1 /*
2 * Copyright (C) 2016-2020 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 *
5 * Network Block Device Server Side
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; under version 2 of the License.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
18 */
19
20 #include "qemu/osdep.h"
21
22 #include "block/export.h"
23 #include "qapi/error.h"
24 #include "qemu/queue.h"
25 #include "trace.h"
26 #include "nbd-internal.h"
27 #include "qemu/units.h"
28
29 #define NBD_META_ID_BASE_ALLOCATION 0
30 #define NBD_META_ID_ALLOCATION_DEPTH 1
31 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
32 #define NBD_META_ID_DIRTY_BITMAP 2
33
34 /*
35 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
36 * constant. If an increase is needed, note that the NBD protocol
37 * recommends no larger than 32 mb, so that the client won't consider
38 * the reply as a denial of service attack.
39 */
40 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
41
system_errno_to_nbd_errno(int err)42 static int system_errno_to_nbd_errno(int err)
43 {
44 switch (err) {
45 case 0:
46 return NBD_SUCCESS;
47 case EPERM:
48 case EROFS:
49 return NBD_EPERM;
50 case EIO:
51 return NBD_EIO;
52 case ENOMEM:
53 return NBD_ENOMEM;
54 #ifdef EDQUOT
55 case EDQUOT:
56 #endif
57 case EFBIG:
58 case ENOSPC:
59 return NBD_ENOSPC;
60 case EOVERFLOW:
61 return NBD_EOVERFLOW;
62 case ENOTSUP:
63 #if ENOTSUP != EOPNOTSUPP
64 case EOPNOTSUPP:
65 #endif
66 return NBD_ENOTSUP;
67 case ESHUTDOWN:
68 return NBD_ESHUTDOWN;
69 case EINVAL:
70 default:
71 return NBD_EINVAL;
72 }
73 }
74
75 /* Definitions for opaque data types */
76
77 typedef struct NBDRequestData NBDRequestData;
78
79 struct NBDRequestData {
80 QSIMPLEQ_ENTRY(NBDRequestData) entry;
81 NBDClient *client;
82 uint8_t *data;
83 bool complete;
84 };
85
86 struct NBDExport {
87 BlockExport common;
88
89 char *name;
90 char *description;
91 uint64_t size;
92 uint16_t nbdflags;
93 QTAILQ_HEAD(, NBDClient) clients;
94 QTAILQ_ENTRY(NBDExport) next;
95
96 BlockBackend *eject_notifier_blk;
97 Notifier eject_notifier;
98
99 bool allocation_depth;
100 BdrvDirtyBitmap **export_bitmaps;
101 size_t nr_export_bitmaps;
102 };
103
104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
105
106 /* NBDExportMetaContexts represents a list of contexts to be exported,
107 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
108 * NBD_OPT_LIST_META_CONTEXT. */
109 typedef struct NBDExportMetaContexts {
110 NBDExport *exp;
111 size_t count; /* number of negotiated contexts */
112 bool base_allocation; /* export base:allocation context (block status) */
113 bool allocation_depth; /* export qemu:allocation-depth */
114 bool *bitmaps; /*
115 * export qemu:dirty-bitmap:<export bitmap name>,
116 * sized by exp->nr_export_bitmaps
117 */
118 } NBDExportMetaContexts;
119
120 struct NBDClient {
121 int refcount;
122 void (*close_fn)(NBDClient *client, bool negotiated);
123
124 NBDExport *exp;
125 QCryptoTLSCreds *tlscreds;
126 char *tlsauthz;
127 QIOChannelSocket *sioc; /* The underlying data channel */
128 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
129
130 Coroutine *recv_coroutine;
131
132 CoMutex send_lock;
133 Coroutine *send_coroutine;
134
135 bool read_yielding;
136 bool quiescing;
137
138 QTAILQ_ENTRY(NBDClient) next;
139 int nb_requests;
140 bool closing;
141
142 uint32_t check_align; /* If non-zero, check for aligned client requests */
143
144 bool structured_reply;
145 NBDExportMetaContexts export_meta;
146
147 uint32_t opt; /* Current option being negotiated */
148 uint32_t optlen; /* remaining length of data in ioc for the option being
149 negotiated now */
150 };
151
152 static void nbd_client_receive_next_request(NBDClient *client);
153
154 /* Basic flow for negotiation
155
156 Server Client
157 Negotiate
158
159 or
160
161 Server Client
162 Negotiate #1
163 Option
164 Negotiate #2
165
166 ----
167
168 followed by
169
170 Server Client
171 Request
172 Response
173 Request
174 Response
175 ...
176 ...
177 Request (type == 2)
178
179 */
180
set_be_option_rep(NBDOptionReply * rep,uint32_t option,uint32_t type,uint32_t length)181 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
182 uint32_t type, uint32_t length)
183 {
184 stq_be_p(&rep->magic, NBD_REP_MAGIC);
185 stl_be_p(&rep->option, option);
186 stl_be_p(&rep->type, type);
187 stl_be_p(&rep->length, length);
188 }
189
190 /* Send a reply header, including length, but no payload.
191 * Return -errno on error, 0 on success. */
nbd_negotiate_send_rep_len(NBDClient * client,uint32_t type,uint32_t len,Error ** errp)192 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
193 uint32_t len, Error **errp)
194 {
195 NBDOptionReply rep;
196
197 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
198 type, nbd_rep_lookup(type), len);
199
200 assert(len < NBD_MAX_BUFFER_SIZE);
201
202 set_be_option_rep(&rep, client->opt, type, len);
203 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
204 }
205
206 /* Send a reply header with default 0 length.
207 * Return -errno on error, 0 on success. */
nbd_negotiate_send_rep(NBDClient * client,uint32_t type,Error ** errp)208 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
209 Error **errp)
210 {
211 return nbd_negotiate_send_rep_len(client, type, 0, errp);
212 }
213
214 /* Send an error reply.
215 * Return -errno on error, 0 on success. */
216 static int GCC_FMT_ATTR(4, 0)
nbd_negotiate_send_rep_verr(NBDClient * client,uint32_t type,Error ** errp,const char * fmt,va_list va)217 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
218 Error **errp, const char *fmt, va_list va)
219 {
220 ERRP_GUARD();
221 g_autofree char *msg = NULL;
222 int ret;
223 size_t len;
224
225 msg = g_strdup_vprintf(fmt, va);
226 len = strlen(msg);
227 assert(len < NBD_MAX_STRING_SIZE);
228 trace_nbd_negotiate_send_rep_err(msg);
229 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
230 if (ret < 0) {
231 return ret;
232 }
233 if (nbd_write(client->ioc, msg, len, errp) < 0) {
234 error_prepend(errp, "write failed (error message): ");
235 return -EIO;
236 }
237
238 return 0;
239 }
240
241 /*
242 * Return a malloc'd copy of @name suitable for use in an error reply.
243 */
244 static char *
nbd_sanitize_name(const char * name)245 nbd_sanitize_name(const char *name)
246 {
247 if (strnlen(name, 80) < 80) {
248 return g_strdup(name);
249 }
250 /* XXX Should we also try to sanitize any control characters? */
251 return g_strdup_printf("%.80s...", name);
252 }
253
254 /* Send an error reply.
255 * Return -errno on error, 0 on success. */
256 static int GCC_FMT_ATTR(4, 5)
nbd_negotiate_send_rep_err(NBDClient * client,uint32_t type,Error ** errp,const char * fmt,...)257 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
258 Error **errp, const char *fmt, ...)
259 {
260 va_list va;
261 int ret;
262
263 va_start(va, fmt);
264 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
265 va_end(va);
266 return ret;
267 }
268
269 /* Drop remainder of the current option, and send a reply with the
270 * given error type and message. Return -errno on read or write
271 * failure; or 0 if connection is still live. */
272 static int GCC_FMT_ATTR(4, 0)
nbd_opt_vdrop(NBDClient * client,uint32_t type,Error ** errp,const char * fmt,va_list va)273 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
274 const char *fmt, va_list va)
275 {
276 int ret = nbd_drop(client->ioc, client->optlen, errp);
277
278 client->optlen = 0;
279 if (!ret) {
280 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
281 }
282 return ret;
283 }
284
285 static int GCC_FMT_ATTR(4, 5)
nbd_opt_drop(NBDClient * client,uint32_t type,Error ** errp,const char * fmt,...)286 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
287 const char *fmt, ...)
288 {
289 int ret;
290 va_list va;
291
292 va_start(va, fmt);
293 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
294 va_end(va);
295
296 return ret;
297 }
298
299 static int GCC_FMT_ATTR(3, 4)
nbd_opt_invalid(NBDClient * client,Error ** errp,const char * fmt,...)300 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
301 {
302 int ret;
303 va_list va;
304
305 va_start(va, fmt);
306 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
307 va_end(va);
308
309 return ret;
310 }
311
312 /* Read size bytes from the unparsed payload of the current option.
313 * If @check_nul, require that no NUL bytes appear in buffer.
314 * Return -errno on I/O error, 0 if option was completely handled by
315 * sending a reply about inconsistent lengths, or 1 on success. */
nbd_opt_read(NBDClient * client,void * buffer,size_t size,bool check_nul,Error ** errp)316 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
317 bool check_nul, Error **errp)
318 {
319 if (size > client->optlen) {
320 return nbd_opt_invalid(client, errp,
321 "Inconsistent lengths in option %s",
322 nbd_opt_lookup(client->opt));
323 }
324 client->optlen -= size;
325 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
326 return -EIO;
327 }
328
329 if (check_nul && strnlen(buffer, size) != size) {
330 return nbd_opt_invalid(client, errp,
331 "Unexpected embedded NUL in option %s",
332 nbd_opt_lookup(client->opt));
333 }
334 return 1;
335 }
336
337 /* Drop size bytes from the unparsed payload of the current option.
338 * Return -errno on I/O error, 0 if option was completely handled by
339 * sending a reply about inconsistent lengths, or 1 on success. */
nbd_opt_skip(NBDClient * client,size_t size,Error ** errp)340 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
341 {
342 if (size > client->optlen) {
343 return nbd_opt_invalid(client, errp,
344 "Inconsistent lengths in option %s",
345 nbd_opt_lookup(client->opt));
346 }
347 client->optlen -= size;
348 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
349 }
350
351 /* nbd_opt_read_name
352 *
353 * Read a string with the format:
354 * uint32_t len (<= NBD_MAX_STRING_SIZE)
355 * len bytes string (not 0-terminated)
356 *
357 * On success, @name will be allocated.
358 * If @length is non-null, it will be set to the actual string length.
359 *
360 * Return -errno on I/O error, 0 if option was completely handled by
361 * sending a reply about inconsistent lengths, or 1 on success.
362 */
nbd_opt_read_name(NBDClient * client,char ** name,uint32_t * length,Error ** errp)363 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
364 Error **errp)
365 {
366 int ret;
367 uint32_t len;
368 g_autofree char *local_name = NULL;
369
370 *name = NULL;
371 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
372 if (ret <= 0) {
373 return ret;
374 }
375 len = cpu_to_be32(len);
376
377 if (len > NBD_MAX_STRING_SIZE) {
378 return nbd_opt_invalid(client, errp,
379 "Invalid name length: %" PRIu32, len);
380 }
381
382 local_name = g_malloc(len + 1);
383 ret = nbd_opt_read(client, local_name, len, true, errp);
384 if (ret <= 0) {
385 return ret;
386 }
387 local_name[len] = '\0';
388
389 if (length) {
390 *length = len;
391 }
392 *name = g_steal_pointer(&local_name);
393
394 return 1;
395 }
396
397 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
398 * Return -errno on error, 0 on success. */
nbd_negotiate_send_rep_list(NBDClient * client,NBDExport * exp,Error ** errp)399 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
400 Error **errp)
401 {
402 ERRP_GUARD();
403 size_t name_len, desc_len;
404 uint32_t len;
405 const char *name = exp->name ? exp->name : "";
406 const char *desc = exp->description ? exp->description : "";
407 QIOChannel *ioc = client->ioc;
408 int ret;
409
410 trace_nbd_negotiate_send_rep_list(name, desc);
411 name_len = strlen(name);
412 desc_len = strlen(desc);
413 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
414 len = name_len + desc_len + sizeof(len);
415 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
416 if (ret < 0) {
417 return ret;
418 }
419
420 len = cpu_to_be32(name_len);
421 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
422 error_prepend(errp, "write failed (name length): ");
423 return -EINVAL;
424 }
425
426 if (nbd_write(ioc, name, name_len, errp) < 0) {
427 error_prepend(errp, "write failed (name buffer): ");
428 return -EINVAL;
429 }
430
431 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
432 error_prepend(errp, "write failed (description buffer): ");
433 return -EINVAL;
434 }
435
436 return 0;
437 }
438
439 /* Process the NBD_OPT_LIST command, with a potential series of replies.
440 * Return -errno on error, 0 on success. */
nbd_negotiate_handle_list(NBDClient * client,Error ** errp)441 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
442 {
443 NBDExport *exp;
444 assert(client->opt == NBD_OPT_LIST);
445
446 /* For each export, send a NBD_REP_SERVER reply. */
447 QTAILQ_FOREACH(exp, &exports, next) {
448 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
449 return -EINVAL;
450 }
451 }
452 /* Finish with a NBD_REP_ACK. */
453 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
454 }
455
nbd_check_meta_export(NBDClient * client)456 static void nbd_check_meta_export(NBDClient *client)
457 {
458 if (client->exp != client->export_meta.exp) {
459 client->export_meta.count = 0;
460 }
461 }
462
463 /* Send a reply to NBD_OPT_EXPORT_NAME.
464 * Return -errno on error, 0 on success. */
nbd_negotiate_handle_export_name(NBDClient * client,bool no_zeroes,Error ** errp)465 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
466 Error **errp)
467 {
468 ERRP_GUARD();
469 g_autofree char *name = NULL;
470 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
471 size_t len;
472 int ret;
473 uint16_t myflags;
474
475 /* Client sends:
476 [20 .. xx] export name (length bytes)
477 Server replies:
478 [ 0 .. 7] size
479 [ 8 .. 9] export flags
480 [10 .. 133] reserved (0) [unless no_zeroes]
481 */
482 trace_nbd_negotiate_handle_export_name();
483 if (client->optlen > NBD_MAX_STRING_SIZE) {
484 error_setg(errp, "Bad length received");
485 return -EINVAL;
486 }
487 name = g_malloc(client->optlen + 1);
488 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
489 return -EIO;
490 }
491 name[client->optlen] = '\0';
492 client->optlen = 0;
493
494 trace_nbd_negotiate_handle_export_name_request(name);
495
496 client->exp = nbd_export_find(name);
497 if (!client->exp) {
498 error_setg(errp, "export not found");
499 return -EINVAL;
500 }
501
502 myflags = client->exp->nbdflags;
503 if (client->structured_reply) {
504 myflags |= NBD_FLAG_SEND_DF;
505 }
506 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
507 stq_be_p(buf, client->exp->size);
508 stw_be_p(buf + 8, myflags);
509 len = no_zeroes ? 10 : sizeof(buf);
510 ret = nbd_write(client->ioc, buf, len, errp);
511 if (ret < 0) {
512 error_prepend(errp, "write failed: ");
513 return ret;
514 }
515
516 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
517 blk_exp_ref(&client->exp->common);
518 nbd_check_meta_export(client);
519
520 return 0;
521 }
522
523 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
524 * The buffer does NOT include the info type prefix.
525 * Return -errno on error, 0 if ready to send more. */
nbd_negotiate_send_info(NBDClient * client,uint16_t info,uint32_t length,void * buf,Error ** errp)526 static int nbd_negotiate_send_info(NBDClient *client,
527 uint16_t info, uint32_t length, void *buf,
528 Error **errp)
529 {
530 int rc;
531
532 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
533 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
534 sizeof(info) + length, errp);
535 if (rc < 0) {
536 return rc;
537 }
538 info = cpu_to_be16(info);
539 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
540 return -EIO;
541 }
542 if (nbd_write(client->ioc, buf, length, errp) < 0) {
543 return -EIO;
544 }
545 return 0;
546 }
547
548 /* nbd_reject_length: Handle any unexpected payload.
549 * @fatal requests that we quit talking to the client, even if we are able
550 * to successfully send an error reply.
551 * Return:
552 * -errno transmission error occurred or @fatal was requested, errp is set
553 * 0 error message successfully sent to client, errp is not set
554 */
nbd_reject_length(NBDClient * client,bool fatal,Error ** errp)555 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
556 {
557 int ret;
558
559 assert(client->optlen);
560 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
561 nbd_opt_lookup(client->opt));
562 if (fatal && !ret) {
563 error_setg(errp, "option '%s' has unexpected length",
564 nbd_opt_lookup(client->opt));
565 return -EINVAL;
566 }
567 return ret;
568 }
569
570 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
571 * Return -errno on error, 0 if ready for next option, and 1 to move
572 * into transmission phase. */
nbd_negotiate_handle_info(NBDClient * client,Error ** errp)573 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
574 {
575 int rc;
576 g_autofree char *name = NULL;
577 NBDExport *exp;
578 uint16_t requests;
579 uint16_t request;
580 uint32_t namelen = 0;
581 bool sendname = false;
582 bool blocksize = false;
583 uint32_t sizes[3];
584 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
585 uint32_t check_align = 0;
586 uint16_t myflags;
587
588 /* Client sends:
589 4 bytes: L, name length (can be 0)
590 L bytes: export name
591 2 bytes: N, number of requests (can be 0)
592 N * 2 bytes: N requests
593 */
594 rc = nbd_opt_read_name(client, &name, &namelen, errp);
595 if (rc <= 0) {
596 return rc;
597 }
598 trace_nbd_negotiate_handle_export_name_request(name);
599
600 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
601 if (rc <= 0) {
602 return rc;
603 }
604 requests = be16_to_cpu(requests);
605 trace_nbd_negotiate_handle_info_requests(requests);
606 while (requests--) {
607 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
608 if (rc <= 0) {
609 return rc;
610 }
611 request = be16_to_cpu(request);
612 trace_nbd_negotiate_handle_info_request(request,
613 nbd_info_lookup(request));
614 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
615 * everything else is either a request we don't know or
616 * something we send regardless of request */
617 switch (request) {
618 case NBD_INFO_NAME:
619 sendname = true;
620 break;
621 case NBD_INFO_BLOCK_SIZE:
622 blocksize = true;
623 break;
624 }
625 }
626 if (client->optlen) {
627 return nbd_reject_length(client, false, errp);
628 }
629
630 exp = nbd_export_find(name);
631 if (!exp) {
632 g_autofree char *sane_name = nbd_sanitize_name(name);
633
634 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
635 errp, "export '%s' not present",
636 sane_name);
637 }
638
639 /* Don't bother sending NBD_INFO_NAME unless client requested it */
640 if (sendname) {
641 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
642 errp);
643 if (rc < 0) {
644 return rc;
645 }
646 }
647
648 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
649 * client request */
650 if (exp->description) {
651 size_t len = strlen(exp->description);
652
653 assert(len <= NBD_MAX_STRING_SIZE);
654 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
655 len, exp->description, errp);
656 if (rc < 0) {
657 return rc;
658 }
659 }
660
661 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
662 * according to whether the client requested it, and according to
663 * whether this is OPT_INFO or OPT_GO. */
664 /* minimum - 1 for back-compat, or actual if client will obey it. */
665 if (client->opt == NBD_OPT_INFO || blocksize) {
666 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
667 } else {
668 sizes[0] = 1;
669 }
670 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
671 /* preferred - Hard-code to 4096 for now.
672 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
673 sizes[1] = MAX(4096, sizes[0]);
674 /* maximum - At most 32M, but smaller as appropriate. */
675 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
676 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
677 sizes[0] = cpu_to_be32(sizes[0]);
678 sizes[1] = cpu_to_be32(sizes[1]);
679 sizes[2] = cpu_to_be32(sizes[2]);
680 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
681 sizeof(sizes), sizes, errp);
682 if (rc < 0) {
683 return rc;
684 }
685
686 /* Send NBD_INFO_EXPORT always */
687 myflags = exp->nbdflags;
688 if (client->structured_reply) {
689 myflags |= NBD_FLAG_SEND_DF;
690 }
691 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
692 stq_be_p(buf, exp->size);
693 stw_be_p(buf + 8, myflags);
694 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
695 sizeof(buf), buf, errp);
696 if (rc < 0) {
697 return rc;
698 }
699
700 /*
701 * If the client is just asking for NBD_OPT_INFO, but forgot to
702 * request block sizes in a situation that would impact
703 * performance, then return an error. But for NBD_OPT_GO, we
704 * tolerate all clients, regardless of alignments.
705 */
706 if (client->opt == NBD_OPT_INFO && !blocksize &&
707 blk_get_request_alignment(exp->common.blk) > 1) {
708 return nbd_negotiate_send_rep_err(client,
709 NBD_REP_ERR_BLOCK_SIZE_REQD,
710 errp,
711 "request NBD_INFO_BLOCK_SIZE to "
712 "use this export");
713 }
714
715 /* Final reply */
716 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
717 if (rc < 0) {
718 return rc;
719 }
720
721 if (client->opt == NBD_OPT_GO) {
722 client->exp = exp;
723 client->check_align = check_align;
724 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
725 blk_exp_ref(&client->exp->common);
726 nbd_check_meta_export(client);
727 rc = 1;
728 }
729 return rc;
730 }
731
732
733 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
734 * new channel for all further (now-encrypted) communication. */
nbd_negotiate_handle_starttls(NBDClient * client,Error ** errp)735 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
736 Error **errp)
737 {
738 QIOChannel *ioc;
739 QIOChannelTLS *tioc;
740 struct NBDTLSHandshakeData data = { 0 };
741
742 assert(client->opt == NBD_OPT_STARTTLS);
743
744 trace_nbd_negotiate_handle_starttls();
745 ioc = client->ioc;
746
747 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
748 return NULL;
749 }
750
751 tioc = qio_channel_tls_new_server(ioc,
752 client->tlscreds,
753 client->tlsauthz,
754 errp);
755 if (!tioc) {
756 return NULL;
757 }
758
759 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
760 trace_nbd_negotiate_handle_starttls_handshake();
761 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
762 qio_channel_tls_handshake(tioc,
763 nbd_tls_handshake,
764 &data,
765 NULL,
766 NULL);
767
768 if (!data.complete) {
769 g_main_loop_run(data.loop);
770 }
771 g_main_loop_unref(data.loop);
772 if (data.error) {
773 object_unref(OBJECT(tioc));
774 error_propagate(errp, data.error);
775 return NULL;
776 }
777
778 return QIO_CHANNEL(tioc);
779 }
780
781 /* nbd_negotiate_send_meta_context
782 *
783 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
784 *
785 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
786 */
nbd_negotiate_send_meta_context(NBDClient * client,const char * context,uint32_t context_id,Error ** errp)787 static int nbd_negotiate_send_meta_context(NBDClient *client,
788 const char *context,
789 uint32_t context_id,
790 Error **errp)
791 {
792 NBDOptionReplyMetaContext opt;
793 struct iovec iov[] = {
794 {.iov_base = &opt, .iov_len = sizeof(opt)},
795 {.iov_base = (void *)context, .iov_len = strlen(context)}
796 };
797
798 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
799 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
800 context_id = 0;
801 }
802
803 trace_nbd_negotiate_meta_query_reply(context, context_id);
804 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
805 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
806 stl_be_p(&opt.context_id, context_id);
807
808 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
809 }
810
811 /*
812 * Return true if @query matches @pattern, or if @query is empty when
813 * the @client is performing _LIST_.
814 */
nbd_meta_empty_or_pattern(NBDClient * client,const char * pattern,const char * query)815 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
816 const char *query)
817 {
818 if (!*query) {
819 trace_nbd_negotiate_meta_query_parse("empty");
820 return client->opt == NBD_OPT_LIST_META_CONTEXT;
821 }
822 if (strcmp(query, pattern) == 0) {
823 trace_nbd_negotiate_meta_query_parse(pattern);
824 return true;
825 }
826 trace_nbd_negotiate_meta_query_skip("pattern not matched");
827 return false;
828 }
829
830 /*
831 * Return true and adjust @str in place if it begins with @prefix.
832 */
nbd_strshift(const char ** str,const char * prefix)833 static bool nbd_strshift(const char **str, const char *prefix)
834 {
835 size_t len = strlen(prefix);
836
837 if (strncmp(*str, prefix, len) == 0) {
838 *str += len;
839 return true;
840 }
841 return false;
842 }
843
844 /* nbd_meta_base_query
845 *
846 * Handle queries to 'base' namespace. For now, only the base:allocation
847 * context is available. Return true if @query has been handled.
848 */
nbd_meta_base_query(NBDClient * client,NBDExportMetaContexts * meta,const char * query)849 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
850 const char *query)
851 {
852 if (!nbd_strshift(&query, "base:")) {
853 return false;
854 }
855 trace_nbd_negotiate_meta_query_parse("base:");
856
857 if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
858 meta->base_allocation = true;
859 }
860 return true;
861 }
862
863 /* nbd_meta_qemu_query
864 *
865 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
866 * and qemu:allocation-depth contexts are available. Return true if @query
867 * has been handled.
868 */
nbd_meta_qemu_query(NBDClient * client,NBDExportMetaContexts * meta,const char * query)869 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
870 const char *query)
871 {
872 size_t i;
873
874 if (!nbd_strshift(&query, "qemu:")) {
875 return false;
876 }
877 trace_nbd_negotiate_meta_query_parse("qemu:");
878
879 if (!*query) {
880 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
881 meta->allocation_depth = meta->exp->allocation_depth;
882 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
883 }
884 trace_nbd_negotiate_meta_query_parse("empty");
885 return true;
886 }
887
888 if (strcmp(query, "allocation-depth") == 0) {
889 trace_nbd_negotiate_meta_query_parse("allocation-depth");
890 meta->allocation_depth = meta->exp->allocation_depth;
891 return true;
892 }
893
894 if (nbd_strshift(&query, "dirty-bitmap:")) {
895 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
896 if (!*query) {
897 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
898 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
899 }
900 trace_nbd_negotiate_meta_query_parse("empty");
901 return true;
902 }
903
904 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
905 const char *bm_name;
906
907 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
908 if (strcmp(bm_name, query) == 0) {
909 meta->bitmaps[i] = true;
910 trace_nbd_negotiate_meta_query_parse(query);
911 return true;
912 }
913 }
914 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
915 return true;
916 }
917
918 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
919 return true;
920 }
921
922 /* nbd_negotiate_meta_query
923 *
924 * Parse namespace name and call corresponding function to parse body of the
925 * query.
926 *
927 * The only supported namespaces are 'base' and 'qemu'.
928 *
929 * Return -errno on I/O error, 0 if option was completely handled by
930 * sending a reply about inconsistent lengths, or 1 on success. */
nbd_negotiate_meta_query(NBDClient * client,NBDExportMetaContexts * meta,Error ** errp)931 static int nbd_negotiate_meta_query(NBDClient *client,
932 NBDExportMetaContexts *meta, Error **errp)
933 {
934 int ret;
935 g_autofree char *query = NULL;
936 uint32_t len;
937
938 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
939 if (ret <= 0) {
940 return ret;
941 }
942 len = cpu_to_be32(len);
943
944 if (len > NBD_MAX_STRING_SIZE) {
945 trace_nbd_negotiate_meta_query_skip("length too long");
946 return nbd_opt_skip(client, len, errp);
947 }
948
949 query = g_malloc(len + 1);
950 ret = nbd_opt_read(client, query, len, true, errp);
951 if (ret <= 0) {
952 return ret;
953 }
954 query[len] = '\0';
955
956 if (nbd_meta_base_query(client, meta, query)) {
957 return 1;
958 }
959 if (nbd_meta_qemu_query(client, meta, query)) {
960 return 1;
961 }
962
963 trace_nbd_negotiate_meta_query_skip("unknown namespace");
964 return 1;
965 }
966
967 /* nbd_negotiate_meta_queries
968 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
969 *
970 * Return -errno on I/O error, or 0 if option was completely handled. */
nbd_negotiate_meta_queries(NBDClient * client,NBDExportMetaContexts * meta,Error ** errp)971 static int nbd_negotiate_meta_queries(NBDClient *client,
972 NBDExportMetaContexts *meta, Error **errp)
973 {
974 int ret;
975 g_autofree char *export_name = NULL;
976 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */
977 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL;
978 NBDExportMetaContexts local_meta = {0};
979 uint32_t nb_queries;
980 size_t i;
981 size_t count = 0;
982
983 if (!client->structured_reply) {
984 return nbd_opt_invalid(client, errp,
985 "request option '%s' when structured reply "
986 "is not negotiated",
987 nbd_opt_lookup(client->opt));
988 }
989
990 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
991 /* Only change the caller's meta on SET. */
992 meta = &local_meta;
993 }
994
995 g_free(meta->bitmaps);
996 memset(meta, 0, sizeof(*meta));
997
998 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
999 if (ret <= 0) {
1000 return ret;
1001 }
1002
1003 meta->exp = nbd_export_find(export_name);
1004 if (meta->exp == NULL) {
1005 g_autofree char *sane_name = nbd_sanitize_name(export_name);
1006
1007 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1008 "export '%s' not present", sane_name);
1009 }
1010 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1011 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1012 bitmaps = meta->bitmaps;
1013 }
1014
1015 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1016 if (ret <= 0) {
1017 return ret;
1018 }
1019 nb_queries = cpu_to_be32(nb_queries);
1020 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1021 export_name, nb_queries);
1022
1023 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1024 /* enable all known contexts */
1025 meta->base_allocation = true;
1026 meta->allocation_depth = meta->exp->allocation_depth;
1027 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1028 } else {
1029 for (i = 0; i < nb_queries; ++i) {
1030 ret = nbd_negotiate_meta_query(client, meta, errp);
1031 if (ret <= 0) {
1032 return ret;
1033 }
1034 }
1035 }
1036
1037 if (meta->base_allocation) {
1038 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1039 NBD_META_ID_BASE_ALLOCATION,
1040 errp);
1041 if (ret < 0) {
1042 return ret;
1043 }
1044 count++;
1045 }
1046
1047 if (meta->allocation_depth) {
1048 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1049 NBD_META_ID_ALLOCATION_DEPTH,
1050 errp);
1051 if (ret < 0) {
1052 return ret;
1053 }
1054 count++;
1055 }
1056
1057 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1058 const char *bm_name;
1059 g_autofree char *context = NULL;
1060
1061 if (!meta->bitmaps[i]) {
1062 continue;
1063 }
1064
1065 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1066 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1067
1068 ret = nbd_negotiate_send_meta_context(client, context,
1069 NBD_META_ID_DIRTY_BITMAP + i,
1070 errp);
1071 if (ret < 0) {
1072 return ret;
1073 }
1074 count++;
1075 }
1076
1077 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1078 if (ret == 0) {
1079 meta->count = count;
1080 }
1081
1082 return ret;
1083 }
1084
1085 /* nbd_negotiate_options
1086 * Process all NBD_OPT_* client option commands, during fixed newstyle
1087 * negotiation.
1088 * Return:
1089 * -errno on error, errp is set
1090 * 0 on successful negotiation, errp is not set
1091 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1092 * errp is not set
1093 */
nbd_negotiate_options(NBDClient * client,Error ** errp)1094 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1095 {
1096 uint32_t flags;
1097 bool fixedNewstyle = false;
1098 bool no_zeroes = false;
1099
1100 /* Client sends:
1101 [ 0 .. 3] client flags
1102
1103 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1104 [ 0 .. 7] NBD_OPTS_MAGIC
1105 [ 8 .. 11] NBD option
1106 [12 .. 15] Data length
1107 ... Rest of request
1108
1109 [ 0 .. 7] NBD_OPTS_MAGIC
1110 [ 8 .. 11] Second NBD option
1111 [12 .. 15] Data length
1112 ... Rest of request
1113 */
1114
1115 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1116 return -EIO;
1117 }
1118 trace_nbd_negotiate_options_flags(flags);
1119 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1120 fixedNewstyle = true;
1121 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1122 }
1123 if (flags & NBD_FLAG_C_NO_ZEROES) {
1124 no_zeroes = true;
1125 flags &= ~NBD_FLAG_C_NO_ZEROES;
1126 }
1127 if (flags != 0) {
1128 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1129 return -EINVAL;
1130 }
1131
1132 while (1) {
1133 int ret;
1134 uint32_t option, length;
1135 uint64_t magic;
1136
1137 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1138 return -EINVAL;
1139 }
1140 trace_nbd_negotiate_options_check_magic(magic);
1141 if (magic != NBD_OPTS_MAGIC) {
1142 error_setg(errp, "Bad magic received");
1143 return -EINVAL;
1144 }
1145
1146 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1147 return -EINVAL;
1148 }
1149 client->opt = option;
1150
1151 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1152 return -EINVAL;
1153 }
1154 assert(!client->optlen);
1155 client->optlen = length;
1156
1157 if (length > NBD_MAX_BUFFER_SIZE) {
1158 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1159 length, NBD_MAX_BUFFER_SIZE);
1160 return -EINVAL;
1161 }
1162
1163 trace_nbd_negotiate_options_check_option(option,
1164 nbd_opt_lookup(option));
1165 if (client->tlscreds &&
1166 client->ioc == (QIOChannel *)client->sioc) {
1167 QIOChannel *tioc;
1168 if (!fixedNewstyle) {
1169 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1170 return -EINVAL;
1171 }
1172 switch (option) {
1173 case NBD_OPT_STARTTLS:
1174 if (length) {
1175 /* Unconditionally drop the connection if the client
1176 * can't start a TLS negotiation correctly */
1177 return nbd_reject_length(client, true, errp);
1178 }
1179 tioc = nbd_negotiate_handle_starttls(client, errp);
1180 if (!tioc) {
1181 return -EIO;
1182 }
1183 ret = 0;
1184 object_unref(OBJECT(client->ioc));
1185 client->ioc = QIO_CHANNEL(tioc);
1186 break;
1187
1188 case NBD_OPT_EXPORT_NAME:
1189 /* No way to return an error to client, so drop connection */
1190 error_setg(errp, "Option 0x%x not permitted before TLS",
1191 option);
1192 return -EINVAL;
1193
1194 default:
1195 /* Let the client keep trying, unless they asked to
1196 * quit. Always try to give an error back to the
1197 * client; but when replying to OPT_ABORT, be aware
1198 * that the client may hang up before receiving the
1199 * error, in which case we are fine ignoring the
1200 * resulting EPIPE. */
1201 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1202 option == NBD_OPT_ABORT ? NULL : errp,
1203 "Option 0x%" PRIx32
1204 " not permitted before TLS", option);
1205 if (option == NBD_OPT_ABORT) {
1206 return 1;
1207 }
1208 break;
1209 }
1210 } else if (fixedNewstyle) {
1211 switch (option) {
1212 case NBD_OPT_LIST:
1213 if (length) {
1214 ret = nbd_reject_length(client, false, errp);
1215 } else {
1216 ret = nbd_negotiate_handle_list(client, errp);
1217 }
1218 break;
1219
1220 case NBD_OPT_ABORT:
1221 /* NBD spec says we must try to reply before
1222 * disconnecting, but that we must also tolerate
1223 * guests that don't wait for our reply. */
1224 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1225 return 1;
1226
1227 case NBD_OPT_EXPORT_NAME:
1228 return nbd_negotiate_handle_export_name(client, no_zeroes,
1229 errp);
1230
1231 case NBD_OPT_INFO:
1232 case NBD_OPT_GO:
1233 ret = nbd_negotiate_handle_info(client, errp);
1234 if (ret == 1) {
1235 assert(option == NBD_OPT_GO);
1236 return 0;
1237 }
1238 break;
1239
1240 case NBD_OPT_STARTTLS:
1241 if (length) {
1242 ret = nbd_reject_length(client, false, errp);
1243 } else if (client->tlscreds) {
1244 ret = nbd_negotiate_send_rep_err(client,
1245 NBD_REP_ERR_INVALID, errp,
1246 "TLS already enabled");
1247 } else {
1248 ret = nbd_negotiate_send_rep_err(client,
1249 NBD_REP_ERR_POLICY, errp,
1250 "TLS not configured");
1251 }
1252 break;
1253
1254 case NBD_OPT_STRUCTURED_REPLY:
1255 if (length) {
1256 ret = nbd_reject_length(client, false, errp);
1257 } else if (client->structured_reply) {
1258 ret = nbd_negotiate_send_rep_err(
1259 client, NBD_REP_ERR_INVALID, errp,
1260 "structured reply already negotiated");
1261 } else {
1262 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1263 client->structured_reply = true;
1264 }
1265 break;
1266
1267 case NBD_OPT_LIST_META_CONTEXT:
1268 case NBD_OPT_SET_META_CONTEXT:
1269 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1270 errp);
1271 break;
1272
1273 default:
1274 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1275 "Unsupported option %" PRIu32 " (%s)",
1276 option, nbd_opt_lookup(option));
1277 break;
1278 }
1279 } else {
1280 /*
1281 * If broken new-style we should drop the connection
1282 * for anything except NBD_OPT_EXPORT_NAME
1283 */
1284 switch (option) {
1285 case NBD_OPT_EXPORT_NAME:
1286 return nbd_negotiate_handle_export_name(client, no_zeroes,
1287 errp);
1288
1289 default:
1290 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1291 option, nbd_opt_lookup(option));
1292 return -EINVAL;
1293 }
1294 }
1295 if (ret < 0) {
1296 return ret;
1297 }
1298 }
1299 }
1300
1301 /* nbd_negotiate
1302 * Return:
1303 * -errno on error, errp is set
1304 * 0 on successful negotiation, errp is not set
1305 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1306 * errp is not set
1307 */
nbd_negotiate(NBDClient * client,Error ** errp)1308 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1309 {
1310 ERRP_GUARD();
1311 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1312 int ret;
1313
1314 /* Old style negotiation header, no room for options
1315 [ 0 .. 7] passwd ("NBDMAGIC")
1316 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1317 [16 .. 23] size
1318 [24 .. 27] export flags (zero-extended)
1319 [28 .. 151] reserved (0)
1320
1321 New style negotiation header, client can send options
1322 [ 0 .. 7] passwd ("NBDMAGIC")
1323 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1324 [16 .. 17] server flags (0)
1325 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1326 */
1327
1328 qio_channel_set_blocking(client->ioc, false, NULL);
1329
1330 trace_nbd_negotiate_begin();
1331 memcpy(buf, "NBDMAGIC", 8);
1332
1333 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1334 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1335
1336 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1337 error_prepend(errp, "write failed: ");
1338 return -EINVAL;
1339 }
1340 ret = nbd_negotiate_options(client, errp);
1341 if (ret != 0) {
1342 if (ret < 0) {
1343 error_prepend(errp, "option negotiation failed: ");
1344 }
1345 return ret;
1346 }
1347
1348 /* Attach the channel to the same AioContext as the export */
1349 if (client->exp && client->exp->common.ctx) {
1350 qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
1351 }
1352
1353 assert(!client->optlen);
1354 trace_nbd_negotiate_success();
1355
1356 return 0;
1357 }
1358
1359 /* nbd_read_eof
1360 * Tries to read @size bytes from @ioc. This is a local implementation of
1361 * qio_channel_readv_all_eof. We have it here because we need it to be
1362 * interruptible and to know when the coroutine is yielding.
1363 * Returns 1 on success
1364 * 0 on eof, when no data was read (errp is not set)
1365 * negative errno on failure (errp is set)
1366 */
1367 static inline int coroutine_fn
nbd_read_eof(NBDClient * client,void * buffer,size_t size,Error ** errp)1368 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1369 {
1370 bool partial = false;
1371
1372 assert(size);
1373 while (size > 0) {
1374 struct iovec iov = { .iov_base = buffer, .iov_len = size };
1375 ssize_t len;
1376
1377 len = qio_channel_readv(client->ioc, &iov, 1, errp);
1378 if (len == QIO_CHANNEL_ERR_BLOCK) {
1379 client->read_yielding = true;
1380 qio_channel_yield(client->ioc, G_IO_IN);
1381 client->read_yielding = false;
1382 if (client->quiescing) {
1383 return -EAGAIN;
1384 }
1385 continue;
1386 } else if (len < 0) {
1387 return -EIO;
1388 } else if (len == 0) {
1389 if (partial) {
1390 error_setg(errp,
1391 "Unexpected end-of-file before all bytes were read");
1392 return -EIO;
1393 } else {
1394 return 0;
1395 }
1396 }
1397
1398 partial = true;
1399 size -= len;
1400 buffer = (uint8_t *) buffer + len;
1401 }
1402 return 1;
1403 }
1404
nbd_receive_request(NBDClient * client,NBDRequest * request,Error ** errp)1405 static int nbd_receive_request(NBDClient *client, NBDRequest *request,
1406 Error **errp)
1407 {
1408 uint8_t buf[NBD_REQUEST_SIZE];
1409 uint32_t magic;
1410 int ret;
1411
1412 ret = nbd_read_eof(client, buf, sizeof(buf), errp);
1413 if (ret < 0) {
1414 return ret;
1415 }
1416
1417 /* Request
1418 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1419 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1420 [ 6 .. 7] type (NBD_CMD_READ, ...)
1421 [ 8 .. 15] handle
1422 [16 .. 23] from
1423 [24 .. 27] len
1424 */
1425
1426 magic = ldl_be_p(buf);
1427 request->flags = lduw_be_p(buf + 4);
1428 request->type = lduw_be_p(buf + 6);
1429 request->handle = ldq_be_p(buf + 8);
1430 request->from = ldq_be_p(buf + 16);
1431 request->len = ldl_be_p(buf + 24);
1432
1433 trace_nbd_receive_request(magic, request->flags, request->type,
1434 request->from, request->len);
1435
1436 if (magic != NBD_REQUEST_MAGIC) {
1437 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1438 return -EINVAL;
1439 }
1440 return 0;
1441 }
1442
1443 #define MAX_NBD_REQUESTS 16
1444
nbd_client_get(NBDClient * client)1445 void nbd_client_get(NBDClient *client)
1446 {
1447 client->refcount++;
1448 }
1449
nbd_client_put(NBDClient * client)1450 void nbd_client_put(NBDClient *client)
1451 {
1452 if (--client->refcount == 0) {
1453 /* The last reference should be dropped by client->close,
1454 * which is called by client_close.
1455 */
1456 assert(client->closing);
1457
1458 qio_channel_detach_aio_context(client->ioc);
1459 object_unref(OBJECT(client->sioc));
1460 object_unref(OBJECT(client->ioc));
1461 if (client->tlscreds) {
1462 object_unref(OBJECT(client->tlscreds));
1463 }
1464 g_free(client->tlsauthz);
1465 if (client->exp) {
1466 QTAILQ_REMOVE(&client->exp->clients, client, next);
1467 blk_exp_unref(&client->exp->common);
1468 }
1469 g_free(client->export_meta.bitmaps);
1470 g_free(client);
1471 }
1472 }
1473
client_close(NBDClient * client,bool negotiated)1474 static void client_close(NBDClient *client, bool negotiated)
1475 {
1476 if (client->closing) {
1477 return;
1478 }
1479
1480 client->closing = true;
1481
1482 /* Force requests to finish. They will drop their own references,
1483 * then we'll close the socket and free the NBDClient.
1484 */
1485 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1486 NULL);
1487
1488 /* Also tell the client, so that they release their reference. */
1489 if (client->close_fn) {
1490 client->close_fn(client, negotiated);
1491 }
1492 }
1493
nbd_request_get(NBDClient * client)1494 static NBDRequestData *nbd_request_get(NBDClient *client)
1495 {
1496 NBDRequestData *req;
1497
1498 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1499 client->nb_requests++;
1500
1501 req = g_new0(NBDRequestData, 1);
1502 nbd_client_get(client);
1503 req->client = client;
1504 return req;
1505 }
1506
nbd_request_put(NBDRequestData * req)1507 static void nbd_request_put(NBDRequestData *req)
1508 {
1509 NBDClient *client = req->client;
1510
1511 if (req->data) {
1512 qemu_vfree(req->data);
1513 }
1514 g_free(req);
1515
1516 client->nb_requests--;
1517
1518 if (client->quiescing && client->nb_requests == 0) {
1519 aio_wait_kick();
1520 }
1521
1522 nbd_client_receive_next_request(client);
1523
1524 nbd_client_put(client);
1525 }
1526
blk_aio_attached(AioContext * ctx,void * opaque)1527 static void blk_aio_attached(AioContext *ctx, void *opaque)
1528 {
1529 NBDExport *exp = opaque;
1530 NBDClient *client;
1531
1532 trace_nbd_blk_aio_attached(exp->name, ctx);
1533
1534 exp->common.ctx = ctx;
1535
1536 QTAILQ_FOREACH(client, &exp->clients, next) {
1537 qio_channel_attach_aio_context(client->ioc, ctx);
1538
1539 assert(client->nb_requests == 0);
1540 assert(client->recv_coroutine == NULL);
1541 assert(client->send_coroutine == NULL);
1542 }
1543 }
1544
blk_aio_detach(void * opaque)1545 static void blk_aio_detach(void *opaque)
1546 {
1547 NBDExport *exp = opaque;
1548 NBDClient *client;
1549
1550 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1551
1552 QTAILQ_FOREACH(client, &exp->clients, next) {
1553 qio_channel_detach_aio_context(client->ioc);
1554 }
1555
1556 exp->common.ctx = NULL;
1557 }
1558
nbd_drained_begin(void * opaque)1559 static void nbd_drained_begin(void *opaque)
1560 {
1561 NBDExport *exp = opaque;
1562 NBDClient *client;
1563
1564 QTAILQ_FOREACH(client, &exp->clients, next) {
1565 client->quiescing = true;
1566 }
1567 }
1568
nbd_drained_end(void * opaque)1569 static void nbd_drained_end(void *opaque)
1570 {
1571 NBDExport *exp = opaque;
1572 NBDClient *client;
1573
1574 QTAILQ_FOREACH(client, &exp->clients, next) {
1575 client->quiescing = false;
1576 nbd_client_receive_next_request(client);
1577 }
1578 }
1579
nbd_drained_poll(void * opaque)1580 static bool nbd_drained_poll(void *opaque)
1581 {
1582 NBDExport *exp = opaque;
1583 NBDClient *client;
1584
1585 QTAILQ_FOREACH(client, &exp->clients, next) {
1586 if (client->nb_requests != 0) {
1587 /*
1588 * If there's a coroutine waiting for a request on nbd_read_eof()
1589 * enter it here so we don't depend on the client to wake it up.
1590 */
1591 if (client->recv_coroutine != NULL && client->read_yielding) {
1592 qemu_aio_coroutine_enter(exp->common.ctx,
1593 client->recv_coroutine);
1594 }
1595
1596 return true;
1597 }
1598 }
1599
1600 return false;
1601 }
1602
nbd_eject_notifier(Notifier * n,void * data)1603 static void nbd_eject_notifier(Notifier *n, void *data)
1604 {
1605 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1606
1607 blk_exp_request_shutdown(&exp->common);
1608 }
1609
nbd_export_set_on_eject_blk(BlockExport * exp,BlockBackend * blk)1610 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1611 {
1612 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1613 assert(exp->drv == &blk_exp_nbd);
1614 assert(nbd_exp->eject_notifier_blk == NULL);
1615
1616 blk_ref(blk);
1617 nbd_exp->eject_notifier_blk = blk;
1618 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1619 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1620 }
1621
1622 static const BlockDevOps nbd_block_ops = {
1623 .drained_begin = nbd_drained_begin,
1624 .drained_end = nbd_drained_end,
1625 .drained_poll = nbd_drained_poll,
1626 };
1627
nbd_export_create(BlockExport * blk_exp,BlockExportOptions * exp_args,Error ** errp)1628 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1629 Error **errp)
1630 {
1631 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1632 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1633 BlockBackend *blk = blk_exp->blk;
1634 int64_t size;
1635 uint64_t perm, shared_perm;
1636 bool readonly = !exp_args->writable;
1637 bool shared = !exp_args->writable;
1638 strList *bitmaps;
1639 size_t i;
1640 int ret;
1641
1642 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1643
1644 if (!nbd_server_is_running()) {
1645 error_setg(errp, "NBD server not running");
1646 return -EINVAL;
1647 }
1648
1649 if (!arg->has_name) {
1650 arg->name = exp_args->node_name;
1651 }
1652
1653 if (strlen(arg->name) > NBD_MAX_STRING_SIZE) {
1654 error_setg(errp, "export name '%s' too long", arg->name);
1655 return -EINVAL;
1656 }
1657
1658 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1659 error_setg(errp, "description '%s' too long", arg->description);
1660 return -EINVAL;
1661 }
1662
1663 if (nbd_export_find(arg->name)) {
1664 error_setg(errp, "NBD server already has export named '%s'", arg->name);
1665 return -EEXIST;
1666 }
1667
1668 size = blk_getlength(blk);
1669 if (size < 0) {
1670 error_setg_errno(errp, -size,
1671 "Failed to determine the NBD export's length");
1672 return size;
1673 }
1674
1675 /* Don't allow resize while the NBD server is running, otherwise we don't
1676 * care what happens with the node. */
1677 blk_get_perm(blk, &perm, &shared_perm);
1678 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1679 if (ret < 0) {
1680 return ret;
1681 }
1682
1683 QTAILQ_INIT(&exp->clients);
1684 exp->name = g_strdup(arg->name);
1685 exp->description = g_strdup(arg->description);
1686 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1687 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1688 if (readonly) {
1689 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1690 if (shared) {
1691 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1692 }
1693 } else {
1694 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1695 NBD_FLAG_SEND_FAST_ZERO);
1696 }
1697 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1698
1699 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1700 exp->nr_export_bitmaps++;
1701 }
1702 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1703 for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1704 i++, bitmaps = bitmaps->next) {
1705 const char *bitmap = bitmaps->value;
1706 BlockDriverState *bs = blk_bs(blk);
1707 BdrvDirtyBitmap *bm = NULL;
1708
1709 while (bs) {
1710 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1711 if (bm != NULL) {
1712 break;
1713 }
1714
1715 bs = bdrv_filter_or_cow_bs(bs);
1716 }
1717
1718 if (bm == NULL) {
1719 ret = -ENOENT;
1720 error_setg(errp, "Bitmap '%s' is not found", bitmap);
1721 goto fail;
1722 }
1723
1724 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1725 ret = -EINVAL;
1726 goto fail;
1727 }
1728
1729 if (readonly && bdrv_is_writable(bs) &&
1730 bdrv_dirty_bitmap_enabled(bm)) {
1731 ret = -EINVAL;
1732 error_setg(errp,
1733 "Enabled bitmap '%s' incompatible with readonly export",
1734 bitmap);
1735 goto fail;
1736 }
1737
1738 exp->export_bitmaps[i] = bm;
1739 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1740 }
1741
1742 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1743 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1744 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1745 }
1746
1747 exp->allocation_depth = arg->allocation_depth;
1748
1749 /*
1750 * We need to inhibit request queuing in the block layer to ensure we can
1751 * be properly quiesced when entering a drained section, as our coroutines
1752 * servicing pending requests might enter blk_pread().
1753 */
1754 blk_set_disable_request_queuing(blk, true);
1755
1756 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1757
1758 blk_set_dev_ops(blk, &nbd_block_ops, exp);
1759
1760 QTAILQ_INSERT_TAIL(&exports, exp, next);
1761
1762 return 0;
1763
1764 fail:
1765 g_free(exp->export_bitmaps);
1766 g_free(exp->name);
1767 g_free(exp->description);
1768 return ret;
1769 }
1770
nbd_export_find(const char * name)1771 NBDExport *nbd_export_find(const char *name)
1772 {
1773 NBDExport *exp;
1774 QTAILQ_FOREACH(exp, &exports, next) {
1775 if (strcmp(name, exp->name) == 0) {
1776 return exp;
1777 }
1778 }
1779
1780 return NULL;
1781 }
1782
1783 AioContext *
nbd_export_aio_context(NBDExport * exp)1784 nbd_export_aio_context(NBDExport *exp)
1785 {
1786 return exp->common.ctx;
1787 }
1788
nbd_export_request_shutdown(BlockExport * blk_exp)1789 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1790 {
1791 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1792 NBDClient *client, *next;
1793
1794 blk_exp_ref(&exp->common);
1795 /*
1796 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1797 * close mode that stops advertising the export to new clients but
1798 * still permits existing clients to run to completion? Because of
1799 * that possibility, nbd_export_close() can be called more than
1800 * once on an export.
1801 */
1802 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1803 client_close(client, true);
1804 }
1805 if (exp->name) {
1806 g_free(exp->name);
1807 exp->name = NULL;
1808 QTAILQ_REMOVE(&exports, exp, next);
1809 }
1810 blk_exp_unref(&exp->common);
1811 }
1812
nbd_export_delete(BlockExport * blk_exp)1813 static void nbd_export_delete(BlockExport *blk_exp)
1814 {
1815 size_t i;
1816 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1817
1818 assert(exp->name == NULL);
1819 assert(QTAILQ_EMPTY(&exp->clients));
1820
1821 g_free(exp->description);
1822 exp->description = NULL;
1823
1824 if (exp->common.blk) {
1825 if (exp->eject_notifier_blk) {
1826 notifier_remove(&exp->eject_notifier);
1827 blk_unref(exp->eject_notifier_blk);
1828 }
1829 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1830 blk_aio_detach, exp);
1831 blk_set_disable_request_queuing(exp->common.blk, false);
1832 }
1833
1834 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1835 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1836 }
1837 }
1838
1839 const BlockExportDriver blk_exp_nbd = {
1840 .type = BLOCK_EXPORT_TYPE_NBD,
1841 .instance_size = sizeof(NBDExport),
1842 .create = nbd_export_create,
1843 .delete = nbd_export_delete,
1844 .request_shutdown = nbd_export_request_shutdown,
1845 };
1846
nbd_co_send_iov(NBDClient * client,struct iovec * iov,unsigned niov,Error ** errp)1847 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1848 unsigned niov, Error **errp)
1849 {
1850 int ret;
1851
1852 g_assert(qemu_in_coroutine());
1853 qemu_co_mutex_lock(&client->send_lock);
1854 client->send_coroutine = qemu_coroutine_self();
1855
1856 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1857
1858 client->send_coroutine = NULL;
1859 qemu_co_mutex_unlock(&client->send_lock);
1860
1861 return ret;
1862 }
1863
set_be_simple_reply(NBDSimpleReply * reply,uint64_t error,uint64_t handle)1864 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1865 uint64_t handle)
1866 {
1867 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1868 stl_be_p(&reply->error, error);
1869 stq_be_p(&reply->handle, handle);
1870 }
1871
nbd_co_send_simple_reply(NBDClient * client,uint64_t handle,uint32_t error,void * data,size_t len,Error ** errp)1872 static int nbd_co_send_simple_reply(NBDClient *client,
1873 uint64_t handle,
1874 uint32_t error,
1875 void *data,
1876 size_t len,
1877 Error **errp)
1878 {
1879 NBDSimpleReply reply;
1880 int nbd_err = system_errno_to_nbd_errno(error);
1881 struct iovec iov[] = {
1882 {.iov_base = &reply, .iov_len = sizeof(reply)},
1883 {.iov_base = data, .iov_len = len}
1884 };
1885
1886 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1887 len);
1888 set_be_simple_reply(&reply, nbd_err, handle);
1889
1890 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1891 }
1892
set_be_chunk(NBDStructuredReplyChunk * chunk,uint16_t flags,uint16_t type,uint64_t handle,uint32_t length)1893 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1894 uint16_t type, uint64_t handle, uint32_t length)
1895 {
1896 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1897 stw_be_p(&chunk->flags, flags);
1898 stw_be_p(&chunk->type, type);
1899 stq_be_p(&chunk->handle, handle);
1900 stl_be_p(&chunk->length, length);
1901 }
1902
nbd_co_send_structured_done(NBDClient * client,uint64_t handle,Error ** errp)1903 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1904 uint64_t handle,
1905 Error **errp)
1906 {
1907 NBDStructuredReplyChunk chunk;
1908 struct iovec iov[] = {
1909 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1910 };
1911
1912 trace_nbd_co_send_structured_done(handle);
1913 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1914
1915 return nbd_co_send_iov(client, iov, 1, errp);
1916 }
1917
nbd_co_send_structured_read(NBDClient * client,uint64_t handle,uint64_t offset,void * data,size_t size,bool final,Error ** errp)1918 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1919 uint64_t handle,
1920 uint64_t offset,
1921 void *data,
1922 size_t size,
1923 bool final,
1924 Error **errp)
1925 {
1926 NBDStructuredReadData chunk;
1927 struct iovec iov[] = {
1928 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1929 {.iov_base = data, .iov_len = size}
1930 };
1931
1932 assert(size);
1933 trace_nbd_co_send_structured_read(handle, offset, data, size);
1934 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1935 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1936 sizeof(chunk) - sizeof(chunk.h) + size);
1937 stq_be_p(&chunk.offset, offset);
1938
1939 return nbd_co_send_iov(client, iov, 2, errp);
1940 }
1941
nbd_co_send_structured_error(NBDClient * client,uint64_t handle,uint32_t error,const char * msg,Error ** errp)1942 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1943 uint64_t handle,
1944 uint32_t error,
1945 const char *msg,
1946 Error **errp)
1947 {
1948 NBDStructuredError chunk;
1949 int nbd_err = system_errno_to_nbd_errno(error);
1950 struct iovec iov[] = {
1951 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1952 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1953 };
1954
1955 assert(nbd_err);
1956 trace_nbd_co_send_structured_error(handle, nbd_err,
1957 nbd_err_lookup(nbd_err), msg ? msg : "");
1958 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1959 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1960 stl_be_p(&chunk.error, nbd_err);
1961 stw_be_p(&chunk.message_length, iov[1].iov_len);
1962
1963 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1964 }
1965
1966 /* Do a sparse read and send the structured reply to the client.
1967 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1968 * reported to the client, at which point this function succeeds.
1969 */
nbd_co_send_sparse_read(NBDClient * client,uint64_t handle,uint64_t offset,uint8_t * data,size_t size,Error ** errp)1970 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1971 uint64_t handle,
1972 uint64_t offset,
1973 uint8_t *data,
1974 size_t size,
1975 Error **errp)
1976 {
1977 int ret = 0;
1978 NBDExport *exp = client->exp;
1979 size_t progress = 0;
1980
1981 while (progress < size) {
1982 int64_t pnum;
1983 int status = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
1984 offset + progress,
1985 size - progress, &pnum, NULL,
1986 NULL);
1987 bool final;
1988
1989 if (status < 0) {
1990 char *msg = g_strdup_printf("unable to check for holes: %s",
1991 strerror(-status));
1992
1993 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1994 errp);
1995 g_free(msg);
1996 return ret;
1997 }
1998 assert(pnum && pnum <= size - progress);
1999 final = progress + pnum == size;
2000 if (status & BDRV_BLOCK_ZERO) {
2001 NBDStructuredReadHole chunk;
2002 struct iovec iov[] = {
2003 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2004 };
2005
2006 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
2007 pnum);
2008 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
2009 NBD_REPLY_TYPE_OFFSET_HOLE,
2010 handle, sizeof(chunk) - sizeof(chunk.h));
2011 stq_be_p(&chunk.offset, offset + progress);
2012 stl_be_p(&chunk.length, pnum);
2013 ret = nbd_co_send_iov(client, iov, 1, errp);
2014 } else {
2015 ret = blk_pread(exp->common.blk, offset + progress,
2016 data + progress, pnum);
2017 if (ret < 0) {
2018 error_setg_errno(errp, -ret, "reading from file failed");
2019 break;
2020 }
2021 ret = nbd_co_send_structured_read(client, handle, offset + progress,
2022 data + progress, pnum, final,
2023 errp);
2024 }
2025
2026 if (ret < 0) {
2027 break;
2028 }
2029 progress += pnum;
2030 }
2031 return ret;
2032 }
2033
2034 typedef struct NBDExtentArray {
2035 NBDExtent *extents;
2036 unsigned int nb_alloc;
2037 unsigned int count;
2038 uint64_t total_length;
2039 bool can_add;
2040 bool converted_to_be;
2041 } NBDExtentArray;
2042
nbd_extent_array_new(unsigned int nb_alloc)2043 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
2044 {
2045 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2046
2047 ea->nb_alloc = nb_alloc;
2048 ea->extents = g_new(NBDExtent, nb_alloc);
2049 ea->can_add = true;
2050
2051 return ea;
2052 }
2053
nbd_extent_array_free(NBDExtentArray * ea)2054 static void nbd_extent_array_free(NBDExtentArray *ea)
2055 {
2056 g_free(ea->extents);
2057 g_free(ea);
2058 }
2059 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free);
2060
2061 /* Further modifications of the array after conversion are abandoned */
nbd_extent_array_convert_to_be(NBDExtentArray * ea)2062 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2063 {
2064 int i;
2065
2066 assert(!ea->converted_to_be);
2067 ea->can_add = false;
2068 ea->converted_to_be = true;
2069
2070 for (i = 0; i < ea->count; i++) {
2071 ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2072 ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
2073 }
2074 }
2075
2076 /*
2077 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2078 * return -1.
2079 * For safety, when returning -1 for the first time, .can_add is set to false,
2080 * further call to nbd_extent_array_add() will crash.
2081 * (to avoid the situation, when after failing to add an extent (returned -1),
2082 * user miss this failure and add another extent, which is successfully added
2083 * (array is full, but new extent may be squashed into the last one), then we
2084 * have invalid array with skipped extent)
2085 */
nbd_extent_array_add(NBDExtentArray * ea,uint32_t length,uint32_t flags)2086 static int nbd_extent_array_add(NBDExtentArray *ea,
2087 uint32_t length, uint32_t flags)
2088 {
2089 assert(ea->can_add);
2090
2091 if (!length) {
2092 return 0;
2093 }
2094
2095 /* Extend previous extent if flags are the same */
2096 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2097 uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
2098
2099 if (sum <= UINT32_MAX) {
2100 ea->extents[ea->count - 1].length = sum;
2101 ea->total_length += length;
2102 return 0;
2103 }
2104 }
2105
2106 if (ea->count >= ea->nb_alloc) {
2107 ea->can_add = false;
2108 return -1;
2109 }
2110
2111 ea->total_length += length;
2112 ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
2113 ea->count++;
2114
2115 return 0;
2116 }
2117
blockstatus_to_extents(BlockDriverState * bs,uint64_t offset,uint64_t bytes,NBDExtentArray * ea)2118 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
2119 uint64_t bytes, NBDExtentArray *ea)
2120 {
2121 while (bytes) {
2122 uint32_t flags;
2123 int64_t num;
2124 int ret = bdrv_block_status_above(bs, NULL, offset, bytes, &num,
2125 NULL, NULL);
2126
2127 if (ret < 0) {
2128 return ret;
2129 }
2130
2131 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2132 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2133
2134 if (nbd_extent_array_add(ea, num, flags) < 0) {
2135 return 0;
2136 }
2137
2138 offset += num;
2139 bytes -= num;
2140 }
2141
2142 return 0;
2143 }
2144
blockalloc_to_extents(BlockDriverState * bs,uint64_t offset,uint64_t bytes,NBDExtentArray * ea)2145 static int blockalloc_to_extents(BlockDriverState *bs, uint64_t offset,
2146 uint64_t bytes, NBDExtentArray *ea)
2147 {
2148 while (bytes) {
2149 int64_t num;
2150 int ret = bdrv_is_allocated_above(bs, NULL, false, offset, bytes,
2151 &num);
2152
2153 if (ret < 0) {
2154 return ret;
2155 }
2156
2157 if (nbd_extent_array_add(ea, num, ret) < 0) {
2158 return 0;
2159 }
2160
2161 offset += num;
2162 bytes -= num;
2163 }
2164
2165 return 0;
2166 }
2167
2168 /*
2169 * nbd_co_send_extents
2170 *
2171 * @ea is converted to BE by the function
2172 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2173 */
nbd_co_send_extents(NBDClient * client,uint64_t handle,NBDExtentArray * ea,bool last,uint32_t context_id,Error ** errp)2174 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
2175 NBDExtentArray *ea,
2176 bool last, uint32_t context_id, Error **errp)
2177 {
2178 NBDStructuredMeta chunk;
2179 struct iovec iov[] = {
2180 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2181 {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
2182 };
2183
2184 nbd_extent_array_convert_to_be(ea);
2185
2186 trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
2187 last);
2188 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
2189 NBD_REPLY_TYPE_BLOCK_STATUS,
2190 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
2191 stl_be_p(&chunk.context_id, context_id);
2192
2193 return nbd_co_send_iov(client, iov, 2, errp);
2194 }
2195
2196 /* Get block status from the exported device and send it to the client */
nbd_co_send_block_status(NBDClient * client,uint64_t handle,BlockDriverState * bs,uint64_t offset,uint32_t length,bool dont_fragment,bool last,uint32_t context_id,Error ** errp)2197 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2198 BlockDriverState *bs, uint64_t offset,
2199 uint32_t length, bool dont_fragment,
2200 bool last, uint32_t context_id,
2201 Error **errp)
2202 {
2203 int ret;
2204 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2205 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2206
2207 if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2208 ret = blockstatus_to_extents(bs, offset, length, ea);
2209 } else {
2210 ret = blockalloc_to_extents(bs, offset, length, ea);
2211 }
2212 if (ret < 0) {
2213 return nbd_co_send_structured_error(
2214 client, handle, -ret, "can't get block status", errp);
2215 }
2216
2217 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2218 }
2219
2220 /* Populate @ea from a dirty bitmap. */
bitmap_to_extents(BdrvDirtyBitmap * bitmap,uint64_t offset,uint64_t length,NBDExtentArray * es)2221 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2222 uint64_t offset, uint64_t length,
2223 NBDExtentArray *es)
2224 {
2225 int64_t start, dirty_start, dirty_count;
2226 int64_t end = offset + length;
2227 bool full = false;
2228
2229 bdrv_dirty_bitmap_lock(bitmap);
2230
2231 for (start = offset;
2232 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2233 &dirty_start, &dirty_count);
2234 start = dirty_start + dirty_count)
2235 {
2236 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2237 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2238 {
2239 full = true;
2240 break;
2241 }
2242 }
2243
2244 if (!full) {
2245 /* last non dirty extent, nothing to do if array is now full */
2246 (void) nbd_extent_array_add(es, end - start, 0);
2247 }
2248
2249 bdrv_dirty_bitmap_unlock(bitmap);
2250 }
2251
nbd_co_send_bitmap(NBDClient * client,uint64_t handle,BdrvDirtyBitmap * bitmap,uint64_t offset,uint32_t length,bool dont_fragment,bool last,uint32_t context_id,Error ** errp)2252 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2253 BdrvDirtyBitmap *bitmap, uint64_t offset,
2254 uint32_t length, bool dont_fragment, bool last,
2255 uint32_t context_id, Error **errp)
2256 {
2257 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2258 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2259
2260 bitmap_to_extents(bitmap, offset, length, ea);
2261
2262 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2263 }
2264
2265 /* nbd_co_receive_request
2266 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2267 * connection right away, -EAGAIN to indicate we were interrupted and the
2268 * channel should be quiesced, and any other negative value to report an error
2269 * to the client (although the caller may still need to disconnect after
2270 * reporting the error).
2271 */
nbd_co_receive_request(NBDRequestData * req,NBDRequest * request,Error ** errp)2272 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2273 Error **errp)
2274 {
2275 NBDClient *client = req->client;
2276 int valid_flags;
2277 int ret;
2278
2279 g_assert(qemu_in_coroutine());
2280 assert(client->recv_coroutine == qemu_coroutine_self());
2281 ret = nbd_receive_request(client, request, errp);
2282 if (ret < 0) {
2283 return ret;
2284 }
2285
2286 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2287 nbd_cmd_lookup(request->type));
2288
2289 if (request->type != NBD_CMD_WRITE) {
2290 /* No payload, we are ready to read the next request. */
2291 req->complete = true;
2292 }
2293
2294 if (request->type == NBD_CMD_DISC) {
2295 /* Special case: we're going to disconnect without a reply,
2296 * whether or not flags, from, or len are bogus */
2297 return -EIO;
2298 }
2299
2300 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2301 request->type == NBD_CMD_CACHE)
2302 {
2303 if (request->len > NBD_MAX_BUFFER_SIZE) {
2304 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2305 request->len, NBD_MAX_BUFFER_SIZE);
2306 return -EINVAL;
2307 }
2308
2309 if (request->type != NBD_CMD_CACHE) {
2310 req->data = blk_try_blockalign(client->exp->common.blk,
2311 request->len);
2312 if (req->data == NULL) {
2313 error_setg(errp, "No memory");
2314 return -ENOMEM;
2315 }
2316 }
2317 }
2318
2319 if (request->type == NBD_CMD_WRITE) {
2320 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2321 errp) < 0)
2322 {
2323 return -EIO;
2324 }
2325 req->complete = true;
2326
2327 trace_nbd_co_receive_request_payload_received(request->handle,
2328 request->len);
2329 }
2330
2331 /* Sanity checks. */
2332 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2333 (request->type == NBD_CMD_WRITE ||
2334 request->type == NBD_CMD_WRITE_ZEROES ||
2335 request->type == NBD_CMD_TRIM)) {
2336 error_setg(errp, "Export is read-only");
2337 return -EROFS;
2338 }
2339 if (request->from > client->exp->size ||
2340 request->len > client->exp->size - request->from) {
2341 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2342 ", Size: %" PRIu64, request->from, request->len,
2343 client->exp->size);
2344 return (request->type == NBD_CMD_WRITE ||
2345 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2346 }
2347 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2348 client->check_align)) {
2349 /*
2350 * The block layer gracefully handles unaligned requests, but
2351 * it's still worth tracing client non-compliance
2352 */
2353 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2354 request->from,
2355 request->len,
2356 client->check_align);
2357 }
2358 valid_flags = NBD_CMD_FLAG_FUA;
2359 if (request->type == NBD_CMD_READ && client->structured_reply) {
2360 valid_flags |= NBD_CMD_FLAG_DF;
2361 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2362 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2363 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2364 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2365 }
2366 if (request->flags & ~valid_flags) {
2367 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2368 nbd_cmd_lookup(request->type), request->flags);
2369 return -EINVAL;
2370 }
2371
2372 return 0;
2373 }
2374
2375 /* Send simple reply without a payload, or a structured error
2376 * @error_msg is ignored if @ret >= 0
2377 * Returns 0 if connection is still live, -errno on failure to talk to client
2378 */
nbd_send_generic_reply(NBDClient * client,uint64_t handle,int ret,const char * error_msg,Error ** errp)2379 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2380 uint64_t handle,
2381 int ret,
2382 const char *error_msg,
2383 Error **errp)
2384 {
2385 if (client->structured_reply && ret < 0) {
2386 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2387 errp);
2388 } else {
2389 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2390 NULL, 0, errp);
2391 }
2392 }
2393
2394 /* Handle NBD_CMD_READ request.
2395 * Return -errno if sending fails. Other errors are reported directly to the
2396 * client as an error reply. */
nbd_do_cmd_read(NBDClient * client,NBDRequest * request,uint8_t * data,Error ** errp)2397 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2398 uint8_t *data, Error **errp)
2399 {
2400 int ret;
2401 NBDExport *exp = client->exp;
2402
2403 assert(request->type == NBD_CMD_READ);
2404
2405 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2406 if (request->flags & NBD_CMD_FLAG_FUA) {
2407 ret = blk_co_flush(exp->common.blk);
2408 if (ret < 0) {
2409 return nbd_send_generic_reply(client, request->handle, ret,
2410 "flush failed", errp);
2411 }
2412 }
2413
2414 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2415 request->len)
2416 {
2417 return nbd_co_send_sparse_read(client, request->handle, request->from,
2418 data, request->len, errp);
2419 }
2420
2421 ret = blk_pread(exp->common.blk, request->from, data, request->len);
2422 if (ret < 0) {
2423 return nbd_send_generic_reply(client, request->handle, ret,
2424 "reading from file failed", errp);
2425 }
2426
2427 if (client->structured_reply) {
2428 if (request->len) {
2429 return nbd_co_send_structured_read(client, request->handle,
2430 request->from, data,
2431 request->len, true, errp);
2432 } else {
2433 return nbd_co_send_structured_done(client, request->handle, errp);
2434 }
2435 } else {
2436 return nbd_co_send_simple_reply(client, request->handle, 0,
2437 data, request->len, errp);
2438 }
2439 }
2440
2441 /*
2442 * nbd_do_cmd_cache
2443 *
2444 * Handle NBD_CMD_CACHE request.
2445 * Return -errno if sending fails. Other errors are reported directly to the
2446 * client as an error reply.
2447 */
nbd_do_cmd_cache(NBDClient * client,NBDRequest * request,Error ** errp)2448 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2449 Error **errp)
2450 {
2451 int ret;
2452 NBDExport *exp = client->exp;
2453
2454 assert(request->type == NBD_CMD_CACHE);
2455
2456 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2457 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2458
2459 return nbd_send_generic_reply(client, request->handle, ret,
2460 "caching data failed", errp);
2461 }
2462
2463 /* Handle NBD request.
2464 * Return -errno if sending fails. Other errors are reported directly to the
2465 * client as an error reply. */
nbd_handle_request(NBDClient * client,NBDRequest * request,uint8_t * data,Error ** errp)2466 static coroutine_fn int nbd_handle_request(NBDClient *client,
2467 NBDRequest *request,
2468 uint8_t *data, Error **errp)
2469 {
2470 int ret;
2471 int flags;
2472 NBDExport *exp = client->exp;
2473 char *msg;
2474 size_t i;
2475
2476 switch (request->type) {
2477 case NBD_CMD_CACHE:
2478 return nbd_do_cmd_cache(client, request, errp);
2479
2480 case NBD_CMD_READ:
2481 return nbd_do_cmd_read(client, request, data, errp);
2482
2483 case NBD_CMD_WRITE:
2484 flags = 0;
2485 if (request->flags & NBD_CMD_FLAG_FUA) {
2486 flags |= BDRV_REQ_FUA;
2487 }
2488 ret = blk_pwrite(exp->common.blk, request->from, data, request->len,
2489 flags);
2490 return nbd_send_generic_reply(client, request->handle, ret,
2491 "writing to file failed", errp);
2492
2493 case NBD_CMD_WRITE_ZEROES:
2494 flags = 0;
2495 if (request->flags & NBD_CMD_FLAG_FUA) {
2496 flags |= BDRV_REQ_FUA;
2497 }
2498 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2499 flags |= BDRV_REQ_MAY_UNMAP;
2500 }
2501 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2502 flags |= BDRV_REQ_NO_FALLBACK;
2503 }
2504 ret = 0;
2505 /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2506 while (ret >= 0 && request->len) {
2507 int align = client->check_align ?: 1;
2508 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2509 align));
2510 ret = blk_pwrite_zeroes(exp->common.blk, request->from, len, flags);
2511 request->len -= len;
2512 request->from += len;
2513 }
2514 return nbd_send_generic_reply(client, request->handle, ret,
2515 "writing to file failed", errp);
2516
2517 case NBD_CMD_DISC:
2518 /* unreachable, thanks to special case in nbd_co_receive_request() */
2519 abort();
2520
2521 case NBD_CMD_FLUSH:
2522 ret = blk_co_flush(exp->common.blk);
2523 return nbd_send_generic_reply(client, request->handle, ret,
2524 "flush failed", errp);
2525
2526 case NBD_CMD_TRIM:
2527 ret = 0;
2528 /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2529 while (ret >= 0 && request->len) {
2530 int align = client->check_align ?: 1;
2531 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2532 align));
2533 ret = blk_co_pdiscard(exp->common.blk, request->from, len);
2534 request->len -= len;
2535 request->from += len;
2536 }
2537 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2538 ret = blk_co_flush(exp->common.blk);
2539 }
2540 return nbd_send_generic_reply(client, request->handle, ret,
2541 "discard failed", errp);
2542
2543 case NBD_CMD_BLOCK_STATUS:
2544 if (!request->len) {
2545 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2546 "need non-zero length", errp);
2547 }
2548 if (client->export_meta.count) {
2549 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2550 int contexts_remaining = client->export_meta.count;
2551
2552 if (client->export_meta.base_allocation) {
2553 ret = nbd_co_send_block_status(client, request->handle,
2554 blk_bs(exp->common.blk),
2555 request->from,
2556 request->len, dont_fragment,
2557 !--contexts_remaining,
2558 NBD_META_ID_BASE_ALLOCATION,
2559 errp);
2560 if (ret < 0) {
2561 return ret;
2562 }
2563 }
2564
2565 if (client->export_meta.allocation_depth) {
2566 ret = nbd_co_send_block_status(client, request->handle,
2567 blk_bs(exp->common.blk),
2568 request->from, request->len,
2569 dont_fragment,
2570 !--contexts_remaining,
2571 NBD_META_ID_ALLOCATION_DEPTH,
2572 errp);
2573 if (ret < 0) {
2574 return ret;
2575 }
2576 }
2577
2578 for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2579 if (!client->export_meta.bitmaps[i]) {
2580 continue;
2581 }
2582 ret = nbd_co_send_bitmap(client, request->handle,
2583 client->exp->export_bitmaps[i],
2584 request->from, request->len,
2585 dont_fragment, !--contexts_remaining,
2586 NBD_META_ID_DIRTY_BITMAP + i, errp);
2587 if (ret < 0) {
2588 return ret;
2589 }
2590 }
2591
2592 assert(!contexts_remaining);
2593
2594 return 0;
2595 } else {
2596 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2597 "CMD_BLOCK_STATUS not negotiated",
2598 errp);
2599 }
2600
2601 default:
2602 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2603 request->type);
2604 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2605 errp);
2606 g_free(msg);
2607 return ret;
2608 }
2609 }
2610
2611 /* Owns a reference to the NBDClient passed as opaque. */
nbd_trip(void * opaque)2612 static coroutine_fn void nbd_trip(void *opaque)
2613 {
2614 NBDClient *client = opaque;
2615 NBDRequestData *req;
2616 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2617 int ret;
2618 Error *local_err = NULL;
2619
2620 trace_nbd_trip();
2621 if (client->closing) {
2622 nbd_client_put(client);
2623 return;
2624 }
2625
2626 if (client->quiescing) {
2627 /*
2628 * We're switching between AIO contexts. Don't attempt to receive a new
2629 * request and kick the main context which may be waiting for us.
2630 */
2631 nbd_client_put(client);
2632 client->recv_coroutine = NULL;
2633 aio_wait_kick();
2634 return;
2635 }
2636
2637 req = nbd_request_get(client);
2638 ret = nbd_co_receive_request(req, &request, &local_err);
2639 client->recv_coroutine = NULL;
2640
2641 if (client->closing) {
2642 /*
2643 * The client may be closed when we are blocked in
2644 * nbd_co_receive_request()
2645 */
2646 goto done;
2647 }
2648
2649 if (ret == -EAGAIN) {
2650 assert(client->quiescing);
2651 goto done;
2652 }
2653
2654 nbd_client_receive_next_request(client);
2655 if (ret == -EIO) {
2656 goto disconnect;
2657 }
2658
2659 if (ret < 0) {
2660 /* It wans't -EIO, so, according to nbd_co_receive_request()
2661 * semantics, we should return the error to the client. */
2662 Error *export_err = local_err;
2663
2664 local_err = NULL;
2665 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2666 error_get_pretty(export_err), &local_err);
2667 error_free(export_err);
2668 } else {
2669 ret = nbd_handle_request(client, &request, req->data, &local_err);
2670 }
2671 if (ret < 0) {
2672 error_prepend(&local_err, "Failed to send reply: ");
2673 goto disconnect;
2674 }
2675
2676 /* We must disconnect after NBD_CMD_WRITE if we did not
2677 * read the payload.
2678 */
2679 if (!req->complete) {
2680 error_setg(&local_err, "Request handling failed in intermediate state");
2681 goto disconnect;
2682 }
2683
2684 done:
2685 nbd_request_put(req);
2686 nbd_client_put(client);
2687 return;
2688
2689 disconnect:
2690 if (local_err) {
2691 error_reportf_err(local_err, "Disconnect client, due to: ");
2692 }
2693 nbd_request_put(req);
2694 client_close(client, true);
2695 nbd_client_put(client);
2696 }
2697
nbd_client_receive_next_request(NBDClient * client)2698 static void nbd_client_receive_next_request(NBDClient *client)
2699 {
2700 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
2701 !client->quiescing) {
2702 nbd_client_get(client);
2703 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2704 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
2705 }
2706 }
2707
nbd_co_client_start(void * opaque)2708 static coroutine_fn void nbd_co_client_start(void *opaque)
2709 {
2710 NBDClient *client = opaque;
2711 Error *local_err = NULL;
2712
2713 qemu_co_mutex_init(&client->send_lock);
2714
2715 if (nbd_negotiate(client, &local_err)) {
2716 if (local_err) {
2717 error_report_err(local_err);
2718 }
2719 client_close(client, false);
2720 return;
2721 }
2722
2723 nbd_client_receive_next_request(client);
2724 }
2725
2726 /*
2727 * Create a new client listener using the given channel @sioc.
2728 * Begin servicing it in a coroutine. When the connection closes, call
2729 * @close_fn with an indication of whether the client completed negotiation.
2730 */
nbd_client_new(QIOChannelSocket * sioc,QCryptoTLSCreds * tlscreds,const char * tlsauthz,void (* close_fn)(NBDClient *,bool))2731 void nbd_client_new(QIOChannelSocket *sioc,
2732 QCryptoTLSCreds *tlscreds,
2733 const char *tlsauthz,
2734 void (*close_fn)(NBDClient *, bool))
2735 {
2736 NBDClient *client;
2737 Coroutine *co;
2738
2739 client = g_new0(NBDClient, 1);
2740 client->refcount = 1;
2741 client->tlscreds = tlscreds;
2742 if (tlscreds) {
2743 object_ref(OBJECT(client->tlscreds));
2744 }
2745 client->tlsauthz = g_strdup(tlsauthz);
2746 client->sioc = sioc;
2747 object_ref(OBJECT(client->sioc));
2748 client->ioc = QIO_CHANNEL(sioc);
2749 object_ref(OBJECT(client->ioc));
2750 client->close_fn = close_fn;
2751
2752 co = qemu_coroutine_create(nbd_co_client_start, client);
2753 qemu_coroutine_enter(co);
2754 }
2755