xref: /openbmc/qemu/qemu-nbd.c (revision 6016b7b4)
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 "sysemu/runstate.h" /* for qemu_system_killed() prototype */
29 #include "block/block_int.h"
30 #include "block/nbd.h"
31 #include "qemu/main-loop.h"
32 #include "qemu/module.h"
33 #include "qemu/option.h"
34 #include "qemu/error-report.h"
35 #include "qemu/config-file.h"
36 #include "qemu/bswap.h"
37 #include "qemu/log.h"
38 #include "qemu/systemd.h"
39 #include "block/snapshot.h"
40 #include "qapi/qmp/qdict.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qom/object_interfaces.h"
43 #include "io/channel-socket.h"
44 #include "io/net-listener.h"
45 #include "crypto/init.h"
46 #include "crypto/tlscreds.h"
47 #include "trace/control.h"
48 #include "qemu-version.h"
49 
50 #ifdef CONFIG_SELINUX
51 #include <selinux/selinux.h>
52 #endif
53 
54 #ifdef __linux__
55 #define HAVE_NBD_DEVICE 1
56 #else
57 #define HAVE_NBD_DEVICE 0
58 #endif
59 
60 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
61 #define QEMU_NBD_OPT_CACHE         256
62 #define QEMU_NBD_OPT_AIO           257
63 #define QEMU_NBD_OPT_DISCARD       258
64 #define QEMU_NBD_OPT_DETECT_ZEROES 259
65 #define QEMU_NBD_OPT_OBJECT        260
66 #define QEMU_NBD_OPT_TLSCREDS      261
67 #define QEMU_NBD_OPT_IMAGE_OPTS    262
68 #define QEMU_NBD_OPT_FORK          263
69 #define QEMU_NBD_OPT_TLSAUTHZ      264
70 #define QEMU_NBD_OPT_PID_FILE      265
71 #define QEMU_NBD_OPT_SELINUX_LABEL 266
72 
73 #define MBR_SIZE 512
74 
75 static int verbose;
76 static char *srcpath;
77 static SocketAddress *saddr;
78 static int persistent = 0;
79 static enum { RUNNING, TERMINATE, TERMINATED } state;
80 static int shared = 1;
81 static int nb_fds;
82 static QIONetListener *server;
83 static QCryptoTLSCreds *tlscreds;
84 static const char *tlsauthz;
85 
86 static void usage(const char *name)
87 {
88     (printf) (
89 "Usage: %s [OPTIONS] FILE\n"
90 "  or:  %s -L [OPTIONS]\n"
91 "QEMU Disk Network Block Device Utility\n"
92 "\n"
93 "  -h, --help                display this help and exit\n"
94 "  -V, --version             output version information and exit\n"
95 "\n"
96 "Connection properties:\n"
97 "  -p, --port=PORT           port to listen on (default `%d')\n"
98 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
99 "  -k, --socket=PATH         path to the unix socket\n"
100 "                            (default '"SOCKET_PATH"')\n"
101 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
102 "  -t, --persistent          don't exit on the last connection\n"
103 "  -v, --verbose             display extra debugging information\n"
104 "  -x, --export-name=NAME    expose export by name (default is empty string)\n"
105 "  -D, --description=TEXT    export a human-readable description\n"
106 "\n"
107 "Exposing part of the image:\n"
108 "  -o, --offset=OFFSET       offset into the image\n"
109 "  -A, --allocation-depth    expose the allocation depth\n"
110 "  -B, --bitmap=NAME         expose a persistent dirty bitmap\n"
111 "\n"
112 "General purpose options:\n"
113 "  -L, --list                list exports available from another NBD server\n"
114 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
115 "                            passwords and/or encryption keys\n"
116 "  --tls-creds=ID            use id of an earlier --object to provide TLS\n"
117 "  --tls-authz=ID            use id of an earlier --object to provide\n"
118 "                            authorization\n"
119 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
120 "                            specify tracing options\n"
121 "  --fork                    fork off the server process and exit the parent\n"
122 "                            once the server is running\n"
123 "  --pid-file=PATH           store the server's process ID in the given file\n"
124 #ifdef CONFIG_SELINUX
125 "  --selinux-label=LABEL     set SELinux process label on listening socket\n"
126 #endif
127 #if HAVE_NBD_DEVICE
128 "\n"
129 "Kernel NBD client support:\n"
130 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
131 "  -d, --disconnect          disconnect the specified device\n"
132 #endif
133 "\n"
134 "Block device options:\n"
135 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
136 "  -r, --read-only           export read-only\n"
137 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
138 "                            file with backing_file=FILE, redirect the write to\n"
139 "                            the temporary one\n"
140 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
141 "                            load an internal snapshot inside FILE and export it\n"
142 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
143 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
144 "                            '[ID_OR_NAME]'\n"
145 "  -n, --nocache             disable host cache\n"
146 "      --cache=MODE          set cache mode used to access the disk image, the\n"
147 "                            valid options are: 'none', 'writeback' (default),\n"
148 "                            'writethrough', 'directsync' and 'unsafe'\n"
149 "      --aio=MODE            set AIO mode (native, io_uring or threads)\n"
150 "      --discard=MODE        set discard mode (ignore, unmap)\n"
151 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
152 "      --image-opts          treat FILE as a full set of image options\n"
153 "\n"
154 QEMU_HELP_BOTTOM "\n"
155     , name, name, NBD_DEFAULT_PORT, "DEVICE");
156 }
157 
158 static void version(const char *name)
159 {
160     printf(
161 "%s " QEMU_FULL_VERSION "\n"
162 "Written by Anthony Liguori.\n"
163 "\n"
164 QEMU_COPYRIGHT "\n"
165 "This is free software; see the source for copying conditions.  There is NO\n"
166 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
167     , name);
168 }
169 
170 #ifdef CONFIG_POSIX
171 /*
172  * The client thread uses SIGTERM to interrupt the server.  A signal
173  * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
174  */
175 void qemu_system_killed(int signum, pid_t pid)
176 {
177     qatomic_cmpxchg(&state, RUNNING, TERMINATE);
178     qemu_notify_event();
179 }
180 #endif /* CONFIG_POSIX */
181 
182 static int qemu_nbd_client_list(SocketAddress *saddr, QCryptoTLSCreds *tls,
183                                 const char *hostname)
184 {
185     int ret = EXIT_FAILURE;
186     int rc;
187     Error *err = NULL;
188     QIOChannelSocket *sioc;
189     NBDExportInfo *list;
190     int i, j;
191 
192     sioc = qio_channel_socket_new();
193     if (qio_channel_socket_connect_sync(sioc, saddr, &err) < 0) {
194         error_report_err(err);
195         goto out;
196     }
197     rc = nbd_receive_export_list(QIO_CHANNEL(sioc), tls, hostname, &list,
198                                  &err);
199     if (rc < 0) {
200         if (err) {
201             error_report_err(err);
202         }
203         goto out;
204     }
205     printf("exports available: %d\n", rc);
206     for (i = 0; i < rc; i++) {
207         printf(" export: '%s'\n", list[i].name);
208         if (list[i].description && *list[i].description) {
209             printf("  description: %s\n", list[i].description);
210         }
211         if (list[i].flags & NBD_FLAG_HAS_FLAGS) {
212             static const char *const flag_names[] = {
213                 [NBD_FLAG_READ_ONLY_BIT]            = "readonly",
214                 [NBD_FLAG_SEND_FLUSH_BIT]           = "flush",
215                 [NBD_FLAG_SEND_FUA_BIT]             = "fua",
216                 [NBD_FLAG_ROTATIONAL_BIT]           = "rotational",
217                 [NBD_FLAG_SEND_TRIM_BIT]            = "trim",
218                 [NBD_FLAG_SEND_WRITE_ZEROES_BIT]    = "zeroes",
219                 [NBD_FLAG_SEND_DF_BIT]              = "df",
220                 [NBD_FLAG_CAN_MULTI_CONN_BIT]       = "multi",
221                 [NBD_FLAG_SEND_RESIZE_BIT]          = "resize",
222                 [NBD_FLAG_SEND_CACHE_BIT]           = "cache",
223                 [NBD_FLAG_SEND_FAST_ZERO_BIT]       = "fast-zero",
224             };
225 
226             printf("  size:  %" PRIu64 "\n", list[i].size);
227             printf("  flags: 0x%x (", list[i].flags);
228             for (size_t bit = 0; bit < ARRAY_SIZE(flag_names); bit++) {
229                 if (flag_names[bit] && (list[i].flags & (1 << bit))) {
230                     printf(" %s", flag_names[bit]);
231                 }
232             }
233             printf(" )\n");
234         }
235         if (list[i].min_block) {
236             printf("  min block: %u\n", list[i].min_block);
237             printf("  opt block: %u\n", list[i].opt_block);
238             printf("  max block: %u\n", list[i].max_block);
239         }
240         if (list[i].n_contexts) {
241             printf("  available meta contexts: %d\n", list[i].n_contexts);
242             for (j = 0; j < list[i].n_contexts; j++) {
243                 printf("   %s\n", list[i].contexts[j]);
244             }
245         }
246     }
247     nbd_free_export_list(list, rc);
248 
249     ret = EXIT_SUCCESS;
250  out:
251     object_unref(OBJECT(sioc));
252     return ret;
253 }
254 
255 
256 #if HAVE_NBD_DEVICE
257 static void *show_parts(void *arg)
258 {
259     char *device = arg;
260     int nbd;
261 
262     /* linux just needs an open() to trigger
263      * the partition table update
264      * but remember to load the module with max_part != 0 :
265      *     modprobe nbd max_part=63
266      */
267     nbd = open(device, O_RDWR);
268     if (nbd >= 0) {
269         close(nbd);
270     }
271     return NULL;
272 }
273 
274 static void *nbd_client_thread(void *arg)
275 {
276     char *device = arg;
277     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
278     QIOChannelSocket *sioc;
279     int fd = -1;
280     int ret = EXIT_FAILURE;
281     pthread_t show_parts_thread;
282     Error *local_error = NULL;
283 
284     sioc = qio_channel_socket_new();
285     if (qio_channel_socket_connect_sync(sioc,
286                                         saddr,
287                                         &local_error) < 0) {
288         error_report_err(local_error);
289         goto out;
290     }
291 
292     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
293                               NULL, NULL, NULL, &info, &local_error) < 0) {
294         if (local_error) {
295             error_report_err(local_error);
296         }
297         goto out;
298     }
299 
300     fd = open(device, O_RDWR);
301     if (fd < 0) {
302         /* Linux-only, we can use %m in printf.  */
303         error_report("Failed to open %s: %m", device);
304         goto out;
305     }
306 
307     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
308         error_report_err(local_error);
309         goto out;
310     }
311 
312     /* update partition table */
313     pthread_create(&show_parts_thread, NULL, show_parts, device);
314 
315     if (verbose) {
316         fprintf(stderr, "NBD device %s is now connected to %s\n",
317                 device, srcpath);
318     } else {
319         /* Close stderr so that the qemu-nbd process exits.  */
320         dup2(STDOUT_FILENO, STDERR_FILENO);
321     }
322 
323     if (nbd_client(fd) < 0) {
324         goto out;
325     }
326 
327     ret = EXIT_SUCCESS;
328 
329  out:
330     if (fd >= 0) {
331         close(fd);
332     }
333     object_unref(OBJECT(sioc));
334     g_free(info.name);
335     kill(getpid(), SIGTERM);
336     return (void *) (intptr_t) ret;
337 }
338 #endif /* HAVE_NBD_DEVICE */
339 
340 static int nbd_can_accept(void)
341 {
342     return state == RUNNING && (shared == 0 || nb_fds < shared);
343 }
344 
345 static void nbd_update_server_watch(void);
346 
347 static void nbd_client_closed(NBDClient *client, bool negotiated)
348 {
349     nb_fds--;
350     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
351         state = TERMINATE;
352     }
353     nbd_update_server_watch();
354     nbd_client_put(client);
355 }
356 
357 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
358                        gpointer opaque)
359 {
360     if (state >= TERMINATE) {
361         return;
362     }
363 
364     nb_fds++;
365     nbd_update_server_watch();
366     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
367 }
368 
369 static void nbd_update_server_watch(void)
370 {
371     if (nbd_can_accept()) {
372         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
373     } else {
374         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
375     }
376 }
377 
378 
379 static SocketAddress *nbd_build_socket_address(const char *sockpath,
380                                                const char *bindto,
381                                                const char *port)
382 {
383     SocketAddress *saddr;
384 
385     saddr = g_new0(SocketAddress, 1);
386     if (sockpath) {
387         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
388         saddr->u.q_unix.path = g_strdup(sockpath);
389     } else {
390         InetSocketAddress *inet;
391         saddr->type = SOCKET_ADDRESS_TYPE_INET;
392         inet = &saddr->u.inet;
393         inet->host = g_strdup(bindto);
394         if (port) {
395             inet->port = g_strdup(port);
396         } else  {
397             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
398         }
399     }
400 
401     return saddr;
402 }
403 
404 
405 static QemuOptsList file_opts = {
406     .name = "file",
407     .implied_opt_name = "file",
408     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
409     .desc = {
410         /* no elements => accept any params */
411         { /* end of list */ }
412     },
413 };
414 
415 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
416                                           Error **errp)
417 {
418     Object *obj;
419     QCryptoTLSCreds *creds;
420 
421     obj = object_resolve_path_component(
422         object_get_objects_root(), id);
423     if (!obj) {
424         error_setg(errp, "No TLS credentials with id '%s'",
425                    id);
426         return NULL;
427     }
428     creds = (QCryptoTLSCreds *)
429         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
430     if (!creds) {
431         error_setg(errp, "Object with id '%s' is not TLS credentials",
432                    id);
433         return NULL;
434     }
435 
436     if (!qcrypto_tls_creds_check_endpoint(creds,
437                                           list
438                                           ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
439                                           : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
440                                           errp)) {
441         return NULL;
442     }
443     object_ref(obj);
444     return creds;
445 }
446 
447 static void setup_address_and_port(const char **address, const char **port)
448 {
449     if (*address == NULL) {
450         *address = "0.0.0.0";
451     }
452 
453     if (*port == NULL) {
454         *port = stringify(NBD_DEFAULT_PORT);
455     }
456 }
457 
458 /*
459  * Check socket parameters compatibility when socket activation is used.
460  */
461 static const char *socket_activation_validate_opts(const char *device,
462                                                    const char *sockpath,
463                                                    const char *address,
464                                                    const char *port,
465                                                    const char *selinux,
466                                                    bool list)
467 {
468     if (device != NULL) {
469         return "NBD device can't be set when using socket activation";
470     }
471 
472     if (sockpath != NULL) {
473         return "Unix socket can't be set when using socket activation";
474     }
475 
476     if (address != NULL) {
477         return "The interface can't be set when using socket activation";
478     }
479 
480     if (port != NULL) {
481         return "TCP port number can't be set when using socket activation";
482     }
483 
484     if (selinux != NULL) {
485         return "SELinux label can't be set when using socket activation";
486     }
487 
488     if (list) {
489         return "List mode is incompatible with socket activation";
490     }
491 
492     return NULL;
493 }
494 
495 static void qemu_nbd_shutdown(void)
496 {
497     job_cancel_sync_all();
498     blk_exp_close_all();
499     bdrv_close_all();
500 }
501 
502 int main(int argc, char **argv)
503 {
504     BlockBackend *blk;
505     BlockDriverState *bs;
506     uint64_t dev_offset = 0;
507     bool readonly = false;
508     bool disconnect = false;
509     const char *bindto = NULL;
510     const char *port = NULL;
511     char *sockpath = NULL;
512     char *device = NULL;
513     QemuOpts *sn_opts = NULL;
514     const char *sn_id_or_name = NULL;
515     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
516     struct option lopt[] = {
517         { "help", no_argument, NULL, 'h' },
518         { "version", no_argument, NULL, 'V' },
519         { "bind", required_argument, NULL, 'b' },
520         { "port", required_argument, NULL, 'p' },
521         { "socket", required_argument, NULL, 'k' },
522         { "offset", required_argument, NULL, 'o' },
523         { "read-only", no_argument, NULL, 'r' },
524         { "allocation-depth", no_argument, NULL, 'A' },
525         { "bitmap", required_argument, NULL, 'B' },
526         { "connect", required_argument, NULL, 'c' },
527         { "disconnect", no_argument, NULL, 'd' },
528         { "list", no_argument, NULL, 'L' },
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' },
543         { "description", required_argument, NULL, 'D' },
544         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
545         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
546         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
547         { "trace", required_argument, NULL, 'T' },
548         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
549         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
550         { "selinux-label", required_argument, NULL,
551           QEMU_NBD_OPT_SELINUX_LABEL },
552         { NULL, 0, NULL, 0 }
553     };
554     int ch;
555     int opt_ind = 0;
556     int flags = BDRV_O_RDWR;
557     int ret = 0;
558     bool seen_cache = false;
559     bool seen_discard = false;
560     bool seen_aio = false;
561     pthread_t client_thread;
562     const char *fmt = NULL;
563     Error *local_err = NULL;
564     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
565     QDict *options = NULL;
566     const char *export_name = NULL; /* defaults to "" later for server mode */
567     const char *export_description = NULL;
568     strList *bitmaps = NULL;
569     bool alloc_depth = false;
570     const char *tlscredsid = NULL;
571     bool imageOpts = false;
572     bool writethrough = false; /* Client will flush as needed. */
573     bool fork_process = false;
574     bool list = false;
575     int old_stderr = -1;
576     unsigned socket_activation;
577     const char *pid_file_name = NULL;
578     const char *selinux_label = NULL;
579     BlockExportOptions *export_opts;
580 
581 #ifdef CONFIG_POSIX
582     os_setup_early_signal_handling();
583     os_setup_signal_handling();
584 #endif
585 
586     socket_init();
587     error_init(argv[0]);
588     module_call_init(MODULE_INIT_TRACE);
589     qcrypto_init(&error_fatal);
590 
591     module_call_init(MODULE_INIT_QOM);
592     qemu_add_opts(&qemu_trace_opts);
593     qemu_init_exec_dir(argv[0]);
594 
595     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
596         switch (ch) {
597         case 's':
598             flags |= BDRV_O_SNAPSHOT;
599             break;
600         case 'n':
601             optarg = (char *) "none";
602             /* fallthrough */
603         case QEMU_NBD_OPT_CACHE:
604             if (seen_cache) {
605                 error_report("-n and --cache can only be specified once");
606                 exit(EXIT_FAILURE);
607             }
608             seen_cache = true;
609             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
610                 error_report("Invalid cache mode `%s'", optarg);
611                 exit(EXIT_FAILURE);
612             }
613             break;
614         case QEMU_NBD_OPT_AIO:
615             if (seen_aio) {
616                 error_report("--aio can only be specified once");
617                 exit(EXIT_FAILURE);
618             }
619             seen_aio = true;
620             if (bdrv_parse_aio(optarg, &flags) < 0) {
621                 error_report("Invalid aio mode '%s'", optarg);
622                 exit(EXIT_FAILURE);
623             }
624             break;
625         case QEMU_NBD_OPT_DISCARD:
626             if (seen_discard) {
627                 error_report("--discard can only be specified once");
628                 exit(EXIT_FAILURE);
629             }
630             seen_discard = true;
631             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
632                 error_report("Invalid discard mode `%s'", optarg);
633                 exit(EXIT_FAILURE);
634             }
635             break;
636         case QEMU_NBD_OPT_DETECT_ZEROES:
637             detect_zeroes =
638                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
639                                 optarg,
640                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
641                                 &local_err);
642             if (local_err) {
643                 error_reportf_err(local_err,
644                                   "Failed to parse detect_zeroes mode: ");
645                 exit(EXIT_FAILURE);
646             }
647             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
648                 !(flags & BDRV_O_UNMAP)) {
649                 error_report("setting detect-zeroes to unmap is not allowed "
650                              "without setting discard operation to unmap");
651                 exit(EXIT_FAILURE);
652             }
653             break;
654         case 'b':
655             bindto = optarg;
656             break;
657         case 'p':
658             port = optarg;
659             break;
660         case 'o':
661             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
662                 error_report("Invalid offset '%s'", optarg);
663                 exit(EXIT_FAILURE);
664             }
665             break;
666         case 'l':
667             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
668                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
669                                                   optarg, false);
670                 if (!sn_opts) {
671                     error_report("Failed in parsing snapshot param `%s'",
672                                  optarg);
673                     exit(EXIT_FAILURE);
674                 }
675             } else {
676                 sn_id_or_name = optarg;
677             }
678             /* fall through */
679         case 'r':
680             readonly = true;
681             flags &= ~BDRV_O_RDWR;
682             break;
683         case 'A':
684             alloc_depth = true;
685             break;
686         case 'B':
687             QAPI_LIST_PREPEND(bitmaps, g_strdup(optarg));
688             break;
689         case 'k':
690             sockpath = optarg;
691             if (sockpath[0] != '/') {
692                 error_report("socket path must be absolute");
693                 exit(EXIT_FAILURE);
694             }
695             break;
696         case 'd':
697             disconnect = true;
698             break;
699         case 'c':
700             device = optarg;
701             break;
702         case 'e':
703             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
704                 shared < 0) {
705                 error_report("Invalid shared device number '%s'", optarg);
706                 exit(EXIT_FAILURE);
707             }
708             break;
709         case 'f':
710             fmt = optarg;
711             break;
712         case 't':
713             persistent = 1;
714             break;
715         case 'x':
716             export_name = optarg;
717             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
718                 error_report("export name '%s' too long", export_name);
719                 exit(EXIT_FAILURE);
720             }
721             break;
722         case 'D':
723             export_description = optarg;
724             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
725                 error_report("export description '%s' too long",
726                              export_description);
727                 exit(EXIT_FAILURE);
728             }
729             break;
730         case 'v':
731             verbose = 1;
732             break;
733         case 'V':
734             version(argv[0]);
735             exit(0);
736             break;
737         case 'h':
738             usage(argv[0]);
739             exit(0);
740             break;
741         case '?':
742             error_report("Try `%s --help' for more information.", argv[0]);
743             exit(EXIT_FAILURE);
744         case QEMU_NBD_OPT_OBJECT:
745             user_creatable_process_cmdline(optarg);
746             break;
747         case QEMU_NBD_OPT_TLSCREDS:
748             tlscredsid = optarg;
749             break;
750         case QEMU_NBD_OPT_IMAGE_OPTS:
751             imageOpts = true;
752             break;
753         case 'T':
754             trace_opt_parse(optarg);
755             break;
756         case QEMU_NBD_OPT_TLSAUTHZ:
757             tlsauthz = optarg;
758             break;
759         case QEMU_NBD_OPT_FORK:
760             fork_process = true;
761             break;
762         case 'L':
763             list = true;
764             break;
765         case QEMU_NBD_OPT_PID_FILE:
766             pid_file_name = optarg;
767             break;
768         case QEMU_NBD_OPT_SELINUX_LABEL:
769             selinux_label = optarg;
770             break;
771         }
772     }
773 
774     if (list) {
775         if (argc != optind) {
776             error_report("List mode is incompatible with a file name");
777             exit(EXIT_FAILURE);
778         }
779         if (export_name || export_description || dev_offset ||
780             device || disconnect || fmt || sn_id_or_name || bitmaps ||
781             alloc_depth || seen_aio || seen_discard || seen_cache) {
782             error_report("List mode is incompatible with per-device settings");
783             exit(EXIT_FAILURE);
784         }
785         if (fork_process) {
786             error_report("List mode is incompatible with forking");
787             exit(EXIT_FAILURE);
788         }
789     } else if ((argc - optind) != 1) {
790         error_report("Invalid number of arguments");
791         error_printf("Try `%s --help' for more information.\n", argv[0]);
792         exit(EXIT_FAILURE);
793     } else if (!export_name) {
794         export_name = "";
795     }
796 
797     if (!trace_init_backends()) {
798         exit(1);
799     }
800     trace_init_file();
801     qemu_set_log(LOG_TRACE);
802 
803     socket_activation = check_socket_activation();
804     if (socket_activation == 0) {
805         setup_address_and_port(&bindto, &port);
806     } else {
807         /* Using socket activation - check user didn't use -p etc. */
808         const char *err_msg = socket_activation_validate_opts(device, sockpath,
809                                                               bindto, port,
810                                                               selinux_label,
811                                                               list);
812         if (err_msg != NULL) {
813             error_report("%s", err_msg);
814             exit(EXIT_FAILURE);
815         }
816 
817         /* qemu-nbd can only listen on a single socket.  */
818         if (socket_activation > 1) {
819             error_report("qemu-nbd does not support socket activation with %s > 1",
820                          "LISTEN_FDS");
821             exit(EXIT_FAILURE);
822         }
823     }
824 
825     if (tlscredsid) {
826         if (sockpath) {
827             error_report("TLS is only supported with IPv4/IPv6");
828             exit(EXIT_FAILURE);
829         }
830         if (device) {
831             error_report("TLS is not supported with a host device");
832             exit(EXIT_FAILURE);
833         }
834         if (tlsauthz && list) {
835             error_report("TLS authorization is incompatible with export list");
836             exit(EXIT_FAILURE);
837         }
838         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
839         if (local_err) {
840             error_reportf_err(local_err, "Failed to get TLS creds: ");
841             exit(EXIT_FAILURE);
842         }
843     } else {
844         if (tlsauthz) {
845             error_report("--tls-authz is not permitted without --tls-creds");
846             exit(EXIT_FAILURE);
847         }
848     }
849 
850     if (selinux_label) {
851 #ifdef CONFIG_SELINUX
852         if (sockpath == NULL && device == NULL) {
853             error_report("--selinux-label is not permitted without --socket");
854             exit(EXIT_FAILURE);
855         }
856 #else
857         error_report("SELinux support not enabled in this binary");
858         exit(EXIT_FAILURE);
859 #endif
860     }
861 
862     if (list) {
863         saddr = nbd_build_socket_address(sockpath, bindto, port);
864         return qemu_nbd_client_list(saddr, tlscreds, bindto);
865     }
866 
867 #if !HAVE_NBD_DEVICE
868     if (disconnect || device) {
869         error_report("Kernel /dev/nbdN support not available");
870         exit(EXIT_FAILURE);
871     }
872 #else /* HAVE_NBD_DEVICE */
873     if (disconnect) {
874         int nbdfd = open(argv[optind], O_RDWR);
875         if (nbdfd < 0) {
876             error_report("Cannot open %s: %s", argv[optind],
877                          strerror(errno));
878             exit(EXIT_FAILURE);
879         }
880         nbd_disconnect(nbdfd);
881 
882         close(nbdfd);
883 
884         printf("%s disconnected\n", argv[optind]);
885 
886         return 0;
887     }
888 #endif
889 
890     if ((device && !verbose) || fork_process) {
891 #ifndef WIN32
892         int stderr_fd[2];
893         pid_t pid;
894         int ret;
895 
896         if (qemu_pipe(stderr_fd) < 0) {
897             error_report("Error setting up communication pipe: %s",
898                          strerror(errno));
899             exit(EXIT_FAILURE);
900         }
901 
902         /* Now daemonize, but keep a communication channel open to
903          * print errors and exit with the proper status code.
904          */
905         pid = fork();
906         if (pid < 0) {
907             error_report("Failed to fork: %s", strerror(errno));
908             exit(EXIT_FAILURE);
909         } else if (pid == 0) {
910             close(stderr_fd[0]);
911 
912             /* Remember parent's stderr if we will be restoring it. */
913             if (fork_process) {
914                 old_stderr = dup(STDERR_FILENO);
915             }
916 
917             ret = qemu_daemon(1, 0);
918 
919             /* Temporarily redirect stderr to the parent's pipe...  */
920             dup2(stderr_fd[1], STDERR_FILENO);
921             if (ret < 0) {
922                 error_report("Failed to daemonize: %s", strerror(errno));
923                 exit(EXIT_FAILURE);
924             }
925 
926             /* ... close the descriptor we inherited and go on.  */
927             close(stderr_fd[1]);
928         } else {
929             bool errors = false;
930             char *buf;
931 
932             /* In the parent.  Print error messages from the child until
933              * it closes the pipe.
934              */
935             close(stderr_fd[1]);
936             buf = g_malloc(1024);
937             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
938                 errors = true;
939                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
940                 if (ret < 0) {
941                     exit(EXIT_FAILURE);
942                 }
943             }
944             if (ret < 0) {
945                 error_report("Cannot read from daemon: %s",
946                              strerror(errno));
947                 exit(EXIT_FAILURE);
948             }
949 
950             /* Usually the daemon should not print any message.
951              * Exit with zero status in that case.
952              */
953             exit(errors);
954         }
955 #else /* WIN32 */
956         error_report("Unable to fork into background on Windows hosts");
957         exit(EXIT_FAILURE);
958 #endif /* WIN32 */
959     }
960 
961     if (device != NULL && sockpath == NULL) {
962         sockpath = g_malloc(128);
963         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
964     }
965 
966     server = qio_net_listener_new();
967     if (socket_activation == 0) {
968         int backlog;
969 
970         if (persistent || shared == 0) {
971             backlog = SOMAXCONN;
972         } else {
973             backlog = MIN(shared, SOMAXCONN);
974         }
975 #ifdef CONFIG_SELINUX
976         if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
977             error_report("Cannot set SELinux socket create context to %s: %s",
978                          selinux_label, strerror(errno));
979             exit(EXIT_FAILURE);
980         }
981 #endif
982         saddr = nbd_build_socket_address(sockpath, bindto, port);
983         if (qio_net_listener_open_sync(server, saddr, backlog,
984                                        &local_err) < 0) {
985             object_unref(OBJECT(server));
986             error_report_err(local_err);
987             exit(EXIT_FAILURE);
988         }
989 #ifdef CONFIG_SELINUX
990         if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
991             error_report("Cannot clear SELinux socket create context: %s",
992                          strerror(errno));
993             exit(EXIT_FAILURE);
994         }
995 #endif
996     } else {
997         size_t i;
998         /* See comment in check_socket_activation above. */
999         for (i = 0; i < socket_activation; i++) {
1000             QIOChannelSocket *sioc;
1001             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1002                                              &local_err);
1003             if (sioc == NULL) {
1004                 object_unref(OBJECT(server));
1005                 error_reportf_err(local_err,
1006                                   "Failed to use socket activation: ");
1007                 exit(EXIT_FAILURE);
1008             }
1009             qio_net_listener_add(server, sioc);
1010             object_unref(OBJECT(sioc));
1011         }
1012     }
1013 
1014     qemu_init_main_loop(&error_fatal);
1015     bdrv_init();
1016     atexit(qemu_nbd_shutdown);
1017 
1018     srcpath = argv[optind];
1019     if (imageOpts) {
1020         QemuOpts *opts;
1021         if (fmt) {
1022             error_report("--image-opts and -f are mutually exclusive");
1023             exit(EXIT_FAILURE);
1024         }
1025         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1026         if (!opts) {
1027             qemu_opts_reset(&file_opts);
1028             exit(EXIT_FAILURE);
1029         }
1030         options = qemu_opts_to_qdict(opts, NULL);
1031         qemu_opts_reset(&file_opts);
1032         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1033     } else {
1034         if (fmt) {
1035             options = qdict_new();
1036             qdict_put_str(options, "driver", fmt);
1037         }
1038         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1039     }
1040 
1041     if (!blk) {
1042         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1043                           argv[optind]);
1044         exit(EXIT_FAILURE);
1045     }
1046     bs = blk_bs(blk);
1047 
1048     if (dev_offset) {
1049         QDict *raw_opts = qdict_new();
1050         qdict_put_str(raw_opts, "driver", "raw");
1051         qdict_put_str(raw_opts, "file", bs->node_name);
1052         qdict_put_int(raw_opts, "offset", dev_offset);
1053         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1054         blk_remove_bs(blk);
1055         blk_insert_bs(blk, bs, &error_fatal);
1056         bdrv_unref(bs);
1057     }
1058 
1059     blk_set_enable_write_cache(blk, !writethrough);
1060 
1061     if (sn_opts) {
1062         ret = bdrv_snapshot_load_tmp(bs,
1063                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1064                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1065                                      &local_err);
1066     } else if (sn_id_or_name) {
1067         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1068                                                    &local_err);
1069     }
1070     if (ret < 0) {
1071         error_reportf_err(local_err, "Failed to load snapshot: ");
1072         exit(EXIT_FAILURE);
1073     }
1074 
1075     bs->detect_zeroes = detect_zeroes;
1076 
1077     nbd_server_is_qemu_nbd(true);
1078 
1079     export_opts = g_new(BlockExportOptions, 1);
1080     *export_opts = (BlockExportOptions) {
1081         .type               = BLOCK_EXPORT_TYPE_NBD,
1082         .id                 = g_strdup("qemu-nbd-export"),
1083         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1084         .has_writethrough   = true,
1085         .writethrough       = writethrough,
1086         .has_writable       = true,
1087         .writable           = !readonly,
1088         .u.nbd = {
1089             .has_name             = true,
1090             .name                 = g_strdup(export_name),
1091             .has_description      = !!export_description,
1092             .description          = g_strdup(export_description),
1093             .has_bitmaps          = !!bitmaps,
1094             .bitmaps              = bitmaps,
1095             .has_allocation_depth = alloc_depth,
1096             .allocation_depth     = alloc_depth,
1097         },
1098     };
1099     blk_exp_add(export_opts, &error_fatal);
1100     qapi_free_BlockExportOptions(export_opts);
1101 
1102     if (device) {
1103 #if HAVE_NBD_DEVICE
1104         int ret;
1105 
1106         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1107         if (ret != 0) {
1108             error_report("Failed to create client thread: %s", strerror(ret));
1109             exit(EXIT_FAILURE);
1110         }
1111 #endif
1112     } else {
1113         /* Shut up GCC warnings.  */
1114         memset(&client_thread, 0, sizeof(client_thread));
1115     }
1116 
1117     nbd_update_server_watch();
1118 
1119     if (pid_file_name) {
1120         qemu_write_pidfile(pid_file_name, &error_fatal);
1121     }
1122 
1123     /* now when the initialization is (almost) complete, chdir("/")
1124      * to free any busy filesystems */
1125     if (chdir("/") < 0) {
1126         error_report("Could not chdir to root directory: %s",
1127                      strerror(errno));
1128         exit(EXIT_FAILURE);
1129     }
1130 
1131     if (fork_process) {
1132         dup2(old_stderr, STDERR_FILENO);
1133         close(old_stderr);
1134     }
1135 
1136     state = RUNNING;
1137     do {
1138         main_loop_wait(false);
1139         if (state == TERMINATE) {
1140             blk_exp_close_all();
1141             state = TERMINATED;
1142         }
1143     } while (state != TERMINATED);
1144 
1145     blk_unref(blk);
1146     if (sockpath) {
1147         unlink(sockpath);
1148     }
1149 
1150     qemu_opts_del(sn_opts);
1151 
1152     if (device) {
1153         void *ret;
1154         pthread_join(client_thread, &ret);
1155         exit(ret != NULL);
1156     } else {
1157         exit(EXIT_SUCCESS);
1158     }
1159 }
1160