Modify trace function from generic_file_read_iter() to ext4_file_read_iter()

[kernel-z]

The new Linux version (since form 4.10) uses ext4_file_read_iter(), And if the 'CONFIG_FS_DAX'
is not set ,then ext4_file_read_iter() will call generic_file_read_iter(), else it will call
ext4_dax_read_iter(), and trace generic_file_read_iter() will fail.

[pchaigno]

Instead of a new command line flag, couldn't we try attaching to ext4_file_read_iter, and if it fails, fall-back to the old behavior? Or even, simply check that ext4_file_read_iter exists instead of trying to attach...

[kernel-z]

There are too many scripts, and will be more.
I think one script for one function is better.
If two scripts for one function, when the kernel evolving ,it may be need another one. At last it will be a lots of scripts for the same function.

[yonghong-song]

agree with @pchaigno A new option is unnecessary. bcc tools have numerous example for such detecting-and-attach. For example, in tcplife.py, we have

if (BPF.tracepoint_exists("sock", "inet_sock_set_state")):
    bpf_text += bpf_text_tracepoint
else:
    bpf_text += bpf_text_kprobe

In zfsdist.py, we have

if BPF.get_kprobe_functions(b'zpl_iter'):
    b.attach_kprobe(event="zpl_iter_read", fn_name="trace_entry")
    b.attach_kprobe(event="zpl_iter_write", fn_name="trace_entry")
elif BPF.get_kprobe_functions(b'zpl_aio'):
    b.attach_kprobe(event="zpl_aio_read", fn_name="trace_entry")
    b.attach_kprobe(event="zpl_aio_write", fn_name="trace_entry")
else:
    b.attach_kprobe(event="zpl_read", fn_name="trace_entry")
    b.attach_kprobe(event="zpl_write", fn_name="trace_entry")
b.attach_kprobe(event="zpl_open", fn_name="trace_entry")
b.attach_kprobe(event="zpl_fsync", fn_name="trace_entry")
if BPF.get_kprobe_functions(b'zpl_iter'):
    b.attach_kretprobe(event="zpl_iter_read", fn_name="trace_read_return")
    b.attach_kretprobe(event="zpl_iter_write", fn_name="trace_write_return")
elif BPF.get_kprobe_functions(b'zpl_aio'):
    b.attach_kretprobe(event="zpl_aio_read", fn_name="trace_read_return")
    b.attach_kretprobe(event="zpl_aio_write", fn_name="trace_write_return")
else:
    b.attach_kretprobe(event="zpl_read", fn_name="trace_read_return")
    b.attach_kretprobe(event="zpl_write", fn_name="trace_write_return")
b.attach_kretprobe(event="zpl_open", fn_name="trace_open_return")

Similar mechanism can be applied here.

[yonghong-song]

[buildbot, ok to test]

[kernel-z]

Thank you ,@yonghong-song,@pchaigno, I have changed the code as zfsdist.py type.

[yonghong-song]

LGTM. Thanks!