[bpf-next,v6,1/1] bpf, docs: document BPF_MAP_TYPE_ARRAY

Commit Message

Donald Hunter Oct. 5, 2022, 10:46 a.m. UTC

From: Dave Tucker <[e-mail]>

Add documentation for the BPF_MAP_TYPE_ARRAY including kernel version
introduced, usage and examples. Also document BPF_MAP_TYPE_PERCPU_ARRAY
which is similar.

Signed-off-by: Dave Tucker <[e-mail]>
Signed-off-by: Donald Hunter <[e-mail]>
---
 Documentation/bpf/map_array.rst | 231 ++++++++++++++++++++++++++++++++
 1 file changed, 231 insertions(+)
 create mode 100644 Documentation/bpf/map_array.rst

Comments

Jiri Olsa Oct. 6, 2022, 11:36 a.m. UTC | #1

On Wed, Oct 05, 2022 at 11:46:34AM +0100, Donald Hunter wrote:
> From: Dave Tucker <[e-mail]>
> 
> Add documentation for the BPF_MAP_TYPE_ARRAY including kernel version
> introduced, usage and examples. Also document BPF_MAP_TYPE_PERCPU_ARRAY
> which is similar.
> 
> Signed-off-by: Dave Tucker <[e-mail]>
> Signed-off-by: Donald Hunter <[e-mail]>
> ---
>  Documentation/bpf/map_array.rst | 231 ++++++++++++++++++++++++++++++++
>  1 file changed, 231 insertions(+)
>  create mode 100644 Documentation/bpf/map_array.rst
> 
> diff --git a/Documentation/bpf/map_array.rst b/Documentation/bpf/map_array.rst
> new file mode 100644
> index 000000000000..9d2da884c41e
> --- /dev/null
> +++ b/Documentation/bpf/map_array.rst
> @@ -0,0 +1,231 @@
> +.. SPDX-License-Identifier: GPL-2.0-only
> +.. Copyright (C) 2022 Red Hat, Inc.
> +
> +================================================
> +BPF_MAP_TYPE_ARRAY and BPF_MAP_TYPE_PERCPU_ARRAY
> +================================================
> +
> +.. note::
> +   - ``BPF_MAP_TYPE_ARRAY`` was introduced in kernel version 3.19
> +   - ``BPF_MAP_TYPE_PERCPU_ARRAY`` was introduced in version 4.6
> +
> +``BPF_MAP_TYPE_ARRAY`` and ``BPF_MAP_TYPE_PERCPU_ARRAY`` provide generic array
> +storage. The key type is an unsigned 32-bit integer (4 bytes) and the map is of
> +constant size. All array elements are pre-allocated and zero initialized when
> +created. ``BPF_MAP_TYPE_PERCPU_ARRAY`` uses a different memory region for each
> +CPU whereas ``BPF_MAP_TYPE_ARRAY`` uses the same memory region. The maximum
> +size of an array, defined in max_entries, is limited to 2^32. The value stored
> +can be of any size, however, small values will be rounded up to 8 bytes.

I recently hit 32k size limit for per-cpu map value.. it seems to be
size limit for generic per cpu allocation, but would be great to have
it confirmed by somebody who knows mm better ;-)

jirka

> +
> +Since kernel 5.5, memory mapping may be enabled for ``BPF_MAP_TYPE_ARRAY`` by
> +setting the flag ``BPF_F_MMAPABLE``. The map definition is page-aligned and
> +starts on the first page. Sufficient page-sized and page-aligned blocks of
> +memory are allocated to store all array values, starting on the second page,
> +which in some cases will result in over-allocation of memory. The benefit of
> +using this is increased performance and ease of use since userspace programs
> +would not be required to use helper functions to access and mutate data.
> +
> +Usage
> +=====
> +
> +.. c:function::
> +   void *bpf_map_lookup_elem(struct bpf_map *map, const void *key)
> +
> +Array elements can be retrieved using the ``bpf_map_lookup_elem()`` helper.
> +This helper returns a pointer into the array element, so to avoid data races
> +with userspace reading the value, the user must use primitives like
> +``__sync_fetch_and_add()`` when updating the value in-place. Access from
> +userspace uses the libbpf API of the same name.
> +
> +.. c:function::
> +   long bpf_map_update_elem(struct bpf_map *map, const void *key, const void *value, u64 flags)
> +
> +Array elements can also be added using the ``bpf_map_update_elem()`` helper or
> +libbpf API.
> +
> +``bpf_map_update_elem()`` returns 0 on success, or negative error in case of
> +failure.
> +
> +Since the array is of constant size, ``bpf_map_delete_elem()`` is not supported.
> +To clear an array element, you may use ``bpf_map_update_elem()`` to insert a
> +zero value to that index.
> +
> +Per CPU Array
> +-------------
> +
> +Values stored in ``BPF_MAP_TYPE_ARRAY`` can be accessed by multiple programs
> +across different CPUs. To restrict storage to a single CPU, you may use a
> +``BPF_MAP_TYPE_PERCPU_ARRAY``.
> +
> +When using a ``BPF_MAP_TYPE_PERCPU_ARRAY`` the ``bpf_map_update_elem()`` and
> +``bpf_map_lookup_elem()`` helpers automatically access the hash slot for the
> +current CPU.
> +
> +.. c:function::
> +   void *bpf_map_lookup_percpu_elem(struct bpf_map *map, const void *key, u32 cpu)
> +
> +The ``bpf_map_lookup_percpu_elem()`` helper can be used to lookup the array
> +value for a specific CPU. Returns value on success , or ``NULL`` if no entry was
> +found or ``cpu`` is invalid.
> +
> +Concurrency
> +-----------
> +
> +Since kernel version 5.1, the BPF infrastructure provides ``struct bpf_spin_lock``
> +to synchronize access.
> +
> +Examples
> +========
> +
> +Please see the ``tools/testing/selftests/bpf`` directory for functional
> +examples. The code samples below demonstrate API usage.
> +
> +Kernel BPF
> +----------
> +
> +This snippet shows how to declare an array in a BPF program.
> +
> +.. code-block:: c
> +
> +    struct {
> +            __uint(type, BPF_MAP_TYPE_ARRAY);
> +            __type(key, u32);
> +            __type(value, long);
> +            __uint(max_entries, 256);
> +    } my_map SEC(".maps");
> +
> +
> +This example BPF program shows how to access an array element.
> +
> +.. code-block:: c
> +
> +    int bpf_prog(struct __sk_buff *skb)
> +    {
> +            int index = load_byte(skb,
> +                                  ETH_HLEN + offsetof(struct iphdr, protocol));
> +            long *value;
> +
> +            if (skb->pkt_type != PACKET_OUTGOING)
> +                    return 0;
> +
> +            value = bpf_map_lookup_elem(&my_map, &index);
> +            if (value)
> +                    __sync_fetch_and_add(value, skb->len);
> +
> +            return 0;
> +    }
> +
> +Userspace
> +---------
> +
> +BPF_MAP_TYPE_ARRAY
> +~~~~~~~~~~~~~~~~~~
> +
> +This snippet shows how to create an array, using ``bpf_map_create_opts`` to
> +set flags.
> +
> +.. code-block:: c
> +
> +    #include <bpf/libbpf.h>
> +    #include <bpf/bpf.h>
> +
> +    int create_array() {
> +            int fd;
> +            LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
> +            fd = bpf_map_create(BPF_MAP_TYPE_ARRAY,
> +                                "example_array",       /* name */
> +                                sizeof(__u32),         /* key size */
> +                                sizeof(long),          /* value size */
> +                                256,                   /* max entries */
> +                                &opts);                /* create opts */
> +            return fd;
> +    }
> +
> +This snippet shows how to initialize the elements of an array.
> +
> +.. code-block:: c
> +
> +    int initialize_array(int fd) {
> +            __u32 i;
> +            long value;
> +            int ret;
> +
> +            for (i = 0; i < 256; i++) {
> +                    value = i;
> +                    ret = bpf_map_update_elem(fd, &i, &value, BPF_ANY);
> +                    if (ret < 0)
> +                            return ret;
> +            }
> +
> +            return ret;
> +    }
> +
> +This snippet shows how to retrieve an element value from an array.
> +
> +.. code-block:: c
> +
> +    int lookup(int fd) {
> +            __u32 index = 42;
> +            long value;
> +            int ret = bpf_map_lookup_elem(fd, &index, &value);
> +            if (ret < 0)
> +                    return ret;
> +
> +            /* use value here */
> +            assert(value == 42);
> +
> +            return ret;
> +    }
> +
> +BPF_MAP_TYPE_PERCPU_ARRAY
> +~~~~~~~~~~~~~~~~~~~~~~~~~
> +
> +This snippet shows how to initialize the elements of a per CPU array.
> +
> +.. code-block:: c
> +
> +    int initialize_array(int fd) {
> +            int ncpus = libbpf_num_possible_cpus();
> +            long values[ncpus];
> +            __u32 i, j;
> +            int ret;
> +
> +            for (i = 0; i < 256 ; i++) {
> +                    for (j = 0; j < ncpus; j++)
> +                            values[j] = i;
> +                    ret = bpf_map_update_elem(fd, &i, &values, BPF_ANY);
> +                    if (ret < 0)
> +                            return ret;
> +            }
> +
> +            return ret;
> +    }
> +
> +This snippet shows how to access the per CPU elements of an array value.
> +
> +.. code-block:: c
> +
> +    int lookup(int fd) {
> +            int ncpus = libbpf_num_possible_cpus();
> +            __u32 index = 42, j;
> +            long values[ncpus];
> +            int ret = bpf_map_lookup_elem(fd, &index, &values);
> +            if (ret < 0)
> +                    return ret;
> +
> +            for (j = 0; j < ncpus; j++) {
> +                    /* Use per CPU value here */
> +                    assert(values[j] == 42);
> +            }
> +
> +            return ret;
> +    }
> +
> +Semantics
> +=========
> +
> +As shown in the example above, when accessing a ``BPF_MAP_TYPE_PERCPU_ARRAY``
> +in userspace, each value is an array with ``ncpus`` elements.
> +
> +When calling ``bpf_map_update_elem()`` the flag ``BPF_NOEXIST`` can not be used
> +for these maps.
> -- 
> 2.35.1
>

Daniel Borkmann Oct. 7, 2022, 2:54 p.m. UTC | #2

On 10/6/22 1:36 PM, Jiri Olsa wrote:
> On Wed, Oct 05, 2022 at 11:46:34AM +0100, Donald Hunter wrote:
>> From: Dave Tucker <[e-mail]>
>>
>> Add documentation for the BPF_MAP_TYPE_ARRAY including kernel version
>> introduced, usage and examples. Also document BPF_MAP_TYPE_PERCPU_ARRAY
>> which is similar.
>>
>> Signed-off-by: Dave Tucker <[e-mail]>
>> Signed-off-by: Donald Hunter <[e-mail]>
>> ---
>>   Documentation/bpf/map_array.rst | 231 ++++++++++++++++++++++++++++++++
>>   1 file changed, 231 insertions(+)
>>   create mode 100644 Documentation/bpf/map_array.rst
>>
>> diff --git a/Documentation/bpf/map_array.rst b/Documentation/bpf/map_array.rst
>> new file mode 100644
>> index 000000000000..9d2da884c41e
>> --- /dev/null
>> +++ b/Documentation/bpf/map_array.rst
>> @@ -0,0 +1,231 @@
>> +.. SPDX-License-Identifier: GPL-2.0-only
>> +.. Copyright (C) 2022 Red Hat, Inc.
>> +
>> +================================================
>> +BPF_MAP_TYPE_ARRAY and BPF_MAP_TYPE_PERCPU_ARRAY
>> +================================================
>> +
>> +.. note::
>> +   - ``BPF_MAP_TYPE_ARRAY`` was introduced in kernel version 3.19
>> +   - ``BPF_MAP_TYPE_PERCPU_ARRAY`` was introduced in version 4.6
>> +
>> +``BPF_MAP_TYPE_ARRAY`` and ``BPF_MAP_TYPE_PERCPU_ARRAY`` provide generic array
>> +storage. The key type is an unsigned 32-bit integer (4 bytes) and the map is of
>> +constant size. All array elements are pre-allocated and zero initialized when
>> +created. ``BPF_MAP_TYPE_PERCPU_ARRAY`` uses a different memory region for each
>> +CPU whereas ``BPF_MAP_TYPE_ARRAY`` uses the same memory region. The maximum
>> +size of an array, defined in max_entries, is limited to 2^32. The value stored
>> +can be of any size, however, small values will be rounded up to 8 bytes.
> 
> I recently hit 32k size limit for per-cpu map value.. it seems to be
> size limit for generic per cpu allocation, but would be great to have
> it confirmed by somebody who knows mm better ;-)

Yes, for percpu the max is PCPU_MIN_UNIT_SIZE which is 32k, see mm/percpu.c +1756.
In many cases it's implementation specific, so it probably does not make too much
sense to state limits like 2^32, or at least it should say that its theoretical/uapi
limit and actual limits may be implementation/config specific.

Donald Hunter Oct. 7, 2022, 4:07 p.m. UTC | #3

Daniel Borkmann <[e-mail]> writes:

> On 10/6/22 1:36 PM, Jiri Olsa wrote:
>> I recently hit 32k size limit for per-cpu map value.. it seems to be
>> size limit for generic per cpu allocation, but would be great to have
>> it confirmed by somebody who knows mm better ;-)
>
> Yes, for percpu the max is PCPU_MIN_UNIT_SIZE which is 32k, see mm/percpu.c +1756.
> In many cases it's implementation specific, so it probably does not make too much
> sense to state limits like 2^32, or at least it should say that its theoretical/uapi
> limit and actual limits may be implementation/config specific.

Yes, good point. I will just drop the 2^32 bit. Same issue was
raised by Stanislav Fomichev for BPF_MAP_TYPE_HASH here:

https://patchwork.kernel.org/project/netdevbpf/patch/[e-mail]/#24936386

Patch

diff --git a/Documentation/bpf/map_array.rst b/Documentation/bpf/map_array.rst
new file mode 100644
index 000000000000..9d2da884c41e
--- /dev/null
+++ b/Documentation/bpf/map_array.rst
@@ -0,0 +1,231 @@ 
+.. SPDX-License-Identifier: GPL-2.0-only
+.. Copyright (C) 2022 Red Hat, Inc.
+
+================================================
+BPF_MAP_TYPE_ARRAY and BPF_MAP_TYPE_PERCPU_ARRAY
+================================================
+
+.. note::
+   - ``BPF_MAP_TYPE_ARRAY`` was introduced in kernel version 3.19
+   - ``BPF_MAP_TYPE_PERCPU_ARRAY`` was introduced in version 4.6
+
+``BPF_MAP_TYPE_ARRAY`` and ``BPF_MAP_TYPE_PERCPU_ARRAY`` provide generic array
+storage. The key type is an unsigned 32-bit integer (4 bytes) and the map is of
+constant size. All array elements are pre-allocated and zero initialized when
+created. ``BPF_MAP_TYPE_PERCPU_ARRAY`` uses a different memory region for each
+CPU whereas ``BPF_MAP_TYPE_ARRAY`` uses the same memory region. The maximum
+size of an array, defined in max_entries, is limited to 2^32. The value stored
+can be of any size, however, small values will be rounded up to 8 bytes.
+
+Since kernel 5.5, memory mapping may be enabled for ``BPF_MAP_TYPE_ARRAY`` by
+setting the flag ``BPF_F_MMAPABLE``. The map definition is page-aligned and
+starts on the first page. Sufficient page-sized and page-aligned blocks of
+memory are allocated to store all array values, starting on the second page,
+which in some cases will result in over-allocation of memory. The benefit of
+using this is increased performance and ease of use since userspace programs
+would not be required to use helper functions to access and mutate data.
+
+Usage
+=====
+
+.. c:function::
+   void *bpf_map_lookup_elem(struct bpf_map *map, const void *key)
+
+Array elements can be retrieved using the ``bpf_map_lookup_elem()`` helper.
+This helper returns a pointer into the array element, so to avoid data races
+with userspace reading the value, the user must use primitives like
+``__sync_fetch_and_add()`` when updating the value in-place. Access from
+userspace uses the libbpf API of the same name.
+
+.. c:function::
+   long bpf_map_update_elem(struct bpf_map *map, const void *key, const void *value, u64 flags)
+
+Array elements can also be added using the ``bpf_map_update_elem()`` helper or
+libbpf API.
+
+``bpf_map_update_elem()`` returns 0 on success, or negative error in case of
+failure.
+
+Since the array is of constant size, ``bpf_map_delete_elem()`` is not supported.
+To clear an array element, you may use ``bpf_map_update_elem()`` to insert a
+zero value to that index.
+
+Per CPU Array
+-------------
+
+Values stored in ``BPF_MAP_TYPE_ARRAY`` can be accessed by multiple programs
+across different CPUs. To restrict storage to a single CPU, you may use a
+``BPF_MAP_TYPE_PERCPU_ARRAY``.
+
+When using a ``BPF_MAP_TYPE_PERCPU_ARRAY`` the ``bpf_map_update_elem()`` and
+``bpf_map_lookup_elem()`` helpers automatically access the hash slot for the
+current CPU.
+
+.. c:function::
+   void *bpf_map_lookup_percpu_elem(struct bpf_map *map, const void *key, u32 cpu)
+
+The ``bpf_map_lookup_percpu_elem()`` helper can be used to lookup the array
+value for a specific CPU. Returns value on success , or ``NULL`` if no entry was
+found or ``cpu`` is invalid.
+
+Concurrency
+-----------
+
+Since kernel version 5.1, the BPF infrastructure provides ``struct bpf_spin_lock``
+to synchronize access.
+
+Examples
+========
+
+Please see the ``tools/testing/selftests/bpf`` directory for functional
+examples. The code samples below demonstrate API usage.
+
+Kernel BPF
+----------
+
+This snippet shows how to declare an array in a BPF program.
+
+.. code-block:: c
+
+    struct {
+            __uint(type, BPF_MAP_TYPE_ARRAY);
+            __type(key, u32);
+            __type(value, long);
+            __uint(max_entries, 256);
+    } my_map SEC(".maps");
+
+
+This example BPF program shows how to access an array element.
+
+.. code-block:: c
+
+    int bpf_prog(struct __sk_buff *skb)
+    {
+            int index = load_byte(skb,
+                                  ETH_HLEN + offsetof(struct iphdr, protocol));
+            long *value;
+
+            if (skb->pkt_type != PACKET_OUTGOING)
+                    return 0;
+
+            value = bpf_map_lookup_elem(&my_map, &index);
+            if (value)
+                    __sync_fetch_and_add(value, skb->len);
+
+            return 0;
+    }
+
+Userspace
+---------
+
+BPF_MAP_TYPE_ARRAY
+~~~~~~~~~~~~~~~~~~
+
+This snippet shows how to create an array, using ``bpf_map_create_opts`` to
+set flags.
+
+.. code-block:: c
+
+    #include <bpf/libbpf.h>
+    #include <bpf/bpf.h>
+
+    int create_array() {
+            int fd;
+            LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
+            fd = bpf_map_create(BPF_MAP_TYPE_ARRAY,
+                                "example_array",       /* name */
+                                sizeof(__u32),         /* key size */
+                                sizeof(long),          /* value size */
+                                256,                   /* max entries */
+                                &opts);                /* create opts */
+            return fd;
+    }
+
+This snippet shows how to initialize the elements of an array.
+
+.. code-block:: c
+
+    int initialize_array(int fd) {
+            __u32 i;
+            long value;
+            int ret;
+
+            for (i = 0; i < 256; i++) {
+                    value = i;
+                    ret = bpf_map_update_elem(fd, &i, &value, BPF_ANY);
+                    if (ret < 0)
+                            return ret;
+            }
+
+            return ret;
+    }
+
+This snippet shows how to retrieve an element value from an array.
+
+.. code-block:: c
+
+    int lookup(int fd) {
+            __u32 index = 42;
+            long value;
+            int ret = bpf_map_lookup_elem(fd, &index, &value);
+            if (ret < 0)
+                    return ret;
+
+            /* use value here */
+            assert(value == 42);
+
+            return ret;
+    }
+
+BPF_MAP_TYPE_PERCPU_ARRAY
+~~~~~~~~~~~~~~~~~~~~~~~~~
+
+This snippet shows how to initialize the elements of a per CPU array.
+
+.. code-block:: c
+
+    int initialize_array(int fd) {
+            int ncpus = libbpf_num_possible_cpus();
+            long values[ncpus];
+            __u32 i, j;
+            int ret;
+
+            for (i = 0; i < 256 ; i++) {
+                    for (j = 0; j < ncpus; j++)
+                            values[j] = i;
+                    ret = bpf_map_update_elem(fd, &i, &values, BPF_ANY);
+                    if (ret < 0)
+                            return ret;
+            }
+
+            return ret;
+    }
+
+This snippet shows how to access the per CPU elements of an array value.
+
+.. code-block:: c
+
+    int lookup(int fd) {
+            int ncpus = libbpf_num_possible_cpus();
+            __u32 index = 42, j;
+            long values[ncpus];
+            int ret = bpf_map_lookup_elem(fd, &index, &values);
+            if (ret < 0)
+                    return ret;
+
+            for (j = 0; j < ncpus; j++) {
+                    /* Use per CPU value here */
+                    assert(values[j] == 42);
+            }
+
+            return ret;
+    }
+
+Semantics
+=========
+
+As shown in the example above, when accessing a ``BPF_MAP_TYPE_PERCPU_ARRAY``
+in userspace, each value is an array with ``ncpus`` elements.
+
+When calling ``bpf_map_update_elem()`` the flag ``BPF_NOEXIST`` can not be used
+for these maps.