2 * Copyright (C) 2016-2020 Red Hat, Inc.
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/export.h"
23 #include "qapi/error.h"
24 #include "qemu/queue.h"
26 #include "nbd-internal.h"
27 #include "qemu/units.h"
29 #define NBD_META_ID_BASE_ALLOCATION 0
30 #define NBD_META_ID_DIRTY_BITMAP 1
33 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
34 * constant. If an increase is needed, note that the NBD protocol
35 * recommends no larger than 32 mb, so that the client won't consider
36 * the reply as a denial of service attack.
38 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
40 static int system_errno_to_nbd_errno(int err)
61 #if ENOTSUP != EOPNOTSUPP
73 /* Definitions for opaque data types */
75 typedef struct NBDRequestData NBDRequestData;
77 struct NBDRequestData {
78 QSIMPLEQ_ENTRY(NBDRequestData) entry;
91 QTAILQ_HEAD(, NBDClient) clients;
92 QTAILQ_ENTRY(NBDExport) next;
94 BlockBackend *eject_notifier_blk;
95 Notifier eject_notifier;
97 BdrvDirtyBitmap *export_bitmap;
98 char *export_bitmap_context;
101 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
103 /* NBDExportMetaContexts represents a list of contexts to be exported,
104 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
105 * NBD_OPT_LIST_META_CONTEXT. */
106 typedef struct NBDExportMetaContexts {
108 bool valid; /* means that negotiation of the option finished without
110 bool base_allocation; /* export base:allocation context (block status) */
111 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
112 } NBDExportMetaContexts;
116 void (*close_fn)(NBDClient *client, bool negotiated);
119 QCryptoTLSCreds *tlscreds;
121 QIOChannelSocket *sioc; /* The underlying data channel */
122 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
124 Coroutine *recv_coroutine;
127 Coroutine *send_coroutine;
129 QTAILQ_ENTRY(NBDClient) next;
133 uint32_t check_align; /* If non-zero, check for aligned client requests */
135 bool structured_reply;
136 NBDExportMetaContexts export_meta;
138 uint32_t opt; /* Current option being negotiated */
139 uint32_t optlen; /* remaining length of data in ioc for the option being
143 static void nbd_client_receive_next_request(NBDClient *client);
145 /* Basic flow for negotiation
172 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
173 uint32_t type, uint32_t length)
175 stq_be_p(&rep->magic, NBD_REP_MAGIC);
176 stl_be_p(&rep->option, option);
177 stl_be_p(&rep->type, type);
178 stl_be_p(&rep->length, length);
181 /* Send a reply header, including length, but no payload.
182 * Return -errno on error, 0 on success. */
183 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
184 uint32_t len, Error **errp)
188 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
189 type, nbd_rep_lookup(type), len);
191 assert(len < NBD_MAX_BUFFER_SIZE);
193 set_be_option_rep(&rep, client->opt, type, len);
194 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
197 /* Send a reply header with default 0 length.
198 * Return -errno on error, 0 on success. */
199 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
202 return nbd_negotiate_send_rep_len(client, type, 0, errp);
205 /* Send an error reply.
206 * Return -errno on error, 0 on success. */
207 static int GCC_FMT_ATTR(4, 0)
208 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
209 Error **errp, const char *fmt, va_list va)
212 g_autofree char *msg = NULL;
216 msg = g_strdup_vprintf(fmt, va);
218 assert(len < NBD_MAX_STRING_SIZE);
219 trace_nbd_negotiate_send_rep_err(msg);
220 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
224 if (nbd_write(client->ioc, msg, len, errp) < 0) {
225 error_prepend(errp, "write failed (error message): ");
233 * Return a malloc'd copy of @name suitable for use in an error reply.
236 nbd_sanitize_name(const char *name)
238 if (strnlen(name, 80) < 80) {
239 return g_strdup(name);
241 /* XXX Should we also try to sanitize any control characters? */
242 return g_strdup_printf("%.80s...", name);
245 /* Send an error reply.
246 * Return -errno on error, 0 on success. */
247 static int GCC_FMT_ATTR(4, 5)
248 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
249 Error **errp, const char *fmt, ...)
255 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
260 /* Drop remainder of the current option, and send a reply with the
261 * given error type and message. Return -errno on read or write
262 * failure; or 0 if connection is still live. */
263 static int GCC_FMT_ATTR(4, 0)
264 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
265 const char *fmt, va_list va)
267 int ret = nbd_drop(client->ioc, client->optlen, errp);
271 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
276 static int GCC_FMT_ATTR(4, 5)
277 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
278 const char *fmt, ...)
284 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
290 static int GCC_FMT_ATTR(3, 4)
291 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
297 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
303 /* Read size bytes from the unparsed payload of the current option.
304 * If @check_nul, require that no NUL bytes appear in buffer.
305 * Return -errno on I/O error, 0 if option was completely handled by
306 * sending a reply about inconsistent lengths, or 1 on success. */
307 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
308 bool check_nul, Error **errp)
310 if (size > client->optlen) {
311 return nbd_opt_invalid(client, errp,
312 "Inconsistent lengths in option %s",
313 nbd_opt_lookup(client->opt));
315 client->optlen -= size;
316 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
320 if (check_nul && strnlen(buffer, size) != size) {
321 return nbd_opt_invalid(client, errp,
322 "Unexpected embedded NUL in option %s",
323 nbd_opt_lookup(client->opt));
328 /* Drop size bytes from the unparsed payload of the current option.
329 * Return -errno on I/O error, 0 if option was completely handled by
330 * sending a reply about inconsistent lengths, or 1 on success. */
331 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
333 if (size > client->optlen) {
334 return nbd_opt_invalid(client, errp,
335 "Inconsistent lengths in option %s",
336 nbd_opt_lookup(client->opt));
338 client->optlen -= size;
339 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
344 * Read a string with the format:
345 * uint32_t len (<= NBD_MAX_STRING_SIZE)
346 * len bytes string (not 0-terminated)
348 * On success, @name will be allocated.
349 * If @length is non-null, it will be set to the actual string length.
351 * Return -errno on I/O error, 0 if option was completely handled by
352 * sending a reply about inconsistent lengths, or 1 on success.
354 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
359 g_autofree char *local_name = NULL;
362 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
366 len = cpu_to_be32(len);
368 if (len > NBD_MAX_STRING_SIZE) {
369 return nbd_opt_invalid(client, errp,
370 "Invalid name length: %" PRIu32, len);
373 local_name = g_malloc(len + 1);
374 ret = nbd_opt_read(client, local_name, len, true, errp);
378 local_name[len] = '\0';
383 *name = g_steal_pointer(&local_name);
388 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
389 * Return -errno on error, 0 on success. */
390 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
394 size_t name_len, desc_len;
396 const char *name = exp->name ? exp->name : "";
397 const char *desc = exp->description ? exp->description : "";
398 QIOChannel *ioc = client->ioc;
401 trace_nbd_negotiate_send_rep_list(name, desc);
402 name_len = strlen(name);
403 desc_len = strlen(desc);
404 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
405 len = name_len + desc_len + sizeof(len);
406 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
411 len = cpu_to_be32(name_len);
412 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
413 error_prepend(errp, "write failed (name length): ");
417 if (nbd_write(ioc, name, name_len, errp) < 0) {
418 error_prepend(errp, "write failed (name buffer): ");
422 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
423 error_prepend(errp, "write failed (description buffer): ");
430 /* Process the NBD_OPT_LIST command, with a potential series of replies.
431 * Return -errno on error, 0 on success. */
432 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
435 assert(client->opt == NBD_OPT_LIST);
437 /* For each export, send a NBD_REP_SERVER reply. */
438 QTAILQ_FOREACH(exp, &exports, next) {
439 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
443 /* Finish with a NBD_REP_ACK. */
444 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
447 static void nbd_check_meta_export(NBDClient *client)
449 client->export_meta.valid &= client->exp == client->export_meta.exp;
452 /* Send a reply to NBD_OPT_EXPORT_NAME.
453 * Return -errno on error, 0 on success. */
454 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
458 g_autofree char *name = NULL;
459 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
465 [20 .. xx] export name (length bytes)
468 [ 8 .. 9] export flags
469 [10 .. 133] reserved (0) [unless no_zeroes]
471 trace_nbd_negotiate_handle_export_name();
472 if (client->optlen > NBD_MAX_STRING_SIZE) {
473 error_setg(errp, "Bad length received");
476 name = g_malloc(client->optlen + 1);
477 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
480 name[client->optlen] = '\0';
483 trace_nbd_negotiate_handle_export_name_request(name);
485 client->exp = nbd_export_find(name);
487 error_setg(errp, "export not found");
491 myflags = client->exp->nbdflags;
492 if (client->structured_reply) {
493 myflags |= NBD_FLAG_SEND_DF;
495 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
496 stq_be_p(buf, client->exp->size);
497 stw_be_p(buf + 8, myflags);
498 len = no_zeroes ? 10 : sizeof(buf);
499 ret = nbd_write(client->ioc, buf, len, errp);
501 error_prepend(errp, "write failed: ");
505 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
506 blk_exp_ref(&client->exp->common);
507 nbd_check_meta_export(client);
512 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
513 * The buffer does NOT include the info type prefix.
514 * Return -errno on error, 0 if ready to send more. */
515 static int nbd_negotiate_send_info(NBDClient *client,
516 uint16_t info, uint32_t length, void *buf,
521 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
522 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
523 sizeof(info) + length, errp);
527 info = cpu_to_be16(info);
528 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
531 if (nbd_write(client->ioc, buf, length, errp) < 0) {
537 /* nbd_reject_length: Handle any unexpected payload.
538 * @fatal requests that we quit talking to the client, even if we are able
539 * to successfully send an error reply.
541 * -errno transmission error occurred or @fatal was requested, errp is set
542 * 0 error message successfully sent to client, errp is not set
544 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
548 assert(client->optlen);
549 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
550 nbd_opt_lookup(client->opt));
552 error_setg(errp, "option '%s' has unexpected length",
553 nbd_opt_lookup(client->opt));
559 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
560 * Return -errno on error, 0 if ready for next option, and 1 to move
561 * into transmission phase. */
562 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
565 g_autofree char *name = NULL;
569 uint32_t namelen = 0;
570 bool sendname = false;
571 bool blocksize = false;
573 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
574 uint32_t check_align = 0;
578 4 bytes: L, name length (can be 0)
580 2 bytes: N, number of requests (can be 0)
581 N * 2 bytes: N requests
583 rc = nbd_opt_read_name(client, &name, &namelen, errp);
587 trace_nbd_negotiate_handle_export_name_request(name);
589 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
593 requests = be16_to_cpu(requests);
594 trace_nbd_negotiate_handle_info_requests(requests);
596 rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
600 request = be16_to_cpu(request);
601 trace_nbd_negotiate_handle_info_request(request,
602 nbd_info_lookup(request));
603 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
604 * everything else is either a request we don't know or
605 * something we send regardless of request */
610 case NBD_INFO_BLOCK_SIZE:
615 if (client->optlen) {
616 return nbd_reject_length(client, false, errp);
619 exp = nbd_export_find(name);
621 g_autofree char *sane_name = nbd_sanitize_name(name);
623 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
624 errp, "export '%s' not present",
628 /* Don't bother sending NBD_INFO_NAME unless client requested it */
630 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
637 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
639 if (exp->description) {
640 size_t len = strlen(exp->description);
642 assert(len <= NBD_MAX_STRING_SIZE);
643 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
644 len, exp->description, errp);
650 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
651 * according to whether the client requested it, and according to
652 * whether this is OPT_INFO or OPT_GO. */
653 /* minimum - 1 for back-compat, or actual if client will obey it. */
654 if (client->opt == NBD_OPT_INFO || blocksize) {
655 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
659 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
660 /* preferred - Hard-code to 4096 for now.
661 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
662 sizes[1] = MAX(4096, sizes[0]);
663 /* maximum - At most 32M, but smaller as appropriate. */
664 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
665 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
666 sizes[0] = cpu_to_be32(sizes[0]);
667 sizes[1] = cpu_to_be32(sizes[1]);
668 sizes[2] = cpu_to_be32(sizes[2]);
669 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
670 sizeof(sizes), sizes, errp);
675 /* Send NBD_INFO_EXPORT always */
676 myflags = exp->nbdflags;
677 if (client->structured_reply) {
678 myflags |= NBD_FLAG_SEND_DF;
680 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
681 stq_be_p(buf, exp->size);
682 stw_be_p(buf + 8, myflags);
683 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
684 sizeof(buf), buf, errp);
690 * If the client is just asking for NBD_OPT_INFO, but forgot to
691 * request block sizes in a situation that would impact
692 * performance, then return an error. But for NBD_OPT_GO, we
693 * tolerate all clients, regardless of alignments.
695 if (client->opt == NBD_OPT_INFO && !blocksize &&
696 blk_get_request_alignment(exp->common.blk) > 1) {
697 return nbd_negotiate_send_rep_err(client,
698 NBD_REP_ERR_BLOCK_SIZE_REQD,
700 "request NBD_INFO_BLOCK_SIZE to "
705 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
710 if (client->opt == NBD_OPT_GO) {
712 client->check_align = check_align;
713 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
714 blk_exp_ref(&client->exp->common);
715 nbd_check_meta_export(client);
722 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
723 * new channel for all further (now-encrypted) communication. */
724 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
729 struct NBDTLSHandshakeData data = { 0 };
731 assert(client->opt == NBD_OPT_STARTTLS);
733 trace_nbd_negotiate_handle_starttls();
736 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
740 tioc = qio_channel_tls_new_server(ioc,
748 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
749 trace_nbd_negotiate_handle_starttls_handshake();
750 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
751 qio_channel_tls_handshake(tioc,
757 if (!data.complete) {
758 g_main_loop_run(data.loop);
760 g_main_loop_unref(data.loop);
762 object_unref(OBJECT(tioc));
763 error_propagate(errp, data.error);
767 return QIO_CHANNEL(tioc);
770 /* nbd_negotiate_send_meta_context
772 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
774 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
776 static int nbd_negotiate_send_meta_context(NBDClient *client,
781 NBDOptionReplyMetaContext opt;
782 struct iovec iov[] = {
783 {.iov_base = &opt, .iov_len = sizeof(opt)},
784 {.iov_base = (void *)context, .iov_len = strlen(context)}
787 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
788 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
792 trace_nbd_negotiate_meta_query_reply(context, context_id);
793 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
794 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
795 stl_be_p(&opt.context_id, context_id);
797 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
801 * Return true if @query matches @pattern, or if @query is empty when
802 * the @client is performing _LIST_.
804 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
808 trace_nbd_negotiate_meta_query_parse("empty");
809 return client->opt == NBD_OPT_LIST_META_CONTEXT;
811 if (strcmp(query, pattern) == 0) {
812 trace_nbd_negotiate_meta_query_parse(pattern);
815 trace_nbd_negotiate_meta_query_skip("pattern not matched");
820 * Return true and adjust @str in place if it begins with @prefix.
822 static bool nbd_strshift(const char **str, const char *prefix)
824 size_t len = strlen(prefix);
826 if (strncmp(*str, prefix, len) == 0) {
833 /* nbd_meta_base_query
835 * Handle queries to 'base' namespace. For now, only the base:allocation
836 * context is available. Return true if @query has been handled.
838 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
841 if (!nbd_strshift(&query, "base:")) {
844 trace_nbd_negotiate_meta_query_parse("base:");
846 if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
847 meta->base_allocation = true;
852 /* nbd_meta_qemu_query
854 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
855 * context is available. Return true if @query has been handled.
857 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
860 if (!nbd_strshift(&query, "qemu:")) {
863 trace_nbd_negotiate_meta_query_parse("qemu:");
866 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
867 meta->bitmap = !!meta->exp->export_bitmap;
869 trace_nbd_negotiate_meta_query_parse("empty");
873 if (nbd_strshift(&query, "dirty-bitmap:")) {
874 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
875 if (!meta->exp->export_bitmap) {
876 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
879 if (nbd_meta_empty_or_pattern(client,
880 meta->exp->export_bitmap_context +
881 strlen("qemu:dirty-bitmap:"), query)) {
887 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap");
891 /* nbd_negotiate_meta_query
893 * Parse namespace name and call corresponding function to parse body of the
896 * The only supported namespaces are 'base' and 'qemu'.
898 * Return -errno on I/O error, 0 if option was completely handled by
899 * sending a reply about inconsistent lengths, or 1 on success. */
900 static int nbd_negotiate_meta_query(NBDClient *client,
901 NBDExportMetaContexts *meta, Error **errp)
904 g_autofree char *query = NULL;
907 ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
911 len = cpu_to_be32(len);
913 if (len > NBD_MAX_STRING_SIZE) {
914 trace_nbd_negotiate_meta_query_skip("length too long");
915 return nbd_opt_skip(client, len, errp);
918 query = g_malloc(len + 1);
919 ret = nbd_opt_read(client, query, len, true, errp);
925 if (nbd_meta_base_query(client, meta, query)) {
928 if (nbd_meta_qemu_query(client, meta, query)) {
932 trace_nbd_negotiate_meta_query_skip("unknown namespace");
936 /* nbd_negotiate_meta_queries
937 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
939 * Return -errno on I/O error, or 0 if option was completely handled. */
940 static int nbd_negotiate_meta_queries(NBDClient *client,
941 NBDExportMetaContexts *meta, Error **errp)
944 g_autofree char *export_name = NULL;
945 NBDExportMetaContexts local_meta;
949 if (!client->structured_reply) {
950 return nbd_opt_invalid(client, errp,
951 "request option '%s' when structured reply "
953 nbd_opt_lookup(client->opt));
956 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
957 /* Only change the caller's meta on SET. */
961 memset(meta, 0, sizeof(*meta));
963 ret = nbd_opt_read_name(client, &export_name, NULL, errp);
968 meta->exp = nbd_export_find(export_name);
969 if (meta->exp == NULL) {
970 g_autofree char *sane_name = nbd_sanitize_name(export_name);
972 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
973 "export '%s' not present", sane_name);
976 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
980 nb_queries = cpu_to_be32(nb_queries);
981 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
982 export_name, nb_queries);
984 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
985 /* enable all known contexts */
986 meta->base_allocation = true;
987 meta->bitmap = !!meta->exp->export_bitmap;
989 for (i = 0; i < nb_queries; ++i) {
990 ret = nbd_negotiate_meta_query(client, meta, errp);
997 if (meta->base_allocation) {
998 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
999 NBD_META_ID_BASE_ALLOCATION,
1007 ret = nbd_negotiate_send_meta_context(client,
1008 meta->exp->export_bitmap_context,
1009 NBD_META_ID_DIRTY_BITMAP,
1016 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1024 /* nbd_negotiate_options
1025 * Process all NBD_OPT_* client option commands, during fixed newstyle
1028 * -errno on error, errp is set
1029 * 0 on successful negotiation, errp is not set
1030 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1033 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1036 bool fixedNewstyle = false;
1037 bool no_zeroes = false;
1040 [ 0 .. 3] client flags
1042 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1043 [ 0 .. 7] NBD_OPTS_MAGIC
1044 [ 8 .. 11] NBD option
1045 [12 .. 15] Data length
1048 [ 0 .. 7] NBD_OPTS_MAGIC
1049 [ 8 .. 11] Second NBD option
1050 [12 .. 15] Data length
1054 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1057 trace_nbd_negotiate_options_flags(flags);
1058 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1059 fixedNewstyle = true;
1060 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1062 if (flags & NBD_FLAG_C_NO_ZEROES) {
1064 flags &= ~NBD_FLAG_C_NO_ZEROES;
1067 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1073 uint32_t option, length;
1076 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1079 trace_nbd_negotiate_options_check_magic(magic);
1080 if (magic != NBD_OPTS_MAGIC) {
1081 error_setg(errp, "Bad magic received");
1085 if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1088 client->opt = option;
1090 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1093 assert(!client->optlen);
1094 client->optlen = length;
1096 if (length > NBD_MAX_BUFFER_SIZE) {
1097 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1098 length, NBD_MAX_BUFFER_SIZE);
1102 trace_nbd_negotiate_options_check_option(option,
1103 nbd_opt_lookup(option));
1104 if (client->tlscreds &&
1105 client->ioc == (QIOChannel *)client->sioc) {
1107 if (!fixedNewstyle) {
1108 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1112 case NBD_OPT_STARTTLS:
1114 /* Unconditionally drop the connection if the client
1115 * can't start a TLS negotiation correctly */
1116 return nbd_reject_length(client, true, errp);
1118 tioc = nbd_negotiate_handle_starttls(client, errp);
1123 object_unref(OBJECT(client->ioc));
1124 client->ioc = QIO_CHANNEL(tioc);
1127 case NBD_OPT_EXPORT_NAME:
1128 /* No way to return an error to client, so drop connection */
1129 error_setg(errp, "Option 0x%x not permitted before TLS",
1134 /* Let the client keep trying, unless they asked to
1135 * quit. Always try to give an error back to the
1136 * client; but when replying to OPT_ABORT, be aware
1137 * that the client may hang up before receiving the
1138 * error, in which case we are fine ignoring the
1139 * resulting EPIPE. */
1140 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1141 option == NBD_OPT_ABORT ? NULL : errp,
1143 " not permitted before TLS", option);
1144 if (option == NBD_OPT_ABORT) {
1149 } else if (fixedNewstyle) {
1153 ret = nbd_reject_length(client, false, errp);
1155 ret = nbd_negotiate_handle_list(client, errp);
1160 /* NBD spec says we must try to reply before
1161 * disconnecting, but that we must also tolerate
1162 * guests that don't wait for our reply. */
1163 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1166 case NBD_OPT_EXPORT_NAME:
1167 return nbd_negotiate_handle_export_name(client, no_zeroes,
1172 ret = nbd_negotiate_handle_info(client, errp);
1174 assert(option == NBD_OPT_GO);
1179 case NBD_OPT_STARTTLS:
1181 ret = nbd_reject_length(client, false, errp);
1182 } else if (client->tlscreds) {
1183 ret = nbd_negotiate_send_rep_err(client,
1184 NBD_REP_ERR_INVALID, errp,
1185 "TLS already enabled");
1187 ret = nbd_negotiate_send_rep_err(client,
1188 NBD_REP_ERR_POLICY, errp,
1189 "TLS not configured");
1193 case NBD_OPT_STRUCTURED_REPLY:
1195 ret = nbd_reject_length(client, false, errp);
1196 } else if (client->structured_reply) {
1197 ret = nbd_negotiate_send_rep_err(
1198 client, NBD_REP_ERR_INVALID, errp,
1199 "structured reply already negotiated");
1201 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1202 client->structured_reply = true;
1206 case NBD_OPT_LIST_META_CONTEXT:
1207 case NBD_OPT_SET_META_CONTEXT:
1208 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1213 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1214 "Unsupported option %" PRIu32 " (%s)",
1215 option, nbd_opt_lookup(option));
1220 * If broken new-style we should drop the connection
1221 * for anything except NBD_OPT_EXPORT_NAME
1224 case NBD_OPT_EXPORT_NAME:
1225 return nbd_negotiate_handle_export_name(client, no_zeroes,
1229 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1230 option, nbd_opt_lookup(option));
1242 * -errno on error, errp is set
1243 * 0 on successful negotiation, errp is not set
1244 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1247 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1250 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1253 /* Old style negotiation header, no room for options
1254 [ 0 .. 7] passwd ("NBDMAGIC")
1255 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1257 [24 .. 27] export flags (zero-extended)
1258 [28 .. 151] reserved (0)
1260 New style negotiation header, client can send options
1261 [ 0 .. 7] passwd ("NBDMAGIC")
1262 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1263 [16 .. 17] server flags (0)
1264 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1267 qio_channel_set_blocking(client->ioc, false, NULL);
1269 trace_nbd_negotiate_begin();
1270 memcpy(buf, "NBDMAGIC", 8);
1272 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1273 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1275 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1276 error_prepend(errp, "write failed: ");
1279 ret = nbd_negotiate_options(client, errp);
1282 error_prepend(errp, "option negotiation failed: ");
1287 /* Attach the channel to the same AioContext as the export */
1288 if (client->exp && client->exp->common.ctx) {
1289 qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
1292 assert(!client->optlen);
1293 trace_nbd_negotiate_success();
1298 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1301 uint8_t buf[NBD_REQUEST_SIZE];
1305 ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1311 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1312 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1313 [ 6 .. 7] type (NBD_CMD_READ, ...)
1319 magic = ldl_be_p(buf);
1320 request->flags = lduw_be_p(buf + 4);
1321 request->type = lduw_be_p(buf + 6);
1322 request->handle = ldq_be_p(buf + 8);
1323 request->from = ldq_be_p(buf + 16);
1324 request->len = ldl_be_p(buf + 24);
1326 trace_nbd_receive_request(magic, request->flags, request->type,
1327 request->from, request->len);
1329 if (magic != NBD_REQUEST_MAGIC) {
1330 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1336 #define MAX_NBD_REQUESTS 16
1338 void nbd_client_get(NBDClient *client)
1343 void nbd_client_put(NBDClient *client)
1345 if (--client->refcount == 0) {
1346 /* The last reference should be dropped by client->close,
1347 * which is called by client_close.
1349 assert(client->closing);
1351 qio_channel_detach_aio_context(client->ioc);
1352 object_unref(OBJECT(client->sioc));
1353 object_unref(OBJECT(client->ioc));
1354 if (client->tlscreds) {
1355 object_unref(OBJECT(client->tlscreds));
1357 g_free(client->tlsauthz);
1359 QTAILQ_REMOVE(&client->exp->clients, client, next);
1360 blk_exp_unref(&client->exp->common);
1366 static void client_close(NBDClient *client, bool negotiated)
1368 if (client->closing) {
1372 client->closing = true;
1374 /* Force requests to finish. They will drop their own references,
1375 * then we'll close the socket and free the NBDClient.
1377 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1380 /* Also tell the client, so that they release their reference. */
1381 if (client->close_fn) {
1382 client->close_fn(client, negotiated);
1386 static NBDRequestData *nbd_request_get(NBDClient *client)
1388 NBDRequestData *req;
1390 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1391 client->nb_requests++;
1393 req = g_new0(NBDRequestData, 1);
1394 nbd_client_get(client);
1395 req->client = client;
1399 static void nbd_request_put(NBDRequestData *req)
1401 NBDClient *client = req->client;
1404 qemu_vfree(req->data);
1408 client->nb_requests--;
1409 nbd_client_receive_next_request(client);
1411 nbd_client_put(client);
1414 static void blk_aio_attached(AioContext *ctx, void *opaque)
1416 NBDExport *exp = opaque;
1419 trace_nbd_blk_aio_attached(exp->name, ctx);
1421 exp->common.ctx = ctx;
1423 QTAILQ_FOREACH(client, &exp->clients, next) {
1424 qio_channel_attach_aio_context(client->ioc, ctx);
1425 if (client->recv_coroutine) {
1426 aio_co_schedule(ctx, client->recv_coroutine);
1428 if (client->send_coroutine) {
1429 aio_co_schedule(ctx, client->send_coroutine);
1434 static void blk_aio_detach(void *opaque)
1436 NBDExport *exp = opaque;
1439 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1441 QTAILQ_FOREACH(client, &exp->clients, next) {
1442 qio_channel_detach_aio_context(client->ioc);
1445 exp->common.ctx = NULL;
1448 static void nbd_eject_notifier(Notifier *n, void *data)
1450 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1452 blk_exp_request_shutdown(&exp->common);
1455 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1457 NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1458 assert(exp->drv == &blk_exp_nbd);
1459 assert(nbd_exp->eject_notifier_blk == NULL);
1462 nbd_exp->eject_notifier_blk = blk;
1463 nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1464 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1467 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1470 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1471 BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1472 BlockBackend *blk = blk_exp->blk;
1474 uint64_t perm, shared_perm;
1475 bool readonly = !exp_args->writable;
1476 bool shared = !exp_args->writable;
1479 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1481 if (!nbd_server_is_running()) {
1482 error_setg(errp, "NBD server not running");
1486 if (!arg->has_name) {
1487 arg->name = exp_args->node_name;
1490 if (strlen(arg->name) > NBD_MAX_STRING_SIZE) {
1491 error_setg(errp, "export name '%s' too long", arg->name);
1495 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1496 error_setg(errp, "description '%s' too long", arg->description);
1500 if (nbd_export_find(arg->name)) {
1501 error_setg(errp, "NBD server already has export named '%s'", arg->name);
1505 size = blk_getlength(blk);
1507 error_setg_errno(errp, -size,
1508 "Failed to determine the NBD export's length");
1512 /* Don't allow resize while the NBD server is running, otherwise we don't
1513 * care what happens with the node. */
1514 blk_get_perm(blk, &perm, &shared_perm);
1515 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1520 QTAILQ_INIT(&exp->clients);
1521 exp->name = g_strdup(arg->name);
1522 exp->description = g_strdup(arg->description);
1523 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1524 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1526 exp->nbdflags |= NBD_FLAG_READ_ONLY;
1528 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1531 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1532 NBD_FLAG_SEND_FAST_ZERO);
1534 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1537 BlockDriverState *bs = blk_bs(blk);
1538 BdrvDirtyBitmap *bm = NULL;
1541 bm = bdrv_find_dirty_bitmap(bs, arg->bitmap);
1546 bs = bdrv_filter_or_cow_bs(bs);
1551 error_setg(errp, "Bitmap '%s' is not found", arg->bitmap);
1555 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1560 if (readonly && bdrv_is_writable(bs) &&
1561 bdrv_dirty_bitmap_enabled(bm)) {
1564 "Enabled bitmap '%s' incompatible with readonly export",
1569 bdrv_dirty_bitmap_set_busy(bm, true);
1570 exp->export_bitmap = bm;
1571 assert(strlen(arg->bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1572 exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1574 assert(strlen(exp->export_bitmap_context) < NBD_MAX_STRING_SIZE);
1577 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1579 QTAILQ_INSERT_TAIL(&exports, exp, next);
1585 g_free(exp->description);
1589 NBDExport *nbd_export_find(const char *name)
1592 QTAILQ_FOREACH(exp, &exports, next) {
1593 if (strcmp(name, exp->name) == 0) {
1602 nbd_export_aio_context(NBDExport *exp)
1604 return exp->common.ctx;
1607 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1609 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1610 NBDClient *client, *next;
1612 blk_exp_ref(&exp->common);
1614 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1615 * close mode that stops advertising the export to new clients but
1616 * still permits existing clients to run to completion? Because of
1617 * that possibility, nbd_export_close() can be called more than
1618 * once on an export.
1620 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1621 client_close(client, true);
1626 QTAILQ_REMOVE(&exports, exp, next);
1628 blk_exp_unref(&exp->common);
1631 static void nbd_export_delete(BlockExport *blk_exp)
1633 NBDExport *exp = container_of(blk_exp, NBDExport, common);
1635 assert(exp->name == NULL);
1636 assert(QTAILQ_EMPTY(&exp->clients));
1638 g_free(exp->description);
1639 exp->description = NULL;
1641 if (exp->common.blk) {
1642 if (exp->eject_notifier_blk) {
1643 notifier_remove(&exp->eject_notifier);
1644 blk_unref(exp->eject_notifier_blk);
1646 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1647 blk_aio_detach, exp);
1650 if (exp->export_bitmap) {
1651 bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1652 g_free(exp->export_bitmap_context);
1656 const BlockExportDriver blk_exp_nbd = {
1657 .type = BLOCK_EXPORT_TYPE_NBD,
1658 .instance_size = sizeof(NBDExport),
1659 .create = nbd_export_create,
1660 .delete = nbd_export_delete,
1661 .request_shutdown = nbd_export_request_shutdown,
1664 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1665 unsigned niov, Error **errp)
1669 g_assert(qemu_in_coroutine());
1670 qemu_co_mutex_lock(&client->send_lock);
1671 client->send_coroutine = qemu_coroutine_self();
1673 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1675 client->send_coroutine = NULL;
1676 qemu_co_mutex_unlock(&client->send_lock);
1681 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1684 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1685 stl_be_p(&reply->error, error);
1686 stq_be_p(&reply->handle, handle);
1689 static int nbd_co_send_simple_reply(NBDClient *client,
1696 NBDSimpleReply reply;
1697 int nbd_err = system_errno_to_nbd_errno(error);
1698 struct iovec iov[] = {
1699 {.iov_base = &reply, .iov_len = sizeof(reply)},
1700 {.iov_base = data, .iov_len = len}
1703 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1705 set_be_simple_reply(&reply, nbd_err, handle);
1707 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1710 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1711 uint16_t type, uint64_t handle, uint32_t length)
1713 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1714 stw_be_p(&chunk->flags, flags);
1715 stw_be_p(&chunk->type, type);
1716 stq_be_p(&chunk->handle, handle);
1717 stl_be_p(&chunk->length, length);
1720 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1724 NBDStructuredReplyChunk chunk;
1725 struct iovec iov[] = {
1726 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1729 trace_nbd_co_send_structured_done(handle);
1730 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1732 return nbd_co_send_iov(client, iov, 1, errp);
1735 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1743 NBDStructuredReadData chunk;
1744 struct iovec iov[] = {
1745 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1746 {.iov_base = data, .iov_len = size}
1750 trace_nbd_co_send_structured_read(handle, offset, data, size);
1751 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1752 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1753 sizeof(chunk) - sizeof(chunk.h) + size);
1754 stq_be_p(&chunk.offset, offset);
1756 return nbd_co_send_iov(client, iov, 2, errp);
1759 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1765 NBDStructuredError chunk;
1766 int nbd_err = system_errno_to_nbd_errno(error);
1767 struct iovec iov[] = {
1768 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1769 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1773 trace_nbd_co_send_structured_error(handle, nbd_err,
1774 nbd_err_lookup(nbd_err), msg ? msg : "");
1775 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1776 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1777 stl_be_p(&chunk.error, nbd_err);
1778 stw_be_p(&chunk.message_length, iov[1].iov_len);
1780 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1783 /* Do a sparse read and send the structured reply to the client.
1784 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1785 * reported to the client, at which point this function succeeds.
1787 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1795 NBDExport *exp = client->exp;
1796 size_t progress = 0;
1798 while (progress < size) {
1800 int status = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
1802 size - progress, &pnum, NULL,
1807 char *msg = g_strdup_printf("unable to check for holes: %s",
1810 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1815 assert(pnum && pnum <= size - progress);
1816 final = progress + pnum == size;
1817 if (status & BDRV_BLOCK_ZERO) {
1818 NBDStructuredReadHole chunk;
1819 struct iovec iov[] = {
1820 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1823 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1825 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1826 NBD_REPLY_TYPE_OFFSET_HOLE,
1827 handle, sizeof(chunk) - sizeof(chunk.h));
1828 stq_be_p(&chunk.offset, offset + progress);
1829 stl_be_p(&chunk.length, pnum);
1830 ret = nbd_co_send_iov(client, iov, 1, errp);
1832 ret = blk_pread(exp->common.blk, offset + progress,
1833 data + progress, pnum);
1835 error_setg_errno(errp, -ret, "reading from file failed");
1838 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1839 data + progress, pnum, final,
1851 typedef struct NBDExtentArray {
1853 unsigned int nb_alloc;
1855 uint64_t total_length;
1857 bool converted_to_be;
1860 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
1862 NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
1864 ea->nb_alloc = nb_alloc;
1865 ea->extents = g_new(NBDExtent, nb_alloc);
1871 static void nbd_extent_array_free(NBDExtentArray *ea)
1873 g_free(ea->extents);
1876 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free);
1878 /* Further modifications of the array after conversion are abandoned */
1879 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
1883 assert(!ea->converted_to_be);
1884 ea->can_add = false;
1885 ea->converted_to_be = true;
1887 for (i = 0; i < ea->count; i++) {
1888 ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
1889 ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
1894 * Add extent to NBDExtentArray. If extent can't be added (no available space),
1896 * For safety, when returning -1 for the first time, .can_add is set to false,
1897 * further call to nbd_extent_array_add() will crash.
1898 * (to avoid the situation, when after failing to add an extent (returned -1),
1899 * user miss this failure and add another extent, which is successfully added
1900 * (array is full, but new extent may be squashed into the last one), then we
1901 * have invalid array with skipped extent)
1903 static int nbd_extent_array_add(NBDExtentArray *ea,
1904 uint32_t length, uint32_t flags)
1906 assert(ea->can_add);
1912 /* Extend previous extent if flags are the same */
1913 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
1914 uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
1916 if (sum <= UINT32_MAX) {
1917 ea->extents[ea->count - 1].length = sum;
1918 ea->total_length += length;
1923 if (ea->count >= ea->nb_alloc) {
1924 ea->can_add = false;
1928 ea->total_length += length;
1929 ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
1935 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1936 uint64_t bytes, NBDExtentArray *ea)
1941 int ret = bdrv_block_status_above(bs, NULL, offset, bytes, &num,
1948 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1949 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1951 if (nbd_extent_array_add(ea, num, flags) < 0) {
1963 * nbd_co_send_extents
1965 * @ea is converted to BE by the function
1966 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
1968 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1970 bool last, uint32_t context_id, Error **errp)
1972 NBDStructuredMeta chunk;
1973 struct iovec iov[] = {
1974 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1975 {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
1978 nbd_extent_array_convert_to_be(ea);
1980 trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
1982 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1983 NBD_REPLY_TYPE_BLOCK_STATUS,
1984 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1985 stl_be_p(&chunk.context_id, context_id);
1987 return nbd_co_send_iov(client, iov, 2, errp);
1990 /* Get block status from the exported device and send it to the client */
1991 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1992 BlockDriverState *bs, uint64_t offset,
1993 uint32_t length, bool dont_fragment,
1994 bool last, uint32_t context_id,
1998 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
1999 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2001 ret = blockstatus_to_extents(bs, offset, length, ea);
2003 return nbd_co_send_structured_error(
2004 client, handle, -ret, "can't get block status", errp);
2007 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2010 /* Populate @ea from a dirty bitmap. */
2011 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2012 uint64_t offset, uint64_t length,
2015 int64_t start, dirty_start, dirty_count;
2016 int64_t end = offset + length;
2019 bdrv_dirty_bitmap_lock(bitmap);
2021 for (start = offset;
2022 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2023 &dirty_start, &dirty_count);
2024 start = dirty_start + dirty_count)
2026 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2027 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2035 /* last non dirty extent */
2036 nbd_extent_array_add(es, end - start, 0);
2039 bdrv_dirty_bitmap_unlock(bitmap);
2042 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2043 BdrvDirtyBitmap *bitmap, uint64_t offset,
2044 uint32_t length, bool dont_fragment, bool last,
2045 uint32_t context_id, Error **errp)
2047 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2048 g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2050 bitmap_to_extents(bitmap, offset, length, ea);
2052 return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2055 /* nbd_co_receive_request
2056 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2057 * connection right away, and any other negative value to report an error to
2058 * the client (although the caller may still need to disconnect after reporting
2061 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2064 NBDClient *client = req->client;
2067 g_assert(qemu_in_coroutine());
2068 assert(client->recv_coroutine == qemu_coroutine_self());
2069 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2073 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2074 nbd_cmd_lookup(request->type));
2076 if (request->type != NBD_CMD_WRITE) {
2077 /* No payload, we are ready to read the next request. */
2078 req->complete = true;
2081 if (request->type == NBD_CMD_DISC) {
2082 /* Special case: we're going to disconnect without a reply,
2083 * whether or not flags, from, or len are bogus */
2087 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2088 request->type == NBD_CMD_CACHE)
2090 if (request->len > NBD_MAX_BUFFER_SIZE) {
2091 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2092 request->len, NBD_MAX_BUFFER_SIZE);
2096 if (request->type != NBD_CMD_CACHE) {
2097 req->data = blk_try_blockalign(client->exp->common.blk,
2099 if (req->data == NULL) {
2100 error_setg(errp, "No memory");
2106 if (request->type == NBD_CMD_WRITE) {
2107 if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2112 req->complete = true;
2114 trace_nbd_co_receive_request_payload_received(request->handle,
2118 /* Sanity checks. */
2119 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2120 (request->type == NBD_CMD_WRITE ||
2121 request->type == NBD_CMD_WRITE_ZEROES ||
2122 request->type == NBD_CMD_TRIM)) {
2123 error_setg(errp, "Export is read-only");
2126 if (request->from > client->exp->size ||
2127 request->len > client->exp->size - request->from) {
2128 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2129 ", Size: %" PRIu64, request->from, request->len,
2131 return (request->type == NBD_CMD_WRITE ||
2132 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2134 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2135 client->check_align)) {
2137 * The block layer gracefully handles unaligned requests, but
2138 * it's still worth tracing client non-compliance
2140 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2143 client->check_align);
2145 valid_flags = NBD_CMD_FLAG_FUA;
2146 if (request->type == NBD_CMD_READ && client->structured_reply) {
2147 valid_flags |= NBD_CMD_FLAG_DF;
2148 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2149 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2150 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2151 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2153 if (request->flags & ~valid_flags) {
2154 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2155 nbd_cmd_lookup(request->type), request->flags);
2162 /* Send simple reply without a payload, or a structured error
2163 * @error_msg is ignored if @ret >= 0
2164 * Returns 0 if connection is still live, -errno on failure to talk to client
2166 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2169 const char *error_msg,
2172 if (client->structured_reply && ret < 0) {
2173 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2176 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2181 /* Handle NBD_CMD_READ request.
2182 * Return -errno if sending fails. Other errors are reported directly to the
2183 * client as an error reply. */
2184 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2185 uint8_t *data, Error **errp)
2188 NBDExport *exp = client->exp;
2190 assert(request->type == NBD_CMD_READ);
2192 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2193 if (request->flags & NBD_CMD_FLAG_FUA) {
2194 ret = blk_co_flush(exp->common.blk);
2196 return nbd_send_generic_reply(client, request->handle, ret,
2197 "flush failed", errp);
2201 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2204 return nbd_co_send_sparse_read(client, request->handle, request->from,
2205 data, request->len, errp);
2208 ret = blk_pread(exp->common.blk, request->from, data, request->len);
2210 return nbd_send_generic_reply(client, request->handle, ret,
2211 "reading from file failed", errp);
2214 if (client->structured_reply) {
2216 return nbd_co_send_structured_read(client, request->handle,
2217 request->from, data,
2218 request->len, true, errp);
2220 return nbd_co_send_structured_done(client, request->handle, errp);
2223 return nbd_co_send_simple_reply(client, request->handle, 0,
2224 data, request->len, errp);
2231 * Handle NBD_CMD_CACHE request.
2232 * Return -errno if sending fails. Other errors are reported directly to the
2233 * client as an error reply.
2235 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2239 NBDExport *exp = client->exp;
2241 assert(request->type == NBD_CMD_CACHE);
2243 ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2244 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2246 return nbd_send_generic_reply(client, request->handle, ret,
2247 "caching data failed", errp);
2250 /* Handle NBD request.
2251 * Return -errno if sending fails. Other errors are reported directly to the
2252 * client as an error reply. */
2253 static coroutine_fn int nbd_handle_request(NBDClient *client,
2254 NBDRequest *request,
2255 uint8_t *data, Error **errp)
2259 NBDExport *exp = client->exp;
2262 switch (request->type) {
2264 return nbd_do_cmd_cache(client, request, errp);
2267 return nbd_do_cmd_read(client, request, data, errp);
2271 if (request->flags & NBD_CMD_FLAG_FUA) {
2272 flags |= BDRV_REQ_FUA;
2274 ret = blk_pwrite(exp->common.blk, request->from, data, request->len,
2276 return nbd_send_generic_reply(client, request->handle, ret,
2277 "writing to file failed", errp);
2279 case NBD_CMD_WRITE_ZEROES:
2281 if (request->flags & NBD_CMD_FLAG_FUA) {
2282 flags |= BDRV_REQ_FUA;
2284 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2285 flags |= BDRV_REQ_MAY_UNMAP;
2287 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2288 flags |= BDRV_REQ_NO_FALLBACK;
2291 /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2292 while (ret >= 0 && request->len) {
2293 int align = client->check_align ?: 1;
2294 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2296 ret = blk_pwrite_zeroes(exp->common.blk, request->from, len, flags);
2297 request->len -= len;
2298 request->from += len;
2300 return nbd_send_generic_reply(client, request->handle, ret,
2301 "writing to file failed", errp);
2304 /* unreachable, thanks to special case in nbd_co_receive_request() */
2308 ret = blk_co_flush(exp->common.blk);
2309 return nbd_send_generic_reply(client, request->handle, ret,
2310 "flush failed", errp);
2314 /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2315 while (ret >= 0 && request->len) {
2316 int align = client->check_align ?: 1;
2317 int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2319 ret = blk_co_pdiscard(exp->common.blk, request->from, len);
2320 request->len -= len;
2321 request->from += len;
2323 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2324 ret = blk_co_flush(exp->common.blk);
2326 return nbd_send_generic_reply(client, request->handle, ret,
2327 "discard failed", errp);
2329 case NBD_CMD_BLOCK_STATUS:
2330 if (!request->len) {
2331 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2332 "need non-zero length", errp);
2334 if (client->export_meta.valid &&
2335 (client->export_meta.base_allocation ||
2336 client->export_meta.bitmap))
2338 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2340 if (client->export_meta.base_allocation) {
2341 ret = nbd_co_send_block_status(client, request->handle,
2342 blk_bs(exp->common.blk),
2344 request->len, dont_fragment,
2345 !client->export_meta.bitmap,
2346 NBD_META_ID_BASE_ALLOCATION,
2353 if (client->export_meta.bitmap) {
2354 ret = nbd_co_send_bitmap(client, request->handle,
2355 client->exp->export_bitmap,
2356 request->from, request->len,
2358 true, NBD_META_ID_DIRTY_BITMAP, errp);
2366 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2367 "CMD_BLOCK_STATUS not negotiated",
2372 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2374 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2381 /* Owns a reference to the NBDClient passed as opaque. */
2382 static coroutine_fn void nbd_trip(void *opaque)
2384 NBDClient *client = opaque;
2385 NBDRequestData *req;
2386 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2388 Error *local_err = NULL;
2391 if (client->closing) {
2392 nbd_client_put(client);
2396 req = nbd_request_get(client);
2397 ret = nbd_co_receive_request(req, &request, &local_err);
2398 client->recv_coroutine = NULL;
2400 if (client->closing) {
2402 * The client may be closed when we are blocked in
2403 * nbd_co_receive_request()
2408 nbd_client_receive_next_request(client);
2414 /* It wans't -EIO, so, according to nbd_co_receive_request()
2415 * semantics, we should return the error to the client. */
2416 Error *export_err = local_err;
2419 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2420 error_get_pretty(export_err), &local_err);
2421 error_free(export_err);
2423 ret = nbd_handle_request(client, &request, req->data, &local_err);
2426 error_prepend(&local_err, "Failed to send reply: ");
2430 /* We must disconnect after NBD_CMD_WRITE if we did not
2433 if (!req->complete) {
2434 error_setg(&local_err, "Request handling failed in intermediate state");
2439 nbd_request_put(req);
2440 nbd_client_put(client);
2445 error_reportf_err(local_err, "Disconnect client, due to: ");
2447 nbd_request_put(req);
2448 client_close(client, true);
2449 nbd_client_put(client);
2452 static void nbd_client_receive_next_request(NBDClient *client)
2454 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2455 nbd_client_get(client);
2456 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2457 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
2461 static coroutine_fn void nbd_co_client_start(void *opaque)
2463 NBDClient *client = opaque;
2464 Error *local_err = NULL;
2466 qemu_co_mutex_init(&client->send_lock);
2468 if (nbd_negotiate(client, &local_err)) {
2470 error_report_err(local_err);
2472 client_close(client, false);
2476 nbd_client_receive_next_request(client);
2480 * Create a new client listener using the given channel @sioc.
2481 * Begin servicing it in a coroutine. When the connection closes, call
2482 * @close_fn with an indication of whether the client completed negotiation.
2484 void nbd_client_new(QIOChannelSocket *sioc,
2485 QCryptoTLSCreds *tlscreds,
2486 const char *tlsauthz,
2487 void (*close_fn)(NBDClient *, bool))
2492 client = g_new0(NBDClient, 1);
2493 client->refcount = 1;
2494 client->tlscreds = tlscreds;
2496 object_ref(OBJECT(client->tlscreds));
2498 client->tlsauthz = g_strdup(tlsauthz);
2499 client->sioc = sioc;
2500 object_ref(OBJECT(client->sioc));
2501 client->ioc = QIO_CHANNEL(sioc);
2502 object_ref(OBJECT(client->ioc));
2503 client->close_fn = close_fn;
2505 co = qemu_coroutine_create(nbd_co_client_start, client);
2506 qemu_coroutine_enter(co);