audio: replace the resampling loop in audio_pcm_sw_write()
[qemu/ar7.git] / nbd / server.c
bloba4750e41880ab8a244df60277c1713d2d45478d7
1 /*
2 * Copyright (C) 2016-2022 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
5 * Network Block Device Server Side
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.
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.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
22 #include "block/block_int.h"
23 #include "block/export.h"
24 #include "block/dirty-bitmap.h"
25 #include "qapi/error.h"
26 #include "qemu/queue.h"
27 #include "trace.h"
28 #include "nbd-internal.h"
29 #include "qemu/units.h"
30 #include "qemu/memalign.h"
32 #define NBD_META_ID_BASE_ALLOCATION 0
33 #define NBD_META_ID_ALLOCATION_DEPTH 1
34 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
35 #define NBD_META_ID_DIRTY_BITMAP 2
38 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
39 * constant. If an increase is needed, note that the NBD protocol
40 * recommends no larger than 32 mb, so that the client won't consider
41 * the reply as a denial of service attack.
43 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
45 static int system_errno_to_nbd_errno(int err)
47 switch (err) {
48 case 0:
49 return NBD_SUCCESS;
50 case EPERM:
51 case EROFS:
52 return NBD_EPERM;
53 case EIO:
54 return NBD_EIO;
55 case ENOMEM:
56 return NBD_ENOMEM;
57 #ifdef EDQUOT
58 case EDQUOT:
59 #endif
60 case EFBIG:
61 case ENOSPC:
62 return NBD_ENOSPC;
63 case EOVERFLOW:
64 return NBD_EOVERFLOW;
65 case ENOTSUP:
66 #if ENOTSUP != EOPNOTSUPP
67 case EOPNOTSUPP:
68 #endif
69 return NBD_ENOTSUP;
70 case ESHUTDOWN:
71 return NBD_ESHUTDOWN;
72 case EINVAL:
73 default:
74 return NBD_EINVAL;
78 /* Definitions for opaque data types */
80 typedef struct NBDRequestData NBDRequestData;
82 struct NBDRequestData {
83 NBDClient *client;
84 uint8_t *data;
85 bool complete;
88 struct NBDExport {
89 BlockExport common;
91 char *name;
92 char *description;
93 uint64_t size;
94 uint16_t nbdflags;
95 QTAILQ_HEAD(, NBDClient) clients;
96 QTAILQ_ENTRY(NBDExport) next;
98 BlockBackend *eject_notifier_blk;
99 Notifier eject_notifier;
101 bool allocation_depth;
102 BdrvDirtyBitmap **export_bitmaps;
103 size_t nr_export_bitmaps;
106 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
108 /* NBDExportMetaContexts represents a list of contexts to be exported,
109 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
110 * NBD_OPT_LIST_META_CONTEXT. */
111 typedef struct NBDExportMetaContexts {
112 NBDExport *exp;
113 size_t count; /* number of negotiated contexts */
114 bool base_allocation; /* export base:allocation context (block status) */
115 bool allocation_depth; /* export qemu:allocation-depth */
116 bool *bitmaps; /*
117 * export qemu:dirty-bitmap:<export bitmap name>,
118 * sized by exp->nr_export_bitmaps
120 } NBDExportMetaContexts;
122 struct NBDClient {
123 int refcount;
124 void (*close_fn)(NBDClient *client, bool negotiated);
126 NBDExport *exp;
127 QCryptoTLSCreds *tlscreds;
128 char *tlsauthz;
129 QIOChannelSocket *sioc; /* The underlying data channel */
130 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
132 Coroutine *recv_coroutine;
134 CoMutex send_lock;
135 Coroutine *send_coroutine;
137 bool read_yielding;
138 bool quiescing;
140 QTAILQ_ENTRY(NBDClient) next;
141 int nb_requests;
142 bool closing;
144 uint32_t check_align; /* If non-zero, check for aligned client requests */
146 bool structured_reply;
147 NBDExportMetaContexts export_meta;
149 uint32_t opt; /* Current option being negotiated */
150 uint32_t optlen; /* remaining length of data in ioc for the option being
151 negotiated now */
154 static void nbd_client_receive_next_request(NBDClient *client);
156 /* Basic flow for negotiation
158 Server Client
159 Negotiate
163 Server Client
164 Negotiate #1
165 Option
166 Negotiate #2
168 ----
170 followed by
172 Server Client
173 Request
174 Response
175 Request
176 Response
179 Request (type == 2)
183 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
184 uint32_t type, uint32_t length)
186 stq_be_p(&rep->magic, NBD_REP_MAGIC);
187 stl_be_p(&rep->option, option);
188 stl_be_p(&rep->type, type);
189 stl_be_p(&rep->length, length);
192 /* Send a reply header, including length, but no payload.
193 * Return -errno on error, 0 on success. */
194 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
195 uint32_t len, Error **errp)
197 NBDOptionReply rep;
199 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
200 type, nbd_rep_lookup(type), len);
202 assert(len < NBD_MAX_BUFFER_SIZE);
204 set_be_option_rep(&rep, client->opt, type, len);
205 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
208 /* Send a reply header with default 0 length.
209 * Return -errno on error, 0 on success. */
210 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
211 Error **errp)
213 return nbd_negotiate_send_rep_len(client, type, 0, errp);
216 /* Send an error reply.
217 * Return -errno on error, 0 on success. */
218 static int G_GNUC_PRINTF(4, 0)
219 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
220 Error **errp, const char *fmt, va_list va)
222 ERRP_GUARD();
223 g_autofree char *msg = NULL;
224 int ret;
225 size_t len;
227 msg = g_strdup_vprintf(fmt, va);
228 len = strlen(msg);
229 assert(len < NBD_MAX_STRING_SIZE);
230 trace_nbd_negotiate_send_rep_err(msg);
231 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
232 if (ret < 0) {
233 return ret;
235 if (nbd_write(client->ioc, msg, len, errp) < 0) {
236 error_prepend(errp, "write failed (error message): ");
237 return -EIO;
240 return 0;
244 * Return a malloc'd copy of @name suitable for use in an error reply.
246 static char *
247 nbd_sanitize_name(const char *name)
249 if (strnlen(name, 80) < 80) {
250 return g_strdup(name);
252 /* XXX Should we also try to sanitize any control characters? */
253 return g_strdup_printf("%.80s...", name);
256 /* Send an error reply.
257 * Return -errno on error, 0 on success. */
258 static int G_GNUC_PRINTF(4, 5)
259 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
260 Error **errp, const char *fmt, ...)
262 va_list va;
263 int ret;
265 va_start(va, fmt);
266 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
267 va_end(va);
268 return ret;
271 /* Drop remainder of the current option, and send a reply with the
272 * given error type and message. Return -errno on read or write
273 * failure; or 0 if connection is still live. */
274 static int G_GNUC_PRINTF(4, 0)
275 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
276 const char *fmt, va_list va)
278 int ret = nbd_drop(client->ioc, client->optlen, errp);
280 client->optlen = 0;
281 if (!ret) {
282 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
284 return ret;
287 static int G_GNUC_PRINTF(4, 5)
288 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
289 const char *fmt, ...)
291 int ret;
292 va_list va;
294 va_start(va, fmt);
295 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
296 va_end(va);
298 return ret;
301 static int G_GNUC_PRINTF(3, 4)
302 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
304 int ret;
305 va_list va;
307 va_start(va, fmt);
308 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
309 va_end(va);
311 return ret;
314 /* Read size bytes from the unparsed payload of the current option.
315 * If @check_nul, require that no NUL bytes appear in buffer.
316 * Return -errno on I/O error, 0 if option was completely handled by
317 * sending a reply about inconsistent lengths, or 1 on success. */
318 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
319 bool check_nul, Error **errp)
321 if (size > client->optlen) {
322 return nbd_opt_invalid(client, errp,
323 "Inconsistent lengths in option %s",
324 nbd_opt_lookup(client->opt));
326 client->optlen -= size;
327 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
328 return -EIO;
331 if (check_nul && strnlen(buffer, size) != size) {
332 return nbd_opt_invalid(client, errp,
333 "Unexpected embedded NUL in option %s",
334 nbd_opt_lookup(client->opt));
336 return 1;
339 /* Drop size bytes from the unparsed payload of the current option.
340 * Return -errno on I/O error, 0 if option was completely handled by
341 * sending a reply about inconsistent lengths, or 1 on success. */
342 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
344 if (size > client->optlen) {
345 return nbd_opt_invalid(client, errp,
346 "Inconsistent lengths in option %s",
347 nbd_opt_lookup(client->opt));
349 client->optlen -= size;
350 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
353 /* nbd_opt_read_name
355 * Read a string with the format:
356 * uint32_t len (<= NBD_MAX_STRING_SIZE)
357 * len bytes string (not 0-terminated)
359 * On success, @name will be allocated.
360 * If @length is non-null, it will be set to the actual string length.
362 * Return -errno on I/O error, 0 if option was completely handled by
363 * sending a reply about inconsistent lengths, or 1 on success.
365 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
366 Error **errp)
368 int ret;
369 uint32_t len;
370 g_autofree char *local_name = NULL;
372 *name = NULL;
373 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
374 if (ret <= 0) {
375 return ret;
377 len = cpu_to_be32(len);
379 if (len > NBD_MAX_STRING_SIZE) {
380 return nbd_opt_invalid(client, errp,
381 "Invalid name length: %" PRIu32, len);
384 local_name = g_malloc(len + 1);
385 ret = nbd_opt_read(client, local_name, len, true, errp);
386 if (ret <= 0) {
387 return ret;
389 local_name[len] = '\0';
391 if (length) {
392 *length = len;
394 *name = g_steal_pointer(&local_name);
396 return 1;
399 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
400 * Return -errno on error, 0 on success. */
401 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
402 Error **errp)
404 ERRP_GUARD();
405 size_t name_len, desc_len;
406 uint32_t len;
407 const char *name = exp->name ? exp->name : "";
408 const char *desc = exp->description ? exp->description : "";
409 QIOChannel *ioc = client->ioc;
410 int ret;
412 trace_nbd_negotiate_send_rep_list(name, desc);
413 name_len = strlen(name);
414 desc_len = strlen(desc);
415 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
416 len = name_len + desc_len + sizeof(len);
417 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
418 if (ret < 0) {
419 return ret;
422 len = cpu_to_be32(name_len);
423 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
424 error_prepend(errp, "write failed (name length): ");
425 return -EINVAL;
428 if (nbd_write(ioc, name, name_len, errp) < 0) {
429 error_prepend(errp, "write failed (name buffer): ");
430 return -EINVAL;
433 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
434 error_prepend(errp, "write failed (description buffer): ");
435 return -EINVAL;
438 return 0;
441 /* Process the NBD_OPT_LIST command, with a potential series of replies.
442 * Return -errno on error, 0 on success. */
443 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
445 NBDExport *exp;
446 assert(client->opt == NBD_OPT_LIST);
448 /* For each export, send a NBD_REP_SERVER reply. */
449 QTAILQ_FOREACH(exp, &exports, next) {
450 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
451 return -EINVAL;
454 /* Finish with a NBD_REP_ACK. */
455 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
458 static void nbd_check_meta_export(NBDClient *client)
460 if (client->exp != client->export_meta.exp) {
461 client->export_meta.count = 0;
465 /* Send a reply to NBD_OPT_EXPORT_NAME.
466 * Return -errno on error, 0 on success. */
467 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
468 Error **errp)
470 ERRP_GUARD();
471 g_autofree char *name = NULL;
472 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
473 size_t len;
474 int ret;
475 uint16_t myflags;
477 /* Client sends:
478 [20 .. xx] export name (length bytes)
479 Server replies:
480 [ 0 .. 7] size
481 [ 8 .. 9] export flags
482 [10 .. 133] reserved (0) [unless no_zeroes]
484 trace_nbd_negotiate_handle_export_name();
485 if (client->optlen > NBD_MAX_STRING_SIZE) {
486 error_setg(errp, "Bad length received");
487 return -EINVAL;
489 name = g_malloc(client->optlen + 1);
490 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
491 return -EIO;
493 name[client->optlen] = '\0';
494 client->optlen = 0;
496 trace_nbd_negotiate_handle_export_name_request(name);
498 client->exp = nbd_export_find(name);
499 if (!client->exp) {
500 error_setg(errp, "export not found");
501 return -EINVAL;
504 myflags = client->exp->nbdflags;
505 if (client->structured_reply) {
506 myflags |= NBD_FLAG_SEND_DF;
508 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
509 stq_be_p(buf, client->exp->size);
510 stw_be_p(buf + 8, myflags);
511 len = no_zeroes ? 10 : sizeof(buf);
512 ret = nbd_write(client->ioc, buf, len, errp);
513 if (ret < 0) {
514 error_prepend(errp, "write failed: ");
515 return ret;
518 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
519 blk_exp_ref(&client->exp->common);
520 nbd_check_meta_export(client);
522 return 0;
525 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
526 * The buffer does NOT include the info type prefix.
527 * Return -errno on error, 0 if ready to send more. */
528 static int nbd_negotiate_send_info(NBDClient *client,
529 uint16_t info, uint32_t length, void *buf,
530 Error **errp)
532 int rc;
534 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
535 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
536 sizeof(info) + length, errp);
537 if (rc < 0) {
538 return rc;
540 info = cpu_to_be16(info);
541 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
542 return -EIO;
544 if (nbd_write(client->ioc, buf, length, errp) < 0) {
545 return -EIO;
547 return 0;
550 /* nbd_reject_length: Handle any unexpected payload.
551 * @fatal requests that we quit talking to the client, even if we are able
552 * to successfully send an error reply.
553 * Return:
554 * -errno transmission error occurred or @fatal was requested, errp is set
555 * 0 error message successfully sent to client, errp is not set
557 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
559 int ret;
561 assert(client->optlen);
562 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
563 nbd_opt_lookup(client->opt));
564 if (fatal && !ret) {
565 error_setg(errp, "option '%s' has unexpected length",
566 nbd_opt_lookup(client->opt));
567 return -EINVAL;
569 return ret;
572 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
573 * Return -errno on error, 0 if ready for next option, and 1 to move
574 * into transmission phase. */
575 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
577 int rc;
578 g_autofree char *name = NULL;
579 NBDExport *exp;
580 uint16_t requests;
581 uint16_t request;
582 uint32_t namelen = 0;
583 bool sendname = false;
584 bool blocksize = false;
585 uint32_t sizes[3];
586 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
587 uint32_t check_align = 0;
588 uint16_t myflags;
590 /* Client sends:
591 4 bytes: L, name length (can be 0)
592 L bytes: export name
593 2 bytes: N, number of requests (can be 0)
594 N * 2 bytes: N requests
596 rc = nbd_opt_read_name(client, &name, &namelen, errp);
597 if (rc <= 0) {
598 return rc;
600 trace_nbd_negotiate_handle_export_name_request(name);
602 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
603 if (rc <= 0) {
604 return rc;
606 requests = be16_to_cpu(requests);
607 trace_nbd_negotiate_handle_info_requests(requests);
608 while (requests--) {
609 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
610 if (rc <= 0) {
611 return rc;
613 request = be16_to_cpu(request);
614 trace_nbd_negotiate_handle_info_request(request,
615 nbd_info_lookup(request));
616 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
617 * everything else is either a request we don't know or
618 * something we send regardless of request */
619 switch (request) {
620 case NBD_INFO_NAME:
621 sendname = true;
622 break;
623 case NBD_INFO_BLOCK_SIZE:
624 blocksize = true;
625 break;
628 if (client->optlen) {
629 return nbd_reject_length(client, false, errp);
632 exp = nbd_export_find(name);
633 if (!exp) {
634 g_autofree char *sane_name = nbd_sanitize_name(name);
636 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
637 errp, "export '%s' not present",
638 sane_name);
641 /* Don't bother sending NBD_INFO_NAME unless client requested it */
642 if (sendname) {
643 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
644 errp);
645 if (rc < 0) {
646 return rc;
650 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
651 * client request */
652 if (exp->description) {
653 size_t len = strlen(exp->description);
655 assert(len <= NBD_MAX_STRING_SIZE);
656 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
657 len, exp->description, errp);
658 if (rc < 0) {
659 return rc;
663 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
664 * according to whether the client requested it, and according to
665 * whether this is OPT_INFO or OPT_GO. */
666 /* minimum - 1 for back-compat, or actual if client will obey it. */
667 if (client->opt == NBD_OPT_INFO || blocksize) {
668 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
669 } else {
670 sizes[0] = 1;
672 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
673 /* preferred - Hard-code to 4096 for now.
674 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
675 sizes[1] = MAX(4096, sizes[0]);
676 /* maximum - At most 32M, but smaller as appropriate. */
677 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
678 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
679 sizes[0] = cpu_to_be32(sizes[0]);
680 sizes[1] = cpu_to_be32(sizes[1]);
681 sizes[2] = cpu_to_be32(sizes[2]);
682 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
683 sizeof(sizes), sizes, errp);
684 if (rc < 0) {
685 return rc;
688 /* Send NBD_INFO_EXPORT always */
689 myflags = exp->nbdflags;
690 if (client->structured_reply) {
691 myflags |= NBD_FLAG_SEND_DF;
693 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
694 stq_be_p(buf, exp->size);
695 stw_be_p(buf + 8, myflags);
696 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
697 sizeof(buf), buf, errp);
698 if (rc < 0) {
699 return rc;
703 * If the client is just asking for NBD_OPT_INFO, but forgot to
704 * request block sizes in a situation that would impact
705 * performance, then return an error. But for NBD_OPT_GO, we
706 * tolerate all clients, regardless of alignments.
708 if (client->opt == NBD_OPT_INFO && !blocksize &&
709 blk_get_request_alignment(exp->common.blk) > 1) {
710 return nbd_negotiate_send_rep_err(client,
711 NBD_REP_ERR_BLOCK_SIZE_REQD,
712 errp,
713 "request NBD_INFO_BLOCK_SIZE to "
714 "use this export");
717 /* Final reply */
718 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
719 if (rc < 0) {
720 return rc;
723 if (client->opt == NBD_OPT_GO) {
724 client->exp = exp;
725 client->check_align = check_align;
726 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
727 blk_exp_ref(&client->exp->common);
728 nbd_check_meta_export(client);
729 rc = 1;
731 return rc;
735 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
736 * new channel for all further (now-encrypted) communication. */
737 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
738 Error **errp)
740 QIOChannel *ioc;
741 QIOChannelTLS *tioc;
742 struct NBDTLSHandshakeData data = { 0 };
744 assert(client->opt == NBD_OPT_STARTTLS);
746 trace_nbd_negotiate_handle_starttls();
747 ioc = client->ioc;
749 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
750 return NULL;
753 tioc = qio_channel_tls_new_server(ioc,
754 client->tlscreds,
755 client->tlsauthz,
756 errp);
757 if (!tioc) {
758 return NULL;
761 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
762 trace_nbd_negotiate_handle_starttls_handshake();
763 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
764 qio_channel_tls_handshake(tioc,
765 nbd_tls_handshake,
766 &data,
767 NULL,
768 NULL);
770 if (!data.complete) {
771 g_main_loop_run(data.loop);
773 g_main_loop_unref(data.loop);
774 if (data.error) {
775 object_unref(OBJECT(tioc));
776 error_propagate(errp, data.error);
777 return NULL;
780 return QIO_CHANNEL(tioc);
783 /* nbd_negotiate_send_meta_context
785 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
787 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
789 static int nbd_negotiate_send_meta_context(NBDClient *client,
790 const char *context,
791 uint32_t context_id,
792 Error **errp)
794 NBDOptionReplyMetaContext opt;
795 struct iovec iov[] = {
796 {.iov_base = &opt, .iov_len = sizeof(opt)},
797 {.iov_base = (void *)context, .iov_len = strlen(context)}
800 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
801 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
802 context_id = 0;
805 trace_nbd_negotiate_meta_query_reply(context, context_id);
806 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
807 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
808 stl_be_p(&opt.context_id, context_id);
810 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
814 * Return true if @query matches @pattern, or if @query is empty when
815 * the @client is performing _LIST_.
817 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
818 const char *query)
820 if (!*query) {
821 trace_nbd_negotiate_meta_query_parse("empty");
822 return client->opt == NBD_OPT_LIST_META_CONTEXT;
824 if (strcmp(query, pattern) == 0) {
825 trace_nbd_negotiate_meta_query_parse(pattern);
826 return true;
828 trace_nbd_negotiate_meta_query_skip("pattern not matched");
829 return false;
833 * Return true and adjust @str in place if it begins with @prefix.
835 static bool nbd_strshift(const char **str, const char *prefix)
837 size_t len = strlen(prefix);
839 if (strncmp(*str, prefix, len) == 0) {
840 *str += len;
841 return true;
843 return false;
846 /* nbd_meta_base_query
848 * Handle queries to 'base' namespace. For now, only the base:allocation
849 * context is available. Return true if @query has been handled.
851 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
852 const char *query)
854 if (!nbd_strshift(&query, "base:")) {
855 return false;
857 trace_nbd_negotiate_meta_query_parse("base:");
859 if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
860 meta->base_allocation = true;
862 return true;
865 /* nbd_meta_qemu_query
867 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
868 * and qemu:allocation-depth contexts are available. Return true if @query
869 * has been handled.
871 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
872 const char *query)
874 size_t i;
876 if (!nbd_strshift(&query, "qemu:")) {
877 return false;
879 trace_nbd_negotiate_meta_query_parse("qemu:");
881 if (!*query) {
882 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
883 meta->allocation_depth = meta->exp->allocation_depth;
884 if (meta->exp->nr_export_bitmaps) {
885 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
888 trace_nbd_negotiate_meta_query_parse("empty");
889 return true;
892 if (strcmp(query, "allocation-depth") == 0) {
893 trace_nbd_negotiate_meta_query_parse("allocation-depth");
894 meta->allocation_depth = meta->exp->allocation_depth;
895 return true;
898 if (nbd_strshift(&query, "dirty-bitmap:")) {
899 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
900 if (!*query) {
901 if (client->opt == NBD_OPT_LIST_META_CONTEXT &&
902 meta->exp->nr_export_bitmaps) {
903 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
905 trace_nbd_negotiate_meta_query_parse("empty");
906 return true;
909 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
910 const char *bm_name;
912 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
913 if (strcmp(bm_name, query) == 0) {
914 meta->bitmaps[i] = true;
915 trace_nbd_negotiate_meta_query_parse(query);
916 return true;
919 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
920 return true;
923 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
924 return true;
927 /* nbd_negotiate_meta_query
929 * Parse namespace name and call corresponding function to parse body of the
930 * query.
932 * The only supported namespaces are 'base' and 'qemu'.
934 * Return -errno on I/O error, 0 if option was completely handled by
935 * sending a reply about inconsistent lengths, or 1 on success. */
936 static int nbd_negotiate_meta_query(NBDClient *client,
937 NBDExportMetaContexts *meta, Error **errp)
939 int ret;
940 g_autofree char *query = NULL;
941 uint32_t len;
943 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
944 if (ret <= 0) {
945 return ret;
947 len = cpu_to_be32(len);
949 if (len > NBD_MAX_STRING_SIZE) {
950 trace_nbd_negotiate_meta_query_skip("length too long");
951 return nbd_opt_skip(client, len, errp);
954 query = g_malloc(len + 1);
955 ret = nbd_opt_read(client, query, len, true, errp);
956 if (ret <= 0) {
957 return ret;
959 query[len] = '\0';
961 if (nbd_meta_base_query(client, meta, query)) {
962 return 1;
964 if (nbd_meta_qemu_query(client, meta, query)) {
965 return 1;
968 trace_nbd_negotiate_meta_query_skip("unknown namespace");
969 return 1;
972 /* nbd_negotiate_meta_queries
973 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
975 * Return -errno on I/O error, or 0 if option was completely handled. */
976 static int nbd_negotiate_meta_queries(NBDClient *client,
977 NBDExportMetaContexts *meta, Error **errp)
979 int ret;
980 g_autofree char *export_name = NULL;
981 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */
982 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL;
983 NBDExportMetaContexts local_meta = {0};
984 uint32_t nb_queries;
985 size_t i;
986 size_t count = 0;
988 if (client->opt == NBD_OPT_SET_META_CONTEXT && !client->structured_reply) {
989 return nbd_opt_invalid(client, errp,
990 "request option '%s' when structured reply "
991 "is not negotiated",
992 nbd_opt_lookup(client->opt));
995 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
996 /* Only change the caller's meta on SET. */
997 meta = &local_meta;
1000 g_free(meta->bitmaps);
1001 memset(meta, 0, sizeof(*meta));
1003 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
1004 if (ret <= 0) {
1005 return ret;
1008 meta->exp = nbd_export_find(export_name);
1009 if (meta->exp == NULL) {
1010 g_autofree char *sane_name = nbd_sanitize_name(export_name);
1012 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1013 "export '%s' not present", sane_name);
1015 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1016 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1017 bitmaps = meta->bitmaps;
1020 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1021 if (ret <= 0) {
1022 return ret;
1024 nb_queries = cpu_to_be32(nb_queries);
1025 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1026 export_name, nb_queries);
1028 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1029 /* enable all known contexts */
1030 meta->base_allocation = true;
1031 meta->allocation_depth = meta->exp->allocation_depth;
1032 if (meta->exp->nr_export_bitmaps) {
1033 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1035 } else {
1036 for (i = 0; i < nb_queries; ++i) {
1037 ret = nbd_negotiate_meta_query(client, meta, errp);
1038 if (ret <= 0) {
1039 return ret;
1044 if (meta->base_allocation) {
1045 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1046 NBD_META_ID_BASE_ALLOCATION,
1047 errp);
1048 if (ret < 0) {
1049 return ret;
1051 count++;
1054 if (meta->allocation_depth) {
1055 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1056 NBD_META_ID_ALLOCATION_DEPTH,
1057 errp);
1058 if (ret < 0) {
1059 return ret;
1061 count++;
1064 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1065 const char *bm_name;
1066 g_autofree char *context = NULL;
1068 if (!meta->bitmaps[i]) {
1069 continue;
1072 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1073 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1075 ret = nbd_negotiate_send_meta_context(client, context,
1076 NBD_META_ID_DIRTY_BITMAP + i,
1077 errp);
1078 if (ret < 0) {
1079 return ret;
1081 count++;
1084 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1085 if (ret == 0) {
1086 meta->count = count;
1089 return ret;
1092 /* nbd_negotiate_options
1093 * Process all NBD_OPT_* client option commands, during fixed newstyle
1094 * negotiation.
1095 * Return:
1096 * -errno on error, errp is set
1097 * 0 on successful negotiation, errp is not set
1098 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1099 * errp is not set
1101 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1103 uint32_t flags;
1104 bool fixedNewstyle = false;
1105 bool no_zeroes = false;
1107 /* Client sends:
1108 [ 0 .. 3] client flags
1110 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1111 [ 0 .. 7] NBD_OPTS_MAGIC
1112 [ 8 .. 11] NBD option
1113 [12 .. 15] Data length
1114 ... Rest of request
1116 [ 0 .. 7] NBD_OPTS_MAGIC
1117 [ 8 .. 11] Second NBD option
1118 [12 .. 15] Data length
1119 ... Rest of request
1122 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1123 return -EIO;
1125 trace_nbd_negotiate_options_flags(flags);
1126 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1127 fixedNewstyle = true;
1128 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1130 if (flags & NBD_FLAG_C_NO_ZEROES) {
1131 no_zeroes = true;
1132 flags &= ~NBD_FLAG_C_NO_ZEROES;
1134 if (flags != 0) {
1135 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1136 return -EINVAL;
1139 while (1) {
1140 int ret;
1141 uint32_t option, length;
1142 uint64_t magic;
1144 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1145 return -EINVAL;
1147 trace_nbd_negotiate_options_check_magic(magic);
1148 if (magic != NBD_OPTS_MAGIC) {
1149 error_setg(errp, "Bad magic received");
1150 return -EINVAL;
1153 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1154 return -EINVAL;
1156 client->opt = option;
1158 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1159 return -EINVAL;
1161 assert(!client->optlen);
1162 client->optlen = length;
1164 if (length > NBD_MAX_BUFFER_SIZE) {
1165 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1166 length, NBD_MAX_BUFFER_SIZE);
1167 return -EINVAL;
1170 trace_nbd_negotiate_options_check_option(option,
1171 nbd_opt_lookup(option));
1172 if (client->tlscreds &&
1173 client->ioc == (QIOChannel *)client->sioc) {
1174 QIOChannel *tioc;
1175 if (!fixedNewstyle) {
1176 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1177 return -EINVAL;
1179 switch (option) {
1180 case NBD_OPT_STARTTLS:
1181 if (length) {
1182 /* Unconditionally drop the connection if the client
1183 * can't start a TLS negotiation correctly */
1184 return nbd_reject_length(client, true, errp);
1186 tioc = nbd_negotiate_handle_starttls(client, errp);
1187 if (!tioc) {
1188 return -EIO;
1190 ret = 0;
1191 object_unref(OBJECT(client->ioc));
1192 client->ioc = QIO_CHANNEL(tioc);
1193 break;
1195 case NBD_OPT_EXPORT_NAME:
1196 /* No way to return an error to client, so drop connection */
1197 error_setg(errp, "Option 0x%x not permitted before TLS",
1198 option);
1199 return -EINVAL;
1201 default:
1202 /* Let the client keep trying, unless they asked to
1203 * quit. Always try to give an error back to the
1204 * client; but when replying to OPT_ABORT, be aware
1205 * that the client may hang up before receiving the
1206 * error, in which case we are fine ignoring the
1207 * resulting EPIPE. */
1208 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1209 option == NBD_OPT_ABORT ? NULL : errp,
1210 "Option 0x%" PRIx32
1211 " not permitted before TLS", option);
1212 if (option == NBD_OPT_ABORT) {
1213 return 1;
1215 break;
1217 } else if (fixedNewstyle) {
1218 switch (option) {
1219 case NBD_OPT_LIST:
1220 if (length) {
1221 ret = nbd_reject_length(client, false, errp);
1222 } else {
1223 ret = nbd_negotiate_handle_list(client, errp);
1225 break;
1227 case NBD_OPT_ABORT:
1228 /* NBD spec says we must try to reply before
1229 * disconnecting, but that we must also tolerate
1230 * guests that don't wait for our reply. */
1231 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1232 return 1;
1234 case NBD_OPT_EXPORT_NAME:
1235 return nbd_negotiate_handle_export_name(client, no_zeroes,
1236 errp);
1238 case NBD_OPT_INFO:
1239 case NBD_OPT_GO:
1240 ret = nbd_negotiate_handle_info(client, errp);
1241 if (ret == 1) {
1242 assert(option == NBD_OPT_GO);
1243 return 0;
1245 break;
1247 case NBD_OPT_STARTTLS:
1248 if (length) {
1249 ret = nbd_reject_length(client, false, errp);
1250 } else if (client->tlscreds) {
1251 ret = nbd_negotiate_send_rep_err(client,
1252 NBD_REP_ERR_INVALID, errp,
1253 "TLS already enabled");
1254 } else {
1255 ret = nbd_negotiate_send_rep_err(client,
1256 NBD_REP_ERR_POLICY, errp,
1257 "TLS not configured");
1259 break;
1261 case NBD_OPT_STRUCTURED_REPLY:
1262 if (length) {
1263 ret = nbd_reject_length(client, false, errp);
1264 } else if (client->structured_reply) {
1265 ret = nbd_negotiate_send_rep_err(
1266 client, NBD_REP_ERR_INVALID, errp,
1267 "structured reply already negotiated");
1268 } else {
1269 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1270 client->structured_reply = true;
1272 break;
1274 case NBD_OPT_LIST_META_CONTEXT:
1275 case NBD_OPT_SET_META_CONTEXT:
1276 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1277 errp);
1278 break;
1280 default:
1281 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1282 "Unsupported option %" PRIu32 " (%s)",
1283 option, nbd_opt_lookup(option));
1284 break;
1286 } else {
1288 * If broken new-style we should drop the connection
1289 * for anything except NBD_OPT_EXPORT_NAME
1291 switch (option) {
1292 case NBD_OPT_EXPORT_NAME:
1293 return nbd_negotiate_handle_export_name(client, no_zeroes,
1294 errp);
1296 default:
1297 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1298 option, nbd_opt_lookup(option));
1299 return -EINVAL;
1302 if (ret < 0) {
1303 return ret;
1308 /* nbd_negotiate
1309 * Return:
1310 * -errno on error, errp is set
1311 * 0 on successful negotiation, errp is not set
1312 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1313 * errp is not set
1315 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1317 ERRP_GUARD();
1318 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1319 int ret;
1321 /* Old style negotiation header, no room for options
1322 [ 0 .. 7] passwd ("NBDMAGIC")
1323 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1324 [16 .. 23] size
1325 [24 .. 27] export flags (zero-extended)
1326 [28 .. 151] reserved (0)
1328 New style negotiation header, client can send options
1329 [ 0 .. 7] passwd ("NBDMAGIC")
1330 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1331 [16 .. 17] server flags (0)
1332 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1335 qio_channel_set_blocking(client->ioc, false, NULL);
1337 trace_nbd_negotiate_begin();
1338 memcpy(buf, "NBDMAGIC", 8);
1340 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1341 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1343 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1344 error_prepend(errp, "write failed: ");
1345 return -EINVAL;
1347 ret = nbd_negotiate_options(client, errp);
1348 if (ret != 0) {
1349 if (ret < 0) {
1350 error_prepend(errp, "option negotiation failed: ");
1352 return ret;
1355 /* Attach the channel to the same AioContext as the export */
1356 if (client->exp && client->exp->common.ctx) {
1357 qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
1360 assert(!client->optlen);
1361 trace_nbd_negotiate_success();
1363 return 0;
1366 /* nbd_read_eof
1367 * Tries to read @size bytes from @ioc. This is a local implementation of
1368 * qio_channel_readv_all_eof. We have it here because we need it to be
1369 * interruptible and to know when the coroutine is yielding.
1370 * Returns 1 on success
1371 * 0 on eof, when no data was read (errp is not set)
1372 * negative errno on failure (errp is set)
1374 static inline int coroutine_fn
1375 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1377 bool partial = false;
1379 assert(size);
1380 while (size > 0) {
1381 struct iovec iov = { .iov_base = buffer, .iov_len = size };
1382 ssize_t len;
1384 len = qio_channel_readv(client->ioc, &iov, 1, errp);
1385 if (len == QIO_CHANNEL_ERR_BLOCK) {
1386 client->read_yielding = true;
1387 qio_channel_yield(client->ioc, G_IO_IN);
1388 client->read_yielding = false;
1389 if (client->quiescing) {
1390 return -EAGAIN;
1392 continue;
1393 } else if (len < 0) {
1394 return -EIO;
1395 } else if (len == 0) {
1396 if (partial) {
1397 error_setg(errp,
1398 "Unexpected end-of-file before all bytes were read");
1399 return -EIO;
1400 } else {
1401 return 0;
1405 partial = true;
1406 size -= len;
1407 buffer = (uint8_t *) buffer + len;
1409 return 1;
1412 static int nbd_receive_request(NBDClient *client, NBDRequest *request,
1413 Error **errp)
1415 uint8_t buf[NBD_REQUEST_SIZE];
1416 uint32_t magic;
1417 int ret;
1419 ret = nbd_read_eof(client, buf, sizeof(buf), errp);
1420 if (ret < 0) {
1421 return ret;
1423 if (ret == 0) {
1424 return -EIO;
1427 /* Request
1428 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1429 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1430 [ 6 .. 7] type (NBD_CMD_READ, ...)
1431 [ 8 .. 15] handle
1432 [16 .. 23] from
1433 [24 .. 27] len
1436 magic = ldl_be_p(buf);
1437 request->flags = lduw_be_p(buf + 4);
1438 request->type = lduw_be_p(buf + 6);
1439 request->handle = ldq_be_p(buf + 8);
1440 request->from = ldq_be_p(buf + 16);
1441 request->len = ldl_be_p(buf + 24);
1443 trace_nbd_receive_request(magic, request->flags, request->type,
1444 request->from, request->len);
1446 if (magic != NBD_REQUEST_MAGIC) {
1447 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1448 return -EINVAL;
1450 return 0;
1453 #define MAX_NBD_REQUESTS 16
1455 void nbd_client_get(NBDClient *client)
1457 client->refcount++;
1460 void nbd_client_put(NBDClient *client)
1462 if (--client->refcount == 0) {
1463 /* The last reference should be dropped by client->close,
1464 * which is called by client_close.
1466 assert(client->closing);
1468 qio_channel_detach_aio_context(client->ioc);
1469 object_unref(OBJECT(client->sioc));
1470 object_unref(OBJECT(client->ioc));
1471 if (client->tlscreds) {
1472 object_unref(OBJECT(client->tlscreds));
1474 g_free(client->tlsauthz);
1475 if (client->exp) {
1476 QTAILQ_REMOVE(&client->exp->clients, client, next);
1477 blk_exp_unref(&client->exp->common);
1479 g_free(client->export_meta.bitmaps);
1480 g_free(client);
1484 static void client_close(NBDClient *client, bool negotiated)
1486 if (client->closing) {
1487 return;
1490 client->closing = true;
1492 /* Force requests to finish. They will drop their own references,
1493 * then we'll close the socket and free the NBDClient.
1495 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1496 NULL);
1498 /* Also tell the client, so that they release their reference. */
1499 if (client->close_fn) {
1500 client->close_fn(client, negotiated);
1504 static NBDRequestData *nbd_request_get(NBDClient *client)
1506 NBDRequestData *req;
1508 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1509 client->nb_requests++;
1511 req = g_new0(NBDRequestData, 1);
1512 nbd_client_get(client);
1513 req->client = client;
1514 return req;
1517 static void nbd_request_put(NBDRequestData *req)
1519 NBDClient *client = req->client;
1521 if (req->data) {
1522 qemu_vfree(req->data);
1524 g_free(req);
1526 client->nb_requests--;
1528 if (client->quiescing && client->nb_requests == 0) {
1529 aio_wait_kick();
1532 nbd_client_receive_next_request(client);
1534 nbd_client_put(client);
1537 static void blk_aio_attached(AioContext *ctx, void *opaque)
1539 NBDExport *exp = opaque;
1540 NBDClient *client;
1542 trace_nbd_blk_aio_attached(exp->name, ctx);
1544 exp->common.ctx = ctx;
1546 QTAILQ_FOREACH(client, &exp->clients, next) {
1547 qio_channel_attach_aio_context(client->ioc, ctx);
1549 assert(client->nb_requests == 0);
1550 assert(client->recv_coroutine == NULL);
1551 assert(client->send_coroutine == NULL);
1555 static void blk_aio_detach(void *opaque)
1557 NBDExport *exp = opaque;
1558 NBDClient *client;
1560 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1562 QTAILQ_FOREACH(client, &exp->clients, next) {
1563 qio_channel_detach_aio_context(client->ioc);
1566 exp->common.ctx = NULL;
1569 static void nbd_drained_begin(void *opaque)
1571 NBDExport *exp = opaque;
1572 NBDClient *client;
1574 QTAILQ_FOREACH(client, &exp->clients, next) {
1575 client->quiescing = true;
1579 static void nbd_drained_end(void *opaque)
1581 NBDExport *exp = opaque;
1582 NBDClient *client;
1584 QTAILQ_FOREACH(client, &exp->clients, next) {
1585 client->quiescing = false;
1586 nbd_client_receive_next_request(client);
1590 static bool nbd_drained_poll(void *opaque)
1592 NBDExport *exp = opaque;
1593 NBDClient *client;
1595 QTAILQ_FOREACH(client, &exp->clients, next) {
1596 if (client->nb_requests != 0) {
1598 * If there's a coroutine waiting for a request on nbd_read_eof()
1599 * enter it here so we don't depend on the client to wake it up.
1601 if (client->recv_coroutine != NULL && client->read_yielding) {
1602 qemu_aio_coroutine_enter(exp->common.ctx,
1603 client->recv_coroutine);
1606 return true;
1610 return false;
1613 static void nbd_eject_notifier(Notifier *n, void *data)
1615 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1617 blk_exp_request_shutdown(&exp->common);
1620 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1622 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1623 assert(exp->drv == &blk_exp_nbd);
1624 assert(nbd_exp->eject_notifier_blk == NULL);
1626 blk_ref(blk);
1627 nbd_exp->eject_notifier_blk = blk;
1628 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1629 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1632 static const BlockDevOps nbd_block_ops = {
1633 .drained_begin = nbd_drained_begin,
1634 .drained_end = nbd_drained_end,
1635 .drained_poll = nbd_drained_poll,
1638 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1639 Error **errp)
1641 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1642 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1643 const char *name = arg->name ?: exp_args->node_name;
1644 BlockBackend *blk = blk_exp->blk;
1645 int64_t size;
1646 uint64_t perm, shared_perm;
1647 bool readonly = !exp_args->writable;
1648 BlockDirtyBitmapOrStrList *bitmaps;
1649 size_t i;
1650 int ret;
1652 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1654 if (!nbd_server_is_running()) {
1655 error_setg(errp, "NBD server not running");
1656 return -EINVAL;
1659 if (strlen(name) > NBD_MAX_STRING_SIZE) {
1660 error_setg(errp, "export name '%s' too long", name);
1661 return -EINVAL;
1664 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1665 error_setg(errp, "description '%s' too long", arg->description);
1666 return -EINVAL;
1669 if (nbd_export_find(name)) {
1670 error_setg(errp, "NBD server already has export named '%s'", name);
1671 return -EEXIST;
1674 size = blk_getlength(blk);
1675 if (size < 0) {
1676 error_setg_errno(errp, -size,
1677 "Failed to determine the NBD export's length");
1678 return size;
1681 /* Don't allow resize while the NBD server is running, otherwise we don't
1682 * care what happens with the node. */
1683 blk_get_perm(blk, &perm, &shared_perm);
1684 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1685 if (ret < 0) {
1686 return ret;
1689 QTAILQ_INIT(&exp->clients);
1690 exp->name = g_strdup(name);
1691 exp->description = g_strdup(arg->description);
1692 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1693 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1695 if (nbd_server_max_connections() != 1) {
1696 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1698 if (readonly) {
1699 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1700 } else {
1701 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1702 NBD_FLAG_SEND_FAST_ZERO);
1704 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1706 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1707 exp->nr_export_bitmaps++;
1709 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1710 for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1711 i++, bitmaps = bitmaps->next)
1713 const char *bitmap;
1714 BlockDriverState *bs = blk_bs(blk);
1715 BdrvDirtyBitmap *bm = NULL;
1717 switch (bitmaps->value->type) {
1718 case QTYPE_QSTRING:
1719 bitmap = bitmaps->value->u.local;
1720 while (bs) {
1721 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1722 if (bm != NULL) {
1723 break;
1726 bs = bdrv_filter_or_cow_bs(bs);
1729 if (bm == NULL) {
1730 ret = -ENOENT;
1731 error_setg(errp, "Bitmap '%s' is not found",
1732 bitmaps->value->u.local);
1733 goto fail;
1736 if (readonly && bdrv_is_writable(bs) &&
1737 bdrv_dirty_bitmap_enabled(bm)) {
1738 ret = -EINVAL;
1739 error_setg(errp, "Enabled bitmap '%s' incompatible with "
1740 "readonly export", bitmap);
1741 goto fail;
1743 break;
1744 case QTYPE_QDICT:
1745 bitmap = bitmaps->value->u.external.name;
1746 bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node,
1747 bitmap, NULL, errp);
1748 if (!bm) {
1749 ret = -ENOENT;
1750 goto fail;
1752 break;
1753 default:
1754 abort();
1757 assert(bm);
1759 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1760 ret = -EINVAL;
1761 goto fail;
1764 exp->export_bitmaps[i] = bm;
1765 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1768 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1769 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1770 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1773 exp->allocation_depth = arg->allocation_depth;
1776 * We need to inhibit request queuing in the block layer to ensure we can
1777 * be properly quiesced when entering a drained section, as our coroutines
1778 * servicing pending requests might enter blk_pread().
1780 blk_set_disable_request_queuing(blk, true);
1782 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1784 blk_set_dev_ops(blk, &nbd_block_ops, exp);
1786 QTAILQ_INSERT_TAIL(&exports, exp, next);
1788 return 0;
1790 fail:
1791 g_free(exp->export_bitmaps);
1792 g_free(exp->name);
1793 g_free(exp->description);
1794 return ret;
1797 NBDExport *nbd_export_find(const char *name)
1799 NBDExport *exp;
1800 QTAILQ_FOREACH(exp, &exports, next) {
1801 if (strcmp(name, exp->name) == 0) {
1802 return exp;
1806 return NULL;
1809 AioContext *
1810 nbd_export_aio_context(NBDExport *exp)
1812 return exp->common.ctx;
1815 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1817 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1818 NBDClient *client, *next;
1820 blk_exp_ref(&exp->common);
1822 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1823 * close mode that stops advertising the export to new clients but
1824 * still permits existing clients to run to completion? Because of
1825 * that possibility, nbd_export_close() can be called more than
1826 * once on an export.
1828 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1829 client_close(client, true);
1831 if (exp->name) {
1832 g_free(exp->name);
1833 exp->name = NULL;
1834 QTAILQ_REMOVE(&exports, exp, next);
1836 blk_exp_unref(&exp->common);
1839 static void nbd_export_delete(BlockExport *blk_exp)
1841 size_t i;
1842 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1844 assert(exp->name == NULL);
1845 assert(QTAILQ_EMPTY(&exp->clients));
1847 g_free(exp->description);
1848 exp->description = NULL;
1850 if (exp->common.blk) {
1851 if (exp->eject_notifier_blk) {
1852 notifier_remove(&exp->eject_notifier);
1853 blk_unref(exp->eject_notifier_blk);
1855 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1856 blk_aio_detach, exp);
1857 blk_set_disable_request_queuing(exp->common.blk, false);
1860 for (i = 0; i < exp->nr_export_bitmaps; i++) {
1861 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1865 const BlockExportDriver blk_exp_nbd = {
1866 .type = BLOCK_EXPORT_TYPE_NBD,
1867 .instance_size = sizeof(NBDExport),
1868 .create = nbd_export_create,
1869 .delete = nbd_export_delete,
1870 .request_shutdown = nbd_export_request_shutdown,
1873 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1874 unsigned niov, Error **errp)
1876 int ret;
1878 g_assert(qemu_in_coroutine());
1879 qemu_co_mutex_lock(&client->send_lock);
1880 client->send_coroutine = qemu_coroutine_self();
1882 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1884 client->send_coroutine = NULL;
1885 qemu_co_mutex_unlock(&client->send_lock);
1887 return ret;
1890 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1891 uint64_t handle)
1893 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1894 stl_be_p(&reply->error, error);
1895 stq_be_p(&reply->handle, handle);
1898 static int nbd_co_send_simple_reply(NBDClient *client,
1899 uint64_t handle,
1900 uint32_t error,
1901 void *data,
1902 size_t len,
1903 Error **errp)
1905 NBDSimpleReply reply;
1906 int nbd_err = system_errno_to_nbd_errno(error);
1907 struct iovec iov[] = {
1908 {.iov_base = &reply, .iov_len = sizeof(reply)},
1909 {.iov_base = data, .iov_len = len}
1912 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1913 len);
1914 set_be_simple_reply(&reply, nbd_err, handle);
1916 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1919 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1920 uint16_t type, uint64_t handle, uint32_t length)
1922 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1923 stw_be_p(&chunk->flags, flags);
1924 stw_be_p(&chunk->type, type);
1925 stq_be_p(&chunk->handle, handle);
1926 stl_be_p(&chunk->length, length);
1929 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1930 uint64_t handle,
1931 Error **errp)
1933 NBDStructuredReplyChunk chunk;
1934 struct iovec iov[] = {
1935 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1938 trace_nbd_co_send_structured_done(handle);
1939 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1941 return nbd_co_send_iov(client, iov, 1, errp);
1944 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1945 uint64_t handle,
1946 uint64_t offset,
1947 void *data,
1948 size_t size,
1949 bool final,
1950 Error **errp)
1952 NBDStructuredReadData chunk;
1953 struct iovec iov[] = {
1954 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1955 {.iov_base = data, .iov_len = size}
1958 assert(size);
1959 trace_nbd_co_send_structured_read(handle, offset, data, size);
1960 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1961 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1962 sizeof(chunk) - sizeof(chunk.h) + size);
1963 stq_be_p(&chunk.offset, offset);
1965 return nbd_co_send_iov(client, iov, 2, errp);
1968 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1969 uint64_t handle,
1970 uint32_t error,
1971 const char *msg,
1972 Error **errp)
1974 NBDStructuredError chunk;
1975 int nbd_err = system_errno_to_nbd_errno(error);
1976 struct iovec iov[] = {
1977 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1978 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1981 assert(nbd_err);
1982 trace_nbd_co_send_structured_error(handle, nbd_err,
1983 nbd_err_lookup(nbd_err), msg ? msg : "");
1984 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1985 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1986 stl_be_p(&chunk.error, nbd_err);
1987 stw_be_p(&chunk.message_length, iov[1].iov_len);
1989 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1992 /* Do a sparse read and send the structured reply to the client.
1993 * Returns -errno if sending fails. blk_co_block_status_above() failure is
1994 * reported to the client, at which point this function succeeds.
1996 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1997 uint64_t handle,
1998 uint64_t offset,
1999 uint8_t *data,
2000 size_t size,
2001 Error **errp)
2003 int ret = 0;
2004 NBDExport *exp = client->exp;
2005 size_t progress = 0;
2007 while (progress < size) {
2008 int64_t pnum;
2009 int status = blk_co_block_status_above(exp->common.blk, NULL,
2010 offset + progress,
2011 size - progress, &pnum, NULL,
2012 NULL);
2013 bool final;
2015 if (status < 0) {
2016 char *msg = g_strdup_printf("unable to check for holes: %s",
2017 strerror(-status));
2019 ret = nbd_co_send_structured_error(client, handle, -status, msg,
2020 errp);
2021 g_free(msg);
2022 return ret;
2024 assert(pnum && pnum <= size - progress);
2025 final = progress + pnum == size;
2026 if (status & BDRV_BLOCK_ZERO) {
2027 NBDStructuredReadHole chunk;
2028 struct iovec iov[] = {
2029 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2032 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
2033 pnum);
2034 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
2035 NBD_REPLY_TYPE_OFFSET_HOLE,
2036 handle, sizeof(chunk) - sizeof(chunk.h));
2037 stq_be_p(&chunk.offset, offset + progress);
2038 stl_be_p(&chunk.length, pnum);
2039 ret = nbd_co_send_iov(client, iov, 1, errp);
2040 } else {
2041 ret = blk_pread(exp->common.blk, offset + progress, pnum,
2042 data + progress, 0);
2043 if (ret < 0) {
2044 error_setg_errno(errp, -ret, "reading from file failed");
2045 break;
2047 ret = nbd_co_send_structured_read(client, handle, offset + progress,
2048 data + progress, pnum, final,
2049 errp);
2052 if (ret < 0) {
2053 break;
2055 progress += pnum;
2057 return ret;
2060 typedef struct NBDExtentArray {
2061 NBDExtent *extents;
2062 unsigned int nb_alloc;
2063 unsigned int count;
2064 uint64_t total_length;
2065 bool can_add;
2066 bool converted_to_be;
2067 } NBDExtentArray;
2069 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
2071 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2073 ea->nb_alloc = nb_alloc;
2074 ea->extents = g_new(NBDExtent, nb_alloc);
2075 ea->can_add = true;
2077 return ea;
2080 static void nbd_extent_array_free(NBDExtentArray *ea)
2082 g_free(ea->extents);
2083 g_free(ea);
2085 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free)
2087 /* Further modifications of the array after conversion are abandoned */
2088 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2090 int i;
2092 assert(!ea->converted_to_be);
2093 ea->can_add = false;
2094 ea->converted_to_be = true;
2096 for (i = 0; i < ea->count; i++) {
2097 ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2098 ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
2103 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2104 * return -1.
2105 * For safety, when returning -1 for the first time, .can_add is set to false,
2106 * and further calls to nbd_extent_array_add() will crash.
2107 * (this avoids the situation where a caller ignores failure to add one extent,
2108 * where adding another extent that would squash into the last array entry
2109 * would result in an incorrect range reported to the client)
2111 static int nbd_extent_array_add(NBDExtentArray *ea,
2112 uint32_t length, uint32_t flags)
2114 assert(ea->can_add);
2116 if (!length) {
2117 return 0;
2120 /* Extend previous extent if flags are the same */
2121 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2122 uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
2124 if (sum <= UINT32_MAX) {
2125 ea->extents[ea->count - 1].length = sum;
2126 ea->total_length += length;
2127 return 0;
2131 if (ea->count >= ea->nb_alloc) {
2132 ea->can_add = false;
2133 return -1;
2136 ea->total_length += length;
2137 ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
2138 ea->count++;
2140 return 0;
2143 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk,
2144 uint64_t offset, uint64_t bytes,
2145 NBDExtentArray *ea)
2147 while (bytes) {
2148 uint32_t flags;
2149 int64_t num;
2150 int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num,
2151 NULL, NULL);
2153 if (ret < 0) {
2154 return ret;
2157 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2158 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2160 if (nbd_extent_array_add(ea, num, flags) < 0) {
2161 return 0;
2164 offset += num;
2165 bytes -= num;
2168 return 0;
2171 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk,
2172 uint64_t offset, uint64_t bytes,
2173 NBDExtentArray *ea)
2175 while (bytes) {
2176 int64_t num;
2177 int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes,
2178 &num);
2180 if (ret < 0) {
2181 return ret;
2184 if (nbd_extent_array_add(ea, num, ret) < 0) {
2185 return 0;
2188 offset += num;
2189 bytes -= num;
2192 return 0;
2196 * nbd_co_send_extents
2198 * @ea is converted to BE by the function
2199 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2201 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
2202 NBDExtentArray *ea,
2203 bool last, uint32_t context_id, Error **errp)
2205 NBDStructuredMeta chunk;
2206 struct iovec iov[] = {
2207 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2208 {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
2211 nbd_extent_array_convert_to_be(ea);
2213 trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
2214 last);
2215 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
2216 NBD_REPLY_TYPE_BLOCK_STATUS,
2217 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
2218 stl_be_p(&chunk.context_id, context_id);
2220 return nbd_co_send_iov(client, iov, 2, errp);
2223 /* Get block status from the exported device and send it to the client */
2224 static int
2225 coroutine_fn nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2226 BlockBackend *blk, uint64_t offset,
2227 uint32_t length, bool dont_fragment,
2228 bool last, uint32_t context_id,
2229 Error **errp)
2231 int ret;
2232 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2233 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2235 if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2236 ret = blockstatus_to_extents(blk, offset, length, ea);
2237 } else {
2238 ret = blockalloc_to_extents(blk, offset, length, ea);
2240 if (ret < 0) {
2241 return nbd_co_send_structured_error(
2242 client, handle, -ret, "can't get block status", errp);
2245 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2248 /* Populate @ea from a dirty bitmap. */
2249 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2250 uint64_t offset, uint64_t length,
2251 NBDExtentArray *es)
2253 int64_t start, dirty_start, dirty_count;
2254 int64_t end = offset + length;
2255 bool full = false;
2257 bdrv_dirty_bitmap_lock(bitmap);
2259 for (start = offset;
2260 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2261 &dirty_start, &dirty_count);
2262 start = dirty_start + dirty_count)
2264 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2265 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2267 full = true;
2268 break;
2272 if (!full) {
2273 /* last non dirty extent, nothing to do if array is now full */
2274 (void) nbd_extent_array_add(es, end - start, 0);
2277 bdrv_dirty_bitmap_unlock(bitmap);
2280 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2281 BdrvDirtyBitmap *bitmap, uint64_t offset,
2282 uint32_t length, bool dont_fragment, bool last,
2283 uint32_t context_id, Error **errp)
2285 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2286 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2288 bitmap_to_extents(bitmap, offset, length, ea);
2290 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2293 /* nbd_co_receive_request
2294 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2295 * connection right away, -EAGAIN to indicate we were interrupted and the
2296 * channel should be quiesced, and any other negative value to report an error
2297 * to the client (although the caller may still need to disconnect after
2298 * reporting the error).
2300 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2301 Error **errp)
2303 NBDClient *client = req->client;
2304 int valid_flags;
2305 int ret;
2307 g_assert(qemu_in_coroutine());
2308 assert(client->recv_coroutine == qemu_coroutine_self());
2309 ret = nbd_receive_request(client, request, errp);
2310 if (ret < 0) {
2311 return ret;
2314 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2315 nbd_cmd_lookup(request->type));
2317 if (request->type != NBD_CMD_WRITE) {
2318 /* No payload, we are ready to read the next request. */
2319 req->complete = true;
2322 if (request->type == NBD_CMD_DISC) {
2323 /* Special case: we're going to disconnect without a reply,
2324 * whether or not flags, from, or len are bogus */
2325 return -EIO;
2328 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2329 request->type == NBD_CMD_CACHE)
2331 if (request->len > NBD_MAX_BUFFER_SIZE) {
2332 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2333 request->len, NBD_MAX_BUFFER_SIZE);
2334 return -EINVAL;
2337 if (request->type != NBD_CMD_CACHE) {
2338 req->data = blk_try_blockalign(client->exp->common.blk,
2339 request->len);
2340 if (req->data == NULL) {
2341 error_setg(errp, "No memory");
2342 return -ENOMEM;
2347 if (request->type == NBD_CMD_WRITE) {
2348 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2349 errp) < 0)
2351 return -EIO;
2353 req->complete = true;
2355 trace_nbd_co_receive_request_payload_received(request->handle,
2356 request->len);
2359 /* Sanity checks. */
2360 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2361 (request->type == NBD_CMD_WRITE ||
2362 request->type == NBD_CMD_WRITE_ZEROES ||
2363 request->type == NBD_CMD_TRIM)) {
2364 error_setg(errp, "Export is read-only");
2365 return -EROFS;
2367 if (request->from > client->exp->size ||
2368 request->len > client->exp->size - request->from) {
2369 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2370 ", Size: %" PRIu64, request->from, request->len,
2371 client->exp->size);
2372 return (request->type == NBD_CMD_WRITE ||
2373 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2375 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2376 client->check_align)) {
2378 * The block layer gracefully handles unaligned requests, but
2379 * it's still worth tracing client non-compliance
2381 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2382 request->from,
2383 request->len,
2384 client->check_align);
2386 valid_flags = NBD_CMD_FLAG_FUA;
2387 if (request->type == NBD_CMD_READ && client->structured_reply) {
2388 valid_flags |= NBD_CMD_FLAG_DF;
2389 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2390 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2391 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2392 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2394 if (request->flags & ~valid_flags) {
2395 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2396 nbd_cmd_lookup(request->type), request->flags);
2397 return -EINVAL;
2400 return 0;
2403 /* Send simple reply without a payload, or a structured error
2404 * @error_msg is ignored if @ret >= 0
2405 * Returns 0 if connection is still live, -errno on failure to talk to client
2407 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2408 uint64_t handle,
2409 int ret,
2410 const char *error_msg,
2411 Error **errp)
2413 if (client->structured_reply && ret < 0) {
2414 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2415 errp);
2416 } else {
2417 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2418 NULL, 0, errp);
2422 /* Handle NBD_CMD_READ request.
2423 * Return -errno if sending fails. Other errors are reported directly to the
2424 * client as an error reply. */
2425 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2426 uint8_t *data, Error **errp)
2428 int ret;
2429 NBDExport *exp = client->exp;
2431 assert(request->type == NBD_CMD_READ);
2433 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2434 if (request->flags & NBD_CMD_FLAG_FUA) {
2435 ret = blk_co_flush(exp->common.blk);
2436 if (ret < 0) {
2437 return nbd_send_generic_reply(client, request->handle, ret,
2438 "flush failed", errp);
2442 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2443 request->len)
2445 return nbd_co_send_sparse_read(client, request->handle, request->from,
2446 data, request->len, errp);
2449 ret = blk_pread(exp->common.blk, request->from, request->len, data, 0);
2450 if (ret < 0) {
2451 return nbd_send_generic_reply(client, request->handle, ret,
2452 "reading from file failed", errp);
2455 if (client->structured_reply) {
2456 if (request->len) {
2457 return nbd_co_send_structured_read(client, request->handle,
2458 request->from, data,
2459 request->len, true, errp);
2460 } else {
2461 return nbd_co_send_structured_done(client, request->handle, errp);
2463 } else {
2464 return nbd_co_send_simple_reply(client, request->handle, 0,
2465 data, request->len, errp);
2470 * nbd_do_cmd_cache
2472 * Handle NBD_CMD_CACHE request.
2473 * Return -errno if sending fails. Other errors are reported directly to the
2474 * client as an error reply.
2476 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2477 Error **errp)
2479 int ret;
2480 NBDExport *exp = client->exp;
2482 assert(request->type == NBD_CMD_CACHE);
2484 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2485 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2487 return nbd_send_generic_reply(client, request->handle, ret,
2488 "caching data failed", errp);
2491 /* Handle NBD request.
2492 * Return -errno if sending fails. Other errors are reported directly to the
2493 * client as an error reply. */
2494 static coroutine_fn int nbd_handle_request(NBDClient *client,
2495 NBDRequest *request,
2496 uint8_t *data, Error **errp)
2498 int ret;
2499 int flags;
2500 NBDExport *exp = client->exp;
2501 char *msg;
2502 size_t i;
2504 switch (request->type) {
2505 case NBD_CMD_CACHE:
2506 return nbd_do_cmd_cache(client, request, errp);
2508 case NBD_CMD_READ:
2509 return nbd_do_cmd_read(client, request, data, errp);
2511 case NBD_CMD_WRITE:
2512 flags = 0;
2513 if (request->flags & NBD_CMD_FLAG_FUA) {
2514 flags |= BDRV_REQ_FUA;
2516 ret = blk_pwrite(exp->common.blk, request->from, request->len, data,
2517 flags);
2518 return nbd_send_generic_reply(client, request->handle, ret,
2519 "writing to file failed", errp);
2521 case NBD_CMD_WRITE_ZEROES:
2522 flags = 0;
2523 if (request->flags & NBD_CMD_FLAG_FUA) {
2524 flags |= BDRV_REQ_FUA;
2526 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2527 flags |= BDRV_REQ_MAY_UNMAP;
2529 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2530 flags |= BDRV_REQ_NO_FALLBACK;
2532 ret = blk_pwrite_zeroes(exp->common.blk, request->from, request->len,
2533 flags);
2534 return nbd_send_generic_reply(client, request->handle, ret,
2535 "writing to file failed", errp);
2537 case NBD_CMD_DISC:
2538 /* unreachable, thanks to special case in nbd_co_receive_request() */
2539 abort();
2541 case NBD_CMD_FLUSH:
2542 ret = blk_co_flush(exp->common.blk);
2543 return nbd_send_generic_reply(client, request->handle, ret,
2544 "flush failed", errp);
2546 case NBD_CMD_TRIM:
2547 ret = blk_co_pdiscard(exp->common.blk, request->from, request->len);
2548 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2549 ret = blk_co_flush(exp->common.blk);
2551 return nbd_send_generic_reply(client, request->handle, ret,
2552 "discard failed", errp);
2554 case NBD_CMD_BLOCK_STATUS:
2555 if (!request->len) {
2556 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2557 "need non-zero length", errp);
2559 if (client->export_meta.count) {
2560 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2561 int contexts_remaining = client->export_meta.count;
2563 if (client->export_meta.base_allocation) {
2564 ret = nbd_co_send_block_status(client, request->handle,
2565 exp->common.blk,
2566 request->from,
2567 request->len, dont_fragment,
2568 !--contexts_remaining,
2569 NBD_META_ID_BASE_ALLOCATION,
2570 errp);
2571 if (ret < 0) {
2572 return ret;
2576 if (client->export_meta.allocation_depth) {
2577 ret = nbd_co_send_block_status(client, request->handle,
2578 exp->common.blk,
2579 request->from, request->len,
2580 dont_fragment,
2581 !--contexts_remaining,
2582 NBD_META_ID_ALLOCATION_DEPTH,
2583 errp);
2584 if (ret < 0) {
2585 return ret;
2589 for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2590 if (!client->export_meta.bitmaps[i]) {
2591 continue;
2593 ret = nbd_co_send_bitmap(client, request->handle,
2594 client->exp->export_bitmaps[i],
2595 request->from, request->len,
2596 dont_fragment, !--contexts_remaining,
2597 NBD_META_ID_DIRTY_BITMAP + i, errp);
2598 if (ret < 0) {
2599 return ret;
2603 assert(!contexts_remaining);
2605 return 0;
2606 } else {
2607 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2608 "CMD_BLOCK_STATUS not negotiated",
2609 errp);
2612 default:
2613 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2614 request->type);
2615 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2616 errp);
2617 g_free(msg);
2618 return ret;
2622 /* Owns a reference to the NBDClient passed as opaque. */
2623 static coroutine_fn void nbd_trip(void *opaque)
2625 NBDClient *client = opaque;
2626 NBDRequestData *req;
2627 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2628 int ret;
2629 Error *local_err = NULL;
2631 trace_nbd_trip();
2632 if (client->closing) {
2633 nbd_client_put(client);
2634 return;
2637 if (client->quiescing) {
2639 * We're switching between AIO contexts. Don't attempt to receive a new
2640 * request and kick the main context which may be waiting for us.
2642 nbd_client_put(client);
2643 client->recv_coroutine = NULL;
2644 aio_wait_kick();
2645 return;
2648 req = nbd_request_get(client);
2649 ret = nbd_co_receive_request(req, &request, &local_err);
2650 client->recv_coroutine = NULL;
2652 if (client->closing) {
2654 * The client may be closed when we are blocked in
2655 * nbd_co_receive_request()
2657 goto done;
2660 if (ret == -EAGAIN) {
2661 assert(client->quiescing);
2662 goto done;
2665 nbd_client_receive_next_request(client);
2666 if (ret == -EIO) {
2667 goto disconnect;
2670 if (ret < 0) {
2671 /* It wasn't -EIO, so, according to nbd_co_receive_request()
2672 * semantics, we should return the error to the client. */
2673 Error *export_err = local_err;
2675 local_err = NULL;
2676 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2677 error_get_pretty(export_err), &local_err);
2678 error_free(export_err);
2679 } else {
2680 ret = nbd_handle_request(client, &request, req->data, &local_err);
2682 if (ret < 0) {
2683 error_prepend(&local_err, "Failed to send reply: ");
2684 goto disconnect;
2687 /* We must disconnect after NBD_CMD_WRITE if we did not
2688 * read the payload.
2690 if (!req->complete) {
2691 error_setg(&local_err, "Request handling failed in intermediate state");
2692 goto disconnect;
2695 done:
2696 nbd_request_put(req);
2697 nbd_client_put(client);
2698 return;
2700 disconnect:
2701 if (local_err) {
2702 error_reportf_err(local_err, "Disconnect client, due to: ");
2704 nbd_request_put(req);
2705 client_close(client, true);
2706 nbd_client_put(client);
2709 static void nbd_client_receive_next_request(NBDClient *client)
2711 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
2712 !client->quiescing) {
2713 nbd_client_get(client);
2714 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2715 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
2719 static coroutine_fn void nbd_co_client_start(void *opaque)
2721 NBDClient *client = opaque;
2722 Error *local_err = NULL;
2724 qemu_co_mutex_init(&client->send_lock);
2726 if (nbd_negotiate(client, &local_err)) {
2727 if (local_err) {
2728 error_report_err(local_err);
2730 client_close(client, false);
2731 return;
2734 nbd_client_receive_next_request(client);
2738 * Create a new client listener using the given channel @sioc.
2739 * Begin servicing it in a coroutine. When the connection closes, call
2740 * @close_fn with an indication of whether the client completed negotiation.
2742 void nbd_client_new(QIOChannelSocket *sioc,
2743 QCryptoTLSCreds *tlscreds,
2744 const char *tlsauthz,
2745 void (*close_fn)(NBDClient *, bool))
2747 NBDClient *client;
2748 Coroutine *co;
2750 client = g_new0(NBDClient, 1);
2751 client->refcount = 1;
2752 client->tlscreds = tlscreds;
2753 if (tlscreds) {
2754 object_ref(OBJECT(client->tlscreds));
2756 client->tlsauthz = g_strdup(tlsauthz);
2757 client->sioc = sioc;
2758 object_ref(OBJECT(client->sioc));
2759 client->ioc = QIO_CHANNEL(sioc);
2760 object_ref(OBJECT(client->ioc));
2761 client->close_fn = close_fn;
2763 co = qemu_coroutine_create(nbd_co_client_start, client);
2764 qemu_coroutine_enter(co);