2 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 * Network Block Device Server Side
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 of the License.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include "qemu/osdep.h"
20 #include "qapi/error.h"
21 #include "nbd-internal.h"
23 static int system_errno_to_nbd_errno(int err
)
47 /* Definitions for opaque data types */
49 typedef struct NBDRequest NBDRequest
;
52 QSIMPLEQ_ENTRY(NBDRequest
) entry
;
59 void (*close
)(NBDExport
*exp
);
66 QTAILQ_HEAD(, NBDClient
) clients
;
67 QTAILQ_ENTRY(NBDExport
) next
;
71 Notifier eject_notifier
;
74 static QTAILQ_HEAD(, NBDExport
) exports
= QTAILQ_HEAD_INITIALIZER(exports
);
78 void (*close
)(NBDClient
*client
);
81 QCryptoTLSCreds
*tlscreds
;
83 QIOChannelSocket
*sioc
; /* The underlying data channel */
84 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
86 Coroutine
*recv_coroutine
;
89 Coroutine
*send_coroutine
;
93 QTAILQ_ENTRY(NBDClient
) next
;
98 /* That's all folks */
100 static void nbd_set_handlers(NBDClient
*client
);
101 static void nbd_unset_handlers(NBDClient
*client
);
102 static void nbd_update_can_read(NBDClient
*client
);
104 static gboolean
nbd_negotiate_continue(QIOChannel
*ioc
,
105 GIOCondition condition
,
108 qemu_coroutine_enter(opaque
, NULL
);
112 static ssize_t
nbd_negotiate_read(QIOChannel
*ioc
, void *buffer
, size_t size
)
117 assert(qemu_in_coroutine());
118 /* Negotiation are always in main loop. */
119 watch
= qio_channel_add_watch(ioc
,
121 nbd_negotiate_continue
,
122 qemu_coroutine_self(),
124 ret
= read_sync(ioc
, buffer
, size
);
125 g_source_remove(watch
);
130 static ssize_t
nbd_negotiate_write(QIOChannel
*ioc
, void *buffer
, size_t size
)
135 assert(qemu_in_coroutine());
136 /* Negotiation are always in main loop. */
137 watch
= qio_channel_add_watch(ioc
,
139 nbd_negotiate_continue
,
140 qemu_coroutine_self(),
142 ret
= write_sync(ioc
, buffer
, size
);
143 g_source_remove(watch
);
147 static ssize_t
nbd_negotiate_drop_sync(QIOChannel
*ioc
, size_t size
)
149 ssize_t ret
, dropped
= size
;
150 uint8_t *buffer
= g_malloc(MIN(65536, size
));
153 ret
= nbd_negotiate_read(ioc
, buffer
, MIN(65536, size
));
167 /* Basic flow for negotiation
194 static int nbd_negotiate_send_rep(QIOChannel
*ioc
, uint32_t type
, uint32_t opt
)
199 TRACE("Reply opt=%x type=%x", type
, opt
);
201 magic
= cpu_to_be64(NBD_REP_MAGIC
);
202 if (nbd_negotiate_write(ioc
, &magic
, sizeof(magic
)) != sizeof(magic
)) {
203 LOG("write failed (rep magic)");
206 opt
= cpu_to_be32(opt
);
207 if (nbd_negotiate_write(ioc
, &opt
, sizeof(opt
)) != sizeof(opt
)) {
208 LOG("write failed (rep opt)");
211 type
= cpu_to_be32(type
);
212 if (nbd_negotiate_write(ioc
, &type
, sizeof(type
)) != sizeof(type
)) {
213 LOG("write failed (rep type)");
216 len
= cpu_to_be32(0);
217 if (nbd_negotiate_write(ioc
, &len
, sizeof(len
)) != sizeof(len
)) {
218 LOG("write failed (rep data length)");
224 static int nbd_negotiate_send_rep_list(QIOChannel
*ioc
, NBDExport
*exp
)
226 uint64_t magic
, name_len
;
227 uint32_t opt
, type
, len
;
229 TRACE("Advertizing export name '%s'", exp
->name
? exp
->name
: "");
230 name_len
= strlen(exp
->name
);
231 magic
= cpu_to_be64(NBD_REP_MAGIC
);
232 if (nbd_negotiate_write(ioc
, &magic
, sizeof(magic
)) != sizeof(magic
)) {
233 LOG("write failed (magic)");
236 opt
= cpu_to_be32(NBD_OPT_LIST
);
237 if (nbd_negotiate_write(ioc
, &opt
, sizeof(opt
)) != sizeof(opt
)) {
238 LOG("write failed (opt)");
241 type
= cpu_to_be32(NBD_REP_SERVER
);
242 if (nbd_negotiate_write(ioc
, &type
, sizeof(type
)) != sizeof(type
)) {
243 LOG("write failed (reply type)");
246 len
= cpu_to_be32(name_len
+ sizeof(len
));
247 if (nbd_negotiate_write(ioc
, &len
, sizeof(len
)) != sizeof(len
)) {
248 LOG("write failed (length)");
251 len
= cpu_to_be32(name_len
);
252 if (nbd_negotiate_write(ioc
, &len
, sizeof(len
)) != sizeof(len
)) {
253 LOG("write failed (length)");
256 if (nbd_negotiate_write(ioc
, exp
->name
, name_len
) != name_len
) {
257 LOG("write failed (buffer)");
263 static int nbd_negotiate_handle_list(NBDClient
*client
, uint32_t length
)
268 if (nbd_negotiate_drop_sync(client
->ioc
, length
) != length
) {
271 return nbd_negotiate_send_rep(client
->ioc
,
272 NBD_REP_ERR_INVALID
, NBD_OPT_LIST
);
275 /* For each export, send a NBD_REP_SERVER reply. */
276 QTAILQ_FOREACH(exp
, &exports
, next
) {
277 if (nbd_negotiate_send_rep_list(client
->ioc
, exp
)) {
281 /* Finish with a NBD_REP_ACK. */
282 return nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, NBD_OPT_LIST
);
285 static int nbd_negotiate_handle_export_name(NBDClient
*client
, uint32_t length
)
291 [20 .. xx] export name (length bytes)
293 TRACE("Checking length");
295 LOG("Bad length received");
298 if (nbd_negotiate_read(client
->ioc
, name
, length
) != length
) {
304 TRACE("Client requested export '%s'", name
);
306 client
->exp
= nbd_export_find(name
);
308 LOG("export not found");
312 QTAILQ_INSERT_TAIL(&client
->exp
->clients
, client
, next
);
313 nbd_export_get(client
->exp
);
320 static QIOChannel
*nbd_negotiate_handle_starttls(NBDClient
*client
,
325 struct NBDTLSHandshakeData data
= { 0 };
327 TRACE("Setting up TLS");
330 if (nbd_negotiate_drop_sync(ioc
, length
) != length
) {
333 nbd_negotiate_send_rep(ioc
, NBD_REP_ERR_INVALID
, NBD_OPT_STARTTLS
);
337 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ACK
, NBD_OPT_STARTTLS
);
339 tioc
= qio_channel_tls_new_server(ioc
,
347 TRACE("Starting TLS handshake");
348 data
.loop
= g_main_loop_new(g_main_context_default(), FALSE
);
349 qio_channel_tls_handshake(tioc
,
354 if (!data
.complete
) {
355 g_main_loop_run(data
.loop
);
357 g_main_loop_unref(data
.loop
);
359 object_unref(OBJECT(tioc
));
360 error_free(data
.error
);
364 return QIO_CHANNEL(tioc
);
368 static int nbd_negotiate_options(NBDClient
*client
)
371 bool fixedNewstyle
= false;
374 [ 0 .. 3] client flags
376 [ 0 .. 7] NBD_OPTS_MAGIC
377 [ 8 .. 11] NBD option
378 [12 .. 15] Data length
381 [ 0 .. 7] NBD_OPTS_MAGIC
382 [ 8 .. 11] Second NBD option
383 [12 .. 15] Data length
387 if (nbd_negotiate_read(client
->ioc
, &flags
, sizeof(flags
)) !=
392 TRACE("Checking client flags");
393 be32_to_cpus(&flags
);
394 if (flags
& NBD_FLAG_C_FIXED_NEWSTYLE
) {
395 TRACE("Support supports fixed newstyle handshake");
396 fixedNewstyle
= true;
397 flags
&= ~NBD_FLAG_C_FIXED_NEWSTYLE
;
400 TRACE("Unknown client flags 0x%x received", flags
);
406 uint32_t clientflags
, length
;
409 if (nbd_negotiate_read(client
->ioc
, &magic
, sizeof(magic
)) !=
414 TRACE("Checking opts magic");
415 if (magic
!= be64_to_cpu(NBD_OPTS_MAGIC
)) {
416 LOG("Bad magic received");
420 if (nbd_negotiate_read(client
->ioc
, &clientflags
,
421 sizeof(clientflags
)) != sizeof(clientflags
)) {
425 clientflags
= be32_to_cpu(clientflags
);
427 if (nbd_negotiate_read(client
->ioc
, &length
, sizeof(length
)) !=
432 length
= be32_to_cpu(length
);
434 TRACE("Checking option 0x%x", clientflags
);
435 if (client
->tlscreds
&&
436 client
->ioc
== (QIOChannel
*)client
->sioc
) {
438 if (!fixedNewstyle
) {
439 TRACE("Unsupported option 0x%x", clientflags
);
442 switch (clientflags
) {
443 case NBD_OPT_STARTTLS
:
444 tioc
= nbd_negotiate_handle_starttls(client
, length
);
448 object_unref(OBJECT(client
->ioc
));
449 client
->ioc
= QIO_CHANNEL(tioc
);
452 case NBD_OPT_EXPORT_NAME
:
453 /* No way to return an error to client, so drop connection */
454 TRACE("Option 0x%x not permitted before TLS", clientflags
);
458 TRACE("Option 0x%x not permitted before TLS", clientflags
);
459 if (nbd_negotiate_drop_sync(client
->ioc
, length
) != length
) {
462 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ERR_TLS_REQD
,
466 } else if (fixedNewstyle
) {
467 switch (clientflags
) {
469 ret
= nbd_negotiate_handle_list(client
, length
);
478 case NBD_OPT_EXPORT_NAME
:
479 return nbd_negotiate_handle_export_name(client
, length
);
481 case NBD_OPT_STARTTLS
:
482 if (nbd_negotiate_drop_sync(client
->ioc
, length
) != length
) {
485 if (client
->tlscreds
) {
486 TRACE("TLS already enabled");
487 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ERR_INVALID
,
490 TRACE("TLS not configured");
491 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ERR_POLICY
,
496 TRACE("Unsupported option 0x%x", clientflags
);
497 if (nbd_negotiate_drop_sync(client
->ioc
, length
) != length
) {
500 nbd_negotiate_send_rep(client
->ioc
, NBD_REP_ERR_UNSUP
,
506 * If broken new-style we should drop the connection
507 * for anything except NBD_OPT_EXPORT_NAME
509 switch (clientflags
) {
510 case NBD_OPT_EXPORT_NAME
:
511 return nbd_negotiate_handle_export_name(client
, length
);
514 TRACE("Unsupported option 0x%x", clientflags
);
526 static coroutine_fn
int nbd_negotiate(NBDClientNewData
*data
)
528 NBDClient
*client
= data
->client
;
529 char buf
[8 + 8 + 8 + 128];
531 const int myflags
= (NBD_FLAG_HAS_FLAGS
| NBD_FLAG_SEND_TRIM
|
532 NBD_FLAG_SEND_FLUSH
| NBD_FLAG_SEND_FUA
);
535 /* Old style negotiation header without options
536 [ 0 .. 7] passwd ("NBDMAGIC")
537 [ 8 .. 15] magic (NBD_CLIENT_MAGIC)
539 [24 .. 25] server flags (0)
540 [26 .. 27] export flags
541 [28 .. 151] reserved (0)
543 New style negotiation header with options
544 [ 0 .. 7] passwd ("NBDMAGIC")
545 [ 8 .. 15] magic (NBD_OPTS_MAGIC)
546 [16 .. 17] server flags (0)
549 [26 .. 27] export flags
550 [28 .. 151] reserved (0)
553 qio_channel_set_blocking(client
->ioc
, false, NULL
);
556 TRACE("Beginning negotiation.");
557 memset(buf
, 0, sizeof(buf
));
558 memcpy(buf
, "NBDMAGIC", 8);
560 oldStyle
= client
->exp
!= NULL
&& !client
->tlscreds
;
562 assert ((client
->exp
->nbdflags
& ~65535) == 0);
563 stq_be_p(buf
+ 8, NBD_CLIENT_MAGIC
);
564 stq_be_p(buf
+ 16, client
->exp
->size
);
565 stw_be_p(buf
+ 26, client
->exp
->nbdflags
| myflags
);
567 stq_be_p(buf
+ 8, NBD_OPTS_MAGIC
);
568 stw_be_p(buf
+ 16, NBD_FLAG_FIXED_NEWSTYLE
);
572 if (client
->tlscreds
) {
573 TRACE("TLS cannot be enabled with oldstyle protocol");
576 if (nbd_negotiate_write(client
->ioc
, buf
, sizeof(buf
)) != sizeof(buf
)) {
581 if (nbd_negotiate_write(client
->ioc
, buf
, 18) != 18) {
585 rc
= nbd_negotiate_options(client
);
587 LOG("option negotiation failed");
591 assert ((client
->exp
->nbdflags
& ~65535) == 0);
592 stq_be_p(buf
+ 18, client
->exp
->size
);
593 stw_be_p(buf
+ 26, client
->exp
->nbdflags
| myflags
);
594 if (nbd_negotiate_write(client
->ioc
, buf
+ 18, sizeof(buf
) - 18) !=
601 TRACE("Negotiation succeeded.");
609 int nbd_disconnect(int fd
)
611 ioctl(fd
, NBD_CLEAR_QUE
);
612 ioctl(fd
, NBD_DISCONNECT
);
613 ioctl(fd
, NBD_CLEAR_SOCK
);
619 int nbd_disconnect(int fd
)
625 static ssize_t
nbd_receive_request(QIOChannel
*ioc
, struct nbd_request
*request
)
627 uint8_t buf
[NBD_REQUEST_SIZE
];
631 ret
= read_sync(ioc
, buf
, sizeof(buf
));
636 if (ret
!= sizeof(buf
)) {
642 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
643 [ 4 .. 7] type (0 == READ, 1 == WRITE)
649 magic
= be32_to_cpup((uint32_t*)buf
);
650 request
->type
= be32_to_cpup((uint32_t*)(buf
+ 4));
651 request
->handle
= be64_to_cpup((uint64_t*)(buf
+ 8));
652 request
->from
= be64_to_cpup((uint64_t*)(buf
+ 16));
653 request
->len
= be32_to_cpup((uint32_t*)(buf
+ 24));
655 TRACE("Got request: "
656 "{ magic = 0x%x, .type = %d, from = %" PRIu64
" , len = %u }",
657 magic
, request
->type
, request
->from
, request
->len
);
659 if (magic
!= NBD_REQUEST_MAGIC
) {
660 LOG("invalid magic (got 0x%x)", magic
);
666 static ssize_t
nbd_send_reply(QIOChannel
*ioc
, struct nbd_reply
*reply
)
668 uint8_t buf
[NBD_REPLY_SIZE
];
671 reply
->error
= system_errno_to_nbd_errno(reply
->error
);
673 TRACE("Sending response to client: { .error = %d, handle = %" PRIu64
" }",
674 reply
->error
, reply
->handle
);
677 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
678 [ 4 .. 7] error (0 == no error)
681 stl_be_p(buf
, NBD_REPLY_MAGIC
);
682 stl_be_p(buf
+ 4, reply
->error
);
683 stq_be_p(buf
+ 8, reply
->handle
);
685 ret
= write_sync(ioc
, buf
, sizeof(buf
));
690 if (ret
!= sizeof(buf
)) {
691 LOG("writing to socket failed");
697 #define MAX_NBD_REQUESTS 16
699 void nbd_client_get(NBDClient
*client
)
704 void nbd_client_put(NBDClient
*client
)
706 if (--client
->refcount
== 0) {
707 /* The last reference should be dropped by client->close,
708 * which is called by client_close.
710 assert(client
->closing
);
712 nbd_unset_handlers(client
);
713 object_unref(OBJECT(client
->sioc
));
714 object_unref(OBJECT(client
->ioc
));
715 if (client
->tlscreds
) {
716 object_unref(OBJECT(client
->tlscreds
));
718 g_free(client
->tlsaclname
);
720 QTAILQ_REMOVE(&client
->exp
->clients
, client
, next
);
721 nbd_export_put(client
->exp
);
727 static void client_close(NBDClient
*client
)
729 if (client
->closing
) {
733 client
->closing
= true;
735 /* Force requests to finish. They will drop their own references,
736 * then we'll close the socket and free the NBDClient.
738 qio_channel_shutdown(client
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
,
741 /* Also tell the client, so that they release their reference. */
743 client
->close(client
);
747 static NBDRequest
*nbd_request_get(NBDClient
*client
)
751 assert(client
->nb_requests
<= MAX_NBD_REQUESTS
- 1);
752 client
->nb_requests
++;
753 nbd_update_can_read(client
);
755 req
= g_new0(NBDRequest
, 1);
756 nbd_client_get(client
);
757 req
->client
= client
;
761 static void nbd_request_put(NBDRequest
*req
)
763 NBDClient
*client
= req
->client
;
766 qemu_vfree(req
->data
);
770 client
->nb_requests
--;
771 nbd_update_can_read(client
);
772 nbd_client_put(client
);
775 static void blk_aio_attached(AioContext
*ctx
, void *opaque
)
777 NBDExport
*exp
= opaque
;
780 TRACE("Export %s: Attaching clients to AIO context %p\n", exp
->name
, ctx
);
784 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
785 nbd_set_handlers(client
);
789 static void blk_aio_detach(void *opaque
)
791 NBDExport
*exp
= opaque
;
794 TRACE("Export %s: Detaching clients from AIO context %p\n", exp
->name
, exp
->ctx
);
796 QTAILQ_FOREACH(client
, &exp
->clients
, next
) {
797 nbd_unset_handlers(client
);
803 static void nbd_eject_notifier(Notifier
*n
, void *data
)
805 NBDExport
*exp
= container_of(n
, NBDExport
, eject_notifier
);
806 nbd_export_close(exp
);
809 NBDExport
*nbd_export_new(BlockBackend
*blk
, off_t dev_offset
, off_t size
,
810 uint32_t nbdflags
, void (*close
)(NBDExport
*),
813 NBDExport
*exp
= g_malloc0(sizeof(NBDExport
));
815 QTAILQ_INIT(&exp
->clients
);
817 exp
->dev_offset
= dev_offset
;
818 exp
->nbdflags
= nbdflags
;
819 exp
->size
= size
< 0 ? blk_getlength(blk
) : size
;
821 error_setg_errno(errp
, -exp
->size
,
822 "Failed to determine the NBD export's length");
825 exp
->size
-= exp
->size
% BDRV_SECTOR_SIZE
;
828 exp
->ctx
= blk_get_aio_context(blk
);
830 blk_add_aio_context_notifier(blk
, blk_aio_attached
, blk_aio_detach
, exp
);
832 exp
->eject_notifier
.notify
= nbd_eject_notifier
;
833 blk_add_remove_bs_notifier(blk
, &exp
->eject_notifier
);
836 * NBD exports are used for non-shared storage migration. Make sure
837 * that BDRV_O_INACTIVE is cleared and the image is ready for write
838 * access since the export could be available before migration handover.
840 aio_context_acquire(exp
->ctx
);
841 blk_invalidate_cache(blk
, NULL
);
842 aio_context_release(exp
->ctx
);
850 NBDExport
*nbd_export_find(const char *name
)
853 QTAILQ_FOREACH(exp
, &exports
, next
) {
854 if (strcmp(name
, exp
->name
) == 0) {
862 void nbd_export_set_name(NBDExport
*exp
, const char *name
)
864 if (exp
->name
== name
) {
869 if (exp
->name
!= NULL
) {
872 QTAILQ_REMOVE(&exports
, exp
, next
);
877 exp
->name
= g_strdup(name
);
878 QTAILQ_INSERT_TAIL(&exports
, exp
, next
);
883 void nbd_export_close(NBDExport
*exp
)
885 NBDClient
*client
, *next
;
888 QTAILQ_FOREACH_SAFE(client
, &exp
->clients
, next
, next
) {
889 client_close(client
);
891 nbd_export_set_name(exp
, NULL
);
895 void nbd_export_get(NBDExport
*exp
)
897 assert(exp
->refcount
> 0);
901 void nbd_export_put(NBDExport
*exp
)
903 assert(exp
->refcount
> 0);
904 if (exp
->refcount
== 1) {
905 nbd_export_close(exp
);
908 if (--exp
->refcount
== 0) {
909 assert(exp
->name
== NULL
);
916 notifier_remove(&exp
->eject_notifier
);
917 blk_remove_aio_context_notifier(exp
->blk
, blk_aio_attached
,
918 blk_aio_detach
, exp
);
927 BlockBackend
*nbd_export_get_blockdev(NBDExport
*exp
)
932 void nbd_export_close_all(void)
934 NBDExport
*exp
, *next
;
936 QTAILQ_FOREACH_SAFE(exp
, &exports
, next
, next
) {
937 nbd_export_close(exp
);
941 static ssize_t
nbd_co_send_reply(NBDRequest
*req
, struct nbd_reply
*reply
,
944 NBDClient
*client
= req
->client
;
947 g_assert(qemu_in_coroutine());
948 qemu_co_mutex_lock(&client
->send_lock
);
949 client
->send_coroutine
= qemu_coroutine_self();
950 nbd_set_handlers(client
);
953 rc
= nbd_send_reply(client
->ioc
, reply
);
955 qio_channel_set_cork(client
->ioc
, true);
956 rc
= nbd_send_reply(client
->ioc
, reply
);
958 ret
= write_sync(client
->ioc
, req
->data
, len
);
963 qio_channel_set_cork(client
->ioc
, false);
966 client
->send_coroutine
= NULL
;
967 nbd_set_handlers(client
);
968 qemu_co_mutex_unlock(&client
->send_lock
);
972 static ssize_t
nbd_co_receive_request(NBDRequest
*req
, struct nbd_request
*request
)
974 NBDClient
*client
= req
->client
;
978 g_assert(qemu_in_coroutine());
979 client
->recv_coroutine
= qemu_coroutine_self();
980 nbd_update_can_read(client
);
982 rc
= nbd_receive_request(client
->ioc
, request
);
990 if ((request
->from
+ request
->len
) < request
->from
) {
991 LOG("integer overflow detected! "
992 "you're probably being attacked");
997 TRACE("Decoding type");
999 command
= request
->type
& NBD_CMD_MASK_COMMAND
;
1000 if (command
== NBD_CMD_READ
|| command
== NBD_CMD_WRITE
) {
1001 if (request
->len
> NBD_MAX_BUFFER_SIZE
) {
1002 LOG("len (%u) is larger than max len (%u)",
1003 request
->len
, NBD_MAX_BUFFER_SIZE
);
1008 req
->data
= blk_try_blockalign(client
->exp
->blk
, request
->len
);
1009 if (req
->data
== NULL
) {
1014 if (command
== NBD_CMD_WRITE
) {
1015 TRACE("Reading %u byte(s)", request
->len
);
1017 if (read_sync(client
->ioc
, req
->data
, request
->len
) != request
->len
) {
1018 LOG("reading from socket failed");
1026 client
->recv_coroutine
= NULL
;
1027 nbd_update_can_read(client
);
1032 static void nbd_trip(void *opaque
)
1034 NBDClient
*client
= opaque
;
1035 NBDExport
*exp
= client
->exp
;
1037 struct nbd_request request
;
1038 struct nbd_reply reply
;
1042 TRACE("Reading request.");
1043 if (client
->closing
) {
1047 req
= nbd_request_get(client
);
1048 ret
= nbd_co_receive_request(req
, &request
);
1049 if (ret
== -EAGAIN
) {
1056 reply
.handle
= request
.handle
;
1063 command
= request
.type
& NBD_CMD_MASK_COMMAND
;
1064 if (command
!= NBD_CMD_DISC
&& (request
.from
+ request
.len
) > exp
->size
) {
1065 LOG("From: %" PRIu64
", Len: %u, Size: %" PRIu64
1066 ", Offset: %" PRIu64
"\n",
1067 request
.from
, request
.len
,
1068 (uint64_t)exp
->size
, (uint64_t)exp
->dev_offset
);
1069 LOG("requested operation past EOF--bad client?");
1070 goto invalid_request
;
1073 if (client
->closing
) {
1075 * The client may be closed when we are blocked in
1076 * nbd_co_receive_request()
1083 TRACE("Request type is READ");
1085 if (request
.type
& NBD_CMD_FLAG_FUA
) {
1086 ret
= blk_co_flush(exp
->blk
);
1088 LOG("flush failed");
1094 ret
= blk_pread(exp
->blk
, request
.from
+ exp
->dev_offset
,
1095 req
->data
, request
.len
);
1097 LOG("reading from file failed");
1102 TRACE("Read %u byte(s)", request
.len
);
1103 if (nbd_co_send_reply(req
, &reply
, request
.len
) < 0)
1107 TRACE("Request type is WRITE");
1109 if (exp
->nbdflags
& NBD_FLAG_READ_ONLY
) {
1110 TRACE("Server is read-only, return error");
1111 reply
.error
= EROFS
;
1115 TRACE("Writing to device");
1117 ret
= blk_pwrite(exp
->blk
, request
.from
+ exp
->dev_offset
,
1118 req
->data
, request
.len
, 0);
1120 LOG("writing to file failed");
1125 if (request
.type
& NBD_CMD_FLAG_FUA
) {
1126 ret
= blk_co_flush(exp
->blk
);
1128 LOG("flush failed");
1134 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
1139 TRACE("Request type is DISCONNECT");
1143 TRACE("Request type is FLUSH");
1145 ret
= blk_co_flush(exp
->blk
);
1147 LOG("flush failed");
1150 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
1155 TRACE("Request type is TRIM");
1156 /* Ignore unaligned head or tail, until block layer adds byte
1158 if (request
.len
>= BDRV_SECTOR_SIZE
) {
1159 request
.len
-= (request
.from
+ request
.len
) % BDRV_SECTOR_SIZE
;
1160 ret
= blk_co_discard(exp
->blk
,
1161 DIV_ROUND_UP(request
.from
+ exp
->dev_offset
,
1163 request
.len
/ BDRV_SECTOR_SIZE
);
1165 LOG("discard failed");
1169 TRACE("trim request too small, ignoring");
1171 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
1176 LOG("invalid request type (%u) received", request
.type
);
1178 reply
.error
= EINVAL
;
1180 if (nbd_co_send_reply(req
, &reply
, 0) < 0) {
1186 TRACE("Request/Reply complete");
1189 nbd_request_put(req
);
1193 nbd_request_put(req
);
1194 client_close(client
);
1197 static void nbd_read(void *opaque
)
1199 NBDClient
*client
= opaque
;
1201 if (client
->recv_coroutine
) {
1202 qemu_coroutine_enter(client
->recv_coroutine
, NULL
);
1204 qemu_coroutine_enter(qemu_coroutine_create(nbd_trip
), client
);
1208 static void nbd_restart_write(void *opaque
)
1210 NBDClient
*client
= opaque
;
1212 qemu_coroutine_enter(client
->send_coroutine
, NULL
);
1215 static void nbd_set_handlers(NBDClient
*client
)
1217 if (client
->exp
&& client
->exp
->ctx
) {
1218 aio_set_fd_handler(client
->exp
->ctx
, client
->sioc
->fd
,
1220 client
->can_read
? nbd_read
: NULL
,
1221 client
->send_coroutine
? nbd_restart_write
: NULL
,
1226 static void nbd_unset_handlers(NBDClient
*client
)
1228 if (client
->exp
&& client
->exp
->ctx
) {
1229 aio_set_fd_handler(client
->exp
->ctx
, client
->sioc
->fd
,
1230 true, NULL
, NULL
, NULL
);
1234 static void nbd_update_can_read(NBDClient
*client
)
1236 bool can_read
= client
->recv_coroutine
||
1237 client
->nb_requests
< MAX_NBD_REQUESTS
;
1239 if (can_read
!= client
->can_read
) {
1240 client
->can_read
= can_read
;
1241 nbd_set_handlers(client
);
1243 /* There is no need to invoke aio_notify(), since aio_set_fd_handler()
1244 * in nbd_set_handlers() will have taken care of that */
1248 static coroutine_fn
void nbd_co_client_start(void *opaque
)
1250 NBDClientNewData
*data
= opaque
;
1251 NBDClient
*client
= data
->client
;
1252 NBDExport
*exp
= client
->exp
;
1255 nbd_export_get(exp
);
1257 if (nbd_negotiate(data
)) {
1258 client_close(client
);
1261 qemu_co_mutex_init(&client
->send_lock
);
1262 nbd_set_handlers(client
);
1265 QTAILQ_INSERT_TAIL(&exp
->clients
, client
, next
);
1271 void nbd_client_new(NBDExport
*exp
,
1272 QIOChannelSocket
*sioc
,
1273 QCryptoTLSCreds
*tlscreds
,
1274 const char *tlsaclname
,
1275 void (*close_fn
)(NBDClient
*))
1278 NBDClientNewData
*data
= g_new(NBDClientNewData
, 1);
1280 client
= g_malloc0(sizeof(NBDClient
));
1281 client
->refcount
= 1;
1283 client
->tlscreds
= tlscreds
;
1285 object_ref(OBJECT(client
->tlscreds
));
1287 client
->tlsaclname
= g_strdup(tlsaclname
);
1288 client
->sioc
= sioc
;
1289 object_ref(OBJECT(client
->sioc
));
1290 client
->ioc
= QIO_CHANNEL(sioc
);
1291 object_ref(OBJECT(client
->ioc
));
1292 client
->can_read
= true;
1293 client
->close
= close_fn
;
1295 data
->client
= client
;
1296 data
->co
= qemu_coroutine_create(nbd_co_client_start
);
1297 qemu_coroutine_enter(data
->co
, data
);