Carefully crafted BTRFS-image causes kernel to crash

Carefully crafted BTRFS-image causes kernel to crash

[Lukas Lueg]

[-- Attachment #1: Type: text/plain, Size: 3165 bytes --]

See also https://bugzilla.kernel.org/show_bug.cgi?id=96971


I've identified some problems in the btrfs code and attached a
btrfs-image which causes the userland tools to crash and the kernel to
immediately freeze once the filesystem get's mounted and one of the
files is accessed. Putting the image onto a usb-drive gives you a
freeze-on-a-stick :-)

"btrfs check" crashes due to a SIGFPE in count_csum_range(). The
culprit is struct btrfs_root->fs_info->super_copy->csum_size being 0,
which goes unchecked before entering a division.
I was not able to identify where the kernel crashes (system goes down
the tubes), yet the problem is probably the same.

"btrfs version" is v3.19.1; bug is also present in latest git (kdave
and unstable) as of 2015/04/21


Full gdb output:

gdb btrfs
GNU gdb (GDB) Fedora 7.8.2-38.fc21
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-redhat-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from btrfs...Reading symbols from
/usr/lib/debug/usr/sbin/btrfs.debug...done.
done.
(gdb) run check btrfs_fukked.bin
Starting program: /usr/sbin/btrfs check btrfs_fukked.bin
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
Checking filesystem on btrfs_fukked.bin
UUID: cdd8684f-9eb1-40a4-91ec-1ed7c3cb444c
checking extents
checking free space cache
checking fs roots

Program received signal SIGFPE, Arithmetic exception.
count_csum_range (root=<optimized out>, root=<optimized out>,
    found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
1455 csum_end = key.offset + (size / csum_size) * root->sectorsize;
(gdb) bt
#0  count_csum_range (root=<optimized out>, root=<optimized out>,
    found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
#1  process_file_extent (active_node=0x7fffffffd710, key=0x7fffffffd680,
    slot=11, eb=<optimized out>, root=0x894b10) at cmds-check.c:1551
#2  process_one_leaf (wc=0x7fffffffd6c0, eb=<optimized out>, root=0x894b10)
    at cmds-check.c:1617
#3  walk_down_tree (level=<synthetic pointer>, wc=0x7fffffffd6c0,
    path=0x7fffffffd7f0, root=0x894b10) at cmds-check.c:1742
#4  check_fs_root (wc=0x7fffffffd6c0, root_cache=0x7fffffffdb20, root=0x894b10)
    at cmds-check.c:3380
#5  check_fs_roots (root_cache=root_cache@entry=0x7fffffffdb20, root=0x894b10)
    at cmds-check.c:3516
#6  0x0000000000428aea in cmd_check (argc=<optimized out>,
    argv=<optimized out>) at cmds-check.c:9465
#7  0x000000000040e5a2 in main (argc=2, argv=0x7fffffffdeb0) at btrfs.c:245
(gdb) p csum_size
$2 = 0

[-- Attachment #2: btrfs_fukked.bin.gz --]
[-- Type: application/x-gzip, Size: 25237 bytes --]

Re: Carefully crafted BTRFS-image causes kernel to crash

[Qu Wenruo]

-------- Original Message  --------
Subject: Carefully crafted BTRFS-image causes kernel to crash
From: Lukas Lueg <lukas.lueg@gmail.com>
To: <linux-btrfs@vger.kernel.org>
Date: 2015年04月21日 07:04

> See also https://bugzilla.kernel.org/show_bug.cgi?id=96971
>
>
> I've identified some problems in the btrfs code and attached a
> btrfs-image which causes the userland tools to crash and the kernel to
> immediately freeze once the filesystem get's mounted and one of the
> files is accessed. Putting the image onto a usb-drive gives you a
> freeze-on-a-stick :-)
>
> "btrfs check" crashes due to a SIGFPE in count_csum_range(). The
> culprit is struct btrfs_root->fs_info->super_copy->csum_size being 0,
> which goes unchecked before entering a division.
> I was not able to identify where the kernel crashes (system goes down
> the tubes), yet the problem is probably the same.
Thanks for the bug report.

Although we may add extra check for such problem to improve robustness,
but IMHO it's not a real world problem.

For normal case, csum will not be zero for sure.
And even for corrupted superblock, sb csum will prevent btrfs from using it.

This problem will only happen, as your mentioned, by a specially crafted 
image or a known bug which btrfs will still use corruped sb even sb csum 
dismatches.

Thanks,
Qu
>
> "btrfs version" is v3.19.1; bug is also present in latest git (kdave
> and unstable) as of 2015/04/21
>
>
> Full gdb output:
>
> gdb btrfs
> GNU gdb (GDB) Fedora 7.8.2-38.fc21
> Copyright (C) 2014 Free Software Foundation, Inc.
> License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
> This is free software: you are free to change and redistribute it.
> There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
> and "show warranty" for details.
> This GDB was configured as "x86_64-redhat-linux-gnu".
> Type "show configuration" for configuration details.
> For bug reporting instructions, please see:
> <http://www.gnu.org/software/gdb/bugs/>.
> Find the GDB manual and other documentation resources online at:
> <http://www.gnu.org/software/gdb/documentation/>.
> For help, type "help".
> Type "apropos word" to search for commands related to "word"...
> Reading symbols from btrfs...Reading symbols from
> /usr/lib/debug/usr/sbin/btrfs.debug...done.
> done.
> (gdb) run check btrfs_fukked.bin
> Starting program: /usr/sbin/btrfs check btrfs_fukked.bin
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib64/libthread_db.so.1".
> Checking filesystem on btrfs_fukked.bin
> UUID: cdd8684f-9eb1-40a4-91ec-1ed7c3cb444c
> checking extents
> checking free space cache
> checking fs roots
>
> Program received signal SIGFPE, Arithmetic exception.
> count_csum_range (root=<optimized out>, root=<optimized out>,
>      found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
> 1455 csum_end = key.offset + (size / csum_size) * root->sectorsize;
> (gdb) bt
> #0  count_csum_range (root=<optimized out>, root=<optimized out>,
>      found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
> #1  process_file_extent (active_node=0x7fffffffd710, key=0x7fffffffd680,
>      slot=11, eb=<optimized out>, root=0x894b10) at cmds-check.c:1551
> #2  process_one_leaf (wc=0x7fffffffd6c0, eb=<optimized out>, root=0x894b10)
>      at cmds-check.c:1617
> #3  walk_down_tree (level=<synthetic pointer>, wc=0x7fffffffd6c0,
>      path=0x7fffffffd7f0, root=0x894b10) at cmds-check.c:1742
> #4  check_fs_root (wc=0x7fffffffd6c0, root_cache=0x7fffffffdb20, root=0x894b10)
>      at cmds-check.c:3380
> #5  check_fs_roots (root_cache=root_cache@entry=0x7fffffffdb20, root=0x894b10)
>      at cmds-check.c:3516
> #6  0x0000000000428aea in cmd_check (argc=<optimized out>,
>      argv=<optimized out>) at cmds-check.c:9465
> #7  0x000000000040e5a2 in main (argc=2, argv=0x7fffffffdeb0) at btrfs.c:245
> (gdb) p csum_size
> $2 = 0
>

Re: Carefully crafted BTRFS-image causes kernel to crash

[Russell Coker]

On Tue, 21 Apr 2015, Qu Wenruo <quwenruo@cn.fujitsu.com> wrote:
> Although we may add extra check for such problem to improve robustness,
> but IMHO it's not a real world problem.

Some of the ReiserFS developers gave a similar reaction to some of my bug 
reports.  ReiserFS wasn't the most robust filesystem.

I think that it should be EXECTED that a kernel will have to occasionally deal 
with filesystem images that are created by hostile parties.  Userspace crash 
and kernel freeze is not a suitable way of dealing with it.

-- 
My Main Blog         http://etbe.coker.com.au/
My Documents Blog    http://doc.coker.com.au/

Re: Carefully crafted BTRFS-image causes kernel to crash

[Austin S Hemmelgarn]

[-- Attachment #1: Type: text/plain, Size: 955 bytes --]

On 2015-04-21 05:38, Russell Coker wrote:
> On Tue, 21 Apr 2015, Qu Wenruo <quwenruo@cn.fujitsu.com> wrote:
>> Although we may add extra check for such problem to improve robustness,
>> but IMHO it's not a real world problem.
>
> Some of the ReiserFS developers gave a similar reaction to some of my bug
> reports.  ReiserFS wasn't the most robust filesystem.
>
> I think that it should be EXECTED that a kernel will have to occasionally deal
> with filesystem images that are created by hostile parties.  Userspace crash
> and kernel freeze is not a suitable way of dealing with it.
>
And that kind of reaction is why ReiserFS isn't taken very seriously in 
an enterprise environment.

Just because something shouldn't be possible in a 'real world' 
environment, doesn't mean it won't happen; never underestimate the 
ability of hardware to fail in new and unexpected ways, or the ability 
of administrators to make stupid mistakes.


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 2967 bytes --]

Re: Carefully crafted BTRFS-image causes kernel to crash

[Zygo Blaxell]

[-- Attachment #1: Type: text/plain, Size: 5156 bytes --]

On Tue, Apr 21, 2015 at 11:16:44AM +0800, Qu Wenruo wrote:
> -------- Original Message  --------
> Subject: Carefully crafted BTRFS-image causes kernel to crash
> From: Lukas Lueg <lukas.lueg@gmail.com>
> To: <linux-btrfs@vger.kernel.org>
> Date: 2015年04月21日 07:04
> 
> >See also https://bugzilla.kernel.org/show_bug.cgi?id=96971
> >
> >
> >I've identified some problems in the btrfs code and attached a
> >btrfs-image which causes the userland tools to crash and the kernel to
> >immediately freeze once the filesystem get's mounted and one of the
> >files is accessed. Putting the image onto a usb-drive gives you a
> >freeze-on-a-stick :-)
> >
> >"btrfs check" crashes due to a SIGFPE in count_csum_range(). The
> >culprit is struct btrfs_root->fs_info->super_copy->csum_size being 0,
> >which goes unchecked before entering a division.
> >I was not able to identify where the kernel crashes (system goes down
> >the tubes), yet the problem is probably the same.
> Thanks for the bug report.
> 
> Although we may add extra check for such problem to improve robustness,
> but IMHO it's not a real world problem.

It depends on context.

If your real world involves putting hostile USB sticks into your computer
(e.g. you are a lawyer and your day job involves receiving electronic
documents from hostile and sometimes technically sophisticated opponents),
then this is definitely a real world problem.  It needs to be treated as a
security vulnerability (a minor one, given the requirement for physical
access and no data loss except for the kernel crash), maybe assigned a
CVE, and fixed with some urgency.

If your real world involves running racks of machines with no USB ports
in a locked data center then this issue is irrelevant noise.  Other btrfs
issues are *much* more likely to crash the kernel before this one.  ;)

> For normal case, csum will not be zero for sure.
> And even for corrupted superblock, sb csum will prevent btrfs from using it.
> 
> This problem will only happen, as your mentioned, by a specially
> crafted image or a known bug which btrfs will still use corruped sb
> even sb csum dismatches.
> 
> Thanks,
> Qu
> >
> >"btrfs version" is v3.19.1; bug is also present in latest git (kdave
> >and unstable) as of 2015/04/21
> >
> >
> >Full gdb output:
> >
> >gdb btrfs
> >GNU gdb (GDB) Fedora 7.8.2-38.fc21
> >Copyright (C) 2014 Free Software Foundation, Inc.
> >License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
> >This is free software: you are free to change and redistribute it.
> >There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
> >and "show warranty" for details.
> >This GDB was configured as "x86_64-redhat-linux-gnu".
> >Type "show configuration" for configuration details.
> >For bug reporting instructions, please see:
> ><http://www.gnu.org/software/gdb/bugs/>.
> >Find the GDB manual and other documentation resources online at:
> ><http://www.gnu.org/software/gdb/documentation/>.
> >For help, type "help".
> >Type "apropos word" to search for commands related to "word"...
> >Reading symbols from btrfs...Reading symbols from
> >/usr/lib/debug/usr/sbin/btrfs.debug...done.
> >done.
> >(gdb) run check btrfs_fukked.bin
> >Starting program: /usr/sbin/btrfs check btrfs_fukked.bin
> >[Thread debugging using libthread_db enabled]
> >Using host libthread_db library "/lib64/libthread_db.so.1".
> >Checking filesystem on btrfs_fukked.bin
> >UUID: cdd8684f-9eb1-40a4-91ec-1ed7c3cb444c
> >checking extents
> >checking free space cache
> >checking fs roots
> >
> >Program received signal SIGFPE, Arithmetic exception.
> >count_csum_range (root=<optimized out>, root=<optimized out>,
> >     found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
> >1455 csum_end = key.offset + (size / csum_size) * root->sectorsize;
> >(gdb) bt
> >#0  count_csum_range (root=<optimized out>, root=<optimized out>,
> >     found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
> >#1  process_file_extent (active_node=0x7fffffffd710, key=0x7fffffffd680,
> >     slot=11, eb=<optimized out>, root=0x894b10) at cmds-check.c:1551
> >#2  process_one_leaf (wc=0x7fffffffd6c0, eb=<optimized out>, root=0x894b10)
> >     at cmds-check.c:1617
> >#3  walk_down_tree (level=<synthetic pointer>, wc=0x7fffffffd6c0,
> >     path=0x7fffffffd7f0, root=0x894b10) at cmds-check.c:1742
> >#4  check_fs_root (wc=0x7fffffffd6c0, root_cache=0x7fffffffdb20, root=0x894b10)
> >     at cmds-check.c:3380
> >#5  check_fs_roots (root_cache=root_cache@entry=0x7fffffffdb20, root=0x894b10)
> >     at cmds-check.c:3516
> >#6  0x0000000000428aea in cmd_check (argc=<optimized out>,
> >     argv=<optimized out>) at cmds-check.c:9465
> >#7  0x000000000040e5a2 in main (argc=2, argv=0x7fffffffdeb0) at btrfs.c:245
> >(gdb) p csum_size
> >$2 = 0
> >
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: Carefully crafted BTRFS-image causes kernel to crash

[Qu Wenruo]

-------- Original Message  --------
Subject: Re: Carefully crafted BTRFS-image causes kernel to crash
From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
To: Qu Wenruo <quwenruo@cn.fujitsu.com>
Date: 2015年04月21日 23:17

> On Tue, Apr 21, 2015 at 11:16:44AM +0800, Qu Wenruo wrote:
>> -------- Original Message  --------
>> Subject: Carefully crafted BTRFS-image causes kernel to crash
>> From: Lukas Lueg <lukas.lueg@gmail.com>
>> To: <linux-btrfs@vger.kernel.org>
>> Date: 2015年04月21日 07:04
>>
>>> See also https://bugzilla.kernel.org/show_bug.cgi?id=96971
>>>
>>>
>>> I've identified some problems in the btrfs code and attached a
>>> btrfs-image which causes the userland tools to crash and the kernel to
>>> immediately freeze once the filesystem get's mounted and one of the
>>> files is accessed. Putting the image onto a usb-drive gives you a
>>> freeze-on-a-stick :-)
>>>
>>> "btrfs check" crashes due to a SIGFPE in count_csum_range(). The
>>> culprit is struct btrfs_root->fs_info->super_copy->csum_size being 0,
>>> which goes unchecked before entering a division.
>>> I was not able to identify where the kernel crashes (system goes down
>>> the tubes), yet the problem is probably the same.
>> Thanks for the bug report.
>>
>> Although we may add extra check for such problem to improve robustness,
>> but IMHO it's not a real world problem.
>
> It depends on context.
>
> If your real world involves putting hostile USB sticks into your computer
> (e.g. you are a lawyer and your day job involves receiving electronic
> documents from hostile and sometimes technically sophisticated opponents),
> then this is definitely a real world problem.  It needs to be treated as a
> security vulnerability (a minor one, given the requirement for physical
> access and no data loss except for the kernel crash), maybe assigned a
> CVE, and fixed with some urgency.
Oh, I forgot the hostile USB case.

And for that case, I think is is worthy to fix.
For kernel, IIRC, there is a function to check superblock validation,
so just one or two new lines to check csum size.

And for btrfs-progs, we can add such function too.

I'll send the patch recently

BUT I also want to show the fact that, BTRFS is still under heavy 
development for its features and stability, so robustness may not be the 
primary concern yet.
(More daily operation can cause problems, see fstests, run with old
kernels and you should get a lot of hangs)
Although any such report is still welcomed and I'll try to fix them if I 
have spare time.

Thanks,
Qu
>
> If your real world involves running racks of machines with no USB ports
> in a locked data center then this issue is irrelevant noise.  Other btrfs
> issues are *much* more likely to crash the kernel before this one.  ;)
>
>> For normal case, csum will not be zero for sure.
>> And even for corrupted superblock, sb csum will prevent btrfs from using it.
>>
>> This problem will only happen, as your mentioned, by a specially
>> crafted image or a known bug which btrfs will still use corruped sb
>> even sb csum dismatches.
>>
>> Thanks,
>> Qu
>>>
>>> "btrfs version" is v3.19.1; bug is also present in latest git (kdave
>>> and unstable) as of 2015/04/21
>>>
>>>
>>> Full gdb output:
>>>
>>> gdb btrfs
>>> GNU gdb (GDB) Fedora 7.8.2-38.fc21
>>> Copyright (C) 2014 Free Software Foundation, Inc.
>>> License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
>>> This is free software: you are free to change and redistribute it.
>>> There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
>>> and "show warranty" for details.
>>> This GDB was configured as "x86_64-redhat-linux-gnu".
>>> Type "show configuration" for configuration details.
>>> For bug reporting instructions, please see:
>>> <http://www.gnu.org/software/gdb/bugs/>.
>>> Find the GDB manual and other documentation resources online at:
>>> <http://www.gnu.org/software/gdb/documentation/>.
>>> For help, type "help".
>>> Type "apropos word" to search for commands related to "word"...
>>> Reading symbols from btrfs...Reading symbols from
>>> /usr/lib/debug/usr/sbin/btrfs.debug...done.
>>> done.
>>> (gdb) run check btrfs_fukked.bin
>>> Starting program: /usr/sbin/btrfs check btrfs_fukked.bin
>>> [Thread debugging using libthread_db enabled]
>>> Using host libthread_db library "/lib64/libthread_db.so.1".
>>> Checking filesystem on btrfs_fukked.bin
>>> UUID: cdd8684f-9eb1-40a4-91ec-1ed7c3cb444c
>>> checking extents
>>> checking free space cache
>>> checking fs roots
>>>
>>> Program received signal SIGFPE, Arithmetic exception.
>>> count_csum_range (root=<optimized out>, root=<optimized out>,
>>>      found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
>>> 1455 csum_end = key.offset + (size / csum_size) * root->sectorsize;
>>> (gdb) bt
>>> #0  count_csum_range (root=<optimized out>, root=<optimized out>,
>>>      found=<synthetic pointer>, len=7385088, start=7471104) at cmds-check.c:1455
>>> #1  process_file_extent (active_node=0x7fffffffd710, key=0x7fffffffd680,
>>>      slot=11, eb=<optimized out>, root=0x894b10) at cmds-check.c:1551
>>> #2  process_one_leaf (wc=0x7fffffffd6c0, eb=<optimized out>, root=0x894b10)
>>>      at cmds-check.c:1617
>>> #3  walk_down_tree (level=<synthetic pointer>, wc=0x7fffffffd6c0,
>>>      path=0x7fffffffd7f0, root=0x894b10) at cmds-check.c:1742
>>> #4  check_fs_root (wc=0x7fffffffd6c0, root_cache=0x7fffffffdb20, root=0x894b10)
>>>      at cmds-check.c:3380
>>> #5  check_fs_roots (root_cache=root_cache@entry=0x7fffffffdb20, root=0x894b10)
>>>      at cmds-check.c:3516
>>> #6  0x0000000000428aea in cmd_check (argc=<optimized out>,
>>>      argv=<optimized out>) at cmds-check.c:9465
>>> #7  0x000000000040e5a2 in main (argc=2, argv=0x7fffffffdeb0) at btrfs.c:245
>>> (gdb) p csum_size
>>> $2 = 0
>>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Carefully crafted BTRFS-image causes kernel to crash

[Lukas Lueg]

Hi,

a patch was posted by Qu which should fix the issue, I did not check
though. To give some context: Testing the userland tools to some
depths comes back with around 13 percent of all execution paths in
around 40 million runs to result in a segmentation fault, an uncatched
divide-by-zero or self-destruction due to a heap buffer overflow
catched by glibc. That is to be compared with zero crashes in the
ext2fs userland tools.

I'll pull in Qu's patch once there are patches for the other problems
that were already posted and continue testing then.

Best regards
Lukas

2015-04-22 2:28 GMT+02:00 Qu Wenruo <quwenruo@cn.fujitsu.com>:
>
>
> -------- Original Message  --------
> Subject: Re: Carefully crafted BTRFS-image causes kernel to crash
> From: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
> To: Qu Wenruo <quwenruo@cn.fujitsu.com>
> Date: 2015年04月21日 23:17
>
>> On Tue, Apr 21, 2015 at 11:16:44AM +0800, Qu Wenruo wrote:
>>>
>>> -------- Original Message  --------
>>> Subject: Carefully crafted BTRFS-image causes kernel to crash
>>> From: Lukas Lueg <lukas.lueg@gmail.com>
>>> To: <linux-btrfs@vger.kernel.org>
>>> Date: 2015年04月21日 07:04
>>>
>>>> See also https://bugzilla.kernel.org/show_bug.cgi?id=96971
>>>>
>>>>
>>>> I've identified some problems in the btrfs code and attached a
>>>> btrfs-image which causes the userland tools to crash and the kernel to
>>>> immediately freeze once the filesystem get's mounted and one of the
>>>> files is accessed. Putting the image onto a usb-drive gives you a
>>>> freeze-on-a-stick :-)
>>>>
>>>> "btrfs check" crashes due to a SIGFPE in count_csum_range(). The
>>>> culprit is struct btrfs_root->fs_info->super_copy->csum_size being 0,
>>>> which goes unchecked before entering a division.
>>>> I was not able to identify where the kernel crashes (system goes down
>>>> the tubes), yet the problem is probably the same.
>>>
>>> Thanks for the bug report.
>>>
>>> Although we may add extra check for such problem to improve robustness,
>>> but IMHO it's not a real world problem.
>>
>>
>> It depends on context.
>>
>> If your real world involves putting hostile USB sticks into your computer
>> (e.g. you are a lawyer and your day job involves receiving electronic
>> documents from hostile and sometimes technically sophisticated opponents),
>> then this is definitely a real world problem.  It needs to be treated as a
>> security vulnerability (a minor one, given the requirement for physical
>> access and no data loss except for the kernel crash), maybe assigned a
>> CVE, and fixed with some urgency.
>
> Oh, I forgot the hostile USB case.
>
> And for that case, I think is is worthy to fix.
> For kernel, IIRC, there is a function to check superblock validation,
> so just one or two new lines to check csum size.
>
> And for btrfs-progs, we can add such function too.
>
> I'll send the patch recently
>
> BUT I also want to show the fact that, BTRFS is still under heavy
> development for its features and stability, so robustness may not be the
> primary concern yet.
> (More daily operation can cause problems, see fstests, run with old
> kernels and you should get a lot of hangs)
> Although any such report is still welcomed and I'll try to fix them if I
> have spare time.
>
> Thanks,
> Qu
>
>>
>> If your real world involves running racks of machines with no USB ports
>> in a locked data center then this issue is irrelevant noise.  Other btrfs
>> issues are *much* more likely to crash the kernel before this one.  ;)
>>
>>> For normal case, csum will not be zero for sure.
>>> And even for corrupted superblock, sb csum will prevent btrfs from using
>>> it.
>>>
>>> This problem will only happen, as your mentioned, by a specially
>>> crafted image or a known bug which btrfs will still use corruped sb
>>> even sb csum dismatches.
>>>
>>> Thanks,
>>> Qu
>>>>
>>>>
>>>> "btrfs version" is v3.19.1; bug is also present in latest git (kdave
>>>> and unstable) as of 2015/04/21
>>>>
>>>>
>>>> Full gdb output:
>>>>
>>>> gdb btrfs
>>>> GNU gdb (GDB) Fedora 7.8.2-38.fc21
>>>> Copyright (C) 2014 Free Software Foundation, Inc.
>>>> License GPLv3+: GNU GPL version 3 or later
>>>> <http://gnu.org/licenses/gpl.html>
>>>> This is free software: you are free to change and redistribute it.
>>>> There is NO WARRANTY, to the extent permitted by law.  Type "show
>>>> copying"
>>>> and "show warranty" for details.
>>>> This GDB was configured as "x86_64-redhat-linux-gnu".
>>>> Type "show configuration" for configuration details.
>>>> For bug reporting instructions, please see:
>>>> <http://www.gnu.org/software/gdb/bugs/>.
>>>> Find the GDB manual and other documentation resources online at:
>>>> <http://www.gnu.org/software/gdb/documentation/>.
>>>> For help, type "help".
>>>> Type "apropos word" to search for commands related to "word"...
>>>> Reading symbols from btrfs...Reading symbols from
>>>> /usr/lib/debug/usr/sbin/btrfs.debug...done.
>>>> done.
>>>> (gdb) run check btrfs_fukked.bin
>>>> Starting program: /usr/sbin/btrfs check btrfs_fukked.bin
>>>> [Thread debugging using libthread_db enabled]
>>>> Using host libthread_db library "/lib64/libthread_db.so.1".
>>>> Checking filesystem on btrfs_fukked.bin
>>>> UUID: cdd8684f-9eb1-40a4-91ec-1ed7c3cb444c
>>>> checking extents
>>>> checking free space cache
>>>> checking fs roots
>>>>
>>>> Program received signal SIGFPE, Arithmetic exception.
>>>> count_csum_range (root=<optimized out>, root=<optimized out>,
>>>>      found=<synthetic pointer>, len=7385088, start=7471104) at
>>>> cmds-check.c:1455
>>>> 1455 csum_end = key.offset + (size / csum_size) * root->sectorsize;
>>>> (gdb) bt
>>>> #0  count_csum_range (root=<optimized out>, root=<optimized out>,
>>>>      found=<synthetic pointer>, len=7385088, start=7471104) at
>>>> cmds-check.c:1455
>>>> #1  process_file_extent (active_node=0x7fffffffd710, key=0x7fffffffd680,
>>>>      slot=11, eb=<optimized out>, root=0x894b10) at cmds-check.c:1551
>>>> #2  process_one_leaf (wc=0x7fffffffd6c0, eb=<optimized out>,
>>>> root=0x894b10)
>>>>      at cmds-check.c:1617
>>>> #3  walk_down_tree (level=<synthetic pointer>, wc=0x7fffffffd6c0,
>>>>      path=0x7fffffffd7f0, root=0x894b10) at cmds-check.c:1742
>>>> #4  check_fs_root (wc=0x7fffffffd6c0, root_cache=0x7fffffffdb20,
>>>> root=0x894b10)
>>>>      at cmds-check.c:3380
>>>> #5  check_fs_roots (root_cache=root_cache@entry=0x7fffffffdb20,
>>>> root=0x894b10)
>>>>      at cmds-check.c:3516
>>>> #6  0x0000000000428aea in cmd_check (argc=<optimized out>,
>>>>      argv=<optimized out>) at cmds-check.c:9465
>>>> #7  0x000000000040e5a2 in main (argc=2, argv=0x7fffffffdeb0) at
>>>> btrfs.c:245
>>>> (gdb) p csum_size
>>>> $2 = 0
>>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Carefully crafted BTRFS-image causes kernel to crash

[Duncan]

Lukas Lueg posted on Wed, 29 Apr 2015 21:50:23 +0200 as excerpted:

> To give some context: Testing the userland tools to some depths comes
> back with around 13 percent of all execution paths in around 40 million
> runs to result in a segmentation fault, an uncatched divide-by-zero or
> self-destruction due to a heap buffer overflow catched by glibc. That is
> to be compared with zero crashes in the ext2fs userland tools.

In a field of 40 million trinity fuzz tester runs, one out of eight btrfs-
tools userspace runs crashed due to some abnormality, compared to 0 
crashes for e2fsprogs.

That's _very_ impressive for e2fs.  I'm sure it's because people have 
been doing just that, beating on it with trinity and the like, for 
awhile, and the problems have been fixed already, but that makes it no 
less impressive.  It takes both work and loving care to get to that 
point, and this demonstrates just how much of that it gets, to get to the 
_zero_ crash (in tens of millions of runs) point.

Meanwhile, one of eight for btrfs-progs.  Of course the focus has so far 
been pretty much on just getting the tools and basic functionality, so 
there's a reason, but it has a _long_ way to go from one of eight to 
_zero_ of 40-million!

Just one more demonstration of the fact that btrfs really isn't and can't 
be called stable yet, even if from about a year ago they _have_ pretty 
well stripped all the warnings telling people that.

Of course userspace is only the half of it, but _zero_ fuzzer crashes in 
40 million runs... userspace or kernelspace regardless, that's quite some 
work showing off there, and they have a right to be proud of it.  Btrfs 
would be so lucky!  But your beating on it and Qu's fixup patches are a 
start at getting us there.  Thanks! =:^)

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

Re: Carefully crafted BTRFS-image causes kernel to crash

[David Sterba]

On Tue, Apr 21, 2015 at 07:44:12AM -0400, Austin S Hemmelgarn wrote:
> On 2015-04-21 05:38, Russell Coker wrote:
> > On Tue, 21 Apr 2015, Qu Wenruo <quwenruo@cn.fujitsu.com> wrote:
> >> Although we may add extra check for such problem to improve robustness,
> >> but IMHO it's not a real world problem.
> >
> > Some of the ReiserFS developers gave a similar reaction to some of my bug
> > reports.  ReiserFS wasn't the most robust filesystem.
> >
> > I think that it should be EXECTED that a kernel will have to occasionally deal
> > with filesystem images that are created by hostile parties.  Userspace crash
> > and kernel freeze is not a suitable way of dealing with it.
> >
> And that kind of reaction is why ReiserFS isn't taken very seriously in 
> an enterprise environment.
> 
> Just because something shouldn't be possible in a 'real world' 
> environment, doesn't mean it won't happen; never underestimate the 
> ability of hardware to fail in new and unexpected ways, or the ability 
> of administrators to make stupid mistakes.

I agree, given the high number of random memory bitflips that I've seen
in past years, that were stored to the disk with valid checksums.
Whether it's a real-world or potentially a hw problem IMO adjusts the
priority, but the issues should be fixed if possible.