]> git.proxmox.com Git - qemu.git/blob - nbd.c
nbd: fix error handling in the server
[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), flags | NBD_FLAG_HAS_FLAGS);
206 memset(buf + 28, 0, 124);
207
208 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
209 LOG("write failed");
210 errno = EINVAL;
211 return -1;
212 }
213
214 TRACE("Negotiation succeeded.");
215
216 return 0;
217 }
218
219 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
220 off_t *size, size_t *blocksize)
221 {
222 char buf[256];
223 uint64_t magic, s;
224 uint16_t tmp;
225
226 TRACE("Receiving negotiation.");
227
228 if (read_sync(csock, buf, 8) != 8) {
229 LOG("read failed");
230 errno = EINVAL;
231 return -1;
232 }
233
234 buf[8] = '\0';
235 if (strlen(buf) == 0) {
236 LOG("server connection closed");
237 errno = EINVAL;
238 return -1;
239 }
240
241 TRACE("Magic is %c%c%c%c%c%c%c%c",
242 qemu_isprint(buf[0]) ? buf[0] : '.',
243 qemu_isprint(buf[1]) ? buf[1] : '.',
244 qemu_isprint(buf[2]) ? buf[2] : '.',
245 qemu_isprint(buf[3]) ? buf[3] : '.',
246 qemu_isprint(buf[4]) ? buf[4] : '.',
247 qemu_isprint(buf[5]) ? buf[5] : '.',
248 qemu_isprint(buf[6]) ? buf[6] : '.',
249 qemu_isprint(buf[7]) ? buf[7] : '.');
250
251 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
252 LOG("Invalid magic received");
253 errno = EINVAL;
254 return -1;
255 }
256
257 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
258 LOG("read failed");
259 errno = EINVAL;
260 return -1;
261 }
262 magic = be64_to_cpu(magic);
263 TRACE("Magic is 0x%" PRIx64, magic);
264
265 if (name) {
266 uint32_t reserved = 0;
267 uint32_t opt;
268 uint32_t namesize;
269
270 TRACE("Checking magic (opts_magic)");
271 if (magic != 0x49484156454F5054LL) {
272 LOG("Bad magic received");
273 errno = EINVAL;
274 return -1;
275 }
276 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
277 LOG("flags read failed");
278 errno = EINVAL;
279 return -1;
280 }
281 *flags = be16_to_cpu(tmp) << 16;
282 /* reserved for future use */
283 if (write_sync(csock, &reserved, sizeof(reserved)) !=
284 sizeof(reserved)) {
285 LOG("write failed (reserved)");
286 errno = EINVAL;
287 return -1;
288 }
289 /* write the export name */
290 magic = cpu_to_be64(magic);
291 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
292 LOG("write failed (magic)");
293 errno = EINVAL;
294 return -1;
295 }
296 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
297 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
298 LOG("write failed (opt)");
299 errno = EINVAL;
300 return -1;
301 }
302 namesize = cpu_to_be32(strlen(name));
303 if (write_sync(csock, &namesize, sizeof(namesize)) !=
304 sizeof(namesize)) {
305 LOG("write failed (namesize)");
306 errno = EINVAL;
307 return -1;
308 }
309 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
310 LOG("write failed (name)");
311 errno = EINVAL;
312 return -1;
313 }
314 } else {
315 TRACE("Checking magic (cli_magic)");
316
317 if (magic != 0x00420281861253LL) {
318 LOG("Bad magic received");
319 errno = EINVAL;
320 return -1;
321 }
322 }
323
324 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
325 LOG("read failed");
326 errno = EINVAL;
327 return -1;
328 }
329 *size = be64_to_cpu(s);
330 *blocksize = 1024;
331 TRACE("Size is %" PRIu64, *size);
332
333 if (!name) {
334 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
335 LOG("read failed (flags)");
336 errno = EINVAL;
337 return -1;
338 }
339 *flags = be32_to_cpup(flags);
340 } else {
341 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
342 LOG("read failed (tmp)");
343 errno = EINVAL;
344 return -1;
345 }
346 *flags |= be32_to_cpu(tmp);
347 }
348 if (read_sync(csock, &buf, 124) != 124) {
349 LOG("read failed (buf)");
350 errno = EINVAL;
351 return -1;
352 }
353 return 0;
354 }
355
356 #ifdef __linux__
357 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
358 {
359 TRACE("Setting block size to %lu", (unsigned long)blocksize);
360
361 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) == -1) {
362 int serrno = errno;
363 LOG("Failed setting NBD block size");
364 errno = serrno;
365 return -1;
366 }
367
368 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
369
370 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) == -1) {
371 int serrno = errno;
372 LOG("Failed setting size (in blocks)");
373 errno = serrno;
374 return -1;
375 }
376
377 if (flags & NBD_FLAG_READ_ONLY) {
378 int read_only = 1;
379 TRACE("Setting readonly attribute");
380
381 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
382 int serrno = errno;
383 LOG("Failed setting read-only attribute");
384 errno = serrno;
385 return -1;
386 }
387 }
388
389 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
390 && errno != ENOTTY) {
391 int serrno = errno;
392 LOG("Failed setting flags");
393 errno = serrno;
394 return -1;
395 }
396
397 TRACE("Clearing NBD socket");
398
399 if (ioctl(fd, NBD_CLEAR_SOCK) == -1) {
400 int serrno = errno;
401 LOG("Failed clearing NBD socket");
402 errno = serrno;
403 return -1;
404 }
405
406 TRACE("Setting NBD socket");
407
408 if (ioctl(fd, NBD_SET_SOCK, csock) == -1) {
409 int serrno = errno;
410 LOG("Failed to set NBD socket");
411 errno = serrno;
412 return -1;
413 }
414
415 TRACE("Negotiation ended");
416
417 return 0;
418 }
419
420 int nbd_disconnect(int fd)
421 {
422 ioctl(fd, NBD_CLEAR_QUE);
423 ioctl(fd, NBD_DISCONNECT);
424 ioctl(fd, NBD_CLEAR_SOCK);
425 return 0;
426 }
427
428 int nbd_client(int fd)
429 {
430 int ret;
431 int serrno;
432
433 TRACE("Doing NBD loop");
434
435 ret = ioctl(fd, NBD_DO_IT);
436 if (ret == -1 && errno == EPIPE) {
437 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
438 * the socket via NBD_DISCONNECT. We do not want to return 1 in
439 * that case.
440 */
441 ret = 0;
442 }
443 serrno = errno;
444
445 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
446
447 TRACE("Clearing NBD queue");
448 ioctl(fd, NBD_CLEAR_QUE);
449
450 TRACE("Clearing NBD socket");
451 ioctl(fd, NBD_CLEAR_SOCK);
452
453 errno = serrno;
454 return ret;
455 }
456 #else
457 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
458 {
459 errno = ENOTSUP;
460 return -1;
461 }
462
463 int nbd_disconnect(int fd)
464 {
465 errno = ENOTSUP;
466 return -1;
467 }
468
469 int nbd_client(int fd)
470 {
471 errno = ENOTSUP;
472 return -1;
473 }
474 #endif
475
476 int nbd_send_request(int csock, struct nbd_request *request)
477 {
478 uint8_t buf[4 + 4 + 8 + 8 + 4];
479
480 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
481 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
482 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
483 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
484 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
485
486 TRACE("Sending request to client: "
487 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
488 request->from, request->len, request->handle, request->type);
489
490 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
491 LOG("writing to socket failed");
492 errno = EINVAL;
493 return -1;
494 }
495 return 0;
496 }
497
498 static int nbd_receive_request(int csock, struct nbd_request *request)
499 {
500 uint8_t buf[4 + 4 + 8 + 8 + 4];
501 uint32_t magic;
502
503 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
504 LOG("read failed");
505 errno = EINVAL;
506 return -1;
507 }
508
509 /* Request
510 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
511 [ 4 .. 7] type (0 == READ, 1 == WRITE)
512 [ 8 .. 15] handle
513 [16 .. 23] from
514 [24 .. 27] len
515 */
516
517 magic = be32_to_cpup((uint32_t*)buf);
518 request->type = be32_to_cpup((uint32_t*)(buf + 4));
519 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
520 request->from = be64_to_cpup((uint64_t*)(buf + 16));
521 request->len = be32_to_cpup((uint32_t*)(buf + 24));
522
523 TRACE("Got request: "
524 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
525 magic, request->type, request->from, request->len);
526
527 if (magic != NBD_REQUEST_MAGIC) {
528 LOG("invalid magic (got 0x%x)", magic);
529 errno = EINVAL;
530 return -1;
531 }
532 return 0;
533 }
534
535 int nbd_receive_reply(int csock, struct nbd_reply *reply)
536 {
537 uint8_t buf[NBD_REPLY_SIZE];
538 uint32_t magic;
539
540 memset(buf, 0xAA, sizeof(buf));
541
542 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
543 LOG("read failed");
544 errno = EINVAL;
545 return -1;
546 }
547
548 /* Reply
549 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
550 [ 4 .. 7] error (0 == no error)
551 [ 7 .. 15] handle
552 */
553
554 magic = be32_to_cpup((uint32_t*)buf);
555 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
556 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
557
558 TRACE("Got reply: "
559 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
560 magic, reply->error, reply->handle);
561
562 if (magic != NBD_REPLY_MAGIC) {
563 LOG("invalid magic (got 0x%x)", magic);
564 errno = EINVAL;
565 return -1;
566 }
567 return 0;
568 }
569
570 static int nbd_send_reply(int csock, struct nbd_reply *reply)
571 {
572 uint8_t buf[4 + 4 + 8];
573
574 /* Reply
575 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
576 [ 4 .. 7] error (0 == no error)
577 [ 7 .. 15] handle
578 */
579 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
580 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
581 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
582
583 TRACE("Sending response to client");
584
585 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
586 LOG("writing to socket failed");
587 errno = EINVAL;
588 return -1;
589 }
590 return 0;
591 }
592
593 int nbd_trip(BlockDriverState *bs, int csock, off_t size, uint64_t dev_offset,
594 off_t *offset, uint32_t nbdflags, uint8_t *data, int data_size)
595 {
596 struct nbd_request request;
597 struct nbd_reply reply;
598 int ret;
599
600 TRACE("Reading request.");
601
602 if (nbd_receive_request(csock, &request) == -1)
603 return -1;
604
605 if (request.len + NBD_REPLY_SIZE > data_size) {
606 LOG("len (%u) is larger than max len (%u)",
607 request.len + NBD_REPLY_SIZE, data_size);
608 errno = EINVAL;
609 return -1;
610 }
611
612 if ((request.from + request.len) < request.from) {
613 LOG("integer overflow detected! "
614 "you're probably being attacked");
615 errno = EINVAL;
616 return -1;
617 }
618
619 if ((request.from + request.len) > size) {
620 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
621 ", Offset: %" PRIu64 "\n",
622 request.from, request.len, (uint64_t)size, dev_offset);
623 LOG("requested operation past EOF--bad client?");
624 errno = EINVAL;
625 return -1;
626 }
627
628 TRACE("Decoding type");
629
630 reply.handle = request.handle;
631 reply.error = 0;
632
633 switch (request.type) {
634 case NBD_CMD_READ:
635 TRACE("Request type is READ");
636
637 ret = bdrv_read(bs, (request.from + dev_offset) / 512,
638 data + NBD_REPLY_SIZE,
639 request.len / 512);
640 if (ret < 0) {
641 LOG("reading from file failed");
642 reply.error = -ret;
643 request.len = 0;
644 }
645 *offset += request.len;
646
647 TRACE("Read %u byte(s)", request.len);
648
649 /* Reply
650 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
651 [ 4 .. 7] error (0 == no error)
652 [ 7 .. 15] handle
653 */
654
655 cpu_to_be32w((uint32_t*)data, NBD_REPLY_MAGIC);
656 cpu_to_be32w((uint32_t*)(data + 4), reply.error);
657 cpu_to_be64w((uint64_t*)(data + 8), reply.handle);
658
659 TRACE("Sending data to client");
660
661 if (write_sync(csock, data,
662 request.len + NBD_REPLY_SIZE) !=
663 request.len + NBD_REPLY_SIZE) {
664 LOG("writing to socket failed");
665 errno = EINVAL;
666 return -1;
667 }
668 break;
669 case NBD_CMD_WRITE:
670 TRACE("Request type is WRITE");
671
672 TRACE("Reading %u byte(s)", request.len);
673
674 if (read_sync(csock, data, request.len) != request.len) {
675 LOG("reading from socket failed");
676 errno = EINVAL;
677 return -1;
678 }
679
680 if (nbdflags & NBD_FLAG_READ_ONLY) {
681 TRACE("Server is read-only, return error");
682 reply.error = 1;
683 } else {
684 TRACE("Writing to device");
685
686 ret = bdrv_write(bs, (request.from + dev_offset) / 512,
687 data, request.len / 512);
688 if (ret < 0) {
689 LOG("writing to file failed");
690 reply.error = -ret;
691 request.len = 0;
692 }
693
694 *offset += request.len;
695 }
696
697 if (nbd_send_reply(csock, &reply) == -1)
698 return -1;
699 break;
700 case NBD_CMD_DISC:
701 TRACE("Request type is DISCONNECT");
702 errno = 0;
703 return 1;
704 default:
705 LOG("invalid request type (%u) received", request.type);
706 errno = EINVAL;
707 return -1;
708 }
709
710 TRACE("Request/Reply complete");
711
712 return 0;
713 }