?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'remotes/ehabkost/tags/python-next-pull-request' into...
[qemu/ar7.git] / nbd / server.c
blob10faedcfc55de9eba89fa0c123fe689b6f600b98
1 /*
2 * Copyright (C) 2016-2018 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 #include "qemu/units.h"
25
26 #define NBD_META_ID_BASE_ALLOCATION 0
27 #define NBD_META_ID_DIRTY_BITMAP 1
28
29 /*
30 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
31 * constant. If an increase is needed, note that the NBD protocol
32 * recommends no larger than 32 mb, so that the client won't consider
33 * the reply as a denial of service attack.
34 */
35 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
36
37 static int system_errno_to_nbd_errno(int err)
38 {
39 switch (err) {
40 case 0:
41 return NBD_SUCCESS;
42 case EPERM:
43 case EROFS:
44 return NBD_EPERM;
45 case EIO:
46 return NBD_EIO;
47 case ENOMEM:
48 return NBD_ENOMEM;
49 #ifdef EDQUOT
50 case EDQUOT:
51 #endif
52 case EFBIG:
53 case ENOSPC:
54 return NBD_ENOSPC;
55 case EOVERFLOW:
56 return NBD_EOVERFLOW;
57 case ESHUTDOWN:
58 return NBD_ESHUTDOWN;
59 case EINVAL:
60 default:
61 return NBD_EINVAL;
62 }
63 }
64
65 /* Definitions for opaque data types */
66
67 typedef struct NBDRequestData NBDRequestData;
68
69 struct NBDRequestData {
70 QSIMPLEQ_ENTRY(NBDRequestData) entry;
71 NBDClient *client;
72 uint8_t *data;
73 bool complete;
74 };
75
76 struct NBDExport {
77 int refcount;
78 void (*close)(NBDExport *exp);
79
80 BlockBackend *blk;
81 char *name;
82 char *description;
83 uint64_t dev_offset;
84 uint64_t size;
85 uint16_t nbdflags;
86 QTAILQ_HEAD(, NBDClient) clients;
87 QTAILQ_ENTRY(NBDExport) next;
88
89 AioContext *ctx;
90
91 BlockBackend *eject_notifier_blk;
92 Notifier eject_notifier;
93
94 BdrvDirtyBitmap *export_bitmap;
95 char *export_bitmap_context;
96 };
97
98 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
99
100 /* NBDExportMetaContexts represents a list of contexts to be exported,
101 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
102 * NBD_OPT_LIST_META_CONTEXT. */
103 typedef struct NBDExportMetaContexts {
104 NBDExport *exp;
105 bool valid; /* means that negotiation of the option finished without
106 errors */
107 bool base_allocation; /* export base:allocation context (block status) */
108 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
109 } NBDExportMetaContexts;
110
111 struct NBDClient {
112 int refcount;
113 void (*close_fn)(NBDClient *client, bool negotiated);
114
115 NBDExport *exp;
116 QCryptoTLSCreds *tlscreds;
117 char *tlsauthz;
118 QIOChannelSocket *sioc; /* The underlying data channel */
119 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
120
121 Coroutine *recv_coroutine;
122
123 CoMutex send_lock;
124 Coroutine *send_coroutine;
125
126 QTAILQ_ENTRY(NBDClient) next;
127 int nb_requests;
128 bool closing;
129
130 uint32_t check_align; /* If non-zero, check for aligned client requests */
131
132 bool structured_reply;
133 NBDExportMetaContexts export_meta;
134
135 uint32_t opt; /* Current option being negotiated */
136 uint32_t optlen; /* remaining length of data in ioc for the option being
137 negotiated now */
138 };
139
140 static void nbd_client_receive_next_request(NBDClient *client);
141
142 /* Basic flow for negotiation
143
144 Server Client
145 Negotiate
146
147 or
148
149 Server Client
150 Negotiate #1
151 Option
152 Negotiate #2
153
154 ----
155
156 followed by
157
158 Server Client
159 Request
160 Response
161 Request
162 Response
163 ...
164 ...
165 Request (type == 2)
166
167 */
168
169 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
170 uint32_t type, uint32_t length)
171 {
172 stq_be_p(&rep->magic, NBD_REP_MAGIC);
173 stl_be_p(&rep->option, option);
174 stl_be_p(&rep->type, type);
175 stl_be_p(&rep->length, length);
176 }
177
178 /* Send a reply header, including length, but no payload.
179 * Return -errno on error, 0 on success. */
180 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
181 uint32_t len, Error **errp)
182 {
183 NBDOptionReply rep;
184
185 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
186 type, nbd_rep_lookup(type), len);
187
188 assert(len < NBD_MAX_BUFFER_SIZE);
189
190 set_be_option_rep(&rep, client->opt, type, len);
191 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
192 }
193
194 /* Send a reply header with default 0 length.
195 * Return -errno on error, 0 on success. */
196 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
197 Error **errp)
198 {
199 return nbd_negotiate_send_rep_len(client, type, 0, errp);
200 }
201
202 /* Send an error reply.
203 * Return -errno on error, 0 on success. */
204 static int GCC_FMT_ATTR(4, 0)
205 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
206 Error **errp, const char *fmt, va_list va)
207 {
208 char *msg;
209 int ret;
210 size_t len;
211
212 msg = g_strdup_vprintf(fmt, va);
213 len = strlen(msg);
214 assert(len < 4096);
215 trace_nbd_negotiate_send_rep_err(msg);
216 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
217 if (ret < 0) {
218 goto out;
219 }
220 if (nbd_write(client->ioc, msg, len, errp) < 0) {
221 error_prepend(errp, "write failed (error message): ");
222 ret = -EIO;
223 } else {
224 ret = 0;
225 }
226
227 out:
228 g_free(msg);
229 return ret;
230 }
231
232 /* Send an error reply.
233 * Return -errno on error, 0 on success. */
234 static int GCC_FMT_ATTR(4, 5)
235 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
236 Error **errp, const char *fmt, ...)
237 {
238 va_list va;
239 int ret;
240
241 va_start(va, fmt);
242 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
243 va_end(va);
244 return ret;
245 }
246
247 /* Drop remainder of the current option, and send a reply with the
248 * given error type and message. Return -errno on read or write
249 * failure; or 0 if connection is still live. */
250 static int GCC_FMT_ATTR(4, 0)
251 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
252 const char *fmt, va_list va)
253 {
254 int ret = nbd_drop(client->ioc, client->optlen, errp);
255
256 client->optlen = 0;
257 if (!ret) {
258 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
259 }
260 return ret;
261 }
262
263 static int GCC_FMT_ATTR(4, 5)
264 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
265 const char *fmt, ...)
266 {
267 int ret;
268 va_list va;
269
270 va_start(va, fmt);
271 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
272 va_end(va);
273
274 return ret;
275 }
276
277 static int GCC_FMT_ATTR(3, 4)
278 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
279 {
280 int ret;
281 va_list va;
282
283 va_start(va, fmt);
284 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
285 va_end(va);
286
287 return ret;
288 }
289
290 /* Read size bytes from the unparsed payload of the current option.
291 * Return -errno on I/O error, 0 if option was completely handled by
292 * sending a reply about inconsistent lengths, or 1 on success. */
293 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
294 Error **errp)
295 {
296 if (size > client->optlen) {
297 return nbd_opt_invalid(client, errp,
298 "Inconsistent lengths in option %s",
299 nbd_opt_lookup(client->opt));
300 }
301 client->optlen -= size;
302 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
303 }
304
305 /* Drop size bytes from the unparsed payload of the current option.
306 * Return -errno on I/O error, 0 if option was completely handled by
307 * sending a reply about inconsistent lengths, or 1 on success. */
308 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
309 {
310 if (size > client->optlen) {
311 return nbd_opt_invalid(client, errp,
312 "Inconsistent lengths in option %s",
313 nbd_opt_lookup(client->opt));
314 }
315 client->optlen -= size;
316 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
317 }
318
319 /* nbd_opt_read_name
320 *
321 * Read a string with the format:
322 * uint32_t len (<= NBD_MAX_NAME_SIZE)
323 * len bytes string (not 0-terminated)
324 *
325 * @name should be enough to store NBD_MAX_NAME_SIZE+1.
326 * If @length is non-null, it will be set to the actual string length.
327 *
328 * Return -errno on I/O error, 0 if option was completely handled by
329 * sending a reply about inconsistent lengths, or 1 on success.
330 */
331 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
332 Error **errp)
333 {
334 int ret;
335 uint32_t len;
336
337 ret = nbd_opt_read(client, &len, sizeof(len), errp);
338 if (ret <= 0) {
339 return ret;
340 }
341 len = cpu_to_be32(len);
342
343 if (len > NBD_MAX_NAME_SIZE) {
344 return nbd_opt_invalid(client, errp,
345 "Invalid name length: %" PRIu32, len);
346 }
347
348 ret = nbd_opt_read(client, name, len, errp);
349 if (ret <= 0) {
350 return ret;
351 }
352 name[len] = '\0';
353
354 if (length) {
355 *length = len;
356 }
357
358 return 1;
359 }
360
361 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
362 * Return -errno on error, 0 on success. */
363 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
364 Error **errp)
365 {
366 size_t name_len, desc_len;
367 uint32_t len;
368 const char *name = exp->name ? exp->name : "";
369 const char *desc = exp->description ? exp->description : "";
370 QIOChannel *ioc = client->ioc;
371 int ret;
372
373 trace_nbd_negotiate_send_rep_list(name, desc);
374 name_len = strlen(name);
375 desc_len = strlen(desc);
376 len = name_len + desc_len + sizeof(len);
377 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
378 if (ret < 0) {
379 return ret;
380 }
381
382 len = cpu_to_be32(name_len);
383 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
384 error_prepend(errp, "write failed (name length): ");
385 return -EINVAL;
386 }
387
388 if (nbd_write(ioc, name, name_len, errp) < 0) {
389 error_prepend(errp, "write failed (name buffer): ");
390 return -EINVAL;
391 }
392
393 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
394 error_prepend(errp, "write failed (description buffer): ");
395 return -EINVAL;
396 }
397
398 return 0;
399 }
400
401 /* Process the NBD_OPT_LIST command, with a potential series of replies.
402 * Return -errno on error, 0 on success. */
403 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
404 {
405 NBDExport *exp;
406 assert(client->opt == NBD_OPT_LIST);
407
408 /* For each export, send a NBD_REP_SERVER reply. */
409 QTAILQ_FOREACH(exp, &exports, next) {
410 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
411 return -EINVAL;
412 }
413 }
414 /* Finish with a NBD_REP_ACK. */
415 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
416 }
417
418 static void nbd_check_meta_export(NBDClient *client)
419 {
420 client->export_meta.valid &= client->exp == client->export_meta.exp;
421 }
422
423 /* Send a reply to NBD_OPT_EXPORT_NAME.
424 * Return -errno on error, 0 on success. */
425 static int nbd_negotiate_handle_export_name(NBDClient *client,
426 uint16_t myflags, bool no_zeroes,
427 Error **errp)
428 {
429 char name[NBD_MAX_NAME_SIZE + 1];
430 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
431 size_t len;
432 int ret;
433
434 /* Client sends:
435 [20 .. xx] export name (length bytes)
436 Server replies:
437 [ 0 .. 7] size
438 [ 8 .. 9] export flags
439 [10 .. 133] reserved (0) [unless no_zeroes]
440 */
441 trace_nbd_negotiate_handle_export_name();
442 if (client->optlen >= sizeof(name)) {
443 error_setg(errp, "Bad length received");
444 return -EINVAL;
445 }
446 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
447 return -EIO;
448 }
449 name[client->optlen] = '\0';
450 client->optlen = 0;
451
452 trace_nbd_negotiate_handle_export_name_request(name);
453
454 client->exp = nbd_export_find(name);
455 if (!client->exp) {
456 error_setg(errp, "export not found");
457 return -EINVAL;
458 }
459
460 trace_nbd_negotiate_new_style_size_flags(client->exp->size,
461 client->exp->nbdflags | myflags);
462 stq_be_p(buf, client->exp->size);
463 stw_be_p(buf + 8, client->exp->nbdflags | myflags);
464 len = no_zeroes ? 10 : sizeof(buf);
465 ret = nbd_write(client->ioc, buf, len, errp);
466 if (ret < 0) {
467 error_prepend(errp, "write failed: ");
468 return ret;
469 }
470
471 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
472 nbd_export_get(client->exp);
473 nbd_check_meta_export(client);
474
475 return 0;
476 }
477
478 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
479 * The buffer does NOT include the info type prefix.
480 * Return -errno on error, 0 if ready to send more. */
481 static int nbd_negotiate_send_info(NBDClient *client,
482 uint16_t info, uint32_t length, void *buf,
483 Error **errp)
484 {
485 int rc;
486
487 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
488 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
489 sizeof(info) + length, errp);
490 if (rc < 0) {
491 return rc;
492 }
493 info = cpu_to_be16(info);
494 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
495 return -EIO;
496 }
497 if (nbd_write(client->ioc, buf, length, errp) < 0) {
498 return -EIO;
499 }
500 return 0;
501 }
502
503 /* nbd_reject_length: Handle any unexpected payload.
504 * @fatal requests that we quit talking to the client, even if we are able
505 * to successfully send an error reply.
506 * Return:
507 * -errno transmission error occurred or @fatal was requested, errp is set
508 * 0 error message successfully sent to client, errp is not set
509 */
510 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
511 {
512 int ret;
513
514 assert(client->optlen);
515 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
516 nbd_opt_lookup(client->opt));
517 if (fatal && !ret) {
518 error_setg(errp, "option '%s' has unexpected length",
519 nbd_opt_lookup(client->opt));
520 return -EINVAL;
521 }
522 return ret;
523 }
524
525 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
526 * Return -errno on error, 0 if ready for next option, and 1 to move
527 * into transmission phase. */
528 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
529 Error **errp)
530 {
531 int rc;
532 char name[NBD_MAX_NAME_SIZE + 1];
533 NBDExport *exp;
534 uint16_t requests;
535 uint16_t request;
536 uint32_t namelen;
537 bool sendname = false;
538 bool blocksize = false;
539 uint32_t sizes[3];
540 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
541 uint32_t check_align = 0;
542
543 /* Client sends:
544 4 bytes: L, name length (can be 0)
545 L bytes: export name
546 2 bytes: N, number of requests (can be 0)
547 N * 2 bytes: N requests
548 */
549 rc = nbd_opt_read_name(client, name, &namelen, errp);
550 if (rc <= 0) {
551 return rc;
552 }
553 trace_nbd_negotiate_handle_export_name_request(name);
554
555 rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
556 if (rc <= 0) {
557 return rc;
558 }
559 requests = be16_to_cpu(requests);
560 trace_nbd_negotiate_handle_info_requests(requests);
561 while (requests--) {
562 rc = nbd_opt_read(client, &request, sizeof(request), errp);
563 if (rc <= 0) {
564 return rc;
565 }
566 request = be16_to_cpu(request);
567 trace_nbd_negotiate_handle_info_request(request,
568 nbd_info_lookup(request));
569 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
570 * everything else is either a request we don't know or
571 * something we send regardless of request */
572 switch (request) {
573 case NBD_INFO_NAME:
574 sendname = true;
575 break;
576 case NBD_INFO_BLOCK_SIZE:
577 blocksize = true;
578 break;
579 }
580 }
581 if (client->optlen) {
582 return nbd_reject_length(client, false, errp);
583 }
584
585 exp = nbd_export_find(name);
586 if (!exp) {
587 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
588 errp, "export '%s' not present",
589 name);
590 }
591
592 /* Don't bother sending NBD_INFO_NAME unless client requested it */
593 if (sendname) {
594 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
595 errp);
596 if (rc < 0) {
597 return rc;
598 }
599 }
600
601 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
602 * client request */
603 if (exp->description) {
604 size_t len = strlen(exp->description);
605
606 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
607 len, exp->description, errp);
608 if (rc < 0) {
609 return rc;
610 }
611 }
612
613 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
614 * according to whether the client requested it, and according to
615 * whether this is OPT_INFO or OPT_GO. */
616 /* minimum - 1 for back-compat, or actual if client will obey it. */
617 if (client->opt == NBD_OPT_INFO || blocksize) {
618 check_align = sizes[0] = blk_get_request_alignment(exp->blk);
619 } else {
620 sizes[0] = 1;
621 }
622 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
623 /* preferred - Hard-code to 4096 for now.
624 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
625 sizes[1] = MAX(4096, sizes[0]);
626 /* maximum - At most 32M, but smaller as appropriate. */
627 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
628 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
629 sizes[0] = cpu_to_be32(sizes[0]);
630 sizes[1] = cpu_to_be32(sizes[1]);
631 sizes[2] = cpu_to_be32(sizes[2]);
632 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
633 sizeof(sizes), sizes, errp);
634 if (rc < 0) {
635 return rc;
636 }
637
638 /* Send NBD_INFO_EXPORT always */
639 trace_nbd_negotiate_new_style_size_flags(exp->size,
640 exp->nbdflags | myflags);
641 stq_be_p(buf, exp->size);
642 stw_be_p(buf + 8, exp->nbdflags | myflags);
643 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
644 sizeof(buf), buf, errp);
645 if (rc < 0) {
646 return rc;
647 }
648
649 /*
650 * If the client is just asking for NBD_OPT_INFO, but forgot to
651 * request block sizes in a situation that would impact
652 * performance, then return an error. But for NBD_OPT_GO, we
653 * tolerate all clients, regardless of alignments.
654 */
655 if (client->opt == NBD_OPT_INFO && !blocksize &&
656 blk_get_request_alignment(exp->blk) > 1) {
657 return nbd_negotiate_send_rep_err(client,
658 NBD_REP_ERR_BLOCK_SIZE_REQD,
659 errp,
660 "request NBD_INFO_BLOCK_SIZE to "
661 "use this export");
662 }
663
664 /* Final reply */
665 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
666 if (rc < 0) {
667 return rc;
668 }
669
670 if (client->opt == NBD_OPT_GO) {
671 client->exp = exp;
672 client->check_align = check_align;
673 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
674 nbd_export_get(client->exp);
675 nbd_check_meta_export(client);
676 rc = 1;
677 }
678 return rc;
679 }
680
681
682 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
683 * new channel for all further (now-encrypted) communication. */
684 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
685 Error **errp)
686 {
687 QIOChannel *ioc;
688 QIOChannelTLS *tioc;
689 struct NBDTLSHandshakeData data = { 0 };
690
691 assert(client->opt == NBD_OPT_STARTTLS);
692
693 trace_nbd_negotiate_handle_starttls();
694 ioc = client->ioc;
695
696 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
697 return NULL;
698 }
699
700 tioc = qio_channel_tls_new_server(ioc,
701 client->tlscreds,
702 client->tlsauthz,
703 errp);
704 if (!tioc) {
705 return NULL;
706 }
707
708 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
709 trace_nbd_negotiate_handle_starttls_handshake();
710 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
711 qio_channel_tls_handshake(tioc,
712 nbd_tls_handshake,
713 &data,
714 NULL,
715 NULL);
716
717 if (!data.complete) {
718 g_main_loop_run(data.loop);
719 }
720 g_main_loop_unref(data.loop);
721 if (data.error) {
722 object_unref(OBJECT(tioc));
723 error_propagate(errp, data.error);
724 return NULL;
725 }
726
727 return QIO_CHANNEL(tioc);
728 }
729
730 /* nbd_negotiate_send_meta_context
731 *
732 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
733 *
734 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
735 */
736 static int nbd_negotiate_send_meta_context(NBDClient *client,
737 const char *context,
738 uint32_t context_id,
739 Error **errp)
740 {
741 NBDOptionReplyMetaContext opt;
742 struct iovec iov[] = {
743 {.iov_base = &opt, .iov_len = sizeof(opt)},
744 {.iov_base = (void *)context, .iov_len = strlen(context)}
745 };
746
747 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
748 context_id = 0;
749 }
750
751 trace_nbd_negotiate_meta_query_reply(context, context_id);
752 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
753 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
754 stl_be_p(&opt.context_id, context_id);
755
756 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
757 }
758
759 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
760 * @match is never set to false.
761 *
762 * Return -errno on I/O error, 0 if option was completely handled by
763 * sending a reply about inconsistent lengths, or 1 on success.
764 *
765 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
766 * It only means that there are no errors.
767 */
768 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
769 Error **errp)
770 {
771 int ret;
772 char *query;
773 size_t len = strlen(pattern);
774
775 assert(len);
776
777 query = g_malloc(len);
778 ret = nbd_opt_read(client, query, len, errp);
779 if (ret <= 0) {
780 g_free(query);
781 return ret;
782 }
783
784 if (strncmp(query, pattern, len) == 0) {
785 trace_nbd_negotiate_meta_query_parse(pattern);
786 *match = true;
787 } else {
788 trace_nbd_negotiate_meta_query_skip("pattern not matched");
789 }
790 g_free(query);
791
792 return 1;
793 }
794
795 /*
796 * Read @len bytes, and set @match to true if they match @pattern, or if @len
797 * is 0 and the client is performing _LIST_. @match is never set to false.
798 *
799 * Return -errno on I/O error, 0 if option was completely handled by
800 * sending a reply about inconsistent lengths, or 1 on success.
801 *
802 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
803 * It only means that there are no errors.
804 */
805 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
806 uint32_t len, bool *match, Error **errp)
807 {
808 if (len == 0) {
809 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
810 *match = true;
811 }
812 trace_nbd_negotiate_meta_query_parse("empty");
813 return 1;
814 }
815
816 if (len != strlen(pattern)) {
817 trace_nbd_negotiate_meta_query_skip("different lengths");
818 return nbd_opt_skip(client, len, errp);
819 }
820
821 return nbd_meta_pattern(client, pattern, match, errp);
822 }
823
824 /* nbd_meta_base_query
825 *
826 * Handle queries to 'base' namespace. For now, only the base:allocation
827 * context is available. 'len' is the amount of text remaining to be read from
828 * the current name, after the 'base:' portion has been stripped.
829 *
830 * Return -errno on I/O error, 0 if option was completely handled by
831 * sending a reply about inconsistent lengths, or 1 on success.
832 */
833 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
834 uint32_t len, Error **errp)
835 {
836 return nbd_meta_empty_or_pattern(client, "allocation", len,
837 &meta->base_allocation, errp);
838 }
839
840 /* nbd_meta_bitmap_query
841 *
842 * Handle query to 'qemu:' namespace.
843 * @len is the amount of text remaining to be read from the current name, after
844 * the 'qemu:' portion has been stripped.
845 *
846 * Return -errno on I/O error, 0 if option was completely handled by
847 * sending a reply about inconsistent lengths, or 1 on success. */
848 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
849 uint32_t len, Error **errp)
850 {
851 bool dirty_bitmap = false;
852 size_t dirty_bitmap_len = strlen("dirty-bitmap:");
853 int ret;
854
855 if (!meta->exp->export_bitmap) {
856 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
857 return nbd_opt_skip(client, len, errp);
858 }
859
860 if (len == 0) {
861 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
862 meta->bitmap = true;
863 }
864 trace_nbd_negotiate_meta_query_parse("empty");
865 return 1;
866 }
867
868 if (len < dirty_bitmap_len) {
869 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
870 return nbd_opt_skip(client, len, errp);
871 }
872
873 len -= dirty_bitmap_len;
874 ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
875 if (ret <= 0) {
876 return ret;
877 }
878 if (!dirty_bitmap) {
879 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
880 return nbd_opt_skip(client, len, errp);
881 }
882
883 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
884
885 return nbd_meta_empty_or_pattern(
886 client, meta->exp->export_bitmap_context +
887 strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
888 }
889
890 /* nbd_negotiate_meta_query
891 *
892 * Parse namespace name and call corresponding function to parse body of the
893 * query.
894 *
895 * The only supported namespace now is 'base'.
896 *
897 * The function aims not wasting time and memory to read long unknown namespace
898 * names.
899 *
900 * Return -errno on I/O error, 0 if option was completely handled by
901 * sending a reply about inconsistent lengths, or 1 on success. */
902 static int nbd_negotiate_meta_query(NBDClient *client,
903 NBDExportMetaContexts *meta, Error **errp)
904 {
905 /*
906 * Both 'qemu' and 'base' namespaces have length = 5 including a
907 * colon. If another length namespace is later introduced, this
908 * should certainly be refactored.
909 */
910 int ret;
911 size_t ns_len = 5;
912 char ns[5];
913 uint32_t len;
914
915 ret = nbd_opt_read(client, &len, sizeof(len), errp);
916 if (ret <= 0) {
917 return ret;
918 }
919 len = cpu_to_be32(len);
920
921 if (len < ns_len) {
922 trace_nbd_negotiate_meta_query_skip("length too short");
923 return nbd_opt_skip(client, len, errp);
924 }
925
926 len -= ns_len;
927 ret = nbd_opt_read(client, ns, ns_len, errp);
928 if (ret <= 0) {
929 return ret;
930 }
931
932 if (!strncmp(ns, "base:", ns_len)) {
933 trace_nbd_negotiate_meta_query_parse("base:");
934 return nbd_meta_base_query(client, meta, len, errp);
935 } else if (!strncmp(ns, "qemu:", ns_len)) {
936 trace_nbd_negotiate_meta_query_parse("qemu:");
937 return nbd_meta_qemu_query(client, meta, len, errp);
938 }
939
940 trace_nbd_negotiate_meta_query_skip("unknown namespace");
941 return nbd_opt_skip(client, len, errp);
942 }
943
944 /* nbd_negotiate_meta_queries
945 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
946 *
947 * Return -errno on I/O error, or 0 if option was completely handled. */
948 static int nbd_negotiate_meta_queries(NBDClient *client,
949 NBDExportMetaContexts *meta, Error **errp)
950 {
951 int ret;
952 char export_name[NBD_MAX_NAME_SIZE + 1];
953 NBDExportMetaContexts local_meta;
954 uint32_t nb_queries;
955 int i;
956
957 if (!client->structured_reply) {
958 return nbd_opt_invalid(client, errp,
959 "request option '%s' when structured reply "
960 "is not negotiated",
961 nbd_opt_lookup(client->opt));
962 }
963
964 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
965 /* Only change the caller's meta on SET. */
966 meta = &local_meta;
967 }
968
969 memset(meta, 0, sizeof(*meta));
970
971 ret = nbd_opt_read_name(client, export_name, NULL, errp);
972 if (ret <= 0) {
973 return ret;
974 }
975
976 meta->exp = nbd_export_find(export_name);
977 if (meta->exp == NULL) {
978 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
979 "export '%s' not present", export_name);
980 }
981
982 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
983 if (ret <= 0) {
984 return ret;
985 }
986 nb_queries = cpu_to_be32(nb_queries);
987 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
988 export_name, nb_queries);
989
990 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
991 /* enable all known contexts */
992 meta->base_allocation = true;
993 meta->bitmap = !!meta->exp->export_bitmap;
994 } else {
995 for (i = 0; i < nb_queries; ++i) {
996 ret = nbd_negotiate_meta_query(client, meta, errp);
997 if (ret <= 0) {
998 return ret;
999 }
1000 }
1001 }
1002
1003 if (meta->base_allocation) {
1004 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1005 NBD_META_ID_BASE_ALLOCATION,
1006 errp);
1007 if (ret < 0) {
1008 return ret;
1009 }
1010 }
1011
1012 if (meta->bitmap) {
1013 ret = nbd_negotiate_send_meta_context(client,
1014 meta->exp->export_bitmap_context,
1015 NBD_META_ID_DIRTY_BITMAP,
1016 errp);
1017 if (ret < 0) {
1018 return ret;
1019 }
1020 }
1021
1022 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1023 if (ret == 0) {
1024 meta->valid = true;
1025 }
1026
1027 return ret;
1028 }
1029
1030 /* nbd_negotiate_options
1031 * Process all NBD_OPT_* client option commands, during fixed newstyle
1032 * negotiation.
1033 * Return:
1034 * -errno on error, errp is set
1035 * 0 on successful negotiation, errp is not set
1036 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1037 * errp is not set
1038 */
1039 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1040 Error **errp)
1041 {
1042 uint32_t flags;
1043 bool fixedNewstyle = false;
1044 bool no_zeroes = false;
1045
1046 /* Client sends:
1047 [ 0 .. 3] client flags
1048
1049 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1050 [ 0 .. 7] NBD_OPTS_MAGIC
1051 [ 8 .. 11] NBD option
1052 [12 .. 15] Data length
1053 ... Rest of request
1054
1055 [ 0 .. 7] NBD_OPTS_MAGIC
1056 [ 8 .. 11] Second NBD option
1057 [12 .. 15] Data length
1058 ... Rest of request
1059 */
1060
1061 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1062 return -EIO;
1063 }
1064 trace_nbd_negotiate_options_flags(flags);
1065 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1066 fixedNewstyle = true;
1067 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1068 }
1069 if (flags & NBD_FLAG_C_NO_ZEROES) {
1070 no_zeroes = true;
1071 flags &= ~NBD_FLAG_C_NO_ZEROES;
1072 }
1073 if (flags != 0) {
1074 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1075 return -EINVAL;
1076 }
1077
1078 while (1) {
1079 int ret;
1080 uint32_t option, length;
1081 uint64_t magic;
1082
1083 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1084 return -EINVAL;
1085 }
1086 trace_nbd_negotiate_options_check_magic(magic);
1087 if (magic != NBD_OPTS_MAGIC) {
1088 error_setg(errp, "Bad magic received");
1089 return -EINVAL;
1090 }
1091
1092 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1093 return -EINVAL;
1094 }
1095 client->opt = option;
1096
1097 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1098 return -EINVAL;
1099 }
1100 assert(!client->optlen);
1101 client->optlen = length;
1102
1103 if (length > NBD_MAX_BUFFER_SIZE) {
1104 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1105 length, NBD_MAX_BUFFER_SIZE);
1106 return -EINVAL;
1107 }
1108
1109 trace_nbd_negotiate_options_check_option(option,
1110 nbd_opt_lookup(option));
1111 if (client->tlscreds &&
1112 client->ioc == (QIOChannel *)client->sioc) {
1113 QIOChannel *tioc;
1114 if (!fixedNewstyle) {
1115 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1116 return -EINVAL;
1117 }
1118 switch (option) {
1119 case NBD_OPT_STARTTLS:
1120 if (length) {
1121 /* Unconditionally drop the connection if the client
1122 * can't start a TLS negotiation correctly */
1123 return nbd_reject_length(client, true, errp);
1124 }
1125 tioc = nbd_negotiate_handle_starttls(client, errp);
1126 if (!tioc) {
1127 return -EIO;
1128 }
1129 ret = 0;
1130 object_unref(OBJECT(client->ioc));
1131 client->ioc = QIO_CHANNEL(tioc);
1132 break;
1133
1134 case NBD_OPT_EXPORT_NAME:
1135 /* No way to return an error to client, so drop connection */
1136 error_setg(errp, "Option 0x%x not permitted before TLS",
1137 option);
1138 return -EINVAL;
1139
1140 default:
1141 /* Let the client keep trying, unless they asked to
1142 * quit. Always try to give an error back to the
1143 * client; but when replying to OPT_ABORT, be aware
1144 * that the client may hang up before receiving the
1145 * error, in which case we are fine ignoring the
1146 * resulting EPIPE. */
1147 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1148 option == NBD_OPT_ABORT ? NULL : errp,
1149 "Option 0x%" PRIx32
1150 " not permitted before TLS", option);
1151 if (option == NBD_OPT_ABORT) {
1152 return 1;
1153 }
1154 break;
1155 }
1156 } else if (fixedNewstyle) {
1157 switch (option) {
1158 case NBD_OPT_LIST:
1159 if (length) {
1160 ret = nbd_reject_length(client, false, errp);
1161 } else {
1162 ret = nbd_negotiate_handle_list(client, errp);
1163 }
1164 break;
1165
1166 case NBD_OPT_ABORT:
1167 /* NBD spec says we must try to reply before
1168 * disconnecting, but that we must also tolerate
1169 * guests that don't wait for our reply. */
1170 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1171 return 1;
1172
1173 case NBD_OPT_EXPORT_NAME:
1174 return nbd_negotiate_handle_export_name(client,
1175 myflags, no_zeroes,
1176 errp);
1177
1178 case NBD_OPT_INFO:
1179 case NBD_OPT_GO:
1180 ret = nbd_negotiate_handle_info(client, myflags, errp);
1181 if (ret == 1) {
1182 assert(option == NBD_OPT_GO);
1183 return 0;
1184 }
1185 break;
1186
1187 case NBD_OPT_STARTTLS:
1188 if (length) {
1189 ret = nbd_reject_length(client, false, errp);
1190 } else if (client->tlscreds) {
1191 ret = nbd_negotiate_send_rep_err(client,
1192 NBD_REP_ERR_INVALID, errp,
1193 "TLS already enabled");
1194 } else {
1195 ret = nbd_negotiate_send_rep_err(client,
1196 NBD_REP_ERR_POLICY, errp,
1197 "TLS not configured");
1198 }
1199 break;
1200
1201 case NBD_OPT_STRUCTURED_REPLY:
1202 if (length) {
1203 ret = nbd_reject_length(client, false, errp);
1204 } else if (client->structured_reply) {
1205 ret = nbd_negotiate_send_rep_err(
1206 client, NBD_REP_ERR_INVALID, errp,
1207 "structured reply already negotiated");
1208 } else {
1209 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1210 client->structured_reply = true;
1211 myflags |= NBD_FLAG_SEND_DF;
1212 }
1213 break;
1214
1215 case NBD_OPT_LIST_META_CONTEXT:
1216 case NBD_OPT_SET_META_CONTEXT:
1217 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1218 errp);
1219 break;
1220
1221 default:
1222 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1223 "Unsupported option %" PRIu32 " (%s)",
1224 option, nbd_opt_lookup(option));
1225 break;
1226 }
1227 } else {
1228 /*
1229 * If broken new-style we should drop the connection
1230 * for anything except NBD_OPT_EXPORT_NAME
1231 */
1232 switch (option) {
1233 case NBD_OPT_EXPORT_NAME:
1234 return nbd_negotiate_handle_export_name(client,
1235 myflags, no_zeroes,
1236 errp);
1237
1238 default:
1239 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1240 option, nbd_opt_lookup(option));
1241 return -EINVAL;
1242 }
1243 }
1244 if (ret < 0) {
1245 return ret;
1246 }
1247 }
1248 }
1249
1250 /* nbd_negotiate
1251 * Return:
1252 * -errno on error, errp is set
1253 * 0 on successful negotiation, errp is not set
1254 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1255 * errp is not set
1256 */
1257 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1258 {
1259 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1260 int ret;
1261 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1262 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1263 NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1264
1265 /* Old style negotiation header, no room for options
1266 [ 0 .. 7] passwd ("NBDMAGIC")
1267 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1268 [16 .. 23] size
1269 [24 .. 27] export flags (zero-extended)
1270 [28 .. 151] reserved (0)
1271
1272 New style negotiation header, client can send options
1273 [ 0 .. 7] passwd ("NBDMAGIC")
1274 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1275 [16 .. 17] server flags (0)
1276 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1277 */
1278
1279 qio_channel_set_blocking(client->ioc, false, NULL);
1280
1281 trace_nbd_negotiate_begin();
1282 memcpy(buf, "NBDMAGIC", 8);
1283
1284 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1285 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1286
1287 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1288 error_prepend(errp, "write failed: ");
1289 return -EINVAL;
1290 }
1291 ret = nbd_negotiate_options(client, myflags, errp);
1292 if (ret != 0) {
1293 if (ret < 0) {
1294 error_prepend(errp, "option negotiation failed: ");
1295 }
1296 return ret;
1297 }
1298
1299 assert(!client->optlen);
1300 trace_nbd_negotiate_success();
1301
1302 return 0;
1303 }
1304
1305 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1306 Error **errp)
1307 {
1308 uint8_t buf[NBD_REQUEST_SIZE];
1309 uint32_t magic;
1310 int ret;
1311
1312 ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1313 if (ret < 0) {
1314 return ret;
1315 }
1316
1317 /* Request
1318 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1319 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1320 [ 6 .. 7] type (NBD_CMD_READ, ...)
1321 [ 8 .. 15] handle
1322 [16 .. 23] from
1323 [24 .. 27] len
1324 */
1325
1326 magic = ldl_be_p(buf);
1327 request->flags = lduw_be_p(buf + 4);
1328 request->type = lduw_be_p(buf + 6);
1329 request->handle = ldq_be_p(buf + 8);
1330 request->from = ldq_be_p(buf + 16);
1331 request->len = ldl_be_p(buf + 24);
1332
1333 trace_nbd_receive_request(magic, request->flags, request->type,
1334 request->from, request->len);
1335
1336 if (magic != NBD_REQUEST_MAGIC) {
1337 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1338 return -EINVAL;
1339 }
1340 return 0;
1341 }
1342
1343 #define MAX_NBD_REQUESTS 16
1344
1345 void nbd_client_get(NBDClient *client)
1346 {
1347 client->refcount++;
1348 }
1349
1350 void nbd_client_put(NBDClient *client)
1351 {
1352 if (--client->refcount == 0) {
1353 /* The last reference should be dropped by client->close,
1354 * which is called by client_close.
1355 */
1356 assert(client->closing);
1357
1358 qio_channel_detach_aio_context(client->ioc);
1359 object_unref(OBJECT(client->sioc));
1360 object_unref(OBJECT(client->ioc));
1361 if (client->tlscreds) {
1362 object_unref(OBJECT(client->tlscreds));
1363 }
1364 g_free(client->tlsauthz);
1365 if (client->exp) {
1366 QTAILQ_REMOVE(&client->exp->clients, client, next);
1367 nbd_export_put(client->exp);
1368 }
1369 g_free(client);
1370 }
1371 }
1372
1373 static void client_close(NBDClient *client, bool negotiated)
1374 {
1375 if (client->closing) {
1376 return;
1377 }
1378
1379 client->closing = true;
1380
1381 /* Force requests to finish. They will drop their own references,
1382 * then we'll close the socket and free the NBDClient.
1383 */
1384 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1385 NULL);
1386
1387 /* Also tell the client, so that they release their reference. */
1388 if (client->close_fn) {
1389 client->close_fn(client, negotiated);
1390 }
1391 }
1392
1393 static NBDRequestData *nbd_request_get(NBDClient *client)
1394 {
1395 NBDRequestData *req;
1396
1397 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1398 client->nb_requests++;
1399
1400 req = g_new0(NBDRequestData, 1);
1401 nbd_client_get(client);
1402 req->client = client;
1403 return req;
1404 }
1405
1406 static void nbd_request_put(NBDRequestData *req)
1407 {
1408 NBDClient *client = req->client;
1409
1410 if (req->data) {
1411 qemu_vfree(req->data);
1412 }
1413 g_free(req);
1414
1415 client->nb_requests--;
1416 nbd_client_receive_next_request(client);
1417
1418 nbd_client_put(client);
1419 }
1420
1421 static void blk_aio_attached(AioContext *ctx, void *opaque)
1422 {
1423 NBDExport *exp = opaque;
1424 NBDClient *client;
1425
1426 trace_nbd_blk_aio_attached(exp->name, ctx);
1427
1428 exp->ctx = ctx;
1429
1430 QTAILQ_FOREACH(client, &exp->clients, next) {
1431 qio_channel_attach_aio_context(client->ioc, ctx);
1432 if (client->recv_coroutine) {
1433 aio_co_schedule(ctx, client->recv_coroutine);
1434 }
1435 if (client->send_coroutine) {
1436 aio_co_schedule(ctx, client->send_coroutine);
1437 }
1438 }
1439 }
1440
1441 static void blk_aio_detach(void *opaque)
1442 {
1443 NBDExport *exp = opaque;
1444 NBDClient *client;
1445
1446 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1447
1448 QTAILQ_FOREACH(client, &exp->clients, next) {
1449 qio_channel_detach_aio_context(client->ioc);
1450 }
1451
1452 exp->ctx = NULL;
1453 }
1454
1455 static void nbd_eject_notifier(Notifier *n, void *data)
1456 {
1457 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1458 nbd_export_close(exp);
1459 }
1460
1461 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1462 uint64_t size, const char *name, const char *desc,
1463 const char *bitmap, uint16_t nbdflags,
1464 void (*close)(NBDExport *), bool writethrough,
1465 BlockBackend *on_eject_blk, Error **errp)
1466 {
1467 AioContext *ctx;
1468 BlockBackend *blk;
1469 NBDExport *exp = g_new0(NBDExport, 1);
1470 uint64_t perm;
1471 int ret;
1472
1473 /*
1474 * NBD exports are used for non-shared storage migration. Make sure
1475 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1476 * access since the export could be available before migration handover.
1477 */
1478 assert(name);
1479 ctx = bdrv_get_aio_context(bs);
1480 aio_context_acquire(ctx);
1481 bdrv_invalidate_cache(bs, NULL);
1482 aio_context_release(ctx);
1483
1484 /* Don't allow resize while the NBD server is running, otherwise we don't
1485 * care what happens with the node. */
1486 perm = BLK_PERM_CONSISTENT_READ;
1487 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1488 perm |= BLK_PERM_WRITE;
1489 }
1490 blk = blk_new(bdrv_get_aio_context(bs), perm,
1491 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1492 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1493 ret = blk_insert_bs(blk, bs, errp);
1494 if (ret < 0) {
1495 goto fail;
1496 }
1497 blk_set_enable_write_cache(blk, !writethrough);
1498 blk_set_allow_aio_context_change(blk, true);
1499
1500 exp->refcount = 1;
1501 QTAILQ_INIT(&exp->clients);
1502 exp->blk = blk;
1503 assert(dev_offset <= INT64_MAX);
1504 exp->dev_offset = dev_offset;
1505 exp->name = g_strdup(name);
1506 exp->description = g_strdup(desc);
1507 exp->nbdflags = nbdflags;
1508 assert(size <= INT64_MAX - dev_offset);
1509 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1510
1511 if (bitmap) {
1512 BdrvDirtyBitmap *bm = NULL;
1513
1514 while (true) {
1515 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1516 if (bm != NULL || bs->backing == NULL) {
1517 break;
1518 }
1519
1520 bs = bs->backing->bs;
1521 }
1522
1523 if (bm == NULL) {
1524 error_setg(errp, "Bitmap '%s' is not found", bitmap);
1525 goto fail;
1526 }
1527
1528 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1529 goto fail;
1530 }
1531
1532 if ((nbdflags & NBD_FLAG_READ_ONLY) && bdrv_is_writable(bs) &&
1533 bdrv_dirty_bitmap_enabled(bm)) {
1534 error_setg(errp,
1535 "Enabled bitmap '%s' incompatible with readonly export",
1536 bitmap);
1537 goto fail;
1538 }
1539
1540 bdrv_dirty_bitmap_set_busy(bm, true);
1541 exp->export_bitmap = bm;
1542 exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1543 bitmap);
1544 }
1545
1546 exp->close = close;
1547 exp->ctx = blk_get_aio_context(blk);
1548 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1549
1550 if (on_eject_blk) {
1551 blk_ref(on_eject_blk);
1552 exp->eject_notifier_blk = on_eject_blk;
1553 exp->eject_notifier.notify = nbd_eject_notifier;
1554 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1555 }
1556 QTAILQ_INSERT_TAIL(&exports, exp, next);
1557 nbd_export_get(exp);
1558 return exp;
1559
1560 fail:
1561 blk_unref(blk);
1562 g_free(exp->name);
1563 g_free(exp->description);
1564 g_free(exp);
1565 return NULL;
1566 }
1567
1568 NBDExport *nbd_export_find(const char *name)
1569 {
1570 NBDExport *exp;
1571 QTAILQ_FOREACH(exp, &exports, next) {
1572 if (strcmp(name, exp->name) == 0) {
1573 return exp;
1574 }
1575 }
1576
1577 return NULL;
1578 }
1579
1580 void nbd_export_close(NBDExport *exp)
1581 {
1582 NBDClient *client, *next;
1583
1584 nbd_export_get(exp);
1585 /*
1586 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1587 * close mode that stops advertising the export to new clients but
1588 * still permits existing clients to run to completion? Because of
1589 * that possibility, nbd_export_close() can be called more than
1590 * once on an export.
1591 */
1592 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1593 client_close(client, true);
1594 }
1595 if (exp->name) {
1596 nbd_export_put(exp);
1597 g_free(exp->name);
1598 exp->name = NULL;
1599 QTAILQ_REMOVE(&exports, exp, next);
1600 }
1601 g_free(exp->description);
1602 exp->description = NULL;
1603 nbd_export_put(exp);
1604 }
1605
1606 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1607 {
1608 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1609 nbd_export_close(exp);
1610 return;
1611 }
1612
1613 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1614
1615 error_setg(errp, "export '%s' still in use", exp->name);
1616 error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1617 }
1618
1619 void nbd_export_get(NBDExport *exp)
1620 {
1621 assert(exp->refcount > 0);
1622 exp->refcount++;
1623 }
1624
1625 void nbd_export_put(NBDExport *exp)
1626 {
1627 assert(exp->refcount > 0);
1628 if (exp->refcount == 1) {
1629 nbd_export_close(exp);
1630 }
1631
1632 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1633 * if someone calls nbd_export_put() on named export not through
1634 * nbd_export_set_name() when refcount is 1. So, let's assert that
1635 * it is > 0.
1636 */
1637 assert(exp->refcount > 0);
1638 if (--exp->refcount == 0) {
1639 assert(exp->name == NULL);
1640 assert(exp->description == NULL);
1641
1642 if (exp->close) {
1643 exp->close(exp);
1644 }
1645
1646 if (exp->blk) {
1647 if (exp->eject_notifier_blk) {
1648 notifier_remove(&exp->eject_notifier);
1649 blk_unref(exp->eject_notifier_blk);
1650 }
1651 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1652 blk_aio_detach, exp);
1653 blk_unref(exp->blk);
1654 exp->blk = NULL;
1655 }
1656
1657 if (exp->export_bitmap) {
1658 bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1659 g_free(exp->export_bitmap_context);
1660 }
1661
1662 g_free(exp);
1663 }
1664 }
1665
1666 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1667 {
1668 return exp->blk;
1669 }
1670
1671 void nbd_export_close_all(void)
1672 {
1673 NBDExport *exp, *next;
1674
1675 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1676 nbd_export_close(exp);
1677 }
1678 }
1679
1680 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1681 unsigned niov, Error **errp)
1682 {
1683 int ret;
1684
1685 g_assert(qemu_in_coroutine());
1686 qemu_co_mutex_lock(&client->send_lock);
1687 client->send_coroutine = qemu_coroutine_self();
1688
1689 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1690
1691 client->send_coroutine = NULL;
1692 qemu_co_mutex_unlock(&client->send_lock);
1693
1694 return ret;
1695 }
1696
1697 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1698 uint64_t handle)
1699 {
1700 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1701 stl_be_p(&reply->error, error);
1702 stq_be_p(&reply->handle, handle);
1703 }
1704
1705 static int nbd_co_send_simple_reply(NBDClient *client,
1706 uint64_t handle,
1707 uint32_t error,
1708 void *data,
1709 size_t len,
1710 Error **errp)
1711 {
1712 NBDSimpleReply reply;
1713 int nbd_err = system_errno_to_nbd_errno(error);
1714 struct iovec iov[] = {
1715 {.iov_base = &reply, .iov_len = sizeof(reply)},
1716 {.iov_base = data, .iov_len = len}
1717 };
1718
1719 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1720 len);
1721 set_be_simple_reply(&reply, nbd_err, handle);
1722
1723 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1724 }
1725
1726 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1727 uint16_t type, uint64_t handle, uint32_t length)
1728 {
1729 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1730 stw_be_p(&chunk->flags, flags);
1731 stw_be_p(&chunk->type, type);
1732 stq_be_p(&chunk->handle, handle);
1733 stl_be_p(&chunk->length, length);
1734 }
1735
1736 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1737 uint64_t handle,
1738 Error **errp)
1739 {
1740 NBDStructuredReplyChunk chunk;
1741 struct iovec iov[] = {
1742 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1743 };
1744
1745 trace_nbd_co_send_structured_done(handle);
1746 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1747
1748 return nbd_co_send_iov(client, iov, 1, errp);
1749 }
1750
1751 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1752 uint64_t handle,
1753 uint64_t offset,
1754 void *data,
1755 size_t size,
1756 bool final,
1757 Error **errp)
1758 {
1759 NBDStructuredReadData chunk;
1760 struct iovec iov[] = {
1761 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1762 {.iov_base = data, .iov_len = size}
1763 };
1764
1765 assert(size);
1766 trace_nbd_co_send_structured_read(handle, offset, data, size);
1767 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1768 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1769 sizeof(chunk) - sizeof(chunk.h) + size);
1770 stq_be_p(&chunk.offset, offset);
1771
1772 return nbd_co_send_iov(client, iov, 2, errp);
1773 }
1774
1775 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1776 uint64_t handle,
1777 uint32_t error,
1778 const char *msg,
1779 Error **errp)
1780 {
1781 NBDStructuredError chunk;
1782 int nbd_err = system_errno_to_nbd_errno(error);
1783 struct iovec iov[] = {
1784 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1785 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1786 };
1787
1788 assert(nbd_err);
1789 trace_nbd_co_send_structured_error(handle, nbd_err,
1790 nbd_err_lookup(nbd_err), msg ? msg : "");
1791 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1792 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1793 stl_be_p(&chunk.error, nbd_err);
1794 stw_be_p(&chunk.message_length, iov[1].iov_len);
1795
1796 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1797 }
1798
1799 /* Do a sparse read and send the structured reply to the client.
1800 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1801 * reported to the client, at which point this function succeeds.
1802 */
1803 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1804 uint64_t handle,
1805 uint64_t offset,
1806 uint8_t *data,
1807 size_t size,
1808 Error **errp)
1809 {
1810 int ret = 0;
1811 NBDExport *exp = client->exp;
1812 size_t progress = 0;
1813
1814 while (progress < size) {
1815 int64_t pnum;
1816 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1817 offset + progress,
1818 size - progress, &pnum, NULL,
1819 NULL);
1820 bool final;
1821
1822 if (status < 0) {
1823 char *msg = g_strdup_printf("unable to check for holes: %s",
1824 strerror(-status));
1825
1826 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1827 errp);
1828 g_free(msg);
1829 return ret;
1830 }
1831 assert(pnum && pnum <= size - progress);
1832 final = progress + pnum == size;
1833 if (status & BDRV_BLOCK_ZERO) {
1834 NBDStructuredReadHole chunk;
1835 struct iovec iov[] = {
1836 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1837 };
1838
1839 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1840 pnum);
1841 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1842 NBD_REPLY_TYPE_OFFSET_HOLE,
1843 handle, sizeof(chunk) - sizeof(chunk.h));
1844 stq_be_p(&chunk.offset, offset + progress);
1845 stl_be_p(&chunk.length, pnum);
1846 ret = nbd_co_send_iov(client, iov, 1, errp);
1847 } else {
1848 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1849 data + progress, pnum);
1850 if (ret < 0) {
1851 error_setg_errno(errp, -ret, "reading from file failed");
1852 break;
1853 }
1854 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1855 data + progress, pnum, final,
1856 errp);
1857 }
1858
1859 if (ret < 0) {
1860 break;
1861 }
1862 progress += pnum;
1863 }
1864 return ret;
1865 }
1866
1867 /*
1868 * Populate @extents from block status. Update @bytes to be the actual
1869 * length encoded (which may be smaller than the original), and update
1870 * @nb_extents to the number of extents used.
1871 *
1872 * Returns zero on success and -errno on bdrv_block_status_above failure.
1873 */
1874 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1875 uint64_t *bytes, NBDExtent *extents,
1876 unsigned int *nb_extents)
1877 {
1878 uint64_t remaining_bytes = *bytes;
1879 NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1880 bool first_extent = true;
1881
1882 assert(*nb_extents);
1883 while (remaining_bytes) {
1884 uint32_t flags;
1885 int64_t num;
1886 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1887 &num, NULL, NULL);
1888
1889 if (ret < 0) {
1890 return ret;
1891 }
1892
1893 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1894 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1895
1896 if (first_extent) {
1897 extent->flags = flags;
1898 extent->length = num;
1899 first_extent = false;
1900 } else if (flags == extent->flags) {
1901 /* extend current extent */
1902 extent->length += num;
1903 } else {
1904 if (extent + 1 == extents_end) {
1905 break;
1906 }
1907
1908 /* start new extent */
1909 extent++;
1910 extent->flags = flags;
1911 extent->length = num;
1912 }
1913 offset += num;
1914 remaining_bytes -= num;
1915 }
1916
1917 extents_end = extent + 1;
1918
1919 for (extent = extents; extent < extents_end; extent++) {
1920 extent->flags = cpu_to_be32(extent->flags);
1921 extent->length = cpu_to_be32(extent->length);
1922 }
1923
1924 *bytes -= remaining_bytes;
1925 *nb_extents = extents_end - extents;
1926
1927 return 0;
1928 }
1929
1930 /* nbd_co_send_extents
1931 *
1932 * @length is only for tracing purposes (and may be smaller or larger
1933 * than the client's original request). @last controls whether
1934 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1935 * big-endian format.
1936 */
1937 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1938 NBDExtent *extents, unsigned int nb_extents,
1939 uint64_t length, bool last,
1940 uint32_t context_id, Error **errp)
1941 {
1942 NBDStructuredMeta chunk;
1943
1944 struct iovec iov[] = {
1945 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1946 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1947 };
1948
1949 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1950 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1951 NBD_REPLY_TYPE_BLOCK_STATUS,
1952 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1953 stl_be_p(&chunk.context_id, context_id);
1954
1955 return nbd_co_send_iov(client, iov, 2, errp);
1956 }
1957
1958 /* Get block status from the exported device and send it to the client */
1959 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1960 BlockDriverState *bs, uint64_t offset,
1961 uint32_t length, bool dont_fragment,
1962 bool last, uint32_t context_id,
1963 Error **errp)
1964 {
1965 int ret;
1966 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
1967 NBDExtent *extents = g_new(NBDExtent, nb_extents);
1968 uint64_t final_length = length;
1969
1970 ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1971 &nb_extents);
1972 if (ret < 0) {
1973 g_free(extents);
1974 return nbd_co_send_structured_error(
1975 client, handle, -ret, "can't get block status", errp);
1976 }
1977
1978 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1979 final_length, last, context_id, errp);
1980
1981 g_free(extents);
1982
1983 return ret;
1984 }
1985
1986 /*
1987 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1988 * final extent may exceed the original @length. Store in @length the
1989 * byte length encoded (which may be smaller or larger than the
1990 * original), and return the number of extents used.
1991 */
1992 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1993 uint64_t *length, NBDExtent *extents,
1994 unsigned int nb_extents,
1995 bool dont_fragment)
1996 {
1997 uint64_t begin = offset, end = offset;
1998 uint64_t overall_end = offset + *length;
1999 unsigned int i = 0;
2000 BdrvDirtyBitmapIter *it;
2001 bool dirty;
2002
2003 bdrv_dirty_bitmap_lock(bitmap);
2004
2005 it = bdrv_dirty_iter_new(bitmap);
2006 dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
2007
2008 assert(begin < overall_end && nb_extents);
2009 while (begin < overall_end && i < nb_extents) {
2010 bool next_dirty = !dirty;
2011
2012 if (dirty) {
2013 end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2014 } else {
2015 bdrv_set_dirty_iter(it, begin);
2016 end = bdrv_dirty_iter_next(it);
2017 }
2018 if (end == -1 || end - begin > UINT32_MAX) {
2019 /* Cap to an aligned value < 4G beyond begin. */
2020 end = MIN(bdrv_dirty_bitmap_size(bitmap),
2021 begin + UINT32_MAX + 1 -
2022 bdrv_dirty_bitmap_granularity(bitmap));
2023 next_dirty = dirty;
2024 }
2025 if (dont_fragment && end > overall_end) {
2026 end = overall_end;
2027 }
2028
2029 extents[i].length = cpu_to_be32(end - begin);
2030 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2031 i++;
2032 begin = end;
2033 dirty = next_dirty;
2034 }
2035
2036 bdrv_dirty_iter_free(it);
2037
2038 bdrv_dirty_bitmap_unlock(bitmap);
2039
2040 assert(offset < end);
2041 *length = end - offset;
2042 return i;
2043 }
2044
2045 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2046 BdrvDirtyBitmap *bitmap, uint64_t offset,
2047 uint32_t length, bool dont_fragment, bool last,
2048 uint32_t context_id, Error **errp)
2049 {
2050 int ret;
2051 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2052 NBDExtent *extents = g_new(NBDExtent, nb_extents);
2053 uint64_t final_length = length;
2054
2055 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2056 nb_extents, dont_fragment);
2057
2058 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2059 final_length, last, context_id, errp);
2060
2061 g_free(extents);
2062
2063 return ret;
2064 }
2065
2066 /* nbd_co_receive_request
2067 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2068 * connection right away, and any other negative value to report an error to
2069 * the client (although the caller may still need to disconnect after reporting
2070 * the error).
2071 */
2072 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2073 Error **errp)
2074 {
2075 NBDClient *client = req->client;
2076 int valid_flags;
2077
2078 g_assert(qemu_in_coroutine());
2079 assert(client->recv_coroutine == qemu_coroutine_self());
2080 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2081 return -EIO;
2082 }
2083
2084 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2085 nbd_cmd_lookup(request->type));
2086
2087 if (request->type != NBD_CMD_WRITE) {
2088 /* No payload, we are ready to read the next request. */
2089 req->complete = true;
2090 }
2091
2092 if (request->type == NBD_CMD_DISC) {
2093 /* Special case: we're going to disconnect without a reply,
2094 * whether or not flags, from, or len are bogus */
2095 return -EIO;
2096 }
2097
2098 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2099 request->type == NBD_CMD_CACHE)
2100 {
2101 if (request->len > NBD_MAX_BUFFER_SIZE) {
2102 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2103 request->len, NBD_MAX_BUFFER_SIZE);
2104 return -EINVAL;
2105 }
2106
2107 req->data = blk_try_blockalign(client->exp->blk, request->len);
2108 if (req->data == NULL) {
2109 error_setg(errp, "No memory");
2110 return -ENOMEM;
2111 }
2112 }
2113 if (request->type == NBD_CMD_WRITE) {
2114 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2115 errp) < 0)
2116 {
2117 return -EIO;
2118 }
2119 req->complete = true;
2120
2121 trace_nbd_co_receive_request_payload_received(request->handle,
2122 request->len);
2123 }
2124
2125 /* Sanity checks. */
2126 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2127 (request->type == NBD_CMD_WRITE ||
2128 request->type == NBD_CMD_WRITE_ZEROES ||
2129 request->type == NBD_CMD_TRIM)) {
2130 error_setg(errp, "Export is read-only");
2131 return -EROFS;
2132 }
2133 if (request->from > client->exp->size ||
2134 request->len > client->exp->size - request->from) {
2135 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2136 ", Size: %" PRIu64, request->from, request->len,
2137 client->exp->size);
2138 return (request->type == NBD_CMD_WRITE ||
2139 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2140 }
2141 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2142 client->check_align)) {
2143 /*
2144 * The block layer gracefully handles unaligned requests, but
2145 * it's still worth tracing client non-compliance
2146 */
2147 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2148 request->from,
2149 request->len,
2150 client->check_align);
2151 }
2152 valid_flags = NBD_CMD_FLAG_FUA;
2153 if (request->type == NBD_CMD_READ && client->structured_reply) {
2154 valid_flags |= NBD_CMD_FLAG_DF;
2155 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2156 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2157 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2158 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2159 }
2160 if (request->flags & ~valid_flags) {
2161 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2162 nbd_cmd_lookup(request->type), request->flags);
2163 return -EINVAL;
2164 }
2165
2166 return 0;
2167 }
2168
2169 /* Send simple reply without a payload, or a structured error
2170 * @error_msg is ignored if @ret >= 0
2171 * Returns 0 if connection is still live, -errno on failure to talk to client
2172 */
2173 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2174 uint64_t handle,
2175 int ret,
2176 const char *error_msg,
2177 Error **errp)
2178 {
2179 if (client->structured_reply && ret < 0) {
2180 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2181 errp);
2182 } else {
2183 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2184 NULL, 0, errp);
2185 }
2186 }
2187
2188 /* Handle NBD_CMD_READ request.
2189 * Return -errno if sending fails. Other errors are reported directly to the
2190 * client as an error reply. */
2191 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2192 uint8_t *data, Error **errp)
2193 {
2194 int ret;
2195 NBDExport *exp = client->exp;
2196
2197 assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2198
2199 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2200 if (request->flags & NBD_CMD_FLAG_FUA) {
2201 ret = blk_co_flush(exp->blk);
2202 if (ret < 0) {
2203 return nbd_send_generic_reply(client, request->handle, ret,
2204 "flush failed", errp);
2205 }
2206 }
2207
2208 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2209 request->len && request->type != NBD_CMD_CACHE)
2210 {
2211 return nbd_co_send_sparse_read(client, request->handle, request->from,
2212 data, request->len, errp);
2213 }
2214
2215 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2216 request->len);
2217 if (ret < 0 || request->type == NBD_CMD_CACHE) {
2218 return nbd_send_generic_reply(client, request->handle, ret,
2219 "reading from file failed", errp);
2220 }
2221
2222 if (client->structured_reply) {
2223 if (request->len) {
2224 return nbd_co_send_structured_read(client, request->handle,
2225 request->from, data,
2226 request->len, true, errp);
2227 } else {
2228 return nbd_co_send_structured_done(client, request->handle, errp);
2229 }
2230 } else {
2231 return nbd_co_send_simple_reply(client, request->handle, 0,
2232 data, request->len, errp);
2233 }
2234 }
2235
2236 /* Handle NBD request.
2237 * Return -errno if sending fails. Other errors are reported directly to the
2238 * client as an error reply. */
2239 static coroutine_fn int nbd_handle_request(NBDClient *client,
2240 NBDRequest *request,
2241 uint8_t *data, Error **errp)
2242 {
2243 int ret;
2244 int flags;
2245 NBDExport *exp = client->exp;
2246 char *msg;
2247
2248 switch (request->type) {
2249 case NBD_CMD_READ:
2250 case NBD_CMD_CACHE:
2251 return nbd_do_cmd_read(client, request, data, errp);
2252
2253 case NBD_CMD_WRITE:
2254 flags = 0;
2255 if (request->flags & NBD_CMD_FLAG_FUA) {
2256 flags |= BDRV_REQ_FUA;
2257 }
2258 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2259 data, request->len, flags);
2260 return nbd_send_generic_reply(client, request->handle, ret,
2261 "writing to file failed", errp);
2262
2263 case NBD_CMD_WRITE_ZEROES:
2264 flags = 0;
2265 if (request->flags & NBD_CMD_FLAG_FUA) {
2266 flags |= BDRV_REQ_FUA;
2267 }
2268 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2269 flags |= BDRV_REQ_MAY_UNMAP;
2270 }
2271 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2272 request->len, flags);
2273 return nbd_send_generic_reply(client, request->handle, ret,
2274 "writing to file failed", errp);
2275
2276 case NBD_CMD_DISC:
2277 /* unreachable, thanks to special case in nbd_co_receive_request() */
2278 abort();
2279
2280 case NBD_CMD_FLUSH:
2281 ret = blk_co_flush(exp->blk);
2282 return nbd_send_generic_reply(client, request->handle, ret,
2283 "flush failed", errp);
2284
2285 case NBD_CMD_TRIM:
2286 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2287 request->len);
2288 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2289 ret = blk_co_flush(exp->blk);
2290 }
2291 return nbd_send_generic_reply(client, request->handle, ret,
2292 "discard failed", errp);
2293
2294 case NBD_CMD_BLOCK_STATUS:
2295 if (!request->len) {
2296 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2297 "need non-zero length", errp);
2298 }
2299 if (client->export_meta.valid &&
2300 (client->export_meta.base_allocation ||
2301 client->export_meta.bitmap))
2302 {
2303 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2304
2305 if (client->export_meta.base_allocation) {
2306 ret = nbd_co_send_block_status(client, request->handle,
2307 blk_bs(exp->blk), request->from,
2308 request->len, dont_fragment,
2309 !client->export_meta.bitmap,
2310 NBD_META_ID_BASE_ALLOCATION,
2311 errp);
2312 if (ret < 0) {
2313 return ret;
2314 }
2315 }
2316
2317 if (client->export_meta.bitmap) {
2318 ret = nbd_co_send_bitmap(client, request->handle,
2319 client->exp->export_bitmap,
2320 request->from, request->len,
2321 dont_fragment,
2322 true, NBD_META_ID_DIRTY_BITMAP, errp);
2323 if (ret < 0) {
2324 return ret;
2325 }
2326 }
2327
2328 return ret;
2329 } else {
2330 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2331 "CMD_BLOCK_STATUS not negotiated",
2332 errp);
2333 }
2334
2335 default:
2336 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2337 request->type);
2338 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2339 errp);
2340 g_free(msg);
2341 return ret;
2342 }
2343 }
2344
2345 /* Owns a reference to the NBDClient passed as opaque. */
2346 static coroutine_fn void nbd_trip(void *opaque)
2347 {
2348 NBDClient *client = opaque;
2349 NBDRequestData *req;
2350 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2351 int ret;
2352 Error *local_err = NULL;
2353
2354 trace_nbd_trip();
2355 if (client->closing) {
2356 nbd_client_put(client);
2357 return;
2358 }
2359
2360 req = nbd_request_get(client);
2361 ret = nbd_co_receive_request(req, &request, &local_err);
2362 client->recv_coroutine = NULL;
2363
2364 if (client->closing) {
2365 /*
2366 * The client may be closed when we are blocked in
2367 * nbd_co_receive_request()
2368 */
2369 goto done;
2370 }
2371
2372 nbd_client_receive_next_request(client);
2373 if (ret == -EIO) {
2374 goto disconnect;
2375 }
2376
2377 if (ret < 0) {
2378 /* It wans't -EIO, so, according to nbd_co_receive_request()
2379 * semantics, we should return the error to the client. */
2380 Error *export_err = local_err;
2381
2382 local_err = NULL;
2383 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2384 error_get_pretty(export_err), &local_err);
2385 error_free(export_err);
2386 } else {
2387 ret = nbd_handle_request(client, &request, req->data, &local_err);
2388 }
2389 if (ret < 0) {
2390 error_prepend(&local_err, "Failed to send reply: ");
2391 goto disconnect;
2392 }
2393
2394 /* We must disconnect after NBD_CMD_WRITE if we did not
2395 * read the payload.
2396 */
2397 if (!req->complete) {
2398 error_setg(&local_err, "Request handling failed in intermediate state");
2399 goto disconnect;
2400 }
2401
2402 done:
2403 nbd_request_put(req);
2404 nbd_client_put(client);
2405 return;
2406
2407 disconnect:
2408 if (local_err) {
2409 error_reportf_err(local_err, "Disconnect client, due to: ");
2410 }
2411 nbd_request_put(req);
2412 client_close(client, true);
2413 nbd_client_put(client);
2414 }
2415
2416 static void nbd_client_receive_next_request(NBDClient *client)
2417 {
2418 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2419 nbd_client_get(client);
2420 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2421 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2422 }
2423 }
2424
2425 static coroutine_fn void nbd_co_client_start(void *opaque)
2426 {
2427 NBDClient *client = opaque;
2428 Error *local_err = NULL;
2429
2430 qemu_co_mutex_init(&client->send_lock);
2431
2432 if (nbd_negotiate(client, &local_err)) {
2433 if (local_err) {
2434 error_report_err(local_err);
2435 }
2436 client_close(client, false);
2437 return;
2438 }
2439
2440 nbd_client_receive_next_request(client);
2441 }
2442
2443 /*
2444 * Create a new client listener using the given channel @sioc.
2445 * Begin servicing it in a coroutine. When the connection closes, call
2446 * @close_fn with an indication of whether the client completed negotiation.
2447 */
2448 void nbd_client_new(QIOChannelSocket *sioc,
2449 QCryptoTLSCreds *tlscreds,
2450 const char *tlsauthz,
2451 void (*close_fn)(NBDClient *, bool))
2452 {
2453 NBDClient *client;
2454 Coroutine *co;
2455
2456 client = g_new0(NBDClient, 1);
2457 client->refcount = 1;
2458 client->tlscreds = tlscreds;
2459 if (tlscreds) {
2460 object_ref(OBJECT(client->tlscreds));
2461 }
2462 client->tlsauthz = g_strdup(tlsauthz);
2463 client->sioc = sioc;
2464 object_ref(OBJECT(client->sioc));
2465 client->ioc = QIO_CHANNEL(sioc);
2466 object_ref(OBJECT(client->ioc));
2467 client->close_fn = close_fn;
2468
2469 co = qemu_coroutine_create(nbd_co_client_start, client);
2470 qemu_coroutine_enter(co);
2471 }
AR7 emulation for QEMU