blob: 55298db36b2d61a113f25c22905fffb0f22ddd32 [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";
Paul Clements2f012502012-10-04 17:16:15 -070081 case NBD_SET_TIMEOUT: return "set-timeout";
82 case NBD_SET_FLAGS: return "set-flags";
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 case NBD_DO_IT: return "do-it";
84 case NBD_CLEAR_SOCK: return "clear-sock";
85 case NBD_CLEAR_QUE: return "clear-que";
86 case NBD_PRINT_DEBUG: return "print-debug";
87 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
88 case NBD_DISCONNECT: return "disconnect";
89 case BLKROSET: return "set-read-only";
90 case BLKFLSBUF: return "flush-buffer-cache";
91 }
92 return "unknown";
93}
94
95static const char *nbdcmd_to_ascii(int cmd)
96{
97 switch (cmd) {
98 case NBD_CMD_READ: return "read";
99 case NBD_CMD_WRITE: return "write";
100 case NBD_CMD_DISC: return "disconnect";
Alex Bligh75f187a2013-02-27 17:05:23 -0800101 case NBD_CMD_FLUSH: return "flush";
Paul Clementsa336d292012-10-04 17:16:18 -0700102 case NBD_CMD_TRIM: return "trim/discard";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 }
104 return "invalid";
105}
106#endif /* NDEBUG */
107
108static void nbd_end_request(struct request *req)
109{
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500110 int error = req->errors ? -EIO : 0;
Jens Axboe165125e2007-07-24 09:28:11 +0200111 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 unsigned long flags;
113
114 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
Kiyoshi Ueda097c94a2007-12-11 17:44:06 -0500115 req, error ? "failed" : "done");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo1011c1b2009-05-07 22:24:45 +0900118 __blk_end_request_all(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 spin_unlock_irqrestore(q->queue_lock, flags);
120}
121
Wanlong Gaof4507162012-03-28 14:42:51 -0700122static void sock_shutdown(struct nbd_device *nbd, int lock)
Paul Clements7fdfd402007-10-16 23:27:37 -0700123{
124 /* Forcibly shutdown the socket causing all listeners
125 * to error
126 *
127 * FIXME: This code is duplicated from sys_shutdown, but
128 * there should be a more generic interface rather than
129 * calling socket ops directly here */
130 if (lock)
Wanlong Gaof4507162012-03-28 14:42:51 -0700131 mutex_lock(&nbd->tx_lock);
132 if (nbd->sock) {
133 dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n");
134 kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
135 nbd->sock = NULL;
Paul Clements7fdfd402007-10-16 23:27:37 -0700136 }
137 if (lock)
Wanlong Gaof4507162012-03-28 14:42:51 -0700138 mutex_unlock(&nbd->tx_lock);
Paul Clements7fdfd402007-10-16 23:27:37 -0700139}
140
141static void nbd_xmit_timeout(unsigned long arg)
142{
143 struct task_struct *task = (struct task_struct *)arg;
144
145 printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
146 task->comm, task->pid);
147 force_sig(SIGKILL, task);
148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/*
151 * Send or receive packet.
152 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700153static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 int msg_flags)
155{
Wanlong Gaof4507162012-03-28 14:42:51 -0700156 struct socket *sock = nbd->sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 int result;
158 struct msghdr msg;
159 struct kvec iov;
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700160 sigset_t blocked, oldset;
Mel Gorman7f338fe2012-07-31 16:44:32 -0700161 unsigned long pflags = current->flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Mike Snitzerffc41cf2008-04-02 13:04:47 -0700163 if (unlikely(!sock)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700164 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200165 "Attempted %s on closed socket in sock_xmit\n",
166 (send ? "send" : "recv"));
Mike Snitzerffc41cf2008-04-02 13:04:47 -0700167 return -EINVAL;
168 }
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 /* Allow interception of SIGKILL only
171 * Don't allow other signals to interrupt the transmission */
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700172 siginitsetinv(&blocked, sigmask(SIGKILL));
173 sigprocmask(SIG_SETMASK, &blocked, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Mel Gorman7f338fe2012-07-31 16:44:32 -0700175 current->flags |= PF_MEMALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 do {
Mel Gorman7f338fe2012-07-31 16:44:32 -0700177 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 iov.iov_base = buf;
179 iov.iov_len = size;
180 msg.msg_name = NULL;
181 msg.msg_namelen = 0;
182 msg.msg_control = NULL;
183 msg.msg_controllen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
185
Paul Clements7fdfd402007-10-16 23:27:37 -0700186 if (send) {
187 struct timer_list ti;
188
Wanlong Gaof4507162012-03-28 14:42:51 -0700189 if (nbd->xmit_timeout) {
Paul Clements7fdfd402007-10-16 23:27:37 -0700190 init_timer(&ti);
191 ti.function = nbd_xmit_timeout;
192 ti.data = (unsigned long)current;
Wanlong Gaof4507162012-03-28 14:42:51 -0700193 ti.expires = jiffies + nbd->xmit_timeout;
Paul Clements7fdfd402007-10-16 23:27:37 -0700194 add_timer(&ti);
195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
Wanlong Gaof4507162012-03-28 14:42:51 -0700197 if (nbd->xmit_timeout)
Paul Clements7fdfd402007-10-16 23:27:37 -0700198 del_timer_sync(&ti);
199 } else
Namhyung Kim35fbf5b2011-05-28 14:44:46 +0200200 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
201 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 if (signal_pending(current)) {
204 siginfo_t info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700206 task_pid_nr(current), current->comm,
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700207 dequeue_signal_lock(current, &current->blocked, &info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 result = -EINTR;
Wanlong Gaof4507162012-03-28 14:42:51 -0700209 sock_shutdown(nbd, !send);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 break;
211 }
212
213 if (result <= 0) {
214 if (result == 0)
215 result = -EPIPE; /* short read */
216 break;
217 }
218 size -= result;
219 buf += result;
220 } while (size > 0);
221
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700222 sigprocmask(SIG_SETMASK, &oldset, NULL);
Mel Gorman7f338fe2012-07-31 16:44:32 -0700223 tsk_restore_flags(current, pflags, PF_MEMALLOC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
225 return result;
226}
227
Wanlong Gaof4507162012-03-28 14:42:51 -0700228static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 int flags)
230{
231 int result;
232 void *kaddr = kmap(bvec->bv_page);
Wanlong Gaof4507162012-03-28 14:42:51 -0700233 result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset,
234 bvec->bv_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 kunmap(bvec->bv_page);
236 return result;
237}
238
Paul Clements7fdfd402007-10-16 23:27:37 -0700239/* always call with the tx_lock held */
Wanlong Gaof4507162012-03-28 14:42:51 -0700240static int nbd_send_req(struct nbd_device *nbd, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
NeilBrown5705f702007-09-25 12:35:59 +0200242 int result, flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 struct nbd_request request;
Tejun Heo1011c1b2009-05-07 22:24:45 +0900244 unsigned long size = blk_rq_bytes(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246 request.magic = htonl(NBD_REQUEST_MAGIC);
247 request.type = htonl(nbd_cmd(req));
Alex Bligh75f187a2013-02-27 17:05:23 -0800248
249 if (nbd_cmd(req) == NBD_CMD_FLUSH) {
250 /* Other values are reserved for FLUSH requests. */
251 request.from = 0;
252 request.len = 0;
253 } else {
254 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
255 request.len = htonl(size);
256 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 memcpy(request.handle, &req, sizeof(req));
258
Tejun Heo83096eb2009-05-07 22:24:39 +0900259 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700260 nbd->disk->disk_name, req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 nbdcmd_to_ascii(nbd_cmd(req)),
Tejun Heo83096eb2009-05-07 22:24:39 +0900262 (unsigned long long)blk_rq_pos(req) << 9,
Tejun Heo1011c1b2009-05-07 22:24:45 +0900263 blk_rq_bytes(req));
Wanlong Gaof4507162012-03-28 14:42:51 -0700264 result = sock_xmit(nbd, 1, &request, sizeof(request),
Paul Clements7fdfd402007-10-16 23:27:37 -0700265 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700267 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200268 "Send control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 goto error_out;
270 }
271
272 if (nbd_cmd(req) == NBD_CMD_WRITE) {
NeilBrown5705f702007-09-25 12:35:59 +0200273 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -0800274 struct bio_vec bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 /*
276 * we are really probing at internals to determine
277 * whether to set MSG_MORE or not...
278 */
NeilBrown5705f702007-09-25 12:35:59 +0200279 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200280 flags = 0;
Kent Overstreet4550dd62013-08-07 14:26:21 -0700281 if (!rq_iter_last(bvec, iter))
Jens Axboe6c92e692007-08-16 13:43:12 +0200282 flags = MSG_MORE;
283 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
Kent Overstreet79886132013-11-23 17:19:00 -0800284 nbd->disk->disk_name, req, bvec.bv_len);
285 result = sock_send_bvec(nbd, &bvec, flags);
Jens Axboe6c92e692007-08-16 13:43:12 +0200286 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700287 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200288 "Send data failed (result %d)\n",
289 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200290 goto error_out;
291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 return 0;
295
296error_out:
Pavel Machek15746fc2009-04-02 16:58:42 -0700297 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
299
Wanlong Gaof4507162012-03-28 14:42:51 -0700300static struct request *nbd_find_request(struct nbd_device *nbd,
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700301 struct request *xreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Denis Chengd2c97402007-10-16 23:26:14 -0700303 struct request *req, *tmp;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800304 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Wanlong Gaof4507162012-03-28 14:42:51 -0700306 err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800307 if (unlikely(err))
308 goto out;
309
Wanlong Gaof4507162012-03-28 14:42:51 -0700310 spin_lock(&nbd->queue_lock);
311 list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 if (req != xreq)
313 continue;
314 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700315 spin_unlock(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 return req;
317 }
Wanlong Gaof4507162012-03-28 14:42:51 -0700318 spin_unlock(&nbd->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800319
320 err = -ENOENT;
321
322out:
323 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
Wanlong Gaof4507162012-03-28 14:42:51 -0700326static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 int result;
329 void *kaddr = kmap(bvec->bv_page);
Wanlong Gaof4507162012-03-28 14:42:51 -0700330 result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 MSG_WAITALL);
332 kunmap(bvec->bv_page);
333 return result;
334}
335
336/* NULL returned = something went wrong, inform userspace */
Wanlong Gaof4507162012-03-28 14:42:51 -0700337static struct request *nbd_read_stat(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
339 int result;
340 struct nbd_reply reply;
341 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 reply.magic = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700344 result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700346 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200347 "Receive control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 goto harderror;
349 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700350
351 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700352 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
Michal Feixe4b57e02006-07-30 03:03:31 -0700353 (unsigned long)ntohl(reply.magic));
354 result = -EPROTO;
355 goto harderror;
356 }
357
Wanlong Gaof4507162012-03-28 14:42:51 -0700358 req = nbd_find_request(nbd, *(struct request **)reply.handle);
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -0700359 if (IS_ERR(req)) {
Herbert Xu4b2f0262006-01-06 00:09:47 -0800360 result = PTR_ERR(req);
361 if (result != -ENOENT)
362 goto harderror;
363
Wanlong Gaof4507162012-03-28 14:42:51 -0700364 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200365 reply.handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 result = -EBADR;
367 goto harderror;
368 }
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 if (ntohl(reply.error)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700371 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200372 ntohl(reply.error));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 req->errors++;
374 return req;
375 }
376
377 dprintk(DBG_RX, "%s: request %p: got reply\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700378 nbd->disk->disk_name, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (nbd_cmd(req) == NBD_CMD_READ) {
NeilBrown5705f702007-09-25 12:35:59 +0200380 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -0800381 struct bio_vec bvec;
NeilBrown5705f702007-09-25 12:35:59 +0200382
383 rq_for_each_segment(bvec, req, iter) {
Kent Overstreet79886132013-11-23 17:19:00 -0800384 result = sock_recv_bvec(nbd, &bvec);
Jens Axboe6c92e692007-08-16 13:43:12 +0200385 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700386 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200387 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200388 req->errors++;
389 return req;
390 }
391 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
Kent Overstreet79886132013-11-23 17:19:00 -0800392 nbd->disk->disk_name, req, bvec.bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
394 }
395 return req;
396harderror:
Wanlong Gaof4507162012-03-28 14:42:51 -0700397 nbd->harderror = result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 return NULL;
399}
400
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200401static ssize_t pid_show(struct device *dev,
402 struct device_attribute *attr, char *buf)
Paul Clements6b39bb62006-12-06 20:40:53 -0800403{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200404 struct gendisk *disk = dev_to_disk(dev);
405
406 return sprintf(buf, "%ld\n",
Paul Clements6b39bb62006-12-06 20:40:53 -0800407 (long) ((struct nbd_device *)disk->private_data)->pid);
408}
409
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200410static struct device_attribute pid_attr = {
Parag Warudkar01e8ef12008-10-18 20:28:50 -0700411 .attr = { .name = "pid", .mode = S_IRUGO},
Paul Clements6b39bb62006-12-06 20:40:53 -0800412 .show = pid_show,
413};
414
Wanlong Gaof4507162012-03-28 14:42:51 -0700415static int nbd_do_it(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
417 struct request *req;
WANG Cong84963042007-05-09 02:33:36 -0700418 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Wanlong Gaof4507162012-03-28 14:42:51 -0700420 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Mel Gorman7f338fe2012-07-31 16:44:32 -0700422 sk_set_memalloc(nbd->sock->sk);
Wanlong Gaof4507162012-03-28 14:42:51 -0700423 nbd->pid = task_pid_nr(current);
424 ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
WANG Cong84963042007-05-09 02:33:36 -0700425 if (ret) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700426 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
427 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700428 return ret;
429 }
Paul Clements6b39bb62006-12-06 20:40:53 -0800430
Wanlong Gaof4507162012-03-28 14:42:51 -0700431 while ((req = nbd_read_stat(nbd)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800433
Wanlong Gaof4507162012-03-28 14:42:51 -0700434 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
435 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700436 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
Wanlong Gaof4507162012-03-28 14:42:51 -0700439static void nbd_clear_que(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 struct request *req;
442
Wanlong Gaof4507162012-03-28 14:42:51 -0700443 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Herbert Xu4b2f0262006-01-06 00:09:47 -0800445 /*
Wanlong Gaof4507162012-03-28 14:42:51 -0700446 * Because we have set nbd->sock to NULL under the tx_lock, all
Herbert Xu4b2f0262006-01-06 00:09:47 -0800447 * modifications to the list must have completed by now. For
448 * the same reason, the active_req must be NULL.
449 *
450 * As a consequence, we don't need to take the spin lock while
451 * purging the list here.
452 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700453 BUG_ON(nbd->sock);
454 BUG_ON(nbd->active_req);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800455
Wanlong Gaof4507162012-03-28 14:42:51 -0700456 while (!list_empty(&nbd->queue_head)) {
457 req = list_entry(nbd->queue_head.next, struct request,
Herbert Xu4b2f0262006-01-06 00:09:47 -0800458 queuelist);
459 list_del_init(&req->queuelist);
460 req->errors++;
461 nbd_end_request(req);
462 }
Paul Clementsfded4e02012-09-17 14:09:02 -0700463
464 while (!list_empty(&nbd->waiting_queue)) {
465 req = list_entry(nbd->waiting_queue.next, struct request,
466 queuelist);
467 list_del_init(&req->queuelist);
468 req->errors++;
469 nbd_end_request(req);
470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
Paul Clements7fdfd402007-10-16 23:27:37 -0700473
Wanlong Gaof4507162012-03-28 14:42:51 -0700474static void nbd_handle_req(struct nbd_device *nbd, struct request *req)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700475{
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200476 if (req->cmd_type != REQ_TYPE_FS)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700477 goto error_out;
478
479 nbd_cmd(req) = NBD_CMD_READ;
480 if (rq_data_dir(req) == WRITE) {
Paul Clementsa336d292012-10-04 17:16:18 -0700481 if ((req->cmd_flags & REQ_DISCARD)) {
482 WARN_ON(!(nbd->flags & NBD_FLAG_SEND_TRIM));
483 nbd_cmd(req) = NBD_CMD_TRIM;
484 } else
485 nbd_cmd(req) = NBD_CMD_WRITE;
Paul Clements2f012502012-10-04 17:16:15 -0700486 if (nbd->flags & NBD_FLAG_READ_ONLY) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700487 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200488 "Write on read-only\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700489 goto error_out;
490 }
491 }
492
Alex Bligh75f187a2013-02-27 17:05:23 -0800493 if (req->cmd_flags & REQ_FLUSH) {
494 BUG_ON(unlikely(blk_rq_sectors(req)));
495 nbd_cmd(req) = NBD_CMD_FLUSH;
496 }
497
Laurent Vivier48cf6062008-04-29 01:02:46 -0700498 req->errors = 0;
499
Wanlong Gaof4507162012-03-28 14:42:51 -0700500 mutex_lock(&nbd->tx_lock);
501 if (unlikely(!nbd->sock)) {
502 mutex_unlock(&nbd->tx_lock);
503 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200504 "Attempted send on closed socket\n");
Pavel Machek15746fc2009-04-02 16:58:42 -0700505 goto error_out;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700506 }
507
Wanlong Gaof4507162012-03-28 14:42:51 -0700508 nbd->active_req = req;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700509
Wanlong Gaof4507162012-03-28 14:42:51 -0700510 if (nbd_send_req(nbd, req) != 0) {
511 dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700512 req->errors++;
513 nbd_end_request(req);
514 } else {
Wanlong Gaof4507162012-03-28 14:42:51 -0700515 spin_lock(&nbd->queue_lock);
Chetan Loke01ff5db2012-07-31 08:47:13 +0200516 list_add_tail(&req->queuelist, &nbd->queue_head);
Wanlong Gaof4507162012-03-28 14:42:51 -0700517 spin_unlock(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700518 }
519
Wanlong Gaof4507162012-03-28 14:42:51 -0700520 nbd->active_req = NULL;
521 mutex_unlock(&nbd->tx_lock);
522 wake_up_all(&nbd->active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700523
524 return;
525
526error_out:
527 req->errors++;
528 nbd_end_request(req);
529}
530
531static int nbd_thread(void *data)
532{
Wanlong Gaof4507162012-03-28 14:42:51 -0700533 struct nbd_device *nbd = data;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700534 struct request *req;
535
536 set_user_nice(current, -20);
Wanlong Gaof4507162012-03-28 14:42:51 -0700537 while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) {
Laurent Vivier48cf6062008-04-29 01:02:46 -0700538 /* wait for something to do */
Wanlong Gaof4507162012-03-28 14:42:51 -0700539 wait_event_interruptible(nbd->waiting_wq,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700540 kthread_should_stop() ||
Wanlong Gaof4507162012-03-28 14:42:51 -0700541 !list_empty(&nbd->waiting_queue));
Laurent Vivier48cf6062008-04-29 01:02:46 -0700542
543 /* extract request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700544 if (list_empty(&nbd->waiting_queue))
Laurent Vivier48cf6062008-04-29 01:02:46 -0700545 continue;
546
Wanlong Gaof4507162012-03-28 14:42:51 -0700547 spin_lock_irq(&nbd->queue_lock);
548 req = list_entry(nbd->waiting_queue.next, struct request,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700549 queuelist);
550 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700551 spin_unlock_irq(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700552
553 /* handle request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700554 nbd_handle_req(nbd, req);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700555 }
556 return 0;
557}
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559/*
560 * We always wait for result of write, for now. It would be nice to make it optional
561 * in future
Wanlong Gaof4507162012-03-28 14:42:51 -0700562 * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
564 */
565
Pavel Machek15746fc2009-04-02 16:58:42 -0700566static void do_nbd_request(struct request_queue *q)
Alex Elder398eb082013-02-27 17:05:28 -0800567 __releases(q->queue_lock) __acquires(q->queue_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 struct request *req;
570
Tejun Heo9934c8c2009-05-08 11:54:16 +0900571 while ((req = blk_fetch_request(q)) != NULL) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700572 struct nbd_device *nbd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Laurent Vivier48cf6062008-04-29 01:02:46 -0700574 spin_unlock_irq(q->queue_lock);
575
Jens Axboe4aff5e22006-08-10 08:44:47 +0200576 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
577 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Wanlong Gaof4507162012-03-28 14:42:51 -0700579 nbd = req->rq_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Wanlong Gaof4507162012-03-28 14:42:51 -0700581 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Wanlong Gaof4507162012-03-28 14:42:51 -0700583 if (unlikely(!nbd->sock)) {
584 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200585 "Attempted send on closed socket\n");
Paul Clements4d48a542009-02-11 13:04:45 -0800586 req->errors++;
587 nbd_end_request(req);
588 spin_lock_irq(q->queue_lock);
589 continue;
590 }
591
Wanlong Gaof4507162012-03-28 14:42:51 -0700592 spin_lock_irq(&nbd->queue_lock);
593 list_add_tail(&req->queuelist, &nbd->waiting_queue);
594 spin_unlock_irq(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Wanlong Gaof4507162012-03-28 14:42:51 -0700596 wake_up(&nbd->waiting_wq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
Pavel Machek1a2ad212009-04-02 16:58:41 -0700602/* Must be called with tx_lock held */
603
Wanlong Gaof4507162012-03-28 14:42:51 -0700604static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
Pavel Machek1a2ad212009-04-02 16:58:41 -0700605 unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 switch (cmd) {
Pavel Machek1a2ad212009-04-02 16:58:41 -0700608 case NBD_DISCONNECT: {
609 struct request sreq;
610
Wanlong Gaof4507162012-03-28 14:42:51 -0700611 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800612 if (!nbd->sock)
613 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700614
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800615 mutex_unlock(&nbd->tx_lock);
616 fsync_bdev(bdev);
617 mutex_lock(&nbd->tx_lock);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +0200618 blk_rq_init(NULL, &sreq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200619 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 nbd_cmd(&sreq) = NBD_CMD_DISC;
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800621
622 /* Check again after getting mutex back. */
Wanlong Gaof4507162012-03-28 14:42:51 -0700623 if (!nbd->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return -EINVAL;
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800625
Paul Clementsc378f702013-07-03 15:09:04 -0700626 nbd->disconnect = 1;
627
Wanlong Gaof4507162012-03-28 14:42:51 -0700628 nbd_send_req(nbd, &sreq);
Paul Clementsc378f702013-07-03 15:09:04 -0700629 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Pavel Machek1a2ad212009-04-02 16:58:41 -0700632 case NBD_CLEAR_SOCK: {
633 struct file *file;
634
Wanlong Gaof4507162012-03-28 14:42:51 -0700635 nbd->sock = NULL;
636 file = nbd->file;
637 nbd->file = NULL;
638 nbd_clear_que(nbd);
639 BUG_ON(!list_empty(&nbd->queue_head));
Paul Clementsfded4e02012-09-17 14:09:02 -0700640 BUG_ON(!list_empty(&nbd->waiting_queue));
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800641 kill_bdev(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 if (file)
643 fput(file);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700644 return 0;
645 }
646
647 case NBD_SET_SOCK: {
648 struct file *file;
Wanlong Gaof4507162012-03-28 14:42:51 -0700649 if (nbd->file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 file = fget(arg);
652 if (file) {
Al Viro496ad9a2013-01-23 17:07:38 -0500653 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (S_ISSOCK(inode->i_mode)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700655 nbd->file = file;
656 nbd->sock = SOCKET_I(inode);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700657 if (max_part > 0)
658 bdev->bd_invalidated = 1;
Paul Clementsc378f702013-07-03 15:09:04 -0700659 nbd->disconnect = 0; /* we're connected now */
Pavel Machek1a2ad212009-04-02 16:58:41 -0700660 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 } else {
662 fput(file);
663 }
664 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700665 return -EINVAL;
666 }
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 case NBD_SET_BLKSIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700669 nbd->blksize = arg;
670 nbd->bytesize &= ~(nbd->blksize-1);
671 bdev->bd_inode->i_size = nbd->bytesize;
672 set_blocksize(bdev, nbd->blksize);
673 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 case NBD_SET_SIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700677 nbd->bytesize = arg & ~(nbd->blksize-1);
678 bdev->bd_inode->i_size = nbd->bytesize;
679 set_blocksize(bdev, nbd->blksize);
680 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700682
Paul Clements7fdfd402007-10-16 23:27:37 -0700683 case NBD_SET_TIMEOUT:
Wanlong Gaof4507162012-03-28 14:42:51 -0700684 nbd->xmit_timeout = arg * HZ;
Paul Clements7fdfd402007-10-16 23:27:37 -0700685 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700686
Paul Clements2f012502012-10-04 17:16:15 -0700687 case NBD_SET_FLAGS:
688 nbd->flags = arg;
689 return 0;
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 case NBD_SET_SIZE_BLOCKS:
Wanlong Gaof4507162012-03-28 14:42:51 -0700692 nbd->bytesize = ((u64) arg) * nbd->blksize;
693 bdev->bd_inode->i_size = nbd->bytesize;
694 set_blocksize(bdev, nbd->blksize);
695 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700697
698 case NBD_DO_IT: {
699 struct task_struct *thread;
700 struct file *file;
701 int error;
702
Wanlong Gaof4507162012-03-28 14:42:51 -0700703 if (nbd->pid)
Pavel Machekc91192d2009-01-15 13:51:03 -0800704 return -EBUSY;
Wanlong Gaof4507162012-03-28 14:42:51 -0700705 if (!nbd->file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700707
Wanlong Gaof4507162012-03-28 14:42:51 -0700708 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700709
Paolo Bonzinia83e8142013-02-27 17:05:26 -0800710 if (nbd->flags & NBD_FLAG_READ_ONLY)
711 set_device_ro(bdev, true);
Paul Clementsa336d292012-10-04 17:16:18 -0700712 if (nbd->flags & NBD_FLAG_SEND_TRIM)
713 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
714 nbd->disk->queue);
Alex Bligh75f187a2013-02-27 17:05:23 -0800715 if (nbd->flags & NBD_FLAG_SEND_FLUSH)
716 blk_queue_flush(nbd->disk->queue, REQ_FLUSH);
717 else
718 blk_queue_flush(nbd->disk->queue, 0);
Paul Clementsa336d292012-10-04 17:16:18 -0700719
Kees Cookffc8b302013-07-03 15:01:14 -0700720 thread = kthread_create(nbd_thread, nbd, "%s",
721 nbd->disk->disk_name);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700722 if (IS_ERR(thread)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700723 mutex_lock(&nbd->tx_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700724 return PTR_ERR(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700725 }
Laurent Vivier48cf6062008-04-29 01:02:46 -0700726 wake_up_process(thread);
Wanlong Gaof4507162012-03-28 14:42:51 -0700727 error = nbd_do_it(nbd);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700728 kthread_stop(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700729
Wanlong Gaof4507162012-03-28 14:42:51 -0700730 mutex_lock(&nbd->tx_lock);
WANG Cong84963042007-05-09 02:33:36 -0700731 if (error)
732 return error;
Wanlong Gaof4507162012-03-28 14:42:51 -0700733 sock_shutdown(nbd, 0);
734 file = nbd->file;
735 nbd->file = NULL;
736 nbd_clear_que(nbd);
737 dev_warn(disk_to_dev(nbd->disk), "queue cleared\n");
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800738 kill_bdev(bdev);
Paul Clementsa336d292012-10-04 17:16:18 -0700739 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
Paolo Bonzinia83e8142013-02-27 17:05:26 -0800740 set_device_ro(bdev, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 if (file)
742 fput(file);
Alex Bligh75f187a2013-02-27 17:05:23 -0800743 nbd->flags = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700744 nbd->bytesize = 0;
Al Viroa8cdc302008-03-02 09:33:33 -0500745 bdev->bd_inode->i_size = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700746 set_capacity(nbd->disk, 0);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700747 if (max_part > 0)
Al Viroa8cdc302008-03-02 09:33:33 -0500748 ioctl_by_bdev(bdev, BLKRRPART, 0);
Paul Clementsc378f702013-07-03 15:09:04 -0700749 if (nbd->disconnect) /* user requested, ignore socket errors */
750 return 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700751 return nbd->harderror;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700752 }
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800755 /*
756 * This is for compatibility only. The queue is always cleared
757 * by NBD_DO_IT or NBD_CLEAR_SOCK.
758 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700760
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 case NBD_PRINT_DEBUG:
Wanlong Gaof4507162012-03-28 14:42:51 -0700762 dev_info(disk_to_dev(nbd->disk),
WANG Cong5eedf542011-08-19 14:48:28 +0200763 "next = %p, prev = %p, head = %p\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700764 nbd->queue_head.next, nbd->queue_head.prev,
765 &nbd->queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return 0;
767 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700768 return -ENOTTY;
769}
770
771static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
772 unsigned int cmd, unsigned long arg)
773{
Wanlong Gaof4507162012-03-28 14:42:51 -0700774 struct nbd_device *nbd = bdev->bd_disk->private_data;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700775 int error;
776
777 if (!capable(CAP_SYS_ADMIN))
778 return -EPERM;
779
Wanlong Gaof4507162012-03-28 14:42:51 -0700780 BUG_ON(nbd->magic != NBD_MAGIC);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700781
782 /* Anyone capable of this syscall can do *real bad* things */
783 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700784 nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700785
Wanlong Gaof4507162012-03-28 14:42:51 -0700786 mutex_lock(&nbd->tx_lock);
787 error = __nbd_ioctl(bdev, nbd, cmd, arg);
788 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700789
790 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700793static const struct block_device_operations nbd_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
795 .owner = THIS_MODULE,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +0200796 .ioctl = nbd_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797};
798
799/*
800 * And here should be modules and kernel interface
801 * (Just smiley confuses emacs :-)
802 */
803
804static int __init nbd_init(void)
805{
806 int err = -ENOMEM;
807 int i;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700808 int part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800810 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700812 if (max_part < 0) {
WANG Cong7742ce42011-08-19 14:48:28 +0200813 printk(KERN_ERR "nbd: max_part must be >= 0\n");
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700814 return -EINVAL;
815 }
816
Sven Wegenerf3944d62008-08-20 14:09:07 -0700817 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
818 if (!nbd_dev)
819 return -ENOMEM;
820
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700821 part_shift = 0;
Namhyung Kim5988ce22011-05-28 14:44:46 +0200822 if (max_part > 0) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700823 part_shift = fls(max_part);
824
Namhyung Kim5988ce22011-05-28 14:44:46 +0200825 /*
826 * Adjust max_part according to part_shift as it is exported
827 * to user space so that user can know the max number of
828 * partition kernel should be able to manage.
829 *
830 * Note that -1 is required because partition 0 is reserved
831 * for the whole disk.
832 */
833 max_part = (1UL << part_shift) - 1;
834 }
835
Namhyung Kim3b271082011-05-28 14:44:46 +0200836 if ((1UL << part_shift) > DISK_MAX_PARTS)
837 return -EINVAL;
838
839 if (nbds_max > 1UL << (MINORBITS - part_shift))
840 return -EINVAL;
841
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700842 for (i = 0; i < nbds_max; i++) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700843 struct gendisk *disk = alloc_disk(1 << part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (!disk)
845 goto out;
846 nbd_dev[i].disk = disk;
847 /*
848 * The new linux 2.5 block layer implementation requires
849 * every gendisk to have its very own request_queue struct.
850 * These structs are big so we dynamically allocate them.
851 */
852 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
853 if (!disk->queue) {
854 put_disk(disk);
855 goto out;
856 }
Jens Axboe31dcfab2008-10-31 10:06:37 +0100857 /*
858 * Tell the block layer that we are not a rotational device
859 */
860 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
Paul Clementsa336d292012-10-04 17:16:18 -0700861 disk->queue->limits.discard_granularity = 512;
862 disk->queue->limits.max_discard_sectors = UINT_MAX;
863 disk->queue->limits.discard_zeroes_data = 0;
Michal Belczyk078be022013-04-30 15:28:28 -0700864 blk_queue_max_hw_sectors(disk->queue, 65536);
865 disk->queue->limits.max_sectors = 256;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 }
867
868 if (register_blkdev(NBD_MAJOR, "nbd")) {
869 err = -EIO;
870 goto out;
871 }
872
873 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
874 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
875
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700876 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 struct gendisk *disk = nbd_dev[i].disk;
878 nbd_dev[i].file = NULL;
Wanlong Gaof4507162012-03-28 14:42:51 -0700879 nbd_dev[i].magic = NBD_MAGIC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 nbd_dev[i].flags = 0;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700881 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 spin_lock_init(&nbd_dev[i].queue_lock);
883 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800884 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800885 init_waitqueue_head(&nbd_dev[i].active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700886 init_waitqueue_head(&nbd_dev[i].waiting_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 nbd_dev[i].blksize = 1024;
Paul Clements4b86a872007-10-16 23:27:36 -0700888 nbd_dev[i].bytesize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 disk->major = NBD_MAJOR;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700890 disk->first_minor = i << part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 disk->fops = &nbd_fops;
892 disk->private_data = &nbd_dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 sprintf(disk->disk_name, "nbd%d", i);
Paul Clements4b86a872007-10-16 23:27:36 -0700894 set_capacity(disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 add_disk(disk);
896 }
897
898 return 0;
899out:
900 while (i--) {
901 blk_cleanup_queue(nbd_dev[i].disk->queue);
902 put_disk(nbd_dev[i].disk);
903 }
Sven Wegenerf3944d62008-08-20 14:09:07 -0700904 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 return err;
906}
907
908static void __exit nbd_cleanup(void)
909{
910 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700911 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700913 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (disk) {
915 del_gendisk(disk);
916 blk_cleanup_queue(disk->queue);
917 put_disk(disk);
918 }
919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 unregister_blkdev(NBD_MAJOR, "nbd");
Sven Wegenerf3944d62008-08-20 14:09:07 -0700921 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
923}
924
925module_init(nbd_init);
926module_exit(nbd_cleanup);
927
928MODULE_DESCRIPTION("Network Block Device");
929MODULE_LICENSE("GPL");
930
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700931module_param(nbds_max, int, 0444);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700932MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
933module_param(max_part, int, 0444);
934MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935#ifndef NDEBUG
936module_param(debugflags, int, 0644);
937MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
938#endif