blob: 6d854d3c83658ea6070bd450002e57f822fa29bc [file] [log] [blame]
thscd831bd2008-07-03 10:23:51 +00001/*
bellard7a5ca862008-05-27 21:13:40 +00002 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
3 *
4 * Network Block Device
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 of the License.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
Blue Swirl8167ee82009-07-16 20:47:01 +000016 * along with this program; if not, see <http://www.gnu.org/licenses/>.
bellard7a5ca862008-05-27 21:13:40 +000017 */
18
19#include <qemu-common.h>
20#include "block_int.h"
21#include "nbd.h"
22
bellard7a5ca862008-05-27 21:13:40 +000023#include <stdarg.h>
24#include <stdio.h>
25#include <getopt.h>
26#include <err.h>
thscd831bd2008-07-03 10:23:51 +000027#include <sys/types.h>
bellard7a5ca862008-05-27 21:13:40 +000028#include <sys/socket.h>
29#include <netinet/in.h>
30#include <netinet/tcp.h>
31#include <arpa/inet.h>
thscd831bd2008-07-03 10:23:51 +000032#include <signal.h>
Blue Swirl2bff4b62009-12-23 15:34:04 +000033#include <libgen.h>
thscd831bd2008-07-03 10:23:51 +000034
35#define SOCKET_PATH "/var/lock/qemu-nbd-%s"
bellard7a5ca862008-05-27 21:13:40 +000036
ths2f726482008-07-03 11:47:46 +000037#define NBD_BUFFER_SIZE (1024*1024)
38
blueswir1b1d8e522008-10-26 13:43:07 +000039static int verbose;
bellard7a5ca862008-05-27 21:13:40 +000040
41static void usage(const char *name)
42{
43 printf(
44"Usage: %s [OPTIONS] FILE\n"
45"QEMU Disk Network Block Device Server\n"
46"\n"
47" -p, --port=PORT port to listen on (default `1024')\n"
48" -o, --offset=OFFSET offset into the image\n"
49" -b, --bind=IFACE interface to bind to (default `0.0.0.0')\n"
thscd831bd2008-07-03 10:23:51 +000050" -k, --socket=PATH path to the unix socket\n"
51" (default '"SOCKET_PATH"')\n"
bellard7a5ca862008-05-27 21:13:40 +000052" -r, --read-only export read-only\n"
53" -P, --partition=NUM only expose partition NUM\n"
ths2f726482008-07-03 11:47:46 +000054" -s, --snapshot use snapshot file\n"
55" -n, --nocache disable host cache\n"
thscd831bd2008-07-03 10:23:51 +000056" -c, --connect=DEV connect FILE to the local NBD device DEV\n"
57" -d, --disconnect disconnect the specified device\n"
ths3b05a8e2008-07-03 12:45:02 +000058" -e, --shared=NUM device can be shared by NUM clients (default '1')\n"
ths75818252008-07-03 13:41:03 +000059" -t, --persistent don't exit on the last connection\n"
bellard7a5ca862008-05-27 21:13:40 +000060" -v, --verbose display extra debugging information\n"
61" -h, --help display this help and exit\n"
62" -V, --version output version information and exit\n"
63"\n"
64"Report bugs to <anthony@codemonkey.ws>\n"
thscd831bd2008-07-03 10:23:51 +000065 , name, "DEVICE");
bellard7a5ca862008-05-27 21:13:40 +000066}
67
68static void version(const char *name)
69{
70 printf(
ths315bc7a2008-07-18 18:06:23 +000071"%s version 0.0.1\n"
bellard7a5ca862008-05-27 21:13:40 +000072"Written by Anthony Liguori.\n"
73"\n"
74"Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
75"This is free software; see the source for copying conditions. There is NO\n"
76"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
ths315bc7a2008-07-18 18:06:23 +000077 , name);
bellard7a5ca862008-05-27 21:13:40 +000078}
79
80struct partition_record
81{
82 uint8_t bootable;
83 uint8_t start_head;
84 uint32_t start_cylinder;
85 uint8_t start_sector;
86 uint8_t system;
87 uint8_t end_head;
88 uint8_t end_cylinder;
89 uint8_t end_sector;
90 uint32_t start_sector_abs;
91 uint32_t nb_sectors_abs;
92};
93
94static void read_partition(uint8_t *p, struct partition_record *r)
95{
96 r->bootable = p[0];
97 r->start_head = p[1];
98 r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
99 r->start_sector = p[2] & 0x3f;
100 r->system = p[4];
101 r->end_head = p[5];
102 r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
103 r->end_sector = p[6] & 0x3f;
104 r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
105 r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
106}
107
108static int find_partition(BlockDriverState *bs, int partition,
109 off_t *offset, off_t *size)
110{
111 struct partition_record mbr[4];
112 uint8_t data[512];
113 int i;
114 int ext_partnum = 4;
115
116 if (bdrv_read(bs, 0, data, 1))
Ryota Ozakib6353be2010-03-20 15:23:23 +0900117 errx(EXIT_FAILURE, "error while reading");
bellard7a5ca862008-05-27 21:13:40 +0000118
119 if (data[510] != 0x55 || data[511] != 0xaa) {
120 errno = -EINVAL;
121 return -1;
122 }
123
124 for (i = 0; i < 4; i++) {
125 read_partition(&data[446 + 16 * i], &mbr[i]);
126
127 if (!mbr[i].nb_sectors_abs)
128 continue;
129
130 if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
131 struct partition_record ext[4];
132 uint8_t data1[512];
133 int j;
134
135 if (bdrv_read(bs, mbr[i].start_sector_abs, data1, 1))
Ryota Ozakib6353be2010-03-20 15:23:23 +0900136 errx(EXIT_FAILURE, "error while reading");
bellard7a5ca862008-05-27 21:13:40 +0000137
138 for (j = 0; j < 4; j++) {
139 read_partition(&data1[446 + 16 * j], &ext[j]);
140 if (!ext[j].nb_sectors_abs)
141 continue;
142
143 if ((ext_partnum + j + 1) == partition) {
144 *offset = (uint64_t)ext[j].start_sector_abs << 9;
145 *size = (uint64_t)ext[j].nb_sectors_abs << 9;
146 return 0;
147 }
148 }
149 ext_partnum += 4;
150 } else if ((i + 1) == partition) {
151 *offset = (uint64_t)mbr[i].start_sector_abs << 9;
152 *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
153 return 0;
154 }
155 }
156
157 errno = -ENOENT;
158 return -1;
159}
160
thscd831bd2008-07-03 10:23:51 +0000161static void show_parts(const char *device)
162{
163 if (fork() == 0) {
164 int nbd;
165
166 /* linux just needs an open() to trigger
167 * the partition table update
168 * but remember to load the module with max_part != 0 :
169 * modprobe nbd max_part=63
170 */
171 nbd = open(device, O_RDWR);
172 if (nbd != -1)
173 close(nbd);
174 exit(0);
175 }
176}
177
bellard7a5ca862008-05-27 21:13:40 +0000178int main(int argc, char **argv)
179{
180 BlockDriverState *bs;
181 off_t dev_offset = 0;
182 off_t offset = 0;
183 bool readonly = false;
thscd831bd2008-07-03 10:23:51 +0000184 bool disconnect = false;
bellard7a5ca862008-05-27 21:13:40 +0000185 const char *bindto = "0.0.0.0";
186 int port = 1024;
bellard7a5ca862008-05-27 21:13:40 +0000187 struct sockaddr_in addr;
188 socklen_t addr_len = sizeof(addr);
189 off_t fd_size;
thscd831bd2008-07-03 10:23:51 +0000190 char *device = NULL;
191 char *socket = NULL;
192 char sockpath[128];
aliguorid6aa6712009-01-08 19:34:35 +0000193 const char *sopt = "hVb:o:p:rsnP:c:dvk:e:t";
bellard7a5ca862008-05-27 21:13:40 +0000194 struct option lopt[] = {
Blue Swirl660f11b2009-07-31 21:16:51 +0000195 { "help", 0, NULL, 'h' },
196 { "version", 0, NULL, 'V' },
197 { "bind", 1, NULL, 'b' },
198 { "port", 1, NULL, 'p' },
199 { "socket", 1, NULL, 'k' },
200 { "offset", 1, NULL, 'o' },
201 { "read-only", 0, NULL, 'r' },
202 { "partition", 1, NULL, 'P' },
203 { "connect", 1, NULL, 'c' },
204 { "disconnect", 0, NULL, 'd' },
205 { "snapshot", 0, NULL, 's' },
206 { "nocache", 0, NULL, 'n' },
207 { "shared", 1, NULL, 'e' },
208 { "persistent", 0, NULL, 't' },
209 { "verbose", 0, NULL, 'v' },
210 { NULL, 0, NULL, 0 }
bellard7a5ca862008-05-27 21:13:40 +0000211 };
212 int ch;
213 int opt_ind = 0;
214 int li;
215 char *end;
Naphtali Spreif5edb012010-01-17 16:48:13 +0200216 int flags = BDRV_O_RDWR;
bellard7a5ca862008-05-27 21:13:40 +0000217 int partition = -1;
thscd831bd2008-07-03 10:23:51 +0000218 int ret;
ths3b05a8e2008-07-03 12:45:02 +0000219 int shared = 1;
ths2f726482008-07-03 11:47:46 +0000220 uint8_t *data;
ths3b05a8e2008-07-03 12:45:02 +0000221 fd_set fds;
222 int *sharing_fds;
223 int fd;
224 int i;
225 int nb_fds = 0;
226 int max_fd;
ths75818252008-07-03 13:41:03 +0000227 int persistent = 0;
bellard7a5ca862008-05-27 21:13:40 +0000228
229 while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
230 switch (ch) {
231 case 's':
ths2f726482008-07-03 11:47:46 +0000232 flags |= BDRV_O_SNAPSHOT;
233 break;
234 case 'n':
aliguori9f7965c2008-10-14 14:42:54 +0000235 flags |= BDRV_O_NOCACHE;
bellard7a5ca862008-05-27 21:13:40 +0000236 break;
237 case 'b':
238 bindto = optarg;
239 break;
240 case 'p':
241 li = strtol(optarg, &end, 0);
242 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900243 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000244 }
245 if (li < 1 || li > 65535) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900246 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000247 }
248 port = (uint16_t)li;
249 break;
250 case 'o':
251 dev_offset = strtoll (optarg, &end, 0);
252 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900253 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000254 }
255 if (dev_offset < 0) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900256 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000257 }
258 break;
259 case 'r':
260 readonly = true;
Naphtali Sprei07108b22010-03-14 15:19:57 +0200261 flags &= ~BDRV_O_RDWR;
bellard7a5ca862008-05-27 21:13:40 +0000262 break;
263 case 'P':
264 partition = strtol(optarg, &end, 0);
265 if (*end)
Ryota Ozakib6353be2010-03-20 15:23:23 +0900266 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000267 if (partition < 1 || partition > 8)
Ryota Ozakib6353be2010-03-20 15:23:23 +0900268 errx(EXIT_FAILURE, "Invalid partition %d", partition);
bellard7a5ca862008-05-27 21:13:40 +0000269 break;
thscd831bd2008-07-03 10:23:51 +0000270 case 'k':
271 socket = optarg;
272 if (socket[0] != '/')
Ryota Ozakib6353be2010-03-20 15:23:23 +0900273 errx(EXIT_FAILURE, "socket path must be absolute\n");
thscd831bd2008-07-03 10:23:51 +0000274 break;
275 case 'd':
276 disconnect = true;
277 break;
278 case 'c':
279 device = optarg;
280 break;
ths3b05a8e2008-07-03 12:45:02 +0000281 case 'e':
282 shared = strtol(optarg, &end, 0);
283 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900284 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
ths3b05a8e2008-07-03 12:45:02 +0000285 }
286 if (shared < 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900287 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
ths3b05a8e2008-07-03 12:45:02 +0000288 }
289 break;
ths75818252008-07-03 13:41:03 +0000290 case 't':
291 persistent = 1;
292 break;
bellard7a5ca862008-05-27 21:13:40 +0000293 case 'v':
294 verbose = 1;
295 break;
296 case 'V':
297 version(argv[0]);
298 exit(0);
299 break;
300 case 'h':
301 usage(argv[0]);
302 exit(0);
303 break;
304 case '?':
Ryota Ozakib6353be2010-03-20 15:23:23 +0900305 errx(EXIT_FAILURE, "Try `%s --help' for more information.",
bellard7a5ca862008-05-27 21:13:40 +0000306 argv[0]);
307 }
308 }
309
310 if ((argc - optind) != 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900311 errx(EXIT_FAILURE, "Invalid number of argument.\n"
bellard7a5ca862008-05-27 21:13:40 +0000312 "Try `%s --help' for more information.",
313 argv[0]);
314 }
315
thscd831bd2008-07-03 10:23:51 +0000316 if (disconnect) {
317 fd = open(argv[optind], O_RDWR);
318 if (fd == -1)
Ryota Ozakib6353be2010-03-20 15:23:23 +0900319 errx(EXIT_FAILURE, "Cannot open %s", argv[optind]);
thscd831bd2008-07-03 10:23:51 +0000320
321 nbd_disconnect(fd);
322
323 close(fd);
324
325 printf("%s disconnected\n", argv[optind]);
326
327 return 0;
328 }
329
bellard7a5ca862008-05-27 21:13:40 +0000330 bdrv_init();
331
332 bs = bdrv_new("hda");
333 if (bs == NULL)
334 return 1;
335
Ryota Ozakia16c1742010-03-20 15:23:22 +0900336 if (bdrv_open(bs, argv[optind], flags) < 0)
bellard7a5ca862008-05-27 21:13:40 +0000337 return 1;
338
339 fd_size = bs->total_sectors * 512;
340
341 if (partition != -1 &&
342 find_partition(bs, partition, &dev_offset, &fd_size))
Ryota Ozakib6353be2010-03-20 15:23:23 +0900343 errx(EXIT_FAILURE, "Could not find partition %d", partition);
bellard7a5ca862008-05-27 21:13:40 +0000344
thscd831bd2008-07-03 10:23:51 +0000345 if (device) {
ths3b05a8e2008-07-03 12:45:02 +0000346 pid_t pid;
347 int sock;
348
Anthony Liguorif5de1412009-06-15 12:49:59 -0500349 if (!verbose) {
350 /* detach client and server */
351 if (daemon(0, 0) == -1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900352 errx(EXIT_FAILURE, "Failed to daemonize");
Anthony Liguorif5de1412009-06-15 12:49:59 -0500353 }
354 }
thscd831bd2008-07-03 10:23:51 +0000355
356 if (socket == NULL) {
Blue Swirl22ff51e2009-12-23 15:45:30 +0000357 snprintf(sockpath, sizeof(sockpath), SOCKET_PATH,
358 basename(device));
thscd831bd2008-07-03 10:23:51 +0000359 socket = sockpath;
360 }
361
362 pid = fork();
363 if (pid < 0)
364 return 1;
365 if (pid != 0) {
366 off_t size;
367 size_t blocksize;
368
369 ret = 0;
370 bdrv_close(bs);
371
372 do {
373 sock = unix_socket_outgoing(socket);
374 if (sock == -1) {
375 if (errno != ENOENT && errno != ECONNREFUSED)
376 goto out;
377 sleep(1); /* wait children */
378 }
379 } while (sock == -1);
380
381 fd = open(device, O_RDWR);
382 if (fd == -1) {
383 ret = 1;
384 goto out;
385 }
386
387 ret = nbd_receive_negotiate(sock, &size, &blocksize);
388 if (ret == -1) {
389 ret = 1;
390 goto out;
391 }
392
393 ret = nbd_init(fd, sock, size, blocksize);
394 if (ret == -1) {
395 ret = 1;
396 goto out;
397 }
398
399 printf("NBD device %s is now connected to file %s\n",
400 device, argv[optind]);
401
402 /* update partition table */
403
404 show_parts(device);
405
406 nbd_client(fd, sock);
407 close(fd);
408 out:
409 kill(pid, SIGTERM);
410 unlink(socket);
411
412 return ret;
413 }
414 /* children */
415 }
416
ths3b05a8e2008-07-03 12:45:02 +0000417 sharing_fds = qemu_malloc((shared + 1) * sizeof(int));
ths3b05a8e2008-07-03 12:45:02 +0000418
thscd831bd2008-07-03 10:23:51 +0000419 if (socket) {
ths3b05a8e2008-07-03 12:45:02 +0000420 sharing_fds[0] = unix_socket_incoming(socket);
thscd831bd2008-07-03 10:23:51 +0000421 } else {
ths3b05a8e2008-07-03 12:45:02 +0000422 sharing_fds[0] = tcp_socket_incoming(bindto, port);
thscd831bd2008-07-03 10:23:51 +0000423 }
424
ths3b05a8e2008-07-03 12:45:02 +0000425 if (sharing_fds[0] == -1)
bellard7a5ca862008-05-27 21:13:40 +0000426 return 1;
ths3b05a8e2008-07-03 12:45:02 +0000427 max_fd = sharing_fds[0];
428 nb_fds++;
bellard7a5ca862008-05-27 21:13:40 +0000429
ths2f726482008-07-03 11:47:46 +0000430 data = qemu_memalign(512, NBD_BUFFER_SIZE);
ths3b05a8e2008-07-03 12:45:02 +0000431 if (data == NULL)
Ryota Ozakib6353be2010-03-20 15:23:23 +0900432 errx(EXIT_FAILURE, "Cannot allocate data buffer");
ths3b05a8e2008-07-03 12:45:02 +0000433
434 do {
435
436 FD_ZERO(&fds);
437 for (i = 0; i < nb_fds; i++)
438 FD_SET(sharing_fds[i], &fds);
439
440 ret = select(max_fd + 1, &fds, NULL, NULL, NULL);
441 if (ret == -1)
442 break;
443
444 if (FD_ISSET(sharing_fds[0], &fds))
445 ret--;
446 for (i = 1; i < nb_fds && ret; i++) {
447 if (FD_ISSET(sharing_fds[i], &fds)) {
448 if (nbd_trip(bs, sharing_fds[i], fd_size, dev_offset,
449 &offset, readonly, data, NBD_BUFFER_SIZE) != 0) {
450 close(sharing_fds[i]);
451 nb_fds--;
452 sharing_fds[i] = sharing_fds[nb_fds];
453 i--;
454 }
455 ret--;
456 }
457 }
458 /* new connection ? */
459 if (FD_ISSET(sharing_fds[0], &fds)) {
460 if (nb_fds < shared + 1) {
461 sharing_fds[nb_fds] = accept(sharing_fds[0],
462 (struct sockaddr *)&addr,
463 &addr_len);
464 if (sharing_fds[nb_fds] != -1 &&
aliguori27982662008-09-10 15:23:19 +0000465 nbd_negotiate(sharing_fds[nb_fds], fd_size) != -1) {
ths3b05a8e2008-07-03 12:45:02 +0000466 if (sharing_fds[nb_fds] > max_fd)
467 max_fd = sharing_fds[nb_fds];
468 nb_fds++;
469 }
470 }
471 }
ths75818252008-07-03 13:41:03 +0000472 } while (persistent || nb_fds > 1);
Herve Poussineauf8a83242010-01-24 21:23:56 +0000473 qemu_vfree(data);
bellard7a5ca862008-05-27 21:13:40 +0000474
ths3b05a8e2008-07-03 12:45:02 +0000475 close(sharing_fds[0]);
bellard7a5ca862008-05-27 21:13:40 +0000476 bdrv_close(bs);
ths3b05a8e2008-07-03 12:45:02 +0000477 qemu_free(sharing_fds);
thscd831bd2008-07-03 10:23:51 +0000478 if (socket)
479 unlink(socket);
bellard7a5ca862008-05-27 21:13:40 +0000480
481 return 0;
482}