diff mbox series

[bpf,v3] bpf: zero-fill re-used per-cpu map element

Message ID 20201103154738.29809-1-david.verbeiren@tessares.net
State Not Applicable
Delegated to: BPF Maintainers
Headers show
Series [bpf,v3] bpf: zero-fill re-used per-cpu map element | expand

Checks

Context Check Description
jkicinski/cover_letter success Link
jkicinski/fixes_present success Link
jkicinski/patch_count success Link
jkicinski/tree_selection success Clearly marked for bpf
jkicinski/subject_prefix success Link
jkicinski/source_inline success Was 0 now: 0
jkicinski/verify_signedoff success Link
jkicinski/module_param success Was 0 now: 0
jkicinski/build_32bit fail Errors and warnings before: 18 this patch: 18
jkicinski/kdoc success Errors and warnings before: 0 this patch: 0
jkicinski/verify_fixes success Link
jkicinski/checkpatch fail Link
jkicinski/build_allmodconfig_warn success Errors and warnings before: 14 this patch: 14
jkicinski/header_inline success Link
jkicinski/stable success Stable not CCed

Commit Message

David Verbeiren Nov. 3, 2020, 3:47 p.m. UTC
Zero-fill element values for all other cpus than current, just as
when not using prealloc. This is the only way the bpf program can
ensure known initial values for all cpus ('onallcpus' cannot be
set when coming from the bpf program).

The scenario is: bpf program inserts some elements in a per-cpu
map, then deletes some (or userspace does). When later adding
new elements using bpf_map_update_elem(), the bpf program can
only set the value of the new elements for the current cpu.
When prealloc is enabled, previously deleted elements are re-used.
Without the fix, values for other cpus remain whatever they were
when the re-used entry was previously freed.

A selftest is added to validate correct operation in above
scenario as well as in case of LRU per-cpu map element re-use.

Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements")
Acked-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David Verbeiren <david.verbeiren@tessares.net>
---

Notes:
    v3:
      - Added selftest that was initially provided as separate
        patch, and reworked to
        * use skeleton (Andrii, Song Liu)
        * skip test if <=1 CPU (Song Liu)
    
    v2:
      - Moved memset() to separate pcpu_init_value() function,
        which replaces pcpu_copy_value() but delegates to it
        for the cases where no memset() is needed (Andrii).
      - This function now also avoids doing the memset() for
        the current cpu for which the value must be set
        anyhow (Andrii).
      - Same pcpu_init_value() used for per-cpu LRU map
        (Andrii).

 kernel/bpf/hashtab.c                          |  30 ++-
 .../selftests/bpf/prog_tests/map_init.c       | 213 ++++++++++++++++++
 .../selftests/bpf/progs/test_map_init.c       |  34 +++
 3 files changed, 275 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/map_init.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_map_init.c

Comments

Andrii Nakryiko Nov. 3, 2020, 6:19 p.m. UTC | #1
On Tue, Nov 3, 2020 at 7:49 AM David Verbeiren
<david.verbeiren@tessares.net> wrote:
>
> Zero-fill element values for all other cpus than current, just as
> when not using prealloc. This is the only way the bpf program can
> ensure known initial values for all cpus ('onallcpus' cannot be
> set when coming from the bpf program).
>
> The scenario is: bpf program inserts some elements in a per-cpu
> map, then deletes some (or userspace does). When later adding
> new elements using bpf_map_update_elem(), the bpf program can
> only set the value of the new elements for the current cpu.
> When prealloc is enabled, previously deleted elements are re-used.
> Without the fix, values for other cpus remain whatever they were
> when the re-used entry was previously freed.
>
> A selftest is added to validate correct operation in above
> scenario as well as in case of LRU per-cpu map element re-use.
>
> Fixes: 6c9059817432 ("bpf: pre-allocate hash map elements")
> Acked-by: Matthieu Baerts <matthieu.baerts@tessares.net>
> Signed-off-by: David Verbeiren <david.verbeiren@tessares.net>
> ---
>

Tests look really nice, thanks! I'm worried about still racy once
check, see suggestions below. Otherwise looks great!

> Notes:
>     v3:
>       - Added selftest that was initially provided as separate
>         patch, and reworked to
>         * use skeleton (Andrii, Song Liu)
>         * skip test if <=1 CPU (Song Liu)
>
>     v2:
>       - Moved memset() to separate pcpu_init_value() function,
>         which replaces pcpu_copy_value() but delegates to it
>         for the cases where no memset() is needed (Andrii).
>       - This function now also avoids doing the memset() for
>         the current cpu for which the value must be set
>         anyhow (Andrii).
>       - Same pcpu_init_value() used for per-cpu LRU map
>         (Andrii).
>
>  kernel/bpf/hashtab.c                          |  30 ++-
>  .../selftests/bpf/prog_tests/map_init.c       | 213 ++++++++++++++++++
>  .../selftests/bpf/progs/test_map_init.c       |  34 +++
>  3 files changed, 275 insertions(+), 2 deletions(-)
>  create mode 100644 tools/testing/selftests/bpf/prog_tests/map_init.c
>  create mode 100644 tools/testing/selftests/bpf/progs/test_map_init.c
>

[...]

> diff --git a/tools/testing/selftests/bpf/prog_tests/map_init.c b/tools/testing/selftests/bpf/prog_tests/map_init.c
> new file mode 100644
> index 000000000000..386d9439bad9
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/map_init.c
> @@ -0,0 +1,213 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +// Copyright (c) 2020 Tessares SA <http://www.tessares.net>
> +

nit: see below, /* */

> +#include <test_progs.h>
> +#include "test_map_init.skel.h"
> +

[...]

> diff --git a/tools/testing/selftests/bpf/progs/test_map_init.c b/tools/testing/selftests/bpf/progs/test_map_init.c
> new file mode 100644
> index 000000000000..280a45e366d6
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/test_map_init.c
> @@ -0,0 +1,34 @@
> +// SPDX-License-Identifier: GPL-2.0
> +// Copyright (c) 2020 Tessares SA <http://www.tessares.net>

nit: I think copyright line has to be in /* */ comment block

> +
> +#include "vmlinux.h"
> +#include <bpf/bpf_helpers.h>
> +
> +__u64 inKey = 0;
> +__u64 inValue = 0;
> +__u32 once = 0;
> +
> +struct {
> +       __uint(type, BPF_MAP_TYPE_PERCPU_HASH);
> +       __uint(max_entries, 2);
> +       __type(key, __u64);
> +       __type(value, __u64);
> +} hashmap1 SEC(".maps");
> +
> +
> +SEC("raw_tp/sys_enter")
> +int sys_enter(const void *ctx)
> +{
> +       /* Just do it once so the value is only updated for a single CPU.
> +        * Indeed, this tracepoint will quickly be hit from different CPUs.
> +        */
> +       if (!once) {
> +               __sync_fetch_and_add(&once, 1);

This is quite racy, actually, especially for the generic sys_enter
tracepoint. The way I did this before (see progs/trigger_bench.c) was
through doing a "tp/syscalls/sys_enter_getpgid" tracepoint program and
checking for thread id. Or you can use bpf_test_run, probably, with a
different type of BPF program. I just find bpf_test_run() too
inconvenient with all the extra setup, so I usually stick to
tracepoints.

> +
> +               bpf_map_update_elem(&hashmap1, &inKey, &inValue, BPF_NOEXIST);
> +       }
> +
> +       return 0;
> +}
> +
> +char _license[] SEC("license") = "GPL";
> --
> 2.29.0
>
diff mbox series

Patch

diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index 1815e97d4c9c..1fccba6e88c4 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -821,6 +821,32 @@  static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
 	}
 }
 
+static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
+			    void *value, bool onallcpus)
+{
+	/* When using prealloc and not setting the initial value on all cpus,
+	 * zero-fill element values for other cpus (just as what happens when
+	 * not using prealloc). Otherwise, bpf program has no way to ensure
+	 * known initial values for cpus other than current one
+	 * (onallcpus=false always when coming from bpf prog).
+	 */
+	if (htab_is_prealloc(htab) && !onallcpus) {
+		u32 size = round_up(htab->map.value_size, 8);
+		int current_cpu = raw_smp_processor_id();
+		int cpu;
+
+		for_each_possible_cpu(cpu) {
+			if (cpu == current_cpu)
+				bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value,
+						size);
+			else
+				memset(per_cpu_ptr(pptr, cpu), 0, size);
+		}
+	} else {
+		pcpu_copy_value(htab, pptr, value, onallcpus);
+	}
+}
+
 static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
 {
 	return htab->map.map_type == BPF_MAP_TYPE_HASH_OF_MAPS &&
@@ -891,7 +917,7 @@  static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 			}
 		}
 
-		pcpu_copy_value(htab, pptr, value, onallcpus);
+		pcpu_init_value(htab, pptr, value, onallcpus);
 
 		if (!prealloc)
 			htab_elem_set_ptr(l_new, key_size, pptr);
@@ -1183,7 +1209,7 @@  static int __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
 				value, onallcpus);
 	} else {
-		pcpu_copy_value(htab, htab_elem_get_ptr(l_new, key_size),
+		pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
 				value, onallcpus);
 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
 		l_new = NULL;
diff --git a/tools/testing/selftests/bpf/prog_tests/map_init.c b/tools/testing/selftests/bpf/prog_tests/map_init.c
new file mode 100644
index 000000000000..386d9439bad9
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/map_init.c
@@ -0,0 +1,213 @@ 
+// SPDX-License-Identifier: GPL-2.0-only
+// Copyright (c) 2020 Tessares SA <http://www.tessares.net>
+
+#include <test_progs.h>
+#include "test_map_init.skel.h"
+
+#define TEST_VALUE 0x1234
+#define FILL_VALUE 0xdeadbeef
+
+static int nr_cpus;
+static int duration;
+
+typedef unsigned long long map_key_t;
+typedef unsigned long long map_value_t;
+typedef struct {
+	map_value_t v; /* padding */
+} __bpf_percpu_val_align pcpu_map_value_t;
+
+
+static int map_populate(int map_fd, int num)
+{
+	pcpu_map_value_t value[nr_cpus];
+	int i, err;
+	map_key_t key;
+
+	for (i = 0; i < nr_cpus; i++)
+		bpf_percpu(value, i) = FILL_VALUE;
+
+	for (key = 1; key <= num; key++) {
+		err = bpf_map_update_elem(map_fd, &key, value, BPF_NOEXIST);
+		if (!ASSERT_OK(err, "bpf_map_update_elem"))
+			return -1;
+	}
+
+	return 0;
+}
+
+static struct test_map_init *setup(enum bpf_map_type map_type, int map_sz,
+			    int *map_fd, int populate)
+{
+	struct test_map_init *skel;
+	int err;
+
+	skel = test_map_init__open();
+	if (!ASSERT_OK_PTR(skel, "skel_open"))
+		return NULL;
+
+	err = bpf_map__set_type(skel->maps.hashmap1, map_type);
+	if (!ASSERT_OK(err, "bpf_map__set_type"))
+		goto error;
+
+	err = bpf_map__set_max_entries(skel->maps.hashmap1, map_sz);
+	if (!ASSERT_OK(err, "bpf_map__set_max_entries"))
+		goto error;
+
+	err = test_map_init__load(skel);
+	if (!ASSERT_OK(err, "skel_load"))
+		goto error;
+
+	*map_fd = bpf_map__fd(skel->maps.hashmap1);
+	if (CHECK(*map_fd < 0, "bpf_map__fd", "failed\n"))
+		goto error;
+
+	err = map_populate(*map_fd, populate);
+	if (!ASSERT_OK(err, "map_populate"))
+		goto error_map;
+
+	return skel;
+
+error_map:
+	close(*map_fd);
+error:
+	test_map_init__destroy(skel);
+	return NULL;
+}
+
+/* executes bpf program that updates map with key, value */
+static int prog_run_insert_elem(struct test_map_init *skel, map_key_t key,
+				map_value_t value)
+{
+	struct test_map_init__bss *bss;
+
+	bss = skel->bss;
+
+	bss->inKey = key;
+	bss->inValue = value;
+
+	if (!ASSERT_OK(test_map_init__attach(skel), "skel_attach"))
+		return -1;
+
+	/* Let tracepoint trigger */
+	usleep(1);
+
+	test_map_init__detach(skel);
+
+	return 0;
+}
+
+static int check_values_one_cpu(pcpu_map_value_t *value, map_value_t expected)
+{
+	int i, nzCnt = 0;
+	map_value_t val;
+
+	for (i = 0; i < nr_cpus; i++) {
+		val = bpf_percpu(value, i);
+		if (val) {
+			if (CHECK(val != expected, "map value",
+				  "unexpected for cpu %d: 0x%llx\n", i, val))
+				return -1;
+			nzCnt++;
+		}
+	}
+
+	if (CHECK(nzCnt != 1, "map value", "set for %d CPUs instead of 1!\n",
+		  nzCnt))
+		return -1;
+
+	return 0;
+}
+
+/* Add key=1 elem with values set for all CPUs
+ * Delete elem key=1
+ * Run bpf prog that inserts new key=1 elem with value=0x1234
+ *   (bpf prog can only set value for current CPU)
+ * Lookup Key=1 and check value is as expected for all CPUs:
+ *   value set by bpf prog for one CPU, 0 for all others
+ */
+static void test_pcpu_map_init(void)
+{
+	pcpu_map_value_t value[nr_cpus];
+	struct test_map_init *skel;
+	int map_fd, err;
+	map_key_t key;
+
+	/* max 1 elem in map so insertion is forced to reuse freed entry */
+	skel = setup(BPF_MAP_TYPE_PERCPU_HASH, 1, &map_fd, 1);
+	if (!ASSERT_OK_PTR(skel, "prog_setup"))
+		return;
+
+	/* delete element so the entry can be re-used*/
+	key = 1;
+	err = bpf_map_delete_elem(map_fd, &key);
+	if (!ASSERT_OK(err, "bpf_map_delete_elem"))
+		goto cleanup;
+
+	/* run bpf prog that inserts new elem, re-using the slot just freed */
+	err = prog_run_insert_elem(skel, key, TEST_VALUE);
+	if (!ASSERT_OK(err, "prog_run_insert_elem"))
+		goto cleanup;
+
+	/* check that key=1 was re-created by bpf prog */
+	err = bpf_map_lookup_elem(map_fd, &key, value);
+	if (!ASSERT_OK(err, "bpf_map_lookup_elem"))
+		goto cleanup;
+
+	/* and has expected values */
+	check_values_one_cpu(value, TEST_VALUE);
+
+cleanup:
+	test_map_init__destroy(skel);
+}
+
+/* Add key=1 and key=2 elems with values set for all CPUs
+ * Run bpf prog that inserts new key=3 elem
+ *   (only for current cpu; other cpus should have initial value = 0)
+ * Lookup Key=1 and check value is as expected for all CPUs
+ */
+static void test_pcpu_lru_map_init(void)
+{
+	pcpu_map_value_t value[nr_cpus];
+	struct test_map_init *skel;
+	int map_fd, err;
+	map_key_t key;
+
+	/* Set up LRU map with 2 elements, values filled for all CPUs.
+	 * With these 2 elements, the LRU map is full
+	 */
+	skel = setup(BPF_MAP_TYPE_LRU_PERCPU_HASH, 2, &map_fd, 2);
+	if (!ASSERT_OK_PTR(skel, "prog_setup"))
+		return;
+
+	/* run bpf prog that inserts new key=3 element, re-using LRU slot */
+	key = 3;
+	err = prog_run_insert_elem(skel, key, TEST_VALUE);
+	if (!ASSERT_OK(err, "prog_run_insert_elem"))
+		goto cleanup;
+
+	/* check that key=3 replaced one of earlier elements */
+	err = bpf_map_lookup_elem(map_fd, &key, value);
+	if (!ASSERT_OK(err, "bpf_map_lookup_elem"))
+		goto cleanup;
+
+	/* and has expected values */
+	check_values_one_cpu(value, TEST_VALUE);
+
+cleanup:
+	test_map_init__destroy(skel);
+}
+
+void test_map_init(void)
+{
+	nr_cpus = bpf_num_possible_cpus();
+	if (nr_cpus <= 1) {
+		printf("%s:SKIP: >1 cpu needed for this test\n", __func__);
+		test__skip();
+		return;
+	}
+
+	if (test__start_subtest("pcpu_map_init"))
+		test_pcpu_map_init();
+	if (test__start_subtest("pcpu_lru_map_init"))
+		test_pcpu_lru_map_init();
+}
diff --git a/tools/testing/selftests/bpf/progs/test_map_init.c b/tools/testing/selftests/bpf/progs/test_map_init.c
new file mode 100644
index 000000000000..280a45e366d6
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/test_map_init.c
@@ -0,0 +1,34 @@ 
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2020 Tessares SA <http://www.tessares.net>
+
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+
+__u64 inKey = 0;
+__u64 inValue = 0;
+__u32 once = 0;
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_HASH);
+	__uint(max_entries, 2);
+	__type(key, __u64);
+	__type(value, __u64);
+} hashmap1 SEC(".maps");
+
+
+SEC("raw_tp/sys_enter")
+int sys_enter(const void *ctx)
+{
+	/* Just do it once so the value is only updated for a single CPU.
+	 * Indeed, this tracepoint will quickly be hit from different CPUs.
+	 */
+	if (!once) {
+		__sync_fetch_and_add(&once, 1);
+
+		bpf_map_update_elem(&hashmap1, &inKey, &inValue, BPF_NOEXIST);
+	}
+
+	return 0;
+}
+
+char _license[] SEC("license") = "GPL";