Re: [PATCH] vfs: Add AT_EMPTY_PATH_NOCHECK as unchecked AT_EMPTY_PATH

On Sun, Jun 23, 2024 at 2:59 AM Xi Ruoyao [off-list ref] wrote:

On Sat, 2024-06-22 at 15:41 -0700, Linus Torvalds wrote:

quoted

I do think that we should make AT_EMPTY_PATH with a NULL path
"JustWork(tm)", because the stupid "look if the pathname is empty" is
horrible.

But moving that check into getname() is *NOT* the right answer,
because by the time you get to getname(), you have already lost.

Oops.  I'll try to get around of getname() too...

quoted

So the short-cut in vfs_fstatat() to never get a pathname is
disgusting - people should have used 'fstat()' - but it's _important_
disgusting.

The problem is we don't have fstat() for LoongArch, and it'll be
unusable on all 32-bit arch after 2037.

And Arnd hates the idea adding fstat() for LoongArch because there would
be one more 32-bit arch broken in 2037.

Or should we just add something like "fstat_2037()"?

In that case fstat is out of the question, but no problem.

It was suggested to make AT_EMPTY_PATH + NULL pathname do the right
thing and have the syscalls short-circuit as needed.

for statx it would look like this (except you are going to have
implement do_statx_by_fd):

diff --git a/fs/stat.c b/fs/stat.c
index 16aa1f5ceec4..0afe72b320cc 100644
--- a/fs/stat.c
+++ b/fs/stat.c

@@ -710,6 +710,9 @@ SYSCALL_DEFINE5(statx,
        int ret;
        struct filename *name;

+       if (flags == AT_EMPTY_PATH && filename == NULL)
+               return do_statx_by_fd(...);
+
        name = getname_flags(filename, getname_statx_lookup_flags(flags));
        ret = do_statx(dfd, name, flags, mask, buffer);
        putname(name);

and so on

Personally I would prefer if fstatx was added instead, fwiw most
massaging in the area will be the same regardless.

-- 
Mateusz Guzik <mjguzik gmail.com>