blob: 061427a75d375a5ed0655bdac8422e668a252a5e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Pavel Macheka2531292010-07-18 14:27:13 +02007 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
9 *
Pavel Machekdbf492d2006-06-25 05:47:42 -070010 * This file is released under GPLv2 or later.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
Pavel Machekdbf492d2006-06-25 05:47:42 -070012 * (part of code stolen from loop.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 */
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>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020027#include <linux/mutex.h>
Herbert Xu4b2f0262006-01-06 00:09:47 -080028#include <linux/compiler.h>
29#include <linux/err.h>
30#include <linux/kernel.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090031#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <net/sock.h>
Trond Myklebust91cf45f2007-11-12 18:10:39 -080033#include <linux/net.h>
Laurent Vivier48cf6062008-04-29 01:02:46 -070034#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/uaccess.h>
37#include <asm/types.h>
38
39#include <linux/nbd.h>
40
Wanlong Gaof4507162012-03-28 14:42:51 -070041#define NBD_MAGIC 0x68797548
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#ifdef NDEBUG
44#define dprintk(flags, fmt...)
45#else /* NDEBUG */
46#define dprintk(flags, fmt...) do { \
47 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
48} while (0)
49#define DBG_IOCTL 0x0004
50#define DBG_INIT 0x0010
51#define DBG_EXIT 0x0020
52#define DBG_BLKDEV 0x0100
53#define DBG_RX 0x0200
54#define DBG_TX 0x0400
55static unsigned int debugflags;
56#endif /* NDEBUG */
57
Ingo van Lil9c7a4162006-07-01 04:36:36 -070058static unsigned int nbds_max = 16;
Paul Clements20a81432008-02-08 04:21:51 -080059static struct nbd_device *nbd_dev;
Laurent Vivierd71a6d72008-04-29 01:02:51 -070060static int max_part;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62/*
63 * Use just one lock (or at most 1 per NIC). Two arguments for this:
64 * 1. Each NIC is essentially a synchronization point for all servers
65 * accessed through that NIC so there's no need to have more locks
66 * than NICs anyway.
67 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
68 * down each lock to the point where they're actually slower than just
69 * a single lock.
70 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
71 */
72static DEFINE_SPINLOCK(nbd_lock);
73
74#ifndef NDEBUG
75static const char *ioctl_cmd_to_ascii(int cmd)
76{
77 switch (cmd) {
78 case NBD_SET_SOCK: return "set-sock";
79 case NBD_SET_BLKSIZE: return "set-blksize";
80 case NBD_SET_SIZE: return "set-size";
81 case NBD_DO_IT: return "do-it";
82 case NBD_CLEAR_SOCK: return "clear-sock";
83 case NBD_CLEAR_QUE: return "clear-que";
84 case NBD_PRINT_DEBUG: return "print-debug";
85 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
86 case NBD_DISCONNECT: return "disconnect";
87 case BLKROSET: return "set-read-only";
88 case BLKFLSBUF: return "flush-buffer-cache";
89 }
90 return "unknown";
91}
92
93static const char *nbdcmd_to_ascii(int cmd)
94{
95 switch (cmd) {
96 case NBD_CMD_READ: return "read";
97 case NBD_CMD_WRITE: return "write";
98 case NBD_CMD_DISC: return "disconnect";
99 }
100 return "invalid";
101}
102#endif /* NDEBUG */
103
104static void nbd_end_request(struct request *req)
105{
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500106 int error = req->errors ? -EIO : 0;
Jens Axboe165125e2007-07-24 09:28:11 +0200107 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 unsigned long flags;
109
110 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500111 req, error ? "failed" : "done");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo1011c1b2009-05-07 22:24:45 +0900114 __blk_end_request_all(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 spin_unlock_irqrestore(q->queue_lock, flags);
116}
117
Wanlong Gaof4507162012-03-28 14:42:51 -0700118static void sock_shutdown(struct nbd_device *nbd, int lock)
Paul Clements7fdfd402007-10-16 23:27:37 -0700119{
120 /* Forcibly shutdown the socket causing all listeners
121 * to error
122 *
123 * FIXME: This code is duplicated from sys_shutdown, but
124 * there should be a more generic interface rather than
125 * calling socket ops directly here */
126 if (lock)
Wanlong Gaof4507162012-03-28 14:42:51 -0700127 mutex_lock(&nbd->tx_lock);
128 if (nbd->sock) {
129 dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n");
130 kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
131 nbd->sock = NULL;
Paul Clements7fdfd402007-10-16 23:27:37 -0700132 }
133 if (lock)
Wanlong Gaof4507162012-03-28 14:42:51 -0700134 mutex_unlock(&nbd->tx_lock);
Paul Clements7fdfd402007-10-16 23:27:37 -0700135}
136
137static void nbd_xmit_timeout(unsigned long arg)
138{
139 struct task_struct *task = (struct task_struct *)arg;
140
141 printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
142 task->comm, task->pid);
143 force_sig(SIGKILL, task);
144}
145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146/*
147 * Send or receive packet.
148 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700149static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 int msg_flags)
151{
Wanlong Gaof4507162012-03-28 14:42:51 -0700152 struct socket *sock = nbd->sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 int result;
154 struct msghdr msg;
155 struct kvec iov;
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700156 sigset_t blocked, oldset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Mike Snitzerffc41cf2008-04-02 13:04:47 -0700158 if (unlikely(!sock)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700159 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200160 "Attempted %s on closed socket in sock_xmit\n",
161 (send ? "send" : "recv"));
Mike Snitzerffc41cf2008-04-02 13:04:47 -0700162 return -EINVAL;
163 }
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 /* Allow interception of SIGKILL only
166 * Don't allow other signals to interrupt the transmission */
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700167 siginitsetinv(&blocked, sigmask(SIGKILL));
168 sigprocmask(SIG_SETMASK, &blocked, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170 do {
171 sock->sk->sk_allocation = GFP_NOIO;
172 iov.iov_base = buf;
173 iov.iov_len = size;
174 msg.msg_name = NULL;
175 msg.msg_namelen = 0;
176 msg.msg_control = NULL;
177 msg.msg_controllen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
179
Paul Clements7fdfd402007-10-16 23:27:37 -0700180 if (send) {
181 struct timer_list ti;
182
Wanlong Gaof4507162012-03-28 14:42:51 -0700183 if (nbd->xmit_timeout) {
Paul Clements7fdfd402007-10-16 23:27:37 -0700184 init_timer(&ti);
185 ti.function = nbd_xmit_timeout;
186 ti.data = (unsigned long)current;
Wanlong Gaof4507162012-03-28 14:42:51 -0700187 ti.expires = jiffies + nbd->xmit_timeout;
Paul Clements7fdfd402007-10-16 23:27:37 -0700188 add_timer(&ti);
189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
Wanlong Gaof4507162012-03-28 14:42:51 -0700191 if (nbd->xmit_timeout)
Paul Clements7fdfd402007-10-16 23:27:37 -0700192 del_timer_sync(&ti);
193 } else
Namhyung Kim35fbf5b2011-05-28 14:44:46 +0200194 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
195 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 if (signal_pending(current)) {
198 siginfo_t info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700200 task_pid_nr(current), current->comm,
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700201 dequeue_signal_lock(current, &current->blocked, &info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 result = -EINTR;
Wanlong Gaof4507162012-03-28 14:42:51 -0700203 sock_shutdown(nbd, !send);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 break;
205 }
206
207 if (result <= 0) {
208 if (result == 0)
209 result = -EPIPE; /* short read */
210 break;
211 }
212 size -= result;
213 buf += result;
214 } while (size > 0);
215
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700216 sigprocmask(SIG_SETMASK, &oldset, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 return result;
219}
220
Wanlong Gaof4507162012-03-28 14:42:51 -0700221static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 int flags)
223{
224 int result;
225 void *kaddr = kmap(bvec->bv_page);
Wanlong Gaof4507162012-03-28 14:42:51 -0700226 result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset,
227 bvec->bv_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 kunmap(bvec->bv_page);
229 return result;
230}
231
Paul Clements7fdfd402007-10-16 23:27:37 -0700232/* always call with the tx_lock held */
Wanlong Gaof4507162012-03-28 14:42:51 -0700233static int nbd_send_req(struct nbd_device *nbd, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
NeilBrown5705f702007-09-25 12:35:59 +0200235 int result, flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 struct nbd_request request;
Tejun Heo1011c1b2009-05-07 22:24:45 +0900237 unsigned long size = blk_rq_bytes(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 request.magic = htonl(NBD_REQUEST_MAGIC);
240 request.type = htonl(nbd_cmd(req));
Tejun Heo83096eb2009-05-07 22:24:39 +0900241 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 request.len = htonl(size);
243 memcpy(request.handle, &req, sizeof(req));
244
Tejun Heo83096eb2009-05-07 22:24:39 +0900245 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700246 nbd->disk->disk_name, req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 nbdcmd_to_ascii(nbd_cmd(req)),
Tejun Heo83096eb2009-05-07 22:24:39 +0900248 (unsigned long long)blk_rq_pos(req) << 9,
Tejun Heo1011c1b2009-05-07 22:24:45 +0900249 blk_rq_bytes(req));
Wanlong Gaof4507162012-03-28 14:42:51 -0700250 result = sock_xmit(nbd, 1, &request, sizeof(request),
Paul Clements7fdfd402007-10-16 23:27:37 -0700251 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700253 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200254 "Send control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 goto error_out;
256 }
257
258 if (nbd_cmd(req) == NBD_CMD_WRITE) {
NeilBrown5705f702007-09-25 12:35:59 +0200259 struct req_iterator iter;
260 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 /*
262 * we are really probing at internals to determine
263 * whether to set MSG_MORE or not...
264 */
NeilBrown5705f702007-09-25 12:35:59 +0200265 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200266 flags = 0;
267 if (!rq_iter_last(req, iter))
268 flags = MSG_MORE;
269 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700270 nbd->disk->disk_name, req, bvec->bv_len);
271 result = sock_send_bvec(nbd, bvec, flags);
Jens Axboe6c92e692007-08-16 13:43:12 +0200272 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700273 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200274 "Send data failed (result %d)\n",
275 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200276 goto error_out;
277 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 }
279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return 0;
281
282error_out:
Pavel Machek15746fc2009-04-02 16:58:42 -0700283 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
Wanlong Gaof4507162012-03-28 14:42:51 -0700286static struct request *nbd_find_request(struct nbd_device *nbd,
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700287 struct request *xreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
Denis Chengd2c97402007-10-16 23:26:14 -0700289 struct request *req, *tmp;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800290 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Wanlong Gaof4507162012-03-28 14:42:51 -0700292 err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800293 if (unlikely(err))
294 goto out;
295
Wanlong Gaof4507162012-03-28 14:42:51 -0700296 spin_lock(&nbd->queue_lock);
297 list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 if (req != xreq)
299 continue;
300 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700301 spin_unlock(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 return req;
303 }
Wanlong Gaof4507162012-03-28 14:42:51 -0700304 spin_unlock(&nbd->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800305
306 err = -ENOENT;
307
308out:
309 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Wanlong Gaof4507162012-03-28 14:42:51 -0700312static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
314 int result;
315 void *kaddr = kmap(bvec->bv_page);
Wanlong Gaof4507162012-03-28 14:42:51 -0700316 result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 MSG_WAITALL);
318 kunmap(bvec->bv_page);
319 return result;
320}
321
322/* NULL returned = something went wrong, inform userspace */
Wanlong Gaof4507162012-03-28 14:42:51 -0700323static struct request *nbd_read_stat(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
325 int result;
326 struct nbd_reply reply;
327 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 reply.magic = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700330 result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700332 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200333 "Receive control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 goto harderror;
335 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700336
337 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700338 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
Michal Feixe4b57e02006-07-30 03:03:31 -0700339 (unsigned long)ntohl(reply.magic));
340 result = -EPROTO;
341 goto harderror;
342 }
343
Wanlong Gaof4507162012-03-28 14:42:51 -0700344 req = nbd_find_request(nbd, *(struct request **)reply.handle);
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -0700345 if (IS_ERR(req)) {
Herbert Xu4b2f0262006-01-06 00:09:47 -0800346 result = PTR_ERR(req);
347 if (result != -ENOENT)
348 goto harderror;
349
Wanlong Gaof4507162012-03-28 14:42:51 -0700350 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200351 reply.handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 result = -EBADR;
353 goto harderror;
354 }
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 if (ntohl(reply.error)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700357 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200358 ntohl(reply.error));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 req->errors++;
360 return req;
361 }
362
363 dprintk(DBG_RX, "%s: request %p: got reply\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700364 nbd->disk->disk_name, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (nbd_cmd(req) == NBD_CMD_READ) {
NeilBrown5705f702007-09-25 12:35:59 +0200366 struct req_iterator iter;
367 struct bio_vec *bvec;
368
369 rq_for_each_segment(bvec, req, iter) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700370 result = sock_recv_bvec(nbd, bvec);
Jens Axboe6c92e692007-08-16 13:43:12 +0200371 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700372 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200373 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200374 req->errors++;
375 return req;
376 }
377 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700378 nbd->disk->disk_name, req, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
380 }
381 return req;
382harderror:
Wanlong Gaof4507162012-03-28 14:42:51 -0700383 nbd->harderror = result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return NULL;
385}
386
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200387static ssize_t pid_show(struct device *dev,
388 struct device_attribute *attr, char *buf)
Paul Clements6b39bb62006-12-06 20:40:53 -0800389{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200390 struct gendisk *disk = dev_to_disk(dev);
391
392 return sprintf(buf, "%ld\n",
Paul Clements6b39bb62006-12-06 20:40:53 -0800393 (long) ((struct nbd_device *)disk->private_data)->pid);
394}
395
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200396static struct device_attribute pid_attr = {
Parag Warudkar01e8ef12008-10-18 20:28:50 -0700397 .attr = { .name = "pid", .mode = S_IRUGO},
Paul Clements6b39bb62006-12-06 20:40:53 -0800398 .show = pid_show,
399};
400
Wanlong Gaof4507162012-03-28 14:42:51 -0700401static int nbd_do_it(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 struct request *req;
WANG Cong84963042007-05-09 02:33:36 -0700404 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Wanlong Gaof4507162012-03-28 14:42:51 -0700406 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Wanlong Gaof4507162012-03-28 14:42:51 -0700408 nbd->pid = task_pid_nr(current);
409 ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
WANG Cong84963042007-05-09 02:33:36 -0700410 if (ret) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700411 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
412 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700413 return ret;
414 }
Paul Clements6b39bb62006-12-06 20:40:53 -0800415
Wanlong Gaof4507162012-03-28 14:42:51 -0700416 while ((req = nbd_read_stat(nbd)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800418
Wanlong Gaof4507162012-03-28 14:42:51 -0700419 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
420 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
Wanlong Gaof4507162012-03-28 14:42:51 -0700424static void nbd_clear_que(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
426 struct request *req;
427
Wanlong Gaof4507162012-03-28 14:42:51 -0700428 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Herbert Xu4b2f0262006-01-06 00:09:47 -0800430 /*
Wanlong Gaof4507162012-03-28 14:42:51 -0700431 * Because we have set nbd->sock to NULL under the tx_lock, all
Herbert Xu4b2f0262006-01-06 00:09:47 -0800432 * modifications to the list must have completed by now. For
433 * the same reason, the active_req must be NULL.
434 *
435 * As a consequence, we don't need to take the spin lock while
436 * purging the list here.
437 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700438 BUG_ON(nbd->sock);
439 BUG_ON(nbd->active_req);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800440
Wanlong Gaof4507162012-03-28 14:42:51 -0700441 while (!list_empty(&nbd->queue_head)) {
442 req = list_entry(nbd->queue_head.next, struct request,
Herbert Xu4b2f0262006-01-06 00:09:47 -0800443 queuelist);
444 list_del_init(&req->queuelist);
445 req->errors++;
446 nbd_end_request(req);
447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Paul Clements7fdfd402007-10-16 23:27:37 -0700450
Wanlong Gaof4507162012-03-28 14:42:51 -0700451static void nbd_handle_req(struct nbd_device *nbd, struct request *req)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700452{
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200453 if (req->cmd_type != REQ_TYPE_FS)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700454 goto error_out;
455
456 nbd_cmd(req) = NBD_CMD_READ;
457 if (rq_data_dir(req) == WRITE) {
458 nbd_cmd(req) = NBD_CMD_WRITE;
Wanlong Gaof4507162012-03-28 14:42:51 -0700459 if (nbd->flags & NBD_READ_ONLY) {
460 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200461 "Write on read-only\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700462 goto error_out;
463 }
464 }
465
466 req->errors = 0;
467
Wanlong Gaof4507162012-03-28 14:42:51 -0700468 mutex_lock(&nbd->tx_lock);
469 if (unlikely(!nbd->sock)) {
470 mutex_unlock(&nbd->tx_lock);
471 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200472 "Attempted send on closed socket\n");
Pavel Machek15746fc2009-04-02 16:58:42 -0700473 goto error_out;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700474 }
475
Wanlong Gaof4507162012-03-28 14:42:51 -0700476 nbd->active_req = req;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700477
Wanlong Gaof4507162012-03-28 14:42:51 -0700478 if (nbd_send_req(nbd, req) != 0) {
479 dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700480 req->errors++;
481 nbd_end_request(req);
482 } else {
Wanlong Gaof4507162012-03-28 14:42:51 -0700483 spin_lock(&nbd->queue_lock);
484 list_add(&req->queuelist, &nbd->queue_head);
485 spin_unlock(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700486 }
487
Wanlong Gaof4507162012-03-28 14:42:51 -0700488 nbd->active_req = NULL;
489 mutex_unlock(&nbd->tx_lock);
490 wake_up_all(&nbd->active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700491
492 return;
493
494error_out:
495 req->errors++;
496 nbd_end_request(req);
497}
498
499static int nbd_thread(void *data)
500{
Wanlong Gaof4507162012-03-28 14:42:51 -0700501 struct nbd_device *nbd = data;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700502 struct request *req;
503
504 set_user_nice(current, -20);
Wanlong Gaof4507162012-03-28 14:42:51 -0700505 while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) {
Laurent Vivier48cf6062008-04-29 01:02:46 -0700506 /* wait for something to do */
Wanlong Gaof4507162012-03-28 14:42:51 -0700507 wait_event_interruptible(nbd->waiting_wq,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700508 kthread_should_stop() ||
Wanlong Gaof4507162012-03-28 14:42:51 -0700509 !list_empty(&nbd->waiting_queue));
Laurent Vivier48cf6062008-04-29 01:02:46 -0700510
511 /* extract request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700512 if (list_empty(&nbd->waiting_queue))
Laurent Vivier48cf6062008-04-29 01:02:46 -0700513 continue;
514
Wanlong Gaof4507162012-03-28 14:42:51 -0700515 spin_lock_irq(&nbd->queue_lock);
516 req = list_entry(nbd->waiting_queue.next, struct request,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700517 queuelist);
518 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700519 spin_unlock_irq(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700520
521 /* handle request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700522 nbd_handle_req(nbd, req);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700523 }
524 return 0;
525}
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527/*
528 * We always wait for result of write, for now. It would be nice to make it optional
529 * in future
Wanlong Gaof4507162012-03-28 14:42:51 -0700530 * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
532 */
533
Pavel Machek15746fc2009-04-02 16:58:42 -0700534static void do_nbd_request(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535{
536 struct request *req;
537
Tejun Heo9934c8c2009-05-08 11:54:16 +0900538 while ((req = blk_fetch_request(q)) != NULL) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700539 struct nbd_device *nbd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Laurent Vivier48cf6062008-04-29 01:02:46 -0700541 spin_unlock_irq(q->queue_lock);
542
Jens Axboe4aff5e22006-08-10 08:44:47 +0200543 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
544 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Wanlong Gaof4507162012-03-28 14:42:51 -0700546 nbd = req->rq_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Wanlong Gaof4507162012-03-28 14:42:51 -0700548 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Wanlong Gaof4507162012-03-28 14:42:51 -0700550 if (unlikely(!nbd->sock)) {
551 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200552 "Attempted send on closed socket\n");
Paul Clements4d48a542009-02-11 13:04:45 -0800553 req->errors++;
554 nbd_end_request(req);
555 spin_lock_irq(q->queue_lock);
556 continue;
557 }
558
Wanlong Gaof4507162012-03-28 14:42:51 -0700559 spin_lock_irq(&nbd->queue_lock);
560 list_add_tail(&req->queuelist, &nbd->waiting_queue);
561 spin_unlock_irq(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Wanlong Gaof4507162012-03-28 14:42:51 -0700563 wake_up(&nbd->waiting_wq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
Pavel Machek1a2ad212009-04-02 16:58:41 -0700569/* Must be called with tx_lock held */
570
Wanlong Gaof4507162012-03-28 14:42:51 -0700571static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
Pavel Machek1a2ad212009-04-02 16:58:41 -0700572 unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 switch (cmd) {
Pavel Machek1a2ad212009-04-02 16:58:41 -0700575 case NBD_DISCONNECT: {
576 struct request sreq;
577
Wanlong Gaof4507162012-03-28 14:42:51 -0700578 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
Pavel Machek1a2ad212009-04-02 16:58:41 -0700579
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +0200580 blk_rq_init(NULL, &sreq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200581 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 nbd_cmd(&sreq) = NBD_CMD_DISC;
Wanlong Gaof4507162012-03-28 14:42:51 -0700583 if (!nbd->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return -EINVAL;
Wanlong Gaof4507162012-03-28 14:42:51 -0700585 nbd_send_req(nbd, &sreq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Pavel Machek1a2ad212009-04-02 16:58:41 -0700589 case NBD_CLEAR_SOCK: {
590 struct file *file;
591
Wanlong Gaof4507162012-03-28 14:42:51 -0700592 nbd->sock = NULL;
593 file = nbd->file;
594 nbd->file = NULL;
595 nbd_clear_que(nbd);
596 BUG_ON(!list_empty(&nbd->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 if (file)
598 fput(file);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700599 return 0;
600 }
601
602 case NBD_SET_SOCK: {
603 struct file *file;
Wanlong Gaof4507162012-03-28 14:42:51 -0700604 if (nbd->file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 file = fget(arg);
607 if (file) {
Al Viroa8cdc302008-03-02 09:33:33 -0500608 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (S_ISSOCK(inode->i_mode)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700610 nbd->file = file;
611 nbd->sock = SOCKET_I(inode);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700612 if (max_part > 0)
613 bdev->bd_invalidated = 1;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700614 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 } else {
616 fput(file);
617 }
618 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700619 return -EINVAL;
620 }
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 case NBD_SET_BLKSIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700623 nbd->blksize = arg;
624 nbd->bytesize &= ~(nbd->blksize-1);
625 bdev->bd_inode->i_size = nbd->bytesize;
626 set_blocksize(bdev, nbd->blksize);
627 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 case NBD_SET_SIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700631 nbd->bytesize = arg & ~(nbd->blksize-1);
632 bdev->bd_inode->i_size = nbd->bytesize;
633 set_blocksize(bdev, nbd->blksize);
634 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700636
Paul Clements7fdfd402007-10-16 23:27:37 -0700637 case NBD_SET_TIMEOUT:
Wanlong Gaof4507162012-03-28 14:42:51 -0700638 nbd->xmit_timeout = arg * HZ;
Paul Clements7fdfd402007-10-16 23:27:37 -0700639 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 case NBD_SET_SIZE_BLOCKS:
Wanlong Gaof4507162012-03-28 14:42:51 -0700642 nbd->bytesize = ((u64) arg) * nbd->blksize;
643 bdev->bd_inode->i_size = nbd->bytesize;
644 set_blocksize(bdev, nbd->blksize);
645 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700647
648 case NBD_DO_IT: {
649 struct task_struct *thread;
650 struct file *file;
651 int error;
652
Wanlong Gaof4507162012-03-28 14:42:51 -0700653 if (nbd->pid)
Pavel Machekc91192d2009-01-15 13:51:03 -0800654 return -EBUSY;
Wanlong Gaof4507162012-03-28 14:42:51 -0700655 if (!nbd->file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700657
Wanlong Gaof4507162012-03-28 14:42:51 -0700658 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700659
Wanlong Gaof4507162012-03-28 14:42:51 -0700660 thread = kthread_create(nbd_thread, nbd, nbd->disk->disk_name);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700661 if (IS_ERR(thread)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700662 mutex_lock(&nbd->tx_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700663 return PTR_ERR(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700664 }
Laurent Vivier48cf6062008-04-29 01:02:46 -0700665 wake_up_process(thread);
Wanlong Gaof4507162012-03-28 14:42:51 -0700666 error = nbd_do_it(nbd);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700667 kthread_stop(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700668
Wanlong Gaof4507162012-03-28 14:42:51 -0700669 mutex_lock(&nbd->tx_lock);
WANG Cong84963042007-05-09 02:33:36 -0700670 if (error)
671 return error;
Wanlong Gaof4507162012-03-28 14:42:51 -0700672 sock_shutdown(nbd, 0);
673 file = nbd->file;
674 nbd->file = NULL;
675 nbd_clear_que(nbd);
676 dev_warn(disk_to_dev(nbd->disk), "queue cleared\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 if (file)
678 fput(file);
Wanlong Gaof4507162012-03-28 14:42:51 -0700679 nbd->bytesize = 0;
Al Viroa8cdc302008-03-02 09:33:33 -0500680 bdev->bd_inode->i_size = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700681 set_capacity(nbd->disk, 0);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700682 if (max_part > 0)
Al Viroa8cdc302008-03-02 09:33:33 -0500683 ioctl_by_bdev(bdev, BLKRRPART, 0);
Wanlong Gaof4507162012-03-28 14:42:51 -0700684 return nbd->harderror;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700685 }
686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800688 /*
689 * This is for compatibility only. The queue is always cleared
690 * by NBD_DO_IT or NBD_CLEAR_SOCK.
691 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700692 BUG_ON(!nbd->sock && !list_empty(&nbd->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 case NBD_PRINT_DEBUG:
Wanlong Gaof4507162012-03-28 14:42:51 -0700696 dev_info(disk_to_dev(nbd->disk),
WANG Cong5eedf542011-08-19 14:48:28 +0200697 "next = %p, prev = %p, head = %p\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700698 nbd->queue_head.next, nbd->queue_head.prev,
699 &nbd->queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700702 return -ENOTTY;
703}
704
705static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
706 unsigned int cmd, unsigned long arg)
707{
Wanlong Gaof4507162012-03-28 14:42:51 -0700708 struct nbd_device *nbd = bdev->bd_disk->private_data;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700709 int error;
710
711 if (!capable(CAP_SYS_ADMIN))
712 return -EPERM;
713
Wanlong Gaof4507162012-03-28 14:42:51 -0700714 BUG_ON(nbd->magic != NBD_MAGIC);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700715
716 /* Anyone capable of this syscall can do *real bad* things */
717 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700718 nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700719
Wanlong Gaof4507162012-03-28 14:42:51 -0700720 mutex_lock(&nbd->tx_lock);
721 error = __nbd_ioctl(bdev, nbd, cmd, arg);
722 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700723
724 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700727static const struct block_device_operations nbd_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 .owner = THIS_MODULE,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +0200730 .ioctl = nbd_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731};
732
733/*
734 * And here should be modules and kernel interface
735 * (Just smiley confuses emacs :-)
736 */
737
738static int __init nbd_init(void)
739{
740 int err = -ENOMEM;
741 int i;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700742 int part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800744 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700746 if (max_part < 0) {
WANG Cong7742ce42011-08-19 14:48:28 +0200747 printk(KERN_ERR "nbd: max_part must be >= 0\n");
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700748 return -EINVAL;
749 }
750
Sven Wegenerf3944d62008-08-20 14:09:07 -0700751 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
752 if (!nbd_dev)
753 return -ENOMEM;
754
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700755 part_shift = 0;
Namhyung Kim5988ce22011-05-28 14:44:46 +0200756 if (max_part > 0) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700757 part_shift = fls(max_part);
758
Namhyung Kim5988ce22011-05-28 14:44:46 +0200759 /*
760 * Adjust max_part according to part_shift as it is exported
761 * to user space so that user can know the max number of
762 * partition kernel should be able to manage.
763 *
764 * Note that -1 is required because partition 0 is reserved
765 * for the whole disk.
766 */
767 max_part = (1UL << part_shift) - 1;
768 }
769
Namhyung Kim3b271082011-05-28 14:44:46 +0200770 if ((1UL << part_shift) > DISK_MAX_PARTS)
771 return -EINVAL;
772
773 if (nbds_max > 1UL << (MINORBITS - part_shift))
774 return -EINVAL;
775
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700776 for (i = 0; i < nbds_max; i++) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700777 struct gendisk *disk = alloc_disk(1 << part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (!disk)
779 goto out;
780 nbd_dev[i].disk = disk;
781 /*
782 * The new linux 2.5 block layer implementation requires
783 * every gendisk to have its very own request_queue struct.
784 * These structs are big so we dynamically allocate them.
785 */
786 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
787 if (!disk->queue) {
788 put_disk(disk);
789 goto out;
790 }
Jens Axboe31dcfab2008-10-31 10:06:37 +0100791 /*
792 * Tell the block layer that we are not a rotational device
793 */
794 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796
797 if (register_blkdev(NBD_MAJOR, "nbd")) {
798 err = -EIO;
799 goto out;
800 }
801
802 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
803 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
804
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700805 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 struct gendisk *disk = nbd_dev[i].disk;
807 nbd_dev[i].file = NULL;
Wanlong Gaof4507162012-03-28 14:42:51 -0700808 nbd_dev[i].magic = NBD_MAGIC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 nbd_dev[i].flags = 0;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700810 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 spin_lock_init(&nbd_dev[i].queue_lock);
812 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800813 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800814 init_waitqueue_head(&nbd_dev[i].active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700815 init_waitqueue_head(&nbd_dev[i].waiting_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 nbd_dev[i].blksize = 1024;
Paul Clements4b86a872007-10-16 23:27:36 -0700817 nbd_dev[i].bytesize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 disk->major = NBD_MAJOR;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700819 disk->first_minor = i << part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 disk->fops = &nbd_fops;
821 disk->private_data = &nbd_dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 sprintf(disk->disk_name, "nbd%d", i);
Paul Clements4b86a872007-10-16 23:27:36 -0700823 set_capacity(disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 add_disk(disk);
825 }
826
827 return 0;
828out:
829 while (i--) {
830 blk_cleanup_queue(nbd_dev[i].disk->queue);
831 put_disk(nbd_dev[i].disk);
832 }
Sven Wegenerf3944d62008-08-20 14:09:07 -0700833 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 return err;
835}
836
837static void __exit nbd_cleanup(void)
838{
839 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700840 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700842 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (disk) {
844 del_gendisk(disk);
845 blk_cleanup_queue(disk->queue);
846 put_disk(disk);
847 }
848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 unregister_blkdev(NBD_MAJOR, "nbd");
Sven Wegenerf3944d62008-08-20 14:09:07 -0700850 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
852}
853
854module_init(nbd_init);
855module_exit(nbd_cleanup);
856
857MODULE_DESCRIPTION("Network Block Device");
858MODULE_LICENSE("GPL");
859
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700860module_param(nbds_max, int, 0444);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700861MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
862module_param(max_part, int, 0444);
863MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864#ifndef NDEBUG
865module_param(debugflags, int, 0644);
866MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
867#endif