]> git.proxmox.com Git - mirror_qemu.git/blame - qemu-nbd.c
cpus: Fix CPU unplug for MTTCG
[mirror_qemu.git] / qemu-nbd.c
CommitLineData
cd831bd7 1/*
7a5ca864
FB
2 * 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
8167ee88 16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
7a5ca864
FB
17 */
18
d38ea87a 19#include "qemu/osdep.h"
da34e65c 20#include "qapi/error.h"
5a61cb60 21#include "qemu-common.h"
f348b6d1 22#include "qemu/cutils.h"
26f54e9a 23#include "sysemu/block-backend.h"
b3838a40 24#include "block/block_int.h"
737e150e 25#include "block/nbd.h"
6a1751b7 26#include "qemu/main-loop.h"
537b41f5 27#include "qemu/error-report.h"
0ab3b337 28#include "qemu/config-file.h"
58369e22 29#include "qemu/bswap.h"
39ca463e 30#include "qemu/log.h"
53fabd4b 31#include "qemu/systemd.h"
8c116b0e 32#include "block/snapshot.h"
b3838a40 33#include "qapi/util.h"
d49b6836 34#include "qapi/qmp/qstring.h"
0ab3b337 35#include "qom/object_interfaces.h"
d0d6ff58 36#include "io/channel-socket.h"
c2297088 37#include "crypto/init.h"
39ca463e 38#include "trace/control.h"
7a5ca864 39
7a5ca864 40#include <getopt.h>
2bff4b6f 41#include <libgen.h>
a517e88b 42#include <pthread.h>
cd831bd7 43
713cc671 44#define SOCKET_PATH "/var/lock/qemu-nbd-%s"
fa8b7ce2
DB
45#define QEMU_NBD_OPT_CACHE 256
46#define QEMU_NBD_OPT_AIO 257
47#define QEMU_NBD_OPT_DISCARD 258
48#define QEMU_NBD_OPT_DETECT_ZEROES 259
49#define QEMU_NBD_OPT_OBJECT 260
50#define QEMU_NBD_OPT_TLSCREDS 261
51#define QEMU_NBD_OPT_IMAGE_OPTS 262
ffb31e1d 52#define QEMU_NBD_OPT_FORK 263
7a5ca864 53
bd31c214
EB
54#define MBR_SIZE 512
55
af49bbbe 56static NBDExport *exp;
3d4b2f9c 57static bool newproto;
b1d8e52e 58static int verbose;
a517e88b 59static char *srcpath;
bd269ebc 60static SocketAddress *saddr;
7860a380
PB
61static int persistent = 0;
62static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
a61c6782
PB
63static int shared = 1;
64static int nb_fds;
d0d6ff58
DB
65static QIOChannelSocket *server_ioc;
66static int server_watch = -1;
145614a1 67static QCryptoTLSCreds *tlscreds;
7a5ca864
FB
68
69static void usage(const char *name)
70{
b033cd86 71 (printf) (
7a5ca864
FB
72"Usage: %s [OPTIONS] FILE\n"
73"QEMU Disk Network Block Device Server\n"
74"\n"
713cc671
PL
75" -h, --help display this help and exit\n"
76" -V, --version output version information and exit\n"
b033cd86
PB
77"\n"
78"Connection properties:\n"
713cc671
PL
79" -p, --port=PORT port to listen on (default `%d')\n"
80" -b, --bind=IFACE interface to bind to (default `0.0.0.0')\n"
81" -k, --socket=PATH path to the unix socket\n"
82" (default '"SOCKET_PATH"')\n"
83" -e, --shared=NUM device can be shared by NUM clients (default '1')\n"
84" -t, --persistent don't exit on the last connection\n"
85" -v, --verbose display extra debugging information\n"
332a254b 86" -x, --export-name=NAME expose export by name\n"
b1a75b33 87" -D, --description=TEXT with -x, also export a human-readable description\n"
7a5ca864 88"\n"
b033cd86 89"Exposing part of the image:\n"
713cc671
PL
90" -o, --offset=OFFSET offset into the image\n"
91" -P, --partition=NUM only expose partition NUM\n"
b033cd86 92"\n"
0ab3b337
DB
93"General purpose options:\n"
94" --object type,id=ID,... define an object such as 'secret' for providing\n"
95" passwords and/or encryption keys\n"
39ca463e
DL
96" -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
97" specify tracing options\n"
ffb31e1d
HR
98" --fork fork off the server process and exit the parent\n"
99" once the server is running\n"
b033cd86
PB
100#ifdef __linux__
101"Kernel NBD client support:\n"
713cc671
PL
102" -c, --connect=DEV connect FILE to the local NBD device DEV\n"
103" -d, --disconnect disconnect the specified device\n"
b033cd86
PB
104"\n"
105#endif
106"\n"
107"Block device options:\n"
713cc671
PL
108" -f, --format=FORMAT set image format (raw, qcow2, ...)\n"
109" -r, --read-only export read-only\n"
110" -s, --snapshot use FILE as an external snapshot, create a temporary\n"
111" file with backing_file=FILE, redirect the write to\n"
112" the temporary one\n"
8c116b0e 113" -l, --load-snapshot=SNAPSHOT_PARAM\n"
713cc671
PL
114" load an internal snapshot inside FILE and export it\n"
115" as an read-only device, SNAPSHOT_PARAM format is\n"
116" 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
117" '[ID_OR_NAME]'\n"
118" -n, --nocache disable host cache\n"
119" --cache=MODE set cache mode (none, writeback, ...)\n"
713cc671 120" --aio=MODE set AIO mode (native or threads)\n"
b3838a40 121" --discard=MODE set discard mode (ignore, unmap)\n"
6883de6c 122" --detect-zeroes=MODE set detect-zeroes mode (off, on, unmap)\n"
77c9aaef 123" --image-opts treat FILE as a full set of image options\n"
b033cd86
PB
124"\n"
125"Report bugs to <qemu-devel@nongnu.org>\n"
c2e2872b 126 , name, NBD_DEFAULT_PORT, "DEVICE");
7a5ca864
FB
127}
128
129static void version(const char *name)
130{
131 printf(
315bc7aa 132"%s version 0.0.1\n"
7a5ca864
FB
133"Written by Anthony Liguori.\n"
134"\n"
135"Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
136"This is free software; see the source for copying conditions. There is NO\n"
137"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
315bc7aa 138 , name);
7a5ca864
FB
139}
140
141struct partition_record
142{
143 uint8_t bootable;
144 uint8_t start_head;
145 uint32_t start_cylinder;
146 uint8_t start_sector;
147 uint8_t system;
148 uint8_t end_head;
149 uint8_t end_cylinder;
150 uint8_t end_sector;
151 uint32_t start_sector_abs;
152 uint32_t nb_sectors_abs;
153};
154
155static void read_partition(uint8_t *p, struct partition_record *r)
156{
157 r->bootable = p[0];
158 r->start_head = p[1];
159 r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
160 r->start_sector = p[2] & 0x3f;
161 r->system = p[4];
162 r->end_head = p[5];
163 r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
164 r->end_sector = p[6] & 0x3f;
ac97393d 165
773dce3c
PM
166 r->start_sector_abs = ldl_le_p(p + 8);
167 r->nb_sectors_abs = ldl_le_p(p + 12);
7a5ca864
FB
168}
169
4c58e80a 170static int find_partition(BlockBackend *blk, int partition,
7a5ca864
FB
171 off_t *offset, off_t *size)
172{
173 struct partition_record mbr[4];
bd31c214 174 uint8_t data[MBR_SIZE];
7a5ca864
FB
175 int i;
176 int ext_partnum = 4;
cb7cf0e3 177 int ret;
7a5ca864 178
bd31c214
EB
179 ret = blk_pread(blk, 0, data, sizeof(data));
180 if (ret < 0) {
a4699e55 181 error_report("error while reading: %s", strerror(-ret));
85b01e09 182 exit(EXIT_FAILURE);
cb7cf0e3 183 }
7a5ca864
FB
184
185 if (data[510] != 0x55 || data[511] != 0xaa) {
185b4338 186 return -EINVAL;
7a5ca864
FB
187 }
188
189 for (i = 0; i < 4; i++) {
190 read_partition(&data[446 + 16 * i], &mbr[i]);
191
453b07b1 192 if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
7a5ca864 193 continue;
453b07b1 194 }
7a5ca864
FB
195
196 if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
197 struct partition_record ext[4];
bd31c214 198 uint8_t data1[MBR_SIZE];
7a5ca864
FB
199 int j;
200
bd31c214
EB
201 ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
202 data1, sizeof(data1));
203 if (ret < 0) {
a4699e55 204 error_report("error while reading: %s", strerror(-ret));
85b01e09 205 exit(EXIT_FAILURE);
cb7cf0e3 206 }
7a5ca864
FB
207
208 for (j = 0; j < 4; j++) {
209 read_partition(&data1[446 + 16 * j], &ext[j]);
453b07b1 210 if (!ext[j].system || !ext[j].nb_sectors_abs) {
7a5ca864 211 continue;
453b07b1 212 }
7a5ca864
FB
213
214 if ((ext_partnum + j + 1) == partition) {
215 *offset = (uint64_t)ext[j].start_sector_abs << 9;
216 *size = (uint64_t)ext[j].nb_sectors_abs << 9;
217 return 0;
218 }
219 }
220 ext_partnum += 4;
221 } else if ((i + 1) == partition) {
222 *offset = (uint64_t)mbr[i].start_sector_abs << 9;
223 *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
224 return 0;
225 }
226 }
227
185b4338 228 return -ENOENT;
7a5ca864
FB
229}
230
bb345110
PB
231static void termsig_handler(int signum)
232{
23994a5f 233 atomic_cmpxchg(&state, RUNNING, TERMINATE);
a61c6782 234 qemu_notify_event();
bb345110
PB
235}
236
537b41f5 237
a517e88b 238static void *show_parts(void *arg)
cd831bd7 239{
a6ac2313 240 char *device = arg;
a517e88b
PB
241 int nbd;
242
243 /* linux just needs an open() to trigger
244 * the partition table update
245 * but remember to load the module with max_part != 0 :
246 * modprobe nbd max_part=63
247 */
248 nbd = open(device, O_RDWR);
fc19f8a0 249 if (nbd >= 0) {
a517e88b
PB
250 close(nbd);
251 }
252 return NULL;
253}
cd831bd7 254
a517e88b
PB
255static void *nbd_client_thread(void *arg)
256{
a6ac2313 257 char *device = arg;
a517e88b 258 off_t size;
7423f417 259 uint16_t nbdflags;
d0d6ff58
DB
260 QIOChannelSocket *sioc;
261 int fd;
a517e88b
PB
262 int ret;
263 pthread_t show_parts_thread;
1ce52846 264 Error *local_error = NULL;
a517e88b 265
d0d6ff58
DB
266 sioc = qio_channel_socket_new();
267 if (qio_channel_socket_connect_sync(sioc,
268 saddr,
269 &local_error) < 0) {
48bec07e 270 error_report_err(local_error);
dc10e8b3
SH
271 goto out;
272 }
a517e88b 273
1c778ef7 274 ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL, &nbdflags,
f95910fe 275 NULL, NULL, NULL,
3f472659 276 &size, &local_error);
fc19f8a0 277 if (ret < 0) {
1ce52846 278 if (local_error) {
78288671 279 error_report_err(local_error);
1ce52846 280 }
0c544d73 281 goto out_socket;
a517e88b
PB
282 }
283
a6ac2313 284 fd = open(device, O_RDWR);
fc19f8a0 285 if (fd < 0) {
a6ac2313 286 /* Linux-only, we can use %m in printf. */
b9884681 287 error_report("Failed to open %s: %m", device);
0c544d73 288 goto out_socket;
a6ac2313
PB
289 }
290
1c778ef7 291 ret = nbd_init(fd, sioc, nbdflags, size);
fc19f8a0 292 if (ret < 0) {
0c544d73 293 goto out_fd;
a517e88b
PB
294 }
295
296 /* update partition table */
a6ac2313 297 pthread_create(&show_parts_thread, NULL, show_parts, device);
a517e88b 298
c1f8fdc3
PB
299 if (verbose) {
300 fprintf(stderr, "NBD device %s is now connected to %s\n",
301 device, srcpath);
302 } else {
303 /* Close stderr so that the qemu-nbd process exits. */
304 dup2(STDOUT_FILENO, STDERR_FILENO);
305 }
a517e88b
PB
306
307 ret = nbd_client(fd);
308 if (ret) {
0c544d73 309 goto out_fd;
cd831bd7 310 }
a517e88b 311 close(fd);
d0d6ff58 312 object_unref(OBJECT(sioc));
a517e88b
PB
313 kill(getpid(), SIGTERM);
314 return (void *) EXIT_SUCCESS;
315
0c544d73
PB
316out_fd:
317 close(fd);
318out_socket:
d0d6ff58 319 object_unref(OBJECT(sioc));
a517e88b
PB
320out:
321 kill(getpid(), SIGTERM);
322 return (void *) EXIT_FAILURE;
cd831bd7
TS
323}
324
e4afbf4f 325static int nbd_can_accept(void)
a61c6782
PB
326{
327 return nb_fds < shared;
328}
329
7860a380
PB
330static void nbd_export_closed(NBDExport *exp)
331{
332 assert(state == TERMINATING);
333 state = TERMINATED;
334}
335
d0d6ff58 336static void nbd_update_server_watch(void);
e4afbf4f 337
1743b515 338static void nbd_client_closed(NBDClient *client)
a61c6782 339{
1743b515 340 nb_fds--;
7860a380
PB
341 if (nb_fds == 0 && !persistent && state == RUNNING) {
342 state = TERMINATE;
343 }
d0d6ff58 344 nbd_update_server_watch();
7860a380 345 nbd_client_put(client);
a61c6782
PB
346}
347
d0d6ff58 348static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
a61c6782 349{
d0d6ff58 350 QIOChannelSocket *cioc;
a61c6782 351
d0d6ff58
DB
352 cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
353 NULL);
354 if (!cioc) {
355 return TRUE;
0c544d73
PB
356 }
357
7860a380 358 if (state >= TERMINATE) {
d0d6ff58
DB
359 object_unref(OBJECT(cioc));
360 return TRUE;
7860a380
PB
361 }
362
ee7d7aab 363 nb_fds++;
d0d6ff58 364 nbd_update_server_watch();
f95910fe 365 nbd_client_new(newproto ? NULL : exp, cioc,
145614a1 366 tlscreds, NULL, nbd_client_closed);
d0d6ff58
DB
367 object_unref(OBJECT(cioc));
368
369 return TRUE;
a61c6782
PB
370}
371
d0d6ff58 372static void nbd_update_server_watch(void)
e4afbf4f
FZ
373{
374 if (nbd_can_accept()) {
d0d6ff58
DB
375 if (server_watch == -1) {
376 server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
377 G_IO_IN,
378 nbd_accept,
379 NULL, NULL);
380 }
e4afbf4f 381 } else {
d0d6ff58
DB
382 if (server_watch != -1) {
383 g_source_remove(server_watch);
384 server_watch = -1;
385 }
e4afbf4f
FZ
386 }
387}
388
48bec07e 389
bd269ebc 390static SocketAddress *nbd_build_socket_address(const char *sockpath,
48bec07e
DB
391 const char *bindto,
392 const char *port)
393{
bd269ebc 394 SocketAddress *saddr;
48bec07e 395
bd269ebc 396 saddr = g_new0(SocketAddress, 1);
48bec07e 397 if (sockpath) {
bd269ebc
MA
398 saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
399 saddr->u.q_unix.path = g_strdup(sockpath);
48bec07e 400 } else {
0399293e 401 InetSocketAddress *inet;
bd269ebc
MA
402 saddr->type = SOCKET_ADDRESS_TYPE_INET;
403 inet = &saddr->u.inet;
0399293e 404 inet->host = g_strdup(bindto);
48bec07e 405 if (port) {
0399293e 406 inet->port = g_strdup(port);
48bec07e 407 } else {
0399293e 408 inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
48bec07e
DB
409 }
410 }
411
412 return saddr;
413}
414
415
77c9aaef
DB
416static QemuOptsList file_opts = {
417 .name = "file",
418 .implied_opt_name = "file",
419 .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
420 .desc = {
421 /* no elements => accept any params */
422 { /* end of list */ }
423 },
424};
425
0ab3b337
DB
426static QemuOptsList qemu_object_opts = {
427 .name = "object",
428 .implied_opt_name = "qom-type",
429 .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
430 .desc = {
431 { }
432 },
433};
434
435
145614a1
DB
436
437static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
438{
439 Object *obj;
440 QCryptoTLSCreds *creds;
441
442 obj = object_resolve_path_component(
443 object_get_objects_root(), id);
444 if (!obj) {
445 error_setg(errp, "No TLS credentials with id '%s'",
446 id);
447 return NULL;
448 }
449 creds = (QCryptoTLSCreds *)
450 object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
451 if (!creds) {
452 error_setg(errp, "Object with id '%s' is not TLS credentials",
453 id);
454 return NULL;
455 }
456
457 if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
458 error_setg(errp,
459 "Expecting TLS credentials with a server endpoint");
460 return NULL;
461 }
462 object_ref(obj);
463 return creds;
464}
465
a721f53b
RJ
466static void setup_address_and_port(const char **address, const char **port)
467{
468 if (*address == NULL) {
469 *address = "0.0.0.0";
470 }
471
472 if (*port == NULL) {
473 *port = stringify(NBD_DEFAULT_PORT);
474 }
475}
476
a721f53b
RJ
477/*
478 * Check socket parameters compatibility when socket activation is used.
479 */
480static const char *socket_activation_validate_opts(const char *device,
481 const char *sockpath,
482 const char *address,
483 const char *port)
484{
485 if (device != NULL) {
486 return "NBD device can't be set when using socket activation";
487 }
488
489 if (sockpath != NULL) {
490 return "Unix socket can't be set when using socket activation";
491 }
492
493 if (address != NULL) {
494 return "The interface can't be set when using socket activation";
495 }
496
497 if (port != NULL) {
498 return "TCP port number can't be set when using socket activation";
499 }
500
501 return NULL;
502}
145614a1 503
7a5ca864
FB
504int main(int argc, char **argv)
505{
26f54e9a 506 BlockBackend *blk;
7a5ca864
FB
507 BlockDriverState *bs;
508 off_t dev_offset = 0;
7423f417 509 uint16_t nbdflags = 0;
cd831bd7 510 bool disconnect = false;
a721f53b 511 const char *bindto = NULL;
48bec07e
DB
512 const char *port = NULL;
513 char *sockpath = NULL;
a6ac2313 514 char *device = NULL;
7a5ca864 515 off_t fd_size;
8c116b0e
WX
516 QemuOpts *sn_opts = NULL;
517 const char *sn_id_or_name = NULL;
b1a75b33 518 const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
7a5ca864 519 struct option lopt[] = {
aa6e546c
DB
520 { "help", no_argument, NULL, 'h' },
521 { "version", no_argument, NULL, 'V' },
522 { "bind", required_argument, NULL, 'b' },
523 { "port", required_argument, NULL, 'p' },
524 { "socket", required_argument, NULL, 'k' },
525 { "offset", required_argument, NULL, 'o' },
526 { "read-only", no_argument, NULL, 'r' },
527 { "partition", required_argument, NULL, 'P' },
528 { "connect", required_argument, NULL, 'c' },
529 { "disconnect", no_argument, NULL, 'd' },
530 { "snapshot", no_argument, NULL, 's' },
531 { "load-snapshot", required_argument, NULL, 'l' },
532 { "nocache", no_argument, NULL, 'n' },
533 { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
534 { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
535 { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
536 { "detect-zeroes", required_argument, NULL,
537 QEMU_NBD_OPT_DETECT_ZEROES },
538 { "shared", required_argument, NULL, 'e' },
539 { "format", required_argument, NULL, 'f' },
540 { "persistent", no_argument, NULL, 't' },
541 { "verbose", no_argument, NULL, 'v' },
542 { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
543 { "export-name", required_argument, NULL, 'x' },
b1a75b33 544 { "description", required_argument, NULL, 'D' },
aa6e546c
DB
545 { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
546 { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
39ca463e 547 { "trace", required_argument, NULL, 'T' },
ffb31e1d 548 { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
660f11be 549 { NULL, 0, NULL, 0 }
7a5ca864
FB
550 };
551 int ch;
552 int opt_ind = 0;
7a5ca864 553 char *end;
f5edb014 554 int flags = BDRV_O_RDWR;
7a5ca864 555 int partition = -1;
4fbec260 556 int ret = 0;
39a5235c 557 bool seen_cache = false;
ded9d2d5 558 bool seen_discard = false;
39a5235c 559 bool seen_aio = false;
a517e88b 560 pthread_t client_thread;
e6b63677 561 const char *fmt = NULL;
34b5d2c6 562 Error *local_err = NULL;
b3838a40 563 BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
4fbec260 564 QDict *options = NULL;
3d4b2f9c 565 const char *export_name = NULL;
b1a75b33 566 const char *export_description = NULL;
145614a1 567 const char *tlscredsid = NULL;
77c9aaef 568 bool imageOpts = false;
6effd5bf 569 bool writethrough = true;
39ca463e 570 char *trace_file = NULL;
ffb31e1d
HR
571 bool fork_process = false;
572 int old_stderr = -1;
a721f53b 573 unsigned socket_activation;
7a5ca864 574
a517e88b
PB
575 /* The client thread uses SIGTERM to interrupt the server. A signal
576 * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
577 */
bb345110 578 struct sigaction sa_sigterm;
bb345110
PB
579 memset(&sa_sigterm, 0, sizeof(sa_sigterm));
580 sa_sigterm.sa_handler = termsig_handler;
581 sigaction(SIGTERM, &sa_sigterm, NULL);
c2297088 582
fe4db84d 583 module_call_init(MODULE_INIT_TRACE);
e8f2d272 584 qcrypto_init(&error_fatal);
c2297088 585
0ab3b337
DB
586 module_call_init(MODULE_INIT_QOM);
587 qemu_add_opts(&qemu_object_opts);
39ca463e 588 qemu_add_opts(&qemu_trace_opts);
10f5bff6 589 qemu_init_exec_dir(argv[0]);
bb345110 590
7a5ca864
FB
591 while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
592 switch (ch) {
593 case 's':
2f726488
TS
594 flags |= BDRV_O_SNAPSHOT;
595 break;
596 case 'n':
39a5235c
PB
597 optarg = (char *) "none";
598 /* fallthrough */
599 case QEMU_NBD_OPT_CACHE:
600 if (seen_cache) {
85b01e09
MA
601 error_report("-n and --cache can only be specified once");
602 exit(EXIT_FAILURE);
39a5235c
PB
603 }
604 seen_cache = true;
6effd5bf 605 if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
85b01e09
MA
606 error_report("Invalid cache mode `%s'", optarg);
607 exit(EXIT_FAILURE);
39a5235c 608 }
7a5ca864 609 break;
39a5235c
PB
610 case QEMU_NBD_OPT_AIO:
611 if (seen_aio) {
85b01e09
MA
612 error_report("--aio can only be specified once");
613 exit(EXIT_FAILURE);
39a5235c
PB
614 }
615 seen_aio = true;
616 if (!strcmp(optarg, "native")) {
617 flags |= BDRV_O_NATIVE_AIO;
618 } else if (!strcmp(optarg, "threads")) {
619 /* this is the default */
620 } else {
85b01e09
MA
621 error_report("invalid aio mode `%s'", optarg);
622 exit(EXIT_FAILURE);
39a5235c
PB
623 }
624 break;
ded9d2d5
PB
625 case QEMU_NBD_OPT_DISCARD:
626 if (seen_discard) {
85b01e09
MA
627 error_report("--discard can only be specified once");
628 exit(EXIT_FAILURE);
ded9d2d5
PB
629 }
630 seen_discard = true;
631 if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
85b01e09
MA
632 error_report("Invalid discard mode `%s'", optarg);
633 exit(EXIT_FAILURE);
ded9d2d5
PB
634 }
635 break;
b3838a40
PL
636 case QEMU_NBD_OPT_DETECT_ZEROES:
637 detect_zeroes =
638 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
639 optarg,
7fb1cf16 640 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
b3838a40
PL
641 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
642 &local_err);
643 if (local_err) {
c29b77f9
MA
644 error_reportf_err(local_err,
645 "Failed to parse detect_zeroes mode: ");
85b01e09 646 exit(EXIT_FAILURE);
b3838a40
PL
647 }
648 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
649 !(flags & BDRV_O_UNMAP)) {
85b01e09
MA
650 error_report("setting detect-zeroes to unmap is not allowed "
651 "without setting discard operation to unmap");
652 exit(EXIT_FAILURE);
b3838a40
PL
653 }
654 break;
7a5ca864
FB
655 case 'b':
656 bindto = optarg;
657 break;
658 case 'p':
48bec07e 659 port = optarg;
7a5ca864
FB
660 break;
661 case 'o':
662 dev_offset = strtoll (optarg, &end, 0);
663 if (*end) {
85b01e09
MA
664 error_report("Invalid offset `%s'", optarg);
665 exit(EXIT_FAILURE);
7a5ca864
FB
666 }
667 if (dev_offset < 0) {
85b01e09
MA
668 error_report("Offset must be positive `%s'", optarg);
669 exit(EXIT_FAILURE);
7a5ca864
FB
670 }
671 break;
8c116b0e
WX
672 case 'l':
673 if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
70b94331
MA
674 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
675 optarg, false);
8c116b0e 676 if (!sn_opts) {
85b01e09
MA
677 error_report("Failed in parsing snapshot param `%s'",
678 optarg);
679 exit(EXIT_FAILURE);
8c116b0e
WX
680 }
681 } else {
682 sn_id_or_name = optarg;
683 }
684 /* fall through */
7a5ca864 685 case 'r':
b90fb4b8 686 nbdflags |= NBD_FLAG_READ_ONLY;
07108b29 687 flags &= ~BDRV_O_RDWR;
7a5ca864
FB
688 break;
689 case 'P':
690 partition = strtol(optarg, &end, 0);
713cc671 691 if (*end) {
85b01e09
MA
692 error_report("Invalid partition `%s'", optarg);
693 exit(EXIT_FAILURE);
713cc671
PL
694 }
695 if (partition < 1 || partition > 8) {
85b01e09
MA
696 error_report("Invalid partition %d", partition);
697 exit(EXIT_FAILURE);
713cc671 698 }
7a5ca864 699 break;
cd831bd7 700 case 'k':
b32f6c28 701 sockpath = optarg;
713cc671 702 if (sockpath[0] != '/') {
9af9e0fe 703 error_report("socket path must be absolute");
85b01e09 704 exit(EXIT_FAILURE);
713cc671 705 }
cd831bd7
TS
706 break;
707 case 'd':
708 disconnect = true;
709 break;
710 case 'c':
711 device = optarg;
712 break;
3b05a8e9
TS
713 case 'e':
714 shared = strtol(optarg, &end, 0);
715 if (*end) {
85b01e09
MA
716 error_report("Invalid shared device number '%s'", optarg);
717 exit(EXIT_FAILURE);
3b05a8e9
TS
718 }
719 if (shared < 1) {
9af9e0fe 720 error_report("Shared device number must be greater than 0");
85b01e09 721 exit(EXIT_FAILURE);
3b05a8e9
TS
722 }
723 break;
e6b63677
DB
724 case 'f':
725 fmt = optarg;
726 break;
713cc671
PL
727 case 't':
728 persistent = 1;
729 break;
3d4b2f9c
DB
730 case 'x':
731 export_name = optarg;
732 break;
b1a75b33
EB
733 case 'D':
734 export_description = optarg;
735 break;
7a5ca864
FB
736 case 'v':
737 verbose = 1;
738 break;
739 case 'V':
740 version(argv[0]);
741 exit(0);
742 break;
743 case 'h':
744 usage(argv[0]);
745 exit(0);
746 break;
747 case '?':
85b01e09
MA
748 error_report("Try `%s --help' for more information.", argv[0]);
749 exit(EXIT_FAILURE);
0ab3b337
DB
750 case QEMU_NBD_OPT_OBJECT: {
751 QemuOpts *opts;
752 opts = qemu_opts_parse_noisily(&qemu_object_opts,
753 optarg, true);
754 if (!opts) {
755 exit(EXIT_FAILURE);
756 }
757 } break;
145614a1
DB
758 case QEMU_NBD_OPT_TLSCREDS:
759 tlscredsid = optarg;
760 break;
77c9aaef
DB
761 case QEMU_NBD_OPT_IMAGE_OPTS:
762 imageOpts = true;
763 break;
39ca463e
DL
764 case 'T':
765 g_free(trace_file);
766 trace_file = trace_opt_parse(optarg);
767 break;
ffb31e1d
HR
768 case QEMU_NBD_OPT_FORK:
769 fork_process = true;
770 break;
7a5ca864
FB
771 }
772 }
773
774 if ((argc - optind) != 1) {
433672b0
MA
775 error_report("Invalid number of arguments");
776 error_printf("Try `%s --help' for more information.\n", argv[0]);
85b01e09 777 exit(EXIT_FAILURE);
7a5ca864
FB
778 }
779
0ab3b337
DB
780 if (qemu_opts_foreach(&qemu_object_opts,
781 user_creatable_add_opts_foreach,
51b9b478 782 NULL, NULL)) {
0ab3b337
DB
783 exit(EXIT_FAILURE);
784 }
785
39ca463e
DL
786 if (!trace_init_backends()) {
787 exit(1);
788 }
789 trace_init_file(trace_file);
790 qemu_set_log(LOG_TRACE);
791
a721f53b
RJ
792 socket_activation = check_socket_activation();
793 if (socket_activation == 0) {
794 setup_address_and_port(&bindto, &port);
795 } else {
796 /* Using socket activation - check user didn't use -p etc. */
797 const char *err_msg = socket_activation_validate_opts(device, sockpath,
798 bindto, port);
799 if (err_msg != NULL) {
800 error_report("%s", err_msg);
801 exit(EXIT_FAILURE);
802 }
53fabd4b
PB
803
804 /* qemu-nbd can only listen on a single socket. */
805 if (socket_activation > 1) {
806 error_report("qemu-nbd does not support socket activation with %s > 1",
807 "LISTEN_FDS");
808 exit(EXIT_FAILURE);
809 }
a721f53b
RJ
810 }
811
145614a1
DB
812 if (tlscredsid) {
813 if (sockpath) {
814 error_report("TLS is only supported with IPv4/IPv6");
815 exit(EXIT_FAILURE);
816 }
817 if (device) {
818 error_report("TLS is not supported with a host device");
819 exit(EXIT_FAILURE);
820 }
821 if (!export_name) {
822 /* Set the default NBD protocol export name, since
823 * we *must* use new style protocol for TLS */
824 export_name = "";
825 }
826 tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
827 if (local_err) {
828 error_report("Failed to get TLS creds %s",
829 error_get_pretty(local_err));
830 exit(EXIT_FAILURE);
831 }
832 }
833
cd831bd7 834 if (disconnect) {
d0d6ff58
DB
835 int nbdfd = open(argv[optind], O_RDWR);
836 if (nbdfd < 0) {
85b01e09
MA
837 error_report("Cannot open %s: %s", argv[optind],
838 strerror(errno));
839 exit(EXIT_FAILURE);
fc19f8a0 840 }
d0d6ff58 841 nbd_disconnect(nbdfd);
cd831bd7 842
d0d6ff58 843 close(nbdfd);
cd831bd7
TS
844
845 printf("%s disconnected\n", argv[optind]);
846
713cc671 847 return 0;
cd831bd7
TS
848 }
849
ffb31e1d 850 if ((device && !verbose) || fork_process) {
c1f8fdc3
PB
851 int stderr_fd[2];
852 pid_t pid;
853 int ret;
854
fc19f8a0 855 if (qemu_pipe(stderr_fd) < 0) {
85b01e09
MA
856 error_report("Error setting up communication pipe: %s",
857 strerror(errno));
858 exit(EXIT_FAILURE);
c1f8fdc3
PB
859 }
860
861 /* Now daemonize, but keep a communication channel open to
862 * print errors and exit with the proper status code.
863 */
864 pid = fork();
70d4739e 865 if (pid < 0) {
85b01e09
MA
866 error_report("Failed to fork: %s", strerror(errno));
867 exit(EXIT_FAILURE);
70d4739e 868 } else if (pid == 0) {
c1f8fdc3 869 close(stderr_fd[0]);
9faf31b6 870 ret = qemu_daemon(1, 0);
c1f8fdc3
PB
871
872 /* Temporarily redirect stderr to the parent's pipe... */
ffb31e1d 873 old_stderr = dup(STDERR_FILENO);
c1f8fdc3 874 dup2(stderr_fd[1], STDERR_FILENO);
fc19f8a0 875 if (ret < 0) {
85b01e09
MA
876 error_report("Failed to daemonize: %s", strerror(errno));
877 exit(EXIT_FAILURE);
c1f8fdc3
PB
878 }
879
880 /* ... close the descriptor we inherited and go on. */
881 close(stderr_fd[1]);
882 } else {
883 bool errors = false;
884 char *buf;
885
886 /* In the parent. Print error messages from the child until
887 * it closes the pipe.
888 */
889 close(stderr_fd[1]);
890 buf = g_malloc(1024);
891 while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
892 errors = true;
893 ret = qemu_write_full(STDERR_FILENO, buf, ret);
fc19f8a0 894 if (ret < 0) {
c1f8fdc3
PB
895 exit(EXIT_FAILURE);
896 }
897 }
fc19f8a0 898 if (ret < 0) {
85b01e09
MA
899 error_report("Cannot read from daemon: %s",
900 strerror(errno));
901 exit(EXIT_FAILURE);
c1f8fdc3
PB
902 }
903
904 /* Usually the daemon should not print any message.
905 * Exit with zero status in that case.
906 */
907 exit(errors);
908 }
909 }
910
a6ac2313
PB
911 if (device != NULL && sockpath == NULL) {
912 sockpath = g_malloc(128);
913 snprintf(sockpath, 128, SOCKET_PATH, basename(device));
cd831bd7
TS
914 }
915
a721f53b
RJ
916 if (socket_activation == 0) {
917 server_ioc = qio_channel_socket_new();
918 saddr = nbd_build_socket_address(sockpath, bindto, port);
919 if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
920 object_unref(OBJECT(server_ioc));
921 error_report_err(local_err);
922 return 1;
923 }
924 } else {
925 /* See comment in check_socket_activation above. */
926 assert(socket_activation == 1);
927 server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
928 &local_err);
929 if (server_ioc == NULL) {
930 error_report("Failed to use socket activation: %s",
931 error_get_pretty(local_err));
932 exit(EXIT_FAILURE);
933 }
934 }
48bec07e 935
2f78e491 936 if (qemu_init_main_loop(&local_err)) {
565f65d2 937 error_report_err(local_err);
2f78e491
CN
938 exit(EXIT_FAILURE);
939 }
802ddc37
PB
940 bdrv_init();
941 atexit(bdrv_close_all);
942
77c9aaef
DB
943 srcpath = argv[optind];
944 if (imageOpts) {
945 QemuOpts *opts;
946 if (fmt) {
947 error_report("--image-opts and -f are mutually exclusive");
948 exit(EXIT_FAILURE);
949 }
950 opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
951 if (!opts) {
952 qemu_opts_reset(&file_opts);
953 exit(EXIT_FAILURE);
954 }
955 options = qemu_opts_to_qdict(opts, NULL);
956 qemu_opts_reset(&file_opts);
efaa7c4e 957 blk = blk_new_open(NULL, NULL, options, flags, &local_err);
77c9aaef
DB
958 } else {
959 if (fmt) {
960 options = qdict_new();
46f5ac20 961 qdict_put_str(options, "driver", fmt);
77c9aaef 962 }
efaa7c4e 963 blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
e6b63677
DB
964 }
965
4fbec260 966 if (!blk) {
c29b77f9
MA
967 error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
968 argv[optind]);
85b01e09 969 exit(EXIT_FAILURE);
802ddc37 970 }
4fbec260 971 bs = blk_bs(blk);
802ddc37 972
6effd5bf
KW
973 blk_set_enable_write_cache(blk, !writethrough);
974
8c116b0e
WX
975 if (sn_opts) {
976 ret = bdrv_snapshot_load_tmp(bs,
977 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
978 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
979 &local_err);
980 } else if (sn_id_or_name) {
981 ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
982 &local_err);
983 }
984 if (ret < 0) {
c29b77f9 985 error_reportf_err(local_err, "Failed to load snapshot: ");
85b01e09 986 exit(EXIT_FAILURE);
8c116b0e
WX
987 }
988
b3838a40 989 bs->detect_zeroes = detect_zeroes;
4c58e80a 990 fd_size = blk_getlength(blk);
98f44bbe 991 if (fd_size < 0) {
85b01e09
MA
992 error_report("Failed to determine the image length: %s",
993 strerror(-fd_size));
994 exit(EXIT_FAILURE);
98f44bbe 995 }
802ddc37 996
e424b655
TG
997 if (dev_offset >= fd_size) {
998 error_report("Offset (%lld) has to be smaller than the image size "
999 "(%lld)",
1000 (long long int)dev_offset, (long long int)fd_size);
1001 exit(EXIT_FAILURE);
1002 }
1003 fd_size -= dev_offset;
1004
185b4338 1005 if (partition != -1) {
4c58e80a 1006 ret = find_partition(blk, partition, &dev_offset, &fd_size);
185b4338 1007 if (ret < 0) {
85b01e09 1008 error_report("Could not find partition %d: %s", partition,
a4699e55 1009 strerror(-ret));
85b01e09 1010 exit(EXIT_FAILURE);
185b4338 1011 }
802ddc37
PB
1012 }
1013
cd7fca95
KW
1014 exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1015 writethrough, NULL, &local_err);
98f44bbe 1016 if (!exp) {
4fffeb5e 1017 error_report_err(local_err);
85b01e09 1018 exit(EXIT_FAILURE);
98f44bbe 1019 }
3d4b2f9c
DB
1020 if (export_name) {
1021 nbd_export_set_name(exp, export_name);
b1a75b33 1022 nbd_export_set_description(exp, export_description);
3d4b2f9c 1023 newproto = true;
b1a75b33
EB
1024 } else if (export_description) {
1025 error_report("Export description requires an export name");
1026 exit(EXIT_FAILURE);
3d4b2f9c 1027 }
3b05a8e9 1028
f1ef5555
PB
1029 if (device) {
1030 int ret;
1031
a6ac2313 1032 ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
f1ef5555 1033 if (ret != 0) {
85b01e09
MA
1034 error_report("Failed to create client thread: %s", strerror(ret));
1035 exit(EXIT_FAILURE);
f1ef5555
PB
1036 }
1037 } else {
1038 /* Shut up GCC warnings. */
1039 memset(&client_thread, 0, sizeof(client_thread));
1040 }
1041
d0d6ff58 1042 nbd_update_server_watch();
7a5ca864 1043
9faf31b6
MT
1044 /* now when the initialization is (almost) complete, chdir("/")
1045 * to free any busy filesystems */
1046 if (chdir("/") < 0) {
85b01e09
MA
1047 error_report("Could not chdir to root directory: %s",
1048 strerror(errno));
1049 exit(EXIT_FAILURE);
9faf31b6
MT
1050 }
1051
ffb31e1d
HR
1052 if (fork_process) {
1053 dup2(old_stderr, STDERR_FILENO);
1054 close(old_stderr);
1055 }
1056
7860a380 1057 state = RUNNING;
3b05a8e9 1058 do {
a61c6782 1059 main_loop_wait(false);
7860a380
PB
1060 if (state == TERMINATE) {
1061 state = TERMINATING;
1062 nbd_export_close(exp);
1063 nbd_export_put(exp);
1064 exp = NULL;
1065 }
1066 } while (state != TERMINATED);
7a5ca864 1067
26f54e9a 1068 blk_unref(blk);
b32f6c28
PB
1069 if (sockpath) {
1070 unlink(sockpath);
1071 }
7a5ca864 1072
fbf28a43 1073 qemu_opts_del(sn_opts);
8c116b0e 1074
a517e88b
PB
1075 if (device) {
1076 void *ret;
1077 pthread_join(client_thread, &ret);
1078 exit(ret != NULL);
1079 } else {
1080 exit(EXIT_SUCCESS);
1081 }
7a5ca864 1082}