xfs: I/O completion handlers must use NOFS allocations

When completing I/O requests we must not allow the memory allocator to
recurse into the filesystem, as we might deadlock on waiting for the
I/O completion otherwise.  The only thing currently allocating normal
GFP_KERNEL memory is the allocation of the transaction structure for
the unwritten extent conversion.  Add a memflags argument to
_xfs_trans_alloc to allow controlling the allocator behaviour.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Thomas Neumann <tneumann@users.sourceforge.net>
Tested-by: Thomas Neumann <tneumann@users.sourceforge.net>
Reviewed-by: Alex Elder <aelder@sgi.com>
Signed-off-by: Alex Elder <aelder@sgi.com>

1 files changed, 8 insertions, 1 deletions

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 67ae5555a30a..7294abce6ef2 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -860,8 +860,15 @@ xfs_iomap_write_unwritten(
                  * set up a transaction to convert the range of extents
                  * from unwritten to real. Do allocations in a loop until
                  * we have covered the range passed in.
+                 *
+                 * Note that we open code the transaction allocation here
+                 * to pass KM_NOFS--we can't risk to recursing back into
+                 * the filesystem here as we might be asked to write out
+                 * the same inode that we complete here and might deadlock
+                 * on the iolock.
                  */
-                tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
+                xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
+                tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
                 tp->t_flags |= XFS_TRANS_RESERVE;
                 error = xfs_trans_reserve(tp, resblks,
                                 XFS_WRITE_LOG_RES(mp), 0,