kernel/linux.git/block/blk-flush.c, branch v6.6.132

block: fix request.queuelist usage in flush

2024-06-21T12:38:35+00:00

[ Upstream commit d0321c812d89c5910d8da8e4b10c891c6b96ff70 ] Friedrich Weber reported a kernel crash problem and bisected to commit 81ada09cc25e ("blk-flush: reuse rq queuelist in flush state machine"). The root cause is that we use "list_move_tail(&rq->queuelist, pending)" in the PREFLUSH/POSTFLUSH sequences. But rq->queuelist.next == xxx since it's popped out from plug->cached_rq in __blk_mq_alloc_requests_batch(). We don't initialize its queuelist just for this first request, although the queuelist of all later popped requests will be initialized. Fix it by changing to use "list_add_tail(&rq->queuelist, pending)" so rq->queuelist doesn't need to be initialized. It should be ok since rq can't be on any list when PREFLUSH or POSTFLUSH, has no move actually. Please note the commit 81ada09cc25e ("blk-flush: reuse rq queuelist in flush state machine") also has another requirement that no drivers would touch rq->queuelist after blk_mq_end_request() since we will reuse it to add rq to the post-flush pending list in POSTFLUSH. If this is not true, we will have to revert that commit IMHO. This updated version adds "list_del_init(&rq->queuelist)" in flush rq callback since the dm layer may submit request of a weird invalid format (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH), which causes double list_add if without this "list_del_init(&rq->queuelist)". The weird invalid format problem should be fixed in dm layer. Reported-by: Friedrich Weber Closes: https://lore.kernel.org/lkml/14b89dfb-505c-49f7-aebb-01c54451db40@proxmox.com/ Closes: https://lore.kernel.org/lkml/c9d03ff7-27c5-4ebd-b3f6-5a90d96f35ba@proxmox.com/ Fixes: 81ada09cc25e ("blk-flush: reuse rq queuelist in flush state machine") Cc: Christoph Hellwig Cc: ming.lei@redhat.com Cc: bvanassche@acm.org Tested-by: Friedrich Weber Signed-off-by: Chengming Zhou Reviewed-by: Christoph Hellwig Link: https://lore.kernel.org/r/20240608143115.972486-1-chengming.zhou@linux.dev Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

blk-flush: reuse rq queuelist in flush state machine

2023-07-17T14:18:21+00:00

Since we don't need to maintain inflight flush_data requests list anymore, we can reuse rq->queuelist for flush pending list. Note in mq_flush_data_end_io(), we need to re-initialize rq->queuelist before reusing it in the state machine when end, since the rq->rq_next also reuse it, may have corrupted rq->queuelist by the driver. This patch decrease the size of struct request by 16 bytes. Signed-off-by: Chengming Zhou Reviewed-by: Christoph Hellwig Reviewed-by: Ming Lei Link: https://lore.kernel.org/r/20230717040058.3993930-5-chengming.zhou@linux.dev Signed-off-by: Jens Axboe

blk-flush: count inflight flush_data requests

2023-07-17T14:18:21+00:00

The flush state machine use a double list to link all inflight flush_data requests, to avoid issuing separate post-flushes for these flush_data requests which shared PREFLUSH. So we can't reuse rq->queuelist, this is why we need rq->flush.list In preparation of the next patch that reuse rq->queuelist for flush state machine, we change the double linked list to unsigned long counter, which count all inflight flush_data requests. This is ok since we only need to know if there is any inflight flush_data request, so unsigned long counter is good. Signed-off-by: Chengming Zhou Reviewed-by: Ming Lei Reviewed-by: Christoph Hellwig Link: https://lore.kernel.org/r/20230717040058.3993930-4-chengming.zhou@linux.dev Signed-off-by: Jens Axboe

blk-flush: fix rq->flush.seq for post-flush requests

2023-07-17T14:18:21+00:00

If the policy == (REQ_FSEQ_DATA | REQ_FSEQ_POSTFLUSH), it means that the data sequence and post-flush sequence need to be done for this request. The rq->flush.seq should record what sequences have been done (or don't need to be done). So in this case, pre-flush doesn't need to be done, we should init rq->flush.seq to REQ_FSEQ_PREFLUSH not REQ_FSEQ_POSTFLUSH. Fixes: 615939a2ae73 ("blk-mq: defer to the normal submission path for post-flush requests") Signed-off-by: Chengming Zhou Reviewed-by: Christoph Hellwig Link: https://lore.kernel.org/r/20230717040058.3993930-3-chengming.zhou@linux.dev Signed-off-by: Jens Axboe

block: queue data commands from the flush state machine at the head

2023-07-14T14:42:58+00:00

We used to insert the data commands following a pre-flush to the head of the queue until commit 1e82fadfc6b ("blk-mq: do not do head insertions post-pre-flush commands"). Not doing this seems to cause hangs of such commands on NFS workloads when exported from file systems with SATA SSDs. I have no idea why this would starve these workloads, but doing a semantic revert of this patch (which looks quite different due to various other changes) fixes the hangs. Fixes: 1e82fadfc6b ("blk-mq: do not do head insertions post-pre-flush commands") Reported-by: Chuck Lever Signed-off-by: Christoph Hellwig Tested-by: Chuck Lever Link: https://lore.kernel.org/r/20230714143014.11879-1-hch@lst.de Signed-off-by: Jens Axboe

blk-mq: don't use the requeue list to queue flush commands

2023-05-20T01:52:42+00:00

Currently both requeues of commands that were already sent to the driver and flush commands submitted from the flush state machine share the same requeue_list struct request_queue, despite requeues doing head insertions and flushes not. Switch to using two separate lists instead. Signed-off-by: Christoph Hellwig Reviewed-by: Damien Le Moal Link: https://lore.kernel.org/r/20230519044050.107790-8-hch@lst.de Signed-off-by: Jens Axboe

blk-mq: do not do head insertions post-pre-flush commands

2023-05-20T01:52:29+00:00

blk_flush_complete_seq currently queues requests that write data after a pre-flush from the flush state machine at the head of the queue. This doesn't really make sense, as the original request bypassed all queue lists by directly diverting to blk_insert_flush from blk_mq_submit_bio. Signed-off-by: Christoph Hellwig Reviewed-by: Bart Van Assche Reviewed-by: Damien Le Moal Link: https://lore.kernel.org/r/20230519044050.107790-7-hch@lst.de Signed-off-by: Jens Axboe

blk-mq: defer to the normal submission path for post-flush requests

2023-05-20T01:52:29+00:00

Requests with the FUA bit on hardware without FUA support need a post flush before returning to the caller, but they can still be sent using the normal I/O path after initializing the flush-related fields and end I/O handler. Signed-off-by: Christoph Hellwig Reviewed-by: Bart Van Assche Link: https://lore.kernel.org/r/20230519044050.107790-6-hch@lst.de Signed-off-by: Jens Axboe

blk-mq: defer to the normal submission path for non-flush flush commands

2023-05-20T01:52:29+00:00

If blk_insert_flush decides that a command does not need to use the flush state machine, return false and let blk_mq_submit_bio handle it the normal way (including using an I/O scheduler) instead of doing a bypass insert. Signed-off-by: Christoph Hellwig Reviewed-by: Bart Van Assche Reviewed-by: Damien Le Moal Link: https://lore.kernel.org/r/20230519044050.107790-4-hch@lst.de Signed-off-by: Jens Axboe

blk-mq: reflow blk_insert_flush

2023-05-20T01:52:29+00:00

Use a switch statement to decide on the disposition of a flush request instead of multiple if statements, out of which one does checks that are more complex than required. Also warn on a malformed request early on instead of doing a BUG_ON later. Signed-off-by: Christoph Hellwig Reviewed-by: Damien Le Moal Reviewed-by: Bart Van Assche Link: https://lore.kernel.org/r/20230519044050.107790-3-hch@lst.de Signed-off-by: Jens Axboe