]> git.proxmox.com Git - qemu.git/blob - nbd.c
qemu-nbd: introduce nbd_do_send_reply
[qemu.git] / nbd.c
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 "nbd.h"
20 #include "block.h"
21
22 #include <errno.h>
23 #include <string.h>
24 #ifndef _WIN32
25 #include <sys/ioctl.h>
26 #endif
27 #if defined(__sun__) || defined(__HAIKU__)
28 #include <sys/ioccom.h>
29 #endif
30 #include <ctype.h>
31 #include <inttypes.h>
32
33 #ifdef __linux__
34 #include <linux/fs.h>
35 #endif
36
37 #include "qemu_socket.h"
38
39 //#define DEBUG_NBD
40
41 #ifdef DEBUG_NBD
42 #define TRACE(msg, ...) do { \
43 LOG(msg, ## __VA_ARGS__); \
44 } while(0)
45 #else
46 #define TRACE(msg, ...) \
47 do { } while (0)
48 #endif
49
50 #define LOG(msg, ...) do { \
51 fprintf(stderr, "%s:%s():L%d: " msg "\n", \
52 __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
53 } while(0)
54
55 /* This is all part of the "official" NBD API */
56
57 #define NBD_REPLY_SIZE (4 + 4 + 8)
58 #define NBD_REQUEST_MAGIC 0x25609513
59 #define NBD_REPLY_MAGIC 0x67446698
60
61 #define NBD_SET_SOCK _IO(0xab, 0)
62 #define NBD_SET_BLKSIZE _IO(0xab, 1)
63 #define NBD_SET_SIZE _IO(0xab, 2)
64 #define NBD_DO_IT _IO(0xab, 3)
65 #define NBD_CLEAR_SOCK _IO(0xab, 4)
66 #define NBD_CLEAR_QUE _IO(0xab, 5)
67 #define NBD_PRINT_DEBUG _IO(0xab, 6)
68 #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7)
69 #define NBD_DISCONNECT _IO(0xab, 8)
70 #define NBD_SET_TIMEOUT _IO(0xab, 9)
71 #define NBD_SET_FLAGS _IO(0xab, 10)
72
73 #define NBD_OPT_EXPORT_NAME (1 << 0)
74
75 /* That's all folks */
76
77 #define read_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, true)
78 #define write_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, false)
79
80 size_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
81 {
82 size_t offset = 0;
83
84 if (qemu_in_coroutine()) {
85 if (do_read) {
86 return qemu_co_recv(fd, buffer, size);
87 } else {
88 return qemu_co_send(fd, buffer, size);
89 }
90 }
91
92 while (offset < size) {
93 ssize_t len;
94
95 if (do_read) {
96 len = qemu_recv(fd, buffer + offset, size - offset, 0);
97 } else {
98 len = send(fd, buffer + offset, size - offset, 0);
99 }
100
101 if (len == -1)
102 errno = socket_error();
103
104 /* recoverable error */
105 if (len == -1 && (errno == EAGAIN || errno == EINTR)) {
106 continue;
107 }
108
109 /* eof */
110 if (len == 0) {
111 break;
112 }
113
114 /* unrecoverable error */
115 if (len == -1) {
116 return 0;
117 }
118
119 offset += len;
120 }
121
122 return offset;
123 }
124
125 static void combine_addr(char *buf, size_t len, const char* address,
126 uint16_t port)
127 {
128 /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
129 if (strstr(address, ":")) {
130 snprintf(buf, len, "[%s]:%u", address, port);
131 } else {
132 snprintf(buf, len, "%s:%u", address, port);
133 }
134 }
135
136 int tcp_socket_outgoing(const char *address, uint16_t port)
137 {
138 char address_and_port[128];
139 combine_addr(address_and_port, 128, address, port);
140 return tcp_socket_outgoing_spec(address_and_port);
141 }
142
143 int tcp_socket_outgoing_spec(const char *address_and_port)
144 {
145 return inet_connect(address_and_port, SOCK_STREAM);
146 }
147
148 int tcp_socket_incoming(const char *address, uint16_t port)
149 {
150 char address_and_port[128];
151 combine_addr(address_and_port, 128, address, port);
152 return tcp_socket_incoming_spec(address_and_port);
153 }
154
155 int tcp_socket_incoming_spec(const char *address_and_port)
156 {
157 char *ostr = NULL;
158 int olen = 0;
159 return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0);
160 }
161
162 int unix_socket_incoming(const char *path)
163 {
164 char *ostr = NULL;
165 int olen = 0;
166
167 return unix_listen(path, ostr, olen);
168 }
169
170 int unix_socket_outgoing(const char *path)
171 {
172 return unix_connect(path);
173 }
174
175 /* Basic flow
176
177 Server Client
178
179 Negotiate
180 Request
181 Response
182 Request
183 Response
184 ...
185 ...
186 Request (type == 2)
187 */
188
189 int nbd_negotiate(int csock, off_t size, uint32_t flags)
190 {
191 char buf[8 + 8 + 8 + 128];
192
193 /* Negotiate
194 [ 0 .. 7] passwd ("NBDMAGIC")
195 [ 8 .. 15] magic (0x00420281861253)
196 [16 .. 23] size
197 [24 .. 27] flags
198 [28 .. 151] reserved (0)
199 */
200
201 TRACE("Beginning negotiation.");
202 memcpy(buf, "NBDMAGIC", 8);
203 cpu_to_be64w((uint64_t*)(buf + 8), 0x00420281861253LL);
204 cpu_to_be64w((uint64_t*)(buf + 16), size);
205 cpu_to_be32w((uint32_t*)(buf + 24),
206 flags | NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
207 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
208 memset(buf + 28, 0, 124);
209
210 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
211 LOG("write failed");
212 errno = EINVAL;
213 return -1;
214 }
215
216 TRACE("Negotiation succeeded.");
217
218 return 0;
219 }
220
221 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
222 off_t *size, size_t *blocksize)
223 {
224 char buf[256];
225 uint64_t magic, s;
226 uint16_t tmp;
227
228 TRACE("Receiving negotiation.");
229
230 if (read_sync(csock, buf, 8) != 8) {
231 LOG("read failed");
232 errno = EINVAL;
233 return -1;
234 }
235
236 buf[8] = '\0';
237 if (strlen(buf) == 0) {
238 LOG("server connection closed");
239 errno = EINVAL;
240 return -1;
241 }
242
243 TRACE("Magic is %c%c%c%c%c%c%c%c",
244 qemu_isprint(buf[0]) ? buf[0] : '.',
245 qemu_isprint(buf[1]) ? buf[1] : '.',
246 qemu_isprint(buf[2]) ? buf[2] : '.',
247 qemu_isprint(buf[3]) ? buf[3] : '.',
248 qemu_isprint(buf[4]) ? buf[4] : '.',
249 qemu_isprint(buf[5]) ? buf[5] : '.',
250 qemu_isprint(buf[6]) ? buf[6] : '.',
251 qemu_isprint(buf[7]) ? buf[7] : '.');
252
253 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
254 LOG("Invalid magic received");
255 errno = EINVAL;
256 return -1;
257 }
258
259 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
260 LOG("read failed");
261 errno = EINVAL;
262 return -1;
263 }
264 magic = be64_to_cpu(magic);
265 TRACE("Magic is 0x%" PRIx64, magic);
266
267 if (name) {
268 uint32_t reserved = 0;
269 uint32_t opt;
270 uint32_t namesize;
271
272 TRACE("Checking magic (opts_magic)");
273 if (magic != 0x49484156454F5054LL) {
274 LOG("Bad magic received");
275 errno = EINVAL;
276 return -1;
277 }
278 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
279 LOG("flags read failed");
280 errno = EINVAL;
281 return -1;
282 }
283 *flags = be16_to_cpu(tmp) << 16;
284 /* reserved for future use */
285 if (write_sync(csock, &reserved, sizeof(reserved)) !=
286 sizeof(reserved)) {
287 LOG("write failed (reserved)");
288 errno = EINVAL;
289 return -1;
290 }
291 /* write the export name */
292 magic = cpu_to_be64(magic);
293 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
294 LOG("write failed (magic)");
295 errno = EINVAL;
296 return -1;
297 }
298 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
299 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
300 LOG("write failed (opt)");
301 errno = EINVAL;
302 return -1;
303 }
304 namesize = cpu_to_be32(strlen(name));
305 if (write_sync(csock, &namesize, sizeof(namesize)) !=
306 sizeof(namesize)) {
307 LOG("write failed (namesize)");
308 errno = EINVAL;
309 return -1;
310 }
311 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
312 LOG("write failed (name)");
313 errno = EINVAL;
314 return -1;
315 }
316 } else {
317 TRACE("Checking magic (cli_magic)");
318
319 if (magic != 0x00420281861253LL) {
320 LOG("Bad magic received");
321 errno = EINVAL;
322 return -1;
323 }
324 }
325
326 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
327 LOG("read failed");
328 errno = EINVAL;
329 return -1;
330 }
331 *size = be64_to_cpu(s);
332 *blocksize = 1024;
333 TRACE("Size is %" PRIu64, *size);
334
335 if (!name) {
336 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
337 LOG("read failed (flags)");
338 errno = EINVAL;
339 return -1;
340 }
341 *flags = be32_to_cpup(flags);
342 } else {
343 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
344 LOG("read failed (tmp)");
345 errno = EINVAL;
346 return -1;
347 }
348 *flags |= be32_to_cpu(tmp);
349 }
350 if (read_sync(csock, &buf, 124) != 124) {
351 LOG("read failed (buf)");
352 errno = EINVAL;
353 return -1;
354 }
355 return 0;
356 }
357
358 #ifdef __linux__
359 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
360 {
361 TRACE("Setting NBD socket");
362
363 if (ioctl(fd, NBD_SET_SOCK, csock) == -1) {
364 int serrno = errno;
365 LOG("Failed to set NBD socket");
366 errno = serrno;
367 return -1;
368 }
369
370 TRACE("Setting block size to %lu", (unsigned long)blocksize);
371
372 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) == -1) {
373 int serrno = errno;
374 LOG("Failed setting NBD block size");
375 errno = serrno;
376 return -1;
377 }
378
379 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
380
381 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) == -1) {
382 int serrno = errno;
383 LOG("Failed setting size (in blocks)");
384 errno = serrno;
385 return -1;
386 }
387
388 if (flags & NBD_FLAG_READ_ONLY) {
389 int read_only = 1;
390 TRACE("Setting readonly attribute");
391
392 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
393 int serrno = errno;
394 LOG("Failed setting read-only attribute");
395 errno = serrno;
396 return -1;
397 }
398 }
399
400 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
401 && errno != ENOTTY) {
402 int serrno = errno;
403 LOG("Failed setting flags");
404 errno = serrno;
405 return -1;
406 }
407
408 TRACE("Negotiation ended");
409
410 return 0;
411 }
412
413 int nbd_disconnect(int fd)
414 {
415 ioctl(fd, NBD_CLEAR_QUE);
416 ioctl(fd, NBD_DISCONNECT);
417 ioctl(fd, NBD_CLEAR_SOCK);
418 return 0;
419 }
420
421 int nbd_client(int fd)
422 {
423 int ret;
424 int serrno;
425
426 TRACE("Doing NBD loop");
427
428 ret = ioctl(fd, NBD_DO_IT);
429 if (ret == -1 && errno == EPIPE) {
430 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
431 * the socket via NBD_DISCONNECT. We do not want to return 1 in
432 * that case.
433 */
434 ret = 0;
435 }
436 serrno = errno;
437
438 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
439
440 TRACE("Clearing NBD queue");
441 ioctl(fd, NBD_CLEAR_QUE);
442
443 TRACE("Clearing NBD socket");
444 ioctl(fd, NBD_CLEAR_SOCK);
445
446 errno = serrno;
447 return ret;
448 }
449 #else
450 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
451 {
452 errno = ENOTSUP;
453 return -1;
454 }
455
456 int nbd_disconnect(int fd)
457 {
458 errno = ENOTSUP;
459 return -1;
460 }
461
462 int nbd_client(int fd)
463 {
464 errno = ENOTSUP;
465 return -1;
466 }
467 #endif
468
469 int nbd_send_request(int csock, struct nbd_request *request)
470 {
471 uint8_t buf[4 + 4 + 8 + 8 + 4];
472
473 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
474 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
475 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
476 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
477 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
478
479 TRACE("Sending request to client: "
480 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
481 request->from, request->len, request->handle, request->type);
482
483 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
484 LOG("writing to socket failed");
485 errno = EINVAL;
486 return -1;
487 }
488 return 0;
489 }
490
491 static int nbd_receive_request(int csock, struct nbd_request *request)
492 {
493 uint8_t buf[4 + 4 + 8 + 8 + 4];
494 uint32_t magic;
495
496 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
497 LOG("read failed");
498 errno = EINVAL;
499 return -1;
500 }
501
502 /* Request
503 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
504 [ 4 .. 7] type (0 == READ, 1 == WRITE)
505 [ 8 .. 15] handle
506 [16 .. 23] from
507 [24 .. 27] len
508 */
509
510 magic = be32_to_cpup((uint32_t*)buf);
511 request->type = be32_to_cpup((uint32_t*)(buf + 4));
512 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
513 request->from = be64_to_cpup((uint64_t*)(buf + 16));
514 request->len = be32_to_cpup((uint32_t*)(buf + 24));
515
516 TRACE("Got request: "
517 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
518 magic, request->type, request->from, request->len);
519
520 if (magic != NBD_REQUEST_MAGIC) {
521 LOG("invalid magic (got 0x%x)", magic);
522 errno = EINVAL;
523 return -1;
524 }
525 return 0;
526 }
527
528 int nbd_receive_reply(int csock, struct nbd_reply *reply)
529 {
530 uint8_t buf[NBD_REPLY_SIZE];
531 uint32_t magic;
532
533 memset(buf, 0xAA, sizeof(buf));
534
535 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
536 LOG("read failed");
537 errno = EINVAL;
538 return -1;
539 }
540
541 /* Reply
542 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
543 [ 4 .. 7] error (0 == no error)
544 [ 7 .. 15] handle
545 */
546
547 magic = be32_to_cpup((uint32_t*)buf);
548 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
549 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
550
551 TRACE("Got reply: "
552 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
553 magic, reply->error, reply->handle);
554
555 if (magic != NBD_REPLY_MAGIC) {
556 LOG("invalid magic (got 0x%x)", magic);
557 errno = EINVAL;
558 return -1;
559 }
560 return 0;
561 }
562
563 static int nbd_send_reply(int csock, struct nbd_reply *reply)
564 {
565 uint8_t buf[4 + 4 + 8];
566
567 /* Reply
568 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
569 [ 4 .. 7] error (0 == no error)
570 [ 7 .. 15] handle
571 */
572 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
573 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
574 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
575
576 TRACE("Sending response to client");
577
578 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
579 LOG("writing to socket failed");
580 errno = EINVAL;
581 return -1;
582 }
583 return 0;
584 }
585
586 static int nbd_do_send_reply(int csock, struct nbd_reply *reply,
587 uint8_t *data, int len)
588 {
589 int rc, ret;
590
591 if (!len) {
592 rc = nbd_send_reply(csock, reply);
593 if (rc == -1) {
594 rc = -errno;
595 }
596 } else {
597 socket_set_cork(csock, 1);
598 rc = nbd_send_reply(csock, reply);
599 if (rc != -1) {
600 ret = write_sync(csock, data, len);
601 if (ret != len) {
602 errno = EIO;
603 rc = -1;
604 }
605 }
606 if (rc == -1) {
607 rc = -errno;
608 }
609 socket_set_cork(csock, 0);
610 }
611 return rc;
612 }
613
614 int nbd_trip(BlockDriverState *bs, int csock, off_t size,
615 uint64_t dev_offset, uint32_t nbdflags,
616 uint8_t *data)
617 {
618 struct nbd_request request;
619 struct nbd_reply reply;
620 int ret;
621
622 TRACE("Reading request.");
623
624 if (nbd_receive_request(csock, &request) == -1)
625 return -1;
626
627 if (request.len > NBD_BUFFER_SIZE) {
628 LOG("len (%u) is larger than max len (%u)",
629 request.len, NBD_BUFFER_SIZE);
630 errno = EINVAL;
631 return -1;
632 }
633
634 if ((request.from + request.len) < request.from) {
635 LOG("integer overflow detected! "
636 "you're probably being attacked");
637 errno = EINVAL;
638 return -1;
639 }
640
641 if ((request.from + request.len) > size) {
642 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
643 ", Offset: %" PRIu64 "\n",
644 request.from, request.len, (uint64_t)size, dev_offset);
645 LOG("requested operation past EOF--bad client?");
646 errno = EINVAL;
647 return -1;
648 }
649
650 TRACE("Decoding type");
651
652 reply.handle = request.handle;
653 reply.error = 0;
654
655 switch (request.type & NBD_CMD_MASK_COMMAND) {
656 case NBD_CMD_READ:
657 TRACE("Request type is READ");
658
659 ret = bdrv_read(bs, (request.from + dev_offset) / 512,
660 data, request.len / 512);
661 if (ret < 0) {
662 LOG("reading from file failed");
663 reply.error = -ret;
664 request.len = 0;
665 }
666
667 TRACE("Read %u byte(s)", request.len);
668 if (nbd_do_send_reply(csock, &reply, data, request.len) < 0)
669 return -1;
670 break;
671 case NBD_CMD_WRITE:
672 TRACE("Request type is WRITE");
673
674 TRACE("Reading %u byte(s)", request.len);
675
676 if (read_sync(csock, data, request.len) != request.len) {
677 LOG("reading from socket failed");
678 errno = EINVAL;
679 return -1;
680 }
681
682 if (nbdflags & NBD_FLAG_READ_ONLY) {
683 TRACE("Server is read-only, return error");
684 reply.error = 1;
685 } else {
686 TRACE("Writing to device");
687
688 ret = bdrv_write(bs, (request.from + dev_offset) / 512,
689 data, request.len / 512);
690 if (ret < 0) {
691 LOG("writing to file failed");
692 reply.error = -ret;
693 request.len = 0;
694 }
695
696 if (request.type & NBD_CMD_FLAG_FUA) {
697 ret = bdrv_flush(bs);
698 if (ret < 0) {
699 LOG("flush failed");
700 reply.error = -ret;
701 }
702 }
703 }
704
705 if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0)
706 return -1;
707 break;
708 case NBD_CMD_DISC:
709 TRACE("Request type is DISCONNECT");
710 errno = 0;
711 return 1;
712 case NBD_CMD_FLUSH:
713 TRACE("Request type is FLUSH");
714
715 ret = bdrv_flush(bs);
716 if (ret < 0) {
717 LOG("flush failed");
718 reply.error = -ret;
719 }
720
721 if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0)
722 return -1;
723 break;
724 case NBD_CMD_TRIM:
725 TRACE("Request type is TRIM");
726 ret = bdrv_discard(bs, (request.from + dev_offset) / 512,
727 request.len / 512);
728 if (ret < 0) {
729 LOG("discard failed");
730 reply.error = -ret;
731 }
732 if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0)
733 return -1;
734 break;
735 default:
736 LOG("invalid request type (%u) received", request.type);
737 errno = EINVAL;
738 return -1;
739 }
740
741 TRACE("Request/Reply complete");
742
743 return 0;
744 }