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; |
696697cb | 60 | bool disconnect; /* a disconnect has been requested by user */ |
7e2893a1 MP |
61 | |
62 | struct timer_list timeout_timer; | |
dcc909d9 | 63 | spinlock_t tasks_lock; |
7e2893a1 MP |
64 | struct task_struct *task_recv; |
65 | struct task_struct *task_send; | |
30d53d9c MP |
66 | |
67 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
68 | struct dentry *dbg_dir; | |
69 | #endif | |
13e71d69 MP |
70 | }; |
71 | ||
30d53d9c MP |
72 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
73 | static struct dentry *nbd_dbg_dir; | |
74 | #endif | |
75 | ||
76 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
77 | ||
f4507164 | 78 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 79 | |
9c7a4169 | 80 | static unsigned int nbds_max = 16; |
20a8143e | 81 | static struct nbd_device *nbd_dev; |
d71a6d73 | 82 | static int max_part; |
1da177e4 LT |
83 | |
84 | /* | |
85 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
86 | * 1. Each NIC is essentially a synchronization point for all servers | |
87 | * accessed through that NIC so there's no need to have more locks | |
88 | * than NICs anyway. | |
89 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
90 | * down each lock to the point where they're actually slower than just | |
91 | * a single lock. | |
92 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
93 | */ | |
94 | static DEFINE_SPINLOCK(nbd_lock); | |
95 | ||
d18509f5 | 96 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 97 | { |
d18509f5 | 98 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
99 | } |
100 | ||
101 | static const char *nbdcmd_to_ascii(int cmd) | |
102 | { | |
103 | switch (cmd) { | |
104 | case NBD_CMD_READ: return "read"; | |
105 | case NBD_CMD_WRITE: return "write"; | |
106 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 107 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 108 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
109 | } |
110 | return "invalid"; | |
111 | } | |
1da177e4 | 112 | |
d18509f5 | 113 | static void nbd_end_request(struct nbd_device *nbd, struct request *req) |
1da177e4 | 114 | { |
097c94a4 | 115 | int error = req->errors ? -EIO : 0; |
165125e1 | 116 | struct request_queue *q = req->q; |
1da177e4 LT |
117 | unsigned long flags; |
118 | ||
d18509f5 MP |
119 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req, |
120 | error ? "failed" : "done"); | |
1da177e4 LT |
121 | |
122 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 123 | __blk_end_request_all(req, error); |
1da177e4 LT |
124 | spin_unlock_irqrestore(q->queue_lock, flags); |
125 | } | |
126 | ||
e018e757 MP |
127 | /* |
128 | * Forcibly shutdown the socket causing all listeners to error | |
129 | */ | |
36e47bee | 130 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 131 | { |
260bbce4 MP |
132 | if (!nbd->sock) |
133 | return; | |
134 | ||
135 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
136 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
137 | nbd->sock = NULL; | |
138 | del_timer_sync(&nbd->timeout_timer); | |
7fdfd406 PC |
139 | } |
140 | ||
141 | static void nbd_xmit_timeout(unsigned long arg) | |
142 | { | |
7e2893a1 | 143 | struct nbd_device *nbd = (struct nbd_device *)arg; |
dcc909d9 | 144 | unsigned long flags; |
7e2893a1 MP |
145 | |
146 | if (list_empty(&nbd->queue_head)) | |
147 | return; | |
148 | ||
696697cb | 149 | nbd->disconnect = true; |
7e2893a1 | 150 | |
dcc909d9 MP |
151 | spin_lock_irqsave(&nbd->tasks_lock, flags); |
152 | ||
153 | if (nbd->task_recv) | |
154 | force_sig(SIGKILL, nbd->task_recv); | |
7fdfd406 | 155 | |
dcc909d9 | 156 | if (nbd->task_send) |
7e2893a1 MP |
157 | force_sig(SIGKILL, nbd->task_send); |
158 | ||
dcc909d9 MP |
159 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
160 | ||
7e2893a1 | 161 | dev_err(nbd_to_dev(nbd), "Connection timed out, killed receiver and sender, shutting down connection\n"); |
7fdfd406 PC |
162 | } |
163 | ||
1da177e4 LT |
164 | /* |
165 | * Send or receive packet. | |
166 | */ | |
f4507164 | 167 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
168 | int msg_flags) |
169 | { | |
f4507164 | 170 | struct socket *sock = nbd->sock; |
1da177e4 LT |
171 | int result; |
172 | struct msghdr msg; | |
173 | struct kvec iov; | |
be0ef957 | 174 | sigset_t blocked, oldset; |
7f338fe4 | 175 | unsigned long pflags = current->flags; |
1da177e4 | 176 | |
ffc41cf8 | 177 | if (unlikely(!sock)) { |
f4507164 | 178 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
179 | "Attempted %s on closed socket in sock_xmit\n", |
180 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
181 | return -EINVAL; |
182 | } | |
183 | ||
1da177e4 LT |
184 | /* Allow interception of SIGKILL only |
185 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
186 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
187 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 | 188 | |
7f338fe4 | 189 | current->flags |= PF_MEMALLOC; |
1da177e4 | 190 | do { |
7f338fe4 | 191 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
192 | iov.iov_base = buf; |
193 | iov.iov_len = size; | |
194 | msg.msg_name = NULL; | |
195 | msg.msg_namelen = 0; | |
196 | msg.msg_control = NULL; | |
197 | msg.msg_controllen = 0; | |
1da177e4 LT |
198 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
199 | ||
7e2893a1 | 200 | if (send) |
1da177e4 | 201 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7e2893a1 | 202 | else |
35fbf5bc NK |
203 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
204 | msg.msg_flags); | |
1da177e4 | 205 | |
1da177e4 LT |
206 | if (result <= 0) { |
207 | if (result == 0) | |
208 | result = -EPIPE; /* short read */ | |
209 | break; | |
210 | } | |
211 | size -= result; | |
212 | buf += result; | |
213 | } while (size > 0); | |
214 | ||
be0ef957 | 215 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 216 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 | 217 | |
7e2893a1 MP |
218 | if (!send && nbd->xmit_timeout) |
219 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
220 | ||
1da177e4 LT |
221 | return result; |
222 | } | |
223 | ||
f4507164 | 224 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
225 | int flags) |
226 | { | |
227 | int result; | |
228 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
229 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
230 | bvec->bv_len, flags); | |
1da177e4 LT |
231 | kunmap(bvec->bv_page); |
232 | return result; | |
233 | } | |
234 | ||
7fdfd406 | 235 | /* always call with the tx_lock held */ |
f4507164 | 236 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 237 | { |
5705f702 | 238 | int result, flags; |
1da177e4 | 239 | struct nbd_request request; |
1011c1b9 | 240 | unsigned long size = blk_rq_bytes(req); |
9dc6c806 CH |
241 | u32 type; |
242 | ||
243 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
244 | type = NBD_CMD_DISC; | |
245 | else if (req->cmd_flags & REQ_DISCARD) | |
246 | type = NBD_CMD_TRIM; | |
247 | else if (req->cmd_flags & REQ_FLUSH) | |
248 | type = NBD_CMD_FLUSH; | |
249 | else if (rq_data_dir(req) == WRITE) | |
250 | type = NBD_CMD_WRITE; | |
251 | else | |
252 | type = NBD_CMD_READ; | |
1da177e4 | 253 | |
04cfac4e | 254 | memset(&request, 0, sizeof(request)); |
1da177e4 | 255 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 CH |
256 | request.type = htonl(type); |
257 | if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) { | |
75f187ab AB |
258 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
259 | request.len = htonl(size); | |
260 | } | |
1da177e4 LT |
261 | memcpy(request.handle, &req, sizeof(req)); |
262 | ||
d18509f5 | 263 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
9dc6c806 | 264 | req, nbdcmd_to_ascii(type), |
d18509f5 | 265 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
f4507164 | 266 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
9dc6c806 | 267 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 268 | if (result <= 0) { |
f4507164 | 269 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 270 | "Send control failed (result %d)\n", result); |
dab5313a | 271 | return -EIO; |
1da177e4 LT |
272 | } |
273 | ||
9dc6c806 | 274 | if (type == NBD_CMD_WRITE) { |
5705f702 | 275 | struct req_iterator iter; |
7988613b | 276 | struct bio_vec bvec; |
1da177e4 LT |
277 | /* |
278 | * we are really probing at internals to determine | |
279 | * whether to set MSG_MORE or not... | |
280 | */ | |
5705f702 | 281 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 | 282 | flags = 0; |
4550dd6c | 283 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 284 | flags = MSG_MORE; |
d18509f5 MP |
285 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
286 | req, bvec.bv_len); | |
7988613b | 287 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 288 | if (result <= 0) { |
f4507164 | 289 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
290 | "Send data failed (result %d)\n", |
291 | result); | |
dab5313a | 292 | return -EIO; |
6c92e699 | 293 | } |
1da177e4 LT |
294 | } |
295 | } | |
1da177e4 | 296 | return 0; |
1da177e4 LT |
297 | } |
298 | ||
f4507164 | 299 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 300 | struct request *xreq) |
1da177e4 | 301 | { |
d2c9740b | 302 | struct request *req, *tmp; |
4b2f0260 | 303 | int err; |
1da177e4 | 304 | |
f4507164 | 305 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 306 | if (unlikely(err)) |
de9ad6d4 | 307 | return ERR_PTR(err); |
4b2f0260 | 308 | |
f4507164 WG |
309 | spin_lock(&nbd->queue_lock); |
310 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
311 | if (req != xreq) |
312 | continue; | |
313 | list_del_init(&req->queuelist); | |
f4507164 | 314 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
315 | return req; |
316 | } | |
f4507164 | 317 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 318 | |
de9ad6d4 | 319 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
320 | } |
321 | ||
f4507164 | 322 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
323 | { |
324 | int result; | |
325 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 326 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
327 | MSG_WAITALL); |
328 | kunmap(bvec->bv_page); | |
329 | return result; | |
330 | } | |
331 | ||
332 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 333 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
334 | { |
335 | int result; | |
336 | struct nbd_reply reply; | |
337 | struct request *req; | |
1da177e4 LT |
338 | |
339 | reply.magic = 0; | |
f4507164 | 340 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 341 | if (result <= 0) { |
f4507164 | 342 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 343 | "Receive control failed (result %d)\n", result); |
19391830 | 344 | return ERR_PTR(result); |
1da177e4 | 345 | } |
e4b57e08 MF |
346 | |
347 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 348 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 349 | (unsigned long)ntohl(reply.magic)); |
19391830 | 350 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
351 | } |
352 | ||
f4507164 | 353 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 354 | if (IS_ERR(req)) { |
4b2f0260 HX |
355 | result = PTR_ERR(req); |
356 | if (result != -ENOENT) | |
19391830 | 357 | return ERR_PTR(result); |
4b2f0260 | 358 | |
f4507164 | 359 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 360 | reply.handle); |
19391830 | 361 | return ERR_PTR(-EBADR); |
1da177e4 LT |
362 | } |
363 | ||
1da177e4 | 364 | if (ntohl(reply.error)) { |
f4507164 | 365 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 366 | ntohl(reply.error)); |
1da177e4 LT |
367 | req->errors++; |
368 | return req; | |
369 | } | |
370 | ||
d18509f5 | 371 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 372 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 373 | struct req_iterator iter; |
7988613b | 374 | struct bio_vec bvec; |
5705f702 N |
375 | |
376 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 377 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 378 | if (result <= 0) { |
f4507164 | 379 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 380 | result); |
6c92e699 JA |
381 | req->errors++; |
382 | return req; | |
383 | } | |
d18509f5 MP |
384 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
385 | req, bvec.bv_len); | |
1da177e4 LT |
386 | } |
387 | } | |
388 | return req; | |
1da177e4 LT |
389 | } |
390 | ||
edfaa7c3 KS |
391 | static ssize_t pid_show(struct device *dev, |
392 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 393 | { |
edfaa7c3 | 394 | struct gendisk *disk = dev_to_disk(dev); |
6521d39a | 395 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; |
edfaa7c3 | 396 | |
6521d39a | 397 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); |
6b39bb65 PC |
398 | } |
399 | ||
edfaa7c3 | 400 | static struct device_attribute pid_attr = { |
01e8ef11 | 401 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
402 | .show = pid_show, |
403 | }; | |
404 | ||
cad73b27 | 405 | static int nbd_thread_recv(struct nbd_device *nbd) |
1da177e4 LT |
406 | { |
407 | struct request *req; | |
84963048 | 408 | int ret; |
dcc909d9 | 409 | unsigned long flags; |
1da177e4 | 410 | |
f4507164 | 411 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 412 | |
7f338fe4 | 413 | sk_set_memalloc(nbd->sock->sk); |
6521d39a | 414 | |
dcc909d9 | 415 | spin_lock_irqsave(&nbd->tasks_lock, flags); |
6521d39a | 416 | nbd->task_recv = current; |
dcc909d9 | 417 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
6521d39a | 418 | |
f4507164 | 419 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
84963048 | 420 | if (ret) { |
f4507164 | 421 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
dcc909d9 MP |
422 | |
423 | spin_lock_irqsave(&nbd->tasks_lock, flags); | |
6521d39a | 424 | nbd->task_recv = NULL; |
dcc909d9 MP |
425 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
426 | ||
84963048 WC |
427 | return ret; |
428 | } | |
6b39bb65 | 429 | |
19391830 MP |
430 | while (1) { |
431 | req = nbd_read_stat(nbd); | |
432 | if (IS_ERR(req)) { | |
433 | ret = PTR_ERR(req); | |
434 | break; | |
435 | } | |
436 | ||
d18509f5 | 437 | nbd_end_request(nbd, req); |
19391830 | 438 | } |
6b39bb65 | 439 | |
6521d39a MP |
440 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
441 | ||
dcc909d9 | 442 | spin_lock_irqsave(&nbd->tasks_lock, flags); |
7e2893a1 | 443 | nbd->task_recv = NULL; |
dcc909d9 | 444 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
7e2893a1 MP |
445 | |
446 | if (signal_pending(current)) { | |
be0e6f29 | 447 | ret = kernel_dequeue_signal(NULL); |
7e2893a1 MP |
448 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", |
449 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
450 | mutex_lock(&nbd->tx_lock); |
451 | sock_shutdown(nbd); | |
452 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
453 | ret = -ETIMEDOUT; |
454 | } | |
455 | ||
7e2893a1 | 456 | return ret; |
1da177e4 LT |
457 | } |
458 | ||
f4507164 | 459 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
460 | { |
461 | struct request *req; | |
462 | ||
f4507164 | 463 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 464 | |
4b2f0260 | 465 | /* |
f4507164 | 466 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
467 | * modifications to the list must have completed by now. For |
468 | * the same reason, the active_req must be NULL. | |
469 | * | |
470 | * As a consequence, we don't need to take the spin lock while | |
471 | * purging the list here. | |
472 | */ | |
f4507164 WG |
473 | BUG_ON(nbd->sock); |
474 | BUG_ON(nbd->active_req); | |
4b2f0260 | 475 | |
f4507164 WG |
476 | while (!list_empty(&nbd->queue_head)) { |
477 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
478 | queuelist); |
479 | list_del_init(&req->queuelist); | |
480 | req->errors++; | |
d18509f5 | 481 | nbd_end_request(nbd, req); |
4b2f0260 | 482 | } |
fded4e09 PC |
483 | |
484 | while (!list_empty(&nbd->waiting_queue)) { | |
485 | req = list_entry(nbd->waiting_queue.next, struct request, | |
486 | queuelist); | |
487 | list_del_init(&req->queuelist); | |
488 | req->errors++; | |
d18509f5 | 489 | nbd_end_request(nbd, req); |
fded4e09 | 490 | } |
e78273c8 | 491 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
492 | } |
493 | ||
7fdfd406 | 494 | |
f4507164 | 495 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 496 | { |
33659ebb | 497 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
498 | goto error_out; |
499 | ||
9dc6c806 CH |
500 | if (rq_data_dir(req) == WRITE && |
501 | (nbd->flags & NBD_FLAG_READ_ONLY)) { | |
502 | dev_err(disk_to_dev(nbd->disk), | |
503 | "Write on read-only\n"); | |
504 | goto error_out; | |
75f187ab AB |
505 | } |
506 | ||
48cf6061 LV |
507 | req->errors = 0; |
508 | ||
f4507164 WG |
509 | mutex_lock(&nbd->tx_lock); |
510 | if (unlikely(!nbd->sock)) { | |
511 | mutex_unlock(&nbd->tx_lock); | |
512 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 513 | "Attempted send on closed socket\n"); |
15746fca | 514 | goto error_out; |
48cf6061 LV |
515 | } |
516 | ||
f4507164 | 517 | nbd->active_req = req; |
48cf6061 | 518 | |
7e2893a1 MP |
519 | if (nbd->xmit_timeout && list_empty_careful(&nbd->queue_head)) |
520 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
521 | ||
f4507164 WG |
522 | if (nbd_send_req(nbd, req) != 0) { |
523 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 524 | req->errors++; |
d18509f5 | 525 | nbd_end_request(nbd, req); |
48cf6061 | 526 | } else { |
f4507164 | 527 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 528 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 529 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
530 | } |
531 | ||
f4507164 WG |
532 | nbd->active_req = NULL; |
533 | mutex_unlock(&nbd->tx_lock); | |
534 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
535 | |
536 | return; | |
537 | ||
538 | error_out: | |
539 | req->errors++; | |
d18509f5 | 540 | nbd_end_request(nbd, req); |
48cf6061 LV |
541 | } |
542 | ||
cad73b27 | 543 | static int nbd_thread_send(void *data) |
48cf6061 | 544 | { |
f4507164 | 545 | struct nbd_device *nbd = data; |
48cf6061 | 546 | struct request *req; |
dcc909d9 | 547 | unsigned long flags; |
48cf6061 | 548 | |
dcc909d9 | 549 | spin_lock_irqsave(&nbd->tasks_lock, flags); |
7e2893a1 | 550 | nbd->task_send = current; |
dcc909d9 | 551 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
7e2893a1 | 552 | |
8698a745 | 553 | set_user_nice(current, MIN_NICE); |
f4507164 | 554 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 555 | /* wait for something to do */ |
f4507164 | 556 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 557 | kthread_should_stop() || |
f4507164 | 558 | !list_empty(&nbd->waiting_queue)); |
48cf6061 | 559 | |
7e2893a1 | 560 | if (signal_pending(current)) { |
be0e6f29 | 561 | int ret = kernel_dequeue_signal(NULL); |
7e2893a1 | 562 | |
7e2893a1 MP |
563 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", |
564 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
565 | mutex_lock(&nbd->tx_lock); |
566 | sock_shutdown(nbd); | |
567 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
568 | break; |
569 | } | |
570 | ||
48cf6061 | 571 | /* extract request */ |
f4507164 | 572 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
573 | continue; |
574 | ||
f4507164 WG |
575 | spin_lock_irq(&nbd->queue_lock); |
576 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
577 | queuelist); |
578 | list_del_init(&req->queuelist); | |
f4507164 | 579 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
580 | |
581 | /* handle request */ | |
f4507164 | 582 | nbd_handle_req(nbd, req); |
48cf6061 | 583 | } |
7e2893a1 | 584 | |
dcc909d9 | 585 | spin_lock_irqsave(&nbd->tasks_lock, flags); |
7e2893a1 | 586 | nbd->task_send = NULL; |
dcc909d9 MP |
587 | spin_unlock_irqrestore(&nbd->tasks_lock, flags); |
588 | ||
589 | /* Clear maybe pending signals */ | |
be0e6f29 ON |
590 | if (signal_pending(current)) |
591 | kernel_dequeue_signal(NULL); | |
7e2893a1 | 592 | |
48cf6061 LV |
593 | return 0; |
594 | } | |
595 | ||
1da177e4 LT |
596 | /* |
597 | * We always wait for result of write, for now. It would be nice to make it optional | |
598 | * in future | |
f4507164 | 599 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
600 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
601 | */ | |
602 | ||
cad73b27 | 603 | static void nbd_request_handler(struct request_queue *q) |
398eb085 | 604 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
605 | { |
606 | struct request *req; | |
607 | ||
9934c8c0 | 608 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 609 | struct nbd_device *nbd; |
1da177e4 | 610 | |
48cf6061 LV |
611 | spin_unlock_irq(q->queue_lock); |
612 | ||
f4507164 | 613 | nbd = req->rq_disk->private_data; |
1da177e4 | 614 | |
f4507164 | 615 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 616 | |
d18509f5 MP |
617 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
618 | req, req->cmd_type); | |
619 | ||
f4507164 WG |
620 | if (unlikely(!nbd->sock)) { |
621 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 622 | "Attempted send on closed socket\n"); |
4d48a542 | 623 | req->errors++; |
d18509f5 | 624 | nbd_end_request(nbd, req); |
4d48a542 PC |
625 | spin_lock_irq(q->queue_lock); |
626 | continue; | |
627 | } | |
628 | ||
f4507164 WG |
629 | spin_lock_irq(&nbd->queue_lock); |
630 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
631 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 632 | |
f4507164 | 633 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 634 | |
1da177e4 | 635 | spin_lock_irq(q->queue_lock); |
1da177e4 | 636 | } |
1da177e4 LT |
637 | } |
638 | ||
30d53d9c MP |
639 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
640 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
641 | ||
1a2ad211 | 642 | /* Must be called with tx_lock held */ |
1da177e4 | 643 | |
f4507164 | 644 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
645 | unsigned int cmd, unsigned long arg) |
646 | { | |
1da177e4 | 647 | switch (cmd) { |
1a2ad211 PM |
648 | case NBD_DISCONNECT: { |
649 | struct request sreq; | |
650 | ||
f4507164 | 651 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
652 | if (!nbd->sock) |
653 | return -EINVAL; | |
1a2ad211 | 654 | |
3a2d63f8 PB |
655 | mutex_unlock(&nbd->tx_lock); |
656 | fsync_bdev(bdev); | |
657 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 658 | blk_rq_init(NULL, &sreq); |
4f8c9510 | 659 | sreq.cmd_type = REQ_TYPE_DRV_PRIV; |
3a2d63f8 PB |
660 | |
661 | /* Check again after getting mutex back. */ | |
f4507164 | 662 | if (!nbd->sock) |
1da177e4 | 663 | return -EINVAL; |
3a2d63f8 | 664 | |
696697cb | 665 | nbd->disconnect = true; |
c378f70a | 666 | |
f4507164 | 667 | nbd_send_req(nbd, &sreq); |
c378f70a | 668 | return 0; |
1a2ad211 | 669 | } |
1da177e4 | 670 | |
1a2ad211 | 671 | case NBD_CLEAR_SOCK: { |
e2511578 | 672 | struct socket *sock = nbd->sock; |
f4507164 | 673 | nbd->sock = NULL; |
f4507164 WG |
674 | nbd_clear_que(nbd); |
675 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 676 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 677 | kill_bdev(bdev); |
e2511578 AV |
678 | if (sock) |
679 | sockfd_put(sock); | |
1a2ad211 PM |
680 | return 0; |
681 | } | |
682 | ||
683 | case NBD_SET_SOCK: { | |
e2511578 AV |
684 | struct socket *sock; |
685 | int err; | |
686 | if (nbd->sock) | |
1da177e4 | 687 | return -EBUSY; |
e2511578 AV |
688 | sock = sockfd_lookup(arg, &err); |
689 | if (sock) { | |
690 | nbd->sock = sock; | |
691 | if (max_part > 0) | |
692 | bdev->bd_invalidated = 1; | |
696697cb | 693 | nbd->disconnect = false; /* we're connected now */ |
e2511578 | 694 | return 0; |
1da177e4 | 695 | } |
1a2ad211 PM |
696 | return -EINVAL; |
697 | } | |
698 | ||
1da177e4 | 699 | case NBD_SET_BLKSIZE: |
f4507164 WG |
700 | nbd->blksize = arg; |
701 | nbd->bytesize &= ~(nbd->blksize-1); | |
702 | bdev->bd_inode->i_size = nbd->bytesize; | |
703 | set_blocksize(bdev, nbd->blksize); | |
704 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 705 | return 0; |
1a2ad211 | 706 | |
1da177e4 | 707 | case NBD_SET_SIZE: |
f4507164 WG |
708 | nbd->bytesize = arg & ~(nbd->blksize-1); |
709 | bdev->bd_inode->i_size = nbd->bytesize; | |
710 | set_blocksize(bdev, nbd->blksize); | |
711 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 712 | return 0; |
1a2ad211 | 713 | |
7fdfd406 | 714 | case NBD_SET_TIMEOUT: |
f4507164 | 715 | nbd->xmit_timeout = arg * HZ; |
7e2893a1 MP |
716 | if (arg) |
717 | mod_timer(&nbd->timeout_timer, | |
718 | jiffies + nbd->xmit_timeout); | |
719 | else | |
720 | del_timer_sync(&nbd->timeout_timer); | |
721 | ||
7fdfd406 | 722 | return 0; |
1a2ad211 | 723 | |
2f012508 PC |
724 | case NBD_SET_FLAGS: |
725 | nbd->flags = arg; | |
726 | return 0; | |
727 | ||
1da177e4 | 728 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
729 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
730 | bdev->bd_inode->i_size = nbd->bytesize; | |
731 | set_blocksize(bdev, nbd->blksize); | |
732 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 733 | return 0; |
1a2ad211 PM |
734 | |
735 | case NBD_DO_IT: { | |
736 | struct task_struct *thread; | |
e2511578 | 737 | struct socket *sock; |
1a2ad211 PM |
738 | int error; |
739 | ||
6521d39a | 740 | if (nbd->task_recv) |
c91192d6 | 741 | return -EBUSY; |
e2511578 | 742 | if (!nbd->sock) |
1da177e4 | 743 | return -EINVAL; |
1a2ad211 | 744 | |
f4507164 | 745 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 746 | |
a83e814b PB |
747 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
748 | set_device_ro(bdev, true); | |
a336d298 PC |
749 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
750 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
751 | nbd->disk->queue); | |
75f187ab AB |
752 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
753 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
754 | else | |
755 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 756 | |
cad73b27 | 757 | thread = kthread_run(nbd_thread_send, nbd, "%s", |
30d53d9c | 758 | nbd_name(nbd)); |
1a2ad211 | 759 | if (IS_ERR(thread)) { |
f4507164 | 760 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 761 | return PTR_ERR(thread); |
1a2ad211 | 762 | } |
d06df60b | 763 | |
30d53d9c | 764 | nbd_dev_dbg_init(nbd); |
cad73b27 | 765 | error = nbd_thread_recv(nbd); |
30d53d9c | 766 | nbd_dev_dbg_close(nbd); |
48cf6061 | 767 | kthread_stop(thread); |
1a2ad211 | 768 | |
f4507164 | 769 | mutex_lock(&nbd->tx_lock); |
19391830 | 770 | |
36e47bee | 771 | sock_shutdown(nbd); |
e2511578 AV |
772 | sock = nbd->sock; |
773 | nbd->sock = NULL; | |
f4507164 | 774 | nbd_clear_que(nbd); |
3a2d63f8 | 775 | kill_bdev(bdev); |
a336d298 | 776 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 777 | set_device_ro(bdev, false); |
e2511578 AV |
778 | if (sock) |
779 | sockfd_put(sock); | |
75f187ab | 780 | nbd->flags = 0; |
f4507164 | 781 | nbd->bytesize = 0; |
a8cdc308 | 782 | bdev->bd_inode->i_size = 0; |
f4507164 | 783 | set_capacity(nbd->disk, 0); |
d71a6d73 | 784 | if (max_part > 0) |
9dcd1379 | 785 | blkdev_reread_part(bdev); |
c378f70a PC |
786 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
787 | return 0; | |
19391830 | 788 | return error; |
1a2ad211 PM |
789 | } |
790 | ||
1da177e4 | 791 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
792 | /* |
793 | * This is for compatibility only. The queue is always cleared | |
794 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
795 | */ | |
1da177e4 | 796 | return 0; |
1a2ad211 | 797 | |
1da177e4 | 798 | case NBD_PRINT_DEBUG: |
f4507164 | 799 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 800 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
801 | nbd->queue_head.next, nbd->queue_head.prev, |
802 | &nbd->queue_head); | |
1da177e4 LT |
803 | return 0; |
804 | } | |
1a2ad211 PM |
805 | return -ENOTTY; |
806 | } | |
807 | ||
808 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
809 | unsigned int cmd, unsigned long arg) | |
810 | { | |
f4507164 | 811 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
812 | int error; |
813 | ||
814 | if (!capable(CAP_SYS_ADMIN)) | |
815 | return -EPERM; | |
816 | ||
f4507164 | 817 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 818 | |
f4507164 WG |
819 | mutex_lock(&nbd->tx_lock); |
820 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
821 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
822 | |
823 | return error; | |
1da177e4 LT |
824 | } |
825 | ||
83d5cde4 | 826 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
827 | { |
828 | .owner = THIS_MODULE, | |
8a6cfeb6 | 829 | .ioctl = nbd_ioctl, |
263a3df1 | 830 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
831 | }; |
832 | ||
30d53d9c MP |
833 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
834 | ||
835 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
836 | { | |
837 | struct nbd_device *nbd = s->private; | |
838 | ||
839 | if (nbd->task_recv) | |
840 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
841 | if (nbd->task_send) | |
842 | seq_printf(s, "send: %d\n", task_pid_nr(nbd->task_send)); | |
843 | ||
844 | return 0; | |
845 | } | |
846 | ||
847 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
848 | { | |
849 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
850 | } | |
851 | ||
852 | static const struct file_operations nbd_dbg_tasks_ops = { | |
853 | .open = nbd_dbg_tasks_open, | |
854 | .read = seq_read, | |
855 | .llseek = seq_lseek, | |
856 | .release = single_release, | |
857 | }; | |
858 | ||
859 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
860 | { | |
861 | struct nbd_device *nbd = s->private; | |
862 | u32 flags = nbd->flags; | |
863 | ||
864 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
865 | ||
866 | seq_puts(s, "Known flags:\n"); | |
867 | ||
868 | if (flags & NBD_FLAG_HAS_FLAGS) | |
869 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
870 | if (flags & NBD_FLAG_READ_ONLY) | |
871 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
872 | if (flags & NBD_FLAG_SEND_FLUSH) | |
873 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
874 | if (flags & NBD_FLAG_SEND_TRIM) | |
875 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
880 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
881 | { | |
882 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
883 | } | |
884 | ||
885 | static const struct file_operations nbd_dbg_flags_ops = { | |
886 | .open = nbd_dbg_flags_open, | |
887 | .read = seq_read, | |
888 | .llseek = seq_lseek, | |
889 | .release = single_release, | |
890 | }; | |
891 | ||
892 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
893 | { | |
894 | struct dentry *dir; | |
895 | struct dentry *f; | |
896 | ||
897 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
898 | if (IS_ERR_OR_NULL(dir)) { | |
899 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s' (%ld)\n", | |
900 | nbd_name(nbd), PTR_ERR(dir)); | |
901 | return PTR_ERR(dir); | |
902 | } | |
903 | nbd->dbg_dir = dir; | |
904 | ||
905 | f = debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); | |
906 | if (IS_ERR_OR_NULL(f)) { | |
907 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'tasks', %ld\n", | |
908 | PTR_ERR(f)); | |
909 | return PTR_ERR(f); | |
910 | } | |
911 | ||
912 | f = debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize); | |
913 | if (IS_ERR_OR_NULL(f)) { | |
914 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'size_bytes', %ld\n", | |
915 | PTR_ERR(f)); | |
916 | return PTR_ERR(f); | |
917 | } | |
918 | ||
919 | f = debugfs_create_u32("timeout", 0444, dir, &nbd->xmit_timeout); | |
920 | if (IS_ERR_OR_NULL(f)) { | |
921 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'timeout', %ld\n", | |
922 | PTR_ERR(f)); | |
923 | return PTR_ERR(f); | |
924 | } | |
925 | ||
926 | f = debugfs_create_u32("blocksize", 0444, dir, &nbd->blksize); | |
927 | if (IS_ERR_OR_NULL(f)) { | |
928 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'blocksize', %ld\n", | |
929 | PTR_ERR(f)); | |
930 | return PTR_ERR(f); | |
931 | } | |
932 | ||
933 | f = debugfs_create_file("flags", 0444, dir, &nbd, &nbd_dbg_flags_ops); | |
934 | if (IS_ERR_OR_NULL(f)) { | |
935 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'flags', %ld\n", | |
936 | PTR_ERR(f)); | |
937 | return PTR_ERR(f); | |
938 | } | |
939 | ||
940 | return 0; | |
941 | } | |
942 | ||
943 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
944 | { | |
945 | debugfs_remove_recursive(nbd->dbg_dir); | |
946 | } | |
947 | ||
948 | static int nbd_dbg_init(void) | |
949 | { | |
950 | struct dentry *dbg_dir; | |
951 | ||
952 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
953 | if (IS_ERR(dbg_dir)) | |
954 | return PTR_ERR(dbg_dir); | |
955 | ||
956 | nbd_dbg_dir = dbg_dir; | |
957 | ||
958 | return 0; | |
959 | } | |
960 | ||
961 | static void nbd_dbg_close(void) | |
962 | { | |
963 | debugfs_remove_recursive(nbd_dbg_dir); | |
964 | } | |
965 | ||
966 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
967 | ||
968 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
969 | { | |
970 | return 0; | |
971 | } | |
972 | ||
973 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
974 | { | |
975 | } | |
976 | ||
977 | static int nbd_dbg_init(void) | |
978 | { | |
979 | return 0; | |
980 | } | |
981 | ||
982 | static void nbd_dbg_close(void) | |
983 | { | |
984 | } | |
985 | ||
986 | #endif | |
987 | ||
1da177e4 LT |
988 | /* |
989 | * And here should be modules and kernel interface | |
990 | * (Just smiley confuses emacs :-) | |
991 | */ | |
992 | ||
993 | static int __init nbd_init(void) | |
994 | { | |
995 | int err = -ENOMEM; | |
996 | int i; | |
d71a6d73 | 997 | int part_shift; |
1da177e4 | 998 | |
5b7b18cc | 999 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 1000 | |
d71a6d73 | 1001 | if (max_part < 0) { |
7742ce4a | 1002 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
1003 | return -EINVAL; |
1004 | } | |
1005 | ||
1006 | part_shift = 0; | |
5988ce23 | 1007 | if (max_part > 0) { |
d71a6d73 LV |
1008 | part_shift = fls(max_part); |
1009 | ||
5988ce23 NK |
1010 | /* |
1011 | * Adjust max_part according to part_shift as it is exported | |
1012 | * to user space so that user can know the max number of | |
1013 | * partition kernel should be able to manage. | |
1014 | * | |
1015 | * Note that -1 is required because partition 0 is reserved | |
1016 | * for the whole disk. | |
1017 | */ | |
1018 | max_part = (1UL << part_shift) - 1; | |
1019 | } | |
1020 | ||
3b271082 NK |
1021 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1022 | return -EINVAL; | |
1023 | ||
1024 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
1025 | return -EINVAL; | |
1026 | ||
ff6b8090 SM |
1027 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
1028 | if (!nbd_dev) | |
1029 | return -ENOMEM; | |
1030 | ||
40be0c28 | 1031 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 1032 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
1033 | if (!disk) |
1034 | goto out; | |
1035 | nbd_dev[i].disk = disk; | |
1036 | /* | |
1037 | * The new linux 2.5 block layer implementation requires | |
1038 | * every gendisk to have its very own request_queue struct. | |
1039 | * These structs are big so we dynamically allocate them. | |
1040 | */ | |
cad73b27 | 1041 | disk->queue = blk_init_queue(nbd_request_handler, &nbd_lock); |
1da177e4 LT |
1042 | if (!disk->queue) { |
1043 | put_disk(disk); | |
1044 | goto out; | |
1045 | } | |
31dcfab0 JA |
1046 | /* |
1047 | * Tell the block layer that we are not a rotational device | |
1048 | */ | |
1049 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 1050 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 | 1051 | disk->queue->limits.discard_granularity = 512; |
2bb4cd5c | 1052 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); |
a336d298 | 1053 | disk->queue->limits.discard_zeroes_data = 0; |
078be02b MB |
1054 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1055 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
1056 | } |
1057 | ||
1058 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
1059 | err = -EIO; | |
1060 | goto out; | |
1061 | } | |
1062 | ||
1063 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 1064 | |
30d53d9c MP |
1065 | nbd_dbg_init(); |
1066 | ||
40be0c28 | 1067 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1068 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 1069 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 1070 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 | 1071 | spin_lock_init(&nbd_dev[i].queue_lock); |
dcc909d9 | 1072 | spin_lock_init(&nbd_dev[i].tasks_lock); |
1da177e4 | 1073 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); |
82d4dc5a | 1074 | mutex_init(&nbd_dev[i].tx_lock); |
7e2893a1 MP |
1075 | init_timer(&nbd_dev[i].timeout_timer); |
1076 | nbd_dev[i].timeout_timer.function = nbd_xmit_timeout; | |
1077 | nbd_dev[i].timeout_timer.data = (unsigned long)&nbd_dev[i]; | |
4b2f0260 | 1078 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 1079 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 1080 | nbd_dev[i].blksize = 1024; |
4b86a872 | 1081 | nbd_dev[i].bytesize = 0; |
1da177e4 | 1082 | disk->major = NBD_MAJOR; |
d71a6d73 | 1083 | disk->first_minor = i << part_shift; |
1da177e4 LT |
1084 | disk->fops = &nbd_fops; |
1085 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 1086 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 1087 | set_capacity(disk, 0); |
1da177e4 LT |
1088 | add_disk(disk); |
1089 | } | |
1090 | ||
1091 | return 0; | |
1092 | out: | |
1093 | while (i--) { | |
1094 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
1095 | put_disk(nbd_dev[i].disk); | |
1096 | } | |
f3944d61 | 1097 | kfree(nbd_dev); |
1da177e4 LT |
1098 | return err; |
1099 | } | |
1100 | ||
1101 | static void __exit nbd_cleanup(void) | |
1102 | { | |
1103 | int i; | |
30d53d9c MP |
1104 | |
1105 | nbd_dbg_close(); | |
1106 | ||
40be0c28 | 1107 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1108 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 1109 | nbd_dev[i].magic = 0; |
1da177e4 LT |
1110 | if (disk) { |
1111 | del_gendisk(disk); | |
1112 | blk_cleanup_queue(disk->queue); | |
1113 | put_disk(disk); | |
1114 | } | |
1115 | } | |
1da177e4 | 1116 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 1117 | kfree(nbd_dev); |
1da177e4 LT |
1118 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
1119 | } | |
1120 | ||
1121 | module_init(nbd_init); | |
1122 | module_exit(nbd_cleanup); | |
1123 | ||
1124 | MODULE_DESCRIPTION("Network Block Device"); | |
1125 | MODULE_LICENSE("GPL"); | |
1126 | ||
40be0c28 | 1127 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
1128 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
1129 | module_param(max_part, int, 0444); | |
1130 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |