]> git.proxmox.com Git - mirror_qemu.git/blame - nbd/server.c
projects / mirror_qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bitops.h: revert db1ffc32dd ("qemu/bitops.h: add bitrev8 implementation")
[mirror_qemu.git] / nbd / server.c
CommitLineData
75818250 1/*
ebd57062 2 * Copyright (C) 2016-2020 Red Hat, Inc.
7a5ca864
FB		 3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 *
798bfe00 5 * Network Block Device Server Side
7a5ca864
FB		 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
8167ee88 17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
75818250 18 */
7a5ca864 19
d38ea87a 20#include "qemu/osdep.h"
56ee8626
KW		 21
22#include "block/export.h"
da34e65c 23#include "qapi/error.h"
dc5e9ac7 24#include "qemu/queue.h"
9588463e 25#include "trace.h"
798bfe00 26#include "nbd-internal.h"
416e34bd 27#include "qemu/units.h"
ca441480 28
e7b1948d 29#define NBD_META_ID_BASE_ALLOCATION 0
71719cd5 30#define NBD_META_ID_ALLOCATION_DEPTH 1
3b1f244c 31/* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
71719cd5 32#define NBD_META_ID_DIRTY_BITMAP 2
3d068aff 33
416e34bd
EB		 34/*
35 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
3d068aff
VSO		 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
416e34bd
EB		 38 * the reply as a denial of service attack.
39 */
40#define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
e7b1948d 41
ca441480
PB		 42static int system_errno_to_nbd_errno(int err)
43{
44 switch (err) {
45 case 0:
46 return NBD_SUCCESS;
47 case EPERM:
c0301fcc 48 case EROFS:
ca441480
PB		 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;
bae245d1
EB		 60 case EOVERFLOW:
61 return NBD_EOVERFLOW;
0a479545
EB		 62 case ENOTSUP:
63#if ENOTSUP != EOPNOTSUPP
64 case EOPNOTSUPP:
65#endif
66 return NBD_ENOTSUP;
b6f5d3b5
EB		 67 case ESHUTDOWN:
68 return NBD_ESHUTDOWN;
ca441480
PB		 69 case EINVAL:
70 default:
71 return NBD_EINVAL;
72 }
73}
74
9a304d29
PB		 75/* Definitions for opaque data types */
76
315f78ab 77typedef struct NBDRequestData NBDRequestData;
9a304d29 78
315f78ab
EB		 79struct NBDRequestData {
80 QSIMPLEQ_ENTRY(NBDRequestData) entry;
9a304d29
PB		 81 NBDClient *client;
82 uint8_t *data;
29b6c3b3 83 bool complete;
9a304d29
PB		 84};
85
86struct NBDExport {
56ee8626 87 BlockExport common;
0ddf08db 88
ee0a19ec 89 char *name;
b1a75b33 90 char *description;
9d26dfcb 91 uint64_t size;
7423f417 92 uint16_t nbdflags;
4b9441f6 93 QTAILQ_HEAD(, NBDClient) clients;
ee0a19ec 94 QTAILQ_ENTRY(NBDExport) next;
958c717d 95
cd7fca95 96 BlockBackend *eject_notifier_blk;
741cc431 97 Notifier eject_notifier;
3d068aff 98
71719cd5 99 bool allocation_depth;
3b1f244c
EB		 100 BdrvDirtyBitmap **export_bitmaps;
101 size_t nr_export_bitmaps;
9a304d29
PB		 102};
103
ee0a19ec
PB		 104static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
105
e7b1948d
VSO		 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. */
109typedef struct NBDExportMetaContexts {
af736e54 110 NBDExport *exp;
47ec485e 111 size_t count; /* number of negotiated contexts */
e7b1948d 112 bool base_allocation; /* export base:allocation context (block status) */
71719cd5 113 bool allocation_depth; /* export qemu:allocation-depth */
3b1f244c
EB		 114 bool *bitmaps; /*
115 * export qemu:dirty-bitmap:<export bitmap name>,
116 * sized by exp->nr_export_bitmaps
117 */
e7b1948d
VSO		 118} NBDExportMetaContexts;
119
9a304d29
PB		 120struct NBDClient {
121 int refcount;
0c9390d9 122 void (*close_fn)(NBDClient *client, bool negotiated);
9a304d29
PB		 123
124 NBDExport *exp;
f95910fe 125 QCryptoTLSCreds *tlscreds;
b25e12da 126 char *tlsauthz;
1c778ef7
DB		 127 QIOChannelSocket *sioc; /* The underlying data channel */
128 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
9a304d29
PB		 129
130 Coroutine *recv_coroutine;
131
132 CoMutex send_lock;
133 Coroutine *send_coroutine;
134
f148ae7d
SL		 135 bool read_yielding;
136 bool quiescing;
137
4b9441f6 138 QTAILQ_ENTRY(NBDClient) next;
9a304d29 139 int nb_requests;
ff2b68aa 140 bool closing;
5c54e7fa 141
6e280648
EB		 142 uint32_t check_align; /* If non-zero, check for aligned client requests */
143
5c54e7fa 144 bool structured_reply;
e7b1948d 145 NBDExportMetaContexts export_meta;
9a304d29 146
0cfae925
VSO		 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};
7a5ca864 151
ff82911c 152static void nbd_client_receive_next_request(NBDClient *client);
958c717d 153
6b8c01e7 154/* Basic flow for negotiation
7a5ca864
FB		 155
156 Server Client
7a5ca864 157 Negotiate
6b8c01e7
PB		 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
7a5ca864
FB		 171 Request
172 Response
173 Request
174 Response
175 ...
176 ...
177 Request (type == 2)
6b8c01e7 178
7a5ca864
FB		 179*/
180
1d17922a
VSO		 181static 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
526e5c65
EB		 190/* Send a reply header, including length, but no payload.
191 * Return -errno on error, 0 on success. */
0cfae925
VSO		 192static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
193 uint32_t len, Error **errp)
6b8c01e7 194{
1d17922a 195 NBDOptionReply rep;
6b8c01e7 196
1d17922a 197 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
3736cc5b 198 type, nbd_rep_lookup(type), len);
f95910fe 199
f37708f6 200 assert(len < NBD_MAX_BUFFER_SIZE);
2fd2c840 201
1d17922a
VSO		 202 set_be_option_rep(&rep, client->opt, type, len);
203 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
f5076b5a 204}
6b8c01e7 205
526e5c65
EB		 206/* Send a reply header with default 0 length.
207 * Return -errno on error, 0 on success. */
0cfae925 208static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
2fd2c840 209 Error **errp)
526e5c65 210{
0cfae925 211 return nbd_negotiate_send_rep_len(client, type, 0, errp);
526e5c65
EB		 212}
213
36683283
EB		 214/* Send an error reply.
215 * Return -errno on error, 0 on success. */
41f5dfaf
EB		 216static int GCC_FMT_ATTR(4, 0)
217nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
218 Error **errp, const char *fmt, va_list va)
36683283 219{
795d946d 220 ERRP_GUARD();
df18c04e 221 g_autofree char *msg = NULL;
36683283
EB		 222 int ret;
223 size_t len;
224
36683283 225 msg = g_strdup_vprintf(fmt, va);
36683283 226 len = strlen(msg);
5c4fe018 227 assert(len < NBD_MAX_STRING_SIZE);
9588463e 228 trace_nbd_negotiate_send_rep_err(msg);
0cfae925 229 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
36683283 230 if (ret < 0) {
df18c04e 231 return ret;
36683283 232 }
0cfae925 233 if (nbd_write(client->ioc, msg, len, errp) < 0) {
2fd2c840 234 error_prepend(errp, "write failed (error message): ");
df18c04e 235 return -EIO;
36683283 236 }
2fd2c840 237
df18c04e 238 return 0;
36683283
EB		 239}
240
5c4fe018
EB		 241/*
242 * Return a malloc'd copy of @name suitable for use in an error reply.
243 */
244static char *
245nbd_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
41f5dfaf
EB		 254/* Send an error reply.
255 * Return -errno on error, 0 on success. */
256static int GCC_FMT_ATTR(4, 5)
257nbd_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
894e0280
EB		 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. */
2e425fd5
VSO		 272static int GCC_FMT_ATTR(4, 0)
273nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
274 const char *fmt, va_list va)
894e0280
EB		 275{
276 int ret = nbd_drop(client->ioc, client->optlen, errp);
894e0280
EB		 277
278 client->optlen = 0;
279 if (!ret) {
894e0280 280 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
894e0280
EB		 281 }
282 return ret;
283}
284
2e425fd5
VSO		 285static int GCC_FMT_ATTR(4, 5)
286nbd_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
299static int GCC_FMT_ATTR(3, 4)
300nbd_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
894e0280 312/* Read size bytes from the unparsed payload of the current option.
d1e2c3e7 313 * If @check_nul, require that no NUL bytes appear in buffer.
894e0280
EB		 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. */
316static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
d1e2c3e7 317 bool check_nul, Error **errp)
894e0280
EB		 318{
319 if (size > client->optlen) {
2e425fd5
VSO		 320 return nbd_opt_invalid(client, errp,
321 "Inconsistent lengths in option %s",
322 nbd_opt_lookup(client->opt));
894e0280
EB		 323 }
324 client->optlen -= size;
d1e2c3e7
EB		 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;
894e0280
EB		 335}
336
e7b1948d
VSO		 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. */
340static 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
12296459
VSO		 351/* nbd_opt_read_name
352 *
353 * Read a string with the format:
93676c88 354 * uint32_t len (<= NBD_MAX_STRING_SIZE)
12296459
VSO		 355 * len bytes string (not 0-terminated)
356 *
9d7ab222 357 * On success, @name will be allocated.
12296459
VSO		 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 */
9d7ab222 363static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
12296459
VSO		 364 Error **errp)
365{
366 int ret;
367 uint32_t len;
9d7ab222 368 g_autofree char *local_name = NULL;
12296459 369
9d7ab222 370 *name = NULL;
d1e2c3e7 371 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
12296459
VSO		 372 if (ret <= 0) {
373 return ret;
374 }
80c7c2b0 375 len = cpu_to_be32(len);
12296459 376
93676c88 377 if (len > NBD_MAX_STRING_SIZE) {
12296459
VSO		 378 return nbd_opt_invalid(client, errp,
379 "Invalid name length: %" PRIu32, len);
380 }
381
9d7ab222 382 local_name = g_malloc(len + 1);
d1e2c3e7 383 ret = nbd_opt_read(client, local_name, len, true, errp);
12296459
VSO		 384 if (ret <= 0) {
385 return ret;
386 }
9d7ab222 387 local_name[len] = '\0';
12296459
VSO		 388
389 if (length) {
390 *length = len;
391 }
9d7ab222 392 *name = g_steal_pointer(&local_name);
12296459
VSO		 393
394 return 1;
395}
396
526e5c65
EB		 397/* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
398 * Return -errno on error, 0 on success. */
0cfae925 399static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
2fd2c840 400 Error **errp)
32d7d2e0 401{
795d946d 402 ERRP_GUARD();
b1a75b33 403 size_t name_len, desc_len;
526e5c65 404 uint32_t len;
b1a75b33
EB		 405 const char *name = exp->name ? exp->name : "";
406 const char *desc = exp->description ? exp->description : "";
0cfae925 407 QIOChannel *ioc = client->ioc;
2e5c9ad6 408 int ret;
32d7d2e0 409
9588463e 410 trace_nbd_negotiate_send_rep_list(name, desc);
b1a75b33
EB		 411 name_len = strlen(name);
412 desc_len = strlen(desc);
93676c88 413 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
526e5c65 414 len = name_len + desc_len + sizeof(len);
0cfae925 415 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
2e5c9ad6
VSO		 416 if (ret < 0) {
417 return ret;
32d7d2e0 418 }
526e5c65 419
32d7d2e0 420 len = cpu_to_be32(name_len);
2fd2c840
VSO		 421 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
422 error_prepend(errp, "write failed (name length): ");
b1a75b33
EB		 423 return -EINVAL;
424 }
2fd2c840
VSO		 425
426 if (nbd_write(ioc, name, name_len, errp) < 0) {
427 error_prepend(errp, "write failed (name buffer): ");
32d7d2e0
HB		 428 return -EINVAL;
429 }
2fd2c840
VSO		 430
431 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
432 error_prepend(errp, "write failed (description buffer): ");
32d7d2e0
HB		 433 return -EINVAL;
434 }
2fd2c840 435
32d7d2e0
HB		 436 return 0;
437}
438
526e5c65
EB		 439/* Process the NBD_OPT_LIST command, with a potential series of replies.
440 * Return -errno on error, 0 on success. */
e68c35cf 441static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
32d7d2e0 442{
32d7d2e0 443 NBDExport *exp;
0cfae925 444 assert(client->opt == NBD_OPT_LIST);
32d7d2e0 445
32d7d2e0
HB		 446 /* For each export, send a NBD_REP_SERVER reply. */
447 QTAILQ_FOREACH(exp, &exports, next) {
0cfae925 448 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
32d7d2e0
HB		 449 return -EINVAL;
450 }
451 }
452 /* Finish with a NBD_REP_ACK. */
0cfae925 453 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
32d7d2e0
HB		 454}
455
af736e54 456static void nbd_check_meta_export(NBDClient *client)
e7b1948d 457{
47ec485e
EB		 458 if (client->exp != client->export_meta.exp) {
459 client->export_meta.count = 0;
460 }
e7b1948d
VSO		 461}
462
f37708f6
EB		 463/* Send a reply to NBD_OPT_EXPORT_NAME.
464 * Return -errno on error, 0 on success. */
dbb38caa 465static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
2fd2c840 466 Error **errp)
f5076b5a 467{
795d946d 468 ERRP_GUARD();
9d7ab222 469 g_autofree char *name = NULL;
5f66d060 470 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
23e099c3
EB		 471 size_t len;
472 int ret;
dbb38caa 473 uint16_t myflags;
6b8c01e7 474
f5076b5a
HB		 475 /* Client sends:
476 [20 .. xx] export name (length bytes)
5f66d060
EB		 477 Server replies:
478 [ 0 .. 7] size
479 [ 8 .. 9] export flags
480 [10 .. 133] reserved (0) [unless no_zeroes]
f5076b5a 481 */
9588463e 482 trace_nbd_negotiate_handle_export_name();
93676c88 483 if (client->optlen > NBD_MAX_STRING_SIZE) {
2fd2c840 484 error_setg(errp, "Bad length received");
d9faeed8 485 return -EINVAL;
6b8c01e7 486 }
9d7ab222 487 name = g_malloc(client->optlen + 1);
e6798f06 488 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
32f158a6 489 return -EIO;
6b8c01e7 490 }
0cfae925
VSO		 491 name[client->optlen] = '\0';
492 client->optlen = 0;
6b8c01e7 493
9588463e 494 trace_nbd_negotiate_handle_export_name_request(name);
9344e5f5 495
6b8c01e7
PB		 496 client->exp = nbd_export_find(name);
497 if (!client->exp) {
2fd2c840 498 error_setg(errp, "export not found");
d9faeed8 499 return -EINVAL;
6b8c01e7
PB		 500 }
501
dbb38caa
EB		 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);
23e099c3 507 stq_be_p(buf, client->exp->size);
dbb38caa 508 stw_be_p(buf + 8, myflags);
23e099c3
EB		 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
6b8c01e7 516 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
c69de1be 517 blk_exp_ref(&client->exp->common);
af736e54 518 nbd_check_meta_export(client);
d9faeed8
VSO		 519
520 return 0;
6b8c01e7
PB		 521}
522
f37708f6
EB		 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. */
0cfae925 526static int nbd_negotiate_send_info(NBDClient *client,
f37708f6
EB		 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);
0cfae925 533 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
f37708f6
EB		 534 sizeof(info) + length, errp);
535 if (rc < 0) {
536 return rc;
537 }
80c7c2b0 538 info = cpu_to_be16(info);
f37708f6
EB		 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
a16a7907
EB		 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 */
0cfae925 555static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
a16a7907
EB		 556{
557 int ret;
558
0cfae925 559 assert(client->optlen);
2e425fd5
VSO		 560 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
561 nbd_opt_lookup(client->opt));
a16a7907 562 if (fatal && !ret) {
894e0280 563 error_setg(errp, "option '%s' has unexpected length",
0cfae925 564 nbd_opt_lookup(client->opt));
a16a7907
EB		 565 return -EINVAL;
566 }
567 return ret;
568}
569
f37708f6
EB		 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. */
dbb38caa 573static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
f37708f6
EB		 574{
575 int rc;
9d7ab222 576 g_autofree char *name = NULL;
f37708f6
EB		 577 NBDExport *exp;
578 uint16_t requests;
579 uint16_t request;
bbc35fc2 580 uint32_t namelen = 0;
f37708f6 581 bool sendname = false;
0c1d50bd
EB		 582 bool blocksize = false;
583 uint32_t sizes[3];
f37708f6 584 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
6e280648 585 uint32_t check_align = 0;
dbb38caa 586 uint16_t myflags;
f37708f6
EB		 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 */
9d7ab222 594 rc = nbd_opt_read_name(client, &name, &namelen, errp);
894e0280
EB		 595 if (rc <= 0) {
596 return rc;
f37708f6 597 }
f37708f6
EB		 598 trace_nbd_negotiate_handle_export_name_request(name);
599
d1e2c3e7 600 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
894e0280
EB		 601 if (rc <= 0) {
602 return rc;
f37708f6 603 }
80c7c2b0 604 requests = be16_to_cpu(requests);
f37708f6 605 trace_nbd_negotiate_handle_info_requests(requests);
f37708f6 606 while (requests--) {
d1e2c3e7 607 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
894e0280
EB		 608 if (rc <= 0) {
609 return rc;
f37708f6 610 }
80c7c2b0 611 request = be16_to_cpu(request);
f37708f6
EB		 612 trace_nbd_negotiate_handle_info_request(request,
613 nbd_info_lookup(request));
0c1d50bd
EB		 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:
f37708f6 619 sendname = true;
0c1d50bd
EB		 620 break;
621 case NBD_INFO_BLOCK_SIZE:
622 blocksize = true;
623 break;
f37708f6
EB		 624 }
625 }
894e0280
EB		 626 if (client->optlen) {
627 return nbd_reject_length(client, false, errp);
628 }
f37708f6
EB		 629
630 exp = nbd_export_find(name);
631 if (!exp) {
5c4fe018
EB		 632 g_autofree char *sane_name = nbd_sanitize_name(name);
633
0cfae925
VSO		 634 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
635 errp, "export '%s' not present",
5c4fe018 636 sane_name);
f37708f6
EB		 637 }
638
639 /* Don't bother sending NBD_INFO_NAME unless client requested it */
640 if (sendname) {
0cfae925 641 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
f37708f6
EB		 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
93676c88 653 assert(len <= NBD_MAX_STRING_SIZE);
0cfae925 654 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
f37708f6
EB		 655 len, exp->description, errp);
656 if (rc < 0) {
657 return rc;
658 }
659 }
660
0c1d50bd
EB		 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. */
b0245d64
EB		 664 /* minimum - 1 for back-compat, or actual if client will obey it. */
665 if (client->opt == NBD_OPT_INFO || blocksize) {
37a4f70c 666 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
b0245d64
EB		 667 } else {
668 sizes[0] = 1;
669 }
670 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
0c1d50bd
EB		 671 /* preferred - Hard-code to 4096 for now.
672 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
b0245d64 673 sizes[1] = MAX(4096, sizes[0]);
0c1d50bd 674 /* maximum - At most 32M, but smaller as appropriate. */
37a4f70c 675 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
0c1d50bd 676 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
80c7c2b0
PM		 677 sizes[0] = cpu_to_be32(sizes[0]);
678 sizes[1] = cpu_to_be32(sizes[1]);
679 sizes[2] = cpu_to_be32(sizes[2]);
0cfae925 680 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
0c1d50bd
EB		 681 sizeof(sizes), sizes, errp);
682 if (rc < 0) {
683 return rc;
684 }
685
f37708f6 686 /* Send NBD_INFO_EXPORT always */
dbb38caa
EB		 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);
f37708f6 692 stq_be_p(buf, exp->size);
dbb38caa 693 stw_be_p(buf + 8, myflags);
0cfae925 694 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
f37708f6
EB		 695 sizeof(buf), buf, errp);
696 if (rc < 0) {
697 return rc;
698 }
699
099fbcd6
EB		 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 &&
37a4f70c 707 blk_get_request_alignment(exp->common.blk) > 1) {
0cfae925
VSO		 708 return nbd_negotiate_send_rep_err(client,
709 NBD_REP_ERR_BLOCK_SIZE_REQD,
0c1d50bd
EB		 710 errp,
711 "request NBD_INFO_BLOCK_SIZE to "
712 "use this export");
713 }
714
f37708f6 715 /* Final reply */
0cfae925 716 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
f37708f6
EB		 717 if (rc < 0) {
718 return rc;
719 }
720
0cfae925 721 if (client->opt == NBD_OPT_GO) {
f37708f6 722 client->exp = exp;
6e280648 723 client->check_align = check_align;
f37708f6 724 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
c69de1be 725 blk_exp_ref(&client->exp->common);
af736e54 726 nbd_check_meta_export(client);
f37708f6
EB		 727 rc = 1;
728 }
729 return rc;
f37708f6
EB		 730}
731
732
36683283
EB		 733/* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
734 * new channel for all further (now-encrypted) communication. */
f95910fe 735static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
2fd2c840 736 Error **errp)
f95910fe
DB		 737{
738 QIOChannel *ioc;
739 QIOChannelTLS *tioc;
740 struct NBDTLSHandshakeData data = { 0 };
741
0cfae925
VSO		 742 assert(client->opt == NBD_OPT_STARTTLS);
743
9588463e 744 trace_nbd_negotiate_handle_starttls();
f95910fe 745 ioc = client->ioc;
f95910fe 746
0cfae925 747 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
63d5ef86
EB		 748 return NULL;
749 }
f95910fe
DB		 750
751 tioc = qio_channel_tls_new_server(ioc,
752 client->tlscreds,
b25e12da 753 client->tlsauthz,
2fd2c840 754 errp);
f95910fe
DB		 755 if (!tioc) {
756 return NULL;
757 }
758
0d73f725 759 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
9588463e 760 trace_nbd_negotiate_handle_starttls_handshake();
f95910fe
DB		 761 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
762 qio_channel_tls_handshake(tioc,
763 nbd_tls_handshake,
764 &data,
1939ccda 765 NULL,
f95910fe
DB		 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));
2fd2c840 774 error_propagate(errp, data.error);
f95910fe
DB		 775 return NULL;
776 }
777
778 return QIO_CHANNEL(tioc);
779}
780
e7b1948d
VSO		 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 */
787static 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
93676c88 798 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
e7b1948d
VSO		 799 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
800 context_id = 0;
801 }
802
2b53af25 803 trace_nbd_negotiate_meta_query_reply(context, context_id);
e7b1948d
VSO		 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
ebd57062
EB		 811/*
812 * Return true if @query matches @pattern, or if @query is empty when
813 * the @client is performing _LIST_.
dbb8b396 814 */
ebd57062
EB		 815static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
816 const char *query)
e7b1948d 817{
ebd57062
EB		 818 if (!*query) {
819 trace_nbd_negotiate_meta_query_parse("empty");
820 return client->opt == NBD_OPT_LIST_META_CONTEXT;
e7b1948d 821 }
ebd57062 822 if (strcmp(query, pattern) == 0) {
b0769d8f 823 trace_nbd_negotiate_meta_query_parse(pattern);
ebd57062 824 return true;
e7b1948d 825 }
ebd57062
EB		 826 trace_nbd_negotiate_meta_query_skip("pattern not matched");
827 return false;
e7b1948d
VSO		 828}
829
b0769d8f 830/*
ebd57062 831 * Return true and adjust @str in place if it begins with @prefix.
b0769d8f 832 */
ebd57062 833static bool nbd_strshift(const char **str, const char *prefix)
b0769d8f 834{
ebd57062 835 size_t len = strlen(prefix);
b0769d8f 836
ebd57062
EB		 837 if (strncmp(*str, prefix, len) == 0) {
838 *str += len;
839 return true;
b0769d8f 840 }
ebd57062 841 return false;
b0769d8f
VSO		 842}
843
844/* nbd_meta_base_query
845 *
846 * Handle queries to 'base' namespace. For now, only the base:allocation
ebd57062 847 * context is available. Return true if @query has been handled.
b0769d8f 848 */
ebd57062
EB		 849static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
850 const char *query)
b0769d8f 851{
ebd57062
EB		 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;
b0769d8f
VSO		 861}
862
ebd57062 863/* nbd_meta_qemu_query
3d068aff 864 *
ebd57062 865 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
71719cd5
EB		 866 * and qemu:allocation-depth contexts are available. Return true if @query
867 * has been handled.
ebd57062
EB		 868 */
869static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
870 const char *query)
3d068aff 871{
3b1f244c
EB		 872 size_t i;
873
ebd57062
EB		 874 if (!nbd_strshift(&query, "qemu:")) {
875 return false;
3d068aff 876 }
ebd57062 877 trace_nbd_negotiate_meta_query_parse("qemu:");
3d068aff 878
ebd57062 879 if (!*query) {
3d068aff 880 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
71719cd5 881 meta->allocation_depth = meta->exp->allocation_depth;
3b1f244c 882 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
3d068aff
VSO		 883 }
884 trace_nbd_negotiate_meta_query_parse("empty");
ebd57062 885 return true;
3d068aff
VSO		 886 }
887
71719cd5
EB		 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
ebd57062
EB		 894 if (nbd_strshift(&query, "dirty-bitmap:")) {
895 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
3b1f244c
EB		 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");
ebd57062
EB		 901 return true;
902 }
3b1f244c
EB		 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 }
ebd57062 913 }
3b1f244c 914 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
ebd57062 915 return true;
3d068aff
VSO		 916 }
917
71719cd5 918 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
ebd57062 919 return true;
3d068aff
VSO		 920}
921
e7b1948d
VSO		 922/* nbd_negotiate_meta_query
923 *
924 * Parse namespace name and call corresponding function to parse body of the
925 * query.
926 *
93676c88 927 * The only supported namespaces are 'base' and 'qemu'.
e7b1948d 928 *
e7b1948d
VSO		 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. */
931static int nbd_negotiate_meta_query(NBDClient *client,
932 NBDExportMetaContexts *meta, Error **errp)
933{
934 int ret;
ebd57062 935 g_autofree char *query = NULL;
e7b1948d
VSO		 936 uint32_t len;
937
d1e2c3e7 938 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
e7b1948d
VSO		 939 if (ret <= 0) {
940 return ret;
941 }
80c7c2b0 942 len = cpu_to_be32(len);
e7b1948d 943
93676c88
EB		 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 }
e7b1948d 948
ebd57062
EB		 949 query = g_malloc(len + 1);
950 ret = nbd_opt_read(client, query, len, true, errp);
e7b1948d
VSO		 951 if (ret <= 0) {
952 return ret;
953 }
ebd57062 954 query[len] = '\0';
3d068aff 955
ebd57062
EB		 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;
e7b1948d
VSO		 961 }
962
3d068aff 963 trace_nbd_negotiate_meta_query_skip("unknown namespace");
ebd57062 964 return 1;
e7b1948d
VSO		 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. */
971static int nbd_negotiate_meta_queries(NBDClient *client,
972 NBDExportMetaContexts *meta, Error **errp)
973{
974 int ret;
9d7ab222 975 g_autofree char *export_name = NULL;
3b1f244c
EB		 976 g_autofree bool *bitmaps = NULL;
977 NBDExportMetaContexts local_meta = {0};
e7b1948d 978 uint32_t nb_queries;
3b1f244c 979 size_t i;
47ec485e 980 size_t count = 0;
e7b1948d
VSO		 981
982 if (!client->structured_reply) {
983 return nbd_opt_invalid(client, errp,
984 "request option '%s' when structured reply "
985 "is not negotiated",
986 nbd_opt_lookup(client->opt));
987 }
988
989 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
990 /* Only change the caller's meta on SET. */
991 meta = &local_meta;
992 }
993
3b1f244c 994 g_free(meta->bitmaps);
e7b1948d
VSO		 995 memset(meta, 0, sizeof(*meta));
996
9d7ab222 997 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
e7b1948d
VSO		 998 if (ret <= 0) {
999 return ret;
1000 }
1001
af736e54
VSO		 1002 meta->exp = nbd_export_find(export_name);
1003 if (meta->exp == NULL) {
5c4fe018
EB		 1004 g_autofree char *sane_name = nbd_sanitize_name(export_name);
1005
e7b1948d 1006 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
5c4fe018 1007 "export '%s' not present", sane_name);
e7b1948d 1008 }
3b1f244c
EB		 1009 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1010 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1011 bitmaps = meta->bitmaps;
1012 }
e7b1948d 1013
d1e2c3e7 1014 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
e7b1948d
VSO		 1015 if (ret <= 0) {
1016 return ret;
1017 }
80c7c2b0 1018 nb_queries = cpu_to_be32(nb_queries);
2b53af25 1019 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
af736e54 1020 export_name, nb_queries);
e7b1948d
VSO		 1021
1022 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1023 /* enable all known contexts */
1024 meta->base_allocation = true;
71719cd5 1025 meta->allocation_depth = meta->exp->allocation_depth;
3b1f244c 1026 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
e7b1948d
VSO		 1027 } else {
1028 for (i = 0; i < nb_queries; ++i) {
1029 ret = nbd_negotiate_meta_query(client, meta, errp);
1030 if (ret <= 0) {
1031 return ret;
1032 }
1033 }
1034 }
1035
1036 if (meta->base_allocation) {
1037 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1038 NBD_META_ID_BASE_ALLOCATION,
1039 errp);
1040 if (ret < 0) {
1041 return ret;
1042 }
47ec485e 1043 count++;
e7b1948d
VSO		 1044 }
1045
71719cd5
EB		 1046 if (meta->allocation_depth) {
1047 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1048 NBD_META_ID_ALLOCATION_DEPTH,
1049 errp);
1050 if (ret < 0) {
1051 return ret;
1052 }
1053 count++;
1054 }
1055
3b1f244c
EB		 1056 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1057 const char *bm_name;
1058 g_autofree char *context = NULL;
1059
1060 if (!meta->bitmaps[i]) {
1061 continue;
1062 }
1063
1064 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1065 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
02e87e3b
EB		 1066
1067 ret = nbd_negotiate_send_meta_context(client, context,
3b1f244c 1068 NBD_META_ID_DIRTY_BITMAP + i,
3d068aff
VSO		 1069 errp);
1070 if (ret < 0) {
1071 return ret;
1072 }
47ec485e 1073 count++;
3d068aff
VSO		 1074 }
1075
e7b1948d
VSO		 1076 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1077 if (ret == 0) {
47ec485e 1078 meta->count = count;
e7b1948d
VSO		 1079 }
1080
1081 return ret;
1082}
1083
1e120ffe 1084/* nbd_negotiate_options
f37708f6
EB		 1085 * Process all NBD_OPT_* client option commands, during fixed newstyle
1086 * negotiation.
1e120ffe 1087 * Return:
2fd2c840
VSO		 1088 * -errno on error, errp is set
1089 * 0 on successful negotiation, errp is not set
1090 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1091 * errp is not set
1e120ffe 1092 */
dbb38caa 1093static int nbd_negotiate_options(NBDClient *client, Error **errp)
f5076b5a 1094{
9c122ada 1095 uint32_t flags;
26afa868 1096 bool fixedNewstyle = false;
23e099c3 1097 bool no_zeroes = false;
9c122ada
HR		 1098
1099 /* Client sends:
1100 [ 0 .. 3] client flags
1101
f37708f6 1102 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
9c122ada
HR		 1103 [ 0 .. 7] NBD_OPTS_MAGIC
1104 [ 8 .. 11] NBD option
1105 [12 .. 15] Data length
1106 ... Rest of request
1107
1108 [ 0 .. 7] NBD_OPTS_MAGIC
1109 [ 8 .. 11] Second NBD option
1110 [12 .. 15] Data length
1111 ... Rest of request
1112 */
1113
e6798f06 1114 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
9c122ada
HR		 1115 return -EIO;
1116 }
621c4f4e 1117 trace_nbd_negotiate_options_flags(flags);
26afa868 1118 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
26afa868
DB		 1119 fixedNewstyle = true;
1120 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1121 }
c203c59a 1122 if (flags & NBD_FLAG_C_NO_ZEROES) {
23e099c3 1123 no_zeroes = true;
c203c59a
EB		 1124 flags &= ~NBD_FLAG_C_NO_ZEROES;
1125 }
26afa868 1126 if (flags != 0) {
2fd2c840 1127 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
621c4f4e 1128 return -EINVAL;
9c122ada
HR		 1129 }
1130
f5076b5a 1131 while (1) {
9c122ada 1132 int ret;
7f9039cd 1133 uint32_t option, length;
f5076b5a
HB		 1134 uint64_t magic;
1135
e6798f06 1136 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
f5076b5a
HB		 1137 return -EINVAL;
1138 }
9588463e
VSO		 1139 trace_nbd_negotiate_options_check_magic(magic);
1140 if (magic != NBD_OPTS_MAGIC) {
2fd2c840 1141 error_setg(errp, "Bad magic received");
f5076b5a
HB		 1142 return -EINVAL;
1143 }
1144
e6798f06 1145 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
f5076b5a
HB		 1146 return -EINVAL;
1147 }
0cfae925 1148 client->opt = option;
f5076b5a 1149
e6798f06 1150 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
f5076b5a
HB		 1151 return -EINVAL;
1152 }
894e0280 1153 assert(!client->optlen);
0cfae925 1154 client->optlen = length;
f5076b5a 1155
fdad35ef
EB		 1156 if (length > NBD_MAX_BUFFER_SIZE) {
1157 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1158 length, NBD_MAX_BUFFER_SIZE);
1159 return -EINVAL;
1160 }
1161
3736cc5b
EB		 1162 trace_nbd_negotiate_options_check_option(option,
1163 nbd_opt_lookup(option));
f95910fe
DB		 1164 if (client->tlscreds &&
1165 client->ioc == (QIOChannel *)client->sioc) {
1166 QIOChannel *tioc;
1167 if (!fixedNewstyle) {
7f9039cd 1168 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
f95910fe
DB		 1169 return -EINVAL;
1170 }
7f9039cd 1171 switch (option) {
f95910fe 1172 case NBD_OPT_STARTTLS:
e68c35cf
EB		 1173 if (length) {
1174 /* Unconditionally drop the connection if the client
1175 * can't start a TLS negotiation correctly */
0cfae925 1176 return nbd_reject_length(client, true, errp);
e68c35cf
EB		 1177 }
1178 tioc = nbd_negotiate_handle_starttls(client, errp);
f95910fe
DB		 1179 if (!tioc) {
1180 return -EIO;
1181 }
8cbee49e 1182 ret = 0;
f95910fe
DB		 1183 object_unref(OBJECT(client->ioc));
1184 client->ioc = QIO_CHANNEL(tioc);
1185 break;
1186
d1129a8a
EB		 1187 case NBD_OPT_EXPORT_NAME:
1188 /* No way to return an error to client, so drop connection */
2fd2c840 1189 error_setg(errp, "Option 0x%x not permitted before TLS",
7f9039cd 1190 option);
d1129a8a
EB		 1191 return -EINVAL;
1192
f95910fe 1193 default:
3e99ebb9
EB		 1194 /* Let the client keep trying, unless they asked to
1195 * quit. Always try to give an error back to the
1196 * client; but when replying to OPT_ABORT, be aware
1197 * that the client may hang up before receiving the
1198 * error, in which case we are fine ignoring the
1199 * resulting EPIPE. */
1200 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1201 option == NBD_OPT_ABORT ? NULL : errp,
894e0280 1202 "Option 0x%" PRIx32
0b0bb124 1203 " not permitted before TLS", option);
7f9039cd 1204 if (option == NBD_OPT_ABORT) {
1e120ffe 1205 return 1;
b6f5d3b5 1206 }
d1129a8a 1207 break;
f95910fe
DB		 1208 }
1209 } else if (fixedNewstyle) {
7f9039cd 1210 switch (option) {
26afa868 1211 case NBD_OPT_LIST:
e68c35cf 1212 if (length) {
0cfae925 1213 ret = nbd_reject_length(client, false, errp);
e68c35cf
EB		 1214 } else {
1215 ret = nbd_negotiate_handle_list(client, errp);
1216 }
26afa868
DB		 1217 break;
1218
1219 case NBD_OPT_ABORT:
b6f5d3b5
EB		 1220 /* NBD spec says we must try to reply before
1221 * disconnecting, but that we must also tolerate
1222 * guests that don't wait for our reply. */
0cfae925 1223 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1e120ffe 1224 return 1;
26afa868
DB		 1225
1226 case NBD_OPT_EXPORT_NAME:
dbb38caa 1227 return nbd_negotiate_handle_export_name(client, no_zeroes,
23e099c3 1228 errp);
26afa868 1229
f37708f6
EB		 1230 case NBD_OPT_INFO:
1231 case NBD_OPT_GO:
dbb38caa 1232 ret = nbd_negotiate_handle_info(client, errp);
f37708f6
EB		 1233 if (ret == 1) {
1234 assert(option == NBD_OPT_GO);
1235 return 0;
1236 }
f37708f6
EB		 1237 break;
1238
f95910fe 1239 case NBD_OPT_STARTTLS:
e68c35cf 1240 if (length) {
0cfae925 1241 ret = nbd_reject_length(client, false, errp);
e68c35cf 1242 } else if (client->tlscreds) {
0cfae925
VSO		 1243 ret = nbd_negotiate_send_rep_err(client,
1244 NBD_REP_ERR_INVALID, errp,
36683283 1245 "TLS already enabled");
f95910fe 1246 } else {
0cfae925
VSO		 1247 ret = nbd_negotiate_send_rep_err(client,
1248 NBD_REP_ERR_POLICY, errp,
36683283 1249 "TLS not configured");
63d5ef86 1250 }
d1129a8a 1251 break;
5c54e7fa
VSO		 1252
1253 case NBD_OPT_STRUCTURED_REPLY:
1254 if (length) {
0cfae925 1255 ret = nbd_reject_length(client, false, errp);
5c54e7fa
VSO		 1256 } else if (client->structured_reply) {
1257 ret = nbd_negotiate_send_rep_err(
0cfae925 1258 client, NBD_REP_ERR_INVALID, errp,
5c54e7fa
VSO		 1259 "structured reply already negotiated");
1260 } else {
0cfae925 1261 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
5c54e7fa 1262 client->structured_reply = true;
5c54e7fa
VSO		 1263 }
1264 break;
1265
e7b1948d
VSO		 1266 case NBD_OPT_LIST_META_CONTEXT:
1267 case NBD_OPT_SET_META_CONTEXT:
1268 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1269 errp);
1270 break;
1271
26afa868 1272 default:
894e0280 1273 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
28fb494f 1274 "Unsupported option %" PRIu32 " (%s)",
894e0280 1275 option, nbd_opt_lookup(option));
156f6a10 1276 break;
26afa868
DB		 1277 }
1278 } else {
1279 /*
1280 * If broken new-style we should drop the connection
1281 * for anything except NBD_OPT_EXPORT_NAME
1282 */
7f9039cd 1283 switch (option) {
26afa868 1284 case NBD_OPT_EXPORT_NAME:
dbb38caa 1285 return nbd_negotiate_handle_export_name(client, no_zeroes,
23e099c3 1286 errp);
26afa868
DB		 1287
1288 default:
28fb494f 1289 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
3736cc5b 1290 option, nbd_opt_lookup(option));
26afa868 1291 return -EINVAL;
32d7d2e0 1292 }
f5076b5a 1293 }
8cbee49e
EB		 1294 if (ret < 0) {
1295 return ret;
1296 }
f5076b5a
HB		 1297 }
1298}
1299
1e120ffe
VSO		 1300/* nbd_negotiate
1301 * Return:
2fd2c840
VSO		 1302 * -errno on error, errp is set
1303 * 0 on successful negotiation, errp is not set
1304 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1305 * errp is not set
1e120ffe 1306 */
2fd2c840 1307static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
7a5ca864 1308{
795d946d 1309 ERRP_GUARD();
5f66d060 1310 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
2e5c9ad6 1311 int ret;
b2e3d87f 1312
5f66d060 1313 /* Old style negotiation header, no room for options
6b8c01e7
PB		 1314 [ 0 .. 7] passwd ("NBDMAGIC")
1315 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
b2e3d87f 1316 [16 .. 23] size
5f66d060 1317 [24 .. 27] export flags (zero-extended)
6b8c01e7
PB		 1318 [28 .. 151] reserved (0)
1319
5f66d060 1320 New style negotiation header, client can send options
6b8c01e7
PB		 1321 [ 0 .. 7] passwd ("NBDMAGIC")
1322 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1323 [16 .. 17] server flags (0)
f37708f6 1324 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
b2e3d87f
NT		 1325 */
1326
1c778ef7 1327 qio_channel_set_blocking(client->ioc, false, NULL);
185b4338 1328
9588463e 1329 trace_nbd_negotiate_begin();
b2e3d87f 1330 memcpy(buf, "NBDMAGIC", 8);
f95910fe 1331
7f7dfe2a
VSO		 1332 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1333 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
b2e3d87f 1334
7f7dfe2a
VSO		 1335 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1336 error_prepend(errp, "write failed: ");
1337 return -EINVAL;
1338 }
dbb38caa 1339 ret = nbd_negotiate_options(client, errp);
7f7dfe2a
VSO		 1340 if (ret != 0) {
1341 if (ret < 0) {
1342 error_prepend(errp, "option negotiation failed: ");
6b8c01e7 1343 }
7f7dfe2a 1344 return ret;
b2e3d87f
NT		 1345 }
1346
b4961249 1347 /* Attach the channel to the same AioContext as the export */
8612c686
KW		 1348 if (client->exp && client->exp->common.ctx) {
1349 qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
b4961249
SL		 1350 }
1351
0cfae925 1352 assert(!client->optlen);
9588463e 1353 trace_nbd_negotiate_success();
d9faeed8
VSO		 1354
1355 return 0;
7a5ca864
FB		 1356}
1357
f148ae7d
SL		 1358/* nbd_read_eof
1359 * Tries to read @size bytes from @ioc. This is a local implementation of
1360 * qio_channel_readv_all_eof. We have it here because we need it to be
1361 * interruptible and to know when the coroutine is yielding.
1362 * Returns 1 on success
1363 * 0 on eof, when no data was read (errp is not set)
1364 * negative errno on failure (errp is set)
1365 */
1366static inline int coroutine_fn
1367nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1368{
1369 bool partial = false;
1370
1371 assert(size);
1372 while (size > 0) {
1373 struct iovec iov = { .iov_base = buffer, .iov_len = size };
1374 ssize_t len;
1375
1376 len = qio_channel_readv(client->ioc, &iov, 1, errp);
1377 if (len == QIO_CHANNEL_ERR_BLOCK) {
1378 client->read_yielding = true;
1379 qio_channel_yield(client->ioc, G_IO_IN);
1380 client->read_yielding = false;
1381 if (client->quiescing) {
1382 return -EAGAIN;
1383 }
1384 continue;
1385 } else if (len < 0) {
1386 return -EIO;
1387 } else if (len == 0) {
1388 if (partial) {
1389 error_setg(errp,
1390 "Unexpected end-of-file before all bytes were read");
1391 return -EIO;
1392 } else {
1393 return 0;
1394 }
1395 }
1396
1397 partial = true;
1398 size -= len;
1399 buffer = (uint8_t *) buffer + len;
1400 }
1401 return 1;
1402}
1403
1404static int nbd_receive_request(NBDClient *client, NBDRequest *request,
2fd2c840 1405 Error **errp)
75818250 1406{
fa26c26b 1407 uint8_t buf[NBD_REQUEST_SIZE];
b2e3d87f 1408 uint32_t magic;
a0dc63a6 1409 int ret;
b2e3d87f 1410
f148ae7d 1411 ret = nbd_read_eof(client, buf, sizeof(buf), errp);
185b4338
PB		 1412 if (ret < 0) {
1413 return ret;
1414 }
1415
b2e3d87f
NT		 1416 /* Request
1417 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
b626b51a
EB		 1418 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1419 [ 6 .. 7] type (NBD_CMD_READ, ...)
b2e3d87f
NT		 1420 [ 8 .. 15] handle
1421 [16 .. 23] from
1422 [24 .. 27] len
1423 */
1424
773dce3c 1425 magic = ldl_be_p(buf);
b626b51a
EB		 1426 request->flags = lduw_be_p(buf + 4);
1427 request->type = lduw_be_p(buf + 6);
773dce3c
PM		 1428 request->handle = ldq_be_p(buf + 8);
1429 request->from = ldq_be_p(buf + 16);
1430 request->len = ldl_be_p(buf + 24);
b2e3d87f 1431
9588463e
VSO		 1432 trace_nbd_receive_request(magic, request->flags, request->type,
1433 request->from, request->len);
b2e3d87f
NT		 1434
1435 if (magic != NBD_REQUEST_MAGIC) {
2fd2c840 1436 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
185b4338 1437 return -EINVAL;
b2e3d87f
NT		 1438 }
1439 return 0;
75818250
TS		 1440}
1441
41996e38
PB		 1442#define MAX_NBD_REQUESTS 16
1443
ce33967a 1444void nbd_client_get(NBDClient *client)
1743b515
PB		 1445{
1446 client->refcount++;
1447}
1448
ce33967a 1449void nbd_client_put(NBDClient *client)
1743b515
PB		 1450{
1451 if (--client->refcount == 0) {
ff2b68aa 1452 /* The last reference should be dropped by client->close,
f53a829b 1453 * which is called by client_close.
ff2b68aa
PB		 1454 */
1455 assert(client->closing);
1456
ff82911c 1457 qio_channel_detach_aio_context(client->ioc);
1c778ef7
DB		 1458 object_unref(OBJECT(client->sioc));
1459 object_unref(OBJECT(client->ioc));
f95910fe
DB		 1460 if (client->tlscreds) {
1461 object_unref(OBJECT(client->tlscreds));
1462 }
b25e12da 1463 g_free(client->tlsauthz);
6b8c01e7
PB		 1464 if (client->exp) {
1465 QTAILQ_REMOVE(&client->exp->clients, client, next);
c69de1be 1466 blk_exp_unref(&client->exp->common);
6b8c01e7 1467 }
3b1f244c 1468 g_free(client->export_meta.bitmaps);
1743b515
PB		 1469 g_free(client);
1470 }
1471}
1472
0c9390d9 1473static void client_close(NBDClient *client, bool negotiated)
1743b515 1474{
ff2b68aa
PB		 1475 if (client->closing) {
1476 return;
1477 }
1478
1479 client->closing = true;
1480
1481 /* Force requests to finish. They will drop their own references,
1482 * then we'll close the socket and free the NBDClient.
1483 */
1c778ef7
DB		 1484 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1485 NULL);
ff2b68aa
PB		 1486
1487 /* Also tell the client, so that they release their reference. */
0c9390d9
EB		 1488 if (client->close_fn) {
1489 client->close_fn(client, negotiated);
1743b515 1490 }
1743b515
PB		 1491}
1492
315f78ab 1493static NBDRequestData *nbd_request_get(NBDClient *client)
d9a73806 1494{
315f78ab 1495 NBDRequestData *req;
72deddc5 1496
41996e38
PB		 1497 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1498 client->nb_requests++;
1499
315f78ab 1500 req = g_new0(NBDRequestData, 1);
72deddc5
PB		 1501 nbd_client_get(client);
1502 req->client = client;
d9a73806
PB		 1503 return req;
1504}
1505
315f78ab 1506static void nbd_request_put(NBDRequestData *req)
d9a73806 1507{
72deddc5 1508 NBDClient *client = req->client;
e1adb27a 1509
2d821488
SH		 1510 if (req->data) {
1511 qemu_vfree(req->data);
1512 }
1729404c 1513 g_free(req);
e1adb27a 1514
958c717d 1515 client->nb_requests--;
fd6afc50
SL		 1516
1517 if (client->quiescing && client->nb_requests == 0) {
1518 aio_wait_kick();
1519 }
1520
ff82911c
PB		 1521 nbd_client_receive_next_request(client);
1522
72deddc5 1523 nbd_client_put(client);
d9a73806
PB		 1524}
1525
aadf99a7 1526static void blk_aio_attached(AioContext *ctx, void *opaque)
f2149281
HR		 1527{
1528 NBDExport *exp = opaque;
1529 NBDClient *client;
1530
9588463e 1531 trace_nbd_blk_aio_attached(exp->name, ctx);
f2149281 1532
8612c686 1533 exp->common.ctx = ctx;
f2149281
HR		 1534
1535 QTAILQ_FOREACH(client, &exp->clients, next) {
ff82911c 1536 qio_channel_attach_aio_context(client->ioc, ctx);
f148ae7d 1537
fd6afc50 1538 assert(client->nb_requests == 0);
f148ae7d
SL		 1539 assert(client->recv_coroutine == NULL);
1540 assert(client->send_coroutine == NULL);
f148ae7d
SL		 1541 }
1542}
1543
fd6afc50 1544static void blk_aio_detach(void *opaque)
f148ae7d
SL		 1545{
1546 NBDExport *exp = opaque;
1547 NBDClient *client;
1548
fd6afc50
SL		 1549 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1550
f148ae7d
SL		 1551 QTAILQ_FOREACH(client, &exp->clients, next) {
1552 qio_channel_detach_aio_context(client->ioc);
fd6afc50
SL		 1553 }
1554
1555 exp->common.ctx = NULL;
1556}
1557
1558static void nbd_drained_begin(void *opaque)
1559{
1560 NBDExport *exp = opaque;
1561 NBDClient *client;
1562
1563 QTAILQ_FOREACH(client, &exp->clients, next) {
f148ae7d 1564 client->quiescing = true;
fd6afc50
SL		 1565 }
1566}
f148ae7d 1567
fd6afc50
SL		 1568static void nbd_drained_end(void *opaque)
1569{
1570 NBDExport *exp = opaque;
1571 NBDClient *client;
f148ae7d 1572
fd6afc50
SL		 1573 QTAILQ_FOREACH(client, &exp->clients, next) {
1574 client->quiescing = false;
1575 nbd_client_receive_next_request(client);
f2149281
HR		 1576 }
1577}
1578
fd6afc50 1579static bool nbd_drained_poll(void *opaque)
f2149281
HR		 1580{
1581 NBDExport *exp = opaque;
fd6afc50 1582 NBDClient *client;
f2149281 1583
fd6afc50
SL		 1584 QTAILQ_FOREACH(client, &exp->clients, next) {
1585 if (client->nb_requests != 0) {
1586 /*
1587 * If there's a coroutine waiting for a request on nbd_read_eof()
1588 * enter it here so we don't depend on the client to wake it up.
1589 */
1590 if (client->recv_coroutine != NULL && client->read_yielding) {
1591 qemu_aio_coroutine_enter(exp->common.ctx,
1592 client->recv_coroutine);
1593 }
f2149281 1594
fd6afc50
SL		 1595 return true;
1596 }
1597 }
f2149281 1598
fd6afc50 1599 return false;
f2149281
HR		 1600}
1601
741cc431
HR		 1602static void nbd_eject_notifier(Notifier *n, void *data)
1603{
1604 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
61bc846d 1605
bc4ee65b 1606 blk_exp_request_shutdown(&exp->common);
741cc431
HR		 1607}
1608
9b562c64
KW		 1609void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1610{
1611 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1612 assert(exp->drv == &blk_exp_nbd);
1613 assert(nbd_exp->eject_notifier_blk == NULL);
1614
1615 blk_ref(blk);
1616 nbd_exp->eject_notifier_blk = blk;
1617 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1618 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1619}
1620
fd6afc50
SL		 1621static const BlockDevOps nbd_block_ops = {
1622 .drained_begin = nbd_drained_begin,
1623 .drained_end = nbd_drained_end,
1624 .drained_poll = nbd_drained_poll,
1625};
1626
5b1cb497
KW		 1627static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1628 Error **errp)
af49bbbe 1629{
a6ff7989 1630 NBDExport *exp = container_of(blk_exp, NBDExport, common);
5b1cb497 1631 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
331170e0 1632 BlockBackend *blk = blk_exp->blk;
b57e4de0 1633 int64_t size;
331170e0 1634 uint64_t perm, shared_perm;
5b1cb497
KW		 1635 bool readonly = !exp_args->writable;
1636 bool shared = !exp_args->writable;
cbad81ce 1637 strList *bitmaps;
3b1f244c 1638 size_t i;
d7086422 1639 int ret;
cd7fca95 1640
5b1cb497
KW		 1641 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1642
1643 if (!nbd_server_is_running()) {
1644 error_setg(errp, "NBD server not running");
1645 return -EINVAL;
1646 }
1647
1648 if (!arg->has_name) {
1649 arg->name = exp_args->node_name;
1650 }
1651
1652 if (strlen(arg->name) > NBD_MAX_STRING_SIZE) {
1653 error_setg(errp, "export name '%s' too long", arg->name);
1654 return -EINVAL;
1655 }
1656
1657 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1658 error_setg(errp, "description '%s' too long", arg->description);
1659 return -EINVAL;
1660 }
1661
1662 if (nbd_export_find(arg->name)) {
1663 error_setg(errp, "NBD server already has export named '%s'", arg->name);
1664 return -EEXIST;
1665 }
1666
331170e0 1667 size = blk_getlength(blk);
b57e4de0
KW		 1668 if (size < 0) {
1669 error_setg_errno(errp, -size,
1670 "Failed to determine the NBD export's length");
a6ff7989 1671 return size;
b57e4de0
KW		 1672 }
1673
8a7ce4f9
KW		 1674 /* Don't allow resize while the NBD server is running, otherwise we don't
1675 * care what happens with the node. */
331170e0 1676 blk_get_perm(blk, &perm, &shared_perm);
331170e0 1677 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
d7086422 1678 if (ret < 0) {
331170e0 1679 return ret;
d7086422 1680 }
331170e0 1681
4b9441f6 1682 QTAILQ_INIT(&exp->clients);
5b1cb497
KW		 1683 exp->name = g_strdup(arg->name);
1684 exp->description = g_strdup(arg->description);
dbb38caa
EB		 1685 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1686 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1687 if (readonly) {
1688 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1689 if (shared) {
1690 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1691 }
1692 } else {
b491dbb7
EB		 1693 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1694 NBD_FLAG_SEND_FAST_ZERO);
dbb38caa 1695 }
7596bbb3 1696 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
98f44bbe 1697
cbad81ce 1698 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
3b1f244c
EB		 1699 exp->nr_export_bitmaps++;
1700 }
1701 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1702 for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1703 i++, bitmaps = bitmaps->next) {
cbad81ce 1704 const char *bitmap = bitmaps->value;
331170e0 1705 BlockDriverState *bs = blk_bs(blk);
678ba275 1706 BdrvDirtyBitmap *bm = NULL;
678ba275 1707
ee2f94ca 1708 while (bs) {
cbad81ce 1709 bm = bdrv_find_dirty_bitmap(bs, bitmap);
ee2f94ca 1710 if (bm != NULL) {
678ba275
EB		 1711 break;
1712 }
1713
ee2f94ca 1714 bs = bdrv_filter_or_cow_bs(bs);
678ba275
EB		 1715 }
1716
1717 if (bm == NULL) {
a6ff7989 1718 ret = -ENOENT;
cbad81ce 1719 error_setg(errp, "Bitmap '%s' is not found", bitmap);
678ba275
EB		 1720 goto fail;
1721 }
1722
3ae96d66 1723 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
a6ff7989 1724 ret = -EINVAL;
3b78a927
JS		 1725 goto fail;
1726 }
1727
dbb38caa 1728 if (readonly && bdrv_is_writable(bs) &&
678ba275 1729 bdrv_dirty_bitmap_enabled(bm)) {
a6ff7989 1730 ret = -EINVAL;
678ba275
EB		 1731 error_setg(errp,
1732 "Enabled bitmap '%s' incompatible with readonly export",
cbad81ce 1733 bitmap);
678ba275
EB		 1734 goto fail;
1735 }
1736
3b1f244c 1737 exp->export_bitmaps[i] = bm;
cbad81ce 1738 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
678ba275
EB		 1739 }
1740
3b1f244c
EB		 1741 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1742 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1743 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1744 }
1745
dbc7b014
EB		 1746 exp->allocation_depth = arg->allocation_depth;
1747
fd6afc50
SL		 1748 /*
1749 * We need to inhibit request queuing in the block layer to ensure we can
1750 * be properly quiesced when entering a drained section, as our coroutines
1751 * servicing pending requests might enter blk_pread().
1752 */
1753 blk_set_disable_request_queuing(blk, true);
1754
aadf99a7 1755 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
741cc431 1756
fd6afc50
SL		 1757 blk_set_dev_ops(blk, &nbd_block_ops, exp);
1758
3fa4c765 1759 QTAILQ_INSERT_TAIL(&exports, exp, next);
c69de1be 1760
a6ff7989 1761 return 0;
98f44bbe
HR		 1762
1763fail:
3b1f244c 1764 g_free(exp->export_bitmaps);
3fa4c765
EB		 1765 g_free(exp->name);
1766 g_free(exp->description);
a6ff7989 1767 return ret;
af49bbbe
PB		 1768}
1769
ee0a19ec
PB		 1770NBDExport *nbd_export_find(const char *name)
1771{
1772 NBDExport *exp;
1773 QTAILQ_FOREACH(exp, &exports, next) {
1774 if (strcmp(name, exp->name) == 0) {
1775 return exp;
1776 }
1777 }
1778
1779 return NULL;
1780}
1781
61bc846d
EB		 1782AioContext *
1783nbd_export_aio_context(NBDExport *exp)
1784{
8612c686 1785 return exp->common.ctx;
61bc846d
EB		 1786}
1787
bc4ee65b 1788static void nbd_export_request_shutdown(BlockExport *blk_exp)
af49bbbe 1789{
bc4ee65b 1790 NBDExport *exp = container_of(blk_exp, NBDExport, common);
4b9441f6 1791 NBDClient *client, *next;
2c8d9f06 1792
c69de1be 1793 blk_exp_ref(&exp->common);
3fa4c765
EB		 1794 /*
1795 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1796 * close mode that stops advertising the export to new clients but
1797 * still permits existing clients to run to completion? Because of
1798 * that possibility, nbd_export_close() can be called more than
1799 * once on an export.
1800 */
4b9441f6 1801 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
0c9390d9 1802 client_close(client, true);
4b9441f6 1803 }
3fa4c765 1804 if (exp->name) {
3fa4c765
EB		 1805 g_free(exp->name);
1806 exp->name = NULL;
1807 QTAILQ_REMOVE(&exports, exp, next);
1808 }
c69de1be 1809 blk_exp_unref(&exp->common);
2c8d9f06
PB		 1810}
1811
c69de1be 1812static void nbd_export_delete(BlockExport *blk_exp)
2c8d9f06 1813{
3b1f244c 1814 size_t i;
c69de1be 1815 NBDExport *exp = container_of(blk_exp, NBDExport, common);
2c8d9f06 1816
c69de1be
KW		 1817 assert(exp->name == NULL);
1818 assert(QTAILQ_EMPTY(&exp->clients));
d6268348 1819
c69de1be
KW		 1820 g_free(exp->description);
1821 exp->description = NULL;
1822
37a4f70c 1823 if (exp->common.blk) {
c69de1be
KW		 1824 if (exp->eject_notifier_blk) {
1825 notifier_remove(&exp->eject_notifier);
1826 blk_unref(exp->eject_notifier_blk);
3d068aff 1827 }
37a4f70c 1828 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
c69de1be 1829 blk_aio_detach, exp);
fd6afc50 1830 blk_set_disable_request_queuing(exp->common.blk, false);
c69de1be 1831 }
3d068aff 1832
3b1f244c
EB		 1833 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1834 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
2c8d9f06 1835 }
af49bbbe
PB		 1836}
1837
56ee8626
KW		 1838const BlockExportDriver blk_exp_nbd = {
1839 .type = BLOCK_EXPORT_TYPE_NBD,
a6ff7989 1840 .instance_size = sizeof(NBDExport),
56ee8626 1841 .create = nbd_export_create,
c69de1be 1842 .delete = nbd_export_delete,
bc4ee65b 1843 .request_shutdown = nbd_export_request_shutdown,
56ee8626
KW		 1844};
1845
de79bfc3
VSO		 1846static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1847 unsigned niov, Error **errp)
1848{
1849 int ret;
1850
1851 g_assert(qemu_in_coroutine());
1852 qemu_co_mutex_lock(&client->send_lock);
1853 client->send_coroutine = qemu_coroutine_self();
1854
1855 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1856
1857 client->send_coroutine = NULL;
1858 qemu_co_mutex_unlock(&client->send_lock);
1859
1860 return ret;
1861}
1862
caad5384
VSO		 1863static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1864 uint64_t handle)
1865{
1866 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1867 stl_be_p(&reply->error, error);
1868 stq_be_p(&reply->handle, handle);
1869}
1870
978df1b6 1871static int nbd_co_send_simple_reply(NBDClient *client,
14cea41d
VSO		 1872 uint64_t handle,
1873 uint32_t error,
978df1b6
VSO		 1874 void *data,
1875 size_t len,
1876 Error **errp)
22045592 1877{
de79bfc3 1878 NBDSimpleReply reply;
14cea41d 1879 int nbd_err = system_errno_to_nbd_errno(error);
de79bfc3
VSO		 1880 struct iovec iov[] = {
1881 {.iov_base = &reply, .iov_len = sizeof(reply)},
1882 {.iov_base = data, .iov_len = len}
1883 };
6fb2b972 1884
e7a78d0e
EB		 1885 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1886 len);
de79bfc3 1887 set_be_simple_reply(&reply, nbd_err, handle);
262db388 1888
de79bfc3 1889 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
22045592
PB		 1890}
1891
5c54e7fa
VSO		 1892static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1893 uint16_t type, uint64_t handle, uint32_t length)
1894{
1895 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1896 stw_be_p(&chunk->flags, flags);
1897 stw_be_p(&chunk->type, type);
1898 stq_be_p(&chunk->handle, handle);
1899 stl_be_p(&chunk->length, length);
1900}
1901
ef8c887e
EB		 1902static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1903 uint64_t handle,
1904 Error **errp)
1905{
1906 NBDStructuredReplyChunk chunk;
1907 struct iovec iov[] = {
1908 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1909 };
1910
1911 trace_nbd_co_send_structured_done(handle);
1912 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1913
1914 return nbd_co_send_iov(client, iov, 1, errp);
1915}
1916
5c54e7fa
VSO		 1917static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1918 uint64_t handle,
1919 uint64_t offset,
1920 void *data,
1921 size_t size,
418638d3 1922 bool final,
5c54e7fa
VSO		 1923 Error **errp)
1924{
efdc0c10 1925 NBDStructuredReadData chunk;
5c54e7fa
VSO		 1926 struct iovec iov[] = {
1927 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1928 {.iov_base = data, .iov_len = size}
1929 };
1930
ef8c887e 1931 assert(size);
5c54e7fa 1932 trace_nbd_co_send_structured_read(handle, offset, data, size);
418638d3
EB		 1933 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1934 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1935 sizeof(chunk) - sizeof(chunk.h) + size);
5c54e7fa
VSO		 1936 stq_be_p(&chunk.offset, offset);
1937
1938 return nbd_co_send_iov(client, iov, 2, errp);
1939}
1940
60ace2ba
VSO		 1941static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1942 uint64_t handle,
1943 uint32_t error,
1944 const char *msg,
1945 Error **errp)
1946{
1947 NBDStructuredError chunk;
1948 int nbd_err = system_errno_to_nbd_errno(error);
1949 struct iovec iov[] = {
1950 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1951 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1952 };
1953
1954 assert(nbd_err);
1955 trace_nbd_co_send_structured_error(handle, nbd_err,
1956 nbd_err_lookup(nbd_err), msg ? msg : "");
1957 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1958 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1959 stl_be_p(&chunk.error, nbd_err);
1960 stw_be_p(&chunk.message_length, iov[1].iov_len);
1961
1962 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1963}
1964
37e02aeb
VSO		 1965/* Do a sparse read and send the structured reply to the client.
1966 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1967 * reported to the client, at which point this function succeeds.
1968 */
418638d3
EB		 1969static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1970 uint64_t handle,
1971 uint64_t offset,
1972 uint8_t *data,
1973 size_t size,
1974 Error **errp)
1975{
1976 int ret = 0;
1977 NBDExport *exp = client->exp;
1978 size_t progress = 0;
1979
1980 while (progress < size) {
1981 int64_t pnum;
37a4f70c 1982 int status = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
418638d3
EB		 1983 offset + progress,
1984 size - progress, &pnum, NULL,
1985 NULL);
e2de3256 1986 bool final;
418638d3
EB		 1987
1988 if (status < 0) {
37e02aeb
VSO		 1989 char *msg = g_strdup_printf("unable to check for holes: %s",
1990 strerror(-status));
1991
1992 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1993 errp);
1994 g_free(msg);
1995 return ret;
418638d3
EB		 1996 }
1997 assert(pnum && pnum <= size - progress);
e2de3256 1998 final = progress + pnum == size;
418638d3
EB		 1999 if (status & BDRV_BLOCK_ZERO) {
2000 NBDStructuredReadHole chunk;
2001 struct iovec iov[] = {
2002 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2003 };
2004
2005 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
2006 pnum);
e2de3256
EB		 2007 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
2008 NBD_REPLY_TYPE_OFFSET_HOLE,
418638d3
EB		 2009 handle, sizeof(chunk) - sizeof(chunk.h));
2010 stq_be_p(&chunk.offset, offset + progress);
2011 stl_be_p(&chunk.length, pnum);
2012 ret = nbd_co_send_iov(client, iov, 1, errp);
2013 } else {
37a4f70c
KW		 2014 ret = blk_pread(exp->common.blk, offset + progress,
2015 data + progress, pnum);
418638d3
EB		 2016 if (ret < 0) {
2017 error_setg_errno(errp, -ret, "reading from file failed");
2018 break;
2019 }
2020 ret = nbd_co_send_structured_read(client, handle, offset + progress,
e2de3256 2021 data + progress, pnum, final,
418638d3
EB		 2022 errp);
2023 }
2024
2025 if (ret < 0) {
2026 break;
2027 }
2028 progress += pnum;
2029 }
418638d3
EB		 2030 return ret;
2031}
2032
89cbc7e3
VSO		 2033typedef struct NBDExtentArray {
2034 NBDExtent *extents;
2035 unsigned int nb_alloc;
2036 unsigned int count;
2037 uint64_t total_length;
2038 bool can_add;
2039 bool converted_to_be;
2040} NBDExtentArray;
2041
2042static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
2043{
2044 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2045
2046 ea->nb_alloc = nb_alloc;
2047 ea->extents = g_new(NBDExtent, nb_alloc);
2048 ea->can_add = true;
2049
2050 return ea;
2051}
2052
2053static void nbd_extent_array_free(NBDExtentArray *ea)
2054{
2055 g_free(ea->extents);
2056 g_free(ea);
2057}
2058G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free);
2059
2060/* Further modifications of the array after conversion are abandoned */
2061static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2062{
2063 int i;
2064
2065 assert(!ea->converted_to_be);
2066 ea->can_add = false;
2067 ea->converted_to_be = true;
2068
2069 for (i = 0; i < ea->count; i++) {
2070 ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2071 ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
2072 }
2073}
2074
fb7afc79 2075/*
89cbc7e3
VSO		 2076 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2077 * return -1.
2078 * For safety, when returning -1 for the first time, .can_add is set to false,
2079 * further call to nbd_extent_array_add() will crash.
2080 * (to avoid the situation, when after failing to add an extent (returned -1),
2081 * user miss this failure and add another extent, which is successfully added
2082 * (array is full, but new extent may be squashed into the last one), then we
2083 * have invalid array with skipped extent)
fb7afc79 2084 */
89cbc7e3
VSO		 2085static int nbd_extent_array_add(NBDExtentArray *ea,
2086 uint32_t length, uint32_t flags)
e7b1948d 2087{
89cbc7e3
VSO		 2088 assert(ea->can_add);
2089
2090 if (!length) {
2091 return 0;
2092 }
2093
2094 /* Extend previous extent if flags are the same */
2095 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2096 uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
2097
2098 if (sum <= UINT32_MAX) {
2099 ea->extents[ea->count - 1].length = sum;
2100 ea->total_length += length;
2101 return 0;
2102 }
2103 }
2104
2105 if (ea->count >= ea->nb_alloc) {
2106 ea->can_add = false;
2107 return -1;
2108 }
2109
2110 ea->total_length += length;
2111 ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
2112 ea->count++;
e7b1948d 2113
89cbc7e3
VSO		 2114 return 0;
2115}
2116
2117static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
2118 uint64_t bytes, NBDExtentArray *ea)
2119{
2120 while (bytes) {
e7b1948d
VSO		 2121 uint32_t flags;
2122 int64_t num;
89cbc7e3
VSO		 2123 int ret = bdrv_block_status_above(bs, NULL, offset, bytes, &num,
2124 NULL, NULL);
fb7afc79 2125
e7b1948d
VSO		 2126 if (ret < 0) {
2127 return ret;
2128 }
2129
0da98568
NS		 2130 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2131 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
e7b1948d 2132
89cbc7e3
VSO		 2133 if (nbd_extent_array_add(ea, num, flags) < 0) {
2134 return 0;
e7b1948d 2135 }
fb7afc79 2136
89cbc7e3
VSO		 2137 offset += num;
2138 bytes -= num;
e7b1948d
VSO		 2139 }
2140
e7b1948d
VSO		 2141 return 0;
2142}
2143
71719cd5
EB		 2144static int blockalloc_to_extents(BlockDriverState *bs, uint64_t offset,
2145 uint64_t bytes, NBDExtentArray *ea)
2146{
2147 while (bytes) {
2148 int64_t num;
2149 int ret = bdrv_is_allocated_above(bs, NULL, false, offset, bytes,
2150 &num);
2151
2152 if (ret < 0) {
2153 return ret;
2154 }
2155
2156 if (nbd_extent_array_add(ea, num, ret) < 0) {
2157 return 0;
2158 }
2159
2160 offset += num;
2161 bytes -= num;
2162 }
2163
2164 return 0;
2165}
2166
89cbc7e3
VSO		 2167/*
2168 * nbd_co_send_extents
3d068aff 2169 *
89cbc7e3
VSO		 2170 * @ea is converted to BE by the function
2171 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
3d068aff 2172 */
e7b1948d 2173static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
89cbc7e3
VSO		 2174 NBDExtentArray *ea,
2175 bool last, uint32_t context_id, Error **errp)
e7b1948d
VSO		 2176{
2177 NBDStructuredMeta chunk;
e7b1948d
VSO		 2178 struct iovec iov[] = {
2179 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
89cbc7e3 2180 {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
e7b1948d
VSO		 2181 };
2182
89cbc7e3
VSO		 2183 nbd_extent_array_convert_to_be(ea);
2184
2185 trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
2186 last);
3d068aff
VSO		 2187 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
2188 NBD_REPLY_TYPE_BLOCK_STATUS,
e7b1948d
VSO		 2189 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
2190 stl_be_p(&chunk.context_id, context_id);
2191
2192 return nbd_co_send_iov(client, iov, 2, errp);
2193}
2194
2195/* Get block status from the exported device and send it to the client */
2196static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2197 BlockDriverState *bs, uint64_t offset,
fb7afc79
VSO		 2198 uint32_t length, bool dont_fragment,
2199 bool last, uint32_t context_id,
2200 Error **errp)
e7b1948d
VSO		 2201{
2202 int ret;
416e34bd 2203 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
89cbc7e3 2204 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
e7b1948d 2205
71719cd5
EB		 2206 if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2207 ret = blockstatus_to_extents(bs, offset, length, ea);
2208 } else {
2209 ret = blockalloc_to_extents(bs, offset, length, ea);
2210 }
e7b1948d
VSO		 2211 if (ret < 0) {
2212 return nbd_co_send_structured_error(
2213 client, handle, -ret, "can't get block status", errp);
2214 }
2215
89cbc7e3 2216 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
3d068aff
VSO		 2217}
2218
dacbb6eb 2219/* Populate @ea from a dirty bitmap. */
89cbc7e3
VSO		 2220static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2221 uint64_t offset, uint64_t length,
dacbb6eb 2222 NBDExtentArray *es)
3d068aff 2223{
dacbb6eb
VSO		 2224 int64_t start, dirty_start, dirty_count;
2225 int64_t end = offset + length;
2226 bool full = false;
3d068aff
VSO		 2227
2228 bdrv_dirty_bitmap_lock(bitmap);
2229
dacbb6eb
VSO		 2230 for (start = offset;
2231 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2232 &dirty_start, &dirty_count);
2233 start = dirty_start + dirty_count)
2234 {
2235 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2236 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2237 {
2238 full = true;
89cbc7e3
VSO		 2239 break;
2240 }
3d068aff
VSO		 2241 }
2242
dacbb6eb 2243 if (!full) {
c0b21f2e
EB		 2244 /* last non dirty extent, nothing to do if array is now full */
2245 (void) nbd_extent_array_add(es, end - start, 0);
dacbb6eb 2246 }
3d068aff
VSO		 2247
2248 bdrv_dirty_bitmap_unlock(bitmap);
3d068aff
VSO		 2249}
2250
2251static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2252 BdrvDirtyBitmap *bitmap, uint64_t offset,
2253 uint32_t length, bool dont_fragment, bool last,
2254 uint32_t context_id, Error **errp)
2255{
416e34bd 2256 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
89cbc7e3 2257 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
3d068aff 2258
dacbb6eb 2259 bitmap_to_extents(bitmap, offset, length, ea);
3d068aff 2260
89cbc7e3 2261 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
e7b1948d
VSO		 2262}
2263
2a6e128b
VSO		 2264/* nbd_co_receive_request
2265 * Collect a client request. Return 0 if request looks valid, -EIO to drop
f148ae7d
SL		 2266 * connection right away, -EAGAIN to indicate we were interrupted and the
2267 * channel should be quiesced, and any other negative value to report an error
2268 * to the client (although the caller may still need to disconnect after
2269 * reporting the error).
2a6e128b 2270 */
2fd2c840
VSO		 2271static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2272 Error **errp)
a030b347 2273{
72deddc5 2274 NBDClient *client = req->client;
5c54e7fa 2275 int valid_flags;
f148ae7d 2276 int ret;
a030b347 2277
1c778ef7 2278 g_assert(qemu_in_coroutine());
ff82911c 2279 assert(client->recv_coroutine == qemu_coroutine_self());
f148ae7d
SL		 2280 ret = nbd_receive_request(client, request, errp);
2281 if (ret < 0) {
2282 return ret;
a030b347
PB		 2283 }
2284
3736cc5b
EB		 2285 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2286 nbd_cmd_lookup(request->type));
29b6c3b3 2287
b626b51a 2288 if (request->type != NBD_CMD_WRITE) {
29b6c3b3
EB		 2289 /* No payload, we are ready to read the next request. */
2290 req->complete = true;
2291 }
2292
b626b51a 2293 if (request->type == NBD_CMD_DISC) {
29b6c3b3
EB		 2294 /* Special case: we're going to disconnect without a reply,
2295 * whether or not flags, from, or len are bogus */
ee898b87 2296 return -EIO;
29b6c3b3
EB		 2297 }
2298
bc37b06a
VSO		 2299 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2300 request->type == NBD_CMD_CACHE)
2301 {
eb38c3b6 2302 if (request->len > NBD_MAX_BUFFER_SIZE) {
2fd2c840
VSO		 2303 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2304 request->len, NBD_MAX_BUFFER_SIZE);
ee898b87 2305 return -EINVAL;
eb38c3b6
PB		 2306 }
2307
7fa5c565 2308 if (request->type != NBD_CMD_CACHE) {
37a4f70c
KW		 2309 req->data = blk_try_blockalign(client->exp->common.blk,
2310 request->len);
7fa5c565
VSO		 2311 if (req->data == NULL) {
2312 error_setg(errp, "No memory");
2313 return -ENOMEM;
2314 }
f1c17521 2315 }
2d821488 2316 }
7fa5c565 2317
b626b51a 2318 if (request->type == NBD_CMD_WRITE) {
e6798f06
VSO		 2319 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2320 errp) < 0)
2321 {
ee898b87 2322 return -EIO;
a030b347 2323 }
29b6c3b3 2324 req->complete = true;
6fb2b972 2325
9588463e
VSO		 2326 trace_nbd_co_receive_request_payload_received(request->handle,
2327 request->len);
a030b347 2328 }
29b6c3b3 2329
fed5f8f8
EB		 2330 /* Sanity checks. */
2331 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2332 (request->type == NBD_CMD_WRITE ||
2333 request->type == NBD_CMD_WRITE_ZEROES ||
2334 request->type == NBD_CMD_TRIM)) {
2335 error_setg(errp, "Export is read-only");
2336 return -EROFS;
2337 }
2338 if (request->from > client->exp->size ||
9d26dfcb 2339 request->len > client->exp->size - request->from) {
2fd2c840
VSO		 2340 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2341 ", Size: %" PRIu64, request->from, request->len,
9d26dfcb 2342 client->exp->size);
fed5f8f8
EB		 2343 return (request->type == NBD_CMD_WRITE ||
2344 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
29b6c3b3 2345 }
6e280648
EB		 2346 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2347 client->check_align)) {
2348 /*
2349 * The block layer gracefully handles unaligned requests, but
2350 * it's still worth tracing client non-compliance
2351 */
2352 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2353 request->from,
2354 request->len,
2355 client->check_align);
2356 }
5c54e7fa
VSO		 2357 valid_flags = NBD_CMD_FLAG_FUA;
2358 if (request->type == NBD_CMD_READ && client->structured_reply) {
2359 valid_flags |= NBD_CMD_FLAG_DF;
2360 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
b491dbb7 2361 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
e7b1948d
VSO		 2362 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2363 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
ab7c548e 2364 }
5c54e7fa
VSO		 2365 if (request->flags & ~valid_flags) {
2366 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2367 nbd_cmd_lookup(request->type), request->flags);
ee898b87 2368 return -EINVAL;
1f4d6d18 2369 }
29b6c3b3 2370
ee898b87 2371 return 0;
a030b347
PB		 2372}
2373
6a417599
VSO		 2374/* Send simple reply without a payload, or a structured error
2375 * @error_msg is ignored if @ret >= 0
2376 * Returns 0 if connection is still live, -errno on failure to talk to client
2377 */
2378static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2379 uint64_t handle,
2380 int ret,
2381 const char *error_msg,
2382 Error **errp)
2383{
2384 if (client->structured_reply && ret < 0) {
2385 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2386 errp);
2387 } else {
2388 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2389 NULL, 0, errp);
2390 }
2391}
2392
2393/* Handle NBD_CMD_READ request.
2394 * Return -errno if sending fails. Other errors are reported directly to the
2395 * client as an error reply. */
2396static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2397 uint8_t *data, Error **errp)
2398{
2399 int ret;
2400 NBDExport *exp = client->exp;
2401
7fa5c565 2402 assert(request->type == NBD_CMD_READ);
6a417599
VSO		 2403
2404 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2405 if (request->flags & NBD_CMD_FLAG_FUA) {
37a4f70c 2406 ret = blk_co_flush(exp->common.blk);
6a417599
VSO		 2407 if (ret < 0) {
2408 return nbd_send_generic_reply(client, request->handle, ret,
2409 "flush failed", errp);
2410 }
2411 }
2412
2413 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
7fa5c565 2414 request->len)
2f454def 2415 {
6a417599
VSO		 2416 return nbd_co_send_sparse_read(client, request->handle, request->from,
2417 data, request->len, errp);
2418 }
2419
37a4f70c 2420 ret = blk_pread(exp->common.blk, request->from, data, request->len);
7fa5c565 2421 if (ret < 0) {
6a417599
VSO		 2422 return nbd_send_generic_reply(client, request->handle, ret,
2423 "reading from file failed", errp);
2424 }
2425
2426 if (client->structured_reply) {
2427 if (request->len) {
2428 return nbd_co_send_structured_read(client, request->handle,
2429 request->from, data,
2430 request->len, true, errp);
2431 } else {
2432 return nbd_co_send_structured_done(client, request->handle, errp);
2433 }
2434 } else {
2435 return nbd_co_send_simple_reply(client, request->handle, 0,
2436 data, request->len, errp);
2437 }
2438}
2439
7fa5c565
VSO		 2440/*
2441 * nbd_do_cmd_cache
2442 *
2443 * Handle NBD_CMD_CACHE request.
2444 * Return -errno if sending fails. Other errors are reported directly to the
2445 * client as an error reply.
2446 */
2447static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2448 Error **errp)
2449{
2450 int ret;
2451 NBDExport *exp = client->exp;
2452
2453 assert(request->type == NBD_CMD_CACHE);
2454
37a4f70c 2455 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
7fa5c565
VSO		 2456 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2457
2458 return nbd_send_generic_reply(client, request->handle, ret,
2459 "caching data failed", errp);
2460}
2461
6f302e60
VSO		 2462/* Handle NBD request.
2463 * Return -errno if sending fails. Other errors are reported directly to the
2464 * client as an error reply. */
2465static coroutine_fn int nbd_handle_request(NBDClient *client,
2466 NBDRequest *request,
2467 uint8_t *data, Error **errp)
2468{
2469 int ret;
2470 int flags;
2471 NBDExport *exp = client->exp;
2472 char *msg;
3b1f244c 2473 size_t i;
6f302e60
VSO		 2474
2475 switch (request->type) {
bc37b06a 2476 case NBD_CMD_CACHE:
7fa5c565
VSO		 2477 return nbd_do_cmd_cache(client, request, errp);
2478
2479 case NBD_CMD_READ:
6f302e60
VSO		 2480 return nbd_do_cmd_read(client, request, data, errp);
2481
2482 case NBD_CMD_WRITE:
2483 flags = 0;
2484 if (request->flags & NBD_CMD_FLAG_FUA) {
2485 flags |= BDRV_REQ_FUA;
2486 }
37a4f70c
KW		 2487 ret = blk_pwrite(exp->common.blk, request->from, data, request->len,
2488 flags);
6f302e60
VSO		 2489 return nbd_send_generic_reply(client, request->handle, ret,
2490 "writing to file failed", errp);
2491
2492 case NBD_CMD_WRITE_ZEROES:
2493 flags = 0;
2494 if (request->flags & NBD_CMD_FLAG_FUA) {
2495 flags |= BDRV_REQ_FUA;
2496 }
2497 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2498 flags |= BDRV_REQ_MAY_UNMAP;
2499 }
b491dbb7
EB		 2500 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2501 flags |= BDRV_REQ_NO_FALLBACK;
2502 }
890cbccb
EB		 2503 ret = 0;
2504 /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2505 while (ret >= 0 && request->len) {
2506 int align = client->check_align ?: 1;
2507 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2508 align));
37a4f70c 2509 ret = blk_pwrite_zeroes(exp->common.blk, request->from, len, flags);
890cbccb
EB		 2510 request->len -= len;
2511 request->from += len;
2512 }
6f302e60
VSO		 2513 return nbd_send_generic_reply(client, request->handle, ret,
2514 "writing to file failed", errp);
2515
2516 case NBD_CMD_DISC:
2517 /* unreachable, thanks to special case in nbd_co_receive_request() */
2518 abort();
2519
2520 case NBD_CMD_FLUSH:
37a4f70c 2521 ret = blk_co_flush(exp->common.blk);
6f302e60
VSO		 2522 return nbd_send_generic_reply(client, request->handle, ret,
2523 "flush failed", errp);
2524
2525 case NBD_CMD_TRIM:
890cbccb
EB		 2526 ret = 0;
2527 /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2528 while (ret >= 0 && request->len) {
2529 int align = client->check_align ?: 1;
2530 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2531 align));
37a4f70c 2532 ret = blk_co_pdiscard(exp->common.blk, request->from, len);
890cbccb
EB		 2533 request->len -= len;
2534 request->from += len;
2535 }
2536 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
37a4f70c 2537 ret = blk_co_flush(exp->common.blk);
65529782 2538 }
6f302e60
VSO		 2539 return nbd_send_generic_reply(client, request->handle, ret,
2540 "discard failed", errp);
2541
e7b1948d 2542 case NBD_CMD_BLOCK_STATUS:
d8b20291
EB		 2543 if (!request->len) {
2544 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2545 "need non-zero length", errp);
2546 }
47ec485e 2547 if (client->export_meta.count) {
fb7afc79 2548 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
47ec485e 2549 int contexts_remaining = client->export_meta.count;
fb7afc79 2550
3d068aff
VSO		 2551 if (client->export_meta.base_allocation) {
2552 ret = nbd_co_send_block_status(client, request->handle,
37a4f70c
KW		 2553 blk_bs(exp->common.blk),
2554 request->from,
fb7afc79 2555 request->len, dont_fragment,
47ec485e 2556 !--contexts_remaining,
3d068aff
VSO		 2557 NBD_META_ID_BASE_ALLOCATION,
2558 errp);
73e064cc
EB		 2559 if (ret < 0) {
2560 return ret;
2561 }
2562 }
2563
71719cd5
EB		 2564 if (client->export_meta.allocation_depth) {
2565 ret = nbd_co_send_block_status(client, request->handle,
2566 blk_bs(exp->common.blk),
2567 request->from, request->len,
2568 dont_fragment,
2569 !--contexts_remaining,
2570 NBD_META_ID_ALLOCATION_DEPTH,
2571 errp);
2572 if (ret < 0) {
2573 return ret;
2574 }
2575 }
2576
3b1f244c
EB		 2577 for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2578 if (!client->export_meta.bitmaps[i]) {
2579 continue;
2580 }
3d068aff 2581 ret = nbd_co_send_bitmap(client, request->handle,
3b1f244c 2582 client->exp->export_bitmaps[i],
3d068aff 2583 request->from, request->len,
47ec485e 2584 dont_fragment, !--contexts_remaining,
3b1f244c 2585 NBD_META_ID_DIRTY_BITMAP + i, errp);
73e064cc
EB		 2586 if (ret < 0) {
2587 return ret;
2588 }
3d068aff
VSO		 2589 }
2590
47ec485e
EB		 2591 assert(!contexts_remaining);
2592
73e064cc 2593 return 0;
e7b1948d
VSO		 2594 } else {
2595 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2596 "CMD_BLOCK_STATUS not negotiated",
2597 errp);
2598 }
2599
6f302e60
VSO		 2600 default:
2601 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2602 request->type);
2603 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2604 errp);
2605 g_free(msg);
2606 return ret;
2607 }
2608}
2609
ff82911c
PB		 2610/* Owns a reference to the NBDClient passed as opaque. */
2611static coroutine_fn void nbd_trip(void *opaque)
75818250 2612{
262db388 2613 NBDClient *client = opaque;
315f78ab 2614 NBDRequestData *req;
ff82911c 2615 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
a0dc63a6 2616 int ret;
2fd2c840 2617 Error *local_err = NULL;
b2e3d87f 2618
9588463e 2619 trace_nbd_trip();
ff2b68aa 2620 if (client->closing) {
ff82911c 2621 nbd_client_put(client);
ff2b68aa
PB		 2622 return;
2623 }
b2e3d87f 2624
f148ae7d
SL		 2625 if (client->quiescing) {
2626 /*
2627 * We're switching between AIO contexts. Don't attempt to receive a new
2628 * request and kick the main context which may be waiting for us.
2629 */
2630 nbd_client_put(client);
2631 client->recv_coroutine = NULL;
2632 aio_wait_kick();
2633 return;
2634 }
2635
ff2b68aa 2636 req = nbd_request_get(client);
2fd2c840 2637 ret = nbd_co_receive_request(req, &request, &local_err);
ee898b87 2638 client->recv_coroutine = NULL;
b2e3d87f 2639
d6268348
WC		 2640 if (client->closing) {
2641 /*
2642 * The client may be closed when we are blocked in
2643 * nbd_co_receive_request()
2644 */
2645 goto done;
2646 }
2647
f148ae7d
SL		 2648 if (ret == -EAGAIN) {
2649 assert(client->quiescing);
2650 goto done;
2651 }
2652
a0d7ce20
VSO		 2653 nbd_client_receive_next_request(client);
2654 if (ret == -EIO) {
2655 goto disconnect;
2656 }
2657
2658 if (ret < 0) {
6a417599
VSO		 2659 /* It wans't -EIO, so, according to nbd_co_receive_request()
2660 * semantics, we should return the error to the client. */
2661 Error *export_err = local_err;
2662
2663 local_err = NULL;
2664 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2665 error_get_pretty(export_err), &local_err);
2666 error_free(export_err);
6f302e60
VSO		 2667 } else {
2668 ret = nbd_handle_request(client, &request, req->data, &local_err);
5c54e7fa
VSO		 2669 }
2670 if (ret < 0) {
c7b97282 2671 error_prepend(&local_err, "Failed to send reply: ");
2fd2c840
VSO		 2672 goto disconnect;
2673 }
2674
8c372a02
VSO		 2675 /* We must disconnect after NBD_CMD_WRITE if we did not
2676 * read the payload.
2677 */
2fd2c840
VSO		 2678 if (!req->complete) {
2679 error_setg(&local_err, "Request handling failed in intermediate state");
8c372a02 2680 goto disconnect;
b2e3d87f
NT		 2681 }
2682
7fe7b68b 2683done:
262db388 2684 nbd_request_put(req);
ff82911c 2685 nbd_client_put(client);
262db388
PB		 2686 return;
2687
8c372a02 2688disconnect:
2fd2c840
VSO		 2689 if (local_err) {
2690 error_reportf_err(local_err, "Disconnect client, due to: ");
2691 }
72deddc5 2692 nbd_request_put(req);
0c9390d9 2693 client_close(client, true);
ff82911c 2694 nbd_client_put(client);
7a5ca864 2695}
af49bbbe 2696
ff82911c 2697static void nbd_client_receive_next_request(NBDClient *client)
958c717d 2698{
f148ae7d
SL		 2699 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
2700 !client->quiescing) {
ff82911c
PB		 2701 nbd_client_get(client);
2702 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
8612c686 2703 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
958c717d
HR		 2704 }
2705}
2706
1a6245a5
FZ		 2707static coroutine_fn void nbd_co_client_start(void *opaque)
2708{
c84087f2 2709 NBDClient *client = opaque;
2fd2c840 2710 Error *local_err = NULL;
1a6245a5 2711
df8ad9f1
EB		 2712 qemu_co_mutex_init(&client->send_lock);
2713
2fd2c840
VSO		 2714 if (nbd_negotiate(client, &local_err)) {
2715 if (local_err) {
2716 error_report_err(local_err);
2717 }
0c9390d9 2718 client_close(client, false);
c84087f2 2719 return;
1a6245a5 2720 }
ff82911c
PB		 2721
2722 nbd_client_receive_next_request(client);
1a6245a5
FZ		 2723}
2724
0c9390d9 2725/*
7f7dfe2a
VSO		 2726 * Create a new client listener using the given channel @sioc.
2727 * Begin servicing it in a coroutine. When the connection closes, call
2728 * @close_fn with an indication of whether the client completed negotiation.
0c9390d9 2729 */
7f7dfe2a 2730void nbd_client_new(QIOChannelSocket *sioc,
f95910fe 2731 QCryptoTLSCreds *tlscreds,
b25e12da 2732 const char *tlsauthz,
0c9390d9 2733 void (*close_fn)(NBDClient *, bool))
af49bbbe 2734{
1743b515 2735 NBDClient *client;
c84087f2 2736 Coroutine *co;
1a6245a5 2737
e8d3eb74 2738 client = g_new0(NBDClient, 1);
1743b515 2739 client->refcount = 1;
f95910fe
DB		 2740 client->tlscreds = tlscreds;
2741 if (tlscreds) {
2742 object_ref(OBJECT(client->tlscreds));
2743 }
b25e12da 2744 client->tlsauthz = g_strdup(tlsauthz);
1c778ef7
DB		 2745 client->sioc = sioc;
2746 object_ref(OBJECT(client->sioc));
2747 client->ioc = QIO_CHANNEL(sioc);
2748 object_ref(OBJECT(client->ioc));
0c9390d9 2749 client->close_fn = close_fn;
2c8d9f06 2750
c84087f2
VSO		 2751 co = qemu_coroutine_create(nbd_co_client_start, client);
2752 qemu_coroutine_enter(co);
af49bbbe 2753}
QEMU mirror