Skip to contentSkip to conversation

perf tools: Assorted fixes

PATCH Superseded Arnaldo Carvalho de Melo 2 participants 18 messages Jun 7 perf events

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Arnaldo Carvalho de Melo

author maintainer
#0 lore raw

cover perf tools: Assorted fixes

Hi,

Sixth batch of pre-existing bug fixes found by sashiko-bot AI review during the perf-data-validation hardening series. All bugs are latent in existing code — none were introduced by the hardening patches.

Three broad categories:

1. snprintf() accumulation overflows (patches 2, 9, 10, 11):

Several functions accumulate formatted output via ret += snprintf(). snprintf() returns the would-have-been-written count, so on truncation ret overshoots the buffer size and the next 'size - ret' underflows to a huge unsigned value, disabling bounds checking. Switched to scnprintf() which returns actual bytes written.

Affected: cpu_map__snprint(), snprintf_hex(), synthesize_bpf_prog_name(), hists__scnprintf_title(), build_id__snprintf(), hwmon_pmu__describe_items().

2. Missing safety checks on untrusted data (patches 1, 3, 6, 7):

- get_max_num(): size_t underflow on empty sysfs file causes heap over-read.
- machine__resolve(): unguarded env->cpu[] access with untrusted CPU index — switched to perf_env__get_cpu_topology() accessor.
- timehist: test_bit(prio, ...) without bounds check on untrusted tracepoint priority.
- idle-hist: rb_first_cached() on a tree populated with plain rb_insert_color() — rb_leftmost never set, callchains silently dropped.

3. Resource hygiene (patches 4, 5, 8):

- mbind() bitmap allocation one bit short of what kernel reads.
- bitmap_free() without NULLing the pointer (3 call sites).
- O_CLOEXEC missing from open() calls in DSO and ELF code (12 call sites across 2 files).

Also expanded the libperf ABI TODO (tools/lib/perf/TODO) to emphasize the code simplification argument for widening struct perf_cpu.cpu from int16_t to int — the narrow type forces defensive truncation checks at every boundary where wider CPU indices are narrowed.

 tools/lib/perf/TODO          |  7 +++++++
 tools/perf/builtin-record.c  |  1 +
 tools/perf/builtin-sched.c   |  7 +++++--
 tools/perf/util/bpf-event.c  | 11 ++++++-----
 tools/perf/util/build-id.c   |  2 +-
 tools/perf/util/cpumap.c     | 24 +++++++++++++++---------
 tools/perf/util/dso.c        |  4 ++--
 tools/perf/util/event.c      |  9 +++++++--
 tools/perf/util/header.c     |  4 +++-
 tools/perf/util/hist.c       |  7 ++++---
 tools/perf/util/hwmon_pmu.c  | 12 ++++++------
 tools/perf/util/mmap.c       |  4 +++-
 tools/perf/util/symbol-elf.c | 20 ++++++++++----------
 13 files changed, 70 insertions(+), 42 deletions(-)

Thanks,

- Arnaldo

author maintainer
#1 lore raw

PATCH 1/11 perf tools: Fix get_max_num() size_t underflow on empty sysfs file

From: Arnaldo Carvalho de Melo <[email protected]>

get_max_num() reads a sysfs file (cpu/possible, cpu/present, or node/possible) and scans backward from the end to find the last number. If the file is empty, filename__read_str() returns num == 0. The loop while (--num) decrements the size_t from 0 to SIZE_MAX, reading backward across the heap until a comma or hyphen is found or unmapped memory is hit.

Add an early return for empty files before the backward scan.

Fixes 7780c25bae59fd04 ("perf tools: Allow ability to map cpus to nodes easily")
Cc Don Zickus <[email protected]>
Cc Ian Rogers <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/cpumap.c | 6 ++++++
 1 file changed, 6 insertions(+)
tools/perf/util/cpumap.c +6 -0
index 21fa781b03cc7409..1fab00ec4a59a0c7 100644
@@ -448,6 +448,12 @@ static int get_max_num(char *path, int *max)
 
 	buf[num] = '\0';
 
+	/* empty file — nothing to parse */
+	if (num == 0) {
+		err = -1;
+		goto out;
+	}
+
 	/* start on the right, to find highest node num */
 	while (--num) {
 		if ((buf[num] == ',') || (buf[num] == '-')) {
Signature · 1 line
2.54.0
#2 lore raw

Re: [PATCH 01/11] perf tools: Fix get_max_num() size_t underflow on empty sysfs file

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Pre-existing issues:
- [High] Heap buffer underflow due to unsafe length manipulation after filename__read_str on empty PMU identifier file.
- [High] Heap buffer underflow due to unsafe length manipulation after procfs__read_str on empty procfs file.
- [High] Reading an empty PMU scale file leads to a stack buffer underflow.
- [High] Reading an empty PMU unit file leads to an out-of-bounds array access inside the perf_pmu_alias struct. --

commit 3cacfb3c562f81ba051910331484d58d28297996
Author: Arnaldo Carvalho de Melo <[email protected]>

perf tools: Fix get_max_num() size_t underflow on empty sysfs file

This commit fixes a heap buffer underflow in get_max_num() by adding an early return when an empty sysfs file is read. This prevents the size_t variable from underflowing during a backward scan of the buffer.

Quoted 5 lines
tools/perf/util/cpumap.c +0 -0
index 21fa781b03cc7..1fab00ec4a59a 100644

tools/perf/util/cpumap.c static int get_max_num(char *path, int *max) · line 456

Quoted 9 lines
tools/perf/util/cpumap.c +6 -0
@@ -448,6 +448,12 @@ static int get_max_num(char *path, int *max)
 
 	buf[num] = '\0';
 
+	/* empty file — nothing to parse */
+	if (num == 0) {
+		err = -1;
+		goto out;
+	}
+

[Severity: High]
This isn't a bug introduced by this patch, but I noticed a similar pattern in tools/perf/util/pmu.c:pmu_id(). Since filename__read_str() returns success (0) and sets len to 0 when reading an empty file, does the following code underflow the len variable?

tools/perf/util/pmu.c:pmu_id() {
...

	if (filename__read_str(path, &str, &len) < 0)
		return NULL;

	str[len - 1] = 0; /* remove line feed */

...
}

[Severity: High]
This is a pre-existing issue, but another instance seems to exist in tools/perf/util/thread.c:thread__set_comm_from_proc(). If the procfs file is empty (e.g., due to a race), procfs__read_str() returns 0 and sets sz to 0. Could this lead to a heap buffer underflow when setting the NUL byte?

tools/perf/util/thread.c:thread__set_comm_from_proc() {
...

	if (!(snprintf(path, sizeof(path), "%d/task/%d/comm",
		       thread__pid(thread), thread__tid(thread)) >= (int)sizeof(path)) &&
	    procfs__read_str(path, &comm, &sz) == 0) {
		comm[sz - 1] = '\0';

...
}

[Severity: High]
This isn't a bug introduced by this patch, but reading an empty PMU scale file might cause a stack buffer underflow in tools/perf/util/pmu.c:perf_pmu__parse_scale().

If the file is empty, read() returns 0, setting sret to 0. The error check passes, and we evaluate scale[-1]. Could this read uninitialized stack memory and potentially write a NUL byte out-of-bounds?

tools/perf/util/pmu.c:perf_pmu__parse_scale() {
...

	sret = read(fd, scale, sizeof(scale)-1);
	if (sret < 0)
		goto error;

	if (scale[sret - 1] == '\n')
		scale[sret - 1] = '\0';

...
}

[Severity: High]
This is a pre-existing issue, but there appears to be a similar out-of-bounds access in tools/perf/util/pmu.c:perf_pmu__parse_unit().

If read() returns 0 for an empty file, sret is set to 0. The code then evaluates alias->unit[-1]. Since unit is part of the heap-allocated struct perf_pmu_alias, could this overwrite the adjacent pmu_name pointer field if the preceding byte happens to be a newline?

tools/perf/util/pmu.c:perf_pmu__parse_unit() {
...

	sret = read(fd, alias->unit, UNIT_MAX_LEN);
	if (sret < 0)
		goto error;

	close(fd);

	if (alias->unit[sret - 1] == '\n')
		alias->unit[sret - 1] = '\0';

...
}

Quoted 3 lines
tools/perf/util/cpumap.c +0 -0
 	/* start on the right, to find highest node num */
 	while (--num) {
 		if ((buf[num] == ',') || (buf[num] == '-')) {
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
author maintainer
#3 lore raw

PATCH 2/11 perf tools: Use scnprintf() in cpu_map__snprint() to prevent overflow

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

cpu_map__snprint() accumulates snprintf() return values in ret. snprintf() returns the number of characters that *would have been written* on truncation, not the actual count. When a fragmented CPU list exceeds the buffer, ret grows past size, causing size - ret to underflow (both are size_t), and subsequent snprintf() calls write past the end of the caller's stack buffer.

Switch to scnprintf() which returns the actual number of characters written, making ret accumulation safe by construction.

Fixes a24020e6b7cf6eb8 ("perf tools: Change cpu_map__fprintf output")
Cc Jiri Olsa <[email protected]>
Cc Ian Rogers <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/cpumap.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)
tools/perf/util/cpumap.c +9 -9
index 1fab00ec4a59a0c7..23ebe9b97f8e58af 100644
@@ -692,21 +692,21 @@ size_t cpu_map__snprint(struct perf_cpu_map *map, char *buf, size_t size)
 		if (start == -1) {
 			start = i;
 			if (last) {
-				ret += snprintf(buf + ret, size - ret,
-						"%s%d", COMMA,
-						perf_cpu_map__cpu(map, i).cpu);
+				ret += scnprintf(buf + ret, size - ret,
+						 "%s%d", COMMA,
+						 perf_cpu_map__cpu(map, i).cpu);
 			}
 		} else if (((i - start) != (cpu.cpu - perf_cpu_map__cpu(map, start).cpu)) || last) {
 			int end = i - 1;
 
 			if (start == end) {
-				ret += snprintf(buf + ret, size - ret,
-						"%s%d", COMMA,
-						perf_cpu_map__cpu(map, start).cpu);
+				ret += scnprintf(buf + ret, size - ret,
+						 "%s%d", COMMA,
+						 perf_cpu_map__cpu(map, start).cpu);
 			} else {
-				ret += snprintf(buf + ret, size - ret,
-						"%s%d-%d", COMMA,
-						perf_cpu_map__cpu(map, start).cpu, perf_cpu_map__cpu(map, end).cpu);
+				ret += scnprintf(buf + ret, size - ret,
+						 "%s%d-%d", COMMA,
+						 perf_cpu_map__cpu(map, start).cpu, perf_cpu_map__cpu(map, end).cpu);
 			}
 			first = false;
 			start = i;
Signature · 1 line
2.54.0
author maintainer
#4 lore raw

PATCH 3/11 perf tools: Use perf_env__get_cpu_topology() in machine__resolve()

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

machine__resolve() accesses env->cpu[al->cpu].socket_id after checking al->cpu >= 0 and env->cpu != NULL, but without validating al->cpu against env->nr_cpus_avail. Since al->cpu comes from the untrusted perf.data sample, a crafted file with a large CPU index causes an out-of-bounds heap read.

Use perf_env__get_cpu_topology() which validates both NULL and bounds.

Fixes 0c4c4debb0adda4c ("perf tools: Add processor socket info to hist_entry and addr_location")
Cc Kan Liang <[email protected]>
Cc Ian Rogers <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/event.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
tools/perf/util/event.c +7 -2
index 66f4843bb235df53..66293fea64fde9fd 100644
@@ -14,6 +14,7 @@
 #include <linux/perf_event.h>
 #include "cpumap.h"
 #include "dso.h"
+#include "env.h"
 #include "event.h"
 #include "debug.h"
 #include "hist.h"
@@ -835,9 +836,13 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
 
 	if (al->cpu >= 0) {
 		struct perf_env *env = machine->env;
+		struct cpu_topology_map *topo;
 
-		if (env && env->cpu)
-			al->socket = env->cpu[al->cpu].socket_id;
+		if (env) {
+			topo = perf_env__get_cpu_topology(env, (struct perf_cpu){ al->cpu });
+			if (topo)
+				al->socket = topo->socket_id;
+		}
 	}
 
 	/* Account for possible out-of-order switch events. */
Signature · 1 line
2.54.0
author maintainer
#5 lore raw

PATCH 4/11 perf mmap: Fix mbind() maxnode vs bitmap allocation mismatch in aio_bind

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

perf_mmap__aio_bind() allocates a node mask bitmap with bitmap_zalloc(node_index + 1) bits, but passes node_index + 2 as the maxnode argument to mbind(). The mbind syscall interprets maxnode as the number of bits to read from the mask.

When node_index + 2 crosses a BITS_PER_LONG boundary (e.g. node_index = 63 on 64-bit), the bitmap occupies 8 bytes but mbind reads 16 — an out-of-bounds read of user heap memory into kernel space.

Allocate node_index + 2 bits to match what mbind will actually read.

Fixes 44d462acc0bf3eab ("perf record: Fix binding of AIO user space buffers to nodes")
Cc Alexey Budankov <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/mmap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/perf/util/mmap.c +2 -1
index d64aec6c7c843e81..8012301d3cf2ac9a 100644
@@ -113,7 +113,8 @@ static int perf_mmap__aio_bind(struct mmap *map, int idx, struct perf_cpu cpu, i
 		if (node < 0)
 			return 0;
 		node_index = node;
-		node_mask = bitmap_zalloc(node_index + 1);
+		/* mbind's maxnode is node_index + 2 — allocate to match */
+		node_mask = bitmap_zalloc(node_index + 2);
 		if (!node_mask) {
 			pr_err("Failed to allocate node mask for mbind: error %m\n");
 			return -1;
Signature · 1 line
2.54.0
author maintainer
#6 lore raw

PATCH 5/11 perf tools: NULL bitmap pointers after bitmap_free()

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

Three call sites free bitmaps without NULLing the pointer, risking double-free if the structure is reused or cleanup is called twice:

 - mmap__munmap():                 map->affinity_mask.bits
 - record__mmap_cpu_mask_free():   mask->bits
 - memory_node__delete_nodes():    nodesp[i].set

Set each pointer to NULL after bitmap_free().

Fixes 8384a2600c7ddfc8 ("perf record: Adapt affinity to machines with #CPUs > 1K")
Fixes f466e5ed6c356d1d ("perf record: Extend --threads command line option")
Fixes 36d8658618c2505f ("perf header: Validate bitmap size before allocating in do_read_bitmap()")
Cc Alexey Budankov <[email protected]>
Cc Alexey Bayduraev <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-record.c | 1 +
 tools/perf/util/header.c    | 4 +++-
 tools/perf/util/mmap.c      | 1 +
 3 files changed, 5 insertions(+), 1 deletion(-)
tools/perf/builtin-record.c +1 -0
index a33c78f030d91012..e915390556752b9e 100644
@@ -3084,6 +3084,7 @@ static int record__mmap_cpu_mask_alloc(struct mmap_cpu_mask *mask, int nr_bits)
 static void record__mmap_cpu_mask_free(struct mmap_cpu_mask *mask)
 {
 	bitmap_free(mask->bits);
+	mask->bits = NULL;
 	mask->nbits = 0;
 }
 
tools/perf/util/header.c +3 -1
index d7f41db7322cbcb4..8d2ab440a1c8ee4a 100644
@@ -1481,8 +1481,10 @@ static int memory_node__read(struct memory_node *n, unsigned long idx)
 
 static void memory_node__delete_nodes(struct memory_node *nodesp, u64 cnt)
 {
-	for (u64 i = 0; i < cnt; i++)
+	for (u64 i = 0; i < cnt; i++) {
 		bitmap_free(nodesp[i].set);
+		nodesp[i].set = NULL;
+	}
 
 	free(nodesp);
 }
tools/perf/util/mmap.c +1 -0
index 8012301d3cf2ac9a..ee3ebdf53e15291e 100644
@@ -239,6 +239,7 @@ static void perf_mmap__aio_munmap(struct mmap *map __maybe_unused)
 void mmap__munmap(struct mmap *map)
 {
 	bitmap_free(map->affinity_mask.bits);
+	map->affinity_mask.bits = NULL;
 
 	zstd_fini(&map->zstd_data);
 
Signature · 1 line
2.54.0
author maintainer
#7 lore raw

PATCH 6/11 perf sched: Bounds-check prio before test_bit() in timehist

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

timehist_skip_sample() reads prio from untrusted tracepoint data via perf_sample__intval(sample, "prev_prio") without bounds validation. A crafted perf.data with prev_prio >= MAX_PRIO (140) causes test_bit() to read past the end of the prio_bitmap, which is only MAX_PRIO bits.

Add a prio >= 0 && prio < MAX_PRIO check before the test_bit() call. This also makes the != -1 sentinel check explicit as >= 0.

Fixes 9b3a48bbe20d9692 ("perf sched timehist: Add --prio option")
Cc Yang Jihong <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/lib/perf/TODO        | 7 +++++++
 tools/perf/builtin-sched.c | 4 +++-
 2 files changed, 10 insertions(+), 1 deletion(-)
tools/lib/perf/TODO +7 -0
index 486dd95dc57208a8..1a3644aa1f38dde4 100644
@@ -11,6 +11,13 @@ together.
      (x86_64 max is 8192, arm64 is 4096), but NR_CPUS limits keep
      growing.  perf clamps to INT16_MAX in set_max_cpu_num() as a
      safety net.
+   - Code simplification: the int16_t forces defensive truncation
+     checks at every boundary where a wider CPU index (int from
+     sample->cpu, al->cpu, etc.) is narrowed into struct perf_cpu.
+     Without these checks, values > 32767 silently wrap to small
+     positive numbers, bypassing bounds validation.  Widening to int
+     eliminates this entire class of silent truncation bugs and
+     removes the need for the INT16_MAX clamp in set_max_cpu_num().
    - Scope: struct perf_cpu is embedded everywhere — perf_cpu_map__cpu(),
      perf_cpu_map__min(), perf_cpu_map__max(), perf_cpu_map__has(), the
      for_each_cpu macros, and all internal callers.  The perf_cpu_map
tools/perf/builtin-sched.c +3 -1
index 1ff01f03d2ad1ad3..5f3510f9ca249132 100644
@@ -2645,7 +2645,9 @@ static bool timehist_skip_sample(struct perf_sched *sched,
 		else if (evsel__name_is(sample->evsel, "sched:sched_switch"))
 			prio = perf_sample__intval(sample, "prev_prio");
 
-		if (prio != -1 && !test_bit(prio, sched->prio_bitmap)) {
+		/* prio comes from untrusted tracepoint data — skip invalid values */
+		if (prio < 0 || prio >= MAX_PRIO ||
+		    !test_bit(prio, sched->prio_bitmap)) {
 			rc = true;
 			sched->skipped_samples++;
 		}
Signature · 1 line
2.54.0
#8 lore raw

Re: [PATCH 06/11] perf sched: Bounds-check prio before test_bit() in timehist

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] Logic inversion in priority bounds check breaks --prio filtering for initial events. --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Logic inversion in priority bounds check breaks --prio filtering for initial events. --

commit 1cea9987d6681e3cb6b64fbd144fd651abc72351
Author: Arnaldo Carvalho de Melo <[email protected]>

perf sched: Bounds-check prio before test_bit() in timehist

This commit adds bounds checking for the priority variable in timehist_skip_sample to prevent out-of-bounds memory accesses in test_bit. It attempts to skip samples if the priority falls outside the valid range.

Quoted 4 lines
tools/perf/builtin-sched.c +0 -0
index 1ff01f03d2ad1..5f3510f9ca249 100644

[ ... ]

tools/perf/builtin-sched.c static bool timehist_skip_sample(struct perf_sched *sched, · line 2650

Quoted 8 lines
tools/perf/builtin-sched.c +3 -1
@@ -2645,7 +2645,9 @@ static bool timehist_skip_sample(struct perf_sched *sched,
 		else if (evsel__name_is(sample->evsel, "sched:sched_switch"))
 			prio = perf_sample__intval(sample, "prev_prio");
 
-		if (prio != -1 && !test_bit(prio, sched->prio_bitmap)) {
+		/* prio comes from untrusted tracepoint data — skip invalid values */
+		if (prio < 0 || prio >= MAX_PRIO ||
+		    !test_bit(prio, sched->prio_bitmap)) {

[Severity: Medium]
Does this logic inversion unintentionally skip valid initial events that do not have their priority recorded yet?

In timehist_skip_sample(), prio is initialized to -1 to represent an unknown priority:

    int prio = -1;

With the previous logic, an unknown priority of -1 caused the condition to evaluate to false, allowing the event to pass the filter:

    if (prio != -1 && !test_bit(prio, sched->prio_bitmap)) {
        rc = true;
        sched->skipped_samples++;
    }

With the new logic, the condition prio < 0 evaluates to true when prio is -1, causing the event to be unexpectedly skipped.

Quoted 3 lines
tools/perf/builtin-sched.c +0 -0
 			rc = true;
 			sched->skipped_samples++;
 		}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6
author maintainer
#9 lore raw

PATCH 7/11 perf sched: Fix idle-hist callchain display using wrong rb_first variant

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

timehist_print_idlehist_callchain() calls rb_first_cached() on sorted_root, but the sort function (callchain_param.sort) populates it via rb_insert_color() on the plain rb_root member — not the cached variant. This means rb_leftmost is never set, so rb_first_cached() always returns NULL and the entire callchain summary is silently dropped from --idle-hist output.

Use rb_first(&root->rb_root) to match how the tree was populated.

Fixes ba957ebb54893aca ("perf sched timehist: Show callchains for idle stat")
Cc Namhyung Kim <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-sched.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tools/perf/builtin-sched.c +2 -1
index 5f3510f9ca249132..30de749e066335a5 100644
@@ -3130,7 +3130,8 @@ static size_t timehist_print_idlehist_callchain(struct rb_root_cached *root)
 	size_t ret = 0;
 	FILE *fp = stdout;
 	struct callchain_node *chain;
-	struct rb_node *rb_node = rb_first_cached(root);
+	/* sort() uses rb_insert_color() on rb_root, not rb_root_cached */
+	struct rb_node *rb_node = rb_first(&root->rb_root);
 
 	printf("  %16s  %8s  %s\n", "Idle time (msec)", "Count", "Callchains");
 	printf("  %.16s  %.8s  %.50s\n", graph_dotted_line, graph_dotted_line,
Signature · 1 line
2.54.0
author maintainer
#10 lore raw

PATCH 8/11 perf tools: Add O_CLOEXEC to open() calls in DSO and ELF code

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

open() calls in dso.c and symbol-elf.c omit O_CLOEXEC, which leaks file descriptors to child processes spawned during symbol resolution (e.g., addr2line, objdump). This can exhaust the fd limit during long profiling sessions or when processing many DSOs.

Add O_CLOEXEC to all open() calls in both files (12 call sites).

Fixes cdd059d731eeb466 ("perf tools: Move dso_* related functions into dso object")
Fixes e5a1845fc0aeca85 ("perf symbols: Split out util/symbol-elf.c")
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/dso.c        |  4 ++--
 tools/perf/util/symbol-elf.c | 20 ++++++++++----------
 2 files changed, 12 insertions(+), 12 deletions(-)
tools/perf/util/dso.c +2 -2
index 7dced896c64eafd7..fb2e78fe2aa8eb94 100644
@@ -344,7 +344,7 @@ int filename__decompress(const char *name, char *pathname,
 	 * descriptor to the uncompressed file.
 	 */
 	if (!compressions[comp].is_compressed(name))
-		return open(name, O_RDONLY);
+		return open(name, O_RDONLY | O_CLOEXEC);
 
 	fd = mkstemp(tmpbuf);
 	if (fd < 0) {
@@ -1911,7 +1911,7 @@ static const u8 *__dso__read_symbol(struct dso *dso, const char *symfs_filename,
 	int saved_errno;
 
 	nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
-	fd = open(symfs_filename, O_RDONLY);
+	fd = open(symfs_filename, O_RDONLY | O_CLOEXEC);
 	saved_errno = errno;
 	nsinfo__mountns_exit(&nsc);
 	if (fd < 0) {
tools/perf/util/symbol-elf.c +10 -10
index 186e6d92ac3d7742..c2bdfd0003df2abe 100644
@@ -217,7 +217,7 @@ bool filename__has_section(const char *filename, const char *sec)
 	GElf_Shdr shdr;
 	bool found = false;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);
 	if (fd < 0)
 		return false;
 
@@ -872,7 +872,7 @@ static int read_build_id(const char *filename, struct build_id *bid)
 	if (size < BUILD_ID_SIZE)
 		goto out;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);
 	if (fd < 0)
 		goto out;
 
@@ -935,7 +935,7 @@ int sysfs__read_build_id(const char *filename, struct build_id *bid)
 	size_t size = sizeof(bid->data);
 	int fd, err = -1;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);
 	if (fd < 0)
 		goto out;
 
@@ -995,7 +995,7 @@ int filename__read_debuglink(const char *filename, char *debuglink,
 	if (err >= 0)
 		goto out;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);
 	if (fd < 0)
 		goto out;
 
@@ -1153,7 +1153,7 @@ int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
 
 		type = dso__symtab_type(dso);
 	} else {
-		fd = open(name, O_RDONLY);
+		fd = open(name, O_RDONLY | O_CLOEXEC);
 		if (fd < 0) {
 			*dso__load_errno(dso) = errno;
 			return -1;
@@ -1952,7 +1952,7 @@ static int kcore__open(struct kcore *kcore, const char *filename)
 {
 	GElf_Ehdr *ehdr;
 
-	kcore->fd = open(filename, O_RDONLY);
+	kcore->fd = open(filename, O_RDONLY | O_CLOEXEC);
 	if (kcore->fd == -1)
 		return -1;
 
@@ -1985,7 +1985,7 @@ static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
 	if (temp)
 		kcore->fd = mkstemp(filename);
 	else
-		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
+		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL | O_CLOEXEC, 0400);
 	if (kcore->fd == -1)
 		return -1;
 
@@ -2461,11 +2461,11 @@ static int kcore_copy__compare_files(const char *from_filename,
 {
 	int from, to, err = -1;
 
-	from = open(from_filename, O_RDONLY);
+	from = open(from_filename, O_RDONLY | O_CLOEXEC);
 	if (from < 0)
 		return -1;
 
-	to = open(to_filename, O_RDONLY);
+	to = open(to_filename, O_RDONLY | O_CLOEXEC);
 	if (to < 0)
 		goto out_close_from;
 
@@ -2883,7 +2883,7 @@ int get_sdt_note_list(struct list_head *head, const char *target)
 	Elf *elf;
 	int fd, ret;
 
-	fd = open(target, O_RDONLY);
+	fd = open(target, O_RDONLY | O_CLOEXEC);
 	if (fd < 0)
 		return -EBADF;
 
Signature · 1 line
2.54.0
#11 lore raw

Re: [PATCH 08/11] perf tools: Add O_CLOEXEC to open() calls in DSO and ELF code

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Stack buffer overflow in sysfs__read_build_id due to integer overflow.
- [High] Incomplete fix for file descriptor leaks; mkstemp() calls still lack O_CLOEXEC.
- [High] Out-of-bounds memory read in filename__read_debuglink when parsing malformed ELF files. --

commit ebdb5f81eceb8366c0e12ca6798c1cd5dfc35089
Author: Arnaldo Carvalho de Melo <[email protected]>

perf tools: Add O_CLOEXEC to open() calls in DSO and ELF code

This commit adds the O_CLOEXEC flag to open() calls in dso.c and symbol-elf.c. The goal is to prevent file descriptors from leaking to child processes spawned during symbol resolution.

tools/perf/util/dso.c int filename__decompress(const char *name, char *pathname, · line 349

Quoted 12 lines
tools/perf/util/dso.c +1 -1
index 7dced896c64ea..fb2e78fe2aa8e 100644
@@ -344,7 +344,7 @@ int filename__decompress(const char *name, char *pathname,
 	 * descriptor to the uncompressed file.
 	 */
 	if (!compressions[comp].is_compressed(name))
-		return open(name, O_RDONLY);
+		return open(name, O_RDONLY | O_CLOEXEC);
 
 	fd = mkstemp(tmpbuf);

[Severity: High]
This isn't a bug introduced by this patch, but since the goal is to prevent file descriptor leaks to child processes, should we also update the mkstemp() calls?

Here in filename__decompress(), mkstemp() creates a temporary file descriptor without the close-on-exec flag. This FD will be inherited by spawned child processes. Could this be changed to use mkostemp(tmpbuf, O_CLOEXEC)?

[ ... ]

Quoted 4 lines
tools/perf/util/symbol-elf.c +0 -0
index 186e6d92ac3d7..c2bdfd0003df2 100644

[ ... ]

tools/perf/util/symbol-elf.c int sysfs__read_build_id(const char *filename, struct build_id *bid) · line 938

Quoted 6 lines
tools/perf/util/symbol-elf.c +1 -1
@@ -935,7 +935,7 @@ int sysfs__read_build_id(const char *filename, struct build_id *bid)
 	size_t size = sizeof(bid->data);
 	int fd, err = -1;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);

[Severity: Critical]
This is a pre-existing issue, but there appears to be an integer overflow risk further down in sysfs__read_build_id() that could lead to a stack buffer overflow.

The function reads nhdr.n_namesz and nhdr.n_descsz directly from the file. Since these are 32-bit values, they can be arbitrarily large. The calculation:

    int n = namesz + descsz;

can overflow the 32-bit signed integer n, resulting in a negative value. The size check:

    if (n > (int)sizeof(bf)) {
        n = sizeof(bf);
        ...
    }

bypasses truncation because a negative number is smaller than sizeof(bf). When read(fd, bf, n) is called, the negative n is interpreted as a massive size_t count, potentially reading up to 2GB of data into the 8KB stack buffer bf[BUFSIZ].

Could this be exploited if perf processes data from an untrusted root directory?

Quoted 3 lines
tools/perf/util/symbol-elf.c +0 -0
 	if (fd < 0)
 		goto out;
 

[ ... ]

tools/perf/util/symbol-elf.c int filename__read_debuglink(const char *filename, char *debuglink, · line 998

Quoted 6 lines
tools/perf/util/symbol-elf.c +1 -1
@@ -995,7 +995,7 @@ int filename__read_debuglink(const char *filename, char *debuglink,
 	if (err >= 0)
 		goto out;
 
-	fd = open(filename, O_RDONLY);
+	fd = open(filename, O_RDONLY | O_CLOEXEC);

[Severity: High]
This isn't a bug introduced by this patch, but is there a risk of an out-of-bounds read further down in filename__read_debuglink() when parsing malformed ELF files?

Later in this function, the code blindly copies the section data:

    /* the start of this section is a zero-terminated string */
    strncpy(debuglink, data->d_buf, size);

If a corrupted ELF file provides a small .gnu_debuglink section without a null terminator, strncpy() lacks bounds validation against the actual section size (data->d_size) and could read past the end of the mapped section data.

If this crosses into unmapped memory, could it cause a segmentation fault and crash perf?

Quoted 3 lines
tools/perf/util/symbol-elf.c +0 -0
 	if (fd < 0)
 		goto out;
 

[ ... ]

tools/perf/util/symbol-elf.c static int kcore__init(struct kcore *kcore, char *filename, int elfclass, · line 1988

Quoted 6 lines
tools/perf/util/symbol-elf.c +1 -1
@@ -1985,7 +1985,7 @@ static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
 	if (temp)
 		kcore->fd = mkstemp(filename);
 	else
-		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
+		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL | O_CLOEXEC, 0400);

[Severity: High]
This is also a pre-existing issue, but the if (temp) branch uses mkstemp() which lacks O_CLOEXEC. This creates an inconsistent state compared to the else branch that correctly adds O_CLOEXEC.

Does this mkstemp() need to be replaced with mkostemp(..., O_CLOEXEC) to fully prevent file descriptor exhaustion? (There is also another mkstemp() call in read_gnu_debugdata() that exhibits the same behavior).

Quoted 3 lines
tools/perf/util/symbol-elf.c +0 -0
 	if (kcore->fd == -1)
 		return -1;
 
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
author maintainer
#12 lore raw

PATCH 9/11 perf bpf: Use scnprintf() in snprintf_hex() and synthesize_bpf_prog_name()

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

Both functions accumulate formatted output via ret += snprintf(buf + ret, size - ret, ...). If the buffer is too small and snprintf() returns more than the remaining space, ret exceeds size and the next 'size - ret' underflows, causing snprintf() to write past the buffer end.

Switch to scnprintf() which returns the actual number of bytes written, making the accumulation safe.

Fixes 7b612e291a5affb1 ("perf tools: Synthesize PERF_RECORD_* for loaded BPF programs")
Cc Song Liu <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/bpf-event.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
tools/perf/util/bpf-event.c +6 -5
index a27945c279efb779..2c09842469f1f28c 100644
@@ -36,7 +36,7 @@ static int snprintf_hex(char *buf, size_t size, unsigned char *data, size_t len)
 	size_t i;
 
 	for (i = 0; i < len; i++)
-		ret += snprintf(buf + ret, size - ret, "%02x", data[i]);
+		ret += scnprintf(buf + ret, size - ret, "%02x", data[i]);
 	return ret;
 }
 
@@ -140,7 +140,7 @@ static int synthesize_bpf_prog_name(char *buf, int size,
 	const struct btf_type *t;
 	int name_len;
 
-	name_len = snprintf(buf, size, "bpf_prog_");
+	name_len = scnprintf(buf, size, "bpf_prog_");
 	name_len += snprintf_hex(buf + name_len, size - name_len,
 				 prog_tags[sub_id], BPF_TAG_SIZE);
 	if (btf) {
@@ -153,9 +153,10 @@ static int synthesize_bpf_prog_name(char *buf, int size,
 			short_name = info->name;
 	} else
 		short_name = "F";
-	if (short_name)
-		name_len += snprintf(buf + name_len, size - name_len,
-				     "_%s", short_name);
+	if (short_name) {
+		name_len += scnprintf(buf + name_len, size - name_len,
+				      "_%s", short_name);
+	}
 	return name_len;
 }
 
Signature · 1 line
2.54.0
#13 lore raw

Re: [PATCH 09/11] perf bpf: Use scnprintf() in snprintf_hex() and synthesize_bpf_prog_name()

Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Buffer overflow in hwmon_pmu__for_each_event due to unsafe snprintf accumulation.
- [Critical] Buffer overflow in perf_pmu__for_each_event due to unsafe snprintf accumulation.
- [Medium] Missing validation for btf__type_by_id() return value in synthesize_bpf_prog_name leads to a NULL pointer dereference.
- [Medium] Missing bounds checking when accessing BPF function info and program tags allows out-of-bounds memory reads.
- [Medium] Leaked map.btf and map.rodata in bpf_metadata_create() on error path.
- [Medium] Leaked info_node->metadata when perf_env__insert_bpf_prog_info fails in perf_env__add_bpf_info(). --

commit 36b9c4933bfd61cb63ed28852ac600bdb93eebb7
Author: Arnaldo Carvalho de Melo <[email protected]>

perf bpf: Use scnprintf() in snprintf_hex() and synthesize_bpf_prog_name()

This commit fixes buffer overflows by switching from snprintf() to scnprintf() when accumulating formatted strings. This prevents writing past the end of the buffer if truncation previously occurred.

Quoted 4 lines
tools/perf/util/bpf-event.c +0 -0
index a27945c279efb..2c09842469f1f 100644

[ ... ]

tools/perf/util/bpf-event.c static int synthesize_bpf_prog_name(char *buf, int size, · line 146

Quoted 9 lines
tools/perf/util/bpf-event.c +1 -1
@@ -140,7 +140,7 @@ static int synthesize_bpf_prog_name(char *buf, int size,
 	const struct btf_type *t;
 	int name_len;
 
-	name_len = snprintf(buf, size, "bpf_prog_");
+	name_len = scnprintf(buf, size, "bpf_prog_");
 	name_len += snprintf_hex(buf + name_len, size - name_len,
 				 prog_tags[sub_id], BPF_TAG_SIZE);
 	if (btf) {

While reviewing this function, I noticed a couple of pre-existing issues just below this diff hunk:

tools/perf/util/bpf-event.c:synthesize_bpf_prog_name() {
        if (btf) {
                finfo = func_infos + sub_id * info->func_info_rec_size;
                t = btf__type_by_id(btf, finfo->type_id);
                short_name = btf__name_by_offset(btf, t->name_off);

...

[Severity: Medium]
This isn't a bug introduced by this patch, but does calculating the finfo pointer using the unvalidated info->func_info_rec_size from the untrusted perf.data file allow out-of-bounds memory reads?

[Severity: Medium]
This is a pre-existing issue, but btf__type_by_id() can return NULL if finfo->type_id is invalid or out of bounds. Will accessing t->name_off immediately after cause a NULL pointer dereference here?

Additionally, I noticed a few other pre-existing issues in the perf tools codebase during this review:

tools/perf/util/bpf-event.c:bpf_metadata_create() {
...

        metadata = bpf_metadata_alloc(info->nr_prog_tags, map.num_vars);
        if (!metadata)
                continue;

...

[Severity: Medium]
This isn't a bug introduced by this patch, but does this error path permanently leak the memory allocated for map.btf and map.rodata by skipping the bpf_metadata_free_map_data() cleanup?

tools/perf/util/bpf-event.c:perf_env__add_bpf_info() {
...

        if (!perf_env__insert_bpf_prog_info(env, info_node)) {
                pr_debug("%s: duplicate add bpf info request for id %u\n",
                         __func__, btf_id);
                free(info_linear);
                free(info_node);
                goto out;
        }

...

[Severity: Medium]
This is a pre-existing issue, but does freeing info_node here without freeing info_node->metadata leak the dynamically allocated metadata?

tools/perf/util/hwmon_pmu.c:hwmon_pmu__for_each_event() {
...

        len += hwmon_pmu__describe_items(hwm, desc_buf + len, sizeof(desc_buf) - len,
                                         key, value->items, /*is_alarm=*/false);

...

[Severity: Critical]
This isn't a bug introduced by this patch, but it is a similar issue to the ones being fixed.

Does accumulating the length from snprintf() without checking for truncation cause sizeof(desc_buf) - len to wrap to a huge value, leading to a stack buffer overflow in subsequent calls?

Note that this issue was not fixed by the later patches in the series.

tools/perf/util/pmu.c:perf_pmu__for_each_event() {
...

        buf_used += snprintf(buf + buf_used, sizeof(buf) - buf_used,
                        "%.*s/%s/", (int)pmu_name_len, info.pmu_name, event->terms) + 1;

...

[Severity: Critical]
This is a pre-existing issue, but does accumulating the return value of snprintf() into buf_used risk truncation wrapping?

If buf_used exceeds sizeof(buf), sizeof(buf) - buf_used underflows, leading to a stack buffer overflow on the next write.

Note that this issue was also not fixed by the later patches in the series.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=9
author maintainer
#14 lore raw

PATCH 10/11 perf hists: Fix snprintf() in hists__scnprintf_title() UID filter path

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

hists__scnprintf_title() accumulates formatted output into a buffer using scnprintf() for all filter clauses except the UID filter, which uses snprintf(). If the buffer fills up and snprintf() returns more than the remaining space, printed exceeds size and the next 'size - printed' underflows, causing later scnprintf() calls to write past the buffer.

Switch the UID filter clause to scnprintf() to match the rest of the function.

Fixes 25c312dbf88ca402 ("perf hists: Move hists__scnprintf_title() away from the TUI code")
Cc Arnaldo Carvalho de Melo <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/hist.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
tools/perf/util/hist.c +4 -3
index 811d68fa6770c5b7..df978c996b6c2262 100644
@@ -2963,9 +2963,10 @@ int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh
 			   ev_name, sample_freq_str, enable_ref ? ref : " ", nr_events);
 
 
-	if (hists->uid_filter_str)
-		printed += snprintf(bf + printed, size - printed,
-				    ", UID: %s", hists->uid_filter_str);
+	if (hists->uid_filter_str) {
+		printed += scnprintf(bf + printed, size - printed,
+				     ", UID: %s", hists->uid_filter_str);
+	}
 	if (thread) {
 		if (hists__has(hists, thread)) {
 			printed += scnprintf(bf + printed, size - printed,
Signature · 1 line
2.54.0
#15 lore raw

Re: [PATCH 10/11] perf hists: Fix snprintf() in hists__scnprintf_title() UID filter path

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

Pre-existing issues:
- [High] Buffer overflow in config_show() via test_driver/test_fs sysfs attributes due to snprintf underflow.
- [High] Buffer overflow in pm80xx_get_non_fatal_dump() when reading sysfs attributes due to unbounded snprintf usage.
- [Medium] Stack buffer overflow in hwmon_pmu__for_each_event() due to snprintf underflow.
- [Medium] Stack buffer overflow in perf_pmu__for_each_event() when parsing PMU aliases due to snprintf accumulation. --

commit 68f841f9ed7f1af9b87ee345589a137d67d74612
Author: Arnaldo Carvalho de Melo <[email protected]>

perf hists: Fix snprintf() in hists__scnprintf_title() UID filter path

This commit replaces snprintf() with scnprintf() in the UID filter path of hists__scnprintf_title() to prevent buffer size underflow. It ensures consistent buffer length calculation, matching the rest of the function.

Quoted 8 lines
tools/perf/util/hist.c +0 -0
index 811d68fa6770c..df978c996b6c2 100644
@@ -2963,9 +2963,10 @@ int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh
 			   ev_name, sample_freq_str, enable_ref ? ref : " ", nr_events);
 
 

tools/perf/util/hist.c int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh · line 2972

Quoted 10 lines
tools/perf/util/hist.c +4 -3
-	if (hists->uid_filter_str)
-		printed += snprintf(bf + printed, size - printed,
-				    ", UID: %s", hists->uid_filter_str);
+	if (hists->uid_filter_str) {
+		printed += scnprintf(bf + printed, size - printed,
+				     ", UID: %s", hists->uid_filter_str);
+	}
 	if (thread) {
 		if (hists__has(hists, thread)) {
 			printed += scnprintf(bf + printed, size - printed,

While reviewing this fix, I noticed a few similar pre-existing issues in other areas of the codebase where snprintf() return values might cause buffer overflows.

lib/test_kmod.c:config_show() {
    ...
	if (config->test_driver)
		len += snprintf(buf+len, PAGE_SIZE - len,
				"driver:\t%s\n",
				config->test_driver);
    ...
}

[Severity: High]
This is a pre-existing issue, but could this calculation underflow PAGE_SIZE - len?

Since test_driver can be up to PAGE_SIZE bytes long, the accumulated length can exceed PAGE_SIZE. If snprintf() returns the would-have-written length, does the subtraction result in a large positive value that allows subsequent snprintf() calls to write out of bounds of the sysfs buffer?

drivers/scsi/pm8001/pm80xx_hwi.c:pm80xx_get_non_fatal_dump() {
    ...
	for (index = 0; index < output_length/4; index++)
		buf_copy += snprintf(buf_copy, PAGE_SIZE,
				"%08x ", *(temp+index));
    ...
}

[Severity: High]
This isn't a bug introduced by this patch, but might this loop write beyond the bounds of the sysfs buffer?

The pointer buf_copy advances with each write, but the size parameter passed to snprintf() is constantly PAGE_SIZE instead of the remaining buffer space. If the hardware provides a large output_length, would this continuously write up to PAGE_SIZE bytes and cause a heap buffer overflow?

tools/perf/util/hwmon_pmu.c:hwmon_pmu__for_each_event() {
    ...
		char desc_buf[256];
    ...
		len = snprintf(desc_buf, sizeof(desc_buf), "%s in unit %s named %s.",
			hwmon_desc[key.type],
			pmu->name + 6,
			value->label ?: info.name);

		len += hwmon_pmu__describe_items(hwm, desc_buf + len, sizeof(desc_buf) - len,
						key, value->items, /*is_alarm=*/false);
    ...
}

[Severity: Medium]
This is a pre-existing issue, but is there a risk of a stack buffer overflow here?

Since value->label can be up to 255 bytes, snprintf() might return a length greater than sizeof(desc_buf). Would passing sizeof(desc_buf) - len to hwmon_pmu__describe_items() then underflow to a massive size_t value, allowing scnprintf() to write out-of-bounds on the stack?

tools/perf/util/pmu.c:perf_pmu__for_each_event() {
    ...
	char buf[1024];
    ...
			buf_used += snprintf(buf + buf_used, sizeof(buf) - buf_used,
					"%G%s", event->scale, event->unit) + 1;
		}
		info.desc = event->desc;
		info.long_desc = event->long_desc;
		info.encoding_desc = buf + buf_used;
		buf_used += snprintf(buf + buf_used, sizeof(buf) - buf_used,
				"%.*s/%s/", (int)pmu_name_len, info.pmu_name, event->terms) + 1;
    ...
}

[Severity: Medium]
This isn't a bug introduced by this patch, but does accumulating the snprintf() return values into buf_used risk underflowing sizeof(buf) - buf_used?

If an earlier snprintf() truncates, it returns the number of bytes it would have written. Could this cause buf_used to exceed sizeof(buf), making the size limit on the next snprintf() underflow and write beyond the end of the stack buffer?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=10
author maintainer
#16 lore raw

PATCH 11/11 perf tools: Use scnprintf() in build_id__snprintf() and hwmon read_events()

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

build_id__snprintf() and hwmon_pmu__read_events() accumulate formatted output via snprintf(), which returns the would-have-been-written count on truncation. In build_id__snprintf(), this inflates the return value beyond the buffer size. In hwmon_pmu__read_events(), len overshoots out_buf_len and the next 'out_buf_len - len' underflows.

Switch both to scnprintf() which returns actual bytes written.

Fixes fccaaf6fbbc59910 ("perf build-id: Change sprintf functions to snprintf")
Fixes 53cc0b351ec99278 ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
Cc Ian Rogers <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/util/build-id.c  |  2 +-
 tools/perf/util/hwmon_pmu.c | 12 ++++++------
 2 files changed, 7 insertions(+), 7 deletions(-)
tools/perf/util/build-id.c +1 -1
index 8c0a9ae932aa5798..a3b92108f96263c6 100644
@@ -94,7 +94,7 @@ int build_id__snprintf(const struct build_id *build_id, char *bf, size_t bf_size
 	}
 
 	for (size_t i = 0; i < build_id->size && offs < bf_size; ++i)
-		offs += snprintf(bf + offs, bf_size - offs, "%02x", build_id->data[i]);
+		offs += scnprintf(bf + offs, bf_size - offs, "%02x", build_id->data[i]);
 
 	return offs;
 }
tools/perf/util/hwmon_pmu.c +6 -6
index fb3ffa8d32ad2a93..dbf6a71af47f9a42 100644
@@ -442,12 +442,12 @@ static size_t hwmon_pmu__describe_items(struct hwmon_pmu *hwm, char *out_buf, si
 
 				buf[read_len] = '\0';
 				val = strtoll(buf, /*endptr=*/NULL, 10);
-				len += snprintf(out_buf + len, out_buf_len - len, "%s%s%s=%g%s",
-						len == 0 ? " " : ", ",
-						hwmon_item_strs[bit],
-						is_alarm ? "_alarm" : "",
-						(double)val / 1000.0,
-						hwmon_units[key.type]);
+				len += scnprintf(out_buf + len, out_buf_len - len, "%s%s%s=%g%s",
+						 len == 0 ? " " : ", ",
+						 hwmon_item_strs[bit],
+						 is_alarm ? "_alarm" : "",
+						 (double)val / 1000.0,
+						 hwmon_units[key.type]);
 			}
 			close(fd);
 		}
Signature · 1 line
2.54.0
#17 lore raw

Re: [PATCH 11/11] perf tools: Use scnprintf() in build_id__snprintf() and hwmon read_events()

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Reading sysfs file contents into fixed-size stack buffers causes a 1-byte out-of-bounds write when appending a null terminator if the read fills the buffer.
- [High] User-controlled stack buffer overflow via misused strlcpy in parse_hwmon_filename().
--

commit 48628bf245e27619381dcbb4cf7723991e4f5a11
Author: Arnaldo Carvalho de Melo <[email protected]>

perf tools: Use scnprintf() in build_id__snprintf() and hwmon read_events()

build_id__snprintf() and hwmon_pmu__read_events() accumulate formatted output via snprintf(), which returns the would-have-been-written count on truncation. In build_id__snprintf(), this inflates the return value beyond the buffer size. In hwmon_pmu__read_events(), len overshoots out_buf_len and the next 'out_buf_len - len' underflows.

Switch both to scnprintf() which returns actual bytes written.

Fixes fccaaf6fbbc59910 ("perf build-id: Change sprintf functions to snprintf")
Fixes 53cc0b351ec99278 ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
Cc Ian Rogers <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>

[ ... ]

tools/perf/util/hwmon_pmu.c static size_t hwmon_pmu__describe_items(struct hwmon_pmu *hwm, char *out_buf, si · line 443

Quoted 7 lines
tools/perf/util/hwmon_pmu.c +0 -0
index fb3ffa8d32ad2..dbf6a71af47f9 100644
@@ -442,12 +442,12 @@ static size_t hwmon_pmu__describe_items(struct hwmon_pmu *hwm, char *out_buf, si
 
 				buf[read_len] = '\0';

[Severity: High]
This isn't a bug introduced by this patch, but does this write out of bounds?

In hwmon_pmu__describe_items(), the code reads from sysfs directly into a stack buffer (buf) using sizeof(buf) as the maximum length. If the read fills the buffer exactly and lacks a trailing newline, read_len will equal sizeof(buf).

Writing to buf[read_len] would then write one byte past the end of the array, corrupting an adjacent byte on the stack with a null byte. Should the read size be sizeof(buf) - 1 to safely null-terminate?

This same pattern appears to exist in evsel__hwmon_pmu_read() and hwmon_pmu__read_events().

tools/perf/util/hwmon_pmu.c static size_t hwmon_pmu__describe_items(struct hwmon_pmu *hwm, char *out_buf, si · line 450

Quoted 13 lines
tools/perf/util/hwmon_pmu.c +6 -6
 				val = strtoll(buf, /*endptr=*/NULL, 10);
-				len += snprintf(out_buf + len, out_buf_len - len, "%s%s%s=%g%s",
-						len == 0 ? " " : ", ",
-						hwmon_item_strs[bit],
-						is_alarm ? "_alarm" : "",
-						(double)val / 1000.0,
-						hwmon_units[key.type]);
+				len += scnprintf(out_buf + len, out_buf_len - len, "%s%s%s=%g%s",
+						 len == 0 ? " " : ", ",
+						 hwmon_item_strs[bit],
+						 is_alarm ? "_alarm" : "",
+						 (double)val / 1000.0,
+						 hwmon_units[key.type]);

[Severity: High]
This is a pre-existing issue, but I also noticed a potential stack buffer overflow in parse_hwmon_filename() in this same file.

When copying an event prefix into fn_type:

tools/perf/util/hwmon_pmu.c:parse_hwmon_filename() {
    ...
	char fn_type[24];
    ...
	*alarm = false;
	fn_item_len = strlen(fn_item);
	if (fn_item_len > 6 && !strcmp(&fn_item[fn_item_len - 6], "_alarm")) {
		assert(strlen(LONGEST_HWMON_ITEM_STR) < sizeof(fn_type));
		strlcpy(fn_type, fn_item, fn_item_len - 5);
		fn_item = fn_type;
		*alarm = true;
	}
    ...
}

The third argument to strlcpy() is passed as fn_item_len - 5 instead of the destination buffer capacity (sizeof(fn_type)). Since fn_item originates from user-controlled event input, an arbitrarily long name ending in "_alarm" could cause strlcpy() to copy more than 24 characters into fn_type, overflowing the stack buffer.

Could this be exploited for memory corruption if perf is run with elevated capabilities?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=11