Make one call to prof_active_get_unlocked() per allocation event.

Make one call to prof_active_get_unlocked() per allocation event, and
use the result throughout the relevant functions that handle an
allocation event.  Also add a missing check in prof_realloc().  These
fixes protect allocation events against concurrent prof_active changes.

1 files changed, 10 insertions, 8 deletions

diff --git a/include/jemalloc/internal/prof.h b/include/jemalloc/internal/prof.h
index eca8aa8..c66611c 100644
--- a/include/jemalloc/internal/prof.h
+++ b/include/jemalloc/internal/prof.h
@@ -331,14 +331,16 @@ bool	prof_gdump_get_unlocked(void);
 prof_tdata_t	*prof_tdata_get(tsd_t *tsd, bool create);
 bool	prof_sample_accum_update(tsd_t *tsd, size_t usize, bool commit,
     prof_tdata_t **tdata_out);
-prof_tctx_t	*prof_alloc_prep(tsd_t *tsd, size_t usize, bool update);
+prof_tctx_t	*prof_alloc_prep(tsd_t *tsd, size_t usize, bool prof_active,
+    bool update);
 prof_tctx_t	*prof_tctx_get(const void *ptr);
 void	prof_tctx_set(const void *ptr, size_t usize, prof_tctx_t *tctx);
 void	prof_malloc_sample_object(const void *ptr, size_t usize,
     prof_tctx_t *tctx);
 void	prof_malloc(const void *ptr, size_t usize, prof_tctx_t *tctx);
 void	prof_realloc(tsd_t *tsd, const void *ptr, size_t usize,
-    prof_tctx_t *tctx, bool updated, size_t old_usize, prof_tctx_t *old_tctx);
+    prof_tctx_t *tctx, bool prof_active, bool updated,
+    size_t old_usize, prof_tctx_t *old_tctx);
 void	prof_free(tsd_t *tsd, const void *ptr, size_t usize);
 #endif
 
@@ -443,7 +445,7 @@ prof_sample_accum_update(tsd_t *tsd, size_t usize, bool update,
 }
 
 JEMALLOC_ALWAYS_INLINE prof_tctx_t *
-prof_alloc_prep(tsd_t *tsd, size_t usize, bool update)
+prof_alloc_prep(tsd_t *tsd, size_t usize, bool prof_active, bool update)
 {
 	prof_tctx_t *ret;
 	prof_tdata_t *tdata;
@@ -451,8 +453,8 @@ prof_alloc_prep(tsd_t *tsd, size_t usize, bool update)
 
 	assert(usize == s2u(usize));
 
-	if (!prof_active_get_unlocked() || likely(prof_sample_accum_update(tsd,
-	    usize, update, &tdata)))
+	if (!prof_active || likely(prof_sample_accum_update(tsd, usize, update,
+	    &tdata)))
 		ret = (prof_tctx_t *)(uintptr_t)1U;
 	else {
 		bt_init(&bt, tdata->vec);
@@ -479,17 +481,17 @@ prof_malloc(const void *ptr, size_t usize, prof_tctx_t *tctx)
 
 JEMALLOC_ALWAYS_INLINE void
 prof_realloc(tsd_t *tsd, const void *ptr, size_t usize, prof_tctx_t *tctx,
-    bool updated, size_t old_usize, prof_tctx_t *old_tctx)
+    bool prof_active, bool updated, size_t old_usize, prof_tctx_t *old_tctx)
 {
 
 	cassert(config_prof);
 	assert(ptr != NULL || (uintptr_t)tctx <= (uintptr_t)1U);
 
-	if (!updated && ptr != NULL) {
+	if (prof_active && !updated && ptr != NULL) {
 		assert(usize == isalloc(ptr, true));
 		if (prof_sample_accum_update(tsd, usize, true, NULL)) {
 			/*
-			 * Don't sample.  The usize passed to PROF_ALLOC_PREP()
+			 * Don't sample.  The usize passed to prof_alloc_prep()
 			 * was larger than what actually got allocated, so a
 			 * backtrace was captured for this allocation, even
 			 * though its actual usize was insufficient to cross the