?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'v9.0.0-rc3'
[qemu/ar7.git] / nbd / server.c
blobc3484cc1ebc61990491aa60681ba2337b83bd8ce
1 /*
2 * Copyright Red Hat
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/block_int.h"
23 #include "block/export.h"
24 #include "block/dirty-bitmap.h"
25 #include "qapi/error.h"
26 #include "qemu/queue.h"
27 #include "trace.h"
28 #include "nbd-internal.h"
29 #include "qemu/units.h"
30 #include "qemu/memalign.h"
31
32 #define NBD_META_ID_BASE_ALLOCATION 0
33 #define NBD_META_ID_ALLOCATION_DEPTH 1
34 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
35 #define NBD_META_ID_DIRTY_BITMAP 2
36
37 /*
38 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
39 * constant. If an increase is needed, note that the NBD protocol
40 * recommends no larger than 32 mb, so that the client won't consider
41 * the reply as a denial of service attack.
42 */
43 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
44
45 static int system_errno_to_nbd_errno(int err)
46 {
47 switch (err) {
48 case 0:
49 return NBD_SUCCESS;
50 case EPERM:
51 case EROFS:
52 return NBD_EPERM;
53 case EIO:
54 return NBD_EIO;
55 case ENOMEM:
56 return NBD_ENOMEM;
57 #ifdef EDQUOT
58 case EDQUOT:
59 #endif
60 case EFBIG:
61 case ENOSPC:
62 return NBD_ENOSPC;
63 case EOVERFLOW:
64 return NBD_EOVERFLOW;
65 case ENOTSUP:
66 #if ENOTSUP != EOPNOTSUPP
67 case EOPNOTSUPP:
68 #endif
69 return NBD_ENOTSUP;
70 case ESHUTDOWN:
71 return NBD_ESHUTDOWN;
72 case EINVAL:
73 default:
74 return NBD_EINVAL;
75 }
76 }
77
78 /* Definitions for opaque data types */
79
80 typedef struct NBDRequestData NBDRequestData;
81
82 struct NBDRequestData {
83 NBDClient *client;
84 uint8_t *data;
85 bool complete;
86 };
87
88 struct NBDExport {
89 BlockExport common;
90
91 char *name;
92 char *description;
93 uint64_t size;
94 uint16_t nbdflags;
95 QTAILQ_HEAD(, NBDClient) clients;
96 QTAILQ_ENTRY(NBDExport) next;
97
98 BlockBackend *eject_notifier_blk;
99 Notifier eject_notifier;
100
101 bool allocation_depth;
102 BdrvDirtyBitmap **export_bitmaps;
103 size_t nr_export_bitmaps;
104 };
105
106 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
107
108 /*
109 * NBDMetaContexts represents a list of meta contexts in use,
110 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
111 * NBD_OPT_LIST_META_CONTEXT.
112 */
113 struct NBDMetaContexts {
114 const NBDExport *exp; /* associated export */
115 size_t count; /* number of negotiated contexts */
116 bool base_allocation; /* export base:allocation context (block status) */
117 bool allocation_depth; /* export qemu:allocation-depth */
118 bool *bitmaps; /*
119 * export qemu:dirty-bitmap:<export bitmap name>,
120 * sized by exp->nr_export_bitmaps
121 */
122 };
123
124 struct NBDClient {
125 int refcount; /* atomic */
126 void (*close_fn)(NBDClient *client, bool negotiated);
127
128 QemuMutex lock;
129
130 NBDExport *exp;
131 QCryptoTLSCreds *tlscreds;
132 char *tlsauthz;
133 QIOChannelSocket *sioc; /* The underlying data channel */
134 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
135
136 Coroutine *recv_coroutine; /* protected by lock */
137
138 CoMutex send_lock;
139 Coroutine *send_coroutine;
140
141 bool read_yielding; /* protected by lock */
142 bool quiescing; /* protected by lock */
143
144 QTAILQ_ENTRY(NBDClient) next;
145 int nb_requests; /* protected by lock */
146 bool closing; /* protected by lock */
147
148 uint32_t check_align; /* If non-zero, check for aligned client requests */
149
150 NBDMode mode;
151 NBDMetaContexts contexts; /* Negotiated meta contexts */
152
153 uint32_t opt; /* Current option being negotiated */
154 uint32_t optlen; /* remaining length of data in ioc for the option being
155 negotiated now */
156 };
157
158 static void nbd_client_receive_next_request(NBDClient *client);
159
160 /* Basic flow for negotiation
161
162 Server Client
163 Negotiate
164
165 or
166
167 Server Client
168 Negotiate #1
169 Option
170 Negotiate #2
171
172 ----
173
174 followed by
175
176 Server Client
177 Request
178 Response
179 Request
180 Response
181 ...
182 ...
183 Request (type == 2)
184
185 */
186
187 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
188 uint32_t type, uint32_t length)
189 {
190 stq_be_p(&rep->magic, NBD_REP_MAGIC);
191 stl_be_p(&rep->option, option);
192 stl_be_p(&rep->type, type);
193 stl_be_p(&rep->length, length);
194 }
195
196 /* Send a reply header, including length, but no payload.
197 * Return -errno on error, 0 on success. */
198 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
199 uint32_t len, Error **errp)
200 {
201 NBDOptionReply rep;
202
203 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
204 type, nbd_rep_lookup(type), len);
205
206 assert(len < NBD_MAX_BUFFER_SIZE);
207
208 set_be_option_rep(&rep, client->opt, type, len);
209 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
210 }
211
212 /* Send a reply header with default 0 length.
213 * Return -errno on error, 0 on success. */
214 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
215 Error **errp)
216 {
217 return nbd_negotiate_send_rep_len(client, type, 0, errp);
218 }
219
220 /* Send an error reply.
221 * Return -errno on error, 0 on success. */
222 static int G_GNUC_PRINTF(4, 0)
223 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
224 Error **errp, const char *fmt, va_list va)
225 {
226 ERRP_GUARD();
227 g_autofree char *msg = NULL;
228 int ret;
229 size_t len;
230
231 msg = g_strdup_vprintf(fmt, va);
232 len = strlen(msg);
233 assert(len < NBD_MAX_STRING_SIZE);
234 trace_nbd_negotiate_send_rep_err(msg);
235 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
236 if (ret < 0) {
237 return ret;
238 }
239 if (nbd_write(client->ioc, msg, len, errp) < 0) {
240 error_prepend(errp, "write failed (error message): ");
241 return -EIO;
242 }
243
244 return 0;
245 }
246
247 /*
248 * Return a malloc'd copy of @name suitable for use in an error reply.
249 */
250 static char *
251 nbd_sanitize_name(const char *name)
252 {
253 if (strnlen(name, 80) < 80) {
254 return g_strdup(name);
255 }
256 /* XXX Should we also try to sanitize any control characters? */
257 return g_strdup_printf("%.80s...", name);
258 }
259
260 /* Send an error reply.
261 * Return -errno on error, 0 on success. */
262 static int G_GNUC_PRINTF(4, 5)
263 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
264 Error **errp, const char *fmt, ...)
265 {
266 va_list va;
267 int ret;
268
269 va_start(va, fmt);
270 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
271 va_end(va);
272 return ret;
273 }
274
275 /* Drop remainder of the current option, and send a reply with the
276 * given error type and message. Return -errno on read or write
277 * failure; or 0 if connection is still live. */
278 static int G_GNUC_PRINTF(4, 0)
279 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
280 const char *fmt, va_list va)
281 {
282 int ret = nbd_drop(client->ioc, client->optlen, errp);
283
284 client->optlen = 0;
285 if (!ret) {
286 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
287 }
288 return ret;
289 }
290
291 static int G_GNUC_PRINTF(4, 5)
292 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
293 const char *fmt, ...)
294 {
295 int ret;
296 va_list va;
297
298 va_start(va, fmt);
299 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
300 va_end(va);
301
302 return ret;
303 }
304
305 static int G_GNUC_PRINTF(3, 4)
306 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
307 {
308 int ret;
309 va_list va;
310
311 va_start(va, fmt);
312 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
313 va_end(va);
314
315 return ret;
316 }
317
318 /* Read size bytes from the unparsed payload of the current option.
319 * If @check_nul, require that no NUL bytes appear in buffer.
320 * Return -errno on I/O error, 0 if option was completely handled by
321 * sending a reply about inconsistent lengths, or 1 on success. */
322 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
323 bool check_nul, Error **errp)
324 {
325 if (size > client->optlen) {
326 return nbd_opt_invalid(client, errp,
327 "Inconsistent lengths in option %s",
328 nbd_opt_lookup(client->opt));
329 }
330 client->optlen -= size;
331 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
332 return -EIO;
333 }
334
335 if (check_nul && strnlen(buffer, size) != size) {
336 return nbd_opt_invalid(client, errp,
337 "Unexpected embedded NUL in option %s",
338 nbd_opt_lookup(client->opt));
339 }
340 return 1;
341 }
342
343 /* Drop size bytes from the unparsed payload of the current option.
344 * Return -errno on I/O error, 0 if option was completely handled by
345 * sending a reply about inconsistent lengths, or 1 on success. */
346 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
347 {
348 if (size > client->optlen) {
349 return nbd_opt_invalid(client, errp,
350 "Inconsistent lengths in option %s",
351 nbd_opt_lookup(client->opt));
352 }
353 client->optlen -= size;
354 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
355 }
356
357 /* nbd_opt_read_name
358 *
359 * Read a string with the format:
360 * uint32_t len (<= NBD_MAX_STRING_SIZE)
361 * len bytes string (not 0-terminated)
362 *
363 * On success, @name will be allocated.
364 * If @length is non-null, it will be set to the actual string length.
365 *
366 * Return -errno on I/O error, 0 if option was completely handled by
367 * sending a reply about inconsistent lengths, or 1 on success.
368 */
369 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
370 Error **errp)
371 {
372 int ret;
373 uint32_t len;
374 g_autofree char *local_name = NULL;
375
376 *name = NULL;
377 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
378 if (ret <= 0) {
379 return ret;
380 }
381 len = cpu_to_be32(len);
382
383 if (len > NBD_MAX_STRING_SIZE) {
384 return nbd_opt_invalid(client, errp,
385 "Invalid name length: %" PRIu32, len);
386 }
387
388 local_name = g_malloc(len + 1);
389 ret = nbd_opt_read(client, local_name, len, true, errp);
390 if (ret <= 0) {
391 return ret;
392 }
393 local_name[len] = '\0';
394
395 if (length) {
396 *length = len;
397 }
398 *name = g_steal_pointer(&local_name);
399
400 return 1;
401 }
402
403 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
404 * Return -errno on error, 0 on success. */
405 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
406 Error **errp)
407 {
408 ERRP_GUARD();
409 size_t name_len, desc_len;
410 uint32_t len;
411 const char *name = exp->name ? exp->name : "";
412 const char *desc = exp->description ? exp->description : "";
413 QIOChannel *ioc = client->ioc;
414 int ret;
415
416 trace_nbd_negotiate_send_rep_list(name, desc);
417 name_len = strlen(name);
418 desc_len = strlen(desc);
419 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
420 len = name_len + desc_len + sizeof(len);
421 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
422 if (ret < 0) {
423 return ret;
424 }
425
426 len = cpu_to_be32(name_len);
427 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
428 error_prepend(errp, "write failed (name length): ");
429 return -EINVAL;
430 }
431
432 if (nbd_write(ioc, name, name_len, errp) < 0) {
433 error_prepend(errp, "write failed (name buffer): ");
434 return -EINVAL;
435 }
436
437 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
438 error_prepend(errp, "write failed (description buffer): ");
439 return -EINVAL;
440 }
441
442 return 0;
443 }
444
445 /* Process the NBD_OPT_LIST command, with a potential series of replies.
446 * Return -errno on error, 0 on success. */
447 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
448 {
449 NBDExport *exp;
450 assert(client->opt == NBD_OPT_LIST);
451
452 /* For each export, send a NBD_REP_SERVER reply. */
453 QTAILQ_FOREACH(exp, &exports, next) {
454 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
455 return -EINVAL;
456 }
457 }
458 /* Finish with a NBD_REP_ACK. */
459 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
460 }
461
462 static void nbd_check_meta_export(NBDClient *client, NBDExport *exp)
463 {
464 if (exp != client->contexts.exp) {
465 client->contexts.count = 0;
466 }
467 }
468
469 /* Send a reply to NBD_OPT_EXPORT_NAME.
470 * Return -errno on error, 0 on success. */
471 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
472 Error **errp)
473 {
474 ERRP_GUARD();
475 g_autofree char *name = NULL;
476 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
477 size_t len;
478 int ret;
479 uint16_t myflags;
480
481 /* Client sends:
482 [20 .. xx] export name (length bytes)
483 Server replies:
484 [ 0 .. 7] size
485 [ 8 .. 9] export flags
486 [10 .. 133] reserved (0) [unless no_zeroes]
487 */
488 trace_nbd_negotiate_handle_export_name();
489 if (client->mode >= NBD_MODE_EXTENDED) {
490 error_setg(errp, "Extended headers already negotiated");
491 return -EINVAL;
492 }
493 if (client->optlen > NBD_MAX_STRING_SIZE) {
494 error_setg(errp, "Bad length received");
495 return -EINVAL;
496 }
497 name = g_malloc(client->optlen + 1);
498 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
499 return -EIO;
500 }
501 name[client->optlen] = '\0';
502 client->optlen = 0;
503
504 trace_nbd_negotiate_handle_export_name_request(name);
505
506 client->exp = nbd_export_find(name);
507 if (!client->exp) {
508 error_setg(errp, "export not found");
509 return -EINVAL;
510 }
511 nbd_check_meta_export(client, client->exp);
512
513 myflags = client->exp->nbdflags;
514 if (client->mode >= NBD_MODE_STRUCTURED) {
515 myflags |= NBD_FLAG_SEND_DF;
516 }
517 if (client->mode >= NBD_MODE_EXTENDED && client->contexts.count) {
518 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD;
519 }
520 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
521 stq_be_p(buf, client->exp->size);
522 stw_be_p(buf + 8, myflags);
523 len = no_zeroes ? 10 : sizeof(buf);
524 ret = nbd_write(client->ioc, buf, len, errp);
525 if (ret < 0) {
526 error_prepend(errp, "write failed: ");
527 return ret;
528 }
529
530 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
531 blk_exp_ref(&client->exp->common);
532
533 return 0;
534 }
535
536 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
537 * The buffer does NOT include the info type prefix.
538 * Return -errno on error, 0 if ready to send more. */
539 static int nbd_negotiate_send_info(NBDClient *client,
540 uint16_t info, uint32_t length, void *buf,
541 Error **errp)
542 {
543 int rc;
544
545 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
546 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
547 sizeof(info) + length, errp);
548 if (rc < 0) {
549 return rc;
550 }
551 info = cpu_to_be16(info);
552 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
553 return -EIO;
554 }
555 if (nbd_write(client->ioc, buf, length, errp) < 0) {
556 return -EIO;
557 }
558 return 0;
559 }
560
561 /* nbd_reject_length: Handle any unexpected payload.
562 * @fatal requests that we quit talking to the client, even if we are able
563 * to successfully send an error reply.
564 * Return:
565 * -errno transmission error occurred or @fatal was requested, errp is set
566 * 0 error message successfully sent to client, errp is not set
567 */
568 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
569 {
570 int ret;
571
572 assert(client->optlen);
573 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
574 nbd_opt_lookup(client->opt));
575 if (fatal && !ret) {
576 error_setg(errp, "option '%s' has unexpected length",
577 nbd_opt_lookup(client->opt));
578 return -EINVAL;
579 }
580 return ret;
581 }
582
583 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
584 * Return -errno on error, 0 if ready for next option, and 1 to move
585 * into transmission phase. */
586 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
587 {
588 int rc;
589 g_autofree char *name = NULL;
590 NBDExport *exp;
591 uint16_t requests;
592 uint16_t request;
593 uint32_t namelen = 0;
594 bool sendname = false;
595 bool blocksize = false;
596 uint32_t sizes[3];
597 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
598 uint32_t check_align = 0;
599 uint16_t myflags;
600
601 /* Client sends:
602 4 bytes: L, name length (can be 0)
603 L bytes: export name
604 2 bytes: N, number of requests (can be 0)
605 N * 2 bytes: N requests
606 */
607 rc = nbd_opt_read_name(client, &name, &namelen, errp);
608 if (rc <= 0) {
609 return rc;
610 }
611 trace_nbd_negotiate_handle_export_name_request(name);
612
613 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
614 if (rc <= 0) {
615 return rc;
616 }
617 requests = be16_to_cpu(requests);
618 trace_nbd_negotiate_handle_info_requests(requests);
619 while (requests--) {
620 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
621 if (rc <= 0) {
622 return rc;
623 }
624 request = be16_to_cpu(request);
625 trace_nbd_negotiate_handle_info_request(request,
626 nbd_info_lookup(request));
627 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
628 * everything else is either a request we don't know or
629 * something we send regardless of request */
630 switch (request) {
631 case NBD_INFO_NAME:
632 sendname = true;
633 break;
634 case NBD_INFO_BLOCK_SIZE:
635 blocksize = true;
636 break;
637 }
638 }
639 if (client->optlen) {
640 return nbd_reject_length(client, false, errp);
641 }
642
643 exp = nbd_export_find(name);
644 if (!exp) {
645 g_autofree char *sane_name = nbd_sanitize_name(name);
646
647 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
648 errp, "export '%s' not present",
649 sane_name);
650 }
651 if (client->opt == NBD_OPT_GO) {
652 nbd_check_meta_export(client, exp);
653 }
654
655 /* Don't bother sending NBD_INFO_NAME unless client requested it */
656 if (sendname) {
657 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
658 errp);
659 if (rc < 0) {
660 return rc;
661 }
662 }
663
664 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
665 * client request */
666 if (exp->description) {
667 size_t len = strlen(exp->description);
668
669 assert(len <= NBD_MAX_STRING_SIZE);
670 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
671 len, exp->description, errp);
672 if (rc < 0) {
673 return rc;
674 }
675 }
676
677 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
678 * according to whether the client requested it, and according to
679 * whether this is OPT_INFO or OPT_GO. */
680 /* minimum - 1 for back-compat, or actual if client will obey it. */
681 if (client->opt == NBD_OPT_INFO || blocksize) {
682 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
683 } else {
684 sizes[0] = 1;
685 }
686 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
687 /* preferred - Hard-code to 4096 for now.
688 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
689 sizes[1] = MAX(4096, sizes[0]);
690 /* maximum - At most 32M, but smaller as appropriate. */
691 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
692 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
693 sizes[0] = cpu_to_be32(sizes[0]);
694 sizes[1] = cpu_to_be32(sizes[1]);
695 sizes[2] = cpu_to_be32(sizes[2]);
696 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
697 sizeof(sizes), sizes, errp);
698 if (rc < 0) {
699 return rc;
700 }
701
702 /* Send NBD_INFO_EXPORT always */
703 myflags = exp->nbdflags;
704 if (client->mode >= NBD_MODE_STRUCTURED) {
705 myflags |= NBD_FLAG_SEND_DF;
706 }
707 if (client->mode >= NBD_MODE_EXTENDED &&
708 (client->contexts.count || client->opt == NBD_OPT_INFO)) {
709 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD;
710 }
711 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
712 stq_be_p(buf, exp->size);
713 stw_be_p(buf + 8, myflags);
714 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
715 sizeof(buf), buf, errp);
716 if (rc < 0) {
717 return rc;
718 }
719
720 /*
721 * If the client is just asking for NBD_OPT_INFO, but forgot to
722 * request block sizes in a situation that would impact
723 * performance, then return an error. But for NBD_OPT_GO, we
724 * tolerate all clients, regardless of alignments.
725 */
726 if (client->opt == NBD_OPT_INFO && !blocksize &&
727 blk_get_request_alignment(exp->common.blk) > 1) {
728 return nbd_negotiate_send_rep_err(client,
729 NBD_REP_ERR_BLOCK_SIZE_REQD,
730 errp,
731 "request NBD_INFO_BLOCK_SIZE to "
732 "use this export");
733 }
734
735 /* Final reply */
736 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
737 if (rc < 0) {
738 return rc;
739 }
740
741 if (client->opt == NBD_OPT_GO) {
742 client->exp = exp;
743 client->check_align = check_align;
744 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
745 blk_exp_ref(&client->exp->common);
746 rc = 1;
747 }
748 return rc;
749 }
750
751
752 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
753 * new channel for all further (now-encrypted) communication. */
754 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
755 Error **errp)
756 {
757 QIOChannel *ioc;
758 QIOChannelTLS *tioc;
759 struct NBDTLSHandshakeData data = { 0 };
760
761 assert(client->opt == NBD_OPT_STARTTLS);
762
763 trace_nbd_negotiate_handle_starttls();
764 ioc = client->ioc;
765
766 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
767 return NULL;
768 }
769
770 tioc = qio_channel_tls_new_server(ioc,
771 client->tlscreds,
772 client->tlsauthz,
773 errp);
774 if (!tioc) {
775 return NULL;
776 }
777
778 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
779 trace_nbd_negotiate_handle_starttls_handshake();
780 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
781 qio_channel_tls_handshake(tioc,
782 nbd_tls_handshake,
783 &data,
784 NULL,
785 NULL);
786
787 if (!data.complete) {
788 g_main_loop_run(data.loop);
789 }
790 g_main_loop_unref(data.loop);
791 if (data.error) {
792 object_unref(OBJECT(tioc));
793 error_propagate(errp, data.error);
794 return NULL;
795 }
796
797 return QIO_CHANNEL(tioc);
798 }
799
800 /* nbd_negotiate_send_meta_context
801 *
802 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
803 *
804 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
805 */
806 static int nbd_negotiate_send_meta_context(NBDClient *client,
807 const char *context,
808 uint32_t context_id,
809 Error **errp)
810 {
811 NBDOptionReplyMetaContext opt;
812 struct iovec iov[] = {
813 {.iov_base = &opt, .iov_len = sizeof(opt)},
814 {.iov_base = (void *)context, .iov_len = strlen(context)}
815 };
816
817 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
818 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
819 context_id = 0;
820 }
821
822 trace_nbd_negotiate_meta_query_reply(context, context_id);
823 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
824 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
825 stl_be_p(&opt.context_id, context_id);
826
827 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
828 }
829
830 /*
831 * Return true if @query matches @pattern, or if @query is empty when
832 * the @client is performing _LIST_.
833 */
834 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
835 const char *query)
836 {
837 if (!*query) {
838 trace_nbd_negotiate_meta_query_parse("empty");
839 return client->opt == NBD_OPT_LIST_META_CONTEXT;
840 }
841 if (strcmp(query, pattern) == 0) {
842 trace_nbd_negotiate_meta_query_parse(pattern);
843 return true;
844 }
845 trace_nbd_negotiate_meta_query_skip("pattern not matched");
846 return false;
847 }
848
849 /*
850 * Return true and adjust @str in place if it begins with @prefix.
851 */
852 static bool nbd_strshift(const char **str, const char *prefix)
853 {
854 size_t len = strlen(prefix);
855
856 if (strncmp(*str, prefix, len) == 0) {
857 *str += len;
858 return true;
859 }
860 return false;
861 }
862
863 /* nbd_meta_base_query
864 *
865 * Handle queries to 'base' namespace. For now, only the base:allocation
866 * context is available. Return true if @query has been handled.
867 */
868 static bool nbd_meta_base_query(NBDClient *client, NBDMetaContexts *meta,
869 const char *query)
870 {
871 if (!nbd_strshift(&query, "base:")) {
872 return false;
873 }
874 trace_nbd_negotiate_meta_query_parse("base:");
875
876 if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
877 meta->base_allocation = true;
878 }
879 return true;
880 }
881
882 /* nbd_meta_qemu_query
883 *
884 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
885 * and qemu:allocation-depth contexts are available. Return true if @query
886 * has been handled.
887 */
888 static bool nbd_meta_qemu_query(NBDClient *client, NBDMetaContexts *meta,
889 const char *query)
890 {
891 size_t i;
892
893 if (!nbd_strshift(&query, "qemu:")) {
894 return false;
895 }
896 trace_nbd_negotiate_meta_query_parse("qemu:");
897
898 if (!*query) {
899 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
900 meta->allocation_depth = meta->exp->allocation_depth;
901 if (meta->exp->nr_export_bitmaps) {
902 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
903 }
904 }
905 trace_nbd_negotiate_meta_query_parse("empty");
906 return true;
907 }
908
909 if (strcmp(query, "allocation-depth") == 0) {
910 trace_nbd_negotiate_meta_query_parse("allocation-depth");
911 meta->allocation_depth = meta->exp->allocation_depth;
912 return true;
913 }
914
915 if (nbd_strshift(&query, "dirty-bitmap:")) {
916 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
917 if (!*query) {
918 if (client->opt == NBD_OPT_LIST_META_CONTEXT &&
919 meta->exp->nr_export_bitmaps) {
920 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
921 }
922 trace_nbd_negotiate_meta_query_parse("empty");
923 return true;
924 }
925
926 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
927 const char *bm_name;
928
929 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
930 if (strcmp(bm_name, query) == 0) {
931 meta->bitmaps[i] = true;
932 trace_nbd_negotiate_meta_query_parse(query);
933 return true;
934 }
935 }
936 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
937 return true;
938 }
939
940 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
941 return true;
942 }
943
944 /* nbd_negotiate_meta_query
945 *
946 * Parse namespace name and call corresponding function to parse body of the
947 * query.
948 *
949 * The only supported namespaces are 'base' and 'qemu'.
950 *
951 * Return -errno on I/O error, 0 if option was completely handled by
952 * sending a reply about inconsistent lengths, or 1 on success. */
953 static int nbd_negotiate_meta_query(NBDClient *client,
954 NBDMetaContexts *meta, Error **errp)
955 {
956 int ret;
957 g_autofree char *query = NULL;
958 uint32_t len;
959
960 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
961 if (ret <= 0) {
962 return ret;
963 }
964 len = cpu_to_be32(len);
965
966 if (len > NBD_MAX_STRING_SIZE) {
967 trace_nbd_negotiate_meta_query_skip("length too long");
968 return nbd_opt_skip(client, len, errp);
969 }
970
971 query = g_malloc(len + 1);
972 ret = nbd_opt_read(client, query, len, true, errp);
973 if (ret <= 0) {
974 return ret;
975 }
976 query[len] = '\0';
977
978 if (nbd_meta_base_query(client, meta, query)) {
979 return 1;
980 }
981 if (nbd_meta_qemu_query(client, meta, query)) {
982 return 1;
983 }
984
985 trace_nbd_negotiate_meta_query_skip("unknown namespace");
986 return 1;
987 }
988
989 /* nbd_negotiate_meta_queries
990 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
991 *
992 * Return -errno on I/O error, or 0 if option was completely handled. */
993 static int nbd_negotiate_meta_queries(NBDClient *client, Error **errp)
994 {
995 int ret;
996 g_autofree char *export_name = NULL;
997 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */
998 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL;
999 NBDMetaContexts local_meta = {0};
1000 NBDMetaContexts *meta;
1001 uint32_t nb_queries;
1002 size_t i;
1003 size_t count = 0;
1004
1005 if (client->opt == NBD_OPT_SET_META_CONTEXT &&
1006 client->mode < NBD_MODE_STRUCTURED) {
1007 return nbd_opt_invalid(client, errp,
1008 "request option '%s' when structured reply "
1009 "is not negotiated",
1010 nbd_opt_lookup(client->opt));
1011 }
1012
1013 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1014 /* Only change the caller's meta on SET. */
1015 meta = &local_meta;
1016 } else {
1017 meta = &client->contexts;
1018 }
1019
1020 g_free(meta->bitmaps);
1021 memset(meta, 0, sizeof(*meta));
1022
1023 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
1024 if (ret <= 0) {
1025 return ret;
1026 }
1027
1028 meta->exp = nbd_export_find(export_name);
1029 if (meta->exp == NULL) {
1030 g_autofree char *sane_name = nbd_sanitize_name(export_name);
1031
1032 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1033 "export '%s' not present", sane_name);
1034 }
1035 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1036 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1037 bitmaps = meta->bitmaps;
1038 }
1039
1040 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1041 if (ret <= 0) {
1042 return ret;
1043 }
1044 nb_queries = cpu_to_be32(nb_queries);
1045 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1046 export_name, nb_queries);
1047
1048 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1049 /* enable all known contexts */
1050 meta->base_allocation = true;
1051 meta->allocation_depth = meta->exp->allocation_depth;
1052 if (meta->exp->nr_export_bitmaps) {
1053 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1054 }
1055 } else {
1056 for (i = 0; i < nb_queries; ++i) {
1057 ret = nbd_negotiate_meta_query(client, meta, errp);
1058 if (ret <= 0) {
1059 return ret;
1060 }
1061 }
1062 }
1063
1064 if (meta->base_allocation) {
1065 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1066 NBD_META_ID_BASE_ALLOCATION,
1067 errp);
1068 if (ret < 0) {
1069 return ret;
1070 }
1071 count++;
1072 }
1073
1074 if (meta->allocation_depth) {
1075 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1076 NBD_META_ID_ALLOCATION_DEPTH,
1077 errp);
1078 if (ret < 0) {
1079 return ret;
1080 }
1081 count++;
1082 }
1083
1084 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1085 const char *bm_name;
1086 g_autofree char *context = NULL;
1087
1088 if (!meta->bitmaps[i]) {
1089 continue;
1090 }
1091
1092 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1093 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1094
1095 ret = nbd_negotiate_send_meta_context(client, context,
1096 NBD_META_ID_DIRTY_BITMAP + i,
1097 errp);
1098 if (ret < 0) {
1099 return ret;
1100 }
1101 count++;
1102 }
1103
1104 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1105 if (ret == 0) {
1106 meta->count = count;
1107 }
1108
1109 return ret;
1110 }
1111
1112 /* nbd_negotiate_options
1113 * Process all NBD_OPT_* client option commands, during fixed newstyle
1114 * negotiation.
1115 * Return:
1116 * -errno on error, errp is set
1117 * 0 on successful negotiation, errp is not set
1118 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1119 * errp is not set
1120 */
1121 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1122 {
1123 uint32_t flags;
1124 bool fixedNewstyle = false;
1125 bool no_zeroes = false;
1126
1127 /* Client sends:
1128 [ 0 .. 3] client flags
1129
1130 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1131 [ 0 .. 7] NBD_OPTS_MAGIC
1132 [ 8 .. 11] NBD option
1133 [12 .. 15] Data length
1134 ... Rest of request
1135
1136 [ 0 .. 7] NBD_OPTS_MAGIC
1137 [ 8 .. 11] Second NBD option
1138 [12 .. 15] Data length
1139 ... Rest of request
1140 */
1141
1142 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1143 return -EIO;
1144 }
1145 client->mode = NBD_MODE_EXPORT_NAME;
1146 trace_nbd_negotiate_options_flags(flags);
1147 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1148 fixedNewstyle = true;
1149 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1150 client->mode = NBD_MODE_SIMPLE;
1151 }
1152 if (flags & NBD_FLAG_C_NO_ZEROES) {
1153 no_zeroes = true;
1154 flags &= ~NBD_FLAG_C_NO_ZEROES;
1155 }
1156 if (flags != 0) {
1157 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1158 return -EINVAL;
1159 }
1160
1161 while (1) {
1162 int ret;
1163 uint32_t option, length;
1164 uint64_t magic;
1165
1166 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1167 return -EINVAL;
1168 }
1169 trace_nbd_negotiate_options_check_magic(magic);
1170 if (magic != NBD_OPTS_MAGIC) {
1171 error_setg(errp, "Bad magic received");
1172 return -EINVAL;
1173 }
1174
1175 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1176 return -EINVAL;
1177 }
1178 client->opt = option;
1179
1180 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1181 return -EINVAL;
1182 }
1183 assert(!client->optlen);
1184 client->optlen = length;
1185
1186 if (length > NBD_MAX_BUFFER_SIZE) {
1187 error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)",
1188 length, NBD_MAX_BUFFER_SIZE);
1189 return -EINVAL;
1190 }
1191
1192 trace_nbd_negotiate_options_check_option(option,
1193 nbd_opt_lookup(option));
1194 if (client->tlscreds &&
1195 client->ioc == (QIOChannel *)client->sioc) {
1196 QIOChannel *tioc;
1197 if (!fixedNewstyle) {
1198 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1199 return -EINVAL;
1200 }
1201 switch (option) {
1202 case NBD_OPT_STARTTLS:
1203 if (length) {
1204 /* Unconditionally drop the connection if the client
1205 * can't start a TLS negotiation correctly */
1206 return nbd_reject_length(client, true, errp);
1207 }
1208 tioc = nbd_negotiate_handle_starttls(client, errp);
1209 if (!tioc) {
1210 return -EIO;
1211 }
1212 ret = 0;
1213 object_unref(OBJECT(client->ioc));
1214 client->ioc = tioc;
1215 break;
1216
1217 case NBD_OPT_EXPORT_NAME:
1218 /* No way to return an error to client, so drop connection */
1219 error_setg(errp, "Option 0x%x not permitted before TLS",
1220 option);
1221 return -EINVAL;
1222
1223 default:
1224 /* Let the client keep trying, unless they asked to
1225 * quit. Always try to give an error back to the
1226 * client; but when replying to OPT_ABORT, be aware
1227 * that the client may hang up before receiving the
1228 * error, in which case we are fine ignoring the
1229 * resulting EPIPE. */
1230 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1231 option == NBD_OPT_ABORT ? NULL : errp,
1232 "Option 0x%" PRIx32
1233 " not permitted before TLS", option);
1234 if (option == NBD_OPT_ABORT) {
1235 return 1;
1236 }
1237 break;
1238 }
1239 } else if (fixedNewstyle) {
1240 switch (option) {
1241 case NBD_OPT_LIST:
1242 if (length) {
1243 ret = nbd_reject_length(client, false, errp);
1244 } else {
1245 ret = nbd_negotiate_handle_list(client, errp);
1246 }
1247 break;
1248
1249 case NBD_OPT_ABORT:
1250 /* NBD spec says we must try to reply before
1251 * disconnecting, but that we must also tolerate
1252 * guests that don't wait for our reply. */
1253 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1254 return 1;
1255
1256 case NBD_OPT_EXPORT_NAME:
1257 return nbd_negotiate_handle_export_name(client, no_zeroes,
1258 errp);
1259
1260 case NBD_OPT_INFO:
1261 case NBD_OPT_GO:
1262 ret = nbd_negotiate_handle_info(client, errp);
1263 if (ret == 1) {
1264 assert(option == NBD_OPT_GO);
1265 return 0;
1266 }
1267 break;
1268
1269 case NBD_OPT_STARTTLS:
1270 if (length) {
1271 ret = nbd_reject_length(client, false, errp);
1272 } else if (client->tlscreds) {
1273 ret = nbd_negotiate_send_rep_err(client,
1274 NBD_REP_ERR_INVALID, errp,
1275 "TLS already enabled");
1276 } else {
1277 ret = nbd_negotiate_send_rep_err(client,
1278 NBD_REP_ERR_POLICY, errp,
1279 "TLS not configured");
1280 }
1281 break;
1282
1283 case NBD_OPT_STRUCTURED_REPLY:
1284 if (length) {
1285 ret = nbd_reject_length(client, false, errp);
1286 } else if (client->mode >= NBD_MODE_EXTENDED) {
1287 ret = nbd_negotiate_send_rep_err(
1288 client, NBD_REP_ERR_EXT_HEADER_REQD, errp,
1289 "extended headers already negotiated");
1290 } else if (client->mode >= NBD_MODE_STRUCTURED) {
1291 ret = nbd_negotiate_send_rep_err(
1292 client, NBD_REP_ERR_INVALID, errp,
1293 "structured reply already negotiated");
1294 } else {
1295 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1296 client->mode = NBD_MODE_STRUCTURED;
1297 }
1298 break;
1299
1300 case NBD_OPT_LIST_META_CONTEXT:
1301 case NBD_OPT_SET_META_CONTEXT:
1302 ret = nbd_negotiate_meta_queries(client, errp);
1303 break;
1304
1305 case NBD_OPT_EXTENDED_HEADERS:
1306 if (length) {
1307 ret = nbd_reject_length(client, false, errp);
1308 } else if (client->mode >= NBD_MODE_EXTENDED) {
1309 ret = nbd_negotiate_send_rep_err(
1310 client, NBD_REP_ERR_INVALID, errp,
1311 "extended headers already negotiated");
1312 } else {
1313 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1314 client->mode = NBD_MODE_EXTENDED;
1315 }
1316 break;
1317
1318 default:
1319 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1320 "Unsupported option %" PRIu32 " (%s)",
1321 option, nbd_opt_lookup(option));
1322 break;
1323 }
1324 } else {
1325 /*
1326 * If broken new-style we should drop the connection
1327 * for anything except NBD_OPT_EXPORT_NAME
1328 */
1329 switch (option) {
1330 case NBD_OPT_EXPORT_NAME:
1331 return nbd_negotiate_handle_export_name(client, no_zeroes,
1332 errp);
1333
1334 default:
1335 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1336 option, nbd_opt_lookup(option));
1337 return -EINVAL;
1338 }
1339 }
1340 if (ret < 0) {
1341 return ret;
1342 }
1343 }
1344 }
1345
1346 /* nbd_negotiate
1347 * Return:
1348 * -errno on error, errp is set
1349 * 0 on successful negotiation, errp is not set
1350 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1351 * errp is not set
1352 */
1353 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1354 {
1355 ERRP_GUARD();
1356 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1357 int ret;
1358
1359 /* Old style negotiation header, no room for options
1360 [ 0 .. 7] passwd ("NBDMAGIC")
1361 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1362 [16 .. 23] size
1363 [24 .. 27] export flags (zero-extended)
1364 [28 .. 151] reserved (0)
1365
1366 New style negotiation header, client can send options
1367 [ 0 .. 7] passwd ("NBDMAGIC")
1368 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1369 [16 .. 17] server flags (0)
1370 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1371 */
1372
1373 qio_channel_set_blocking(client->ioc, false, NULL);
1374 qio_channel_set_follow_coroutine_ctx(client->ioc, true);
1375
1376 trace_nbd_negotiate_begin();
1377 memcpy(buf, "NBDMAGIC", 8);
1378
1379 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1380 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1381
1382 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1383 error_prepend(errp, "write failed: ");
1384 return -EINVAL;
1385 }
1386 ret = nbd_negotiate_options(client, errp);
1387 if (ret != 0) {
1388 if (ret < 0) {
1389 error_prepend(errp, "option negotiation failed: ");
1390 }
1391 return ret;
1392 }
1393
1394 assert(!client->optlen);
1395 trace_nbd_negotiate_success();
1396
1397 return 0;
1398 }
1399
1400 /* nbd_read_eof
1401 * Tries to read @size bytes from @ioc. This is a local implementation of
1402 * qio_channel_readv_all_eof. We have it here because we need it to be
1403 * interruptible and to know when the coroutine is yielding.
1404 * Returns 1 on success
1405 * 0 on eof, when no data was read (errp is not set)
1406 * negative errno on failure (errp is set)
1407 */
1408 static inline int coroutine_fn
1409 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1410 {
1411 bool partial = false;
1412
1413 assert(size);
1414 while (size > 0) {
1415 struct iovec iov = { .iov_base = buffer, .iov_len = size };
1416 ssize_t len;
1417
1418 len = qio_channel_readv(client->ioc, &iov, 1, errp);
1419 if (len == QIO_CHANNEL_ERR_BLOCK) {
1420 WITH_QEMU_LOCK_GUARD(&client->lock) {
1421 client->read_yielding = true;
1422
1423 /* Prompt main loop thread to re-run nbd_drained_poll() */
1424 aio_wait_kick();
1425 }
1426 qio_channel_yield(client->ioc, G_IO_IN);
1427 WITH_QEMU_LOCK_GUARD(&client->lock) {
1428 client->read_yielding = false;
1429 if (client->quiescing) {
1430 return -EAGAIN;
1431 }
1432 }
1433 continue;
1434 } else if (len < 0) {
1435 return -EIO;
1436 } else if (len == 0) {
1437 if (partial) {
1438 error_setg(errp,
1439 "Unexpected end-of-file before all bytes were read");
1440 return -EIO;
1441 } else {
1442 return 0;
1443 }
1444 }
1445
1446 partial = true;
1447 size -= len;
1448 buffer = (uint8_t *) buffer + len;
1449 }
1450 return 1;
1451 }
1452
1453 static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request,
1454 Error **errp)
1455 {
1456 uint8_t buf[NBD_EXTENDED_REQUEST_SIZE];
1457 uint32_t magic, expect;
1458 int ret;
1459 size_t size = client->mode >= NBD_MODE_EXTENDED ?
1460 NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE;
1461
1462 ret = nbd_read_eof(client, buf, size, errp);
1463 if (ret < 0) {
1464 return ret;
1465 }
1466 if (ret == 0) {
1467 return -EIO;
1468 }
1469
1470 /*
1471 * Compact request
1472 * [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1473 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1474 * [ 6 .. 7] type (NBD_CMD_READ, ...)
1475 * [ 8 .. 15] cookie
1476 * [16 .. 23] from
1477 * [24 .. 27] len
1478 * Extended request
1479 * [ 0 .. 3] magic (NBD_EXTENDED_REQUEST_MAGIC)
1480 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...)
1481 * [ 6 .. 7] type (NBD_CMD_READ, ...)
1482 * [ 8 .. 15] cookie
1483 * [16 .. 23] from
1484 * [24 .. 31] len
1485 */
1486
1487 magic = ldl_be_p(buf);
1488 request->flags = lduw_be_p(buf + 4);
1489 request->type = lduw_be_p(buf + 6);
1490 request->cookie = ldq_be_p(buf + 8);
1491 request->from = ldq_be_p(buf + 16);
1492 if (client->mode >= NBD_MODE_EXTENDED) {
1493 request->len = ldq_be_p(buf + 24);
1494 expect = NBD_EXTENDED_REQUEST_MAGIC;
1495 } else {
1496 request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */
1497 expect = NBD_REQUEST_MAGIC;
1498 }
1499
1500 trace_nbd_receive_request(magic, request->flags, request->type,
1501 request->from, request->len);
1502
1503 if (magic != expect) {
1504 error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%"
1505 PRIx32 ")", magic, expect);
1506 return -EINVAL;
1507 }
1508 return 0;
1509 }
1510
1511 #define MAX_NBD_REQUESTS 16
1512
1513 /* Runs in export AioContext and main loop thread */
1514 void nbd_client_get(NBDClient *client)
1515 {
1516 qatomic_inc(&client->refcount);
1517 }
1518
1519 void nbd_client_put(NBDClient *client)
1520 {
1521 assert(qemu_in_main_thread());
1522
1523 if (qatomic_fetch_dec(&client->refcount) == 1) {
1524 /* The last reference should be dropped by client->close,
1525 * which is called by client_close.
1526 */
1527 assert(client->closing);
1528
1529 object_unref(OBJECT(client->sioc));
1530 object_unref(OBJECT(client->ioc));
1531 if (client->tlscreds) {
1532 object_unref(OBJECT(client->tlscreds));
1533 }
1534 g_free(client->tlsauthz);
1535 if (client->exp) {
1536 QTAILQ_REMOVE(&client->exp->clients, client, next);
1537 blk_exp_unref(&client->exp->common);
1538 }
1539 g_free(client->contexts.bitmaps);
1540 qemu_mutex_destroy(&client->lock);
1541 g_free(client);
1542 }
1543 }
1544
1545 /*
1546 * Tries to release the reference to @client, but only if other references
1547 * remain. This is an optimization for the common case where we want to avoid
1548 * the expense of scheduling nbd_client_put() in the main loop thread.
1549 *
1550 * Returns true upon success or false if the reference was not released because
1551 * it is the last reference.
1552 */
1553 static bool nbd_client_put_nonzero(NBDClient *client)
1554 {
1555 int old = qatomic_read(&client->refcount);
1556 int expected;
1557
1558 do {
1559 if (old == 1) {
1560 return false;
1561 }
1562
1563 expected = old;
1564 old = qatomic_cmpxchg(&client->refcount, expected, expected - 1);
1565 } while (old != expected);
1566
1567 return true;
1568 }
1569
1570 static void client_close(NBDClient *client, bool negotiated)
1571 {
1572 assert(qemu_in_main_thread());
1573
1574 WITH_QEMU_LOCK_GUARD(&client->lock) {
1575 if (client->closing) {
1576 return;
1577 }
1578
1579 client->closing = true;
1580 }
1581
1582 /* Force requests to finish. They will drop their own references,
1583 * then we'll close the socket and free the NBDClient.
1584 */
1585 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1586 NULL);
1587
1588 /* Also tell the client, so that they release their reference. */
1589 if (client->close_fn) {
1590 client->close_fn(client, negotiated);
1591 }
1592 }
1593
1594 /* Runs in export AioContext with client->lock held */
1595 static NBDRequestData *nbd_request_get(NBDClient *client)
1596 {
1597 NBDRequestData *req;
1598
1599 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1600 client->nb_requests++;
1601
1602 req = g_new0(NBDRequestData, 1);
1603 req->client = client;
1604 return req;
1605 }
1606
1607 /* Runs in export AioContext with client->lock held */
1608 static void nbd_request_put(NBDRequestData *req)
1609 {
1610 NBDClient *client = req->client;
1611
1612 if (req->data) {
1613 qemu_vfree(req->data);
1614 }
1615 g_free(req);
1616
1617 client->nb_requests--;
1618
1619 if (client->quiescing && client->nb_requests == 0) {
1620 aio_wait_kick();
1621 }
1622
1623 nbd_client_receive_next_request(client);
1624 }
1625
1626 static void blk_aio_attached(AioContext *ctx, void *opaque)
1627 {
1628 NBDExport *exp = opaque;
1629 NBDClient *client;
1630
1631 assert(qemu_in_main_thread());
1632
1633 trace_nbd_blk_aio_attached(exp->name, ctx);
1634
1635 exp->common.ctx = ctx;
1636
1637 QTAILQ_FOREACH(client, &exp->clients, next) {
1638 WITH_QEMU_LOCK_GUARD(&client->lock) {
1639 assert(client->nb_requests == 0);
1640 assert(client->recv_coroutine == NULL);
1641 assert(client->send_coroutine == NULL);
1642 }
1643 }
1644 }
1645
1646 static void blk_aio_detach(void *opaque)
1647 {
1648 NBDExport *exp = opaque;
1649
1650 assert(qemu_in_main_thread());
1651
1652 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1653
1654 exp->common.ctx = NULL;
1655 }
1656
1657 static void nbd_drained_begin(void *opaque)
1658 {
1659 NBDExport *exp = opaque;
1660 NBDClient *client;
1661
1662 assert(qemu_in_main_thread());
1663
1664 QTAILQ_FOREACH(client, &exp->clients, next) {
1665 WITH_QEMU_LOCK_GUARD(&client->lock) {
1666 client->quiescing = true;
1667 }
1668 }
1669 }
1670
1671 static void nbd_drained_end(void *opaque)
1672 {
1673 NBDExport *exp = opaque;
1674 NBDClient *client;
1675
1676 assert(qemu_in_main_thread());
1677
1678 QTAILQ_FOREACH(client, &exp->clients, next) {
1679 WITH_QEMU_LOCK_GUARD(&client->lock) {
1680 client->quiescing = false;
1681 nbd_client_receive_next_request(client);
1682 }
1683 }
1684 }
1685
1686 /* Runs in export AioContext */
1687 static void nbd_wake_read_bh(void *opaque)
1688 {
1689 NBDClient *client = opaque;
1690 qio_channel_wake_read(client->ioc);
1691 }
1692
1693 static bool nbd_drained_poll(void *opaque)
1694 {
1695 NBDExport *exp = opaque;
1696 NBDClient *client;
1697
1698 assert(qemu_in_main_thread());
1699
1700 QTAILQ_FOREACH(client, &exp->clients, next) {
1701 WITH_QEMU_LOCK_GUARD(&client->lock) {
1702 if (client->nb_requests != 0) {
1703 /*
1704 * If there's a coroutine waiting for a request on nbd_read_eof()
1705 * enter it here so we don't depend on the client to wake it up.
1706 *
1707 * Schedule a BH in the export AioContext to avoid missing the
1708 * wake up due to the race between qio_channel_wake_read() and
1709 * qio_channel_yield().
1710 */
1711 if (client->recv_coroutine != NULL && client->read_yielding) {
1712 aio_bh_schedule_oneshot(nbd_export_aio_context(client->exp),
1713 nbd_wake_read_bh, client);
1714 }
1715
1716 return true;
1717 }
1718 }
1719 }
1720
1721 return false;
1722 }
1723
1724 static void nbd_eject_notifier(Notifier *n, void *data)
1725 {
1726 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1727
1728 assert(qemu_in_main_thread());
1729
1730 blk_exp_request_shutdown(&exp->common);
1731 }
1732
1733 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1734 {
1735 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1736 assert(exp->drv == &blk_exp_nbd);
1737 assert(nbd_exp->eject_notifier_blk == NULL);
1738
1739 blk_ref(blk);
1740 nbd_exp->eject_notifier_blk = blk;
1741 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1742 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1743 }
1744
1745 static const BlockDevOps nbd_block_ops = {
1746 .drained_begin = nbd_drained_begin,
1747 .drained_end = nbd_drained_end,
1748 .drained_poll = nbd_drained_poll,
1749 };
1750
1751 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1752 Error **errp)
1753 {
1754 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1755 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1756 const char *name = arg->name ?: exp_args->node_name;
1757 BlockBackend *blk = blk_exp->blk;
1758 int64_t size;
1759 uint64_t perm, shared_perm;
1760 bool readonly = !exp_args->writable;
1761 BlockDirtyBitmapOrStrList *bitmaps;
1762 size_t i;
1763 int ret;
1764
1765 GLOBAL_STATE_CODE();
1766 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1767
1768 if (!nbd_server_is_running()) {
1769 error_setg(errp, "NBD server not running");
1770 return -EINVAL;
1771 }
1772
1773 if (strlen(name) > NBD_MAX_STRING_SIZE) {
1774 error_setg(errp, "export name '%s' too long", name);
1775 return -EINVAL;
1776 }
1777
1778 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1779 error_setg(errp, "description '%s' too long", arg->description);
1780 return -EINVAL;
1781 }
1782
1783 if (nbd_export_find(name)) {
1784 error_setg(errp, "NBD server already has export named '%s'", name);
1785 return -EEXIST;
1786 }
1787
1788 size = blk_getlength(blk);
1789 if (size < 0) {
1790 error_setg_errno(errp, -size,
1791 "Failed to determine the NBD export's length");
1792 return size;
1793 }
1794
1795 /* Don't allow resize while the NBD server is running, otherwise we don't
1796 * care what happens with the node. */
1797 blk_get_perm(blk, &perm, &shared_perm);
1798 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1799 if (ret < 0) {
1800 return ret;
1801 }
1802
1803 QTAILQ_INIT(&exp->clients);
1804 exp->name = g_strdup(name);
1805 exp->description = g_strdup(arg->description);
1806 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1807 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1808
1809 if (nbd_server_max_connections() != 1) {
1810 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1811 }
1812 if (readonly) {
1813 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1814 } else {
1815 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1816 NBD_FLAG_SEND_FAST_ZERO);
1817 }
1818 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1819
1820 bdrv_graph_rdlock_main_loop();
1821
1822 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1823 exp->nr_export_bitmaps++;
1824 }
1825 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1826 for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1827 i++, bitmaps = bitmaps->next)
1828 {
1829 const char *bitmap;
1830 BlockDriverState *bs = blk_bs(blk);
1831 BdrvDirtyBitmap *bm = NULL;
1832
1833 switch (bitmaps->value->type) {
1834 case QTYPE_QSTRING:
1835 bitmap = bitmaps->value->u.local;
1836 while (bs) {
1837 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1838 if (bm != NULL) {
1839 break;
1840 }
1841
1842 bs = bdrv_filter_or_cow_bs(bs);
1843 }
1844
1845 if (bm == NULL) {
1846 ret = -ENOENT;
1847 error_setg(errp, "Bitmap '%s' is not found",
1848 bitmaps->value->u.local);
1849 goto fail;
1850 }
1851
1852 if (readonly && bdrv_is_writable(bs) &&
1853 bdrv_dirty_bitmap_enabled(bm)) {
1854 ret = -EINVAL;
1855 error_setg(errp, "Enabled bitmap '%s' incompatible with "
1856 "readonly export", bitmap);
1857 goto fail;
1858 }
1859 break;
1860 case QTYPE_QDICT:
1861 bitmap = bitmaps->value->u.external.name;
1862 bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node,
1863 bitmap, NULL, errp);
1864 if (!bm) {
1865 ret = -ENOENT;
1866 goto fail;
1867 }
1868 break;
1869 default:
1870 abort();
1871 }
1872
1873 assert(bm);
1874
1875 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1876 ret = -EINVAL;
1877 goto fail;
1878 }
1879
1880 exp->export_bitmaps[i] = bm;
1881 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1882 }
1883
1884 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1885 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1886 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1887 }
1888
1889 exp->allocation_depth = arg->allocation_depth;
1890
1891 /*
1892 * We need to inhibit request queuing in the block layer to ensure we can
1893 * be properly quiesced when entering a drained section, as our coroutines
1894 * servicing pending requests might enter blk_pread().
1895 */
1896 blk_set_disable_request_queuing(blk, true);
1897
1898 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1899
1900 blk_set_dev_ops(blk, &nbd_block_ops, exp);
1901
1902 QTAILQ_INSERT_TAIL(&exports, exp, next);
1903
1904 bdrv_graph_rdunlock_main_loop();
1905
1906 return 0;
1907
1908 fail:
1909 bdrv_graph_rdunlock_main_loop();
1910 g_free(exp->export_bitmaps);
1911 g_free(exp->name);
1912 g_free(exp->description);
1913 return ret;
1914 }
1915
1916 NBDExport *nbd_export_find(const char *name)
1917 {
1918 NBDExport *exp;
1919 QTAILQ_FOREACH(exp, &exports, next) {
1920 if (strcmp(name, exp->name) == 0) {
1921 return exp;
1922 }
1923 }
1924
1925 return NULL;
1926 }
1927
1928 AioContext *
1929 nbd_export_aio_context(NBDExport *exp)
1930 {
1931 return exp->common.ctx;
1932 }
1933
1934 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1935 {
1936 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1937 NBDClient *client, *next;
1938
1939 blk_exp_ref(&exp->common);
1940 /*
1941 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1942 * close mode that stops advertising the export to new clients but
1943 * still permits existing clients to run to completion? Because of
1944 * that possibility, nbd_export_close() can be called more than
1945 * once on an export.
1946 */
1947 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1948 client_close(client, true);
1949 }
1950 if (exp->name) {
1951 g_free(exp->name);
1952 exp->name = NULL;
1953 QTAILQ_REMOVE(&exports, exp, next);
1954 }
1955 blk_exp_unref(&exp->common);
1956 }
1957
1958 static void nbd_export_delete(BlockExport *blk_exp)
1959 {
1960 size_t i;
1961 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1962
1963 assert(exp->name == NULL);
1964 assert(QTAILQ_EMPTY(&exp->clients));
1965
1966 g_free(exp->description);
1967 exp->description = NULL;
1968
1969 if (exp->eject_notifier_blk) {
1970 notifier_remove(&exp->eject_notifier);
1971 blk_unref(exp->eject_notifier_blk);
1972 }
1973 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1974 blk_aio_detach, exp);
1975 blk_set_disable_request_queuing(exp->common.blk, false);
1976
1977 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1978 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1979 }
1980 }
1981
1982 const BlockExportDriver blk_exp_nbd = {
1983 .type = BLOCK_EXPORT_TYPE_NBD,
1984 .instance_size = sizeof(NBDExport),
1985 .create = nbd_export_create,
1986 .delete = nbd_export_delete,
1987 .request_shutdown = nbd_export_request_shutdown,
1988 };
1989
1990 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1991 unsigned niov, Error **errp)
1992 {
1993 int ret;
1994
1995 g_assert(qemu_in_coroutine());
1996 qemu_co_mutex_lock(&client->send_lock);
1997 client->send_coroutine = qemu_coroutine_self();
1998
1999 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
2000
2001 client->send_coroutine = NULL;
2002 qemu_co_mutex_unlock(&client->send_lock);
2003
2004 return ret;
2005 }
2006
2007 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
2008 uint64_t cookie)
2009 {
2010 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
2011 stl_be_p(&reply->error, error);
2012 stq_be_p(&reply->cookie, cookie);
2013 }
2014
2015 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client,
2016 NBDRequest *request,
2017 uint32_t error,
2018 void *data,
2019 uint64_t len,
2020 Error **errp)
2021 {
2022 NBDSimpleReply reply;
2023 int nbd_err = system_errno_to_nbd_errno(error);
2024 struct iovec iov[] = {
2025 {.iov_base = &reply, .iov_len = sizeof(reply)},
2026 {.iov_base = data, .iov_len = len}
2027 };
2028
2029 assert(!len || !nbd_err);
2030 assert(len <= NBD_MAX_BUFFER_SIZE);
2031 assert(client->mode < NBD_MODE_STRUCTURED ||
2032 (client->mode == NBD_MODE_STRUCTURED &&
2033 request->type != NBD_CMD_READ));
2034 trace_nbd_co_send_simple_reply(request->cookie, nbd_err,
2035 nbd_err_lookup(nbd_err), len);
2036 set_be_simple_reply(&reply, nbd_err, request->cookie);
2037
2038 return nbd_co_send_iov(client, iov, 2, errp);
2039 }
2040
2041 /*
2042 * Prepare the header of a reply chunk for network transmission.
2043 *
2044 * On input, @iov is partially initialized: iov[0].iov_base must point
2045 * to an uninitialized NBDReply, while the remaining @niov elements
2046 * (if any) must be ready for transmission. This function then
2047 * populates iov[0] for transmission.
2048 */
2049 static inline void set_be_chunk(NBDClient *client, struct iovec *iov,
2050 size_t niov, uint16_t flags, uint16_t type,
2051 NBDRequest *request)
2052 {
2053 size_t i, length = 0;
2054
2055 for (i = 1; i < niov; i++) {
2056 length += iov[i].iov_len;
2057 }
2058 assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData));
2059
2060 if (client->mode >= NBD_MODE_EXTENDED) {
2061 NBDExtendedReplyChunk *chunk = iov->iov_base;
2062
2063 iov[0].iov_len = sizeof(*chunk);
2064 stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC);
2065 stw_be_p(&chunk->flags, flags);
2066 stw_be_p(&chunk->type, type);
2067 stq_be_p(&chunk->cookie, request->cookie);
2068 stq_be_p(&chunk->offset, request->from);
2069 stq_be_p(&chunk->length, length);
2070 } else {
2071 NBDStructuredReplyChunk *chunk = iov->iov_base;
2072
2073 iov[0].iov_len = sizeof(*chunk);
2074 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
2075 stw_be_p(&chunk->flags, flags);
2076 stw_be_p(&chunk->type, type);
2077 stq_be_p(&chunk->cookie, request->cookie);
2078 stl_be_p(&chunk->length, length);
2079 }
2080 }
2081
2082 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client,
2083 NBDRequest *request,
2084 Error **errp)
2085 {
2086 NBDReply hdr;
2087 struct iovec iov[] = {
2088 {.iov_base = &hdr},
2089 };
2090
2091 trace_nbd_co_send_chunk_done(request->cookie);
2092 set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE,
2093 NBD_REPLY_TYPE_NONE, request);
2094 return nbd_co_send_iov(client, iov, 1, errp);
2095 }
2096
2097 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client,
2098 NBDRequest *request,
2099 uint64_t offset,
2100 void *data,
2101 uint64_t size,
2102 bool final,
2103 Error **errp)
2104 {
2105 NBDReply hdr;
2106 NBDStructuredReadData chunk;
2107 struct iovec iov[] = {
2108 {.iov_base = &hdr},
2109 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2110 {.iov_base = data, .iov_len = size}
2111 };
2112
2113 assert(size && size <= NBD_MAX_BUFFER_SIZE);
2114 trace_nbd_co_send_chunk_read(request->cookie, offset, data, size);
2115 set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0,
2116 NBD_REPLY_TYPE_OFFSET_DATA, request);
2117 stq_be_p(&chunk.offset, offset);
2118
2119 return nbd_co_send_iov(client, iov, 3, errp);
2120 }
2121
2122 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client,
2123 NBDRequest *request,
2124 uint32_t error,
2125 const char *msg,
2126 Error **errp)
2127 {
2128 NBDReply hdr;
2129 NBDStructuredError chunk;
2130 int nbd_err = system_errno_to_nbd_errno(error);
2131 struct iovec iov[] = {
2132 {.iov_base = &hdr},
2133 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2134 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
2135 };
2136
2137 assert(nbd_err);
2138 trace_nbd_co_send_chunk_error(request->cookie, nbd_err,
2139 nbd_err_lookup(nbd_err), msg ? msg : "");
2140 set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE,
2141 NBD_REPLY_TYPE_ERROR, request);
2142 stl_be_p(&chunk.error, nbd_err);
2143 stw_be_p(&chunk.message_length, iov[2].iov_len);
2144
2145 return nbd_co_send_iov(client, iov, 3, errp);
2146 }
2147
2148 /* Do a sparse read and send the structured reply to the client.
2149 * Returns -errno if sending fails. blk_co_block_status_above() failure is
2150 * reported to the client, at which point this function succeeds.
2151 */
2152 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
2153 NBDRequest *request,
2154 uint64_t offset,
2155 uint8_t *data,
2156 uint64_t size,
2157 Error **errp)
2158 {
2159 int ret = 0;
2160 NBDExport *exp = client->exp;
2161 size_t progress = 0;
2162
2163 assert(size <= NBD_MAX_BUFFER_SIZE);
2164 while (progress < size) {
2165 int64_t pnum;
2166 int status = blk_co_block_status_above(exp->common.blk, NULL,
2167 offset + progress,
2168 size - progress, &pnum, NULL,
2169 NULL);
2170 bool final;
2171
2172 if (status < 0) {
2173 char *msg = g_strdup_printf("unable to check for holes: %s",
2174 strerror(-status));
2175
2176 ret = nbd_co_send_chunk_error(client, request, -status, msg, errp);
2177 g_free(msg);
2178 return ret;
2179 }
2180 assert(pnum && pnum <= size - progress);
2181 final = progress + pnum == size;
2182 if (status & BDRV_BLOCK_ZERO) {
2183 NBDReply hdr;
2184 NBDStructuredReadHole chunk;
2185 struct iovec iov[] = {
2186 {.iov_base = &hdr},
2187 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2188 };
2189
2190 trace_nbd_co_send_chunk_read_hole(request->cookie,
2191 offset + progress, pnum);
2192 set_be_chunk(client, iov, 2,
2193 final ? NBD_REPLY_FLAG_DONE : 0,
2194 NBD_REPLY_TYPE_OFFSET_HOLE, request);
2195 stq_be_p(&chunk.offset, offset + progress);
2196 stl_be_p(&chunk.length, pnum);
2197 ret = nbd_co_send_iov(client, iov, 2, errp);
2198 } else {
2199 ret = blk_co_pread(exp->common.blk, offset + progress, pnum,
2200 data + progress, 0);
2201 if (ret < 0) {
2202 error_setg_errno(errp, -ret, "reading from file failed");
2203 break;
2204 }
2205 ret = nbd_co_send_chunk_read(client, request, offset + progress,
2206 data + progress, pnum, final, errp);
2207 }
2208
2209 if (ret < 0) {
2210 break;
2211 }
2212 progress += pnum;
2213 }
2214 return ret;
2215 }
2216
2217 typedef struct NBDExtentArray {
2218 NBDExtent64 *extents;
2219 unsigned int nb_alloc;
2220 unsigned int count;
2221 uint64_t total_length;
2222 bool extended;
2223 bool can_add;
2224 bool converted_to_be;
2225 } NBDExtentArray;
2226
2227 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc,
2228 NBDMode mode)
2229 {
2230 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2231
2232 assert(mode >= NBD_MODE_STRUCTURED);
2233 ea->nb_alloc = nb_alloc;
2234 ea->extents = g_new(NBDExtent64, nb_alloc);
2235 ea->extended = mode >= NBD_MODE_EXTENDED;
2236 ea->can_add = true;
2237
2238 return ea;
2239 }
2240
2241 static void nbd_extent_array_free(NBDExtentArray *ea)
2242 {
2243 g_free(ea->extents);
2244 g_free(ea);
2245 }
2246 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free)
2247
2248 /* Further modifications of the array after conversion are abandoned */
2249 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2250 {
2251 int i;
2252
2253 assert(!ea->converted_to_be);
2254 assert(ea->extended);
2255 ea->can_add = false;
2256 ea->converted_to_be = true;
2257
2258 for (i = 0; i < ea->count; i++) {
2259 ea->extents[i].length = cpu_to_be64(ea->extents[i].length);
2260 ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags);
2261 }
2262 }
2263
2264 /* Further modifications of the array after conversion are abandoned */
2265 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea)
2266 {
2267 int i;
2268 NBDExtent32 *extents = g_new(NBDExtent32, ea->count);
2269
2270 assert(!ea->converted_to_be);
2271 assert(!ea->extended);
2272 ea->can_add = false;
2273 ea->converted_to_be = true;
2274
2275 for (i = 0; i < ea->count; i++) {
2276 assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX);
2277 extents[i].length = cpu_to_be32(ea->extents[i].length);
2278 extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2279 }
2280
2281 return extents;
2282 }
2283
2284 /*
2285 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2286 * return -1.
2287 * For safety, when returning -1 for the first time, .can_add is set to false,
2288 * and further calls to nbd_extent_array_add() will crash.
2289 * (this avoids the situation where a caller ignores failure to add one extent,
2290 * where adding another extent that would squash into the last array entry
2291 * would result in an incorrect range reported to the client)
2292 */
2293 static int nbd_extent_array_add(NBDExtentArray *ea,
2294 uint64_t length, uint32_t flags)
2295 {
2296 assert(ea->can_add);
2297
2298 if (!length) {
2299 return 0;
2300 }
2301 if (!ea->extended) {
2302 assert(length <= UINT32_MAX);
2303 }
2304
2305 /* Extend previous extent if flags are the same */
2306 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2307 uint64_t sum = length + ea->extents[ea->count - 1].length;
2308
2309 /*
2310 * sum cannot overflow: the block layer bounds image size at
2311 * 2^63, and ea->extents[].length comes from the block layer.
2312 */
2313 assert(sum >= length);
2314 if (sum <= UINT32_MAX || ea->extended) {
2315 ea->extents[ea->count - 1].length = sum;
2316 ea->total_length += length;
2317 return 0;
2318 }
2319 }
2320
2321 if (ea->count >= ea->nb_alloc) {
2322 ea->can_add = false;
2323 return -1;
2324 }
2325
2326 ea->total_length += length;
2327 ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags};
2328 ea->count++;
2329
2330 return 0;
2331 }
2332
2333 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk,
2334 uint64_t offset, uint64_t bytes,
2335 NBDExtentArray *ea)
2336 {
2337 while (bytes) {
2338 uint32_t flags;
2339 int64_t num;
2340 int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num,
2341 NULL, NULL);
2342
2343 if (ret < 0) {
2344 return ret;
2345 }
2346
2347 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2348 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2349
2350 if (nbd_extent_array_add(ea, num, flags) < 0) {
2351 return 0;
2352 }
2353
2354 offset += num;
2355 bytes -= num;
2356 }
2357
2358 return 0;
2359 }
2360
2361 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk,
2362 uint64_t offset, uint64_t bytes,
2363 NBDExtentArray *ea)
2364 {
2365 while (bytes) {
2366 int64_t num;
2367 int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes,
2368 &num);
2369
2370 if (ret < 0) {
2371 return ret;
2372 }
2373
2374 if (nbd_extent_array_add(ea, num, ret) < 0) {
2375 return 0;
2376 }
2377
2378 offset += num;
2379 bytes -= num;
2380 }
2381
2382 return 0;
2383 }
2384
2385 /*
2386 * nbd_co_send_extents
2387 *
2388 * @ea is converted to BE by the function
2389 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2390 */
2391 static int coroutine_fn
2392 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea,
2393 bool last, uint32_t context_id, Error **errp)
2394 {
2395 NBDReply hdr;
2396 NBDStructuredMeta meta;
2397 NBDExtendedMeta meta_ext;
2398 g_autofree NBDExtent32 *extents = NULL;
2399 uint16_t type;
2400 struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} };
2401
2402 if (client->mode >= NBD_MODE_EXTENDED) {
2403 type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT;
2404
2405 iov[1].iov_base = &meta_ext;
2406 iov[1].iov_len = sizeof(meta_ext);
2407 stl_be_p(&meta_ext.context_id, context_id);
2408 stl_be_p(&meta_ext.count, ea->count);
2409
2410 nbd_extent_array_convert_to_be(ea);
2411 iov[2].iov_base = ea->extents;
2412 iov[2].iov_len = ea->count * sizeof(ea->extents[0]);
2413 } else {
2414 type = NBD_REPLY_TYPE_BLOCK_STATUS;
2415
2416 iov[1].iov_base = &meta;
2417 iov[1].iov_len = sizeof(meta);
2418 stl_be_p(&meta.context_id, context_id);
2419
2420 extents = nbd_extent_array_convert_to_narrow(ea);
2421 iov[2].iov_base = extents;
2422 iov[2].iov_len = ea->count * sizeof(extents[0]);
2423 }
2424
2425 trace_nbd_co_send_extents(request->cookie, ea->count, context_id,
2426 ea->total_length, last);
2427 set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type,
2428 request);
2429
2430 return nbd_co_send_iov(client, iov, 3, errp);
2431 }
2432
2433 /* Get block status from the exported device and send it to the client */
2434 static int
2435 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request,
2436 BlockBackend *blk, uint64_t offset,
2437 uint64_t length, bool dont_fragment,
2438 bool last, uint32_t context_id,
2439 Error **errp)
2440 {
2441 int ret;
2442 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2443 g_autoptr(NBDExtentArray) ea =
2444 nbd_extent_array_new(nb_extents, client->mode);
2445
2446 if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2447 ret = blockstatus_to_extents(blk, offset, length, ea);
2448 } else {
2449 ret = blockalloc_to_extents(blk, offset, length, ea);
2450 }
2451 if (ret < 0) {
2452 return nbd_co_send_chunk_error(client, request, -ret,
2453 "can't get block status", errp);
2454 }
2455
2456 return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2457 }
2458
2459 /* Populate @ea from a dirty bitmap. */
2460 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2461 uint64_t offset, uint64_t length,
2462 NBDExtentArray *es)
2463 {
2464 int64_t start, dirty_start, dirty_count;
2465 int64_t end = offset + length;
2466 bool full = false;
2467 int64_t bound = es->extended ? INT64_MAX : INT32_MAX;
2468
2469 bdrv_dirty_bitmap_lock(bitmap);
2470
2471 for (start = offset;
2472 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound,
2473 &dirty_start, &dirty_count);
2474 start = dirty_start + dirty_count)
2475 {
2476 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2477 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2478 {
2479 full = true;
2480 break;
2481 }
2482 }
2483
2484 if (!full) {
2485 /* last non dirty extent, nothing to do if array is now full */
2486 (void) nbd_extent_array_add(es, end - start, 0);
2487 }
2488
2489 bdrv_dirty_bitmap_unlock(bitmap);
2490 }
2491
2492 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client,
2493 NBDRequest *request,
2494 BdrvDirtyBitmap *bitmap,
2495 uint64_t offset,
2496 uint64_t length, bool dont_fragment,
2497 bool last, uint32_t context_id,
2498 Error **errp)
2499 {
2500 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2501 g_autoptr(NBDExtentArray) ea =
2502 nbd_extent_array_new(nb_extents, client->mode);
2503
2504 bitmap_to_extents(bitmap, offset, length, ea);
2505
2506 return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2507 }
2508
2509 /*
2510 * nbd_co_block_status_payload_read
2511 * Called when a client wants a subset of negotiated contexts via a
2512 * BLOCK_STATUS payload. Check the payload for valid length and
2513 * contents. On success, return 0 with request updated to effective
2514 * length. If request was invalid but all payload consumed, return 0
2515 * with request->len and request->contexts->count set to 0 (which will
2516 * trigger an appropriate NBD_EINVAL response later on). Return
2517 * negative errno if the payload was not fully consumed.
2518 */
2519 static int
2520 nbd_co_block_status_payload_read(NBDClient *client, NBDRequest *request,
2521 Error **errp)
2522 {
2523 uint64_t payload_len = request->len;
2524 g_autofree char *buf = NULL;
2525 size_t count, i, nr_bitmaps;
2526 uint32_t id;
2527
2528 if (payload_len > NBD_MAX_BUFFER_SIZE) {
2529 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2530 request->len, NBD_MAX_BUFFER_SIZE);
2531 return -EINVAL;
2532 }
2533
2534 assert(client->contexts.exp == client->exp);
2535 nr_bitmaps = client->exp->nr_export_bitmaps;
2536 request->contexts = g_new0(NBDMetaContexts, 1);
2537 request->contexts->exp = client->exp;
2538
2539 if (payload_len % sizeof(uint32_t) ||
2540 payload_len < sizeof(NBDBlockStatusPayload) ||
2541 payload_len > (sizeof(NBDBlockStatusPayload) +
2542 sizeof(id) * client->contexts.count)) {
2543 goto skip;
2544 }
2545
2546 buf = g_malloc(payload_len);
2547 if (nbd_read(client->ioc, buf, payload_len,
2548 "CMD_BLOCK_STATUS data", errp) < 0) {
2549 return -EIO;
2550 }
2551 trace_nbd_co_receive_request_payload_received(request->cookie,
2552 payload_len);
2553 request->contexts->bitmaps = g_new0(bool, nr_bitmaps);
2554 count = (payload_len - sizeof(NBDBlockStatusPayload)) / sizeof(id);
2555 payload_len = 0;
2556
2557 for (i = 0; i < count; i++) {
2558 id = ldl_be_p(buf + sizeof(NBDBlockStatusPayload) + sizeof(id) * i);
2559 if (id == NBD_META_ID_BASE_ALLOCATION) {
2560 if (!client->contexts.base_allocation ||
2561 request->contexts->base_allocation) {
2562 goto skip;
2563 }
2564 request->contexts->base_allocation = true;
2565 } else if (id == NBD_META_ID_ALLOCATION_DEPTH) {
2566 if (!client->contexts.allocation_depth ||
2567 request->contexts->allocation_depth) {
2568 goto skip;
2569 }
2570 request->contexts->allocation_depth = true;
2571 } else {
2572 unsigned idx = id - NBD_META_ID_DIRTY_BITMAP;
2573
2574 if (idx >= nr_bitmaps || !client->contexts.bitmaps[idx] ||
2575 request->contexts->bitmaps[idx]) {
2576 goto skip;
2577 }
2578 request->contexts->bitmaps[idx] = true;
2579 }
2580 }
2581
2582 request->len = ldq_be_p(buf);
2583 request->contexts->count = count;
2584 return 0;
2585
2586 skip:
2587 trace_nbd_co_receive_block_status_payload_compliance(request->from,
2588 request->len);
2589 request->len = request->contexts->count = 0;
2590 return nbd_drop(client->ioc, payload_len, errp);
2591 }
2592
2593 /* nbd_co_receive_request
2594 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2595 * connection right away, -EAGAIN to indicate we were interrupted and the
2596 * channel should be quiesced, and any other negative value to report an error
2597 * to the client (although the caller may still need to disconnect after
2598 * reporting the error).
2599 */
2600 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req,
2601 NBDRequest *request,
2602 Error **errp)
2603 {
2604 NBDClient *client = req->client;
2605 bool extended_with_payload;
2606 bool check_length = false;
2607 bool check_rofs = false;
2608 bool allocate_buffer = false;
2609 bool payload_okay = false;
2610 uint64_t payload_len = 0;
2611 int valid_flags = NBD_CMD_FLAG_FUA;
2612 int ret;
2613
2614 g_assert(qemu_in_coroutine());
2615 ret = nbd_receive_request(client, request, errp);
2616 if (ret < 0) {
2617 return ret;
2618 }
2619
2620 trace_nbd_co_receive_request_decode_type(request->cookie, request->type,
2621 nbd_cmd_lookup(request->type));
2622 extended_with_payload = client->mode >= NBD_MODE_EXTENDED &&
2623 request->flags & NBD_CMD_FLAG_PAYLOAD_LEN;
2624 if (extended_with_payload) {
2625 payload_len = request->len;
2626 check_length = true;
2627 }
2628
2629 switch (request->type) {
2630 case NBD_CMD_DISC:
2631 /* Special case: we're going to disconnect without a reply,
2632 * whether or not flags, from, or len are bogus */
2633 req->complete = true;
2634 return -EIO;
2635
2636 case NBD_CMD_READ:
2637 if (client->mode >= NBD_MODE_STRUCTURED) {
2638 valid_flags |= NBD_CMD_FLAG_DF;
2639 }
2640 check_length = true;
2641 allocate_buffer = true;
2642 break;
2643
2644 case NBD_CMD_WRITE:
2645 if (client->mode >= NBD_MODE_EXTENDED) {
2646 if (!extended_with_payload) {
2647 /* The client is noncompliant. Trace it, but proceed. */
2648 trace_nbd_co_receive_ext_payload_compliance(request->from,
2649 request->len);
2650 }
2651 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2652 }
2653 payload_okay = true;
2654 payload_len = request->len;
2655 check_length = true;
2656 allocate_buffer = true;
2657 check_rofs = true;
2658 break;
2659
2660 case NBD_CMD_FLUSH:
2661 break;
2662
2663 case NBD_CMD_TRIM:
2664 check_rofs = true;
2665 break;
2666
2667 case NBD_CMD_CACHE:
2668 check_length = true;
2669 break;
2670
2671 case NBD_CMD_WRITE_ZEROES:
2672 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2673 check_rofs = true;
2674 break;
2675
2676 case NBD_CMD_BLOCK_STATUS:
2677 if (extended_with_payload) {
2678 ret = nbd_co_block_status_payload_read(client, request, errp);
2679 if (ret < 0) {
2680 return ret;
2681 }
2682 /* payload now consumed */
2683 check_length = false;
2684 payload_len = 0;
2685 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2686 } else {
2687 request->contexts = &client->contexts;
2688 }
2689 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2690 break;
2691
2692 default:
2693 /* Unrecognized, will fail later */
2694 ;
2695 }
2696
2697 /* Payload and buffer handling. */
2698 if (!payload_len) {
2699 req->complete = true;
2700 }
2701 if (check_length && request->len > NBD_MAX_BUFFER_SIZE) {
2702 /* READ, WRITE, CACHE */
2703 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2704 request->len, NBD_MAX_BUFFER_SIZE);
2705 return -EINVAL;
2706 }
2707 if (payload_len && !payload_okay) {
2708 /*
2709 * For now, we don't support payloads on other commands; but
2710 * we can keep the connection alive by ignoring the payload.
2711 * We will fail the command later with NBD_EINVAL for the use
2712 * of an unsupported flag (and not for access beyond bounds).
2713 */
2714 assert(request->type != NBD_CMD_WRITE);
2715 request->len = 0;
2716 }
2717 if (allocate_buffer) {
2718 /* READ, WRITE */
2719 req->data = blk_try_blockalign(client->exp->common.blk,
2720 request->len);
2721 if (req->data == NULL) {
2722 error_setg(errp, "No memory");
2723 return -ENOMEM;
2724 }
2725 }
2726 if (payload_len) {
2727 if (payload_okay) {
2728 /* WRITE */
2729 assert(req->data);
2730 ret = nbd_read(client->ioc, req->data, payload_len,
2731 "CMD_WRITE data", errp);
2732 } else {
2733 ret = nbd_drop(client->ioc, payload_len, errp);
2734 }
2735 if (ret < 0) {
2736 return -EIO;
2737 }
2738 req->complete = true;
2739 trace_nbd_co_receive_request_payload_received(request->cookie,
2740 payload_len);
2741 }
2742
2743 /* Sanity checks. */
2744 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) {
2745 /* WRITE, TRIM, WRITE_ZEROES */
2746 error_setg(errp, "Export is read-only");
2747 return -EROFS;
2748 }
2749 if (request->from > client->exp->size ||
2750 request->len > client->exp->size - request->from) {
2751 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64
2752 ", Size: %" PRIu64, request->from, request->len,
2753 client->exp->size);
2754 return (request->type == NBD_CMD_WRITE ||
2755 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2756 }
2757 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2758 client->check_align)) {
2759 /*
2760 * The block layer gracefully handles unaligned requests, but
2761 * it's still worth tracing client non-compliance
2762 */
2763 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2764 request->from,
2765 request->len,
2766 client->check_align);
2767 }
2768 if (request->flags & ~valid_flags) {
2769 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2770 nbd_cmd_lookup(request->type), request->flags);
2771 return -EINVAL;
2772 }
2773
2774 return 0;
2775 }
2776
2777 /* Send simple reply without a payload, or a structured error
2778 * @error_msg is ignored if @ret >= 0
2779 * Returns 0 if connection is still live, -errno on failure to talk to client
2780 */
2781 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2782 NBDRequest *request,
2783 int ret,
2784 const char *error_msg,
2785 Error **errp)
2786 {
2787 if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) {
2788 return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp);
2789 } else if (client->mode >= NBD_MODE_EXTENDED) {
2790 return nbd_co_send_chunk_done(client, request, errp);
2791 } else {
2792 return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0,
2793 NULL, 0, errp);
2794 }
2795 }
2796
2797 /* Handle NBD_CMD_READ request.
2798 * Return -errno if sending fails. Other errors are reported directly to the
2799 * client as an error reply. */
2800 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2801 uint8_t *data, Error **errp)
2802 {
2803 int ret;
2804 NBDExport *exp = client->exp;
2805
2806 assert(request->type == NBD_CMD_READ);
2807 assert(request->len <= NBD_MAX_BUFFER_SIZE);
2808
2809 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2810 if (request->flags & NBD_CMD_FLAG_FUA) {
2811 ret = blk_co_flush(exp->common.blk);
2812 if (ret < 0) {
2813 return nbd_send_generic_reply(client, request, ret,
2814 "flush failed", errp);
2815 }
2816 }
2817
2818 if (client->mode >= NBD_MODE_STRUCTURED &&
2819 !(request->flags & NBD_CMD_FLAG_DF) && request->len)
2820 {
2821 return nbd_co_send_sparse_read(client, request, request->from,
2822 data, request->len, errp);
2823 }
2824
2825 ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0);
2826 if (ret < 0) {
2827 return nbd_send_generic_reply(client, request, ret,
2828 "reading from file failed", errp);
2829 }
2830
2831 if (client->mode >= NBD_MODE_STRUCTURED) {
2832 if (request->len) {
2833 return nbd_co_send_chunk_read(client, request, request->from, data,
2834 request->len, true, errp);
2835 } else {
2836 return nbd_co_send_chunk_done(client, request, errp);
2837 }
2838 } else {
2839 return nbd_co_send_simple_reply(client, request, 0,
2840 data, request->len, errp);
2841 }
2842 }
2843
2844 /*
2845 * nbd_do_cmd_cache
2846 *
2847 * Handle NBD_CMD_CACHE request.
2848 * Return -errno if sending fails. Other errors are reported directly to the
2849 * client as an error reply.
2850 */
2851 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2852 Error **errp)
2853 {
2854 int ret;
2855 NBDExport *exp = client->exp;
2856
2857 assert(request->type == NBD_CMD_CACHE);
2858 assert(request->len <= NBD_MAX_BUFFER_SIZE);
2859
2860 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2861 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2862
2863 return nbd_send_generic_reply(client, request, ret,
2864 "caching data failed", errp);
2865 }
2866
2867 /* Handle NBD request.
2868 * Return -errno if sending fails. Other errors are reported directly to the
2869 * client as an error reply. */
2870 static coroutine_fn int nbd_handle_request(NBDClient *client,
2871 NBDRequest *request,
2872 uint8_t *data, Error **errp)
2873 {
2874 int ret;
2875 int flags;
2876 NBDExport *exp = client->exp;
2877 char *msg;
2878 size_t i;
2879
2880 switch (request->type) {
2881 case NBD_CMD_CACHE:
2882 return nbd_do_cmd_cache(client, request, errp);
2883
2884 case NBD_CMD_READ:
2885 return nbd_do_cmd_read(client, request, data, errp);
2886
2887 case NBD_CMD_WRITE:
2888 flags = 0;
2889 if (request->flags & NBD_CMD_FLAG_FUA) {
2890 flags |= BDRV_REQ_FUA;
2891 }
2892 assert(request->len <= NBD_MAX_BUFFER_SIZE);
2893 ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data,
2894 flags);
2895 return nbd_send_generic_reply(client, request, ret,
2896 "writing to file failed", errp);
2897
2898 case NBD_CMD_WRITE_ZEROES:
2899 flags = 0;
2900 if (request->flags & NBD_CMD_FLAG_FUA) {
2901 flags |= BDRV_REQ_FUA;
2902 }
2903 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2904 flags |= BDRV_REQ_MAY_UNMAP;
2905 }
2906 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2907 flags |= BDRV_REQ_NO_FALLBACK;
2908 }
2909 ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len,
2910 flags);
2911 return nbd_send_generic_reply(client, request, ret,
2912 "writing to file failed", errp);
2913
2914 case NBD_CMD_DISC:
2915 /* unreachable, thanks to special case in nbd_co_receive_request() */
2916 abort();
2917
2918 case NBD_CMD_FLUSH:
2919 ret = blk_co_flush(exp->common.blk);
2920 return nbd_send_generic_reply(client, request, ret,
2921 "flush failed", errp);
2922
2923 case NBD_CMD_TRIM:
2924 ret = blk_co_pdiscard(exp->common.blk, request->from, request->len);
2925 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2926 ret = blk_co_flush(exp->common.blk);
2927 }
2928 return nbd_send_generic_reply(client, request, ret,
2929 "discard failed", errp);
2930
2931 case NBD_CMD_BLOCK_STATUS:
2932 assert(request->contexts);
2933 assert(client->mode >= NBD_MODE_EXTENDED ||
2934 request->len <= UINT32_MAX);
2935 if (request->contexts->count) {
2936 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2937 int contexts_remaining = request->contexts->count;
2938
2939 if (!request->len) {
2940 return nbd_send_generic_reply(client, request, -EINVAL,
2941 "need non-zero length", errp);
2942 }
2943 if (request->contexts->base_allocation) {
2944 ret = nbd_co_send_block_status(client, request,
2945 exp->common.blk,
2946 request->from,
2947 request->len, dont_fragment,
2948 !--contexts_remaining,
2949 NBD_META_ID_BASE_ALLOCATION,
2950 errp);
2951 if (ret < 0) {
2952 return ret;
2953 }
2954 }
2955
2956 if (request->contexts->allocation_depth) {
2957 ret = nbd_co_send_block_status(client, request,
2958 exp->common.blk,
2959 request->from, request->len,
2960 dont_fragment,
2961 !--contexts_remaining,
2962 NBD_META_ID_ALLOCATION_DEPTH,
2963 errp);
2964 if (ret < 0) {
2965 return ret;
2966 }
2967 }
2968
2969 assert(request->contexts->exp == client->exp);
2970 for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2971 if (!request->contexts->bitmaps[i]) {
2972 continue;
2973 }
2974 ret = nbd_co_send_bitmap(client, request,
2975 client->exp->export_bitmaps[i],
2976 request->from, request->len,
2977 dont_fragment, !--contexts_remaining,
2978 NBD_META_ID_DIRTY_BITMAP + i, errp);
2979 if (ret < 0) {
2980 return ret;
2981 }
2982 }
2983
2984 assert(!contexts_remaining);
2985
2986 return 0;
2987 } else if (client->contexts.count) {
2988 return nbd_send_generic_reply(client, request, -EINVAL,
2989 "CMD_BLOCK_STATUS payload not valid",
2990 errp);
2991 } else {
2992 return nbd_send_generic_reply(client, request, -EINVAL,
2993 "CMD_BLOCK_STATUS not negotiated",
2994 errp);
2995 }
2996
2997 default:
2998 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2999 request->type);
3000 ret = nbd_send_generic_reply(client, request, -EINVAL, msg,
3001 errp);
3002 g_free(msg);
3003 return ret;
3004 }
3005 }
3006
3007 /* Owns a reference to the NBDClient passed as opaque. */
3008 static coroutine_fn void nbd_trip(void *opaque)
3009 {
3010 NBDRequestData *req = opaque;
3011 NBDClient *client = req->client;
3012 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
3013 int ret;
3014 Error *local_err = NULL;
3015
3016 /*
3017 * Note that nbd_client_put() and client_close() must be called from the
3018 * main loop thread. Use aio_co_reschedule_self() to switch AioContext
3019 * before calling these functions.
3020 */
3021
3022 trace_nbd_trip();
3023
3024 qemu_mutex_lock(&client->lock);
3025
3026 if (client->closing) {
3027 goto done;
3028 }
3029
3030 if (client->quiescing) {
3031 /*
3032 * We're switching between AIO contexts. Don't attempt to receive a new
3033 * request and kick the main context which may be waiting for us.
3034 */
3035 client->recv_coroutine = NULL;
3036 aio_wait_kick();
3037 goto done;
3038 }
3039
3040 /*
3041 * nbd_co_receive_request() returns -EAGAIN when nbd_drained_begin() has
3042 * set client->quiescing but by the time we get back nbd_drained_end() may
3043 * have already cleared client->quiescing. In that case we try again
3044 * because nothing else will spawn an nbd_trip() coroutine until we set
3045 * client->recv_coroutine = NULL further down.
3046 */
3047 do {
3048 assert(client->recv_coroutine == qemu_coroutine_self());
3049 qemu_mutex_unlock(&client->lock);
3050 ret = nbd_co_receive_request(req, &request, &local_err);
3051 qemu_mutex_lock(&client->lock);
3052 } while (ret == -EAGAIN && !client->quiescing);
3053
3054 client->recv_coroutine = NULL;
3055
3056 if (client->closing) {
3057 /*
3058 * The client may be closed when we are blocked in
3059 * nbd_co_receive_request()
3060 */
3061 goto done;
3062 }
3063
3064 if (ret == -EAGAIN) {
3065 goto done;
3066 }
3067
3068 nbd_client_receive_next_request(client);
3069
3070 if (ret == -EIO) {
3071 goto disconnect;
3072 }
3073
3074 qemu_mutex_unlock(&client->lock);
3075 qio_channel_set_cork(client->ioc, true);
3076
3077 if (ret < 0) {
3078 /* It wasn't -EIO, so, according to nbd_co_receive_request()
3079 * semantics, we should return the error to the client. */
3080 Error *export_err = local_err;
3081
3082 local_err = NULL;
3083 ret = nbd_send_generic_reply(client, &request, -EINVAL,
3084 error_get_pretty(export_err), &local_err);
3085 error_free(export_err);
3086 } else {
3087 ret = nbd_handle_request(client, &request, req->data, &local_err);
3088 }
3089 if (request.contexts && request.contexts != &client->contexts) {
3090 assert(request.type == NBD_CMD_BLOCK_STATUS);
3091 g_free(request.contexts->bitmaps);
3092 g_free(request.contexts);
3093 }
3094
3095 qio_channel_set_cork(client->ioc, false);
3096 qemu_mutex_lock(&client->lock);
3097
3098 if (ret < 0) {
3099 error_prepend(&local_err, "Failed to send reply: ");
3100 goto disconnect;
3101 }
3102
3103 /*
3104 * We must disconnect after NBD_CMD_WRITE or BLOCK_STATUS with
3105 * payload if we did not read the payload.
3106 */
3107 if (!req->complete) {
3108 error_setg(&local_err, "Request handling failed in intermediate state");
3109 goto disconnect;
3110 }
3111
3112 done:
3113 nbd_request_put(req);
3114
3115 qemu_mutex_unlock(&client->lock);
3116
3117 if (!nbd_client_put_nonzero(client)) {
3118 aio_co_reschedule_self(qemu_get_aio_context());
3119 nbd_client_put(client);
3120 }
3121 return;
3122
3123 disconnect:
3124 if (local_err) {
3125 error_reportf_err(local_err, "Disconnect client, due to: ");
3126 }
3127
3128 nbd_request_put(req);
3129 qemu_mutex_unlock(&client->lock);
3130
3131 aio_co_reschedule_self(qemu_get_aio_context());
3132 client_close(client, true);
3133 nbd_client_put(client);
3134 }
3135
3136 /*
3137 * Runs in export AioContext and main loop thread. Caller must hold
3138 * client->lock.
3139 */
3140 static void nbd_client_receive_next_request(NBDClient *client)
3141 {
3142 NBDRequestData *req;
3143
3144 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
3145 !client->quiescing) {
3146 nbd_client_get(client);
3147 req = nbd_request_get(client);
3148 client->recv_coroutine = qemu_coroutine_create(nbd_trip, req);
3149 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
3150 }
3151 }
3152
3153 static coroutine_fn void nbd_co_client_start(void *opaque)
3154 {
3155 NBDClient *client = opaque;
3156 Error *local_err = NULL;
3157
3158 qemu_co_mutex_init(&client->send_lock);
3159
3160 if (nbd_negotiate(client, &local_err)) {
3161 if (local_err) {
3162 error_report_err(local_err);
3163 }
3164 client_close(client, false);
3165 return;
3166 }
3167
3168 WITH_QEMU_LOCK_GUARD(&client->lock) {
3169 nbd_client_receive_next_request(client);
3170 }
3171 }
3172
3173 /*
3174 * Create a new client listener using the given channel @sioc.
3175 * Begin servicing it in a coroutine. When the connection closes, call
3176 * @close_fn with an indication of whether the client completed negotiation.
3177 */
3178 void nbd_client_new(QIOChannelSocket *sioc,
3179 QCryptoTLSCreds *tlscreds,
3180 const char *tlsauthz,
3181 void (*close_fn)(NBDClient *, bool))
3182 {
3183 NBDClient *client;
3184 Coroutine *co;
3185
3186 client = g_new0(NBDClient, 1);
3187 qemu_mutex_init(&client->lock);
3188 client->refcount = 1;
3189 client->tlscreds = tlscreds;
3190 if (tlscreds) {
3191 object_ref(OBJECT(client->tlscreds));
3192 }
3193 client->tlsauthz = g_strdup(tlsauthz);
3194 client->sioc = sioc;
3195 qio_channel_set_delay(QIO_CHANNEL(sioc), false);
3196 object_ref(OBJECT(client->sioc));
3197 client->ioc = QIO_CHANNEL(sioc);
3198 object_ref(OBJECT(client->ioc));
3199 client->close_fn = close_fn;
3200
3201 co = qemu_coroutine_create(nbd_co_client_start, client);
3202 qemu_coroutine_enter(co);
3203 }
AR7 emulation for QEMU