]> git.proxmox.com Git - mirror_qemu.git/blame_incremental - nbd/client.c
qemu-nbd: Fail earlier for -c/-d on non-linux
[mirror_qemu.git] / nbd / client.c
... / ...
CommitLineData
1/*
2 * Copyright (C) 2016-2018 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 *
5 * Network Block Device Client Side
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; under version 2 of the License.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
18 */
19
20#include "qemu/osdep.h"
21#include "qapi/error.h"
22#include "trace.h"
23#include "nbd-internal.h"
24
25/* Definitions for opaque data types */
26
27static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
28
29/* That's all folks */
30
31/* Basic flow for negotiation
32
33 Server Client
34 Negotiate
35
36 or
37
38 Server Client
39 Negotiate #1
40 Option
41 Negotiate #2
42
43 ----
44
45 followed by
46
47 Server Client
48 Request
49 Response
50 Request
51 Response
52 ...
53 ...
54 Request (type == 2)
55
56*/
57
58/* Send an option request.
59 *
60 * The request is for option @opt, with @data containing @len bytes of
61 * additional payload for the request (@len may be -1 to treat @data as
62 * a C string; and @data may be NULL if @len is 0).
63 * Return 0 if successful, -1 with errp set if it is impossible to
64 * continue. */
65static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
66 uint32_t len, const char *data,
67 Error **errp)
68{
69 NBDOption req;
70 QEMU_BUILD_BUG_ON(sizeof(req) != 16);
71
72 if (len == -1) {
73 req.length = len = strlen(data);
74 }
75 trace_nbd_send_option_request(opt, nbd_opt_lookup(opt), len);
76
77 stq_be_p(&req.magic, NBD_OPTS_MAGIC);
78 stl_be_p(&req.option, opt);
79 stl_be_p(&req.length, len);
80
81 if (nbd_write(ioc, &req, sizeof(req), errp) < 0) {
82 error_prepend(errp, "Failed to send option request header: ");
83 return -1;
84 }
85
86 if (len && nbd_write(ioc, (char *) data, len, errp) < 0) {
87 error_prepend(errp, "Failed to send option request data: ");
88 return -1;
89 }
90
91 return 0;
92}
93
94/* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
95 * not going to attempt further negotiation. */
96static void nbd_send_opt_abort(QIOChannel *ioc)
97{
98 /* Technically, a compliant server is supposed to reply to us; but
99 * older servers disconnected instead. At any rate, we're allowed
100 * to disconnect without waiting for the server reply, so we don't
101 * even care if the request makes it to the server, let alone
102 * waiting around for whether the server replies. */
103 nbd_send_option_request(ioc, NBD_OPT_ABORT, 0, NULL, NULL);
104}
105
106
107/* Receive the header of an option reply, which should match the given
108 * opt. Read through the length field, but NOT the length bytes of
109 * payload. Return 0 if successful, -1 with errp set if it is
110 * impossible to continue. */
111static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
112 NBDOptionReply *reply, Error **errp)
113{
114 QEMU_BUILD_BUG_ON(sizeof(*reply) != 20);
115 if (nbd_read(ioc, reply, sizeof(*reply), errp) < 0) {
116 error_prepend(errp, "failed to read option reply: ");
117 nbd_send_opt_abort(ioc);
118 return -1;
119 }
120 reply->magic = be64_to_cpu(reply->magic);
121 reply->option = be32_to_cpu(reply->option);
122 reply->type = be32_to_cpu(reply->type);
123 reply->length = be32_to_cpu(reply->length);
124
125 trace_nbd_receive_option_reply(reply->option, nbd_opt_lookup(reply->option),
126 reply->type, nbd_rep_lookup(reply->type),
127 reply->length);
128
129 if (reply->magic != NBD_REP_MAGIC) {
130 error_setg(errp, "Unexpected option reply magic");
131 nbd_send_opt_abort(ioc);
132 return -1;
133 }
134 if (reply->option != opt) {
135 error_setg(errp, "Unexpected option type %u (%s), expected %u (%s)",
136 reply->option, nbd_opt_lookup(reply->option),
137 opt, nbd_opt_lookup(opt));
138 nbd_send_opt_abort(ioc);
139 return -1;
140 }
141 return 0;
142}
143
144/* If reply represents success, return 1 without further action.
145 * If reply represents an error, consume the optional payload of
146 * the packet on ioc. Then return 0 for unsupported (so the client
147 * can fall back to other approaches), or -1 with errp set for other
148 * errors.
149 */
150static int nbd_handle_reply_err(QIOChannel *ioc, NBDOptionReply *reply,
151 Error **errp)
152{
153 char *msg = NULL;
154 int result = -1;
155
156 if (!(reply->type & (1 << 31))) {
157 return 1;
158 }
159
160 if (reply->length) {
161 if (reply->length > NBD_MAX_BUFFER_SIZE) {
162 error_setg(errp, "server error %" PRIu32
163 " (%s) message is too long",
164 reply->type, nbd_rep_lookup(reply->type));
165 goto cleanup;
166 }
167 msg = g_malloc(reply->length + 1);
168 if (nbd_read(ioc, msg, reply->length, errp) < 0) {
169 error_prepend(errp, "failed to read option error %" PRIu32
170 " (%s) message: ",
171 reply->type, nbd_rep_lookup(reply->type));
172 goto cleanup;
173 }
174 msg[reply->length] = '\0';
175 trace_nbd_server_error_msg(reply->type,
176 nbd_reply_type_lookup(reply->type), msg);
177 }
178
179 switch (reply->type) {
180 case NBD_REP_ERR_UNSUP:
181 trace_nbd_reply_err_unsup(reply->option, nbd_opt_lookup(reply->option));
182 result = 0;
183 goto cleanup;
184
185 case NBD_REP_ERR_POLICY:
186 error_setg(errp, "Denied by server for option %" PRIu32 " (%s)",
187 reply->option, nbd_opt_lookup(reply->option));
188 break;
189
190 case NBD_REP_ERR_INVALID:
191 error_setg(errp, "Invalid parameters for option %" PRIu32 " (%s)",
192 reply->option, nbd_opt_lookup(reply->option));
193 break;
194
195 case NBD_REP_ERR_PLATFORM:
196 error_setg(errp, "Server lacks support for option %" PRIu32 " (%s)",
197 reply->option, nbd_opt_lookup(reply->option));
198 break;
199
200 case NBD_REP_ERR_TLS_REQD:
201 error_setg(errp, "TLS negotiation required before option %" PRIu32
202 " (%s)", reply->option, nbd_opt_lookup(reply->option));
203 break;
204
205 case NBD_REP_ERR_UNKNOWN:
206 error_setg(errp, "Requested export not available");
207 break;
208
209 case NBD_REP_ERR_SHUTDOWN:
210 error_setg(errp, "Server shutting down before option %" PRIu32 " (%s)",
211 reply->option, nbd_opt_lookup(reply->option));
212 break;
213
214 case NBD_REP_ERR_BLOCK_SIZE_REQD:
215 error_setg(errp, "Server requires INFO_BLOCK_SIZE for option %" PRIu32
216 " (%s)", reply->option, nbd_opt_lookup(reply->option));
217 break;
218
219 default:
220 error_setg(errp, "Unknown error code when asking for option %" PRIu32
221 " (%s)", reply->option, nbd_opt_lookup(reply->option));
222 break;
223 }
224
225 if (msg) {
226 error_append_hint(errp, "server reported: %s\n", msg);
227 }
228
229 cleanup:
230 g_free(msg);
231 if (result < 0) {
232 nbd_send_opt_abort(ioc);
233 }
234 return result;
235}
236
237/* Process another portion of the NBD_OPT_LIST reply. Set *@match if
238 * the current reply matches @want or if the server does not support
239 * NBD_OPT_LIST, otherwise leave @match alone. Return 0 if iteration
240 * is complete, positive if more replies are expected, or negative
241 * with @errp set if an unrecoverable error occurred. */
242static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
243 Error **errp)
244{
245 NBDOptionReply reply;
246 uint32_t len;
247 uint32_t namelen;
248 char name[NBD_MAX_NAME_SIZE + 1];
249 int error;
250
251 if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
252 return -1;
253 }
254 error = nbd_handle_reply_err(ioc, &reply, errp);
255 if (error <= 0) {
256 /* The server did not support NBD_OPT_LIST, so set *match on
257 * the assumption that any name will be accepted. */
258 *match = true;
259 return error;
260 }
261 len = reply.length;
262
263 if (reply.type == NBD_REP_ACK) {
264 if (len != 0) {
265 error_setg(errp, "length too long for option end");
266 nbd_send_opt_abort(ioc);
267 return -1;
268 }
269 return 0;
270 } else if (reply.type != NBD_REP_SERVER) {
271 error_setg(errp, "Unexpected reply type %u (%s), expected %u (%s)",
272 reply.type, nbd_rep_lookup(reply.type),
273 NBD_REP_SERVER, nbd_rep_lookup(NBD_REP_SERVER));
274 nbd_send_opt_abort(ioc);
275 return -1;
276 }
277
278 if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
279 error_setg(errp, "incorrect option length %" PRIu32, len);
280 nbd_send_opt_abort(ioc);
281 return -1;
282 }
283 if (nbd_read(ioc, &namelen, sizeof(namelen), errp) < 0) {
284 error_prepend(errp, "failed to read option name length: ");
285 nbd_send_opt_abort(ioc);
286 return -1;
287 }
288 namelen = be32_to_cpu(namelen);
289 len -= sizeof(namelen);
290 if (len < namelen) {
291 error_setg(errp, "incorrect option name length");
292 nbd_send_opt_abort(ioc);
293 return -1;
294 }
295 if (namelen != strlen(want)) {
296 if (nbd_drop(ioc, len, errp) < 0) {
297 error_prepend(errp,
298 "failed to skip export name with wrong length: ");
299 nbd_send_opt_abort(ioc);
300 return -1;
301 }
302 return 1;
303 }
304
305 assert(namelen < sizeof(name));
306 if (nbd_read(ioc, name, namelen, errp) < 0) {
307 error_prepend(errp, "failed to read export name: ");
308 nbd_send_opt_abort(ioc);
309 return -1;
310 }
311 name[namelen] = '\0';
312 len -= namelen;
313 if (nbd_drop(ioc, len, errp) < 0) {
314 error_prepend(errp, "failed to read export description: ");
315 nbd_send_opt_abort(ioc);
316 return -1;
317 }
318 if (!strcmp(name, want)) {
319 *match = true;
320 }
321 return 1;
322}
323
324
325/* Returns -1 if NBD_OPT_GO proves the export @wantname cannot be
326 * used, 0 if NBD_OPT_GO is unsupported (fall back to NBD_OPT_LIST and
327 * NBD_OPT_EXPORT_NAME in that case), and > 0 if the export is good to
328 * go (with @info populated). */
329static int nbd_opt_go(QIOChannel *ioc, const char *wantname,
330 NBDExportInfo *info, Error **errp)
331{
332 NBDOptionReply reply;
333 uint32_t len = strlen(wantname);
334 uint16_t type;
335 int error;
336 char *buf;
337
338 /* The protocol requires that the server send NBD_INFO_EXPORT with
339 * a non-zero flags (at least NBD_FLAG_HAS_FLAGS must be set); so
340 * flags still 0 is a witness of a broken server. */
341 info->flags = 0;
342
343 trace_nbd_opt_go_start(wantname);
344 buf = g_malloc(4 + len + 2 + 2 * info->request_sizes + 1);
345 stl_be_p(buf, len);
346 memcpy(buf + 4, wantname, len);
347 /* At most one request, everything else up to server */
348 stw_be_p(buf + 4 + len, info->request_sizes);
349 if (info->request_sizes) {
350 stw_be_p(buf + 4 + len + 2, NBD_INFO_BLOCK_SIZE);
351 }
352 error = nbd_send_option_request(ioc, NBD_OPT_GO,
353 4 + len + 2 + 2 * info->request_sizes,
354 buf, errp);
355 g_free(buf);
356 if (error < 0) {
357 return -1;
358 }
359
360 while (1) {
361 if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
362 return -1;
363 }
364 error = nbd_handle_reply_err(ioc, &reply, errp);
365 if (error <= 0) {
366 return error;
367 }
368 len = reply.length;
369
370 if (reply.type == NBD_REP_ACK) {
371 /* Server is done sending info and moved into transmission
372 phase, but make sure it sent flags */
373 if (len) {
374 error_setg(errp, "server sent invalid NBD_REP_ACK");
375 return -1;
376 }
377 if (!info->flags) {
378 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
379 return -1;
380 }
381 trace_nbd_opt_go_success();
382 return 1;
383 }
384 if (reply.type != NBD_REP_INFO) {
385 error_setg(errp, "unexpected reply type %u (%s), expected %u (%s)",
386 reply.type, nbd_rep_lookup(reply.type),
387 NBD_REP_INFO, nbd_rep_lookup(NBD_REP_INFO));
388 nbd_send_opt_abort(ioc);
389 return -1;
390 }
391 if (len < sizeof(type)) {
392 error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
393 len);
394 nbd_send_opt_abort(ioc);
395 return -1;
396 }
397 if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
398 error_prepend(errp, "failed to read info type: ");
399 nbd_send_opt_abort(ioc);
400 return -1;
401 }
402 len -= sizeof(type);
403 type = be16_to_cpu(type);
404 switch (type) {
405 case NBD_INFO_EXPORT:
406 if (len != sizeof(info->size) + sizeof(info->flags)) {
407 error_setg(errp, "remaining export info len %" PRIu32
408 " is unexpected size", len);
409 nbd_send_opt_abort(ioc);
410 return -1;
411 }
412 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
413 error_prepend(errp, "failed to read info size: ");
414 nbd_send_opt_abort(ioc);
415 return -1;
416 }
417 info->size = be64_to_cpu(info->size);
418 if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
419 error_prepend(errp, "failed to read info flags: ");
420 nbd_send_opt_abort(ioc);
421 return -1;
422 }
423 info->flags = be16_to_cpu(info->flags);
424 trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
425 break;
426
427 case NBD_INFO_BLOCK_SIZE:
428 if (len != sizeof(info->min_block) * 3) {
429 error_setg(errp, "remaining export info len %" PRIu32
430 " is unexpected size", len);
431 nbd_send_opt_abort(ioc);
432 return -1;
433 }
434 if (nbd_read(ioc, &info->min_block, sizeof(info->min_block),
435 errp) < 0) {
436 error_prepend(errp, "failed to read info minimum block size: ");
437 nbd_send_opt_abort(ioc);
438 return -1;
439 }
440 info->min_block = be32_to_cpu(info->min_block);
441 if (!is_power_of_2(info->min_block)) {
442 error_setg(errp, "server minimum block size %" PRIu32
443 " is not a power of two", info->min_block);
444 nbd_send_opt_abort(ioc);
445 return -1;
446 }
447 if (nbd_read(ioc, &info->opt_block, sizeof(info->opt_block),
448 errp) < 0) {
449 error_prepend(errp,
450 "failed to read info preferred block size: ");
451 nbd_send_opt_abort(ioc);
452 return -1;
453 }
454 info->opt_block = be32_to_cpu(info->opt_block);
455 if (!is_power_of_2(info->opt_block) ||
456 info->opt_block < info->min_block) {
457 error_setg(errp, "server preferred block size %" PRIu32
458 " is not valid", info->opt_block);
459 nbd_send_opt_abort(ioc);
460 return -1;
461 }
462 if (nbd_read(ioc, &info->max_block, sizeof(info->max_block),
463 errp) < 0) {
464 error_prepend(errp, "failed to read info maximum block size: ");
465 nbd_send_opt_abort(ioc);
466 return -1;
467 }
468 info->max_block = be32_to_cpu(info->max_block);
469 if (info->max_block < info->min_block) {
470 error_setg(errp, "server maximum block size %" PRIu32
471 " is not valid", info->max_block);
472 nbd_send_opt_abort(ioc);
473 return -1;
474 }
475 trace_nbd_opt_go_info_block_size(info->min_block, info->opt_block,
476 info->max_block);
477 break;
478
479 default:
480 trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
481 if (nbd_drop(ioc, len, errp) < 0) {
482 error_prepend(errp, "Failed to read info payload: ");
483 nbd_send_opt_abort(ioc);
484 return -1;
485 }
486 break;
487 }
488 }
489}
490
491/* Return -1 on failure, 0 if wantname is an available export. */
492static int nbd_receive_query_exports(QIOChannel *ioc,
493 const char *wantname,
494 Error **errp)
495{
496 bool foundExport = false;
497
498 trace_nbd_receive_query_exports_start(wantname);
499 if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
500 return -1;
501 }
502
503 while (1) {
504 int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
505
506 if (ret < 0) {
507 /* Server gave unexpected reply */
508 return -1;
509 } else if (ret == 0) {
510 /* Done iterating. */
511 if (!foundExport) {
512 error_setg(errp, "No export with name '%s' available",
513 wantname);
514 nbd_send_opt_abort(ioc);
515 return -1;
516 }
517 trace_nbd_receive_query_exports_success(wantname);
518 return 0;
519 }
520 }
521}
522
523/* nbd_request_simple_option: Send an option request, and parse the reply
524 * return 1 for successful negotiation,
525 * 0 if operation is unsupported,
526 * -1 with errp set for any other error
527 */
528static int nbd_request_simple_option(QIOChannel *ioc, int opt, Error **errp)
529{
530 NBDOptionReply reply;
531 int error;
532
533 if (nbd_send_option_request(ioc, opt, 0, NULL, errp) < 0) {
534 return -1;
535 }
536
537 if (nbd_receive_option_reply(ioc, opt, &reply, errp) < 0) {
538 return -1;
539 }
540 error = nbd_handle_reply_err(ioc, &reply, errp);
541 if (error <= 0) {
542 return error;
543 }
544
545 if (reply.type != NBD_REP_ACK) {
546 error_setg(errp, "Server answered option %d (%s) with unexpected "
547 "reply %" PRIu32 " (%s)", opt, nbd_opt_lookup(opt),
548 reply.type, nbd_rep_lookup(reply.type));
549 nbd_send_opt_abort(ioc);
550 return -1;
551 }
552
553 if (reply.length != 0) {
554 error_setg(errp, "Option %d ('%s') response length is %" PRIu32
555 " (it should be zero)", opt, nbd_opt_lookup(opt),
556 reply.length);
557 nbd_send_opt_abort(ioc);
558 return -1;
559 }
560
561 return 1;
562}
563
564static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
565 QCryptoTLSCreds *tlscreds,
566 const char *hostname, Error **errp)
567{
568 int ret;
569 QIOChannelTLS *tioc;
570 struct NBDTLSHandshakeData data = { 0 };
571
572 ret = nbd_request_simple_option(ioc, NBD_OPT_STARTTLS, errp);
573 if (ret <= 0) {
574 if (ret == 0) {
575 error_setg(errp, "Server don't support STARTTLS option");
576 nbd_send_opt_abort(ioc);
577 }
578 return NULL;
579 }
580
581 trace_nbd_receive_starttls_new_client();
582 tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
583 if (!tioc) {
584 return NULL;
585 }
586 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
587 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
588 trace_nbd_receive_starttls_tls_handshake();
589 qio_channel_tls_handshake(tioc,
590 nbd_tls_handshake,
591 &data,
592 NULL,
593 NULL);
594
595 if (!data.complete) {
596 g_main_loop_run(data.loop);
597 }
598 g_main_loop_unref(data.loop);
599 if (data.error) {
600 error_propagate(errp, data.error);
601 object_unref(OBJECT(tioc));
602 return NULL;
603 }
604
605 return QIO_CHANNEL(tioc);
606}
607
608/* nbd_negotiate_simple_meta_context:
609 * Set one meta context. Simple means that reply must contain zero (not
610 * negotiated) or one (negotiated) contexts. More contexts would be considered
611 * as a protocol error. It's also implied that meta-data query equals queried
612 * context name, so, if server replies with something different than @context,
613 * it is considered an error too.
614 * return 1 for successful negotiation, context_id is set
615 * 0 if operation is unsupported,
616 * -1 with errp set for any other error
617 */
618static int nbd_negotiate_simple_meta_context(QIOChannel *ioc,
619 const char *export,
620 const char *context,
621 uint32_t *context_id,
622 Error **errp)
623{
624 int ret;
625 NBDOptionReply reply;
626 uint32_t received_id = 0;
627 bool received = false;
628 uint32_t export_len = strlen(export);
629 uint32_t context_len = strlen(context);
630 uint32_t data_len = sizeof(export_len) + export_len +
631 sizeof(uint32_t) + /* number of queries */
632 sizeof(context_len) + context_len;
633 char *data = g_malloc(data_len);
634 char *p = data;
635
636 trace_nbd_opt_meta_request(context, export);
637 stl_be_p(p, export_len);
638 memcpy(p += sizeof(export_len), export, export_len);
639 stl_be_p(p += export_len, 1);
640 stl_be_p(p += sizeof(uint32_t), context_len);
641 memcpy(p += sizeof(context_len), context, context_len);
642
643 ret = nbd_send_option_request(ioc, NBD_OPT_SET_META_CONTEXT, data_len, data,
644 errp);
645 g_free(data);
646 if (ret < 0) {
647 return ret;
648 }
649
650 if (nbd_receive_option_reply(ioc, NBD_OPT_SET_META_CONTEXT, &reply,
651 errp) < 0)
652 {
653 return -1;
654 }
655
656 ret = nbd_handle_reply_err(ioc, &reply, errp);
657 if (ret <= 0) {
658 return ret;
659 }
660
661 if (reply.type == NBD_REP_META_CONTEXT) {
662 char *name;
663
664 if (reply.length != sizeof(received_id) + context_len) {
665 error_setg(errp, "Failed to negotiate meta context '%s', server "
666 "answered with unexpected length %" PRIu32, context,
667 reply.length);
668 nbd_send_opt_abort(ioc);
669 return -1;
670 }
671
672 if (nbd_read(ioc, &received_id, sizeof(received_id), errp) < 0) {
673 return -1;
674 }
675 received_id = be32_to_cpu(received_id);
676
677 reply.length -= sizeof(received_id);
678 name = g_malloc(reply.length + 1);
679 if (nbd_read(ioc, name, reply.length, errp) < 0) {
680 g_free(name);
681 return -1;
682 }
683 name[reply.length] = '\0';
684 if (strcmp(context, name)) {
685 error_setg(errp, "Failed to negotiate meta context '%s', server "
686 "answered with different context '%s'", context,
687 name);
688 g_free(name);
689 nbd_send_opt_abort(ioc);
690 return -1;
691 }
692 g_free(name);
693
694 trace_nbd_opt_meta_reply(context, received_id);
695 received = true;
696
697 /* receive NBD_REP_ACK */
698 if (nbd_receive_option_reply(ioc, NBD_OPT_SET_META_CONTEXT, &reply,
699 errp) < 0)
700 {
701 return -1;
702 }
703
704 ret = nbd_handle_reply_err(ioc, &reply, errp);
705 if (ret <= 0) {
706 return ret;
707 }
708 }
709
710 if (reply.type != NBD_REP_ACK) {
711 error_setg(errp, "Unexpected reply type %u (%s), expected %u (%s)",
712 reply.type, nbd_rep_lookup(reply.type),
713 NBD_REP_ACK, nbd_rep_lookup(NBD_REP_ACK));
714 nbd_send_opt_abort(ioc);
715 return -1;
716 }
717 if (reply.length) {
718 error_setg(errp, "Unexpected length to ACK response");
719 nbd_send_opt_abort(ioc);
720 return -1;
721 }
722
723 if (received) {
724 *context_id = received_id;
725 return 1;
726 }
727
728 return 0;
729}
730
731int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
732 QCryptoTLSCreds *tlscreds, const char *hostname,
733 QIOChannel **outioc, NBDExportInfo *info,
734 Error **errp)
735{
736 char buf[256];
737 uint64_t magic;
738 int rc;
739 bool zeroes = true;
740 bool structured_reply = info->structured_reply;
741 bool base_allocation = info->base_allocation;
742
743 trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
744
745 info->structured_reply = false;
746 info->base_allocation = false;
747 rc = -EINVAL;
748
749 if (outioc) {
750 *outioc = NULL;
751 }
752 if (tlscreds && !outioc) {
753 error_setg(errp, "Output I/O channel required for TLS");
754 goto fail;
755 }
756
757 if (nbd_read(ioc, buf, 8, errp) < 0) {
758 error_prepend(errp, "Failed to read data: ");
759 goto fail;
760 }
761
762 buf[8] = '\0';
763 if (strlen(buf) == 0) {
764 error_setg(errp, "Server connection closed unexpectedly");
765 goto fail;
766 }
767
768 magic = ldq_be_p(buf);
769 trace_nbd_receive_negotiate_magic(magic);
770
771 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
772 error_setg(errp, "Invalid magic received");
773 goto fail;
774 }
775
776 if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
777 error_prepend(errp, "Failed to read magic: ");
778 goto fail;
779 }
780 magic = be64_to_cpu(magic);
781 trace_nbd_receive_negotiate_magic(magic);
782
783 if (magic == NBD_OPTS_MAGIC) {
784 uint32_t clientflags = 0;
785 uint16_t globalflags;
786 bool fixedNewStyle = false;
787
788 if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
789 error_prepend(errp, "Failed to read server flags: ");
790 goto fail;
791 }
792 globalflags = be16_to_cpu(globalflags);
793 trace_nbd_receive_negotiate_server_flags(globalflags);
794 if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
795 fixedNewStyle = true;
796 clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
797 }
798 if (globalflags & NBD_FLAG_NO_ZEROES) {
799 zeroes = false;
800 clientflags |= NBD_FLAG_C_NO_ZEROES;
801 }
802 /* client requested flags */
803 clientflags = cpu_to_be32(clientflags);
804 if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
805 error_prepend(errp, "Failed to send clientflags field: ");
806 goto fail;
807 }
808 if (tlscreds) {
809 if (fixedNewStyle) {
810 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
811 if (!*outioc) {
812 goto fail;
813 }
814 ioc = *outioc;
815 } else {
816 error_setg(errp, "Server does not support STARTTLS");
817 goto fail;
818 }
819 }
820 if (!name) {
821 trace_nbd_receive_negotiate_default_name();
822 name = "";
823 }
824 if (fixedNewStyle) {
825 int result;
826
827 if (structured_reply) {
828 result = nbd_request_simple_option(ioc,
829 NBD_OPT_STRUCTURED_REPLY,
830 errp);
831 if (result < 0) {
832 goto fail;
833 }
834 info->structured_reply = result == 1;
835 }
836
837 if (info->structured_reply && base_allocation) {
838 result = nbd_negotiate_simple_meta_context(
839 ioc, name, info->x_dirty_bitmap ?: "base:allocation",
840 &info->meta_base_allocation_id, errp);
841 if (result < 0) {
842 goto fail;
843 }
844 info->base_allocation = result == 1;
845 }
846
847 /* Try NBD_OPT_GO first - if it works, we are done (it
848 * also gives us a good message if the server requires
849 * TLS). If it is not available, fall back to
850 * NBD_OPT_LIST for nicer error messages about a missing
851 * export, then use NBD_OPT_EXPORT_NAME. */
852 result = nbd_opt_go(ioc, name, info, errp);
853 if (result < 0) {
854 goto fail;
855 }
856 if (result > 0) {
857 return 0;
858 }
859 /* Check our desired export is present in the
860 * server export list. Since NBD_OPT_EXPORT_NAME
861 * cannot return an error message, running this
862 * query gives us better error reporting if the
863 * export name is not available.
864 */
865 if (nbd_receive_query_exports(ioc, name, errp) < 0) {
866 goto fail;
867 }
868 }
869 /* write the export name request */
870 if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
871 errp) < 0) {
872 goto fail;
873 }
874
875 /* Read the response */
876 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
877 error_prepend(errp, "Failed to read export length: ");
878 goto fail;
879 }
880 info->size = be64_to_cpu(info->size);
881
882 if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
883 error_prepend(errp, "Failed to read export flags: ");
884 goto fail;
885 }
886 info->flags = be16_to_cpu(info->flags);
887 } else if (magic == NBD_CLIENT_MAGIC) {
888 uint32_t oldflags;
889
890 if (name) {
891 error_setg(errp, "Server does not support export names");
892 goto fail;
893 }
894 if (tlscreds) {
895 error_setg(errp, "Server does not support STARTTLS");
896 goto fail;
897 }
898
899 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
900 error_prepend(errp, "Failed to read export length: ");
901 goto fail;
902 }
903 info->size = be64_to_cpu(info->size);
904
905 if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
906 error_prepend(errp, "Failed to read export flags: ");
907 goto fail;
908 }
909 oldflags = be32_to_cpu(oldflags);
910 if (oldflags & ~0xffff) {
911 error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
912 goto fail;
913 }
914 info->flags = oldflags;
915 } else {
916 error_setg(errp, "Bad magic received");
917 goto fail;
918 }
919
920 trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
921 if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
922 error_prepend(errp, "Failed to read reserved block: ");
923 goto fail;
924 }
925 rc = 0;
926
927fail:
928 return rc;
929}
930
931#ifdef __linux__
932int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
933 Error **errp)
934{
935 unsigned long sector_size = MAX(BDRV_SECTOR_SIZE, info->min_block);
936 unsigned long sectors = info->size / sector_size;
937
938 /* FIXME: Once the kernel module is patched to honor block sizes,
939 * and to advertise that fact to user space, we should update the
940 * hand-off to the kernel to use any block sizes we learned. */
941 assert(!info->request_sizes);
942 if (info->size / sector_size != sectors) {
943 error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
944 info->size);
945 return -E2BIG;
946 }
947
948 trace_nbd_init_set_socket();
949
950 if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
951 int serrno = errno;
952 error_setg(errp, "Failed to set NBD socket");
953 return -serrno;
954 }
955
956 trace_nbd_init_set_block_size(sector_size);
957
958 if (ioctl(fd, NBD_SET_BLKSIZE, sector_size) < 0) {
959 int serrno = errno;
960 error_setg(errp, "Failed setting NBD block size");
961 return -serrno;
962 }
963
964 trace_nbd_init_set_size(sectors);
965 if (info->size % sector_size) {
966 trace_nbd_init_trailing_bytes(info->size % sector_size);
967 }
968
969 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
970 int serrno = errno;
971 error_setg(errp, "Failed setting size (in blocks)");
972 return -serrno;
973 }
974
975 if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
976 if (errno == ENOTTY) {
977 int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
978 trace_nbd_init_set_readonly();
979
980 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
981 int serrno = errno;
982 error_setg(errp, "Failed setting read-only attribute");
983 return -serrno;
984 }
985 } else {
986 int serrno = errno;
987 error_setg(errp, "Failed setting flags");
988 return -serrno;
989 }
990 }
991
992 trace_nbd_init_finish();
993
994 return 0;
995}
996
997int nbd_client(int fd)
998{
999 int ret;
1000 int serrno;
1001
1002 trace_nbd_client_loop();
1003
1004 ret = ioctl(fd, NBD_DO_IT);
1005 if (ret < 0 && errno == EPIPE) {
1006 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
1007 * the socket via NBD_DISCONNECT. We do not want to return 1 in
1008 * that case.
1009 */
1010 ret = 0;
1011 }
1012 serrno = errno;
1013
1014 trace_nbd_client_loop_ret(ret, strerror(serrno));
1015
1016 trace_nbd_client_clear_queue();
1017 ioctl(fd, NBD_CLEAR_QUE);
1018
1019 trace_nbd_client_clear_socket();
1020 ioctl(fd, NBD_CLEAR_SOCK);
1021
1022 errno = serrno;
1023 return ret;
1024}
1025
1026int nbd_disconnect(int fd)
1027{
1028 ioctl(fd, NBD_CLEAR_QUE);
1029 ioctl(fd, NBD_DISCONNECT);
1030 ioctl(fd, NBD_CLEAR_SOCK);
1031 return 0;
1032}
1033
1034#endif /* __linux__ */
1035
1036int nbd_send_request(QIOChannel *ioc, NBDRequest *request)
1037{
1038 uint8_t buf[NBD_REQUEST_SIZE];
1039
1040 trace_nbd_send_request(request->from, request->len, request->handle,
1041 request->flags, request->type,
1042 nbd_cmd_lookup(request->type));
1043
1044 stl_be_p(buf, NBD_REQUEST_MAGIC);
1045 stw_be_p(buf + 4, request->flags);
1046 stw_be_p(buf + 6, request->type);
1047 stq_be_p(buf + 8, request->handle);
1048 stq_be_p(buf + 16, request->from);
1049 stl_be_p(buf + 24, request->len);
1050
1051 return nbd_write(ioc, buf, sizeof(buf), NULL);
1052}
1053
1054/* nbd_receive_simple_reply
1055 * Read simple reply except magic field (which should be already read).
1056 * Payload is not read (payload is possible for CMD_READ, but here we even
1057 * don't know whether it take place or not).
1058 */
1059static int nbd_receive_simple_reply(QIOChannel *ioc, NBDSimpleReply *reply,
1060 Error **errp)
1061{
1062 int ret;
1063
1064 assert(reply->magic == NBD_SIMPLE_REPLY_MAGIC);
1065
1066 ret = nbd_read(ioc, (uint8_t *)reply + sizeof(reply->magic),
1067 sizeof(*reply) - sizeof(reply->magic), errp);
1068 if (ret < 0) {
1069 return ret;
1070 }
1071
1072 reply->error = be32_to_cpu(reply->error);
1073 reply->handle = be64_to_cpu(reply->handle);
1074
1075 return 0;
1076}
1077
1078/* nbd_receive_structured_reply_chunk
1079 * Read structured reply chunk except magic field (which should be already
1080 * read).
1081 * Payload is not read.
1082 */
1083static int nbd_receive_structured_reply_chunk(QIOChannel *ioc,
1084 NBDStructuredReplyChunk *chunk,
1085 Error **errp)
1086{
1087 int ret;
1088
1089 assert(chunk->magic == NBD_STRUCTURED_REPLY_MAGIC);
1090
1091 ret = nbd_read(ioc, (uint8_t *)chunk + sizeof(chunk->magic),
1092 sizeof(*chunk) - sizeof(chunk->magic), errp);
1093 if (ret < 0) {
1094 return ret;
1095 }
1096
1097 chunk->flags = be16_to_cpu(chunk->flags);
1098 chunk->type = be16_to_cpu(chunk->type);
1099 chunk->handle = be64_to_cpu(chunk->handle);
1100 chunk->length = be32_to_cpu(chunk->length);
1101
1102 return 0;
1103}
1104
1105/* nbd_receive_reply
1106 * Returns 1 on success
1107 * 0 on eof, when no data was read (errp is not set)
1108 * negative errno on failure (errp is set)
1109 */
1110int nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
1111{
1112 int ret;
1113 const char *type;
1114
1115 ret = nbd_read_eof(ioc, &reply->magic, sizeof(reply->magic), errp);
1116 if (ret <= 0) {
1117 return ret;
1118 }
1119
1120 reply->magic = be32_to_cpu(reply->magic);
1121
1122 switch (reply->magic) {
1123 case NBD_SIMPLE_REPLY_MAGIC:
1124 ret = nbd_receive_simple_reply(ioc, &reply->simple, errp);
1125 if (ret < 0) {
1126 break;
1127 }
1128 trace_nbd_receive_simple_reply(reply->simple.error,
1129 nbd_err_lookup(reply->simple.error),
1130 reply->handle);
1131 break;
1132 case NBD_STRUCTURED_REPLY_MAGIC:
1133 ret = nbd_receive_structured_reply_chunk(ioc, &reply->structured, errp);
1134 if (ret < 0) {
1135 break;
1136 }
1137 type = nbd_reply_type_lookup(reply->structured.type);
1138 trace_nbd_receive_structured_reply_chunk(reply->structured.flags,
1139 reply->structured.type, type,
1140 reply->structured.handle,
1141 reply->structured.length);
1142 break;
1143 default:
1144 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", reply->magic);
1145 return -EINVAL;
1146 }
1147 if (ret < 0) {
1148 return ret;
1149 }
1150
1151 return 1;
1152}
1153