blob: f533f3375e24751feaf4e065b920291c7abe149d [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>
Herbert Xu4b2f0262006-01-06 00:09:47 -080037#include <asm/system.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/types.h>
39
40#include <linux/nbd.h>
41
42#define LO_MAGIC 0x68797548
43
44#ifdef NDEBUG
45#define dprintk(flags, fmt...)
46#else /* NDEBUG */
47#define dprintk(flags, fmt...) do { \
48 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
49} while (0)
50#define DBG_IOCTL 0x0004
51#define DBG_INIT 0x0010
52#define DBG_EXIT 0x0020
53#define DBG_BLKDEV 0x0100
54#define DBG_RX 0x0200
55#define DBG_TX 0x0400
56static unsigned int debugflags;
57#endif /* NDEBUG */
58
Ingo van Lil9c7a4162006-07-01 04:36:36 -070059static unsigned int nbds_max = 16;
Paul Clements20a81432008-02-08 04:21:51 -080060static struct nbd_device *nbd_dev;
Laurent Vivierd71a6d72008-04-29 01:02:51 -070061static int max_part;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/*
64 * Use just one lock (or at most 1 per NIC). Two arguments for this:
65 * 1. Each NIC is essentially a synchronization point for all servers
66 * accessed through that NIC so there's no need to have more locks
67 * than NICs anyway.
68 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
69 * down each lock to the point where they're actually slower than just
70 * a single lock.
71 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
72 */
73static DEFINE_SPINLOCK(nbd_lock);
74
75#ifndef NDEBUG
76static const char *ioctl_cmd_to_ascii(int cmd)
77{
78 switch (cmd) {
79 case NBD_SET_SOCK: return "set-sock";
80 case NBD_SET_BLKSIZE: return "set-blksize";
81 case NBD_SET_SIZE: return "set-size";
82 case NBD_DO_IT: return "do-it";
83 case NBD_CLEAR_SOCK: return "clear-sock";
84 case NBD_CLEAR_QUE: return "clear-que";
85 case NBD_PRINT_DEBUG: return "print-debug";
86 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
87 case NBD_DISCONNECT: return "disconnect";
88 case BLKROSET: return "set-read-only";
89 case BLKFLSBUF: return "flush-buffer-cache";
90 }
91 return "unknown";
92}
93
94static const char *nbdcmd_to_ascii(int cmd)
95{
96 switch (cmd) {
97 case NBD_CMD_READ: return "read";
98 case NBD_CMD_WRITE: return "write";
99 case NBD_CMD_DISC: return "disconnect";
100 }
101 return "invalid";
102}
103#endif /* NDEBUG */
104
105static void nbd_end_request(struct request *req)
106{
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500107 int error = req->errors ? -EIO : 0;
Jens Axboe165125e2007-07-24 09:28:11 +0200108 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 unsigned long flags;
110
111 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500112 req, error ? "failed" : "done");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo1011c1b2009-05-07 22:24:45 +0900115 __blk_end_request_all(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 spin_unlock_irqrestore(q->queue_lock, flags);
117}
118
Paul Clements7fdfd402007-10-16 23:27:37 -0700119static void sock_shutdown(struct nbd_device *lo, int lock)
120{
121 /* Forcibly shutdown the socket causing all listeners
122 * to error
123 *
124 * FIXME: This code is duplicated from sys_shutdown, but
125 * there should be a more generic interface rather than
126 * calling socket ops directly here */
127 if (lock)
128 mutex_lock(&lo->tx_lock);
129 if (lo->sock) {
130 printk(KERN_WARNING "%s: shutting down socket\n",
131 lo->disk->disk_name);
Trond Myklebust91cf45f2007-11-12 18:10:39 -0800132 kernel_sock_shutdown(lo->sock, SHUT_RDWR);
Paul Clements7fdfd402007-10-16 23:27:37 -0700133 lo->sock = NULL;
134 }
135 if (lock)
136 mutex_unlock(&lo->tx_lock);
137}
138
139static void nbd_xmit_timeout(unsigned long arg)
140{
141 struct task_struct *task = (struct task_struct *)arg;
142
143 printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
144 task->comm, task->pid);
145 force_sig(SIGKILL, task);
146}
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148/*
149 * Send or receive packet.
150 */
Paul Clements7fdfd402007-10-16 23:27:37 -0700151static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 int msg_flags)
153{
Paul Clements7fdfd402007-10-16 23:27:37 -0700154 struct socket *sock = lo->sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 int result;
156 struct msghdr msg;
157 struct kvec iov;
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700158 sigset_t blocked, oldset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Mike Snitzerffc41cf2008-04-02 13:04:47 -0700160 if (unlikely(!sock)) {
161 printk(KERN_ERR "%s: Attempted %s on closed socket in sock_xmit\n",
162 lo->disk->disk_name, (send ? "send" : "recv"));
163 return -EINVAL;
164 }
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 /* Allow interception of SIGKILL only
167 * Don't allow other signals to interrupt the transmission */
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700168 siginitsetinv(&blocked, sigmask(SIGKILL));
169 sigprocmask(SIG_SETMASK, &blocked, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 do {
172 sock->sk->sk_allocation = GFP_NOIO;
173 iov.iov_base = buf;
174 iov.iov_len = size;
175 msg.msg_name = NULL;
176 msg.msg_namelen = 0;
177 msg.msg_control = NULL;
178 msg.msg_controllen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
180
Paul Clements7fdfd402007-10-16 23:27:37 -0700181 if (send) {
182 struct timer_list ti;
183
184 if (lo->xmit_timeout) {
185 init_timer(&ti);
186 ti.function = nbd_xmit_timeout;
187 ti.data = (unsigned long)current;
188 ti.expires = jiffies + lo->xmit_timeout;
189 add_timer(&ti);
190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
Paul Clements7fdfd402007-10-16 23:27:37 -0700192 if (lo->xmit_timeout)
193 del_timer_sync(&ti);
194 } else
Namhyung Kim35fbf5b2011-05-28 14:44:46 +0200195 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
196 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 if (signal_pending(current)) {
199 siginfo_t info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700201 task_pid_nr(current), current->comm,
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700202 dequeue_signal_lock(current, &current->blocked, &info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 result = -EINTR;
Paul Clements7fdfd402007-10-16 23:27:37 -0700204 sock_shutdown(lo, !send);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 break;
206 }
207
208 if (result <= 0) {
209 if (result == 0)
210 result = -EPIPE; /* short read */
211 break;
212 }
213 size -= result;
214 buf += result;
215 } while (size > 0);
216
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700217 sigprocmask(SIG_SETMASK, &oldset, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 return result;
220}
221
Paul Clements7fdfd402007-10-16 23:27:37 -0700222static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 int flags)
224{
225 int result;
226 void *kaddr = kmap(bvec->bv_page);
Paul Clements7fdfd402007-10-16 23:27:37 -0700227 result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233static int nbd_send_req(struct nbd_device *lo, struct request *req)
234{
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",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 lo->disk->disk_name, req,
247 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));
Paul Clements7fdfd402007-10-16 23:27:37 -0700250 result = sock_xmit(lo, 1, &request, sizeof(request),
251 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (result <= 0) {
253 printk(KERN_ERR "%s: Send control failed (result %d)\n",
254 lo->disk->disk_name, result);
255 goto error_out;
256 }
257
258 if (nbd_cmd(req) == NBD_CMD_WRITE) {
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",
270 lo->disk->disk_name, req, bvec->bv_len);
Paul Clements7fdfd402007-10-16 23:27:37 -0700271 result = sock_send_bvec(lo, bvec, flags);
Jens Axboe6c92e692007-08-16 13:43:12 +0200272 if (result <= 0) {
273 printk(KERN_ERR "%s: Send data failed (result %d)\n",
274 lo->disk->disk_name, result);
275 goto error_out;
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 }
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return 0;
280
281error_out:
Pavel Machek15746fc2009-04-02 16:58:42 -0700282 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}
284
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700285static struct request *nbd_find_request(struct nbd_device *lo,
286 struct request *xreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Denis Chengd2c97402007-10-16 23:26:14 -0700288 struct request *req, *tmp;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800289 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Herbert Xu4b2f0262006-01-06 00:09:47 -0800291 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
292 if (unlikely(err))
293 goto out;
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 spin_lock(&lo->queue_lock);
Denis Chengd2c97402007-10-16 23:26:14 -0700296 list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 if (req != xreq)
298 continue;
299 list_del_init(&req->queuelist);
300 spin_unlock(&lo->queue_lock);
301 return req;
302 }
303 spin_unlock(&lo->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800304
305 err = -ENOENT;
306
307out:
308 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Paul Clements7fdfd402007-10-16 23:27:37 -0700311static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 int result;
314 void *kaddr = kmap(bvec->bv_page);
Paul Clements7fdfd402007-10-16 23:27:37 -0700315 result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 MSG_WAITALL);
317 kunmap(bvec->bv_page);
318 return result;
319}
320
321/* NULL returned = something went wrong, inform userspace */
322static struct request *nbd_read_stat(struct nbd_device *lo)
323{
324 int result;
325 struct nbd_reply reply;
326 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 reply.magic = 0;
Paul Clements7fdfd402007-10-16 23:27:37 -0700329 result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 if (result <= 0) {
331 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
332 lo->disk->disk_name, result);
333 goto harderror;
334 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700335
336 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
337 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
338 lo->disk->disk_name,
339 (unsigned long)ntohl(reply.magic));
340 result = -EPROTO;
341 goto harderror;
342 }
343
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700344 req = nbd_find_request(lo, *(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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
351 lo->disk->disk_name, reply.handle);
352 result = -EBADR;
353 goto harderror;
354 }
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 if (ntohl(reply.error)) {
357 printk(KERN_ERR "%s: Other side returned error (%d)\n",
358 lo->disk->disk_name, ntohl(reply.error));
359 req->errors++;
360 return req;
361 }
362
363 dprintk(DBG_RX, "%s: request %p: got reply\n",
364 lo->disk->disk_name, req);
365 if (nbd_cmd(req) == NBD_CMD_READ) {
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) {
Paul Clements7fdfd402007-10-16 23:27:37 -0700370 result = sock_recv_bvec(lo, bvec);
Jens Axboe6c92e692007-08-16 13:43:12 +0200371 if (result <= 0) {
372 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
373 lo->disk->disk_name, result);
374 req->errors++;
375 return req;
376 }
377 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
378 lo->disk->disk_name, req, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
380 }
381 return req;
382harderror:
383 lo->harderror = result;
384 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
WANG Cong84963042007-05-09 02:33:36 -0700401static int nbd_do_it(struct nbd_device *lo)
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
406 BUG_ON(lo->magic != LO_MAGIC);
407
Paul Clements6b39bb62006-12-06 20:40:53 -0800408 lo->pid = current->pid;
Tejun Heoed9e1982008-08-25 19:56:05 +0900409 ret = sysfs_create_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr);
WANG Cong84963042007-05-09 02:33:36 -0700410 if (ret) {
411 printk(KERN_ERR "nbd: sysfs_create_file failed!");
Pavel Machekc91192d2009-01-15 13:51:03 -0800412 lo->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700413 return ret;
414 }
Paul Clements6b39bb62006-12-06 20:40:53 -0800415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 while ((req = nbd_read_stat(lo)) != NULL)
417 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800418
Tejun Heoed9e1982008-08-25 19:56:05 +0900419 sysfs_remove_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr);
Pavel Machekc91192d2009-01-15 13:51:03 -0800420 lo->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
424static void nbd_clear_que(struct nbd_device *lo)
425{
426 struct request *req;
427
428 BUG_ON(lo->magic != LO_MAGIC);
429
Herbert Xu4b2f0262006-01-06 00:09:47 -0800430 /*
431 * Because we have set lo->sock to NULL under the tx_lock, all
432 * modifications to the list must have completed by now. For
433 * the same reason, the active_req must be NULL.
434 *
435 * As a consequence, we don't need to take the spin lock while
436 * purging the list here.
437 */
438 BUG_ON(lo->sock);
439 BUG_ON(lo->active_req);
440
441 while (!list_empty(&lo->queue_head)) {
442 req = list_entry(lo->queue_head.next, struct request,
443 queuelist);
444 list_del_init(&req->queuelist);
445 req->errors++;
446 nbd_end_request(req);
447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Paul Clements7fdfd402007-10-16 23:27:37 -0700450
Laurent Vivier48cf6062008-04-29 01:02:46 -0700451static void nbd_handle_req(struct nbd_device *lo, struct request *req)
452{
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;
459 if (lo->flags & NBD_READ_ONLY) {
460 printk(KERN_ERR "%s: Write on read-only\n",
461 lo->disk->disk_name);
462 goto error_out;
463 }
464 }
465
466 req->errors = 0;
467
468 mutex_lock(&lo->tx_lock);
469 if (unlikely(!lo->sock)) {
470 mutex_unlock(&lo->tx_lock);
471 printk(KERN_ERR "%s: Attempted send on closed socket\n",
472 lo->disk->disk_name);
Pavel Machek15746fc2009-04-02 16:58:42 -0700473 goto error_out;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700474 }
475
476 lo->active_req = req;
477
478 if (nbd_send_req(lo, req) != 0) {
479 printk(KERN_ERR "%s: Request send failed\n",
480 lo->disk->disk_name);
481 req->errors++;
482 nbd_end_request(req);
483 } else {
484 spin_lock(&lo->queue_lock);
485 list_add(&req->queuelist, &lo->queue_head);
486 spin_unlock(&lo->queue_lock);
487 }
488
489 lo->active_req = NULL;
490 mutex_unlock(&lo->tx_lock);
491 wake_up_all(&lo->active_wq);
492
493 return;
494
495error_out:
496 req->errors++;
497 nbd_end_request(req);
498}
499
500static int nbd_thread(void *data)
501{
502 struct nbd_device *lo = data;
503 struct request *req;
504
505 set_user_nice(current, -20);
506 while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) {
507 /* wait for something to do */
508 wait_event_interruptible(lo->waiting_wq,
509 kthread_should_stop() ||
510 !list_empty(&lo->waiting_queue));
511
512 /* extract request */
513 if (list_empty(&lo->waiting_queue))
514 continue;
515
516 spin_lock_irq(&lo->queue_lock);
517 req = list_entry(lo->waiting_queue.next, struct request,
518 queuelist);
519 list_del_init(&req->queuelist);
520 spin_unlock_irq(&lo->queue_lock);
521
522 /* handle request */
523 nbd_handle_req(lo, req);
524 }
525 return 0;
526}
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528/*
529 * We always wait for result of write, for now. It would be nice to make it optional
530 * in future
Boaz Harroshe654bc42007-06-20 13:53:23 +0200531 * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
533 */
534
Pavel Machek15746fc2009-04-02 16:58:42 -0700535static void do_nbd_request(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 struct request *req;
538
Tejun Heo9934c8c2009-05-08 11:54:16 +0900539 while ((req = blk_fetch_request(q)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 struct nbd_device *lo;
541
Laurent Vivier48cf6062008-04-29 01:02:46 -0700542 spin_unlock_irq(q->queue_lock);
543
Jens Axboe4aff5e22006-08-10 08:44:47 +0200544 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
545 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 lo = req->rq_disk->private_data;
548
549 BUG_ON(lo->magic != LO_MAGIC);
550
Paul Clements4d48a542009-02-11 13:04:45 -0800551 if (unlikely(!lo->sock)) {
552 printk(KERN_ERR "%s: Attempted send on closed socket\n",
553 lo->disk->disk_name);
554 req->errors++;
555 nbd_end_request(req);
556 spin_lock_irq(q->queue_lock);
557 continue;
558 }
559
Laurent Vivier48cf6062008-04-29 01:02:46 -0700560 spin_lock_irq(&lo->queue_lock);
561 list_add_tail(&req->queuelist, &lo->waiting_queue);
562 spin_unlock_irq(&lo->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Laurent Vivier48cf6062008-04-29 01:02:46 -0700564 wake_up(&lo->waiting_wq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
Pavel Machek1a2ad212009-04-02 16:58:41 -0700570/* Must be called with tx_lock held */
571
572static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *lo,
573 unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 switch (cmd) {
Pavel Machek1a2ad212009-04-02 16:58:41 -0700576 case NBD_DISCONNECT: {
577 struct request sreq;
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700580
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +0200581 blk_rq_init(NULL, &sreq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200582 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 nbd_cmd(&sreq) = NBD_CMD_DISC;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700584 if (!lo->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700586 nbd_send_req(lo, &sreq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Pavel Machek1a2ad212009-04-02 16:58:41 -0700590 case NBD_CLEAR_SOCK: {
591 struct file *file;
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 lo->sock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 file = lo->file;
595 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 nbd_clear_que(lo);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800597 BUG_ON(!list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (file)
599 fput(file);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700600 return 0;
601 }
602
603 case NBD_SET_SOCK: {
604 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (lo->file)
606 return -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 file = fget(arg);
608 if (file) {
Al Viroa8cdc302008-03-02 09:33:33 -0500609 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (S_ISSOCK(inode->i_mode)) {
611 lo->file = file;
612 lo->sock = SOCKET_I(inode);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700613 if (max_part > 0)
614 bdev->bd_invalidated = 1;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700615 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 } else {
617 fput(file);
618 }
619 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700620 return -EINVAL;
621 }
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 case NBD_SET_BLKSIZE:
624 lo->blksize = arg;
625 lo->bytesize &= ~(lo->blksize-1);
Al Viroa8cdc302008-03-02 09:33:33 -0500626 bdev->bd_inode->i_size = lo->bytesize;
627 set_blocksize(bdev, lo->blksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 set_capacity(lo->disk, lo->bytesize >> 9);
629 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 case NBD_SET_SIZE:
632 lo->bytesize = arg & ~(lo->blksize-1);
Al Viroa8cdc302008-03-02 09:33:33 -0500633 bdev->bd_inode->i_size = lo->bytesize;
634 set_blocksize(bdev, lo->blksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 set_capacity(lo->disk, lo->bytesize >> 9);
636 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700637
Paul Clements7fdfd402007-10-16 23:27:37 -0700638 case NBD_SET_TIMEOUT:
639 lo->xmit_timeout = arg * HZ;
640 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 case NBD_SET_SIZE_BLOCKS:
643 lo->bytesize = ((u64) arg) * lo->blksize;
Al Viroa8cdc302008-03-02 09:33:33 -0500644 bdev->bd_inode->i_size = lo->bytesize;
645 set_blocksize(bdev, lo->blksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 set_capacity(lo->disk, lo->bytesize >> 9);
647 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700648
649 case NBD_DO_IT: {
650 struct task_struct *thread;
651 struct file *file;
652 int error;
653
Pavel Machekc91192d2009-01-15 13:51:03 -0800654 if (lo->pid)
655 return -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 if (!lo->file)
657 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700658
659 mutex_unlock(&lo->tx_lock);
660
Laurent Vivier48cf6062008-04-29 01:02:46 -0700661 thread = kthread_create(nbd_thread, lo, lo->disk->disk_name);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700662 if (IS_ERR(thread)) {
663 mutex_lock(&lo->tx_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700664 return PTR_ERR(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700665 }
Laurent Vivier48cf6062008-04-29 01:02:46 -0700666 wake_up_process(thread);
WANG Cong84963042007-05-09 02:33:36 -0700667 error = nbd_do_it(lo);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700668 kthread_stop(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700669
670 mutex_lock(&lo->tx_lock);
WANG Cong84963042007-05-09 02:33:36 -0700671 if (error)
672 return error;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700673 sock_shutdown(lo, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 file = lo->file;
675 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 nbd_clear_que(lo);
677 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
678 if (file)
679 fput(file);
Paul Clements4b86a872007-10-16 23:27:36 -0700680 lo->bytesize = 0;
Al Viroa8cdc302008-03-02 09:33:33 -0500681 bdev->bd_inode->i_size = 0;
Paul Clements4b86a872007-10-16 23:27:36 -0700682 set_capacity(lo->disk, 0);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700683 if (max_part > 0)
Al Viroa8cdc302008-03-02 09:33:33 -0500684 ioctl_by_bdev(bdev, BLKRRPART, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 return lo->harderror;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700686 }
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800689 /*
690 * This is for compatibility only. The queue is always cleared
691 * by NBD_DO_IT or NBD_CLEAR_SOCK.
692 */
693 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 case NBD_PRINT_DEBUG:
697 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
Al Viroa8cdc302008-03-02 09:33:33 -0500698 bdev->bd_disk->disk_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 lo->queue_head.next, lo->queue_head.prev,
700 &lo->queue_head);
701 return 0;
702 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700703 return -ENOTTY;
704}
705
706static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
707 unsigned int cmd, unsigned long arg)
708{
709 struct nbd_device *lo = bdev->bd_disk->private_data;
710 int error;
711
712 if (!capable(CAP_SYS_ADMIN))
713 return -EPERM;
714
715 BUG_ON(lo->magic != LO_MAGIC);
716
717 /* Anyone capable of this syscall can do *real bad* things */
718 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
719 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
720
721 mutex_lock(&lo->tx_lock);
722 error = __nbd_ioctl(bdev, lo, cmd, arg);
723 mutex_unlock(&lo->tx_lock);
724
725 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700728static const struct block_device_operations nbd_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
730 .owner = THIS_MODULE,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +0200731 .ioctl = nbd_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732};
733
734/*
735 * And here should be modules and kernel interface
736 * (Just smiley confuses emacs :-)
737 */
738
739static int __init nbd_init(void)
740{
741 int err = -ENOMEM;
742 int i;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700743 int part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800745 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700747 if (max_part < 0) {
748 printk(KERN_CRIT "nbd: max_part must be >= 0\n");
749 return -EINVAL;
750 }
751
Sven Wegenerf3944d62008-08-20 14:09:07 -0700752 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
753 if (!nbd_dev)
754 return -ENOMEM;
755
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700756 part_shift = 0;
Namhyung Kim5988ce22011-05-28 14:44:46 +0200757 if (max_part > 0) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700758 part_shift = fls(max_part);
759
Namhyung Kim5988ce22011-05-28 14:44:46 +0200760 /*
761 * Adjust max_part according to part_shift as it is exported
762 * to user space so that user can know the max number of
763 * partition kernel should be able to manage.
764 *
765 * Note that -1 is required because partition 0 is reserved
766 * for the whole disk.
767 */
768 max_part = (1UL << part_shift) - 1;
769 }
770
Namhyung Kim3b271082011-05-28 14:44:46 +0200771 if ((1UL << part_shift) > DISK_MAX_PARTS)
772 return -EINVAL;
773
774 if (nbds_max > 1UL << (MINORBITS - part_shift))
775 return -EINVAL;
776
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700777 for (i = 0; i < nbds_max; i++) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700778 struct gendisk *disk = alloc_disk(1 << part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (!disk)
780 goto out;
781 nbd_dev[i].disk = disk;
782 /*
783 * The new linux 2.5 block layer implementation requires
784 * every gendisk to have its very own request_queue struct.
785 * These structs are big so we dynamically allocate them.
786 */
787 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
788 if (!disk->queue) {
789 put_disk(disk);
790 goto out;
791 }
Jens Axboe31dcfab2008-10-31 10:06:37 +0100792 /*
793 * Tell the block layer that we are not a rotational device
794 */
795 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
797
798 if (register_blkdev(NBD_MAJOR, "nbd")) {
799 err = -EIO;
800 goto out;
801 }
802
803 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
804 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
805
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700806 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 struct gendisk *disk = nbd_dev[i].disk;
808 nbd_dev[i].file = NULL;
809 nbd_dev[i].magic = LO_MAGIC;
810 nbd_dev[i].flags = 0;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700811 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 spin_lock_init(&nbd_dev[i].queue_lock);
813 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800814 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800815 init_waitqueue_head(&nbd_dev[i].active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700816 init_waitqueue_head(&nbd_dev[i].waiting_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 nbd_dev[i].blksize = 1024;
Paul Clements4b86a872007-10-16 23:27:36 -0700818 nbd_dev[i].bytesize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 disk->major = NBD_MAJOR;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700820 disk->first_minor = i << part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 disk->fops = &nbd_fops;
822 disk->private_data = &nbd_dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 sprintf(disk->disk_name, "nbd%d", i);
Paul Clements4b86a872007-10-16 23:27:36 -0700824 set_capacity(disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 add_disk(disk);
826 }
827
828 return 0;
829out:
830 while (i--) {
831 blk_cleanup_queue(nbd_dev[i].disk->queue);
832 put_disk(nbd_dev[i].disk);
833 }
Sven Wegenerf3944d62008-08-20 14:09:07 -0700834 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 return err;
836}
837
838static void __exit nbd_cleanup(void)
839{
840 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700841 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700843 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (disk) {
845 del_gendisk(disk);
846 blk_cleanup_queue(disk->queue);
847 put_disk(disk);
848 }
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 unregister_blkdev(NBD_MAJOR, "nbd");
Sven Wegenerf3944d62008-08-20 14:09:07 -0700851 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
853}
854
855module_init(nbd_init);
856module_exit(nbd_cleanup);
857
858MODULE_DESCRIPTION("Network Block Device");
859MODULE_LICENSE("GPL");
860
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700861module_param(nbds_max, int, 0444);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700862MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
863module_param(max_part, int, 0444);
864MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865#ifndef NDEBUG
866module_param(debugflags, int, 0644);
867MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
868#endif