blob: fb31b8ee4372323a4a93eddb0aa9141ea97d1021 [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
Hani Benhabiles04cfac42014-06-06 14:38:30 -0700246 memset(&request, 0, sizeof(request));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 request.magic = htonl(NBD_REQUEST_MAGIC);
248 request.type = htonl(nbd_cmd(req));
Alex Bligh75f187a2013-02-27 17:05:23 -0800249
Hani Benhabiles04cfac42014-06-06 14:38:30 -0700250 if (nbd_cmd(req) != NBD_CMD_FLUSH && nbd_cmd(req) != NBD_CMD_DISC) {
Alex Bligh75f187a2013-02-27 17:05:23 -0800251 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
252 request.len = htonl(size);
253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 memcpy(request.handle, &req, sizeof(req));
255
Tejun Heo83096eb2009-05-07 22:24:39 +0900256 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700257 nbd->disk->disk_name, req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 nbdcmd_to_ascii(nbd_cmd(req)),
Tejun Heo83096eb2009-05-07 22:24:39 +0900259 (unsigned long long)blk_rq_pos(req) << 9,
Tejun Heo1011c1b2009-05-07 22:24:45 +0900260 blk_rq_bytes(req));
Wanlong Gaof4507162012-03-28 14:42:51 -0700261 result = sock_xmit(nbd, 1, &request, sizeof(request),
Paul Clements7fdfd402007-10-16 23:27:37 -0700262 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700264 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200265 "Send control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 goto error_out;
267 }
268
269 if (nbd_cmd(req) == NBD_CMD_WRITE) {
NeilBrown5705f702007-09-25 12:35:59 +0200270 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -0800271 struct bio_vec bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 /*
273 * we are really probing at internals to determine
274 * whether to set MSG_MORE or not...
275 */
NeilBrown5705f702007-09-25 12:35:59 +0200276 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200277 flags = 0;
Kent Overstreet4550dd62013-08-07 14:26:21 -0700278 if (!rq_iter_last(bvec, iter))
Jens Axboe6c92e692007-08-16 13:43:12 +0200279 flags = MSG_MORE;
280 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
Kent Overstreet79886132013-11-23 17:19:00 -0800281 nbd->disk->disk_name, req, bvec.bv_len);
282 result = sock_send_bvec(nbd, &bvec, flags);
Jens Axboe6c92e692007-08-16 13:43:12 +0200283 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700284 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200285 "Send data failed (result %d)\n",
286 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200287 goto error_out;
288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 }
290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return 0;
292
293error_out:
Pavel Machek15746fc2009-04-02 16:58:42 -0700294 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
Wanlong Gaof4507162012-03-28 14:42:51 -0700297static struct request *nbd_find_request(struct nbd_device *nbd,
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700298 struct request *xreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Denis Chengd2c97402007-10-16 23:26:14 -0700300 struct request *req, *tmp;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800301 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Wanlong Gaof4507162012-03-28 14:42:51 -0700303 err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800304 if (unlikely(err))
305 goto out;
306
Wanlong Gaof4507162012-03-28 14:42:51 -0700307 spin_lock(&nbd->queue_lock);
308 list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 if (req != xreq)
310 continue;
311 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700312 spin_unlock(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 return req;
314 }
Wanlong Gaof4507162012-03-28 14:42:51 -0700315 spin_unlock(&nbd->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800316
317 err = -ENOENT;
318
319out:
320 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Wanlong Gaof4507162012-03-28 14:42:51 -0700323static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
325 int result;
326 void *kaddr = kmap(bvec->bv_page);
Wanlong Gaof4507162012-03-28 14:42:51 -0700327 result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 MSG_WAITALL);
329 kunmap(bvec->bv_page);
330 return result;
331}
332
333/* NULL returned = something went wrong, inform userspace */
Wanlong Gaof4507162012-03-28 14:42:51 -0700334static struct request *nbd_read_stat(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
336 int result;
337 struct nbd_reply reply;
338 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 reply.magic = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700341 result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700343 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200344 "Receive control failed (result %d)\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 goto harderror;
346 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700347
348 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700349 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
Michal Feixe4b57e02006-07-30 03:03:31 -0700350 (unsigned long)ntohl(reply.magic));
351 result = -EPROTO;
352 goto harderror;
353 }
354
Wanlong Gaof4507162012-03-28 14:42:51 -0700355 req = nbd_find_request(nbd, *(struct request **)reply.handle);
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -0700356 if (IS_ERR(req)) {
Herbert Xu4b2f0262006-01-06 00:09:47 -0800357 result = PTR_ERR(req);
358 if (result != -ENOENT)
359 goto harderror;
360
Wanlong Gaof4507162012-03-28 14:42:51 -0700361 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200362 reply.handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 result = -EBADR;
364 goto harderror;
365 }
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (ntohl(reply.error)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700368 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200369 ntohl(reply.error));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 req->errors++;
371 return req;
372 }
373
374 dprintk(DBG_RX, "%s: request %p: got reply\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700375 nbd->disk->disk_name, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (nbd_cmd(req) == NBD_CMD_READ) {
NeilBrown5705f702007-09-25 12:35:59 +0200377 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -0800378 struct bio_vec bvec;
NeilBrown5705f702007-09-25 12:35:59 +0200379
380 rq_for_each_segment(bvec, req, iter) {
Kent Overstreet79886132013-11-23 17:19:00 -0800381 result = sock_recv_bvec(nbd, &bvec);
Jens Axboe6c92e692007-08-16 13:43:12 +0200382 if (result <= 0) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700383 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
WANG Cong7f1b90f2011-08-19 14:48:22 +0200384 result);
Jens Axboe6c92e692007-08-16 13:43:12 +0200385 req->errors++;
386 return req;
387 }
388 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
Kent Overstreet79886132013-11-23 17:19:00 -0800389 nbd->disk->disk_name, req, bvec.bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 }
391 }
392 return req;
393harderror:
Wanlong Gaof4507162012-03-28 14:42:51 -0700394 nbd->harderror = result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 return NULL;
396}
397
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200398static ssize_t pid_show(struct device *dev,
399 struct device_attribute *attr, char *buf)
Paul Clements6b39bb62006-12-06 20:40:53 -0800400{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200401 struct gendisk *disk = dev_to_disk(dev);
402
403 return sprintf(buf, "%ld\n",
Paul Clements6b39bb62006-12-06 20:40:53 -0800404 (long) ((struct nbd_device *)disk->private_data)->pid);
405}
406
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200407static struct device_attribute pid_attr = {
Parag Warudkar01e8ef12008-10-18 20:28:50 -0700408 .attr = { .name = "pid", .mode = S_IRUGO},
Paul Clements6b39bb62006-12-06 20:40:53 -0800409 .show = pid_show,
410};
411
Wanlong Gaof4507162012-03-28 14:42:51 -0700412static int nbd_do_it(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
414 struct request *req;
WANG Cong84963042007-05-09 02:33:36 -0700415 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Wanlong Gaof4507162012-03-28 14:42:51 -0700417 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Mel Gorman7f338fe2012-07-31 16:44:32 -0700419 sk_set_memalloc(nbd->sock->sk);
Wanlong Gaof4507162012-03-28 14:42:51 -0700420 nbd->pid = task_pid_nr(current);
421 ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
WANG Cong84963042007-05-09 02:33:36 -0700422 if (ret) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700423 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
424 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700425 return ret;
426 }
Paul Clements6b39bb62006-12-06 20:40:53 -0800427
Wanlong Gaof4507162012-03-28 14:42:51 -0700428 while ((req = nbd_read_stat(nbd)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800430
Wanlong Gaof4507162012-03-28 14:42:51 -0700431 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
432 nbd->pid = 0;
WANG Cong84963042007-05-09 02:33:36 -0700433 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
Wanlong Gaof4507162012-03-28 14:42:51 -0700436static void nbd_clear_que(struct nbd_device *nbd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 struct request *req;
439
Wanlong Gaof4507162012-03-28 14:42:51 -0700440 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Herbert Xu4b2f0262006-01-06 00:09:47 -0800442 /*
Wanlong Gaof4507162012-03-28 14:42:51 -0700443 * Because we have set nbd->sock to NULL under the tx_lock, all
Herbert Xu4b2f0262006-01-06 00:09:47 -0800444 * modifications to the list must have completed by now. For
445 * the same reason, the active_req must be NULL.
446 *
447 * As a consequence, we don't need to take the spin lock while
448 * purging the list here.
449 */
Wanlong Gaof4507162012-03-28 14:42:51 -0700450 BUG_ON(nbd->sock);
451 BUG_ON(nbd->active_req);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800452
Wanlong Gaof4507162012-03-28 14:42:51 -0700453 while (!list_empty(&nbd->queue_head)) {
454 req = list_entry(nbd->queue_head.next, struct request,
Herbert Xu4b2f0262006-01-06 00:09:47 -0800455 queuelist);
456 list_del_init(&req->queuelist);
457 req->errors++;
458 nbd_end_request(req);
459 }
Paul Clementsfded4e02012-09-17 14:09:02 -0700460
461 while (!list_empty(&nbd->waiting_queue)) {
462 req = list_entry(nbd->waiting_queue.next, struct request,
463 queuelist);
464 list_del_init(&req->queuelist);
465 req->errors++;
466 nbd_end_request(req);
467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
Paul Clements7fdfd402007-10-16 23:27:37 -0700470
Wanlong Gaof4507162012-03-28 14:42:51 -0700471static void nbd_handle_req(struct nbd_device *nbd, struct request *req)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700472{
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200473 if (req->cmd_type != REQ_TYPE_FS)
Laurent Vivier48cf6062008-04-29 01:02:46 -0700474 goto error_out;
475
476 nbd_cmd(req) = NBD_CMD_READ;
477 if (rq_data_dir(req) == WRITE) {
Paul Clementsa336d292012-10-04 17:16:18 -0700478 if ((req->cmd_flags & REQ_DISCARD)) {
479 WARN_ON(!(nbd->flags & NBD_FLAG_SEND_TRIM));
480 nbd_cmd(req) = NBD_CMD_TRIM;
481 } else
482 nbd_cmd(req) = NBD_CMD_WRITE;
Paul Clements2f012502012-10-04 17:16:15 -0700483 if (nbd->flags & NBD_FLAG_READ_ONLY) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700484 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200485 "Write on read-only\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700486 goto error_out;
487 }
488 }
489
Alex Bligh75f187a2013-02-27 17:05:23 -0800490 if (req->cmd_flags & REQ_FLUSH) {
491 BUG_ON(unlikely(blk_rq_sectors(req)));
492 nbd_cmd(req) = NBD_CMD_FLUSH;
493 }
494
Laurent Vivier48cf6062008-04-29 01:02:46 -0700495 req->errors = 0;
496
Wanlong Gaof4507162012-03-28 14:42:51 -0700497 mutex_lock(&nbd->tx_lock);
498 if (unlikely(!nbd->sock)) {
499 mutex_unlock(&nbd->tx_lock);
500 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200501 "Attempted send on closed socket\n");
Pavel Machek15746fc2009-04-02 16:58:42 -0700502 goto error_out;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700503 }
504
Wanlong Gaof4507162012-03-28 14:42:51 -0700505 nbd->active_req = req;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700506
Wanlong Gaof4507162012-03-28 14:42:51 -0700507 if (nbd_send_req(nbd, req) != 0) {
508 dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
Laurent Vivier48cf6062008-04-29 01:02:46 -0700509 req->errors++;
510 nbd_end_request(req);
511 } else {
Wanlong Gaof4507162012-03-28 14:42:51 -0700512 spin_lock(&nbd->queue_lock);
Chetan Loke01ff5db2012-07-31 08:47:13 +0200513 list_add_tail(&req->queuelist, &nbd->queue_head);
Wanlong Gaof4507162012-03-28 14:42:51 -0700514 spin_unlock(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700515 }
516
Wanlong Gaof4507162012-03-28 14:42:51 -0700517 nbd->active_req = NULL;
518 mutex_unlock(&nbd->tx_lock);
519 wake_up_all(&nbd->active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700520
521 return;
522
523error_out:
524 req->errors++;
525 nbd_end_request(req);
526}
527
528static int nbd_thread(void *data)
529{
Wanlong Gaof4507162012-03-28 14:42:51 -0700530 struct nbd_device *nbd = data;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700531 struct request *req;
532
Dongsheng Yang8698a742014-03-11 18:09:12 +0800533 set_user_nice(current, MIN_NICE);
Wanlong Gaof4507162012-03-28 14:42:51 -0700534 while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) {
Laurent Vivier48cf6062008-04-29 01:02:46 -0700535 /* wait for something to do */
Wanlong Gaof4507162012-03-28 14:42:51 -0700536 wait_event_interruptible(nbd->waiting_wq,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700537 kthread_should_stop() ||
Wanlong Gaof4507162012-03-28 14:42:51 -0700538 !list_empty(&nbd->waiting_queue));
Laurent Vivier48cf6062008-04-29 01:02:46 -0700539
540 /* extract request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700541 if (list_empty(&nbd->waiting_queue))
Laurent Vivier48cf6062008-04-29 01:02:46 -0700542 continue;
543
Wanlong Gaof4507162012-03-28 14:42:51 -0700544 spin_lock_irq(&nbd->queue_lock);
545 req = list_entry(nbd->waiting_queue.next, struct request,
Laurent Vivier48cf6062008-04-29 01:02:46 -0700546 queuelist);
547 list_del_init(&req->queuelist);
Wanlong Gaof4507162012-03-28 14:42:51 -0700548 spin_unlock_irq(&nbd->queue_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700549
550 /* handle request */
Wanlong Gaof4507162012-03-28 14:42:51 -0700551 nbd_handle_req(nbd, req);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700552 }
553 return 0;
554}
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556/*
557 * We always wait for result of write, for now. It would be nice to make it optional
558 * in future
Wanlong Gaof4507162012-03-28 14:42:51 -0700559 * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
561 */
562
Pavel Machek15746fc2009-04-02 16:58:42 -0700563static void do_nbd_request(struct request_queue *q)
Alex Elder398eb082013-02-27 17:05:28 -0800564 __releases(q->queue_lock) __acquires(q->queue_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
566 struct request *req;
567
Tejun Heo9934c8c2009-05-08 11:54:16 +0900568 while ((req = blk_fetch_request(q)) != NULL) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700569 struct nbd_device *nbd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Laurent Vivier48cf6062008-04-29 01:02:46 -0700571 spin_unlock_irq(q->queue_lock);
572
Jens Axboe4aff5e22006-08-10 08:44:47 +0200573 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
574 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
Wanlong Gaof4507162012-03-28 14:42:51 -0700576 nbd = req->rq_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Wanlong Gaof4507162012-03-28 14:42:51 -0700578 BUG_ON(nbd->magic != NBD_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Wanlong Gaof4507162012-03-28 14:42:51 -0700580 if (unlikely(!nbd->sock)) {
581 dev_err(disk_to_dev(nbd->disk),
WANG Cong7f1b90f2011-08-19 14:48:22 +0200582 "Attempted send on closed socket\n");
Paul Clements4d48a542009-02-11 13:04:45 -0800583 req->errors++;
584 nbd_end_request(req);
585 spin_lock_irq(q->queue_lock);
586 continue;
587 }
588
Wanlong Gaof4507162012-03-28 14:42:51 -0700589 spin_lock_irq(&nbd->queue_lock);
590 list_add_tail(&req->queuelist, &nbd->waiting_queue);
591 spin_unlock_irq(&nbd->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Wanlong Gaof4507162012-03-28 14:42:51 -0700593 wake_up(&nbd->waiting_wq);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Pavel Machek1a2ad212009-04-02 16:58:41 -0700599/* Must be called with tx_lock held */
600
Wanlong Gaof4507162012-03-28 14:42:51 -0700601static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
Pavel Machek1a2ad212009-04-02 16:58:41 -0700602 unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 switch (cmd) {
Pavel Machek1a2ad212009-04-02 16:58:41 -0700605 case NBD_DISCONNECT: {
606 struct request sreq;
607
Wanlong Gaof4507162012-03-28 14:42:51 -0700608 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800609 if (!nbd->sock)
610 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700611
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800612 mutex_unlock(&nbd->tx_lock);
613 fsync_bdev(bdev);
614 mutex_lock(&nbd->tx_lock);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +0200615 blk_rq_init(NULL, &sreq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200616 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 nbd_cmd(&sreq) = NBD_CMD_DISC;
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800618
619 /* Check again after getting mutex back. */
Wanlong Gaof4507162012-03-28 14:42:51 -0700620 if (!nbd->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return -EINVAL;
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800622
Paul Clementsc378f702013-07-03 15:09:04 -0700623 nbd->disconnect = 1;
624
Wanlong Gaof4507162012-03-28 14:42:51 -0700625 nbd_send_req(nbd, &sreq);
Paul Clementsc378f702013-07-03 15:09:04 -0700626 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Pavel Machek1a2ad212009-04-02 16:58:41 -0700629 case NBD_CLEAR_SOCK: {
Al Viroe2511572014-03-05 20:41:36 -0500630 struct socket *sock = nbd->sock;
Wanlong Gaof4507162012-03-28 14:42:51 -0700631 nbd->sock = NULL;
Wanlong Gaof4507162012-03-28 14:42:51 -0700632 nbd_clear_que(nbd);
633 BUG_ON(!list_empty(&nbd->queue_head));
Paul Clementsfded4e02012-09-17 14:09:02 -0700634 BUG_ON(!list_empty(&nbd->waiting_queue));
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800635 kill_bdev(bdev);
Al Viroe2511572014-03-05 20:41:36 -0500636 if (sock)
637 sockfd_put(sock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700638 return 0;
639 }
640
641 case NBD_SET_SOCK: {
Al Viroe2511572014-03-05 20:41:36 -0500642 struct socket *sock;
643 int err;
644 if (nbd->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return -EBUSY;
Al Viroe2511572014-03-05 20:41:36 -0500646 sock = sockfd_lookup(arg, &err);
647 if (sock) {
648 nbd->sock = sock;
649 if (max_part > 0)
650 bdev->bd_invalidated = 1;
651 nbd->disconnect = 0; /* we're connected now */
652 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700654 return -EINVAL;
655 }
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 case NBD_SET_BLKSIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700658 nbd->blksize = arg;
659 nbd->bytesize &= ~(nbd->blksize-1);
660 bdev->bd_inode->i_size = nbd->bytesize;
661 set_blocksize(bdev, nbd->blksize);
662 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 case NBD_SET_SIZE:
Wanlong Gaof4507162012-03-28 14:42:51 -0700666 nbd->bytesize = arg & ~(nbd->blksize-1);
667 bdev->bd_inode->i_size = nbd->bytesize;
668 set_blocksize(bdev, nbd->blksize);
669 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700671
Paul Clements7fdfd402007-10-16 23:27:37 -0700672 case NBD_SET_TIMEOUT:
Wanlong Gaof4507162012-03-28 14:42:51 -0700673 nbd->xmit_timeout = arg * HZ;
Paul Clements7fdfd402007-10-16 23:27:37 -0700674 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700675
Paul Clements2f012502012-10-04 17:16:15 -0700676 case NBD_SET_FLAGS:
677 nbd->flags = arg;
678 return 0;
679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 case NBD_SET_SIZE_BLOCKS:
Wanlong Gaof4507162012-03-28 14:42:51 -0700681 nbd->bytesize = ((u64) arg) * nbd->blksize;
682 bdev->bd_inode->i_size = nbd->bytesize;
683 set_blocksize(bdev, nbd->blksize);
684 set_capacity(nbd->disk, nbd->bytesize >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700686
687 case NBD_DO_IT: {
688 struct task_struct *thread;
Al Viroe2511572014-03-05 20:41:36 -0500689 struct socket *sock;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700690 int error;
691
Wanlong Gaof4507162012-03-28 14:42:51 -0700692 if (nbd->pid)
Pavel Machekc91192d2009-01-15 13:51:03 -0800693 return -EBUSY;
Al Viroe2511572014-03-05 20:41:36 -0500694 if (!nbd->sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return -EINVAL;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700696
Wanlong Gaof4507162012-03-28 14:42:51 -0700697 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700698
Paolo Bonzinia83e8142013-02-27 17:05:26 -0800699 if (nbd->flags & NBD_FLAG_READ_ONLY)
700 set_device_ro(bdev, true);
Paul Clementsa336d292012-10-04 17:16:18 -0700701 if (nbd->flags & NBD_FLAG_SEND_TRIM)
702 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
703 nbd->disk->queue);
Alex Bligh75f187a2013-02-27 17:05:23 -0800704 if (nbd->flags & NBD_FLAG_SEND_FLUSH)
705 blk_queue_flush(nbd->disk->queue, REQ_FLUSH);
706 else
707 blk_queue_flush(nbd->disk->queue, 0);
Paul Clementsa336d292012-10-04 17:16:18 -0700708
Kees Cookffc8b302013-07-03 15:01:14 -0700709 thread = kthread_create(nbd_thread, nbd, "%s",
710 nbd->disk->disk_name);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700711 if (IS_ERR(thread)) {
Wanlong Gaof4507162012-03-28 14:42:51 -0700712 mutex_lock(&nbd->tx_lock);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700713 return PTR_ERR(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700714 }
Laurent Vivier48cf6062008-04-29 01:02:46 -0700715 wake_up_process(thread);
Wanlong Gaof4507162012-03-28 14:42:51 -0700716 error = nbd_do_it(nbd);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700717 kthread_stop(thread);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700718
Wanlong Gaof4507162012-03-28 14:42:51 -0700719 mutex_lock(&nbd->tx_lock);
WANG Cong84963042007-05-09 02:33:36 -0700720 if (error)
721 return error;
Wanlong Gaof4507162012-03-28 14:42:51 -0700722 sock_shutdown(nbd, 0);
Al Viroe2511572014-03-05 20:41:36 -0500723 sock = nbd->sock;
724 nbd->sock = NULL;
Wanlong Gaof4507162012-03-28 14:42:51 -0700725 nbd_clear_que(nbd);
726 dev_warn(disk_to_dev(nbd->disk), "queue cleared\n");
Paolo Bonzini3a2d63f82013-02-27 17:05:25 -0800727 kill_bdev(bdev);
Paul Clementsa336d292012-10-04 17:16:18 -0700728 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
Paolo Bonzinia83e8142013-02-27 17:05:26 -0800729 set_device_ro(bdev, false);
Al Viroe2511572014-03-05 20:41:36 -0500730 if (sock)
731 sockfd_put(sock);
Alex Bligh75f187a2013-02-27 17:05:23 -0800732 nbd->flags = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700733 nbd->bytesize = 0;
Al Viroa8cdc302008-03-02 09:33:33 -0500734 bdev->bd_inode->i_size = 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700735 set_capacity(nbd->disk, 0);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700736 if (max_part > 0)
Al Viroa8cdc302008-03-02 09:33:33 -0500737 ioctl_by_bdev(bdev, BLKRRPART, 0);
Paul Clementsc378f702013-07-03 15:09:04 -0700738 if (nbd->disconnect) /* user requested, ignore socket errors */
739 return 0;
Wanlong Gaof4507162012-03-28 14:42:51 -0700740 return nbd->harderror;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700741 }
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800744 /*
745 * This is for compatibility only. The queue is always cleared
746 * by NBD_DO_IT or NBD_CLEAR_SOCK.
747 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 return 0;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 case NBD_PRINT_DEBUG:
Wanlong Gaof4507162012-03-28 14:42:51 -0700751 dev_info(disk_to_dev(nbd->disk),
WANG Cong5eedf542011-08-19 14:48:28 +0200752 "next = %p, prev = %p, head = %p\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700753 nbd->queue_head.next, nbd->queue_head.prev,
754 &nbd->queue_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 return 0;
756 }
Pavel Machek1a2ad212009-04-02 16:58:41 -0700757 return -ENOTTY;
758}
759
760static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
761 unsigned int cmd, unsigned long arg)
762{
Wanlong Gaof4507162012-03-28 14:42:51 -0700763 struct nbd_device *nbd = bdev->bd_disk->private_data;
Pavel Machek1a2ad212009-04-02 16:58:41 -0700764 int error;
765
766 if (!capable(CAP_SYS_ADMIN))
767 return -EPERM;
768
Wanlong Gaof4507162012-03-28 14:42:51 -0700769 BUG_ON(nbd->magic != NBD_MAGIC);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700770
771 /* Anyone capable of this syscall can do *real bad* things */
772 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
Wanlong Gaof4507162012-03-28 14:42:51 -0700773 nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700774
Wanlong Gaof4507162012-03-28 14:42:51 -0700775 mutex_lock(&nbd->tx_lock);
776 error = __nbd_ioctl(bdev, nbd, cmd, arg);
777 mutex_unlock(&nbd->tx_lock);
Pavel Machek1a2ad212009-04-02 16:58:41 -0700778
779 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700782static const struct block_device_operations nbd_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
784 .owner = THIS_MODULE,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +0200785 .ioctl = nbd_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786};
787
788/*
789 * And here should be modules and kernel interface
790 * (Just smiley confuses emacs :-)
791 */
792
793static int __init nbd_init(void)
794{
795 int err = -ENOMEM;
796 int i;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700797 int part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800799 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700801 if (max_part < 0) {
WANG Cong7742ce42011-08-19 14:48:28 +0200802 printk(KERN_ERR "nbd: max_part must be >= 0\n");
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700803 return -EINVAL;
804 }
805
Sven Wegenerf3944d62008-08-20 14:09:07 -0700806 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
807 if (!nbd_dev)
808 return -ENOMEM;
809
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700810 part_shift = 0;
Namhyung Kim5988ce22011-05-28 14:44:46 +0200811 if (max_part > 0) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700812 part_shift = fls(max_part);
813
Namhyung Kim5988ce22011-05-28 14:44:46 +0200814 /*
815 * Adjust max_part according to part_shift as it is exported
816 * to user space so that user can know the max number of
817 * partition kernel should be able to manage.
818 *
819 * Note that -1 is required because partition 0 is reserved
820 * for the whole disk.
821 */
822 max_part = (1UL << part_shift) - 1;
823 }
824
Namhyung Kim3b271082011-05-28 14:44:46 +0200825 if ((1UL << part_shift) > DISK_MAX_PARTS)
826 return -EINVAL;
827
828 if (nbds_max > 1UL << (MINORBITS - part_shift))
829 return -EINVAL;
830
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700831 for (i = 0; i < nbds_max; i++) {
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700832 struct gendisk *disk = alloc_disk(1 << part_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!disk)
834 goto out;
835 nbd_dev[i].disk = disk;
836 /*
837 * The new linux 2.5 block layer implementation requires
838 * every gendisk to have its very own request_queue struct.
839 * These structs are big so we dynamically allocate them.
840 */
841 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
842 if (!disk->queue) {
843 put_disk(disk);
844 goto out;
845 }
Jens Axboe31dcfab2008-10-31 10:06:37 +0100846 /*
847 * Tell the block layer that we are not a rotational device
848 */
849 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
Paul Clementsa336d292012-10-04 17:16:18 -0700850 disk->queue->limits.discard_granularity = 512;
851 disk->queue->limits.max_discard_sectors = UINT_MAX;
852 disk->queue->limits.discard_zeroes_data = 0;
Michal Belczyk078be022013-04-30 15:28:28 -0700853 blk_queue_max_hw_sectors(disk->queue, 65536);
854 disk->queue->limits.max_sectors = 256;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856
857 if (register_blkdev(NBD_MAJOR, "nbd")) {
858 err = -EIO;
859 goto out;
860 }
861
862 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
863 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
864
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700865 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 struct gendisk *disk = nbd_dev[i].disk;
Wanlong Gaof4507162012-03-28 14:42:51 -0700867 nbd_dev[i].magic = NBD_MAGIC;
Laurent Vivier48cf6062008-04-29 01:02:46 -0700868 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 spin_lock_init(&nbd_dev[i].queue_lock);
870 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800871 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800872 init_waitqueue_head(&nbd_dev[i].active_wq);
Laurent Vivier48cf6062008-04-29 01:02:46 -0700873 init_waitqueue_head(&nbd_dev[i].waiting_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 nbd_dev[i].blksize = 1024;
Paul Clements4b86a872007-10-16 23:27:36 -0700875 nbd_dev[i].bytesize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 disk->major = NBD_MAJOR;
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700877 disk->first_minor = i << part_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 disk->fops = &nbd_fops;
879 disk->private_data = &nbd_dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 sprintf(disk->disk_name, "nbd%d", i);
Paul Clements4b86a872007-10-16 23:27:36 -0700881 set_capacity(disk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 add_disk(disk);
883 }
884
885 return 0;
886out:
887 while (i--) {
888 blk_cleanup_queue(nbd_dev[i].disk->queue);
889 put_disk(nbd_dev[i].disk);
890 }
Sven Wegenerf3944d62008-08-20 14:09:07 -0700891 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 return err;
893}
894
895static void __exit nbd_cleanup(void)
896{
897 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700898 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700900 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 if (disk) {
902 del_gendisk(disk);
903 blk_cleanup_queue(disk->queue);
904 put_disk(disk);
905 }
906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 unregister_blkdev(NBD_MAJOR, "nbd");
Sven Wegenerf3944d62008-08-20 14:09:07 -0700908 kfree(nbd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
910}
911
912module_init(nbd_init);
913module_exit(nbd_cleanup);
914
915MODULE_DESCRIPTION("Network Block Device");
916MODULE_LICENSE("GPL");
917
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700918module_param(nbds_max, int, 0444);
Laurent Vivierd71a6d72008-04-29 01:02:51 -0700919MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
920module_param(max_part, int, 0444);
921MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922#ifndef NDEBUG
923module_param(debugflags, int, 0644);
924MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
925#endif