perf/bpf: Don't call bpf_overflow_handler() for tracing events
authorKyle Huey <me@kylehuey.com>
Tue, 13 Aug 2024 15:17:27 +0000 (15:17 +0000)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 13 Aug 2024 17:25:28 +0000 (10:25 -0700)
The regressing commit is new in 6.10. It assumed that anytime event->prog
is set bpf_overflow_handler() should be invoked to execute the attached bpf
program. This assumption is false for tracing events, and as a result the
regressing commit broke bpftrace by invoking the bpf handler with garbage
inputs on overflow.

Prior to the regression the overflow handlers formed a chain (of length 0,
1, or 2) and perf_event_set_bpf_handler() (the !tracing case) added
bpf_overflow_handler() to that chain, while perf_event_attach_bpf_prog()
(the tracing case) did not. Both set event->prog. The chain of overflow
handlers was replaced by a single overflow handler slot and a fixed call to
bpf_overflow_handler() when appropriate. This modifies the condition there
to check event->prog->type == BPF_PROG_TYPE_PERF_EVENT, restoring the
previous behavior and fixing bpftrace.

Signed-off-by: Kyle Huey <khuey@kylehuey.com>
Suggested-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Reported-by: Joe Damato <jdamato@fastly.com>
Closes: https://lore.kernel.org/lkml/ZpFfocvyF3KHaSzF@LQ3V64L9R2/
Fixes: f11f10bfa1ca ("perf/bpf: Call BPF handler directly, not through overflow machinery")
Cc: stable@vger.kernel.org
Tested-by: Joe Damato <jdamato@fastly.com> # bpftrace
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20240813151727.28797-1-jdamato@fastly.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
kernel/events/core.c

index aa3450bdc2276cd47d749ac9631600f7963ecfa9..c973e3c11e036f6ed948431f95ebafab5011663f 100644 (file)
@@ -9706,7 +9706,8 @@ static int __perf_event_overflow(struct perf_event *event,
 
        ret = __perf_event_account_interrupt(event, throttle);
 
-       if (event->prog && !bpf_overflow_handler(event, data, regs))
+       if (event->prog && event->prog->type == BPF_PROG_TYPE_PERF_EVENT &&
+           !bpf_overflow_handler(event, data, regs))
                return ret;
 
        /*