Messages in this thread Patch in this message | | | From | Song Liu <> | Subject | Re: perf: fuzzer leads to trace_kprobe: Could not insert message flood | Date | Wed, 11 Apr 2018 18:02:37 +0000 |
| |
> On Apr 11, 2018, at 5:04 AM, Ingo Molnar <mingo@kernel.org> wrote: > > > * Song Liu <songliubraving@fb.com> wrote: > >> >> >>> On Apr 10, 2018, at 7:48 AM, Vince Weaver <vincent.weaver@maine.edu> wrote: >>> >>> Author: Song Liu <songliubraving@fb.com> >>> Date: Wed Dec 6 14:45:15 2017 -0800 >>> >>> When running the perf_fuzzer on a current git checkout my logs are flooded >>> with messages such as this: >>> [71487.869077] trace_kprobe: Could not insert probe at unknown+0: -22 >>> [71488.174479] trace_kprobe: Could not insert probe at unknown+0: -22 >>> >>> Presumably this is due to the introduction of the perf_kprobe PMU in >>> commit e12f03d7031a977356e3d7b75a68c2185ff8d155 >>> Author: Song Liu <songliubraving@fb.com> >>> Date: Wed Dec 6 14:45:15 2017 -0800 >>> >>> Is there a way to get this error disabled, or else rate-limited? >>> >>> Vince >> >> Hi Vince, >> >> Thanks for the report. >> >> This is a new API that creates probe together with perf_event_open(). Based on >> my limited understanding of perf_fuzzer, it doesn't understand this API, and uses >> it in an abnormal way. [...] > > Vince's point is valid: the kernel log should not be flooded with pointless > messages as a response to user-space ABI uses ... > > Why is there a kernel log message at all, isn't an error returned? > >> [...] I would recommend perf_fuzzer to understand this new API and test it. >> [...] > > This bug needs to be fixed: a new API must not effectively DoS fuzzing efforts by > spamming the kernel log ...
Yeah, the new API allows non-root user to trigger this message. We should only allow root to create kprobe with perf_event_open().
On the other hand, do we need to fix this for root? In fact, a simple bash loop can create something similar through the text interface (with root):
root@virt-test:~# for x in {0..5} ; do echo p:xx xx+$x >> /sys/kernel/debug/tracing/kprobe_events ; done -bash: echo: write error: No such file or directory -bash: echo: write error: No such file or directory -bash: echo: write error: No such file or directory -bash: echo: write error: No such file or directory -bash: echo: write error: No such file or directory -bash: echo: write error: No such file or directory root@virt-test:~# dmesg | tail -n 5 [ 664.208374] trace_kprobe: Could not insert probe at xx+1: -2 [ 664.237882] trace_kprobe: Could not insert probe at xx+2: -2 [ 664.268067] trace_kprobe: Could not insert probe at xx+3: -2 [ 664.297395] trace_kprobe: Could not insert probe at xx+4: -2 [ 664.327614] trace_kprobe: Could not insert probe at xx+5: -2
This happens before the new API is introduced.
The following patch does capable(CAP_SYS_ADMIN) for perf_kprobe and perf_uprobe at an earlier stage, so non-root user cannot trigger this error message. Please let me know whether we need to fix this for root.
Thanks, Song
From c6708e9e3cd5ba7afb5a7f693b04abf64fec031e Mon Sep 17 00:00:00 2001 From: Song Liu <songliubraving@fb.com> Date: Wed, 11 Apr 2018 10:37:00 -0700 Subject: [PATCH] perf: need CAP_SYS_ADMIN to create k/uprobe with perf_event_open()
Non-root user cannot create kprobe or uprobe through the text-based interface (kprobe_events, uprobe_events). So they cannot create the probes with perf_event_open(). To ensure this, we check capable(CAP_SYS_ADMIN) at perf_[k,u]probe_event_init().
Fixes: e12f03d7031a ("perf/core: Implement the 'perf_kprobe' PMU") Fixes: 33ea4b24277b ("perf/core: Implement the 'perf_uprobe' PMU")
Signed-off-by: Song Liu <songliubraving@fb.com> Reported-by: Vince Weaver <vincent.weaver@maine.edu> Cc: Ingo Molnar <mingo@kernel.org> --- kernel/events/core.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/kernel/events/core.c b/kernel/events/core.c index d7af828..2d5fe26 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -8400,6 +8400,10 @@ static int perf_kprobe_event_init(struct perf_event *event)
if (event->attr.type != perf_kprobe.type) return -ENOENT; + + if (!capable(CAP_SYS_ADMIN)) + return -EACCES; + /* * no branch sampling for probe events */ @@ -8437,6 +8441,10 @@ static int perf_uprobe_event_init(struct perf_event *event)
if (event->attr.type != perf_uprobe.type) return -ENOENT; + + if (!capable(CAP_SYS_ADMIN)) + return -EACCES; + /* * no branch sampling for probe events */ -- 2.9.5
| |