Re: Fix build ID parsing logic in stable trees

[Jiri Olsa <olsajiri@gmail.com>]

On Wed, Nov 13, 2024 at 12:07:39PM -0800, Andrii Nakryiko wrote:
> On Fri, Nov 8, 2024 at 3:26 PM Jiri Olsa <olsajiri@gmail.com> wrote:
> >
> > On Fri, Nov 08, 2024 at 09:35:34AM +0100, Jiri Olsa wrote:
> > > On Thu, Nov 07, 2024 at 12:04:05PM -0800, Omar Sandoval wrote:
> > > > On Wed, Nov 06, 2024 at 12:57:34PM +0100, Jiri Olsa wrote:
> > > > > On Wed, Nov 06, 2024 at 07:12:05AM +0100, Greg KH wrote:
> > > > > > On Tue, Nov 05, 2024 at 10:15:04AM +0100, Jiri Olsa wrote:
> > > > > > > On Tue, Nov 05, 2024 at 07:54:48AM +0100, Greg KH wrote:
> > > > > > > > On Mon, Nov 04, 2024 at 06:52:52PM +0100, Jiri Olsa wrote:
> > > > > > > > > hi,
> > > > > > > > > sending fix for buildid parsing that affects only stable trees
> > > > > > > > > after merging upstream fix [1].
> > > > > > > > >
> > > > > > > > > Upstream then factored out the whole buildid parsing code, so it
> > > > > > > > > does not have the problem.
> > > > > > > >
> > > > > > > > Why not just take those patches instead?
> > > > > > >
> > > > > > > I guess we could, but I thought it's too big for stable
> > > > > > >
> > > > > > > we'd need following 2 changes to fix the issue:
> > > > > > >   de3ec364c3c3 lib/buildid: add single folio-based file reader abstraction
> > > > > > >   60c845b4896b lib/buildid: take into account e_phoff when fetching program headers
> > > > > > >
> > > > > > > and there's also few other follow ups:
> > > > > > >   5ac9b4e935df lib/buildid: Handle memfd_secret() files in build_id_parse()
> > > > > > >   cdbb44f9a74f lib/buildid: don't limit .note.gnu.build-id to the first page in ELF
> > > > > > >   ad41251c290d lib/buildid: implement sleepable build_id_parse() API
> > > > > > >   45b8fc309654 lib/buildid: rename build_id_parse() into build_id_parse_nofault()
> > > > > > >   4e9d360c4cdf lib/buildid: remove single-page limit for PHDR search
> > > > > > >
> > > > > > > which I guess are not strictly needed
> > > > > >
> > > > > > Can you verify what exact ones are needed here?  We'll be glad to take
> > > > > > them if you can verify that they work properly.
> > > > >
> > > > > ok, will check
> > > >
> > > > Hello,
> > > >
> > > > I noticed that the BUILD-ID field in vmcoreinfo is broken on
> > > > stable/longterm kernels and found this thread. Can we please get this
> > > > fixed soon?
> > > >
> > > > I tried cherry-picking the patches mentioned above ("lib/buildid: add
> > > > single folio-based file reader abstraction" and "lib/buildid: take into
> > > > account e_phoff when fetching program headers"), but they don't apply
> > > > cleanly before 6.11, and they'd need to be reworked for 5.15, which was
> > > > before folios were introduced. Jiri's minimal fix works for me and seems
> > > > like a much safer option.
> > >
> > > hi,
> > > thanks for testing
> > >
> > > I think for 6.11 we could go with backport of:
> > >   de3ec364c3c3 lib/buildid: add single folio-based file reader abstraction
> > >   60c845b4896b lib/buildid: take into account e_phoff when fetching program headers
> > >
> > > and with the small fix for the rest
> > >
> > > but I still need to figure out why also 60c845b4896b is needed
> > > to fix the issue on 6.11.. hopefully today
> >
> > ok, so the fix the issue in 6.11 with upstream backports we'd need both:
> >
> >   1) de3ec364c3c3 lib/buildid: add single folio-based file reader abstraction
> >   2) 60c845b4896b lib/buildid: take into account e_phoff when fetching program headers
> >
> > 2) is needed because 1) seems to omit ehdr->e_phoff addition (patch below)
> > which is added back in 2)
> >
> > IMO 6.11 is close to upstream and by taking above upstream fixes it will be
> > easier to backport other possible fixes in the future, for other trees I'd
> > take the original one line fix I posted
> 
> I still maintain that very minimal is the way to go instead of risking
> bringing new potential regressions by partially backporting folio
> rework patchset.
> 
> Jiri, there is no point in risking this, best to fix this quickly and
> minimally. If we ever need to backport further fixes, *then* we can
> think about folio-based implementation backport.

ok, make sense, the original plan works for me as well

jirka

> 
> >
> > jirka
> >
> >
> > ---
> > diff --git a/lib/buildid.c b/lib/buildid.c
> > index bfe00b66b1e8..19d9a0f6ce99 100644
> > --- a/lib/buildid.c
> > +++ b/lib/buildid.c
> > @@ -234,7 +234,7 @@ static int get_build_id_32(struct freader *r, unsigned char *build_id, __u32 *si
> >                 return -EINVAL;
> >
> >         for (i = 0; i < phnum; ++i) {
> > -               phdr = freader_fetch(r, i * sizeof(Elf32_Phdr), sizeof(Elf32_Phdr));
> > +               phdr = freader_fetch(r, sizeof(Elf32_Ehdr) + i * sizeof(Elf32_Phdr), sizeof(Elf32_Phdr));
> >                 if (!phdr)
> >                         return r->err;
> >
> > @@ -272,7 +272,7 @@ static int get_build_id_64(struct freader *r, unsigned char *build_id, __u32 *si
> >                 return -EINVAL;
> >
> >         for (i = 0; i < phnum; ++i) {
> > -               phdr = freader_fetch(r, i * sizeof(Elf64_Phdr), sizeof(Elf64_Phdr));
> > +               phdr = freader_fetch(r, sizeof(Elf64_Ehdr) + i * sizeof(Elf64_Phdr), sizeof(Elf64_Phdr));
> >                 if (!phdr)
> >                         return r->err;
> >