block: add assert_spin_locked() to ensure lock is held
PATCH 1/1 block: add assert_spin_locked() to ensure lock is held
Use assert_spin_locked() macro in order to ensure that queue_lock must be held prior to calling some EXPORTed functions.
--- block/blk-core.c | 12 ++++++++++++ 1 files changed, 12 insertions(+), 0 deletions(-)
block/blk-core.c +12 -0
@@ -975,6 +975,8 @@ EXPORT_SYMBOL(blk_make_request);*/ void blk_requeue_request (struct request_queue *q,struct request *rq) { +assert_spin_locked (q->queue_lock); +blk_delete_timer (rq);blk_clear_rq_complete (rq);trace_block_rq_requeue (q, rq);
@@ -1098,6 +1100,8 @@ EXPORT_SYMBOL_GPL(part_round_stats);*/ void __blk_put_request (struct request_queue *q,struct request *req) { +assert_spin_locked (q->queue_lock); +if (unlikely (!q))return ;if (unlikely (--req->ref_count))
@@ -1729,6 +1733,8 @@ unsigned int blk_rq_err_bytes(const struct request *rq)unsigned int bytes =0 ;struct bio *bio; +assert_spin_locked (rq->q->queue_lock); +if (!(rq->cmd_flags &REQ_MIXED_MERGE ))return blk_rq_bytes (rq);
@@ -1811,6 +1817,8 @@ struct request *blk_peek_request(struct request_queue *q)struct request *rq;int ret; +assert_spin_locked (q->queue_lock); +while ((rq =__elv_next_request (q)) !=NULL ) {if (!(rq->cmd_flags &REQ_STARTED )) {/*
@@ -1926,6 +1934,8 @@ void blk_dequeue_request(struct request *rq)*/ void blk_start_request (struct request *req) { +assert_spin_locked (req->q->queue_lock); +blk_dequeue_request (req);/*
@@ -1959,6 +1969,8 @@ struct request *blk_fetch_request(struct request_queue *q) {struct request *rq; +assert_spin_locked (q->queue_lock); + rq =blk_peek_request (q);if (rq)blk_start_request (rq);
Signature · 1 line
1.7.2.2
Jens Axboe
maintainerQuoted 2 lines · Namhyung Kim
Use
assert_spin_locked()macro in order to ensure thatqueue_lockmust be held prior to calling some EXPORTed functions.
As far as I know, no such bug has ever occurred that I know of. So while I don't mind adding such instrumentation, there's little point to doing it when you are not seeing any usability problems in there. And all these paths (requeue less) are heavily used, so problems would appear quickly.
Locking in the block layer is still fairly trivial, we just have the one lock per queue.
Signature · 1 line
Jens Axboe
Quoted 6 lines · Jens Axboe
As far as I know, no such bug has ever occurred that I know of. So while I don't mind adding such instrumentation, there's little point to doing it when you are not seeing any usability problems in there. And all these paths (requeue less) are heavily used, so problems would appear quickly.
I didn't see any problems in there. I just thought if a code has such a restriction, generally it is better to have an explicit assertion in the code also not only in comments. Since those functions are EXPORTed ones I thought there will be a possibility of misusing them, although they could be found quickly without this facility. :-)
Signature · 2 lines
Regards, Namhyung Kim