2 * QEMU Block driver for NBD
4 * Copyright (c) 2019 Virtuozzo International GmbH.
5 * Copyright (C) 2016 Red Hat, Inc.
6 * Copyright (C) 2008 Bull S.A.S.
7 * Author: Laurent Vivier <Laurent.Vivier@bull.net>
10 * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
12 * Permission is hereby granted, free of charge, to any person obtaining a copy
13 * of this software and associated documentation files (the "Software"), to deal
14 * in the Software without restriction, including without limitation the rights
15 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
16 * copies of the Software, and to permit persons to whom the Software is
17 * furnished to do so, subject to the following conditions:
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
25 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
27 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
31 #include "qemu/osdep.h"
35 #include "qemu/option.h"
36 #include "qemu/cutils.h"
37 #include "qemu/main-loop.h"
39 #include "qapi/qapi-visit-sockets.h"
40 #include "qapi/qmp/qstring.h"
42 #include "block/qdict.h"
43 #include "block/nbd.h"
44 #include "block/block_int.h"
46 #define EN_OPTSTR ":exportname="
47 #define MAX_NBD_REQUESTS 16
49 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
50 #define INDEX_TO_HANDLE(bs, index) ((index) ^ (uint64_t)(intptr_t)(bs))
54 uint64_t offset
; /* original offset of the request */
55 bool receiving
; /* waiting for connection_co? */
58 typedef enum NBDClientState
{
59 NBD_CLIENT_CONNECTING_WAIT
,
60 NBD_CLIENT_CONNECTING_NOWAIT
,
65 typedef struct BDRVNBDState
{
66 QIOChannelSocket
*sioc
; /* The master data channel */
67 QIOChannel
*ioc
; /* The current I/O channel which may differ (eg TLS) */
72 Coroutine
*connection_co
;
73 QemuCoSleepState
*connection_co_sleep_ns_state
;
75 bool wait_drained_end
;
82 NBDClientRequest requests
[MAX_NBD_REQUESTS
];
86 /* Connection parameters */
87 uint32_t reconnect_delay
;
89 char *export
, *tlscredsid
;
90 QCryptoTLSCreds
*tlscreds
;
95 static int nbd_client_connect(BlockDriverState
*bs
, Error
**errp
);
97 static void nbd_channel_error(BDRVNBDState
*s
, int ret
)
100 if (s
->state
== NBD_CLIENT_CONNECTED
) {
101 s
->state
= s
->reconnect_delay
? NBD_CLIENT_CONNECTING_WAIT
:
102 NBD_CLIENT_CONNECTING_NOWAIT
;
105 if (s
->state
== NBD_CLIENT_CONNECTED
) {
106 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
108 s
->state
= NBD_CLIENT_QUIT
;
112 static void nbd_recv_coroutines_wake_all(BDRVNBDState
*s
)
116 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
117 NBDClientRequest
*req
= &s
->requests
[i
];
119 if (req
->coroutine
&& req
->receiving
) {
120 aio_co_wake(req
->coroutine
);
125 static void nbd_client_detach_aio_context(BlockDriverState
*bs
)
127 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
129 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
132 static void nbd_client_attach_aio_context_bh(void *opaque
)
134 BlockDriverState
*bs
= opaque
;
135 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
138 * The node is still drained, so we know the coroutine has yielded in
139 * nbd_read_eof(), the only place where bs->in_flight can reach 0, or it is
140 * entered for the first time. Both places are safe for entering the
143 qemu_aio_coroutine_enter(bs
->aio_context
, s
->connection_co
);
144 bdrv_dec_in_flight(bs
);
147 static void nbd_client_attach_aio_context(BlockDriverState
*bs
,
148 AioContext
*new_context
)
150 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
153 * s->connection_co is either yielded from nbd_receive_reply or from
154 * nbd_co_reconnect_loop()
156 if (s
->state
== NBD_CLIENT_CONNECTED
) {
157 qio_channel_attach_aio_context(QIO_CHANNEL(s
->ioc
), new_context
);
160 bdrv_inc_in_flight(bs
);
163 * Need to wait here for the BH to run because the BH must run while the
164 * node is still drained.
166 aio_wait_bh_oneshot(new_context
, nbd_client_attach_aio_context_bh
, bs
);
169 static void coroutine_fn
nbd_client_co_drain_begin(BlockDriverState
*bs
)
171 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
174 if (s
->connection_co_sleep_ns_state
) {
175 qemu_co_sleep_wake(s
->connection_co_sleep_ns_state
);
179 static void coroutine_fn
nbd_client_co_drain_end(BlockDriverState
*bs
)
181 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
184 if (s
->wait_drained_end
) {
185 s
->wait_drained_end
= false;
186 aio_co_wake(s
->connection_co
);
191 static void nbd_teardown_connection(BlockDriverState
*bs
)
193 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
195 if (s
->state
== NBD_CLIENT_CONNECTED
) {
196 /* finish any pending coroutines */
198 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
200 s
->state
= NBD_CLIENT_QUIT
;
201 if (s
->connection_co
) {
202 if (s
->connection_co_sleep_ns_state
) {
203 qemu_co_sleep_wake(s
->connection_co_sleep_ns_state
);
206 BDRV_POLL_WHILE(bs
, s
->connection_co
);
209 static bool nbd_client_connecting(BDRVNBDState
*s
)
211 return s
->state
== NBD_CLIENT_CONNECTING_WAIT
||
212 s
->state
== NBD_CLIENT_CONNECTING_NOWAIT
;
215 static bool nbd_client_connecting_wait(BDRVNBDState
*s
)
217 return s
->state
== NBD_CLIENT_CONNECTING_WAIT
;
220 static coroutine_fn
void nbd_reconnect_attempt(BDRVNBDState
*s
)
222 Error
*local_err
= NULL
;
224 if (!nbd_client_connecting(s
)) {
228 /* Wait for completion of all in-flight requests */
230 qemu_co_mutex_lock(&s
->send_mutex
);
232 while (s
->in_flight
> 0) {
233 qemu_co_mutex_unlock(&s
->send_mutex
);
234 nbd_recv_coroutines_wake_all(s
);
235 s
->wait_in_flight
= true;
236 qemu_coroutine_yield();
237 s
->wait_in_flight
= false;
238 qemu_co_mutex_lock(&s
->send_mutex
);
241 qemu_co_mutex_unlock(&s
->send_mutex
);
243 if (!nbd_client_connecting(s
)) {
248 * Now we are sure that nobody is accessing the channel, and no one will
249 * try until we set the state to CONNECTED.
252 /* Finalize previous connection if any */
254 nbd_client_detach_aio_context(s
->bs
);
255 object_unref(OBJECT(s
->sioc
));
257 object_unref(OBJECT(s
->ioc
));
261 s
->connect_status
= nbd_client_connect(s
->bs
, &local_err
);
262 error_free(s
->connect_err
);
263 s
->connect_err
= NULL
;
264 error_propagate(&s
->connect_err
, local_err
);
266 if (s
->connect_status
< 0) {
271 /* successfully connected */
272 s
->state
= NBD_CLIENT_CONNECTED
;
273 qemu_co_queue_restart_all(&s
->free_sema
);
276 static coroutine_fn
void nbd_co_reconnect_loop(BDRVNBDState
*s
)
278 uint64_t start_time_ns
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
279 uint64_t delay_ns
= s
->reconnect_delay
* NANOSECONDS_PER_SECOND
;
280 uint64_t timeout
= 1 * NANOSECONDS_PER_SECOND
;
281 uint64_t max_timeout
= 16 * NANOSECONDS_PER_SECOND
;
283 nbd_reconnect_attempt(s
);
285 while (nbd_client_connecting(s
)) {
286 if (s
->state
== NBD_CLIENT_CONNECTING_WAIT
&&
287 qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - start_time_ns
> delay_ns
)
289 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
290 qemu_co_queue_restart_all(&s
->free_sema
);
293 qemu_co_sleep_ns_wakeable(QEMU_CLOCK_REALTIME
, timeout
,
294 &s
->connection_co_sleep_ns_state
);
296 bdrv_dec_in_flight(s
->bs
);
297 s
->wait_drained_end
= true;
300 * We may be entered once from nbd_client_attach_aio_context_bh
301 * and then from nbd_client_co_drain_end. So here is a loop.
303 qemu_coroutine_yield();
305 bdrv_inc_in_flight(s
->bs
);
307 if (timeout
< max_timeout
) {
311 nbd_reconnect_attempt(s
);
315 static coroutine_fn
void nbd_connection_entry(void *opaque
)
317 BDRVNBDState
*s
= opaque
;
320 Error
*local_err
= NULL
;
322 while (s
->state
!= NBD_CLIENT_QUIT
) {
324 * The NBD client can only really be considered idle when it has
325 * yielded from qio_channel_readv_all_eof(), waiting for data. This is
326 * the point where the additional scheduled coroutine entry happens
327 * after nbd_client_attach_aio_context().
329 * Therefore we keep an additional in_flight reference all the time and
330 * only drop it temporarily here.
333 if (nbd_client_connecting(s
)) {
334 nbd_co_reconnect_loop(s
);
337 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
341 assert(s
->reply
.handle
== 0);
342 ret
= nbd_receive_reply(s
->bs
, s
->ioc
, &s
->reply
, &local_err
);
345 trace_nbd_read_reply_entry_fail(ret
, error_get_pretty(local_err
));
346 error_free(local_err
);
350 nbd_channel_error(s
, ret
? ret
: -EIO
);
355 * There's no need for a mutex on the receive side, because the
356 * handler acts as a synchronization point and ensures that only
357 * one coroutine is called until the reply finishes.
359 i
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
360 if (i
>= MAX_NBD_REQUESTS
||
361 !s
->requests
[i
].coroutine
||
362 !s
->requests
[i
].receiving
||
363 (nbd_reply_is_structured(&s
->reply
) && !s
->info
.structured_reply
))
365 nbd_channel_error(s
, -EINVAL
);
370 * We're woken up again by the request itself. Note that there
371 * is no race between yielding and reentering connection_co. This
374 * - if the request runs on the same AioContext, it is only
375 * entered after we yield
377 * - if the request runs on a different AioContext, reentering
378 * connection_co happens through a bottom half, which can only
379 * run after we yield.
381 aio_co_wake(s
->requests
[i
].coroutine
);
382 qemu_coroutine_yield();
385 qemu_co_queue_restart_all(&s
->free_sema
);
386 nbd_recv_coroutines_wake_all(s
);
387 bdrv_dec_in_flight(s
->bs
);
389 s
->connection_co
= NULL
;
391 nbd_client_detach_aio_context(s
->bs
);
392 object_unref(OBJECT(s
->sioc
));
394 object_unref(OBJECT(s
->ioc
));
401 static int nbd_co_send_request(BlockDriverState
*bs
,
405 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
408 qemu_co_mutex_lock(&s
->send_mutex
);
409 while (s
->in_flight
== MAX_NBD_REQUESTS
|| nbd_client_connecting_wait(s
)) {
410 qemu_co_queue_wait(&s
->free_sema
, &s
->send_mutex
);
413 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
420 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
421 if (s
->requests
[i
].coroutine
== NULL
) {
426 g_assert(qemu_in_coroutine());
427 assert(i
< MAX_NBD_REQUESTS
);
429 s
->requests
[i
].coroutine
= qemu_coroutine_self();
430 s
->requests
[i
].offset
= request
->from
;
431 s
->requests
[i
].receiving
= false;
433 request
->handle
= INDEX_TO_HANDLE(s
, i
);
438 qio_channel_set_cork(s
->ioc
, true);
439 rc
= nbd_send_request(s
->ioc
, request
);
440 if (rc
>= 0 && s
->state
== NBD_CLIENT_CONNECTED
) {
441 if (qio_channel_writev_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
445 } else if (rc
>= 0) {
448 qio_channel_set_cork(s
->ioc
, false);
450 rc
= nbd_send_request(s
->ioc
, request
);
455 nbd_channel_error(s
, rc
);
457 s
->requests
[i
].coroutine
= NULL
;
460 if (s
->in_flight
== 0 && s
->wait_in_flight
) {
461 aio_co_wake(s
->connection_co
);
463 qemu_co_queue_next(&s
->free_sema
);
466 qemu_co_mutex_unlock(&s
->send_mutex
);
470 static inline uint16_t payload_advance16(uint8_t **payload
)
473 return lduw_be_p(*payload
- 2);
476 static inline uint32_t payload_advance32(uint8_t **payload
)
479 return ldl_be_p(*payload
- 4);
482 static inline uint64_t payload_advance64(uint8_t **payload
)
485 return ldq_be_p(*payload
- 8);
488 static int nbd_parse_offset_hole_payload(BDRVNBDState
*s
,
489 NBDStructuredReplyChunk
*chunk
,
490 uint8_t *payload
, uint64_t orig_offset
,
491 QEMUIOVector
*qiov
, Error
**errp
)
496 if (chunk
->length
!= sizeof(offset
) + sizeof(hole_size
)) {
497 error_setg(errp
, "Protocol error: invalid payload for "
498 "NBD_REPLY_TYPE_OFFSET_HOLE");
502 offset
= payload_advance64(&payload
);
503 hole_size
= payload_advance32(&payload
);
505 if (!hole_size
|| offset
< orig_offset
|| hole_size
> qiov
->size
||
506 offset
> orig_offset
+ qiov
->size
- hole_size
) {
507 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
511 if (s
->info
.min_block
&&
512 !QEMU_IS_ALIGNED(hole_size
, s
->info
.min_block
)) {
513 trace_nbd_structured_read_compliance("hole");
516 qemu_iovec_memset(qiov
, offset
- orig_offset
, 0, hole_size
);
522 * nbd_parse_blockstatus_payload
523 * Based on our request, we expect only one extent in reply, for the
524 * base:allocation context.
526 static int nbd_parse_blockstatus_payload(BDRVNBDState
*s
,
527 NBDStructuredReplyChunk
*chunk
,
528 uint8_t *payload
, uint64_t orig_length
,
529 NBDExtent
*extent
, Error
**errp
)
533 /* The server succeeded, so it must have sent [at least] one extent */
534 if (chunk
->length
< sizeof(context_id
) + sizeof(*extent
)) {
535 error_setg(errp
, "Protocol error: invalid payload for "
536 "NBD_REPLY_TYPE_BLOCK_STATUS");
540 context_id
= payload_advance32(&payload
);
541 if (s
->info
.context_id
!= context_id
) {
542 error_setg(errp
, "Protocol error: unexpected context id %d for "
543 "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
544 "id is %d", context_id
,
549 extent
->length
= payload_advance32(&payload
);
550 extent
->flags
= payload_advance32(&payload
);
552 if (extent
->length
== 0) {
553 error_setg(errp
, "Protocol error: server sent status chunk with "
559 * A server sending unaligned block status is in violation of the
560 * protocol, but as qemu-nbd 3.1 is such a server (at least for
561 * POSIX files that are not a multiple of 512 bytes, since qemu
562 * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
563 * still sees an implicit hole beyond the real EOF), it's nicer to
564 * work around the misbehaving server. If the request included
565 * more than the final unaligned block, truncate it back to an
566 * aligned result; if the request was only the final block, round
567 * up to the full block and change the status to fully-allocated
568 * (always a safe status, even if it loses information).
570 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(extent
->length
,
571 s
->info
.min_block
)) {
572 trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
573 if (extent
->length
> s
->info
.min_block
) {
574 extent
->length
= QEMU_ALIGN_DOWN(extent
->length
,
577 extent
->length
= s
->info
.min_block
;
583 * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
584 * sent us any more than one extent, nor should it have included
585 * status beyond our request in that extent. However, it's easy
586 * enough to ignore the server's noncompliance without killing the
587 * connection; just ignore trailing extents, and clamp things to
588 * the length of our request.
590 if (chunk
->length
> sizeof(context_id
) + sizeof(*extent
)) {
591 trace_nbd_parse_blockstatus_compliance("more than one extent");
593 if (extent
->length
> orig_length
) {
594 extent
->length
= orig_length
;
595 trace_nbd_parse_blockstatus_compliance("extent length too large");
602 * nbd_parse_error_payload
603 * on success @errp contains message describing nbd error reply
605 static int nbd_parse_error_payload(NBDStructuredReplyChunk
*chunk
,
606 uint8_t *payload
, int *request_ret
,
610 uint16_t message_size
;
612 assert(chunk
->type
& (1 << 15));
614 if (chunk
->length
< sizeof(error
) + sizeof(message_size
)) {
616 "Protocol error: invalid payload for structured error");
620 error
= nbd_errno_to_system_errno(payload_advance32(&payload
));
622 error_setg(errp
, "Protocol error: server sent structured error chunk "
627 *request_ret
= -error
;
628 message_size
= payload_advance16(&payload
);
630 if (message_size
> chunk
->length
- sizeof(error
) - sizeof(message_size
)) {
631 error_setg(errp
, "Protocol error: server sent structured error chunk "
632 "with incorrect message size");
636 /* TODO: Add a trace point to mention the server complaint */
638 /* TODO handle ERROR_OFFSET */
643 static int nbd_co_receive_offset_data_payload(BDRVNBDState
*s
,
644 uint64_t orig_offset
,
645 QEMUIOVector
*qiov
, Error
**errp
)
647 QEMUIOVector sub_qiov
;
651 NBDStructuredReplyChunk
*chunk
= &s
->reply
.structured
;
653 assert(nbd_reply_is_structured(&s
->reply
));
655 /* The NBD spec requires at least one byte of payload */
656 if (chunk
->length
<= sizeof(offset
)) {
657 error_setg(errp
, "Protocol error: invalid payload for "
658 "NBD_REPLY_TYPE_OFFSET_DATA");
662 if (nbd_read64(s
->ioc
, &offset
, "OFFSET_DATA offset", errp
) < 0) {
666 data_size
= chunk
->length
- sizeof(offset
);
668 if (offset
< orig_offset
|| data_size
> qiov
->size
||
669 offset
> orig_offset
+ qiov
->size
- data_size
) {
670 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
674 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(data_size
, s
->info
.min_block
)) {
675 trace_nbd_structured_read_compliance("data");
678 qemu_iovec_init(&sub_qiov
, qiov
->niov
);
679 qemu_iovec_concat(&sub_qiov
, qiov
, offset
- orig_offset
, data_size
);
680 ret
= qio_channel_readv_all(s
->ioc
, sub_qiov
.iov
, sub_qiov
.niov
, errp
);
681 qemu_iovec_destroy(&sub_qiov
);
683 return ret
< 0 ? -EIO
: 0;
686 #define NBD_MAX_MALLOC_PAYLOAD 1000
687 static coroutine_fn
int nbd_co_receive_structured_payload(
688 BDRVNBDState
*s
, void **payload
, Error
**errp
)
693 assert(nbd_reply_is_structured(&s
->reply
));
695 len
= s
->reply
.structured
.length
;
701 if (payload
== NULL
) {
702 error_setg(errp
, "Unexpected structured payload");
706 if (len
> NBD_MAX_MALLOC_PAYLOAD
) {
707 error_setg(errp
, "Payload too large");
711 *payload
= g_new(char, len
);
712 ret
= nbd_read(s
->ioc
, *payload
, len
, "structured payload", errp
);
723 * nbd_co_do_receive_one_chunk
725 * set request_ret to received reply error
726 * if qiov is not NULL: read payload to @qiov
727 * for structured reply chunk:
728 * if error chunk: read payload, set @request_ret, do not set @payload
729 * else if offset_data chunk: read payload data to @qiov, do not set @payload
730 * else: read payload to @payload
732 * If function fails, @errp contains corresponding error message, and the
733 * connection with the server is suspect. If it returns 0, then the
734 * transaction succeeded (although @request_ret may be a negative errno
735 * corresponding to the server's error reply), and errp is unchanged.
737 static coroutine_fn
int nbd_co_do_receive_one_chunk(
738 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
739 int *request_ret
, QEMUIOVector
*qiov
, void **payload
, Error
**errp
)
742 int i
= HANDLE_TO_INDEX(s
, handle
);
743 void *local_payload
= NULL
;
744 NBDStructuredReplyChunk
*chunk
;
751 /* Wait until we're woken up by nbd_connection_entry. */
752 s
->requests
[i
].receiving
= true;
753 qemu_coroutine_yield();
754 s
->requests
[i
].receiving
= false;
755 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
756 error_setg(errp
, "Connection closed");
761 assert(s
->reply
.handle
== handle
);
763 if (nbd_reply_is_simple(&s
->reply
)) {
764 if (only_structured
) {
765 error_setg(errp
, "Protocol error: simple reply when structured "
766 "reply chunk was expected");
770 *request_ret
= -nbd_errno_to_system_errno(s
->reply
.simple
.error
);
771 if (*request_ret
< 0 || !qiov
) {
775 return qio_channel_readv_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
776 errp
) < 0 ? -EIO
: 0;
779 /* handle structured reply chunk */
780 assert(s
->info
.structured_reply
);
781 chunk
= &s
->reply
.structured
;
783 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
784 if (!(chunk
->flags
& NBD_REPLY_FLAG_DONE
)) {
785 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
786 " NBD_REPLY_FLAG_DONE flag set");
790 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
797 if (chunk
->type
== NBD_REPLY_TYPE_OFFSET_DATA
) {
799 error_setg(errp
, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
803 return nbd_co_receive_offset_data_payload(s
, s
->requests
[i
].offset
,
807 if (nbd_reply_type_is_error(chunk
->type
)) {
808 payload
= &local_payload
;
811 ret
= nbd_co_receive_structured_payload(s
, payload
, errp
);
816 if (nbd_reply_type_is_error(chunk
->type
)) {
817 ret
= nbd_parse_error_payload(chunk
, local_payload
, request_ret
, errp
);
818 g_free(local_payload
);
826 * nbd_co_receive_one_chunk
827 * Read reply, wake up connection_co and set s->quit if needed.
828 * Return value is a fatal error code or normal nbd reply error code
830 static coroutine_fn
int nbd_co_receive_one_chunk(
831 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
832 int *request_ret
, QEMUIOVector
*qiov
, NBDReply
*reply
, void **payload
,
835 int ret
= nbd_co_do_receive_one_chunk(s
, handle
, only_structured
,
836 request_ret
, qiov
, payload
, errp
);
839 memset(reply
, 0, sizeof(*reply
));
840 nbd_channel_error(s
, ret
);
842 /* For assert at loop start in nbd_connection_entry */
847 if (s
->connection_co
&& !s
->wait_in_flight
) {
849 * We must check s->wait_in_flight, because we may entered by
850 * nbd_recv_coroutines_wake_all(), in this case we should not
851 * wake connection_co here, it will woken by last request.
853 aio_co_wake(s
->connection_co
);
859 typedef struct NBDReplyChunkIter
{
863 bool done
, only_structured
;
866 static void nbd_iter_channel_error(NBDReplyChunkIter
*iter
,
867 int ret
, Error
**local_err
)
869 assert(local_err
&& *local_err
);
874 error_propagate(&iter
->err
, *local_err
);
876 error_free(*local_err
);
882 static void nbd_iter_request_error(NBDReplyChunkIter
*iter
, int ret
)
886 if (!iter
->request_ret
) {
887 iter
->request_ret
= ret
;
892 * NBD_FOREACH_REPLY_CHUNK
893 * The pointer stored in @payload requires g_free() to free it.
895 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
896 qiov, reply, payload) \
897 for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
898 nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
901 * nbd_reply_chunk_iter_receive
902 * The pointer stored in @payload requires g_free() to free it.
904 static bool nbd_reply_chunk_iter_receive(BDRVNBDState
*s
,
905 NBDReplyChunkIter
*iter
,
907 QEMUIOVector
*qiov
, NBDReply
*reply
,
910 int ret
, request_ret
;
911 NBDReply local_reply
;
912 NBDStructuredReplyChunk
*chunk
;
913 Error
*local_err
= NULL
;
914 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
915 error_setg(&local_err
, "Connection closed");
916 nbd_iter_channel_error(iter
, -EIO
, &local_err
);
921 /* Previous iteration was last. */
926 reply
= &local_reply
;
929 ret
= nbd_co_receive_one_chunk(s
, handle
, iter
->only_structured
,
930 &request_ret
, qiov
, reply
, payload
,
933 nbd_iter_channel_error(iter
, ret
, &local_err
);
934 } else if (request_ret
< 0) {
935 nbd_iter_request_error(iter
, request_ret
);
938 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
939 if (nbd_reply_is_simple(reply
) || s
->state
!= NBD_CLIENT_CONNECTED
) {
943 chunk
= &reply
->structured
;
944 iter
->only_structured
= true;
946 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
947 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
948 assert(chunk
->flags
& NBD_REPLY_FLAG_DONE
);
952 if (chunk
->flags
& NBD_REPLY_FLAG_DONE
) {
953 /* This iteration is last. */
957 /* Execute the loop body */
961 s
->requests
[HANDLE_TO_INDEX(s
, handle
)].coroutine
= NULL
;
963 qemu_co_mutex_lock(&s
->send_mutex
);
965 if (s
->in_flight
== 0 && s
->wait_in_flight
) {
966 aio_co_wake(s
->connection_co
);
968 qemu_co_queue_next(&s
->free_sema
);
970 qemu_co_mutex_unlock(&s
->send_mutex
);
975 static int nbd_co_receive_return_code(BDRVNBDState
*s
, uint64_t handle
,
976 int *request_ret
, Error
**errp
)
978 NBDReplyChunkIter iter
;
980 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, NULL
, NULL
) {
981 /* nbd_reply_chunk_iter_receive does all the work */
984 error_propagate(errp
, iter
.err
);
985 *request_ret
= iter
.request_ret
;
989 static int nbd_co_receive_cmdread_reply(BDRVNBDState
*s
, uint64_t handle
,
990 uint64_t offset
, QEMUIOVector
*qiov
,
991 int *request_ret
, Error
**errp
)
993 NBDReplyChunkIter iter
;
995 void *payload
= NULL
;
996 Error
*local_err
= NULL
;
998 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, s
->info
.structured_reply
,
999 qiov
, &reply
, &payload
)
1002 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1004 assert(nbd_reply_is_structured(&reply
));
1006 switch (chunk
->type
) {
1007 case NBD_REPLY_TYPE_OFFSET_DATA
:
1009 * special cased in nbd_co_receive_one_chunk, data is already
1013 case NBD_REPLY_TYPE_OFFSET_HOLE
:
1014 ret
= nbd_parse_offset_hole_payload(s
, &reply
.structured
, payload
,
1015 offset
, qiov
, &local_err
);
1017 nbd_channel_error(s
, ret
);
1018 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1022 if (!nbd_reply_type_is_error(chunk
->type
)) {
1023 /* not allowed reply type */
1024 nbd_channel_error(s
, -EINVAL
);
1025 error_setg(&local_err
,
1026 "Unexpected reply type: %d (%s) for CMD_READ",
1027 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1028 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1036 error_propagate(errp
, iter
.err
);
1037 *request_ret
= iter
.request_ret
;
1041 static int nbd_co_receive_blockstatus_reply(BDRVNBDState
*s
,
1042 uint64_t handle
, uint64_t length
,
1044 int *request_ret
, Error
**errp
)
1046 NBDReplyChunkIter iter
;
1048 void *payload
= NULL
;
1049 Error
*local_err
= NULL
;
1050 bool received
= false;
1052 assert(!extent
->length
);
1053 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, &reply
, &payload
) {
1055 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1057 assert(nbd_reply_is_structured(&reply
));
1059 switch (chunk
->type
) {
1060 case NBD_REPLY_TYPE_BLOCK_STATUS
:
1062 nbd_channel_error(s
, -EINVAL
);
1063 error_setg(&local_err
, "Several BLOCK_STATUS chunks in reply");
1064 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1068 ret
= nbd_parse_blockstatus_payload(s
, &reply
.structured
,
1069 payload
, length
, extent
,
1072 nbd_channel_error(s
, ret
);
1073 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1077 if (!nbd_reply_type_is_error(chunk
->type
)) {
1078 nbd_channel_error(s
, -EINVAL
);
1079 error_setg(&local_err
,
1080 "Unexpected reply type: %d (%s) "
1081 "for CMD_BLOCK_STATUS",
1082 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1083 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1091 if (!extent
->length
&& !iter
.request_ret
) {
1092 error_setg(&local_err
, "Server did not reply with any status extents");
1093 nbd_iter_channel_error(&iter
, -EIO
, &local_err
);
1096 error_propagate(errp
, iter
.err
);
1097 *request_ret
= iter
.request_ret
;
1101 static int nbd_co_request(BlockDriverState
*bs
, NBDRequest
*request
,
1102 QEMUIOVector
*write_qiov
)
1104 int ret
, request_ret
;
1105 Error
*local_err
= NULL
;
1106 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1108 assert(request
->type
!= NBD_CMD_READ
);
1110 assert(request
->type
== NBD_CMD_WRITE
);
1111 assert(request
->len
== iov_size(write_qiov
->iov
, write_qiov
->niov
));
1113 assert(request
->type
!= NBD_CMD_WRITE
);
1117 ret
= nbd_co_send_request(bs
, request
, write_qiov
);
1122 ret
= nbd_co_receive_return_code(s
, request
->handle
,
1123 &request_ret
, &local_err
);
1125 trace_nbd_co_request_fail(request
->from
, request
->len
,
1126 request
->handle
, request
->flags
,
1128 nbd_cmd_lookup(request
->type
),
1129 ret
, error_get_pretty(local_err
));
1130 error_free(local_err
);
1133 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1135 return ret
? ret
: request_ret
;
1138 static int nbd_client_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1139 uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1141 int ret
, request_ret
;
1142 Error
*local_err
= NULL
;
1143 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1144 NBDRequest request
= {
1145 .type
= NBD_CMD_READ
,
1150 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1157 * Work around the fact that the block layer doesn't do
1158 * byte-accurate sizing yet - if the read exceeds the server's
1159 * advertised size because the block layer rounded size up, then
1160 * truncate the request to the server and tail-pad with zero.
1162 if (offset
>= s
->info
.size
) {
1163 assert(bytes
< BDRV_SECTOR_SIZE
);
1164 qemu_iovec_memset(qiov
, 0, 0, bytes
);
1167 if (offset
+ bytes
> s
->info
.size
) {
1168 uint64_t slop
= offset
+ bytes
- s
->info
.size
;
1170 assert(slop
< BDRV_SECTOR_SIZE
);
1171 qemu_iovec_memset(qiov
, bytes
- slop
, 0, slop
);
1172 request
.len
-= slop
;
1176 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1181 ret
= nbd_co_receive_cmdread_reply(s
, request
.handle
, offset
, qiov
,
1182 &request_ret
, &local_err
);
1184 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1185 request
.flags
, request
.type
,
1186 nbd_cmd_lookup(request
.type
),
1187 ret
, error_get_pretty(local_err
));
1188 error_free(local_err
);
1191 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1193 return ret
? ret
: request_ret
;
1196 static int nbd_client_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1197 uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1199 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1200 NBDRequest request
= {
1201 .type
= NBD_CMD_WRITE
,
1206 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1207 if (flags
& BDRV_REQ_FUA
) {
1208 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1209 request
.flags
|= NBD_CMD_FLAG_FUA
;
1212 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1217 return nbd_co_request(bs
, &request
, qiov
);
1220 static int nbd_client_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1221 int bytes
, BdrvRequestFlags flags
)
1223 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1224 NBDRequest request
= {
1225 .type
= NBD_CMD_WRITE_ZEROES
,
1230 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1231 if (!(s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
)) {
1235 if (flags
& BDRV_REQ_FUA
) {
1236 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1237 request
.flags
|= NBD_CMD_FLAG_FUA
;
1239 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1240 request
.flags
|= NBD_CMD_FLAG_NO_HOLE
;
1242 if (flags
& BDRV_REQ_NO_FALLBACK
) {
1243 assert(s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
);
1244 request
.flags
|= NBD_CMD_FLAG_FAST_ZERO
;
1250 return nbd_co_request(bs
, &request
, NULL
);
1253 static int nbd_client_co_flush(BlockDriverState
*bs
)
1255 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1256 NBDRequest request
= { .type
= NBD_CMD_FLUSH
};
1258 if (!(s
->info
.flags
& NBD_FLAG_SEND_FLUSH
)) {
1265 return nbd_co_request(bs
, &request
, NULL
);
1268 static int nbd_client_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
1271 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1272 NBDRequest request
= {
1273 .type
= NBD_CMD_TRIM
,
1278 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1279 if (!(s
->info
.flags
& NBD_FLAG_SEND_TRIM
) || !bytes
) {
1283 return nbd_co_request(bs
, &request
, NULL
);
1286 static int coroutine_fn
nbd_client_co_block_status(
1287 BlockDriverState
*bs
, bool want_zero
, int64_t offset
, int64_t bytes
,
1288 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
1290 int ret
, request_ret
;
1291 NBDExtent extent
= { 0 };
1292 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1293 Error
*local_err
= NULL
;
1295 NBDRequest request
= {
1296 .type
= NBD_CMD_BLOCK_STATUS
,
1298 .len
= MIN(MIN_NON_ZERO(QEMU_ALIGN_DOWN(INT_MAX
,
1299 bs
->bl
.request_alignment
),
1301 MIN(bytes
, s
->info
.size
- offset
)),
1302 .flags
= NBD_CMD_FLAG_REQ_ONE
,
1305 if (!s
->info
.base_allocation
) {
1309 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
1313 * Work around the fact that the block layer doesn't do
1314 * byte-accurate sizing yet - if the status request exceeds the
1315 * server's advertised size because the block layer rounded size
1316 * up, we truncated the request to the server (above), or are
1317 * called on just the hole.
1319 if (offset
>= s
->info
.size
) {
1321 assert(bytes
< BDRV_SECTOR_SIZE
);
1322 /* Intentionally don't report offset_valid for the hole */
1323 return BDRV_BLOCK_ZERO
;
1326 if (s
->info
.min_block
) {
1327 assert(QEMU_IS_ALIGNED(request
.len
, s
->info
.min_block
));
1330 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1335 ret
= nbd_co_receive_blockstatus_reply(s
, request
.handle
, bytes
,
1336 &extent
, &request_ret
,
1339 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1340 request
.flags
, request
.type
,
1341 nbd_cmd_lookup(request
.type
),
1342 ret
, error_get_pretty(local_err
));
1343 error_free(local_err
);
1346 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1348 if (ret
< 0 || request_ret
< 0) {
1349 return ret
? ret
: request_ret
;
1352 assert(extent
.length
);
1353 *pnum
= extent
.length
;
1356 return (extent
.flags
& NBD_STATE_HOLE
? 0 : BDRV_BLOCK_DATA
) |
1357 (extent
.flags
& NBD_STATE_ZERO
? BDRV_BLOCK_ZERO
: 0) |
1358 BDRV_BLOCK_OFFSET_VALID
;
1361 static int nbd_client_reopen_prepare(BDRVReopenState
*state
,
1362 BlockReopenQueue
*queue
, Error
**errp
)
1364 BDRVNBDState
*s
= (BDRVNBDState
*)state
->bs
->opaque
;
1366 if ((state
->flags
& BDRV_O_RDWR
) && (s
->info
.flags
& NBD_FLAG_READ_ONLY
)) {
1367 error_setg(errp
, "Can't reopen read-only NBD mount as read/write");
1373 static void nbd_client_close(BlockDriverState
*bs
)
1375 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1376 NBDRequest request
= { .type
= NBD_CMD_DISC
};
1379 nbd_send_request(s
->ioc
, &request
);
1382 nbd_teardown_connection(bs
);
1385 static QIOChannelSocket
*nbd_establish_connection(SocketAddress
*saddr
,
1388 QIOChannelSocket
*sioc
;
1389 Error
*local_err
= NULL
;
1391 sioc
= qio_channel_socket_new();
1392 qio_channel_set_name(QIO_CHANNEL(sioc
), "nbd-client");
1394 qio_channel_socket_connect_sync(sioc
, saddr
, &local_err
);
1396 object_unref(OBJECT(sioc
));
1397 error_propagate(errp
, local_err
);
1401 qio_channel_set_delay(QIO_CHANNEL(sioc
), false);
1406 static int nbd_client_connect(BlockDriverState
*bs
, Error
**errp
)
1408 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1409 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1413 * establish TCP connection, return error if it fails
1414 * TODO: Configurable retry-until-timeout behaviour.
1416 QIOChannelSocket
*sioc
= nbd_establish_connection(s
->saddr
, errp
);
1419 return -ECONNREFUSED
;
1423 trace_nbd_client_connect(s
->export
);
1424 qio_channel_set_blocking(QIO_CHANNEL(sioc
), false, NULL
);
1425 qio_channel_attach_aio_context(QIO_CHANNEL(sioc
), aio_context
);
1427 s
->info
.request_sizes
= true;
1428 s
->info
.structured_reply
= true;
1429 s
->info
.base_allocation
= true;
1430 s
->info
.x_dirty_bitmap
= g_strdup(s
->x_dirty_bitmap
);
1431 s
->info
.name
= g_strdup(s
->export
?: "");
1432 ret
= nbd_receive_negotiate(aio_context
, QIO_CHANNEL(sioc
), s
->tlscreds
,
1433 s
->hostname
, &s
->ioc
, &s
->info
, errp
);
1434 g_free(s
->info
.x_dirty_bitmap
);
1435 g_free(s
->info
.name
);
1437 object_unref(OBJECT(sioc
));
1440 if (s
->x_dirty_bitmap
&& !s
->info
.base_allocation
) {
1441 error_setg(errp
, "requested x-dirty-bitmap %s not found",
1446 if (s
->info
.flags
& NBD_FLAG_READ_ONLY
) {
1447 ret
= bdrv_apply_auto_read_only(bs
, "NBD export is read-only", errp
);
1452 if (s
->info
.flags
& NBD_FLAG_SEND_FUA
) {
1453 bs
->supported_write_flags
= BDRV_REQ_FUA
;
1454 bs
->supported_zero_flags
|= BDRV_REQ_FUA
;
1456 if (s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
) {
1457 bs
->supported_zero_flags
|= BDRV_REQ_MAY_UNMAP
;
1458 if (s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
) {
1459 bs
->supported_zero_flags
|= BDRV_REQ_NO_FALLBACK
;
1466 s
->ioc
= QIO_CHANNEL(sioc
);
1467 object_ref(OBJECT(s
->ioc
));
1470 trace_nbd_client_connect_success(s
->export
);
1476 * We have connected, but must fail for other reasons.
1477 * Send NBD_CMD_DISC as a courtesy to the server.
1480 NBDRequest request
= { .type
= NBD_CMD_DISC
};
1482 nbd_send_request(s
->ioc
?: QIO_CHANNEL(sioc
), &request
);
1484 object_unref(OBJECT(sioc
));
1491 * Parse nbd_open options
1494 static int nbd_parse_uri(const char *filename
, QDict
*options
)
1498 QueryParams
*qp
= NULL
;
1502 uri
= uri_parse(filename
);
1508 if (!g_strcmp0(uri
->scheme
, "nbd")) {
1510 } else if (!g_strcmp0(uri
->scheme
, "nbd+tcp")) {
1512 } else if (!g_strcmp0(uri
->scheme
, "nbd+unix")) {
1519 p
= uri
->path
? uri
->path
: "/";
1520 p
+= strspn(p
, "/");
1522 qdict_put_str(options
, "export", p
);
1525 qp
= query_params_parse(uri
->query
);
1526 if (qp
->n
> 1 || (is_unix
&& !qp
->n
) || (!is_unix
&& qp
->n
)) {
1532 /* nbd+unix:///export?socket=path */
1533 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
1537 qdict_put_str(options
, "server.type", "unix");
1538 qdict_put_str(options
, "server.path", qp
->p
[0].value
);
1543 /* nbd[+tcp]://host[:port]/export */
1549 /* strip braces from literal IPv6 address */
1550 if (uri
->server
[0] == '[') {
1551 host
= qstring_from_substr(uri
->server
, 1,
1552 strlen(uri
->server
) - 1);
1554 host
= qstring_from_str(uri
->server
);
1557 qdict_put_str(options
, "server.type", "inet");
1558 qdict_put(options
, "server.host", host
);
1560 port_str
= g_strdup_printf("%d", uri
->port
?: NBD_DEFAULT_PORT
);
1561 qdict_put_str(options
, "server.port", port_str
);
1567 query_params_free(qp
);
1573 static bool nbd_has_filename_options_conflict(QDict
*options
, Error
**errp
)
1575 const QDictEntry
*e
;
1577 for (e
= qdict_first(options
); e
; e
= qdict_next(options
, e
)) {
1578 if (!strcmp(e
->key
, "host") ||
1579 !strcmp(e
->key
, "port") ||
1580 !strcmp(e
->key
, "path") ||
1581 !strcmp(e
->key
, "export") ||
1582 strstart(e
->key
, "server.", NULL
))
1584 error_setg(errp
, "Option '%s' cannot be used with a file name",
1593 static void nbd_parse_filename(const char *filename
, QDict
*options
,
1596 g_autofree
char *file
= NULL
;
1598 const char *host_spec
;
1599 const char *unixpath
;
1601 if (nbd_has_filename_options_conflict(options
, errp
)) {
1605 if (strstr(filename
, "://")) {
1606 int ret
= nbd_parse_uri(filename
, options
);
1608 error_setg(errp
, "No valid URL specified");
1613 file
= g_strdup(filename
);
1615 export_name
= strstr(file
, EN_OPTSTR
);
1617 if (export_name
[strlen(EN_OPTSTR
)] == 0) {
1620 export_name
[0] = 0; /* truncate 'file' */
1621 export_name
+= strlen(EN_OPTSTR
);
1623 qdict_put_str(options
, "export", export_name
);
1626 /* extract the host_spec - fail if it's not nbd:... */
1627 if (!strstart(file
, "nbd:", &host_spec
)) {
1628 error_setg(errp
, "File name string for NBD must start with 'nbd:'");
1636 /* are we a UNIX or TCP socket? */
1637 if (strstart(host_spec
, "unix:", &unixpath
)) {
1638 qdict_put_str(options
, "server.type", "unix");
1639 qdict_put_str(options
, "server.path", unixpath
);
1641 InetSocketAddress
*addr
= g_new(InetSocketAddress
, 1);
1643 if (inet_parse(addr
, host_spec
, errp
)) {
1647 qdict_put_str(options
, "server.type", "inet");
1648 qdict_put_str(options
, "server.host", addr
->host
);
1649 qdict_put_str(options
, "server.port", addr
->port
);
1651 qapi_free_InetSocketAddress(addr
);
1655 static bool nbd_process_legacy_socket_options(QDict
*output_options
,
1656 QemuOpts
*legacy_opts
,
1659 const char *path
= qemu_opt_get(legacy_opts
, "path");
1660 const char *host
= qemu_opt_get(legacy_opts
, "host");
1661 const char *port
= qemu_opt_get(legacy_opts
, "port");
1662 const QDictEntry
*e
;
1664 if (!path
&& !host
&& !port
) {
1668 for (e
= qdict_first(output_options
); e
; e
= qdict_next(output_options
, e
))
1670 if (strstart(e
->key
, "server.", NULL
)) {
1671 error_setg(errp
, "Cannot use 'server' and path/host/port at the "
1678 error_setg(errp
, "path and host may not be used at the same time");
1682 error_setg(errp
, "port may not be used without host");
1686 qdict_put_str(output_options
, "server.type", "unix");
1687 qdict_put_str(output_options
, "server.path", path
);
1689 qdict_put_str(output_options
, "server.type", "inet");
1690 qdict_put_str(output_options
, "server.host", host
);
1691 qdict_put_str(output_options
, "server.port",
1692 port
?: stringify(NBD_DEFAULT_PORT
));
1698 static SocketAddress
*nbd_config(BDRVNBDState
*s
, QDict
*options
,
1701 SocketAddress
*saddr
= NULL
;
1704 Error
*local_err
= NULL
;
1706 qdict_extract_subqdict(options
, &addr
, "server.");
1707 if (!qdict_size(addr
)) {
1708 error_setg(errp
, "NBD server address missing");
1712 iv
= qobject_input_visitor_new_flat_confused(addr
, errp
);
1717 visit_type_SocketAddress(iv
, NULL
, &saddr
, &local_err
);
1719 error_propagate(errp
, local_err
);
1724 qobject_unref(addr
);
1729 static QCryptoTLSCreds
*nbd_get_tls_creds(const char *id
, Error
**errp
)
1732 QCryptoTLSCreds
*creds
;
1734 obj
= object_resolve_path_component(
1735 object_get_objects_root(), id
);
1737 error_setg(errp
, "No TLS credentials with id '%s'",
1741 creds
= (QCryptoTLSCreds
*)
1742 object_dynamic_cast(obj
, TYPE_QCRYPTO_TLS_CREDS
);
1744 error_setg(errp
, "Object with id '%s' is not TLS credentials",
1749 if (creds
->endpoint
!= QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
) {
1751 "Expecting TLS credentials with a client endpoint");
1759 static QemuOptsList nbd_runtime_opts
= {
1761 .head
= QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts
.head
),
1765 .type
= QEMU_OPT_STRING
,
1766 .help
= "TCP host to connect to",
1770 .type
= QEMU_OPT_STRING
,
1771 .help
= "TCP port to connect to",
1775 .type
= QEMU_OPT_STRING
,
1776 .help
= "Unix socket path to connect to",
1780 .type
= QEMU_OPT_STRING
,
1781 .help
= "Name of the NBD export to open",
1784 .name
= "tls-creds",
1785 .type
= QEMU_OPT_STRING
,
1786 .help
= "ID of the TLS credentials to use",
1789 .name
= "x-dirty-bitmap",
1790 .type
= QEMU_OPT_STRING
,
1791 .help
= "experimental: expose named dirty bitmap in place of "
1795 .name
= "reconnect-delay",
1796 .type
= QEMU_OPT_NUMBER
,
1797 .help
= "On an unexpected disconnect, the nbd client tries to "
1798 "connect again until succeeding or encountering a serious "
1799 "error. During the first @reconnect-delay seconds, all "
1800 "requests are paused and will be rerun on a successful "
1801 "reconnect. After that time, any delayed requests and all "
1802 "future requests before a successful reconnect will "
1803 "immediately fail. Default 0",
1805 { /* end of list */ }
1809 static int nbd_process_options(BlockDriverState
*bs
, QDict
*options
,
1812 BDRVNBDState
*s
= bs
->opaque
;
1814 Error
*local_err
= NULL
;
1817 opts
= qemu_opts_create(&nbd_runtime_opts
, NULL
, 0, &error_abort
);
1818 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1820 error_propagate(errp
, local_err
);
1824 /* Translate @host, @port, and @path to a SocketAddress */
1825 if (!nbd_process_legacy_socket_options(options
, opts
, errp
)) {
1829 /* Pop the config into our state object. Exit if invalid. */
1830 s
->saddr
= nbd_config(s
, options
, errp
);
1835 s
->export
= g_strdup(qemu_opt_get(opts
, "export"));
1836 if (s
->export
&& strlen(s
->export
) > NBD_MAX_STRING_SIZE
) {
1837 error_setg(errp
, "export name too long to send to server");
1841 s
->tlscredsid
= g_strdup(qemu_opt_get(opts
, "tls-creds"));
1842 if (s
->tlscredsid
) {
1843 s
->tlscreds
= nbd_get_tls_creds(s
->tlscredsid
, errp
);
1848 /* TODO SOCKET_ADDRESS_KIND_FD where fd has AF_INET or AF_INET6 */
1849 if (s
->saddr
->type
!= SOCKET_ADDRESS_TYPE_INET
) {
1850 error_setg(errp
, "TLS only supported over IP sockets");
1853 s
->hostname
= s
->saddr
->u
.inet
.host
;
1856 s
->x_dirty_bitmap
= g_strdup(qemu_opt_get(opts
, "x-dirty-bitmap"));
1857 if (s
->x_dirty_bitmap
&& strlen(s
->x_dirty_bitmap
) > NBD_MAX_STRING_SIZE
) {
1858 error_setg(errp
, "x-dirty-bitmap query too long to send to server");
1862 s
->reconnect_delay
= qemu_opt_get_number(opts
, "reconnect-delay", 0);
1868 object_unref(OBJECT(s
->tlscreds
));
1869 qapi_free_SocketAddress(s
->saddr
);
1871 g_free(s
->tlscredsid
);
1872 g_free(s
->x_dirty_bitmap
);
1874 qemu_opts_del(opts
);
1878 static int nbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1882 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1884 ret
= nbd_process_options(bs
, options
, errp
);
1890 qemu_co_mutex_init(&s
->send_mutex
);
1891 qemu_co_queue_init(&s
->free_sema
);
1893 ret
= nbd_client_connect(bs
, errp
);
1897 /* successfully connected */
1898 s
->state
= NBD_CLIENT_CONNECTED
;
1900 s
->connection_co
= qemu_coroutine_create(nbd_connection_entry
, s
);
1901 bdrv_inc_in_flight(bs
);
1902 aio_co_schedule(bdrv_get_aio_context(bs
), s
->connection_co
);
1907 static int nbd_co_flush(BlockDriverState
*bs
)
1909 return nbd_client_co_flush(bs
);
1912 static void nbd_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1914 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1915 uint32_t min
= s
->info
.min_block
;
1916 uint32_t max
= MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE
, s
->info
.max_block
);
1919 * If the server did not advertise an alignment:
1920 * - a size that is not sector-aligned implies that an alignment
1921 * of 1 can be used to access those tail bytes
1922 * - advertisement of block status requires an alignment of 1, so
1923 * that we don't violate block layer constraints that block
1924 * status is always aligned (as we can't control whether the
1925 * server will report sub-sector extents, such as a hole at EOF
1926 * on an unaligned POSIX file)
1927 * - otherwise, assume the server is so old that we are safer avoiding
1928 * sub-sector requests
1931 min
= (!QEMU_IS_ALIGNED(s
->info
.size
, BDRV_SECTOR_SIZE
) ||
1932 s
->info
.base_allocation
) ? 1 : BDRV_SECTOR_SIZE
;
1935 bs
->bl
.request_alignment
= min
;
1936 bs
->bl
.max_pdiscard
= max
;
1937 bs
->bl
.max_pwrite_zeroes
= max
;
1938 bs
->bl
.max_transfer
= max
;
1940 if (s
->info
.opt_block
&&
1941 s
->info
.opt_block
> bs
->bl
.opt_transfer
) {
1942 bs
->bl
.opt_transfer
= s
->info
.opt_block
;
1946 static void nbd_close(BlockDriverState
*bs
)
1948 BDRVNBDState
*s
= bs
->opaque
;
1950 nbd_client_close(bs
);
1952 object_unref(OBJECT(s
->tlscreds
));
1953 qapi_free_SocketAddress(s
->saddr
);
1955 g_free(s
->tlscredsid
);
1956 g_free(s
->x_dirty_bitmap
);
1959 static int64_t nbd_getlength(BlockDriverState
*bs
)
1961 BDRVNBDState
*s
= bs
->opaque
;
1963 return s
->info
.size
;
1966 static void nbd_refresh_filename(BlockDriverState
*bs
)
1968 BDRVNBDState
*s
= bs
->opaque
;
1969 const char *host
= NULL
, *port
= NULL
, *path
= NULL
;
1971 if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
1972 const InetSocketAddress
*inet
= &s
->saddr
->u
.inet
;
1973 if (!inet
->has_ipv4
&& !inet
->has_ipv6
&& !inet
->has_to
) {
1977 } else if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_UNIX
) {
1978 path
= s
->saddr
->u
.q_unix
.path
;
1979 } /* else can't represent as pseudo-filename */
1981 if (path
&& s
->export
) {
1982 snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
1983 "nbd+unix:///%s?socket=%s", s
->export
, path
);
1984 } else if (path
&& !s
->export
) {
1985 snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
1986 "nbd+unix://?socket=%s", path
);
1987 } else if (host
&& s
->export
) {
1988 snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
1989 "nbd://%s:%s/%s", host
, port
, s
->export
);
1990 } else if (host
&& !s
->export
) {
1991 snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
1992 "nbd://%s:%s", host
, port
);
1996 static char *nbd_dirname(BlockDriverState
*bs
, Error
**errp
)
1998 /* The generic bdrv_dirname() implementation is able to work out some
1999 * directory name for NBD nodes, but that would be wrong. So far there is no
2000 * specification for how "export paths" would work, so NBD does not have
2001 * directory names. */
2002 error_setg(errp
, "Cannot generate a base directory for NBD nodes");
2006 static const char *const nbd_strong_runtime_opts
[] = {
2017 static BlockDriver bdrv_nbd
= {
2018 .format_name
= "nbd",
2019 .protocol_name
= "nbd",
2020 .instance_size
= sizeof(BDRVNBDState
),
2021 .bdrv_parse_filename
= nbd_parse_filename
,
2022 .bdrv_file_open
= nbd_open
,
2023 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2024 .bdrv_co_preadv
= nbd_client_co_preadv
,
2025 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2026 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2027 .bdrv_close
= nbd_close
,
2028 .bdrv_co_flush_to_os
= nbd_co_flush
,
2029 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2030 .bdrv_refresh_limits
= nbd_refresh_limits
,
2031 .bdrv_getlength
= nbd_getlength
,
2032 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2033 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2034 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2035 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2036 .bdrv_refresh_filename
= nbd_refresh_filename
,
2037 .bdrv_co_block_status
= nbd_client_co_block_status
,
2038 .bdrv_dirname
= nbd_dirname
,
2039 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2042 static BlockDriver bdrv_nbd_tcp
= {
2043 .format_name
= "nbd",
2044 .protocol_name
= "nbd+tcp",
2045 .instance_size
= sizeof(BDRVNBDState
),
2046 .bdrv_parse_filename
= nbd_parse_filename
,
2047 .bdrv_file_open
= nbd_open
,
2048 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2049 .bdrv_co_preadv
= nbd_client_co_preadv
,
2050 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2051 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2052 .bdrv_close
= nbd_close
,
2053 .bdrv_co_flush_to_os
= nbd_co_flush
,
2054 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2055 .bdrv_refresh_limits
= nbd_refresh_limits
,
2056 .bdrv_getlength
= nbd_getlength
,
2057 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2058 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2059 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2060 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2061 .bdrv_refresh_filename
= nbd_refresh_filename
,
2062 .bdrv_co_block_status
= nbd_client_co_block_status
,
2063 .bdrv_dirname
= nbd_dirname
,
2064 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2067 static BlockDriver bdrv_nbd_unix
= {
2068 .format_name
= "nbd",
2069 .protocol_name
= "nbd+unix",
2070 .instance_size
= sizeof(BDRVNBDState
),
2071 .bdrv_parse_filename
= nbd_parse_filename
,
2072 .bdrv_file_open
= nbd_open
,
2073 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2074 .bdrv_co_preadv
= nbd_client_co_preadv
,
2075 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2076 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2077 .bdrv_close
= nbd_close
,
2078 .bdrv_co_flush_to_os
= nbd_co_flush
,
2079 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2080 .bdrv_refresh_limits
= nbd_refresh_limits
,
2081 .bdrv_getlength
= nbd_getlength
,
2082 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2083 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2084 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2085 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2086 .bdrv_refresh_filename
= nbd_refresh_filename
,
2087 .bdrv_co_block_status
= nbd_client_co_block_status
,
2088 .bdrv_dirname
= nbd_dirname
,
2089 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2092 static void bdrv_nbd_init(void)
2094 bdrv_register(&bdrv_nbd
);
2095 bdrv_register(&bdrv_nbd_tcp
);
2096 bdrv_register(&bdrv_nbd_unix
);
2099 block_init(bdrv_nbd_init
);