svcrdma: Fix race between svc_rdma_recvfrom thread and the dto_tasklet
Subsystems (2)
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 11 lines · Tom Tucker
RDMA_READcompletions are kept on a separate queue from the general I/O request queue. Since a separate lock is used to protect theRDMA_READcompletion queue, a race exists between thedto_taskletand thesvc_rdma_recvfromthread where thedto_taskletsets theXPT_DATAbit and adds I/O to the read-completion queue. Concurrently, the recvfrom thread checks the generic queue, finds it empty and resets theXPT_DATAbit. A subsequentsvc_xprt_enqueuewill fail to enqueue the transport for I/O and cause the transport to "stall".The fix is to protect both lists with the same lock and set the
XPT_DATAbit with this lock held.
Thanks. Assuming this race has existed from the start (so it's not a recent regression), and the consequences are no worse than a stall, I'm inclined to queue this up for 2.6.32--but if I'm wrong about either of those, we could try for 2.6.31 and/or stable.
--b.
Quoted 74 lines
Signed-off-by Tom Tucker <[email protected]>--- include/linux/sunrpc/svc_rdma.h | 1 - net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 8 ++++---- net/sunrpc/xprtrdma/svc_rdma_transport.c | 5 ++--- 3 files changed, 6 insertions(+), 8 deletions(-)include/linux/sunrpc/svc_rdma.h +0
-1@@ -143,7 +143,6 @@ struct svcxprt_rdma {unsigned long sc_flags;struct list_head sc_dto_q;/* DTO tasklet I/O pending Q */ struct list_head sc_read_complete_q; -spinlock_t sc_read_complete_lock;struct work_struct sc_work; };/* sc_flags */ net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +4
-4@@ -443,18 +443,18 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)dprintk ("svcrdma: rqstp=%p\n" , rqstp); -spin_lock_bh (&rdma_xprt->sc_read_complete_lock); +spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_read_complete_q)) { ctxt =list_entry (rdma_xprt->sc_read_complete_q.next, struct svc_rdma_op_ctxt, dto_q);list_del_init (&ctxt->dto_q); } -spin_unlock_bh (&rdma_xprt->sc_read_complete_lock); -if (ctxt) +if (ctxt) { +spin_unlock_bh (&rdma_xprt->sc_rq_dto_lock);return rdma_read_complete (rqstp, ctxt); + } -spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_rq_dto_q)) { ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next, struct svc_rdma_op_ctxt,net/sunrpc/xprtrdma/svc_rdma_transport.c +2
-3@@ -359,11 +359,11 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt)if (test_bit (RDMACTXT_F_LAST_CTXT , &ctxt->flags)) {struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;BUG_ON (!read_hdr); +spin_lock_bh (&xprt->sc_rq_dto_lock);set_bit (XPT_DATA , &xprt->sc_xprt.xpt_flags); -spin_lock_bh (&xprt->sc_read_complete_lock);list_add_tail (&read_hdr->dto_q, &xprt->sc_read_complete_q); -spin_unlock_bh (&xprt->sc_read_complete_lock); +spin_unlock_bh (&xprt->sc_rq_dto_lock);svc_xprt_enqueue (&xprt->sc_xprt); }svc_rdma_put_context (ctxt,0 );@@ -428,7 +428,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,init_waitqueue_head (&cma_xprt->sc_send_wait);spin_lock_init (&cma_xprt->sc_lock); -spin_lock_init (&cma_xprt->sc_read_complete_lock);spin_lock_init (&cma_xprt->sc_rq_dto_lock); cma_xprt->sc_ord = svcrdma_ord;
Quoted 10 lines · bfields
Quoted 9 lines · Tom Tucker
RDMA_READcompletions are kept on a separate queue from the general I/O request queue. Since a separate lock is used to protect theRDMA_READcompletion queue, a race exists between thedto_taskletand thesvc_rdma_recvfromthread where thedto_taskletsets theXPT_DATAbit and adds I/O to the read-completion queue. Concurrently, the recvfrom thread checks the generic queue, finds it empty and resets theXPT_DATAbit. A subsequentsvc_xprt_enqueuewill fail to enqueue the transport for I/O and cause the transport to "stall".
Quoted 7 lines
Quoted 2 lines
The fix is to protect both lists with the same lock and set the
XPT_DATAbit with this lock held.Thanks. Assuming this race has existed from the start (so it's not a recent regression), and the consequences are no worse than a stall, I'm inclined to queue this up for 2.6.32--but if I'm wrong about either of those, we could try for 2.6.31 and/or stable.
Erp, cripes, sorry--only just noticed all this mail seems to have been delayed a year! OK....--b.
Quoted 77 lines
--b.
Quoted 74 lines
Signed-off-by Tom Tucker <[email protected]>--- include/linux/sunrpc/svc_rdma.h | 1 - net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 8 ++++---- net/sunrpc/xprtrdma/svc_rdma_transport.c | 5 ++--- 3 files changed, 6 insertions(+), 8 deletions(-)include/linux/sunrpc/svc_rdma.h +0
-1@@ -143,7 +143,6 @@ struct svcxprt_rdma {unsigned long sc_flags;struct list_head sc_dto_q;/* DTO tasklet I/O pending Q */ struct list_head sc_read_complete_q; -spinlock_t sc_read_complete_lock;struct work_struct sc_work; };/* sc_flags */ net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +4
-4@@ -443,18 +443,18 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)dprintk ("svcrdma: rqstp=%p\n" , rqstp); -spin_lock_bh (&rdma_xprt->sc_read_complete_lock); +spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_read_complete_q)) { ctxt =list_entry (rdma_xprt->sc_read_complete_q.next, struct svc_rdma_op_ctxt, dto_q);list_del_init (&ctxt->dto_q); } -spin_unlock_bh (&rdma_xprt->sc_read_complete_lock); -if (ctxt) +if (ctxt) { +spin_unlock_bh (&rdma_xprt->sc_rq_dto_lock);return rdma_read_complete (rqstp, ctxt); + } -spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_rq_dto_q)) { ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next, struct svc_rdma_op_ctxt,net/sunrpc/xprtrdma/svc_rdma_transport.c +2
-3@@ -359,11 +359,11 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt)if (test_bit (RDMACTXT_F_LAST_CTXT , &ctxt->flags)) {struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;BUG_ON (!read_hdr); +spin_lock_bh (&xprt->sc_rq_dto_lock);set_bit (XPT_DATA , &xprt->sc_xprt.xpt_flags); -spin_lock_bh (&xprt->sc_read_complete_lock);list_add_tail (&read_hdr->dto_q, &xprt->sc_read_complete_q); -spin_unlock_bh (&xprt->sc_read_complete_lock); +spin_unlock_bh (&xprt->sc_rq_dto_lock);svc_xprt_enqueue (&xprt->sc_xprt); }svc_rdma_put_context (ctxt,0 );@@ -428,7 +428,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,init_waitqueue_head (&cma_xprt->sc_send_wait);spin_lock_init (&cma_xprt->sc_lock); -spin_lock_init (&cma_xprt->sc_read_complete_lock);spin_lock_init (&cma_xprt->sc_rq_dto_lock); cma_xprt->sc_ord = svcrdma_ord;
Please disregard this mail! I fixed a relayhost parameter in a lab machine and this queued mail was sent. My apologies.
Tom
Quoted 108 lines · J. Bruce Fields
Quoted 11 lines · bfields
Quoted 9 lines · Tom Tucker
RDMA_READcompletions are kept on a separate queue from the general I/O request queue. Since a separate lock is used to protect theRDMA_READcompletion queue, a race exists between thedto_taskletand thesvc_rdma_recvfromthread where thedto_taskletsets theXPT_DATAbit and adds I/O to the read-completion queue. Concurrently, the recvfrom thread checks the generic queue, finds it empty and resets theXPT_DATAbit. A subsequentsvc_xprt_enqueuewill fail to enqueue the transport for I/O and cause the transport to "stall".Quoted 8 lines
Quoted 3 lines
The fix is to protect both lists with the same lock and set the
XPT_DATAbit with this lock held.Thanks. Assuming this race has existed from the start (so it's not a recent regression), and the consequences are no worse than a stall, I'm inclined to queue this up for 2.6.32--but if I'm wrong about either of those, we could try for 2.6.31 and/or stable.
Erp, cripes, sorry--only just noticed all this mail seems to have been delayed a year! OK....--b.
Quoted 71 lines
--b.
Quoted 68 lines
Signed-off-by Tom Tucker <[email protected]>--- include/linux/sunrpc/svc_rdma.h | 1 - net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 8 ++++---- net/sunrpc/xprtrdma/svc_rdma_transport.c | 5 ++--- 3 files changed, 6 insertions(+), 8 deletions(-)include/linux/sunrpc/svc_rdma.h +0
-1@@ -143,7 +143,6 @@ struct svcxprt_rdma {unsigned long sc_flags;struct list_head sc_dto_q;/* DTO tasklet I/O pending Q */ struct list_head sc_read_complete_q; -spinlock_t sc_read_complete_lock;struct work_struct sc_work; };/* sc_flags */ net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +4
-4@@ -443,18 +443,18 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)dprintk ("svcrdma: rqstp=%p\n" , rqstp); -spin_lock_bh (&rdma_xprt->sc_read_complete_lock); +spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_read_complete_q)) { ctxt =list_entry (rdma_xprt->sc_read_complete_q.next, struct svc_rdma_op_ctxt, dto_q);list_del_init (&ctxt->dto_q); } -spin_unlock_bh (&rdma_xprt->sc_read_complete_lock); -if (ctxt) +if (ctxt) { +spin_unlock_bh (&rdma_xprt->sc_rq_dto_lock);return rdma_read_complete (rqstp, ctxt); + } -spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_rq_dto_q)) { ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next, struct svc_rdma_op_ctxt,net/sunrpc/xprtrdma/svc_rdma_transport.c +2
-2@@ -359,11 +359,11 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt)if (test_bit (RDMACTXT_F_LAST_CTXT , &ctxt->flags)) {struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;BUG_ON (!read_hdr); +spin_lock_bh (&xprt->sc_rq_dto_lock);set_bit (XPT_DATA , &xprt->sc_xprt.xpt_flags); -spin_lock_bh (&xprt->sc_read_complete_lock);list_add_tail (&read_hdr->dto_q, &xprt->sc_read_complete_q); -spin_unlock_bh (&xprt->sc_read_complete_lock); +spin_unlock_bh (&xprt->sc_rq_dto_lock);svc_xprt_enqueue (&xprt->sc_xprt); }svc_rdma_put_context (ctxt,0 );@@ -428,7 +428,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,init_waitqueue_head (&cma_xprt->sc_send_wait);Quoted 6 lines
net/sunrpc/xprtrdma/svc_rdma_transport.c static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, · line 433
Quoted 6 lines
net/sunrpc/xprtrdma/svc_rdma_transport.c +0
-1spin_lock_init (&cma_xprt->sc_lock); -spin_lock_init (&cma_xprt->sc_read_complete_lock);spin_lock_init (&cma_xprt->sc_rq_dto_lock); cma_xprt->sc_ord = svcrdma_ord;--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 2 lines · Tom Tucker
Please disregard this mail! I fixed a relayhost parameter in a lab machine and this queued mail was sent. My apologies.
No problem! I didn't recognize that first patch, but then got to the following series and said hey, wait a minute, we already did this....
--b.
Quoted 108 lines
Tom
Quoted 103 lines · J. Bruce Fields
Quoted 11 lines · bfields
Quoted 9 lines · Tom Tucker
RDMA_READcompletions are kept on a separate queue from the general I/O request queue. Since a separate lock is used to protect theRDMA_READcompletion queue, a race exists between thedto_taskletand thesvc_rdma_recvfromthread where thedto_taskletsets theXPT_DATAbit and adds I/O to the read-completion queue. Concurrently, the recvfrom thread checks the generic queue, finds it empty and resets theXPT_DATAbit. A subsequentsvc_xprt_enqueuewill fail to enqueue the transport for I/O and cause the transport to "stall".Quoted 8 lines
Quoted 3 lines
The fix is to protect both lists with the same lock and set the
XPT_DATAbit with this lock held.Thanks. Assuming this race has existed from the start (so it's not a recent regression), and the consequences are no worse than a stall, I'm inclined to queue this up for 2.6.32--but if I'm wrong about either of those, we could try for 2.6.31 and/or stable.
Erp, cripes, sorry--only just noticed all this mail seems to have been delayed a year! OK....--b.
Quoted 64 lines
--b.
Quoted 61 lines
Signed-off-by Tom Tucker <[email protected]>--- include/linux/sunrpc/svc_rdma.h | 1 - net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 8 ++++---- net/sunrpc/xprtrdma/svc_rdma_transport.c | 5 ++--- 3 files changed, 6 insertions(+), 8 deletions(-)include/linux/sunrpc/svc_rdma.h +0
-1@@ -143,7 +143,6 @@ struct svcxprt_rdma {unsigned long sc_flags;struct list_head sc_dto_q;/* DTO tasklet I/O pending Q */ struct list_head sc_read_complete_q; -spinlock_t sc_read_complete_lock;struct work_struct sc_work; };/* sc_flags */ net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +4
-2@@ -443,18 +443,18 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)dprintk ("svcrdma: rqstp=%p\n" , rqstp); -spin_lock_bh (&rdma_xprt->sc_read_complete_lock); +spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_read_complete_q)) { ctxt =list_entry (rdma_xprt->sc_read_complete_q.next, struct svc_rdma_op_ctxt, dto_q);list_del_init (&ctxt->dto_q); } -spin_unlock_bh (&rdma_xprt->sc_read_complete_lock); -if (ctxt) +if (ctxt) { +spin_unlock_bh (&rdma_xprt->sc_rq_dto_lock);return rdma_read_complete (rqstp, ctxt); + } -spin_lock_bh (&rdma_xprt->sc_rq_dto_lock);if (!list_empty (&rdma_xprt->sc_rq_dto_q)) { ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next, struct svc_rdma_op_ctxt,net/sunrpc/xprtrdma/svc_rdma_transport.c +2
-2@@ -359,11 +359,11 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt)if (test_bit (RDMACTXT_F_LAST_CTXT , &ctxt->flags)) {struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;BUG_ON (!read_hdr); +spin_lock_bh (&xprt->sc_rq_dto_lock);set_bit (XPT_DATA , &xprt->sc_xprt.xpt_flags); -spin_lock_bh (&xprt->sc_read_complete_lock);list_add_tail (&read_hdr->dto_q, &xprt->sc_read_complete_q); -spin_unlock_bh (&xprt->sc_read_complete_lock); +spin_unlock_bh (&xprt->sc_rq_dto_lock);svc_xprt_enqueue (&xprt->sc_xprt); }Quoted 8 lines
net/sunrpc/xprtrdma/svc_rdma_transport.c static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, · line 432
Quoted 8 lines
net/sunrpc/xprtrdma/svc_rdma_transport.c +0
-1svc_rdma_put_context (ctxt,0 );@@ -428,7 +428,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,init_waitqueue_head (&cma_xprt->sc_send_wait);spin_lock_init (&cma_xprt->sc_lock); -spin_lock_init (&cma_xprt->sc_read_complete_lock);spin_lock_init (&cma_xprt->sc_rq_dto_lock); cma_xprt->sc_ord = svcrdma_ord;--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html