unknown type name 'xfs_ino_t'

[Rtoax]

What reproduces the bug?

when i execute the following bpftrace scription:

#!/usr/bin/env bpftrace

/**
 *	fs/xfs/xfs_linux.h 
 *
 *	See https://github.com/iovisor/bpftrace/issues/2094
 */
//typedef unsigned long long	xfs_ino_t;

BEGIN
{
	printf("Tracing XFS calls... Hit Ctrl-C to end.\n");
}

tracepoint:xfs:xfs_setfilesize
{
	$offset = args->offset;
	$size = args->size;
	printf("%-8d %-16s %-8d %-8d \n", pid, comm, $offset, $size);
}

END
{
	printf("Bye.\n");
}

get

$ sudo ./xfs_setfilesize-tp.bt
definitions.h:13:3: error: unknown type name 'xfs_ino_t'
ERROR: Include headers with missing type definitions or install BTF information to your system.

bpftrace --info

$ sudo bpftrace --info
System
  OS: Linux 5.14.0 #7 SMP Thu Nov 11 15:15:29 CST 2021
  Arch: x86_64

Build
  version: v0.14.0-13-gd431-dirty
  LLVM: 12.0.1
  ORC: v2
  foreach_sym: yes
  unsafe uprobe: no
  bfd: yes
  bpf_attach_kfunc: yes
  bcc_usdt_addsem: yes
  bcc bpf_attach_uprobe refcount: yes
  bcc library path resolution: yes
  libbpf: yes
  libbpf btf dump: yes
  libbpf btf dump type decl: yes
  libdw (DWARF support): yes

Kernel helpers
  probe_read: yes
  probe_read_str: yes
  probe_read_user: yes
  probe_read_user_str: yes
  probe_read_kernel: yes
  probe_read_kernel_str: yes
  get_current_cgroup_id: yes
  send_signal: yes
  override_return: no
  get_boot_ns: yes
  dpath: yes

Kernel features
  Instruction limit: 1000000
  Loop support: yes
  btf (depends on Build:libbpf): yes
  map batch (depends on Build:libbpf): yes
  uprobe refcount (depends on Build:bcc bpf_attach_uprobe refcount): yes

Map types
  hash: yes
  percpu hash: yes
  array: yes
  percpu array: yes
  stack_trace: yes
  perf_event_array: yes

Probe types
  kprobe: yes
  tracepoint: yes
  perf_event: yes
  kfunc: yes
  iter:task: yes
  iter:task_file: yes

[fbs]

Did you try #include <fs/xfs/xfs_linux.h> or what ever the correct import is?

[viktormalik]

fs/xfs/xfs_linux.h is an internal kernel header, so bpftrace won't find it. A possible way would be to install the XFS development package (xfsprogs-devel on Fedora) and then include xfs/linux.h which contains a user-space definition of xfs_ino_t.

Unfortunately, that won't currently work either as the file includes stdbool.h which we don't have, yet (that should be easy to fix, though).

[StingerXIII]

I used to include a custom header file with typedef uint64_t xfs_ino_t; as a crutch. At least this allows you to use the tracepoint.

[viktormalik]

I investigated this issue a bit more. Since xfs_ino_t is defined in an internal header, it cannot be included from a bpftrace script.

On the other hand, it should be possible to pull the type definition from BTF. The problem is that xfs is usually loaded as a kernel module and BTF for modules is not supported for tracepoints at the moment. I tried to implement it in #2479 so it would be great if you could give it a try.

I tried to run the above script but it didn't work since the tracepoint doesn't seem to have the size parameter:

# bpftrace -lv 'tracepoint:xfs:xfs_setfilesize'
tracepoint:xfs:xfs_setfilesize
    dev_t dev
    xfs_ino_t ino
    loff_t isize
    loff_t disize
    loff_t offset
    size_t count

When using count instead of size, I was able to run the script.

Btw. the script would also work if you replaced the tracepoint by kfunc:xfs_setfilesize.

[viktormalik]

Fixed by #2479.