f2fs: compress: don't overwrite populated compress cache entry

f2fs_cache_compressed_page() looks up the entry of blkaddr w/
f2fs_find_cache(), and then grabs it w/ f2fs_grab_cache() and
F2FS_CACHE_LOCK_CREATE. If another reader populates the entry in
between, f2fs_grab_cache() returns that uptodate entry, and we will
overwrite its ino and data.

Before the compress cache was converted to the generic cache,
add_to_page_cache_lru() failed w/ -EEXIST in this case, and the
existing page was left untouched.

Normally, the racer is a reader of the same block of the same inode,
so both copies are the same, let's just back off, as
f2fs_ra_meta_caches() does.

A block address can only be owned by one inode at a time, the block
can not be freed and reused by another inode during population, since
the reader holds the locks of the cluster's pagecache pages until
decompression completes, and truncation must lock these pages before
freeing blocks. So if the uptodate entry was cached by another inode,
it's a bug, and one of the copies must be stale, let's report the bug,
and drop the entry, then readers will load the block from disk.

Cc: stable@kernel.org
Fixes: 6a78d0488116 ("f2fs: cache: use compress cache")
Signed-off-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

0 files changed, 0 insertions, 0 deletions