Al Viro cdc59a62bc kernel/acct.c: saner struct file treatment ...

Instead of switching ->f_path.mnt of an opened file to internal
clone, get a struct path with ->mnt set to internal clone of that
->f_path.mnt, then dentry_open() that to get the file with right ->f_path.mnt
from the very beginning.

	The only subtle part here is that on failure exits we need to
close the file with __fput_sync() and make sure we do that *before*
dropping the original mount.

	With that done, only fs/{file_table,open,namei}.c ever store
anything to file->f_path and only prior to file->f_mode & FMODE_OPENED
becoming true.  Analysis of mount write count handling also becomes
less brittle and convoluted...

[AV: folded a fix for a bug spotted by Jan Kara - we do need a full-blown
open of the original file, not just user_path_at() or we end up skipping
permission checks]

Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

2025-09-27 20:13:56 -04:00

17 KiB

Raw Blame History

View Raw