xref: /openbmc/qemu/qemu-nbd.c (revision 14a650ec) 
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-common.h"
20 #include "block/block.h"
21 #include "block/nbd.h"
22 #include "qemu/main-loop.h"
23 
24 #include <stdarg.h>
25 #include <stdio.h>
26 #include <getopt.h>
27 #include <err.h>
28 #include <sys/types.h>
29 #include <sys/socket.h>
30 #include <netinet/in.h>
31 #include <netinet/tcp.h>
32 #include <arpa/inet.h>
33 #include <signal.h>
34 #include <libgen.h>
35 #include <pthread.h>
36 
37 #define SOCKET_PATH          "/var/lock/qemu-nbd-%s"
38 #define QEMU_NBD_OPT_CACHE   1
39 #define QEMU_NBD_OPT_AIO     2
40 #define QEMU_NBD_OPT_DISCARD 3
41 
42 static NBDExport *exp;
43 static int verbose;
44 static char *srcpath;
45 static char *sockpath;
46 static int persistent = 0;
47 static enum { RUNNING, TERMINATE, TERMINATING, TERMINATED } state;
48 static int shared = 1;
49 static int nb_fds;
50 
51 static void usage(const char *name)
52 {
53     (printf) (
54 "Usage: %s [OPTIONS] FILE\n"
55 "QEMU Disk Network Block Device Server\n"
56 "\n"
57 "  -h, --help           display this help and exit\n"
58 "  -V, --version        output version information and exit\n"
59 "\n"
60 "Connection properties:\n"
61 "  -p, --port=PORT      port to listen on (default `%d')\n"
62 "  -b, --bind=IFACE     interface to bind to (default `0.0.0.0')\n"
63 "  -k, --socket=PATH    path to the unix socket\n"
64 "                       (default '"SOCKET_PATH"')\n"
65 "  -e, --shared=NUM     device can be shared by NUM clients (default '1')\n"
66 "  -t, --persistent     don't exit on the last connection\n"
67 "  -v, --verbose        display extra debugging information\n"
68 "\n"
69 "Exposing part of the image:\n"
70 "  -o, --offset=OFFSET  offset into the image\n"
71 "  -P, --partition=NUM  only expose partition NUM\n"
72 "\n"
73 #ifdef __linux__
74 "Kernel NBD client support:\n"
75 "  -c, --connect=DEV    connect FILE to the local NBD device DEV\n"
76 "  -d, --disconnect     disconnect the specified device\n"
77 "\n"
78 #endif
79 "\n"
80 "Block device options:\n"
81 "  -r, --read-only      export read-only\n"
82 "  -s, --snapshot       use snapshot file\n"
83 "  -n, --nocache        disable host cache\n"
84 "      --cache=MODE     set cache mode (none, writeback, ...)\n"
85 #ifdef CONFIG_LINUX_AIO
86 "      --aio=MODE       set AIO mode (native or threads)\n"
87 #endif
88 "\n"
89 "Report bugs to <qemu-devel@nongnu.org>\n"
90     , name, NBD_DEFAULT_PORT, "DEVICE");
91 }
92 
93 static void version(const char *name)
94 {
95     printf(
96 "%s version 0.0.1\n"
97 "Written by Anthony Liguori.\n"
98 "\n"
99 "Copyright (C) 2006 Anthony Liguori <anthony@codemonkey.ws>.\n"
100 "This is free software; see the source for copying conditions.  There is NO\n"
101 "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
102     , name);
103 }
104 
105 struct partition_record
106 {
107     uint8_t bootable;
108     uint8_t start_head;
109     uint32_t start_cylinder;
110     uint8_t start_sector;
111     uint8_t system;
112     uint8_t end_head;
113     uint8_t end_cylinder;
114     uint8_t end_sector;
115     uint32_t start_sector_abs;
116     uint32_t nb_sectors_abs;
117 };
118 
119 static void read_partition(uint8_t *p, struct partition_record *r)
120 {
121     r->bootable = p[0];
122     r->start_head = p[1];
123     r->start_cylinder = p[3] | ((p[2] << 2) & 0x0300);
124     r->start_sector = p[2] & 0x3f;
125     r->system = p[4];
126     r->end_head = p[5];
127     r->end_cylinder = p[7] | ((p[6] << 2) & 0x300);
128     r->end_sector = p[6] & 0x3f;
129     r->start_sector_abs = p[8] | p[9] << 8 | p[10] << 16 | p[11] << 24;
130     r->nb_sectors_abs = p[12] | p[13] << 8 | p[14] << 16 | p[15] << 24;
131 }
132 
133 static int find_partition(BlockDriverState *bs, int partition,
134                           off_t *offset, off_t *size)
135 {
136     struct partition_record mbr[4];
137     uint8_t data[512];
138     int i;
139     int ext_partnum = 4;
140     int ret;
141 
142     if ((ret = bdrv_read(bs, 0, data, 1)) < 0) {
143         errno = -ret;
144         err(EXIT_FAILURE, "error while reading");
145     }
146 
147     if (data[510] != 0x55 || data[511] != 0xaa) {
148         return -EINVAL;
149     }
150 
151     for (i = 0; i < 4; i++) {
152         read_partition(&data[446 + 16 * i], &mbr[i]);
153 
154         if (!mbr[i].nb_sectors_abs)
155             continue;
156 
157         if (mbr[i].system == 0xF || mbr[i].system == 0x5) {
158             struct partition_record ext[4];
159             uint8_t data1[512];
160             int j;
161 
162             if ((ret = bdrv_read(bs, mbr[i].start_sector_abs, data1, 1)) < 0) {
163                 errno = -ret;
164                 err(EXIT_FAILURE, "error while reading");
165             }
166 
167             for (j = 0; j < 4; j++) {
168                 read_partition(&data1[446 + 16 * j], &ext[j]);
169                 if (!ext[j].nb_sectors_abs)
170                     continue;
171 
172                 if ((ext_partnum + j + 1) == partition) {
173                     *offset = (uint64_t)ext[j].start_sector_abs << 9;
174                     *size = (uint64_t)ext[j].nb_sectors_abs << 9;
175                     return 0;
176                 }
177             }
178             ext_partnum += 4;
179         } else if ((i + 1) == partition) {
180             *offset = (uint64_t)mbr[i].start_sector_abs << 9;
181             *size = (uint64_t)mbr[i].nb_sectors_abs << 9;
182             return 0;
183         }
184     }
185 
186     return -ENOENT;
187 }
188 
189 static void termsig_handler(int signum)
190 {
191     state = TERMINATE;
192     qemu_notify_event();
193 }
194 
195 static void *show_parts(void *arg)
196 {
197     char *device = arg;
198     int nbd;
199 
200     /* linux just needs an open() to trigger
201      * the partition table update
202      * but remember to load the module with max_part != 0 :
203      *     modprobe nbd max_part=63
204      */
205     nbd = open(device, O_RDWR);
206     if (nbd >= 0) {
207         close(nbd);
208     }
209     return NULL;
210 }
211 
212 static void *nbd_client_thread(void *arg)
213 {
214     char *device = arg;
215     off_t size;
216     size_t blocksize;
217     uint32_t nbdflags;
218     int fd, sock;
219     int ret;
220     pthread_t show_parts_thread;
221 
222     sock = unix_socket_outgoing(sockpath);
223     if (sock < 0) {
224         goto out;
225     }
226 
227     ret = nbd_receive_negotiate(sock, NULL, &nbdflags,
228                                 &size, &blocksize);
229     if (ret < 0) {
230         goto out;
231     }
232 
233     fd = open(device, O_RDWR);
234     if (fd < 0) {
235         /* Linux-only, we can use %m in printf.  */
236         fprintf(stderr, "Failed to open %s: %m", device);
237         goto out;
238     }
239 
240     ret = nbd_init(fd, sock, nbdflags, size, blocksize);
241     if (ret < 0) {
242         goto out;
243     }
244 
245     /* update partition table */
246     pthread_create(&show_parts_thread, NULL, show_parts, device);
247 
248     if (verbose) {
249         fprintf(stderr, "NBD device %s is now connected to %s\n",
250                 device, srcpath);
251     } else {
252         /* Close stderr so that the qemu-nbd process exits.  */
253         dup2(STDOUT_FILENO, STDERR_FILENO);
254     }
255 
256     ret = nbd_client(fd);
257     if (ret) {
258         goto out;
259     }
260     close(fd);
261     kill(getpid(), SIGTERM);
262     return (void *) EXIT_SUCCESS;
263 
264 out:
265     kill(getpid(), SIGTERM);
266     return (void *) EXIT_FAILURE;
267 }
268 
269 static int nbd_can_accept(void *opaque)
270 {
271     return nb_fds < shared;
272 }
273 
274 static void nbd_export_closed(NBDExport *exp)
275 {
276     assert(state == TERMINATING);
277     state = TERMINATED;
278 }
279 
280 static void nbd_client_closed(NBDClient *client)
281 {
282     nb_fds--;
283     if (nb_fds == 0 && !persistent && state == RUNNING) {
284         state = TERMINATE;
285     }
286     qemu_notify_event();
287     nbd_client_put(client);
288 }
289 
290 static void nbd_accept(void *opaque)
291 {
292     int server_fd = (uintptr_t) opaque;
293     struct sockaddr_in addr;
294     socklen_t addr_len = sizeof(addr);
295 
296     int fd = accept(server_fd, (struct sockaddr *)&addr, &addr_len);
297     if (state >= TERMINATE) {
298         close(fd);
299         return;
300     }
301 
302     if (fd >= 0 && nbd_client_new(exp, fd, nbd_client_closed)) {
303         nb_fds++;
304     }
305 }
306 
307 int main(int argc, char **argv)
308 {
309     BlockDriverState *bs;
310     BlockDriver *drv;
311     off_t dev_offset = 0;
312     uint32_t nbdflags = 0;
313     bool disconnect = false;
314     const char *bindto = "0.0.0.0";
315     char *device = NULL;
316     int port = NBD_DEFAULT_PORT;
317     off_t fd_size;
318     const char *sopt = "hVb:o:p:rsnP:c:dvk:e:f:t";
319     struct option lopt[] = {
320         { "help", 0, NULL, 'h' },
321         { "version", 0, NULL, 'V' },
322         { "bind", 1, NULL, 'b' },
323         { "port", 1, NULL, 'p' },
324         { "socket", 1, NULL, 'k' },
325         { "offset", 1, NULL, 'o' },
326         { "read-only", 0, NULL, 'r' },
327         { "partition", 1, NULL, 'P' },
328         { "connect", 1, NULL, 'c' },
329         { "disconnect", 0, NULL, 'd' },
330         { "snapshot", 0, NULL, 's' },
331         { "nocache", 0, NULL, 'n' },
332         { "cache", 1, NULL, QEMU_NBD_OPT_CACHE },
333 #ifdef CONFIG_LINUX_AIO
334         { "aio", 1, NULL, QEMU_NBD_OPT_AIO },
335 #endif
336         { "discard", 1, NULL, QEMU_NBD_OPT_DISCARD },
337         { "shared", 1, NULL, 'e' },
338         { "format", 1, NULL, 'f' },
339         { "persistent", 0, NULL, 't' },
340         { "verbose", 0, NULL, 'v' },
341         { NULL, 0, NULL, 0 }
342     };
343     int ch;
344     int opt_ind = 0;
345     int li;
346     char *end;
347     int flags = BDRV_O_RDWR;
348     int partition = -1;
349     int ret;
350     int fd;
351     bool seen_cache = false;
352     bool seen_discard = false;
353 #ifdef CONFIG_LINUX_AIO
354     bool seen_aio = false;
355 #endif
356     pthread_t client_thread;
357     const char *fmt = NULL;
358     Error *local_err = NULL;
359 
360     /* The client thread uses SIGTERM to interrupt the server.  A signal
361      * handler ensures that "qemu-nbd -v -c" exits with a nice status code.
362      */
363     struct sigaction sa_sigterm;
364     memset(&sa_sigterm, 0, sizeof(sa_sigterm));
365     sa_sigterm.sa_handler = termsig_handler;
366     sigaction(SIGTERM, &sa_sigterm, NULL);
367 
368     while ((ch = getopt_long(argc, argv, sopt, lopt, &opt_ind)) != -1) {
369         switch (ch) {
370         case 's':
371             flags |= BDRV_O_SNAPSHOT;
372             break;
373         case 'n':
374             optarg = (char *) "none";
375             /* fallthrough */
376         case QEMU_NBD_OPT_CACHE:
377             if (seen_cache) {
378                 errx(EXIT_FAILURE, "-n and --cache can only be specified once");
379             }
380             seen_cache = true;
381             if (bdrv_parse_cache_flags(optarg, &flags) == -1) {
382                 errx(EXIT_FAILURE, "Invalid cache mode `%s'", optarg);
383             }
384             break;
385 #ifdef CONFIG_LINUX_AIO
386         case QEMU_NBD_OPT_AIO:
387             if (seen_aio) {
388                 errx(EXIT_FAILURE, "--aio can only be specified once");
389             }
390             seen_aio = true;
391             if (!strcmp(optarg, "native")) {
392                 flags |= BDRV_O_NATIVE_AIO;
393             } else if (!strcmp(optarg, "threads")) {
394                 /* this is the default */
395             } else {
396                errx(EXIT_FAILURE, "invalid aio mode `%s'", optarg);
397             }
398             break;
399 #endif
400         case QEMU_NBD_OPT_DISCARD:
401             if (seen_discard) {
402                 errx(EXIT_FAILURE, "--discard can only be specified once");
403             }
404             seen_discard = true;
405             if (bdrv_parse_discard_flags(optarg, &flags) == -1) {
406                 errx(EXIT_FAILURE, "Invalid discard mode `%s'", optarg);
407             }
408             break;
409         case 'b':
410             bindto = optarg;
411             break;
412         case 'p':
413             li = strtol(optarg, &end, 0);
414             if (*end) {
415                 errx(EXIT_FAILURE, "Invalid port `%s'", optarg);
416             }
417             if (li < 1 || li > 65535) {
418                 errx(EXIT_FAILURE, "Port out of range `%s'", optarg);
419             }
420             port = (uint16_t)li;
421             break;
422         case 'o':
423                 dev_offset = strtoll (optarg, &end, 0);
424             if (*end) {
425                 errx(EXIT_FAILURE, "Invalid offset `%s'", optarg);
426             }
427             if (dev_offset < 0) {
428                 errx(EXIT_FAILURE, "Offset must be positive `%s'", optarg);
429             }
430             break;
431         case 'r':
432             nbdflags |= NBD_FLAG_READ_ONLY;
433             flags &= ~BDRV_O_RDWR;
434             break;
435         case 'P':
436             partition = strtol(optarg, &end, 0);
437             if (*end)
438                 errx(EXIT_FAILURE, "Invalid partition `%s'", optarg);
439             if (partition < 1 || partition > 8)
440                 errx(EXIT_FAILURE, "Invalid partition %d", partition);
441             break;
442         case 'k':
443             sockpath = optarg;
444             if (sockpath[0] != '/')
445                 errx(EXIT_FAILURE, "socket path must be absolute\n");
446             break;
447         case 'd':
448             disconnect = true;
449             break;
450         case 'c':
451             device = optarg;
452             break;
453         case 'e':
454             shared = strtol(optarg, &end, 0);
455             if (*end) {
456                 errx(EXIT_FAILURE, "Invalid shared device number '%s'", optarg);
457             }
458             if (shared < 1) {
459                 errx(EXIT_FAILURE, "Shared device number must be greater than 0\n");
460             }
461             break;
462         case 'f':
463             fmt = optarg;
464             break;
465 	case 't':
466 	    persistent = 1;
467 	    break;
468         case 'v':
469             verbose = 1;
470             break;
471         case 'V':
472             version(argv[0]);
473             exit(0);
474             break;
475         case 'h':
476             usage(argv[0]);
477             exit(0);
478             break;
479         case '?':
480             errx(EXIT_FAILURE, "Try `%s --help' for more information.",
481                  argv[0]);
482         }
483     }
484 
485     if ((argc - optind) != 1) {
486         errx(EXIT_FAILURE, "Invalid number of argument.\n"
487              "Try `%s --help' for more information.",
488              argv[0]);
489     }
490 
491     if (disconnect) {
492         fd = open(argv[optind], O_RDWR);
493         if (fd < 0) {
494             err(EXIT_FAILURE, "Cannot open %s", argv[optind]);
495         }
496         nbd_disconnect(fd);
497 
498         close(fd);
499 
500         printf("%s disconnected\n", argv[optind]);
501 
502 	return 0;
503     }
504 
505     if (device && !verbose) {
506         int stderr_fd[2];
507         pid_t pid;
508         int ret;
509 
510         if (qemu_pipe(stderr_fd) < 0) {
511             err(EXIT_FAILURE, "Error setting up communication pipe");
512         }
513 
514         /* Now daemonize, but keep a communication channel open to
515          * print errors and exit with the proper status code.
516          */
517         pid = fork();
518         if (pid == 0) {
519             close(stderr_fd[0]);
520             ret = qemu_daemon(1, 0);
521 
522             /* Temporarily redirect stderr to the parent's pipe...  */
523             dup2(stderr_fd[1], STDERR_FILENO);
524             if (ret < 0) {
525                 err(EXIT_FAILURE, "Failed to daemonize");
526             }
527 
528             /* ... close the descriptor we inherited and go on.  */
529             close(stderr_fd[1]);
530         } else {
531             bool errors = false;
532             char *buf;
533 
534             /* In the parent.  Print error messages from the child until
535              * it closes the pipe.
536              */
537             close(stderr_fd[1]);
538             buf = g_malloc(1024);
539             while ((ret = read(stderr_fd[0], buf, 1024)) > 0) {
540                 errors = true;
541                 ret = qemu_write_full(STDERR_FILENO, buf, ret);
542                 if (ret < 0) {
543                     exit(EXIT_FAILURE);
544                 }
545             }
546             if (ret < 0) {
547                 err(EXIT_FAILURE, "Cannot read from daemon");
548             }
549 
550             /* Usually the daemon should not print any message.
551              * Exit with zero status in that case.
552              */
553             exit(errors);
554         }
555     }
556 
557     if (device != NULL && sockpath == NULL) {
558         sockpath = g_malloc(128);
559         snprintf(sockpath, 128, SOCKET_PATH, basename(device));
560     }
561 
562     qemu_init_main_loop();
563     bdrv_init();
564     atexit(bdrv_close_all);
565 
566     if (fmt) {
567         drv = bdrv_find_format(fmt);
568         if (!drv) {
569             errx(EXIT_FAILURE, "Unknown file format '%s'", fmt);
570         }
571     } else {
572         drv = NULL;
573     }
574 
575     bs = bdrv_new("hda");
576     srcpath = argv[optind];
577     ret = bdrv_open(bs, srcpath, NULL, flags, drv, &local_err);
578     if (ret < 0) {
579         errno = -ret;
580         err(EXIT_FAILURE, "Failed to bdrv_open '%s': %s", argv[optind],
581             error_get_pretty(local_err));
582     }
583 
584     fd_size = bdrv_getlength(bs);
585 
586     if (partition != -1) {
587         ret = find_partition(bs, partition, &dev_offset, &fd_size);
588         if (ret < 0) {
589             errno = -ret;
590             err(EXIT_FAILURE, "Could not find partition %d", partition);
591         }
592     }
593 
594     exp = nbd_export_new(bs, dev_offset, fd_size, nbdflags, nbd_export_closed);
595 
596     if (sockpath) {
597         fd = unix_socket_incoming(sockpath);
598     } else {
599         fd = tcp_socket_incoming(bindto, port);
600     }
601 
602     if (fd < 0) {
603         return 1;
604     }
605 
606     if (device) {
607         int ret;
608 
609         ret = pthread_create(&client_thread, NULL, nbd_client_thread, device);
610         if (ret != 0) {
611             errx(EXIT_FAILURE, "Failed to create client thread: %s",
612                  strerror(ret));
613         }
614     } else {
615         /* Shut up GCC warnings.  */
616         memset(&client_thread, 0, sizeof(client_thread));
617     }
618 
619     qemu_set_fd_handler2(fd, nbd_can_accept, nbd_accept, NULL,
620                          (void *)(uintptr_t)fd);
621 
622     /* now when the initialization is (almost) complete, chdir("/")
623      * to free any busy filesystems */
624     if (chdir("/") < 0) {
625         err(EXIT_FAILURE, "Could not chdir to root directory");
626     }
627 
628     state = RUNNING;
629     do {
630         main_loop_wait(false);
631         if (state == TERMINATE) {
632             state = TERMINATING;
633             nbd_export_close(exp);
634             nbd_export_put(exp);
635             exp = NULL;
636         }
637     } while (state != TERMINATED);
638 
639     bdrv_close(bs);
640     if (sockpath) {
641         unlink(sockpath);
642     }
643 
644     if (device) {
645         void *ret;
646         pthread_join(client_thread, &ret);
647         exit(ret != NULL);
648     } else {
649         exit(EXIT_SUCCESS);
650     }
651 }
652