block: do not pass disk names as format strings
[pandora-kernel.git] / drivers / block / nbd.c
1 /*
2  * Network block device - make block devices work over TCP
3  *
4  * Note that you can not swap over this thing, yet. Seems to work but
5  * deadlocks sometimes - you can not swap over TCP in general.
6  * 
7  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
9  *
10  * This file is released under GPLv2 or later.
11  *
12  * (part of code stolen from loop.c)
13  */
14
15 #include <linux/major.h>
16
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/sched.h>
21 #include <linux/fs.h>
22 #include <linux/bio.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/file.h>
26 #include <linux/ioctl.h>
27 #include <linux/mutex.h>
28 #include <linux/compiler.h>
29 #include <linux/err.h>
30 #include <linux/kernel.h>
31 #include <linux/slab.h>
32 #include <net/sock.h>
33 #include <linux/net.h>
34 #include <linux/kthread.h>
35
36 #include <asm/uaccess.h>
37 #include <asm/system.h>
38 #include <asm/types.h>
39
40 #include <linux/nbd.h>
41
42 #define LO_MAGIC 0x68797548
43
44 #ifdef NDEBUG
45 #define dprintk(flags, fmt...)
46 #else /* NDEBUG */
47 #define dprintk(flags, fmt...) do { \
48         if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
49 } while (0)
50 #define DBG_IOCTL       0x0004
51 #define DBG_INIT        0x0010
52 #define DBG_EXIT        0x0020
53 #define DBG_BLKDEV      0x0100
54 #define DBG_RX          0x0200
55 #define DBG_TX          0x0400
56 static unsigned int debugflags;
57 #endif /* NDEBUG */
58
59 static unsigned int nbds_max = 16;
60 static struct nbd_device *nbd_dev;
61 static int max_part;
62
63 /*
64  * Use just one lock (or at most 1 per NIC). Two arguments for this:
65  * 1. Each NIC is essentially a synchronization point for all servers
66  *    accessed through that NIC so there's no need to have more locks
67  *    than NICs anyway.
68  * 2. More locks lead to more "Dirty cache line bouncing" which will slow
69  *    down each lock to the point where they're actually slower than just
70  *    a single lock.
71  * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
72  */
73 static DEFINE_SPINLOCK(nbd_lock);
74
75 #ifndef NDEBUG
76 static const char *ioctl_cmd_to_ascii(int cmd)
77 {
78         switch (cmd) {
79         case NBD_SET_SOCK: return "set-sock";
80         case NBD_SET_BLKSIZE: return "set-blksize";
81         case NBD_SET_SIZE: return "set-size";
82         case NBD_DO_IT: return "do-it";
83         case NBD_CLEAR_SOCK: return "clear-sock";
84         case NBD_CLEAR_QUE: return "clear-que";
85         case NBD_PRINT_DEBUG: return "print-debug";
86         case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
87         case NBD_DISCONNECT: return "disconnect";
88         case BLKROSET: return "set-read-only";
89         case BLKFLSBUF: return "flush-buffer-cache";
90         }
91         return "unknown";
92 }
93
94 static const char *nbdcmd_to_ascii(int cmd)
95 {
96         switch (cmd) {
97         case  NBD_CMD_READ: return "read";
98         case NBD_CMD_WRITE: return "write";
99         case  NBD_CMD_DISC: return "disconnect";
100         }
101         return "invalid";
102 }
103 #endif /* NDEBUG */
104
105 static void nbd_end_request(struct request *req)
106 {
107         int error = req->errors ? -EIO : 0;
108         struct request_queue *q = req->q;
109         unsigned long flags;
110
111         dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
112                         req, error ? "failed" : "done");
113
114         spin_lock_irqsave(q->queue_lock, flags);
115         __blk_end_request_all(req, error);
116         spin_unlock_irqrestore(q->queue_lock, flags);
117 }
118
119 static void sock_shutdown(struct nbd_device *lo, int lock)
120 {
121         /* Forcibly shutdown the socket causing all listeners
122          * to error
123          *
124          * FIXME: This code is duplicated from sys_shutdown, but
125          * there should be a more generic interface rather than
126          * calling socket ops directly here */
127         if (lock)
128                 mutex_lock(&lo->tx_lock);
129         if (lo->sock) {
130                 dev_warn(disk_to_dev(lo->disk), "shutting down socket\n");
131                 kernel_sock_shutdown(lo->sock, SHUT_RDWR);
132                 lo->sock = NULL;
133         }
134         if (lock)
135                 mutex_unlock(&lo->tx_lock);
136 }
137
138 static void nbd_xmit_timeout(unsigned long arg)
139 {
140         struct task_struct *task = (struct task_struct *)arg;
141
142         printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
143                 task->comm, task->pid);
144         force_sig(SIGKILL, task);
145 }
146
147 /*
148  *  Send or receive packet.
149  */
150 static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
151                 int msg_flags)
152 {
153         struct socket *sock = lo->sock;
154         int result;
155         struct msghdr msg;
156         struct kvec iov;
157         sigset_t blocked, oldset;
158
159         if (unlikely(!sock)) {
160                 dev_err(disk_to_dev(lo->disk),
161                         "Attempted %s on closed socket in sock_xmit\n",
162                         (send ? "send" : "recv"));
163                 return -EINVAL;
164         }
165
166         /* Allow interception of SIGKILL only
167          * Don't allow other signals to interrupt the transmission */
168         siginitsetinv(&blocked, sigmask(SIGKILL));
169         sigprocmask(SIG_SETMASK, &blocked, &oldset);
170
171         do {
172                 sock->sk->sk_allocation = GFP_NOIO;
173                 iov.iov_base = buf;
174                 iov.iov_len = size;
175                 msg.msg_name = NULL;
176                 msg.msg_namelen = 0;
177                 msg.msg_control = NULL;
178                 msg.msg_controllen = 0;
179                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
180
181                 if (send) {
182                         struct timer_list ti;
183
184                         if (lo->xmit_timeout) {
185                                 init_timer(&ti);
186                                 ti.function = nbd_xmit_timeout;
187                                 ti.data = (unsigned long)current;
188                                 ti.expires = jiffies + lo->xmit_timeout;
189                                 add_timer(&ti);
190                         }
191                         result = kernel_sendmsg(sock, &msg, &iov, 1, size);
192                         if (lo->xmit_timeout)
193                                 del_timer_sync(&ti);
194                 } else
195                         result = kernel_recvmsg(sock, &msg, &iov, 1, size,
196                                                 msg.msg_flags);
197
198                 if (signal_pending(current)) {
199                         siginfo_t info;
200                         printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
201                                 task_pid_nr(current), current->comm,
202                                 dequeue_signal_lock(current, &current->blocked, &info));
203                         result = -EINTR;
204                         sock_shutdown(lo, !send);
205                         break;
206                 }
207
208                 if (result <= 0) {
209                         if (result == 0)
210                                 result = -EPIPE; /* short read */
211                         break;
212                 }
213                 size -= result;
214                 buf += result;
215         } while (size > 0);
216
217         sigprocmask(SIG_SETMASK, &oldset, NULL);
218
219         return result;
220 }
221
222 static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
223                 int flags)
224 {
225         int result;
226         void *kaddr = kmap(bvec->bv_page);
227         result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags);
228         kunmap(bvec->bv_page);
229         return result;
230 }
231
232 /* always call with the tx_lock held */
233 static int nbd_send_req(struct nbd_device *lo, struct request *req)
234 {
235         int result, flags;
236         struct nbd_request request;
237         unsigned long size = blk_rq_bytes(req);
238
239         request.magic = htonl(NBD_REQUEST_MAGIC);
240         request.type = htonl(nbd_cmd(req));
241         request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
242         request.len = htonl(size);
243         memcpy(request.handle, &req, sizeof(req));
244
245         dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
246                         lo->disk->disk_name, req,
247                         nbdcmd_to_ascii(nbd_cmd(req)),
248                         (unsigned long long)blk_rq_pos(req) << 9,
249                         blk_rq_bytes(req));
250         result = sock_xmit(lo, 1, &request, sizeof(request),
251                         (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
252         if (result <= 0) {
253                 dev_err(disk_to_dev(lo->disk),
254                         "Send control failed (result %d)\n", result);
255                 goto error_out;
256         }
257
258         if (nbd_cmd(req) == NBD_CMD_WRITE) {
259                 struct req_iterator iter;
260                 struct bio_vec *bvec;
261                 /*
262                  * we are really probing at internals to determine
263                  * whether to set MSG_MORE or not...
264                  */
265                 rq_for_each_segment(bvec, req, iter) {
266                         flags = 0;
267                         if (!rq_iter_last(req, iter))
268                                 flags = MSG_MORE;
269                         dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
270                                         lo->disk->disk_name, req, bvec->bv_len);
271                         result = sock_send_bvec(lo, bvec, flags);
272                         if (result <= 0) {
273                                 dev_err(disk_to_dev(lo->disk),
274                                         "Send data failed (result %d)\n",
275                                         result);
276                                 goto error_out;
277                         }
278                 }
279         }
280         return 0;
281
282 error_out:
283         return -EIO;
284 }
285
286 static struct request *nbd_find_request(struct nbd_device *lo,
287                                         struct request *xreq)
288 {
289         struct request *req, *tmp;
290         int err;
291
292         err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
293         if (unlikely(err))
294                 goto out;
295
296         spin_lock(&lo->queue_lock);
297         list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
298                 if (req != xreq)
299                         continue;
300                 list_del_init(&req->queuelist);
301                 spin_unlock(&lo->queue_lock);
302                 return req;
303         }
304         spin_unlock(&lo->queue_lock);
305
306         err = -ENOENT;
307
308 out:
309         return ERR_PTR(err);
310 }
311
312 static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
313 {
314         int result;
315         void *kaddr = kmap(bvec->bv_page);
316         result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
317                         MSG_WAITALL);
318         kunmap(bvec->bv_page);
319         return result;
320 }
321
322 /* NULL returned = something went wrong, inform userspace */
323 static struct request *nbd_read_stat(struct nbd_device *lo)
324 {
325         int result;
326         struct nbd_reply reply;
327         struct request *req;
328
329         reply.magic = 0;
330         result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
331         if (result <= 0) {
332                 dev_err(disk_to_dev(lo->disk),
333                         "Receive control failed (result %d)\n", result);
334                 goto harderror;
335         }
336
337         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
338                 dev_err(disk_to_dev(lo->disk), "Wrong magic (0x%lx)\n",
339                                 (unsigned long)ntohl(reply.magic));
340                 result = -EPROTO;
341                 goto harderror;
342         }
343
344         req = nbd_find_request(lo, *(struct request **)reply.handle);
345         if (IS_ERR(req)) {
346                 result = PTR_ERR(req);
347                 if (result != -ENOENT)
348                         goto harderror;
349
350                 dev_err(disk_to_dev(lo->disk), "Unexpected reply (%p)\n",
351                         reply.handle);
352                 result = -EBADR;
353                 goto harderror;
354         }
355
356         if (ntohl(reply.error)) {
357                 dev_err(disk_to_dev(lo->disk), "Other side returned error (%d)\n",
358                         ntohl(reply.error));
359                 req->errors++;
360                 return req;
361         }
362
363         dprintk(DBG_RX, "%s: request %p: got reply\n",
364                         lo->disk->disk_name, req);
365         if (nbd_cmd(req) == NBD_CMD_READ) {
366                 struct req_iterator iter;
367                 struct bio_vec *bvec;
368
369                 rq_for_each_segment(bvec, req, iter) {
370                         result = sock_recv_bvec(lo, bvec);
371                         if (result <= 0) {
372                                 dev_err(disk_to_dev(lo->disk), "Receive data failed (result %d)\n",
373                                         result);
374                                 req->errors++;
375                                 return req;
376                         }
377                         dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
378                                 lo->disk->disk_name, req, bvec->bv_len);
379                 }
380         }
381         return req;
382 harderror:
383         lo->harderror = result;
384         return NULL;
385 }
386
387 static ssize_t pid_show(struct device *dev,
388                         struct device_attribute *attr, char *buf)
389 {
390         struct gendisk *disk = dev_to_disk(dev);
391
392         return sprintf(buf, "%ld\n",
393                 (long) ((struct nbd_device *)disk->private_data)->pid);
394 }
395
396 static struct device_attribute pid_attr = {
397         .attr = { .name = "pid", .mode = S_IRUGO},
398         .show = pid_show,
399 };
400
401 static int nbd_do_it(struct nbd_device *lo)
402 {
403         struct request *req;
404         int ret;
405
406         BUG_ON(lo->magic != LO_MAGIC);
407
408         lo->pid = task_pid_nr(current);
409         ret = device_create_file(disk_to_dev(lo->disk), &pid_attr);
410         if (ret) {
411                 dev_err(disk_to_dev(lo->disk), "device_create_file failed!\n");
412                 lo->pid = 0;
413                 return ret;
414         }
415
416         while ((req = nbd_read_stat(lo)) != NULL)
417                 nbd_end_request(req);
418
419         device_remove_file(disk_to_dev(lo->disk), &pid_attr);
420         lo->pid = 0;
421         return 0;
422 }
423
424 static void nbd_clear_que(struct nbd_device *lo)
425 {
426         struct request *req;
427
428         BUG_ON(lo->magic != LO_MAGIC);
429
430         /*
431          * Because we have set lo->sock to NULL under the tx_lock, all
432          * modifications to the list must have completed by now.  For
433          * the same reason, the active_req must be NULL.
434          *
435          * As a consequence, we don't need to take the spin lock while
436          * purging the list here.
437          */
438         BUG_ON(lo->sock);
439         BUG_ON(lo->active_req);
440
441         while (!list_empty(&lo->queue_head)) {
442                 req = list_entry(lo->queue_head.next, struct request,
443                                  queuelist);
444                 list_del_init(&req->queuelist);
445                 req->errors++;
446                 nbd_end_request(req);
447         }
448
449         while (!list_empty(&lo->waiting_queue)) {
450                 req = list_entry(lo->waiting_queue.next, struct request,
451                                  queuelist);
452                 list_del_init(&req->queuelist);
453                 req->errors++;
454                 nbd_end_request(req);
455         }
456 }
457
458
459 static void nbd_handle_req(struct nbd_device *lo, struct request *req)
460 {
461         if (req->cmd_type != REQ_TYPE_FS)
462                 goto error_out;
463
464         nbd_cmd(req) = NBD_CMD_READ;
465         if (rq_data_dir(req) == WRITE) {
466                 nbd_cmd(req) = NBD_CMD_WRITE;
467                 if (lo->flags & NBD_READ_ONLY) {
468                         dev_err(disk_to_dev(lo->disk),
469                                 "Write on read-only\n");
470                         goto error_out;
471                 }
472         }
473
474         req->errors = 0;
475
476         mutex_lock(&lo->tx_lock);
477         if (unlikely(!lo->sock)) {
478                 mutex_unlock(&lo->tx_lock);
479                 dev_err(disk_to_dev(lo->disk),
480                         "Attempted send on closed socket\n");
481                 goto error_out;
482         }
483
484         lo->active_req = req;
485
486         if (nbd_send_req(lo, req) != 0) {
487                 dev_err(disk_to_dev(lo->disk), "Request send failed\n");
488                 req->errors++;
489                 nbd_end_request(req);
490         } else {
491                 spin_lock(&lo->queue_lock);
492                 list_add(&req->queuelist, &lo->queue_head);
493                 spin_unlock(&lo->queue_lock);
494         }
495
496         lo->active_req = NULL;
497         mutex_unlock(&lo->tx_lock);
498         wake_up_all(&lo->active_wq);
499
500         return;
501
502 error_out:
503         req->errors++;
504         nbd_end_request(req);
505 }
506
507 static int nbd_thread(void *data)
508 {
509         struct nbd_device *lo = data;
510         struct request *req;
511
512         set_user_nice(current, -20);
513         while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) {
514                 /* wait for something to do */
515                 wait_event_interruptible(lo->waiting_wq,
516                                          kthread_should_stop() ||
517                                          !list_empty(&lo->waiting_queue));
518
519                 /* extract request */
520                 if (list_empty(&lo->waiting_queue))
521                         continue;
522
523                 spin_lock_irq(&lo->queue_lock);
524                 req = list_entry(lo->waiting_queue.next, struct request,
525                                  queuelist);
526                 list_del_init(&req->queuelist);
527                 spin_unlock_irq(&lo->queue_lock);
528
529                 /* handle request */
530                 nbd_handle_req(lo, req);
531         }
532         return 0;
533 }
534
535 /*
536  * We always wait for result of write, for now. It would be nice to make it optional
537  * in future
538  * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
539  *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
540  */
541
542 static void do_nbd_request(struct request_queue *q)
543 {
544         struct request *req;
545         
546         while ((req = blk_fetch_request(q)) != NULL) {
547                 struct nbd_device *lo;
548
549                 spin_unlock_irq(q->queue_lock);
550
551                 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
552                                 req->rq_disk->disk_name, req, req->cmd_type);
553
554                 lo = req->rq_disk->private_data;
555
556                 BUG_ON(lo->magic != LO_MAGIC);
557
558                 if (unlikely(!lo->sock)) {
559                         dev_err(disk_to_dev(lo->disk),
560                                 "Attempted send on closed socket\n");
561                         req->errors++;
562                         nbd_end_request(req);
563                         spin_lock_irq(q->queue_lock);
564                         continue;
565                 }
566
567                 spin_lock_irq(&lo->queue_lock);
568                 list_add_tail(&req->queuelist, &lo->waiting_queue);
569                 spin_unlock_irq(&lo->queue_lock);
570
571                 wake_up(&lo->waiting_wq);
572
573                 spin_lock_irq(q->queue_lock);
574         }
575 }
576
577 /* Must be called with tx_lock held */
578
579 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *lo,
580                        unsigned int cmd, unsigned long arg)
581 {
582         switch (cmd) {
583         case NBD_DISCONNECT: {
584                 struct request sreq;
585
586                 dev_info(disk_to_dev(lo->disk), "NBD_DISCONNECT\n");
587                 if (!lo->sock)
588                         return -EINVAL;
589
590                 mutex_unlock(&lo->tx_lock);
591                 fsync_bdev(bdev);
592                 mutex_lock(&lo->tx_lock);
593                 blk_rq_init(NULL, &sreq);
594                 sreq.cmd_type = REQ_TYPE_SPECIAL;
595                 nbd_cmd(&sreq) = NBD_CMD_DISC;
596
597                 /* Check again after getting mutex back.  */
598                 if (!lo->sock)
599                         return -EINVAL;
600
601                 nbd_send_req(lo, &sreq);
602                 return 0;
603         }
604  
605         case NBD_CLEAR_SOCK: {
606                 struct file *file;
607
608                 lo->sock = NULL;
609                 file = lo->file;
610                 lo->file = NULL;
611                 nbd_clear_que(lo);
612                 BUG_ON(!list_empty(&lo->queue_head));
613                 BUG_ON(!list_empty(&lo->waiting_queue));
614                 kill_bdev(bdev);
615                 if (file)
616                         fput(file);
617                 return 0;
618         }
619
620         case NBD_SET_SOCK: {
621                 struct file *file;
622                 if (lo->file)
623                         return -EBUSY;
624                 file = fget(arg);
625                 if (file) {
626                         struct inode *inode = file->f_path.dentry->d_inode;
627                         if (S_ISSOCK(inode->i_mode)) {
628                                 lo->file = file;
629                                 lo->sock = SOCKET_I(inode);
630                                 if (max_part > 0)
631                                         bdev->bd_invalidated = 1;
632                                 return 0;
633                         } else {
634                                 fput(file);
635                         }
636                 }
637                 return -EINVAL;
638         }
639
640         case NBD_SET_BLKSIZE:
641                 lo->blksize = arg;
642                 lo->bytesize &= ~(lo->blksize-1);
643                 bdev->bd_inode->i_size = lo->bytesize;
644                 set_blocksize(bdev, lo->blksize);
645                 set_capacity(lo->disk, lo->bytesize >> 9);
646                 return 0;
647
648         case NBD_SET_SIZE:
649                 lo->bytesize = arg & ~(lo->blksize-1);
650                 bdev->bd_inode->i_size = lo->bytesize;
651                 set_blocksize(bdev, lo->blksize);
652                 set_capacity(lo->disk, lo->bytesize >> 9);
653                 return 0;
654
655         case NBD_SET_TIMEOUT:
656                 lo->xmit_timeout = arg * HZ;
657                 return 0;
658
659         case NBD_SET_SIZE_BLOCKS:
660                 lo->bytesize = ((u64) arg) * lo->blksize;
661                 bdev->bd_inode->i_size = lo->bytesize;
662                 set_blocksize(bdev, lo->blksize);
663                 set_capacity(lo->disk, lo->bytesize >> 9);
664                 return 0;
665
666         case NBD_DO_IT: {
667                 struct task_struct *thread;
668                 struct file *file;
669                 int error;
670
671                 if (lo->pid)
672                         return -EBUSY;
673                 if (!lo->file)
674                         return -EINVAL;
675
676                 mutex_unlock(&lo->tx_lock);
677
678                 thread = kthread_create(nbd_thread, lo, "%s",
679                                         lo->disk->disk_name);
680                 if (IS_ERR(thread)) {
681                         mutex_lock(&lo->tx_lock);
682                         return PTR_ERR(thread);
683                 }
684                 wake_up_process(thread);
685                 error = nbd_do_it(lo);
686                 kthread_stop(thread);
687
688                 mutex_lock(&lo->tx_lock);
689                 if (error)
690                         return error;
691                 sock_shutdown(lo, 0);
692                 file = lo->file;
693                 lo->file = NULL;
694                 nbd_clear_que(lo);
695                 dev_warn(disk_to_dev(lo->disk), "queue cleared\n");
696                 kill_bdev(bdev);
697                 if (file)
698                         fput(file);
699                 lo->bytesize = 0;
700                 bdev->bd_inode->i_size = 0;
701                 set_capacity(lo->disk, 0);
702                 if (max_part > 0)
703                         ioctl_by_bdev(bdev, BLKRRPART, 0);
704                 return lo->harderror;
705         }
706
707         case NBD_CLEAR_QUE:
708                 /*
709                  * This is for compatibility only.  The queue is always cleared
710                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
711                  */
712                 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
713                 return 0;
714
715         case NBD_PRINT_DEBUG:
716                 dev_info(disk_to_dev(lo->disk),
717                         "next = %p, prev = %p, head = %p\n",
718                         lo->queue_head.next, lo->queue_head.prev,
719                         &lo->queue_head);
720                 return 0;
721         }
722         return -ENOTTY;
723 }
724
725 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
726                      unsigned int cmd, unsigned long arg)
727 {
728         struct nbd_device *lo = bdev->bd_disk->private_data;
729         int error;
730
731         if (!capable(CAP_SYS_ADMIN))
732                 return -EPERM;
733
734         BUG_ON(lo->magic != LO_MAGIC);
735
736         /* Anyone capable of this syscall can do *real bad* things */
737         dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
738                         lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
739
740         mutex_lock(&lo->tx_lock);
741         error = __nbd_ioctl(bdev, lo, cmd, arg);
742         mutex_unlock(&lo->tx_lock);
743
744         return error;
745 }
746
747 static const struct block_device_operations nbd_fops =
748 {
749         .owner =        THIS_MODULE,
750         .ioctl =        nbd_ioctl,
751 };
752
753 /*
754  * And here should be modules and kernel interface 
755  *  (Just smiley confuses emacs :-)
756  */
757
758 static int __init nbd_init(void)
759 {
760         int err = -ENOMEM;
761         int i;
762         int part_shift;
763
764         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
765
766         if (max_part < 0) {
767                 printk(KERN_ERR "nbd: max_part must be >= 0\n");
768                 return -EINVAL;
769         }
770
771         nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
772         if (!nbd_dev)
773                 return -ENOMEM;
774
775         part_shift = 0;
776         if (max_part > 0) {
777                 part_shift = fls(max_part);
778
779                 /*
780                  * Adjust max_part according to part_shift as it is exported
781                  * to user space so that user can know the max number of
782                  * partition kernel should be able to manage.
783                  *
784                  * Note that -1 is required because partition 0 is reserved
785                  * for the whole disk.
786                  */
787                 max_part = (1UL << part_shift) - 1;
788         }
789
790         if ((1UL << part_shift) > DISK_MAX_PARTS)
791                 return -EINVAL;
792
793         if (nbds_max > 1UL << (MINORBITS - part_shift))
794                 return -EINVAL;
795
796         for (i = 0; i < nbds_max; i++) {
797                 struct gendisk *disk = alloc_disk(1 << part_shift);
798                 if (!disk)
799                         goto out;
800                 nbd_dev[i].disk = disk;
801                 /*
802                  * The new linux 2.5 block layer implementation requires
803                  * every gendisk to have its very own request_queue struct.
804                  * These structs are big so we dynamically allocate them.
805                  */
806                 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
807                 if (!disk->queue) {
808                         put_disk(disk);
809                         goto out;
810                 }
811                 /*
812                  * Tell the block layer that we are not a rotational device
813                  */
814                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
815         }
816
817         if (register_blkdev(NBD_MAJOR, "nbd")) {
818                 err = -EIO;
819                 goto out;
820         }
821
822         printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
823         dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
824
825         for (i = 0; i < nbds_max; i++) {
826                 struct gendisk *disk = nbd_dev[i].disk;
827                 nbd_dev[i].file = NULL;
828                 nbd_dev[i].magic = LO_MAGIC;
829                 nbd_dev[i].flags = 0;
830                 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
831                 spin_lock_init(&nbd_dev[i].queue_lock);
832                 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
833                 mutex_init(&nbd_dev[i].tx_lock);
834                 init_waitqueue_head(&nbd_dev[i].active_wq);
835                 init_waitqueue_head(&nbd_dev[i].waiting_wq);
836                 nbd_dev[i].blksize = 1024;
837                 nbd_dev[i].bytesize = 0;
838                 disk->major = NBD_MAJOR;
839                 disk->first_minor = i << part_shift;
840                 disk->fops = &nbd_fops;
841                 disk->private_data = &nbd_dev[i];
842                 sprintf(disk->disk_name, "nbd%d", i);
843                 set_capacity(disk, 0);
844                 add_disk(disk);
845         }
846
847         return 0;
848 out:
849         while (i--) {
850                 blk_cleanup_queue(nbd_dev[i].disk->queue);
851                 put_disk(nbd_dev[i].disk);
852         }
853         kfree(nbd_dev);
854         return err;
855 }
856
857 static void __exit nbd_cleanup(void)
858 {
859         int i;
860         for (i = 0; i < nbds_max; i++) {
861                 struct gendisk *disk = nbd_dev[i].disk;
862                 nbd_dev[i].magic = 0;
863                 if (disk) {
864                         del_gendisk(disk);
865                         blk_cleanup_queue(disk->queue);
866                         put_disk(disk);
867                 }
868         }
869         unregister_blkdev(NBD_MAJOR, "nbd");
870         kfree(nbd_dev);
871         printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
872 }
873
874 module_init(nbd_init);
875 module_exit(nbd_cleanup);
876
877 MODULE_DESCRIPTION("Network Block Device");
878 MODULE_LICENSE("GPL");
879
880 module_param(nbds_max, int, 0444);
881 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
882 module_param(max_part, int, 0444);
883 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
884 #ifndef NDEBUG
885 module_param(debugflags, int, 0644);
886 MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
887 #endif