linux-kernel - Re: [GIT PULL] Ceph fixes for 5.1-rc7

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: <20190425200941.GW2217@ZenIV.linux.org.uk>
Date:   Thu, 25 Apr 2019 21:09:42 +0100
From:   Al Viro <viro@...iv.linux.org.uk>
To:     Jeff Layton <jlayton@...nel.org>
Cc:     Linus Torvalds <torvalds@...ux-foundation.org>,
        Ilya Dryomov <idryomov@...il.com>, ceph-devel@...r.kernel.org,
        Linux List Kernel Mailing <linux-kernel@...r.kernel.org>
Subject: Re: [GIT PULL] Ceph fixes for 5.1-rc7

On Thu, Apr 25, 2019 at 02:23:59PM -0400, Jeff Layton wrote:

> I took a quick look at the dcache code to see if we had something like
> that before I did this, but I guess I didn't look closely enough. Those
> routines do look nicer than my hand-rolled version.
> 
> I'll look at spinning up a patch to switch that over in the near future.

Jeff asked to put the name length in there; looking at the existing users,
it does make sense.  I propose turning struct name_snapshot into

struct name_snapshot {
        struct qstr name;
        unsigned char inline_name[DNAME_INLINE_LEN];
};

with
void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
{
        spin_lock(&dentry->d_lock);
	name->name = dentry->d_name;
        if (unlikely(dname_external(dentry))) {
                struct external_name *p = external_name(dentry);
                atomic_inc(&p->u.count);
        } else {
                memcpy(name->inline_name, dentry->d_iname,
                       dentry->d_name.len + 1);
                name->name.name = name->inline_name;
        }
	spin_unlock(&dentry->d_lock);
}

and callers adjusted, initially simply by turning snapshot.name into
snapshot.name.name.  Next step: overlayfs call site becoming
        take_dentry_name_snapshot(&name, real);
        this = lookup_one_len(name.name.name, connected, name.name.len);
Next: snotify stuff switched to passing struct qstr * - fsnotify_move()
first, then fsnotify().  That one would
	* leave callers passing NULL alone
	* have the callers passing snapshot.name.name pass &snapshot.name
	* fsnotify_dirent() pass the entire &dentry->d_name, not just
dentry->d_name.name (that's dependent upon parent being held exclusive;
devpts plays fast and loose here, relying upon the lack of any kind of
renames, explicit or implicit, in that fs)
	* ditto for remaining call in fsnotify_move() (both parents
are locked in all callers, thankfully)
	* ditto for fsnotify_link()
	* kernfs callers in kernfs_notify_workfn() would grow strlen().
Next: send_to_group() and ->handle_event() instances switched to passing
struct qstr *.
Next: inotify_handle_event() doesn't need to bother with strlen().
Next: audit_update_watch() and audit_compare_dname_path() switched to
struct qstr *.  Callers in __audit_inode_child() pass the entire
&dentry->d_name.  strlen() inside audit_compare_dname_path() goes away.

Objections?