As with evlist this a no-op for most of the perf tool. The reference
count is set to 1 at allocation, the put will see the 1, decrement it
and perform the delete.
The purpose for adding the reference count is for the python code. Prior
to this change the python code would clone evsels, but this has issues
if events are opened, etc. leading to assertion failures.
With a reference count the same evsel can be used and the reference
count incremented for the python usage. To not change the python evsel
API getset functions are added for the evsel members, no set function is
provided for size as it doesn't make sense to alter this.
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alice Rogers <alice.mei.rogers@gmail.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
This a no-op for most of the perf tool. The reference count is set to
1 at allocation, the put will see the 1, decrement it and perform the
delete.
The purpose for adding the reference count is for the python code. Prior
to this change the python code would clone evlists, but this has issues
if events are opened, etc.
This change adds a reference count for the evlists and a later change
will add it to evsels. The combination is needed for the python code to
operate correctly (not hit asserts in the evsel clone), but the changes
are broken apart for the sake of smaller patches.
Assisted-by: Gemini:gemini-3.1-pro-preview
Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alice Rogers <alice.mei.rogers@gmail.com>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
machine__init() always returns 0 even when memory allocation fails,
because commit 81f981d7ec ("perf machine: Free root_dir in
machine__init() error path") introduced 'int err = -ENOMEM' and an
error cleanup path but left the final 'return 0' instead of
'return err'.
Fix by returning err, check the return value in __machine__new_host()
which was ignoring it, and change machines__init() from void to int so
it too can propagate the error to perf_session__new(), aslr_tool__init()
and test callers.
The error cleanup also used zfree(&machine->kmaps), but kmaps is a
refcounted maps structure — use maps__zput() to properly drop the
reference, matching machine__exit().
Move dsos__init() and threads__init() before the first fallible
allocation (maps__new) so that machine__exit() is safe to call on
any machine struct that machine__init() touched, even on early failure.
Fixes: 81f981d7ec ("perf machine: Free root_dir in machine__init() error path")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a helper that lazily computes the e_machine and falls back to EM_HOST.
Use the perf_env's arch to compute the e_machine if available, using a
binary search for efficiency while handling duplicate rules.
Switch perf_env__arch to be derived from e_machine for consistency.
To support 32-bit compat binaries on 64-bit hosts during dynamic local
or live operations, unpopulated arch fallback paths query uname() at
runtime to dynamically resolve the correct host e_machine, safely
preventing bitness misclassification regressions.
Update session and header to use the helper to safely record e_machine
and flags without forcing premature thread scanning.
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Honglei Wang <jameshongleiwang@126.com>
Cc: Jan Polensky <japo@linux.ibm.com>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add 'at offset %#<hex>' to all warning and error messages in session.c
that fire when events are skipped or processing stops due to validation
failures. This lets users cross-reference with 'perf report -D' output
to inspect the surrounding records and understand the corruption context.
Covers messages in perf_session__process_event() (alignment, min size,
swap failure), perf_session__deliver_event() (no evsel, parse failure,
CPU clamping), machines__deliver_event() (NAMESPACES, TEXT_POKE,
null-terminated string checks for MMAP/MMAP2/COMM/CGROUP/KSYMBOL), and
perf_session__process_user_event() (THREAD_MAP, CPU_MAP, STAT_CONFIG,
BPF_METADATA, HEADER_BUILD_ID).
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a file_offset field to struct perf_sample so that event processing
callbacks can report the byte offset of the problematic event in
perf.data, letting users cross-reference with 'perf report -D' output.
Set sample.file_offset in perf_session__deliver_event(), which is the
common entry point for both file mode (mmap'd offset) and pipe mode
(running byte counter from __perf_session__process_pipe_events).
The assignment is placed after evsel__parse_sample(), which zeroes
the struct via memset.
Preserve file_offset through the deferred callchain delivery path by
storing it in struct deferred_event and restoring it after
evlist__parse_sample() in both evlist__deliver_deferred_callchain()
and session__flush_deferred_samples().
Subsequent patches will use this field in skip/stop warning messages.
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
On native-endian files, events are read from MAP_SHARED memory.
Multiple reads of event->header.size can return different values
if the file is concurrently modified, allowing an attacker to
bypass bounds checks performed on an earlier read.
Snapshot header.size into a local variable at function entry using
READ_ONCE() to prevent compiler rematerialization, and use it for
all size-dependent arithmetic within the function. This ensures
every bounds calculation uses the same value that was validated
by the reader.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Several downstream consumers (timechart, kwork, sched) use fixed-size
arrays indexed by CPU. A crafted perf.data can supply arbitrary CPU
values that index past these arrays, causing out-of-bounds access.
Validate sample.cpu against min(nr_cpus_avail, MAX_NR_CPUS) in
perf_session__deliver_event() before any tool callback runs. The
cap at MAX_NR_CPUS protects fixed-size downstream arrays; the true
nr_cpus_avail is preserved in env for header parsing (e.g.
process_cpu_topology) which needs the real count.
Fall back to MAX_NR_CPUS when HEADER_NRCPUS is missing (truncated
files, pipe mode, pre-2017 perf).
Only validate when PERF_SAMPLE_CPU is set in sample_type — when
absent, evsel__parse_sample() leaves sample.cpu as (u32)-1, a
sentinel that downstream tools (script, inject) check to identify
events without CPU info. Clamping it to 0 would break those checks.
Inline evlist__parse_sample() into perf_session__deliver_event()
so the evsel lookup needed for sample_type checking reuses the same
evsel that parsed the sample, avoiding a second evlist__event2evsel()
call on every event.
For pipe-mode streams where HEADER_NRCPUS may arrive late or not at
all, the MAX_NR_CPUS fallback ensures the bounds check is still
effective against the fixed-size downstream arrays.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
PERF_RECORD_COMPRESSED2 events carry a data_size field that must be
byte-swapped when reading cross-endian perf.data files. Without a
swap handler, reading COMPRESSED2 events on a different-endian machine
would misinterpret data_size as a garbage value, causing the
decompression path to read the wrong number of bytes.
The compressed payload itself is a raw byte stream and needs no
swapping.
Fixes: 208c0e1683 ("perf record: Add 8-byte aligned event type PERF_RECORD_COMPRESSED2")
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Chun-Tse Shao <ctshao@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
strdup(ev->unit) and strdup(ev->name) read until '\0' with no
guarantee the string is null-terminated within event->header.size.
The dump_trace fprintf path has the same problem with %s.
Validate before either path runs — same class of bug fixed for
MMAP/MMAP2/COMM/CGROUP by perf_event__check_nul().
Also harden the event_update swap handler to:
- Validate SCALE event size before swapping the double at
offset 24, which exceeds the 24-byte min_size.
- Validate CPUS event size before accessing the cpu_map
type/nr/long_size fields, which also start at the min_size
boundary.
- Swap CPUS variant fields (type, nr, long_size) so the
processing path sees native byte order.
Add validation in perf_event__process_event_update() for all
event update variants (UNIT, NAME, SCALE, CPUS) before
dump_trace or processing.
Validate CPUS nr against payload size for both PERF_CPU_MAP__CPUS
and PERF_CPU_MAP__MASK types on the fprintf (dump_trace) path:
- CPUS: check nr does not exceed available cpu entries
- MASK: check nr does not exceed available mask entries for
both mask32 (long_size == 4) and mask64 (long_size == 8)
layouts, with underflow guards on the offsetof subtraction
Fix a missing break before the default case in the CPUS
switch path.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
PERF_RECORD_BPF_METADATA has no entry in perf_event__swap_ops[], so its
nr_entries field is never byte-swapped when reading a cross-endian
perf.data file. Downstream processing in
perf_event__fprintf_bpf_metadata() loops over nr_entries, so a
foreign-endian value causes out-of-bounds reads.
Add a swap handler that byte-swaps nr_entries after validating that
header.size is large enough. The entries[] array contains only char
arrays (key/value strings), so no per-entry swap is needed — but ensure
NUL-termination on the writable cross-endian path.
Validate header.size, nr_entries, and string NUL-termination in the
common event delivery path so that native-endian files with malicious
values are also rejected. Snapshot nr_entries via READ_ONCE() before
validation — the event is on a MAP_SHARED mmap that could theoretically
change between the bounds check and the loop.
Changes in v2:
- Snapshot event->header.size via READ_ONCE() into a local variable
to prevent a double-fetch underflow in the max_entries calculation
(Reported-by: sashiko-bot@kernel.org)
- Write back clamped nr_entries to the event on the swap path,
consistent with NAMESPACES and STAT_CONFIG handlers — without
writeback the native path sees the inflated nr and skips the
event entirely (Reported-by: sashiko-bot@kernel.org)
Fixes: ab38e84ba9 ("perf record: collect BPF metadata from existing BPF programs")
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Blake Jones <blakejones@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fix four issues in PERF_RECORD_AUXTRACE_ERROR handling:
1. auxtrace_error_name() takes a signed int parameter, but e->type
is __u32. A crafted value like 0xFFFFFFFF converts to -1, passes
the bounds check, and causes a negative array index. Fix by
changing the parameter to unsigned int.
2. The msg field is printed via %s without a length bound. The
min_size table only guarantees fields up to msg (offset 48), so
a truncated event has zero msg bytes within the event boundary.
Compute the available msg length from header.size, cap at
sizeof(e->msg), and use %.*s.
3. fmt >= 2 adds machine_pid and vcpu fields after msg[64]. Older
files may have fmt >= 2 but an event size that doesn't include
these fields. Add a size check in the swap handler to downgrade
fmt before the conditional field access, and a matching size
guard in the fprintf path for native-endian events (which are
mmap'd read-only and can't be modified in place).
4. python_process_auxtrace_error() had the same issues: msg was
passed to tuple_set_string() unbounded, and machine_pid/vcpu
were accessed unconditionally without checking fmt or event
size. Apply the same bounds checks.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf_header__read_build_ids() swaps the event header fields for cross-endian
perf.data files but not bev.pid. This causes perf_session__findnew_machine()
to look up the wrong machine for guest VM build IDs, misattributing them.
Swap bev.pid alongside the header fields.
Also add a build_id_swap callback for stream-mode build ID events,
and validate NUL-termination of build_id.filename on the native-endian
delivery path (perf_session__process_user_event) — events with
unterminated filenames are skipped.
Harden perf_header__read_build_ids() against crafted perf.data files:
- Add overflow check on offset + size to prevent wrap past ULLONG_MAX.
- Reject bev.header.size == 0 which would loop forever.
- Reject bev.header.size > remaining section to prevent reading past
the section boundary.
- Guard memcmp(filename, "nel.kallsyms]", 13) with len >= 13 to avoid
reading uninitialized stack memory on short filenames.
- Force NUL-termination of filename before passing it to functions
like machine__findnew_dso() that use strlen/strcmp.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Several event types use an nr field to control iteration over
variable-length arrays. The swap handlers byte-swap and loop using
these fields without bounds checks, and the native processing path
trusts them as well.
Add bounds checks on both paths for:
- PERF_RECORD_THREAD_MAP: validate nr against payload, return -1
on the swap path. On the native path, reject with -EINVAL.
- PERF_RECORD_NAMESPACES: clamp nr on the swap path (safe because
each entry is indexed by type; missing entries just won't be
resolved). Skip the event on the native path.
- PERF_RECORD_CPU_MAP: clamp nr for CPUS and MASK sub-types on
the swap path. Add bounds checks for mask64 which previously
had no nr validation. Skip the event on the native path.
- PERF_RECORD_STAT_CONFIG: clamp nr on the swap path (safe because
each config entry is self-describing via its tag). Skip the
event on the native path.
The swap path (cross-endian, writable MAP_PRIVATE mapping) can
safely clamp by writing back to the event. The native path
(read-only MAP_SHARED mapping) must skip instead of clamping
because writing to the mmap'd event would segfault.
Also fix stat_config swap range: change size += 1 to
size += sizeof(event->stat_config.nr) for clarity. The old +1
happened to work because mem_bswap_64 processes 8-byte chunks,
but the intent is to include the 8-byte nr field in the swap
range.
Changes in v2:
- Document that PERF_RECORD_NAMESPACES max_nr includes trailing
sample_id space when sample_id_all is present — harmless on the
swap path because both per-element bswap_64 and swap_sample_id_all()
perform the same u64 byte swap (Reported-by: sashiko-bot@kernel.org)
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Harden PERF_RECORD_HEADER_ATTR handling against crafted perf.data:
- Validate attr.size: must be >= PERF_ATTR_SIZE_VER0, a multiple
of sizeof(u64), and fit within the event payload.
- Copy only min(attr.size, sizeof(struct perf_event_attr)) bytes
into a local attr, zeroing the rest so legacy files don't leak
adjacent event data into new fields.
- Keep the original attr.size so perf_event__synthesize_attr()
uses it for both allocation and ID-array placement.
Fix perf_event__synthesize_attr() to use attr->size (not the
compiled sizeof) for event allocation and layout, so perf inject
correctly re-synthesizes attrs from files recorded by a different
perf version. Without this, the ID array destination pointer
(computed via perf_record_header_attr_id()) would be inconsistent
with the allocation when attr->size differs from sizeof.
Also fix the parse-no-sample-id-all test to set attr.size, which
is now validated, and improve error handling in read_attr() for
short reads and invalid attr sizes.
Handle ABI0 pipe/inject events where attr.size is 0: use a local
attr_size variable set to PERF_ATTR_SIZE_VER0 for both the bounded
copy and ID array position, instead of writing back to the event.
Native-endian files may be MAP_SHARED (read-only mmap), so writing
to the event buffer would SIGSEGV. The swap path handles ABI0 in
perf_event__attr_swap() which writes to the MAP_PRIVATE copy.
header.size alignment is now validated centrally in
perf_session__process_event() (see "Add minimum event size and
alignment validation").
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
session->time_conv = event->time_conv copies sizeof(struct
perf_record_time_conv) bytes unconditionally, but older kernels
emit shorter TIME_CONV events without the time_cycles, time_mask,
cap_user_time_zero, and cap_user_time_short fields.
For a 32-byte event (the original format), this reads 24 bytes
past the event boundary into adjacent mmap'd data. The garbage
values end up in session->time_conv and can cause incorrect TSC
conversion if cap_user_time_zero happens to be non-zero.
Replace the struct assignment with a bounded memcpy capped at
event->header.size, zeroing the remainder so extended fields
default to off when absent.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Change swap callbacks from void to int return so handlers can
propagate errors. All 28 existing handlers are converted to
return 0 on success, -1 on error. Three new handlers (KSYMBOL,
BPF_EVENT, HEADER_FEATURE) are added returning int from the
start, with sample_id_all handling for the kernel event types.
event_swap() propagates the return to its callers (process_event
and peek_event), which skip events that fail to swap.
Add perf_event__check_nul() for null-termination enforcement
on the common event delivery path for MMAP, MMAP2, COMM,
CGROUP, and KSYMBOL events. Events with
unterminated strings are skipped — native-endian files are
mapped read-only, so writing a NUL byte in place would segfault.
Swap handler hardening:
- Use strnlen bounded by event size (instead of strlen) in
COMM/MMAP/MMAP2/CGROUP swap handlers, returning -1 on
unterminated strings.
- Bounds check text_poke old_len+new_len before computing the
sample_id offset, returning -1 on overflow. Use offsetof()
for the native-path check in machines__deliver_event() since
sizeof() includes struct padding past the flexible array.
- Fix PERF_RECORD_SWITCH sample_id_all: non-CPU_WIDE SWITCH
events have sample_id immediately after the 8-byte header,
not at sizeof(struct perf_record_switch) which is the
CPU_WIDE variant size.
- Fix perf_event__time_conv_swap(): decouple time_cycles and
time_mask into independent per-field event_contains() checks,
so each field is only swapped when the event is large enough
to contain it. The original code guarded both fields under
a single time_cycles check, which would swap time_mask on a
short event that contains time_cycles but not time_mask.
- Handle ABI0 (attr.size == 0) in perf_event__attr_swap()
by substituting PERF_ATTR_SIZE_VER0, so bswap_safe()
correctly swaps VER0 fields instead of skipping everything.
- peek_events: on swap failure, advance past the malformed
entry instead of aborting the loop.
Note: the nr-field bounds checks for namespaces, thread_map,
cpu_map, and stat_config arrays are added by a subsequent
patch ("perf session: Validate nr fields against event size
on both swap and common paths"). The HEADER_ATTR attr.size
validation is added by ("perf session: Validate HEADER_ATTR
attr.size before swapping").
By establishing the int-returning swap infrastructure first,
all subsequent hardening patches can use direct error returns
from day one — no poison values, no workarounds for void return.
Changes in v2:
- peek_events: abort instead of skip for AUXTRACE events on
validation failure — skipping only header.size would land
inside the raw trace payload, causing subsequent iterations
to misparse data as events (Reported-by: sashiko-bot@kernel.org)
Fixes: 9aa0bfa370 ("perf tools: Handle PERF_RECORD_KSYMBOL")
Fixes: 45178a928a ("perf tools: Handle PERF_RECORD_BPF_EVENT")
Fixes: e9def1b2e7 ("perf tools: Add feature header record to pipe-mode")
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Song Liu <songliubraving@fb.com>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
swap_sample_id_all() calls BUG_ON(size % sizeof(u64)) which kills
perf on any event where the sample_id_all tail is not 8-byte aligned.
A crafted perf.data can trigger this trivially.
Replace BUG_ON with a bounds check: skip the swap if the data pointer
is past the end of the event, and only swap when there are bytes
remaining.
Note: the strlen calls in string-field swap handlers (comm,
mmap, mmap2, cgroup) are replaced with bounded strnlen by the
next patch in this series ("perf session: Add validated swap
infrastructure with null-termination checks").
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The kernel dynamically sizes PERF_RECORD_READ based on
attr.read_format: only the fields enabled by PERF_FORMAT_TOTAL_TIME_ENABLED,
PERF_FORMAT_TOTAL_TIME_RUNNING, PERF_FORMAT_ID, and PERF_FORMAT_LOST
are emitted, packed with no gaps.
perf_event__read_swap() unconditionally byte-swapped time_enabled,
time_running, and id at their fixed struct offsets, causing
out-of-bounds access on smaller events and swapping the wrong
bytes when not all format fields are present. It also swapped
sample_id_all at a fixed offset past the full struct, which is
wrong for shorter events.
Replace the individual field swaps with a single mem_bswap_64()
over the entire tail from value onward. Since every field after
pid/tid is u64 regardless of which combination is present, this
correctly handles any read_format combination and any trailing
sample_id_all fields.
Similarly, dump_read() accessed optional fields via fixed struct
offsets, displaying values from wrong positions when not all
format bits are set. Walk the packed u64 array sequentially
instead, with bounds checks against event->header.size.
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf_session__peek_event() computes an event pointer directly from
file_offset when one_mmap is active, without verifying that file_offset
and the subsequent event->header.size fall within the mapped region.
A corrupted perf.data file could cause out-of-bounds memory reads.
Add one_mmap_size to the session struct and validate both the header
and full event fit within the mmap before dereferencing.
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a per-type minimum size table (perf_event__min_size[]) and
enforce it before swap and processing, so that both cross-endian
and native-endian paths are protected from accessing fields past
the event boundary.
The table uses offsetof() for types with trailing variable-length
fields (filenames, strings, msg arrays) and sizeof() for
fixed-size types. Zero entries mean no minimum beyond the 8-byte
header already enforced by the reader.
Undersized events are skipped with a warning in process_event
and rejected in peek_event — both checked before the swap
handler runs, preventing OOB access on crafted event fields.
Also reject events whose header.size is not 8-byte aligned. The
kernel aligns all event sizes to sizeof(u64) — see
perf_event_comm_event() (ALIGN), perf_event_mmap_event(),
perf_event_cgroup(), perf_event_ksymbol() (IS_ALIGNED loops),
and perf_event_text_poke() (ALIGN) in kernel/events/core.c.
An unaligned size means the file is corrupted or crafted; reject
early so downstream code that divides by sizeof(u64) to compute
array element counts gets exact results.
Three legacy user events are exempted from the alignment check:
TRACING_DATA (66) had a 12-byte struct before commit b39c915a4f
("libperf event: Ensure tracing data is multiple of 8 sized")
added padding, COMPRESSED (81) carries raw ZSTD output (already
superseded by COMPRESSED2 with PERF_ALIGN), and HEADER_FEATURE
(80) uses do_write_string() with a 4-byte length prefix.
Also guard event_swap() against crafted event types >=
PERF_RECORD_HEADER_MAX to prevent OOB reads on the
perf_event__swap_ops[] array.
Changes in v2:
- Fix double-skip for unsupported event types: return 0 instead
of event->header.size in perf_session__process_event() for
HEADER_MAX, since reader__read_event() already advances by
event->header.size (Reported-by: sashiko-bot@kernel.org)
- Exempt TRACING_DATA, COMPRESSED, and HEADER_FEATURE from the
alignment check — these legacy user events predate the 8-byte
alignment rule (Reported-by: sashiko-bot@kernel.org)
- peek_event: return 0 (skip) for unknown event types instead of
-1 (error), consistent with process_event which already skips
unsupported types gracefully (Reported-by: sashiko-bot@kernel.org)
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Assisted-by: Claude:claude-opus-4.6-1m
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
use_stdio was associated with struct perf_data and not perf_data_file
meaning there was implicit use of fd rather than fptr that may not be
safe. For example, in perf_data_file__write. Reorganize perf_data_file
to better abstract use_stdio, add kernel-doc and more consistently use
perf_data__ accessors so that use_stdio is better respected.
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Instead of using zalloc(nr_entries * sizeof_entry) that is what calloc()
does.
In some places where linux/zalloc.h isn't needed, remove it, add when
needed and was getting it indirectly.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add the evsel from evsel__parse_sample into the struct
perf_sample. Sometimes we want to alter the evsel associated with a
sample, such as with off-cpu bpf-output events. In general the evsel
and perf_sample are passed as a pair, but this makes an altered evsel
something of a chore to keep checking for and setting up. Later
patches will remove passing an evsel with the perf_sample and switch
to just using the perf_sample's value.
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The deferred stack trace code wasn't using perf_sample__init/exit. Add
the deferred stack trace clean up to perf_sample__exit which requires
proper NULL initialization in perf_sample__init. Make the
perf_sample__exit robust to being called more than once by using
zfree. Make the error paths in evsel__parse_sample exit the
sample. Add a merged_callchain boolean to capture that callchain is
allocated, deferred_callchain doen't suffice for this. Pack the struct
variables to avoid padding bytes for this.
Similiarly powerpc_vpadtl_sample wasn't using perf_sample__init/exit,
use it for consistency and potential issues with uninitialized
variables.
Similarly guest_session__inject_events in builtin-inject wasn't using
perf_sample_init/exit. The lifetime management for fetched events is
somewhat complex there, but when an event is fetched the sample should
be initialized and needs exiting on error. The sample may be left in
place so that future injects have access to it.
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Print log information in ordered event processing so that the cause of
finished round failing is clearer. Print the event name along with its
number when an event isn't processed. Add extra detail about where the
failure happened.
The following log lines come from running `perf data convert`. Before:
0xa250 [0x10]: failed to process type: 80
After:
0xa250 [0x10]: piped event processing failed for event of type: FEATURE (80)
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The index into the cpumap array and the number of entries within the
array can never be negative, so let's make them unsigned. This is
prompted by reports that gcc 13 with -O6 is giving a
alloc-size-larger-than errors. The change makes the cpumap changes and
then updates the declaration of index variables throughout perf and
libperf to be unsigned. The two things are hard to separate as
compiler warnings about mixing signed and unsigned types breaks the
build.
Reported-by: Chingbin Li <liqb365@163.com>
Closes: https://lore.kernel.org/lkml/20260212025127.841090-1-liqb365@163.com/
Tested-by: Chingbin Li <liqb365@163.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Since it is freshly allocated just attribute it to a non-const pointer
and then change it via that pointer.
That way we avoid const-correctness warnings in recent glibc versions.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf_session__fprintf() prints only the host.
This has been changed to print details of host and all guests, by
traversing through the RB-Tree.
These are visible when using high verbosity (-vvvv) in KVM environments,
during perf report dumps.
Testing:
- Test 1: Record the local machine and guest VM using 'perf kvm record' and
generate the report using 'perf kvm report -vvvv -D'. The dump should show
the threads and other details related to local and guest machine.
- 1 Ubuntu VM running on Fedora host
- VM is running a noisy program =>
$ dd if=/dev/urandom of=/dev/null
- On host run =>
$ sudo ./perf kvm --guestvmlinux=/tmp/shared/guest_vmlinux \
--guestkallsyms=/tmp/shared/guest_kallsyms \
--guestmodules=/tmp/shared/guest_modules \
record -a -g -o perf.data.guest
and exit after a few seconds.
[ perf record: Woken up 9 times to write data ]
[ perf record: Captured and wrote 3.150 MB perf.data.guest \
(29311 samples) ]
- Generate dump =>
$ sudo ./perf kvm --guestkallsyms /tmp/shared/guest_kallsyms \
report -vvvv -D -i perf.data.guest > output.txt
- Check for threads associated with guest machine.
$ grep "Thread 0" output.txt
Thread 0 swapper
Thread 0 [guest/0]
PASS
- Test 2: Record the local machine and guest VM using 'perf kvm record' and
generate the report using 'perf kvm report'. The functions running on
guest VM should be seen in the report.
- Same setup as Test 1 but the test looks at the performance profile,
to check if the function names are visible.
- Peek into profile using =>
$ sudo ./perf kvm --guestkallsyms /tmp/shared/guest_kallsyms \
report -i perf.data.guest
- Samples: 29K of event 'cycles', Event count (approx.): 28711693142
Children Self Command Shared Object Symbol
35.69% 35.69% :5820 [guest.kernel.kallsyms] [g] chacha_permute
11.56% 11.56% :5820 [guest.kernel.kallsyms] [g] entry_SYSRETQ_unsXXX
11.12% 11.12% :5820 [guest.kernel.kallsyms] [g] syscall_return_viXXX
7.36% 7.36% :5820 [guest.kernel.kallsyms] [g] entry_SYSCALL_64_XXX
6.07% 6.07% :5820 [guest.kernel.kallsyms] [g] chacha_block_generic
5.40% 5.40% :5820 [guest.kernel.kallsyms] [g] _copy_to_iter
....
PASS
- Test 3: Record the local and 2 guest VMs using 'perf kvm record' and
generate the report using 'perf kvm report -vvvv -D'. The dump should show
the threads and other details related to local and guest machines.
- 1 Ubuntu and 1 Alpine VMs running on Fedora host.
- Find PIDs of qemu instances and use them during record and report
$ pgrep qemu
5816
25098
- Record the activity =>
$ sudo ./perf kvm record -p 5816,25098 -a -g -o perf.data.guests
Warning:
PID/TID switch overriding SYSTEM
[ perf record: Woken up 325927 times to write data ]
[ perf record: Captured and wrote 3.692 MB perf.data.guests \
(57389 samples) ]
- Generate dump =>
$ sudo ./perf kvm report -vvvv -D -i perf.data.guests > output.txt
- Check if the threads related to the local machine and guest VMs
are present =>
$ grep "Thread 0" output.txt
Thread 0 swapper
Thread 0 [guest/0]
NOTE: Threads from Ubuntu and Alpine VMs are bundled together and
appear as one guest machine.
Looking into output.txt =>
Threads: 6
Thread 0 [guest/0]
Thread 5816 :5816
Thread 25098 :25098
Thread 5819 :5819
Thread 5820 :5820
Thread 25103 :25103
To conclude, information is collected for both VMs and not listed
as two different guest machines.
PASS
- Test 4: Check if any guest-related information is printed in
perf annotate. This test is included because the command calls
perf_session__fprintf() in its code path when using -vvvv option.
This could be explained by inability / lack of options for 'perf annotate'
to look into guest VM from host machine, due to no option to specify the
guest's kallsyms or modules. A similar explanation for 'perf mem' could
be used, as perf_session__fprintf() is also present in its code path.
- Run annotate =>
$ sudo ./perf annotate -i perf.data.guest -vvvv > output.txt
- Check for threads from local machine or guest VM =>
$ grep "Thread 0" output.txt
Thread 0 swapper
Threads from local machine are found while threads from guest VM
are not found. It is possibly because of a lack of a guest kallsyms
option for DSO matching in perf annotate.
PASS
- Test 5: Run kvm test available on perf path
- $ sudo ./perf test kvm
89: perf kvm tests : Ok
PASS
Signed-off-by: Hrishikesh Suresh <hrishikesh123s@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Blake Jones <blakejones@google.com>
Cc: Chun-Tse Shao <ctshao@google.com>
Cc: Dmitriy Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
[ Declare 'nd' in the 'for' line and and 'pos' inside the loop body, to make it more compact ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
These are hard to interpret in the raw output because they are printed
as hex but are defined in perf_event.h as decimal. Make it much easier
to read the raw callchains by just printing their names.
For example:
$ perf report -D
1798195372321 0x4638 [0xb0]: PERF_RECORD_SAMPLE(IP, 0x4002): 44922/44922: 0x7c8046dd3400 period: 120218 addr: 0
... FP chain: nr:12
..... 0: fffffffffffffe00 (PERF_CONTEXT_USER)
..... 1: 00007c8046dd3400
..... 2: 00007c8046db86d3
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
[ Add PERF_CONTEXT_USER_DEFERRED too, as per Namhyung's review comment ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It's possible that some kernel samples don't have matching deferred
callchain records when the profiling session was ended before the
threads came back to userspace. Let's flush the samples before
finish the session.
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Save samples with deferred callchains in a separate list and deliver
them after merging the user callchains. If users don't want to merge
they can set tool->merge_deferred_callchains to false to prevent the
behavior.
With previous result, now perf script will show the merged callchains.
$ perf script
...
pwd 2312 121.163435: 249113 cpu/cycles/P:
ffffffff845b78d8 __build_id_parse.isra.0+0x218 ([kernel.kallsyms])
ffffffff83bb5bf6 perf_event_mmap+0x2e6 ([kernel.kallsyms])
ffffffff83c31959 mprotect_fixup+0x1e9 ([kernel.kallsyms])
ffffffff83c31dc5 do_mprotect_pkey+0x2b5 ([kernel.kallsyms])
ffffffff83c3206f __x64_sys_mprotect+0x1f ([kernel.kallsyms])
ffffffff845e6692 do_syscall_64+0x62 ([kernel.kallsyms])
ffffffff8360012f entry_SYSCALL_64_after_hwframe+0x76 ([kernel.kallsyms])
7f18fe337fa7 mprotect+0x7 (/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2)
7f18fe330e0f _dl_sysdep_start+0x7f (/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2)
7f18fe331448 _dl_start_user+0x0 (/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2)
...
The old output can be get using --no-merge-callchain option.
Also perf report can get the user callchain entry at the end.
$ perf report --no-children --stdio -q -S __build_id_parse.isra.0
# symbol: __build_id_parse.isra.0
8.40% pwd [kernel.kallsyms]
|
---__build_id_parse.isra.0
perf_event_mmap
mprotect_fixup
do_mprotect_pkey
__x64_sys_mprotect
do_syscall_64
entry_SYSCALL_64_after_hwframe
mprotect
_dl_sysdep_start
_dl_start_user
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add a new event type for deferred callchains and a new callback for the
struct perf_tool. For now it doesn't actually handle the deferred
callchains but it just marks the sample if it has the PERF_CONTEXT_
USER_DEFFERED in the callchain array.
At least, perf report can dump the raw data with this change. Actually
this requires the next commit to enable attr.defer_callchain, but if you
already have a data file, it'll show the following result.
$ perf report -D
...
0x2158@perf.data [0x40]: event: 22
.
. ... raw event: size 64 bytes
. 0000: 16 00 00 00 02 00 40 00 06 00 00 00 0b 00 00 00 ......@.........
. 0010: 03 00 00 00 00 00 00 00 a7 7f 33 fe 18 7f 00 00 ..........3.....
. 0020: 0f 0e 33 fe 18 7f 00 00 48 14 33 fe 18 7f 00 00 ..3.....H.3.....
. 0030: 08 09 00 00 08 09 00 00 e6 7a e7 35 1c 00 00 00 .........z.5....
121163447014 0x2158 [0x40]: PERF_RECORD_CALLCHAIN_DEFERRED(IP, 0x2): 2312/2312: 0xb00000006
... FP chain: nr:3
..... 0: 00007f18fe337fa7
..... 1: 00007f18fe330e0f
..... 2: 00007f18fe331448
: unhandled!
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Getting context for what a tool is doing, such as the perf_inject
instance, using container_of the tool is a common pattern in the
code. This isn't possible event_op2, event_op3 and event_op4 callbacks
as the tool isn't passed. Add the argument and then fix function
signatures to match. As tools maybe reading a tool from somewhere
else, change that code to use the passed in tool.
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
If a user specifies an AUX buffer larger than 2 GiB, the returned size
may exceed 0x80000000. Since the err variable is defined as a signed
32-bit integer, such a value overflows and becomes negative.
As a result, the perf record command reports an error:
0x146e8 [0x30]: failed to process type: 71 [Unknown error 183711232]
Change the type of the err variable to a signed 64-bit integer to
accommodate large buffer sizes correctly.
Fixes: d5652d865e ("perf session: Add ability to skip 4GiB or more")
Reported-by: Tamas Zsoldos <tamas.zsoldos@arm.com>
Signed-off-by: Leo Yan <leo.yan@arm.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20250808-perf_fix_big_buffer_size-v1-1-45f45444a9a4@arm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
By definition arch sample parsing and synthesis will inhibit certain
kinds of cross-platform record then analysis (report, script,
etc.). Remove arch_perf_parse_sample_weight and
arch_perf_synthesize_sample_weight replacing with a common
implementation. Combine perf_sample p_stage_cyc and retire_lat as
weight3 to capture the differing uses regardless of compiled for
architecture.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250724163302.596743-21-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The global perf_env was used for the host, but if a perf_env wasn't
easy to come by it was used in a lot of places where potentially
recorded and host data could be confused. Remove the global variable
as now the majority of accesses retrieve the perf_env for the host
from the session.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250724163302.596743-20-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When creating a perf_session the host perf_env may or may not want to
be used. For example, `perf top` uses a host perf_env while `perf
inject` does not. Add a host_env argument to perf_session__new so that
sessions requiring a host perf_env can pass it in. Currently if none
is specified the global perf_env variable is used, but this will
change in later patches.
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20250724163302.596743-14-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
