linux-ext4 - Re: generic_permission() optimization

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAGudoHEJZ32rDUt4+n2-L-RU=bpGgkYMroxtdMF6MQjKRsW24w@mail.gmail.com>
Date: Sun, 13 Apr 2025 11:41:47 +0200
From: Mateusz Guzik <mjguzik@...il.com>
To: "Theodore Ts'o" <tytso@....edu>
Cc: Linus Torvalds <torvalds@...ux-foundation.org>, Christian Brauner <brauner@...nel.org>, 
	Al Viro <viro@...iv.linux.org.uk>, linux-fsdevel <linux-fsdevel@...r.kernel.org>, 
	Jan Kara <jack@...e.cz>, Ext4 Developers List <linux-ext4@...r.kernel.org>
Subject: Re: generic_permission() optimization

On Sun, Apr 13, 2025 at 1:55 AM Theodore Ts'o <tytso@....edu> wrote:
>
> On Sat, Apr 12, 2025 at 03:36:00PM -0700, Linus Torvalds wrote:
> > Indeed. I sent a query to the ext4 list (and I think you) about
> > whether my test was even the right one.
>
> Sorry, I must have not seen that message; at least, I don't have any
> memory of it.
>
> > Also, while I did a "getfattr -dR" to see if there are any *existing*
> > attributes (and couldn't find any), I also assume that if a file has
> > ever *had* any attributes, the filesystem may have the attribute block
> > allocated even if it's now empty.
>
> Well, getfattr will only show user xattrs.  It won't show security.*
> xattr's that might have been set by SELinux, or a
> system.posix_acl_access xattr.
>
> > I assume there's some trivial e2fstools thing to show things like
> > that, but it needs more ext4 specific knowledge than I have.
>
> Yes, we can test for this using the debugfs command.  For exaple:
>
> root@...-xfstests:~# debugfs /dev/vdc
> debugfs 1.47.2-rc1 (28-Nov-2024)
> debugfs:  stat <13>
> Inode: 13   Type: regular    Mode:  0644   Flags: 0x80000
> Generation: 1672288850    Version: 0x00000000:00000003
> User:     0   Group:     0   Project:     0   Size: 286
> File ACL: 0
> Links: 1   Blockcount: 8
> Fragment:  Address: 0    Number: 0    Size: 0
>  ctime: 0x67faf5d0:30d0b2e4 -- Sat Apr 12 19:22:56 2025
>  atime: 0x67faf571:7064bd50 -- Sat Apr 12 19:21:21 2025
>  mtime: 0x67faf571:71236aa8 -- Sat Apr 12 19:21:21 2025
> crtime: 0x67faf571:7064bd50 -- Sat Apr 12 19:21:21 2025
> Size of extra inode fields: 32
> Extended attributes:
>   system.posix_acl_access (28) = 01 00 00 00 01 00 06 00 02 00 04 00 b7 7a 00 00 04 00 04 00 10 00 04 00 20 00 04 00
> Inode checksum: 0xc8f7f1a7
> EXTENTS:
> (0):33792
>
> (If you know the pathname instead of the inode number, you can also
> give that to debugfs's stat command, e.g., "stat /lost+found")
>
> I tested it with a simple variant of your patch, and seems to do the right
> thing.  Mateusz, if you want, try the following patch, and then mount
> your test file system with "mount -o debug".  (The test_opt is to
> avoid a huge amount of noise on your root file system; you can skip it
> if it's more trouble than it's worth.)  The patch has a reversed
> seense of the test, so it will print a message for every one where
> cache_no_acl *wouldn't* be called.  You casn then use debugfs's "stat
> <ino#>" to verify whether it has some kind of extended attribute.
>
>                                       - Ted
>
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index f386de8c12f6..3e0ba7c4723a 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -5109,6 +5109,11 @@ struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
>                 goto bad_inode;
>         brelse(iloc.bh);
>
> +       if (test_opt(sb, DEBUG) &&
> +           (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
> +            ei->i_file_acl))
> +               ext4_msg(sb, KERN_DEBUG, "has xattr ino %lu", inode->i_ino);
> +
>         unlock_new_inode(inode);
>         return inode;
>

This is the rootfs of the thing, so I tried it out with merely
printing it. I got 70 entries at boot time. I don't think figuring out
what this is specifically is warranted (it is on debian though).

With a little bit of cumbersome bpf tracing I verified newly created
dirs (as in mkdir) also have i_acl == NULL, which is the bit important
here (apart from dirs instantiated from storage which now also have
it).

So... I think this is good enough to commit? I had no part in writing
the patch and I'm not an ext4 person, so I'm not submitting it myself.

Ted, you seem fine with the patch, so perhaps you could do the needful(tm)?

The patch is valuable in its own right and is a soft blocker for my
own patch which adds IOP_FAST_MAY_EXEC to skip a bunch of crap work in
inode_permission() which for MAY_EXEC wont be needed [I have not
posted it yet].
-- 
Mateusz Guzik <mjguzik gmail.com>