Skip to contentSkip to conversation

perf tools: Fix OOB reads, NULL deref, and resource leaks

PATCH Superseded Arnaldo Carvalho de Melo 2 participants 9 messages Jun 4 perf events

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Arnaldo Carvalho de Melo

author maintainer
#0 lore raw

cover perf tools: Fix OOB reads, NULL deref, and resource leaks

Hi,

Four pre-existing bugs found by sashiko-bot during AI-assisted review of the perf-data-validation hardening series. All are independent of that series -- they are latent bugs in surrounding code exposed during review.

1. test_bit(sample->cpu, cpu_bitmap) reads out of bounds when PERF_SAMPLE_CPU is absent (sample->cpu == (u32)-1) in annotate, diff, report, and sched timehist.

2. Thread reference leak in perf sched latency_switch_event() -- one error path does 'return -1' instead of 'goto out_put'.

3. NULL pointer dereference in perf sched latency_runtime_event() -- thread_atoms_search() called before the NULL check on the thread returned by machine__findnew_thread().

4. Heap buffer overflow in perf sched map --compact mode -- comp_cpus allocated based on host CPU count but indexed using a MAX_CPUS-sized bitmap, overflowing when analyzing recordings from machines with more CPUs than the host. Also removes the now-dead sysconf(_SC_NPROCESSORS_CONF) initialization whose sole consumer was the replaced allocation, since max_cpu is dynamically updated from the recording's events during processing.

All four require crafted or unusual perf.data inputs to trigger. Verified with gcc and clang builds, checkpatch, and perf test.

Arnaldo Carvalho de Melo (4):
perf tools: Guard test_bit from out-of-bounds sample CPU perf sched: Fix thread reference leak in latency_switch_event perf sched: Fix NULL dereference in latency_runtime_event perf sched: Fix comp_cpus heap overflow with cross-machine recordings

 tools/perf/builtin-annotate.c |  3 ++-
 tools/perf/builtin-diff.c     |  3 ++-
 tools/perf/builtin-report.c   |  3 ++-
 tools/perf/builtin-sched.c    | 16 +++++++++-------
 4 files changed, 15 insertions(+), 10 deletions(-)

Developed with AI assistance (Claude/sashiko), tagged in commits.

Thanks a lot,

- Arnaldo

author maintainer
#1 lore raw

PATCH 1/4 perf tools: Guard test_bit from out-of-bounds sample CPU

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

When PERF_SAMPLE_CPU is absent from a perf.data file, sample->cpu is initialized to (u32)-1 by evsel__parse_sample(). Five call sites pass this value directly to test_bit(sample->cpu, cpu_bitmap), reading massively out of bounds past the DECLARE_BITMAP(..., MAX_NR_CPUS) allocation of 4096 bits.

Add a sample->cpu >= MAX_NR_CPUS guard before each test_bit() call, matching the existing safe pattern in builtin-kwork.c. This catches both the (u32)-1 sentinel and any corrupted CPU value exceeding the bitmap size.

Fixes 5d67be97f890 ("perf report/annotate/script: Add option to specify a CPU range")
Cc Anton Blanchard <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-annotate.c | 3 ++-
 tools/perf/builtin-diff.c     | 3 ++-
 tools/perf/builtin-report.c   | 3 ++-
 tools/perf/builtin-sched.c    | 6 ++++--
 4 files changed, 10 insertions(+), 5 deletions(-)
tools/perf/builtin-annotate.c +2 -1
index b918f9eed5fd2441..8a0eb30eac24fdbc 100644
@@ -295,7 +295,8 @@ static int process_sample_event(const struct perf_tool *tool,
 		goto out_put;
 	}
 
-	if (ann->cpu_list && !test_bit(sample->cpu, ann->cpu_bitmap))
+	if (ann->cpu_list && (sample->cpu >= MAX_NR_CPUS ||
+			     !test_bit(sample->cpu, ann->cpu_bitmap)))
 		goto out_put;
 
 	if (!al.filtered &&
tools/perf/builtin-diff.c +2 -1
index 9592f44b6545bab6..9fa8e900637b0d71 100644
@@ -416,7 +416,8 @@ static int diff__process_sample_event(const struct perf_tool *tool,
 		goto out;
 	}
 
-	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap)) {
+	if (cpu_list && (sample->cpu >= MAX_NR_CPUS ||
+			!test_bit(sample->cpu, cpu_bitmap))) {
 		ret = 0;
 		goto out;
 	}
tools/perf/builtin-report.c +2 -1
index 6f044c3df8937dc5..dd1309c320943ea4 100644
@@ -298,7 +298,8 @@ static int process_sample_event(const struct perf_tool *tool,
 	if (symbol_conf.hide_unresolved && al.sym == NULL)
 		goto out_put;
 
-	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
+	if (rep->cpu_list && (sample->cpu >= MAX_NR_CPUS ||
+			     !test_bit(sample->cpu, rep->cpu_bitmap)))
 		goto out_put;
 
 	if (sort__mode == SORT_MODE__BRANCH) {
tools/perf/builtin-sched.c +4 -2
index 4de2baf03c5036dc..e7bd3f331cb8e889 100644
@@ -2192,7 +2192,8 @@ static void timehist_print_sample(struct perf_sched *sched,
 	char nstr[30];
 	u64 wait_time;
 
-	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
+	if (cpu_list && (sample->cpu >= MAX_NR_CPUS ||
+			!test_bit(sample->cpu, cpu_bitmap)))
 		return;
 
 	timestamp__scnprintf_usec(t, tstr, sizeof(tstr));
@@ -2871,7 +2872,8 @@ static int timehist_sched_change_event(const struct perf_tool *tool,
 	}
 
 	if (!sched->idle_hist || thread__tid(thread) == 0) {
-		if (!cpu_list || test_bit(sample->cpu, cpu_bitmap))
+		if (!cpu_list || (sample->cpu < MAX_NR_CPUS &&
+				 test_bit(sample->cpu, cpu_bitmap)))
 			timehist_update_runtime_stats(tr, t, tprev);
 
 		if (sched->idle_hist) {
Signature · 1 line
2.54.0
author maintainer
#2 lore raw

PATCH 2/4 perf sched: Fix thread reference leak in latency_switch_event

In reply to Arnaldo Carvalho de Melo

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

In latency_switch_event(), after acquiring thread references for sched_out and sched_in via machine__findnew_thread(), the first add_sched_out_event() failure path does 'return -1', bypassing the out_put label that calls thread__put() on both references.

The second and third add_sched_out_event() failures correctly use 'goto out_put'. Fix the first one to match.

Fixes b91fc39f4ad7 ("perf machine: Protect the machine->threads with a rwlock")
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-sched.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-sched.c +1 -1
index e7bd3f331cb8e889..13b801496a01271e 100644
@@ -1180,7 +1180,7 @@ static int latency_switch_event(struct perf_sched *sched,
 		}
 	}
 	if (add_sched_out_event(out_events, prev_state, timestamp))
-		return -1;
+		goto out_put;
 
 	in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
 	if (!in_events) {
Signature · 1 line
2.54.0
author maintainer
#3 lore raw

PATCH 3/4 perf sched: Fix NULL dereference in latency_runtime_event

In reply to Arnaldo Carvalho de Melo

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

latency_runtime_event() passes the return value of machine__findnew_thread() directly to thread_atoms_search() at line 1216, before checking for NULL at line 1220. thread_atoms_search() calls pid_cmp() which dereferences the thread pointer via thread__tid(), causing a NULL pointer dereference if the allocation fails.

All other callers of thread_atoms_search() in this file (latency_switch_event, latency_wakeup_event, latency_migrate_task_event) correctly check for NULL first.

Move the atoms assignment after the NULL check to match the pattern used by the other callers.

Fixes b91fc39f4ad7 ("perf machine: Protect the machine->threads with a rwlock")
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-sched.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
tools/perf/builtin-sched.c +3 -1
index 13b801496a01271e..36da451447b5e59f 100644
@@ -1213,13 +1213,15 @@ static int latency_runtime_event(struct perf_sched *sched,
 	const u32 pid	   = perf_sample__intval(sample, "pid");
 	const u64 runtime  = perf_sample__intval(sample, "runtime");
 	struct thread *thread = machine__findnew_thread(machine, -1, pid);
-	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+	struct work_atoms *atoms;
 	u64 timestamp = sample->time;
 	int cpu = sample->cpu, err = -1;
 
 	if (thread == NULL)
 		return -1;
 
+	atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+
 	/* perf.data is untrusted input — CPU may be absent or corrupted */
 	if (cpu >= MAX_CPUS || cpu < 0) {
 		pr_warning("WARNING: at offset %#" PRIx64 ": out-of-bound sample CPU %d, skipping sample\n",
Signature · 1 line
2.54.0
#4 lore raw

Re: [PATCH 3/4] perf sched: Fix NULL dereference in latency_runtime_event

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:
- [Critical] Using strcpy() to copy untrusted string payloads into a fixed-size buffer leads to a heap buffer overflow.
- [Critical] Unbounded attacker-controlled PID values are used to calculate memory allocation sizes, leading to integer overflows and OOB writes. --

commit ab6f421859203bf939f45ab57b4ced38c5193ef5
Author: Arnaldo Carvalho de Melo <[email protected]>

perf sched: Fix NULL dereference in latency_runtime_event

This commit fixes a NULL pointer dereference in latency_runtime_event(). It moves the assignment to atoms using thread_atoms_search() after the thread NULL check to match patterns used elsewhere in the file.

Quoted 11 lines
tools/perf/builtin-sched.c +1 -1
index 13b801496a012..36da451447b5e 100644
@@ -1213,13 +1213,15 @@ static int latency_runtime_event(struct perf_sched *sched,
 	const u32 pid	   = perf_sample__intval(sample, "pid");
 	const u64 runtime  = perf_sample__intval(sample, "runtime");
 	struct thread *thread = machine__findnew_thread(machine, -1, pid);
-	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+	struct work_atoms *atoms;
 	u64 timestamp = sample->time;

tools/perf/builtin-sched.c static int latency_runtime_event(struct perf_sched *sched, · line 1227

Quoted 10 lines
tools/perf/builtin-sched.c +2 -0
 	int cpu = sample->cpu, err = -1;
 
 	if (thread == NULL)
 		return -1;
 
+	atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+
 	/* perf.data is untrusted input — CPU may be absent or corrupted */
 	if (cpu >= MAX_CPUS || cpu < 0) {
 		pr_warning("WARNING: at offset %#" PRIx64 ": out-of-bound sample CPU %d, skipping sample\n",

[ ... ]

While reviewing this file, I noticed a couple of pre-existing issues in register_pid():

tools/perf/builtin-sched.c:register_pid() {
...

	if (pid >= (unsigned long)pid_max) {
		BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
			sizeof(struct task_desc *))) == NULL);
		while (pid >= (unsigned long)pid_max)
			sched->pid_to_task[pid_max++] = NULL;
	}

...

[Severity: Critical]
This isn't a bug introduced by this patch, but could an attacker-controlled pid from an untrusted perf.data file overflow the size calculation here?

If pid is set to 0xFFFFFFFF, the pid + 1 calculation wraps to 0 on 32-bit systems, resulting in a 0-byte allocation.

The subsequent while loop would then write NULL pointers sequentially out of bounds, which could corrupt heap memory.

Also in the same function:

tools/perf/builtin-sched.c:register_pid() {
...

	task = zalloc(sizeof(*task));
	task->pid = pid;
	task->nr = sched->nr_tasks;
	strcpy(task->comm, comm);

...

[Severity: Critical]
This is a pre-existing issue, but does this strcpy() allow a heap buffer overflow?

The comm string extracted from the perf.data payload is copied into the fixed 20-byte task->comm array.

If the untrusted string is longer than 19 characters, this could overflow the struct task_desc heap allocation and overwrite adjacent fields.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
author maintainer
#5 lore raw

Re: [PATCH 3/4] perf sched: Fix NULL dereference in latency_runtime_event

Quoted 6 lines · On Thu, Jun 04, 2026 at 08:30:10PM +0000, [email protected] wrote:

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

Pre-existing issues:
- [Critical] Using strcpy() to copy untrusted string payloads into a fixed-size buffer leads to a heap buffer overflow.
- [Critical] Unbounded attacker-controlled PID values are used to calculate memory allocation sizes, leading to integer overflows and OOB writes. --

Both pre-existing, added to TODO files for future series.

- Arnaldo

Quoted 79 lines

commit ab6f421859203bf939f45ab57b4ced38c5193ef5
Author: Arnaldo Carvalho de Melo <[email protected]>

perf sched: Fix NULL dereference in latency_runtime_event

This commit fixes a NULL pointer dereference in latency_runtime_event(). It moves the assignment to atoms using thread_atoms_search() after the thread NULL check to match patterns used elsewhere in the file.

Quoted 11 lines
tools/perf/builtin-sched.c +1 -1
index 13b801496a012..36da451447b5e 100644
@@ -1213,13 +1213,15 @@ static int latency_runtime_event(struct perf_sched *sched,
 	const u32 pid	   = perf_sample__intval(sample, "pid");
 	const u64 runtime  = perf_sample__intval(sample, "runtime");
 	struct thread *thread = machine__findnew_thread(machine, -1, pid);
-	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+	struct work_atoms *atoms;
 	u64 timestamp = sample->time;

tools/perf/builtin-sched.c static int latency_runtime_event(struct perf_sched *sched, · line 1227

Quoted 10 lines
tools/perf/builtin-sched.c +2 -0
 	int cpu = sample->cpu, err = -1;
 
 	if (thread == NULL)
 		return -1;
 
+	atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
+
 	/* perf.data is untrusted input — CPU may be absent or corrupted */
 	if (cpu >= MAX_CPUS || cpu < 0) {
 		pr_warning("WARNING: at offset %#" PRIx64 ": out-of-bound sample CPU %d, skipping sample\n",

[ ... ]

While reviewing this file, I noticed a couple of pre-existing issues in register_pid():

tools/perf/builtin-sched.c:register_pid() {
...

	if (pid >= (unsigned long)pid_max) {
		BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
			sizeof(struct task_desc *))) == NULL);
		while (pid >= (unsigned long)pid_max)
			sched->pid_to_task[pid_max++] = NULL;
	}

...

[Severity: Critical]
This isn't a bug introduced by this patch, but could an attacker-controlled pid from an untrusted perf.data file overflow the size calculation here?

If pid is set to 0xFFFFFFFF, the pid + 1 calculation wraps to 0 on 32-bit systems, resulting in a 0-byte allocation.

The subsequent while loop would then write NULL pointers sequentially out of bounds, which could corrupt heap memory.

Also in the same function:

tools/perf/builtin-sched.c:register_pid() {
...

	task = zalloc(sizeof(*task));
	task->pid = pid;
	task->nr = sched->nr_tasks;
	strcpy(task->comm, comm);

...

[Severity: Critical]
This is a pre-existing issue, but does this strcpy() allow a heap buffer overflow?

The comm string extracted from the perf.data payload is copied into the fixed 20-byte task->comm array.

If the untrusted string is longer than 19 characters, this could overflow the struct task_desc heap allocation and overwrite adjacent fields.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
author maintainer
#6 lore raw

PATCH 4/4 perf sched: Fix comp_cpus heap overflow with cross-machine recordings

In reply to Arnaldo Carvalho de Melo

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

setup_map_cpus() allocates comp_cpus based on sysconf(_SC_NPROCESSORS_CONF), the host machine's CPU count. But map_switch_event() indexes comp_cpus using cpus_nr derived from bitmap_weight(comp_cpus_mask, MAX_CPUS), where comp_cpus_mask is declared as DECLARE_BITMAP(..., MAX_CPUS) with MAX_CPUS=4096.

When analyzing a perf.data recording from a machine with more CPUs than the analysis host (e.g. 128-CPU server recording analyzed on an 8-CPU laptop), cpus_nr exceeds the allocation size, causing a heap buffer overflow.

Also fix a type mismatch: comp_cpus is 'struct perf_cpu *' (2 bytes per element) but was allocated with sizeof(int) (4 bytes per element).

Allocate comp_cpus with MAX_CPUS entries using the correct element size, matching the comp_cpus_mask bitmap bounds. Remove the sysconf(_SC_NPROCESSORS_CONF) initialization of max_cpu — it was only used for the comp_cpus allocation, and max_cpu is dynamically updated from the recording's events during processing.

Fixes 99623c628f54 ("perf sched: Add compact display option")
Cc Jiri Olsa <[email protected]>
Assisted-by Claude Opus 4.6 <[email protected]>
---
 tools/perf/builtin-sched.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tools/perf/builtin-sched.c +1 -3
index 36da451447b5e59f..80e42536400a91c2 100644
@@ -3573,10 +3573,8 @@ static int perf_sched__lat(struct perf_sched *sched)
 
 static int setup_map_cpus(struct perf_sched *sched)
 {
-	sched->max_cpu.cpu  = sysconf(_SC_NPROCESSORS_CONF);
-
 	if (sched->map.comp) {
-		sched->map.comp_cpus = calloc(sched->max_cpu.cpu, sizeof(int));
+		sched->map.comp_cpus = calloc(MAX_CPUS, sizeof(*sched->map.comp_cpus));
 		if (!sched->map.comp_cpus)
 			return -1;
 	}
Signature · 1 line
2.54.0
#7 lore raw

Re: [PATCH 4/4] perf sched: Fix comp_cpus heap overflow with cross-machine recordings

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] Removing the `sched->max_cpu.cpu` initialization introduces a severe off-by-one error, causing…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Removing the sched->max_cpu.cpu initialization introduces a severe off-by-one error, causing perf sched map to truncate the highest CPU from its output and print nothing for single-CPU traces. --

commit 4c1487bde6fa91cfb4bca7de01bfd611065c7c9a
Author: Arnaldo Carvalho de Melo <[email protected]>

perf sched: Fix comp_cpus heap overflow with cross-machine recordings

This commit fixes a heap buffer overflow by allocating comp_cpus with MAX_CPUS and the correct element size. It also removes the initial assignment of max_cpu from the host's CPU count.

tools/perf/builtin-sched.c static int perf_sched__lat(struct perf_sched *sched) · line 3577

Quoted 10 lines
tools/perf/builtin-sched.c +0 -2
index 36da451447b5e..80e42536400a9 100644
@@ -3573,10 +3573,8 @@ static int perf_sched__lat(struct perf_sched *sched)
 
 static int setup_map_cpus(struct perf_sched *sched)
 {
-	sched->max_cpu.cpu  = sysconf(_SC_NPROCESSORS_CONF);
-

[Severity: Medium]
Does removing this initialization cause an off-by-one error when rendering the schedule map?

Without this initialization, sched->max_cpu.cpu tracks the maximum 0-based CPU index seen in the trace. However, when compact mode is disabled, map_switch_event() assigns this 0-based index directly to cpus_nr:

tools/perf/builtin-sched.c:map_switch_event() {
        ...
        } else
                cpus_nr = sched->max_cpu.cpu;
        ...
}

Then print_sched_map() iterates strictly less than cpus_nr:

tools/perf/builtin-sched.c:print_sched_map() {
        ...
        for (int i = 0; i < cpus_nr; i++) {
        ...
}

Will this skip the highest CPU column entirely, and print nothing for single-CPU traces where cpus_nr becomes 0?

Quoted 6 lines
tools/perf/builtin-sched.c +1 -1
 	if (sched->map.comp) {
-		sched->map.comp_cpus = calloc(sched->max_cpu.cpu, sizeof(int));
+		sched->map.comp_cpus = calloc(MAX_CPUS, sizeof(*sched->map.comp_cpus));
 		if (!sched->map.comp_cpus)
 			return -1;
 	}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
author maintainer
#8 lore raw

Re: [PATCH 4/4] perf sched: Fix comp_cpus heap overflow with cross-machine recordings

Quoted 41 lines · On Thu, Jun 04, 2026 at 08:26:05PM +0000, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Removing the sched->max_cpu.cpu initialization introduces a severe off-by-one error, causing perf sched map to truncate the highest CPU from its output and print nothing for single-CPU traces. --

commit 4c1487bde6fa91cfb4bca7de01bfd611065c7c9a
Author: Arnaldo Carvalho de Melo <[email protected]>

perf sched: Fix comp_cpus heap overflow with cross-machine recordings

This commit fixes a heap buffer overflow by allocating comp_cpus with MAX_CPUS and the correct element size. It also removes the initial assignment of max_cpu from the host's CPU count.

tools/perf/builtin-sched.c static int perf_sched__lat(struct perf_sched *sched) · line 3577

Quoted 10 lines
tools/perf/builtin-sched.c +0 -2
index 36da451447b5e..80e42536400a9 100644
@@ -3573,10 +3573,8 @@ static int perf_sched__lat(struct perf_sched *sched)
 
 static int setup_map_cpus(struct perf_sched *sched)
 {
-	sched->max_cpu.cpu  = sysconf(_SC_NPROCESSORS_CONF);
-

[Severity: Medium]
Does removing this initialization cause an off-by-one error when rendering the schedule map?

Without this initialization, sched->max_cpu.cpu tracks the maximum 0-based CPU index seen in the trace. However, when compact mode is disabled, map_switch_event() assigns this 0-based index directly to cpus_nr:

tools/perf/builtin-sched.c:map_switch_event() {
        ...
        } else
                cpus_nr = sched->max_cpu.cpu;
        ...
}

Then print_sched_map() iterates strictly less than cpus_nr:

Quoted 8 lines
tools/perf/builtin-sched.c:print_sched_map() {
        ...
        for (int i = 0; i < cpus_nr; i++) {
        ...
}

Will this skip the highest CPU column entirely, and print nothing for single-CPU traces where cpus_nr becomes 0?

Legit, v2 will have a fix for this, the sysconf() was just masking a pre-existing problem.

- Arnaldo

Quoted 9 lines
Quoted 6 lines
tools/perf/builtin-sched.c +1 -1
 	if (sched->map.comp) {
-		sched->map.comp_cpus = calloc(sched->max_cpu.cpu, sizeof(int));
+		sched->map.comp_cpus = calloc(MAX_CPUS, sizeof(*sched->map.comp_cpus));
 		if (!sched->map.comp_cpus)
 			return -1;
 	}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4