This is a resubmit of this patch, originally sent on the 19 Oct 2020. If squashfs encounters a read error when reading a block this error will be cached. When the underlying problem that caused this error is fixed, squashfs will still report the previous error, even though, a re-read would now be successful.
This patch fixes this by setting the block field of the cache entry to SQUASHFS_INVALID_BLK if an error was encountered. This prevents the cache entry lookup from simply returning the cache entry with the previous error. With this patch a mounted squashfs file system can recover from read errors. Whereas previously this would have required a full remount. Signed-off-by: Mikael Szreder <[email protected]> --- fs/squashfs/cache.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/fs/squashfs/cache.c b/fs/squashfs/cache.c index 5062326d0efb..020fded42a6b 100644 --- a/fs/squashfs/cache.c +++ b/fs/squashfs/cache.c @@ -112,8 +112,10 @@ struct squashfs_cache_entry *squashfs_cache_get(struct super_block *sb, spin_lock(&cache->lock); - if (entry->length < 0) + if (entry->length < 0) { + entry->block = SQUASHFS_INVALID_BLK; entry->error = entry->length; + } entry->pending = 0; -- 2.28.0
