]> git.proxmox.com Git - mirror_qemu.git/blame - qemu-nbd.c
blockdev: convert internal NBD server to QIONetListener
[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"
c2a3d7da
EB
20#include <getopt.h>
21#include <libgen.h>
22#include <pthread.h>
23
da34e65c 24#include "qapi/error.h"
5a61cb60 25#include "qemu-common.h"
f348b6d1 26#include "qemu/cutils.h"
26f54e9a 27#include "sysemu/block-backend.h"
b3838a40 28#include "block/block_int.h"
737e150e 29#include "block/nbd.h"
6a1751b7 30#include "qemu/main-loop.h"
537b41f5 31#include "qemu/error-report.h"
0ab3b337 32#include "qemu/config-file.h"
58369e22 33#include "qemu/bswap.h"
39ca463e 34#include "qemu/log.h"
53fabd4b 35#include "qemu/systemd.h"
8c116b0e 36#include "block/snapshot.h"
d49b6836 37#include "qapi/qmp/qstring.h"
0ab3b337 38#include "qom/object_interfaces.h"
d0d6ff58 39#include "io/channel-socket.h"
c2297088 40#include "crypto/init.h"
39ca463e 41#include "trace/control.h"
be377133 42#include "qemu-version.h"
7a5ca864 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 124"\n"
f5048cb7 125QEMU_HELP_BOTTOM "\n"
c2e2872b 126 , name, NBD_DEFAULT_PORT, "DEVICE");
7a5ca864
FB
127}
128
129static void version(const char *name)
130{
131 printf(
be377133 132"%s " QEMU_VERSION QEMU_PKGVERSION "\n"
7a5ca864
FB
133"Written by Anthony Liguori.\n"
134"\n"
be377133 135QEMU_COPYRIGHT "\n"
7a5ca864
FB
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;
081dd1fe 258 NBDExportInfo info = { .request_sizes = false, };
d0d6ff58
DB
259 QIOChannelSocket *sioc;
260 int fd;
a517e88b
PB
261 int ret;
262 pthread_t show_parts_thread;
1ce52846 263 Error *local_error = NULL;
a517e88b 264
d0d6ff58
DB
265 sioc = qio_channel_socket_new();
266 if (qio_channel_socket_connect_sync(sioc,
267 saddr,
268 &local_error) < 0) {
48bec07e 269 error_report_err(local_error);
dc10e8b3
SH
270 goto out;
271 }
a517e88b 272
004a89fc
EB
273 ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL,
274 NULL, NULL, NULL, &info, &local_error);
fc19f8a0 275 if (ret < 0) {
1ce52846 276 if (local_error) {
78288671 277 error_report_err(local_error);
1ce52846 278 }
0c544d73 279 goto out_socket;
a517e88b
PB
280 }
281
a6ac2313 282 fd = open(device, O_RDWR);
fc19f8a0 283 if (fd < 0) {
a6ac2313 284 /* Linux-only, we can use %m in printf. */
b9884681 285 error_report("Failed to open %s: %m", device);
0c544d73 286 goto out_socket;
a6ac2313
PB
287 }
288
004a89fc 289 ret = nbd_init(fd, sioc, &info, &local_error);
fc19f8a0 290 if (ret < 0) {
be41c100 291 error_report_err(local_error);
0c544d73 292 goto out_fd;
a517e88b
PB
293 }
294
295 /* update partition table */
a6ac2313 296 pthread_create(&show_parts_thread, NULL, show_parts, device);
a517e88b 297
c1f8fdc3
PB
298 if (verbose) {
299 fprintf(stderr, "NBD device %s is now connected to %s\n",
300 device, srcpath);
301 } else {
302 /* Close stderr so that the qemu-nbd process exits. */
303 dup2(STDOUT_FILENO, STDERR_FILENO);
304 }
a517e88b
PB
305
306 ret = nbd_client(fd);
307 if (ret) {
0c544d73 308 goto out_fd;
cd831bd7 309 }
a517e88b 310 close(fd);
d0d6ff58 311 object_unref(OBJECT(sioc));
a517e88b
PB
312 kill(getpid(), SIGTERM);
313 return (void *) EXIT_SUCCESS;
314
0c544d73
PB
315out_fd:
316 close(fd);
317out_socket:
d0d6ff58 318 object_unref(OBJECT(sioc));
a517e88b
PB
319out:
320 kill(getpid(), SIGTERM);
321 return (void *) EXIT_FAILURE;
cd831bd7
TS
322}
323
e4afbf4f 324static int nbd_can_accept(void)
a61c6782 325{
df8ad9f1 326 return state == RUNNING && nb_fds < shared;
a61c6782
PB
327}
328
7860a380
PB
329static void nbd_export_closed(NBDExport *exp)
330{
331 assert(state == TERMINATING);
332 state = TERMINATED;
333}
334
d0d6ff58 335static void nbd_update_server_watch(void);
e4afbf4f 336
0c9390d9 337static void nbd_client_closed(NBDClient *client, bool negotiated)
a61c6782 338{
1743b515 339 nb_fds--;
0c9390d9 340 if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
7860a380
PB
341 state = TERMINATE;
342 }
d0d6ff58 343 nbd_update_server_watch();
7860a380 344 nbd_client_put(client);
a61c6782
PB
345}
346
d0d6ff58 347static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
a61c6782 348{
d0d6ff58 349 QIOChannelSocket *cioc;
a61c6782 350
d0d6ff58
DB
351 cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
352 NULL);
353 if (!cioc) {
354 return TRUE;
0c544d73
PB
355 }
356
7860a380 357 if (state >= TERMINATE) {
d0d6ff58
DB
358 object_unref(OBJECT(cioc));
359 return TRUE;
7860a380
PB
360 }
361
ee7d7aab 362 nb_fds++;
d0d6ff58 363 nbd_update_server_watch();
f95910fe 364 nbd_client_new(newproto ? NULL : exp, cioc,
145614a1 365 tlscreds, NULL, nbd_client_closed);
d0d6ff58
DB
366 object_unref(OBJECT(cioc));
367
368 return TRUE;
a61c6782
PB
369}
370
d0d6ff58 371static void nbd_update_server_watch(void)
e4afbf4f
FZ
372{
373 if (nbd_can_accept()) {
d0d6ff58
DB
374 if (server_watch == -1) {
375 server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
376 G_IO_IN,
377 nbd_accept,
378 NULL, NULL);
379 }
e4afbf4f 380 } else {
d0d6ff58
DB
381 if (server_watch != -1) {
382 g_source_remove(server_watch);
383 server_watch = -1;
384 }
e4afbf4f
FZ
385 }
386}
387
48bec07e 388
bd269ebc 389static SocketAddress *nbd_build_socket_address(const char *sockpath,
48bec07e
DB
390 const char *bindto,
391 const char *port)
392{
bd269ebc 393 SocketAddress *saddr;
48bec07e 394
bd269ebc 395 saddr = g_new0(SocketAddress, 1);
48bec07e 396 if (sockpath) {
bd269ebc
MA
397 saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
398 saddr->u.q_unix.path = g_strdup(sockpath);
48bec07e 399 } else {
0399293e 400 InetSocketAddress *inet;
bd269ebc
MA
401 saddr->type = SOCKET_ADDRESS_TYPE_INET;
402 inet = &saddr->u.inet;
0399293e 403 inet->host = g_strdup(bindto);
48bec07e 404 if (port) {
0399293e 405 inet->port = g_strdup(port);
48bec07e 406 } else {
0399293e 407 inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
48bec07e
DB
408 }
409 }
410
411 return saddr;
412}
413
414
77c9aaef
DB
415static QemuOptsList file_opts = {
416 .name = "file",
417 .implied_opt_name = "file",
418 .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
419 .desc = {
420 /* no elements => accept any params */
421 { /* end of list */ }
422 },
423};
424
0ab3b337
DB
425static QemuOptsList qemu_object_opts = {
426 .name = "object",
427 .implied_opt_name = "qom-type",
428 .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
429 .desc = {
430 { }
431 },
432};
433
434
145614a1
DB
435
436static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
437{
438 Object *obj;
439 QCryptoTLSCreds *creds;
440
441 obj = object_resolve_path_component(
442 object_get_objects_root(), id);
443 if (!obj) {
444 error_setg(errp, "No TLS credentials with id '%s'",
445 id);
446 return NULL;
447 }
448 creds = (QCryptoTLSCreds *)
449 object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
450 if (!creds) {
451 error_setg(errp, "Object with id '%s' is not TLS credentials",
452 id);
453 return NULL;
454 }
455
456 if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
457 error_setg(errp,
458 "Expecting TLS credentials with a server endpoint");
459 return NULL;
460 }
461 object_ref(obj);
462 return creds;
463}
464
a721f53b
RJ
465static void setup_address_and_port(const char **address, const char **port)
466{
467 if (*address == NULL) {
468 *address = "0.0.0.0";
469 }
470
471 if (*port == NULL) {
472 *port = stringify(NBD_DEFAULT_PORT);
473 }
474}
475
a721f53b
RJ
476/*
477 * Check socket parameters compatibility when socket activation is used.
478 */
479static const char *socket_activation_validate_opts(const char *device,
480 const char *sockpath,
481 const char *address,
482 const char *port)
483{
484 if (device != NULL) {
485 return "NBD device can't be set when using socket activation";
486 }
487
488 if (sockpath != NULL) {
489 return "Unix socket can't be set when using socket activation";
490 }
491
492 if (address != NULL) {
493 return "The interface can't be set when using socket activation";
494 }
495
496 if (port != NULL) {
497 return "TCP port number can't be set when using socket activation";
498 }
499
500 return NULL;
501}
145614a1 502
7a5ca864
FB
503int main(int argc, char **argv)
504{
26f54e9a 505 BlockBackend *blk;
7a5ca864
FB
506 BlockDriverState *bs;
507 off_t dev_offset = 0;
7423f417 508 uint16_t nbdflags = 0;
cd831bd7 509 bool disconnect = false;
a721f53b 510 const char *bindto = NULL;
48bec07e
DB
511 const char *port = NULL;
512 char *sockpath = NULL;
a6ac2313 513 char *device = NULL;
7a5ca864 514 off_t fd_size;
8c116b0e
WX
515 QemuOpts *sn_opts = NULL;
516 const char *sn_id_or_name = NULL;
b1a75b33 517 const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
7a5ca864 518 struct option lopt[] = {
aa6e546c
DB
519 { "help", no_argument, NULL, 'h' },
520 { "version", no_argument, NULL, 'V' },
521 { "bind", required_argument, NULL, 'b' },
522 { "port", required_argument, NULL, 'p' },
523 { "socket", required_argument, NULL, 'k' },
524 { "offset", required_argument, NULL, 'o' },
525 { "read-only", no_argument, NULL, 'r' },
526 { "partition", required_argument, NULL, 'P' },
527 { "connect", required_argument, NULL, 'c' },
528 { "disconnect", no_argument, NULL, 'd' },
529 { "snapshot", no_argument, NULL, 's' },
530 { "load-snapshot", required_argument, NULL, 'l' },
531 { "nocache", no_argument, NULL, 'n' },
532 { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
533 { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
534 { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
535 { "detect-zeroes", required_argument, NULL,
536 QEMU_NBD_OPT_DETECT_ZEROES },
537 { "shared", required_argument, NULL, 'e' },
538 { "format", required_argument, NULL, 'f' },
539 { "persistent", no_argument, NULL, 't' },
540 { "verbose", no_argument, NULL, 'v' },
541 { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
542 { "export-name", required_argument, NULL, 'x' },
b1a75b33 543 { "description", required_argument, NULL, 'D' },
aa6e546c
DB
544 { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
545 { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
39ca463e 546 { "trace", required_argument, NULL, 'T' },
ffb31e1d 547 { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
660f11be 548 { NULL, 0, NULL, 0 }
7a5ca864
FB
549 };
550 int ch;
551 int opt_ind = 0;
7a5ca864 552 char *end;
f5edb014 553 int flags = BDRV_O_RDWR;
7a5ca864 554 int partition = -1;
4fbec260 555 int ret = 0;
39a5235c 556 bool seen_cache = false;
ded9d2d5 557 bool seen_discard = false;
39a5235c 558 bool seen_aio = false;
a517e88b 559 pthread_t client_thread;
e6b63677 560 const char *fmt = NULL;
34b5d2c6 561 Error *local_err = NULL;
b3838a40 562 BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
4fbec260 563 QDict *options = NULL;
3d4b2f9c 564 const char *export_name = NULL;
b1a75b33 565 const char *export_description = NULL;
145614a1 566 const char *tlscredsid = NULL;
77c9aaef 567 bool imageOpts = false;
6effd5bf 568 bool writethrough = true;
39ca463e 569 char *trace_file = NULL;
ffb31e1d
HR
570 bool fork_process = false;
571 int old_stderr = -1;
a721f53b 572 unsigned socket_activation;
7a5ca864 573
a517e88b
PB
574 /* The client thread uses SIGTERM to interrupt the server. A signal
575 * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
576 */
bb345110 577 struct sigaction sa_sigterm;
bb345110
PB
578 memset(&sa_sigterm, 0, sizeof(sa_sigterm));
579 sa_sigterm.sa_handler = termsig_handler;
580 sigaction(SIGTERM, &sa_sigterm, NULL);
c2297088 581
041e32b8
HR
582#ifdef CONFIG_POSIX
583 signal(SIGPIPE, SIG_IGN);
584#endif
585
fe4db84d 586 module_call_init(MODULE_INIT_TRACE);
e8f2d272 587 qcrypto_init(&error_fatal);
c2297088 588
0ab3b337
DB
589 module_call_init(MODULE_INIT_QOM);
590 qemu_add_opts(&qemu_object_opts);
39ca463e 591 qemu_add_opts(&qemu_trace_opts);
10f5bff6 592 qemu_init_exec_dir(argv[0]);
bb345110 593
7a5ca864
FB
594 while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
595 switch (ch) {
596 case 's':
2f726488
TS
597 flags |= BDRV_O_SNAPSHOT;
598 break;
599 case 'n':
39a5235c
PB
600 optarg = (char *) "none";
601 /* fallthrough */
602 case QEMU_NBD_OPT_CACHE:
603 if (seen_cache) {
85b01e09
MA
604 error_report("-n and --cache can only be specified once");
605 exit(EXIT_FAILURE);
39a5235c
PB
606 }
607 seen_cache = true;
6effd5bf 608 if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
85b01e09
MA
609 error_report("Invalid cache mode `%s'", optarg);
610 exit(EXIT_FAILURE);
39a5235c 611 }
7a5ca864 612 break;
39a5235c
PB
613 case QEMU_NBD_OPT_AIO:
614 if (seen_aio) {
85b01e09
MA
615 error_report("--aio can only be specified once");
616 exit(EXIT_FAILURE);
39a5235c
PB
617 }
618 seen_aio = true;
619 if (!strcmp(optarg, "native")) {
620 flags |= BDRV_O_NATIVE_AIO;
621 } else if (!strcmp(optarg, "threads")) {
622 /* this is the default */
623 } else {
85b01e09
MA
624 error_report("invalid aio mode `%s'", optarg);
625 exit(EXIT_FAILURE);
39a5235c
PB
626 }
627 break;
ded9d2d5
PB
628 case QEMU_NBD_OPT_DISCARD:
629 if (seen_discard) {
85b01e09
MA
630 error_report("--discard can only be specified once");
631 exit(EXIT_FAILURE);
ded9d2d5
PB
632 }
633 seen_discard = true;
634 if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
85b01e09
MA
635 error_report("Invalid discard mode `%s'", optarg);
636 exit(EXIT_FAILURE);
ded9d2d5
PB
637 }
638 break;
b3838a40
PL
639 case QEMU_NBD_OPT_DETECT_ZEROES:
640 detect_zeroes =
f7abe0ec 641 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
b3838a40 642 optarg,
b3838a40
PL
643 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
644 &local_err);
645 if (local_err) {
c29b77f9
MA
646 error_reportf_err(local_err,
647 "Failed to parse detect_zeroes mode: ");
85b01e09 648 exit(EXIT_FAILURE);
b3838a40
PL
649 }
650 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
651 !(flags & BDRV_O_UNMAP)) {
85b01e09
MA
652 error_report("setting detect-zeroes to unmap is not allowed "
653 "without setting discard operation to unmap");
654 exit(EXIT_FAILURE);
b3838a40
PL
655 }
656 break;
7a5ca864
FB
657 case 'b':
658 bindto = optarg;
659 break;
660 case 'p':
48bec07e 661 port = optarg;
7a5ca864
FB
662 break;
663 case 'o':
664 dev_offset = strtoll (optarg, &end, 0);
665 if (*end) {
85b01e09
MA
666 error_report("Invalid offset `%s'", optarg);
667 exit(EXIT_FAILURE);
7a5ca864
FB
668 }
669 if (dev_offset < 0) {
85b01e09
MA
670 error_report("Offset must be positive `%s'", optarg);
671 exit(EXIT_FAILURE);
7a5ca864
FB
672 }
673 break;
8c116b0e
WX
674 case 'l':
675 if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
70b94331
MA
676 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
677 optarg, false);
8c116b0e 678 if (!sn_opts) {
85b01e09
MA
679 error_report("Failed in parsing snapshot param `%s'",
680 optarg);
681 exit(EXIT_FAILURE);
8c116b0e
WX
682 }
683 } else {
684 sn_id_or_name = optarg;
685 }
686 /* fall through */
7a5ca864 687 case 'r':
b90fb4b8 688 nbdflags |= NBD_FLAG_READ_ONLY;
07108b29 689 flags &= ~BDRV_O_RDWR;
7a5ca864
FB
690 break;
691 case 'P':
692 partition = strtol(optarg, &end, 0);
713cc671 693 if (*end) {
85b01e09
MA
694 error_report("Invalid partition `%s'", optarg);
695 exit(EXIT_FAILURE);
713cc671
PL
696 }
697 if (partition < 1 || partition > 8) {
85b01e09
MA
698 error_report("Invalid partition %d", partition);
699 exit(EXIT_FAILURE);
713cc671 700 }
7a5ca864 701 break;
cd831bd7 702 case 'k':
b32f6c28 703 sockpath = optarg;
713cc671 704 if (sockpath[0] != '/') {
9af9e0fe 705 error_report("socket path must be absolute");
85b01e09 706 exit(EXIT_FAILURE);
713cc671 707 }
cd831bd7
TS
708 break;
709 case 'd':
710 disconnect = true;
711 break;
712 case 'c':
713 device = optarg;
714 break;
3b05a8e9
TS
715 case 'e':
716 shared = strtol(optarg, &end, 0);
717 if (*end) {
85b01e09
MA
718 error_report("Invalid shared device number '%s'", optarg);
719 exit(EXIT_FAILURE);
3b05a8e9
TS
720 }
721 if (shared < 1) {
9af9e0fe 722 error_report("Shared device number must be greater than 0");
85b01e09 723 exit(EXIT_FAILURE);
3b05a8e9
TS
724 }
725 break;
e6b63677
DB
726 case 'f':
727 fmt = optarg;
728 break;
713cc671
PL
729 case 't':
730 persistent = 1;
731 break;
3d4b2f9c
DB
732 case 'x':
733 export_name = optarg;
734 break;
b1a75b33
EB
735 case 'D':
736 export_description = optarg;
737 break;
7a5ca864
FB
738 case 'v':
739 verbose = 1;
740 break;
741 case 'V':
742 version(argv[0]);
743 exit(0);
744 break;
745 case 'h':
746 usage(argv[0]);
747 exit(0);
748 break;
749 case '?':
85b01e09
MA
750 error_report("Try `%s --help' for more information.", argv[0]);
751 exit(EXIT_FAILURE);
0ab3b337
DB
752 case QEMU_NBD_OPT_OBJECT: {
753 QemuOpts *opts;
754 opts = qemu_opts_parse_noisily(&qemu_object_opts,
755 optarg, true);
756 if (!opts) {
757 exit(EXIT_FAILURE);
758 }
759 } break;
145614a1
DB
760 case QEMU_NBD_OPT_TLSCREDS:
761 tlscredsid = optarg;
762 break;
77c9aaef
DB
763 case QEMU_NBD_OPT_IMAGE_OPTS:
764 imageOpts = true;
765 break;
39ca463e
DL
766 case 'T':
767 g_free(trace_file);
768 trace_file = trace_opt_parse(optarg);
769 break;
ffb31e1d
HR
770 case QEMU_NBD_OPT_FORK:
771 fork_process = true;
772 break;
7a5ca864
FB
773 }
774 }
775
776 if ((argc - optind) != 1) {
433672b0
MA
777 error_report("Invalid number of arguments");
778 error_printf("Try `%s --help' for more information.\n", argv[0]);
85b01e09 779 exit(EXIT_FAILURE);
7a5ca864
FB
780 }
781
0ab3b337
DB
782 if (qemu_opts_foreach(&qemu_object_opts,
783 user_creatable_add_opts_foreach,
51b9b478 784 NULL, NULL)) {
0ab3b337
DB
785 exit(EXIT_FAILURE);
786 }
787
39ca463e
DL
788 if (!trace_init_backends()) {
789 exit(1);
790 }
791 trace_init_file(trace_file);
792 qemu_set_log(LOG_TRACE);
793
a721f53b
RJ
794 socket_activation = check_socket_activation();
795 if (socket_activation == 0) {
796 setup_address_and_port(&bindto, &port);
797 } else {
798 /* Using socket activation - check user didn't use -p etc. */
799 const char *err_msg = socket_activation_validate_opts(device, sockpath,
800 bindto, port);
801 if (err_msg != NULL) {
802 error_report("%s", err_msg);
803 exit(EXIT_FAILURE);
804 }
53fabd4b
PB
805
806 /* qemu-nbd can only listen on a single socket. */
807 if (socket_activation > 1) {
808 error_report("qemu-nbd does not support socket activation with %s > 1",
809 "LISTEN_FDS");
810 exit(EXIT_FAILURE);
811 }
a721f53b
RJ
812 }
813
145614a1
DB
814 if (tlscredsid) {
815 if (sockpath) {
816 error_report("TLS is only supported with IPv4/IPv6");
817 exit(EXIT_FAILURE);
818 }
819 if (device) {
820 error_report("TLS is not supported with a host device");
821 exit(EXIT_FAILURE);
822 }
823 if (!export_name) {
824 /* Set the default NBD protocol export name, since
825 * we *must* use new style protocol for TLS */
826 export_name = "";
827 }
828 tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
829 if (local_err) {
830 error_report("Failed to get TLS creds %s",
831 error_get_pretty(local_err));
832 exit(EXIT_FAILURE);
833 }
834 }
835
cd831bd7 836 if (disconnect) {
d0d6ff58
DB
837 int nbdfd = open(argv[optind], O_RDWR);
838 if (nbdfd < 0) {
85b01e09
MA
839 error_report("Cannot open %s: %s", argv[optind],
840 strerror(errno));
841 exit(EXIT_FAILURE);
fc19f8a0 842 }
d0d6ff58 843 nbd_disconnect(nbdfd);
cd831bd7 844
d0d6ff58 845 close(nbdfd);
cd831bd7
TS
846
847 printf("%s disconnected\n", argv[optind]);
848
713cc671 849 return 0;
cd831bd7
TS
850 }
851
ffb31e1d 852 if ((device && !verbose) || fork_process) {
c1f8fdc3
PB
853 int stderr_fd[2];
854 pid_t pid;
855 int ret;
856
fc19f8a0 857 if (qemu_pipe(stderr_fd) < 0) {
85b01e09
MA
858 error_report("Error setting up communication pipe: %s",
859 strerror(errno));
860 exit(EXIT_FAILURE);
c1f8fdc3
PB
861 }
862
863 /* Now daemonize, but keep a communication channel open to
864 * print errors and exit with the proper status code.
865 */
866 pid = fork();
70d4739e 867 if (pid < 0) {
85b01e09
MA
868 error_report("Failed to fork: %s", strerror(errno));
869 exit(EXIT_FAILURE);
70d4739e 870 } else if (pid == 0) {
c1f8fdc3 871 close(stderr_fd[0]);
9faf31b6 872 ret = qemu_daemon(1, 0);
c1f8fdc3
PB
873
874 /* Temporarily redirect stderr to the parent's pipe... */
ffb31e1d 875 old_stderr = dup(STDERR_FILENO);
c1f8fdc3 876 dup2(stderr_fd[1], STDERR_FILENO);
fc19f8a0 877 if (ret < 0) {
85b01e09
MA
878 error_report("Failed to daemonize: %s", strerror(errno));
879 exit(EXIT_FAILURE);
c1f8fdc3
PB
880 }
881
882 /* ... close the descriptor we inherited and go on. */
883 close(stderr_fd[1]);
884 } else {
885 bool errors = false;
886 char *buf;
887
888 /* In the parent. Print error messages from the child until
889 * it closes the pipe.
890 */
891 close(stderr_fd[1]);
892 buf = g_malloc(1024);
893 while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
894 errors = true;
895 ret = qemu_write_full(STDERR_FILENO, buf, ret);
fc19f8a0 896 if (ret < 0) {
c1f8fdc3
PB
897 exit(EXIT_FAILURE);
898 }
899 }
fc19f8a0 900 if (ret < 0) {
85b01e09
MA
901 error_report("Cannot read from daemon: %s",
902 strerror(errno));
903 exit(EXIT_FAILURE);
c1f8fdc3
PB
904 }
905
906 /* Usually the daemon should not print any message.
907 * Exit with zero status in that case.
908 */
909 exit(errors);
910 }
911 }
912
a6ac2313
PB
913 if (device != NULL && sockpath == NULL) {
914 sockpath = g_malloc(128);
915 snprintf(sockpath, 128, SOCKET_PATH, basename(device));
cd831bd7
TS
916 }
917
a721f53b
RJ
918 if (socket_activation == 0) {
919 server_ioc = qio_channel_socket_new();
920 saddr = nbd_build_socket_address(sockpath, bindto, port);
921 if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
922 object_unref(OBJECT(server_ioc));
923 error_report_err(local_err);
924 return 1;
925 }
926 } else {
927 /* See comment in check_socket_activation above. */
928 assert(socket_activation == 1);
929 server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
930 &local_err);
931 if (server_ioc == NULL) {
932 error_report("Failed to use socket activation: %s",
933 error_get_pretty(local_err));
934 exit(EXIT_FAILURE);
935 }
936 }
48bec07e 937
2f78e491 938 if (qemu_init_main_loop(&local_err)) {
565f65d2 939 error_report_err(local_err);
2f78e491
CN
940 exit(EXIT_FAILURE);
941 }
802ddc37
PB
942 bdrv_init();
943 atexit(bdrv_close_all);
944
77c9aaef
DB
945 srcpath = argv[optind];
946 if (imageOpts) {
947 QemuOpts *opts;
948 if (fmt) {
949 error_report("--image-opts and -f are mutually exclusive");
950 exit(EXIT_FAILURE);
951 }
952 opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
953 if (!opts) {
954 qemu_opts_reset(&file_opts);
955 exit(EXIT_FAILURE);
956 }
957 options = qemu_opts_to_qdict(opts, NULL);
958 qemu_opts_reset(&file_opts);
efaa7c4e 959 blk = blk_new_open(NULL, NULL, options, flags, &local_err);
77c9aaef
DB
960 } else {
961 if (fmt) {
962 options = qdict_new();
46f5ac20 963 qdict_put_str(options, "driver", fmt);
77c9aaef 964 }
efaa7c4e 965 blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
e6b63677
DB
966 }
967
4fbec260 968 if (!blk) {
c29b77f9
MA
969 error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
970 argv[optind]);
85b01e09 971 exit(EXIT_FAILURE);
802ddc37 972 }
4fbec260 973 bs = blk_bs(blk);
802ddc37 974
6effd5bf
KW
975 blk_set_enable_write_cache(blk, !writethrough);
976
8c116b0e
WX
977 if (sn_opts) {
978 ret = bdrv_snapshot_load_tmp(bs,
979 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
980 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
981 &local_err);
982 } else if (sn_id_or_name) {
983 ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
984 &local_err);
985 }
986 if (ret < 0) {
c29b77f9 987 error_reportf_err(local_err, "Failed to load snapshot: ");
85b01e09 988 exit(EXIT_FAILURE);
8c116b0e
WX
989 }
990
b3838a40 991 bs->detect_zeroes = detect_zeroes;
4c58e80a 992 fd_size = blk_getlength(blk);
98f44bbe 993 if (fd_size < 0) {
85b01e09
MA
994 error_report("Failed to determine the image length: %s",
995 strerror(-fd_size));
996 exit(EXIT_FAILURE);
98f44bbe 997 }
802ddc37 998
e424b655
TG
999 if (dev_offset >= fd_size) {
1000 error_report("Offset (%lld) has to be smaller than the image size "
1001 "(%lld)",
1002 (long long int)dev_offset, (long long int)fd_size);
1003 exit(EXIT_FAILURE);
1004 }
1005 fd_size -= dev_offset;
1006
185b4338 1007 if (partition != -1) {
4c58e80a 1008 ret = find_partition(blk, partition, &dev_offset, &fd_size);
185b4338 1009 if (ret < 0) {
85b01e09 1010 error_report("Could not find partition %d: %s", partition,
a4699e55 1011 strerror(-ret));
85b01e09 1012 exit(EXIT_FAILURE);
185b4338 1013 }
802ddc37
PB
1014 }
1015
cd7fca95
KW
1016 exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1017 writethrough, NULL, &local_err);
98f44bbe 1018 if (!exp) {
4fffeb5e 1019 error_report_err(local_err);
85b01e09 1020 exit(EXIT_FAILURE);
98f44bbe 1021 }
3d4b2f9c
DB
1022 if (export_name) {
1023 nbd_export_set_name(exp, export_name);
b1a75b33 1024 nbd_export_set_description(exp, export_description);
3d4b2f9c 1025 newproto = true;
b1a75b33
EB
1026 } else if (export_description) {
1027 error_report("Export description requires an export name");
1028 exit(EXIT_FAILURE);
3d4b2f9c 1029 }
3b05a8e9 1030
f1ef5555
PB
1031 if (device) {
1032 int ret;
1033
a6ac2313 1034 ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
f1ef5555 1035 if (ret != 0) {
85b01e09
MA
1036 error_report("Failed to create client thread: %s", strerror(ret));
1037 exit(EXIT_FAILURE);
f1ef5555
PB
1038 }
1039 } else {
1040 /* Shut up GCC warnings. */
1041 memset(&client_thread, 0, sizeof(client_thread));
1042 }
1043
d0d6ff58 1044 nbd_update_server_watch();
7a5ca864 1045
9faf31b6
MT
1046 /* now when the initialization is (almost) complete, chdir("/")
1047 * to free any busy filesystems */
1048 if (chdir("/") < 0) {
85b01e09
MA
1049 error_report("Could not chdir to root directory: %s",
1050 strerror(errno));
1051 exit(EXIT_FAILURE);
9faf31b6
MT
1052 }
1053
ffb31e1d
HR
1054 if (fork_process) {
1055 dup2(old_stderr, STDERR_FILENO);
1056 close(old_stderr);
1057 }
1058
7860a380 1059 state = RUNNING;
3b05a8e9 1060 do {
a61c6782 1061 main_loop_wait(false);
7860a380
PB
1062 if (state == TERMINATE) {
1063 state = TERMINATING;
1064 nbd_export_close(exp);
1065 nbd_export_put(exp);
1066 exp = NULL;
1067 }
1068 } while (state != TERMINATED);
7a5ca864 1069
26f54e9a 1070 blk_unref(blk);
b32f6c28
PB
1071 if (sockpath) {
1072 unlink(sockpath);
1073 }
7a5ca864 1074
fbf28a43 1075 qemu_opts_del(sn_opts);
8c116b0e 1076
a517e88b
PB
1077 if (device) {
1078 void *ret;
1079 pthread_join(client_thread, &ret);
1080 exit(ret != NULL);
1081 } else {
1082 exit(EXIT_SUCCESS);
1083 }
7a5ca864 1084}