2 * Copyright (C) 2016-2017 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
5 * Network Block Device Server Side
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; under version 2 of the License.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
23 #include "nbd-internal.h"
25 static int system_errno_to_nbd_errno(int err
)
51 /* Definitions for opaque data types */
53 typedef struct NBDRequestData NBDRequestData
;
55 struct NBDRequestData
{
56 QSIMPLEQ_ENTRY(NBDRequestData
) entry
;
64 void (*close
)(NBDExport
*exp
);
72 QTAILQ_HEAD(, NBDClient
) clients
;
73 QTAILQ_ENTRY(NBDExport
) next
;
77 BlockBackend
*eject_notifier_blk
;
78 Notifier eject_notifier
;
81 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
85 void (*close_fn
)(NBDClient
*client
, bool negotiated
);
88 QCryptoTLSCreds
*tlscreds
;
90 QIOChannelSocket
*sioc
; /* The underlying data channel */
91 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
93 Coroutine
*recv_coroutine
;
96 Coroutine
*send_coroutine
;
98 QTAILQ_ENTRY(NBDClient
) next
;
103 /* That's all folks */
105 static void nbd_client_receive_next_request(NBDClient
*client
);
107 /* Basic flow for negotiation
134 /* Send a reply header, including length, but no payload.
135 * Return -errno on error, 0 on success. */
136 static int nbd_negotiate_send_rep_len(QIOChannel
*ioc
, uint32_t type
,
137 uint32_t opt
, uint32_t len
, Error
**errp
)
141 trace_nbd_negotiate_send_rep_len(opt
, nbd_opt_lookup(opt
),
142 type
, nbd_rep_lookup(type
), len
);
144 assert(len
< NBD_MAX_BUFFER_SIZE
);
145 magic
= cpu_to_be64(NBD_REP_MAGIC
);
146 if (nbd_write(ioc
, &magic
, sizeof(magic
), errp
) < 0) {
147 error_prepend(errp
, "write failed (rep magic): ");
151 opt
= cpu_to_be32(opt
);
152 if (nbd_write(ioc
, &opt
, sizeof(opt
), errp
) < 0) {
153 error_prepend(errp
, "write failed (rep opt): ");
157 type
= cpu_to_be32(type
);
158 if (nbd_write(ioc
, &type
, sizeof(type
), errp
) < 0) {
159 error_prepend(errp
, "write failed (rep type): ");
163 len
= cpu_to_be32(len
);
164 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
165 error_prepend(errp
, "write failed (rep data length): ");
171 /* Send a reply header with default 0 length.
172 * Return -errno on error, 0 on success. */
173 static int nbd_negotiate_send_rep(QIOChannel
*ioc
, uint32_t type
, uint32_t opt
,
176 return nbd_negotiate_send_rep_len(ioc
, type
, opt
, 0, errp
);
179 /* Send an error reply.
180 * Return -errno on error, 0 on success. */
181 static int GCC_FMT_ATTR(5, 6)
182 nbd_negotiate_send_rep_err(QIOChannel
*ioc
, uint32_t type
,
183 uint32_t opt
, Error
**errp
, const char *fmt
, ...)
191 msg
= g_strdup_vprintf(fmt
, va
);
195 trace_nbd_negotiate_send_rep_err(msg
);
196 ret
= nbd_negotiate_send_rep_len(ioc
, type
, opt
, len
, errp
);
200 if (nbd_write(ioc
, msg
, len
, errp
) < 0) {
201 error_prepend(errp
, "write failed (error message): ");
212 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
213 * Return -errno on error, 0 on success. */
214 static int nbd_negotiate_send_rep_list(QIOChannel
*ioc
, NBDExport
*exp
,
217 size_t name_len
, desc_len
;
219 const char *name
= exp
->name
? exp
->name
: "";
220 const char *desc
= exp
->description
? exp
->description
: "";
223 trace_nbd_negotiate_send_rep_list(name
, desc
);
224 name_len
= strlen(name
);
225 desc_len
= strlen(desc
);
226 len
= name_len
+ desc_len
+ sizeof(len
);
227 ret
= nbd_negotiate_send_rep_len(ioc
, NBD_REP_SERVER
, NBD_OPT_LIST
, len
,
233 len
= cpu_to_be32(name_len
);
234 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
235 error_prepend(errp
, "write failed (name length): ");
239 if (nbd_write(ioc
, name
, name_len
, errp
) < 0) {
240 error_prepend(errp
, "write failed (name buffer): ");
244 if (nbd_write(ioc
, desc
, desc_len
, errp
) < 0) {
245 error_prepend(errp
, "write failed (description buffer): ");
252 /* Process the NBD_OPT_LIST command, with a potential series of replies.
253 * Return -errno on error, 0 on success. */
254 static int nbd_negotiate_handle_list(NBDClient
*client
, uint32_t length
,
260 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
263 return nbd_negotiate_send_rep_err(client
->ioc
,
264 NBD_REP_ERR_INVALID
, NBD_OPT_LIST
,
266 "OPT_LIST should not have length");
269 /* For each export, send a NBD_REP_SERVER reply. */
270 QTAILQ_FOREACH(exp
, &exports
, next
) {
271 if (nbd_negotiate_send_rep_list(client
->ioc
, exp
, errp
)) {
275 /* Finish with a NBD_REP_ACK. */
276 return nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, NBD_OPT_LIST
, errp
);
279 /* Send a reply to NBD_OPT_EXPORT_NAME.
280 * Return -errno on error, 0 on success. */
281 static int nbd_negotiate_handle_export_name(NBDClient
*client
, uint32_t length
,
282 uint16_t myflags
, bool no_zeroes
,
285 char name
[NBD_MAX_NAME_SIZE
+ 1];
286 char buf
[8 + 4 + 124] = "";
291 [20 .. xx] export name (length bytes)
293 trace_nbd_negotiate_handle_export_name();
294 if (length
>= sizeof(name
)) {
295 error_setg(errp
, "Bad length received");
298 if (nbd_read(client
->ioc
, name
, length
, errp
) < 0) {
299 error_prepend(errp
, "read failed: ");
304 trace_nbd_negotiate_handle_export_name_request(name
);
306 client
->exp
= nbd_export_find(name
);
308 error_setg(errp
, "export not found");
312 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
,
313 client
->exp
->nbdflags
| myflags
);
314 stq_be_p(buf
, client
->exp
->size
);
315 stw_be_p(buf
+ 8, client
->exp
->nbdflags
| myflags
);
316 len
= no_zeroes
? 10 : sizeof(buf
);
317 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
319 error_prepend(errp
, "write failed: ");
323 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
324 nbd_export_get(client
->exp
);
329 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
330 * The buffer does NOT include the info type prefix.
331 * Return -errno on error, 0 if ready to send more. */
332 static int nbd_negotiate_send_info(NBDClient
*client
, uint32_t opt
,
333 uint16_t info
, uint32_t length
, void *buf
,
338 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
339 rc
= nbd_negotiate_send_rep_len(client
->ioc
, NBD_REP_INFO
, opt
,
340 sizeof(info
) + length
, errp
);
345 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
348 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
354 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
355 * Return -errno on error, 0 if ready for next option, and 1 to move
356 * into transmission phase. */
357 static int nbd_negotiate_handle_info(NBDClient
*client
, uint32_t length
,
358 uint32_t opt
, uint16_t myflags
,
362 char name
[NBD_MAX_NAME_SIZE
+ 1];
367 bool sendname
= false;
368 bool blocksize
= false;
370 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
374 4 bytes: L, name length (can be 0)
376 2 bytes: N, number of requests (can be 0)
377 N * 2 bytes: N requests
379 if (length
< sizeof(namelen
) + sizeof(requests
)) {
380 msg
= "overall request too short";
383 if (nbd_read(client
->ioc
, &namelen
, sizeof(namelen
), errp
) < 0) {
386 be32_to_cpus(&namelen
);
387 length
-= sizeof(namelen
);
388 if (namelen
> length
- sizeof(requests
) || (length
- namelen
) % 2) {
389 msg
= "name length is incorrect";
392 if (nbd_read(client
->ioc
, name
, namelen
, errp
) < 0) {
395 name
[namelen
] = '\0';
397 trace_nbd_negotiate_handle_export_name_request(name
);
399 if (nbd_read(client
->ioc
, &requests
, sizeof(requests
), errp
) < 0) {
402 be16_to_cpus(&requests
);
403 length
-= sizeof(requests
);
404 trace_nbd_negotiate_handle_info_requests(requests
);
405 if (requests
!= length
/ sizeof(request
)) {
406 msg
= "incorrect number of requests for overall length";
410 if (nbd_read(client
->ioc
, &request
, sizeof(request
), errp
) < 0) {
413 be16_to_cpus(&request
);
414 length
-= sizeof(request
);
415 trace_nbd_negotiate_handle_info_request(request
,
416 nbd_info_lookup(request
));
417 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
418 * everything else is either a request we don't know or
419 * something we send regardless of request */
424 case NBD_INFO_BLOCK_SIZE
:
430 exp
= nbd_export_find(name
);
432 return nbd_negotiate_send_rep_err(client
->ioc
, NBD_REP_ERR_UNKNOWN
,
433 opt
, errp
, "export '%s' not present",
437 /* Don't bother sending NBD_INFO_NAME unless client requested it */
439 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_NAME
, length
, name
,
446 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
448 if (exp
->description
) {
449 size_t len
= strlen(exp
->description
);
451 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_DESCRIPTION
,
452 len
, exp
->description
, errp
);
458 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
459 * according to whether the client requested it, and according to
460 * whether this is OPT_INFO or OPT_GO. */
461 /* minimum - 1 for back-compat, or 512 if client is new enough.
462 * TODO: consult blk_bs(blk)->bl.request_alignment? */
463 sizes
[0] = (opt
== NBD_OPT_INFO
|| blocksize
) ? BDRV_SECTOR_SIZE
: 1;
464 /* preferred - Hard-code to 4096 for now.
465 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
467 /* maximum - At most 32M, but smaller as appropriate. */
468 sizes
[2] = MIN(blk_get_max_transfer(exp
->blk
), NBD_MAX_BUFFER_SIZE
);
469 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
470 cpu_to_be32s(&sizes
[0]);
471 cpu_to_be32s(&sizes
[1]);
472 cpu_to_be32s(&sizes
[2]);
473 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_BLOCK_SIZE
,
474 sizeof(sizes
), sizes
, errp
);
479 /* Send NBD_INFO_EXPORT always */
480 trace_nbd_negotiate_new_style_size_flags(exp
->size
,
481 exp
->nbdflags
| myflags
);
482 stq_be_p(buf
, exp
->size
);
483 stw_be_p(buf
+ 8, exp
->nbdflags
| myflags
);
484 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_EXPORT
,
485 sizeof(buf
), buf
, errp
);
490 /* If the client is just asking for NBD_OPT_INFO, but forgot to
491 * request block sizes, return an error.
492 * TODO: consult blk_bs(blk)->request_align, and only error if it
494 if (opt
== NBD_OPT_INFO
&& !blocksize
) {
495 return nbd_negotiate_send_rep_err(client
->ioc
,
496 NBD_REP_ERR_BLOCK_SIZE_REQD
, opt
,
498 "request NBD_INFO_BLOCK_SIZE to "
503 rc
= nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, opt
, errp
);
508 if (opt
== NBD_OPT_GO
) {
510 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
511 nbd_export_get(client
->exp
);
517 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
520 return nbd_negotiate_send_rep_err(client
->ioc
, NBD_REP_ERR_INVALID
, opt
,
525 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
526 * new channel for all further (now-encrypted) communication. */
527 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
533 struct NBDTLSHandshakeData data
= { 0 };
535 trace_nbd_negotiate_handle_starttls();
538 if (nbd_drop(ioc
, length
, errp
) < 0) {
541 nbd_negotiate_send_rep_err(ioc
, NBD_REP_ERR_INVALID
, NBD_OPT_STARTTLS
,
543 "OPT_STARTTLS should not have length");
547 if (nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
,
548 NBD_OPT_STARTTLS
, errp
) < 0) {
552 tioc
= qio_channel_tls_new_server(ioc
,
560 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
561 trace_nbd_negotiate_handle_starttls_handshake();
562 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
563 qio_channel_tls_handshake(tioc
,
568 if (!data
.complete
) {
569 g_main_loop_run(data
.loop
);
571 g_main_loop_unref(data
.loop
);
573 object_unref(OBJECT(tioc
));
574 error_propagate(errp
, data
.error
);
578 return QIO_CHANNEL(tioc
);
581 /* nbd_negotiate_options
582 * Process all NBD_OPT_* client option commands, during fixed newstyle
585 * -errno on error, errp is set
586 * 0 on successful negotiation, errp is not set
587 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
590 static int nbd_negotiate_options(NBDClient
*client
, uint16_t myflags
,
594 bool fixedNewstyle
= false;
595 bool no_zeroes
= false;
598 [ 0 .. 3] client flags
600 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
601 [ 0 .. 7] NBD_OPTS_MAGIC
602 [ 8 .. 11] NBD option
603 [12 .. 15] Data length
606 [ 0 .. 7] NBD_OPTS_MAGIC
607 [ 8 .. 11] Second NBD option
608 [12 .. 15] Data length
612 if (nbd_read(client
->ioc
, &flags
, sizeof(flags
), errp
) < 0) {
613 error_prepend(errp
, "read failed: ");
616 be32_to_cpus(&flags
);
617 trace_nbd_negotiate_options_flags(flags
);
618 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
619 fixedNewstyle
= true;
620 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
622 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
624 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
627 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
633 uint32_t option
, length
;
636 if (nbd_read(client
->ioc
, &magic
, sizeof(magic
), errp
) < 0) {
637 error_prepend(errp
, "read failed: ");
640 magic
= be64_to_cpu(magic
);
641 trace_nbd_negotiate_options_check_magic(magic
);
642 if (magic
!= NBD_OPTS_MAGIC
) {
643 error_setg(errp
, "Bad magic received");
647 if (nbd_read(client
->ioc
, &option
,
648 sizeof(option
), errp
) < 0) {
649 error_prepend(errp
, "read failed: ");
652 option
= be32_to_cpu(option
);
654 if (nbd_read(client
->ioc
, &length
, sizeof(length
), errp
) < 0) {
655 error_prepend(errp
, "read failed: ");
658 length
= be32_to_cpu(length
);
660 trace_nbd_negotiate_options_check_option(option
,
661 nbd_opt_lookup(option
));
662 if (client
->tlscreds
&&
663 client
->ioc
== (QIOChannel
*)client
->sioc
) {
665 if (!fixedNewstyle
) {
666 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
670 case NBD_OPT_STARTTLS
:
671 tioc
= nbd_negotiate_handle_starttls(client
, length
, errp
);
675 object_unref(OBJECT(client
->ioc
));
676 client
->ioc
= QIO_CHANNEL(tioc
);
679 case NBD_OPT_EXPORT_NAME
:
680 /* No way to return an error to client, so drop connection */
681 error_setg(errp
, "Option 0x%x not permitted before TLS",
686 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
689 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
690 NBD_REP_ERR_TLS_REQD
,
693 "not permitted before TLS",
698 /* Let the client keep trying, unless they asked to
699 * quit. In this mode, we've already sent an error, so
700 * we can't ack the abort. */
701 if (option
== NBD_OPT_ABORT
) {
706 } else if (fixedNewstyle
) {
709 ret
= nbd_negotiate_handle_list(client
, length
, errp
);
716 /* NBD spec says we must try to reply before
717 * disconnecting, but that we must also tolerate
718 * guests that don't wait for our reply. */
719 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, option
, NULL
);
722 case NBD_OPT_EXPORT_NAME
:
723 return nbd_negotiate_handle_export_name(client
, length
,
729 ret
= nbd_negotiate_handle_info(client
, length
, option
,
732 assert(option
== NBD_OPT_GO
);
740 case NBD_OPT_STARTTLS
:
741 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
744 if (client
->tlscreds
) {
745 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
748 "TLS already enabled");
750 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
753 "TLS not configured");
760 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
763 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
766 "Unsupported option 0x%"
767 PRIx32
" (%s)", option
,
768 nbd_opt_lookup(option
));
776 * If broken new-style we should drop the connection
777 * for anything except NBD_OPT_EXPORT_NAME
780 case NBD_OPT_EXPORT_NAME
:
781 return nbd_negotiate_handle_export_name(client
, length
,
786 error_setg(errp
, "Unsupported option 0x%" PRIx32
" (%s)",
787 option
, nbd_opt_lookup(option
));
796 * -errno on error, errp is set
797 * 0 on successful negotiation, errp is not set
798 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
801 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
803 char buf
[8 + 8 + 8 + 128];
805 const uint16_t myflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
806 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
|
807 NBD_FLAG_SEND_WRITE_ZEROES
);
810 /* Old style negotiation header without options
811 [ 0 .. 7] passwd ("NBDMAGIC")
812 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
814 [24 .. 25] server flags (0)
815 [26 .. 27] export flags
816 [28 .. 151] reserved (0)
818 New style negotiation header with options
819 [ 0 .. 7] passwd ("NBDMAGIC")
820 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
821 [16 .. 17] server flags (0)
822 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
825 qio_channel_set_blocking(client
->ioc
, false, NULL
);
827 trace_nbd_negotiate_begin();
828 memset(buf
, 0, sizeof(buf
));
829 memcpy(buf
, "NBDMAGIC", 8);
831 oldStyle
= client
->exp
!= NULL
&& !client
->tlscreds
;
833 trace_nbd_negotiate_old_style(client
->exp
->size
,
834 client
->exp
->nbdflags
| myflags
);
835 stq_be_p(buf
+ 8, NBD_CLIENT_MAGIC
);
836 stq_be_p(buf
+ 16, client
->exp
->size
);
837 stw_be_p(buf
+ 26, client
->exp
->nbdflags
| myflags
);
839 if (nbd_write(client
->ioc
, buf
, sizeof(buf
), errp
) < 0) {
840 error_prepend(errp
, "write failed: ");
844 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
845 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
847 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
848 error_prepend(errp
, "write failed: ");
851 ret
= nbd_negotiate_options(client
, myflags
, errp
);
854 error_prepend(errp
, "option negotiation failed: ");
860 trace_nbd_negotiate_success();
865 static int nbd_receive_request(QIOChannel
*ioc
, NBDRequest
*request
,
868 uint8_t buf
[NBD_REQUEST_SIZE
];
872 ret
= nbd_read(ioc
, buf
, sizeof(buf
), errp
);
878 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
879 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
880 [ 6 .. 7] type (NBD_CMD_READ, ...)
886 magic
= ldl_be_p(buf
);
887 request
->flags
= lduw_be_p(buf
+ 4);
888 request
->type
= lduw_be_p(buf
+ 6);
889 request
->handle
= ldq_be_p(buf
+ 8);
890 request
->from
= ldq_be_p(buf
+ 16);
891 request
->len
= ldl_be_p(buf
+ 24);
893 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
894 request
->from
, request
->len
);
896 if (magic
!= NBD_REQUEST_MAGIC
) {
897 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
903 static int nbd_send_reply(QIOChannel
*ioc
, NBDReply
*reply
, Error
**errp
)
905 uint8_t buf
[NBD_REPLY_SIZE
];
907 reply
->error
= system_errno_to_nbd_errno(reply
->error
);
909 trace_nbd_send_reply(reply
->error
, reply
->handle
);
912 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
913 [ 4 .. 7] error (0 == no error)
916 stl_be_p(buf
, NBD_REPLY_MAGIC
);
917 stl_be_p(buf
+ 4, reply
->error
);
918 stq_be_p(buf
+ 8, reply
->handle
);
920 return nbd_write(ioc
, buf
, sizeof(buf
), errp
);
923 #define MAX_NBD_REQUESTS 16
925 void nbd_client_get(NBDClient
*client
)
930 void nbd_client_put(NBDClient
*client
)
932 if (--client
->refcount
== 0) {
933 /* The last reference should be dropped by client->close,
934 * which is called by client_close.
936 assert(client
->closing
);
938 qio_channel_detach_aio_context(client
->ioc
);
939 object_unref(OBJECT(client
->sioc
));
940 object_unref(OBJECT(client
->ioc
));
941 if (client
->tlscreds
) {
942 object_unref(OBJECT(client
->tlscreds
));
944 g_free(client
->tlsaclname
);
946 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
947 nbd_export_put(client
->exp
);
953 static void client_close(NBDClient
*client
, bool negotiated
)
955 if (client
->closing
) {
959 client
->closing
= true;
961 /* Force requests to finish. They will drop their own references,
962 * then we'll close the socket and free the NBDClient.
964 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
967 /* Also tell the client, so that they release their reference. */
968 if (client
->close_fn
) {
969 client
->close_fn(client
, negotiated
);
973 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
977 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
978 client
->nb_requests
++;
980 req
= g_new0(NBDRequestData
, 1);
981 nbd_client_get(client
);
982 req
->client
= client
;
986 static void nbd_request_put(NBDRequestData
*req
)
988 NBDClient
*client
= req
->client
;
991 qemu_vfree(req
->data
);
995 client
->nb_requests
--;
996 nbd_client_receive_next_request(client
);
998 nbd_client_put(client
);
1001 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1003 NBDExport
*exp
= opaque
;
1006 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1010 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1011 qio_channel_attach_aio_context(client
->ioc
, ctx
);
1012 if (client
->recv_coroutine
) {
1013 aio_co_schedule(ctx
, client
->recv_coroutine
);
1015 if (client
->send_coroutine
) {
1016 aio_co_schedule(ctx
, client
->send_coroutine
);
1021 static void blk_aio_detach(void *opaque
)
1023 NBDExport
*exp
= opaque
;
1026 trace_nbd_blk_aio_detach(exp
->name
, exp
->ctx
);
1028 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1029 qio_channel_detach_aio_context(client
->ioc
);
1035 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1037 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1038 nbd_export_close(exp
);
1041 NBDExport
*nbd_export_new(BlockDriverState
*bs
, off_t dev_offset
, off_t size
,
1042 uint16_t nbdflags
, void (*close
)(NBDExport
*),
1043 bool writethrough
, BlockBackend
*on_eject_blk
,
1047 NBDExport
*exp
= g_malloc0(sizeof(NBDExport
));
1051 /* Don't allow resize while the NBD server is running, otherwise we don't
1052 * care what happens with the node. */
1053 perm
= BLK_PERM_CONSISTENT_READ
;
1054 if ((nbdflags
& NBD_FLAG_READ_ONLY
) == 0) {
1055 perm
|= BLK_PERM_WRITE
;
1057 blk
= blk_new(perm
, BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
|
1058 BLK_PERM_WRITE
| BLK_PERM_GRAPH_MOD
);
1059 ret
= blk_insert_bs(blk
, bs
, errp
);
1063 blk_set_enable_write_cache(blk
, !writethrough
);
1066 QTAILQ_INIT(&exp
->clients
);
1068 exp
->dev_offset
= dev_offset
;
1069 exp
->nbdflags
= nbdflags
;
1070 exp
->size
= size
< 0 ? blk_getlength(blk
) : size
;
1071 if (exp
->size
< 0) {
1072 error_setg_errno(errp
, -exp
->size
,
1073 "Failed to determine the NBD export's length");
1076 exp
->size
-= exp
->size
% BDRV_SECTOR_SIZE
;
1079 exp
->ctx
= blk_get_aio_context(blk
);
1080 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1083 blk_ref(on_eject_blk
);
1084 exp
->eject_notifier_blk
= on_eject_blk
;
1085 exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1086 blk_add_remove_bs_notifier(on_eject_blk
, &exp
->eject_notifier
);
1090 * NBD exports are used for non-shared storage migration. Make sure
1091 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1092 * access since the export could be available before migration handover.
1094 aio_context_acquire(exp
->ctx
);
1095 blk_invalidate_cache(blk
, NULL
);
1096 aio_context_release(exp
->ctx
);
1105 NBDExport
*nbd_export_find(const char *name
)
1108 QTAILQ_FOREACH(exp
, &exports
, next
) {
1109 if (strcmp(name
, exp
->name
) == 0) {
1117 void nbd_export_set_name(NBDExport
*exp
, const char *name
)
1119 if (exp
->name
== name
) {
1123 nbd_export_get(exp
);
1124 if (exp
->name
!= NULL
) {
1127 QTAILQ_REMOVE(&exports
, exp
, next
);
1128 nbd_export_put(exp
);
1131 nbd_export_get(exp
);
1132 exp
->name
= g_strdup(name
);
1133 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1135 nbd_export_put(exp
);
1138 void nbd_export_set_description(NBDExport
*exp
, const char *description
)
1140 g_free(exp
->description
);
1141 exp
->description
= g_strdup(description
);
1144 void nbd_export_close(NBDExport
*exp
)
1146 NBDClient
*client
, *next
;
1148 nbd_export_get(exp
);
1149 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1150 client_close(client
, true);
1152 nbd_export_set_name(exp
, NULL
);
1153 nbd_export_set_description(exp
, NULL
);
1154 nbd_export_put(exp
);
1157 void nbd_export_get(NBDExport
*exp
)
1159 assert(exp
->refcount
> 0);
1163 void nbd_export_put(NBDExport
*exp
)
1165 assert(exp
->refcount
> 0);
1166 if (exp
->refcount
== 1) {
1167 nbd_export_close(exp
);
1170 if (--exp
->refcount
== 0) {
1171 assert(exp
->name
== NULL
);
1172 assert(exp
->description
== NULL
);
1179 if (exp
->eject_notifier_blk
) {
1180 notifier_remove(&exp
->eject_notifier
);
1181 blk_unref(exp
->eject_notifier_blk
);
1183 blk_remove_aio_context_notifier(exp
->blk
, blk_aio_attached
,
1184 blk_aio_detach
, exp
);
1185 blk_unref(exp
->blk
);
1193 BlockBackend
*nbd_export_get_blockdev(NBDExport
*exp
)
1198 void nbd_export_close_all(void)
1200 NBDExport
*exp
, *next
;
1202 QTAILQ_FOREACH_SAFE(exp
, &exports
, next
, next
) {
1203 nbd_export_close(exp
);
1207 static int nbd_co_send_reply(NBDRequestData
*req
, NBDReply
*reply
, int len
,
1210 NBDClient
*client
= req
->client
;
1213 g_assert(qemu_in_coroutine());
1215 trace_nbd_co_send_reply(reply
->handle
, reply
->error
, len
);
1217 qemu_co_mutex_lock(&client
->send_lock
);
1218 client
->send_coroutine
= qemu_coroutine_self();
1221 ret
= nbd_send_reply(client
->ioc
, reply
, errp
);
1223 qio_channel_set_cork(client
->ioc
, true);
1224 ret
= nbd_send_reply(client
->ioc
, reply
, errp
);
1226 ret
= nbd_write(client
->ioc
, req
->data
, len
, errp
);
1231 qio_channel_set_cork(client
->ioc
, false);
1234 client
->send_coroutine
= NULL
;
1235 qemu_co_mutex_unlock(&client
->send_lock
);
1239 /* nbd_co_receive_request
1240 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1241 * connection right away, and any other negative value to report an error to
1242 * the client (although the caller may still need to disconnect after reporting
1245 static int nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
1248 NBDClient
*client
= req
->client
;
1250 g_assert(qemu_in_coroutine());
1251 assert(client
->recv_coroutine
== qemu_coroutine_self());
1252 if (nbd_receive_request(client
->ioc
, request
, errp
) < 0) {
1256 trace_nbd_co_receive_request_decode_type(request
->handle
, request
->type
,
1257 nbd_cmd_lookup(request
->type
));
1259 if (request
->type
!= NBD_CMD_WRITE
) {
1260 /* No payload, we are ready to read the next request. */
1261 req
->complete
= true;
1264 if (request
->type
== NBD_CMD_DISC
) {
1265 /* Special case: we're going to disconnect without a reply,
1266 * whether or not flags, from, or len are bogus */
1270 /* Check for sanity in the parameters, part 1. Defer as many
1271 * checks as possible until after reading any NBD_CMD_WRITE
1272 * payload, so we can try and keep the connection alive. */
1273 if ((request
->from
+ request
->len
) < request
->from
) {
1275 "integer overflow detected, you're probably being attacked");
1279 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
) {
1280 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
1281 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1282 request
->len
, NBD_MAX_BUFFER_SIZE
);
1286 req
->data
= blk_try_blockalign(client
->exp
->blk
, request
->len
);
1287 if (req
->data
== NULL
) {
1288 error_setg(errp
, "No memory");
1292 if (request
->type
== NBD_CMD_WRITE
) {
1293 if (nbd_read(client
->ioc
, req
->data
, request
->len
, errp
) < 0) {
1294 error_prepend(errp
, "reading from socket failed: ");
1297 req
->complete
= true;
1299 trace_nbd_co_receive_request_payload_received(request
->handle
,
1303 /* Sanity checks, part 2. */
1304 if (request
->from
+ request
->len
> client
->exp
->size
) {
1305 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
1306 ", Size: %" PRIu64
, request
->from
, request
->len
,
1307 (uint64_t)client
->exp
->size
);
1308 return request
->type
== NBD_CMD_WRITE
? -ENOSPC
: -EINVAL
;
1310 if (request
->flags
& ~(NBD_CMD_FLAG_FUA
| NBD_CMD_FLAG_NO_HOLE
)) {
1311 error_setg(errp
, "unsupported flags (got 0x%x)", request
->flags
);
1314 if (request
->type
!= NBD_CMD_WRITE_ZEROES
&&
1315 (request
->flags
& NBD_CMD_FLAG_NO_HOLE
)) {
1316 error_setg(errp
, "unexpected flags (got 0x%x)", request
->flags
);
1323 /* Owns a reference to the NBDClient passed as opaque. */
1324 static coroutine_fn
void nbd_trip(void *opaque
)
1326 NBDClient
*client
= opaque
;
1327 NBDExport
*exp
= client
->exp
;
1328 NBDRequestData
*req
;
1329 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
1333 int reply_data_len
= 0;
1334 Error
*local_err
= NULL
;
1337 if (client
->closing
) {
1338 nbd_client_put(client
);
1342 req
= nbd_request_get(client
);
1343 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
1344 client
->recv_coroutine
= NULL
;
1345 nbd_client_receive_next_request(client
);
1350 reply
.handle
= request
.handle
;
1358 if (client
->closing
) {
1360 * The client may be closed when we are blocked in
1361 * nbd_co_receive_request()
1366 switch (request
.type
) {
1368 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1369 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1370 ret
= blk_co_flush(exp
->blk
);
1372 error_setg_errno(&local_err
, -ret
, "flush failed");
1378 ret
= blk_pread(exp
->blk
, request
.from
+ exp
->dev_offset
,
1379 req
->data
, request
.len
);
1381 error_setg_errno(&local_err
, -ret
, "reading from file failed");
1386 reply_data_len
= request
.len
;
1390 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
1391 reply
.error
= EROFS
;
1396 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1397 flags
|= BDRV_REQ_FUA
;
1399 ret
= blk_pwrite(exp
->blk
, request
.from
+ exp
->dev_offset
,
1400 req
->data
, request
.len
, flags
);
1402 error_setg_errno(&local_err
, -ret
, "writing to file failed");
1407 case NBD_CMD_WRITE_ZEROES
:
1408 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
1409 error_setg(&local_err
, "Server is read-only, return error");
1410 reply
.error
= EROFS
;
1415 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1416 flags
|= BDRV_REQ_FUA
;
1418 if (!(request
.flags
& NBD_CMD_FLAG_NO_HOLE
)) {
1419 flags
|= BDRV_REQ_MAY_UNMAP
;
1421 ret
= blk_pwrite_zeroes(exp
->blk
, request
.from
+ exp
->dev_offset
,
1422 request
.len
, flags
);
1424 error_setg_errno(&local_err
, -ret
, "writing to file failed");
1430 /* unreachable, thanks to special case in nbd_co_receive_request() */
1434 ret
= blk_co_flush(exp
->blk
);
1436 error_setg_errno(&local_err
, -ret
, "flush failed");
1442 ret
= blk_co_pdiscard(exp
->blk
, request
.from
+ exp
->dev_offset
,
1445 error_setg_errno(&local_err
, -ret
, "discard failed");
1451 error_setg(&local_err
, "invalid request type (%" PRIu32
") received",
1453 reply
.error
= EINVAL
;
1458 /* If we are here local_err is not fatal error, already stored in
1460 error_report_err(local_err
);
1464 if (nbd_co_send_reply(req
, &reply
, reply_data_len
, &local_err
) < 0) {
1465 error_prepend(&local_err
, "Failed to send reply: ");
1469 /* We must disconnect after NBD_CMD_WRITE if we did not
1472 if (!req
->complete
) {
1473 error_setg(&local_err
, "Request handling failed in intermediate state");
1478 nbd_request_put(req
);
1479 nbd_client_put(client
);
1484 error_reportf_err(local_err
, "Disconnect client, due to: ");
1486 nbd_request_put(req
);
1487 client_close(client
, true);
1488 nbd_client_put(client
);
1491 static void nbd_client_receive_next_request(NBDClient
*client
)
1493 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
) {
1494 nbd_client_get(client
);
1495 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
1496 aio_co_schedule(client
->exp
->ctx
, client
->recv_coroutine
);
1500 static coroutine_fn
void nbd_co_client_start(void *opaque
)
1502 NBDClient
*client
= opaque
;
1503 NBDExport
*exp
= client
->exp
;
1504 Error
*local_err
= NULL
;
1507 nbd_export_get(exp
);
1508 QTAILQ_INSERT_TAIL(&exp
->clients
, client
, next
);
1510 qemu_co_mutex_init(&client
->send_lock
);
1512 if (nbd_negotiate(client
, &local_err
)) {
1514 error_report_err(local_err
);
1516 client_close(client
, false);
1520 nbd_client_receive_next_request(client
);
1524 * Create a new client listener on the given export @exp, using the
1525 * given channel @sioc. Begin servicing it in a coroutine. When the
1526 * connection closes, call @close_fn with an indication of whether the
1527 * client completed negotiation.
1529 void nbd_client_new(NBDExport
*exp
,
1530 QIOChannelSocket
*sioc
,
1531 QCryptoTLSCreds
*tlscreds
,
1532 const char *tlsaclname
,
1533 void (*close_fn
)(NBDClient
*, bool))
1538 client
= g_malloc0(sizeof(NBDClient
));
1539 client
->refcount
= 1;
1541 client
->tlscreds
= tlscreds
;
1543 object_ref(OBJECT(client
->tlscreds
));
1545 client
->tlsaclname
= g_strdup(tlsaclname
);
1546 client
->sioc
= sioc
;
1547 object_ref(OBJECT(client
->sioc
));
1548 client
->ioc
= QIO_CHANNEL(sioc
);
1549 object_ref(OBJECT(client
->ioc
));
1550 client
->close_fn
= close_fn
;
1552 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
1553 qemu_coroutine_enter(co
);