Jeremiah Mahler <jmmahler@xxxxxxxxx> writes: > all, > > On Thu, Dec 04, 2014 at 03:34:11PM -0800, Jeremiah Mahler wrote: >> all, >> >> On Thu, Dec 04, 2014 at 07:59:28PM +1100, Stephen Rothwell wrote: >> > Hi all, >> > >> > Changes since 20141203: >> > > [...] >> > >> > -- >> > Cheers, >> > Stephen Rothwell sfr@xxxxxxxxxxxxxxxx >> > >> [...] >> >> linux-next 20141204 is broken on all my machines, 20141203 worked fine. >> >> They fail very early in the boot process and are unable to mount root. >> There are lots of "ext4_check_descriptors: Checksum for group ... >> failed" messages. >> >> Any suggestions for where to start looking? >> >> -- >> - Jeremiah Mahler > > I ran a bisect and found this patch to be faulty. > > commit fdfe073987619ec375da8d8a2701ab271d1b1339 > Author: Dmitry Monakhov <dmonakhov@xxxxxxxxxx> > Date: Tue Dec 2 16:09:16 2014 -0500 > > ext4: fix potential use after free during resize > > We need some sort of synchronization while updating ->s_group_desc > because there are a lot of users which can access old ->s_group_desc > array after it was released. > > Signed-off-by: Dmitry Monakhov <dmonakhov@xxxxxxxxxx> > Signed-off-by: Theodore Ts'o <tytso@xxxxxxx> + rcu_read_lock(); + gd_bh = *rcu_dereference(sbi->s_group_desc) + group_desc; + rcu_read_unlock(); Yes this is true crap. The only question is how it passed xfstess? > > -- > - Jeremiah Mahler
Attachment:
signature.asc
Description: PGP signature