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
; /* waiting for connection_co? */
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 */
76 Coroutine
*connection_co
;
77 Coroutine
*teardown_co
;
78 QemuCoSleep reconnect_sleep
;
80 bool wait_drained_end
;
85 QEMUTimer
*reconnect_delay_timer
;
87 NBDClientRequest requests
[MAX_NBD_REQUESTS
];
91 /* Connection parameters */
92 uint32_t reconnect_delay
;
94 char *export
, *tlscredsid
;
95 QCryptoTLSCreds
*tlscreds
;
100 NBDClientConnection
*conn
;
103 static void nbd_yank(void *opaque
);
105 static void nbd_clear_bdrvstate(BlockDriverState
*bs
)
107 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
109 nbd_client_connection_release(s
->conn
);
112 yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
));
114 object_unref(OBJECT(s
->tlscreds
));
115 qapi_free_SocketAddress(s
->saddr
);
119 g_free(s
->tlscredsid
);
120 s
->tlscredsid
= NULL
;
121 g_free(s
->x_dirty_bitmap
);
122 s
->x_dirty_bitmap
= NULL
;
125 static bool nbd_client_connected(BDRVNBDState
*s
)
127 return qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTED
;
130 static void nbd_channel_error(BDRVNBDState
*s
, int ret
)
133 if (nbd_client_connected(s
)) {
134 s
->state
= s
->reconnect_delay
? NBD_CLIENT_CONNECTING_WAIT
:
135 NBD_CLIENT_CONNECTING_NOWAIT
;
138 if (nbd_client_connected(s
)) {
139 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
141 s
->state
= NBD_CLIENT_QUIT
;
145 static void nbd_recv_coroutines_wake_all(BDRVNBDState
*s
)
149 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
150 NBDClientRequest
*req
= &s
->requests
[i
];
152 if (req
->coroutine
&& req
->receiving
) {
153 req
->receiving
= false;
154 aio_co_wake(req
->coroutine
);
159 static void reconnect_delay_timer_del(BDRVNBDState
*s
)
161 if (s
->reconnect_delay_timer
) {
162 timer_free(s
->reconnect_delay_timer
);
163 s
->reconnect_delay_timer
= NULL
;
167 static void reconnect_delay_timer_cb(void *opaque
)
169 BDRVNBDState
*s
= opaque
;
171 if (qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
) {
172 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
173 while (qemu_co_enter_next(&s
->free_sema
, NULL
)) {
174 /* Resume all queued requests */
178 reconnect_delay_timer_del(s
);
181 static void reconnect_delay_timer_init(BDRVNBDState
*s
, uint64_t expire_time_ns
)
183 if (qatomic_load_acquire(&s
->state
) != NBD_CLIENT_CONNECTING_WAIT
) {
187 assert(!s
->reconnect_delay_timer
);
188 s
->reconnect_delay_timer
= aio_timer_new(bdrv_get_aio_context(s
->bs
),
191 reconnect_delay_timer_cb
, s
);
192 timer_mod(s
->reconnect_delay_timer
, expire_time_ns
);
195 static void nbd_client_detach_aio_context(BlockDriverState
*bs
)
197 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
199 /* Timer is deleted in nbd_client_co_drain_begin() */
200 assert(!s
->reconnect_delay_timer
);
202 * If reconnect is in progress we may have no ->ioc. It will be
203 * re-instantiated in the proper aio context once the connection is
207 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
211 static void nbd_client_attach_aio_context_bh(void *opaque
)
213 BlockDriverState
*bs
= opaque
;
214 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
216 if (s
->connection_co
) {
218 * The node is still drained, so we know the coroutine has yielded in
219 * nbd_read_eof(), the only place where bs->in_flight can reach 0, or
220 * it is entered for the first time. Both places are safe for entering
223 qemu_aio_coroutine_enter(bs
->aio_context
, s
->connection_co
);
225 bdrv_dec_in_flight(bs
);
228 static void nbd_client_attach_aio_context(BlockDriverState
*bs
,
229 AioContext
*new_context
)
231 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
234 * s->connection_co is either yielded from nbd_receive_reply or from
235 * nbd_co_reconnect_loop()
237 if (nbd_client_connected(s
)) {
238 qio_channel_attach_aio_context(QIO_CHANNEL(s
->ioc
), new_context
);
241 bdrv_inc_in_flight(bs
);
244 * Need to wait here for the BH to run because the BH must run while the
245 * node is still drained.
247 aio_wait_bh_oneshot(new_context
, nbd_client_attach_aio_context_bh
, bs
);
250 static void coroutine_fn
nbd_client_co_drain_begin(BlockDriverState
*bs
)
252 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
255 qemu_co_sleep_wake(&s
->reconnect_sleep
);
257 nbd_co_establish_connection_cancel(s
->conn
);
259 reconnect_delay_timer_del(s
);
261 if (qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
) {
262 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
263 qemu_co_queue_restart_all(&s
->free_sema
);
267 static void coroutine_fn
nbd_client_co_drain_end(BlockDriverState
*bs
)
269 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
272 if (s
->wait_drained_end
) {
273 s
->wait_drained_end
= false;
274 aio_co_wake(s
->connection_co
);
279 static void nbd_teardown_connection(BlockDriverState
*bs
)
281 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
284 /* finish any pending coroutines */
285 qio_channel_shutdown(s
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
288 s
->state
= NBD_CLIENT_QUIT
;
289 if (s
->connection_co
) {
290 qemu_co_sleep_wake(&s
->reconnect_sleep
);
291 nbd_co_establish_connection_cancel(s
->conn
);
293 if (qemu_in_coroutine()) {
294 s
->teardown_co
= qemu_coroutine_self();
295 /* connection_co resumes us when it terminates */
296 qemu_coroutine_yield();
297 s
->teardown_co
= NULL
;
299 BDRV_POLL_WHILE(bs
, s
->connection_co
);
301 assert(!s
->connection_co
);
304 static bool nbd_client_connecting(BDRVNBDState
*s
)
306 NBDClientState state
= qatomic_load_acquire(&s
->state
);
307 return state
== NBD_CLIENT_CONNECTING_WAIT
||
308 state
== NBD_CLIENT_CONNECTING_NOWAIT
;
311 static bool nbd_client_connecting_wait(BDRVNBDState
*s
)
313 return qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
;
317 * Update @bs with information learned during a completed negotiation process.
318 * Return failure if the server's advertised options are incompatible with the
321 static int nbd_handle_updated_info(BlockDriverState
*bs
, Error
**errp
)
323 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
326 if (s
->x_dirty_bitmap
) {
327 if (!s
->info
.base_allocation
) {
328 error_setg(errp
, "requested x-dirty-bitmap %s not found",
332 if (strcmp(s
->x_dirty_bitmap
, "qemu:allocation-depth") == 0) {
333 s
->alloc_depth
= true;
337 if (s
->info
.flags
& NBD_FLAG_READ_ONLY
) {
338 ret
= bdrv_apply_auto_read_only(bs
, "NBD export is read-only", errp
);
344 if (s
->info
.flags
& NBD_FLAG_SEND_FUA
) {
345 bs
->supported_write_flags
= BDRV_REQ_FUA
;
346 bs
->supported_zero_flags
|= BDRV_REQ_FUA
;
349 if (s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
) {
350 bs
->supported_zero_flags
|= BDRV_REQ_MAY_UNMAP
;
351 if (s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
) {
352 bs
->supported_zero_flags
|= BDRV_REQ_NO_FALLBACK
;
356 trace_nbd_client_handshake_success(s
->export
);
361 int coroutine_fn
nbd_co_do_establish_connection(BlockDriverState
*bs
,
364 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
369 s
->ioc
= nbd_co_establish_connection(s
->conn
, &s
->info
, true, errp
);
371 return -ECONNREFUSED
;
374 ret
= nbd_handle_updated_info(s
->bs
, NULL
);
377 * We have connected, but must fail for other reasons.
378 * Send NBD_CMD_DISC as a courtesy to the server.
380 NBDRequest request
= { .type
= NBD_CMD_DISC
};
382 nbd_send_request(s
->ioc
, &request
);
384 object_unref(OBJECT(s
->ioc
));
390 qio_channel_set_blocking(s
->ioc
, false, NULL
);
391 qio_channel_attach_aio_context(s
->ioc
, bdrv_get_aio_context(bs
));
393 yank_register_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
), nbd_yank
,
396 /* successfully connected */
397 s
->state
= NBD_CLIENT_CONNECTED
;
398 qemu_co_queue_restart_all(&s
->free_sema
);
403 static coroutine_fn
void nbd_reconnect_attempt(BDRVNBDState
*s
)
405 if (!nbd_client_connecting(s
)) {
409 /* Wait for completion of all in-flight requests */
411 qemu_co_mutex_lock(&s
->send_mutex
);
413 while (s
->in_flight
> 0) {
414 qemu_co_mutex_unlock(&s
->send_mutex
);
415 nbd_recv_coroutines_wake_all(s
);
416 s
->wait_in_flight
= true;
417 qemu_coroutine_yield();
418 s
->wait_in_flight
= false;
419 qemu_co_mutex_lock(&s
->send_mutex
);
422 qemu_co_mutex_unlock(&s
->send_mutex
);
424 if (!nbd_client_connecting(s
)) {
429 * Now we are sure that nobody is accessing the channel, and no one will
430 * try until we set the state to CONNECTED.
433 /* Finalize previous connection if any */
435 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
436 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
438 object_unref(OBJECT(s
->ioc
));
442 nbd_co_do_establish_connection(s
->bs
, NULL
);
445 static coroutine_fn
void nbd_co_reconnect_loop(BDRVNBDState
*s
)
447 uint64_t timeout
= 1 * NANOSECONDS_PER_SECOND
;
448 uint64_t max_timeout
= 16 * NANOSECONDS_PER_SECOND
;
450 if (qatomic_load_acquire(&s
->state
) == NBD_CLIENT_CONNECTING_WAIT
) {
451 reconnect_delay_timer_init(s
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) +
452 s
->reconnect_delay
* NANOSECONDS_PER_SECOND
);
455 nbd_reconnect_attempt(s
);
457 while (nbd_client_connecting(s
)) {
459 bdrv_dec_in_flight(s
->bs
);
460 s
->wait_drained_end
= true;
463 * We may be entered once from nbd_client_attach_aio_context_bh
464 * and then from nbd_client_co_drain_end. So here is a loop.
466 qemu_coroutine_yield();
468 bdrv_inc_in_flight(s
->bs
);
470 qemu_co_sleep_ns_wakeable(&s
->reconnect_sleep
,
471 QEMU_CLOCK_REALTIME
, timeout
);
475 if (timeout
< max_timeout
) {
480 nbd_reconnect_attempt(s
);
483 reconnect_delay_timer_del(s
);
486 static coroutine_fn
void nbd_connection_entry(void *opaque
)
488 BDRVNBDState
*s
= opaque
;
491 Error
*local_err
= NULL
;
493 while (qatomic_load_acquire(&s
->state
) != NBD_CLIENT_QUIT
) {
495 * The NBD client can only really be considered idle when it has
496 * yielded from qio_channel_readv_all_eof(), waiting for data. This is
497 * the point where the additional scheduled coroutine entry happens
498 * after nbd_client_attach_aio_context().
500 * Therefore we keep an additional in_flight reference all the time and
501 * only drop it temporarily here.
504 if (nbd_client_connecting(s
)) {
505 nbd_co_reconnect_loop(s
);
508 if (!nbd_client_connected(s
)) {
512 assert(s
->reply
.handle
== 0);
513 ret
= nbd_receive_reply(s
->bs
, s
->ioc
, &s
->reply
, &local_err
);
516 trace_nbd_read_reply_entry_fail(ret
, error_get_pretty(local_err
));
517 error_free(local_err
);
521 nbd_channel_error(s
, ret
? ret
: -EIO
);
526 * There's no need for a mutex on the receive side, because the
527 * handler acts as a synchronization point and ensures that only
528 * one coroutine is called until the reply finishes.
530 i
= HANDLE_TO_INDEX(s
, s
->reply
.handle
);
531 if (i
>= MAX_NBD_REQUESTS
||
532 !s
->requests
[i
].coroutine
||
533 !s
->requests
[i
].receiving
||
534 (nbd_reply_is_structured(&s
->reply
) && !s
->info
.structured_reply
))
536 nbd_channel_error(s
, -EINVAL
);
541 * We're woken up again by the request itself. Note that there
542 * is no race between yielding and reentering connection_co. This
545 * - if the request runs on the same AioContext, it is only
546 * entered after we yield
548 * - if the request runs on a different AioContext, reentering
549 * connection_co happens through a bottom half, which can only
550 * run after we yield.
552 s
->requests
[i
].receiving
= false;
553 aio_co_wake(s
->requests
[i
].coroutine
);
554 qemu_coroutine_yield();
557 qemu_co_queue_restart_all(&s
->free_sema
);
558 nbd_recv_coroutines_wake_all(s
);
559 bdrv_dec_in_flight(s
->bs
);
561 s
->connection_co
= NULL
;
563 qio_channel_detach_aio_context(QIO_CHANNEL(s
->ioc
));
564 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s
->bs
->node_name
),
566 object_unref(OBJECT(s
->ioc
));
570 if (s
->teardown_co
) {
571 aio_co_wake(s
->teardown_co
);
576 static int nbd_co_send_request(BlockDriverState
*bs
,
580 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
583 qemu_co_mutex_lock(&s
->send_mutex
);
584 while (s
->in_flight
== MAX_NBD_REQUESTS
|| nbd_client_connecting_wait(s
)) {
585 qemu_co_queue_wait(&s
->free_sema
, &s
->send_mutex
);
588 if (!nbd_client_connected(s
)) {
595 for (i
= 0; i
< MAX_NBD_REQUESTS
; i
++) {
596 if (s
->requests
[i
].coroutine
== NULL
) {
601 g_assert(qemu_in_coroutine());
602 assert(i
< MAX_NBD_REQUESTS
);
604 s
->requests
[i
].coroutine
= qemu_coroutine_self();
605 s
->requests
[i
].offset
= request
->from
;
606 s
->requests
[i
].receiving
= false;
608 request
->handle
= INDEX_TO_HANDLE(s
, i
);
613 qio_channel_set_cork(s
->ioc
, true);
614 rc
= nbd_send_request(s
->ioc
, request
);
615 if (nbd_client_connected(s
) && rc
>= 0) {
616 if (qio_channel_writev_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
620 } else if (rc
>= 0) {
623 qio_channel_set_cork(s
->ioc
, false);
625 rc
= nbd_send_request(s
->ioc
, request
);
630 nbd_channel_error(s
, rc
);
632 s
->requests
[i
].coroutine
= NULL
;
635 if (s
->in_flight
== 0 && s
->wait_in_flight
) {
636 aio_co_wake(s
->connection_co
);
638 qemu_co_queue_next(&s
->free_sema
);
641 qemu_co_mutex_unlock(&s
->send_mutex
);
645 static inline uint16_t payload_advance16(uint8_t **payload
)
648 return lduw_be_p(*payload
- 2);
651 static inline uint32_t payload_advance32(uint8_t **payload
)
654 return ldl_be_p(*payload
- 4);
657 static inline uint64_t payload_advance64(uint8_t **payload
)
660 return ldq_be_p(*payload
- 8);
663 static int nbd_parse_offset_hole_payload(BDRVNBDState
*s
,
664 NBDStructuredReplyChunk
*chunk
,
665 uint8_t *payload
, uint64_t orig_offset
,
666 QEMUIOVector
*qiov
, Error
**errp
)
671 if (chunk
->length
!= sizeof(offset
) + sizeof(hole_size
)) {
672 error_setg(errp
, "Protocol error: invalid payload for "
673 "NBD_REPLY_TYPE_OFFSET_HOLE");
677 offset
= payload_advance64(&payload
);
678 hole_size
= payload_advance32(&payload
);
680 if (!hole_size
|| offset
< orig_offset
|| hole_size
> qiov
->size
||
681 offset
> orig_offset
+ qiov
->size
- hole_size
) {
682 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
686 if (s
->info
.min_block
&&
687 !QEMU_IS_ALIGNED(hole_size
, s
->info
.min_block
)) {
688 trace_nbd_structured_read_compliance("hole");
691 qemu_iovec_memset(qiov
, offset
- orig_offset
, 0, hole_size
);
697 * nbd_parse_blockstatus_payload
698 * Based on our request, we expect only one extent in reply, for the
699 * base:allocation context.
701 static int nbd_parse_blockstatus_payload(BDRVNBDState
*s
,
702 NBDStructuredReplyChunk
*chunk
,
703 uint8_t *payload
, uint64_t orig_length
,
704 NBDExtent
*extent
, Error
**errp
)
708 /* The server succeeded, so it must have sent [at least] one extent */
709 if (chunk
->length
< sizeof(context_id
) + sizeof(*extent
)) {
710 error_setg(errp
, "Protocol error: invalid payload for "
711 "NBD_REPLY_TYPE_BLOCK_STATUS");
715 context_id
= payload_advance32(&payload
);
716 if (s
->info
.context_id
!= context_id
) {
717 error_setg(errp
, "Protocol error: unexpected context id %d for "
718 "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
719 "id is %d", context_id
,
724 extent
->length
= payload_advance32(&payload
);
725 extent
->flags
= payload_advance32(&payload
);
727 if (extent
->length
== 0) {
728 error_setg(errp
, "Protocol error: server sent status chunk with "
734 * A server sending unaligned block status is in violation of the
735 * protocol, but as qemu-nbd 3.1 is such a server (at least for
736 * POSIX files that are not a multiple of 512 bytes, since qemu
737 * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
738 * still sees an implicit hole beyond the real EOF), it's nicer to
739 * work around the misbehaving server. If the request included
740 * more than the final unaligned block, truncate it back to an
741 * aligned result; if the request was only the final block, round
742 * up to the full block and change the status to fully-allocated
743 * (always a safe status, even if it loses information).
745 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(extent
->length
,
746 s
->info
.min_block
)) {
747 trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
748 if (extent
->length
> s
->info
.min_block
) {
749 extent
->length
= QEMU_ALIGN_DOWN(extent
->length
,
752 extent
->length
= s
->info
.min_block
;
758 * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
759 * sent us any more than one extent, nor should it have included
760 * status beyond our request in that extent. However, it's easy
761 * enough to ignore the server's noncompliance without killing the
762 * connection; just ignore trailing extents, and clamp things to
763 * the length of our request.
765 if (chunk
->length
> sizeof(context_id
) + sizeof(*extent
)) {
766 trace_nbd_parse_blockstatus_compliance("more than one extent");
768 if (extent
->length
> orig_length
) {
769 extent
->length
= orig_length
;
770 trace_nbd_parse_blockstatus_compliance("extent length too large");
774 * HACK: if we are using x-dirty-bitmaps to access
775 * qemu:allocation-depth, treat all depths > 2 the same as 2,
776 * since nbd_client_co_block_status is only expecting the low two
779 if (s
->alloc_depth
&& extent
->flags
> 2) {
787 * nbd_parse_error_payload
788 * on success @errp contains message describing nbd error reply
790 static int nbd_parse_error_payload(NBDStructuredReplyChunk
*chunk
,
791 uint8_t *payload
, int *request_ret
,
795 uint16_t message_size
;
797 assert(chunk
->type
& (1 << 15));
799 if (chunk
->length
< sizeof(error
) + sizeof(message_size
)) {
801 "Protocol error: invalid payload for structured error");
805 error
= nbd_errno_to_system_errno(payload_advance32(&payload
));
807 error_setg(errp
, "Protocol error: server sent structured error chunk "
812 *request_ret
= -error
;
813 message_size
= payload_advance16(&payload
);
815 if (message_size
> chunk
->length
- sizeof(error
) - sizeof(message_size
)) {
816 error_setg(errp
, "Protocol error: server sent structured error chunk "
817 "with incorrect message size");
821 /* TODO: Add a trace point to mention the server complaint */
823 /* TODO handle ERROR_OFFSET */
828 static int nbd_co_receive_offset_data_payload(BDRVNBDState
*s
,
829 uint64_t orig_offset
,
830 QEMUIOVector
*qiov
, Error
**errp
)
832 QEMUIOVector sub_qiov
;
836 NBDStructuredReplyChunk
*chunk
= &s
->reply
.structured
;
838 assert(nbd_reply_is_structured(&s
->reply
));
840 /* The NBD spec requires at least one byte of payload */
841 if (chunk
->length
<= sizeof(offset
)) {
842 error_setg(errp
, "Protocol error: invalid payload for "
843 "NBD_REPLY_TYPE_OFFSET_DATA");
847 if (nbd_read64(s
->ioc
, &offset
, "OFFSET_DATA offset", errp
) < 0) {
851 data_size
= chunk
->length
- sizeof(offset
);
853 if (offset
< orig_offset
|| data_size
> qiov
->size
||
854 offset
> orig_offset
+ qiov
->size
- data_size
) {
855 error_setg(errp
, "Protocol error: server sent chunk exceeding requested"
859 if (s
->info
.min_block
&& !QEMU_IS_ALIGNED(data_size
, s
->info
.min_block
)) {
860 trace_nbd_structured_read_compliance("data");
863 qemu_iovec_init(&sub_qiov
, qiov
->niov
);
864 qemu_iovec_concat(&sub_qiov
, qiov
, offset
- orig_offset
, data_size
);
865 ret
= qio_channel_readv_all(s
->ioc
, sub_qiov
.iov
, sub_qiov
.niov
, errp
);
866 qemu_iovec_destroy(&sub_qiov
);
868 return ret
< 0 ? -EIO
: 0;
871 #define NBD_MAX_MALLOC_PAYLOAD 1000
872 static coroutine_fn
int nbd_co_receive_structured_payload(
873 BDRVNBDState
*s
, void **payload
, Error
**errp
)
878 assert(nbd_reply_is_structured(&s
->reply
));
880 len
= s
->reply
.structured
.length
;
886 if (payload
== NULL
) {
887 error_setg(errp
, "Unexpected structured payload");
891 if (len
> NBD_MAX_MALLOC_PAYLOAD
) {
892 error_setg(errp
, "Payload too large");
896 *payload
= g_new(char, len
);
897 ret
= nbd_read(s
->ioc
, *payload
, len
, "structured payload", errp
);
908 * nbd_co_do_receive_one_chunk
910 * set request_ret to received reply error
911 * if qiov is not NULL: read payload to @qiov
912 * for structured reply chunk:
913 * if error chunk: read payload, set @request_ret, do not set @payload
914 * else if offset_data chunk: read payload data to @qiov, do not set @payload
915 * else: read payload to @payload
917 * If function fails, @errp contains corresponding error message, and the
918 * connection with the server is suspect. If it returns 0, then the
919 * transaction succeeded (although @request_ret may be a negative errno
920 * corresponding to the server's error reply), and errp is unchanged.
922 static coroutine_fn
int nbd_co_do_receive_one_chunk(
923 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
924 int *request_ret
, QEMUIOVector
*qiov
, void **payload
, Error
**errp
)
927 int i
= HANDLE_TO_INDEX(s
, handle
);
928 void *local_payload
= NULL
;
929 NBDStructuredReplyChunk
*chunk
;
936 /* Wait until we're woken up by nbd_connection_entry. */
937 s
->requests
[i
].receiving
= true;
938 qemu_coroutine_yield();
939 assert(!s
->requests
[i
].receiving
);
940 if (!nbd_client_connected(s
)) {
941 error_setg(errp
, "Connection closed");
946 assert(s
->reply
.handle
== handle
);
948 if (nbd_reply_is_simple(&s
->reply
)) {
949 if (only_structured
) {
950 error_setg(errp
, "Protocol error: simple reply when structured "
951 "reply chunk was expected");
955 *request_ret
= -nbd_errno_to_system_errno(s
->reply
.simple
.error
);
956 if (*request_ret
< 0 || !qiov
) {
960 return qio_channel_readv_all(s
->ioc
, qiov
->iov
, qiov
->niov
,
961 errp
) < 0 ? -EIO
: 0;
964 /* handle structured reply chunk */
965 assert(s
->info
.structured_reply
);
966 chunk
= &s
->reply
.structured
;
968 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
969 if (!(chunk
->flags
& NBD_REPLY_FLAG_DONE
)) {
970 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
971 " NBD_REPLY_FLAG_DONE flag set");
975 error_setg(errp
, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
982 if (chunk
->type
== NBD_REPLY_TYPE_OFFSET_DATA
) {
984 error_setg(errp
, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
988 return nbd_co_receive_offset_data_payload(s
, s
->requests
[i
].offset
,
992 if (nbd_reply_type_is_error(chunk
->type
)) {
993 payload
= &local_payload
;
996 ret
= nbd_co_receive_structured_payload(s
, payload
, errp
);
1001 if (nbd_reply_type_is_error(chunk
->type
)) {
1002 ret
= nbd_parse_error_payload(chunk
, local_payload
, request_ret
, errp
);
1003 g_free(local_payload
);
1011 * nbd_co_receive_one_chunk
1012 * Read reply, wake up connection_co and set s->quit if needed.
1013 * Return value is a fatal error code or normal nbd reply error code
1015 static coroutine_fn
int nbd_co_receive_one_chunk(
1016 BDRVNBDState
*s
, uint64_t handle
, bool only_structured
,
1017 int *request_ret
, QEMUIOVector
*qiov
, NBDReply
*reply
, void **payload
,
1020 int ret
= nbd_co_do_receive_one_chunk(s
, handle
, only_structured
,
1021 request_ret
, qiov
, payload
, errp
);
1024 memset(reply
, 0, sizeof(*reply
));
1025 nbd_channel_error(s
, ret
);
1027 /* For assert at loop start in nbd_connection_entry */
1030 s
->reply
.handle
= 0;
1032 if (s
->connection_co
&& !s
->wait_in_flight
) {
1034 * We must check s->wait_in_flight, because we may entered by
1035 * nbd_recv_coroutines_wake_all(), in this case we should not
1036 * wake connection_co here, it will woken by last request.
1038 aio_co_wake(s
->connection_co
);
1044 typedef struct NBDReplyChunkIter
{
1048 bool done
, only_structured
;
1049 } NBDReplyChunkIter
;
1051 static void nbd_iter_channel_error(NBDReplyChunkIter
*iter
,
1052 int ret
, Error
**local_err
)
1054 assert(local_err
&& *local_err
);
1059 error_propagate(&iter
->err
, *local_err
);
1061 error_free(*local_err
);
1067 static void nbd_iter_request_error(NBDReplyChunkIter
*iter
, int ret
)
1071 if (!iter
->request_ret
) {
1072 iter
->request_ret
= ret
;
1077 * NBD_FOREACH_REPLY_CHUNK
1078 * The pointer stored in @payload requires g_free() to free it.
1080 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
1081 qiov, reply, payload) \
1082 for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
1083 nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
1086 * nbd_reply_chunk_iter_receive
1087 * The pointer stored in @payload requires g_free() to free it.
1089 static bool nbd_reply_chunk_iter_receive(BDRVNBDState
*s
,
1090 NBDReplyChunkIter
*iter
,
1092 QEMUIOVector
*qiov
, NBDReply
*reply
,
1095 int ret
, request_ret
;
1096 NBDReply local_reply
;
1097 NBDStructuredReplyChunk
*chunk
;
1098 Error
*local_err
= NULL
;
1099 if (!nbd_client_connected(s
)) {
1100 error_setg(&local_err
, "Connection closed");
1101 nbd_iter_channel_error(iter
, -EIO
, &local_err
);
1106 /* Previous iteration was last. */
1110 if (reply
== NULL
) {
1111 reply
= &local_reply
;
1114 ret
= nbd_co_receive_one_chunk(s
, handle
, iter
->only_structured
,
1115 &request_ret
, qiov
, reply
, payload
,
1118 nbd_iter_channel_error(iter
, ret
, &local_err
);
1119 } else if (request_ret
< 0) {
1120 nbd_iter_request_error(iter
, request_ret
);
1123 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1124 if (nbd_reply_is_simple(reply
) || !nbd_client_connected(s
)) {
1128 chunk
= &reply
->structured
;
1129 iter
->only_structured
= true;
1131 if (chunk
->type
== NBD_REPLY_TYPE_NONE
) {
1132 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1133 assert(chunk
->flags
& NBD_REPLY_FLAG_DONE
);
1137 if (chunk
->flags
& NBD_REPLY_FLAG_DONE
) {
1138 /* This iteration is last. */
1142 /* Execute the loop body */
1146 s
->requests
[HANDLE_TO_INDEX(s
, handle
)].coroutine
= NULL
;
1148 qemu_co_mutex_lock(&s
->send_mutex
);
1150 if (s
->in_flight
== 0 && s
->wait_in_flight
) {
1151 aio_co_wake(s
->connection_co
);
1153 qemu_co_queue_next(&s
->free_sema
);
1155 qemu_co_mutex_unlock(&s
->send_mutex
);
1160 static int nbd_co_receive_return_code(BDRVNBDState
*s
, uint64_t handle
,
1161 int *request_ret
, Error
**errp
)
1163 NBDReplyChunkIter iter
;
1165 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, NULL
, NULL
) {
1166 /* nbd_reply_chunk_iter_receive does all the work */
1169 error_propagate(errp
, iter
.err
);
1170 *request_ret
= iter
.request_ret
;
1174 static int nbd_co_receive_cmdread_reply(BDRVNBDState
*s
, uint64_t handle
,
1175 uint64_t offset
, QEMUIOVector
*qiov
,
1176 int *request_ret
, Error
**errp
)
1178 NBDReplyChunkIter iter
;
1180 void *payload
= NULL
;
1181 Error
*local_err
= NULL
;
1183 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, s
->info
.structured_reply
,
1184 qiov
, &reply
, &payload
)
1187 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1189 assert(nbd_reply_is_structured(&reply
));
1191 switch (chunk
->type
) {
1192 case NBD_REPLY_TYPE_OFFSET_DATA
:
1194 * special cased in nbd_co_receive_one_chunk, data is already
1198 case NBD_REPLY_TYPE_OFFSET_HOLE
:
1199 ret
= nbd_parse_offset_hole_payload(s
, &reply
.structured
, payload
,
1200 offset
, qiov
, &local_err
);
1202 nbd_channel_error(s
, ret
);
1203 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1207 if (!nbd_reply_type_is_error(chunk
->type
)) {
1208 /* not allowed reply type */
1209 nbd_channel_error(s
, -EINVAL
);
1210 error_setg(&local_err
,
1211 "Unexpected reply type: %d (%s) for CMD_READ",
1212 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1213 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1221 error_propagate(errp
, iter
.err
);
1222 *request_ret
= iter
.request_ret
;
1226 static int nbd_co_receive_blockstatus_reply(BDRVNBDState
*s
,
1227 uint64_t handle
, uint64_t length
,
1229 int *request_ret
, Error
**errp
)
1231 NBDReplyChunkIter iter
;
1233 void *payload
= NULL
;
1234 Error
*local_err
= NULL
;
1235 bool received
= false;
1237 assert(!extent
->length
);
1238 NBD_FOREACH_REPLY_CHUNK(s
, iter
, handle
, false, NULL
, &reply
, &payload
) {
1240 NBDStructuredReplyChunk
*chunk
= &reply
.structured
;
1242 assert(nbd_reply_is_structured(&reply
));
1244 switch (chunk
->type
) {
1245 case NBD_REPLY_TYPE_BLOCK_STATUS
:
1247 nbd_channel_error(s
, -EINVAL
);
1248 error_setg(&local_err
, "Several BLOCK_STATUS chunks in reply");
1249 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1253 ret
= nbd_parse_blockstatus_payload(s
, &reply
.structured
,
1254 payload
, length
, extent
,
1257 nbd_channel_error(s
, ret
);
1258 nbd_iter_channel_error(&iter
, ret
, &local_err
);
1262 if (!nbd_reply_type_is_error(chunk
->type
)) {
1263 nbd_channel_error(s
, -EINVAL
);
1264 error_setg(&local_err
,
1265 "Unexpected reply type: %d (%s) "
1266 "for CMD_BLOCK_STATUS",
1267 chunk
->type
, nbd_reply_type_lookup(chunk
->type
));
1268 nbd_iter_channel_error(&iter
, -EINVAL
, &local_err
);
1276 if (!extent
->length
&& !iter
.request_ret
) {
1277 error_setg(&local_err
, "Server did not reply with any status extents");
1278 nbd_iter_channel_error(&iter
, -EIO
, &local_err
);
1281 error_propagate(errp
, iter
.err
);
1282 *request_ret
= iter
.request_ret
;
1286 static int nbd_co_request(BlockDriverState
*bs
, NBDRequest
*request
,
1287 QEMUIOVector
*write_qiov
)
1289 int ret
, request_ret
;
1290 Error
*local_err
= NULL
;
1291 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1293 assert(request
->type
!= NBD_CMD_READ
);
1295 assert(request
->type
== NBD_CMD_WRITE
);
1296 assert(request
->len
== iov_size(write_qiov
->iov
, write_qiov
->niov
));
1298 assert(request
->type
!= NBD_CMD_WRITE
);
1302 ret
= nbd_co_send_request(bs
, request
, write_qiov
);
1307 ret
= nbd_co_receive_return_code(s
, request
->handle
,
1308 &request_ret
, &local_err
);
1310 trace_nbd_co_request_fail(request
->from
, request
->len
,
1311 request
->handle
, request
->flags
,
1313 nbd_cmd_lookup(request
->type
),
1314 ret
, error_get_pretty(local_err
));
1315 error_free(local_err
);
1318 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1320 return ret
? ret
: request_ret
;
1323 static int nbd_client_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1324 uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1326 int ret
, request_ret
;
1327 Error
*local_err
= NULL
;
1328 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1329 NBDRequest request
= {
1330 .type
= NBD_CMD_READ
,
1335 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1342 * Work around the fact that the block layer doesn't do
1343 * byte-accurate sizing yet - if the read exceeds the server's
1344 * advertised size because the block layer rounded size up, then
1345 * truncate the request to the server and tail-pad with zero.
1347 if (offset
>= s
->info
.size
) {
1348 assert(bytes
< BDRV_SECTOR_SIZE
);
1349 qemu_iovec_memset(qiov
, 0, 0, bytes
);
1352 if (offset
+ bytes
> s
->info
.size
) {
1353 uint64_t slop
= offset
+ bytes
- s
->info
.size
;
1355 assert(slop
< BDRV_SECTOR_SIZE
);
1356 qemu_iovec_memset(qiov
, bytes
- slop
, 0, slop
);
1357 request
.len
-= slop
;
1361 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1366 ret
= nbd_co_receive_cmdread_reply(s
, request
.handle
, offset
, qiov
,
1367 &request_ret
, &local_err
);
1369 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1370 request
.flags
, request
.type
,
1371 nbd_cmd_lookup(request
.type
),
1372 ret
, error_get_pretty(local_err
));
1373 error_free(local_err
);
1376 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1378 return ret
? ret
: request_ret
;
1381 static int nbd_client_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1382 uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1384 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1385 NBDRequest request
= {
1386 .type
= NBD_CMD_WRITE
,
1391 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1392 if (flags
& BDRV_REQ_FUA
) {
1393 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1394 request
.flags
|= NBD_CMD_FLAG_FUA
;
1397 assert(bytes
<= NBD_MAX_BUFFER_SIZE
);
1402 return nbd_co_request(bs
, &request
, qiov
);
1405 static int nbd_client_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1406 int bytes
, BdrvRequestFlags flags
)
1408 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1409 NBDRequest request
= {
1410 .type
= NBD_CMD_WRITE_ZEROES
,
1415 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1416 if (!(s
->info
.flags
& NBD_FLAG_SEND_WRITE_ZEROES
)) {
1420 if (flags
& BDRV_REQ_FUA
) {
1421 assert(s
->info
.flags
& NBD_FLAG_SEND_FUA
);
1422 request
.flags
|= NBD_CMD_FLAG_FUA
;
1424 if (!(flags
& BDRV_REQ_MAY_UNMAP
)) {
1425 request
.flags
|= NBD_CMD_FLAG_NO_HOLE
;
1427 if (flags
& BDRV_REQ_NO_FALLBACK
) {
1428 assert(s
->info
.flags
& NBD_FLAG_SEND_FAST_ZERO
);
1429 request
.flags
|= NBD_CMD_FLAG_FAST_ZERO
;
1435 return nbd_co_request(bs
, &request
, NULL
);
1438 static int nbd_client_co_flush(BlockDriverState
*bs
)
1440 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1441 NBDRequest request
= { .type
= NBD_CMD_FLUSH
};
1443 if (!(s
->info
.flags
& NBD_FLAG_SEND_FLUSH
)) {
1450 return nbd_co_request(bs
, &request
, NULL
);
1453 static int nbd_client_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
1456 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1457 NBDRequest request
= {
1458 .type
= NBD_CMD_TRIM
,
1463 assert(!(s
->info
.flags
& NBD_FLAG_READ_ONLY
));
1464 if (!(s
->info
.flags
& NBD_FLAG_SEND_TRIM
) || !bytes
) {
1468 return nbd_co_request(bs
, &request
, NULL
);
1471 static int coroutine_fn
nbd_client_co_block_status(
1472 BlockDriverState
*bs
, bool want_zero
, int64_t offset
, int64_t bytes
,
1473 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
1475 int ret
, request_ret
;
1476 NBDExtent extent
= { 0 };
1477 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1478 Error
*local_err
= NULL
;
1480 NBDRequest request
= {
1481 .type
= NBD_CMD_BLOCK_STATUS
,
1483 .len
= MIN(QEMU_ALIGN_DOWN(INT_MAX
, bs
->bl
.request_alignment
),
1484 MIN(bytes
, s
->info
.size
- offset
)),
1485 .flags
= NBD_CMD_FLAG_REQ_ONE
,
1488 if (!s
->info
.base_allocation
) {
1492 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
1496 * Work around the fact that the block layer doesn't do
1497 * byte-accurate sizing yet - if the status request exceeds the
1498 * server's advertised size because the block layer rounded size
1499 * up, we truncated the request to the server (above), or are
1500 * called on just the hole.
1502 if (offset
>= s
->info
.size
) {
1504 assert(bytes
< BDRV_SECTOR_SIZE
);
1505 /* Intentionally don't report offset_valid for the hole */
1506 return BDRV_BLOCK_ZERO
;
1509 if (s
->info
.min_block
) {
1510 assert(QEMU_IS_ALIGNED(request
.len
, s
->info
.min_block
));
1513 ret
= nbd_co_send_request(bs
, &request
, NULL
);
1518 ret
= nbd_co_receive_blockstatus_reply(s
, request
.handle
, bytes
,
1519 &extent
, &request_ret
,
1522 trace_nbd_co_request_fail(request
.from
, request
.len
, request
.handle
,
1523 request
.flags
, request
.type
,
1524 nbd_cmd_lookup(request
.type
),
1525 ret
, error_get_pretty(local_err
));
1526 error_free(local_err
);
1529 } while (ret
< 0 && nbd_client_connecting_wait(s
));
1531 if (ret
< 0 || request_ret
< 0) {
1532 return ret
? ret
: request_ret
;
1535 assert(extent
.length
);
1536 *pnum
= extent
.length
;
1539 return (extent
.flags
& NBD_STATE_HOLE
? 0 : BDRV_BLOCK_DATA
) |
1540 (extent
.flags
& NBD_STATE_ZERO
? BDRV_BLOCK_ZERO
: 0) |
1541 BDRV_BLOCK_OFFSET_VALID
;
1544 static int nbd_client_reopen_prepare(BDRVReopenState
*state
,
1545 BlockReopenQueue
*queue
, Error
**errp
)
1547 BDRVNBDState
*s
= (BDRVNBDState
*)state
->bs
->opaque
;
1549 if ((state
->flags
& BDRV_O_RDWR
) && (s
->info
.flags
& NBD_FLAG_READ_ONLY
)) {
1550 error_setg(errp
, "Can't reopen read-only NBD mount as read/write");
1556 static void nbd_yank(void *opaque
)
1558 BlockDriverState
*bs
= opaque
;
1559 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1561 qatomic_store_release(&s
->state
, NBD_CLIENT_QUIT
);
1562 qio_channel_shutdown(QIO_CHANNEL(s
->ioc
), QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
);
1565 static void nbd_client_close(BlockDriverState
*bs
)
1567 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1568 NBDRequest request
= { .type
= NBD_CMD_DISC
};
1571 nbd_send_request(s
->ioc
, &request
);
1574 nbd_teardown_connection(bs
);
1579 * Parse nbd_open options
1582 static int nbd_parse_uri(const char *filename
, QDict
*options
)
1586 QueryParams
*qp
= NULL
;
1590 uri
= uri_parse(filename
);
1596 if (!g_strcmp0(uri
->scheme
, "nbd")) {
1598 } else if (!g_strcmp0(uri
->scheme
, "nbd+tcp")) {
1600 } else if (!g_strcmp0(uri
->scheme
, "nbd+unix")) {
1607 p
= uri
->path
? uri
->path
: "";
1612 qdict_put_str(options
, "export", p
);
1615 qp
= query_params_parse(uri
->query
);
1616 if (qp
->n
> 1 || (is_unix
&& !qp
->n
) || (!is_unix
&& qp
->n
)) {
1622 /* nbd+unix:///export?socket=path */
1623 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
1627 qdict_put_str(options
, "server.type", "unix");
1628 qdict_put_str(options
, "server.path", qp
->p
[0].value
);
1633 /* nbd[+tcp]://host[:port]/export */
1639 /* strip braces from literal IPv6 address */
1640 if (uri
->server
[0] == '[') {
1641 host
= qstring_from_substr(uri
->server
, 1,
1642 strlen(uri
->server
) - 1);
1644 host
= qstring_from_str(uri
->server
);
1647 qdict_put_str(options
, "server.type", "inet");
1648 qdict_put(options
, "server.host", host
);
1650 port_str
= g_strdup_printf("%d", uri
->port
?: NBD_DEFAULT_PORT
);
1651 qdict_put_str(options
, "server.port", port_str
);
1657 query_params_free(qp
);
1663 static bool nbd_has_filename_options_conflict(QDict
*options
, Error
**errp
)
1665 const QDictEntry
*e
;
1667 for (e
= qdict_first(options
); e
; e
= qdict_next(options
, e
)) {
1668 if (!strcmp(e
->key
, "host") ||
1669 !strcmp(e
->key
, "port") ||
1670 !strcmp(e
->key
, "path") ||
1671 !strcmp(e
->key
, "export") ||
1672 strstart(e
->key
, "server.", NULL
))
1674 error_setg(errp
, "Option '%s' cannot be used with a file name",
1683 static void nbd_parse_filename(const char *filename
, QDict
*options
,
1686 g_autofree
char *file
= NULL
;
1688 const char *host_spec
;
1689 const char *unixpath
;
1691 if (nbd_has_filename_options_conflict(options
, errp
)) {
1695 if (strstr(filename
, "://")) {
1696 int ret
= nbd_parse_uri(filename
, options
);
1698 error_setg(errp
, "No valid URL specified");
1703 file
= g_strdup(filename
);
1705 export_name
= strstr(file
, EN_OPTSTR
);
1707 if (export_name
[strlen(EN_OPTSTR
)] == 0) {
1710 export_name
[0] = 0; /* truncate 'file' */
1711 export_name
+= strlen(EN_OPTSTR
);
1713 qdict_put_str(options
, "export", export_name
);
1716 /* extract the host_spec - fail if it's not nbd:... */
1717 if (!strstart(file
, "nbd:", &host_spec
)) {
1718 error_setg(errp
, "File name string for NBD must start with 'nbd:'");
1726 /* are we a UNIX or TCP socket? */
1727 if (strstart(host_spec
, "unix:", &unixpath
)) {
1728 qdict_put_str(options
, "server.type", "unix");
1729 qdict_put_str(options
, "server.path", unixpath
);
1731 InetSocketAddress
*addr
= g_new(InetSocketAddress
, 1);
1733 if (inet_parse(addr
, host_spec
, errp
)) {
1737 qdict_put_str(options
, "server.type", "inet");
1738 qdict_put_str(options
, "server.host", addr
->host
);
1739 qdict_put_str(options
, "server.port", addr
->port
);
1741 qapi_free_InetSocketAddress(addr
);
1745 static bool nbd_process_legacy_socket_options(QDict
*output_options
,
1746 QemuOpts
*legacy_opts
,
1749 const char *path
= qemu_opt_get(legacy_opts
, "path");
1750 const char *host
= qemu_opt_get(legacy_opts
, "host");
1751 const char *port
= qemu_opt_get(legacy_opts
, "port");
1752 const QDictEntry
*e
;
1754 if (!path
&& !host
&& !port
) {
1758 for (e
= qdict_first(output_options
); e
; e
= qdict_next(output_options
, e
))
1760 if (strstart(e
->key
, "server.", NULL
)) {
1761 error_setg(errp
, "Cannot use 'server' and path/host/port at the "
1768 error_setg(errp
, "path and host may not be used at the same time");
1772 error_setg(errp
, "port may not be used without host");
1776 qdict_put_str(output_options
, "server.type", "unix");
1777 qdict_put_str(output_options
, "server.path", path
);
1779 qdict_put_str(output_options
, "server.type", "inet");
1780 qdict_put_str(output_options
, "server.host", host
);
1781 qdict_put_str(output_options
, "server.port",
1782 port
?: stringify(NBD_DEFAULT_PORT
));
1788 static SocketAddress
*nbd_config(BDRVNBDState
*s
, QDict
*options
,
1791 SocketAddress
*saddr
= NULL
;
1795 qdict_extract_subqdict(options
, &addr
, "server.");
1796 if (!qdict_size(addr
)) {
1797 error_setg(errp
, "NBD server address missing");
1801 iv
= qobject_input_visitor_new_flat_confused(addr
, errp
);
1806 if (!visit_type_SocketAddress(iv
, NULL
, &saddr
, errp
)) {
1810 if (socket_address_parse_named_fd(saddr
, errp
) < 0) {
1811 qapi_free_SocketAddress(saddr
);
1817 qobject_unref(addr
);
1822 static QCryptoTLSCreds
*nbd_get_tls_creds(const char *id
, Error
**errp
)
1825 QCryptoTLSCreds
*creds
;
1827 obj
= object_resolve_path_component(
1828 object_get_objects_root(), id
);
1830 error_setg(errp
, "No TLS credentials with id '%s'",
1834 creds
= (QCryptoTLSCreds
*)
1835 object_dynamic_cast(obj
, TYPE_QCRYPTO_TLS_CREDS
);
1837 error_setg(errp
, "Object with id '%s' is not TLS credentials",
1842 if (creds
->endpoint
!= QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT
) {
1844 "Expecting TLS credentials with a client endpoint");
1852 static QemuOptsList nbd_runtime_opts
= {
1854 .head
= QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts
.head
),
1858 .type
= QEMU_OPT_STRING
,
1859 .help
= "TCP host to connect to",
1863 .type
= QEMU_OPT_STRING
,
1864 .help
= "TCP port to connect to",
1868 .type
= QEMU_OPT_STRING
,
1869 .help
= "Unix socket path to connect to",
1873 .type
= QEMU_OPT_STRING
,
1874 .help
= "Name of the NBD export to open",
1877 .name
= "tls-creds",
1878 .type
= QEMU_OPT_STRING
,
1879 .help
= "ID of the TLS credentials to use",
1882 .name
= "x-dirty-bitmap",
1883 .type
= QEMU_OPT_STRING
,
1884 .help
= "experimental: expose named dirty bitmap in place of "
1888 .name
= "reconnect-delay",
1889 .type
= QEMU_OPT_NUMBER
,
1890 .help
= "On an unexpected disconnect, the nbd client tries to "
1891 "connect again until succeeding or encountering a serious "
1892 "error. During the first @reconnect-delay seconds, all "
1893 "requests are paused and will be rerun on a successful "
1894 "reconnect. After that time, any delayed requests and all "
1895 "future requests before a successful reconnect will "
1896 "immediately fail. Default 0",
1898 { /* end of list */ }
1902 static int nbd_process_options(BlockDriverState
*bs
, QDict
*options
,
1905 BDRVNBDState
*s
= bs
->opaque
;
1909 opts
= qemu_opts_create(&nbd_runtime_opts
, NULL
, 0, &error_abort
);
1910 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1914 /* Translate @host, @port, and @path to a SocketAddress */
1915 if (!nbd_process_legacy_socket_options(options
, opts
, errp
)) {
1919 /* Pop the config into our state object. Exit if invalid. */
1920 s
->saddr
= nbd_config(s
, options
, errp
);
1925 s
->export
= g_strdup(qemu_opt_get(opts
, "export"));
1926 if (s
->export
&& strlen(s
->export
) > NBD_MAX_STRING_SIZE
) {
1927 error_setg(errp
, "export name too long to send to server");
1931 s
->tlscredsid
= g_strdup(qemu_opt_get(opts
, "tls-creds"));
1932 if (s
->tlscredsid
) {
1933 s
->tlscreds
= nbd_get_tls_creds(s
->tlscredsid
, errp
);
1938 /* TODO SOCKET_ADDRESS_KIND_FD where fd has AF_INET or AF_INET6 */
1939 if (s
->saddr
->type
!= SOCKET_ADDRESS_TYPE_INET
) {
1940 error_setg(errp
, "TLS only supported over IP sockets");
1943 s
->hostname
= s
->saddr
->u
.inet
.host
;
1946 s
->x_dirty_bitmap
= g_strdup(qemu_opt_get(opts
, "x-dirty-bitmap"));
1947 if (s
->x_dirty_bitmap
&& strlen(s
->x_dirty_bitmap
) > NBD_MAX_STRING_SIZE
) {
1948 error_setg(errp
, "x-dirty-bitmap query too long to send to server");
1952 s
->reconnect_delay
= qemu_opt_get_number(opts
, "reconnect-delay", 0);
1957 qemu_opts_del(opts
);
1961 static int nbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1965 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
1968 qemu_co_mutex_init(&s
->send_mutex
);
1969 qemu_co_queue_init(&s
->free_sema
);
1971 if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs
->node_name
), errp
)) {
1975 ret
= nbd_process_options(bs
, options
, errp
);
1980 s
->conn
= nbd_client_connection_new(s
->saddr
, true, s
->export
,
1981 s
->x_dirty_bitmap
, s
->tlscreds
);
1983 /* TODO: Configurable retry-until-timeout behaviour. */
1984 ret
= nbd_do_establish_connection(bs
, errp
);
1989 s
->connection_co
= qemu_coroutine_create(nbd_connection_entry
, s
);
1990 bdrv_inc_in_flight(bs
);
1991 aio_co_schedule(bdrv_get_aio_context(bs
), s
->connection_co
);
1996 nbd_clear_bdrvstate(bs
);
2000 static int nbd_co_flush(BlockDriverState
*bs
)
2002 return nbd_client_co_flush(bs
);
2005 static void nbd_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
2007 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
2008 uint32_t min
= s
->info
.min_block
;
2009 uint32_t max
= MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE
, s
->info
.max_block
);
2012 * If the server did not advertise an alignment:
2013 * - a size that is not sector-aligned implies that an alignment
2014 * of 1 can be used to access those tail bytes
2015 * - advertisement of block status requires an alignment of 1, so
2016 * that we don't violate block layer constraints that block
2017 * status is always aligned (as we can't control whether the
2018 * server will report sub-sector extents, such as a hole at EOF
2019 * on an unaligned POSIX file)
2020 * - otherwise, assume the server is so old that we are safer avoiding
2021 * sub-sector requests
2024 min
= (!QEMU_IS_ALIGNED(s
->info
.size
, BDRV_SECTOR_SIZE
) ||
2025 s
->info
.base_allocation
) ? 1 : BDRV_SECTOR_SIZE
;
2028 bs
->bl
.request_alignment
= min
;
2029 bs
->bl
.max_pdiscard
= QEMU_ALIGN_DOWN(INT_MAX
, min
);
2030 bs
->bl
.max_pwrite_zeroes
= max
;
2031 bs
->bl
.max_transfer
= max
;
2033 if (s
->info
.opt_block
&&
2034 s
->info
.opt_block
> bs
->bl
.opt_transfer
) {
2035 bs
->bl
.opt_transfer
= s
->info
.opt_block
;
2039 static void nbd_close(BlockDriverState
*bs
)
2041 nbd_client_close(bs
);
2042 nbd_clear_bdrvstate(bs
);
2046 * NBD cannot truncate, but if the caller asks to truncate to the same size, or
2047 * to a smaller size with exact=false, there is no reason to fail the
2050 * Preallocation mode is ignored since it does not seems useful to fail when
2051 * we never change anything.
2053 static int coroutine_fn
nbd_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2054 bool exact
, PreallocMode prealloc
,
2055 BdrvRequestFlags flags
, Error
**errp
)
2057 BDRVNBDState
*s
= bs
->opaque
;
2059 if (offset
!= s
->info
.size
&& exact
) {
2060 error_setg(errp
, "Cannot resize NBD nodes");
2064 if (offset
> s
->info
.size
) {
2065 error_setg(errp
, "Cannot grow NBD nodes");
2072 static int64_t nbd_getlength(BlockDriverState
*bs
)
2074 BDRVNBDState
*s
= bs
->opaque
;
2076 return s
->info
.size
;
2079 static void nbd_refresh_filename(BlockDriverState
*bs
)
2081 BDRVNBDState
*s
= bs
->opaque
;
2082 const char *host
= NULL
, *port
= NULL
, *path
= NULL
;
2085 if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_INET
) {
2086 const InetSocketAddress
*inet
= &s
->saddr
->u
.inet
;
2087 if (!inet
->has_ipv4
&& !inet
->has_ipv6
&& !inet
->has_to
) {
2091 } else if (s
->saddr
->type
== SOCKET_ADDRESS_TYPE_UNIX
) {
2092 path
= s
->saddr
->u
.q_unix
.path
;
2093 } /* else can't represent as pseudo-filename */
2095 if (path
&& s
->export
) {
2096 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2097 "nbd+unix:///%s?socket=%s", s
->export
, path
);
2098 } else if (path
&& !s
->export
) {
2099 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2100 "nbd+unix://?socket=%s", path
);
2101 } else if (host
&& s
->export
) {
2102 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2103 "nbd://%s:%s/%s", host
, port
, s
->export
);
2104 } else if (host
&& !s
->export
) {
2105 len
= snprintf(bs
->exact_filename
, sizeof(bs
->exact_filename
),
2106 "nbd://%s:%s", host
, port
);
2108 if (len
>= sizeof(bs
->exact_filename
)) {
2109 /* Name is too long to represent exactly, so leave it empty. */
2110 bs
->exact_filename
[0] = '\0';
2114 static char *nbd_dirname(BlockDriverState
*bs
, Error
**errp
)
2116 /* The generic bdrv_dirname() implementation is able to work out some
2117 * directory name for NBD nodes, but that would be wrong. So far there is no
2118 * specification for how "export paths" would work, so NBD does not have
2119 * directory names. */
2120 error_setg(errp
, "Cannot generate a base directory for NBD nodes");
2124 static const char *const nbd_strong_runtime_opts
[] = {
2135 static void nbd_cancel_in_flight(BlockDriverState
*bs
)
2137 BDRVNBDState
*s
= (BDRVNBDState
*)bs
->opaque
;
2139 reconnect_delay_timer_del(s
);
2141 if (s
->state
== NBD_CLIENT_CONNECTING_WAIT
) {
2142 s
->state
= NBD_CLIENT_CONNECTING_NOWAIT
;
2143 qemu_co_queue_restart_all(&s
->free_sema
);
2147 static BlockDriver bdrv_nbd
= {
2148 .format_name
= "nbd",
2149 .protocol_name
= "nbd",
2150 .instance_size
= sizeof(BDRVNBDState
),
2151 .bdrv_parse_filename
= nbd_parse_filename
,
2152 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2153 .create_opts
= &bdrv_create_opts_simple
,
2154 .bdrv_file_open
= nbd_open
,
2155 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2156 .bdrv_co_preadv
= nbd_client_co_preadv
,
2157 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2158 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2159 .bdrv_close
= nbd_close
,
2160 .bdrv_co_flush_to_os
= nbd_co_flush
,
2161 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2162 .bdrv_refresh_limits
= nbd_refresh_limits
,
2163 .bdrv_co_truncate
= nbd_co_truncate
,
2164 .bdrv_getlength
= nbd_getlength
,
2165 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2166 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2167 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2168 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2169 .bdrv_refresh_filename
= nbd_refresh_filename
,
2170 .bdrv_co_block_status
= nbd_client_co_block_status
,
2171 .bdrv_dirname
= nbd_dirname
,
2172 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2173 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2176 static BlockDriver bdrv_nbd_tcp
= {
2177 .format_name
= "nbd",
2178 .protocol_name
= "nbd+tcp",
2179 .instance_size
= sizeof(BDRVNBDState
),
2180 .bdrv_parse_filename
= nbd_parse_filename
,
2181 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2182 .create_opts
= &bdrv_create_opts_simple
,
2183 .bdrv_file_open
= nbd_open
,
2184 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2185 .bdrv_co_preadv
= nbd_client_co_preadv
,
2186 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2187 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2188 .bdrv_close
= nbd_close
,
2189 .bdrv_co_flush_to_os
= nbd_co_flush
,
2190 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2191 .bdrv_refresh_limits
= nbd_refresh_limits
,
2192 .bdrv_co_truncate
= nbd_co_truncate
,
2193 .bdrv_getlength
= nbd_getlength
,
2194 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2195 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2196 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2197 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2198 .bdrv_refresh_filename
= nbd_refresh_filename
,
2199 .bdrv_co_block_status
= nbd_client_co_block_status
,
2200 .bdrv_dirname
= nbd_dirname
,
2201 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2202 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2205 static BlockDriver bdrv_nbd_unix
= {
2206 .format_name
= "nbd",
2207 .protocol_name
= "nbd+unix",
2208 .instance_size
= sizeof(BDRVNBDState
),
2209 .bdrv_parse_filename
= nbd_parse_filename
,
2210 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2211 .create_opts
= &bdrv_create_opts_simple
,
2212 .bdrv_file_open
= nbd_open
,
2213 .bdrv_reopen_prepare
= nbd_client_reopen_prepare
,
2214 .bdrv_co_preadv
= nbd_client_co_preadv
,
2215 .bdrv_co_pwritev
= nbd_client_co_pwritev
,
2216 .bdrv_co_pwrite_zeroes
= nbd_client_co_pwrite_zeroes
,
2217 .bdrv_close
= nbd_close
,
2218 .bdrv_co_flush_to_os
= nbd_co_flush
,
2219 .bdrv_co_pdiscard
= nbd_client_co_pdiscard
,
2220 .bdrv_refresh_limits
= nbd_refresh_limits
,
2221 .bdrv_co_truncate
= nbd_co_truncate
,
2222 .bdrv_getlength
= nbd_getlength
,
2223 .bdrv_detach_aio_context
= nbd_client_detach_aio_context
,
2224 .bdrv_attach_aio_context
= nbd_client_attach_aio_context
,
2225 .bdrv_co_drain_begin
= nbd_client_co_drain_begin
,
2226 .bdrv_co_drain_end
= nbd_client_co_drain_end
,
2227 .bdrv_refresh_filename
= nbd_refresh_filename
,
2228 .bdrv_co_block_status
= nbd_client_co_block_status
,
2229 .bdrv_dirname
= nbd_dirname
,
2230 .strong_runtime_opts
= nbd_strong_runtime_opts
,
2231 .bdrv_cancel_in_flight
= nbd_cancel_in_flight
,
2234 static void bdrv_nbd_init(void)
2236 bdrv_register(&bdrv_nbd
);
2237 bdrv_register(&bdrv_nbd_tcp
);
2238 bdrv_register(&bdrv_nbd_unix
);
2241 block_init(bdrv_nbd_init
);