?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
MAINTAINERS: Drop Aneesh as 9pfs maintainer
[qemu.git] / nbd / server.c
blob92c0fdd03b651872fda5f5c0a52227c1b77996d1
1 /*
2 * Copyright (C) 2016-2017 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 *
5 * Network Block Device Server 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 static int system_errno_to_nbd_errno(int err)
26 {
27 switch (err) {
28 case 0:
29 return NBD_SUCCESS;
30 case EPERM:
31 case EROFS:
32 return NBD_EPERM;
33 case EIO:
34 return NBD_EIO;
35 case ENOMEM:
36 return NBD_ENOMEM;
37 #ifdef EDQUOT
38 case EDQUOT:
39 #endif
40 case EFBIG:
41 case ENOSPC:
42 return NBD_ENOSPC;
43 case EOVERFLOW:
44 return NBD_EOVERFLOW;
45 case ESHUTDOWN:
46 return NBD_ESHUTDOWN;
47 case EINVAL:
48 default:
49 return NBD_EINVAL;
50 }
51 }
52
53 /* Definitions for opaque data types */
54
55 typedef struct NBDRequestData NBDRequestData;
56
57 struct NBDRequestData {
58 QSIMPLEQ_ENTRY(NBDRequestData) entry;
59 NBDClient *client;
60 uint8_t *data;
61 bool complete;
62 };
63
64 struct NBDExport {
65 int refcount;
66 void (*close)(NBDExport *exp);
67
68 BlockBackend *blk;
69 char *name;
70 char *description;
71 off_t dev_offset;
72 off_t size;
73 uint16_t nbdflags;
74 QTAILQ_HEAD(, NBDClient) clients;
75 QTAILQ_ENTRY(NBDExport) next;
76
77 AioContext *ctx;
78
79 BlockBackend *eject_notifier_blk;
80 Notifier eject_notifier;
81 };
82
83 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
84
85 struct NBDClient {
86 int refcount;
87 void (*close_fn)(NBDClient *client, bool negotiated);
88
89 NBDExport *exp;
90 QCryptoTLSCreds *tlscreds;
91 char *tlsaclname;
92 QIOChannelSocket *sioc; /* The underlying data channel */
93 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
94
95 Coroutine *recv_coroutine;
96
97 CoMutex send_lock;
98 Coroutine *send_coroutine;
99
100 QTAILQ_ENTRY(NBDClient) next;
101 int nb_requests;
102 bool closing;
103
104 bool structured_reply;
105 };
106
107 /* That's all folks */
108
109 static void nbd_client_receive_next_request(NBDClient *client);
110
111 /* Basic flow for negotiation
112
113 Server Client
114 Negotiate
115
116 or
117
118 Server Client
119 Negotiate #1
120 Option
121 Negotiate #2
122
123 ----
124
125 followed by
126
127 Server Client
128 Request
129 Response
130 Request
131 Response
132 ...
133 ...
134 Request (type == 2)
135
136 */
137
138 /* Send a reply header, including length, but no payload.
139 * Return -errno on error, 0 on success. */
140 static int nbd_negotiate_send_rep_len(QIOChannel *ioc, uint32_t type,
141 uint32_t opt, uint32_t len, Error **errp)
142 {
143 uint64_t magic;
144
145 trace_nbd_negotiate_send_rep_len(opt, nbd_opt_lookup(opt),
146 type, nbd_rep_lookup(type), len);
147
148 assert(len < NBD_MAX_BUFFER_SIZE);
149 magic = cpu_to_be64(NBD_REP_MAGIC);
150 if (nbd_write(ioc, &magic, sizeof(magic), errp) < 0) {
151 error_prepend(errp, "write failed (rep magic): ");
152 return -EINVAL;
153 }
154
155 opt = cpu_to_be32(opt);
156 if (nbd_write(ioc, &opt, sizeof(opt), errp) < 0) {
157 error_prepend(errp, "write failed (rep opt): ");
158 return -EINVAL;
159 }
160
161 type = cpu_to_be32(type);
162 if (nbd_write(ioc, &type, sizeof(type), errp) < 0) {
163 error_prepend(errp, "write failed (rep type): ");
164 return -EINVAL;
165 }
166
167 len = cpu_to_be32(len);
168 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
169 error_prepend(errp, "write failed (rep data length): ");
170 return -EINVAL;
171 }
172 return 0;
173 }
174
175 /* Send a reply header with default 0 length.
176 * Return -errno on error, 0 on success. */
177 static int nbd_negotiate_send_rep(QIOChannel *ioc, uint32_t type, uint32_t opt,
178 Error **errp)
179 {
180 return nbd_negotiate_send_rep_len(ioc, type, opt, 0, errp);
181 }
182
183 /* Send an error reply.
184 * Return -errno on error, 0 on success. */
185 static int GCC_FMT_ATTR(5, 6)
186 nbd_negotiate_send_rep_err(QIOChannel *ioc, uint32_t type,
187 uint32_t opt, Error **errp, const char *fmt, ...)
188 {
189 va_list va;
190 char *msg;
191 int ret;
192 size_t len;
193
194 va_start(va, fmt);
195 msg = g_strdup_vprintf(fmt, va);
196 va_end(va);
197 len = strlen(msg);
198 assert(len < 4096);
199 trace_nbd_negotiate_send_rep_err(msg);
200 ret = nbd_negotiate_send_rep_len(ioc, type, opt, len, errp);
201 if (ret < 0) {
202 goto out;
203 }
204 if (nbd_write(ioc, msg, len, errp) < 0) {
205 error_prepend(errp, "write failed (error message): ");
206 ret = -EIO;
207 } else {
208 ret = 0;
209 }
210
211 out:
212 g_free(msg);
213 return ret;
214 }
215
216 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
217 * Return -errno on error, 0 on success. */
218 static int nbd_negotiate_send_rep_list(QIOChannel *ioc, NBDExport *exp,
219 Error **errp)
220 {
221 size_t name_len, desc_len;
222 uint32_t len;
223 const char *name = exp->name ? exp->name : "";
224 const char *desc = exp->description ? exp->description : "";
225 int ret;
226
227 trace_nbd_negotiate_send_rep_list(name, desc);
228 name_len = strlen(name);
229 desc_len = strlen(desc);
230 len = name_len + desc_len + sizeof(len);
231 ret = nbd_negotiate_send_rep_len(ioc, NBD_REP_SERVER, NBD_OPT_LIST, len,
232 errp);
233 if (ret < 0) {
234 return ret;
235 }
236
237 len = cpu_to_be32(name_len);
238 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
239 error_prepend(errp, "write failed (name length): ");
240 return -EINVAL;
241 }
242
243 if (nbd_write(ioc, name, name_len, errp) < 0) {
244 error_prepend(errp, "write failed (name buffer): ");
245 return -EINVAL;
246 }
247
248 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
249 error_prepend(errp, "write failed (description buffer): ");
250 return -EINVAL;
251 }
252
253 return 0;
254 }
255
256 /* Process the NBD_OPT_LIST command, with a potential series of replies.
257 * Return -errno on error, 0 on success. */
258 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
259 {
260 NBDExport *exp;
261
262 /* For each export, send a NBD_REP_SERVER reply. */
263 QTAILQ_FOREACH(exp, &exports, next) {
264 if (nbd_negotiate_send_rep_list(client->ioc, exp, errp)) {
265 return -EINVAL;
266 }
267 }
268 /* Finish with a NBD_REP_ACK. */
269 return nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, NBD_OPT_LIST, errp);
270 }
271
272 /* Send a reply to NBD_OPT_EXPORT_NAME.
273 * Return -errno on error, 0 on success. */
274 static int nbd_negotiate_handle_export_name(NBDClient *client, uint32_t length,
275 uint16_t myflags, bool no_zeroes,
276 Error **errp)
277 {
278 char name[NBD_MAX_NAME_SIZE + 1];
279 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
280 size_t len;
281 int ret;
282
283 /* Client sends:
284 [20 .. xx] export name (length bytes)
285 Server replies:
286 [ 0 .. 7] size
287 [ 8 .. 9] export flags
288 [10 .. 133] reserved (0) [unless no_zeroes]
289 */
290 trace_nbd_negotiate_handle_export_name();
291 if (length >= sizeof(name)) {
292 error_setg(errp, "Bad length received");
293 return -EINVAL;
294 }
295 if (nbd_read(client->ioc, name, length, errp) < 0) {
296 error_prepend(errp, "read failed: ");
297 return -EINVAL;
298 }
299 name[length] = '\0';
300
301 trace_nbd_negotiate_handle_export_name_request(name);
302
303 client->exp = nbd_export_find(name);
304 if (!client->exp) {
305 error_setg(errp, "export not found");
306 return -EINVAL;
307 }
308
309 trace_nbd_negotiate_new_style_size_flags(client->exp->size,
310 client->exp->nbdflags | myflags);
311 stq_be_p(buf, client->exp->size);
312 stw_be_p(buf + 8, client->exp->nbdflags | myflags);
313 len = no_zeroes ? 10 : sizeof(buf);
314 ret = nbd_write(client->ioc, buf, len, errp);
315 if (ret < 0) {
316 error_prepend(errp, "write failed: ");
317 return ret;
318 }
319
320 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
321 nbd_export_get(client->exp);
322
323 return 0;
324 }
325
326 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
327 * The buffer does NOT include the info type prefix.
328 * Return -errno on error, 0 if ready to send more. */
329 static int nbd_negotiate_send_info(NBDClient *client, uint32_t opt,
330 uint16_t info, uint32_t length, void *buf,
331 Error **errp)
332 {
333 int rc;
334
335 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
336 rc = nbd_negotiate_send_rep_len(client->ioc, NBD_REP_INFO, opt,
337 sizeof(info) + length, errp);
338 if (rc < 0) {
339 return rc;
340 }
341 cpu_to_be16s(&info);
342 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
343 return -EIO;
344 }
345 if (nbd_write(client->ioc, buf, length, errp) < 0) {
346 return -EIO;
347 }
348 return 0;
349 }
350
351 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
352 * Return -errno on error, 0 if ready for next option, and 1 to move
353 * into transmission phase. */
354 static int nbd_negotiate_handle_info(NBDClient *client, uint32_t length,
355 uint32_t opt, uint16_t myflags,
356 Error **errp)
357 {
358 int rc;
359 char name[NBD_MAX_NAME_SIZE + 1];
360 NBDExport *exp;
361 uint16_t requests;
362 uint16_t request;
363 uint32_t namelen;
364 bool sendname = false;
365 bool blocksize = false;
366 uint32_t sizes[3];
367 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
368 const char *msg;
369
370 /* Client sends:
371 4 bytes: L, name length (can be 0)
372 L bytes: export name
373 2 bytes: N, number of requests (can be 0)
374 N * 2 bytes: N requests
375 */
376 if (length < sizeof(namelen) + sizeof(requests)) {
377 msg = "overall request too short";
378 goto invalid;
379 }
380 if (nbd_read(client->ioc, &namelen, sizeof(namelen), errp) < 0) {
381 return -EIO;
382 }
383 be32_to_cpus(&namelen);
384 length -= sizeof(namelen);
385 if (namelen > length - sizeof(requests) || (length - namelen) % 2) {
386 msg = "name length is incorrect";
387 goto invalid;
388 }
389 if (namelen >= sizeof(name)) {
390 msg = "name too long for qemu";
391 goto invalid;
392 }
393 if (nbd_read(client->ioc, name, namelen, errp) < 0) {
394 return -EIO;
395 }
396 name[namelen] = '\0';
397 length -= namelen;
398 trace_nbd_negotiate_handle_export_name_request(name);
399
400 if (nbd_read(client->ioc, &requests, sizeof(requests), errp) < 0) {
401 return -EIO;
402 }
403 be16_to_cpus(&requests);
404 length -= sizeof(requests);
405 trace_nbd_negotiate_handle_info_requests(requests);
406 if (requests != length / sizeof(request)) {
407 msg = "incorrect number of requests for overall length";
408 goto invalid;
409 }
410 while (requests--) {
411 if (nbd_read(client->ioc, &request, sizeof(request), errp) < 0) {
412 return -EIO;
413 }
414 be16_to_cpus(&request);
415 length -= sizeof(request);
416 trace_nbd_negotiate_handle_info_request(request,
417 nbd_info_lookup(request));
418 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
419 * everything else is either a request we don't know or
420 * something we send regardless of request */
421 switch (request) {
422 case NBD_INFO_NAME:
423 sendname = true;
424 break;
425 case NBD_INFO_BLOCK_SIZE:
426 blocksize = true;
427 break;
428 }
429 }
430 assert(length == 0);
431
432 exp = nbd_export_find(name);
433 if (!exp) {
434 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_UNKNOWN,
435 opt, errp, "export '%s' not present",
436 name);
437 }
438
439 /* Don't bother sending NBD_INFO_NAME unless client requested it */
440 if (sendname) {
441 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, namelen, name,
442 errp);
443 if (rc < 0) {
444 return rc;
445 }
446 }
447
448 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
449 * client request */
450 if (exp->description) {
451 size_t len = strlen(exp->description);
452
453 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_DESCRIPTION,
454 len, exp->description, errp);
455 if (rc < 0) {
456 return rc;
457 }
458 }
459
460 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
461 * according to whether the client requested it, and according to
462 * whether this is OPT_INFO or OPT_GO. */
463 /* minimum - 1 for back-compat, or 512 if client is new enough.
464 * TODO: consult blk_bs(blk)->bl.request_alignment? */
465 sizes[0] = (opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
466 /* preferred - Hard-code to 4096 for now.
467 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
468 sizes[1] = 4096;
469 /* maximum - At most 32M, but smaller as appropriate. */
470 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
471 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
472 cpu_to_be32s(&sizes[0]);
473 cpu_to_be32s(&sizes[1]);
474 cpu_to_be32s(&sizes[2]);
475 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_BLOCK_SIZE,
476 sizeof(sizes), sizes, errp);
477 if (rc < 0) {
478 return rc;
479 }
480
481 /* Send NBD_INFO_EXPORT always */
482 trace_nbd_negotiate_new_style_size_flags(exp->size,
483 exp->nbdflags | myflags);
484 stq_be_p(buf, exp->size);
485 stw_be_p(buf + 8, exp->nbdflags | myflags);
486 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_EXPORT,
487 sizeof(buf), buf, errp);
488 if (rc < 0) {
489 return rc;
490 }
491
492 /* If the client is just asking for NBD_OPT_INFO, but forgot to
493 * request block sizes, return an error.
494 * TODO: consult blk_bs(blk)->request_align, and only error if it
495 * is not 1? */
496 if (opt == NBD_OPT_INFO && !blocksize) {
497 return nbd_negotiate_send_rep_err(client->ioc,
498 NBD_REP_ERR_BLOCK_SIZE_REQD, opt,
499 errp,
500 "request NBD_INFO_BLOCK_SIZE to "
501 "use this export");
502 }
503
504 /* Final reply */
505 rc = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, opt, errp);
506 if (rc < 0) {
507 return rc;
508 }
509
510 if (opt == NBD_OPT_GO) {
511 client->exp = exp;
512 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
513 nbd_export_get(client->exp);
514 rc = 1;
515 }
516 return rc;
517
518 invalid:
519 if (nbd_drop(client->ioc, length, errp) < 0) {
520 return -EIO;
521 }
522 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID, opt,
523 errp, "%s", msg);
524 }
525
526
527 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
528 * new channel for all further (now-encrypted) communication. */
529 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
530 Error **errp)
531 {
532 QIOChannel *ioc;
533 QIOChannelTLS *tioc;
534 struct NBDTLSHandshakeData data = { 0 };
535
536 trace_nbd_negotiate_handle_starttls();
537 ioc = client->ioc;
538
539 if (nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
540 NBD_OPT_STARTTLS, errp) < 0) {
541 return NULL;
542 }
543
544 tioc = qio_channel_tls_new_server(ioc,
545 client->tlscreds,
546 client->tlsaclname,
547 errp);
548 if (!tioc) {
549 return NULL;
550 }
551
552 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
553 trace_nbd_negotiate_handle_starttls_handshake();
554 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
555 qio_channel_tls_handshake(tioc,
556 nbd_tls_handshake,
557 &data,
558 NULL);
559
560 if (!data.complete) {
561 g_main_loop_run(data.loop);
562 }
563 g_main_loop_unref(data.loop);
564 if (data.error) {
565 object_unref(OBJECT(tioc));
566 error_propagate(errp, data.error);
567 return NULL;
568 }
569
570 return QIO_CHANNEL(tioc);
571 }
572
573 /* nbd_reject_length: Handle any unexpected payload.
574 * @fatal requests that we quit talking to the client, even if we are able
575 * to successfully send an error to the guest.
576 * Return:
577 * -errno transmission error occurred or @fatal was requested, errp is set
578 * 0 error message successfully sent to client, errp is not set
579 */
580 static int nbd_reject_length(NBDClient *client, uint32_t length,
581 uint32_t option, bool fatal, Error **errp)
582 {
583 int ret;
584
585 assert(length);
586 if (nbd_drop(client->ioc, length, errp) < 0) {
587 return -EIO;
588 }
589 ret = nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID,
590 option, errp,
591 "option '%s' should have zero length",
592 nbd_opt_lookup(option));
593 if (fatal && !ret) {
594 error_setg(errp, "option '%s' should have zero length",
595 nbd_opt_lookup(option));
596 return -EINVAL;
597 }
598 return ret;
599 }
600
601 /* nbd_negotiate_options
602 * Process all NBD_OPT_* client option commands, during fixed newstyle
603 * negotiation.
604 * Return:
605 * -errno on error, errp is set
606 * 0 on successful negotiation, errp is not set
607 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
608 * errp is not set
609 */
610 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
611 Error **errp)
612 {
613 uint32_t flags;
614 bool fixedNewstyle = false;
615 bool no_zeroes = false;
616
617 /* Client sends:
618 [ 0 .. 3] client flags
619
620 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
621 [ 0 .. 7] NBD_OPTS_MAGIC
622 [ 8 .. 11] NBD option
623 [12 .. 15] Data length
624 ... Rest of request
625
626 [ 0 .. 7] NBD_OPTS_MAGIC
627 [ 8 .. 11] Second NBD option
628 [12 .. 15] Data length
629 ... Rest of request
630 */
631
632 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
633 error_prepend(errp, "read failed: ");
634 return -EIO;
635 }
636 be32_to_cpus(&flags);
637 trace_nbd_negotiate_options_flags(flags);
638 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
639 fixedNewstyle = true;
640 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
641 }
642 if (flags & NBD_FLAG_C_NO_ZEROES) {
643 no_zeroes = true;
644 flags &= ~NBD_FLAG_C_NO_ZEROES;
645 }
646 if (flags != 0) {
647 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
648 return -EINVAL;
649 }
650
651 while (1) {
652 int ret;
653 uint32_t option, length;
654 uint64_t magic;
655
656 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
657 error_prepend(errp, "read failed: ");
658 return -EINVAL;
659 }
660 magic = be64_to_cpu(magic);
661 trace_nbd_negotiate_options_check_magic(magic);
662 if (magic != NBD_OPTS_MAGIC) {
663 error_setg(errp, "Bad magic received");
664 return -EINVAL;
665 }
666
667 if (nbd_read(client->ioc, &option,
668 sizeof(option), errp) < 0) {
669 error_prepend(errp, "read failed: ");
670 return -EINVAL;
671 }
672 option = be32_to_cpu(option);
673
674 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
675 error_prepend(errp, "read failed: ");
676 return -EINVAL;
677 }
678 length = be32_to_cpu(length);
679
680 if (length > NBD_MAX_BUFFER_SIZE) {
681 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
682 length, NBD_MAX_BUFFER_SIZE);
683 return -EINVAL;
684 }
685
686 trace_nbd_negotiate_options_check_option(option,
687 nbd_opt_lookup(option));
688 if (client->tlscreds &&
689 client->ioc == (QIOChannel *)client->sioc) {
690 QIOChannel *tioc;
691 if (!fixedNewstyle) {
692 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
693 return -EINVAL;
694 }
695 switch (option) {
696 case NBD_OPT_STARTTLS:
697 if (length) {
698 /* Unconditionally drop the connection if the client
699 * can't start a TLS negotiation correctly */
700 return nbd_reject_length(client, length, option, true,
701 errp);
702 }
703 tioc = nbd_negotiate_handle_starttls(client, errp);
704 if (!tioc) {
705 return -EIO;
706 }
707 ret = 0;
708 object_unref(OBJECT(client->ioc));
709 client->ioc = QIO_CHANNEL(tioc);
710 break;
711
712 case NBD_OPT_EXPORT_NAME:
713 /* No way to return an error to client, so drop connection */
714 error_setg(errp, "Option 0x%x not permitted before TLS",
715 option);
716 return -EINVAL;
717
718 default:
719 if (nbd_drop(client->ioc, length, errp) < 0) {
720 return -EIO;
721 }
722 ret = nbd_negotiate_send_rep_err(client->ioc,
723 NBD_REP_ERR_TLS_REQD,
724 option, errp,
725 "Option 0x%" PRIx32
726 "not permitted before TLS",
727 option);
728 /* Let the client keep trying, unless they asked to
729 * quit. In this mode, we've already sent an error, so
730 * we can't ack the abort. */
731 if (option == NBD_OPT_ABORT) {
732 return 1;
733 }
734 break;
735 }
736 } else if (fixedNewstyle) {
737 switch (option) {
738 case NBD_OPT_LIST:
739 if (length) {
740 ret = nbd_reject_length(client, length, option, false,
741 errp);
742 } else {
743 ret = nbd_negotiate_handle_list(client, errp);
744 }
745 break;
746
747 case NBD_OPT_ABORT:
748 /* NBD spec says we must try to reply before
749 * disconnecting, but that we must also tolerate
750 * guests that don't wait for our reply. */
751 nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, option, NULL);
752 return 1;
753
754 case NBD_OPT_EXPORT_NAME:
755 return nbd_negotiate_handle_export_name(client, length,
756 myflags, no_zeroes,
757 errp);
758
759 case NBD_OPT_INFO:
760 case NBD_OPT_GO:
761 ret = nbd_negotiate_handle_info(client, length, option,
762 myflags, errp);
763 if (ret == 1) {
764 assert(option == NBD_OPT_GO);
765 return 0;
766 }
767 break;
768
769 case NBD_OPT_STARTTLS:
770 if (length) {
771 ret = nbd_reject_length(client, length, option, false,
772 errp);
773 } else if (client->tlscreds) {
774 ret = nbd_negotiate_send_rep_err(client->ioc,
775 NBD_REP_ERR_INVALID,
776 option, errp,
777 "TLS already enabled");
778 } else {
779 ret = nbd_negotiate_send_rep_err(client->ioc,
780 NBD_REP_ERR_POLICY,
781 option, errp,
782 "TLS not configured");
783 }
784 break;
785
786 case NBD_OPT_STRUCTURED_REPLY:
787 if (length) {
788 ret = nbd_reject_length(client, length, option, false,
789 errp);
790 } else if (client->structured_reply) {
791 ret = nbd_negotiate_send_rep_err(
792 client->ioc, NBD_REP_ERR_INVALID, option, errp,
793 "structured reply already negotiated");
794 } else {
795 ret = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
796 option, errp);
797 client->structured_reply = true;
798 myflags |= NBD_FLAG_SEND_DF;
799 }
800 break;
801
802 default:
803 if (nbd_drop(client->ioc, length, errp) < 0) {
804 return -EIO;
805 }
806 ret = nbd_negotiate_send_rep_err(client->ioc,
807 NBD_REP_ERR_UNSUP,
808 option, errp,
809 "Unsupported option 0x%"
810 PRIx32 " (%s)", option,
811 nbd_opt_lookup(option));
812 break;
813 }
814 } else {
815 /*
816 * If broken new-style we should drop the connection
817 * for anything except NBD_OPT_EXPORT_NAME
818 */
819 switch (option) {
820 case NBD_OPT_EXPORT_NAME:
821 return nbd_negotiate_handle_export_name(client, length,
822 myflags, no_zeroes,
823 errp);
824
825 default:
826 error_setg(errp, "Unsupported option 0x%" PRIx32 " (%s)",
827 option, nbd_opt_lookup(option));
828 return -EINVAL;
829 }
830 }
831 if (ret < 0) {
832 return ret;
833 }
834 }
835 }
836
837 /* nbd_negotiate
838 * Return:
839 * -errno on error, errp is set
840 * 0 on successful negotiation, errp is not set
841 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
842 * errp is not set
843 */
844 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
845 {
846 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
847 int ret;
848 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
849 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
850 NBD_FLAG_SEND_WRITE_ZEROES);
851 bool oldStyle;
852
853 /* Old style negotiation header, no room for options
854 [ 0 .. 7] passwd ("NBDMAGIC")
855 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
856 [16 .. 23] size
857 [24 .. 27] export flags (zero-extended)
858 [28 .. 151] reserved (0)
859
860 New style negotiation header, client can send options
861 [ 0 .. 7] passwd ("NBDMAGIC")
862 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
863 [16 .. 17] server flags (0)
864 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
865 */
866
867 qio_channel_set_blocking(client->ioc, false, NULL);
868
869 trace_nbd_negotiate_begin();
870 memcpy(buf, "NBDMAGIC", 8);
871
872 oldStyle = client->exp != NULL && !client->tlscreds;
873 if (oldStyle) {
874 trace_nbd_negotiate_old_style(client->exp->size,
875 client->exp->nbdflags | myflags);
876 stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
877 stq_be_p(buf + 16, client->exp->size);
878 stl_be_p(buf + 24, client->exp->nbdflags | myflags);
879
880 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
881 error_prepend(errp, "write failed: ");
882 return -EINVAL;
883 }
884 } else {
885 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
886 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
887
888 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
889 error_prepend(errp, "write failed: ");
890 return -EINVAL;
891 }
892 ret = nbd_negotiate_options(client, myflags, errp);
893 if (ret != 0) {
894 if (ret < 0) {
895 error_prepend(errp, "option negotiation failed: ");
896 }
897 return ret;
898 }
899 }
900
901 trace_nbd_negotiate_success();
902
903 return 0;
904 }
905
906 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
907 Error **errp)
908 {
909 uint8_t buf[NBD_REQUEST_SIZE];
910 uint32_t magic;
911 int ret;
912
913 ret = nbd_read(ioc, buf, sizeof(buf), errp);
914 if (ret < 0) {
915 return ret;
916 }
917
918 /* Request
919 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
920 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
921 [ 6 .. 7] type (NBD_CMD_READ, ...)
922 [ 8 .. 15] handle
923 [16 .. 23] from
924 [24 .. 27] len
925 */
926
927 magic = ldl_be_p(buf);
928 request->flags = lduw_be_p(buf + 4);
929 request->type = lduw_be_p(buf + 6);
930 request->handle = ldq_be_p(buf + 8);
931 request->from = ldq_be_p(buf + 16);
932 request->len = ldl_be_p(buf + 24);
933
934 trace_nbd_receive_request(magic, request->flags, request->type,
935 request->from, request->len);
936
937 if (magic != NBD_REQUEST_MAGIC) {
938 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
939 return -EINVAL;
940 }
941 return 0;
942 }
943
944 #define MAX_NBD_REQUESTS 16
945
946 void nbd_client_get(NBDClient *client)
947 {
948 client->refcount++;
949 }
950
951 void nbd_client_put(NBDClient *client)
952 {
953 if (--client->refcount == 0) {
954 /* The last reference should be dropped by client->close,
955 * which is called by client_close.
956 */
957 assert(client->closing);
958
959 qio_channel_detach_aio_context(client->ioc);
960 object_unref(OBJECT(client->sioc));
961 object_unref(OBJECT(client->ioc));
962 if (client->tlscreds) {
963 object_unref(OBJECT(client->tlscreds));
964 }
965 g_free(client->tlsaclname);
966 if (client->exp) {
967 QTAILQ_REMOVE(&client->exp->clients, client, next);
968 nbd_export_put(client->exp);
969 }
970 g_free(client);
971 }
972 }
973
974 static void client_close(NBDClient *client, bool negotiated)
975 {
976 if (client->closing) {
977 return;
978 }
979
980 client->closing = true;
981
982 /* Force requests to finish. They will drop their own references,
983 * then we'll close the socket and free the NBDClient.
984 */
985 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
986 NULL);
987
988 /* Also tell the client, so that they release their reference. */
989 if (client->close_fn) {
990 client->close_fn(client, negotiated);
991 }
992 }
993
994 static NBDRequestData *nbd_request_get(NBDClient *client)
995 {
996 NBDRequestData *req;
997
998 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
999 client->nb_requests++;
1000
1001 req = g_new0(NBDRequestData, 1);
1002 nbd_client_get(client);
1003 req->client = client;
1004 return req;
1005 }
1006
1007 static void nbd_request_put(NBDRequestData *req)
1008 {
1009 NBDClient *client = req->client;
1010
1011 if (req->data) {
1012 qemu_vfree(req->data);
1013 }
1014 g_free(req);
1015
1016 client->nb_requests--;
1017 nbd_client_receive_next_request(client);
1018
1019 nbd_client_put(client);
1020 }
1021
1022 static void blk_aio_attached(AioContext *ctx, void *opaque)
1023 {
1024 NBDExport *exp = opaque;
1025 NBDClient *client;
1026
1027 trace_nbd_blk_aio_attached(exp->name, ctx);
1028
1029 exp->ctx = ctx;
1030
1031 QTAILQ_FOREACH(client, &exp->clients, next) {
1032 qio_channel_attach_aio_context(client->ioc, ctx);
1033 if (client->recv_coroutine) {
1034 aio_co_schedule(ctx, client->recv_coroutine);
1035 }
1036 if (client->send_coroutine) {
1037 aio_co_schedule(ctx, client->send_coroutine);
1038 }
1039 }
1040 }
1041
1042 static void blk_aio_detach(void *opaque)
1043 {
1044 NBDExport *exp = opaque;
1045 NBDClient *client;
1046
1047 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1048
1049 QTAILQ_FOREACH(client, &exp->clients, next) {
1050 qio_channel_detach_aio_context(client->ioc);
1051 }
1052
1053 exp->ctx = NULL;
1054 }
1055
1056 static void nbd_eject_notifier(Notifier *n, void *data)
1057 {
1058 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1059 nbd_export_close(exp);
1060 }
1061
1062 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1063 uint16_t nbdflags, void (*close)(NBDExport *),
1064 bool writethrough, BlockBackend *on_eject_blk,
1065 Error **errp)
1066 {
1067 AioContext *ctx;
1068 BlockBackend *blk;
1069 NBDExport *exp = g_new0(NBDExport, 1);
1070 uint64_t perm;
1071 int ret;
1072
1073 /*
1074 * NBD exports are used for non-shared storage migration. Make sure
1075 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1076 * access since the export could be available before migration handover.
1077 */
1078 ctx = bdrv_get_aio_context(bs);
1079 aio_context_acquire(ctx);
1080 bdrv_invalidate_cache(bs, NULL);
1081 aio_context_release(ctx);
1082
1083 /* Don't allow resize while the NBD server is running, otherwise we don't
1084 * care what happens with the node. */
1085 perm = BLK_PERM_CONSISTENT_READ;
1086 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1087 perm |= BLK_PERM_WRITE;
1088 }
1089 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1090 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1091 ret = blk_insert_bs(blk, bs, errp);
1092 if (ret < 0) {
1093 goto fail;
1094 }
1095 blk_set_enable_write_cache(blk, !writethrough);
1096
1097 exp->refcount = 1;
1098 QTAILQ_INIT(&exp->clients);
1099 exp->blk = blk;
1100 exp->dev_offset = dev_offset;
1101 exp->nbdflags = nbdflags;
1102 exp->size = size < 0 ? blk_getlength(blk) : size;
1103 if (exp->size < 0) {
1104 error_setg_errno(errp, -exp->size,
1105 "Failed to determine the NBD export's length");
1106 goto fail;
1107 }
1108 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1109
1110 exp->close = close;
1111 exp->ctx = blk_get_aio_context(blk);
1112 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1113
1114 if (on_eject_blk) {
1115 blk_ref(on_eject_blk);
1116 exp->eject_notifier_blk = on_eject_blk;
1117 exp->eject_notifier.notify = nbd_eject_notifier;
1118 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1119 }
1120 return exp;
1121
1122 fail:
1123 blk_unref(blk);
1124 g_free(exp);
1125 return NULL;
1126 }
1127
1128 NBDExport *nbd_export_find(const char *name)
1129 {
1130 NBDExport *exp;
1131 QTAILQ_FOREACH(exp, &exports, next) {
1132 if (strcmp(name, exp->name) == 0) {
1133 return exp;
1134 }
1135 }
1136
1137 return NULL;
1138 }
1139
1140 void nbd_export_set_name(NBDExport *exp, const char *name)
1141 {
1142 if (exp->name == name) {
1143 return;
1144 }
1145
1146 nbd_export_get(exp);
1147 if (exp->name != NULL) {
1148 g_free(exp->name);
1149 exp->name = NULL;
1150 QTAILQ_REMOVE(&exports, exp, next);
1151 nbd_export_put(exp);
1152 }
1153 if (name != NULL) {
1154 nbd_export_get(exp);
1155 exp->name = g_strdup(name);
1156 QTAILQ_INSERT_TAIL(&exports, exp, next);
1157 }
1158 nbd_export_put(exp);
1159 }
1160
1161 void nbd_export_set_description(NBDExport *exp, const char *description)
1162 {
1163 g_free(exp->description);
1164 exp->description = g_strdup(description);
1165 }
1166
1167 void nbd_export_close(NBDExport *exp)
1168 {
1169 NBDClient *client, *next;
1170
1171 nbd_export_get(exp);
1172 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1173 client_close(client, true);
1174 }
1175 nbd_export_set_name(exp, NULL);
1176 nbd_export_set_description(exp, NULL);
1177 nbd_export_put(exp);
1178 }
1179
1180 void nbd_export_get(NBDExport *exp)
1181 {
1182 assert(exp->refcount > 0);
1183 exp->refcount++;
1184 }
1185
1186 void nbd_export_put(NBDExport *exp)
1187 {
1188 assert(exp->refcount > 0);
1189 if (exp->refcount == 1) {
1190 nbd_export_close(exp);
1191 }
1192
1193 if (--exp->refcount == 0) {
1194 assert(exp->name == NULL);
1195 assert(exp->description == NULL);
1196
1197 if (exp->close) {
1198 exp->close(exp);
1199 }
1200
1201 if (exp->blk) {
1202 if (exp->eject_notifier_blk) {
1203 notifier_remove(&exp->eject_notifier);
1204 blk_unref(exp->eject_notifier_blk);
1205 }
1206 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1207 blk_aio_detach, exp);
1208 blk_unref(exp->blk);
1209 exp->blk = NULL;
1210 }
1211
1212 g_free(exp);
1213 }
1214 }
1215
1216 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1217 {
1218 return exp->blk;
1219 }
1220
1221 void nbd_export_close_all(void)
1222 {
1223 NBDExport *exp, *next;
1224
1225 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1226 nbd_export_close(exp);
1227 }
1228 }
1229
1230 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1231 unsigned niov, Error **errp)
1232 {
1233 int ret;
1234
1235 g_assert(qemu_in_coroutine());
1236 qemu_co_mutex_lock(&client->send_lock);
1237 client->send_coroutine = qemu_coroutine_self();
1238
1239 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1240
1241 client->send_coroutine = NULL;
1242 qemu_co_mutex_unlock(&client->send_lock);
1243
1244 return ret;
1245 }
1246
1247 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1248 uint64_t handle)
1249 {
1250 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1251 stl_be_p(&reply->error, error);
1252 stq_be_p(&reply->handle, handle);
1253 }
1254
1255 static int nbd_co_send_simple_reply(NBDClient *client,
1256 uint64_t handle,
1257 uint32_t error,
1258 void *data,
1259 size_t len,
1260 Error **errp)
1261 {
1262 NBDSimpleReply reply;
1263 int nbd_err = system_errno_to_nbd_errno(error);
1264 struct iovec iov[] = {
1265 {.iov_base = &reply, .iov_len = sizeof(reply)},
1266 {.iov_base = data, .iov_len = len}
1267 };
1268
1269 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1270 len);
1271 set_be_simple_reply(&reply, nbd_err, handle);
1272
1273 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1274 }
1275
1276 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1277 uint16_t type, uint64_t handle, uint32_t length)
1278 {
1279 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1280 stw_be_p(&chunk->flags, flags);
1281 stw_be_p(&chunk->type, type);
1282 stq_be_p(&chunk->handle, handle);
1283 stl_be_p(&chunk->length, length);
1284 }
1285
1286 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1287 uint64_t handle,
1288 Error **errp)
1289 {
1290 NBDStructuredReplyChunk chunk;
1291 struct iovec iov[] = {
1292 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1293 };
1294
1295 trace_nbd_co_send_structured_done(handle);
1296 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1297
1298 return nbd_co_send_iov(client, iov, 1, errp);
1299 }
1300
1301 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1302 uint64_t handle,
1303 uint64_t offset,
1304 void *data,
1305 size_t size,
1306 Error **errp)
1307 {
1308 NBDStructuredReadData chunk;
1309 struct iovec iov[] = {
1310 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1311 {.iov_base = data, .iov_len = size}
1312 };
1313
1314 assert(size);
1315 trace_nbd_co_send_structured_read(handle, offset, data, size);
1316 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_OFFSET_DATA,
1317 handle, sizeof(chunk) - sizeof(chunk.h) + size);
1318 stq_be_p(&chunk.offset, offset);
1319
1320 return nbd_co_send_iov(client, iov, 2, errp);
1321 }
1322
1323 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1324 uint64_t handle,
1325 uint32_t error,
1326 const char *msg,
1327 Error **errp)
1328 {
1329 NBDStructuredError chunk;
1330 int nbd_err = system_errno_to_nbd_errno(error);
1331 struct iovec iov[] = {
1332 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1333 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1334 };
1335
1336 assert(nbd_err);
1337 trace_nbd_co_send_structured_error(handle, nbd_err,
1338 nbd_err_lookup(nbd_err), msg ? msg : "");
1339 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1340 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1341 stl_be_p(&chunk.error, nbd_err);
1342 stw_be_p(&chunk.message_length, iov[1].iov_len);
1343
1344 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1345 }
1346
1347 /* nbd_co_receive_request
1348 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1349 * connection right away, and any other negative value to report an error to
1350 * the client (although the caller may still need to disconnect after reporting
1351 * the error).
1352 */
1353 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
1354 Error **errp)
1355 {
1356 NBDClient *client = req->client;
1357 int valid_flags;
1358
1359 g_assert(qemu_in_coroutine());
1360 assert(client->recv_coroutine == qemu_coroutine_self());
1361 if (nbd_receive_request(client->ioc, request, errp) < 0) {
1362 return -EIO;
1363 }
1364
1365 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
1366 nbd_cmd_lookup(request->type));
1367
1368 if (request->type != NBD_CMD_WRITE) {
1369 /* No payload, we are ready to read the next request. */
1370 req->complete = true;
1371 }
1372
1373 if (request->type == NBD_CMD_DISC) {
1374 /* Special case: we're going to disconnect without a reply,
1375 * whether or not flags, from, or len are bogus */
1376 return -EIO;
1377 }
1378
1379 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) {
1380 if (request->len > NBD_MAX_BUFFER_SIZE) {
1381 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1382 request->len, NBD_MAX_BUFFER_SIZE);
1383 return -EINVAL;
1384 }
1385
1386 req->data = blk_try_blockalign(client->exp->blk, request->len);
1387 if (req->data == NULL) {
1388 error_setg(errp, "No memory");
1389 return -ENOMEM;
1390 }
1391 }
1392 if (request->type == NBD_CMD_WRITE) {
1393 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
1394 error_prepend(errp, "reading from socket failed: ");
1395 return -EIO;
1396 }
1397 req->complete = true;
1398
1399 trace_nbd_co_receive_request_payload_received(request->handle,
1400 request->len);
1401 }
1402
1403 /* Sanity checks. */
1404 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
1405 (request->type == NBD_CMD_WRITE ||
1406 request->type == NBD_CMD_WRITE_ZEROES ||
1407 request->type == NBD_CMD_TRIM)) {
1408 error_setg(errp, "Export is read-only");
1409 return -EROFS;
1410 }
1411 if (request->from > client->exp->size ||
1412 request->from + request->len > client->exp->size) {
1413 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
1414 ", Size: %" PRIu64, request->from, request->len,
1415 (uint64_t)client->exp->size);
1416 return (request->type == NBD_CMD_WRITE ||
1417 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
1418 }
1419 valid_flags = NBD_CMD_FLAG_FUA;
1420 if (request->type == NBD_CMD_READ && client->structured_reply) {
1421 valid_flags |= NBD_CMD_FLAG_DF;
1422 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
1423 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
1424 }
1425 if (request->flags & ~valid_flags) {
1426 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
1427 nbd_cmd_lookup(request->type), request->flags);
1428 return -EINVAL;
1429 }
1430
1431 return 0;
1432 }
1433
1434 /* Owns a reference to the NBDClient passed as opaque. */
1435 static coroutine_fn void nbd_trip(void *opaque)
1436 {
1437 NBDClient *client = opaque;
1438 NBDExport *exp = client->exp;
1439 NBDRequestData *req;
1440 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
1441 int ret;
1442 int flags;
1443 int reply_data_len = 0;
1444 Error *local_err = NULL;
1445 char *msg = NULL;
1446
1447 trace_nbd_trip();
1448 if (client->closing) {
1449 nbd_client_put(client);
1450 return;
1451 }
1452
1453 req = nbd_request_get(client);
1454 ret = nbd_co_receive_request(req, &request, &local_err);
1455 client->recv_coroutine = NULL;
1456 nbd_client_receive_next_request(client);
1457 if (ret == -EIO) {
1458 goto disconnect;
1459 }
1460
1461 if (ret < 0) {
1462 goto reply;
1463 }
1464
1465 if (client->closing) {
1466 /*
1467 * The client may be closed when we are blocked in
1468 * nbd_co_receive_request()
1469 */
1470 goto done;
1471 }
1472
1473 switch (request.type) {
1474 case NBD_CMD_READ:
1475 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1476 if (request.flags & NBD_CMD_FLAG_FUA) {
1477 ret = blk_co_flush(exp->blk);
1478 if (ret < 0) {
1479 error_setg_errno(&local_err, -ret, "flush failed");
1480 break;
1481 }
1482 }
1483
1484 ret = blk_pread(exp->blk, request.from + exp->dev_offset,
1485 req->data, request.len);
1486 if (ret < 0) {
1487 error_setg_errno(&local_err, -ret, "reading from file failed");
1488 break;
1489 }
1490
1491 reply_data_len = request.len;
1492
1493 break;
1494 case NBD_CMD_WRITE:
1495 flags = 0;
1496 if (request.flags & NBD_CMD_FLAG_FUA) {
1497 flags |= BDRV_REQ_FUA;
1498 }
1499 ret = blk_pwrite(exp->blk, request.from + exp->dev_offset,
1500 req->data, request.len, flags);
1501 if (ret < 0) {
1502 error_setg_errno(&local_err, -ret, "writing to file failed");
1503 }
1504
1505 break;
1506 case NBD_CMD_WRITE_ZEROES:
1507 flags = 0;
1508 if (request.flags & NBD_CMD_FLAG_FUA) {
1509 flags |= BDRV_REQ_FUA;
1510 }
1511 if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) {
1512 flags |= BDRV_REQ_MAY_UNMAP;
1513 }
1514 ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset,
1515 request.len, flags);
1516 if (ret < 0) {
1517 error_setg_errno(&local_err, -ret, "writing to file failed");
1518 }
1519
1520 break;
1521 case NBD_CMD_DISC:
1522 /* unreachable, thanks to special case in nbd_co_receive_request() */
1523 abort();
1524
1525 case NBD_CMD_FLUSH:
1526 ret = blk_co_flush(exp->blk);
1527 if (ret < 0) {
1528 error_setg_errno(&local_err, -ret, "flush failed");
1529 }
1530
1531 break;
1532 case NBD_CMD_TRIM:
1533 ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset,
1534 request.len);
1535 if (ret < 0) {
1536 error_setg_errno(&local_err, -ret, "discard failed");
1537 }
1538
1539 break;
1540 default:
1541 error_setg(&local_err, "invalid request type (%" PRIu32 ") received",
1542 request.type);
1543 ret = -EINVAL;
1544 }
1545
1546 reply:
1547 if (local_err) {
1548 /* If we get here, local_err was not a fatal error, and should be sent
1549 * to the client. */
1550 assert(ret < 0);
1551 msg = g_strdup(error_get_pretty(local_err));
1552 error_report_err(local_err);
1553 local_err = NULL;
1554 }
1555
1556 if (client->structured_reply &&
1557 (ret < 0 || request.type == NBD_CMD_READ)) {
1558 if (ret < 0) {
1559 ret = nbd_co_send_structured_error(req->client, request.handle,
1560 -ret, msg, &local_err);
1561 } else if (reply_data_len) {
1562 ret = nbd_co_send_structured_read(req->client, request.handle,
1563 request.from, req->data,
1564 reply_data_len, &local_err);
1565 } else {
1566 ret = nbd_co_send_structured_done(req->client, request.handle,
1567 &local_err);
1568 }
1569 } else {
1570 ret = nbd_co_send_simple_reply(req->client, request.handle,
1571 ret < 0 ? -ret : 0,
1572 req->data, reply_data_len, &local_err);
1573 }
1574 g_free(msg);
1575 if (ret < 0) {
1576 error_prepend(&local_err, "Failed to send reply: ");
1577 goto disconnect;
1578 }
1579
1580 /* We must disconnect after NBD_CMD_WRITE if we did not
1581 * read the payload.
1582 */
1583 if (!req->complete) {
1584 error_setg(&local_err, "Request handling failed in intermediate state");
1585 goto disconnect;
1586 }
1587
1588 done:
1589 nbd_request_put(req);
1590 nbd_client_put(client);
1591 return;
1592
1593 disconnect:
1594 if (local_err) {
1595 error_reportf_err(local_err, "Disconnect client, due to: ");
1596 }
1597 nbd_request_put(req);
1598 client_close(client, true);
1599 nbd_client_put(client);
1600 }
1601
1602 static void nbd_client_receive_next_request(NBDClient *client)
1603 {
1604 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
1605 nbd_client_get(client);
1606 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
1607 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
1608 }
1609 }
1610
1611 static coroutine_fn void nbd_co_client_start(void *opaque)
1612 {
1613 NBDClient *client = opaque;
1614 NBDExport *exp = client->exp;
1615 Error *local_err = NULL;
1616
1617 if (exp) {
1618 nbd_export_get(exp);
1619 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1620 }
1621 qemu_co_mutex_init(&client->send_lock);
1622
1623 if (nbd_negotiate(client, &local_err)) {
1624 if (local_err) {
1625 error_report_err(local_err);
1626 }
1627 client_close(client, false);
1628 return;
1629 }
1630
1631 nbd_client_receive_next_request(client);
1632 }
1633
1634 /*
1635 * Create a new client listener on the given export @exp, using the
1636 * given channel @sioc. Begin servicing it in a coroutine. When the
1637 * connection closes, call @close_fn with an indication of whether the
1638 * client completed negotiation.
1639 */
1640 void nbd_client_new(NBDExport *exp,
1641 QIOChannelSocket *sioc,
1642 QCryptoTLSCreds *tlscreds,
1643 const char *tlsaclname,
1644 void (*close_fn)(NBDClient *, bool))
1645 {
1646 NBDClient *client;
1647 Coroutine *co;
1648
1649 client = g_new0(NBDClient, 1);
1650 client->refcount = 1;
1651 client->exp = exp;
1652 client->tlscreds = tlscreds;
1653 if (tlscreds) {
1654 object_ref(OBJECT(client->tlscreds));
1655 }
1656 client->tlsaclname = g_strdup(tlsaclname);
1657 client->sioc = sioc;
1658 object_ref(OBJECT(client->sioc));
1659 client->ioc = QIO_CHANNEL(sioc);
1660 object_ref(OBJECT(client->ioc));
1661 client->close_fn = close_fn;
1662
1663 co = qemu_coroutine_create(nbd_co_client_start, client);
1664 qemu_coroutine_enter(co);
1665 }
QEmu