xref: /qemu/qemu-nbd.c (revision dbc9e94a)
1 /*
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
16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
17  */
18 
19 #include "qemu/osdep.h"
20 #include <getopt.h>
21 #include <libgen.h>
22 #include <pthread.h>
23 
24 #include "qemu-common.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "sysemu/block-backend.h"
28 #include "block/block_int.h"
29 #include "block/nbd.h"
30 #include "qemu/main-loop.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "qemu/error-report.h"
34 #include "qemu/config-file.h"
35 #include "qemu/bswap.h"
36 #include "qemu/log.h"
37 #include "qemu/systemd.h"
38 #include "block/snapshot.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qom/object_interfaces.h"
42 #include "io/channel-socket.h"
43 #include "io/net-listener.h"
44 #include "crypto/init.h"
45 #include "trace/control.h"
46 #include "qemu-version.h"
47 
48 #ifdef __linux__
49 #define HAVE_NBD_DEVICE 1
50 #else
51 #define HAVE_NBD_DEVICE 0
52 #endif
53 
54 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
55 #define QEMU_NBD_OPT_CACHE         256
56 #define QEMU_NBD_OPT_AIO           257
57 #define QEMU_NBD_OPT_DISCARD       258
58 #define QEMU_NBD_OPT_DETECT_ZEROES 259
59 #define QEMU_NBD_OPT_OBJECT        260
60 #define QEMU_NBD_OPT_TLSCREDS      261
61 #define QEMU_NBD_OPT_IMAGE_OPTS    262
62 #define QEMU_NBD_OPT_FORK          263
63 #define QEMU_NBD_OPT_TLSAUTHZ      264
64 #define QEMU_NBD_OPT_PID_FILE      265
65 
66 #define MBR_SIZE 512
67 
68 static int verbose;
69 static char *srcpath;
70 static SocketAddress *saddr;
71 static int persistent = 0;
72 static enum { RUNNING, TERMINATE, TERMINATED } state;
73 static int shared = 1;
74 static int nb_fds;
75 static QIONetListener *server;
76 static QCryptoTLSCreds *tlscreds;
77 static const char *tlsauthz;
78 
79 static void usage(const char *name)
80 {
81     (printf) (
82 "Usage: %s [OPTIONS] FILE\n"
83 "  or:  %s -L [OPTIONS]\n"
84 "QEMU Disk Network Block Device Utility\n"
85 "\n"
86 "  -h, --help                display this help and exit\n"
87 "  -V, --version             output version information and exit\n"
88 "\n"
89 "Connection properties:\n"
90 "  -p, --port=PORT           port to listen on (default `%d')\n"
91 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
92 "  -k, --socket=PATH         path to the unix socket\n"
93 "                            (default '"SOCKET_PATH"')\n"
94 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
95 "  -t, --persistent          don't exit on the last connection\n"
96 "  -v, --verbose             display extra debugging information\n"
97 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
98 "  -D, --description=TEXT    export a human-readable description\n"
99 "\n"
100 "Exposing part of the image:\n"
101 "  -o, --offset=OFFSET       offset into the image\n"
102 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
103 "\n"
104 "General purpose options:\n"
105 "  -L, --list                list exports available from another NBD server\n"
106 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
107 "                            passwords and/or encryption keys\n"
108 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
109 "  --tls-authz=ID            use id of an earlier --object to provide\n"
110 "                            authorization\n"
111 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
112 "                            specify tracing options\n"
113 "  --fork                    fork off the server process and exit the parent\n"
114 "                            once the server is running\n"
115 "  --pid-file=PATH           store the server's process ID in the given file\n"
116 #if HAVE_NBD_DEVICE
117 "\n"
118 "Kernel NBD client support:\n"
119 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
120 "  -d, --disconnect          disconnect the specified device\n"
121 #endif
122 "\n"
123 "Block device options:\n"
124 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
125 "  -r, --read-only           export read-only\n"
126 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
127 "                            file with backing_file=FILE, redirect the write to\n"
128 "                            the temporary one\n"
129 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
130 "                            load an internal snapshot inside FILE and export it\n"
131 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
132 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
133 "                            '[ID_OR_NAME]'\n"
134 "  -n, --nocache             disable host cache\n"
135 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
136 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
137 "      --discard=MODE        set discard mode (ignore, unmap)\n"
138 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
139 "      --image-opts          treat FILE as a full set of image options\n"
140 "\n"
141 QEMU_HELP_BOTTOM "\n"
142     , name, name, NBD_DEFAULT_PORT, "DEVICE");
143 }
144 
145 static void version(const char *name)
146 {
147     printf(
148 "%s " QEMU_FULL_VERSION "\n"
149 "Written by Anthony Liguori.\n"
150 "\n"
151 QEMU_COPYRIGHT "\n"
152 "This is free software; see the source for copying conditions.  There is NO\n"
153 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
154     , name);
155 }
156 
157 #if HAVE_NBD_DEVICE
158 static void termsig_handler(int signum)
159 {
160     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
161     qemu_notify_event();
162 }
163 #endif /* HAVE_NBD_DEVICE */
164 
165 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
166                                 const char *hostname)
167 {
168     int ret = EXIT_FAILURE;
169     int rc;
170     Error *err = NULL;
171     QIOChannelSocket *sioc;
172     NBDExportInfo *list;
173     int i, j;
174 
175     sioc = qio_channel_socket_new();
176     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
177         error_report_err(err);
178         return EXIT_FAILURE;
179     }
180     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
181                                  &err);
182     if (rc < 0) {
183         if (err) {
184             error_report_err(err);
185         }
186         goto out;
187     }
188     printf("exports available: %d\n", rc);
189     for (i = 0; i < rc; i++) {
190         printf(" export: '%s'\n", list[i].name);
191         if (list[i].description && *list[i].description) {
192             printf("  description: %s\n", list[i].description);
193         }
194         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
195             static const char *const flag_names[] = {
196                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
197                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
198                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
199                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
200                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
201                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
202                 [NBD_FLAG_SEND_DF_BIT]              = "df",
203                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
204                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
205                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
206                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
207             };
208 
209             printf("  size:  %" PRIu64 "\n", list[i].size);
210             printf("  flags: 0x%x (", list[i].flags);
211             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
212                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
213                     printf(" %s", flag_names[bit]);
214                 }
215             }
216             printf(" )\n");
217         }
218         if (list[i].min_block) {
219             printf("  min block: %u\n", list[i].min_block);
220             printf("  opt block: %u\n", list[i].opt_block);
221             printf("  max block: %u\n", list[i].max_block);
222         }
223         if (list[i].n_contexts) {
224             printf("  available meta contexts: %d\n", list[i].n_contexts);
225             for (j = 0; j < list[i].n_contexts; j++) {
226                 printf("   %s\n", list[i].contexts[j]);
227             }
228         }
229     }
230     nbd_free_export_list(list, rc);
231 
232     ret = EXIT_SUCCESS;
233  out:
234     object_unref(OBJECT(sioc));
235     return ret;
236 }
237 
238 
239 #if HAVE_NBD_DEVICE
240 static void *show_parts(void *arg)
241 {
242     char *device = arg;
243     int nbd;
244 
245     /* linux just needs an open() to trigger
246      * the partition table update
247      * but remember to load the module with max_part != 0 :
248      *     modprobe nbd max_part=63
249      */
250     nbd = open(device, O_RDWR);
251     if (nbd >= 0) {
252         close(nbd);
253     }
254     return NULL;
255 }
256 
257 static void *nbd_client_thread(void *arg)
258 {
259     char *device = arg;
260     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
261     QIOChannelSocket *sioc;
262     int fd;
263     int ret;
264     pthread_t show_parts_thread;
265     Error *local_error = NULL;
266 
267     sioc = qio_channel_socket_new();
268     if (qio_channel_socket_connect_sync(sioc,
269                                         saddr,
270                                         &local_error) < 0) {
271         error_report_err(local_error);
272         goto out;
273     }
274 
275     ret = nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
276                                 NULL, NULL, NULL, &info, &local_error);
277     if (ret < 0) {
278         if (local_error) {
279             error_report_err(local_error);
280         }
281         goto out_socket;
282     }
283 
284     fd = open(device, O_RDWR);
285     if (fd < 0) {
286         /* Linux-only, we can use %m in printf.  */
287         error_report("Failed to open %s: %m", device);
288         goto out_socket;
289     }
290 
291     ret = nbd_init(fd, sioc, &info, &local_error);
292     if (ret < 0) {
293         error_report_err(local_error);
294         goto out_fd;
295     }
296 
297     /* update partition table */
298     pthread_create(&show_parts_thread, NULL, show_parts, device);
299 
300     if (verbose) {
301         fprintf(stderr, "NBD device %s is now connected to %s\n",
302                 device, srcpath);
303     } else {
304         /* Close stderr so that the qemu-nbd process exits.  */
305         dup2(STDOUT_FILENO, STDERR_FILENO);
306     }
307 
308     ret = nbd_client(fd);
309     if (ret) {
310         goto out_fd;
311     }
312     close(fd);
313     object_unref(OBJECT(sioc));
314     g_free(info.name);
315     kill(getpid(), SIGTERM);
316     return (void *) EXIT_SUCCESS;
317 
318 out_fd:
319     close(fd);
320 out_socket:
321     object_unref(OBJECT(sioc));
322 out:
323     g_free(info.name);
324     kill(getpid(), SIGTERM);
325     return (void *) EXIT_FAILURE;
326 }
327 #endif /* HAVE_NBD_DEVICE */
328 
329 static int nbd_can_accept(void)
330 {
331     return state == RUNNING && nb_fds < shared;
332 }
333 
334 static void nbd_update_server_watch(void);
335 
336 static void nbd_client_closed(NBDClient *client, bool negotiated)
337 {
338     nb_fds--;
339     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
340         state = TERMINATE;
341     }
342     nbd_update_server_watch();
343     nbd_client_put(client);
344 }
345 
346 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
347                        gpointer opaque)
348 {
349     if (state >= TERMINATE) {
350         return;
351     }
352 
353     nb_fds++;
354     nbd_update_server_watch();
355     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
356 }
357 
358 static void nbd_update_server_watch(void)
359 {
360     if (nbd_can_accept()) {
361         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
362     } else {
363         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
364     }
365 }
366 
367 
368 static SocketAddress *nbd_build_socket_address(const char *sockpath,
369                                                const char *bindto,
370                                                const char *port)
371 {
372     SocketAddress *saddr;
373 
374     saddr = g_new0(SocketAddress, 1);
375     if (sockpath) {
376         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
377         saddr->u.q_unix.path = g_strdup(sockpath);
378     } else {
379         InetSocketAddress *inet;
380         saddr->type = SOCKET_ADDRESS_TYPE_INET;
381         inet = &saddr->u.inet;
382         inet->host = g_strdup(bindto);
383         if (port) {
384             inet->port = g_strdup(port);
385         } else  {
386             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
387         }
388     }
389 
390     return saddr;
391 }
392 
393 
394 static QemuOptsList file_opts = {
395     .name = "file",
396     .implied_opt_name = "file",
397     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
398     .desc = {
399         /* no elements => accept any params */
400         { /* end of list */ }
401     },
402 };
403 
404 static QemuOptsList qemu_object_opts = {
405     .name = "object",
406     .implied_opt_name = "qom-type",
407     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
408     .desc = {
409         { }
410     },
411 };
412 
413 static bool qemu_nbd_object_print_help(const char *type, QemuOpts *opts)
414 {
415     if (user_creatable_print_help(type, opts)) {
416         exit(0);
417     }
418     return true;
419 }
420 
421 
422 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
423                                           Error **errp)
424 {
425     Object *obj;
426     QCryptoTLSCreds *creds;
427 
428     obj = object_resolve_path_component(
429         object_get_objects_root(), id);
430     if (!obj) {
431         error_setg(errp, "No TLS credentials with id '%s'",
432                    id);
433         return NULL;
434     }
435     creds = (QCryptoTLSCreds *)
436         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
437     if (!creds) {
438         error_setg(errp, "Object with id '%s' is not TLS credentials",
439                    id);
440         return NULL;
441     }
442 
443     if (list) {
444         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
445             error_setg(errp,
446                        "Expecting TLS credentials with a client endpoint");
447             return NULL;
448         }
449     } else {
450         if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
451             error_setg(errp,
452                        "Expecting TLS credentials with a server endpoint");
453             return NULL;
454         }
455     }
456     object_ref(obj);
457     return creds;
458 }
459 
460 static void setup_address_and_port(const char **address, const char **port)
461 {
462     if (*address == NULL) {
463         *address = "0.0.0.0";
464     }
465 
466     if (*port == NULL) {
467         *port = stringify(NBD_DEFAULT_PORT);
468     }
469 }
470 
471 /*
472  * Check socket parameters compatibility when socket activation is used.
473  */
474 static const char *socket_activation_validate_opts(const char *device,
475                                                    const char *sockpath,
476                                                    const char *address,
477                                                    const char *port,
478                                                    bool list)
479 {
480     if (device != NULL) {
481         return "NBD device can't be set when using socket activation";
482     }
483 
484     if (sockpath != NULL) {
485         return "Unix socket can't be set when using socket activation";
486     }
487 
488     if (address != NULL) {
489         return "The interface can't be set when using socket activation";
490     }
491 
492     if (port != NULL) {
493         return "TCP port number can't be set when using socket activation";
494     }
495 
496     if (list) {
497         return "List mode is incompatible with socket activation";
498     }
499 
500     return NULL;
501 }
502 
503 static void qemu_nbd_shutdown(void)
504 {
505     job_cancel_sync_all();
506     bdrv_close_all();
507 }
508 
509 int main(int argc, char **argv)
510 {
511     BlockBackend *blk;
512     BlockDriverState *bs;
513     uint64_t dev_offset = 0;
514     bool readonly = false;
515     bool disconnect = false;
516     const char *bindto = NULL;
517     const char *port = NULL;
518     char *sockpath = NULL;
519     char *device = NULL;
520     QemuOpts *sn_opts = NULL;
521     const char *sn_id_or_name = NULL;
522     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:B:L";
523     struct option lopt[] = {
524         { "help", no_argument, NULL, 'h' },
525         { "version", no_argument, NULL, 'V' },
526         { "bind", required_argument, NULL, 'b' },
527         { "port", required_argument, NULL, 'p' },
528         { "socket", required_argument, NULL, 'k' },
529         { "offset", required_argument, NULL, 'o' },
530         { "read-only", no_argument, NULL, 'r' },
531         { "bitmap", required_argument, NULL, 'B' },
532         { "connect", required_argument, NULL, 'c' },
533         { "disconnect", no_argument, NULL, 'd' },
534         { "list", no_argument, NULL, 'L' },
535         { "snapshot", no_argument, NULL, 's' },
536         { "load-snapshot", required_argument, NULL, 'l' },
537         { "nocache", no_argument, NULL, 'n' },
538         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
539         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
540         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
541         { "detect-zeroes", required_argument, NULL,
542           QEMU_NBD_OPT_DETECT_ZEROES },
543         { "shared", required_argument, NULL, 'e' },
544         { "format", required_argument, NULL, 'f' },
545         { "persistent", no_argument, NULL, 't' },
546         { "verbose", no_argument, NULL, 'v' },
547         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
548         { "export-name", required_argument, NULL, 'x' },
549         { "description", required_argument, NULL, 'D' },
550         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
551         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
552         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
553         { "trace", required_argument, NULL, 'T' },
554         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
555         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
556         { NULL, 0, NULL, 0 }
557     };
558     int ch;
559     int opt_ind = 0;
560     int flags = BDRV_O_RDWR;
561     int ret = 0;
562     bool seen_cache = false;
563     bool seen_discard = false;
564     bool seen_aio = false;
565     pthread_t client_thread;
566     const char *fmt = NULL;
567     Error *local_err = NULL;
568     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
569     QDict *options = NULL;
570     const char *export_name = NULL; /* defaults to "" later for server mode */
571     const char *export_description = NULL;
572     const char *bitmap = NULL;
573     const char *tlscredsid = NULL;
574     bool imageOpts = false;
575     bool writethrough = true;
576     char *trace_file = NULL;
577     bool fork_process = false;
578     bool list = false;
579     int old_stderr = -1;
580     unsigned socket_activation;
581     const char *pid_file_name = NULL;
582     BlockExportOptions *export_opts;
583 
584 #if HAVE_NBD_DEVICE
585     /* The client thread uses SIGTERM to interrupt the server.  A signal
586      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
587      */
588     struct sigaction sa_sigterm;
589     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
590     sa_sigterm.sa_handler = termsig_handler;
591     sigaction(SIGTERM, &sa_sigterm, NULL);
592 #endif /* HAVE_NBD_DEVICE */
593 
594 #ifdef CONFIG_POSIX
595     signal(SIGPIPE, SIG_IGN);
596 #endif
597 
598     socket_init();
599     error_init(argv[0]);
600     module_call_init(MODULE_INIT_TRACE);
601     qcrypto_init(&error_fatal);
602 
603     module_call_init(MODULE_INIT_QOM);
604     qemu_add_opts(&qemu_object_opts);
605     qemu_add_opts(&qemu_trace_opts);
606     qemu_init_exec_dir(argv[0]);
607 
608     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
609         switch (ch) {
610         case 's':
611             flags |= BDRV_O_SNAPSHOT;
612             break;
613         case 'n':
614             optarg = (char *) "none";
615             /* fallthrough */
616         case QEMU_NBD_OPT_CACHE:
617             if (seen_cache) {
618                 error_report("-n and --cache can only be specified once");
619                 exit(EXIT_FAILURE);
620             }
621             seen_cache = true;
622             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
623                 error_report("Invalid cache mode `%s'", optarg);
624                 exit(EXIT_FAILURE);
625             }
626             break;
627         case QEMU_NBD_OPT_AIO:
628             if (seen_aio) {
629                 error_report("--aio can only be specified once");
630                 exit(EXIT_FAILURE);
631             }
632             seen_aio = true;
633             if (bdrv_parse_aio(optarg, &flags) < 0) {
634                 error_report("Invalid aio mode '%s'", optarg);
635                 exit(EXIT_FAILURE);
636             }
637             break;
638         case QEMU_NBD_OPT_DISCARD:
639             if (seen_discard) {
640                 error_report("--discard can only be specified once");
641                 exit(EXIT_FAILURE);
642             }
643             seen_discard = true;
644             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
645                 error_report("Invalid discard mode `%s'", optarg);
646                 exit(EXIT_FAILURE);
647             }
648             break;
649         case QEMU_NBD_OPT_DETECT_ZEROES:
650             detect_zeroes =
651                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
652                                 optarg,
653                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
654                                 &local_err);
655             if (local_err) {
656                 error_reportf_err(local_err,
657                                   "Failed to parse detect_zeroes mode: ");
658                 exit(EXIT_FAILURE);
659             }
660             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
661                 !(flags & BDRV_O_UNMAP)) {
662                 error_report("setting detect-zeroes to unmap is not allowed "
663                              "without setting discard operation to unmap");
664                 exit(EXIT_FAILURE);
665             }
666             break;
667         case 'b':
668             bindto = optarg;
669             break;
670         case 'p':
671             port = optarg;
672             break;
673         case 'o':
674             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
675                 error_report("Invalid offset '%s'", optarg);
676                 exit(EXIT_FAILURE);
677             }
678             break;
679         case 'l':
680             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
681                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
682                                                   optarg, false);
683                 if (!sn_opts) {
684                     error_report("Failed in parsing snapshot param `%s'",
685                                  optarg);
686                     exit(EXIT_FAILURE);
687                 }
688             } else {
689                 sn_id_or_name = optarg;
690             }
691             /* fall through */
692         case 'r':
693             readonly = true;
694             flags &= ~BDRV_O_RDWR;
695             break;
696         case 'B':
697             bitmap = optarg;
698             break;
699         case 'k':
700             sockpath = optarg;
701             if (sockpath[0] != '/') {
702                 error_report("socket path must be absolute");
703                 exit(EXIT_FAILURE);
704             }
705             break;
706         case 'd':
707             disconnect = true;
708             break;
709         case 'c':
710             device = optarg;
711             break;
712         case 'e':
713             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
714                 shared < 1) {
715                 error_report("Invalid shared device number '%s'", optarg);
716                 exit(EXIT_FAILURE);
717             }
718             break;
719         case 'f':
720             fmt = optarg;
721             break;
722         case 't':
723             persistent = 1;
724             break;
725         case 'x':
726             export_name = optarg;
727             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
728                 error_report("export name '%s' too long", export_name);
729                 exit(EXIT_FAILURE);
730             }
731             break;
732         case 'D':
733             export_description = optarg;
734             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
735                 error_report("export description '%s' too long",
736                              export_description);
737                 exit(EXIT_FAILURE);
738             }
739             break;
740         case 'v':
741             verbose = 1;
742             break;
743         case 'V':
744             version(argv[0]);
745             exit(0);
746             break;
747         case 'h':
748             usage(argv[0]);
749             exit(0);
750             break;
751         case '?':
752             error_report("Try `%s --help' for more information.", argv[0]);
753             exit(EXIT_FAILURE);
754         case QEMU_NBD_OPT_OBJECT: {
755             QemuOpts *opts;
756             opts = qemu_opts_parse_noisily(&qemu_object_opts,
757                                            optarg, true);
758             if (!opts) {
759                 exit(EXIT_FAILURE);
760             }
761         }   break;
762         case QEMU_NBD_OPT_TLSCREDS:
763             tlscredsid = optarg;
764             break;
765         case QEMU_NBD_OPT_IMAGE_OPTS:
766             imageOpts = true;
767             break;
768         case 'T':
769             g_free(trace_file);
770             trace_file = trace_opt_parse(optarg);
771             break;
772         case QEMU_NBD_OPT_TLSAUTHZ:
773             tlsauthz = optarg;
774             break;
775         case QEMU_NBD_OPT_FORK:
776             fork_process = true;
777             break;
778         case 'L':
779             list = true;
780             break;
781         case QEMU_NBD_OPT_PID_FILE:
782             pid_file_name = optarg;
783             break;
784         }
785     }
786 
787     if (list) {
788         if (argc != optind) {
789             error_report("List mode is incompatible with a file name");
790             exit(EXIT_FAILURE);
791         }
792         if (export_name || export_description || dev_offset ||
793             device || disconnect || fmt || sn_id_or_name || bitmap ||
794             seen_aio || seen_discard || seen_cache) {
795             error_report("List mode is incompatible with per-device settings");
796             exit(EXIT_FAILURE);
797         }
798         if (fork_process) {
799             error_report("List mode is incompatible with forking");
800             exit(EXIT_FAILURE);
801         }
802     } else if ((argc - optind) != 1) {
803         error_report("Invalid number of arguments");
804         error_printf("Try `%s --help' for more information.\n", argv[0]);
805         exit(EXIT_FAILURE);
806     } else if (!export_name) {
807         export_name = "";
808     }
809 
810     qemu_opts_foreach(&qemu_object_opts,
811                       user_creatable_add_opts_foreach,
812                       qemu_nbd_object_print_help, &error_fatal);
813 
814     if (!trace_init_backends()) {
815         exit(1);
816     }
817     trace_init_file(trace_file);
818     qemu_set_log(LOG_TRACE);
819 
820     socket_activation = check_socket_activation();
821     if (socket_activation == 0) {
822         setup_address_and_port(&bindto, &port);
823     } else {
824         /* Using socket activation - check user didn't use -p etc. */
825         const char *err_msg = socket_activation_validate_opts(device, sockpath,
826                                                               bindto, port,
827                                                               list);
828         if (err_msg != NULL) {
829             error_report("%s", err_msg);
830             exit(EXIT_FAILURE);
831         }
832 
833         /* qemu-nbd can only listen on a single socket.  */
834         if (socket_activation > 1) {
835             error_report("qemu-nbd does not support socket activation with %s > 1",
836                          "LISTEN_FDS");
837             exit(EXIT_FAILURE);
838         }
839     }
840 
841     if (tlscredsid) {
842         if (sockpath) {
843             error_report("TLS is only supported with IPv4/IPv6");
844             exit(EXIT_FAILURE);
845         }
846         if (device) {
847             error_report("TLS is not supported with a host device");
848             exit(EXIT_FAILURE);
849         }
850         if (tlsauthz && list) {
851             error_report("TLS authorization is incompatible with export list");
852             exit(EXIT_FAILURE);
853         }
854         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
855         if (local_err) {
856             error_reportf_err(local_err, "Failed to get TLS creds: ");
857             exit(EXIT_FAILURE);
858         }
859     } else {
860         if (tlsauthz) {
861             error_report("--tls-authz is not permitted without --tls-creds");
862             exit(EXIT_FAILURE);
863         }
864     }
865 
866     if (list) {
867         saddr = nbd_build_socket_address(sockpath, bindto, port);
868         return qemu_nbd_client_list(saddr, tlscreds, bindto);
869     }
870 
871 #if !HAVE_NBD_DEVICE
872     if (disconnect || device) {
873         error_report("Kernel /dev/nbdN support not available");
874         exit(EXIT_FAILURE);
875     }
876 #else /* HAVE_NBD_DEVICE */
877     if (disconnect) {
878         int nbdfd = open(argv[optind], O_RDWR);
879         if (nbdfd < 0) {
880             error_report("Cannot open %s: %s", argv[optind],
881                          strerror(errno));
882             exit(EXIT_FAILURE);
883         }
884         nbd_disconnect(nbdfd);
885 
886         close(nbdfd);
887 
888         printf("%s disconnected\n", argv[optind]);
889 
890         return 0;
891     }
892 #endif
893 
894     if ((device && !verbose) || fork_process) {
895 #ifndef WIN32
896         int stderr_fd[2];
897         pid_t pid;
898         int ret;
899 
900         if (qemu_pipe(stderr_fd) < 0) {
901             error_report("Error setting up communication pipe: %s",
902                          strerror(errno));
903             exit(EXIT_FAILURE);
904         }
905 
906         /* Now daemonize, but keep a communication channel open to
907          * print errors and exit with the proper status code.
908          */
909         pid = fork();
910         if (pid < 0) {
911             error_report("Failed to fork: %s", strerror(errno));
912             exit(EXIT_FAILURE);
913         } else if (pid == 0) {
914             close(stderr_fd[0]);
915 
916             /* Remember parent's stderr if we will be restoring it. */
917             if (fork_process) {
918                 old_stderr = dup(STDERR_FILENO);
919             }
920 
921             ret = qemu_daemon(1, 0);
922 
923             /* Temporarily redirect stderr to the parent's pipe...  */
924             dup2(stderr_fd[1], STDERR_FILENO);
925             if (ret < 0) {
926                 error_report("Failed to daemonize: %s", strerror(errno));
927                 exit(EXIT_FAILURE);
928             }
929 
930             /* ... close the descriptor we inherited and go on.  */
931             close(stderr_fd[1]);
932         } else {
933             bool errors = false;
934             char *buf;
935 
936             /* In the parent.  Print error messages from the child until
937              * it closes the pipe.
938              */
939             close(stderr_fd[1]);
940             buf = g_malloc(1024);
941             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
942                 errors = true;
943                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
944                 if (ret < 0) {
945                     exit(EXIT_FAILURE);
946                 }
947             }
948             if (ret < 0) {
949                 error_report("Cannot read from daemon: %s",
950                              strerror(errno));
951                 exit(EXIT_FAILURE);
952             }
953 
954             /* Usually the daemon should not print any message.
955              * Exit with zero status in that case.
956              */
957             exit(errors);
958         }
959 #else /* WIN32 */
960         error_report("Unable to fork into background on Windows hosts");
961         exit(EXIT_FAILURE);
962 #endif /* WIN32 */
963     }
964 
965     if (device != NULL && sockpath == NULL) {
966         sockpath = g_malloc(128);
967         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
968     }
969 
970     server = qio_net_listener_new();
971     if (socket_activation == 0) {
972         saddr = nbd_build_socket_address(sockpath, bindto, port);
973         if (qio_net_listener_open_sync(server, saddr, 1, &local_err) < 0) {
974             object_unref(OBJECT(server));
975             error_report_err(local_err);
976             exit(EXIT_FAILURE);
977         }
978     } else {
979         size_t i;
980         /* See comment in check_socket_activation above. */
981         for (i = 0; i < socket_activation; i++) {
982             QIOChannelSocket *sioc;
983             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
984                                              &local_err);
985             if (sioc == NULL) {
986                 object_unref(OBJECT(server));
987                 error_reportf_err(local_err,
988                                   "Failed to use socket activation: ");
989                 exit(EXIT_FAILURE);
990             }
991             qio_net_listener_add(server, sioc);
992             object_unref(OBJECT(sioc));
993         }
994     }
995 
996     if (qemu_init_main_loop(&local_err)) {
997         error_report_err(local_err);
998         exit(EXIT_FAILURE);
999     }
1000     bdrv_init();
1001     atexit(qemu_nbd_shutdown);
1002 
1003     srcpath = argv[optind];
1004     if (imageOpts) {
1005         QemuOpts *opts;
1006         if (fmt) {
1007             error_report("--image-opts and -f are mutually exclusive");
1008             exit(EXIT_FAILURE);
1009         }
1010         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1011         if (!opts) {
1012             qemu_opts_reset(&file_opts);
1013             exit(EXIT_FAILURE);
1014         }
1015         options = qemu_opts_to_qdict(opts, NULL);
1016         qemu_opts_reset(&file_opts);
1017         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1018     } else {
1019         if (fmt) {
1020             options = qdict_new();
1021             qdict_put_str(options, "driver", fmt);
1022         }
1023         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1024     }
1025 
1026     if (!blk) {
1027         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1028                           argv[optind]);
1029         exit(EXIT_FAILURE);
1030     }
1031     bs = blk_bs(blk);
1032 
1033     if (dev_offset) {
1034         QDict *raw_opts = qdict_new();
1035         qdict_put_str(raw_opts, "driver", "raw");
1036         qdict_put_str(raw_opts, "file", bs->node_name);
1037         qdict_put_int(raw_opts, "offset", dev_offset);
1038         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1039         blk_remove_bs(blk);
1040         blk_insert_bs(blk, bs, &error_fatal);
1041         bdrv_unref(bs);
1042     }
1043 
1044     blk_set_enable_write_cache(blk, !writethrough);
1045 
1046     if (sn_opts) {
1047         ret = bdrv_snapshot_load_tmp(bs,
1048                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1049                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1050                                      &local_err);
1051     } else if (sn_id_or_name) {
1052         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1053                                                    &local_err);
1054     }
1055     if (ret < 0) {
1056         error_reportf_err(local_err, "Failed to load snapshot: ");
1057         exit(EXIT_FAILURE);
1058     }
1059 
1060     bs->detect_zeroes = detect_zeroes;
1061 
1062     nbd_server_is_qemu_nbd(true);
1063 
1064     export_opts = g_new(BlockExportOptions, 1);
1065     *export_opts = (BlockExportOptions) {
1066         .type               = BLOCK_EXPORT_TYPE_NBD,
1067         .has_writethrough   = true,
1068         .writethrough       = writethrough,
1069         .u.nbd = {
1070             .device             = g_strdup(bdrv_get_node_name(bs)),
1071             .has_name           = true,
1072             .name               = g_strdup(export_name),
1073             .has_description    = !!export_description,
1074             .description        = g_strdup(export_description),
1075             .has_writable       = true,
1076             .writable           = !readonly,
1077             .has_bitmap         = !!bitmap,
1078             .bitmap             = g_strdup(bitmap),
1079         },
1080     };
1081     blk_exp_add(export_opts, &error_fatal);
1082     qapi_free_BlockExportOptions(export_opts);
1083 
1084     if (device) {
1085 #if HAVE_NBD_DEVICE
1086         int ret;
1087 
1088         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1089         if (ret != 0) {
1090             error_report("Failed to create client thread: %s", strerror(ret));
1091             exit(EXIT_FAILURE);
1092         }
1093 #endif
1094     } else {
1095         /* Shut up GCC warnings.  */
1096         memset(&client_thread, 0, sizeof(client_thread));
1097     }
1098 
1099     nbd_update_server_watch();
1100 
1101     if (pid_file_name) {
1102         qemu_write_pidfile(pid_file_name, &error_fatal);
1103     }
1104 
1105     /* now when the initialization is (almost) complete, chdir("/")
1106      * to free any busy filesystems */
1107     if (chdir("/") < 0) {
1108         error_report("Could not chdir to root directory: %s",
1109                      strerror(errno));
1110         exit(EXIT_FAILURE);
1111     }
1112 
1113     if (fork_process) {
1114         dup2(old_stderr, STDERR_FILENO);
1115         close(old_stderr);
1116     }
1117 
1118     state = RUNNING;
1119     do {
1120         main_loop_wait(false);
1121         if (state == TERMINATE) {
1122             nbd_export_close_all();
1123             state = TERMINATED;
1124         }
1125     } while (state != TERMINATED);
1126 
1127     blk_unref(blk);
1128     if (sockpath) {
1129         unlink(sockpath);
1130     }
1131 
1132     qemu_opts_del(sn_opts);
1133 
1134     if (device) {
1135         void *ret;
1136         pthread_join(client_thread, &ret);
1137         exit(ret != NULL);
1138     } else {
1139         exit(EXIT_SUCCESS);
1140     }
1141 }
1142