Message ID | 3350eaf14a073538bf491f93fc852cd02ab0875a.1699280113.git.pabeni@redhat.com (mailing list archive) |
---|---|
State | Superseded, archived |
Commit | 40b5e6f7d0b7bd404ec5c17be31e549b3aa5b465 |
Delegated to: | Paolo Abeni |
Headers | show |
Series | [mptcp-net] mptcp: fix possible NULL pointer dereference on close | expand |
Context | Check | Description |
---|---|---|
matttbe/build | success | Build and static analysis OK |
matttbe/checkpatch | warning | total: 0 errors, 3 warnings, 0 checks, 12 lines checked |
matttbe/KVM_Validation__normal__except_selftest_mptcp_join_ | success | Success! ✅ |
matttbe/KVM_Validation__debug__only_selftest_mptcp_join_ | success | Success! ✅ |
matttbe/KVM_Validation__debug__except_selftest_mptcp_join_ | success | Success! ✅ |
matttbe/KVM_Validation__normal__only_selftest_mptcp_join_ | success | Success! ✅ |
Hi Paolo, Thank you for your modifications, that's great! Our CI did some validations and here is its report: - KVM Validation: normal (except selftest_mptcp_join): - Success! ✅: - Task: https://cirrus-ci.com/task/5967971751821312 - Summary: https://api.cirrus-ci.com/v1/artifact/task/5967971751821312/summary/summary.txt - KVM Validation: debug (only selftest_mptcp_join): - Success! ✅: - Task: https://cirrus-ci.com/task/4701334356623360 - Summary: https://api.cirrus-ci.com/v1/artifact/task/4701334356623360/summary/summary.txt - KVM Validation: debug (except selftest_mptcp_join): - Success! ✅: - Task: https://cirrus-ci.com/task/6530921705242624 - Summary: https://api.cirrus-ci.com/v1/artifact/task/6530921705242624/summary/summary.txt - KVM Validation: normal (only selftest_mptcp_join): - Success! ✅: - Task: https://cirrus-ci.com/task/5405021798400000 - Summary: https://api.cirrus-ci.com/v1/artifact/task/5405021798400000/summary/summary.txt Initiator: Patchew Applier Commits: https://github.com/multipath-tcp/mptcp_net-next/commits/78fb6e638f64 If there are some issues, you can reproduce them using the same environment as the one used by the CI thanks to a docker image, e.g.: $ cd [kernel source code] $ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \ --pull always mptcp/mptcp-upstream-virtme-docker:latest \ auto-debug For more details: https://github.com/multipath-tcp/mptcp-upstream-virtme-docker Please note that despite all the efforts that have been already done to have a stable tests suite when executed on a public CI like here, it is possible some reported issues are not due to your modifications. Still, do not hesitate to help us improve that ;-) Cheers, MPTCP GH Action bot Bot operated by Matthieu Baerts (Tessares)
On Mon, 6 Nov 2023, Paolo Abeni wrote: > After the blamed commit below, the MPTCP release callback > can dereference the first subflow pointer, even at socket > release time. > > In such scenario such subflow is already released and the pointer > zeroed, causing the following splat: > > general protection fault, probably for non-canonical address 0xdffffc00000000f2: 0000 [#1] PREEMPT SMP KASAN > KASAN: null-ptr-deref in range [0x0000000000000790-0x0000000000000797] > CPU: 1 PID: 26719 Comm: syz-executor.2 Not tainted 6.6.0-syzkaller-10102-gff269e2cd5ad #0 > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/09/2023 > RIP: 0010:mptcp_subflow_ctx net/mptcp/protocol.h:542 [inline] > RIP: 0010:__mptcp_propagate_sndbuf net/mptcp/protocol.h:813 [inline] > RIP: 0010:__mptcp_set_connected+0x57/0x3e0 net/mptcp/subflow.c:424 > RAX: dffffc0000000000 RBX: 0000000000000000 RCX: ffffffff8a62323c > RDX: 00000000000000f2 RSI: ffffffff8a630116 RDI: 0000000000000790 > RBP: ffff88803334b100 R08: 0000000000000001 R09: 0000000000000000 > R10: 0000000000000001 R11: 0000000000000034 R12: ffff88803334b198 > R13: ffff888054f0b018 R14: 0000000000000000 R15: ffff88803334b100 > FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 00007fbcb4f75198 CR3: 000000006afb5000 CR4: 00000000003506f0 > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 > Call Trace: > <TASK> > mptcp_release_cb+0xa2c/0xc40 net/mptcp/protocol.c:3405 > release_sock+0xba/0x1f0 net/core/sock.c:3537 > mptcp_close+0x32/0xf0 net/mptcp/protocol.c:3084 > inet_release+0x132/0x270 net/ipv4/af_inet.c:433 > inet6_release+0x4f/0x70 net/ipv6/af_inet6.c:485 > __sock_release+0xae/0x260 net/socket.c:659 > sock_close+0x1c/0x20 net/socket.c:1419 > __fput+0x270/0xbb0 fs/file_table.c:394 > task_work_run+0x14d/0x240 kernel/task_work.c:180 > exit_task_work include/linux/task_work.h:38 [inline] > do_exit+0xa92/0x2a20 kernel/exit.c:876 > do_group_exit+0xd4/0x2a0 kernel/exit.c:1026 > get_signal+0x23ba/0x2790 kernel/signal.c:2900 > arch_do_signal_or_restart+0x90/0x7f0 arch/x86/kernel/signal.c:309 > exit_to_user_mode_loop kernel/entry/common.c:168 [inline] > exit_to_user_mode_prepare+0x11f/0x240 kernel/entry/common.c:204 > __syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline] > syscall_exit_to_user_mode+0x1d/0x60 kernel/entry/common.c:296 > do_syscall_64+0x4b/0x110 arch/x86/entry/common.c:88 > entry_SYSCALL_64_after_hwframe+0x63/0x6b > RIP: 0033:0x7fb515e7cae9 > Code: Unable to access opcode bytes at 0x7fb515e7cabf. > RSP: 002b:00007fb516c560c8 EFLAGS: 00000246 ORIG_RAX: 000000000000002e > RAX: 000000000000003c RBX: 00007fb515f9c120 RCX: 00007fb515e7cae9 > RDX: 0000000000000000 RSI: 0000000020000140 RDI: 0000000000000006 > RBP: 00007fb515ec847a R08: 0000000000000000 R09: 0000000000000000 > R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 > R13: 000000000000006e R14: 00007fb515f9c120 R15: 00007ffc631eb968 > </TASK> > > Address the issue explicitly checking msk->first before tacking > action. > > Fixes: 8005184fd1ca ("mptcp: refactor sndbuf auto-tuning") > Reported-by: syzbot+9dfbaedb6e6baca57a32@syzkaller.appspotmail.com > Reported-by: Eric Dumazet <edumazet@google.com> > Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/454 > Signed-off-by: Paolo Abeni <pabeni@redhat.com> > --- > net/mptcp/protocol.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c > index 0ad507ac6bc7..1490fd724036 100644 > --- a/net/mptcp/protocol.c > +++ b/net/mptcp/protocol.c > @@ -3399,9 +3399,10 @@ static void mptcp_release_cb(struct sock *sk) > __mptcp_clean_una_wakeup(sk); > if (unlikely(msk->cb_flags)) { > /* be sure to set the current sk state before tacking actions (taking?) > - * depending on sk_state, that is processing MPTCP_ERROR_REPORT > + * depending on sk_state (MPTCP_ERROR_REPORT) > + * On sk release avoid actions depending on the first subflow > */ > - if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags)) > + if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags) && msk->first) > __mptcp_set_connected(sk); > if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags)) > __mptcp_error_report(sk); > -- > 2.41.0 Hi Paolo - This does address the call stack that syzbot triggered. Are you sure enough that the only way to reach __mptcp_set_connected() with msk->first == NULL is when mptcp_set_connected() encountered a locked socket and had to defer the call? It seems like that *should* be the case but it's not obvious to me that it's impossible in every other scenario. Checking for NULL inside __mptcp_set_connected() could catch more edge cases. If you're confident that there are no other edge cases it would help to note that in either the commit message or comment. - Mat
On Mon, 2023-11-06 at 17:44 -0800, Mat Martineau wrote: > This does address the call stack that syzbot triggered. Are you sure > enough that the only way to reach __mptcp_set_connected() with msk->first > == NULL is when mptcp_set_connected() encountered a locked socket and had > to defer the call? It seems like that *should* be the case but it's not > obvious to me that it's impossible in every other scenario. AFAIK, yes. msk->first is freed and the pointer is set to NULL only at sk destruction time, by: __mptcp_destroy_sock() -> mptcp_destroy() -> mptcp_destroy_common() -> __mptcp_close_ssk() The commit message intended to point to the above even if a very very terse way. Putting the check in release callback - in an unlikely branch - was intentional to avoid unneeded conditional in all the other cases. Please let me know if I should repost with a more verbose commit message (and addressing the typo you noticed). Thanks, Paolo
On Tue, 7 Nov 2023, Paolo Abeni wrote: > On Mon, 2023-11-06 at 17:44 -0800, Mat Martineau wrote: >> This does address the call stack that syzbot triggered. Are you sure >> enough that the only way to reach __mptcp_set_connected() with msk->first >> == NULL is when mptcp_set_connected() encountered a locked socket and had >> to defer the call? It seems like that *should* be the case but it's not >> obvious to me that it's impossible in every other scenario. > > AFAIK, yes. msk->first is freed and the pointer is set to NULL only at > sk destruction time, by: __mptcp_destroy_sock() -> mptcp_destroy() -> > mptcp_destroy_common() -> __mptcp_close_ssk() > > The commit message intended to point to the above even if a very very > terse way. > > Putting the check in release callback - in an unlikely branch - was > intentional to avoid unneeded conditional in all the other cases. > > > Please let me know if I should repost with a more verbose commit > message (and addressing the typo you noticed). > Hi Paolo - Yes, I think a v2 with more specifics would be helpful. Thanks! Mat
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c index 0ad507ac6bc7..1490fd724036 100644 --- a/net/mptcp/protocol.c +++ b/net/mptcp/protocol.c @@ -3399,9 +3399,10 @@ static void mptcp_release_cb(struct sock *sk) __mptcp_clean_una_wakeup(sk); if (unlikely(msk->cb_flags)) { /* be sure to set the current sk state before tacking actions - * depending on sk_state, that is processing MPTCP_ERROR_REPORT + * depending on sk_state (MPTCP_ERROR_REPORT) + * On sk release avoid actions depending on the first subflow */ - if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags)) + if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags) && msk->first) __mptcp_set_connected(sk); if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags)) __mptcp_error_report(sk);
After the blamed commit below, the MPTCP release callback can dereference the first subflow pointer, even at socket release time. In such scenario such subflow is already released and the pointer zeroed, causing the following splat: general protection fault, probably for non-canonical address 0xdffffc00000000f2: 0000 [#1] PREEMPT SMP KASAN KASAN: null-ptr-deref in range [0x0000000000000790-0x0000000000000797] CPU: 1 PID: 26719 Comm: syz-executor.2 Not tainted 6.6.0-syzkaller-10102-gff269e2cd5ad #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/09/2023 RIP: 0010:mptcp_subflow_ctx net/mptcp/protocol.h:542 [inline] RIP: 0010:__mptcp_propagate_sndbuf net/mptcp/protocol.h:813 [inline] RIP: 0010:__mptcp_set_connected+0x57/0x3e0 net/mptcp/subflow.c:424 RAX: dffffc0000000000 RBX: 0000000000000000 RCX: ffffffff8a62323c RDX: 00000000000000f2 RSI: ffffffff8a630116 RDI: 0000000000000790 RBP: ffff88803334b100 R08: 0000000000000001 R09: 0000000000000000 R10: 0000000000000001 R11: 0000000000000034 R12: ffff88803334b198 R13: ffff888054f0b018 R14: 0000000000000000 R15: ffff88803334b100 FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fbcb4f75198 CR3: 000000006afb5000 CR4: 00000000003506f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: <TASK> mptcp_release_cb+0xa2c/0xc40 net/mptcp/protocol.c:3405 release_sock+0xba/0x1f0 net/core/sock.c:3537 mptcp_close+0x32/0xf0 net/mptcp/protocol.c:3084 inet_release+0x132/0x270 net/ipv4/af_inet.c:433 inet6_release+0x4f/0x70 net/ipv6/af_inet6.c:485 __sock_release+0xae/0x260 net/socket.c:659 sock_close+0x1c/0x20 net/socket.c:1419 __fput+0x270/0xbb0 fs/file_table.c:394 task_work_run+0x14d/0x240 kernel/task_work.c:180 exit_task_work include/linux/task_work.h:38 [inline] do_exit+0xa92/0x2a20 kernel/exit.c:876 do_group_exit+0xd4/0x2a0 kernel/exit.c:1026 get_signal+0x23ba/0x2790 kernel/signal.c:2900 arch_do_signal_or_restart+0x90/0x7f0 arch/x86/kernel/signal.c:309 exit_to_user_mode_loop kernel/entry/common.c:168 [inline] exit_to_user_mode_prepare+0x11f/0x240 kernel/entry/common.c:204 __syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline] syscall_exit_to_user_mode+0x1d/0x60 kernel/entry/common.c:296 do_syscall_64+0x4b/0x110 arch/x86/entry/common.c:88 entry_SYSCALL_64_after_hwframe+0x63/0x6b RIP: 0033:0x7fb515e7cae9 Code: Unable to access opcode bytes at 0x7fb515e7cabf. RSP: 002b:00007fb516c560c8 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: 000000000000003c RBX: 00007fb515f9c120 RCX: 00007fb515e7cae9 RDX: 0000000000000000 RSI: 0000000020000140 RDI: 0000000000000006 RBP: 00007fb515ec847a R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 000000000000006e R14: 00007fb515f9c120 R15: 00007ffc631eb968 </TASK> Address the issue explicitly checking msk->first before tacking action. Fixes: 8005184fd1ca ("mptcp: refactor sndbuf auto-tuning") Reported-by: syzbot+9dfbaedb6e6baca57a32@syzkaller.appspotmail.com Reported-by: Eric Dumazet <edumazet@google.com> Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/454 Signed-off-by: Paolo Abeni <pabeni@redhat.com> --- net/mptcp/protocol.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)