* [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling
@ 2026-11-16 5:57 Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature Namhyung Kim
` (10 more replies)
0 siblings, 11 replies; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Hello,
This series improves namespace and chroot handling when perf accesses
files from /proc. It defers reading /proc until namespace info is
actually needed, verifies thread liveness, matching boot IDs, and
mount namespace inodes against PID/TID recycling, and avoids entering
target mount namespaces for host-side files such as the build-id cache.
It's assisted by LLM.
v2 changes:
* Keep dso__filename_with_chroot() (guarded by nsinfo__check_proc())
instead of calling chroot() inside nsinfo__mountns_enter().
* Add a mutex lock to struct nsinfo and replace thread->nsinfo safely on
exec/namespace events rather than mutating shared instances in place.
* Mark offline sessions explicitly, propagate exit on fork replacement
and perf_event__exit_del_thread(), and guard sub-thread fallback in
thread__e_machine_endian().
* Gate mount namespace entry for debuginfo/binary probing callers,
include distro debuginfo types, and handle build-id lookups in
symbol-minimal's dso__load_sym().
* Add unit and shell tests for namespace, nsinfo, chroot, and boot ID
handling.
v1: https://lore.kernel.org/r/20261023071141.134149-1-namhyung@kernel.org
Thanks,
Namhyung
Namhyung Kim (11):
perf tools: Add HEADER_BOOT_ID feature
perf tools: Add perf_env__is_same_boot()
perf tools: Add nsinfo__new_thread() to track tid
perf tools: Add lock to struct nsinfo
perf tools: Defer /proc lookup until nsinfo is used
perf tools: Verify thread status and mount namespace inode
perf tools: Check thread liveness when accessing /proc
perf tools: Export filename__read_build_id_ns()
perf tools: Only enter mount namespace if necessary
perf tools: Scope mount namespace entry in dso__load()
perf tests: Add namespace, nsinfo, and boot ID tests
.../Documentation/perf.data-file-format.txt | 5 +
tools/perf/builtin-inject.c | 15 +-
tools/perf/builtin-trace.c | 2 +-
tools/perf/tests/Build | 1 +
tools/perf/tests/builtin-test.c | 1 +
tools/perf/tests/namespaces.c | 1035 +++++++++++++++++
tools/perf/tests/shell/header.sh | 9 +
tools/perf/tests/tests.h | 1 +
tools/perf/util/build-id.c | 14 +-
tools/perf/util/capstone.c | 8 +-
tools/perf/util/debuginfo.c | 19 +-
tools/perf/util/disasm.c | 8 +-
tools/perf/util/dso.c | 59 +-
tools/perf/util/dso.h | 34 +
tools/perf/util/dsos.c | 17 +-
tools/perf/util/env.c | 69 ++
tools/perf/util/env.h | 4 +
tools/perf/util/event.c | 1 +
tools/perf/util/header.c | 28 +
tools/perf/util/header.h | 1 +
tools/perf/util/jitdump.c | 2 +-
tools/perf/util/machine.c | 16 +-
tools/perf/util/map.c | 2 +-
tools/perf/util/namespaces.c | 273 ++++-
tools/perf/util/namespaces.h | 10 +
tools/perf/util/session.c | 1 +
tools/perf/util/symbol-minimal.c | 4 +-
tools/perf/util/symbol.c | 45 +-
tools/perf/util/symbol.h | 3 +
tools/perf/util/thread.c | 95 +-
tools/perf/util/thread.h | 13 +-
31 files changed, 1651 insertions(+), 144 deletions(-)
create mode 100644 tools/perf/tests/namespaces.c
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply [flat|nested] 20+ messages in thread
* [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:09 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot() Namhyung Kim
` (9 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Save /proc/sys/kernel/random/boot_id in the perf.data header and
struct perf_env so post-processing tools can identify if it's the
same kernel boot on which the data was recorded.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
.../Documentation/perf.data-file-format.txt | 5 +++
tools/perf/builtin-inject.c | 1 +
tools/perf/util/env.c | 44 +++++++++++++++++++
tools/perf/util/env.h | 3 ++
| 28 ++++++++++++
| 1 +
tools/perf/util/session.c | 1 +
7 files changed, 83 insertions(+)
diff --git a/tools/perf/Documentation/perf.data-file-format.txt b/tools/perf/Documentation/perf.data-file-format.txt
index 1f8e7d8e9ae21d73..359d6ef0c80aac26 100644
--- a/tools/perf/Documentation/perf.data-file-format.txt
+++ b/tools/perf/Documentation/perf.data-file-format.txt
@@ -491,6 +491,11 @@ struct memory_range {
u8 remote_region_id;
}; [nr_memory_ranges]
+ HEADER_BOOT_ID = 36,
+
+ A perf_header_string with the kernel boot ID read from
+ /proc/sys/kernel/random/boot_id.
+
other bits are reserved and should be ignored for now
HEADER_FEAT_BITS = 256,
diff --git a/tools/perf/builtin-inject.c b/tools/perf/builtin-inject.c
index 3f3108fd39811017..0e01b14fa51676ab 100644
--- a/tools/perf/builtin-inject.c
+++ b/tools/perf/builtin-inject.c
@@ -2307,6 +2307,7 @@ static bool keep_feat(struct perf_inject *inject, int feat)
case HEADER_CPU_DOMAIN_INFO:
case HEADER_CLN_SIZE:
case HEADER_MEMORY_RANGES:
+ case HEADER_BOOT_ID:
return true;
/* Information that can be updated */
case HEADER_BUILD_ID:
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 50993b0dedc577c7..24de685486af9bd5 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -12,6 +12,7 @@
#include <linux/string.h>
#include <linux/zalloc.h>
#include "cgroup.h"
+#include <api/fs/fs.h>
#include <errno.h>
#include <sys/utsname.h>
#include <stdlib.h>
@@ -258,6 +259,7 @@ void perf_env__exit(struct perf_env *env)
zfree(&env->os_release);
zfree(&env->version);
zfree(&env->arch);
+ zfree(&env->boot_id);
zfree(&env->cpu_desc);
zfree(&env->cpuid);
zfree(&env->cmdline);
@@ -400,6 +402,48 @@ const char *perf_env__os_release(struct perf_env *env)
return release;
}
+static char *read_host_boot_id(void)
+{
+ char *buf = NULL;
+ size_t sz;
+
+ if (procfs__read_str("sys/kernel/random/boot_id", &buf, &sz) < 0)
+ return NULL;
+
+ strim(buf);
+ return buf;
+}
+
+const char *perf_env__boot_id(struct perf_env *env)
+{
+ const char *boot_id;
+
+ if (!env)
+ return NULL;
+
+ mutex_lock(&env->lock);
+ if (env->boot_id) {
+ boot_id = env->boot_id;
+ goto out;
+ }
+
+ /*
+ * If this is an offline target environment (set by perf_session on
+ * read or via HEADER_ARCH / HEADER_HOSTNAME), do not fall back to the
+ * host's boot_id when boot_id is not populated in the file.
+ */
+ if (env->offline || env->arch || env->hostname) {
+ boot_id = NULL;
+ goto out;
+ }
+
+ env->boot_id = read_host_boot_id();
+ boot_id = env->boot_id;
+out:
+ mutex_unlock(&env->lock);
+ return boot_id;
+}
+
int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
{
int i;
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 23d35425efe38b24..aafd213c7523c018 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -81,6 +81,7 @@ struct perf_env {
char *os_release;
char *version;
char *arch;
+ char *boot_id;
/* e_machine expanded from 16 to 32-bits for alignment. */
u32 e_machine;
u32 e_flags;
@@ -154,6 +155,7 @@ struct perf_env {
/* For fast cpu to numa node lookup via perf_env__numa_node */
int *numa_map;
int nr_numa_map;
+ bool offline;
/* For real clock time reference. */
struct {
@@ -187,6 +189,7 @@ void perf_env__exit(struct perf_env *env);
int perf_env__kernel_is_64_bit(struct perf_env *env);
bool perf_arch_is_big_endian(const char *arch);
const char *perf_env__os_release(struct perf_env *env);
+const char *perf_env__boot_id(struct perf_env *env);
int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[]);
--git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 9eec6deee190ff21..3e0b6fce55973dd1 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -494,6 +494,27 @@ static int write_version(struct feat_fd *ff,
return do_write_string(ff, perf_version_string);
}
+static int write_boot_id(struct feat_fd *ff, struct evlist *evlist)
+{
+ const char *boot_id = NULL;
+ char *buf = NULL;
+ size_t sz;
+ int ret;
+
+ if (evlist__session(evlist))
+ boot_id = perf_env__boot_id(perf_session__env(evlist__session(evlist)));
+
+ if (!boot_id) {
+ if (procfs__read_str("sys/kernel/random/boot_id", &buf, &sz) < 0)
+ return -1;
+ boot_id = strim(buf);
+ }
+
+ ret = do_write_string(ff, boot_id);
+ free(buf);
+ return ret;
+}
+
static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
{
FILE *file;
@@ -2079,6 +2100,11 @@ static void print_version(struct feat_fd *ff, FILE *fp)
fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
}
+static void print_boot_id(struct feat_fd *ff, FILE *fp)
+{
+ fprintf(fp, "# boot id : %s\n", ff->ph->env.boot_id);
+}
+
static void print_cmdline(struct feat_fd *ff, FILE *fp)
{
int nr, i;
@@ -3022,6 +3048,7 @@ FEAT_PROCESS_STR_FUN(osrelease, os_release);
FEAT_PROCESS_STR_FUN(version, version);
FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
FEAT_PROCESS_STR_FUN(cpuid, cpuid);
+FEAT_PROCESS_STR_FUN(boot_id, boot_id);
static int process_arch(struct feat_fd *ff, void *data __maybe_unused)
{
@@ -4496,6 +4523,7 @@ const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
FEAT_OPR(E_MACHINE, e_machine, false),
FEAT_OPR(CLN_SIZE, cln_size, false),
FEAT_OPR(MEMORY_RANGES, memory_ranges, false),
+ FEAT_OPR(BOOT_ID, boot_id, false),
};
struct header_print_data {
--git a/tools/perf/util/header.h b/tools/perf/util/header.h
index 765037762758ca11..5bbf16342b4d26d8 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -57,6 +57,7 @@ enum {
HEADER_E_MACHINE,
HEADER_CLN_SIZE,
HEADER_MEMORY_RANGES,
+ HEADER_BOOT_ID,
HEADER_LAST_FEATURE,
HEADER_FEAT_BITS = 256,
};
diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index a58b4fe44b62e3bb..578abcb14de84a39 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -179,6 +179,7 @@ struct perf_session *__perf_session__new(struct perf_data *data,
session->data = data;
if (perf_data__is_read(data)) {
+ session->header.env.offline = true;
ret = perf_session__open(session);
if (ret < 0)
goto out_delete;
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot()
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:02 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid Namhyung Kim
` (8 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Add perf_env__is_same_boot() to compare the recorded boot_id in struct
perf_env against the running host's /proc/sys/kernel/random/boot_id.
For live sessions where no offline header was populated, it falls back
to reading the host boot_id and returns true.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/env.c | 25 +++++++++++++++++++++++++
tools/perf/util/env.h | 1 +
2 files changed, 26 insertions(+)
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 24de685486af9bd5..91ab4ccaf5a9475c 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -444,6 +444,31 @@ const char *perf_env__boot_id(struct perf_env *env)
return boot_id;
}
+static char *host_boot_id;
+
+static void init_host_boot_id(void)
+{
+ host_boot_id = read_host_boot_id();
+}
+
+bool perf_env__is_same_boot(struct perf_env *env)
+{
+ const char *env_boot_id;
+
+ DO_ONCE(init_host_boot_id);
+ if (!host_boot_id)
+ return false;
+
+ if (!env)
+ return true;
+
+ env_boot_id = perf_env__boot_id(env);
+ if (!env_boot_id)
+ return false;
+
+ return !strcmp(env_boot_id, host_boot_id);
+}
+
int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
{
int i;
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index aafd213c7523c018..856385624cee6adf 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -190,6 +190,7 @@ int perf_env__kernel_is_64_bit(struct perf_env *env);
bool perf_arch_is_big_endian(const char *arch);
const char *perf_env__os_release(struct perf_env *env);
const char *perf_env__boot_id(struct perf_env *env);
+bool perf_env__is_same_boot(struct perf_env *env);
int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[]);
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot() Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 04/11] perf tools: Add lock to struct nsinfo Namhyung Kim
` (7 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Record the thread ID (tid) in struct nsinfo via nsinfo__new_thread() and
access /proc/<pid>/task/<tid>/ status and ns/mnt paths instead of only
/proc/<pid>/. Also recreate thread->nsinfo when the pid of the thread
is updated.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/machine.c | 3 +++
tools/perf/util/namespaces.c | 22 ++++++++++++++++++----
tools/perf/util/namespaces.h | 3 +++
tools/perf/util/thread.c | 2 +-
4 files changed, 25 insertions(+), 5 deletions(-)
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index 5b4550e97172db54..ac857b395882f569 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -19,6 +19,7 @@
#include "branch.h"
#include "mem-events.h"
#include "mem-info.h"
+#include "namespaces.h"
#include "path.h"
#include "srcline.h"
#include "symbol.h"
@@ -516,6 +517,8 @@ static void machine__update_thread_pid(struct machine *machine,
return;
thread__set_pid(th, pid);
+ nsinfo__zput(RC_CHK_ACCESS(th)->nsinfo);
+ RC_CHK_ACCESS(th)->nsinfo = nsinfo__new_thread(pid, thread__tid(th));
if (thread__pid(th) == thread__tid(th))
return;
diff --git a/tools/perf/util/namespaces.c b/tools/perf/util/namespaces.c
index 01502570b32d02ab..f426eba04d6ef449 100644
--- a/tools/perf/util/namespaces.c
+++ b/tools/perf/util/namespaces.c
@@ -106,7 +106,8 @@ int nsinfo__init(struct nsinfo *nsi)
if (snprintf(oldns, PATH_MAX, "/proc/self/ns/mnt") >= PATH_MAX)
return rv;
- if (asprintf(&newns, "/proc/%d/ns/mnt", nsinfo__pid(nsi)) == -1)
+ if (asprintf(&newns, "/proc/%d/task/%d/ns/mnt",
+ nsinfo__pid(nsi), nsinfo__tid(nsi)) == -1)
return rv;
if (stat(oldns, &old_stat) < 0)
@@ -127,7 +128,8 @@ int nsinfo__init(struct nsinfo *nsi)
/* If we're dealing with a process that is in a different PID namespace,
* attempt to work out the innermost tgid for the process.
*/
- if (snprintf(spath, PATH_MAX, "/proc/%d/status", nsinfo__pid(nsi)) >= PATH_MAX)
+ if (snprintf(spath, PATH_MAX, "/proc/%d/task/%d/status",
+ nsinfo__pid(nsi), nsinfo__tid(nsi)) >= PATH_MAX)
goto out;
rv = nsinfo__get_nspid(&RC_CHK_ACCESS(nsi)->tgid, &RC_CHK_ACCESS(nsi)->nstgid,
@@ -150,11 +152,11 @@ static struct nsinfo *nsinfo__alloc(void)
return res;
}
-struct nsinfo *nsinfo__new(pid_t pid)
+struct nsinfo *nsinfo__new_thread(pid_t pid, pid_t tid)
{
struct nsinfo *nsi;
- if (pid == 0)
+ if (pid <= 0 || tid <= 0)
return NULL;
nsi = nsinfo__alloc();
@@ -162,6 +164,7 @@ struct nsinfo *nsinfo__new(pid_t pid)
return NULL;
RC_CHK_ACCESS(nsi)->pid = pid;
+ RC_CHK_ACCESS(nsi)->tid = tid;
RC_CHK_ACCESS(nsi)->tgid = pid;
RC_CHK_ACCESS(nsi)->nstgid = pid;
nsinfo__clear_need_setns(nsi);
@@ -176,6 +179,11 @@ struct nsinfo *nsinfo__new(pid_t pid)
return nsi;
}
+struct nsinfo *nsinfo__new(pid_t pid)
+{
+ return nsinfo__new_thread(pid, pid);
+}
+
static const char *nsinfo__mntns_path(const struct nsinfo *nsi)
{
return RC_CHK_ACCESS(nsi)->mntns_path;
@@ -193,6 +201,7 @@ struct nsinfo *nsinfo__copy(const struct nsinfo *nsi)
return NULL;
RC_CHK_ACCESS(nnsi)->pid = nsinfo__pid(nsi);
+ RC_CHK_ACCESS(nnsi)->tid = nsinfo__tid(nsi);
RC_CHK_ACCESS(nnsi)->tgid = nsinfo__tgid(nsi);
RC_CHK_ACCESS(nnsi)->nstgid = nsinfo__nstgid(nsi);
RC_CHK_ACCESS(nnsi)->need_setns = nsinfo__need_setns(nsi);
@@ -265,6 +274,11 @@ pid_t nsinfo__pid(const struct nsinfo *nsi)
return RC_CHK_ACCESS(nsi)->pid;
}
+pid_t nsinfo__tid(const struct nsinfo *nsi)
+{
+ return RC_CHK_ACCESS(nsi)->tid;
+}
+
bool nsinfo__in_pidns(const struct nsinfo *nsi)
{
return RC_CHK_ACCESS(nsi)->in_pidns;
diff --git a/tools/perf/util/namespaces.h b/tools/perf/util/namespaces.h
index e95c79b80e27c8fc..ea645954944de4be 100644
--- a/tools/perf/util/namespaces.h
+++ b/tools/perf/util/namespaces.h
@@ -32,6 +32,7 @@ void namespaces__free(struct namespaces *namespaces);
DECLARE_RC_STRUCT(nsinfo) {
pid_t pid;
+ pid_t tid;
pid_t tgid;
pid_t nstgid;
bool need_setns;
@@ -48,6 +49,7 @@ struct nscookie {
int nsinfo__init(struct nsinfo *nsi);
struct nsinfo *nsinfo__new(pid_t pid);
+struct nsinfo *nsinfo__new_thread(pid_t pid, pid_t tid);
struct nsinfo *nsinfo__copy(const struct nsinfo *nsi);
struct nsinfo *nsinfo__get(struct nsinfo *nsi);
@@ -58,6 +60,7 @@ void nsinfo__clear_need_setns(struct nsinfo *nsi);
pid_t nsinfo__tgid(const struct nsinfo *nsi);
pid_t nsinfo__nstgid(const struct nsinfo *nsi);
pid_t nsinfo__pid(const struct nsinfo *nsi);
+pid_t nsinfo__tid(const struct nsinfo *nsi);
bool nsinfo__in_pidns(const struct nsinfo *nsi);
void nsinfo__set_in_pidns(struct nsinfo *nsi);
diff --git a/tools/perf/util/thread.c b/tools/perf/util/thread.c
index 91118c301913b834..d12eb6bc63313b00 100644
--- a/tools/perf/util/thread.c
+++ b/tools/perf/util/thread.c
@@ -71,7 +71,7 @@ struct thread *thread__new(pid_t pid, pid_t tid)
list_add(&comm->list, thread__comm_list(thread));
refcount_set(thread__refcnt(thread), 1);
/* Thread holds first ref to nsdata. */
- RC_CHK_ACCESS(thread)->nsinfo = nsinfo__new(pid);
+ RC_CHK_ACCESS(thread)->nsinfo = nsinfo__new_thread(pid, tid);
srccode_state_init(thread__srccode_state(thread));
}
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 04/11] perf tools: Add lock to struct nsinfo
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (2 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:17 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used Namhyung Kim
` (6 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
A struct nsinfo instance can be shared across threads via
thread->nsinfo and dso->nsinfo. Add a mutex lock inside struct nsinfo
to synchronize access to its fields, initialization, copying, and mount
namespace path duplication in nsinfo__mountns_enter().
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/namespaces.c | 106 +++++++++++++++++++++++++++--------
tools/perf/util/namespaces.h | 2 +
2 files changed, 86 insertions(+), 22 deletions(-)
diff --git a/tools/perf/util/namespaces.c b/tools/perf/util/namespaces.c
index f426eba04d6ef449..845a62045405c80a 100644
--- a/tools/perf/util/namespaces.c
+++ b/tools/perf/util/namespaces.c
@@ -94,7 +94,7 @@ static int nsinfo__get_nspid(pid_t *tgid, pid_t *nstgid, bool *in_pidns, const c
return 0;
}
-int nsinfo__init(struct nsinfo *nsi)
+static int nsinfo__init_locked(struct nsinfo *nsi)
{
char oldns[PATH_MAX];
char spath[PATH_MAX];
@@ -140,14 +140,29 @@ int nsinfo__init(struct nsinfo *nsi)
return rv;
}
+int nsinfo__init(struct nsinfo *nsi)
+{
+ int rv;
+
+ if (!nsi)
+ return -1;
+
+ mutex_lock(&RC_CHK_ACCESS(nsi)->lock);
+ rv = nsinfo__init_locked(nsi);
+ mutex_unlock(&RC_CHK_ACCESS(nsi)->lock);
+ return rv;
+}
+
static struct nsinfo *nsinfo__alloc(void)
{
struct nsinfo *res;
RC_STRUCT(nsinfo) *nsi;
nsi = calloc(1, sizeof(*nsi));
- if (ADD_RC_CHK(res, nsi))
+ if (ADD_RC_CHK(res, nsi)) {
refcount_set(&nsi->refcnt, 1);
+ mutex_init(&nsi->lock);
+ }
return res;
}
@@ -184,9 +199,9 @@ struct nsinfo *nsinfo__new(pid_t pid)
return nsinfo__new_thread(pid, pid);
}
-static const char *nsinfo__mntns_path(const struct nsinfo *nsi)
+static struct mutex *nsinfo__lock(const struct nsinfo *nsi)
{
- return RC_CHK_ACCESS(nsi)->mntns_path;
+ return (struct mutex *)&RC_CHK_ACCESS(nsi)->lock;
}
struct nsinfo *nsinfo__copy(const struct nsinfo *nsi)
@@ -200,19 +215,22 @@ struct nsinfo *nsinfo__copy(const struct nsinfo *nsi)
if (!nnsi)
return NULL;
- RC_CHK_ACCESS(nnsi)->pid = nsinfo__pid(nsi);
- RC_CHK_ACCESS(nnsi)->tid = nsinfo__tid(nsi);
- RC_CHK_ACCESS(nnsi)->tgid = nsinfo__tgid(nsi);
- RC_CHK_ACCESS(nnsi)->nstgid = nsinfo__nstgid(nsi);
- RC_CHK_ACCESS(nnsi)->need_setns = nsinfo__need_setns(nsi);
- RC_CHK_ACCESS(nnsi)->in_pidns = nsinfo__in_pidns(nsi);
- if (nsinfo__mntns_path(nsi)) {
- RC_CHK_ACCESS(nnsi)->mntns_path = strdup(nsinfo__mntns_path(nsi));
+ mutex_lock(nsinfo__lock(nsi));
+ RC_CHK_ACCESS(nnsi)->pid = RC_CHK_ACCESS(nsi)->pid;
+ RC_CHK_ACCESS(nnsi)->tid = RC_CHK_ACCESS(nsi)->tid;
+ RC_CHK_ACCESS(nnsi)->tgid = RC_CHK_ACCESS(nsi)->tgid;
+ RC_CHK_ACCESS(nnsi)->nstgid = RC_CHK_ACCESS(nsi)->nstgid;
+ RC_CHK_ACCESS(nnsi)->need_setns = RC_CHK_ACCESS(nsi)->need_setns;
+ RC_CHK_ACCESS(nnsi)->in_pidns = RC_CHK_ACCESS(nsi)->in_pidns;
+ if (RC_CHK_ACCESS(nsi)->mntns_path) {
+ RC_CHK_ACCESS(nnsi)->mntns_path = strdup(RC_CHK_ACCESS(nsi)->mntns_path);
if (!RC_CHK_ACCESS(nnsi)->mntns_path) {
+ mutex_unlock(nsinfo__lock(nsi));
nsinfo__put(nnsi);
return NULL;
}
}
+ mutex_unlock(nsinfo__lock(nsi));
return nnsi;
}
@@ -227,6 +245,7 @@ static void nsinfo__delete(struct nsinfo *nsi)
if (nsi) {
WARN_ONCE(refcount_read(nsinfo__refcnt(nsi)) != 0, "nsinfo refcnt unbalanced\n");
zfree(&RC_CHK_ACCESS(nsi)->mntns_path);
+ mutex_destroy(&RC_CHK_ACCESS(nsi)->lock);
RC_CHK_FREE(nsi);
}
}
@@ -251,22 +270,46 @@ void nsinfo__put(struct nsinfo *nsi)
bool nsinfo__need_setns(const struct nsinfo *nsi)
{
- return RC_CHK_ACCESS(nsi)->need_setns;
+ bool ret;
+
+ if (!nsi)
+ return false;
+
+ mutex_lock(nsinfo__lock(nsi));
+ ret = RC_CHK_ACCESS(nsi)->need_setns;
+ mutex_unlock(nsinfo__lock(nsi));
+ return ret;
}
void nsinfo__clear_need_setns(struct nsinfo *nsi)
{
+ if (!nsi)
+ return;
+
+ mutex_lock(nsinfo__lock(nsi));
RC_CHK_ACCESS(nsi)->need_setns = false;
+ zfree(&RC_CHK_ACCESS(nsi)->mntns_path);
+ mutex_unlock(nsinfo__lock(nsi));
}
-pid_t nsinfo__tgid(const struct nsinfo *nsi)
+pid_t nsinfo__tgid(const struct nsinfo *nsi)
{
- return RC_CHK_ACCESS(nsi)->tgid;
+ pid_t ret;
+
+ mutex_lock(nsinfo__lock(nsi));
+ ret = RC_CHK_ACCESS(nsi)->tgid;
+ mutex_unlock(nsinfo__lock(nsi));
+ return ret;
}
-pid_t nsinfo__nstgid(const struct nsinfo *nsi)
+pid_t nsinfo__nstgid(const struct nsinfo *nsi)
{
- return RC_CHK_ACCESS(nsi)->nstgid;
+ pid_t ret;
+
+ mutex_lock(nsinfo__lock(nsi));
+ ret = RC_CHK_ACCESS(nsi)->nstgid;
+ mutex_unlock(nsinfo__lock(nsi));
+ return ret;
}
pid_t nsinfo__pid(const struct nsinfo *nsi)
@@ -281,18 +324,26 @@ pid_t nsinfo__tid(const struct nsinfo *nsi)
bool nsinfo__in_pidns(const struct nsinfo *nsi)
{
- return RC_CHK_ACCESS(nsi)->in_pidns;
+ bool ret;
+
+ mutex_lock(nsinfo__lock(nsi));
+ ret = RC_CHK_ACCESS(nsi)->in_pidns;
+ mutex_unlock(nsinfo__lock(nsi));
+ return ret;
}
void nsinfo__set_in_pidns(struct nsinfo *nsi)
{
+ mutex_lock(&RC_CHK_ACCESS(nsi)->lock);
RC_CHK_ACCESS(nsi)->in_pidns = true;
+ mutex_unlock(&RC_CHK_ACCESS(nsi)->lock);
}
void nsinfo__mountns_enter(struct nsinfo *nsi,
struct nscookie *nc)
{
char curpath[PATH_MAX];
+ char *mntns_path = NULL;
int oldns = -1;
int newns = -1;
char *oldcwd = NULL;
@@ -303,27 +354,36 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
nc->oldns = -1;
nc->newns = -1;
- if (!nsi || !nsinfo__need_setns(nsi))
+ if (!nsi)
+ return;
+
+ mutex_lock(&RC_CHK_ACCESS(nsi)->lock);
+ if (RC_CHK_ACCESS(nsi)->need_setns && RC_CHK_ACCESS(nsi)->mntns_path)
+ mntns_path = strdup(RC_CHK_ACCESS(nsi)->mntns_path);
+ mutex_unlock(&RC_CHK_ACCESS(nsi)->lock);
+
+ if (!mntns_path)
return;
if (!getcwd(curpath, sizeof(curpath)))
- return;
+ goto out;
oldcwd = strdup(curpath);
if (!oldcwd)
- return;
+ goto out;
oldns = open("/proc/self/ns/mnt", O_RDONLY);
if (oldns < 0)
goto errout;
- newns = open(nsinfo__mntns_path(nsi), O_RDONLY);
+ newns = open(mntns_path, O_RDONLY);
if (newns < 0)
goto errout;
if (setns(newns, CLONE_NEWNS) < 0)
goto errout;
+ free(mntns_path);
nc->oldcwd = oldcwd;
nc->oldns = oldns;
nc->newns = newns;
@@ -335,6 +395,8 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
close(oldns);
if (newns > -1)
close(newns);
+out:
+ free(mntns_path);
}
void nsinfo__mountns_exit(struct nscookie *nc)
diff --git a/tools/perf/util/namespaces.h b/tools/perf/util/namespaces.h
index ea645954944de4be..bbb47a6f9e8323f2 100644
--- a/tools/perf/util/namespaces.h
+++ b/tools/perf/util/namespaces.h
@@ -14,6 +14,7 @@
#include <linux/refcount.h>
#include <linux/types.h>
#include <internal/rc_check.h>
+#include "mutex.h"
#ifndef HAVE_SETNS_SUPPORT
int setns(int fd, int nstype);
@@ -31,6 +32,7 @@ struct namespaces *namespaces__new(struct perf_record_namespaces *event);
void namespaces__free(struct namespaces *namespaces);
DECLARE_RC_STRUCT(nsinfo) {
+ struct mutex lock;
pid_t pid;
pid_t tid;
pid_t tgid;
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (3 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 04/11] perf tools: Add lock to struct nsinfo Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:11 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode Namhyung Kim
` (5 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Do not read /proc/<pid>/task/<tid>/ eagerly in nsinfo__new_thread().
Instead, initialize struct nsinfo on demand via
nsinfo__ensure_init_locked() when namespace information is queried or
entered, protected by nsinfo->lock.
Replace thread__nsinfo() with thread__get_nsinfo() and add
thread__reset_nsinfo() to replace thread->nsinfo safely under
thread->namespaces_lock on exec and PERF_RECORD_NAMESPACES events so
namespace changes are picked up without racing existing readers.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/builtin-inject.c | 2 +-
tools/perf/util/jitdump.c | 2 +-
tools/perf/util/map.c | 2 +-
tools/perf/util/namespaces.c | 24 +++++++++++++++---------
tools/perf/util/namespaces.h | 1 +
tools/perf/util/thread.c | 32 ++++++++++++++++++++++++++++++++
tools/perf/util/thread.h | 6 ++----
7 files changed, 53 insertions(+), 16 deletions(-)
diff --git a/tools/perf/builtin-inject.c b/tools/perf/builtin-inject.c
index 0e01b14fa51676ab..05bf1aae61778952 100644
--- a/tools/perf/builtin-inject.c
+++ b/tools/perf/builtin-inject.c
@@ -626,7 +626,7 @@ static struct dso *findnew_dso(int pid, int tid, const char *filename,
}
vdso = is_vdso_map(filename);
- nsi = nsinfo__get(thread__nsinfo(thread));
+ nsi = thread__get_nsinfo(thread);
if (vdso) {
/* The vdso maps are always on the host and not the
diff --git a/tools/perf/util/jitdump.c b/tools/perf/util/jitdump.c
index 45a05316b3ca65b7..95a08ead1f380747 100644
--- a/tools/perf/util/jitdump.c
+++ b/tools/perf/util/jitdump.c
@@ -985,7 +985,7 @@ jit_process(struct perf_session *session,
return 0;
}
- nsi = nsinfo__get(thread__nsinfo(thread));
+ nsi = thread__get_nsinfo(thread);
thread__put(thread);
/*
diff --git a/tools/perf/util/map.c b/tools/perf/util/map.c
index d5319ffa88a7477c..91c1fbeff9d2446f 100644
--- a/tools/perf/util/map.c
+++ b/tools/perf/util/map.c
@@ -139,7 +139,7 @@ struct map *map__new(struct machine *machine, u64 start, u64 len,
anon = is_anon_memory(filename) || flags & MAP_HUGETLB;
vdso = is_vdso_map(filename);
no_dso = is_no_dso_memory(filename);
- nsi = nsinfo__get(thread__nsinfo(thread));
+ nsi = thread__get_nsinfo(thread);
if ((anon || no_dso) && nsi && (prot & PROT_EXEC)) {
snprintf(newfilename, sizeof(newfilename),
diff --git a/tools/perf/util/namespaces.c b/tools/perf/util/namespaces.c
index 845a62045405c80a..afd8ad7f843ec09d 100644
--- a/tools/perf/util/namespaces.c
+++ b/tools/perf/util/namespaces.c
@@ -103,6 +103,8 @@ static int nsinfo__init_locked(struct nsinfo *nsi)
struct stat new_stat;
int rv = -1;
+ RC_CHK_ACCESS(nsi)->inited = true;
+
if (snprintf(oldns, PATH_MAX, "/proc/self/ns/mnt") >= PATH_MAX)
return rv;
@@ -153,6 +155,12 @@ int nsinfo__init(struct nsinfo *nsi)
return rv;
}
+static void nsinfo__ensure_init_locked(struct nsinfo *nsi)
+{
+ if (!RC_CHK_ACCESS(nsi)->inited)
+ nsinfo__init_locked(nsi);
+}
+
static struct nsinfo *nsinfo__alloc(void)
{
struct nsinfo *res;
@@ -182,15 +190,6 @@ struct nsinfo *nsinfo__new_thread(pid_t pid, pid_t tid)
RC_CHK_ACCESS(nsi)->tid = tid;
RC_CHK_ACCESS(nsi)->tgid = pid;
RC_CHK_ACCESS(nsi)->nstgid = pid;
- nsinfo__clear_need_setns(nsi);
- RC_CHK_ACCESS(nsi)->in_pidns = false;
- /* Init may fail if the process exits while we're trying to look at its
- * proc information. In that case, save the pid but don't try to enter
- * the namespace.
- */
- if (nsinfo__init(nsi) == -1)
- nsinfo__clear_need_setns(nsi);
-
return nsi;
}
@@ -220,6 +219,7 @@ struct nsinfo *nsinfo__copy(const struct nsinfo *nsi)
RC_CHK_ACCESS(nnsi)->tid = RC_CHK_ACCESS(nsi)->tid;
RC_CHK_ACCESS(nnsi)->tgid = RC_CHK_ACCESS(nsi)->tgid;
RC_CHK_ACCESS(nnsi)->nstgid = RC_CHK_ACCESS(nsi)->nstgid;
+ RC_CHK_ACCESS(nnsi)->inited = RC_CHK_ACCESS(nsi)->inited;
RC_CHK_ACCESS(nnsi)->need_setns = RC_CHK_ACCESS(nsi)->need_setns;
RC_CHK_ACCESS(nnsi)->in_pidns = RC_CHK_ACCESS(nsi)->in_pidns;
if (RC_CHK_ACCESS(nsi)->mntns_path) {
@@ -276,6 +276,7 @@ bool nsinfo__need_setns(const struct nsinfo *nsi)
return false;
mutex_lock(nsinfo__lock(nsi));
+ nsinfo__ensure_init_locked((struct nsinfo *)nsi);
ret = RC_CHK_ACCESS(nsi)->need_setns;
mutex_unlock(nsinfo__lock(nsi));
return ret;
@@ -287,6 +288,7 @@ void nsinfo__clear_need_setns(struct nsinfo *nsi)
return;
mutex_lock(nsinfo__lock(nsi));
+ RC_CHK_ACCESS(nsi)->inited = true;
RC_CHK_ACCESS(nsi)->need_setns = false;
zfree(&RC_CHK_ACCESS(nsi)->mntns_path);
mutex_unlock(nsinfo__lock(nsi));
@@ -297,6 +299,7 @@ pid_t nsinfo__tgid(const struct nsinfo *nsi)
pid_t ret;
mutex_lock(nsinfo__lock(nsi));
+ nsinfo__ensure_init_locked((struct nsinfo *)nsi);
ret = RC_CHK_ACCESS(nsi)->tgid;
mutex_unlock(nsinfo__lock(nsi));
return ret;
@@ -307,6 +310,7 @@ pid_t nsinfo__nstgid(const struct nsinfo *nsi)
pid_t ret;
mutex_lock(nsinfo__lock(nsi));
+ nsinfo__ensure_init_locked((struct nsinfo *)nsi);
ret = RC_CHK_ACCESS(nsi)->nstgid;
mutex_unlock(nsinfo__lock(nsi));
return ret;
@@ -327,6 +331,7 @@ bool nsinfo__in_pidns(const struct nsinfo *nsi)
bool ret;
mutex_lock(nsinfo__lock(nsi));
+ nsinfo__ensure_init_locked((struct nsinfo *)nsi);
ret = RC_CHK_ACCESS(nsi)->in_pidns;
mutex_unlock(nsinfo__lock(nsi));
return ret;
@@ -358,6 +363,7 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
return;
mutex_lock(&RC_CHK_ACCESS(nsi)->lock);
+ nsinfo__ensure_init_locked(nsi);
if (RC_CHK_ACCESS(nsi)->need_setns && RC_CHK_ACCESS(nsi)->mntns_path)
mntns_path = strdup(RC_CHK_ACCESS(nsi)->mntns_path);
mutex_unlock(&RC_CHK_ACCESS(nsi)->lock);
diff --git a/tools/perf/util/namespaces.h b/tools/perf/util/namespaces.h
index bbb47a6f9e8323f2..42dcca13b002f153 100644
--- a/tools/perf/util/namespaces.h
+++ b/tools/perf/util/namespaces.h
@@ -37,6 +37,7 @@ DECLARE_RC_STRUCT(nsinfo) {
pid_t tid;
pid_t tgid;
pid_t nstgid;
+ bool inited;
bool need_setns;
bool in_pidns;
char *mntns_path;
diff --git a/tools/perf/util/thread.c b/tools/perf/util/thread.c
index d12eb6bc63313b00..5d1c886add3e61eb 100644
--- a/tools/perf/util/thread.c
+++ b/tools/perf/util/thread.c
@@ -191,6 +191,34 @@ static int __thread__set_namespaces(struct thread *thread, u64 timestamp,
return 0;
}
+static void __thread__reset_nsinfo(struct thread *thread)
+ EXCLUSIVE_LOCKS_REQUIRED(thread__namespaces_lock(thread))
+{
+ nsinfo__zput(RC_CHK_ACCESS(thread)->nsinfo);
+ RC_CHK_ACCESS(thread)->nsinfo = nsinfo__new_thread(thread__pid(thread),
+ thread__tid(thread));
+}
+
+void thread__reset_nsinfo(struct thread *thread)
+{
+ down_write(thread__namespaces_lock(thread));
+ __thread__reset_nsinfo(thread);
+ up_write(thread__namespaces_lock(thread));
+}
+
+struct nsinfo *thread__get_nsinfo(struct thread *thread)
+{
+ struct nsinfo *nsi;
+
+ if (!thread)
+ return NULL;
+
+ down_read(thread__namespaces_lock(thread));
+ nsi = nsinfo__get(RC_CHK_ACCESS(thread)->nsinfo);
+ up_read(thread__namespaces_lock(thread));
+ return nsi;
+}
+
int thread__set_namespaces(struct thread *thread, u64 timestamp,
struct perf_record_namespaces *event)
{
@@ -198,6 +226,8 @@ int thread__set_namespaces(struct thread *thread, u64 timestamp,
down_write(thread__namespaces_lock(thread));
ret = __thread__set_namespaces(thread, timestamp, event);
+ if (!ret && event)
+ __thread__reset_nsinfo(thread);
up_write(thread__namespaces_lock(thread));
return ret;
}
@@ -282,6 +312,8 @@ int __thread__set_comm(struct thread *thread, const char *str, u64 timestamp,
down_write(thread__comm_lock(thread));
ret = ____thread__set_comm(thread, str, timestamp, exec);
up_write(thread__comm_lock(thread));
+ if (!ret && exec)
+ thread__reset_nsinfo(thread);
return ret;
}
diff --git a/tools/perf/util/thread.h b/tools/perf/util/thread.h
index d82fce8173aefedc..36f2932422df8988 100644
--- a/tools/perf/util/thread.h
+++ b/tools/perf/util/thread.h
@@ -286,10 +286,8 @@ static inline void thread__set_ts(struct thread *thread, struct thread_stack *ts
RC_CHK_ACCESS(thread)->ts = ts;
}
-static inline struct nsinfo *thread__nsinfo(struct thread *thread)
-{
- return RC_CHK_ACCESS(thread)->nsinfo;
-}
+struct nsinfo *thread__get_nsinfo(struct thread *thread);
+void thread__reset_nsinfo(struct thread *thread);
static inline struct srccode_state *thread__srccode_state(struct thread *thread)
{
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (4 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:14 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 07/11] perf tools: Check thread liveness when accessing /proc Namhyung Kim
` (4 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Verify /proc/<pid>/task/<tid>/status exists and has Tgid == pid before
reading or entering a thread's namespaces to guard against PID/TID
recycling.
Also record the mount namespace inode (mntns_ino) during initialization
and verify via fstat() that the opened namespace file descriptor still
refers to the same mount namespace.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/namespaces.c | 139 +++++++++++++++++++++++++++--------
tools/perf/util/namespaces.h | 4 +
2 files changed, 114 insertions(+), 29 deletions(-)
diff --git a/tools/perf/util/namespaces.c b/tools/perf/util/namespaces.c
index afd8ad7f843ec09d..860b90e8384ed1f2 100644
--- a/tools/perf/util/namespaces.c
+++ b/tools/perf/util/namespaces.c
@@ -65,6 +65,7 @@ static int nsinfo__get_nspid(pid_t *tgid, pid_t *nstgid, bool *in_pidns, const c
char *statln = NULL;
size_t linesz = 0;
char *nspid;
+ bool got_tgid = false;
f = fopen(path, "r");
if (f == NULL)
@@ -75,6 +76,7 @@ static int nsinfo__get_nspid(pid_t *tgid, pid_t *nstgid, bool *in_pidns, const c
if (strstr(statln, "Tgid:") != NULL) {
*tgid = (pid_t)strtol(strrchr(statln, '\t'), NULL, 10);
*nstgid = *tgid;
+ got_tgid = true;
}
if (strstr(statln, "NStgid:") != NULL) {
@@ -91,7 +93,20 @@ static int nsinfo__get_nspid(pid_t *tgid, pid_t *nstgid, bool *in_pidns, const c
fclose(f);
free(statln);
- return 0;
+ return got_tgid ? 0 : -1;
+}
+
+static void nsinfo__clear_need_setns_locked(struct nsinfo *nsi)
+{
+ RC_CHK_ACCESS(nsi)->inited = true;
+ RC_CHK_ACCESS(nsi)->need_setns = false;
+ zfree(&RC_CHK_ACCESS(nsi)->mntns_path);
+}
+
+static void nsinfo__set_exited_locked(struct nsinfo *nsi)
+{
+ RC_CHK_ACCESS(nsi)->exited = true;
+ nsinfo__clear_need_setns_locked(nsi);
}
static int nsinfo__init_locked(struct nsinfo *nsi)
@@ -101,16 +116,41 @@ static int nsinfo__init_locked(struct nsinfo *nsi)
char *newns = NULL;
struct stat old_stat;
struct stat new_stat;
+ pid_t tgid = 0, nstgid = 0;
+ bool in_pidns = false;
int rv = -1;
+ if (RC_CHK_ACCESS(nsi)->exited)
+ return -1;
+
RC_CHK_ACCESS(nsi)->inited = true;
+ /*
+ * Verify that the thread (pid, tid) still exists in /proc and its Tgid
+ * matches pid (guarding against PID/TID recycling), while resolving the
+ * innermost tgid if the process is in a different PID namespace.
+ */
+ if (snprintf(spath, PATH_MAX, "/proc/%d/task/%d/status",
+ nsinfo__pid(nsi), nsinfo__tid(nsi)) >= PATH_MAX)
+ goto out;
+
+ rv = nsinfo__get_nspid(&tgid, &nstgid, &in_pidns, spath);
+ if (rv < 0 || tgid != nsinfo__pid(nsi)) {
+ nsinfo__set_exited_locked(nsi);
+ rv = -1;
+ goto out;
+ }
+
+ RC_CHK_ACCESS(nsi)->tgid = tgid;
+ RC_CHK_ACCESS(nsi)->nstgid = nstgid;
+ RC_CHK_ACCESS(nsi)->in_pidns = in_pidns;
+
if (snprintf(oldns, PATH_MAX, "/proc/self/ns/mnt") >= PATH_MAX)
- return rv;
+ return -1;
if (asprintf(&newns, "/proc/%d/task/%d/ns/mnt",
nsinfo__pid(nsi), nsinfo__tid(nsi)) == -1)
- return rv;
+ return -1;
if (stat(oldns, &old_stat) < 0)
goto out;
@@ -124,19 +164,10 @@ static int nsinfo__init_locked(struct nsinfo *nsi)
if (old_stat.st_ino != new_stat.st_ino) {
RC_CHK_ACCESS(nsi)->need_setns = true;
RC_CHK_ACCESS(nsi)->mntns_path = newns;
+ RC_CHK_ACCESS(nsi)->mntns_ino = new_stat.st_ino;
newns = NULL;
}
- /* If we're dealing with a process that is in a different PID namespace,
- * attempt to work out the innermost tgid for the process.
- */
- if (snprintf(spath, PATH_MAX, "/proc/%d/task/%d/status",
- nsinfo__pid(nsi), nsinfo__tid(nsi)) >= PATH_MAX)
- goto out;
-
- rv = nsinfo__get_nspid(&RC_CHK_ACCESS(nsi)->tgid, &RC_CHK_ACCESS(nsi)->nstgid,
- &RC_CHK_ACCESS(nsi)->in_pidns, spath);
-
out:
free(newns);
return rv;
@@ -157,7 +188,7 @@ int nsinfo__init(struct nsinfo *nsi)
static void nsinfo__ensure_init_locked(struct nsinfo *nsi)
{
- if (!RC_CHK_ACCESS(nsi)->inited)
+ if (!RC_CHK_ACCESS(nsi)->inited && !RC_CHK_ACCESS(nsi)->exited)
nsinfo__init_locked(nsi);
}
@@ -219,7 +250,9 @@ struct nsinfo *nsinfo__copy(const struct nsinfo *nsi)
RC_CHK_ACCESS(nnsi)->tid = RC_CHK_ACCESS(nsi)->tid;
RC_CHK_ACCESS(nnsi)->tgid = RC_CHK_ACCESS(nsi)->tgid;
RC_CHK_ACCESS(nnsi)->nstgid = RC_CHK_ACCESS(nsi)->nstgid;
+ RC_CHK_ACCESS(nnsi)->mntns_ino = RC_CHK_ACCESS(nsi)->mntns_ino;
RC_CHK_ACCESS(nnsi)->inited = RC_CHK_ACCESS(nsi)->inited;
+ RC_CHK_ACCESS(nnsi)->exited = RC_CHK_ACCESS(nsi)->exited;
RC_CHK_ACCESS(nnsi)->need_setns = RC_CHK_ACCESS(nsi)->need_setns;
RC_CHK_ACCESS(nnsi)->in_pidns = RC_CHK_ACCESS(nsi)->in_pidns;
if (RC_CHK_ACCESS(nsi)->mntns_path) {
@@ -277,7 +310,7 @@ bool nsinfo__need_setns(const struct nsinfo *nsi)
mutex_lock(nsinfo__lock(nsi));
nsinfo__ensure_init_locked((struct nsinfo *)nsi);
- ret = RC_CHK_ACCESS(nsi)->need_setns;
+ ret = !RC_CHK_ACCESS(nsi)->exited && RC_CHK_ACCESS(nsi)->need_setns;
mutex_unlock(nsinfo__lock(nsi));
return ret;
}
@@ -288,12 +321,47 @@ void nsinfo__clear_need_setns(struct nsinfo *nsi)
return;
mutex_lock(nsinfo__lock(nsi));
- RC_CHK_ACCESS(nsi)->inited = true;
- RC_CHK_ACCESS(nsi)->need_setns = false;
- zfree(&RC_CHK_ACCESS(nsi)->mntns_path);
+ nsinfo__clear_need_setns_locked(nsi);
+ mutex_unlock(nsinfo__lock(nsi));
+}
+
+void nsinfo__set_exited(struct nsinfo *nsi)
+{
+ if (!nsi)
+ return;
+
+ mutex_lock(nsinfo__lock(nsi));
+ nsinfo__set_exited_locked(nsi);
mutex_unlock(nsinfo__lock(nsi));
}
+bool nsinfo__check_proc(struct nsinfo *nsi)
+{
+ char spath[64];
+ pid_t tgid = 0, nstgid = 0;
+ bool in_pidns = false;
+ bool exited;
+
+ if (!nsi)
+ return false;
+
+ mutex_lock(nsinfo__lock(nsi));
+ exited = RC_CHK_ACCESS(nsi)->exited;
+ mutex_unlock(nsinfo__lock(nsi));
+ if (exited)
+ return false;
+
+ scnprintf(spath, sizeof(spath), "/proc/%d/task/%d/status",
+ nsinfo__pid(nsi), nsinfo__tid(nsi));
+ if (nsinfo__get_nspid(&tgid, &nstgid, &in_pidns, spath) < 0 ||
+ tgid != nsinfo__pid(nsi)) {
+ nsinfo__set_exited(nsi);
+ return false;
+ }
+
+ return true;
+}
+
pid_t nsinfo__tgid(const struct nsinfo *nsi)
{
pid_t ret;
@@ -349,6 +417,8 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
{
char curpath[PATH_MAX];
char *mntns_path = NULL;
+ ino_t mntns_ino = 0;
+ struct stat st;
int oldns = -1;
int newns = -1;
char *oldcwd = NULL;
@@ -358,19 +428,26 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
nc->oldns = -1;
nc->newns = -1;
+ nc->oldcwd = NULL;
if (!nsi)
return;
mutex_lock(&RC_CHK_ACCESS(nsi)->lock);
nsinfo__ensure_init_locked(nsi);
- if (RC_CHK_ACCESS(nsi)->need_setns && RC_CHK_ACCESS(nsi)->mntns_path)
+ if (!RC_CHK_ACCESS(nsi)->exited && RC_CHK_ACCESS(nsi)->need_setns &&
+ RC_CHK_ACCESS(nsi)->mntns_path) {
mntns_path = strdup(RC_CHK_ACCESS(nsi)->mntns_path);
+ mntns_ino = RC_CHK_ACCESS(nsi)->mntns_ino;
+ }
mutex_unlock(&RC_CHK_ACCESS(nsi)->lock);
if (!mntns_path)
return;
+ if (!nsinfo__check_proc(nsi))
+ goto out;
+
if (!getcwd(curpath, sizeof(curpath)))
goto out;
@@ -378,14 +455,19 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
if (!oldcwd)
goto out;
- oldns = open("/proc/self/ns/mnt", O_RDONLY);
+ oldns = open("/proc/self/ns/mnt", O_RDONLY | O_CLOEXEC);
if (oldns < 0)
goto errout;
- newns = open(mntns_path, O_RDONLY);
+ newns = open(mntns_path, O_RDONLY | O_CLOEXEC);
if (newns < 0)
goto errout;
+ if (fstat(newns, &st) < 0 || st.st_ino != mntns_ino) {
+ nsinfo__clear_need_setns(nsi);
+ goto errout;
+ }
+
if (setns(newns, CLONE_NEWNS) < 0)
goto errout;
@@ -407,17 +489,11 @@ void nsinfo__mountns_enter(struct nsinfo *nsi,
void nsinfo__mountns_exit(struct nscookie *nc)
{
- if (nc == NULL || nc->oldns == -1 || nc->newns == -1 || !nc->oldcwd)
+ if (nc == NULL || !nc->oldcwd)
return;
- setns(nc->oldns, CLONE_NEWNS);
-
- if (nc->oldcwd) {
- WARN_ON_ONCE(chdir(nc->oldcwd));
- zfree(&nc->oldcwd);
- }
-
if (nc->oldns > -1) {
+ setns(nc->oldns, CLONE_NEWNS);
close(nc->oldns);
nc->oldns = -1;
}
@@ -426,6 +502,11 @@ void nsinfo__mountns_exit(struct nscookie *nc)
close(nc->newns);
nc->newns = -1;
}
+
+ if (nc->oldcwd) {
+ WARN_ON_ONCE(chdir(nc->oldcwd));
+ zfree(&nc->oldcwd);
+ }
}
char *nsinfo__realpath(const char *path, struct nsinfo *nsi)
diff --git a/tools/perf/util/namespaces.h b/tools/perf/util/namespaces.h
index 42dcca13b002f153..7bba217aa3fde3d7 100644
--- a/tools/perf/util/namespaces.h
+++ b/tools/perf/util/namespaces.h
@@ -37,7 +37,9 @@ DECLARE_RC_STRUCT(nsinfo) {
pid_t tid;
pid_t tgid;
pid_t nstgid;
+ ino_t mntns_ino;
bool inited;
+ bool exited;
bool need_setns;
bool in_pidns;
char *mntns_path;
@@ -60,6 +62,8 @@ void nsinfo__put(struct nsinfo *nsi);
bool nsinfo__need_setns(const struct nsinfo *nsi);
void nsinfo__clear_need_setns(struct nsinfo *nsi);
+void nsinfo__set_exited(struct nsinfo *nsi);
+bool nsinfo__check_proc(struct nsinfo *nsi);
pid_t nsinfo__tgid(const struct nsinfo *nsi);
pid_t nsinfo__nstgid(const struct nsinfo *nsi);
pid_t nsinfo__pid(const struct nsinfo *nsi);
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 07/11] perf tools: Check thread liveness when accessing /proc
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (5 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns() Namhyung Kim
` (3 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Add thread__exited() and thread__is_alive(), and mark threads as exited
on PERF_RECORD_EXIT, replaced parent and child threads on
PERF_RECORD_FORK, and perf_event__exit_del_thread().
Propagate thread exit and boot ID mismatch to thread->nsinfo, guard
/proc accesses with thread__is_alive() and nsinfo__check_proc() (for
dso__filename_with_chroot()), and avoid synthesizing a live parent
thread in thread__e_machine_endian() when a sub-thread is no longer
alive.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/builtin-trace.c | 2 +-
tools/perf/util/dso.c | 7 +++-
tools/perf/util/event.c | 1 +
tools/perf/util/machine.c | 17 +++++++---
tools/perf/util/thread.c | 65 +++++++++++++++++++++++++++++++++++---
tools/perf/util/thread.h | 7 ++--
6 files changed, 86 insertions(+), 13 deletions(-)
diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
index 4094141461f91c32..a21bbf5926617633 100644
--- a/tools/perf/builtin-trace.c
+++ b/tools/perf/builtin-trace.c
@@ -1858,7 +1858,7 @@ static const char *thread__fd_path(struct thread *thread, int fd,
return NULL;
if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
- if (!trace->live)
+ if (!trace->live || !thread__is_alive(thread))
return NULL;
++trace->stats.proc_getname;
if (thread__read_fd_path(thread, fd))
diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index e46cd614eb21184b..36e6c45dd0383073 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -587,7 +587,12 @@ static int do_open(char *name) EXCLUSIVE_LOCKS_REQUIRED(_dso__data_open_lock)
char *dso__filename_with_chroot(const struct dso *dso, const char *filename)
{
- return filename_with_chroot(nsinfo__pid(dso__nsinfo_const(dso)), filename);
+ struct nsinfo *nsi = (struct nsinfo *)dso__nsinfo_const(dso);
+
+ if (!nsinfo__check_proc(nsi))
+ return NULL;
+
+ return filename_with_chroot(nsinfo__pid(nsi), filename);
}
char *dso__get_filename(struct dso *dso, const char *root_dir, bool *decomp,
diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index 2f80e364659fea71..ae0ebda42fa47e8c 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -447,6 +447,7 @@ int perf_event__exit_del_thread(const struct perf_tool *tool __maybe_unused,
event->fork.ppid, event->fork.ptid);
if (thread) {
+ thread__set_exited(thread, /*exited=*/true);
machine__remove_thread(machine, thread);
thread__put(thread);
}
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index ac857b395882f569..53cda0046d24799c 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -517,8 +517,14 @@ static void machine__update_thread_pid(struct machine *machine,
return;
thread__set_pid(th, pid);
- nsinfo__zput(RC_CHK_ACCESS(th)->nsinfo);
- RC_CHK_ACCESS(th)->nsinfo = nsinfo__new_thread(pid, thread__tid(th));
+ thread__reset_nsinfo(th);
+ if (thread__exited(th) || !machine__is_host(machine) ||
+ !perf_env__is_same_boot(machine->env)) {
+ struct nsinfo *nsi = thread__get_nsinfo(th);
+
+ nsinfo__set_exited(nsi);
+ nsinfo__put(nsi);
+ }
if (thread__pid(th) == thread__tid(th))
return;
@@ -1975,6 +1981,7 @@ int machine__process_fork_event(struct machine *machine, union perf_event *event
thread__pid(parent) != (pid_t)event->fork.ppid) {
dump_printf("removing erroneous parent thread %d/%d\n",
thread__pid(parent), thread__tid(parent));
+ thread__set_exited(parent, /*exited=*/true);
machine__remove_thread(machine, parent);
thread__put(parent);
parent = machine__findnew_thread(machine, event->fork.ppid,
@@ -1983,6 +1990,7 @@ int machine__process_fork_event(struct machine *machine, union perf_event *event
/* if a thread currently exists for the thread id remove it */
if (thread != NULL) {
+ thread__set_exited(thread, /*exited=*/true);
machine__remove_thread(machine, thread);
thread__put(thread);
}
@@ -2030,9 +2038,8 @@ int machine__process_exit_event(struct machine *machine, union perf_event *event
/* There is no context switch out before exit, so we decrement here. */
machine->parallelism--;
if (thread != NULL) {
- if (symbol_conf.keep_exited_threads)
- thread__set_exited(thread, /*exited=*/true);
- else
+ thread__set_exited(thread, /*exited=*/true);
+ if (!symbol_conf.keep_exited_threads)
machine__remove_thread(machine, thread);
}
thread__put(thread);
diff --git a/tools/perf/util/thread.c b/tools/perf/util/thread.c
index 5d1c886add3e61eb..fda718ea5d7e6276 100644
--- a/tools/perf/util/thread.c
+++ b/tools/perf/util/thread.c
@@ -37,6 +37,16 @@ int thread__init_maps(struct thread *thread, struct machine *machine)
}
}
+ if (thread__pid(thread) != 0 || thread__tid(thread) != 0) {
+ if (!machine || !machine__is_host(machine) ||
+ !perf_env__is_same_boot(machine->env)) {
+ struct nsinfo *nsi = thread__get_nsinfo(thread);
+
+ nsinfo__set_exited(nsi);
+ nsinfo__put(nsi);
+ }
+ }
+
return thread__maps(thread) ? 0 : -1;
}
@@ -194,9 +204,17 @@ static int __thread__set_namespaces(struct thread *thread, u64 timestamp,
static void __thread__reset_nsinfo(struct thread *thread)
EXCLUSIVE_LOCKS_REQUIRED(thread__namespaces_lock(thread))
{
+ struct maps *maps = thread__maps(thread);
+ struct machine *machine = maps ? maps__machine(maps) : NULL;
+ struct nsinfo *nsi;
+
nsinfo__zput(RC_CHK_ACCESS(thread)->nsinfo);
- RC_CHK_ACCESS(thread)->nsinfo = nsinfo__new_thread(thread__pid(thread),
- thread__tid(thread));
+ nsi = nsinfo__new_thread(thread__pid(thread), thread__tid(thread));
+ if (nsi && (thread__exited(thread) ||
+ (machine && (!machine__is_host(machine) ||
+ !perf_env__is_same_boot(machine->env)))))
+ nsinfo__set_exited(nsi);
+ RC_CHK_ACCESS(thread)->nsinfo = nsi;
}
void thread__reset_nsinfo(struct thread *thread)
@@ -317,6 +335,39 @@ int __thread__set_comm(struct thread *thread, const char *str, u64 timestamp,
return ret;
}
+void thread__set_exited(struct thread *thread, bool exited)
+{
+ RC_CHK_ACCESS(thread)->exited = exited;
+ if (exited) {
+ struct nsinfo *nsi = thread__get_nsinfo(thread);
+
+ nsinfo__set_exited(nsi);
+ nsinfo__put(nsi);
+ }
+}
+
+bool thread__is_alive(struct thread *thread)
+{
+ struct maps *maps;
+ struct machine *machine;
+ struct nsinfo *nsi;
+ bool ret;
+
+ if (!thread || thread__exited(thread))
+ return false;
+
+ maps = thread__maps(thread);
+ machine = maps ? maps__machine(maps) : NULL;
+ if (machine && (!machine__is_host(machine) ||
+ !perf_env__is_same_boot(machine->env)))
+ return false;
+
+ nsi = thread__get_nsinfo(thread);
+ ret = nsinfo__check_proc(nsi);
+ nsinfo__put(nsi);
+ return ret;
+}
+
int thread__set_comm_from_proc(struct thread *thread)
{
char path[64];
@@ -324,6 +375,9 @@ int thread__set_comm_from_proc(struct thread *thread)
size_t sz;
int err = -1;
+ if (!thread__is_alive(thread))
+ return -1;
+
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) {
@@ -542,7 +596,10 @@ uint16_t thread__e_machine_endian(struct thread *thread, struct machine *machine
tid = thread__tid(thread);
pid = thread__pid(thread);
if (pid != tid) {
- struct thread *parent = machine__findnew_thread(machine, pid, pid);
+ struct thread *parent = machine__find_thread(machine, pid, pid);
+
+ if (!parent && thread__is_alive(thread))
+ parent = machine__findnew_thread(machine, pid, pid);
if (parent) {
e_machine = thread__e_machine_endian(parent, machine, &local_e_flags,
@@ -569,7 +626,7 @@ uint16_t thread__e_machine_endian(struct thread *thread, struct machine *machine
is_live = !session || !session->data;
}
/* Read from /proc/pid/exe if live. */
- if (is_live) {
+ if (is_live && thread__is_alive(thread)) {
e_machine = read_proc_e_machine_for_pid(pid, &local_e_flags,
&args.is_big_endian);
} else if (machine && machine->env) {
diff --git a/tools/perf/util/thread.h b/tools/perf/util/thread.h
index 36f2932422df8988..927b2ddccfc8433d 100644
--- a/tools/perf/util/thread.h
+++ b/tools/perf/util/thread.h
@@ -205,11 +205,14 @@ static inline refcount_t *thread__refcnt(struct thread *thread)
return &RC_CHK_ACCESS(thread)->refcnt;
}
-static inline void thread__set_exited(struct thread *thread, bool exited)
+static inline bool thread__exited(const struct thread *thread)
{
- RC_CHK_ACCESS(thread)->exited = exited;
+ return RC_CHK_ACCESS(thread)->exited;
}
+void thread__set_exited(struct thread *thread, bool exited);
+bool thread__is_alive(struct thread *thread);
+
static inline bool thread__comm_set(const struct thread *thread)
{
return RC_CHK_ACCESS(thread)->comm_set;
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns()
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (6 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 07/11] perf tools: Check thread liveness when accessing /proc Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 09/11] perf tools: Only enter mount namespace if necessary Namhyung Kim
` (2 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Export filename__read_build_id_ns(), including a chroot fallback guarded
by nsinfo__check_proc(), and use it in the dsos callback and
perf-inject's dso__read_build_id() instead of open-coding
nsinfo__mountns_enter() and dso__filename_with_chroot() fallbacks.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/builtin-inject.c | 12 +-----------
tools/perf/util/build-id.c | 14 +++++++++++---
tools/perf/util/dsos.c | 17 +++--------------
tools/perf/util/symbol.h | 3 +++
4 files changed, 18 insertions(+), 28 deletions(-)
diff --git a/tools/perf/builtin-inject.c b/tools/perf/builtin-inject.c
index 05bf1aae61778952..71ebb5b9ab84cbf7 100644
--- a/tools/perf/builtin-inject.c
+++ b/tools/perf/builtin-inject.c
@@ -894,24 +894,14 @@ static int perf_event__repipe_tracing_data(const struct perf_tool *tool,
static int dso__read_build_id(struct dso *dso)
{
- struct nscookie nsc;
struct build_id bid = { .size = 0, };
if (dso__has_build_id(dso))
return 0;
mutex_lock(dso__lock(dso));
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
- if (filename__read_build_id(dso__long_name(dso), &bid) > 0)
+ if (filename__read_build_id_ns(dso__long_name(dso), &bid, dso__nsinfo(dso)) > 0)
dso__set_build_id(dso, &bid);
- else if (dso__nsinfo(dso)) {
- char *new_name = dso__filename_with_chroot(dso, dso__long_name(dso));
-
- if (new_name && filename__read_build_id(new_name, &bid) > 0)
- dso__set_build_id(dso, &bid);
- free(new_name);
- }
- nsinfo__mountns_exit(&nsc);
mutex_unlock(dso__lock(dso));
return dso__has_build_id(dso) ? 0 : -1;
diff --git a/tools/perf/util/build-id.c b/tools/perf/util/build-id.c
index 02233ef0eafff2c2..ff083a594b405208 100644
--- a/tools/perf/util/build-id.c
+++ b/tools/perf/util/build-id.c
@@ -847,15 +847,23 @@ int build_id_cache__remove_s(const char *sbuild_id)
return err;
}
-static int filename__read_build_id_ns(const char *filename,
- struct build_id *bid,
- struct nsinfo *nsi)
+int filename__read_build_id_ns(const char *filename,
+ struct build_id *bid,
+ struct nsinfo *nsi)
{
struct nscookie nsc;
int ret;
nsinfo__mountns_enter(nsi, &nsc);
ret = filename__read_build_id(filename, bid);
+ if (ret < 0 && errno == ENOENT && nsinfo__check_proc(nsi)) {
+ char *new_name = filename_with_chroot(nsinfo__pid(nsi), filename);
+
+ if (new_name) {
+ ret = filename__read_build_id(new_name, bid);
+ free(new_name);
+ }
+ }
nsinfo__mountns_exit(&nsc);
return ret;
diff --git a/tools/perf/util/dsos.c b/tools/perf/util/dsos.c
index e927e707abace9c6..a97fed55286aefca 100644
--- a/tools/perf/util/dsos.c
+++ b/tools/perf/util/dsos.c
@@ -4,11 +4,11 @@
#include "dso.h"
#include "util.h"
#include "vdso.h"
-#include "namespaces.h"
#include <errno.h>
+#include <linux/perf_event.h>
#include <stdlib.h>
#include <string.h>
-#include <symbol.h> // filename__read_build_id
+#include <symbol.h> // filename__read_build_id_ns
#include <unistd.h>
void dsos__init(struct dsos *dsos)
@@ -70,7 +70,6 @@ struct dsos__read_build_ids_cb_args {
static int dsos__read_build_ids_cb(struct dso *dso, void *data)
{
struct dsos__read_build_ids_cb_args *args = data;
- struct nscookie nsc;
struct build_id bid = { .size = 0, };
if (args->with_hits && !dso__hit(dso) && !dso__is_vdso(dso))
@@ -79,20 +78,10 @@ static int dsos__read_build_ids_cb(struct dso *dso, void *data)
args->have_build_id = true;
return 0;
}
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
- if (filename__read_build_id(dso__long_name(dso), &bid) > 0) {
+ if (filename__read_build_id_ns(dso__long_name(dso), &bid, dso__nsinfo(dso)) > 0) {
dso__set_build_id(dso, &bid);
args->have_build_id = true;
- } else if (errno == ENOENT && dso__nsinfo(dso)) {
- char *new_name = dso__filename_with_chroot(dso, dso__long_name(dso));
-
- if (new_name && filename__read_build_id(new_name, &bid) > 0) {
- dso__set_build_id(dso, &bid);
- args->have_build_id = true;
- }
- free(new_name);
}
- nsinfo__mountns_exit(&nsc);
return 0;
}
diff --git a/tools/perf/util/symbol.h b/tools/perf/util/symbol.h
index 6347bd218ad5eac4..7ac362b5d53530b0 100644
--- a/tools/perf/util/symbol.h
+++ b/tools/perf/util/symbol.h
@@ -28,6 +28,7 @@
struct dso;
struct map;
struct maps;
+struct nsinfo;
struct option;
struct build_id;
struct perf_env;
@@ -211,6 +212,8 @@ struct symbol *dso__next_symbol(struct symbol *sym);
enum dso_type dso__type_fd(int fd);
int filename__read_build_id(const char *filename, struct build_id *id);
+int filename__read_build_id_ns(const char *filename, struct build_id *bid,
+ struct nsinfo *nsi);
int sysfs__read_build_id(const char *filename, struct build_id *bid);
int modules__parse(const char *filename, void *arg,
int (*process_module)(void *arg, const char *name,
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 09/11] perf tools: Only enter mount namespace if necessary
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (7 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns() Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load() Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 11/11] perf tests: Add namespace, nsinfo, and boot ID tests Namhyung Kim
10 siblings, 0 replies; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Add dso_binary_type__needs_ns() to distinguish DSO binary and debuginfo
types that reside inside the target process's mount namespace from
host-side paths.
Gate mount namespace entry in open_dso(), dso__find_binary_type_locked(),
dso__find_dbginfo_type(), __dso__read_symbol(), debuginfo__new(),
symbol__disassemble(), and symbol__disassemble_capstone_powerpc() using
dso_binary_type__needs_ns().
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/capstone.c | 8 ++++--
tools/perf/util/debuginfo.c | 19 +++++++++-----
tools/perf/util/disasm.c | 8 +++++-
tools/perf/util/dso.c | 52 ++++++++++++++++++++++++++++++-------
tools/perf/util/dso.h | 34 ++++++++++++++++++++++++
5 files changed, 102 insertions(+), 19 deletions(-)
diff --git a/tools/perf/util/capstone.c b/tools/perf/util/capstone.c
index 95881f15324aa614..e1de77879c5e5f97 100644
--- a/tools/perf/util/capstone.c
+++ b/tools/perf/util/capstone.c
@@ -649,13 +649,17 @@ int symbol__disassemble_capstone_powerpc(const char *filename __maybe_unused,
bool disassembler_style = false;
uint16_t e_machine = EM_NONE;
bool is_big_endian = false;
+ bool nsenter;
if (args->options->objdump_path)
return -1;
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
+ nsenter = dso_binary_type__needs_ns(dso__binary_type(dso));
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
fd = open(filename, O_RDONLY);
- nsinfo__mountns_exit(&nsc);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (fd < 0)
return -1;
diff --git a/tools/perf/util/debuginfo.c b/tools/perf/util/debuginfo.c
index db188c397d63880c..d339906c28b41425 100644
--- a/tools/perf/util/debuginfo.c
+++ b/tools/perf/util/debuginfo.c
@@ -17,6 +17,7 @@
#include "dso.h"
#include "debug.h"
#include "debuginfo.h"
+#include "namespaces.h"
#include "symbol.h"
#ifdef HAVE_DEBUGINFOD_SUPPORT
@@ -93,17 +94,23 @@ static struct debuginfo *__debuginfo__new(const char *path)
struct debuginfo *debuginfo__new(struct dso *dso)
{
- struct debuginfo *dinfo;
+ struct debuginfo *dinfo = NULL;
+ struct nscookie nsc;
char *path;
bool decomp = false;
+ bool nsenter;
dso__find_dbginfo_type(dso);
+ nsenter = dso_binary_type__needs_ns(dso__dbginfo_type(dso));
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
path = dso__get_filename(dso, "", &decomp, dso__dbginfo_type(dso));
- if (path == NULL)
- return NULL;
-
- dinfo = __debuginfo__new(path);
- dso__put_filename(dso, path, decomp);
+ if (path != NULL) {
+ dinfo = __debuginfo__new(path);
+ dso__put_filename(dso, path, decomp);
+ }
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
return dinfo;
}
diff --git a/tools/perf/util/disasm.c b/tools/perf/util/disasm.c
index ca478191813cf46f..5b9bc94be55ef088 100644
--- a/tools/perf/util/disasm.c
+++ b/tools/perf/util/disasm.c
@@ -1499,7 +1499,8 @@ int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
bool delete_extract = false;
struct kcore_extract kce;
enum dso_binary_type dbt;
- bool decomp = false;
+ struct nscookie nsc;
+ bool decomp = false, nsenter;
int err;
dbt = dso__find_binary_type(dso);
@@ -1507,8 +1508,13 @@ int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
if (dbt == DSO_BINARY_TYPE__KALLSYMS)
return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
+ nsenter = dso_binary_type__needs_ns(dbt);
mutex_lock(dso__lock(dso));
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
symfs_filename = dso__get_filename(dso, "", &decomp, dbt);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
mutex_unlock(dso__lock(dso));
if (symfs_filename == NULL) {
diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index 36e6c45dd0383073..d4fc702b4c338d96 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -712,14 +712,15 @@ static int open_dso(struct dso *dso, struct machine *machine)
{
int fd;
struct nscookie nsc;
+ bool nsenter = dso_binary_type__needs_ns(dso__binary_type(dso));
- if (dso__binary_type(dso) != DSO_BINARY_TYPE__BUILD_ID_CACHE) {
+ if (nsenter) {
mutex_lock(dso__lock(dso));
nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
mutex_unlock(dso__lock(dso));
}
fd = __open_dso(dso, machine);
- if (dso__binary_type(dso) != DSO_BINARY_TYPE__BUILD_ID_CACHE)
+ if (nsenter)
nsinfo__mountns_exit(&nsc);
if (fd >= 0) {
@@ -897,12 +898,22 @@ static int dso__find_binary_type_locked(struct dso *dso)
}
do {
- dso__set_binary_type(dso, binary_type_data[i++]);
+ struct nscookie nsc;
+ enum dso_binary_type type = binary_type_data[i++];
+ bool nsenter = dso_binary_type__needs_ns(type);
+
+ dso__set_binary_type(dso, type);
+
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
errno = 0;
- path = dso__get_filename(dso, "", &decomp, dso__binary_type(dso));
+ path = dso__get_filename(dso, "", &decomp, type);
if (path == NULL) {
- if (errno == EMFILE)
+ saved_errno = errno;
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
+ if (saved_errno == EMFILE)
goto open_err;
continue;
}
@@ -912,9 +923,11 @@ static int dso__find_binary_type_locked(struct dso *dso)
if (!found)
saved_errno = errno;
dso__put_filename(dso, path, decomp);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (found)
- return dso__binary_type(dso);
+ return type;
else if (saved_errno == EMFILE)
goto open_err;
@@ -2119,11 +2132,14 @@ static const u8 *__dso__read_symbol(struct dso *dso, const char *symfs_filename,
};
u8 *code_buf = NULL;
int saved_errno;
+ bool nsenter = dso_binary_type__needs_ns(dso__binary_type(dso));
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
fd = open(symfs_filename, O_RDONLY | O_CLOEXEC);
saved_errno = errno;
- nsinfo__mountns_exit(&nsc);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (fd < 0) {
errno = saved_errno;
return NULL;
@@ -2244,8 +2260,9 @@ void dso__find_dbginfo_type(struct dso *dso)
DSO_BINARY_TYPE__NOT_FOUND,
};
const enum dso_binary_type *type;
+ struct nscookie nsc;
char *path;
- bool found, decomp = false;
+ bool found, decomp = false, nsenter;
if (dso__dbginfo_type(dso) != DSO_BINARY_TYPE__NOT_FOUND)
return;
@@ -2255,28 +2272,43 @@ void dso__find_dbginfo_type(struct dso *dso)
dso__set_debuginfo_searched(dso);
/* Maybe debug info is in the same file with the symbol table */
+ nsenter = dso_binary_type__needs_ns(dso__symtab_type(dso));
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
path = dso__get_filename(dso, "", &decomp, dso__symtab_type(dso));
if (path) {
found = filename__has_section(path, ".debug_info") ||
filename__has_section(path, ".zdebug_info");
dso__put_filename(dso, path, decomp);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (found) {
dso__set_dbginfo_type(dso, dso__symtab_type(dso));
return;
}
+ } else if (nsenter) {
+ nsinfo__mountns_exit(&nsc);
}
/* Otherwise check distro debug locations */
for (type = dbginfo_types; *type != DSO_BINARY_TYPE__NOT_FOUND; type++) {
+ nsenter = dso_binary_type__needs_ns(*type);
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
path = dso__get_filename(dso, "", &decomp, *type);
- if (path == NULL)
+ if (path == NULL) {
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
continue;
+ }
found = filename__has_section(path, ".debug_info") ||
filename__has_section(path, ".zdebug_info");
dso__put_filename(dso, path, decomp);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (found)
break;
}
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index d5f7382d4e13a371..7cac0960e1a6dd32 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -141,6 +141,40 @@ enum dso_binary_type {
DSO_BINARY_TYPE__NOT_FOUND,
};
+static inline bool dso_binary_type__needs_ns(enum dso_binary_type type)
+{
+ switch (type) {
+ case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
+ case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
+ case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
+ case DSO_BINARY_TYPE__DEBUGLINK:
+ case DSO_BINARY_TYPE__GNU_DEBUGDATA:
+ case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
+ case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
+ case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
+ case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
+ case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
+ return true;
+ case DSO_BINARY_TYPE__KALLSYMS:
+ case DSO_BINARY_TYPE__GUEST_KALLSYMS:
+ case DSO_BINARY_TYPE__VMLINUX:
+ case DSO_BINARY_TYPE__GUEST_VMLINUX:
+ case DSO_BINARY_TYPE__JAVA_JIT:
+ case DSO_BINARY_TYPE__BUILD_ID_CACHE:
+ case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
+ case DSO_BINARY_TYPE__GUEST_KMODULE:
+ case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
+ case DSO_BINARY_TYPE__KCORE:
+ case DSO_BINARY_TYPE__GUEST_KCORE:
+ case DSO_BINARY_TYPE__BPF_PROG_INFO:
+ case DSO_BINARY_TYPE__BPF_IMAGE:
+ case DSO_BINARY_TYPE__OOL:
+ case DSO_BINARY_TYPE__NOT_FOUND:
+ default:
+ return false;
+ }
+}
+
enum dso_space_type {
DSO_SPACE__USER = 0,
DSO_SPACE__KERNEL,
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load()
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (8 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 09/11] perf tools: Only enter mount namespace if necessary Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 11/11] perf tests: Add namespace, nsinfo, and boot ID tests Namhyung Kim
10 siblings, 1 reply; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Enter the target mount namespace only for the specific binary types
that need it in dso__load(), and use filename__read_build_id_ns() for
BUILDID_DEBUGINFO lookup and in symbol-minimal's dso__load_sym() so it
does not read the wrong build-id from the host filesystem.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/util/symbol-minimal.c | 4 ++-
tools/perf/util/symbol.c | 45 +++++++++++++++-----------------
2 files changed, 24 insertions(+), 25 deletions(-)
diff --git a/tools/perf/util/symbol-minimal.c b/tools/perf/util/symbol-minimal.c
index 0a71d146395271a6..5beb73fd508ad079 100644
--- a/tools/perf/util/symbol-minimal.c
+++ b/tools/perf/util/symbol-minimal.c
@@ -342,7 +342,9 @@ int dso__load_sym(struct dso *dso, struct map *map __maybe_unused,
if (ret >= 0)
RC_CHK_ACCESS(dso)->is_64_bit = ret;
- if (filename__read_build_id(ss->name, &bid) > 0)
+ if (filename__read_build_id_ns(ss->name, &bid,
+ dso_binary_type__needs_ns(ss->type) ?
+ dso__nsinfo(dso) : NULL) > 0)
dso__set_build_id(dso, &bid);
return 0;
}
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index dc70bbe8514f4d37..4c0ff9f0c32f8293 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -1811,17 +1811,6 @@ int dso__load(struct dso *dso, struct map *map)
const char *map_path = dso__long_name(dso);
mutex_lock(dso__lock(dso));
- perfmap = is_perf_pid_map_name(map_path);
-
- if (perfmap) {
- if (dso__nsinfo(dso) &&
- (dso__find_perf_map(newmapname, sizeof(newmapname),
- dso__nsinfo_ptr(dso)) == 0)) {
- map_path = newmapname;
- }
- }
-
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
/* check again under the dso->lock */
if (dso__loaded(dso)) {
@@ -1845,8 +1834,16 @@ int dso__load(struct dso *dso, struct map *map)
dso__set_adjust_symbols(dso, false);
+ perfmap = is_perf_pid_map_name(map_path);
if (perfmap) {
+ if (dso__nsinfo(dso) &&
+ (dso__find_perf_map(newmapname, sizeof(newmapname),
+ dso__nsinfo_ptr(dso)) == 0)) {
+ map_path = newmapname;
+ }
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
ret = dso__load_perf_map(map_path, dso);
+ nsinfo__mountns_exit(&nsc);
dso__set_binary_type(dso, DSO_BINARY_TYPE__JAVA_JIT);
dso__set_symtab_type(dso, ret > 0
? DSO_BINARY_TYPE__JAVA_JIT
@@ -1869,7 +1866,7 @@ int dso__load(struct dso *dso, struct map *map)
struct build_id bid = { .size = 0, };
__symbol__join_symfs(name, PATH_MAX, dso__long_name(dso));
- if (filename__read_build_id(name, &bid) > 0)
+ if (filename__read_build_id_ns(name, &bid, dso__nsinfo(dso)) > 0)
dso__set_build_id(dso, &bid);
}
@@ -1882,27 +1879,28 @@ int dso__load(struct dso *dso, struct map *map)
struct symsrc *ss = &ss_[ss_pos];
bool next_slot = false;
bool is_reg;
- bool nsexit;
+ bool nsenter;
int bfdrc = -1;
int sirc = -1;
enum dso_binary_type symtab_type = binary_type_symtab[i];
- nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
- symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
-
if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
continue;
+ nsenter = dso_binary_type__needs_ns(symtab_type);
+ if (nsenter)
+ nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
+
if (dso__read_binary_type_filename(dso, symtab_type,
- root_dir, name, PATH_MAX))
+ root_dir, name, PATH_MAX)) {
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
continue;
-
- if (nsexit)
- nsinfo__mountns_exit(&nsc);
+ }
is_reg = is_regular_file(name);
- if (!is_reg && errno == ENOENT && dso__nsinfo(dso)) {
+ if (!is_reg && errno == ENOENT && nsenter && dso__nsinfo(dso)) {
char *new_name = dso__filename_with_chroot(dso, name);
if (new_name) {
is_reg = is_regular_file(new_name);
@@ -1918,8 +1916,8 @@ int dso__load(struct dso *dso, struct map *map)
if (is_reg && bfdrc < 0)
sirc = symsrc__init(ss, dso, name, symtab_type);
- if (nsexit)
- nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
+ if (nsenter)
+ nsinfo__mountns_exit(&nsc);
if (bfdrc > 0) {
dso__set_symtab_type(dso, symtab_type);
@@ -1986,7 +1984,6 @@ int dso__load(struct dso *dso, struct map *map)
out:
dso__set_loaded(dso);
mutex_unlock(dso__lock(dso));
- nsinfo__mountns_exit(&nsc);
return ret;
}
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v2 11/11] perf tests: Add namespace, nsinfo, and boot ID tests
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
` (9 preceding siblings ...)
2026-11-16 5:57 ` [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load() Namhyung Kim
@ 2026-11-16 5:57 ` Namhyung Kim
10 siblings, 0 replies; 20+ messages in thread
From: Namhyung Kim @ 2026-11-16 5:57 UTC (permalink / raw)
To: Arnaldo Carvalho de Melo
Cc: Ian Rogers, Jiri Olsa, Adrian Hunter, James Clark, Peter Zijlstra,
Ingo Molnar, LKML, linux-perf-users
Add a "Namespace and nsinfo tests" suite covering HEADER_BOOT_ID,
perf_env__is_same_boot(), deferred /proc lookup, raw nsinfo__copy(),
PID/TID recycling detection, thread exit and fork liveness tracking,
dso_binary_type__needs_ns(), and isolated mountns and chroot resolution.
Also verify the boot ID header matches /proc/sys/kernel/random/boot_id
in the perf header shell test for both file and pipe modes.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
tools/perf/tests/Build | 1 +
tools/perf/tests/builtin-test.c | 1 +
tools/perf/tests/namespaces.c | 1035 ++++++++++++++++++++++++++++++
| 9 +
tools/perf/tests/tests.h | 1 +
5 files changed, 1047 insertions(+)
create mode 100644 tools/perf/tests/namespaces.c
diff --git a/tools/perf/tests/Build b/tools/perf/tests/Build
index 5eebb2ec5ddefcfd..7f2e926eb0bc7674 100644
--- a/tools/perf/tests/Build
+++ b/tools/perf/tests/Build
@@ -74,6 +74,7 @@ perf-test-y += hwmon_pmu.o
perf-test-y += tool_pmu.o
perf-test-y += subcmd-help.o
perf-test-y += kallsyms-split.o
+perf-test-y += namespaces.o
ifeq ($(SRCARCH),$(filter $(SRCARCH),x86 arm arm64 powerpc riscv))
perf-test-$(CONFIG_DWARF_UNWIND) += dwarf-unwind.o
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c
index 327483f96401b580..f3c041ff1acf0289 100644
--- a/tools/perf/tests/builtin-test.c
+++ b/tools/perf/tests/builtin-test.c
@@ -192,6 +192,7 @@ static struct test_suite *generic_tests[] = {
&suite__util,
&suite__subcmd_help,
&suite__kallsyms_split,
+ &suite__namespaces,
NULL,
};
diff --git a/tools/perf/tests/namespaces.c b/tools/perf/tests/namespaces.c
new file mode 100644
index 0000000000000000..cfc654db48f92292
--- /dev/null
+++ b/tools/perf/tests/namespaces.c
@@ -0,0 +1,1035 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <errno.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <pthread.h>
+#include <sched.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <sys/mount.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <api/fs/fs.h>
+#include <linux/compiler.h>
+#include <linux/kernel.h>
+#include <linux/perf_event.h>
+#include <linux/string.h>
+#include <linux/zalloc.h>
+
+#include "debug.h"
+#include "tests.h"
+#include "util/build-id.h"
+#include "util/dso.h"
+#include "util/env.h"
+#include "util/event.h"
+#include "util/machine.h"
+#include "util/map.h"
+#include "util/namespaces.h"
+#include "util/symbol.h"
+#include "util/thread.h"
+#include "util/tool.h"
+#include "util/util.h"
+
+#define CHILD_EXIT_SKIP 2
+
+static int read_host_boot_id_str(char *buf, size_t sz)
+{
+ char *tmp = NULL;
+ size_t len = 0;
+
+ if (procfs__read_str("sys/kernel/random/boot_id", &tmp, &len) < 0)
+ return -1;
+
+ strlcpy(buf, strim(tmp), sz);
+ free(tmp);
+ return buf[0] ? 0 : -1;
+}
+
+static int test__nsinfo_boot_id(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ struct perf_env live_env, off_env;
+ char host_boot_id[64];
+ const char *bid;
+ int ret = TEST_FAIL;
+
+ if (read_host_boot_id_str(host_boot_id, sizeof(host_boot_id)) < 0) {
+ pr_debug("SKIP: cannot read host boot_id from procfs\n");
+ return TEST_SKIP;
+ }
+
+ perf_env__init(&live_env);
+ perf_env__init(&off_env);
+
+ TEST_ASSERT_VAL("NULL env should match current host boot",
+ perf_env__is_same_boot(NULL));
+
+ TEST_ASSERT_VAL("live env should match current host boot",
+ perf_env__is_same_boot(&live_env));
+ bid = perf_env__boot_id(&live_env);
+ if (!bid || strcmp(bid, host_boot_id) != 0) {
+ pr_debug("unexpected live env boot_id: %s vs %s\n",
+ bid ? bid : "(null)", host_boot_id);
+ goto out;
+ }
+
+ /* Offline environments must not fall back to or cache the host boot_id. */
+ off_env.offline = true;
+ if (perf_env__boot_id(&off_env) != NULL) {
+ pr_debug("offline env without boot_id unexpectedly returned host boot_id\n");
+ goto out;
+ }
+ if (off_env.boot_id != NULL) {
+ pr_debug("offline env->boot_id was poisoned\n");
+ goto out;
+ }
+ if (perf_env__is_same_boot(&off_env)) {
+ pr_debug("offline env without boot_id should not match current boot\n");
+ goto out;
+ }
+
+ off_env.boot_id = strdup(host_boot_id);
+ if (!off_env.boot_id)
+ goto out;
+ if (!perf_env__is_same_boot(&off_env)) {
+ pr_debug("offline env with matching boot_id failed to match\n");
+ goto out;
+ }
+
+ zfree(&off_env.boot_id);
+ off_env.boot_id = strdup("00000000-0000-0000-0000-000000000000");
+ if (!off_env.boot_id)
+ goto out;
+ if (perf_env__is_same_boot(&off_env)) {
+ pr_debug("offline env with mismatched boot_id unexpectedly matched\n");
+ goto out;
+ }
+
+ ret = TEST_OK;
+out:
+ perf_env__exit(&off_env);
+ perf_env__exit(&live_env);
+ return ret;
+}
+
+static int test__nsinfo_init_and_copy(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ struct nsinfo *nsi = NULL, *copy = NULL, *copy2 = NULL;
+ pid_t pid = getpid();
+ int ret = TEST_FAIL;
+
+ if (nsinfo__new_thread(0, pid) != NULL ||
+ nsinfo__new_thread(pid, 0) != NULL ||
+ nsinfo__new_thread(-1, -1) != NULL ||
+ nsinfo__copy(NULL) != NULL ||
+ nsinfo__need_setns(NULL) ||
+ nsinfo__check_proc(NULL)) {
+ pr_debug("invalid nsinfo boundary condition succeeded\n");
+ return TEST_FAIL;
+ }
+
+ nsi = nsinfo__new_thread(pid, pid);
+ if (!nsi)
+ return TEST_FAIL;
+
+ if (RC_CHK_ACCESS(nsi)->inited || RC_CHK_ACCESS(nsi)->exited) {
+ pr_debug("nsinfo__new_thread() unexpectedly ran /proc init eagerly\n");
+ goto out;
+ }
+ if (nsinfo__pid(nsi) != pid || nsinfo__tid(nsi) != pid) {
+ pr_debug("unexpected pid/tid on new nsinfo\n");
+ goto out;
+ }
+
+ /*
+ * Copying before initialization (e.g., for [vdso] mappings) must not
+ * force eager /proc lookups on the source nsinfo.
+ */
+ copy = nsinfo__copy(nsi);
+ if (!copy)
+ goto out;
+ if (RC_CHK_ACCESS(nsi)->inited || RC_CHK_ACCESS(copy)->inited) {
+ pr_debug("nsinfo__copy() unexpectedly triggered lazy init\n");
+ goto out;
+ }
+
+ nsinfo__clear_need_setns(copy);
+ if (!RC_CHK_ACCESS(copy)->inited || RC_CHK_ACCESS(copy)->need_setns) {
+ pr_debug("nsinfo__clear_need_setns() state mismatch\n");
+ goto out;
+ }
+
+ /* Querying triggers lazy initialization cleanly. */
+ if (nsinfo__need_setns(nsi)) {
+ pr_debug("same-namespace task unexpectedly needs setns\n");
+ goto out;
+ }
+ if (!RC_CHK_ACCESS(nsi)->inited || RC_CHK_ACCESS(nsi)->exited ||
+ nsinfo__tgid(nsi) != pid || nsinfo__nstgid(nsi) <= 0) {
+ pr_debug("lazy init failed to populate expected fields\n");
+ goto out;
+ }
+
+ copy2 = nsinfo__copy(nsi);
+ if (!copy2)
+ goto out;
+ if (!RC_CHK_ACCESS(copy2)->inited ||
+ nsinfo__tgid(copy2) != nsinfo__tgid(nsi) ||
+ nsinfo__nstgid(copy2) != nsinfo__nstgid(nsi) ||
+ nsinfo__in_pidns(copy2) != nsinfo__in_pidns(nsi)) {
+ pr_debug("nsinfo__copy() did not preserve initialized fields\n");
+ goto out;
+ }
+
+ ret = TEST_OK;
+out:
+ nsinfo__put(copy2);
+ nsinfo__put(copy);
+ nsinfo__put(nsi);
+ return ret;
+}
+
+struct worker_ctx {
+ pthread_mutex_t lock;
+ pthread_cond_t cond;
+ bool ready;
+ bool stop;
+ pid_t tid;
+};
+
+static void *worker_thread_fn(void *arg)
+{
+ struct worker_ctx *ctx = arg;
+
+ pthread_mutex_lock(&ctx->lock);
+ ctx->tid = syscall(SYS_gettid);
+ ctx->ready = true;
+ pthread_cond_broadcast(&ctx->cond);
+ while (!ctx->stop)
+ pthread_cond_wait(&ctx->cond, &ctx->lock);
+ pthread_mutex_unlock(&ctx->lock);
+
+ return NULL;
+}
+
+static pid_t spawn_reaped_child(void)
+{
+ pid_t child;
+ int status;
+
+ child = fork();
+ if (child < 0)
+ return -1;
+ if (child == 0)
+ _exit(0);
+
+ if (waitpid(child, &status, 0) != child)
+ return -1;
+
+ return child;
+}
+
+static int test__nsinfo_proc_and_recycled_tid(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ struct worker_ctx wctx = {
+ .lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ };
+ struct nsinfo *dead1 = NULL, *dead2 = NULL;
+ struct nsinfo *valid = NULL, *recycled1 = NULL, *recycled2 = NULL;
+ pthread_t thr;
+ bool started = false;
+ pid_t dead_pid, self_pid = getpid();
+ int ret = TEST_FAIL;
+
+ dead_pid = spawn_reaped_child();
+ if (dead_pid <= 0) {
+ pr_debug("SKIP: cannot fork dead child\n");
+ return TEST_SKIP;
+ }
+
+ dead1 = nsinfo__new_thread(dead_pid, dead_pid);
+ dead2 = nsinfo__new_thread(dead_pid, dead_pid);
+ if (!dead1 || !dead2)
+ goto out;
+
+ if (nsinfo__check_proc(dead1)) {
+ pr_debug("reaped child unexpectedly passed nsinfo__check_proc()\n");
+ goto out;
+ }
+ if (!RC_CHK_ACCESS(dead1)->exited || RC_CHK_ACCESS(dead1)->need_setns) {
+ pr_debug("failed nsinfo__check_proc() did not mark dead1 exited\n");
+ goto out;
+ }
+
+ if (nsinfo__need_setns(dead2)) {
+ pr_debug("reaped child unexpectedly needed setns\n");
+ goto out;
+ }
+ if (!RC_CHK_ACCESS(dead2)->exited || !RC_CHK_ACCESS(dead2)->inited ||
+ nsinfo__tgid(dead2) != dead_pid || nsinfo__nstgid(dead2) != dead_pid ||
+ nsinfo__in_pidns(dead2)) {
+ pr_debug("failed lazy init corrupted fallback fields on dead nsinfo\n");
+ goto out;
+ }
+
+ if (pthread_create(&thr, NULL, worker_thread_fn, &wctx) != 0)
+ goto out;
+ started = true;
+
+ pthread_mutex_lock(&wctx.lock);
+ while (!wctx.ready)
+ pthread_cond_wait(&wctx.cond, &wctx.lock);
+ pthread_mutex_unlock(&wctx.lock);
+
+ if (wctx.tid == self_pid) {
+ pr_debug("worker tid unexpectedly matched tgid\n");
+ goto out;
+ }
+
+ valid = nsinfo__new_thread(self_pid, wctx.tid);
+ if (!valid)
+ goto out;
+ if (!nsinfo__check_proc(valid) || nsinfo__need_setns(valid) ||
+ RC_CHK_ACCESS(valid)->exited || nsinfo__tgid(valid) != self_pid) {
+ pr_debug("valid live thread (pid, tid) failed proc checks\n");
+ goto out;
+ }
+
+ /*
+ * Simulate TID recycling: /proc/<tid>/task/<tid>/status exists in
+ * procfs for our live worker thread, but its Tgid is self_pid != tid.
+ * Both check_proc and lazy init must reject it and preserve fallback
+ * fields without overwriting tgid/nstgid/in_pidns.
+ */
+ recycled1 = nsinfo__new_thread(wctx.tid, wctx.tid);
+ recycled2 = nsinfo__new_thread(wctx.tid, wctx.tid);
+ if (!recycled1 || !recycled2)
+ goto out;
+
+ if (nsinfo__check_proc(recycled1) || !RC_CHK_ACCESS(recycled1)->exited) {
+ pr_debug("mismatched Tgid passed nsinfo__check_proc()\n");
+ goto out;
+ }
+ if (nsinfo__need_setns(recycled2) || !RC_CHK_ACCESS(recycled2)->exited) {
+ pr_debug("mismatched Tgid passed lazy init\n");
+ goto out;
+ }
+ if (nsinfo__tgid(recycled2) != wctx.tid ||
+ nsinfo__nstgid(recycled2) != wctx.tid ||
+ nsinfo__in_pidns(recycled2)) {
+ pr_debug("mismatched Tgid overwrote nsinfo tgid/pidns fields\n");
+ goto out;
+ }
+
+ ret = TEST_OK;
+out:
+ if (started) {
+ pthread_mutex_lock(&wctx.lock);
+ wctx.stop = true;
+ pthread_cond_broadcast(&wctx.cond);
+ pthread_mutex_unlock(&wctx.lock);
+ pthread_join(thr, NULL);
+ }
+ nsinfo__put(recycled2);
+ nsinfo__put(recycled1);
+ nsinfo__put(valid);
+ nsinfo__put(dead2);
+ nsinfo__put(dead1);
+ return ret;
+}
+
+static int test__nsinfo_thread_liveness(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ struct machines live_machines, off_machines;
+ struct machine *live_m, *off_m;
+ struct perf_env live_env, off_env;
+ struct thread *idle_th = NULL, *off_th = NULL, *off_sub = NULL;
+ struct thread *th = NULL, *bad_parent = NULL;
+ struct nsinfo *nsi1 = NULL, *nsi2 = NULL, *nsi3 = NULL;
+ struct nsinfo *old_parent_nsi = NULL, *del_nsi = NULL, *off_nsi = NULL;
+ struct perf_record_namespaces *ns_ev;
+ struct perf_sample sample = { 0 };
+ struct perf_tool tool;
+ union perf_event ev;
+ pid_t self_pid = getpid();
+ int ret = TEST_FAIL;
+
+ perf_env__init(&live_env);
+ perf_env__init(&off_env);
+ off_env.offline = true;
+ perf_tool__init(&tool, /*ordered_events=*/false);
+
+ if (machines__init(&live_machines, NULL) != 0)
+ goto out_env;
+ if (machines__init(&off_machines, NULL) != 0) {
+ machines__exit(&live_machines);
+ goto out_env;
+ }
+
+ live_m = &live_machines.host;
+ live_m->env = &live_env;
+ off_m = &off_machines.host;
+ off_m->env = &off_env;
+
+ /* Idle (0, 0) thread must not evaluate boot_id or populate env->boot_id. */
+ idle_th = machine__findnew_thread(off_m, 0, 0);
+ if (!idle_th)
+ goto out;
+ if (off_env.boot_id != NULL) {
+ pr_debug("idle thread init poisoned offline env->boot_id\n");
+ goto out;
+ }
+ thread__put(idle_th);
+ idle_th = NULL;
+
+ /*
+ * Create both leader and sub-thread before removing the leader so
+ * thread__init_maps() does not re-create (self_pid, self_pid) later,
+ * then verify thread__e_machine_endian() does not recreate it either.
+ */
+ off_th = machine__findnew_thread(off_m, self_pid, self_pid);
+ off_sub = machine__findnew_thread(off_m, self_pid, self_pid + 1);
+ if (!off_th || !off_sub)
+ goto out;
+ off_nsi = thread__get_nsinfo(off_th);
+ if (!off_nsi || !RC_CHK_ACCESS(off_nsi)->exited || thread__is_alive(off_th)) {
+ pr_debug("offline session thread stayed alive\n");
+ goto out;
+ }
+ machine__remove_thread(off_m, off_th);
+ thread__put(off_th);
+ off_th = NULL;
+
+ thread__e_machine_endian(off_sub, off_m, NULL, NULL);
+ if (machine__find_thread(off_m, self_pid, self_pid) != NULL) {
+ pr_debug("thread__e_machine_endian() unexpectedly recreated exited leader\n");
+ goto out;
+ }
+ machine__remove_thread(off_m, off_sub);
+ thread__put(off_sub);
+ off_sub = NULL;
+
+ /* Live thread tracking, reset on exec/namespaces, and exit propagation. */
+ th = machine__findnew_thread(live_m, self_pid, self_pid);
+ if (!th || !thread__is_alive(th)) {
+ pr_debug("current live thread reported as dead\n");
+ goto out;
+ }
+
+ nsi1 = thread__get_nsinfo(th);
+ if (!nsi1)
+ goto out;
+
+ if (__thread__set_comm(th, "execed", 1000, /*exec=*/true) != 0)
+ goto out;
+ nsi2 = thread__get_nsinfo(th);
+ if (!nsi2 || nsi2 == nsi1) {
+ pr_debug("exec did not replace thread->nsinfo\n");
+ goto out;
+ }
+
+ ns_ev = calloc(1, sizeof(*ns_ev) + sizeof(ns_ev->link_info[0]));
+ if (!ns_ev)
+ goto out;
+ ns_ev->nr_namespaces = 1;
+ if (thread__set_namespaces(th, 2000, ns_ev) != 0) {
+ free(ns_ev);
+ goto out;
+ }
+ free(ns_ev);
+
+ nsi3 = thread__get_nsinfo(th);
+ if (!nsi3 || nsi3 == nsi2) {
+ pr_debug("PERF_RECORD_NAMESPACES did not replace thread->nsinfo\n");
+ goto out;
+ }
+
+ thread__set_exited(th, true);
+ if (!thread__exited(th) || thread__is_alive(th) ||
+ !RC_CHK_ACCESS(nsi3)->exited || nsinfo__check_proc(nsi3) ||
+ RC_CHK_ACCESS(nsi1)->exited) {
+ pr_debug("exit propagation or prior DSO nsinfo isolation failed\n");
+ goto out;
+ }
+
+ machine__remove_thread(live_m, th);
+ thread__put(th);
+ th = NULL;
+
+ /* Replaced erroneous parent in machine__process_fork_event() gets marked exited. */
+ bad_parent = machine__findnew_thread(live_m, self_pid + 1, self_pid);
+ if (!bad_parent)
+ goto out;
+ old_parent_nsi = thread__get_nsinfo(bad_parent);
+ thread__put(bad_parent);
+ bad_parent = NULL;
+ if (!old_parent_nsi)
+ goto out;
+
+ memset(&ev, 0, sizeof(ev));
+ ev.fork.header.type = PERF_RECORD_FORK;
+ ev.fork.header.size = sizeof(ev.fork);
+ ev.fork.ppid = self_pid;
+ ev.fork.ptid = self_pid;
+ ev.fork.pid = self_pid + 2;
+ ev.fork.tid = self_pid + 2;
+ if (machine__process_fork_event(live_m, &ev, &sample) != 0)
+ goto out;
+ if (!RC_CHK_ACCESS(old_parent_nsi)->exited || nsinfo__check_proc(old_parent_nsi)) {
+ pr_debug("replaced parent thread nsinfo was not marked exited\n");
+ goto out;
+ }
+
+ /* perf_event__exit_del_thread() also marks removed thread's nsinfo exited. */
+ th = machine__find_thread(live_m, self_pid, self_pid);
+ if (!th)
+ goto out;
+ del_nsi = thread__get_nsinfo(th);
+ thread__put(th);
+ th = NULL;
+ if (!del_nsi)
+ goto out;
+
+ memset(&ev, 0, sizeof(ev));
+ ev.fork.header.type = PERF_RECORD_EXIT;
+ ev.fork.header.size = sizeof(ev.fork);
+ ev.fork.pid = self_pid;
+ ev.fork.tid = self_pid;
+ ev.fork.ppid = 1;
+ ev.fork.ptid = 1;
+ if (perf_event__exit_del_thread(&tool, &ev, &sample, live_m) != 0)
+ goto out;
+ if (!RC_CHK_ACCESS(del_nsi)->exited || nsinfo__check_proc(del_nsi) ||
+ machine__find_thread(live_m, self_pid, self_pid) != NULL) {
+ pr_debug("perf_event__exit_del_thread() did not mark nsinfo exited\n");
+ goto out;
+ }
+
+ ret = TEST_OK;
+out:
+ thread__put(th);
+ thread__put(off_sub);
+ thread__put(off_th);
+ thread__put(idle_th);
+ thread__put(bad_parent);
+ nsinfo__put(del_nsi);
+ nsinfo__put(old_parent_nsi);
+ nsinfo__put(off_nsi);
+ nsinfo__put(nsi3);
+ nsinfo__put(nsi2);
+ nsinfo__put(nsi1);
+ machines__exit(&off_machines);
+ machines__exit(&live_machines);
+out_env:
+ perf_env__exit(&off_env);
+ perf_env__exit(&live_env);
+ return ret;
+}
+
+static int test__nsinfo_dso_needs_ns(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ static const enum dso_binary_type needs_ns_types[] = {
+ DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
+ DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
+ DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
+ DSO_BINARY_TYPE__DEBUGLINK,
+ DSO_BINARY_TYPE__GNU_DEBUGDATA,
+ DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
+ DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
+ DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
+ DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
+ DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
+ };
+ static const enum dso_binary_type host_only_types[] = {
+ DSO_BINARY_TYPE__KALLSYMS,
+ DSO_BINARY_TYPE__GUEST_KALLSYMS,
+ DSO_BINARY_TYPE__VMLINUX,
+ DSO_BINARY_TYPE__GUEST_VMLINUX,
+ DSO_BINARY_TYPE__JAVA_JIT,
+ DSO_BINARY_TYPE__BUILD_ID_CACHE,
+ DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
+ DSO_BINARY_TYPE__GUEST_KMODULE,
+ DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
+ DSO_BINARY_TYPE__KCORE,
+ DSO_BINARY_TYPE__GUEST_KCORE,
+ DSO_BINARY_TYPE__BPF_PROG_INFO,
+ DSO_BINARY_TYPE__BPF_IMAGE,
+ DSO_BINARY_TYPE__OOL,
+ DSO_BINARY_TYPE__NOT_FOUND,
+ };
+ struct dso *dso = NULL;
+ struct nsinfo *nsi = NULL;
+ struct build_id bid = { .size = 0 };
+ int ret = TEST_FAIL;
+
+ for (size_t i = 0; i < ARRAY_SIZE(needs_ns_types); i++) {
+ if (!dso_binary_type__needs_ns(needs_ns_types[i])) {
+ pr_debug("binary type %d should need ns\n", needs_ns_types[i]);
+ return TEST_FAIL;
+ }
+ }
+
+ for (size_t i = 0; i < ARRAY_SIZE(host_only_types); i++) {
+ if (dso_binary_type__needs_ns(host_only_types[i])) {
+ pr_debug("binary type %d should not need ns\n", host_only_types[i]);
+ return TEST_FAIL;
+ }
+ }
+
+ /* Guarded chroot/build-id lookups must bail out immediately when exited. */
+ dso = dso__new("/bin/sh");
+ nsi = nsinfo__new_thread(getpid(), getpid());
+ if (!dso || !nsi)
+ goto out;
+ nsinfo__set_exited(nsi);
+ dso__set_nsinfo(dso, nsinfo__get(nsi));
+
+ if (dso__filename_with_chroot(dso, "/bin/sh") != NULL) {
+ pr_debug("dso__filename_with_chroot() did not respect exited nsinfo\n");
+ goto out;
+ }
+ if (filename__read_build_id_ns("/nonexistent_perf_chroot_test", &bid, nsi) >= 0) {
+ pr_debug("filename__read_build_id_ns() succeeded on nonexistent path\n");
+ goto out;
+ }
+
+ ret = TEST_OK;
+out:
+ nsinfo__put(nsi);
+ dso__put(dso);
+ return ret;
+}
+
+static int copy_file_mode(const char *src, const char *dst, mode_t mode)
+{
+ char buf[4096];
+ int in_fd = -1, out_fd = -1;
+ ssize_t n;
+ int err = -1;
+
+ in_fd = open(src, O_RDONLY);
+ if (in_fd < 0)
+ return -1;
+ out_fd = open(dst, O_WRONLY | O_CREAT | O_TRUNC, mode);
+ if (out_fd < 0)
+ goto out;
+
+ while ((n = read(in_fd, buf, sizeof(buf))) > 0) {
+ ssize_t off = 0;
+
+ while (off < n) {
+ ssize_t w = write(out_fd, buf + off, n - off);
+
+ if (w <= 0)
+ goto out;
+ off += w;
+ }
+ }
+ if (n == 0)
+ err = 0;
+out:
+ if (out_fd >= 0)
+ close(out_fd);
+ close(in_fd);
+ return err;
+}
+
+static int write_proc_map(const char *path, const char *text)
+{
+ int fd = open(path, O_WRONLY);
+ ssize_t len = strlen(text);
+ ssize_t n;
+
+ if (fd < 0)
+ return -1;
+ n = write(fd, text, len);
+ close(fd);
+ return n == len ? 0 : -1;
+}
+
+static int enter_private_userns(void)
+{
+ uid_t uid = getuid();
+ gid_t gid = getgid();
+ char map_buf[64];
+
+ if (unshare(CLONE_NEWUSER) < 0)
+ return -1;
+
+ /* Required before writing gid_map in unprivileged user namespaces. */
+ if (write_proc_map("/proc/self/setgroups", "deny\n") < 0)
+ return -1;
+ snprintf(map_buf, sizeof(map_buf), "0 %u 1\n", uid);
+ if (write_proc_map("/proc/self/uid_map", map_buf) < 0)
+ return -1;
+ snprintf(map_buf, sizeof(map_buf), "0 %u 1\n", gid);
+ if (write_proc_map("/proc/self/gid_map", map_buf) < 0)
+ return -1;
+
+ return 0;
+}
+
+static int run_mountns_and_chroot_tests(void)
+{
+ char tmpdir[] = "/tmp/perf-test-mntns-XXXXXX";
+ char exe_path[PATH_MAX], target_bin[PATH_MAX];
+ char chroot_dir[PATH_MAX], chroot_bin[PATH_MAX];
+ struct build_id exe_bid = { .size = 0 }, ns_bid = { .size = 0 };
+ int to_child[2] = { -1, -1 }, to_parent[2] = { -1, -1 };
+ struct machine machine;
+ struct nscookie nsc;
+ struct dso *dso = NULL, *chroot_dso = NULL;
+ struct nsinfo *nsi = NULL;
+ bool has_exe_bid = false;
+ pid_t child = -1;
+ char sig = 0;
+ u8 magic[4];
+ int status;
+ int ret = TEST_FAIL;
+
+ if (geteuid() != 0 && enter_private_userns() < 0) {
+ pr_debug("SKIP: user namespace not permitted\n");
+ return TEST_SKIP;
+ }
+
+ if (mkdtemp(tmpdir) == NULL)
+ return TEST_FAIL;
+
+ perf_exe(exe_path, sizeof(exe_path));
+ has_exe_bid = filename__read_build_id(exe_path, &exe_bid) > 0;
+ snprintf(target_bin, sizeof(target_bin), "%s/testbin", tmpdir);
+ snprintf(chroot_dir, sizeof(chroot_dir), "%s/chroot", tmpdir);
+ snprintf(chroot_bin, sizeof(chroot_bin), "%s/chroot/chroot_elf", tmpdir);
+
+ if (mkdir(chroot_dir, 0700) < 0 ||
+ copy_file_mode(exe_path, chroot_bin, 0700) < 0)
+ goto out_rmdir;
+
+ if (pipe(to_child) < 0 || pipe(to_parent) < 0)
+ goto out_rmdir;
+
+ child = fork();
+ if (child < 0)
+ goto out_rmdir;
+
+ if (child == 0) {
+ char cmd = 0;
+
+ close(to_child[1]);
+ close(to_parent[0]);
+
+ if (unshare(CLONE_NEWNS) < 0 ||
+ mount(NULL, "/", NULL, MS_REC | MS_SLAVE, NULL) < 0 ||
+ mount("tmpfs", tmpdir, "tmpfs", 0, NULL) < 0 ||
+ copy_file_mode(exe_path, target_bin, 0700) < 0) {
+ sig = 'S';
+ if (write(to_parent[1], &sig, 1) != 1)
+ _exit(1);
+ _exit(CHILD_EXIT_SKIP);
+ }
+
+ sig = '1';
+ if (write(to_parent[1], &sig, 1) != 1)
+ _exit(1);
+
+ /* Phase 2: switch to another mount namespace to test inode check. */
+ if (read(to_child[0], &cmd, 1) != 1)
+ _exit(1);
+ if (unshare(CLONE_NEWNS) < 0) {
+ sig = 'S';
+ if (write(to_parent[1], &sig, 1) != 1)
+ _exit(1);
+ _exit(CHILD_EXIT_SKIP);
+ }
+ sig = '2';
+ if (write(to_parent[1], &sig, 1) != 1)
+ _exit(1);
+
+ /* Wait until parent releases us. */
+ if (read(to_child[0], &cmd, 1) < 0)
+ _exit(1);
+ _exit(0);
+ }
+
+ close(to_child[0]);
+ to_child[0] = -1;
+ close(to_parent[1]);
+ to_parent[1] = -1;
+
+ if (read(to_parent[0], &sig, 1) != 1)
+ goto out_wait;
+ if (sig == 'S') {
+ pr_debug("SKIP: mount namespace operations not permitted\n");
+ ret = TEST_SKIP;
+ goto out_wait;
+ }
+ if (sig != '1')
+ goto out_wait;
+
+ /* File in child's tmpfs must not be visible on the host filesystem. */
+ if (access(target_bin, F_OK) == 0) {
+ pr_debug("child mount leaked to host namespace\n");
+ goto out_wait;
+ }
+
+ nsi = nsinfo__new_thread(child, child);
+ if (!nsi)
+ goto out_wait;
+ if (!nsinfo__need_setns(nsi) || RC_CHK_ACCESS(nsi)->mntns_ino == 0) {
+ pr_debug("nsinfo did not detect differing mount namespace or inode\n");
+ goto out_wait;
+ }
+
+ if (has_exe_bid) {
+ if (filename__read_build_id_ns(target_bin, &ns_bid, nsi) <= 0 ||
+ memcmp(&exe_bid, &ns_bid, sizeof(exe_bid)) != 0) {
+ pr_debug("failed to read build-id inside child mountns\n");
+ goto out_wait;
+ }
+ }
+
+ memset(&machine, 0, sizeof(machine));
+ if (machine__init(&machine, "", HOST_KERNEL_ID) != 0)
+ goto out_wait;
+
+ dso = dso__new(target_bin);
+ if (!dso) {
+ machine__exit(&machine);
+ goto out_wait;
+ }
+ dso__set_nsinfo(dso, nsinfo__get(nsi));
+#ifndef HAVE_LIBELF_SUPPORT
+ /* Without libelf, filename__has_section() cannot auto-detect binary_type. */
+ dso__set_binary_type(dso, DSO_BINARY_TYPE__SYSTEM_PATH_DSO);
+ dso__set_bintype_searched(dso);
+#endif
+ if (dso__data_read_offset(dso, &machine, 0, magic, sizeof(magic)) != sizeof(magic) ||
+ memcmp(magic, "\x7f" "ELF", 4) != 0) {
+ pr_debug("failed to read ELF header from binary inside mountns\n");
+ dso__data_close(dso);
+ dso__put(dso);
+ dso = NULL;
+ machine__exit(&machine);
+ goto out_wait;
+ }
+ dso__data_close(dso);
+ dso__put(dso);
+ dso = NULL;
+
+ {
+ struct thread *cth = machine__findnew_thread(&machine, child, child);
+ struct map *map = NULL;
+ struct dso *ldso;
+
+ if (!cth) {
+ machine__exit(&machine);
+ goto out_wait;
+ }
+ map = map__new(&machine, 0x100000, 0xffffffff, 0, &dso_id_empty,
+ PROT_EXEC, /*flags=*/0, target_bin, cth);
+ thread__put(cth);
+ if (!map) {
+ machine__exit(&machine);
+ goto out_wait;
+ }
+ ldso = map__dso(map);
+ if (dso__load(ldso, map) < 0 ||
+ (has_exe_bid && (!dso__has_build_id(ldso) ||
+ !dso__build_id_equal(ldso, &exe_bid)))) {
+ pr_debug("dso__load() failed inside child mountns\n");
+ map__put(map);
+ machine__exit(&machine);
+ goto out_wait;
+ }
+ map__put(map);
+ }
+ machine__exit(&machine);
+
+ /* Have the child unshare again so mntns_ino changes while staying alive. */
+ sig = 'n';
+ if (write(to_child[1], &sig, 1) != 1 || read(to_parent[0], &sig, 1) != 1)
+ goto out_wait;
+ if (sig == 'S') {
+ ret = TEST_SKIP;
+ goto out_wait;
+ }
+
+ nsinfo__mountns_enter(nsi, &nsc);
+ nsinfo__mountns_exit(&nsc);
+ if (nsinfo__need_setns(nsi) || RC_CHK_ACCESS(nsi)->exited || !nsinfo__check_proc(nsi)) {
+ pr_debug("mntns_ino change should clear need_setns without setting exited\n");
+ goto out_wait;
+ }
+
+ sig = 'q';
+ if (write(to_child[1], &sig, 1) != 1)
+ goto out_wait;
+ waitpid(child, &status, 0);
+ child = -1;
+
+ /* Once reaped, nsinfo__check_proc() must detect exit. */
+ if (nsinfo__check_proc(nsi) || !RC_CHK_ACCESS(nsi)->exited) {
+ pr_debug("exited child still passed nsinfo__check_proc()\n");
+ goto out_wait;
+ }
+
+ /* Also verify chroot resolution when chroot(2) is permitted. */
+ if (geteuid() == 0 && access("/chroot_elf", F_OK) < 0) {
+ int cpipe[2] = { -1, -1 };
+
+ if (pipe(cpipe) == 0) {
+ pid_t cpid = fork();
+
+ if (cpid == 0) {
+ close(cpipe[0]);
+ if (chroot(chroot_dir) < 0 || chdir("/") < 0) {
+ sig = 'S';
+ if (write(cpipe[1], &sig, 1) != 1)
+ _exit(1);
+ _exit(CHILD_EXIT_SKIP);
+ }
+ sig = '1';
+ if (write(cpipe[1], &sig, 1) != 1)
+ _exit(1);
+ pause();
+ _exit(0);
+ }
+ close(cpipe[1]);
+ if (cpid > 0 && read(cpipe[0], &sig, 1) == 1 && sig == '1') {
+ struct nsinfo *cnsi = nsinfo__new_thread(cpid, cpid);
+ struct stat st1, st2;
+ char *resolved = NULL;
+
+ chroot_dso = dso__new("/chroot_elf");
+ if (chroot_dso && cnsi) {
+ dso__set_nsinfo(chroot_dso, nsinfo__get(cnsi));
+ resolved = dso__filename_with_chroot(chroot_dso,
+ "/chroot_elf");
+ if (!resolved ||
+ stat(resolved, &st1) < 0 ||
+ stat(chroot_bin, &st2) < 0 ||
+ st1.st_dev != st2.st_dev ||
+ st1.st_ino != st2.st_ino) {
+ pr_debug("chroot resolution mismatch: %s vs %s\n",
+ resolved ? resolved : "(null)",
+ chroot_bin);
+ free(resolved);
+ kill(cpid, SIGKILL);
+ waitpid(cpid, NULL, 0);
+ nsinfo__put(cnsi);
+ close(cpipe[0]);
+ goto out_wait;
+ }
+ free(resolved);
+ if (has_exe_bid) {
+ memset(&ns_bid, 0, sizeof(ns_bid));
+ if (filename__read_build_id_ns("/chroot_elf",
+ &ns_bid, cnsi) <= 0 ||
+ memcmp(&exe_bid, &ns_bid,
+ sizeof(exe_bid)) != 0) {
+ pr_debug("chroot build-id mismatch\n");
+ kill(cpid, SIGKILL);
+ waitpid(cpid, NULL, 0);
+ nsinfo__put(cnsi);
+ close(cpipe[0]);
+ goto out_wait;
+ }
+ }
+ }
+ kill(cpid, SIGKILL);
+ waitpid(cpid, NULL, 0);
+ if (chroot_dso &&
+ dso__filename_with_chroot(chroot_dso, "/chroot_elf") != NULL) {
+ pr_debug("exited chroot task resolved /proc/<pid>/root\n");
+ nsinfo__put(cnsi);
+ close(cpipe[0]);
+ goto out_wait;
+ }
+ nsinfo__put(cnsi);
+ } else if (cpid > 0) {
+ waitpid(cpid, NULL, 0);
+ }
+ close(cpipe[0]);
+ }
+ }
+
+ ret = TEST_OK;
+out_wait:
+ if (child > 0) {
+ kill(child, SIGKILL);
+ waitpid(child, &status, 0);
+ }
+ nsinfo__put(nsi);
+ dso__put(chroot_dso);
+ dso__put(dso);
+out_rmdir:
+ if (to_child[0] >= 0)
+ close(to_child[0]);
+ if (to_child[1] >= 0)
+ close(to_child[1]);
+ if (to_parent[0] >= 0)
+ close(to_parent[0]);
+ if (to_parent[1] >= 0)
+ close(to_parent[1]);
+ unlink(chroot_bin);
+ rmdir(chroot_dir);
+ unlink(target_bin);
+ rmdir(tmpdir);
+ return ret;
+}
+
+static int test__nsinfo_mountns(struct test_suite *test __maybe_unused,
+ int subtest __maybe_unused)
+{
+ int status;
+ pid_t pid;
+
+ if (geteuid() == 0)
+ return run_mountns_and_chroot_tests();
+
+ /* Run in a dedicated child so unshare(CLONE_NEWUSER) never mutates the caller. */
+ pid = fork();
+ if (pid < 0)
+ return TEST_FAIL;
+ if (pid == 0) {
+ int rc = run_mountns_and_chroot_tests();
+
+ if (rc == TEST_SKIP)
+ _exit(CHILD_EXIT_SKIP);
+ _exit(rc == TEST_OK ? 0 : 1);
+ }
+
+ if (waitpid(pid, &status, 0) != pid)
+ return TEST_FAIL;
+ if (!WIFEXITED(status))
+ return TEST_FAIL;
+ if (WEXITSTATUS(status) == CHILD_EXIT_SKIP)
+ return TEST_SKIP;
+ return WEXITSTATUS(status) == 0 ? TEST_OK : TEST_FAIL;
+}
+
+static struct test_case tests__namespaces[] = {
+ TEST_CASE("boot_id", nsinfo_boot_id),
+ TEST_CASE("init_and_copy", nsinfo_init_and_copy),
+ TEST_CASE("proc_and_recycled_tid", nsinfo_proc_and_recycled_tid),
+ TEST_CASE("thread_liveness", nsinfo_thread_liveness),
+ TEST_CASE("dso_needs_ns", nsinfo_dso_needs_ns),
+ TEST_CASE("mountns_and_chroot", nsinfo_mountns),
+ { .name = NULL, }
+};
+
+struct test_suite suite__namespaces = {
+ .desc = "Namespace and nsinfo tests",
+ .test_cases = tests__namespaces,
+};
--git a/tools/perf/tests/shell/header.sh b/tools/perf/tests/shell/header.sh
index e1628ac0a61401a7..ae3cd2b50d1c3437 100755
--- a/tools/perf/tests/shell/header.sh
+++ b/tools/perf/tests/shell/header.sh
@@ -37,6 +37,7 @@ check_header_output() {
"sibling (cores|dies|threads)"
"sibling threads"
"total memory"
+ "boot id"
)
for i in "${fields[@]}"
do
@@ -46,6 +47,14 @@ check_header_output() {
err=1
fi
done
+
+ if [ -r /proc/sys/kernel/random/boot_id ]; then
+ host_boot_id=$(tr -d '\n' < /proc/sys/kernel/random/boot_id)
+ if ! grep -q "boot id : ${host_boot_id}" "${script_output}"; then
+ echo "Failed to find matching host boot id ${host_boot_id} in output"
+ err=1
+ fi
+ fi
}
test_file() {
diff --git a/tools/perf/tests/tests.h b/tools/perf/tests/tests.h
index af822885890dda73..b75fc1ef1c13bdd2 100644
--- a/tools/perf/tests/tests.h
+++ b/tools/perf/tests/tests.h
@@ -203,6 +203,7 @@ DECLARE_SUITE(util);
DECLARE_SUITE(uncore_event_sorting);
DECLARE_SUITE(subcmd_help);
DECLARE_SUITE(kallsyms_split);
+DECLARE_SUITE(namespaces);
/*
* PowerPC and S390 do not support creation of instruction breakpoints using the
--
2.56.0.611.g1030f0b940-goog
^ permalink raw reply related [flat|nested] 20+ messages in thread
* Re: [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot()
2026-11-16 5:57 ` [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot() Namhyung Kim
@ 2026-11-16 6:02 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:02 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> Add perf_env__is_same_boot() to compare the recorded boot_id in struct
> perf_env against the running host's /proc/sys/kernel/random/boot_id.
> For live sessions where no offline header was populated, it falls back
> to reading the host boot_id and returns true.
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=2
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature
2026-11-16 5:57 ` [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature Namhyung Kim
@ 2026-11-16 6:09 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:09 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> Save /proc/sys/kernel/random/boot_id in the perf.data header and
> struct perf_env so post-processing tools can identify if it's the
> same kernel boot on which the data was recorded.
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=1
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used
2026-11-16 5:57 ` [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used Namhyung Kim
@ 2026-11-16 6:11 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:11 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> Do not read /proc/<pid>/task/<tid>/ eagerly in nsinfo__new_thread().
> Instead, initialize struct nsinfo on demand via
> nsinfo__ensure_init_locked() when namespace information is queried or
> entered, protected by nsinfo->lock.
>
> Replace thread__nsinfo() with thread__get_nsinfo() and add
> thread__reset_nsinfo() to replace thread->nsinfo safely under
> thread->namespaces_lock on exec and PERF_RECORD_NAMESPACES events so
> namespace changes are picked up without racing existing readers.
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=5
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode
2026-11-16 5:57 ` [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode Namhyung Kim
@ 2026-11-16 6:14 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:14 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Low] perf namespaces: nsinfo__init_locked() returns 0 when stat()
fails
--
commit e6cbb7bd6fc00579ed776b5b31e5e5cfe0cc41a2
Author: Namhyung Kim <namhyung@kernel.org>
perf tools: Verify thread status and mount namespace inode
This commit verifies that /proc/<pid>/task/<tid>/status exists with a
matching Tgid and checks the mount namespace inode via fstat() before
entering a thread's mount namespace.
> diff --git a/tools/perf/util/namespaces.c b/tools/perf/util/namespaces.c
> index afd8ad7f843ec..860b90e8384ed 100644
> --- a/tools/perf/util/namespaces.c
> +++ b/tools/perf/util/namespaces.c
[ ... ]
> @@ -101,16 +116,41 @@ static int nsinfo__init_locked(struct nsinfo *nsi)
> char *newns = NULL;
> struct stat old_stat;
> struct stat new_stat;
> + pid_t tgid = 0, nstgid = 0;
> + bool in_pidns = false;
> int rv = -1;
>
> + if (RC_CHK_ACCESS(nsi)->exited)
> + return -1;
> +
> RC_CHK_ACCESS(nsi)->inited = true;
>
> + /*
> + * Verify that the thread (pid, tid) still exists in /proc and its Tgid
> + * matches pid (guarding against PID/TID recycling), while resolving the
> + * innermost tgid if the process is in a different PID namespace.
> + */
> + if (snprintf(spath, PATH_MAX, "/proc/%d/task/%d/status",
> + nsinfo__pid(nsi), nsinfo__tid(nsi)) >= PATH_MAX)
> + goto out;
> +
> + rv = nsinfo__get_nspid(&tgid, &nstgid, &in_pidns, spath);
> + if (rv < 0 || tgid != nsinfo__pid(nsi)) {
> + nsinfo__set_exited_locked(nsi);
> + rv = -1;
> + goto out;
> + }
> +
> + RC_CHK_ACCESS(nsi)->tgid = tgid;
> + RC_CHK_ACCESS(nsi)->nstgid = nstgid;
> + RC_CHK_ACCESS(nsi)->in_pidns = in_pidns;
> +
> if (snprintf(oldns, PATH_MAX, "/proc/self/ns/mnt") >= PATH_MAX)
> - return rv;
> + return -1;
>
> if (asprintf(&newns, "/proc/%d/task/%d/ns/mnt",
> nsinfo__pid(nsi), nsinfo__tid(nsi)) == -1)
> - return rv;
> + return -1;
>
> if (stat(oldns, &old_stat) < 0)
> goto out;
> @@ -124,19 +164,10 @@ static int nsinfo__init_locked(struct nsinfo *nsi)
> if (old_stat.st_ino != new_stat.st_ino) {
> RC_CHK_ACCESS(nsi)->need_setns = true;
> RC_CHK_ACCESS(nsi)->mntns_path = newns;
> + RC_CHK_ACCESS(nsi)->mntns_ino = new_stat.st_ino;
> newns = NULL;
> }
>
> - /* If we're dealing with a process that is in a different PID namespace,
> - * attempt to work out the innermost tgid for the process.
> - */
> - if (snprintf(spath, PATH_MAX, "/proc/%d/task/%d/status",
> - nsinfo__pid(nsi), nsinfo__tid(nsi)) >= PATH_MAX)
> - goto out;
> -
> - rv = nsinfo__get_nspid(&RC_CHK_ACCESS(nsi)->tgid, &RC_CHK_ACCESS(nsi)->nstgid,
> - &RC_CHK_ACCESS(nsi)->in_pidns, spath);
> -
> out:
> free(newns);
> return rv;
[Severity: Low]
Does nsinfo__init_locked() return 0 instead of -1 when stat() fails?
Moving nsinfo__get_nspid() ahead of the stat() checks sets rv to 0 on
success. If stat(newns, &new_stat) then fails (for example with EACCES
because /proc/<pid>/task/<tid>/ns/mnt requires PTRACE_MODE_READ_FSCREDS
while status is world-readable), jumping to out returns 0 to callers of
nsinfo__init().
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=6
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns()
2026-11-16 5:57 ` [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns() Namhyung Kim
@ 2026-11-16 6:16 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:16 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Medium] perf build-id: filename__read_build_id_ns() checks host /proc
inside mountns
--
commit 32054987b7b4b030ff39e7d9b8a53cb7044261a7
Author: Namhyung Kim <namhyung@kernel.org>
perf tools: Export filename__read_build_id_ns()
This commit exports filename__read_build_id_ns() with a chroot fallback
and uses it in dsos__read_build_ids_cb() and perf-inject's
dso__read_build_id().
> diff --git a/tools/perf/util/build-id.c b/tools/perf/util/build-id.c
> index 02233ef0eafff..ff083a594b405 100644
> --- a/tools/perf/util/build-id.c
> +++ b/tools/perf/util/build-id.c
> @@ -847,15 +847,23 @@ int build_id_cache__remove_s(const char *sbuild_id)
> return err;
> }
>
> -static int filename__read_build_id_ns(const char *filename,
> - struct build_id *bid,
> - struct nsinfo *nsi)
> +int filename__read_build_id_ns(const char *filename,
> + struct build_id *bid,
> + struct nsinfo *nsi)
> {
> struct nscookie nsc;
> int ret;
>
> nsinfo__mountns_enter(nsi, &nsc);
> ret = filename__read_build_id(filename, bid);
> + if (ret < 0 && errno == ENOENT && nsinfo__check_proc(nsi)) {
> + char *new_name = filename_with_chroot(nsinfo__pid(nsi), filename);
> +
> + if (new_name) {
> + ret = filename__read_build_id(new_name, bid);
> + free(new_name);
> + }
> + }
> nsinfo__mountns_exit(&nsc);
[Severity: Medium]
Does calling nsinfo__check_proc() and filename_with_chroot() before
nsinfo__mountns_exit() look up the host PID in the target mount
namespace's /proc?
When profiling a container with its own PID namespace and /proc mount,
any ENOENT from filename__read_build_id() (for example on [heap] or a
deleted binary) causes nsinfo__check_proc() to inspect
/proc/<host_pid>/task/<host_tid>/status inside the container's /proc:
tools/perf/util/namespaces.c:nsinfo__check_proc() {
...
scnprintf(spath, sizeof(spath), "/proc/%d/task/%d/status",
nsinfo__pid(nsi), nsinfo__tid(nsi));
if (nsinfo__get_nspid(&tgid, &nstgid, &in_pidns, spath) < 0 ||
tgid != nsinfo__pid(nsi)) {
nsinfo__set_exited(nsi);
return false;
}
...
}
Because the host PID is absent or belongs to a different process there,
nsinfo__check_proc() calls nsinfo__set_exited(nsi), clearing need_setns
and freeing mntns_path on the shared struct nsinfo, which disables mount
namespace entry for subsequent DSOs of that thread. Moving
nsinfo__mountns_exit(&nsc) before the fallback (or resolving new_name in
the host mount namespace) would avoid this.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=8
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid
2026-11-16 5:57 ` [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid Namhyung Kim
@ 2026-11-16 6:16 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:16 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> Record the thread ID (tid) in struct nsinfo via nsinfo__new_thread() and
> access /proc/<pid>/task/<tid>/ status and ns/mnt paths instead of only
> /proc/<pid>/. Also recreate thread->nsinfo when the pid of the thread
> is updated.
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=3
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load()
2026-11-16 5:57 ` [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load() Namhyung Kim
@ 2026-11-16 6:16 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:16 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> Enter the target mount namespace only for the specific binary types
> that need it in dso__load(), and use filename__read_build_id_ns() for
> BUILDID_DEBUGINFO lookup and in symbol-minimal's dso__load_sym() so it
> does not read the wrong build-id from the host filesystem.
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=10
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v2 04/11] perf tools: Add lock to struct nsinfo
2026-11-16 5:57 ` [PATCH v2 04/11] perf tools: Add lock to struct nsinfo Namhyung Kim
@ 2026-11-16 6:17 ` sashiko-bot
0 siblings, 0 replies; 20+ messages in thread
From: sashiko-bot @ 2026-11-16 6:17 UTC (permalink / raw)
To: Namhyung Kim; +Cc: linux-perf-users
> A struct nsinfo instance can be shared across threads via
> thread->nsinfo and dso->nsinfo. Add a mutex lock inside struct nsinfo
> to synchronize access to its fields, initialization, copying, and mount
> namespace path duplication in nsinfo__mountns_enter().
>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116055731.707164-1-namhyung@kernel.org?part=4
^ permalink raw reply [flat|nested] 20+ messages in thread
end of thread, other threads:[~2026-11-16 6:17 UTC | newest]
Thread overview: 20+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 5:57 [PATCH v2 00/11] perf tools: Enhance namespace and chroot handling Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 01/11] perf tools: Add HEADER_BOOT_ID feature Namhyung Kim
2026-11-16 6:09 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 02/11] perf tools: Add perf_env__is_same_boot() Namhyung Kim
2026-11-16 6:02 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 03/11] perf tools: Add nsinfo__new_thread() to track tid Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 04/11] perf tools: Add lock to struct nsinfo Namhyung Kim
2026-11-16 6:17 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 05/11] perf tools: Defer /proc lookup until nsinfo is used Namhyung Kim
2026-11-16 6:11 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 06/11] perf tools: Verify thread status and mount namespace inode Namhyung Kim
2026-11-16 6:14 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 07/11] perf tools: Check thread liveness when accessing /proc Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 08/11] perf tools: Export filename__read_build_id_ns() Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 09/11] perf tools: Only enter mount namespace if necessary Namhyung Kim
2026-11-16 5:57 ` [PATCH v2 10/11] perf tools: Scope mount namespace entry in dso__load() Namhyung Kim
2026-11-16 6:16 ` sashiko-bot
2026-11-16 5:57 ` [PATCH v2 11/11] perf tests: Add namespace, nsinfo, and boot ID tests Namhyung Kim
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox