- 12 Apr, 2017 2 commits
-
-
Jaegeuk Kim authored
This patch fixes the following scenario. - f2fs_create/f2fs_mkdir - write_checkpoint - f2fs_mark_inode_dirty_sync - block_operations - f2fs_lock_all - f2fs_sync_inode_meta - f2fs_unlock_all - sync_inode_metadata - f2fs_lock_op - f2fs_write_inode - update_inode_page - get_node_page return -ENOENT - new_inode_page - fill_node_footer - f2fs_mark_inode_dirty_sync - ... - f2fs_unlock_op - f2fs_inode_synced - f2fs_lock_all - do_checkpoint In this checkpoint, we can get an inode page which contains zeros having valid node footer only. Cc: <stable@vger.kernel.org> Signed-off-by:Jaegeuk Kim <jaegeuk@kernel.org>
-
Chao Yu authored
Don't use blk plug covering area where there won't be any IOs being issued. Signed-off-by:
Chao Yu <yuchao0@huawei.com> Signed-off-by:
-
- 11 Apr, 2017 11 commits
-
-
Chao Yu authored
rb-tree lookup/update functions are deeply coupled into extent cache codes, it's very hard to reuse these basic functions, this patch extracts common rb-tree operation infrastructure for latter reusing. Signed-off-by:
-
Jaegeuk Kim authored
Now we're doing SSR aggressively more than ever before, so once we reach to the reserved_segment, f2fs_balance_fs will call f2fs_gc, which triggers checkpoint everytime. We actually must avoid that. Reviewed-by:
-
Jaegeuk Kim authored
This patch cleans several macros by introducing: - BLKS_PER_SEC - GET_SEC_FROM_SEG - GET_SEG_FROM_SEC - GET_ZONE_FROM_SEC - GET_ZONE_FROM_SEG Signed-off-by: -
Jaegeuk Kim authored
This patch cleans up get_valid_blocks, which has no functional change. Signed-off-by: -
Jaegeuk Kim authored
This patch fixes to submit a segment number for get_valid_blocks. Signed-off-by: -
Tomohiro Kusumi authored
Add braces around variables used within macros for those make sense to do it. Many of the macros in f2fs already do this. What this commit doesn't do is anything that changes line# as a result of adding braces, which usually affects the binary via __LINE__. Confirmed no diff in fs/f2fs/f2fs.ko before/after this commit on x86_64, to make sure this has no functional change as well as there's been no unexpected side effect due to callers' arithmetics within the existing code. Signed-off-by:
Tomohiro Kusumi <tkusumi@tuxera.com> Signed-off-by:
-
Tomohiro Kusumi authored
Callers are to unlock the page on failure after 86531d6b. Signed-off-by:
-
Chao Yu authored
In f2fs_submit_discard_endio, we will wake up waiter before setting discard command states, so waiter may use incorrect states. Change the order between complete() and states setting to fix this issue. Signed-off-by:
-
Chao Yu authored
Split f2fs_wait_discard_bios from f2fs_wait_discard_bio, just for cleanup, no logic change. Signed-off-by:
-
Chao Yu authored
Split discard_cmd_list to discard_{pend,wait}_list, so while sending/waiting discard command, we can avoid traversing unneeded entries in original list. Signed-off-by: -
Jaegeuk Kim authored
This reverts commit 3436c4bd. This makes a leak to register dirty segments. I reproduced the issue by modified postmark which injects a lot of file create/delete/update and finally triggers huge number of SSR allocations. Cc: <stable@vger.kernel.org> # v4.10+ [Jaegeuk Kim: Change missing incorrect comment] Signed-off-by:
-
- 05 Apr, 2017 9 commits
-
-
Tomohiro Kusumi authored
Since callers statically know which type to use, make_dentry_ptr() can simply be splitted into two inline functions. This way, the code has less inlined, fewer arguments, and no cast. Signed-off-by:
-
Jaegeuk Kim authored
This patch tries to split in-place-update bios from sequential bios. Suggested-by:
Yunlei He <heyunlei@huawei.com> Signed-off-by:
-
Kaixu Xia authored
The variable 'i' has been defined before, so here we can use it directly. Signed-off-by:
Kaixu Xia <xiakaixu@huawei.com> Signed-off-by:
-
Jaegeuk Kim authored
If two threads try to flush dirty pages in different inodes respectively, f2fs_write_data_pages() will produce WRITE and WRITE_SYNC one at a time, resulting in a lot of 4KB seperated IOs. So, this patch gives higher priority to WB_SYNC_ALL IOs and gathers write IOs with a big WRITE_SYNC'ed bio. Signed-off-by: -
Jaegeuk Kim authored
It would better split small and large IOs separately in order to get more consecutive big writes. The default threshold is set to 64KB, but configurable by sysfs/min_hot_blocks. Signed-off-by: -
Chao Yu authored
This patch changes to use bitmap instead of extent in struct discard_entry to indicate discard range in one segment, for fragmented space, this implementation can save memory footprint. Signed-off-by:
-
Chao Yu authored
Remove unneeded parameter and simply change flow in destroy_discard_cmd_control. Signed-off-by:
-
Chao Yu authored
Adds to count discard command entry and show the number in debugfs, also fix to add cost of discard command cache into total comsumed memory footprint. Signed-off-by:
-
Chao Yu authored
Show historical count of flush command and discard command. Signed-off-by:
-
- 30 Mar, 2017 3 commits
-
-
Jaegeuk Kim authored
- has_not_enough_free_secs node_secs: 0 dent_secs: 0 freed:0 free_segments:103 reserved:104 - f2fs_gc - get_victim_by_default alloc_mode 0, gc_mode 1, max_search 2672, offset 4654, ofs_unit 1 - do_garbage_collect start_segno 3976, end_segno 3977 type 0 - is_alive nid 22797, blkaddr 2131882, ofs_in_node 0, version 0x8/0x0 - gc_data_segment 766, segno 3976, block 512/426 not alive So, this patch fixes subtle corrupted case where node version does not match to summary version which results in infinite loop by gc. Reported-by: -
Jaegeuk Kim authored
This patch initiates SSR much eariler, resulting in less FG_GC. Signed-off-by: -
Jaegeuk Kim authored
In order to give more spatial locality, this patch changes the block allocation policy which assigns beginning of partition for small and hot data/node blocks. In order to do this, we set noheap allocation by default and introduce another mount option, heap, to reset it back. Signed-off-by:
-
- 28 Mar, 2017 1 commit
-
-
Jaegeuk Kim authored
This patch fixes missing increased max cost caused by a patch that we increased cose of data segments in greedy algorithm. Cc: <stable@vger.kernel.org> # v4.10+ Fixes: b9cd2061 "f2fs: node segment is prior to data segment selected victim" Signed-off-by:
-
- 25 Mar, 2017 1 commit
-
-
Yunlei He authored
This patch allow write data to normal file when writting new checkpoint. We relax three limitations for write_begin path: 1. data allocation 2. node allocation 3. variables in checkpoint Signed-off-by:
-
- 24 Mar, 2017 7 commits
-
-
Chao Yu authored
In this patch, we change xattr block disk layout as below: Before: xattr node block layout +---------------------------------------------+---------------+-------------+ | node block xattr entries | reserved | node footer | | 4068 Bytes | 4 Bytes | 24 Bytes | In memory layout +--------------------+---------------------------------+--------------------+ | inline xattr | node block xattr entries | reserved | | 200 Bytes | 4068 Bytes | 4 Bytes | After: xattr node block layout +-------------------------------------------------------------+-------------+ | node block xattr entries | node footer | | 4072 Bytes | 24 Bytes | In memory layout +--------------------+---------------------------------+--------------------+ | inline xattr | node block xattr entries | reserved | | 200 Bytes | 4072 Bytes | 4 Bytes | With this change, we don't need to reserve additional space in node block, just keep reserved space in logical in-memory layout. So that it would help to enlarge valid free space of xattr node block. As tested, generic/026 shows max stored xattr entires number increases from 531 to 532 when inline_xattr option is enabled. Signed-off-by:
-
Chao Yu authored
1. don't allocate redundant memory in read_all_xattrs. 2. introduce RESERVED_XATTR_SIZE for cleanup. Signed-off-by:
Kinglong Mee <kinglongmee@gmail.com> Signed-off-by:
-
Chao Yu authored
Don't track volatile file in dirty inode list, otherwise with data_flush option, background thread will entry into endless loop for flushing journal file's pages. Signed-off-by:
-
Chao Yu authored
This patch adds to show the max number of volatile operations which are conducting concurrently. Signed-off-by:
-
Chao Yu authored
In below concurrent case, allocated nid can be loaded into free nid cache and be allocated again. Thread A Thread B - f2fs_create - f2fs_new_inode - alloc_nid - __insert_nid_to_list(ALLOC_NID_LIST) - f2fs_balance_fs_bg - build_free_nids - __build_free_nids - scan_nat_page - add_free_nid - __lookup_nat_cache - f2fs_add_link - init_inode_metadata - new_inode_page - new_node_page - set_node_addr - alloc_nid_done - __remove_nid_from_list(ALLOC_NID_LIST) - __insert_nid_to_list(FREE_NID_LIST) This patch makes nat cache lookup and free nid list operation being atomical to avoid this race condition. Signed-off-by: -
Yunlei He authored
Use set_page_private marcro instead of operte page struct directly Signed-off-by:
-
Chao Yu authored
When doing garbage collection, we try to record segment offset which locates at next one of last victim, using it as the start offset in next searching. But in some corner cases, recorded offset may cross the end of main segment area, it will cause incorrectly searching in dirty_segmap bitmap. This patch adds modular operation to avoid this issue. Reported-by:
-
- 22 Mar, 2017 6 commits
-
-
Kinglong Mee authored
Signed-off-by:
-
Kinglong Mee authored
The memory size of f2fs_stat_info also should be calculated. Signed-off-by:
-
Kinglong Mee authored
After filemap_write_and_wait_range fail, the FI_ATOMIC_FILE flags is removed, so that f2fs should not increase the stat of atomic_write. Signed-off-by:
-
Chao Yu authored
Signed-off-by:
-
Kinglong Mee authored
The crc_offset towards or beyond the end of block is wrong, sanity check it. Signed-off-by:
-
Kinglong Mee authored
As discuss with Jaegeuk and Chao, "Once checkpoint is done, f2fs doesn't need to update there-in filename at all." The disk-level filename is used only one case, 1. create a file A under a dir 2. sync A 3. godown 4. umount 5. mount (roll_forward) Only the rename/cross_rename changes the filename, if it happens, a. between step 1 and 2, the sync A will caused checkpoint, so that, the roll_forward at step 5 never happens. b. after step 2, the roll_forward happens, file A will roll forward to the result as after step 1. So that, any updating the disk filename is useless, just cleanup it. Signed-off-by:
-
