Function flocks_remove_flock changed from 2.6.12 to 2.6.13: IS_POSIX from "if (IS_FLOCK(fl) || IS_POSIX(fl))" has been deleted. Because of that, code runs into BUG() which is below two 'if' instructions. Reproducible: Always Steps to Reproduce: 1.emerge gentoo-sources-2.6.13 && emerge unionfs 2.build live cd with gentoo 3.kernel crashes with BUG() in flocks_remove_flock Patch for kernel: --- fs/locks.c 2005-09-02 16:38:40.770509784 +0200 +++ fs/locks-fixed.c 2005-09-02 16:37:22.000000000 +0200 @@ -1908,7 +1908,7 @@ while ((fl = *before) != NULL) { if (fl->fl_file == filp) { - if (IS_FLOCK(fl)) { + if (IS_FLOCK(fl) || IS_POSIX(fl)) { locks_delete_lock(before); continue; } Patch for unionfs to compile on 2.6.13: --- unionfs-1.0.13/inode.c 2005-07-22 19:16:48.000000000 +0200 +++ unionfs-1.0.13-0/inode.c 2005-09-02 01:57:04.850697936 +0200 @@ -884,7 +884,7 @@ return err; } -static int unionfs_follow_link(struct dentry *dentry, struct nameidata *nd) +static void* unionfs_follow_link(struct dentry *dentry, struct nameidata *nd) { char *buf; int len = PAGE_SIZE, err; @@ -922,11 +922,11 @@ KFREE(buf); #endif print_exit_status(err); - return err; + return ERR_PTR(err); } #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) -void unionfs_put_link(struct dentry *dentry, struct nameidata *nd) +void unionfs_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) { char *link; print_entry_location(); --- unionfs-1.0.13/Makefile 2005-07-22 19:16:48.000000000 +0200 +++ unionfs-1.0.13-0/Makefile 2005-09-02 02:07:13.145223112 +0200 @@ -62,7 +62,7 @@ ld -o ${obj-m} -r ${unionfs-objs} unionctl: unionctl.c usercommon.c - ${CC} -o $@ $^ ${UCFLAGS} + ${CC} -o $@ $^ ${UCFLAGS} -static uniondbg: uniondbg.c ${CC} -o $@ $^ ${UCFLAGS} --- unionfs-1.0.13/stale_inode.c 2005-07-22 19:16:48.000000000 +0200 +++ unionfs-1.0.13-0/stale_inode.c 2005-09-02 02:15:40.772052104 +0200 @@ -30,9 +30,10 @@ * so that a stale root inode can at least be unmounted. To do this * we must dput() the base and return the dentry with a dget(). */ -static int stale_follow_link(struct dentry *dent, struct nameidata *nd) +static void* stale_follow_link(struct dentry *dent, struct nameidata *nd) { - return vfs_follow_link(nd, ERR_PTR(-ESTALE)); + int err = vfs_follow_link(nd, ERR_PTR(-ESTALE)); + return ERR_PTR(err); } static int return_ESTALE(void)
Where has the kernel patch come from? Was it written by you? Has it been submitted upstream?
I wrote patch for kernel during work on Hackett Linux LiveCD (hackett.linux.pl). I sent it to linux-fsdevel@vger.kernel.org, but it appears that it was eaten by antispam or something. Patch for unionfs is in unionfs CVS.
Please discuss this patch with Peter Staubach <staubach@redhat.com> (CC Linux kernel mailing list) as by the look of his recent change, he specifically did _not_ check IS_POSIX(fl). See his patch and a long description: http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff_plain;h=c293621bbf678a3d85e3ed721c3921c8a670610d;hp=3e5ea098446e19175fdee4c2c4ec9366b0217db4
Please reopen this once a fix has been applied upstream
