JFFS2 OOPS in linus tree

[Joakim Tjernlund]

> -----Original Message-----
> From: linux-mtd-bounces at lists.infradead.org 
> [mailto:linux-mtd-bounces at lists.infradead.org] On Behalf Of 
> Joakim Tjernlund
> Sent: den 19 juni 2007 15:00
> To: Linux MTD mailing list
> Subject: Re: JFFS2 OOPS in linus tree
> 
> On Thu, 2007-06-14 at 13:15 +0200, Joakim Tjernlund wrote:
> > Two times I have gotten this OOPS at boot:
> > 
> > argh. node added in wrong place
> > Unable to handle kernel paging request for data at address 
> 0xffffffee
> > Faulting instruction address: 0xc00d5244
> > Oops: Kernel access of bad area, sig: 11 [#1]
> > TMCUTU
> > NIP: c00d5244 LR: c00dcd34 CTR: c014145c
> > REGS: cfd23d00 TRAP: 0300   Not tainted  (2.6.22-rc4)
> > MSR: 00009032 <EE,ME,IR,DR>  CR: 84000428  XER: 20000000
> > DAR: ffffffee, DSISR: 20000000
> > TASK = cff6a0f0[201] 'getty' THREAD: cfd22000
> > GPR00: c00dcd34 cfd23db0 cff6a0f0 cff2d000 ffffffea 
> 02c885c4 c060e524 4671100e 
> > GPR08: 00000000 00000000 00000000 46713eb7 00000000 
> 10078b2c 0fffd100 007fff87 
> > GPR16: 00000001 ffffffff 00000000 0fff79d8 00000000 
> c0250000 cfd23e08 00000000 
> > GPR24: cff2d000 00000002 cfd23ee0 cff8ea38 cff62aac 
> ffffffea cff2d000 c024b4b4 
> > NIP [c00d5244] jffs2_mark_node_obsolete+0x1c/0x508
> > LR [c00dcd34] jffs2_do_setattr+0x29c/0x654
> > Call Trace:
> > [cfd23db0] [c07ec000] 0xc07ec000 (unreliable)
> > [cfd23e00] [c00dcd34] jffs2_do_setattr+0x29c/0x654
> > [cfd23e40] [c00dd100] jffs2_setattr+0x14/0x24
> > [cfd23e50] [c0073818] notify_change+0x28c/0x2c0
> > [cfd23e80] [c005ac70] sys_fchmodat+0xc0/0xdc
> > [cfd23f40] [c000f148] ret_from_syscall+0x0/0x38
> > --- Exception: c01 at 0xff5e428
> >     LR = 0x1003f584
> > Instruction dump:
> > 38210020 7c0803a6 4e800020 4bf2ede1 4bffffe8 9421ffb0 
> 7c0802a6 bf61003c 
> > 7c9d2379 7c7e1b78 90010054 41820450 <809d0004> 548007be 
> 2f800001 419e0364 
> > Node totlen on flash (0x00000006) != totlen from node ref 
> (0x00000090)
> > 
> > I can't reproduce it, but I figured I should report it.
> 
> Keep getting these oops on 2.6.22-{rc4,rc5}. Once it stayed and didn't
> go away until I booted a 2.6.20 kernel and swapped back afterwards to
> 2.6.22-rcX.
> 
> I get these traces on boot too:
> jffs2_scan_eraseblock():
> Node at 0x02c87e44 {0x1985, 0xe002, 0x00000006) has invalid 
> CRC 0x4a400110 (calculated 0x3c3d0282)
> ...
> jffs2_scan_eraseblock():
> Node at 0x06997cac {0x1985, 0xc002, 0x00000006) has invalid 
> CRC 0x4a400110 (calculated 0x3c3d0282)
> ...
> jffs2_scan_eraseblock():
> Node at 0x073617c4 {0x1985, 0xe002, 0x00000006) has invalid 
> CRC 0x08410102 (calculated 0x3c3d0282)
> 
> To repeat, boot a JFFS2 root fs enabled board, do reboot -f
> Repeat few times.
> 
> Don't have time to track this down ATM as I have to find a 
> few other bugs in our 2.6.20
> kernel. I do think JFFS2 is broken in 2.6.22 though
> 
>  Jocke
> 

The reason for the OOPS is that jffs2_add_physical_node_ref()
can return an error ptr, return ERR_PTR(-EINVAL), and several
callers never check for an error ptr, instead they assume
that the return value is always good. What to do at
these call sites in case of error is another matter,
I sure don't know :(

 Jocke