mirror of
https://github.com/hashcat/hashcat.git
synced 2024-11-21 15:48:08 +00:00
Fixed cleanup of password candidate buffers on GPU set from autotune in case -n was used
This commit is contained in:
Jens Steube
parent
1dc3469f6e
commit
9b64a405d1
@ -77,8 +77,9 @@
|
||||
|
||||
- Fixed buffer overflow in build_plain() function
|
||||
- Fixed buffer overflow in mp_add_cs_buf() function
|
||||
- Fixed copy/paste error leading to invalid "Integer overflow detected in keyspace of mask" in attack-mode 6 and 7
|
||||
- Fixed calculation of brain-session ID, only the first hash of the hashset was taken into account
|
||||
- Fixed cleanup of password candidate buffers on GPU set from autotune in case -n was used
|
||||
- Fixed copy/paste error leading to invalid "Integer overflow detected in keyspace of mask" in attack-mode 6 and 7
|
||||
- Fixed cracking multiple Office hashes (modes 9500, 9600) with the same salt
|
||||
- Fixed cracking of Blockchain, My Wallet (V1 and V2) hashes with unexpected decrypted data
|
||||
- Fixed cracking of Cisco-PIX and Cisco-ASA MD5 passwords in mask-attack mode if mask > length 16
|
||||
@ -100,9 +101,9 @@
|
||||
- Fixed race condition in maskfile mode by using a dedicated flag for restore execution
|
||||
- Fixed some memory leaks in case hashcat is shutting down due to some file error
|
||||
- Fixed some memory leaks in case mask-files are used in optimized mode
|
||||
- Fixed --status-json to correctly escape certain characters in hashes
|
||||
- Fixed the 7-Zip parser to allow the entire supported range of encrypted and decrypted data lengths
|
||||
- Fixed the validation of the --brain-client-features command line argument (only values 1, 2 or 3 are allowed)
|
||||
- Fixed --status-json to correctly escape certain characters in hashes
|
||||
|
||||
##
|
||||
## Improvements
|
||||
|
||||
398
src/autotune.c
398
src/autotune.c
@ -136,233 +136,227 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
device_param->kernel_accel = kernel_accel;
|
||||
device_param->kernel_loops = kernel_loops;
|
||||
|
||||
const u32 kernel_power = device_param->hardware_power * device_param->kernel_accel;
|
||||
|
||||
device_param->kernel_power = kernel_power;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
// from here it's clear we are allowed to autotune
|
||||
// so let's init some fake words
|
||||
|
||||
const u32 kernel_power_max = device_param->hardware_power * kernel_accel_max;
|
||||
|
||||
int CL_rc;
|
||||
int CU_rc;
|
||||
|
||||
if (device_param->is_cuda == true)
|
||||
{
|
||||
CU_rc = run_cuda_kernel_atinit (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, kernel_power_max);
|
||||
|
||||
if (CU_rc == -1) return -1;
|
||||
}
|
||||
|
||||
if (device_param->is_opencl == true)
|
||||
{
|
||||
CL_rc = run_opencl_kernel_atinit (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, kernel_power_max);
|
||||
|
||||
if (CL_rc == -1) return -1;
|
||||
}
|
||||
|
||||
if (user_options->slow_candidates == true)
|
||||
{
|
||||
}
|
||||
else
|
||||
{
|
||||
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
|
||||
// from here it's clear we are allowed to autotune
|
||||
// so let's init some fake words
|
||||
|
||||
const u32 kernel_power_max = device_param->hardware_power * kernel_accel_max;
|
||||
|
||||
int CL_rc;
|
||||
int CU_rc;
|
||||
|
||||
if (device_param->is_cuda == true)
|
||||
{
|
||||
if (straight_ctx->kernel_rules_cnt > 1)
|
||||
{
|
||||
if (device_param->is_cuda == true)
|
||||
{
|
||||
CU_rc = hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t));
|
||||
CU_rc = run_cuda_kernel_atinit (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, kernel_power_max);
|
||||
|
||||
if (CU_rc == -1) return -1;
|
||||
}
|
||||
|
||||
if (device_param->is_opencl == true)
|
||||
{
|
||||
CL_rc = hc_clEnqueueCopyBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, device_param->opencl_d_rules_c, 0, 0, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), 0, NULL, NULL);
|
||||
|
||||
if (CL_rc == -1) return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Do a pre-autotune test run to find out if kernel runtime is above some TDR limit
|
||||
|
||||
u32 kernel_loops_max_reduced = kernel_loops_max;
|
||||
|
||||
if (true)
|
||||
{
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
|
||||
|
||||
if (exec_msec > 2000)
|
||||
{
|
||||
event_log_error (hashcat_ctx, "Kernel minimum runtime larger than default TDR");
|
||||
|
||||
return -1;
|
||||
if (CU_rc == -1) return -1;
|
||||
}
|
||||
|
||||
exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
|
||||
|
||||
const u32 mm = kernel_loops_max / kernel_loops_min;
|
||||
|
||||
if ((exec_msec * mm) > target_msec)
|
||||
if (device_param->is_opencl == true)
|
||||
{
|
||||
const u32 loops_valid = (const u32) (target_msec / exec_msec);
|
||||
CL_rc = run_opencl_kernel_atinit (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, kernel_power_max);
|
||||
|
||||
kernel_loops_max_reduced = kernel_loops_min * loops_valid;
|
||||
if (CL_rc == -1) return -1;
|
||||
}
|
||||
}
|
||||
|
||||
// first find out highest kernel-loops that stays below target_msec
|
||||
|
||||
if (kernel_loops_min < kernel_loops_max)
|
||||
{
|
||||
for (kernel_loops = kernel_loops_max; kernel_loops > kernel_loops_min; kernel_loops >>= 1)
|
||||
if (user_options->slow_candidates == true)
|
||||
{
|
||||
if (kernel_loops > kernel_loops_max_reduced) continue;
|
||||
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops);
|
||||
|
||||
if (exec_msec < target_msec) break;
|
||||
}
|
||||
}
|
||||
|
||||
// now the same for kernel-accel but with the new kernel-loops from previous loop set
|
||||
|
||||
#define STEPS_CNT 16
|
||||
|
||||
if (kernel_accel_min < kernel_accel_max)
|
||||
{
|
||||
for (int i = 0; i < STEPS_CNT; i++)
|
||||
{
|
||||
const u32 kernel_accel_try = 1U << i;
|
||||
|
||||
if (kernel_accel_try < kernel_accel_min) continue;
|
||||
if (kernel_accel_try > kernel_accel_max) break;
|
||||
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops);
|
||||
|
||||
if (exec_msec > target_msec) break;
|
||||
|
||||
kernel_accel = kernel_accel_try;
|
||||
}
|
||||
}
|
||||
|
||||
// now find the middle balance between kernel_accel and kernel_loops
|
||||
// while respecting allowed ranges at the same time
|
||||
|
||||
if (kernel_accel < kernel_loops)
|
||||
{
|
||||
const u32 kernel_accel_orig = kernel_accel;
|
||||
const u32 kernel_loops_orig = kernel_loops;
|
||||
|
||||
double exec_msec_prev = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
for (int i = 1; i < STEPS_CNT; i++)
|
||||
{
|
||||
const u32 kernel_accel_try = kernel_accel_orig * (1U << i);
|
||||
const u32 kernel_loops_try = kernel_loops_orig / (1U << i);
|
||||
|
||||
if (kernel_accel_try < kernel_accel_min) continue;
|
||||
if (kernel_accel_try > kernel_accel_max) break;
|
||||
|
||||
if (kernel_loops_try > kernel_loops_max) continue;
|
||||
if (kernel_loops_try < kernel_loops_min) break;
|
||||
|
||||
// do a real test
|
||||
|
||||
const double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops_try);
|
||||
|
||||
if (exec_msec_prev < exec_msec) break;
|
||||
|
||||
exec_msec_prev = exec_msec;
|
||||
|
||||
// so far, so good! save
|
||||
|
||||
kernel_accel = kernel_accel_try;
|
||||
kernel_loops = kernel_loops_try;
|
||||
|
||||
// too much if the next test is true
|
||||
|
||||
if (kernel_loops_try < kernel_accel_try) break;
|
||||
}
|
||||
}
|
||||
|
||||
double exec_msec_pre_final = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
const u32 exec_left = (const u32) (target_msec / exec_msec_pre_final);
|
||||
|
||||
const u32 accel_left = kernel_accel_max / kernel_accel;
|
||||
|
||||
const u32 exec_accel_min = MIN (exec_left, accel_left); // we want that to be int
|
||||
|
||||
if (exec_accel_min >= 1)
|
||||
{
|
||||
// this is safe to not overflow kernel_accel_max because of accel_left
|
||||
|
||||
kernel_accel *= exec_accel_min;
|
||||
}
|
||||
|
||||
// start finding best thread count is easier.
|
||||
// it's either the preferred or the maximum thread count
|
||||
|
||||
/*
|
||||
const u32 kernel_threads_min = device_param->kernel_threads_min;
|
||||
const u32 kernel_threads_max = device_param->kernel_threads_max;
|
||||
|
||||
if (kernel_threads_min < kernel_threads_max)
|
||||
{
|
||||
const double exec_msec_max = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
u32 preferred_threads = 0;
|
||||
|
||||
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
|
||||
{
|
||||
if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple1;
|
||||
}
|
||||
else
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple4;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple2;
|
||||
}
|
||||
|
||||
if ((preferred_threads >= kernel_threads_min) && (preferred_threads <= kernel_threads_max))
|
||||
{
|
||||
const double exec_msec_preferred = try_run_preferred (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
if (exec_msec_preferred < exec_msec_max)
|
||||
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
|
||||
{
|
||||
device_param->kernel_threads = preferred_threads;
|
||||
if (straight_ctx->kernel_rules_cnt > 1)
|
||||
{
|
||||
if (device_param->is_cuda == true)
|
||||
{
|
||||
CU_rc = hc_cuMemcpyDtoD (hashcat_ctx, device_param->cuda_d_rules_c, device_param->cuda_d_rules, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t));
|
||||
|
||||
if (CU_rc == -1) return -1;
|
||||
}
|
||||
|
||||
if (device_param->is_opencl == true)
|
||||
{
|
||||
CL_rc = hc_clEnqueueCopyBuffer (hashcat_ctx, device_param->opencl_command_queue, device_param->opencl_d_rules, device_param->opencl_d_rules_c, 0, 0, MIN (kernel_loops_max, KERNEL_RULES) * sizeof (kernel_rule_t), 0, NULL, NULL);
|
||||
|
||||
if (CL_rc == -1) return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Do a pre-autotune test run to find out if kernel runtime is above some TDR limit
|
||||
|
||||
u32 kernel_loops_max_reduced = kernel_loops_max;
|
||||
|
||||
if (true)
|
||||
{
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
|
||||
|
||||
if (exec_msec > 2000)
|
||||
{
|
||||
event_log_error (hashcat_ctx, "Kernel minimum runtime larger than default TDR");
|
||||
|
||||
return -1;
|
||||
}
|
||||
|
||||
exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops_min);
|
||||
|
||||
const u32 mm = kernel_loops_max / kernel_loops_min;
|
||||
|
||||
if ((exec_msec * mm) > target_msec)
|
||||
{
|
||||
const u32 loops_valid = (const u32) (target_msec / exec_msec);
|
||||
|
||||
kernel_loops_max_reduced = kernel_loops_min * loops_valid;
|
||||
}
|
||||
}
|
||||
|
||||
// first find out highest kernel-loops that stays below target_msec
|
||||
|
||||
if (kernel_loops_min < kernel_loops_max)
|
||||
{
|
||||
for (kernel_loops = kernel_loops_max; kernel_loops > kernel_loops_min; kernel_loops >>= 1)
|
||||
{
|
||||
if (kernel_loops > kernel_loops_max_reduced) continue;
|
||||
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_min, kernel_loops);
|
||||
|
||||
if (exec_msec < target_msec) break;
|
||||
}
|
||||
}
|
||||
|
||||
// now the same for kernel-accel but with the new kernel-loops from previous loop set
|
||||
|
||||
#define STEPS_CNT 16
|
||||
|
||||
if (kernel_accel_min < kernel_accel_max)
|
||||
{
|
||||
for (int i = 0; i < STEPS_CNT; i++)
|
||||
{
|
||||
const u32 kernel_accel_try = 1U << i;
|
||||
|
||||
if (kernel_accel_try < kernel_accel_min) continue;
|
||||
if (kernel_accel_try > kernel_accel_max) break;
|
||||
|
||||
double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops);
|
||||
|
||||
if (exec_msec > target_msec) break;
|
||||
|
||||
kernel_accel = kernel_accel_try;
|
||||
}
|
||||
}
|
||||
|
||||
// now find the middle balance between kernel_accel and kernel_loops
|
||||
// while respecting allowed ranges at the same time
|
||||
|
||||
if (kernel_accel < kernel_loops)
|
||||
{
|
||||
const u32 kernel_accel_orig = kernel_accel;
|
||||
const u32 kernel_loops_orig = kernel_loops;
|
||||
|
||||
double exec_msec_prev = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
for (int i = 1; i < STEPS_CNT; i++)
|
||||
{
|
||||
const u32 kernel_accel_try = kernel_accel_orig * (1U << i);
|
||||
const u32 kernel_loops_try = kernel_loops_orig / (1U << i);
|
||||
|
||||
if (kernel_accel_try < kernel_accel_min) continue;
|
||||
if (kernel_accel_try > kernel_accel_max) break;
|
||||
|
||||
if (kernel_loops_try > kernel_loops_max) continue;
|
||||
if (kernel_loops_try < kernel_loops_min) break;
|
||||
|
||||
// do a real test
|
||||
|
||||
const double exec_msec = try_run (hashcat_ctx, device_param, kernel_accel_try, kernel_loops_try);
|
||||
|
||||
if (exec_msec_prev < exec_msec) break;
|
||||
|
||||
exec_msec_prev = exec_msec;
|
||||
|
||||
// so far, so good! save
|
||||
|
||||
kernel_accel = kernel_accel_try;
|
||||
kernel_loops = kernel_loops_try;
|
||||
|
||||
// too much if the next test is true
|
||||
|
||||
if (kernel_loops_try < kernel_accel_try) break;
|
||||
}
|
||||
}
|
||||
|
||||
double exec_msec_pre_final = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
const u32 exec_left = (const u32) (target_msec / exec_msec_pre_final);
|
||||
|
||||
const u32 accel_left = kernel_accel_max / kernel_accel;
|
||||
|
||||
const u32 exec_accel_min = MIN (exec_left, accel_left); // we want that to be int
|
||||
|
||||
if (exec_accel_min >= 1)
|
||||
{
|
||||
// this is safe to not overflow kernel_accel_max because of accel_left
|
||||
|
||||
kernel_accel *= exec_accel_min;
|
||||
}
|
||||
|
||||
// start finding best thread count is easier.
|
||||
// it's either the preferred or the maximum thread count
|
||||
|
||||
/*
|
||||
const u32 kernel_threads_min = device_param->kernel_threads_min;
|
||||
const u32 kernel_threads_max = device_param->kernel_threads_max;
|
||||
|
||||
if (kernel_threads_min < kernel_threads_max)
|
||||
{
|
||||
const double exec_msec_max = try_run (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
u32 preferred_threads = 0;
|
||||
|
||||
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
|
||||
{
|
||||
if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple1;
|
||||
}
|
||||
else
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple4;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
preferred_threads = device_param->kernel_preferred_wgs_multiple2;
|
||||
}
|
||||
|
||||
if ((preferred_threads >= kernel_threads_min) && (preferred_threads <= kernel_threads_max))
|
||||
{
|
||||
const double exec_msec_preferred = try_run_preferred (hashcat_ctx, device_param, kernel_accel, kernel_loops);
|
||||
|
||||
if (exec_msec_preferred < exec_msec_max)
|
||||
{
|
||||
device_param->kernel_threads = preferred_threads;
|
||||
}
|
||||
}
|
||||
}
|
||||
*/
|
||||
}
|
||||
*/
|
||||
|
||||
// reset them fake words
|
||||
// reset other buffers in case autotune cracked something
|
||||
|
||||
if (device_param->is_cuda == true)
|
||||
{
|
||||
// reset them fake words
|
||||
int CU_rc;
|
||||
|
||||
CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_pws_buf, 0, device_param->size_pws);
|
||||
|
||||
if (CU_rc == -1) return -1;
|
||||
|
||||
// reset other buffers in case autotune cracked something
|
||||
|
||||
CU_rc = run_cuda_kernel_memset (hashcat_ctx, device_param, device_param->cuda_d_plain_bufs, 0, device_param->size_plains);
|
||||
|
||||
if (CU_rc == -1) return -1;
|
||||
@ -378,14 +372,12 @@ static int autotune (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param
|
||||
|
||||
if (device_param->is_opencl == true)
|
||||
{
|
||||
// reset them fake words
|
||||
int CL_rc;
|
||||
|
||||
CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_pws_buf, 0, device_param->size_pws);
|
||||
|
||||
if (CL_rc == -1) return -1;
|
||||
|
||||
// reset other buffers in case autotune cracked something
|
||||
|
||||
CL_rc = run_opencl_kernel_memset (hashcat_ctx, device_param, device_param->opencl_d_plain_bufs, 0, device_param->size_plains);
|
||||
|
||||
if (CL_rc == -1) return -1;
|
||||
|
||||
Loading…
Reference in New Issue
Block a user