]> git.proxmox.com Git - mirror_qemu.git/blame - qemu-nbd.c
Merge tag 'pull-jobs-2021-11-09' of https://src.openvz.org/scm/~vsementsov/qemu into...
[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
a8d25326 24#include "qemu-common.h"
da34e65c 25#include "qapi/error.h"
f348b6d1 26#include "qemu/cutils.h"
26f54e9a 27#include "sysemu/block-backend.h"
cbc20bfb 28#include "sysemu/runstate.h" /* for qemu_system_killed() prototype */
b3838a40 29#include "block/block_int.h"
737e150e 30#include "block/nbd.h"
6a1751b7 31#include "qemu/main-loop.h"
0b8fa32f 32#include "qemu/module.h"
922a01a0 33#include "qemu/option.h"
537b41f5 34#include "qemu/error-report.h"
0ab3b337 35#include "qemu/config-file.h"
58369e22 36#include "qemu/bswap.h"
39ca463e 37#include "qemu/log.h"
53fabd4b 38#include "qemu/systemd.h"
8c116b0e 39#include "block/snapshot.h"
452fcdbc 40#include "qapi/qmp/qdict.h"
d49b6836 41#include "qapi/qmp/qstring.h"
0ab3b337 42#include "qom/object_interfaces.h"
d0d6ff58 43#include "io/channel-socket.h"
e4849c1d 44#include "io/net-listener.h"
c2297088 45#include "crypto/init.h"
0279cd95 46#include "crypto/tlscreds.h"
39ca463e 47#include "trace/control.h"
be377133 48#include "qemu-version.h"
7a5ca864 49
3c1fa35d
EB
50#ifdef __linux__
51#define HAVE_NBD_DEVICE 1
52#else
53#define HAVE_NBD_DEVICE 0
54#endif
55
713cc671 56#define SOCKET_PATH "/var/lock/qemu-nbd-%s"
fa8b7ce2
DB
57#define QEMU_NBD_OPT_CACHE 256
58#define QEMU_NBD_OPT_AIO 257
59#define QEMU_NBD_OPT_DISCARD 258
60#define QEMU_NBD_OPT_DETECT_ZEROES 259
61#define QEMU_NBD_OPT_OBJECT 260
62#define QEMU_NBD_OPT_TLSCREDS 261
63#define QEMU_NBD_OPT_IMAGE_OPTS 262
ffb31e1d 64#define QEMU_NBD_OPT_FORK 263
b25e12da 65#define QEMU_NBD_OPT_TLSAUTHZ 264
637bc5a5 66#define QEMU_NBD_OPT_PID_FILE 265
7a5ca864 67
bd31c214
EB
68#define MBR_SIZE 512
69
b1d8e52e 70static int verbose;
a517e88b 71static char *srcpath;
bd269ebc 72static SocketAddress *saddr;
7860a380 73static int persistent = 0;
d794f7f3 74static enum { RUNNING, TERMINATE, TERMINATED } state;
a61c6782
PB
75static int shared = 1;
76static int nb_fds;
e4849c1d 77static QIONetListener *server;
145614a1 78static QCryptoTLSCreds *tlscreds;
b25e12da 79static const char *tlsauthz;
7a5ca864
FB
80
81static void usage(const char *name)
82{
b033cd86 83 (printf) (
7a5ca864 84"Usage: %s [OPTIONS] FILE\n"
68b96f15
EB
85" or: %s -L [OPTIONS]\n"
86"QEMU Disk Network Block Device Utility\n"
7a5ca864 87"\n"
713cc671
PL
88" -h, --help display this help and exit\n"
89" -V, --version output version information and exit\n"
b033cd86
PB
90"\n"
91"Connection properties:\n"
713cc671
PL
92" -p, --port=PORT port to listen on (default `%d')\n"
93" -b, --bind=IFACE interface to bind to (default `0.0.0.0')\n"
94" -k, --socket=PATH path to the unix socket\n"
95" (default '"SOCKET_PATH"')\n"
96" -e, --shared=NUM device can be shared by NUM clients (default '1')\n"
97" -t, --persistent don't exit on the last connection\n"
98" -v, --verbose display extra debugging information\n"
f5cd0bb5
VSO
99" -x, --export-name=NAME expose export by name (default is empty string)\n"
100" -D, --description=TEXT export a human-readable description\n"
7a5ca864 101"\n"
b033cd86 102"Exposing part of the image:\n"
713cc671 103" -o, --offset=OFFSET offset into the image\n"
dbc7b014 104" -A, --allocation-depth expose the allocation depth\n"
636192c4 105" -B, --bitmap=NAME expose a persistent dirty bitmap\n"
b033cd86 106"\n"
0ab3b337 107"General purpose options:\n"
68b96f15 108" -L, --list list exports available from another NBD server\n"
0ab3b337
DB
109" --object type,id=ID,... define an object such as 'secret' for providing\n"
110" passwords and/or encryption keys\n"
f7812df7 111" --tls-creds=ID use id of an earlier --object to provide TLS\n"
b25e12da
DB
112" --tls-authz=ID use id of an earlier --object to provide\n"
113" authorization\n"
39ca463e
DL
114" -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
115" specify tracing options\n"
ffb31e1d
HR
116" --fork fork off the server process and exit the parent\n"
117" once the server is running\n"
637bc5a5 118" --pid-file=PATH store the server's process ID in the given file\n"
3c1fa35d
EB
119#if HAVE_NBD_DEVICE
120"\n"
b033cd86 121"Kernel NBD client support:\n"
713cc671
PL
122" -c, --connect=DEV connect FILE to the local NBD device DEV\n"
123" -d, --disconnect disconnect the specified device\n"
b033cd86
PB
124#endif
125"\n"
126"Block device options:\n"
713cc671
PL
127" -f, --format=FORMAT set image format (raw, qcow2, ...)\n"
128" -r, --read-only export read-only\n"
129" -s, --snapshot use FILE as an external snapshot, create a temporary\n"
130" file with backing_file=FILE, redirect the write to\n"
131" the temporary one\n"
8c116b0e 132" -l, --load-snapshot=SNAPSHOT_PARAM\n"
713cc671
PL
133" load an internal snapshot inside FILE and export it\n"
134" as an read-only device, SNAPSHOT_PARAM format is\n"
135" 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
136" '[ID_OR_NAME]'\n"
137" -n, --nocache disable host cache\n"
09615257
NS
138" --cache=MODE set cache mode used to access the disk image, the\n"
139" valid options are: 'none', 'writeback' (default),\n"
140" 'writethrough', 'directsync' and 'unsafe'\n"
7680274d 141" --aio=MODE set AIO mode (native, io_uring or threads)\n"
b3838a40 142" --discard=MODE set discard mode (ignore, unmap)\n"
6883de6c 143" --detect-zeroes=MODE set detect-zeroes mode (off, on, unmap)\n"
77c9aaef 144" --image-opts treat FILE as a full set of image options\n"
b033cd86 145"\n"
f5048cb7 146QEMU_HELP_BOTTOM "\n"
68b96f15 147 , name, name, NBD_DEFAULT_PORT, "DEVICE");
7a5ca864
FB
148}
149
150static void version(const char *name)
151{
152 printf(
7e563bfb 153"%s " QEMU_FULL_VERSION "\n"
7a5ca864
FB
154"Written by Anthony Liguori.\n"
155"\n"
be377133 156QEMU_COPYRIGHT "\n"
7a5ca864
FB
157"This is free software; see the source for copying conditions. There is NO\n"
158"warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
315bc7aa 159 , name);
7a5ca864
FB
160}
161
029a88c9 162#ifdef CONFIG_POSIX
cbc20bfb
SH
163/*
164 * The client thread uses SIGTERM to interrupt the server. A signal
165 * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
166 */
167void qemu_system_killed(int signum, pid_t pid)
bb345110 168{
d73415a3 169 qatomic_cmpxchg(&state, RUNNING, TERMINATE);
a61c6782 170 qemu_notify_event();
bb345110 171}
029a88c9 172#endif /* CONFIG_POSIX */
537b41f5 173
68b96f15
EB
174static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
175 const char *hostname)
176{
177 int ret = EXIT_FAILURE;
178 int rc;
179 Error *err = NULL;
180 QIOChannelSocket *sioc;
181 NBDExportInfo *list;
182 int i, j;
183
184 sioc = qio_channel_socket_new();
185 if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
186 error_report_err(err);
992809bf 187 goto out;
68b96f15
EB
188 }
189 rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
190 &err);
191 if (rc < 0) {
192 if (err) {
193 error_report_err(err);
194 }
195 goto out;
196 }
197 printf("exports available: %d\n", rc);
198 for (i = 0; i < rc; i++) {
199 printf(" export: '%s'\n", list[i].name);
200 if (list[i].description && *list[i].description) {
201 printf(" description: %s\n", list[i].description);
202 }
203 if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
c4e2aff8
HR
204 static const char *const flag_names[] = {
205 [NBD_FLAG_READ_ONLY_BIT] = "readonly",
206 [NBD_FLAG_SEND_FLUSH_BIT] = "flush",
207 [NBD_FLAG_SEND_FUA_BIT] = "fua",
208 [NBD_FLAG_ROTATIONAL_BIT] = "rotational",
209 [NBD_FLAG_SEND_TRIM_BIT] = "trim",
210 [NBD_FLAG_SEND_WRITE_ZEROES_BIT] = "zeroes",
211 [NBD_FLAG_SEND_DF_BIT] = "df",
212 [NBD_FLAG_CAN_MULTI_CONN_BIT] = "multi",
213 [NBD_FLAG_SEND_RESIZE_BIT] = "resize",
214 [NBD_FLAG_SEND_CACHE_BIT] = "cache",
0a479545 215 [NBD_FLAG_SEND_FAST_ZERO_BIT] = "fast-zero",
c4e2aff8
HR
216 };
217
68b96f15
EB
218 printf(" size: %" PRIu64 "\n", list[i].size);
219 printf(" flags: 0x%x (", list[i].flags);
c4e2aff8
HR
220 for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
221 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
222 printf(" %s", flag_names[bit]);
223 }
68b96f15
EB
224 }
225 printf(" )\n");
226 }
227 if (list[i].min_block) {
228 printf(" min block: %u\n", list[i].min_block);
229 printf(" opt block: %u\n", list[i].opt_block);
230 printf(" max block: %u\n", list[i].max_block);
231 }
232 if (list[i].n_contexts) {
233 printf(" available meta contexts: %d\n", list[i].n_contexts);
234 for (j = 0; j < list[i].n_contexts; j++) {
235 printf(" %s\n", list[i].contexts[j]);
236 }
237 }
238 }
239 nbd_free_export_list(list, rc);
240
241 ret = EXIT_SUCCESS;
242 out:
243 object_unref(OBJECT(sioc));
244 return ret;
245}
246
247
3c1fa35d 248#if HAVE_NBD_DEVICE
a517e88b 249static void *show_parts(void *arg)
cd831bd7 250{
a6ac2313 251 char *device = arg;
a517e88b
PB
252 int nbd;
253
254 /* linux just needs an open() to trigger
255 * the partition table update
256 * but remember to load the module with max_part != 0 :
257 * modprobe nbd max_part=63
258 */
259 nbd = open(device, O_RDWR);
fc19f8a0 260 if (nbd >= 0) {
a517e88b
PB
261 close(nbd);
262 }
263 return NULL;
264}
cd831bd7 265
a517e88b
PB
266static void *nbd_client_thread(void *arg)
267{
a6ac2313 268 char *device = arg;
6dc1667d 269 NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
d0d6ff58 270 QIOChannelSocket *sioc;
af74b550
AC
271 int fd = -1;
272 int ret = EXIT_FAILURE;
a517e88b 273 pthread_t show_parts_thread;
1ce52846 274 Error *local_error = NULL;
a517e88b 275
d0d6ff58
DB
276 sioc = qio_channel_socket_new();
277 if (qio_channel_socket_connect_sync(sioc,
278 saddr,
279 &local_error) < 0) {
48bec07e 280 error_report_err(local_error);
dc10e8b3
SH
281 goto out;
282 }
a517e88b 283
af74b550
AC
284 if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
285 NULL, NULL, NULL, &info, &local_error) < 0) {
1ce52846 286 if (local_error) {
78288671 287 error_report_err(local_error);
1ce52846 288 }
af74b550 289 goto out;
a517e88b
PB
290 }
291
a6ac2313 292 fd = open(device, O_RDWR);
fc19f8a0 293 if (fd < 0) {
a6ac2313 294 /* Linux-only, we can use %m in printf. */
b9884681 295 error_report("Failed to open %s: %m", device);
af74b550 296 goto out;
a6ac2313
PB
297 }
298
af74b550 299 if (nbd_init(fd, sioc, &info, &local_error) < 0) {
be41c100 300 error_report_err(local_error);
af74b550 301 goto out;
a517e88b
PB
302 }
303
304 /* update partition table */
a6ac2313 305 pthread_create(&show_parts_thread, NULL, show_parts, device);
a517e88b 306
c1f8fdc3
PB
307 if (verbose) {
308 fprintf(stderr, "NBD device %s is now connected to %s\n",
309 device, srcpath);
310 } else {
311 /* Close stderr so that the qemu-nbd process exits. */
312 dup2(STDOUT_FILENO, STDERR_FILENO);
313 }
a517e88b 314
af74b550
AC
315 if (nbd_client(fd) < 0) {
316 goto out;
cd831bd7 317 }
a517e88b 318
af74b550
AC
319 ret = EXIT_SUCCESS;
320
321 out:
322 if (fd >= 0) {
323 close(fd);
324 }
d0d6ff58 325 object_unref(OBJECT(sioc));
6dc1667d 326 g_free(info.name);
a517e88b 327 kill(getpid(), SIGTERM);
af74b550 328 return (void *) (intptr_t) ret;
cd831bd7 329}
3c1fa35d 330#endif /* HAVE_NBD_DEVICE */
cd831bd7 331
e4afbf4f 332static int nbd_can_accept(void)
a61c6782 333{
3dcf56e6 334 return state == RUNNING && (shared == 0 || nb_fds < shared);
a61c6782
PB
335}
336
d0d6ff58 337static void nbd_update_server_watch(void);
e4afbf4f 338
0c9390d9 339static void nbd_client_closed(NBDClient *client, bool negotiated)
a61c6782 340{
1743b515 341 nb_fds--;
0c9390d9 342 if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
7860a380
PB
343 state = TERMINATE;
344 }
d0d6ff58 345 nbd_update_server_watch();
7860a380 346 nbd_client_put(client);
a61c6782
PB
347}
348
e4849c1d
DB
349static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
350 gpointer opaque)
a61c6782 351{
7860a380 352 if (state >= TERMINATE) {
e4849c1d 353 return;
7860a380
PB
354 }
355
ee7d7aab 356 nb_fds++;
d0d6ff58 357 nbd_update_server_watch();
b25e12da 358 nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
a61c6782
PB
359}
360
d0d6ff58 361static void nbd_update_server_watch(void)
e4afbf4f
FZ
362{
363 if (nbd_can_accept()) {
e4849c1d 364 qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
e4afbf4f 365 } else {
e4849c1d 366 qio_net_listener_set_client_func(server, NULL, NULL, NULL);
e4afbf4f
FZ
367 }
368}
369
48bec07e 370
bd269ebc 371static SocketAddress *nbd_build_socket_address(const char *sockpath,
48bec07e
DB
372 const char *bindto,
373 const char *port)
374{
bd269ebc 375 SocketAddress *saddr;
48bec07e 376
bd269ebc 377 saddr = g_new0(SocketAddress, 1);
48bec07e 378 if (sockpath) {
bd269ebc
MA
379 saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
380 saddr->u.q_unix.path = g_strdup(sockpath);
48bec07e 381 } else {
0399293e 382 InetSocketAddress *inet;
bd269ebc
MA
383 saddr->type = SOCKET_ADDRESS_TYPE_INET;
384 inet = &saddr->u.inet;
0399293e 385 inet->host = g_strdup(bindto);
48bec07e 386 if (port) {
0399293e 387 inet->port = g_strdup(port);
48bec07e 388 } else {
0399293e 389 inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
48bec07e
DB
390 }
391 }
392
393 return saddr;
394}
395
396
77c9aaef
DB
397static QemuOptsList file_opts = {
398 .name = "file",
399 .implied_opt_name = "file",
400 .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
401 .desc = {
402 /* no elements => accept any params */
403 { /* end of list */ }
404 },
405};
406
68b96f15
EB
407static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
408 Error **errp)
145614a1
DB
409{
410 Object *obj;
411 QCryptoTLSCreds *creds;
412
413 obj = object_resolve_path_component(
414 object_get_objects_root(), id);
415 if (!obj) {
416 error_setg(errp, "No TLS credentials with id '%s'",
417 id);
418 return NULL;
419 }
420 creds = (QCryptoTLSCreds *)
421 object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
422 if (!creds) {
423 error_setg(errp, "Object with id '%s' is not TLS credentials",
424 id);
425 return NULL;
426 }
427
0279cd95
PMD
428 if (!qcrypto_tls_creds_check_endpoint(creds,
429 list
430 ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
431 : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
432 errp)) {
433 return NULL;
145614a1
DB
434 }
435 object_ref(obj);
436 return creds;
437}
438
a721f53b
RJ
439static void setup_address_and_port(const char **address, const char **port)
440{
441 if (*address == NULL) {
442 *address = "0.0.0.0";
443 }
444
445 if (*port == NULL) {
446 *port = stringify(NBD_DEFAULT_PORT);
447 }
448}
449
a721f53b
RJ
450/*
451 * Check socket parameters compatibility when socket activation is used.
452 */
453static const char *socket_activation_validate_opts(const char *device,
454 const char *sockpath,
455 const char *address,
68b96f15
EB
456 const char *port,
457 bool list)
a721f53b
RJ
458{
459 if (device != NULL) {
460 return "NBD device can't be set when using socket activation";
461 }
462
463 if (sockpath != NULL) {
464 return "Unix socket can't be set when using socket activation";
465 }
466
467 if (address != NULL) {
468 return "The interface can't be set when using socket activation";
469 }
470
471 if (port != NULL) {
472 return "TCP port number can't be set when using socket activation";
473 }
474
68b96f15
EB
475 if (list) {
476 return "List mode is incompatible with socket activation";
477 }
478
a721f53b
RJ
479 return NULL;
480}
145614a1 481
b3b5299d
KW
482static void qemu_nbd_shutdown(void)
483{
484 job_cancel_sync_all();
1895b977 485 blk_exp_close_all();
b3b5299d
KW
486 bdrv_close_all();
487}
488
7a5ca864
FB
489int main(int argc, char **argv)
490{
26f54e9a 491 BlockBackend *blk;
7a5ca864 492 BlockDriverState *bs;
9d26dfcb 493 uint64_t dev_offset = 0;
dbb38caa 494 bool readonly = false;
cd831bd7 495 bool disconnect = false;
a721f53b 496 const char *bindto = NULL;
48bec07e
DB
497 const char *port = NULL;
498 char *sockpath = NULL;
a6ac2313 499 char *device = NULL;
8c116b0e
WX
500 QemuOpts *sn_opts = NULL;
501 const char *sn_id_or_name = NULL;
dbc7b014 502 const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
7a5ca864 503 struct option lopt[] = {
aa6e546c
DB
504 { "help", no_argument, NULL, 'h' },
505 { "version", no_argument, NULL, 'V' },
506 { "bind", required_argument, NULL, 'b' },
507 { "port", required_argument, NULL, 'p' },
508 { "socket", required_argument, NULL, 'k' },
509 { "offset", required_argument, NULL, 'o' },
510 { "read-only", no_argument, NULL, 'r' },
dbc7b014 511 { "allocation-depth", no_argument, NULL, 'A' },
636192c4 512 { "bitmap", required_argument, NULL, 'B' },
aa6e546c
DB
513 { "connect", required_argument, NULL, 'c' },
514 { "disconnect", no_argument, NULL, 'd' },
68b96f15 515 { "list", no_argument, NULL, 'L' },
aa6e546c
DB
516 { "snapshot", no_argument, NULL, 's' },
517 { "load-snapshot", required_argument, NULL, 'l' },
518 { "nocache", no_argument, NULL, 'n' },
519 { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
520 { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
521 { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
522 { "detect-zeroes", required_argument, NULL,
523 QEMU_NBD_OPT_DETECT_ZEROES },
524 { "shared", required_argument, NULL, 'e' },
525 { "format", required_argument, NULL, 'f' },
526 { "persistent", no_argument, NULL, 't' },
527 { "verbose", no_argument, NULL, 'v' },
528 { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
529 { "export-name", required_argument, NULL, 'x' },
b1a75b33 530 { "description", required_argument, NULL, 'D' },
aa6e546c 531 { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
b25e12da 532 { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
aa6e546c 533 { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
39ca463e 534 { "trace", required_argument, NULL, 'T' },
ffb31e1d 535 { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
637bc5a5 536 { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
660f11be 537 { NULL, 0, NULL, 0 }
7a5ca864
FB
538 };
539 int ch;
540 int opt_ind = 0;
f5edb014 541 int flags = BDRV_O_RDWR;
4fbec260 542 int ret = 0;
39a5235c 543 bool seen_cache = false;
ded9d2d5 544 bool seen_discard = false;
39a5235c 545 bool seen_aio = false;
a517e88b 546 pthread_t client_thread;
e6b63677 547 const char *fmt = NULL;
34b5d2c6 548 Error *local_err = NULL;
b3838a40 549 BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
4fbec260 550 QDict *options = NULL;
68b96f15 551 const char *export_name = NULL; /* defaults to "" later for server mode */
b1a75b33 552 const char *export_description = NULL;
cbad81ce 553 strList *bitmaps = NULL;
dbc7b014 554 bool alloc_depth = false;
145614a1 555 const char *tlscredsid = NULL;
77c9aaef 556 bool imageOpts = false;
09615257 557 bool writethrough = false; /* Client will flush as needed. */
ffb31e1d 558 bool fork_process = false;
68b96f15 559 bool list = false;
ffb31e1d 560 int old_stderr = -1;
a721f53b 561 unsigned socket_activation;
637bc5a5 562 const char *pid_file_name = NULL;
00917172 563 BlockExportOptions *export_opts;
7a5ca864 564
029a88c9 565#ifdef CONFIG_POSIX
cbc20bfb
SH
566 os_setup_early_signal_handling();
567 os_setup_signal_handling();
041e32b8
HR
568#endif
569
98c5d2e7 570 socket_init();
f5852efa 571 error_init(argv[0]);
fe4db84d 572 module_call_init(MODULE_INIT_TRACE);
e8f2d272 573 qcrypto_init(&error_fatal);
c2297088 574
0ab3b337 575 module_call_init(MODULE_INIT_QOM);
39ca463e 576 qemu_add_opts(&qemu_trace_opts);
10f5bff6 577 qemu_init_exec_dir(argv[0]);
bb345110 578
7a5ca864
FB
579 while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
580 switch (ch) {
581 case 's':
2f726488
TS
582 flags |= BDRV_O_SNAPSHOT;
583 break;
584 case 'n':
39a5235c
PB
585 optarg = (char *) "none";
586 /* fallthrough */
587 case QEMU_NBD_OPT_CACHE:
588 if (seen_cache) {
85b01e09
MA
589 error_report("-n and --cache can only be specified once");
590 exit(EXIT_FAILURE);
39a5235c
PB
591 }
592 seen_cache = true;
6effd5bf 593 if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
85b01e09
MA
594 error_report("Invalid cache mode `%s'", optarg);
595 exit(EXIT_FAILURE);
39a5235c 596 }
7a5ca864 597 break;
39a5235c
PB
598 case QEMU_NBD_OPT_AIO:
599 if (seen_aio) {
85b01e09
MA
600 error_report("--aio can only be specified once");
601 exit(EXIT_FAILURE);
39a5235c
PB
602 }
603 seen_aio = true;
7680274d
AM
604 if (bdrv_parse_aio(optarg, &flags) < 0) {
605 error_report("Invalid aio mode '%s'", optarg);
606 exit(EXIT_FAILURE);
39a5235c
PB
607 }
608 break;
ded9d2d5
PB
609 case QEMU_NBD_OPT_DISCARD:
610 if (seen_discard) {
85b01e09
MA
611 error_report("--discard can only be specified once");
612 exit(EXIT_FAILURE);
ded9d2d5
PB
613 }
614 seen_discard = true;
615 if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
85b01e09
MA
616 error_report("Invalid discard mode `%s'", optarg);
617 exit(EXIT_FAILURE);
ded9d2d5
PB
618 }
619 break;
b3838a40
PL
620 case QEMU_NBD_OPT_DETECT_ZEROES:
621 detect_zeroes =
f7abe0ec 622 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
b3838a40 623 optarg,
b3838a40
PL
624 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
625 &local_err);
626 if (local_err) {
c29b77f9
MA
627 error_reportf_err(local_err,
628 "Failed to parse detect_zeroes mode: ");
85b01e09 629 exit(EXIT_FAILURE);
b3838a40
PL
630 }
631 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
632 !(flags & BDRV_O_UNMAP)) {
85b01e09
MA
633 error_report("setting detect-zeroes to unmap is not allowed "
634 "without setting discard operation to unmap");
635 exit(EXIT_FAILURE);
b3838a40
PL
636 }
637 break;
7a5ca864
FB
638 case 'b':
639 bindto = optarg;
640 break;
641 case 'p':
48bec07e 642 port = optarg;
7a5ca864
FB
643 break;
644 case 'o':
43b51011
EB
645 if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
646 error_report("Invalid offset '%s'", optarg);
85b01e09 647 exit(EXIT_FAILURE);
7a5ca864 648 }
7a5ca864 649 break;
8c116b0e
WX
650 case 'l':
651 if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
70b94331
MA
652 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
653 optarg, false);
8c116b0e 654 if (!sn_opts) {
85b01e09
MA
655 error_report("Failed in parsing snapshot param `%s'",
656 optarg);
657 exit(EXIT_FAILURE);
8c116b0e
WX
658 }
659 } else {
660 sn_id_or_name = optarg;
661 }
662 /* fall through */
7a5ca864 663 case 'r':
dbb38caa 664 readonly = true;
07108b29 665 flags &= ~BDRV_O_RDWR;
7a5ca864 666 break;
dbc7b014
EB
667 case 'A':
668 alloc_depth = true;
669 break;
636192c4 670 case 'B':
cbad81ce 671 QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
636192c4 672 break;
cd831bd7 673 case 'k':
b32f6c28 674 sockpath = optarg;
713cc671 675 if (sockpath[0] != '/') {
9af9e0fe 676 error_report("socket path must be absolute");
85b01e09 677 exit(EXIT_FAILURE);
713cc671 678 }
cd831bd7
TS
679 break;
680 case 'd':
681 disconnect = true;
682 break;
683 case 'c':
684 device = optarg;
685 break;
3b05a8e9 686 case 'e':
43b51011 687 if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
3dcf56e6 688 shared < 0) {
85b01e09
MA
689 error_report("Invalid shared device number '%s'", optarg);
690 exit(EXIT_FAILURE);
3b05a8e9 691 }
3b05a8e9 692 break;
e6b63677
DB
693 case 'f':
694 fmt = optarg;
695 break;
713cc671
PL
696 case 't':
697 persistent = 1;
698 break;
3d4b2f9c
DB
699 case 'x':
700 export_name = optarg;
93676c88
EB
701 if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
702 error_report("export name '%s' too long", export_name);
703 exit(EXIT_FAILURE);
704 }
3d4b2f9c 705 break;
b1a75b33
EB
706 case 'D':
707 export_description = optarg;
93676c88
EB
708 if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
709 error_report("export description '%s' too long",
710 export_description);
711 exit(EXIT_FAILURE);
712 }
b1a75b33 713 break;
7a5ca864
FB
714 case 'v':
715 verbose = 1;
716 break;
717 case 'V':
718 version(argv[0]);
719 exit(0);
720 break;
721 case 'h':
722 usage(argv[0]);
723 exit(0);
724 break;
725 case '?':
85b01e09
MA
726 error_report("Try `%s --help' for more information.", argv[0]);
727 exit(EXIT_FAILURE);
fa40e43c
KW
728 case QEMU_NBD_OPT_OBJECT:
729 user_creatable_process_cmdline(optarg);
730 break;
145614a1
DB
731 case QEMU_NBD_OPT_TLSCREDS:
732 tlscredsid = optarg;
733 break;
77c9aaef
DB
734 case QEMU_NBD_OPT_IMAGE_OPTS:
735 imageOpts = true;
736 break;
39ca463e 737 case 'T':
92eecfff 738 trace_opt_parse(optarg);
39ca463e 739 break;
b25e12da
DB
740 case QEMU_NBD_OPT_TLSAUTHZ:
741 tlsauthz = optarg;
742 break;
ffb31e1d
HR
743 case QEMU_NBD_OPT_FORK:
744 fork_process = true;
745 break;
68b96f15
EB
746 case 'L':
747 list = true;
748 break;
637bc5a5
HR
749 case QEMU_NBD_OPT_PID_FILE:
750 pid_file_name = optarg;
751 break;
7a5ca864
FB
752 }
753 }
754
68b96f15
EB
755 if (list) {
756 if (argc != optind) {
757 error_report("List mode is incompatible with a file name");
758 exit(EXIT_FAILURE);
759 }
0bc16997 760 if (export_name || export_description || dev_offset ||
cbad81ce 761 device || disconnect || fmt || sn_id_or_name || bitmaps ||
dbc7b014 762 alloc_depth || seen_aio || seen_discard || seen_cache) {
68b96f15
EB
763 error_report("List mode is incompatible with per-device settings");
764 exit(EXIT_FAILURE);
765 }
766 if (fork_process) {
767 error_report("List mode is incompatible with forking");
768 exit(EXIT_FAILURE);
769 }
770 } else if ((argc - optind) != 1) {
433672b0
MA
771 error_report("Invalid number of arguments");
772 error_printf("Try `%s --help' for more information.\n", argv[0]);
85b01e09 773 exit(EXIT_FAILURE);
68b96f15
EB
774 } else if (!export_name) {
775 export_name = "";
7a5ca864
FB
776 }
777
39ca463e
DL
778 if (!trace_init_backends()) {
779 exit(1);
780 }
92eecfff 781 trace_init_file();
39ca463e
DL
782 qemu_set_log(LOG_TRACE);
783
a721f53b
RJ
784 socket_activation = check_socket_activation();
785 if (socket_activation == 0) {
786 setup_address_and_port(&bindto, &port);
787 } else {
788 /* Using socket activation - check user didn't use -p etc. */
789 const char *err_msg = socket_activation_validate_opts(device, sockpath,
68b96f15
EB
790 bindto, port,
791 list);
a721f53b
RJ
792 if (err_msg != NULL) {
793 error_report("%s", err_msg);
794 exit(EXIT_FAILURE);
795 }
53fabd4b
PB
796
797 /* qemu-nbd can only listen on a single socket. */
798 if (socket_activation > 1) {
799 error_report("qemu-nbd does not support socket activation with %s > 1",
800 "LISTEN_FDS");
801 exit(EXIT_FAILURE);
802 }
a721f53b
RJ
803 }
804
145614a1
DB
805 if (tlscredsid) {
806 if (sockpath) {
807 error_report("TLS is only supported with IPv4/IPv6");
808 exit(EXIT_FAILURE);
809 }
810 if (device) {
811 error_report("TLS is not supported with a host device");
812 exit(EXIT_FAILURE);
813 }
b25e12da
DB
814 if (tlsauthz && list) {
815 error_report("TLS authorization is incompatible with export list");
816 exit(EXIT_FAILURE);
817 }
68b96f15 818 tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
145614a1 819 if (local_err) {
5217f188 820 error_reportf_err(local_err, "Failed to get TLS creds: ");
145614a1
DB
821 exit(EXIT_FAILURE);
822 }
b25e12da
DB
823 } else {
824 if (tlsauthz) {
825 error_report("--tls-authz is not permitted without --tls-creds");
826 exit(EXIT_FAILURE);
827 }
145614a1
DB
828 }
829
68b96f15
EB
830 if (list) {
831 saddr = nbd_build_socket_address(sockpath, bindto, port);
832 return qemu_nbd_client_list(saddr, tlscreds, bindto);
833 }
834
3c1fa35d
EB
835#if !HAVE_NBD_DEVICE
836 if (disconnect || device) {
837 error_report("Kernel /dev/nbdN support not available");
838 exit(EXIT_FAILURE);
839 }
840#else /* HAVE_NBD_DEVICE */
cd831bd7 841 if (disconnect) {
d0d6ff58
DB
842 int nbdfd = open(argv[optind], O_RDWR);
843 if (nbdfd < 0) {
85b01e09
MA
844 error_report("Cannot open %s: %s", argv[optind],
845 strerror(errno));
846 exit(EXIT_FAILURE);
fc19f8a0 847 }
d0d6ff58 848 nbd_disconnect(nbdfd);
cd831bd7 849
d0d6ff58 850 close(nbdfd);
cd831bd7
TS
851
852 printf("%s disconnected\n", argv[optind]);
853
713cc671 854 return 0;
cd831bd7 855 }
3c1fa35d 856#endif
cd831bd7 857
ffb31e1d 858 if ((device && !verbose) || fork_process) {
eb705985 859#ifndef WIN32
c1f8fdc3
PB
860 int stderr_fd[2];
861 pid_t pid;
862 int ret;
863
fc19f8a0 864 if (qemu_pipe(stderr_fd) < 0) {
85b01e09
MA
865 error_report("Error setting up communication pipe: %s",
866 strerror(errno));
867 exit(EXIT_FAILURE);
c1f8fdc3
PB
868 }
869
870 /* Now daemonize, but keep a communication channel open to
871 * print errors and exit with the proper status code.
872 */
873 pid = fork();
70d4739e 874 if (pid < 0) {
85b01e09
MA
875 error_report("Failed to fork: %s", strerror(errno));
876 exit(EXIT_FAILURE);
70d4739e 877 } else if (pid == 0) {
c1f8fdc3 878 close(stderr_fd[0]);
e6df58a5 879
0eaf453e
RP
880 /* Remember parent's stderr if we will be restoring it. */
881 if (fork_process) {
882 old_stderr = dup(STDERR_FILENO);
883 }
884
9faf31b6 885 ret = qemu_daemon(1, 0);
c1f8fdc3
PB
886
887 /* Temporarily redirect stderr to the parent's pipe... */
888 dup2(stderr_fd[1], STDERR_FILENO);
fc19f8a0 889 if (ret < 0) {
85b01e09
MA
890 error_report("Failed to daemonize: %s", strerror(errno));
891 exit(EXIT_FAILURE);
c1f8fdc3
PB
892 }
893
894 /* ... close the descriptor we inherited and go on. */
895 close(stderr_fd[1]);
896 } else {
897 bool errors = false;
898 char *buf;
899
900 /* In the parent. Print error messages from the child until
901 * it closes the pipe.
902 */
903 close(stderr_fd[1]);
904 buf = g_malloc(1024);
905 while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
906 errors = true;
907 ret = qemu_write_full(STDERR_FILENO, buf, ret);
fc19f8a0 908 if (ret < 0) {
c1f8fdc3
PB
909 exit(EXIT_FAILURE);
910 }
911 }
fc19f8a0 912 if (ret < 0) {
85b01e09
MA
913 error_report("Cannot read from daemon: %s",
914 strerror(errno));
915 exit(EXIT_FAILURE);
c1f8fdc3
PB
916 }
917
918 /* Usually the daemon should not print any message.
919 * Exit with zero status in that case.
920 */
921 exit(errors);
922 }
eb705985
DB
923#else /* WIN32 */
924 error_report("Unable to fork into background on Windows hosts");
925 exit(EXIT_FAILURE);
926#endif /* WIN32 */
c1f8fdc3
PB
927 }
928
a6ac2313
PB
929 if (device != NULL && sockpath == NULL) {
930 sockpath = g_malloc(128);
931 snprintf(sockpath, 128, SOCKET_PATH, basename(device));
cd831bd7
TS
932 }
933
e4849c1d 934 server = qio_net_listener_new();
a721f53b 935 if (socket_activation == 0) {
582d4210
EB
936 int backlog;
937
3dcf56e6 938 if (persistent || shared == 0) {
582d4210
EB
939 backlog = SOMAXCONN;
940 } else {
941 backlog = MIN(shared, SOMAXCONN);
942 }
a721f53b 943 saddr = nbd_build_socket_address(sockpath, bindto, port);
582d4210
EB
944 if (qio_net_listener_open_sync(server, saddr, backlog,
945 &local_err) < 0) {
e4849c1d 946 object_unref(OBJECT(server));
a721f53b 947 error_report_err(local_err);
e4849c1d 948 exit(EXIT_FAILURE);
a721f53b
RJ
949 }
950 } else {
e4849c1d 951 size_t i;
a721f53b 952 /* See comment in check_socket_activation above. */
e4849c1d
DB
953 for (i = 0; i < socket_activation; i++) {
954 QIOChannelSocket *sioc;
955 sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
956 &local_err);
957 if (sioc == NULL) {
958 object_unref(OBJECT(server));
5217f188
MA
959 error_reportf_err(local_err,
960 "Failed to use socket activation: ");
e4849c1d
DB
961 exit(EXIT_FAILURE);
962 }
963 qio_net_listener_add(server, sioc);
964 object_unref(OBJECT(sioc));
a721f53b
RJ
965 }
966 }
48bec07e 967
f9734d5d 968 qemu_init_main_loop(&error_fatal);
802ddc37 969 bdrv_init();
b3b5299d 970 atexit(qemu_nbd_shutdown);
802ddc37 971
77c9aaef
DB
972 srcpath = argv[optind];
973 if (imageOpts) {
974 QemuOpts *opts;
975 if (fmt) {
976 error_report("--image-opts and -f are mutually exclusive");
977 exit(EXIT_FAILURE);
978 }
979 opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
980 if (!opts) {
981 qemu_opts_reset(&file_opts);
982 exit(EXIT_FAILURE);
983 }
984 options = qemu_opts_to_qdict(opts, NULL);
985 qemu_opts_reset(&file_opts);
efaa7c4e 986 blk = blk_new_open(NULL, NULL, options, flags, &local_err);
77c9aaef
DB
987 } else {
988 if (fmt) {
989 options = qdict_new();
46f5ac20 990 qdict_put_str(options, "driver", fmt);
77c9aaef 991 }
efaa7c4e 992 blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
e6b63677
DB
993 }
994
4fbec260 995 if (!blk) {
c29b77f9
MA
996 error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
997 argv[optind]);
85b01e09 998 exit(EXIT_FAILURE);
802ddc37 999 }
4fbec260 1000 bs = blk_bs(blk);
802ddc37 1001
b57e4de0
KW
1002 if (dev_offset) {
1003 QDict *raw_opts = qdict_new();
1004 qdict_put_str(raw_opts, "driver", "raw");
1005 qdict_put_str(raw_opts, "file", bs->node_name);
1006 qdict_put_int(raw_opts, "offset", dev_offset);
1007 bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1008 blk_remove_bs(blk);
1009 blk_insert_bs(blk, bs, &error_fatal);
1010 bdrv_unref(bs);
1011 }
1012
6effd5bf
KW
1013 blk_set_enable_write_cache(blk, !writethrough);
1014
8c116b0e
WX
1015 if (sn_opts) {
1016 ret = bdrv_snapshot_load_tmp(bs,
1017 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1018 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1019 &local_err);
1020 } else if (sn_id_or_name) {
1021 ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1022 &local_err);
1023 }
1024 if (ret < 0) {
c29b77f9 1025 error_reportf_err(local_err, "Failed to load snapshot: ");
85b01e09 1026 exit(EXIT_FAILURE);
8c116b0e
WX
1027 }
1028
b3838a40 1029 bs->detect_zeroes = detect_zeroes;
e424b655 1030
00917172
KW
1031 nbd_server_is_qemu_nbd(true);
1032
1033 export_opts = g_new(BlockExportOptions, 1);
1034 *export_opts = (BlockExportOptions) {
1035 .type = BLOCK_EXPORT_TYPE_NBD,
d53be9ce 1036 .id = g_strdup("qemu-nbd-export"),
b6076afc 1037 .node_name = g_strdup(bdrv_get_node_name(bs)),
00917172
KW
1038 .has_writethrough = true,
1039 .writethrough = writethrough,
30dbc81d
KW
1040 .has_writable = true,
1041 .writable = !readonly,
00917172 1042 .u.nbd = {
cbad81ce
EB
1043 .has_name = true,
1044 .name = g_strdup(export_name),
1045 .has_description = !!export_description,
1046 .description = g_strdup(export_description),
1047 .has_bitmaps = !!bitmaps,
1048 .bitmaps = bitmaps,
dbc7b014
EB
1049 .has_allocation_depth = alloc_depth,
1050 .allocation_depth = alloc_depth,
00917172
KW
1051 },
1052 };
1053 blk_exp_add(export_opts, &error_fatal);
1054 qapi_free_BlockExportOptions(export_opts);
3b05a8e9 1055
f1ef5555 1056 if (device) {
3c1fa35d 1057#if HAVE_NBD_DEVICE
f1ef5555
PB
1058 int ret;
1059
a6ac2313 1060 ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
f1ef5555 1061 if (ret != 0) {
85b01e09
MA
1062 error_report("Failed to create client thread: %s", strerror(ret));
1063 exit(EXIT_FAILURE);
f1ef5555 1064 }
3c1fa35d 1065#endif
f1ef5555
PB
1066 } else {
1067 /* Shut up GCC warnings. */
1068 memset(&client_thread, 0, sizeof(client_thread));
1069 }
1070
d0d6ff58 1071 nbd_update_server_watch();
7a5ca864 1072
637bc5a5
HR
1073 if (pid_file_name) {
1074 qemu_write_pidfile(pid_file_name, &error_fatal);
1075 }
1076
9faf31b6
MT
1077 /* now when the initialization is (almost) complete, chdir("/")
1078 * to free any busy filesystems */
1079 if (chdir("/") < 0) {
85b01e09
MA
1080 error_report("Could not chdir to root directory: %s",
1081 strerror(errno));
1082 exit(EXIT_FAILURE);
9faf31b6
MT
1083 }
1084
ffb31e1d
HR
1085 if (fork_process) {
1086 dup2(old_stderr, STDERR_FILENO);
1087 close(old_stderr);
1088 }
1089
7860a380 1090 state = RUNNING;
3b05a8e9 1091 do {
a61c6782 1092 main_loop_wait(false);
7860a380 1093 if (state == TERMINATE) {
bc4ee65b 1094 blk_exp_close_all();
d794f7f3 1095 state = TERMINATED;
7860a380
PB
1096 }
1097 } while (state != TERMINATED);
7a5ca864 1098
26f54e9a 1099 blk_unref(blk);
b32f6c28
PB
1100 if (sockpath) {
1101 unlink(sockpath);
1102 }
7a5ca864 1103
fbf28a43 1104 qemu_opts_del(sn_opts);
8c116b0e 1105
a517e88b
PB
1106 if (device) {
1107 void *ret;
1108 pthread_join(client_thread, &ret);
1109 exit(ret != NULL);
1110 } else {
1111 exit(EXIT_SUCCESS);
1112 }
7a5ca864 1113}