- 30 May, 2020 1 commit
-
-
Chao Yu authored
Under heavy fsstress, we may triggle panic while issuing discard, because __check_sit_bitmap() detects that discard command may earse valid data blocks, the root cause is as below race stack described, since we removed lock when flushing quota data, quota data writeback may race with write_checkpoint(), so that it causes inconsistency in between cached discard entry and segment bitmap. - f2fs_write_checkpoint - block_operations - set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH) - f2fs_flush_sit_entries - add_discard_addrs - __set_bit_le(i, (void *)de->discard_map); - f2fs_write_data_pages - f2fs_write_single_data_page : inode is quota one, cp_rwsem won't be locked - f2fs_do_write_data_page - f2fs_allocate_data_block - f2fs_wait_discard_bio : discard entry has not been added yet. - update_sit_entry - f2fs_clear_prefree_segments - f2fs_issue_discard : add discard entry In order to fix this, this patch uses node_write to serialize f2fs_allocate_data_block and checkpoint. Fixes: 435cbab9 ("f2fs: fix quota_sync failure due to f2fs_lock_op") Signed-off-by:
Chao Yu <yuchao0@huawei.com> Signed-off-by:
Jaegeuk Kim <jaegeuk@kernel.org>
-
- 28 May, 2020 4 commits
-
-
Chao Yu authored
While compressed data writeback, we need to drop dirty pages like we did for non-compressed pages if cp stops, however it's not needed to compress any data in such case, so let's detect cp stop condition in cluster_may_compress() to avoid redundant compressing and let following f2fs_write_raw_pages() drops dirty pages correctly. Signed-off-by:
-
Chao Yu authored
We never use return value of __insert_discard_tree(), so remove it. Signed-off-by:
-
Chao Yu authored
In f2fs_lookup(), we should set @err correctly before printing it in tracepoint. Signed-off-by:
-
Daeho Jeong authored
Found a new segemnt allocation without f2fs_lock_op() in expand_inode_data(). So, when we do fallocate() for a pinned file and trigger checkpoint very frequently and simultaneously. F2FS gets stuck in the below code of do_checkpoint() forever. f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); /* Wait for all dirty meta pages to be submitted for IO */ <= if fallocate() here, f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META); <= it'll wait forever. Signed-off-by:Daeho Jeong <daehojeong@google.com> Reviewed-by:
-
- 27 May, 2020 1 commit
-
-
Chengguang Xu authored
Define f2fs_listxattr and to NULL when CONFIG_F2FS_FS_XATTR is not enabled, then we can remove many ugly ifdef macros in the code. Signed-off-by:
Chengguang Xu <cgxu519@mykernel.net> Reviewed-by:
-
- 25 May, 2020 1 commit
-
-
Jaegeuk Kim authored
Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages in an inifinite loop. Let's drop dirty pages at umount in that case. Signed-off-by:
-
- 18 May, 2020 2 commits
-
-
Jaegeuk Kim authored
Let's guarantee flusing dirty meta pages to avoid infinite loop. Signed-off-by: -
Jaegeuk Kim authored
When parsing the mount option, we don't have sbi->user_block_count. Should do it after getting it. Cc: <stable@vger.kernel.org> Reviewed-by:
-
- 12 May, 2020 21 commits
-
-
Chao Yu authored
During zstd compression, ZSTD_endStream() may return non-zero value because distination buffer is full, but there is still compressed data remained in intermediate buffer, it means that zstd algorithm can not save at last one block space, let's just writeback raw data instead of compressed one, this can fix data corruption when decompressing incomplete stored compression data. Fixes: 50cfa66f ("f2fs: compress: support zstd compress algorithm") Signed-off-by:
-
Chao Yu authored
in order to account data read IOs more accurately. Signed-off-by:
-
Chao Yu authored
In error path of f2fs_read_multi_pages(), it should let last referrer release decompress io context memory, otherwise, other referrer will cause use-after-free issue. Fixes: 4c8ff709 ("f2fs: support data compression") Signed-off-by:
-
Chao Yu authored
If bio has no compressed data, we don't need to handle end_io work in workqueue, instead, it should just let interrupter handle it directly to speed up IO response. Signed-off-by:
-
Colin Ian King authored
The variable err is being assigned with a value that is never read and it is being updated later with a new value. The initialization is redundant and can be removed. Addresses-Coverity: ("Unused value") Signed-off-by:Colin Ian King <colin.king@canonical.com> Reviewed-by:
-
Jaegeuk Kim authored
Sahitya raised an issue: - prevent meta updates while checkpoint is in progress allocate_segment_for_resize() can cause metapage updates if it requires to change the current node/data segments for resizing. Stop these meta updates when there is a checkpoint already in progress to prevent inconsistent CP data. Signed-off-by:
Sahitya Tummala <stummala@codeaurora.org> Reviewed-by:
-
Chao Yu authored
.i_cluster_size should be power of 2, so we can use round_up() instead of roundup() to enhance the calculation. Signed-off-by:
-
Chao Yu authored
This patch introduces a new ioctl to rollback all compress inode status: - add reserved blocks in dnode blocks - increase i_compr_blocks, i_blocks, total_valid_block_count - remove immutable flag Then compress inode can be restored to support overwrite functionality again. Signee-off-by:
-
Sayali Lokhande authored
There could be a scenario where f2fs_sync_node_pages gets called during checkpoint, which in turn tries to flush inline data and calls iput(). This results in deadlock as iput() tries to hold cp_rwsem, which is already held at the beginning by checkpoint->block_operations(). Call stack : Thread A Thread B f2fs_write_checkpoint() - block_operations(sbi) - f2fs_lock_all(sbi); - down_write(&sbi->cp_rwsem); - open() - igrab() - write() write inline data - unlink() - f2fs_sync_node_pages() - if (is_inline_node(page)) - flush_inline_data() - ilookup() page = f2fs_pagecache_get_page() if (!page) goto iput_out; iput_out: -close() -iput() iput(inode); - f2fs_evict_inode() - f2fs_truncate_blocks() - f2fs_lock_op() - down_read(&sbi->cp_rwsem); Fixes: 2049d4fc ("f2fs: avoid multiple node page writes due to inline_data") Signed-off-by:Sayali Lokhande <sayalil@codeaurora.org> Signed-off-by:
-
Konstantin Khlebnikov authored
This reserved space isn't committed yet but cannot be used for allocations. For userspace it has no difference from used space. See the same fix in ext4 commit f06925c7 ("ext4: report delalloc reserve as non-free in statfs for project quota"). Fixes: ddc34e32 ("f2fs: introduce f2fs_statfs_project") Signed-off-by:
Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Reviewed-by:
-
YueHaibing authored
update_sit_info should be f2fs_update_sit_info, otherwise build fails while no CONFIG_F2FS_STAT_FS. Fixes: fc7100ea ("f2fs: Add f2fs stats to sysfs") Signed-off-by:
YueHaibing <yuehaibing@huawei.com> Reviewed-by:
-
Chao Yu authored
Commonly, in order to handle lz4 worst compress case, caller should allocate buffer with size of LZ4_compressBound(inputsize) for target compressed data storing, however in this case, if caller didn't allocate enough space, lz4 compressor still can handle output buffer budget properly, and end up compressing when left space in output buffer is not enough. So we don't have to allocate buffer with size for worst case, then we can avoid 2 * 4KB size intermediate buffer allocation when log_cluster_size is 2, and avoid unnecessary compressing work of compressor if we can not save at least 4KB space. Suggested-by:
-
Jaegeuk Kim authored
blk_plugging doesn't seem to give any benefit. Signed-off-by: -
Chao Yu authored
There are still reserved blocks on compressed inode, this patch introduce a new ioctl to help release reserved blocks back to filesystem, so that userspace can reuse those freed space. ---- Daeho fixed a bug like below. Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged. We have to execute releasing compress blocks only one time for a file without being interfered by writepages path. --- Signed-off-by:
-
Chao Yu authored
In f2fs_try_to_free_nids(), .nid_list_lock spinlock critical region will increase as expected shrink number increase, to avoid spining other CPUs for long time, we change to release nid caches with small batch each time under .nid_list_lock coverage. Signed-off-by:
-
Eric Biggers authored
fsync() may be called on a deleted file that's still open. So when fsync() tries to set the parent inode number when the inode has LOST_PINO and i_nlink == 1 (to avoid later checkpoints), it needs to make sure to get the parent directory via a non-deleted alias. Also remove the unnecessary igrab() and iput(), as the caller already holds a reference to the inode. Signed-off-by:
Eric Biggers <ebiggers@google.com> Signed-off-by:
-
Eric Biggers authored
Rework f2fs's handling of filenames to use a new 'struct f2fs_filename'. Similar to 'struct ext4_filename', this stores the usr_fname, disk_name, dirhash, crypto_buf, and casefolded name. Some of these names can be NULL in some cases. 'struct f2fs_filename' differs from 'struct fscrypt_name' mainly in that the casefolded name is included. For user-initiated directory operations like lookup() and create(), initialize the f2fs_filename by translating the corresponding fscrypt_name, then computing the dirhash and casefolded name if needed. This makes the dirhash and casefolded name be cached for each syscall, so we don't have to recompute them repeatedly. (Previously, f2fs computed the dirhash once per directory level, and the casefolded name once per directory block.) This improves performance. This rework also makes it much easier to correctly handle all combinations of normal, encrypted, casefolded, and encrypted+casefolded directories. (The fourth isn't supported yet but is being worked on.) The only other cases where an f2fs_filename gets initialized are for two filesystem-internal operations: (1) when converting an inline directory to a regular one, we grab the needed disk_name and hash from an existing f2fs_dir_entry; and (2) when roll-forward recovering a new dentry, we grab the needed disk_name from f2fs_inode::i_name and compute the hash. Signed-off-by:
-
Eric Biggers authored
Sharing f2fs_ci_compare() between comparing cached dentries (f2fs_d_compare()) and comparing on-disk dentries (f2fs_match_name()) doesn't work as well as intended, as these actions fundamentally differ in several ways (e.g. whether the task may sleep, whether the directory is stable, whether the casefolded name was precomputed, whether the dentry will need to be decrypted once we allow casefold+encrypt, etc.) Just make f2fs_d_compare() implement what it needs directly, and rework f2fs_ci_compare() to be specialized for f2fs_match_name(). Signed-off-by:
-
Eric Biggers authored
We need to call fscrypt_free_filename() to free the memory allocated by fscrypt_setup_filename(). Fixes: b06af2af ("f2fs: convert inline_dir early before starting rename") Cc: <stable@vger.kernel.org> # v5.6+ Signed-off-by:
-
Chao Yu authored
LZO-RLE extension (run length encoding) was introduced to improve performance of LZO algorithm in scenario of data contains many zeros, zram has changed to use this extended algorithm by default, this patch adds to support this algorithm extension, to enable this extension, it needs to enable F2FS_FS_LZO and F2FS_FS_LZORLE config, and specifies "compress_algorithm=lzo-rle" mountoption. Signed-off-by:
-
Chao Yu authored
If compression feature is on, in scenario of no enough free memory, page refault ratio is higher than before, the root cause is: - {,de}compression flow needs to allocate intermediate pages to store compressed data in cluster, so during their allocation, vm may reclaim mmaped pages. - if above reclaimed pages belong to compressed cluster, during its refault, it may cause more intermediate pages allocation, result in reclaiming more mmaped pages. So this patch introduces a mempool for intermediate page allocation, in order to avoid high refault ratio, by default, number of preallocated page in pool is 512, user can change the number by assigning 'num_compress_pages' parameter during module initialization. Ma Feng found warnings in the original patch and fixed like below. Fix the following sparse warning: fs/f2fs/compress.c:501:5: warning: symbol 'num_compress_pages' was not declared. Should it be static? fs/f2fs/compress.c:530:6: warning: symbol 'f2fs_compress_free_page' was not declared. Should it be static? Reported-by:Hulk Robot <hulkci@huawei.com> Signed-off-by:
Ma Feng <mafeng.ma@huawei.com> Signed-off-by:
-
- 08 May, 2020 6 commits
-
-
Chao Yu authored
to support bmap() on compressed inode: if queried block locates in non-compressed cluster, return its physical block address. Signed-off-by:
-
Chao Yu authored
Map normal/compressed cluster of compressed inode correctly, and give the right fiemap flag FIEMAP_EXTENT_ENCODED on mapped compressed extent. Signed-off-by:
-
Chao Yu authored
Supports to truncate compressed/normal cluster partially on compressed inode. Signed-off-by:
-
Chao Yu authored
due to f2fs_post_read_required() has did that. Signed-off-by:
-
Eric Biggers authored
Remove the pointless string length checks. Just use strcmp(). Signed-off-by:
-
Nishad Kamdar authored
This patch corrects the SPDX License Identifier style in header files related to F2FS File System support. For C header files Documentation/process/license-rules.rst mandates C-like comments (opposed to C source files where C++ style should be used). Changes made by using a script provided by Joe Perches here: https://lkml.org/lkml/2019/2/7/46. Suggested-by:
Joe Perches <joe@perches.com> Signed-off-by:
Nishad Kamdar <nishadkamdar@gmail.com> Reviewed-by:
-
- 24 Apr, 2020 1 commit
-
-
Jaegeuk Kim authored
f2fs_quota_sync() uses f2fs_lock_op() before flushing dirty pages, but f2fs_write_data_page() returns EAGAIN. Likewise dentry blocks, we can just bypass getting the lock, since quota blocks are also maintained by checkpoint. Reviewed-by:
-
- 17 Apr, 2020 3 commits
-
-
Chao Yu authored
Adds to support accounting read IOs from userspace/kernel. Signed-off-by:
-
Sahitya Tummala authored
When a discard_cmd needs to be split due to dpolicy->max_requests, then for the remaining length it will be either merged into another cmd or a new discard_cmd will be created. In this case, there is double accounting of dcc->undiscard_blks for the remaining len, due to which it shows incorrect value in stats. Signed-off-by:
-
Chao Yu authored
In f2fs_ra_meta_pages(), if f2fs_submit_page_bio() failed, we need to unlock page, fix it. Signed-off-by:
-
