This is the mail archive of the
mailing list for the GDB project.
Re: [patch] Set bfd field in target_section
Pedro Alves wrote:
On Tuesday 28 July 2009 16:16:23, Aleksandar Ristovski wrote:
Pedro Alves wrote:
I think now you broke it in a different way. Now we can end
up trying to read from a closed bfd.
Ooops, you're right. There are uses of the bfd after transfering
its ownership to the bfd target with target_bfd_reopen, and closing
the bfd target (target_close)... This target is modeled on fdopen,
which is why I blindly assumed it wasn't happening. I've reverted
And just wondering, why not simply:
RCS file: /cvs/src/src/gdb/exec.c,v
retrieving revision 1.90
diff -u -p -r1.90 exec.c
--- gdb/exec.c 2 Jul 2009 17:21:06 -0000 1.90
+++ gdb/exec.c 28 Jul 2009 14:58:16 -0000
@@ -381,6 +381,7 @@ add_to_section_table (bfd *abfd, struct
struct target_section **table_pp = (struct
target_section **) table_pp_char;
+ (*table_pp)->bfd = abfd;
/* Check the section flags, but do not discard
zero-length sections, since
some symbols may still be attached to this section.
For instance, we
encountered on sparc-solaris 2.10 a shared library
with an empty .bss
@@ -390,7 +391,6 @@ add_to_section_table (bfd *abfd, struct
if (!(aflag & SEC_ALLOC))
- (*table_pp)->bfd = abfd;
(*table_pp)->the_bfd_section = asect;
(*table_pp)->addr = bfd_section_vma (abfd, asect);
(*table_pp)->endaddr = (*table_pp)->addr +
bfd_section_size (abfd, asect);
Still not right, because that's still papering over the problem, and
still hitting undefined behaviour, in the case of the
table ending up being empty. When table->sections == table->sections_end,
we should never dereference table->sections, like in 'table->sections->bfd'.
Notice how the end pointer isn't advanced if !SEC_ALLOC.
Consider the fact that we usually xmalloc enough memory that
makes that not misbehave a coincidence. E.g., see that with
a resize_section_table call, you can end up with table->sections == NULL,
and table->section_end == NULL, which is legal, because table->sections
is still equal to table->sections_end in that case (meaning empty table).
There used to be an extra structure in bfd_target.c to hold
the sections and the bfd, which didn't look necessary when I
reimplemented bfd-target.c on top of exec.c's functions, but
it is clear now that we need it anyway. Here's a patch that
brings something like that back.
Could you try it? Thanks.
I tried it, it works.
QNX Software Systems