2 * Copyright (C) 2016-2018 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
5 * Network Block Device Server Side
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; under version 2 of the License.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
23 #include "nbd-internal.h"
25 #define NBD_META_ID_BASE_ALLOCATION 0
27 static int system_errno_to_nbd_errno(int err
)
55 /* Definitions for opaque data types */
57 typedef struct NBDRequestData NBDRequestData
;
59 struct NBDRequestData
{
60 QSIMPLEQ_ENTRY(NBDRequestData
) entry
;
68 void (*close
)(NBDExport
*exp
);
76 QTAILQ_HEAD(, NBDClient
) clients
;
77 QTAILQ_ENTRY(NBDExport
) next
;
81 BlockBackend
*eject_notifier_blk
;
82 Notifier eject_notifier
;
85 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
87 /* NBDExportMetaContexts represents a list of contexts to be exported,
88 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
89 * NBD_OPT_LIST_META_CONTEXT. */
90 typedef struct NBDExportMetaContexts
{
91 char export_name
[NBD_MAX_NAME_SIZE
+ 1];
92 bool valid
; /* means that negotiation of the option finished without
94 bool base_allocation
; /* export base:allocation context (block status) */
95 } NBDExportMetaContexts
;
99 void (*close_fn
)(NBDClient
*client
, bool negotiated
);
102 QCryptoTLSCreds
*tlscreds
;
104 QIOChannelSocket
*sioc
; /* The underlying data channel */
105 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
107 Coroutine
*recv_coroutine
;
110 Coroutine
*send_coroutine
;
112 QTAILQ_ENTRY(NBDClient
) next
;
116 bool structured_reply
;
117 NBDExportMetaContexts export_meta
;
119 uint32_t opt
; /* Current option being negotiated */
120 uint32_t optlen
; /* remaining length of data in ioc for the option being
124 static void nbd_client_receive_next_request(NBDClient
*client
);
126 /* Basic flow for negotiation
153 static inline void set_be_option_rep(NBDOptionReply
*rep
, uint32_t option
,
154 uint32_t type
, uint32_t length
)
156 stq_be_p(&rep
->magic
, NBD_REP_MAGIC
);
157 stl_be_p(&rep
->option
, option
);
158 stl_be_p(&rep
->type
, type
);
159 stl_be_p(&rep
->length
, length
);
162 /* Send a reply header, including length, but no payload.
163 * Return -errno on error, 0 on success. */
164 static int nbd_negotiate_send_rep_len(NBDClient
*client
, uint32_t type
,
165 uint32_t len
, Error
**errp
)
169 trace_nbd_negotiate_send_rep_len(client
->opt
, nbd_opt_lookup(client
->opt
),
170 type
, nbd_rep_lookup(type
), len
);
172 assert(len
< NBD_MAX_BUFFER_SIZE
);
174 set_be_option_rep(&rep
, client
->opt
, type
, len
);
175 return nbd_write(client
->ioc
, &rep
, sizeof(rep
), errp
);
178 /* Send a reply header with default 0 length.
179 * Return -errno on error, 0 on success. */
180 static int nbd_negotiate_send_rep(NBDClient
*client
, uint32_t type
,
183 return nbd_negotiate_send_rep_len(client
, type
, 0, errp
);
186 /* Send an error reply.
187 * Return -errno on error, 0 on success. */
188 static int GCC_FMT_ATTR(4, 0)
189 nbd_negotiate_send_rep_verr(NBDClient
*client
, uint32_t type
,
190 Error
**errp
, const char *fmt
, va_list va
)
196 msg
= g_strdup_vprintf(fmt
, va
);
199 trace_nbd_negotiate_send_rep_err(msg
);
200 ret
= nbd_negotiate_send_rep_len(client
, type
, len
, errp
);
204 if (nbd_write(client
->ioc
, msg
, len
, errp
) < 0) {
205 error_prepend(errp
, "write failed (error message): ");
216 /* Send an error reply.
217 * Return -errno on error, 0 on success. */
218 static int GCC_FMT_ATTR(4, 5)
219 nbd_negotiate_send_rep_err(NBDClient
*client
, uint32_t type
,
220 Error
**errp
, const char *fmt
, ...)
226 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
231 /* Drop remainder of the current option, and send a reply with the
232 * given error type and message. Return -errno on read or write
233 * failure; or 0 if connection is still live. */
234 static int GCC_FMT_ATTR(4, 0)
235 nbd_opt_vdrop(NBDClient
*client
, uint32_t type
, Error
**errp
,
236 const char *fmt
, va_list va
)
238 int ret
= nbd_drop(client
->ioc
, client
->optlen
, errp
);
242 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
247 static int GCC_FMT_ATTR(4, 5)
248 nbd_opt_drop(NBDClient
*client
, uint32_t type
, Error
**errp
,
249 const char *fmt
, ...)
255 ret
= nbd_opt_vdrop(client
, type
, errp
, fmt
, va
);
261 static int GCC_FMT_ATTR(3, 4)
262 nbd_opt_invalid(NBDClient
*client
, Error
**errp
, const char *fmt
, ...)
268 ret
= nbd_opt_vdrop(client
, NBD_REP_ERR_INVALID
, errp
, fmt
, va
);
274 /* Read size bytes from the unparsed payload of the current option.
275 * Return -errno on I/O error, 0 if option was completely handled by
276 * sending a reply about inconsistent lengths, or 1 on success. */
277 static int nbd_opt_read(NBDClient
*client
, void *buffer
, size_t size
,
280 if (size
> client
->optlen
) {
281 return nbd_opt_invalid(client
, errp
,
282 "Inconsistent lengths in option %s",
283 nbd_opt_lookup(client
->opt
));
285 client
->optlen
-= size
;
286 return qio_channel_read_all(client
->ioc
, buffer
, size
, errp
) < 0 ? -EIO
: 1;
289 /* Drop size bytes from the unparsed payload of the current option.
290 * Return -errno on I/O error, 0 if option was completely handled by
291 * sending a reply about inconsistent lengths, or 1 on success. */
292 static int nbd_opt_skip(NBDClient
*client
, size_t size
, Error
**errp
)
294 if (size
> client
->optlen
) {
295 return nbd_opt_invalid(client
, errp
,
296 "Inconsistent lengths in option %s",
297 nbd_opt_lookup(client
->opt
));
299 client
->optlen
-= size
;
300 return nbd_drop(client
->ioc
, size
, errp
) < 0 ? -EIO
: 1;
305 * Read a string with the format:
306 * uint32_t len (<= NBD_MAX_NAME_SIZE)
307 * len bytes string (not 0-terminated)
309 * @name should be enough to store NBD_MAX_NAME_SIZE+1.
310 * If @length is non-null, it will be set to the actual string length.
312 * Return -errno on I/O error, 0 if option was completely handled by
313 * sending a reply about inconsistent lengths, or 1 on success.
315 static int nbd_opt_read_name(NBDClient
*client
, char *name
, uint32_t *length
,
321 ret
= nbd_opt_read(client
, &len
, sizeof(len
), errp
);
327 if (len
> NBD_MAX_NAME_SIZE
) {
328 return nbd_opt_invalid(client
, errp
,
329 "Invalid name length: %" PRIu32
, len
);
332 ret
= nbd_opt_read(client
, name
, len
, errp
);
345 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
346 * Return -errno on error, 0 on success. */
347 static int nbd_negotiate_send_rep_list(NBDClient
*client
, NBDExport
*exp
,
350 size_t name_len
, desc_len
;
352 const char *name
= exp
->name
? exp
->name
: "";
353 const char *desc
= exp
->description
? exp
->description
: "";
354 QIOChannel
*ioc
= client
->ioc
;
357 trace_nbd_negotiate_send_rep_list(name
, desc
);
358 name_len
= strlen(name
);
359 desc_len
= strlen(desc
);
360 len
= name_len
+ desc_len
+ sizeof(len
);
361 ret
= nbd_negotiate_send_rep_len(client
, NBD_REP_SERVER
, len
, errp
);
366 len
= cpu_to_be32(name_len
);
367 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
368 error_prepend(errp
, "write failed (name length): ");
372 if (nbd_write(ioc
, name
, name_len
, errp
) < 0) {
373 error_prepend(errp
, "write failed (name buffer): ");
377 if (nbd_write(ioc
, desc
, desc_len
, errp
) < 0) {
378 error_prepend(errp
, "write failed (description buffer): ");
385 /* Process the NBD_OPT_LIST command, with a potential series of replies.
386 * Return -errno on error, 0 on success. */
387 static int nbd_negotiate_handle_list(NBDClient
*client
, Error
**errp
)
390 assert(client
->opt
== NBD_OPT_LIST
);
392 /* For each export, send a NBD_REP_SERVER reply. */
393 QTAILQ_FOREACH(exp
, &exports
, next
) {
394 if (nbd_negotiate_send_rep_list(client
, exp
, errp
)) {
398 /* Finish with a NBD_REP_ACK. */
399 return nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
402 static void nbd_check_meta_export_name(NBDClient
*client
)
404 client
->export_meta
.valid
&= !strcmp(client
->exp
->name
,
405 client
->export_meta
.export_name
);
408 /* Send a reply to NBD_OPT_EXPORT_NAME.
409 * Return -errno on error, 0 on success. */
410 static int nbd_negotiate_handle_export_name(NBDClient
*client
,
411 uint16_t myflags
, bool no_zeroes
,
414 char name
[NBD_MAX_NAME_SIZE
+ 1];
415 char buf
[NBD_REPLY_EXPORT_NAME_SIZE
] = "";
420 [20 .. xx] export name (length bytes)
423 [ 8 .. 9] export flags
424 [10 .. 133] reserved (0) [unless no_zeroes]
426 trace_nbd_negotiate_handle_export_name();
427 if (client
->optlen
>= sizeof(name
)) {
428 error_setg(errp
, "Bad length received");
431 if (nbd_read(client
->ioc
, name
, client
->optlen
, errp
) < 0) {
432 error_prepend(errp
, "read failed: ");
435 name
[client
->optlen
] = '\0';
438 trace_nbd_negotiate_handle_export_name_request(name
);
440 client
->exp
= nbd_export_find(name
);
442 error_setg(errp
, "export not found");
446 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
,
447 client
->exp
->nbdflags
| myflags
);
448 stq_be_p(buf
, client
->exp
->size
);
449 stw_be_p(buf
+ 8, client
->exp
->nbdflags
| myflags
);
450 len
= no_zeroes
? 10 : sizeof(buf
);
451 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
453 error_prepend(errp
, "write failed: ");
457 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
458 nbd_export_get(client
->exp
);
459 nbd_check_meta_export_name(client
);
464 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
465 * The buffer does NOT include the info type prefix.
466 * Return -errno on error, 0 if ready to send more. */
467 static int nbd_negotiate_send_info(NBDClient
*client
,
468 uint16_t info
, uint32_t length
, void *buf
,
473 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
474 rc
= nbd_negotiate_send_rep_len(client
, NBD_REP_INFO
,
475 sizeof(info
) + length
, errp
);
480 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
483 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
489 /* nbd_reject_length: Handle any unexpected payload.
490 * @fatal requests that we quit talking to the client, even if we are able
491 * to successfully send an error reply.
493 * -errno transmission error occurred or @fatal was requested, errp is set
494 * 0 error message successfully sent to client, errp is not set
496 static int nbd_reject_length(NBDClient
*client
, bool fatal
, Error
**errp
)
500 assert(client
->optlen
);
501 ret
= nbd_opt_invalid(client
, errp
, "option '%s' has unexpected length",
502 nbd_opt_lookup(client
->opt
));
504 error_setg(errp
, "option '%s' has unexpected length",
505 nbd_opt_lookup(client
->opt
));
511 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
512 * Return -errno on error, 0 if ready for next option, and 1 to move
513 * into transmission phase. */
514 static int nbd_negotiate_handle_info(NBDClient
*client
, uint16_t myflags
,
518 char name
[NBD_MAX_NAME_SIZE
+ 1];
523 bool sendname
= false;
524 bool blocksize
= false;
526 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
529 4 bytes: L, name length (can be 0)
531 2 bytes: N, number of requests (can be 0)
532 N * 2 bytes: N requests
534 rc
= nbd_opt_read_name(client
, name
, &namelen
, errp
);
538 trace_nbd_negotiate_handle_export_name_request(name
);
540 rc
= nbd_opt_read(client
, &requests
, sizeof(requests
), errp
);
544 be16_to_cpus(&requests
);
545 trace_nbd_negotiate_handle_info_requests(requests
);
547 rc
= nbd_opt_read(client
, &request
, sizeof(request
), errp
);
551 be16_to_cpus(&request
);
552 trace_nbd_negotiate_handle_info_request(request
,
553 nbd_info_lookup(request
));
554 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
555 * everything else is either a request we don't know or
556 * something we send regardless of request */
561 case NBD_INFO_BLOCK_SIZE
:
566 if (client
->optlen
) {
567 return nbd_reject_length(client
, false, errp
);
570 exp
= nbd_export_find(name
);
572 return nbd_negotiate_send_rep_err(client
, NBD_REP_ERR_UNKNOWN
,
573 errp
, "export '%s' not present",
577 /* Don't bother sending NBD_INFO_NAME unless client requested it */
579 rc
= nbd_negotiate_send_info(client
, NBD_INFO_NAME
, namelen
, name
,
586 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
588 if (exp
->description
) {
589 size_t len
= strlen(exp
->description
);
591 rc
= nbd_negotiate_send_info(client
, NBD_INFO_DESCRIPTION
,
592 len
, exp
->description
, errp
);
598 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
599 * according to whether the client requested it, and according to
600 * whether this is OPT_INFO or OPT_GO. */
601 /* minimum - 1 for back-compat, or 512 if client is new enough.
602 * TODO: consult blk_bs(blk)->bl.request_alignment? */
604 (client
->opt
== NBD_OPT_INFO
|| blocksize
) ? BDRV_SECTOR_SIZE
: 1;
605 /* preferred - Hard-code to 4096 for now.
606 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
608 /* maximum - At most 32M, but smaller as appropriate. */
609 sizes
[2] = MIN(blk_get_max_transfer(exp
->blk
), NBD_MAX_BUFFER_SIZE
);
610 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
611 cpu_to_be32s(&sizes
[0]);
612 cpu_to_be32s(&sizes
[1]);
613 cpu_to_be32s(&sizes
[2]);
614 rc
= nbd_negotiate_send_info(client
, NBD_INFO_BLOCK_SIZE
,
615 sizeof(sizes
), sizes
, errp
);
620 /* Send NBD_INFO_EXPORT always */
621 trace_nbd_negotiate_new_style_size_flags(exp
->size
,
622 exp
->nbdflags
| myflags
);
623 stq_be_p(buf
, exp
->size
);
624 stw_be_p(buf
+ 8, exp
->nbdflags
| myflags
);
625 rc
= nbd_negotiate_send_info(client
, NBD_INFO_EXPORT
,
626 sizeof(buf
), buf
, errp
);
631 /* If the client is just asking for NBD_OPT_INFO, but forgot to
632 * request block sizes, return an error.
633 * TODO: consult blk_bs(blk)->request_align, and only error if it
635 if (client
->opt
== NBD_OPT_INFO
&& !blocksize
) {
636 return nbd_negotiate_send_rep_err(client
,
637 NBD_REP_ERR_BLOCK_SIZE_REQD
,
639 "request NBD_INFO_BLOCK_SIZE to "
644 rc
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
649 if (client
->opt
== NBD_OPT_GO
) {
651 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
652 nbd_export_get(client
->exp
);
653 nbd_check_meta_export_name(client
);
660 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
661 * new channel for all further (now-encrypted) communication. */
662 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
667 struct NBDTLSHandshakeData data
= { 0 };
669 assert(client
->opt
== NBD_OPT_STARTTLS
);
671 trace_nbd_negotiate_handle_starttls();
674 if (nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
) < 0) {
678 tioc
= qio_channel_tls_new_server(ioc
,
686 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
687 trace_nbd_negotiate_handle_starttls_handshake();
688 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
689 qio_channel_tls_handshake(tioc
,
695 if (!data
.complete
) {
696 g_main_loop_run(data
.loop
);
698 g_main_loop_unref(data
.loop
);
700 object_unref(OBJECT(tioc
));
701 error_propagate(errp
, data
.error
);
705 return QIO_CHANNEL(tioc
);
708 /* nbd_negotiate_send_meta_context
710 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
712 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
714 static int nbd_negotiate_send_meta_context(NBDClient
*client
,
719 NBDOptionReplyMetaContext opt
;
720 struct iovec iov
[] = {
721 {.iov_base
= &opt
, .iov_len
= sizeof(opt
)},
722 {.iov_base
= (void *)context
, .iov_len
= strlen(context
)}
725 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
729 trace_nbd_negotiate_meta_query_reply(context
, context_id
);
730 set_be_option_rep(&opt
.h
, client
->opt
, NBD_REP_META_CONTEXT
,
731 sizeof(opt
) - sizeof(opt
.h
) + iov
[1].iov_len
);
732 stl_be_p(&opt
.context_id
, context_id
);
734 return qio_channel_writev_all(client
->ioc
, iov
, 2, errp
) < 0 ? -EIO
: 0;
737 /* nbd_meta_base_query
739 * Handle queries to 'base' namespace. For now, only the base:allocation
740 * context is available. 'len' is the amount of text remaining to be read from
741 * the current name, after the 'base:' portion has been stripped.
743 * Return -errno on I/O error, 0 if option was completely handled by
744 * sending a reply about inconsistent lengths, or 1 on success.
746 * Note: return code = 1 doesn't mean that we've parsed the "base:allocation"
747 * namespace. It only means that there are no errors.
749 static int nbd_meta_base_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
750 uint32_t len
, Error
**errp
)
753 char query
[sizeof("allocation") - 1];
754 size_t alen
= strlen("allocation");
757 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
758 meta
->base_allocation
= true;
760 trace_nbd_negotiate_meta_query_parse("base:");
765 trace_nbd_negotiate_meta_query_skip("not base:allocation");
766 return nbd_opt_skip(client
, len
, errp
);
769 ret
= nbd_opt_read(client
, query
, len
, errp
);
774 if (strncmp(query
, "allocation", alen
) == 0) {
775 trace_nbd_negotiate_meta_query_parse("base:allocation");
776 meta
->base_allocation
= true;
778 trace_nbd_negotiate_meta_query_skip("not base:allocation");
784 /* nbd_negotiate_meta_query
786 * Parse namespace name and call corresponding function to parse body of the
789 * The only supported namespace now is 'base'.
791 * The function aims not wasting time and memory to read long unknown namespace
794 * Return -errno on I/O error, 0 if option was completely handled by
795 * sending a reply about inconsistent lengths, or 1 on success. */
796 static int nbd_negotiate_meta_query(NBDClient
*client
,
797 NBDExportMetaContexts
*meta
, Error
**errp
)
800 char query
[sizeof("base:") - 1];
801 size_t baselen
= strlen("base:");
804 ret
= nbd_opt_read(client
, &len
, sizeof(len
), errp
);
810 /* The only supported namespace for now is 'base'. So query should start
811 * with 'base:'. Otherwise, we can ignore it and skip the remainder. */
813 trace_nbd_negotiate_meta_query_skip("length too short");
814 return nbd_opt_skip(client
, len
, errp
);
818 ret
= nbd_opt_read(client
, query
, baselen
, errp
);
822 if (strncmp(query
, "base:", baselen
) != 0) {
823 trace_nbd_negotiate_meta_query_skip("not for base: namespace");
824 return nbd_opt_skip(client
, len
, errp
);
827 return nbd_meta_base_query(client
, meta
, len
, errp
);
830 /* nbd_negotiate_meta_queries
831 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
833 * Return -errno on I/O error, or 0 if option was completely handled. */
834 static int nbd_negotiate_meta_queries(NBDClient
*client
,
835 NBDExportMetaContexts
*meta
, Error
**errp
)
839 NBDExportMetaContexts local_meta
;
843 if (!client
->structured_reply
) {
844 return nbd_opt_invalid(client
, errp
,
845 "request option '%s' when structured reply "
847 nbd_opt_lookup(client
->opt
));
850 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
851 /* Only change the caller's meta on SET. */
855 memset(meta
, 0, sizeof(*meta
));
857 ret
= nbd_opt_read_name(client
, meta
->export_name
, NULL
, errp
);
862 exp
= nbd_export_find(meta
->export_name
);
864 return nbd_opt_drop(client
, NBD_REP_ERR_UNKNOWN
, errp
,
865 "export '%s' not present", meta
->export_name
);
868 ret
= nbd_opt_read(client
, &nb_queries
, sizeof(nb_queries
), errp
);
872 cpu_to_be32s(&nb_queries
);
873 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client
->opt
),
874 meta
->export_name
, nb_queries
);
876 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
&& !nb_queries
) {
877 /* enable all known contexts */
878 meta
->base_allocation
= true;
880 for (i
= 0; i
< nb_queries
; ++i
) {
881 ret
= nbd_negotiate_meta_query(client
, meta
, errp
);
888 if (meta
->base_allocation
) {
889 ret
= nbd_negotiate_send_meta_context(client
, "base:allocation",
890 NBD_META_ID_BASE_ALLOCATION
,
897 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
905 /* nbd_negotiate_options
906 * Process all NBD_OPT_* client option commands, during fixed newstyle
909 * -errno on error, errp is set
910 * 0 on successful negotiation, errp is not set
911 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
914 static int nbd_negotiate_options(NBDClient
*client
, uint16_t myflags
,
918 bool fixedNewstyle
= false;
919 bool no_zeroes
= false;
922 [ 0 .. 3] client flags
924 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
925 [ 0 .. 7] NBD_OPTS_MAGIC
926 [ 8 .. 11] NBD option
927 [12 .. 15] Data length
930 [ 0 .. 7] NBD_OPTS_MAGIC
931 [ 8 .. 11] Second NBD option
932 [12 .. 15] Data length
936 if (nbd_read(client
->ioc
, &flags
, sizeof(flags
), errp
) < 0) {
937 error_prepend(errp
, "read failed: ");
940 be32_to_cpus(&flags
);
941 trace_nbd_negotiate_options_flags(flags
);
942 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
943 fixedNewstyle
= true;
944 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
946 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
948 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
951 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
957 uint32_t option
, length
;
960 if (nbd_read(client
->ioc
, &magic
, sizeof(magic
), errp
) < 0) {
961 error_prepend(errp
, "read failed: ");
964 magic
= be64_to_cpu(magic
);
965 trace_nbd_negotiate_options_check_magic(magic
);
966 if (magic
!= NBD_OPTS_MAGIC
) {
967 error_setg(errp
, "Bad magic received");
971 if (nbd_read(client
->ioc
, &option
,
972 sizeof(option
), errp
) < 0) {
973 error_prepend(errp
, "read failed: ");
976 option
= be32_to_cpu(option
);
977 client
->opt
= option
;
979 if (nbd_read(client
->ioc
, &length
, sizeof(length
), errp
) < 0) {
980 error_prepend(errp
, "read failed: ");
983 length
= be32_to_cpu(length
);
984 assert(!client
->optlen
);
985 client
->optlen
= length
;
987 if (length
> NBD_MAX_BUFFER_SIZE
) {
988 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
989 length
, NBD_MAX_BUFFER_SIZE
);
993 trace_nbd_negotiate_options_check_option(option
,
994 nbd_opt_lookup(option
));
995 if (client
->tlscreds
&&
996 client
->ioc
== (QIOChannel
*)client
->sioc
) {
998 if (!fixedNewstyle
) {
999 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
1003 case NBD_OPT_STARTTLS
:
1005 /* Unconditionally drop the connection if the client
1006 * can't start a TLS negotiation correctly */
1007 return nbd_reject_length(client
, true, errp
);
1009 tioc
= nbd_negotiate_handle_starttls(client
, errp
);
1014 object_unref(OBJECT(client
->ioc
));
1015 client
->ioc
= QIO_CHANNEL(tioc
);
1018 case NBD_OPT_EXPORT_NAME
:
1019 /* No way to return an error to client, so drop connection */
1020 error_setg(errp
, "Option 0x%x not permitted before TLS",
1025 ret
= nbd_opt_drop(client
, NBD_REP_ERR_TLS_REQD
, errp
,
1027 "not permitted before TLS", option
);
1028 /* Let the client keep trying, unless they asked to
1029 * quit. In this mode, we've already sent an error, so
1030 * we can't ack the abort. */
1031 if (option
== NBD_OPT_ABORT
) {
1036 } else if (fixedNewstyle
) {
1040 ret
= nbd_reject_length(client
, false, errp
);
1042 ret
= nbd_negotiate_handle_list(client
, errp
);
1047 /* NBD spec says we must try to reply before
1048 * disconnecting, but that we must also tolerate
1049 * guests that don't wait for our reply. */
1050 nbd_negotiate_send_rep(client
, NBD_REP_ACK
, NULL
);
1053 case NBD_OPT_EXPORT_NAME
:
1054 return nbd_negotiate_handle_export_name(client
,
1060 ret
= nbd_negotiate_handle_info(client
, myflags
, errp
);
1062 assert(option
== NBD_OPT_GO
);
1067 case NBD_OPT_STARTTLS
:
1069 ret
= nbd_reject_length(client
, false, errp
);
1070 } else if (client
->tlscreds
) {
1071 ret
= nbd_negotiate_send_rep_err(client
,
1072 NBD_REP_ERR_INVALID
, errp
,
1073 "TLS already enabled");
1075 ret
= nbd_negotiate_send_rep_err(client
,
1076 NBD_REP_ERR_POLICY
, errp
,
1077 "TLS not configured");
1081 case NBD_OPT_STRUCTURED_REPLY
:
1083 ret
= nbd_reject_length(client
, false, errp
);
1084 } else if (client
->structured_reply
) {
1085 ret
= nbd_negotiate_send_rep_err(
1086 client
, NBD_REP_ERR_INVALID
, errp
,
1087 "structured reply already negotiated");
1089 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1090 client
->structured_reply
= true;
1091 myflags
|= NBD_FLAG_SEND_DF
;
1095 case NBD_OPT_LIST_META_CONTEXT
:
1096 case NBD_OPT_SET_META_CONTEXT
:
1097 ret
= nbd_negotiate_meta_queries(client
, &client
->export_meta
,
1102 ret
= nbd_opt_drop(client
, NBD_REP_ERR_UNSUP
, errp
,
1103 "Unsupported option %" PRIu32
" (%s)",
1104 option
, nbd_opt_lookup(option
));
1109 * If broken new-style we should drop the connection
1110 * for anything except NBD_OPT_EXPORT_NAME
1113 case NBD_OPT_EXPORT_NAME
:
1114 return nbd_negotiate_handle_export_name(client
,
1119 error_setg(errp
, "Unsupported option %" PRIu32
" (%s)",
1120 option
, nbd_opt_lookup(option
));
1132 * -errno on error, errp is set
1133 * 0 on successful negotiation, errp is not set
1134 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1137 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
1139 char buf
[NBD_OLDSTYLE_NEGOTIATE_SIZE
] = "";
1141 const uint16_t myflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
1142 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
|
1143 NBD_FLAG_SEND_WRITE_ZEROES
);
1146 /* Old style negotiation header, no room for options
1147 [ 0 .. 7] passwd ("NBDMAGIC")
1148 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1150 [24 .. 27] export flags (zero-extended)
1151 [28 .. 151] reserved (0)
1153 New style negotiation header, client can send options
1154 [ 0 .. 7] passwd ("NBDMAGIC")
1155 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1156 [16 .. 17] server flags (0)
1157 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1160 qio_channel_set_blocking(client
->ioc
, false, NULL
);
1162 trace_nbd_negotiate_begin();
1163 memcpy(buf
, "NBDMAGIC", 8);
1165 oldStyle
= client
->exp
!= NULL
&& !client
->tlscreds
;
1167 trace_nbd_negotiate_old_style(client
->exp
->size
,
1168 client
->exp
->nbdflags
| myflags
);
1169 stq_be_p(buf
+ 8, NBD_CLIENT_MAGIC
);
1170 stq_be_p(buf
+ 16, client
->exp
->size
);
1171 stl_be_p(buf
+ 24, client
->exp
->nbdflags
| myflags
);
1173 if (nbd_write(client
->ioc
, buf
, sizeof(buf
), errp
) < 0) {
1174 error_prepend(errp
, "write failed: ");
1178 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
1179 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
1181 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
1182 error_prepend(errp
, "write failed: ");
1185 ret
= nbd_negotiate_options(client
, myflags
, errp
);
1188 error_prepend(errp
, "option negotiation failed: ");
1194 assert(!client
->optlen
);
1195 trace_nbd_negotiate_success();
1200 static int nbd_receive_request(QIOChannel
*ioc
, NBDRequest
*request
,
1203 uint8_t buf
[NBD_REQUEST_SIZE
];
1207 ret
= nbd_read(ioc
, buf
, sizeof(buf
), errp
);
1213 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1214 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1215 [ 6 .. 7] type (NBD_CMD_READ, ...)
1221 magic
= ldl_be_p(buf
);
1222 request
->flags
= lduw_be_p(buf
+ 4);
1223 request
->type
= lduw_be_p(buf
+ 6);
1224 request
->handle
= ldq_be_p(buf
+ 8);
1225 request
->from
= ldq_be_p(buf
+ 16);
1226 request
->len
= ldl_be_p(buf
+ 24);
1228 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
1229 request
->from
, request
->len
);
1231 if (magic
!= NBD_REQUEST_MAGIC
) {
1232 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
1238 #define MAX_NBD_REQUESTS 16
1240 void nbd_client_get(NBDClient
*client
)
1245 void nbd_client_put(NBDClient
*client
)
1247 if (--client
->refcount
== 0) {
1248 /* The last reference should be dropped by client->close,
1249 * which is called by client_close.
1251 assert(client
->closing
);
1253 qio_channel_detach_aio_context(client
->ioc
);
1254 object_unref(OBJECT(client
->sioc
));
1255 object_unref(OBJECT(client
->ioc
));
1256 if (client
->tlscreds
) {
1257 object_unref(OBJECT(client
->tlscreds
));
1259 g_free(client
->tlsaclname
);
1261 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
1262 nbd_export_put(client
->exp
);
1268 static void client_close(NBDClient
*client
, bool negotiated
)
1270 if (client
->closing
) {
1274 client
->closing
= true;
1276 /* Force requests to finish. They will drop their own references,
1277 * then we'll close the socket and free the NBDClient.
1279 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
1282 /* Also tell the client, so that they release their reference. */
1283 if (client
->close_fn
) {
1284 client
->close_fn(client
, negotiated
);
1288 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
1290 NBDRequestData
*req
;
1292 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
1293 client
->nb_requests
++;
1295 req
= g_new0(NBDRequestData
, 1);
1296 nbd_client_get(client
);
1297 req
->client
= client
;
1301 static void nbd_request_put(NBDRequestData
*req
)
1303 NBDClient
*client
= req
->client
;
1306 qemu_vfree(req
->data
);
1310 client
->nb_requests
--;
1311 nbd_client_receive_next_request(client
);
1313 nbd_client_put(client
);
1316 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1318 NBDExport
*exp
= opaque
;
1321 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1325 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1326 qio_channel_attach_aio_context(client
->ioc
, ctx
);
1327 if (client
->recv_coroutine
) {
1328 aio_co_schedule(ctx
, client
->recv_coroutine
);
1330 if (client
->send_coroutine
) {
1331 aio_co_schedule(ctx
, client
->send_coroutine
);
1336 static void blk_aio_detach(void *opaque
)
1338 NBDExport
*exp
= opaque
;
1341 trace_nbd_blk_aio_detach(exp
->name
, exp
->ctx
);
1343 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1344 qio_channel_detach_aio_context(client
->ioc
);
1350 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1352 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1353 nbd_export_close(exp
);
1356 NBDExport
*nbd_export_new(BlockDriverState
*bs
, off_t dev_offset
, off_t size
,
1357 uint16_t nbdflags
, void (*close
)(NBDExport
*),
1358 bool writethrough
, BlockBackend
*on_eject_blk
,
1363 NBDExport
*exp
= g_new0(NBDExport
, 1);
1368 * NBD exports are used for non-shared storage migration. Make sure
1369 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1370 * access since the export could be available before migration handover.
1372 ctx
= bdrv_get_aio_context(bs
);
1373 aio_context_acquire(ctx
);
1374 bdrv_invalidate_cache(bs
, NULL
);
1375 aio_context_release(ctx
);
1377 /* Don't allow resize while the NBD server is running, otherwise we don't
1378 * care what happens with the node. */
1379 perm
= BLK_PERM_CONSISTENT_READ
;
1380 if ((nbdflags
& NBD_FLAG_READ_ONLY
) == 0) {
1381 perm
|= BLK_PERM_WRITE
;
1383 blk
= blk_new(perm
, BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
|
1384 BLK_PERM_WRITE
| BLK_PERM_GRAPH_MOD
);
1385 ret
= blk_insert_bs(blk
, bs
, errp
);
1389 blk_set_enable_write_cache(blk
, !writethrough
);
1392 QTAILQ_INIT(&exp
->clients
);
1394 exp
->dev_offset
= dev_offset
;
1395 exp
->nbdflags
= nbdflags
;
1396 exp
->size
= size
< 0 ? blk_getlength(blk
) : size
;
1397 if (exp
->size
< 0) {
1398 error_setg_errno(errp
, -exp
->size
,
1399 "Failed to determine the NBD export's length");
1402 exp
->size
-= exp
->size
% BDRV_SECTOR_SIZE
;
1405 exp
->ctx
= blk_get_aio_context(blk
);
1406 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1409 blk_ref(on_eject_blk
);
1410 exp
->eject_notifier_blk
= on_eject_blk
;
1411 exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1412 blk_add_remove_bs_notifier(on_eject_blk
, &exp
->eject_notifier
);
1422 NBDExport
*nbd_export_find(const char *name
)
1425 QTAILQ_FOREACH(exp
, &exports
, next
) {
1426 if (strcmp(name
, exp
->name
) == 0) {
1434 void nbd_export_set_name(NBDExport
*exp
, const char *name
)
1436 if (exp
->name
== name
) {
1440 nbd_export_get(exp
);
1441 if (exp
->name
!= NULL
) {
1444 QTAILQ_REMOVE(&exports
, exp
, next
);
1445 nbd_export_put(exp
);
1448 nbd_export_get(exp
);
1449 exp
->name
= g_strdup(name
);
1450 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1452 nbd_export_put(exp
);
1455 void nbd_export_set_description(NBDExport
*exp
, const char *description
)
1457 g_free(exp
->description
);
1458 exp
->description
= g_strdup(description
);
1461 void nbd_export_close(NBDExport
*exp
)
1463 NBDClient
*client
, *next
;
1465 nbd_export_get(exp
);
1466 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1467 client_close(client
, true);
1469 nbd_export_set_name(exp
, NULL
);
1470 nbd_export_set_description(exp
, NULL
);
1471 nbd_export_put(exp
);
1474 void nbd_export_remove(NBDExport
*exp
, NbdServerRemoveMode mode
, Error
**errp
)
1476 if (mode
== NBD_SERVER_REMOVE_MODE_HARD
|| QTAILQ_EMPTY(&exp
->clients
)) {
1477 nbd_export_close(exp
);
1481 assert(mode
== NBD_SERVER_REMOVE_MODE_SAFE
);
1483 error_setg(errp
, "export '%s' still in use", exp
->name
);
1484 error_append_hint(errp
, "Use mode='hard' to force client disconnect\n");
1487 void nbd_export_get(NBDExport
*exp
)
1489 assert(exp
->refcount
> 0);
1493 void nbd_export_put(NBDExport
*exp
)
1495 assert(exp
->refcount
> 0);
1496 if (exp
->refcount
== 1) {
1497 nbd_export_close(exp
);
1500 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1501 * if someone calls nbd_export_put() on named export not through
1502 * nbd_export_set_name() when refcount is 1. So, let's assert that
1505 assert(exp
->refcount
> 0);
1506 if (--exp
->refcount
== 0) {
1507 assert(exp
->name
== NULL
);
1508 assert(exp
->description
== NULL
);
1515 if (exp
->eject_notifier_blk
) {
1516 notifier_remove(&exp
->eject_notifier
);
1517 blk_unref(exp
->eject_notifier_blk
);
1519 blk_remove_aio_context_notifier(exp
->blk
, blk_aio_attached
,
1520 blk_aio_detach
, exp
);
1521 blk_unref(exp
->blk
);
1529 BlockBackend
*nbd_export_get_blockdev(NBDExport
*exp
)
1534 void nbd_export_close_all(void)
1536 NBDExport
*exp
, *next
;
1538 QTAILQ_FOREACH_SAFE(exp
, &exports
, next
, next
) {
1539 nbd_export_close(exp
);
1543 static int coroutine_fn
nbd_co_send_iov(NBDClient
*client
, struct iovec
*iov
,
1544 unsigned niov
, Error
**errp
)
1548 g_assert(qemu_in_coroutine());
1549 qemu_co_mutex_lock(&client
->send_lock
);
1550 client
->send_coroutine
= qemu_coroutine_self();
1552 ret
= qio_channel_writev_all(client
->ioc
, iov
, niov
, errp
) < 0 ? -EIO
: 0;
1554 client
->send_coroutine
= NULL
;
1555 qemu_co_mutex_unlock(&client
->send_lock
);
1560 static inline void set_be_simple_reply(NBDSimpleReply
*reply
, uint64_t error
,
1563 stl_be_p(&reply
->magic
, NBD_SIMPLE_REPLY_MAGIC
);
1564 stl_be_p(&reply
->error
, error
);
1565 stq_be_p(&reply
->handle
, handle
);
1568 static int nbd_co_send_simple_reply(NBDClient
*client
,
1575 NBDSimpleReply reply
;
1576 int nbd_err
= system_errno_to_nbd_errno(error
);
1577 struct iovec iov
[] = {
1578 {.iov_base
= &reply
, .iov_len
= sizeof(reply
)},
1579 {.iov_base
= data
, .iov_len
= len
}
1582 trace_nbd_co_send_simple_reply(handle
, nbd_err
, nbd_err_lookup(nbd_err
),
1584 set_be_simple_reply(&reply
, nbd_err
, handle
);
1586 return nbd_co_send_iov(client
, iov
, len
? 2 : 1, errp
);
1589 static inline void set_be_chunk(NBDStructuredReplyChunk
*chunk
, uint16_t flags
,
1590 uint16_t type
, uint64_t handle
, uint32_t length
)
1592 stl_be_p(&chunk
->magic
, NBD_STRUCTURED_REPLY_MAGIC
);
1593 stw_be_p(&chunk
->flags
, flags
);
1594 stw_be_p(&chunk
->type
, type
);
1595 stq_be_p(&chunk
->handle
, handle
);
1596 stl_be_p(&chunk
->length
, length
);
1599 static int coroutine_fn
nbd_co_send_structured_done(NBDClient
*client
,
1603 NBDStructuredReplyChunk chunk
;
1604 struct iovec iov
[] = {
1605 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1608 trace_nbd_co_send_structured_done(handle
);
1609 set_be_chunk(&chunk
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_NONE
, handle
, 0);
1611 return nbd_co_send_iov(client
, iov
, 1, errp
);
1614 static int coroutine_fn
nbd_co_send_structured_read(NBDClient
*client
,
1622 NBDStructuredReadData chunk
;
1623 struct iovec iov
[] = {
1624 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1625 {.iov_base
= data
, .iov_len
= size
}
1629 trace_nbd_co_send_structured_read(handle
, offset
, data
, size
);
1630 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1631 NBD_REPLY_TYPE_OFFSET_DATA
, handle
,
1632 sizeof(chunk
) - sizeof(chunk
.h
) + size
);
1633 stq_be_p(&chunk
.offset
, offset
);
1635 return nbd_co_send_iov(client
, iov
, 2, errp
);
1638 static int coroutine_fn
nbd_co_send_structured_error(NBDClient
*client
,
1644 NBDStructuredError chunk
;
1645 int nbd_err
= system_errno_to_nbd_errno(error
);
1646 struct iovec iov
[] = {
1647 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1648 {.iov_base
= (char *)msg
, .iov_len
= msg
? strlen(msg
) : 0},
1652 trace_nbd_co_send_structured_error(handle
, nbd_err
,
1653 nbd_err_lookup(nbd_err
), msg
? msg
: "");
1654 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_ERROR
, handle
,
1655 sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1656 stl_be_p(&chunk
.error
, nbd_err
);
1657 stw_be_p(&chunk
.message_length
, iov
[1].iov_len
);
1659 return nbd_co_send_iov(client
, iov
, 1 + !!iov
[1].iov_len
, errp
);
1662 /* Do a sparse read and send the structured reply to the client.
1663 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1664 * reported to the client, at which point this function succeeds.
1666 static int coroutine_fn
nbd_co_send_sparse_read(NBDClient
*client
,
1674 NBDExport
*exp
= client
->exp
;
1675 size_t progress
= 0;
1677 while (progress
< size
) {
1679 int status
= bdrv_block_status_above(blk_bs(exp
->blk
), NULL
,
1681 size
- progress
, &pnum
, NULL
,
1686 char *msg
= g_strdup_printf("unable to check for holes: %s",
1689 ret
= nbd_co_send_structured_error(client
, handle
, -status
, msg
,
1694 assert(pnum
&& pnum
<= size
- progress
);
1695 final
= progress
+ pnum
== size
;
1696 if (status
& BDRV_BLOCK_ZERO
) {
1697 NBDStructuredReadHole chunk
;
1698 struct iovec iov
[] = {
1699 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1702 trace_nbd_co_send_structured_read_hole(handle
, offset
+ progress
,
1704 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1705 NBD_REPLY_TYPE_OFFSET_HOLE
,
1706 handle
, sizeof(chunk
) - sizeof(chunk
.h
));
1707 stq_be_p(&chunk
.offset
, offset
+ progress
);
1708 stl_be_p(&chunk
.length
, pnum
);
1709 ret
= nbd_co_send_iov(client
, iov
, 1, errp
);
1711 ret
= blk_pread(exp
->blk
, offset
+ progress
+ exp
->dev_offset
,
1712 data
+ progress
, pnum
);
1714 error_setg_errno(errp
, -ret
, "reading from file failed");
1717 ret
= nbd_co_send_structured_read(client
, handle
, offset
+ progress
,
1718 data
+ progress
, pnum
, final
,
1730 static int blockstatus_to_extent_be(BlockDriverState
*bs
, uint64_t offset
,
1731 uint64_t bytes
, NBDExtent
*extent
)
1733 uint64_t remaining_bytes
= bytes
;
1735 while (remaining_bytes
) {
1738 int ret
= bdrv_block_status_above(bs
, NULL
, offset
, remaining_bytes
,
1744 flags
= (ret
& BDRV_BLOCK_ALLOCATED
? 0 : NBD_STATE_HOLE
) |
1745 (ret
& BDRV_BLOCK_ZERO
? NBD_STATE_ZERO
: 0);
1747 if (remaining_bytes
== bytes
) {
1748 extent
->flags
= flags
;
1751 if (flags
!= extent
->flags
) {
1756 remaining_bytes
-= num
;
1759 cpu_to_be32s(&extent
->flags
);
1760 extent
->length
= cpu_to_be32(bytes
- remaining_bytes
);
1765 /* nbd_co_send_extents
1766 * @extents should be in big-endian */
1767 static int nbd_co_send_extents(NBDClient
*client
, uint64_t handle
,
1768 NBDExtent
*extents
, unsigned nb_extents
,
1769 uint32_t context_id
, Error
**errp
)
1771 NBDStructuredMeta chunk
;
1773 struct iovec iov
[] = {
1774 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1775 {.iov_base
= extents
, .iov_len
= nb_extents
* sizeof(extents
[0])}
1778 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_BLOCK_STATUS
,
1779 handle
, sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1780 stl_be_p(&chunk
.context_id
, context_id
);
1782 return nbd_co_send_iov(client
, iov
, 2, errp
);
1785 /* Get block status from the exported device and send it to the client */
1786 static int nbd_co_send_block_status(NBDClient
*client
, uint64_t handle
,
1787 BlockDriverState
*bs
, uint64_t offset
,
1788 uint64_t length
, uint32_t context_id
,
1794 ret
= blockstatus_to_extent_be(bs
, offset
, length
, &extent
);
1796 return nbd_co_send_structured_error(
1797 client
, handle
, -ret
, "can't get block status", errp
);
1800 return nbd_co_send_extents(client
, handle
, &extent
, 1, context_id
, errp
);
1803 /* nbd_co_receive_request
1804 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1805 * connection right away, and any other negative value to report an error to
1806 * the client (although the caller may still need to disconnect after reporting
1809 static int nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
1812 NBDClient
*client
= req
->client
;
1815 g_assert(qemu_in_coroutine());
1816 assert(client
->recv_coroutine
== qemu_coroutine_self());
1817 if (nbd_receive_request(client
->ioc
, request
, errp
) < 0) {
1821 trace_nbd_co_receive_request_decode_type(request
->handle
, request
->type
,
1822 nbd_cmd_lookup(request
->type
));
1824 if (request
->type
!= NBD_CMD_WRITE
) {
1825 /* No payload, we are ready to read the next request. */
1826 req
->complete
= true;
1829 if (request
->type
== NBD_CMD_DISC
) {
1830 /* Special case: we're going to disconnect without a reply,
1831 * whether or not flags, from, or len are bogus */
1835 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
) {
1836 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
1837 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1838 request
->len
, NBD_MAX_BUFFER_SIZE
);
1842 req
->data
= blk_try_blockalign(client
->exp
->blk
, request
->len
);
1843 if (req
->data
== NULL
) {
1844 error_setg(errp
, "No memory");
1848 if (request
->type
== NBD_CMD_WRITE
) {
1849 if (nbd_read(client
->ioc
, req
->data
, request
->len
, errp
) < 0) {
1850 error_prepend(errp
, "reading from socket failed: ");
1853 req
->complete
= true;
1855 trace_nbd_co_receive_request_payload_received(request
->handle
,
1859 /* Sanity checks. */
1860 if (client
->exp
->nbdflags
& NBD_FLAG_READ_ONLY
&&
1861 (request
->type
== NBD_CMD_WRITE
||
1862 request
->type
== NBD_CMD_WRITE_ZEROES
||
1863 request
->type
== NBD_CMD_TRIM
)) {
1864 error_setg(errp
, "Export is read-only");
1867 if (request
->from
> client
->exp
->size
||
1868 request
->from
+ request
->len
> client
->exp
->size
) {
1869 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
1870 ", Size: %" PRIu64
, request
->from
, request
->len
,
1871 (uint64_t)client
->exp
->size
);
1872 return (request
->type
== NBD_CMD_WRITE
||
1873 request
->type
== NBD_CMD_WRITE_ZEROES
) ? -ENOSPC
: -EINVAL
;
1875 valid_flags
= NBD_CMD_FLAG_FUA
;
1876 if (request
->type
== NBD_CMD_READ
&& client
->structured_reply
) {
1877 valid_flags
|= NBD_CMD_FLAG_DF
;
1878 } else if (request
->type
== NBD_CMD_WRITE_ZEROES
) {
1879 valid_flags
|= NBD_CMD_FLAG_NO_HOLE
;
1880 } else if (request
->type
== NBD_CMD_BLOCK_STATUS
) {
1881 valid_flags
|= NBD_CMD_FLAG_REQ_ONE
;
1883 if (request
->flags
& ~valid_flags
) {
1884 error_setg(errp
, "unsupported flags for command %s (got 0x%x)",
1885 nbd_cmd_lookup(request
->type
), request
->flags
);
1892 /* Send simple reply without a payload, or a structured error
1893 * @error_msg is ignored if @ret >= 0
1894 * Returns 0 if connection is still live, -errno on failure to talk to client
1896 static coroutine_fn
int nbd_send_generic_reply(NBDClient
*client
,
1899 const char *error_msg
,
1902 if (client
->structured_reply
&& ret
< 0) {
1903 return nbd_co_send_structured_error(client
, handle
, -ret
, error_msg
,
1906 return nbd_co_send_simple_reply(client
, handle
, ret
< 0 ? -ret
: 0,
1911 /* Handle NBD_CMD_READ request.
1912 * Return -errno if sending fails. Other errors are reported directly to the
1913 * client as an error reply. */
1914 static coroutine_fn
int nbd_do_cmd_read(NBDClient
*client
, NBDRequest
*request
,
1915 uint8_t *data
, Error
**errp
)
1918 NBDExport
*exp
= client
->exp
;
1920 assert(request
->type
== NBD_CMD_READ
);
1922 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1923 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1924 ret
= blk_co_flush(exp
->blk
);
1926 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1927 "flush failed", errp
);
1931 if (client
->structured_reply
&& !(request
->flags
& NBD_CMD_FLAG_DF
) &&
1933 return nbd_co_send_sparse_read(client
, request
->handle
, request
->from
,
1934 data
, request
->len
, errp
);
1937 ret
= blk_pread(exp
->blk
, request
->from
+ exp
->dev_offset
, data
,
1940 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1941 "reading from file failed", errp
);
1944 if (client
->structured_reply
) {
1946 return nbd_co_send_structured_read(client
, request
->handle
,
1947 request
->from
, data
,
1948 request
->len
, true, errp
);
1950 return nbd_co_send_structured_done(client
, request
->handle
, errp
);
1953 return nbd_co_send_simple_reply(client
, request
->handle
, 0,
1954 data
, request
->len
, errp
);
1958 /* Handle NBD request.
1959 * Return -errno if sending fails. Other errors are reported directly to the
1960 * client as an error reply. */
1961 static coroutine_fn
int nbd_handle_request(NBDClient
*client
,
1962 NBDRequest
*request
,
1963 uint8_t *data
, Error
**errp
)
1967 NBDExport
*exp
= client
->exp
;
1970 switch (request
->type
) {
1972 return nbd_do_cmd_read(client
, request
, data
, errp
);
1976 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1977 flags
|= BDRV_REQ_FUA
;
1979 ret
= blk_pwrite(exp
->blk
, request
->from
+ exp
->dev_offset
,
1980 data
, request
->len
, flags
);
1981 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1982 "writing to file failed", errp
);
1984 case NBD_CMD_WRITE_ZEROES
:
1986 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1987 flags
|= BDRV_REQ_FUA
;
1989 if (!(request
->flags
& NBD_CMD_FLAG_NO_HOLE
)) {
1990 flags
|= BDRV_REQ_MAY_UNMAP
;
1992 ret
= blk_pwrite_zeroes(exp
->blk
, request
->from
+ exp
->dev_offset
,
1993 request
->len
, flags
);
1994 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1995 "writing to file failed", errp
);
1998 /* unreachable, thanks to special case in nbd_co_receive_request() */
2002 ret
= blk_co_flush(exp
->blk
);
2003 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2004 "flush failed", errp
);
2007 ret
= blk_co_pdiscard(exp
->blk
, request
->from
+ exp
->dev_offset
,
2009 if (ret
== 0 && request
->flags
& NBD_CMD_FLAG_FUA
) {
2010 ret
= blk_co_flush(exp
->blk
);
2012 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2013 "discard failed", errp
);
2015 case NBD_CMD_BLOCK_STATUS
:
2016 if (!request
->len
) {
2017 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2018 "need non-zero length", errp
);
2020 if (client
->export_meta
.valid
&& client
->export_meta
.base_allocation
) {
2021 return nbd_co_send_block_status(client
, request
->handle
,
2022 blk_bs(exp
->blk
), request
->from
,
2024 NBD_META_ID_BASE_ALLOCATION
, errp
);
2026 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2027 "CMD_BLOCK_STATUS not negotiated",
2032 msg
= g_strdup_printf("invalid request type (%" PRIu32
") received",
2034 ret
= nbd_send_generic_reply(client
, request
->handle
, -EINVAL
, msg
,
2041 /* Owns a reference to the NBDClient passed as opaque. */
2042 static coroutine_fn
void nbd_trip(void *opaque
)
2044 NBDClient
*client
= opaque
;
2045 NBDRequestData
*req
;
2046 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
2048 Error
*local_err
= NULL
;
2051 if (client
->closing
) {
2052 nbd_client_put(client
);
2056 req
= nbd_request_get(client
);
2057 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
2058 client
->recv_coroutine
= NULL
;
2060 if (client
->closing
) {
2062 * The client may be closed when we are blocked in
2063 * nbd_co_receive_request()
2068 nbd_client_receive_next_request(client
);
2074 /* It wans't -EIO, so, according to nbd_co_receive_request()
2075 * semantics, we should return the error to the client. */
2076 Error
*export_err
= local_err
;
2079 ret
= nbd_send_generic_reply(client
, request
.handle
, -EINVAL
,
2080 error_get_pretty(export_err
), &local_err
);
2081 error_free(export_err
);
2083 ret
= nbd_handle_request(client
, &request
, req
->data
, &local_err
);
2086 error_prepend(&local_err
, "Failed to send reply: ");
2090 /* We must disconnect after NBD_CMD_WRITE if we did not
2093 if (!req
->complete
) {
2094 error_setg(&local_err
, "Request handling failed in intermediate state");
2099 nbd_request_put(req
);
2100 nbd_client_put(client
);
2105 error_reportf_err(local_err
, "Disconnect client, due to: ");
2107 nbd_request_put(req
);
2108 client_close(client
, true);
2109 nbd_client_put(client
);
2112 static void nbd_client_receive_next_request(NBDClient
*client
)
2114 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
) {
2115 nbd_client_get(client
);
2116 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
2117 aio_co_schedule(client
->exp
->ctx
, client
->recv_coroutine
);
2121 static coroutine_fn
void nbd_co_client_start(void *opaque
)
2123 NBDClient
*client
= opaque
;
2124 NBDExport
*exp
= client
->exp
;
2125 Error
*local_err
= NULL
;
2128 nbd_export_get(exp
);
2129 QTAILQ_INSERT_TAIL(&exp
->clients
, client
, next
);
2131 qemu_co_mutex_init(&client
->send_lock
);
2133 if (nbd_negotiate(client
, &local_err
)) {
2135 error_report_err(local_err
);
2137 client_close(client
, false);
2141 nbd_client_receive_next_request(client
);
2145 * Create a new client listener on the given export @exp, using the
2146 * given channel @sioc. Begin servicing it in a coroutine. When the
2147 * connection closes, call @close_fn with an indication of whether the
2148 * client completed negotiation.
2150 void nbd_client_new(NBDExport
*exp
,
2151 QIOChannelSocket
*sioc
,
2152 QCryptoTLSCreds
*tlscreds
,
2153 const char *tlsaclname
,
2154 void (*close_fn
)(NBDClient
*, bool))
2159 client
= g_new0(NBDClient
, 1);
2160 client
->refcount
= 1;
2162 client
->tlscreds
= tlscreds
;
2164 object_ref(OBJECT(client
->tlscreds
));
2166 client
->tlsaclname
= g_strdup(tlsaclname
);
2167 client
->sioc
= sioc
;
2168 object_ref(OBJECT(client
->sioc
));
2169 client
->ioc
= QIO_CHANNEL(sioc
);
2170 object_ref(OBJECT(client
->ioc
));
2171 client
->close_fn
= close_fn
;
2173 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
2174 qemu_coroutine_enter(co
);