Fix potential TLS-related memory corruption.

Avoid writing to uninitialized TLS as a side effect of deallocation.
Initializing TLS during deallocation is unsafe because it is possible
that a thread never did any allocation, and that TLS has already been
deallocated by the threads library, resulting in write-after-free
corruption.  These fixes affect prof_tdata and quarantine; all other
uses of TLS are already safe, whether intentionally (as for tcache) or
unintentionally (as for arenas).

src/jemalloc.c

@@ -282,12 +282,30 @@ arenas_cleanup(void *arg)
 	malloc_mutex_unlock(&arenas_lock);
 }
 
+static JEMALLOC_ATTR(always_inline) void
+malloc_thread_init(void)
+{
+
+	/*
+	 * TSD initialization can't be safely done as a side effect of
+	 * deallocation, because it is possible for a thread to do nothing but
+	 * deallocate its TLS data via free(), in which case writing to TLS
+	 * would cause write-after-free memory corruption.  The quarantine
+	 * facility *only* gets used as a side effect of deallocation, so make
+	 * a best effort attempt at initializing its TSD by hooking all
+	 * allocation events.
+	 */
+	if (config_fill && opt_quarantine)
+		quarantine_alloc_hook();
+}
+
 static JEMALLOC_ATTR(always_inline) bool
 malloc_init(void)
 {
 
-	if (malloc_initialized == false)
-		return (malloc_init_hard());
+	if (malloc_initialized == false && malloc_init_hard())
+		return (true);
+	malloc_thread_init();
 
 	return (false);
 }
@@ -1095,6 +1113,7 @@ je_realloc(void *ptr, size_t size)
 	if (size == 0) {
 		if (ptr != NULL) {
 			/* realloc(ptr, 0) is equivalent to free(p). */
+			assert(malloc_initialized || IS_INITIALIZER);
 			if (config_prof) {
 				old_size = isalloc(ptr, true);
 				if (config_valgrind && opt_valgrind)
@@ -1120,6 +1139,7 @@ je_realloc(void *ptr, size_t size)
 
 	if (ptr != NULL) {
 		assert(malloc_initialized || IS_INITIALIZER);
+		malloc_thread_init();
 
 		if (config_prof) {
 			old_size = isalloc(ptr, true);
@@ -1323,6 +1343,7 @@ je_malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *ptr)
 	size_t ret;
 
 	assert(malloc_initialized || IS_INITIALIZER);
+	malloc_thread_init();
 
 	if (config_ivsalloc)
 		ret = ivsalloc(ptr, config_prof);
@@ -1497,6 +1518,7 @@ je_rallocm(void **ptr, size_t *rsize, size_t size, size_t extra, int flags)
 	assert(size != 0);
 	assert(SIZE_T_MAX - size >= extra);
 	assert(malloc_initialized || IS_INITIALIZER);
+	malloc_thread_init();
 
 	if (arena_ind != UINT_MAX) {
 		arena_chunk_t *chunk;
@@ -1611,6 +1633,7 @@ je_sallocm(const void *ptr, size_t *rsize, int flags)
 	size_t sz;
 
 	assert(malloc_initialized || IS_INITIALIZER);
+	malloc_thread_init();
 
 	if (config_ivsalloc)
 		sz = ivsalloc(ptr, config_prof);

src/prof.c

@@ -438,7 +438,7 @@ prof_lookup(prof_bt_t *bt)
 
 	cassert(config_prof);
 
-	prof_tdata = prof_tdata_get();
+	prof_tdata = prof_tdata_get(false);
 	if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
 		return (NULL);
 
@@ -684,7 +684,7 @@ prof_ctx_destroy(prof_ctx_t *ctx)
 	 * avoid a race between the main body of prof_ctx_merge() and entry
 	 * into this function.
 	 */
-	prof_tdata = *prof_tdata_tsd_get();
+	prof_tdata = prof_tdata_get(false);
 	assert((uintptr_t)prof_tdata > (uintptr_t)PROF_TDATA_STATE_MAX);
 	prof_enter(prof_tdata);
 	malloc_mutex_lock(ctx->lock);
@@ -844,7 +844,7 @@ prof_dump(bool propagate_err, const char *filename, bool leakcheck)
 
 	cassert(config_prof);
 
-	prof_tdata = prof_tdata_get();
+	prof_tdata = prof_tdata_get(false);
 	if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
 		return (true);
 	prof_enter(prof_tdata);
@@ -966,11 +966,7 @@ prof_idump(void)
 
 	if (prof_booted == false)
 		return;
-	/*
-	 * Don't call prof_tdata_get() here, because it could cause recursive
-	 * allocation.
-	 */
-	prof_tdata = *prof_tdata_tsd_get();
+	prof_tdata = prof_tdata_get(false);
 	if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
 		return;
 	if (prof_tdata->enq) {
@@ -1020,11 +1016,7 @@ prof_gdump(void)
 
 	if (prof_booted == false)
 		return;
-	/*
-	 * Don't call prof_tdata_get() here, because it could cause recursive
-	 * allocation.
-	 */
-	prof_tdata = *prof_tdata_tsd_get();
+	prof_tdata = prof_tdata_get(false);
 	if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
 		return;
 	if (prof_tdata->enq) {

src/quarantine.c

@@ -1,3 +1,4 @@
+#define	JEMALLOC_QUARANTINE_C_
 #include "jemalloc/internal/jemalloc_internal.h"
 
 /*
@@ -11,39 +12,17 @@
 /******************************************************************************/
 /* Data. */
 
-typedef struct quarantine_obj_s quarantine_obj_t;
-typedef struct quarantine_s quarantine_t;
-
-struct quarantine_obj_s {
-	void	*ptr;
-	size_t	usize;
-};
-
-struct quarantine_s {
-	size_t			curbytes;
-	size_t			curobjs;
-	size_t			first;
-#define	LG_MAXOBJS_INIT 10
-	size_t			lg_maxobjs;
-	quarantine_obj_t	objs[1]; /* Dynamically sized ring buffer. */
-};
-
-static void	quarantine_cleanup(void *arg);
-
-malloc_tsd_data(static, quarantine, quarantine_t *, NULL)
-malloc_tsd_funcs(JEMALLOC_INLINE, quarantine, quarantine_t *, NULL,
-    quarantine_cleanup)
+malloc_tsd_data(, quarantine, quarantine_t *, NULL)
 
 /******************************************************************************/
 /* Function prototypes for non-inline static functions. */
 
-static quarantine_t	*quarantine_init(size_t lg_maxobjs);
 static quarantine_t	*quarantine_grow(quarantine_t *quarantine);
 static void	quarantine_drain(quarantine_t *quarantine, size_t upper_bound);
 
 /******************************************************************************/
 
-static quarantine_t *
+quarantine_t *
 quarantine_init(size_t lg_maxobjs)
 {
 	quarantine_t *quarantine;
@@ -119,24 +98,16 @@ quarantine(void *ptr)
 
 	quarantine = *quarantine_tsd_get();
 	if ((uintptr_t)quarantine <= (uintptr_t)QUARANTINE_STATE_MAX) {
-		if (quarantine == NULL) {
-			if ((quarantine = quarantine_init(LG_MAXOBJS_INIT)) ==
-			    NULL) {
-				idalloc(ptr);
-				return;
-			}
-		} else {
-			if (quarantine == QUARANTINE_STATE_PURGATORY) {
-				/*
-				 * Make a note that quarantine() was called
-				 * after quarantine_cleanup() was called.
-				 */
-				quarantine = QUARANTINE_STATE_REINCARNATED;
-				quarantine_tsd_set(&quarantine);
-			}
-			idalloc(ptr);
-			return;
+		if (quarantine == QUARANTINE_STATE_PURGATORY) {
+			/*
+			 * Make a note that quarantine() was called after
+			 * quarantine_cleanup() was called.
+			 */
+			quarantine = QUARANTINE_STATE_REINCARNATED;
+			quarantine_tsd_set(&quarantine);
 		}
+		idalloc(ptr);
+		return;
 	}
 	/*
 	 * Drain one or more objects if the quarantine size limit would be
@@ -169,7 +140,7 @@ quarantine(void *ptr)
 	}
 }
 
-static void
+void
 quarantine_cleanup(void *arg)
 {
 	quarantine_t *quarantine = *(quarantine_t **)arg;