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 nbd_reply_chunk_iter_receive(BDRVNBDState *s,
987 NBDReplyChunkIter *iter,
989 QEMUIOVector *qiov, NBDReply *reply,
992 int ret, request_ret;
993 NBDReply local_reply;
994 NBDStructuredReplyChunk *chunk;
995 Error *local_err = NULL;
998 /* Previous iteration was last. */
1002 if (reply == NULL) {
1003 reply = &local_reply;
1006 ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
1007 &request_ret, qiov, reply, payload,
1010 nbd_iter_channel_error(iter, ret, &local_err);
1011 } else if (request_ret < 0) {
1012 nbd_iter_request_error(iter, request_ret);
1015 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1016 if (nbd_reply_is_simple(reply) || iter->ret < 0) {
1020 chunk = &reply->structured;
1021 iter->only_structured = true;
1023 if (chunk->type == NBD_REPLY_TYPE_NONE) {
1024 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1025 assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1029 if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1030 /* This iteration is last. */
1034 /* Execute the loop body */
1038 qemu_mutex_lock(&s->requests_lock);
1039 s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
1041 qemu_co_queue_next(&s->free_sema);
1042 qemu_mutex_unlock(&s->requests_lock);
1047 static int coroutine_fn nbd_co_receive_return_code(BDRVNBDState *s, uint64_t handle,
1048 int *request_ret, Error **errp)
1050 NBDReplyChunkIter iter;
1052 NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
1053 /* nbd_reply_chunk_iter_receive does all the work */
1056 error_propagate(errp, iter.err);
1057 *request_ret = iter.request_ret;
1061 static int coroutine_fn nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t handle,
1062 uint64_t offset, QEMUIOVector *qiov,
1063 int *request_ret, Error **errp)
1065 NBDReplyChunkIter iter;
1067 void *payload = NULL;
1068 Error *local_err = NULL;
1070 NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
1071 qiov, &reply, &payload)
1074 NBDStructuredReplyChunk *chunk = &reply.structured;
1076 assert(nbd_reply_is_structured(&reply));
1078 switch (chunk->type) {
1079 case NBD_REPLY_TYPE_OFFSET_DATA:
1081 * special cased in nbd_co_receive_one_chunk, data is already
1085 case NBD_REPLY_TYPE_OFFSET_HOLE:
1086 ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1087 offset, qiov, &local_err);
1089 nbd_channel_error(s, ret);
1090 nbd_iter_channel_error(&iter, ret, &local_err);
1094 if (!nbd_reply_type_is_error(chunk->type)) {
1095 /* not allowed reply type */
1096 nbd_channel_error(s, -EINVAL);
1097 error_setg(&local_err,
1098 "Unexpected reply type: %d (%s) for CMD_READ",
1099 chunk->type, nbd_reply_type_lookup(chunk->type));
1100 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1108 error_propagate(errp, iter.err);
1109 *request_ret = iter.request_ret;
1113 static int coroutine_fn nbd_co_receive_blockstatus_reply(BDRVNBDState *s,
1114 uint64_t handle, uint64_t length,
1116 int *request_ret, Error **errp)
1118 NBDReplyChunkIter iter;
1120 void *payload = NULL;
1121 Error *local_err = NULL;
1122 bool received = false;
1124 assert(!extent->length);
1125 NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) {
1127 NBDStructuredReplyChunk *chunk = &reply.structured;
1129 assert(nbd_reply_is_structured(&reply));
1131 switch (chunk->type) {
1132 case NBD_REPLY_TYPE_BLOCK_STATUS:
1134 nbd_channel_error(s, -EINVAL);
1135 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1136 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1140 ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1141 payload, length, extent,
1144 nbd_channel_error(s, ret);
1145 nbd_iter_channel_error(&iter, ret, &local_err);
1149 if (!nbd_reply_type_is_error(chunk->type)) {
1150 nbd_channel_error(s, -EINVAL);
1151 error_setg(&local_err,
1152 "Unexpected reply type: %d (%s) "
1153 "for CMD_BLOCK_STATUS",
1154 chunk->type, nbd_reply_type_lookup(chunk->type));
1155 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1163 if (!extent->length && !iter.request_ret) {
1164 error_setg(&local_err, "Server did not reply with any status extents");
1165 nbd_iter_channel_error(&iter, -EIO, &local_err);
1168 error_propagate(errp, iter.err);
1169 *request_ret = iter.request_ret;
1173 static int coroutine_fn nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1174 QEMUIOVector *write_qiov)
1176 int ret, request_ret;
1177 Error *local_err = NULL;
1178 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1180 assert(request->type != NBD_CMD_READ);
1182 assert(request->type == NBD_CMD_WRITE);
1183 assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1185 assert(request->type != NBD_CMD_WRITE);
1189 ret = nbd_co_send_request(bs, request, write_qiov);
1194 ret = nbd_co_receive_return_code(s, request->handle,
1195 &request_ret, &local_err);
1197 trace_nbd_co_request_fail(request->from, request->len,
1198 request->handle, request->flags,
1200 nbd_cmd_lookup(request->type),
1201 ret, error_get_pretty(local_err));
1202 error_free(local_err);
1205 } while (ret < 0 && nbd_client_will_reconnect(s));
1207 return ret ? ret : request_ret;
1210 static int coroutine_fn nbd_client_co_preadv(BlockDriverState *bs, int64_t offset,
1211 int64_t bytes, QEMUIOVector *qiov,
1212 BdrvRequestFlags flags)
1214 int ret, request_ret;
1215 Error *local_err = NULL;
1216 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1217 NBDRequest request = {
1218 .type = NBD_CMD_READ,
1223 assert(bytes <= NBD_MAX_BUFFER_SIZE);
1230 * Work around the fact that the block layer doesn't do
1231 * byte-accurate sizing yet - if the read exceeds the server's
1232 * advertised size because the block layer rounded size up, then
1233 * truncate the request to the server and tail-pad with zero.
1235 if (offset >= s->info.size) {
1236 assert(bytes < BDRV_SECTOR_SIZE);
1237 qemu_iovec_memset(qiov, 0, 0, bytes);
1240 if (offset + bytes > s->info.size) {
1241 uint64_t slop = offset + bytes - s->info.size;
1243 assert(slop < BDRV_SECTOR_SIZE);
1244 qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1245 request.len -= slop;
1249 ret = nbd_co_send_request(bs, &request, NULL);
1254 ret = nbd_co_receive_cmdread_reply(s, request.handle, offset, qiov,
1255 &request_ret, &local_err);
1257 trace_nbd_co_request_fail(request.from, request.len, request.handle,
1258 request.flags, request.type,
1259 nbd_cmd_lookup(request.type),
1260 ret, error_get_pretty(local_err));
1261 error_free(local_err);
1264 } while (ret < 0 && nbd_client_will_reconnect(s));
1266 return ret ? ret : request_ret;
1269 static int coroutine_fn nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset,
1270 int64_t bytes, QEMUIOVector *qiov,
1271 BdrvRequestFlags flags)
1273 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1274 NBDRequest request = {
1275 .type = NBD_CMD_WRITE,
1280 assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1281 if (flags & BDRV_REQ_FUA) {
1282 assert(s->info.flags & NBD_FLAG_SEND_FUA);
1283 request.flags |= NBD_CMD_FLAG_FUA;
1286 assert(bytes <= NBD_MAX_BUFFER_SIZE);
1291 return nbd_co_request(bs, &request, qiov);
1294 static int coroutine_fn nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1295 int64_t bytes, BdrvRequestFlags flags)
1297 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1298 NBDRequest request = {
1299 .type = NBD_CMD_WRITE_ZEROES,
1301 .len = bytes, /* .len is uint32_t actually */
1304 assert(bytes <= UINT32_MAX); /* rely on max_pwrite_zeroes */
1306 assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1307 if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1311 if (flags & BDRV_REQ_FUA) {
1312 assert(s->info.flags & NBD_FLAG_SEND_FUA);
1313 request.flags |= NBD_CMD_FLAG_FUA;
1315 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1316 request.flags |= NBD_CMD_FLAG_NO_HOLE;
1318 if (flags & BDRV_REQ_NO_FALLBACK) {
1319 assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1320 request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1326 return nbd_co_request(bs, &request, NULL);
1329 static int coroutine_fn nbd_client_co_flush(BlockDriverState *bs)
1331 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1332 NBDRequest request = { .type = NBD_CMD_FLUSH };
1334 if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1341 return nbd_co_request(bs, &request, NULL);
1344 static int coroutine_fn nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset,
1347 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1348 NBDRequest request = {
1349 .type = NBD_CMD_TRIM,
1351 .len = bytes, /* len is uint32_t */
1354 assert(bytes <= UINT32_MAX); /* rely on max_pdiscard */
1356 assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1357 if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1361 return nbd_co_request(bs, &request, NULL);
1364 static int coroutine_fn nbd_client_co_block_status(
1365 BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1366 int64_t *pnum, int64_t *map, BlockDriverState **file)
1368 int ret, request_ret;
1369 NBDExtent extent = { 0 };
1370 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1371 Error *local_err = NULL;
1373 NBDRequest request = {
1374 .type = NBD_CMD_BLOCK_STATUS,
1376 .len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1377 MIN(bytes, s->info.size - offset)),
1378 .flags = NBD_CMD_FLAG_REQ_ONE,
1381 if (!s->info.base_allocation) {
1385 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1389 * Work around the fact that the block layer doesn't do
1390 * byte-accurate sizing yet - if the status request exceeds the
1391 * server's advertised size because the block layer rounded size
1392 * up, we truncated the request to the server (above), or are
1393 * called on just the hole.
1395 if (offset >= s->info.size) {
1397 assert(bytes < BDRV_SECTOR_SIZE);
1398 /* Intentionally don't report offset_valid for the hole */
1399 return BDRV_BLOCK_ZERO;
1402 if (s->info.min_block) {
1403 assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1406 ret = nbd_co_send_request(bs, &request, NULL);
1411 ret = nbd_co_receive_blockstatus_reply(s, request.handle, bytes,
1412 &extent, &request_ret,
1415 trace_nbd_co_request_fail(request.from, request.len, request.handle,
1416 request.flags, request.type,
1417 nbd_cmd_lookup(request.type),
1418 ret, error_get_pretty(local_err));
1419 error_free(local_err);
1422 } while (ret < 0 && nbd_client_will_reconnect(s));
1424 if (ret < 0 || request_ret < 0) {
1425 return ret ? ret : request_ret;
1428 assert(extent.length);
1429 *pnum = extent.length;
1432 return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1433 (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1434 BDRV_BLOCK_OFFSET_VALID;
1437 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1438 BlockReopenQueue *queue, Error **errp)
1440 BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1442 if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1443 error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1449 static void nbd_yank(void *opaque)
1451 BlockDriverState *bs = opaque;
1452 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1454 QEMU_LOCK_GUARD(&s->requests_lock);
1455 qio_channel_shutdown(QIO_CHANNEL(s->ioc), QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1456 s->state = NBD_CLIENT_QUIT;
1459 static void nbd_client_close(BlockDriverState *bs)
1461 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1462 NBDRequest request = { .type = NBD_CMD_DISC };
1465 nbd_send_request(s->ioc, &request);
1468 nbd_teardown_connection(bs);
1473 * Parse nbd_open options
1476 static int nbd_parse_uri(const char *filename, QDict *options)
1480 QueryParams *qp = NULL;
1484 uri = uri_parse(filename);
1490 if (!g_strcmp0(uri->scheme, "nbd")) {
1492 } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1494 } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1501 p = uri->path ? uri->path : "";
1506 qdict_put_str(options, "export", p);
1509 qp = query_params_parse(uri->query);
1510 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1516 /* nbd+unix:///export?socket=path */
1517 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1521 qdict_put_str(options, "server.type", "unix");
1522 qdict_put_str(options, "server.path", qp->p[0].value);
1527 /* nbd[+tcp]://host[:port]/export */
1533 /* strip braces from literal IPv6 address */
1534 if (uri->server[0] == '[') {
1535 host = qstring_from_substr(uri->server, 1,
1536 strlen(uri->server) - 1);
1538 host = qstring_from_str(uri->server);
1541 qdict_put_str(options, "server.type", "inet");
1542 qdict_put(options, "server.host", host);
1544 port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1545 qdict_put_str(options, "server.port", port_str);
1551 query_params_free(qp);
1557 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1559 const QDictEntry *e;
1561 for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1562 if (!strcmp(e->key, "host") ||
1563 !strcmp(e->key, "port") ||
1564 !strcmp(e->key, "path") ||
1565 !strcmp(e->key, "export") ||
1566 strstart(e->key, "server.", NULL))
1568 error_setg(errp, "Option '%s' cannot be used with a file name",
1577 static void nbd_parse_filename(const char *filename, QDict *options,
1580 g_autofree char *file = NULL;
1582 const char *host_spec;
1583 const char *unixpath;
1585 if (nbd_has_filename_options_conflict(options, errp)) {
1589 if (strstr(filename, "://")) {
1590 int ret = nbd_parse_uri(filename, options);
1592 error_setg(errp, "No valid URL specified");
1597 file = g_strdup(filename);
1599 export_name = strstr(file, EN_OPTSTR);
1601 if (export_name[strlen(EN_OPTSTR)] == 0) {
1604 export_name[0] = 0; /* truncate 'file' */
1605 export_name += strlen(EN_OPTSTR);
1607 qdict_put_str(options, "export", export_name);
1610 /* extract the host_spec - fail if it's not nbd:... */
1611 if (!strstart(file, "nbd:", &host_spec)) {
1612 error_setg(errp, "File name string for NBD must start with 'nbd:'");
1620 /* are we a UNIX or TCP socket? */
1621 if (strstart(host_spec, "unix:", &unixpath)) {
1622 qdict_put_str(options, "server.type", "unix");
1623 qdict_put_str(options, "server.path", unixpath);
1625 InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1627 if (inet_parse(addr, host_spec, errp)) {
1631 qdict_put_str(options, "server.type", "inet");
1632 qdict_put_str(options, "server.host", addr->host);
1633 qdict_put_str(options, "server.port", addr->port);
1635 qapi_free_InetSocketAddress(addr);
1639 static bool nbd_process_legacy_socket_options(QDict *output_options,
1640 QemuOpts *legacy_opts,
1643 const char *path = qemu_opt_get(legacy_opts, "path");
1644 const char *host = qemu_opt_get(legacy_opts, "host");
1645 const char *port = qemu_opt_get(legacy_opts, "port");
1646 const QDictEntry *e;
1648 if (!path && !host && !port) {
1652 for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1654 if (strstart(e->key, "server.", NULL)) {
1655 error_setg(errp, "Cannot use 'server' and path/host/port at the "
1662 error_setg(errp, "path and host may not be used at the same time");
1666 error_setg(errp, "port may not be used without host");
1670 qdict_put_str(output_options, "server.type", "unix");
1671 qdict_put_str(output_options, "server.path", path);
1673 qdict_put_str(output_options, "server.type", "inet");
1674 qdict_put_str(output_options, "server.host", host);
1675 qdict_put_str(output_options, "server.port",
1676 port ?: stringify(NBD_DEFAULT_PORT));
1682 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1685 SocketAddress *saddr = NULL;
1689 qdict_extract_subqdict(options, &addr, "server.");
1690 if (!qdict_size(addr)) {
1691 error_setg(errp, "NBD server address missing");
1695 iv = qobject_input_visitor_new_flat_confused(addr, errp);
1700 if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1704 if (socket_address_parse_named_fd(saddr, errp) < 0) {
1705 qapi_free_SocketAddress(saddr);
1711 qobject_unref(addr);
1716 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1719 QCryptoTLSCreds *creds;
1721 obj = object_resolve_path_component(
1722 object_get_objects_root(), id);
1724 error_setg(errp, "No TLS credentials with id '%s'",
1728 creds = (QCryptoTLSCreds *)
1729 object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1731 error_setg(errp, "Object with id '%s' is not TLS credentials",
1736 if (!qcrypto_tls_creds_check_endpoint(creds,
1737 QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1746 static QemuOptsList nbd_runtime_opts = {
1748 .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1752 .type = QEMU_OPT_STRING,
1753 .help = "TCP host to connect to",
1757 .type = QEMU_OPT_STRING,
1758 .help = "TCP port to connect to",
1762 .type = QEMU_OPT_STRING,
1763 .help = "Unix socket path to connect to",
1767 .type = QEMU_OPT_STRING,
1768 .help = "Name of the NBD export to open",
1771 .name = "tls-creds",
1772 .type = QEMU_OPT_STRING,
1773 .help = "ID of the TLS credentials to use",
1776 .name = "tls-hostname",
1777 .type = QEMU_OPT_STRING,
1778 .help = "Override hostname for validating TLS x509 certificate",
1781 .name = "x-dirty-bitmap",
1782 .type = QEMU_OPT_STRING,
1783 .help = "experimental: expose named dirty bitmap in place of "
1787 .name = "reconnect-delay",
1788 .type = QEMU_OPT_NUMBER,
1789 .help = "On an unexpected disconnect, the nbd client tries to "
1790 "connect again until succeeding or encountering a serious "
1791 "error. During the first @reconnect-delay seconds, all "
1792 "requests are paused and will be rerun on a successful "
1793 "reconnect. After that time, any delayed requests and all "
1794 "future requests before a successful reconnect will "
1795 "immediately fail. Default 0",
1798 .name = "open-timeout",
1799 .type = QEMU_OPT_NUMBER,
1800 .help = "In seconds. If zero, the nbd driver tries the connection "
1801 "only once, and fails to open if the connection fails. "
1802 "If non-zero, the nbd driver will repeat connection "
1803 "attempts until successful or until @open-timeout seconds "
1804 "have elapsed. Default 0",
1806 { /* end of list */ }
1810 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1813 BDRVNBDState *s = bs->opaque;
1817 opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1818 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1822 /* Translate @host, @port, and @path to a SocketAddress */
1823 if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1827 /* Pop the config into our state object. Exit if invalid. */
1828 s->saddr = nbd_config(s, options, errp);
1833 s->export = g_strdup(qemu_opt_get(opts, "export"));
1834 if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1835 error_setg(errp, "export name too long to send to server");
1839 s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1840 if (s->tlscredsid) {
1841 s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1846 s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname"));
1847 if (!s->tlshostname &&
1848 s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1849 s->tlshostname = g_strdup(s->saddr->u.inet.host);
1853 s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1854 if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1855 error_setg(errp, "x-dirty-bitmap query too long to send to server");
1859 s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1860 s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1865 qemu_opts_del(opts);
1869 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1873 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1876 qemu_mutex_init(&s->requests_lock);
1877 qemu_co_queue_init(&s->free_sema);
1878 qemu_co_mutex_init(&s->send_mutex);
1879 qemu_co_mutex_init(&s->receive_mutex);
1881 if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1885 ret = nbd_process_options(bs, options, errp);
1890 s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1891 s->x_dirty_bitmap, s->tlscreds,
1894 if (s->open_timeout) {
1895 nbd_client_connection_enable_retry(s->conn);
1896 open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1897 s->open_timeout * NANOSECONDS_PER_SECOND);
1900 s->state = NBD_CLIENT_CONNECTING_WAIT;
1901 ret = nbd_do_establish_connection(bs, true, errp);
1907 * The connect attempt is done, so we no longer need this timer.
1908 * Delete it, because we do not want it to be around when this node
1909 * is drained or closed.
1913 nbd_client_connection_enable_retry(s->conn);
1919 nbd_clear_bdrvstate(bs);
1923 static int coroutine_fn nbd_co_flush(BlockDriverState *bs)
1925 return nbd_client_co_flush(bs);
1928 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1930 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1931 uint32_t min = s->info.min_block;
1932 uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1935 * If the server did not advertise an alignment:
1936 * - a size that is not sector-aligned implies that an alignment
1937 * of 1 can be used to access those tail bytes
1938 * - advertisement of block status requires an alignment of 1, so
1939 * that we don't violate block layer constraints that block
1940 * status is always aligned (as we can't control whether the
1941 * server will report sub-sector extents, such as a hole at EOF
1942 * on an unaligned POSIX file)
1943 * - otherwise, assume the server is so old that we are safer avoiding
1944 * sub-sector requests
1947 min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1948 s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1951 bs->bl.request_alignment = min;
1952 bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1953 bs->bl.max_pwrite_zeroes = max;
1954 bs->bl.max_transfer = max;
1956 if (s->info.opt_block &&
1957 s->info.opt_block > bs->bl.opt_transfer) {
1958 bs->bl.opt_transfer = s->info.opt_block;
1962 static void nbd_close(BlockDriverState *bs)
1964 nbd_client_close(bs);
1965 nbd_clear_bdrvstate(bs);
1969 * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1970 * to a smaller size with exact=false, there is no reason to fail the
1973 * Preallocation mode is ignored since it does not seems useful to fail when
1974 * we never change anything.
1976 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
1977 bool exact, PreallocMode prealloc,
1978 BdrvRequestFlags flags, Error **errp)
1980 BDRVNBDState *s = bs->opaque;
1982 if (offset != s->info.size && exact) {
1983 error_setg(errp, "Cannot resize NBD nodes");
1987 if (offset > s->info.size) {
1988 error_setg(errp, "Cannot grow NBD nodes");
1995 static int64_t nbd_getlength(BlockDriverState *bs)
1997 BDRVNBDState *s = bs->opaque;
1999 return s->info.size;
2002 static void nbd_refresh_filename(BlockDriverState *bs)
2004 BDRVNBDState *s = bs->opaque;
2005 const char *host = NULL, *port = NULL, *path = NULL;
2008 if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
2009 const InetSocketAddress *inet = &s->saddr->u.inet;
2010 if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
2014 } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2015 path = s->saddr->u.q_unix.path;
2016 } /* else can't represent as pseudo-filename */
2018 if (path && s->export) {
2019 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2020 "nbd+unix:///%s?socket=%s", s->export, path);
2021 } else if (path && !s->export) {
2022 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2023 "nbd+unix://?socket=%s", path);
2024 } else if (host && s->export) {
2025 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2026 "nbd://%s:%s/%s", host, port, s->export);
2027 } else if (host && !s->export) {
2028 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2029 "nbd://%s:%s", host, port);
2031 if (len >= sizeof(bs->exact_filename)) {
2032 /* Name is too long to represent exactly, so leave it empty. */
2033 bs->exact_filename[0] = '\0';
2037 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2039 /* The generic bdrv_dirname() implementation is able to work out some
2040 * directory name for NBD nodes, but that would be wrong. So far there is no
2041 * specification for how "export paths" would work, so NBD does not have
2042 * directory names. */
2043 error_setg(errp, "Cannot generate a base directory for NBD nodes");
2047 static const char *const nbd_strong_runtime_opts[] = {
2059 static void nbd_cancel_in_flight(BlockDriverState *bs)
2061 BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2063 reconnect_delay_timer_del(s);
2065 qemu_mutex_lock(&s->requests_lock);
2066 if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2067 s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2069 qemu_mutex_unlock(&s->requests_lock);
2071 nbd_co_establish_connection_cancel(s->conn);
2074 static void nbd_attach_aio_context(BlockDriverState *bs,
2075 AioContext *new_context)
2077 BDRVNBDState *s = bs->opaque;
2079 /* The open_timer is used only during nbd_open() */
2080 assert(!s->open_timer);
2083 * The reconnect_delay_timer is scheduled in I/O paths when the
2084 * connection is lost, to cancel the reconnection attempt after a
2085 * given time. Once this attempt is done (successfully or not),
2086 * nbd_reconnect_attempt() ensures the timer is deleted before the
2087 * respective I/O request is resumed.
2088 * Since the AioContext can only be changed when a node is drained,
2089 * the reconnect_delay_timer cannot be active here.
2091 assert(!s->reconnect_delay_timer);
2094 qio_channel_attach_aio_context(s->ioc, new_context);
2098 static void nbd_detach_aio_context(BlockDriverState *bs)
2100 BDRVNBDState *s = bs->opaque;
2102 assert(!s->open_timer);
2103 assert(!s->reconnect_delay_timer);
2106 qio_channel_detach_aio_context(s->ioc);
2110 static BlockDriver bdrv_nbd = {
2111 .format_name = "nbd",
2112 .protocol_name = "nbd",
2113 .instance_size = sizeof(BDRVNBDState),
2114 .bdrv_parse_filename = nbd_parse_filename,
2115 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2116 .create_opts = &bdrv_create_opts_simple,
2117 .bdrv_file_open = nbd_open,
2118 .bdrv_reopen_prepare = nbd_client_reopen_prepare,
2119 .bdrv_co_preadv = nbd_client_co_preadv,
2120 .bdrv_co_pwritev = nbd_client_co_pwritev,
2121 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes,
2122 .bdrv_close = nbd_close,
2123 .bdrv_co_flush_to_os = nbd_co_flush,
2124 .bdrv_co_pdiscard = nbd_client_co_pdiscard,
2125 .bdrv_refresh_limits = nbd_refresh_limits,
2126 .bdrv_co_truncate = nbd_co_truncate,
2127 .bdrv_getlength = nbd_getlength,
2128 .bdrv_refresh_filename = nbd_refresh_filename,
2129 .bdrv_co_block_status = nbd_client_co_block_status,
2130 .bdrv_dirname = nbd_dirname,
2131 .strong_runtime_opts = nbd_strong_runtime_opts,
2132 .bdrv_cancel_in_flight = nbd_cancel_in_flight,
2134 .bdrv_attach_aio_context = nbd_attach_aio_context,
2135 .bdrv_detach_aio_context = nbd_detach_aio_context,
2138 static BlockDriver bdrv_nbd_tcp = {
2139 .format_name = "nbd",
2140 .protocol_name = "nbd+tcp",
2141 .instance_size = sizeof(BDRVNBDState),
2142 .bdrv_parse_filename = nbd_parse_filename,
2143 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2144 .create_opts = &bdrv_create_opts_simple,
2145 .bdrv_file_open = nbd_open,
2146 .bdrv_reopen_prepare = nbd_client_reopen_prepare,
2147 .bdrv_co_preadv = nbd_client_co_preadv,
2148 .bdrv_co_pwritev = nbd_client_co_pwritev,
2149 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes,
2150 .bdrv_close = nbd_close,
2151 .bdrv_co_flush_to_os = nbd_co_flush,
2152 .bdrv_co_pdiscard = nbd_client_co_pdiscard,
2153 .bdrv_refresh_limits = nbd_refresh_limits,
2154 .bdrv_co_truncate = nbd_co_truncate,
2155 .bdrv_getlength = nbd_getlength,
2156 .bdrv_refresh_filename = nbd_refresh_filename,
2157 .bdrv_co_block_status = nbd_client_co_block_status,
2158 .bdrv_dirname = nbd_dirname,
2159 .strong_runtime_opts = nbd_strong_runtime_opts,
2160 .bdrv_cancel_in_flight = nbd_cancel_in_flight,
2162 .bdrv_attach_aio_context = nbd_attach_aio_context,
2163 .bdrv_detach_aio_context = nbd_detach_aio_context,
2166 static BlockDriver bdrv_nbd_unix = {
2167 .format_name = "nbd",
2168 .protocol_name = "nbd+unix",
2169 .instance_size = sizeof(BDRVNBDState),
2170 .bdrv_parse_filename = nbd_parse_filename,
2171 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2172 .create_opts = &bdrv_create_opts_simple,
2173 .bdrv_file_open = nbd_open,
2174 .bdrv_reopen_prepare = nbd_client_reopen_prepare,
2175 .bdrv_co_preadv = nbd_client_co_preadv,
2176 .bdrv_co_pwritev = nbd_client_co_pwritev,
2177 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes,
2178 .bdrv_close = nbd_close,
2179 .bdrv_co_flush_to_os = nbd_co_flush,
2180 .bdrv_co_pdiscard = nbd_client_co_pdiscard,
2181 .bdrv_refresh_limits = nbd_refresh_limits,
2182 .bdrv_co_truncate = nbd_co_truncate,
2183 .bdrv_getlength = nbd_getlength,
2184 .bdrv_refresh_filename = nbd_refresh_filename,
2185 .bdrv_co_block_status = nbd_client_co_block_status,
2186 .bdrv_dirname = nbd_dirname,
2187 .strong_runtime_opts = nbd_strong_runtime_opts,
2188 .bdrv_cancel_in_flight = nbd_cancel_in_flight,
2190 .bdrv_attach_aio_context = nbd_attach_aio_context,
2191 .bdrv_detach_aio_context = nbd_detach_aio_context,
2194 static void bdrv_nbd_init(void)
2196 bdrv_register(&bdrv_nbd);
2197 bdrv_register(&bdrv_nbd_tcp);
2198 bdrv_register(&bdrv_nbd_unix);
2201 block_init(bdrv_nbd_init);