Re: memory leak in vhost_net_ioctl

[Jason Wang]

On 2019/6/6 äå10:40, Hillf Danton wrote:
> On Wed, 05 Jun 2019 16:42:05 -0700 (PDT) syzbot wrote:
> > Hello,
> >
> > syzbot found the following crash on:
> >
> > HEAD commit:ÂÂÂ 788a0249 Merge tag 'arc-5.2-rc4' of 
> > git://git.kernel.org/p..
> > git tree:ÂÂÂÂÂÂ upstream
> > console output: https://syzkaller.appspot.com/x/log.txt?x=15dc9ea6a00000
> > kernel config: 
> > https://syzkaller.appspot.com/x/.config?x=d5c73825cbdc7326
> > dashboard link: 
> > https://syzkaller.appspot.com/bug?extid=0789f0c7e45efd7bb643
> > compiler:ÂÂÂÂÂÂ gcc (GCC) 9.0.0 20181231 (experimental)
> > syz repro: https://syzkaller.appspot.com/x/repro.syz?x=10b31761a00000
> > C reproducer: https://syzkaller.appspot.com/x/repro.c?x=124892c1a00000
> >
> > IMPORTANT: if you fix the bug, please add the following tag to the 
> > commit:
> > Reported-by: syzbot+0789f0c7e45efd7bb643@xxxxxxxxxxxxxxxxxxxxxxxxx
> >
> > udit: type=1400 audit(1559768703.229:36): avc: denied { map } for 
> > pid=7116 comm="syz-executor330" path="/root/syz-executor330334897" 
> > dev="sda1" ino=16461 
> > scontext=unconfined_u:system_r:insmod_t:s0-s0:c0.c1023 
> > tcontext=unconfined_u:object_r:user_home_t:s0 tclass=file permissive=1
> > executing program
> > executing program
> > BUG: memory leak
> > unreferenced object 0xffff88812421fe40 (size 64):
> > ÂÂ comm "syz-executor330", pid 7117, jiffies 4294949245 (age 13.030s)
> > ÂÂ hex dump (first 32 bytes):
> >  01 00 00 00 20 69 6f 63 00 00 00 00 64 65 76 2f .... ioc....dev/
> > ÂÂÂÂ 50 fe 21 24 81 88 ff ff 50 fe 21 24 81 88 ff ff P.!$....P.!$....
> > ÂÂ backtrace:
> > ÂÂÂÂ [<00000000ae0c4ae0>] kmemleak_alloc_recursive 
> > include/linux/kmemleak.h:55 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] slab_post_alloc_hook mm/slab.h:439 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] slab_alloc mm/slab.c:3326 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] kmem_cache_alloc_trace+0x13d/0x280 
> > mm/slab.c:3553
> > ÂÂÂÂ [<0000000079ebab38>] kmalloc include/linux/slab.h:547 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_ubuf_alloc 
> > drivers/vhost/net.c:241 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_set_backend 
> > drivers/vhost/net.c:1534 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_ioctl+0xb43/0xc10 
> > drivers/vhost/net.c:1716
> > ÂÂÂÂ [<000000009f6204a2>] vfs_ioctl fs/ioctl.c:46 [inline]
> > ÂÂÂÂ [<000000009f6204a2>] file_ioctl fs/ioctl.c:509 [inline]
> > ÂÂÂÂ [<000000009f6204a2>] do_vfs_ioctl+0x62a/0x810 fs/ioctl.c:696
> > ÂÂÂÂ [<00000000b45866de>] ksys_ioctl+0x86/0xb0 fs/ioctl.c:713
> > ÂÂÂÂ [<00000000dfb41eb8>] __do_sys_ioctl fs/ioctl.c:720 [inline]
> > ÂÂÂÂ [<00000000dfb41eb8>] __se_sys_ioctl fs/ioctl.c:718 [inline]
> > ÂÂÂÂ [<00000000dfb41eb8>] __x64_sys_ioctl+0x1e/0x30 fs/ioctl.c:718
> > ÂÂÂÂ [<0000000049c1f547>] do_syscall_64+0x76/0x1a0 
> > arch/x86/entry/common.c:301
> > ÂÂÂÂ [<0000000029cc8ca7>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >
> > BUG: memory leak
> > unreferenced object 0xffff88812421fa80 (size 64):
> > ÂÂ comm "syz-executor330", pid 7130, jiffies 4294949755 (age 7.930s)
> > ÂÂ hex dump (first 32 bytes):
> > ÂÂÂÂ 01 00 00 00 01 00 00 00 00 00 00 00 2f 76 69 72 ............/vir
> > ÂÂÂÂ 90 fa 21 24 81 88 ff ff 90 fa 21 24 81 88 ff ff ..!$......!$....
> > ÂÂ backtrace:
> > ÂÂÂÂ [<00000000ae0c4ae0>] kmemleak_alloc_recursive 
> > include/linux/kmemleak.h:55 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] slab_post_alloc_hook mm/slab.h:439 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] slab_alloc mm/slab.c:3326 [inline]
> > ÂÂÂÂ [<00000000ae0c4ae0>] kmem_cache_alloc_trace+0x13d/0x280 
> > mm/slab.c:3553
> > ÂÂÂÂ [<0000000079ebab38>] kmalloc include/linux/slab.h:547 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_ubuf_alloc 
> > drivers/vhost/net.c:241 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_set_backend 
> > drivers/vhost/net.c:1534 [inline]
> > ÂÂÂÂ [<0000000079ebab38>] vhost_net_ioctl+0xb43/0xc10 
> > drivers/vhost/net.c:1716
> > ÂÂÂÂ [<000000009f6204a2>] vfs_ioctl fs/ioctl.c:46 [inline]
> > ÂÂÂÂ [<000000009f6204a2>] file_ioctl fs/ioctl.c:509 [inline]
> > ÂÂÂÂ [<000000009f6204a2>] do_vfs_ioctl+0x62a/0x810 fs/ioctl.c:696
> > ÂÂÂÂ [<00000000b45866de>] ksys_ioctl+0x86/0xb0 fs/ioctl.c:713
> > ÂÂÂÂ [<00000000dfb41eb8>] __do_sys_ioctl fs/ioctl.c:720 [inline]
> > ÂÂÂÂ [<00000000dfb41eb8>] __se_sys_ioctl fs/ioctl.c:718 [inline]
> > ÂÂÂÂ [<00000000dfb41eb8>] __x64_sys_ioctl+0x1e/0x30 fs/ioctl.c:718
> > ÂÂÂÂ [<0000000049c1f547>] do_syscall_64+0x76/0x1a0 
> > arch/x86/entry/common.c:301
> > ÂÂÂÂ [<0000000029cc8ca7>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >
> >
> >
> > ---
> > This bug is generated by a bot. It may contain errors.
> > See https://goo.gl/tpsmEJ for more information about syzbot.
> > syzbot engineers can be reached at syzkaller@xxxxxxxxxxxxxxxxx
> >
> > syzbot will keep track of this bug report. See:
> > https://goo.gl/tpsmEJ#status for how to communicate with syzbot.
> > syzbot can test patches for this bug, for details see:
> > https://goo.gl/tpsmEJ#testing-patches
> Ignore my noise if you have no interest seeing the syzbot report.
>
> After commit c38e39c378f46f ("vhost-net: fix use-after-free in
> vhost_net_flush") flush would no longer free ubuf, just wait until 
> ubuf users
> disappear instead.
>
> The following diff, in hope that may perhaps help you handle the 
> memory leak,
> makes flush able to free ubuf in the path of file release.
>
> Thanks
> Hillf
> ---
> drivers/vhost/net.c | 8 +++++++-
> 1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index 3beb401..dcf20b6 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -141,6 +141,7 @@ struct vhost_net {
> ÂÂÂÂunsigned tx_zcopy_err;
> ÂÂÂÂ/* Flush in progress. Protected by tx vq lock. */
> ÂÂÂÂbool tx_flush;
> +ÂÂÂ bool ld;ÂÂÂ /* Last dinner */
> ÂÂÂÂ/* Private page frag */
> ÂÂÂÂstruct page_frag page_frag;
> ÂÂÂÂ/* Refcount bias of page frag */
> @@ -1283,6 +1284,7 @@ static int vhost_net_open(struct inode *inode, 
> struct file *f)
> ÂÂÂÂn = kvmalloc(sizeof *n, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
> ÂÂÂÂif (!n)
> ÂÂÂÂÂÂÂ return -ENOMEM;
> +ÂÂÂ n->ld = false;
> ÂÂÂÂvqs = kmalloc_array(VHOST_NET_VQ_MAX, sizeof(*vqs), GFP_KERNEL);
> ÂÂÂÂif (!vqs) {
> ÂÂÂÂÂÂÂ kvfree(n);
> @@ -1376,7 +1378,10 @@ static void vhost_net_flush(struct vhost_net *n)
> ÂÂÂÂÂÂÂ n->tx_flush = true;
> ÂÂÂÂÂÂÂ mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
> ÂÂÂÂÂÂÂ /* Wait for all lower device DMAs done. */
> - vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
> +ÂÂÂÂÂÂÂ if (n->ld)
> + vhost_net_ubuf_put_wait_and_free(n->vqs[VHOST_NET_VQ_TX].ubufs);
> +ÂÂÂÂÂÂÂ else
> + vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
> ÂÂÂÂÂÂÂ mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
> ÂÂÂÂÂÂÂ n->tx_flush = false;
> atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1);
> @@ -1403,6 +1408,7 @@ static int vhost_net_release(struct inode 
> *inode, struct file *f)
> ÂÂÂÂsynchronize_rcu();
> ÂÂÂÂ/* We do an extra flush before freeing memory,
> ÂÂÂÂ * since jobs can re-queue themselves. */
> +ÂÂÂ n->ld = true;
> ÂÂÂÂvhost_net_flush(n);
> ÂÂÂÂkfree(n->vqs[VHOST_NET_VQ_RX].rxq.queue);
> ÂÂÂÂkfree(n->vqs[VHOST_NET_VQ_TX].xdp);
> --


This is basically a kfree(ubuf) after the second vhost_net_flush() in 
vhost_net_release().

Could you please post a formal patch?

Thanks