Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20160614-2' into...
[qemu/kevin.git] / nbd / client.c
blob31b88f3a31c00f8be0febfc7b7d52b2c01d2acff
1 /*
2 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 * Network Block Device Client Side
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.
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.
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/>.
19 #include "qemu/osdep.h"
20 #include "qapi/error.h"
21 #include "nbd-internal.h"
23 static int nbd_errno_to_system_errno(int err)
25 switch (err) {
26 case NBD_SUCCESS:
27 return 0;
28 case NBD_EPERM:
29 return EPERM;
30 case NBD_EIO:
31 return EIO;
32 case NBD_ENOMEM:
33 return ENOMEM;
34 case NBD_ENOSPC:
35 return ENOSPC;
36 case NBD_EINVAL:
37 default:
38 return EINVAL;
42 /* Definitions for opaque data types */
44 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
46 /* That's all folks */
48 /* Basic flow for negotiation
50 Server Client
51 Negotiate
55 Server Client
56 Negotiate #1
57 Option
58 Negotiate #2
60 ----
62 followed by
64 Server Client
65 Request
66 Response
67 Request
68 Response
69 ...
70 ...
71 Request (type == 2)
76 /* If type represents success, return 1 without further action.
77 * If type represents an error reply, consume the rest of the packet on ioc.
78 * Then return 0 for unsupported (so the client can fall back to
79 * other approaches), or -1 with errp set for other errors.
81 static int nbd_handle_reply_err(QIOChannel *ioc, uint32_t opt, uint32_t type,
82 Error **errp)
84 uint32_t len;
85 char *msg = NULL;
86 int result = -1;
88 if (!(type & (1 << 31))) {
89 return 1;
92 if (read_sync(ioc, &len, sizeof(len)) != sizeof(len)) {
93 error_setg(errp, "failed to read option length");
94 return -1;
96 len = be32_to_cpu(len);
97 if (len) {
98 if (len > NBD_MAX_BUFFER_SIZE) {
99 error_setg(errp, "server's error message is too long");
100 goto cleanup;
102 msg = g_malloc(len + 1);
103 if (read_sync(ioc, msg, len) != len) {
104 error_setg(errp, "failed to read option error message");
105 goto cleanup;
107 msg[len] = '\0';
110 switch (type) {
111 case NBD_REP_ERR_UNSUP:
112 TRACE("server doesn't understand request %d, attempting fallback",
113 opt);
114 result = 0;
115 goto cleanup;
117 case NBD_REP_ERR_POLICY:
118 error_setg(errp, "Denied by server for option %x", opt);
119 break;
121 case NBD_REP_ERR_INVALID:
122 error_setg(errp, "Invalid data length for option %x", opt);
123 break;
125 case NBD_REP_ERR_TLS_REQD:
126 error_setg(errp, "TLS negotiation required before option %x", opt);
127 break;
129 default:
130 error_setg(errp, "Unknown error code when asking for option %x", opt);
131 break;
134 if (msg) {
135 error_append_hint(errp, "%s\n", msg);
138 cleanup:
139 g_free(msg);
140 return result;
143 static int nbd_receive_list(QIOChannel *ioc, char **name, Error **errp)
145 uint64_t magic;
146 uint32_t opt;
147 uint32_t type;
148 uint32_t len;
149 uint32_t namelen;
150 int error;
152 *name = NULL;
153 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
154 error_setg(errp, "failed to read list option magic");
155 return -1;
157 magic = be64_to_cpu(magic);
158 if (magic != NBD_REP_MAGIC) {
159 error_setg(errp, "Unexpected option list magic");
160 return -1;
162 if (read_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
163 error_setg(errp, "failed to read list option");
164 return -1;
166 opt = be32_to_cpu(opt);
167 if (opt != NBD_OPT_LIST) {
168 error_setg(errp, "Unexpected option type %x expected %x",
169 opt, NBD_OPT_LIST);
170 return -1;
173 if (read_sync(ioc, &type, sizeof(type)) != sizeof(type)) {
174 error_setg(errp, "failed to read list option type");
175 return -1;
177 type = be32_to_cpu(type);
178 error = nbd_handle_reply_err(ioc, opt, type, errp);
179 if (error <= 0) {
180 return error;
183 if (read_sync(ioc, &len, sizeof(len)) != sizeof(len)) {
184 error_setg(errp, "failed to read option length");
185 return -1;
187 len = be32_to_cpu(len);
189 if (type == NBD_REP_ACK) {
190 if (len != 0) {
191 error_setg(errp, "length too long for option end");
192 return -1;
194 } else if (type == NBD_REP_SERVER) {
195 if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
196 error_setg(errp, "incorrect option length");
197 return -1;
199 if (read_sync(ioc, &namelen, sizeof(namelen)) != sizeof(namelen)) {
200 error_setg(errp, "failed to read option name length");
201 return -1;
203 namelen = be32_to_cpu(namelen);
204 len -= sizeof(namelen);
205 if (len < namelen) {
206 error_setg(errp, "incorrect option name length");
207 return -1;
209 if (namelen > 255) {
210 error_setg(errp, "export name length too long %d", namelen);
211 return -1;
214 *name = g_new0(char, namelen + 1);
215 if (read_sync(ioc, *name, namelen) != namelen) {
216 error_setg(errp, "failed to read export name");
217 g_free(*name);
218 *name = NULL;
219 return -1;
221 (*name)[namelen] = '\0';
222 len -= namelen;
223 if (len) {
224 char *buf = g_malloc(len + 1);
225 if (read_sync(ioc, buf, len) != len) {
226 error_setg(errp, "failed to read export description");
227 g_free(*name);
228 g_free(buf);
229 *name = NULL;
230 return -1;
232 buf[len] = '\0';
233 TRACE("Ignoring export description: %s", buf);
234 g_free(buf);
236 } else {
237 error_setg(errp, "Unexpected reply type %x expected %x",
238 type, NBD_REP_SERVER);
239 return -1;
241 return 1;
245 static int nbd_receive_query_exports(QIOChannel *ioc,
246 const char *wantname,
247 Error **errp)
249 uint64_t magic = cpu_to_be64(NBD_OPTS_MAGIC);
250 uint32_t opt = cpu_to_be32(NBD_OPT_LIST);
251 uint32_t length = 0;
252 bool foundExport = false;
254 TRACE("Querying export list");
255 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
256 error_setg(errp, "Failed to send list option magic");
257 return -1;
260 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
261 error_setg(errp, "Failed to send list option number");
262 return -1;
265 if (write_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
266 error_setg(errp, "Failed to send list option length");
267 return -1;
270 TRACE("Reading available export names");
271 while (1) {
272 char *name = NULL;
273 int ret = nbd_receive_list(ioc, &name, errp);
275 if (ret < 0) {
276 g_free(name);
277 name = NULL;
278 return -1;
280 if (ret == 0) {
281 /* Server doesn't support export listing, so
282 * we will just assume an export with our
283 * wanted name exists */
284 foundExport = true;
285 break;
287 if (name == NULL) {
288 TRACE("End of export name list");
289 break;
291 if (g_str_equal(name, wantname)) {
292 foundExport = true;
293 TRACE("Found desired export name '%s'", name);
294 } else {
295 TRACE("Ignored export name '%s'", name);
297 g_free(name);
300 if (!foundExport) {
301 error_setg(errp, "No export with name '%s' available", wantname);
302 return -1;
305 return 0;
308 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
309 QCryptoTLSCreds *tlscreds,
310 const char *hostname, Error **errp)
312 uint64_t magic = cpu_to_be64(NBD_OPTS_MAGIC);
313 uint32_t opt = cpu_to_be32(NBD_OPT_STARTTLS);
314 uint32_t length = 0;
315 uint32_t type;
316 QIOChannelTLS *tioc;
317 struct NBDTLSHandshakeData data = { 0 };
319 TRACE("Requesting TLS from server");
320 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
321 error_setg(errp, "Failed to send option magic");
322 return NULL;
325 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
326 error_setg(errp, "Failed to send option number");
327 return NULL;
330 if (write_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
331 error_setg(errp, "Failed to send option length");
332 return NULL;
335 TRACE("Getting TLS reply from server1");
336 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
337 error_setg(errp, "failed to read option magic");
338 return NULL;
340 magic = be64_to_cpu(magic);
341 if (magic != NBD_REP_MAGIC) {
342 error_setg(errp, "Unexpected option magic");
343 return NULL;
345 TRACE("Getting TLS reply from server2");
346 if (read_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
347 error_setg(errp, "failed to read option");
348 return NULL;
350 opt = be32_to_cpu(opt);
351 if (opt != NBD_OPT_STARTTLS) {
352 error_setg(errp, "Unexpected option type %x expected %x",
353 opt, NBD_OPT_STARTTLS);
354 return NULL;
357 TRACE("Getting TLS reply from server");
358 if (read_sync(ioc, &type, sizeof(type)) != sizeof(type)) {
359 error_setg(errp, "failed to read option type");
360 return NULL;
362 type = be32_to_cpu(type);
363 if (type != NBD_REP_ACK) {
364 error_setg(errp, "Server rejected request to start TLS %x",
365 type);
366 return NULL;
369 TRACE("Getting TLS reply from server");
370 if (read_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
371 error_setg(errp, "failed to read option length");
372 return NULL;
374 length = be32_to_cpu(length);
375 if (length != 0) {
376 error_setg(errp, "Start TLS response was not zero %x",
377 length);
378 return NULL;
381 TRACE("TLS request approved, setting up TLS");
382 tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
383 if (!tioc) {
384 return NULL;
386 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
387 TRACE("Starting TLS hanshake");
388 qio_channel_tls_handshake(tioc,
389 nbd_tls_handshake,
390 &data,
391 NULL);
393 if (!data.complete) {
394 g_main_loop_run(data.loop);
396 g_main_loop_unref(data.loop);
397 if (data.error) {
398 error_propagate(errp, data.error);
399 object_unref(OBJECT(tioc));
400 return NULL;
403 return QIO_CHANNEL(tioc);
407 int nbd_receive_negotiate(QIOChannel *ioc, const char *name, uint32_t *flags,
408 QCryptoTLSCreds *tlscreds, const char *hostname,
409 QIOChannel **outioc,
410 off_t *size, Error **errp)
412 char buf[256];
413 uint64_t magic, s;
414 int rc;
416 TRACE("Receiving negotiation tlscreds=%p hostname=%s.",
417 tlscreds, hostname ? hostname : "<null>");
419 rc = -EINVAL;
421 if (outioc) {
422 *outioc = NULL;
424 if (tlscreds && !outioc) {
425 error_setg(errp, "Output I/O channel required for TLS");
426 goto fail;
429 if (read_sync(ioc, buf, 8) != 8) {
430 error_setg(errp, "Failed to read data");
431 goto fail;
434 buf[8] = '\0';
435 if (strlen(buf) == 0) {
436 error_setg(errp, "Server connection closed unexpectedly");
437 goto fail;
440 TRACE("Magic is %c%c%c%c%c%c%c%c",
441 qemu_isprint(buf[0]) ? buf[0] : '.',
442 qemu_isprint(buf[1]) ? buf[1] : '.',
443 qemu_isprint(buf[2]) ? buf[2] : '.',
444 qemu_isprint(buf[3]) ? buf[3] : '.',
445 qemu_isprint(buf[4]) ? buf[4] : '.',
446 qemu_isprint(buf[5]) ? buf[5] : '.',
447 qemu_isprint(buf[6]) ? buf[6] : '.',
448 qemu_isprint(buf[7]) ? buf[7] : '.');
450 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
451 error_setg(errp, "Invalid magic received");
452 goto fail;
455 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
456 error_setg(errp, "Failed to read magic");
457 goto fail;
459 magic = be64_to_cpu(magic);
460 TRACE("Magic is 0x%" PRIx64, magic);
462 if (magic == NBD_OPTS_MAGIC) {
463 uint32_t clientflags = 0;
464 uint32_t opt;
465 uint32_t namesize;
466 uint16_t globalflags;
467 uint16_t exportflags;
468 bool fixedNewStyle = false;
470 if (read_sync(ioc, &globalflags, sizeof(globalflags)) !=
471 sizeof(globalflags)) {
472 error_setg(errp, "Failed to read server flags");
473 goto fail;
475 globalflags = be16_to_cpu(globalflags);
476 *flags = globalflags << 16;
477 TRACE("Global flags are %x", globalflags);
478 if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
479 fixedNewStyle = true;
480 TRACE("Server supports fixed new style");
481 clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
483 /* client requested flags */
484 clientflags = cpu_to_be32(clientflags);
485 if (write_sync(ioc, &clientflags, sizeof(clientflags)) !=
486 sizeof(clientflags)) {
487 error_setg(errp, "Failed to send clientflags field");
488 goto fail;
490 if (tlscreds) {
491 if (fixedNewStyle) {
492 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
493 if (!*outioc) {
494 goto fail;
496 ioc = *outioc;
497 } else {
498 error_setg(errp, "Server does not support STARTTLS");
499 goto fail;
502 if (!name) {
503 TRACE("Using default NBD export name \"\"");
504 name = "";
506 if (fixedNewStyle) {
507 /* Check our desired export is present in the
508 * server export list. Since NBD_OPT_EXPORT_NAME
509 * cannot return an error message, running this
510 * query gives us good error reporting if the
511 * server required TLS
513 if (nbd_receive_query_exports(ioc, name, errp) < 0) {
514 goto fail;
517 /* write the export name */
518 magic = cpu_to_be64(magic);
519 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
520 error_setg(errp, "Failed to send export name magic");
521 goto fail;
523 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
524 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
525 error_setg(errp, "Failed to send export name option number");
526 goto fail;
528 namesize = cpu_to_be32(strlen(name));
529 if (write_sync(ioc, &namesize, sizeof(namesize)) !=
530 sizeof(namesize)) {
531 error_setg(errp, "Failed to send export name length");
532 goto fail;
534 if (write_sync(ioc, (char *)name, strlen(name)) != strlen(name)) {
535 error_setg(errp, "Failed to send export name");
536 goto fail;
539 if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
540 error_setg(errp, "Failed to read export length");
541 goto fail;
543 *size = be64_to_cpu(s);
544 TRACE("Size is %" PRIu64, *size);
546 if (read_sync(ioc, &exportflags, sizeof(exportflags)) !=
547 sizeof(exportflags)) {
548 error_setg(errp, "Failed to read export flags");
549 goto fail;
551 exportflags = be16_to_cpu(exportflags);
552 *flags |= exportflags;
553 TRACE("Export flags are %x", exportflags);
554 } else if (magic == NBD_CLIENT_MAGIC) {
555 if (name) {
556 error_setg(errp, "Server does not support export names");
557 goto fail;
559 if (tlscreds) {
560 error_setg(errp, "Server does not support STARTTLS");
561 goto fail;
564 if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
565 error_setg(errp, "Failed to read export length");
566 goto fail;
568 *size = be64_to_cpu(s);
569 TRACE("Size is %" PRIu64, *size);
571 if (read_sync(ioc, flags, sizeof(*flags)) != sizeof(*flags)) {
572 error_setg(errp, "Failed to read export flags");
573 goto fail;
575 *flags = be32_to_cpup(flags);
576 } else {
577 error_setg(errp, "Bad magic received");
578 goto fail;
581 if (read_sync(ioc, &buf, 124) != 124) {
582 error_setg(errp, "Failed to read reserved block");
583 goto fail;
585 rc = 0;
587 fail:
588 return rc;
591 #ifdef __linux__
592 int nbd_init(int fd, QIOChannelSocket *sioc, uint32_t flags, off_t size)
594 TRACE("Setting NBD socket");
596 if (ioctl(fd, NBD_SET_SOCK, sioc->fd) < 0) {
597 int serrno = errno;
598 LOG("Failed to set NBD socket");
599 return -serrno;
602 TRACE("Setting block size to %lu", (unsigned long)BDRV_SECTOR_SIZE);
604 if (ioctl(fd, NBD_SET_BLKSIZE, (size_t)BDRV_SECTOR_SIZE) < 0) {
605 int serrno = errno;
606 LOG("Failed setting NBD block size");
607 return -serrno;
610 TRACE("Setting size to %zd block(s)", (size_t)(size / BDRV_SECTOR_SIZE));
612 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, (size_t)(size / BDRV_SECTOR_SIZE)) < 0) {
613 int serrno = errno;
614 LOG("Failed setting size (in blocks)");
615 return -serrno;
618 if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) {
619 if (errno == ENOTTY) {
620 int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
621 TRACE("Setting readonly attribute");
623 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
624 int serrno = errno;
625 LOG("Failed setting read-only attribute");
626 return -serrno;
628 } else {
629 int serrno = errno;
630 LOG("Failed setting flags");
631 return -serrno;
635 TRACE("Negotiation ended");
637 return 0;
640 int nbd_client(int fd)
642 int ret;
643 int serrno;
645 TRACE("Doing NBD loop");
647 ret = ioctl(fd, NBD_DO_IT);
648 if (ret < 0 && errno == EPIPE) {
649 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
650 * the socket via NBD_DISCONNECT. We do not want to return 1 in
651 * that case.
653 ret = 0;
655 serrno = errno;
657 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
659 TRACE("Clearing NBD queue");
660 ioctl(fd, NBD_CLEAR_QUE);
662 TRACE("Clearing NBD socket");
663 ioctl(fd, NBD_CLEAR_SOCK);
665 errno = serrno;
666 return ret;
668 #else
669 int nbd_init(int fd, QIOChannelSocket *ioc, uint32_t flags, off_t size)
671 return -ENOTSUP;
674 int nbd_client(int fd)
676 return -ENOTSUP;
678 #endif
680 ssize_t nbd_send_request(QIOChannel *ioc, struct nbd_request *request)
682 uint8_t buf[NBD_REQUEST_SIZE];
683 ssize_t ret;
685 TRACE("Sending request to server: "
686 "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
687 request->from, request->len, request->handle, request->type);
689 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
690 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
691 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
692 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
693 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
695 ret = write_sync(ioc, buf, sizeof(buf));
696 if (ret < 0) {
697 return ret;
700 if (ret != sizeof(buf)) {
701 LOG("writing to socket failed");
702 return -EINVAL;
704 return 0;
707 ssize_t nbd_receive_reply(QIOChannel *ioc, struct nbd_reply *reply)
709 uint8_t buf[NBD_REPLY_SIZE];
710 uint32_t magic;
711 ssize_t ret;
713 ret = read_sync(ioc, buf, sizeof(buf));
714 if (ret < 0) {
715 return ret;
718 if (ret != sizeof(buf)) {
719 LOG("read failed");
720 return -EINVAL;
723 /* Reply
724 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
725 [ 4 .. 7] error (0 == no error)
726 [ 7 .. 15] handle
729 magic = be32_to_cpup((uint32_t*)buf);
730 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
731 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
733 reply->error = nbd_errno_to_system_errno(reply->error);
735 TRACE("Got reply: "
736 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
737 magic, reply->error, reply->handle);
739 if (magic != NBD_REPLY_MAGIC) {
740 LOG("invalid magic (got 0x%x)", magic);
741 return -EINVAL;
743 return 0;