Drivers: hv: vmbus: Fix rescind handling issues
authorK. Y. Srinivasan <kys@microsoft.com>
Fri, 11 Aug 2017 17:03:59 +0000 (10:03 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 16 Aug 2017 16:16:29 +0000 (09:16 -0700)
This patch handles the following issues that were observed when we are
handling racing channel offer message and rescind message for the same
offer:

1. Since the host does not respond to messages on a rescinded channel,
in the current code, we could be indefinitely blocked on the vmbus_open() call.

2. When a rescinded channel is being closed, if there is a pending interrupt on the
channel, we could end up freeing the channel that the interrupt handler would run on.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Reviewed-by: Dexuan Cui <decui@microsoft.com>
Tested-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/hv/channel.c
drivers/hv/channel_mgmt.c
drivers/hv/vmbus_drv.c
include/linux/hyperv.h

index e57cc40..63ac1c6 100644 (file)
@@ -177,6 +177,11 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
                      &vmbus_connection.chn_msg_list);
        spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 
+       if (newchannel->rescind) {
+               err = -ENODEV;
+               goto error_free_gpadl;
+       }
+
        ret = vmbus_post_msg(open_msg,
                             sizeof(struct vmbus_channel_open_channel), true);
 
@@ -421,6 +426,11 @@ int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 
        spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 
+       if (channel->rescind) {
+               ret = -ENODEV;
+               goto cleanup;
+       }
+
        ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
                             sizeof(*msginfo), true);
        if (ret != 0)
@@ -494,6 +504,10 @@ int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
        list_add_tail(&info->msglistentry,
                      &vmbus_connection.chn_msg_list);
        spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
+
+       if (channel->rescind)
+               goto post_msg_err;
+
        ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
                             true);
 
index 4bbb8de..968af17 100644 (file)
@@ -451,6 +451,12 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
        /* Make sure this is a new offer */
        mutex_lock(&vmbus_connection.channel_mutex);
 
+       /*
+        * Now that we have acquired the channel_mutex,
+        * we can release the potentially racing rescind thread.
+        */
+       atomic_dec(&vmbus_connection.offer_in_progress);
+
        list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
                if (!uuid_le_cmp(channel->offermsg.offer.if_type,
                        newchannel->offermsg.offer.if_type) &&
@@ -481,7 +487,6 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
                        channel->num_sc++;
                        spin_unlock_irqrestore(&channel->lock, flags);
                } else {
-                       atomic_dec(&vmbus_connection.offer_in_progress);
                        goto err_free_chan;
                }
        }
@@ -510,7 +515,6 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
        if (!fnew) {
                if (channel->sc_creation_callback != NULL)
                        channel->sc_creation_callback(newchannel);
-               atomic_dec(&vmbus_connection.offer_in_progress);
                return;
        }
 
@@ -541,7 +545,7 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
                goto err_deq_chan;
        }
 
-       atomic_dec(&vmbus_connection.offer_in_progress);
+       newchannel->probe_done = true;
        return;
 
 err_deq_chan:
@@ -882,8 +886,27 @@ static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
        channel->rescind = true;
        spin_unlock_irqrestore(&channel->lock, flags);
 
+       /*
+        * Now that we have posted the rescind state, perform
+        * rescind related cleanup.
+        */
        vmbus_rescind_cleanup(channel);
 
+       /*
+        * Now wait for offer handling to complete.
+        */
+       while (READ_ONCE(channel->probe_done) == false) {
+               /*
+                * We wait here until any channel offer is currently
+                * being processed.
+                */
+               msleep(1);
+       }
+
+       /*
+        * At this point, the rescind handling can proceed safely.
+        */
+
        if (channel->device_obj) {
                if (channel->chn_rescind_callback) {
                        channel->chn_rescind_callback(channel);
index ed84e96..43160a2 100644 (file)
@@ -940,6 +940,9 @@ static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
                        if (channel->offermsg.child_relid != relid)
                                continue;
 
+                       if (channel->rescind)
+                               continue;
+
                        switch (channel->callback_mode) {
                        case HV_CALL_ISR:
                                vmbus_channel_isr(channel);
index 27db4e6..07650d0 100644 (file)
@@ -879,6 +879,8 @@ struct vmbus_channel {
         */
        enum hv_numa_policy affinity_policy;
 
+       bool probe_done;
+
 };
 
 static inline bool is_hvsock_channel(const struct vmbus_channel *c)