bpf on x64

perform a command on an extended BPF map or program

sys_bpf(int cmd, union bpf_attr *attr, unsigned int size, struct bpf_common_attr *attr_common, unsigned int size_common)
number321 (0x141)
abicommon
entry pointsys_bpf
sinceLinux 3.18
man pagebpf(2)
sourceelixir.bootlin.com

Registers

registervalue
rax321, the syscall number
rdiint cmd
rsiunion bpf_attr *attr
rdxunsigned int size
r10struct bpf_common_attr *attr_common
r8unsigned int size_common
raxthe return value

Then run syscall to enter the kernel.

The syscall instruction clobbers rcx (the return address lands in rcx) and r11 (the flags land in r11).

Since Linux 5.14 the kernel reads the syscall number from eax and ignores the upper 32 bits of rax. Older kernels fail with ENOSYS when those bits are not zero, and mov eax, N clears them, so it works on both.

Same syscall on other architectures

Get this as JSON: https://api.syscall.sh/v2/syscalls/x64/bpf