projects / platform / kernel / linux-rpi.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: 9a45984)
tun: fix use-after-free when register netdev failed
authorYang Yingliang <yangyingliang@huawei.com>
Tue, 10 Sep 2019 10:56:57 +0000 (18:56 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 19 Sep 2019 07:09:33 +0000 (09:09 +0200)
[ Upstream commit 77f22f92dff8e7b45c7786a430626d38071d4670 ]

I got a UAF repport in tun driver when doing fuzzy test:

[  466.269490] ==================================================================
[  466.271792] BUG: KASAN: use-after-free in tun_chr_read_iter+0x2ca/0x2d0
[  466.271806] Read of size 8 at addr ffff888372139250 by task tun-test/2699
[  466.271810]
[  466.271824] CPU: 1 PID: 2699 Comm: tun-test Not tainted 5.3.0-rc1-00001-g5a9433db2614-dirty #427
[  466.271833] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
[  466.271838] Call Trace:
[  466.271858]  dump_stack+0xca/0x13e
[  466.271871]  ? tun_chr_read_iter+0x2ca/0x2d0
[  466.271890]  print_address_description+0x79/0x440
[  466.271906]  ? vprintk_func+0x5e/0xf0
[  466.271920]  ? tun_chr_read_iter+0x2ca/0x2d0
[  466.271935]  __kasan_report+0x15c/0x1df
[  466.271958]  ? tun_chr_read_iter+0x2ca/0x2d0
[  466.271976]  kasan_report+0xe/0x20
[  466.271987]  tun_chr_read_iter+0x2ca/0x2d0
[  466.272013]  do_iter_readv_writev+0x4b7/0x740
[  466.272032]  ? default_llseek+0x2d0/0x2d0
[  466.272072]  do_iter_read+0x1c5/0x5e0
[  466.272110]  vfs_readv+0x108/0x180
[  466.299007]  ? compat_rw_copy_check_uvector+0x440/0x440
[  466.299020]  ? fsnotify+0x888/0xd50
[  466.299040]  ? __fsnotify_parent+0xd0/0x350
[  466.299064]  ? fsnotify_first_mark+0x1e0/0x1e0
[  466.304548]  ? vfs_write+0x264/0x510
[  466.304569]  ? ksys_write+0x101/0x210
[  466.304591]  ? do_preadv+0x116/0x1a0
[  466.304609]  do_preadv+0x116/0x1a0
[  466.309829]  do_syscall_64+0xc8/0x600
[  466.309849]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  466.309861] RIP: 0033:0x4560f9
[  466.309875] Code: 00 00 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48
[  466.309889] RSP: 002b:00007ffffa5166e8 EFLAGS: 00000206 ORIG_RAX: 0000000000000127
[  466.322992] RAX: ffffffffffffffda RBX: 0000000000400460 RCX: 00000000004560f9
[  466.322999] RDX: 0000000000000003 RSI: 00000000200008c0 RDI: 0000000000000003
[  466.323007] RBP: 00007ffffa516700 R08: 0000000000000004 R09: 0000000000000000
[  466.323014] R10: 0000000000000000 R11: 0000000000000206 R12: 000000000040cb10
[  466.323021] R13: 0000000000000000 R14: 00000000006d7018 R15: 0000000000000000
[  466.323057]
[  466.323064] Allocated by task 2605:
[  466.335165]  save_stack+0x19/0x80
[  466.336240]  __kasan_kmalloc.constprop.8+0xa0/0xd0
[  466.337755]  kmem_cache_alloc+0xe8/0x320
[  466.339050]  getname_flags+0xca/0x560
[  466.340229]  user_path_at_empty+0x2c/0x50
[  466.341508]  vfs_statx+0xe6/0x190
[  466.342619]  __do_sys_newstat+0x81/0x100
[  466.343908]  do_syscall_64+0xc8/0x600
[  466.345303]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  466.347034]
[  466.347517] Freed by task 2605:
[  466.348471]  save_stack+0x19/0x80
[  466.349476]  __kasan_slab_free+0x12e/0x180
[  466.350726]  kmem_cache_free+0xc8/0x430
[  466.351874]  putname+0xe2/0x120
[  466.352921]  filename_lookup+0x257/0x3e0
[  466.354319]  vfs_statx+0xe6/0x190
[  466.355498]  __do_sys_newstat+0x81/0x100
[  466.356889]  do_syscall_64+0xc8/0x600
[  466.358037]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  466.359567]
[  466.360050] The buggy address belongs to the object at ffff888372139100
[  466.360050]  which belongs to the cache names_cache of size 4096
[  466.363735] The buggy address is located 336 bytes inside of
[  466.363735]  4096-byte region [ffff888372139100, ffff88837213a100)
[  466.367179] The buggy address belongs to the page:
[  466.368604] page:ffffea000dc84e00 refcount:1 mapcount:0 mapping:ffff8883df1b4f00 index:0x0 compound_mapcount: 0
[  466.371582] flags: 0x2fffff80010200(slab|head)
[  466.372910] raw: 002fffff80010200 dead000000000100 dead000000000122 ffff8883df1b4f00
[  466.375209] raw: 0000000000000000 0000000000070007 00000001ffffffff 0000000000000000
[  466.377778] page dumped because: kasan: bad access detected
[  466.379730]
[  466.380288] Memory state around the buggy address:
[  466.381844]  ffff888372139100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  466.384009]  ffff888372139180: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  466.386131] >ffff888372139200: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  466.388257]                                                  ^
[  466.390234]  ffff888372139280: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  466.392512]  ffff888372139300: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  466.394667] ==================================================================

tun_chr_read_iter() accessed the memory which freed by free_netdev()
called by tun_set_iff():

        CPUA                                           CPUB
  tun_set_iff()
    alloc_netdev_mqs()
    tun_attach()
                                                  tun_chr_read_iter()
                                                    tun_get()
                                                    tun_do_read()
                                                      tun_ring_recv()
    register_netdevice() <-- inject error
    goto err_detach
    tun_detach_all() <-- set RCV_SHUTDOWN
    free_netdev() <-- called from
                     err_free_dev path
      netdev_freemem() <-- free the memory
                        without check refcount
      (In this path, the refcount cannot prevent
       freeing the memory of dev, and the memory
       will be used by dev_put() called by
       tun_chr_read_iter() on CPUB.)
                                                     (Break from tun_ring_recv(),
                                                     because RCV_SHUTDOWN is set)
                                                   tun_put()
                                                     dev_put() <-- use the memory
                                                                   freed by netdev_freemem()

Put the publishing of tfile->tun after register_netdevice(),
so tun_get() won't get the tun pointer that freed by
err_detach path if register_netdevice() failed.

Fixes: eb0fb363f920 ("tuntap: attach queue 0 before registering netdevice")
Reported-by: Hulk Robot <hulkci@huawei.com>
Suggested-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/tun.c patch | blob | history

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 5fa7047..e1ac1c5 100644 (file)
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -801,7 +801,8 @@ static void tun_detach_all(struct net_device *dev)
 }
 
 static int tun_attach(struct tun_struct *tun, struct file *file,
-                     bool skip_filter, bool napi, bool napi_frags)
+                     bool skip_filter, bool napi, bool napi_frags,
+                     bool publish_tun)
 {
        struct tun_file *tfile = file->private_data;
        struct net_device *dev = tun->dev;
@@ -881,7 +882,8 @@ static int tun_attach(struct tun_struct *tun, struct file *file,
         * initialized tfile; otherwise we risk using half-initialized
         * object.
         */
-       rcu_assign_pointer(tfile->tun, tun);
+       if (publish_tun)
+               rcu_assign_pointer(tfile->tun, tun);
        rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
        tun->numqueues++;
        tun_set_real_num_queues(tun);
@@ -2553,7 +2555,7 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 
                err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
                                 ifr->ifr_flags & IFF_NAPI,
-                                ifr->ifr_flags & IFF_NAPI_FRAGS);
+                                ifr->ifr_flags & IFF_NAPI_FRAGS, true);
                if (err < 0)
                        return err;
 
@@ -2652,13 +2654,17 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
 
                INIT_LIST_HEAD(&tun->disabled);
                err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
-                                ifr->ifr_flags & IFF_NAPI_FRAGS);
+                                ifr->ifr_flags & IFF_NAPI_FRAGS, false);
                if (err < 0)
                        goto err_free_flow;
 
                err = register_netdevice(tun->dev);
                if (err < 0)
                        goto err_detach;
+               /* free_netdev() won't check refcnt, to aovid race
+                * with dev_put() we need publish tun after registration.
+                */
+               rcu_assign_pointer(tfile->tun, tun);
        }
 
        netif_carrier_on(tun->dev);
@@ -2802,7 +2808,7 @@ static int tun_set_queue(struct file *file, struct ifreq *ifr)
                if (ret < 0)
                        goto unlock;
                ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
-                                tun->flags & IFF_NAPI_FRAGS);
+                                tun->flags & IFF_NAPI_FRAGS, true);
        } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
                tun = rtnl_dereference(tfile->tun);
                if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
Domain: System / Kernel;