blob: 9518b75a3a0f2cc19ac253f1dcad4319faef6a28 [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
Stefan Weil5a61cb62011-09-08 17:55:32 +020019#include "qemu-common.h"
Markus Armbruster26f54e92014-10-07 13:59:04 +020020#include "sysemu/block-backend.h"
Peter Lievenb3838a42014-08-13 19:20:18 +020021#include "block/block_int.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010022#include "block/nbd.h"
Alex Bligh6a1751b2013-08-21 16:02:47 +010023#include "qemu/main-loop.h"
Paolo Bonzini537b41f2014-02-17 14:43:51 +010024#include "qemu/sockets.h"
25#include "qemu/error-report.h"
Wenchao Xia8c116b02013-12-04 17:10:55 +080026#include "block/snapshot.h"
Peter Lievenb3838a42014-08-13 19:20:18 +020027#include "qapi/util.h"
Markus Armbrusterd49b6832015-03-17 18:29:20 +010028#include "qapi/qmp/qstring.h"
bellard7a5ca862008-05-27 21:13:40 +000029
bellard7a5ca862008-05-27 21:13:40 +000030#include <stdarg.h>
31#include <stdio.h>
32#include <getopt.h>
33#include <err.h>
thscd831bd2008-07-03 10:23:51 +000034#include <sys/types.h>
bellard7a5ca862008-05-27 21:13:40 +000035#include <sys/socket.h>
36#include <netinet/in.h>
37#include <netinet/tcp.h>
38#include <arpa/inet.h>
thscd831bd2008-07-03 10:23:51 +000039#include <signal.h>
Blue Swirl2bff4b62009-12-23 15:34:04 +000040#include <libgen.h>
Paolo Bonzinia517e882011-11-04 15:51:21 +010041#include <pthread.h>
thscd831bd2008-07-03 10:23:51 +000042
Peter Lieven713cc672014-08-13 19:20:19 +020043#define SOCKET_PATH "/var/lock/qemu-nbd-%s"
44#define QEMU_NBD_OPT_CACHE 1
45#define QEMU_NBD_OPT_AIO 2
46#define QEMU_NBD_OPT_DISCARD 3
Peter Lievenb3838a42014-08-13 19:20:18 +020047#define QEMU_NBD_OPT_DETECT_ZEROES 4
bellard7a5ca862008-05-27 21:13:40 +000048
Paolo Bonziniaf49bbb2011-09-19 14:03:37 +020049static NBDExport *exp;
blueswir1b1d8e522008-10-26 13:43:07 +000050static int verbose;
Paolo Bonzinia517e882011-11-04 15:51:21 +010051static char *srcpath;
52static char *sockpath;
Paolo Bonzini7860a382012-09-18 13:31:56 +020053static int persistent = 0;
54static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
Paolo Bonzinia61c6782011-09-12 17:28:11 +020055static int shared = 1;
56static int nb_fds;
Fam Zhenge4afbf42015-05-19 10:50:59 +000057static int server_fd;
bellard7a5ca862008-05-27 21:13:40 +000058
59static void usage(const char *name)
60{
Paolo Bonzinib033cd82012-07-18 14:50:52 +020061 (printf) (
bellard7a5ca862008-05-27 21:13:40 +000062"Usage: %s [OPTIONS] FILE\n"
63"QEMU Disk Network Block Device Server\n"
64"\n"
Peter Lieven713cc672014-08-13 19:20:19 +020065" -h, --help display this help and exit\n"
66" -V, --version output version information and exit\n"
bellard7a5ca862008-05-27 21:13:40 +000067"\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020068"Connection properties:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020069" -p, --port=PORT port to listen on (default `%d')\n"
70" -b, --bind=IFACE interface to bind to (default `0.0.0.0')\n"
71" -k, --socket=PATH path to the unix socket\n"
72" (default '"SOCKET_PATH"')\n"
73" -e, --shared=NUM device can be shared by NUM clients (default '1')\n"
74" -t, --persistent don't exit on the last connection\n"
75" -v, --verbose display extra debugging information\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020076"\n"
77"Exposing part of the image:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020078" -o, --offset=OFFSET offset into the image\n"
79" -P, --partition=NUM only expose partition NUM\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020080"\n"
81#ifdef __linux__
82"Kernel NBD client support:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020083" -c, --connect=DEV connect FILE to the local NBD device DEV\n"
84" -d, --disconnect disconnect the specified device\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020085"\n"
86#endif
87"\n"
88"Block device options:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020089" -f, --format=FORMAT set image format (raw, qcow2, ...)\n"
90" -r, --read-only export read-only\n"
91" -s, --snapshot use FILE as an external snapshot, create a temporary\n"
92" file with backing_file=FILE, redirect the write to\n"
93" the temporary one\n"
Wenchao Xia8c116b02013-12-04 17:10:55 +080094" -l, --load-snapshot=SNAPSHOT_PARAM\n"
Peter Lieven713cc672014-08-13 19:20:19 +020095" load an internal snapshot inside FILE and export it\n"
96" as an read-only device, SNAPSHOT_PARAM format is\n"
97" 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
98" '[ID_OR_NAME]'\n"
99" -n, --nocache disable host cache\n"
100" --cache=MODE set cache mode (none, writeback, ...)\n"
Paolo Bonzini39a52352012-07-18 14:57:15 +0200101#ifdef CONFIG_LINUX_AIO
Peter Lieven713cc672014-08-13 19:20:19 +0200102" --aio=MODE set AIO mode (native or threads)\n"
Paolo Bonzini39a52352012-07-18 14:57:15 +0200103#endif
Peter Lievenb3838a42014-08-13 19:20:18 +0200104" --discard=MODE set discard mode (ignore, unmap)\n"
Andrey Korolyov6883de62015-07-31 22:12:54 +0300105" --detect-zeroes=MODE set detect-zeroes mode (off, on, unmap)\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +0200106"\n"
107"Report bugs to <qemu-devel@nongnu.org>\n"
Laurent Vivierc2e28722010-09-17 20:37:25 +0200108 , name, NBD_DEFAULT_PORT, "DEVICE");
bellard7a5ca862008-05-27 21:13:40 +0000109}
110
111static void version(const char *name)
112{
113 printf(
ths315bc7a2008-07-18 18:06:23 +0000114"%s version 0.0.1\n"
bellard7a5ca862008-05-27 21:13:40 +0000115"Written by Anthony Liguori.\n"
116"\n"
117"Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
118"This is free software; see the source for copying conditions. There is NO\n"
119"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
ths315bc7a2008-07-18 18:06:23 +0000120 , name);
bellard7a5ca862008-05-27 21:13:40 +0000121}
122
123struct partition_record
124{
125 uint8_t bootable;
126 uint8_t start_head;
127 uint32_t start_cylinder;
128 uint8_t start_sector;
129 uint8_t system;
130 uint8_t end_head;
131 uint8_t end_cylinder;
132 uint8_t end_sector;
133 uint32_t start_sector_abs;
134 uint32_t nb_sectors_abs;
135};
136
137static void read_partition(uint8_t *p, struct partition_record *r)
138{
139 r->bootable = p[0];
140 r->start_head = p[1];
141 r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
142 r->start_sector = p[2] & 0x3f;
143 r->system = p[4];
144 r->end_head = p[5];
145 r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
146 r->end_sector = p[6] & 0x3f;
Max Reitzac973932015-02-25 13:08:23 -0500147
148 r->start_sector_abs = le32_to_cpup((uint32_t *)(p + 8));
149 r->nb_sectors_abs = le32_to_cpup((uint32_t *)(p + 12));
bellard7a5ca862008-05-27 21:13:40 +0000150}
151
Max Reitz4c58e802014-11-18 12:21:19 +0100152static int find_partition(BlockBackend *blk, int partition,
bellard7a5ca862008-05-27 21:13:40 +0000153 off_t *offset, off_t *size)
154{
155 struct partition_record mbr[4];
156 uint8_t data[512];
157 int i;
158 int ext_partnum = 4;
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900159 int ret;
bellard7a5ca862008-05-27 21:13:40 +0000160
Max Reitz4c58e802014-11-18 12:21:19 +0100161 if ((ret = blk_read(blk, 0, data, 1)) < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900162 errno = -ret;
163 err(EXIT_FAILURE, "error while reading");
164 }
bellard7a5ca862008-05-27 21:13:40 +0000165
166 if (data[510] != 0x55 || data[511] != 0xaa) {
Paolo Bonzini185b4332012-03-05 08:56:10 +0100167 return -EINVAL;
bellard7a5ca862008-05-27 21:13:40 +0000168 }
169
170 for (i = 0; i < 4; i++) {
171 read_partition(&data[446 + 16 * i], &mbr[i]);
172
Max Reitz453b07b2015-02-25 13:08:15 -0500173 if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
bellard7a5ca862008-05-27 21:13:40 +0000174 continue;
Max Reitz453b07b2015-02-25 13:08:15 -0500175 }
bellard7a5ca862008-05-27 21:13:40 +0000176
177 if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
178 struct partition_record ext[4];
179 uint8_t data1[512];
180 int j;
181
Max Reitz4c58e802014-11-18 12:21:19 +0100182 if ((ret = blk_read(blk, mbr[i].start_sector_abs, data1, 1)) < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900183 errno = -ret;
184 err(EXIT_FAILURE, "error while reading");
185 }
bellard7a5ca862008-05-27 21:13:40 +0000186
187 for (j = 0; j < 4; j++) {
188 read_partition(&data1[446 + 16 * j], &ext[j]);
Max Reitz453b07b2015-02-25 13:08:15 -0500189 if (!ext[j].system || !ext[j].nb_sectors_abs) {
bellard7a5ca862008-05-27 21:13:40 +0000190 continue;
Max Reitz453b07b2015-02-25 13:08:15 -0500191 }
bellard7a5ca862008-05-27 21:13:40 +0000192
193 if ((ext_partnum + j + 1) == partition) {
194 *offset = (uint64_t)ext[j].start_sector_abs << 9;
195 *size = (uint64_t)ext[j].nb_sectors_abs << 9;
196 return 0;
197 }
198 }
199 ext_partnum += 4;
200 } else if ((i + 1) == partition) {
201 *offset = (uint64_t)mbr[i].start_sector_abs << 9;
202 *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
203 return 0;
204 }
205 }
206
Paolo Bonzini185b4332012-03-05 08:56:10 +0100207 return -ENOENT;
bellard7a5ca862008-05-27 21:13:40 +0000208}
209
Paolo Bonzinibb345112011-11-04 15:51:19 +0100210static void termsig_handler(int signum)
211{
Paolo Bonzini7860a382012-09-18 13:31:56 +0200212 state = TERMINATE;
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200213 qemu_notify_event();
Paolo Bonzinibb345112011-11-04 15:51:19 +0100214}
215
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100216static void combine_addr(char *buf, size_t len, const char* address,
217 uint16_t port)
218{
219 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
220 if (strstr(address, ":")) {
221 snprintf(buf, len, "[%s]:%u", address, port);
222 } else {
223 snprintf(buf, len, "%s:%u", address, port);
224 }
225}
226
227static int tcp_socket_incoming(const char *address, uint16_t port)
228{
229 char address_and_port[128];
230 Error *local_err = NULL;
231
232 combine_addr(address_and_port, 128, address, port);
233 int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
234
235 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100236 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100237 }
238 return fd;
239}
240
241static int unix_socket_incoming(const char *path)
242{
243 Error *local_err = NULL;
244 int fd = unix_listen(path, NULL, 0, &local_err);
245
246 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100247 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100248 }
249 return fd;
250}
251
252static int unix_socket_outgoing(const char *path)
253{
254 Error *local_err = NULL;
255 int fd = unix_connect(path, &local_err);
256
257 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100258 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100259 }
260 return fd;
261}
262
Paolo Bonzinia517e882011-11-04 15:51:21 +0100263static void *show_parts(void *arg)
thscd831bd2008-07-03 10:23:51 +0000264{
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100265 char *device = arg;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100266 int nbd;
thscd831bd2008-07-03 10:23:51 +0000267
Paolo Bonzinia517e882011-11-04 15:51:21 +0100268 /* linux just needs an open() to trigger
269 * the partition table update
270 * but remember to load the module with max_part != 0 :
271 * modprobe nbd max_part=63
272 */
273 nbd = open(device, O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100274 if (nbd >= 0) {
Paolo Bonzinia517e882011-11-04 15:51:21 +0100275 close(nbd);
thscd831bd2008-07-03 10:23:51 +0000276 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100277 return NULL;
278}
279
280static void *nbd_client_thread(void *arg)
281{
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100282 char *device = arg;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100283 off_t size;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100284 uint32_t nbdflags;
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100285 int fd, sock;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100286 int ret;
287 pthread_t show_parts_thread;
Max Reitz1ce52842015-01-26 21:02:59 -0500288 Error *local_error = NULL;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100289
Stefan Hajnoczidc10e8b2012-01-05 13:16:07 +0000290 sock = unix_socket_outgoing(sockpath);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100291 if (sock < 0) {
Stefan Hajnoczidc10e8b2012-01-05 13:16:07 +0000292 goto out;
293 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100294
295 ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
Max Reitz3f472652015-02-25 13:08:25 -0500296 &size, &local_error);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100297 if (ret < 0) {
Max Reitz1ce52842015-01-26 21:02:59 -0500298 if (local_error) {
299 fprintf(stderr, "%s\n", error_get_pretty(local_error));
300 error_free(local_error);
301 }
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100302 goto out_socket;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100303 }
304
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100305 fd = open(device, O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100306 if (fd < 0) {
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100307 /* Linux-only, we can use %m in printf. */
Hani Benhabiles5672ee52014-05-13 00:35:16 +0100308 fprintf(stderr, "Failed to open %s: %m\n", device);
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100309 goto out_socket;
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100310 }
311
Max Reitz3f472652015-02-25 13:08:25 -0500312 ret = nbd_init(fd, sock, nbdflags, size);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100313 if (ret < 0) {
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100314 goto out_fd;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100315 }
316
317 /* update partition table */
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100318 pthread_create(&show_parts_thread, NULL, show_parts, device);
Paolo Bonzinia517e882011-11-04 15:51:21 +0100319
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100320 if (verbose) {
321 fprintf(stderr, "NBD device %s is now connected to %s\n",
322 device, srcpath);
323 } else {
324 /* Close stderr so that the qemu-nbd process exits. */
325 dup2(STDOUT_FILENO, STDERR_FILENO);
326 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100327
328 ret = nbd_client(fd);
329 if (ret) {
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100330 goto out_fd;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100331 }
332 close(fd);
333 kill(getpid(), SIGTERM);
334 return (void *) EXIT_SUCCESS;
335
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100336out_fd:
337 close(fd);
338out_socket:
339 closesocket(sock);
Paolo Bonzinia517e882011-11-04 15:51:21 +0100340out:
341 kill(getpid(), SIGTERM);
342 return (void *) EXIT_FAILURE;
thscd831bd2008-07-03 10:23:51 +0000343}
344
Fam Zhenge4afbf42015-05-19 10:50:59 +0000345static int nbd_can_accept(void)
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200346{
347 return nb_fds < shared;
348}
349
Paolo Bonzini7860a382012-09-18 13:31:56 +0200350static void nbd_export_closed(NBDExport *exp)
351{
352 assert(state == TERMINATING);
353 state = TERMINATED;
354}
355
Fam Zhenge4afbf42015-05-19 10:50:59 +0000356static void nbd_update_server_fd_handler(int fd);
357
Paolo Bonzini1743b512011-09-19 14:33:23 +0200358static void nbd_client_closed(NBDClient *client)
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200359{
Paolo Bonzini1743b512011-09-19 14:33:23 +0200360 nb_fds--;
Paolo Bonzini7860a382012-09-18 13:31:56 +0200361 if (nb_fds == 0 && !persistent && state == RUNNING) {
362 state = TERMINATE;
363 }
Fam Zhenge4afbf42015-05-19 10:50:59 +0000364 nbd_update_server_fd_handler(server_fd);
Paolo Bonzini7860a382012-09-18 13:31:56 +0200365 nbd_client_put(client);
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200366}
367
368static void nbd_accept(void *opaque)
369{
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200370 struct sockaddr_in addr;
371 socklen_t addr_len = sizeof(addr);
372
373 int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100374 if (fd < 0) {
375 perror("accept");
376 return;
377 }
378
Paolo Bonzini7860a382012-09-18 13:31:56 +0200379 if (state >= TERMINATE) {
380 close(fd);
381 return;
382 }
383
Hani Benhabiles36af5992014-05-13 00:35:15 +0100384 if (nbd_client_new(exp, fd, nbd_client_closed)) {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200385 nb_fds++;
Fam Zhenge4afbf42015-05-19 10:50:59 +0000386 nbd_update_server_fd_handler(server_fd);
Hani Benhabiles36af5992014-05-13 00:35:15 +0100387 } else {
Hani Benhabiles27e5eae2014-05-31 22:39:42 +0100388 shutdown(fd, 2);
Hani Benhabiles36af5992014-05-13 00:35:15 +0100389 close(fd);
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200390 }
391}
392
Fam Zhenge4afbf42015-05-19 10:50:59 +0000393static void nbd_update_server_fd_handler(int fd)
394{
395 if (nbd_can_accept()) {
396 qemu_set_fd_handler(fd, nbd_accept, NULL, (void *)(uintptr_t)fd);
397 } else {
398 qemu_set_fd_handler(fd, NULL, NULL, NULL);
399 }
400}
401
bellard7a5ca862008-05-27 21:13:40 +0000402int main(int argc, char **argv)
403{
Markus Armbruster26f54e92014-10-07 13:59:04 +0200404 BlockBackend *blk;
bellard7a5ca862008-05-27 21:13:40 +0000405 BlockDriverState *bs;
406 off_t dev_offset = 0;
Paolo Bonzinib90fb4b2011-09-08 17:24:54 +0200407 uint32_t nbdflags = 0;
thscd831bd2008-07-03 10:23:51 +0000408 bool disconnect = false;
bellard7a5ca862008-05-27 21:13:40 +0000409 const char *bindto = "0.0.0.0";
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100410 char *device = NULL;
Laurent Vivierc2e28722010-09-17 20:37:25 +0200411 int port = NBD_DEFAULT_PORT;
bellard7a5ca862008-05-27 21:13:40 +0000412 off_t fd_size;
Wenchao Xia8c116b02013-12-04 17:10:55 +0800413 QemuOpts *sn_opts = NULL;
414 const char *sn_id_or_name = NULL;
415 const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:";
bellard7a5ca862008-05-27 21:13:40 +0000416 struct option lopt[] = {
Blue Swirl660f11b2009-07-31 21:16:51 +0000417 { "help", 0, NULL, 'h' },
418 { "version", 0, NULL, 'V' },
419 { "bind", 1, NULL, 'b' },
420 { "port", 1, NULL, 'p' },
421 { "socket", 1, NULL, 'k' },
422 { "offset", 1, NULL, 'o' },
423 { "read-only", 0, NULL, 'r' },
424 { "partition", 1, NULL, 'P' },
425 { "connect", 1, NULL, 'c' },
426 { "disconnect", 0, NULL, 'd' },
427 { "snapshot", 0, NULL, 's' },
Wenchao Xia8c116b02013-12-04 17:10:55 +0800428 { "load-snapshot", 1, NULL, 'l' },
Blue Swirl660f11b2009-07-31 21:16:51 +0000429 { "nocache", 0, NULL, 'n' },
Paolo Bonzini39a52352012-07-18 14:57:15 +0200430 { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
431#ifdef CONFIG_LINUX_AIO
432 { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
433#endif
Paolo Bonzinided9d2d2013-02-08 14:06:13 +0100434 { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
Peter Lievenb3838a42014-08-13 19:20:18 +0200435 { "detect-zeroes", 1, NULL, QEMU_NBD_OPT_DETECT_ZEROES },
Blue Swirl660f11b2009-07-31 21:16:51 +0000436 { "shared", 1, NULL, 'e' },
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000437 { "format", 1, NULL, 'f' },
Blue Swirl660f11b2009-07-31 21:16:51 +0000438 { "persistent", 0, NULL, 't' },
439 { "verbose", 0, NULL, 'v' },
440 { NULL, 0, NULL, 0 }
bellard7a5ca862008-05-27 21:13:40 +0000441 };
442 int ch;
443 int opt_ind = 0;
444 int li;
445 char *end;
Naphtali Spreif5edb012010-01-17 16:48:13 +0200446 int flags = BDRV_O_RDWR;
bellard7a5ca862008-05-27 21:13:40 +0000447 int partition = -1;
Max Reitz4fbec262015-02-05 13:58:19 -0500448 int ret = 0;
ths3b05a8e2008-07-03 12:45:02 +0000449 int fd;
Paolo Bonzini39a52352012-07-18 14:57:15 +0200450 bool seen_cache = false;
Paolo Bonzinided9d2d2013-02-08 14:06:13 +0100451 bool seen_discard = false;
Paolo Bonzini39a52352012-07-18 14:57:15 +0200452#ifdef CONFIG_LINUX_AIO
453 bool seen_aio = false;
454#endif
Paolo Bonzinia517e882011-11-04 15:51:21 +0100455 pthread_t client_thread;
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000456 const char *fmt = NULL;
Max Reitz34b5d2c2013-09-05 14:45:29 +0200457 Error *local_err = NULL;
Peter Lievenb3838a42014-08-13 19:20:18 +0200458 BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
Max Reitz4fbec262015-02-05 13:58:19 -0500459 QDict *options = NULL;
bellard7a5ca862008-05-27 21:13:40 +0000460
Paolo Bonzinia517e882011-11-04 15:51:21 +0100461 /* The client thread uses SIGTERM to interrupt the server. A signal
462 * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
463 */
Paolo Bonzinibb345112011-11-04 15:51:19 +0100464 struct sigaction sa_sigterm;
Paolo Bonzinibb345112011-11-04 15:51:19 +0100465 memset(&sa_sigterm, 0, sizeof(sa_sigterm));
466 sa_sigterm.sa_handler = termsig_handler;
467 sigaction(SIGTERM, &sa_sigterm, NULL);
Fam Zheng10f5bff2014-02-10 14:48:51 +0800468 qemu_init_exec_dir(argv[0]);
Paolo Bonzinibb345112011-11-04 15:51:19 +0100469
bellard7a5ca862008-05-27 21:13:40 +0000470 while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
471 switch (ch) {
472 case 's':
ths2f726482008-07-03 11:47:46 +0000473 flags |= BDRV_O_SNAPSHOT;
474 break;
475 case 'n':
Paolo Bonzini39a52352012-07-18 14:57:15 +0200476 optarg = (char *) "none";
477 /* fallthrough */
478 case QEMU_NBD_OPT_CACHE:
479 if (seen_cache) {
480 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
481 }
482 seen_cache = true;
483 if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
484 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
485 }
bellard7a5ca862008-05-27 21:13:40 +0000486 break;
Paolo Bonzini39a52352012-07-18 14:57:15 +0200487#ifdef CONFIG_LINUX_AIO
488 case QEMU_NBD_OPT_AIO:
489 if (seen_aio) {
490 errx(EXIT_FAILURE, "--aio can only be specified once");
491 }
492 seen_aio = true;
493 if (!strcmp(optarg, "native")) {
494 flags |= BDRV_O_NATIVE_AIO;
495 } else if (!strcmp(optarg, "threads")) {
496 /* this is the default */
497 } else {
498 errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
499 }
500 break;
501#endif
Paolo Bonzinided9d2d2013-02-08 14:06:13 +0100502 case QEMU_NBD_OPT_DISCARD:
503 if (seen_discard) {
504 errx(EXIT_FAILURE, "--discard can only be specified once");
505 }
506 seen_discard = true;
507 if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
508 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
509 }
510 break;
Peter Lievenb3838a42014-08-13 19:20:18 +0200511 case QEMU_NBD_OPT_DETECT_ZEROES:
512 detect_zeroes =
513 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
514 optarg,
515 BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
516 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
517 &local_err);
518 if (local_err) {
519 errx(EXIT_FAILURE, "Failed to parse detect_zeroes mode: %s",
520 error_get_pretty(local_err));
521 }
522 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
523 !(flags & BDRV_O_UNMAP)) {
524 errx(EXIT_FAILURE, "setting detect-zeroes to unmap is not allowed "
525 "without setting discard operation to unmap");
526 }
527 break;
bellard7a5ca862008-05-27 21:13:40 +0000528 case 'b':
529 bindto = optarg;
530 break;
531 case 'p':
532 li = strtol(optarg, &end, 0);
533 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900534 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000535 }
536 if (li < 1 || li > 65535) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900537 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000538 }
539 port = (uint16_t)li;
540 break;
541 case 'o':
542 dev_offset = strtoll (optarg, &end, 0);
543 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900544 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000545 }
546 if (dev_offset < 0) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900547 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
bellard7a5ca862008-05-27 21:13:40 +0000548 }
549 break;
Wenchao Xia8c116b02013-12-04 17:10:55 +0800550 case 'l':
551 if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
Markus Armbruster70b94332015-02-13 12:50:26 +0100552 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
553 optarg, false);
Wenchao Xia8c116b02013-12-04 17:10:55 +0800554 if (!sn_opts) {
555 errx(EXIT_FAILURE, "Failed in parsing snapshot param `%s'",
556 optarg);
557 }
558 } else {
559 sn_id_or_name = optarg;
560 }
561 /* fall through */
bellard7a5ca862008-05-27 21:13:40 +0000562 case 'r':
Paolo Bonzinib90fb4b2011-09-08 17:24:54 +0200563 nbdflags |= NBD_FLAG_READ_ONLY;
Naphtali Sprei07108b22010-03-14 15:19:57 +0200564 flags &= ~BDRV_O_RDWR;
bellard7a5ca862008-05-27 21:13:40 +0000565 break;
566 case 'P':
567 partition = strtol(optarg, &end, 0);
Peter Lieven713cc672014-08-13 19:20:19 +0200568 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900569 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
Peter Lieven713cc672014-08-13 19:20:19 +0200570 }
571 if (partition < 1 || partition > 8) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900572 errx(EXIT_FAILURE, "Invalid partition %d", partition);
Peter Lieven713cc672014-08-13 19:20:19 +0200573 }
bellard7a5ca862008-05-27 21:13:40 +0000574 break;
thscd831bd2008-07-03 10:23:51 +0000575 case 'k':
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100576 sockpath = optarg;
Peter Lieven713cc672014-08-13 19:20:19 +0200577 if (sockpath[0] != '/') {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900578 errx(EXIT_FAILURE, "socket path must be absolute\n");
Peter Lieven713cc672014-08-13 19:20:19 +0200579 }
thscd831bd2008-07-03 10:23:51 +0000580 break;
581 case 'd':
582 disconnect = true;
583 break;
584 case 'c':
585 device = optarg;
586 break;
ths3b05a8e2008-07-03 12:45:02 +0000587 case 'e':
588 shared = strtol(optarg, &end, 0);
589 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900590 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
ths3b05a8e2008-07-03 12:45:02 +0000591 }
592 if (shared < 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900593 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
ths3b05a8e2008-07-03 12:45:02 +0000594 }
595 break;
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000596 case 'f':
597 fmt = optarg;
598 break;
Peter Lieven713cc672014-08-13 19:20:19 +0200599 case 't':
600 persistent = 1;
601 break;
bellard7a5ca862008-05-27 21:13:40 +0000602 case 'v':
603 verbose = 1;
604 break;
605 case 'V':
606 version(argv[0]);
607 exit(0);
608 break;
609 case 'h':
610 usage(argv[0]);
611 exit(0);
612 break;
613 case '?':
Ryota Ozakib6353be2010-03-20 15:23:23 +0900614 errx(EXIT_FAILURE, "Try `%s --help' for more information.",
bellard7a5ca862008-05-27 21:13:40 +0000615 argv[0]);
616 }
617 }
618
619 if ((argc - optind) != 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900620 errx(EXIT_FAILURE, "Invalid number of argument.\n"
bellard7a5ca862008-05-27 21:13:40 +0000621 "Try `%s --help' for more information.",
622 argv[0]);
623 }
624
thscd831bd2008-07-03 10:23:51 +0000625 if (disconnect) {
626 fd = open(argv[optind], O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100627 if (fd < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900628 err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100629 }
thscd831bd2008-07-03 10:23:51 +0000630 nbd_disconnect(fd);
631
632 close(fd);
633
634 printf("%s disconnected\n", argv[optind]);
635
Peter Lieven713cc672014-08-13 19:20:19 +0200636 return 0;
thscd831bd2008-07-03 10:23:51 +0000637 }
638
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100639 if (device && !verbose) {
640 int stderr_fd[2];
641 pid_t pid;
642 int ret;
643
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100644 if (qemu_pipe(stderr_fd) < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100645 err(EXIT_FAILURE, "Error setting up communication pipe");
646 }
647
648 /* Now daemonize, but keep a communication channel open to
649 * print errors and exit with the proper status code.
650 */
651 pid = fork();
Max Reitz70d47392015-02-25 13:08:22 -0500652 if (pid < 0) {
653 err(EXIT_FAILURE, "Failed to fork");
654 } else if (pid == 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100655 close(stderr_fd[0]);
Michael Tokarev9faf31b2012-01-16 18:37:44 +0400656 ret = qemu_daemon(1, 0);
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100657
658 /* Temporarily redirect stderr to the parent's pipe... */
659 dup2(stderr_fd[1], STDERR_FILENO);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100660 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100661 err(EXIT_FAILURE, "Failed to daemonize");
662 }
663
664 /* ... close the descriptor we inherited and go on. */
665 close(stderr_fd[1]);
666 } else {
667 bool errors = false;
668 char *buf;
669
670 /* In the parent. Print error messages from the child until
671 * it closes the pipe.
672 */
673 close(stderr_fd[1]);
674 buf = g_malloc(1024);
675 while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
676 errors = true;
677 ret = qemu_write_full(STDERR_FILENO, buf, ret);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100678 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100679 exit(EXIT_FAILURE);
680 }
681 }
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100682 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100683 err(EXIT_FAILURE, "Cannot read from daemon");
684 }
685
686 /* Usually the daemon should not print any message.
687 * Exit with zero status in that case.
688 */
689 exit(errors);
690 }
691 }
692
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100693 if (device != NULL && sockpath == NULL) {
694 sockpath = g_malloc(128);
695 snprintf(sockpath, 128, SOCKET_PATH, basename(device));
thscd831bd2008-07-03 10:23:51 +0000696 }
697
Chrysostomos Nanakos2f78e492014-09-18 14:30:49 +0300698 if (qemu_init_main_loop(&local_err)) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100699 error_report_err(local_err);
Chrysostomos Nanakos2f78e492014-09-18 14:30:49 +0300700 exit(EXIT_FAILURE);
701 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100702 bdrv_init();
703 atexit(bdrv_close_all);
704
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000705 if (fmt) {
Max Reitz4fbec262015-02-05 13:58:19 -0500706 options = qdict_new();
707 qdict_put(options, "driver", qstring_from_str(fmt));
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000708 }
709
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100710 srcpath = argv[optind];
Max Reitz4fbec262015-02-05 13:58:19 -0500711 blk = blk_new_open("hda", srcpath, NULL, options, flags, &local_err);
712 if (!blk) {
713 errx(EXIT_FAILURE, "Failed to blk_new_open '%s': %s", argv[optind],
714 error_get_pretty(local_err));
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100715 }
Max Reitz4fbec262015-02-05 13:58:19 -0500716 bs = blk_bs(blk);
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100717
Wenchao Xia8c116b02013-12-04 17:10:55 +0800718 if (sn_opts) {
719 ret = bdrv_snapshot_load_tmp(bs,
720 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
721 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
722 &local_err);
723 } else if (sn_id_or_name) {
724 ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
725 &local_err);
726 }
727 if (ret < 0) {
728 errno = -ret;
729 err(EXIT_FAILURE,
730 "Failed to load snapshot: %s",
731 error_get_pretty(local_err));
732 }
733
Peter Lievenb3838a42014-08-13 19:20:18 +0200734 bs->detect_zeroes = detect_zeroes;
Max Reitz4c58e802014-11-18 12:21:19 +0100735 fd_size = blk_getlength(blk);
Max Reitz98f44bb2015-02-25 13:08:21 -0500736 if (fd_size < 0) {
737 errx(EXIT_FAILURE, "Failed to determine the image length: %s",
738 strerror(-fd_size));
739 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100740
Paolo Bonzini185b4332012-03-05 08:56:10 +0100741 if (partition != -1) {
Max Reitz4c58e802014-11-18 12:21:19 +0100742 ret = find_partition(blk, partition, &dev_offset, &fd_size);
Paolo Bonzini185b4332012-03-05 08:56:10 +0100743 if (ret < 0) {
744 errno = -ret;
745 err(EXIT_FAILURE, "Could not find partition %d", partition);
746 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100747 }
748
Max Reitz98f44bb2015-02-25 13:08:21 -0500749 exp = nbd_export_new(blk, dev_offset, fd_size, nbdflags, nbd_export_closed,
750 &local_err);
751 if (!exp) {
752 errx(EXIT_FAILURE, "%s", error_get_pretty(local_err));
753 }
ths3b05a8e2008-07-03 12:45:02 +0000754
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100755 if (sockpath) {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200756 fd = unix_socket_incoming(sockpath);
thscd831bd2008-07-03 10:23:51 +0000757 } else {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200758 fd = tcp_socket_incoming(bindto, port);
thscd831bd2008-07-03 10:23:51 +0000759 }
760
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100761 if (fd < 0) {
bellard7a5ca862008-05-27 21:13:40 +0000762 return 1;
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200763 }
Paolo Bonzinif1ef5552011-11-04 15:51:23 +0100764
765 if (device) {
766 int ret;
767
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100768 ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
Paolo Bonzinif1ef5552011-11-04 15:51:23 +0100769 if (ret != 0) {
770 errx(EXIT_FAILURE, "Failed to create client thread: %s",
771 strerror(ret));
772 }
773 } else {
774 /* Shut up GCC warnings. */
775 memset(&client_thread, 0, sizeof(client_thread));
776 }
777
Fam Zhenge4afbf42015-05-19 10:50:59 +0000778 server_fd = fd;
779 nbd_update_server_fd_handler(fd);
bellard7a5ca862008-05-27 21:13:40 +0000780
Michael Tokarev9faf31b2012-01-16 18:37:44 +0400781 /* now when the initialization is (almost) complete, chdir("/")
782 * to free any busy filesystems */
783 if (chdir("/") < 0) {
784 err(EXIT_FAILURE, "Could not chdir to root directory");
785 }
786
Paolo Bonzini7860a382012-09-18 13:31:56 +0200787 state = RUNNING;
ths3b05a8e2008-07-03 12:45:02 +0000788 do {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200789 main_loop_wait(false);
Paolo Bonzini7860a382012-09-18 13:31:56 +0200790 if (state == TERMINATE) {
791 state = TERMINATING;
792 nbd_export_close(exp);
793 nbd_export_put(exp);
794 exp = NULL;
795 }
796 } while (state != TERMINATED);
ths3b05a8e2008-07-03 12:45:02 +0000797
Markus Armbruster26f54e92014-10-07 13:59:04 +0200798 blk_unref(blk);
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100799 if (sockpath) {
800 unlink(sockpath);
801 }
bellard7a5ca862008-05-27 21:13:40 +0000802
Markus Armbrusterfbf28a42014-09-29 16:07:55 +0200803 qemu_opts_del(sn_opts);
Wenchao Xia8c116b02013-12-04 17:10:55 +0800804
Paolo Bonzinia517e882011-11-04 15:51:21 +0100805 if (device) {
806 void *ret;
807 pthread_join(client_thread, &ret);
808 exit(ret != NULL);
809 } else {
810 exit(EXIT_SUCCESS);
811 }
bellard7a5ca862008-05-27 21:13:40 +0000812}