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 | * | |
a2531293 | 7 | * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz> |
1da177e4 LT |
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> | |
2a48fc0a | 27 | #include <linux/mutex.h> |
4b2f0260 HX |
28 | #include <linux/compiler.h> |
29 | #include <linux/err.h> | |
30 | #include <linux/kernel.h> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
1da177e4 | 32 | #include <net/sock.h> |
91cf45f0 | 33 | #include <linux/net.h> |
48cf6061 | 34 | #include <linux/kthread.h> |
1da177e4 | 35 | |
1da177e4 | 36 | #include <asm/uaccess.h> |
4b2f0260 | 37 | #include <asm/system.h> |
1da177e4 LT |
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 | ||
9c7a4169 | 59 | static unsigned int nbds_max = 16; |
20a8143e | 60 | static struct nbd_device *nbd_dev; |
d71a6d73 | 61 | static int max_part; |
1da177e4 LT |
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 | { | |
097c94a4 | 107 | int error = req->errors ? -EIO : 0; |
165125e1 | 108 | struct request_queue *q = req->q; |
1da177e4 LT |
109 | unsigned long flags; |
110 | ||
111 | dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name, | |
097c94a4 | 112 | req, error ? "failed" : "done"); |
1da177e4 LT |
113 | |
114 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 115 | __blk_end_request_all(req, error); |
1da177e4 LT |
116 | spin_unlock_irqrestore(q->queue_lock, flags); |
117 | } | |
118 | ||
7fdfd406 PC |
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 | printk(KERN_WARNING "%s: shutting down socket\n", | |
131 | lo->disk->disk_name); | |
91cf45f0 | 132 | kernel_sock_shutdown(lo->sock, SHUT_RDWR); |
7fdfd406 PC |
133 | lo->sock = NULL; |
134 | } | |
135 | if (lock) | |
136 | mutex_unlock(&lo->tx_lock); | |
137 | } | |
138 | ||
139 | static void nbd_xmit_timeout(unsigned long arg) | |
140 | { | |
141 | struct task_struct *task = (struct task_struct *)arg; | |
142 | ||
143 | printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n", | |
144 | task->comm, task->pid); | |
145 | force_sig(SIGKILL, task); | |
146 | } | |
147 | ||
1da177e4 LT |
148 | /* |
149 | * Send or receive packet. | |
150 | */ | |
7fdfd406 | 151 | static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size, |
1da177e4 LT |
152 | int msg_flags) |
153 | { | |
7fdfd406 | 154 | struct socket *sock = lo->sock; |
1da177e4 LT |
155 | int result; |
156 | struct msghdr msg; | |
157 | struct kvec iov; | |
be0ef957 | 158 | sigset_t blocked, oldset; |
1da177e4 | 159 | |
ffc41cf8 MS |
160 | if (unlikely(!sock)) { |
161 | printk(KERN_ERR "%s: Attempted %s on closed socket in sock_xmit\n", | |
162 | lo->disk->disk_name, (send ? "send" : "recv")); | |
163 | return -EINVAL; | |
164 | } | |
165 | ||
1da177e4 LT |
166 | /* Allow interception of SIGKILL only |
167 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
168 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
169 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 LT |
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; | |
1da177e4 LT |
179 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
180 | ||
7fdfd406 PC |
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 | } | |
1da177e4 | 191 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7fdfd406 PC |
192 | if (lo->xmit_timeout) |
193 | del_timer_sync(&ti); | |
194 | } else | |
35fbf5bc NK |
195 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
196 | msg.msg_flags); | |
1da177e4 LT |
197 | |
198 | if (signal_pending(current)) { | |
199 | siginfo_t info; | |
1da177e4 | 200 | printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", |
ba25f9dc | 201 | task_pid_nr(current), current->comm, |
be0ef957 | 202 | dequeue_signal_lock(current, ¤t->blocked, &info)); |
1da177e4 | 203 | result = -EINTR; |
7fdfd406 | 204 | sock_shutdown(lo, !send); |
1da177e4 LT |
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 | ||
be0ef957 | 217 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
1da177e4 LT |
218 | |
219 | return result; | |
220 | } | |
221 | ||
7fdfd406 | 222 | static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec, |
1da177e4 LT |
223 | int flags) |
224 | { | |
225 | int result; | |
226 | void *kaddr = kmap(bvec->bv_page); | |
7fdfd406 | 227 | result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags); |
1da177e4 LT |
228 | kunmap(bvec->bv_page); |
229 | return result; | |
230 | } | |
231 | ||
7fdfd406 | 232 | /* always call with the tx_lock held */ |
1da177e4 LT |
233 | static int nbd_send_req(struct nbd_device *lo, struct request *req) |
234 | { | |
5705f702 | 235 | int result, flags; |
1da177e4 | 236 | struct nbd_request request; |
1011c1b9 | 237 | unsigned long size = blk_rq_bytes(req); |
1da177e4 LT |
238 | |
239 | request.magic = htonl(NBD_REQUEST_MAGIC); | |
240 | request.type = htonl(nbd_cmd(req)); | |
83096ebf | 241 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
1da177e4 LT |
242 | request.len = htonl(size); |
243 | memcpy(request.handle, &req, sizeof(req)); | |
244 | ||
83096ebf | 245 | dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n", |
1da177e4 LT |
246 | lo->disk->disk_name, req, |
247 | nbdcmd_to_ascii(nbd_cmd(req)), | |
83096ebf | 248 | (unsigned long long)blk_rq_pos(req) << 9, |
1011c1b9 | 249 | blk_rq_bytes(req)); |
7fdfd406 PC |
250 | result = sock_xmit(lo, 1, &request, sizeof(request), |
251 | (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0); | |
1da177e4 LT |
252 | if (result <= 0) { |
253 | printk(KERN_ERR "%s: Send control failed (result %d)\n", | |
254 | lo->disk->disk_name, result); | |
255 | goto error_out; | |
256 | } | |
257 | ||
258 | if (nbd_cmd(req) == NBD_CMD_WRITE) { | |
5705f702 N |
259 | struct req_iterator iter; |
260 | struct bio_vec *bvec; | |
1da177e4 LT |
261 | /* |
262 | * we are really probing at internals to determine | |
263 | * whether to set MSG_MORE or not... | |
264 | */ | |
5705f702 | 265 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 JA |
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); | |
7fdfd406 | 271 | result = sock_send_bvec(lo, bvec, flags); |
6c92e699 JA |
272 | if (result <= 0) { |
273 | printk(KERN_ERR "%s: Send data failed (result %d)\n", | |
274 | lo->disk->disk_name, result); | |
275 | goto error_out; | |
276 | } | |
1da177e4 LT |
277 | } |
278 | } | |
1da177e4 LT |
279 | return 0; |
280 | ||
281 | error_out: | |
15746fca | 282 | return -EIO; |
1da177e4 LT |
283 | } |
284 | ||
0cbc591b DC |
285 | static struct request *nbd_find_request(struct nbd_device *lo, |
286 | struct request *xreq) | |
1da177e4 | 287 | { |
d2c9740b | 288 | struct request *req, *tmp; |
4b2f0260 | 289 | int err; |
1da177e4 | 290 | |
4b2f0260 HX |
291 | err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq); |
292 | if (unlikely(err)) | |
293 | goto out; | |
294 | ||
1da177e4 | 295 | spin_lock(&lo->queue_lock); |
d2c9740b | 296 | list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) { |
1da177e4 LT |
297 | if (req != xreq) |
298 | continue; | |
299 | list_del_init(&req->queuelist); | |
300 | spin_unlock(&lo->queue_lock); | |
301 | return req; | |
302 | } | |
303 | spin_unlock(&lo->queue_lock); | |
4b2f0260 HX |
304 | |
305 | err = -ENOENT; | |
306 | ||
307 | out: | |
308 | return ERR_PTR(err); | |
1da177e4 LT |
309 | } |
310 | ||
7fdfd406 | 311 | static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec) |
1da177e4 LT |
312 | { |
313 | int result; | |
314 | void *kaddr = kmap(bvec->bv_page); | |
7fdfd406 | 315 | result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
316 | MSG_WAITALL); |
317 | kunmap(bvec->bv_page); | |
318 | return result; | |
319 | } | |
320 | ||
321 | /* NULL returned = something went wrong, inform userspace */ | |
322 | static struct request *nbd_read_stat(struct nbd_device *lo) | |
323 | { | |
324 | int result; | |
325 | struct nbd_reply reply; | |
326 | struct request *req; | |
1da177e4 LT |
327 | |
328 | reply.magic = 0; | |
7fdfd406 | 329 | result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 LT |
330 | if (result <= 0) { |
331 | printk(KERN_ERR "%s: Receive control failed (result %d)\n", | |
332 | lo->disk->disk_name, result); | |
333 | goto harderror; | |
334 | } | |
e4b57e08 MF |
335 | |
336 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
337 | printk(KERN_ERR "%s: Wrong magic (0x%lx)\n", | |
338 | lo->disk->disk_name, | |
339 | (unsigned long)ntohl(reply.magic)); | |
340 | result = -EPROTO; | |
341 | goto harderror; | |
342 | } | |
343 | ||
0cbc591b | 344 | req = nbd_find_request(lo, *(struct request **)reply.handle); |
801678c5 | 345 | if (IS_ERR(req)) { |
4b2f0260 HX |
346 | result = PTR_ERR(req); |
347 | if (result != -ENOENT) | |
348 | goto harderror; | |
349 | ||
1da177e4 LT |
350 | printk(KERN_ERR "%s: Unexpected reply (%p)\n", |
351 | lo->disk->disk_name, reply.handle); | |
352 | result = -EBADR; | |
353 | goto harderror; | |
354 | } | |
355 | ||
1da177e4 LT |
356 | if (ntohl(reply.error)) { |
357 | printk(KERN_ERR "%s: Other side returned error (%d)\n", | |
358 | lo->disk->disk_name, 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) { | |
5705f702 N |
366 | struct req_iterator iter; |
367 | struct bio_vec *bvec; | |
368 | ||
369 | rq_for_each_segment(bvec, req, iter) { | |
7fdfd406 | 370 | result = sock_recv_bvec(lo, bvec); |
6c92e699 JA |
371 | if (result <= 0) { |
372 | printk(KERN_ERR "%s: Receive data failed (result %d)\n", | |
373 | lo->disk->disk_name, 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); | |
1da177e4 LT |
379 | } |
380 | } | |
381 | return req; | |
382 | harderror: | |
383 | lo->harderror = result; | |
384 | return NULL; | |
385 | } | |
386 | ||
edfaa7c3 KS |
387 | static ssize_t pid_show(struct device *dev, |
388 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 389 | { |
edfaa7c3 KS |
390 | struct gendisk *disk = dev_to_disk(dev); |
391 | ||
392 | return sprintf(buf, "%ld\n", | |
6b39bb65 PC |
393 | (long) ((struct nbd_device *)disk->private_data)->pid); |
394 | } | |
395 | ||
edfaa7c3 | 396 | static struct device_attribute pid_attr = { |
01e8ef11 | 397 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
398 | .show = pid_show, |
399 | }; | |
400 | ||
84963048 | 401 | static int nbd_do_it(struct nbd_device *lo) |
1da177e4 LT |
402 | { |
403 | struct request *req; | |
84963048 | 404 | int ret; |
1da177e4 LT |
405 | |
406 | BUG_ON(lo->magic != LO_MAGIC); | |
407 | ||
6b39bb65 | 408 | lo->pid = current->pid; |
ed9e1982 | 409 | ret = sysfs_create_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr); |
84963048 WC |
410 | if (ret) { |
411 | printk(KERN_ERR "nbd: sysfs_create_file failed!"); | |
c91192d6 | 412 | lo->pid = 0; |
84963048 WC |
413 | return ret; |
414 | } | |
6b39bb65 | 415 | |
1da177e4 LT |
416 | while ((req = nbd_read_stat(lo)) != NULL) |
417 | nbd_end_request(req); | |
6b39bb65 | 418 | |
ed9e1982 | 419 | sysfs_remove_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr); |
c91192d6 | 420 | lo->pid = 0; |
84963048 | 421 | return 0; |
1da177e4 LT |
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 | ||
4b2f0260 HX |
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 | } | |
1da177e4 LT |
448 | } |
449 | ||
7fdfd406 | 450 | |
48cf6061 LV |
451 | static void nbd_handle_req(struct nbd_device *lo, struct request *req) |
452 | { | |
33659ebb | 453 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
454 | goto error_out; |
455 | ||
456 | nbd_cmd(req) = NBD_CMD_READ; | |
457 | if (rq_data_dir(req) == WRITE) { | |
458 | nbd_cmd(req) = NBD_CMD_WRITE; | |
459 | if (lo->flags & NBD_READ_ONLY) { | |
460 | printk(KERN_ERR "%s: Write on read-only\n", | |
461 | lo->disk->disk_name); | |
462 | goto error_out; | |
463 | } | |
464 | } | |
465 | ||
466 | req->errors = 0; | |
467 | ||
468 | mutex_lock(&lo->tx_lock); | |
469 | if (unlikely(!lo->sock)) { | |
470 | mutex_unlock(&lo->tx_lock); | |
471 | printk(KERN_ERR "%s: Attempted send on closed socket\n", | |
472 | lo->disk->disk_name); | |
15746fca | 473 | goto error_out; |
48cf6061 LV |
474 | } |
475 | ||
476 | lo->active_req = req; | |
477 | ||
478 | if (nbd_send_req(lo, req) != 0) { | |
479 | printk(KERN_ERR "%s: Request send failed\n", | |
480 | lo->disk->disk_name); | |
481 | req->errors++; | |
482 | nbd_end_request(req); | |
483 | } else { | |
484 | spin_lock(&lo->queue_lock); | |
485 | list_add(&req->queuelist, &lo->queue_head); | |
486 | spin_unlock(&lo->queue_lock); | |
487 | } | |
488 | ||
489 | lo->active_req = NULL; | |
490 | mutex_unlock(&lo->tx_lock); | |
491 | wake_up_all(&lo->active_wq); | |
492 | ||
493 | return; | |
494 | ||
495 | error_out: | |
496 | req->errors++; | |
497 | nbd_end_request(req); | |
498 | } | |
499 | ||
500 | static int nbd_thread(void *data) | |
501 | { | |
502 | struct nbd_device *lo = data; | |
503 | struct request *req; | |
504 | ||
505 | set_user_nice(current, -20); | |
506 | while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) { | |
507 | /* wait for something to do */ | |
508 | wait_event_interruptible(lo->waiting_wq, | |
509 | kthread_should_stop() || | |
510 | !list_empty(&lo->waiting_queue)); | |
511 | ||
512 | /* extract request */ | |
513 | if (list_empty(&lo->waiting_queue)) | |
514 | continue; | |
515 | ||
516 | spin_lock_irq(&lo->queue_lock); | |
517 | req = list_entry(lo->waiting_queue.next, struct request, | |
518 | queuelist); | |
519 | list_del_init(&req->queuelist); | |
520 | spin_unlock_irq(&lo->queue_lock); | |
521 | ||
522 | /* handle request */ | |
523 | nbd_handle_req(lo, req); | |
524 | } | |
525 | return 0; | |
526 | } | |
527 | ||
1da177e4 LT |
528 | /* |
529 | * We always wait for result of write, for now. It would be nice to make it optional | |
530 | * in future | |
e654bc43 | 531 | * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
532 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
533 | */ | |
534 | ||
15746fca | 535 | static void do_nbd_request(struct request_queue *q) |
1da177e4 LT |
536 | { |
537 | struct request *req; | |
538 | ||
9934c8c0 | 539 | while ((req = blk_fetch_request(q)) != NULL) { |
1da177e4 LT |
540 | struct nbd_device *lo; |
541 | ||
48cf6061 LV |
542 | spin_unlock_irq(q->queue_lock); |
543 | ||
4aff5e23 JA |
544 | dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n", |
545 | req->rq_disk->disk_name, req, req->cmd_type); | |
1da177e4 | 546 | |
1da177e4 LT |
547 | lo = req->rq_disk->private_data; |
548 | ||
549 | BUG_ON(lo->magic != LO_MAGIC); | |
550 | ||
4d48a542 PC |
551 | if (unlikely(!lo->sock)) { |
552 | printk(KERN_ERR "%s: Attempted send on closed socket\n", | |
553 | lo->disk->disk_name); | |
554 | req->errors++; | |
555 | nbd_end_request(req); | |
556 | spin_lock_irq(q->queue_lock); | |
557 | continue; | |
558 | } | |
559 | ||
48cf6061 LV |
560 | spin_lock_irq(&lo->queue_lock); |
561 | list_add_tail(&req->queuelist, &lo->waiting_queue); | |
562 | spin_unlock_irq(&lo->queue_lock); | |
1da177e4 | 563 | |
48cf6061 | 564 | wake_up(&lo->waiting_wq); |
4b2f0260 | 565 | |
1da177e4 | 566 | spin_lock_irq(q->queue_lock); |
1da177e4 | 567 | } |
1da177e4 LT |
568 | } |
569 | ||
1a2ad211 | 570 | /* Must be called with tx_lock held */ |
1da177e4 | 571 | |
1a2ad211 PM |
572 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *lo, |
573 | unsigned int cmd, unsigned long arg) | |
574 | { | |
1da177e4 | 575 | switch (cmd) { |
1a2ad211 PM |
576 | case NBD_DISCONNECT: { |
577 | struct request sreq; | |
578 | ||
1da177e4 | 579 | printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name); |
1a2ad211 | 580 | |
4f54eec8 | 581 | blk_rq_init(NULL, &sreq); |
4aff5e23 | 582 | sreq.cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 | 583 | nbd_cmd(&sreq) = NBD_CMD_DISC; |
1a2ad211 | 584 | if (!lo->sock) |
1da177e4 | 585 | return -EINVAL; |
1a2ad211 | 586 | nbd_send_req(lo, &sreq); |
1da177e4 | 587 | return 0; |
1a2ad211 | 588 | } |
1da177e4 | 589 | |
1a2ad211 PM |
590 | case NBD_CLEAR_SOCK: { |
591 | struct file *file; | |
592 | ||
1da177e4 | 593 | lo->sock = NULL; |
1da177e4 LT |
594 | file = lo->file; |
595 | lo->file = NULL; | |
1da177e4 | 596 | nbd_clear_que(lo); |
4b2f0260 | 597 | BUG_ON(!list_empty(&lo->queue_head)); |
1da177e4 LT |
598 | if (file) |
599 | fput(file); | |
1a2ad211 PM |
600 | return 0; |
601 | } | |
602 | ||
603 | case NBD_SET_SOCK: { | |
604 | struct file *file; | |
1da177e4 LT |
605 | if (lo->file) |
606 | return -EBUSY; | |
1da177e4 LT |
607 | file = fget(arg); |
608 | if (file) { | |
a8cdc308 | 609 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
610 | if (S_ISSOCK(inode->i_mode)) { |
611 | lo->file = file; | |
612 | lo->sock = SOCKET_I(inode); | |
d71a6d73 LV |
613 | if (max_part > 0) |
614 | bdev->bd_invalidated = 1; | |
1a2ad211 | 615 | return 0; |
1da177e4 LT |
616 | } else { |
617 | fput(file); | |
618 | } | |
619 | } | |
1a2ad211 PM |
620 | return -EINVAL; |
621 | } | |
622 | ||
1da177e4 LT |
623 | case NBD_SET_BLKSIZE: |
624 | lo->blksize = arg; | |
625 | lo->bytesize &= ~(lo->blksize-1); | |
a8cdc308 AV |
626 | bdev->bd_inode->i_size = lo->bytesize; |
627 | set_blocksize(bdev, lo->blksize); | |
1da177e4 LT |
628 | set_capacity(lo->disk, lo->bytesize >> 9); |
629 | return 0; | |
1a2ad211 | 630 | |
1da177e4 LT |
631 | case NBD_SET_SIZE: |
632 | lo->bytesize = arg & ~(lo->blksize-1); | |
a8cdc308 AV |
633 | bdev->bd_inode->i_size = lo->bytesize; |
634 | set_blocksize(bdev, lo->blksize); | |
1da177e4 LT |
635 | set_capacity(lo->disk, lo->bytesize >> 9); |
636 | return 0; | |
1a2ad211 | 637 | |
7fdfd406 PC |
638 | case NBD_SET_TIMEOUT: |
639 | lo->xmit_timeout = arg * HZ; | |
640 | return 0; | |
1a2ad211 | 641 | |
1da177e4 LT |
642 | case NBD_SET_SIZE_BLOCKS: |
643 | lo->bytesize = ((u64) arg) * lo->blksize; | |
a8cdc308 AV |
644 | bdev->bd_inode->i_size = lo->bytesize; |
645 | set_blocksize(bdev, lo->blksize); | |
1da177e4 LT |
646 | set_capacity(lo->disk, lo->bytesize >> 9); |
647 | return 0; | |
1a2ad211 PM |
648 | |
649 | case NBD_DO_IT: { | |
650 | struct task_struct *thread; | |
651 | struct file *file; | |
652 | int error; | |
653 | ||
c91192d6 PM |
654 | if (lo->pid) |
655 | return -EBUSY; | |
1da177e4 LT |
656 | if (!lo->file) |
657 | return -EINVAL; | |
1a2ad211 PM |
658 | |
659 | mutex_unlock(&lo->tx_lock); | |
660 | ||
48cf6061 | 661 | thread = kthread_create(nbd_thread, lo, lo->disk->disk_name); |
1a2ad211 PM |
662 | if (IS_ERR(thread)) { |
663 | mutex_lock(&lo->tx_lock); | |
48cf6061 | 664 | return PTR_ERR(thread); |
1a2ad211 | 665 | } |
48cf6061 | 666 | wake_up_process(thread); |
84963048 | 667 | error = nbd_do_it(lo); |
48cf6061 | 668 | kthread_stop(thread); |
1a2ad211 PM |
669 | |
670 | mutex_lock(&lo->tx_lock); | |
84963048 WC |
671 | if (error) |
672 | return error; | |
1a2ad211 | 673 | sock_shutdown(lo, 0); |
1da177e4 LT |
674 | file = lo->file; |
675 | lo->file = NULL; | |
1da177e4 LT |
676 | nbd_clear_que(lo); |
677 | printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name); | |
678 | if (file) | |
679 | fput(file); | |
4b86a872 | 680 | lo->bytesize = 0; |
a8cdc308 | 681 | bdev->bd_inode->i_size = 0; |
4b86a872 | 682 | set_capacity(lo->disk, 0); |
d71a6d73 | 683 | if (max_part > 0) |
a8cdc308 | 684 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
1da177e4 | 685 | return lo->harderror; |
1a2ad211 PM |
686 | } |
687 | ||
1da177e4 | 688 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
689 | /* |
690 | * This is for compatibility only. The queue is always cleared | |
691 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
692 | */ | |
693 | BUG_ON(!lo->sock && !list_empty(&lo->queue_head)); | |
1da177e4 | 694 | return 0; |
1a2ad211 | 695 | |
1da177e4 LT |
696 | case NBD_PRINT_DEBUG: |
697 | printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n", | |
a8cdc308 | 698 | bdev->bd_disk->disk_name, |
1da177e4 LT |
699 | lo->queue_head.next, lo->queue_head.prev, |
700 | &lo->queue_head); | |
701 | return 0; | |
702 | } | |
1a2ad211 PM |
703 | return -ENOTTY; |
704 | } | |
705 | ||
706 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
707 | unsigned int cmd, unsigned long arg) | |
708 | { | |
709 | struct nbd_device *lo = bdev->bd_disk->private_data; | |
710 | int error; | |
711 | ||
712 | if (!capable(CAP_SYS_ADMIN)) | |
713 | return -EPERM; | |
714 | ||
715 | BUG_ON(lo->magic != LO_MAGIC); | |
716 | ||
717 | /* Anyone capable of this syscall can do *real bad* things */ | |
718 | dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n", | |
719 | lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg); | |
720 | ||
721 | mutex_lock(&lo->tx_lock); | |
722 | error = __nbd_ioctl(bdev, lo, cmd, arg); | |
723 | mutex_unlock(&lo->tx_lock); | |
724 | ||
725 | return error; | |
1da177e4 LT |
726 | } |
727 | ||
83d5cde4 | 728 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
729 | { |
730 | .owner = THIS_MODULE, | |
8a6cfeb6 | 731 | .ioctl = nbd_ioctl, |
1da177e4 LT |
732 | }; |
733 | ||
734 | /* | |
735 | * And here should be modules and kernel interface | |
736 | * (Just smiley confuses emacs :-) | |
737 | */ | |
738 | ||
739 | static int __init nbd_init(void) | |
740 | { | |
741 | int err = -ENOMEM; | |
742 | int i; | |
d71a6d73 | 743 | int part_shift; |
1da177e4 | 744 | |
5b7b18cc | 745 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 746 | |
d71a6d73 LV |
747 | if (max_part < 0) { |
748 | printk(KERN_CRIT "nbd: max_part must be >= 0\n"); | |
749 | return -EINVAL; | |
750 | } | |
751 | ||
f3944d61 SW |
752 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
753 | if (!nbd_dev) | |
754 | return -ENOMEM; | |
755 | ||
d71a6d73 | 756 | part_shift = 0; |
5988ce23 | 757 | if (max_part > 0) { |
d71a6d73 LV |
758 | part_shift = fls(max_part); |
759 | ||
5988ce23 NK |
760 | /* |
761 | * Adjust max_part according to part_shift as it is exported | |
762 | * to user space so that user can know the max number of | |
763 | * partition kernel should be able to manage. | |
764 | * | |
765 | * Note that -1 is required because partition 0 is reserved | |
766 | * for the whole disk. | |
767 | */ | |
768 | max_part = (1UL << part_shift) - 1; | |
769 | } | |
770 | ||
3b271082 NK |
771 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
772 | return -EINVAL; | |
773 | ||
774 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
775 | return -EINVAL; | |
776 | ||
40be0c28 | 777 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 778 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
779 | if (!disk) |
780 | goto out; | |
781 | nbd_dev[i].disk = disk; | |
782 | /* | |
783 | * The new linux 2.5 block layer implementation requires | |
784 | * every gendisk to have its very own request_queue struct. | |
785 | * These structs are big so we dynamically allocate them. | |
786 | */ | |
787 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
788 | if (!disk->queue) { | |
789 | put_disk(disk); | |
790 | goto out; | |
791 | } | |
31dcfab0 JA |
792 | /* |
793 | * Tell the block layer that we are not a rotational device | |
794 | */ | |
795 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1da177e4 LT |
796 | } |
797 | ||
798 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
799 | err = -EIO; | |
800 | goto out; | |
801 | } | |
802 | ||
803 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
804 | dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags); | |
805 | ||
40be0c28 | 806 | for (i = 0; i < nbds_max; i++) { |
1da177e4 LT |
807 | struct gendisk *disk = nbd_dev[i].disk; |
808 | nbd_dev[i].file = NULL; | |
809 | nbd_dev[i].magic = LO_MAGIC; | |
810 | nbd_dev[i].flags = 0; | |
48cf6061 | 811 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
812 | spin_lock_init(&nbd_dev[i].queue_lock); |
813 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 814 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 815 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 816 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 817 | nbd_dev[i].blksize = 1024; |
4b86a872 | 818 | nbd_dev[i].bytesize = 0; |
1da177e4 | 819 | disk->major = NBD_MAJOR; |
d71a6d73 | 820 | disk->first_minor = i << part_shift; |
1da177e4 LT |
821 | disk->fops = &nbd_fops; |
822 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 823 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 824 | set_capacity(disk, 0); |
1da177e4 LT |
825 | add_disk(disk); |
826 | } | |
827 | ||
828 | return 0; | |
829 | out: | |
830 | while (i--) { | |
831 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
832 | put_disk(nbd_dev[i].disk); | |
833 | } | |
f3944d61 | 834 | kfree(nbd_dev); |
1da177e4 LT |
835 | return err; |
836 | } | |
837 | ||
838 | static void __exit nbd_cleanup(void) | |
839 | { | |
840 | int i; | |
40be0c28 | 841 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 842 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 843 | nbd_dev[i].magic = 0; |
1da177e4 LT |
844 | if (disk) { |
845 | del_gendisk(disk); | |
846 | blk_cleanup_queue(disk->queue); | |
847 | put_disk(disk); | |
848 | } | |
849 | } | |
1da177e4 | 850 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 851 | kfree(nbd_dev); |
1da177e4 LT |
852 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
853 | } | |
854 | ||
855 | module_init(nbd_init); | |
856 | module_exit(nbd_cleanup); | |
857 | ||
858 | MODULE_DESCRIPTION("Network Block Device"); | |
859 | MODULE_LICENSE("GPL"); | |
860 | ||
40be0c28 | 861 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
862 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
863 | module_param(max_part, int, 0444); | |
864 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); | |
1da177e4 LT |
865 | #ifndef NDEBUG |
866 | module_param(debugflags, int, 0644); | |
867 | MODULE_PARM_DESC(debugflags, "flags for controlling debug output"); | |
868 | #endif |