Re: [PATCH -tip 2/2] kprobes: Set unoptimized flag after unoptimizing code

[Alexei Starovoitov]

On Tue, Nov 26, 2019 at 10:49 PM Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
>
> * Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> wrote:
>
> > On Wed, Nov 27, 2019 at 02:57:04PM +0900, Masami Hiramatsu wrote:
> > > Fix to set unoptimized flag after confirming the code is completely
> > > unoptimized. Without this fix, when a kprobe hits the intermediate
> > > modified instruction (the first byte is replaced by int3, but
> > > latter bytes still be a jump address operand) while unoptimizing,
> > > it can return to the middle byte of the modified code. And it causes
> > > an invalid instruction exception in the kernel.
> > >
> > > Usually, this is a rare case, but if we put a probe on the function
> > > called while text patching, it always causes a kernel panic as below.
> > > (text_poke() is used for patching the code in optprobe)
> > >
> > >  # echo p text_poke+5 > kprobe_events
> > >  # echo 1 > events/kprobes/enable
> > >  # echo 0 > events/kprobes/enable
> > >  invalid opcode: 0000 [#1] PREEMPT SMP PTI
> > >  CPU: 7 PID: 137 Comm: kworker/7:1 Not tainted 5.4.0-rc8+ #29
> > >  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
> > >  Workqueue: events kprobe_optimizer
> > >  RIP: 0010:text_poke+0x9/0x50
> > >  Code: 01 00 00 5b 5d 41 5c 41 5d c3 89 c0 0f b7 4c 02 fe 66 89 4c 05 fe e9 31 ff ff ff e8 71 ac 03 00 90 55 48 89 f5 53 cc 30 cb fd <1e> ec 08 8b 05 72 98 31 01 85 c0 75 11 48 83 c4 08 48 89 ee 48 89
> > >  RSP: 0018:ffffc90000343df0 EFLAGS: 00010686
> > >  RAX: 0000000000000000 RBX: ffffffff81025796 RCX: 0000000000000000
> > >  RDX: 0000000000000004 RSI: ffff88807c983148 RDI: ffffffff81025796
> > >  RBP: ffff88807c983148 R08: 0000000000000001 R09: 0000000000000000
> > >  R10: 0000000000000000 R11: 0000000000000000 R12: ffffffff82284fe0
> > >  R13: ffff88807c983138 R14: ffffffff82284ff0 R15: 0ffff88807d9eee0
> > >  FS:  0000000000000000(0000) GS:ffff88807d9c0000(0000) knlGS:0000000000000000
> > >  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > >  CR2: 000000000058158b CR3: 000000007b372000 CR4: 00000000000006a0
> > >  Call Trace:
> > >   arch_unoptimize_kprobe+0x22/0x28
> > >   arch_unoptimize_kprobes+0x39/0x87
> > >   kprobe_optimizer+0x6e/0x290
> > >   process_one_work+0x2a0/0x610
> > >   worker_thread+0x28/0x3d0
> > >   ? process_one_work+0x610/0x610
> > >   kthread+0x10d/0x130
> > >   ? kthread_park+0x80/0x80
> > >   ret_from_fork+0x3a/0x50
> > >  Modules linked in:
> > >  ---[ end trace 83b34b22a228711b ]---
> > >
> > > This can happen even if we blacklist text_poke() and other functions,
> > > because there is a small time window which showing the intermediate
> > > code to other CPUs.
> > >
> > > Fixes: 6274de4984a6 ("kprobes: Support delayed unoptimizing")
> > > Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> >
> > Awesome. It fixes the crash for me.
> > Tested-by: Alexei Starovoitov <ast@xxxxxxxxxx>
>
> Thanks guys - I just pushed out a rebased tree, based on an upstream
> version that has both the BPF tree and most x86 trees merged, into
> tip:WIP.core/kprobes. This includes these two fixes as well.

fwiw. I merged WIP.core/kprobes into my local tree and have been
testing everything with it. No issues found.