Don't we already have VFS tracepoints?

Posted Apr 19, 2025 5:13 UTC (Sat) by alison (subscriber, #63752)
Parent article: Tracepoints for the VFS?

What am I missing? Aren't these VFS syscall tracepoints?

$ sudo bpftrace -l | grep tracepoint:syscalls | grep read
tracepoint:syscalls:sys_enter_read
tracepoint:syscalls:sys_enter_readahead
tracepoint:syscalls:sys_enter_readlink
tracepoint:syscalls:sys_enter_readlinkat
tracepoint:syscalls:sys_enter_readv
. . .

$ uname -r
6.12.21-amd64

And there are kfuncs too.

Don't we already have VFS tracepoints?

Posted Apr 19, 2025 7:16 UTC (Sat) by kxxt (subscriber, #172895) [Link] (1 responses)

> What am I missing? Aren't these VFS syscall tracepoints?

These are syscall tracepoints, definitely not VFS tracepoints.

For example, if you make x32 or x86_32 syscalls on x86_64 linux, they won't hit those syscall tracepoints you mentioned. But they will hit the VFS tracepoints (in the future).

Don't we already have VFS tracepoints?

Posted Apr 19, 2025 18:23 UTC (Sat) by iabervon (subscriber, #722) [Link]

Presumably, "[T'so] thinks that could be worked around by keeping the tracepoints at the system-call level" means that they're in functions implementing VFS operations that each do the work that one syscall does, but it's easy to interpret as meaning that they'd be in the code that exposes that functionality as syscalls.