Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 Pavel Machek <pavel@ucw.cz> | |
8 | * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com> | |
9 | * | |
dbf492d6 | 10 | * This file is released under GPLv2 or later. |
1da177e4 | 11 | * |
dbf492d6 | 12 | * (part of code stolen from loop.c) |
1da177e4 LT |
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> | |
4b2f0260 HX |
27 | #include <linux/compiler.h> |
28 | #include <linux/err.h> | |
29 | #include <linux/kernel.h> | |
1da177e4 LT |
30 | #include <net/sock.h> |
31 | ||
1da177e4 | 32 | #include <asm/uaccess.h> |
4b2f0260 | 33 | #include <asm/system.h> |
1da177e4 LT |
34 | #include <asm/types.h> |
35 | ||
36 | #include <linux/nbd.h> | |
37 | ||
38 | #define LO_MAGIC 0x68797548 | |
39 | ||
40 | #ifdef NDEBUG | |
41 | #define dprintk(flags, fmt...) | |
42 | #else /* NDEBUG */ | |
43 | #define dprintk(flags, fmt...) do { \ | |
44 | if (debugflags & (flags)) printk(KERN_DEBUG fmt); \ | |
45 | } while (0) | |
46 | #define DBG_IOCTL 0x0004 | |
47 | #define DBG_INIT 0x0010 | |
48 | #define DBG_EXIT 0x0020 | |
49 | #define DBG_BLKDEV 0x0100 | |
50 | #define DBG_RX 0x0200 | |
51 | #define DBG_TX 0x0400 | |
52 | static unsigned int debugflags; | |
53 | #endif /* NDEBUG */ | |
54 | ||
9c7a4169 | 55 | static unsigned int nbds_max = 16; |
1da177e4 LT |
56 | static struct nbd_device nbd_dev[MAX_NBD]; |
57 | ||
58 | /* | |
59 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
60 | * 1. Each NIC is essentially a synchronization point for all servers | |
61 | * accessed through that NIC so there's no need to have more locks | |
62 | * than NICs anyway. | |
63 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
64 | * down each lock to the point where they're actually slower than just | |
65 | * a single lock. | |
66 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
67 | */ | |
68 | static DEFINE_SPINLOCK(nbd_lock); | |
69 | ||
70 | #ifndef NDEBUG | |
71 | static const char *ioctl_cmd_to_ascii(int cmd) | |
72 | { | |
73 | switch (cmd) { | |
74 | case NBD_SET_SOCK: return "set-sock"; | |
75 | case NBD_SET_BLKSIZE: return "set-blksize"; | |
76 | case NBD_SET_SIZE: return "set-size"; | |
77 | case NBD_DO_IT: return "do-it"; | |
78 | case NBD_CLEAR_SOCK: return "clear-sock"; | |
79 | case NBD_CLEAR_QUE: return "clear-que"; | |
80 | case NBD_PRINT_DEBUG: return "print-debug"; | |
81 | case NBD_SET_SIZE_BLOCKS: return "set-size-blocks"; | |
82 | case NBD_DISCONNECT: return "disconnect"; | |
83 | case BLKROSET: return "set-read-only"; | |
84 | case BLKFLSBUF: return "flush-buffer-cache"; | |
85 | } | |
86 | return "unknown"; | |
87 | } | |
88 | ||
89 | static const char *nbdcmd_to_ascii(int cmd) | |
90 | { | |
91 | switch (cmd) { | |
92 | case NBD_CMD_READ: return "read"; | |
93 | case NBD_CMD_WRITE: return "write"; | |
94 | case NBD_CMD_DISC: return "disconnect"; | |
95 | } | |
96 | return "invalid"; | |
97 | } | |
98 | #endif /* NDEBUG */ | |
99 | ||
100 | static void nbd_end_request(struct request *req) | |
101 | { | |
102 | int uptodate = (req->errors == 0) ? 1 : 0; | |
165125e1 | 103 | struct request_queue *q = req->q; |
1da177e4 LT |
104 | unsigned long flags; |
105 | ||
106 | dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name, | |
107 | req, uptodate? "done": "failed"); | |
108 | ||
109 | spin_lock_irqsave(q->queue_lock, flags); | |
110 | if (!end_that_request_first(req, uptodate, req->nr_sectors)) { | |
8ffdc655 | 111 | end_that_request_last(req, uptodate); |
1da177e4 LT |
112 | } |
113 | spin_unlock_irqrestore(q->queue_lock, flags); | |
114 | } | |
115 | ||
116 | /* | |
117 | * Send or receive packet. | |
118 | */ | |
119 | static int sock_xmit(struct socket *sock, int send, void *buf, int size, | |
120 | int msg_flags) | |
121 | { | |
122 | int result; | |
123 | struct msghdr msg; | |
124 | struct kvec iov; | |
be0ef957 | 125 | sigset_t blocked, oldset; |
1da177e4 LT |
126 | |
127 | /* Allow interception of SIGKILL only | |
128 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
129 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
130 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 LT |
131 | |
132 | do { | |
133 | sock->sk->sk_allocation = GFP_NOIO; | |
134 | iov.iov_base = buf; | |
135 | iov.iov_len = size; | |
136 | msg.msg_name = NULL; | |
137 | msg.msg_namelen = 0; | |
138 | msg.msg_control = NULL; | |
139 | msg.msg_controllen = 0; | |
1da177e4 LT |
140 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
141 | ||
142 | if (send) | |
143 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
144 | else | |
145 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0); | |
146 | ||
147 | if (signal_pending(current)) { | |
148 | siginfo_t info; | |
1da177e4 | 149 | printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", |
be0ef957 ON |
150 | current->pid, current->comm, |
151 | dequeue_signal_lock(current, ¤t->blocked, &info)); | |
1da177e4 LT |
152 | result = -EINTR; |
153 | break; | |
154 | } | |
155 | ||
156 | if (result <= 0) { | |
157 | if (result == 0) | |
158 | result = -EPIPE; /* short read */ | |
159 | break; | |
160 | } | |
161 | size -= result; | |
162 | buf += result; | |
163 | } while (size > 0); | |
164 | ||
be0ef957 | 165 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
1da177e4 LT |
166 | |
167 | return result; | |
168 | } | |
169 | ||
170 | static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec, | |
171 | int flags) | |
172 | { | |
173 | int result; | |
174 | void *kaddr = kmap(bvec->bv_page); | |
175 | result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len, | |
176 | flags); | |
177 | kunmap(bvec->bv_page); | |
178 | return result; | |
179 | } | |
180 | ||
181 | static int nbd_send_req(struct nbd_device *lo, struct request *req) | |
182 | { | |
5705f702 | 183 | int result, flags; |
1da177e4 LT |
184 | struct nbd_request request; |
185 | unsigned long size = req->nr_sectors << 9; | |
186 | struct socket *sock = lo->sock; | |
187 | ||
188 | request.magic = htonl(NBD_REQUEST_MAGIC); | |
189 | request.type = htonl(nbd_cmd(req)); | |
190 | request.from = cpu_to_be64((u64) req->sector << 9); | |
191 | request.len = htonl(size); | |
192 | memcpy(request.handle, &req, sizeof(req)); | |
193 | ||
1da177e4 LT |
194 | dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n", |
195 | lo->disk->disk_name, req, | |
196 | nbdcmd_to_ascii(nbd_cmd(req)), | |
197 | (unsigned long long)req->sector << 9, | |
198 | req->nr_sectors << 9); | |
199 | result = sock_xmit(sock, 1, &request, sizeof(request), | |
200 | (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0); | |
201 | if (result <= 0) { | |
202 | printk(KERN_ERR "%s: Send control failed (result %d)\n", | |
203 | lo->disk->disk_name, result); | |
204 | goto error_out; | |
205 | } | |
206 | ||
207 | if (nbd_cmd(req) == NBD_CMD_WRITE) { | |
5705f702 N |
208 | struct req_iterator iter; |
209 | struct bio_vec *bvec; | |
1da177e4 LT |
210 | /* |
211 | * we are really probing at internals to determine | |
212 | * whether to set MSG_MORE or not... | |
213 | */ | |
5705f702 | 214 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 JA |
215 | flags = 0; |
216 | if (!rq_iter_last(req, iter)) | |
217 | flags = MSG_MORE; | |
218 | dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n", | |
219 | lo->disk->disk_name, req, bvec->bv_len); | |
220 | result = sock_send_bvec(sock, bvec, flags); | |
221 | if (result <= 0) { | |
222 | printk(KERN_ERR "%s: Send data failed (result %d)\n", | |
223 | lo->disk->disk_name, result); | |
224 | goto error_out; | |
225 | } | |
1da177e4 LT |
226 | } |
227 | } | |
1da177e4 LT |
228 | return 0; |
229 | ||
230 | error_out: | |
1da177e4 LT |
231 | return 1; |
232 | } | |
233 | ||
234 | static struct request *nbd_find_request(struct nbd_device *lo, char *handle) | |
235 | { | |
236 | struct request *req; | |
237 | struct list_head *tmp; | |
238 | struct request *xreq; | |
4b2f0260 | 239 | int err; |
1da177e4 LT |
240 | |
241 | memcpy(&xreq, handle, sizeof(xreq)); | |
242 | ||
4b2f0260 HX |
243 | err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq); |
244 | if (unlikely(err)) | |
245 | goto out; | |
246 | ||
1da177e4 LT |
247 | spin_lock(&lo->queue_lock); |
248 | list_for_each(tmp, &lo->queue_head) { | |
249 | req = list_entry(tmp, struct request, queuelist); | |
250 | if (req != xreq) | |
251 | continue; | |
252 | list_del_init(&req->queuelist); | |
253 | spin_unlock(&lo->queue_lock); | |
254 | return req; | |
255 | } | |
256 | spin_unlock(&lo->queue_lock); | |
4b2f0260 HX |
257 | |
258 | err = -ENOENT; | |
259 | ||
260 | out: | |
261 | return ERR_PTR(err); | |
1da177e4 LT |
262 | } |
263 | ||
264 | static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec) | |
265 | { | |
266 | int result; | |
267 | void *kaddr = kmap(bvec->bv_page); | |
268 | result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len, | |
269 | MSG_WAITALL); | |
270 | kunmap(bvec->bv_page); | |
271 | return result; | |
272 | } | |
273 | ||
274 | /* NULL returned = something went wrong, inform userspace */ | |
275 | static struct request *nbd_read_stat(struct nbd_device *lo) | |
276 | { | |
277 | int result; | |
278 | struct nbd_reply reply; | |
279 | struct request *req; | |
280 | struct socket *sock = lo->sock; | |
281 | ||
282 | reply.magic = 0; | |
283 | result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL); | |
284 | if (result <= 0) { | |
285 | printk(KERN_ERR "%s: Receive control failed (result %d)\n", | |
286 | lo->disk->disk_name, result); | |
287 | goto harderror; | |
288 | } | |
e4b57e08 MF |
289 | |
290 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
291 | printk(KERN_ERR "%s: Wrong magic (0x%lx)\n", | |
292 | lo->disk->disk_name, | |
293 | (unsigned long)ntohl(reply.magic)); | |
294 | result = -EPROTO; | |
295 | goto harderror; | |
296 | } | |
297 | ||
1da177e4 | 298 | req = nbd_find_request(lo, reply.handle); |
4b2f0260 HX |
299 | if (unlikely(IS_ERR(req))) { |
300 | result = PTR_ERR(req); | |
301 | if (result != -ENOENT) | |
302 | goto harderror; | |
303 | ||
1da177e4 LT |
304 | printk(KERN_ERR "%s: Unexpected reply (%p)\n", |
305 | lo->disk->disk_name, reply.handle); | |
306 | result = -EBADR; | |
307 | goto harderror; | |
308 | } | |
309 | ||
1da177e4 LT |
310 | if (ntohl(reply.error)) { |
311 | printk(KERN_ERR "%s: Other side returned error (%d)\n", | |
312 | lo->disk->disk_name, ntohl(reply.error)); | |
313 | req->errors++; | |
314 | return req; | |
315 | } | |
316 | ||
317 | dprintk(DBG_RX, "%s: request %p: got reply\n", | |
318 | lo->disk->disk_name, req); | |
319 | if (nbd_cmd(req) == NBD_CMD_READ) { | |
5705f702 N |
320 | struct req_iterator iter; |
321 | struct bio_vec *bvec; | |
322 | ||
323 | rq_for_each_segment(bvec, req, iter) { | |
6c92e699 JA |
324 | result = sock_recv_bvec(sock, bvec); |
325 | if (result <= 0) { | |
326 | printk(KERN_ERR "%s: Receive data failed (result %d)\n", | |
327 | lo->disk->disk_name, result); | |
328 | req->errors++; | |
329 | return req; | |
330 | } | |
331 | dprintk(DBG_RX, "%s: request %p: got %d bytes data\n", | |
332 | lo->disk->disk_name, req, bvec->bv_len); | |
1da177e4 LT |
333 | } |
334 | } | |
335 | return req; | |
336 | harderror: | |
337 | lo->harderror = result; | |
338 | return NULL; | |
339 | } | |
340 | ||
6b39bb65 PC |
341 | static ssize_t pid_show(struct gendisk *disk, char *page) |
342 | { | |
343 | return sprintf(page, "%ld\n", | |
344 | (long) ((struct nbd_device *)disk->private_data)->pid); | |
345 | } | |
346 | ||
347 | static struct disk_attribute pid_attr = { | |
348 | .attr = { .name = "pid", .mode = S_IRUGO }, | |
349 | .show = pid_show, | |
350 | }; | |
351 | ||
84963048 | 352 | static int nbd_do_it(struct nbd_device *lo) |
1da177e4 LT |
353 | { |
354 | struct request *req; | |
84963048 | 355 | int ret; |
1da177e4 LT |
356 | |
357 | BUG_ON(lo->magic != LO_MAGIC); | |
358 | ||
6b39bb65 | 359 | lo->pid = current->pid; |
84963048 WC |
360 | ret = sysfs_create_file(&lo->disk->kobj, &pid_attr.attr); |
361 | if (ret) { | |
362 | printk(KERN_ERR "nbd: sysfs_create_file failed!"); | |
363 | return ret; | |
364 | } | |
6b39bb65 | 365 | |
1da177e4 LT |
366 | while ((req = nbd_read_stat(lo)) != NULL) |
367 | nbd_end_request(req); | |
6b39bb65 PC |
368 | |
369 | sysfs_remove_file(&lo->disk->kobj, &pid_attr.attr); | |
84963048 | 370 | return 0; |
1da177e4 LT |
371 | } |
372 | ||
373 | static void nbd_clear_que(struct nbd_device *lo) | |
374 | { | |
375 | struct request *req; | |
376 | ||
377 | BUG_ON(lo->magic != LO_MAGIC); | |
378 | ||
4b2f0260 HX |
379 | /* |
380 | * Because we have set lo->sock to NULL under the tx_lock, all | |
381 | * modifications to the list must have completed by now. For | |
382 | * the same reason, the active_req must be NULL. | |
383 | * | |
384 | * As a consequence, we don't need to take the spin lock while | |
385 | * purging the list here. | |
386 | */ | |
387 | BUG_ON(lo->sock); | |
388 | BUG_ON(lo->active_req); | |
389 | ||
390 | while (!list_empty(&lo->queue_head)) { | |
391 | req = list_entry(lo->queue_head.next, struct request, | |
392 | queuelist); | |
393 | list_del_init(&req->queuelist); | |
394 | req->errors++; | |
395 | nbd_end_request(req); | |
396 | } | |
1da177e4 LT |
397 | } |
398 | ||
399 | /* | |
400 | * We always wait for result of write, for now. It would be nice to make it optional | |
401 | * in future | |
e654bc43 | 402 | * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
403 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
404 | */ | |
405 | ||
165125e1 | 406 | static void do_nbd_request(struct request_queue * q) |
1da177e4 LT |
407 | { |
408 | struct request *req; | |
409 | ||
410 | while ((req = elv_next_request(q)) != NULL) { | |
411 | struct nbd_device *lo; | |
412 | ||
413 | blkdev_dequeue_request(req); | |
4aff5e23 JA |
414 | dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n", |
415 | req->rq_disk->disk_name, req, req->cmd_type); | |
1da177e4 | 416 | |
4aff5e23 | 417 | if (!blk_fs_request(req)) |
1da177e4 LT |
418 | goto error_out; |
419 | ||
420 | lo = req->rq_disk->private_data; | |
421 | ||
422 | BUG_ON(lo->magic != LO_MAGIC); | |
423 | ||
1da177e4 LT |
424 | nbd_cmd(req) = NBD_CMD_READ; |
425 | if (rq_data_dir(req) == WRITE) { | |
426 | nbd_cmd(req) = NBD_CMD_WRITE; | |
427 | if (lo->flags & NBD_READ_ONLY) { | |
428 | printk(KERN_ERR "%s: Write on read-only\n", | |
429 | lo->disk->disk_name); | |
430 | goto error_out; | |
431 | } | |
432 | } | |
433 | ||
434 | req->errors = 0; | |
435 | spin_unlock_irq(q->queue_lock); | |
436 | ||
82d4dc5a | 437 | mutex_lock(&lo->tx_lock); |
4b2f0260 | 438 | if (unlikely(!lo->sock)) { |
82d4dc5a | 439 | mutex_unlock(&lo->tx_lock); |
4b2f0260 HX |
440 | printk(KERN_ERR "%s: Attempted send on closed socket\n", |
441 | lo->disk->disk_name); | |
1da177e4 LT |
442 | req->errors++; |
443 | nbd_end_request(req); | |
444 | spin_lock_irq(q->queue_lock); | |
445 | continue; | |
446 | } | |
447 | ||
4b2f0260 | 448 | lo->active_req = req; |
1da177e4 LT |
449 | |
450 | if (nbd_send_req(lo, req) != 0) { | |
451 | printk(KERN_ERR "%s: Request send failed\n", | |
452 | lo->disk->disk_name); | |
4b2f0260 HX |
453 | req->errors++; |
454 | nbd_end_request(req); | |
455 | } else { | |
456 | spin_lock(&lo->queue_lock); | |
457 | list_add(&req->queuelist, &lo->queue_head); | |
458 | spin_unlock(&lo->queue_lock); | |
1da177e4 LT |
459 | } |
460 | ||
4b2f0260 | 461 | lo->active_req = NULL; |
82d4dc5a | 462 | mutex_unlock(&lo->tx_lock); |
4b2f0260 HX |
463 | wake_up_all(&lo->active_wq); |
464 | ||
1da177e4 LT |
465 | spin_lock_irq(q->queue_lock); |
466 | continue; | |
467 | ||
468 | error_out: | |
469 | req->errors++; | |
470 | spin_unlock(q->queue_lock); | |
471 | nbd_end_request(req); | |
472 | spin_lock(q->queue_lock); | |
473 | } | |
474 | return; | |
475 | } | |
476 | ||
477 | static int nbd_ioctl(struct inode *inode, struct file *file, | |
478 | unsigned int cmd, unsigned long arg) | |
479 | { | |
480 | struct nbd_device *lo = inode->i_bdev->bd_disk->private_data; | |
481 | int error; | |
482 | struct request sreq ; | |
483 | ||
484 | if (!capable(CAP_SYS_ADMIN)) | |
485 | return -EPERM; | |
486 | ||
487 | BUG_ON(lo->magic != LO_MAGIC); | |
488 | ||
489 | /* Anyone capable of this syscall can do *real bad* things */ | |
490 | dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n", | |
491 | lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg); | |
492 | ||
493 | switch (cmd) { | |
494 | case NBD_DISCONNECT: | |
495 | printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name); | |
4aff5e23 | 496 | sreq.cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
497 | nbd_cmd(&sreq) = NBD_CMD_DISC; |
498 | /* | |
499 | * Set these to sane values in case server implementation | |
500 | * fails to check the request type first and also to keep | |
501 | * debugging output cleaner. | |
502 | */ | |
503 | sreq.sector = 0; | |
504 | sreq.nr_sectors = 0; | |
505 | if (!lo->sock) | |
506 | return -EINVAL; | |
507 | nbd_send_req(lo, &sreq); | |
508 | return 0; | |
509 | ||
510 | case NBD_CLEAR_SOCK: | |
511 | error = 0; | |
82d4dc5a | 512 | mutex_lock(&lo->tx_lock); |
1da177e4 | 513 | lo->sock = NULL; |
82d4dc5a | 514 | mutex_unlock(&lo->tx_lock); |
1da177e4 LT |
515 | file = lo->file; |
516 | lo->file = NULL; | |
1da177e4 | 517 | nbd_clear_que(lo); |
4b2f0260 | 518 | BUG_ON(!list_empty(&lo->queue_head)); |
1da177e4 LT |
519 | if (file) |
520 | fput(file); | |
521 | return error; | |
522 | case NBD_SET_SOCK: | |
523 | if (lo->file) | |
524 | return -EBUSY; | |
525 | error = -EINVAL; | |
526 | file = fget(arg); | |
527 | if (file) { | |
17506041 | 528 | inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
529 | if (S_ISSOCK(inode->i_mode)) { |
530 | lo->file = file; | |
531 | lo->sock = SOCKET_I(inode); | |
532 | error = 0; | |
533 | } else { | |
534 | fput(file); | |
535 | } | |
536 | } | |
537 | return error; | |
538 | case NBD_SET_BLKSIZE: | |
539 | lo->blksize = arg; | |
540 | lo->bytesize &= ~(lo->blksize-1); | |
541 | inode->i_bdev->bd_inode->i_size = lo->bytesize; | |
542 | set_blocksize(inode->i_bdev, lo->blksize); | |
543 | set_capacity(lo->disk, lo->bytesize >> 9); | |
544 | return 0; | |
545 | case NBD_SET_SIZE: | |
546 | lo->bytesize = arg & ~(lo->blksize-1); | |
547 | inode->i_bdev->bd_inode->i_size = lo->bytesize; | |
548 | set_blocksize(inode->i_bdev, lo->blksize); | |
549 | set_capacity(lo->disk, lo->bytesize >> 9); | |
550 | return 0; | |
551 | case NBD_SET_SIZE_BLOCKS: | |
552 | lo->bytesize = ((u64) arg) * lo->blksize; | |
553 | inode->i_bdev->bd_inode->i_size = lo->bytesize; | |
554 | set_blocksize(inode->i_bdev, lo->blksize); | |
555 | set_capacity(lo->disk, lo->bytesize >> 9); | |
556 | return 0; | |
557 | case NBD_DO_IT: | |
558 | if (!lo->file) | |
559 | return -EINVAL; | |
84963048 WC |
560 | error = nbd_do_it(lo); |
561 | if (error) | |
562 | return error; | |
1da177e4 LT |
563 | /* on return tidy up in case we have a signal */ |
564 | /* Forcibly shutdown the socket causing all listeners | |
565 | * to error | |
566 | * | |
567 | * FIXME: This code is duplicated from sys_shutdown, but | |
568 | * there should be a more generic interface rather than | |
569 | * calling socket ops directly here */ | |
82d4dc5a | 570 | mutex_lock(&lo->tx_lock); |
1da177e4 LT |
571 | if (lo->sock) { |
572 | printk(KERN_WARNING "%s: shutting down socket\n", | |
573 | lo->disk->disk_name); | |
574 | lo->sock->ops->shutdown(lo->sock, | |
575 | SEND_SHUTDOWN|RCV_SHUTDOWN); | |
576 | lo->sock = NULL; | |
577 | } | |
82d4dc5a | 578 | mutex_unlock(&lo->tx_lock); |
1da177e4 LT |
579 | file = lo->file; |
580 | lo->file = NULL; | |
1da177e4 LT |
581 | nbd_clear_que(lo); |
582 | printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name); | |
583 | if (file) | |
584 | fput(file); | |
585 | return lo->harderror; | |
586 | case NBD_CLEAR_QUE: | |
4b2f0260 HX |
587 | /* |
588 | * This is for compatibility only. The queue is always cleared | |
589 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
590 | */ | |
591 | BUG_ON(!lo->sock && !list_empty(&lo->queue_head)); | |
1da177e4 LT |
592 | return 0; |
593 | case NBD_PRINT_DEBUG: | |
594 | printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n", | |
595 | inode->i_bdev->bd_disk->disk_name, | |
596 | lo->queue_head.next, lo->queue_head.prev, | |
597 | &lo->queue_head); | |
598 | return 0; | |
599 | } | |
600 | return -EINVAL; | |
601 | } | |
602 | ||
603 | static struct block_device_operations nbd_fops = | |
604 | { | |
605 | .owner = THIS_MODULE, | |
606 | .ioctl = nbd_ioctl, | |
607 | }; | |
608 | ||
609 | /* | |
610 | * And here should be modules and kernel interface | |
611 | * (Just smiley confuses emacs :-) | |
612 | */ | |
613 | ||
614 | static int __init nbd_init(void) | |
615 | { | |
616 | int err = -ENOMEM; | |
617 | int i; | |
618 | ||
5b7b18cc | 619 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 620 | |
40be0c28 LMB |
621 | if (nbds_max > MAX_NBD) { |
622 | printk(KERN_CRIT "nbd: cannot allocate more than %u nbds; %u requested.\n", MAX_NBD, | |
623 | nbds_max); | |
624 | return -EINVAL; | |
625 | } | |
626 | ||
627 | for (i = 0; i < nbds_max; i++) { | |
1da177e4 LT |
628 | struct gendisk *disk = alloc_disk(1); |
629 | if (!disk) | |
630 | goto out; | |
631 | nbd_dev[i].disk = disk; | |
632 | /* | |
633 | * The new linux 2.5 block layer implementation requires | |
634 | * every gendisk to have its very own request_queue struct. | |
635 | * These structs are big so we dynamically allocate them. | |
636 | */ | |
637 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
638 | if (!disk->queue) { | |
639 | put_disk(disk); | |
640 | goto out; | |
641 | } | |
642 | } | |
643 | ||
644 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
645 | err = -EIO; | |
646 | goto out; | |
647 | } | |
648 | ||
649 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
650 | dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags); | |
651 | ||
40be0c28 | 652 | for (i = 0; i < nbds_max; i++) { |
1da177e4 LT |
653 | struct gendisk *disk = nbd_dev[i].disk; |
654 | nbd_dev[i].file = NULL; | |
655 | nbd_dev[i].magic = LO_MAGIC; | |
656 | nbd_dev[i].flags = 0; | |
657 | spin_lock_init(&nbd_dev[i].queue_lock); | |
658 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 659 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 660 | init_waitqueue_head(&nbd_dev[i].active_wq); |
1da177e4 LT |
661 | nbd_dev[i].blksize = 1024; |
662 | nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */ | |
663 | disk->major = NBD_MAJOR; | |
664 | disk->first_minor = i; | |
665 | disk->fops = &nbd_fops; | |
666 | disk->private_data = &nbd_dev[i]; | |
667 | disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO; | |
668 | sprintf(disk->disk_name, "nbd%d", i); | |
1da177e4 LT |
669 | set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */ |
670 | add_disk(disk); | |
671 | } | |
672 | ||
673 | return 0; | |
674 | out: | |
675 | while (i--) { | |
676 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
677 | put_disk(nbd_dev[i].disk); | |
678 | } | |
679 | return err; | |
680 | } | |
681 | ||
682 | static void __exit nbd_cleanup(void) | |
683 | { | |
684 | int i; | |
40be0c28 | 685 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 686 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 687 | nbd_dev[i].magic = 0; |
1da177e4 LT |
688 | if (disk) { |
689 | del_gendisk(disk); | |
690 | blk_cleanup_queue(disk->queue); | |
691 | put_disk(disk); | |
692 | } | |
693 | } | |
1da177e4 LT |
694 | unregister_blkdev(NBD_MAJOR, "nbd"); |
695 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); | |
696 | } | |
697 | ||
698 | module_init(nbd_init); | |
699 | module_exit(nbd_cleanup); | |
700 | ||
701 | MODULE_DESCRIPTION("Network Block Device"); | |
702 | MODULE_LICENSE("GPL"); | |
703 | ||
40be0c28 LMB |
704 | module_param(nbds_max, int, 0444); |
705 | MODULE_PARM_DESC(nbds_max, "How many network block devices to initialize."); | |
1da177e4 LT |
706 | #ifndef NDEBUG |
707 | module_param(debugflags, int, 0644); | |
708 | MODULE_PARM_DESC(debugflags, "flags for controlling debug output"); | |
709 | #endif |