2020-03-24 18:12:44

by Stephane Eranian

[permalink] [raw]
Subject: [PATCH] perf/tools: fix perf_evsel__fallback() for paranoid=2 and -b

When perf_event_paranoid=2, regular users are prevented from sampling
at the kernel level. However, it the user passes an event without
a privilege level, the tool will force :u when it detects paranoid>1.
This works well, except when branch sampling is requested. It has a more
stringent requirement especially with exclude_hv.
$ perf record ls
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.001 MB /tmp/perf.data ]

But:
$ perf record -b ls
Error:
You may not have permission to collect stats.

Consider tweaking /proc/sys/kernel/perf_event_paranoid,
which controls use of the performance events system by
unprivileged users (without CAP_SYS_ADMIN).

The current value is 2:

-1: Allow use of (almost) all events by all users
>= 0: Disallow raw tracepoint access by users without CAP_IOC_LOCK
>= 1: Disallow CPU event access by users without CAP_SYS_ADMIN
>= 2: Disallow kernel profiling by users without CAP_SYS_ADMIN

To make this setting permanent, edit /etc/sysctl.conf too, e.g.:

kernel.perf_event_paranoid = -1

The problem is that in the fallback cod only exclude_kernel is checked and
if set, then exclude_hv is not forced to 1. When branch sampling is enabled
exclude_hv must be set.

This patch fixes the bug in the fallback code by considering the value of
exclude_hv and not just exclude_kernel. We prefer this approach to give a
chance to exclude_hv=0.

Signed-off-by: Stephane Eranian <[email protected]>
---
tools/perf/util/evsel.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 816d930d774e7..db0e6112992e5 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -2424,7 +2424,8 @@ bool perf_evsel__fallback(struct evsel *evsel, int err,

zfree(&evsel->name);
return true;
- } else if (err == EACCES && !evsel->core.attr.exclude_kernel &&
+ } else if (err == EACCES &&
+ (!evsel->core.attr.exclude_kernel || !evsel->core.attr.exclude_hv) &&
(paranoid = perf_event_paranoid()) > 1) {
const char *name = perf_evsel__name(evsel);
char *new_name;
--
2.25.1.696.g5e7596f4ac-goog


2020-03-25 11:09:50

by Jiri Olsa

[permalink] [raw]
Subject: Re: [PATCH] perf/tools: fix perf_evsel__fallback() for paranoid=2 and -b

On Tue, Mar 24, 2020 at 11:10:20AM -0700, Stephane Eranian wrote:
> When perf_event_paranoid=2, regular users are prevented from sampling
> at the kernel level. However, it the user passes an event without
> a privilege level, the tool will force :u when it detects paranoid>1.
> This works well, except when branch sampling is requested. It has a more
> stringent requirement especially with exclude_hv.
> $ perf record ls
> [ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 0.001 MB /tmp/perf.data ]
>
> But:
> $ perf record -b ls
> Error:
> You may not have permission to collect stats.
>
> Consider tweaking /proc/sys/kernel/perf_event_paranoid,
> which controls use of the performance events system by
> unprivileged users (without CAP_SYS_ADMIN).
>
> The current value is 2:
>
> -1: Allow use of (almost) all events by all users
> >= 0: Disallow raw tracepoint access by users without CAP_IOC_LOCK
> >= 1: Disallow CPU event access by users without CAP_SYS_ADMIN
> >= 2: Disallow kernel profiling by users without CAP_SYS_ADMIN
>
> To make this setting permanent, edit /etc/sysctl.conf too, e.g.:
>
> kernel.perf_event_paranoid = -1
>
> The problem is that in the fallback cod only exclude_kernel is checked and
> if set, then exclude_hv is not forced to 1. When branch sampling is enabled
> exclude_hv must be set.
>
> This patch fixes the bug in the fallback code by considering the value of
> exclude_hv and not just exclude_kernel. We prefer this approach to give a
> chance to exclude_hv=0.
>
> Signed-off-by: Stephane Eranian <[email protected]>

Acked-by: Jiri Olsa <[email protected]>

thanks,
jirka

> ---
> tools/perf/util/evsel.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 816d930d774e7..db0e6112992e5 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -2424,7 +2424,8 @@ bool perf_evsel__fallback(struct evsel *evsel, int err,
>
> zfree(&evsel->name);
> return true;
> - } else if (err == EACCES && !evsel->core.attr.exclude_kernel &&
> + } else if (err == EACCES &&
> + (!evsel->core.attr.exclude_kernel || !evsel->core.attr.exclude_hv) &&
> (paranoid = perf_event_paranoid()) > 1) {
> const char *name = perf_evsel__name(evsel);
> char *new_name;
> --
> 2.25.1.696.g5e7596f4ac-goog
>