?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
linux-user: Pass CPUState to MAX_RESERVED_VA
[qemu/ar7.git] / nbd / server.c
blob28c3c8be854c22079d85f72752406bf0595bc9cc
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 "qemu/queue.h"
23 #include "trace.h"
24 #include "nbd-internal.h"
25 #include "qemu/units.h"
26
27 #define NBD_META_ID_BASE_ALLOCATION 0
28 #define NBD_META_ID_DIRTY_BITMAP 1
29
30 /*
31 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
32 * constant. If an increase is needed, note that the NBD protocol
33 * recommends no larger than 32 mb, so that the client won't consider
34 * the reply as a denial of service attack.
35 */
36 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
37
38 static int system_errno_to_nbd_errno(int err)
39 {
40 switch (err) {
41 case 0:
42 return NBD_SUCCESS;
43 case EPERM:
44 case EROFS:
45 return NBD_EPERM;
46 case EIO:
47 return NBD_EIO;
48 case ENOMEM:
49 return NBD_ENOMEM;
50 #ifdef EDQUOT
51 case EDQUOT:
52 #endif
53 case EFBIG:
54 case ENOSPC:
55 return NBD_ENOSPC;
56 case EOVERFLOW:
57 return NBD_EOVERFLOW;
58 case ENOTSUP:
59 #if ENOTSUP != EOPNOTSUPP
60 case EOPNOTSUPP:
61 #endif
62 return NBD_ENOTSUP;
63 case ESHUTDOWN:
64 return NBD_ESHUTDOWN;
65 case EINVAL:
66 default:
67 return NBD_EINVAL;
68 }
69 }
70
71 /* Definitions for opaque data types */
72
73 typedef struct NBDRequestData NBDRequestData;
74
75 struct NBDRequestData {
76 QSIMPLEQ_ENTRY(NBDRequestData) entry;
77 NBDClient *client;
78 uint8_t *data;
79 bool complete;
80 };
81
82 struct NBDExport {
83 int refcount;
84 void (*close)(NBDExport *exp);
85
86 BlockBackend *blk;
87 char *name;
88 char *description;
89 uint64_t dev_offset;
90 uint64_t size;
91 uint16_t nbdflags;
92 QTAILQ_HEAD(, NBDClient) clients;
93 QTAILQ_ENTRY(NBDExport) next;
94
95 AioContext *ctx;
96
97 BlockBackend *eject_notifier_blk;
98 Notifier eject_notifier;
99
100 BdrvDirtyBitmap *export_bitmap;
101 char *export_bitmap_context;
102 };
103
104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
105
106 /* NBDExportMetaContexts represents a list of contexts to be exported,
107 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
108 * NBD_OPT_LIST_META_CONTEXT. */
109 typedef struct NBDExportMetaContexts {
110 NBDExport *exp;
111 bool valid; /* means that negotiation of the option finished without
112 errors */
113 bool base_allocation; /* export base:allocation context (block status) */
114 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
115 } NBDExportMetaContexts;
116
117 struct NBDClient {
118 int refcount;
119 void (*close_fn)(NBDClient *client, bool negotiated);
120
121 NBDExport *exp;
122 QCryptoTLSCreds *tlscreds;
123 char *tlsauthz;
124 QIOChannelSocket *sioc; /* The underlying data channel */
125 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
126
127 Coroutine *recv_coroutine;
128
129 CoMutex send_lock;
130 Coroutine *send_coroutine;
131
132 QTAILQ_ENTRY(NBDClient) next;
133 int nb_requests;
134 bool closing;
135
136 uint32_t check_align; /* If non-zero, check for aligned client requests */
137
138 bool structured_reply;
139 NBDExportMetaContexts export_meta;
140
141 uint32_t opt; /* Current option being negotiated */
142 uint32_t optlen; /* remaining length of data in ioc for the option being
143 negotiated now */
144 };
145
146 static void nbd_client_receive_next_request(NBDClient *client);
147
148 /* Basic flow for negotiation
149
150 Server Client
151 Negotiate
152
153 or
154
155 Server Client
156 Negotiate #1
157 Option
158 Negotiate #2
159
160 ----
161
162 followed by
163
164 Server Client
165 Request
166 Response
167 Request
168 Response
169 ...
170 ...
171 Request (type == 2)
172
173 */
174
175 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
176 uint32_t type, uint32_t length)
177 {
178 stq_be_p(&rep->magic, NBD_REP_MAGIC);
179 stl_be_p(&rep->option, option);
180 stl_be_p(&rep->type, type);
181 stl_be_p(&rep->length, length);
182 }
183
184 /* Send a reply header, including length, but no payload.
185 * Return -errno on error, 0 on success. */
186 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
187 uint32_t len, Error **errp)
188 {
189 NBDOptionReply rep;
190
191 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
192 type, nbd_rep_lookup(type), len);
193
194 assert(len < NBD_MAX_BUFFER_SIZE);
195
196 set_be_option_rep(&rep, client->opt, type, len);
197 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
198 }
199
200 /* Send a reply header with default 0 length.
201 * Return -errno on error, 0 on success. */
202 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
203 Error **errp)
204 {
205 return nbd_negotiate_send_rep_len(client, type, 0, errp);
206 }
207
208 /* Send an error reply.
209 * Return -errno on error, 0 on success. */
210 static int GCC_FMT_ATTR(4, 0)
211 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
212 Error **errp, const char *fmt, va_list va)
213 {
214 g_autofree char *msg = NULL;
215 int ret;
216 size_t len;
217
218 msg = g_strdup_vprintf(fmt, va);
219 len = strlen(msg);
220 assert(len < 4096);
221 trace_nbd_negotiate_send_rep_err(msg);
222 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
223 if (ret < 0) {
224 return ret;
225 }
226 if (nbd_write(client->ioc, msg, len, errp) < 0) {
227 error_prepend(errp, "write failed (error message): ");
228 return -EIO;
229 }
230
231 return 0;
232 }
233
234 /* Send an error reply.
235 * Return -errno on error, 0 on success. */
236 static int GCC_FMT_ATTR(4, 5)
237 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
238 Error **errp, const char *fmt, ...)
239 {
240 va_list va;
241 int ret;
242
243 va_start(va, fmt);
244 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
245 va_end(va);
246 return ret;
247 }
248
249 /* Drop remainder of the current option, and send a reply with the
250 * given error type and message. Return -errno on read or write
251 * failure; or 0 if connection is still live. */
252 static int GCC_FMT_ATTR(4, 0)
253 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
254 const char *fmt, va_list va)
255 {
256 int ret = nbd_drop(client->ioc, client->optlen, errp);
257
258 client->optlen = 0;
259 if (!ret) {
260 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
261 }
262 return ret;
263 }
264
265 static int GCC_FMT_ATTR(4, 5)
266 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
267 const char *fmt, ...)
268 {
269 int ret;
270 va_list va;
271
272 va_start(va, fmt);
273 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
274 va_end(va);
275
276 return ret;
277 }
278
279 static int GCC_FMT_ATTR(3, 4)
280 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
281 {
282 int ret;
283 va_list va;
284
285 va_start(va, fmt);
286 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
287 va_end(va);
288
289 return ret;
290 }
291
292 /* Read size bytes from the unparsed payload of the current option.
293 * Return -errno on I/O error, 0 if option was completely handled by
294 * sending a reply about inconsistent lengths, or 1 on success. */
295 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
296 Error **errp)
297 {
298 if (size > client->optlen) {
299 return nbd_opt_invalid(client, errp,
300 "Inconsistent lengths in option %s",
301 nbd_opt_lookup(client->opt));
302 }
303 client->optlen -= size;
304 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
305 }
306
307 /* Drop size bytes from the unparsed payload of the current option.
308 * Return -errno on I/O error, 0 if option was completely handled by
309 * sending a reply about inconsistent lengths, or 1 on success. */
310 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
311 {
312 if (size > client->optlen) {
313 return nbd_opt_invalid(client, errp,
314 "Inconsistent lengths in option %s",
315 nbd_opt_lookup(client->opt));
316 }
317 client->optlen -= size;
318 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
319 }
320
321 /* nbd_opt_read_name
322 *
323 * Read a string with the format:
324 * uint32_t len (<= NBD_MAX_NAME_SIZE)
325 * len bytes string (not 0-terminated)
326 *
327 * @name should be enough to store NBD_MAX_NAME_SIZE+1.
328 * If @length is non-null, it will be set to the actual string length.
329 *
330 * Return -errno on I/O error, 0 if option was completely handled by
331 * sending a reply about inconsistent lengths, or 1 on success.
332 */
333 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
334 Error **errp)
335 {
336 int ret;
337 uint32_t len;
338
339 ret = nbd_opt_read(client, &len, sizeof(len), errp);
340 if (ret <= 0) {
341 return ret;
342 }
343 len = cpu_to_be32(len);
344
345 if (len > NBD_MAX_NAME_SIZE) {
346 return nbd_opt_invalid(client, errp,
347 "Invalid name length: %" PRIu32, len);
348 }
349
350 ret = nbd_opt_read(client, name, len, errp);
351 if (ret <= 0) {
352 return ret;
353 }
354 name[len] = '\0';
355
356 if (length) {
357 *length = len;
358 }
359
360 return 1;
361 }
362
363 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
364 * Return -errno on error, 0 on success. */
365 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
366 Error **errp)
367 {
368 size_t name_len, desc_len;
369 uint32_t len;
370 const char *name = exp->name ? exp->name : "";
371 const char *desc = exp->description ? exp->description : "";
372 QIOChannel *ioc = client->ioc;
373 int ret;
374
375 trace_nbd_negotiate_send_rep_list(name, desc);
376 name_len = strlen(name);
377 desc_len = strlen(desc);
378 len = name_len + desc_len + sizeof(len);
379 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
380 if (ret < 0) {
381 return ret;
382 }
383
384 len = cpu_to_be32(name_len);
385 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
386 error_prepend(errp, "write failed (name length): ");
387 return -EINVAL;
388 }
389
390 if (nbd_write(ioc, name, name_len, errp) < 0) {
391 error_prepend(errp, "write failed (name buffer): ");
392 return -EINVAL;
393 }
394
395 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
396 error_prepend(errp, "write failed (description buffer): ");
397 return -EINVAL;
398 }
399
400 return 0;
401 }
402
403 /* Process the NBD_OPT_LIST command, with a potential series of replies.
404 * Return -errno on error, 0 on success. */
405 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
406 {
407 NBDExport *exp;
408 assert(client->opt == NBD_OPT_LIST);
409
410 /* For each export, send a NBD_REP_SERVER reply. */
411 QTAILQ_FOREACH(exp, &exports, next) {
412 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
413 return -EINVAL;
414 }
415 }
416 /* Finish with a NBD_REP_ACK. */
417 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
418 }
419
420 static void nbd_check_meta_export(NBDClient *client)
421 {
422 client->export_meta.valid &= client->exp == client->export_meta.exp;
423 }
424
425 /* Send a reply to NBD_OPT_EXPORT_NAME.
426 * Return -errno on error, 0 on success. */
427 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
428 Error **errp)
429 {
430 char name[NBD_MAX_NAME_SIZE + 1];
431 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
432 size_t len;
433 int ret;
434 uint16_t myflags;
435
436 /* Client sends:
437 [20 .. xx] export name (length bytes)
438 Server replies:
439 [ 0 .. 7] size
440 [ 8 .. 9] export flags
441 [10 .. 133] reserved (0) [unless no_zeroes]
442 */
443 trace_nbd_negotiate_handle_export_name();
444 if (client->optlen >= sizeof(name)) {
445 error_setg(errp, "Bad length received");
446 return -EINVAL;
447 }
448 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
449 return -EIO;
450 }
451 name[client->optlen] = '\0';
452 client->optlen = 0;
453
454 trace_nbd_negotiate_handle_export_name_request(name);
455
456 client->exp = nbd_export_find(name);
457 if (!client->exp) {
458 error_setg(errp, "export not found");
459 return -EINVAL;
460 }
461
462 myflags = client->exp->nbdflags;
463 if (client->structured_reply) {
464 myflags |= NBD_FLAG_SEND_DF;
465 }
466 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
467 stq_be_p(buf, client->exp->size);
468 stw_be_p(buf + 8, myflags);
469 len = no_zeroes ? 10 : sizeof(buf);
470 ret = nbd_write(client->ioc, buf, len, errp);
471 if (ret < 0) {
472 error_prepend(errp, "write failed: ");
473 return ret;
474 }
475
476 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
477 nbd_export_get(client->exp);
478 nbd_check_meta_export(client);
479
480 return 0;
481 }
482
483 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
484 * The buffer does NOT include the info type prefix.
485 * Return -errno on error, 0 if ready to send more. */
486 static int nbd_negotiate_send_info(NBDClient *client,
487 uint16_t info, uint32_t length, void *buf,
488 Error **errp)
489 {
490 int rc;
491
492 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
493 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
494 sizeof(info) + length, errp);
495 if (rc < 0) {
496 return rc;
497 }
498 info = cpu_to_be16(info);
499 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
500 return -EIO;
501 }
502 if (nbd_write(client->ioc, buf, length, errp) < 0) {
503 return -EIO;
504 }
505 return 0;
506 }
507
508 /* nbd_reject_length: Handle any unexpected payload.
509 * @fatal requests that we quit talking to the client, even if we are able
510 * to successfully send an error reply.
511 * Return:
512 * -errno transmission error occurred or @fatal was requested, errp is set
513 * 0 error message successfully sent to client, errp is not set
514 */
515 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
516 {
517 int ret;
518
519 assert(client->optlen);
520 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
521 nbd_opt_lookup(client->opt));
522 if (fatal && !ret) {
523 error_setg(errp, "option '%s' has unexpected length",
524 nbd_opt_lookup(client->opt));
525 return -EINVAL;
526 }
527 return ret;
528 }
529
530 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
531 * Return -errno on error, 0 if ready for next option, and 1 to move
532 * into transmission phase. */
533 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
534 {
535 int rc;
536 char name[NBD_MAX_NAME_SIZE + 1];
537 NBDExport *exp;
538 uint16_t requests;
539 uint16_t request;
540 uint32_t namelen;
541 bool sendname = false;
542 bool blocksize = false;
543 uint32_t sizes[3];
544 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
545 uint32_t check_align = 0;
546 uint16_t myflags;
547
548 /* Client sends:
549 4 bytes: L, name length (can be 0)
550 L bytes: export name
551 2 bytes: N, number of requests (can be 0)
552 N * 2 bytes: N requests
553 */
554 rc = nbd_opt_read_name(client, name, &namelen, errp);
555 if (rc <= 0) {
556 return rc;
557 }
558 trace_nbd_negotiate_handle_export_name_request(name);
559
560 rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
561 if (rc <= 0) {
562 return rc;
563 }
564 requests = be16_to_cpu(requests);
565 trace_nbd_negotiate_handle_info_requests(requests);
566 while (requests--) {
567 rc = nbd_opt_read(client, &request, sizeof(request), errp);
568 if (rc <= 0) {
569 return rc;
570 }
571 request = be16_to_cpu(request);
572 trace_nbd_negotiate_handle_info_request(request,
573 nbd_info_lookup(request));
574 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
575 * everything else is either a request we don't know or
576 * something we send regardless of request */
577 switch (request) {
578 case NBD_INFO_NAME:
579 sendname = true;
580 break;
581 case NBD_INFO_BLOCK_SIZE:
582 blocksize = true;
583 break;
584 }
585 }
586 if (client->optlen) {
587 return nbd_reject_length(client, false, errp);
588 }
589
590 exp = nbd_export_find(name);
591 if (!exp) {
592 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
593 errp, "export '%s' not present",
594 name);
595 }
596
597 /* Don't bother sending NBD_INFO_NAME unless client requested it */
598 if (sendname) {
599 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
600 errp);
601 if (rc < 0) {
602 return rc;
603 }
604 }
605
606 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
607 * client request */
608 if (exp->description) {
609 size_t len = strlen(exp->description);
610
611 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
612 len, exp->description, errp);
613 if (rc < 0) {
614 return rc;
615 }
616 }
617
618 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
619 * according to whether the client requested it, and according to
620 * whether this is OPT_INFO or OPT_GO. */
621 /* minimum - 1 for back-compat, or actual if client will obey it. */
622 if (client->opt == NBD_OPT_INFO || blocksize) {
623 check_align = sizes[0] = blk_get_request_alignment(exp->blk);
624 } else {
625 sizes[0] = 1;
626 }
627 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
628 /* preferred - Hard-code to 4096 for now.
629 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
630 sizes[1] = MAX(4096, sizes[0]);
631 /* maximum - At most 32M, but smaller as appropriate. */
632 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
633 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
634 sizes[0] = cpu_to_be32(sizes[0]);
635 sizes[1] = cpu_to_be32(sizes[1]);
636 sizes[2] = cpu_to_be32(sizes[2]);
637 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
638 sizeof(sizes), sizes, errp);
639 if (rc < 0) {
640 return rc;
641 }
642
643 /* Send NBD_INFO_EXPORT always */
644 myflags = exp->nbdflags;
645 if (client->structured_reply) {
646 myflags |= NBD_FLAG_SEND_DF;
647 }
648 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
649 stq_be_p(buf, exp->size);
650 stw_be_p(buf + 8, myflags);
651 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
652 sizeof(buf), buf, errp);
653 if (rc < 0) {
654 return rc;
655 }
656
657 /*
658 * If the client is just asking for NBD_OPT_INFO, but forgot to
659 * request block sizes in a situation that would impact
660 * performance, then return an error. But for NBD_OPT_GO, we
661 * tolerate all clients, regardless of alignments.
662 */
663 if (client->opt == NBD_OPT_INFO && !blocksize &&
664 blk_get_request_alignment(exp->blk) > 1) {
665 return nbd_negotiate_send_rep_err(client,
666 NBD_REP_ERR_BLOCK_SIZE_REQD,
667 errp,
668 "request NBD_INFO_BLOCK_SIZE to "
669 "use this export");
670 }
671
672 /* Final reply */
673 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
674 if (rc < 0) {
675 return rc;
676 }
677
678 if (client->opt == NBD_OPT_GO) {
679 client->exp = exp;
680 client->check_align = check_align;
681 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
682 nbd_export_get(client->exp);
683 nbd_check_meta_export(client);
684 rc = 1;
685 }
686 return rc;
687 }
688
689
690 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
691 * new channel for all further (now-encrypted) communication. */
692 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
693 Error **errp)
694 {
695 QIOChannel *ioc;
696 QIOChannelTLS *tioc;
697 struct NBDTLSHandshakeData data = { 0 };
698
699 assert(client->opt == NBD_OPT_STARTTLS);
700
701 trace_nbd_negotiate_handle_starttls();
702 ioc = client->ioc;
703
704 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
705 return NULL;
706 }
707
708 tioc = qio_channel_tls_new_server(ioc,
709 client->tlscreds,
710 client->tlsauthz,
711 errp);
712 if (!tioc) {
713 return NULL;
714 }
715
716 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
717 trace_nbd_negotiate_handle_starttls_handshake();
718 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
719 qio_channel_tls_handshake(tioc,
720 nbd_tls_handshake,
721 &data,
722 NULL,
723 NULL);
724
725 if (!data.complete) {
726 g_main_loop_run(data.loop);
727 }
728 g_main_loop_unref(data.loop);
729 if (data.error) {
730 object_unref(OBJECT(tioc));
731 error_propagate(errp, data.error);
732 return NULL;
733 }
734
735 return QIO_CHANNEL(tioc);
736 }
737
738 /* nbd_negotiate_send_meta_context
739 *
740 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
741 *
742 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
743 */
744 static int nbd_negotiate_send_meta_context(NBDClient *client,
745 const char *context,
746 uint32_t context_id,
747 Error **errp)
748 {
749 NBDOptionReplyMetaContext opt;
750 struct iovec iov[] = {
751 {.iov_base = &opt, .iov_len = sizeof(opt)},
752 {.iov_base = (void *)context, .iov_len = strlen(context)}
753 };
754
755 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
756 context_id = 0;
757 }
758
759 trace_nbd_negotiate_meta_query_reply(context, context_id);
760 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
761 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
762 stl_be_p(&opt.context_id, context_id);
763
764 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
765 }
766
767 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
768 * @match is never set to false.
769 *
770 * Return -errno on I/O error, 0 if option was completely handled by
771 * sending a reply about inconsistent lengths, or 1 on success.
772 *
773 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
774 * It only means that there are no errors.
775 */
776 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
777 Error **errp)
778 {
779 int ret;
780 char *query;
781 size_t len = strlen(pattern);
782
783 assert(len);
784
785 query = g_malloc(len);
786 ret = nbd_opt_read(client, query, len, errp);
787 if (ret <= 0) {
788 g_free(query);
789 return ret;
790 }
791
792 if (strncmp(query, pattern, len) == 0) {
793 trace_nbd_negotiate_meta_query_parse(pattern);
794 *match = true;
795 } else {
796 trace_nbd_negotiate_meta_query_skip("pattern not matched");
797 }
798 g_free(query);
799
800 return 1;
801 }
802
803 /*
804 * Read @len bytes, and set @match to true if they match @pattern, or if @len
805 * is 0 and the client is performing _LIST_. @match is never set to false.
806 *
807 * Return -errno on I/O error, 0 if option was completely handled by
808 * sending a reply about inconsistent lengths, or 1 on success.
809 *
810 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
811 * It only means that there are no errors.
812 */
813 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
814 uint32_t len, bool *match, Error **errp)
815 {
816 if (len == 0) {
817 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
818 *match = true;
819 }
820 trace_nbd_negotiate_meta_query_parse("empty");
821 return 1;
822 }
823
824 if (len != strlen(pattern)) {
825 trace_nbd_negotiate_meta_query_skip("different lengths");
826 return nbd_opt_skip(client, len, errp);
827 }
828
829 return nbd_meta_pattern(client, pattern, match, errp);
830 }
831
832 /* nbd_meta_base_query
833 *
834 * Handle queries to 'base' namespace. For now, only the base:allocation
835 * context is available. 'len' is the amount of text remaining to be read from
836 * the current name, after the 'base:' portion has been stripped.
837 *
838 * Return -errno on I/O error, 0 if option was completely handled by
839 * sending a reply about inconsistent lengths, or 1 on success.
840 */
841 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
842 uint32_t len, Error **errp)
843 {
844 return nbd_meta_empty_or_pattern(client, "allocation", len,
845 &meta->base_allocation, errp);
846 }
847
848 /* nbd_meta_bitmap_query
849 *
850 * Handle query to 'qemu:' namespace.
851 * @len is the amount of text remaining to be read from the current name, after
852 * the 'qemu:' portion has been stripped.
853 *
854 * Return -errno on I/O error, 0 if option was completely handled by
855 * sending a reply about inconsistent lengths, or 1 on success. */
856 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
857 uint32_t len, Error **errp)
858 {
859 bool dirty_bitmap = false;
860 size_t dirty_bitmap_len = strlen("dirty-bitmap:");
861 int ret;
862
863 if (!meta->exp->export_bitmap) {
864 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
865 return nbd_opt_skip(client, len, errp);
866 }
867
868 if (len == 0) {
869 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
870 meta->bitmap = true;
871 }
872 trace_nbd_negotiate_meta_query_parse("empty");
873 return 1;
874 }
875
876 if (len < dirty_bitmap_len) {
877 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
878 return nbd_opt_skip(client, len, errp);
879 }
880
881 len -= dirty_bitmap_len;
882 ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
883 if (ret <= 0) {
884 return ret;
885 }
886 if (!dirty_bitmap) {
887 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
888 return nbd_opt_skip(client, len, errp);
889 }
890
891 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
892
893 return nbd_meta_empty_or_pattern(
894 client, meta->exp->export_bitmap_context +
895 strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
896 }
897
898 /* nbd_negotiate_meta_query
899 *
900 * Parse namespace name and call corresponding function to parse body of the
901 * query.
902 *
903 * The only supported namespace now is 'base'.
904 *
905 * The function aims not wasting time and memory to read long unknown namespace
906 * names.
907 *
908 * Return -errno on I/O error, 0 if option was completely handled by
909 * sending a reply about inconsistent lengths, or 1 on success. */
910 static int nbd_negotiate_meta_query(NBDClient *client,
911 NBDExportMetaContexts *meta, Error **errp)
912 {
913 /*
914 * Both 'qemu' and 'base' namespaces have length = 5 including a
915 * colon. If another length namespace is later introduced, this
916 * should certainly be refactored.
917 */
918 int ret;
919 size_t ns_len = 5;
920 char ns[5];
921 uint32_t len;
922
923 ret = nbd_opt_read(client, &len, sizeof(len), errp);
924 if (ret <= 0) {
925 return ret;
926 }
927 len = cpu_to_be32(len);
928
929 if (len < ns_len) {
930 trace_nbd_negotiate_meta_query_skip("length too short");
931 return nbd_opt_skip(client, len, errp);
932 }
933
934 len -= ns_len;
935 ret = nbd_opt_read(client, ns, ns_len, errp);
936 if (ret <= 0) {
937 return ret;
938 }
939
940 if (!strncmp(ns, "base:", ns_len)) {
941 trace_nbd_negotiate_meta_query_parse("base:");
942 return nbd_meta_base_query(client, meta, len, errp);
943 } else if (!strncmp(ns, "qemu:", ns_len)) {
944 trace_nbd_negotiate_meta_query_parse("qemu:");
945 return nbd_meta_qemu_query(client, meta, len, errp);
946 }
947
948 trace_nbd_negotiate_meta_query_skip("unknown namespace");
949 return nbd_opt_skip(client, len, errp);
950 }
951
952 /* nbd_negotiate_meta_queries
953 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
954 *
955 * Return -errno on I/O error, or 0 if option was completely handled. */
956 static int nbd_negotiate_meta_queries(NBDClient *client,
957 NBDExportMetaContexts *meta, Error **errp)
958 {
959 int ret;
960 char export_name[NBD_MAX_NAME_SIZE + 1];
961 NBDExportMetaContexts local_meta;
962 uint32_t nb_queries;
963 int i;
964
965 if (!client->structured_reply) {
966 return nbd_opt_invalid(client, errp,
967 "request option '%s' when structured reply "
968 "is not negotiated",
969 nbd_opt_lookup(client->opt));
970 }
971
972 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
973 /* Only change the caller's meta on SET. */
974 meta = &local_meta;
975 }
976
977 memset(meta, 0, sizeof(*meta));
978
979 ret = nbd_opt_read_name(client, export_name, NULL, errp);
980 if (ret <= 0) {
981 return ret;
982 }
983
984 meta->exp = nbd_export_find(export_name);
985 if (meta->exp == NULL) {
986 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
987 "export '%s' not present", export_name);
988 }
989
990 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
991 if (ret <= 0) {
992 return ret;
993 }
994 nb_queries = cpu_to_be32(nb_queries);
995 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
996 export_name, nb_queries);
997
998 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
999 /* enable all known contexts */
1000 meta->base_allocation = true;
1001 meta->bitmap = !!meta->exp->export_bitmap;
1002 } else {
1003 for (i = 0; i < nb_queries; ++i) {
1004 ret = nbd_negotiate_meta_query(client, meta, errp);
1005 if (ret <= 0) {
1006 return ret;
1007 }
1008 }
1009 }
1010
1011 if (meta->base_allocation) {
1012 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1013 NBD_META_ID_BASE_ALLOCATION,
1014 errp);
1015 if (ret < 0) {
1016 return ret;
1017 }
1018 }
1019
1020 if (meta->bitmap) {
1021 ret = nbd_negotiate_send_meta_context(client,
1022 meta->exp->export_bitmap_context,
1023 NBD_META_ID_DIRTY_BITMAP,
1024 errp);
1025 if (ret < 0) {
1026 return ret;
1027 }
1028 }
1029
1030 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1031 if (ret == 0) {
1032 meta->valid = true;
1033 }
1034
1035 return ret;
1036 }
1037
1038 /* nbd_negotiate_options
1039 * Process all NBD_OPT_* client option commands, during fixed newstyle
1040 * negotiation.
1041 * Return:
1042 * -errno on error, errp is set
1043 * 0 on successful negotiation, errp is not set
1044 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1045 * errp is not set
1046 */
1047 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1048 {
1049 uint32_t flags;
1050 bool fixedNewstyle = false;
1051 bool no_zeroes = false;
1052
1053 /* Client sends:
1054 [ 0 .. 3] client flags
1055
1056 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1057 [ 0 .. 7] NBD_OPTS_MAGIC
1058 [ 8 .. 11] NBD option
1059 [12 .. 15] Data length
1060 ... Rest of request
1061
1062 [ 0 .. 7] NBD_OPTS_MAGIC
1063 [ 8 .. 11] Second NBD option
1064 [12 .. 15] Data length
1065 ... Rest of request
1066 */
1067
1068 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1069 return -EIO;
1070 }
1071 trace_nbd_negotiate_options_flags(flags);
1072 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1073 fixedNewstyle = true;
1074 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1075 }
1076 if (flags & NBD_FLAG_C_NO_ZEROES) {
1077 no_zeroes = true;
1078 flags &= ~NBD_FLAG_C_NO_ZEROES;
1079 }
1080 if (flags != 0) {
1081 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1082 return -EINVAL;
1083 }
1084
1085 while (1) {
1086 int ret;
1087 uint32_t option, length;
1088 uint64_t magic;
1089
1090 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1091 return -EINVAL;
1092 }
1093 trace_nbd_negotiate_options_check_magic(magic);
1094 if (magic != NBD_OPTS_MAGIC) {
1095 error_setg(errp, "Bad magic received");
1096 return -EINVAL;
1097 }
1098
1099 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1100 return -EINVAL;
1101 }
1102 client->opt = option;
1103
1104 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1105 return -EINVAL;
1106 }
1107 assert(!client->optlen);
1108 client->optlen = length;
1109
1110 if (length > NBD_MAX_BUFFER_SIZE) {
1111 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1112 length, NBD_MAX_BUFFER_SIZE);
1113 return -EINVAL;
1114 }
1115
1116 trace_nbd_negotiate_options_check_option(option,
1117 nbd_opt_lookup(option));
1118 if (client->tlscreds &&
1119 client->ioc == (QIOChannel *)client->sioc) {
1120 QIOChannel *tioc;
1121 if (!fixedNewstyle) {
1122 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1123 return -EINVAL;
1124 }
1125 switch (option) {
1126 case NBD_OPT_STARTTLS:
1127 if (length) {
1128 /* Unconditionally drop the connection if the client
1129 * can't start a TLS negotiation correctly */
1130 return nbd_reject_length(client, true, errp);
1131 }
1132 tioc = nbd_negotiate_handle_starttls(client, errp);
1133 if (!tioc) {
1134 return -EIO;
1135 }
1136 ret = 0;
1137 object_unref(OBJECT(client->ioc));
1138 client->ioc = QIO_CHANNEL(tioc);
1139 break;
1140
1141 case NBD_OPT_EXPORT_NAME:
1142 /* No way to return an error to client, so drop connection */
1143 error_setg(errp, "Option 0x%x not permitted before TLS",
1144 option);
1145 return -EINVAL;
1146
1147 default:
1148 /* Let the client keep trying, unless they asked to
1149 * quit. Always try to give an error back to the
1150 * client; but when replying to OPT_ABORT, be aware
1151 * that the client may hang up before receiving the
1152 * error, in which case we are fine ignoring the
1153 * resulting EPIPE. */
1154 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1155 option == NBD_OPT_ABORT ? NULL : errp,
1156 "Option 0x%" PRIx32
1157 " not permitted before TLS", option);
1158 if (option == NBD_OPT_ABORT) {
1159 return 1;
1160 }
1161 break;
1162 }
1163 } else if (fixedNewstyle) {
1164 switch (option) {
1165 case NBD_OPT_LIST:
1166 if (length) {
1167 ret = nbd_reject_length(client, false, errp);
1168 } else {
1169 ret = nbd_negotiate_handle_list(client, errp);
1170 }
1171 break;
1172
1173 case NBD_OPT_ABORT:
1174 /* NBD spec says we must try to reply before
1175 * disconnecting, but that we must also tolerate
1176 * guests that don't wait for our reply. */
1177 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1178 return 1;
1179
1180 case NBD_OPT_EXPORT_NAME:
1181 return nbd_negotiate_handle_export_name(client, no_zeroes,
1182 errp);
1183
1184 case NBD_OPT_INFO:
1185 case NBD_OPT_GO:
1186 ret = nbd_negotiate_handle_info(client, errp);
1187 if (ret == 1) {
1188 assert(option == NBD_OPT_GO);
1189 return 0;
1190 }
1191 break;
1192
1193 case NBD_OPT_STARTTLS:
1194 if (length) {
1195 ret = nbd_reject_length(client, false, errp);
1196 } else if (client->tlscreds) {
1197 ret = nbd_negotiate_send_rep_err(client,
1198 NBD_REP_ERR_INVALID, errp,
1199 "TLS already enabled");
1200 } else {
1201 ret = nbd_negotiate_send_rep_err(client,
1202 NBD_REP_ERR_POLICY, errp,
1203 "TLS not configured");
1204 }
1205 break;
1206
1207 case NBD_OPT_STRUCTURED_REPLY:
1208 if (length) {
1209 ret = nbd_reject_length(client, false, errp);
1210 } else if (client->structured_reply) {
1211 ret = nbd_negotiate_send_rep_err(
1212 client, NBD_REP_ERR_INVALID, errp,
1213 "structured reply already negotiated");
1214 } else {
1215 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1216 client->structured_reply = true;
1217 }
1218 break;
1219
1220 case NBD_OPT_LIST_META_CONTEXT:
1221 case NBD_OPT_SET_META_CONTEXT:
1222 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1223 errp);
1224 break;
1225
1226 default:
1227 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1228 "Unsupported option %" PRIu32 " (%s)",
1229 option, nbd_opt_lookup(option));
1230 break;
1231 }
1232 } else {
1233 /*
1234 * If broken new-style we should drop the connection
1235 * for anything except NBD_OPT_EXPORT_NAME
1236 */
1237 switch (option) {
1238 case NBD_OPT_EXPORT_NAME:
1239 return nbd_negotiate_handle_export_name(client, no_zeroes,
1240 errp);
1241
1242 default:
1243 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1244 option, nbd_opt_lookup(option));
1245 return -EINVAL;
1246 }
1247 }
1248 if (ret < 0) {
1249 return ret;
1250 }
1251 }
1252 }
1253
1254 /* nbd_negotiate
1255 * Return:
1256 * -errno on error, errp is set
1257 * 0 on successful negotiation, errp is not set
1258 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1259 * errp is not set
1260 */
1261 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1262 {
1263 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1264 int ret;
1265
1266 /* Old style negotiation header, no room for options
1267 [ 0 .. 7] passwd ("NBDMAGIC")
1268 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1269 [16 .. 23] size
1270 [24 .. 27] export flags (zero-extended)
1271 [28 .. 151] reserved (0)
1272
1273 New style negotiation header, client can send options
1274 [ 0 .. 7] passwd ("NBDMAGIC")
1275 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1276 [16 .. 17] server flags (0)
1277 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1278 */
1279
1280 qio_channel_set_blocking(client->ioc, false, NULL);
1281
1282 trace_nbd_negotiate_begin();
1283 memcpy(buf, "NBDMAGIC", 8);
1284
1285 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1286 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1287
1288 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1289 error_prepend(errp, "write failed: ");
1290 return -EINVAL;
1291 }
1292 ret = nbd_negotiate_options(client, errp);
1293 if (ret != 0) {
1294 if (ret < 0) {
1295 error_prepend(errp, "option negotiation failed: ");
1296 }
1297 return ret;
1298 }
1299
1300 assert(!client->optlen);
1301 trace_nbd_negotiate_success();
1302
1303 return 0;
1304 }
1305
1306 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1307 Error **errp)
1308 {
1309 uint8_t buf[NBD_REQUEST_SIZE];
1310 uint32_t magic;
1311 int ret;
1312
1313 ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1314 if (ret < 0) {
1315 return ret;
1316 }
1317
1318 /* Request
1319 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1320 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1321 [ 6 .. 7] type (NBD_CMD_READ, ...)
1322 [ 8 .. 15] handle
1323 [16 .. 23] from
1324 [24 .. 27] len
1325 */
1326
1327 magic = ldl_be_p(buf);
1328 request->flags = lduw_be_p(buf + 4);
1329 request->type = lduw_be_p(buf + 6);
1330 request->handle = ldq_be_p(buf + 8);
1331 request->from = ldq_be_p(buf + 16);
1332 request->len = ldl_be_p(buf + 24);
1333
1334 trace_nbd_receive_request(magic, request->flags, request->type,
1335 request->from, request->len);
1336
1337 if (magic != NBD_REQUEST_MAGIC) {
1338 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1339 return -EINVAL;
1340 }
1341 return 0;
1342 }
1343
1344 #define MAX_NBD_REQUESTS 16
1345
1346 void nbd_client_get(NBDClient *client)
1347 {
1348 client->refcount++;
1349 }
1350
1351 void nbd_client_put(NBDClient *client)
1352 {
1353 if (--client->refcount == 0) {
1354 /* The last reference should be dropped by client->close,
1355 * which is called by client_close.
1356 */
1357 assert(client->closing);
1358
1359 qio_channel_detach_aio_context(client->ioc);
1360 object_unref(OBJECT(client->sioc));
1361 object_unref(OBJECT(client->ioc));
1362 if (client->tlscreds) {
1363 object_unref(OBJECT(client->tlscreds));
1364 }
1365 g_free(client->tlsauthz);
1366 if (client->exp) {
1367 QTAILQ_REMOVE(&client->exp->clients, client, next);
1368 nbd_export_put(client->exp);
1369 }
1370 g_free(client);
1371 }
1372 }
1373
1374 static void client_close(NBDClient *client, bool negotiated)
1375 {
1376 if (client->closing) {
1377 return;
1378 }
1379
1380 client->closing = true;
1381
1382 /* Force requests to finish. They will drop their own references,
1383 * then we'll close the socket and free the NBDClient.
1384 */
1385 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1386 NULL);
1387
1388 /* Also tell the client, so that they release their reference. */
1389 if (client->close_fn) {
1390 client->close_fn(client, negotiated);
1391 }
1392 }
1393
1394 static NBDRequestData *nbd_request_get(NBDClient *client)
1395 {
1396 NBDRequestData *req;
1397
1398 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1399 client->nb_requests++;
1400
1401 req = g_new0(NBDRequestData, 1);
1402 nbd_client_get(client);
1403 req->client = client;
1404 return req;
1405 }
1406
1407 static void nbd_request_put(NBDRequestData *req)
1408 {
1409 NBDClient *client = req->client;
1410
1411 if (req->data) {
1412 qemu_vfree(req->data);
1413 }
1414 g_free(req);
1415
1416 client->nb_requests--;
1417 nbd_client_receive_next_request(client);
1418
1419 nbd_client_put(client);
1420 }
1421
1422 static void blk_aio_attached(AioContext *ctx, void *opaque)
1423 {
1424 NBDExport *exp = opaque;
1425 NBDClient *client;
1426
1427 trace_nbd_blk_aio_attached(exp->name, ctx);
1428
1429 exp->ctx = ctx;
1430
1431 QTAILQ_FOREACH(client, &exp->clients, next) {
1432 qio_channel_attach_aio_context(client->ioc, ctx);
1433 if (client->recv_coroutine) {
1434 aio_co_schedule(ctx, client->recv_coroutine);
1435 }
1436 if (client->send_coroutine) {
1437 aio_co_schedule(ctx, client->send_coroutine);
1438 }
1439 }
1440 }
1441
1442 static void blk_aio_detach(void *opaque)
1443 {
1444 NBDExport *exp = opaque;
1445 NBDClient *client;
1446
1447 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1448
1449 QTAILQ_FOREACH(client, &exp->clients, next) {
1450 qio_channel_detach_aio_context(client->ioc);
1451 }
1452
1453 exp->ctx = NULL;
1454 }
1455
1456 static void nbd_eject_notifier(Notifier *n, void *data)
1457 {
1458 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1459 nbd_export_close(exp);
1460 }
1461
1462 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1463 uint64_t size, const char *name, const char *desc,
1464 const char *bitmap, bool readonly, bool shared,
1465 void (*close)(NBDExport *), bool writethrough,
1466 BlockBackend *on_eject_blk, Error **errp)
1467 {
1468 AioContext *ctx;
1469 BlockBackend *blk;
1470 NBDExport *exp = g_new0(NBDExport, 1);
1471 uint64_t perm;
1472 int ret;
1473
1474 /*
1475 * NBD exports are used for non-shared storage migration. Make sure
1476 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1477 * access since the export could be available before migration handover.
1478 */
1479 assert(name);
1480 ctx = bdrv_get_aio_context(bs);
1481 aio_context_acquire(ctx);
1482 bdrv_invalidate_cache(bs, NULL);
1483 aio_context_release(ctx);
1484
1485 /* Don't allow resize while the NBD server is running, otherwise we don't
1486 * care what happens with the node. */
1487 perm = BLK_PERM_CONSISTENT_READ;
1488 if (!readonly) {
1489 perm |= BLK_PERM_WRITE;
1490 }
1491 blk = blk_new(bdrv_get_aio_context(bs), perm,
1492 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1493 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1494 ret = blk_insert_bs(blk, bs, errp);
1495 if (ret < 0) {
1496 goto fail;
1497 }
1498 blk_set_enable_write_cache(blk, !writethrough);
1499 blk_set_allow_aio_context_change(blk, true);
1500
1501 exp->refcount = 1;
1502 QTAILQ_INIT(&exp->clients);
1503 exp->blk = blk;
1504 assert(dev_offset <= INT64_MAX);
1505 exp->dev_offset = dev_offset;
1506 exp->name = g_strdup(name);
1507 exp->description = g_strdup(desc);
1508 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1509 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1510 if (readonly) {
1511 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1512 if (shared) {
1513 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1514 }
1515 } else {
1516 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1517 NBD_FLAG_SEND_FAST_ZERO);
1518 }
1519 assert(size <= INT64_MAX - dev_offset);
1520 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1521
1522 if (bitmap) {
1523 BdrvDirtyBitmap *bm = NULL;
1524
1525 while (true) {
1526 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1527 if (bm != NULL || bs->backing == NULL) {
1528 break;
1529 }
1530
1531 bs = bs->backing->bs;
1532 }
1533
1534 if (bm == NULL) {
1535 error_setg(errp, "Bitmap '%s' is not found", bitmap);
1536 goto fail;
1537 }
1538
1539 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1540 goto fail;
1541 }
1542
1543 if (readonly && bdrv_is_writable(bs) &&
1544 bdrv_dirty_bitmap_enabled(bm)) {
1545 error_setg(errp,
1546 "Enabled bitmap '%s' incompatible with readonly export",
1547 bitmap);
1548 goto fail;
1549 }
1550
1551 bdrv_dirty_bitmap_set_busy(bm, true);
1552 exp->export_bitmap = bm;
1553 exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1554 bitmap);
1555 }
1556
1557 exp->close = close;
1558 exp->ctx = blk_get_aio_context(blk);
1559 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1560
1561 if (on_eject_blk) {
1562 blk_ref(on_eject_blk);
1563 exp->eject_notifier_blk = on_eject_blk;
1564 exp->eject_notifier.notify = nbd_eject_notifier;
1565 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1566 }
1567 QTAILQ_INSERT_TAIL(&exports, exp, next);
1568 nbd_export_get(exp);
1569 return exp;
1570
1571 fail:
1572 blk_unref(blk);
1573 g_free(exp->name);
1574 g_free(exp->description);
1575 g_free(exp);
1576 return NULL;
1577 }
1578
1579 NBDExport *nbd_export_find(const char *name)
1580 {
1581 NBDExport *exp;
1582 QTAILQ_FOREACH(exp, &exports, next) {
1583 if (strcmp(name, exp->name) == 0) {
1584 return exp;
1585 }
1586 }
1587
1588 return NULL;
1589 }
1590
1591 void nbd_export_close(NBDExport *exp)
1592 {
1593 NBDClient *client, *next;
1594
1595 nbd_export_get(exp);
1596 /*
1597 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1598 * close mode that stops advertising the export to new clients but
1599 * still permits existing clients to run to completion? Because of
1600 * that possibility, nbd_export_close() can be called more than
1601 * once on an export.
1602 */
1603 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1604 client_close(client, true);
1605 }
1606 if (exp->name) {
1607 nbd_export_put(exp);
1608 g_free(exp->name);
1609 exp->name = NULL;
1610 QTAILQ_REMOVE(&exports, exp, next);
1611 }
1612 g_free(exp->description);
1613 exp->description = NULL;
1614 nbd_export_put(exp);
1615 }
1616
1617 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1618 {
1619 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1620 nbd_export_close(exp);
1621 return;
1622 }
1623
1624 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1625
1626 error_setg(errp, "export '%s' still in use", exp->name);
1627 error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1628 }
1629
1630 void nbd_export_get(NBDExport *exp)
1631 {
1632 assert(exp->refcount > 0);
1633 exp->refcount++;
1634 }
1635
1636 void nbd_export_put(NBDExport *exp)
1637 {
1638 assert(exp->refcount > 0);
1639 if (exp->refcount == 1) {
1640 nbd_export_close(exp);
1641 }
1642
1643 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1644 * if someone calls nbd_export_put() on named export not through
1645 * nbd_export_set_name() when refcount is 1. So, let's assert that
1646 * it is > 0.
1647 */
1648 assert(exp->refcount > 0);
1649 if (--exp->refcount == 0) {
1650 assert(exp->name == NULL);
1651 assert(exp->description == NULL);
1652
1653 if (exp->close) {
1654 exp->close(exp);
1655 }
1656
1657 if (exp->blk) {
1658 if (exp->eject_notifier_blk) {
1659 notifier_remove(&exp->eject_notifier);
1660 blk_unref(exp->eject_notifier_blk);
1661 }
1662 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1663 blk_aio_detach, exp);
1664 blk_unref(exp->blk);
1665 exp->blk = NULL;
1666 }
1667
1668 if (exp->export_bitmap) {
1669 bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1670 g_free(exp->export_bitmap_context);
1671 }
1672
1673 g_free(exp);
1674 }
1675 }
1676
1677 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1678 {
1679 return exp->blk;
1680 }
1681
1682 void nbd_export_close_all(void)
1683 {
1684 NBDExport *exp, *next;
1685
1686 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1687 nbd_export_close(exp);
1688 }
1689 }
1690
1691 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1692 unsigned niov, Error **errp)
1693 {
1694 int ret;
1695
1696 g_assert(qemu_in_coroutine());
1697 qemu_co_mutex_lock(&client->send_lock);
1698 client->send_coroutine = qemu_coroutine_self();
1699
1700 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1701
1702 client->send_coroutine = NULL;
1703 qemu_co_mutex_unlock(&client->send_lock);
1704
1705 return ret;
1706 }
1707
1708 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1709 uint64_t handle)
1710 {
1711 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1712 stl_be_p(&reply->error, error);
1713 stq_be_p(&reply->handle, handle);
1714 }
1715
1716 static int nbd_co_send_simple_reply(NBDClient *client,
1717 uint64_t handle,
1718 uint32_t error,
1719 void *data,
1720 size_t len,
1721 Error **errp)
1722 {
1723 NBDSimpleReply reply;
1724 int nbd_err = system_errno_to_nbd_errno(error);
1725 struct iovec iov[] = {
1726 {.iov_base = &reply, .iov_len = sizeof(reply)},
1727 {.iov_base = data, .iov_len = len}
1728 };
1729
1730 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1731 len);
1732 set_be_simple_reply(&reply, nbd_err, handle);
1733
1734 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1735 }
1736
1737 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1738 uint16_t type, uint64_t handle, uint32_t length)
1739 {
1740 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1741 stw_be_p(&chunk->flags, flags);
1742 stw_be_p(&chunk->type, type);
1743 stq_be_p(&chunk->handle, handle);
1744 stl_be_p(&chunk->length, length);
1745 }
1746
1747 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1748 uint64_t handle,
1749 Error **errp)
1750 {
1751 NBDStructuredReplyChunk chunk;
1752 struct iovec iov[] = {
1753 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1754 };
1755
1756 trace_nbd_co_send_structured_done(handle);
1757 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1758
1759 return nbd_co_send_iov(client, iov, 1, errp);
1760 }
1761
1762 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1763 uint64_t handle,
1764 uint64_t offset,
1765 void *data,
1766 size_t size,
1767 bool final,
1768 Error **errp)
1769 {
1770 NBDStructuredReadData chunk;
1771 struct iovec iov[] = {
1772 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1773 {.iov_base = data, .iov_len = size}
1774 };
1775
1776 assert(size);
1777 trace_nbd_co_send_structured_read(handle, offset, data, size);
1778 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1779 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1780 sizeof(chunk) - sizeof(chunk.h) + size);
1781 stq_be_p(&chunk.offset, offset);
1782
1783 return nbd_co_send_iov(client, iov, 2, errp);
1784 }
1785
1786 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1787 uint64_t handle,
1788 uint32_t error,
1789 const char *msg,
1790 Error **errp)
1791 {
1792 NBDStructuredError chunk;
1793 int nbd_err = system_errno_to_nbd_errno(error);
1794 struct iovec iov[] = {
1795 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1796 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1797 };
1798
1799 assert(nbd_err);
1800 trace_nbd_co_send_structured_error(handle, nbd_err,
1801 nbd_err_lookup(nbd_err), msg ? msg : "");
1802 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1803 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1804 stl_be_p(&chunk.error, nbd_err);
1805 stw_be_p(&chunk.message_length, iov[1].iov_len);
1806
1807 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1808 }
1809
1810 /* Do a sparse read and send the structured reply to the client.
1811 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1812 * reported to the client, at which point this function succeeds.
1813 */
1814 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1815 uint64_t handle,
1816 uint64_t offset,
1817 uint8_t *data,
1818 size_t size,
1819 Error **errp)
1820 {
1821 int ret = 0;
1822 NBDExport *exp = client->exp;
1823 size_t progress = 0;
1824
1825 while (progress < size) {
1826 int64_t pnum;
1827 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1828 offset + progress,
1829 size - progress, &pnum, NULL,
1830 NULL);
1831 bool final;
1832
1833 if (status < 0) {
1834 char *msg = g_strdup_printf("unable to check for holes: %s",
1835 strerror(-status));
1836
1837 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1838 errp);
1839 g_free(msg);
1840 return ret;
1841 }
1842 assert(pnum && pnum <= size - progress);
1843 final = progress + pnum == size;
1844 if (status & BDRV_BLOCK_ZERO) {
1845 NBDStructuredReadHole chunk;
1846 struct iovec iov[] = {
1847 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1848 };
1849
1850 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1851 pnum);
1852 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1853 NBD_REPLY_TYPE_OFFSET_HOLE,
1854 handle, sizeof(chunk) - sizeof(chunk.h));
1855 stq_be_p(&chunk.offset, offset + progress);
1856 stl_be_p(&chunk.length, pnum);
1857 ret = nbd_co_send_iov(client, iov, 1, errp);
1858 } else {
1859 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1860 data + progress, pnum);
1861 if (ret < 0) {
1862 error_setg_errno(errp, -ret, "reading from file failed");
1863 break;
1864 }
1865 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1866 data + progress, pnum, final,
1867 errp);
1868 }
1869
1870 if (ret < 0) {
1871 break;
1872 }
1873 progress += pnum;
1874 }
1875 return ret;
1876 }
1877
1878 /*
1879 * Populate @extents from block status. Update @bytes to be the actual
1880 * length encoded (which may be smaller than the original), and update
1881 * @nb_extents to the number of extents used.
1882 *
1883 * Returns zero on success and -errno on bdrv_block_status_above failure.
1884 */
1885 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1886 uint64_t *bytes, NBDExtent *extents,
1887 unsigned int *nb_extents)
1888 {
1889 uint64_t remaining_bytes = *bytes;
1890 NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1891 bool first_extent = true;
1892
1893 assert(*nb_extents);
1894 while (remaining_bytes) {
1895 uint32_t flags;
1896 int64_t num;
1897 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1898 &num, NULL, NULL);
1899
1900 if (ret < 0) {
1901 return ret;
1902 }
1903
1904 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1905 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1906
1907 if (first_extent) {
1908 extent->flags = flags;
1909 extent->length = num;
1910 first_extent = false;
1911 } else if (flags == extent->flags) {
1912 /* extend current extent */
1913 extent->length += num;
1914 } else {
1915 if (extent + 1 == extents_end) {
1916 break;
1917 }
1918
1919 /* start new extent */
1920 extent++;
1921 extent->flags = flags;
1922 extent->length = num;
1923 }
1924 offset += num;
1925 remaining_bytes -= num;
1926 }
1927
1928 extents_end = extent + 1;
1929
1930 for (extent = extents; extent < extents_end; extent++) {
1931 extent->flags = cpu_to_be32(extent->flags);
1932 extent->length = cpu_to_be32(extent->length);
1933 }
1934
1935 *bytes -= remaining_bytes;
1936 *nb_extents = extents_end - extents;
1937
1938 return 0;
1939 }
1940
1941 /* nbd_co_send_extents
1942 *
1943 * @length is only for tracing purposes (and may be smaller or larger
1944 * than the client's original request). @last controls whether
1945 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1946 * big-endian format.
1947 */
1948 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1949 NBDExtent *extents, unsigned int nb_extents,
1950 uint64_t length, bool last,
1951 uint32_t context_id, Error **errp)
1952 {
1953 NBDStructuredMeta chunk;
1954
1955 struct iovec iov[] = {
1956 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1957 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1958 };
1959
1960 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1961 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1962 NBD_REPLY_TYPE_BLOCK_STATUS,
1963 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1964 stl_be_p(&chunk.context_id, context_id);
1965
1966 return nbd_co_send_iov(client, iov, 2, errp);
1967 }
1968
1969 /* Get block status from the exported device and send it to the client */
1970 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1971 BlockDriverState *bs, uint64_t offset,
1972 uint32_t length, bool dont_fragment,
1973 bool last, uint32_t context_id,
1974 Error **errp)
1975 {
1976 int ret;
1977 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
1978 NBDExtent *extents = g_new(NBDExtent, nb_extents);
1979 uint64_t final_length = length;
1980
1981 ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1982 &nb_extents);
1983 if (ret < 0) {
1984 g_free(extents);
1985 return nbd_co_send_structured_error(
1986 client, handle, -ret, "can't get block status", errp);
1987 }
1988
1989 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1990 final_length, last, context_id, errp);
1991
1992 g_free(extents);
1993
1994 return ret;
1995 }
1996
1997 /*
1998 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1999 * final extent may exceed the original @length. Store in @length the
2000 * byte length encoded (which may be smaller or larger than the
2001 * original), and return the number of extents used.
2002 */
2003 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
2004 uint64_t *length, NBDExtent *extents,
2005 unsigned int nb_extents,
2006 bool dont_fragment)
2007 {
2008 uint64_t begin = offset, end = offset;
2009 uint64_t overall_end = offset + *length;
2010 unsigned int i = 0;
2011 BdrvDirtyBitmapIter *it;
2012 bool dirty;
2013
2014 bdrv_dirty_bitmap_lock(bitmap);
2015
2016 it = bdrv_dirty_iter_new(bitmap);
2017 dirty = bdrv_dirty_bitmap_get_locked(bitmap, offset);
2018
2019 assert(begin < overall_end && nb_extents);
2020 while (begin < overall_end && i < nb_extents) {
2021 bool next_dirty = !dirty;
2022
2023 if (dirty) {
2024 end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2025 } else {
2026 bdrv_set_dirty_iter(it, begin);
2027 end = bdrv_dirty_iter_next(it);
2028 }
2029 if (end == -1 || end - begin > UINT32_MAX) {
2030 /* Cap to an aligned value < 4G beyond begin. */
2031 end = MIN(bdrv_dirty_bitmap_size(bitmap),
2032 begin + UINT32_MAX + 1 -
2033 bdrv_dirty_bitmap_granularity(bitmap));
2034 next_dirty = dirty;
2035 }
2036 if (dont_fragment && end > overall_end) {
2037 end = overall_end;
2038 }
2039
2040 extents[i].length = cpu_to_be32(end - begin);
2041 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2042 i++;
2043 begin = end;
2044 dirty = next_dirty;
2045 }
2046
2047 bdrv_dirty_iter_free(it);
2048
2049 bdrv_dirty_bitmap_unlock(bitmap);
2050
2051 assert(offset < end);
2052 *length = end - offset;
2053 return i;
2054 }
2055
2056 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2057 BdrvDirtyBitmap *bitmap, uint64_t offset,
2058 uint32_t length, bool dont_fragment, bool last,
2059 uint32_t context_id, Error **errp)
2060 {
2061 int ret;
2062 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2063 NBDExtent *extents = g_new(NBDExtent, nb_extents);
2064 uint64_t final_length = length;
2065
2066 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2067 nb_extents, dont_fragment);
2068
2069 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2070 final_length, last, context_id, errp);
2071
2072 g_free(extents);
2073
2074 return ret;
2075 }
2076
2077 /* nbd_co_receive_request
2078 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2079 * connection right away, and any other negative value to report an error to
2080 * the client (although the caller may still need to disconnect after reporting
2081 * the error).
2082 */
2083 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2084 Error **errp)
2085 {
2086 NBDClient *client = req->client;
2087 int valid_flags;
2088
2089 g_assert(qemu_in_coroutine());
2090 assert(client->recv_coroutine == qemu_coroutine_self());
2091 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2092 return -EIO;
2093 }
2094
2095 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2096 nbd_cmd_lookup(request->type));
2097
2098 if (request->type != NBD_CMD_WRITE) {
2099 /* No payload, we are ready to read the next request. */
2100 req->complete = true;
2101 }
2102
2103 if (request->type == NBD_CMD_DISC) {
2104 /* Special case: we're going to disconnect without a reply,
2105 * whether or not flags, from, or len are bogus */
2106 return -EIO;
2107 }
2108
2109 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2110 request->type == NBD_CMD_CACHE)
2111 {
2112 if (request->len > NBD_MAX_BUFFER_SIZE) {
2113 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2114 request->len, NBD_MAX_BUFFER_SIZE);
2115 return -EINVAL;
2116 }
2117
2118 if (request->type != NBD_CMD_CACHE) {
2119 req->data = blk_try_blockalign(client->exp->blk, request->len);
2120 if (req->data == NULL) {
2121 error_setg(errp, "No memory");
2122 return -ENOMEM;
2123 }
2124 }
2125 }
2126
2127 if (request->type == NBD_CMD_WRITE) {
2128 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2129 errp) < 0)
2130 {
2131 return -EIO;
2132 }
2133 req->complete = true;
2134
2135 trace_nbd_co_receive_request_payload_received(request->handle,
2136 request->len);
2137 }
2138
2139 /* Sanity checks. */
2140 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2141 (request->type == NBD_CMD_WRITE ||
2142 request->type == NBD_CMD_WRITE_ZEROES ||
2143 request->type == NBD_CMD_TRIM)) {
2144 error_setg(errp, "Export is read-only");
2145 return -EROFS;
2146 }
2147 if (request->from > client->exp->size ||
2148 request->len > client->exp->size - request->from) {
2149 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2150 ", Size: %" PRIu64, request->from, request->len,
2151 client->exp->size);
2152 return (request->type == NBD_CMD_WRITE ||
2153 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2154 }
2155 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2156 client->check_align)) {
2157 /*
2158 * The block layer gracefully handles unaligned requests, but
2159 * it's still worth tracing client non-compliance
2160 */
2161 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2162 request->from,
2163 request->len,
2164 client->check_align);
2165 }
2166 valid_flags = NBD_CMD_FLAG_FUA;
2167 if (request->type == NBD_CMD_READ && client->structured_reply) {
2168 valid_flags |= NBD_CMD_FLAG_DF;
2169 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2170 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2171 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2172 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2173 }
2174 if (request->flags & ~valid_flags) {
2175 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2176 nbd_cmd_lookup(request->type), request->flags);
2177 return -EINVAL;
2178 }
2179
2180 return 0;
2181 }
2182
2183 /* Send simple reply without a payload, or a structured error
2184 * @error_msg is ignored if @ret >= 0
2185 * Returns 0 if connection is still live, -errno on failure to talk to client
2186 */
2187 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2188 uint64_t handle,
2189 int ret,
2190 const char *error_msg,
2191 Error **errp)
2192 {
2193 if (client->structured_reply && ret < 0) {
2194 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2195 errp);
2196 } else {
2197 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2198 NULL, 0, errp);
2199 }
2200 }
2201
2202 /* Handle NBD_CMD_READ request.
2203 * Return -errno if sending fails. Other errors are reported directly to the
2204 * client as an error reply. */
2205 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2206 uint8_t *data, Error **errp)
2207 {
2208 int ret;
2209 NBDExport *exp = client->exp;
2210
2211 assert(request->type == NBD_CMD_READ);
2212
2213 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2214 if (request->flags & NBD_CMD_FLAG_FUA) {
2215 ret = blk_co_flush(exp->blk);
2216 if (ret < 0) {
2217 return nbd_send_generic_reply(client, request->handle, ret,
2218 "flush failed", errp);
2219 }
2220 }
2221
2222 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2223 request->len)
2224 {
2225 return nbd_co_send_sparse_read(client, request->handle, request->from,
2226 data, request->len, errp);
2227 }
2228
2229 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2230 request->len);
2231 if (ret < 0) {
2232 return nbd_send_generic_reply(client, request->handle, ret,
2233 "reading from file failed", errp);
2234 }
2235
2236 if (client->structured_reply) {
2237 if (request->len) {
2238 return nbd_co_send_structured_read(client, request->handle,
2239 request->from, data,
2240 request->len, true, errp);
2241 } else {
2242 return nbd_co_send_structured_done(client, request->handle, errp);
2243 }
2244 } else {
2245 return nbd_co_send_simple_reply(client, request->handle, 0,
2246 data, request->len, errp);
2247 }
2248 }
2249
2250 /*
2251 * nbd_do_cmd_cache
2252 *
2253 * Handle NBD_CMD_CACHE request.
2254 * Return -errno if sending fails. Other errors are reported directly to the
2255 * client as an error reply.
2256 */
2257 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2258 Error **errp)
2259 {
2260 int ret;
2261 NBDExport *exp = client->exp;
2262
2263 assert(request->type == NBD_CMD_CACHE);
2264
2265 ret = blk_co_preadv(exp->blk, request->from + exp->dev_offset, request->len,
2266 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2267
2268 return nbd_send_generic_reply(client, request->handle, ret,
2269 "caching data failed", errp);
2270 }
2271
2272 /* Handle NBD request.
2273 * Return -errno if sending fails. Other errors are reported directly to the
2274 * client as an error reply. */
2275 static coroutine_fn int nbd_handle_request(NBDClient *client,
2276 NBDRequest *request,
2277 uint8_t *data, Error **errp)
2278 {
2279 int ret;
2280 int flags;
2281 NBDExport *exp = client->exp;
2282 char *msg;
2283
2284 switch (request->type) {
2285 case NBD_CMD_CACHE:
2286 return nbd_do_cmd_cache(client, request, errp);
2287
2288 case NBD_CMD_READ:
2289 return nbd_do_cmd_read(client, request, data, errp);
2290
2291 case NBD_CMD_WRITE:
2292 flags = 0;
2293 if (request->flags & NBD_CMD_FLAG_FUA) {
2294 flags |= BDRV_REQ_FUA;
2295 }
2296 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2297 data, request->len, flags);
2298 return nbd_send_generic_reply(client, request->handle, ret,
2299 "writing to file failed", errp);
2300
2301 case NBD_CMD_WRITE_ZEROES:
2302 flags = 0;
2303 if (request->flags & NBD_CMD_FLAG_FUA) {
2304 flags |= BDRV_REQ_FUA;
2305 }
2306 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2307 flags |= BDRV_REQ_MAY_UNMAP;
2308 }
2309 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2310 flags |= BDRV_REQ_NO_FALLBACK;
2311 }
2312 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2313 request->len, flags);
2314 return nbd_send_generic_reply(client, request->handle, ret,
2315 "writing to file failed", errp);
2316
2317 case NBD_CMD_DISC:
2318 /* unreachable, thanks to special case in nbd_co_receive_request() */
2319 abort();
2320
2321 case NBD_CMD_FLUSH:
2322 ret = blk_co_flush(exp->blk);
2323 return nbd_send_generic_reply(client, request->handle, ret,
2324 "flush failed", errp);
2325
2326 case NBD_CMD_TRIM:
2327 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2328 request->len);
2329 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2330 ret = blk_co_flush(exp->blk);
2331 }
2332 return nbd_send_generic_reply(client, request->handle, ret,
2333 "discard failed", errp);
2334
2335 case NBD_CMD_BLOCK_STATUS:
2336 if (!request->len) {
2337 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2338 "need non-zero length", errp);
2339 }
2340 if (client->export_meta.valid &&
2341 (client->export_meta.base_allocation ||
2342 client->export_meta.bitmap))
2343 {
2344 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2345
2346 if (client->export_meta.base_allocation) {
2347 ret = nbd_co_send_block_status(client, request->handle,
2348 blk_bs(exp->blk), request->from,
2349 request->len, dont_fragment,
2350 !client->export_meta.bitmap,
2351 NBD_META_ID_BASE_ALLOCATION,
2352 errp);
2353 if (ret < 0) {
2354 return ret;
2355 }
2356 }
2357
2358 if (client->export_meta.bitmap) {
2359 ret = nbd_co_send_bitmap(client, request->handle,
2360 client->exp->export_bitmap,
2361 request->from, request->len,
2362 dont_fragment,
2363 true, NBD_META_ID_DIRTY_BITMAP, errp);
2364 if (ret < 0) {
2365 return ret;
2366 }
2367 }
2368
2369 return ret;
2370 } else {
2371 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2372 "CMD_BLOCK_STATUS not negotiated",
2373 errp);
2374 }
2375
2376 default:
2377 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2378 request->type);
2379 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2380 errp);
2381 g_free(msg);
2382 return ret;
2383 }
2384 }
2385
2386 /* Owns a reference to the NBDClient passed as opaque. */
2387 static coroutine_fn void nbd_trip(void *opaque)
2388 {
2389 NBDClient *client = opaque;
2390 NBDRequestData *req;
2391 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2392 int ret;
2393 Error *local_err = NULL;
2394
2395 trace_nbd_trip();
2396 if (client->closing) {
2397 nbd_client_put(client);
2398 return;
2399 }
2400
2401 req = nbd_request_get(client);
2402 ret = nbd_co_receive_request(req, &request, &local_err);
2403 client->recv_coroutine = NULL;
2404
2405 if (client->closing) {
2406 /*
2407 * The client may be closed when we are blocked in
2408 * nbd_co_receive_request()
2409 */
2410 goto done;
2411 }
2412
2413 nbd_client_receive_next_request(client);
2414 if (ret == -EIO) {
2415 goto disconnect;
2416 }
2417
2418 if (ret < 0) {
2419 /* It wans't -EIO, so, according to nbd_co_receive_request()
2420 * semantics, we should return the error to the client. */
2421 Error *export_err = local_err;
2422
2423 local_err = NULL;
2424 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2425 error_get_pretty(export_err), &local_err);
2426 error_free(export_err);
2427 } else {
2428 ret = nbd_handle_request(client, &request, req->data, &local_err);
2429 }
2430 if (ret < 0) {
2431 error_prepend(&local_err, "Failed to send reply: ");
2432 goto disconnect;
2433 }
2434
2435 /* We must disconnect after NBD_CMD_WRITE if we did not
2436 * read the payload.
2437 */
2438 if (!req->complete) {
2439 error_setg(&local_err, "Request handling failed in intermediate state");
2440 goto disconnect;
2441 }
2442
2443 done:
2444 nbd_request_put(req);
2445 nbd_client_put(client);
2446 return;
2447
2448 disconnect:
2449 if (local_err) {
2450 error_reportf_err(local_err, "Disconnect client, due to: ");
2451 }
2452 nbd_request_put(req);
2453 client_close(client, true);
2454 nbd_client_put(client);
2455 }
2456
2457 static void nbd_client_receive_next_request(NBDClient *client)
2458 {
2459 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2460 nbd_client_get(client);
2461 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2462 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2463 }
2464 }
2465
2466 static coroutine_fn void nbd_co_client_start(void *opaque)
2467 {
2468 NBDClient *client = opaque;
2469 Error *local_err = NULL;
2470
2471 qemu_co_mutex_init(&client->send_lock);
2472
2473 if (nbd_negotiate(client, &local_err)) {
2474 if (local_err) {
2475 error_report_err(local_err);
2476 }
2477 client_close(client, false);
2478 return;
2479 }
2480
2481 nbd_client_receive_next_request(client);
2482 }
2483
2484 /*
2485 * Create a new client listener using the given channel @sioc.
2486 * Begin servicing it in a coroutine. When the connection closes, call
2487 * @close_fn with an indication of whether the client completed negotiation.
2488 */
2489 void nbd_client_new(QIOChannelSocket *sioc,
2490 QCryptoTLSCreds *tlscreds,
2491 const char *tlsauthz,
2492 void (*close_fn)(NBDClient *, bool))
2493 {
2494 NBDClient *client;
2495 Coroutine *co;
2496
2497 client = g_new0(NBDClient, 1);
2498 client->refcount = 1;
2499 client->tlscreds = tlscreds;
2500 if (tlscreds) {
2501 object_ref(OBJECT(client->tlscreds));
2502 }
2503 client->tlsauthz = g_strdup(tlsauthz);
2504 client->sioc = sioc;
2505 object_ref(OBJECT(client->sioc));
2506 client->ioc = QIO_CHANNEL(sioc);
2507 object_ref(OBJECT(client->ioc));
2508 client->close_fn = close_fn;
2509
2510 co = qemu_coroutine_create(nbd_co_client_start, client);
2511 qemu_coroutine_enter(co);
2512 }
AR7 emulation for QEMU