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> |
b9c495bb | 35 | #include <linux/types.h> |
30d53d9c | 36 | #include <linux/debugfs.h> |
1da177e4 | 37 | |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | #include <asm/types.h> | |
40 | ||
41 | #include <linux/nbd.h> | |
42 | ||
13e71d69 | 43 | struct nbd_device { |
22d109c1 | 44 | u32 flags; |
13e71d69 MP |
45 | struct socket * sock; /* If == NULL, device is not ready, yet */ |
46 | int magic; | |
47 | ||
48 | spinlock_t queue_lock; | |
49 | struct list_head queue_head; /* Requests waiting result */ | |
50 | struct request *active_req; | |
51 | wait_queue_head_t active_wq; | |
52 | struct list_head waiting_queue; /* Requests to be sent */ | |
53 | wait_queue_head_t waiting_wq; | |
54 | ||
55 | struct mutex tx_lock; | |
56 | struct gendisk *disk; | |
57 | int blksize; | |
b9c495bb | 58 | loff_t bytesize; |
13e71d69 | 59 | int xmit_timeout; |
1f7b5cf1 | 60 | bool timedout; |
696697cb | 61 | bool disconnect; /* a disconnect has been requested by user */ |
7e2893a1 MP |
62 | |
63 | struct timer_list timeout_timer; | |
23272a67 MP |
64 | /* protects initialization and shutdown of the socket */ |
65 | spinlock_t sock_lock; | |
7e2893a1 MP |
66 | struct task_struct *task_recv; |
67 | struct task_struct *task_send; | |
30d53d9c MP |
68 | |
69 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
70 | struct dentry *dbg_dir; | |
71 | #endif | |
13e71d69 MP |
72 | }; |
73 | ||
30d53d9c MP |
74 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
75 | static struct dentry *nbd_dbg_dir; | |
76 | #endif | |
77 | ||
78 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
79 | ||
f4507164 | 80 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 81 | |
9c7a4169 | 82 | static unsigned int nbds_max = 16; |
20a8143e | 83 | static struct nbd_device *nbd_dev; |
d71a6d73 | 84 | static int max_part; |
1da177e4 LT |
85 | |
86 | /* | |
87 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
88 | * 1. Each NIC is essentially a synchronization point for all servers | |
89 | * accessed through that NIC so there's no need to have more locks | |
90 | * than NICs anyway. | |
91 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
92 | * down each lock to the point where they're actually slower than just | |
93 | * a single lock. | |
94 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
95 | */ | |
96 | static DEFINE_SPINLOCK(nbd_lock); | |
97 | ||
d18509f5 | 98 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 99 | { |
d18509f5 | 100 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
101 | } |
102 | ||
37091fdd MP |
103 | static bool nbd_is_connected(struct nbd_device *nbd) |
104 | { | |
105 | return !!nbd->task_recv; | |
106 | } | |
107 | ||
1da177e4 LT |
108 | static const char *nbdcmd_to_ascii(int cmd) |
109 | { | |
110 | switch (cmd) { | |
111 | case NBD_CMD_READ: return "read"; | |
112 | case NBD_CMD_WRITE: return "write"; | |
113 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 114 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 115 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
116 | } |
117 | return "invalid"; | |
118 | } | |
1da177e4 | 119 | |
37091fdd MP |
120 | static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev) |
121 | { | |
122 | bdev->bd_inode->i_size = 0; | |
123 | set_capacity(nbd->disk, 0); | |
124 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
125 | ||
126 | return 0; | |
127 | } | |
128 | ||
129 | static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev) | |
130 | { | |
131 | if (!nbd_is_connected(nbd)) | |
132 | return; | |
133 | ||
134 | bdev->bd_inode->i_size = nbd->bytesize; | |
135 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
136 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
137 | } | |
138 | ||
139 | static int nbd_size_set(struct nbd_device *nbd, struct block_device *bdev, | |
140 | int blocksize, int nr_blocks) | |
141 | { | |
142 | int ret; | |
143 | ||
144 | ret = set_blocksize(bdev, blocksize); | |
145 | if (ret) | |
146 | return ret; | |
147 | ||
148 | nbd->blksize = blocksize; | |
149 | nbd->bytesize = (loff_t)blocksize * (loff_t)nr_blocks; | |
150 | ||
151 | nbd_size_update(nbd, bdev); | |
152 | ||
153 | return 0; | |
154 | } | |
155 | ||
d18509f5 | 156 | static void nbd_end_request(struct nbd_device *nbd, struct request *req) |
1da177e4 | 157 | { |
097c94a4 | 158 | int error = req->errors ? -EIO : 0; |
165125e1 | 159 | struct request_queue *q = req->q; |
1da177e4 LT |
160 | unsigned long flags; |
161 | ||
d18509f5 MP |
162 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req, |
163 | error ? "failed" : "done"); | |
1da177e4 LT |
164 | |
165 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 166 | __blk_end_request_all(req, error); |
1da177e4 LT |
167 | spin_unlock_irqrestore(q->queue_lock, flags); |
168 | } | |
169 | ||
e018e757 MP |
170 | /* |
171 | * Forcibly shutdown the socket causing all listeners to error | |
172 | */ | |
36e47bee | 173 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 174 | { |
23272a67 MP |
175 | spin_lock_irq(&nbd->sock_lock); |
176 | ||
177 | if (!nbd->sock) { | |
178 | spin_unlock_irq(&nbd->sock_lock); | |
260bbce4 | 179 | return; |
23272a67 | 180 | } |
260bbce4 MP |
181 | |
182 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
183 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
23272a67 | 184 | sockfd_put(nbd->sock); |
260bbce4 | 185 | nbd->sock = NULL; |
23272a67 MP |
186 | spin_unlock_irq(&nbd->sock_lock); |
187 | ||
188 | del_timer(&nbd->timeout_timer); | |
7fdfd406 PC |
189 | } |
190 | ||
191 | static void nbd_xmit_timeout(unsigned long arg) | |
192 | { | |
7e2893a1 | 193 | struct nbd_device *nbd = (struct nbd_device *)arg; |
dcc909d9 | 194 | unsigned long flags; |
7e2893a1 MP |
195 | |
196 | if (list_empty(&nbd->queue_head)) | |
197 | return; | |
198 | ||
23272a67 | 199 | spin_lock_irqsave(&nbd->sock_lock, flags); |
dcc909d9 | 200 | |
1f7b5cf1 | 201 | nbd->timedout = true; |
7fdfd406 | 202 | |
23272a67 MP |
203 | if (nbd->sock) |
204 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
7e2893a1 | 205 | |
23272a67 | 206 | spin_unlock_irqrestore(&nbd->sock_lock, flags); |
dcc909d9 | 207 | |
23272a67 | 208 | dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n"); |
7fdfd406 PC |
209 | } |
210 | ||
1da177e4 LT |
211 | /* |
212 | * Send or receive packet. | |
213 | */ | |
f4507164 | 214 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
215 | int msg_flags) |
216 | { | |
f4507164 | 217 | struct socket *sock = nbd->sock; |
1da177e4 LT |
218 | int result; |
219 | struct msghdr msg; | |
220 | struct kvec iov; | |
7f338fe4 | 221 | unsigned long pflags = current->flags; |
1da177e4 | 222 | |
ffc41cf8 | 223 | if (unlikely(!sock)) { |
f4507164 | 224 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
225 | "Attempted %s on closed socket in sock_xmit\n", |
226 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
227 | return -EINVAL; |
228 | } | |
229 | ||
7f338fe4 | 230 | current->flags |= PF_MEMALLOC; |
1da177e4 | 231 | do { |
7f338fe4 | 232 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
233 | iov.iov_base = buf; |
234 | iov.iov_len = size; | |
235 | msg.msg_name = NULL; | |
236 | msg.msg_namelen = 0; | |
237 | msg.msg_control = NULL; | |
238 | msg.msg_controllen = 0; | |
1da177e4 LT |
239 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
240 | ||
7e2893a1 | 241 | if (send) |
1da177e4 | 242 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7e2893a1 | 243 | else |
35fbf5bc NK |
244 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
245 | msg.msg_flags); | |
1da177e4 | 246 | |
1da177e4 LT |
247 | if (result <= 0) { |
248 | if (result == 0) | |
249 | result = -EPIPE; /* short read */ | |
250 | break; | |
251 | } | |
252 | size -= result; | |
253 | buf += result; | |
254 | } while (size > 0); | |
255 | ||
7f338fe4 | 256 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 | 257 | |
7e2893a1 MP |
258 | if (!send && nbd->xmit_timeout) |
259 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
260 | ||
1da177e4 LT |
261 | return result; |
262 | } | |
263 | ||
f4507164 | 264 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
265 | int flags) |
266 | { | |
267 | int result; | |
268 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
269 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
270 | bvec->bv_len, flags); | |
1da177e4 LT |
271 | kunmap(bvec->bv_page); |
272 | return result; | |
273 | } | |
274 | ||
7fdfd406 | 275 | /* always call with the tx_lock held */ |
f4507164 | 276 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 277 | { |
5705f702 | 278 | int result, flags; |
1da177e4 | 279 | struct nbd_request request; |
1011c1b9 | 280 | unsigned long size = blk_rq_bytes(req); |
9dc6c806 CH |
281 | u32 type; |
282 | ||
283 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
284 | type = NBD_CMD_DISC; | |
285 | else if (req->cmd_flags & REQ_DISCARD) | |
286 | type = NBD_CMD_TRIM; | |
287 | else if (req->cmd_flags & REQ_FLUSH) | |
288 | type = NBD_CMD_FLUSH; | |
289 | else if (rq_data_dir(req) == WRITE) | |
290 | type = NBD_CMD_WRITE; | |
291 | else | |
292 | type = NBD_CMD_READ; | |
1da177e4 | 293 | |
04cfac4e | 294 | memset(&request, 0, sizeof(request)); |
1da177e4 | 295 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 CH |
296 | request.type = htonl(type); |
297 | if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) { | |
75f187ab AB |
298 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
299 | request.len = htonl(size); | |
300 | } | |
1da177e4 LT |
301 | memcpy(request.handle, &req, sizeof(req)); |
302 | ||
d18509f5 | 303 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
9dc6c806 | 304 | req, nbdcmd_to_ascii(type), |
d18509f5 | 305 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
f4507164 | 306 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
9dc6c806 | 307 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 308 | if (result <= 0) { |
f4507164 | 309 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 310 | "Send control failed (result %d)\n", result); |
dab5313a | 311 | return -EIO; |
1da177e4 LT |
312 | } |
313 | ||
9dc6c806 | 314 | if (type == NBD_CMD_WRITE) { |
5705f702 | 315 | struct req_iterator iter; |
7988613b | 316 | struct bio_vec bvec; |
1da177e4 LT |
317 | /* |
318 | * we are really probing at internals to determine | |
319 | * whether to set MSG_MORE or not... | |
320 | */ | |
5705f702 | 321 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 | 322 | flags = 0; |
4550dd6c | 323 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 324 | flags = MSG_MORE; |
d18509f5 MP |
325 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
326 | req, bvec.bv_len); | |
7988613b | 327 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 328 | if (result <= 0) { |
f4507164 | 329 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
330 | "Send data failed (result %d)\n", |
331 | result); | |
dab5313a | 332 | return -EIO; |
6c92e699 | 333 | } |
1da177e4 LT |
334 | } |
335 | } | |
1da177e4 | 336 | return 0; |
1da177e4 LT |
337 | } |
338 | ||
f4507164 | 339 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 340 | struct request *xreq) |
1da177e4 | 341 | { |
d2c9740b | 342 | struct request *req, *tmp; |
4b2f0260 | 343 | int err; |
1da177e4 | 344 | |
f4507164 | 345 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 346 | if (unlikely(err)) |
de9ad6d4 | 347 | return ERR_PTR(err); |
4b2f0260 | 348 | |
f4507164 WG |
349 | spin_lock(&nbd->queue_lock); |
350 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
351 | if (req != xreq) |
352 | continue; | |
353 | list_del_init(&req->queuelist); | |
f4507164 | 354 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
355 | return req; |
356 | } | |
f4507164 | 357 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 358 | |
de9ad6d4 | 359 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
360 | } |
361 | ||
f4507164 | 362 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
363 | { |
364 | int result; | |
365 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 366 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
367 | MSG_WAITALL); |
368 | kunmap(bvec->bv_page); | |
369 | return result; | |
370 | } | |
371 | ||
372 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 373 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
374 | { |
375 | int result; | |
376 | struct nbd_reply reply; | |
377 | struct request *req; | |
1da177e4 LT |
378 | |
379 | reply.magic = 0; | |
f4507164 | 380 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 381 | if (result <= 0) { |
f4507164 | 382 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 383 | "Receive control failed (result %d)\n", result); |
19391830 | 384 | return ERR_PTR(result); |
1da177e4 | 385 | } |
e4b57e08 MF |
386 | |
387 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 388 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 389 | (unsigned long)ntohl(reply.magic)); |
19391830 | 390 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
391 | } |
392 | ||
f4507164 | 393 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 394 | if (IS_ERR(req)) { |
4b2f0260 HX |
395 | result = PTR_ERR(req); |
396 | if (result != -ENOENT) | |
19391830 | 397 | return ERR_PTR(result); |
4b2f0260 | 398 | |
f4507164 | 399 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 400 | reply.handle); |
19391830 | 401 | return ERR_PTR(-EBADR); |
1da177e4 LT |
402 | } |
403 | ||
1da177e4 | 404 | if (ntohl(reply.error)) { |
f4507164 | 405 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 406 | ntohl(reply.error)); |
1da177e4 LT |
407 | req->errors++; |
408 | return req; | |
409 | } | |
410 | ||
d18509f5 | 411 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 412 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 413 | struct req_iterator iter; |
7988613b | 414 | struct bio_vec bvec; |
5705f702 N |
415 | |
416 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 417 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 418 | if (result <= 0) { |
f4507164 | 419 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 420 | result); |
6c92e699 JA |
421 | req->errors++; |
422 | return req; | |
423 | } | |
d18509f5 MP |
424 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
425 | req, bvec.bv_len); | |
1da177e4 LT |
426 | } |
427 | } | |
428 | return req; | |
1da177e4 LT |
429 | } |
430 | ||
edfaa7c3 KS |
431 | static ssize_t pid_show(struct device *dev, |
432 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 433 | { |
edfaa7c3 | 434 | struct gendisk *disk = dev_to_disk(dev); |
6521d39a | 435 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; |
edfaa7c3 | 436 | |
6521d39a | 437 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); |
6b39bb65 PC |
438 | } |
439 | ||
edfaa7c3 | 440 | static struct device_attribute pid_attr = { |
01e8ef11 | 441 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
442 | .show = pid_show, |
443 | }; | |
444 | ||
37091fdd | 445 | static int nbd_thread_recv(struct nbd_device *nbd, struct block_device *bdev) |
1da177e4 LT |
446 | { |
447 | struct request *req; | |
84963048 | 448 | int ret; |
1da177e4 | 449 | |
f4507164 | 450 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 451 | |
7f338fe4 | 452 | sk_set_memalloc(nbd->sock->sk); |
6521d39a MP |
453 | |
454 | nbd->task_recv = current; | |
455 | ||
f4507164 | 456 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
84963048 | 457 | if (ret) { |
f4507164 | 458 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
dcc909d9 | 459 | |
6521d39a | 460 | nbd->task_recv = NULL; |
dcc909d9 | 461 | |
84963048 WC |
462 | return ret; |
463 | } | |
6b39bb65 | 464 | |
37091fdd MP |
465 | nbd_size_update(nbd, bdev); |
466 | ||
19391830 MP |
467 | while (1) { |
468 | req = nbd_read_stat(nbd); | |
469 | if (IS_ERR(req)) { | |
470 | ret = PTR_ERR(req); | |
471 | break; | |
472 | } | |
473 | ||
d18509f5 | 474 | nbd_end_request(nbd, req); |
19391830 | 475 | } |
6b39bb65 | 476 | |
37091fdd MP |
477 | nbd_size_clear(nbd, bdev); |
478 | ||
6521d39a MP |
479 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
480 | ||
7e2893a1 | 481 | nbd->task_recv = NULL; |
7e2893a1 | 482 | |
7e2893a1 | 483 | return ret; |
1da177e4 LT |
484 | } |
485 | ||
f4507164 | 486 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
487 | { |
488 | struct request *req; | |
489 | ||
f4507164 | 490 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 491 | |
4b2f0260 | 492 | /* |
f4507164 | 493 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
494 | * modifications to the list must have completed by now. For |
495 | * the same reason, the active_req must be NULL. | |
496 | * | |
497 | * As a consequence, we don't need to take the spin lock while | |
498 | * purging the list here. | |
499 | */ | |
f4507164 WG |
500 | BUG_ON(nbd->sock); |
501 | BUG_ON(nbd->active_req); | |
4b2f0260 | 502 | |
f4507164 WG |
503 | while (!list_empty(&nbd->queue_head)) { |
504 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
505 | queuelist); |
506 | list_del_init(&req->queuelist); | |
507 | req->errors++; | |
d18509f5 | 508 | nbd_end_request(nbd, req); |
4b2f0260 | 509 | } |
fded4e09 PC |
510 | |
511 | while (!list_empty(&nbd->waiting_queue)) { | |
512 | req = list_entry(nbd->waiting_queue.next, struct request, | |
513 | queuelist); | |
514 | list_del_init(&req->queuelist); | |
515 | req->errors++; | |
d18509f5 | 516 | nbd_end_request(nbd, req); |
fded4e09 | 517 | } |
e78273c8 | 518 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
519 | } |
520 | ||
7fdfd406 | 521 | |
f4507164 | 522 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 523 | { |
33659ebb | 524 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
525 | goto error_out; |
526 | ||
9dc6c806 CH |
527 | if (rq_data_dir(req) == WRITE && |
528 | (nbd->flags & NBD_FLAG_READ_ONLY)) { | |
529 | dev_err(disk_to_dev(nbd->disk), | |
530 | "Write on read-only\n"); | |
531 | goto error_out; | |
75f187ab AB |
532 | } |
533 | ||
48cf6061 LV |
534 | req->errors = 0; |
535 | ||
f4507164 WG |
536 | mutex_lock(&nbd->tx_lock); |
537 | if (unlikely(!nbd->sock)) { | |
538 | mutex_unlock(&nbd->tx_lock); | |
539 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 540 | "Attempted send on closed socket\n"); |
15746fca | 541 | goto error_out; |
48cf6061 LV |
542 | } |
543 | ||
f4507164 | 544 | nbd->active_req = req; |
48cf6061 | 545 | |
7e2893a1 MP |
546 | if (nbd->xmit_timeout && list_empty_careful(&nbd->queue_head)) |
547 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
548 | ||
f4507164 WG |
549 | if (nbd_send_req(nbd, req) != 0) { |
550 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 551 | req->errors++; |
d18509f5 | 552 | nbd_end_request(nbd, req); |
48cf6061 | 553 | } else { |
f4507164 | 554 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 555 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 556 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
557 | } |
558 | ||
f4507164 WG |
559 | nbd->active_req = NULL; |
560 | mutex_unlock(&nbd->tx_lock); | |
561 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
562 | |
563 | return; | |
564 | ||
565 | error_out: | |
566 | req->errors++; | |
d18509f5 | 567 | nbd_end_request(nbd, req); |
48cf6061 LV |
568 | } |
569 | ||
cad73b27 | 570 | static int nbd_thread_send(void *data) |
48cf6061 | 571 | { |
f4507164 | 572 | struct nbd_device *nbd = data; |
48cf6061 LV |
573 | struct request *req; |
574 | ||
7e2893a1 MP |
575 | nbd->task_send = current; |
576 | ||
8698a745 | 577 | set_user_nice(current, MIN_NICE); |
f4507164 | 578 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 579 | /* wait for something to do */ |
f4507164 | 580 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 581 | kthread_should_stop() || |
f4507164 | 582 | !list_empty(&nbd->waiting_queue)); |
48cf6061 LV |
583 | |
584 | /* extract request */ | |
f4507164 | 585 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
586 | continue; |
587 | ||
f4507164 WG |
588 | spin_lock_irq(&nbd->queue_lock); |
589 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
590 | queuelist); |
591 | list_del_init(&req->queuelist); | |
f4507164 | 592 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
593 | |
594 | /* handle request */ | |
f4507164 | 595 | nbd_handle_req(nbd, req); |
48cf6061 | 596 | } |
7e2893a1 MP |
597 | |
598 | nbd->task_send = NULL; | |
599 | ||
48cf6061 LV |
600 | return 0; |
601 | } | |
602 | ||
1da177e4 LT |
603 | /* |
604 | * We always wait for result of write, for now. It would be nice to make it optional | |
605 | * in future | |
f4507164 | 606 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
607 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
608 | */ | |
609 | ||
cad73b27 | 610 | static void nbd_request_handler(struct request_queue *q) |
398eb085 | 611 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
612 | { |
613 | struct request *req; | |
614 | ||
9934c8c0 | 615 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 616 | struct nbd_device *nbd; |
1da177e4 | 617 | |
48cf6061 LV |
618 | spin_unlock_irq(q->queue_lock); |
619 | ||
f4507164 | 620 | nbd = req->rq_disk->private_data; |
1da177e4 | 621 | |
f4507164 | 622 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 623 | |
d18509f5 MP |
624 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
625 | req, req->cmd_type); | |
626 | ||
f4507164 | 627 | if (unlikely(!nbd->sock)) { |
da6ccaaa DS |
628 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
629 | "Attempted send on closed socket\n"); | |
4d48a542 | 630 | req->errors++; |
d18509f5 | 631 | nbd_end_request(nbd, req); |
4d48a542 PC |
632 | spin_lock_irq(q->queue_lock); |
633 | continue; | |
634 | } | |
635 | ||
f4507164 WG |
636 | spin_lock_irq(&nbd->queue_lock); |
637 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
638 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 639 | |
f4507164 | 640 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 641 | |
1da177e4 | 642 | spin_lock_irq(q->queue_lock); |
1da177e4 | 643 | } |
1da177e4 LT |
644 | } |
645 | ||
23272a67 MP |
646 | static int nbd_set_socket(struct nbd_device *nbd, struct socket *sock) |
647 | { | |
648 | int ret = 0; | |
649 | ||
650 | spin_lock_irq(&nbd->sock_lock); | |
651 | ||
652 | if (nbd->sock) { | |
653 | ret = -EBUSY; | |
654 | goto out; | |
655 | } | |
656 | ||
657 | nbd->sock = sock; | |
658 | ||
659 | out: | |
660 | spin_unlock_irq(&nbd->sock_lock); | |
661 | ||
662 | return ret; | |
663 | } | |
664 | ||
0e4f0f6f MP |
665 | /* Reset all properties of an NBD device */ |
666 | static void nbd_reset(struct nbd_device *nbd) | |
667 | { | |
668 | nbd->disconnect = false; | |
669 | nbd->timedout = false; | |
670 | nbd->blksize = 1024; | |
671 | nbd->bytesize = 0; | |
672 | set_capacity(nbd->disk, 0); | |
673 | nbd->flags = 0; | |
674 | nbd->xmit_timeout = 0; | |
675 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
676 | del_timer_sync(&nbd->timeout_timer); | |
677 | } | |
678 | ||
679 | static void nbd_bdev_reset(struct block_device *bdev) | |
680 | { | |
681 | set_device_ro(bdev, false); | |
682 | bdev->bd_inode->i_size = 0; | |
683 | if (max_part > 0) { | |
684 | blkdev_reread_part(bdev); | |
685 | bdev->bd_invalidated = 1; | |
686 | } | |
687 | } | |
688 | ||
d02cf531 MP |
689 | static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev) |
690 | { | |
691 | if (nbd->flags & NBD_FLAG_READ_ONLY) | |
692 | set_device_ro(bdev, true); | |
693 | if (nbd->flags & NBD_FLAG_SEND_TRIM) | |
694 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
695 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) | |
aafb1eec | 696 | blk_queue_write_cache(nbd->disk->queue, true, false); |
d02cf531 | 697 | else |
aafb1eec | 698 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
699 | } |
700 | ||
30d53d9c MP |
701 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
702 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
703 | ||
1a2ad211 | 704 | /* Must be called with tx_lock held */ |
1da177e4 | 705 | |
f4507164 | 706 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
707 | unsigned int cmd, unsigned long arg) |
708 | { | |
1da177e4 | 709 | switch (cmd) { |
1a2ad211 PM |
710 | case NBD_DISCONNECT: { |
711 | struct request sreq; | |
712 | ||
f4507164 | 713 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
714 | if (!nbd->sock) |
715 | return -EINVAL; | |
1a2ad211 | 716 | |
3a2d63f8 PB |
717 | mutex_unlock(&nbd->tx_lock); |
718 | fsync_bdev(bdev); | |
719 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 720 | blk_rq_init(NULL, &sreq); |
4f8c9510 | 721 | sreq.cmd_type = REQ_TYPE_DRV_PRIV; |
3a2d63f8 PB |
722 | |
723 | /* Check again after getting mutex back. */ | |
f4507164 | 724 | if (!nbd->sock) |
1da177e4 | 725 | return -EINVAL; |
3a2d63f8 | 726 | |
696697cb | 727 | nbd->disconnect = true; |
c378f70a | 728 | |
f4507164 | 729 | nbd_send_req(nbd, &sreq); |
c378f70a | 730 | return 0; |
1a2ad211 | 731 | } |
1da177e4 | 732 | |
23272a67 MP |
733 | case NBD_CLEAR_SOCK: |
734 | sock_shutdown(nbd); | |
f4507164 WG |
735 | nbd_clear_que(nbd); |
736 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 737 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 738 | kill_bdev(bdev); |
1a2ad211 | 739 | return 0; |
1a2ad211 PM |
740 | |
741 | case NBD_SET_SOCK: { | |
e2511578 | 742 | int err; |
23272a67 MP |
743 | struct socket *sock = sockfd_lookup(arg, &err); |
744 | ||
745 | if (!sock) | |
746 | return err; | |
747 | ||
748 | err = nbd_set_socket(nbd, sock); | |
749 | if (!err && max_part) | |
750 | bdev->bd_invalidated = 1; | |
751 | ||
752 | return err; | |
1a2ad211 PM |
753 | } |
754 | ||
37091fdd | 755 | case NBD_SET_BLKSIZE: { |
5e454c67 | 756 | loff_t bsize = div_s64(nbd->bytesize, arg); |
37091fdd MP |
757 | |
758 | return nbd_size_set(nbd, bdev, arg, bsize); | |
759 | } | |
1a2ad211 | 760 | |
1da177e4 | 761 | case NBD_SET_SIZE: |
37091fdd MP |
762 | return nbd_size_set(nbd, bdev, nbd->blksize, |
763 | arg / nbd->blksize); | |
764 | ||
765 | case NBD_SET_SIZE_BLOCKS: | |
766 | return nbd_size_set(nbd, bdev, nbd->blksize, arg); | |
1a2ad211 | 767 | |
7fdfd406 | 768 | case NBD_SET_TIMEOUT: |
f4507164 | 769 | nbd->xmit_timeout = arg * HZ; |
7e2893a1 MP |
770 | if (arg) |
771 | mod_timer(&nbd->timeout_timer, | |
772 | jiffies + nbd->xmit_timeout); | |
773 | else | |
774 | del_timer_sync(&nbd->timeout_timer); | |
775 | ||
7fdfd406 | 776 | return 0; |
1a2ad211 | 777 | |
2f012508 PC |
778 | case NBD_SET_FLAGS: |
779 | nbd->flags = arg; | |
780 | return 0; | |
781 | ||
1a2ad211 PM |
782 | case NBD_DO_IT: { |
783 | struct task_struct *thread; | |
1a2ad211 PM |
784 | int error; |
785 | ||
6521d39a | 786 | if (nbd->task_recv) |
c91192d6 | 787 | return -EBUSY; |
e2511578 | 788 | if (!nbd->sock) |
1da177e4 | 789 | return -EINVAL; |
1a2ad211 | 790 | |
f4507164 | 791 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 792 | |
d02cf531 | 793 | nbd_parse_flags(nbd, bdev); |
a336d298 | 794 | |
cad73b27 | 795 | thread = kthread_run(nbd_thread_send, nbd, "%s", |
30d53d9c | 796 | nbd_name(nbd)); |
1a2ad211 | 797 | if (IS_ERR(thread)) { |
f4507164 | 798 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 799 | return PTR_ERR(thread); |
1a2ad211 | 800 | } |
d06df60b | 801 | |
30d53d9c | 802 | nbd_dev_dbg_init(nbd); |
37091fdd | 803 | error = nbd_thread_recv(nbd, bdev); |
30d53d9c | 804 | nbd_dev_dbg_close(nbd); |
48cf6061 | 805 | kthread_stop(thread); |
1a2ad211 | 806 | |
f4507164 | 807 | mutex_lock(&nbd->tx_lock); |
19391830 | 808 | |
36e47bee | 809 | sock_shutdown(nbd); |
f4507164 | 810 | nbd_clear_que(nbd); |
3a2d63f8 | 811 | kill_bdev(bdev); |
0e4f0f6f MP |
812 | nbd_bdev_reset(bdev); |
813 | ||
c378f70a | 814 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
1f7b5cf1 MP |
815 | error = 0; |
816 | if (nbd->timedout) | |
817 | error = -ETIMEDOUT; | |
818 | ||
0e4f0f6f MP |
819 | nbd_reset(nbd); |
820 | ||
19391830 | 821 | return error; |
1a2ad211 PM |
822 | } |
823 | ||
1da177e4 | 824 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
825 | /* |
826 | * This is for compatibility only. The queue is always cleared | |
827 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
828 | */ | |
1da177e4 | 829 | return 0; |
1a2ad211 | 830 | |
1da177e4 | 831 | case NBD_PRINT_DEBUG: |
f4507164 | 832 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 833 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
834 | nbd->queue_head.next, nbd->queue_head.prev, |
835 | &nbd->queue_head); | |
1da177e4 LT |
836 | return 0; |
837 | } | |
1a2ad211 PM |
838 | return -ENOTTY; |
839 | } | |
840 | ||
841 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
842 | unsigned int cmd, unsigned long arg) | |
843 | { | |
f4507164 | 844 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
845 | int error; |
846 | ||
847 | if (!capable(CAP_SYS_ADMIN)) | |
848 | return -EPERM; | |
849 | ||
f4507164 | 850 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 851 | |
f4507164 WG |
852 | mutex_lock(&nbd->tx_lock); |
853 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
854 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
855 | |
856 | return error; | |
1da177e4 LT |
857 | } |
858 | ||
83d5cde4 | 859 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
860 | { |
861 | .owner = THIS_MODULE, | |
8a6cfeb6 | 862 | .ioctl = nbd_ioctl, |
263a3df1 | 863 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
864 | }; |
865 | ||
30d53d9c MP |
866 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
867 | ||
868 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
869 | { | |
870 | struct nbd_device *nbd = s->private; | |
871 | ||
872 | if (nbd->task_recv) | |
873 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
874 | if (nbd->task_send) | |
875 | seq_printf(s, "send: %d\n", task_pid_nr(nbd->task_send)); | |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
880 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
881 | { | |
882 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
883 | } | |
884 | ||
885 | static const struct file_operations nbd_dbg_tasks_ops = { | |
886 | .open = nbd_dbg_tasks_open, | |
887 | .read = seq_read, | |
888 | .llseek = seq_lseek, | |
889 | .release = single_release, | |
890 | }; | |
891 | ||
892 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
893 | { | |
894 | struct nbd_device *nbd = s->private; | |
895 | u32 flags = nbd->flags; | |
896 | ||
897 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
898 | ||
899 | seq_puts(s, "Known flags:\n"); | |
900 | ||
901 | if (flags & NBD_FLAG_HAS_FLAGS) | |
902 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
903 | if (flags & NBD_FLAG_READ_ONLY) | |
904 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
905 | if (flags & NBD_FLAG_SEND_FLUSH) | |
906 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
907 | if (flags & NBD_FLAG_SEND_TRIM) | |
908 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
909 | ||
910 | return 0; | |
911 | } | |
912 | ||
913 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
914 | { | |
915 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
916 | } | |
917 | ||
918 | static const struct file_operations nbd_dbg_flags_ops = { | |
919 | .open = nbd_dbg_flags_open, | |
920 | .read = seq_read, | |
921 | .llseek = seq_lseek, | |
922 | .release = single_release, | |
923 | }; | |
924 | ||
925 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
926 | { | |
927 | struct dentry *dir; | |
27ea43fe MP |
928 | |
929 | if (!nbd_dbg_dir) | |
930 | return -EIO; | |
30d53d9c MP |
931 | |
932 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
933 | if (!dir) { |
934 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
935 | nbd_name(nbd)); | |
936 | return -EIO; | |
30d53d9c MP |
937 | } |
938 | nbd->dbg_dir = dir; | |
939 | ||
27ea43fe MP |
940 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
941 | debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize); | |
942 | debugfs_create_u32("timeout", 0444, dir, &nbd->xmit_timeout); | |
943 | debugfs_create_u32("blocksize", 0444, dir, &nbd->blksize); | |
d366a0ff | 944 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
945 | |
946 | return 0; | |
947 | } | |
948 | ||
949 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
950 | { | |
951 | debugfs_remove_recursive(nbd->dbg_dir); | |
952 | } | |
953 | ||
954 | static int nbd_dbg_init(void) | |
955 | { | |
956 | struct dentry *dbg_dir; | |
957 | ||
958 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
959 | if (!dbg_dir) |
960 | return -EIO; | |
30d53d9c MP |
961 | |
962 | nbd_dbg_dir = dbg_dir; | |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
967 | static void nbd_dbg_close(void) | |
968 | { | |
969 | debugfs_remove_recursive(nbd_dbg_dir); | |
970 | } | |
971 | ||
972 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
973 | ||
974 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
975 | { | |
976 | return 0; | |
977 | } | |
978 | ||
979 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
980 | { | |
981 | } | |
982 | ||
983 | static int nbd_dbg_init(void) | |
984 | { | |
985 | return 0; | |
986 | } | |
987 | ||
988 | static void nbd_dbg_close(void) | |
989 | { | |
990 | } | |
991 | ||
992 | #endif | |
993 | ||
1da177e4 LT |
994 | /* |
995 | * And here should be modules and kernel interface | |
996 | * (Just smiley confuses emacs :-) | |
997 | */ | |
998 | ||
999 | static int __init nbd_init(void) | |
1000 | { | |
1001 | int err = -ENOMEM; | |
1002 | int i; | |
d71a6d73 | 1003 | int part_shift; |
1da177e4 | 1004 | |
5b7b18cc | 1005 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 1006 | |
d71a6d73 | 1007 | if (max_part < 0) { |
7742ce4a | 1008 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
1009 | return -EINVAL; |
1010 | } | |
1011 | ||
1012 | part_shift = 0; | |
5988ce23 | 1013 | if (max_part > 0) { |
d71a6d73 LV |
1014 | part_shift = fls(max_part); |
1015 | ||
5988ce23 NK |
1016 | /* |
1017 | * Adjust max_part according to part_shift as it is exported | |
1018 | * to user space so that user can know the max number of | |
1019 | * partition kernel should be able to manage. | |
1020 | * | |
1021 | * Note that -1 is required because partition 0 is reserved | |
1022 | * for the whole disk. | |
1023 | */ | |
1024 | max_part = (1UL << part_shift) - 1; | |
1025 | } | |
1026 | ||
3b271082 NK |
1027 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1028 | return -EINVAL; | |
1029 | ||
1030 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
1031 | return -EINVAL; | |
1032 | ||
ff6b8090 SM |
1033 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
1034 | if (!nbd_dev) | |
1035 | return -ENOMEM; | |
1036 | ||
40be0c28 | 1037 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 1038 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
1039 | if (!disk) |
1040 | goto out; | |
1041 | nbd_dev[i].disk = disk; | |
1042 | /* | |
1043 | * The new linux 2.5 block layer implementation requires | |
1044 | * every gendisk to have its very own request_queue struct. | |
1045 | * These structs are big so we dynamically allocate them. | |
1046 | */ | |
cad73b27 | 1047 | disk->queue = blk_init_queue(nbd_request_handler, &nbd_lock); |
1da177e4 LT |
1048 | if (!disk->queue) { |
1049 | put_disk(disk); | |
1050 | goto out; | |
1051 | } | |
31dcfab0 JA |
1052 | /* |
1053 | * Tell the block layer that we are not a rotational device | |
1054 | */ | |
1055 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 1056 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 | 1057 | disk->queue->limits.discard_granularity = 512; |
2bb4cd5c | 1058 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); |
a336d298 | 1059 | disk->queue->limits.discard_zeroes_data = 0; |
078be02b MB |
1060 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1061 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
1062 | } |
1063 | ||
1064 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
1065 | err = -EIO; | |
1066 | goto out; | |
1067 | } | |
1068 | ||
1069 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 1070 | |
30d53d9c MP |
1071 | nbd_dbg_init(); |
1072 | ||
40be0c28 | 1073 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1074 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 1075 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 1076 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 | 1077 | spin_lock_init(&nbd_dev[i].queue_lock); |
23272a67 | 1078 | spin_lock_init(&nbd_dev[i].sock_lock); |
1da177e4 | 1079 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); |
82d4dc5a | 1080 | mutex_init(&nbd_dev[i].tx_lock); |
7e2893a1 MP |
1081 | init_timer(&nbd_dev[i].timeout_timer); |
1082 | nbd_dev[i].timeout_timer.function = nbd_xmit_timeout; | |
1083 | nbd_dev[i].timeout_timer.data = (unsigned long)&nbd_dev[i]; | |
4b2f0260 | 1084 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 1085 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 1086 | disk->major = NBD_MAJOR; |
d71a6d73 | 1087 | disk->first_minor = i << part_shift; |
1da177e4 LT |
1088 | disk->fops = &nbd_fops; |
1089 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 1090 | sprintf(disk->disk_name, "nbd%d", i); |
0e4f0f6f | 1091 | nbd_reset(&nbd_dev[i]); |
1da177e4 LT |
1092 | add_disk(disk); |
1093 | } | |
1094 | ||
1095 | return 0; | |
1096 | out: | |
1097 | while (i--) { | |
1098 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
1099 | put_disk(nbd_dev[i].disk); | |
1100 | } | |
f3944d61 | 1101 | kfree(nbd_dev); |
1da177e4 LT |
1102 | return err; |
1103 | } | |
1104 | ||
1105 | static void __exit nbd_cleanup(void) | |
1106 | { | |
1107 | int i; | |
30d53d9c MP |
1108 | |
1109 | nbd_dbg_close(); | |
1110 | ||
40be0c28 | 1111 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1112 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 1113 | nbd_dev[i].magic = 0; |
1da177e4 LT |
1114 | if (disk) { |
1115 | del_gendisk(disk); | |
1116 | blk_cleanup_queue(disk->queue); | |
1117 | put_disk(disk); | |
1118 | } | |
1119 | } | |
1da177e4 | 1120 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 1121 | kfree(nbd_dev); |
1da177e4 LT |
1122 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
1123 | } | |
1124 | ||
1125 | module_init(nbd_init); | |
1126 | module_exit(nbd_cleanup); | |
1127 | ||
1128 | MODULE_DESCRIPTION("Network Block Device"); | |
1129 | MODULE_LICENSE("GPL"); | |
1130 | ||
40be0c28 | 1131 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
1132 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
1133 | module_param(max_part, int, 0444); | |
1134 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |