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"
22 #include "block/export.h"
23 #include "qapi/error.h"
24 #include "qemu/queue.h"
26 #include "nbd-internal.h"
27 #include "qemu/units.h"
29 #define NBD_META_ID_BASE_ALLOCATION 0
30 #define NBD_META_ID_DIRTY_BITMAP 1
33 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
34 * constant. If an increase is needed, note that the NBD protocol
35 * recommends no larger than 32 mb, so that the client won't consider
36 * the reply as a denial of service attack.
38 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
40 static int system_errno_to_nbd_errno(int err
)
61 #if ENOTSUP != EOPNOTSUPP
73 /* Definitions for opaque data types */
75 typedef struct NBDRequestData NBDRequestData
;
77 struct NBDRequestData
{
78 QSIMPLEQ_ENTRY(NBDRequestData
) entry
;
91 QTAILQ_HEAD(, NBDClient
) clients
;
92 QTAILQ_ENTRY(NBDExport
) next
;
94 BlockBackend
*eject_notifier_blk
;
95 Notifier eject_notifier
;
97 BdrvDirtyBitmap
*export_bitmap
;
98 char *export_bitmap_context
;
101 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
103 /* NBDExportMetaContexts represents a list of contexts to be exported,
104 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
105 * NBD_OPT_LIST_META_CONTEXT. */
106 typedef struct NBDExportMetaContexts
{
108 bool valid
; /* means that negotiation of the option finished without
110 bool base_allocation
; /* export base:allocation context (block status) */
111 bool bitmap
; /* export qemu:dirty-bitmap:<export bitmap name> */
112 } NBDExportMetaContexts
;
116 void (*close_fn
)(NBDClient
*client
, bool negotiated
);
119 QCryptoTLSCreds
*tlscreds
;
121 QIOChannelSocket
*sioc
; /* The underlying data channel */
122 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
124 Coroutine
*recv_coroutine
;
127 Coroutine
*send_coroutine
;
129 QTAILQ_ENTRY(NBDClient
) next
;
133 uint32_t check_align
; /* If non-zero, check for aligned client requests */
135 bool structured_reply
;
136 NBDExportMetaContexts export_meta
;
138 uint32_t opt
; /* Current option being negotiated */
139 uint32_t optlen
; /* remaining length of data in ioc for the option being
143 static void nbd_client_receive_next_request(NBDClient
*client
);
145 /* Basic flow for negotiation
172 static inline void set_be_option_rep(NBDOptionReply
*rep
, uint32_t option
,
173 uint32_t type
, uint32_t length
)
175 stq_be_p(&rep
->magic
, NBD_REP_MAGIC
);
176 stl_be_p(&rep
->option
, option
);
177 stl_be_p(&rep
->type
, type
);
178 stl_be_p(&rep
->length
, length
);
181 /* Send a reply header, including length, but no payload.
182 * Return -errno on error, 0 on success. */
183 static int nbd_negotiate_send_rep_len(NBDClient
*client
, uint32_t type
,
184 uint32_t len
, Error
**errp
)
188 trace_nbd_negotiate_send_rep_len(client
->opt
, nbd_opt_lookup(client
->opt
),
189 type
, nbd_rep_lookup(type
), len
);
191 assert(len
< NBD_MAX_BUFFER_SIZE
);
193 set_be_option_rep(&rep
, client
->opt
, type
, len
);
194 return nbd_write(client
->ioc
, &rep
, sizeof(rep
), errp
);
197 /* Send a reply header with default 0 length.
198 * Return -errno on error, 0 on success. */
199 static int nbd_negotiate_send_rep(NBDClient
*client
, uint32_t type
,
202 return nbd_negotiate_send_rep_len(client
, type
, 0, errp
);
205 /* Send an error reply.
206 * Return -errno on error, 0 on success. */
207 static int GCC_FMT_ATTR(4, 0)
208 nbd_negotiate_send_rep_verr(NBDClient
*client
, uint32_t type
,
209 Error
**errp
, const char *fmt
, va_list va
)
212 g_autofree
char *msg
= NULL
;
216 msg
= g_strdup_vprintf(fmt
, va
);
218 assert(len
< NBD_MAX_STRING_SIZE
);
219 trace_nbd_negotiate_send_rep_err(msg
);
220 ret
= nbd_negotiate_send_rep_len(client
, type
, len
, errp
);
224 if (nbd_write(client
->ioc
, msg
, len
, errp
) < 0) {
225 error_prepend(errp
, "write failed (error message): ");
233 * Return a malloc'd copy of @name suitable for use in an error reply.
236 nbd_sanitize_name(const char *name
)
238 if (strnlen(name
, 80) < 80) {
239 return g_strdup(name
);
241 /* XXX Should we also try to sanitize any control characters? */
242 return g_strdup_printf("%.80s...", name
);
245 /* Send an error reply.
246 * Return -errno on error, 0 on success. */
247 static int GCC_FMT_ATTR(4, 5)
248 nbd_negotiate_send_rep_err(NBDClient
*client
, uint32_t type
,
249 Error
**errp
, const char *fmt
, ...)
255 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
260 /* Drop remainder of the current option, and send a reply with the
261 * given error type and message. Return -errno on read or write
262 * failure; or 0 if connection is still live. */
263 static int GCC_FMT_ATTR(4, 0)
264 nbd_opt_vdrop(NBDClient
*client
, uint32_t type
, Error
**errp
,
265 const char *fmt
, va_list va
)
267 int ret
= nbd_drop(client
->ioc
, client
->optlen
, errp
);
271 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
276 static int GCC_FMT_ATTR(4, 5)
277 nbd_opt_drop(NBDClient
*client
, uint32_t type
, Error
**errp
,
278 const char *fmt
, ...)
284 ret
= nbd_opt_vdrop(client
, type
, errp
, fmt
, va
);
290 static int GCC_FMT_ATTR(3, 4)
291 nbd_opt_invalid(NBDClient
*client
, Error
**errp
, const char *fmt
, ...)
297 ret
= nbd_opt_vdrop(client
, NBD_REP_ERR_INVALID
, errp
, fmt
, va
);
303 /* Read size bytes from the unparsed payload of the current option.
304 * Return -errno on I/O error, 0 if option was completely handled by
305 * sending a reply about inconsistent lengths, or 1 on success. */
306 static int nbd_opt_read(NBDClient
*client
, void *buffer
, size_t size
,
309 if (size
> client
->optlen
) {
310 return nbd_opt_invalid(client
, errp
,
311 "Inconsistent lengths in option %s",
312 nbd_opt_lookup(client
->opt
));
314 client
->optlen
-= size
;
315 return qio_channel_read_all(client
->ioc
, buffer
, size
, errp
) < 0 ? -EIO
: 1;
318 /* Drop size bytes from the unparsed payload of the current option.
319 * Return -errno on I/O error, 0 if option was completely handled by
320 * sending a reply about inconsistent lengths, or 1 on success. */
321 static int nbd_opt_skip(NBDClient
*client
, size_t size
, Error
**errp
)
323 if (size
> client
->optlen
) {
324 return nbd_opt_invalid(client
, errp
,
325 "Inconsistent lengths in option %s",
326 nbd_opt_lookup(client
->opt
));
328 client
->optlen
-= size
;
329 return nbd_drop(client
->ioc
, size
, errp
) < 0 ? -EIO
: 1;
334 * Read a string with the format:
335 * uint32_t len (<= NBD_MAX_STRING_SIZE)
336 * len bytes string (not 0-terminated)
338 * On success, @name will be allocated.
339 * If @length is non-null, it will be set to the actual string length.
341 * Return -errno on I/O error, 0 if option was completely handled by
342 * sending a reply about inconsistent lengths, or 1 on success.
344 static int nbd_opt_read_name(NBDClient
*client
, char **name
, uint32_t *length
,
349 g_autofree
char *local_name
= NULL
;
352 ret
= nbd_opt_read(client
, &len
, sizeof(len
), errp
);
356 len
= cpu_to_be32(len
);
358 if (len
> NBD_MAX_STRING_SIZE
) {
359 return nbd_opt_invalid(client
, errp
,
360 "Invalid name length: %" PRIu32
, len
);
363 local_name
= g_malloc(len
+ 1);
364 ret
= nbd_opt_read(client
, local_name
, len
, errp
);
368 local_name
[len
] = '\0';
373 *name
= g_steal_pointer(&local_name
);
378 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
379 * Return -errno on error, 0 on success. */
380 static int nbd_negotiate_send_rep_list(NBDClient
*client
, NBDExport
*exp
,
384 size_t name_len
, desc_len
;
386 const char *name
= exp
->name
? exp
->name
: "";
387 const char *desc
= exp
->description
? exp
->description
: "";
388 QIOChannel
*ioc
= client
->ioc
;
391 trace_nbd_negotiate_send_rep_list(name
, desc
);
392 name_len
= strlen(name
);
393 desc_len
= strlen(desc
);
394 assert(name_len
<= NBD_MAX_STRING_SIZE
&& desc_len
<= NBD_MAX_STRING_SIZE
);
395 len
= name_len
+ desc_len
+ sizeof(len
);
396 ret
= nbd_negotiate_send_rep_len(client
, NBD_REP_SERVER
, len
, errp
);
401 len
= cpu_to_be32(name_len
);
402 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
403 error_prepend(errp
, "write failed (name length): ");
407 if (nbd_write(ioc
, name
, name_len
, errp
) < 0) {
408 error_prepend(errp
, "write failed (name buffer): ");
412 if (nbd_write(ioc
, desc
, desc_len
, errp
) < 0) {
413 error_prepend(errp
, "write failed (description buffer): ");
420 /* Process the NBD_OPT_LIST command, with a potential series of replies.
421 * Return -errno on error, 0 on success. */
422 static int nbd_negotiate_handle_list(NBDClient
*client
, Error
**errp
)
425 assert(client
->opt
== NBD_OPT_LIST
);
427 /* For each export, send a NBD_REP_SERVER reply. */
428 QTAILQ_FOREACH(exp
, &exports
, next
) {
429 if (nbd_negotiate_send_rep_list(client
, exp
, errp
)) {
433 /* Finish with a NBD_REP_ACK. */
434 return nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
437 static void nbd_check_meta_export(NBDClient
*client
)
439 client
->export_meta
.valid
&= client
->exp
== client
->export_meta
.exp
;
442 /* Send a reply to NBD_OPT_EXPORT_NAME.
443 * Return -errno on error, 0 on success. */
444 static int nbd_negotiate_handle_export_name(NBDClient
*client
, bool no_zeroes
,
448 g_autofree
char *name
= NULL
;
449 char buf
[NBD_REPLY_EXPORT_NAME_SIZE
] = "";
455 [20 .. xx] export name (length bytes)
458 [ 8 .. 9] export flags
459 [10 .. 133] reserved (0) [unless no_zeroes]
461 trace_nbd_negotiate_handle_export_name();
462 if (client
->optlen
> NBD_MAX_STRING_SIZE
) {
463 error_setg(errp
, "Bad length received");
466 name
= g_malloc(client
->optlen
+ 1);
467 if (nbd_read(client
->ioc
, name
, client
->optlen
, "export name", errp
) < 0) {
470 name
[client
->optlen
] = '\0';
473 trace_nbd_negotiate_handle_export_name_request(name
);
475 client
->exp
= nbd_export_find(name
);
477 error_setg(errp
, "export not found");
481 myflags
= client
->exp
->nbdflags
;
482 if (client
->structured_reply
) {
483 myflags
|= NBD_FLAG_SEND_DF
;
485 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
, myflags
);
486 stq_be_p(buf
, client
->exp
->size
);
487 stw_be_p(buf
+ 8, myflags
);
488 len
= no_zeroes
? 10 : sizeof(buf
);
489 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
491 error_prepend(errp
, "write failed: ");
495 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
496 blk_exp_ref(&client
->exp
->common
);
497 nbd_check_meta_export(client
);
502 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
503 * The buffer does NOT include the info type prefix.
504 * Return -errno on error, 0 if ready to send more. */
505 static int nbd_negotiate_send_info(NBDClient
*client
,
506 uint16_t info
, uint32_t length
, void *buf
,
511 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
512 rc
= nbd_negotiate_send_rep_len(client
, NBD_REP_INFO
,
513 sizeof(info
) + length
, errp
);
517 info
= cpu_to_be16(info
);
518 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
521 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
527 /* nbd_reject_length: Handle any unexpected payload.
528 * @fatal requests that we quit talking to the client, even if we are able
529 * to successfully send an error reply.
531 * -errno transmission error occurred or @fatal was requested, errp is set
532 * 0 error message successfully sent to client, errp is not set
534 static int nbd_reject_length(NBDClient
*client
, bool fatal
, Error
**errp
)
538 assert(client
->optlen
);
539 ret
= nbd_opt_invalid(client
, errp
, "option '%s' has unexpected length",
540 nbd_opt_lookup(client
->opt
));
542 error_setg(errp
, "option '%s' has unexpected length",
543 nbd_opt_lookup(client
->opt
));
549 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
550 * Return -errno on error, 0 if ready for next option, and 1 to move
551 * into transmission phase. */
552 static int nbd_negotiate_handle_info(NBDClient
*client
, Error
**errp
)
555 g_autofree
char *name
= NULL
;
560 bool sendname
= false;
561 bool blocksize
= false;
563 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
564 uint32_t check_align
= 0;
568 4 bytes: L, name length (can be 0)
570 2 bytes: N, number of requests (can be 0)
571 N * 2 bytes: N requests
573 rc
= nbd_opt_read_name(client
, &name
, &namelen
, errp
);
577 trace_nbd_negotiate_handle_export_name_request(name
);
579 rc
= nbd_opt_read(client
, &requests
, sizeof(requests
), errp
);
583 requests
= be16_to_cpu(requests
);
584 trace_nbd_negotiate_handle_info_requests(requests
);
586 rc
= nbd_opt_read(client
, &request
, sizeof(request
), errp
);
590 request
= be16_to_cpu(request
);
591 trace_nbd_negotiate_handle_info_request(request
,
592 nbd_info_lookup(request
));
593 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
594 * everything else is either a request we don't know or
595 * something we send regardless of request */
600 case NBD_INFO_BLOCK_SIZE
:
605 if (client
->optlen
) {
606 return nbd_reject_length(client
, false, errp
);
609 exp
= nbd_export_find(name
);
611 g_autofree
char *sane_name
= nbd_sanitize_name(name
);
613 return nbd_negotiate_send_rep_err(client
, NBD_REP_ERR_UNKNOWN
,
614 errp
, "export '%s' not present",
618 /* Don't bother sending NBD_INFO_NAME unless client requested it */
620 rc
= nbd_negotiate_send_info(client
, NBD_INFO_NAME
, namelen
, name
,
627 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
629 if (exp
->description
) {
630 size_t len
= strlen(exp
->description
);
632 assert(len
<= NBD_MAX_STRING_SIZE
);
633 rc
= nbd_negotiate_send_info(client
, NBD_INFO_DESCRIPTION
,
634 len
, exp
->description
, errp
);
640 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
641 * according to whether the client requested it, and according to
642 * whether this is OPT_INFO or OPT_GO. */
643 /* minimum - 1 for back-compat, or actual if client will obey it. */
644 if (client
->opt
== NBD_OPT_INFO
|| blocksize
) {
645 check_align
= sizes
[0] = blk_get_request_alignment(exp
->common
.blk
);
649 assert(sizes
[0] <= NBD_MAX_BUFFER_SIZE
);
650 /* preferred - Hard-code to 4096 for now.
651 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
652 sizes
[1] = MAX(4096, sizes
[0]);
653 /* maximum - At most 32M, but smaller as appropriate. */
654 sizes
[2] = MIN(blk_get_max_transfer(exp
->common
.blk
), NBD_MAX_BUFFER_SIZE
);
655 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
656 sizes
[0] = cpu_to_be32(sizes
[0]);
657 sizes
[1] = cpu_to_be32(sizes
[1]);
658 sizes
[2] = cpu_to_be32(sizes
[2]);
659 rc
= nbd_negotiate_send_info(client
, NBD_INFO_BLOCK_SIZE
,
660 sizeof(sizes
), sizes
, errp
);
665 /* Send NBD_INFO_EXPORT always */
666 myflags
= exp
->nbdflags
;
667 if (client
->structured_reply
) {
668 myflags
|= NBD_FLAG_SEND_DF
;
670 trace_nbd_negotiate_new_style_size_flags(exp
->size
, myflags
);
671 stq_be_p(buf
, exp
->size
);
672 stw_be_p(buf
+ 8, myflags
);
673 rc
= nbd_negotiate_send_info(client
, NBD_INFO_EXPORT
,
674 sizeof(buf
), buf
, errp
);
680 * If the client is just asking for NBD_OPT_INFO, but forgot to
681 * request block sizes in a situation that would impact
682 * performance, then return an error. But for NBD_OPT_GO, we
683 * tolerate all clients, regardless of alignments.
685 if (client
->opt
== NBD_OPT_INFO
&& !blocksize
&&
686 blk_get_request_alignment(exp
->common
.blk
) > 1) {
687 return nbd_negotiate_send_rep_err(client
,
688 NBD_REP_ERR_BLOCK_SIZE_REQD
,
690 "request NBD_INFO_BLOCK_SIZE to "
695 rc
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
700 if (client
->opt
== NBD_OPT_GO
) {
702 client
->check_align
= check_align
;
703 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
704 blk_exp_ref(&client
->exp
->common
);
705 nbd_check_meta_export(client
);
712 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
713 * new channel for all further (now-encrypted) communication. */
714 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
719 struct NBDTLSHandshakeData data
= { 0 };
721 assert(client
->opt
== NBD_OPT_STARTTLS
);
723 trace_nbd_negotiate_handle_starttls();
726 if (nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
) < 0) {
730 tioc
= qio_channel_tls_new_server(ioc
,
738 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
739 trace_nbd_negotiate_handle_starttls_handshake();
740 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
741 qio_channel_tls_handshake(tioc
,
747 if (!data
.complete
) {
748 g_main_loop_run(data
.loop
);
750 g_main_loop_unref(data
.loop
);
752 object_unref(OBJECT(tioc
));
753 error_propagate(errp
, data
.error
);
757 return QIO_CHANNEL(tioc
);
760 /* nbd_negotiate_send_meta_context
762 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
764 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
766 static int nbd_negotiate_send_meta_context(NBDClient
*client
,
771 NBDOptionReplyMetaContext opt
;
772 struct iovec iov
[] = {
773 {.iov_base
= &opt
, .iov_len
= sizeof(opt
)},
774 {.iov_base
= (void *)context
, .iov_len
= strlen(context
)}
777 assert(iov
[1].iov_len
<= NBD_MAX_STRING_SIZE
);
778 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
782 trace_nbd_negotiate_meta_query_reply(context
, context_id
);
783 set_be_option_rep(&opt
.h
, client
->opt
, NBD_REP_META_CONTEXT
,
784 sizeof(opt
) - sizeof(opt
.h
) + iov
[1].iov_len
);
785 stl_be_p(&opt
.context_id
, context_id
);
787 return qio_channel_writev_all(client
->ioc
, iov
, 2, errp
) < 0 ? -EIO
: 0;
790 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
791 * @match is never set to false.
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.
796 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
797 * It only means that there are no errors.
799 static int nbd_meta_pattern(NBDClient
*client
, const char *pattern
, bool *match
,
804 size_t len
= strlen(pattern
);
808 query
= g_malloc(len
);
809 ret
= nbd_opt_read(client
, query
, len
, errp
);
815 if (strncmp(query
, pattern
, len
) == 0) {
816 trace_nbd_negotiate_meta_query_parse(pattern
);
819 trace_nbd_negotiate_meta_query_skip("pattern not matched");
827 * Read @len bytes, and set @match to true if they match @pattern, or if @len
828 * is 0 and the client is performing _LIST_. @match is never set to false.
830 * Return -errno on I/O error, 0 if option was completely handled by
831 * sending a reply about inconsistent lengths, or 1 on success.
833 * Note: return code = 1 doesn't mean that we've read exactly @pattern.
834 * It only means that there are no errors.
836 static int nbd_meta_empty_or_pattern(NBDClient
*client
, const char *pattern
,
837 uint32_t len
, bool *match
, Error
**errp
)
840 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
843 trace_nbd_negotiate_meta_query_parse("empty");
847 if (len
!= strlen(pattern
)) {
848 trace_nbd_negotiate_meta_query_skip("different lengths");
849 return nbd_opt_skip(client
, len
, errp
);
852 return nbd_meta_pattern(client
, pattern
, match
, errp
);
855 /* nbd_meta_base_query
857 * Handle queries to 'base' namespace. For now, only the base:allocation
858 * context is available. 'len' is the amount of text remaining to be read from
859 * the current name, after the 'base:' portion has been stripped.
861 * Return -errno on I/O error, 0 if option was completely handled by
862 * sending a reply about inconsistent lengths, or 1 on success.
864 static int nbd_meta_base_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
865 uint32_t len
, Error
**errp
)
867 return nbd_meta_empty_or_pattern(client
, "allocation", len
,
868 &meta
->base_allocation
, errp
);
871 /* nbd_meta_bitmap_query
873 * Handle query to 'qemu:' namespace.
874 * @len is the amount of text remaining to be read from the current name, after
875 * the 'qemu:' portion has been stripped.
877 * Return -errno on I/O error, 0 if option was completely handled by
878 * sending a reply about inconsistent lengths, or 1 on success. */
879 static int nbd_meta_qemu_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
880 uint32_t len
, Error
**errp
)
882 bool dirty_bitmap
= false;
883 size_t dirty_bitmap_len
= strlen("dirty-bitmap:");
886 if (!meta
->exp
->export_bitmap
) {
887 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
888 return nbd_opt_skip(client
, len
, errp
);
892 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
895 trace_nbd_negotiate_meta_query_parse("empty");
899 if (len
< dirty_bitmap_len
) {
900 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
901 return nbd_opt_skip(client
, len
, errp
);
904 len
-= dirty_bitmap_len
;
905 ret
= nbd_meta_pattern(client
, "dirty-bitmap:", &dirty_bitmap
, errp
);
910 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
911 return nbd_opt_skip(client
, len
, errp
);
914 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
916 return nbd_meta_empty_or_pattern(
917 client
, meta
->exp
->export_bitmap_context
+
918 strlen("qemu:dirty_bitmap:"), len
, &meta
->bitmap
, errp
);
921 /* nbd_negotiate_meta_query
923 * Parse namespace name and call corresponding function to parse body of the
926 * The only supported namespaces are 'base' and 'qemu'.
928 * The function aims not wasting time and memory to read long unknown namespace
931 * Return -errno on I/O error, 0 if option was completely handled by
932 * sending a reply about inconsistent lengths, or 1 on success. */
933 static int nbd_negotiate_meta_query(NBDClient
*client
,
934 NBDExportMetaContexts
*meta
, Error
**errp
)
937 * Both 'qemu' and 'base' namespaces have length = 5 including a
938 * colon. If another length namespace is later introduced, this
939 * should certainly be refactored.
946 ret
= nbd_opt_read(client
, &len
, sizeof(len
), errp
);
950 len
= cpu_to_be32(len
);
952 if (len
> NBD_MAX_STRING_SIZE
) {
953 trace_nbd_negotiate_meta_query_skip("length too long");
954 return nbd_opt_skip(client
, len
, errp
);
957 trace_nbd_negotiate_meta_query_skip("length too short");
958 return nbd_opt_skip(client
, len
, errp
);
962 ret
= nbd_opt_read(client
, ns
, ns_len
, errp
);
967 if (!strncmp(ns
, "base:", ns_len
)) {
968 trace_nbd_negotiate_meta_query_parse("base:");
969 return nbd_meta_base_query(client
, meta
, len
, errp
);
970 } else if (!strncmp(ns
, "qemu:", ns_len
)) {
971 trace_nbd_negotiate_meta_query_parse("qemu:");
972 return nbd_meta_qemu_query(client
, meta
, len
, errp
);
975 trace_nbd_negotiate_meta_query_skip("unknown namespace");
976 return nbd_opt_skip(client
, len
, errp
);
979 /* nbd_negotiate_meta_queries
980 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
982 * Return -errno on I/O error, or 0 if option was completely handled. */
983 static int nbd_negotiate_meta_queries(NBDClient
*client
,
984 NBDExportMetaContexts
*meta
, Error
**errp
)
987 g_autofree
char *export_name
= NULL
;
988 NBDExportMetaContexts local_meta
;
992 if (!client
->structured_reply
) {
993 return nbd_opt_invalid(client
, errp
,
994 "request option '%s' when structured reply "
996 nbd_opt_lookup(client
->opt
));
999 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
1000 /* Only change the caller's meta on SET. */
1004 memset(meta
, 0, sizeof(*meta
));
1006 ret
= nbd_opt_read_name(client
, &export_name
, NULL
, errp
);
1011 meta
->exp
= nbd_export_find(export_name
);
1012 if (meta
->exp
== NULL
) {
1013 g_autofree
char *sane_name
= nbd_sanitize_name(export_name
);
1015 return nbd_opt_drop(client
, NBD_REP_ERR_UNKNOWN
, errp
,
1016 "export '%s' not present", sane_name
);
1019 ret
= nbd_opt_read(client
, &nb_queries
, sizeof(nb_queries
), errp
);
1023 nb_queries
= cpu_to_be32(nb_queries
);
1024 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client
->opt
),
1025 export_name
, nb_queries
);
1027 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
&& !nb_queries
) {
1028 /* enable all known contexts */
1029 meta
->base_allocation
= true;
1030 meta
->bitmap
= !!meta
->exp
->export_bitmap
;
1032 for (i
= 0; i
< nb_queries
; ++i
) {
1033 ret
= nbd_negotiate_meta_query(client
, meta
, errp
);
1040 if (meta
->base_allocation
) {
1041 ret
= nbd_negotiate_send_meta_context(client
, "base:allocation",
1042 NBD_META_ID_BASE_ALLOCATION
,
1050 ret
= nbd_negotiate_send_meta_context(client
,
1051 meta
->exp
->export_bitmap_context
,
1052 NBD_META_ID_DIRTY_BITMAP
,
1059 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1067 /* nbd_negotiate_options
1068 * Process all NBD_OPT_* client option commands, during fixed newstyle
1071 * -errno on error, errp is set
1072 * 0 on successful negotiation, errp is not set
1073 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1076 static int nbd_negotiate_options(NBDClient
*client
, Error
**errp
)
1079 bool fixedNewstyle
= false;
1080 bool no_zeroes
= false;
1083 [ 0 .. 3] client flags
1085 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1086 [ 0 .. 7] NBD_OPTS_MAGIC
1087 [ 8 .. 11] NBD option
1088 [12 .. 15] Data length
1091 [ 0 .. 7] NBD_OPTS_MAGIC
1092 [ 8 .. 11] Second NBD option
1093 [12 .. 15] Data length
1097 if (nbd_read32(client
->ioc
, &flags
, "flags", errp
) < 0) {
1100 trace_nbd_negotiate_options_flags(flags
);
1101 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
1102 fixedNewstyle
= true;
1103 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
1105 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
1107 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
1110 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
1116 uint32_t option
, length
;
1119 if (nbd_read64(client
->ioc
, &magic
, "opts magic", errp
) < 0) {
1122 trace_nbd_negotiate_options_check_magic(magic
);
1123 if (magic
!= NBD_OPTS_MAGIC
) {
1124 error_setg(errp
, "Bad magic received");
1128 if (nbd_read32(client
->ioc
, &option
, "option", errp
) < 0) {
1131 client
->opt
= option
;
1133 if (nbd_read32(client
->ioc
, &length
, "option length", errp
) < 0) {
1136 assert(!client
->optlen
);
1137 client
->optlen
= length
;
1139 if (length
> NBD_MAX_BUFFER_SIZE
) {
1140 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1141 length
, NBD_MAX_BUFFER_SIZE
);
1145 trace_nbd_negotiate_options_check_option(option
,
1146 nbd_opt_lookup(option
));
1147 if (client
->tlscreds
&&
1148 client
->ioc
== (QIOChannel
*)client
->sioc
) {
1150 if (!fixedNewstyle
) {
1151 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
1155 case NBD_OPT_STARTTLS
:
1157 /* Unconditionally drop the connection if the client
1158 * can't start a TLS negotiation correctly */
1159 return nbd_reject_length(client
, true, errp
);
1161 tioc
= nbd_negotiate_handle_starttls(client
, errp
);
1166 object_unref(OBJECT(client
->ioc
));
1167 client
->ioc
= QIO_CHANNEL(tioc
);
1170 case NBD_OPT_EXPORT_NAME
:
1171 /* No way to return an error to client, so drop connection */
1172 error_setg(errp
, "Option 0x%x not permitted before TLS",
1177 /* Let the client keep trying, unless they asked to
1178 * quit. Always try to give an error back to the
1179 * client; but when replying to OPT_ABORT, be aware
1180 * that the client may hang up before receiving the
1181 * error, in which case we are fine ignoring the
1182 * resulting EPIPE. */
1183 ret
= nbd_opt_drop(client
, NBD_REP_ERR_TLS_REQD
,
1184 option
== NBD_OPT_ABORT
? NULL
: errp
,
1186 " not permitted before TLS", option
);
1187 if (option
== NBD_OPT_ABORT
) {
1192 } else if (fixedNewstyle
) {
1196 ret
= nbd_reject_length(client
, false, errp
);
1198 ret
= nbd_negotiate_handle_list(client
, errp
);
1203 /* NBD spec says we must try to reply before
1204 * disconnecting, but that we must also tolerate
1205 * guests that don't wait for our reply. */
1206 nbd_negotiate_send_rep(client
, NBD_REP_ACK
, NULL
);
1209 case NBD_OPT_EXPORT_NAME
:
1210 return nbd_negotiate_handle_export_name(client
, no_zeroes
,
1215 ret
= nbd_negotiate_handle_info(client
, errp
);
1217 assert(option
== NBD_OPT_GO
);
1222 case NBD_OPT_STARTTLS
:
1224 ret
= nbd_reject_length(client
, false, errp
);
1225 } else if (client
->tlscreds
) {
1226 ret
= nbd_negotiate_send_rep_err(client
,
1227 NBD_REP_ERR_INVALID
, errp
,
1228 "TLS already enabled");
1230 ret
= nbd_negotiate_send_rep_err(client
,
1231 NBD_REP_ERR_POLICY
, errp
,
1232 "TLS not configured");
1236 case NBD_OPT_STRUCTURED_REPLY
:
1238 ret
= nbd_reject_length(client
, false, errp
);
1239 } else if (client
->structured_reply
) {
1240 ret
= nbd_negotiate_send_rep_err(
1241 client
, NBD_REP_ERR_INVALID
, errp
,
1242 "structured reply already negotiated");
1244 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1245 client
->structured_reply
= true;
1249 case NBD_OPT_LIST_META_CONTEXT
:
1250 case NBD_OPT_SET_META_CONTEXT
:
1251 ret
= nbd_negotiate_meta_queries(client
, &client
->export_meta
,
1256 ret
= nbd_opt_drop(client
, NBD_REP_ERR_UNSUP
, errp
,
1257 "Unsupported option %" PRIu32
" (%s)",
1258 option
, nbd_opt_lookup(option
));
1263 * If broken new-style we should drop the connection
1264 * for anything except NBD_OPT_EXPORT_NAME
1267 case NBD_OPT_EXPORT_NAME
:
1268 return nbd_negotiate_handle_export_name(client
, no_zeroes
,
1272 error_setg(errp
, "Unsupported option %" PRIu32
" (%s)",
1273 option
, nbd_opt_lookup(option
));
1285 * -errno on error, errp is set
1286 * 0 on successful negotiation, errp is not set
1287 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1290 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
1293 char buf
[NBD_OLDSTYLE_NEGOTIATE_SIZE
] = "";
1296 /* Old style negotiation header, no room for options
1297 [ 0 .. 7] passwd ("NBDMAGIC")
1298 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1300 [24 .. 27] export flags (zero-extended)
1301 [28 .. 151] reserved (0)
1303 New style negotiation header, client can send options
1304 [ 0 .. 7] passwd ("NBDMAGIC")
1305 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1306 [16 .. 17] server flags (0)
1307 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1310 qio_channel_set_blocking(client
->ioc
, false, NULL
);
1312 trace_nbd_negotiate_begin();
1313 memcpy(buf
, "NBDMAGIC", 8);
1315 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
1316 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
1318 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
1319 error_prepend(errp
, "write failed: ");
1322 ret
= nbd_negotiate_options(client
, errp
);
1325 error_prepend(errp
, "option negotiation failed: ");
1330 /* Attach the channel to the same AioContext as the export */
1331 if (client
->exp
&& client
->exp
->common
.ctx
) {
1332 qio_channel_attach_aio_context(client
->ioc
, client
->exp
->common
.ctx
);
1335 assert(!client
->optlen
);
1336 trace_nbd_negotiate_success();
1341 static int nbd_receive_request(QIOChannel
*ioc
, NBDRequest
*request
,
1344 uint8_t buf
[NBD_REQUEST_SIZE
];
1348 ret
= nbd_read(ioc
, buf
, sizeof(buf
), "request", errp
);
1354 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1355 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1356 [ 6 .. 7] type (NBD_CMD_READ, ...)
1362 magic
= ldl_be_p(buf
);
1363 request
->flags
= lduw_be_p(buf
+ 4);
1364 request
->type
= lduw_be_p(buf
+ 6);
1365 request
->handle
= ldq_be_p(buf
+ 8);
1366 request
->from
= ldq_be_p(buf
+ 16);
1367 request
->len
= ldl_be_p(buf
+ 24);
1369 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
1370 request
->from
, request
->len
);
1372 if (magic
!= NBD_REQUEST_MAGIC
) {
1373 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
1379 #define MAX_NBD_REQUESTS 16
1381 void nbd_client_get(NBDClient
*client
)
1386 void nbd_client_put(NBDClient
*client
)
1388 if (--client
->refcount
== 0) {
1389 /* The last reference should be dropped by client->close,
1390 * which is called by client_close.
1392 assert(client
->closing
);
1394 qio_channel_detach_aio_context(client
->ioc
);
1395 object_unref(OBJECT(client
->sioc
));
1396 object_unref(OBJECT(client
->ioc
));
1397 if (client
->tlscreds
) {
1398 object_unref(OBJECT(client
->tlscreds
));
1400 g_free(client
->tlsauthz
);
1402 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
1403 blk_exp_unref(&client
->exp
->common
);
1409 static void client_close(NBDClient
*client
, bool negotiated
)
1411 if (client
->closing
) {
1415 client
->closing
= true;
1417 /* Force requests to finish. They will drop their own references,
1418 * then we'll close the socket and free the NBDClient.
1420 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
1423 /* Also tell the client, so that they release their reference. */
1424 if (client
->close_fn
) {
1425 client
->close_fn(client
, negotiated
);
1429 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
1431 NBDRequestData
*req
;
1433 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
1434 client
->nb_requests
++;
1436 req
= g_new0(NBDRequestData
, 1);
1437 nbd_client_get(client
);
1438 req
->client
= client
;
1442 static void nbd_request_put(NBDRequestData
*req
)
1444 NBDClient
*client
= req
->client
;
1447 qemu_vfree(req
->data
);
1451 client
->nb_requests
--;
1452 nbd_client_receive_next_request(client
);
1454 nbd_client_put(client
);
1457 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1459 NBDExport
*exp
= opaque
;
1462 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1464 exp
->common
.ctx
= ctx
;
1466 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1467 qio_channel_attach_aio_context(client
->ioc
, ctx
);
1468 if (client
->recv_coroutine
) {
1469 aio_co_schedule(ctx
, client
->recv_coroutine
);
1471 if (client
->send_coroutine
) {
1472 aio_co_schedule(ctx
, client
->send_coroutine
);
1477 static void blk_aio_detach(void *opaque
)
1479 NBDExport
*exp
= opaque
;
1482 trace_nbd_blk_aio_detach(exp
->name
, exp
->common
.ctx
);
1484 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1485 qio_channel_detach_aio_context(client
->ioc
);
1488 exp
->common
.ctx
= NULL
;
1491 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1493 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1495 blk_exp_request_shutdown(&exp
->common
);
1498 void nbd_export_set_on_eject_blk(BlockExport
*exp
, BlockBackend
*blk
)
1500 NBDExport
*nbd_exp
= container_of(exp
, NBDExport
, common
);
1501 assert(exp
->drv
== &blk_exp_nbd
);
1502 assert(nbd_exp
->eject_notifier_blk
== NULL
);
1505 nbd_exp
->eject_notifier_blk
= blk
;
1506 nbd_exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1507 blk_add_remove_bs_notifier(blk
, &nbd_exp
->eject_notifier
);
1510 static int nbd_export_create(BlockExport
*blk_exp
, BlockExportOptions
*exp_args
,
1513 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1514 BlockExportOptionsNbd
*arg
= &exp_args
->u
.nbd
;
1515 BlockBackend
*blk
= blk_exp
->blk
;
1517 uint64_t perm
, shared_perm
;
1518 bool readonly
= !exp_args
->writable
;
1519 bool shared
= !exp_args
->writable
;
1522 assert(exp_args
->type
== BLOCK_EXPORT_TYPE_NBD
);
1524 if (!nbd_server_is_running()) {
1525 error_setg(errp
, "NBD server not running");
1529 if (!arg
->has_name
) {
1530 arg
->name
= exp_args
->node_name
;
1533 if (strlen(arg
->name
) > NBD_MAX_STRING_SIZE
) {
1534 error_setg(errp
, "export name '%s' too long", arg
->name
);
1538 if (arg
->description
&& strlen(arg
->description
) > NBD_MAX_STRING_SIZE
) {
1539 error_setg(errp
, "description '%s' too long", arg
->description
);
1543 if (nbd_export_find(arg
->name
)) {
1544 error_setg(errp
, "NBD server already has export named '%s'", arg
->name
);
1548 size
= blk_getlength(blk
);
1550 error_setg_errno(errp
, -size
,
1551 "Failed to determine the NBD export's length");
1555 /* Don't allow resize while the NBD server is running, otherwise we don't
1556 * care what happens with the node. */
1557 blk_get_perm(blk
, &perm
, &shared_perm
);
1558 ret
= blk_set_perm(blk
, perm
, shared_perm
& ~BLK_PERM_RESIZE
, errp
);
1563 blk_set_allow_aio_context_change(blk
, true);
1565 QTAILQ_INIT(&exp
->clients
);
1566 exp
->name
= g_strdup(arg
->name
);
1567 exp
->description
= g_strdup(arg
->description
);
1568 exp
->nbdflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_FLUSH
|
1569 NBD_FLAG_SEND_FUA
| NBD_FLAG_SEND_CACHE
);
1571 exp
->nbdflags
|= NBD_FLAG_READ_ONLY
;
1573 exp
->nbdflags
|= NBD_FLAG_CAN_MULTI_CONN
;
1576 exp
->nbdflags
|= (NBD_FLAG_SEND_TRIM
| NBD_FLAG_SEND_WRITE_ZEROES
|
1577 NBD_FLAG_SEND_FAST_ZERO
);
1579 exp
->size
= QEMU_ALIGN_DOWN(size
, BDRV_SECTOR_SIZE
);
1582 BlockDriverState
*bs
= blk_bs(blk
);
1583 BdrvDirtyBitmap
*bm
= NULL
;
1586 bm
= bdrv_find_dirty_bitmap(bs
, arg
->bitmap
);
1591 bs
= bdrv_filter_or_cow_bs(bs
);
1596 error_setg(errp
, "Bitmap '%s' is not found", arg
->bitmap
);
1600 if (bdrv_dirty_bitmap_check(bm
, BDRV_BITMAP_ALLOW_RO
, errp
)) {
1605 if (readonly
&& bdrv_is_writable(bs
) &&
1606 bdrv_dirty_bitmap_enabled(bm
)) {
1609 "Enabled bitmap '%s' incompatible with readonly export",
1614 bdrv_dirty_bitmap_set_busy(bm
, true);
1615 exp
->export_bitmap
= bm
;
1616 assert(strlen(arg
->bitmap
) <= BDRV_BITMAP_MAX_NAME_SIZE
);
1617 exp
->export_bitmap_context
= g_strdup_printf("qemu:dirty-bitmap:%s",
1619 assert(strlen(exp
->export_bitmap_context
) < NBD_MAX_STRING_SIZE
);
1622 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1624 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1630 g_free(exp
->description
);
1634 NBDExport
*nbd_export_find(const char *name
)
1637 QTAILQ_FOREACH(exp
, &exports
, next
) {
1638 if (strcmp(name
, exp
->name
) == 0) {
1647 nbd_export_aio_context(NBDExport
*exp
)
1649 return exp
->common
.ctx
;
1652 static void nbd_export_request_shutdown(BlockExport
*blk_exp
)
1654 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1655 NBDClient
*client
, *next
;
1657 blk_exp_ref(&exp
->common
);
1659 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1660 * close mode that stops advertising the export to new clients but
1661 * still permits existing clients to run to completion? Because of
1662 * that possibility, nbd_export_close() can be called more than
1663 * once on an export.
1665 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1666 client_close(client
, true);
1671 QTAILQ_REMOVE(&exports
, exp
, next
);
1673 blk_exp_unref(&exp
->common
);
1676 static void nbd_export_delete(BlockExport
*blk_exp
)
1678 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1680 assert(exp
->name
== NULL
);
1681 assert(QTAILQ_EMPTY(&exp
->clients
));
1683 g_free(exp
->description
);
1684 exp
->description
= NULL
;
1686 if (exp
->common
.blk
) {
1687 if (exp
->eject_notifier_blk
) {
1688 notifier_remove(&exp
->eject_notifier
);
1689 blk_unref(exp
->eject_notifier_blk
);
1691 blk_remove_aio_context_notifier(exp
->common
.blk
, blk_aio_attached
,
1692 blk_aio_detach
, exp
);
1695 if (exp
->export_bitmap
) {
1696 bdrv_dirty_bitmap_set_busy(exp
->export_bitmap
, false);
1697 g_free(exp
->export_bitmap_context
);
1701 const BlockExportDriver blk_exp_nbd
= {
1702 .type
= BLOCK_EXPORT_TYPE_NBD
,
1703 .instance_size
= sizeof(NBDExport
),
1704 .create
= nbd_export_create
,
1705 .delete = nbd_export_delete
,
1706 .request_shutdown
= nbd_export_request_shutdown
,
1709 static int coroutine_fn
nbd_co_send_iov(NBDClient
*client
, struct iovec
*iov
,
1710 unsigned niov
, Error
**errp
)
1714 g_assert(qemu_in_coroutine());
1715 qemu_co_mutex_lock(&client
->send_lock
);
1716 client
->send_coroutine
= qemu_coroutine_self();
1718 ret
= qio_channel_writev_all(client
->ioc
, iov
, niov
, errp
) < 0 ? -EIO
: 0;
1720 client
->send_coroutine
= NULL
;
1721 qemu_co_mutex_unlock(&client
->send_lock
);
1726 static inline void set_be_simple_reply(NBDSimpleReply
*reply
, uint64_t error
,
1729 stl_be_p(&reply
->magic
, NBD_SIMPLE_REPLY_MAGIC
);
1730 stl_be_p(&reply
->error
, error
);
1731 stq_be_p(&reply
->handle
, handle
);
1734 static int nbd_co_send_simple_reply(NBDClient
*client
,
1741 NBDSimpleReply reply
;
1742 int nbd_err
= system_errno_to_nbd_errno(error
);
1743 struct iovec iov
[] = {
1744 {.iov_base
= &reply
, .iov_len
= sizeof(reply
)},
1745 {.iov_base
= data
, .iov_len
= len
}
1748 trace_nbd_co_send_simple_reply(handle
, nbd_err
, nbd_err_lookup(nbd_err
),
1750 set_be_simple_reply(&reply
, nbd_err
, handle
);
1752 return nbd_co_send_iov(client
, iov
, len
? 2 : 1, errp
);
1755 static inline void set_be_chunk(NBDStructuredReplyChunk
*chunk
, uint16_t flags
,
1756 uint16_t type
, uint64_t handle
, uint32_t length
)
1758 stl_be_p(&chunk
->magic
, NBD_STRUCTURED_REPLY_MAGIC
);
1759 stw_be_p(&chunk
->flags
, flags
);
1760 stw_be_p(&chunk
->type
, type
);
1761 stq_be_p(&chunk
->handle
, handle
);
1762 stl_be_p(&chunk
->length
, length
);
1765 static int coroutine_fn
nbd_co_send_structured_done(NBDClient
*client
,
1769 NBDStructuredReplyChunk chunk
;
1770 struct iovec iov
[] = {
1771 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1774 trace_nbd_co_send_structured_done(handle
);
1775 set_be_chunk(&chunk
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_NONE
, handle
, 0);
1777 return nbd_co_send_iov(client
, iov
, 1, errp
);
1780 static int coroutine_fn
nbd_co_send_structured_read(NBDClient
*client
,
1788 NBDStructuredReadData chunk
;
1789 struct iovec iov
[] = {
1790 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1791 {.iov_base
= data
, .iov_len
= size
}
1795 trace_nbd_co_send_structured_read(handle
, offset
, data
, size
);
1796 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1797 NBD_REPLY_TYPE_OFFSET_DATA
, handle
,
1798 sizeof(chunk
) - sizeof(chunk
.h
) + size
);
1799 stq_be_p(&chunk
.offset
, offset
);
1801 return nbd_co_send_iov(client
, iov
, 2, errp
);
1804 static int coroutine_fn
nbd_co_send_structured_error(NBDClient
*client
,
1810 NBDStructuredError chunk
;
1811 int nbd_err
= system_errno_to_nbd_errno(error
);
1812 struct iovec iov
[] = {
1813 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1814 {.iov_base
= (char *)msg
, .iov_len
= msg
? strlen(msg
) : 0},
1818 trace_nbd_co_send_structured_error(handle
, nbd_err
,
1819 nbd_err_lookup(nbd_err
), msg
? msg
: "");
1820 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_ERROR
, handle
,
1821 sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1822 stl_be_p(&chunk
.error
, nbd_err
);
1823 stw_be_p(&chunk
.message_length
, iov
[1].iov_len
);
1825 return nbd_co_send_iov(client
, iov
, 1 + !!iov
[1].iov_len
, errp
);
1828 /* Do a sparse read and send the structured reply to the client.
1829 * Returns -errno if sending fails. bdrv_block_status_above() failure is
1830 * reported to the client, at which point this function succeeds.
1832 static int coroutine_fn
nbd_co_send_sparse_read(NBDClient
*client
,
1840 NBDExport
*exp
= client
->exp
;
1841 size_t progress
= 0;
1843 while (progress
< size
) {
1845 int status
= bdrv_block_status_above(blk_bs(exp
->common
.blk
), NULL
,
1847 size
- progress
, &pnum
, NULL
,
1852 char *msg
= g_strdup_printf("unable to check for holes: %s",
1855 ret
= nbd_co_send_structured_error(client
, handle
, -status
, msg
,
1860 assert(pnum
&& pnum
<= size
- progress
);
1861 final
= progress
+ pnum
== size
;
1862 if (status
& BDRV_BLOCK_ZERO
) {
1863 NBDStructuredReadHole chunk
;
1864 struct iovec iov
[] = {
1865 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1868 trace_nbd_co_send_structured_read_hole(handle
, offset
+ progress
,
1870 set_be_chunk(&chunk
.h
, final
? NBD_REPLY_FLAG_DONE
: 0,
1871 NBD_REPLY_TYPE_OFFSET_HOLE
,
1872 handle
, sizeof(chunk
) - sizeof(chunk
.h
));
1873 stq_be_p(&chunk
.offset
, offset
+ progress
);
1874 stl_be_p(&chunk
.length
, pnum
);
1875 ret
= nbd_co_send_iov(client
, iov
, 1, errp
);
1877 ret
= blk_pread(exp
->common
.blk
, offset
+ progress
,
1878 data
+ progress
, pnum
);
1880 error_setg_errno(errp
, -ret
, "reading from file failed");
1883 ret
= nbd_co_send_structured_read(client
, handle
, offset
+ progress
,
1884 data
+ progress
, pnum
, final
,
1896 typedef struct NBDExtentArray
{
1898 unsigned int nb_alloc
;
1900 uint64_t total_length
;
1902 bool converted_to_be
;
1905 static NBDExtentArray
*nbd_extent_array_new(unsigned int nb_alloc
)
1907 NBDExtentArray
*ea
= g_new0(NBDExtentArray
, 1);
1909 ea
->nb_alloc
= nb_alloc
;
1910 ea
->extents
= g_new(NBDExtent
, nb_alloc
);
1916 static void nbd_extent_array_free(NBDExtentArray
*ea
)
1918 g_free(ea
->extents
);
1921 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray
, nbd_extent_array_free
);
1923 /* Further modifications of the array after conversion are abandoned */
1924 static void nbd_extent_array_convert_to_be(NBDExtentArray
*ea
)
1928 assert(!ea
->converted_to_be
);
1929 ea
->can_add
= false;
1930 ea
->converted_to_be
= true;
1932 for (i
= 0; i
< ea
->count
; i
++) {
1933 ea
->extents
[i
].flags
= cpu_to_be32(ea
->extents
[i
].flags
);
1934 ea
->extents
[i
].length
= cpu_to_be32(ea
->extents
[i
].length
);
1939 * Add extent to NBDExtentArray. If extent can't be added (no available space),
1941 * For safety, when returning -1 for the first time, .can_add is set to false,
1942 * further call to nbd_extent_array_add() will crash.
1943 * (to avoid the situation, when after failing to add an extent (returned -1),
1944 * user miss this failure and add another extent, which is successfully added
1945 * (array is full, but new extent may be squashed into the last one), then we
1946 * have invalid array with skipped extent)
1948 static int nbd_extent_array_add(NBDExtentArray
*ea
,
1949 uint32_t length
, uint32_t flags
)
1951 assert(ea
->can_add
);
1957 /* Extend previous extent if flags are the same */
1958 if (ea
->count
> 0 && flags
== ea
->extents
[ea
->count
- 1].flags
) {
1959 uint64_t sum
= (uint64_t)length
+ ea
->extents
[ea
->count
- 1].length
;
1961 if (sum
<= UINT32_MAX
) {
1962 ea
->extents
[ea
->count
- 1].length
= sum
;
1963 ea
->total_length
+= length
;
1968 if (ea
->count
>= ea
->nb_alloc
) {
1969 ea
->can_add
= false;
1973 ea
->total_length
+= length
;
1974 ea
->extents
[ea
->count
] = (NBDExtent
) {.length
= length
, .flags
= flags
};
1980 static int blockstatus_to_extents(BlockDriverState
*bs
, uint64_t offset
,
1981 uint64_t bytes
, NBDExtentArray
*ea
)
1986 int ret
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &num
,
1993 flags
= (ret
& BDRV_BLOCK_ALLOCATED
? 0 : NBD_STATE_HOLE
) |
1994 (ret
& BDRV_BLOCK_ZERO
? NBD_STATE_ZERO
: 0);
1996 if (nbd_extent_array_add(ea
, num
, flags
) < 0) {
2008 * nbd_co_send_extents
2010 * @ea is converted to BE by the function
2011 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2013 static int nbd_co_send_extents(NBDClient
*client
, uint64_t handle
,
2015 bool last
, uint32_t context_id
, Error
**errp
)
2017 NBDStructuredMeta chunk
;
2018 struct iovec iov
[] = {
2019 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
2020 {.iov_base
= ea
->extents
, .iov_len
= ea
->count
* sizeof(ea
->extents
[0])}
2023 nbd_extent_array_convert_to_be(ea
);
2025 trace_nbd_co_send_extents(handle
, ea
->count
, context_id
, ea
->total_length
,
2027 set_be_chunk(&chunk
.h
, last
? NBD_REPLY_FLAG_DONE
: 0,
2028 NBD_REPLY_TYPE_BLOCK_STATUS
,
2029 handle
, sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
2030 stl_be_p(&chunk
.context_id
, context_id
);
2032 return nbd_co_send_iov(client
, iov
, 2, errp
);
2035 /* Get block status from the exported device and send it to the client */
2036 static int nbd_co_send_block_status(NBDClient
*client
, uint64_t handle
,
2037 BlockDriverState
*bs
, uint64_t offset
,
2038 uint32_t length
, bool dont_fragment
,
2039 bool last
, uint32_t context_id
,
2043 unsigned int nb_extents
= dont_fragment
? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS
;
2044 g_autoptr(NBDExtentArray
) ea
= nbd_extent_array_new(nb_extents
);
2046 ret
= blockstatus_to_extents(bs
, offset
, length
, ea
);
2048 return nbd_co_send_structured_error(
2049 client
, handle
, -ret
, "can't get block status", errp
);
2052 return nbd_co_send_extents(client
, handle
, ea
, last
, context_id
, errp
);
2055 /* Populate @ea from a dirty bitmap. */
2056 static void bitmap_to_extents(BdrvDirtyBitmap
*bitmap
,
2057 uint64_t offset
, uint64_t length
,
2060 int64_t start
, dirty_start
, dirty_count
;
2061 int64_t end
= offset
+ length
;
2064 bdrv_dirty_bitmap_lock(bitmap
);
2066 for (start
= offset
;
2067 bdrv_dirty_bitmap_next_dirty_area(bitmap
, start
, end
, INT32_MAX
,
2068 &dirty_start
, &dirty_count
);
2069 start
= dirty_start
+ dirty_count
)
2071 if ((nbd_extent_array_add(es
, dirty_start
- start
, 0) < 0) ||
2072 (nbd_extent_array_add(es
, dirty_count
, NBD_STATE_DIRTY
) < 0))
2080 /* last non dirty extent */
2081 nbd_extent_array_add(es
, end
- start
, 0);
2084 bdrv_dirty_bitmap_unlock(bitmap
);
2087 static int nbd_co_send_bitmap(NBDClient
*client
, uint64_t handle
,
2088 BdrvDirtyBitmap
*bitmap
, uint64_t offset
,
2089 uint32_t length
, bool dont_fragment
, bool last
,
2090 uint32_t context_id
, Error
**errp
)
2092 unsigned int nb_extents
= dont_fragment
? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS
;
2093 g_autoptr(NBDExtentArray
) ea
= nbd_extent_array_new(nb_extents
);
2095 bitmap_to_extents(bitmap
, offset
, length
, ea
);
2097 return nbd_co_send_extents(client
, handle
, ea
, last
, context_id
, errp
);
2100 /* nbd_co_receive_request
2101 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2102 * connection right away, and any other negative value to report an error to
2103 * the client (although the caller may still need to disconnect after reporting
2106 static int nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
2109 NBDClient
*client
= req
->client
;
2112 g_assert(qemu_in_coroutine());
2113 assert(client
->recv_coroutine
== qemu_coroutine_self());
2114 if (nbd_receive_request(client
->ioc
, request
, errp
) < 0) {
2118 trace_nbd_co_receive_request_decode_type(request
->handle
, request
->type
,
2119 nbd_cmd_lookup(request
->type
));
2121 if (request
->type
!= NBD_CMD_WRITE
) {
2122 /* No payload, we are ready to read the next request. */
2123 req
->complete
= true;
2126 if (request
->type
== NBD_CMD_DISC
) {
2127 /* Special case: we're going to disconnect without a reply,
2128 * whether or not flags, from, or len are bogus */
2132 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
||
2133 request
->type
== NBD_CMD_CACHE
)
2135 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
2136 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
2137 request
->len
, NBD_MAX_BUFFER_SIZE
);
2141 if (request
->type
!= NBD_CMD_CACHE
) {
2142 req
->data
= blk_try_blockalign(client
->exp
->common
.blk
,
2144 if (req
->data
== NULL
) {
2145 error_setg(errp
, "No memory");
2151 if (request
->type
== NBD_CMD_WRITE
) {
2152 if (nbd_read(client
->ioc
, req
->data
, request
->len
, "CMD_WRITE data",
2157 req
->complete
= true;
2159 trace_nbd_co_receive_request_payload_received(request
->handle
,
2163 /* Sanity checks. */
2164 if (client
->exp
->nbdflags
& NBD_FLAG_READ_ONLY
&&
2165 (request
->type
== NBD_CMD_WRITE
||
2166 request
->type
== NBD_CMD_WRITE_ZEROES
||
2167 request
->type
== NBD_CMD_TRIM
)) {
2168 error_setg(errp
, "Export is read-only");
2171 if (request
->from
> client
->exp
->size
||
2172 request
->len
> client
->exp
->size
- request
->from
) {
2173 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
2174 ", Size: %" PRIu64
, request
->from
, request
->len
,
2176 return (request
->type
== NBD_CMD_WRITE
||
2177 request
->type
== NBD_CMD_WRITE_ZEROES
) ? -ENOSPC
: -EINVAL
;
2179 if (client
->check_align
&& !QEMU_IS_ALIGNED(request
->from
| request
->len
,
2180 client
->check_align
)) {
2182 * The block layer gracefully handles unaligned requests, but
2183 * it's still worth tracing client non-compliance
2185 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request
->type
),
2188 client
->check_align
);
2190 valid_flags
= NBD_CMD_FLAG_FUA
;
2191 if (request
->type
== NBD_CMD_READ
&& client
->structured_reply
) {
2192 valid_flags
|= NBD_CMD_FLAG_DF
;
2193 } else if (request
->type
== NBD_CMD_WRITE_ZEROES
) {
2194 valid_flags
|= NBD_CMD_FLAG_NO_HOLE
| NBD_CMD_FLAG_FAST_ZERO
;
2195 } else if (request
->type
== NBD_CMD_BLOCK_STATUS
) {
2196 valid_flags
|= NBD_CMD_FLAG_REQ_ONE
;
2198 if (request
->flags
& ~valid_flags
) {
2199 error_setg(errp
, "unsupported flags for command %s (got 0x%x)",
2200 nbd_cmd_lookup(request
->type
), request
->flags
);
2207 /* Send simple reply without a payload, or a structured error
2208 * @error_msg is ignored if @ret >= 0
2209 * Returns 0 if connection is still live, -errno on failure to talk to client
2211 static coroutine_fn
int nbd_send_generic_reply(NBDClient
*client
,
2214 const char *error_msg
,
2217 if (client
->structured_reply
&& ret
< 0) {
2218 return nbd_co_send_structured_error(client
, handle
, -ret
, error_msg
,
2221 return nbd_co_send_simple_reply(client
, handle
, ret
< 0 ? -ret
: 0,
2226 /* Handle NBD_CMD_READ request.
2227 * Return -errno if sending fails. Other errors are reported directly to the
2228 * client as an error reply. */
2229 static coroutine_fn
int nbd_do_cmd_read(NBDClient
*client
, NBDRequest
*request
,
2230 uint8_t *data
, Error
**errp
)
2233 NBDExport
*exp
= client
->exp
;
2235 assert(request
->type
== NBD_CMD_READ
);
2237 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2238 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2239 ret
= blk_co_flush(exp
->common
.blk
);
2241 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2242 "flush failed", errp
);
2246 if (client
->structured_reply
&& !(request
->flags
& NBD_CMD_FLAG_DF
) &&
2249 return nbd_co_send_sparse_read(client
, request
->handle
, request
->from
,
2250 data
, request
->len
, errp
);
2253 ret
= blk_pread(exp
->common
.blk
, request
->from
, data
, request
->len
);
2255 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2256 "reading from file failed", errp
);
2259 if (client
->structured_reply
) {
2261 return nbd_co_send_structured_read(client
, request
->handle
,
2262 request
->from
, data
,
2263 request
->len
, true, errp
);
2265 return nbd_co_send_structured_done(client
, request
->handle
, errp
);
2268 return nbd_co_send_simple_reply(client
, request
->handle
, 0,
2269 data
, request
->len
, errp
);
2276 * Handle NBD_CMD_CACHE request.
2277 * Return -errno if sending fails. Other errors are reported directly to the
2278 * client as an error reply.
2280 static coroutine_fn
int nbd_do_cmd_cache(NBDClient
*client
, NBDRequest
*request
,
2284 NBDExport
*exp
= client
->exp
;
2286 assert(request
->type
== NBD_CMD_CACHE
);
2288 ret
= blk_co_preadv(exp
->common
.blk
, request
->from
, request
->len
,
2289 NULL
, BDRV_REQ_COPY_ON_READ
| BDRV_REQ_PREFETCH
);
2291 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2292 "caching data failed", errp
);
2295 /* Handle NBD request.
2296 * Return -errno if sending fails. Other errors are reported directly to the
2297 * client as an error reply. */
2298 static coroutine_fn
int nbd_handle_request(NBDClient
*client
,
2299 NBDRequest
*request
,
2300 uint8_t *data
, Error
**errp
)
2304 NBDExport
*exp
= client
->exp
;
2307 switch (request
->type
) {
2309 return nbd_do_cmd_cache(client
, request
, errp
);
2312 return nbd_do_cmd_read(client
, request
, data
, errp
);
2316 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2317 flags
|= BDRV_REQ_FUA
;
2319 ret
= blk_pwrite(exp
->common
.blk
, request
->from
, data
, request
->len
,
2321 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2322 "writing to file failed", errp
);
2324 case NBD_CMD_WRITE_ZEROES
:
2326 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2327 flags
|= BDRV_REQ_FUA
;
2329 if (!(request
->flags
& NBD_CMD_FLAG_NO_HOLE
)) {
2330 flags
|= BDRV_REQ_MAY_UNMAP
;
2332 if (request
->flags
& NBD_CMD_FLAG_FAST_ZERO
) {
2333 flags
|= BDRV_REQ_NO_FALLBACK
;
2336 /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2337 while (ret
>= 0 && request
->len
) {
2338 int align
= client
->check_align
?: 1;
2339 int len
= MIN(request
->len
, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES
,
2341 ret
= blk_pwrite_zeroes(exp
->common
.blk
, request
->from
, len
, flags
);
2342 request
->len
-= len
;
2343 request
->from
+= len
;
2345 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2346 "writing to file failed", errp
);
2349 /* unreachable, thanks to special case in nbd_co_receive_request() */
2353 ret
= blk_co_flush(exp
->common
.blk
);
2354 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2355 "flush failed", errp
);
2359 /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2360 while (ret
>= 0 && request
->len
) {
2361 int align
= client
->check_align
?: 1;
2362 int len
= MIN(request
->len
, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES
,
2364 ret
= blk_co_pdiscard(exp
->common
.blk
, request
->from
, len
);
2365 request
->len
-= len
;
2366 request
->from
+= len
;
2368 if (ret
>= 0 && request
->flags
& NBD_CMD_FLAG_FUA
) {
2369 ret
= blk_co_flush(exp
->common
.blk
);
2371 return nbd_send_generic_reply(client
, request
->handle
, ret
,
2372 "discard failed", errp
);
2374 case NBD_CMD_BLOCK_STATUS
:
2375 if (!request
->len
) {
2376 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2377 "need non-zero length", errp
);
2379 if (client
->export_meta
.valid
&&
2380 (client
->export_meta
.base_allocation
||
2381 client
->export_meta
.bitmap
))
2383 bool dont_fragment
= request
->flags
& NBD_CMD_FLAG_REQ_ONE
;
2385 if (client
->export_meta
.base_allocation
) {
2386 ret
= nbd_co_send_block_status(client
, request
->handle
,
2387 blk_bs(exp
->common
.blk
),
2389 request
->len
, dont_fragment
,
2390 !client
->export_meta
.bitmap
,
2391 NBD_META_ID_BASE_ALLOCATION
,
2398 if (client
->export_meta
.bitmap
) {
2399 ret
= nbd_co_send_bitmap(client
, request
->handle
,
2400 client
->exp
->export_bitmap
,
2401 request
->from
, request
->len
,
2403 true, NBD_META_ID_DIRTY_BITMAP
, errp
);
2411 return nbd_send_generic_reply(client
, request
->handle
, -EINVAL
,
2412 "CMD_BLOCK_STATUS not negotiated",
2417 msg
= g_strdup_printf("invalid request type (%" PRIu32
") received",
2419 ret
= nbd_send_generic_reply(client
, request
->handle
, -EINVAL
, msg
,
2426 /* Owns a reference to the NBDClient passed as opaque. */
2427 static coroutine_fn
void nbd_trip(void *opaque
)
2429 NBDClient
*client
= opaque
;
2430 NBDRequestData
*req
;
2431 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
2433 Error
*local_err
= NULL
;
2436 if (client
->closing
) {
2437 nbd_client_put(client
);
2441 req
= nbd_request_get(client
);
2442 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
2443 client
->recv_coroutine
= NULL
;
2445 if (client
->closing
) {
2447 * The client may be closed when we are blocked in
2448 * nbd_co_receive_request()
2453 nbd_client_receive_next_request(client
);
2459 /* It wans't -EIO, so, according to nbd_co_receive_request()
2460 * semantics, we should return the error to the client. */
2461 Error
*export_err
= local_err
;
2464 ret
= nbd_send_generic_reply(client
, request
.handle
, -EINVAL
,
2465 error_get_pretty(export_err
), &local_err
);
2466 error_free(export_err
);
2468 ret
= nbd_handle_request(client
, &request
, req
->data
, &local_err
);
2471 error_prepend(&local_err
, "Failed to send reply: ");
2475 /* We must disconnect after NBD_CMD_WRITE if we did not
2478 if (!req
->complete
) {
2479 error_setg(&local_err
, "Request handling failed in intermediate state");
2484 nbd_request_put(req
);
2485 nbd_client_put(client
);
2490 error_reportf_err(local_err
, "Disconnect client, due to: ");
2492 nbd_request_put(req
);
2493 client_close(client
, true);
2494 nbd_client_put(client
);
2497 static void nbd_client_receive_next_request(NBDClient
*client
)
2499 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
) {
2500 nbd_client_get(client
);
2501 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
2502 aio_co_schedule(client
->exp
->common
.ctx
, client
->recv_coroutine
);
2506 static coroutine_fn
void nbd_co_client_start(void *opaque
)
2508 NBDClient
*client
= opaque
;
2509 Error
*local_err
= NULL
;
2511 qemu_co_mutex_init(&client
->send_lock
);
2513 if (nbd_negotiate(client
, &local_err
)) {
2515 error_report_err(local_err
);
2517 client_close(client
, false);
2521 nbd_client_receive_next_request(client
);
2525 * Create a new client listener using the given channel @sioc.
2526 * Begin servicing it in a coroutine. When the connection closes, call
2527 * @close_fn with an indication of whether the client completed negotiation.
2529 void nbd_client_new(QIOChannelSocket
*sioc
,
2530 QCryptoTLSCreds
*tlscreds
,
2531 const char *tlsauthz
,
2532 void (*close_fn
)(NBDClient
*, bool))
2537 client
= g_new0(NBDClient
, 1);
2538 client
->refcount
= 1;
2539 client
->tlscreds
= tlscreds
;
2541 object_ref(OBJECT(client
->tlscreds
));
2543 client
->tlsauthz
= g_strdup(tlsauthz
);
2544 client
->sioc
= sioc
;
2545 object_ref(OBJECT(client
->sioc
));
2546 client
->ioc
= QIO_CHANNEL(sioc
);
2547 object_ref(OBJECT(client
->ioc
));
2548 client
->close_fn
= close_fn
;
2550 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
2551 qemu_coroutine_enter(co
);