Skip to contentSkip to conversation

bpf: Add open coded version of kmem_cache iterator

PATCH Superseded Namhyung Kim 3 participants 11 messages Oct 17, 2024 – Oct 24, 2024

Subsystems (5)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 bpf: Add open coded version of kmem_cache iterator

Add a new open coded iterator for kmem_cache which can be called from a BPF program like below. It doesn't take any argument and traverses all kmem_cache entries.

struct kmem_cache *pos;

  bpf_for_each(kmem_cache, pos) {
      ...
  }

As it needs to grab slab_mutex, it should be called from sleepable BPF programs only.

---
 kernel/bpf/helpers.c         |  3 ++
 kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
 2 files changed, 90 insertions(+)
kernel/bpf/helpers.c +3 -0
index 073e6f04f4d765ff..d1dfa4f335577914 100644
@@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
 BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_get_kmem_cache)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/kmem_cache_iter.c +87 -0

93 lines not on this page · raw message

Signature · 1 line
2.47.0.rc1.288.g06298d1525-goog
#1 lore raw

PATCH 2/2 selftests/bpf: Add a test for open coded kmem_cache iter

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

  $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
  ...
  #130/1   kmem_cache_iter/check_task_struct:OK
  #130/2   kmem_cache_iter/check_slabinfo:OK
  #130/3   kmem_cache_iter/open_coded_iter:OK
  #130     kmem_cache_iter:OK
  Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
 .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
 .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
 .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
 3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +6 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
 		unsigned int flags__k, void *aux__ign) __ksym;
 #define bpf_wq_set_callback(timer, cb, flags) \
 	bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+
 #endif
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +16 -12
index 848d8fc9171fae45..a1fd3bc57c0b21bb 100644
@@ -68,12 +68,18 @@ static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel)
 	fclose(fp);
 }
 
+static void subtest_kmem_cache_iter_open_coded(struct kmem_cache_iter *skel)
+{
+	/* To trigger the open coded iterator attached to the syscall */
+	syncfs(0);
+
+	/* It should be same as we've seen from the explicit iterator */
+	ASSERT_EQ(skel->bss->open_coded_seen, skel->bss->kmem_cache_seen, "open_code_seen_eq");
+}
+
 void test_kmem_cache_iter(void)
 {
-	DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
 	struct kmem_cache_iter *skel = NULL;
-	union bpf_iter_link_info linfo = {};
-	struct bpf_link *link;
 	char buf[256];
 	int iter_fd;
 
@@ -81,16 +87,12 @@ void test_kmem_cache_iter(void)
 	if (!ASSERT_OK_PTR(skel, "kmem_cache_iter__open_and_load"))
 		return;
 
-	opts.link_info = &linfo;
-	opts.link_info_len = sizeof(linfo);
-
-	link = bpf_program__attach_iter(skel->progs.slab_info_collector, &opts);
-	if (!ASSERT_OK_PTR(link, "attach_iter"))
+	if (!ASSERT_OK(kmem_cache_iter__attach(skel), "skel_attach"))
 		goto destroy;
 
-	iter_fd = bpf_iter_create(bpf_link__fd(link));
+	iter_fd = bpf_iter_create(bpf_link__fd(skel->links.slab_info_collector));
 	if (!ASSERT_GE(iter_fd, 0, "iter_create"))
-		goto free_link;
+		goto detach;
 
 	memset(buf, 0, sizeof(buf));
 	while (read(iter_fd, buf, sizeof(buf) > 0)) {
@@ -105,11 +107,13 @@ void test_kmem_cache_iter(void)
 		subtest_kmem_cache_iter_check_task_struct(skel);
 	if (test__start_subtest("check_slabinfo"))
 		subtest_kmem_cache_iter_check_slabinfo(skel);
+	if (test__start_subtest("open_coded_iter"))
+		subtest_kmem_cache_iter_open_coded(skel);
 
 	close(iter_fd);
 
-free_link:
-	bpf_link__destroy(link);
+detach:
+	kmem_cache_iter__detach(skel);
 destroy:
 	kmem_cache_iter__destroy(skel);
 }
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +24 -0
index 72c9dafecd98406b..4c44aa279a5328fe 100644
@@ -2,6 +2,8 @@
 /* Copyright (c) 2024 Google */
 
 #include "bpf_iter.h"
+#include "bpf_experimental.h"
+#include "bpf_misc.h"
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 
@@ -33,6 +35,7 @@ extern struct kmem_cache *bpf_get_kmem_cache(u64 addr) __ksym;
 /* Result, will be checked by userspace */
 int task_struct_found;
 int kmem_cache_seen;
+int open_coded_seen;
 
 SEC("iter/kmem_cache")
 int slab_info_collector(struct bpf_iter__kmem_cache *ctx)
@@ -85,3 +88,24 @@ int BPF_PROG(check_task_struct)
 		task_struct_found = -2;
 	return 0;
 }
+
+SEC("fentry.s/" SYS_PREFIX "sys_syncfs")
+int open_coded_iter(const void *ctx)
+{
+	struct kmem_cache *s;
+
+	bpf_for_each(kmem_cache, s) {
+		struct kmem_cache_result *r;
+		int idx = open_coded_seen;
+
+		r = bpf_map_lookup_elem(&slab_result, &idx);
+		if (r == NULL)
+			break;
+
+		open_coded_seen++;
+
+		if (r->obj_size != s->size)
+			break;
+	}
+	return 0;
+}
Signature · 1 line
2.47.0.rc1.288.g06298d1525-goog
#2 lore raw

Re: [PATCH bpf-next 2/2] selftests/bpf: Add a test for open coded kmem_cache iter

Quoted 64 lines · Namhyung Kim

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

   $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
   ...
   #130/1   kmem_cache_iter/check_task_struct:OK
   #130/2   kmem_cache_iter/check_slabinfo:OK
   #130/3   kmem_cache_iter/open_coded_iter:OK
   #130     kmem_cache_iter:OK
   Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
  .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
  .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
  .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
  3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +6 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
  		unsigned int flags__k, void *aux__ign) __ksym;
  #define bpf_wq_set_callback(timer, cb, flags) \
  	bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+
  #endif
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +9 -3
index 848d8fc9171fae45..a1fd3bc57c0b21bb 100644
@@ -68,12 +68,18 @@ static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel)
  	fclose(fp);
  }
 
+static void subtest_kmem_cache_iter_open_coded(struct kmem_cache_iter *skel)
+{
+	/* To trigger the open coded iterator attached to the syscall */
+	syncfs(0);
+
+	/* It should be same as we've seen from the explicit iterator */
+	ASSERT_EQ(skel->bss->open_coded_seen, skel->bss->kmem_cache_seen, "open_code_seen_eq");
+}
+
  void test_kmem_cache_iter(void)
  {
-	DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
  	struct kmem_cache_iter *skel = NULL;
-	union bpf_iter_link_info linfo = {};
-	struct bpf_link *link;
  	char buf[256];
  	int iter_fd;
 

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c void test_kmem_cache_iter(void) · line 90

Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +1 -5
@@ -81,16 +87,12 @@ void test_kmem_cache_iter(void)
  	if (!ASSERT_OK_PTR(skel, "kmem_cache_iter__open_and_load"))
  		return;
 
-	opts.link_info = &linfo;
-	opts.link_info_len = sizeof(linfo);
-
-	link = bpf_program__attach_iter(skel->progs.slab_info_collector, &opts);
-	if (!ASSERT_OK_PTR(link, "attach_iter"))
+	if (!ASSERT_OK(kmem_cache_iter__attach(skel), "skel_attach"))

with this change.

Quoted 13 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +2 -2
  		goto destroy;
 
-	iter_fd = bpf_iter_create(bpf_link__fd(link));
+	iter_fd = bpf_iter_create(bpf_link__fd(skel->links.slab_info_collector));
  	if (!ASSERT_GE(iter_fd, 0, "iter_create"))
-		goto free_link;
+		goto detach;
 
  	memset(buf, 0, sizeof(buf));
  	while (read(iter_fd, buf, sizeof(buf) > 0)) {
@@ -105,11 +107,13 @@ void test_kmem_cache_iter(void)
  		subtest_kmem_cache_iter_check_task_struct(skel);
  	if (test__start_subtest("check_slabinfo"))

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c void test_kmem_cache_iter(void) · line 116

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +4 -2
  		subtest_kmem_cache_iter_check_slabinfo(skel);
+	if (test__start_subtest("open_coded_iter"))
+		subtest_kmem_cache_iter_open_coded(skel);
 
  	close(iter_fd);
 
-free_link:
-	bpf_link__destroy(link);
+detach:
+	kmem_cache_iter__detach(skel);

nit. I think the kmem_cache_iter__destroy() below will also detach, so no need to explicit kmem_cache_iter__detach().

Quoted 33 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +0 -0
  destroy:
  	kmem_cache_iter__destroy(skel);
  }
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +8 -0
index 72c9dafecd98406b..4c44aa279a5328fe 100644
@@ -2,6 +2,8 @@
  /* Copyright (c) 2024 Google */
 
  #include "bpf_iter.h"
+#include "bpf_experimental.h"
+#include "bpf_misc.h"
  #include <bpf/bpf_helpers.h>
  #include <bpf/bpf_tracing.h>
 
@@ -33,6 +35,7 @@ extern struct kmem_cache *bpf_get_kmem_cache(u64 addr) __ksym;
  /* Result, will be checked by userspace */
  int task_struct_found;
  int kmem_cache_seen;
+int open_coded_seen;
 
  SEC("iter/kmem_cache")
  int slab_info_collector(struct bpf_iter__kmem_cache *ctx)
@@ -85,3 +88,24 @@ int BPF_PROG(check_task_struct)
  		task_struct_found = -2;
  	return 0;
  }
+
+SEC("fentry.s/" SYS_PREFIX "sys_syncfs")
+int open_coded_iter(const void *ctx)
+{
+	struct kmem_cache *s;

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 105

Quoted 10 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +10 -0
+
+	bpf_for_each(kmem_cache, s) {
+		struct kmem_cache_result *r;
+		int idx = open_coded_seen;
+
+		r = bpf_map_lookup_elem(&slab_result, &idx);
+		if (r == NULL)
+			break;
+
+		open_coded_seen++;

I am not sure if this will work well if the testing system somehow has another process calling syncfs. It is probably a good idea to guard this by checking the tid of the test_progs at the beginning of this bpf prog.

Quoted 6 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +6 -0
+
+		if (r->obj_size != s->size)
+			break;
+	}
+	return 0;
+}
#3 lore raw

Re: [PATCH bpf-next 2/2] selftests/bpf: Add a test for open coded kmem_cache iter

Quoted 88 lines · Martin KaFai Lau
Quoted 62 lines · Namhyung Kim

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

   $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
   ...
   #130/1   kmem_cache_iter/check_task_struct:OK
   #130/2   kmem_cache_iter/check_slabinfo:OK
   #130/3   kmem_cache_iter/open_coded_iter:OK
   #130     kmem_cache_iter:OK
   Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
  .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
  .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
  .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
  3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +6 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
  		unsigned int flags__k, void *aux__ign) __ksym;
  #define bpf_wq_set_callback(timer, cb, flags) \
  	bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+
  #endif
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +9 -3
index 848d8fc9171fae45..a1fd3bc57c0b21bb 100644
@@ -68,12 +68,18 @@ static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel)
  	fclose(fp);
  }
+static void subtest_kmem_cache_iter_open_coded(struct kmem_cache_iter *skel)
+{
+	/* To trigger the open coded iterator attached to the syscall */
+	syncfs(0);
+
+	/* It should be same as we've seen from the explicit iterator */
+	ASSERT_EQ(skel->bss->open_coded_seen, skel->bss->kmem_cache_seen, "open_code_seen_eq");
+}
+
  void test_kmem_cache_iter(void)
  {
-	DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
  	struct kmem_cache_iter *skel = NULL;
-	union bpf_iter_link_info linfo = {};
-	struct bpf_link *link;
  	char buf[256];
  	int iter_fd;

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c void test_kmem_cache_iter(void) · line 89

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +1 -5
@@ -81,16 +87,12 @@ void test_kmem_cache_iter(void)
  	if (!ASSERT_OK_PTR(skel, "kmem_cache_iter__open_and_load"))
  		return;
-	opts.link_info = &linfo;
-	opts.link_info_len = sizeof(linfo);
-
-	link = bpf_program__attach_iter(skel->progs.slab_info_collector, &opts);
-	if (!ASSERT_OK_PTR(link, "attach_iter"))
+	if (!ASSERT_OK(kmem_cache_iter__attach(skel), "skel_attach"))

with this change.

Quoted 14 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +4 -2
  		goto destroy;
-	iter_fd = bpf_iter_create(bpf_link__fd(link));
+	iter_fd = bpf_iter_create(bpf_link__fd(skel->links.slab_info_collector));
  	if (!ASSERT_GE(iter_fd, 0, "iter_create"))
-		goto free_link;
+		goto detach;
  	memset(buf, 0, sizeof(buf));
  	while (read(iter_fd, buf, sizeof(buf) > 0)) {
@@ -105,11 +107,13 @@ void test_kmem_cache_iter(void)
  		subtest_kmem_cache_iter_check_task_struct(skel);
  	if (test__start_subtest("check_slabinfo"))
  		subtest_kmem_cache_iter_check_slabinfo(skel);
+	if (test__start_subtest("open_coded_iter"))
+		subtest_kmem_cache_iter_open_coded(skel);
Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c void test_kmem_cache_iter(void) · line 116

Quoted 5 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +2 -2
  	close(iter_fd);
-free_link:
-	bpf_link__destroy(link);
+detach:
+	kmem_cache_iter__detach(skel);

nit. I think the kmem_cache_iter__destroy() below will also detach, so no need to explicit kmem_cache_iter__detach().

Ok, will remove.

Quoted 37 lines
Quoted 37 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +0 -0
  destroy:
  	kmem_cache_iter__destroy(skel);
  }
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +14 -0
index 72c9dafecd98406b..4c44aa279a5328fe 100644
@@ -2,6 +2,8 @@
  /* Copyright (c) 2024 Google */
  #include "bpf_iter.h"
+#include "bpf_experimental.h"
+#include "bpf_misc.h"
  #include <bpf/bpf_helpers.h>
  #include <bpf/bpf_tracing.h>
@@ -33,6 +35,7 @@ extern struct kmem_cache *bpf_get_kmem_cache(u64 addr) __ksym;
  /* Result, will be checked by userspace */
  int task_struct_found;
  int kmem_cache_seen;
+int open_coded_seen;
  SEC("iter/kmem_cache")
  int slab_info_collector(struct bpf_iter__kmem_cache *ctx)
@@ -85,3 +88,24 @@ int BPF_PROG(check_task_struct)
  		task_struct_found = -2;
  	return 0;
  }
+
+SEC("fentry.s/" SYS_PREFIX "sys_syncfs")
+int open_coded_iter(const void *ctx)
+{
+	struct kmem_cache *s;
+
+	bpf_for_each(kmem_cache, s) {
+		struct kmem_cache_result *r;
+		int idx = open_coded_seen;
+
+		r = bpf_map_lookup_elem(&slab_result, &idx);
Quoted 8 lines

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 105

Quoted 4 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +4 -0
+		if (r == NULL)
+			break;
+
+		open_coded_seen++;

I am not sure if this will work well if the testing system somehow has another process calling syncfs. It is probably a good idea to guard this by checking the tid of the test_progs at the beginning of this bpf prog.

Right, I'll add the tid check.

Thanks for the review,
Namhyung

#4 lore raw

Re: [PATCH bpf-next 2/2] selftests/bpf: Add a test for open coded kmem_cache iter

In reply to Namhyung Kim
Quoted 28 lines · Namhyung Kim

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

  $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
  ...
  #130/1   kmem_cache_iter/check_task_struct:OK
  #130/2   kmem_cache_iter/check_slabinfo:OK
  #130/3   kmem_cache_iter/open_coded_iter:OK
  #130     kmem_cache_iter:OK
  Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
 .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
 .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
 .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
 3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +0 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644

tools/testing/selftests/bpf/bpf_experimental.h extern int bpf_wq_set_callback_impl(struct bpf_wq *wq, · line 590

Quoted 9 lines
tools/testing/selftests/bpf/bpf_experimental.h +6 -0
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
                unsigned int flags__k, void *aux__ign) __ksym;
 #define bpf_wq_set_callback(timer, cb, flags) \
        bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+

we should be getting this from vmlinux.h nowadays, so this is probably unnecessary

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel) · line 74

Quoted 13 lines
tools/testing/selftests/bpf/bpf_experimental.h +0 -0
 #endif
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +4 -0
index 848d8fc9171fae45..a1fd3bc57c0b21bb 100644
@@ -68,12 +68,18 @@ static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel)
        fclose(fp);
 }
 
+static void subtest_kmem_cache_iter_open_coded(struct kmem_cache_iter *skel)
+{
+       /* To trigger the open coded iterator attached to the syscall */
+       syncfs(0);

what Martin said, you still need to filter by PID

Quoted 76 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +12 -12
+
+       /* It should be same as we've seen from the explicit iterator */
+       ASSERT_EQ(skel->bss->open_coded_seen, skel->bss->kmem_cache_seen, "open_code_seen_eq");
+}
+
 void test_kmem_cache_iter(void)
 {
-       DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
        struct kmem_cache_iter *skel = NULL;
-       union bpf_iter_link_info linfo = {};
-       struct bpf_link *link;
        char buf[256];
        int iter_fd;
 
@@ -81,16 +87,12 @@ void test_kmem_cache_iter(void)
        if (!ASSERT_OK_PTR(skel, "kmem_cache_iter__open_and_load"))
                return;
 
-       opts.link_info = &linfo;
-       opts.link_info_len = sizeof(linfo);
-
-       link = bpf_program__attach_iter(skel->progs.slab_info_collector, &opts);
-       if (!ASSERT_OK_PTR(link, "attach_iter"))
+       if (!ASSERT_OK(kmem_cache_iter__attach(skel), "skel_attach"))
                goto destroy;
 
-       iter_fd = bpf_iter_create(bpf_link__fd(link));
+       iter_fd = bpf_iter_create(bpf_link__fd(skel->links.slab_info_collector));
        if (!ASSERT_GE(iter_fd, 0, "iter_create"))
-               goto free_link;
+               goto detach;
 
        memset(buf, 0, sizeof(buf));
        while (read(iter_fd, buf, sizeof(buf) > 0)) {
@@ -105,11 +107,13 @@ void test_kmem_cache_iter(void)
                subtest_kmem_cache_iter_check_task_struct(skel);
        if (test__start_subtest("check_slabinfo"))
                subtest_kmem_cache_iter_check_slabinfo(skel);
+       if (test__start_subtest("open_coded_iter"))
+               subtest_kmem_cache_iter_open_coded(skel);
 
        close(iter_fd);
 
-free_link:
-       bpf_link__destroy(link);
+detach:
+       kmem_cache_iter__detach(skel);
 destroy:
        kmem_cache_iter__destroy(skel);
 }
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +4 -0
index 72c9dafecd98406b..4c44aa279a5328fe 100644
@@ -2,6 +2,8 @@
 /* Copyright (c) 2024 Google */
 
 #include "bpf_iter.h"
+#include "bpf_experimental.h"
+#include "bpf_misc.h"
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 
@@ -33,6 +35,7 @@ extern struct kmem_cache *bpf_get_kmem_cache(u64 addr) __ksym;
 /* Result, will be checked by userspace */
 int task_struct_found;
 int kmem_cache_seen;
+int open_coded_seen;
 
 SEC("iter/kmem_cache")
 int slab_info_collector(struct bpf_iter__kmem_cache *ctx)
@@ -85,3 +88,24 @@ int BPF_PROG(check_task_struct)
                task_struct_found = -2;
        return 0;
 }
+

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 101

Quoted 10 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +10 -0
+SEC("fentry.s/" SYS_PREFIX "sys_syncfs")
+int open_coded_iter(const void *ctx)
+{
+       struct kmem_cache *s;
+
+       bpf_for_each(kmem_cache, s) {
+               struct kmem_cache_result *r;
+               int idx = open_coded_seen;
+
+               r = bpf_map_lookup_elem(&slab_result, &idx);

nit: you don't need idx, just &open_coded_seen should be fine, I think

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 102

Quoted 1 line
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +1 -0
+               if (r == NULL)

nit: !r

Quoted 12 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +9 -1
+                       break;
+
+               open_coded_seen++;
+
+               if (r->obj_size != s->size)
+                       break;
+       }
+       return 0;
+}
--

2.47.0.rc1.288.g06298d1525-goog

#5 lore raw

Re: [PATCH bpf-next 2/2] selftests/bpf: Add a test for open coded kmem_cache iter

Quoted 33 lines · Andrii Nakryiko
Quoted 32 lines · Namhyung Kim

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

  $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
  ...
  #130/1   kmem_cache_iter/check_task_struct:OK
  #130/2   kmem_cache_iter/check_slabinfo:OK
  #130/3   kmem_cache_iter/open_coded_iter:OK
  #130     kmem_cache_iter:OK
  Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
 .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
 .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
 .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
 3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
                unsigned int flags__k, void *aux__ign) __ksym;
 #define bpf_wq_set_callback(timer, cb, flags) \
        bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
Quoted 8 lines

tools/testing/selftests/bpf/bpf_experimental.h extern int bpf_wq_set_callback_impl(struct bpf_wq *wq, · line 590

Quoted 5 lines
tools/testing/selftests/bpf/bpf_experimental.h +5 -0
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+

we should be getting this from vmlinux.h nowadays, so this is probably unnecessary

That'd be nice. Will remove.

Quoted 10 lines
Quoted 10 lines
tools/testing/selftests/bpf/bpf_experimental.h +0 -0
 #endif
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +0 -0
index 848d8fc9171fae45..a1fd3bc57c0b21bb 100644
@@ -68,12 +68,18 @@ static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel)
        fclose(fp);
 }
 
Quoted 6 lines

tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c static void subtest_kmem_cache_iter_check_slabinfo(struct kmem_cache_iter *skel) · line 74

Quoted 4 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +4 -0
+static void subtest_kmem_cache_iter_open_coded(struct kmem_cache_iter *skel)
+{
+       /* To trigger the open coded iterator attached to the syscall */
+       syncfs(0);

what Martin said, you still need to filter by PID

Yep.

Quoted 81 lines
Quoted 81 lines
tools/testing/selftests/bpf/prog_tests/kmem_cache_iter.c +12 -12
+
+       /* It should be same as we've seen from the explicit iterator */
+       ASSERT_EQ(skel->bss->open_coded_seen, skel->bss->kmem_cache_seen, "open_code_seen_eq");
+}
+
 void test_kmem_cache_iter(void)
 {
-       DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts);
        struct kmem_cache_iter *skel = NULL;
-       union bpf_iter_link_info linfo = {};
-       struct bpf_link *link;
        char buf[256];
        int iter_fd;
 
@@ -81,16 +87,12 @@ void test_kmem_cache_iter(void)
        if (!ASSERT_OK_PTR(skel, "kmem_cache_iter__open_and_load"))
                return;
 
-       opts.link_info = &linfo;
-       opts.link_info_len = sizeof(linfo);
-
-       link = bpf_program__attach_iter(skel->progs.slab_info_collector, &opts);
-       if (!ASSERT_OK_PTR(link, "attach_iter"))
+       if (!ASSERT_OK(kmem_cache_iter__attach(skel), "skel_attach"))
                goto destroy;
 
-       iter_fd = bpf_iter_create(bpf_link__fd(link));
+       iter_fd = bpf_iter_create(bpf_link__fd(skel->links.slab_info_collector));
        if (!ASSERT_GE(iter_fd, 0, "iter_create"))
-               goto free_link;
+               goto detach;
 
        memset(buf, 0, sizeof(buf));
        while (read(iter_fd, buf, sizeof(buf) > 0)) {
@@ -105,11 +107,13 @@ void test_kmem_cache_iter(void)
                subtest_kmem_cache_iter_check_task_struct(skel);
        if (test__start_subtest("check_slabinfo"))
                subtest_kmem_cache_iter_check_slabinfo(skel);
+       if (test__start_subtest("open_coded_iter"))
+               subtest_kmem_cache_iter_open_coded(skel);
 
        close(iter_fd);
 
-free_link:
-       bpf_link__destroy(link);
+detach:
+       kmem_cache_iter__detach(skel);
 destroy:
        kmem_cache_iter__destroy(skel);
 }
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +8 -0
index 72c9dafecd98406b..4c44aa279a5328fe 100644
@@ -2,6 +2,8 @@
 /* Copyright (c) 2024 Google */
 
 #include "bpf_iter.h"
+#include "bpf_experimental.h"
+#include "bpf_misc.h"
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 
@@ -33,6 +35,7 @@ extern struct kmem_cache *bpf_get_kmem_cache(u64 addr) __ksym;
 /* Result, will be checked by userspace */
 int task_struct_found;
 int kmem_cache_seen;
+int open_coded_seen;
 
 SEC("iter/kmem_cache")
 int slab_info_collector(struct bpf_iter__kmem_cache *ctx)
@@ -85,3 +88,24 @@ int BPF_PROG(check_task_struct)
                task_struct_found = -2;
        return 0;
 }
+
+SEC("fentry.s/" SYS_PREFIX "sys_syncfs")
+int open_coded_iter(const void *ctx)
+{
+       struct kmem_cache *s;
Quoted 8 lines

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 101

Quoted 6 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +6 -0
+
+       bpf_for_each(kmem_cache, s) {
+               struct kmem_cache_result *r;
+               int idx = open_coded_seen;
+
+               r = bpf_map_lookup_elem(&slab_result, &idx);

nit: you don't need idx, just &open_coded_seen should be fine, I think

Ok.

Quoted 4 lines

tools/testing/selftests/bpf/progs/kmem_cache_iter.c int BPF_PROG(check_task_struct) · line 102

Quoted 2 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +1 -0
+               if (r == NULL)

nit: !r

Will change!

Thanks,
Namhyung

Quoted 13 lines
Quoted 13 lines
tools/testing/selftests/bpf/progs/kmem_cache_iter.c +9 -1
+                       break;
+
+               open_coded_seen++;
+
+               if (r->obj_size != s->size)
+                       break;
+       }
+       return 0;
+}
--

2.47.0.rc1.288.g06298d1525-goog

#6 lore raw

Re: [PATCH bpf-next 2/2] selftests/bpf: Add a test for open coded kmem_cache iter

In reply to Andrii Nakryiko
Quoted 33 lines · Andrii Nakryiko
Quoted 32 lines · Namhyung Kim

The new subtest is attached to sleepable fentry of syncfs() syscall. It iterates the kmem_cache using bpf_for_each loop and count the number of entries. Finally it checks it with the number of entries from the regular iterator.

  $ ./vmtest.sh -- ./test_progs -t kmem_cache_iter
  ...
  #130/1   kmem_cache_iter/check_task_struct:OK
  #130/2   kmem_cache_iter/check_slabinfo:OK
  #130/3   kmem_cache_iter/open_coded_iter:OK
  #130     kmem_cache_iter:OK
  Summary: 1/3 PASSED, 0 SKIPPED, 0 FAILED

Also simplify the code by using attach routine of the skeleton.

---
 .../testing/selftests/bpf/bpf_experimental.h  |  6 ++++
 .../bpf/prog_tests/kmem_cache_iter.c          | 28 +++++++++++--------
 .../selftests/bpf/progs/kmem_cache_iter.c     | 24 ++++++++++++++++
 3 files changed, 46 insertions(+), 12 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index b0668f29f7b394eb..cd8ecd39c3f3c68d 100644
@@ -582,4 +582,10 @@ extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
                unsigned int flags__k, void *aux__ign) __ksym;
 #define bpf_wq_set_callback(timer, cb, flags) \
        bpf_wq_set_callback_impl(timer, cb, flags, NULL)
+
Quoted 8 lines

tools/testing/selftests/bpf/bpf_experimental.h extern int bpf_wq_set_callback_impl(struct bpf_wq *wq, · line 590

Quoted 5 lines
tools/testing/selftests/bpf/bpf_experimental.h +5 -0
+struct bpf_iter_kmem_cache;
+extern int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it) __weak __ksym;
+extern void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it) __weak __ksym;
+

we should be getting this from vmlinux.h nowadays, so this is probably unnecessary

I got some build errors without this. I'll leave it for v2.

Thanks,
Namhyung

#7 lore raw
In reply to Namhyung Kim
Quoted 61 lines · Namhyung Kim

Add a new open coded iterator for kmem_cache which can be called from a BPF program like below. It doesn't take any argument and traverses all kmem_cache entries.

   struct kmem_cache *pos;

   bpf_for_each(kmem_cache, pos) {
       ...
   }

As it needs to grab slab_mutex, it should be called from sleepable BPF programs only.

---
  kernel/bpf/helpers.c         |  3 ++
  kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
  2 files changed, 90 insertions(+)
kernel/bpf/helpers.c +3 -0
index 073e6f04f4d765ff..d1dfa4f335577914 100644
@@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
  BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
  BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
  BTF_ID_FLAGS(func, bpf_get_kmem_cache)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
  BTF_KFUNCS_END(common_btf_ids)
 
  static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/kmem_cache_iter.c +20 -0
index ebc101d7da51b57c..31ddaf452b20a458 100644
@@ -145,6 +145,93 @@ static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = {
  	.seq_ops		= &kmem_cache_iter_seq_ops,
  };
 
+/* open-coded version */
+struct bpf_iter_kmem_cache {
+	__u64 __opaque[1];
+} __attribute__((aligned(8)));
+
+struct bpf_iter_kmem_cache_kern {
+	struct kmem_cache *pos;
+} __attribute__((aligned(8)));
+
+__bpf_kfunc_start_defs();
+
+__bpf_kfunc int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+
+	BUILD_BUG_ON(sizeof(*kit) > sizeof(*it));
+	BUILD_BUG_ON(__alignof__(*kit) != __alignof__(*it));
+
+	kit->pos = NULL;
+	return 0;

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 177

Quoted 10 lines
kernel/bpf/kmem_cache_iter.c +10 -0
+}
+
+__bpf_kfunc struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+	struct kmem_cache *prev = kit->pos;
+	struct kmem_cache *next;
+	bool destroy = false;
+
+	mutex_lock(&slab_mutex);

I think taking mutex_lock here should be fine since sleepable tracing prog should be limited to the error injection whitelist. Those functions should not have held the mutex afaict.

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 187

Quoted 10 lines
kernel/bpf/kmem_cache_iter.c +10 -0
+
+	if (list_empty(&slab_caches)) {
+		mutex_unlock(&slab_mutex);
+		return NULL;
+	}
+
+	if (prev == NULL)
+		next = list_first_entry(&slab_caches, struct kmem_cache, list);
+	else if (list_last_entry(&slab_caches, struct kmem_cache, list) == prev)
+		next = NULL;

At the last entry, next is NULL.

Quoted 9 lines
kernel/bpf/kmem_cache_iter.c +9 -0
+	else
+		next = list_next_entry(prev, list);
+
+	/* boot_caches have negative refcount, don't touch them */
+	if (next && next->refcount > 0)
+		next->refcount++;
+
+	/* Skip kmem_cache_destroy() for active entries */
+	if (prev && prev->refcount > 1)

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 206

Quoted 10 lines
kernel/bpf/kmem_cache_iter.c +10 -0
+		prev->refcount--;
+	else if (prev && prev->refcount == 1)
+		destroy = true;
+
+	mutex_unlock(&slab_mutex);
+
+	if (destroy)
+		kmem_cache_destroy(prev);
+
+	kit->pos = next;

so kit->pos will be NULL also. Does it mean the bpf prog will be able to call bpf_iter_kmem_cache_next() again and re-loop from the beginning of the slab_caches list?

Quoted 31 lines
kernel/bpf/kmem_cache_iter.c +28 -0
+	return next;
+}
+
+__bpf_kfunc void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+	struct kmem_cache *s = kit->pos;
+	bool destroy = false;
+
+	if (s == NULL)
+		return;
+
+	mutex_lock(&slab_mutex);
+
+	/* Skip kmem_cache_destroy() for active entries */
+	if (s->refcount > 1)
+		s->refcount--;
+	else if (s->refcount == 1)
+		destroy = true;
+
+	mutex_unlock(&slab_mutex);
+
+	if (destroy)
+		kmem_cache_destroy(s);
+}
+
+__bpf_kfunc_end_defs();
+
  static void bpf_iter_kmem_cache_show_fdinfo(const struct bpf_iter_aux_info *aux,
  					    struct seq_file *seq)
  {
#8 lore raw

Hello,

Quoted 103 lines · Martin KaFai Lau
Quoted 59 lines · Namhyung Kim

Add a new open coded iterator for kmem_cache which can be called from a BPF program like below. It doesn't take any argument and traverses all kmem_cache entries.

   struct kmem_cache *pos;

   bpf_for_each(kmem_cache, pos) {
       ...
   }

As it needs to grab slab_mutex, it should be called from sleepable BPF programs only.

---
  kernel/bpf/helpers.c         |  3 ++
  kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
  2 files changed, 90 insertions(+)
kernel/bpf/helpers.c +3 -0
index 073e6f04f4d765ff..d1dfa4f335577914 100644
@@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
  BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
  BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
  BTF_ID_FLAGS(func, bpf_get_kmem_cache)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
  BTF_KFUNCS_END(common_btf_ids)
  static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/kmem_cache_iter.c +20 -0
index ebc101d7da51b57c..31ddaf452b20a458 100644
@@ -145,6 +145,93 @@ static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = {
  	.seq_ops		= &kmem_cache_iter_seq_ops,
  };
+/* open-coded version */
+struct bpf_iter_kmem_cache {
+	__u64 __opaque[1];
+} __attribute__((aligned(8)));
+
+struct bpf_iter_kmem_cache_kern {
+	struct kmem_cache *pos;
+} __attribute__((aligned(8)));
+
+__bpf_kfunc_start_defs();
+
+__bpf_kfunc int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+
+	BUILD_BUG_ON(sizeof(*kit) > sizeof(*it));
+	BUILD_BUG_ON(__alignof__(*kit) != __alignof__(*it));
+
+	kit->pos = NULL;
+	return 0;

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 176

Quoted 10 lines
kernel/bpf/kmem_cache_iter.c +10 -0
+}
+
+__bpf_kfunc struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+	struct kmem_cache *prev = kit->pos;
+	struct kmem_cache *next;
+	bool destroy = false;
+
+	mutex_lock(&slab_mutex);

I think taking mutex_lock here should be fine since sleepable tracing prog should be limited to the error injection whitelist. Those functions should not have held the mutex afaict.

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 187

Quoted 10 lines
kernel/bpf/kmem_cache_iter.c +10 -0
+
+	if (list_empty(&slab_caches)) {
+		mutex_unlock(&slab_mutex);
+		return NULL;
+	}
+
+	if (prev == NULL)
+		next = list_first_entry(&slab_caches, struct kmem_cache, list);
+	else if (list_last_entry(&slab_caches, struct kmem_cache, list) == prev)
+		next = NULL;

At the last entry, next is NULL.

Quoted 15 lines
kernel/bpf/kmem_cache_iter.c +15 -0
+	else
+		next = list_next_entry(prev, list);
+
+	/* boot_caches have negative refcount, don't touch them */
+	if (next && next->refcount > 0)
+		next->refcount++;
+
+	/* Skip kmem_cache_destroy() for active entries */
+	if (prev && prev->refcount > 1)
+		prev->refcount--;
+	else if (prev && prev->refcount == 1)
+		destroy = true;
+
+	mutex_unlock(&slab_mutex);
+
Quoted 8 lines

kernel/bpf/kmem_cache_iter.c static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = { · line 206

Quoted 4 lines
kernel/bpf/kmem_cache_iter.c +4 -0
+	if (destroy)
+		kmem_cache_destroy(prev);
+
+	kit->pos = next;

so kit->pos will be NULL also. Does it mean the bpf prog will be able to call bpf_iter_kmem_cache_next() again and re-loop from the beginning of the slab_caches list?

Right, I'll mark the start pos differently to prevent that.

Thanks,
Namhyung

Quoted 33 lines
Quoted 32 lines
kernel/bpf/kmem_cache_iter.c +28 -0
+	return next;
+}
+
+__bpf_kfunc void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it)
+{
+	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
+	struct kmem_cache *s = kit->pos;
+	bool destroy = false;
+
+	if (s == NULL)
+		return;
+
+	mutex_lock(&slab_mutex);
+
+	/* Skip kmem_cache_destroy() for active entries */
+	if (s->refcount > 1)
+		s->refcount--;
+	else if (s->refcount == 1)
+		destroy = true;
+
+	mutex_unlock(&slab_mutex);
+
+	if (destroy)
+		kmem_cache_destroy(s);
+}
+
+__bpf_kfunc_end_defs();
+
  static void bpf_iter_kmem_cache_show_fdinfo(const struct bpf_iter_aux_info *aux,
  					    struct seq_file *seq)
  {
#9 lore raw
In reply to Namhyung Kim
Quoted 25 lines · Namhyung Kim

Add a new open coded iterator for kmem_cache which can be called from a BPF program like below. It doesn't take any argument and traverses all kmem_cache entries.

struct kmem_cache *pos;

  bpf_for_each(kmem_cache, pos) {
      ...
  }

As it needs to grab slab_mutex, it should be called from sleepable BPF programs only.

---
 kernel/bpf/helpers.c         |  3 ++
 kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
 2 files changed, 90 insertions(+)
kernel/bpf/helpers.c +0 -0
index 073e6f04f4d765ff..d1dfa4f335577914 100644

kernel/bpf/helpers.c BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL) · line 3116

Quoted 6 lines
kernel/bpf/helpers.c +3 -0
@@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
 BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_get_kmem_cache)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)

I'm curious. Having bpf_iter_kmem_cache_{new,next,destroy} functions, can we rewrite kmem_cache_iter_seq_next in terms of these ones, so that we have less duplication of iteration logic? Or there will be some locking concerns preventing this? (I haven't looked into the actual logic much, sorry, lazy question)

Quoted 11 lines
kernel/bpf/helpers.c +0 -0
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/kmem_cache_iter.c +0 -0
index ebc101d7da51b57c..31ddaf452b20a458 100644
@@ -145,6 +145,93 @@ static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = {
        .seq_ops                = &kmem_cache_iter_seq_ops,
 };
 

[...]

#10 lore raw

Hello,

Quoted 30 lines · Andrii Nakryiko
Quoted 29 lines · Namhyung Kim

Add a new open coded iterator for kmem_cache which can be called from a BPF program like below. It doesn't take any argument and traverses all kmem_cache entries.

struct kmem_cache *pos;

  bpf_for_each(kmem_cache, pos) {
      ...
  }

As it needs to grab slab_mutex, it should be called from sleepable BPF programs only.

---
 kernel/bpf/helpers.c         |  3 ++
 kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
 2 files changed, 90 insertions(+)
kernel/bpf/helpers.c +1 -0
index 073e6f04f4d765ff..d1dfa4f335577914 100644
@@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
 BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_get_kmem_cache)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
Quoted 8 lines

kernel/bpf/helpers.c BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL) · line 3116

Quoted 2 lines
kernel/bpf/helpers.c +2 -0
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)

I'm curious. Having bpf_iter_kmem_cache_{new,next,destroy} functions, can we rewrite kmem_cache_iter_seq_next in terms of these ones, so that we have less duplication of iteration logic? Or there will be some locking concerns preventing this? (I haven't looked into the actual logic much, sorry, lazy question)

It should be fine with locking, I think there's a subtle difference between seq interface and the open coded iterator. But I'll think about how to reduce the duplication.

Thanks for your review!
Namhyung