeuk Kim authored
This patch exposes what features are supported by current f2fs build to sysfs entry via: /sys/fs/f2fs/features/ /sys/fs/f2fs/dev/features Reviewed-by:
-
Chao Yu authored
This patch adds to support inode checksum in f2fs. Signed-off-by:
-
Jaegeuk Kim authored
This must return size, not error number. Signed-off-by: -
Yunlong Song authored
Let node writeback also do f2fs_balance_fs to ensure there are always enough free segments. Signed-off-by:
-
- 31 Jul, 2017 10 commits
-
-
Chao Yu authored
This patch introduces f2fs_statfs_project, it enables to show usage status of directory tree which is limited with project quota. Signed-off-by:
-
Chao Yu authored
This patch adds FS_IOC_FSSETXATTR/FS_IOC_FSGETXATTR ioctl interface support for f2fs. The interface is kept consistent with the one of ext4/xfs. Signed-off-by:
-
Jaegeuk Kim authored
We have a node chain to serialize node block writes, so if any IOs for node block writes are reordered, we'll get broken node chain. IOWs, roll-forward recovery will see all or none node blocks given fsync mark. E.g., Node chain consists of: N1 -> N2 -> N3 -> NFSYNC -> N1' -> N2' -> N'FSYNC Reordered to: 1) N1 -> N2 -> N3 -> N2' -> NFSYNC -> N'FSYNC -> power-cut 2) N1 -> N2 -> N3 -> N1' -> NFSYNC -> power-cut 3) N1 -> N2 -> NFSYNC -> N1' -> N'FSYNC -> N3 -> power-cut 4) N1 -> NFSYNC -> N1' -> N2' -> N'FSYNC -> N3 -> power-cut Roll-forward recovery can proceed to: 1) N1 -> N2 -> N3 -> NFSYNC -> X 2) N1 -> N2 -> N3 -> NFSYNC -> N1' -> X 3) N1 -> N2 -> N3 -> FSYNC -> N1' -> X 4) N1 -> X Signed-off-by: -
Jaegeuk Kim authored
This patch changes the function names of sysfs init to follow ext4. f2fs_init_sysfs <-> f2fs_register_sysfs f2fs_exit_sysfs <-> f2fs_unregister_sysfs Suggested-by:
-
Chao Yu authored
This patch adds to support plain project quota. Signed-off-by:
-
Chao Yu authored
In ->lookup(), we will have a try to recover dot or dotdot for corrupted directory, once disk quota is on, if it allocates new block during dotdot recovery, we need to record disk quota info for the allocation, so this patch fixes this issue by adding missing dquot_initialize() in __recover_dot_dentries. Signed-off-by:
-
Chao Yu authored
This patch add new flag F2FS_EXTRA_ATTR storing in inode.i_inline to indicate that on-disk structure of current inode is extended. In order to extend, we changed the inode structure a bit: Original one: struct f2fs_inode { ... struct f2fs_extent i_ext; __le32 i_addr[DEF_ADDRS_PER_INODE]; __le32 i_nid[DEF_NIDS_PER_INODE]; } Extended one: struct f2fs_inode { ... struct f2fs_extent i_ext; union { struct { __le16 i_extra_isize; __le16 i_padding; __le32 i_extra_end[0]; }; __le32 i_addr[DEF_ADDRS_PER_INODE]; }; __le32 i_nid[DEF_NIDS_PER_INODE]; } Once F2FS_EXTRA_ATTR is set, we will steal four bytes in the head of i_addr field for storing i_extra_isize and i_padding. with i_extra_isize, we can calculate actual size of reserved space in i_addr, available attribute fields included in total extra attribute fields for current inode can be described as below: +--------------------+ | .i_mode | | ... | | .i_ext | +--------------------+ | .i_extra_isize |-----+ | .i_padding | | | .i_prjid | | | .i_atime_extra | | | .i_ctime_extra | | | .i_mtime_extra |<----+ | .i_inode_cs |<----- store blkaddr/inline from here | .i_xattr_cs | | ... | +--------------------+ | | | block address | | | +--------------------+ | .i_nid | +--------------------+ | node_footer | | (nid, ino, offset) | +--------------------+ Hence, with this patch, we would enhance scalability of f2fs inode for storing more newly added attribute. Signed-off-by: -
Chao Yu authored
This patch tries to make below macros calculating max inline size, inline dentry field size considerring reserving size-changeable space: - MAX_INLINE_DATA - NR_INLINE_DENTRY - INLINE_DENTRY_BITMAP_SIZE - INLINE_RESERVED_SIZE Then, when inline_{data,dentry} options is enabled, it allows us to reserve inline space with different size flexibly for adding newly introduced inode attribute. Signed-off-by: -
Jaegeuk Kim authored
This patch adds an ioctl to provide feature information to user. For exapmle, SQLite can use this ioctl to detect whether f2fs support atomic write or not. Reviewed-by:
-
Chao Yu authored
When ->freeze_fs is called from lvm for doing snapshot, it needs to make sure there will be no more changes in filesystem's data, however, previously, background threads like GC thread wasn't aware of freezing, so in environment with active background threads, data of snapshot becomes unstable. This patch fixes this issue by adding sb_{start,end}_intwrite in below background threads: - GC thread - flush thread - discard thread Note that, don't use sb_start_intwrite() in gc_thread_func() due to: generic/241 reports below bug: ====================================================== WARNING: possible circular locking dependency detected 4.13.0-rc1+ #32 Tainted: G O ------------------------------------------------------ f2fs_gc-250:0/22186 is trying to acquire lock: (&sbi->gc_mutex){+.+...}, at: [<f8fa7f0b>] f2fs_sync_fs+0x7b/0x1b0 [f2fs] but task is already holding lock: (sb_internal#2){++++.-}, at: [<f8fb5609>] gc_thread_func+0x159/0x4a0 [f2fs] which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (sb_internal#2){++++.-}: __lock_acquire+0x405/0x7b0 lock_acquire+0xae/0x220 __sb_start_write+0x11d/0x1f0 f2fs_evict_inode+0x2d6/0x4e0 [f2fs] evict+0xa8/0x170 iput+0x1fb/0x2c0 f2fs_sync_inode_meta+0x3f/0xf0 [f2fs] write_checkpoint+0x1b1/0x750 [f2fs] f2fs_sync_fs+0x85/0x1b0 [f2fs] f2fs_do_sync_file.isra.24+0x137/0xa30 [f2fs] f2fs_sync_file+0x34/0x40 [f2fs] vfs_fsync_range+0x4a/0xa0 do_fsync+0x3c/0x60 SyS_fdatasync+0x15/0x20 do_fast_syscall_32+0xa1/0x1b0 entry_SYSENTER_32+0x4c/0x7b -> #1 (&sbi->cp_mutex){+.+...}: __lock_acquire+0x405/0x7b0 lock_acquire+0xae/0x220 __mutex_lock+0x4f/0x830 mutex_lock_nested+0x25/0x30 write_checkpoint+0x2f/0x750 [f2fs] f2fs_sync_fs+0x85/0x1b0 [f2fs] sync_filesystem+0x67/0x80 generic_shutdown_super+0x27/0x100 kill_block_super+0x22/0x50 kill_f2fs_super+0x3a/0x40 [f2fs] deactivate_locked_super+0x3d/0x70 deactivate_super+0x40/0x60 cleanup_mnt+0x39/0x70 __cleanup_mnt+0x10/0x20 task_work_run+0x69/0x80 exit_to_usermode_loop+0x57/0x92 do_fast_syscall_32+0x18c/0x1b0 entry_SYSENTER_32+0x4c/0x7b -> #0 (&sbi->gc_mutex){+.+...}: validate_chain.isra.36+0xc50/0xdb0 __lock_acquire+0x405/0x7b0 lock_acquire+0xae/0x220 __mutex_lock+0x4f/0x830 mutex_lock_nested+0x25/0x30 f2fs_sync_fs+0x7b/0x1b0 [f2fs] f2fs_balance_fs_bg+0xb9/0x200 [f2fs] gc_thread_func+0x302/0x4a0 [f2fs] kthread+0xe9/0x120 ret_from_fork+0x19/0x24 other info that might help us debug this: Chain exists of: &sbi->gc_mutex --> &sbi->cp_mutex --> sb_internal#2 Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(sb_internal#2); lock(&sbi->cp_mutex); lock(sb_internal#2); lock(&sbi->gc_mutex); *** DEADLOCK *** 1 lock held by f2fs_gc-250:0/22186: #0: (sb_internal#2){++++.-}, at: [<f8fb5609>] gc_thread_func+0x159/0x4a0 [f2fs] stack backtrace: CPU: 2 PID: 22186 Comm: f2fs_gc-250:0 Tainted: G O 4.13.0-rc1+ #32 Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006 Call Trace: dump_stack+0x5f/0x92 print_circular_bug+0x1b3/0x1bd validate_chain.isra.36+0xc50/0xdb0 ? __this_cpu_preempt_check+0xf/0x20 __lock_acquire+0x405/0x7b0 lock_acquire+0xae/0x220 ? f2fs_sync_fs+0x7b/0x1b0 [f2fs] __mutex_lock+0x4f/0x830 ? f2fs_sync_fs+0x7b/0x1b0 [f2fs] mutex_lock_nested+0x25/0x30 ? f2fs_sync_fs+0x7b/0x1b0 [f2fs] f2fs_sync_fs+0x7b/0x1b0 [f2fs] f2fs_balance_fs_bg+0xb9/0x200 [f2fs] gc_thread_func+0x302/0x4a0 [f2fs] ? preempt_schedule_common+0x2f/0x4d ? f2fs_gc+0x540/0x540 [f2fs] kthread+0xe9/0x120 ? f2fs_gc+0x540/0x540 [f2fs] ? kthread_create_on_node+0x30/0x30 ret_from_fork+0x19/0x24 The deadlock occurs in below condition: GC Thread Thread B - sb_start_intwrite - f2fs_sync_file - f2fs_sync_fs - mutex_lock(&sbi->gc_mutex) - write_checkpoint - block_operations - f2fs_sync_inode_meta - iput - sb_start_intwrite - mutex_lock(&sbi->gc_mutex) Fix this by altering sb_start_intwrite to sb_start_write_trylock. Signed-off-by:
-
- 29 Jul, 2017 3 commits
-
-
Jaegeuk Kim authored
This patch resolves the below scenario. == Process 1 == == Process 2 == open(w) open(rw) begin write(new_#1) process_crash f_op->flush locks_remove_posix f_op>release read (new_#1) In order to avoid corrupted database caused by new_#1, we must do roll-back at process_crash time. In order to check that, this patch keeps task which triggers transaction begin, and does roll-back in f_op->flush before removing file locks. Reviewed-by: -
Jaegeuk Kim authored
It'd be better to retry writing atomic pages when we get -ENOMEM. Reviewed-by:
-
Ernesto A. Fernández authored
When changing a file's acl mask, __f2fs_set_acl() will first set the group bits of i_mode to the value of the mask, and only then set the actual extended attribute representing the new acl. If the second part fails (due to lack of space, for example) and the file had no acl attribute to begin with, the system will from now on assume that the mask permission bits are actual group permission bits, potentially granting access to the wrong users. Prevent this by only changing the inode mode after the acl has been set. Signed-off-by:
Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com> Reviewed-by:
-
