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
{
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(NBDClient
*client
)
404 client
->export_meta
.valid
&= client
->exp
== client
->export_meta
.exp
;
407 /* Send a reply to NBD_OPT_EXPORT_NAME.
408 * Return -errno on error, 0 on success. */
409 static int nbd_negotiate_handle_export_name(NBDClient
*client
,
410 uint16_t myflags
, bool no_zeroes
,
413 char name
[NBD_MAX_NAME_SIZE
+ 1];
414 char buf
[NBD_REPLY_EXPORT_NAME_SIZE
] = "";
419 [20 .. xx] export name (length bytes)
422 [ 8 .. 9] export flags
423 [10 .. 133] reserved (0) [unless no_zeroes]
425 trace_nbd_negotiate_handle_export_name();
426 if (client
->optlen
>= sizeof(name
)) {
427 error_setg(errp
, "Bad length received");
430 if (nbd_read(client
->ioc
, name
, client
->optlen
, errp
) < 0) {
431 error_prepend(errp
, "read failed: ");
434 name
[client
->optlen
] = '\0';
437 trace_nbd_negotiate_handle_export_name_request(name
);
439 client
->exp
= nbd_export_find(name
);
441 error_setg(errp
, "export not found");
445 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
,
446 client
->exp
->nbdflags
| myflags
);
447 stq_be_p(buf
, client
->exp
->size
);
448 stw_be_p(buf
+ 8, client
->exp
->nbdflags
| myflags
);
449 len
= no_zeroes
? 10 : sizeof(buf
);
450 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
452 error_prepend(errp
, "write failed: ");
456 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
457 nbd_export_get(client
->exp
);
458 nbd_check_meta_export(client
);
463 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
464 * The buffer does NOT include the info type prefix.
465 * Return -errno on error, 0 if ready to send more. */
466 static int nbd_negotiate_send_info(NBDClient
*client
,
467 uint16_t info
, uint32_t length
, void *buf
,
472 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
473 rc
= nbd_negotiate_send_rep_len(client
, NBD_REP_INFO
,
474 sizeof(info
) + length
, errp
);
479 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
482 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
488 /* nbd_reject_length: Handle any unexpected payload.
489 * @fatal requests that we quit talking to the client, even if we are able
490 * to successfully send an error reply.
492 * -errno transmission error occurred or @fatal was requested, errp is set
493 * 0 error message successfully sent to client, errp is not set
495 static int nbd_reject_length(NBDClient
*client
, bool fatal
, Error
**errp
)
499 assert(client
->optlen
);
500 ret
= nbd_opt_invalid(client
, errp
, "option '%s' has unexpected length",
501 nbd_opt_lookup(client
->opt
));
503 error_setg(errp
, "option '%s' has unexpected length",
504 nbd_opt_lookup(client
->opt
));
510 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
511 * Return -errno on error, 0 if ready for next option, and 1 to move
512 * into transmission phase. */
513 static int nbd_negotiate_handle_info(NBDClient
*client
, uint16_t myflags
,
517 char name
[NBD_MAX_NAME_SIZE
+ 1];
522 bool sendname
= false;
523 bool blocksize
= false;
525 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
528 4 bytes: L, name length (can be 0)
530 2 bytes: N, number of requests (can be 0)
531 N * 2 bytes: N requests
533 rc
= nbd_opt_read_name(client
, name
, &namelen
, errp
);
537 trace_nbd_negotiate_handle_export_name_request(name
);
539 rc
= nbd_opt_read(client
, &requests
, sizeof(requests
), errp
);
543 be16_to_cpus(&requests
);
544 trace_nbd_negotiate_handle_info_requests(requests
);
546 rc
= nbd_opt_read(client
, &request
, sizeof(request
), errp
);
550 be16_to_cpus(&request
);
551 trace_nbd_negotiate_handle_info_request(request
,
552 nbd_info_lookup(request
));
553 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
554 * everything else is either a request we don't know or
555 * something we send regardless of request */
560 case NBD_INFO_BLOCK_SIZE
:
565 if (client
->optlen
) {
566 return nbd_reject_length(client
, false, errp
);
569 exp
= nbd_export_find(name
);
571 return nbd_negotiate_send_rep_err(client
, NBD_REP_ERR_UNKNOWN
,
572 errp
, "export '%s' not present",
576 /* Don't bother sending NBD_INFO_NAME unless client requested it */
578 rc
= nbd_negotiate_send_info(client
, NBD_INFO_NAME
, namelen
, name
,
585 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
587 if (exp
->description
) {
588 size_t len
= strlen(exp
->description
);
590 rc
= nbd_negotiate_send_info(client
, NBD_INFO_DESCRIPTION
,
591 len
, exp
->description
, errp
);
597 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
598 * according to whether the client requested it, and according to
599 * whether this is OPT_INFO or OPT_GO. */
600 /* minimum - 1 for back-compat, or 512 if client is new enough.
601 * TODO: consult blk_bs(blk)->bl.request_alignment? */
603 (client
->opt
== NBD_OPT_INFO
|| blocksize
) ? BDRV_SECTOR_SIZE
: 1;
604 /* preferred - Hard-code to 4096 for now.
605 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
607 /* maximum - At most 32M, but smaller as appropriate. */
608 sizes
[2] = MIN(blk_get_max_transfer(exp
->blk
), NBD_MAX_BUFFER_SIZE
);
609 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
610 cpu_to_be32s(&sizes
[0]);
611 cpu_to_be32s(&sizes
[1]);
612 cpu_to_be32s(&sizes
[2]);
613 rc
= nbd_negotiate_send_info(client
, NBD_INFO_BLOCK_SIZE
,
614 sizeof(sizes
), sizes
, errp
);
619 /* Send NBD_INFO_EXPORT always */
620 trace_nbd_negotiate_new_style_size_flags(exp
->size
,
621 exp
->nbdflags
| myflags
);
622 stq_be_p(buf
, exp
->size
);
623 stw_be_p(buf
+ 8, exp
->nbdflags
| myflags
);
624 rc
= nbd_negotiate_send_info(client
, NBD_INFO_EXPORT
,
625 sizeof(buf
), buf
, errp
);
630 /* If the client is just asking for NBD_OPT_INFO, but forgot to
631 * request block sizes, return an error.
632 * TODO: consult blk_bs(blk)->request_align, and only error if it
634 if (client
->opt
== NBD_OPT_INFO
&& !blocksize
) {
635 return nbd_negotiate_send_rep_err(client
,
636 NBD_REP_ERR_BLOCK_SIZE_REQD
,
638 "request NBD_INFO_BLOCK_SIZE to "
643 rc
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
648 if (client
->opt
== NBD_OPT_GO
) {
650 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
651 nbd_export_get(client
->exp
);
652 nbd_check_meta_export(client
);
659 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
660 * new channel for all further (now-encrypted) communication. */
661 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
666 struct NBDTLSHandshakeData data
= { 0 };
668 assert(client
->opt
== NBD_OPT_STARTTLS
);
670 trace_nbd_negotiate_handle_starttls();
673 if (nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
) < 0) {
677 tioc
= qio_channel_tls_new_server(ioc
,
685 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
686 trace_nbd_negotiate_handle_starttls_handshake();
687 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
688 qio_channel_tls_handshake(tioc
,
694 if (!data
.complete
) {
695 g_main_loop_run(data
.loop
);
697 g_main_loop_unref(data
.loop
);
699 object_unref(OBJECT(tioc
));
700 error_propagate(errp
, data
.error
);
704 return QIO_CHANNEL(tioc
);
707 /* nbd_negotiate_send_meta_context
709 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
711 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
713 static int nbd_negotiate_send_meta_context(NBDClient
*client
,
718 NBDOptionReplyMetaContext opt
;
719 struct iovec iov
[] = {
720 {.iov_base
= &opt
, .iov_len
= sizeof(opt
)},
721 {.iov_base
= (void *)context
, .iov_len
= strlen(context
)}
724 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
728 trace_nbd_negotiate_meta_query_reply(context
, context_id
);
729 set_be_option_rep(&opt
.h
, client
->opt
, NBD_REP_META_CONTEXT
,
730 sizeof(opt
) - sizeof(opt
.h
) + iov
[1].iov_len
);
731 stl_be_p(&opt
.context_id
, context_id
);
733 return qio_channel_writev_all(client
->ioc
, iov
, 2, errp
) < 0 ? -EIO
: 0;
736 /* nbd_meta_base_query
738 * Handle queries to 'base' namespace. For now, only the base:allocation
739 * context is available. 'len' is the amount of text remaining to be read from
740 * the current name, after the 'base:' portion has been stripped.
742 * Return -errno on I/O error, 0 if option was completely handled by
743 * sending a reply about inconsistent lengths, or 1 on success.
745 * Note: return code = 1 doesn't mean that we've parsed the "base:allocation"
746 * namespace. It only means that there are no errors.
748 static int nbd_meta_base_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
749 uint32_t len
, Error
**errp
)
752 char query
[sizeof("allocation") - 1];
753 size_t alen
= strlen("allocation");
756 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
757 meta
->base_allocation
= true;
759 trace_nbd_negotiate_meta_query_parse("base:");
764 trace_nbd_negotiate_meta_query_skip("not base:allocation");
765 return nbd_opt_skip(client
, len
, errp
);
768 ret
= nbd_opt_read(client
, query
, len
, errp
);
773 if (strncmp(query
, "allocation", alen
) == 0) {
774 trace_nbd_negotiate_meta_query_parse("base:allocation");
775 meta
->base_allocation
= true;
777 trace_nbd_negotiate_meta_query_skip("not base:allocation");
783 /* nbd_negotiate_meta_query
785 * Parse namespace name and call corresponding function to parse body of the
788 * The only supported namespace now is 'base'.
790 * The function aims not wasting time and memory to read long unknown namespace
793 * Return -errno on I/O error, 0 if option was completely handled by
794 * sending a reply about inconsistent lengths, or 1 on success. */
795 static int nbd_negotiate_meta_query(NBDClient
*client
,
796 NBDExportMetaContexts
*meta
, Error
**errp
)
799 char query
[sizeof("base:") - 1];
800 size_t baselen
= strlen("base:");
803 ret
= nbd_opt_read(client
, &len
, sizeof(len
), errp
);
809 /* The only supported namespace for now is 'base'. So query should start
810 * with 'base:'. Otherwise, we can ignore it and skip the remainder. */
812 trace_nbd_negotiate_meta_query_skip("length too short");
813 return nbd_opt_skip(client
, len
, errp
);
817 ret
= nbd_opt_read(client
, query
, baselen
, errp
);
821 if (strncmp(query
, "base:", baselen
) != 0) {
822 trace_nbd_negotiate_meta_query_skip("not for base: namespace");
823 return nbd_opt_skip(client
, len
, errp
);
826 return nbd_meta_base_query(client
, meta
, len
, errp
);
829 /* nbd_negotiate_meta_queries
830 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
832 * Return -errno on I/O error, or 0 if option was completely handled. */
833 static int nbd_negotiate_meta_queries(NBDClient
*client
,
834 NBDExportMetaContexts
*meta
, Error
**errp
)
837 char export_name
[NBD_MAX_NAME_SIZE
+ 1];
838 NBDExportMetaContexts local_meta
;
842 if (!client
->structured_reply
) {
843 return nbd_opt_invalid(client
, errp
,
844 "request option '%s' when structured reply "
846 nbd_opt_lookup(client
->opt
));
849 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
850 /* Only change the caller's meta on SET. */
854 memset(meta
, 0, sizeof(*meta
));
856 ret
= nbd_opt_read_name(client
, export_name
, NULL
, errp
);
861 meta
->exp
= nbd_export_find(export_name
);
862 if (meta
->exp
== NULL
) {
863 return nbd_opt_drop(client
, NBD_REP_ERR_UNKNOWN
, errp
,
864 "export '%s' not present", export_name
);
867 ret
= nbd_opt_read(client
, &nb_queries
, sizeof(nb_queries
), errp
);
871 cpu_to_be32s(&nb_queries
);
872 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client
->opt
),
873 export_name
, nb_queries
);
875 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
&& !nb_queries
) {
876 /* enable all known contexts */
877 meta
->base_allocation
= true;
879 for (i
= 0; i
< nb_queries
; ++i
) {
880 ret
= nbd_negotiate_meta_query(client
, meta
, errp
);
887 if (meta
->base_allocation
) {
888 ret
= nbd_negotiate_send_meta_context(client
, "base:allocation",
889 NBD_META_ID_BASE_ALLOCATION
,
896 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
904 /* nbd_negotiate_options
905 * Process all NBD_OPT_* client option commands, during fixed newstyle
908 * -errno on error, errp is set
909 * 0 on successful negotiation, errp is not set
910 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
913 static int nbd_negotiate_options(NBDClient
*client
, uint16_t myflags
,
917 bool fixedNewstyle
= false;
918 bool no_zeroes
= false;
921 [ 0 .. 3] client flags
923 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
924 [ 0 .. 7] NBD_OPTS_MAGIC
925 [ 8 .. 11] NBD option
926 [12 .. 15] Data length
929 [ 0 .. 7] NBD_OPTS_MAGIC
930 [ 8 .. 11] Second NBD option
931 [12 .. 15] Data length
935 if (nbd_read(client
->ioc
, &flags
, sizeof(flags
), errp
) < 0) {
936 error_prepend(errp
, "read failed: ");
939 be32_to_cpus(&flags
);
940 trace_nbd_negotiate_options_flags(flags
);
941 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
942 fixedNewstyle
= true;
943 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
945 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
947 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
950 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
956 uint32_t option
, length
;
959 if (nbd_read(client
->ioc
, &magic
, sizeof(magic
), errp
) < 0) {
960 error_prepend(errp
, "read failed: ");
963 magic
= be64_to_cpu(magic
);
964 trace_nbd_negotiate_options_check_magic(magic
);
965 if (magic
!= NBD_OPTS_MAGIC
) {
966 error_setg(errp
, "Bad magic received");
970 if (nbd_read(client
->ioc
, &option
,
971 sizeof(option
), errp
) < 0) {
972 error_prepend(errp
, "read failed: ");
975 option
= be32_to_cpu(option
);
976 client
->opt
= option
;
978 if (nbd_read(client
->ioc
, &length
, sizeof(length
), errp
) < 0) {
979 error_prepend(errp
, "read failed: ");
982 length
= be32_to_cpu(length
);
983 assert(!client
->optlen
);
984 client
->optlen
= length
;
986 if (length
> NBD_MAX_BUFFER_SIZE
) {
987 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
988 length
, NBD_MAX_BUFFER_SIZE
);
992 trace_nbd_negotiate_options_check_option(option
,
993 nbd_opt_lookup(option
));
994 if (client
->tlscreds
&&
995 client
->ioc
== (QIOChannel
*)client
->sioc
) {
997 if (!fixedNewstyle
) {
998 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
1002 case NBD_OPT_STARTTLS
:
1004 /* Unconditionally drop the connection if the client
1005 * can't start a TLS negotiation correctly */
1006 return nbd_reject_length(client
, true, errp
);
1008 tioc
= nbd_negotiate_handle_starttls(client
, errp
);
1013 object_unref(OBJECT(client
->ioc
));
1014 client
->ioc
= QIO_CHANNEL(tioc
);
1017 case NBD_OPT_EXPORT_NAME
:
1018 /* No way to return an error to client, so drop connection */
1019 error_setg(errp
, "Option 0x%x not permitted before TLS",
1024 ret
= nbd_opt_drop(client
, NBD_REP_ERR_TLS_REQD
, errp
,
1026 "not permitted before TLS", option
);
1027 /* Let the client keep trying, unless they asked to
1028 * quit. In this mode, we've already sent an error, so
1029 * we can't ack the abort. */
1030 if (option
== NBD_OPT_ABORT
) {
1035 } else if (fixedNewstyle
) {
1039 ret
= nbd_reject_length(client
, false, errp
);
1041 ret
= nbd_negotiate_handle_list(client
, errp
);
1046 /* NBD spec says we must try to reply before
1047 * disconnecting, but that we must also tolerate
1048 * guests that don't wait for our reply. */
1049 nbd_negotiate_send_rep(client
, NBD_REP_ACK
, NULL
);
1052 case NBD_OPT_EXPORT_NAME
:
1053 return nbd_negotiate_handle_export_name(client
,
1059 ret
= nbd_negotiate_handle_info(client
, myflags
, errp
);
1061 assert(option
== NBD_OPT_GO
);
1066 case NBD_OPT_STARTTLS
:
1068 ret
= nbd_reject_length(client
, false, errp
);
1069 } else if (client
->tlscreds
) {
1070 ret
= nbd_negotiate_send_rep_err(client
,
1071 NBD_REP_ERR_INVALID
, errp
,
1072 "TLS already enabled");
1074 ret
= nbd_negotiate_send_rep_err(client
,
1075 NBD_REP_ERR_POLICY
, errp
,
1076 "TLS not configured");
1080 case NBD_OPT_STRUCTURED_REPLY
:
1082 ret
= nbd_reject_length(client
, false, errp
);
1083 } else if (client
->structured_reply
) {
1084 ret
= nbd_negotiate_send_rep_err(
1085 client
, NBD_REP_ERR_INVALID
, errp
,
1086 "structured reply already negotiated");
1088 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1089 client
->structured_reply
= true;
1090 myflags
|= NBD_FLAG_SEND_DF
;
1094 case NBD_OPT_LIST_META_CONTEXT
:
1095 case NBD_OPT_SET_META_CONTEXT
:
1096 ret
= nbd_negotiate_meta_queries(client
, &client
->export_meta
,
1101 ret
= nbd_opt_drop(client
, NBD_REP_ERR_UNSUP
, errp
,
1102 "Unsupported option %" PRIu32
" (%s)",
1103 option
, nbd_opt_lookup(option
));
1108 * If broken new-style we should drop the connection
1109 * for anything except NBD_OPT_EXPORT_NAME
1112 case NBD_OPT_EXPORT_NAME
:
1113 return nbd_negotiate_handle_export_name(client
,
1118 error_setg(errp
, "Unsupported option %" PRIu32
" (%s)",
1119 option
, nbd_opt_lookup(option
));
1131 * -errno on error, errp is set
1132 * 0 on successful negotiation, errp is not set
1133 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1136 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
1138 char buf
[NBD_OLDSTYLE_NEGOTIATE_SIZE
] = "";
1140 const uint16_t myflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
1141 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
|
1142 NBD_FLAG_SEND_WRITE_ZEROES
);
1145 /* Old style negotiation header, no room for options
1146 [ 0 .. 7] passwd ("NBDMAGIC")
1147 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1149 [24 .. 27] export flags (zero-extended)
1150 [28 .. 151] reserved (0)
1152 New style negotiation header, client can send options
1153 [ 0 .. 7] passwd ("NBDMAGIC")
1154 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1155 [16 .. 17] server flags (0)
1156 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1159 qio_channel_set_blocking(client
->ioc
, false, NULL
);
1161 trace_nbd_negotiate_begin();
1162 memcpy(buf
, "NBDMAGIC", 8);
1164 oldStyle
= client
->exp
!= NULL
&& !client
->tlscreds
;
1166 trace_nbd_negotiate_old_style(client
->exp
->size
,
1167 client
->exp
->nbdflags
| myflags
);
1168 stq_be_p(buf
+ 8, NBD_CLIENT_MAGIC
);
1169 stq_be_p(buf
+ 16, client
->exp
->size
);
1170 stl_be_p(buf
+ 24, client
->exp
->nbdflags
| myflags
);
1172 if (nbd_write(client
->ioc
, buf
, sizeof(buf
), errp
) < 0) {
1173 error_prepend(errp
, "write failed: ");
1177 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
1178 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
1180 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
1181 error_prepend(errp
, "write failed: ");
1184 ret
= nbd_negotiate_options(client
, myflags
, errp
);
1187 error_prepend(errp
, "option negotiation failed: ");
1193 assert(!client
->optlen
);
1194 trace_nbd_negotiate_success();
1199 static int nbd_receive_request(QIOChannel
*ioc
, NBDRequest
*request
,
1202 uint8_t buf
[NBD_REQUEST_SIZE
];
1206 ret
= nbd_read(ioc
, buf
, sizeof(buf
), errp
);
1212 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1213 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1214 [ 6 .. 7] type (NBD_CMD_READ, ...)
1220 magic
= ldl_be_p(buf
);
1221 request
->flags
= lduw_be_p(buf
+ 4);
1222 request
->type
= lduw_be_p(buf
+ 6);
1223 request
->handle
= ldq_be_p(buf
+ 8);
1224 request
->from
= ldq_be_p(buf
+ 16);
1225 request
->len
= ldl_be_p(buf
+ 24);
1227 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
1228 request
->from
, request
->len
);
1230 if (magic
!= NBD_REQUEST_MAGIC
) {
1231 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
1237 #define MAX_NBD_REQUESTS 16
1239 void nbd_client_get(NBDClient
*client
)
1244 void nbd_client_put(NBDClient
*client
)
1246 if (--client
->refcount
== 0) {
1247 /* The last reference should be dropped by client->close,
1248 * which is called by client_close.
1250 assert(client
->closing
);
1252 qio_channel_detach_aio_context(client
->ioc
);
1253 object_unref(OBJECT(client
->sioc
));
1254 object_unref(OBJECT(client
->ioc
));
1255 if (client
->tlscreds
) {
1256 object_unref(OBJECT(client
->tlscreds
));
1258 g_free(client
->tlsaclname
);
1260 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
1261 nbd_export_put(client
->exp
);
1267 static void client_close(NBDClient
*client
, bool negotiated
)
1269 if (client
->closing
) {
1273 client
->closing
= true;
1275 /* Force requests to finish. They will drop their own references,
1276 * then we'll close the socket and free the NBDClient.
1278 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
1281 /* Also tell the client, so that they release their reference. */
1282 if (client
->close_fn
) {
1283 client
->close_fn(client
, negotiated
);
1287 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
1289 NBDRequestData
*req
;
1291 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
1292 client
->nb_requests
++;
1294 req
= g_new0(NBDRequestData
, 1);
1295 nbd_client_get(client
);
1296 req
->client
= client
;
1300 static void nbd_request_put(NBDRequestData
*req
)
1302 NBDClient
*client
= req
->client
;
1305 qemu_vfree(req
->data
);
1309 client
->nb_requests
--;
1310 nbd_client_receive_next_request(client
);
1312 nbd_client_put(client
);
1315 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1317 NBDExport
*exp
= opaque
;
1320 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1324 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1325 qio_channel_attach_aio_context(client
->ioc
, ctx
);
1326 if (client
->recv_coroutine
) {
1327 aio_co_schedule(ctx
, client
->recv_coroutine
);
1329 if (client
->send_coroutine
) {
1330 aio_co_schedule(ctx
, client
->send_coroutine
);
1335 static void blk_aio_detach(void *opaque
)
1337 NBDExport
*exp
= opaque
;
1340 trace_nbd_blk_aio_detach(exp
->name
, exp
->ctx
);
1342 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1343 qio_channel_detach_aio_context(client
->ioc
);
1349 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1351 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1352 nbd_export_close(exp
);
1355 NBDExport
*nbd_export_new(BlockDriverState
*bs
, off_t dev_offset
, off_t size
,
1356 uint16_t nbdflags
, void (*close
)(NBDExport
*),
1357 bool writethrough
, BlockBackend
*on_eject_blk
,
1362 NBDExport
*exp
= g_new0(NBDExport
, 1);
1367 * NBD exports are used for non-shared storage migration. Make sure
1368 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1369 * access since the export could be available before migration handover.
1371 ctx
= bdrv_get_aio_context(bs
);
1372 aio_context_acquire(ctx
);
1373 bdrv_invalidate_cache(bs
, NULL
);
1374 aio_context_release(ctx
);
1376 /* Don't allow resize while the NBD server is running, otherwise we don't
1377 * care what happens with the node. */
1378 perm
= BLK_PERM_CONSISTENT_READ
;
1379 if ((nbdflags
& NBD_FLAG_READ_ONLY
) == 0) {
1380 perm
|= BLK_PERM_WRITE
;
1382 blk
= blk_new(perm
, BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
|
1383 BLK_PERM_WRITE
| BLK_PERM_GRAPH_MOD
);
1384 ret
= blk_insert_bs(blk
, bs
, errp
);
1388 blk_set_enable_write_cache(blk
, !writethrough
);
1391 QTAILQ_INIT(&exp
->clients
);
1393 exp
->dev_offset
= dev_offset
;
1394 exp
->nbdflags
= nbdflags
;
1395 exp
->size
= size
< 0 ? blk_getlength(blk
) : size
;
1396 if (exp
->size
< 0) {
1397 error_setg_errno(errp
, -exp
->size
,
1398 "Failed to determine the NBD export's length");
1401 exp
->size
-= exp
->size
% BDRV_SECTOR_SIZE
;
1404 exp
->ctx
= blk_get_aio_context(blk
);
1405 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1408 blk_ref(on_eject_blk
);
1409 exp
->eject_notifier_blk
= on_eject_blk
;
1410 exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1411 blk_add_remove_bs_notifier(on_eject_blk
, &exp
->eject_notifier
);
1421 NBDExport
*nbd_export_find(const char *name
)
1424 QTAILQ_FOREACH(exp
, &exports
, next
) {
1425 if (strcmp(name
, exp
->name
) == 0) {
1433 void nbd_export_set_name(NBDExport
*exp
, const char *name
)
1435 if (exp
->name
== name
) {
1439 nbd_export_get(exp
);
1440 if (exp
->name
!= NULL
) {
1443 QTAILQ_REMOVE(&exports
, exp
, next
);
1444 nbd_export_put(exp
);
1447 nbd_export_get(exp
);
1448 exp
->name
= g_strdup(name
);
1449 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1451 nbd_export_put(exp
);
1454 void nbd_export_set_description(NBDExport
*exp
, const char *description
)
1456 g_free(exp
->description
);
1457 exp
->description
= g_strdup(description
);
1460 void nbd_export_close(NBDExport
*exp
)
1462 NBDClient
*client
, *next
;
1464 nbd_export_get(exp
);
1465 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1466 client_close(client
, true);
1468 nbd_export_set_name(exp
, NULL
);
1469 nbd_export_set_description(exp
, NULL
);
1470 nbd_export_put(exp
);
1473 void nbd_export_remove(NBDExport
*exp
, NbdServerRemoveMode mode
, Error
**errp
)
1475 if (mode
== NBD_SERVER_REMOVE_MODE_HARD
|| QTAILQ_EMPTY(&exp
->clients
)) {
1476 nbd_export_close(exp
);
1480 assert(mode
== NBD_SERVER_REMOVE_MODE_SAFE
);
1482 error_setg(errp
, "export '%s' still in use", exp
->name
);
1483 error_append_hint(errp
, "Use mode='hard' to force client disconnect\n");
1486 void nbd_export_get(NBDExport
*exp
)
1488 assert(exp
->refcount
> 0);
1492 void nbd_export_put(NBDExport
*exp
)
1494 assert(exp
->refcount
> 0);
1495 if (exp
->refcount
== 1) {
1496 nbd_export_close(exp
);
1499 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1500 * if someone calls nbd_export_put() on named export not through
1501 * nbd_export_set_name() when refcount is 1. So, let's assert that
1504 assert(exp
->refcount
> 0);
1505 if (--exp
->refcount
== 0) {
1506 assert(exp
->name
== NULL
);
1507 assert(exp
->description
== NULL
);
1514 if (exp
->eject_notifier_blk
) {
1515 notifier_remove(&exp
->eject_notifier
);
1516 blk_unref(exp
->eject_notifier_blk
);
1518 blk_remove_aio_context_notifier(exp
->blk
, blk_aio_attached
,
1519 blk_aio_detach
, exp
);
1520 blk_unref(exp
->blk
);
1528 BlockBackend
*nbd_export_get_blockdev(NBDExport
*exp
)
1533 void nbd_export_close_all(void)
1535 NBDExport
*exp
, *next
;
1537 QTAILQ_FOREACH_SAFE(exp
, &exports
, next
, next
) {
1538 nbd_export_close(exp
);
1542 static int coroutine_fn
nbd_co_send_iov(NBDClient
*client
, struct iovec
*iov
,
1543 unsigned niov
, Error
**errp
)
1547 g_assert(qemu_in_coroutine());
1548 qemu_co_mutex_lock(&client
->send_lock
);
1549 client
->send_coroutine
= qemu_coroutine_self();
1551 ret
= qio_channel_writev_all(client
->ioc
, iov
, niov
, errp
) < 0 ? -EIO
: 0;
1553 client
->send_coroutine
= NULL
;
1554 qemu_co_mutex_unlock(&client
->send_lock
);
1559 static inline void set_be_simple_reply(NBDSimpleReply
*reply
, uint64_t error
,
1562 stl_be_p(&reply
->magic
, NBD_SIMPLE_REPLY_MAGIC
);
1563 stl_be_p(&reply
->error
, error
);
1564 stq_be_p(&reply
->handle
, handle
);
1567 static int nbd_co_send_simple_reply(NBDClient
*client
,
1574 NBDSimpleReply reply
;
1575 int nbd_err
= system_errno_to_nbd_errno(error
);
1576 struct iovec iov
[] = {
1577 {.iov_base
= &reply
, .iov_len
= sizeof(reply
)},
1578 {.iov_base
= data
, .iov_len
= len
}
1581 trace_nbd_co_send_simple_reply(handle
, nbd_err
, nbd_err_lookup(nbd_err
),
1583 set_be_simple_reply(&reply
, nbd_err
, handle
);
1585 return nbd_co_send_iov(client
, iov
, len
? 2 : 1, errp
);
1588 static inline void set_be_chunk(NBDStructuredReplyChunk
*chunk
, uint16_t flags
,
1589 uint16_t type
, uint64_t handle
, uint32_t length
)
1591 stl_be_p(&chunk
->magic
, NBD_STRUCTURED_REPLY_MAGIC
);
1592 stw_be_p(&chunk
->flags
, flags
);
1593 stw_be_p(&chunk
->type
, type
);
1594 stq_be_p(&chunk
->handle
, handle
);
1595 stl_be_p(&chunk
->length
, length
);
1598 static int coroutine_fn
nbd_co_send_structured_done(NBDClient
*client
,
1602 NBDStructuredReplyChunk chunk
;
1603 struct iovec iov
[] = {
1604 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1607 trace_nbd_co_send_structured_done(handle
);
1608 set_be_chunk(&chunk
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_NONE
, handle
, 0);
1610 return nbd_co_send_iov(client
, iov
, 1, errp
);
1613 static int coroutine_fn
nbd_co_send_structured_read(NBDClient
*client
,
1621 NBDStructuredReadData chunk
;
1622 struct iovec iov
[] = {
1623 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1624 {.iov_base
= data
, .iov_len
= size
}
1628 trace_nbd_co_send_structured_read(handle
, offset
, data
, size
);
1629 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1630 NBD_REPLY_TYPE_OFFSET_DATA
, handle
,
1631 sizeof(chunk
) - sizeof(chunk
.h
) + size
);
1632 stq_be_p(&chunk
.offset
, offset
);
1634 return nbd_co_send_iov(client
, iov
, 2, errp
);
1637 static int coroutine_fn
nbd_co_send_structured_error(NBDClient
*client
,
1643 NBDStructuredError chunk
;
1644 int nbd_err
= system_errno_to_nbd_errno(error
);
1645 struct iovec iov
[] = {
1646 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1647 {.iov_base
= (char *)msg
, .iov_len
= msg
? strlen(msg
) : 0},
1651 trace_nbd_co_send_structured_error(handle
, nbd_err
,
1652 nbd_err_lookup(nbd_err
), msg
? msg
: "");
1653 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_ERROR
, handle
,
1654 sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1655 stl_be_p(&chunk
.error
, nbd_err
);
1656 stw_be_p(&chunk
.message_length
, iov
[1].iov_len
);
1658 return nbd_co_send_iov(client
, iov
, 1 + !!iov
[1].iov_len
, errp
);
1661 /* Do a sparse read and send the structured reply to the client.
1662 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1663 * reported to the client, at which point this function succeeds.
1665 static int coroutine_fn
nbd_co_send_sparse_read(NBDClient
*client
,
1673 NBDExport
*exp
= client
->exp
;
1674 size_t progress
= 0;
1676 while (progress
< size
) {
1678 int status
= bdrv_block_status_above(blk_bs(exp
->blk
), NULL
,
1680 size
- progress
, &pnum
, NULL
,
1685 char *msg
= g_strdup_printf("unable to check for holes: %s",
1688 ret
= nbd_co_send_structured_error(client
, handle
, -status
, msg
,
1693 assert(pnum
&& pnum
<= size
- progress
);
1694 final
= progress
+ pnum
== size
;
1695 if (status
& BDRV_BLOCK_ZERO
) {
1696 NBDStructuredReadHole chunk
;
1697 struct iovec iov
[] = {
1698 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1701 trace_nbd_co_send_structured_read_hole(handle
, offset
+ progress
,
1703 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1704 NBD_REPLY_TYPE_OFFSET_HOLE
,
1705 handle
, sizeof(chunk
) - sizeof(chunk
.h
));
1706 stq_be_p(&chunk
.offset
, offset
+ progress
);
1707 stl_be_p(&chunk
.length
, pnum
);
1708 ret
= nbd_co_send_iov(client
, iov
, 1, errp
);
1710 ret
= blk_pread(exp
->blk
, offset
+ progress
+ exp
->dev_offset
,
1711 data
+ progress
, pnum
);
1713 error_setg_errno(errp
, -ret
, "reading from file failed");
1716 ret
= nbd_co_send_structured_read(client
, handle
, offset
+ progress
,
1717 data
+ progress
, pnum
, final
,
1729 static int blockstatus_to_extent_be(BlockDriverState
*bs
, uint64_t offset
,
1730 uint64_t bytes
, NBDExtent
*extent
)
1732 uint64_t remaining_bytes
= bytes
;
1734 while (remaining_bytes
) {
1737 int ret
= bdrv_block_status_above(bs
, NULL
, offset
, remaining_bytes
,
1743 flags
= (ret
& BDRV_BLOCK_ALLOCATED
? 0 : NBD_STATE_HOLE
) |
1744 (ret
& BDRV_BLOCK_ZERO
? NBD_STATE_ZERO
: 0);
1746 if (remaining_bytes
== bytes
) {
1747 extent
->flags
= flags
;
1750 if (flags
!= extent
->flags
) {
1755 remaining_bytes
-= num
;
1758 cpu_to_be32s(&extent
->flags
);
1759 extent
->length
= cpu_to_be32(bytes
- remaining_bytes
);
1764 /* nbd_co_send_extents
1765 * @extents should be in big-endian */
1766 static int nbd_co_send_extents(NBDClient
*client
, uint64_t handle
,
1767 NBDExtent
*extents
, unsigned nb_extents
,
1768 uint32_t context_id
, Error
**errp
)
1770 NBDStructuredMeta chunk
;
1772 struct iovec iov
[] = {
1773 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1774 {.iov_base
= extents
, .iov_len
= nb_extents
* sizeof(extents
[0])}
1777 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_BLOCK_STATUS
,
1778 handle
, sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1779 stl_be_p(&chunk
.context_id
, context_id
);
1781 return nbd_co_send_iov(client
, iov
, 2, errp
);
1784 /* Get block status from the exported device and send it to the client */
1785 static int nbd_co_send_block_status(NBDClient
*client
, uint64_t handle
,
1786 BlockDriverState
*bs
, uint64_t offset
,
1787 uint64_t length
, uint32_t context_id
,
1793 ret
= blockstatus_to_extent_be(bs
, offset
, length
, &extent
);
1795 return nbd_co_send_structured_error(
1796 client
, handle
, -ret
, "can't get block status", errp
);
1799 return nbd_co_send_extents(client
, handle
, &extent
, 1, context_id
, errp
);
1802 /* nbd_co_receive_request
1803 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1804 * connection right away, and any other negative value to report an error to
1805 * the client (although the caller may still need to disconnect after reporting
1808 static int nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
1811 NBDClient
*client
= req
->client
;
1814 g_assert(qemu_in_coroutine());
1815 assert(client
->recv_coroutine
== qemu_coroutine_self());
1816 if (nbd_receive_request(client
->ioc
, request
, errp
) < 0) {
1820 trace_nbd_co_receive_request_decode_type(request
->handle
, request
->type
,
1821 nbd_cmd_lookup(request
->type
));
1823 if (request
->type
!= NBD_CMD_WRITE
) {
1824 /* No payload, we are ready to read the next request. */
1825 req
->complete
= true;
1828 if (request
->type
== NBD_CMD_DISC
) {
1829 /* Special case: we're going to disconnect without a reply,
1830 * whether or not flags, from, or len are bogus */
1834 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
) {
1835 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
1836 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1837 request
->len
, NBD_MAX_BUFFER_SIZE
);
1841 req
->data
= blk_try_blockalign(client
->exp
->blk
, request
->len
);
1842 if (req
->data
== NULL
) {
1843 error_setg(errp
, "No memory");
1847 if (request
->type
== NBD_CMD_WRITE
) {
1848 if (nbd_read(client
->ioc
, req
->data
, request
->len
, errp
) < 0) {
1849 error_prepend(errp
, "reading from socket failed: ");
1852 req
->complete
= true;
1854 trace_nbd_co_receive_request_payload_received(request
->handle
,
1858 /* Sanity checks. */
1859 if (client
->exp
->nbdflags
& NBD_FLAG_READ_ONLY
&&
1860 (request
->type
== NBD_CMD_WRITE
||
1861 request
->type
== NBD_CMD_WRITE_ZEROES
||
1862 request
->type
== NBD_CMD_TRIM
)) {
1863 error_setg(errp
, "Export is read-only");
1866 if (request
->from
> client
->exp
->size
||
1867 request
->from
+ request
->len
> client
->exp
->size
) {
1868 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
1869 ", Size: %" PRIu64
, request
->from
, request
->len
,
1870 (uint64_t)client
->exp
->size
);
1871 return (request
->type
== NBD_CMD_WRITE
||
1872 request
->type
== NBD_CMD_WRITE_ZEROES
) ? -ENOSPC
: -EINVAL
;
1874 valid_flags
= NBD_CMD_FLAG_FUA
;
1875 if (request
->type
== NBD_CMD_READ
&& client
->structured_reply
) {
1876 valid_flags
|= NBD_CMD_FLAG_DF
;
1877 } else if (request
->type
== NBD_CMD_WRITE_ZEROES
) {
1878 valid_flags
|= NBD_CMD_FLAG_NO_HOLE
;
1879 } else if (request
->type
== NBD_CMD_BLOCK_STATUS
) {
1880 valid_flags
|= NBD_CMD_FLAG_REQ_ONE
;
1882 if (request
->flags
& ~valid_flags
) {
1883 error_setg(errp
, "unsupported flags for command %s (got 0x%x)",
1884 nbd_cmd_lookup(request
->type
), request
->flags
);
1891 /* Send simple reply without a payload, or a structured error
1892 * @error_msg is ignored if @ret >= 0
1893 * Returns 0 if connection is still live, -errno on failure to talk to client
1895 static coroutine_fn
int nbd_send_generic_reply(NBDClient
*client
,
1898 const char *error_msg
,
1901 if (client
->structured_reply
&& ret
< 0) {
1902 return nbd_co_send_structured_error(client
, handle
, -ret
, error_msg
,
1905 return nbd_co_send_simple_reply(client
, handle
, ret
< 0 ? -ret
: 0,
1910 /* Handle NBD_CMD_READ request.
1911 * Return -errno if sending fails. Other errors are reported directly to the
1912 * client as an error reply. */
1913 static coroutine_fn
int nbd_do_cmd_read(NBDClient
*client
, NBDRequest
*request
,
1914 uint8_t *data
, Error
**errp
)
1917 NBDExport
*exp
= client
->exp
;
1919 assert(request
->type
== NBD_CMD_READ
);
1921 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1922 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1923 ret
= blk_co_flush(exp
->blk
);
1925 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1926 "flush failed", errp
);
1930 if (client
->structured_reply
&& !(request
->flags
& NBD_CMD_FLAG_DF
) &&
1932 return nbd_co_send_sparse_read(client
, request
->handle
, request
->from
,
1933 data
, request
->len
, errp
);
1936 ret
= blk_pread(exp
->blk
, request
->from
+ exp
->dev_offset
, data
,
1939 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1940 "reading from file failed", errp
);
1943 if (client
->structured_reply
) {
1945 return nbd_co_send_structured_read(client
, request
->handle
,
1946 request
->from
, data
,
1947 request
->len
, true, errp
);
1949 return nbd_co_send_structured_done(client
, request
->handle
, errp
);
1952 return nbd_co_send_simple_reply(client
, request
->handle
, 0,
1953 data
, request
->len
, errp
);
1957 /* Handle NBD request.
1958 * Return -errno if sending fails. Other errors are reported directly to the
1959 * client as an error reply. */
1960 static coroutine_fn
int nbd_handle_request(NBDClient
*client
,
1961 NBDRequest
*request
,
1962 uint8_t *data
, Error
**errp
)
1966 NBDExport
*exp
= client
->exp
;
1969 switch (request
->type
) {
1971 return nbd_do_cmd_read(client
, request
, data
, errp
);
1975 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1976 flags
|= BDRV_REQ_FUA
;
1978 ret
= blk_pwrite(exp
->blk
, request
->from
+ exp
->dev_offset
,
1979 data
, request
->len
, flags
);
1980 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1981 "writing to file failed", errp
);
1983 case NBD_CMD_WRITE_ZEROES
:
1985 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
1986 flags
|= BDRV_REQ_FUA
;
1988 if (!(request
->flags
& NBD_CMD_FLAG_NO_HOLE
)) {
1989 flags
|= BDRV_REQ_MAY_UNMAP
;
1991 ret
= blk_pwrite_zeroes(exp
->blk
, request
->from
+ exp
->dev_offset
,
1992 request
->len
, flags
);
1993 return nbd_send_generic_reply(client
, request
->handle
, ret
,
1994 "writing to file failed", errp
);
1997 /* unreachable, thanks to special case in nbd_co_receive_request() */
2001 ret
= blk_co_flush(exp
->blk
);
2002 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2003 "flush failed", errp
);
2006 ret
= blk_co_pdiscard(exp
->blk
, request
->from
+ exp
->dev_offset
,
2008 if (ret
== 0 && request
->flags
& NBD_CMD_FLAG_FUA
) {
2009 ret
= blk_co_flush(exp
->blk
);
2011 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2012 "discard failed", errp
);
2014 case NBD_CMD_BLOCK_STATUS
:
2015 if (!request
->len
) {
2016 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2017 "need non-zero length", errp
);
2019 if (client
->export_meta
.valid
&& client
->export_meta
.base_allocation
) {
2020 return nbd_co_send_block_status(client
, request
->handle
,
2021 blk_bs(exp
->blk
), request
->from
,
2023 NBD_META_ID_BASE_ALLOCATION
, errp
);
2025 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2026 "CMD_BLOCK_STATUS not negotiated",
2031 msg
= g_strdup_printf("invalid request type (%" PRIu32
") received",
2033 ret
= nbd_send_generic_reply(client
, request
->handle
, -EINVAL
, msg
,
2040 /* Owns a reference to the NBDClient passed as opaque. */
2041 static coroutine_fn
void nbd_trip(void *opaque
)
2043 NBDClient
*client
= opaque
;
2044 NBDRequestData
*req
;
2045 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
2047 Error
*local_err
= NULL
;
2050 if (client
->closing
) {
2051 nbd_client_put(client
);
2055 req
= nbd_request_get(client
);
2056 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
2057 client
->recv_coroutine
= NULL
;
2059 if (client
->closing
) {
2061 * The client may be closed when we are blocked in
2062 * nbd_co_receive_request()
2067 nbd_client_receive_next_request(client
);
2073 /* It wans't -EIO, so, according to nbd_co_receive_request()
2074 * semantics, we should return the error to the client. */
2075 Error
*export_err
= local_err
;
2078 ret
= nbd_send_generic_reply(client
, request
.handle
, -EINVAL
,
2079 error_get_pretty(export_err
), &local_err
);
2080 error_free(export_err
);
2082 ret
= nbd_handle_request(client
, &request
, req
->data
, &local_err
);
2085 error_prepend(&local_err
, "Failed to send reply: ");
2089 /* We must disconnect after NBD_CMD_WRITE if we did not
2092 if (!req
->complete
) {
2093 error_setg(&local_err
, "Request handling failed in intermediate state");
2098 nbd_request_put(req
);
2099 nbd_client_put(client
);
2104 error_reportf_err(local_err
, "Disconnect client, due to: ");
2106 nbd_request_put(req
);
2107 client_close(client
, true);
2108 nbd_client_put(client
);
2111 static void nbd_client_receive_next_request(NBDClient
*client
)
2113 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
) {
2114 nbd_client_get(client
);
2115 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
2116 aio_co_schedule(client
->exp
->ctx
, client
->recv_coroutine
);
2120 static coroutine_fn
void nbd_co_client_start(void *opaque
)
2122 NBDClient
*client
= opaque
;
2123 NBDExport
*exp
= client
->exp
;
2124 Error
*local_err
= NULL
;
2127 nbd_export_get(exp
);
2128 QTAILQ_INSERT_TAIL(&exp
->clients
, client
, next
);
2130 qemu_co_mutex_init(&client
->send_lock
);
2132 if (nbd_negotiate(client
, &local_err
)) {
2134 error_report_err(local_err
);
2136 client_close(client
, false);
2140 nbd_client_receive_next_request(client
);
2144 * Create a new client listener on the given export @exp, using the
2145 * given channel @sioc. Begin servicing it in a coroutine. When the
2146 * connection closes, call @close_fn with an indication of whether the
2147 * client completed negotiation.
2149 void nbd_client_new(NBDExport
*exp
,
2150 QIOChannelSocket
*sioc
,
2151 QCryptoTLSCreds
*tlscreds
,
2152 const char *tlsaclname
,
2153 void (*close_fn
)(NBDClient
*, bool))
2158 client
= g_new0(NBDClient
, 1);
2159 client
->refcount
= 1;
2161 client
->tlscreds
= tlscreds
;
2163 object_ref(OBJECT(client
->tlscreds
));
2165 client
->tlsaclname
= g_strdup(tlsaclname
);
2166 client
->sioc
= sioc
;
2167 object_ref(OBJECT(client
->sioc
));
2168 client
->ioc
= QIO_CHANNEL(sioc
);
2169 object_ref(OBJECT(client
->ioc
));
2170 client
->close_fn
= close_fn
;
2172 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
2173 qemu_coroutine_enter(co
);