----- Original Message ----- > From: "Steven Rostedt" <[email protected]> > To: "Sasha Levin" <[email protected]> > Cc: "Oleg Nesterov" <[email protected]>, [email protected], "LKML" > <[email protected]>, "Dave Jones" > <[email protected]>, "Mathieu Desnoyers" <[email protected]> > Sent: Wednesday, May 7, 2014 12:06:27 PM > Subject: Re: ptrace: gpf in syscall_trace_enter > > [ adding Mathieu as well, as this is tracepoint code ] > > On Wed, 07 May 2014 11:23:36 -0400 > Sasha Levin <[email protected]> wrote: > > > On 05/07/2014 10:31 AM, Steven Rostedt wrote: > > > On Wed, 7 May 2014 16:04:22 +0200 > > > Oleg Nesterov <[email protected]> wrote: > > > > > >> On 05/06, Sasha Levin wrote: > > >>> > > >>> On 05/06/2014 08:36 PM, Sasha Levin wrote: > > >>>> Hi all, > > >>>> > > >>>> While fuzzing with trinity inside a KVM tools guest running the latest > > >>>> -next > > >>>> kernel I've stumbled on the following spew: > > >>> > > >>> And another similar trace: > > >> > > >> Again, this looks like __DO_TRACE() trying to call it_func_ptr->func(). > > > > > > Really? Can I see an objdump of the location of the crash. Preferably > > > the entire function. > > > > 0000000000002740 <syscall_trace_leave>: > > 2740: e8 00 00 00 00 callq 2745 <syscall_trace_leave+0x5> > > 2741: R_X86_64_PC32 __fentry__-0x4 > > 2745: 55 push %rbp > > 2746: 48 89 e5 mov %rsp,%rbp > > 2749: 48 83 ec 20 sub $0x20,%rsp > > 274d: 48 89 5d e8 mov %rbx,-0x18(%rbp) > > 2751: 48 89 fb mov %rdi,%rbx > > 2754: 4c 89 65 f0 mov %r12,-0x10(%rbp) > > 2758: 4c 89 6d f8 mov %r13,-0x8(%rbp) > > 275c: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) > > 2761: 65 48 8b 04 25 00 00 mov %gs:0x0,%rax > > 2768: 00 00 > > 2766: R_X86_64_32S current_task > > 276a: 48 83 b8 b8 0b 00 00 cmpq $0x0,0xbb8(%rax) > > 2771: 00 > > 2772: 74 1c je 2790 <syscall_trace_leave+0x50> > > 2774: 48 8b 73 50 mov 0x50(%rbx),%rsi > > 2778: 31 ff xor %edi,%edi > > 277a: 48 81 fe 00 f0 ff ff cmp $0xfffffffffffff000,%rsi > > 2781: 40 0f 96 c7 setbe %dil > > 2785: e8 00 00 00 00 callq 278a <syscall_trace_leave+0x4a> > > 2786: R_X86_64_PC32 __audit_syscall_exit-0x4 > > 278a: 66 0f 1f 44 00 00 nopw 0x0(%rax,%rax,1) > > 2790: 65 48 8b 04 25 00 00 mov %gs:0x0,%rax > > 2797: 00 00 > > 2795: R_X86_64_32S kernel_stack > > 2799: 48 8b 80 38 e0 ff ff mov -0x1fc8(%rax),%rax > > 27a0: a9 00 00 00 10 test $0x10000000,%eax > > 27a5: 74 71 je 2818 <syscall_trace_leave+0xd8> > > 27a7: 4c 8b 6b 50 mov 0x50(%rbx),%r13 > > 27ab: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) > > Here's the static_key branch > > > 27b0: eb 62 jmp 2814 <syscall_trace_leave+0xd4> > > 27b2: 80 3d 00 00 00 00 00 cmpb $0x0,0x0(%rip) # 27b9 > > <syscall_trace_leave+0x79> > > 27b4: R_X86_64_PC32 .data.unlikely-0x4 > > 27b9: 75 28 jne 27e3 <syscall_trace_leave+0xa3> > > 27bb: e8 00 00 00 00 callq 27c0 <syscall_trace_leave+0x80> > > 27bc: R_X86_64_PC32 .text.unlikely-0x4 > > Interesting that we have a "callq" to the next instruction. > > > 27c0: 85 c0 test %eax,%eax > > 27c2: 75 1f jne 27e3 <syscall_trace_leave+0xa3> > > 27c4: 48 c7 c2 00 00 00 00 mov $0x0,%rdx > > 27c7: R_X86_64_32S .rodata.str1.8+0x60 > > 27cb: be 3e 00 00 00 mov $0x3e,%esi > > 27d0: 48 c7 c7 00 00 00 00 mov $0x0,%rdi > > 27d3: R_X86_64_32S .rodata.str1.8+0x90 > > 27d7: c6 05 00 00 00 00 01 movb $0x1,0x0(%rip) # 27de > > <syscall_trace_leave+0x9e> > > 27d9: R_X86_64_PC32 .data.unlikely-0x4 > > 27de: e8 00 00 00 00 callq 27e3 <syscall_trace_leave+0xa3> > > 27df: R_X86_64_PC32 lockdep_rcu_suspicious-0x4 > > OK, rcu debugging is on. Not really a factor, just making notes. > > > > 27e3: 4d 85 e4 test %r12,%r12 > > %r12 is the it_func_ptr > > > 27e6: 75 10 jne 27f8 <syscall_trace_leave+0xb8> > > 27e8: 65 ff 0c 25 00 00 00 decl %gs:0x0 > > 27ef: 00 > > 27ec: R_X86_64_32S __preempt_count > > 27f0: 0f 84 85 00 00 00 je 287b <syscall_trace_leave+0x13b> > > 27f6: eb 1c jmp 2814 <syscall_trace_leave+0xd4> > > 27f8: 49 8b 7c 24 08 mov 0x8(%r12),%rdi > > 27fd: 4c 89 ea mov %r13,%rdx > > 2800: 48 89 de mov %rbx,%rsi > > 2803: 41 ff 14 24 callq *(%r12) > > As I stated, I didn't need the offset that I asked for, but the > machine code had the information I needed: > > 24 08 4c 89 ea 48 89 de <41> ff 14 24 49 83 c4 10 49 > > Which matches 2803. > > From your dump: > > R10: 0000000000000000 R11: 0000000000000000 R12: 000000000000000c > > Yeah, that's a bad pointer. > > OK, for some reason, funcs got 0xc?
Oh crap, I think I made a stupid mistake there. Calling call_rcu to free the old func before rcu_assign_pointer publishes the new func. Can you try the attached patch ? (sorry for no inlining, using a dumb mail client here) Thanks, Mathieu > > -- Steve > > > > 2807: 49 83 c4 10 add $0x10,%r12 > > 280b: 49 83 3c 24 00 cmpq $0x0,(%r12) > > 2810: 75 e6 jne 27f8 <syscall_trace_leave+0xb8> > > 2812: eb d4 jmp 27e8 <syscall_trace_leave+0xa8> > > 2814: 0f 1f 40 00 nopl 0x0(%rax) > > 2818: 65 48 8b 04 25 00 00 mov %gs:0x0,%rax > > 281f: 00 00 > > 281d: R_X86_64_32S kernel_stack > > 2821: 48 8b 90 38 e0 ff ff mov -0x1fc8(%rax),%rdx > > 2828: 83 e2 10 and $0x10,%edx > > 282b: 74 5b je 2888 <syscall_trace_leave+0x148> > > 282d: 48 8b 80 38 e0 ff ff mov -0x1fc8(%rax),%rax > > 2834: a8 40 test $0x40,%al > > 2836: 75 50 jne 2888 <syscall_trace_leave+0x148> > > 2838: be 01 00 00 00 mov $0x1,%esi > > 283d: 0f 1f 00 nopl (%rax) > > 2840: 48 89 df mov %rbx,%rdi > > 2843: e8 f8 fa ff ff callq 2340 > > <tracehook_report_syscall_exit> > > 2848: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) > > 284d: eb 56 jmp 28a5 <syscall_trace_leave+0x165> > > 284f: 90 nop > > 2850: e8 00 00 00 00 callq 2855 <syscall_trace_leave+0x115> > > 2851: R_X86_64_PC32 context_tracking_user_exit-0x4 > > 2855: e9 07 ff ff ff jmpq 2761 <syscall_trace_leave+0x21> > > 285a: 65 ff 04 25 00 00 00 incl %gs:0x0 > > 2861: 00 > > 285e: R_X86_64_32S __preempt_count > > 2862: 4c 8b 25 00 00 00 00 mov 0x0(%rip),%r12 # 2869 > > <syscall_trace_leave+0x129> > > 2865: R_X86_64_PC32 __tracepoint_sys_exit+0x2c > > 2869: e8 00 00 00 00 callq 286e <syscall_trace_leave+0x12e> > > 286a: R_X86_64_PC32 debug_lockdep_rcu_enabled-0x4 > > 286e: 85 c0 test %eax,%eax > > 2870: 0f 85 3c ff ff ff jne 27b2 <syscall_trace_leave+0x72> > > 2876: e9 68 ff ff ff jmpq 27e3 <syscall_trace_leave+0xa3> > > 287b: e8 00 00 00 00 callq 2880 <syscall_trace_leave+0x140> > > 287c: R_X86_64_PC32 ___preempt_schedule_context-0x4 > > 2880: eb 96 jmp 2818 <syscall_trace_leave+0xd8> > > 2882: 66 0f 1f 44 00 00 nopw 0x0(%rax,%rax,1) > > 2888: 65 48 8b 04 25 00 00 mov %gs:0x0,%rax > > 288f: 00 00 > > 288d: R_X86_64_32S kernel_stack > > 2891: 48 8b 80 38 e0 ff ff mov -0x1fc8(%rax),%rax > > 2898: 31 f6 xor %esi,%esi > > 289a: a8 01 test $0x1,%al > > 289c: 75 a2 jne 2840 <syscall_trace_leave+0x100> > > 289e: eb a8 jmp 2848 <syscall_trace_leave+0x108> > > 28a0: e8 00 00 00 00 callq 28a5 <syscall_trace_leave+0x165> > > 28a1: R_X86_64_PC32 context_tracking_user_enter-0x4 > > 28a5: 48 8b 5d e8 mov -0x18(%rbp),%rbx > > 28a9: 4c 8b 65 f0 mov -0x10(%rbp),%r12 > > 28ad: 4c 8b 6d f8 mov -0x8(%rbp),%r13 > > 28b1: c9 leaveq > > 28b2: c3 retq > > > > > > Thanks, > > Sasha > > -- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
commit 13c8bda5154665f4499b31030d74665207025d24 Author: Mathieu Desnoyers <[email protected]> Date: Wed May 7 18:48:53 2014 -0400 Fix: tracepoint: release old probe after rcu assign pointer Signed-off-by: Mathieu Desnoyers <[email protected]> diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c index ac5b23c..6620e58 100644 --- a/kernel/tracepoint.c +++ b/kernel/tracepoint.c @@ -188,7 +188,6 @@ static int tracepoint_add_func(struct tracepoint *tp, WARN_ON_ONCE(1); return PTR_ERR(old); } - release_probes(old); /* * rcu_assign_pointer has a smp_wmb() which makes sure that the new @@ -200,6 +199,7 @@ static int tracepoint_add_func(struct tracepoint *tp, rcu_assign_pointer(tp->funcs, tp_funcs); if (!static_key_enabled(&tp->key)) static_key_slow_inc(&tp->key); + release_probes(old); return 0; } @@ -221,7 +221,6 @@ static int tracepoint_remove_func(struct tracepoint *tp, WARN_ON_ONCE(1); return PTR_ERR(old); } - release_probes(old); if (!tp_funcs) { /* Removed last function */ @@ -232,6 +231,7 @@ static int tracepoint_remove_func(struct tracepoint *tp, static_key_slow_dec(&tp->key); } rcu_assign_pointer(tp->funcs, tp_funcs); + release_probes(old); return 0; }
