fuse: disable FOPEN_PARALLEL_DIRECT_WRITES with FUSE_DIRECT_IO_ALLOW_MMAP
authorAmir Goldstein <amir73il@gmail.com>
Sun, 3 Dec 2023 07:42:33 +0000 (09:42 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 20 Dec 2023 16:01:51 +0000 (17:01 +0100)
commit 3f29f1c336c0e8a4bec52f1e5217f88835553e5b upstream.

The new fuse init flag FUSE_DIRECT_IO_ALLOW_MMAP breaks assumptions made by
FOPEN_PARALLEL_DIRECT_WRITES and causes test generic/095 to hit
BUG_ON(fi->writectr < 0) assertions in fuse_set_nowrite():

generic/095 5s ...
  kernel BUG at fs/fuse/dir.c:1756!
...
  ? fuse_set_nowrite+0x3d/0xdd
  ? do_raw_spin_unlock+0x88/0x8f
  ? _raw_spin_unlock+0x2d/0x43
  ? fuse_range_is_writeback+0x71/0x84
  fuse_sync_writes+0xf/0x19
  fuse_direct_io+0x167/0x5bd
  fuse_direct_write_iter+0xf0/0x146

Auto disable FOPEN_PARALLEL_DIRECT_WRITES when server negotiated
FUSE_DIRECT_IO_ALLOW_MMAP.

Fixes: e78662e818f9 ("fuse: add a new fuse init flag to relax restrictions in no cache mode")
Cc: <stable@vger.kernel.org> # v6.6
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/fuse/file.c

index 89e870d..a660f1f 100644 (file)
@@ -1574,6 +1574,7 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
        ssize_t res;
        bool exclusive_lock =
                !(ff->open_flags & FOPEN_PARALLEL_DIRECT_WRITES) ||
+               get_fuse_conn(inode)->direct_io_allow_mmap ||
                iocb->ki_flags & IOCB_APPEND ||
                fuse_direct_write_extending_i_size(iocb, from);
 
@@ -1581,6 +1582,7 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
         * Take exclusive lock if
         * - Parallel direct writes are disabled - a user space decision
         * - Parallel direct writes are enabled and i_size is being extended.
+        * - Shared mmap on direct_io file is supported (FUSE_DIRECT_IO_ALLOW_MMAP).
         *   This might not be needed at all, but needs further investigation.
         */
        if (exclusive_lock)