NBD: allow hung network I/O to be cancelled
[pandora-kernel.git] / drivers / block / nbd.c
index be92c65..cb136a9 100644 (file)
@@ -113,12 +113,42 @@ static void nbd_end_request(struct request *req)
        spin_unlock_irqrestore(q->queue_lock, flags);
 }
 
+static void sock_shutdown(struct nbd_device *lo, int lock)
+{
+       /* Forcibly shutdown the socket causing all listeners
+        * to error
+        *
+        * FIXME: This code is duplicated from sys_shutdown, but
+        * there should be a more generic interface rather than
+        * calling socket ops directly here */
+       if (lock)
+               mutex_lock(&lo->tx_lock);
+       if (lo->sock) {
+               printk(KERN_WARNING "%s: shutting down socket\n",
+                       lo->disk->disk_name);
+               lo->sock->ops->shutdown(lo->sock, SEND_SHUTDOWN|RCV_SHUTDOWN);
+               lo->sock = NULL;
+       }
+       if (lock)
+               mutex_unlock(&lo->tx_lock);
+}
+
+static void nbd_xmit_timeout(unsigned long arg)
+{
+       struct task_struct *task = (struct task_struct *)arg;
+
+       printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
+               task->comm, task->pid);
+       force_sig(SIGKILL, task);
+}
+
 /*
  *  Send or receive packet.
  */
-static int sock_xmit(struct socket *sock, int send, void *buf, int size,
+static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
                int msg_flags)
 {
+       struct socket *sock = lo->sock;
        int result;
        struct msghdr msg;
        struct kvec iov;
@@ -139,9 +169,20 @@ static int sock_xmit(struct socket *sock, int send, void *buf, int size,
                msg.msg_controllen = 0;
                msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 
-               if (send)
+               if (send) {
+                       struct timer_list ti;
+
+                       if (lo->xmit_timeout) {
+                               init_timer(&ti);
+                               ti.function = nbd_xmit_timeout;
+                               ti.data = (unsigned long)current;
+                               ti.expires = jiffies + lo->xmit_timeout;
+                               add_timer(&ti);
+                       }
                        result = kernel_sendmsg(sock, &msg, &iov, 1, size);
-               else
+                       if (lo->xmit_timeout)
+                               del_timer_sync(&ti);
+               } else
                        result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
 
                if (signal_pending(current)) {
@@ -150,6 +191,7 @@ static int sock_xmit(struct socket *sock, int send, void *buf, int size,
                                current->pid, current->comm,
                                dequeue_signal_lock(current, &current->blocked, &info));
                        result = -EINTR;
+                       sock_shutdown(lo, !send);
                        break;
                }
 
@@ -167,23 +209,22 @@ static int sock_xmit(struct socket *sock, int send, void *buf, int size,
        return result;
 }
 
-static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec,
+static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
                int flags)
 {
        int result;
        void *kaddr = kmap(bvec->bv_page);
-       result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len,
-                       flags);
+       result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags);
        kunmap(bvec->bv_page);
        return result;
 }
 
+/* always call with the tx_lock held */
 static int nbd_send_req(struct nbd_device *lo, struct request *req)
 {
-       int result, i, flags;
+       int result, flags;
        struct nbd_request request;
        unsigned long size = req->nr_sectors << 9;
-       struct socket *sock = lo->sock;
 
        request.magic = htonl(NBD_REQUEST_MAGIC);
        request.type = htonl(nbd_cmd(req));
@@ -196,8 +237,8 @@ static int nbd_send_req(struct nbd_device *lo, struct request *req)
                        nbdcmd_to_ascii(nbd_cmd(req)),
                        (unsigned long long)req->sector << 9,
                        req->nr_sectors << 9);
-       result = sock_xmit(sock, 1, &request, sizeof(request),
-                       (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0);
+       result = sock_xmit(lo, 1, &request, sizeof(request),
+                       (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
        if (result <= 0) {
                printk(KERN_ERR "%s: Send control failed (result %d)\n",
                                lo->disk->disk_name, result);
@@ -205,27 +246,23 @@ static int nbd_send_req(struct nbd_device *lo, struct request *req)
        }
 
        if (nbd_cmd(req) == NBD_CMD_WRITE) {
-               struct bio *bio;
+               struct req_iterator iter;
+               struct bio_vec *bvec;
                /*
                 * we are really probing at internals to determine
                 * whether to set MSG_MORE or not...
                 */
-               rq_for_each_bio(bio, req) {
-                       struct bio_vec *bvec;
-                       bio_for_each_segment(bvec, bio, i) {
-                               flags = 0;
-                               if ((i < (bio->bi_vcnt - 1)) || bio->bi_next)
-                                       flags = MSG_MORE;
-                               dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
-                                               lo->disk->disk_name, req,
-                                               bvec->bv_len);
-                               result = sock_send_bvec(sock, bvec, flags);
-                               if (result <= 0) {
-                                       printk(KERN_ERR "%s: Send data failed (result %d)\n",
-                                                       lo->disk->disk_name,
-                                                       result);
-                                       goto error_out;
-                               }
+               rq_for_each_segment(bvec, req, iter) {
+                       flags = 0;
+                       if (!rq_iter_last(req, iter))
+                               flags = MSG_MORE;
+                       dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
+                                       lo->disk->disk_name, req, bvec->bv_len);
+                       result = sock_send_bvec(lo, bvec, flags);
+                       if (result <= 0) {
+                               printk(KERN_ERR "%s: Send data failed (result %d)\n",
+                                               lo->disk->disk_name, result);
+                               goto error_out;
                        }
                }
        }
@@ -235,22 +272,18 @@ error_out:
        return 1;
 }
 
-static struct request *nbd_find_request(struct nbd_device *lo, char *handle)
+static struct request *nbd_find_request(struct nbd_device *lo,
+                                       struct request *xreq)
 {
-       struct request *req;
-       struct list_head *tmp;
-       struct request *xreq;
+       struct request *req, *tmp;
        int err;
 
-       memcpy(&xreq, handle, sizeof(xreq));
-
        err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
        if (unlikely(err))
                goto out;
 
        spin_lock(&lo->queue_lock);
-       list_for_each(tmp, &lo->queue_head) {
-               req = list_entry(tmp, struct request, queuelist);
+       list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
                if (req != xreq)
                        continue;
                list_del_init(&req->queuelist);
@@ -265,11 +298,11 @@ out:
        return ERR_PTR(err);
 }
 
-static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec)
+static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
 {
        int result;
        void *kaddr = kmap(bvec->bv_page);
-       result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len,
+       result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
                        MSG_WAITALL);
        kunmap(bvec->bv_page);
        return result;
@@ -281,10 +314,9 @@ static struct request *nbd_read_stat(struct nbd_device *lo)
        int result;
        struct nbd_reply reply;
        struct request *req;
-       struct socket *sock = lo->sock;
 
        reply.magic = 0;
-       result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL);
+       result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
        if (result <= 0) {
                printk(KERN_ERR "%s: Receive control failed (result %d)\n",
                                lo->disk->disk_name, result);
@@ -299,7 +331,7 @@ static struct request *nbd_read_stat(struct nbd_device *lo)
                goto harderror;
        }
 
-       req = nbd_find_request(lo, reply.handle);
+       req = nbd_find_request(lo, *(struct request **)reply.handle);
        if (unlikely(IS_ERR(req))) {
                result = PTR_ERR(req);
                if (result != -ENOENT)
@@ -321,22 +353,19 @@ static struct request *nbd_read_stat(struct nbd_device *lo)
        dprintk(DBG_RX, "%s: request %p: got reply\n",
                        lo->disk->disk_name, req);
        if (nbd_cmd(req) == NBD_CMD_READ) {
-               int i;
-               struct bio *bio;
-               rq_for_each_bio(bio, req) {
-                       struct bio_vec *bvec;
-                       bio_for_each_segment(bvec, bio, i) {
-                               result = sock_recv_bvec(sock, bvec);
-                               if (result <= 0) {
-                                       printk(KERN_ERR "%s: Receive data failed (result %d)\n",
-                                                       lo->disk->disk_name,
-                                                       result);
-                                       req->errors++;
-                                       return req;
-                               }
-                               dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
-                                       lo->disk->disk_name, req, bvec->bv_len);
+               struct req_iterator iter;
+               struct bio_vec *bvec;
+
+               rq_for_each_segment(bvec, req, iter) {
+                       result = sock_recv_bvec(lo, bvec);
+                       if (result <= 0) {
+                               printk(KERN_ERR "%s: Receive data failed (result %d)\n",
+                                               lo->disk->disk_name, result);
+                               req->errors++;
+                               return req;
                        }
+                       dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
+                               lo->disk->disk_name, req, bvec->bv_len);
                }
        }
        return req;
@@ -403,6 +432,7 @@ static void nbd_clear_que(struct nbd_device *lo)
        }
 }
 
+
 /*
  * We always wait for result of write, for now. It would be nice to make it optional
  * in future
@@ -511,7 +541,9 @@ static int nbd_ioctl(struct inode *inode, struct file *file,
                sreq.nr_sectors = 0;
                 if (!lo->sock)
                        return -EINVAL;
+               mutex_lock(&lo->tx_lock);
                 nbd_send_req(lo, &sreq);
+               mutex_unlock(&lo->tx_lock);
                 return 0;
  
        case NBD_CLEAR_SOCK:
@@ -555,6 +587,9 @@ static int nbd_ioctl(struct inode *inode, struct file *file,
                set_blocksize(inode->i_bdev, lo->blksize);
                set_capacity(lo->disk, lo->bytesize >> 9);
                return 0;
+       case NBD_SET_TIMEOUT:
+               lo->xmit_timeout = arg * HZ;
+               return 0;
        case NBD_SET_SIZE_BLOCKS:
                lo->bytesize = ((u64) arg) * lo->blksize;
                inode->i_bdev->bd_inode->i_size = lo->bytesize;
@@ -567,28 +602,16 @@ static int nbd_ioctl(struct inode *inode, struct file *file,
                error = nbd_do_it(lo);
                if (error)
                        return error;
-               /* on return tidy up in case we have a signal */
-               /* Forcibly shutdown the socket causing all listeners
-                * to error
-                *
-                * FIXME: This code is duplicated from sys_shutdown, but
-                * there should be a more generic interface rather than
-                * calling socket ops directly here */
-               mutex_lock(&lo->tx_lock);
-               if (lo->sock) {
-                       printk(KERN_WARNING "%s: shutting down socket\n",
-                               lo->disk->disk_name);
-                       lo->sock->ops->shutdown(lo->sock,
-                               SEND_SHUTDOWN|RCV_SHUTDOWN);
-                       lo->sock = NULL;
-               }
-               mutex_unlock(&lo->tx_lock);
+               sock_shutdown(lo, 1);
                file = lo->file;
                lo->file = NULL;
                nbd_clear_que(lo);
                printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
                if (file)
                        fput(file);
+               lo->bytesize = 0;
+               inode->i_bdev->bd_inode->i_size = 0;
+               set_capacity(lo->disk, 0);
                return lo->harderror;
        case NBD_CLEAR_QUE:
                /*
@@ -666,14 +689,14 @@ static int __init nbd_init(void)
                mutex_init(&nbd_dev[i].tx_lock);
                init_waitqueue_head(&nbd_dev[i].active_wq);
                nbd_dev[i].blksize = 1024;
-               nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */
+               nbd_dev[i].bytesize = 0;
                disk->major = NBD_MAJOR;
                disk->first_minor = i;
                disk->fops = &nbd_fops;
                disk->private_data = &nbd_dev[i];
                disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
                sprintf(disk->disk_name, "nbd%d", i);
-               set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */
+               set_capacity(disk, 0);
                add_disk(disk);
        }