Am 07.04.2020 um 08:52 hat Vladimir Sementsov-Ogievskiy geschrieben:
06.04.2020 20:14, Kevin Wolf wrote:
Waiting in blk_wait_while_drained() while blk->in_flight is increased
for the current request is wrong because it will cause the drain
operation to deadlock.
This patch makes sure that blk_wait_while_drained() is called with
blk->in_flight increased exactly once for the current request, and that
it temporarily decreases the counter while it waits.
Fixes: cf3129323f900ef5ddbccbe86e4fa801e88c566e
Signed-off-by: Kevin Wolf <address@hidden>
---
block/block-backend.c | 17 +++++------------
1 file changed, 5 insertions(+), 12 deletions(-)
diff --git a/block/block-backend.c b/block/block-backend.c
index d330e08b05..f621435f0b 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -1140,10 +1140,15 @@ static int blk_check_byte_request(BlockBackend *blk,
int64_t offset,
return 0;
}
+/* To be called between exactly one pair of blk_inc/dec_in_flight() */
static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
{
+ assert(blk->in_flight > 0);
Hmm. You promise to make sure that in_flight increased exactly once.
Shouldn't it be assert(blk->in_flight == 1) ?
Exactly once for this specific request, but if you have multiple
requests in flight, blk->in_flight will be the sum of all requests.
Just asserting > 0 should still catch potential bugs because you won't
always have multiple requests in flight.
+
if (blk->quiesce_counter && !blk->disable_request_queuing) {
+ blk_dec_in_flight(blk);
qemu_co_queue_wait(&blk->queued_requests, NULL);
+ blk_inc_in_flight(blk);
}
}
@@ -1416,12 +1421,6 @@ static void blk_aio_read_entry(void *opaque)
BlkRwCo *rwco = &acb->rwco;
QEMUIOVector *qiov = rwco->iobuf;
- if (rwco->blk->quiesce_counter) {
- blk_dec_in_flight(rwco->blk);
- blk_wait_while_drained(rwco->blk);
- blk_inc_in_flight(rwco->blk);
- }
Hm, you drop it as it's called from blk_do_preadv too. I think it
worth mentioning in commit message still.
Okay, I can add a sentence like "The blk_wait_while_drained() call in
blk_aio_read/write_entry is redundant with the one in blk_co_*(), so
drop it."