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"
41 #include "qapi/clone-visitor.h"
43 #include "block/qdict.h"
44 #include "block/nbd.h"
45 #include "block/block_int.h"
46 #include "block/coroutines.h"
48 #include "qemu/yank.h"
50 #define EN_OPTSTR ":exportname="
51 #define MAX_NBD_REQUESTS 16
53 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
54 #define INDEX_TO_HANDLE(bs, index) ((index) ^ (uint64_t)(intptr_t)(bs))
58 uint64_t offset
; /* original offset of the request */
59 bool receiving
; /* sleeping in the yield in nbd_receive_replies */
62 typedef enum NBDClientState
{
63 NBD_CLIENT_CONNECTING_WAIT
,
64 NBD_CLIENT_CONNECTING_NOWAIT
,
69 typedef struct BDRVNBDState
{
70 QIOChannel
*ioc
; /* The current I/O channel */
74 * Protects state, free_sema, in_flight, requests[].coroutine,
75 * reconnect_delay_timer.
77 QemuMutex requests_lock
;
81 NBDClientRequest requests
[MAX_NBD_REQUESTS
];
82 QEMUTimer
*reconnect_delay_timer
;
84 /* Protects sending data on the socket. */
88 * Protects receiving reply headers from the socket, as well as the
89 * fields reply and requests[].receiving
91 CoMutex receive_mutex
;
94 QEMUTimer
*open_timer
;
98 /* Connection parameters */
99 uint32_t reconnect_delay
;
100 uint32_t open_timeout
;
101 SocketAddress
*saddr
;
104 QCryptoTLSCreds
*tlscreds
;
106 char *x_dirty_bitmap
;
109 NBDClientConnection
*conn
;
112 static void nbd_yank(void *opaque
);
114 static void nbd_clear_bdrvstate(BlockDriverState
*bs
)
116 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
118 nbd_client_connection_release(s
->conn
);
121 yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
));
123 /* Must not leave timers behind that would access freed data */
124 assert(!s
->reconnect_delay_timer
);
125 assert(!s
->open_timer
);
127 object_unref(OBJECT(s
->tlscreds
));
128 qapi_free_SocketAddress(s
->saddr
);
132 g_free(s
->tlscredsid
);
133 s
->tlscredsid
= NULL
;
134 g_free(s
->tlshostname
);
135 s
->tlshostname
= NULL
;
136 g_free(s
->x_dirty_bitmap
);
137 s
->x_dirty_bitmap
= NULL
;
140 /* Called with s->receive_mutex taken. */
141 static bool coroutine_fn
nbd_recv_coroutine_wake_one(NBDClientRequest
*req
)
143 if (req
->receiving
) {
144 req
->receiving
= false;
145 aio_co_wake(req
->coroutine
);
152 static void coroutine_fn
nbd_recv_coroutines_wake(BDRVNBDState
*s
)
156 QEMU_LOCK_GUARD(&s
->receive_mutex
);
157 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
158 if (nbd_recv_coroutine_wake_one(&s
->requests
[i
])) {
164 /* Called with s->requests_lock held. */
165 static void coroutine_fn
nbd_channel_error_locked(BDRVNBDState
*s
, int ret
)
167 if (s
->state
== NBD_CLIENT_CONNECTED
) {
168 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
172 if (s
->state
== NBD_CLIENT_CONNECTED
) {
173 s
->state
= s
->reconnect_delay
? NBD_CLIENT_CONNECTING_WAIT
:
174 NBD_CLIENT_CONNECTING_NOWAIT
;
177 s
->state
= NBD_CLIENT_QUIT
;
181 static void coroutine_fn
nbd_channel_error(BDRVNBDState
*s
, int ret
)
183 QEMU_LOCK_GUARD(&s
->requests_lock
);
184 nbd_channel_error_locked(s
, ret
);
187 static void reconnect_delay_timer_del(BDRVNBDState
*s
)
189 if (s
->reconnect_delay_timer
) {
190 timer_free(s
->reconnect_delay_timer
);
191 s
->reconnect_delay_timer
= NULL
;
195 static void reconnect_delay_timer_cb(void *opaque
)
197 BDRVNBDState
*s
= opaque
;
199 reconnect_delay_timer_del(s
);
200 WITH_QEMU_LOCK_GUARD(&s
->requests_lock
) {
201 if (s
->state
!= NBD_CLIENT_CONNECTING_WAIT
) {
204 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
206 nbd_co_establish_connection_cancel(s
->conn
);
209 static void reconnect_delay_timer_init(BDRVNBDState
*s
, uint64_t expire_time_ns
)
211 assert(!s
->reconnect_delay_timer
);
212 s
->reconnect_delay_timer
= aio_timer_new(bdrv_get_aio_context(s
->bs
),
215 reconnect_delay_timer_cb
, s
);
216 timer_mod(s
->reconnect_delay_timer
, expire_time_ns
);
219 static void nbd_teardown_connection(BlockDriverState
*bs
)
221 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
223 assert(!s
->in_flight
);
226 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
227 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
229 object_unref(OBJECT(s
->ioc
));
233 WITH_QEMU_LOCK_GUARD(&s
->requests_lock
) {
234 s
->state
= NBD_CLIENT_QUIT
;
238 static void open_timer_del(BDRVNBDState
*s
)
241 timer_free(s
->open_timer
);
242 s
->open_timer
= NULL
;
246 static void open_timer_cb(void *opaque
)
248 BDRVNBDState
*s
= opaque
;
250 nbd_co_establish_connection_cancel(s
->conn
);
254 static void open_timer_init(BDRVNBDState
*s
, uint64_t expire_time_ns
)
256 assert(!s
->open_timer
);
257 s
->open_timer
= aio_timer_new(bdrv_get_aio_context(s
->bs
),
261 timer_mod(s
->open_timer
, expire_time_ns
);
264 static bool nbd_client_will_reconnect(BDRVNBDState
*s
)
267 * Called only after a socket error, so this is not performance sensitive.
269 QEMU_LOCK_GUARD(&s
->requests_lock
);
270 return s
->state
== NBD_CLIENT_CONNECTING_WAIT
;
274 * Update @bs with information learned during a completed negotiation process.
275 * Return failure if the server's advertised options are incompatible with the
278 static int nbd_handle_updated_info(BlockDriverState
*bs
, Error
**errp
)
280 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
283 if (s
->x_dirty_bitmap
) {
284 if (!s
->info
.base_allocation
) {
285 error_setg(errp
, "requested x-dirty-bitmap %s not found",
289 if (strcmp(s
->x_dirty_bitmap
, "qemu:allocation-depth") == 0) {
290 s
->alloc_depth
= true;
294 if (s
->info
.flags
& NBD_FLAG_READ_ONLY
) {
295 ret
= bdrv_apply_auto_read_only(bs
, "NBD export is read-only", errp
);
301 if (s
->info
.flags
& NBD_FLAG_SEND_FUA
) {
302 bs
->supported_write_flags
= BDRV_REQ_FUA
;
303 bs
->supported_zero_flags
|= BDRV_REQ_FUA
;
306 if (s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
) {
307 bs
->supported_zero_flags
|= BDRV_REQ_MAY_UNMAP
;
308 if (s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
) {
309 bs
->supported_zero_flags
|= BDRV_REQ_NO_FALLBACK
;
313 trace_nbd_client_handshake_success(s
->export
);
318 int coroutine_fn
nbd_co_do_establish_connection(BlockDriverState
*bs
,
319 bool blocking
, Error
**errp
)
321 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
327 s
->ioc
= nbd_co_establish_connection(s
->conn
, &s
->info
, blocking
, errp
);
329 return -ECONNREFUSED
;
332 yank_register_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
), nbd_yank
,
335 ret
= nbd_handle_updated_info(s
->bs
, NULL
);
338 * We have connected, but must fail for other reasons.
339 * Send NBD_CMD_DISC as a courtesy to the server.
341 NBDRequest request
= { .type
= NBD_CMD_DISC
};
343 nbd_send_request(s
->ioc
, &request
);
345 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
347 object_unref(OBJECT(s
->ioc
));
353 qio_channel_set_blocking(s
->ioc
, false, NULL
);
354 qio_channel_attach_aio_context(s
->ioc
, bdrv_get_aio_context(bs
));
356 /* successfully connected */
357 WITH_QEMU_LOCK_GUARD(&s
->requests_lock
) {
358 s
->state
= NBD_CLIENT_CONNECTED
;
364 /* Called with s->requests_lock held. */
365 static bool nbd_client_connecting(BDRVNBDState
*s
)
367 return s
->state
== NBD_CLIENT_CONNECTING_WAIT
||
368 s
->state
== NBD_CLIENT_CONNECTING_NOWAIT
;
371 /* Called with s->requests_lock taken. */
372 static coroutine_fn
void nbd_reconnect_attempt(BDRVNBDState
*s
)
375 bool blocking
= s
->state
== NBD_CLIENT_CONNECTING_WAIT
;
378 * Now we are sure that nobody is accessing the channel, and no one will
379 * try until we set the state to CONNECTED.
381 assert(nbd_client_connecting(s
));
382 assert(s
->in_flight
== 1);
384 trace_nbd_reconnect_attempt(s
->bs
->in_flight
);
386 if (blocking
&& !s
->reconnect_delay_timer
) {
388 * It's the first reconnect attempt after switching to
389 * NBD_CLIENT_CONNECTING_WAIT
391 g_assert(s
->reconnect_delay
);
392 reconnect_delay_timer_init(s
,
393 qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) +
394 s
->reconnect_delay
* NANOSECONDS_PER_SECOND
);
397 /* Finalize previous connection if any */
399 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
400 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
402 object_unref(OBJECT(s
->ioc
));
406 qemu_mutex_unlock(&s
->requests_lock
);
407 ret
= nbd_co_do_establish_connection(s
->bs
, blocking
, NULL
);
408 trace_nbd_reconnect_attempt_result(ret
, s
->bs
->in_flight
);
409 qemu_mutex_lock(&s
->requests_lock
);
412 * The reconnect attempt is done (maybe successfully, maybe not), so
413 * we no longer need this timer. Delete it so it will not outlive
414 * this I/O request (so draining removes all timers).
416 reconnect_delay_timer_del(s
);
419 static coroutine_fn
int nbd_receive_replies(BDRVNBDState
*s
, uint64_t handle
)
422 uint64_t ind
= HANDLE_TO_INDEX(s
, handle
), ind2
;
423 QEMU_LOCK_GUARD(&s
->receive_mutex
);
426 if (s
->reply
.handle
== handle
) {
431 if (s
->reply
.handle
!= 0) {
433 * Some other request is being handled now. It should already be
434 * woken by whoever set s->reply.handle (or never wait in this
435 * yield). So, we should not wake it here.
437 ind2
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
438 assert(!s
->requests
[ind2
].receiving
);
440 s
->requests
[ind
].receiving
= true;
441 qemu_co_mutex_unlock(&s
->receive_mutex
);
443 qemu_coroutine_yield();
445 * We may be woken for 2 reasons:
446 * 1. From this function, executing in parallel coroutine, when our
447 * handle is received.
448 * 2. From nbd_co_receive_one_chunk(), when previous request is
449 * finished and s->reply.handle set to 0.
450 * Anyway, it's OK to lock the mutex and go to the next iteration.
453 qemu_co_mutex_lock(&s
->receive_mutex
);
454 assert(!s
->requests
[ind
].receiving
);
458 /* We are under mutex and handle is 0. We have to do the dirty work. */
459 assert(s
->reply
.handle
== 0);
460 ret
= nbd_receive_reply(s
->bs
, s
->ioc
, &s
->reply
, NULL
);
462 ret
= ret
? ret
: -EIO
;
463 nbd_channel_error(s
, ret
);
466 if (nbd_reply_is_structured(&s
->reply
) && !s
->info
.structured_reply
) {
467 nbd_channel_error(s
, -EINVAL
);
470 ind2
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
471 if (ind2
>= MAX_NBD_REQUESTS
|| !s
->requests
[ind2
].coroutine
) {
472 nbd_channel_error(s
, -EINVAL
);
475 if (s
->reply
.handle
== handle
) {
479 nbd_recv_coroutine_wake_one(&s
->requests
[ind2
]);
483 static int coroutine_fn
nbd_co_send_request(BlockDriverState
*bs
,
487 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
490 qemu_mutex_lock(&s
->requests_lock
);
491 while (s
->in_flight
== MAX_NBD_REQUESTS
||
492 (s
->state
!= NBD_CLIENT_CONNECTED
&& s
->in_flight
> 0)) {
493 qemu_co_queue_wait(&s
->free_sema
, &s
->requests_lock
);
497 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
498 if (nbd_client_connecting(s
)) {
499 nbd_reconnect_attempt(s
);
500 qemu_co_queue_restart_all(&s
->free_sema
);
502 if (s
->state
!= NBD_CLIENT_CONNECTED
) {
508 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
509 if (s
->requests
[i
].coroutine
== NULL
) {
514 assert(i
< MAX_NBD_REQUESTS
);
515 s
->requests
[i
].coroutine
= qemu_coroutine_self();
516 s
->requests
[i
].offset
= request
->from
;
517 s
->requests
[i
].receiving
= false;
518 qemu_mutex_unlock(&s
->requests_lock
);
520 qemu_co_mutex_lock(&s
->send_mutex
);
521 request
->handle
= INDEX_TO_HANDLE(s
, i
);
526 qio_channel_set_cork(s
->ioc
, true);
527 rc
= nbd_send_request(s
->ioc
, request
);
528 if (rc
>= 0 && qio_channel_writev_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
532 qio_channel_set_cork(s
->ioc
, false);
534 rc
= nbd_send_request(s
->ioc
, request
);
536 qemu_co_mutex_unlock(&s
->send_mutex
);
539 qemu_mutex_lock(&s
->requests_lock
);
541 nbd_channel_error_locked(s
, rc
);
543 s
->requests
[i
].coroutine
= NULL
;
546 qemu_co_queue_next(&s
->free_sema
);
547 qemu_mutex_unlock(&s
->requests_lock
);
552 static inline uint16_t payload_advance16(uint8_t **payload
)
555 return lduw_be_p(*payload
- 2);
558 static inline uint32_t payload_advance32(uint8_t **payload
)
561 return ldl_be_p(*payload
- 4);
564 static inline uint64_t payload_advance64(uint8_t **payload
)
567 return ldq_be_p(*payload
- 8);
570 static int nbd_parse_offset_hole_payload(BDRVNBDState
*s
,
571 NBDStructuredReplyChunk
*chunk
,
572 uint8_t *payload
, uint64_t orig_offset
,
573 QEMUIOVector
*qiov
, Error
**errp
)
578 if (chunk
->length
!= sizeof(offset
) + sizeof(hole_size
)) {
579 error_setg(errp
, "Protocol error: invalid payload for "
580 "NBD_REPLY_TYPE_OFFSET_HOLE");
584 offset
= payload_advance64(&payload
);
585 hole_size
= payload_advance32(&payload
);
587 if (!hole_size
|| offset
< orig_offset
|| hole_size
> qiov
->size
||
588 offset
> orig_offset
+ qiov
->size
- hole_size
) {
589 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
593 if (s
->info
.min_block
&&
594 !QEMU_IS_ALIGNED(hole_size
, s
->info
.min_block
)) {
595 trace_nbd_structured_read_compliance("hole");
598 qemu_iovec_memset(qiov
, offset
- orig_offset
, 0, hole_size
);
604 * nbd_parse_blockstatus_payload
605 * Based on our request, we expect only one extent in reply, for the
606 * base:allocation context.
608 static int nbd_parse_blockstatus_payload(BDRVNBDState
*s
,
609 NBDStructuredReplyChunk
*chunk
,
610 uint8_t *payload
, uint64_t orig_length
,
611 NBDExtent
*extent
, Error
**errp
)
615 /* The server succeeded, so it must have sent [at least] one extent */
616 if (chunk
->length
< sizeof(context_id
) + sizeof(*extent
)) {
617 error_setg(errp
, "Protocol error: invalid payload for "
618 "NBD_REPLY_TYPE_BLOCK_STATUS");
622 context_id
= payload_advance32(&payload
);
623 if (s
->info
.context_id
!= context_id
) {
624 error_setg(errp
, "Protocol error: unexpected context id %d for "
625 "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
626 "id is %d", context_id
,
631 extent
->length
= payload_advance32(&payload
);
632 extent
->flags
= payload_advance32(&payload
);
634 if (extent
->length
== 0) {
635 error_setg(errp
, "Protocol error: server sent status chunk with "
641 * A server sending unaligned block status is in violation of the
642 * protocol, but as qemu-nbd 3.1 is such a server (at least for
643 * POSIX files that are not a multiple of 512 bytes, since qemu
644 * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
645 * still sees an implicit hole beyond the real EOF), it's nicer to
646 * work around the misbehaving server. If the request included
647 * more than the final unaligned block, truncate it back to an
648 * aligned result; if the request was only the final block, round
649 * up to the full block and change the status to fully-allocated
650 * (always a safe status, even if it loses information).
652 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(extent
->length
,
653 s
->info
.min_block
)) {
654 trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
655 if (extent
->length
> s
->info
.min_block
) {
656 extent
->length
= QEMU_ALIGN_DOWN(extent
->length
,
659 extent
->length
= s
->info
.min_block
;
665 * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
666 * sent us any more than one extent, nor should it have included
667 * status beyond our request in that extent. However, it's easy
668 * enough to ignore the server's noncompliance without killing the
669 * connection; just ignore trailing extents, and clamp things to
670 * the length of our request.
672 if (chunk
->length
> sizeof(context_id
) + sizeof(*extent
)) {
673 trace_nbd_parse_blockstatus_compliance("more than one extent");
675 if (extent
->length
> orig_length
) {
676 extent
->length
= orig_length
;
677 trace_nbd_parse_blockstatus_compliance("extent length too large");
681 * HACK: if we are using x-dirty-bitmaps to access
682 * qemu:allocation-depth, treat all depths > 2 the same as 2,
683 * since nbd_client_co_block_status is only expecting the low two
686 if (s
->alloc_depth
&& extent
->flags
> 2) {
694 * nbd_parse_error_payload
695 * on success @errp contains message describing nbd error reply
697 static int nbd_parse_error_payload(NBDStructuredReplyChunk
*chunk
,
698 uint8_t *payload
, int *request_ret
,
702 uint16_t message_size
;
704 assert(chunk
->type
& (1 << 15));
706 if (chunk
->length
< sizeof(error
) + sizeof(message_size
)) {
708 "Protocol error: invalid payload for structured error");
712 error
= nbd_errno_to_system_errno(payload_advance32(&payload
));
714 error_setg(errp
, "Protocol error: server sent structured error chunk "
719 *request_ret
= -error
;
720 message_size
= payload_advance16(&payload
);
722 if (message_size
> chunk
->length
- sizeof(error
) - sizeof(message_size
)) {
723 error_setg(errp
, "Protocol error: server sent structured error chunk "
724 "with incorrect message size");
728 /* TODO: Add a trace point to mention the server complaint */
730 /* TODO handle ERROR_OFFSET */
735 static int coroutine_fn
736 nbd_co_receive_offset_data_payload(BDRVNBDState
*s
, uint64_t orig_offset
,
737 QEMUIOVector
*qiov
, Error
**errp
)
739 QEMUIOVector sub_qiov
;
743 NBDStructuredReplyChunk
*chunk
= &s
->reply
.structured
;
745 assert(nbd_reply_is_structured(&s
->reply
));
747 /* The NBD spec requires at least one byte of payload */
748 if (chunk
->length
<= sizeof(offset
)) {
749 error_setg(errp
, "Protocol error: invalid payload for "
750 "NBD_REPLY_TYPE_OFFSET_DATA");
754 if (nbd_read64(s
->ioc
, &offset
, "OFFSET_DATA offset", errp
) < 0) {
758 data_size
= chunk
->length
- sizeof(offset
);
760 if (offset
< orig_offset
|| data_size
> qiov
->size
||
761 offset
> orig_offset
+ qiov
->size
- data_size
) {
762 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
766 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(data_size
, s
->info
.min_block
)) {
767 trace_nbd_structured_read_compliance("data");
770 qemu_iovec_init(&sub_qiov
, qiov
->niov
);
771 qemu_iovec_concat(&sub_qiov
, qiov
, offset
- orig_offset
, data_size
);
772 ret
= qio_channel_readv_all(s
->ioc
, sub_qiov
.iov
, sub_qiov
.niov
, errp
);
773 qemu_iovec_destroy(&sub_qiov
);
775 return ret
< 0 ? -EIO
: 0;
778 #define NBD_MAX_MALLOC_PAYLOAD 1000
779 static coroutine_fn
int nbd_co_receive_structured_payload(
780 BDRVNBDState
*s
, void **payload
, Error
**errp
)
785 assert(nbd_reply_is_structured(&s
->reply
));
787 len
= s
->reply
.structured
.length
;
793 if (payload
== NULL
) {
794 error_setg(errp
, "Unexpected structured payload");
798 if (len
> NBD_MAX_MALLOC_PAYLOAD
) {
799 error_setg(errp
, "Payload too large");
803 *payload
= g_new(char, len
);
804 ret
= nbd_read(s
->ioc
, *payload
, len
, "structured payload", errp
);
815 * nbd_co_do_receive_one_chunk
817 * set request_ret to received reply error
818 * if qiov is not NULL: read payload to @qiov
819 * for structured reply chunk:
820 * if error chunk: read payload, set @request_ret, do not set @payload
821 * else if offset_data chunk: read payload data to @qiov, do not set @payload
822 * else: read payload to @payload
824 * If function fails, @errp contains corresponding error message, and the
825 * connection with the server is suspect. If it returns 0, then the
826 * transaction succeeded (although @request_ret may be a negative errno
827 * corresponding to the server's error reply), and errp is unchanged.
829 static coroutine_fn
int nbd_co_do_receive_one_chunk(
830 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
831 int *request_ret
, QEMUIOVector
*qiov
, void **payload
, Error
**errp
)
834 int i
= HANDLE_TO_INDEX(s
, handle
);
835 void *local_payload
= NULL
;
836 NBDStructuredReplyChunk
*chunk
;
843 ret
= nbd_receive_replies(s
, handle
);
845 error_setg(errp
, "Connection closed");
850 assert(s
->reply
.handle
== handle
);
852 if (nbd_reply_is_simple(&s
->reply
)) {
853 if (only_structured
) {
854 error_setg(errp
, "Protocol error: simple reply when structured "
855 "reply chunk was expected");
859 *request_ret
= -nbd_errno_to_system_errno(s
->reply
.simple
.error
);
860 if (*request_ret
< 0 || !qiov
) {
864 return qio_channel_readv_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
865 errp
) < 0 ? -EIO
: 0;
868 /* handle structured reply chunk */
869 assert(s
->info
.structured_reply
);
870 chunk
= &s
->reply
.structured
;
872 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
873 if (!(chunk
->flags
& NBD_REPLY_FLAG_DONE
)) {
874 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
875 " NBD_REPLY_FLAG_DONE flag set");
879 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
886 if (chunk
->type
== NBD_REPLY_TYPE_OFFSET_DATA
) {
888 error_setg(errp
, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
892 return nbd_co_receive_offset_data_payload(s
, s
->requests
[i
].offset
,
896 if (nbd_reply_type_is_error(chunk
->type
)) {
897 payload
= &local_payload
;
900 ret
= nbd_co_receive_structured_payload(s
, payload
, errp
);
905 if (nbd_reply_type_is_error(chunk
->type
)) {
906 ret
= nbd_parse_error_payload(chunk
, local_payload
, request_ret
, errp
);
907 g_free(local_payload
);
915 * nbd_co_receive_one_chunk
916 * Read reply, wake up connection_co and set s->quit if needed.
917 * Return value is a fatal error code or normal nbd reply error code
919 static coroutine_fn
int nbd_co_receive_one_chunk(
920 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
921 int *request_ret
, QEMUIOVector
*qiov
, NBDReply
*reply
, void **payload
,
924 int ret
= nbd_co_do_receive_one_chunk(s
, handle
, only_structured
,
925 request_ret
, qiov
, payload
, errp
);
928 memset(reply
, 0, sizeof(*reply
));
929 nbd_channel_error(s
, ret
);
931 /* For assert at loop start in nbd_connection_entry */
936 nbd_recv_coroutines_wake(s
);
941 typedef struct NBDReplyChunkIter
{
945 bool done
, only_structured
;
948 static void nbd_iter_channel_error(NBDReplyChunkIter
*iter
,
949 int ret
, Error
**local_err
)
951 assert(local_err
&& *local_err
);
956 error_propagate(&iter
->err
, *local_err
);
958 error_free(*local_err
);
964 static void nbd_iter_request_error(NBDReplyChunkIter
*iter
, int ret
)
968 if (!iter
->request_ret
) {
969 iter
->request_ret
= ret
;
974 * NBD_FOREACH_REPLY_CHUNK
975 * The pointer stored in @payload requires g_free() to free it.
977 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
978 qiov, reply, payload) \
979 for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
980 nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
983 * nbd_reply_chunk_iter_receive
984 * The pointer stored in @payload requires g_free() to free it.
986 static bool coroutine_fn
nbd_reply_chunk_iter_receive(BDRVNBDState
*s
,
987 NBDReplyChunkIter
*iter
,
993 int ret
, request_ret
;
994 NBDReply local_reply
;
995 NBDStructuredReplyChunk
*chunk
;
996 Error
*local_err
= NULL
;
999 /* Previous iteration was last. */
1003 if (reply
== NULL
) {
1004 reply
= &local_reply
;
1007 ret
= nbd_co_receive_one_chunk(s
, handle
, iter
->only_structured
,
1008 &request_ret
, qiov
, reply
, payload
,
1011 nbd_iter_channel_error(iter
, ret
, &local_err
);
1012 } else if (request_ret
< 0) {
1013 nbd_iter_request_error(iter
, request_ret
);
1016 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1017 if (nbd_reply_is_simple(reply
) || iter
->ret
< 0) {
1021 chunk
= &reply
->structured
;
1022 iter
->only_structured
= true;
1024 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
1025 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1026 assert(chunk
->flags
& NBD_REPLY_FLAG_DONE
);
1030 if (chunk
->flags
& NBD_REPLY_FLAG_DONE
) {
1031 /* This iteration is last. */
1035 /* Execute the loop body */
1039 qemu_mutex_lock(&s
->requests_lock
);
1040 s
->requests
[HANDLE_TO_INDEX(s
, handle
)].coroutine
= NULL
;
1042 qemu_co_queue_next(&s
->free_sema
);
1043 qemu_mutex_unlock(&s
->requests_lock
);
1048 static int coroutine_fn
nbd_co_receive_return_code(BDRVNBDState
*s
, uint64_t handle
,
1049 int *request_ret
, Error
**errp
)
1051 NBDReplyChunkIter iter
;
1053 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, NULL
, NULL
) {
1054 /* nbd_reply_chunk_iter_receive does all the work */
1057 error_propagate(errp
, iter
.err
);
1058 *request_ret
= iter
.request_ret
;
1062 static int coroutine_fn
nbd_co_receive_cmdread_reply(BDRVNBDState
*s
, uint64_t handle
,
1063 uint64_t offset
, QEMUIOVector
*qiov
,
1064 int *request_ret
, Error
**errp
)
1066 NBDReplyChunkIter iter
;
1068 void *payload
= NULL
;
1069 Error
*local_err
= NULL
;
1071 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, s
->info
.structured_reply
,
1072 qiov
, &reply
, &payload
)
1075 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1077 assert(nbd_reply_is_structured(&reply
));
1079 switch (chunk
->type
) {
1080 case NBD_REPLY_TYPE_OFFSET_DATA
:
1082 * special cased in nbd_co_receive_one_chunk, data is already
1086 case NBD_REPLY_TYPE_OFFSET_HOLE
:
1087 ret
= nbd_parse_offset_hole_payload(s
, &reply
.structured
, payload
,
1088 offset
, qiov
, &local_err
);
1090 nbd_channel_error(s
, ret
);
1091 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1095 if (!nbd_reply_type_is_error(chunk
->type
)) {
1096 /* not allowed reply type */
1097 nbd_channel_error(s
, -EINVAL
);
1098 error_setg(&local_err
,
1099 "Unexpected reply type: %d (%s) for CMD_READ",
1100 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1101 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1109 error_propagate(errp
, iter
.err
);
1110 *request_ret
= iter
.request_ret
;
1114 static int coroutine_fn
nbd_co_receive_blockstatus_reply(BDRVNBDState
*s
,
1115 uint64_t handle
, uint64_t length
,
1117 int *request_ret
, Error
**errp
)
1119 NBDReplyChunkIter iter
;
1121 void *payload
= NULL
;
1122 Error
*local_err
= NULL
;
1123 bool received
= false;
1125 assert(!extent
->length
);
1126 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, &reply
, &payload
) {
1128 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1130 assert(nbd_reply_is_structured(&reply
));
1132 switch (chunk
->type
) {
1133 case NBD_REPLY_TYPE_BLOCK_STATUS
:
1135 nbd_channel_error(s
, -EINVAL
);
1136 error_setg(&local_err
, "Several BLOCK_STATUS chunks in reply");
1137 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1141 ret
= nbd_parse_blockstatus_payload(s
, &reply
.structured
,
1142 payload
, length
, extent
,
1145 nbd_channel_error(s
, ret
);
1146 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1150 if (!nbd_reply_type_is_error(chunk
->type
)) {
1151 nbd_channel_error(s
, -EINVAL
);
1152 error_setg(&local_err
,
1153 "Unexpected reply type: %d (%s) "
1154 "for CMD_BLOCK_STATUS",
1155 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1156 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1164 if (!extent
->length
&& !iter
.request_ret
) {
1165 error_setg(&local_err
, "Server did not reply with any status extents");
1166 nbd_iter_channel_error(&iter
, -EIO
, &local_err
);
1169 error_propagate(errp
, iter
.err
);
1170 *request_ret
= iter
.request_ret
;
1174 static int coroutine_fn
nbd_co_request(BlockDriverState
*bs
, NBDRequest
*request
,
1175 QEMUIOVector
*write_qiov
)
1177 int ret
, request_ret
;
1178 Error
*local_err
= NULL
;
1179 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1181 assert(request
->type
!= NBD_CMD_READ
);
1183 assert(request
->type
== NBD_CMD_WRITE
);
1184 assert(request
->len
== iov_size(write_qiov
->iov
, write_qiov
->niov
));
1186 assert(request
->type
!= NBD_CMD_WRITE
);
1190 ret
= nbd_co_send_request(bs
, request
, write_qiov
);
1195 ret
= nbd_co_receive_return_code(s
, request
->handle
,
1196 &request_ret
, &local_err
);
1198 trace_nbd_co_request_fail(request
->from
, request
->len
,
1199 request
->handle
, request
->flags
,
1201 nbd_cmd_lookup(request
->type
),
1202 ret
, error_get_pretty(local_err
));
1203 error_free(local_err
);
1206 } while (ret
< 0 && nbd_client_will_reconnect(s
));
1208 return ret
? ret
: request_ret
;
1211 static int coroutine_fn
nbd_client_co_preadv(BlockDriverState
*bs
, int64_t offset
,
1212 int64_t bytes
, QEMUIOVector
*qiov
,
1213 BdrvRequestFlags flags
)
1215 int ret
, request_ret
;
1216 Error
*local_err
= NULL
;
1217 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1218 NBDRequest request
= {
1219 .type
= NBD_CMD_READ
,
1224 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1231 * Work around the fact that the block layer doesn't do
1232 * byte-accurate sizing yet - if the read exceeds the server's
1233 * advertised size because the block layer rounded size up, then
1234 * truncate the request to the server and tail-pad with zero.
1236 if (offset
>= s
->info
.size
) {
1237 assert(bytes
< BDRV_SECTOR_SIZE
);
1238 qemu_iovec_memset(qiov
, 0, 0, bytes
);
1241 if (offset
+ bytes
> s
->info
.size
) {
1242 uint64_t slop
= offset
+ bytes
- s
->info
.size
;
1244 assert(slop
< BDRV_SECTOR_SIZE
);
1245 qemu_iovec_memset(qiov
, bytes
- slop
, 0, slop
);
1246 request
.len
-= slop
;
1250 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1255 ret
= nbd_co_receive_cmdread_reply(s
, request
.handle
, offset
, qiov
,
1256 &request_ret
, &local_err
);
1258 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1259 request
.flags
, request
.type
,
1260 nbd_cmd_lookup(request
.type
),
1261 ret
, error_get_pretty(local_err
));
1262 error_free(local_err
);
1265 } while (ret
< 0 && nbd_client_will_reconnect(s
));
1267 return ret
? ret
: request_ret
;
1270 static int coroutine_fn
nbd_client_co_pwritev(BlockDriverState
*bs
, int64_t offset
,
1271 int64_t bytes
, QEMUIOVector
*qiov
,
1272 BdrvRequestFlags flags
)
1274 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1275 NBDRequest request
= {
1276 .type
= NBD_CMD_WRITE
,
1281 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1282 if (flags
& BDRV_REQ_FUA
) {
1283 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1284 request
.flags
|= NBD_CMD_FLAG_FUA
;
1287 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1292 return nbd_co_request(bs
, &request
, qiov
);
1295 static int coroutine_fn
nbd_client_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1296 int64_t bytes
, BdrvRequestFlags flags
)
1298 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1299 NBDRequest request
= {
1300 .type
= NBD_CMD_WRITE_ZEROES
,
1302 .len
= bytes
, /* .len is uint32_t actually */
1305 assert(bytes
<= UINT32_MAX
); /* rely on max_pwrite_zeroes */
1307 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1308 if (!(s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
)) {
1312 if (flags
& BDRV_REQ_FUA
) {
1313 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1314 request
.flags
|= NBD_CMD_FLAG_FUA
;
1316 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1317 request
.flags
|= NBD_CMD_FLAG_NO_HOLE
;
1319 if (flags
& BDRV_REQ_NO_FALLBACK
) {
1320 assert(s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
);
1321 request
.flags
|= NBD_CMD_FLAG_FAST_ZERO
;
1327 return nbd_co_request(bs
, &request
, NULL
);
1330 static int coroutine_fn
nbd_client_co_flush(BlockDriverState
*bs
)
1332 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1333 NBDRequest request
= { .type
= NBD_CMD_FLUSH
};
1335 if (!(s
->info
.flags
& NBD_FLAG_SEND_FLUSH
)) {
1342 return nbd_co_request(bs
, &request
, NULL
);
1345 static int coroutine_fn
nbd_client_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
1348 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1349 NBDRequest request
= {
1350 .type
= NBD_CMD_TRIM
,
1352 .len
= bytes
, /* len is uint32_t */
1355 assert(bytes
<= UINT32_MAX
); /* rely on max_pdiscard */
1357 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1358 if (!(s
->info
.flags
& NBD_FLAG_SEND_TRIM
) || !bytes
) {
1362 return nbd_co_request(bs
, &request
, NULL
);
1365 static int coroutine_fn
nbd_client_co_block_status(
1366 BlockDriverState
*bs
, bool want_zero
, int64_t offset
, int64_t bytes
,
1367 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
1369 int ret
, request_ret
;
1370 NBDExtent extent
= { 0 };
1371 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1372 Error
*local_err
= NULL
;
1374 NBDRequest request
= {
1375 .type
= NBD_CMD_BLOCK_STATUS
,
1377 .len
= MIN(QEMU_ALIGN_DOWN(INT_MAX
, bs
->bl
.request_alignment
),
1378 MIN(bytes
, s
->info
.size
- offset
)),
1379 .flags
= NBD_CMD_FLAG_REQ_ONE
,
1382 if (!s
->info
.base_allocation
) {
1386 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
1390 * Work around the fact that the block layer doesn't do
1391 * byte-accurate sizing yet - if the status request exceeds the
1392 * server's advertised size because the block layer rounded size
1393 * up, we truncated the request to the server (above), or are
1394 * called on just the hole.
1396 if (offset
>= s
->info
.size
) {
1398 assert(bytes
< BDRV_SECTOR_SIZE
);
1399 /* Intentionally don't report offset_valid for the hole */
1400 return BDRV_BLOCK_ZERO
;
1403 if (s
->info
.min_block
) {
1404 assert(QEMU_IS_ALIGNED(request
.len
, s
->info
.min_block
));
1407 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1412 ret
= nbd_co_receive_blockstatus_reply(s
, request
.handle
, bytes
,
1413 &extent
, &request_ret
,
1416 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1417 request
.flags
, request
.type
,
1418 nbd_cmd_lookup(request
.type
),
1419 ret
, error_get_pretty(local_err
));
1420 error_free(local_err
);
1423 } while (ret
< 0 && nbd_client_will_reconnect(s
));
1425 if (ret
< 0 || request_ret
< 0) {
1426 return ret
? ret
: request_ret
;
1429 assert(extent
.length
);
1430 *pnum
= extent
.length
;
1433 return (extent
.flags
& NBD_STATE_HOLE
? 0 : BDRV_BLOCK_DATA
) |
1434 (extent
.flags
& NBD_STATE_ZERO
? BDRV_BLOCK_ZERO
: 0) |
1435 BDRV_BLOCK_OFFSET_VALID
;
1438 static int nbd_client_reopen_prepare(BDRVReopenState
*state
,
1439 BlockReopenQueue
*queue
, Error
**errp
)
1441 BDRVNBDState
*s
= (BDRVNBDState
*)state
->bs
->opaque
;
1443 if ((state
->flags
& BDRV_O_RDWR
) && (s
->info
.flags
& NBD_FLAG_READ_ONLY
)) {
1444 error_setg(errp
, "Can't reopen read-only NBD mount as read/write");
1450 static void nbd_yank(void *opaque
)
1452 BlockDriverState
*bs
= opaque
;
1453 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1455 QEMU_LOCK_GUARD(&s
->requests_lock
);
1456 qio_channel_shutdown(QIO_CHANNEL(s
->ioc
), QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
1457 s
->state
= NBD_CLIENT_QUIT
;
1460 static void nbd_client_close(BlockDriverState
*bs
)
1462 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1463 NBDRequest request
= { .type
= NBD_CMD_DISC
};
1466 nbd_send_request(s
->ioc
, &request
);
1469 nbd_teardown_connection(bs
);
1474 * Parse nbd_open options
1477 static int nbd_parse_uri(const char *filename
, QDict
*options
)
1481 QueryParams
*qp
= NULL
;
1485 uri
= uri_parse(filename
);
1491 if (!g_strcmp0(uri
->scheme
, "nbd")) {
1493 } else if (!g_strcmp0(uri
->scheme
, "nbd+tcp")) {
1495 } else if (!g_strcmp0(uri
->scheme
, "nbd+unix")) {
1502 p
= uri
->path
? uri
->path
: "";
1507 qdict_put_str(options
, "export", p
);
1510 qp
= query_params_parse(uri
->query
);
1511 if (qp
->n
> 1 || (is_unix
&& !qp
->n
) || (!is_unix
&& qp
->n
)) {
1517 /* nbd+unix:///export?socket=path */
1518 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
1522 qdict_put_str(options
, "server.type", "unix");
1523 qdict_put_str(options
, "server.path", qp
->p
[0].value
);
1528 /* nbd[+tcp]://host[:port]/export */
1534 /* strip braces from literal IPv6 address */
1535 if (uri
->server
[0] == '[') {
1536 host
= qstring_from_substr(uri
->server
, 1,
1537 strlen(uri
->server
) - 1);
1539 host
= qstring_from_str(uri
->server
);
1542 qdict_put_str(options
, "server.type", "inet");
1543 qdict_put(options
, "server.host", host
);
1545 port_str
= g_strdup_printf("%d", uri
->port
?: NBD_DEFAULT_PORT
);
1546 qdict_put_str(options
, "server.port", port_str
);
1552 query_params_free(qp
);
1558 static bool nbd_has_filename_options_conflict(QDict
*options
, Error
**errp
)
1560 const QDictEntry
*e
;
1562 for (e
= qdict_first(options
); e
; e
= qdict_next(options
, e
)) {
1563 if (!strcmp(e
->key
, "host") ||
1564 !strcmp(e
->key
, "port") ||
1565 !strcmp(e
->key
, "path") ||
1566 !strcmp(e
->key
, "export") ||
1567 strstart(e
->key
, "server.", NULL
))
1569 error_setg(errp
, "Option '%s' cannot be used with a file name",
1578 static void nbd_parse_filename(const char *filename
, QDict
*options
,
1581 g_autofree
char *file
= NULL
;
1583 const char *host_spec
;
1584 const char *unixpath
;
1586 if (nbd_has_filename_options_conflict(options
, errp
)) {
1590 if (strstr(filename
, "://")) {
1591 int ret
= nbd_parse_uri(filename
, options
);
1593 error_setg(errp
, "No valid URL specified");
1598 file
= g_strdup(filename
);
1600 export_name
= strstr(file
, EN_OPTSTR
);
1602 if (export_name
[strlen(EN_OPTSTR
)] == 0) {
1605 export_name
[0] = 0; /* truncate 'file' */
1606 export_name
+= strlen(EN_OPTSTR
);
1608 qdict_put_str(options
, "export", export_name
);
1611 /* extract the host_spec - fail if it's not nbd:... */
1612 if (!strstart(file
, "nbd:", &host_spec
)) {
1613 error_setg(errp
, "File name string for NBD must start with 'nbd:'");
1621 /* are we a UNIX or TCP socket? */
1622 if (strstart(host_spec
, "unix:", &unixpath
)) {
1623 qdict_put_str(options
, "server.type", "unix");
1624 qdict_put_str(options
, "server.path", unixpath
);
1626 InetSocketAddress
*addr
= g_new(InetSocketAddress
, 1);
1628 if (inet_parse(addr
, host_spec
, errp
)) {
1632 qdict_put_str(options
, "server.type", "inet");
1633 qdict_put_str(options
, "server.host", addr
->host
);
1634 qdict_put_str(options
, "server.port", addr
->port
);
1636 qapi_free_InetSocketAddress(addr
);
1640 static bool nbd_process_legacy_socket_options(QDict
*output_options
,
1641 QemuOpts
*legacy_opts
,
1644 const char *path
= qemu_opt_get(legacy_opts
, "path");
1645 const char *host
= qemu_opt_get(legacy_opts
, "host");
1646 const char *port
= qemu_opt_get(legacy_opts
, "port");
1647 const QDictEntry
*e
;
1649 if (!path
&& !host
&& !port
) {
1653 for (e
= qdict_first(output_options
); e
; e
= qdict_next(output_options
, e
))
1655 if (strstart(e
->key
, "server.", NULL
)) {
1656 error_setg(errp
, "Cannot use 'server' and path/host/port at the "
1663 error_setg(errp
, "path and host may not be used at the same time");
1667 error_setg(errp
, "port may not be used without host");
1671 qdict_put_str(output_options
, "server.type", "unix");
1672 qdict_put_str(output_options
, "server.path", path
);
1674 qdict_put_str(output_options
, "server.type", "inet");
1675 qdict_put_str(output_options
, "server.host", host
);
1676 qdict_put_str(output_options
, "server.port",
1677 port
?: stringify(NBD_DEFAULT_PORT
));
1683 static SocketAddress
*nbd_config(BDRVNBDState
*s
, QDict
*options
,
1686 SocketAddress
*saddr
= NULL
;
1690 qdict_extract_subqdict(options
, &addr
, "server.");
1691 if (!qdict_size(addr
)) {
1692 error_setg(errp
, "NBD server address missing");
1696 iv
= qobject_input_visitor_new_flat_confused(addr
, errp
);
1701 if (!visit_type_SocketAddress(iv
, NULL
, &saddr
, errp
)) {
1705 if (socket_address_parse_named_fd(saddr
, errp
) < 0) {
1706 qapi_free_SocketAddress(saddr
);
1712 qobject_unref(addr
);
1717 static QCryptoTLSCreds
*nbd_get_tls_creds(const char *id
, Error
**errp
)
1720 QCryptoTLSCreds
*creds
;
1722 obj
= object_resolve_path_component(
1723 object_get_objects_root(), id
);
1725 error_setg(errp
, "No TLS credentials with id '%s'",
1729 creds
= (QCryptoTLSCreds
*)
1730 object_dynamic_cast(obj
, TYPE_QCRYPTO_TLS_CREDS
);
1732 error_setg(errp
, "Object with id '%s' is not TLS credentials",
1737 if (!qcrypto_tls_creds_check_endpoint(creds
,
1738 QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
,
1747 static QemuOptsList nbd_runtime_opts
= {
1749 .head
= QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts
.head
),
1753 .type
= QEMU_OPT_STRING
,
1754 .help
= "TCP host to connect to",
1758 .type
= QEMU_OPT_STRING
,
1759 .help
= "TCP port to connect to",
1763 .type
= QEMU_OPT_STRING
,
1764 .help
= "Unix socket path to connect to",
1768 .type
= QEMU_OPT_STRING
,
1769 .help
= "Name of the NBD export to open",
1772 .name
= "tls-creds",
1773 .type
= QEMU_OPT_STRING
,
1774 .help
= "ID of the TLS credentials to use",
1777 .name
= "tls-hostname",
1778 .type
= QEMU_OPT_STRING
,
1779 .help
= "Override hostname for validating TLS x509 certificate",
1782 .name
= "x-dirty-bitmap",
1783 .type
= QEMU_OPT_STRING
,
1784 .help
= "experimental: expose named dirty bitmap in place of "
1788 .name
= "reconnect-delay",
1789 .type
= QEMU_OPT_NUMBER
,
1790 .help
= "On an unexpected disconnect, the nbd client tries to "
1791 "connect again until succeeding or encountering a serious "
1792 "error. During the first @reconnect-delay seconds, all "
1793 "requests are paused and will be rerun on a successful "
1794 "reconnect. After that time, any delayed requests and all "
1795 "future requests before a successful reconnect will "
1796 "immediately fail. Default 0",
1799 .name
= "open-timeout",
1800 .type
= QEMU_OPT_NUMBER
,
1801 .help
= "In seconds. If zero, the nbd driver tries the connection "
1802 "only once, and fails to open if the connection fails. "
1803 "If non-zero, the nbd driver will repeat connection "
1804 "attempts until successful or until @open-timeout seconds "
1805 "have elapsed. Default 0",
1807 { /* end of list */ }
1811 static int nbd_process_options(BlockDriverState
*bs
, QDict
*options
,
1814 BDRVNBDState
*s
= bs
->opaque
;
1818 opts
= qemu_opts_create(&nbd_runtime_opts
, NULL
, 0, &error_abort
);
1819 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1823 /* Translate @host, @port, and @path to a SocketAddress */
1824 if (!nbd_process_legacy_socket_options(options
, opts
, errp
)) {
1828 /* Pop the config into our state object. Exit if invalid. */
1829 s
->saddr
= nbd_config(s
, options
, errp
);
1834 s
->export
= g_strdup(qemu_opt_get(opts
, "export"));
1835 if (s
->export
&& strlen(s
->export
) > NBD_MAX_STRING_SIZE
) {
1836 error_setg(errp
, "export name too long to send to server");
1840 s
->tlscredsid
= g_strdup(qemu_opt_get(opts
, "tls-creds"));
1841 if (s
->tlscredsid
) {
1842 s
->tlscreds
= nbd_get_tls_creds(s
->tlscredsid
, errp
);
1847 s
->tlshostname
= g_strdup(qemu_opt_get(opts
, "tls-hostname"));
1848 if (!s
->tlshostname
&&
1849 s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
1850 s
->tlshostname
= g_strdup(s
->saddr
->u
.inet
.host
);
1854 s
->x_dirty_bitmap
= g_strdup(qemu_opt_get(opts
, "x-dirty-bitmap"));
1855 if (s
->x_dirty_bitmap
&& strlen(s
->x_dirty_bitmap
) > NBD_MAX_STRING_SIZE
) {
1856 error_setg(errp
, "x-dirty-bitmap query too long to send to server");
1860 s
->reconnect_delay
= qemu_opt_get_number(opts
, "reconnect-delay", 0);
1861 s
->open_timeout
= qemu_opt_get_number(opts
, "open-timeout", 0);
1866 qemu_opts_del(opts
);
1870 static int nbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1874 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1877 qemu_mutex_init(&s
->requests_lock
);
1878 qemu_co_queue_init(&s
->free_sema
);
1879 qemu_co_mutex_init(&s
->send_mutex
);
1880 qemu_co_mutex_init(&s
->receive_mutex
);
1882 if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
), errp
)) {
1886 ret
= nbd_process_options(bs
, options
, errp
);
1891 s
->conn
= nbd_client_connection_new(s
->saddr
, true, s
->export
,
1892 s
->x_dirty_bitmap
, s
->tlscreds
,
1895 if (s
->open_timeout
) {
1896 nbd_client_connection_enable_retry(s
->conn
);
1897 open_timer_init(s
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) +
1898 s
->open_timeout
* NANOSECONDS_PER_SECOND
);
1901 s
->state
= NBD_CLIENT_CONNECTING_WAIT
;
1902 ret
= nbd_do_establish_connection(bs
, true, errp
);
1908 * The connect attempt is done, so we no longer need this timer.
1909 * Delete it, because we do not want it to be around when this node
1910 * is drained or closed.
1914 nbd_client_connection_enable_retry(s
->conn
);
1920 nbd_clear_bdrvstate(bs
);
1924 static int coroutine_fn
nbd_co_flush(BlockDriverState
*bs
)
1926 return nbd_client_co_flush(bs
);
1929 static void nbd_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1931 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1932 uint32_t min
= s
->info
.min_block
;
1933 uint32_t max
= MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE
, s
->info
.max_block
);
1936 * If the server did not advertise an alignment:
1937 * - a size that is not sector-aligned implies that an alignment
1938 * of 1 can be used to access those tail bytes
1939 * - advertisement of block status requires an alignment of 1, so
1940 * that we don't violate block layer constraints that block
1941 * status is always aligned (as we can't control whether the
1942 * server will report sub-sector extents, such as a hole at EOF
1943 * on an unaligned POSIX file)
1944 * - otherwise, assume the server is so old that we are safer avoiding
1945 * sub-sector requests
1948 min
= (!QEMU_IS_ALIGNED(s
->info
.size
, BDRV_SECTOR_SIZE
) ||
1949 s
->info
.base_allocation
) ? 1 : BDRV_SECTOR_SIZE
;
1952 bs
->bl
.request_alignment
= min
;
1953 bs
->bl
.max_pdiscard
= QEMU_ALIGN_DOWN(INT_MAX
, min
);
1954 bs
->bl
.max_pwrite_zeroes
= max
;
1955 bs
->bl
.max_transfer
= max
;
1957 if (s
->info
.opt_block
&&
1958 s
->info
.opt_block
> bs
->bl
.opt_transfer
) {
1959 bs
->bl
.opt_transfer
= s
->info
.opt_block
;
1963 static void nbd_close(BlockDriverState
*bs
)
1965 nbd_client_close(bs
);
1966 nbd_clear_bdrvstate(bs
);
1970 * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1971 * to a smaller size with exact=false, there is no reason to fail the
1974 * Preallocation mode is ignored since it does not seems useful to fail when
1975 * we never change anything.
1977 static int coroutine_fn
nbd_co_truncate(BlockDriverState
*bs
, int64_t offset
,
1978 bool exact
, PreallocMode prealloc
,
1979 BdrvRequestFlags flags
, Error
**errp
)
1981 BDRVNBDState
*s
= bs
->opaque
;
1983 if (offset
!= s
->info
.size
&& exact
) {
1984 error_setg(errp
, "Cannot resize NBD nodes");
1988 if (offset
> s
->info
.size
) {
1989 error_setg(errp
, "Cannot grow NBD nodes");
1996 static int64_t nbd_getlength(BlockDriverState
*bs
)
1998 BDRVNBDState
*s
= bs
->opaque
;
2000 return s
->info
.size
;
2003 static void nbd_refresh_filename(BlockDriverState
*bs
)
2005 BDRVNBDState
*s
= bs
->opaque
;
2006 const char *host
= NULL
, *port
= NULL
, *path
= NULL
;
2009 if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
2010 const InetSocketAddress
*inet
= &s
->saddr
->u
.inet
;
2011 if (!inet
->has_ipv4
&& !inet
->has_ipv6
&& !inet
->has_to
) {
2015 } else if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_UNIX
) {
2016 path
= s
->saddr
->u
.q_unix
.path
;
2017 } /* else can't represent as pseudo-filename */
2019 if (path
&& s
->export
) {
2020 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2021 "nbd+unix:///%s?socket=%s", s
->export
, path
);
2022 } else if (path
&& !s
->export
) {
2023 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2024 "nbd+unix://?socket=%s", path
);
2025 } else if (host
&& s
->export
) {
2026 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2027 "nbd://%s:%s/%s", host
, port
, s
->export
);
2028 } else if (host
&& !s
->export
) {
2029 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2030 "nbd://%s:%s", host
, port
);
2032 if (len
>= sizeof(bs
->exact_filename
)) {
2033 /* Name is too long to represent exactly, so leave it empty. */
2034 bs
->exact_filename
[0] = '\0';
2038 static char *nbd_dirname(BlockDriverState
*bs
, Error
**errp
)
2040 /* The generic bdrv_dirname() implementation is able to work out some
2041 * directory name for NBD nodes, but that would be wrong. So far there is no
2042 * specification for how "export paths" would work, so NBD does not have
2043 * directory names. */
2044 error_setg(errp
, "Cannot generate a base directory for NBD nodes");
2048 static const char *const nbd_strong_runtime_opts
[] = {
2060 static void nbd_cancel_in_flight(BlockDriverState
*bs
)
2062 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
2064 reconnect_delay_timer_del(s
);
2066 qemu_mutex_lock(&s
->requests_lock
);
2067 if (s
->state
== NBD_CLIENT_CONNECTING_WAIT
) {
2068 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
2070 qemu_mutex_unlock(&s
->requests_lock
);
2072 nbd_co_establish_connection_cancel(s
->conn
);
2075 static void nbd_attach_aio_context(BlockDriverState
*bs
,
2076 AioContext
*new_context
)
2078 BDRVNBDState
*s
= bs
->opaque
;
2080 /* The open_timer is used only during nbd_open() */
2081 assert(!s
->open_timer
);
2084 * The reconnect_delay_timer is scheduled in I/O paths when the
2085 * connection is lost, to cancel the reconnection attempt after a
2086 * given time. Once this attempt is done (successfully or not),
2087 * nbd_reconnect_attempt() ensures the timer is deleted before the
2088 * respective I/O request is resumed.
2089 * Since the AioContext can only be changed when a node is drained,
2090 * the reconnect_delay_timer cannot be active here.
2092 assert(!s
->reconnect_delay_timer
);
2095 qio_channel_attach_aio_context(s
->ioc
, new_context
);
2099 static void nbd_detach_aio_context(BlockDriverState
*bs
)
2101 BDRVNBDState
*s
= bs
->opaque
;
2103 assert(!s
->open_timer
);
2104 assert(!s
->reconnect_delay_timer
);
2107 qio_channel_detach_aio_context(s
->ioc
);
2111 static BlockDriver bdrv_nbd
= {
2112 .format_name
= "nbd",
2113 .protocol_name
= "nbd",
2114 .instance_size
= sizeof(BDRVNBDState
),
2115 .bdrv_parse_filename
= nbd_parse_filename
,
2116 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2117 .create_opts
= &bdrv_create_opts_simple
,
2118 .bdrv_file_open
= nbd_open
,
2119 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2120 .bdrv_co_preadv
= nbd_client_co_preadv
,
2121 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2122 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2123 .bdrv_close
= nbd_close
,
2124 .bdrv_co_flush_to_os
= nbd_co_flush
,
2125 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2126 .bdrv_refresh_limits
= nbd_refresh_limits
,
2127 .bdrv_co_truncate
= nbd_co_truncate
,
2128 .bdrv_getlength
= nbd_getlength
,
2129 .bdrv_refresh_filename
= nbd_refresh_filename
,
2130 .bdrv_co_block_status
= nbd_client_co_block_status
,
2131 .bdrv_dirname
= nbd_dirname
,
2132 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2133 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2135 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2136 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2139 static BlockDriver bdrv_nbd_tcp
= {
2140 .format_name
= "nbd",
2141 .protocol_name
= "nbd+tcp",
2142 .instance_size
= sizeof(BDRVNBDState
),
2143 .bdrv_parse_filename
= nbd_parse_filename
,
2144 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2145 .create_opts
= &bdrv_create_opts_simple
,
2146 .bdrv_file_open
= nbd_open
,
2147 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2148 .bdrv_co_preadv
= nbd_client_co_preadv
,
2149 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2150 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2151 .bdrv_close
= nbd_close
,
2152 .bdrv_co_flush_to_os
= nbd_co_flush
,
2153 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2154 .bdrv_refresh_limits
= nbd_refresh_limits
,
2155 .bdrv_co_truncate
= nbd_co_truncate
,
2156 .bdrv_getlength
= nbd_getlength
,
2157 .bdrv_refresh_filename
= nbd_refresh_filename
,
2158 .bdrv_co_block_status
= nbd_client_co_block_status
,
2159 .bdrv_dirname
= nbd_dirname
,
2160 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2161 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2163 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2164 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2167 static BlockDriver bdrv_nbd_unix
= {
2168 .format_name
= "nbd",
2169 .protocol_name
= "nbd+unix",
2170 .instance_size
= sizeof(BDRVNBDState
),
2171 .bdrv_parse_filename
= nbd_parse_filename
,
2172 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2173 .create_opts
= &bdrv_create_opts_simple
,
2174 .bdrv_file_open
= nbd_open
,
2175 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2176 .bdrv_co_preadv
= nbd_client_co_preadv
,
2177 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2178 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2179 .bdrv_close
= nbd_close
,
2180 .bdrv_co_flush_to_os
= nbd_co_flush
,
2181 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2182 .bdrv_refresh_limits
= nbd_refresh_limits
,
2183 .bdrv_co_truncate
= nbd_co_truncate
,
2184 .bdrv_getlength
= nbd_getlength
,
2185 .bdrv_refresh_filename
= nbd_refresh_filename
,
2186 .bdrv_co_block_status
= nbd_client_co_block_status
,
2187 .bdrv_dirname
= nbd_dirname
,
2188 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2189 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2191 .bdrv_attach_aio_context
= nbd_attach_aio_context
,
2192 .bdrv_detach_aio_context
= nbd_detach_aio_context
,
2195 static void bdrv_nbd_init(void)
2197 bdrv_register(&bdrv_nbd
);
2198 bdrv_register(&bdrv_nbd_tcp
);
2199 bdrv_register(&bdrv_nbd_unix
);
2202 block_init(bdrv_nbd_init
);