Bug 30872

On Thu, Mar 10, 2011 at 10:21 PM,  <bugzilla-daemon@bugzilla.kernel.org> wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=30872
>
>           Summary: Calling kfree() for uninitialized pointer in
>                    ext4_mb_init_backend()
>           Product: File System
>           Version: 2.5
>    Kernel Version: 2.6.38-rc5 (ext4 subsystem tree)
>          Platform: All
>        OS/Version: Linux
>              Tree: Mainline
>            Status: NEW
>          Severity: normal
>          Priority: P1
>         Component: ext4
>        AssignedTo: fs_ext4@kernel-bugs.osdl.org
>        ReportedBy: dame_eugene@mail.ru
>        Regression: No
>
>
> Tested on ext4 module from git tree
> git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4.git,
> tip: b616844310a6c8a4ab405d3436bbb6e53cfd852f
>
> Arch: x86
>
> At fs/ext4/mballoc.c:2389, memory is allocated for sbi->s_group_info array.
> The
> elements of this array (pointers themselves) seem to be initialized when
> ext4_mb_add_groupinfo() is called (line 2408).
>
> If ext4_mb_add_groupinfo() fails for some reason (e.g. if memory allocation
> at
> line 2296 fails), ext4_mb_init_backend() tries to call kfree() for each
> element in sbi->s_group_info array, including the ones that have not been
> initialized yet:
>
> fs/ext4/mballoc.c:2414:
> err_freebuddy:
>    cachep = get_groupinfo_cache(sb->s_blocksize_bits);
>    while (i-- > 0)
>        kmem_cache_free(cachep, ext4_get_group_info(sb, i));
>    i = num_meta_group_infos;
>    while (i-- > 0)

literally understand, should be
while (--i >= 0)

Could you try with above?


>        kfree(sbi->s_group_info[i]); /* <= oops here */
>    iput(sbi->s_buddy_cache);
>
> 'num_meta_group_infos' seems to be the total number of the elements that
> should
> have been created.
>
> The problem showed up when I ran tests for ext4 from Linux Test Project
> (ext4-alloc-test, test #7, to be exact).
>
> 'num_meta_group_infos' was 12 on my system. The first 2 calls to
> ext4_mb_add_groupinfo() (ln 2408) succeeded but the 3rd one failed.
> kfree(sbi->s_group_info[11]) resulted in a kernel oops:
>
> --------------------------------------------------
> [ 6349.953315] EXT4-fs: can't allocate buddy mem
> [ 6349.953587] BUG: unable to handle kernel paging request at f7853a00
> [ 6349.953591] IP: [<c02f1273>] kfree+0x43/0xf0
> [ 6349.953613] *pde = 00000000
> [ 6349.953615] Oops: 0000 [#1] SMP
> [ 6349.953617] last sysfs file:
> /sys/devices/virtual/block/loop1/queue/rotational
> [ 6349.953623] Modules linked in: ext4 jbd2 crc16 kedr_controller
> kedr_fsim_indicator_kmalloc kedr_fsim_indicator_common kedr_fsim_vmm
> kedr_fsim_mem_util kedr_fsim_cmm kedr_fault_simulation kedr_trace kedr_base
> fuse snd_pcm_oss snd_mixer_oss snd_seq snd_seq_device edd af_packet mperf
> loop
> dm_mod ppdev snd_intel8x0 snd_ac97_codec ac97_bus snd_pcm parport_pc
> snd_timer
> parport snd sr_mod cdrom ac e1000 sg i2c_piix4 soundcore snd_page_alloc
> button
> pcspkr ohci_hcd ehci_hcd rtc_cmos rtc_core rtc_lib usbcore sd_mod fan
> processor
> ata_generic ata_piix thermal thermal_sys hwmon ahci libahci libata scsi_mod
> [last unloaded: kedr_base]
> [ 6349.953655]
> [ 6349.953657] Pid: 8379, comm: mount Not tainted 2.6.38-rc5-testbox-ext4+ #1
> innotek GmbH VirtualBox
> [ 6349.953664] EIP: 0060:[<c02f1273>] EFLAGS: 00010086 CPU: 0
> [ 6349.953666] EIP is at kfree+0x43/0xf0
> [ 6349.953667] EAX: f7853a00 EBX: 02b50001 ECX: 00000000 EDX: 000000bb
> [ 6349.953669] ESI: f8a15391 EDI: 00000202 EBP: e7e03d70 ESP: e7e03d60
> [ 6349.953670]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
> [ 6349.953672] Process mount (pid: 8379, ti=e7e02000 task=f66bed30
> task.ti=e7e02000)
> [ 6349.953674] Stack:
> [ 6349.953675]  eea9cae8 02b50001 f446f400 f446f400 e7e03d8c f8a15391
> f8f01609
> 00000002
> [ 6349.953679]  00032609 02b50001 0000000a e7e03dbc f8f01609 00020000
> 00000000
> f47ce1c0
> [ 6349.953683]  f446f400 f446f400 0000000c f446f400 f446f400 e4866bf0
> f446f400
> e7e03e98
> [ 6349.953687] Call Trace:
> [ 6349.953691]  [<f8a15391>] repl_kfree+0x51/0xa0 [kedr_fsim_cmm]
> [ 6349.953709]  [<f8f01609>] ext4_mb_init+0x2a9/0x4b0 [ext4]
> [ 6349.953716]  [<f8ef1b22>] ext4_fill_super+0x2602/0x2ae0 [ext4]
> [ 6349.953726]  [<c02fd920>] mount_bdev+0x170/0x1b0
> [ 6349.953732]  [<f8ee706a>] ext4_mount+0x1a/0x20 [ext4]
> [ 6349.953741]  [<c02fd0f0>] vfs_kern_mount+0x70/0x230
> [ 6349.953753]  [<c02fd309>] do_kern_mount+0x39/0xd0
> [ 6349.953755]  [<c0314d72>] do_mount+0x432/0x6c0
> [ 6349.953768]  [<c0315326>] sys_mount+0x66/0xa0
> [ 6349.953771]  [<c0202ddc>] sysenter_do_call+0x12/0x28
> [ 6349.953772] Code: 7d fc 85 c9 75 73 83 fb 10 76 61 9c 58 8d 74 26 00 89 c7
> fa 90 8d 74 26 00 8d 83 00 00 00 40 c1 e8 0c c1 e0 05 03 05 64 ff 86 c0 <8b>
> 10
> 80 e6 80 0f 85 8b 00 00 00 8b 10 80 e6 80 75 7c 8b 10 81
> [ 6349.953796] EIP: [<c02f1273>] kfree+0x43/0xf0 SS:ESP 0068:e7e03d60
> [ 6349.953799] CR2: 00000000f7853a00
> [ 6349.953801] ---[ end trace 8e09ff66f4f48163 ]---
> --------------------------------------------------
>
> If I understand correctly what happens in ext4_mb_init_backend(), zeroing
> sbi->s_group_info on allocation could fix the problem:
> fs/ext4/mballoc.c:2389:
> - sbi->s_group_info = kmalloc(array_size, GFP_KERNEL);
> + sbi->s_group_info = kzalloc(array_size, GFP_KERNEL);
>
> This issue was detected with the help of KEDR framework
> (http://kedr.berlios.de/). repl_kfree() that can be seen in the call stack
> above is a thin wrapper around kfree() that simply output its argument to a
> trace.
>
> --
> Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are watching the assignee of the bug.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

(In reply to comment #1)

...

> > If ext4_mb_add_groupinfo() fails for some reason (e.g. if memory allocation
> at
> > line 2296 fails), ext4_mb_init_backend() tries to call kfree() for each
> > element in sbi->s_group_info array, including the ones that have not been
> > initialized yet:
> >
> > fs/ext4/mballoc.c:2414:
> > err_freebuddy:
> >    cachep = get_groupinfo_cache(sb->s_blocksize_bits);
> >    while (i-- > 0)
> >        kmem_cache_free(cachep, ext4_get_group_info(sb, i));
> >    i = num_meta_group_infos;
> >    while (i-- > 0)
> 
> literally understand, should be
> while (--i >= 0)
> 
> Could you try with above?

Hm, that probably is the most direct fix.

I sent a patch here:
http://marc.info/?l=linux-ext4&m=129979820301087&w=2
which initializes all the pointers to NULL and only frees if they are non-NULL, 
which may be more foolproof, but might also be overkill.

-Eric

(In reply to comment #2)
> I sent a patch here:
> http://marc.info/?l=linux-ext4&m=129979820301087&w=2
> which initializes all the pointers to NULL and only frees if they are
> non-NULL, 
> which may be more foolproof, but might also be overkill.

As far as I can see, it is not overkill. 

On the system where I observed the problem, 'i' began from 12 ('num_meta_group_infos' was 12, the total number of groups to be initialized). But only the first two elements of the sbi->s_group_info array were actually initialized. So without setting the remaining pointers to NULL, 
    while (--i >= 0)
        kfree(sbi->s_group_info[i]);
would still result in an oops when it attempted to kfree() sbi->s_group_info[11].

Apart from that, checking if the pointer is not NULL before kfree() could probably be omitted in the patch as kfree(NULL) is a no-op anyway. But still, it is a matter of coding style and personal preference.

A patch referencing this bug report has been merged in v2.6.38-8876-g036a982:

commit 4596fe07679ff0fae904515691ea747467614871
Author: Eric Sandeen <sandeen@redhat.com>
Date:   Mon Mar 21 21:25:13 2011 -0400

    ext4: don't kfree uninitialized s_group_info members