xref: /openbmc/qemu/qemu-nbd.c (revision 136cb9cc)
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/help-texts.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 #define QEMU_NBD_OPT_TLSHOSTNAME   267
73 
74 #define MBR_SIZE 512
75 
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 struct NbdClientOpts {
275     char *device;
276     bool fork_process;
277     bool verbose;
278 };
279 
280 static void *nbd_client_thread(void *arg)
281 {
282     struct NbdClientOpts *opts = arg;
283     NBDExportInfo info = { .request_sizes = false, .name = g_strdup("") };
284     QIOChannelSocket *sioc;
285     int fd = -1;
286     int ret = EXIT_FAILURE;
287     pthread_t show_parts_thread;
288     Error *local_error = NULL;
289 
290     sioc = qio_channel_socket_new();
291     if (qio_channel_socket_connect_sync(sioc,
292                                         saddr,
293                                         &local_error) < 0) {
294         error_report_err(local_error);
295         goto out;
296     }
297 
298     if (nbd_receive_negotiate(NULL, QIO_CHANNEL(sioc),
299                               NULL, NULL, NULL, &info, &local_error) < 0) {
300         if (local_error) {
301             error_report_err(local_error);
302         }
303         goto out;
304     }
305 
306     fd = open(opts->device, O_RDWR);
307     if (fd < 0) {
308         /* Linux-only, we can use %m in printf.  */
309         error_report("Failed to open %s: %m", opts->device);
310         goto out;
311     }
312 
313     if (nbd_init(fd, sioc, &info, &local_error) < 0) {
314         error_report_err(local_error);
315         goto out;
316     }
317 
318     /* update partition table */
319     pthread_create(&show_parts_thread, NULL, show_parts, opts->device);
320 
321     if (opts->verbose && !opts->fork_process) {
322         fprintf(stderr, "NBD device %s is now connected to %s\n",
323                 opts->device, srcpath);
324     } else {
325         /* Close stderr so that the qemu-nbd process exits.  */
326         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
327             error_report("Could not set stderr to /dev/null: %s",
328                          strerror(errno));
329             exit(EXIT_FAILURE);
330         }
331     }
332 
333     if (nbd_client(fd) < 0) {
334         goto out;
335     }
336 
337     ret = EXIT_SUCCESS;
338 
339  out:
340     if (fd >= 0) {
341         close(fd);
342     }
343     object_unref(OBJECT(sioc));
344     g_free(info.name);
345     kill(getpid(), SIGTERM);
346     return (void *) (intptr_t) ret;
347 }
348 #endif /* HAVE_NBD_DEVICE */
349 
350 static int nbd_can_accept(void)
351 {
352     return state == RUNNING && (shared == 0 || nb_fds < shared);
353 }
354 
355 static void nbd_update_server_watch(void);
356 
357 static void nbd_client_closed(NBDClient *client, bool negotiated)
358 {
359     nb_fds--;
360     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
361         state = TERMINATE;
362     }
363     nbd_update_server_watch();
364     nbd_client_put(client);
365 }
366 
367 static void nbd_accept(QIONetListener *listener, QIOChannelSocket *cioc,
368                        gpointer opaque)
369 {
370     if (state >= TERMINATE) {
371         return;
372     }
373 
374     nb_fds++;
375     nbd_update_server_watch();
376     nbd_client_new(cioc, tlscreds, tlsauthz, nbd_client_closed);
377 }
378 
379 static void nbd_update_server_watch(void)
380 {
381     if (nbd_can_accept()) {
382         qio_net_listener_set_client_func(server, nbd_accept, NULL, NULL);
383     } else {
384         qio_net_listener_set_client_func(server, NULL, NULL, NULL);
385     }
386 }
387 
388 
389 static SocketAddress *nbd_build_socket_address(const char *sockpath,
390                                                const char *bindto,
391                                                const char *port)
392 {
393     SocketAddress *saddr;
394 
395     saddr = g_new0(SocketAddress, 1);
396     if (sockpath) {
397         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
398         saddr->u.q_unix.path = g_strdup(sockpath);
399     } else {
400         InetSocketAddress *inet;
401         saddr->type = SOCKET_ADDRESS_TYPE_INET;
402         inet = &saddr->u.inet;
403         inet->host = g_strdup(bindto);
404         if (port) {
405             inet->port = g_strdup(port);
406         } else  {
407             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
408         }
409     }
410 
411     return saddr;
412 }
413 
414 
415 static QemuOptsList file_opts = {
416     .name = "file",
417     .implied_opt_name = "file",
418     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
419     .desc = {
420         /* no elements => accept any params */
421         { /* end of list */ }
422     },
423 };
424 
425 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, bool list,
426                                           Error **errp)
427 {
428     Object *obj;
429     QCryptoTLSCreds *creds;
430 
431     obj = object_resolve_path_component(
432         object_get_objects_root(), id);
433     if (!obj) {
434         error_setg(errp, "No TLS credentials with id '%s'",
435                    id);
436         return NULL;
437     }
438     creds = (QCryptoTLSCreds *)
439         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
440     if (!creds) {
441         error_setg(errp, "Object with id '%s' is not TLS credentials",
442                    id);
443         return NULL;
444     }
445 
446     if (!qcrypto_tls_creds_check_endpoint(creds,
447                                           list
448                                           ? QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
449                                           : QCRYPTO_TLS_CREDS_ENDPOINT_SERVER,
450                                           errp)) {
451         return NULL;
452     }
453     object_ref(obj);
454     return creds;
455 }
456 
457 static void setup_address_and_port(const char **address, const char **port)
458 {
459     if (*address == NULL) {
460         *address = "0.0.0.0";
461     }
462 
463     if (*port == NULL) {
464         *port = stringify(NBD_DEFAULT_PORT);
465     }
466 }
467 
468 /*
469  * Check socket parameters compatibility when socket activation is used.
470  */
471 static const char *socket_activation_validate_opts(const char *device,
472                                                    const char *sockpath,
473                                                    const char *address,
474                                                    const char *port,
475                                                    const char *selinux,
476                                                    bool list)
477 {
478     if (device != NULL) {
479         return "NBD device can't be set when using socket activation";
480     }
481 
482     if (sockpath != NULL) {
483         return "Unix socket can't be set when using socket activation";
484     }
485 
486     if (address != NULL) {
487         return "The interface can't be set when using socket activation";
488     }
489 
490     if (port != NULL) {
491         return "TCP port number can't be set when using socket activation";
492     }
493 
494     if (selinux != NULL) {
495         return "SELinux label can't be set when using socket activation";
496     }
497 
498     if (list) {
499         return "List mode is incompatible with socket activation";
500     }
501 
502     return NULL;
503 }
504 
505 static void qemu_nbd_shutdown(void)
506 {
507     job_cancel_sync_all();
508     blk_exp_close_all();
509     bdrv_close_all();
510 }
511 
512 int main(int argc, char **argv)
513 {
514     BlockBackend *blk;
515     BlockDriverState *bs;
516     uint64_t dev_offset = 0;
517     bool readonly = false;
518     bool disconnect = false;
519     const char *bindto = NULL;
520     const char *port = NULL;
521     char *sockpath = NULL;
522     char *device = NULL;
523     QemuOpts *sn_opts = NULL;
524     const char *sn_id_or_name = NULL;
525     const char *sopt = "hVb:o:p:rsnc:dvk:e:f:tl:x:T:D:AB:L";
526     struct option lopt[] = {
527         { "help", no_argument, NULL, 'h' },
528         { "version", no_argument, NULL, 'V' },
529         { "bind", required_argument, NULL, 'b' },
530         { "port", required_argument, NULL, 'p' },
531         { "socket", required_argument, NULL, 'k' },
532         { "offset", required_argument, NULL, 'o' },
533         { "read-only", no_argument, NULL, 'r' },
534         { "allocation-depth", no_argument, NULL, 'A' },
535         { "bitmap", required_argument, NULL, 'B' },
536         { "connect", required_argument, NULL, 'c' },
537         { "disconnect", no_argument, NULL, 'd' },
538         { "list", no_argument, NULL, 'L' },
539         { "snapshot", no_argument, NULL, 's' },
540         { "load-snapshot", required_argument, NULL, 'l' },
541         { "nocache", no_argument, NULL, 'n' },
542         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
543         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
544         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
545         { "detect-zeroes", required_argument, NULL,
546           QEMU_NBD_OPT_DETECT_ZEROES },
547         { "shared", required_argument, NULL, 'e' },
548         { "format", required_argument, NULL, 'f' },
549         { "persistent", no_argument, NULL, 't' },
550         { "verbose", no_argument, NULL, 'v' },
551         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
552         { "export-name", required_argument, NULL, 'x' },
553         { "description", required_argument, NULL, 'D' },
554         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
555         { "tls-hostname", required_argument, NULL, QEMU_NBD_OPT_TLSHOSTNAME },
556         { "tls-authz", required_argument, NULL, QEMU_NBD_OPT_TLSAUTHZ },
557         { "image-opts", no_argument, NULL, QEMU_NBD_OPT_IMAGE_OPTS },
558         { "trace", required_argument, NULL, 'T' },
559         { "fork", no_argument, NULL, QEMU_NBD_OPT_FORK },
560         { "pid-file", required_argument, NULL, QEMU_NBD_OPT_PID_FILE },
561         { "selinux-label", required_argument, NULL,
562           QEMU_NBD_OPT_SELINUX_LABEL },
563         { NULL, 0, NULL, 0 }
564     };
565     int ch;
566     int opt_ind = 0;
567     int flags = BDRV_O_RDWR;
568     int ret = 0;
569     bool seen_cache = false;
570     bool seen_discard = false;
571     bool seen_aio = false;
572     pthread_t client_thread;
573     const char *fmt = NULL;
574     Error *local_err = NULL;
575     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
576     QDict *options = NULL;
577     const char *export_name = NULL; /* defaults to "" later for server mode */
578     const char *export_description = NULL;
579     BlockDirtyBitmapOrStrList *bitmaps = NULL;
580     bool alloc_depth = false;
581     const char *tlscredsid = NULL;
582     const char *tlshostname = NULL;
583     bool imageOpts = false;
584     bool writethrough = false; /* Client will flush as needed. */
585     bool verbose = false;
586     bool fork_process = false;
587     bool list = false;
588     unsigned socket_activation;
589     const char *pid_file_name = NULL;
590     const char *selinux_label = NULL;
591     BlockExportOptions *export_opts;
592 #if HAVE_NBD_DEVICE
593     struct NbdClientOpts opts;
594 #endif
595 
596 #ifdef CONFIG_POSIX
597     os_setup_early_signal_handling();
598     os_setup_signal_handling();
599 #endif
600 
601     socket_init();
602     error_init(argv[0]);
603     module_call_init(MODULE_INIT_TRACE);
604     qcrypto_init(&error_fatal);
605 
606     module_call_init(MODULE_INIT_QOM);
607     qemu_add_opts(&qemu_trace_opts);
608     qemu_init_exec_dir(argv[0]);
609 
610     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
611         switch (ch) {
612         case 's':
613             flags |= BDRV_O_SNAPSHOT;
614             break;
615         case 'n':
616             optarg = (char *) "none";
617             /* fallthrough */
618         case QEMU_NBD_OPT_CACHE:
619             if (seen_cache) {
620                 error_report("-n and --cache can only be specified once");
621                 exit(EXIT_FAILURE);
622             }
623             seen_cache = true;
624             if (bdrv_parse_cache_mode(optarg, &flags, &writethrough) == -1) {
625                 error_report("Invalid cache mode `%s'", optarg);
626                 exit(EXIT_FAILURE);
627             }
628             break;
629         case QEMU_NBD_OPT_AIO:
630             if (seen_aio) {
631                 error_report("--aio can only be specified once");
632                 exit(EXIT_FAILURE);
633             }
634             seen_aio = true;
635             if (bdrv_parse_aio(optarg, &flags) < 0) {
636                 error_report("Invalid aio mode '%s'", optarg);
637                 exit(EXIT_FAILURE);
638             }
639             break;
640         case QEMU_NBD_OPT_DISCARD:
641             if (seen_discard) {
642                 error_report("--discard can only be specified once");
643                 exit(EXIT_FAILURE);
644             }
645             seen_discard = true;
646             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
647                 error_report("Invalid discard mode `%s'", optarg);
648                 exit(EXIT_FAILURE);
649             }
650             break;
651         case QEMU_NBD_OPT_DETECT_ZEROES:
652             detect_zeroes =
653                 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
654                                 optarg,
655                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
656                                 &local_err);
657             if (local_err) {
658                 error_reportf_err(local_err,
659                                   "Failed to parse detect_zeroes mode: ");
660                 exit(EXIT_FAILURE);
661             }
662             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
663                 !(flags & BDRV_O_UNMAP)) {
664                 error_report("setting detect-zeroes to unmap is not allowed "
665                              "without setting discard operation to unmap");
666                 exit(EXIT_FAILURE);
667             }
668             break;
669         case 'b':
670             bindto = optarg;
671             break;
672         case 'p':
673             port = optarg;
674             break;
675         case 'o':
676             if (qemu_strtou64(optarg, NULL, 0, &dev_offset) < 0) {
677                 error_report("Invalid offset '%s'", optarg);
678                 exit(EXIT_FAILURE);
679             }
680             break;
681         case 'l':
682             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
683                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
684                                                   optarg, false);
685                 if (!sn_opts) {
686                     error_report("Failed in parsing snapshot param `%s'",
687                                  optarg);
688                     exit(EXIT_FAILURE);
689                 }
690             } else {
691                 sn_id_or_name = optarg;
692             }
693             /* fall through */
694         case 'r':
695             readonly = true;
696             flags &= ~BDRV_O_RDWR;
697             break;
698         case 'A':
699             alloc_depth = true;
700             break;
701         case 'B':
702             {
703                 BlockDirtyBitmapOrStr *el = g_new(BlockDirtyBitmapOrStr, 1);
704                 *el = (BlockDirtyBitmapOrStr) {
705                     .type = QTYPE_QSTRING,
706                     .u.local = g_strdup(optarg),
707                 };
708                 QAPI_LIST_PREPEND(bitmaps, el);
709             }
710             break;
711         case 'k':
712             sockpath = optarg;
713             if (sockpath[0] != '/') {
714                 error_report("socket path must be absolute");
715                 exit(EXIT_FAILURE);
716             }
717             break;
718         case 'd':
719             disconnect = true;
720             break;
721         case 'c':
722             device = optarg;
723             break;
724         case 'e':
725             if (qemu_strtoi(optarg, NULL, 0, &shared) < 0 ||
726                 shared < 0) {
727                 error_report("Invalid shared device number '%s'", optarg);
728                 exit(EXIT_FAILURE);
729             }
730             break;
731         case 'f':
732             fmt = optarg;
733             break;
734         case 't':
735             persistent = 1;
736             break;
737         case 'x':
738             export_name = optarg;
739             if (strlen(export_name) > NBD_MAX_STRING_SIZE) {
740                 error_report("export name '%s' too long", export_name);
741                 exit(EXIT_FAILURE);
742             }
743             break;
744         case 'D':
745             export_description = optarg;
746             if (strlen(export_description) > NBD_MAX_STRING_SIZE) {
747                 error_report("export description '%s' too long",
748                              export_description);
749                 exit(EXIT_FAILURE);
750             }
751             break;
752         case 'v':
753             verbose = true;
754             break;
755         case 'V':
756             version(argv[0]);
757             exit(0);
758             break;
759         case 'h':
760             usage(argv[0]);
761             exit(0);
762             break;
763         case '?':
764             error_report("Try `%s --help' for more information.", argv[0]);
765             exit(EXIT_FAILURE);
766         case QEMU_NBD_OPT_OBJECT:
767             user_creatable_process_cmdline(optarg);
768             break;
769         case QEMU_NBD_OPT_TLSCREDS:
770             tlscredsid = optarg;
771             break;
772         case QEMU_NBD_OPT_TLSHOSTNAME:
773             tlshostname = optarg;
774             break;
775         case QEMU_NBD_OPT_IMAGE_OPTS:
776             imageOpts = true;
777             break;
778         case 'T':
779             trace_opt_parse(optarg);
780             break;
781         case QEMU_NBD_OPT_TLSAUTHZ:
782             tlsauthz = optarg;
783             break;
784         case QEMU_NBD_OPT_FORK:
785             fork_process = true;
786             break;
787         case 'L':
788             list = true;
789             break;
790         case QEMU_NBD_OPT_PID_FILE:
791             pid_file_name = optarg;
792             break;
793         case QEMU_NBD_OPT_SELINUX_LABEL:
794             selinux_label = optarg;
795             break;
796         }
797     }
798 
799     if (list) {
800         if (argc != optind) {
801             error_report("List mode is incompatible with a file name");
802             exit(EXIT_FAILURE);
803         }
804         if (export_name || export_description || dev_offset ||
805             device || disconnect || fmt || sn_id_or_name || bitmaps ||
806             alloc_depth || seen_aio || seen_discard || seen_cache) {
807             error_report("List mode is incompatible with per-device settings");
808             exit(EXIT_FAILURE);
809         }
810         if (fork_process) {
811             error_report("List mode is incompatible with forking");
812             exit(EXIT_FAILURE);
813         }
814     } else if ((argc - optind) != 1) {
815         error_report("Invalid number of arguments");
816         error_printf("Try `%s --help' for more information.\n", argv[0]);
817         exit(EXIT_FAILURE);
818     } else if (!export_name) {
819         export_name = "";
820     }
821 
822     if (!trace_init_backends()) {
823         exit(1);
824     }
825     trace_init_file();
826     qemu_set_log(LOG_TRACE, &error_fatal);
827 
828     socket_activation = check_socket_activation();
829     if (socket_activation == 0) {
830         if (!sockpath) {
831             setup_address_and_port(&bindto, &port);
832         }
833     } else {
834         /* Using socket activation - check user didn't use -p etc. */
835         const char *err_msg = socket_activation_validate_opts(device, sockpath,
836                                                               bindto, port,
837                                                               selinux_label,
838                                                               list);
839         if (err_msg != NULL) {
840             error_report("%s", err_msg);
841             exit(EXIT_FAILURE);
842         }
843 
844         /* qemu-nbd can only listen on a single socket.  */
845         if (socket_activation > 1) {
846             error_report("qemu-nbd does not support socket activation with %s > 1",
847                          "LISTEN_FDS");
848             exit(EXIT_FAILURE);
849         }
850     }
851 
852     if (tlscredsid) {
853         if (device) {
854             error_report("TLS is not supported with a host device");
855             exit(EXIT_FAILURE);
856         }
857         if (tlsauthz && list) {
858             error_report("TLS authorization is incompatible with export list");
859             exit(EXIT_FAILURE);
860         }
861         if (tlshostname && !list) {
862             error_report("TLS hostname is only supported with export list");
863             exit(EXIT_FAILURE);
864         }
865         tlscreds = nbd_get_tls_creds(tlscredsid, list, &local_err);
866         if (local_err) {
867             error_reportf_err(local_err, "Failed to get TLS creds: ");
868             exit(EXIT_FAILURE);
869         }
870     } else {
871         if (tlsauthz) {
872             error_report("--tls-authz is not permitted without --tls-creds");
873             exit(EXIT_FAILURE);
874         }
875         if (tlshostname) {
876             error_report("--tls-hostname is not permitted without --tls-creds");
877             exit(EXIT_FAILURE);
878         }
879     }
880 
881     if (selinux_label) {
882 #ifdef CONFIG_SELINUX
883         if (sockpath == NULL && device == NULL) {
884             error_report("--selinux-label is not permitted without --socket");
885             exit(EXIT_FAILURE);
886         }
887 #else
888         error_report("SELinux support not enabled in this binary");
889         exit(EXIT_FAILURE);
890 #endif
891     }
892 
893     if (list) {
894         saddr = nbd_build_socket_address(sockpath, bindto, port);
895         return qemu_nbd_client_list(saddr, tlscreds,
896                                     tlshostname ? tlshostname : bindto);
897     }
898 
899 #if !HAVE_NBD_DEVICE
900     if (disconnect || device) {
901         error_report("Kernel /dev/nbdN support not available");
902         exit(EXIT_FAILURE);
903     }
904 #else /* HAVE_NBD_DEVICE */
905     if (disconnect) {
906         int nbdfd = open(argv[optind], O_RDWR);
907         if (nbdfd < 0) {
908             error_report("Cannot open %s: %s", argv[optind],
909                          strerror(errno));
910             exit(EXIT_FAILURE);
911         }
912         nbd_disconnect(nbdfd);
913 
914         close(nbdfd);
915 
916         printf("%s disconnected\n", argv[optind]);
917 
918         return 0;
919     }
920 #endif
921 
922     if ((device && !verbose) || fork_process) {
923 #ifndef WIN32
924         g_autoptr(GError) err = NULL;
925         int stderr_fd[2];
926         pid_t pid;
927         int ret;
928 
929         if (!g_unix_open_pipe(stderr_fd, FD_CLOEXEC, &err)) {
930             error_report("Error setting up communication pipe: %s",
931                          err->message);
932             exit(EXIT_FAILURE);
933         }
934 
935         /* Now daemonize, but keep a communication channel open to
936          * print errors and exit with the proper status code.
937          */
938         pid = fork();
939         if (pid < 0) {
940             error_report("Failed to fork: %s", strerror(errno));
941             exit(EXIT_FAILURE);
942         } else if (pid == 0) {
943             int saved_errno;
944 
945             close(stderr_fd[0]);
946 
947             ret = qemu_daemon(1, 0);
948             saved_errno = errno;    /* dup2 will overwrite error below */
949 
950             /* Temporarily redirect stderr to the parent's pipe...  */
951             if (dup2(stderr_fd[1], STDERR_FILENO) < 0) {
952                 char str[256];
953                 snprintf(str, sizeof(str),
954                          "%s: Failed to link stderr to the pipe: %s\n",
955                          g_get_prgname(), strerror(errno));
956                 /*
957                  * We are unable to use error_report() here as we need to get
958                  * stderr pointed to the parent's pipe. Write to that pipe
959                  * manually.
960                  */
961                 ret = write(stderr_fd[1], str, strlen(str));
962                 exit(EXIT_FAILURE);
963             }
964 
965             if (ret < 0) {
966                 error_report("Failed to daemonize: %s", strerror(saved_errno));
967                 exit(EXIT_FAILURE);
968             }
969 
970             /* ... close the descriptor we inherited and go on.  */
971             close(stderr_fd[1]);
972         } else {
973             bool errors = false;
974             char *buf;
975 
976             /* In the parent.  Print error messages from the child until
977              * it closes the pipe.
978              */
979             close(stderr_fd[1]);
980             buf = g_malloc(1024);
981             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
982                 errors = true;
983                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
984                 if (ret < 0) {
985                     exit(EXIT_FAILURE);
986                 }
987             }
988             if (ret < 0) {
989                 error_report("Cannot read from daemon: %s",
990                              strerror(errno));
991                 exit(EXIT_FAILURE);
992             }
993 
994             /* Usually the daemon should not print any message.
995              * Exit with zero status in that case.
996              */
997             exit(errors);
998         }
999 #else /* WIN32 */
1000         error_report("Unable to fork into background on Windows hosts");
1001         exit(EXIT_FAILURE);
1002 #endif /* WIN32 */
1003     }
1004 
1005     if (device != NULL && sockpath == NULL) {
1006         sockpath = g_malloc(128);
1007         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
1008     }
1009 
1010     server = qio_net_listener_new();
1011     if (socket_activation == 0) {
1012         int backlog;
1013 
1014         if (persistent || shared == 0) {
1015             backlog = SOMAXCONN;
1016         } else {
1017             backlog = MIN(shared, SOMAXCONN);
1018         }
1019 #ifdef CONFIG_SELINUX
1020         if (selinux_label && setsockcreatecon_raw(selinux_label) == -1) {
1021             error_report("Cannot set SELinux socket create context to %s: %s",
1022                          selinux_label, strerror(errno));
1023             exit(EXIT_FAILURE);
1024         }
1025 #endif
1026         saddr = nbd_build_socket_address(sockpath, bindto, port);
1027         if (qio_net_listener_open_sync(server, saddr, backlog,
1028                                        &local_err) < 0) {
1029             object_unref(OBJECT(server));
1030             error_report_err(local_err);
1031             exit(EXIT_FAILURE);
1032         }
1033 #ifdef CONFIG_SELINUX
1034         if (selinux_label && setsockcreatecon_raw(NULL) == -1) {
1035             error_report("Cannot clear SELinux socket create context: %s",
1036                          strerror(errno));
1037             exit(EXIT_FAILURE);
1038         }
1039 #endif
1040     } else {
1041         size_t i;
1042         /* See comment in check_socket_activation above. */
1043         for (i = 0; i < socket_activation; i++) {
1044             QIOChannelSocket *sioc;
1045             sioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD + i,
1046                                              &local_err);
1047             if (sioc == NULL) {
1048                 object_unref(OBJECT(server));
1049                 error_reportf_err(local_err,
1050                                   "Failed to use socket activation: ");
1051                 exit(EXIT_FAILURE);
1052             }
1053             qio_net_listener_add(server, sioc);
1054             object_unref(OBJECT(sioc));
1055         }
1056     }
1057 
1058     qemu_init_main_loop(&error_fatal);
1059     bdrv_init();
1060     atexit(qemu_nbd_shutdown);
1061 
1062     srcpath = argv[optind];
1063     if (imageOpts) {
1064         QemuOpts *opts;
1065         if (fmt) {
1066             error_report("--image-opts and -f are mutually exclusive");
1067             exit(EXIT_FAILURE);
1068         }
1069         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
1070         if (!opts) {
1071             qemu_opts_reset(&file_opts);
1072             exit(EXIT_FAILURE);
1073         }
1074         options = qemu_opts_to_qdict(opts, NULL);
1075         qemu_opts_reset(&file_opts);
1076         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
1077     } else {
1078         if (fmt) {
1079             options = qdict_new();
1080             qdict_put_str(options, "driver", fmt);
1081         }
1082         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
1083     }
1084 
1085     if (!blk) {
1086         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
1087                           argv[optind]);
1088         exit(EXIT_FAILURE);
1089     }
1090     bs = blk_bs(blk);
1091 
1092     if (dev_offset) {
1093         QDict *raw_opts = qdict_new();
1094         qdict_put_str(raw_opts, "driver", "raw");
1095         qdict_put_str(raw_opts, "file", bs->node_name);
1096         qdict_put_int(raw_opts, "offset", dev_offset);
1097 
1098         aio_context_acquire(qemu_get_aio_context());
1099         bs = bdrv_open(NULL, NULL, raw_opts, flags, &error_fatal);
1100         aio_context_release(qemu_get_aio_context());
1101 
1102         blk_remove_bs(blk);
1103         blk_insert_bs(blk, bs, &error_fatal);
1104         bdrv_unref(bs);
1105     }
1106 
1107     blk_set_enable_write_cache(blk, !writethrough);
1108 
1109     if (sn_opts) {
1110         ret = bdrv_snapshot_load_tmp(bs,
1111                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1112                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1113                                      &local_err);
1114     } else if (sn_id_or_name) {
1115         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
1116                                                    &local_err);
1117     }
1118     if (ret < 0) {
1119         error_reportf_err(local_err, "Failed to load snapshot: ");
1120         exit(EXIT_FAILURE);
1121     }
1122 
1123     bs->detect_zeroes = detect_zeroes;
1124 
1125     nbd_server_is_qemu_nbd(shared);
1126 
1127     export_opts = g_new(BlockExportOptions, 1);
1128     *export_opts = (BlockExportOptions) {
1129         .type               = BLOCK_EXPORT_TYPE_NBD,
1130         .id                 = g_strdup("qemu-nbd-export"),
1131         .node_name          = g_strdup(bdrv_get_node_name(bs)),
1132         .has_writethrough   = true,
1133         .writethrough       = writethrough,
1134         .has_writable       = true,
1135         .writable           = !readonly,
1136         .u.nbd = {
1137             .name                 = g_strdup(export_name),
1138             .description          = g_strdup(export_description),
1139             .has_bitmaps          = !!bitmaps,
1140             .bitmaps              = bitmaps,
1141             .has_allocation_depth = alloc_depth,
1142             .allocation_depth     = alloc_depth,
1143         },
1144     };
1145     blk_exp_add(export_opts, &error_fatal);
1146     qapi_free_BlockExportOptions(export_opts);
1147 
1148     if (device) {
1149 #if HAVE_NBD_DEVICE
1150         int ret;
1151         opts = (struct NbdClientOpts) {
1152             .device = device,
1153             .fork_process = fork_process,
1154             .verbose = verbose,
1155         };
1156 
1157         ret = pthread_create(&client_thread, NULL, nbd_client_thread, &opts);
1158         if (ret != 0) {
1159             error_report("Failed to create client thread: %s", strerror(ret));
1160             exit(EXIT_FAILURE);
1161         }
1162 #endif
1163     } else {
1164         /* Shut up GCC warnings.  */
1165         memset(&client_thread, 0, sizeof(client_thread));
1166     }
1167 
1168     nbd_update_server_watch();
1169 
1170     if (pid_file_name) {
1171         qemu_write_pidfile(pid_file_name, &error_fatal);
1172     }
1173 
1174     /* now when the initialization is (almost) complete, chdir("/")
1175      * to free any busy filesystems */
1176     if (chdir("/") < 0) {
1177         error_report("Could not chdir to root directory: %s",
1178                      strerror(errno));
1179         exit(EXIT_FAILURE);
1180     }
1181 
1182     if (fork_process) {
1183         if (dup2(STDOUT_FILENO, STDERR_FILENO) < 0) {
1184             error_report("Could not set stderr to /dev/null: %s",
1185                          strerror(errno));
1186             exit(EXIT_FAILURE);
1187         }
1188     }
1189 
1190     state = RUNNING;
1191     do {
1192         main_loop_wait(false);
1193         if (state == TERMINATE) {
1194             blk_exp_close_all();
1195             state = TERMINATED;
1196         }
1197     } while (state != TERMINATED);
1198 
1199     blk_unref(blk);
1200     if (sockpath) {
1201         unlink(sockpath);
1202     }
1203 
1204     qemu_opts_del(sn_opts);
1205 
1206     if (device) {
1207         void *ret;
1208         pthread_join(client_thread, &ret);
1209         exit(ret != NULL);
1210     } else {
1211         exit(EXIT_SUCCESS);
1212     }
1213 }
1214