summaryrefslogtreecommitdiff
path: root/src/lib/libcrypto/bn (follow)
Commit message (Collapse)AuthorAgeFilesLines
...
* Refactor BN_mul().jsing2023-01-211-67/+81
| | | | | | | | | | | | | This splits BN_mul() into two parts, one of which is a separate bn_mul() function. This makes the code more readable and managable, while also providing a better entry point for assembly optimisation. A separate bn_mul() is provided for the BN_RECURSION implementation, to reduce complexity. This also enables bn_mul_comba4() for four word long bignums - this was disabled for unknown reasons. ok tb@
* Refactor BN_sqr().jsing2023-01-211-50/+80
| | | | | | | | This splits BN_sqr() into two parts, one of which is a separate bn_sqr() function. This makes the code more readable and managable, while also providing a better entry point for assembly optimisation. ok tb@
* Reorder functions and drop unnessary static prototypes.jsing2023-01-211-372/+363
| | | | No functional change.
* Remove various #ifndef BN_SQR_COMBA.jsing2023-01-201-19/+1
| | | | | | bn_sqr_comba{4,8}() is now always available. ok tb@
* Move bn_{mul,sqr}_comba{4,8}() from bn_asm.c to bn_mul.c/bn_sqr.c.jsing2023-01-208-306/+302
| | | | | | | Wrap these in HAVE_BN_{MUL,SQR}_COMBA{4,8} defines. Add these defines to bn_arch.h where the architecture currently provides its own version. ok tb@
* Move {mul,sqr}_add_c{,2} macros from bn_asm.c to bn_local.h.jsing2023-01-203-153/+160
| | | | | | | | | | | These depend on other macros that are in already in bn_local.h and this makes them available to other source files. A lot more clean up will be needed in the future. Of course x86_64-gcc.c makes use of the same macro names - sprinkle some undef in there for the time being. ok tb@
* Remove unused code.jsing2023-01-202-395/+13
| | | | ok tb@
* Replace BN_DIV3W with HAVE_BN_DIV_3_WORDS (in bn_arch.h).jsing2023-01-202-4/+7
| | | | ok tb@
* Provide a per machine bn_arch.h.jsing2023-01-2014-0/+336
| | | | | | | This will provide a location for machine specific defines, prototypes and inline functions. ok tb@
* Reorder functions.jsing2023-01-201-344/+344
| | | | No functional change.
* Reorder functions for easier maintenance.jsing2023-01-201-30/+30
| | | | No functional change.
* Remove more unused assembly generation scripts.jsing2023-01-198-696/+0
|
* Remove various unused assembly files and assembly generation scripts.jsing2023-01-196-6507/+0
| | | | | | These are just creating clutter and cause grep noise. ok miod@
* Bring in a description of bn_words_3_div().jsing2023-01-181-1/+9
| | | | | | This comes from OpenSSL commit 3da2e9c4ee45989a426ff513dc6c6250d1e460de. ok tb@
* Start cleaning up BN_div_internal().jsing2023-01-181-158/+195
| | | | | | | | | | Always provide a bn_div_3_words() function, rather than having deeply nested compiler conditionals. Use readable variable names, clean up formatting and use a single exit path. Tested on various platforms by miod@ ok tb@
* Move BN_sqr() to the bottom of the file.jsing2023-01-161-85/+85
| | | | | | This will simplify review/upcoming changes. No functional change.
* Mop up debug code that escaped previously.jsing2023-01-163-82/+3
| | | | This is the result of `unifdef -m -U BN_COUNT'.
* Move constants out of text segment into rodata to prepare for xonly supportderaadt2023-01-143-6/+2
| | | | | on amd64. no pic handling is neccessary since amd64 has full reach. ok kettenis
* Rewrite BN_CTX.jsing2023-01-141-410/+98
| | | | | | | | | | | | | | | | | | | | | The current BN_CTX implementation is an incredibly overengineered piece of code, which even includes its own debug system. Rewrite BN_CTX from scratch, simplifying things things considerably by having a "stack" of BIGNUM pointers and a matching array of group assignments. This means that BN_CTX_start() and BN_CTX_end() effectively do not fail. Unlike the previous implementation, if a failure occurs nothing will work and the BN_CTX must be freed/recreated, instead of trying to pick up at the point where the failure occurred (which does not make sense given its intended usage). Additionally, it has long been documented that BN_CTX_start() must be called before BN_CTX_get() can be used, however the previous implementation did not actually enforce this. Now that missing BN_CTX_start() and BN_CTX_end() calls have been added to DSA and EC, we can actually make this a hard requirement. ok tb@
* Greatly simplify bn_expand_internal().jsing2023-01-141-103/+26
| | | | | | | | We have a function called recallocarray() - make use of it rather than handrolling a version of it. Also have bn_expand() call bn_wexpand(), which avoids some duplication. ok tb@
* Move all data tables from .text section to .rodata, and update the code tomiod2023-01-132-4/+0
| | | | | | | fetch them correctly when building PIC. Also drop unused data, and remove --no-execute-only from linker flags. ok kettenis@
* Move all data tables from .text section to .rodata, and update the code tomiod2023-01-131-9/+1
| | | | | | | fetch them correctly when building PIC. Also drop unused data, and remove --no-execute-only from linker flags. ok jsing@ kettenis@
* Replace BN_lshift1()/BN_rshift1() with calls to BN_lshift()/BN_rshift().jsing2023-01-111-125/+42
| | | | | | | | | | | | | | | | | | | Currently, BN_lshift1() and BN_rshift1() are separate implementations that are intended to be faster since the shift is known (and only one bit crosses a word boundary). However, with the rewrite of BN_lshift() and BN_rshift(), they are either slower or only minimally faster (depending on architecture). Avoid duplication and turn BN_lshift1()/BN_rshift1() into functions that call inlined versions of BN_lshift()/BN_rshift(), making BN_lshift() and BN_rshift() call the same inlined implementation. This results in a single implementation and BN_lshift1()/BN_rshift1() that out perform the previous versions (in part due to compiler optimisation). Now that none of the original code exists, replace the license and copyright for this file. ok tb@
* Rewrite BN_lshift()jsing2023-01-101-26/+57
| | | | | | | | This improves readability and eliminates special handling for various cases, making the code cleaner and closer to constant time. Basic benchmarking shows a performance gain on modern 64 bit architectures. ok tb@
* Rewrite/simplify BN_free().jsing2023-01-071-10/+12
| | | | ok tb@
* Flip BN_clear_free() and BN_free()jsing2023-01-071-4/+4
| | | | | | | | All of our BIGNUMs are cleared when we free them - move the code to BN_free() and have BN_clear_free() call BN_free(), rather than the other way around. ok tb@
* Use calloc() in BN_new(), rather than malloc() and then manually zeroing.jsing2023-01-071-10/+7
| | | | ok tb@
* Rewrite BN_rshift()jsing2023-01-051-37/+42
| | | | | | | | | | | | This improves readability and eliminates special handling for various cases, making the code cleaner and closer to constant time. Basic benchmarking shows a performance gain on modern 64 bit architectures, while there is a decrease on legacy 32 bit architectures (i386), particularly for the zero bit shift case (which is now handled in the same code path). ok tb@
* spelling fixes; from paul tagliamontejmc2022-12-2616-29/+29
| | | | | | | i removed the arithmetics -> arithmetic changes, as i felt they were not clearly correct ok tb
* Simplify BN_cmp() and BN_ucmp().jsing2022-12-231-46/+15
| | | | | | | | | | The only real difference between BN_cmp() and BN_ucmp() is that one has to respect the sign of the BN (although BN_cmp() also gets to deal with some insanity from accepting NULLs). Rewrite/cleanup BN_ucmp() and turn BN_cmp() into code that handles differences in sign, before calling BN_ucmp(). ok tb@
* Provide BN_zero()/BN_one() as functions and make BN_zero() always succeed.jsing2022-12-173-7/+22
| | | | | | | | | | | | | | BN_zero() is currently implemented using BN_set_word(), which means it can fail, however almost nothing ever checks the return value. A long time ago OpenSSL changed BN_zero() to always succeed and return void, however kept BN_zero as a macro that calls a new BN_zero_ex() function, so that it can be switched back to the "can fail" version. Take a simpler approach - change BN_zero()/BN_one() to functions and make BN_zero() always succeed. This will be exposed in the next bump, at which point we can hopefully also remove the BN_zero_ex() function. ok tb@
* Update reference to table generationtb2022-12-011-2/+2
|
* BN_one() can fail, check its return value.jsing2022-12-011-4/+7
| | | | ok tb@
* Rewrite bn_correct_top().jsing2022-11-302-14/+10
| | | | | | | bn_correct_top() is currently a macro and far more complex than it needs to be - rewrite it as a function. ok tb@
* Fix return values bug in BN_ucmp().jsing2022-11-301-4/+6
| | | | | | | | BN_ucmp() is supposed to return -1/0/1 on a < b, a == b and a > b, however it currently returns other negative and positive values when the top of a and b differ. Correct this. ok tb@
* Mop up more BN_DEBUG related code.jsing2022-11-307-48/+7
|
* Make header guards of internal headers consistenttb2022-11-261-4/+4
| | | | | Not all of them, only those that didn't leak into a public header... Yes.
* bn_lcl.h wanted special treatment.tb2022-11-261-567/+0
|
* Make internal header file names consistenttb2022-11-2630-59/+626
| | | | | | | | | | | | | | | | Libcrypto currently has a mess of *_lcl.h, *_locl.h, and *_local.h names used for internal headers. Move all these headers we inherited from OpenSSL to *_local.h, reserving the name *_internal.h for our own code. Similarly, move dtls_locl.h and ssl_locl.h to dtls_local and ssl_local.h. constant_time_locl.h is moved to constant_time.h since it's special. Adjust all .c files in libcrypto, libssl and regress. The diff is mechanical with the exception of tls13_quic.c, where #include <ssl_locl.h> was fixed manually. discussed with jsing, no objection bcook
* Remove BIGNUM consistency macros.jsing2022-11-2623-328/+24
| | | | | | | | | | | | Compiling with BN_DEBUG (and if you want to take it further, BN_DEBUG_RAND) supposedly adds consistency checks to the BN code. These are rarely if ever used and introduce a bunch of clutter in the code. Furthermore, there are hacks in place to undo things that the debugging code does. Remove all of this mess and instead rely on always enabled checks, more readable code and proper regress coverage to ensure correct behaviour. "Good riddance." tb@
* Change bn_expand()/bn_wexpand() to indicate failure/success via 0/1.jsing2022-11-2412-67/+67
| | | | | | | | | Currently bn_expand()/bn_wexpand() return a BIGNUM *, however none of the callers use this (and many already treat it as a true/false value). Change these functions to return 0 on failure and 1 on success, revising callers that test against NULL in the process. ok tb@
* Use bn_wexpand() rather than bn_expand() with sizeof(BN_ULONG).jsing2022-11-232-4/+4
| | | | | | | This also fixes a bug in BN_MONT_CTX_set(), where the sizeof(BN_ULONG) in the call to bn_expand() was not multiplied by eight (to get bits). ok tb@
* Ensure that bn_expand()/bn_wexpand() fail on negative sizes.jsing2022-11-231-1/+7
| | | | ok tb@
* Turn bn_wexpand() into a function.jsing2022-11-232-5/+13
| | | | | | | | Any sensible compiler will likely inline this anyway (and even if it does not, one extra function call/return is the least of the performance overhead for this code). ok tb@
* Move bn_expand() under bn_expand2().jsing2022-11-231-13/+13
| | | | | | No functional change. ok tb@
* Remove unused bn_dup_expand().jsing2022-11-232-56/+2
| | | | ok tb@
* Move #ifndef OPENSSL_NO_DEPRECATED.jsing2022-11-231-21/+21
| | | | | | | The BN_set_params()/BN_get_params() and associated unused variables are meant to be in this block, not things like BN_new() and BN_free(). ok tb@
* Remove incorrect "r must not be a" commenttb2022-11-221-2/+1
| | | | | This was fixed by Eric A. Young in "a C2Net version of SSLeay" and committed to OpenSSL by Mark J. Cox in January 1999 (OpenSSL a0a54079).
* Fix segfaults in BN_dec2bn() and BN_hex2bn()tb2022-11-221-3/+3
| | | | | | | | | bn_print.c r1.29 added length checks to avoid overflowing the BIGNUM. If these checks are hit in length-only mode, i.e., bn is NULL, the error path dereferences bn. Change goto err to an early return to avoid this. ok jsing
* Fix a surprising quirk in BN_GF2m_mod(3).schwarze2022-11-201-5/+11
| | | | | | | | | | | | | | | | | | | | | | | | All other wrappers in the same file that use a temporary array of degrees size that array dynamically, such that they are able to handle reducing polynomials of arbitrary lengths. BN_GF2m_mod(3) was the only one that used a static array of size 6 instead, limiting it to trinomials and pentanomials and causing it to fail for longer reducing polynomials. Make this more uniform and less surprising by using exactly the same code as in all the other wrappers, such that BN_GF2m_mod(3) works with reducing polynomials of arbitrary length, too, just like the others. Again, tb@ points out this quirk is very unlikely to cause vulnerabilities in practice because cryptographic applications do not use longer reducing polynomials. This patch is not expected to significantly impact performance because the relevant caller, BN_GF2m_mod_div(3), already uses dynamic allocation via BN_GF2m_mod_mul(3). OK tb@
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-08-26 20:35:03 +0000