nvme-tcp: don't access released socket during error recovery
authorAkinobu Mita <akinobu.mita@gmail.com>
Sun, 26 Feb 2023 12:42:54 +0000 (21:42 +0900)
committerChristoph Hellwig <hch@lst.de>
Tue, 28 Feb 2023 13:14:44 +0000 (06:14 -0700)
While the error recovery work is temporarily failing reconnect attempts,
running the 'nvme list' command causes a kernel NULL pointer dereference
by calling getsockname() with a released socket.

During error recovery work, the nvme tcp socket is released and a new one
created, so it is not safe to access the socket without proper check.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Fixes: 02c57a82c008 ("nvme-tcp: print actual source IP address through sysfs "address" attr")
Reviewed-by: Martin Belanger <martin.belanger@dell.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/host/tcp.c

index d6100a7..ef27122 100644 (file)
@@ -2489,6 +2489,10 @@ static int nvme_tcp_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
 
        len = nvmf_get_address(ctrl, buf, size);
 
+       mutex_lock(&queue->queue_lock);
+
+       if (!test_bit(NVME_TCP_Q_LIVE, &queue->flags))
+               goto done;
        ret = kernel_getsockname(queue->sock, (struct sockaddr *)&src_addr);
        if (ret > 0) {
                if (len > 0)
@@ -2496,6 +2500,8 @@ static int nvme_tcp_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
                len += scnprintf(buf + len, size - len, "%ssrc_addr=%pISc\n",
                                (len) ? "," : "", &src_addr);
        }
+done:
+       mutex_unlock(&queue->queue_lock);
 
        return len;
 }