Currently, this change merely addresses where ex_data indexes are stored
and managed, and thus fixes the thread-safety issues that existed at that
level. "Class" code (eg. RSA, DSA, etc) no longer store their own STACKS
and per-class index counters - all such data is stored inside ex_data.c. So
rather than passing both STACK+counter to index-management ex_data
functions, a 'class_index' is instead passed to indicate the class (eg.
CRYPTO_EX_INDEX_RSA). New classes can be dynamically registered on-the-fly
and this is also thread-safe inside ex_data.c (though whether the caller
manages the return value in a thread-safe way is not addressed).
This does not change the "get/set" functions on individual "ex_data"
structures, and so thread-safety at that level isn't (yet) assured.
Likewise, the method of getting and storing per-class indexes has not
changed, so locking may still be required at the "caller" end but is
nonetheless thread-safe inside "ex_data"'s internal implementation.
Typically this occurs when code implements a new method of some kind and
stores its own per-class index in a global variable without locking the
setting and usage of that variable. If the code in question is likely to be
used in multiple threads, locking the setting and use of that index is
still up to the code in question. Possible fixes to this are being
sketched, but definitely require more major changes to the API itself than
this change undertakes.
The underlying implementation in ex_data.c has also been modularised so
that alternative "ex_data" implementations (that control all access to
state) can be plugged in. Eg. a loaded module can have its implementation
set to that of the application loaded it - the result being that
thread-safety and consistency of "ex_data" classes and indexes can be
maintained in the same place rather than the loaded module using its own
copy of ex_data support code and state.
Due to the centralisation of "state" with this change, cleanup of all
"ex_data" state can now be performed properly. Previously all allocation of
ex_data state was guaranteed to leak - and MemCheck_off() had been used to
avoid it flagging up the memory debugging. A new function has been added to
perfrom all this cleanup, CRYPTO_cleanup_all_ex_data(). The "openssl"
command(s) have been changed to use this cleanup, as have the relevant test
programs. External application code may want to do so too - failure to
cleanup will not induce more memory leaking than was the case before, but
the memory debugging is not tricked into hiding it any more so it may
"appear" where it previously did not.
like des_read_password and friends (backward compatibility functions
using this new API are provided). The purpose is to remove prompting
functions from the DES code section as well as provide for prompting
through dialog boxes in a window system and the like.
sure they are available in opensslconf.h, by giving them names starting
with "OPENSSL_" to avoid conflicts with other packages and by making
sure e_os2.h will cover all platform-specific cases together with
opensslconf.h.
I've checked fairly well that nothing breaks with this (apart from
external software that will adapt if they have used something like
NO_KRB5), but I can't guarantee it completely, so a review of this
change would be a good thing.
handling routines that need file name and line number information,
I've added a call level to our memory handling routines to allow that
kind of hooking.
Also, make sure empty slots of the dynamic lock stack are used.
Actually, I'm not really sure this is the right thing to do, and may
remove it, with an endlessly growing stack as result...
insecure, so a static lock is added to isolate the sensitive parts.
Also, to avoid one thread freeing a lock that is used by another, a
reference counter is added.
be needed in some ENGINE code, and might serve elsewhere as well.
Note that it's implemented in such a way that the locking itself is
done through the same CRYPTO_lock function as the static locks.
WARNING: This is currently experimental and untested code (it will get
tested soon, though :-)).
After some messing around this seems to work but needs
a few more tests. Working out the syntax for sk_set_cmp_func()
(cast it to a function that itself returns a function pointer)
was painful :-(
Needs some testing to see what other compilers think of this
syntax.
Also needs similar stuff for ASN1_SET_OF etc etc.
like Malloc, Realloc and especially Free conflict with already existing names
on some operating systems or other packages. That is reason enough to change
the names of the OpenSSL memory allocation macros to something that has a
better chance of being unique, like prepending them with OPENSSL_.
This change includes all the name changes needed throughout all C files.
"Jan Mikkelsen" <janm@transactionsite.com> correctly states that the
OpenSSL header files have #include's and extern "C"'s in an incorrect
order. Thusly fixed.
Also, make the memory debugging routines defined and declared with
prototypes, and use void* instead of char* for memory blobs.
And last of all, redo the ugly callback construct for elegance and
better definition (with prototypes).
variety of platforms. A few are missing, and they will be added in
eventually, but as this is new stuff, it was better to not break lots of
platforms in one go that we can't easily test. The changes to "Configure"
should illustrate how to add support to other systems if you feel like
having a go.
NB: I'll add something shortly to allow you to add "dlfcn.h" support on
those platforms that don't have (or need) a dlfcn.h header file. (The
symbol for Configure will probably by "dlfcn_no_h").
Thanks to Richard Levitte, who is responsible for the dso_dl.c support,
understanding the trickier aspects of the build process, and giving great
feedback on everything else.
[Don't use this stuff if you're easily offended by changes to the
interface or behaviour - it's still work in progress.]
PR:
also needed to get around C compilers on VMS that set the symbol limit
to 31 characters. Bot the macros VMS and __VMS are checked, since
there's no real way to know if e_os.h has been included yet.
(and that malloc can be called with an int argument).
- Use proper prototypes (with argument list) for various function pointers,
avoid casts (however there are still many such cases left in these files).
- Avoid collissions in app_info_cmp if sizeof int != sizeof long.
- Use CRYPTO_LOCK_MALLOC in mem_dbg.c.
- Made CRYPTO_MDEBUG even less used in crypto.h, giving
MemCheck_start() and MemCheck_stop() only one possible definition.
- Made the values of the debug function pointers in mem.c dependent
on the existence of the CRYPTO_MDEBUG macro, and made the rest of
the code understand the NULL case.
That's it. With this code, the old behvior of the debug functionality
is restored, but you can still opt to have it on, even when the
library wasn't compiled with a defined CRYPTO_MDEBUG.
- Moved the handling of compile-time defaults from crypto.h to
mem_dbg.c, since it doesn't make sense for the library users to try
to affect this without recompiling libcrypto.
- Made sure V_CRYPTO_MDEBUG_TIME and V_CRYPTO_MDEBUG_THREAD had clear
and constant definitions.
- Aesthetic correction.
With this change, the following is provided and present at all times
(meaning CRYPTO_MDEBUG is no longer required to get this functionality):
- hooks to provide your own allocation and deallocation routines.
They have to have the same interface as malloc(), realloc() and
free(). They are registered by calling CRYPTO_set_mem_functions()
with the function pointers.
- hooks to provide your own memory debugging routines. The have to
have the same interface as as the CRYPTO_dbg_*() routines. They
are registered by calling CRYPTO_set_mem_debug_functions() with
the function pointers.
I moved everything that was already built into OpenSSL and did memory
debugging to a separate file (mem_dbg.c), to make it clear what is
what.
With this, the relevance of the CRYPTO_MDEBUG has changed. The only
thing in crypto/crypto.h that it affects is the definition of the
MemCheck_start and MemCheck_stop macros.
1. Added code to the memory leak detecting code to give the user the
possibility to add information, thereby forming a traceback.
2. Make the memory leak detecting code multithread-safe.
The idea is that we're actually dealing with two separate critical
sections, one containing the hash tables with the information, the
other containing the current memory checking mode. Those should not
be handled with the same lock, especially since their handling overlap.
Hence, the added second lock.
thread-safe (where thread-safe counterparts are not available on all
platforms), and don't memcpy to NULL-pointers
Submitted by: Anonymous
Reviewed by: Bodo Moeller
Also, clean up htons vs. ntohs confusions.
While modifying the sources, I found some inconsistencies on the use of
s->cert vs. s->session->sess_cert; I don't know if those could
really have caused problems, but possibly this is a proper bug-fix
and not just a clean-up.
script, translates function codes better and doesn't need the K&R function
prototypes to work (NB. the K&R prototypes can't be wiped just yet: they are
still needed by the DEF generator...). I also ran the script with the -rewrite
option to update all the header and source files.
1. The already released version was 0.9.1c and not 0.9.1b
2. The next release should be 0.9.2 and not 0.9.1d, because
first the changes are already too large, second we should avoid any more
0.9.1x confusions and third, the Apache version semantics of
VERSION.REVISION.PATCHLEVEL for the version string is reasonable (and here
.2 is already just a patchlevel and not major change).
tVS: ----------------------------------------------------------------------
