From:   Mohammed Gamal <mgamal@redhat.com>
To:     decui@microsoft.com, linux-hyperv@vger.kernel.org
Cc:     linux-kernel@vger.kernel.org, mikelley@microsoft.com,
        parri.andrea@gmail.com, haiyangz@microsoft.com, wei.liu@kernel.org,
        vkuznets@redhat.com, xxiong@redhat.com,
        Mohammed Gamal <mgamal@redhat.com>
Subject: [PATCH v3] Drivers: vmbus: Check for channel allocation before looking up relids
Date:   Fri, 17 Feb 2023 22:44:11 +0200
Message-Id: <20230217204411.212709-1-mgamal@redhat.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

relid2channel() assumes vmbus channel array to be allocated when called.
However, in cases such as kdump/kexec, not all relids will be reset by the host.
When the second kernel boots and if the guest receives a vmbus interrupt during
vmbus driver initialization before vmbus_connect() is called, before it finishes,
or if it fails, the vmbus interrupt service routine is called which in turn calls
relid2channel() and can cause a null pointer dereference.

Print a warning and error out in relid2channel() for a channel id that's invalid
in the second kernel.

Fixes: 8b6a877c060e ("Drivers: hv: vmbus: Replace the per-CPU channel lists with a global array of channels")

Signed-off-by: Mohammed Gamal <mgamal@redhat.com>
---
Changes from v2:
	* Changed commit message
	* Use pr_warn_once() instead of WARN() with a shorter message
	  for less overall noise
---
 drivers/hv/connection.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/hv/connection.c b/drivers/hv/connection.c
index 9dc27e5d367a..da51b50787df 100644
--- a/drivers/hv/connection.c
+++ b/drivers/hv/connection.c
@@ -409,6 +409,10 @@ void vmbus_disconnect(void)
  */
 struct vmbus_channel *relid2channel(u32 relid)
 {
+	if (vmbus_connection.channels == NULL) {
+		pr_warn_once("relid2channel: relid=%d: No channels mapped!\n", relid);
+		return NULL;
+	}
 	if (WARN_ON(relid >= MAX_CHANNEL_RELIDS))
 		return NULL;
 	return READ_ONCE(vmbus_connection.channels[relid]);
-- 
2.38.1