xref: /openbmc/qemu/qemu-nbd.c (revision 80adf54e)
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 "qapi/error.h"
25 #include "qemu-common.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/error-report.h"
32 #include "qemu/config-file.h"
33 #include "qemu/bswap.h"
34 #include "qemu/log.h"
35 #include "qemu/systemd.h"
36 #include "block/snapshot.h"
37 #include "qapi/util.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qom/object_interfaces.h"
40 #include "io/channel-socket.h"
41 #include "crypto/init.h"
42 #include "trace/control.h"
43 #include "qemu-version.h"
44 
45 #define SOCKET_PATH                "/var/lock/qemu-nbd-%s"
46 #define QEMU_NBD_OPT_CACHE         256
47 #define QEMU_NBD_OPT_AIO           257
48 #define QEMU_NBD_OPT_DISCARD       258
49 #define QEMU_NBD_OPT_DETECT_ZEROES 259
50 #define QEMU_NBD_OPT_OBJECT        260
51 #define QEMU_NBD_OPT_TLSCREDS      261
52 #define QEMU_NBD_OPT_IMAGE_OPTS    262
53 #define QEMU_NBD_OPT_FORK          263
54 
55 #define MBR_SIZE 512
56 
57 static NBDExport *exp;
58 static bool newproto;
59 static int verbose;
60 static char *srcpath;
61 static SocketAddress *saddr;
62 static int persistent = 0;
63 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
64 static int shared = 1;
65 static int nb_fds;
66 static QIOChannelSocket *server_ioc;
67 static int server_watch = -1;
68 static QCryptoTLSCreds *tlscreds;
69 
70 static void usage(const char *name)
71 {
72     (printf) (
73 "Usage: %s [OPTIONS] FILE\n"
74 "QEMU Disk Network Block Device Server\n"
75 "\n"
76 "  -h, --help                display this help and exit\n"
77 "  -V, --version             output version information and exit\n"
78 "\n"
79 "Connection properties:\n"
80 "  -p, --port=PORT           port to listen on (default `%d')\n"
81 "  -b, --bind=IFACE          interface to bind to (default `0.0.0.0')\n"
82 "  -k, --socket=PATH         path to the unix socket\n"
83 "                            (default '"SOCKET_PATH"')\n"
84 "  -e, --shared=NUM          device can be shared by NUM clients (default '1')\n"
85 "  -t, --persistent          don't exit on the last connection\n"
86 "  -v, --verbose             display extra debugging information\n"
87 "  -x, --export-name=NAME    expose export by name\n"
88 "  -D, --description=TEXT    with -x, also export a human-readable description\n"
89 "\n"
90 "Exposing part of the image:\n"
91 "  -o, --offset=OFFSET       offset into the image\n"
92 "  -P, --partition=NUM       only expose partition NUM\n"
93 "\n"
94 "General purpose options:\n"
95 "  --object type,id=ID,...   define an object such as 'secret' for providing\n"
96 "                            passwords and/or encryption keys\n"
97 "  -T, --trace [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
98 "                            specify tracing options\n"
99 "  --fork                    fork off the server process and exit the parent\n"
100 "                            once the server is running\n"
101 #ifdef __linux__
102 "Kernel NBD client support:\n"
103 "  -c, --connect=DEV         connect FILE to the local NBD device DEV\n"
104 "  -d, --disconnect          disconnect the specified device\n"
105 "\n"
106 #endif
107 "\n"
108 "Block device options:\n"
109 "  -f, --format=FORMAT       set image format (raw, qcow2, ...)\n"
110 "  -r, --read-only           export read-only\n"
111 "  -s, --snapshot            use FILE as an external snapshot, create a temporary\n"
112 "                            file with backing_file=FILE, redirect the write to\n"
113 "                            the temporary one\n"
114 "  -l, --load-snapshot=SNAPSHOT_PARAM\n"
115 "                            load an internal snapshot inside FILE and export it\n"
116 "                            as an read-only device, SNAPSHOT_PARAM format is\n"
117 "                            'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
118 "                            '[ID_OR_NAME]'\n"
119 "  -n, --nocache             disable host cache\n"
120 "      --cache=MODE          set cache mode (none, writeback, ...)\n"
121 "      --aio=MODE            set AIO mode (native or threads)\n"
122 "      --discard=MODE        set discard mode (ignore, unmap)\n"
123 "      --detect-zeroes=MODE  set detect-zeroes mode (off, on, unmap)\n"
124 "      --image-opts          treat FILE as a full set of image options\n"
125 "\n"
126 QEMU_HELP_BOTTOM "\n"
127     , name, NBD_DEFAULT_PORT, "DEVICE");
128 }
129 
130 static void version(const char *name)
131 {
132     printf(
133 "%s " QEMU_VERSION QEMU_PKGVERSION "\n"
134 "Written by Anthony Liguori.\n"
135 "\n"
136 QEMU_COPYRIGHT "\n"
137 "This is free software; see the source for copying conditions.  There is NO\n"
138 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
139     , name);
140 }
141 
142 struct partition_record
143 {
144     uint8_t bootable;
145     uint8_t start_head;
146     uint32_t start_cylinder;
147     uint8_t start_sector;
148     uint8_t system;
149     uint8_t end_head;
150     uint8_t end_cylinder;
151     uint8_t end_sector;
152     uint32_t start_sector_abs;
153     uint32_t nb_sectors_abs;
154 };
155 
156 static void read_partition(uint8_t *p, struct partition_record *r)
157 {
158     r->bootable = p[0];
159     r->start_head = p[1];
160     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
161     r->start_sector = p[2] & 0x3f;
162     r->system = p[4];
163     r->end_head = p[5];
164     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
165     r->end_sector = p[6] & 0x3f;
166 
167     r->start_sector_abs = ldl_le_p(p + 8);
168     r->nb_sectors_abs   = ldl_le_p(p + 12);
169 }
170 
171 static int find_partition(BlockBackend *blk, int partition,
172                           off_t *offset, off_t *size)
173 {
174     struct partition_record mbr[4];
175     uint8_t data[MBR_SIZE];
176     int i;
177     int ext_partnum = 4;
178     int ret;
179 
180     ret = blk_pread(blk, 0, data, sizeof(data));
181     if (ret < 0) {
182         error_report("error while reading: %s", strerror(-ret));
183         exit(EXIT_FAILURE);
184     }
185 
186     if (data[510] != 0x55 || data[511] != 0xaa) {
187         return -EINVAL;
188     }
189 
190     for (i = 0; i < 4; i++) {
191         read_partition(&data[446 + 16 * i], &mbr[i]);
192 
193         if (!mbr[i].system || !mbr[i].nb_sectors_abs) {
194             continue;
195         }
196 
197         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
198             struct partition_record ext[4];
199             uint8_t data1[MBR_SIZE];
200             int j;
201 
202             ret = blk_pread(blk, mbr[i].start_sector_abs * MBR_SIZE,
203                             data1, sizeof(data1));
204             if (ret < 0) {
205                 error_report("error while reading: %s", strerror(-ret));
206                 exit(EXIT_FAILURE);
207             }
208 
209             for (j = 0; j < 4; j++) {
210                 read_partition(&data1[446 + 16 * j], &ext[j]);
211                 if (!ext[j].system || !ext[j].nb_sectors_abs) {
212                     continue;
213                 }
214 
215                 if ((ext_partnum + j + 1) == partition) {
216                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
217                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
218                     return 0;
219                 }
220             }
221             ext_partnum += 4;
222         } else if ((i + 1) == partition) {
223             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
224             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
225             return 0;
226         }
227     }
228 
229     return -ENOENT;
230 }
231 
232 static void termsig_handler(int signum)
233 {
234     atomic_cmpxchg(&state, RUNNING, TERMINATE);
235     qemu_notify_event();
236 }
237 
238 
239 static void *show_parts(void *arg)
240 {
241     char *device = arg;
242     int nbd;
243 
244     /* linux just needs an open() to trigger
245      * the partition table update
246      * but remember to load the module with max_part != 0 :
247      *     modprobe nbd max_part=63
248      */
249     nbd = open(device, O_RDWR);
250     if (nbd >= 0) {
251         close(nbd);
252     }
253     return NULL;
254 }
255 
256 static void *nbd_client_thread(void *arg)
257 {
258     char *device = arg;
259     NBDExportInfo info = { .request_sizes = false, };
260     QIOChannelSocket *sioc;
261     int fd;
262     int ret;
263     pthread_t show_parts_thread;
264     Error *local_error = NULL;
265 
266     sioc = qio_channel_socket_new();
267     if (qio_channel_socket_connect_sync(sioc,
268                                         saddr,
269                                         &local_error) < 0) {
270         error_report_err(local_error);
271         goto out;
272     }
273 
274     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), NULL,
275                                 NULL, NULL, NULL, &info, &local_error);
276     if (ret < 0) {
277         if (local_error) {
278             error_report_err(local_error);
279         }
280         goto out_socket;
281     }
282 
283     fd = open(device, O_RDWR);
284     if (fd < 0) {
285         /* Linux-only, we can use %m in printf.  */
286         error_report("Failed to open %s: %m", device);
287         goto out_socket;
288     }
289 
290     ret = nbd_init(fd, sioc, &info, &local_error);
291     if (ret < 0) {
292         error_report_err(local_error);
293         goto out_fd;
294     }
295 
296     /* update partition table */
297     pthread_create(&show_parts_thread, NULL, show_parts, device);
298 
299     if (verbose) {
300         fprintf(stderr, "NBD device %s is now connected to %s\n",
301                 device, srcpath);
302     } else {
303         /* Close stderr so that the qemu-nbd process exits.  */
304         dup2(STDOUT_FILENO, STDERR_FILENO);
305     }
306 
307     ret = nbd_client(fd);
308     if (ret) {
309         goto out_fd;
310     }
311     close(fd);
312     object_unref(OBJECT(sioc));
313     kill(getpid(), SIGTERM);
314     return (void *) EXIT_SUCCESS;
315 
316 out_fd:
317     close(fd);
318 out_socket:
319     object_unref(OBJECT(sioc));
320 out:
321     kill(getpid(), SIGTERM);
322     return (void *) EXIT_FAILURE;
323 }
324 
325 static int nbd_can_accept(void)
326 {
327     return state == RUNNING && nb_fds < shared;
328 }
329 
330 static void nbd_export_closed(NBDExport *exp)
331 {
332     assert(state == TERMINATING);
333     state = TERMINATED;
334 }
335 
336 static void nbd_update_server_watch(void);
337 
338 static void nbd_client_closed(NBDClient *client, bool negotiated)
339 {
340     nb_fds--;
341     if (negotiated && nb_fds == 0 && !persistent && state == RUNNING) {
342         state = TERMINATE;
343     }
344     nbd_update_server_watch();
345     nbd_client_put(client);
346 }
347 
348 static gboolean nbd_accept(QIOChannel *ioc, GIOCondition cond, gpointer opaque)
349 {
350     QIOChannelSocket *cioc;
351 
352     cioc = qio_channel_socket_accept(QIO_CHANNEL_SOCKET(ioc),
353                                      NULL);
354     if (!cioc) {
355         return TRUE;
356     }
357 
358     if (state >= TERMINATE) {
359         object_unref(OBJECT(cioc));
360         return TRUE;
361     }
362 
363     nb_fds++;
364     nbd_update_server_watch();
365     nbd_client_new(newproto ? NULL : exp, cioc,
366                    tlscreds, NULL, nbd_client_closed);
367     object_unref(OBJECT(cioc));
368 
369     return TRUE;
370 }
371 
372 static void nbd_update_server_watch(void)
373 {
374     if (nbd_can_accept()) {
375         if (server_watch == -1) {
376             server_watch = qio_channel_add_watch(QIO_CHANNEL(server_ioc),
377                                                  G_IO_IN,
378                                                  nbd_accept,
379                                                  NULL, NULL);
380         }
381     } else {
382         if (server_watch != -1) {
383             g_source_remove(server_watch);
384             server_watch = -1;
385         }
386     }
387 }
388 
389 
390 static SocketAddress *nbd_build_socket_address(const char *sockpath,
391                                                const char *bindto,
392                                                const char *port)
393 {
394     SocketAddress *saddr;
395 
396     saddr = g_new0(SocketAddress, 1);
397     if (sockpath) {
398         saddr->type = SOCKET_ADDRESS_TYPE_UNIX;
399         saddr->u.q_unix.path = g_strdup(sockpath);
400     } else {
401         InetSocketAddress *inet;
402         saddr->type = SOCKET_ADDRESS_TYPE_INET;
403         inet = &saddr->u.inet;
404         inet->host = g_strdup(bindto);
405         if (port) {
406             inet->port = g_strdup(port);
407         } else  {
408             inet->port = g_strdup_printf("%d", NBD_DEFAULT_PORT);
409         }
410     }
411 
412     return saddr;
413 }
414 
415 
416 static QemuOptsList file_opts = {
417     .name = "file",
418     .implied_opt_name = "file",
419     .head = QTAILQ_HEAD_INITIALIZER(file_opts.head),
420     .desc = {
421         /* no elements => accept any params */
422         { /* end of list */ }
423     },
424 };
425 
426 static QemuOptsList qemu_object_opts = {
427     .name = "object",
428     .implied_opt_name = "qom-type",
429     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
430     .desc = {
431         { }
432     },
433 };
434 
435 
436 
437 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
438 {
439     Object *obj;
440     QCryptoTLSCreds *creds;
441 
442     obj = object_resolve_path_component(
443         object_get_objects_root(), id);
444     if (!obj) {
445         error_setg(errp, "No TLS credentials with id '%s'",
446                    id);
447         return NULL;
448     }
449     creds = (QCryptoTLSCreds *)
450         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
451     if (!creds) {
452         error_setg(errp, "Object with id '%s' is not TLS credentials",
453                    id);
454         return NULL;
455     }
456 
457     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_SERVER) {
458         error_setg(errp,
459                    "Expecting TLS credentials with a server endpoint");
460         return NULL;
461     }
462     object_ref(obj);
463     return creds;
464 }
465 
466 static void setup_address_and_port(const char **address, const char **port)
467 {
468     if (*address == NULL) {
469         *address = "0.0.0.0";
470     }
471 
472     if (*port == NULL) {
473         *port = stringify(NBD_DEFAULT_PORT);
474     }
475 }
476 
477 /*
478  * Check socket parameters compatibility when socket activation is used.
479  */
480 static const char *socket_activation_validate_opts(const char *device,
481                                                    const char *sockpath,
482                                                    const char *address,
483                                                    const char *port)
484 {
485     if (device != NULL) {
486         return "NBD device can't be set when using socket activation";
487     }
488 
489     if (sockpath != NULL) {
490         return "Unix socket can't be set when using socket activation";
491     }
492 
493     if (address != NULL) {
494         return "The interface can't be set when using socket activation";
495     }
496 
497     if (port != NULL) {
498         return "TCP port number can't be set when using socket activation";
499     }
500 
501     return NULL;
502 }
503 
504 int main(int argc, char **argv)
505 {
506     BlockBackend *blk;
507     BlockDriverState *bs;
508     off_t dev_offset = 0;
509     uint16_t nbdflags = 0;
510     bool disconnect = false;
511     const char *bindto = NULL;
512     const char *port = NULL;
513     char *sockpath = NULL;
514     char *device = NULL;
515     off_t fd_size;
516     QemuOpts *sn_opts = NULL;
517     const char *sn_id_or_name = NULL;
518     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:tl:x:T:D:";
519     struct option lopt[] = {
520         { "help", no_argument, NULL, 'h' },
521         { "version", no_argument, NULL, 'V' },
522         { "bind", required_argument, NULL, 'b' },
523         { "port", required_argument, NULL, 'p' },
524         { "socket", required_argument, NULL, 'k' },
525         { "offset", required_argument, NULL, 'o' },
526         { "read-only", no_argument, NULL, 'r' },
527         { "partition", required_argument, NULL, 'P' },
528         { "connect", required_argument, NULL, 'c' },
529         { "disconnect", no_argument, NULL, 'd' },
530         { "snapshot", no_argument, NULL, 's' },
531         { "load-snapshot", required_argument, NULL, 'l' },
532         { "nocache", no_argument, NULL, 'n' },
533         { "cache", required_argument, NULL, QEMU_NBD_OPT_CACHE },
534         { "aio", required_argument, NULL, QEMU_NBD_OPT_AIO },
535         { "discard", required_argument, NULL, QEMU_NBD_OPT_DISCARD },
536         { "detect-zeroes", required_argument, NULL,
537           QEMU_NBD_OPT_DETECT_ZEROES },
538         { "shared", required_argument, NULL, 'e' },
539         { "format", required_argument, NULL, 'f' },
540         { "persistent", no_argument, NULL, 't' },
541         { "verbose", no_argument, NULL, 'v' },
542         { "object", required_argument, NULL, QEMU_NBD_OPT_OBJECT },
543         { "export-name", required_argument, NULL, 'x' },
544         { "description", required_argument, NULL, 'D' },
545         { "tls-creds", required_argument, NULL, QEMU_NBD_OPT_TLSCREDS },
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         { NULL, 0, NULL, 0 }
550     };
551     int ch;
552     int opt_ind = 0;
553     char *end;
554     int flags = BDRV_O_RDWR;
555     int partition = -1;
556     int ret = 0;
557     bool seen_cache = false;
558     bool seen_discard = false;
559     bool seen_aio = false;
560     pthread_t client_thread;
561     const char *fmt = NULL;
562     Error *local_err = NULL;
563     BlockdevDetectZeroesOptions detect_zeroes = BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
564     QDict *options = NULL;
565     const char *export_name = NULL;
566     const char *export_description = NULL;
567     const char *tlscredsid = NULL;
568     bool imageOpts = false;
569     bool writethrough = true;
570     char *trace_file = NULL;
571     bool fork_process = false;
572     int old_stderr = -1;
573     unsigned socket_activation;
574 
575     /* The client thread uses SIGTERM to interrupt the server.  A signal
576      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
577      */
578     struct sigaction sa_sigterm;
579     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
580     sa_sigterm.sa_handler = termsig_handler;
581     sigaction(SIGTERM, &sa_sigterm, NULL);
582 
583 #ifdef CONFIG_POSIX
584     signal(SIGPIPE, SIG_IGN);
585 #endif
586 
587     module_call_init(MODULE_INIT_TRACE);
588     qcrypto_init(&error_fatal);
589 
590     module_call_init(MODULE_INIT_QOM);
591     qemu_add_opts(&qemu_object_opts);
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 (!strcmp(optarg, "native")) {
621                 flags |= BDRV_O_NATIVE_AIO;
622             } else if (!strcmp(optarg, "threads")) {
623                 /* this is the default */
624             } else {
625                error_report("invalid aio mode `%s'", optarg);
626                exit(EXIT_FAILURE);
627             }
628             break;
629         case QEMU_NBD_OPT_DISCARD:
630             if (seen_discard) {
631                 error_report("--discard can only be specified once");
632                 exit(EXIT_FAILURE);
633             }
634             seen_discard = true;
635             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
636                 error_report("Invalid discard mode `%s'", optarg);
637                 exit(EXIT_FAILURE);
638             }
639             break;
640         case QEMU_NBD_OPT_DETECT_ZEROES:
641             detect_zeroes =
642                 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
643                                 optarg,
644                                 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
645                                 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
646                                 &local_err);
647             if (local_err) {
648                 error_reportf_err(local_err,
649                                   "Failed to parse detect_zeroes mode: ");
650                 exit(EXIT_FAILURE);
651             }
652             if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
653                 !(flags & BDRV_O_UNMAP)) {
654                 error_report("setting detect-zeroes to unmap is not allowed "
655                              "without setting discard operation to unmap");
656                 exit(EXIT_FAILURE);
657             }
658             break;
659         case 'b':
660             bindto = optarg;
661             break;
662         case 'p':
663             port = optarg;
664             break;
665         case 'o':
666                 dev_offset = strtoll (optarg, &end, 0);
667             if (*end) {
668                 error_report("Invalid offset `%s'", optarg);
669                 exit(EXIT_FAILURE);
670             }
671             if (dev_offset < 0) {
672                 error_report("Offset must be positive `%s'", optarg);
673                 exit(EXIT_FAILURE);
674             }
675             break;
676         case 'l':
677             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
678                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
679                                                   optarg, false);
680                 if (!sn_opts) {
681                     error_report("Failed in parsing snapshot param `%s'",
682                                  optarg);
683                     exit(EXIT_FAILURE);
684                 }
685             } else {
686                 sn_id_or_name = optarg;
687             }
688             /* fall through */
689         case 'r':
690             nbdflags |= NBD_FLAG_READ_ONLY;
691             flags &= ~BDRV_O_RDWR;
692             break;
693         case 'P':
694             partition = strtol(optarg, &end, 0);
695             if (*end) {
696                 error_report("Invalid partition `%s'", optarg);
697                 exit(EXIT_FAILURE);
698             }
699             if (partition < 1 || partition > 8) {
700                 error_report("Invalid partition %d", partition);
701                 exit(EXIT_FAILURE);
702             }
703             break;
704         case 'k':
705             sockpath = optarg;
706             if (sockpath[0] != '/') {
707                 error_report("socket path must be absolute");
708                 exit(EXIT_FAILURE);
709             }
710             break;
711         case 'd':
712             disconnect = true;
713             break;
714         case 'c':
715             device = optarg;
716             break;
717         case 'e':
718             shared = strtol(optarg, &end, 0);
719             if (*end) {
720                 error_report("Invalid shared device number '%s'", optarg);
721                 exit(EXIT_FAILURE);
722             }
723             if (shared < 1) {
724                 error_report("Shared device number must be greater than 0");
725                 exit(EXIT_FAILURE);
726             }
727             break;
728         case 'f':
729             fmt = optarg;
730             break;
731         case 't':
732             persistent = 1;
733             break;
734         case 'x':
735             export_name = optarg;
736             break;
737         case 'D':
738             export_description = optarg;
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_FORK:
773             fork_process = true;
774             break;
775         }
776     }
777 
778     if ((argc - optind) != 1) {
779         error_report("Invalid number of arguments");
780         error_printf("Try `%s --help' for more information.\n", argv[0]);
781         exit(EXIT_FAILURE);
782     }
783 
784     if (qemu_opts_foreach(&qemu_object_opts,
785                           user_creatable_add_opts_foreach,
786                           NULL, NULL)) {
787         exit(EXIT_FAILURE);
788     }
789 
790     if (!trace_init_backends()) {
791         exit(1);
792     }
793     trace_init_file(trace_file);
794     qemu_set_log(LOG_TRACE);
795 
796     socket_activation = check_socket_activation();
797     if (socket_activation == 0) {
798         setup_address_and_port(&bindto, &port);
799     } else {
800         /* Using socket activation - check user didn't use -p etc. */
801         const char *err_msg = socket_activation_validate_opts(device, sockpath,
802                                                               bindto, port);
803         if (err_msg != NULL) {
804             error_report("%s", err_msg);
805             exit(EXIT_FAILURE);
806         }
807 
808         /* qemu-nbd can only listen on a single socket.  */
809         if (socket_activation > 1) {
810             error_report("qemu-nbd does not support socket activation with %s > 1",
811                          "LISTEN_FDS");
812             exit(EXIT_FAILURE);
813         }
814     }
815 
816     if (tlscredsid) {
817         if (sockpath) {
818             error_report("TLS is only supported with IPv4/IPv6");
819             exit(EXIT_FAILURE);
820         }
821         if (device) {
822             error_report("TLS is not supported with a host device");
823             exit(EXIT_FAILURE);
824         }
825         if (!export_name) {
826             /* Set the default NBD protocol export name, since
827              * we *must* use new style protocol for TLS */
828             export_name = "";
829         }
830         tlscreds = nbd_get_tls_creds(tlscredsid, &local_err);
831         if (local_err) {
832             error_report("Failed to get TLS creds %s",
833                          error_get_pretty(local_err));
834             exit(EXIT_FAILURE);
835         }
836     }
837 
838     if (disconnect) {
839         int nbdfd = open(argv[optind], O_RDWR);
840         if (nbdfd < 0) {
841             error_report("Cannot open %s: %s", argv[optind],
842                          strerror(errno));
843             exit(EXIT_FAILURE);
844         }
845         nbd_disconnect(nbdfd);
846 
847         close(nbdfd);
848 
849         printf("%s disconnected\n", argv[optind]);
850 
851         return 0;
852     }
853 
854     if ((device && !verbose) || fork_process) {
855         int stderr_fd[2];
856         pid_t pid;
857         int ret;
858 
859         if (qemu_pipe(stderr_fd) < 0) {
860             error_report("Error setting up communication pipe: %s",
861                          strerror(errno));
862             exit(EXIT_FAILURE);
863         }
864 
865         /* Now daemonize, but keep a communication channel open to
866          * print errors and exit with the proper status code.
867          */
868         pid = fork();
869         if (pid < 0) {
870             error_report("Failed to fork: %s", strerror(errno));
871             exit(EXIT_FAILURE);
872         } else if (pid == 0) {
873             close(stderr_fd[0]);
874             ret = qemu_daemon(1, 0);
875 
876             /* Temporarily redirect stderr to the parent's pipe...  */
877             old_stderr = dup(STDERR_FILENO);
878             dup2(stderr_fd[1], STDERR_FILENO);
879             if (ret < 0) {
880                 error_report("Failed to daemonize: %s", strerror(errno));
881                 exit(EXIT_FAILURE);
882             }
883 
884             /* ... close the descriptor we inherited and go on.  */
885             close(stderr_fd[1]);
886         } else {
887             bool errors = false;
888             char *buf;
889 
890             /* In the parent.  Print error messages from the child until
891              * it closes the pipe.
892              */
893             close(stderr_fd[1]);
894             buf = g_malloc(1024);
895             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
896                 errors = true;
897                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
898                 if (ret < 0) {
899                     exit(EXIT_FAILURE);
900                 }
901             }
902             if (ret < 0) {
903                 error_report("Cannot read from daemon: %s",
904                              strerror(errno));
905                 exit(EXIT_FAILURE);
906             }
907 
908             /* Usually the daemon should not print any message.
909              * Exit with zero status in that case.
910              */
911             exit(errors);
912         }
913     }
914 
915     if (device != NULL && sockpath == NULL) {
916         sockpath = g_malloc(128);
917         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
918     }
919 
920     if (socket_activation == 0) {
921         server_ioc = qio_channel_socket_new();
922         saddr = nbd_build_socket_address(sockpath, bindto, port);
923         if (qio_channel_socket_listen_sync(server_ioc, saddr, &local_err) < 0) {
924             object_unref(OBJECT(server_ioc));
925             error_report_err(local_err);
926             return 1;
927         }
928     } else {
929         /* See comment in check_socket_activation above. */
930         assert(socket_activation == 1);
931         server_ioc = qio_channel_socket_new_fd(FIRST_SOCKET_ACTIVATION_FD,
932                                                &local_err);
933         if (server_ioc == NULL) {
934             error_report("Failed to use socket activation: %s",
935                          error_get_pretty(local_err));
936             exit(EXIT_FAILURE);
937         }
938     }
939 
940     if (qemu_init_main_loop(&local_err)) {
941         error_report_err(local_err);
942         exit(EXIT_FAILURE);
943     }
944     bdrv_init();
945     atexit(bdrv_close_all);
946 
947     srcpath = argv[optind];
948     if (imageOpts) {
949         QemuOpts *opts;
950         if (fmt) {
951             error_report("--image-opts and -f are mutually exclusive");
952             exit(EXIT_FAILURE);
953         }
954         opts = qemu_opts_parse_noisily(&file_opts, srcpath, true);
955         if (!opts) {
956             qemu_opts_reset(&file_opts);
957             exit(EXIT_FAILURE);
958         }
959         options = qemu_opts_to_qdict(opts, NULL);
960         qemu_opts_reset(&file_opts);
961         blk = blk_new_open(NULL, NULL, options, flags, &local_err);
962     } else {
963         if (fmt) {
964             options = qdict_new();
965             qdict_put_str(options, "driver", fmt);
966         }
967         blk = blk_new_open(srcpath, NULL, options, flags, &local_err);
968     }
969 
970     if (!blk) {
971         error_reportf_err(local_err, "Failed to blk_new_open '%s': ",
972                           argv[optind]);
973         exit(EXIT_FAILURE);
974     }
975     bs = blk_bs(blk);
976 
977     blk_set_enable_write_cache(blk, !writethrough);
978 
979     if (sn_opts) {
980         ret = bdrv_snapshot_load_tmp(bs,
981                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
982                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
983                                      &local_err);
984     } else if (sn_id_or_name) {
985         ret = bdrv_snapshot_load_tmp_by_id_or_name(bs, sn_id_or_name,
986                                                    &local_err);
987     }
988     if (ret < 0) {
989         error_reportf_err(local_err, "Failed to load snapshot: ");
990         exit(EXIT_FAILURE);
991     }
992 
993     bs->detect_zeroes = detect_zeroes;
994     fd_size = blk_getlength(blk);
995     if (fd_size < 0) {
996         error_report("Failed to determine the image length: %s",
997                      strerror(-fd_size));
998         exit(EXIT_FAILURE);
999     }
1000 
1001     if (dev_offset >= fd_size) {
1002         error_report("Offset (%lld) has to be smaller than the image size "
1003                      "(%lld)",
1004                      (long long int)dev_offset, (long long int)fd_size);
1005         exit(EXIT_FAILURE);
1006     }
1007     fd_size -= dev_offset;
1008 
1009     if (partition != -1) {
1010         ret = find_partition(blk, partition, &dev_offset, &fd_size);
1011         if (ret < 0) {
1012             error_report("Could not find partition %d: %s", partition,
1013                          strerror(-ret));
1014             exit(EXIT_FAILURE);
1015         }
1016     }
1017 
1018     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed,
1019                          writethrough, NULL, &local_err);
1020     if (!exp) {
1021         error_report_err(local_err);
1022         exit(EXIT_FAILURE);
1023     }
1024     if (export_name) {
1025         nbd_export_set_name(exp, export_name);
1026         nbd_export_set_description(exp, export_description);
1027         newproto = true;
1028     } else if (export_description) {
1029         error_report("Export description requires an export name");
1030         exit(EXIT_FAILURE);
1031     }
1032 
1033     if (device) {
1034         int ret;
1035 
1036         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
1037         if (ret != 0) {
1038             error_report("Failed to create client thread: %s", strerror(ret));
1039             exit(EXIT_FAILURE);
1040         }
1041     } else {
1042         /* Shut up GCC warnings.  */
1043         memset(&client_thread, 0, sizeof(client_thread));
1044     }
1045 
1046     nbd_update_server_watch();
1047 
1048     /* now when the initialization is (almost) complete, chdir("/")
1049      * to free any busy filesystems */
1050     if (chdir("/") < 0) {
1051         error_report("Could not chdir to root directory: %s",
1052                      strerror(errno));
1053         exit(EXIT_FAILURE);
1054     }
1055 
1056     if (fork_process) {
1057         dup2(old_stderr, STDERR_FILENO);
1058         close(old_stderr);
1059     }
1060 
1061     state = RUNNING;
1062     do {
1063         main_loop_wait(false);
1064         if (state == TERMINATE) {
1065             state = TERMINATING;
1066             nbd_export_close(exp);
1067             nbd_export_put(exp);
1068             exp = NULL;
1069         }
1070     } while (state != TERMINATED);
1071 
1072     blk_unref(blk);
1073     if (sockpath) {
1074         unlink(sockpath);
1075     }
1076 
1077     qemu_opts_del(sn_opts);
1078 
1079     if (device) {
1080         void *ret;
1081         pthread_join(client_thread, &ret);
1082         exit(ret != NULL);
1083     } else {
1084         exit(EXIT_SUCCESS);
1085     }
1086 }
1087