code review 70710043: cmd/ld: don't include padding length in size of the not...

cmd/ld: don't include padding length in size of the note section
Recently NetBSD starts to enforce this, and refuses to execute
the program if n is larger than the sum of entry sizes.

Before:
$ readelf -n ../bin/go.old
Notes at offset 0x00000bd0 with length 0x00000019:
  Owner         Data size       Description
  NetBSD                0x00000004      NT_VERSION (version)
readelf: Warning: corrupt note found at offset 18 into core notes
readelf: Warning:  type: 0, namesize: 00000000, descsize: 00000000
$ readelf -n ../bin/go
Notes at offset 0x00000bd0 with length 0x00000018:
Owner         Data size       Description
NetBSD                0x00000004      NT_VERSION (version)

[minux1, 2014-03-03 08:02:39]

Hello iant@golang.org (cc: golang-codereviews@googlegroups.com),

I'd like you to review this change to
https://code.google.com/p/go

[iant, 2014-03-03 15:44:00]

https://codereview.appspot.com/70710043/diff/40001/src/cmd/ld/elf.c
File src/cmd/ld/elf.c (right):

https://codereview.appspot.com/70710043/diff/40001/src/cmd/ld/elf.c#newcode358
src/cmd/ld/elf.c:358: sh->size = n & ~3;
I think this will only produce the correct results when sz is a multiple of 4,
or by coincidence.  There is nothing that forces sz to be a multiple of 4 here;
for the NetBSD and OpenBSD notes it will always be, but for the GNU note it
won't.  I think you need to instead change the first line to something like
    resoff &= ~(uint64)3;
    n = sizeof(Elf_Note) + sz;

[minux1, 2014-03-03 21:10:05]

https://codereview.appspot.com/70710043/diff/40001/src/cmd/ld/elf.c
File src/cmd/ld/elf.c (right):

https://codereview.appspot.com/70710043/diff/40001/src/cmd/ld/elf.c#newcode358
src/cmd/ld/elf.c:358: sh->size = n & ~3;
On 2014/03/03 15:44:00, iant wrote:
> I think this will only produce the correct results when sz is a multiple of 4,
> or by coincidence.  There is nothing that forces sz to be a multiple of 4
here;
> for the NetBSD and OpenBSD notes it will always be, but for the GNU note it
> won't.  I think you need to instead change the first line to something like
>     resoff &= ~(uint64)3;
>     n = sizeof(Elf_Note) + sz;
I think we can probably do this:
sh->size = n - resoff % 4;

the important thing here is that sh->size should reflect the real size of
the note.

[minux1, 2014-03-03 21:12:04]

PTAL.

[iant, 2014-03-03 22:13:32]

LGTM

[rsc, 2014-03-05 19:40:58]

*** Submitted as https://code.google.com/p/go/source/detail?r=db613ec83949 ***

cmd/ld: don't include padding length in size of the note section
Recently NetBSD starts to enforce this, and refuses to execute
the program if n is larger than the sum of entry sizes.

Before:
$ readelf -n ../bin/go.old
Notes at offset 0x00000bd0 with length 0x00000019:
  Owner         Data size       Description
  NetBSD                0x00000004      NT_VERSION (version)
readelf: Warning: corrupt note found at offset 18 into core notes
readelf: Warning:  type: 0, namesize: 00000000, descsize: 00000000
$ readelf -n ../bin/go
Notes at offset 0x00000bd0 with length 0x00000018:
Owner         Data size       Description
NetBSD                0x00000004      NT_VERSION (version)

LGTM=iant
R=iant
CC=golang-codereviews
https://codereview.appspot.com/70710043

Committer: Russ Cox <rsc@golang.org>

[gobot, 2014-03-06 05:19:58]

This CL appears to have broken the plan9-386-cnielsen builder.

[minux1, 2014-03-06 05:23:19]

On Thu, Mar 6, 2014 at 12:19 AM, <gobot@golang.org> wrote:

> This CL appears to have broken the plan9-386-cnielsen builder.
>
# ../misc/goplay

# ../test/bench/go1
ok   _/tmp/gobuilder/plan9-386-cnielsen-db613ec83949/go/test/bench/go1
16.145s

# ../test
Build complete, duration 1h0m0.125292928s. Result: error: timed out after
1h0m0s


Shouldn't be related to this change.