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"
38 #include "qemu/atomic.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/clone-visitor.h"
44 #include "block/qdict.h"
45 #include "block/nbd.h"
46 #include "block/block_int.h"
47 #include "block/coroutines.h"
49 #include "qemu/yank.h"
51 #define EN_OPTSTR ":exportname="
52 #define MAX_NBD_REQUESTS 16
54 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
55 #define INDEX_TO_HANDLE(bs, index) ((index) ^ (uint64_t)(intptr_t)(bs))
59 uint64_t offset
; /* original offset of the request */
60 bool receiving
; /* sleeping in the yield in nbd_receive_replies */
63 typedef enum NBDClientState
{
64 NBD_CLIENT_CONNECTING_WAIT
,
65 NBD_CLIENT_CONNECTING_NOWAIT
,
70 typedef struct BDRVNBDState
{
71 QIOChannel
*ioc
; /* The current I/O channel */
77 CoMutex receive_mutex
;
81 QEMUTimer
*reconnect_delay_timer
;
82 QEMUTimer
*open_timer
;
84 NBDClientRequest requests
[MAX_NBD_REQUESTS
];
88 /* Connection parameters */
89 uint32_t reconnect_delay
;
90 uint32_t open_timeout
;
94 QCryptoTLSCreds
*tlscreds
;
99 NBDClientConnection
*conn
;
102 static void nbd_yank(void *opaque
);
104 static void nbd_clear_bdrvstate(BlockDriverState
*bs
)
106 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
108 nbd_client_connection_release(s
->conn
);
111 yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
));
113 /* Must not leave timers behind that would access freed data */
114 assert(!s
->reconnect_delay_timer
);
115 assert(!s
->open_timer
);
117 object_unref(OBJECT(s
->tlscreds
));
118 qapi_free_SocketAddress(s
->saddr
);
122 g_free(s
->tlscredsid
);
123 s
->tlscredsid
= NULL
;
124 g_free(s
->tlshostname
);
125 s
->tlshostname
= NULL
;
126 g_free(s
->x_dirty_bitmap
);
127 s
->x_dirty_bitmap
= NULL
;
130 static bool nbd_client_connected(BDRVNBDState
*s
)
132 return qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTED
;
135 static bool coroutine_fn
nbd_recv_coroutine_wake_one(NBDClientRequest
*req
)
137 if (req
->receiving
) {
138 req
->receiving
= false;
139 aio_co_wake(req
->coroutine
);
146 static void coroutine_fn
nbd_recv_coroutines_wake(BDRVNBDState
*s
, bool all
)
150 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
151 if (nbd_recv_coroutine_wake_one(&s
->requests
[i
]) && !all
) {
157 static void coroutine_fn
nbd_channel_error(BDRVNBDState
*s
, int ret
)
159 if (nbd_client_connected(s
)) {
160 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
164 if (nbd_client_connected(s
)) {
165 s
->state
= s
->reconnect_delay
? NBD_CLIENT_CONNECTING_WAIT
:
166 NBD_CLIENT_CONNECTING_NOWAIT
;
169 s
->state
= NBD_CLIENT_QUIT
;
172 nbd_recv_coroutines_wake(s
, true);
175 static void reconnect_delay_timer_del(BDRVNBDState
*s
)
177 if (s
->reconnect_delay_timer
) {
178 timer_free(s
->reconnect_delay_timer
);
179 s
->reconnect_delay_timer
= NULL
;
183 static void reconnect_delay_timer_cb(void *opaque
)
185 BDRVNBDState
*s
= opaque
;
187 if (qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
) {
188 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
189 nbd_co_establish_connection_cancel(s
->conn
);
190 while (qemu_co_enter_next(&s
->free_sema
, NULL
)) {
191 /* Resume all queued requests */
195 reconnect_delay_timer_del(s
);
198 static void reconnect_delay_timer_init(BDRVNBDState
*s
, uint64_t expire_time_ns
)
200 if (qatomic_load_acquire(&s
->state
) != NBD_CLIENT_CONNECTING_WAIT
) {
204 assert(!s
->reconnect_delay_timer
);
205 s
->reconnect_delay_timer
= aio_timer_new(bdrv_get_aio_context(s
->bs
),
208 reconnect_delay_timer_cb
, s
);
209 timer_mod(s
->reconnect_delay_timer
, expire_time_ns
);
212 static void nbd_teardown_connection(BlockDriverState
*bs
)
214 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
216 assert(!s
->in_flight
);
219 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
220 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
222 object_unref(OBJECT(s
->ioc
));
226 s
->state
= NBD_CLIENT_QUIT
;
229 static void open_timer_del(BDRVNBDState
*s
)
232 timer_free(s
->open_timer
);
233 s
->open_timer
= NULL
;
237 static void open_timer_cb(void *opaque
)
239 BDRVNBDState
*s
= opaque
;
241 nbd_co_establish_connection_cancel(s
->conn
);
245 static void open_timer_init(BDRVNBDState
*s
, uint64_t expire_time_ns
)
247 assert(!s
->open_timer
);
248 s
->open_timer
= aio_timer_new(bdrv_get_aio_context(s
->bs
),
252 timer_mod(s
->open_timer
, expire_time_ns
);
255 static bool nbd_client_connecting(BDRVNBDState
*s
)
257 NBDClientState state
= qatomic_load_acquire(&s
->state
);
258 return state
== NBD_CLIENT_CONNECTING_WAIT
||
259 state
== NBD_CLIENT_CONNECTING_NOWAIT
;
262 static bool nbd_client_connecting_wait(BDRVNBDState
*s
)
264 return qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
;
268 * Update @bs with information learned during a completed negotiation process.
269 * Return failure if the server's advertised options are incompatible with the
272 static int nbd_handle_updated_info(BlockDriverState
*bs
, Error
**errp
)
274 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
277 if (s
->x_dirty_bitmap
) {
278 if (!s
->info
.base_allocation
) {
279 error_setg(errp
, "requested x-dirty-bitmap %s not found",
283 if (strcmp(s
->x_dirty_bitmap
, "qemu:allocation-depth") == 0) {
284 s
->alloc_depth
= true;
288 if (s
->info
.flags
& NBD_FLAG_READ_ONLY
) {
289 ret
= bdrv_apply_auto_read_only(bs
, "NBD export is read-only", errp
);
295 if (s
->info
.flags
& NBD_FLAG_SEND_FUA
) {
296 bs
->supported_write_flags
= BDRV_REQ_FUA
;
297 bs
->supported_zero_flags
|= BDRV_REQ_FUA
;
300 if (s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
) {
301 bs
->supported_zero_flags
|= BDRV_REQ_MAY_UNMAP
;
302 if (s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
) {
303 bs
->supported_zero_flags
|= BDRV_REQ_NO_FALLBACK
;
307 trace_nbd_client_handshake_success(s
->export
);
312 int coroutine_fn
nbd_co_do_establish_connection(BlockDriverState
*bs
,
315 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
317 bool blocking
= nbd_client_connecting_wait(s
);
322 s
->ioc
= nbd_co_establish_connection(s
->conn
, &s
->info
, blocking
, errp
);
324 return -ECONNREFUSED
;
327 yank_register_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
), nbd_yank
,
330 ret
= nbd_handle_updated_info(s
->bs
, NULL
);
333 * We have connected, but must fail for other reasons.
334 * Send NBD_CMD_DISC as a courtesy to the server.
336 NBDRequest request
= { .type
= NBD_CMD_DISC
};
338 nbd_send_request(s
->ioc
, &request
);
340 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
342 object_unref(OBJECT(s
->ioc
));
348 qio_channel_set_blocking(s
->ioc
, false, NULL
);
349 qio_channel_attach_aio_context(s
->ioc
, bdrv_get_aio_context(bs
));
351 /* successfully connected */
352 s
->state
= NBD_CLIENT_CONNECTED
;
353 qemu_co_queue_restart_all(&s
->free_sema
);
358 /* called under s->send_mutex */
359 static coroutine_fn
void nbd_reconnect_attempt(BDRVNBDState
*s
)
361 assert(nbd_client_connecting(s
));
362 assert(s
->in_flight
== 0);
364 if (nbd_client_connecting_wait(s
) && s
->reconnect_delay
&&
365 !s
->reconnect_delay_timer
)
368 * It's first reconnect attempt after switching to
369 * NBD_CLIENT_CONNECTING_WAIT
371 reconnect_delay_timer_init(s
,
372 qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) +
373 s
->reconnect_delay
* NANOSECONDS_PER_SECOND
);
377 * Now we are sure that nobody is accessing the channel, and no one will
378 * try until we set the state to CONNECTED.
381 /* Finalize previous connection if any */
383 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
384 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
386 object_unref(OBJECT(s
->ioc
));
390 nbd_co_do_establish_connection(s
->bs
, NULL
);
393 * The reconnect attempt is done (maybe successfully, maybe not), so
394 * we no longer need this timer. Delete it so it will not outlive
395 * this I/O request (so draining removes all timers).
397 reconnect_delay_timer_del(s
);
400 static coroutine_fn
int nbd_receive_replies(BDRVNBDState
*s
, uint64_t handle
)
403 uint64_t ind
= HANDLE_TO_INDEX(s
, handle
), ind2
;
404 QEMU_LOCK_GUARD(&s
->receive_mutex
);
407 if (s
->reply
.handle
== handle
) {
412 if (s
->reply
.handle
!= 0) {
414 * Some other request is being handled now. It should already be
415 * woken by whoever set s->reply.handle (or never wait in this
416 * yield). So, we should not wake it here.
418 ind2
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
419 assert(!s
->requests
[ind2
].receiving
);
421 s
->requests
[ind
].receiving
= true;
422 qemu_co_mutex_unlock(&s
->receive_mutex
);
424 qemu_coroutine_yield();
426 * We may be woken for 3 reasons:
427 * 1. From this function, executing in parallel coroutine, when our
428 * handle is received.
429 * 2. From nbd_channel_error(), when connection is lost.
430 * 3. From nbd_co_receive_one_chunk(), when previous request is
431 * finished and s->reply.handle set to 0.
432 * Anyway, it's OK to lock the mutex and go to the next iteration.
435 qemu_co_mutex_lock(&s
->receive_mutex
);
436 assert(!s
->requests
[ind
].receiving
);
440 /* We are under mutex and handle is 0. We have to do the dirty work. */
441 assert(s
->reply
.handle
== 0);
442 ret
= nbd_receive_reply(s
->bs
, s
->ioc
, &s
->reply
, NULL
);
444 ret
= ret
? ret
: -EIO
;
445 nbd_channel_error(s
, ret
);
448 if (nbd_reply_is_structured(&s
->reply
) && !s
->info
.structured_reply
) {
449 nbd_channel_error(s
, -EINVAL
);
452 ind2
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
453 if (ind2
>= MAX_NBD_REQUESTS
|| !s
->requests
[ind2
].coroutine
) {
454 nbd_channel_error(s
, -EINVAL
);
457 if (s
->reply
.handle
== handle
) {
461 nbd_recv_coroutine_wake_one(&s
->requests
[ind2
]);
465 static int coroutine_fn
nbd_co_send_request(BlockDriverState
*bs
,
469 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
472 qemu_co_mutex_lock(&s
->send_mutex
);
474 while (s
->in_flight
== MAX_NBD_REQUESTS
||
475 (!nbd_client_connected(s
) && s
->in_flight
> 0))
477 qemu_co_queue_wait(&s
->free_sema
, &s
->send_mutex
);
480 if (nbd_client_connecting(s
)) {
481 nbd_reconnect_attempt(s
);
484 if (!nbd_client_connected(s
)) {
491 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
492 if (s
->requests
[i
].coroutine
== NULL
) {
497 g_assert(qemu_in_coroutine());
498 assert(i
< MAX_NBD_REQUESTS
);
500 s
->requests
[i
].coroutine
= qemu_coroutine_self();
501 s
->requests
[i
].offset
= request
->from
;
502 s
->requests
[i
].receiving
= false;
504 request
->handle
= INDEX_TO_HANDLE(s
, i
);
509 qio_channel_set_cork(s
->ioc
, true);
510 rc
= nbd_send_request(s
->ioc
, request
);
512 if (qio_channel_writev_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
516 } else if (rc
>= 0) {
519 qio_channel_set_cork(s
->ioc
, false);
521 rc
= nbd_send_request(s
->ioc
, request
);
526 nbd_channel_error(s
, rc
);
528 s
->requests
[i
].coroutine
= NULL
;
531 qemu_co_queue_next(&s
->free_sema
);
533 qemu_co_mutex_unlock(&s
->send_mutex
);
537 static inline uint16_t payload_advance16(uint8_t **payload
)
540 return lduw_be_p(*payload
- 2);
543 static inline uint32_t payload_advance32(uint8_t **payload
)
546 return ldl_be_p(*payload
- 4);
549 static inline uint64_t payload_advance64(uint8_t **payload
)
552 return ldq_be_p(*payload
- 8);
555 static int nbd_parse_offset_hole_payload(BDRVNBDState
*s
,
556 NBDStructuredReplyChunk
*chunk
,
557 uint8_t *payload
, uint64_t orig_offset
,
558 QEMUIOVector
*qiov
, Error
**errp
)
563 if (chunk
->length
!= sizeof(offset
) + sizeof(hole_size
)) {
564 error_setg(errp
, "Protocol error: invalid payload for "
565 "NBD_REPLY_TYPE_OFFSET_HOLE");
569 offset
= payload_advance64(&payload
);
570 hole_size
= payload_advance32(&payload
);
572 if (!hole_size
|| offset
< orig_offset
|| hole_size
> qiov
->size
||
573 offset
> orig_offset
+ qiov
->size
- hole_size
) {
574 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
578 if (s
->info
.min_block
&&
579 !QEMU_IS_ALIGNED(hole_size
, s
->info
.min_block
)) {
580 trace_nbd_structured_read_compliance("hole");
583 qemu_iovec_memset(qiov
, offset
- orig_offset
, 0, hole_size
);
589 * nbd_parse_blockstatus_payload
590 * Based on our request, we expect only one extent in reply, for the
591 * base:allocation context.
593 static int nbd_parse_blockstatus_payload(BDRVNBDState
*s
,
594 NBDStructuredReplyChunk
*chunk
,
595 uint8_t *payload
, uint64_t orig_length
,
596 NBDExtent
*extent
, Error
**errp
)
600 /* The server succeeded, so it must have sent [at least] one extent */
601 if (chunk
->length
< sizeof(context_id
) + sizeof(*extent
)) {
602 error_setg(errp
, "Protocol error: invalid payload for "
603 "NBD_REPLY_TYPE_BLOCK_STATUS");
607 context_id
= payload_advance32(&payload
);
608 if (s
->info
.context_id
!= context_id
) {
609 error_setg(errp
, "Protocol error: unexpected context id %d for "
610 "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
611 "id is %d", context_id
,
616 extent
->length
= payload_advance32(&payload
);
617 extent
->flags
= payload_advance32(&payload
);
619 if (extent
->length
== 0) {
620 error_setg(errp
, "Protocol error: server sent status chunk with "
626 * A server sending unaligned block status is in violation of the
627 * protocol, but as qemu-nbd 3.1 is such a server (at least for
628 * POSIX files that are not a multiple of 512 bytes, since qemu
629 * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
630 * still sees an implicit hole beyond the real EOF), it's nicer to
631 * work around the misbehaving server. If the request included
632 * more than the final unaligned block, truncate it back to an
633 * aligned result; if the request was only the final block, round
634 * up to the full block and change the status to fully-allocated
635 * (always a safe status, even if it loses information).
637 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(extent
->length
,
638 s
->info
.min_block
)) {
639 trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
640 if (extent
->length
> s
->info
.min_block
) {
641 extent
->length
= QEMU_ALIGN_DOWN(extent
->length
,
644 extent
->length
= s
->info
.min_block
;
650 * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
651 * sent us any more than one extent, nor should it have included
652 * status beyond our request in that extent. However, it's easy
653 * enough to ignore the server's noncompliance without killing the
654 * connection; just ignore trailing extents, and clamp things to
655 * the length of our request.
657 if (chunk
->length
> sizeof(context_id
) + sizeof(*extent
)) {
658 trace_nbd_parse_blockstatus_compliance("more than one extent");
660 if (extent
->length
> orig_length
) {
661 extent
->length
= orig_length
;
662 trace_nbd_parse_blockstatus_compliance("extent length too large");
666 * HACK: if we are using x-dirty-bitmaps to access
667 * qemu:allocation-depth, treat all depths > 2 the same as 2,
668 * since nbd_client_co_block_status is only expecting the low two
671 if (s
->alloc_depth
&& extent
->flags
> 2) {
679 * nbd_parse_error_payload
680 * on success @errp contains message describing nbd error reply
682 static int nbd_parse_error_payload(NBDStructuredReplyChunk
*chunk
,
683 uint8_t *payload
, int *request_ret
,
687 uint16_t message_size
;
689 assert(chunk
->type
& (1 << 15));
691 if (chunk
->length
< sizeof(error
) + sizeof(message_size
)) {
693 "Protocol error: invalid payload for structured error");
697 error
= nbd_errno_to_system_errno(payload_advance32(&payload
));
699 error_setg(errp
, "Protocol error: server sent structured error chunk "
704 *request_ret
= -error
;
705 message_size
= payload_advance16(&payload
);
707 if (message_size
> chunk
->length
- sizeof(error
) - sizeof(message_size
)) {
708 error_setg(errp
, "Protocol error: server sent structured error chunk "
709 "with incorrect message size");
713 /* TODO: Add a trace point to mention the server complaint */
715 /* TODO handle ERROR_OFFSET */
720 static int coroutine_fn
721 nbd_co_receive_offset_data_payload(BDRVNBDState
*s
, uint64_t orig_offset
,
722 QEMUIOVector
*qiov
, Error
**errp
)
724 QEMUIOVector sub_qiov
;
728 NBDStructuredReplyChunk
*chunk
= &s
->reply
.structured
;
730 assert(nbd_reply_is_structured(&s
->reply
));
732 /* The NBD spec requires at least one byte of payload */
733 if (chunk
->length
<= sizeof(offset
)) {
734 error_setg(errp
, "Protocol error: invalid payload for "
735 "NBD_REPLY_TYPE_OFFSET_DATA");
739 if (nbd_read64(s
->ioc
, &offset
, "OFFSET_DATA offset", errp
) < 0) {
743 data_size
= chunk
->length
- sizeof(offset
);
745 if (offset
< orig_offset
|| data_size
> qiov
->size
||
746 offset
> orig_offset
+ qiov
->size
- data_size
) {
747 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
751 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(data_size
, s
->info
.min_block
)) {
752 trace_nbd_structured_read_compliance("data");
755 qemu_iovec_init(&sub_qiov
, qiov
->niov
);
756 qemu_iovec_concat(&sub_qiov
, qiov
, offset
- orig_offset
, data_size
);
757 ret
= qio_channel_readv_all(s
->ioc
, sub_qiov
.iov
, sub_qiov
.niov
, errp
);
758 qemu_iovec_destroy(&sub_qiov
);
760 return ret
< 0 ? -EIO
: 0;
763 #define NBD_MAX_MALLOC_PAYLOAD 1000
764 static coroutine_fn
int nbd_co_receive_structured_payload(
765 BDRVNBDState
*s
, void **payload
, Error
**errp
)
770 assert(nbd_reply_is_structured(&s
->reply
));
772 len
= s
->reply
.structured
.length
;
778 if (payload
== NULL
) {
779 error_setg(errp
, "Unexpected structured payload");
783 if (len
> NBD_MAX_MALLOC_PAYLOAD
) {
784 error_setg(errp
, "Payload too large");
788 *payload
= g_new(char, len
);
789 ret
= nbd_read(s
->ioc
, *payload
, len
, "structured payload", errp
);
800 * nbd_co_do_receive_one_chunk
802 * set request_ret to received reply error
803 * if qiov is not NULL: read payload to @qiov
804 * for structured reply chunk:
805 * if error chunk: read payload, set @request_ret, do not set @payload
806 * else if offset_data chunk: read payload data to @qiov, do not set @payload
807 * else: read payload to @payload
809 * If function fails, @errp contains corresponding error message, and the
810 * connection with the server is suspect. If it returns 0, then the
811 * transaction succeeded (although @request_ret may be a negative errno
812 * corresponding to the server's error reply), and errp is unchanged.
814 static coroutine_fn
int nbd_co_do_receive_one_chunk(
815 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
816 int *request_ret
, QEMUIOVector
*qiov
, void **payload
, Error
**errp
)
819 int i
= HANDLE_TO_INDEX(s
, handle
);
820 void *local_payload
= NULL
;
821 NBDStructuredReplyChunk
*chunk
;
828 ret
= nbd_receive_replies(s
, handle
);
830 error_setg(errp
, "Connection closed");
835 assert(s
->reply
.handle
== handle
);
837 if (nbd_reply_is_simple(&s
->reply
)) {
838 if (only_structured
) {
839 error_setg(errp
, "Protocol error: simple reply when structured "
840 "reply chunk was expected");
844 *request_ret
= -nbd_errno_to_system_errno(s
->reply
.simple
.error
);
845 if (*request_ret
< 0 || !qiov
) {
849 return qio_channel_readv_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
850 errp
) < 0 ? -EIO
: 0;
853 /* handle structured reply chunk */
854 assert(s
->info
.structured_reply
);
855 chunk
= &s
->reply
.structured
;
857 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
858 if (!(chunk
->flags
& NBD_REPLY_FLAG_DONE
)) {
859 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
860 " NBD_REPLY_FLAG_DONE flag set");
864 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
871 if (chunk
->type
== NBD_REPLY_TYPE_OFFSET_DATA
) {
873 error_setg(errp
, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
877 return nbd_co_receive_offset_data_payload(s
, s
->requests
[i
].offset
,
881 if (nbd_reply_type_is_error(chunk
->type
)) {
882 payload
= &local_payload
;
885 ret
= nbd_co_receive_structured_payload(s
, payload
, errp
);
890 if (nbd_reply_type_is_error(chunk
->type
)) {
891 ret
= nbd_parse_error_payload(chunk
, local_payload
, request_ret
, errp
);
892 g_free(local_payload
);
900 * nbd_co_receive_one_chunk
901 * Read reply, wake up connection_co and set s->quit if needed.
902 * Return value is a fatal error code or normal nbd reply error code
904 static coroutine_fn
int nbd_co_receive_one_chunk(
905 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
906 int *request_ret
, QEMUIOVector
*qiov
, NBDReply
*reply
, void **payload
,
909 int ret
= nbd_co_do_receive_one_chunk(s
, handle
, only_structured
,
910 request_ret
, qiov
, payload
, errp
);
913 memset(reply
, 0, sizeof(*reply
));
914 nbd_channel_error(s
, ret
);
916 /* For assert at loop start in nbd_connection_entry */
921 nbd_recv_coroutines_wake(s
, false);
926 typedef struct NBDReplyChunkIter
{
930 bool done
, only_structured
;
933 static void nbd_iter_channel_error(NBDReplyChunkIter
*iter
,
934 int ret
, Error
**local_err
)
936 assert(local_err
&& *local_err
);
941 error_propagate(&iter
->err
, *local_err
);
943 error_free(*local_err
);
949 static void nbd_iter_request_error(NBDReplyChunkIter
*iter
, int ret
)
953 if (!iter
->request_ret
) {
954 iter
->request_ret
= ret
;
959 * NBD_FOREACH_REPLY_CHUNK
960 * The pointer stored in @payload requires g_free() to free it.
962 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
963 qiov, reply, payload) \
964 for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
965 nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
968 * nbd_reply_chunk_iter_receive
969 * The pointer stored in @payload requires g_free() to free it.
971 static bool nbd_reply_chunk_iter_receive(BDRVNBDState
*s
,
972 NBDReplyChunkIter
*iter
,
974 QEMUIOVector
*qiov
, NBDReply
*reply
,
977 int ret
, request_ret
;
978 NBDReply local_reply
;
979 NBDStructuredReplyChunk
*chunk
;
980 Error
*local_err
= NULL
;
983 /* Previous iteration was last. */
988 reply
= &local_reply
;
991 ret
= nbd_co_receive_one_chunk(s
, handle
, iter
->only_structured
,
992 &request_ret
, qiov
, reply
, payload
,
995 nbd_iter_channel_error(iter
, ret
, &local_err
);
996 } else if (request_ret
< 0) {
997 nbd_iter_request_error(iter
, request_ret
);
1000 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1001 if (nbd_reply_is_simple(reply
) || iter
->ret
< 0) {
1005 chunk
= &reply
->structured
;
1006 iter
->only_structured
= true;
1008 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
1009 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1010 assert(chunk
->flags
& NBD_REPLY_FLAG_DONE
);
1014 if (chunk
->flags
& NBD_REPLY_FLAG_DONE
) {
1015 /* This iteration is last. */
1019 /* Execute the loop body */
1023 s
->requests
[HANDLE_TO_INDEX(s
, handle
)].coroutine
= NULL
;
1025 qemu_co_mutex_lock(&s
->send_mutex
);
1027 qemu_co_queue_next(&s
->free_sema
);
1028 qemu_co_mutex_unlock(&s
->send_mutex
);
1033 static int coroutine_fn
nbd_co_receive_return_code(BDRVNBDState
*s
, uint64_t handle
,
1034 int *request_ret
, Error
**errp
)
1036 NBDReplyChunkIter iter
;
1038 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, NULL
, NULL
) {
1039 /* nbd_reply_chunk_iter_receive does all the work */
1042 error_propagate(errp
, iter
.err
);
1043 *request_ret
= iter
.request_ret
;
1047 static int coroutine_fn
nbd_co_receive_cmdread_reply(BDRVNBDState
*s
, uint64_t handle
,
1048 uint64_t offset
, QEMUIOVector
*qiov
,
1049 int *request_ret
, Error
**errp
)
1051 NBDReplyChunkIter iter
;
1053 void *payload
= NULL
;
1054 Error
*local_err
= NULL
;
1056 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, s
->info
.structured_reply
,
1057 qiov
, &reply
, &payload
)
1060 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1062 assert(nbd_reply_is_structured(&reply
));
1064 switch (chunk
->type
) {
1065 case NBD_REPLY_TYPE_OFFSET_DATA
:
1067 * special cased in nbd_co_receive_one_chunk, data is already
1071 case NBD_REPLY_TYPE_OFFSET_HOLE
:
1072 ret
= nbd_parse_offset_hole_payload(s
, &reply
.structured
, payload
,
1073 offset
, qiov
, &local_err
);
1075 nbd_channel_error(s
, ret
);
1076 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1080 if (!nbd_reply_type_is_error(chunk
->type
)) {
1081 /* not allowed reply type */
1082 nbd_channel_error(s
, -EINVAL
);
1083 error_setg(&local_err
,
1084 "Unexpected reply type: %d (%s) for CMD_READ",
1085 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1086 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1094 error_propagate(errp
, iter
.err
);
1095 *request_ret
= iter
.request_ret
;
1099 static int coroutine_fn
nbd_co_receive_blockstatus_reply(BDRVNBDState
*s
,
1100 uint64_t handle
, uint64_t length
,
1102 int *request_ret
, Error
**errp
)
1104 NBDReplyChunkIter iter
;
1106 void *payload
= NULL
;
1107 Error
*local_err
= NULL
;
1108 bool received
= false;
1110 assert(!extent
->length
);
1111 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, &reply
, &payload
) {
1113 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1115 assert(nbd_reply_is_structured(&reply
));
1117 switch (chunk
->type
) {
1118 case NBD_REPLY_TYPE_BLOCK_STATUS
:
1120 nbd_channel_error(s
, -EINVAL
);
1121 error_setg(&local_err
, "Several BLOCK_STATUS chunks in reply");
1122 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1126 ret
= nbd_parse_blockstatus_payload(s
, &reply
.structured
,
1127 payload
, length
, extent
,
1130 nbd_channel_error(s
, ret
);
1131 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1135 if (!nbd_reply_type_is_error(chunk
->type
)) {
1136 nbd_channel_error(s
, -EINVAL
);
1137 error_setg(&local_err
,
1138 "Unexpected reply type: %d (%s) "
1139 "for CMD_BLOCK_STATUS",
1140 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1141 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1149 if (!extent
->length
&& !iter
.request_ret
) {
1150 error_setg(&local_err
, "Server did not reply with any status extents");
1151 nbd_iter_channel_error(&iter
, -EIO
, &local_err
);
1154 error_propagate(errp
, iter
.err
);
1155 *request_ret
= iter
.request_ret
;
1159 static int coroutine_fn
nbd_co_request(BlockDriverState
*bs
, NBDRequest
*request
,
1160 QEMUIOVector
*write_qiov
)
1162 int ret
, request_ret
;
1163 Error
*local_err
= NULL
;
1164 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1166 assert(request
->type
!= NBD_CMD_READ
);
1168 assert(request
->type
== NBD_CMD_WRITE
);
1169 assert(request
->len
== iov_size(write_qiov
->iov
, write_qiov
->niov
));
1171 assert(request
->type
!= NBD_CMD_WRITE
);
1175 ret
= nbd_co_send_request(bs
, request
, write_qiov
);
1180 ret
= nbd_co_receive_return_code(s
, request
->handle
,
1181 &request_ret
, &local_err
);
1183 trace_nbd_co_request_fail(request
->from
, request
->len
,
1184 request
->handle
, request
->flags
,
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 coroutine_fn
nbd_client_co_preadv(BlockDriverState
*bs
, int64_t offset
,
1197 int64_t bytes
, QEMUIOVector
*qiov
,
1198 BdrvRequestFlags flags
)
1200 int ret
, request_ret
;
1201 Error
*local_err
= NULL
;
1202 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1203 NBDRequest request
= {
1204 .type
= NBD_CMD_READ
,
1209 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1216 * Work around the fact that the block layer doesn't do
1217 * byte-accurate sizing yet - if the read exceeds the server's
1218 * advertised size because the block layer rounded size up, then
1219 * truncate the request to the server and tail-pad with zero.
1221 if (offset
>= s
->info
.size
) {
1222 assert(bytes
< BDRV_SECTOR_SIZE
);
1223 qemu_iovec_memset(qiov
, 0, 0, bytes
);
1226 if (offset
+ bytes
> s
->info
.size
) {
1227 uint64_t slop
= offset
+ bytes
- s
->info
.size
;
1229 assert(slop
< BDRV_SECTOR_SIZE
);
1230 qemu_iovec_memset(qiov
, bytes
- slop
, 0, slop
);
1231 request
.len
-= slop
;
1235 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1240 ret
= nbd_co_receive_cmdread_reply(s
, request
.handle
, offset
, qiov
,
1241 &request_ret
, &local_err
);
1243 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1244 request
.flags
, request
.type
,
1245 nbd_cmd_lookup(request
.type
),
1246 ret
, error_get_pretty(local_err
));
1247 error_free(local_err
);
1250 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1252 return ret
? ret
: request_ret
;
1255 static int coroutine_fn
nbd_client_co_pwritev(BlockDriverState
*bs
, int64_t offset
,
1256 int64_t bytes
, QEMUIOVector
*qiov
,
1257 BdrvRequestFlags flags
)
1259 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1260 NBDRequest request
= {
1261 .type
= NBD_CMD_WRITE
,
1266 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1267 if (flags
& BDRV_REQ_FUA
) {
1268 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1269 request
.flags
|= NBD_CMD_FLAG_FUA
;
1272 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1277 return nbd_co_request(bs
, &request
, qiov
);
1280 static int coroutine_fn
nbd_client_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1281 int64_t bytes
, BdrvRequestFlags flags
)
1283 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1284 NBDRequest request
= {
1285 .type
= NBD_CMD_WRITE_ZEROES
,
1287 .len
= bytes
, /* .len is uint32_t actually */
1290 assert(bytes
<= UINT32_MAX
); /* rely on max_pwrite_zeroes */
1292 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1293 if (!(s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
)) {
1297 if (flags
& BDRV_REQ_FUA
) {
1298 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1299 request
.flags
|= NBD_CMD_FLAG_FUA
;
1301 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1302 request
.flags
|= NBD_CMD_FLAG_NO_HOLE
;
1304 if (flags
& BDRV_REQ_NO_FALLBACK
) {
1305 assert(s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
);
1306 request
.flags
|= NBD_CMD_FLAG_FAST_ZERO
;
1312 return nbd_co_request(bs
, &request
, NULL
);
1315 static int coroutine_fn
nbd_client_co_flush(BlockDriverState
*bs
)
1317 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1318 NBDRequest request
= { .type
= NBD_CMD_FLUSH
};
1320 if (!(s
->info
.flags
& NBD_FLAG_SEND_FLUSH
)) {
1327 return nbd_co_request(bs
, &request
, NULL
);
1330 static int coroutine_fn
nbd_client_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
1333 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1334 NBDRequest request
= {
1335 .type
= NBD_CMD_TRIM
,
1337 .len
= bytes
, /* len is uint32_t */
1340 assert(bytes
<= UINT32_MAX
); /* rely on max_pdiscard */
1342 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1343 if (!(s
->info
.flags
& NBD_FLAG_SEND_TRIM
) || !bytes
) {
1347 return nbd_co_request(bs
, &request
, NULL
);
1350 static int coroutine_fn
nbd_client_co_block_status(
1351 BlockDriverState
*bs
, bool want_zero
, int64_t offset
, int64_t bytes
,
1352 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
1354 int ret
, request_ret
;
1355 NBDExtent extent
= { 0 };
1356 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1357 Error
*local_err
= NULL
;
1359 NBDRequest request
= {
1360 .type
= NBD_CMD_BLOCK_STATUS
,
1362 .len
= MIN(QEMU_ALIGN_DOWN(INT_MAX
, bs
->bl
.request_alignment
),
1363 MIN(bytes
, s
->info
.size
- offset
)),
1364 .flags
= NBD_CMD_FLAG_REQ_ONE
,
1367 if (!s
->info
.base_allocation
) {
1371 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
1375 * Work around the fact that the block layer doesn't do
1376 * byte-accurate sizing yet - if the status request exceeds the
1377 * server's advertised size because the block layer rounded size
1378 * up, we truncated the request to the server (above), or are
1379 * called on just the hole.
1381 if (offset
>= s
->info
.size
) {
1383 assert(bytes
< BDRV_SECTOR_SIZE
);
1384 /* Intentionally don't report offset_valid for the hole */
1385 return BDRV_BLOCK_ZERO
;
1388 if (s
->info
.min_block
) {
1389 assert(QEMU_IS_ALIGNED(request
.len
, s
->info
.min_block
));
1392 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1397 ret
= nbd_co_receive_blockstatus_reply(s
, request
.handle
, bytes
,
1398 &extent
, &request_ret
,
1401 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1402 request
.flags
, request
.type
,
1403 nbd_cmd_lookup(request
.type
),
1404 ret
, error_get_pretty(local_err
));
1405 error_free(local_err
);
1408 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1410 if (ret
< 0 || request_ret
< 0) {
1411 return ret
? ret
: request_ret
;
1414 assert(extent
.length
);
1415 *pnum
= extent
.length
;
1418 return (extent
.flags
& NBD_STATE_HOLE
? 0 : BDRV_BLOCK_DATA
) |
1419 (extent
.flags
& NBD_STATE_ZERO
? BDRV_BLOCK_ZERO
: 0) |
1420 BDRV_BLOCK_OFFSET_VALID
;
1423 static int nbd_client_reopen_prepare(BDRVReopenState
*state
,
1424 BlockReopenQueue
*queue
, Error
**errp
)
1426 BDRVNBDState
*s
= (BDRVNBDState
*)state
->bs
->opaque
;
1428 if ((state
->flags
& BDRV_O_RDWR
) && (s
->info
.flags
& NBD_FLAG_READ_ONLY
)) {
1429 error_setg(errp
, "Can't reopen read-only NBD mount as read/write");
1435 static void nbd_yank(void *opaque
)
1437 BlockDriverState
*bs
= opaque
;
1438 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1440 qatomic_store_release(&s
->state
, NBD_CLIENT_QUIT
);
1441 qio_channel_shutdown(QIO_CHANNEL(s
->ioc
), QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
1444 static void nbd_client_close(BlockDriverState
*bs
)
1446 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1447 NBDRequest request
= { .type
= NBD_CMD_DISC
};
1450 nbd_send_request(s
->ioc
, &request
);
1453 nbd_teardown_connection(bs
);
1458 * Parse nbd_open options
1461 static int nbd_parse_uri(const char *filename
, QDict
*options
)
1465 QueryParams
*qp
= NULL
;
1469 uri
= uri_parse(filename
);
1475 if (!g_strcmp0(uri
->scheme
, "nbd")) {
1477 } else if (!g_strcmp0(uri
->scheme
, "nbd+tcp")) {
1479 } else if (!g_strcmp0(uri
->scheme
, "nbd+unix")) {
1486 p
= uri
->path
? uri
->path
: "";
1491 qdict_put_str(options
, "export", p
);
1494 qp
= query_params_parse(uri
->query
);
1495 if (qp
->n
> 1 || (is_unix
&& !qp
->n
) || (!is_unix
&& qp
->n
)) {
1501 /* nbd+unix:///export?socket=path */
1502 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
1506 qdict_put_str(options
, "server.type", "unix");
1507 qdict_put_str(options
, "server.path", qp
->p
[0].value
);
1512 /* nbd[+tcp]://host[:port]/export */
1518 /* strip braces from literal IPv6 address */
1519 if (uri
->server
[0] == '[') {
1520 host
= qstring_from_substr(uri
->server
, 1,
1521 strlen(uri
->server
) - 1);
1523 host
= qstring_from_str(uri
->server
);
1526 qdict_put_str(options
, "server.type", "inet");
1527 qdict_put(options
, "server.host", host
);
1529 port_str
= g_strdup_printf("%d", uri
->port
?: NBD_DEFAULT_PORT
);
1530 qdict_put_str(options
, "server.port", port_str
);
1536 query_params_free(qp
);
1542 static bool nbd_has_filename_options_conflict(QDict
*options
, Error
**errp
)
1544 const QDictEntry
*e
;
1546 for (e
= qdict_first(options
); e
; e
= qdict_next(options
, e
)) {
1547 if (!strcmp(e
->key
, "host") ||
1548 !strcmp(e
->key
, "port") ||
1549 !strcmp(e
->key
, "path") ||
1550 !strcmp(e
->key
, "export") ||
1551 strstart(e
->key
, "server.", NULL
))
1553 error_setg(errp
, "Option '%s' cannot be used with a file name",
1562 static void nbd_parse_filename(const char *filename
, QDict
*options
,
1565 g_autofree
char *file
= NULL
;
1567 const char *host_spec
;
1568 const char *unixpath
;
1570 if (nbd_has_filename_options_conflict(options
, errp
)) {
1574 if (strstr(filename
, "://")) {
1575 int ret
= nbd_parse_uri(filename
, options
);
1577 error_setg(errp
, "No valid URL specified");
1582 file
= g_strdup(filename
);
1584 export_name
= strstr(file
, EN_OPTSTR
);
1586 if (export_name
[strlen(EN_OPTSTR
)] == 0) {
1589 export_name
[0] = 0; /* truncate 'file' */
1590 export_name
+= strlen(EN_OPTSTR
);
1592 qdict_put_str(options
, "export", export_name
);
1595 /* extract the host_spec - fail if it's not nbd:... */
1596 if (!strstart(file
, "nbd:", &host_spec
)) {
1597 error_setg(errp
, "File name string for NBD must start with 'nbd:'");
1605 /* are we a UNIX or TCP socket? */
1606 if (strstart(host_spec
, "unix:", &unixpath
)) {
1607 qdict_put_str(options
, "server.type", "unix");
1608 qdict_put_str(options
, "server.path", unixpath
);
1610 InetSocketAddress
*addr
= g_new(InetSocketAddress
, 1);
1612 if (inet_parse(addr
, host_spec
, errp
)) {
1616 qdict_put_str(options
, "server.type", "inet");
1617 qdict_put_str(options
, "server.host", addr
->host
);
1618 qdict_put_str(options
, "server.port", addr
->port
);
1620 qapi_free_InetSocketAddress(addr
);
1624 static bool nbd_process_legacy_socket_options(QDict
*output_options
,
1625 QemuOpts
*legacy_opts
,
1628 const char *path
= qemu_opt_get(legacy_opts
, "path");
1629 const char *host
= qemu_opt_get(legacy_opts
, "host");
1630 const char *port
= qemu_opt_get(legacy_opts
, "port");
1631 const QDictEntry
*e
;
1633 if (!path
&& !host
&& !port
) {
1637 for (e
= qdict_first(output_options
); e
; e
= qdict_next(output_options
, e
))
1639 if (strstart(e
->key
, "server.", NULL
)) {
1640 error_setg(errp
, "Cannot use 'server' and path/host/port at the "
1647 error_setg(errp
, "path and host may not be used at the same time");
1651 error_setg(errp
, "port may not be used without host");
1655 qdict_put_str(output_options
, "server.type", "unix");
1656 qdict_put_str(output_options
, "server.path", path
);
1658 qdict_put_str(output_options
, "server.type", "inet");
1659 qdict_put_str(output_options
, "server.host", host
);
1660 qdict_put_str(output_options
, "server.port",
1661 port
?: stringify(NBD_DEFAULT_PORT
));
1667 static SocketAddress
*nbd_config(BDRVNBDState
*s
, QDict
*options
,
1670 SocketAddress
*saddr
= NULL
;
1674 qdict_extract_subqdict(options
, &addr
, "server.");
1675 if (!qdict_size(addr
)) {
1676 error_setg(errp
, "NBD server address missing");
1680 iv
= qobject_input_visitor_new_flat_confused(addr
, errp
);
1685 if (!visit_type_SocketAddress(iv
, NULL
, &saddr
, errp
)) {
1689 if (socket_address_parse_named_fd(saddr
, errp
) < 0) {
1690 qapi_free_SocketAddress(saddr
);
1696 qobject_unref(addr
);
1701 static QCryptoTLSCreds
*nbd_get_tls_creds(const char *id
, Error
**errp
)
1704 QCryptoTLSCreds
*creds
;
1706 obj
= object_resolve_path_component(
1707 object_get_objects_root(), id
);
1709 error_setg(errp
, "No TLS credentials with id '%s'",
1713 creds
= (QCryptoTLSCreds
*)
1714 object_dynamic_cast(obj
, TYPE_QCRYPTO_TLS_CREDS
);
1716 error_setg(errp
, "Object with id '%s' is not TLS credentials",
1721 if (!qcrypto_tls_creds_check_endpoint(creds
,
1722 QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
,
1731 static QemuOptsList nbd_runtime_opts
= {
1733 .head
= QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts
.head
),
1737 .type
= QEMU_OPT_STRING
,
1738 .help
= "TCP host to connect to",
1742 .type
= QEMU_OPT_STRING
,
1743 .help
= "TCP port to connect to",
1747 .type
= QEMU_OPT_STRING
,
1748 .help
= "Unix socket path to connect to",
1752 .type
= QEMU_OPT_STRING
,
1753 .help
= "Name of the NBD export to open",
1756 .name
= "tls-creds",
1757 .type
= QEMU_OPT_STRING
,
1758 .help
= "ID of the TLS credentials to use",
1761 .name
= "tls-hostname",
1762 .type
= QEMU_OPT_STRING
,
1763 .help
= "Override hostname for validating TLS x509 certificate",
1766 .name
= "x-dirty-bitmap",
1767 .type
= QEMU_OPT_STRING
,
1768 .help
= "experimental: expose named dirty bitmap in place of "
1772 .name
= "reconnect-delay",
1773 .type
= QEMU_OPT_NUMBER
,
1774 .help
= "On an unexpected disconnect, the nbd client tries to "
1775 "connect again until succeeding or encountering a serious "
1776 "error. During the first @reconnect-delay seconds, all "
1777 "requests are paused and will be rerun on a successful "
1778 "reconnect. After that time, any delayed requests and all "
1779 "future requests before a successful reconnect will "
1780 "immediately fail. Default 0",
1783 .name
= "open-timeout",
1784 .type
= QEMU_OPT_NUMBER
,
1785 .help
= "In seconds. If zero, the nbd driver tries the connection "
1786 "only once, and fails to open if the connection fails. "
1787 "If non-zero, the nbd driver will repeat connection "
1788 "attempts until successful or until @open-timeout seconds "
1789 "have elapsed. Default 0",
1791 { /* end of list */ }
1795 static int nbd_process_options(BlockDriverState
*bs
, QDict
*options
,
1798 BDRVNBDState
*s
= bs
->opaque
;
1802 opts
= qemu_opts_create(&nbd_runtime_opts
, NULL
, 0, &error_abort
);
1803 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1807 /* Translate @host, @port, and @path to a SocketAddress */
1808 if (!nbd_process_legacy_socket_options(options
, opts
, errp
)) {
1812 /* Pop the config into our state object. Exit if invalid. */
1813 s
->saddr
= nbd_config(s
, options
, errp
);
1818 s
->export
= g_strdup(qemu_opt_get(opts
, "export"));
1819 if (s
->export
&& strlen(s
->export
) > NBD_MAX_STRING_SIZE
) {
1820 error_setg(errp
, "export name too long to send to server");
1824 s
->tlscredsid
= g_strdup(qemu_opt_get(opts
, "tls-creds"));
1825 if (s
->tlscredsid
) {
1826 s
->tlscreds
= nbd_get_tls_creds(s
->tlscredsid
, errp
);
1831 s
->tlshostname
= g_strdup(qemu_opt_get(opts
, "tls-hostname"));
1832 if (!s
->tlshostname
&&
1833 s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
1834 s
->tlshostname
= g_strdup(s
->saddr
->u
.inet
.host
);
1838 s
->x_dirty_bitmap
= g_strdup(qemu_opt_get(opts
, "x-dirty-bitmap"));
1839 if (s
->x_dirty_bitmap
&& strlen(s
->x_dirty_bitmap
) > NBD_MAX_STRING_SIZE
) {
1840 error_setg(errp
, "x-dirty-bitmap query too long to send to server");
1844 s
->reconnect_delay
= qemu_opt_get_number(opts
, "reconnect-delay", 0);
1845 s
->open_timeout
= qemu_opt_get_number(opts
, "open-timeout", 0);
1850 qemu_opts_del(opts
);
1854 static int nbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1858 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1861 qemu_co_mutex_init(&s
->send_mutex
);
1862 qemu_co_queue_init(&s
->free_sema
);
1863 qemu_co_mutex_init(&s
->receive_mutex
);
1865 if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
), errp
)) {
1869 ret
= nbd_process_options(bs
, options
, errp
);
1874 s
->conn
= nbd_client_connection_new(s
->saddr
, true, s
->export
,
1875 s
->x_dirty_bitmap
, s
->tlscreds
,
1878 if (s
->open_timeout
) {
1879 nbd_client_connection_enable_retry(s
->conn
);
1880 open_timer_init(s
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) +
1881 s
->open_timeout
* NANOSECONDS_PER_SECOND
);
1884 s
->state
= NBD_CLIENT_CONNECTING_WAIT
;
1885 ret
= nbd_do_establish_connection(bs
, errp
);
1891 * The connect attempt is done, so we no longer need this timer.
1892 * Delete it, because we do not want it to be around when this node
1893 * is drained or closed.
1897 nbd_client_connection_enable_retry(s
->conn
);
1903 nbd_clear_bdrvstate(bs
);
1907 static int coroutine_fn
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
= QEMU_ALIGN_DOWN(INT_MAX
, min
);
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 nbd_client_close(bs
);
1949 nbd_clear_bdrvstate(bs
);
1953 * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1954 * to a smaller size with exact=false, there is no reason to fail the
1957 * Preallocation mode is ignored since it does not seems useful to fail when
1958 * we never change anything.
1960 static int coroutine_fn
nbd_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1961 bool exact
, PreallocMode prealloc
,
1962 BdrvRequestFlags flags
, Error
**errp
)
1964 BDRVNBDState
*s
= bs
->opaque
;
1966 if (offset
!= s
->info
.size
&& exact
) {
1967 error_setg(errp
, "Cannot resize NBD nodes");
1971 if (offset
> s
->info
.size
) {
1972 error_setg(errp
, "Cannot grow NBD nodes");
1979 static int64_t nbd_getlength(BlockDriverState
*bs
)
1981 BDRVNBDState
*s
= bs
->opaque
;
1983 return s
->info
.size
;
1986 static void nbd_refresh_filename(BlockDriverState
*bs
)
1988 BDRVNBDState
*s
= bs
->opaque
;
1989 const char *host
= NULL
, *port
= NULL
, *path
= NULL
;
1992 if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
1993 const InetSocketAddress
*inet
= &s
->saddr
->u
.inet
;
1994 if (!inet
->has_ipv4
&& !inet
->has_ipv6
&& !inet
->has_to
) {
1998 } else if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_UNIX
) {
1999 path
= s
->saddr
->u
.q_unix
.path
;
2000 } /* else can't represent as pseudo-filename */
2002 if (path
&& s
->export
) {
2003 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2004 "nbd+unix:///%s?socket=%s", s
->export
, path
);
2005 } else if (path
&& !s
->export
) {
2006 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2007 "nbd+unix://?socket=%s", path
);
2008 } else if (host
&& s
->export
) {
2009 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2010 "nbd://%s:%s/%s", host
, port
, s
->export
);
2011 } else if (host
&& !s
->export
) {
2012 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2013 "nbd://%s:%s", host
, port
);
2015 if (len
>= sizeof(bs
->exact_filename
)) {
2016 /* Name is too long to represent exactly, so leave it empty. */
2017 bs
->exact_filename
[0] = '\0';
2021 static char *nbd_dirname(BlockDriverState
*bs
, Error
**errp
)
2023 /* The generic bdrv_dirname() implementation is able to work out some
2024 * directory name for NBD nodes, but that would be wrong. So far there is no
2025 * specification for how "export paths" would work, so NBD does not have
2026 * directory names. */
2027 error_setg(errp
, "Cannot generate a base directory for NBD nodes");
2031 static const char *const nbd_strong_runtime_opts
[] = {
2043 static void nbd_cancel_in_flight(BlockDriverState
*bs
)
2045 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
2047 reconnect_delay_timer_del(s
);
2049 if (s
->state
== NBD_CLIENT_CONNECTING_WAIT
) {
2050 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
2051 qemu_co_queue_restart_all(&s
->free_sema
);
2054 nbd_co_establish_connection_cancel(s
->conn
);
2057 static void nbd_attach_aio_context(BlockDriverState
*bs
,
2058 AioContext
*new_context
)
2060 BDRVNBDState
*s
= bs
->opaque
;
2062 /* The open_timer is used only during nbd_open() */
2063 assert(!s
->open_timer
);
2066 * The reconnect_delay_timer is scheduled in I/O paths when the
2067 * connection is lost, to cancel the reconnection attempt after a
2068 * given time. Once this attempt is done (successfully or not),
2069 * nbd_reconnect_attempt() ensures the timer is deleted before the
2070 * respective I/O request is resumed.
2071 * Since the AioContext can only be changed when a node is drained,
2072 * the reconnect_delay_timer cannot be active here.
2074 assert(!s
->reconnect_delay_timer
);
2077 qio_channel_attach_aio_context(s
->ioc
, new_context
);
2081 static void nbd_detach_aio_context(BlockDriverState
*bs
)
2083 BDRVNBDState
*s
= bs
->opaque
;
2085 assert(!s
->open_timer
);
2086 assert(!s
->reconnect_delay_timer
);
2089 qio_channel_detach_aio_context(s
->ioc
);
2093 static BlockDriver bdrv_nbd
= {
2094 .format_name
= "nbd",
2095 .protocol_name
= "nbd",
2096 .instance_size
= sizeof(BDRVNBDState
),
2097 .bdrv_parse_filename
= nbd_parse_filename
,
2098 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2099 .create_opts
= &bdrv_create_opts_simple
,
2100 .bdrv_file_open
= nbd_open
,
2101 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2102 .bdrv_co_preadv
= nbd_client_co_preadv
,
2103 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2104 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2105 .bdrv_close
= nbd_close
,
2106 .bdrv_co_flush_to_os
= nbd_co_flush
,
2107 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2108 .bdrv_refresh_limits
= nbd_refresh_limits
,
2109 .bdrv_co_truncate
= nbd_co_truncate
,
2110 .bdrv_getlength
= nbd_getlength
,
2111 .bdrv_refresh_filename
= nbd_refresh_filename
,
2112 .bdrv_co_block_status
= nbd_client_co_block_status
,
2113 .bdrv_dirname
= nbd_dirname
,
2114 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2115 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2117 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2118 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2121 static BlockDriver bdrv_nbd_tcp
= {
2122 .format_name
= "nbd",
2123 .protocol_name
= "nbd+tcp",
2124 .instance_size
= sizeof(BDRVNBDState
),
2125 .bdrv_parse_filename
= nbd_parse_filename
,
2126 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2127 .create_opts
= &bdrv_create_opts_simple
,
2128 .bdrv_file_open
= nbd_open
,
2129 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2130 .bdrv_co_preadv
= nbd_client_co_preadv
,
2131 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2132 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2133 .bdrv_close
= nbd_close
,
2134 .bdrv_co_flush_to_os
= nbd_co_flush
,
2135 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2136 .bdrv_refresh_limits
= nbd_refresh_limits
,
2137 .bdrv_co_truncate
= nbd_co_truncate
,
2138 .bdrv_getlength
= nbd_getlength
,
2139 .bdrv_refresh_filename
= nbd_refresh_filename
,
2140 .bdrv_co_block_status
= nbd_client_co_block_status
,
2141 .bdrv_dirname
= nbd_dirname
,
2142 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2143 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2145 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2146 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2149 static BlockDriver bdrv_nbd_unix
= {
2150 .format_name
= "nbd",
2151 .protocol_name
= "nbd+unix",
2152 .instance_size
= sizeof(BDRVNBDState
),
2153 .bdrv_parse_filename
= nbd_parse_filename
,
2154 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2155 .create_opts
= &bdrv_create_opts_simple
,
2156 .bdrv_file_open
= nbd_open
,
2157 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2158 .bdrv_co_preadv
= nbd_client_co_preadv
,
2159 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2160 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2161 .bdrv_close
= nbd_close
,
2162 .bdrv_co_flush_to_os
= nbd_co_flush
,
2163 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2164 .bdrv_refresh_limits
= nbd_refresh_limits
,
2165 .bdrv_co_truncate
= nbd_co_truncate
,
2166 .bdrv_getlength
= nbd_getlength
,
2167 .bdrv_refresh_filename
= nbd_refresh_filename
,
2168 .bdrv_co_block_status
= nbd_client_co_block_status
,
2169 .bdrv_dirname
= nbd_dirname
,
2170 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2171 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2173 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2174 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2177 static void bdrv_nbd_init(void)
2179 bdrv_register(&bdrv_nbd
);
2180 bdrv_register(&bdrv_nbd_tcp
);
2181 bdrv_register(&bdrv_nbd_unix
);
2184 block_init(bdrv_nbd_init
);