Re: WARNING in vfs_getxattr_alloc

[Christian Brauner]

On Mon, Mar 04, 2024 at 09:11:23PM +0900, Dominique Martinet wrote:
> Christian Brauner wrote on Mon, Mar 04, 2024 at 12:50:12PM +0100:
> > > kernel: lastest linux 6.7.rc8 90d35da658da8cff0d4ecbb5113f5fac9d00eb72
> > > kernel config: https://syzkaller.appspot.com/text?tag=KernelConfig&x=4a65fa9f077ead01
> > > with KASAN enabled
> > > compiler: gcc (GCC) 12.2.0
> > > 
> > > TITLE: WARNING in vfs_getxattr_alloc------------[ cut here ]------------
> > 
> > Very likely a bug in 9p. Report it on that mailing list. It seems that
> > p9_client_xattrwalk() returns questionable values for attr_size:
> > 748310584784038656
> > That's obviously a rather problematic allocation request.
> 
> That's whatever the server requested -- in 9p we don't have the data at
> allocation time (xattrwalk returns the size, then we "read" it out in a
> subsequent request), so we cannot double-check that the size makes sense
> based on a payload at this point.
> 
> We could obviously add a max (the current max of SSIZE_MAX is "a bit"
> too generous), but I honestly have no idea what'd make sense for this
> without breaking some weird usecase somewhere (given the content is
> "read" we're not limited by the size of a single message; I've seen
> someone return large content as synthetic xattrs so it's hard to put an
> actual number for me).
> If the linux VFS has a max hard-wired somewhere plase tell me and I'll
> be glad to change the max.

Surprisingly we have a max limit that exists in a way because the whole
xattr uapi is somewhat broken. So best to limit it at XATTR_SIZE_MAX.
See fs/xattr.c for how it's used.