Hi Bernard, Thanks for the detailed explanation. Now it's clear to me. I was under impression that using read_lock(instead of read_lock_bh) in process context could cause a deadlock due to a write_lock in softirq, which is clearly not as write_lock is always acquired in process context. Jason, please consider the current patch as final. Thanks, Krishna. On Thursday, September 09/26/19, 2019 at 13:38:32 +0000, Bernard Metzler wrote: > -----"Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> wrote: ----- > > >To: "Bernard Metzler" <BMT@xxxxxxxxxxxxxx> > >From: "Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> > >Date: 09/23/2019 10:26PM > >Cc: jgg@xxxxxxxx, linux-rdma@xxxxxxxxxxxxxxx, bharat@xxxxxxxxxxx, > >nirranjan@xxxxxxxxxxx > >Subject: [EXTERNAL] Re: [PATCH for-next] RDMA/siw: fixes > >serialization issue in write_space > > > >Hi Bernard, > > > >write_space callback may not always be called from softirq/tasklet > >context, it may be called from process context. > >Hence, read_lock_bh() is safer than read_lock(). > > > >I will resubmit the patch. > > > > Hi Krishna, > > siw takes a write_lock on the sockets sk_callback_lock > only when it is in a process context, and never from > within an socket upcall. It only takes the write_lock when > changing the assignment of the socket upcalls. > write_lock is always taken using the _bh variant, > which prevents us from an otherwise potentially > concurrently running softirq to spinlock on the same > socket. > > During socket upcall, siw always only takes a read lock > (one was missing, your current patch thankfully fixes that). > If now the upcall may happen from a process context, > it would not be prevented by write_lock_bh, but it > also does not have to. It then just does what it > is supposed to do - the process spins until the > lock becomes available. That's what spinlock's > are for... > > So I don't see a reason to change read_lock()'s to > read_lock_bh()'s: All critical write lock's are > write_lock_bh(), which either prevents the softirq > upcall from running, or lets the up-calling process > context correctly spinning. > > I think your current patch is just fine. > > > I am not sure about the iscsi_tcp code change. Maybe > it is needed since it may take a write_lock from within > softirq. > > > Many thanks, > Bernard. > > > >Also, after looking at the below commit, I feel all other read_lock() > >in SIW driver should be replaced with BH variants. > > > >https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_torva > >lds_linux_commit_7cb001d4c4fa7e1cc1a55388a9544e160dddc610&d=DwIBAg&c= > >jf_iaSHvJObTbx-siA1ZOg&r=2TaYXQ0T-r8ZO1PP1alNwU_QJcRRLfmYTAgd3QCvqSc& > >m=QHDqv1KcjDv-pXFNN-vF1j8b_JkQdzRXbebBg24Mf0U&s=LQmC1X40Q1CFelQASrwaI > >25kLAzhrTdPj28m4x3sxUs&e= > > > > > > > >Shall I also change all other occurances of read_lock() to > >read_lock_bh()? > > > > > >Thanks, > >Krishna. > > > >On Monday, September 09/23/19, 2019 at 11:24:36 +0000, Bernard > >Metzler wrote: > >> -----"Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> wrote: ----- > >> > >> >To: jgg@xxxxxxxx, bmt@xxxxxxxxxxxxxx > >> >From: "Krishnamraju Eraparaju" <krishna2@xxxxxxxxxxx> > >> >Date: 09/23/2019 12:11PM > >> >Cc: linux-rdma@xxxxxxxxxxxxxxx, bharat@xxxxxxxxxxx, > >> >nirranjan@xxxxxxxxxxx, "Krishnamraju Eraparaju" > >> ><krishna2@xxxxxxxxxxx> > >> >Subject: [EXTERNAL] [PATCH for-next] RDMA/siw: fixes serialization > >> >issue in write_space > >> > > >> >In siw_qp_llp_write_space(), 'sock' members should be accessed > >> >with sk_callback_lock held, otherwise, it could race with > >> >siw_sk_restore_upcalls(). And this could cause "NULL deref" panic. > >> >Below panic is due to the NULL cep returned from sk_to_cep(sk): > >> >[14524.030863] Call Trace: > >> >[14524.030868] <IRQ> siw_qp_llp_write_space+0x11/0x40 [siw] > >> >[14524.030873] tcp_check_space+0x4c/0xf0 > >> >[14524.030877] tcp_rcv_established+0x52b/0x630 > >> >[14524.030880] tcp_v4_do_rcv+0xf4/0x1e0 > >> >[14524.030882] tcp_v4_rcv+0x9b8/0xab0 > >> >[14524.030886] ip_protocol_deliver_rcu+0x2c/0x1c0 > >> >[14524.030889] ip_local_deliver_finish+0x44/0x50 > >> >[14524.030891] ip_local_deliver+0x6b/0xf0 > >> >[14524.030893] ? ip_protocol_deliver_rcu+0x1c0/0x1c0 > >> >[14524.030896] ip_rcv+0x52/0xd0 > >> >[14524.030898] ? ip_rcv_finish_core.isra.14+0x390/0x390 > >> >[14524.030903] __netif_receive_skb_one_core+0x83/0xa0 > >> >[14524.030906] netif_receive_skb_internal+0x73/0xb0 > >> >[14524.030909] napi_gro_frags+0x1ff/0x2b0 > >> >[14524.030922] t4_ethrx_handler+0x4a7/0x740 [cxgb4] > >> >[14524.030930] process_responses+0x2c9/0x590 [cxgb4] > >> >[14524.030937] ? t4_sge_intr_msix+0x1d/0x30 [cxgb4] > >> >[14524.030941] ? handle_irq_event_percpu+0x51/0x70 > >> >[14524.030943] ? handle_irq_event+0x41/0x60 > >> >[14524.030946] ? handle_edge_irq+0x97/0x1a0 > >> >[14524.030952] napi_rx_handler+0x14/0xe0 [cxgb4] > >> >[14524.030955] net_rx_action+0x2af/0x410 > >> >[14524.030962] __do_softirq+0xda/0x2a8 > >> >[14524.030965] do_softirq_own_stack+0x2a/0x40 > >> >[14524.030967] </IRQ> > >> >[14524.030969] do_softirq+0x50/0x60 > >> >[14524.030972] __local_bh_enable_ip+0x50/0x60 > >> >[14524.030974] ip_finish_output2+0x18f/0x520 > >> >[14524.030977] ip_output+0x6e/0xf0 > >> >[14524.030979] ? __ip_finish_output+0x1f0/0x1f0 > >> >[14524.030982] __ip_queue_xmit+0x14f/0x3d0 > >> >[14524.030986] ? __slab_alloc+0x4b/0x58 > >> >[14524.030990] __tcp_transmit_skb+0x57d/0xa60 > >> >[14524.030992] tcp_write_xmit+0x23b/0xfd0 > >> >[14524.030995] __tcp_push_pending_frames+0x2e/0xf0 > >> >[14524.030998] tcp_sendmsg_locked+0x939/0xd50 > >> >[14524.031001] tcp_sendmsg+0x27/0x40 > >> >[14524.031004] sock_sendmsg+0x57/0x80 > >> >[14524.031009] siw_tx_hdt+0x894/0xb20 [siw] > >> >[14524.031015] ? find_busiest_group+0x3e/0x5b0 > >> >[14524.031019] ? common_interrupt+0xa/0xf > >> >[14524.031021] ? common_interrupt+0xa/0xf > >> >[14524.031023] ? common_interrupt+0xa/0xf > >> >[14524.031028] siw_qp_sq_process+0xf1/0xe60 [siw] > >> >[14524.031031] ? __wake_up_common_lock+0x87/0xc0 > >> >[14524.031035] siw_sq_resume+0x33/0xe0 [siw] > >> >[14524.031039] siw_run_sq+0xac/0x190 [siw] > >> >[14524.031041] ? remove_wait_queue+0x60/0x60 > >> >[14524.031045] kthread+0xf8/0x130 > >> >[14524.031049] ? siw_sq_resume+0xe0/0xe0 [siw] > >> >[14524.031051] ? kthread_bind+0x10/0x10 > >> >[14524.031053] ret_from_fork+0x35/0x40 > >> > > >> >Fixes: f29dd55b0236 (rdma/siw: queue pair methods) > >> >Signed-off-by: Krishnamraju Eraparaju <krishna2@xxxxxxxxxxx> > >> >--- > >> > drivers/infiniband/sw/siw/siw_qp.c | 15 +++++++++++---- > >> > 1 file changed, 11 insertions(+), 4 deletions(-) > >> > > >> >diff --git a/drivers/infiniband/sw/siw/siw_qp.c > >> >b/drivers/infiniband/sw/siw/siw_qp.c > >> >index 430314c8abd9..52d402f39df9 100644 > >> >--- a/drivers/infiniband/sw/siw/siw_qp.c > >> >+++ b/drivers/infiniband/sw/siw/siw_qp.c > >> >@@ -182,12 +182,19 @@ void siw_qp_llp_close(struct siw_qp *qp) > >> > */ > >> > void siw_qp_llp_write_space(struct sock *sk) > >> > { > >> >- struct siw_cep *cep = sk_to_cep(sk); > >> >+ struct siw_cep *cep; > >> > > >> >- cep->sk_write_space(sk); > >> >+ read_lock(&sk->sk_callback_lock); > >> >+ > >> >+ cep = sk_to_cep(sk); > >> >+ if (cep) { > >> >+ cep->sk_write_space(sk); > >> > > >> >- if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) > >> >- (void)siw_sq_start(cep->qp); > >> >+ if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) > >> >+ (void)siw_sq_start(cep->qp); > >> >+ } > >> >+ > >> >+ read_unlock(&sk->sk_callback_lock); > >> > } > >> > > >> > static int siw_qp_readq_init(struct siw_qp *qp, int irq_size, int > >> >orq_size) > >> >-- > >> >2.23.0.rc0 > >> > > >> > > >> > >> Hi Krishna, > >> > >> Many thanks! I completely agree. This fixes a potential race. > >> I was under the impression the socket layer itself would hold > >> the sk_callback_lock during the writespace upcall, which is > >> obviously not true. > >> > >> Reviewed-by: Bernard Metzler <bmt@xxxxxxxxxxxxxx> > >> > > > > >
