Re: [PATCH][RFC] make take_dentry_name_snapshot() lockless

[Al Viro <viro@zeniv.linux.org.uk>]

On Tue, Dec 24, 2024 at 11:44:55AM -0800, Linus Torvalds wrote:
> On Tue, 24 Dec 2024 at 11:18, Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> >                 if (f)
> >                         seq_file_path(m, f, " \t\n\\<");
> >                 else
> >                         seq_puts(m, "<none>");
> 
> Maybe seq_file_path() could do the "<none>" thing itself.
> 
> Right now it looks like d_path() just oopses with a NULL pointer
> dereference if you pass it a NULL path. Our printk() routines -
> including '%pd' - tend to do "(null)" for this case (and also handle
> ERR_PTR() cases) thanks to the 'check_pointer()' logic.

	Umm...  What about the escape character set?  If we have
seq_file_path() produce "<none>" for a NULL file reference, we'd
need to make sure that file called "<none>" won't get mixed with
that...

	OTOH, there will be leading / in all cases except for
pipes/sockets/anon files/anything that has ->d_dname() and
it looks like the output of all instances has either '/' or
':' in it...  Feels brittle, though.

	Anyway, I'm not at all sure that those <none> lines are
actually useful for anything - we get something like
UserFiles:    512
followed by 512 lines of form index:name or index:<none>, with
indices going from 0 to 511.

	We are dumping an io_uring private descriptor table - all
of it in one call of ->show().  And each line is prepended with
the slot number, so it's not obvious that dumping empty slots
makes sense.