FD uses xyarray__entry that may return NULL if an index is out of
bounds. If NULL is returned then a segv happens as FD unconditionally
dereferences the pointer. This was happening in a case of with perf
iostat as shown below. The fix is to make FD an "int*" rather than an
int and handle the NULL case as either invalid input or a closed fd.
$ sudo gdb --args perf stat --iostat list
...
Breakpoint 1, perf_evsel__alloc_fd (evsel=0x5555560951a0, ncpus=1, nthreads=1) at evsel.c:50
50 {
(gdb) bt
#0 perf_evsel__alloc_fd (evsel=0x5555560951a0, ncpus=1, nthreads=1) at evsel.c:50
#1 0x000055555585c188 in evsel__open_cpu (evsel=0x5555560951a0, cpus=0x555556093410,
threads=0x555556086fb0, start_cpu=0, end_cpu=1) at util/evsel.c:1792
#2 0x000055555585cfb2 in evsel__open (evsel=0x5555560951a0, cpus=0x0, threads=0x555556086fb0)
at util/evsel.c:2045
#3 0x000055555585d0db in evsel__open_per_thread (evsel=0x5555560951a0, threads=0x555556086fb0)
at util/evsel.c:2065
#4 0x00005555558ece64 in create_perf_stat_counter (evsel=0x5555560951a0,
config=0x555555c34700 <stat_config>, target=0x555555c2f1c0 <target>, cpu=0) at util/stat.c:590
#5 0x000055555578e927 in __run_perf_stat (argc=1, argv=0x7fffffffe4a0, run_idx=0)
at builtin-stat.c:833
#6 0x000055555578f3c6 in run_perf_stat (argc=1, argv=0x7fffffffe4a0, run_idx=0)
at builtin-stat.c:1048
#7 0x0000555555792ee5 in cmd_stat (argc=1, argv=0x7fffffffe4a0) at builtin-stat.c:2534
#8 0x0000555555835ed3 in run_builtin (p=0x555555c3f540 <commands+288>, argc=3,
argv=0x7fffffffe4a0) at perf.c:313
#9 0x0000555555836154 in handle_internal_command (argc=3, argv=0x7fffffffe4a0) at perf.c:365
#10 0x000055555583629f in run_argv (argcp=0x7fffffffe2ec, argv=0x7fffffffe2e0) at perf.c:409
#11 0x0000555555836692 in main (argc=3, argv=0x7fffffffe4a0) at perf.c:539
...
(gdb) c
Continuing.
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (uncore_iio_0/event=0x83,umask=0x04,ch_mask=0xF,fc_mask=0x07/).
/bin/dmesg | grep -i perf may provide additional information.
Program received signal SIGSEGV, Segmentation fault.
0x00005555559b03ea in perf_evsel__close_fd_cpu (evsel=0x5555560951a0, cpu=1) at evsel.c:166
166 if (FD(evsel, cpu, thread) >= 0)
v3. fixes a bug in perf_evsel__run_ioctl where the sense of a branch was
backward.
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210918054440.2350466-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf tool updates from Arnaldo Carvalho de Melo:
"New features:
- Improvements for the flamegraph python script, including:
- Display perf.data header
- Display PIDs of user stacks
- Added option to change color scheme
- Default to blue/green color scheme to improve accessibility
- Correctly identify kernel stacks when debuginfo is available
- Improvements for 'perf bench futex':
- Add --mlockall parameter
- Add --broadcast and --pi to the 'requeue' sub benchmark
- Add support for PMU aliases.
- Introduce an ARM Coresight ETE decoder.
- Add a 'perf bench' entry for evlist open/close operations, to help
quantify improvements with multithreading 'perf record'.
- Allow reporting the [un]throttle PERF_RECORD_ meta event in 'perf
script's python scripting.
- Add a 'perf test' entry for PMU aliases.
- Add a 'perf test' entry for 'perf record/perf report/perf script'
pipe mode.
Fixes:
- perf script dlfilter (API for filtering via dynamically loaded
shared object introduced in v5.14) fixes and a 'perf test' entry
for it.
- Fix get_current_dir_name() compilation on Android.
- Fix issues with asciidoc and double dashes uses.
- Fix memory leaks in the BTF handling code.
- Fix leftover problems in the Documentation from the infrastructure
originally lifted from the git codebase.
- Fix *probe_vfs_getname.sh 'perf test' failures.
- Handle fd gaps in 'perf test's test__dso_data_reopen().
- Make sure to show disasembly warnings for 'perf annotate --stdio'.
- Fix output from pipe to file and vice-versa in 'perf
record/report/script'.
- Correct 'perf data -h' output.
- Fix wrong comm in system-wide mode with 'perf record --delay'.
- Do not allow --for-each-cgroup without cpu in 'perf stat'
- Make 'perf test --skip' work on shell tests.
- Fix libperf's verbose printing.
Misc improvements:
- Preparatory patches for multithreading various 'perf record' phases
(synthesizing, opening, recording, etc).
- Add sparse context/locking annotations in compiler-types.h, also to
help with the multithreading effort.
- Optimize the generation of the arch specific erno tables used in
'perf trace'.
- Optimize libperf's perf_cpu_map__max().
- Improve ARM's CoreSight warnings.
- Report collisions in AUX records.
- Improve warnings for the LLVM 'perf test' entry.
- Improve the PMU events 'perf test' codebase.
- perf test: Do not compare overheads in the zstd comp test
- Better support annotation on ARM.
- Update 'perf trace's cmd string table to decode sys_bpf() first
arg.
Vendor events:
- Add JSON events and metrics for Intel's Ice Lake, Tiger Lake and
Elhart Lake.
- Update JSON eventsand metrics for Intel's Cascade Lake and Sky Lake
servers.
Hardware tracing:
- Improvements for the ARM hardware tracing auxtrace support"
* tag 'perf-tools-for-v5.15-2021-09-04' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (130 commits)
perf tests: Add test for PMU aliases
perf pmu: Add PMU alias support
perf session: Report collisions in AUX records
perf script python: Allow reporting the [un]throttle PERF_RECORD_ meta event
perf build: Report failure for testing feature libopencsd
perf cs-etm: Show a warning for an unknown magic number
perf cs-etm: Print the decoder name
perf cs-etm: Create ETE decoder
perf cs-etm: Update OpenCSD decoder for ETE
perf cs-etm: Fix typo
perf cs-etm: Save TRCDEVARCH register
perf cs-etm: Refactor out ETMv4 header saving
perf cs-etm: Initialise architecture based on TRCIDR1
perf cs-etm: Refactor initialisation of decoder params.
tools build: Fix feature detect clean for out of source builds
perf evlist: Add evlist__for_each_entry_from() macro
perf evsel: Handle precise_ip fallback in evsel__open_cpu()
perf evsel: Move bpf_counter__install_pe() to success path in evsel__open_cpu()
perf evsel: Move test_attr__open() to success path in evsel__open_cpu()
perf evsel: Move ignore_missing_thread() to fallback code
...
When attaching to uprobes through perf subsystem, it's possible to specify
offset of a so-called USDT semaphore, which is just a reference counted u16,
used by kernel to keep track of how many tracers are attached to a given
location. Support for this feature was added in [0], so just wire this through
uprobe_opts. This is important to enable implementing USDT attachment and
tracing through libbpf's bpf_program__attach_uprobe_opts() API.
[0] a6ca88b241 ("trace_uprobe: support reference counter in fd-based uprobe")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210815070609.987780-16-andrii@kernel.org
Wire through bpf_cookie for all attach APIs that use perf_event_open under the
hood:
- for kprobes, extend existing bpf_kprobe_opts with bpf_cookie field;
- for perf_event, uprobe, and tracepoint APIs, add their _opts variants and
pass bpf_cookie through opts.
For kernel that don't support BPF_LINK_CREATE for perf_events, and thus
bpf_cookie is not supported either, return error and log warning for user.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210815070609.987780-12-andrii@kernel.org
Add ability to specify bpf_cookie value when creating BPF perf link with
bpf_link_create() low-level API.
Given BPF_LINK_CREATE command is growing and keeps getting new fields that are
specific to the type of BPF_LINK, extend libbpf side of bpf_link_create() API
and corresponding OPTS struct to accomodate such changes. Add extra checks to
prevent using incompatible/unexpected combinations of fields.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210815070609.987780-11-andrii@kernel.org
Detect kernel support for BPF perf link and prefer it when attaching to
perf_event, tracepoint, kprobe/uprobe. Underlying perf_event FD will be kept
open until BPF link is destroyed, at which point both perf_event FD and BPF
link FD will be closed.
This preserves current behavior in which perf_event FD is open for the
duration of bpf_link's lifetime and user is able to "disconnect" bpf_link from
underlying FD (with bpf_link__disconnect()), so that bpf_link__destroy()
doesn't close underlying perf_event FD.When BPF perf link is used, disconnect
will keep both perf_event and bpf_link FDs open, so it will be up to
(advanced) user to close them. This approach is demonstrated in bpf_cookie.c
selftests, added in this patch set.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210815070609.987780-10-andrii@kernel.org
bpf_link->destroy() isn't used by any code, so remove it. Instead, add ability
to override deallocation procedure, with default doing plain free(link). This
is necessary for cases when we want to "subclass" struct bpf_link to keep
extra information, as is the case in the next patch adding struct
bpf_link_perf.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20210815070609.987780-9-andrii@kernel.org
Ensure libbpf.so is re-built whenever libbpf.map is modified. Without this,
changes to libbpf.map are not detected and versioned symbols mismatch error
will be reported until `make clean && make` is used, which is a suboptimal
developer experience.
Fixes: 306b267cb3 ("libbpf: Verify versioned symbols")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20210815070609.987780-8-andrii@kernel.org
Currently weak typeless ksyms have default value zero, when they don't
exist in the kernel. However, weak typed ksyms are rejected by libbpf
if they can not be resolved. This means that if a bpf object contains
the declaration of a nonexistent weak typed ksym, it will be rejected
even if there is no program that references the symbol.
Nonexistent weak typed ksyms can also default to zero just like
typeless ones. This allows programs that access weak typed ksyms to be
accepted by verifier, if the accesses are guarded. For example,
extern const int bpf_link_fops3 __ksym __weak;
/* then in BPF program */
if (&bpf_link_fops3) {
/* use bpf_link_fops3 */
}
If actual use of nonexistent typed ksym is not guarded properly,
verifier would see that register is not PTR_TO_BTF_ID and wouldn't
allow to use it for direct memory reads or passing it to BPF helpers.
Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210812003819.2439037-1-haoluo@google.com
Before this patch, btf_new() was liable to close an arbitrary FD 0 if
BTF parsing failed. This was because:
* btf->fd was initialized to 0 through the calloc()
* btf__free() (in the `done` label) closed any FDs >= 0
* btf->fd is left at 0 if parsing fails
This issue was discovered on a system using libbpf v0.3 (without
BTF_KIND_FLOAT support) but with a kernel that had BTF_KIND_FLOAT types
in BTF. Thus, parsing fails.
While this patch technically doesn't fix any issues b/c upstream libbpf
has BTF_KIND_FLOAT support, it'll help prevent issues in the future if
more BTF types are added. It also allow the fix to be backported to
older libbpf's.
Fixes: 3289959b97 ("libbpf: Support BTF loading and raw data output in both endianness")
Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/5969bb991adedb03c6ae93e051fd2a00d293cf25.1627513670.git.dxu@dxuuu.xyz
Add two new APIs: btf__load_vmlinux_btf and btf__load_module_btf.
btf__load_vmlinux_btf is just an alias to the existing API named
libbpf_find_kernel_btf, rename to be more precisely and consistent
with existing BTF APIs. btf__load_module_btf can be used to load
module BTF, add it for completeness. These two APIs are useful for
implementing tracing tools and introspection tools. This is part
of the effort towards libbpf 1.0 ([0]).
[0] Closes: https://github.com/libbpf/libbpf/issues/280
Signed-off-by: Hengqi Chen <hengqi.chen@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210730114012.494408-1-hengqi.chen@gmail.com
Variable "err" is initialised to -EINVAL so that this error code is
returned when something goes wrong in libbpf_find_prog_btf_id().
However, a recent change in the function made use of the variable in
such a way that it is set to 0 if retrieving linear information on the
program is successful, and this 0 value remains if we error out on
failures at later stages.
Let's fix this by setting err to -EINVAL later in the function.
Fixes: e9fc3ce99b ("libbpf: Streamline error reporting for high-level APIs")
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210729162028.29512-2-quentin@isovalent.com
When loading in parallel multiple programs which use the same to-be
pinned map, it is possible that two instances of the loader will call
bpf_object__create_maps() at the same time. If the map doesn't exist
when both instances call bpf_object__reuse_map(), then one of the
instances will fail with EEXIST when calling bpf_map__pin().
Fix the race by retrying reusing a map if bpf_map__pin() returns
EEXIST. The fix is similar to the one in iproute2: e4c4685fd6e4 ("bpf:
Fix race condition with map pinning").
Before retrying the pinning, we don't do any special cleaning of an
internal map state. The closer code inspection revealed that it's not
required:
- bpf_object__create_map(): map->inner_map is destroyed after a
successful call, map->fd is closed if pinning fails.
- bpf_object__populate_internal_map(): created map elements is
destroyed upon close(map->fd).
- init_map_slots(): slots are freed after their initialization.
Signed-off-by: Martynas Pumputis <m@lambda.lt>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20210726152001.34845-1-m@lambda.lt
