2 * Copyright (C) 2016-2018 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"
21 #include "qapi/error.h"
23 #include "nbd-internal.h"
25 #define NBD_META_ID_BASE_ALLOCATION 0
26 #define NBD_META_ID_DIRTY_BITMAP 1
28 /* NBD_MAX_BITMAP_EXTENTS: 1 mb of extents data. An empirical
29 * constant. If an increase is needed, note that the NBD protocol
30 * recommends no larger than 32 mb, so that the client won't consider
31 * the reply as a denial of service attack. */
32 #define NBD_MAX_BITMAP_EXTENTS (0x100000 / 8)
34 static int system_errno_to_nbd_errno(int err)
62 /* Definitions for opaque data types */
64 typedef struct NBDRequestData NBDRequestData;
66 struct NBDRequestData {
67 QSIMPLEQ_ENTRY(NBDRequestData) entry;
75 void (*close)(NBDExport *exp);
83 QTAILQ_HEAD(, NBDClient) clients;
84 QTAILQ_ENTRY(NBDExport) next;
88 BlockBackend *eject_notifier_blk;
89 Notifier eject_notifier;
91 BdrvDirtyBitmap *export_bitmap;
92 char *export_bitmap_context;
95 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
97 /* NBDExportMetaContexts represents a list of contexts to be exported,
98 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
99 * NBD_OPT_LIST_META_CONTEXT. */
100 typedef struct NBDExportMetaContexts {
102 bool valid; /* means that negotiation of the option finished without
104 bool base_allocation; /* export base:allocation context (block status) */
105 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
106 } NBDExportMetaContexts;
110 void (*close_fn)(NBDClient *client, bool negotiated);
113 QCryptoTLSCreds *tlscreds;
115 QIOChannelSocket *sioc; /* The underlying data channel */
116 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
118 Coroutine *recv_coroutine;
121 Coroutine *send_coroutine;
123 QTAILQ_ENTRY(NBDClient) next;
127 bool structured_reply;
128 NBDExportMetaContexts export_meta;
130 uint32_t opt; /* Current option being negotiated */
131 uint32_t optlen; /* remaining length of data in ioc for the option being
135 static void nbd_client_receive_next_request(NBDClient *client);
137 /* Basic flow for negotiation
164 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
165 uint32_t type, uint32_t length)
167 stq_be_p(&rep->magic, NBD_REP_MAGIC);
168 stl_be_p(&rep->option, option);
169 stl_be_p(&rep->type, type);
170 stl_be_p(&rep->length, length);
173 /* Send a reply header, including length, but no payload.
174 * Return -errno on error, 0 on success. */
175 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
176 uint32_t len, Error **errp)
180 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
181 type, nbd_rep_lookup(type), len);
183 assert(len < NBD_MAX_BUFFER_SIZE);
185 set_be_option_rep(&rep, client->opt, type, len);
186 return nbd_write(client->ioc, &rep, sizeof(rep), errp);
189 /* Send a reply header with default 0 length.
190 * Return -errno on error, 0 on success. */
191 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
194 return nbd_negotiate_send_rep_len(client, type, 0, errp);
197 /* Send an error reply.
198 * Return -errno on error, 0 on success. */
199 static int GCC_FMT_ATTR(4, 0)
200 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
201 Error **errp, const char *fmt, va_list va)
207 msg = g_strdup_vprintf(fmt, va);
210 trace_nbd_negotiate_send_rep_err(msg);
211 ret = nbd_negotiate_send_rep_len(client, type, len, errp);
215 if (nbd_write(client->ioc, msg, len, errp) < 0) {
216 error_prepend(errp, "write failed (error message): ");
227 /* Send an error reply.
228 * Return -errno on error, 0 on success. */
229 static int GCC_FMT_ATTR(4, 5)
230 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
231 Error **errp, const char *fmt, ...)
237 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
242 /* Drop remainder of the current option, and send a reply with the
243 * given error type and message. Return -errno on read or write
244 * failure; or 0 if connection is still live. */
245 static int GCC_FMT_ATTR(4, 0)
246 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
247 const char *fmt, va_list va)
249 int ret = nbd_drop(client->ioc, client->optlen, errp);
253 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
258 static int GCC_FMT_ATTR(4, 5)
259 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
260 const char *fmt, ...)
266 ret = nbd_opt_vdrop(client, type, errp, fmt, va);
272 static int GCC_FMT_ATTR(3, 4)
273 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
279 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
285 /* Read size bytes from the unparsed payload of the current option.
286 * Return -errno on I/O error, 0 if option was completely handled by
287 * sending a reply about inconsistent lengths, or 1 on success. */
288 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
291 if (size > client->optlen) {
292 return nbd_opt_invalid(client, errp,
293 "Inconsistent lengths in option %s",
294 nbd_opt_lookup(client->opt));
296 client->optlen -= size;
297 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
300 /* Drop size bytes from the unparsed payload of the current option.
301 * Return -errno on I/O error, 0 if option was completely handled by
302 * sending a reply about inconsistent lengths, or 1 on success. */
303 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
305 if (size > client->optlen) {
306 return nbd_opt_invalid(client, errp,
307 "Inconsistent lengths in option %s",
308 nbd_opt_lookup(client->opt));
310 client->optlen -= size;
311 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
316 * Read a string with the format:
317 * uint32_t len (<= NBD_MAX_NAME_SIZE)
318 * len bytes string (not 0-terminated)
320 * @name should be enough to store NBD_MAX_NAME_SIZE+1.
321 * If @length is non-null, it will be set to the actual string length.
323 * Return -errno on I/O error, 0 if option was completely handled by
324 * sending a reply about inconsistent lengths, or 1 on success.
326 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
332 ret = nbd_opt_read(client, &len, sizeof(len), errp);
338 if (len > NBD_MAX_NAME_SIZE) {
339 return nbd_opt_invalid(client, errp,
340 "Invalid name length: %" PRIu32, len);
343 ret = nbd_opt_read(client, name, len, errp);
356 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
357 * Return -errno on error, 0 on success. */
358 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
361 size_t name_len, desc_len;
363 const char *name = exp->name ? exp->name : "";
364 const char *desc = exp->description ? exp->description : "";
365 QIOChannel *ioc = client->ioc;
368 trace_nbd_negotiate_send_rep_list(name, desc);
369 name_len = strlen(name);
370 desc_len = strlen(desc);
371 len = name_len + desc_len + sizeof(len);
372 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
377 len = cpu_to_be32(name_len);
378 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
379 error_prepend(errp, "write failed (name length): ");
383 if (nbd_write(ioc, name, name_len, errp) < 0) {
384 error_prepend(errp, "write failed (name buffer): ");
388 if (nbd_write(ioc, desc, desc_len, errp) < 0) {
389 error_prepend(errp, "write failed (description buffer): ");
396 /* Process the NBD_OPT_LIST command, with a potential series of replies.
397 * Return -errno on error, 0 on success. */
398 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
401 assert(client->opt == NBD_OPT_LIST);
403 /* For each export, send a NBD_REP_SERVER reply. */
404 QTAILQ_FOREACH(exp, &exports, next) {
405 if (nbd_negotiate_send_rep_list(client, exp, errp)) {
409 /* Finish with a NBD_REP_ACK. */
410 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
413 static void nbd_check_meta_export(NBDClient *client)
415 client->export_meta.valid &= client->exp == client->export_meta.exp;
418 /* Send a reply to NBD_OPT_EXPORT_NAME.
419 * Return -errno on error, 0 on success. */
420 static int nbd_negotiate_handle_export_name(NBDClient *client,
421 uint16_t myflags, bool no_zeroes,
424 char name[NBD_MAX_NAME_SIZE + 1];
425 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
430 [20 .. xx] export name (length bytes)
433 [ 8 .. 9] export flags
434 [10 .. 133] reserved (0) [unless no_zeroes]
436 trace_nbd_negotiate_handle_export_name();
437 if (client->optlen >= sizeof(name)) {
438 error_setg(errp, "Bad length received");
441 if (nbd_read(client->ioc, name, client->optlen, errp) < 0) {
442 error_prepend(errp, "read failed: ");
445 name[client->optlen] = '\0';
448 trace_nbd_negotiate_handle_export_name_request(name);
450 client->exp = nbd_export_find(name);
452 error_setg(errp, "export not found");
456 trace_nbd_negotiate_new_style_size_flags(client->exp->size,
457 client->exp->nbdflags | myflags);
458 stq_be_p(buf, client->exp->size);
459 stw_be_p(buf + 8, client->exp->nbdflags | myflags);
460 len = no_zeroes ? 10 : sizeof(buf);
461 ret = nbd_write(client->ioc, buf, len, errp);
463 error_prepend(errp, "write failed: ");
467 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
468 nbd_export_get(client->exp);
469 nbd_check_meta_export(client);
474 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
475 * The buffer does NOT include the info type prefix.
476 * Return -errno on error, 0 if ready to send more. */
477 static int nbd_negotiate_send_info(NBDClient *client,
478 uint16_t info, uint32_t length, void *buf,
483 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
484 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
485 sizeof(info) + length, errp);
490 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
493 if (nbd_write(client->ioc, buf, length, errp) < 0) {
499 /* nbd_reject_length: Handle any unexpected payload.
500 * @fatal requests that we quit talking to the client, even if we are able
501 * to successfully send an error reply.
503 * -errno transmission error occurred or @fatal was requested, errp is set
504 * 0 error message successfully sent to client, errp is not set
506 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
510 assert(client->optlen);
511 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
512 nbd_opt_lookup(client->opt));
514 error_setg(errp, "option '%s' has unexpected length",
515 nbd_opt_lookup(client->opt));
521 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
522 * Return -errno on error, 0 if ready for next option, and 1 to move
523 * into transmission phase. */
524 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
528 char name[NBD_MAX_NAME_SIZE + 1];
533 bool sendname = false;
534 bool blocksize = false;
536 char buf[sizeof(uint64_t) + sizeof(uint16_t)];
539 4 bytes: L, name length (can be 0)
541 2 bytes: N, number of requests (can be 0)
542 N * 2 bytes: N requests
544 rc = nbd_opt_read_name(client, name, &namelen, errp);
548 trace_nbd_negotiate_handle_export_name_request(name);
550 rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
554 be16_to_cpus(&requests);
555 trace_nbd_negotiate_handle_info_requests(requests);
557 rc = nbd_opt_read(client, &request, sizeof(request), errp);
561 be16_to_cpus(&request);
562 trace_nbd_negotiate_handle_info_request(request,
563 nbd_info_lookup(request));
564 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
565 * everything else is either a request we don't know or
566 * something we send regardless of request */
571 case NBD_INFO_BLOCK_SIZE:
576 if (client->optlen) {
577 return nbd_reject_length(client, false, errp);
580 exp = nbd_export_find(name);
582 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
583 errp, "export '%s' not present",
587 /* Don't bother sending NBD_INFO_NAME unless client requested it */
589 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
596 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
598 if (exp->description) {
599 size_t len = strlen(exp->description);
601 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
602 len, exp->description, errp);
608 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
609 * according to whether the client requested it, and according to
610 * whether this is OPT_INFO or OPT_GO. */
611 /* minimum - 1 for back-compat, or 512 if client is new enough.
612 * TODO: consult blk_bs(blk)->bl.request_alignment? */
614 (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
615 /* preferred - Hard-code to 4096 for now.
616 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
618 /* maximum - At most 32M, but smaller as appropriate. */
619 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
620 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
621 cpu_to_be32s(&sizes[0]);
622 cpu_to_be32s(&sizes[1]);
623 cpu_to_be32s(&sizes[2]);
624 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
625 sizeof(sizes), sizes, errp);
630 /* Send NBD_INFO_EXPORT always */
631 trace_nbd_negotiate_new_style_size_flags(exp->size,
632 exp->nbdflags | myflags);
633 stq_be_p(buf, exp->size);
634 stw_be_p(buf + 8, exp->nbdflags | myflags);
635 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
636 sizeof(buf), buf, errp);
641 /* If the client is just asking for NBD_OPT_INFO, but forgot to
642 * request block sizes, return an error.
643 * TODO: consult blk_bs(blk)->request_align, and only error if it
645 if (client->opt == NBD_OPT_INFO && !blocksize) {
646 return nbd_negotiate_send_rep_err(client,
647 NBD_REP_ERR_BLOCK_SIZE_REQD,
649 "request NBD_INFO_BLOCK_SIZE to "
654 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
659 if (client->opt == NBD_OPT_GO) {
661 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
662 nbd_export_get(client->exp);
663 nbd_check_meta_export(client);
670 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
671 * new channel for all further (now-encrypted) communication. */
672 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
677 struct NBDTLSHandshakeData data = { 0 };
679 assert(client->opt == NBD_OPT_STARTTLS);
681 trace_nbd_negotiate_handle_starttls();
684 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
688 tioc = qio_channel_tls_new_server(ioc,
696 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
697 trace_nbd_negotiate_handle_starttls_handshake();
698 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
699 qio_channel_tls_handshake(tioc,
705 if (!data.complete) {
706 g_main_loop_run(data.loop);
708 g_main_loop_unref(data.loop);
710 object_unref(OBJECT(tioc));
711 error_propagate(errp, data.error);
715 return QIO_CHANNEL(tioc);
718 /* nbd_negotiate_send_meta_context
720 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
722 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
724 static int nbd_negotiate_send_meta_context(NBDClient *client,
729 NBDOptionReplyMetaContext opt;
730 struct iovec iov[] = {
731 {.iov_base = &opt, .iov_len = sizeof(opt)},
732 {.iov_base = (void *)context, .iov_len = strlen(context)}
735 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
739 trace_nbd_negotiate_meta_query_reply(context, context_id);
740 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
741 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
742 stl_be_p(&opt.context_id, context_id);
744 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
747 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
748 * @match is never set to false.
750 * Return -errno on I/O error, 0 if option was completely handled by
751 * sending a reply about inconsistent lengths, or 1 on success.
753 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
754 * It only means that there are no errors.
756 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
761 size_t len = strlen(pattern);
765 query = g_malloc(len);
766 ret = nbd_opt_read(client, query, len, errp);
772 if (strncmp(query, pattern, len) == 0) {
773 trace_nbd_negotiate_meta_query_parse(pattern);
776 trace_nbd_negotiate_meta_query_skip("pattern not matched");
784 * Read @len bytes, and set @match to true if they match @pattern, or if @len
785 * is 0 and the client is performing _LIST_. @match is never set to false.
787 * Return -errno on I/O error, 0 if option was completely handled by
788 * sending a reply about inconsistent lengths, or 1 on success.
790 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
791 * It only means that there are no errors.
793 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
794 uint32_t len, bool *match, Error **errp)
797 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
800 trace_nbd_negotiate_meta_query_parse("empty");
804 if (len != strlen(pattern)) {
805 trace_nbd_negotiate_meta_query_skip("different lengths");
806 return nbd_opt_skip(client, len, errp);
809 return nbd_meta_pattern(client, pattern, match, errp);
812 /* nbd_meta_base_query
814 * Handle queries to 'base' namespace. For now, only the base:allocation
815 * context is available. 'len' is the amount of text remaining to be read from
816 * the current name, after the 'base:' portion has been stripped.
818 * Return -errno on I/O error, 0 if option was completely handled by
819 * sending a reply about inconsistent lengths, or 1 on success.
821 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
822 uint32_t len, Error **errp)
824 return nbd_meta_empty_or_pattern(client, "allocation", len,
825 &meta->base_allocation, errp);
828 /* nbd_meta_bitmap_query
830 * Handle query to 'qemu:' namespace.
831 * @len is the amount of text remaining to be read from the current name, after
832 * the 'qemu:' portion has been stripped.
834 * Return -errno on I/O error, 0 if option was completely handled by
835 * sending a reply about inconsistent lengths, or 1 on success. */
836 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
837 uint32_t len, Error **errp)
839 bool dirty_bitmap = false;
840 size_t dirty_bitmap_len = strlen("dirty-bitmap:");
843 if (!meta->exp->export_bitmap) {
844 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
845 return nbd_opt_skip(client, len, errp);
849 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
852 trace_nbd_negotiate_meta_query_parse("empty");
856 if (len < dirty_bitmap_len) {
857 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
858 return nbd_opt_skip(client, len, errp);
861 len -= dirty_bitmap_len;
862 ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
867 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
868 return nbd_opt_skip(client, len, errp);
871 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
873 return nbd_meta_empty_or_pattern(
874 client, meta->exp->export_bitmap_context +
875 strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
878 /* nbd_negotiate_meta_query
880 * Parse namespace name and call corresponding function to parse body of the
883 * The only supported namespace now is 'base'.
885 * The function aims not wasting time and memory to read long unknown namespace
888 * Return -errno on I/O error, 0 if option was completely handled by
889 * sending a reply about inconsistent lengths, or 1 on success. */
890 static int nbd_negotiate_meta_query(NBDClient *client,
891 NBDExportMetaContexts *meta, Error **errp)
894 * Both 'qemu' and 'base' namespaces have length = 5 including a
895 * colon. If another length namespace is later introduced, this
896 * should certainly be refactored.
903 ret = nbd_opt_read(client, &len, sizeof(len), errp);
910 trace_nbd_negotiate_meta_query_skip("length too short");
911 return nbd_opt_skip(client, len, errp);
915 ret = nbd_opt_read(client, ns, ns_len, errp);
920 if (!strncmp(ns, "base:", ns_len)) {
921 trace_nbd_negotiate_meta_query_parse("base:");
922 return nbd_meta_base_query(client, meta, len, errp);
923 } else if (!strncmp(ns, "qemu:", ns_len)) {
924 trace_nbd_negotiate_meta_query_parse("qemu:");
925 return nbd_meta_qemu_query(client, meta, len, errp);
928 trace_nbd_negotiate_meta_query_skip("unknown namespace");
929 return nbd_opt_skip(client, len, errp);
932 /* nbd_negotiate_meta_queries
933 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
935 * Return -errno on I/O error, or 0 if option was completely handled. */
936 static int nbd_negotiate_meta_queries(NBDClient *client,
937 NBDExportMetaContexts *meta, Error **errp)
940 char export_name[NBD_MAX_NAME_SIZE + 1];
941 NBDExportMetaContexts local_meta;
945 if (!client->structured_reply) {
946 return nbd_opt_invalid(client, errp,
947 "request option '%s' when structured reply "
949 nbd_opt_lookup(client->opt));
952 if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
953 /* Only change the caller's meta on SET. */
957 memset(meta, 0, sizeof(*meta));
959 ret = nbd_opt_read_name(client, export_name, NULL, errp);
964 meta->exp = nbd_export_find(export_name);
965 if (meta->exp == NULL) {
966 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
967 "export '%s' not present", export_name);
970 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
974 cpu_to_be32s(&nb_queries);
975 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
976 export_name, nb_queries);
978 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
979 /* enable all known contexts */
980 meta->base_allocation = true;
982 for (i = 0; i < nb_queries; ++i) {
983 ret = nbd_negotiate_meta_query(client, meta, errp);
990 if (meta->base_allocation) {
991 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
992 NBD_META_ID_BASE_ALLOCATION,
1000 ret = nbd_negotiate_send_meta_context(client,
1001 meta->exp->export_bitmap_context,
1002 NBD_META_ID_DIRTY_BITMAP,
1009 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1017 /* nbd_negotiate_options
1018 * Process all NBD_OPT_* client option commands, during fixed newstyle
1021 * -errno on error, errp is set
1022 * 0 on successful negotiation, errp is not set
1023 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1026 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1030 bool fixedNewstyle = false;
1031 bool no_zeroes = false;
1034 [ 0 .. 3] client flags
1036 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1037 [ 0 .. 7] NBD_OPTS_MAGIC
1038 [ 8 .. 11] NBD option
1039 [12 .. 15] Data length
1042 [ 0 .. 7] NBD_OPTS_MAGIC
1043 [ 8 .. 11] Second NBD option
1044 [12 .. 15] Data length
1048 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
1049 error_prepend(errp, "read failed: ");
1052 be32_to_cpus(&flags);
1053 trace_nbd_negotiate_options_flags(flags);
1054 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1055 fixedNewstyle = true;
1056 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1058 if (flags & NBD_FLAG_C_NO_ZEROES) {
1060 flags &= ~NBD_FLAG_C_NO_ZEROES;
1063 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1069 uint32_t option, length;
1072 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
1073 error_prepend(errp, "read failed: ");
1076 magic = be64_to_cpu(magic);
1077 trace_nbd_negotiate_options_check_magic(magic);
1078 if (magic != NBD_OPTS_MAGIC) {
1079 error_setg(errp, "Bad magic received");
1083 if (nbd_read(client->ioc, &option,
1084 sizeof(option), errp) < 0) {
1085 error_prepend(errp, "read failed: ");
1088 option = be32_to_cpu(option);
1089 client->opt = option;
1091 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
1092 error_prepend(errp, "read failed: ");
1095 length = be32_to_cpu(length);
1096 assert(!client->optlen);
1097 client->optlen = length;
1099 if (length > NBD_MAX_BUFFER_SIZE) {
1100 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1101 length, NBD_MAX_BUFFER_SIZE);
1105 trace_nbd_negotiate_options_check_option(option,
1106 nbd_opt_lookup(option));
1107 if (client->tlscreds &&
1108 client->ioc == (QIOChannel *)client->sioc) {
1110 if (!fixedNewstyle) {
1111 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1115 case NBD_OPT_STARTTLS:
1117 /* Unconditionally drop the connection if the client
1118 * can't start a TLS negotiation correctly */
1119 return nbd_reject_length(client, true, errp);
1121 tioc = nbd_negotiate_handle_starttls(client, errp);
1126 object_unref(OBJECT(client->ioc));
1127 client->ioc = QIO_CHANNEL(tioc);
1130 case NBD_OPT_EXPORT_NAME:
1131 /* No way to return an error to client, so drop connection */
1132 error_setg(errp, "Option 0x%x not permitted before TLS",
1137 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, errp,
1139 "not permitted before TLS", option);
1140 /* Let the client keep trying, unless they asked to
1141 * quit. In this mode, we've already sent an error, so
1142 * we can't ack the abort. */
1143 if (option == NBD_OPT_ABORT) {
1148 } else if (fixedNewstyle) {
1152 ret = nbd_reject_length(client, false, errp);
1154 ret = nbd_negotiate_handle_list(client, errp);
1159 /* NBD spec says we must try to reply before
1160 * disconnecting, but that we must also tolerate
1161 * guests that don't wait for our reply. */
1162 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1165 case NBD_OPT_EXPORT_NAME:
1166 return nbd_negotiate_handle_export_name(client,
1172 ret = nbd_negotiate_handle_info(client, myflags, 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;
1203 myflags |= NBD_FLAG_SEND_DF;
1207 case NBD_OPT_LIST_META_CONTEXT:
1208 case NBD_OPT_SET_META_CONTEXT:
1209 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1214 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1215 "Unsupported option %" PRIu32 " (%s)",
1216 option, nbd_opt_lookup(option));
1221 * If broken new-style we should drop the connection
1222 * for anything except NBD_OPT_EXPORT_NAME
1225 case NBD_OPT_EXPORT_NAME:
1226 return nbd_negotiate_handle_export_name(client,
1231 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1232 option, nbd_opt_lookup(option));
1244 * -errno on error, errp is set
1245 * 0 on successful negotiation, errp is not set
1246 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1249 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1251 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1253 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1254 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1255 NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1258 /* Old style negotiation header, no room for options
1259 [ 0 .. 7] passwd ("NBDMAGIC")
1260 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1262 [24 .. 27] export flags (zero-extended)
1263 [28 .. 151] reserved (0)
1265 New style negotiation header, client can send options
1266 [ 0 .. 7] passwd ("NBDMAGIC")
1267 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1268 [16 .. 17] server flags (0)
1269 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1272 qio_channel_set_blocking(client->ioc, false, NULL);
1274 trace_nbd_negotiate_begin();
1275 memcpy(buf, "NBDMAGIC", 8);
1277 oldStyle = client->exp != NULL && !client->tlscreds;
1279 trace_nbd_negotiate_old_style(client->exp->size,
1280 client->exp->nbdflags | myflags);
1281 stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
1282 stq_be_p(buf + 16, client->exp->size);
1283 stl_be_p(buf + 24, client->exp->nbdflags | myflags);
1285 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
1286 error_prepend(errp, "write failed: ");
1290 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1291 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1293 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1294 error_prepend(errp, "write failed: ");
1297 ret = nbd_negotiate_options(client, myflags, errp);
1300 error_prepend(errp, "option negotiation failed: ");
1306 assert(!client->optlen);
1307 trace_nbd_negotiate_success();
1312 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1315 uint8_t buf[NBD_REQUEST_SIZE];
1319 ret = nbd_read(ioc, buf, sizeof(buf), errp);
1325 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1326 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1327 [ 6 .. 7] type (NBD_CMD_READ, ...)
1333 magic = ldl_be_p(buf);
1334 request->flags = lduw_be_p(buf + 4);
1335 request->type = lduw_be_p(buf + 6);
1336 request->handle = ldq_be_p(buf + 8);
1337 request->from = ldq_be_p(buf + 16);
1338 request->len = ldl_be_p(buf + 24);
1340 trace_nbd_receive_request(magic, request->flags, request->type,
1341 request->from, request->len);
1343 if (magic != NBD_REQUEST_MAGIC) {
1344 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1350 #define MAX_NBD_REQUESTS 16
1352 void nbd_client_get(NBDClient *client)
1357 void nbd_client_put(NBDClient *client)
1359 if (--client->refcount == 0) {
1360 /* The last reference should be dropped by client->close,
1361 * which is called by client_close.
1363 assert(client->closing);
1365 qio_channel_detach_aio_context(client->ioc);
1366 object_unref(OBJECT(client->sioc));
1367 object_unref(OBJECT(client->ioc));
1368 if (client->tlscreds) {
1369 object_unref(OBJECT(client->tlscreds));
1371 g_free(client->tlsaclname);
1373 QTAILQ_REMOVE(&client->exp->clients, client, next);
1374 nbd_export_put(client->exp);
1380 static void client_close(NBDClient *client, bool negotiated)
1382 if (client->closing) {
1386 client->closing = true;
1388 /* Force requests to finish. They will drop their own references,
1389 * then we'll close the socket and free the NBDClient.
1391 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1394 /* Also tell the client, so that they release their reference. */
1395 if (client->close_fn) {
1396 client->close_fn(client, negotiated);
1400 static NBDRequestData *nbd_request_get(NBDClient *client)
1402 NBDRequestData *req;
1404 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1405 client->nb_requests++;
1407 req = g_new0(NBDRequestData, 1);
1408 nbd_client_get(client);
1409 req->client = client;
1413 static void nbd_request_put(NBDRequestData *req)
1415 NBDClient *client = req->client;
1418 qemu_vfree(req->data);
1422 client->nb_requests--;
1423 nbd_client_receive_next_request(client);
1425 nbd_client_put(client);
1428 static void blk_aio_attached(AioContext *ctx, void *opaque)
1430 NBDExport *exp = opaque;
1433 trace_nbd_blk_aio_attached(exp->name, ctx);
1437 QTAILQ_FOREACH(client, &exp->clients, next) {
1438 qio_channel_attach_aio_context(client->ioc, ctx);
1439 if (client->recv_coroutine) {
1440 aio_co_schedule(ctx, client->recv_coroutine);
1442 if (client->send_coroutine) {
1443 aio_co_schedule(ctx, client->send_coroutine);
1448 static void blk_aio_detach(void *opaque)
1450 NBDExport *exp = opaque;
1453 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1455 QTAILQ_FOREACH(client, &exp->clients, next) {
1456 qio_channel_detach_aio_context(client->ioc);
1462 static void nbd_eject_notifier(Notifier *n, void *data)
1464 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1465 nbd_export_close(exp);
1468 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1469 uint16_t nbdflags, void (*close)(NBDExport *),
1470 bool writethrough, BlockBackend *on_eject_blk,
1475 NBDExport *exp = g_new0(NBDExport, 1);
1480 * NBD exports are used for non-shared storage migration. Make sure
1481 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1482 * access since the export could be available before migration handover.
1484 ctx = bdrv_get_aio_context(bs);
1485 aio_context_acquire(ctx);
1486 bdrv_invalidate_cache(bs, NULL);
1487 aio_context_release(ctx);
1489 /* Don't allow resize while the NBD server is running, otherwise we don't
1490 * care what happens with the node. */
1491 perm = BLK_PERM_CONSISTENT_READ;
1492 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1493 perm |= BLK_PERM_WRITE;
1495 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1496 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1497 ret = blk_insert_bs(blk, bs, errp);
1501 blk_set_enable_write_cache(blk, !writethrough);
1504 QTAILQ_INIT(&exp->clients);
1506 exp->dev_offset = dev_offset;
1507 exp->nbdflags = nbdflags;
1508 exp->size = size < 0 ? blk_getlength(blk) : size;
1509 if (exp->size < 0) {
1510 error_setg_errno(errp, -exp->size,
1511 "Failed to determine the NBD export's length");
1514 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1517 exp->ctx = blk_get_aio_context(blk);
1518 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1521 blk_ref(on_eject_blk);
1522 exp->eject_notifier_blk = on_eject_blk;
1523 exp->eject_notifier.notify = nbd_eject_notifier;
1524 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1534 NBDExport *nbd_export_find(const char *name)
1537 QTAILQ_FOREACH(exp, &exports, next) {
1538 if (strcmp(name, exp->name) == 0) {
1546 void nbd_export_set_name(NBDExport *exp, const char *name)
1548 if (exp->name == name) {
1552 nbd_export_get(exp);
1553 if (exp->name != NULL) {
1556 QTAILQ_REMOVE(&exports, exp, next);
1557 nbd_export_put(exp);
1560 nbd_export_get(exp);
1561 exp->name = g_strdup(name);
1562 QTAILQ_INSERT_TAIL(&exports, exp, next);
1564 nbd_export_put(exp);
1567 void nbd_export_set_description(NBDExport *exp, const char *description)
1569 g_free(exp->description);
1570 exp->description = g_strdup(description);
1573 void nbd_export_close(NBDExport *exp)
1575 NBDClient *client, *next;
1577 nbd_export_get(exp);
1578 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1579 client_close(client, true);
1581 nbd_export_set_name(exp, NULL);
1582 nbd_export_set_description(exp, NULL);
1583 nbd_export_put(exp);
1586 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1588 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1589 nbd_export_close(exp);
1593 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1595 error_setg(errp, "export '%s' still in use", exp->name);
1596 error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1599 void nbd_export_get(NBDExport *exp)
1601 assert(exp->refcount > 0);
1605 void nbd_export_put(NBDExport *exp)
1607 assert(exp->refcount > 0);
1608 if (exp->refcount == 1) {
1609 nbd_export_close(exp);
1612 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1613 * if someone calls nbd_export_put() on named export not through
1614 * nbd_export_set_name() when refcount is 1. So, let's assert that
1617 assert(exp->refcount > 0);
1618 if (--exp->refcount == 0) {
1619 assert(exp->name == NULL);
1620 assert(exp->description == NULL);
1627 if (exp->eject_notifier_blk) {
1628 notifier_remove(&exp->eject_notifier);
1629 blk_unref(exp->eject_notifier_blk);
1631 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1632 blk_aio_detach, exp);
1633 blk_unref(exp->blk);
1637 if (exp->export_bitmap) {
1638 bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1639 g_free(exp->export_bitmap_context);
1646 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1651 void nbd_export_close_all(void)
1653 NBDExport *exp, *next;
1655 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1656 nbd_export_close(exp);
1660 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1661 unsigned niov, Error **errp)
1665 g_assert(qemu_in_coroutine());
1666 qemu_co_mutex_lock(&client->send_lock);
1667 client->send_coroutine = qemu_coroutine_self();
1669 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1671 client->send_coroutine = NULL;
1672 qemu_co_mutex_unlock(&client->send_lock);
1677 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1680 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1681 stl_be_p(&reply->error, error);
1682 stq_be_p(&reply->handle, handle);
1685 static int nbd_co_send_simple_reply(NBDClient *client,
1692 NBDSimpleReply reply;
1693 int nbd_err = system_errno_to_nbd_errno(error);
1694 struct iovec iov[] = {
1695 {.iov_base = &reply, .iov_len = sizeof(reply)},
1696 {.iov_base = data, .iov_len = len}
1699 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1701 set_be_simple_reply(&reply, nbd_err, handle);
1703 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1706 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1707 uint16_t type, uint64_t handle, uint32_t length)
1709 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1710 stw_be_p(&chunk->flags, flags);
1711 stw_be_p(&chunk->type, type);
1712 stq_be_p(&chunk->handle, handle);
1713 stl_be_p(&chunk->length, length);
1716 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1720 NBDStructuredReplyChunk chunk;
1721 struct iovec iov[] = {
1722 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1725 trace_nbd_co_send_structured_done(handle);
1726 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1728 return nbd_co_send_iov(client, iov, 1, errp);
1731 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1739 NBDStructuredReadData chunk;
1740 struct iovec iov[] = {
1741 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1742 {.iov_base = data, .iov_len = size}
1746 trace_nbd_co_send_structured_read(handle, offset, data, size);
1747 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1748 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1749 sizeof(chunk) - sizeof(chunk.h) + size);
1750 stq_be_p(&chunk.offset, offset);
1752 return nbd_co_send_iov(client, iov, 2, errp);
1755 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1761 NBDStructuredError chunk;
1762 int nbd_err = system_errno_to_nbd_errno(error);
1763 struct iovec iov[] = {
1764 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1765 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1769 trace_nbd_co_send_structured_error(handle, nbd_err,
1770 nbd_err_lookup(nbd_err), msg ? msg : "");
1771 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1772 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1773 stl_be_p(&chunk.error, nbd_err);
1774 stw_be_p(&chunk.message_length, iov[1].iov_len);
1776 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1779 /* Do a sparse read and send the structured reply to the client.
1780 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1781 * reported to the client, at which point this function succeeds.
1783 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1791 NBDExport *exp = client->exp;
1792 size_t progress = 0;
1794 while (progress < size) {
1796 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1798 size - progress, &pnum, NULL,
1803 char *msg = g_strdup_printf("unable to check for holes: %s",
1806 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1811 assert(pnum && pnum <= size - progress);
1812 final = progress + pnum == size;
1813 if (status & BDRV_BLOCK_ZERO) {
1814 NBDStructuredReadHole chunk;
1815 struct iovec iov[] = {
1816 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1819 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1821 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1822 NBD_REPLY_TYPE_OFFSET_HOLE,
1823 handle, sizeof(chunk) - sizeof(chunk.h));
1824 stq_be_p(&chunk.offset, offset + progress);
1825 stl_be_p(&chunk.length, pnum);
1826 ret = nbd_co_send_iov(client, iov, 1, errp);
1828 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1829 data + progress, pnum);
1831 error_setg_errno(errp, -ret, "reading from file failed");
1834 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1835 data + progress, pnum, final,
1848 * Populate @extents from block status. Update @bytes to be the actual
1849 * length encoded (which may be smaller than the original), and update
1850 * @nb_extents to the number of extents used.
1852 * Returns zero on success and -errno on bdrv_block_status_above failure.
1854 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1855 uint64_t *bytes, NBDExtent *extents,
1856 unsigned int *nb_extents)
1858 uint64_t remaining_bytes = *bytes;
1859 NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1860 bool first_extent = true;
1862 assert(*nb_extents);
1863 while (remaining_bytes) {
1866 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1873 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1874 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1876 remaining_bytes -= num;
1879 extent->flags = flags;
1880 extent->length = num;
1881 first_extent = false;
1885 if (flags == extent->flags) {
1886 /* extend current extent */
1887 extent->length += num;
1889 if (extent + 1 == extents_end) {
1893 /* start new extent */
1895 extent->flags = flags;
1896 extent->length = num;
1900 extents_end = extent + 1;
1902 for (extent = extents; extent < extents_end; extent++) {
1903 cpu_to_be32s(&extent->flags);
1904 cpu_to_be32s(&extent->length);
1907 *bytes -= remaining_bytes;
1908 *nb_extents = extents_end - extents;
1913 /* nbd_co_send_extents
1915 * @length is only for tracing purposes (and may be smaller or larger
1916 * than the client's original request). @last controls whether
1917 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1918 * big-endian format.
1920 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1921 NBDExtent *extents, unsigned int nb_extents,
1922 uint64_t length, bool last,
1923 uint32_t context_id, Error **errp)
1925 NBDStructuredMeta chunk;
1927 struct iovec iov[] = {
1928 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1929 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1932 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1933 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1934 NBD_REPLY_TYPE_BLOCK_STATUS,
1935 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1936 stl_be_p(&chunk.context_id, context_id);
1938 return nbd_co_send_iov(client, iov, 2, errp);
1941 /* Get block status from the exported device and send it to the client */
1942 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1943 BlockDriverState *bs, uint64_t offset,
1944 uint32_t length, bool dont_fragment,
1945 bool last, uint32_t context_id,
1949 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1950 NBDExtent *extents = g_new(NBDExtent, nb_extents);
1951 uint64_t final_length = length;
1953 ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1957 return nbd_co_send_structured_error(
1958 client, handle, -ret, "can't get block status", errp);
1961 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1962 final_length, last, context_id, errp);
1970 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1971 * final extent may exceed the original @length. Store in @length the
1972 * byte length encoded (which may be smaller or larger than the
1973 * original), and return the number of extents used.
1975 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1976 uint64_t *length, NBDExtent *extents,
1977 unsigned int nb_extents,
1980 uint64_t begin = offset, end = offset;
1981 uint64_t overall_end = offset + *length;
1983 BdrvDirtyBitmapIter *it;
1986 bdrv_dirty_bitmap_lock(bitmap);
1988 it = bdrv_dirty_iter_new(bitmap);
1989 dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1991 assert(begin < overall_end && nb_extents);
1992 while (begin < overall_end && i < nb_extents) {
1993 bool next_dirty = !dirty;
1996 end = bdrv_dirty_bitmap_next_zero(bitmap, begin);
1998 bdrv_set_dirty_iter(it, begin);
1999 end = bdrv_dirty_iter_next(it);
2001 if (end == -1 || end - begin > UINT32_MAX) {
2002 /* Cap to an aligned value < 4G beyond begin. */
2003 end = MIN(bdrv_dirty_bitmap_size(bitmap),
2004 begin + UINT32_MAX + 1 -
2005 bdrv_dirty_bitmap_granularity(bitmap));
2008 if (dont_fragment && end > overall_end) {
2012 extents[i].length = cpu_to_be32(end - begin);
2013 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2019 bdrv_dirty_iter_free(it);
2021 bdrv_dirty_bitmap_unlock(bitmap);
2023 assert(offset < end);
2024 *length = end - offset;
2028 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2029 BdrvDirtyBitmap *bitmap, uint64_t offset,
2030 uint32_t length, bool dont_fragment, bool last,
2031 uint32_t context_id, Error **errp)
2034 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2035 NBDExtent *extents = g_new(NBDExtent, nb_extents);
2036 uint64_t final_length = length;
2038 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2039 nb_extents, dont_fragment);
2041 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2042 final_length, last, context_id, errp);
2049 /* nbd_co_receive_request
2050 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2051 * connection right away, and any other negative value to report an error to
2052 * the client (although the caller may still need to disconnect after reporting
2055 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2058 NBDClient *client = req->client;
2061 g_assert(qemu_in_coroutine());
2062 assert(client->recv_coroutine == qemu_coroutine_self());
2063 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2067 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2068 nbd_cmd_lookup(request->type));
2070 if (request->type != NBD_CMD_WRITE) {
2071 /* No payload, we are ready to read the next request. */
2072 req->complete = true;
2075 if (request->type == NBD_CMD_DISC) {
2076 /* Special case: we're going to disconnect without a reply,
2077 * whether or not flags, from, or len are bogus */
2081 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2082 request->type == NBD_CMD_CACHE)
2084 if (request->len > NBD_MAX_BUFFER_SIZE) {
2085 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2086 request->len, NBD_MAX_BUFFER_SIZE);
2090 req->data = blk_try_blockalign(client->exp->blk, request->len);
2091 if (req->data == NULL) {
2092 error_setg(errp, "No memory");
2096 if (request->type == NBD_CMD_WRITE) {
2097 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
2098 error_prepend(errp, "reading from socket failed: ");
2101 req->complete = true;
2103 trace_nbd_co_receive_request_payload_received(request->handle,
2107 /* Sanity checks. */
2108 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2109 (request->type == NBD_CMD_WRITE ||
2110 request->type == NBD_CMD_WRITE_ZEROES ||
2111 request->type == NBD_CMD_TRIM)) {
2112 error_setg(errp, "Export is read-only");
2115 if (request->from > client->exp->size ||
2116 request->from + request->len > client->exp->size) {
2117 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2118 ", Size: %" PRIu64, request->from, request->len,
2119 (uint64_t)client->exp->size);
2120 return (request->type == NBD_CMD_WRITE ||
2121 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2123 valid_flags = NBD_CMD_FLAG_FUA;
2124 if (request->type == NBD_CMD_READ && client->structured_reply) {
2125 valid_flags |= NBD_CMD_FLAG_DF;
2126 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2127 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2128 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2129 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2131 if (request->flags & ~valid_flags) {
2132 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2133 nbd_cmd_lookup(request->type), request->flags);
2140 /* Send simple reply without a payload, or a structured error
2141 * @error_msg is ignored if @ret >= 0
2142 * Returns 0 if connection is still live, -errno on failure to talk to client
2144 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2147 const char *error_msg,
2150 if (client->structured_reply && ret < 0) {
2151 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2154 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2159 /* Handle NBD_CMD_READ request.
2160 * Return -errno if sending fails. Other errors are reported directly to the
2161 * client as an error reply. */
2162 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2163 uint8_t *data, Error **errp)
2166 NBDExport *exp = client->exp;
2168 assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2170 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2171 if (request->flags & NBD_CMD_FLAG_FUA) {
2172 ret = blk_co_flush(exp->blk);
2174 return nbd_send_generic_reply(client, request->handle, ret,
2175 "flush failed", errp);
2179 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2181 return nbd_co_send_sparse_read(client, request->handle, request->from,
2182 data, request->len, errp);
2185 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2187 if (ret < 0 || request->type == NBD_CMD_CACHE) {
2188 return nbd_send_generic_reply(client, request->handle, ret,
2189 "reading from file failed", errp);
2192 if (client->structured_reply) {
2194 return nbd_co_send_structured_read(client, request->handle,
2195 request->from, data,
2196 request->len, true, errp);
2198 return nbd_co_send_structured_done(client, request->handle, errp);
2201 return nbd_co_send_simple_reply(client, request->handle, 0,
2202 data, request->len, errp);
2206 /* Handle NBD request.
2207 * Return -errno if sending fails. Other errors are reported directly to the
2208 * client as an error reply. */
2209 static coroutine_fn int nbd_handle_request(NBDClient *client,
2210 NBDRequest *request,
2211 uint8_t *data, Error **errp)
2215 NBDExport *exp = client->exp;
2218 switch (request->type) {
2221 return nbd_do_cmd_read(client, request, data, errp);
2225 if (request->flags & NBD_CMD_FLAG_FUA) {
2226 flags |= BDRV_REQ_FUA;
2228 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2229 data, request->len, flags);
2230 return nbd_send_generic_reply(client, request->handle, ret,
2231 "writing to file failed", errp);
2233 case NBD_CMD_WRITE_ZEROES:
2235 if (request->flags & NBD_CMD_FLAG_FUA) {
2236 flags |= BDRV_REQ_FUA;
2238 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2239 flags |= BDRV_REQ_MAY_UNMAP;
2241 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2242 request->len, flags);
2243 return nbd_send_generic_reply(client, request->handle, ret,
2244 "writing to file failed", errp);
2247 /* unreachable, thanks to special case in nbd_co_receive_request() */
2251 ret = blk_co_flush(exp->blk);
2252 return nbd_send_generic_reply(client, request->handle, ret,
2253 "flush failed", errp);
2256 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2258 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2259 ret = blk_co_flush(exp->blk);
2261 return nbd_send_generic_reply(client, request->handle, ret,
2262 "discard failed", errp);
2264 case NBD_CMD_BLOCK_STATUS:
2265 if (!request->len) {
2266 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2267 "need non-zero length", errp);
2269 if (client->export_meta.valid &&
2270 (client->export_meta.base_allocation ||
2271 client->export_meta.bitmap))
2273 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2275 if (client->export_meta.base_allocation) {
2276 ret = nbd_co_send_block_status(client, request->handle,
2277 blk_bs(exp->blk), request->from,
2278 request->len, dont_fragment,
2279 !client->export_meta.bitmap,
2280 NBD_META_ID_BASE_ALLOCATION,
2287 if (client->export_meta.bitmap) {
2288 ret = nbd_co_send_bitmap(client, request->handle,
2289 client->exp->export_bitmap,
2290 request->from, request->len,
2292 true, NBD_META_ID_DIRTY_BITMAP, errp);
2300 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2301 "CMD_BLOCK_STATUS not negotiated",
2306 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2308 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2315 /* Owns a reference to the NBDClient passed as opaque. */
2316 static coroutine_fn void nbd_trip(void *opaque)
2318 NBDClient *client = opaque;
2319 NBDRequestData *req;
2320 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2322 Error *local_err = NULL;
2325 if (client->closing) {
2326 nbd_client_put(client);
2330 req = nbd_request_get(client);
2331 ret = nbd_co_receive_request(req, &request, &local_err);
2332 client->recv_coroutine = NULL;
2334 if (client->closing) {
2336 * The client may be closed when we are blocked in
2337 * nbd_co_receive_request()
2342 nbd_client_receive_next_request(client);
2348 /* It wans't -EIO, so, according to nbd_co_receive_request()
2349 * semantics, we should return the error to the client. */
2350 Error *export_err = local_err;
2353 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2354 error_get_pretty(export_err), &local_err);
2355 error_free(export_err);
2357 ret = nbd_handle_request(client, &request, req->data, &local_err);
2360 error_prepend(&local_err, "Failed to send reply: ");
2364 /* We must disconnect after NBD_CMD_WRITE if we did not
2367 if (!req->complete) {
2368 error_setg(&local_err, "Request handling failed in intermediate state");
2373 nbd_request_put(req);
2374 nbd_client_put(client);
2379 error_reportf_err(local_err, "Disconnect client, due to: ");
2381 nbd_request_put(req);
2382 client_close(client, true);
2383 nbd_client_put(client);
2386 static void nbd_client_receive_next_request(NBDClient *client)
2388 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2389 nbd_client_get(client);
2390 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2391 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2395 static coroutine_fn void nbd_co_client_start(void *opaque)
2397 NBDClient *client = opaque;
2398 NBDExport *exp = client->exp;
2399 Error *local_err = NULL;
2402 nbd_export_get(exp);
2403 QTAILQ_INSERT_TAIL(&exp->clients, client, next);
2405 qemu_co_mutex_init(&client->send_lock);
2407 if (nbd_negotiate(client, &local_err)) {
2409 error_report_err(local_err);
2411 client_close(client, false);
2415 nbd_client_receive_next_request(client);
2419 * Create a new client listener on the given export @exp, using the
2420 * given channel @sioc. Begin servicing it in a coroutine. When the
2421 * connection closes, call @close_fn with an indication of whether the
2422 * client completed negotiation.
2424 void nbd_client_new(NBDExport *exp,
2425 QIOChannelSocket *sioc,
2426 QCryptoTLSCreds *tlscreds,
2427 const char *tlsaclname,
2428 void (*close_fn)(NBDClient *, bool))
2433 client = g_new0(NBDClient, 1);
2434 client->refcount = 1;
2436 client->tlscreds = tlscreds;
2438 object_ref(OBJECT(client->tlscreds));
2440 client->tlsaclname = g_strdup(tlsaclname);
2441 client->sioc = sioc;
2442 object_ref(OBJECT(client->sioc));
2443 client->ioc = QIO_CHANNEL(sioc);
2444 object_ref(OBJECT(client->ioc));
2445 client->close_fn = close_fn;
2447 co = qemu_coroutine_create(nbd_co_client_start, client);
2448 qemu_coroutine_enter(co);
2451 void nbd_export_bitmap(NBDExport *exp, const char *bitmap,
2452 const char *bitmap_export_name, Error **errp)
2454 BdrvDirtyBitmap *bm = NULL;
2455 BlockDriverState *bs = blk_bs(exp->blk);
2457 if (exp->export_bitmap) {
2458 error_setg(errp, "Export bitmap is already set");
2463 bm = bdrv_find_dirty_bitmap(bs, bitmap);
2464 if (bm != NULL || bs->backing == NULL) {
2468 bs = bs->backing->bs;
2472 error_setg(errp, "Bitmap '%s' is not found", bitmap);
2476 if (bdrv_dirty_bitmap_enabled(bm)) {
2477 error_setg(errp, "Bitmap '%s' is enabled", bitmap);
2481 if (bdrv_dirty_bitmap_qmp_locked(bm)) {
2482 error_setg(errp, "Bitmap '%s' is locked", bitmap);
2486 bdrv_dirty_bitmap_set_qmp_locked(bm, true);
2487 exp->export_bitmap = bm;
2488 exp->export_bitmap_context =
2489 g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name);