migration: stop compressing page in migration thread
[qemu/ar7.git] / nbd / client.c
blobb9e175d1c2724ebdc5bac75a98750e61c3278c76
1 /*
2 * Copyright (C) 2016-2017 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
5 * Network Block Device Client Side
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.
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.
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/>.
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "trace.h"
23 #include "nbd-internal.h"
25 /* Definitions for opaque data types */
27 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
29 /* That's all folks */
31 /* Basic flow for negotiation
33 Server Client
34 Negotiate
38 Server Client
39 Negotiate #1
40 Option
41 Negotiate #2
43 ----
45 followed by
47 Server Client
48 Request
49 Response
50 Request
51 Response
52 ...
53 ...
54 Request (type == 2)
58 /* Send an option request.
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. */
65 static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
66 uint32_t len, const char *data,
67 Error **errp)
69 NBDOption req;
70 QEMU_BUILD_BUG_ON(sizeof(req) != 16);
72 if (len == -1) {
73 req.length = len = strlen(data);
75 trace_nbd_send_option_request(opt, nbd_opt_lookup(opt), len);
77 stq_be_p(&req.magic, NBD_OPTS_MAGIC);
78 stl_be_p(&req.option, opt);
79 stl_be_p(&req.length, len);
81 if (nbd_write(ioc, &req, sizeof(req), errp) < 0) {
82 error_prepend(errp, "Failed to send option request header: ");
83 return -1;
86 if (len && nbd_write(ioc, (char *) data, len, errp) < 0) {
87 error_prepend(errp, "Failed to send option request data: ");
88 return -1;
91 return 0;
94 /* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
95 * not going to attempt further negotiation. */
96 static void nbd_send_opt_abort(QIOChannel *ioc)
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);
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. */
111 static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
112 NBDOptionReply *reply, Error **errp)
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;
120 be64_to_cpus(&reply->magic);
121 be32_to_cpus(&reply->option);
122 be32_to_cpus(&reply->type);
123 be32_to_cpus(&reply->length);
125 trace_nbd_receive_option_reply(reply->option, nbd_opt_lookup(reply->option),
126 reply->type, nbd_rep_lookup(reply->type),
127 reply->length);
129 if (reply->magic != NBD_REP_MAGIC) {
130 error_setg(errp, "Unexpected option reply magic");
131 nbd_send_opt_abort(ioc);
132 return -1;
134 if (reply->option != opt) {
135 error_setg(errp, "Unexpected option type %x expected %x",
136 reply->option, opt);
137 nbd_send_opt_abort(ioc);
138 return -1;
140 return 0;
143 /* If reply represents success, return 1 without further action.
144 * If reply represents an error, consume the optional payload of
145 * the packet on ioc. Then return 0 for unsupported (so the client
146 * can fall back to other approaches), or -1 with errp set for other
147 * errors.
149 static int nbd_handle_reply_err(QIOChannel *ioc, NBDOptionReply *reply,
150 Error **errp)
152 char *msg = NULL;
153 int result = -1;
155 if (!(reply->type & (1 << 31))) {
156 return 1;
159 if (reply->length) {
160 if (reply->length > NBD_MAX_BUFFER_SIZE) {
161 error_setg(errp, "server error %" PRIu32
162 " (%s) message is too long",
163 reply->type, nbd_rep_lookup(reply->type));
164 goto cleanup;
166 msg = g_malloc(reply->length + 1);
167 if (nbd_read(ioc, msg, reply->length, errp) < 0) {
168 error_prepend(errp, "failed to read option error %" PRIu32
169 " (%s) message: ",
170 reply->type, nbd_rep_lookup(reply->type));
171 goto cleanup;
173 msg[reply->length] = '\0';
176 switch (reply->type) {
177 case NBD_REP_ERR_UNSUP:
178 trace_nbd_reply_err_unsup(reply->option, nbd_opt_lookup(reply->option));
179 result = 0;
180 goto cleanup;
182 case NBD_REP_ERR_POLICY:
183 error_setg(errp, "Denied by server for option %" PRIu32 " (%s)",
184 reply->option, nbd_opt_lookup(reply->option));
185 break;
187 case NBD_REP_ERR_INVALID:
188 error_setg(errp, "Invalid parameters for option %" PRIu32 " (%s)",
189 reply->option, nbd_opt_lookup(reply->option));
190 break;
192 case NBD_REP_ERR_PLATFORM:
193 error_setg(errp, "Server lacks support for option %" PRIu32 " (%s)",
194 reply->option, nbd_opt_lookup(reply->option));
195 break;
197 case NBD_REP_ERR_TLS_REQD:
198 error_setg(errp, "TLS negotiation required before option %" PRIu32
199 " (%s)", reply->option, nbd_opt_lookup(reply->option));
200 break;
202 case NBD_REP_ERR_UNKNOWN:
203 error_setg(errp, "Requested export not available");
204 break;
206 case NBD_REP_ERR_SHUTDOWN:
207 error_setg(errp, "Server shutting down before option %" PRIu32 " (%s)",
208 reply->option, nbd_opt_lookup(reply->option));
209 break;
211 case NBD_REP_ERR_BLOCK_SIZE_REQD:
212 error_setg(errp, "Server requires INFO_BLOCK_SIZE for option %" PRIu32
213 " (%s)", reply->option, nbd_opt_lookup(reply->option));
214 break;
216 default:
217 error_setg(errp, "Unknown error code when asking for option %" PRIu32
218 " (%s)", reply->option, nbd_opt_lookup(reply->option));
219 break;
222 if (msg) {
223 error_append_hint(errp, "server reported: %s\n", msg);
226 cleanup:
227 g_free(msg);
228 if (result < 0) {
229 nbd_send_opt_abort(ioc);
231 return result;
234 /* Process another portion of the NBD_OPT_LIST reply. Set *@match if
235 * the current reply matches @want or if the server does not support
236 * NBD_OPT_LIST, otherwise leave @match alone. Return 0 if iteration
237 * is complete, positive if more replies are expected, or negative
238 * with @errp set if an unrecoverable error occurred. */
239 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
240 Error **errp)
242 NBDOptionReply reply;
243 uint32_t len;
244 uint32_t namelen;
245 char name[NBD_MAX_NAME_SIZE + 1];
246 int error;
248 if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
249 return -1;
251 error = nbd_handle_reply_err(ioc, &reply, errp);
252 if (error <= 0) {
253 /* The server did not support NBD_OPT_LIST, so set *match on
254 * the assumption that any name will be accepted. */
255 *match = true;
256 return error;
258 len = reply.length;
260 if (reply.type == NBD_REP_ACK) {
261 if (len != 0) {
262 error_setg(errp, "length too long for option end");
263 nbd_send_opt_abort(ioc);
264 return -1;
266 return 0;
267 } else if (reply.type != NBD_REP_SERVER) {
268 error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
269 reply.type, NBD_REP_SERVER);
270 nbd_send_opt_abort(ioc);
271 return -1;
274 if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
275 error_setg(errp, "incorrect option length %" PRIu32, len);
276 nbd_send_opt_abort(ioc);
277 return -1;
279 if (nbd_read(ioc, &namelen, sizeof(namelen), errp) < 0) {
280 error_prepend(errp, "failed to read option name length: ");
281 nbd_send_opt_abort(ioc);
282 return -1;
284 namelen = be32_to_cpu(namelen);
285 len -= sizeof(namelen);
286 if (len < namelen) {
287 error_setg(errp, "incorrect option name length");
288 nbd_send_opt_abort(ioc);
289 return -1;
291 if (namelen != strlen(want)) {
292 if (nbd_drop(ioc, len, errp) < 0) {
293 error_prepend(errp,
294 "failed to skip export name with wrong length: ");
295 nbd_send_opt_abort(ioc);
296 return -1;
298 return 1;
301 assert(namelen < sizeof(name));
302 if (nbd_read(ioc, name, namelen, errp) < 0) {
303 error_prepend(errp, "failed to read export name: ");
304 nbd_send_opt_abort(ioc);
305 return -1;
307 name[namelen] = '\0';
308 len -= namelen;
309 if (nbd_drop(ioc, len, errp) < 0) {
310 error_prepend(errp, "failed to read export description: ");
311 nbd_send_opt_abort(ioc);
312 return -1;
314 if (!strcmp(name, want)) {
315 *match = true;
317 return 1;
321 /* Returns -1 if NBD_OPT_GO proves the export @wantname cannot be
322 * used, 0 if NBD_OPT_GO is unsupported (fall back to NBD_OPT_LIST and
323 * NBD_OPT_EXPORT_NAME in that case), and > 0 if the export is good to
324 * go (with @info populated). */
325 static int nbd_opt_go(QIOChannel *ioc, const char *wantname,
326 NBDExportInfo *info, Error **errp)
328 NBDOptionReply reply;
329 uint32_t len = strlen(wantname);
330 uint16_t type;
331 int error;
332 char *buf;
334 /* The protocol requires that the server send NBD_INFO_EXPORT with
335 * a non-zero flags (at least NBD_FLAG_HAS_FLAGS must be set); so
336 * flags still 0 is a witness of a broken server. */
337 info->flags = 0;
339 trace_nbd_opt_go_start(wantname);
340 buf = g_malloc(4 + len + 2 + 2 * info->request_sizes + 1);
341 stl_be_p(buf, len);
342 memcpy(buf + 4, wantname, len);
343 /* At most one request, everything else up to server */
344 stw_be_p(buf + 4 + len, info->request_sizes);
345 if (info->request_sizes) {
346 stw_be_p(buf + 4 + len + 2, NBD_INFO_BLOCK_SIZE);
348 error = nbd_send_option_request(ioc, NBD_OPT_GO,
349 4 + len + 2 + 2 * info->request_sizes,
350 buf, errp);
351 g_free(buf);
352 if (error < 0) {
353 return -1;
356 while (1) {
357 if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
358 return -1;
360 error = nbd_handle_reply_err(ioc, &reply, errp);
361 if (error <= 0) {
362 return error;
364 len = reply.length;
366 if (reply.type == NBD_REP_ACK) {
367 /* Server is done sending info and moved into transmission
368 phase, but make sure it sent flags */
369 if (len) {
370 error_setg(errp, "server sent invalid NBD_REP_ACK");
371 return -1;
373 if (!info->flags) {
374 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
375 return -1;
377 trace_nbd_opt_go_success();
378 return 1;
380 if (reply.type != NBD_REP_INFO) {
381 error_setg(errp, "unexpected reply type %" PRIu32
382 " (%s), expected %u",
383 reply.type, nbd_rep_lookup(reply.type), NBD_REP_INFO);
384 nbd_send_opt_abort(ioc);
385 return -1;
387 if (len < sizeof(type)) {
388 error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
389 len);
390 nbd_send_opt_abort(ioc);
391 return -1;
393 if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
394 error_prepend(errp, "failed to read info type: ");
395 nbd_send_opt_abort(ioc);
396 return -1;
398 len -= sizeof(type);
399 be16_to_cpus(&type);
400 switch (type) {
401 case NBD_INFO_EXPORT:
402 if (len != sizeof(info->size) + sizeof(info->flags)) {
403 error_setg(errp, "remaining export info len %" PRIu32
404 " is unexpected size", len);
405 nbd_send_opt_abort(ioc);
406 return -1;
408 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
409 error_prepend(errp, "failed to read info size: ");
410 nbd_send_opt_abort(ioc);
411 return -1;
413 be64_to_cpus(&info->size);
414 if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
415 error_prepend(errp, "failed to read info flags: ");
416 nbd_send_opt_abort(ioc);
417 return -1;
419 be16_to_cpus(&info->flags);
420 trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
421 break;
423 case NBD_INFO_BLOCK_SIZE:
424 if (len != sizeof(info->min_block) * 3) {
425 error_setg(errp, "remaining export info len %" PRIu32
426 " is unexpected size", len);
427 nbd_send_opt_abort(ioc);
428 return -1;
430 if (nbd_read(ioc, &info->min_block, sizeof(info->min_block),
431 errp) < 0) {
432 error_prepend(errp, "failed to read info minimum block size: ");
433 nbd_send_opt_abort(ioc);
434 return -1;
436 be32_to_cpus(&info->min_block);
437 if (!is_power_of_2(info->min_block)) {
438 error_setg(errp, "server minimum block size %" PRId32
439 "is not a power of two", info->min_block);
440 nbd_send_opt_abort(ioc);
441 return -1;
443 if (nbd_read(ioc, &info->opt_block, sizeof(info->opt_block),
444 errp) < 0) {
445 error_prepend(errp,
446 "failed to read info preferred block size: ");
447 nbd_send_opt_abort(ioc);
448 return -1;
450 be32_to_cpus(&info->opt_block);
451 if (!is_power_of_2(info->opt_block) ||
452 info->opt_block < info->min_block) {
453 error_setg(errp, "server preferred block size %" PRId32
454 "is not valid", info->opt_block);
455 nbd_send_opt_abort(ioc);
456 return -1;
458 if (nbd_read(ioc, &info->max_block, sizeof(info->max_block),
459 errp) < 0) {
460 error_prepend(errp, "failed to read info maximum block size: ");
461 nbd_send_opt_abort(ioc);
462 return -1;
464 be32_to_cpus(&info->max_block);
465 trace_nbd_opt_go_info_block_size(info->min_block, info->opt_block,
466 info->max_block);
467 break;
469 default:
470 trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
471 if (nbd_drop(ioc, len, errp) < 0) {
472 error_prepend(errp, "Failed to read info payload: ");
473 nbd_send_opt_abort(ioc);
474 return -1;
476 break;
481 /* Return -1 on failure, 0 if wantname is an available export. */
482 static int nbd_receive_query_exports(QIOChannel *ioc,
483 const char *wantname,
484 Error **errp)
486 bool foundExport = false;
488 trace_nbd_receive_query_exports_start(wantname);
489 if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
490 return -1;
493 while (1) {
494 int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
496 if (ret < 0) {
497 /* Server gave unexpected reply */
498 return -1;
499 } else if (ret == 0) {
500 /* Done iterating. */
501 if (!foundExport) {
502 error_setg(errp, "No export with name '%s' available",
503 wantname);
504 nbd_send_opt_abort(ioc);
505 return -1;
507 trace_nbd_receive_query_exports_success(wantname);
508 return 0;
513 /* nbd_request_simple_option: Send an option request, and parse the reply
514 * return 1 for successful negotiation,
515 * 0 if operation is unsupported,
516 * -1 with errp set for any other error
518 static int nbd_request_simple_option(QIOChannel *ioc, int opt, Error **errp)
520 NBDOptionReply reply;
521 int error;
523 if (nbd_send_option_request(ioc, opt, 0, NULL, errp) < 0) {
524 return -1;
527 if (nbd_receive_option_reply(ioc, opt, &reply, errp) < 0) {
528 return -1;
530 error = nbd_handle_reply_err(ioc, &reply, errp);
531 if (error <= 0) {
532 return error;
535 if (reply.type != NBD_REP_ACK) {
536 error_setg(errp, "Server answered option %d (%s) with unexpected "
537 "reply %" PRIu32 " (%s)", opt, nbd_opt_lookup(opt),
538 reply.type, nbd_rep_lookup(reply.type));
539 nbd_send_opt_abort(ioc);
540 return -1;
543 if (reply.length != 0) {
544 error_setg(errp, "Option %d ('%s') response length is %" PRIu32
545 " (it should be zero)", opt, nbd_opt_lookup(opt),
546 reply.length);
547 nbd_send_opt_abort(ioc);
548 return -1;
551 return 1;
554 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
555 QCryptoTLSCreds *tlscreds,
556 const char *hostname, Error **errp)
558 int ret;
559 QIOChannelTLS *tioc;
560 struct NBDTLSHandshakeData data = { 0 };
562 ret = nbd_request_simple_option(ioc, NBD_OPT_STARTTLS, errp);
563 if (ret <= 0) {
564 if (ret == 0) {
565 error_setg(errp, "Server don't support STARTTLS option");
566 nbd_send_opt_abort(ioc);
568 return NULL;
571 trace_nbd_receive_starttls_new_client();
572 tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
573 if (!tioc) {
574 return NULL;
576 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
577 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
578 trace_nbd_receive_starttls_tls_handshake();
579 qio_channel_tls_handshake(tioc,
580 nbd_tls_handshake,
581 &data,
582 NULL,
583 NULL);
585 if (!data.complete) {
586 g_main_loop_run(data.loop);
588 g_main_loop_unref(data.loop);
589 if (data.error) {
590 error_propagate(errp, data.error);
591 object_unref(OBJECT(tioc));
592 return NULL;
595 return QIO_CHANNEL(tioc);
598 /* nbd_negotiate_simple_meta_context:
599 * Set one meta context. Simple means that reply must contain zero (not
600 * negotiated) or one (negotiated) contexts. More contexts would be considered
601 * as a protocol error. It's also implied that meta-data query equals queried
602 * context name, so, if server replies with something different than @context,
603 * it is considered an error too.
604 * return 1 for successful negotiation, context_id is set
605 * 0 if operation is unsupported,
606 * -1 with errp set for any other error
608 static int nbd_negotiate_simple_meta_context(QIOChannel *ioc,
609 const char *export,
610 const char *context,
611 uint32_t *context_id,
612 Error **errp)
614 int ret;
615 NBDOptionReply reply;
616 uint32_t received_id;
617 bool received;
618 uint32_t export_len = strlen(export);
619 uint32_t context_len = strlen(context);
620 uint32_t data_len = sizeof(export_len) + export_len +
621 sizeof(uint32_t) + /* number of queries */
622 sizeof(context_len) + context_len;
623 char *data = g_malloc(data_len);
624 char *p = data;
626 trace_nbd_opt_meta_request(context, export);
627 stl_be_p(p, export_len);
628 memcpy(p += sizeof(export_len), export, export_len);
629 stl_be_p(p += export_len, 1);
630 stl_be_p(p += sizeof(uint32_t), context_len);
631 memcpy(p += sizeof(context_len), context, context_len);
633 ret = nbd_send_option_request(ioc, NBD_OPT_SET_META_CONTEXT, data_len, data,
634 errp);
635 g_free(data);
636 if (ret < 0) {
637 return ret;
640 if (nbd_receive_option_reply(ioc, NBD_OPT_SET_META_CONTEXT, &reply,
641 errp) < 0)
643 return -1;
646 ret = nbd_handle_reply_err(ioc, &reply, errp);
647 if (ret <= 0) {
648 return ret;
651 if (reply.type == NBD_REP_META_CONTEXT) {
652 char *name;
654 if (reply.length != sizeof(received_id) + context_len) {
655 error_setg(errp, "Failed to negotiate meta context '%s', server "
656 "answered with unexpected length %" PRIu32, context,
657 reply.length);
658 nbd_send_opt_abort(ioc);
659 return -1;
662 if (nbd_read(ioc, &received_id, sizeof(received_id), errp) < 0) {
663 return -1;
665 be32_to_cpus(&received_id);
667 reply.length -= sizeof(received_id);
668 name = g_malloc(reply.length + 1);
669 if (nbd_read(ioc, name, reply.length, errp) < 0) {
670 g_free(name);
671 return -1;
673 name[reply.length] = '\0';
674 if (strcmp(context, name)) {
675 error_setg(errp, "Failed to negotiate meta context '%s', server "
676 "answered with different context '%s'", context,
677 name);
678 g_free(name);
679 nbd_send_opt_abort(ioc);
680 return -1;
682 g_free(name);
684 trace_nbd_opt_meta_reply(context, received_id);
685 received = true;
687 /* receive NBD_REP_ACK */
688 if (nbd_receive_option_reply(ioc, NBD_OPT_SET_META_CONTEXT, &reply,
689 errp) < 0)
691 return -1;
694 ret = nbd_handle_reply_err(ioc, &reply, errp);
695 if (ret <= 0) {
696 return ret;
700 if (reply.type != NBD_REP_ACK) {
701 error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
702 reply.type, NBD_REP_ACK);
703 nbd_send_opt_abort(ioc);
704 return -1;
706 if (reply.length) {
707 error_setg(errp, "Unexpected length to ACK response");
708 nbd_send_opt_abort(ioc);
709 return -1;
712 if (received) {
713 *context_id = received_id;
714 return 1;
717 return 0;
720 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
721 QCryptoTLSCreds *tlscreds, const char *hostname,
722 QIOChannel **outioc, NBDExportInfo *info,
723 Error **errp)
725 char buf[256];
726 uint64_t magic;
727 int rc;
728 bool zeroes = true;
729 bool structured_reply = info->structured_reply;
730 bool base_allocation = info->base_allocation;
732 trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
734 info->structured_reply = false;
735 info->base_allocation = false;
736 rc = -EINVAL;
738 if (outioc) {
739 *outioc = NULL;
741 if (tlscreds && !outioc) {
742 error_setg(errp, "Output I/O channel required for TLS");
743 goto fail;
746 if (nbd_read(ioc, buf, 8, errp) < 0) {
747 error_prepend(errp, "Failed to read data: ");
748 goto fail;
751 buf[8] = '\0';
752 if (strlen(buf) == 0) {
753 error_setg(errp, "Server connection closed unexpectedly");
754 goto fail;
757 magic = ldq_be_p(buf);
758 trace_nbd_receive_negotiate_magic(magic);
760 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
761 error_setg(errp, "Invalid magic received");
762 goto fail;
765 if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
766 error_prepend(errp, "Failed to read magic: ");
767 goto fail;
769 magic = be64_to_cpu(magic);
770 trace_nbd_receive_negotiate_magic(magic);
772 if (magic == NBD_OPTS_MAGIC) {
773 uint32_t clientflags = 0;
774 uint16_t globalflags;
775 bool fixedNewStyle = false;
777 if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
778 error_prepend(errp, "Failed to read server flags: ");
779 goto fail;
781 globalflags = be16_to_cpu(globalflags);
782 trace_nbd_receive_negotiate_server_flags(globalflags);
783 if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
784 fixedNewStyle = true;
785 clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
787 if (globalflags & NBD_FLAG_NO_ZEROES) {
788 zeroes = false;
789 clientflags |= NBD_FLAG_C_NO_ZEROES;
791 /* client requested flags */
792 clientflags = cpu_to_be32(clientflags);
793 if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
794 error_prepend(errp, "Failed to send clientflags field: ");
795 goto fail;
797 if (tlscreds) {
798 if (fixedNewStyle) {
799 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
800 if (!*outioc) {
801 goto fail;
803 ioc = *outioc;
804 } else {
805 error_setg(errp, "Server does not support STARTTLS");
806 goto fail;
809 if (!name) {
810 trace_nbd_receive_negotiate_default_name();
811 name = "";
813 if (fixedNewStyle) {
814 int result;
816 if (structured_reply) {
817 result = nbd_request_simple_option(ioc,
818 NBD_OPT_STRUCTURED_REPLY,
819 errp);
820 if (result < 0) {
821 goto fail;
823 info->structured_reply = result == 1;
826 if (info->structured_reply && base_allocation) {
827 result = nbd_negotiate_simple_meta_context(
828 ioc, name, "base:allocation",
829 &info->meta_base_allocation_id, errp);
830 if (result < 0) {
831 goto fail;
833 info->base_allocation = result == 1;
836 /* Try NBD_OPT_GO first - if it works, we are done (it
837 * also gives us a good message if the server requires
838 * TLS). If it is not available, fall back to
839 * NBD_OPT_LIST for nicer error messages about a missing
840 * export, then use NBD_OPT_EXPORT_NAME. */
841 result = nbd_opt_go(ioc, name, info, errp);
842 if (result < 0) {
843 goto fail;
845 if (result > 0) {
846 return 0;
848 /* Check our desired export is present in the
849 * server export list. Since NBD_OPT_EXPORT_NAME
850 * cannot return an error message, running this
851 * query gives us better error reporting if the
852 * export name is not available.
854 if (nbd_receive_query_exports(ioc, name, errp) < 0) {
855 goto fail;
858 /* write the export name request */
859 if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
860 errp) < 0) {
861 goto fail;
864 /* Read the response */
865 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
866 error_prepend(errp, "Failed to read export length: ");
867 goto fail;
869 be64_to_cpus(&info->size);
871 if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
872 error_prepend(errp, "Failed to read export flags: ");
873 goto fail;
875 be16_to_cpus(&info->flags);
876 } else if (magic == NBD_CLIENT_MAGIC) {
877 uint32_t oldflags;
879 if (name) {
880 error_setg(errp, "Server does not support export names");
881 goto fail;
883 if (tlscreds) {
884 error_setg(errp, "Server does not support STARTTLS");
885 goto fail;
888 if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
889 error_prepend(errp, "Failed to read export length: ");
890 goto fail;
892 be64_to_cpus(&info->size);
894 if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
895 error_prepend(errp, "Failed to read export flags: ");
896 goto fail;
898 be32_to_cpus(&oldflags);
899 if (oldflags & ~0xffff) {
900 error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
901 goto fail;
903 info->flags = oldflags;
904 } else {
905 error_setg(errp, "Bad magic received");
906 goto fail;
909 trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
910 if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
911 error_prepend(errp, "Failed to read reserved block: ");
912 goto fail;
914 rc = 0;
916 fail:
917 return rc;
920 #ifdef __linux__
921 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
922 Error **errp)
924 unsigned long sector_size = MAX(BDRV_SECTOR_SIZE, info->min_block);
925 unsigned long sectors = info->size / sector_size;
927 /* FIXME: Once the kernel module is patched to honor block sizes,
928 * and to advertise that fact to user space, we should update the
929 * hand-off to the kernel to use any block sizes we learned. */
930 assert(!info->request_sizes);
931 if (info->size / sector_size != sectors) {
932 error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
933 info->size);
934 return -E2BIG;
937 trace_nbd_init_set_socket();
939 if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
940 int serrno = errno;
941 error_setg(errp, "Failed to set NBD socket");
942 return -serrno;
945 trace_nbd_init_set_block_size(sector_size);
947 if (ioctl(fd, NBD_SET_BLKSIZE, sector_size) < 0) {
948 int serrno = errno;
949 error_setg(errp, "Failed setting NBD block size");
950 return -serrno;
953 trace_nbd_init_set_size(sectors);
954 if (info->size % sector_size) {
955 trace_nbd_init_trailing_bytes(info->size % sector_size);
958 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
959 int serrno = errno;
960 error_setg(errp, "Failed setting size (in blocks)");
961 return -serrno;
964 if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
965 if (errno == ENOTTY) {
966 int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
967 trace_nbd_init_set_readonly();
969 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
970 int serrno = errno;
971 error_setg(errp, "Failed setting read-only attribute");
972 return -serrno;
974 } else {
975 int serrno = errno;
976 error_setg(errp, "Failed setting flags");
977 return -serrno;
981 trace_nbd_init_finish();
983 return 0;
986 int nbd_client(int fd)
988 int ret;
989 int serrno;
991 trace_nbd_client_loop();
993 ret = ioctl(fd, NBD_DO_IT);
994 if (ret < 0 && errno == EPIPE) {
995 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
996 * the socket via NBD_DISCONNECT. We do not want to return 1 in
997 * that case.
999 ret = 0;
1001 serrno = errno;
1003 trace_nbd_client_loop_ret(ret, strerror(serrno));
1005 trace_nbd_client_clear_queue();
1006 ioctl(fd, NBD_CLEAR_QUE);
1008 trace_nbd_client_clear_socket();
1009 ioctl(fd, NBD_CLEAR_SOCK);
1011 errno = serrno;
1012 return ret;
1015 int nbd_disconnect(int fd)
1017 ioctl(fd, NBD_CLEAR_QUE);
1018 ioctl(fd, NBD_DISCONNECT);
1019 ioctl(fd, NBD_CLEAR_SOCK);
1020 return 0;
1023 #else
1024 int nbd_init(int fd, QIOChannelSocket *ioc, NBDExportInfo *info,
1025 Error **errp)
1027 error_setg(errp, "nbd_init is only supported on Linux");
1028 return -ENOTSUP;
1031 int nbd_client(int fd)
1033 return -ENOTSUP;
1035 int nbd_disconnect(int fd)
1037 return -ENOTSUP;
1039 #endif
1041 int nbd_send_request(QIOChannel *ioc, NBDRequest *request)
1043 uint8_t buf[NBD_REQUEST_SIZE];
1045 trace_nbd_send_request(request->from, request->len, request->handle,
1046 request->flags, request->type,
1047 nbd_cmd_lookup(request->type));
1049 stl_be_p(buf, NBD_REQUEST_MAGIC);
1050 stw_be_p(buf + 4, request->flags);
1051 stw_be_p(buf + 6, request->type);
1052 stq_be_p(buf + 8, request->handle);
1053 stq_be_p(buf + 16, request->from);
1054 stl_be_p(buf + 24, request->len);
1056 return nbd_write(ioc, buf, sizeof(buf), NULL);
1059 /* nbd_receive_simple_reply
1060 * Read simple reply except magic field (which should be already read).
1061 * Payload is not read (payload is possible for CMD_READ, but here we even
1062 * don't know whether it take place or not).
1064 static int nbd_receive_simple_reply(QIOChannel *ioc, NBDSimpleReply *reply,
1065 Error **errp)
1067 int ret;
1069 assert(reply->magic == NBD_SIMPLE_REPLY_MAGIC);
1071 ret = nbd_read(ioc, (uint8_t *)reply + sizeof(reply->magic),
1072 sizeof(*reply) - sizeof(reply->magic), errp);
1073 if (ret < 0) {
1074 return ret;
1077 be32_to_cpus(&reply->error);
1078 be64_to_cpus(&reply->handle);
1080 return 0;
1083 /* nbd_receive_structured_reply_chunk
1084 * Read structured reply chunk except magic field (which should be already
1085 * read).
1086 * Payload is not read.
1088 static int nbd_receive_structured_reply_chunk(QIOChannel *ioc,
1089 NBDStructuredReplyChunk *chunk,
1090 Error **errp)
1092 int ret;
1094 assert(chunk->magic == NBD_STRUCTURED_REPLY_MAGIC);
1096 ret = nbd_read(ioc, (uint8_t *)chunk + sizeof(chunk->magic),
1097 sizeof(*chunk) - sizeof(chunk->magic), errp);
1098 if (ret < 0) {
1099 return ret;
1102 be16_to_cpus(&chunk->flags);
1103 be16_to_cpus(&chunk->type);
1104 be64_to_cpus(&chunk->handle);
1105 be32_to_cpus(&chunk->length);
1107 return 0;
1110 /* nbd_receive_reply
1111 * Returns 1 on success
1112 * 0 on eof, when no data was read (errp is not set)
1113 * negative errno on failure (errp is set)
1115 int nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
1117 int ret;
1118 const char *type;
1120 ret = nbd_read_eof(ioc, &reply->magic, sizeof(reply->magic), errp);
1121 if (ret <= 0) {
1122 return ret;
1125 be32_to_cpus(&reply->magic);
1127 switch (reply->magic) {
1128 case NBD_SIMPLE_REPLY_MAGIC:
1129 ret = nbd_receive_simple_reply(ioc, &reply->simple, errp);
1130 if (ret < 0) {
1131 break;
1133 trace_nbd_receive_simple_reply(reply->simple.error,
1134 nbd_err_lookup(reply->simple.error),
1135 reply->handle);
1136 break;
1137 case NBD_STRUCTURED_REPLY_MAGIC:
1138 ret = nbd_receive_structured_reply_chunk(ioc, &reply->structured, errp);
1139 if (ret < 0) {
1140 break;
1142 type = nbd_reply_type_lookup(reply->structured.type);
1143 trace_nbd_receive_structured_reply_chunk(reply->structured.flags,
1144 reply->structured.type, type,
1145 reply->structured.handle,
1146 reply->structured.length);
1147 break;
1148 default:
1149 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", reply->magic);
1150 return -EINVAL;
1152 if (ret < 0) {
1153 return ret;
1156 return 1;