malloc: Decorate malloc maps

Add anonymous mmap annotations on loader malloc, malloc when it
allocates memory with mmap, and on malloc arena.  The /proc/self/maps
will now print:

   [anon: glibc: malloc arena]
   [anon: glibc: malloc]
   [anon: glibc: loader malloc]

On arena allocation, glibc annotates only the read/write mapping.

Checked on x86_64-linux-gnu and aarch64-linux-gnu.
Reviewed-by: DJ Delorie <dj@redhat.com>

elf/Makefile

@@ -2983,3 +2983,7 @@ $(objpfx)tst-dlclose-lazy.out: \
   $(objpfx)tst-dlclose-lazy-mod1.so $(objpfx)tst-dlclose-lazy-mod2.so
 
 $(objpfx)tst-decorate-maps: $(shared-thread-library)
+
+tst-decorate-maps-ENV = \
+  GLIBC_TUNABLES=glibc.malloc.arena_max=8:glibc.malloc.mmap_threshold=1024
+tst-decorate-maps-ARGS = 8

elf/dl-minimal-malloc.c

@@ -26,6 +26,7 @@
 #include <string.h>
 #include <ldsodefs.h>
 #include <malloc/malloc-internal.h>
+#include <setvmaname.h>
 
 static void *alloc_ptr, *alloc_end, *alloc_last_block;
 
@@ -60,6 +61,7 @@ __minimal_malloc (size_t n)
 		     MAP_ANON|MAP_PRIVATE, -1, 0);
       if (page == MAP_FAILED)
 	return NULL;
+      __set_vma_name (page, nup, " glibc: loader malloc");
       if (page != alloc_end)
 	alloc_ptr = page;
       alloc_end = page + nup;

elf/tst-decorate-maps.c

@@ -32,15 +32,21 @@
 
 static pthread_barrier_t b;
 
+static int expected_n_arenas;
+
 static void *
 tf (void *closure)
 {
+  void *p = xmalloc (1024);
+
   /* Wait the thread startup, so thread stack is allocated.  */
   xpthread_barrier_wait (&b);
 
   /* Wait the test to read the process mapping.  */
   xpthread_barrier_wait (&b);
 
+  free (p);
+
   return NULL;
 }
 
@@ -48,6 +54,9 @@ struct proc_maps_t
 {
   int n_def_threads;
   int n_user_threads;
+  int n_arenas;
+  int n_malloc_mmap;
+  int n_loader_malloc_mmap;
 };
 
 static struct proc_maps_t
@@ -69,6 +78,12 @@ read_proc_maps (void)
 	r.n_def_threads++;
       else if (strstr (line, "[anon: glibc: pthread user stack:") != NULL)
 	r.n_user_threads++;
+      else if (strstr (line, "[anon: glibc: malloc arena]") != NULL)
+	r.n_arenas++;
+      else if (strstr (line, "[anon: glibc: malloc]") != NULL)
+	r.n_malloc_mmap++;
+      else if (strstr (line, "[anon: glibc: loader malloc]") != NULL)
+	r.n_loader_malloc_mmap++;
     }
   free (line);
   xfclose (f);
@@ -90,6 +105,9 @@ do_test_threads (bool set_guard)
 
   xpthread_barrier_init (&b, NULL, num_threads + 1);
 
+  /* Issue a large malloc to trigger a mmap call.  */
+  void *p = xmalloc (256 * 1024);
+
   pthread_t thr[num_threads];
   {
     int i = 0;
@@ -128,6 +146,10 @@ do_test_threads (bool set_guard)
     struct proc_maps_t r = read_proc_maps ();
     TEST_COMPARE (r.n_def_threads, num_def_threads);
     TEST_COMPARE (r.n_user_threads, num_user_threads);
+    TEST_COMPARE (r.n_arenas, expected_n_arenas);
+    TEST_COMPARE (r.n_malloc_mmap, 1);
+    /* On some architectures the loader might use more than one page.  */
+    TEST_VERIFY (r.n_loader_malloc_mmap >= 1);
   }
 
   /* Let the threads finish.  */
@@ -140,9 +162,23 @@ do_test_threads (bool set_guard)
     struct proc_maps_t r = read_proc_maps ();
     TEST_COMPARE (r.n_def_threads, 0);
     TEST_COMPARE (r.n_user_threads, 0);
+    TEST_COMPARE (r.n_arenas, expected_n_arenas);
+    TEST_COMPARE (r.n_malloc_mmap, 1);
+    TEST_VERIFY (r.n_loader_malloc_mmap >= 1);
   }
+
+  free (p);
 }
 
+static void
+do_prepare (int argc, char *argv[])
+{
+  TEST_VERIFY_EXIT (argc == 2);
+  expected_n_arenas = strtol (argv[1], NULL, 10);
+  expected_n_arenas = expected_n_arenas - 1;
+}
+#define PREPARE do_prepare
+
 static int
 do_test (void)
 {

malloc/arena.c

@@ -17,6 +17,7 @@
    not, see <https://www.gnu.org/licenses/>.  */
 
 #include <stdbool.h>
+#include <setvmaname.h>
 
 #define TUNABLE_NAMESPACE malloc
 #include <elf/dl-tunables.h>
@@ -436,6 +437,9 @@ alloc_new_heap  (size_t size, size_t top_pad, size_t pagesize,
       return 0;
     }
 
+  /* Only considere the actual usable range.  */
+  __set_vma_name (p2, size, " glibc: malloc arena");
+
   madvise_thp (p2, size);
 
   h = (heap_info *) p2;

malloc/malloc.c

@@ -218,6 +218,7 @@
 #include <sys/sysinfo.h>
 
 #include <ldsodefs.h>
+#include <setvmaname.h>
 
 #include <unistd.h>
 #include <stdio.h>    /* needed for malloc_stats */
@@ -2428,6 +2429,8 @@ sysmalloc_mmap (INTERNAL_SIZE_T nb, size_t pagesize, int extra_flags, mstate av)
     madvise_thp (mm, size);
 #endif
 
+  __set_vma_name (mm, size, " glibc: malloc");
+
   /*
     The offset to the start of the mmapped region is stored in the prev_size
     field of the chunk.  This allows us to adjust returned start address to
@@ -2513,6 +2516,8 @@ sysmalloc_mmap_fallback (long int *s, INTERNAL_SIZE_T nb,
     madvise_thp (mbrk, size);
 #endif
 
+  __set_vma_name (mbrk, size, " glibc: malloc");
+
   /* Record that we no longer have a contiguous sbrk region.  After the first
      time mmap is used as backup, we do not ever rely on contiguous space
      since this could incorrectly bridge regions.  */

nptl/Makefile

@@ -699,6 +699,10 @@ tst-audit-threads-ENV = LD_AUDIT=$(objpfx)tst-audit-threads-mod1.so
 tst-setuid1-static-ENV = \
   LD_LIBRARY_PATH=$(ld-library-path):$(common-objpfx)elf:$(common-objpfx)nss
 
+tst-pthread-proc-maps-ENV = \
+  GLIBC_TUNABLES=glibc.malloc.arena_max=8:glibc.malloc.mmap_threshold=1024
+tst-pthread-proc-maps-ARGS = 8
+
 # The tests here better do not run in parallel.
 ifeq ($(run-built-tests),yes)
 ifneq ($(filter %tests,$(MAKECMDGOALS)),)