Make bpftrace work

[Officeyutong]

Did lots of things, which make bpftrace work under bpftime.

At least the following trace will work:

• bpftrace -e 'tracepoint:syscalls:sys_enter_clock_nanosleep { printf("PID %d sleeping...\n", pid); }'
• bpftrace -e 'tracepoint:raw_syscalls:sys_enter { @[comm] = count(); }'

Updates

• per cpu array/hash map
• global syscall tracepoints (hook the enter and exit of all syscalls, tracepoint/raw_syscalls/sys_{enter,exit})
• Splitted map lookup/update/delete for bpf helper and syscall. per cpu maps behaves differently when being examined from kernel and userspace
• helper ringbuf_output
• atomic operations support for ubpf and llvmjit. Newer bpftrace used this

Usage

a. Console1:

root@mnfe-pve:~/bpftime# bpftime load -- /root/bpftrace/build/src/bpftrace -e 'tracepoint:raw_syscalls:sys_enter { @[comm] = count(); }'
[2023-10-14 23:36:25.519] [info] manager constructed
[2023-10-14 23:36:25.608] [info] Initialize syscall server
[2023-10-14 23:36:25][info][1763968] Global shm constructed. global_shm_open_type 0 for bpftime_maps_shm
[2023-10-14 23:36:25][info][1763968] Enabling helper groups ffi, kernel, shm_map by default
[2023-10-14 23:36:25][info][1763968] bpftime-syscall-server started
Attaching 1 probe...

b.Console2 (or more):

bpftime start -s /bin/ls
bpftime start -s /bin/pwd
bpftime start -s /bin/whoami

c. Press Ctrl + C in console1

Attaching 1 probe...
^C

@[pwd]: 5
@[ls]: 19
@[whoami]: 24
INFO: Global shm destructed