Btrfs Issues

Btrfs Issues

[Ruoxin Jiang]

[-- Attachment #1: Type: text/plain, Size: 954 bytes --]

Hello,

We are researchers from Columbia University, New York. As part of our
current research we have found some semantic discrepancies between
btrfs and other popular filesystems.

We have attached two cases. The first one involves an invalid O_DIRECT
write() that fails back to buffered write instead of failing with
error EINVAL. In directory 2, we discovered that btrfs calculates
write_bytes in __btrfs_buffered_write differently from that in
generic_perform_writes in fs/mmap.c. This can cause inconsistent
behavior between btrfs and other filesystems when program invokes the
same writev/write() syscall.

In each directory, you will find a Readme.md describing the issue and
pointing to the code that may cause the problem. For your convenience,
we also included test programs (min.cpp) and instructions in Readme to
help reproduce the issues.

We would appreciate very much if you could confirm the two cases at
your conveniences.

Thanks,
Amy

[-- Attachment #2: btrfs_issues.tar.gz --]
[-- Type: application/x-gzip, Size: 4825 bytes --]

Re: Btrfs Issues

[Satoru Takeuchi]

At Wed, 13 Sep 2017 11:53:35 -0400,
Ruoxin Jiang wrote:
> 
> [1  <text/plain; UTF-8 (7bit)>]
> Hello,
> 
> We are researchers from Columbia University, New York. As part of our
> current research we have found some semantic discrepancies between
> btrfs and other popular filesystems.
> 
> We have attached two cases. The first one involves an invalid O_DIRECT
> write() that fails back to buffered write instead of failing with
> error EINVAL. In directory 2, we discovered that btrfs calculates
> write_bytes in __btrfs_buffered_write differently from that in
> generic_perform_writes in fs/mmap.c. This can cause inconsistent
> behavior between btrfs and other filesystems when program invokes the
> same writev/write() syscall.
> 
> In each directory, you will find a Readme.md describing the issue and
> pointing to the code that may cause the problem. For your convenience,
> we also included test programs (min.cpp) and instructions in Readme to
> help reproduce the issues.
> 
> We would appreciate very much if you could confirm the two cases at
> your conveniences.

I took a look at your test programs, btrfs_issues/{1,2}/min.cpp. It looks
very hard to read since you call syscalls in odd ways and all flags are
hardcoded as literal hexadecimal numbers. Could rewrite these program
to improve readability?

In addition, I have two questions about btrfs_issues/1/min.cpp.

1. Why you set 'filename' as the 1st argument of mmap()?
2. What's the purpose of mmap() call? I guess mmap() is not related to issue 1.

Thanks,
Satoru

> 
> Thanks,
> Amy
> [2 btrfs_issues.tar.gz <application/x-gzip (base64)>]

Re: Btrfs Issues

[Ruoxin Jiang]

[-- Attachment #1: Type: text/plain, Size: 2472 bytes --]

Hi Satoru,

I'm so sorry for the late reply! I rewrote the test programs to invoke
syscalls in the conventional way and added a few comments. Please find
attached the revised report.

As to your questions on "mmap", you are totally right that the first
mmap in the original 1/min.cpp is irrelevant. The original program was
automatically generated from a fuzzer I used and I hope the revised
ones to be much more understandable.

Thanks,
Amy

On Sun, Sep 17, 2017 at 11:10 PM, Satoru Takeuchi
<satoru.takeuchi@gmail.com> wrote:
> At Wed, 13 Sep 2017 11:53:35 -0400,
> Ruoxin Jiang wrote:
>>
>> [1  <text/plain; UTF-8 (7bit)>]
>> Hello,
>>
>> We are researchers from Columbia University, New York. As part of our
>> current research we have found some semantic discrepancies between
>> btrfs and other popular filesystems.
>>
>> We have attached two cases. The first one involves an invalid O_DIRECT
>> write() that fails back to buffered write instead of failing with
>> error EINVAL. In directory 2, we discovered that btrfs calculates
>> write_bytes in __btrfs_buffered_write differently from that in
>> generic_perform_writes in fs/mmap.c. This can cause inconsistent
>> behavior between btrfs and other filesystems when program invokes the
>> same writev/write() syscall.
>>
>> In each directory, you will find a Readme.md describing the issue and
>> pointing to the code that may cause the problem. For your convenience,
>> we also included test programs (min.cpp) and instructions in Readme to
>> help reproduce the issues.
>>
>> We would appreciate very much if you could confirm the two cases at
>> your conveniences.
>
> I took a look at your test programs, btrfs_issues/{1,2}/min.cpp. It looks
> very hard to read since you call syscalls in odd ways and all flags are
> hardcoded as literal hexadecimal numbers. Could rewrite these program
> to improve readability?
>
> In addition, I have two questions about btrfs_issues/1/min.cpp.
>
> 1. Why you set 'filename' as the 1st argument of mmap()?
> 2. What's the purpose of mmap() call? I guess mmap() is not related to issue 1.
>
> Thanks,
> Satoru
>
>>
>> Thanks,
>> Amy
>> [2 btrfs_issues.tar.gz <application/x-gzip (base64)>]
> --
> 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



-- 
Ruoxin(Amy) Jiang
Columbia University, M.S. Computer Science
347-277-5455

[-- Attachment #2: btrfs_issues_revised.tar.gz --]
[-- Type: application/x-gzip, Size: 1801 bytes --]