On Thu, 28 Aug 2025 at 11:58, Arnaldo Carvalho de Melo
<arnaldo.melo@xxxxxxxxx> wrote:
> >
> >Give the damn thing an actual filename or something *useful*, not a
> >number that user space can't even necessarily match up to anything.
>
> A build ID?
I think that's a better thing than the disgusting inode number, yes.
That said, I think they are problematic too, in that I don't think
they are universally available, so if you want to trace some
executable without build ids - and there are good reasons to do that -
you might hate being limited that way.
So I think you'd be much better off with just actual pathnames.
Are there no trace events for "mmap this path"? Create a good u64 hash
from the contents of a 'struct path' (which is just two pointers: the
dentry and the mnt) when mmap'ing the file, and then you can just
associate the stack trace entry with that hash.
That should be simple and straightforward, and hashing two pointers
should be simple and straightforward.
And then matching that hash against the mmap event where the actual
path was saved off gives you an actual *pathname*. Which is *so* much
better than those horrific inode numbers.
And yes, yes, obviously filenames can go away and aren't some kind of
long-term stable thing. But inode numbers can be re-used too, so
that's no different.
With the "create a hash of 'struct path' contents" you basically have
an ID that can be associated with whatever the file name was at the
time it was mmap'ed into the thing you are tracing, which is I think
what you really want anyway.
Now, what would be even simpler is to not create a hash at all, but
simply just create the whole pathname when the stack trace entry is
created. But it would probably waste too much space, since you'd
probably want to have at least 32 bytes (as opposed to just 64 bits)
for a (truncated) pathname.
And it would be more expensive than just hashing the dentry/mnt
pointers, although '%pD' isn't actually *that* expensive. But probably
expensive enough to not really be acceptable. I'm just throwing it out
as a stupid idea that at least generates much more usable output than
the inode numbers do.
Linus
