Commit 3d18a37c authored by Jonathan Peyton's avatar Jonathan Peyton
Browse files

[OpenMP] Fix race condition in omp_init_lock 

This is a partial fix for bug 34050.

This prevents callers of omp_set_lock (which does not hold __kmp_global_lock)
from ever seeing an uninitialized version of __kmp_i_lock_table.table.

It does not solve a use-after-free race condition if omp_set_lock obtains a
pointer to __kmp_i_lock_table.table before it is updated and then attempts to
dereference afterwards. That race is far less likely and can be handled in a
separate patch.

The unit test usually segfaults on the current trunk revision. It passes with
the patch.

Patch by Adam Azarchs

Differential Revision: https://reviews.llvm.org/D39439

llvm-svn: 317115
parent 435151ad

openmp/runtime/src/kmp_lock.cpp

+4 −3
Original line number Diff line number Diff line
@@ -3058,11 +3058,12 @@ kmp_indirect_lock_t *__kmp_allocate_indirect_lock(void **user_lock, 
    if (idx == __kmp_i_lock_table.size) {

      // Double up the space for block pointers

      int row = __kmp_i_lock_table.size / KMP_I_LOCK_CHUNK;

      kmp_indirect_lock_t **old_table = __kmp_i_lock_table.table;

      __kmp_i_lock_table.table = (kmp_indirect_lock_t **)__kmp_allocate(

      kmp_indirect_lock_t **new_table = (kmp_indirect_lock_t **)__kmp_allocate(

          2 * row * sizeof(kmp_indirect_lock_t *));

      KMP_MEMCPY(__kmp_i_lock_table.table, old_table,

      KMP_MEMCPY(new_table, __kmp_i_lock_table.table,

                 row * sizeof(kmp_indirect_lock_t *));

      kmp_indirect_lock_t **old_table = __kmp_i_lock_table.table;

      __kmp_i_lock_table.table = new_table;

      __kmp_free(old_table);

      // Allocate new objects in the new blocks

      for (int i = row; i < 2 * row; ++i)

openmp/runtime/test/lock/omp_init_lock.c

0 → 100644
+42 −0
Original line number Diff line number Diff line 
// RUN: %libomp-compile-and-run

#include "omp_testsuite.h"

#include <stdio.h>



// This should be slightly less than KMP_I_LOCK_CHUNK, which is 1024

#define LOCKS_PER_ITER 1000

#define ITERATIONS (REPETITIONS + 1)



// This tests concurrently using locks on one thread while initializing new

// ones on another thread.  This exercises the global lock pool.

int test_omp_init_lock() {

  int i;

  omp_lock_t lcks[ITERATIONS * LOCKS_PER_ITER];

#pragma omp parallel for schedule(static) num_threads(NUM_TASKS)

  for (i = 0; i < ITERATIONS; i++) {

    int j;

    omp_lock_t *my_lcks = &lcks[i * LOCKS_PER_ITER];

    for (j = 0; j < LOCKS_PER_ITER; j++) {

      omp_init_lock(&my_lcks[j]);

    }

    for (j = 0; j < LOCKS_PER_ITER * 100; j++) {

      omp_set_lock(&my_lcks[j % LOCKS_PER_ITER]);

      omp_unset_lock(&my_lcks[j % LOCKS_PER_ITER]);

    }

  }

  // Wait until all repititions are done.  The test is exercising growth of

  // the global lock pool, which does not shrink when no locks are allocated.

  {

    int j;

    for (j = 0; j < ITERATIONS * LOCKS_PER_ITER; j++) {

      omp_destroy_lock(&lcks[j]);

    }

  }



  return 0;

}



int main() {

  // No use repeating this test, since it's exercising a private global pool

  // which is not reset between test iterations.

  return test_omp_init_lock();

}