On Tue, Jul 27, 2021 at 8:05 PM G [off-list ref] wrote:

Hi BPF Experts

I'm having an issue with using "BPF_MAP_TYPE_DEVMAP" when pinning it to a global path for sharing with multiple programs.
I used iproute2 to load the program, the parameters for creating a map are as follows:
----------------------------------------------------------------------------------
    struct bpf_elf_map BPFM_SEC_MAPS dev_map_test = {
        .type                = BPF_MAP_TYPE_DEVMAP,
        .size_key       = sizeof(int32),
        .size_value   = sizeof(int32),
        .max_elem   = 1,
        .flags               = 0,
        .pinning         = PIN_GLOBAL_NS,
    };
----------------------------------------------------------------------------------

Here I set the flags to 0, but the actual map is created with flags 0x80, so when the second program is loaded, map parameter check is incorrect.
When I tried to set the flags directly to 0x80, I failed with the parameter error `-EINVA`  when the first program was loaded
The general process for iproute2 to create a map with a pin tag is as follows:
-----------------------------------------------------------------------------------
bpf_map_attach
-> bpf_probe_pinned
    bpf_map_selfcheck_pinned
    -> bpf_derive_elf_map_from_fdinfo
         bpf_map_pin_report
         -> if (obj->flags != pin->flags)
                   fprintf(stderr, " - Flags:        %#x (obj) != %#x (pin)\n", obj->flags, pin->flags);
-----------------------------------------------------------------------------------

I have tried to read the relevant kernel code :
-----------------------------------------------------------------------------------
    // kernel/bpf/devmap.c
    #define DEV_CREATE_FLAG_MASK \
         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)

    static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)
    {
       ...
        /* check sanity of attributes */
        if (attr->max_entries == 0 || attr->key_size != 4 ||
                attr->value_size != 4 || attr->map_flags & ~DEV_CREATE_FLAG_MASK)
                return -EINVAL;

        /* Lookup returns a pointer straight to dev->ifindex, so make sure the
         * verifier prevents writes from the BPF side
         */
        attr->map_flags |= BPF_F_RDONLY_PROG;
        ...
    }
-----------------------------------------------------------------------------------

At the time of check , it does not allow 0x80 to be set (DEV_CREATE_FLAG_MASK) , but then the code sets the 0x80 flag itself.

It looks like the kernel itself is setting the flags to 0x80 when it
initializes a devmap[1]. (Note: BPF_PROG_READONLY is 0x80.[2])

I believe you do not need to worry about setting this flag when you call the
BPF_MAP_CREATE syscall.

[1] https://elixir.bootlin.com/linux/v5.13.5/source/kernel/bpf/devmap.c#L126
[2] https://elixir.bootlin.com/linux/v5.13.5/source/include/uapi/linux/bpf.h#L1179

So I'm not sure if we should allow 0x80 on the check here as well ? If anyone has suggestions please let me know, thanks.
-----------------------------------------------------------------------------------
    #define DEV_CREATE_FLAG_MASK \
         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY | BPF_F_RDONLY_PROG)
-----------------------------------------------------------------------------------

Best regards
W.Gao