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);
336 len = cpu_to_be32(len);
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);
489 info = cpu_to_be16(info);
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 requests = be16_to_cpu(requests);
555 trace_nbd_negotiate_handle_info_requests(requests);
557 rc = nbd_opt_read(client, &request, sizeof(request), errp);
561 request = be16_to_cpu(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 sizes[0] = cpu_to_be32(sizes[0]);
622 sizes[1] = cpu_to_be32(sizes[1]);
623 sizes[2] = cpu_to_be32(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);
907 len = cpu_to_be32(len);
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 nb_queries = cpu_to_be32(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;
981 meta->bitmap = !!meta->exp->export_bitmap;
983 for (i = 0; i < nb_queries; ++i) {
984 ret = nbd_negotiate_meta_query(client, meta, errp);
991 if (meta->base_allocation) {
992 ret = nbd_negotiate_send_meta_context(client, "base:allocation",
993 NBD_META_ID_BASE_ALLOCATION,
1001 ret = nbd_negotiate_send_meta_context(client,
1002 meta->exp->export_bitmap_context,
1003 NBD_META_ID_DIRTY_BITMAP,
1010 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1018 /* nbd_negotiate_options
1019 * Process all NBD_OPT_* client option commands, during fixed newstyle
1022 * -errno on error, errp is set
1023 * 0 on successful negotiation, errp is not set
1024 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1027 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1031 bool fixedNewstyle = false;
1032 bool no_zeroes = false;
1035 [ 0 .. 3] client flags
1037 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1038 [ 0 .. 7] NBD_OPTS_MAGIC
1039 [ 8 .. 11] NBD option
1040 [12 .. 15] Data length
1043 [ 0 .. 7] NBD_OPTS_MAGIC
1044 [ 8 .. 11] Second NBD option
1045 [12 .. 15] Data length
1049 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
1050 error_prepend(errp, "read failed: ");
1053 flags = be32_to_cpu(flags);
1054 trace_nbd_negotiate_options_flags(flags);
1055 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1056 fixedNewstyle = true;
1057 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1059 if (flags & NBD_FLAG_C_NO_ZEROES) {
1061 flags &= ~NBD_FLAG_C_NO_ZEROES;
1064 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1070 uint32_t option, length;
1073 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
1074 error_prepend(errp, "read failed: ");
1077 magic = be64_to_cpu(magic);
1078 trace_nbd_negotiate_options_check_magic(magic);
1079 if (magic != NBD_OPTS_MAGIC) {
1080 error_setg(errp, "Bad magic received");
1084 if (nbd_read(client->ioc, &option,
1085 sizeof(option), errp) < 0) {
1086 error_prepend(errp, "read failed: ");
1089 option = be32_to_cpu(option);
1090 client->opt = option;
1092 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
1093 error_prepend(errp, "read failed: ");
1096 length = be32_to_cpu(length);
1097 assert(!client->optlen);
1098 client->optlen = length;
1100 if (length > NBD_MAX_BUFFER_SIZE) {
1101 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1102 length, NBD_MAX_BUFFER_SIZE);
1106 trace_nbd_negotiate_options_check_option(option,
1107 nbd_opt_lookup(option));
1108 if (client->tlscreds &&
1109 client->ioc == (QIOChannel *)client->sioc) {
1111 if (!fixedNewstyle) {
1112 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1116 case NBD_OPT_STARTTLS:
1118 /* Unconditionally drop the connection if the client
1119 * can't start a TLS negotiation correctly */
1120 return nbd_reject_length(client, true, errp);
1122 tioc = nbd_negotiate_handle_starttls(client, errp);
1127 object_unref(OBJECT(client->ioc));
1128 client->ioc = QIO_CHANNEL(tioc);
1131 case NBD_OPT_EXPORT_NAME:
1132 /* No way to return an error to client, so drop connection */
1133 error_setg(errp, "Option 0x%x not permitted before TLS",
1138 /* Let the client keep trying, unless they asked to
1139 * quit. Always try to give an error back to the
1140 * client; but when replying to OPT_ABORT, be aware
1141 * that the client may hang up before receiving the
1142 * error, in which case we are fine ignoring the
1143 * resulting EPIPE. */
1144 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1145 option == NBD_OPT_ABORT ? NULL : errp,
1147 " not permitted before TLS", option);
1148 if (option == NBD_OPT_ABORT) {
1153 } else if (fixedNewstyle) {
1157 ret = nbd_reject_length(client, false, errp);
1159 ret = nbd_negotiate_handle_list(client, errp);
1164 /* NBD spec says we must try to reply before
1165 * disconnecting, but that we must also tolerate
1166 * guests that don't wait for our reply. */
1167 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1170 case NBD_OPT_EXPORT_NAME:
1171 return nbd_negotiate_handle_export_name(client,
1177 ret = nbd_negotiate_handle_info(client, myflags, errp);
1179 assert(option == NBD_OPT_GO);
1184 case NBD_OPT_STARTTLS:
1186 ret = nbd_reject_length(client, false, errp);
1187 } else if (client->tlscreds) {
1188 ret = nbd_negotiate_send_rep_err(client,
1189 NBD_REP_ERR_INVALID, errp,
1190 "TLS already enabled");
1192 ret = nbd_negotiate_send_rep_err(client,
1193 NBD_REP_ERR_POLICY, errp,
1194 "TLS not configured");
1198 case NBD_OPT_STRUCTURED_REPLY:
1200 ret = nbd_reject_length(client, false, errp);
1201 } else if (client->structured_reply) {
1202 ret = nbd_negotiate_send_rep_err(
1203 client, NBD_REP_ERR_INVALID, errp,
1204 "structured reply already negotiated");
1206 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1207 client->structured_reply = true;
1208 myflags |= NBD_FLAG_SEND_DF;
1212 case NBD_OPT_LIST_META_CONTEXT:
1213 case NBD_OPT_SET_META_CONTEXT:
1214 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1219 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1220 "Unsupported option %" PRIu32 " (%s)",
1221 option, nbd_opt_lookup(option));
1226 * If broken new-style we should drop the connection
1227 * for anything except NBD_OPT_EXPORT_NAME
1230 case NBD_OPT_EXPORT_NAME:
1231 return nbd_negotiate_handle_export_name(client,
1236 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1237 option, nbd_opt_lookup(option));
1249 * -errno on error, errp is set
1250 * 0 on successful negotiation, errp is not set
1251 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1254 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1256 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1258 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1259 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1260 NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1262 /* Old style negotiation header, no room for options
1263 [ 0 .. 7] passwd ("NBDMAGIC")
1264 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1266 [24 .. 27] export flags (zero-extended)
1267 [28 .. 151] reserved (0)
1269 New style negotiation header, client can send options
1270 [ 0 .. 7] passwd ("NBDMAGIC")
1271 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1272 [16 .. 17] server flags (0)
1273 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1276 qio_channel_set_blocking(client->ioc, false, NULL);
1278 trace_nbd_negotiate_begin();
1279 memcpy(buf, "NBDMAGIC", 8);
1281 stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1282 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1284 if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1285 error_prepend(errp, "write failed: ");
1288 ret = nbd_negotiate_options(client, myflags, errp);
1291 error_prepend(errp, "option negotiation failed: ");
1296 assert(!client->optlen);
1297 trace_nbd_negotiate_success();
1302 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1305 uint8_t buf[NBD_REQUEST_SIZE];
1309 ret = nbd_read(ioc, buf, sizeof(buf), errp);
1315 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1316 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1317 [ 6 .. 7] type (NBD_CMD_READ, ...)
1323 magic = ldl_be_p(buf);
1324 request->flags = lduw_be_p(buf + 4);
1325 request->type = lduw_be_p(buf + 6);
1326 request->handle = ldq_be_p(buf + 8);
1327 request->from = ldq_be_p(buf + 16);
1328 request->len = ldl_be_p(buf + 24);
1330 trace_nbd_receive_request(magic, request->flags, request->type,
1331 request->from, request->len);
1333 if (magic != NBD_REQUEST_MAGIC) {
1334 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1340 #define MAX_NBD_REQUESTS 16
1342 void nbd_client_get(NBDClient *client)
1347 void nbd_client_put(NBDClient *client)
1349 if (--client->refcount == 0) {
1350 /* The last reference should be dropped by client->close,
1351 * which is called by client_close.
1353 assert(client->closing);
1355 qio_channel_detach_aio_context(client->ioc);
1356 object_unref(OBJECT(client->sioc));
1357 object_unref(OBJECT(client->ioc));
1358 if (client->tlscreds) {
1359 object_unref(OBJECT(client->tlscreds));
1361 g_free(client->tlsaclname);
1363 QTAILQ_REMOVE(&client->exp->clients, client, next);
1364 nbd_export_put(client->exp);
1370 static void client_close(NBDClient *client, bool negotiated)
1372 if (client->closing) {
1376 client->closing = true;
1378 /* Force requests to finish. They will drop their own references,
1379 * then we'll close the socket and free the NBDClient.
1381 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1384 /* Also tell the client, so that they release their reference. */
1385 if (client->close_fn) {
1386 client->close_fn(client, negotiated);
1390 static NBDRequestData *nbd_request_get(NBDClient *client)
1392 NBDRequestData *req;
1394 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1395 client->nb_requests++;
1397 req = g_new0(NBDRequestData, 1);
1398 nbd_client_get(client);
1399 req->client = client;
1403 static void nbd_request_put(NBDRequestData *req)
1405 NBDClient *client = req->client;
1408 qemu_vfree(req->data);
1412 client->nb_requests--;
1413 nbd_client_receive_next_request(client);
1415 nbd_client_put(client);
1418 static void blk_aio_attached(AioContext *ctx, void *opaque)
1420 NBDExport *exp = opaque;
1423 trace_nbd_blk_aio_attached(exp->name, ctx);
1427 QTAILQ_FOREACH(client, &exp->clients, next) {
1428 qio_channel_attach_aio_context(client->ioc, ctx);
1429 if (client->recv_coroutine) {
1430 aio_co_schedule(ctx, client->recv_coroutine);
1432 if (client->send_coroutine) {
1433 aio_co_schedule(ctx, client->send_coroutine);
1438 static void blk_aio_detach(void *opaque)
1440 NBDExport *exp = opaque;
1443 trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1445 QTAILQ_FOREACH(client, &exp->clients, next) {
1446 qio_channel_detach_aio_context(client->ioc);
1452 static void nbd_eject_notifier(Notifier *n, void *data)
1454 NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1455 nbd_export_close(exp);
1458 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1459 uint16_t nbdflags, void (*close)(NBDExport *),
1460 bool writethrough, BlockBackend *on_eject_blk,
1465 NBDExport *exp = g_new0(NBDExport, 1);
1470 * NBD exports are used for non-shared storage migration. Make sure
1471 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1472 * access since the export could be available before migration handover.
1474 ctx = bdrv_get_aio_context(bs);
1475 aio_context_acquire(ctx);
1476 bdrv_invalidate_cache(bs, NULL);
1477 aio_context_release(ctx);
1479 /* Don't allow resize while the NBD server is running, otherwise we don't
1480 * care what happens with the node. */
1481 perm = BLK_PERM_CONSISTENT_READ;
1482 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1483 perm |= BLK_PERM_WRITE;
1485 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1486 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1487 ret = blk_insert_bs(blk, bs, errp);
1491 blk_set_enable_write_cache(blk, !writethrough);
1494 QTAILQ_INIT(&exp->clients);
1496 exp->dev_offset = dev_offset;
1497 exp->nbdflags = nbdflags;
1498 exp->size = size < 0 ? blk_getlength(blk) : size;
1499 if (exp->size < 0) {
1500 error_setg_errno(errp, -exp->size,
1501 "Failed to determine the NBD export's length");
1504 exp->size -= exp->size % BDRV_SECTOR_SIZE;
1507 exp->ctx = blk_get_aio_context(blk);
1508 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1511 blk_ref(on_eject_blk);
1512 exp->eject_notifier_blk = on_eject_blk;
1513 exp->eject_notifier.notify = nbd_eject_notifier;
1514 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1524 NBDExport *nbd_export_find(const char *name)
1527 QTAILQ_FOREACH(exp, &exports, next) {
1528 if (strcmp(name, exp->name) == 0) {
1536 void nbd_export_set_name(NBDExport *exp, const char *name)
1538 if (exp->name == name) {
1542 nbd_export_get(exp);
1543 if (exp->name != NULL) {
1546 QTAILQ_REMOVE(&exports, exp, next);
1547 nbd_export_put(exp);
1550 nbd_export_get(exp);
1551 exp->name = g_strdup(name);
1552 QTAILQ_INSERT_TAIL(&exports, exp, next);
1554 nbd_export_put(exp);
1557 void nbd_export_set_description(NBDExport *exp, const char *description)
1559 g_free(exp->description);
1560 exp->description = g_strdup(description);
1563 void nbd_export_close(NBDExport *exp)
1565 NBDClient *client, *next;
1567 nbd_export_get(exp);
1568 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1569 client_close(client, true);
1571 nbd_export_set_name(exp, NULL);
1572 nbd_export_set_description(exp, NULL);
1573 nbd_export_put(exp);
1576 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1578 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1579 nbd_export_close(exp);
1583 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1585 error_setg(errp, "export '%s' still in use", exp->name);
1586 error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1589 void nbd_export_get(NBDExport *exp)
1591 assert(exp->refcount > 0);
1595 void nbd_export_put(NBDExport *exp)
1597 assert(exp->refcount > 0);
1598 if (exp->refcount == 1) {
1599 nbd_export_close(exp);
1602 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1603 * if someone calls nbd_export_put() on named export not through
1604 * nbd_export_set_name() when refcount is 1. So, let's assert that
1607 assert(exp->refcount > 0);
1608 if (--exp->refcount == 0) {
1609 assert(exp->name == NULL);
1610 assert(exp->description == NULL);
1617 if (exp->eject_notifier_blk) {
1618 notifier_remove(&exp->eject_notifier);
1619 blk_unref(exp->eject_notifier_blk);
1621 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1622 blk_aio_detach, exp);
1623 blk_unref(exp->blk);
1627 if (exp->export_bitmap) {
1628 bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1629 g_free(exp->export_bitmap_context);
1636 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1641 void nbd_export_close_all(void)
1643 NBDExport *exp, *next;
1645 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1646 nbd_export_close(exp);
1650 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1651 unsigned niov, Error **errp)
1655 g_assert(qemu_in_coroutine());
1656 qemu_co_mutex_lock(&client->send_lock);
1657 client->send_coroutine = qemu_coroutine_self();
1659 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1661 client->send_coroutine = NULL;
1662 qemu_co_mutex_unlock(&client->send_lock);
1667 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1670 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1671 stl_be_p(&reply->error, error);
1672 stq_be_p(&reply->handle, handle);
1675 static int nbd_co_send_simple_reply(NBDClient *client,
1682 NBDSimpleReply reply;
1683 int nbd_err = system_errno_to_nbd_errno(error);
1684 struct iovec iov[] = {
1685 {.iov_base = &reply, .iov_len = sizeof(reply)},
1686 {.iov_base = data, .iov_len = len}
1689 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1691 set_be_simple_reply(&reply, nbd_err, handle);
1693 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1696 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1697 uint16_t type, uint64_t handle, uint32_t length)
1699 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1700 stw_be_p(&chunk->flags, flags);
1701 stw_be_p(&chunk->type, type);
1702 stq_be_p(&chunk->handle, handle);
1703 stl_be_p(&chunk->length, length);
1706 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1710 NBDStructuredReplyChunk chunk;
1711 struct iovec iov[] = {
1712 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1715 trace_nbd_co_send_structured_done(handle);
1716 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1718 return nbd_co_send_iov(client, iov, 1, errp);
1721 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1729 NBDStructuredReadData chunk;
1730 struct iovec iov[] = {
1731 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1732 {.iov_base = data, .iov_len = size}
1736 trace_nbd_co_send_structured_read(handle, offset, data, size);
1737 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1738 NBD_REPLY_TYPE_OFFSET_DATA, handle,
1739 sizeof(chunk) - sizeof(chunk.h) + size);
1740 stq_be_p(&chunk.offset, offset);
1742 return nbd_co_send_iov(client, iov, 2, errp);
1745 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1751 NBDStructuredError chunk;
1752 int nbd_err = system_errno_to_nbd_errno(error);
1753 struct iovec iov[] = {
1754 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1755 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1759 trace_nbd_co_send_structured_error(handle, nbd_err,
1760 nbd_err_lookup(nbd_err), msg ? msg : "");
1761 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1762 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1763 stl_be_p(&chunk.error, nbd_err);
1764 stw_be_p(&chunk.message_length, iov[1].iov_len);
1766 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1769 /* Do a sparse read and send the structured reply to the client.
1770 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1771 * reported to the client, at which point this function succeeds.
1773 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1781 NBDExport *exp = client->exp;
1782 size_t progress = 0;
1784 while (progress < size) {
1786 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1788 size - progress, &pnum, NULL,
1793 char *msg = g_strdup_printf("unable to check for holes: %s",
1796 ret = nbd_co_send_structured_error(client, handle, -status, msg,
1801 assert(pnum && pnum <= size - progress);
1802 final = progress + pnum == size;
1803 if (status & BDRV_BLOCK_ZERO) {
1804 NBDStructuredReadHole chunk;
1805 struct iovec iov[] = {
1806 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1809 trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1811 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1812 NBD_REPLY_TYPE_OFFSET_HOLE,
1813 handle, sizeof(chunk) - sizeof(chunk.h));
1814 stq_be_p(&chunk.offset, offset + progress);
1815 stl_be_p(&chunk.length, pnum);
1816 ret = nbd_co_send_iov(client, iov, 1, errp);
1818 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1819 data + progress, pnum);
1821 error_setg_errno(errp, -ret, "reading from file failed");
1824 ret = nbd_co_send_structured_read(client, handle, offset + progress,
1825 data + progress, pnum, final,
1838 * Populate @extents from block status. Update @bytes to be the actual
1839 * length encoded (which may be smaller than the original), and update
1840 * @nb_extents to the number of extents used.
1842 * Returns zero on success and -errno on bdrv_block_status_above failure.
1844 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1845 uint64_t *bytes, NBDExtent *extents,
1846 unsigned int *nb_extents)
1848 uint64_t remaining_bytes = *bytes;
1849 NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1850 bool first_extent = true;
1852 assert(*nb_extents);
1853 while (remaining_bytes) {
1856 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1863 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1864 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
1866 remaining_bytes -= num;
1869 extent->flags = flags;
1870 extent->length = num;
1871 first_extent = false;
1875 if (flags == extent->flags) {
1876 /* extend current extent */
1877 extent->length += num;
1879 if (extent + 1 == extents_end) {
1883 /* start new extent */
1885 extent->flags = flags;
1886 extent->length = num;
1890 extents_end = extent + 1;
1892 for (extent = extents; extent < extents_end; extent++) {
1893 extent->flags = cpu_to_be32(extent->flags);
1894 extent->length = cpu_to_be32(extent->length);
1897 *bytes -= remaining_bytes;
1898 *nb_extents = extents_end - extents;
1903 /* nbd_co_send_extents
1905 * @length is only for tracing purposes (and may be smaller or larger
1906 * than the client's original request). @last controls whether
1907 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1908 * big-endian format.
1910 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1911 NBDExtent *extents, unsigned int nb_extents,
1912 uint64_t length, bool last,
1913 uint32_t context_id, Error **errp)
1915 NBDStructuredMeta chunk;
1917 struct iovec iov[] = {
1918 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1919 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1922 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1923 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1924 NBD_REPLY_TYPE_BLOCK_STATUS,
1925 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1926 stl_be_p(&chunk.context_id, context_id);
1928 return nbd_co_send_iov(client, iov, 2, errp);
1931 /* Get block status from the exported device and send it to the client */
1932 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1933 BlockDriverState *bs, uint64_t offset,
1934 uint32_t length, bool dont_fragment,
1935 bool last, uint32_t context_id,
1939 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1940 NBDExtent *extents = g_new(NBDExtent, nb_extents);
1941 uint64_t final_length = length;
1943 ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1947 return nbd_co_send_structured_error(
1948 client, handle, -ret, "can't get block status", errp);
1951 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1952 final_length, last, context_id, errp);
1960 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1961 * final extent may exceed the original @length. Store in @length the
1962 * byte length encoded (which may be smaller or larger than the
1963 * original), and return the number of extents used.
1965 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1966 uint64_t *length, NBDExtent *extents,
1967 unsigned int nb_extents,
1970 uint64_t begin = offset, end = offset;
1971 uint64_t overall_end = offset + *length;
1973 BdrvDirtyBitmapIter *it;
1976 bdrv_dirty_bitmap_lock(bitmap);
1978 it = bdrv_dirty_iter_new(bitmap);
1979 dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1981 assert(begin < overall_end && nb_extents);
1982 while (begin < overall_end && i < nb_extents) {
1983 bool next_dirty = !dirty;
1986 end = bdrv_dirty_bitmap_next_zero(bitmap, begin);
1988 bdrv_set_dirty_iter(it, begin);
1989 end = bdrv_dirty_iter_next(it);
1991 if (end == -1 || end - begin > UINT32_MAX) {
1992 /* Cap to an aligned value < 4G beyond begin. */
1993 end = MIN(bdrv_dirty_bitmap_size(bitmap),
1994 begin + UINT32_MAX + 1 -
1995 bdrv_dirty_bitmap_granularity(bitmap));
1998 if (dont_fragment && end > overall_end) {
2002 extents[i].length = cpu_to_be32(end - begin);
2003 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2009 bdrv_dirty_iter_free(it);
2011 bdrv_dirty_bitmap_unlock(bitmap);
2013 assert(offset < end);
2014 *length = end - offset;
2018 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2019 BdrvDirtyBitmap *bitmap, uint64_t offset,
2020 uint32_t length, bool dont_fragment, bool last,
2021 uint32_t context_id, Error **errp)
2024 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2025 NBDExtent *extents = g_new(NBDExtent, nb_extents);
2026 uint64_t final_length = length;
2028 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2029 nb_extents, dont_fragment);
2031 ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2032 final_length, last, context_id, errp);
2039 /* nbd_co_receive_request
2040 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2041 * connection right away, and any other negative value to report an error to
2042 * the client (although the caller may still need to disconnect after reporting
2045 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2048 NBDClient *client = req->client;
2051 g_assert(qemu_in_coroutine());
2052 assert(client->recv_coroutine == qemu_coroutine_self());
2053 if (nbd_receive_request(client->ioc, request, errp) < 0) {
2057 trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2058 nbd_cmd_lookup(request->type));
2060 if (request->type != NBD_CMD_WRITE) {
2061 /* No payload, we are ready to read the next request. */
2062 req->complete = true;
2065 if (request->type == NBD_CMD_DISC) {
2066 /* Special case: we're going to disconnect without a reply,
2067 * whether or not flags, from, or len are bogus */
2071 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2072 request->type == NBD_CMD_CACHE)
2074 if (request->len > NBD_MAX_BUFFER_SIZE) {
2075 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2076 request->len, NBD_MAX_BUFFER_SIZE);
2080 req->data = blk_try_blockalign(client->exp->blk, request->len);
2081 if (req->data == NULL) {
2082 error_setg(errp, "No memory");
2086 if (request->type == NBD_CMD_WRITE) {
2087 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
2088 error_prepend(errp, "reading from socket failed: ");
2091 req->complete = true;
2093 trace_nbd_co_receive_request_payload_received(request->handle,
2097 /* Sanity checks. */
2098 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2099 (request->type == NBD_CMD_WRITE ||
2100 request->type == NBD_CMD_WRITE_ZEROES ||
2101 request->type == NBD_CMD_TRIM)) {
2102 error_setg(errp, "Export is read-only");
2105 if (request->from > client->exp->size ||
2106 request->from + request->len > client->exp->size) {
2107 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2108 ", Size: %" PRIu64, request->from, request->len,
2109 (uint64_t)client->exp->size);
2110 return (request->type == NBD_CMD_WRITE ||
2111 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2113 valid_flags = NBD_CMD_FLAG_FUA;
2114 if (request->type == NBD_CMD_READ && client->structured_reply) {
2115 valid_flags |= NBD_CMD_FLAG_DF;
2116 } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2117 valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2118 } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2119 valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2121 if (request->flags & ~valid_flags) {
2122 error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2123 nbd_cmd_lookup(request->type), request->flags);
2130 /* Send simple reply without a payload, or a structured error
2131 * @error_msg is ignored if @ret >= 0
2132 * Returns 0 if connection is still live, -errno on failure to talk to client
2134 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2137 const char *error_msg,
2140 if (client->structured_reply && ret < 0) {
2141 return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2144 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2149 /* Handle NBD_CMD_READ request.
2150 * Return -errno if sending fails. Other errors are reported directly to the
2151 * client as an error reply. */
2152 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2153 uint8_t *data, Error **errp)
2156 NBDExport *exp = client->exp;
2158 assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2160 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2161 if (request->flags & NBD_CMD_FLAG_FUA) {
2162 ret = blk_co_flush(exp->blk);
2164 return nbd_send_generic_reply(client, request->handle, ret,
2165 "flush failed", errp);
2169 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2170 request->len && request->type != NBD_CMD_CACHE)
2172 return nbd_co_send_sparse_read(client, request->handle, request->from,
2173 data, request->len, errp);
2176 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2178 if (ret < 0 || request->type == NBD_CMD_CACHE) {
2179 return nbd_send_generic_reply(client, request->handle, ret,
2180 "reading from file failed", errp);
2183 if (client->structured_reply) {
2185 return nbd_co_send_structured_read(client, request->handle,
2186 request->from, data,
2187 request->len, true, errp);
2189 return nbd_co_send_structured_done(client, request->handle, errp);
2192 return nbd_co_send_simple_reply(client, request->handle, 0,
2193 data, request->len, errp);
2197 /* Handle NBD request.
2198 * Return -errno if sending fails. Other errors are reported directly to the
2199 * client as an error reply. */
2200 static coroutine_fn int nbd_handle_request(NBDClient *client,
2201 NBDRequest *request,
2202 uint8_t *data, Error **errp)
2206 NBDExport *exp = client->exp;
2209 switch (request->type) {
2212 return nbd_do_cmd_read(client, request, data, errp);
2216 if (request->flags & NBD_CMD_FLAG_FUA) {
2217 flags |= BDRV_REQ_FUA;
2219 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2220 data, request->len, flags);
2221 return nbd_send_generic_reply(client, request->handle, ret,
2222 "writing to file failed", errp);
2224 case NBD_CMD_WRITE_ZEROES:
2226 if (request->flags & NBD_CMD_FLAG_FUA) {
2227 flags |= BDRV_REQ_FUA;
2229 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2230 flags |= BDRV_REQ_MAY_UNMAP;
2232 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2233 request->len, flags);
2234 return nbd_send_generic_reply(client, request->handle, ret,
2235 "writing to file failed", errp);
2238 /* unreachable, thanks to special case in nbd_co_receive_request() */
2242 ret = blk_co_flush(exp->blk);
2243 return nbd_send_generic_reply(client, request->handle, ret,
2244 "flush failed", errp);
2247 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2249 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2250 ret = blk_co_flush(exp->blk);
2252 return nbd_send_generic_reply(client, request->handle, ret,
2253 "discard failed", errp);
2255 case NBD_CMD_BLOCK_STATUS:
2256 if (!request->len) {
2257 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2258 "need non-zero length", errp);
2260 if (client->export_meta.valid &&
2261 (client->export_meta.base_allocation ||
2262 client->export_meta.bitmap))
2264 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2266 if (client->export_meta.base_allocation) {
2267 ret = nbd_co_send_block_status(client, request->handle,
2268 blk_bs(exp->blk), request->from,
2269 request->len, dont_fragment,
2270 !client->export_meta.bitmap,
2271 NBD_META_ID_BASE_ALLOCATION,
2278 if (client->export_meta.bitmap) {
2279 ret = nbd_co_send_bitmap(client, request->handle,
2280 client->exp->export_bitmap,
2281 request->from, request->len,
2283 true, NBD_META_ID_DIRTY_BITMAP, errp);
2291 return nbd_send_generic_reply(client, request->handle, -EINVAL,
2292 "CMD_BLOCK_STATUS not negotiated",
2297 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2299 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2306 /* Owns a reference to the NBDClient passed as opaque. */
2307 static coroutine_fn void nbd_trip(void *opaque)
2309 NBDClient *client = opaque;
2310 NBDRequestData *req;
2311 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
2313 Error *local_err = NULL;
2316 if (client->closing) {
2317 nbd_client_put(client);
2321 req = nbd_request_get(client);
2322 ret = nbd_co_receive_request(req, &request, &local_err);
2323 client->recv_coroutine = NULL;
2325 if (client->closing) {
2327 * The client may be closed when we are blocked in
2328 * nbd_co_receive_request()
2333 nbd_client_receive_next_request(client);
2339 /* It wans't -EIO, so, according to nbd_co_receive_request()
2340 * semantics, we should return the error to the client. */
2341 Error *export_err = local_err;
2344 ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2345 error_get_pretty(export_err), &local_err);
2346 error_free(export_err);
2348 ret = nbd_handle_request(client, &request, req->data, &local_err);
2351 error_prepend(&local_err, "Failed to send reply: ");
2355 /* We must disconnect after NBD_CMD_WRITE if we did not
2358 if (!req->complete) {
2359 error_setg(&local_err, "Request handling failed in intermediate state");
2364 nbd_request_put(req);
2365 nbd_client_put(client);
2370 error_reportf_err(local_err, "Disconnect client, due to: ");
2372 nbd_request_put(req);
2373 client_close(client, true);
2374 nbd_client_put(client);
2377 static void nbd_client_receive_next_request(NBDClient *client)
2379 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2380 nbd_client_get(client);
2381 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2382 aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2386 static coroutine_fn void nbd_co_client_start(void *opaque)
2388 NBDClient *client = opaque;
2389 Error *local_err = NULL;
2391 qemu_co_mutex_init(&client->send_lock);
2393 if (nbd_negotiate(client, &local_err)) {
2395 error_report_err(local_err);
2397 client_close(client, false);
2401 nbd_client_receive_next_request(client);
2405 * Create a new client listener using the given channel @sioc.
2406 * Begin servicing it in a coroutine. When the connection closes, call
2407 * @close_fn with an indication of whether the client completed negotiation.
2409 void nbd_client_new(QIOChannelSocket *sioc,
2410 QCryptoTLSCreds *tlscreds,
2411 const char *tlsaclname,
2412 void (*close_fn)(NBDClient *, bool))
2417 client = g_new0(NBDClient, 1);
2418 client->refcount = 1;
2419 client->tlscreds = tlscreds;
2421 object_ref(OBJECT(client->tlscreds));
2423 client->tlsaclname = g_strdup(tlsaclname);
2424 client->sioc = sioc;
2425 object_ref(OBJECT(client->sioc));
2426 client->ioc = QIO_CHANNEL(sioc);
2427 object_ref(OBJECT(client->ioc));
2428 client->close_fn = close_fn;
2430 co = qemu_coroutine_create(nbd_co_client_start, client);
2431 qemu_coroutine_enter(co);
2434 void nbd_export_bitmap(NBDExport *exp, const char *bitmap,
2435 const char *bitmap_export_name, Error **errp)
2437 BdrvDirtyBitmap *bm = NULL;
2438 BlockDriverState *bs = blk_bs(exp->blk);
2440 if (exp->export_bitmap) {
2441 error_setg(errp, "Export bitmap is already set");
2446 bm = bdrv_find_dirty_bitmap(bs, bitmap);
2447 if (bm != NULL || bs->backing == NULL) {
2451 bs = bs->backing->bs;
2455 error_setg(errp, "Bitmap '%s' is not found", bitmap);
2459 if (bdrv_dirty_bitmap_enabled(bm)) {
2460 error_setg(errp, "Bitmap '%s' is enabled", bitmap);
2464 if (bdrv_dirty_bitmap_user_locked(bm)) {
2465 error_setg(errp, "Bitmap '%s' is in use", bitmap);
2469 bdrv_dirty_bitmap_set_qmp_locked(bm, true);
2470 exp->export_bitmap = bm;
2471 exp->export_bitmap_context =
2472 g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name);