Merge remote-tracking branch 'remotes/ehabkost-gl/tags/x86-next-pull-request' into...
[qemu.git] / nbd / server.c
1 /*
2 * Copyright (C) 2016-2020 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
22 #include "block/export.h"
23 #include "qapi/error.h"
24 #include "qemu/queue.h"
25 #include "trace.h"
26 #include "nbd-internal.h"
27 #include "qemu/units.h"
28
29 #define NBD_META_ID_BASE_ALLOCATION 0
30 #define NBD_META_ID_ALLOCATION_DEPTH 1
31 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
32 #define NBD_META_ID_DIRTY_BITMAP 2
33
34 /*
35 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
36 * constant. If an increase is needed, note that the NBD protocol
37 * recommends no larger than 32 mb, so that the client won't consider
38 * the reply as a denial of service attack.
39 */
40 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
41
42 static int system_errno_to_nbd_errno(int err)
43 {
44 switch (err) {
45 case 0:
46 return NBD_SUCCESS;
47 case EPERM:
48 case EROFS:
49 return NBD_EPERM;
50 case EIO:
51 return NBD_EIO;
52 case ENOMEM:
53 return NBD_ENOMEM;
54 #ifdef EDQUOT
55 case EDQUOT:
56 #endif
57 case EFBIG:
58 case ENOSPC:
59 return NBD_ENOSPC;
60 case EOVERFLOW:
61 return NBD_EOVERFLOW;
62 case ENOTSUP:
63 #if ENOTSUP != EOPNOTSUPP
64 case EOPNOTSUPP:
65 #endif
66 return NBD_ENOTSUP;
67 case ESHUTDOWN:
68 return NBD_ESHUTDOWN;
69 case EINVAL:
70 default:
71 return NBD_EINVAL;
72 }
73 }
74
75 /* Definitions for opaque data types */
76
77 typedef struct NBDRequestData NBDRequestData;
78
79 struct NBDRequestData {
80 QSIMPLEQ_ENTRY(NBDRequestData) entry;
81 NBDClient *client;
82 uint8_t *data;
83 bool complete;
84 };
85
86 struct NBDExport {
87 BlockExport common;
88
89 char *name;
90 char *description;
91 uint64_t size;
92 uint16_t nbdflags;
93 QTAILQ_HEAD(, NBDClient) clients;
94 QTAILQ_ENTRY(NBDExport) next;
95
96 BlockBackend *eject_notifier_blk;
97 Notifier eject_notifier;
98
99 bool allocation_depth;
100 BdrvDirtyBitmap **export_bitmaps;
101 size_t nr_export_bitmaps;
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 size_t count; /* number of negotiated contexts */
112 bool base_allocation; /* export base:allocation context (block status) */
113 bool allocation_depth; /* export qemu:allocation-depth */
114 bool *bitmaps; /*
115 * export qemu:dirty-bitmap:<export bitmap name>,
116 * sized by exp->nr_export_bitmaps
117 */
118 } NBDExportMetaContexts;
119
120 struct NBDClient {
121 int refcount;
122 void (*close_fn)(NBDClient *client, bool negotiated);
123
124 NBDExport *exp;
125 QCryptoTLSCreds *tlscreds;
126 char *tlsauthz;
127 QIOChannelSocket *sioc; /* The underlying data channel */
128 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
129
130 Coroutine *recv_coroutine;
131
132 CoMutex send_lock;
133 Coroutine *send_coroutine;
134
135 bool read_yielding;
136 bool quiescing;
137
138 QTAILQ_ENTRY(NBDClient) next;
139 int nb_requests;
140 bool closing;
141
142 uint32_t check_align; /* If non-zero, check for aligned client requests */
143
144 bool structured_reply;
145 NBDExportMetaContexts export_meta;
146
147 uint32_t opt; /* Current option being negotiated */
148 uint32_t optlen; /* remaining length of data in ioc for the option being
149 negotiated now */
150 };
151
152 static void nbd_client_receive_next_request(NBDClient *client);
153
154 /* Basic flow for negotiation
155
156 Server Client
157 Negotiate
158
159 or
160
161 Server Client
162 Negotiate #1
163 Option
164 Negotiate #2
165
166 ----
167
168 followed by
169
170 Server Client
171 Request
172 Response
173 Request
174 Response
175 ...
176 ...
177 Request (type == 2)
178
179 */
180
181 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
182 uint32_t type, uint32_t length)
183 {
184 stq_be_p(&rep->magic, NBD_REP_MAGIC);
185 stl_be_p(&rep->option, option);
186 stl_be_p(&rep->type, type);
187 stl_be_p(&rep->length, length);
188 }
189
190 /* Send a reply header, including length, but no payload.
191 * Return -errno on error, 0 on success. */
192 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
193 uint32_t len, Error **errp)
194 {
195 NBDOptionReply rep;
196
197 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
198 type, nbd_rep_lookup(type), len);
199
200 assert(len < NBD_MAX_BUFFER_SIZE);
201
202 set_be_option_rep(&rep, client->opt, type, len);
203 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
204 }
205
206 /* Send a reply header with default 0 length.
207 * Return -errno on error, 0 on success. */
208 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
209 Error **errp)
210 {
211 return nbd_negotiate_send_rep_len(client, type, 0, errp);
212 }
213
214 /* Send an error reply.
215 * Return -errno on error, 0 on success. */
216 static int GCC_FMT_ATTR(4, 0)
217 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
218 Error **errp, const char *fmt, va_list va)
219 {
220 ERRP_GUARD();
221 g_autofree char *msg = NULL;
222 int ret;
223 size_t len;
224
225 msg = g_strdup_vprintf(fmt, va);
226 len = strlen(msg);
227 assert(len < NBD_MAX_STRING_SIZE);
228 trace_nbd_negotiate_send_rep_err(msg);
229 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
230 if (ret < 0) {
231 return ret;
232 }
233 if (nbd_write(client->ioc, msg, len, errp) < 0) {
234 error_prepend(errp, "write failed (error message): ");
235 return -EIO;
236 }
237
238 return 0;
239 }
240
241 /*
242 * Return a malloc'd copy of @name suitable for use in an error reply.
243 */
244 static char *
245 nbd_sanitize_name(const char *name)
246 {
247 if (strnlen(name, 80) < 80) {
248 return g_strdup(name);
249 }
250 /* XXX Should we also try to sanitize any control characters? */
251 return g_strdup_printf("%.80s...", name);
252 }
253
254 /* Send an error reply.
255 * Return -errno on error, 0 on success. */
256 static int GCC_FMT_ATTR(4, 5)
257 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
258 Error **errp, const char *fmt, ...)
259 {
260 va_list va;
261 int ret;
262
263 va_start(va, fmt);
264 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
265 va_end(va);
266 return ret;
267 }
268
269 /* Drop remainder of the current option, and send a reply with the
270 * given error type and message. Return -errno on read or write
271 * failure; or 0 if connection is still live. */
272 static int GCC_FMT_ATTR(4, 0)
273 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
274 const char *fmt, va_list va)
275 {
276 int ret = nbd_drop(client->ioc, client->optlen, errp);
277
278 client->optlen = 0;
279 if (!ret) {
280 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
281 }
282 return ret;
283 }
284
285 static int GCC_FMT_ATTR(4, 5)
286 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
287 const char *fmt, ...)
288 {
289 int ret;
290 va_list va;
291
292 va_start(va, fmt);
293 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
294 va_end(va);
295
296 return ret;
297 }
298
299 static int GCC_FMT_ATTR(3, 4)
300 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
301 {
302 int ret;
303 va_list va;
304
305 va_start(va, fmt);
306 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
307 va_end(va);
308
309 return ret;
310 }
311
312 /* Read size bytes from the unparsed payload of the current option.
313 * If @check_nul, require that no NUL bytes appear in buffer.
314 * Return -errno on I/O error, 0 if option was completely handled by
315 * sending a reply about inconsistent lengths, or 1 on success. */
316 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
317 bool check_nul, Error **errp)
318 {
319 if (size > client->optlen) {
320 return nbd_opt_invalid(client, errp,
321 "Inconsistent lengths in option %s",
322 nbd_opt_lookup(client->opt));
323 }
324 client->optlen -= size;
325 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
326 return -EIO;
327 }
328
329 if (check_nul && strnlen(buffer, size) != size) {
330 return nbd_opt_invalid(client, errp,
331 "Unexpected embedded NUL in option %s",
332 nbd_opt_lookup(client->opt));
333 }
334 return 1;
335 }
336
337 /* Drop size bytes from the unparsed payload of the current option.
338 * Return -errno on I/O error, 0 if option was completely handled by
339 * sending a reply about inconsistent lengths, or 1 on success. */
340 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
341 {
342 if (size > client->optlen) {
343 return nbd_opt_invalid(client, errp,
344 "Inconsistent lengths in option %s",
345 nbd_opt_lookup(client->opt));
346 }
347 client->optlen -= size;
348 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
349 }
350
351 /* nbd_opt_read_name
352 *
353 * Read a string with the format:
354 * uint32_t len (<= NBD_MAX_STRING_SIZE)
355 * len bytes string (not 0-terminated)
356 *
357 * On success, @name will be allocated.
358 * If @length is non-null, it will be set to the actual string length.
359 *
360 * Return -errno on I/O error, 0 if option was completely handled by
361 * sending a reply about inconsistent lengths, or 1 on success.
362 */
363 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
364 Error **errp)
365 {
366 int ret;
367 uint32_t len;
368 g_autofree char *local_name = NULL;
369
370 *name = NULL;
371 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
372 if (ret <= 0) {
373 return ret;
374 }
375 len = cpu_to_be32(len);
376
377 if (len > NBD_MAX_STRING_SIZE) {
378 return nbd_opt_invalid(client, errp,
379 "Invalid name length: %" PRIu32, len);
380 }
381
382 local_name = g_malloc(len + 1);
383 ret = nbd_opt_read(client, local_name, len, true, errp);
384 if (ret <= 0) {
385 return ret;
386 }
387 local_name[len] = '\0';
388
389 if (length) {
390 *length = len;
391 }
392 *name = g_steal_pointer(&local_name);
393
394 return 1;
395 }
396
397 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
398 * Return -errno on error, 0 on success. */
399 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
400 Error **errp)
401 {
402 ERRP_GUARD();
403 size_t name_len, desc_len;
404 uint32_t len;
405 const char *name = exp->name ? exp->name : "";
406 const char *desc = exp->description ? exp->description : "";
407 QIOChannel *ioc = client->ioc;
408 int ret;
409
410 trace_nbd_negotiate_send_rep_list(name, desc);
411 name_len = strlen(name);
412 desc_len = strlen(desc);
413 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
414 len = name_len + desc_len + sizeof(len);
415 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
416 if (ret < 0) {
417 return ret;
418 }
419
420 len = cpu_to_be32(name_len);
421 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
422 error_prepend(errp, "write failed (name length): ");
423 return -EINVAL;
424 }
425
426 if (nbd_write(ioc, name, name_len, errp) < 0) {
427 error_prepend(errp, "write failed (name buffer): ");
428 return -EINVAL;
429 }
430
431 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
432 error_prepend(errp, "write failed (description buffer): ");
433 return -EINVAL;
434 }
435
436 return 0;
437 }
438
439 /* Process the NBD_OPT_LIST command, with a potential series of replies.
440 * Return -errno on error, 0 on success. */
441 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
442 {
443 NBDExport *exp;
444 assert(client->opt == NBD_OPT_LIST);
445
446 /* For each export, send a NBD_REP_SERVER reply. */
447 QTAILQ_FOREACH(exp, &exports, next) {
448 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
449 return -EINVAL;
450 }
451 }
452 /* Finish with a NBD_REP_ACK. */
453 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
454 }
455
456 static void nbd_check_meta_export(NBDClient *client)
457 {
458 if (client->exp != client->export_meta.exp) {
459 client->export_meta.count = 0;
460 }
461 }
462
463 /* Send a reply to NBD_OPT_EXPORT_NAME.
464 * Return -errno on error, 0 on success. */
465 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
466 Error **errp)
467 {
468 ERRP_GUARD();
469 g_autofree char *name = NULL;
470 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
471 size_t len;
472 int ret;
473 uint16_t myflags;
474
475 /* Client sends:
476 [20 .. xx] export name (length bytes)
477 Server replies:
478 [ 0 .. 7] size
479 [ 8 .. 9] export flags
480 [10 .. 133] reserved (0) [unless no_zeroes]
481 */
482 trace_nbd_negotiate_handle_export_name();
483 if (client->optlen > NBD_MAX_STRING_SIZE) {
484 error_setg(errp, "Bad length received");
485 return -EINVAL;
486 }
487 name = g_malloc(client->optlen + 1);
488 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
489 return -EIO;
490 }
491 name[client->optlen] = '\0';
492 client->optlen = 0;
493
494 trace_nbd_negotiate_handle_export_name_request(name);
495
496 client->exp = nbd_export_find(name);
497 if (!client->exp) {
498 error_setg(errp, "export not found");
499 return -EINVAL;
500 }
501
502 myflags = client->exp->nbdflags;
503 if (client->structured_reply) {
504 myflags |= NBD_FLAG_SEND_DF;
505 }
506 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
507 stq_be_p(buf, client->exp->size);
508 stw_be_p(buf + 8, myflags);
509 len = no_zeroes ? 10 : sizeof(buf);
510 ret = nbd_write(client->ioc, buf, len, errp);
511 if (ret < 0) {
512 error_prepend(errp, "write failed: ");
513 return ret;
514 }
515
516 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
517 blk_exp_ref(&client->exp->common);
518 nbd_check_meta_export(client);
519
520 return 0;
521 }
522
523 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
524 * The buffer does NOT include the info type prefix.
525 * Return -errno on error, 0 if ready to send more. */
526 static int nbd_negotiate_send_info(NBDClient *client,
527 uint16_t info, uint32_t length, void *buf,
528 Error **errp)
529 {
530 int rc;
531
532 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
533 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
534 sizeof(info) + length, errp);
535 if (rc < 0) {
536 return rc;
537 }
538 info = cpu_to_be16(info);
539 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
540 return -EIO;
541 }
542 if (nbd_write(client->ioc, buf, length, errp) < 0) {
543 return -EIO;
544 }
545 return 0;
546 }
547
548 /* nbd_reject_length: Handle any unexpected payload.
549 * @fatal requests that we quit talking to the client, even if we are able
550 * to successfully send an error reply.
551 * Return:
552 * -errno transmission error occurred or @fatal was requested, errp is set
553 * 0 error message successfully sent to client, errp is not set
554 */
555 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
556 {
557 int ret;
558
559 assert(client->optlen);
560 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
561 nbd_opt_lookup(client->opt));
562 if (fatal && !ret) {
563 error_setg(errp, "option '%s' has unexpected length",
564 nbd_opt_lookup(client->opt));
565 return -EINVAL;
566 }
567 return ret;
568 }
569
570 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
571 * Return -errno on error, 0 if ready for next option, and 1 to move
572 * into transmission phase. */
573 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
574 {
575 int rc;
576 g_autofree char *name = NULL;
577 NBDExport *exp;
578 uint16_t requests;
579 uint16_t request;
580 uint32_t namelen = 0;
581 bool sendname = false;
582 bool blocksize = false;
583 uint32_t sizes[3];
584 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
585 uint32_t check_align = 0;
586 uint16_t myflags;
587
588 /* Client sends:
589 4 bytes: L, name length (can be 0)
590 L bytes: export name
591 2 bytes: N, number of requests (can be 0)
592 N * 2 bytes: N requests
593 */
594 rc = nbd_opt_read_name(client, &name, &namelen, errp);
595 if (rc <= 0) {
596 return rc;
597 }
598 trace_nbd_negotiate_handle_export_name_request(name);
599
600 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
601 if (rc <= 0) {
602 return rc;
603 }
604 requests = be16_to_cpu(requests);
605 trace_nbd_negotiate_handle_info_requests(requests);
606 while (requests--) {
607 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
608 if (rc <= 0) {
609 return rc;
610 }
611 request = be16_to_cpu(request);
612 trace_nbd_negotiate_handle_info_request(request,
613 nbd_info_lookup(request));
614 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
615 * everything else is either a request we don't know or
616 * something we send regardless of request */
617 switch (request) {
618 case NBD_INFO_NAME:
619 sendname = true;
620 break;
621 case NBD_INFO_BLOCK_SIZE:
622 blocksize = true;
623 break;
624 }
625 }
626 if (client->optlen) {
627 return nbd_reject_length(client, false, errp);
628 }
629
630 exp = nbd_export_find(name);
631 if (!exp) {
632 g_autofree char *sane_name = nbd_sanitize_name(name);
633
634 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
635 errp, "export '%s' not present",
636 sane_name);
637 }
638
639 /* Don't bother sending NBD_INFO_NAME unless client requested it */
640 if (sendname) {
641 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
642 errp);
643 if (rc < 0) {
644 return rc;
645 }
646 }
647
648 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
649 * client request */
650 if (exp->description) {
651 size_t len = strlen(exp->description);
652
653 assert(len <= NBD_MAX_STRING_SIZE);
654 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
655 len, exp->description, errp);
656 if (rc < 0) {
657 return rc;
658 }
659 }
660
661 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
662 * according to whether the client requested it, and according to
663 * whether this is OPT_INFO or OPT_GO. */
664 /* minimum - 1 for back-compat, or actual if client will obey it. */
665 if (client->opt == NBD_OPT_INFO || blocksize) {
666 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
667 } else {
668 sizes[0] = 1;
669 }
670 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
671 /* preferred - Hard-code to 4096 for now.
672 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
673 sizes[1] = MAX(4096, sizes[0]);
674 /* maximum - At most 32M, but smaller as appropriate. */
675 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
676 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
677 sizes[0] = cpu_to_be32(sizes[0]);
678 sizes[1] = cpu_to_be32(sizes[1]);
679 sizes[2] = cpu_to_be32(sizes[2]);
680 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
681 sizeof(sizes), sizes, errp);
682 if (rc < 0) {
683 return rc;
684 }
685
686 /* Send NBD_INFO_EXPORT always */
687 myflags = exp->nbdflags;
688 if (client->structured_reply) {
689 myflags |= NBD_FLAG_SEND_DF;
690 }
691 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
692 stq_be_p(buf, exp->size);
693 stw_be_p(buf + 8, myflags);
694 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
695 sizeof(buf), buf, errp);
696 if (rc < 0) {
697 return rc;
698 }
699
700 /*
701 * If the client is just asking for NBD_OPT_INFO, but forgot to
702 * request block sizes in a situation that would impact
703 * performance, then return an error. But for NBD_OPT_GO, we
704 * tolerate all clients, regardless of alignments.
705 */
706 if (client->opt == NBD_OPT_INFO && !blocksize &&
707 blk_get_request_alignment(exp->common.blk) > 1) {
708 return nbd_negotiate_send_rep_err(client,
709 NBD_REP_ERR_BLOCK_SIZE_REQD,
710 errp,
711 "request NBD_INFO_BLOCK_SIZE to "
712 "use this export");
713 }
714
715 /* Final reply */
716 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
717 if (rc < 0) {
718 return rc;
719 }
720
721 if (client->opt == NBD_OPT_GO) {
722 client->exp = exp;
723 client->check_align = check_align;
724 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
725 blk_exp_ref(&client->exp->common);
726 nbd_check_meta_export(client);
727 rc = 1;
728 }
729 return rc;
730 }
731
732
733 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
734 * new channel for all further (now-encrypted) communication. */
735 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
736 Error **errp)
737 {
738 QIOChannel *ioc;
739 QIOChannelTLS *tioc;
740 struct NBDTLSHandshakeData data = { 0 };
741
742 assert(client->opt == NBD_OPT_STARTTLS);
743
744 trace_nbd_negotiate_handle_starttls();
745 ioc = client->ioc;
746
747 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
748 return NULL;
749 }
750
751 tioc = qio_channel_tls_new_server(ioc,
752 client->tlscreds,
753 client->tlsauthz,
754 errp);
755 if (!tioc) {
756 return NULL;
757 }
758
759 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
760 trace_nbd_negotiate_handle_starttls_handshake();
761 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
762 qio_channel_tls_handshake(tioc,
763 nbd_tls_handshake,
764 &data,
765 NULL,
766 NULL);
767
768 if (!data.complete) {
769 g_main_loop_run(data.loop);
770 }
771 g_main_loop_unref(data.loop);
772 if (data.error) {
773 object_unref(OBJECT(tioc));
774 error_propagate(errp, data.error);
775 return NULL;
776 }
777
778 return QIO_CHANNEL(tioc);
779 }
780
781 /* nbd_negotiate_send_meta_context
782 *
783 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
784 *
785 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
786 */
787 static int nbd_negotiate_send_meta_context(NBDClient *client,
788 const char *context,
789 uint32_t context_id,
790 Error **errp)
791 {
792 NBDOptionReplyMetaContext opt;
793 struct iovec iov[] = {
794 {.iov_base = &opt, .iov_len = sizeof(opt)},
795 {.iov_base = (void *)context, .iov_len = strlen(context)}
796 };
797
798 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
799 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
800 context_id = 0;
801 }
802
803 trace_nbd_negotiate_meta_query_reply(context, context_id);
804 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
805 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
806 stl_be_p(&opt.context_id, context_id);
807
808 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
809 }
810
811 /*
812 * Return true if @query matches @pattern, or if @query is empty when
813 * the @client is performing _LIST_.
814 */
815 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
816 const char *query)
817 {
818 if (!*query) {
819 trace_nbd_negotiate_meta_query_parse("empty");
820 return client->opt == NBD_OPT_LIST_META_CONTEXT;
821 }
822 if (strcmp(query, pattern) == 0) {
823 trace_nbd_negotiate_meta_query_parse(pattern);
824 return true;
825 }
826 trace_nbd_negotiate_meta_query_skip("pattern not matched");
827 return false;
828 }
829
830 /*
831 * Return true and adjust @str in place if it begins with @prefix.
832 */
833 static bool nbd_strshift(const char **str, const char *prefix)
834 {
835 size_t len = strlen(prefix);
836
837 if (strncmp(*str, prefix, len) == 0) {
838 *str += len;
839 return true;
840 }
841 return false;
842 }
843
844 /* nbd_meta_base_query
845 *
846 * Handle queries to 'base' namespace. For now, only the base:allocation
847 * context is available. Return true if @query has been handled.
848 */
849 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
850 const char *query)
851 {
852 if (!nbd_strshift(&query, "base:")) {
853 return false;
854 }
855 trace_nbd_negotiate_meta_query_parse("base:");
856
857 if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
858 meta->base_allocation = true;
859 }
860 return true;
861 }
862
863 /* nbd_meta_qemu_query
864 *
865 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
866 * and qemu:allocation-depth contexts are available. Return true if @query
867 * has been handled.
868 */
869 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
870 const char *query)
871 {
872 size_t i;
873
874 if (!nbd_strshift(&query, "qemu:")) {
875 return false;
876 }
877 trace_nbd_negotiate_meta_query_parse("qemu:");
878
879 if (!*query) {
880 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
881 meta->allocation_depth = meta->exp->allocation_depth;
882 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
883 }
884 trace_nbd_negotiate_meta_query_parse("empty");
885 return true;
886 }
887
888 if (strcmp(query, "allocation-depth") == 0) {
889 trace_nbd_negotiate_meta_query_parse("allocation-depth");
890 meta->allocation_depth = meta->exp->allocation_depth;
891 return true;
892 }
893
894 if (nbd_strshift(&query, "dirty-bitmap:")) {
895 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
896 if (!*query) {
897 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
898 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
899 }
900 trace_nbd_negotiate_meta_query_parse("empty");
901 return true;
902 }
903
904 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
905 const char *bm_name;
906
907 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
908 if (strcmp(bm_name, query) == 0) {
909 meta->bitmaps[i] = true;
910 trace_nbd_negotiate_meta_query_parse(query);
911 return true;
912 }
913 }
914 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
915 return true;
916 }
917
918 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
919 return true;
920 }
921
922 /* nbd_negotiate_meta_query
923 *
924 * Parse namespace name and call corresponding function to parse body of the
925 * query.
926 *
927 * The only supported namespaces are 'base' and 'qemu'.
928 *
929 * Return -errno on I/O error, 0 if option was completely handled by
930 * sending a reply about inconsistent lengths, or 1 on success. */
931 static int nbd_negotiate_meta_query(NBDClient *client,
932 NBDExportMetaContexts *meta, Error **errp)
933 {
934 int ret;
935 g_autofree char *query = NULL;
936 uint32_t len;
937
938 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
939 if (ret <= 0) {
940 return ret;
941 }
942 len = cpu_to_be32(len);
943
944 if (len > NBD_MAX_STRING_SIZE) {
945 trace_nbd_negotiate_meta_query_skip("length too long");
946 return nbd_opt_skip(client, len, errp);
947 }
948
949 query = g_malloc(len + 1);
950 ret = nbd_opt_read(client, query, len, true, errp);
951 if (ret <= 0) {
952 return ret;
953 }
954 query[len] = '\0';
955
956 if (nbd_meta_base_query(client, meta, query)) {
957 return 1;
958 }
959 if (nbd_meta_qemu_query(client, meta, query)) {
960 return 1;
961 }
962
963 trace_nbd_negotiate_meta_query_skip("unknown namespace");
964 return 1;
965 }
966
967 /* nbd_negotiate_meta_queries
968 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
969 *
970 * Return -errno on I/O error, or 0 if option was completely handled. */
971 static int nbd_negotiate_meta_queries(NBDClient *client,
972 NBDExportMetaContexts *meta, Error **errp)
973 {
974 int ret;
975 g_autofree char *export_name = NULL;
976 g_autofree bool *bitmaps = NULL;
977 NBDExportMetaContexts local_meta = {0};
978 uint32_t nb_queries;
979 size_t i;
980 size_t count = 0;
981
982 if (!client->structured_reply) {
983 return nbd_opt_invalid(client, errp,
984 "request option '%s' when structured reply "
985 "is not negotiated",
986 nbd_opt_lookup(client->opt));
987 }
988
989 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
990 /* Only change the caller's meta on SET. */
991 meta = &local_meta;
992 }
993
994 g_free(meta->bitmaps);
995 memset(meta, 0, sizeof(*meta));
996
997 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
998 if (ret <= 0) {
999 return ret;
1000 }
1001
1002 meta->exp = nbd_export_find(export_name);
1003 if (meta->exp == NULL) {
1004 g_autofree char *sane_name = nbd_sanitize_name(export_name);
1005
1006 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1007 "export '%s' not present", sane_name);
1008 }
1009 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1010 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1011 bitmaps = meta->bitmaps;
1012 }
1013
1014 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1015 if (ret <= 0) {
1016 return ret;
1017 }
1018 nb_queries = cpu_to_be32(nb_queries);
1019 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1020 export_name, nb_queries);
1021
1022 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1023 /* enable all known contexts */
1024 meta->base_allocation = true;
1025 meta->allocation_depth = meta->exp->allocation_depth;
1026 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1027 } else {
1028 for (i = 0; i < nb_queries; ++i) {
1029 ret = nbd_negotiate_meta_query(client, meta, errp);
1030 if (ret <= 0) {
1031 return ret;
1032 }
1033 }
1034 }
1035
1036 if (meta->base_allocation) {
1037 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1038 NBD_META_ID_BASE_ALLOCATION,
1039 errp);
1040 if (ret < 0) {
1041 return ret;
1042 }
1043 count++;
1044 }
1045
1046 if (meta->allocation_depth) {
1047 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1048 NBD_META_ID_ALLOCATION_DEPTH,
1049 errp);
1050 if (ret < 0) {
1051 return ret;
1052 }
1053 count++;
1054 }
1055
1056 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1057 const char *bm_name;
1058 g_autofree char *context = NULL;
1059
1060 if (!meta->bitmaps[i]) {
1061 continue;
1062 }
1063
1064 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1065 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1066
1067 ret = nbd_negotiate_send_meta_context(client, context,
1068 NBD_META_ID_DIRTY_BITMAP + i,
1069 errp);
1070 if (ret < 0) {
1071 return ret;
1072 }
1073 count++;
1074 }
1075
1076 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1077 if (ret == 0) {
1078 meta->count = count;
1079 }
1080
1081 return ret;
1082 }
1083
1084 /* nbd_negotiate_options
1085 * Process all NBD_OPT_* client option commands, during fixed newstyle
1086 * negotiation.
1087 * Return:
1088 * -errno on error, errp is set
1089 * 0 on successful negotiation, errp is not set
1090 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1091 * errp is not set
1092 */
1093 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1094 {
1095 uint32_t flags;
1096 bool fixedNewstyle = false;
1097 bool no_zeroes = false;
1098
1099 /* Client sends:
1100 [ 0 .. 3] client flags
1101
1102 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1103 [ 0 .. 7] NBD_OPTS_MAGIC
1104 [ 8 .. 11] NBD option
1105 [12 .. 15] Data length
1106 ... Rest of request
1107
1108 [ 0 .. 7] NBD_OPTS_MAGIC
1109 [ 8 .. 11] Second NBD option
1110 [12 .. 15] Data length
1111 ... Rest of request
1112 */
1113
1114 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1115 return -EIO;
1116 }
1117 trace_nbd_negotiate_options_flags(flags);
1118 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1119 fixedNewstyle = true;
1120 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1121 }
1122 if (flags & NBD_FLAG_C_NO_ZEROES) {
1123 no_zeroes = true;
1124 flags &= ~NBD_FLAG_C_NO_ZEROES;
1125 }
1126 if (flags != 0) {
1127 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1128 return -EINVAL;
1129 }
1130
1131 while (1) {
1132 int ret;
1133 uint32_t option, length;
1134 uint64_t magic;
1135
1136 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1137 return -EINVAL;
1138 }
1139 trace_nbd_negotiate_options_check_magic(magic);
1140 if (magic != NBD_OPTS_MAGIC) {
1141 error_setg(errp, "Bad magic received");
1142 return -EINVAL;
1143 }
1144
1145 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1146 return -EINVAL;
1147 }
1148 client->opt = option;
1149
1150 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1151 return -EINVAL;
1152 }
1153 assert(!client->optlen);
1154 client->optlen = length;
1155
1156 if (length > NBD_MAX_BUFFER_SIZE) {
1157 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1158 length, NBD_MAX_BUFFER_SIZE);
1159 return -EINVAL;
1160 }
1161
1162 trace_nbd_negotiate_options_check_option(option,
1163 nbd_opt_lookup(option));
1164 if (client->tlscreds &&
1165 client->ioc == (QIOChannel *)client->sioc) {
1166 QIOChannel *tioc;
1167 if (!fixedNewstyle) {
1168 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1169 return -EINVAL;
1170 }
1171 switch (option) {
1172 case NBD_OPT_STARTTLS:
1173 if (length) {
1174 /* Unconditionally drop the connection if the client
1175 * can't start a TLS negotiation correctly */
1176 return nbd_reject_length(client, true, errp);
1177 }
1178 tioc = nbd_negotiate_handle_starttls(client, errp);
1179 if (!tioc) {
1180 return -EIO;
1181 }
1182 ret = 0;
1183 object_unref(OBJECT(client->ioc));
1184 client->ioc = QIO_CHANNEL(tioc);
1185 break;
1186
1187 case NBD_OPT_EXPORT_NAME:
1188 /* No way to return an error to client, so drop connection */
1189 error_setg(errp, "Option 0x%x not permitted before TLS",
1190 option);
1191 return -EINVAL;
1192
1193 default:
1194 /* Let the client keep trying, unless they asked to
1195 * quit. Always try to give an error back to the
1196 * client; but when replying to OPT_ABORT, be aware
1197 * that the client may hang up before receiving the
1198 * error, in which case we are fine ignoring the
1199 * resulting EPIPE. */
1200 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1201 option == NBD_OPT_ABORT ? NULL : errp,
1202 "Option 0x%" PRIx32
1203 " not permitted before TLS", option);
1204 if (option == NBD_OPT_ABORT) {
1205 return 1;
1206 }
1207 break;
1208 }
1209 } else if (fixedNewstyle) {
1210 switch (option) {
1211 case NBD_OPT_LIST:
1212 if (length) {
1213 ret = nbd_reject_length(client, false, errp);
1214 } else {
1215 ret = nbd_negotiate_handle_list(client, errp);
1216 }
1217 break;
1218
1219 case NBD_OPT_ABORT:
1220 /* NBD spec says we must try to reply before
1221 * disconnecting, but that we must also tolerate
1222 * guests that don't wait for our reply. */
1223 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1224 return 1;
1225
1226 case NBD_OPT_EXPORT_NAME:
1227 return nbd_negotiate_handle_export_name(client, no_zeroes,
1228 errp);
1229
1230 case NBD_OPT_INFO:
1231 case NBD_OPT_GO:
1232 ret = nbd_negotiate_handle_info(client, errp);
1233 if (ret == 1) {
1234 assert(option == NBD_OPT_GO);
1235 return 0;
1236 }
1237 break;
1238
1239 case NBD_OPT_STARTTLS:
1240 if (length) {
1241 ret = nbd_reject_length(client, false, errp);
1242 } else if (client->tlscreds) {
1243 ret = nbd_negotiate_send_rep_err(client,
1244 NBD_REP_ERR_INVALID, errp,
1245 "TLS already enabled");
1246 } else {
1247 ret = nbd_negotiate_send_rep_err(client,
1248 NBD_REP_ERR_POLICY, errp,
1249 "TLS not configured");
1250 }
1251 break;
1252
1253 case NBD_OPT_STRUCTURED_REPLY:
1254 if (length) {
1255 ret = nbd_reject_length(client, false, errp);
1256 } else if (client->structured_reply) {
1257 ret = nbd_negotiate_send_rep_err(
1258 client, NBD_REP_ERR_INVALID, errp,
1259 "structured reply already negotiated");
1260 } else {
1261 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1262 client->structured_reply = true;
1263 }
1264 break;
1265
1266 case NBD_OPT_LIST_META_CONTEXT:
1267 case NBD_OPT_SET_META_CONTEXT:
1268 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1269 errp);
1270 break;
1271
1272 default:
1273 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1274 "Unsupported option %" PRIu32 " (%s)",
1275 option, nbd_opt_lookup(option));
1276 break;
1277 }
1278 } else {
1279 /*
1280 * If broken new-style we should drop the connection
1281 * for anything except NBD_OPT_EXPORT_NAME
1282 */
1283 switch (option) {
1284 case NBD_OPT_EXPORT_NAME:
1285 return nbd_negotiate_handle_export_name(client, no_zeroes,
1286 errp);
1287
1288 default:
1289 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1290 option, nbd_opt_lookup(option));
1291 return -EINVAL;
1292 }
1293 }
1294 if (ret < 0) {
1295 return ret;
1296 }
1297 }
1298 }
1299
1300 /* nbd_negotiate
1301 * Return:
1302 * -errno on error, errp is set
1303 * 0 on successful negotiation, errp is not set
1304 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1305 * errp is not set
1306 */
1307 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1308 {
1309 ERRP_GUARD();
1310 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1311 int ret;
1312
1313 /* Old style negotiation header, no room for options
1314 [ 0 .. 7] passwd ("NBDMAGIC")
1315 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1316 [16 .. 23] size
1317 [24 .. 27] export flags (zero-extended)
1318 [28 .. 151] reserved (0)
1319
1320 New style negotiation header, client can send options
1321 [ 0 .. 7] passwd ("NBDMAGIC")
1322 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1323 [16 .. 17] server flags (0)
1324 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1325 */
1326
1327 qio_channel_set_blocking(client->ioc, false, NULL);
1328
1329 trace_nbd_negotiate_begin();
1330 memcpy(buf, "NBDMAGIC", 8);
1331
1332 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1333 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1334
1335 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1336 error_prepend(errp, "write failed: ");
1337 return -EINVAL;
1338 }
1339 ret = nbd_negotiate_options(client, errp);
1340 if (ret != 0) {
1341 if (ret < 0) {
1342 error_prepend(errp, "option negotiation failed: ");
1343 }
1344 return ret;
1345 }
1346
1347 /* Attach the channel to the same AioContext as the export */
1348 if (client->exp && client->exp->common.ctx) {
1349 qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
1350 }
1351
1352 assert(!client->optlen);
1353 trace_nbd_negotiate_success();
1354
1355 return 0;
1356 }
1357
1358 /* nbd_read_eof
1359 * Tries to read @size bytes from @ioc. This is a local implementation of
1360 * qio_channel_readv_all_eof. We have it here because we need it to be
1361 * interruptible and to know when the coroutine is yielding.
1362 * Returns 1 on success
1363 * 0 on eof, when no data was read (errp is not set)
1364 * negative errno on failure (errp is set)
1365 */
1366 static inline int coroutine_fn
1367 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1368 {
1369 bool partial = false;
1370
1371 assert(size);
1372 while (size > 0) {
1373 struct iovec iov = { .iov_base = buffer, .iov_len = size };
1374 ssize_t len;
1375
1376 len = qio_channel_readv(client->ioc, &iov, 1, errp);
1377 if (len == QIO_CHANNEL_ERR_BLOCK) {
1378 client->read_yielding = true;
1379 qio_channel_yield(client->ioc, G_IO_IN);
1380 client->read_yielding = false;
1381 if (client->quiescing) {
1382 return -EAGAIN;
1383 }
1384 continue;
1385 } else if (len < 0) {
1386 return -EIO;
1387 } else if (len == 0) {
1388 if (partial) {
1389 error_setg(errp,
1390 "Unexpected end-of-file before all bytes were read");
1391 return -EIO;
1392 } else {
1393 return 0;
1394 }
1395 }
1396
1397 partial = true;
1398 size -= len;
1399 buffer = (uint8_t *) buffer + len;
1400 }
1401 return 1;
1402 }
1403
1404 static int nbd_receive_request(NBDClient *client, NBDRequest *request,
1405 Error **errp)
1406 {
1407 uint8_t buf[NBD_REQUEST_SIZE];
1408 uint32_t magic;
1409 int ret;
1410
1411 ret = nbd_read_eof(client, buf, sizeof(buf), errp);
1412 if (ret < 0) {
1413 return ret;
1414 }
1415
1416 /* Request
1417 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1418 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1419 [ 6 .. 7] type (NBD_CMD_READ, ...)
1420 [ 8 .. 15] handle
1421 [16 .. 23] from
1422 [24 .. 27] len
1423 */
1424
1425 magic = ldl_be_p(buf);
1426 request->flags = lduw_be_p(buf + 4);
1427 request->type = lduw_be_p(buf + 6);
1428 request->handle = ldq_be_p(buf + 8);
1429 request->from = ldq_be_p(buf + 16);
1430 request->len = ldl_be_p(buf + 24);
1431
1432 trace_nbd_receive_request(magic, request->flags, request->type,
1433 request->from, request->len);
1434
1435 if (magic != NBD_REQUEST_MAGIC) {
1436 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1437 return -EINVAL;
1438 }
1439 return 0;
1440 }
1441
1442 #define MAX_NBD_REQUESTS 16
1443
1444 void nbd_client_get(NBDClient *client)
1445 {
1446 client->refcount++;
1447 }
1448
1449 void nbd_client_put(NBDClient *client)
1450 {
1451 if (--client->refcount == 0) {
1452 /* The last reference should be dropped by client->close,
1453 * which is called by client_close.
1454 */
1455 assert(client->closing);
1456
1457 qio_channel_detach_aio_context(client->ioc);
1458 object_unref(OBJECT(client->sioc));
1459 object_unref(OBJECT(client->ioc));
1460 if (client->tlscreds) {
1461 object_unref(OBJECT(client->tlscreds));
1462 }
1463 g_free(client->tlsauthz);
1464 if (client->exp) {
1465 QTAILQ_REMOVE(&client->exp->clients, client, next);
1466 blk_exp_unref(&client->exp->common);
1467 }
1468 g_free(client->export_meta.bitmaps);
1469 g_free(client);
1470 }
1471 }
1472
1473 static void client_close(NBDClient *client, bool negotiated)
1474 {
1475 if (client->closing) {
1476 return;
1477 }
1478
1479 client->closing = true;
1480
1481 /* Force requests to finish. They will drop their own references,
1482 * then we'll close the socket and free the NBDClient.
1483 */
1484 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1485 NULL);
1486
1487 /* Also tell the client, so that they release their reference. */
1488 if (client->close_fn) {
1489 client->close_fn(client, negotiated);
1490 }
1491 }
1492
1493 static NBDRequestData *nbd_request_get(NBDClient *client)
1494 {
1495 NBDRequestData *req;
1496
1497 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1498 client->nb_requests++;
1499
1500 req = g_new0(NBDRequestData, 1);
1501 nbd_client_get(client);
1502 req->client = client;
1503 return req;
1504 }
1505
1506 static void nbd_request_put(NBDRequestData *req)
1507 {
1508 NBDClient *client = req->client;
1509
1510 if (req->data) {
1511 qemu_vfree(req->data);
1512 }
1513 g_free(req);
1514
1515 client->nb_requests--;
1516 nbd_client_receive_next_request(client);
1517
1518 nbd_client_put(client);
1519 }
1520
1521 static void blk_aio_attached(AioContext *ctx, void *opaque)
1522 {
1523 NBDExport *exp = opaque;
1524 NBDClient *client;
1525
1526 trace_nbd_blk_aio_attached(exp->name, ctx);
1527
1528 exp->common.ctx = ctx;
1529
1530 QTAILQ_FOREACH(client, &exp->clients, next) {
1531 qio_channel_attach_aio_context(client->ioc, ctx);
1532
1533 assert(client->recv_coroutine == NULL);
1534 assert(client->send_coroutine == NULL);
1535
1536 if (client->quiescing) {
1537 client->quiescing = false;
1538 nbd_client_receive_next_request(client);
1539 }
1540 }
1541 }
1542
1543 static void nbd_aio_detach_bh(void *opaque)
1544 {
1545 NBDExport *exp = opaque;
1546 NBDClient *client;
1547
1548 QTAILQ_FOREACH(client, &exp->clients, next) {
1549 qio_channel_detach_aio_context(client->ioc);
1550 client->quiescing = true;
1551
1552 if (client->recv_coroutine) {
1553 if (client->read_yielding) {
1554 qemu_aio_coroutine_enter(exp->common.ctx,
1555 client->recv_coroutine);
1556 } else {
1557 AIO_WAIT_WHILE(exp->common.ctx, client->recv_coroutine != NULL);
1558 }
1559 }
1560
1561 if (client->send_coroutine) {
1562 AIO_WAIT_WHILE(exp->common.ctx, client->send_coroutine != NULL);
1563 }
1564 }
1565 }
1566
1567 static void blk_aio_detach(void *opaque)
1568 {
1569 NBDExport *exp = opaque;
1570
1571 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1572
1573 aio_wait_bh_oneshot(exp->common.ctx, nbd_aio_detach_bh, exp);
1574
1575 exp->common.ctx = NULL;
1576 }
1577
1578 static void nbd_eject_notifier(Notifier *n, void *data)
1579 {
1580 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1581
1582 blk_exp_request_shutdown(&exp->common);
1583 }
1584
1585 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1586 {
1587 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1588 assert(exp->drv == &blk_exp_nbd);
1589 assert(nbd_exp->eject_notifier_blk == NULL);
1590
1591 blk_ref(blk);
1592 nbd_exp->eject_notifier_blk = blk;
1593 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1594 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1595 }
1596
1597 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1598 Error **errp)
1599 {
1600 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1601 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1602 BlockBackend *blk = blk_exp->blk;
1603 int64_t size;
1604 uint64_t perm, shared_perm;
1605 bool readonly = !exp_args->writable;
1606 bool shared = !exp_args->writable;
1607 strList *bitmaps;
1608 size_t i;
1609 int ret;
1610
1611 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1612
1613 if (!nbd_server_is_running()) {
1614 error_setg(errp, "NBD server not running");
1615 return -EINVAL;
1616 }
1617
1618 if (!arg->has_name) {
1619 arg->name = exp_args->node_name;
1620 }
1621
1622 if (strlen(arg->name) > NBD_MAX_STRING_SIZE) {
1623 error_setg(errp, "export name '%s' too long", arg->name);
1624 return -EINVAL;
1625 }
1626
1627 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1628 error_setg(errp, "description '%s' too long", arg->description);
1629 return -EINVAL;
1630 }
1631
1632 if (nbd_export_find(arg->name)) {
1633 error_setg(errp, "NBD server already has export named '%s'", arg->name);
1634 return -EEXIST;
1635 }
1636
1637 size = blk_getlength(blk);
1638 if (size < 0) {
1639 error_setg_errno(errp, -size,
1640 "Failed to determine the NBD export's length");
1641 return size;
1642 }
1643
1644 /* Don't allow resize while the NBD server is running, otherwise we don't
1645 * care what happens with the node. */
1646 blk_get_perm(blk, &perm, &shared_perm);
1647 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1648 if (ret < 0) {
1649 return ret;
1650 }
1651
1652 QTAILQ_INIT(&exp->clients);
1653 exp->name = g_strdup(arg->name);
1654 exp->description = g_strdup(arg->description);
1655 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1656 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1657 if (readonly) {
1658 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1659 if (shared) {
1660 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1661 }
1662 } else {
1663 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1664 NBD_FLAG_SEND_FAST_ZERO);
1665 }
1666 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1667
1668 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1669 exp->nr_export_bitmaps++;
1670 }
1671 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1672 for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1673 i++, bitmaps = bitmaps->next) {
1674 const char *bitmap = bitmaps->value;
1675 BlockDriverState *bs = blk_bs(blk);
1676 BdrvDirtyBitmap *bm = NULL;
1677
1678 while (bs) {
1679 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1680 if (bm != NULL) {
1681 break;
1682 }
1683
1684 bs = bdrv_filter_or_cow_bs(bs);
1685 }
1686
1687 if (bm == NULL) {
1688 ret = -ENOENT;
1689 error_setg(errp, "Bitmap '%s' is not found", bitmap);
1690 goto fail;
1691 }
1692
1693 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1694 ret = -EINVAL;
1695 goto fail;
1696 }
1697
1698 if (readonly && bdrv_is_writable(bs) &&
1699 bdrv_dirty_bitmap_enabled(bm)) {
1700 ret = -EINVAL;
1701 error_setg(errp,
1702 "Enabled bitmap '%s' incompatible with readonly export",
1703 bitmap);
1704 goto fail;
1705 }
1706
1707 exp->export_bitmaps[i] = bm;
1708 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1709 }
1710
1711 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1712 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1713 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1714 }
1715
1716 exp->allocation_depth = arg->allocation_depth;
1717
1718 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1719
1720 QTAILQ_INSERT_TAIL(&exports, exp, next);
1721
1722 return 0;
1723
1724 fail:
1725 g_free(exp->export_bitmaps);
1726 g_free(exp->name);
1727 g_free(exp->description);
1728 return ret;
1729 }
1730
1731 NBDExport *nbd_export_find(const char *name)
1732 {
1733 NBDExport *exp;
1734 QTAILQ_FOREACH(exp, &exports, next) {
1735 if (strcmp(name, exp->name) == 0) {
1736 return exp;
1737 }
1738 }
1739
1740 return NULL;
1741 }
1742
1743 AioContext *
1744 nbd_export_aio_context(NBDExport *exp)
1745 {
1746 return exp->common.ctx;
1747 }
1748
1749 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1750 {
1751 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1752 NBDClient *client, *next;
1753
1754 blk_exp_ref(&exp->common);
1755 /*
1756 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1757 * close mode that stops advertising the export to new clients but
1758 * still permits existing clients to run to completion? Because of
1759 * that possibility, nbd_export_close() can be called more than
1760 * once on an export.
1761 */
1762 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1763 client_close(client, true);
1764 }
1765 if (exp->name) {
1766 g_free(exp->name);
1767 exp->name = NULL;
1768 QTAILQ_REMOVE(&exports, exp, next);
1769 }
1770 blk_exp_unref(&exp->common);
1771 }
1772
1773 static void nbd_export_delete(BlockExport *blk_exp)
1774 {
1775 size_t i;
1776 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1777
1778 assert(exp->name == NULL);
1779 assert(QTAILQ_EMPTY(&exp->clients));
1780
1781 g_free(exp->description);
1782 exp->description = NULL;
1783
1784 if (exp->common.blk) {
1785 if (exp->eject_notifier_blk) {
1786 notifier_remove(&exp->eject_notifier);
1787 blk_unref(exp->eject_notifier_blk);
1788 }
1789 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1790 blk_aio_detach, exp);
1791 }
1792
1793 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1794 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1795 }
1796 }
1797
1798 const BlockExportDriver blk_exp_nbd = {
1799 .type = BLOCK_EXPORT_TYPE_NBD,
1800 .instance_size = sizeof(NBDExport),
1801 .create = nbd_export_create,
1802 .delete = nbd_export_delete,
1803 .request_shutdown = nbd_export_request_shutdown,
1804 };
1805
1806 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1807 unsigned niov, Error **errp)
1808 {
1809 int ret;
1810
1811 g_assert(qemu_in_coroutine());
1812 qemu_co_mutex_lock(&client->send_lock);
1813 client->send_coroutine = qemu_coroutine_self();
1814
1815 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1816
1817 client->send_coroutine = NULL;
1818 qemu_co_mutex_unlock(&client->send_lock);
1819
1820 return ret;
1821 }
1822
1823 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1824 uint64_t handle)
1825 {
1826 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1827 stl_be_p(&reply->error, error);
1828 stq_be_p(&reply->handle, handle);
1829 }
1830
1831 static int nbd_co_send_simple_reply(NBDClient *client,
1832 uint64_t handle,
1833 uint32_t error,
1834 void *data,
1835 size_t len,
1836 Error **errp)
1837 {
1838 NBDSimpleReply reply;
1839 int nbd_err = system_errno_to_nbd_errno(error);
1840 struct iovec iov[] = {
1841 {.iov_base = &reply, .iov_len = sizeof(reply)},
1842 {.iov_base = data, .iov_len = len}
1843 };
1844
1845 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1846 len);
1847 set_be_simple_reply(&reply, nbd_err, handle);
1848
1849 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1850 }
1851
1852 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1853 uint16_t type, uint64_t handle, uint32_t length)
1854 {
1855 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1856 stw_be_p(&chunk->flags, flags);
1857 stw_be_p(&chunk->type, type);
1858 stq_be_p(&chunk->handle, handle);
1859 stl_be_p(&chunk->length, length);
1860 }
1861
1862 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1863 uint64_t handle,
1864 Error **errp)
1865 {
1866 NBDStructuredReplyChunk chunk;
1867 struct iovec iov[] = {
1868 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1869 };
1870
1871 trace_nbd_co_send_structured_done(handle);
1872 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1873
1874 return nbd_co_send_iov(client, iov, 1, errp);
1875 }
1876
1877 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1878 uint64_t handle,
1879 uint64_t offset,
1880 void *data,
1881 size_t size,
1882 bool final,
1883 Error **errp)
1884 {
1885 NBDStructuredReadData chunk;
1886 struct iovec iov[] = {
1887 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1888 {.iov_base = data, .iov_len = size}
1889 };
1890
1891 assert(size);
1892 trace_nbd_co_send_structured_read(handle, offset, data, size);
1893 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1894 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1895 sizeof(chunk) - sizeof(chunk.h) + size);
1896 stq_be_p(&chunk.offset, offset);
1897
1898 return nbd_co_send_iov(client, iov, 2, errp);
1899 }
1900
1901 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1902 uint64_t handle,
1903 uint32_t error,
1904 const char *msg,
1905 Error **errp)
1906 {
1907 NBDStructuredError chunk;
1908 int nbd_err = system_errno_to_nbd_errno(error);
1909 struct iovec iov[] = {
1910 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1911 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1912 };
1913
1914 assert(nbd_err);
1915 trace_nbd_co_send_structured_error(handle, nbd_err,
1916 nbd_err_lookup(nbd_err), msg ? msg : "");
1917 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1918 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1919 stl_be_p(&chunk.error, nbd_err);
1920 stw_be_p(&chunk.message_length, iov[1].iov_len);
1921
1922 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1923 }
1924
1925 /* Do a sparse read and send the structured reply to the client.
1926 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1927 * reported to the client, at which point this function succeeds.
1928 */
1929 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1930 uint64_t handle,
1931 uint64_t offset,
1932 uint8_t *data,
1933 size_t size,
1934 Error **errp)
1935 {
1936 int ret = 0;
1937 NBDExport *exp = client->exp;
1938 size_t progress = 0;
1939
1940 while (progress < size) {
1941 int64_t pnum;
1942 int status = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
1943 offset + progress,
1944 size - progress, &pnum, NULL,
1945 NULL);
1946 bool final;
1947
1948 if (status < 0) {
1949 char *msg = g_strdup_printf("unable to check for holes: %s",
1950 strerror(-status));
1951
1952 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1953 errp);
1954 g_free(msg);
1955 return ret;
1956 }
1957 assert(pnum && pnum <= size - progress);
1958 final = progress + pnum == size;
1959 if (status & BDRV_BLOCK_ZERO) {
1960 NBDStructuredReadHole chunk;
1961 struct iovec iov[] = {
1962 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1963 };
1964
1965 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1966 pnum);
1967 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1968 NBD_REPLY_TYPE_OFFSET_HOLE,
1969 handle, sizeof(chunk) - sizeof(chunk.h));
1970 stq_be_p(&chunk.offset, offset + progress);
1971 stl_be_p(&chunk.length, pnum);
1972 ret = nbd_co_send_iov(client, iov, 1, errp);
1973 } else {
1974 ret = blk_pread(exp->common.blk, offset + progress,
1975 data + progress, pnum);
1976 if (ret < 0) {
1977 error_setg_errno(errp, -ret, "reading from file failed");
1978 break;
1979 }
1980 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1981 data + progress, pnum, final,
1982 errp);
1983 }
1984
1985 if (ret < 0) {
1986 break;
1987 }
1988 progress += pnum;
1989 }
1990 return ret;
1991 }
1992
1993 typedef struct NBDExtentArray {
1994 NBDExtent *extents;
1995 unsigned int nb_alloc;
1996 unsigned int count;
1997 uint64_t total_length;
1998 bool can_add;
1999 bool converted_to_be;
2000 } NBDExtentArray;
2001
2002 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
2003 {
2004 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2005
2006 ea->nb_alloc = nb_alloc;
2007 ea->extents = g_new(NBDExtent, nb_alloc);
2008 ea->can_add = true;
2009
2010 return ea;
2011 }
2012
2013 static void nbd_extent_array_free(NBDExtentArray *ea)
2014 {
2015 g_free(ea->extents);
2016 g_free(ea);
2017 }
2018 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free);
2019
2020 /* Further modifications of the array after conversion are abandoned */
2021 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2022 {
2023 int i;
2024
2025 assert(!ea->converted_to_be);
2026 ea->can_add = false;
2027 ea->converted_to_be = true;
2028
2029 for (i = 0; i < ea->count; i++) {
2030 ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2031 ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
2032 }
2033 }
2034
2035 /*
2036 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2037 * return -1.
2038 * For safety, when returning -1 for the first time, .can_add is set to false,
2039 * further call to nbd_extent_array_add() will crash.
2040 * (to avoid the situation, when after failing to add an extent (returned -1),
2041 * user miss this failure and add another extent, which is successfully added
2042 * (array is full, but new extent may be squashed into the last one), then we
2043 * have invalid array with skipped extent)
2044 */
2045 static int nbd_extent_array_add(NBDExtentArray *ea,
2046 uint32_t length, uint32_t flags)
2047 {
2048 assert(ea->can_add);
2049
2050 if (!length) {
2051 return 0;
2052 }
2053
2054 /* Extend previous extent if flags are the same */
2055 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2056 uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
2057
2058 if (sum <= UINT32_MAX) {
2059 ea->extents[ea->count - 1].length = sum;
2060 ea->total_length += length;
2061 return 0;
2062 }
2063 }
2064
2065 if (ea->count >= ea->nb_alloc) {
2066 ea->can_add = false;
2067 return -1;
2068 }
2069
2070 ea->total_length += length;
2071 ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
2072 ea->count++;
2073
2074 return 0;
2075 }
2076
2077 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
2078 uint64_t bytes, NBDExtentArray *ea)
2079 {
2080 while (bytes) {
2081 uint32_t flags;
2082 int64_t num;
2083 int ret = bdrv_block_status_above(bs, NULL, offset, bytes, &num,
2084 NULL, NULL);
2085
2086 if (ret < 0) {
2087 return ret;
2088 }
2089
2090 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
2091 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2092
2093 if (nbd_extent_array_add(ea, num, flags) < 0) {
2094 return 0;
2095 }
2096
2097 offset += num;
2098 bytes -= num;
2099 }
2100
2101 return 0;
2102 }
2103
2104 static int blockalloc_to_extents(BlockDriverState *bs, uint64_t offset,
2105 uint64_t bytes, NBDExtentArray *ea)
2106 {
2107 while (bytes) {
2108 int64_t num;
2109 int ret = bdrv_is_allocated_above(bs, NULL, false, offset, bytes,
2110 &num);
2111
2112 if (ret < 0) {
2113 return ret;
2114 }
2115
2116 if (nbd_extent_array_add(ea, num, ret) < 0) {
2117 return 0;
2118 }
2119
2120 offset += num;
2121 bytes -= num;
2122 }
2123
2124 return 0;
2125 }
2126
2127 /*
2128 * nbd_co_send_extents
2129 *
2130 * @ea is converted to BE by the function
2131 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2132 */
2133 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
2134 NBDExtentArray *ea,
2135 bool last, uint32_t context_id, Error **errp)
2136 {
2137 NBDStructuredMeta chunk;
2138 struct iovec iov[] = {
2139 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2140 {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
2141 };
2142
2143 nbd_extent_array_convert_to_be(ea);
2144
2145 trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
2146 last);
2147 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
2148 NBD_REPLY_TYPE_BLOCK_STATUS,
2149 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
2150 stl_be_p(&chunk.context_id, context_id);
2151
2152 return nbd_co_send_iov(client, iov, 2, errp);
2153 }
2154
2155 /* Get block status from the exported device and send it to the client */
2156 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2157 BlockDriverState *bs, uint64_t offset,
2158 uint32_t length, bool dont_fragment,
2159 bool last, uint32_t context_id,
2160 Error **errp)
2161 {
2162 int ret;
2163 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2164 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2165
2166 if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2167 ret = blockstatus_to_extents(bs, offset, length, ea);
2168 } else {
2169 ret = blockalloc_to_extents(bs, offset, length, ea);
2170 }
2171 if (ret < 0) {
2172 return nbd_co_send_structured_error(
2173 client, handle, -ret, "can't get block status", errp);
2174 }
2175
2176 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2177 }
2178
2179 /* Populate @ea from a dirty bitmap. */
2180 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2181 uint64_t offset, uint64_t length,
2182 NBDExtentArray *es)
2183 {
2184 int64_t start, dirty_start, dirty_count;
2185 int64_t end = offset + length;
2186 bool full = false;
2187
2188 bdrv_dirty_bitmap_lock(bitmap);
2189
2190 for (start = offset;
2191 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2192 &dirty_start, &dirty_count);
2193 start = dirty_start + dirty_count)
2194 {
2195 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2196 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2197 {
2198 full = true;
2199 break;
2200 }
2201 }
2202
2203 if (!full) {
2204 /* last non dirty extent, nothing to do if array is now full */
2205 (void) nbd_extent_array_add(es, end - start, 0);
2206 }
2207
2208 bdrv_dirty_bitmap_unlock(bitmap);
2209 }
2210
2211 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2212 BdrvDirtyBitmap *bitmap, uint64_t offset,
2213 uint32_t length, bool dont_fragment, bool last,
2214 uint32_t context_id, Error **errp)
2215 {
2216 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2217 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2218
2219 bitmap_to_extents(bitmap, offset, length, ea);
2220
2221 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2222 }
2223
2224 /* nbd_co_receive_request
2225 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2226 * connection right away, -EAGAIN to indicate we were interrupted and the
2227 * channel should be quiesced, and any other negative value to report an error
2228 * to the client (although the caller may still need to disconnect after
2229 * reporting the error).
2230 */
2231 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2232 Error **errp)
2233 {
2234 NBDClient *client = req->client;
2235 int valid_flags;
2236 int ret;
2237
2238 g_assert(qemu_in_coroutine());
2239 assert(client->recv_coroutine == qemu_coroutine_self());
2240 ret = nbd_receive_request(client, request, errp);
2241 if (ret < 0) {
2242 return ret;
2243 }
2244
2245 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2246 nbd_cmd_lookup(request->type));
2247
2248 if (request->type != NBD_CMD_WRITE) {
2249 /* No payload, we are ready to read the next request. */
2250 req->complete = true;
2251 }
2252
2253 if (request->type == NBD_CMD_DISC) {
2254 /* Special case: we're going to disconnect without a reply,
2255 * whether or not flags, from, or len are bogus */
2256 return -EIO;
2257 }
2258
2259 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2260 request->type == NBD_CMD_CACHE)
2261 {
2262 if (request->len > NBD_MAX_BUFFER_SIZE) {
2263 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2264 request->len, NBD_MAX_BUFFER_SIZE);
2265 return -EINVAL;
2266 }
2267
2268 if (request->type != NBD_CMD_CACHE) {
2269 req->data = blk_try_blockalign(client->exp->common.blk,
2270 request->len);
2271 if (req->data == NULL) {
2272 error_setg(errp, "No memory");
2273 return -ENOMEM;
2274 }
2275 }
2276 }
2277
2278 if (request->type == NBD_CMD_WRITE) {
2279 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2280 errp) < 0)
2281 {
2282 return -EIO;
2283 }
2284 req->complete = true;
2285
2286 trace_nbd_co_receive_request_payload_received(request->handle,
2287 request->len);
2288 }
2289
2290 /* Sanity checks. */
2291 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2292 (request->type == NBD_CMD_WRITE ||
2293 request->type == NBD_CMD_WRITE_ZEROES ||
2294 request->type == NBD_CMD_TRIM)) {
2295 error_setg(errp, "Export is read-only");
2296 return -EROFS;
2297 }
2298 if (request->from > client->exp->size ||
2299 request->len > client->exp->size - request->from) {
2300 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2301 ", Size: %" PRIu64, request->from, request->len,
2302 client->exp->size);
2303 return (request->type == NBD_CMD_WRITE ||
2304 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2305 }
2306 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2307 client->check_align)) {
2308 /*
2309 * The block layer gracefully handles unaligned requests, but
2310 * it's still worth tracing client non-compliance
2311 */
2312 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2313 request->from,
2314 request->len,
2315 client->check_align);
2316 }
2317 valid_flags = NBD_CMD_FLAG_FUA;
2318 if (request->type == NBD_CMD_READ && client->structured_reply) {
2319 valid_flags |= NBD_CMD_FLAG_DF;
2320 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2321 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2322 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2323 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2324 }
2325 if (request->flags & ~valid_flags) {
2326 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2327 nbd_cmd_lookup(request->type), request->flags);
2328 return -EINVAL;
2329 }
2330
2331 return 0;
2332 }
2333
2334 /* Send simple reply without a payload, or a structured error
2335 * @error_msg is ignored if @ret >= 0
2336 * Returns 0 if connection is still live, -errno on failure to talk to client
2337 */
2338 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2339 uint64_t handle,
2340 int ret,
2341 const char *error_msg,
2342 Error **errp)
2343 {
2344 if (client->structured_reply && ret < 0) {
2345 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2346 errp);
2347 } else {
2348 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2349 NULL, 0, errp);
2350 }
2351 }
2352
2353 /* Handle NBD_CMD_READ request.
2354 * Return -errno if sending fails. Other errors are reported directly to the
2355 * client as an error reply. */
2356 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2357 uint8_t *data, Error **errp)
2358 {
2359 int ret;
2360 NBDExport *exp = client->exp;
2361
2362 assert(request->type == NBD_CMD_READ);
2363
2364 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2365 if (request->flags & NBD_CMD_FLAG_FUA) {
2366 ret = blk_co_flush(exp->common.blk);
2367 if (ret < 0) {
2368 return nbd_send_generic_reply(client, request->handle, ret,
2369 "flush failed", errp);
2370 }
2371 }
2372
2373 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2374 request->len)
2375 {
2376 return nbd_co_send_sparse_read(client, request->handle, request->from,
2377 data, request->len, errp);
2378 }
2379
2380 ret = blk_pread(exp->common.blk, request->from, data, request->len);
2381 if (ret < 0) {
2382 return nbd_send_generic_reply(client, request->handle, ret,
2383 "reading from file failed", errp);
2384 }
2385
2386 if (client->structured_reply) {
2387 if (request->len) {
2388 return nbd_co_send_structured_read(client, request->handle,
2389 request->from, data,
2390 request->len, true, errp);
2391 } else {
2392 return nbd_co_send_structured_done(client, request->handle, errp);
2393 }
2394 } else {
2395 return nbd_co_send_simple_reply(client, request->handle, 0,
2396 data, request->len, errp);
2397 }
2398 }
2399
2400 /*
2401 * nbd_do_cmd_cache
2402 *
2403 * Handle NBD_CMD_CACHE request.
2404 * Return -errno if sending fails. Other errors are reported directly to the
2405 * client as an error reply.
2406 */
2407 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2408 Error **errp)
2409 {
2410 int ret;
2411 NBDExport *exp = client->exp;
2412
2413 assert(request->type == NBD_CMD_CACHE);
2414
2415 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2416 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2417
2418 return nbd_send_generic_reply(client, request->handle, ret,
2419 "caching data failed", errp);
2420 }
2421
2422 /* Handle NBD request.
2423 * Return -errno if sending fails. Other errors are reported directly to the
2424 * client as an error reply. */
2425 static coroutine_fn int nbd_handle_request(NBDClient *client,
2426 NBDRequest *request,
2427 uint8_t *data, Error **errp)
2428 {
2429 int ret;
2430 int flags;
2431 NBDExport *exp = client->exp;
2432 char *msg;
2433 size_t i;
2434
2435 switch (request->type) {
2436 case NBD_CMD_CACHE:
2437 return nbd_do_cmd_cache(client, request, errp);
2438
2439 case NBD_CMD_READ:
2440 return nbd_do_cmd_read(client, request, data, errp);
2441
2442 case NBD_CMD_WRITE:
2443 flags = 0;
2444 if (request->flags & NBD_CMD_FLAG_FUA) {
2445 flags |= BDRV_REQ_FUA;
2446 }
2447 ret = blk_pwrite(exp->common.blk, request->from, data, request->len,
2448 flags);
2449 return nbd_send_generic_reply(client, request->handle, ret,
2450 "writing to file failed", errp);
2451
2452 case NBD_CMD_WRITE_ZEROES:
2453 flags = 0;
2454 if (request->flags & NBD_CMD_FLAG_FUA) {
2455 flags |= BDRV_REQ_FUA;
2456 }
2457 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2458 flags |= BDRV_REQ_MAY_UNMAP;
2459 }
2460 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2461 flags |= BDRV_REQ_NO_FALLBACK;
2462 }
2463 ret = 0;
2464 /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2465 while (ret >= 0 && request->len) {
2466 int align = client->check_align ?: 1;
2467 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2468 align));
2469 ret = blk_pwrite_zeroes(exp->common.blk, request->from, len, flags);
2470 request->len -= len;
2471 request->from += len;
2472 }
2473 return nbd_send_generic_reply(client, request->handle, ret,
2474 "writing to file failed", errp);
2475
2476 case NBD_CMD_DISC:
2477 /* unreachable, thanks to special case in nbd_co_receive_request() */
2478 abort();
2479
2480 case NBD_CMD_FLUSH:
2481 ret = blk_co_flush(exp->common.blk);
2482 return nbd_send_generic_reply(client, request->handle, ret,
2483 "flush failed", errp);
2484
2485 case NBD_CMD_TRIM:
2486 ret = 0;
2487 /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2488 while (ret >= 0 && request->len) {
2489 int align = client->check_align ?: 1;
2490 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2491 align));
2492 ret = blk_co_pdiscard(exp->common.blk, request->from, len);
2493 request->len -= len;
2494 request->from += len;
2495 }
2496 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2497 ret = blk_co_flush(exp->common.blk);
2498 }
2499 return nbd_send_generic_reply(client, request->handle, ret,
2500 "discard failed", errp);
2501
2502 case NBD_CMD_BLOCK_STATUS:
2503 if (!request->len) {
2504 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2505 "need non-zero length", errp);
2506 }
2507 if (client->export_meta.count) {
2508 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2509 int contexts_remaining = client->export_meta.count;
2510
2511 if (client->export_meta.base_allocation) {
2512 ret = nbd_co_send_block_status(client, request->handle,
2513 blk_bs(exp->common.blk),
2514 request->from,
2515 request->len, dont_fragment,
2516 !--contexts_remaining,
2517 NBD_META_ID_BASE_ALLOCATION,
2518 errp);
2519 if (ret < 0) {
2520 return ret;
2521 }
2522 }
2523
2524 if (client->export_meta.allocation_depth) {
2525 ret = nbd_co_send_block_status(client, request->handle,
2526 blk_bs(exp->common.blk),
2527 request->from, request->len,
2528 dont_fragment,
2529 !--contexts_remaining,
2530 NBD_META_ID_ALLOCATION_DEPTH,
2531 errp);
2532 if (ret < 0) {
2533 return ret;
2534 }
2535 }
2536
2537 for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2538 if (!client->export_meta.bitmaps[i]) {
2539 continue;
2540 }
2541 ret = nbd_co_send_bitmap(client, request->handle,
2542 client->exp->export_bitmaps[i],
2543 request->from, request->len,
2544 dont_fragment, !--contexts_remaining,
2545 NBD_META_ID_DIRTY_BITMAP + i, errp);
2546 if (ret < 0) {
2547 return ret;
2548 }
2549 }
2550
2551 assert(!contexts_remaining);
2552
2553 return 0;
2554 } else {
2555 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2556 "CMD_BLOCK_STATUS not negotiated",
2557 errp);
2558 }
2559
2560 default:
2561 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2562 request->type);
2563 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2564 errp);
2565 g_free(msg);
2566 return ret;
2567 }
2568 }
2569
2570 /* Owns a reference to the NBDClient passed as opaque. */
2571 static coroutine_fn void nbd_trip(void *opaque)
2572 {
2573 NBDClient *client = opaque;
2574 NBDRequestData *req;
2575 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2576 int ret;
2577 Error *local_err = NULL;
2578
2579 trace_nbd_trip();
2580 if (client->closing) {
2581 nbd_client_put(client);
2582 return;
2583 }
2584
2585 if (client->quiescing) {
2586 /*
2587 * We're switching between AIO contexts. Don't attempt to receive a new
2588 * request and kick the main context which may be waiting for us.
2589 */
2590 nbd_client_put(client);
2591 client->recv_coroutine = NULL;
2592 aio_wait_kick();
2593 return;
2594 }
2595
2596 req = nbd_request_get(client);
2597 ret = nbd_co_receive_request(req, &request, &local_err);
2598 client->recv_coroutine = NULL;
2599
2600 if (client->closing) {
2601 /*
2602 * The client may be closed when we are blocked in
2603 * nbd_co_receive_request()
2604 */
2605 goto done;
2606 }
2607
2608 if (ret == -EAGAIN) {
2609 assert(client->quiescing);
2610 goto done;
2611 }
2612
2613 nbd_client_receive_next_request(client);
2614 if (ret == -EIO) {
2615 goto disconnect;
2616 }
2617
2618 if (ret < 0) {
2619 /* It wans't -EIO, so, according to nbd_co_receive_request()
2620 * semantics, we should return the error to the client. */
2621 Error *export_err = local_err;
2622
2623 local_err = NULL;
2624 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2625 error_get_pretty(export_err), &local_err);
2626 error_free(export_err);
2627 } else {
2628 ret = nbd_handle_request(client, &request, req->data, &local_err);
2629 }
2630 if (ret < 0) {
2631 error_prepend(&local_err, "Failed to send reply: ");
2632 goto disconnect;
2633 }
2634
2635 /* We must disconnect after NBD_CMD_WRITE if we did not
2636 * read the payload.
2637 */
2638 if (!req->complete) {
2639 error_setg(&local_err, "Request handling failed in intermediate state");
2640 goto disconnect;
2641 }
2642
2643 done:
2644 nbd_request_put(req);
2645 nbd_client_put(client);
2646 return;
2647
2648 disconnect:
2649 if (local_err) {
2650 error_reportf_err(local_err, "Disconnect client, due to: ");
2651 }
2652 nbd_request_put(req);
2653 client_close(client, true);
2654 nbd_client_put(client);
2655 }
2656
2657 static void nbd_client_receive_next_request(NBDClient *client)
2658 {
2659 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
2660 !client->quiescing) {
2661 nbd_client_get(client);
2662 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2663 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
2664 }
2665 }
2666
2667 static coroutine_fn void nbd_co_client_start(void *opaque)
2668 {
2669 NBDClient *client = opaque;
2670 Error *local_err = NULL;
2671
2672 qemu_co_mutex_init(&client->send_lock);
2673
2674 if (nbd_negotiate(client, &local_err)) {
2675 if (local_err) {
2676 error_report_err(local_err);
2677 }
2678 client_close(client, false);
2679 return;
2680 }
2681
2682 nbd_client_receive_next_request(client);
2683 }
2684
2685 /*
2686 * Create a new client listener using the given channel @sioc.
2687 * Begin servicing it in a coroutine. When the connection closes, call
2688 * @close_fn with an indication of whether the client completed negotiation.
2689 */
2690 void nbd_client_new(QIOChannelSocket *sioc,
2691 QCryptoTLSCreds *tlscreds,
2692 const char *tlsauthz,
2693 void (*close_fn)(NBDClient *, bool))
2694 {
2695 NBDClient *client;
2696 Coroutine *co;
2697
2698 client = g_new0(NBDClient, 1);
2699 client->refcount = 1;
2700 client->tlscreds = tlscreds;
2701 if (tlscreds) {
2702 object_ref(OBJECT(client->tlscreds));
2703 }
2704 client->tlsauthz = g_strdup(tlsauthz);
2705 client->sioc = sioc;
2706 object_ref(OBJECT(client->sioc));
2707 client->ioc = QIO_CHANNEL(sioc);
2708 object_ref(OBJECT(client->ioc));
2709 client->close_fn = close_fn;
2710
2711 co = qemu_coroutine_create(nbd_co_client_start, client);
2712 qemu_coroutine_enter(co);
2713 }