Disable FOP_DONTCACHE for now due to bugs

This is kind of last-minute, but Al Viro reported that the new
FOP_DONTCACHE flag causes memory corruption due to use-after-free
issues.

This was triggered by commit 974c5e6 ("xfs: flag as supporting
FOP_DONTCACHE"), but that is not the underlying bug - it is just the
first user of the flag.

Vlastimil Babka suspects the underlying problem stems from the
folio_end_writeback() logic introduced in commit fb7d3bc
("mm/filemap: drop streaming/uncached pages when writeback completes").

The most straightforward fix would be to just revert the commit that
exposed this, but Matthew Wilcox points out that other filesystems are
also starting to enable the FOP_DONTCACHE logic, so this instead
disables that bit globally for now.

The fix will hopefully end up being trivial and we can just re-enable
this logic after more testing, but until such a time we'll have to
disable the new FOP_DONTCACHE flag.

Reported-by: Al Viro <[email protected]>
Link: https://lore.kernel.org/all/20250525083209.GS2023217@ZenIV/
Triggered-by: 974c5e6 ("xfs: flag as supporting FOP_DONTCACHE")
Cc: Vlastimil Babka <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Jan Kara <[email protected]>
Cc: Jens Axboe <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Darrick J. Wong <[email protected]>
Cc: Christian Brauner <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Author: torvalds

include/linux/fs.h

@@ -2186,7 +2186,7 @@ struct file_operations {
 /* Supports asynchronous lock callbacks */
 #define FOP_ASYNC_LOCK		((__force fop_flags_t)(1 << 6))
 /* File system supports uncached read/write buffered IO */
-#define FOP_DONTCACHE		((__force fop_flags_t)(1 << 7))
+#define FOP_DONTCACHE		0 /* ((__force fop_flags_t)(1 << 7)) */
 
 /* Wrap a directory iterator that needs exclusive inode access */
 int wrap_directory_iterator(struct file *, struct dir_context *,