?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nbd/server: CVE-2017-15119 Reject options larger than 32M
[qemu/ar7.git] / nbd / server.c
bloba81801e3bc8ef33f034edde13e784bffe16dd433
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 (nbd_read(client->ioc, name, namelen, errp) < 0) {
390 return -EIO;
391 }
392 name[namelen] = '\0';
393 length -= namelen;
394 trace_nbd_negotiate_handle_export_name_request(name);
395
396 if (nbd_read(client->ioc, &requests, sizeof(requests), errp) < 0) {
397 return -EIO;
398 }
399 be16_to_cpus(&requests);
400 length -= sizeof(requests);
401 trace_nbd_negotiate_handle_info_requests(requests);
402 if (requests != length / sizeof(request)) {
403 msg = "incorrect number of requests for overall length";
404 goto invalid;
405 }
406 while (requests--) {
407 if (nbd_read(client->ioc, &request, sizeof(request), errp) < 0) {
408 return -EIO;
409 }
410 be16_to_cpus(&request);
411 length -= sizeof(request);
412 trace_nbd_negotiate_handle_info_request(request,
413 nbd_info_lookup(request));
414 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
415 * everything else is either a request we don't know or
416 * something we send regardless of request */
417 switch (request) {
418 case NBD_INFO_NAME:
419 sendname = true;
420 break;
421 case NBD_INFO_BLOCK_SIZE:
422 blocksize = true;
423 break;
424 }
425 }
426 assert(length == 0);
427
428 exp = nbd_export_find(name);
429 if (!exp) {
430 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_UNKNOWN,
431 opt, errp, "export '%s' not present",
432 name);
433 }
434
435 /* Don't bother sending NBD_INFO_NAME unless client requested it */
436 if (sendname) {
437 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, namelen, name,
438 errp);
439 if (rc < 0) {
440 return rc;
441 }
442 }
443
444 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
445 * client request */
446 if (exp->description) {
447 size_t len = strlen(exp->description);
448
449 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_DESCRIPTION,
450 len, exp->description, errp);
451 if (rc < 0) {
452 return rc;
453 }
454 }
455
456 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
457 * according to whether the client requested it, and according to
458 * whether this is OPT_INFO or OPT_GO. */
459 /* minimum - 1 for back-compat, or 512 if client is new enough.
460 * TODO: consult blk_bs(blk)->bl.request_alignment? */
461 sizes[0] = (opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
462 /* preferred - Hard-code to 4096 for now.
463 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
464 sizes[1] = 4096;
465 /* maximum - At most 32M, but smaller as appropriate. */
466 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
467 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
468 cpu_to_be32s(&sizes[0]);
469 cpu_to_be32s(&sizes[1]);
470 cpu_to_be32s(&sizes[2]);
471 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_BLOCK_SIZE,
472 sizeof(sizes), sizes, errp);
473 if (rc < 0) {
474 return rc;
475 }
476
477 /* Send NBD_INFO_EXPORT always */
478 trace_nbd_negotiate_new_style_size_flags(exp->size,
479 exp->nbdflags | myflags);
480 stq_be_p(buf, exp->size);
481 stw_be_p(buf + 8, exp->nbdflags | myflags);
482 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_EXPORT,
483 sizeof(buf), buf, errp);
484 if (rc < 0) {
485 return rc;
486 }
487
488 /* If the client is just asking for NBD_OPT_INFO, but forgot to
489 * request block sizes, return an error.
490 * TODO: consult blk_bs(blk)->request_align, and only error if it
491 * is not 1? */
492 if (opt == NBD_OPT_INFO && !blocksize) {
493 return nbd_negotiate_send_rep_err(client->ioc,
494 NBD_REP_ERR_BLOCK_SIZE_REQD, opt,
495 errp,
496 "request NBD_INFO_BLOCK_SIZE to "
497 "use this export");
498 }
499
500 /* Final reply */
501 rc = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, opt, errp);
502 if (rc < 0) {
503 return rc;
504 }
505
506 if (opt == NBD_OPT_GO) {
507 client->exp = exp;
508 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
509 nbd_export_get(client->exp);
510 rc = 1;
511 }
512 return rc;
513
514 invalid:
515 if (nbd_drop(client->ioc, length, errp) < 0) {
516 return -EIO;
517 }
518 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID, opt,
519 errp, "%s", msg);
520 }
521
522
523 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
524 * new channel for all further (now-encrypted) communication. */
525 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
526 Error **errp)
527 {
528 QIOChannel *ioc;
529 QIOChannelTLS *tioc;
530 struct NBDTLSHandshakeData data = { 0 };
531
532 trace_nbd_negotiate_handle_starttls();
533 ioc = client->ioc;
534
535 if (nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
536 NBD_OPT_STARTTLS, errp) < 0) {
537 return NULL;
538 }
539
540 tioc = qio_channel_tls_new_server(ioc,
541 client->tlscreds,
542 client->tlsaclname,
543 errp);
544 if (!tioc) {
545 return NULL;
546 }
547
548 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
549 trace_nbd_negotiate_handle_starttls_handshake();
550 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
551 qio_channel_tls_handshake(tioc,
552 nbd_tls_handshake,
553 &data,
554 NULL);
555
556 if (!data.complete) {
557 g_main_loop_run(data.loop);
558 }
559 g_main_loop_unref(data.loop);
560 if (data.error) {
561 object_unref(OBJECT(tioc));
562 error_propagate(errp, data.error);
563 return NULL;
564 }
565
566 return QIO_CHANNEL(tioc);
567 }
568
569 /* nbd_reject_length: Handle any unexpected payload.
570 * @fatal requests that we quit talking to the client, even if we are able
571 * to successfully send an error to the guest.
572 * Return:
573 * -errno transmission error occurred or @fatal was requested, errp is set
574 * 0 error message successfully sent to client, errp is not set
575 */
576 static int nbd_reject_length(NBDClient *client, uint32_t length,
577 uint32_t option, bool fatal, Error **errp)
578 {
579 int ret;
580
581 assert(length);
582 if (nbd_drop(client->ioc, length, errp) < 0) {
583 return -EIO;
584 }
585 ret = nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID,
586 option, errp,
587 "option '%s' should have zero length",
588 nbd_opt_lookup(option));
589 if (fatal && !ret) {
590 error_setg(errp, "option '%s' should have zero length",
591 nbd_opt_lookup(option));
592 return -EINVAL;
593 }
594 return ret;
595 }
596
597 /* nbd_negotiate_options
598 * Process all NBD_OPT_* client option commands, during fixed newstyle
599 * negotiation.
600 * Return:
601 * -errno on error, errp is set
602 * 0 on successful negotiation, errp is not set
603 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
604 * errp is not set
605 */
606 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
607 Error **errp)
608 {
609 uint32_t flags;
610 bool fixedNewstyle = false;
611 bool no_zeroes = false;
612
613 /* Client sends:
614 [ 0 .. 3] client flags
615
616 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
617 [ 0 .. 7] NBD_OPTS_MAGIC
618 [ 8 .. 11] NBD option
619 [12 .. 15] Data length
620 ... Rest of request
621
622 [ 0 .. 7] NBD_OPTS_MAGIC
623 [ 8 .. 11] Second NBD option
624 [12 .. 15] Data length
625 ... Rest of request
626 */
627
628 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
629 error_prepend(errp, "read failed: ");
630 return -EIO;
631 }
632 be32_to_cpus(&flags);
633 trace_nbd_negotiate_options_flags(flags);
634 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
635 fixedNewstyle = true;
636 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
637 }
638 if (flags & NBD_FLAG_C_NO_ZEROES) {
639 no_zeroes = true;
640 flags &= ~NBD_FLAG_C_NO_ZEROES;
641 }
642 if (flags != 0) {
643 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
644 return -EINVAL;
645 }
646
647 while (1) {
648 int ret;
649 uint32_t option, length;
650 uint64_t magic;
651
652 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
653 error_prepend(errp, "read failed: ");
654 return -EINVAL;
655 }
656 magic = be64_to_cpu(magic);
657 trace_nbd_negotiate_options_check_magic(magic);
658 if (magic != NBD_OPTS_MAGIC) {
659 error_setg(errp, "Bad magic received");
660 return -EINVAL;
661 }
662
663 if (nbd_read(client->ioc, &option,
664 sizeof(option), errp) < 0) {
665 error_prepend(errp, "read failed: ");
666 return -EINVAL;
667 }
668 option = be32_to_cpu(option);
669
670 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
671 error_prepend(errp, "read failed: ");
672 return -EINVAL;
673 }
674 length = be32_to_cpu(length);
675
676 if (length > NBD_MAX_BUFFER_SIZE) {
677 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
678 length, NBD_MAX_BUFFER_SIZE);
679 return -EINVAL;
680 }
681
682 trace_nbd_negotiate_options_check_option(option,
683 nbd_opt_lookup(option));
684 if (client->tlscreds &&
685 client->ioc == (QIOChannel *)client->sioc) {
686 QIOChannel *tioc;
687 if (!fixedNewstyle) {
688 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
689 return -EINVAL;
690 }
691 switch (option) {
692 case NBD_OPT_STARTTLS:
693 if (length) {
694 /* Unconditionally drop the connection if the client
695 * can't start a TLS negotiation correctly */
696 return nbd_reject_length(client, length, option, true,
697 errp);
698 }
699 tioc = nbd_negotiate_handle_starttls(client, errp);
700 if (!tioc) {
701 return -EIO;
702 }
703 ret = 0;
704 object_unref(OBJECT(client->ioc));
705 client->ioc = QIO_CHANNEL(tioc);
706 break;
707
708 case NBD_OPT_EXPORT_NAME:
709 /* No way to return an error to client, so drop connection */
710 error_setg(errp, "Option 0x%x not permitted before TLS",
711 option);
712 return -EINVAL;
713
714 default:
715 if (nbd_drop(client->ioc, length, errp) < 0) {
716 return -EIO;
717 }
718 ret = nbd_negotiate_send_rep_err(client->ioc,
719 NBD_REP_ERR_TLS_REQD,
720 option, errp,
721 "Option 0x%" PRIx32
722 "not permitted before TLS",
723 option);
724 /* Let the client keep trying, unless they asked to
725 * quit. In this mode, we've already sent an error, so
726 * we can't ack the abort. */
727 if (option == NBD_OPT_ABORT) {
728 return 1;
729 }
730 break;
731 }
732 } else if (fixedNewstyle) {
733 switch (option) {
734 case NBD_OPT_LIST:
735 if (length) {
736 ret = nbd_reject_length(client, length, option, false,
737 errp);
738 } else {
739 ret = nbd_negotiate_handle_list(client, errp);
740 }
741 break;
742
743 case NBD_OPT_ABORT:
744 /* NBD spec says we must try to reply before
745 * disconnecting, but that we must also tolerate
746 * guests that don't wait for our reply. */
747 nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, option, NULL);
748 return 1;
749
750 case NBD_OPT_EXPORT_NAME:
751 return nbd_negotiate_handle_export_name(client, length,
752 myflags, no_zeroes,
753 errp);
754
755 case NBD_OPT_INFO:
756 case NBD_OPT_GO:
757 ret = nbd_negotiate_handle_info(client, length, option,
758 myflags, errp);
759 if (ret == 1) {
760 assert(option == NBD_OPT_GO);
761 return 0;
762 }
763 break;
764
765 case NBD_OPT_STARTTLS:
766 if (length) {
767 ret = nbd_reject_length(client, length, option, false,
768 errp);
769 } else if (client->tlscreds) {
770 ret = nbd_negotiate_send_rep_err(client->ioc,
771 NBD_REP_ERR_INVALID,
772 option, errp,
773 "TLS already enabled");
774 } else {
775 ret = nbd_negotiate_send_rep_err(client->ioc,
776 NBD_REP_ERR_POLICY,
777 option, errp,
778 "TLS not configured");
779 }
780 break;
781
782 case NBD_OPT_STRUCTURED_REPLY:
783 if (length) {
784 ret = nbd_reject_length(client, length, option, false,
785 errp);
786 } else if (client->structured_reply) {
787 ret = nbd_negotiate_send_rep_err(
788 client->ioc, NBD_REP_ERR_INVALID, option, errp,
789 "structured reply already negotiated");
790 } else {
791 ret = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
792 option, errp);
793 client->structured_reply = true;
794 myflags |= NBD_FLAG_SEND_DF;
795 }
796 break;
797
798 default:
799 if (nbd_drop(client->ioc, length, errp) < 0) {
800 return -EIO;
801 }
802 ret = nbd_negotiate_send_rep_err(client->ioc,
803 NBD_REP_ERR_UNSUP,
804 option, errp,
805 "Unsupported option 0x%"
806 PRIx32 " (%s)", option,
807 nbd_opt_lookup(option));
808 break;
809 }
810 } else {
811 /*
812 * If broken new-style we should drop the connection
813 * for anything except NBD_OPT_EXPORT_NAME
814 */
815 switch (option) {
816 case NBD_OPT_EXPORT_NAME:
817 return nbd_negotiate_handle_export_name(client, length,
818 myflags, no_zeroes,
819 errp);
820
821 default:
822 error_setg(errp, "Unsupported option 0x%" PRIx32 " (%s)",
823 option, nbd_opt_lookup(option));
824 return -EINVAL;
825 }
826 }
827 if (ret < 0) {
828 return ret;
829 }
830 }
831 }
832
833 /* nbd_negotiate
834 * Return:
835 * -errno on error, errp is set
836 * 0 on successful negotiation, errp is not set
837 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
838 * errp is not set
839 */
840 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
841 {
842 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
843 int ret;
844 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
845 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
846 NBD_FLAG_SEND_WRITE_ZEROES);
847 bool oldStyle;
848
849 /* Old style negotiation header, no room for options
850 [ 0 .. 7] passwd ("NBDMAGIC")
851 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
852 [16 .. 23] size
853 [24 .. 27] export flags (zero-extended)
854 [28 .. 151] reserved (0)
855
856 New style negotiation header, client can send options
857 [ 0 .. 7] passwd ("NBDMAGIC")
858 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
859 [16 .. 17] server flags (0)
860 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
861 */
862
863 qio_channel_set_blocking(client->ioc, false, NULL);
864
865 trace_nbd_negotiate_begin();
866 memcpy(buf, "NBDMAGIC", 8);
867
868 oldStyle = client->exp != NULL && !client->tlscreds;
869 if (oldStyle) {
870 trace_nbd_negotiate_old_style(client->exp->size,
871 client->exp->nbdflags | myflags);
872 stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
873 stq_be_p(buf + 16, client->exp->size);
874 stl_be_p(buf + 24, client->exp->nbdflags | myflags);
875
876 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
877 error_prepend(errp, "write failed: ");
878 return -EINVAL;
879 }
880 } else {
881 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
882 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
883
884 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
885 error_prepend(errp, "write failed: ");
886 return -EINVAL;
887 }
888 ret = nbd_negotiate_options(client, myflags, errp);
889 if (ret != 0) {
890 if (ret < 0) {
891 error_prepend(errp, "option negotiation failed: ");
892 }
893 return ret;
894 }
895 }
896
897 trace_nbd_negotiate_success();
898
899 return 0;
900 }
901
902 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
903 Error **errp)
904 {
905 uint8_t buf[NBD_REQUEST_SIZE];
906 uint32_t magic;
907 int ret;
908
909 ret = nbd_read(ioc, buf, sizeof(buf), errp);
910 if (ret < 0) {
911 return ret;
912 }
913
914 /* Request
915 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
916 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
917 [ 6 .. 7] type (NBD_CMD_READ, ...)
918 [ 8 .. 15] handle
919 [16 .. 23] from
920 [24 .. 27] len
921 */
922
923 magic = ldl_be_p(buf);
924 request->flags = lduw_be_p(buf + 4);
925 request->type = lduw_be_p(buf + 6);
926 request->handle = ldq_be_p(buf + 8);
927 request->from = ldq_be_p(buf + 16);
928 request->len = ldl_be_p(buf + 24);
929
930 trace_nbd_receive_request(magic, request->flags, request->type,
931 request->from, request->len);
932
933 if (magic != NBD_REQUEST_MAGIC) {
934 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
935 return -EINVAL;
936 }
937 return 0;
938 }
939
940 #define MAX_NBD_REQUESTS 16
941
942 void nbd_client_get(NBDClient *client)
943 {
944 client->refcount++;
945 }
946
947 void nbd_client_put(NBDClient *client)
948 {
949 if (--client->refcount == 0) {
950 /* The last reference should be dropped by client->close,
951 * which is called by client_close.
952 */
953 assert(client->closing);
954
955 qio_channel_detach_aio_context(client->ioc);
956 object_unref(OBJECT(client->sioc));
957 object_unref(OBJECT(client->ioc));
958 if (client->tlscreds) {
959 object_unref(OBJECT(client->tlscreds));
960 }
961 g_free(client->tlsaclname);
962 if (client->exp) {
963 QTAILQ_REMOVE(&client->exp->clients, client, next);
964 nbd_export_put(client->exp);
965 }
966 g_free(client);
967 }
968 }
969
970 static void client_close(NBDClient *client, bool negotiated)
971 {
972 if (client->closing) {
973 return;
974 }
975
976 client->closing = true;
977
978 /* Force requests to finish. They will drop their own references,
979 * then we'll close the socket and free the NBDClient.
980 */
981 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
982 NULL);
983
984 /* Also tell the client, so that they release their reference. */
985 if (client->close_fn) {
986 client->close_fn(client, negotiated);
987 }
988 }
989
990 static NBDRequestData *nbd_request_get(NBDClient *client)
991 {
992 NBDRequestData *req;
993
994 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
995 client->nb_requests++;
996
997 req = g_new0(NBDRequestData, 1);
998 nbd_client_get(client);
999 req->client = client;
1000 return req;
1001 }
1002
1003 static void nbd_request_put(NBDRequestData *req)
1004 {
1005 NBDClient *client = req->client;
1006
1007 if (req->data) {
1008 qemu_vfree(req->data);
1009 }
1010 g_free(req);
1011
1012 client->nb_requests--;
1013 nbd_client_receive_next_request(client);
1014
1015 nbd_client_put(client);
1016 }
1017
1018 static void blk_aio_attached(AioContext *ctx, void *opaque)
1019 {
1020 NBDExport *exp = opaque;
1021 NBDClient *client;
1022
1023 trace_nbd_blk_aio_attached(exp->name, ctx);
1024
1025 exp->ctx = ctx;
1026
1027 QTAILQ_FOREACH(client, &exp->clients, next) {
1028 qio_channel_attach_aio_context(client->ioc, ctx);
1029 if (client->recv_coroutine) {
1030 aio_co_schedule(ctx, client->recv_coroutine);
1031 }
1032 if (client->send_coroutine) {
1033 aio_co_schedule(ctx, client->send_coroutine);
1034 }
1035 }
1036 }
1037
1038 static void blk_aio_detach(void *opaque)
1039 {
1040 NBDExport *exp = opaque;
1041 NBDClient *client;
1042
1043 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1044
1045 QTAILQ_FOREACH(client, &exp->clients, next) {
1046 qio_channel_detach_aio_context(client->ioc);
1047 }
1048
1049 exp->ctx = NULL;
1050 }
1051
1052 static void nbd_eject_notifier(Notifier *n, void *data)
1053 {
1054 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1055 nbd_export_close(exp);
1056 }
1057
1058 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1059 uint16_t nbdflags, void (*close)(NBDExport *),
1060 bool writethrough, BlockBackend *on_eject_blk,
1061 Error **errp)
1062 {
1063 AioContext *ctx;
1064 BlockBackend *blk;
1065 NBDExport *exp = g_new0(NBDExport, 1);
1066 uint64_t perm;
1067 int ret;
1068
1069 /*
1070 * NBD exports are used for non-shared storage migration. Make sure
1071 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1072 * access since the export could be available before migration handover.
1073 */
1074 ctx = bdrv_get_aio_context(bs);
1075 aio_context_acquire(ctx);
1076 bdrv_invalidate_cache(bs, NULL);
1077 aio_context_release(ctx);
1078
1079 /* Don't allow resize while the NBD server is running, otherwise we don't
1080 * care what happens with the node. */
1081 perm = BLK_PERM_CONSISTENT_READ;
1082 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1083 perm |= BLK_PERM_WRITE;
1084 }
1085 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1086 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1087 ret = blk_insert_bs(blk, bs, errp);
1088 if (ret < 0) {
1089 goto fail;
1090 }
1091 blk_set_enable_write_cache(blk, !writethrough);
1092
1093 exp->refcount = 1;
1094 QTAILQ_INIT(&exp->clients);
1095 exp->blk = blk;
1096 exp->dev_offset = dev_offset;
1097 exp->nbdflags = nbdflags;
1098 exp->size = size < 0 ? blk_getlength(blk) : size;
1099 if (exp->size < 0) {
1100 error_setg_errno(errp, -exp->size,
1101 "Failed to determine the NBD export's length");
1102 goto fail;
1103 }
1104 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1105
1106 exp->close = close;
1107 exp->ctx = blk_get_aio_context(blk);
1108 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1109
1110 if (on_eject_blk) {
1111 blk_ref(on_eject_blk);
1112 exp->eject_notifier_blk = on_eject_blk;
1113 exp->eject_notifier.notify = nbd_eject_notifier;
1114 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1115 }
1116 return exp;
1117
1118 fail:
1119 blk_unref(blk);
1120 g_free(exp);
1121 return NULL;
1122 }
1123
1124 NBDExport *nbd_export_find(const char *name)
1125 {
1126 NBDExport *exp;
1127 QTAILQ_FOREACH(exp, &exports, next) {
1128 if (strcmp(name, exp->name) == 0) {
1129 return exp;
1130 }
1131 }
1132
1133 return NULL;
1134 }
1135
1136 void nbd_export_set_name(NBDExport *exp, const char *name)
1137 {
1138 if (exp->name == name) {
1139 return;
1140 }
1141
1142 nbd_export_get(exp);
1143 if (exp->name != NULL) {
1144 g_free(exp->name);
1145 exp->name = NULL;
1146 QTAILQ_REMOVE(&exports, exp, next);
1147 nbd_export_put(exp);
1148 }
1149 if (name != NULL) {
1150 nbd_export_get(exp);
1151 exp->name = g_strdup(name);
1152 QTAILQ_INSERT_TAIL(&exports, exp, next);
1153 }
1154 nbd_export_put(exp);
1155 }
1156
1157 void nbd_export_set_description(NBDExport *exp, const char *description)
1158 {
1159 g_free(exp->description);
1160 exp->description = g_strdup(description);
1161 }
1162
1163 void nbd_export_close(NBDExport *exp)
1164 {
1165 NBDClient *client, *next;
1166
1167 nbd_export_get(exp);
1168 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1169 client_close(client, true);
1170 }
1171 nbd_export_set_name(exp, NULL);
1172 nbd_export_set_description(exp, NULL);
1173 nbd_export_put(exp);
1174 }
1175
1176 void nbd_export_get(NBDExport *exp)
1177 {
1178 assert(exp->refcount > 0);
1179 exp->refcount++;
1180 }
1181
1182 void nbd_export_put(NBDExport *exp)
1183 {
1184 assert(exp->refcount > 0);
1185 if (exp->refcount == 1) {
1186 nbd_export_close(exp);
1187 }
1188
1189 if (--exp->refcount == 0) {
1190 assert(exp->name == NULL);
1191 assert(exp->description == NULL);
1192
1193 if (exp->close) {
1194 exp->close(exp);
1195 }
1196
1197 if (exp->blk) {
1198 if (exp->eject_notifier_blk) {
1199 notifier_remove(&exp->eject_notifier);
1200 blk_unref(exp->eject_notifier_blk);
1201 }
1202 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1203 blk_aio_detach, exp);
1204 blk_unref(exp->blk);
1205 exp->blk = NULL;
1206 }
1207
1208 g_free(exp);
1209 }
1210 }
1211
1212 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1213 {
1214 return exp->blk;
1215 }
1216
1217 void nbd_export_close_all(void)
1218 {
1219 NBDExport *exp, *next;
1220
1221 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1222 nbd_export_close(exp);
1223 }
1224 }
1225
1226 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1227 unsigned niov, Error **errp)
1228 {
1229 int ret;
1230
1231 g_assert(qemu_in_coroutine());
1232 qemu_co_mutex_lock(&client->send_lock);
1233 client->send_coroutine = qemu_coroutine_self();
1234
1235 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1236
1237 client->send_coroutine = NULL;
1238 qemu_co_mutex_unlock(&client->send_lock);
1239
1240 return ret;
1241 }
1242
1243 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1244 uint64_t handle)
1245 {
1246 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1247 stl_be_p(&reply->error, error);
1248 stq_be_p(&reply->handle, handle);
1249 }
1250
1251 static int nbd_co_send_simple_reply(NBDClient *client,
1252 uint64_t handle,
1253 uint32_t error,
1254 void *data,
1255 size_t len,
1256 Error **errp)
1257 {
1258 NBDSimpleReply reply;
1259 int nbd_err = system_errno_to_nbd_errno(error);
1260 struct iovec iov[] = {
1261 {.iov_base = &reply, .iov_len = sizeof(reply)},
1262 {.iov_base = data, .iov_len = len}
1263 };
1264
1265 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1266 len);
1267 set_be_simple_reply(&reply, nbd_err, handle);
1268
1269 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1270 }
1271
1272 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1273 uint16_t type, uint64_t handle, uint32_t length)
1274 {
1275 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1276 stw_be_p(&chunk->flags, flags);
1277 stw_be_p(&chunk->type, type);
1278 stq_be_p(&chunk->handle, handle);
1279 stl_be_p(&chunk->length, length);
1280 }
1281
1282 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1283 uint64_t handle,
1284 Error **errp)
1285 {
1286 NBDStructuredReplyChunk chunk;
1287 struct iovec iov[] = {
1288 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1289 };
1290
1291 trace_nbd_co_send_structured_done(handle);
1292 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1293
1294 return nbd_co_send_iov(client, iov, 1, errp);
1295 }
1296
1297 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1298 uint64_t handle,
1299 uint64_t offset,
1300 void *data,
1301 size_t size,
1302 Error **errp)
1303 {
1304 NBDStructuredReadData chunk;
1305 struct iovec iov[] = {
1306 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1307 {.iov_base = data, .iov_len = size}
1308 };
1309
1310 assert(size);
1311 trace_nbd_co_send_structured_read(handle, offset, data, size);
1312 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_OFFSET_DATA,
1313 handle, sizeof(chunk) - sizeof(chunk.h) + size);
1314 stq_be_p(&chunk.offset, offset);
1315
1316 return nbd_co_send_iov(client, iov, 2, errp);
1317 }
1318
1319 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1320 uint64_t handle,
1321 uint32_t error,
1322 const char *msg,
1323 Error **errp)
1324 {
1325 NBDStructuredError chunk;
1326 int nbd_err = system_errno_to_nbd_errno(error);
1327 struct iovec iov[] = {
1328 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1329 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1330 };
1331
1332 assert(nbd_err);
1333 trace_nbd_co_send_structured_error(handle, nbd_err,
1334 nbd_err_lookup(nbd_err), msg ? msg : "");
1335 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1336 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1337 stl_be_p(&chunk.error, nbd_err);
1338 stw_be_p(&chunk.message_length, iov[1].iov_len);
1339
1340 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1341 }
1342
1343 /* nbd_co_receive_request
1344 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1345 * connection right away, and any other negative value to report an error to
1346 * the client (although the caller may still need to disconnect after reporting
1347 * the error).
1348 */
1349 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
1350 Error **errp)
1351 {
1352 NBDClient *client = req->client;
1353 int valid_flags;
1354
1355 g_assert(qemu_in_coroutine());
1356 assert(client->recv_coroutine == qemu_coroutine_self());
1357 if (nbd_receive_request(client->ioc, request, errp) < 0) {
1358 return -EIO;
1359 }
1360
1361 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
1362 nbd_cmd_lookup(request->type));
1363
1364 if (request->type != NBD_CMD_WRITE) {
1365 /* No payload, we are ready to read the next request. */
1366 req->complete = true;
1367 }
1368
1369 if (request->type == NBD_CMD_DISC) {
1370 /* Special case: we're going to disconnect without a reply,
1371 * whether or not flags, from, or len are bogus */
1372 return -EIO;
1373 }
1374
1375 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) {
1376 if (request->len > NBD_MAX_BUFFER_SIZE) {
1377 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1378 request->len, NBD_MAX_BUFFER_SIZE);
1379 return -EINVAL;
1380 }
1381
1382 req->data = blk_try_blockalign(client->exp->blk, request->len);
1383 if (req->data == NULL) {
1384 error_setg(errp, "No memory");
1385 return -ENOMEM;
1386 }
1387 }
1388 if (request->type == NBD_CMD_WRITE) {
1389 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
1390 error_prepend(errp, "reading from socket failed: ");
1391 return -EIO;
1392 }
1393 req->complete = true;
1394
1395 trace_nbd_co_receive_request_payload_received(request->handle,
1396 request->len);
1397 }
1398
1399 /* Sanity checks. */
1400 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
1401 (request->type == NBD_CMD_WRITE ||
1402 request->type == NBD_CMD_WRITE_ZEROES ||
1403 request->type == NBD_CMD_TRIM)) {
1404 error_setg(errp, "Export is read-only");
1405 return -EROFS;
1406 }
1407 if (request->from > client->exp->size ||
1408 request->from + request->len > client->exp->size) {
1409 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
1410 ", Size: %" PRIu64, request->from, request->len,
1411 (uint64_t)client->exp->size);
1412 return (request->type == NBD_CMD_WRITE ||
1413 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
1414 }
1415 valid_flags = NBD_CMD_FLAG_FUA;
1416 if (request->type == NBD_CMD_READ && client->structured_reply) {
1417 valid_flags |= NBD_CMD_FLAG_DF;
1418 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
1419 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
1420 }
1421 if (request->flags & ~valid_flags) {
1422 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
1423 nbd_cmd_lookup(request->type), request->flags);
1424 return -EINVAL;
1425 }
1426
1427 return 0;
1428 }
1429
1430 /* Owns a reference to the NBDClient passed as opaque. */
1431 static coroutine_fn void nbd_trip(void *opaque)
1432 {
1433 NBDClient *client = opaque;
1434 NBDExport *exp = client->exp;
1435 NBDRequestData *req;
1436 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
1437 int ret;
1438 int flags;
1439 int reply_data_len = 0;
1440 Error *local_err = NULL;
1441 char *msg = NULL;
1442
1443 trace_nbd_trip();
1444 if (client->closing) {
1445 nbd_client_put(client);
1446 return;
1447 }
1448
1449 req = nbd_request_get(client);
1450 ret = nbd_co_receive_request(req, &request, &local_err);
1451 client->recv_coroutine = NULL;
1452 nbd_client_receive_next_request(client);
1453 if (ret == -EIO) {
1454 goto disconnect;
1455 }
1456
1457 if (ret < 0) {
1458 goto reply;
1459 }
1460
1461 if (client->closing) {
1462 /*
1463 * The client may be closed when we are blocked in
1464 * nbd_co_receive_request()
1465 */
1466 goto done;
1467 }
1468
1469 switch (request.type) {
1470 case NBD_CMD_READ:
1471 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1472 if (request.flags & NBD_CMD_FLAG_FUA) {
1473 ret = blk_co_flush(exp->blk);
1474 if (ret < 0) {
1475 error_setg_errno(&local_err, -ret, "flush failed");
1476 break;
1477 }
1478 }
1479
1480 ret = blk_pread(exp->blk, request.from + exp->dev_offset,
1481 req->data, request.len);
1482 if (ret < 0) {
1483 error_setg_errno(&local_err, -ret, "reading from file failed");
1484 break;
1485 }
1486
1487 reply_data_len = request.len;
1488
1489 break;
1490 case NBD_CMD_WRITE:
1491 flags = 0;
1492 if (request.flags & NBD_CMD_FLAG_FUA) {
1493 flags |= BDRV_REQ_FUA;
1494 }
1495 ret = blk_pwrite(exp->blk, request.from + exp->dev_offset,
1496 req->data, request.len, flags);
1497 if (ret < 0) {
1498 error_setg_errno(&local_err, -ret, "writing to file failed");
1499 }
1500
1501 break;
1502 case NBD_CMD_WRITE_ZEROES:
1503 flags = 0;
1504 if (request.flags & NBD_CMD_FLAG_FUA) {
1505 flags |= BDRV_REQ_FUA;
1506 }
1507 if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) {
1508 flags |= BDRV_REQ_MAY_UNMAP;
1509 }
1510 ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset,
1511 request.len, flags);
1512 if (ret < 0) {
1513 error_setg_errno(&local_err, -ret, "writing to file failed");
1514 }
1515
1516 break;
1517 case NBD_CMD_DISC:
1518 /* unreachable, thanks to special case in nbd_co_receive_request() */
1519 abort();
1520
1521 case NBD_CMD_FLUSH:
1522 ret = blk_co_flush(exp->blk);
1523 if (ret < 0) {
1524 error_setg_errno(&local_err, -ret, "flush failed");
1525 }
1526
1527 break;
1528 case NBD_CMD_TRIM:
1529 ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset,
1530 request.len);
1531 if (ret < 0) {
1532 error_setg_errno(&local_err, -ret, "discard failed");
1533 }
1534
1535 break;
1536 default:
1537 error_setg(&local_err, "invalid request type (%" PRIu32 ") received",
1538 request.type);
1539 ret = -EINVAL;
1540 }
1541
1542 reply:
1543 if (local_err) {
1544 /* If we get here, local_err was not a fatal error, and should be sent
1545 * to the client. */
1546 assert(ret < 0);
1547 msg = g_strdup(error_get_pretty(local_err));
1548 error_report_err(local_err);
1549 local_err = NULL;
1550 }
1551
1552 if (client->structured_reply &&
1553 (ret < 0 || request.type == NBD_CMD_READ)) {
1554 if (ret < 0) {
1555 ret = nbd_co_send_structured_error(req->client, request.handle,
1556 -ret, msg, &local_err);
1557 } else if (reply_data_len) {
1558 ret = nbd_co_send_structured_read(req->client, request.handle,
1559 request.from, req->data,
1560 reply_data_len, &local_err);
1561 } else {
1562 ret = nbd_co_send_structured_done(req->client, request.handle,
1563 &local_err);
1564 }
1565 } else {
1566 ret = nbd_co_send_simple_reply(req->client, request.handle,
1567 ret < 0 ? -ret : 0,
1568 req->data, reply_data_len, &local_err);
1569 }
1570 g_free(msg);
1571 if (ret < 0) {
1572 error_prepend(&local_err, "Failed to send reply: ");
1573 goto disconnect;
1574 }
1575
1576 /* We must disconnect after NBD_CMD_WRITE if we did not
1577 * read the payload.
1578 */
1579 if (!req->complete) {
1580 error_setg(&local_err, "Request handling failed in intermediate state");
1581 goto disconnect;
1582 }
1583
1584 done:
1585 nbd_request_put(req);
1586 nbd_client_put(client);
1587 return;
1588
1589 disconnect:
1590 if (local_err) {
1591 error_reportf_err(local_err, "Disconnect client, due to: ");
1592 }
1593 nbd_request_put(req);
1594 client_close(client, true);
1595 nbd_client_put(client);
1596 }
1597
1598 static void nbd_client_receive_next_request(NBDClient *client)
1599 {
1600 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
1601 nbd_client_get(client);
1602 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
1603 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
1604 }
1605 }
1606
1607 static coroutine_fn void nbd_co_client_start(void *opaque)
1608 {
1609 NBDClient *client = opaque;
1610 NBDExport *exp = client->exp;
1611 Error *local_err = NULL;
1612
1613 if (exp) {
1614 nbd_export_get(exp);
1615 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1616 }
1617 qemu_co_mutex_init(&client->send_lock);
1618
1619 if (nbd_negotiate(client, &local_err)) {
1620 if (local_err) {
1621 error_report_err(local_err);
1622 }
1623 client_close(client, false);
1624 return;
1625 }
1626
1627 nbd_client_receive_next_request(client);
1628 }
1629
1630 /*
1631 * Create a new client listener on the given export @exp, using the
1632 * given channel @sioc. Begin servicing it in a coroutine. When the
1633 * connection closes, call @close_fn with an indication of whether the
1634 * client completed negotiation.
1635 */
1636 void nbd_client_new(NBDExport *exp,
1637 QIOChannelSocket *sioc,
1638 QCryptoTLSCreds *tlscreds,
1639 const char *tlsaclname,
1640 void (*close_fn)(NBDClient *, bool))
1641 {
1642 NBDClient *client;
1643 Coroutine *co;
1644
1645 client = g_new0(NBDClient, 1);
1646 client->refcount = 1;
1647 client->exp = exp;
1648 client->tlscreds = tlscreds;
1649 if (tlscreds) {
1650 object_ref(OBJECT(client->tlscreds));
1651 }
1652 client->tlsaclname = g_strdup(tlsaclname);
1653 client->sioc = sioc;
1654 object_ref(OBJECT(client->sioc));
1655 client->ioc = QIO_CHANNEL(sioc);
1656 object_ref(OBJECT(client->ioc));
1657 client->close_fn = close_fn;
1658
1659 co = qemu_coroutine_create(nbd_co_client_start, client);
1660 qemu_coroutine_enter(co);
1661 }
AR7 emulation for QEMU