[PATCH] Missing BKL in sys_chroot() for 2.6

From: BlaisorBlade
Date: Mon Jun 07 2004 - 12:47:49 EST


(PLEASE cc me on replies as I'm not subscribed).

Set_fs_root *claims* it wants the BKL held:

/*
* Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
* It can block. Requires the big lock held.
*/
void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
struct dentry *dentry)

But sys_chroot ignores this. So I attach this patch (apply with patch -p1 -l).
Maybe the solution is to kill that comment, maybe not (sys_pivot_root() calls
chroot_fs_refs() -> set_fs_root() with the BKL held, but that has a lot of
reasons other than this). If that comment is correct, however, it probably
holds even for set_fs_altroot(), since it's similar. So this patch adds it.


TESTING: none, I'm sure about the inconsistency, a lot less about the solution
(no kernel manual explains the BKL, since it is a relict; except the ones who
still says it blocks all the rest of the kernel, while I think it is just a
recursive and "sleepable" spinlock - am I correct?).

--- vanilla-linux-2.6.6/fs/open.c.saved 2004-05-10 21:37:11.000000000 +0200
+++ vanilla-linux-2.6.6/fs/open.c 2004-06-06 19:40:20.000000000 +0200
@@ -581,8 +581,13 @@
if (!capable(CAP_SYS_CHROOT))
goto dput_and_out;

+ lock_kernel();
+
set_fs_root(current->fs, nd.mnt, nd.dentry);
set_fs_altroot();
+
+ unlock_kernel();
+
error = 0;
dput_and_out:
path_release(&nd);
--- vanilla-linux-2.6.6/fs/namei.c.saved 2004-05-10 21:37:09.000000000
+0200
+++ vanilla-linux-2.6.6/fs/namei.c 2004-06-06 19:49:25.000000000 +0200
@@ -814,6 +814,9 @@
return 1;
}

+/*
+ * Requires the big lock held.
+ */
void set_fs_altroot(void)
{
char *emul = __emul_prefix();

--
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/