blob: 5af6d11e3387b070f0ba411b4a3310960273a609 [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"
bellard7a5ca862008-05-27 21:13:40 +000028
bellard7a5ca862008-05-27 21:13:40 +000029#include <stdarg.h>
30#include <stdio.h>
31#include <getopt.h>
32#include <err.h>
thscd831bd2008-07-03 10:23:51 +000033#include <sys/types.h>
bellard7a5ca862008-05-27 21:13:40 +000034#include <sys/socket.h>
35#include <netinet/in.h>
36#include <netinet/tcp.h>
37#include <arpa/inet.h>
thscd831bd2008-07-03 10:23:51 +000038#include <signal.h>
Blue Swirl2bff4b62009-12-23 15:34:04 +000039#include <libgen.h>
Paolo Bonzinia517e882011-11-04 15:51:21 +010040#include <pthread.h>
thscd831bd2008-07-03 10:23:51 +000041
Peter Lieven713cc672014-08-13 19:20:19 +020042#define SOCKET_PATH "/var/lock/qemu-nbd-%s"
43#define QEMU_NBD_OPT_CACHE 1
44#define QEMU_NBD_OPT_AIO 2
45#define QEMU_NBD_OPT_DISCARD 3
Peter Lievenb3838a42014-08-13 19:20:18 +020046#define QEMU_NBD_OPT_DETECT_ZEROES 4
bellard7a5ca862008-05-27 21:13:40 +000047
Paolo Bonziniaf49bbb2011-09-19 14:03:37 +020048static NBDExport *exp;
blueswir1b1d8e522008-10-26 13:43:07 +000049static int verbose;
Paolo Bonzinia517e882011-11-04 15:51:21 +010050static char *srcpath;
51static char *sockpath;
Paolo Bonzini7860a382012-09-18 13:31:56 +020052static int persistent = 0;
53static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
Paolo Bonzinia61c6782011-09-12 17:28:11 +020054static int shared = 1;
55static int nb_fds;
Fam Zhenge4afbf42015-05-19 10:50:59 +000056static int server_fd;
bellard7a5ca862008-05-27 21:13:40 +000057
58static void usage(const char *name)
59{
Paolo Bonzinib033cd82012-07-18 14:50:52 +020060 (printf) (
bellard7a5ca862008-05-27 21:13:40 +000061"Usage: %s [OPTIONS] FILE\n"
62"QEMU Disk Network Block Device Server\n"
63"\n"
Peter Lieven713cc672014-08-13 19:20:19 +020064" -h, --help display this help and exit\n"
65" -V, --version output version information and exit\n"
bellard7a5ca862008-05-27 21:13:40 +000066"\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020067"Connection properties:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020068" -p, --port=PORT port to listen on (default `%d')\n"
69" -b, --bind=IFACE interface to bind to (default `0.0.0.0')\n"
70" -k, --socket=PATH path to the unix socket\n"
71" (default '"SOCKET_PATH"')\n"
72" -e, --shared=NUM device can be shared by NUM clients (default '1')\n"
73" -t, --persistent don't exit on the last connection\n"
74" -v, --verbose display extra debugging information\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020075"\n"
76"Exposing part of the image:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020077" -o, --offset=OFFSET offset into the image\n"
78" -P, --partition=NUM only expose partition NUM\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020079"\n"
80#ifdef __linux__
81"Kernel NBD client support:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020082" -c, --connect=DEV connect FILE to the local NBD device DEV\n"
83" -d, --disconnect disconnect the specified device\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +020084"\n"
85#endif
86"\n"
87"Block device options:\n"
Peter Lieven713cc672014-08-13 19:20:19 +020088" -f, --format=FORMAT set image format (raw, qcow2, ...)\n"
89" -r, --read-only export read-only\n"
90" -s, --snapshot use FILE as an external snapshot, create a temporary\n"
91" file with backing_file=FILE, redirect the write to\n"
92" the temporary one\n"
Wenchao Xia8c116b02013-12-04 17:10:55 +080093" -l, --load-snapshot=SNAPSHOT_PARAM\n"
Peter Lieven713cc672014-08-13 19:20:19 +020094" load an internal snapshot inside FILE and export it\n"
95" as an read-only device, SNAPSHOT_PARAM format is\n"
96" 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
97" '[ID_OR_NAME]'\n"
98" -n, --nocache disable host cache\n"
99" --cache=MODE set cache mode (none, writeback, ...)\n"
Paolo Bonzini39a52352012-07-18 14:57:15 +0200100#ifdef CONFIG_LINUX_AIO
Peter Lieven713cc672014-08-13 19:20:19 +0200101" --aio=MODE set AIO mode (native or threads)\n"
Paolo Bonzini39a52352012-07-18 14:57:15 +0200102#endif
Peter Lievenb3838a42014-08-13 19:20:18 +0200103" --discard=MODE set discard mode (ignore, unmap)\n"
104" --detect-zeroes=MODE set detect-zeroes mode (off, on, discard)\n"
Paolo Bonzinib033cd82012-07-18 14:50:52 +0200105"\n"
106"Report bugs to <qemu-devel@nongnu.org>\n"
Laurent Vivierc2e28722010-09-17 20:37:25 +0200107 , name, NBD_DEFAULT_PORT, "DEVICE");
bellard7a5ca862008-05-27 21:13:40 +0000108}
109
110static void version(const char *name)
111{
112 printf(
ths315bc7a2008-07-18 18:06:23 +0000113"%s version 0.0.1\n"
bellard7a5ca862008-05-27 21:13:40 +0000114"Written by Anthony Liguori.\n"
115"\n"
116"Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
117"This is free software; see the source for copying conditions. There is NO\n"
118"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
ths315bc7a2008-07-18 18:06:23 +0000119 , name);
bellard7a5ca862008-05-27 21:13:40 +0000120}
121
122struct partition_record
123{
124 uint8_t bootable;
125 uint8_t start_head;
126 uint32_t start_cylinder;
127 uint8_t start_sector;
128 uint8_t system;
129 uint8_t end_head;
130 uint8_t end_cylinder;
131 uint8_t end_sector;
132 uint32_t start_sector_abs;
133 uint32_t nb_sectors_abs;
134};
135
136static void read_partition(uint8_t *p, struct partition_record *r)
137{
138 r->bootable = p[0];
139 r->start_head = p[1];
140 r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
141 r->start_sector = p[2] & 0x3f;
142 r->system = p[4];
143 r->end_head = p[5];
144 r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
145 r->end_sector = p[6] & 0x3f;
Max Reitzac973932015-02-25 13:08:23 -0500146
147 r->start_sector_abs = le32_to_cpup((uint32_t *)(p + 8));
148 r->nb_sectors_abs = le32_to_cpup((uint32_t *)(p + 12));
bellard7a5ca862008-05-27 21:13:40 +0000149}
150
Max Reitz4c58e802014-11-18 12:21:19 +0100151static int find_partition(BlockBackend *blk, int partition,
bellard7a5ca862008-05-27 21:13:40 +0000152 off_t *offset, off_t *size)
153{
154 struct partition_record mbr[4];
155 uint8_t data[512];
156 int i;
157 int ext_partnum = 4;
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900158 int ret;
bellard7a5ca862008-05-27 21:13:40 +0000159
Max Reitz4c58e802014-11-18 12:21:19 +0100160 if ((ret = blk_read(blk, 0, data, 1)) < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900161 errno = -ret;
162 err(EXIT_FAILURE, "error while reading");
163 }
bellard7a5ca862008-05-27 21:13:40 +0000164
165 if (data[510] != 0x55 || data[511] != 0xaa) {
Paolo Bonzini185b4332012-03-05 08:56:10 +0100166 return -EINVAL;
bellard7a5ca862008-05-27 21:13:40 +0000167 }
168
169 for (i = 0; i < 4; i++) {
170 read_partition(&data[446 + 16 * i], &mbr[i]);
171
Max Reitz453b07b2015-02-25 13:08:15 -0500172 if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
bellard7a5ca862008-05-27 21:13:40 +0000173 continue;
Max Reitz453b07b2015-02-25 13:08:15 -0500174 }
bellard7a5ca862008-05-27 21:13:40 +0000175
176 if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
177 struct partition_record ext[4];
178 uint8_t data1[512];
179 int j;
180
Max Reitz4c58e802014-11-18 12:21:19 +0100181 if ((ret = blk_read(blk, mbr[i].start_sector_abs, data1, 1)) < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900182 errno = -ret;
183 err(EXIT_FAILURE, "error while reading");
184 }
bellard7a5ca862008-05-27 21:13:40 +0000185
186 for (j = 0; j < 4; j++) {
187 read_partition(&data1[446 + 16 * j], &ext[j]);
Max Reitz453b07b2015-02-25 13:08:15 -0500188 if (!ext[j].system || !ext[j].nb_sectors_abs) {
bellard7a5ca862008-05-27 21:13:40 +0000189 continue;
Max Reitz453b07b2015-02-25 13:08:15 -0500190 }
bellard7a5ca862008-05-27 21:13:40 +0000191
192 if ((ext_partnum + j + 1) == partition) {
193 *offset = (uint64_t)ext[j].start_sector_abs << 9;
194 *size = (uint64_t)ext[j].nb_sectors_abs << 9;
195 return 0;
196 }
197 }
198 ext_partnum += 4;
199 } else if ((i + 1) == partition) {
200 *offset = (uint64_t)mbr[i].start_sector_abs << 9;
201 *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
202 return 0;
203 }
204 }
205
Paolo Bonzini185b4332012-03-05 08:56:10 +0100206 return -ENOENT;
bellard7a5ca862008-05-27 21:13:40 +0000207}
208
Paolo Bonzinibb345112011-11-04 15:51:19 +0100209static void termsig_handler(int signum)
210{
Paolo Bonzini7860a382012-09-18 13:31:56 +0200211 state = TERMINATE;
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200212 qemu_notify_event();
Paolo Bonzinibb345112011-11-04 15:51:19 +0100213}
214
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100215static void combine_addr(char *buf, size_t len, const char* address,
216 uint16_t port)
217{
218 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
219 if (strstr(address, ":")) {
220 snprintf(buf, len, "[%s]:%u", address, port);
221 } else {
222 snprintf(buf, len, "%s:%u", address, port);
223 }
224}
225
226static int tcp_socket_incoming(const char *address, uint16_t port)
227{
228 char address_and_port[128];
229 Error *local_err = NULL;
230
231 combine_addr(address_and_port, 128, address, port);
232 int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
233
234 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100235 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100236 }
237 return fd;
238}
239
240static int unix_socket_incoming(const char *path)
241{
242 Error *local_err = NULL;
243 int fd = unix_listen(path, NULL, 0, &local_err);
244
245 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100246 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100247 }
248 return fd;
249}
250
251static int unix_socket_outgoing(const char *path)
252{
253 Error *local_err = NULL;
254 int fd = unix_connect(path, &local_err);
255
256 if (local_err != NULL) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100257 error_report_err(local_err);
Paolo Bonzini537b41f2014-02-17 14:43:51 +0100258 }
259 return fd;
260}
261
Paolo Bonzinia517e882011-11-04 15:51:21 +0100262static void *show_parts(void *arg)
thscd831bd2008-07-03 10:23:51 +0000263{
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100264 char *device = arg;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100265 int nbd;
thscd831bd2008-07-03 10:23:51 +0000266
Paolo Bonzinia517e882011-11-04 15:51:21 +0100267 /* linux just needs an open() to trigger
268 * the partition table update
269 * but remember to load the module with max_part != 0 :
270 * modprobe nbd max_part=63
271 */
272 nbd = open(device, O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100273 if (nbd >= 0) {
Paolo Bonzinia517e882011-11-04 15:51:21 +0100274 close(nbd);
thscd831bd2008-07-03 10:23:51 +0000275 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100276 return NULL;
277}
278
279static void *nbd_client_thread(void *arg)
280{
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100281 char *device = arg;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100282 off_t size;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100283 uint32_t nbdflags;
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100284 int fd, sock;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100285 int ret;
286 pthread_t show_parts_thread;
Max Reitz1ce52842015-01-26 21:02:59 -0500287 Error *local_error = NULL;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100288
Stefan Hajnoczidc10e8b2012-01-05 13:16:07 +0000289 sock = unix_socket_outgoing(sockpath);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100290 if (sock < 0) {
Stefan Hajnoczidc10e8b2012-01-05 13:16:07 +0000291 goto out;
292 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100293
294 ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
Max Reitz3f472652015-02-25 13:08:25 -0500295 &size, &local_error);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100296 if (ret < 0) {
Max Reitz1ce52842015-01-26 21:02:59 -0500297 if (local_error) {
298 fprintf(stderr, "%s\n", error_get_pretty(local_error));
299 error_free(local_error);
300 }
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100301 goto out_socket;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100302 }
303
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100304 fd = open(device, O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100305 if (fd < 0) {
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100306 /* Linux-only, we can use %m in printf. */
Hani Benhabiles5672ee52014-05-13 00:35:16 +0100307 fprintf(stderr, "Failed to open %s: %m\n", device);
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100308 goto out_socket;
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100309 }
310
Max Reitz3f472652015-02-25 13:08:25 -0500311 ret = nbd_init(fd, sock, nbdflags, size);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100312 if (ret < 0) {
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100313 goto out_fd;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100314 }
315
316 /* update partition table */
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100317 pthread_create(&show_parts_thread, NULL, show_parts, device);
Paolo Bonzinia517e882011-11-04 15:51:21 +0100318
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100319 if (verbose) {
320 fprintf(stderr, "NBD device %s is now connected to %s\n",
321 device, srcpath);
322 } else {
323 /* Close stderr so that the qemu-nbd process exits. */
324 dup2(STDOUT_FILENO, STDERR_FILENO);
325 }
Paolo Bonzinia517e882011-11-04 15:51:21 +0100326
327 ret = nbd_client(fd);
328 if (ret) {
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100329 goto out_fd;
Paolo Bonzinia517e882011-11-04 15:51:21 +0100330 }
331 close(fd);
332 kill(getpid(), SIGTERM);
333 return (void *) EXIT_SUCCESS;
334
Paolo Bonzini0c544d72014-03-14 18:10:54 +0100335out_fd:
336 close(fd);
337out_socket:
338 closesocket(sock);
Paolo Bonzinia517e882011-11-04 15:51:21 +0100339out:
340 kill(getpid(), SIGTERM);
341 return (void *) EXIT_FAILURE;
thscd831bd2008-07-03 10:23:51 +0000342}
343
Fam Zhenge4afbf42015-05-19 10:50:59 +0000344static int nbd_can_accept(void)
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200345{
346 return nb_fds < shared;
347}
348
Paolo Bonzini7860a382012-09-18 13:31:56 +0200349static void nbd_export_closed(NBDExport *exp)
350{
351 assert(state == TERMINATING);
352 state = TERMINATED;
353}
354
Fam Zhenge4afbf42015-05-19 10:50:59 +0000355static void nbd_update_server_fd_handler(int fd);
356
Paolo Bonzini1743b512011-09-19 14:33:23 +0200357static void nbd_client_closed(NBDClient *client)
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200358{
Paolo Bonzini1743b512011-09-19 14:33:23 +0200359 nb_fds--;
Paolo Bonzini7860a382012-09-18 13:31:56 +0200360 if (nb_fds == 0 && !persistent && state == RUNNING) {
361 state = TERMINATE;
362 }
Fam Zhenge4afbf42015-05-19 10:50:59 +0000363 nbd_update_server_fd_handler(server_fd);
Paolo Bonzini1743b512011-09-19 14:33:23 +0200364 qemu_notify_event();
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)) {
552 sn_opts = qemu_opts_parse(&internal_snapshot_opts, optarg, 0);
553 if (!sn_opts) {
554 errx(EXIT_FAILURE, "Failed in parsing snapshot param `%s'",
555 optarg);
556 }
557 } else {
558 sn_id_or_name = optarg;
559 }
560 /* fall through */
bellard7a5ca862008-05-27 21:13:40 +0000561 case 'r':
Paolo Bonzinib90fb4b2011-09-08 17:24:54 +0200562 nbdflags |= NBD_FLAG_READ_ONLY;
Naphtali Sprei07108b22010-03-14 15:19:57 +0200563 flags &= ~BDRV_O_RDWR;
bellard7a5ca862008-05-27 21:13:40 +0000564 break;
565 case 'P':
566 partition = strtol(optarg, &end, 0);
Peter Lieven713cc672014-08-13 19:20:19 +0200567 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900568 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
Peter Lieven713cc672014-08-13 19:20:19 +0200569 }
570 if (partition < 1 || partition > 8) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900571 errx(EXIT_FAILURE, "Invalid partition %d", partition);
Peter Lieven713cc672014-08-13 19:20:19 +0200572 }
bellard7a5ca862008-05-27 21:13:40 +0000573 break;
thscd831bd2008-07-03 10:23:51 +0000574 case 'k':
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100575 sockpath = optarg;
Peter Lieven713cc672014-08-13 19:20:19 +0200576 if (sockpath[0] != '/') {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900577 errx(EXIT_FAILURE, "socket path must be absolute\n");
Peter Lieven713cc672014-08-13 19:20:19 +0200578 }
thscd831bd2008-07-03 10:23:51 +0000579 break;
580 case 'd':
581 disconnect = true;
582 break;
583 case 'c':
584 device = optarg;
585 break;
ths3b05a8e2008-07-03 12:45:02 +0000586 case 'e':
587 shared = strtol(optarg, &end, 0);
588 if (*end) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900589 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
ths3b05a8e2008-07-03 12:45:02 +0000590 }
591 if (shared < 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900592 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
ths3b05a8e2008-07-03 12:45:02 +0000593 }
594 break;
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000595 case 'f':
596 fmt = optarg;
597 break;
Peter Lieven713cc672014-08-13 19:20:19 +0200598 case 't':
599 persistent = 1;
600 break;
bellard7a5ca862008-05-27 21:13:40 +0000601 case 'v':
602 verbose = 1;
603 break;
604 case 'V':
605 version(argv[0]);
606 exit(0);
607 break;
608 case 'h':
609 usage(argv[0]);
610 exit(0);
611 break;
612 case '?':
Ryota Ozakib6353be2010-03-20 15:23:23 +0900613 errx(EXIT_FAILURE, "Try `%s --help' for more information.",
bellard7a5ca862008-05-27 21:13:40 +0000614 argv[0]);
615 }
616 }
617
618 if ((argc - optind) != 1) {
Ryota Ozakib6353be2010-03-20 15:23:23 +0900619 errx(EXIT_FAILURE, "Invalid number of argument.\n"
bellard7a5ca862008-05-27 21:13:40 +0000620 "Try `%s --help' for more information.",
621 argv[0]);
622 }
623
thscd831bd2008-07-03 10:23:51 +0000624 if (disconnect) {
625 fd = open(argv[optind], O_RDWR);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100626 if (fd < 0) {
Ryota Ozakicb7cf0e2010-05-03 06:50:25 +0900627 err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100628 }
thscd831bd2008-07-03 10:23:51 +0000629 nbd_disconnect(fd);
630
631 close(fd);
632
633 printf("%s disconnected\n", argv[optind]);
634
Peter Lieven713cc672014-08-13 19:20:19 +0200635 return 0;
thscd831bd2008-07-03 10:23:51 +0000636 }
637
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100638 if (device && !verbose) {
639 int stderr_fd[2];
640 pid_t pid;
641 int ret;
642
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100643 if (qemu_pipe(stderr_fd) < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100644 err(EXIT_FAILURE, "Error setting up communication pipe");
645 }
646
647 /* Now daemonize, but keep a communication channel open to
648 * print errors and exit with the proper status code.
649 */
650 pid = fork();
Max Reitz70d47392015-02-25 13:08:22 -0500651 if (pid < 0) {
652 err(EXIT_FAILURE, "Failed to fork");
653 } else if (pid == 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100654 close(stderr_fd[0]);
Michael Tokarev9faf31b2012-01-16 18:37:44 +0400655 ret = qemu_daemon(1, 0);
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100656
657 /* Temporarily redirect stderr to the parent's pipe... */
658 dup2(stderr_fd[1], STDERR_FILENO);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100659 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100660 err(EXIT_FAILURE, "Failed to daemonize");
661 }
662
663 /* ... close the descriptor we inherited and go on. */
664 close(stderr_fd[1]);
665 } else {
666 bool errors = false;
667 char *buf;
668
669 /* In the parent. Print error messages from the child until
670 * it closes the pipe.
671 */
672 close(stderr_fd[1]);
673 buf = g_malloc(1024);
674 while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
675 errors = true;
676 ret = qemu_write_full(STDERR_FILENO, buf, ret);
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100677 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100678 exit(EXIT_FAILURE);
679 }
680 }
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100681 if (ret < 0) {
Paolo Bonzinic1f8fdc2011-11-04 15:51:22 +0100682 err(EXIT_FAILURE, "Cannot read from daemon");
683 }
684
685 /* Usually the daemon should not print any message.
686 * Exit with zero status in that case.
687 */
688 exit(errors);
689 }
690 }
691
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100692 if (device != NULL && sockpath == NULL) {
693 sockpath = g_malloc(128);
694 snprintf(sockpath, 128, SOCKET_PATH, basename(device));
thscd831bd2008-07-03 10:23:51 +0000695 }
696
Chrysostomos Nanakos2f78e492014-09-18 14:30:49 +0300697 if (qemu_init_main_loop(&local_err)) {
Markus Armbruster565f65d2015-02-12 13:55:05 +0100698 error_report_err(local_err);
Chrysostomos Nanakos2f78e492014-09-18 14:30:49 +0300699 exit(EXIT_FAILURE);
700 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100701 bdrv_init();
702 atexit(bdrv_close_all);
703
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000704 if (fmt) {
Max Reitz4fbec262015-02-05 13:58:19 -0500705 options = qdict_new();
706 qdict_put(options, "driver", qstring_from_str(fmt));
Daniel P. Berrangee6b63672013-03-19 11:20:20 +0000707 }
708
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100709 srcpath = argv[optind];
Max Reitz4fbec262015-02-05 13:58:19 -0500710 blk = blk_new_open("hda", srcpath, NULL, options, flags, &local_err);
711 if (!blk) {
712 errx(EXIT_FAILURE, "Failed to blk_new_open '%s': %s", argv[optind],
713 error_get_pretty(local_err));
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100714 }
Max Reitz4fbec262015-02-05 13:58:19 -0500715 bs = blk_bs(blk);
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100716
Wenchao Xia8c116b02013-12-04 17:10:55 +0800717 if (sn_opts) {
718 ret = bdrv_snapshot_load_tmp(bs,
719 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
720 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
721 &local_err);
722 } else if (sn_id_or_name) {
723 ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
724 &local_err);
725 }
726 if (ret < 0) {
727 errno = -ret;
728 err(EXIT_FAILURE,
729 "Failed to load snapshot: %s",
730 error_get_pretty(local_err));
731 }
732
Peter Lievenb3838a42014-08-13 19:20:18 +0200733 bs->detect_zeroes = detect_zeroes;
Max Reitz4c58e802014-11-18 12:21:19 +0100734 fd_size = blk_getlength(blk);
Max Reitz98f44bb2015-02-25 13:08:21 -0500735 if (fd_size < 0) {
736 errx(EXIT_FAILURE, "Failed to determine the image length: %s",
737 strerror(-fd_size));
738 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100739
Paolo Bonzini185b4332012-03-05 08:56:10 +0100740 if (partition != -1) {
Max Reitz4c58e802014-11-18 12:21:19 +0100741 ret = find_partition(blk, partition, &dev_offset, &fd_size);
Paolo Bonzini185b4332012-03-05 08:56:10 +0100742 if (ret < 0) {
743 errno = -ret;
744 err(EXIT_FAILURE, "Could not find partition %d", partition);
745 }
Paolo Bonzini802ddc32011-11-04 15:51:24 +0100746 }
747
Max Reitz98f44bb2015-02-25 13:08:21 -0500748 exp = nbd_export_new(blk, dev_offset, fd_size, nbdflags, nbd_export_closed,
749 &local_err);
750 if (!exp) {
751 errx(EXIT_FAILURE, "%s", error_get_pretty(local_err));
752 }
ths3b05a8e2008-07-03 12:45:02 +0000753
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100754 if (sockpath) {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200755 fd = unix_socket_incoming(sockpath);
thscd831bd2008-07-03 10:23:51 +0000756 } else {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200757 fd = tcp_socket_incoming(bindto, port);
thscd831bd2008-07-03 10:23:51 +0000758 }
759
Paolo Bonzinifc19f8a2012-03-07 11:05:34 +0100760 if (fd < 0) {
bellard7a5ca862008-05-27 21:13:40 +0000761 return 1;
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200762 }
Paolo Bonzinif1ef5552011-11-04 15:51:23 +0100763
764 if (device) {
765 int ret;
766
Paolo Bonzinia6ac2312011-12-06 09:07:00 +0100767 ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
Paolo Bonzinif1ef5552011-11-04 15:51:23 +0100768 if (ret != 0) {
769 errx(EXIT_FAILURE, "Failed to create client thread: %s",
770 strerror(ret));
771 }
772 } else {
773 /* Shut up GCC warnings. */
774 memset(&client_thread, 0, sizeof(client_thread));
775 }
776
Fam Zhenge4afbf42015-05-19 10:50:59 +0000777 server_fd = fd;
778 nbd_update_server_fd_handler(fd);
bellard7a5ca862008-05-27 21:13:40 +0000779
Michael Tokarev9faf31b2012-01-16 18:37:44 +0400780 /* now when the initialization is (almost) complete, chdir("/")
781 * to free any busy filesystems */
782 if (chdir("/") < 0) {
783 err(EXIT_FAILURE, "Could not chdir to root directory");
784 }
785
Paolo Bonzini7860a382012-09-18 13:31:56 +0200786 state = RUNNING;
ths3b05a8e2008-07-03 12:45:02 +0000787 do {
Paolo Bonzinia61c6782011-09-12 17:28:11 +0200788 main_loop_wait(false);
Paolo Bonzini7860a382012-09-18 13:31:56 +0200789 if (state == TERMINATE) {
790 state = TERMINATING;
791 nbd_export_close(exp);
792 nbd_export_put(exp);
793 exp = NULL;
794 }
795 } while (state != TERMINATED);
ths3b05a8e2008-07-03 12:45:02 +0000796
Markus Armbruster26f54e92014-10-07 13:59:04 +0200797 blk_unref(blk);
Paolo Bonzinib32f6c22011-11-04 15:51:20 +0100798 if (sockpath) {
799 unlink(sockpath);
800 }
bellard7a5ca862008-05-27 21:13:40 +0000801
Markus Armbrusterfbf28a42014-09-29 16:07:55 +0200802 qemu_opts_del(sn_opts);
Wenchao Xia8c116b02013-12-04 17:10:55 +0800803
Paolo Bonzinia517e882011-11-04 15:51:21 +0100804 if (device) {
805 void *ret;
806 pthread_join(client_thread, &ret);
807 exit(ret != NULL);
808 } else {
809 exit(EXIT_SUCCESS);
810 }
bellard7a5ca862008-05-27 21:13:40 +0000811}