tun: memory leak in tun_set_iff

[Dmitry Vyukov]

Hello,

syzkaller has hit the following memory leak on 4.15-rc7.
Reproducer is attached.

unreeferenced object 0xffff88002c9ac400 (size 4096):
comm "syz-executor0", pid 12349, jiffies 4295751114 (age 10.067s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<00000000ad172f4e>] kmemleak_alloc_recursive
include/linux/kmemleak.h:55 [inline]
[<00000000ad172f4e>] slab_post_alloc_hook mm/slab.h:440 [inline]
[<00000000ad172f4e>] slab_alloc_node mm/slub.c:2725 [inline]
[<00000000ad172f4e>] slab_alloc mm/slub.c:2733 [inline]
[<00000000ad172f4e>] __kmalloc+0x1a9/0x340 mm/slub.c:3758
[<00000000d66b86d6>] kmalloc_array include/linux/slab.h:618 [inline]
[<00000000d66b86d6>] kcalloc include/linux/slab.h:629 [inline]
[<00000000d66b86d6>] __ptr_ring_init_queue_alloc
include/linux/ptr_ring.h:450 [inline]
[<00000000d66b86d6>] ptr_ring_init include/linux/ptr_ring.h:468 [inline]
[<00000000d66b86d6>] skb_array_init include/linux/skb_array.h:176 [inline]
[<00000000d66b86d6>] tun_attach+0x940/0x10b0 drivers/net/tun.c:754
[<000000007a69e5cb>] tun_set_iff drivers/net/tun.c:2315 [inline]
[<000000007a69e5cb>] __tun_chr_ioctl+0x2435/0x4210 drivers/net/tun.c:2524
[<000000005c75f6a6>] tun_chr_ioctl+0x2a/0x40 drivers/net/tun.c:2773
[<00000000ece2f188>] vfs_ioctl fs/ioctl.c:46 [inline]
[<00000000ece2f188>] file_ioctl fs/ioctl.c:500 [inline]
[<00000000ece2f188>] do_vfs_ioctl+0x1cf/0x16b0 fs/ioctl.c:684
[<0000000021f4fda7>] SYSC_ioctl fs/ioctl.c:701 [inline]
[<0000000021f4fda7>] SyS_ioctl+0xb6/0xe0 fs/ioctl.c:692
[<0000000001148918>] entry_SYSCALL_64_fastpath+0x23/0x9a
[<000000008d0cf26e>] 0xffffffffffffffff

[Xin Long]

we probably just need:
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -831,8 +831,10 @@ static int tun_attach(struct tun_struct *tun,
struct file *file,
/* Setup XDP RX-queue info, for new tfile getting attached */
err = xdp_rxq_info_reg(&tfile->xdp_rxq,
tun->dev, tfile->queue_index);
- if (err < 0)
+ if (err < 0) {
+ ptr_ring_cleanup(&tfile->tx_ring, NULL);
goto out;
+ }

will check it for sure.

[Jason Wang]

Good catch. This looks like a new bug in net-next instead of the what
memory leaker told us here.

Wang Cong has posted a possible fix for this.

Thanks