?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
build: allow setting a custom GIT binary for transparent proxying
[qemu/ar7.git] / nbd / server.c
blob70b40ed27e21c2a63f8d33da8268cd163f3867a0
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
427 exp = nbd_export_find(name);
428 if (!exp) {
429 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_UNKNOWN,
430 opt, errp, "export '%s' not present",
431 name);
432 }
433
434 /* Don't bother sending NBD_INFO_NAME unless client requested it */
435 if (sendname) {
436 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_NAME, length, name,
437 errp);
438 if (rc < 0) {
439 return rc;
440 }
441 }
442
443 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
444 * client request */
445 if (exp->description) {
446 size_t len = strlen(exp->description);
447
448 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_DESCRIPTION,
449 len, exp->description, errp);
450 if (rc < 0) {
451 return rc;
452 }
453 }
454
455 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
456 * according to whether the client requested it, and according to
457 * whether this is OPT_INFO or OPT_GO. */
458 /* minimum - 1 for back-compat, or 512 if client is new enough.
459 * TODO: consult blk_bs(blk)->bl.request_alignment? */
460 sizes[0] = (opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
461 /* preferred - Hard-code to 4096 for now.
462 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
463 sizes[1] = 4096;
464 /* maximum - At most 32M, but smaller as appropriate. */
465 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
466 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
467 cpu_to_be32s(&sizes[0]);
468 cpu_to_be32s(&sizes[1]);
469 cpu_to_be32s(&sizes[2]);
470 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_BLOCK_SIZE,
471 sizeof(sizes), sizes, errp);
472 if (rc < 0) {
473 return rc;
474 }
475
476 /* Send NBD_INFO_EXPORT always */
477 trace_nbd_negotiate_new_style_size_flags(exp->size,
478 exp->nbdflags | myflags);
479 stq_be_p(buf, exp->size);
480 stw_be_p(buf + 8, exp->nbdflags | myflags);
481 rc = nbd_negotiate_send_info(client, opt, NBD_INFO_EXPORT,
482 sizeof(buf), buf, errp);
483 if (rc < 0) {
484 return rc;
485 }
486
487 /* If the client is just asking for NBD_OPT_INFO, but forgot to
488 * request block sizes, return an error.
489 * TODO: consult blk_bs(blk)->request_align, and only error if it
490 * is not 1? */
491 if (opt == NBD_OPT_INFO && !blocksize) {
492 return nbd_negotiate_send_rep_err(client->ioc,
493 NBD_REP_ERR_BLOCK_SIZE_REQD, opt,
494 errp,
495 "request NBD_INFO_BLOCK_SIZE to "
496 "use this export");
497 }
498
499 /* Final reply */
500 rc = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, opt, errp);
501 if (rc < 0) {
502 return rc;
503 }
504
505 if (opt == NBD_OPT_GO) {
506 client->exp = exp;
507 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
508 nbd_export_get(client->exp);
509 rc = 1;
510 }
511 return rc;
512
513 invalid:
514 if (nbd_drop(client->ioc, length, errp) < 0) {
515 return -EIO;
516 }
517 return nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID, opt,
518 errp, "%s", msg);
519 }
520
521
522 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
523 * new channel for all further (now-encrypted) communication. */
524 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
525 Error **errp)
526 {
527 QIOChannel *ioc;
528 QIOChannelTLS *tioc;
529 struct NBDTLSHandshakeData data = { 0 };
530
531 trace_nbd_negotiate_handle_starttls();
532 ioc = client->ioc;
533
534 if (nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
535 NBD_OPT_STARTTLS, errp) < 0) {
536 return NULL;
537 }
538
539 tioc = qio_channel_tls_new_server(ioc,
540 client->tlscreds,
541 client->tlsaclname,
542 errp);
543 if (!tioc) {
544 return NULL;
545 }
546
547 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
548 trace_nbd_negotiate_handle_starttls_handshake();
549 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
550 qio_channel_tls_handshake(tioc,
551 nbd_tls_handshake,
552 &data,
553 NULL);
554
555 if (!data.complete) {
556 g_main_loop_run(data.loop);
557 }
558 g_main_loop_unref(data.loop);
559 if (data.error) {
560 object_unref(OBJECT(tioc));
561 error_propagate(errp, data.error);
562 return NULL;
563 }
564
565 return QIO_CHANNEL(tioc);
566 }
567
568 /* nbd_reject_length: Handle any unexpected payload.
569 * @fatal requests that we quit talking to the client, even if we are able
570 * to successfully send an error to the guest.
571 * Return:
572 * -errno transmission error occurred or @fatal was requested, errp is set
573 * 0 error message successfully sent to client, errp is not set
574 */
575 static int nbd_reject_length(NBDClient *client, uint32_t length,
576 uint32_t option, bool fatal, Error **errp)
577 {
578 int ret;
579
580 assert(length);
581 if (nbd_drop(client->ioc, length, errp) < 0) {
582 return -EIO;
583 }
584 ret = nbd_negotiate_send_rep_err(client->ioc, NBD_REP_ERR_INVALID,
585 option, errp,
586 "option '%s' should have zero length",
587 nbd_opt_lookup(option));
588 if (fatal && !ret) {
589 error_setg(errp, "option '%s' should have zero length",
590 nbd_opt_lookup(option));
591 return -EINVAL;
592 }
593 return ret;
594 }
595
596 /* nbd_negotiate_options
597 * Process all NBD_OPT_* client option commands, during fixed newstyle
598 * negotiation.
599 * Return:
600 * -errno on error, errp is set
601 * 0 on successful negotiation, errp is not set
602 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
603 * errp is not set
604 */
605 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
606 Error **errp)
607 {
608 uint32_t flags;
609 bool fixedNewstyle = false;
610 bool no_zeroes = false;
611
612 /* Client sends:
613 [ 0 .. 3] client flags
614
615 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
616 [ 0 .. 7] NBD_OPTS_MAGIC
617 [ 8 .. 11] NBD option
618 [12 .. 15] Data length
619 ... Rest of request
620
621 [ 0 .. 7] NBD_OPTS_MAGIC
622 [ 8 .. 11] Second NBD option
623 [12 .. 15] Data length
624 ... Rest of request
625 */
626
627 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
628 error_prepend(errp, "read failed: ");
629 return -EIO;
630 }
631 be32_to_cpus(&flags);
632 trace_nbd_negotiate_options_flags(flags);
633 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
634 fixedNewstyle = true;
635 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
636 }
637 if (flags & NBD_FLAG_C_NO_ZEROES) {
638 no_zeroes = true;
639 flags &= ~NBD_FLAG_C_NO_ZEROES;
640 }
641 if (flags != 0) {
642 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
643 return -EINVAL;
644 }
645
646 while (1) {
647 int ret;
648 uint32_t option, length;
649 uint64_t magic;
650
651 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
652 error_prepend(errp, "read failed: ");
653 return -EINVAL;
654 }
655 magic = be64_to_cpu(magic);
656 trace_nbd_negotiate_options_check_magic(magic);
657 if (magic != NBD_OPTS_MAGIC) {
658 error_setg(errp, "Bad magic received");
659 return -EINVAL;
660 }
661
662 if (nbd_read(client->ioc, &option,
663 sizeof(option), errp) < 0) {
664 error_prepend(errp, "read failed: ");
665 return -EINVAL;
666 }
667 option = be32_to_cpu(option);
668
669 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
670 error_prepend(errp, "read failed: ");
671 return -EINVAL;
672 }
673 length = be32_to_cpu(length);
674
675 trace_nbd_negotiate_options_check_option(option,
676 nbd_opt_lookup(option));
677 if (client->tlscreds &&
678 client->ioc == (QIOChannel *)client->sioc) {
679 QIOChannel *tioc;
680 if (!fixedNewstyle) {
681 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
682 return -EINVAL;
683 }
684 switch (option) {
685 case NBD_OPT_STARTTLS:
686 if (length) {
687 /* Unconditionally drop the connection if the client
688 * can't start a TLS negotiation correctly */
689 return nbd_reject_length(client, length, option, true,
690 errp);
691 }
692 tioc = nbd_negotiate_handle_starttls(client, errp);
693 if (!tioc) {
694 return -EIO;
695 }
696 ret = 0;
697 object_unref(OBJECT(client->ioc));
698 client->ioc = QIO_CHANNEL(tioc);
699 break;
700
701 case NBD_OPT_EXPORT_NAME:
702 /* No way to return an error to client, so drop connection */
703 error_setg(errp, "Option 0x%x not permitted before TLS",
704 option);
705 return -EINVAL;
706
707 default:
708 if (nbd_drop(client->ioc, length, errp) < 0) {
709 return -EIO;
710 }
711 ret = nbd_negotiate_send_rep_err(client->ioc,
712 NBD_REP_ERR_TLS_REQD,
713 option, errp,
714 "Option 0x%" PRIx32
715 "not permitted before TLS",
716 option);
717 /* Let the client keep trying, unless they asked to
718 * quit. In this mode, we've already sent an error, so
719 * we can't ack the abort. */
720 if (option == NBD_OPT_ABORT) {
721 return 1;
722 }
723 break;
724 }
725 } else if (fixedNewstyle) {
726 switch (option) {
727 case NBD_OPT_LIST:
728 if (length) {
729 ret = nbd_reject_length(client, length, option, false,
730 errp);
731 } else {
732 ret = nbd_negotiate_handle_list(client, errp);
733 }
734 break;
735
736 case NBD_OPT_ABORT:
737 /* NBD spec says we must try to reply before
738 * disconnecting, but that we must also tolerate
739 * guests that don't wait for our reply. */
740 nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, option, NULL);
741 return 1;
742
743 case NBD_OPT_EXPORT_NAME:
744 return nbd_negotiate_handle_export_name(client, length,
745 myflags, no_zeroes,
746 errp);
747
748 case NBD_OPT_INFO:
749 case NBD_OPT_GO:
750 ret = nbd_negotiate_handle_info(client, length, option,
751 myflags, errp);
752 if (ret == 1) {
753 assert(option == NBD_OPT_GO);
754 return 0;
755 }
756 break;
757
758 case NBD_OPT_STARTTLS:
759 if (length) {
760 ret = nbd_reject_length(client, length, option, false,
761 errp);
762 } else if (client->tlscreds) {
763 ret = nbd_negotiate_send_rep_err(client->ioc,
764 NBD_REP_ERR_INVALID,
765 option, errp,
766 "TLS already enabled");
767 } else {
768 ret = nbd_negotiate_send_rep_err(client->ioc,
769 NBD_REP_ERR_POLICY,
770 option, errp,
771 "TLS not configured");
772 }
773 break;
774
775 case NBD_OPT_STRUCTURED_REPLY:
776 if (length) {
777 ret = nbd_reject_length(client, length, option, false,
778 errp);
779 } else if (client->structured_reply) {
780 ret = nbd_negotiate_send_rep_err(
781 client->ioc, NBD_REP_ERR_INVALID, option, errp,
782 "structured reply already negotiated");
783 } else {
784 ret = nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
785 option, errp);
786 client->structured_reply = true;
787 myflags |= NBD_FLAG_SEND_DF;
788 }
789 break;
790
791 default:
792 if (nbd_drop(client->ioc, length, errp) < 0) {
793 return -EIO;
794 }
795 ret = nbd_negotiate_send_rep_err(client->ioc,
796 NBD_REP_ERR_UNSUP,
797 option, errp,
798 "Unsupported option 0x%"
799 PRIx32 " (%s)", option,
800 nbd_opt_lookup(option));
801 break;
802 }
803 } else {
804 /*
805 * If broken new-style we should drop the connection
806 * for anything except NBD_OPT_EXPORT_NAME
807 */
808 switch (option) {
809 case NBD_OPT_EXPORT_NAME:
810 return nbd_negotiate_handle_export_name(client, length,
811 myflags, no_zeroes,
812 errp);
813
814 default:
815 error_setg(errp, "Unsupported option 0x%" PRIx32 " (%s)",
816 option, nbd_opt_lookup(option));
817 return -EINVAL;
818 }
819 }
820 if (ret < 0) {
821 return ret;
822 }
823 }
824 }
825
826 /* nbd_negotiate
827 * Return:
828 * -errno on error, errp is set
829 * 0 on successful negotiation, errp is not set
830 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
831 * errp is not set
832 */
833 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
834 {
835 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
836 int ret;
837 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
838 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
839 NBD_FLAG_SEND_WRITE_ZEROES);
840 bool oldStyle;
841
842 /* Old style negotiation header, no room for options
843 [ 0 .. 7] passwd ("NBDMAGIC")
844 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
845 [16 .. 23] size
846 [24 .. 27] export flags (zero-extended)
847 [28 .. 151] reserved (0)
848
849 New style negotiation header, client can send options
850 [ 0 .. 7] passwd ("NBDMAGIC")
851 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
852 [16 .. 17] server flags (0)
853 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
854 */
855
856 qio_channel_set_blocking(client->ioc, false, NULL);
857
858 trace_nbd_negotiate_begin();
859 memcpy(buf, "NBDMAGIC", 8);
860
861 oldStyle = client->exp != NULL && !client->tlscreds;
862 if (oldStyle) {
863 trace_nbd_negotiate_old_style(client->exp->size,
864 client->exp->nbdflags | myflags);
865 stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
866 stq_be_p(buf + 16, client->exp->size);
867 stl_be_p(buf + 24, client->exp->nbdflags | myflags);
868
869 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
870 error_prepend(errp, "write failed: ");
871 return -EINVAL;
872 }
873 } else {
874 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
875 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
876
877 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
878 error_prepend(errp, "write failed: ");
879 return -EINVAL;
880 }
881 ret = nbd_negotiate_options(client, myflags, errp);
882 if (ret != 0) {
883 if (ret < 0) {
884 error_prepend(errp, "option negotiation failed: ");
885 }
886 return ret;
887 }
888 }
889
890 trace_nbd_negotiate_success();
891
892 return 0;
893 }
894
895 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
896 Error **errp)
897 {
898 uint8_t buf[NBD_REQUEST_SIZE];
899 uint32_t magic;
900 int ret;
901
902 ret = nbd_read(ioc, buf, sizeof(buf), errp);
903 if (ret < 0) {
904 return ret;
905 }
906
907 /* Request
908 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
909 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
910 [ 6 .. 7] type (NBD_CMD_READ, ...)
911 [ 8 .. 15] handle
912 [16 .. 23] from
913 [24 .. 27] len
914 */
915
916 magic = ldl_be_p(buf);
917 request->flags = lduw_be_p(buf + 4);
918 request->type = lduw_be_p(buf + 6);
919 request->handle = ldq_be_p(buf + 8);
920 request->from = ldq_be_p(buf + 16);
921 request->len = ldl_be_p(buf + 24);
922
923 trace_nbd_receive_request(magic, request->flags, request->type,
924 request->from, request->len);
925
926 if (magic != NBD_REQUEST_MAGIC) {
927 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
928 return -EINVAL;
929 }
930 return 0;
931 }
932
933 #define MAX_NBD_REQUESTS 16
934
935 void nbd_client_get(NBDClient *client)
936 {
937 client->refcount++;
938 }
939
940 void nbd_client_put(NBDClient *client)
941 {
942 if (--client->refcount == 0) {
943 /* The last reference should be dropped by client->close,
944 * which is called by client_close.
945 */
946 assert(client->closing);
947
948 qio_channel_detach_aio_context(client->ioc);
949 object_unref(OBJECT(client->sioc));
950 object_unref(OBJECT(client->ioc));
951 if (client->tlscreds) {
952 object_unref(OBJECT(client->tlscreds));
953 }
954 g_free(client->tlsaclname);
955 if (client->exp) {
956 QTAILQ_REMOVE(&client->exp->clients, client, next);
957 nbd_export_put(client->exp);
958 }
959 g_free(client);
960 }
961 }
962
963 static void client_close(NBDClient *client, bool negotiated)
964 {
965 if (client->closing) {
966 return;
967 }
968
969 client->closing = true;
970
971 /* Force requests to finish. They will drop their own references,
972 * then we'll close the socket and free the NBDClient.
973 */
974 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
975 NULL);
976
977 /* Also tell the client, so that they release their reference. */
978 if (client->close_fn) {
979 client->close_fn(client, negotiated);
980 }
981 }
982
983 static NBDRequestData *nbd_request_get(NBDClient *client)
984 {
985 NBDRequestData *req;
986
987 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
988 client->nb_requests++;
989
990 req = g_new0(NBDRequestData, 1);
991 nbd_client_get(client);
992 req->client = client;
993 return req;
994 }
995
996 static void nbd_request_put(NBDRequestData *req)
997 {
998 NBDClient *client = req->client;
999
1000 if (req->data) {
1001 qemu_vfree(req->data);
1002 }
1003 g_free(req);
1004
1005 client->nb_requests--;
1006 nbd_client_receive_next_request(client);
1007
1008 nbd_client_put(client);
1009 }
1010
1011 static void blk_aio_attached(AioContext *ctx, void *opaque)
1012 {
1013 NBDExport *exp = opaque;
1014 NBDClient *client;
1015
1016 trace_nbd_blk_aio_attached(exp->name, ctx);
1017
1018 exp->ctx = ctx;
1019
1020 QTAILQ_FOREACH(client, &exp->clients, next) {
1021 qio_channel_attach_aio_context(client->ioc, ctx);
1022 if (client->recv_coroutine) {
1023 aio_co_schedule(ctx, client->recv_coroutine);
1024 }
1025 if (client->send_coroutine) {
1026 aio_co_schedule(ctx, client->send_coroutine);
1027 }
1028 }
1029 }
1030
1031 static void blk_aio_detach(void *opaque)
1032 {
1033 NBDExport *exp = opaque;
1034 NBDClient *client;
1035
1036 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1037
1038 QTAILQ_FOREACH(client, &exp->clients, next) {
1039 qio_channel_detach_aio_context(client->ioc);
1040 }
1041
1042 exp->ctx = NULL;
1043 }
1044
1045 static void nbd_eject_notifier(Notifier *n, void *data)
1046 {
1047 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1048 nbd_export_close(exp);
1049 }
1050
1051 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1052 uint16_t nbdflags, void (*close)(NBDExport *),
1053 bool writethrough, BlockBackend *on_eject_blk,
1054 Error **errp)
1055 {
1056 AioContext *ctx;
1057 BlockBackend *blk;
1058 NBDExport *exp = g_new0(NBDExport, 1);
1059 uint64_t perm;
1060 int ret;
1061
1062 /*
1063 * NBD exports are used for non-shared storage migration. Make sure
1064 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1065 * access since the export could be available before migration handover.
1066 */
1067 ctx = bdrv_get_aio_context(bs);
1068 aio_context_acquire(ctx);
1069 bdrv_invalidate_cache(bs, NULL);
1070 aio_context_release(ctx);
1071
1072 /* Don't allow resize while the NBD server is running, otherwise we don't
1073 * care what happens with the node. */
1074 perm = BLK_PERM_CONSISTENT_READ;
1075 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1076 perm |= BLK_PERM_WRITE;
1077 }
1078 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1079 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1080 ret = blk_insert_bs(blk, bs, errp);
1081 if (ret < 0) {
1082 goto fail;
1083 }
1084 blk_set_enable_write_cache(blk, !writethrough);
1085
1086 exp->refcount = 1;
1087 QTAILQ_INIT(&exp->clients);
1088 exp->blk = blk;
1089 exp->dev_offset = dev_offset;
1090 exp->nbdflags = nbdflags;
1091 exp->size = size < 0 ? blk_getlength(blk) : size;
1092 if (exp->size < 0) {
1093 error_setg_errno(errp, -exp->size,
1094 "Failed to determine the NBD export's length");
1095 goto fail;
1096 }
1097 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1098
1099 exp->close = close;
1100 exp->ctx = blk_get_aio_context(blk);
1101 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1102
1103 if (on_eject_blk) {
1104 blk_ref(on_eject_blk);
1105 exp->eject_notifier_blk = on_eject_blk;
1106 exp->eject_notifier.notify = nbd_eject_notifier;
1107 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1108 }
1109 return exp;
1110
1111 fail:
1112 blk_unref(blk);
1113 g_free(exp);
1114 return NULL;
1115 }
1116
1117 NBDExport *nbd_export_find(const char *name)
1118 {
1119 NBDExport *exp;
1120 QTAILQ_FOREACH(exp, &exports, next) {
1121 if (strcmp(name, exp->name) == 0) {
1122 return exp;
1123 }
1124 }
1125
1126 return NULL;
1127 }
1128
1129 void nbd_export_set_name(NBDExport *exp, const char *name)
1130 {
1131 if (exp->name == name) {
1132 return;
1133 }
1134
1135 nbd_export_get(exp);
1136 if (exp->name != NULL) {
1137 g_free(exp->name);
1138 exp->name = NULL;
1139 QTAILQ_REMOVE(&exports, exp, next);
1140 nbd_export_put(exp);
1141 }
1142 if (name != NULL) {
1143 nbd_export_get(exp);
1144 exp->name = g_strdup(name);
1145 QTAILQ_INSERT_TAIL(&exports, exp, next);
1146 }
1147 nbd_export_put(exp);
1148 }
1149
1150 void nbd_export_set_description(NBDExport *exp, const char *description)
1151 {
1152 g_free(exp->description);
1153 exp->description = g_strdup(description);
1154 }
1155
1156 void nbd_export_close(NBDExport *exp)
1157 {
1158 NBDClient *client, *next;
1159
1160 nbd_export_get(exp);
1161 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1162 client_close(client, true);
1163 }
1164 nbd_export_set_name(exp, NULL);
1165 nbd_export_set_description(exp, NULL);
1166 nbd_export_put(exp);
1167 }
1168
1169 void nbd_export_get(NBDExport *exp)
1170 {
1171 assert(exp->refcount > 0);
1172 exp->refcount++;
1173 }
1174
1175 void nbd_export_put(NBDExport *exp)
1176 {
1177 assert(exp->refcount > 0);
1178 if (exp->refcount == 1) {
1179 nbd_export_close(exp);
1180 }
1181
1182 if (--exp->refcount == 0) {
1183 assert(exp->name == NULL);
1184 assert(exp->description == NULL);
1185
1186 if (exp->close) {
1187 exp->close(exp);
1188 }
1189
1190 if (exp->blk) {
1191 if (exp->eject_notifier_blk) {
1192 notifier_remove(&exp->eject_notifier);
1193 blk_unref(exp->eject_notifier_blk);
1194 }
1195 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1196 blk_aio_detach, exp);
1197 blk_unref(exp->blk);
1198 exp->blk = NULL;
1199 }
1200
1201 g_free(exp);
1202 }
1203 }
1204
1205 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1206 {
1207 return exp->blk;
1208 }
1209
1210 void nbd_export_close_all(void)
1211 {
1212 NBDExport *exp, *next;
1213
1214 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1215 nbd_export_close(exp);
1216 }
1217 }
1218
1219 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1220 unsigned niov, Error **errp)
1221 {
1222 int ret;
1223
1224 g_assert(qemu_in_coroutine());
1225 qemu_co_mutex_lock(&client->send_lock);
1226 client->send_coroutine = qemu_coroutine_self();
1227
1228 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1229
1230 client->send_coroutine = NULL;
1231 qemu_co_mutex_unlock(&client->send_lock);
1232
1233 return ret;
1234 }
1235
1236 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1237 uint64_t handle)
1238 {
1239 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1240 stl_be_p(&reply->error, error);
1241 stq_be_p(&reply->handle, handle);
1242 }
1243
1244 static int nbd_co_send_simple_reply(NBDClient *client,
1245 uint64_t handle,
1246 uint32_t error,
1247 void *data,
1248 size_t len,
1249 Error **errp)
1250 {
1251 NBDSimpleReply reply;
1252 int nbd_err = system_errno_to_nbd_errno(error);
1253 struct iovec iov[] = {
1254 {.iov_base = &reply, .iov_len = sizeof(reply)},
1255 {.iov_base = data, .iov_len = len}
1256 };
1257
1258 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1259 len);
1260 set_be_simple_reply(&reply, nbd_err, handle);
1261
1262 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1263 }
1264
1265 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1266 uint16_t type, uint64_t handle, uint32_t length)
1267 {
1268 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1269 stw_be_p(&chunk->flags, flags);
1270 stw_be_p(&chunk->type, type);
1271 stq_be_p(&chunk->handle, handle);
1272 stl_be_p(&chunk->length, length);
1273 }
1274
1275 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1276 uint64_t handle,
1277 uint64_t offset,
1278 void *data,
1279 size_t size,
1280 Error **errp)
1281 {
1282 NBDStructuredRead chunk;
1283 struct iovec iov[] = {
1284 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1285 {.iov_base = data, .iov_len = size}
1286 };
1287
1288 trace_nbd_co_send_structured_read(handle, offset, data, size);
1289 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_OFFSET_DATA,
1290 handle, sizeof(chunk) - sizeof(chunk.h) + size);
1291 stq_be_p(&chunk.offset, offset);
1292
1293 return nbd_co_send_iov(client, iov, 2, errp);
1294 }
1295
1296 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1297 uint64_t handle,
1298 uint32_t error,
1299 const char *msg,
1300 Error **errp)
1301 {
1302 NBDStructuredError chunk;
1303 int nbd_err = system_errno_to_nbd_errno(error);
1304 struct iovec iov[] = {
1305 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1306 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1307 };
1308
1309 assert(nbd_err);
1310 trace_nbd_co_send_structured_error(handle, nbd_err,
1311 nbd_err_lookup(nbd_err), msg ? msg : "");
1312 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1313 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1314 stl_be_p(&chunk.error, nbd_err);
1315 stw_be_p(&chunk.message_length, iov[1].iov_len);
1316
1317 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1318 }
1319
1320 /* nbd_co_receive_request
1321 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1322 * connection right away, and any other negative value to report an error to
1323 * the client (although the caller may still need to disconnect after reporting
1324 * the error).
1325 */
1326 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
1327 Error **errp)
1328 {
1329 NBDClient *client = req->client;
1330 int valid_flags;
1331
1332 g_assert(qemu_in_coroutine());
1333 assert(client->recv_coroutine == qemu_coroutine_self());
1334 if (nbd_receive_request(client->ioc, request, errp) < 0) {
1335 return -EIO;
1336 }
1337
1338 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
1339 nbd_cmd_lookup(request->type));
1340
1341 if (request->type != NBD_CMD_WRITE) {
1342 /* No payload, we are ready to read the next request. */
1343 req->complete = true;
1344 }
1345
1346 if (request->type == NBD_CMD_DISC) {
1347 /* Special case: we're going to disconnect without a reply,
1348 * whether or not flags, from, or len are bogus */
1349 return -EIO;
1350 }
1351
1352 /* Check for sanity in the parameters, part 1. Defer as many
1353 * checks as possible until after reading any NBD_CMD_WRITE
1354 * payload, so we can try and keep the connection alive. */
1355 if ((request->from + request->len) < request->from) {
1356 error_setg(errp,
1357 "integer overflow detected, you're probably being attacked");
1358 return -EINVAL;
1359 }
1360
1361 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) {
1362 if (request->len > NBD_MAX_BUFFER_SIZE) {
1363 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1364 request->len, NBD_MAX_BUFFER_SIZE);
1365 return -EINVAL;
1366 }
1367
1368 req->data = blk_try_blockalign(client->exp->blk, request->len);
1369 if (req->data == NULL) {
1370 error_setg(errp, "No memory");
1371 return -ENOMEM;
1372 }
1373 }
1374 if (request->type == NBD_CMD_WRITE) {
1375 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
1376 error_prepend(errp, "reading from socket failed: ");
1377 return -EIO;
1378 }
1379 req->complete = true;
1380
1381 trace_nbd_co_receive_request_payload_received(request->handle,
1382 request->len);
1383 }
1384
1385 /* Sanity checks, part 2. */
1386 if (request->from + request->len > client->exp->size) {
1387 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
1388 ", Size: %" PRIu64, request->from, request->len,
1389 (uint64_t)client->exp->size);
1390 return request->type == NBD_CMD_WRITE ? -ENOSPC : -EINVAL;
1391 }
1392 valid_flags = NBD_CMD_FLAG_FUA;
1393 if (request->type == NBD_CMD_READ && client->structured_reply) {
1394 valid_flags |= NBD_CMD_FLAG_DF;
1395 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
1396 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
1397 }
1398 if (request->flags & ~valid_flags) {
1399 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
1400 nbd_cmd_lookup(request->type), request->flags);
1401 return -EINVAL;
1402 }
1403
1404 return 0;
1405 }
1406
1407 /* Owns a reference to the NBDClient passed as opaque. */
1408 static coroutine_fn void nbd_trip(void *opaque)
1409 {
1410 NBDClient *client = opaque;
1411 NBDExport *exp = client->exp;
1412 NBDRequestData *req;
1413 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
1414 int ret;
1415 int flags;
1416 int reply_data_len = 0;
1417 Error *local_err = NULL;
1418 char *msg = NULL;
1419
1420 trace_nbd_trip();
1421 if (client->closing) {
1422 nbd_client_put(client);
1423 return;
1424 }
1425
1426 req = nbd_request_get(client);
1427 ret = nbd_co_receive_request(req, &request, &local_err);
1428 client->recv_coroutine = NULL;
1429 nbd_client_receive_next_request(client);
1430 if (ret == -EIO) {
1431 goto disconnect;
1432 }
1433
1434 if (ret < 0) {
1435 goto reply;
1436 }
1437
1438 if (client->closing) {
1439 /*
1440 * The client may be closed when we are blocked in
1441 * nbd_co_receive_request()
1442 */
1443 goto done;
1444 }
1445
1446 switch (request.type) {
1447 case NBD_CMD_READ:
1448 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1449 if (request.flags & NBD_CMD_FLAG_FUA) {
1450 ret = blk_co_flush(exp->blk);
1451 if (ret < 0) {
1452 error_setg_errno(&local_err, -ret, "flush failed");
1453 break;
1454 }
1455 }
1456
1457 ret = blk_pread(exp->blk, request.from + exp->dev_offset,
1458 req->data, request.len);
1459 if (ret < 0) {
1460 error_setg_errno(&local_err, -ret, "reading from file failed");
1461 break;
1462 }
1463
1464 reply_data_len = request.len;
1465
1466 break;
1467 case NBD_CMD_WRITE:
1468 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1469 error_setg(&local_err, "Export is read-only");
1470 ret = -EROFS;
1471 break;
1472 }
1473
1474 flags = 0;
1475 if (request.flags & NBD_CMD_FLAG_FUA) {
1476 flags |= BDRV_REQ_FUA;
1477 }
1478 ret = blk_pwrite(exp->blk, request.from + exp->dev_offset,
1479 req->data, request.len, flags);
1480 if (ret < 0) {
1481 error_setg_errno(&local_err, -ret, "writing to file failed");
1482 }
1483
1484 break;
1485 case NBD_CMD_WRITE_ZEROES:
1486 if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1487 error_setg(&local_err, "Export is read-only");
1488 ret = -EROFS;
1489 break;
1490 }
1491
1492 flags = 0;
1493 if (request.flags & NBD_CMD_FLAG_FUA) {
1494 flags |= BDRV_REQ_FUA;
1495 }
1496 if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) {
1497 flags |= BDRV_REQ_MAY_UNMAP;
1498 }
1499 ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset,
1500 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_DISC:
1507 /* unreachable, thanks to special case in nbd_co_receive_request() */
1508 abort();
1509
1510 case NBD_CMD_FLUSH:
1511 ret = blk_co_flush(exp->blk);
1512 if (ret < 0) {
1513 error_setg_errno(&local_err, -ret, "flush failed");
1514 }
1515
1516 break;
1517 case NBD_CMD_TRIM:
1518 ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset,
1519 request.len);
1520 if (ret < 0) {
1521 error_setg_errno(&local_err, -ret, "discard failed");
1522 }
1523
1524 break;
1525 default:
1526 error_setg(&local_err, "invalid request type (%" PRIu32 ") received",
1527 request.type);
1528 ret = -EINVAL;
1529 }
1530
1531 reply:
1532 if (local_err) {
1533 /* If we get here, local_err was not a fatal error, and should be sent
1534 * to the client. */
1535 assert(ret < 0);
1536 msg = g_strdup(error_get_pretty(local_err));
1537 error_report_err(local_err);
1538 local_err = NULL;
1539 }
1540
1541 if (client->structured_reply &&
1542 (ret < 0 || request.type == NBD_CMD_READ)) {
1543 if (ret < 0) {
1544 ret = nbd_co_send_structured_error(req->client, request.handle,
1545 -ret, msg, &local_err);
1546 } else {
1547 ret = nbd_co_send_structured_read(req->client, request.handle,
1548 request.from, req->data,
1549 reply_data_len, &local_err);
1550 }
1551 } else {
1552 ret = nbd_co_send_simple_reply(req->client, request.handle,
1553 ret < 0 ? -ret : 0,
1554 req->data, reply_data_len, &local_err);
1555 }
1556 g_free(msg);
1557 if (ret < 0) {
1558 error_prepend(&local_err, "Failed to send reply: ");
1559 goto disconnect;
1560 }
1561
1562 /* We must disconnect after NBD_CMD_WRITE if we did not
1563 * read the payload.
1564 */
1565 if (!req->complete) {
1566 error_setg(&local_err, "Request handling failed in intermediate state");
1567 goto disconnect;
1568 }
1569
1570 done:
1571 nbd_request_put(req);
1572 nbd_client_put(client);
1573 return;
1574
1575 disconnect:
1576 if (local_err) {
1577 error_reportf_err(local_err, "Disconnect client, due to: ");
1578 }
1579 nbd_request_put(req);
1580 client_close(client, true);
1581 nbd_client_put(client);
1582 }
1583
1584 static void nbd_client_receive_next_request(NBDClient *client)
1585 {
1586 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
1587 nbd_client_get(client);
1588 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
1589 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
1590 }
1591 }
1592
1593 static coroutine_fn void nbd_co_client_start(void *opaque)
1594 {
1595 NBDClient *client = opaque;
1596 NBDExport *exp = client->exp;
1597 Error *local_err = NULL;
1598
1599 if (exp) {
1600 nbd_export_get(exp);
1601 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1602 }
1603 qemu_co_mutex_init(&client->send_lock);
1604
1605 if (nbd_negotiate(client, &local_err)) {
1606 if (local_err) {
1607 error_report_err(local_err);
1608 }
1609 client_close(client, false);
1610 return;
1611 }
1612
1613 nbd_client_receive_next_request(client);
1614 }
1615
1616 /*
1617 * Create a new client listener on the given export @exp, using the
1618 * given channel @sioc. Begin servicing it in a coroutine. When the
1619 * connection closes, call @close_fn with an indication of whether the
1620 * client completed negotiation.
1621 */
1622 void nbd_client_new(NBDExport *exp,
1623 QIOChannelSocket *sioc,
1624 QCryptoTLSCreds *tlscreds,
1625 const char *tlsaclname,
1626 void (*close_fn)(NBDClient *, bool))
1627 {
1628 NBDClient *client;
1629 Coroutine *co;
1630
1631 client = g_new0(NBDClient, 1);
1632 client->refcount = 1;
1633 client->exp = exp;
1634 client->tlscreds = tlscreds;
1635 if (tlscreds) {
1636 object_ref(OBJECT(client->tlscreds));
1637 }
1638 client->tlsaclname = g_strdup(tlsaclname);
1639 client->sioc = sioc;
1640 object_ref(OBJECT(client->sioc));
1641 client->ioc = QIO_CHANNEL(sioc);
1642 object_ref(OBJECT(client->ioc));
1643 client->close_fn = close_fn;
1644
1645 co = qemu_coroutine_create(nbd_co_client_start, client);
1646 qemu_coroutine_enter(co);
1647 }
AR7 emulation for QEMU