Fix mlockall()/madvise() interaction.

mlockall(2) can cause purging via madvise(2) to fail.  Fix purging code
to check whether madvise() succeeded, and base zeroed page metadata on
the result.

Reported by Olivier Lecomte.

include/jemalloc/internal/chunk_mmap.h

@@ -9,7 +9,7 @@
 /******************************************************************************/
 #ifdef JEMALLOC_H_EXTERNS
 
-void	pages_purge(void *addr, size_t length);
+bool	pages_purge(void *addr, size_t length);
 
 void	*chunk_alloc_mmap(size_t size, size_t alignment, bool *zero);
 bool	chunk_dealloc_mmap(void *chunk, size_t size);

include/jemalloc/internal/extent.h

@@ -23,6 +23,9 @@ struct extent_node_s {
 
 	/* Total region size. */
 	size_t			size;
+
+	/* True if zero-filled; used by chunk recycling code. */
+	bool			zeroed;
 };
 typedef rb_tree(extent_node_t) extent_tree_t;
 

src/arena.c

@@ -551,24 +551,12 @@ arena_chunk_purge(arena_t *arena, arena_chunk_t *chunk)
 {
 	ql_head(arena_chunk_map_t) mapelms;
 	arena_chunk_map_t *mapelm;
-	size_t pageind, flag_unzeroed;
+	size_t pageind;
 	size_t ndirty;
 	size_t nmadvise;
 
 	ql_new(&mapelms);
 
-	flag_unzeroed =
-#ifdef JEMALLOC_PURGE_MADVISE_DONTNEED
-   /*
-    * madvise(..., MADV_DONTNEED) results in zero-filled pages for anonymous
-    * mappings.
-    */
-	    0
-#else
-	    CHUNK_MAP_UNZEROED
-#endif
-	    ;
-
 	/*
 	 * If chunk is the spare, temporarily re-allocate it, 1) so that its
 	 * run is reinserted into runs_avail_dirty, and 2) so that it cannot be
@@ -603,26 +591,12 @@ arena_chunk_purge(arena_t *arena, arena_chunk_t *chunk)
 			assert(arena_mapbits_dirty_get(chunk, pageind) ==
 			    arena_mapbits_dirty_get(chunk, pageind+npages-1));
 			if (arena_mapbits_dirty_get(chunk, pageind) != 0) {
-				size_t i;
-
 				arena_avail_tree_remove(
 				    &arena->runs_avail_dirty, mapelm);
 
-				arena_mapbits_unzeroed_set(chunk, pageind,
-				    flag_unzeroed);
 				arena_mapbits_large_set(chunk, pageind,
 				    (npages << LG_PAGE), 0);
-				/*
-				 * Update internal elements in the page map, so
-				 * that CHUNK_MAP_UNZEROED is properly set.
-				 */
-				for (i = 1; i < npages - 1; i++) {
-					arena_mapbits_unzeroed_set(chunk,
-					    pageind+i, flag_unzeroed);
-				}
 				if (npages > 1) {
-					arena_mapbits_unzeroed_set(chunk,
-					    pageind+npages-1, flag_unzeroed);
 					arena_mapbits_large_set(chunk,
 					    pageind+npages-1, 0, 0);
 				}
@@ -685,14 +659,30 @@ arena_chunk_purge(arena_t *arena, arena_chunk_t *chunk)
 		    sizeof(arena_chunk_map_t)) + map_bias;
 		size_t npages = arena_mapbits_large_size_get(chunk, pageind) >>
 		    LG_PAGE;
+		bool unzeroed;
+		size_t flag_unzeroed, i;
 
 		assert(pageind + npages <= chunk_npages);
 		assert(ndirty >= npages);
 		if (config_debug)
 			ndirty -= npages;
-
-		pages_purge((void *)((uintptr_t)chunk + (pageind << LG_PAGE)),
-		    (npages << LG_PAGE));
+		unzeroed = pages_purge((void *)((uintptr_t)chunk + (pageind <<
+		    LG_PAGE)), (npages << LG_PAGE));
+		flag_unzeroed = unzeroed ? CHUNK_MAP_UNZEROED : 0;
+		/*
+		 * Set the unzeroed flag for all pages, now that pages_purge()
+		 * has returned whether the pages were zeroed as a side effect
+		 * of purging.  This chunk map modification is safe even though
+		 * the arena mutex isn't currently owned by this thread,
+		 * because the run is marked as allocated, thus protecting it
+		 * from being modified by any other thread.  As long as these
+		 * writes don't perturb the first and last elements'
+		 * CHUNK_MAP_ALLOCATED bits, behavior is well defined.
+		 */
+		for (i = 0; i < npages; i++) {
+			arena_mapbits_unzeroed_set(chunk, pageind+i,
+			    flag_unzeroed);
+		}
 		if (config_stats)
 			nmadvise++;
 	}

src/chunk.c

@@ -43,6 +43,7 @@ chunk_recycle(size_t size, size_t alignment, bool base, bool *zero)
 	extent_node_t *node;
 	extent_node_t key;
 	size_t alloc_size, leadsize, trailsize;
+	bool zeroed;
 
 	if (base) {
 		/*
@@ -107,17 +108,18 @@ chunk_recycle(size_t size, size_t alignment, bool base, bool *zero)
 	}
 	malloc_mutex_unlock(&chunks_mtx);
 
-	if (node != NULL)
-		base_node_dealloc(node);
-#ifdef JEMALLOC_PURGE_MADVISE_DONTNEED
-	/* Pages are zeroed as a side effect of pages_purge(). */
+	zeroed = false;
+	if (node != NULL) {
+		if (node->zeroed) {
+			zeroed = true;
 			*zero = true;
-#else
-	if (*zero) {
+		}
+		base_node_dealloc(node);
+	}
+	if (zeroed == false && *zero) {
 		VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 		memset(ret, 0, size);
 	}
-#endif
 	return (ret);
 }
 
@@ -191,9 +193,10 @@ label_return:
 static void
 chunk_record(void *chunk, size_t size)
 {
+	bool unzeroed;
 	extent_node_t *xnode, *node, *prev, key;
 
-	pages_purge(chunk, size);
+	unzeroed = pages_purge(chunk, size);
 
 	/*
 	 * Allocate a node before acquiring chunks_mtx even though it might not
@@ -216,6 +219,7 @@ chunk_record(void *chunk, size_t size)
 		extent_tree_szad_remove(&chunks_szad, node);
 		node->addr = chunk;
 		node->size += size;
+		node->zeroed = (node->zeroed && (unzeroed == false));
 		extent_tree_szad_insert(&chunks_szad, node);
 		if (xnode != NULL)
 			base_node_dealloc(xnode);
@@ -234,6 +238,7 @@ chunk_record(void *chunk, size_t size)
 		node = xnode;
 		node->addr = chunk;
 		node->size = size;
+		node->zeroed = (unzeroed == false);
 		extent_tree_ad_insert(&chunks_ad, node);
 		extent_tree_szad_insert(&chunks_szad, node);
 	}
@@ -253,6 +258,7 @@ chunk_record(void *chunk, size_t size)
 		extent_tree_szad_remove(&chunks_szad, node);
 		node->addr = prev->addr;
 		node->size += prev->size;
+		node->zeroed = (node->zeroed && prev->zeroed);
 		extent_tree_szad_insert(&chunks_szad, node);
 
 		base_node_dealloc(prev);

src/chunk_mmap.c

@@ -113,22 +113,30 @@ pages_trim(void *addr, size_t alloc_size, size_t leadsize, size_t size)
 #endif
 }
 
-void
+bool
 pages_purge(void *addr, size_t length)
 {
+	bool unzeroed;
 
 #ifdef _WIN32
 	VirtualAlloc(addr, length, MEM_RESET, PAGE_READWRITE);
+	unzeroed = true;
 #else
 #  ifdef JEMALLOC_PURGE_MADVISE_DONTNEED
 #    define JEMALLOC_MADV_PURGE MADV_DONTNEED
+#    define JEMALLOC_MADV_ZEROS true
 #  elif defined(JEMALLOC_PURGE_MADVISE_FREE)
 #    define JEMALLOC_MADV_PURGE MADV_FREE
+#    define JEMALLOC_MADV_ZEROS false
 #  else
 #    error "No method defined for purging unused dirty pages."
 #  endif
-	madvise(addr, length, JEMALLOC_MADV_PURGE);
+	int err = madvise(addr, length, JEMALLOC_MADV_PURGE);
+	unzeroed = (JEMALLOC_MADV_ZEROS == false || err != 0);
+#  undef JEMALLOC_MADV_PURGE
+#  undef JEMALLOC_MADV_ZEROS
 #endif
+	return (unzeroed);
 }
 
 static void *