On Fri, 9 Feb 2024 at 05:07, Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> From: Alexei Starovoitov <ast@xxxxxxxxxx>
>
> mmap() bpf_arena right after creation, since the kernel needs to
> remember the address returned from mmap. This is user_vm_start.
> LLVM will generate bpf_arena_cast_user() instructions where
> necessary and JIT will add upper 32-bit of user_vm_start
> to such pointers.
>
> Fix up bpf_map_mmap_sz() to compute mmap size as
> map->value_size * map->max_entries for arrays and
> PAGE_SIZE * map->max_entries for arena.
>
> Don't set BTF at arena creation time, since it doesn't support it.
>
> Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
> ---
> tools/lib/bpf/libbpf.c | 43 ++++++++++++++++++++++++++++++-----
> tools/lib/bpf/libbpf_probes.c | 7 ++++++
> 2 files changed, 44 insertions(+), 6 deletions(-)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index 01f407591a92..4880d623098d 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -185,6 +185,7 @@ static const char * const map_type_name[] = {
> [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter",
> [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf",
> [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage",
> + [BPF_MAP_TYPE_ARENA] = "arena",
> };
>
> static const char * const prog_type_name[] = {
> @@ -1577,7 +1578,7 @@ static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
> return map;
> }
>
> -static size_t bpf_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
> +static size_t __bpf_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
> {
> const long page_sz = sysconf(_SC_PAGE_SIZE);
> size_t map_sz;
> @@ -1587,6 +1588,20 @@ static size_t bpf_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
> return map_sz;
> }
>
> +static size_t bpf_map_mmap_sz(const struct bpf_map *map)
> +{
> + const long page_sz = sysconf(_SC_PAGE_SIZE);
> +
> + switch (map->def.type) {
> + case BPF_MAP_TYPE_ARRAY:
> + return __bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
> + case BPF_MAP_TYPE_ARENA:
> + return page_sz * map->def.max_entries;
> + default:
> + return 0; /* not supported */
> + }
> +}
> +
> static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz)
> {
> void *mmaped;
> @@ -1740,7 +1755,7 @@ bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
> pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
> map->name, map->sec_idx, map->sec_offset, def->map_flags);
>
> - mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
> + mmap_sz = bpf_map_mmap_sz(map);
> map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE,
> MAP_SHARED | MAP_ANONYMOUS, -1, 0);
> if (map->mmaped == MAP_FAILED) {
> @@ -4852,6 +4867,7 @@ static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, b
> case BPF_MAP_TYPE_SOCKHASH:
> case BPF_MAP_TYPE_QUEUE:
> case BPF_MAP_TYPE_STACK:
> + case BPF_MAP_TYPE_ARENA:
> create_attr.btf_fd = 0;
> create_attr.btf_key_type_id = 0;
> create_attr.btf_value_type_id = 0;
> @@ -4908,6 +4924,21 @@ static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, b
> if (map->fd == map_fd)
> return 0;
>
> + if (def->type == BPF_MAP_TYPE_ARENA) {
> + map->mmaped = mmap((void *)map->map_extra, bpf_map_mmap_sz(map),
> + PROT_READ | PROT_WRITE,
> + map->map_extra ? MAP_SHARED | MAP_FIXED : MAP_SHARED,
> + map_fd, 0);
> + if (map->mmaped == MAP_FAILED) {
> + err = -errno;
> + map->mmaped = NULL;
> + close(map_fd);
> + pr_warn("map '%s': failed to mmap bpf_arena: %d\n",
> + bpf_map__name(map), err);
> + return err;
> + }
> + }
> +
Would it be possible to introduce a public API accessor for getting
the value of map->mmaped?
Otherwise one would have to parse through /proc/self/maps in case
map_extra is 0.
The use case is to be able to use the arena as a backing store for
userspace malloc arenas, so that
we can pass through malloc/mallocx calls (or class specific operator
new) directly to malloc arena using the BPF arena.
In such a case a lot of the burden of converting existing data
structures or code can be avoided by making much of the process
transparent.
Userspace malloced objects can also be easily shared to BPF progs as a
pool through bpf_ma style per-CPU allocator.
> [...]
