From: "Darrick J. Wong" djwong@kernel.org
[ Upstream commit 13ae04d8d45227c2ba51e188daf9fc13d08a1b12 ]
While stress-testing online repair of btrees, I noticed periodic assertion failures from the buffer cache about buffers with incorrect DELWRI_Q state. Looking further, I observed this race between the AIL trying to write out a btree block and repair zapping a btree block after the fact:
AIL: Repair0:
pin buffer X delwri_queue: set DELWRI_Q add to delwri list
stale buf X: clear DELWRI_Q does not clear b_list free space X commit
delwri_submit # oops
Worse yet, I discovered that running the same repair over and over in a tight loop can result in a second race that cause data integrity problems with the repair:
AIL: Repair0: Repair1:
pin buffer X delwri_queue: set DELWRI_Q add to delwri list
stale buf X: clear DELWRI_Q does not clear b_list free space X commit
find free space X get buffer rewrite buffer delwri_queue: set DELWRI_Q already on a list, do not add commit
BAD: committed tree root before all blocks written
delwri_submit # too late now
I traced this to my own misunderstanding of how the delwri lists work, particularly with regards to the AIL's buffer list. If a buffer is logged and committed, the buffer can end up on that AIL buffer list. If btree repairs are run twice in rapid succession, it's possible that the first repair will invalidate the buffer and free it before the next time the AIL wakes up. Marking the buffer stale clears DELWRI_Q from the buffer state without removing the buffer from its delwri list. The buffer doesn't know which list it's on, so it cannot know which lock to take to protect the list for a removal.
If the second repair allocates the same block, it will then recycle the buffer to start writing the new btree block. Meanwhile, if the AIL wakes up and walks the buffer list, it will ignore the buffer because it can't lock it, and go back to sleep.
When the second repair calls delwri_queue to put the buffer on the list of buffers to write before committing the new btree, it will set DELWRI_Q again, but since the buffer hasn't been removed from the AIL's buffer list, it won't add it to the bulkload buffer's list.
This is incorrect, because the bulkload caller relies on delwri_submit to ensure that all the buffers have been sent to disk /before/ committing the new btree root pointer. This ordering requirement is required for data consistency.
Worse, the AIL won't clear DELWRI_Q from the buffer when it does finally drop it, so the next thread to walk through the btree will trip over a debug assertion on that flag.
To fix this, create a new function that waits for the buffer to be removed from any other delwri lists before adding the buffer to the caller's delwri list. By waiting for the buffer to clear both the delwri list and any potential delwri wait list, we can be sure that repair will initiate writes of all buffers and report all write errors back to userspace instead of committing the new structure.
Signed-off-by: Darrick J. Wong djwong@kernel.org Reviewed-by: Christoph Hellwig hch@lst.de Signed-off-by: Leah Rumancik leah.rumancik@gmail.com Acked-by: "Darrick J. Wong" djwong@kernel.org --- fs/xfs/libxfs/xfs_btree_staging.c | 4 +-- fs/xfs/xfs_buf.c | 44 ++++++++++++++++++++++++++++--- fs/xfs/xfs_buf.h | 1 + 3 files changed, 42 insertions(+), 7 deletions(-)
diff --git a/fs/xfs/libxfs/xfs_btree_staging.c b/fs/xfs/libxfs/xfs_btree_staging.c index dd75e208b543..29e3f8ccb185 100644 --- a/fs/xfs/libxfs/xfs_btree_staging.c +++ b/fs/xfs/libxfs/xfs_btree_staging.c @@ -340,13 +340,11 @@ xfs_btree_bload_drop_buf( struct xfs_buf **bpp) { if (*bpp == NULL) return;
- if (!xfs_buf_delwri_queue(*bpp, buffers_list)) - ASSERT(0); - + xfs_buf_delwri_queue_here(*bpp, buffers_list); xfs_buf_relse(*bpp); *bpp = NULL; }
/* diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c index 54c774af6e1c..257945cdf63b 100644 --- a/fs/xfs/xfs_buf.c +++ b/fs/xfs/xfs_buf.c @@ -2038,28 +2038,36 @@ xfs_alloc_buftarg( error_free: kmem_free(btp); return NULL; }
+static inline void +xfs_buf_list_del( + struct xfs_buf *bp) +{ + list_del_init(&bp->b_list); + wake_up_var(&bp->b_list); +} + /* * Cancel a delayed write list. * * Remove each buffer from the list, clear the delwri queue flag and drop the * associated buffer reference. */ void xfs_buf_delwri_cancel( struct list_head *list) { struct xfs_buf *bp;
while (!list_empty(list)) { bp = list_first_entry(list, struct xfs_buf, b_list);
xfs_buf_lock(bp); bp->b_flags &= ~_XBF_DELWRI_Q; - list_del_init(&bp->b_list); + xfs_buf_list_del(bp); xfs_buf_relse(bp); } }
/* @@ -2108,10 +2116,38 @@ xfs_buf_delwri_queue( }
return true; }
+/* + * Queue a buffer to this delwri list as part of a data integrity operation. + * If the buffer is on any other delwri list, we'll wait for that to clear + * so that the caller can submit the buffer for IO and wait for the result. + * Callers must ensure the buffer is not already on the list. + */ +void +xfs_buf_delwri_queue_here( + struct xfs_buf *bp, + struct list_head *buffer_list) +{ + /* + * We need this buffer to end up on the /caller's/ delwri list, not any + * old list. This can happen if the buffer is marked stale (which + * clears DELWRI_Q) after the AIL queues the buffer to its list but + * before the AIL has a chance to submit the list. + */ + while (!list_empty(&bp->b_list)) { + xfs_buf_unlock(bp); + wait_var_event(&bp->b_list, list_empty(&bp->b_list)); + xfs_buf_lock(bp); + } + + ASSERT(!(bp->b_flags & _XBF_DELWRI_Q)); + + xfs_buf_delwri_queue(bp, buffer_list); +} + /* * Compare function is more complex than it needs to be because * the return value is only 32 bits and we are doing comparisons * on 64 bit values */ @@ -2170,31 +2206,31 @@ xfs_buf_delwri_submit_buffers( * marked it stale in the meantime. In that case only the * _XBF_DELWRI_Q flag got cleared, and we have to drop the * reference and remove it from the list here. */ if (!(bp->b_flags & _XBF_DELWRI_Q)) { - list_del_init(&bp->b_list); + xfs_buf_list_del(bp); xfs_buf_relse(bp); continue; }
trace_xfs_buf_delwri_split(bp, _RET_IP_);
/* * If we have a wait list, each buffer (and associated delwri * queue reference) transfers to it and is submitted * synchronously. Otherwise, drop the buffer from the delwri * queue and submit async. */ bp->b_flags &= ~_XBF_DELWRI_Q; bp->b_flags |= XBF_WRITE; if (wait_list) { bp->b_flags &= ~XBF_ASYNC; list_move_tail(&bp->b_list, wait_list); } else { bp->b_flags |= XBF_ASYNC; - list_del_init(&bp->b_list); + xfs_buf_list_del(bp); } __xfs_buf_submit(bp, false); } blk_finish_plug(&plug);
@@ -2244,11 +2280,11 @@ xfs_buf_delwri_submit(
/* Wait for IO to complete. */ while (!list_empty(&wait_list)) { bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
- list_del_init(&bp->b_list); + xfs_buf_list_del(bp);
/* * Wait on the locked buffer, check for errors and unlock and * release the delwri queue reference. */ diff --git a/fs/xfs/xfs_buf.h b/fs/xfs/xfs_buf.h index 549c60942208..6cf0332ba62c 100644 --- a/fs/xfs/xfs_buf.h +++ b/fs/xfs/xfs_buf.h @@ -303,10 +303,11 @@ extern void *xfs_buf_offset(struct xfs_buf *, size_t); extern void xfs_buf_stale(struct xfs_buf *bp);
/* Delayed Write Buffer Routines */ extern void xfs_buf_delwri_cancel(struct list_head *); extern bool xfs_buf_delwri_queue(struct xfs_buf *, struct list_head *); +void xfs_buf_delwri_queue_here(struct xfs_buf *bp, struct list_head *bl); extern int xfs_buf_delwri_submit(struct list_head *); extern int xfs_buf_delwri_submit_nowait(struct list_head *); extern int xfs_buf_delwri_pushbuf(struct xfs_buf *, struct list_head *);
static inline xfs_daddr_t xfs_buf_daddr(struct xfs_buf *bp)