linux-kernel - Re: ftrace: trace_pipe

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <AANLkTi=_VJcDmpdfy=nfo_771iHjdVHTHnciUe8DrZme@mail.gmail.com>
Date:	Thu, 23 Dec 2010 15:06:18 -0800
From:	David Sharp <dhsharp@...gle.com>
To:	Steven Rostedt <rostedt@...dmis.org>
Cc:	linux-kernel@...r.kernel.org, Michael Rubin <mrubin@...gle.com>
Subject: Re: ftrace: trace_pipe_raw interface broken

On Thu, Dec 23, 2010 at 7:57 AM, Steven Rostedt <rostedt@...dmis.org> wrote:
> Hi David,
>
> Thanks for looking deeper into this.
>
> On Wed, 2010-12-22 at 16:37 -0800, David Sharp wrote:
>
>> Regarding the strange "commit" values:
>>
>> I noticed just now that it appears the low 16-bits are the correct
>> value for "commit", bits 16-31 are c000, and bits 32-63 are ffffffff.
>
> Ah yeah, I use 32bits so it would be the same on both 32 and 64.
>
>>
>> ring_buffer_read_page can store the number of missing events at the
>> end of the page if there is room. It signals it has done so with two
>> bits in commit, bits 30 and 31. That's c0000000. This points to the
>> ffffffff being a signed math problem, because these bits are added
>> using "local_add".
>>
>> Here's how the bits are defined:
>>
>> /* Flag when events were overwritten */
>> #define RB_MISSED_EVENTS      (1 << 31)
>> /* Missed count stored at end */
>> #define RB_MISSED_STORED      (1 << 30)
>>
>> well, those would come out as signed int, and 1<<31 is 0x80000000, aka
>> INT_MIN. When passed to local_add, which takes signed long, that would
>> be sign-extended to 0xffffffff80000000.
>>
>> Well, mystery solved at least. Now, how should it be fixed? Or is this
>> intended behavior?
>
> Not quite intended, but not something to worry about either. We mask off
> the 30 bits to determine the size.

in trace-cmd you mean?

>> By my count that leaves only one mystery: why are we seeing this extra
>> page at the beginning with pre-overflow data?
>
> If you did not reset the buffer, there's a chance that the writer is on
> the reader page. The reader page is always outside the ring buffer, but
> it points into the ring buffer. If this occurs, then you will get the
> reader page data, plus the rest of the ring buffer (which is the full
> size you asked for).

Okay, that's a surprising behavior to me, but resetting the buffer
seems like a fine workaround.

btw, corroborating this is that reading the formatted output of
"trace" also shows the page of pre-overflow data.

Thanks for your help,
d#
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/