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> |
fd8383fd | 37 | #include <linux/blk-mq.h> |
1da177e4 | 38 | |
7c0f6ba6 | 39 | #include <linux/uaccess.h> |
1da177e4 LT |
40 | #include <asm/types.h> |
41 | ||
42 | #include <linux/nbd.h> | |
43 | ||
9561a7ad JB |
44 | struct nbd_sock { |
45 | struct socket *sock; | |
46 | struct mutex tx_lock; | |
47 | }; | |
48 | ||
9b4a6ba9 JB |
49 | #define NBD_TIMEDOUT 0 |
50 | #define NBD_DISCONNECT_REQUESTED 1 | |
9561a7ad JB |
51 | #define NBD_DISCONNECTED 2 |
52 | #define NBD_RUNNING 3 | |
9b4a6ba9 | 53 | |
13e71d69 | 54 | struct nbd_device { |
22d109c1 | 55 | u32 flags; |
9b4a6ba9 | 56 | unsigned long runtime_flags; |
9561a7ad | 57 | struct nbd_sock **socks; |
13e71d69 MP |
58 | int magic; |
59 | ||
fd8383fd | 60 | struct blk_mq_tag_set tag_set; |
13e71d69 | 61 | |
9561a7ad | 62 | struct mutex config_lock; |
13e71d69 | 63 | struct gendisk *disk; |
9561a7ad JB |
64 | int num_connections; |
65 | atomic_t recv_threads; | |
66 | wait_queue_head_t recv_wq; | |
ef77b515 | 67 | loff_t blksize; |
b9c495bb | 68 | loff_t bytesize; |
7e2893a1 | 69 | |
7e2893a1 | 70 | struct task_struct *task_recv; |
9561a7ad | 71 | struct task_struct *task_setup; |
30d53d9c MP |
72 | |
73 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
74 | struct dentry *dbg_dir; | |
75 | #endif | |
13e71d69 MP |
76 | }; |
77 | ||
fd8383fd JB |
78 | struct nbd_cmd { |
79 | struct nbd_device *nbd; | |
9561a7ad | 80 | struct completion send_complete; |
fd8383fd JB |
81 | }; |
82 | ||
30d53d9c MP |
83 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
84 | static struct dentry *nbd_dbg_dir; | |
85 | #endif | |
86 | ||
87 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
88 | ||
f4507164 | 89 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 90 | |
9c7a4169 | 91 | static unsigned int nbds_max = 16; |
20a8143e | 92 | static struct nbd_device *nbd_dev; |
d71a6d73 | 93 | static int max_part; |
1da177e4 | 94 | |
d18509f5 | 95 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 96 | { |
d18509f5 | 97 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
98 | } |
99 | ||
37091fdd MP |
100 | static bool nbd_is_connected(struct nbd_device *nbd) |
101 | { | |
102 | return !!nbd->task_recv; | |
103 | } | |
104 | ||
1da177e4 LT |
105 | static const char *nbdcmd_to_ascii(int cmd) |
106 | { | |
107 | switch (cmd) { | |
108 | case NBD_CMD_READ: return "read"; | |
109 | case NBD_CMD_WRITE: return "write"; | |
110 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 111 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 112 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
113 | } |
114 | return "invalid"; | |
115 | } | |
1da177e4 | 116 | |
37091fdd MP |
117 | static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev) |
118 | { | |
119 | bdev->bd_inode->i_size = 0; | |
120 | set_capacity(nbd->disk, 0); | |
121 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
122 | ||
123 | return 0; | |
124 | } | |
125 | ||
126 | static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev) | |
127 | { | |
128 | if (!nbd_is_connected(nbd)) | |
129 | return; | |
130 | ||
131 | bdev->bd_inode->i_size = nbd->bytesize; | |
132 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
133 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
134 | } | |
135 | ||
136 | static int nbd_size_set(struct nbd_device *nbd, struct block_device *bdev, | |
ef77b515 | 137 | loff_t blocksize, loff_t nr_blocks) |
37091fdd MP |
138 | { |
139 | int ret; | |
140 | ||
141 | ret = set_blocksize(bdev, blocksize); | |
142 | if (ret) | |
143 | return ret; | |
144 | ||
145 | nbd->blksize = blocksize; | |
ef77b515 | 146 | nbd->bytesize = blocksize * nr_blocks; |
37091fdd MP |
147 | |
148 | nbd_size_update(nbd, bdev); | |
149 | ||
150 | return 0; | |
151 | } | |
152 | ||
fd8383fd | 153 | static void nbd_end_request(struct nbd_cmd *cmd) |
1da177e4 | 154 | { |
fd8383fd JB |
155 | struct nbd_device *nbd = cmd->nbd; |
156 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
097c94a4 | 157 | int error = req->errors ? -EIO : 0; |
1da177e4 | 158 | |
fd8383fd | 159 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd, |
d18509f5 | 160 | error ? "failed" : "done"); |
1da177e4 | 161 | |
fd8383fd | 162 | blk_mq_complete_request(req, error); |
1da177e4 LT |
163 | } |
164 | ||
e018e757 MP |
165 | /* |
166 | * Forcibly shutdown the socket causing all listeners to error | |
167 | */ | |
36e47bee | 168 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 169 | { |
9561a7ad | 170 | int i; |
23272a67 | 171 | |
9561a7ad JB |
172 | if (nbd->num_connections == 0) |
173 | return; | |
174 | if (test_and_set_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) | |
260bbce4 | 175 | return; |
23272a67 | 176 | |
9561a7ad JB |
177 | for (i = 0; i < nbd->num_connections; i++) { |
178 | struct nbd_sock *nsock = nbd->socks[i]; | |
179 | mutex_lock(&nsock->tx_lock); | |
180 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); | |
181 | mutex_unlock(&nsock->tx_lock); | |
182 | } | |
183 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
7fdfd406 PC |
184 | } |
185 | ||
0eadf37a JB |
186 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, |
187 | bool reserved) | |
7fdfd406 | 188 | { |
0eadf37a JB |
189 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
190 | struct nbd_device *nbd = cmd->nbd; | |
dcc909d9 | 191 | |
9561a7ad | 192 | dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n"); |
9b4a6ba9 | 193 | set_bit(NBD_TIMEDOUT, &nbd->runtime_flags); |
0eadf37a | 194 | req->errors++; |
9561a7ad JB |
195 | |
196 | /* | |
197 | * If our disconnect packet times out then we're already holding the | |
198 | * config_lock and could deadlock here, so just set an error and return, | |
199 | * we'll handle shutting everything down later. | |
200 | */ | |
201 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
202 | return BLK_EH_HANDLED; | |
203 | mutex_lock(&nbd->config_lock); | |
204 | sock_shutdown(nbd); | |
205 | mutex_unlock(&nbd->config_lock); | |
0eadf37a | 206 | return BLK_EH_HANDLED; |
7fdfd406 PC |
207 | } |
208 | ||
1da177e4 LT |
209 | /* |
210 | * Send or receive packet. | |
211 | */ | |
9561a7ad JB |
212 | static int sock_xmit(struct nbd_device *nbd, int index, int send, void *buf, |
213 | int size, int msg_flags) | |
1da177e4 | 214 | { |
9561a7ad | 215 | struct socket *sock = nbd->socks[index]->sock; |
1da177e4 LT |
216 | int result; |
217 | struct msghdr msg; | |
218 | struct kvec iov; | |
7f338fe4 | 219 | unsigned long pflags = current->flags; |
1da177e4 | 220 | |
ffc41cf8 | 221 | if (unlikely(!sock)) { |
a897b666 | 222 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
223 | "Attempted %s on closed socket in sock_xmit\n", |
224 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
225 | return -EINVAL; |
226 | } | |
227 | ||
7f338fe4 | 228 | current->flags |= PF_MEMALLOC; |
1da177e4 | 229 | do { |
7f338fe4 | 230 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
231 | iov.iov_base = buf; |
232 | iov.iov_len = size; | |
233 | msg.msg_name = NULL; | |
234 | msg.msg_namelen = 0; | |
235 | msg.msg_control = NULL; | |
236 | msg.msg_controllen = 0; | |
1da177e4 LT |
237 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
238 | ||
7e2893a1 | 239 | if (send) |
1da177e4 | 240 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7e2893a1 | 241 | else |
35fbf5bc NK |
242 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
243 | msg.msg_flags); | |
1da177e4 | 244 | |
1da177e4 LT |
245 | if (result <= 0) { |
246 | if (result == 0) | |
247 | result = -EPIPE; /* short read */ | |
248 | break; | |
249 | } | |
250 | size -= result; | |
251 | buf += result; | |
252 | } while (size > 0); | |
253 | ||
7f338fe4 | 254 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 LT |
255 | |
256 | return result; | |
257 | } | |
258 | ||
9561a7ad JB |
259 | static inline int sock_send_bvec(struct nbd_device *nbd, int index, |
260 | struct bio_vec *bvec, int flags) | |
1da177e4 LT |
261 | { |
262 | int result; | |
263 | void *kaddr = kmap(bvec->bv_page); | |
9561a7ad | 264 | result = sock_xmit(nbd, index, 1, kaddr + bvec->bv_offset, |
f4507164 | 265 | bvec->bv_len, flags); |
1da177e4 LT |
266 | kunmap(bvec->bv_page); |
267 | return result; | |
268 | } | |
269 | ||
7fdfd406 | 270 | /* always call with the tx_lock held */ |
9561a7ad | 271 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) |
1da177e4 | 272 | { |
fd8383fd | 273 | struct request *req = blk_mq_rq_from_pdu(cmd); |
5705f702 | 274 | int result, flags; |
1da177e4 | 275 | struct nbd_request request; |
1011c1b9 | 276 | unsigned long size = blk_rq_bytes(req); |
429a787b | 277 | struct bio *bio; |
9dc6c806 | 278 | u32 type; |
9561a7ad | 279 | u32 tag = blk_mq_unique_tag(req); |
9dc6c806 | 280 | |
9561a7ad | 281 | if (req_op(req) == REQ_OP_DISCARD) |
9dc6c806 | 282 | type = NBD_CMD_TRIM; |
3a5e02ce | 283 | else if (req_op(req) == REQ_OP_FLUSH) |
9dc6c806 CH |
284 | type = NBD_CMD_FLUSH; |
285 | else if (rq_data_dir(req) == WRITE) | |
286 | type = NBD_CMD_WRITE; | |
287 | else | |
288 | type = NBD_CMD_READ; | |
1da177e4 | 289 | |
04cfac4e | 290 | memset(&request, 0, sizeof(request)); |
1da177e4 | 291 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 | 292 | request.type = htonl(type); |
9561a7ad | 293 | if (type != NBD_CMD_FLUSH) { |
75f187ab AB |
294 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
295 | request.len = htonl(size); | |
296 | } | |
9561a7ad | 297 | memcpy(request.handle, &tag, sizeof(tag)); |
1da177e4 | 298 | |
d18509f5 | 299 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
fd8383fd | 300 | cmd, nbdcmd_to_ascii(type), |
d18509f5 | 301 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
9561a7ad | 302 | result = sock_xmit(nbd, index, 1, &request, sizeof(request), |
9dc6c806 | 303 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 304 | if (result <= 0) { |
a897b666 | 305 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 | 306 | "Send control failed (result %d)\n", result); |
dab5313a | 307 | return -EIO; |
1da177e4 LT |
308 | } |
309 | ||
429a787b JA |
310 | if (type != NBD_CMD_WRITE) |
311 | return 0; | |
312 | ||
313 | flags = 0; | |
314 | bio = req->bio; | |
315 | while (bio) { | |
316 | struct bio *next = bio->bi_next; | |
317 | struct bvec_iter iter; | |
7988613b | 318 | struct bio_vec bvec; |
429a787b JA |
319 | |
320 | bio_for_each_segment(bvec, bio, iter) { | |
321 | bool is_last = !next && bio_iter_last(bvec, iter); | |
322 | ||
323 | if (is_last) | |
6c92e699 | 324 | flags = MSG_MORE; |
d18509f5 | 325 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
fd8383fd | 326 | cmd, bvec.bv_len); |
9561a7ad | 327 | result = sock_send_bvec(nbd, index, &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 | } |
429a787b JA |
334 | /* |
335 | * The completion might already have come in, | |
336 | * so break for the last one instead of letting | |
337 | * the iterator do it. This prevents use-after-free | |
338 | * of the bio. | |
339 | */ | |
340 | if (is_last) | |
341 | break; | |
1da177e4 | 342 | } |
429a787b | 343 | bio = next; |
1da177e4 | 344 | } |
1da177e4 | 345 | return 0; |
1da177e4 LT |
346 | } |
347 | ||
9561a7ad JB |
348 | static inline int sock_recv_bvec(struct nbd_device *nbd, int index, |
349 | struct bio_vec *bvec) | |
1da177e4 LT |
350 | { |
351 | int result; | |
352 | void *kaddr = kmap(bvec->bv_page); | |
9561a7ad JB |
353 | result = sock_xmit(nbd, index, 0, kaddr + bvec->bv_offset, |
354 | bvec->bv_len, MSG_WAITALL); | |
1da177e4 LT |
355 | kunmap(bvec->bv_page); |
356 | return result; | |
357 | } | |
358 | ||
359 | /* NULL returned = something went wrong, inform userspace */ | |
9561a7ad | 360 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) |
1da177e4 LT |
361 | { |
362 | int result; | |
363 | struct nbd_reply reply; | |
fd8383fd JB |
364 | struct nbd_cmd *cmd; |
365 | struct request *req = NULL; | |
366 | u16 hwq; | |
9561a7ad | 367 | u32 tag; |
1da177e4 LT |
368 | |
369 | reply.magic = 0; | |
9561a7ad | 370 | result = sock_xmit(nbd, index, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 371 | if (result <= 0) { |
9561a7ad JB |
372 | if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) && |
373 | !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags)) | |
374 | dev_err(disk_to_dev(nbd->disk), | |
375 | "Receive control failed (result %d)\n", result); | |
19391830 | 376 | return ERR_PTR(result); |
1da177e4 | 377 | } |
e4b57e08 MF |
378 | |
379 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 380 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 381 | (unsigned long)ntohl(reply.magic)); |
19391830 | 382 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
383 | } |
384 | ||
9561a7ad | 385 | memcpy(&tag, reply.handle, sizeof(u32)); |
4b2f0260 | 386 | |
fd8383fd JB |
387 | hwq = blk_mq_unique_tag_to_hwq(tag); |
388 | if (hwq < nbd->tag_set.nr_hw_queues) | |
389 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
390 | blk_mq_unique_tag_to_tag(tag)); | |
391 | if (!req || !blk_mq_request_started(req)) { | |
392 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
393 | tag, req); | |
394 | return ERR_PTR(-ENOENT); | |
1da177e4 | 395 | } |
fd8383fd | 396 | cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 397 | if (ntohl(reply.error)) { |
f4507164 | 398 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 399 | ntohl(reply.error)); |
1da177e4 | 400 | req->errors++; |
fd8383fd | 401 | return cmd; |
1da177e4 LT |
402 | } |
403 | ||
fd8383fd | 404 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd); |
9dc6c806 | 405 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 406 | struct req_iterator iter; |
7988613b | 407 | struct bio_vec bvec; |
5705f702 N |
408 | |
409 | rq_for_each_segment(bvec, req, iter) { | |
9561a7ad | 410 | result = sock_recv_bvec(nbd, index, &bvec); |
6c92e699 | 411 | if (result <= 0) { |
f4507164 | 412 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 413 | result); |
6c92e699 | 414 | req->errors++; |
fd8383fd | 415 | return cmd; |
6c92e699 | 416 | } |
d18509f5 | 417 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
fd8383fd | 418 | cmd, bvec.bv_len); |
1da177e4 | 419 | } |
9561a7ad JB |
420 | } else { |
421 | /* See the comment in nbd_queue_rq. */ | |
422 | wait_for_completion(&cmd->send_complete); | |
1da177e4 | 423 | } |
fd8383fd | 424 | return cmd; |
1da177e4 LT |
425 | } |
426 | ||
edfaa7c3 KS |
427 | static ssize_t pid_show(struct device *dev, |
428 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 429 | { |
edfaa7c3 | 430 | struct gendisk *disk = dev_to_disk(dev); |
6521d39a | 431 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; |
edfaa7c3 | 432 | |
6521d39a | 433 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); |
6b39bb65 PC |
434 | } |
435 | ||
edfaa7c3 | 436 | static struct device_attribute pid_attr = { |
01e8ef11 | 437 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
438 | .show = pid_show, |
439 | }; | |
440 | ||
9561a7ad JB |
441 | struct recv_thread_args { |
442 | struct work_struct work; | |
443 | struct nbd_device *nbd; | |
444 | int index; | |
445 | }; | |
446 | ||
447 | static void recv_work(struct work_struct *work) | |
1da177e4 | 448 | { |
9561a7ad JB |
449 | struct recv_thread_args *args = container_of(work, |
450 | struct recv_thread_args, | |
451 | work); | |
452 | struct nbd_device *nbd = args->nbd; | |
fd8383fd | 453 | struct nbd_cmd *cmd; |
9561a7ad | 454 | int ret = 0; |
1da177e4 | 455 | |
f4507164 | 456 | BUG_ON(nbd->magic != NBD_MAGIC); |
19391830 | 457 | while (1) { |
9561a7ad | 458 | cmd = nbd_read_stat(nbd, args->index); |
fd8383fd JB |
459 | if (IS_ERR(cmd)) { |
460 | ret = PTR_ERR(cmd); | |
19391830 MP |
461 | break; |
462 | } | |
463 | ||
fd8383fd | 464 | nbd_end_request(cmd); |
19391830 | 465 | } |
6b39bb65 | 466 | |
9561a7ad JB |
467 | /* |
468 | * We got an error, shut everybody down if this wasn't the result of a | |
469 | * disconnect request. | |
470 | */ | |
471 | if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags)) | |
472 | sock_shutdown(nbd); | |
473 | atomic_dec(&nbd->recv_threads); | |
474 | wake_up(&nbd->recv_wq); | |
1da177e4 LT |
475 | } |
476 | ||
fd8383fd | 477 | static void nbd_clear_req(struct request *req, void *data, bool reserved) |
1da177e4 | 478 | { |
fd8383fd | 479 | struct nbd_cmd *cmd; |
1da177e4 | 480 | |
fd8383fd JB |
481 | if (!blk_mq_request_started(req)) |
482 | return; | |
483 | cmd = blk_mq_rq_to_pdu(req); | |
484 | req->errors++; | |
485 | nbd_end_request(cmd); | |
486 | } | |
487 | ||
488 | static void nbd_clear_que(struct nbd_device *nbd) | |
489 | { | |
f4507164 | 490 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 491 | |
fd8383fd | 492 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); |
e78273c8 | 493 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
494 | } |
495 | ||
7fdfd406 | 496 | |
9561a7ad | 497 | static void nbd_handle_cmd(struct nbd_cmd *cmd, int index) |
48cf6061 | 498 | { |
fd8383fd JB |
499 | struct request *req = blk_mq_rq_from_pdu(cmd); |
500 | struct nbd_device *nbd = cmd->nbd; | |
9561a7ad | 501 | struct nbd_sock *nsock; |
fd8383fd | 502 | |
9561a7ad | 503 | if (index >= nbd->num_connections) { |
a897b666 JB |
504 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
505 | "Attempted send on invalid socket\n"); | |
48cf6061 | 506 | goto error_out; |
9561a7ad | 507 | } |
48cf6061 | 508 | |
9561a7ad | 509 | if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) { |
a897b666 JB |
510 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
511 | "Attempted send on closed socket\n"); | |
9561a7ad JB |
512 | goto error_out; |
513 | } | |
514 | ||
515 | if (req->cmd_type != REQ_TYPE_FS && | |
516 | req->cmd_type != REQ_TYPE_DRV_PRIV) | |
517 | goto error_out; | |
518 | ||
519 | if (req->cmd_type == REQ_TYPE_FS && | |
520 | rq_data_dir(req) == WRITE && | |
9dc6c806 | 521 | (nbd->flags & NBD_FLAG_READ_ONLY)) { |
a897b666 JB |
522 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
523 | "Write on read-only\n"); | |
9dc6c806 | 524 | goto error_out; |
75f187ab AB |
525 | } |
526 | ||
48cf6061 LV |
527 | req->errors = 0; |
528 | ||
9561a7ad JB |
529 | nsock = nbd->socks[index]; |
530 | mutex_lock(&nsock->tx_lock); | |
531 | if (unlikely(!nsock->sock)) { | |
532 | mutex_unlock(&nsock->tx_lock); | |
a897b666 JB |
533 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
534 | "Attempted send on closed socket\n"); | |
15746fca | 535 | goto error_out; |
48cf6061 LV |
536 | } |
537 | ||
9561a7ad | 538 | if (nbd_send_cmd(nbd, cmd, index) != 0) { |
a897b666 JB |
539 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
540 | "Request send failed\n"); | |
48cf6061 | 541 | req->errors++; |
fd8383fd | 542 | nbd_end_request(cmd); |
48cf6061 LV |
543 | } |
544 | ||
9561a7ad | 545 | mutex_unlock(&nsock->tx_lock); |
48cf6061 LV |
546 | |
547 | return; | |
548 | ||
549 | error_out: | |
550 | req->errors++; | |
fd8383fd | 551 | nbd_end_request(cmd); |
48cf6061 LV |
552 | } |
553 | ||
fd8383fd JB |
554 | static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
555 | const struct blk_mq_queue_data *bd) | |
1da177e4 | 556 | { |
fd8383fd | 557 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); |
1da177e4 | 558 | |
9561a7ad JB |
559 | /* |
560 | * Since we look at the bio's to send the request over the network we | |
561 | * need to make sure the completion work doesn't mark this request done | |
562 | * before we are done doing our send. This keeps us from dereferencing | |
563 | * freed data if we have particularly fast completions (ie we get the | |
564 | * completion before we exit sock_xmit on the last bvec) or in the case | |
565 | * that the server is misbehaving (or there was an error) before we're | |
566 | * done sending everything over the wire. | |
567 | */ | |
568 | init_completion(&cmd->send_complete); | |
fd8383fd | 569 | blk_mq_start_request(bd->rq); |
9561a7ad JB |
570 | nbd_handle_cmd(cmd, hctx->queue_num); |
571 | complete(&cmd->send_complete); | |
572 | ||
fd8383fd | 573 | return BLK_MQ_RQ_QUEUE_OK; |
1da177e4 LT |
574 | } |
575 | ||
9561a7ad | 576 | static int nbd_add_socket(struct nbd_device *nbd, struct socket *sock) |
23272a67 | 577 | { |
9561a7ad JB |
578 | struct nbd_sock **socks; |
579 | struct nbd_sock *nsock; | |
23272a67 | 580 | |
9561a7ad JB |
581 | if (!nbd->task_setup) |
582 | nbd->task_setup = current; | |
583 | if (nbd->task_setup != current) { | |
584 | dev_err(disk_to_dev(nbd->disk), | |
585 | "Device being setup by another task"); | |
586 | return -EINVAL; | |
23272a67 MP |
587 | } |
588 | ||
9561a7ad JB |
589 | socks = krealloc(nbd->socks, (nbd->num_connections + 1) * |
590 | sizeof(struct nbd_sock *), GFP_KERNEL); | |
591 | if (!socks) | |
592 | return -ENOMEM; | |
593 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); | |
594 | if (!nsock) | |
595 | return -ENOMEM; | |
596 | ||
597 | nbd->socks = socks; | |
23272a67 | 598 | |
9561a7ad JB |
599 | mutex_init(&nsock->tx_lock); |
600 | nsock->sock = sock; | |
601 | socks[nbd->num_connections++] = nsock; | |
23272a67 | 602 | |
9561a7ad | 603 | return 0; |
23272a67 MP |
604 | } |
605 | ||
0e4f0f6f MP |
606 | /* Reset all properties of an NBD device */ |
607 | static void nbd_reset(struct nbd_device *nbd) | |
608 | { | |
9561a7ad JB |
609 | int i; |
610 | ||
611 | for (i = 0; i < nbd->num_connections; i++) | |
612 | kfree(nbd->socks[i]); | |
613 | kfree(nbd->socks); | |
614 | nbd->socks = NULL; | |
9b4a6ba9 | 615 | nbd->runtime_flags = 0; |
0e4f0f6f MP |
616 | nbd->blksize = 1024; |
617 | nbd->bytesize = 0; | |
618 | set_capacity(nbd->disk, 0); | |
619 | nbd->flags = 0; | |
0eadf37a | 620 | nbd->tag_set.timeout = 0; |
9561a7ad JB |
621 | nbd->num_connections = 0; |
622 | nbd->task_setup = NULL; | |
0e4f0f6f | 623 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
0e4f0f6f MP |
624 | } |
625 | ||
626 | static void nbd_bdev_reset(struct block_device *bdev) | |
627 | { | |
628 | set_device_ro(bdev, false); | |
629 | bdev->bd_inode->i_size = 0; | |
630 | if (max_part > 0) { | |
631 | blkdev_reread_part(bdev); | |
632 | bdev->bd_invalidated = 1; | |
633 | } | |
634 | } | |
635 | ||
d02cf531 MP |
636 | static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev) |
637 | { | |
638 | if (nbd->flags & NBD_FLAG_READ_ONLY) | |
639 | set_device_ro(bdev, true); | |
640 | if (nbd->flags & NBD_FLAG_SEND_TRIM) | |
641 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
642 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) | |
aafb1eec | 643 | blk_queue_write_cache(nbd->disk->queue, true, false); |
d02cf531 | 644 | else |
aafb1eec | 645 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
646 | } |
647 | ||
9561a7ad JB |
648 | static void send_disconnects(struct nbd_device *nbd) |
649 | { | |
650 | struct nbd_request request = {}; | |
651 | int i, ret; | |
652 | ||
653 | request.magic = htonl(NBD_REQUEST_MAGIC); | |
654 | request.type = htonl(NBD_CMD_DISC); | |
655 | ||
656 | for (i = 0; i < nbd->num_connections; i++) { | |
657 | ret = sock_xmit(nbd, i, 1, &request, sizeof(request), 0); | |
658 | if (ret <= 0) | |
659 | dev_err(disk_to_dev(nbd->disk), | |
660 | "Send disconnect failed %d\n", ret); | |
661 | } | |
662 | } | |
663 | ||
30d53d9c MP |
664 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
665 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
666 | ||
9561a7ad | 667 | /* Must be called with config_lock held */ |
f4507164 | 668 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
669 | unsigned int cmd, unsigned long arg) |
670 | { | |
1da177e4 | 671 | switch (cmd) { |
1a2ad211 | 672 | case NBD_DISCONNECT: { |
f4507164 | 673 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
9561a7ad | 674 | if (!nbd->socks) |
3a2d63f8 | 675 | return -EINVAL; |
1a2ad211 | 676 | |
9561a7ad | 677 | mutex_unlock(&nbd->config_lock); |
3a2d63f8 | 678 | fsync_bdev(bdev); |
9561a7ad | 679 | mutex_lock(&nbd->config_lock); |
3a2d63f8 PB |
680 | |
681 | /* Check again after getting mutex back. */ | |
9561a7ad | 682 | if (!nbd->socks) |
1da177e4 | 683 | return -EINVAL; |
c378f70a | 684 | |
9561a7ad JB |
685 | if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED, |
686 | &nbd->runtime_flags)) | |
687 | send_disconnects(nbd); | |
c378f70a | 688 | return 0; |
1a2ad211 | 689 | } |
9561a7ad | 690 | |
23272a67 MP |
691 | case NBD_CLEAR_SOCK: |
692 | sock_shutdown(nbd); | |
f4507164 | 693 | nbd_clear_que(nbd); |
3a2d63f8 | 694 | kill_bdev(bdev); |
9561a7ad JB |
695 | nbd_bdev_reset(bdev); |
696 | /* | |
697 | * We want to give the run thread a chance to wait for everybody | |
698 | * to clean up and then do it's own cleanup. | |
699 | */ | |
700 | if (!test_bit(NBD_RUNNING, &nbd->runtime_flags)) { | |
701 | int i; | |
702 | ||
703 | for (i = 0; i < nbd->num_connections; i++) | |
704 | kfree(nbd->socks[i]); | |
705 | kfree(nbd->socks); | |
706 | nbd->socks = NULL; | |
707 | nbd->num_connections = 0; | |
20032ec3 | 708 | nbd->task_setup = NULL; |
9561a7ad | 709 | } |
1a2ad211 | 710 | return 0; |
1a2ad211 PM |
711 | |
712 | case NBD_SET_SOCK: { | |
e2511578 | 713 | int err; |
23272a67 MP |
714 | struct socket *sock = sockfd_lookup(arg, &err); |
715 | ||
716 | if (!sock) | |
717 | return err; | |
718 | ||
9561a7ad | 719 | err = nbd_add_socket(nbd, sock); |
23272a67 MP |
720 | if (!err && max_part) |
721 | bdev->bd_invalidated = 1; | |
722 | ||
723 | return err; | |
1a2ad211 PM |
724 | } |
725 | ||
37091fdd | 726 | case NBD_SET_BLKSIZE: { |
5e454c67 | 727 | loff_t bsize = div_s64(nbd->bytesize, arg); |
37091fdd MP |
728 | |
729 | return nbd_size_set(nbd, bdev, arg, bsize); | |
730 | } | |
1a2ad211 | 731 | |
1da177e4 | 732 | case NBD_SET_SIZE: |
37091fdd | 733 | return nbd_size_set(nbd, bdev, nbd->blksize, |
e88f72cb | 734 | div_s64(arg, nbd->blksize)); |
37091fdd MP |
735 | |
736 | case NBD_SET_SIZE_BLOCKS: | |
737 | return nbd_size_set(nbd, bdev, nbd->blksize, arg); | |
1a2ad211 | 738 | |
7fdfd406 | 739 | case NBD_SET_TIMEOUT: |
0eadf37a | 740 | nbd->tag_set.timeout = arg * HZ; |
7fdfd406 | 741 | return 0; |
1a2ad211 | 742 | |
2f012508 PC |
743 | case NBD_SET_FLAGS: |
744 | nbd->flags = arg; | |
745 | return 0; | |
746 | ||
1a2ad211 | 747 | case NBD_DO_IT: { |
9561a7ad JB |
748 | struct recv_thread_args *args; |
749 | int num_connections = nbd->num_connections; | |
feffa5cc | 750 | int error = 0, i; |
1a2ad211 | 751 | |
6521d39a | 752 | if (nbd->task_recv) |
c91192d6 | 753 | return -EBUSY; |
9561a7ad | 754 | if (!nbd->socks) |
1da177e4 | 755 | return -EINVAL; |
9561a7ad JB |
756 | if (num_connections > 1 && |
757 | !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) { | |
758 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); | |
feffa5cc | 759 | error = -EINVAL; |
9561a7ad JB |
760 | goto out_err; |
761 | } | |
1a2ad211 | 762 | |
9561a7ad JB |
763 | set_bit(NBD_RUNNING, &nbd->runtime_flags); |
764 | blk_mq_update_nr_hw_queues(&nbd->tag_set, nbd->num_connections); | |
765 | args = kcalloc(num_connections, sizeof(*args), GFP_KERNEL); | |
feffa5cc JA |
766 | if (!args) { |
767 | error = -ENOMEM; | |
9561a7ad | 768 | goto out_err; |
feffa5cc | 769 | } |
97240963 | 770 | nbd->task_recv = current; |
9561a7ad | 771 | mutex_unlock(&nbd->config_lock); |
1a2ad211 | 772 | |
d02cf531 | 773 | nbd_parse_flags(nbd, bdev); |
a336d298 | 774 | |
9561a7ad JB |
775 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
776 | if (error) { | |
777 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
778 | goto out_recv; | |
779 | } | |
780 | ||
781 | nbd_size_update(nbd, bdev); | |
782 | ||
30d53d9c | 783 | nbd_dev_dbg_init(nbd); |
9561a7ad JB |
784 | for (i = 0; i < num_connections; i++) { |
785 | sk_set_memalloc(nbd->socks[i]->sock->sk); | |
786 | atomic_inc(&nbd->recv_threads); | |
787 | INIT_WORK(&args[i].work, recv_work); | |
788 | args[i].nbd = nbd; | |
789 | args[i].index = i; | |
790 | queue_work(system_long_wq, &args[i].work); | |
791 | } | |
792 | wait_event_interruptible(nbd->recv_wq, | |
793 | atomic_read(&nbd->recv_threads) == 0); | |
794 | for (i = 0; i < num_connections; i++) | |
795 | flush_work(&args[i].work); | |
30d53d9c | 796 | nbd_dev_dbg_close(nbd); |
9561a7ad JB |
797 | nbd_size_clear(nbd, bdev); |
798 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
799 | out_recv: | |
800 | mutex_lock(&nbd->config_lock); | |
97240963 | 801 | nbd->task_recv = NULL; |
9561a7ad | 802 | out_err: |
36e47bee | 803 | sock_shutdown(nbd); |
f4507164 | 804 | nbd_clear_que(nbd); |
3a2d63f8 | 805 | kill_bdev(bdev); |
0e4f0f6f MP |
806 | nbd_bdev_reset(bdev); |
807 | ||
9b4a6ba9 JB |
808 | /* user requested, ignore socket errors */ |
809 | if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags)) | |
1f7b5cf1 | 810 | error = 0; |
9b4a6ba9 | 811 | if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags)) |
1f7b5cf1 MP |
812 | error = -ETIMEDOUT; |
813 | ||
0e4f0f6f | 814 | nbd_reset(nbd); |
19391830 | 815 | return error; |
1a2ad211 PM |
816 | } |
817 | ||
1da177e4 | 818 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
819 | /* |
820 | * This is for compatibility only. The queue is always cleared | |
821 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
822 | */ | |
1da177e4 | 823 | return 0; |
1a2ad211 | 824 | |
1da177e4 | 825 | case NBD_PRINT_DEBUG: |
fd8383fd JB |
826 | /* |
827 | * For compatibility only, we no longer keep a list of | |
828 | * outstanding requests. | |
829 | */ | |
1da177e4 LT |
830 | return 0; |
831 | } | |
1a2ad211 PM |
832 | return -ENOTTY; |
833 | } | |
834 | ||
835 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
836 | unsigned int cmd, unsigned long arg) | |
837 | { | |
f4507164 | 838 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
839 | int error; |
840 | ||
841 | if (!capable(CAP_SYS_ADMIN)) | |
842 | return -EPERM; | |
843 | ||
f4507164 | 844 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 845 | |
9561a7ad | 846 | mutex_lock(&nbd->config_lock); |
f4507164 | 847 | error = __nbd_ioctl(bdev, nbd, cmd, arg); |
9561a7ad | 848 | mutex_unlock(&nbd->config_lock); |
1a2ad211 PM |
849 | |
850 | return error; | |
1da177e4 LT |
851 | } |
852 | ||
83d5cde4 | 853 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
854 | { |
855 | .owner = THIS_MODULE, | |
8a6cfeb6 | 856 | .ioctl = nbd_ioctl, |
263a3df1 | 857 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
858 | }; |
859 | ||
30d53d9c MP |
860 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
861 | ||
862 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
863 | { | |
864 | struct nbd_device *nbd = s->private; | |
865 | ||
866 | if (nbd->task_recv) | |
867 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
30d53d9c MP |
868 | |
869 | return 0; | |
870 | } | |
871 | ||
872 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
873 | { | |
874 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
875 | } | |
876 | ||
877 | static const struct file_operations nbd_dbg_tasks_ops = { | |
878 | .open = nbd_dbg_tasks_open, | |
879 | .read = seq_read, | |
880 | .llseek = seq_lseek, | |
881 | .release = single_release, | |
882 | }; | |
883 | ||
884 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
885 | { | |
886 | struct nbd_device *nbd = s->private; | |
887 | u32 flags = nbd->flags; | |
888 | ||
889 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
890 | ||
891 | seq_puts(s, "Known flags:\n"); | |
892 | ||
893 | if (flags & NBD_FLAG_HAS_FLAGS) | |
894 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
895 | if (flags & NBD_FLAG_READ_ONLY) | |
896 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
897 | if (flags & NBD_FLAG_SEND_FLUSH) | |
898 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
899 | if (flags & NBD_FLAG_SEND_TRIM) | |
900 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
901 | ||
902 | return 0; | |
903 | } | |
904 | ||
905 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
906 | { | |
907 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
908 | } | |
909 | ||
910 | static const struct file_operations nbd_dbg_flags_ops = { | |
911 | .open = nbd_dbg_flags_open, | |
912 | .read = seq_read, | |
913 | .llseek = seq_lseek, | |
914 | .release = single_release, | |
915 | }; | |
916 | ||
917 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
918 | { | |
919 | struct dentry *dir; | |
27ea43fe MP |
920 | |
921 | if (!nbd_dbg_dir) | |
922 | return -EIO; | |
30d53d9c MP |
923 | |
924 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
925 | if (!dir) { |
926 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
927 | nbd_name(nbd)); | |
928 | return -EIO; | |
30d53d9c MP |
929 | } |
930 | nbd->dbg_dir = dir; | |
931 | ||
27ea43fe MP |
932 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
933 | debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize); | |
0eadf37a | 934 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); |
ef77b515 | 935 | debugfs_create_u64("blocksize", 0444, dir, &nbd->blksize); |
d366a0ff | 936 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
937 | |
938 | return 0; | |
939 | } | |
940 | ||
941 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
942 | { | |
943 | debugfs_remove_recursive(nbd->dbg_dir); | |
944 | } | |
945 | ||
946 | static int nbd_dbg_init(void) | |
947 | { | |
948 | struct dentry *dbg_dir; | |
949 | ||
950 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
951 | if (!dbg_dir) |
952 | return -EIO; | |
30d53d9c MP |
953 | |
954 | nbd_dbg_dir = dbg_dir; | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
959 | static void nbd_dbg_close(void) | |
960 | { | |
961 | debugfs_remove_recursive(nbd_dbg_dir); | |
962 | } | |
963 | ||
964 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
965 | ||
966 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
967 | { | |
968 | return 0; | |
969 | } | |
970 | ||
971 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
972 | { | |
973 | } | |
974 | ||
975 | static int nbd_dbg_init(void) | |
976 | { | |
977 | return 0; | |
978 | } | |
979 | ||
980 | static void nbd_dbg_close(void) | |
981 | { | |
982 | } | |
983 | ||
984 | #endif | |
985 | ||
fd8383fd JB |
986 | static int nbd_init_request(void *data, struct request *rq, |
987 | unsigned int hctx_idx, unsigned int request_idx, | |
988 | unsigned int numa_node) | |
989 | { | |
990 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
fd8383fd | 991 | cmd->nbd = data; |
fd8383fd JB |
992 | return 0; |
993 | } | |
994 | ||
995 | static struct blk_mq_ops nbd_mq_ops = { | |
996 | .queue_rq = nbd_queue_rq, | |
fd8383fd | 997 | .init_request = nbd_init_request, |
0eadf37a | 998 | .timeout = nbd_xmit_timeout, |
fd8383fd JB |
999 | }; |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * And here should be modules and kernel interface | |
1003 | * (Just smiley confuses emacs :-) | |
1004 | */ | |
1005 | ||
1006 | static int __init nbd_init(void) | |
1007 | { | |
1008 | int err = -ENOMEM; | |
1009 | int i; | |
d71a6d73 | 1010 | int part_shift; |
1da177e4 | 1011 | |
5b7b18cc | 1012 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 1013 | |
d71a6d73 | 1014 | if (max_part < 0) { |
7742ce4a | 1015 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
1016 | return -EINVAL; |
1017 | } | |
1018 | ||
1019 | part_shift = 0; | |
5988ce23 | 1020 | if (max_part > 0) { |
d71a6d73 LV |
1021 | part_shift = fls(max_part); |
1022 | ||
5988ce23 NK |
1023 | /* |
1024 | * Adjust max_part according to part_shift as it is exported | |
1025 | * to user space so that user can know the max number of | |
1026 | * partition kernel should be able to manage. | |
1027 | * | |
1028 | * Note that -1 is required because partition 0 is reserved | |
1029 | * for the whole disk. | |
1030 | */ | |
1031 | max_part = (1UL << part_shift) - 1; | |
1032 | } | |
1033 | ||
3b271082 NK |
1034 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1035 | return -EINVAL; | |
1036 | ||
1037 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
1038 | return -EINVAL; | |
1039 | ||
ff6b8090 SM |
1040 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
1041 | if (!nbd_dev) | |
1042 | return -ENOMEM; | |
1043 | ||
40be0c28 | 1044 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 1045 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
1046 | if (!disk) |
1047 | goto out; | |
1048 | nbd_dev[i].disk = disk; | |
fd8383fd JB |
1049 | |
1050 | nbd_dev[i].tag_set.ops = &nbd_mq_ops; | |
1051 | nbd_dev[i].tag_set.nr_hw_queues = 1; | |
1052 | nbd_dev[i].tag_set.queue_depth = 128; | |
1053 | nbd_dev[i].tag_set.numa_node = NUMA_NO_NODE; | |
1054 | nbd_dev[i].tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1055 | nbd_dev[i].tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
005043ac | 1056 | BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING; |
fd8383fd JB |
1057 | nbd_dev[i].tag_set.driver_data = &nbd_dev[i]; |
1058 | ||
1059 | err = blk_mq_alloc_tag_set(&nbd_dev[i].tag_set); | |
1060 | if (err) { | |
1061 | put_disk(disk); | |
1062 | goto out; | |
1063 | } | |
1064 | ||
1da177e4 LT |
1065 | /* |
1066 | * The new linux 2.5 block layer implementation requires | |
1067 | * every gendisk to have its very own request_queue struct. | |
1068 | * These structs are big so we dynamically allocate them. | |
1069 | */ | |
fd8383fd | 1070 | disk->queue = blk_mq_init_queue(&nbd_dev[i].tag_set); |
1da177e4 | 1071 | if (!disk->queue) { |
fd8383fd | 1072 | blk_mq_free_tag_set(&nbd_dev[i].tag_set); |
1da177e4 LT |
1073 | put_disk(disk); |
1074 | goto out; | |
1075 | } | |
fd8383fd | 1076 | |
31dcfab0 JA |
1077 | /* |
1078 | * Tell the block layer that we are not a rotational device | |
1079 | */ | |
1080 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 1081 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 | 1082 | disk->queue->limits.discard_granularity = 512; |
2bb4cd5c | 1083 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); |
a336d298 | 1084 | disk->queue->limits.discard_zeroes_data = 0; |
078be02b MB |
1085 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1086 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
1087 | } |
1088 | ||
1089 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
1090 | err = -EIO; | |
1091 | goto out; | |
1092 | } | |
1093 | ||
1094 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 1095 | |
30d53d9c MP |
1096 | nbd_dbg_init(); |
1097 | ||
40be0c28 | 1098 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1099 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 1100 | nbd_dev[i].magic = NBD_MAGIC; |
9561a7ad | 1101 | mutex_init(&nbd_dev[i].config_lock); |
1da177e4 | 1102 | disk->major = NBD_MAJOR; |
d71a6d73 | 1103 | disk->first_minor = i << part_shift; |
1da177e4 LT |
1104 | disk->fops = &nbd_fops; |
1105 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 1106 | sprintf(disk->disk_name, "nbd%d", i); |
9561a7ad | 1107 | init_waitqueue_head(&nbd_dev[i].recv_wq); |
0e4f0f6f | 1108 | nbd_reset(&nbd_dev[i]); |
1da177e4 LT |
1109 | add_disk(disk); |
1110 | } | |
1111 | ||
1112 | return 0; | |
1113 | out: | |
1114 | while (i--) { | |
fd8383fd | 1115 | blk_mq_free_tag_set(&nbd_dev[i].tag_set); |
1da177e4 LT |
1116 | blk_cleanup_queue(nbd_dev[i].disk->queue); |
1117 | put_disk(nbd_dev[i].disk); | |
1118 | } | |
f3944d61 | 1119 | kfree(nbd_dev); |
1da177e4 LT |
1120 | return err; |
1121 | } | |
1122 | ||
1123 | static void __exit nbd_cleanup(void) | |
1124 | { | |
1125 | int i; | |
30d53d9c MP |
1126 | |
1127 | nbd_dbg_close(); | |
1128 | ||
40be0c28 | 1129 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1130 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 1131 | nbd_dev[i].magic = 0; |
1da177e4 LT |
1132 | if (disk) { |
1133 | del_gendisk(disk); | |
1134 | blk_cleanup_queue(disk->queue); | |
fd8383fd | 1135 | blk_mq_free_tag_set(&nbd_dev[i].tag_set); |
1da177e4 LT |
1136 | put_disk(disk); |
1137 | } | |
1138 | } | |
1da177e4 | 1139 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 1140 | kfree(nbd_dev); |
1da177e4 LT |
1141 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
1142 | } | |
1143 | ||
1144 | module_init(nbd_init); | |
1145 | module_exit(nbd_cleanup); | |
1146 | ||
1147 | MODULE_DESCRIPTION("Network Block Device"); | |
1148 | MODULE_LICENSE("GPL"); | |
1149 | ||
40be0c28 | 1150 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
1151 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
1152 | module_param(max_part, int, 0444); | |
1153 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |