Bug 511573

Setting to ASSIGNED per Fedora Bug Triage workflow.  https://fedoraproject.org/wiki/BugZappers/BugStatusWorkFlow

Created attachment 352540 [details]
root.log

root.log for i386

Created attachment 352541 [details]
build.log

build.log for i386

Created attachment 352542 [details]
mock.log

mock.log for i386

Created attachment 352543 [details]
root.log

root.log for x86_64

Created attachment 352544 [details]
build.log

build.log for x86_64

Created attachment 352545 [details]
mock.log

mock.log for x86_64

Hmm, __builtin_object_size() doesn't seem to allow for how gnupg allocates and uses "struct para_data_s" in g10/keygen.c.  Changing the way the structure is
laid out will work as a temporary fix, but we should probably ask the toolchain folks about it, too.

Created attachment 354021 [details]
cut-down reproducer

Created attachment 354030 [details]
repro.i

preprocessed version of repro.c (attachment #354021 [details])

Jakub, I boiled down the code in gnupg that crashes during the test suite down to attachment #354030 [details].  Building it with -O2 using gcc-4.4.0-12 triggers __fortify_fail on my x86_64 system.

Do you know if this is something that gcc needs to be allowing for, or should I be looking at reworking the structure definition in gnupg?  I'd appreciate any advice you have.  Thanks!

Yes, this is considered invalid for str*/stp* functions with -D_FORTIFY_SOURCE=2 (valid for -D_FORTIFY_SOURCE=1 and -U_FORTIFY_SOURCE).
Either use memcpy instead of strcpy (memcpy accross field boundaries is valid even with -D_FORTIFY_SOURCE=2, memcpy (l->u.value, "DSA", sizeof "DSA")), or rework the data structures.  Say:
struct list_item {
 struct list_item *next;
 int blah;
};
union list_item_union {
 struct list_item header;
 struct list_item_foo { struct list_item header; foo *foo; } foo;
 struct list_item_bar { struct list_item header; bar *bar; } bar;
 struct list_item_baz { struct list_item header; uint32_t baz; } baz;
 struct list_item_value { struct list_item header; char value[1]; } value;
};
should work (or obviously you can have union for the stuff you don't abuse as wannabe flexible array member).

Is there a reference for what's valid and what isn't?  I've found mention of some things that aren't valid in assorted bug reports and mailing list archives, but haven't found a full list to check against.  I'd like to avoid proposing a change upstream that happens to be accepted now but later turns out to be problematic.

Actually, now that I got another similar report (#512689), after some discussions on upstream IRC I'm considering allowing this:
http://gcc.gnu.org/ml/gcc-patches/2009-07/msg01136.html

Thanks, Jakub!  The scratch build now succeeds with 4.4.0-15, so I'll mark this as fixed in Raw Hide.