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/block_int.h"
23 #include "block/export.h"
24 #include "block/dirty-bitmap.h"
25 #include "qapi/error.h"
26 #include "qemu/queue.h"
28 #include "nbd-internal.h"
29 #include "qemu/units.h"
30 #include "qemu/memalign.h"
32 #define NBD_META_ID_BASE_ALLOCATION 0
33 #define NBD_META_ID_ALLOCATION_DEPTH 1
34 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
35 #define NBD_META_ID_DIRTY_BITMAP 2
38 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
39 * constant. If an increase is needed, note that the NBD protocol
40 * recommends no larger than 32 mb, so that the client won't consider
41 * the reply as a denial of service attack.
43 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
45 static int system_errno_to_nbd_errno(int err
)
66 #if ENOTSUP != EOPNOTSUPP
78 /* Definitions for opaque data types */
80 typedef struct NBDRequestData NBDRequestData
;
82 struct NBDRequestData
{
95 QTAILQ_HEAD(, NBDClient
) clients
;
96 QTAILQ_ENTRY(NBDExport
) next
;
98 BlockBackend
*eject_notifier_blk
;
99 Notifier eject_notifier
;
101 bool allocation_depth
;
102 BdrvDirtyBitmap
**export_bitmaps
;
103 size_t nr_export_bitmaps
;
106 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
108 /* NBDExportMetaContexts represents a list of contexts to be exported,
109 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
110 * NBD_OPT_LIST_META_CONTEXT. */
111 typedef struct NBDExportMetaContexts
{
113 size_t count
; /* number of negotiated contexts */
114 bool base_allocation
; /* export base:allocation context (block status) */
115 bool allocation_depth
; /* export qemu:allocation-depth */
117 * export qemu:dirty-bitmap:<export bitmap name>,
118 * sized by exp->nr_export_bitmaps
120 } NBDExportMetaContexts
;
124 void (*close_fn
)(NBDClient
*client
, bool negotiated
);
127 QCryptoTLSCreds
*tlscreds
;
129 QIOChannelSocket
*sioc
; /* The underlying data channel */
130 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
132 Coroutine
*recv_coroutine
;
135 Coroutine
*send_coroutine
;
140 QTAILQ_ENTRY(NBDClient
) next
;
144 uint32_t check_align
; /* If non-zero, check for aligned client requests */
146 bool structured_reply
;
147 NBDExportMetaContexts export_meta
;
149 uint32_t opt
; /* Current option being negotiated */
150 uint32_t optlen
; /* remaining length of data in ioc for the option being
154 static void nbd_client_receive_next_request(NBDClient
*client
);
156 /* Basic flow for negotiation
183 static inline void set_be_option_rep(NBDOptionReply
*rep
, uint32_t option
,
184 uint32_t type
, uint32_t length
)
186 stq_be_p(&rep
->magic
, NBD_REP_MAGIC
);
187 stl_be_p(&rep
->option
, option
);
188 stl_be_p(&rep
->type
, type
);
189 stl_be_p(&rep
->length
, length
);
192 /* Send a reply header, including length, but no payload.
193 * Return -errno on error, 0 on success. */
194 static int nbd_negotiate_send_rep_len(NBDClient
*client
, uint32_t type
,
195 uint32_t len
, Error
**errp
)
199 trace_nbd_negotiate_send_rep_len(client
->opt
, nbd_opt_lookup(client
->opt
),
200 type
, nbd_rep_lookup(type
), len
);
202 assert(len
< NBD_MAX_BUFFER_SIZE
);
204 set_be_option_rep(&rep
, client
->opt
, type
, len
);
205 return nbd_write(client
->ioc
, &rep
, sizeof(rep
), errp
);
208 /* Send a reply header with default 0 length.
209 * Return -errno on error, 0 on success. */
210 static int nbd_negotiate_send_rep(NBDClient
*client
, uint32_t type
,
213 return nbd_negotiate_send_rep_len(client
, type
, 0, errp
);
216 /* Send an error reply.
217 * Return -errno on error, 0 on success. */
218 static int G_GNUC_PRINTF(4, 0)
219 nbd_negotiate_send_rep_verr(NBDClient
*client
, uint32_t type
,
220 Error
**errp
, const char *fmt
, va_list va
)
223 g_autofree
char *msg
= NULL
;
227 msg
= g_strdup_vprintf(fmt
, va
);
229 assert(len
< NBD_MAX_STRING_SIZE
);
230 trace_nbd_negotiate_send_rep_err(msg
);
231 ret
= nbd_negotiate_send_rep_len(client
, type
, len
, errp
);
235 if (nbd_write(client
->ioc
, msg
, len
, errp
) < 0) {
236 error_prepend(errp
, "write failed (error message): ");
244 * Return a malloc'd copy of @name suitable for use in an error reply.
247 nbd_sanitize_name(const char *name
)
249 if (strnlen(name
, 80) < 80) {
250 return g_strdup(name
);
252 /* XXX Should we also try to sanitize any control characters? */
253 return g_strdup_printf("%.80s...", name
);
256 /* Send an error reply.
257 * Return -errno on error, 0 on success. */
258 static int G_GNUC_PRINTF(4, 5)
259 nbd_negotiate_send_rep_err(NBDClient
*client
, uint32_t type
,
260 Error
**errp
, const char *fmt
, ...)
266 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
271 /* Drop remainder of the current option, and send a reply with the
272 * given error type and message. Return -errno on read or write
273 * failure; or 0 if connection is still live. */
274 static int G_GNUC_PRINTF(4, 0)
275 nbd_opt_vdrop(NBDClient
*client
, uint32_t type
, Error
**errp
,
276 const char *fmt
, va_list va
)
278 int ret
= nbd_drop(client
->ioc
, client
->optlen
, errp
);
282 ret
= nbd_negotiate_send_rep_verr(client
, type
, errp
, fmt
, va
);
287 static int G_GNUC_PRINTF(4, 5)
288 nbd_opt_drop(NBDClient
*client
, uint32_t type
, Error
**errp
,
289 const char *fmt
, ...)
295 ret
= nbd_opt_vdrop(client
, type
, errp
, fmt
, va
);
301 static int G_GNUC_PRINTF(3, 4)
302 nbd_opt_invalid(NBDClient
*client
, Error
**errp
, const char *fmt
, ...)
308 ret
= nbd_opt_vdrop(client
, NBD_REP_ERR_INVALID
, errp
, fmt
, va
);
314 /* Read size bytes from the unparsed payload of the current option.
315 * If @check_nul, require that no NUL bytes appear in buffer.
316 * Return -errno on I/O error, 0 if option was completely handled by
317 * sending a reply about inconsistent lengths, or 1 on success. */
318 static int nbd_opt_read(NBDClient
*client
, void *buffer
, size_t size
,
319 bool check_nul
, Error
**errp
)
321 if (size
> client
->optlen
) {
322 return nbd_opt_invalid(client
, errp
,
323 "Inconsistent lengths in option %s",
324 nbd_opt_lookup(client
->opt
));
326 client
->optlen
-= size
;
327 if (qio_channel_read_all(client
->ioc
, buffer
, size
, errp
) < 0) {
331 if (check_nul
&& strnlen(buffer
, size
) != size
) {
332 return nbd_opt_invalid(client
, errp
,
333 "Unexpected embedded NUL in option %s",
334 nbd_opt_lookup(client
->opt
));
339 /* Drop size bytes from the unparsed payload of the current option.
340 * Return -errno on I/O error, 0 if option was completely handled by
341 * sending a reply about inconsistent lengths, or 1 on success. */
342 static int nbd_opt_skip(NBDClient
*client
, size_t size
, Error
**errp
)
344 if (size
> client
->optlen
) {
345 return nbd_opt_invalid(client
, errp
,
346 "Inconsistent lengths in option %s",
347 nbd_opt_lookup(client
->opt
));
349 client
->optlen
-= size
;
350 return nbd_drop(client
->ioc
, size
, errp
) < 0 ? -EIO
: 1;
355 * Read a string with the format:
356 * uint32_t len (<= NBD_MAX_STRING_SIZE)
357 * len bytes string (not 0-terminated)
359 * On success, @name will be allocated.
360 * If @length is non-null, it will be set to the actual string length.
362 * Return -errno on I/O error, 0 if option was completely handled by
363 * sending a reply about inconsistent lengths, or 1 on success.
365 static int nbd_opt_read_name(NBDClient
*client
, char **name
, uint32_t *length
,
370 g_autofree
char *local_name
= NULL
;
373 ret
= nbd_opt_read(client
, &len
, sizeof(len
), false, errp
);
377 len
= cpu_to_be32(len
);
379 if (len
> NBD_MAX_STRING_SIZE
) {
380 return nbd_opt_invalid(client
, errp
,
381 "Invalid name length: %" PRIu32
, len
);
384 local_name
= g_malloc(len
+ 1);
385 ret
= nbd_opt_read(client
, local_name
, len
, true, errp
);
389 local_name
[len
] = '\0';
394 *name
= g_steal_pointer(&local_name
);
399 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
400 * Return -errno on error, 0 on success. */
401 static int nbd_negotiate_send_rep_list(NBDClient
*client
, NBDExport
*exp
,
405 size_t name_len
, desc_len
;
407 const char *name
= exp
->name
? exp
->name
: "";
408 const char *desc
= exp
->description
? exp
->description
: "";
409 QIOChannel
*ioc
= client
->ioc
;
412 trace_nbd_negotiate_send_rep_list(name
, desc
);
413 name_len
= strlen(name
);
414 desc_len
= strlen(desc
);
415 assert(name_len
<= NBD_MAX_STRING_SIZE
&& desc_len
<= NBD_MAX_STRING_SIZE
);
416 len
= name_len
+ desc_len
+ sizeof(len
);
417 ret
= nbd_negotiate_send_rep_len(client
, NBD_REP_SERVER
, len
, errp
);
422 len
= cpu_to_be32(name_len
);
423 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
424 error_prepend(errp
, "write failed (name length): ");
428 if (nbd_write(ioc
, name
, name_len
, errp
) < 0) {
429 error_prepend(errp
, "write failed (name buffer): ");
433 if (nbd_write(ioc
, desc
, desc_len
, errp
) < 0) {
434 error_prepend(errp
, "write failed (description buffer): ");
441 /* Process the NBD_OPT_LIST command, with a potential series of replies.
442 * Return -errno on error, 0 on success. */
443 static int nbd_negotiate_handle_list(NBDClient
*client
, Error
**errp
)
446 assert(client
->opt
== NBD_OPT_LIST
);
448 /* For each export, send a NBD_REP_SERVER reply. */
449 QTAILQ_FOREACH(exp
, &exports
, next
) {
450 if (nbd_negotiate_send_rep_list(client
, exp
, errp
)) {
454 /* Finish with a NBD_REP_ACK. */
455 return nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
458 static void nbd_check_meta_export(NBDClient
*client
)
460 if (client
->exp
!= client
->export_meta
.exp
) {
461 client
->export_meta
.count
= 0;
465 /* Send a reply to NBD_OPT_EXPORT_NAME.
466 * Return -errno on error, 0 on success. */
467 static int nbd_negotiate_handle_export_name(NBDClient
*client
, bool no_zeroes
,
471 g_autofree
char *name
= NULL
;
472 char buf
[NBD_REPLY_EXPORT_NAME_SIZE
] = "";
478 [20 .. xx] export name (length bytes)
481 [ 8 .. 9] export flags
482 [10 .. 133] reserved (0) [unless no_zeroes]
484 trace_nbd_negotiate_handle_export_name();
485 if (client
->optlen
> NBD_MAX_STRING_SIZE
) {
486 error_setg(errp
, "Bad length received");
489 name
= g_malloc(client
->optlen
+ 1);
490 if (nbd_read(client
->ioc
, name
, client
->optlen
, "export name", errp
) < 0) {
493 name
[client
->optlen
] = '\0';
496 trace_nbd_negotiate_handle_export_name_request(name
);
498 client
->exp
= nbd_export_find(name
);
500 error_setg(errp
, "export not found");
504 myflags
= client
->exp
->nbdflags
;
505 if (client
->structured_reply
) {
506 myflags
|= NBD_FLAG_SEND_DF
;
508 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
, myflags
);
509 stq_be_p(buf
, client
->exp
->size
);
510 stw_be_p(buf
+ 8, myflags
);
511 len
= no_zeroes
? 10 : sizeof(buf
);
512 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
514 error_prepend(errp
, "write failed: ");
518 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
519 blk_exp_ref(&client
->exp
->common
);
520 nbd_check_meta_export(client
);
525 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
526 * The buffer does NOT include the info type prefix.
527 * Return -errno on error, 0 if ready to send more. */
528 static int nbd_negotiate_send_info(NBDClient
*client
,
529 uint16_t info
, uint32_t length
, void *buf
,
534 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
535 rc
= nbd_negotiate_send_rep_len(client
, NBD_REP_INFO
,
536 sizeof(info
) + length
, errp
);
540 info
= cpu_to_be16(info
);
541 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
544 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
550 /* nbd_reject_length: Handle any unexpected payload.
551 * @fatal requests that we quit talking to the client, even if we are able
552 * to successfully send an error reply.
554 * -errno transmission error occurred or @fatal was requested, errp is set
555 * 0 error message successfully sent to client, errp is not set
557 static int nbd_reject_length(NBDClient
*client
, bool fatal
, Error
**errp
)
561 assert(client
->optlen
);
562 ret
= nbd_opt_invalid(client
, errp
, "option '%s' has unexpected length",
563 nbd_opt_lookup(client
->opt
));
565 error_setg(errp
, "option '%s' has unexpected length",
566 nbd_opt_lookup(client
->opt
));
572 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
573 * Return -errno on error, 0 if ready for next option, and 1 to move
574 * into transmission phase. */
575 static int nbd_negotiate_handle_info(NBDClient
*client
, Error
**errp
)
578 g_autofree
char *name
= NULL
;
582 uint32_t namelen
= 0;
583 bool sendname
= false;
584 bool blocksize
= false;
586 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
587 uint32_t check_align
= 0;
591 4 bytes: L, name length (can be 0)
593 2 bytes: N, number of requests (can be 0)
594 N * 2 bytes: N requests
596 rc
= nbd_opt_read_name(client
, &name
, &namelen
, errp
);
600 trace_nbd_negotiate_handle_export_name_request(name
);
602 rc
= nbd_opt_read(client
, &requests
, sizeof(requests
), false, errp
);
606 requests
= be16_to_cpu(requests
);
607 trace_nbd_negotiate_handle_info_requests(requests
);
609 rc
= nbd_opt_read(client
, &request
, sizeof(request
), false, errp
);
613 request
= be16_to_cpu(request
);
614 trace_nbd_negotiate_handle_info_request(request
,
615 nbd_info_lookup(request
));
616 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
617 * everything else is either a request we don't know or
618 * something we send regardless of request */
623 case NBD_INFO_BLOCK_SIZE
:
628 if (client
->optlen
) {
629 return nbd_reject_length(client
, false, errp
);
632 exp
= nbd_export_find(name
);
634 g_autofree
char *sane_name
= nbd_sanitize_name(name
);
636 return nbd_negotiate_send_rep_err(client
, NBD_REP_ERR_UNKNOWN
,
637 errp
, "export '%s' not present",
641 /* Don't bother sending NBD_INFO_NAME unless client requested it */
643 rc
= nbd_negotiate_send_info(client
, NBD_INFO_NAME
, namelen
, name
,
650 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
652 if (exp
->description
) {
653 size_t len
= strlen(exp
->description
);
655 assert(len
<= NBD_MAX_STRING_SIZE
);
656 rc
= nbd_negotiate_send_info(client
, NBD_INFO_DESCRIPTION
,
657 len
, exp
->description
, errp
);
663 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
664 * according to whether the client requested it, and according to
665 * whether this is OPT_INFO or OPT_GO. */
666 /* minimum - 1 for back-compat, or actual if client will obey it. */
667 if (client
->opt
== NBD_OPT_INFO
|| blocksize
) {
668 check_align
= sizes
[0] = blk_get_request_alignment(exp
->common
.blk
);
672 assert(sizes
[0] <= NBD_MAX_BUFFER_SIZE
);
673 /* preferred - Hard-code to 4096 for now.
674 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
675 sizes
[1] = MAX(4096, sizes
[0]);
676 /* maximum - At most 32M, but smaller as appropriate. */
677 sizes
[2] = MIN(blk_get_max_transfer(exp
->common
.blk
), NBD_MAX_BUFFER_SIZE
);
678 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
679 sizes
[0] = cpu_to_be32(sizes
[0]);
680 sizes
[1] = cpu_to_be32(sizes
[1]);
681 sizes
[2] = cpu_to_be32(sizes
[2]);
682 rc
= nbd_negotiate_send_info(client
, NBD_INFO_BLOCK_SIZE
,
683 sizeof(sizes
), sizes
, errp
);
688 /* Send NBD_INFO_EXPORT always */
689 myflags
= exp
->nbdflags
;
690 if (client
->structured_reply
) {
691 myflags
|= NBD_FLAG_SEND_DF
;
693 trace_nbd_negotiate_new_style_size_flags(exp
->size
, myflags
);
694 stq_be_p(buf
, exp
->size
);
695 stw_be_p(buf
+ 8, myflags
);
696 rc
= nbd_negotiate_send_info(client
, NBD_INFO_EXPORT
,
697 sizeof(buf
), buf
, errp
);
703 * If the client is just asking for NBD_OPT_INFO, but forgot to
704 * request block sizes in a situation that would impact
705 * performance, then return an error. But for NBD_OPT_GO, we
706 * tolerate all clients, regardless of alignments.
708 if (client
->opt
== NBD_OPT_INFO
&& !blocksize
&&
709 blk_get_request_alignment(exp
->common
.blk
) > 1) {
710 return nbd_negotiate_send_rep_err(client
,
711 NBD_REP_ERR_BLOCK_SIZE_REQD
,
713 "request NBD_INFO_BLOCK_SIZE to "
718 rc
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
723 if (client
->opt
== NBD_OPT_GO
) {
725 client
->check_align
= check_align
;
726 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
727 blk_exp_ref(&client
->exp
->common
);
728 nbd_check_meta_export(client
);
735 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
736 * new channel for all further (now-encrypted) communication. */
737 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
742 struct NBDTLSHandshakeData data
= { 0 };
744 assert(client
->opt
== NBD_OPT_STARTTLS
);
746 trace_nbd_negotiate_handle_starttls();
749 if (nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
) < 0) {
753 tioc
= qio_channel_tls_new_server(ioc
,
761 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
762 trace_nbd_negotiate_handle_starttls_handshake();
763 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
764 qio_channel_tls_handshake(tioc
,
770 if (!data
.complete
) {
771 g_main_loop_run(data
.loop
);
773 g_main_loop_unref(data
.loop
);
775 object_unref(OBJECT(tioc
));
776 error_propagate(errp
, data
.error
);
780 return QIO_CHANNEL(tioc
);
783 /* nbd_negotiate_send_meta_context
785 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
787 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
789 static int nbd_negotiate_send_meta_context(NBDClient
*client
,
794 NBDOptionReplyMetaContext opt
;
795 struct iovec iov
[] = {
796 {.iov_base
= &opt
, .iov_len
= sizeof(opt
)},
797 {.iov_base
= (void *)context
, .iov_len
= strlen(context
)}
800 assert(iov
[1].iov_len
<= NBD_MAX_STRING_SIZE
);
801 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
805 trace_nbd_negotiate_meta_query_reply(context
, context_id
);
806 set_be_option_rep(&opt
.h
, client
->opt
, NBD_REP_META_CONTEXT
,
807 sizeof(opt
) - sizeof(opt
.h
) + iov
[1].iov_len
);
808 stl_be_p(&opt
.context_id
, context_id
);
810 return qio_channel_writev_all(client
->ioc
, iov
, 2, errp
) < 0 ? -EIO
: 0;
814 * Return true if @query matches @pattern, or if @query is empty when
815 * the @client is performing _LIST_.
817 static bool nbd_meta_empty_or_pattern(NBDClient
*client
, const char *pattern
,
821 trace_nbd_negotiate_meta_query_parse("empty");
822 return client
->opt
== NBD_OPT_LIST_META_CONTEXT
;
824 if (strcmp(query
, pattern
) == 0) {
825 trace_nbd_negotiate_meta_query_parse(pattern
);
828 trace_nbd_negotiate_meta_query_skip("pattern not matched");
833 * Return true and adjust @str in place if it begins with @prefix.
835 static bool nbd_strshift(const char **str
, const char *prefix
)
837 size_t len
= strlen(prefix
);
839 if (strncmp(*str
, prefix
, len
) == 0) {
846 /* nbd_meta_base_query
848 * Handle queries to 'base' namespace. For now, only the base:allocation
849 * context is available. Return true if @query has been handled.
851 static bool nbd_meta_base_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
854 if (!nbd_strshift(&query
, "base:")) {
857 trace_nbd_negotiate_meta_query_parse("base:");
859 if (nbd_meta_empty_or_pattern(client
, "allocation", query
)) {
860 meta
->base_allocation
= true;
865 /* nbd_meta_qemu_query
867 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
868 * and qemu:allocation-depth contexts are available. Return true if @query
871 static bool nbd_meta_qemu_query(NBDClient
*client
, NBDExportMetaContexts
*meta
,
876 if (!nbd_strshift(&query
, "qemu:")) {
879 trace_nbd_negotiate_meta_query_parse("qemu:");
882 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
883 meta
->allocation_depth
= meta
->exp
->allocation_depth
;
884 if (meta
->exp
->nr_export_bitmaps
) {
885 memset(meta
->bitmaps
, 1, meta
->exp
->nr_export_bitmaps
);
888 trace_nbd_negotiate_meta_query_parse("empty");
892 if (strcmp(query
, "allocation-depth") == 0) {
893 trace_nbd_negotiate_meta_query_parse("allocation-depth");
894 meta
->allocation_depth
= meta
->exp
->allocation_depth
;
898 if (nbd_strshift(&query
, "dirty-bitmap:")) {
899 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
901 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
&&
902 meta
->exp
->nr_export_bitmaps
) {
903 memset(meta
->bitmaps
, 1, meta
->exp
->nr_export_bitmaps
);
905 trace_nbd_negotiate_meta_query_parse("empty");
909 for (i
= 0; i
< meta
->exp
->nr_export_bitmaps
; i
++) {
912 bm_name
= bdrv_dirty_bitmap_name(meta
->exp
->export_bitmaps
[i
]);
913 if (strcmp(bm_name
, query
) == 0) {
914 meta
->bitmaps
[i
] = true;
915 trace_nbd_negotiate_meta_query_parse(query
);
919 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
923 trace_nbd_negotiate_meta_query_skip("unknown qemu context");
927 /* nbd_negotiate_meta_query
929 * Parse namespace name and call corresponding function to parse body of the
932 * The only supported namespaces are 'base' and 'qemu'.
934 * Return -errno on I/O error, 0 if option was completely handled by
935 * sending a reply about inconsistent lengths, or 1 on success. */
936 static int nbd_negotiate_meta_query(NBDClient
*client
,
937 NBDExportMetaContexts
*meta
, Error
**errp
)
940 g_autofree
char *query
= NULL
;
943 ret
= nbd_opt_read(client
, &len
, sizeof(len
), false, errp
);
947 len
= cpu_to_be32(len
);
949 if (len
> NBD_MAX_STRING_SIZE
) {
950 trace_nbd_negotiate_meta_query_skip("length too long");
951 return nbd_opt_skip(client
, len
, errp
);
954 query
= g_malloc(len
+ 1);
955 ret
= nbd_opt_read(client
, query
, len
, true, errp
);
961 if (nbd_meta_base_query(client
, meta
, query
)) {
964 if (nbd_meta_qemu_query(client
, meta
, query
)) {
968 trace_nbd_negotiate_meta_query_skip("unknown namespace");
972 /* nbd_negotiate_meta_queries
973 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
975 * Return -errno on I/O error, or 0 if option was completely handled. */
976 static int nbd_negotiate_meta_queries(NBDClient
*client
,
977 NBDExportMetaContexts
*meta
, Error
**errp
)
980 g_autofree
char *export_name
= NULL
;
981 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */
982 g_autofree G_GNUC_UNUSED
bool *bitmaps
= NULL
;
983 NBDExportMetaContexts local_meta
= {0};
988 if (client
->opt
== NBD_OPT_SET_META_CONTEXT
&& !client
->structured_reply
) {
989 return nbd_opt_invalid(client
, errp
,
990 "request option '%s' when structured reply "
992 nbd_opt_lookup(client
->opt
));
995 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
996 /* Only change the caller's meta on SET. */
1000 g_free(meta
->bitmaps
);
1001 memset(meta
, 0, sizeof(*meta
));
1003 ret
= nbd_opt_read_name(client
, &export_name
, NULL
, errp
);
1008 meta
->exp
= nbd_export_find(export_name
);
1009 if (meta
->exp
== NULL
) {
1010 g_autofree
char *sane_name
= nbd_sanitize_name(export_name
);
1012 return nbd_opt_drop(client
, NBD_REP_ERR_UNKNOWN
, errp
,
1013 "export '%s' not present", sane_name
);
1015 meta
->bitmaps
= g_new0(bool, meta
->exp
->nr_export_bitmaps
);
1016 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
) {
1017 bitmaps
= meta
->bitmaps
;
1020 ret
= nbd_opt_read(client
, &nb_queries
, sizeof(nb_queries
), false, errp
);
1024 nb_queries
= cpu_to_be32(nb_queries
);
1025 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client
->opt
),
1026 export_name
, nb_queries
);
1028 if (client
->opt
== NBD_OPT_LIST_META_CONTEXT
&& !nb_queries
) {
1029 /* enable all known contexts */
1030 meta
->base_allocation
= true;
1031 meta
->allocation_depth
= meta
->exp
->allocation_depth
;
1032 if (meta
->exp
->nr_export_bitmaps
) {
1033 memset(meta
->bitmaps
, 1, meta
->exp
->nr_export_bitmaps
);
1036 for (i
= 0; i
< nb_queries
; ++i
) {
1037 ret
= nbd_negotiate_meta_query(client
, meta
, errp
);
1044 if (meta
->base_allocation
) {
1045 ret
= nbd_negotiate_send_meta_context(client
, "base:allocation",
1046 NBD_META_ID_BASE_ALLOCATION
,
1054 if (meta
->allocation_depth
) {
1055 ret
= nbd_negotiate_send_meta_context(client
, "qemu:allocation-depth",
1056 NBD_META_ID_ALLOCATION_DEPTH
,
1064 for (i
= 0; i
< meta
->exp
->nr_export_bitmaps
; i
++) {
1065 const char *bm_name
;
1066 g_autofree
char *context
= NULL
;
1068 if (!meta
->bitmaps
[i
]) {
1072 bm_name
= bdrv_dirty_bitmap_name(meta
->exp
->export_bitmaps
[i
]);
1073 context
= g_strdup_printf("qemu:dirty-bitmap:%s", bm_name
);
1075 ret
= nbd_negotiate_send_meta_context(client
, context
,
1076 NBD_META_ID_DIRTY_BITMAP
+ i
,
1084 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1086 meta
->count
= count
;
1092 /* nbd_negotiate_options
1093 * Process all NBD_OPT_* client option commands, during fixed newstyle
1096 * -errno on error, errp is set
1097 * 0 on successful negotiation, errp is not set
1098 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1101 static int nbd_negotiate_options(NBDClient
*client
, Error
**errp
)
1104 bool fixedNewstyle
= false;
1105 bool no_zeroes
= false;
1108 [ 0 .. 3] client flags
1110 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1111 [ 0 .. 7] NBD_OPTS_MAGIC
1112 [ 8 .. 11] NBD option
1113 [12 .. 15] Data length
1116 [ 0 .. 7] NBD_OPTS_MAGIC
1117 [ 8 .. 11] Second NBD option
1118 [12 .. 15] Data length
1122 if (nbd_read32(client
->ioc
, &flags
, "flags", errp
) < 0) {
1125 trace_nbd_negotiate_options_flags(flags
);
1126 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
1127 fixedNewstyle
= true;
1128 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
1130 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
1132 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
1135 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
1141 uint32_t option
, length
;
1144 if (nbd_read64(client
->ioc
, &magic
, "opts magic", errp
) < 0) {
1147 trace_nbd_negotiate_options_check_magic(magic
);
1148 if (magic
!= NBD_OPTS_MAGIC
) {
1149 error_setg(errp
, "Bad magic received");
1153 if (nbd_read32(client
->ioc
, &option
, "option", errp
) < 0) {
1156 client
->opt
= option
;
1158 if (nbd_read32(client
->ioc
, &length
, "option length", errp
) < 0) {
1161 assert(!client
->optlen
);
1162 client
->optlen
= length
;
1164 if (length
> NBD_MAX_BUFFER_SIZE
) {
1165 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1166 length
, NBD_MAX_BUFFER_SIZE
);
1170 trace_nbd_negotiate_options_check_option(option
,
1171 nbd_opt_lookup(option
));
1172 if (client
->tlscreds
&&
1173 client
->ioc
== (QIOChannel
*)client
->sioc
) {
1175 if (!fixedNewstyle
) {
1176 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
1180 case NBD_OPT_STARTTLS
:
1182 /* Unconditionally drop the connection if the client
1183 * can't start a TLS negotiation correctly */
1184 return nbd_reject_length(client
, true, errp
);
1186 tioc
= nbd_negotiate_handle_starttls(client
, errp
);
1191 object_unref(OBJECT(client
->ioc
));
1195 case NBD_OPT_EXPORT_NAME
:
1196 /* No way to return an error to client, so drop connection */
1197 error_setg(errp
, "Option 0x%x not permitted before TLS",
1202 /* Let the client keep trying, unless they asked to
1203 * quit. Always try to give an error back to the
1204 * client; but when replying to OPT_ABORT, be aware
1205 * that the client may hang up before receiving the
1206 * error, in which case we are fine ignoring the
1207 * resulting EPIPE. */
1208 ret
= nbd_opt_drop(client
, NBD_REP_ERR_TLS_REQD
,
1209 option
== NBD_OPT_ABORT
? NULL
: errp
,
1211 " not permitted before TLS", option
);
1212 if (option
== NBD_OPT_ABORT
) {
1217 } else if (fixedNewstyle
) {
1221 ret
= nbd_reject_length(client
, false, errp
);
1223 ret
= nbd_negotiate_handle_list(client
, errp
);
1228 /* NBD spec says we must try to reply before
1229 * disconnecting, but that we must also tolerate
1230 * guests that don't wait for our reply. */
1231 nbd_negotiate_send_rep(client
, NBD_REP_ACK
, NULL
);
1234 case NBD_OPT_EXPORT_NAME
:
1235 return nbd_negotiate_handle_export_name(client
, no_zeroes
,
1240 ret
= nbd_negotiate_handle_info(client
, errp
);
1242 assert(option
== NBD_OPT_GO
);
1247 case NBD_OPT_STARTTLS
:
1249 ret
= nbd_reject_length(client
, false, errp
);
1250 } else if (client
->tlscreds
) {
1251 ret
= nbd_negotiate_send_rep_err(client
,
1252 NBD_REP_ERR_INVALID
, errp
,
1253 "TLS already enabled");
1255 ret
= nbd_negotiate_send_rep_err(client
,
1256 NBD_REP_ERR_POLICY
, errp
,
1257 "TLS not configured");
1261 case NBD_OPT_STRUCTURED_REPLY
:
1263 ret
= nbd_reject_length(client
, false, errp
);
1264 } else if (client
->structured_reply
) {
1265 ret
= nbd_negotiate_send_rep_err(
1266 client
, NBD_REP_ERR_INVALID
, errp
,
1267 "structured reply already negotiated");
1269 ret
= nbd_negotiate_send_rep(client
, NBD_REP_ACK
, errp
);
1270 client
->structured_reply
= true;
1274 case NBD_OPT_LIST_META_CONTEXT
:
1275 case NBD_OPT_SET_META_CONTEXT
:
1276 ret
= nbd_negotiate_meta_queries(client
, &client
->export_meta
,
1281 ret
= nbd_opt_drop(client
, NBD_REP_ERR_UNSUP
, errp
,
1282 "Unsupported option %" PRIu32
" (%s)",
1283 option
, nbd_opt_lookup(option
));
1288 * If broken new-style we should drop the connection
1289 * for anything except NBD_OPT_EXPORT_NAME
1292 case NBD_OPT_EXPORT_NAME
:
1293 return nbd_negotiate_handle_export_name(client
, no_zeroes
,
1297 error_setg(errp
, "Unsupported option %" PRIu32
" (%s)",
1298 option
, nbd_opt_lookup(option
));
1310 * -errno on error, errp is set
1311 * 0 on successful negotiation, errp is not set
1312 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1315 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
1318 char buf
[NBD_OLDSTYLE_NEGOTIATE_SIZE
] = "";
1321 /* Old style negotiation header, no room for options
1322 [ 0 .. 7] passwd ("NBDMAGIC")
1323 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
1325 [24 .. 27] export flags (zero-extended)
1326 [28 .. 151] reserved (0)
1328 New style negotiation header, client can send options
1329 [ 0 .. 7] passwd ("NBDMAGIC")
1330 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
1331 [16 .. 17] server flags (0)
1332 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1335 qio_channel_set_blocking(client
->ioc
, false, NULL
);
1336 qio_channel_set_follow_coroutine_ctx(client
->ioc
, true);
1338 trace_nbd_negotiate_begin();
1339 memcpy(buf
, "NBDMAGIC", 8);
1341 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
1342 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
1344 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
1345 error_prepend(errp
, "write failed: ");
1348 ret
= nbd_negotiate_options(client
, errp
);
1351 error_prepend(errp
, "option negotiation failed: ");
1356 assert(!client
->optlen
);
1357 trace_nbd_negotiate_success();
1363 * Tries to read @size bytes from @ioc. This is a local implementation of
1364 * qio_channel_readv_all_eof. We have it here because we need it to be
1365 * interruptible and to know when the coroutine is yielding.
1366 * Returns 1 on success
1367 * 0 on eof, when no data was read (errp is not set)
1368 * negative errno on failure (errp is set)
1370 static inline int coroutine_fn
1371 nbd_read_eof(NBDClient
*client
, void *buffer
, size_t size
, Error
**errp
)
1373 bool partial
= false;
1377 struct iovec iov
= { .iov_base
= buffer
, .iov_len
= size
};
1380 len
= qio_channel_readv(client
->ioc
, &iov
, 1, errp
);
1381 if (len
== QIO_CHANNEL_ERR_BLOCK
) {
1382 client
->read_yielding
= true;
1383 qio_channel_yield(client
->ioc
, G_IO_IN
);
1384 client
->read_yielding
= false;
1385 if (client
->quiescing
) {
1389 } else if (len
< 0) {
1391 } else if (len
== 0) {
1394 "Unexpected end-of-file before all bytes were read");
1403 buffer
= (uint8_t *) buffer
+ len
;
1408 static int coroutine_fn
nbd_receive_request(NBDClient
*client
, NBDRequest
*request
,
1411 uint8_t buf
[NBD_REQUEST_SIZE
];
1415 ret
= nbd_read_eof(client
, buf
, sizeof(buf
), errp
);
1424 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
1425 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
1426 [ 6 .. 7] type (NBD_CMD_READ, ...)
1432 magic
= ldl_be_p(buf
);
1433 request
->flags
= lduw_be_p(buf
+ 4);
1434 request
->type
= lduw_be_p(buf
+ 6);
1435 request
->cookie
= ldq_be_p(buf
+ 8);
1436 request
->from
= ldq_be_p(buf
+ 16);
1437 request
->len
= ldl_be_p(buf
+ 24);
1439 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
1440 request
->from
, request
->len
);
1442 if (magic
!= NBD_REQUEST_MAGIC
) {
1443 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
1449 #define MAX_NBD_REQUESTS 16
1451 void nbd_client_get(NBDClient
*client
)
1456 void nbd_client_put(NBDClient
*client
)
1458 if (--client
->refcount
== 0) {
1459 /* The last reference should be dropped by client->close,
1460 * which is called by client_close.
1462 assert(client
->closing
);
1464 object_unref(OBJECT(client
->sioc
));
1465 object_unref(OBJECT(client
->ioc
));
1466 if (client
->tlscreds
) {
1467 object_unref(OBJECT(client
->tlscreds
));
1469 g_free(client
->tlsauthz
);
1471 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
1472 blk_exp_unref(&client
->exp
->common
);
1474 g_free(client
->export_meta
.bitmaps
);
1479 static void client_close(NBDClient
*client
, bool negotiated
)
1481 if (client
->closing
) {
1485 client
->closing
= true;
1487 /* Force requests to finish. They will drop their own references,
1488 * then we'll close the socket and free the NBDClient.
1490 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
1493 /* Also tell the client, so that they release their reference. */
1494 if (client
->close_fn
) {
1495 client
->close_fn(client
, negotiated
);
1499 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
1501 NBDRequestData
*req
;
1503 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
1504 client
->nb_requests
++;
1506 req
= g_new0(NBDRequestData
, 1);
1507 nbd_client_get(client
);
1508 req
->client
= client
;
1512 static void nbd_request_put(NBDRequestData
*req
)
1514 NBDClient
*client
= req
->client
;
1517 qemu_vfree(req
->data
);
1521 client
->nb_requests
--;
1523 if (client
->quiescing
&& client
->nb_requests
== 0) {
1527 nbd_client_receive_next_request(client
);
1529 nbd_client_put(client
);
1532 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1534 NBDExport
*exp
= opaque
;
1537 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1539 exp
->common
.ctx
= ctx
;
1541 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1542 assert(client
->nb_requests
== 0);
1543 assert(client
->recv_coroutine
== NULL
);
1544 assert(client
->send_coroutine
== NULL
);
1548 static void blk_aio_detach(void *opaque
)
1550 NBDExport
*exp
= opaque
;
1552 trace_nbd_blk_aio_detach(exp
->name
, exp
->common
.ctx
);
1554 exp
->common
.ctx
= NULL
;
1557 static void nbd_drained_begin(void *opaque
)
1559 NBDExport
*exp
= opaque
;
1562 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1563 client
->quiescing
= true;
1567 static void nbd_drained_end(void *opaque
)
1569 NBDExport
*exp
= opaque
;
1572 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1573 client
->quiescing
= false;
1574 nbd_client_receive_next_request(client
);
1578 static bool nbd_drained_poll(void *opaque
)
1580 NBDExport
*exp
= opaque
;
1583 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1584 if (client
->nb_requests
!= 0) {
1586 * If there's a coroutine waiting for a request on nbd_read_eof()
1587 * enter it here so we don't depend on the client to wake it up.
1589 if (client
->recv_coroutine
!= NULL
&& client
->read_yielding
) {
1590 qio_channel_wake_read(client
->ioc
);
1600 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1602 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1604 blk_exp_request_shutdown(&exp
->common
);
1607 void nbd_export_set_on_eject_blk(BlockExport
*exp
, BlockBackend
*blk
)
1609 NBDExport
*nbd_exp
= container_of(exp
, NBDExport
, common
);
1610 assert(exp
->drv
== &blk_exp_nbd
);
1611 assert(nbd_exp
->eject_notifier_blk
== NULL
);
1614 nbd_exp
->eject_notifier_blk
= blk
;
1615 nbd_exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1616 blk_add_remove_bs_notifier(blk
, &nbd_exp
->eject_notifier
);
1619 static const BlockDevOps nbd_block_ops
= {
1620 .drained_begin
= nbd_drained_begin
,
1621 .drained_end
= nbd_drained_end
,
1622 .drained_poll
= nbd_drained_poll
,
1625 static int nbd_export_create(BlockExport
*blk_exp
, BlockExportOptions
*exp_args
,
1628 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1629 BlockExportOptionsNbd
*arg
= &exp_args
->u
.nbd
;
1630 const char *name
= arg
->name
?: exp_args
->node_name
;
1631 BlockBackend
*blk
= blk_exp
->blk
;
1633 uint64_t perm
, shared_perm
;
1634 bool readonly
= !exp_args
->writable
;
1635 BlockDirtyBitmapOrStrList
*bitmaps
;
1639 assert(exp_args
->type
== BLOCK_EXPORT_TYPE_NBD
);
1641 if (!nbd_server_is_running()) {
1642 error_setg(errp
, "NBD server not running");
1646 if (strlen(name
) > NBD_MAX_STRING_SIZE
) {
1647 error_setg(errp
, "export name '%s' too long", name
);
1651 if (arg
->description
&& strlen(arg
->description
) > NBD_MAX_STRING_SIZE
) {
1652 error_setg(errp
, "description '%s' too long", arg
->description
);
1656 if (nbd_export_find(name
)) {
1657 error_setg(errp
, "NBD server already has export named '%s'", name
);
1661 size
= blk_getlength(blk
);
1663 error_setg_errno(errp
, -size
,
1664 "Failed to determine the NBD export's length");
1668 /* Don't allow resize while the NBD server is running, otherwise we don't
1669 * care what happens with the node. */
1670 blk_get_perm(blk
, &perm
, &shared_perm
);
1671 ret
= blk_set_perm(blk
, perm
, shared_perm
& ~BLK_PERM_RESIZE
, errp
);
1676 QTAILQ_INIT(&exp
->clients
);
1677 exp
->name
= g_strdup(name
);
1678 exp
->description
= g_strdup(arg
->description
);
1679 exp
->nbdflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_FLUSH
|
1680 NBD_FLAG_SEND_FUA
| NBD_FLAG_SEND_CACHE
);
1682 if (nbd_server_max_connections() != 1) {
1683 exp
->nbdflags
|= NBD_FLAG_CAN_MULTI_CONN
;
1686 exp
->nbdflags
|= NBD_FLAG_READ_ONLY
;
1688 exp
->nbdflags
|= (NBD_FLAG_SEND_TRIM
| NBD_FLAG_SEND_WRITE_ZEROES
|
1689 NBD_FLAG_SEND_FAST_ZERO
);
1691 exp
->size
= QEMU_ALIGN_DOWN(size
, BDRV_SECTOR_SIZE
);
1693 for (bitmaps
= arg
->bitmaps
; bitmaps
; bitmaps
= bitmaps
->next
) {
1694 exp
->nr_export_bitmaps
++;
1696 exp
->export_bitmaps
= g_new0(BdrvDirtyBitmap
*, exp
->nr_export_bitmaps
);
1697 for (i
= 0, bitmaps
= arg
->bitmaps
; bitmaps
;
1698 i
++, bitmaps
= bitmaps
->next
)
1701 BlockDriverState
*bs
= blk_bs(blk
);
1702 BdrvDirtyBitmap
*bm
= NULL
;
1704 switch (bitmaps
->value
->type
) {
1706 bitmap
= bitmaps
->value
->u
.local
;
1708 bm
= bdrv_find_dirty_bitmap(bs
, bitmap
);
1713 bs
= bdrv_filter_or_cow_bs(bs
);
1718 error_setg(errp
, "Bitmap '%s' is not found",
1719 bitmaps
->value
->u
.local
);
1723 if (readonly
&& bdrv_is_writable(bs
) &&
1724 bdrv_dirty_bitmap_enabled(bm
)) {
1726 error_setg(errp
, "Enabled bitmap '%s' incompatible with "
1727 "readonly export", bitmap
);
1732 bitmap
= bitmaps
->value
->u
.external
.name
;
1733 bm
= block_dirty_bitmap_lookup(bitmaps
->value
->u
.external
.node
,
1734 bitmap
, NULL
, errp
);
1746 if (bdrv_dirty_bitmap_check(bm
, BDRV_BITMAP_ALLOW_RO
, errp
)) {
1751 exp
->export_bitmaps
[i
] = bm
;
1752 assert(strlen(bitmap
) <= BDRV_BITMAP_MAX_NAME_SIZE
);
1755 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1756 for (i
= 0; i
< exp
->nr_export_bitmaps
; i
++) {
1757 bdrv_dirty_bitmap_set_busy(exp
->export_bitmaps
[i
], true);
1760 exp
->allocation_depth
= arg
->allocation_depth
;
1763 * We need to inhibit request queuing in the block layer to ensure we can
1764 * be properly quiesced when entering a drained section, as our coroutines
1765 * servicing pending requests might enter blk_pread().
1767 blk_set_disable_request_queuing(blk
, true);
1769 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1771 blk_set_dev_ops(blk
, &nbd_block_ops
, exp
);
1773 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1778 g_free(exp
->export_bitmaps
);
1780 g_free(exp
->description
);
1784 NBDExport
*nbd_export_find(const char *name
)
1787 QTAILQ_FOREACH(exp
, &exports
, next
) {
1788 if (strcmp(name
, exp
->name
) == 0) {
1797 nbd_export_aio_context(NBDExport
*exp
)
1799 return exp
->common
.ctx
;
1802 static void nbd_export_request_shutdown(BlockExport
*blk_exp
)
1804 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1805 NBDClient
*client
, *next
;
1807 blk_exp_ref(&exp
->common
);
1809 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1810 * close mode that stops advertising the export to new clients but
1811 * still permits existing clients to run to completion? Because of
1812 * that possibility, nbd_export_close() can be called more than
1813 * once on an export.
1815 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1816 client_close(client
, true);
1821 QTAILQ_REMOVE(&exports
, exp
, next
);
1823 blk_exp_unref(&exp
->common
);
1826 static void nbd_export_delete(BlockExport
*blk_exp
)
1829 NBDExport
*exp
= container_of(blk_exp
, NBDExport
, common
);
1831 assert(exp
->name
== NULL
);
1832 assert(QTAILQ_EMPTY(&exp
->clients
));
1834 g_free(exp
->description
);
1835 exp
->description
= NULL
;
1837 if (exp
->eject_notifier_blk
) {
1838 notifier_remove(&exp
->eject_notifier
);
1839 blk_unref(exp
->eject_notifier_blk
);
1841 blk_remove_aio_context_notifier(exp
->common
.blk
, blk_aio_attached
,
1842 blk_aio_detach
, exp
);
1843 blk_set_disable_request_queuing(exp
->common
.blk
, false);
1845 for (i
= 0; i
< exp
->nr_export_bitmaps
; i
++) {
1846 bdrv_dirty_bitmap_set_busy(exp
->export_bitmaps
[i
], false);
1850 const BlockExportDriver blk_exp_nbd
= {
1851 .type
= BLOCK_EXPORT_TYPE_NBD
,
1852 .instance_size
= sizeof(NBDExport
),
1853 .create
= nbd_export_create
,
1854 .delete = nbd_export_delete
,
1855 .request_shutdown
= nbd_export_request_shutdown
,
1858 static int coroutine_fn
nbd_co_send_iov(NBDClient
*client
, struct iovec
*iov
,
1859 unsigned niov
, Error
**errp
)
1863 g_assert(qemu_in_coroutine());
1864 qemu_co_mutex_lock(&client
->send_lock
);
1865 client
->send_coroutine
= qemu_coroutine_self();
1867 ret
= qio_channel_writev_all(client
->ioc
, iov
, niov
, errp
) < 0 ? -EIO
: 0;
1869 client
->send_coroutine
= NULL
;
1870 qemu_co_mutex_unlock(&client
->send_lock
);
1875 static inline void set_be_simple_reply(NBDSimpleReply
*reply
, uint64_t error
,
1878 stl_be_p(&reply
->magic
, NBD_SIMPLE_REPLY_MAGIC
);
1879 stl_be_p(&reply
->error
, error
);
1880 stq_be_p(&reply
->cookie
, cookie
);
1883 static int coroutine_fn
nbd_co_send_simple_reply(NBDClient
*client
,
1884 NBDRequest
*request
,
1890 NBDSimpleReply reply
;
1891 int nbd_err
= system_errno_to_nbd_errno(error
);
1892 struct iovec iov
[] = {
1893 {.iov_base
= &reply
, .iov_len
= sizeof(reply
)},
1894 {.iov_base
= data
, .iov_len
= len
}
1897 assert(!len
|| !nbd_err
);
1898 assert(!client
->structured_reply
|| request
->type
!= NBD_CMD_READ
);
1899 trace_nbd_co_send_simple_reply(request
->cookie
, nbd_err
,
1900 nbd_err_lookup(nbd_err
), len
);
1901 set_be_simple_reply(&reply
, nbd_err
, request
->cookie
);
1903 return nbd_co_send_iov(client
, iov
, 2, errp
);
1907 * Prepare the header of a reply chunk for network transmission.
1909 * On input, @iov is partially initialized: iov[0].iov_base must point
1910 * to an uninitialized NBDReply, while the remaining @niov elements
1911 * (if any) must be ready for transmission. This function then
1912 * populates iov[0] for transmission.
1914 static inline void set_be_chunk(NBDClient
*client
, struct iovec
*iov
,
1915 size_t niov
, uint16_t flags
, uint16_t type
,
1916 NBDRequest
*request
)
1918 /* TODO - handle structured vs. extended replies */
1919 NBDStructuredReplyChunk
*chunk
= iov
->iov_base
;
1920 size_t i
, length
= 0;
1922 for (i
= 1; i
< niov
; i
++) {
1923 length
+= iov
[i
].iov_len
;
1925 assert(length
<= NBD_MAX_BUFFER_SIZE
+ sizeof(NBDStructuredReadData
));
1927 iov
[0].iov_len
= sizeof(*chunk
);
1928 stl_be_p(&chunk
->magic
, NBD_STRUCTURED_REPLY_MAGIC
);
1929 stw_be_p(&chunk
->flags
, flags
);
1930 stw_be_p(&chunk
->type
, type
);
1931 stq_be_p(&chunk
->cookie
, request
->cookie
);
1932 stl_be_p(&chunk
->length
, length
);
1935 static int coroutine_fn
nbd_co_send_chunk_done(NBDClient
*client
,
1936 NBDRequest
*request
,
1940 struct iovec iov
[] = {
1944 trace_nbd_co_send_chunk_done(request
->cookie
);
1945 set_be_chunk(client
, iov
, 1, NBD_REPLY_FLAG_DONE
,
1946 NBD_REPLY_TYPE_NONE
, request
);
1947 return nbd_co_send_iov(client
, iov
, 1, errp
);
1950 static int coroutine_fn
nbd_co_send_chunk_read(NBDClient
*client
,
1951 NBDRequest
*request
,
1959 NBDStructuredReadData chunk
;
1960 struct iovec iov
[] = {
1962 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1963 {.iov_base
= data
, .iov_len
= size
}
1967 trace_nbd_co_send_chunk_read(request
->cookie
, offset
, data
, size
);
1968 set_be_chunk(client
, iov
, 3, final
? NBD_REPLY_FLAG_DONE
: 0,
1969 NBD_REPLY_TYPE_OFFSET_DATA
, request
);
1970 stq_be_p(&chunk
.offset
, offset
);
1972 return nbd_co_send_iov(client
, iov
, 3, errp
);
1975 static int coroutine_fn
nbd_co_send_chunk_error(NBDClient
*client
,
1976 NBDRequest
*request
,
1982 NBDStructuredError chunk
;
1983 int nbd_err
= system_errno_to_nbd_errno(error
);
1984 struct iovec iov
[] = {
1986 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1987 {.iov_base
= (char *)msg
, .iov_len
= msg
? strlen(msg
) : 0},
1991 trace_nbd_co_send_chunk_error(request
->cookie
, nbd_err
,
1992 nbd_err_lookup(nbd_err
), msg
? msg
: "");
1993 set_be_chunk(client
, iov
, 3, NBD_REPLY_FLAG_DONE
,
1994 NBD_REPLY_TYPE_ERROR
, request
);
1995 stl_be_p(&chunk
.error
, nbd_err
);
1996 stw_be_p(&chunk
.message_length
, iov
[2].iov_len
);
1998 return nbd_co_send_iov(client
, iov
, 3, errp
);
2001 /* Do a sparse read and send the structured reply to the client.
2002 * Returns -errno if sending fails. blk_co_block_status_above() failure is
2003 * reported to the client, at which point this function succeeds.
2005 static int coroutine_fn
nbd_co_send_sparse_read(NBDClient
*client
,
2006 NBDRequest
*request
,
2013 NBDExport
*exp
= client
->exp
;
2014 size_t progress
= 0;
2016 while (progress
< size
) {
2018 int status
= blk_co_block_status_above(exp
->common
.blk
, NULL
,
2020 size
- progress
, &pnum
, NULL
,
2025 char *msg
= g_strdup_printf("unable to check for holes: %s",
2028 ret
= nbd_co_send_chunk_error(client
, request
, -status
, msg
, errp
);
2032 assert(pnum
&& pnum
<= size
- progress
);
2033 final
= progress
+ pnum
== size
;
2034 if (status
& BDRV_BLOCK_ZERO
) {
2036 NBDStructuredReadHole chunk
;
2037 struct iovec iov
[] = {
2039 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
2042 trace_nbd_co_send_chunk_read_hole(request
->cookie
,
2043 offset
+ progress
, pnum
);
2044 set_be_chunk(client
, iov
, 2,
2045 final
? NBD_REPLY_FLAG_DONE
: 0,
2046 NBD_REPLY_TYPE_OFFSET_HOLE
, request
);
2047 stq_be_p(&chunk
.offset
, offset
+ progress
);
2048 stl_be_p(&chunk
.length
, pnum
);
2049 ret
= nbd_co_send_iov(client
, iov
, 2, errp
);
2051 ret
= blk_co_pread(exp
->common
.blk
, offset
+ progress
, pnum
,
2052 data
+ progress
, 0);
2054 error_setg_errno(errp
, -ret
, "reading from file failed");
2057 ret
= nbd_co_send_chunk_read(client
, request
, offset
+ progress
,
2058 data
+ progress
, pnum
, final
, errp
);
2069 typedef struct NBDExtentArray
{
2071 unsigned int nb_alloc
;
2073 uint64_t total_length
;
2075 bool converted_to_be
;
2078 static NBDExtentArray
*nbd_extent_array_new(unsigned int nb_alloc
)
2080 NBDExtentArray
*ea
= g_new0(NBDExtentArray
, 1);
2082 ea
->nb_alloc
= nb_alloc
;
2083 ea
->extents
= g_new(NBDExtent
, nb_alloc
);
2089 static void nbd_extent_array_free(NBDExtentArray
*ea
)
2091 g_free(ea
->extents
);
2094 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray
, nbd_extent_array_free
)
2096 /* Further modifications of the array after conversion are abandoned */
2097 static void nbd_extent_array_convert_to_be(NBDExtentArray
*ea
)
2101 assert(!ea
->converted_to_be
);
2102 ea
->can_add
= false;
2103 ea
->converted_to_be
= true;
2105 for (i
= 0; i
< ea
->count
; i
++) {
2106 ea
->extents
[i
].flags
= cpu_to_be32(ea
->extents
[i
].flags
);
2107 ea
->extents
[i
].length
= cpu_to_be32(ea
->extents
[i
].length
);
2112 * Add extent to NBDExtentArray. If extent can't be added (no available space),
2114 * For safety, when returning -1 for the first time, .can_add is set to false,
2115 * and further calls to nbd_extent_array_add() will crash.
2116 * (this avoids the situation where a caller ignores failure to add one extent,
2117 * where adding another extent that would squash into the last array entry
2118 * would result in an incorrect range reported to the client)
2120 static int nbd_extent_array_add(NBDExtentArray
*ea
,
2121 uint32_t length
, uint32_t flags
)
2123 assert(ea
->can_add
);
2129 /* Extend previous extent if flags are the same */
2130 if (ea
->count
> 0 && flags
== ea
->extents
[ea
->count
- 1].flags
) {
2131 uint64_t sum
= (uint64_t)length
+ ea
->extents
[ea
->count
- 1].length
;
2133 if (sum
<= UINT32_MAX
) {
2134 ea
->extents
[ea
->count
- 1].length
= sum
;
2135 ea
->total_length
+= length
;
2140 if (ea
->count
>= ea
->nb_alloc
) {
2141 ea
->can_add
= false;
2145 ea
->total_length
+= length
;
2146 ea
->extents
[ea
->count
] = (NBDExtent
) {.length
= length
, .flags
= flags
};
2152 static int coroutine_fn
blockstatus_to_extents(BlockBackend
*blk
,
2153 uint64_t offset
, uint64_t bytes
,
2159 int ret
= blk_co_block_status_above(blk
, NULL
, offset
, bytes
, &num
,
2166 flags
= (ret
& BDRV_BLOCK_DATA
? 0 : NBD_STATE_HOLE
) |
2167 (ret
& BDRV_BLOCK_ZERO
? NBD_STATE_ZERO
: 0);
2169 if (nbd_extent_array_add(ea
, num
, flags
) < 0) {
2180 static int coroutine_fn
blockalloc_to_extents(BlockBackend
*blk
,
2181 uint64_t offset
, uint64_t bytes
,
2186 int ret
= blk_co_is_allocated_above(blk
, NULL
, false, offset
, bytes
,
2193 if (nbd_extent_array_add(ea
, num
, ret
) < 0) {
2205 * nbd_co_send_extents
2207 * @ea is converted to BE by the function
2208 * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2210 static int coroutine_fn
2211 nbd_co_send_extents(NBDClient
*client
, NBDRequest
*request
, NBDExtentArray
*ea
,
2212 bool last
, uint32_t context_id
, Error
**errp
)
2215 NBDStructuredMeta chunk
;
2216 struct iovec iov
[] = {
2218 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
2219 {.iov_base
= ea
->extents
, .iov_len
= ea
->count
* sizeof(ea
->extents
[0])}
2222 nbd_extent_array_convert_to_be(ea
);
2224 trace_nbd_co_send_extents(request
->cookie
, ea
->count
, context_id
,
2225 ea
->total_length
, last
);
2226 set_be_chunk(client
, iov
, 3, last
? NBD_REPLY_FLAG_DONE
: 0,
2227 NBD_REPLY_TYPE_BLOCK_STATUS
, request
);
2228 stl_be_p(&chunk
.context_id
, context_id
);
2230 return nbd_co_send_iov(client
, iov
, 3, errp
);
2233 /* Get block status from the exported device and send it to the client */
2235 coroutine_fn
nbd_co_send_block_status(NBDClient
*client
, NBDRequest
*request
,
2236 BlockBackend
*blk
, uint64_t offset
,
2237 uint32_t length
, bool dont_fragment
,
2238 bool last
, uint32_t context_id
,
2242 unsigned int nb_extents
= dont_fragment
? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS
;
2243 g_autoptr(NBDExtentArray
) ea
= nbd_extent_array_new(nb_extents
);
2245 if (context_id
== NBD_META_ID_BASE_ALLOCATION
) {
2246 ret
= blockstatus_to_extents(blk
, offset
, length
, ea
);
2248 ret
= blockalloc_to_extents(blk
, offset
, length
, ea
);
2251 return nbd_co_send_chunk_error(client
, request
, -ret
,
2252 "can't get block status", errp
);
2255 return nbd_co_send_extents(client
, request
, ea
, last
, context_id
, errp
);
2258 /* Populate @ea from a dirty bitmap. */
2259 static void bitmap_to_extents(BdrvDirtyBitmap
*bitmap
,
2260 uint64_t offset
, uint64_t length
,
2263 int64_t start
, dirty_start
, dirty_count
;
2264 int64_t end
= offset
+ length
;
2267 bdrv_dirty_bitmap_lock(bitmap
);
2269 for (start
= offset
;
2270 bdrv_dirty_bitmap_next_dirty_area(bitmap
, start
, end
, INT32_MAX
,
2271 &dirty_start
, &dirty_count
);
2272 start
= dirty_start
+ dirty_count
)
2274 if ((nbd_extent_array_add(es
, dirty_start
- start
, 0) < 0) ||
2275 (nbd_extent_array_add(es
, dirty_count
, NBD_STATE_DIRTY
) < 0))
2283 /* last non dirty extent, nothing to do if array is now full */
2284 (void) nbd_extent_array_add(es
, end
- start
, 0);
2287 bdrv_dirty_bitmap_unlock(bitmap
);
2290 static int coroutine_fn
nbd_co_send_bitmap(NBDClient
*client
,
2291 NBDRequest
*request
,
2292 BdrvDirtyBitmap
*bitmap
,
2294 uint32_t length
, bool dont_fragment
,
2295 bool last
, uint32_t context_id
,
2298 unsigned int nb_extents
= dont_fragment
? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS
;
2299 g_autoptr(NBDExtentArray
) ea
= nbd_extent_array_new(nb_extents
);
2301 bitmap_to_extents(bitmap
, offset
, length
, ea
);
2303 return nbd_co_send_extents(client
, request
, ea
, last
, context_id
, errp
);
2306 /* nbd_co_receive_request
2307 * Collect a client request. Return 0 if request looks valid, -EIO to drop
2308 * connection right away, -EAGAIN to indicate we were interrupted and the
2309 * channel should be quiesced, and any other negative value to report an error
2310 * to the client (although the caller may still need to disconnect after
2311 * reporting the error).
2313 static int coroutine_fn
nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
2316 NBDClient
*client
= req
->client
;
2320 g_assert(qemu_in_coroutine());
2321 assert(client
->recv_coroutine
== qemu_coroutine_self());
2322 ret
= nbd_receive_request(client
, request
, errp
);
2327 trace_nbd_co_receive_request_decode_type(request
->cookie
, request
->type
,
2328 nbd_cmd_lookup(request
->type
));
2330 if (request
->type
!= NBD_CMD_WRITE
) {
2331 /* No payload, we are ready to read the next request. */
2332 req
->complete
= true;
2335 if (request
->type
== NBD_CMD_DISC
) {
2336 /* Special case: we're going to disconnect without a reply,
2337 * whether or not flags, from, or len are bogus */
2341 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
||
2342 request
->type
== NBD_CMD_CACHE
)
2344 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
2345 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
2346 request
->len
, NBD_MAX_BUFFER_SIZE
);
2350 if (request
->type
!= NBD_CMD_CACHE
) {
2351 req
->data
= blk_try_blockalign(client
->exp
->common
.blk
,
2353 if (req
->data
== NULL
) {
2354 error_setg(errp
, "No memory");
2360 if (request
->type
== NBD_CMD_WRITE
) {
2361 if (nbd_read(client
->ioc
, req
->data
, request
->len
, "CMD_WRITE data",
2366 req
->complete
= true;
2368 trace_nbd_co_receive_request_payload_received(request
->cookie
,
2372 /* Sanity checks. */
2373 if (client
->exp
->nbdflags
& NBD_FLAG_READ_ONLY
&&
2374 (request
->type
== NBD_CMD_WRITE
||
2375 request
->type
== NBD_CMD_WRITE_ZEROES
||
2376 request
->type
== NBD_CMD_TRIM
)) {
2377 error_setg(errp
, "Export is read-only");
2380 if (request
->from
> client
->exp
->size
||
2381 request
->len
> client
->exp
->size
- request
->from
) {
2382 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
2383 ", Size: %" PRIu64
, request
->from
, request
->len
,
2385 return (request
->type
== NBD_CMD_WRITE
||
2386 request
->type
== NBD_CMD_WRITE_ZEROES
) ? -ENOSPC
: -EINVAL
;
2388 if (client
->check_align
&& !QEMU_IS_ALIGNED(request
->from
| request
->len
,
2389 client
->check_align
)) {
2391 * The block layer gracefully handles unaligned requests, but
2392 * it's still worth tracing client non-compliance
2394 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request
->type
),
2397 client
->check_align
);
2399 valid_flags
= NBD_CMD_FLAG_FUA
;
2400 if (request
->type
== NBD_CMD_READ
&& client
->structured_reply
) {
2401 valid_flags
|= NBD_CMD_FLAG_DF
;
2402 } else if (request
->type
== NBD_CMD_WRITE_ZEROES
) {
2403 valid_flags
|= NBD_CMD_FLAG_NO_HOLE
| NBD_CMD_FLAG_FAST_ZERO
;
2404 } else if (request
->type
== NBD_CMD_BLOCK_STATUS
) {
2405 valid_flags
|= NBD_CMD_FLAG_REQ_ONE
;
2407 if (request
->flags
& ~valid_flags
) {
2408 error_setg(errp
, "unsupported flags for command %s (got 0x%x)",
2409 nbd_cmd_lookup(request
->type
), request
->flags
);
2416 /* Send simple reply without a payload, or a structured error
2417 * @error_msg is ignored if @ret >= 0
2418 * Returns 0 if connection is still live, -errno on failure to talk to client
2420 static coroutine_fn
int nbd_send_generic_reply(NBDClient
*client
,
2421 NBDRequest
*request
,
2423 const char *error_msg
,
2426 if (client
->structured_reply
&& ret
< 0) {
2427 return nbd_co_send_chunk_error(client
, request
, -ret
, error_msg
, errp
);
2429 return nbd_co_send_simple_reply(client
, request
, ret
< 0 ? -ret
: 0,
2434 /* Handle NBD_CMD_READ request.
2435 * Return -errno if sending fails. Other errors are reported directly to the
2436 * client as an error reply. */
2437 static coroutine_fn
int nbd_do_cmd_read(NBDClient
*client
, NBDRequest
*request
,
2438 uint8_t *data
, Error
**errp
)
2441 NBDExport
*exp
= client
->exp
;
2443 assert(request
->type
== NBD_CMD_READ
);
2445 /* XXX: NBD Protocol only documents use of FUA with WRITE */
2446 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2447 ret
= blk_co_flush(exp
->common
.blk
);
2449 return nbd_send_generic_reply(client
, request
, ret
,
2450 "flush failed", errp
);
2454 if (client
->structured_reply
&& !(request
->flags
& NBD_CMD_FLAG_DF
) &&
2457 return nbd_co_send_sparse_read(client
, request
, request
->from
,
2458 data
, request
->len
, errp
);
2461 ret
= blk_co_pread(exp
->common
.blk
, request
->from
, request
->len
, data
, 0);
2463 return nbd_send_generic_reply(client
, request
, ret
,
2464 "reading from file failed", errp
);
2467 if (client
->structured_reply
) {
2469 return nbd_co_send_chunk_read(client
, request
, request
->from
, data
,
2470 request
->len
, true, errp
);
2472 return nbd_co_send_chunk_done(client
, request
, errp
);
2475 return nbd_co_send_simple_reply(client
, request
, 0,
2476 data
, request
->len
, errp
);
2483 * Handle NBD_CMD_CACHE request.
2484 * Return -errno if sending fails. Other errors are reported directly to the
2485 * client as an error reply.
2487 static coroutine_fn
int nbd_do_cmd_cache(NBDClient
*client
, NBDRequest
*request
,
2491 NBDExport
*exp
= client
->exp
;
2493 assert(request
->type
== NBD_CMD_CACHE
);
2495 ret
= blk_co_preadv(exp
->common
.blk
, request
->from
, request
->len
,
2496 NULL
, BDRV_REQ_COPY_ON_READ
| BDRV_REQ_PREFETCH
);
2498 return nbd_send_generic_reply(client
, request
, ret
,
2499 "caching data failed", errp
);
2502 /* Handle NBD request.
2503 * Return -errno if sending fails. Other errors are reported directly to the
2504 * client as an error reply. */
2505 static coroutine_fn
int nbd_handle_request(NBDClient
*client
,
2506 NBDRequest
*request
,
2507 uint8_t *data
, Error
**errp
)
2511 NBDExport
*exp
= client
->exp
;
2515 switch (request
->type
) {
2517 return nbd_do_cmd_cache(client
, request
, errp
);
2520 return nbd_do_cmd_read(client
, request
, data
, errp
);
2524 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2525 flags
|= BDRV_REQ_FUA
;
2527 ret
= blk_co_pwrite(exp
->common
.blk
, request
->from
, request
->len
, data
,
2529 return nbd_send_generic_reply(client
, request
, ret
,
2530 "writing to file failed", errp
);
2532 case NBD_CMD_WRITE_ZEROES
:
2534 if (request
->flags
& NBD_CMD_FLAG_FUA
) {
2535 flags
|= BDRV_REQ_FUA
;
2537 if (!(request
->flags
& NBD_CMD_FLAG_NO_HOLE
)) {
2538 flags
|= BDRV_REQ_MAY_UNMAP
;
2540 if (request
->flags
& NBD_CMD_FLAG_FAST_ZERO
) {
2541 flags
|= BDRV_REQ_NO_FALLBACK
;
2543 ret
= blk_co_pwrite_zeroes(exp
->common
.blk
, request
->from
, request
->len
,
2545 return nbd_send_generic_reply(client
, request
, ret
,
2546 "writing to file failed", errp
);
2549 /* unreachable, thanks to special case in nbd_co_receive_request() */
2553 ret
= blk_co_flush(exp
->common
.blk
);
2554 return nbd_send_generic_reply(client
, request
, ret
,
2555 "flush failed", errp
);
2558 ret
= blk_co_pdiscard(exp
->common
.blk
, request
->from
, request
->len
);
2559 if (ret
>= 0 && request
->flags
& NBD_CMD_FLAG_FUA
) {
2560 ret
= blk_co_flush(exp
->common
.blk
);
2562 return nbd_send_generic_reply(client
, request
, ret
,
2563 "discard failed", errp
);
2565 case NBD_CMD_BLOCK_STATUS
:
2566 if (!request
->len
) {
2567 return nbd_send_generic_reply(client
, request
, -EINVAL
,
2568 "need non-zero length", errp
);
2570 if (client
->export_meta
.count
) {
2571 bool dont_fragment
= request
->flags
& NBD_CMD_FLAG_REQ_ONE
;
2572 int contexts_remaining
= client
->export_meta
.count
;
2574 if (client
->export_meta
.base_allocation
) {
2575 ret
= nbd_co_send_block_status(client
, request
,
2578 request
->len
, dont_fragment
,
2579 !--contexts_remaining
,
2580 NBD_META_ID_BASE_ALLOCATION
,
2587 if (client
->export_meta
.allocation_depth
) {
2588 ret
= nbd_co_send_block_status(client
, request
,
2590 request
->from
, request
->len
,
2592 !--contexts_remaining
,
2593 NBD_META_ID_ALLOCATION_DEPTH
,
2600 for (i
= 0; i
< client
->exp
->nr_export_bitmaps
; i
++) {
2601 if (!client
->export_meta
.bitmaps
[i
]) {
2604 ret
= nbd_co_send_bitmap(client
, request
,
2605 client
->exp
->export_bitmaps
[i
],
2606 request
->from
, request
->len
,
2607 dont_fragment
, !--contexts_remaining
,
2608 NBD_META_ID_DIRTY_BITMAP
+ i
, errp
);
2614 assert(!contexts_remaining
);
2618 return nbd_send_generic_reply(client
, request
, -EINVAL
,
2619 "CMD_BLOCK_STATUS not negotiated",
2624 msg
= g_strdup_printf("invalid request type (%" PRIu32
") received",
2626 ret
= nbd_send_generic_reply(client
, request
, -EINVAL
, msg
,
2633 /* Owns a reference to the NBDClient passed as opaque. */
2634 static coroutine_fn
void nbd_trip(void *opaque
)
2636 NBDClient
*client
= opaque
;
2637 NBDRequestData
*req
;
2638 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
2640 Error
*local_err
= NULL
;
2643 if (client
->closing
) {
2644 nbd_client_put(client
);
2648 if (client
->quiescing
) {
2650 * We're switching between AIO contexts. Don't attempt to receive a new
2651 * request and kick the main context which may be waiting for us.
2653 nbd_client_put(client
);
2654 client
->recv_coroutine
= NULL
;
2659 req
= nbd_request_get(client
);
2660 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
2661 client
->recv_coroutine
= NULL
;
2663 if (client
->closing
) {
2665 * The client may be closed when we are blocked in
2666 * nbd_co_receive_request()
2671 if (ret
== -EAGAIN
) {
2672 assert(client
->quiescing
);
2676 nbd_client_receive_next_request(client
);
2681 qio_channel_set_cork(client
->ioc
, true);
2684 /* It wasn't -EIO, so, according to nbd_co_receive_request()
2685 * semantics, we should return the error to the client. */
2686 Error
*export_err
= local_err
;
2689 ret
= nbd_send_generic_reply(client
, &request
, -EINVAL
,
2690 error_get_pretty(export_err
), &local_err
);
2691 error_free(export_err
);
2693 ret
= nbd_handle_request(client
, &request
, req
->data
, &local_err
);
2696 error_prepend(&local_err
, "Failed to send reply: ");
2700 /* We must disconnect after NBD_CMD_WRITE if we did not
2703 if (!req
->complete
) {
2704 error_setg(&local_err
, "Request handling failed in intermediate state");
2708 qio_channel_set_cork(client
->ioc
, false);
2710 nbd_request_put(req
);
2711 nbd_client_put(client
);
2716 error_reportf_err(local_err
, "Disconnect client, due to: ");
2718 nbd_request_put(req
);
2719 client_close(client
, true);
2720 nbd_client_put(client
);
2723 static void nbd_client_receive_next_request(NBDClient
*client
)
2725 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
&&
2726 !client
->quiescing
) {
2727 nbd_client_get(client
);
2728 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
2729 aio_co_schedule(client
->exp
->common
.ctx
, client
->recv_coroutine
);
2733 static coroutine_fn
void nbd_co_client_start(void *opaque
)
2735 NBDClient
*client
= opaque
;
2736 Error
*local_err
= NULL
;
2738 qemu_co_mutex_init(&client
->send_lock
);
2740 if (nbd_negotiate(client
, &local_err
)) {
2742 error_report_err(local_err
);
2744 client_close(client
, false);
2748 nbd_client_receive_next_request(client
);
2752 * Create a new client listener using the given channel @sioc.
2753 * Begin servicing it in a coroutine. When the connection closes, call
2754 * @close_fn with an indication of whether the client completed negotiation.
2756 void nbd_client_new(QIOChannelSocket
*sioc
,
2757 QCryptoTLSCreds
*tlscreds
,
2758 const char *tlsauthz
,
2759 void (*close_fn
)(NBDClient
*, bool))
2764 client
= g_new0(NBDClient
, 1);
2765 client
->refcount
= 1;
2766 client
->tlscreds
= tlscreds
;
2768 object_ref(OBJECT(client
->tlscreds
));
2770 client
->tlsauthz
= g_strdup(tlsauthz
);
2771 client
->sioc
= sioc
;
2772 qio_channel_set_delay(QIO_CHANNEL(sioc
), false);
2773 object_ref(OBJECT(client
->sioc
));
2774 client
->ioc
= QIO_CHANNEL(sioc
);
2775 object_ref(OBJECT(client
->ioc
));
2776 client
->close_fn
= close_fn
;
2778 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
2779 qemu_coroutine_enter(co
);