?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
usb-mtp: outlaw slashes in filenames
[qemu/ar7.git] / nbd / server.c
blobdc04513de705178e392167eb3052faa5b86400ae
1 /*
2 * Copyright (C) 2016-2018 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 *
5 * Network Block Device Server Side
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; under version 2 of the License.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
18 */
19
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "trace.h"
23 #include "nbd-internal.h"
24
25 #define NBD_META_ID_BASE_ALLOCATION 0
26 #define NBD_META_ID_DIRTY_BITMAP 1
27
28 /* NBD_MAX_BITMAP_EXTENTS: 1 mb of extents data. An empirical
29 * constant. If an increase is needed, note that the NBD protocol
30 * recommends no larger than 32 mb, so that the client won't consider
31 * the reply as a denial of service attack. */
32 #define NBD_MAX_BITMAP_EXTENTS (0x100000 / 8)
33
34 static int system_errno_to_nbd_errno(int err)
35 {
36 switch (err) {
37 case 0:
38 return NBD_SUCCESS;
39 case EPERM:
40 case EROFS:
41 return NBD_EPERM;
42 case EIO:
43 return NBD_EIO;
44 case ENOMEM:
45 return NBD_ENOMEM;
46 #ifdef EDQUOT
47 case EDQUOT:
48 #endif
49 case EFBIG:
50 case ENOSPC:
51 return NBD_ENOSPC;
52 case EOVERFLOW:
53 return NBD_EOVERFLOW;
54 case ESHUTDOWN:
55 return NBD_ESHUTDOWN;
56 case EINVAL:
57 default:
58 return NBD_EINVAL;
59 }
60 }
61
62 /* Definitions for opaque data types */
63
64 typedef struct NBDRequestData NBDRequestData;
65
66 struct NBDRequestData {
67 QSIMPLEQ_ENTRY(NBDRequestData) entry;
68 NBDClient *client;
69 uint8_t *data;
70 bool complete;
71 };
72
73 struct NBDExport {
74 int refcount;
75 void (*close)(NBDExport *exp);
76
77 BlockBackend *blk;
78 char *name;
79 char *description;
80 off_t dev_offset;
81 off_t size;
82 uint16_t nbdflags;
83 QTAILQ_HEAD(, NBDClient) clients;
84 QTAILQ_ENTRY(NBDExport) next;
85
86 AioContext *ctx;
87
88 BlockBackend *eject_notifier_blk;
89 Notifier eject_notifier;
90
91 BdrvDirtyBitmap *export_bitmap;
92 char *export_bitmap_context;
93 };
94
95 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
96
97 /* NBDExportMetaContexts represents a list of contexts to be exported,
98 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
99 * NBD_OPT_LIST_META_CONTEXT. */
100 typedef struct NBDExportMetaContexts {
101 NBDExport *exp;
102 bool valid; /* means that negotiation of the option finished without
103 errors */
104 bool base_allocation; /* export base:allocation context (block status) */
105 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
106 } NBDExportMetaContexts;
107
108 struct NBDClient {
109 int refcount;
110 void (*close_fn)(NBDClient *client, bool negotiated);
111
112 NBDExport *exp;
113 QCryptoTLSCreds *tlscreds;
114 char *tlsaclname;
115 QIOChannelSocket *sioc; /* The underlying data channel */
116 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
117
118 Coroutine *recv_coroutine;
119
120 CoMutex send_lock;
121 Coroutine *send_coroutine;
122
123 QTAILQ_ENTRY(NBDClient) next;
124 int nb_requests;
125 bool closing;
126
127 bool structured_reply;
128 NBDExportMetaContexts export_meta;
129
130 uint32_t opt; /* Current option being negotiated */
131 uint32_t optlen; /* remaining length of data in ioc for the option being
132 negotiated now */
133 };
134
135 static void nbd_client_receive_next_request(NBDClient *client);
136
137 /* Basic flow for negotiation
138
139 Server Client
140 Negotiate
141
142 or
143
144 Server Client
145 Negotiate #1
146 Option
147 Negotiate #2
148
149 ----
150
151 followed by
152
153 Server Client
154 Request
155 Response
156 Request
157 Response
158 ...
159 ...
160 Request (type == 2)
161
162 */
163
164 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
165 uint32_t type, uint32_t length)
166 {
167 stq_be_p(&rep->magic, NBD_REP_MAGIC);
168 stl_be_p(&rep->option, option);
169 stl_be_p(&rep->type, type);
170 stl_be_p(&rep->length, length);
171 }
172
173 /* Send a reply header, including length, but no payload.
174 * Return -errno on error, 0 on success. */
175 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
176 uint32_t len, Error **errp)
177 {
178 NBDOptionReply rep;
179
180 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
181 type, nbd_rep_lookup(type), len);
182
183 assert(len < NBD_MAX_BUFFER_SIZE);
184
185 set_be_option_rep(&rep, client->opt, type, len);
186 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
187 }
188
189 /* Send a reply header with default 0 length.
190 * Return -errno on error, 0 on success. */
191 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
192 Error **errp)
193 {
194 return nbd_negotiate_send_rep_len(client, type, 0, errp);
195 }
196
197 /* Send an error reply.
198 * Return -errno on error, 0 on success. */
199 static int GCC_FMT_ATTR(4, 0)
200 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
201 Error **errp, const char *fmt, va_list va)
202 {
203 char *msg;
204 int ret;
205 size_t len;
206
207 msg = g_strdup_vprintf(fmt, va);
208 len = strlen(msg);
209 assert(len < 4096);
210 trace_nbd_negotiate_send_rep_err(msg);
211 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
212 if (ret < 0) {
213 goto out;
214 }
215 if (nbd_write(client->ioc, msg, len, errp) < 0) {
216 error_prepend(errp, "write failed (error message): ");
217 ret = -EIO;
218 } else {
219 ret = 0;
220 }
221
222 out:
223 g_free(msg);
224 return ret;
225 }
226
227 /* Send an error reply.
228 * Return -errno on error, 0 on success. */
229 static int GCC_FMT_ATTR(4, 5)
230 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
231 Error **errp, const char *fmt, ...)
232 {
233 va_list va;
234 int ret;
235
236 va_start(va, fmt);
237 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
238 va_end(va);
239 return ret;
240 }
241
242 /* Drop remainder of the current option, and send a reply with the
243 * given error type and message. Return -errno on read or write
244 * failure; or 0 if connection is still live. */
245 static int GCC_FMT_ATTR(4, 0)
246 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
247 const char *fmt, va_list va)
248 {
249 int ret = nbd_drop(client->ioc, client->optlen, errp);
250
251 client->optlen = 0;
252 if (!ret) {
253 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
254 }
255 return ret;
256 }
257
258 static int GCC_FMT_ATTR(4, 5)
259 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
260 const char *fmt, ...)
261 {
262 int ret;
263 va_list va;
264
265 va_start(va, fmt);
266 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
267 va_end(va);
268
269 return ret;
270 }
271
272 static int GCC_FMT_ATTR(3, 4)
273 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
274 {
275 int ret;
276 va_list va;
277
278 va_start(va, fmt);
279 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
280 va_end(va);
281
282 return ret;
283 }
284
285 /* Read size bytes from the unparsed payload of the current option.
286 * Return -errno on I/O error, 0 if option was completely handled by
287 * sending a reply about inconsistent lengths, or 1 on success. */
288 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
289 Error **errp)
290 {
291 if (size > client->optlen) {
292 return nbd_opt_invalid(client, errp,
293 "Inconsistent lengths in option %s",
294 nbd_opt_lookup(client->opt));
295 }
296 client->optlen -= size;
297 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
298 }
299
300 /* Drop size bytes from the unparsed payload of the current option.
301 * Return -errno on I/O error, 0 if option was completely handled by
302 * sending a reply about inconsistent lengths, or 1 on success. */
303 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
304 {
305 if (size > client->optlen) {
306 return nbd_opt_invalid(client, errp,
307 "Inconsistent lengths in option %s",
308 nbd_opt_lookup(client->opt));
309 }
310 client->optlen -= size;
311 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
312 }
313
314 /* nbd_opt_read_name
315 *
316 * Read a string with the format:
317 * uint32_t len (<= NBD_MAX_NAME_SIZE)
318 * len bytes string (not 0-terminated)
319 *
320 * @name should be enough to store NBD_MAX_NAME_SIZE+1.
321 * If @length is non-null, it will be set to the actual string length.
322 *
323 * Return -errno on I/O error, 0 if option was completely handled by
324 * sending a reply about inconsistent lengths, or 1 on success.
325 */
326 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
327 Error **errp)
328 {
329 int ret;
330 uint32_t len;
331
332 ret = nbd_opt_read(client, &len, sizeof(len), errp);
333 if (ret <= 0) {
334 return ret;
335 }
336 len = cpu_to_be32(len);
337
338 if (len > NBD_MAX_NAME_SIZE) {
339 return nbd_opt_invalid(client, errp,
340 "Invalid name length: %" PRIu32, len);
341 }
342
343 ret = nbd_opt_read(client, name, len, errp);
344 if (ret <= 0) {
345 return ret;
346 }
347 name[len] = '\0';
348
349 if (length) {
350 *length = len;
351 }
352
353 return 1;
354 }
355
356 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
357 * Return -errno on error, 0 on success. */
358 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
359 Error **errp)
360 {
361 size_t name_len, desc_len;
362 uint32_t len;
363 const char *name = exp->name ? exp->name : "";
364 const char *desc = exp->description ? exp->description : "";
365 QIOChannel *ioc = client->ioc;
366 int ret;
367
368 trace_nbd_negotiate_send_rep_list(name, desc);
369 name_len = strlen(name);
370 desc_len = strlen(desc);
371 len = name_len + desc_len + sizeof(len);
372 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
373 if (ret < 0) {
374 return ret;
375 }
376
377 len = cpu_to_be32(name_len);
378 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
379 error_prepend(errp, "write failed (name length): ");
380 return -EINVAL;
381 }
382
383 if (nbd_write(ioc, name, name_len, errp) < 0) {
384 error_prepend(errp, "write failed (name buffer): ");
385 return -EINVAL;
386 }
387
388 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
389 error_prepend(errp, "write failed (description buffer): ");
390 return -EINVAL;
391 }
392
393 return 0;
394 }
395
396 /* Process the NBD_OPT_LIST command, with a potential series of replies.
397 * Return -errno on error, 0 on success. */
398 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
399 {
400 NBDExport *exp;
401 assert(client->opt == NBD_OPT_LIST);
402
403 /* For each export, send a NBD_REP_SERVER reply. */
404 QTAILQ_FOREACH(exp, &exports, next) {
405 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
406 return -EINVAL;
407 }
408 }
409 /* Finish with a NBD_REP_ACK. */
410 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
411 }
412
413 static void nbd_check_meta_export(NBDClient *client)
414 {
415 client->export_meta.valid &= client->exp == client->export_meta.exp;
416 }
417
418 /* Send a reply to NBD_OPT_EXPORT_NAME.
419 * Return -errno on error, 0 on success. */
420 static int nbd_negotiate_handle_export_name(NBDClient *client,
421 uint16_t myflags, bool no_zeroes,
422 Error **errp)
423 {
424 char name[NBD_MAX_NAME_SIZE + 1];
425 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
426 size_t len;
427 int ret;
428
429 /* Client sends:
430 [20 .. xx] export name (length bytes)
431 Server replies:
432 [ 0 .. 7] size
433 [ 8 .. 9] export flags
434 [10 .. 133] reserved (0) [unless no_zeroes]
435 */
436 trace_nbd_negotiate_handle_export_name();
437 if (client->optlen >= sizeof(name)) {
438 error_setg(errp, "Bad length received");
439 return -EINVAL;
440 }
441 if (nbd_read(client->ioc, name, client->optlen, errp) < 0) {
442 error_prepend(errp, "read failed: ");
443 return -EIO;
444 }
445 name[client->optlen] = '\0';
446 client->optlen = 0;
447
448 trace_nbd_negotiate_handle_export_name_request(name);
449
450 client->exp = nbd_export_find(name);
451 if (!client->exp) {
452 error_setg(errp, "export not found");
453 return -EINVAL;
454 }
455
456 trace_nbd_negotiate_new_style_size_flags(client->exp->size,
457 client->exp->nbdflags | myflags);
458 stq_be_p(buf, client->exp->size);
459 stw_be_p(buf + 8, client->exp->nbdflags | myflags);
460 len = no_zeroes ? 10 : sizeof(buf);
461 ret = nbd_write(client->ioc, buf, len, errp);
462 if (ret < 0) {
463 error_prepend(errp, "write failed: ");
464 return ret;
465 }
466
467 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
468 nbd_export_get(client->exp);
469 nbd_check_meta_export(client);
470
471 return 0;
472 }
473
474 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
475 * The buffer does NOT include the info type prefix.
476 * Return -errno on error, 0 if ready to send more. */
477 static int nbd_negotiate_send_info(NBDClient *client,
478 uint16_t info, uint32_t length, void *buf,
479 Error **errp)
480 {
481 int rc;
482
483 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
484 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
485 sizeof(info) + length, errp);
486 if (rc < 0) {
487 return rc;
488 }
489 info = cpu_to_be16(info);
490 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
491 return -EIO;
492 }
493 if (nbd_write(client->ioc, buf, length, errp) < 0) {
494 return -EIO;
495 }
496 return 0;
497 }
498
499 /* nbd_reject_length: Handle any unexpected payload.
500 * @fatal requests that we quit talking to the client, even if we are able
501 * to successfully send an error reply.
502 * Return:
503 * -errno transmission error occurred or @fatal was requested, errp is set
504 * 0 error message successfully sent to client, errp is not set
505 */
506 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
507 {
508 int ret;
509
510 assert(client->optlen);
511 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
512 nbd_opt_lookup(client->opt));
513 if (fatal && !ret) {
514 error_setg(errp, "option '%s' has unexpected length",
515 nbd_opt_lookup(client->opt));
516 return -EINVAL;
517 }
518 return ret;
519 }
520
521 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
522 * Return -errno on error, 0 if ready for next option, and 1 to move
523 * into transmission phase. */
524 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
525 Error **errp)
526 {
527 int rc;
528 char name[NBD_MAX_NAME_SIZE + 1];
529 NBDExport *exp;
530 uint16_t requests;
531 uint16_t request;
532 uint32_t namelen;
533 bool sendname = false;
534 bool blocksize = false;
535 uint32_t sizes[3];
536 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
537
538 /* Client sends:
539 4 bytes: L, name length (can be 0)
540 L bytes: export name
541 2 bytes: N, number of requests (can be 0)
542 N * 2 bytes: N requests
543 */
544 rc = nbd_opt_read_name(client, name, &namelen, errp);
545 if (rc <= 0) {
546 return rc;
547 }
548 trace_nbd_negotiate_handle_export_name_request(name);
549
550 rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
551 if (rc <= 0) {
552 return rc;
553 }
554 requests = be16_to_cpu(requests);
555 trace_nbd_negotiate_handle_info_requests(requests);
556 while (requests--) {
557 rc = nbd_opt_read(client, &request, sizeof(request), errp);
558 if (rc <= 0) {
559 return rc;
560 }
561 request = be16_to_cpu(request);
562 trace_nbd_negotiate_handle_info_request(request,
563 nbd_info_lookup(request));
564 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
565 * everything else is either a request we don't know or
566 * something we send regardless of request */
567 switch (request) {
568 case NBD_INFO_NAME:
569 sendname = true;
570 break;
571 case NBD_INFO_BLOCK_SIZE:
572 blocksize = true;
573 break;
574 }
575 }
576 if (client->optlen) {
577 return nbd_reject_length(client, false, errp);
578 }
579
580 exp = nbd_export_find(name);
581 if (!exp) {
582 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
583 errp, "export '%s' not present",
584 name);
585 }
586
587 /* Don't bother sending NBD_INFO_NAME unless client requested it */
588 if (sendname) {
589 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
590 errp);
591 if (rc < 0) {
592 return rc;
593 }
594 }
595
596 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
597 * client request */
598 if (exp->description) {
599 size_t len = strlen(exp->description);
600
601 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
602 len, exp->description, errp);
603 if (rc < 0) {
604 return rc;
605 }
606 }
607
608 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
609 * according to whether the client requested it, and according to
610 * whether this is OPT_INFO or OPT_GO. */
611 /* minimum - 1 for back-compat, or 512 if client is new enough.
612 * TODO: consult blk_bs(blk)->bl.request_alignment? */
613 sizes[0] =
614 (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
615 /* preferred - Hard-code to 4096 for now.
616 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
617 sizes[1] = 4096;
618 /* maximum - At most 32M, but smaller as appropriate. */
619 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
620 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
621 sizes[0] = cpu_to_be32(sizes[0]);
622 sizes[1] = cpu_to_be32(sizes[1]);
623 sizes[2] = cpu_to_be32(sizes[2]);
624 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
625 sizeof(sizes), sizes, errp);
626 if (rc < 0) {
627 return rc;
628 }
629
630 /* Send NBD_INFO_EXPORT always */
631 trace_nbd_negotiate_new_style_size_flags(exp->size,
632 exp->nbdflags | myflags);
633 stq_be_p(buf, exp->size);
634 stw_be_p(buf + 8, exp->nbdflags | myflags);
635 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
636 sizeof(buf), buf, errp);
637 if (rc < 0) {
638 return rc;
639 }
640
641 /* If the client is just asking for NBD_OPT_INFO, but forgot to
642 * request block sizes, return an error.
643 * TODO: consult blk_bs(blk)->request_align, and only error if it
644 * is not 1? */
645 if (client->opt == NBD_OPT_INFO && !blocksize) {
646 return nbd_negotiate_send_rep_err(client,
647 NBD_REP_ERR_BLOCK_SIZE_REQD,
648 errp,
649 "request NBD_INFO_BLOCK_SIZE to "
650 "use this export");
651 }
652
653 /* Final reply */
654 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
655 if (rc < 0) {
656 return rc;
657 }
658
659 if (client->opt == NBD_OPT_GO) {
660 client->exp = exp;
661 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
662 nbd_export_get(client->exp);
663 nbd_check_meta_export(client);
664 rc = 1;
665 }
666 return rc;
667 }
668
669
670 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
671 * new channel for all further (now-encrypted) communication. */
672 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
673 Error **errp)
674 {
675 QIOChannel *ioc;
676 QIOChannelTLS *tioc;
677 struct NBDTLSHandshakeData data = { 0 };
678
679 assert(client->opt == NBD_OPT_STARTTLS);
680
681 trace_nbd_negotiate_handle_starttls();
682 ioc = client->ioc;
683
684 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
685 return NULL;
686 }
687
688 tioc = qio_channel_tls_new_server(ioc,
689 client->tlscreds,
690 client->tlsaclname,
691 errp);
692 if (!tioc) {
693 return NULL;
694 }
695
696 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
697 trace_nbd_negotiate_handle_starttls_handshake();
698 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
699 qio_channel_tls_handshake(tioc,
700 nbd_tls_handshake,
701 &data,
702 NULL,
703 NULL);
704
705 if (!data.complete) {
706 g_main_loop_run(data.loop);
707 }
708 g_main_loop_unref(data.loop);
709 if (data.error) {
710 object_unref(OBJECT(tioc));
711 error_propagate(errp, data.error);
712 return NULL;
713 }
714
715 return QIO_CHANNEL(tioc);
716 }
717
718 /* nbd_negotiate_send_meta_context
719 *
720 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
721 *
722 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
723 */
724 static int nbd_negotiate_send_meta_context(NBDClient *client,
725 const char *context,
726 uint32_t context_id,
727 Error **errp)
728 {
729 NBDOptionReplyMetaContext opt;
730 struct iovec iov[] = {
731 {.iov_base = &opt, .iov_len = sizeof(opt)},
732 {.iov_base = (void *)context, .iov_len = strlen(context)}
733 };
734
735 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
736 context_id = 0;
737 }
738
739 trace_nbd_negotiate_meta_query_reply(context, context_id);
740 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
741 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
742 stl_be_p(&opt.context_id, context_id);
743
744 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
745 }
746
747 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
748 * @match is never set to false.
749 *
750 * Return -errno on I/O error, 0 if option was completely handled by
751 * sending a reply about inconsistent lengths, or 1 on success.
752 *
753 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
754 * It only means that there are no errors.
755 */
756 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
757 Error **errp)
758 {
759 int ret;
760 char *query;
761 size_t len = strlen(pattern);
762
763 assert(len);
764
765 query = g_malloc(len);
766 ret = nbd_opt_read(client, query, len, errp);
767 if (ret <= 0) {
768 g_free(query);
769 return ret;
770 }
771
772 if (strncmp(query, pattern, len) == 0) {
773 trace_nbd_negotiate_meta_query_parse(pattern);
774 *match = true;
775 } else {
776 trace_nbd_negotiate_meta_query_skip("pattern not matched");
777 }
778 g_free(query);
779
780 return 1;
781 }
782
783 /*
784 * Read @len bytes, and set @match to true if they match @pattern, or if @len
785 * is 0 and the client is performing _LIST_. @match is never set to false.
786 *
787 * Return -errno on I/O error, 0 if option was completely handled by
788 * sending a reply about inconsistent lengths, or 1 on success.
789 *
790 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
791 * It only means that there are no errors.
792 */
793 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
794 uint32_t len, bool *match, Error **errp)
795 {
796 if (len == 0) {
797 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
798 *match = true;
799 }
800 trace_nbd_negotiate_meta_query_parse("empty");
801 return 1;
802 }
803
804 if (len != strlen(pattern)) {
805 trace_nbd_negotiate_meta_query_skip("different lengths");
806 return nbd_opt_skip(client, len, errp);
807 }
808
809 return nbd_meta_pattern(client, pattern, match, errp);
810 }
811
812 /* nbd_meta_base_query
813 *
814 * Handle queries to 'base' namespace. For now, only the base:allocation
815 * context is available. 'len' is the amount of text remaining to be read from
816 * the current name, after the 'base:' portion has been stripped.
817 *
818 * Return -errno on I/O error, 0 if option was completely handled by
819 * sending a reply about inconsistent lengths, or 1 on success.
820 */
821 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
822 uint32_t len, Error **errp)
823 {
824 return nbd_meta_empty_or_pattern(client, "allocation", len,
825 &meta->base_allocation, errp);
826 }
827
828 /* nbd_meta_bitmap_query
829 *
830 * Handle query to 'qemu:' namespace.
831 * @len is the amount of text remaining to be read from the current name, after
832 * the 'qemu:' portion has been stripped.
833 *
834 * Return -errno on I/O error, 0 if option was completely handled by
835 * sending a reply about inconsistent lengths, or 1 on success. */
836 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
837 uint32_t len, Error **errp)
838 {
839 bool dirty_bitmap = false;
840 size_t dirty_bitmap_len = strlen("dirty-bitmap:");
841 int ret;
842
843 if (!meta->exp->export_bitmap) {
844 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
845 return nbd_opt_skip(client, len, errp);
846 }
847
848 if (len == 0) {
849 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
850 meta->bitmap = true;
851 }
852 trace_nbd_negotiate_meta_query_parse("empty");
853 return 1;
854 }
855
856 if (len < dirty_bitmap_len) {
857 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
858 return nbd_opt_skip(client, len, errp);
859 }
860
861 len -= dirty_bitmap_len;
862 ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
863 if (ret <= 0) {
864 return ret;
865 }
866 if (!dirty_bitmap) {
867 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
868 return nbd_opt_skip(client, len, errp);
869 }
870
871 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
872
873 return nbd_meta_empty_or_pattern(
874 client, meta->exp->export_bitmap_context +
875 strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
876 }
877
878 /* nbd_negotiate_meta_query
879 *
880 * Parse namespace name and call corresponding function to parse body of the
881 * query.
882 *
883 * The only supported namespace now is 'base'.
884 *
885 * The function aims not wasting time and memory to read long unknown namespace
886 * names.
887 *
888 * Return -errno on I/O error, 0 if option was completely handled by
889 * sending a reply about inconsistent lengths, or 1 on success. */
890 static int nbd_negotiate_meta_query(NBDClient *client,
891 NBDExportMetaContexts *meta, Error **errp)
892 {
893 /*
894 * Both 'qemu' and 'base' namespaces have length = 5 including a
895 * colon. If another length namespace is later introduced, this
896 * should certainly be refactored.
897 */
898 int ret;
899 size_t ns_len = 5;
900 char ns[5];
901 uint32_t len;
902
903 ret = nbd_opt_read(client, &len, sizeof(len), errp);
904 if (ret <= 0) {
905 return ret;
906 }
907 len = cpu_to_be32(len);
908
909 if (len < ns_len) {
910 trace_nbd_negotiate_meta_query_skip("length too short");
911 return nbd_opt_skip(client, len, errp);
912 }
913
914 len -= ns_len;
915 ret = nbd_opt_read(client, ns, ns_len, errp);
916 if (ret <= 0) {
917 return ret;
918 }
919
920 if (!strncmp(ns, "base:", ns_len)) {
921 trace_nbd_negotiate_meta_query_parse("base:");
922 return nbd_meta_base_query(client, meta, len, errp);
923 } else if (!strncmp(ns, "qemu:", ns_len)) {
924 trace_nbd_negotiate_meta_query_parse("qemu:");
925 return nbd_meta_qemu_query(client, meta, len, errp);
926 }
927
928 trace_nbd_negotiate_meta_query_skip("unknown namespace");
929 return nbd_opt_skip(client, len, errp);
930 }
931
932 /* nbd_negotiate_meta_queries
933 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
934 *
935 * Return -errno on I/O error, or 0 if option was completely handled. */
936 static int nbd_negotiate_meta_queries(NBDClient *client,
937 NBDExportMetaContexts *meta, Error **errp)
938 {
939 int ret;
940 char export_name[NBD_MAX_NAME_SIZE + 1];
941 NBDExportMetaContexts local_meta;
942 uint32_t nb_queries;
943 int i;
944
945 if (!client->structured_reply) {
946 return nbd_opt_invalid(client, errp,
947 "request option '%s' when structured reply "
948 "is not negotiated",
949 nbd_opt_lookup(client->opt));
950 }
951
952 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
953 /* Only change the caller's meta on SET. */
954 meta = &local_meta;
955 }
956
957 memset(meta, 0, sizeof(*meta));
958
959 ret = nbd_opt_read_name(client, export_name, NULL, errp);
960 if (ret <= 0) {
961 return ret;
962 }
963
964 meta->exp = nbd_export_find(export_name);
965 if (meta->exp == NULL) {
966 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
967 "export '%s' not present", export_name);
968 }
969
970 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
971 if (ret <= 0) {
972 return ret;
973 }
974 nb_queries = cpu_to_be32(nb_queries);
975 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
976 export_name, nb_queries);
977
978 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
979 /* enable all known contexts */
980 meta->base_allocation = true;
981 } else {
982 for (i = 0; i < nb_queries; ++i) {
983 ret = nbd_negotiate_meta_query(client, meta, errp);
984 if (ret <= 0) {
985 return ret;
986 }
987 }
988 }
989
990 if (meta->base_allocation) {
991 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
992 NBD_META_ID_BASE_ALLOCATION,
993 errp);
994 if (ret < 0) {
995 return ret;
996 }
997 }
998
999 if (meta->bitmap) {
1000 ret = nbd_negotiate_send_meta_context(client,
1001 meta->exp->export_bitmap_context,
1002 NBD_META_ID_DIRTY_BITMAP,
1003 errp);
1004 if (ret < 0) {
1005 return ret;
1006 }
1007 }
1008
1009 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1010 if (ret == 0) {
1011 meta->valid = true;
1012 }
1013
1014 return ret;
1015 }
1016
1017 /* nbd_negotiate_options
1018 * Process all NBD_OPT_* client option commands, during fixed newstyle
1019 * negotiation.
1020 * Return:
1021 * -errno on error, errp is set
1022 * 0 on successful negotiation, errp is not set
1023 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1024 * errp is not set
1025 */
1026 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1027 Error **errp)
1028 {
1029 uint32_t flags;
1030 bool fixedNewstyle = false;
1031 bool no_zeroes = false;
1032
1033 /* Client sends:
1034 [ 0 .. 3] client flags
1035
1036 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1037 [ 0 .. 7] NBD_OPTS_MAGIC
1038 [ 8 .. 11] NBD option
1039 [12 .. 15] Data length
1040 ... Rest of request
1041
1042 [ 0 .. 7] NBD_OPTS_MAGIC
1043 [ 8 .. 11] Second NBD option
1044 [12 .. 15] Data length
1045 ... Rest of request
1046 */
1047
1048 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
1049 error_prepend(errp, "read failed: ");
1050 return -EIO;
1051 }
1052 flags = be32_to_cpu(flags);
1053 trace_nbd_negotiate_options_flags(flags);
1054 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1055 fixedNewstyle = true;
1056 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1057 }
1058 if (flags & NBD_FLAG_C_NO_ZEROES) {
1059 no_zeroes = true;
1060 flags &= ~NBD_FLAG_C_NO_ZEROES;
1061 }
1062 if (flags != 0) {
1063 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1064 return -EINVAL;
1065 }
1066
1067 while (1) {
1068 int ret;
1069 uint32_t option, length;
1070 uint64_t magic;
1071
1072 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
1073 error_prepend(errp, "read failed: ");
1074 return -EINVAL;
1075 }
1076 magic = be64_to_cpu(magic);
1077 trace_nbd_negotiate_options_check_magic(magic);
1078 if (magic != NBD_OPTS_MAGIC) {
1079 error_setg(errp, "Bad magic received");
1080 return -EINVAL;
1081 }
1082
1083 if (nbd_read(client->ioc, &option,
1084 sizeof(option), errp) < 0) {
1085 error_prepend(errp, "read failed: ");
1086 return -EINVAL;
1087 }
1088 option = be32_to_cpu(option);
1089 client->opt = option;
1090
1091 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
1092 error_prepend(errp, "read failed: ");
1093 return -EINVAL;
1094 }
1095 length = be32_to_cpu(length);
1096 assert(!client->optlen);
1097 client->optlen = length;
1098
1099 if (length > NBD_MAX_BUFFER_SIZE) {
1100 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1101 length, NBD_MAX_BUFFER_SIZE);
1102 return -EINVAL;
1103 }
1104
1105 trace_nbd_negotiate_options_check_option(option,
1106 nbd_opt_lookup(option));
1107 if (client->tlscreds &&
1108 client->ioc == (QIOChannel *)client->sioc) {
1109 QIOChannel *tioc;
1110 if (!fixedNewstyle) {
1111 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1112 return -EINVAL;
1113 }
1114 switch (option) {
1115 case NBD_OPT_STARTTLS:
1116 if (length) {
1117 /* Unconditionally drop the connection if the client
1118 * can't start a TLS negotiation correctly */
1119 return nbd_reject_length(client, true, errp);
1120 }
1121 tioc = nbd_negotiate_handle_starttls(client, errp);
1122 if (!tioc) {
1123 return -EIO;
1124 }
1125 ret = 0;
1126 object_unref(OBJECT(client->ioc));
1127 client->ioc = QIO_CHANNEL(tioc);
1128 break;
1129
1130 case NBD_OPT_EXPORT_NAME:
1131 /* No way to return an error to client, so drop connection */
1132 error_setg(errp, "Option 0x%x not permitted before TLS",
1133 option);
1134 return -EINVAL;
1135
1136 default:
1137 /* Let the client keep trying, unless they asked to
1138 * quit. Always try to give an error back to the
1139 * client; but when replying to OPT_ABORT, be aware
1140 * that the client may hang up before receiving the
1141 * error, in which case we are fine ignoring the
1142 * resulting EPIPE. */
1143 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1144 option == NBD_OPT_ABORT ? NULL : errp,
1145 "Option 0x%" PRIx32
1146 " not permitted before TLS", option);
1147 if (option == NBD_OPT_ABORT) {
1148 return 1;
1149 }
1150 break;
1151 }
1152 } else if (fixedNewstyle) {
1153 switch (option) {
1154 case NBD_OPT_LIST:
1155 if (length) {
1156 ret = nbd_reject_length(client, false, errp);
1157 } else {
1158 ret = nbd_negotiate_handle_list(client, errp);
1159 }
1160 break;
1161
1162 case NBD_OPT_ABORT:
1163 /* NBD spec says we must try to reply before
1164 * disconnecting, but that we must also tolerate
1165 * guests that don't wait for our reply. */
1166 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1167 return 1;
1168
1169 case NBD_OPT_EXPORT_NAME:
1170 return nbd_negotiate_handle_export_name(client,
1171 myflags, no_zeroes,
1172 errp);
1173
1174 case NBD_OPT_INFO:
1175 case NBD_OPT_GO:
1176 ret = nbd_negotiate_handle_info(client, myflags, errp);
1177 if (ret == 1) {
1178 assert(option == NBD_OPT_GO);
1179 return 0;
1180 }
1181 break;
1182
1183 case NBD_OPT_STARTTLS:
1184 if (length) {
1185 ret = nbd_reject_length(client, false, errp);
1186 } else if (client->tlscreds) {
1187 ret = nbd_negotiate_send_rep_err(client,
1188 NBD_REP_ERR_INVALID, errp,
1189 "TLS already enabled");
1190 } else {
1191 ret = nbd_negotiate_send_rep_err(client,
1192 NBD_REP_ERR_POLICY, errp,
1193 "TLS not configured");
1194 }
1195 break;
1196
1197 case NBD_OPT_STRUCTURED_REPLY:
1198 if (length) {
1199 ret = nbd_reject_length(client, false, errp);
1200 } else if (client->structured_reply) {
1201 ret = nbd_negotiate_send_rep_err(
1202 client, NBD_REP_ERR_INVALID, errp,
1203 "structured reply already negotiated");
1204 } else {
1205 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1206 client->structured_reply = true;
1207 myflags |= NBD_FLAG_SEND_DF;
1208 }
1209 break;
1210
1211 case NBD_OPT_LIST_META_CONTEXT:
1212 case NBD_OPT_SET_META_CONTEXT:
1213 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1214 errp);
1215 break;
1216
1217 default:
1218 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1219 "Unsupported option %" PRIu32 " (%s)",
1220 option, nbd_opt_lookup(option));
1221 break;
1222 }
1223 } else {
1224 /*
1225 * If broken new-style we should drop the connection
1226 * for anything except NBD_OPT_EXPORT_NAME
1227 */
1228 switch (option) {
1229 case NBD_OPT_EXPORT_NAME:
1230 return nbd_negotiate_handle_export_name(client,
1231 myflags, no_zeroes,
1232 errp);
1233
1234 default:
1235 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1236 option, nbd_opt_lookup(option));
1237 return -EINVAL;
1238 }
1239 }
1240 if (ret < 0) {
1241 return ret;
1242 }
1243 }
1244 }
1245
1246 /* nbd_negotiate
1247 * Return:
1248 * -errno on error, errp is set
1249 * 0 on successful negotiation, errp is not set
1250 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1251 * errp is not set
1252 */
1253 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1254 {
1255 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1256 int ret;
1257 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1258 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1259 NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1260
1261 /* Old style negotiation header, no room for options
1262 [ 0 .. 7] passwd ("NBDMAGIC")
1263 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1264 [16 .. 23] size
1265 [24 .. 27] export flags (zero-extended)
1266 [28 .. 151] reserved (0)
1267
1268 New style negotiation header, client can send options
1269 [ 0 .. 7] passwd ("NBDMAGIC")
1270 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1271 [16 .. 17] server flags (0)
1272 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1273 */
1274
1275 qio_channel_set_blocking(client->ioc, false, NULL);
1276
1277 trace_nbd_negotiate_begin();
1278 memcpy(buf, "NBDMAGIC", 8);
1279
1280 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1281 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1282
1283 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1284 error_prepend(errp, "write failed: ");
1285 return -EINVAL;
1286 }
1287 ret = nbd_negotiate_options(client, myflags, errp);
1288 if (ret != 0) {
1289 if (ret < 0) {
1290 error_prepend(errp, "option negotiation failed: ");
1291 }
1292 return ret;
1293 }
1294
1295 assert(!client->optlen);
1296 trace_nbd_negotiate_success();
1297
1298 return 0;
1299 }
1300
1301 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1302 Error **errp)
1303 {
1304 uint8_t buf[NBD_REQUEST_SIZE];
1305 uint32_t magic;
1306 int ret;
1307
1308 ret = nbd_read(ioc, buf, sizeof(buf), errp);
1309 if (ret < 0) {
1310 return ret;
1311 }
1312
1313 /* Request
1314 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1315 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1316 [ 6 .. 7] type (NBD_CMD_READ, ...)
1317 [ 8 .. 15] handle
1318 [16 .. 23] from
1319 [24 .. 27] len
1320 */
1321
1322 magic = ldl_be_p(buf);
1323 request->flags = lduw_be_p(buf + 4);
1324 request->type = lduw_be_p(buf + 6);
1325 request->handle = ldq_be_p(buf + 8);
1326 request->from = ldq_be_p(buf + 16);
1327 request->len = ldl_be_p(buf + 24);
1328
1329 trace_nbd_receive_request(magic, request->flags, request->type,
1330 request->from, request->len);
1331
1332 if (magic != NBD_REQUEST_MAGIC) {
1333 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1334 return -EINVAL;
1335 }
1336 return 0;
1337 }
1338
1339 #define MAX_NBD_REQUESTS 16
1340
1341 void nbd_client_get(NBDClient *client)
1342 {
1343 client->refcount++;
1344 }
1345
1346 void nbd_client_put(NBDClient *client)
1347 {
1348 if (--client->refcount == 0) {
1349 /* The last reference should be dropped by client->close,
1350 * which is called by client_close.
1351 */
1352 assert(client->closing);
1353
1354 qio_channel_detach_aio_context(client->ioc);
1355 object_unref(OBJECT(client->sioc));
1356 object_unref(OBJECT(client->ioc));
1357 if (client->tlscreds) {
1358 object_unref(OBJECT(client->tlscreds));
1359 }
1360 g_free(client->tlsaclname);
1361 if (client->exp) {
1362 QTAILQ_REMOVE(&client->exp->clients, client, next);
1363 nbd_export_put(client->exp);
1364 }
1365 g_free(client);
1366 }
1367 }
1368
1369 static void client_close(NBDClient *client, bool negotiated)
1370 {
1371 if (client->closing) {
1372 return;
1373 }
1374
1375 client->closing = true;
1376
1377 /* Force requests to finish. They will drop their own references,
1378 * then we'll close the socket and free the NBDClient.
1379 */
1380 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1381 NULL);
1382
1383 /* Also tell the client, so that they release their reference. */
1384 if (client->close_fn) {
1385 client->close_fn(client, negotiated);
1386 }
1387 }
1388
1389 static NBDRequestData *nbd_request_get(NBDClient *client)
1390 {
1391 NBDRequestData *req;
1392
1393 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1394 client->nb_requests++;
1395
1396 req = g_new0(NBDRequestData, 1);
1397 nbd_client_get(client);
1398 req->client = client;
1399 return req;
1400 }
1401
1402 static void nbd_request_put(NBDRequestData *req)
1403 {
1404 NBDClient *client = req->client;
1405
1406 if (req->data) {
1407 qemu_vfree(req->data);
1408 }
1409 g_free(req);
1410
1411 client->nb_requests--;
1412 nbd_client_receive_next_request(client);
1413
1414 nbd_client_put(client);
1415 }
1416
1417 static void blk_aio_attached(AioContext *ctx, void *opaque)
1418 {
1419 NBDExport *exp = opaque;
1420 NBDClient *client;
1421
1422 trace_nbd_blk_aio_attached(exp->name, ctx);
1423
1424 exp->ctx = ctx;
1425
1426 QTAILQ_FOREACH(client, &exp->clients, next) {
1427 qio_channel_attach_aio_context(client->ioc, ctx);
1428 if (client->recv_coroutine) {
1429 aio_co_schedule(ctx, client->recv_coroutine);
1430 }
1431 if (client->send_coroutine) {
1432 aio_co_schedule(ctx, client->send_coroutine);
1433 }
1434 }
1435 }
1436
1437 static void blk_aio_detach(void *opaque)
1438 {
1439 NBDExport *exp = opaque;
1440 NBDClient *client;
1441
1442 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1443
1444 QTAILQ_FOREACH(client, &exp->clients, next) {
1445 qio_channel_detach_aio_context(client->ioc);
1446 }
1447
1448 exp->ctx = NULL;
1449 }
1450
1451 static void nbd_eject_notifier(Notifier *n, void *data)
1452 {
1453 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1454 nbd_export_close(exp);
1455 }
1456
1457 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1458 uint16_t nbdflags, void (*close)(NBDExport *),
1459 bool writethrough, BlockBackend *on_eject_blk,
1460 Error **errp)
1461 {
1462 AioContext *ctx;
1463 BlockBackend *blk;
1464 NBDExport *exp = g_new0(NBDExport, 1);
1465 uint64_t perm;
1466 int ret;
1467
1468 /*
1469 * NBD exports are used for non-shared storage migration. Make sure
1470 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1471 * access since the export could be available before migration handover.
1472 */
1473 ctx = bdrv_get_aio_context(bs);
1474 aio_context_acquire(ctx);
1475 bdrv_invalidate_cache(bs, NULL);
1476 aio_context_release(ctx);
1477
1478 /* Don't allow resize while the NBD server is running, otherwise we don't
1479 * care what happens with the node. */
1480 perm = BLK_PERM_CONSISTENT_READ;
1481 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1482 perm |= BLK_PERM_WRITE;
1483 }
1484 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1485 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1486 ret = blk_insert_bs(blk, bs, errp);
1487 if (ret < 0) {
1488 goto fail;
1489 }
1490 blk_set_enable_write_cache(blk, !writethrough);
1491
1492 exp->refcount = 1;
1493 QTAILQ_INIT(&exp->clients);
1494 exp->blk = blk;
1495 exp->dev_offset = dev_offset;
1496 exp->nbdflags = nbdflags;
1497 exp->size = size < 0 ? blk_getlength(blk) : size;
1498 if (exp->size < 0) {
1499 error_setg_errno(errp, -exp->size,
1500 "Failed to determine the NBD export's length");
1501 goto fail;
1502 }
1503 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1504
1505 exp->close = close;
1506 exp->ctx = blk_get_aio_context(blk);
1507 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1508
1509 if (on_eject_blk) {
1510 blk_ref(on_eject_blk);
1511 exp->eject_notifier_blk = on_eject_blk;
1512 exp->eject_notifier.notify = nbd_eject_notifier;
1513 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1514 }
1515 return exp;
1516
1517 fail:
1518 blk_unref(blk);
1519 g_free(exp);
1520 return NULL;
1521 }
1522
1523 NBDExport *nbd_export_find(const char *name)
1524 {
1525 NBDExport *exp;
1526 QTAILQ_FOREACH(exp, &exports, next) {
1527 if (strcmp(name, exp->name) == 0) {
1528 return exp;
1529 }
1530 }
1531
1532 return NULL;
1533 }
1534
1535 void nbd_export_set_name(NBDExport *exp, const char *name)
1536 {
1537 if (exp->name == name) {
1538 return;
1539 }
1540
1541 nbd_export_get(exp);
1542 if (exp->name != NULL) {
1543 g_free(exp->name);
1544 exp->name = NULL;
1545 QTAILQ_REMOVE(&exports, exp, next);
1546 nbd_export_put(exp);
1547 }
1548 if (name != NULL) {
1549 nbd_export_get(exp);
1550 exp->name = g_strdup(name);
1551 QTAILQ_INSERT_TAIL(&exports, exp, next);
1552 }
1553 nbd_export_put(exp);
1554 }
1555
1556 void nbd_export_set_description(NBDExport *exp, const char *description)
1557 {
1558 g_free(exp->description);
1559 exp->description = g_strdup(description);
1560 }
1561
1562 void nbd_export_close(NBDExport *exp)
1563 {
1564 NBDClient *client, *next;
1565
1566 nbd_export_get(exp);
1567 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1568 client_close(client, true);
1569 }
1570 nbd_export_set_name(exp, NULL);
1571 nbd_export_set_description(exp, NULL);
1572 nbd_export_put(exp);
1573 }
1574
1575 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1576 {
1577 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1578 nbd_export_close(exp);
1579 return;
1580 }
1581
1582 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1583
1584 error_setg(errp, "export '%s' still in use", exp->name);
1585 error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1586 }
1587
1588 void nbd_export_get(NBDExport *exp)
1589 {
1590 assert(exp->refcount > 0);
1591 exp->refcount++;
1592 }
1593
1594 void nbd_export_put(NBDExport *exp)
1595 {
1596 assert(exp->refcount > 0);
1597 if (exp->refcount == 1) {
1598 nbd_export_close(exp);
1599 }
1600
1601 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1602 * if someone calls nbd_export_put() on named export not through
1603 * nbd_export_set_name() when refcount is 1. So, let's assert that
1604 * it is > 0.
1605 */
1606 assert(exp->refcount > 0);
1607 if (--exp->refcount == 0) {
1608 assert(exp->name == NULL);
1609 assert(exp->description == NULL);
1610
1611 if (exp->close) {
1612 exp->close(exp);
1613 }
1614
1615 if (exp->blk) {
1616 if (exp->eject_notifier_blk) {
1617 notifier_remove(&exp->eject_notifier);
1618 blk_unref(exp->eject_notifier_blk);
1619 }
1620 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1621 blk_aio_detach, exp);
1622 blk_unref(exp->blk);
1623 exp->blk = NULL;
1624 }
1625
1626 if (exp->export_bitmap) {
1627 bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1628 g_free(exp->export_bitmap_context);
1629 }
1630
1631 g_free(exp);
1632 }
1633 }
1634
1635 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1636 {
1637 return exp->blk;
1638 }
1639
1640 void nbd_export_close_all(void)
1641 {
1642 NBDExport *exp, *next;
1643
1644 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1645 nbd_export_close(exp);
1646 }
1647 }
1648
1649 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1650 unsigned niov, Error **errp)
1651 {
1652 int ret;
1653
1654 g_assert(qemu_in_coroutine());
1655 qemu_co_mutex_lock(&client->send_lock);
1656 client->send_coroutine = qemu_coroutine_self();
1657
1658 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1659
1660 client->send_coroutine = NULL;
1661 qemu_co_mutex_unlock(&client->send_lock);
1662
1663 return ret;
1664 }
1665
1666 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1667 uint64_t handle)
1668 {
1669 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1670 stl_be_p(&reply->error, error);
1671 stq_be_p(&reply->handle, handle);
1672 }
1673
1674 static int nbd_co_send_simple_reply(NBDClient *client,
1675 uint64_t handle,
1676 uint32_t error,
1677 void *data,
1678 size_t len,
1679 Error **errp)
1680 {
1681 NBDSimpleReply reply;
1682 int nbd_err = system_errno_to_nbd_errno(error);
1683 struct iovec iov[] = {
1684 {.iov_base = &reply, .iov_len = sizeof(reply)},
1685 {.iov_base = data, .iov_len = len}
1686 };
1687
1688 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1689 len);
1690 set_be_simple_reply(&reply, nbd_err, handle);
1691
1692 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1693 }
1694
1695 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1696 uint16_t type, uint64_t handle, uint32_t length)
1697 {
1698 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1699 stw_be_p(&chunk->flags, flags);
1700 stw_be_p(&chunk->type, type);
1701 stq_be_p(&chunk->handle, handle);
1702 stl_be_p(&chunk->length, length);
1703 }
1704
1705 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1706 uint64_t handle,
1707 Error **errp)
1708 {
1709 NBDStructuredReplyChunk chunk;
1710 struct iovec iov[] = {
1711 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1712 };
1713
1714 trace_nbd_co_send_structured_done(handle);
1715 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1716
1717 return nbd_co_send_iov(client, iov, 1, errp);
1718 }
1719
1720 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1721 uint64_t handle,
1722 uint64_t offset,
1723 void *data,
1724 size_t size,
1725 bool final,
1726 Error **errp)
1727 {
1728 NBDStructuredReadData chunk;
1729 struct iovec iov[] = {
1730 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1731 {.iov_base = data, .iov_len = size}
1732 };
1733
1734 assert(size);
1735 trace_nbd_co_send_structured_read(handle, offset, data, size);
1736 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1737 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1738 sizeof(chunk) - sizeof(chunk.h) + size);
1739 stq_be_p(&chunk.offset, offset);
1740
1741 return nbd_co_send_iov(client, iov, 2, errp);
1742 }
1743
1744 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1745 uint64_t handle,
1746 uint32_t error,
1747 const char *msg,
1748 Error **errp)
1749 {
1750 NBDStructuredError chunk;
1751 int nbd_err = system_errno_to_nbd_errno(error);
1752 struct iovec iov[] = {
1753 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1754 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1755 };
1756
1757 assert(nbd_err);
1758 trace_nbd_co_send_structured_error(handle, nbd_err,
1759 nbd_err_lookup(nbd_err), msg ? msg : "");
1760 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1761 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1762 stl_be_p(&chunk.error, nbd_err);
1763 stw_be_p(&chunk.message_length, iov[1].iov_len);
1764
1765 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1766 }
1767
1768 /* Do a sparse read and send the structured reply to the client.
1769 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1770 * reported to the client, at which point this function succeeds.
1771 */
1772 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1773 uint64_t handle,
1774 uint64_t offset,
1775 uint8_t *data,
1776 size_t size,
1777 Error **errp)
1778 {
1779 int ret = 0;
1780 NBDExport *exp = client->exp;
1781 size_t progress = 0;
1782
1783 while (progress < size) {
1784 int64_t pnum;
1785 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1786 offset + progress,
1787 size - progress, &pnum, NULL,
1788 NULL);
1789 bool final;
1790
1791 if (status < 0) {
1792 char *msg = g_strdup_printf("unable to check for holes: %s",
1793 strerror(-status));
1794
1795 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1796 errp);
1797 g_free(msg);
1798 return ret;
1799 }
1800 assert(pnum && pnum <= size - progress);
1801 final = progress + pnum == size;
1802 if (status & BDRV_BLOCK_ZERO) {
1803 NBDStructuredReadHole chunk;
1804 struct iovec iov[] = {
1805 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1806 };
1807
1808 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1809 pnum);
1810 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1811 NBD_REPLY_TYPE_OFFSET_HOLE,
1812 handle, sizeof(chunk) - sizeof(chunk.h));
1813 stq_be_p(&chunk.offset, offset + progress);
1814 stl_be_p(&chunk.length, pnum);
1815 ret = nbd_co_send_iov(client, iov, 1, errp);
1816 } else {
1817 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1818 data + progress, pnum);
1819 if (ret < 0) {
1820 error_setg_errno(errp, -ret, "reading from file failed");
1821 break;
1822 }
1823 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1824 data + progress, pnum, final,
1825 errp);
1826 }
1827
1828 if (ret < 0) {
1829 break;
1830 }
1831 progress += pnum;
1832 }
1833 return ret;
1834 }
1835
1836 /*
1837 * Populate @extents from block status. Update @bytes to be the actual
1838 * length encoded (which may be smaller than the original), and update
1839 * @nb_extents to the number of extents used.
1840 *
1841 * Returns zero on success and -errno on bdrv_block_status_above failure.
1842 */
1843 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1844 uint64_t *bytes, NBDExtent *extents,
1845 unsigned int *nb_extents)
1846 {
1847 uint64_t remaining_bytes = *bytes;
1848 NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1849 bool first_extent = true;
1850
1851 assert(*nb_extents);
1852 while (remaining_bytes) {
1853 uint32_t flags;
1854 int64_t num;
1855 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1856 &num, NULL, NULL);
1857
1858 if (ret < 0) {
1859 return ret;
1860 }
1861
1862 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1863 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1864 offset += num;
1865 remaining_bytes -= num;
1866
1867 if (first_extent) {
1868 extent->flags = flags;
1869 extent->length = num;
1870 first_extent = false;
1871 continue;
1872 }
1873
1874 if (flags == extent->flags) {
1875 /* extend current extent */
1876 extent->length += num;
1877 } else {
1878 if (extent + 1 == extents_end) {
1879 break;
1880 }
1881
1882 /* start new extent */
1883 extent++;
1884 extent->flags = flags;
1885 extent->length = num;
1886 }
1887 }
1888
1889 extents_end = extent + 1;
1890
1891 for (extent = extents; extent < extents_end; extent++) {
1892 extent->flags = cpu_to_be32(extent->flags);
1893 extent->length = cpu_to_be32(extent->length);
1894 }
1895
1896 *bytes -= remaining_bytes;
1897 *nb_extents = extents_end - extents;
1898
1899 return 0;
1900 }
1901
1902 /* nbd_co_send_extents
1903 *
1904 * @length is only for tracing purposes (and may be smaller or larger
1905 * than the client's original request). @last controls whether
1906 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1907 * big-endian format.
1908 */
1909 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1910 NBDExtent *extents, unsigned int nb_extents,
1911 uint64_t length, bool last,
1912 uint32_t context_id, Error **errp)
1913 {
1914 NBDStructuredMeta chunk;
1915
1916 struct iovec iov[] = {
1917 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1918 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1919 };
1920
1921 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1922 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1923 NBD_REPLY_TYPE_BLOCK_STATUS,
1924 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1925 stl_be_p(&chunk.context_id, context_id);
1926
1927 return nbd_co_send_iov(client, iov, 2, errp);
1928 }
1929
1930 /* Get block status from the exported device and send it to the client */
1931 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1932 BlockDriverState *bs, uint64_t offset,
1933 uint32_t length, bool dont_fragment,
1934 bool last, uint32_t context_id,
1935 Error **errp)
1936 {
1937 int ret;
1938 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1939 NBDExtent *extents = g_new(NBDExtent, nb_extents);
1940 uint64_t final_length = length;
1941
1942 ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1943 &nb_extents);
1944 if (ret < 0) {
1945 g_free(extents);
1946 return nbd_co_send_structured_error(
1947 client, handle, -ret, "can't get block status", errp);
1948 }
1949
1950 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1951 final_length, last, context_id, errp);
1952
1953 g_free(extents);
1954
1955 return ret;
1956 }
1957
1958 /*
1959 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1960 * final extent may exceed the original @length. Store in @length the
1961 * byte length encoded (which may be smaller or larger than the
1962 * original), and return the number of extents used.
1963 */
1964 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1965 uint64_t *length, NBDExtent *extents,
1966 unsigned int nb_extents,
1967 bool dont_fragment)
1968 {
1969 uint64_t begin = offset, end = offset;
1970 uint64_t overall_end = offset + *length;
1971 unsigned int i = 0;
1972 BdrvDirtyBitmapIter *it;
1973 bool dirty;
1974
1975 bdrv_dirty_bitmap_lock(bitmap);
1976
1977 it = bdrv_dirty_iter_new(bitmap);
1978 dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1979
1980 assert(begin < overall_end && nb_extents);
1981 while (begin < overall_end && i < nb_extents) {
1982 bool next_dirty = !dirty;
1983
1984 if (dirty) {
1985 end = bdrv_dirty_bitmap_next_zero(bitmap, begin);
1986 } else {
1987 bdrv_set_dirty_iter(it, begin);
1988 end = bdrv_dirty_iter_next(it);
1989 }
1990 if (end == -1 || end - begin > UINT32_MAX) {
1991 /* Cap to an aligned value < 4G beyond begin. */
1992 end = MIN(bdrv_dirty_bitmap_size(bitmap),
1993 begin + UINT32_MAX + 1 -
1994 bdrv_dirty_bitmap_granularity(bitmap));
1995 next_dirty = dirty;
1996 }
1997 if (dont_fragment && end > overall_end) {
1998 end = overall_end;
1999 }
2000
2001 extents[i].length = cpu_to_be32(end - begin);
2002 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2003 i++;
2004 begin = end;
2005 dirty = next_dirty;
2006 }
2007
2008 bdrv_dirty_iter_free(it);
2009
2010 bdrv_dirty_bitmap_unlock(bitmap);
2011
2012 assert(offset < end);
2013 *length = end - offset;
2014 return i;
2015 }
2016
2017 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2018 BdrvDirtyBitmap *bitmap, uint64_t offset,
2019 uint32_t length, bool dont_fragment, bool last,
2020 uint32_t context_id, Error **errp)
2021 {
2022 int ret;
2023 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2024 NBDExtent *extents = g_new(NBDExtent, nb_extents);
2025 uint64_t final_length = length;
2026
2027 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2028 nb_extents, dont_fragment);
2029
2030 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2031 final_length, last, context_id, errp);
2032
2033 g_free(extents);
2034
2035 return ret;
2036 }
2037
2038 /* nbd_co_receive_request
2039 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2040 * connection right away, and any other negative value to report an error to
2041 * the client (although the caller may still need to disconnect after reporting
2042 * the error).
2043 */
2044 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2045 Error **errp)
2046 {
2047 NBDClient *client = req->client;
2048 int valid_flags;
2049
2050 g_assert(qemu_in_coroutine());
2051 assert(client->recv_coroutine == qemu_coroutine_self());
2052 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2053 return -EIO;
2054 }
2055
2056 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2057 nbd_cmd_lookup(request->type));
2058
2059 if (request->type != NBD_CMD_WRITE) {
2060 /* No payload, we are ready to read the next request. */
2061 req->complete = true;
2062 }
2063
2064 if (request->type == NBD_CMD_DISC) {
2065 /* Special case: we're going to disconnect without a reply,
2066 * whether or not flags, from, or len are bogus */
2067 return -EIO;
2068 }
2069
2070 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2071 request->type == NBD_CMD_CACHE)
2072 {
2073 if (request->len > NBD_MAX_BUFFER_SIZE) {
2074 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2075 request->len, NBD_MAX_BUFFER_SIZE);
2076 return -EINVAL;
2077 }
2078
2079 req->data = blk_try_blockalign(client->exp->blk, request->len);
2080 if (req->data == NULL) {
2081 error_setg(errp, "No memory");
2082 return -ENOMEM;
2083 }
2084 }
2085 if (request->type == NBD_CMD_WRITE) {
2086 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
2087 error_prepend(errp, "reading from socket failed: ");
2088 return -EIO;
2089 }
2090 req->complete = true;
2091
2092 trace_nbd_co_receive_request_payload_received(request->handle,
2093 request->len);
2094 }
2095
2096 /* Sanity checks. */
2097 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2098 (request->type == NBD_CMD_WRITE ||
2099 request->type == NBD_CMD_WRITE_ZEROES ||
2100 request->type == NBD_CMD_TRIM)) {
2101 error_setg(errp, "Export is read-only");
2102 return -EROFS;
2103 }
2104 if (request->from > client->exp->size ||
2105 request->from + request->len > client->exp->size) {
2106 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2107 ", Size: %" PRIu64, request->from, request->len,
2108 (uint64_t)client->exp->size);
2109 return (request->type == NBD_CMD_WRITE ||
2110 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2111 }
2112 valid_flags = NBD_CMD_FLAG_FUA;
2113 if (request->type == NBD_CMD_READ && client->structured_reply) {
2114 valid_flags |= NBD_CMD_FLAG_DF;
2115 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2116 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2117 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2118 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2119 }
2120 if (request->flags & ~valid_flags) {
2121 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2122 nbd_cmd_lookup(request->type), request->flags);
2123 return -EINVAL;
2124 }
2125
2126 return 0;
2127 }
2128
2129 /* Send simple reply without a payload, or a structured error
2130 * @error_msg is ignored if @ret >= 0
2131 * Returns 0 if connection is still live, -errno on failure to talk to client
2132 */
2133 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2134 uint64_t handle,
2135 int ret,
2136 const char *error_msg,
2137 Error **errp)
2138 {
2139 if (client->structured_reply && ret < 0) {
2140 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2141 errp);
2142 } else {
2143 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2144 NULL, 0, errp);
2145 }
2146 }
2147
2148 /* Handle NBD_CMD_READ request.
2149 * Return -errno if sending fails. Other errors are reported directly to the
2150 * client as an error reply. */
2151 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2152 uint8_t *data, Error **errp)
2153 {
2154 int ret;
2155 NBDExport *exp = client->exp;
2156
2157 assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2158
2159 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2160 if (request->flags & NBD_CMD_FLAG_FUA) {
2161 ret = blk_co_flush(exp->blk);
2162 if (ret < 0) {
2163 return nbd_send_generic_reply(client, request->handle, ret,
2164 "flush failed", errp);
2165 }
2166 }
2167
2168 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2169 request->len && request->type != NBD_CMD_CACHE)
2170 {
2171 return nbd_co_send_sparse_read(client, request->handle, request->from,
2172 data, request->len, errp);
2173 }
2174
2175 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2176 request->len);
2177 if (ret < 0 || request->type == NBD_CMD_CACHE) {
2178 return nbd_send_generic_reply(client, request->handle, ret,
2179 "reading from file failed", errp);
2180 }
2181
2182 if (client->structured_reply) {
2183 if (request->len) {
2184 return nbd_co_send_structured_read(client, request->handle,
2185 request->from, data,
2186 request->len, true, errp);
2187 } else {
2188 return nbd_co_send_structured_done(client, request->handle, errp);
2189 }
2190 } else {
2191 return nbd_co_send_simple_reply(client, request->handle, 0,
2192 data, request->len, errp);
2193 }
2194 }
2195
2196 /* Handle NBD request.
2197 * Return -errno if sending fails. Other errors are reported directly to the
2198 * client as an error reply. */
2199 static coroutine_fn int nbd_handle_request(NBDClient *client,
2200 NBDRequest *request,
2201 uint8_t *data, Error **errp)
2202 {
2203 int ret;
2204 int flags;
2205 NBDExport *exp = client->exp;
2206 char *msg;
2207
2208 switch (request->type) {
2209 case NBD_CMD_READ:
2210 case NBD_CMD_CACHE:
2211 return nbd_do_cmd_read(client, request, data, errp);
2212
2213 case NBD_CMD_WRITE:
2214 flags = 0;
2215 if (request->flags & NBD_CMD_FLAG_FUA) {
2216 flags |= BDRV_REQ_FUA;
2217 }
2218 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2219 data, request->len, flags);
2220 return nbd_send_generic_reply(client, request->handle, ret,
2221 "writing to file failed", errp);
2222
2223 case NBD_CMD_WRITE_ZEROES:
2224 flags = 0;
2225 if (request->flags & NBD_CMD_FLAG_FUA) {
2226 flags |= BDRV_REQ_FUA;
2227 }
2228 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2229 flags |= BDRV_REQ_MAY_UNMAP;
2230 }
2231 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2232 request->len, flags);
2233 return nbd_send_generic_reply(client, request->handle, ret,
2234 "writing to file failed", errp);
2235
2236 case NBD_CMD_DISC:
2237 /* unreachable, thanks to special case in nbd_co_receive_request() */
2238 abort();
2239
2240 case NBD_CMD_FLUSH:
2241 ret = blk_co_flush(exp->blk);
2242 return nbd_send_generic_reply(client, request->handle, ret,
2243 "flush failed", errp);
2244
2245 case NBD_CMD_TRIM:
2246 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2247 request->len);
2248 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2249 ret = blk_co_flush(exp->blk);
2250 }
2251 return nbd_send_generic_reply(client, request->handle, ret,
2252 "discard failed", errp);
2253
2254 case NBD_CMD_BLOCK_STATUS:
2255 if (!request->len) {
2256 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2257 "need non-zero length", errp);
2258 }
2259 if (client->export_meta.valid &&
2260 (client->export_meta.base_allocation ||
2261 client->export_meta.bitmap))
2262 {
2263 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2264
2265 if (client->export_meta.base_allocation) {
2266 ret = nbd_co_send_block_status(client, request->handle,
2267 blk_bs(exp->blk), request->from,
2268 request->len, dont_fragment,
2269 !client->export_meta.bitmap,
2270 NBD_META_ID_BASE_ALLOCATION,
2271 errp);
2272 if (ret < 0) {
2273 return ret;
2274 }
2275 }
2276
2277 if (client->export_meta.bitmap) {
2278 ret = nbd_co_send_bitmap(client, request->handle,
2279 client->exp->export_bitmap,
2280 request->from, request->len,
2281 dont_fragment,
2282 true, NBD_META_ID_DIRTY_BITMAP, errp);
2283 if (ret < 0) {
2284 return ret;
2285 }
2286 }
2287
2288 return ret;
2289 } else {
2290 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2291 "CMD_BLOCK_STATUS not negotiated",
2292 errp);
2293 }
2294
2295 default:
2296 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2297 request->type);
2298 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2299 errp);
2300 g_free(msg);
2301 return ret;
2302 }
2303 }
2304
2305 /* Owns a reference to the NBDClient passed as opaque. */
2306 static coroutine_fn void nbd_trip(void *opaque)
2307 {
2308 NBDClient *client = opaque;
2309 NBDRequestData *req;
2310 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2311 int ret;
2312 Error *local_err = NULL;
2313
2314 trace_nbd_trip();
2315 if (client->closing) {
2316 nbd_client_put(client);
2317 return;
2318 }
2319
2320 req = nbd_request_get(client);
2321 ret = nbd_co_receive_request(req, &request, &local_err);
2322 client->recv_coroutine = NULL;
2323
2324 if (client->closing) {
2325 /*
2326 * The client may be closed when we are blocked in
2327 * nbd_co_receive_request()
2328 */
2329 goto done;
2330 }
2331
2332 nbd_client_receive_next_request(client);
2333 if (ret == -EIO) {
2334 goto disconnect;
2335 }
2336
2337 if (ret < 0) {
2338 /* It wans't -EIO, so, according to nbd_co_receive_request()
2339 * semantics, we should return the error to the client. */
2340 Error *export_err = local_err;
2341
2342 local_err = NULL;
2343 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2344 error_get_pretty(export_err), &local_err);
2345 error_free(export_err);
2346 } else {
2347 ret = nbd_handle_request(client, &request, req->data, &local_err);
2348 }
2349 if (ret < 0) {
2350 error_prepend(&local_err, "Failed to send reply: ");
2351 goto disconnect;
2352 }
2353
2354 /* We must disconnect after NBD_CMD_WRITE if we did not
2355 * read the payload.
2356 */
2357 if (!req->complete) {
2358 error_setg(&local_err, "Request handling failed in intermediate state");
2359 goto disconnect;
2360 }
2361
2362 done:
2363 nbd_request_put(req);
2364 nbd_client_put(client);
2365 return;
2366
2367 disconnect:
2368 if (local_err) {
2369 error_reportf_err(local_err, "Disconnect client, due to: ");
2370 }
2371 nbd_request_put(req);
2372 client_close(client, true);
2373 nbd_client_put(client);
2374 }
2375
2376 static void nbd_client_receive_next_request(NBDClient *client)
2377 {
2378 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2379 nbd_client_get(client);
2380 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2381 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2382 }
2383 }
2384
2385 static coroutine_fn void nbd_co_client_start(void *opaque)
2386 {
2387 NBDClient *client = opaque;
2388 Error *local_err = NULL;
2389
2390 qemu_co_mutex_init(&client->send_lock);
2391
2392 if (nbd_negotiate(client, &local_err)) {
2393 if (local_err) {
2394 error_report_err(local_err);
2395 }
2396 client_close(client, false);
2397 return;
2398 }
2399
2400 nbd_client_receive_next_request(client);
2401 }
2402
2403 /*
2404 * Create a new client listener using the given channel @sioc.
2405 * Begin servicing it in a coroutine. When the connection closes, call
2406 * @close_fn with an indication of whether the client completed negotiation.
2407 */
2408 void nbd_client_new(QIOChannelSocket *sioc,
2409 QCryptoTLSCreds *tlscreds,
2410 const char *tlsaclname,
2411 void (*close_fn)(NBDClient *, bool))
2412 {
2413 NBDClient *client;
2414 Coroutine *co;
2415
2416 client = g_new0(NBDClient, 1);
2417 client->refcount = 1;
2418 client->tlscreds = tlscreds;
2419 if (tlscreds) {
2420 object_ref(OBJECT(client->tlscreds));
2421 }
2422 client->tlsaclname = g_strdup(tlsaclname);
2423 client->sioc = sioc;
2424 object_ref(OBJECT(client->sioc));
2425 client->ioc = QIO_CHANNEL(sioc);
2426 object_ref(OBJECT(client->ioc));
2427 client->close_fn = close_fn;
2428
2429 co = qemu_coroutine_create(nbd_co_client_start, client);
2430 qemu_coroutine_enter(co);
2431 }
2432
2433 void nbd_export_bitmap(NBDExport *exp, const char *bitmap,
2434 const char *bitmap_export_name, Error **errp)
2435 {
2436 BdrvDirtyBitmap *bm = NULL;
2437 BlockDriverState *bs = blk_bs(exp->blk);
2438
2439 if (exp->export_bitmap) {
2440 error_setg(errp, "Export bitmap is already set");
2441 return;
2442 }
2443
2444 while (true) {
2445 bm = bdrv_find_dirty_bitmap(bs, bitmap);
2446 if (bm != NULL || bs->backing == NULL) {
2447 break;
2448 }
2449
2450 bs = bs->backing->bs;
2451 }
2452
2453 if (bm == NULL) {
2454 error_setg(errp, "Bitmap '%s' is not found", bitmap);
2455 return;
2456 }
2457
2458 if (bdrv_dirty_bitmap_enabled(bm)) {
2459 error_setg(errp, "Bitmap '%s' is enabled", bitmap);
2460 return;
2461 }
2462
2463 if (bdrv_dirty_bitmap_user_locked(bm)) {
2464 error_setg(errp, "Bitmap '%s' is in use", bitmap);
2465 return;
2466 }
2467
2468 bdrv_dirty_bitmap_set_qmp_locked(bm, true);
2469 exp->export_bitmap = bm;
2470 exp->export_bitmap_context =
2471 g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name);
2472 }
AR7 emulation for QEMU