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
)
53 /* Definitions for opaque data types */
55 typedef struct NBDRequestData NBDRequestData
;
57 struct NBDRequestData
{
58 QSIMPLEQ_ENTRY(NBDRequestData
) entry
;
66 void (*close
)(NBDExport
*exp
);
74 QTAILQ_HEAD(, NBDClient
) clients
;
75 QTAILQ_ENTRY(NBDExport
) next
;
79 BlockBackend
*eject_notifier_blk
;
80 Notifier eject_notifier
;
83 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
87 void (*close_fn
)(NBDClient
*client
, bool negotiated
);
90 QCryptoTLSCreds
*tlscreds
;
92 QIOChannelSocket
*sioc
; /* The underlying data channel */
93 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
95 Coroutine
*recv_coroutine
;
98 Coroutine
*send_coroutine
;
100 QTAILQ_ENTRY(NBDClient
) next
;
104 bool structured_reply
;
107 /* That's all folks */
109 static void nbd_client_receive_next_request(NBDClient
*client
);
111 /* Basic flow for negotiation
138 /* Send a reply header, including length, but no payload.
139 * Return -errno on error, 0 on success. */
140 static int nbd_negotiate_send_rep_len(QIOChannel
*ioc
, uint32_t type
,
141 uint32_t opt
, uint32_t len
, Error
**errp
)
145 trace_nbd_negotiate_send_rep_len(opt
, nbd_opt_lookup(opt
),
146 type
, nbd_rep_lookup(type
), len
);
148 assert(len
< NBD_MAX_BUFFER_SIZE
);
149 magic
= cpu_to_be64(NBD_REP_MAGIC
);
150 if (nbd_write(ioc
, &magic
, sizeof(magic
), errp
) < 0) {
151 error_prepend(errp
, "write failed (rep magic): ");
155 opt
= cpu_to_be32(opt
);
156 if (nbd_write(ioc
, &opt
, sizeof(opt
), errp
) < 0) {
157 error_prepend(errp
, "write failed (rep opt): ");
161 type
= cpu_to_be32(type
);
162 if (nbd_write(ioc
, &type
, sizeof(type
), errp
) < 0) {
163 error_prepend(errp
, "write failed (rep type): ");
167 len
= cpu_to_be32(len
);
168 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
169 error_prepend(errp
, "write failed (rep data length): ");
175 /* Send a reply header with default 0 length.
176 * Return -errno on error, 0 on success. */
177 static int nbd_negotiate_send_rep(QIOChannel
*ioc
, uint32_t type
, uint32_t opt
,
180 return nbd_negotiate_send_rep_len(ioc
, type
, opt
, 0, errp
);
183 /* Send an error reply.
184 * Return -errno on error, 0 on success. */
185 static int GCC_FMT_ATTR(5, 6)
186 nbd_negotiate_send_rep_err(QIOChannel
*ioc
, uint32_t type
,
187 uint32_t opt
, Error
**errp
, const char *fmt
, ...)
195 msg
= g_strdup_vprintf(fmt
, va
);
199 trace_nbd_negotiate_send_rep_err(msg
);
200 ret
= nbd_negotiate_send_rep_len(ioc
, type
, opt
, len
, errp
);
204 if (nbd_write(ioc
, msg
, len
, errp
) < 0) {
205 error_prepend(errp
, "write failed (error message): ");
216 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
217 * Return -errno on error, 0 on success. */
218 static int nbd_negotiate_send_rep_list(QIOChannel
*ioc
, NBDExport
*exp
,
221 size_t name_len
, desc_len
;
223 const char *name
= exp
->name
? exp
->name
: "";
224 const char *desc
= exp
->description
? exp
->description
: "";
227 trace_nbd_negotiate_send_rep_list(name
, desc
);
228 name_len
= strlen(name
);
229 desc_len
= strlen(desc
);
230 len
= name_len
+ desc_len
+ sizeof(len
);
231 ret
= nbd_negotiate_send_rep_len(ioc
, NBD_REP_SERVER
, NBD_OPT_LIST
, len
,
237 len
= cpu_to_be32(name_len
);
238 if (nbd_write(ioc
, &len
, sizeof(len
), errp
) < 0) {
239 error_prepend(errp
, "write failed (name length): ");
243 if (nbd_write(ioc
, name
, name_len
, errp
) < 0) {
244 error_prepend(errp
, "write failed (name buffer): ");
248 if (nbd_write(ioc
, desc
, desc_len
, errp
) < 0) {
249 error_prepend(errp
, "write failed (description buffer): ");
256 /* Process the NBD_OPT_LIST command, with a potential series of replies.
257 * Return -errno on error, 0 on success. */
258 static int nbd_negotiate_handle_list(NBDClient
*client
, Error
**errp
)
262 /* For each export, send a NBD_REP_SERVER reply. */
263 QTAILQ_FOREACH(exp
, &exports
, next
) {
264 if (nbd_negotiate_send_rep_list(client
->ioc
, exp
, errp
)) {
268 /* Finish with a NBD_REP_ACK. */
269 return nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, NBD_OPT_LIST
, errp
);
272 /* Send a reply to NBD_OPT_EXPORT_NAME.
273 * Return -errno on error, 0 on success. */
274 static int nbd_negotiate_handle_export_name(NBDClient
*client
, uint32_t length
,
275 uint16_t myflags
, bool no_zeroes
,
278 char name
[NBD_MAX_NAME_SIZE
+ 1];
279 char buf
[NBD_REPLY_EXPORT_NAME_SIZE
] = "";
284 [20 .. xx] export name (length bytes)
287 [ 8 .. 9] export flags
288 [10 .. 133] reserved (0) [unless no_zeroes]
290 trace_nbd_negotiate_handle_export_name();
291 if (length
>= sizeof(name
)) {
292 error_setg(errp
, "Bad length received");
295 if (nbd_read(client
->ioc
, name
, length
, errp
) < 0) {
296 error_prepend(errp
, "read failed: ");
301 trace_nbd_negotiate_handle_export_name_request(name
);
303 client
->exp
= nbd_export_find(name
);
305 error_setg(errp
, "export not found");
309 trace_nbd_negotiate_new_style_size_flags(client
->exp
->size
,
310 client
->exp
->nbdflags
| myflags
);
311 stq_be_p(buf
, client
->exp
->size
);
312 stw_be_p(buf
+ 8, client
->exp
->nbdflags
| myflags
);
313 len
= no_zeroes
? 10 : sizeof(buf
);
314 ret
= nbd_write(client
->ioc
, buf
, len
, errp
);
316 error_prepend(errp
, "write failed: ");
320 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
321 nbd_export_get(client
->exp
);
326 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
327 * The buffer does NOT include the info type prefix.
328 * Return -errno on error, 0 if ready to send more. */
329 static int nbd_negotiate_send_info(NBDClient
*client
, uint32_t opt
,
330 uint16_t info
, uint32_t length
, void *buf
,
335 trace_nbd_negotiate_send_info(info
, nbd_info_lookup(info
), length
);
336 rc
= nbd_negotiate_send_rep_len(client
->ioc
, NBD_REP_INFO
, opt
,
337 sizeof(info
) + length
, errp
);
342 if (nbd_write(client
->ioc
, &info
, sizeof(info
), errp
) < 0) {
345 if (nbd_write(client
->ioc
, buf
, length
, errp
) < 0) {
351 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
352 * Return -errno on error, 0 if ready for next option, and 1 to move
353 * into transmission phase. */
354 static int nbd_negotiate_handle_info(NBDClient
*client
, uint32_t length
,
355 uint32_t opt
, uint16_t myflags
,
359 char name
[NBD_MAX_NAME_SIZE
+ 1];
364 bool sendname
= false;
365 bool blocksize
= false;
367 char buf
[sizeof(uint64_t) + sizeof(uint16_t)];
371 4 bytes: L, name length (can be 0)
373 2 bytes: N, number of requests (can be 0)
374 N * 2 bytes: N requests
376 if (length
< sizeof(namelen
) + sizeof(requests
)) {
377 msg
= "overall request too short";
380 if (nbd_read(client
->ioc
, &namelen
, sizeof(namelen
), errp
) < 0) {
383 be32_to_cpus(&namelen
);
384 length
-= sizeof(namelen
);
385 if (namelen
> length
- sizeof(requests
) || (length
- namelen
) % 2) {
386 msg
= "name length is incorrect";
389 if (nbd_read(client
->ioc
, name
, namelen
, errp
) < 0) {
392 name
[namelen
] = '\0';
394 trace_nbd_negotiate_handle_export_name_request(name
);
396 if (nbd_read(client
->ioc
, &requests
, sizeof(requests
), errp
) < 0) {
399 be16_to_cpus(&requests
);
400 length
-= sizeof(requests
);
401 trace_nbd_negotiate_handle_info_requests(requests
);
402 if (requests
!= length
/ sizeof(request
)) {
403 msg
= "incorrect number of requests for overall length";
407 if (nbd_read(client
->ioc
, &request
, sizeof(request
), errp
) < 0) {
410 be16_to_cpus(&request
);
411 length
-= sizeof(request
);
412 trace_nbd_negotiate_handle_info_request(request
,
413 nbd_info_lookup(request
));
414 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
415 * everything else is either a request we don't know or
416 * something we send regardless of request */
421 case NBD_INFO_BLOCK_SIZE
:
427 exp
= nbd_export_find(name
);
429 return nbd_negotiate_send_rep_err(client
->ioc
, NBD_REP_ERR_UNKNOWN
,
430 opt
, errp
, "export '%s' not present",
434 /* Don't bother sending NBD_INFO_NAME unless client requested it */
436 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_NAME
, length
, name
,
443 /* Send NBD_INFO_DESCRIPTION only if available, regardless of
445 if (exp
->description
) {
446 size_t len
= strlen(exp
->description
);
448 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_DESCRIPTION
,
449 len
, exp
->description
, errp
);
455 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
456 * according to whether the client requested it, and according to
457 * whether this is OPT_INFO or OPT_GO. */
458 /* minimum - 1 for back-compat, or 512 if client is new enough.
459 * TODO: consult blk_bs(blk)->bl.request_alignment? */
460 sizes
[0] = (opt
== NBD_OPT_INFO
|| blocksize
) ? BDRV_SECTOR_SIZE
: 1;
461 /* preferred - Hard-code to 4096 for now.
462 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
464 /* maximum - At most 32M, but smaller as appropriate. */
465 sizes
[2] = MIN(blk_get_max_transfer(exp
->blk
), NBD_MAX_BUFFER_SIZE
);
466 trace_nbd_negotiate_handle_info_block_size(sizes
[0], sizes
[1], sizes
[2]);
467 cpu_to_be32s(&sizes
[0]);
468 cpu_to_be32s(&sizes
[1]);
469 cpu_to_be32s(&sizes
[2]);
470 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_BLOCK_SIZE
,
471 sizeof(sizes
), sizes
, errp
);
476 /* Send NBD_INFO_EXPORT always */
477 trace_nbd_negotiate_new_style_size_flags(exp
->size
,
478 exp
->nbdflags
| myflags
);
479 stq_be_p(buf
, exp
->size
);
480 stw_be_p(buf
+ 8, exp
->nbdflags
| myflags
);
481 rc
= nbd_negotiate_send_info(client
, opt
, NBD_INFO_EXPORT
,
482 sizeof(buf
), buf
, errp
);
487 /* If the client is just asking for NBD_OPT_INFO, but forgot to
488 * request block sizes, return an error.
489 * TODO: consult blk_bs(blk)->request_align, and only error if it
491 if (opt
== NBD_OPT_INFO
&& !blocksize
) {
492 return nbd_negotiate_send_rep_err(client
->ioc
,
493 NBD_REP_ERR_BLOCK_SIZE_REQD
, opt
,
495 "request NBD_INFO_BLOCK_SIZE to "
500 rc
= nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, opt
, errp
);
505 if (opt
== NBD_OPT_GO
) {
507 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
508 nbd_export_get(client
->exp
);
514 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
517 return nbd_negotiate_send_rep_err(client
->ioc
, NBD_REP_ERR_INVALID
, opt
,
522 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
523 * new channel for all further (now-encrypted) communication. */
524 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
529 struct NBDTLSHandshakeData data
= { 0 };
531 trace_nbd_negotiate_handle_starttls();
534 if (nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
,
535 NBD_OPT_STARTTLS
, errp
) < 0) {
539 tioc
= qio_channel_tls_new_server(ioc
,
547 qio_channel_set_name(QIO_CHANNEL(tioc
), "nbd-server-tls");
548 trace_nbd_negotiate_handle_starttls_handshake();
549 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
550 qio_channel_tls_handshake(tioc
,
555 if (!data
.complete
) {
556 g_main_loop_run(data
.loop
);
558 g_main_loop_unref(data
.loop
);
560 object_unref(OBJECT(tioc
));
561 error_propagate(errp
, data
.error
);
565 return QIO_CHANNEL(tioc
);
568 /* nbd_reject_length: Handle any unexpected payload.
569 * @fatal requests that we quit talking to the client, even if we are able
570 * to successfully send an error to the guest.
572 * -errno transmission error occurred or @fatal was requested, errp is set
573 * 0 error message successfully sent to client, errp is not set
575 static int nbd_reject_length(NBDClient
*client
, uint32_t length
,
576 uint32_t option
, bool fatal
, Error
**errp
)
581 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
584 ret
= nbd_negotiate_send_rep_err(client
->ioc
, NBD_REP_ERR_INVALID
,
586 "option '%s' should have zero length",
587 nbd_opt_lookup(option
));
589 error_setg(errp
, "option '%s' should have zero length",
590 nbd_opt_lookup(option
));
596 /* nbd_negotiate_options
597 * Process all NBD_OPT_* client option commands, during fixed newstyle
600 * -errno on error, errp is set
601 * 0 on successful negotiation, errp is not set
602 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
605 static int nbd_negotiate_options(NBDClient
*client
, uint16_t myflags
,
609 bool fixedNewstyle
= false;
610 bool no_zeroes
= false;
613 [ 0 .. 3] client flags
615 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
616 [ 0 .. 7] NBD_OPTS_MAGIC
617 [ 8 .. 11] NBD option
618 [12 .. 15] Data length
621 [ 0 .. 7] NBD_OPTS_MAGIC
622 [ 8 .. 11] Second NBD option
623 [12 .. 15] Data length
627 if (nbd_read(client
->ioc
, &flags
, sizeof(flags
), errp
) < 0) {
628 error_prepend(errp
, "read failed: ");
631 be32_to_cpus(&flags
);
632 trace_nbd_negotiate_options_flags(flags
);
633 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
634 fixedNewstyle
= true;
635 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
637 if (flags
& NBD_FLAG_C_NO_ZEROES
) {
639 flags
&= ~NBD_FLAG_C_NO_ZEROES
;
642 error_setg(errp
, "Unknown client flags 0x%" PRIx32
" received", flags
);
648 uint32_t option
, length
;
651 if (nbd_read(client
->ioc
, &magic
, sizeof(magic
), errp
) < 0) {
652 error_prepend(errp
, "read failed: ");
655 magic
= be64_to_cpu(magic
);
656 trace_nbd_negotiate_options_check_magic(magic
);
657 if (magic
!= NBD_OPTS_MAGIC
) {
658 error_setg(errp
, "Bad magic received");
662 if (nbd_read(client
->ioc
, &option
,
663 sizeof(option
), errp
) < 0) {
664 error_prepend(errp
, "read failed: ");
667 option
= be32_to_cpu(option
);
669 if (nbd_read(client
->ioc
, &length
, sizeof(length
), errp
) < 0) {
670 error_prepend(errp
, "read failed: ");
673 length
= be32_to_cpu(length
);
675 trace_nbd_negotiate_options_check_option(option
,
676 nbd_opt_lookup(option
));
677 if (client
->tlscreds
&&
678 client
->ioc
== (QIOChannel
*)client
->sioc
) {
680 if (!fixedNewstyle
) {
681 error_setg(errp
, "Unsupported option 0x%" PRIx32
, option
);
685 case NBD_OPT_STARTTLS
:
687 /* Unconditionally drop the connection if the client
688 * can't start a TLS negotiation correctly */
689 return nbd_reject_length(client
, length
, option
, true,
692 tioc
= nbd_negotiate_handle_starttls(client
, errp
);
697 object_unref(OBJECT(client
->ioc
));
698 client
->ioc
= QIO_CHANNEL(tioc
);
701 case NBD_OPT_EXPORT_NAME
:
702 /* No way to return an error to client, so drop connection */
703 error_setg(errp
, "Option 0x%x not permitted before TLS",
708 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
711 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
712 NBD_REP_ERR_TLS_REQD
,
715 "not permitted before TLS",
717 /* Let the client keep trying, unless they asked to
718 * quit. In this mode, we've already sent an error, so
719 * we can't ack the abort. */
720 if (option
== NBD_OPT_ABORT
) {
725 } else if (fixedNewstyle
) {
729 ret
= nbd_reject_length(client
, length
, option
, false,
732 ret
= nbd_negotiate_handle_list(client
, errp
);
737 /* NBD spec says we must try to reply before
738 * disconnecting, but that we must also tolerate
739 * guests that don't wait for our reply. */
740 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, option
, NULL
);
743 case NBD_OPT_EXPORT_NAME
:
744 return nbd_negotiate_handle_export_name(client
, length
,
750 ret
= nbd_negotiate_handle_info(client
, length
, option
,
753 assert(option
== NBD_OPT_GO
);
758 case NBD_OPT_STARTTLS
:
760 ret
= nbd_reject_length(client
, length
, option
, false,
762 } else if (client
->tlscreds
) {
763 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
766 "TLS already enabled");
768 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
771 "TLS not configured");
775 case NBD_OPT_STRUCTURED_REPLY
:
777 ret
= nbd_reject_length(client
, length
, option
, false,
779 } else if (client
->structured_reply
) {
780 ret
= nbd_negotiate_send_rep_err(
781 client
->ioc
, NBD_REP_ERR_INVALID
, option
, errp
,
782 "structured reply already negotiated");
784 ret
= nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
,
786 client
->structured_reply
= true;
787 myflags
|= NBD_FLAG_SEND_DF
;
792 if (nbd_drop(client
->ioc
, length
, errp
) < 0) {
795 ret
= nbd_negotiate_send_rep_err(client
->ioc
,
798 "Unsupported option 0x%"
799 PRIx32
" (%s)", option
,
800 nbd_opt_lookup(option
));
805 * If broken new-style we should drop the connection
806 * for anything except NBD_OPT_EXPORT_NAME
809 case NBD_OPT_EXPORT_NAME
:
810 return nbd_negotiate_handle_export_name(client
, length
,
815 error_setg(errp
, "Unsupported option 0x%" PRIx32
" (%s)",
816 option
, nbd_opt_lookup(option
));
828 * -errno on error, errp is set
829 * 0 on successful negotiation, errp is not set
830 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
833 static coroutine_fn
int nbd_negotiate(NBDClient
*client
, Error
**errp
)
835 char buf
[NBD_OLDSTYLE_NEGOTIATE_SIZE
] = "";
837 const uint16_t myflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
838 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
|
839 NBD_FLAG_SEND_WRITE_ZEROES
);
842 /* Old style negotiation header, no room for options
843 [ 0 .. 7] passwd ("NBDMAGIC")
844 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
846 [24 .. 27] export flags (zero-extended)
847 [28 .. 151] reserved (0)
849 New style negotiation header, client can send options
850 [ 0 .. 7] passwd ("NBDMAGIC")
851 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
852 [16 .. 17] server flags (0)
853 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
856 qio_channel_set_blocking(client
->ioc
, false, NULL
);
858 trace_nbd_negotiate_begin();
859 memcpy(buf
, "NBDMAGIC", 8);
861 oldStyle
= client
->exp
!= NULL
&& !client
->tlscreds
;
863 trace_nbd_negotiate_old_style(client
->exp
->size
,
864 client
->exp
->nbdflags
| myflags
);
865 stq_be_p(buf
+ 8, NBD_CLIENT_MAGIC
);
866 stq_be_p(buf
+ 16, client
->exp
->size
);
867 stl_be_p(buf
+ 24, client
->exp
->nbdflags
| myflags
);
869 if (nbd_write(client
->ioc
, buf
, sizeof(buf
), errp
) < 0) {
870 error_prepend(errp
, "write failed: ");
874 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
875 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
| NBD_FLAG_NO_ZEROES
);
877 if (nbd_write(client
->ioc
, buf
, 18, errp
) < 0) {
878 error_prepend(errp
, "write failed: ");
881 ret
= nbd_negotiate_options(client
, myflags
, errp
);
884 error_prepend(errp
, "option negotiation failed: ");
890 trace_nbd_negotiate_success();
895 static int nbd_receive_request(QIOChannel
*ioc
, NBDRequest
*request
,
898 uint8_t buf
[NBD_REQUEST_SIZE
];
902 ret
= nbd_read(ioc
, buf
, sizeof(buf
), errp
);
908 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
909 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...)
910 [ 6 .. 7] type (NBD_CMD_READ, ...)
916 magic
= ldl_be_p(buf
);
917 request
->flags
= lduw_be_p(buf
+ 4);
918 request
->type
= lduw_be_p(buf
+ 6);
919 request
->handle
= ldq_be_p(buf
+ 8);
920 request
->from
= ldq_be_p(buf
+ 16);
921 request
->len
= ldl_be_p(buf
+ 24);
923 trace_nbd_receive_request(magic
, request
->flags
, request
->type
,
924 request
->from
, request
->len
);
926 if (magic
!= NBD_REQUEST_MAGIC
) {
927 error_setg(errp
, "invalid magic (got 0x%" PRIx32
")", magic
);
933 #define MAX_NBD_REQUESTS 16
935 void nbd_client_get(NBDClient
*client
)
940 void nbd_client_put(NBDClient
*client
)
942 if (--client
->refcount
== 0) {
943 /* The last reference should be dropped by client->close,
944 * which is called by client_close.
946 assert(client
->closing
);
948 qio_channel_detach_aio_context(client
->ioc
);
949 object_unref(OBJECT(client
->sioc
));
950 object_unref(OBJECT(client
->ioc
));
951 if (client
->tlscreds
) {
952 object_unref(OBJECT(client
->tlscreds
));
954 g_free(client
->tlsaclname
);
956 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
957 nbd_export_put(client
->exp
);
963 static void client_close(NBDClient
*client
, bool negotiated
)
965 if (client
->closing
) {
969 client
->closing
= true;
971 /* Force requests to finish. They will drop their own references,
972 * then we'll close the socket and free the NBDClient.
974 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
977 /* Also tell the client, so that they release their reference. */
978 if (client
->close_fn
) {
979 client
->close_fn(client
, negotiated
);
983 static NBDRequestData
*nbd_request_get(NBDClient
*client
)
987 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
988 client
->nb_requests
++;
990 req
= g_new0(NBDRequestData
, 1);
991 nbd_client_get(client
);
992 req
->client
= client
;
996 static void nbd_request_put(NBDRequestData
*req
)
998 NBDClient
*client
= req
->client
;
1001 qemu_vfree(req
->data
);
1005 client
->nb_requests
--;
1006 nbd_client_receive_next_request(client
);
1008 nbd_client_put(client
);
1011 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
1013 NBDExport
*exp
= opaque
;
1016 trace_nbd_blk_aio_attached(exp
->name
, ctx
);
1020 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1021 qio_channel_attach_aio_context(client
->ioc
, ctx
);
1022 if (client
->recv_coroutine
) {
1023 aio_co_schedule(ctx
, client
->recv_coroutine
);
1025 if (client
->send_coroutine
) {
1026 aio_co_schedule(ctx
, client
->send_coroutine
);
1031 static void blk_aio_detach(void *opaque
)
1033 NBDExport
*exp
= opaque
;
1036 trace_nbd_blk_aio_detach(exp
->name
, exp
->ctx
);
1038 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
1039 qio_channel_detach_aio_context(client
->ioc
);
1045 static void nbd_eject_notifier(Notifier
*n
, void *data
)
1047 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
1048 nbd_export_close(exp
);
1051 NBDExport
*nbd_export_new(BlockDriverState
*bs
, off_t dev_offset
, off_t size
,
1052 uint16_t nbdflags
, void (*close
)(NBDExport
*),
1053 bool writethrough
, BlockBackend
*on_eject_blk
,
1058 NBDExport
*exp
= g_new0(NBDExport
, 1);
1063 * NBD exports are used for non-shared storage migration. Make sure
1064 * that BDRV_O_INACTIVE is cleared and the image is ready for write
1065 * access since the export could be available before migration handover.
1067 ctx
= bdrv_get_aio_context(bs
);
1068 aio_context_acquire(ctx
);
1069 bdrv_invalidate_cache(bs
, NULL
);
1070 aio_context_release(ctx
);
1072 /* Don't allow resize while the NBD server is running, otherwise we don't
1073 * care what happens with the node. */
1074 perm
= BLK_PERM_CONSISTENT_READ
;
1075 if ((nbdflags
& NBD_FLAG_READ_ONLY
) == 0) {
1076 perm
|= BLK_PERM_WRITE
;
1078 blk
= blk_new(perm
, BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
|
1079 BLK_PERM_WRITE
| BLK_PERM_GRAPH_MOD
);
1080 ret
= blk_insert_bs(blk
, bs
, errp
);
1084 blk_set_enable_write_cache(blk
, !writethrough
);
1087 QTAILQ_INIT(&exp
->clients
);
1089 exp
->dev_offset
= dev_offset
;
1090 exp
->nbdflags
= nbdflags
;
1091 exp
->size
= size
< 0 ? blk_getlength(blk
) : size
;
1092 if (exp
->size
< 0) {
1093 error_setg_errno(errp
, -exp
->size
,
1094 "Failed to determine the NBD export's length");
1097 exp
->size
-= exp
->size
% BDRV_SECTOR_SIZE
;
1100 exp
->ctx
= blk_get_aio_context(blk
);
1101 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
1104 blk_ref(on_eject_blk
);
1105 exp
->eject_notifier_blk
= on_eject_blk
;
1106 exp
->eject_notifier
.notify
= nbd_eject_notifier
;
1107 blk_add_remove_bs_notifier(on_eject_blk
, &exp
->eject_notifier
);
1117 NBDExport
*nbd_export_find(const char *name
)
1120 QTAILQ_FOREACH(exp
, &exports
, next
) {
1121 if (strcmp(name
, exp
->name
) == 0) {
1129 void nbd_export_set_name(NBDExport
*exp
, const char *name
)
1131 if (exp
->name
== name
) {
1135 nbd_export_get(exp
);
1136 if (exp
->name
!= NULL
) {
1139 QTAILQ_REMOVE(&exports
, exp
, next
);
1140 nbd_export_put(exp
);
1143 nbd_export_get(exp
);
1144 exp
->name
= g_strdup(name
);
1145 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
1147 nbd_export_put(exp
);
1150 void nbd_export_set_description(NBDExport
*exp
, const char *description
)
1152 g_free(exp
->description
);
1153 exp
->description
= g_strdup(description
);
1156 void nbd_export_close(NBDExport
*exp
)
1158 NBDClient
*client
, *next
;
1160 nbd_export_get(exp
);
1161 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
1162 client_close(client
, true);
1164 nbd_export_set_name(exp
, NULL
);
1165 nbd_export_set_description(exp
, NULL
);
1166 nbd_export_put(exp
);
1169 void nbd_export_get(NBDExport
*exp
)
1171 assert(exp
->refcount
> 0);
1175 void nbd_export_put(NBDExport
*exp
)
1177 assert(exp
->refcount
> 0);
1178 if (exp
->refcount
== 1) {
1179 nbd_export_close(exp
);
1182 if (--exp
->refcount
== 0) {
1183 assert(exp
->name
== NULL
);
1184 assert(exp
->description
== NULL
);
1191 if (exp
->eject_notifier_blk
) {
1192 notifier_remove(&exp
->eject_notifier
);
1193 blk_unref(exp
->eject_notifier_blk
);
1195 blk_remove_aio_context_notifier(exp
->blk
, blk_aio_attached
,
1196 blk_aio_detach
, exp
);
1197 blk_unref(exp
->blk
);
1205 BlockBackend
*nbd_export_get_blockdev(NBDExport
*exp
)
1210 void nbd_export_close_all(void)
1212 NBDExport
*exp
, *next
;
1214 QTAILQ_FOREACH_SAFE(exp
, &exports
, next
, next
) {
1215 nbd_export_close(exp
);
1219 static int coroutine_fn
nbd_co_send_iov(NBDClient
*client
, struct iovec
*iov
,
1220 unsigned niov
, Error
**errp
)
1224 g_assert(qemu_in_coroutine());
1225 qemu_co_mutex_lock(&client
->send_lock
);
1226 client
->send_coroutine
= qemu_coroutine_self();
1228 ret
= qio_channel_writev_all(client
->ioc
, iov
, niov
, errp
) < 0 ? -EIO
: 0;
1230 client
->send_coroutine
= NULL
;
1231 qemu_co_mutex_unlock(&client
->send_lock
);
1236 static inline void set_be_simple_reply(NBDSimpleReply
*reply
, uint64_t error
,
1239 stl_be_p(&reply
->magic
, NBD_SIMPLE_REPLY_MAGIC
);
1240 stl_be_p(&reply
->error
, error
);
1241 stq_be_p(&reply
->handle
, handle
);
1244 static int nbd_co_send_simple_reply(NBDClient
*client
,
1251 NBDSimpleReply reply
;
1252 int nbd_err
= system_errno_to_nbd_errno(error
);
1253 struct iovec iov
[] = {
1254 {.iov_base
= &reply
, .iov_len
= sizeof(reply
)},
1255 {.iov_base
= data
, .iov_len
= len
}
1258 trace_nbd_co_send_simple_reply(handle
, nbd_err
, nbd_err_lookup(nbd_err
),
1260 set_be_simple_reply(&reply
, nbd_err
, handle
);
1262 return nbd_co_send_iov(client
, iov
, len
? 2 : 1, errp
);
1265 static inline void set_be_chunk(NBDStructuredReplyChunk
*chunk
, uint16_t flags
,
1266 uint16_t type
, uint64_t handle
, uint32_t length
)
1268 stl_be_p(&chunk
->magic
, NBD_STRUCTURED_REPLY_MAGIC
);
1269 stw_be_p(&chunk
->flags
, flags
);
1270 stw_be_p(&chunk
->type
, type
);
1271 stq_be_p(&chunk
->handle
, handle
);
1272 stl_be_p(&chunk
->length
, length
);
1275 static int coroutine_fn
nbd_co_send_structured_read(NBDClient
*client
,
1282 NBDStructuredRead chunk
;
1283 struct iovec iov
[] = {
1284 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1285 {.iov_base
= data
, .iov_len
= size
}
1288 trace_nbd_co_send_structured_read(handle
, offset
, data
, size
);
1289 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_OFFSET_DATA
,
1290 handle
, sizeof(chunk
) - sizeof(chunk
.h
) + size
);
1291 stq_be_p(&chunk
.offset
, offset
);
1293 return nbd_co_send_iov(client
, iov
, 2, errp
);
1296 static int coroutine_fn
nbd_co_send_structured_error(NBDClient
*client
,
1302 NBDStructuredError chunk
;
1303 int nbd_err
= system_errno_to_nbd_errno(error
);
1304 struct iovec iov
[] = {
1305 {.iov_base
= &chunk
, .iov_len
= sizeof(chunk
)},
1306 {.iov_base
= (char *)msg
, .iov_len
= msg
? strlen(msg
) : 0},
1310 trace_nbd_co_send_structured_error(handle
, nbd_err
,
1311 nbd_err_lookup(nbd_err
), msg
? msg
: "");
1312 set_be_chunk(&chunk
.h
, NBD_REPLY_FLAG_DONE
, NBD_REPLY_TYPE_ERROR
, handle
,
1313 sizeof(chunk
) - sizeof(chunk
.h
) + iov
[1].iov_len
);
1314 stl_be_p(&chunk
.error
, nbd_err
);
1315 stw_be_p(&chunk
.message_length
, iov
[1].iov_len
);
1317 return nbd_co_send_iov(client
, iov
, 1 + !!iov
[1].iov_len
, errp
);
1320 /* nbd_co_receive_request
1321 * Collect a client request. Return 0 if request looks valid, -EIO to drop
1322 * connection right away, and any other negative value to report an error to
1323 * the client (although the caller may still need to disconnect after reporting
1326 static int nbd_co_receive_request(NBDRequestData
*req
, NBDRequest
*request
,
1329 NBDClient
*client
= req
->client
;
1332 g_assert(qemu_in_coroutine());
1333 assert(client
->recv_coroutine
== qemu_coroutine_self());
1334 if (nbd_receive_request(client
->ioc
, request
, errp
) < 0) {
1338 trace_nbd_co_receive_request_decode_type(request
->handle
, request
->type
,
1339 nbd_cmd_lookup(request
->type
));
1341 if (request
->type
!= NBD_CMD_WRITE
) {
1342 /* No payload, we are ready to read the next request. */
1343 req
->complete
= true;
1346 if (request
->type
== NBD_CMD_DISC
) {
1347 /* Special case: we're going to disconnect without a reply,
1348 * whether or not flags, from, or len are bogus */
1352 /* Check for sanity in the parameters, part 1. Defer as many
1353 * checks as possible until after reading any NBD_CMD_WRITE
1354 * payload, so we can try and keep the connection alive. */
1355 if ((request
->from
+ request
->len
) < request
->from
) {
1357 "integer overflow detected, you're probably being attacked");
1361 if (request
->type
== NBD_CMD_READ
|| request
->type
== NBD_CMD_WRITE
) {
1362 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
1363 error_setg(errp
, "len (%" PRIu32
" ) is larger than max len (%u)",
1364 request
->len
, NBD_MAX_BUFFER_SIZE
);
1368 req
->data
= blk_try_blockalign(client
->exp
->blk
, request
->len
);
1369 if (req
->data
== NULL
) {
1370 error_setg(errp
, "No memory");
1374 if (request
->type
== NBD_CMD_WRITE
) {
1375 if (nbd_read(client
->ioc
, req
->data
, request
->len
, errp
) < 0) {
1376 error_prepend(errp
, "reading from socket failed: ");
1379 req
->complete
= true;
1381 trace_nbd_co_receive_request_payload_received(request
->handle
,
1385 /* Sanity checks, part 2. */
1386 if (request
->from
+ request
->len
> client
->exp
->size
) {
1387 error_setg(errp
, "operation past EOF; From: %" PRIu64
", Len: %" PRIu32
1388 ", Size: %" PRIu64
, request
->from
, request
->len
,
1389 (uint64_t)client
->exp
->size
);
1390 return request
->type
== NBD_CMD_WRITE
? -ENOSPC
: -EINVAL
;
1392 valid_flags
= NBD_CMD_FLAG_FUA
;
1393 if (request
->type
== NBD_CMD_READ
&& client
->structured_reply
) {
1394 valid_flags
|= NBD_CMD_FLAG_DF
;
1395 } else if (request
->type
== NBD_CMD_WRITE_ZEROES
) {
1396 valid_flags
|= NBD_CMD_FLAG_NO_HOLE
;
1398 if (request
->flags
& ~valid_flags
) {
1399 error_setg(errp
, "unsupported flags for command %s (got 0x%x)",
1400 nbd_cmd_lookup(request
->type
), request
->flags
);
1407 /* Owns a reference to the NBDClient passed as opaque. */
1408 static coroutine_fn
void nbd_trip(void *opaque
)
1410 NBDClient
*client
= opaque
;
1411 NBDExport
*exp
= client
->exp
;
1412 NBDRequestData
*req
;
1413 NBDRequest request
= { 0 }; /* GCC thinks it can be used uninitialized */
1416 int reply_data_len
= 0;
1417 Error
*local_err
= NULL
;
1421 if (client
->closing
) {
1422 nbd_client_put(client
);
1426 req
= nbd_request_get(client
);
1427 ret
= nbd_co_receive_request(req
, &request
, &local_err
);
1428 client
->recv_coroutine
= NULL
;
1429 nbd_client_receive_next_request(client
);
1438 if (client
->closing
) {
1440 * The client may be closed when we are blocked in
1441 * nbd_co_receive_request()
1446 switch (request
.type
) {
1448 /* XXX: NBD Protocol only documents use of FUA with WRITE */
1449 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1450 ret
= blk_co_flush(exp
->blk
);
1452 error_setg_errno(&local_err
, -ret
, "flush failed");
1457 ret
= blk_pread(exp
->blk
, request
.from
+ exp
->dev_offset
,
1458 req
->data
, request
.len
);
1460 error_setg_errno(&local_err
, -ret
, "reading from file failed");
1464 reply_data_len
= request
.len
;
1468 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
1469 error_setg(&local_err
, "Export is read-only");
1475 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1476 flags
|= BDRV_REQ_FUA
;
1478 ret
= blk_pwrite(exp
->blk
, request
.from
+ exp
->dev_offset
,
1479 req
->data
, request
.len
, flags
);
1481 error_setg_errno(&local_err
, -ret
, "writing to file failed");
1485 case NBD_CMD_WRITE_ZEROES
:
1486 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
1487 error_setg(&local_err
, "Export is read-only");
1493 if (request
.flags
& NBD_CMD_FLAG_FUA
) {
1494 flags
|= BDRV_REQ_FUA
;
1496 if (!(request
.flags
& NBD_CMD_FLAG_NO_HOLE
)) {
1497 flags
|= BDRV_REQ_MAY_UNMAP
;
1499 ret
= blk_pwrite_zeroes(exp
->blk
, request
.from
+ exp
->dev_offset
,
1500 request
.len
, flags
);
1502 error_setg_errno(&local_err
, -ret
, "writing to file failed");
1507 /* unreachable, thanks to special case in nbd_co_receive_request() */
1511 ret
= blk_co_flush(exp
->blk
);
1513 error_setg_errno(&local_err
, -ret
, "flush failed");
1518 ret
= blk_co_pdiscard(exp
->blk
, request
.from
+ exp
->dev_offset
,
1521 error_setg_errno(&local_err
, -ret
, "discard failed");
1526 error_setg(&local_err
, "invalid request type (%" PRIu32
") received",
1533 /* If we get here, local_err was not a fatal error, and should be sent
1536 msg
= g_strdup(error_get_pretty(local_err
));
1537 error_report_err(local_err
);
1541 if (client
->structured_reply
&&
1542 (ret
< 0 || request
.type
== NBD_CMD_READ
)) {
1544 ret
= nbd_co_send_structured_error(req
->client
, request
.handle
,
1545 -ret
, msg
, &local_err
);
1547 ret
= nbd_co_send_structured_read(req
->client
, request
.handle
,
1548 request
.from
, req
->data
,
1549 reply_data_len
, &local_err
);
1552 ret
= nbd_co_send_simple_reply(req
->client
, request
.handle
,
1554 req
->data
, reply_data_len
, &local_err
);
1558 error_prepend(&local_err
, "Failed to send reply: ");
1562 /* We must disconnect after NBD_CMD_WRITE if we did not
1565 if (!req
->complete
) {
1566 error_setg(&local_err
, "Request handling failed in intermediate state");
1571 nbd_request_put(req
);
1572 nbd_client_put(client
);
1577 error_reportf_err(local_err
, "Disconnect client, due to: ");
1579 nbd_request_put(req
);
1580 client_close(client
, true);
1581 nbd_client_put(client
);
1584 static void nbd_client_receive_next_request(NBDClient
*client
)
1586 if (!client
->recv_coroutine
&& client
->nb_requests
< MAX_NBD_REQUESTS
) {
1587 nbd_client_get(client
);
1588 client
->recv_coroutine
= qemu_coroutine_create(nbd_trip
, client
);
1589 aio_co_schedule(client
->exp
->ctx
, client
->recv_coroutine
);
1593 static coroutine_fn
void nbd_co_client_start(void *opaque
)
1595 NBDClient
*client
= opaque
;
1596 NBDExport
*exp
= client
->exp
;
1597 Error
*local_err
= NULL
;
1600 nbd_export_get(exp
);
1601 QTAILQ_INSERT_TAIL(&exp
->clients
, client
, next
);
1603 qemu_co_mutex_init(&client
->send_lock
);
1605 if (nbd_negotiate(client
, &local_err
)) {
1607 error_report_err(local_err
);
1609 client_close(client
, false);
1613 nbd_client_receive_next_request(client
);
1617 * Create a new client listener on the given export @exp, using the
1618 * given channel @sioc. Begin servicing it in a coroutine. When the
1619 * connection closes, call @close_fn with an indication of whether the
1620 * client completed negotiation.
1622 void nbd_client_new(NBDExport
*exp
,
1623 QIOChannelSocket
*sioc
,
1624 QCryptoTLSCreds
*tlscreds
,
1625 const char *tlsaclname
,
1626 void (*close_fn
)(NBDClient
*, bool))
1631 client
= g_new0(NBDClient
, 1);
1632 client
->refcount
= 1;
1634 client
->tlscreds
= tlscreds
;
1636 object_ref(OBJECT(client
->tlscreds
));
1638 client
->tlsaclname
= g_strdup(tlsaclname
);
1639 client
->sioc
= sioc
;
1640 object_ref(OBJECT(client
->sioc
));
1641 client
->ioc
= QIO_CHANNEL(sioc
);
1642 object_ref(OBJECT(client
->ioc
));
1643 client
->close_fn
= close_fn
;
1645 co
= qemu_coroutine_create(nbd_co_client_start
, client
);
1646 qemu_coroutine_enter(co
);