Commit de1cbee which removed b_file_offset in favor of b_bn introduced a bug
causing xfs_buf_allocate_memory() to overestimate the number of necessary
pages. The problem is that xfs_buf_alloc() sets b_bn to -1 and thus effectively
every buffer is straddling a page boundary which causes
xfs_buf_allocate_memory() to allocate two pages and use vmalloc() for access
which is unnecessary.
Dave says xfs_buf_alloc() doesn't need to set b_bn to -1 anymore since the
buffer is inserted into the cache only after being fully initialized now.
So just make xfs_buf_alloc() fill in proper block number from the beginning.
CC: David Chinner <dchinner@xxxxxxxxxx>
Signed-off-by: Jan Kara <jack@xxxxxxx>
fs/xfs/xfs_buf.c | 16 ++--------------
1 files changed, 2 insertions(+), 14 deletions(-)
diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
index 172d3cc..a4beb42 100644
@@ -201,14 +201,7 @@ xfs_buf_alloc(
bp->b_length = numblks;
bp->b_io_length = numblks;
bp->b_flags = flags;
- * We do not set the block number here in the buffer because we have not
- * finished initialising the buffer. We insert the buffer into the cache
- * in this state, so this ensures that we are unable to do IO on a
- * buffer that hasn't been fully initialised.
- bp->b_bn = XFS_BUF_DADDR_NULL;
+ bp->b_bn = blkno;
@@ -567,11 +560,6 @@ xfs_buf_get(
if (bp != new_bp)
- * Now we have a workable buffer, fill in the block number so
- * that we can do IO on it.
- bp->b_bn = blkno;
bp->b_io_length = bp->b_length;
@@ -772,7 +760,7 @@ xfs_buf_get_uncached(
int error, i;
- bp = xfs_buf_alloc(target, 0, numblks, 0);
+ bp = xfs_buf_alloc(target, XFS_BUF_DADDR_NULL, numblks, 0);
if (unlikely(bp == NULL))