]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
mm/vmalloc: fix data race in show_numa_info()
authorJeongjun Park <aha310510@gmail.com>
Thu, 8 May 2025 16:56:20 +0000 (01:56 +0900)
committerAndrew Morton <akpm@linux-foundation.org>
Thu, 22 May 2025 21:55:36 +0000 (14:55 -0700)
The following data-race was found in show_numa_info():

==================================================================
BUG: KCSAN: data-race in vmalloc_info_show / vmalloc_info_show

read to 0xffff88800971fe30 of 4 bytes by task 8289 on cpu 0:
 show_numa_info mm/vmalloc.c:4936 [inline]
 vmalloc_info_show+0x5a8/0x7e0 mm/vmalloc.c:5016
 seq_read_iter+0x373/0xb40 fs/seq_file.c:230
 proc_reg_read_iter+0x11e/0x170 fs/proc/inode.c:299
....

write to 0xffff88800971fe30 of 4 bytes by task 8287 on cpu 1:
 show_numa_info mm/vmalloc.c:4934 [inline]
 vmalloc_info_show+0x38f/0x7e0 mm/vmalloc.c:5016
 seq_read_iter+0x373/0xb40 fs/seq_file.c:230
 proc_reg_read_iter+0x11e/0x170 fs/proc/inode.c:299
....

value changed: 0x0000008f -> 0x00000000
==================================================================

According to this report,there is a read/write data-race because
m->private is accessible to multiple CPUs.  To fix this, instead of
allocating the heap in proc_vmalloc_init() and passing the heap address to
m->private, vmalloc_info_show() should allocate the heap.

Link: https://lkml.kernel.org/r/20250508165620.15321-1-aha310510@gmail.com
Fixes: 8e1d743f2c26 ("mm: vmalloc: support multiple nodes in vmallocinfo")
Signed-off-by: Jeongjun Park <aha310510@gmail.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: "Uladzislau Rezki (Sony)" <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/vmalloc.c

index 701ea4ec8950eadbea9955d323403ed0d5d27ca4..49df04e1fbe179ddb0036a7caac972448ef00ec4 100644 (file)
@@ -3109,7 +3109,7 @@ static void clear_vm_uninitialized_flag(struct vm_struct *vm)
        /*
         * Before removing VM_UNINITIALIZED,
         * we should make sure that vm has proper values.
-        * Pair with smp_rmb() in show_numa_info().
+        * Pair with smp_rmb() in vread_iter() and vmalloc_info_show().
         */
        smp_wmb();
        vm->flags &= ~VM_UNINITIALIZED;
@@ -4939,28 +4939,29 @@ bool vmalloc_dump_obj(void *object)
 #endif
 
 #ifdef CONFIG_PROC_FS
-static void show_numa_info(struct seq_file *m, struct vm_struct *v)
-{
-       if (IS_ENABLED(CONFIG_NUMA)) {
-               unsigned int nr, *counters = m->private;
-               unsigned int step = 1U << vm_area_page_order(v);
 
-               if (!counters)
-                       return;
+/*
+ * Print number of pages allocated on each memory node.
+ *
+ * This function can only be called if CONFIG_NUMA is enabled
+ * and VM_UNINITIALIZED bit in v->flags is disabled.
+ */
+static void show_numa_info(struct seq_file *m, struct vm_struct *v,
+                                unsigned int *counters)
+{
+       unsigned int nr;
+       unsigned int step = 1U << vm_area_page_order(v);
 
-               if (v->flags & VM_UNINITIALIZED)
-                       return;
-               /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
-               smp_rmb();
+       if (!counters)
+               return;
 
-               memset(counters, 0, nr_node_ids * sizeof(unsigned int));
+       memset(counters, 0, nr_node_ids * sizeof(unsigned int));
 
-               for (nr = 0; nr < v->nr_pages; nr += step)
-                       counters[page_to_nid(v->pages[nr])] += step;
-               for_each_node_state(nr, N_HIGH_MEMORY)
-                       if (counters[nr])
-                               seq_printf(m, " N%u=%u", nr, counters[nr]);
-       }
+       for (nr = 0; nr < v->nr_pages; nr += step)
+               counters[page_to_nid(v->pages[nr])] += step;
+       for_each_node_state(nr, N_HIGH_MEMORY)
+               if (counters[nr])
+                       seq_printf(m, " N%u=%u", nr, counters[nr]);
 }
 
 static void show_purge_info(struct seq_file *m)
@@ -4984,6 +4985,10 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
        struct vmap_node *vn;
        struct vmap_area *va;
        struct vm_struct *v;
+       unsigned int *counters;
+
+       if (IS_ENABLED(CONFIG_NUMA))
+               counters = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
 
        for_each_vmap_node(vn) {
                spin_lock(&vn->busy.lock);
@@ -4998,6 +5003,11 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
                        }
 
                        v = va->vm;
+                       if (v->flags & VM_UNINITIALIZED)
+                               continue;
+
+                       /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
+                       smp_rmb();
 
                        seq_printf(m, "0x%pK-0x%pK %7ld",
                                v->addr, v->addr + v->size, v->size);
@@ -5032,7 +5042,9 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
                        if (is_vmalloc_addr(v->pages))
                                seq_puts(m, " vpages");
 
-                       show_numa_info(m, v);
+                       if (IS_ENABLED(CONFIG_NUMA))
+                               show_numa_info(m, v, counters);
+
                        seq_putc(m, '\n');
                }
                spin_unlock(&vn->busy.lock);
@@ -5042,19 +5054,14 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
         * As a final step, dump "unpurged" areas.
         */
        show_purge_info(m);
+       if (IS_ENABLED(CONFIG_NUMA))
+               kfree(counters);
        return 0;
 }
 
 static int __init proc_vmalloc_init(void)
 {
-       void *priv_data = NULL;
-
-       if (IS_ENABLED(CONFIG_NUMA))
-               priv_data = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
-
-       proc_create_single_data("vmallocinfo",
-               0400, NULL, vmalloc_info_show, priv_data);
-
+       proc_create_single("vmallocinfo", 0400, NULL, vmalloc_info_show);
        return 0;
 }
 module_init(proc_vmalloc_init);