linux-kernel - Re: [RFC] perf record: missing buildid for callstack modules

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]

Date:	Tue, 12 Jan 2016 11:34:54 -0300
From:	Arnaldo Carvalho de Melo <acme@...nel.org>
To:	Peter Zijlstra <peterz@...radead.org>
Cc:	Ingo Molnar <mingo@...nel.org>,
	Stephane Eranian <eranian@...gle.com>,
	Namhyung Kim <namhyung@...il.com>,
	LKML <linux-kernel@...r.kernel.org>,
	Jiri Olsa <jolsa@...nel.org>,
	Namhyung Kim <namhyung@...nel.org>,
	Adrian Hunter <adrian.hunter@...el.com>,
	"ak@...ux.intel.com" <ak@...ux.intel.com>
Subject: Re: [RFC] perf record: missing buildid for callstack modules

Em Tue, Jan 12, 2016 at 12:35:21PM +0100, Peter Zijlstra escreveu:
> On Tue, Jan 12, 2016 at 11:39:43AM +0100, Ingo Molnar wrote:
> > > Does the kernel even know about the buildid crap? AFAIK the binfmt stuff doesn't 
> > > know or care about things like that. Heck, we support binfmts that do not even 
> > > have a buildid.
> > 
> > The kernel's exec() code does not care about the past, it will execute whatever is 
> > fit to execute right now.
> > 
> > But perf tooling cares very much: it can lead to subtle bugs and bad data if we 
> > display a profile with the wrong DSO or binary. 'Bad' profiles resulting out of 
> > binary mismatch can be very convincing and can send developers down the wrong path 
> > for hours. I'd expect my tooling to not do that.
> 
> Well, it really is rather a rare case, replacing binaries you're
> profiling. Sure, if it happens (by accident or otherwise) it can be a
> pain, but the cost of fixing this 'problem' is huge.

Humm, for most things today, i.e. ELF, most distros (all? maybe this is
a gcc switch that is default on, haven't checked) come with such
pre-computed cookie its just a way for efficiently passing it to the
tooling via a new record. With that, no post processing, etc. But then
someone would need to prototype this...

[acme@zoo linux]$ perf record -h build

 Usage: perf record [<options>] [<command>]
    or: perf record [<options>] -- <command> [<options>]

    -B, --no-buildid      do not collect buildids in perf.data
    -N, --no-buildid-cache
                          do not update the buildid cache

[acme@zoo linux]$

Have you ever played, when you noticed those overheads, with -N? Or just
used the -B big hammer and moved on?

- Arnaldo
 
> > Path names alone (the thing that exec() cares about) are not unique enough to 
> > identify the binary that was profiled. So we need a content hash - hence the 
> > build-ID.
> > 
> > Can you suggest a better solution than a build-time calculated content hash?
> 
> Not really, but the current 'solution' is a massive pain. The result is
> that perf-record needs to do a full scan of the recorded data after
> completion and look for buildids across the system.
> 
> On my system that pass takes longer than the actual workload (of
> building a kernel). Furthermore, the resulting data is useless for me.
> 
> > As for binary formats that suck and don't allow for a content hash: we do our 
> > best, but of course the risk of data mismatch is there. We could perhaps cache the 
> > binary inode's mtime field to at least produce a 'profile data is older than 
> > binary/DSO modification date!' warning. (Which check won't catch all cases, like 
> > cross-system profiling data matches.)
> 
> So my problem with the kernel side thing is that I fear it will, again,
> be a partial solution, and we'll still end up scanning the perf-record
> output, ie. nothing better than we are now.
> 
> Sure, maybe we can have binfmt_elf read the buildid and cache it
> someplace, maybe we can even have the other binfmt thingies do something
> similar (at small cost, we obviously cannot compute hashes over files at
> exec() time, that would upset people).
> 
> But what do we do for DSOs, does dlopen() ever end up in the binfmt
> code? I would think not, I would fully expect the dynamic linker to just
> mmap() the relevant bits and be done with it.
> 
> And we cannot, at mmap() time, 'assume' the file is ELF and try prodding
> into it to find a buildid or whatnot.
> 
> And all for some weird corner case.
> 
> ~ Peter, who thinks buildid stuff stinks.