IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
We explicitly call free() on a pool which falls to zero, assuming it's
not inside another pool (we crash). Check on creation and explicitly
document this case.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This neatens the code a bit (we should do a similar thing for all the
TALLOC_CHUNK macros).
Two subtler changes:
(1) As a result of the struct, we actually pack object_count into the
talloc header on 32-bit platforms (since the header is 40 bytes, but
needs to be 16-byte aligned).
(2) I avoid VALGRIND_MAKE_MEM_UNDEFINED on memmove when we resize the
only entry in a pool; that's done later anyway.
With -O2 on my 11.04 Ubuntu 32-bit x86 laptop, the talloc_pool speed as
measured by testsuite.c actually increases 10%.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Nested if's are hard to understand to me.
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Autobuild-User: Volker Lendecke <vlendec@samba.org>
Autobuild-Date: Mon Jan 2 19:07:23 CET 2012 on sn-devel-104
This ensures that the sibling list of a pointer doesn't become invalid
during a free operation. It is an alternative fix to the fix in
6f51a1f45b, and avoids the problem of
trying to calculate the parent pointer early
This should fix the subtle spoolss talloc bug that Simo found
Autobuild-User: Andrew Tridgell <tridge@samba.org>
Autobuild-Date: Tue Aug 9 01:53:17 CEST 2011 on sn-devel-104
Otherwise tc->name will end up pointing to garbage when it is not
set to a const but rather to a string allocate as child of the context itself.
Signed-off-by: Andrew Tridgell <tridge@samba.org>
*talloc_pool_objectcount(pool_tc) == 2 doesn't mean the one of the objects
is the pool itself! So we better check for == 1 and calculate the chunk count.
metze
The optimization of the object_count == 1 case should only happen
for when we're not destroying the pool itself. And it should only
happen if the pool itself is still valid.
If the pool isn't valid (it has TALLOC_FLAG_FREE),
object_count == 1 does not mean that the pool is the last object,
which can happen if you use talloc_steal/move() on memory
from the pool and then free the pool itself.
Thanks to Volker for noticing this!
metze
If we optimize on top of raw realloc() we need
TC_INVALIDATE_SHRINK_CHUNK together with TC_UNDEFINE_GROW_CHUNK
(with was missing and caused false positive valgrind warnings).
But that is really slow, as we do a lot of talloc_realloc calls in samba.
That's why we only to TC_UNDEFINE_SHRINK_CHUNK() for now.
metze
this is the minimal fix for the problem Rusty found. I previously
thought that the best fix would be to change tc->parent to be valid
for all pointers, but that is expensive for realloc with large numbers
of child pointers, which is much more commmon than I expected it to
be.
Autobuild-User: Andrew Tridgell <tridge@samba.org>
Autobuild-Date: Wed Jan 5 07:22:27 CET 2011 on sn-devel-104
when this environment variable is set, talloc will fill freed memory
with the value from that environment variable. This can be used to
help find use after free bugs when valgrind is too slow to be used
when a ptr has a single reference and a NULL parent, then
talloc_free(ptr) is not ambiguous, as the caller could not have done a
talloc_free(NULL) to free the memory
Pair-Programmed-With: Rusty Russell <rusty@samba.org>
If ALWASY_REALLOC is defined and we are to 'shrink' memory block,
memcpy() will write outside memory just allocated.
Signed-off-by: Andrew Tridgell <tridge@samba.org>
When we disable null tracking, we need to move any existing objects
that are under the null_context to be parented by the true NULL
context.
We also need a new talloc_enable_null_tracking_no_autofree() function,
as the talloc testsuite cannot cope with the moving of the autofree
context under the null_context as it wants to check exact counts of
objects under the null_context, and smbtorture has a large number of
objects in the autofree_context from .init functions
It's annoying when you use
p talloc_report_full(ctx, fopen("/tmp/xx","w"))
in gdb, and if you don't have write permission on the file then
you get a segv.
If NULL tracking is enabled after the autofree context is initialised
then autofree ends up separate from the null_context. This means that
talloc_report_full() doesn't report the autofree context. Fix this by
reparenting the autofree context when we create the null_context.
When we get a double free abort from talloc it is often hard to work
out where the first free came from. This patch takes advantage of the
fact that _talloc_free() now takes a location the free was called from
to allow the double free abort code to print the location of the first
free that conflicts.
We also use the major and minor versions in the TALLOC_MAGIC,
so that we can detect if two conflicting versions of talloc
are loaded in one process. In this case we use talloc_log() to
output a very useful debug message before we call
talloc_abort().
metze
