Thanks for your detailed suggestions.
在 2020/1/18 1:32, Mike Christie 写道:
On 01/17/2020 05:50 AM, Sun Ke wrote:
Connect and disconnect a nbd device repeatedly, will cause
NULL pointer fault.
It will appear by the steps:
1. Connect the nbd device and disconnect it, but now nbd device
is not disconnected totally.
2. Connect the same nbd device again immediately, it will fail
in nbd_start_device with a EBUSY return value.
3. Wait a second to make sure the last config_refs is reduced
and run nbd_config_put to disconnect the nbd device totally.
4. Start another process to open the nbd_device, config_refs
will increase and at the same time disconnect it.
Just to make sure I understood this, for step 4 the process is doing:
open(/dev/nbdX);
ioctl(NBD_DISCONNECT, /dev/nbdX) or nbd_genl_disconnect(for /dev/nbdX)
?
do nbd_genl_disconnect(for /dev/nbdX)；
I tested it. Connect /dev/nbdX
through ioctl interface by nbd-client -L -N export localhost /dev/nbdX and
through netlink interface by nbd-client localhost XXXX /dev/nbdX,
disconnect /dev/nbdX by nbd-client -d /dev/nbdX.
Both call nbd_genl_disconnect(for /dev/nbdX) and both contain the same
null pointer dereference.
There is no successful NBD_DO_IT / nbd_genl_connect between the open and
disconnect calls at step #4, because it would normally be done at #2 and
that failed. nbd_disconnect_and_put could then reference a null
recv_workq. If we are also racing with a close() then that could free
the device/config from under nbd_disconnect_and_put.
Yes, nbd_disconnect_and_put could then reference a null recv_workq.
To fix it, add a NBD_HAS_STARTED flag. Set it in nbd_start_device_ioctl
I'm not sure if we need the new bit. We could just add a check for a non
null task_recv in nbd_genl_disconnect like how nbd_start_device and
nbd_genl_disconnect do.
I am also not very sure which is better.
because in nbd_config_put, not only recv_workq is null,
nbd->task_recv and nbd->config the same.
so I doubt that if step 4 do something else will also reference a null
pointer.
The new bit might be more clear which is nice. If we got this route,
should the new bit be a runtime_flag like other device state bits?
Yes, I realize it. Just add a check for a non null task_recv in
nbd_genl_disconnect is better, right?
and nbd_genl_connect if nbd device is started successfully.
Clear it in nbd_config_put. Test it in nbd_genl_disconnect and
nbd_genl_reconfigure.
Signed-off-by: Sun Ke <sunke32@xxxxxxxxxx>
---
drivers/block/nbd.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
index b4607dd96185..ddd364e208ab 100644
--- a/drivers/block/nbd.c
+++ b/drivers/block/nbd.c
@@ -83,6 +83,7 @@ struct link_dead_args {
#define NBD_DESTROY_ON_DISCONNECT 0
#define NBD_DISCONNECT_REQUESTED 1
+#define NBD_HAS_STARTED 2
struct nbd_config {
u32 flags;
@@ -1215,6 +1216,7 @@ static void nbd_config_put(struct nbd_device *nbd)
nbd->disk->queue->limits.discard_alignment = 0;
blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
+ clear_bit(NBD_HAS_STARTED, &nbd->flags);
mutex_unlock(&nbd->config_lock);
nbd_put(nbd);
@@ -1290,6 +1292,8 @@ static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *b
ret = nbd_start_device(nbd);
if (ret)
return ret;
+ else
+ set_bit(NBD_HAS_STARTED, &nbd->flags);
if (max_part)
bdev->bd_invalidated = 1;
@@ -1961,6 +1965,7 @@ static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
mutex_unlock(&nbd->config_lock);
if (!ret) {
set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
+ set_bit(NBD_HAS_STARTED, &nbd->flags);
refcount_inc(&nbd->config_refs);
nbd_connect_reply(info, nbd->index);
}
@@ -2008,6 +2013,14 @@ static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
index);
return -EINVAL;
}
+
+ if (!test_bit(NBD_HAS_STARTED, &nbd->flags)) {
+ mutex_unlock(&nbd_index_mutex);
+ printk(KERN_ERR "nbd: device at index %d failed to start\n",
+ index);
+ return -EBUSY;
+ }
+
if (!refcount_inc_not_zero(&nbd->refs)) {
mutex_unlock(&nbd_index_mutex);
printk(KERN_ERR "nbd: device at index %d is going down\n",
@@ -2049,6 +2062,14 @@ static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
index);
return -EINVAL;
}
+
+ if (!test_bit(NBD_HAS_STARTED, &nbd->flags)) {
+ mutex_unlock(&nbd_index_mutex);
+ printk(KERN_ERR "nbd: device at index %d failed to start\n",
+ index);
+ return -EBUSY;
+ }
+
if (!refcount_inc_not_zero(&nbd->refs)) {
mutex_unlock(&nbd_index_mutex);
printk(KERN_ERR "nbd: device at index %d is going down\n",
I thought the changes in nbd_genl_reconfigure is necessary althought my
test do not call it. but now I think it is superfluous,
nbd_genl_reconfigure checks for a non null task_recv.
Thanks,
Ke
.
