Message ID | 20230522121125.2595254-9-dhowells@redhat.com (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | splice, net: Replace sendpage with sendmsg(MSG_SPLICE_PAGES), part 1 | expand |
On 22/05/2023 15:11, David Howells wrote: > do_tcp_sendpages() is now just a small wrapper around tcp_sendmsg_locked(), > so inline it, allowing do_tcp_sendpages() to be removed. This is part of > replacing ->sendpage() with a call to sendmsg() with MSG_SPLICE_PAGES set. > > Signed-off-by: David Howells <dhowells@redhat.com> > cc: Boris Pismenny <borisp@nvidia.com> > cc: John Fastabend <john.fastabend@gmail.com> > cc: Jakub Kicinski <kuba@kernel.org> > cc: "David S. Miller" <davem@davemloft.net> > cc: Eric Dumazet <edumazet@google.com> > cc: Paolo Abeni <pabeni@redhat.com> > cc: Jens Axboe <axboe@kernel.dk> > cc: Matthew Wilcox <willy@infradead.org> > cc: netdev@vger.kernel.org > --- Hi, My team spotted a new degradation in TLS TX device offload, bisected to this patch. From a quick look at the patch, it's not clear to me what's going wrong. Please let us know of any helpful information that we can provide to help in the debug. Regards, Tariq Reproduce Flow: client / server test using nginx and wrk (nothing special/custom about the apps used). client: /opt/mellanox/iproute2/sbin/ip link set dev eth3 up /opt/mellanox/iproute2/sbin/ip addr add 11.141.46.9/16 dev eth3 server: /opt/mellanox/iproute2/sbin/ip link set dev eth3 up /opt/mellanox/iproute2/sbin/ip addr add 11.141.46.10/16 dev eth3 client: /auto/sw/regression/sw_net_ver_tools/ktls/tools/x86_64/nginx_openssl_3_0_0 -p /usr/bin/drivertest_rpms/ktls/nginx/ /opt/mellanox/iproute2/sbin/ss -i src [11.141.46.9] server: /auto/sw/regression/sw_net_ver_tools/ktls/tools/x86_64/wrk_openssl_3_0_0 -b11.141.46.10 -t4 -c874 -d14 --timeout 5s https://11.141.46.9:20443/256000b.img client: dmesg /auto/sw/regression/sw_net_ver_tools/ktls/tools/x86_64/nginx_openssl_3_0_0 -p /usr/bin/drivertest_rpms/ktls/nginx/ -s stop [root@c-141-46-1-009 ~]# dmesg ------------[ cut here ]------------ WARNING: CPU: 1 PID: 977 at net/core/skbuff.c:6957 skb_splice_from_iter+0x102/0x300 Modules linked in: rpcrdma rdma_ucm ib_iser libiscsi scsi_transport_iscsi ib_umad rdma_cm ib_ipoib iw_cm ib_cm mlx5_ib ib_uverbs ib_core xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter rpcsec_gss_krb5 auth_rpcgss oid_registry overlay mlx5_core zram zsmalloc fuse CPU: 1 PID: 977 Comm: nginx_openssl_3 Not tainted 6.4.0-rc3_for_upstream_min_debug_2023_06_01_23_04 #1 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014 RIP: 0010:skb_splice_from_iter+0x102/0x300 Code: ef 48 8b 55 08 f6 c2 01 0f 85 54 01 00 00 8b 0d 98 cf 5f 01 48 89 ea 85 c9 0f 8f 4c 01 00 00 48 8b 12 80 e6 02 74 48 49 89 dd <0f> 0b 48 c7 c1 fb ff ff ff 45 01 65 70 45 01 65 74 45 01 a5 d0 00 RSP: 0018:ffff8881045abaa0 EFLAGS: 00010246 RAX: 0000000000000000 RBX: ffff88814370fe00 RCX: 0000000000000000 RDX: 0000000000000000 RSI: ffffea00051123c0 RDI: ffff88814370fe00 RBP: ffffea0005112400 R08: 0000000000000011 R09: 0000000000003ffd R10: 0000000000003ffd R11: 0000000000000008 R12: 0000000000002e6e R13: ffff88814370fe00 R14: ffff8881045abae8 R15: 000000000000118f FS: 00007f6e23043740(0000) GS:ffff88852c880000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000009c6c00 CR3: 000000013b791001 CR4: 0000000000370ea0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: ? kmalloc_reserve+0x86/0xe0 tcp_sendmsg_locked+0x33e/0xd40 tls_push_sg+0xdd/0x230 tls_push_data+0x673/0x920 tls_device_sendmsg+0x6e/0xc0 sock_sendmsg+0x38/0x60 sock_write_iter+0x97/0x100 vfs_write+0x2df/0x380 ksys_write+0xa7/0xe0 do_syscall_64+0x3d/0x90 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7f6e22f018b7 Code: 0f 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 51 c3 48 83 ec 28 48 89 54 24 18 48 89 74 24 RSP: 002b:00007ffdb528a2f8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004000 RCX: 00007f6e22f018b7 RDX: 0000000000004000 RSI: 00000000025cdef0 RDI: 0000000000000028 RBP: 00000000020103c0 R08: 00007ffdb5289a90 R09: 0000000000000001 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000025cdef0 R13: 000000000204fca0 R14: 0000000000004000 R15: 0000000000004000 ---[ end trace 0000000000000000 ]--- > include/net/tls.h | 2 +- > net/tls/tls_main.c | 24 +++++++++++++++--------- > 2 files changed, 16 insertions(+), 10 deletions(-) > > diff --git a/include/net/tls.h b/include/net/tls.h > index 6056ce5a2aa5..5791ca7a189c 100644 > --- a/include/net/tls.h > +++ b/include/net/tls.h > @@ -258,7 +258,7 @@ struct tls_context { > struct scatterlist *partially_sent_record; > u16 partially_sent_offset; > > - bool in_tcp_sendpages; > + bool splicing_pages; > bool pending_open_record_frags; > > struct mutex tx_lock; /* protects partially_sent_* fields and > diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c > index f2e7302a4d96..3d45fdb5c4e9 100644 > --- a/net/tls/tls_main.c > +++ b/net/tls/tls_main.c > @@ -125,7 +125,10 @@ int tls_push_sg(struct sock *sk, > u16 first_offset, > int flags) > { > - int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST; > + struct bio_vec bvec; > + struct msghdr msg = { > + .msg_flags = MSG_SENDPAGE_NOTLAST | MSG_SPLICE_PAGES | flags, > + }; > int ret = 0; > struct page *p; > size_t size; > @@ -134,16 +137,19 @@ int tls_push_sg(struct sock *sk, > size = sg->length - offset; > offset += sg->offset; > > - ctx->in_tcp_sendpages = true; > + ctx->splicing_pages = true; > while (1) { > if (sg_is_last(sg)) > - sendpage_flags = flags; > + msg.msg_flags = flags; > > /* is sending application-limited? */ > tcp_rate_check_app_limited(sk); > p = sg_page(sg); > retry: > - ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags); > + bvec_set_page(&bvec, p, size, offset); > + iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size); > + > + ret = tcp_sendmsg_locked(sk, &msg, size); > > if (ret != size) { > if (ret > 0) { > @@ -155,7 +161,7 @@ int tls_push_sg(struct sock *sk, > offset -= sg->offset; > ctx->partially_sent_offset = offset; > ctx->partially_sent_record = (void *)sg; > - ctx->in_tcp_sendpages = false; > + ctx->splicing_pages = false; > return ret; > } > > @@ -169,7 +175,7 @@ int tls_push_sg(struct sock *sk, > size = sg->length; > } > > - ctx->in_tcp_sendpages = false; > + ctx->splicing_pages = false; > > return 0; > } > @@ -247,11 +253,11 @@ static void tls_write_space(struct sock *sk) > { > struct tls_context *ctx = tls_get_ctx(sk); > > - /* If in_tcp_sendpages call lower protocol write space handler > + /* If splicing_pages call lower protocol write space handler > * to ensure we wake up any waiting operations there. For example > - * if do_tcp_sendpages where to call sk_wait_event. > + * if splicing pages where to call sk_wait_event. > */ > - if (ctx->in_tcp_sendpages) { > + if (ctx->splicing_pages) { > ctx->sk_write_space(sk); > return; > } > >
Tariq Toukan <ttoukan.linux@gmail.com> wrote: > My team spotted a new degradation in TLS TX device offload, bisected to this > patch. I presume you're using some hardware (I'm guessing Mellanox?) that can actually do TLS offload? Unfortunately, I don't have any hardware that can do this, so I can't test the tls_device stuff. > From a quick look at the patch, it's not clear to me what's going wrong. > Please let us know of any helpful information that we can provide to help in > the debug. Can you find out what source line this corresponds to? RIP: 0010:skb_splice_from_iter+0x102/0x300 Assuming you're building your own kernel, something like the following might do the trick: echo "RIP: 0010:skb_splice_from_iter+0x102/0x300" | ./scripts/decode_stacktrace.sh /my/built/vmlinux /my/build/tree if you run it in the kernel source tree you're using and substitute the paths to vmlinux and the build tree for modules. David
On 07/06/2023 18:03, David Howells wrote: > Tariq Toukan <ttoukan.linux@gmail.com> wrote: > >> My team spotted a new degradation in TLS TX device offload, bisected to this >> patch. > > I presume you're using some hardware (I'm guessing Mellanox?) that can > actually do TLS offload? Unfortunately, I don't have any hardware that can do > this, so I can't test the tls_device stuff. > >> From a quick look at the patch, it's not clear to me what's going wrong. >> Please let us know of any helpful information that we can provide to help in >> the debug. > > Can you find out what source line this corresponds to? > > RIP: 0010:skb_splice_from_iter+0x102/0x300 > > Assuming you're building your own kernel, something like the following might > do the trick: > > echo "RIP: 0010:skb_splice_from_iter+0x102/0x300" | > ./scripts/decode_stacktrace.sh /my/built/vmlinux /my/build/tree > Hi, It's: RIP: 0010:skb_splice_from_iter (/usr/linux/net/core/skbuff.c:6957) which coresponds to this line: if (WARN_ON_ONCE(!sendpage_ok(page))) > if you run it in the kernel source tree you're using and substitute the > paths to vmlinux and the build tree for modules. > > David >
On 13/06/2023 14:15, Tariq Toukan wrote: > > > On 07/06/2023 18:03, David Howells wrote: >> Tariq Toukan <ttoukan.linux@gmail.com> wrote: >> >>> My team spotted a new degradation in TLS TX device offload, bisected >>> to this >>> patch. >> >> I presume you're using some hardware (I'm guessing Mellanox?) that can >> actually do TLS offload? Unfortunately, I don't have any hardware >> that can do >> this, so I can't test the tls_device stuff. >> >>> From a quick look at the patch, it's not clear to me what's going >>> wrong. >>> Please let us know of any helpful information that we can provide to >>> help in >>> the debug. >> >> Can you find out what source line this corresponds to? >> >> RIP: 0010:skb_splice_from_iter+0x102/0x300 >> >> Assuming you're building your own kernel, something like the following >> might >> do the trick: >> >> echo "RIP: 0010:skb_splice_from_iter+0x102/0x300" | >> ./scripts/decode_stacktrace.sh /my/built/vmlinux /my/build/tree >> > > Hi, > > It's: > RIP: 0010:skb_splice_from_iter (/usr/linux/net/core/skbuff.c:6957) > > which coresponds to this line: > if (WARN_ON_ONCE(!sendpage_ok(page))) > Hi David, Any other debug information that we can provide to progress with the analysis?
Tariq Toukan <ttoukan.linux@gmail.com> wrote:
> Any other debug information that we can provide to progress with the analysis?
Can you see if the problem still happens on this branch of my tree?
https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=sendpage-3-frag
It eases the restriction that the WARN_ON is warning about by (in patch 1[*])
copying slab objects into page fragments.
David
[*] "net: Copy slab data for sendmsg(MSG_SPLICE_PAGES)"
On 19/06/2023 12:35, David Howells wrote: > Tariq Toukan <ttoukan.linux@gmail.com> wrote: > >> Any other debug information that we can provide to progress with the analysis? > > Can you see if the problem still happens on this branch of my tree? > > https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=sendpage-3-frag > > It eases the restriction that the WARN_ON is warning about by (in patch 1[*]) > copying slab objects into page fragments. > > David > > [*] "net: Copy slab data for sendmsg(MSG_SPLICE_PAGES)" > Hi David, Unfortunately, it still happens: ------------[ cut here ]------------ WARNING: CPU: 2 PID: 93427 at net/core/skbuff.c:7013 skb_splice_from_iter+0x299/0x550 Modules linked in: bonding nf_tables vfio_pci ip_gre geneve ib_umad rdma_ucm ipip tunnel4 ip6_gre gre ip6_tunnel tunnel6 ib_ipoib mlx5_vfio_pci vfio_pci_core mlx5_ib ib_uverbs mlx5_core sch_mqprio sch_mqprio_lib sch_netem iptable_raw vfio_iommu_type1 vfio openvswitch nsh rpcrdma ib_iser libiscsi scsi_transport_iscsi rdma_cm iw_cm ib_cm ib_core xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter rpcsec_gss_krb5 auth_rpcgss oid_registry overlay zram zsmalloc fuse [last unloaded: nf_tables] CPU: 2 PID: 93427 Comm: nginx_openssl_3 Tainted: G W 6.4.0-rc6_net_next_mlx5_9b6e6b6 #1 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014 RIP: 0010:skb_splice_from_iter+0x299/0x550 Code: 49 8b 57 08 f6 c2 01 0f 85 89 01 00 00 8b 0d 22 b3 4a 01 4c 89 fa 85 c9 0f 8f 81 01 00 00 48 8b 12 80 e6 02 0f 84 a3 00 00 00 <0f> 0b 48 c7 c1 fb ff ff ff 44 01 6b 70 44 01 6b 74 44 01 ab d0 00 RSP: 0018:ffff8882a16d3a80 EFLAGS: 00010246 RAX: 0000000000000000 RBX: ffff88821a89ee00 RCX: 0000000000000000 RDX: 0000000000000000 RSI: ffffea0004c2cfc0 RDI: ffff88821a89ee00 RBP: 0000000000000f34 R08: 0000000000000011 R09: 0000000000000f34 R10: 0000000000000000 R11: 000000000000000d R12: 0000000000000004 R13: 0000000000002d0f R14: 0000000000000f34 R15: ffffea0004c2d000 FS: 00007f5c383eb740(0000) GS:ffff88885f880000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000002ea5000 CR3: 0000000264ffe006 CR4: 0000000000370ea0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: <TASK> ? __warn+0x79/0x120 ? skb_splice_from_iter+0x299/0x550 ? report_bug+0x17c/0x190 ? handle_bug+0x3c/0x60 ? exc_invalid_op+0x14/0x70 ? asm_exc_invalid_op+0x16/0x20 ? skb_splice_from_iter+0x299/0x550 tcp_sendmsg_locked+0x375/0xd00 tls_push_sg+0xdd/0x230 tls_push_data+0x6de/0xb00 tls_device_sendmsg+0x7a/0xd0 sock_sendmsg+0x38/0x60 sock_write_iter+0x97/0x100 vfs_write+0x2df/0x380 ksys_write+0xa7/0xe0 do_syscall_64+0x3d/0x90 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7f5c381018b7 Code: 0f 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 51 c3 48 83 ec 28 48 89 54 24 18 48 89 74 24 RSP: 002b:00007ffee9750848 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 RAX: ffffffffffffffda RBX: 0000000000004000 RCX: 00007f5c381018b7 RDX: 0000000000004000 RSI: 0000000002ea2dc0 RDI: 00000000000000d1 RBP: 0000000001d1bbe0 R08: 00007ffee974ffe0 R09: 0000000000000001 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000002ea2dc0 R13: 0000000001d2a7e0 R14: 0000000000004000 R15: 0000000000004000 </TASK> ---[ end trace 0000000000000000 ]---
Tariq Toukan <ttoukan.linux@gmail.com> wrote:
> WARNING: CPU: 2 PID: 93427 at net/core/skbuff.c:7013
Is that this line for you:
} else if (WARN_ON_ONCE(!sendpage_ok(page))) {
If so, it's not slab data, but we've got a page with a 0 refcount from
somewhere.
David
Can you try net-next/main? Parts of the branch you're trying have been dropped. David
On Tue, 27 Jun 2023 19:49:22 +0300 Tariq Toukan wrote: > Unfortunately, it still happens: > > ------------[ cut here ]------------ > WARNING: CPU: 2 PID: 93427 at net/core/skbuff.c:7013 I can't repro it on net-next with basic TLS 1.2 sendmsg/stream test + device offload, let us know if you still see it.
On 30/06/2023 20:21, Jakub Kicinski wrote: > On Tue, 27 Jun 2023 19:49:22 +0300 Tariq Toukan wrote: >> Unfortunately, it still happens: >> >> ------------[ cut here ]------------ >> WARNING: CPU: 2 PID: 93427 at net/core/skbuff.c:7013 > > I can't repro it on net-next with basic TLS 1.2 sendmsg/stream > test + device offload, let us know if you still see it. Hi, Unfortunately, it still repros for us. We are collecting more info on how the repro is affected by the different parameters. Regards, Tariq
On Tue, 4 Jul 2023 23:06:02 +0300 Tariq Toukan wrote: > Unfortunately, it still repros for us. > > We are collecting more info on how the repro is affected by the > different parameters. Consider configuring kdump for your test env. Debugging is super easy if one has the vmcore available.
On 05/07/2023 19:19, Jakub Kicinski wrote: > On Tue, 4 Jul 2023 23:06:02 +0300 Tariq Toukan wrote: >> Unfortunately, it still repros for us. >> >> We are collecting more info on how the repro is affected by the >> different parameters. > > Consider configuring kdump for your test env. Debugging is super easy > if one has the vmcore available. Hi Jakub, David, We repro the issue on the server side using this client command: $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s https://2.2.2.3:20443/256000b.img Port 20443 is configured with: ssl_ciphers ECDHE-RSA-AES128-GCM-SHA256; sendfile off; Important: 1. Couldn't repro with files smaller than 40KB. 2. Couldn't repro with "sendfile on;" In addition, we collected the vmcore (forced by panic_on_warn), it can be downloaded from here: https://drive.google.com/file/d/1Fi2dzgq6k2hb2L_kwyntRjfLF6_RmbxB/view?usp=sharing Regards, Tariq
On Sun, 23 Jul 2023 09:35:56 +0300 Tariq Toukan wrote: > Hi Jakub, David, > > We repro the issue on the server side using this client command: > $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s > https://2.2.2.3:20443/256000b.img > > Port 20443 is configured with: > ssl_ciphers ECDHE-RSA-AES128-GCM-SHA256; > sendfile off; > > > Important: > 1. Couldn't repro with files smaller than 40KB. > 2. Couldn't repro with "sendfile on;" > > In addition, we collected the vmcore (forced by panic_on_warn), it can > be downloaded from here: > https://drive.google.com/file/d/1Fi2dzgq6k2hb2L_kwyntRjfLF6_RmbxB/view?usp=sharing This has no symbols :( There is a small bug in this commit, we should always set SPLICE. But I don't see how that'd cause the warning you're seeing. Does your build have CONFIG_DEBUG_VM enabled? -->8------------------------- From: Jakub Kicinski <kuba@kernel.org> Date: Tue, 25 Jul 2023 17:03:25 -0700 Subject: net: tls: set MSG_SPLICE_PAGES consistently We used to change the flags for the last segment, because non-last segments had the MSG_SENDPAGE_NOTLAST flag set. That flag is no longer a thing so remove the setting. Since flags most likely don't have MSG_SPLICE_PAGES set this avoids passing parts of the sg as splice and parts as non-splice. ... tags ... Signed-off-by: Jakub Kicinski <kuba@kernel.org> --- net/tls/tls_main.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c index b6896126bb92..4a8ee2f6badb 100644 --- a/net/tls/tls_main.c +++ b/net/tls/tls_main.c @@ -139,9 +139,6 @@ int tls_push_sg(struct sock *sk, ctx->splicing_pages = true; while (1) { - if (sg_is_last(sg)) - msg.msg_flags = flags; - /* is sending application-limited? */ tcp_rate_check_app_limited(sk); p = sg_page(sg);
Tariq Toukan <ttoukan.linux@gmail.com> wrote: > We repro the issue on the server side using this client command: > $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s https://2.2.2.3:20443/256000b.img What's wrk? David
On 26/07/2023 13:51, David Howells wrote: > Tariq Toukan <ttoukan.linux@gmail.com> wrote: > >> We repro the issue on the server side using this client command: >> $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s https://2.2.2.3:20443/256000b.img > > What's wrk? > > David > Pretty known and standard client app. wrk - a HTTP benchmarking tool https://github.com/wg/wrk
On Wed, 26 Jul 2023 14:43:35 +0300 Tariq Toukan wrote: > >> We repro the issue on the server side using this client command: > >> $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s https://2.2.2.3:20443/256000b.img > > > > What's wrk? > > > > David > > > > Pretty known and standard client app. > wrk - a HTTP benchmarking tool > https://github.com/wg/wrk Let us know if your build has CONFIG_DEBUG_VM, please. Because in the old code the warning was gated by this config, so the bug may be older. We just started reporting it.
On 26/07/2023 3:30, Jakub Kicinski wrote: > On Sun, 23 Jul 2023 09:35:56 +0300 Tariq Toukan wrote: >> Hi Jakub, David, >> >> We repro the issue on the server side using this client command: >> $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s >> https://2.2.2.3:20443/256000b.img >> >> Port 20443 is configured with: >> ssl_ciphers ECDHE-RSA-AES128-GCM-SHA256; >> sendfile off; >> >> >> Important: >> 1. Couldn't repro with files smaller than 40KB. >> 2. Couldn't repro with "sendfile on;" >> >> In addition, we collected the vmcore (forced by panic_on_warn), it can >> be downloaded from here: >> https://drive.google.com/file/d/1Fi2dzgq6k2hb2L_kwyntRjfLF6_RmbxB/view?usp=sharing > > This has no symbols :( > Uh.. :/ I'll try to fix this and re-generate. > There is a small bug in this commit, we should always set SPLICE. > But I don't see how that'd cause the warning you're seeing. > Does your build have CONFIG_DEBUG_VM enabled? No. # CONFIG_DEBUG_VM is not set # CONFIG_DEBUG_VM_PGTABLE is not set > > -->8------------------------- > > From: Jakub Kicinski <kuba@kernel.org> > Date: Tue, 25 Jul 2023 17:03:25 -0700 > Subject: net: tls: set MSG_SPLICE_PAGES consistently > > We used to change the flags for the last segment, because > non-last segments had the MSG_SENDPAGE_NOTLAST flag set. > That flag is no longer a thing so remove the setting. > > Since flags most likely don't have MSG_SPLICE_PAGES set > this avoids passing parts of the sg as splice and parts > as non-splice. > > ... tags ... > Signed-off-by: Jakub Kicinski <kuba@kernel.org> > --- > net/tls/tls_main.c | 3 --- > 1 file changed, 3 deletions(-) > > diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c > index b6896126bb92..4a8ee2f6badb 100644 > --- a/net/tls/tls_main.c > +++ b/net/tls/tls_main.c > @@ -139,9 +139,6 @@ int tls_push_sg(struct sock *sk, > > ctx->splicing_pages = true; > while (1) { > - if (sg_is_last(sg)) > - msg.msg_flags = flags; > - > /* is sending application-limited? */ > tcp_rate_check_app_limited(sk); > p = sg_page(sg); I'll test this anyway tomorrow and update. Regards, Tariq
On Wed, 26 Jul 2023 22:20:42 +0300 Tariq Toukan wrote: > > There is a small bug in this commit, we should always set SPLICE. > > But I don't see how that'd cause the warning you're seeing. > > Does your build have CONFIG_DEBUG_VM enabled? > > No. > > # CONFIG_DEBUG_VM is not set > # CONFIG_DEBUG_VM_PGTABLE is not set Try testing v6.3 with DEBUG_VM enabled or just remove the IS_ENABLED() from: https://github.com/torvalds/linux/blob/v6.4/net/ipv4/tcp.c#L1051
On 26/07/2023 3:30, Jakub Kicinski wrote: > On Sun, 23 Jul 2023 09:35:56 +0300 Tariq Toukan wrote: >> Hi Jakub, David, >> >> We repro the issue on the server side using this client command: >> $ wrk -b2.2.2.2 -t4 -c1000 -d5 --timeout 5s >> https://2.2.2.3:20443/256000b.img >> >> Port 20443 is configured with: >> ssl_ciphers ECDHE-RSA-AES128-GCM-SHA256; >> sendfile off; >> >> >> Important: >> 1. Couldn't repro with files smaller than 40KB. >> 2. Couldn't repro with "sendfile on;" >> >> In addition, we collected the vmcore (forced by panic_on_warn), it can >> be downloaded from here: >> https://drive.google.com/file/d/1Fi2dzgq6k2hb2L_kwyntRjfLF6_RmbxB/view?usp=sharing > > This has no symbols :( > > There is a small bug in this commit, we should always set SPLICE. > But I don't see how that'd cause the warning you're seeing. > Does your build have CONFIG_DEBUG_VM enabled? > > -->8------------------------- > > From: Jakub Kicinski <kuba@kernel.org> > Date: Tue, 25 Jul 2023 17:03:25 -0700 > Subject: net: tls: set MSG_SPLICE_PAGES consistently > > We used to change the flags for the last segment, because > non-last segments had the MSG_SENDPAGE_NOTLAST flag set. > That flag is no longer a thing so remove the setting. > > Since flags most likely don't have MSG_SPLICE_PAGES set > this avoids passing parts of the sg as splice and parts > as non-splice. > > ... tags ... > Signed-off-by: Jakub Kicinski <kuba@kernel.org> > --- > net/tls/tls_main.c | 3 --- > 1 file changed, 3 deletions(-) > > diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c > index b6896126bb92..4a8ee2f6badb 100644 > --- a/net/tls/tls_main.c > +++ b/net/tls/tls_main.c > @@ -139,9 +139,6 @@ int tls_push_sg(struct sock *sk, > > ctx->splicing_pages = true; > while (1) { > - if (sg_is_last(sg)) > - msg.msg_flags = flags; > - > /* is sending application-limited? */ > tcp_rate_check_app_limited(sk); > p = sg_page(sg); Hi Jakub, When applying this patch, repro disappears! :) Apparently it is related to the warning. Please go on and submit it. Tested-by: Tariq Toukan <tariqt@nvidia.com> We are going to run more comprehensive tests, I'll let you know if we find anything unusual. Regards, Tariq
On 26/07/2023 23:08, Jakub Kicinski wrote: > On Wed, 26 Jul 2023 22:20:42 +0300 Tariq Toukan wrote: >>> There is a small bug in this commit, we should always set SPLICE. >>> But I don't see how that'd cause the warning you're seeing. >>> Does your build have CONFIG_DEBUG_VM enabled? >> >> No. >> >> # CONFIG_DEBUG_VM is not set >> # CONFIG_DEBUG_VM_PGTABLE is not set > > Try testing v6.3 with DEBUG_VM enabled or just remove the IS_ENABLED() > from: https://github.com/torvalds/linux/blob/v6.4/net/ipv4/tcp.c#L1051 Tested. It doesn't repro.
On Thu, 3 Aug 2023 14:47:35 +0300 Tariq Toukan wrote: > When applying this patch, repro disappears! :) > Apparently it is related to the warning. > Please go on and submit it. I have no idea how. I found a different bug, staring at this code for another hour. But I still don't get how we can avoid UaF on a page by having the TCP take a ref on it rather than copy it. If anything we should have 2 refs on any page in the sg, one because it's on the sg, and another held by the re-tx handling. So I'm afraid we're papering over something here :( We need to keep digging.
On 04/08/2023 6:12, Jakub Kicinski wrote: > On Thu, 3 Aug 2023 14:47:35 +0300 Tariq Toukan wrote: >> When applying this patch, repro disappears! :) >> Apparently it is related to the warning. >> Please go on and submit it. > > I have no idea how. I found a different bug, staring at this code > for another hour. But I still don't get how we can avoid UaF on > a page by having the TCP take a ref on it rather than copy it. > > If anything we should have 2 refs on any page in the sg, one because > it's on the sg, and another held by the re-tx handling. > > So I'm afraid we're papering over something here :( We need to keep > digging. Hi Jakub, I'm glad to see that you already nailed the other bug and merged the fix. I can update that we ran comprehensive TLS testing on a branch that contains your proposed fix (net: tls: set MSG_SPLICE_PAGES consistently), and doesn't contain the other fix (net: tls: avoid discarding data on record close). Except for one "known" issue (we'll discuss it in a second), the runs look clean. No more traces or encrypt/decrypt error counters. Your proposed fix seems to work and causes no degradation. How do you suggest proceeding here? One mysterious remaining issue, which I already reported some time ago but couldn't effectively debug due to other TLS bugs, is the increase of TlsDecryptError / TlsEncryptError counters when running kTLS offloaded traffic during bond creation on some other interface. Weird... We should start giving it the needed attention now that the other issues seem to be resolved. Regards, Tariq
Tariq Toukan <ttoukan.linux@gmail.com> wrote: > We are collecting more info on how the repro is affected by the different > parameters. I'm wondering if userspace is feeding the unspliceable page in somehow. Could you try running with the attached changes? It might help catch the point at which the offending page is first spliced into the pipe and any backtrace might help localise the driver that's producing it. Thanks, David --- diff --git a/fs/splice.c b/fs/splice.c index 3e2a31e1ce6a..877df1de3863 100644 --- a/fs/splice.c +++ b/fs/splice.c @@ -218,6 +218,8 @@ ssize_t splice_to_pipe(struct pipe_inode_info *pipe, while (!pipe_full(head, tail, pipe->max_usage)) { struct pipe_buffer *buf = &pipe->bufs[head & mask]; + WARN_ON_ONCE(!sendpage_ok(spd->pages[page_nr])); + buf->page = spd->pages[page_nr]; buf->offset = spd->partial[page_nr].offset; buf->len = spd->partial[page_nr].len; @@ -252,6 +254,8 @@ ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) unsigned int mask = pipe->ring_size - 1; int ret; + WARN_ON_ONCE(!sendpage_ok(buf->page)); + if (unlikely(!pipe->readers)) { send_sig(SIGPIPE, current, 0); ret = -EPIPE; @@ -861,6 +865,8 @@ ssize_t splice_to_socket(struct pipe_inode_info *pipe, struct file *out, break; } + WARN_ON_ONCE(!sendpage_ok(buf->page)); + bvec_set_page(&bvec[bc++], buf->page, seg, buf->offset); remain -= seg; if (remain == 0 || bc >= ARRAY_SIZE(bvec)) @@ -1411,6 +1417,8 @@ static int iter_to_pipe(struct iov_iter *from, for (i = 0; i < n; i++) { int size = min_t(int, left, PAGE_SIZE - start); + WARN_ON_ONCE(!sendpage_ok(pages[i])); + buf.page = pages[i]; buf.offset = start; buf.len = size;
diff --git a/include/net/tls.h b/include/net/tls.h index 6056ce5a2aa5..5791ca7a189c 100644 --- a/include/net/tls.h +++ b/include/net/tls.h @@ -258,7 +258,7 @@ struct tls_context { struct scatterlist *partially_sent_record; u16 partially_sent_offset; - bool in_tcp_sendpages; + bool splicing_pages; bool pending_open_record_frags; struct mutex tx_lock; /* protects partially_sent_* fields and diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c index f2e7302a4d96..3d45fdb5c4e9 100644 --- a/net/tls/tls_main.c +++ b/net/tls/tls_main.c @@ -125,7 +125,10 @@ int tls_push_sg(struct sock *sk, u16 first_offset, int flags) { - int sendpage_flags = flags | MSG_SENDPAGE_NOTLAST; + struct bio_vec bvec; + struct msghdr msg = { + .msg_flags = MSG_SENDPAGE_NOTLAST | MSG_SPLICE_PAGES | flags, + }; int ret = 0; struct page *p; size_t size; @@ -134,16 +137,19 @@ int tls_push_sg(struct sock *sk, size = sg->length - offset; offset += sg->offset; - ctx->in_tcp_sendpages = true; + ctx->splicing_pages = true; while (1) { if (sg_is_last(sg)) - sendpage_flags = flags; + msg.msg_flags = flags; /* is sending application-limited? */ tcp_rate_check_app_limited(sk); p = sg_page(sg); retry: - ret = do_tcp_sendpages(sk, p, offset, size, sendpage_flags); + bvec_set_page(&bvec, p, size, offset); + iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size); + + ret = tcp_sendmsg_locked(sk, &msg, size); if (ret != size) { if (ret > 0) { @@ -155,7 +161,7 @@ int tls_push_sg(struct sock *sk, offset -= sg->offset; ctx->partially_sent_offset = offset; ctx->partially_sent_record = (void *)sg; - ctx->in_tcp_sendpages = false; + ctx->splicing_pages = false; return ret; } @@ -169,7 +175,7 @@ int tls_push_sg(struct sock *sk, size = sg->length; } - ctx->in_tcp_sendpages = false; + ctx->splicing_pages = false; return 0; } @@ -247,11 +253,11 @@ static void tls_write_space(struct sock *sk) { struct tls_context *ctx = tls_get_ctx(sk); - /* If in_tcp_sendpages call lower protocol write space handler + /* If splicing_pages call lower protocol write space handler * to ensure we wake up any waiting operations there. For example - * if do_tcp_sendpages where to call sk_wait_event. + * if splicing pages where to call sk_wait_event. */ - if (ctx->in_tcp_sendpages) { + if (ctx->splicing_pages) { ctx->sk_write_space(sk); return; }
do_tcp_sendpages() is now just a small wrapper around tcp_sendmsg_locked(), so inline it, allowing do_tcp_sendpages() to be removed. This is part of replacing ->sendpage() with a call to sendmsg() with MSG_SPLICE_PAGES set. Signed-off-by: David Howells <dhowells@redhat.com> cc: Boris Pismenny <borisp@nvidia.com> cc: John Fastabend <john.fastabend@gmail.com> cc: Jakub Kicinski <kuba@kernel.org> cc: "David S. Miller" <davem@davemloft.net> cc: Eric Dumazet <edumazet@google.com> cc: Paolo Abeni <pabeni@redhat.com> cc: Jens Axboe <axboe@kernel.dk> cc: Matthew Wilcox <willy@infradead.org> cc: netdev@vger.kernel.org --- include/net/tls.h | 2 +- net/tls/tls_main.c | 24 +++++++++++++++--------- 2 files changed, 16 insertions(+), 10 deletions(-)