xref: /openbmc/qemu/block/nbd.c (revision 1c8f85d9)
1 /*
2  * QEMU Block driver for  NBD
3  *
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>
8  *
9  * Some parts:
10  *    Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
11  *
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:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
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
28  * THE SOFTWARE.
29  */
30 
31 #include "qemu/osdep.h"
32 
33 #include "trace.h"
34 #include "qemu/uri.h"
35 #include "qemu/option.h"
36 #include "qemu/cutils.h"
37 #include "qemu/main-loop.h"
38 
39 #include "qapi/qapi-visit-sockets.h"
40 #include "qapi/qmp/qstring.h"
41 
42 #include "block/qdict.h"
43 #include "block/nbd.h"
44 #include "block/block_int.h"
45 
46 #define EN_OPTSTR ":exportname="
47 #define MAX_NBD_REQUESTS    16
48 
49 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
50 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ (uint64_t)(intptr_t)(bs))
51 
52 typedef struct {
53     Coroutine *coroutine;
54     uint64_t offset;        /* original offset of the request */
55     bool receiving;         /* waiting for connection_co? */
56 } NBDClientRequest;
57 
58 typedef enum NBDClientState {
59     NBD_CLIENT_CONNECTING_WAIT,
60     NBD_CLIENT_CONNECTING_NOWAIT,
61     NBD_CLIENT_CONNECTED,
62     NBD_CLIENT_QUIT
63 } NBDClientState;
64 
65 typedef struct BDRVNBDState {
66     QIOChannelSocket *sioc; /* The master data channel */
67     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
68     NBDExportInfo info;
69 
70     CoMutex send_mutex;
71     CoQueue free_sema;
72     Coroutine *connection_co;
73     QemuCoSleepState *connection_co_sleep_ns_state;
74     bool drained;
75     bool wait_drained_end;
76     int in_flight;
77     NBDClientState state;
78     int connect_status;
79     Error *connect_err;
80     bool wait_in_flight;
81 
82     NBDClientRequest requests[MAX_NBD_REQUESTS];
83     NBDReply reply;
84     BlockDriverState *bs;
85 
86     /* Connection parameters */
87     uint32_t reconnect_delay;
88     SocketAddress *saddr;
89     char *export, *tlscredsid;
90     QCryptoTLSCreds *tlscreds;
91     const char *hostname;
92     char *x_dirty_bitmap;
93 } BDRVNBDState;
94 
95 static int nbd_client_connect(BlockDriverState *bs, Error **errp);
96 
97 static void nbd_channel_error(BDRVNBDState *s, int ret)
98 {
99     if (ret == -EIO) {
100         if (s->state == NBD_CLIENT_CONNECTED) {
101             s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT :
102                                             NBD_CLIENT_CONNECTING_NOWAIT;
103         }
104     } else {
105         if (s->state == NBD_CLIENT_CONNECTED) {
106             qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
107         }
108         s->state = NBD_CLIENT_QUIT;
109     }
110 }
111 
112 static void nbd_recv_coroutines_wake_all(BDRVNBDState *s)
113 {
114     int i;
115 
116     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
117         NBDClientRequest *req = &s->requests[i];
118 
119         if (req->coroutine && req->receiving) {
120             aio_co_wake(req->coroutine);
121         }
122     }
123 }
124 
125 static void nbd_client_detach_aio_context(BlockDriverState *bs)
126 {
127     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
128 
129     qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc));
130 }
131 
132 static void nbd_client_attach_aio_context_bh(void *opaque)
133 {
134     BlockDriverState *bs = opaque;
135     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
136 
137     /*
138      * The node is still drained, so we know the coroutine has yielded in
139      * nbd_read_eof(), the only place where bs->in_flight can reach 0, or it is
140      * entered for the first time. Both places are safe for entering the
141      * coroutine.
142      */
143     qemu_aio_coroutine_enter(bs->aio_context, s->connection_co);
144     bdrv_dec_in_flight(bs);
145 }
146 
147 static void nbd_client_attach_aio_context(BlockDriverState *bs,
148                                           AioContext *new_context)
149 {
150     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
151 
152     /*
153      * s->connection_co is either yielded from nbd_receive_reply or from
154      * nbd_co_reconnect_loop()
155      */
156     if (s->state == NBD_CLIENT_CONNECTED) {
157         qio_channel_attach_aio_context(QIO_CHANNEL(s->ioc), new_context);
158     }
159 
160     bdrv_inc_in_flight(bs);
161 
162     /*
163      * Need to wait here for the BH to run because the BH must run while the
164      * node is still drained.
165      */
166     aio_wait_bh_oneshot(new_context, nbd_client_attach_aio_context_bh, bs);
167 }
168 
169 static void coroutine_fn nbd_client_co_drain_begin(BlockDriverState *bs)
170 {
171     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
172 
173     s->drained = true;
174     if (s->connection_co_sleep_ns_state) {
175         qemu_co_sleep_wake(s->connection_co_sleep_ns_state);
176     }
177 }
178 
179 static void coroutine_fn nbd_client_co_drain_end(BlockDriverState *bs)
180 {
181     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
182 
183     s->drained = false;
184     if (s->wait_drained_end) {
185         s->wait_drained_end = false;
186         aio_co_wake(s->connection_co);
187     }
188 }
189 
190 
191 static void nbd_teardown_connection(BlockDriverState *bs)
192 {
193     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
194 
195     if (s->state == NBD_CLIENT_CONNECTED) {
196         /* finish any pending coroutines */
197         assert(s->ioc);
198         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
199     }
200     s->state = NBD_CLIENT_QUIT;
201     if (s->connection_co) {
202         if (s->connection_co_sleep_ns_state) {
203             qemu_co_sleep_wake(s->connection_co_sleep_ns_state);
204         }
205     }
206     BDRV_POLL_WHILE(bs, s->connection_co);
207 }
208 
209 static bool nbd_client_connecting(BDRVNBDState *s)
210 {
211     return s->state == NBD_CLIENT_CONNECTING_WAIT ||
212         s->state == NBD_CLIENT_CONNECTING_NOWAIT;
213 }
214 
215 static bool nbd_client_connecting_wait(BDRVNBDState *s)
216 {
217     return s->state == NBD_CLIENT_CONNECTING_WAIT;
218 }
219 
220 static coroutine_fn void nbd_reconnect_attempt(BDRVNBDState *s)
221 {
222     Error *local_err = NULL;
223 
224     if (!nbd_client_connecting(s)) {
225         return;
226     }
227 
228     /* Wait for completion of all in-flight requests */
229 
230     qemu_co_mutex_lock(&s->send_mutex);
231 
232     while (s->in_flight > 0) {
233         qemu_co_mutex_unlock(&s->send_mutex);
234         nbd_recv_coroutines_wake_all(s);
235         s->wait_in_flight = true;
236         qemu_coroutine_yield();
237         s->wait_in_flight = false;
238         qemu_co_mutex_lock(&s->send_mutex);
239     }
240 
241     qemu_co_mutex_unlock(&s->send_mutex);
242 
243     if (!nbd_client_connecting(s)) {
244         return;
245     }
246 
247     /*
248      * Now we are sure that nobody is accessing the channel, and no one will
249      * try until we set the state to CONNECTED.
250      */
251 
252     /* Finalize previous connection if any */
253     if (s->ioc) {
254         nbd_client_detach_aio_context(s->bs);
255         object_unref(OBJECT(s->sioc));
256         s->sioc = NULL;
257         object_unref(OBJECT(s->ioc));
258         s->ioc = NULL;
259     }
260 
261     s->connect_status = nbd_client_connect(s->bs, &local_err);
262     error_free(s->connect_err);
263     s->connect_err = NULL;
264     error_propagate(&s->connect_err, local_err);
265 
266     if (s->connect_status < 0) {
267         /* failed attempt */
268         return;
269     }
270 
271     /* successfully connected */
272     s->state = NBD_CLIENT_CONNECTED;
273     qemu_co_queue_restart_all(&s->free_sema);
274 }
275 
276 static coroutine_fn void nbd_co_reconnect_loop(BDRVNBDState *s)
277 {
278     uint64_t start_time_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
279     uint64_t delay_ns = s->reconnect_delay * NANOSECONDS_PER_SECOND;
280     uint64_t timeout = 1 * NANOSECONDS_PER_SECOND;
281     uint64_t max_timeout = 16 * NANOSECONDS_PER_SECOND;
282 
283     nbd_reconnect_attempt(s);
284 
285     while (nbd_client_connecting(s)) {
286         if (s->state == NBD_CLIENT_CONNECTING_WAIT &&
287             qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start_time_ns > delay_ns)
288         {
289             s->state = NBD_CLIENT_CONNECTING_NOWAIT;
290             qemu_co_queue_restart_all(&s->free_sema);
291         }
292 
293         qemu_co_sleep_ns_wakeable(QEMU_CLOCK_REALTIME, timeout,
294                                   &s->connection_co_sleep_ns_state);
295         if (s->drained) {
296             bdrv_dec_in_flight(s->bs);
297             s->wait_drained_end = true;
298             while (s->drained) {
299                 /*
300                  * We may be entered once from nbd_client_attach_aio_context_bh
301                  * and then from nbd_client_co_drain_end. So here is a loop.
302                  */
303                 qemu_coroutine_yield();
304             }
305             bdrv_inc_in_flight(s->bs);
306         }
307         if (timeout < max_timeout) {
308             timeout *= 2;
309         }
310 
311         nbd_reconnect_attempt(s);
312     }
313 }
314 
315 static coroutine_fn void nbd_connection_entry(void *opaque)
316 {
317     BDRVNBDState *s = opaque;
318     uint64_t i;
319     int ret = 0;
320     Error *local_err = NULL;
321 
322     while (s->state != NBD_CLIENT_QUIT) {
323         /*
324          * The NBD client can only really be considered idle when it has
325          * yielded from qio_channel_readv_all_eof(), waiting for data. This is
326          * the point where the additional scheduled coroutine entry happens
327          * after nbd_client_attach_aio_context().
328          *
329          * Therefore we keep an additional in_flight reference all the time and
330          * only drop it temporarily here.
331          */
332 
333         if (nbd_client_connecting(s)) {
334             nbd_co_reconnect_loop(s);
335         }
336 
337         if (s->state != NBD_CLIENT_CONNECTED) {
338             continue;
339         }
340 
341         assert(s->reply.handle == 0);
342         ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, &local_err);
343 
344         if (local_err) {
345             trace_nbd_read_reply_entry_fail(ret, error_get_pretty(local_err));
346             error_free(local_err);
347             local_err = NULL;
348         }
349         if (ret <= 0) {
350             nbd_channel_error(s, ret ? ret : -EIO);
351             continue;
352         }
353 
354         /*
355          * There's no need for a mutex on the receive side, because the
356          * handler acts as a synchronization point and ensures that only
357          * one coroutine is called until the reply finishes.
358          */
359         i = HANDLE_TO_INDEX(s, s->reply.handle);
360         if (i >= MAX_NBD_REQUESTS ||
361             !s->requests[i].coroutine ||
362             !s->requests[i].receiving ||
363             (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply))
364         {
365             nbd_channel_error(s, -EINVAL);
366             continue;
367         }
368 
369         /*
370          * We're woken up again by the request itself.  Note that there
371          * is no race between yielding and reentering connection_co.  This
372          * is because:
373          *
374          * - if the request runs on the same AioContext, it is only
375          *   entered after we yield
376          *
377          * - if the request runs on a different AioContext, reentering
378          *   connection_co happens through a bottom half, which can only
379          *   run after we yield.
380          */
381         aio_co_wake(s->requests[i].coroutine);
382         qemu_coroutine_yield();
383     }
384 
385     qemu_co_queue_restart_all(&s->free_sema);
386     nbd_recv_coroutines_wake_all(s);
387     bdrv_dec_in_flight(s->bs);
388 
389     s->connection_co = NULL;
390     if (s->ioc) {
391         nbd_client_detach_aio_context(s->bs);
392         object_unref(OBJECT(s->sioc));
393         s->sioc = NULL;
394         object_unref(OBJECT(s->ioc));
395         s->ioc = NULL;
396     }
397 
398     aio_wait_kick();
399 }
400 
401 static int nbd_co_send_request(BlockDriverState *bs,
402                                NBDRequest *request,
403                                QEMUIOVector *qiov)
404 {
405     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
406     int rc, i = -1;
407 
408     qemu_co_mutex_lock(&s->send_mutex);
409     while (s->in_flight == MAX_NBD_REQUESTS || nbd_client_connecting_wait(s)) {
410         qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
411     }
412 
413     if (s->state != NBD_CLIENT_CONNECTED) {
414         rc = -EIO;
415         goto err;
416     }
417 
418     s->in_flight++;
419 
420     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
421         if (s->requests[i].coroutine == NULL) {
422             break;
423         }
424     }
425 
426     g_assert(qemu_in_coroutine());
427     assert(i < MAX_NBD_REQUESTS);
428 
429     s->requests[i].coroutine = qemu_coroutine_self();
430     s->requests[i].offset = request->from;
431     s->requests[i].receiving = false;
432 
433     request->handle = INDEX_TO_HANDLE(s, i);
434 
435     assert(s->ioc);
436 
437     if (qiov) {
438         qio_channel_set_cork(s->ioc, true);
439         rc = nbd_send_request(s->ioc, request);
440         if (rc >= 0 && s->state == NBD_CLIENT_CONNECTED) {
441             if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
442                                        NULL) < 0) {
443                 rc = -EIO;
444             }
445         } else if (rc >= 0) {
446             rc = -EIO;
447         }
448         qio_channel_set_cork(s->ioc, false);
449     } else {
450         rc = nbd_send_request(s->ioc, request);
451     }
452 
453 err:
454     if (rc < 0) {
455         nbd_channel_error(s, rc);
456         if (i != -1) {
457             s->requests[i].coroutine = NULL;
458             s->in_flight--;
459         }
460         if (s->in_flight == 0 && s->wait_in_flight) {
461             aio_co_wake(s->connection_co);
462         } else {
463             qemu_co_queue_next(&s->free_sema);
464         }
465     }
466     qemu_co_mutex_unlock(&s->send_mutex);
467     return rc;
468 }
469 
470 static inline uint16_t payload_advance16(uint8_t **payload)
471 {
472     *payload += 2;
473     return lduw_be_p(*payload - 2);
474 }
475 
476 static inline uint32_t payload_advance32(uint8_t **payload)
477 {
478     *payload += 4;
479     return ldl_be_p(*payload - 4);
480 }
481 
482 static inline uint64_t payload_advance64(uint8_t **payload)
483 {
484     *payload += 8;
485     return ldq_be_p(*payload - 8);
486 }
487 
488 static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
489                                          NBDStructuredReplyChunk *chunk,
490                                          uint8_t *payload, uint64_t orig_offset,
491                                          QEMUIOVector *qiov, Error **errp)
492 {
493     uint64_t offset;
494     uint32_t hole_size;
495 
496     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
497         error_setg(errp, "Protocol error: invalid payload for "
498                          "NBD_REPLY_TYPE_OFFSET_HOLE");
499         return -EINVAL;
500     }
501 
502     offset = payload_advance64(&payload);
503     hole_size = payload_advance32(&payload);
504 
505     if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
506         offset > orig_offset + qiov->size - hole_size) {
507         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
508                          " region");
509         return -EINVAL;
510     }
511     if (s->info.min_block &&
512         !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
513         trace_nbd_structured_read_compliance("hole");
514     }
515 
516     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
517 
518     return 0;
519 }
520 
521 /*
522  * nbd_parse_blockstatus_payload
523  * Based on our request, we expect only one extent in reply, for the
524  * base:allocation context.
525  */
526 static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
527                                          NBDStructuredReplyChunk *chunk,
528                                          uint8_t *payload, uint64_t orig_length,
529                                          NBDExtent *extent, Error **errp)
530 {
531     uint32_t context_id;
532 
533     /* The server succeeded, so it must have sent [at least] one extent */
534     if (chunk->length < sizeof(context_id) + sizeof(*extent)) {
535         error_setg(errp, "Protocol error: invalid payload for "
536                          "NBD_REPLY_TYPE_BLOCK_STATUS");
537         return -EINVAL;
538     }
539 
540     context_id = payload_advance32(&payload);
541     if (s->info.context_id != context_id) {
542         error_setg(errp, "Protocol error: unexpected context id %d for "
543                          "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
544                          "id is %d", context_id,
545                          s->info.context_id);
546         return -EINVAL;
547     }
548 
549     extent->length = payload_advance32(&payload);
550     extent->flags = payload_advance32(&payload);
551 
552     if (extent->length == 0) {
553         error_setg(errp, "Protocol error: server sent status chunk with "
554                    "zero length");
555         return -EINVAL;
556     }
557 
558     /*
559      * A server sending unaligned block status is in violation of the
560      * protocol, but as qemu-nbd 3.1 is such a server (at least for
561      * POSIX files that are not a multiple of 512 bytes, since qemu
562      * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
563      * still sees an implicit hole beyond the real EOF), it's nicer to
564      * work around the misbehaving server. If the request included
565      * more than the final unaligned block, truncate it back to an
566      * aligned result; if the request was only the final block, round
567      * up to the full block and change the status to fully-allocated
568      * (always a safe status, even if it loses information).
569      */
570     if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
571                                                    s->info.min_block)) {
572         trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
573         if (extent->length > s->info.min_block) {
574             extent->length = QEMU_ALIGN_DOWN(extent->length,
575                                              s->info.min_block);
576         } else {
577             extent->length = s->info.min_block;
578             extent->flags = 0;
579         }
580     }
581 
582     /*
583      * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
584      * sent us any more than one extent, nor should it have included
585      * status beyond our request in that extent. However, it's easy
586      * enough to ignore the server's noncompliance without killing the
587      * connection; just ignore trailing extents, and clamp things to
588      * the length of our request.
589      */
590     if (chunk->length > sizeof(context_id) + sizeof(*extent)) {
591         trace_nbd_parse_blockstatus_compliance("more than one extent");
592     }
593     if (extent->length > orig_length) {
594         extent->length = orig_length;
595         trace_nbd_parse_blockstatus_compliance("extent length too large");
596     }
597 
598     return 0;
599 }
600 
601 /*
602  * nbd_parse_error_payload
603  * on success @errp contains message describing nbd error reply
604  */
605 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
606                                    uint8_t *payload, int *request_ret,
607                                    Error **errp)
608 {
609     uint32_t error;
610     uint16_t message_size;
611 
612     assert(chunk->type & (1 << 15));
613 
614     if (chunk->length < sizeof(error) + sizeof(message_size)) {
615         error_setg(errp,
616                    "Protocol error: invalid payload for structured error");
617         return -EINVAL;
618     }
619 
620     error = nbd_errno_to_system_errno(payload_advance32(&payload));
621     if (error == 0) {
622         error_setg(errp, "Protocol error: server sent structured error chunk "
623                          "with error = 0");
624         return -EINVAL;
625     }
626 
627     *request_ret = -error;
628     message_size = payload_advance16(&payload);
629 
630     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
631         error_setg(errp, "Protocol error: server sent structured error chunk "
632                          "with incorrect message size");
633         return -EINVAL;
634     }
635 
636     /* TODO: Add a trace point to mention the server complaint */
637 
638     /* TODO handle ERROR_OFFSET */
639 
640     return 0;
641 }
642 
643 static int nbd_co_receive_offset_data_payload(BDRVNBDState *s,
644                                               uint64_t orig_offset,
645                                               QEMUIOVector *qiov, Error **errp)
646 {
647     QEMUIOVector sub_qiov;
648     uint64_t offset;
649     size_t data_size;
650     int ret;
651     NBDStructuredReplyChunk *chunk = &s->reply.structured;
652 
653     assert(nbd_reply_is_structured(&s->reply));
654 
655     /* The NBD spec requires at least one byte of payload */
656     if (chunk->length <= sizeof(offset)) {
657         error_setg(errp, "Protocol error: invalid payload for "
658                          "NBD_REPLY_TYPE_OFFSET_DATA");
659         return -EINVAL;
660     }
661 
662     if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
663         return -EIO;
664     }
665 
666     data_size = chunk->length - sizeof(offset);
667     assert(data_size);
668     if (offset < orig_offset || data_size > qiov->size ||
669         offset > orig_offset + qiov->size - data_size) {
670         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
671                          " region");
672         return -EINVAL;
673     }
674     if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
675         trace_nbd_structured_read_compliance("data");
676     }
677 
678     qemu_iovec_init(&sub_qiov, qiov->niov);
679     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
680     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
681     qemu_iovec_destroy(&sub_qiov);
682 
683     return ret < 0 ? -EIO : 0;
684 }
685 
686 #define NBD_MAX_MALLOC_PAYLOAD 1000
687 static coroutine_fn int nbd_co_receive_structured_payload(
688         BDRVNBDState *s, void **payload, Error **errp)
689 {
690     int ret;
691     uint32_t len;
692 
693     assert(nbd_reply_is_structured(&s->reply));
694 
695     len = s->reply.structured.length;
696 
697     if (len == 0) {
698         return 0;
699     }
700 
701     if (payload == NULL) {
702         error_setg(errp, "Unexpected structured payload");
703         return -EINVAL;
704     }
705 
706     if (len > NBD_MAX_MALLOC_PAYLOAD) {
707         error_setg(errp, "Payload too large");
708         return -EINVAL;
709     }
710 
711     *payload = g_new(char, len);
712     ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
713     if (ret < 0) {
714         g_free(*payload);
715         *payload = NULL;
716         return ret;
717     }
718 
719     return 0;
720 }
721 
722 /*
723  * nbd_co_do_receive_one_chunk
724  * for simple reply:
725  *   set request_ret to received reply error
726  *   if qiov is not NULL: read payload to @qiov
727  * for structured reply chunk:
728  *   if error chunk: read payload, set @request_ret, do not set @payload
729  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
730  *   else: read payload to @payload
731  *
732  * If function fails, @errp contains corresponding error message, and the
733  * connection with the server is suspect.  If it returns 0, then the
734  * transaction succeeded (although @request_ret may be a negative errno
735  * corresponding to the server's error reply), and errp is unchanged.
736  */
737 static coroutine_fn int nbd_co_do_receive_one_chunk(
738         BDRVNBDState *s, uint64_t handle, bool only_structured,
739         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
740 {
741     int ret;
742     int i = HANDLE_TO_INDEX(s, handle);
743     void *local_payload = NULL;
744     NBDStructuredReplyChunk *chunk;
745 
746     if (payload) {
747         *payload = NULL;
748     }
749     *request_ret = 0;
750 
751     /* Wait until we're woken up by nbd_connection_entry.  */
752     s->requests[i].receiving = true;
753     qemu_coroutine_yield();
754     s->requests[i].receiving = false;
755     if (s->state != NBD_CLIENT_CONNECTED) {
756         error_setg(errp, "Connection closed");
757         return -EIO;
758     }
759     assert(s->ioc);
760 
761     assert(s->reply.handle == handle);
762 
763     if (nbd_reply_is_simple(&s->reply)) {
764         if (only_structured) {
765             error_setg(errp, "Protocol error: simple reply when structured "
766                              "reply chunk was expected");
767             return -EINVAL;
768         }
769 
770         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
771         if (*request_ret < 0 || !qiov) {
772             return 0;
773         }
774 
775         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
776                                      errp) < 0 ? -EIO : 0;
777     }
778 
779     /* handle structured reply chunk */
780     assert(s->info.structured_reply);
781     chunk = &s->reply.structured;
782 
783     if (chunk->type == NBD_REPLY_TYPE_NONE) {
784         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
785             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
786                        " NBD_REPLY_FLAG_DONE flag set");
787             return -EINVAL;
788         }
789         if (chunk->length) {
790             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
791                        " nonzero length");
792             return -EINVAL;
793         }
794         return 0;
795     }
796 
797     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
798         if (!qiov) {
799             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
800             return -EINVAL;
801         }
802 
803         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
804                                                   qiov, errp);
805     }
806 
807     if (nbd_reply_type_is_error(chunk->type)) {
808         payload = &local_payload;
809     }
810 
811     ret = nbd_co_receive_structured_payload(s, payload, errp);
812     if (ret < 0) {
813         return ret;
814     }
815 
816     if (nbd_reply_type_is_error(chunk->type)) {
817         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
818         g_free(local_payload);
819         return ret;
820     }
821 
822     return 0;
823 }
824 
825 /*
826  * nbd_co_receive_one_chunk
827  * Read reply, wake up connection_co and set s->quit if needed.
828  * Return value is a fatal error code or normal nbd reply error code
829  */
830 static coroutine_fn int nbd_co_receive_one_chunk(
831         BDRVNBDState *s, uint64_t handle, bool only_structured,
832         int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
833         Error **errp)
834 {
835     int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
836                                           request_ret, qiov, payload, errp);
837 
838     if (ret < 0) {
839         memset(reply, 0, sizeof(*reply));
840         nbd_channel_error(s, ret);
841     } else {
842         /* For assert at loop start in nbd_connection_entry */
843         *reply = s->reply;
844     }
845     s->reply.handle = 0;
846 
847     if (s->connection_co && !s->wait_in_flight) {
848         /*
849          * We must check s->wait_in_flight, because we may entered by
850          * nbd_recv_coroutines_wake_all(), in this case we should not
851          * wake connection_co here, it will woken by last request.
852          */
853         aio_co_wake(s->connection_co);
854     }
855 
856     return ret;
857 }
858 
859 typedef struct NBDReplyChunkIter {
860     int ret;
861     int request_ret;
862     Error *err;
863     bool done, only_structured;
864 } NBDReplyChunkIter;
865 
866 static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
867                                    int ret, Error **local_err)
868 {
869     assert(local_err && *local_err);
870     assert(ret < 0);
871 
872     if (!iter->ret) {
873         iter->ret = ret;
874         error_propagate(&iter->err, *local_err);
875     } else {
876         error_free(*local_err);
877     }
878 
879     *local_err = NULL;
880 }
881 
882 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
883 {
884     assert(ret < 0);
885 
886     if (!iter->request_ret) {
887         iter->request_ret = ret;
888     }
889 }
890 
891 /*
892  * NBD_FOREACH_REPLY_CHUNK
893  * The pointer stored in @payload requires g_free() to free it.
894  */
895 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
896                                 qiov, reply, payload) \
897     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
898          nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
899 
900 /*
901  * nbd_reply_chunk_iter_receive
902  * The pointer stored in @payload requires g_free() to free it.
903  */
904 static bool nbd_reply_chunk_iter_receive(BDRVNBDState *s,
905                                          NBDReplyChunkIter *iter,
906                                          uint64_t handle,
907                                          QEMUIOVector *qiov, NBDReply *reply,
908                                          void **payload)
909 {
910     int ret, request_ret;
911     NBDReply local_reply;
912     NBDStructuredReplyChunk *chunk;
913     Error *local_err = NULL;
914     if (s->state != NBD_CLIENT_CONNECTED) {
915         error_setg(&local_err, "Connection closed");
916         nbd_iter_channel_error(iter, -EIO, &local_err);
917         goto break_loop;
918     }
919 
920     if (iter->done) {
921         /* Previous iteration was last. */
922         goto break_loop;
923     }
924 
925     if (reply == NULL) {
926         reply = &local_reply;
927     }
928 
929     ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
930                                    &request_ret, qiov, reply, payload,
931                                    &local_err);
932     if (ret < 0) {
933         nbd_iter_channel_error(iter, ret, &local_err);
934     } else if (request_ret < 0) {
935         nbd_iter_request_error(iter, request_ret);
936     }
937 
938     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
939     if (nbd_reply_is_simple(reply) || s->state != NBD_CLIENT_CONNECTED) {
940         goto break_loop;
941     }
942 
943     chunk = &reply->structured;
944     iter->only_structured = true;
945 
946     if (chunk->type == NBD_REPLY_TYPE_NONE) {
947         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
948         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
949         goto break_loop;
950     }
951 
952     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
953         /* This iteration is last. */
954         iter->done = true;
955     }
956 
957     /* Execute the loop body */
958     return true;
959 
960 break_loop:
961     s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
962 
963     qemu_co_mutex_lock(&s->send_mutex);
964     s->in_flight--;
965     if (s->in_flight == 0 && s->wait_in_flight) {
966         aio_co_wake(s->connection_co);
967     } else {
968         qemu_co_queue_next(&s->free_sema);
969     }
970     qemu_co_mutex_unlock(&s->send_mutex);
971 
972     return false;
973 }
974 
975 static int nbd_co_receive_return_code(BDRVNBDState *s, uint64_t handle,
976                                       int *request_ret, Error **errp)
977 {
978     NBDReplyChunkIter iter;
979 
980     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
981         /* nbd_reply_chunk_iter_receive does all the work */
982     }
983 
984     error_propagate(errp, iter.err);
985     *request_ret = iter.request_ret;
986     return iter.ret;
987 }
988 
989 static int nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t handle,
990                                         uint64_t offset, QEMUIOVector *qiov,
991                                         int *request_ret, Error **errp)
992 {
993     NBDReplyChunkIter iter;
994     NBDReply reply;
995     void *payload = NULL;
996     Error *local_err = NULL;
997 
998     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
999                             qiov, &reply, &payload)
1000     {
1001         int ret;
1002         NBDStructuredReplyChunk *chunk = &reply.structured;
1003 
1004         assert(nbd_reply_is_structured(&reply));
1005 
1006         switch (chunk->type) {
1007         case NBD_REPLY_TYPE_OFFSET_DATA:
1008             /*
1009              * special cased in nbd_co_receive_one_chunk, data is already
1010              * in qiov
1011              */
1012             break;
1013         case NBD_REPLY_TYPE_OFFSET_HOLE:
1014             ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1015                                                 offset, qiov, &local_err);
1016             if (ret < 0) {
1017                 nbd_channel_error(s, ret);
1018                 nbd_iter_channel_error(&iter, ret, &local_err);
1019             }
1020             break;
1021         default:
1022             if (!nbd_reply_type_is_error(chunk->type)) {
1023                 /* not allowed reply type */
1024                 nbd_channel_error(s, -EINVAL);
1025                 error_setg(&local_err,
1026                            "Unexpected reply type: %d (%s) for CMD_READ",
1027                            chunk->type, nbd_reply_type_lookup(chunk->type));
1028                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1029             }
1030         }
1031 
1032         g_free(payload);
1033         payload = NULL;
1034     }
1035 
1036     error_propagate(errp, iter.err);
1037     *request_ret = iter.request_ret;
1038     return iter.ret;
1039 }
1040 
1041 static int nbd_co_receive_blockstatus_reply(BDRVNBDState *s,
1042                                             uint64_t handle, uint64_t length,
1043                                             NBDExtent *extent,
1044                                             int *request_ret, Error **errp)
1045 {
1046     NBDReplyChunkIter iter;
1047     NBDReply reply;
1048     void *payload = NULL;
1049     Error *local_err = NULL;
1050     bool received = false;
1051 
1052     assert(!extent->length);
1053     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) {
1054         int ret;
1055         NBDStructuredReplyChunk *chunk = &reply.structured;
1056 
1057         assert(nbd_reply_is_structured(&reply));
1058 
1059         switch (chunk->type) {
1060         case NBD_REPLY_TYPE_BLOCK_STATUS:
1061             if (received) {
1062                 nbd_channel_error(s, -EINVAL);
1063                 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1064                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1065             }
1066             received = true;
1067 
1068             ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1069                                                 payload, length, extent,
1070                                                 &local_err);
1071             if (ret < 0) {
1072                 nbd_channel_error(s, ret);
1073                 nbd_iter_channel_error(&iter, ret, &local_err);
1074             }
1075             break;
1076         default:
1077             if (!nbd_reply_type_is_error(chunk->type)) {
1078                 nbd_channel_error(s, -EINVAL);
1079                 error_setg(&local_err,
1080                            "Unexpected reply type: %d (%s) "
1081                            "for CMD_BLOCK_STATUS",
1082                            chunk->type, nbd_reply_type_lookup(chunk->type));
1083                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1084             }
1085         }
1086 
1087         g_free(payload);
1088         payload = NULL;
1089     }
1090 
1091     if (!extent->length && !iter.request_ret) {
1092         error_setg(&local_err, "Server did not reply with any status extents");
1093         nbd_iter_channel_error(&iter, -EIO, &local_err);
1094     }
1095 
1096     error_propagate(errp, iter.err);
1097     *request_ret = iter.request_ret;
1098     return iter.ret;
1099 }
1100 
1101 static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1102                           QEMUIOVector *write_qiov)
1103 {
1104     int ret, request_ret;
1105     Error *local_err = NULL;
1106     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1107 
1108     assert(request->type != NBD_CMD_READ);
1109     if (write_qiov) {
1110         assert(request->type == NBD_CMD_WRITE);
1111         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1112     } else {
1113         assert(request->type != NBD_CMD_WRITE);
1114     }
1115 
1116     do {
1117         ret = nbd_co_send_request(bs, request, write_qiov);
1118         if (ret < 0) {
1119             continue;
1120         }
1121 
1122         ret = nbd_co_receive_return_code(s, request->handle,
1123                                          &request_ret, &local_err);
1124         if (local_err) {
1125             trace_nbd_co_request_fail(request->from, request->len,
1126                                       request->handle, request->flags,
1127                                       request->type,
1128                                       nbd_cmd_lookup(request->type),
1129                                       ret, error_get_pretty(local_err));
1130             error_free(local_err);
1131             local_err = NULL;
1132         }
1133     } while (ret < 0 && nbd_client_connecting_wait(s));
1134 
1135     return ret ? ret : request_ret;
1136 }
1137 
1138 static int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
1139                                 uint64_t bytes, QEMUIOVector *qiov, int flags)
1140 {
1141     int ret, request_ret;
1142     Error *local_err = NULL;
1143     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1144     NBDRequest request = {
1145         .type = NBD_CMD_READ,
1146         .from = offset,
1147         .len = bytes,
1148     };
1149 
1150     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1151     assert(!flags);
1152 
1153     if (!bytes) {
1154         return 0;
1155     }
1156     /*
1157      * Work around the fact that the block layer doesn't do
1158      * byte-accurate sizing yet - if the read exceeds the server's
1159      * advertised size because the block layer rounded size up, then
1160      * truncate the request to the server and tail-pad with zero.
1161      */
1162     if (offset >= s->info.size) {
1163         assert(bytes < BDRV_SECTOR_SIZE);
1164         qemu_iovec_memset(qiov, 0, 0, bytes);
1165         return 0;
1166     }
1167     if (offset + bytes > s->info.size) {
1168         uint64_t slop = offset + bytes - s->info.size;
1169 
1170         assert(slop < BDRV_SECTOR_SIZE);
1171         qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1172         request.len -= slop;
1173     }
1174 
1175     do {
1176         ret = nbd_co_send_request(bs, &request, NULL);
1177         if (ret < 0) {
1178             continue;
1179         }
1180 
1181         ret = nbd_co_receive_cmdread_reply(s, request.handle, offset, qiov,
1182                                            &request_ret, &local_err);
1183         if (local_err) {
1184             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1185                                       request.flags, request.type,
1186                                       nbd_cmd_lookup(request.type),
1187                                       ret, error_get_pretty(local_err));
1188             error_free(local_err);
1189             local_err = NULL;
1190         }
1191     } while (ret < 0 && nbd_client_connecting_wait(s));
1192 
1193     return ret ? ret : request_ret;
1194 }
1195 
1196 static int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
1197                                  uint64_t bytes, QEMUIOVector *qiov, int flags)
1198 {
1199     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1200     NBDRequest request = {
1201         .type = NBD_CMD_WRITE,
1202         .from = offset,
1203         .len = bytes,
1204     };
1205 
1206     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1207     if (flags & BDRV_REQ_FUA) {
1208         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1209         request.flags |= NBD_CMD_FLAG_FUA;
1210     }
1211 
1212     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1213 
1214     if (!bytes) {
1215         return 0;
1216     }
1217     return nbd_co_request(bs, &request, qiov);
1218 }
1219 
1220 static int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1221                                        int bytes, BdrvRequestFlags flags)
1222 {
1223     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1224     NBDRequest request = {
1225         .type = NBD_CMD_WRITE_ZEROES,
1226         .from = offset,
1227         .len = bytes,
1228     };
1229 
1230     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1231     if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1232         return -ENOTSUP;
1233     }
1234 
1235     if (flags & BDRV_REQ_FUA) {
1236         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1237         request.flags |= NBD_CMD_FLAG_FUA;
1238     }
1239     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1240         request.flags |= NBD_CMD_FLAG_NO_HOLE;
1241     }
1242     if (flags & BDRV_REQ_NO_FALLBACK) {
1243         assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1244         request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1245     }
1246 
1247     if (!bytes) {
1248         return 0;
1249     }
1250     return nbd_co_request(bs, &request, NULL);
1251 }
1252 
1253 static int nbd_client_co_flush(BlockDriverState *bs)
1254 {
1255     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1256     NBDRequest request = { .type = NBD_CMD_FLUSH };
1257 
1258     if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1259         return 0;
1260     }
1261 
1262     request.from = 0;
1263     request.len = 0;
1264 
1265     return nbd_co_request(bs, &request, NULL);
1266 }
1267 
1268 static int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset,
1269                                   int bytes)
1270 {
1271     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1272     NBDRequest request = {
1273         .type = NBD_CMD_TRIM,
1274         .from = offset,
1275         .len = bytes,
1276     };
1277 
1278     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1279     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1280         return 0;
1281     }
1282 
1283     return nbd_co_request(bs, &request, NULL);
1284 }
1285 
1286 static int coroutine_fn nbd_client_co_block_status(
1287         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1288         int64_t *pnum, int64_t *map, BlockDriverState **file)
1289 {
1290     int ret, request_ret;
1291     NBDExtent extent = { 0 };
1292     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1293     Error *local_err = NULL;
1294 
1295     NBDRequest request = {
1296         .type = NBD_CMD_BLOCK_STATUS,
1297         .from = offset,
1298         .len = MIN(MIN_NON_ZERO(QEMU_ALIGN_DOWN(INT_MAX,
1299                                                 bs->bl.request_alignment),
1300                                 s->info.max_block),
1301                    MIN(bytes, s->info.size - offset)),
1302         .flags = NBD_CMD_FLAG_REQ_ONE,
1303     };
1304 
1305     if (!s->info.base_allocation) {
1306         *pnum = bytes;
1307         *map = offset;
1308         *file = bs;
1309         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1310     }
1311 
1312     /*
1313      * Work around the fact that the block layer doesn't do
1314      * byte-accurate sizing yet - if the status request exceeds the
1315      * server's advertised size because the block layer rounded size
1316      * up, we truncated the request to the server (above), or are
1317      * called on just the hole.
1318      */
1319     if (offset >= s->info.size) {
1320         *pnum = bytes;
1321         assert(bytes < BDRV_SECTOR_SIZE);
1322         /* Intentionally don't report offset_valid for the hole */
1323         return BDRV_BLOCK_ZERO;
1324     }
1325 
1326     if (s->info.min_block) {
1327         assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1328     }
1329     do {
1330         ret = nbd_co_send_request(bs, &request, NULL);
1331         if (ret < 0) {
1332             continue;
1333         }
1334 
1335         ret = nbd_co_receive_blockstatus_reply(s, request.handle, bytes,
1336                                                &extent, &request_ret,
1337                                                &local_err);
1338         if (local_err) {
1339             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1340                                       request.flags, request.type,
1341                                       nbd_cmd_lookup(request.type),
1342                                       ret, error_get_pretty(local_err));
1343             error_free(local_err);
1344             local_err = NULL;
1345         }
1346     } while (ret < 0 && nbd_client_connecting_wait(s));
1347 
1348     if (ret < 0 || request_ret < 0) {
1349         return ret ? ret : request_ret;
1350     }
1351 
1352     assert(extent.length);
1353     *pnum = extent.length;
1354     *map = offset;
1355     *file = bs;
1356     return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1357         (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1358         BDRV_BLOCK_OFFSET_VALID;
1359 }
1360 
1361 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1362                                      BlockReopenQueue *queue, Error **errp)
1363 {
1364     BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1365 
1366     if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1367         error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1368         return -EACCES;
1369     }
1370     return 0;
1371 }
1372 
1373 static void nbd_client_close(BlockDriverState *bs)
1374 {
1375     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1376     NBDRequest request = { .type = NBD_CMD_DISC };
1377 
1378     if (s->ioc) {
1379         nbd_send_request(s->ioc, &request);
1380     }
1381 
1382     nbd_teardown_connection(bs);
1383 }
1384 
1385 static QIOChannelSocket *nbd_establish_connection(SocketAddress *saddr,
1386                                                   Error **errp)
1387 {
1388     QIOChannelSocket *sioc;
1389     Error *local_err = NULL;
1390 
1391     sioc = qio_channel_socket_new();
1392     qio_channel_set_name(QIO_CHANNEL(sioc), "nbd-client");
1393 
1394     qio_channel_socket_connect_sync(sioc, saddr, &local_err);
1395     if (local_err) {
1396         object_unref(OBJECT(sioc));
1397         error_propagate(errp, local_err);
1398         return NULL;
1399     }
1400 
1401     qio_channel_set_delay(QIO_CHANNEL(sioc), false);
1402 
1403     return sioc;
1404 }
1405 
1406 static int nbd_client_connect(BlockDriverState *bs, Error **errp)
1407 {
1408     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1409     AioContext *aio_context = bdrv_get_aio_context(bs);
1410     int ret;
1411 
1412     /*
1413      * establish TCP connection, return error if it fails
1414      * TODO: Configurable retry-until-timeout behaviour.
1415      */
1416     QIOChannelSocket *sioc = nbd_establish_connection(s->saddr, errp);
1417 
1418     if (!sioc) {
1419         return -ECONNREFUSED;
1420     }
1421 
1422     /* NBD handshake */
1423     trace_nbd_client_connect(s->export);
1424     qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
1425     qio_channel_attach_aio_context(QIO_CHANNEL(sioc), aio_context);
1426 
1427     s->info.request_sizes = true;
1428     s->info.structured_reply = true;
1429     s->info.base_allocation = true;
1430     s->info.x_dirty_bitmap = g_strdup(s->x_dirty_bitmap);
1431     s->info.name = g_strdup(s->export ?: "");
1432     ret = nbd_receive_negotiate(aio_context, QIO_CHANNEL(sioc), s->tlscreds,
1433                                 s->hostname, &s->ioc, &s->info, errp);
1434     g_free(s->info.x_dirty_bitmap);
1435     g_free(s->info.name);
1436     if (ret < 0) {
1437         object_unref(OBJECT(sioc));
1438         return ret;
1439     }
1440     if (s->x_dirty_bitmap && !s->info.base_allocation) {
1441         error_setg(errp, "requested x-dirty-bitmap %s not found",
1442                    s->x_dirty_bitmap);
1443         ret = -EINVAL;
1444         goto fail;
1445     }
1446     if (s->info.flags & NBD_FLAG_READ_ONLY) {
1447         ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp);
1448         if (ret < 0) {
1449             goto fail;
1450         }
1451     }
1452     if (s->info.flags & NBD_FLAG_SEND_FUA) {
1453         bs->supported_write_flags = BDRV_REQ_FUA;
1454         bs->supported_zero_flags |= BDRV_REQ_FUA;
1455     }
1456     if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
1457         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
1458         if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) {
1459             bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK;
1460         }
1461     }
1462 
1463     s->sioc = sioc;
1464 
1465     if (!s->ioc) {
1466         s->ioc = QIO_CHANNEL(sioc);
1467         object_ref(OBJECT(s->ioc));
1468     }
1469 
1470     trace_nbd_client_connect_success(s->export);
1471 
1472     return 0;
1473 
1474  fail:
1475     /*
1476      * We have connected, but must fail for other reasons.
1477      * Send NBD_CMD_DISC as a courtesy to the server.
1478      */
1479     {
1480         NBDRequest request = { .type = NBD_CMD_DISC };
1481 
1482         nbd_send_request(s->ioc ?: QIO_CHANNEL(sioc), &request);
1483 
1484         object_unref(OBJECT(sioc));
1485 
1486         return ret;
1487     }
1488 }
1489 
1490 /*
1491  * Parse nbd_open options
1492  */
1493 
1494 static int nbd_parse_uri(const char *filename, QDict *options)
1495 {
1496     URI *uri;
1497     const char *p;
1498     QueryParams *qp = NULL;
1499     int ret = 0;
1500     bool is_unix;
1501 
1502     uri = uri_parse(filename);
1503     if (!uri) {
1504         return -EINVAL;
1505     }
1506 
1507     /* transport */
1508     if (!g_strcmp0(uri->scheme, "nbd")) {
1509         is_unix = false;
1510     } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1511         is_unix = false;
1512     } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1513         is_unix = true;
1514     } else {
1515         ret = -EINVAL;
1516         goto out;
1517     }
1518 
1519     p = uri->path ? uri->path : "/";
1520     p += strspn(p, "/");
1521     if (p[0]) {
1522         qdict_put_str(options, "export", p);
1523     }
1524 
1525     qp = query_params_parse(uri->query);
1526     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1527         ret = -EINVAL;
1528         goto out;
1529     }
1530 
1531     if (is_unix) {
1532         /* nbd+unix:///export?socket=path */
1533         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1534             ret = -EINVAL;
1535             goto out;
1536         }
1537         qdict_put_str(options, "server.type", "unix");
1538         qdict_put_str(options, "server.path", qp->p[0].value);
1539     } else {
1540         QString *host;
1541         char *port_str;
1542 
1543         /* nbd[+tcp]://host[:port]/export */
1544         if (!uri->server) {
1545             ret = -EINVAL;
1546             goto out;
1547         }
1548 
1549         /* strip braces from literal IPv6 address */
1550         if (uri->server[0] == '[') {
1551             host = qstring_from_substr(uri->server, 1,
1552                                        strlen(uri->server) - 1);
1553         } else {
1554             host = qstring_from_str(uri->server);
1555         }
1556 
1557         qdict_put_str(options, "server.type", "inet");
1558         qdict_put(options, "server.host", host);
1559 
1560         port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1561         qdict_put_str(options, "server.port", port_str);
1562         g_free(port_str);
1563     }
1564 
1565 out:
1566     if (qp) {
1567         query_params_free(qp);
1568     }
1569     uri_free(uri);
1570     return ret;
1571 }
1572 
1573 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1574 {
1575     const QDictEntry *e;
1576 
1577     for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1578         if (!strcmp(e->key, "host") ||
1579             !strcmp(e->key, "port") ||
1580             !strcmp(e->key, "path") ||
1581             !strcmp(e->key, "export") ||
1582             strstart(e->key, "server.", NULL))
1583         {
1584             error_setg(errp, "Option '%s' cannot be used with a file name",
1585                        e->key);
1586             return true;
1587         }
1588     }
1589 
1590     return false;
1591 }
1592 
1593 static void nbd_parse_filename(const char *filename, QDict *options,
1594                                Error **errp)
1595 {
1596     g_autofree char *file = NULL;
1597     char *export_name;
1598     const char *host_spec;
1599     const char *unixpath;
1600 
1601     if (nbd_has_filename_options_conflict(options, errp)) {
1602         return;
1603     }
1604 
1605     if (strstr(filename, "://")) {
1606         int ret = nbd_parse_uri(filename, options);
1607         if (ret < 0) {
1608             error_setg(errp, "No valid URL specified");
1609         }
1610         return;
1611     }
1612 
1613     file = g_strdup(filename);
1614 
1615     export_name = strstr(file, EN_OPTSTR);
1616     if (export_name) {
1617         if (export_name[strlen(EN_OPTSTR)] == 0) {
1618             return;
1619         }
1620         export_name[0] = 0; /* truncate 'file' */
1621         export_name += strlen(EN_OPTSTR);
1622 
1623         qdict_put_str(options, "export", export_name);
1624     }
1625 
1626     /* extract the host_spec - fail if it's not nbd:... */
1627     if (!strstart(file, "nbd:", &host_spec)) {
1628         error_setg(errp, "File name string for NBD must start with 'nbd:'");
1629         return;
1630     }
1631 
1632     if (!*host_spec) {
1633         return;
1634     }
1635 
1636     /* are we a UNIX or TCP socket? */
1637     if (strstart(host_spec, "unix:", &unixpath)) {
1638         qdict_put_str(options, "server.type", "unix");
1639         qdict_put_str(options, "server.path", unixpath);
1640     } else {
1641         InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1642 
1643         if (inet_parse(addr, host_spec, errp)) {
1644             goto out_inet;
1645         }
1646 
1647         qdict_put_str(options, "server.type", "inet");
1648         qdict_put_str(options, "server.host", addr->host);
1649         qdict_put_str(options, "server.port", addr->port);
1650     out_inet:
1651         qapi_free_InetSocketAddress(addr);
1652     }
1653 }
1654 
1655 static bool nbd_process_legacy_socket_options(QDict *output_options,
1656                                               QemuOpts *legacy_opts,
1657                                               Error **errp)
1658 {
1659     const char *path = qemu_opt_get(legacy_opts, "path");
1660     const char *host = qemu_opt_get(legacy_opts, "host");
1661     const char *port = qemu_opt_get(legacy_opts, "port");
1662     const QDictEntry *e;
1663 
1664     if (!path && !host && !port) {
1665         return true;
1666     }
1667 
1668     for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1669     {
1670         if (strstart(e->key, "server.", NULL)) {
1671             error_setg(errp, "Cannot use 'server' and path/host/port at the "
1672                        "same time");
1673             return false;
1674         }
1675     }
1676 
1677     if (path && host) {
1678         error_setg(errp, "path and host may not be used at the same time");
1679         return false;
1680     } else if (path) {
1681         if (port) {
1682             error_setg(errp, "port may not be used without host");
1683             return false;
1684         }
1685 
1686         qdict_put_str(output_options, "server.type", "unix");
1687         qdict_put_str(output_options, "server.path", path);
1688     } else if (host) {
1689         qdict_put_str(output_options, "server.type", "inet");
1690         qdict_put_str(output_options, "server.host", host);
1691         qdict_put_str(output_options, "server.port",
1692                       port ?: stringify(NBD_DEFAULT_PORT));
1693     }
1694 
1695     return true;
1696 }
1697 
1698 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1699                                  Error **errp)
1700 {
1701     SocketAddress *saddr = NULL;
1702     QDict *addr = NULL;
1703     Visitor *iv = NULL;
1704     Error *local_err = NULL;
1705 
1706     qdict_extract_subqdict(options, &addr, "server.");
1707     if (!qdict_size(addr)) {
1708         error_setg(errp, "NBD server address missing");
1709         goto done;
1710     }
1711 
1712     iv = qobject_input_visitor_new_flat_confused(addr, errp);
1713     if (!iv) {
1714         goto done;
1715     }
1716 
1717     visit_type_SocketAddress(iv, NULL, &saddr, &local_err);
1718     if (local_err) {
1719         error_propagate(errp, local_err);
1720         goto done;
1721     }
1722 
1723 done:
1724     qobject_unref(addr);
1725     visit_free(iv);
1726     return saddr;
1727 }
1728 
1729 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1730 {
1731     Object *obj;
1732     QCryptoTLSCreds *creds;
1733 
1734     obj = object_resolve_path_component(
1735         object_get_objects_root(), id);
1736     if (!obj) {
1737         error_setg(errp, "No TLS credentials with id '%s'",
1738                    id);
1739         return NULL;
1740     }
1741     creds = (QCryptoTLSCreds *)
1742         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1743     if (!creds) {
1744         error_setg(errp, "Object with id '%s' is not TLS credentials",
1745                    id);
1746         return NULL;
1747     }
1748 
1749     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
1750         error_setg(errp,
1751                    "Expecting TLS credentials with a client endpoint");
1752         return NULL;
1753     }
1754     object_ref(obj);
1755     return creds;
1756 }
1757 
1758 
1759 static QemuOptsList nbd_runtime_opts = {
1760     .name = "nbd",
1761     .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1762     .desc = {
1763         {
1764             .name = "host",
1765             .type = QEMU_OPT_STRING,
1766             .help = "TCP host to connect to",
1767         },
1768         {
1769             .name = "port",
1770             .type = QEMU_OPT_STRING,
1771             .help = "TCP port to connect to",
1772         },
1773         {
1774             .name = "path",
1775             .type = QEMU_OPT_STRING,
1776             .help = "Unix socket path to connect to",
1777         },
1778         {
1779             .name = "export",
1780             .type = QEMU_OPT_STRING,
1781             .help = "Name of the NBD export to open",
1782         },
1783         {
1784             .name = "tls-creds",
1785             .type = QEMU_OPT_STRING,
1786             .help = "ID of the TLS credentials to use",
1787         },
1788         {
1789             .name = "x-dirty-bitmap",
1790             .type = QEMU_OPT_STRING,
1791             .help = "experimental: expose named dirty bitmap in place of "
1792                     "block status",
1793         },
1794         {
1795             .name = "reconnect-delay",
1796             .type = QEMU_OPT_NUMBER,
1797             .help = "On an unexpected disconnect, the nbd client tries to "
1798                     "connect again until succeeding or encountering a serious "
1799                     "error.  During the first @reconnect-delay seconds, all "
1800                     "requests are paused and will be rerun on a successful "
1801                     "reconnect. After that time, any delayed requests and all "
1802                     "future requests before a successful reconnect will "
1803                     "immediately fail. Default 0",
1804         },
1805         { /* end of list */ }
1806     },
1807 };
1808 
1809 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1810                                Error **errp)
1811 {
1812     BDRVNBDState *s = bs->opaque;
1813     QemuOpts *opts;
1814     Error *local_err = NULL;
1815     int ret = -EINVAL;
1816 
1817     opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1818     qemu_opts_absorb_qdict(opts, options, &local_err);
1819     if (local_err) {
1820         error_propagate(errp, local_err);
1821         goto error;
1822     }
1823 
1824     /* Translate @host, @port, and @path to a SocketAddress */
1825     if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1826         goto error;
1827     }
1828 
1829     /* Pop the config into our state object. Exit if invalid. */
1830     s->saddr = nbd_config(s, options, errp);
1831     if (!s->saddr) {
1832         goto error;
1833     }
1834 
1835     s->export = g_strdup(qemu_opt_get(opts, "export"));
1836     if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1837         error_setg(errp, "export name too long to send to server");
1838         goto error;
1839     }
1840 
1841     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1842     if (s->tlscredsid) {
1843         s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1844         if (!s->tlscreds) {
1845             goto error;
1846         }
1847 
1848         /* TODO SOCKET_ADDRESS_KIND_FD where fd has AF_INET or AF_INET6 */
1849         if (s->saddr->type != SOCKET_ADDRESS_TYPE_INET) {
1850             error_setg(errp, "TLS only supported over IP sockets");
1851             goto error;
1852         }
1853         s->hostname = s->saddr->u.inet.host;
1854     }
1855 
1856     s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1857     if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1858         error_setg(errp, "x-dirty-bitmap query too long to send to server");
1859         goto error;
1860     }
1861 
1862     s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1863 
1864     ret = 0;
1865 
1866  error:
1867     if (ret < 0) {
1868         object_unref(OBJECT(s->tlscreds));
1869         qapi_free_SocketAddress(s->saddr);
1870         g_free(s->export);
1871         g_free(s->tlscredsid);
1872         g_free(s->x_dirty_bitmap);
1873     }
1874     qemu_opts_del(opts);
1875     return ret;
1876 }
1877 
1878 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1879                     Error **errp)
1880 {
1881     int ret;
1882     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1883 
1884     ret = nbd_process_options(bs, options, errp);
1885     if (ret < 0) {
1886         return ret;
1887     }
1888 
1889     s->bs = bs;
1890     qemu_co_mutex_init(&s->send_mutex);
1891     qemu_co_queue_init(&s->free_sema);
1892 
1893     ret = nbd_client_connect(bs, errp);
1894     if (ret < 0) {
1895         return ret;
1896     }
1897     /* successfully connected */
1898     s->state = NBD_CLIENT_CONNECTED;
1899 
1900     s->connection_co = qemu_coroutine_create(nbd_connection_entry, s);
1901     bdrv_inc_in_flight(bs);
1902     aio_co_schedule(bdrv_get_aio_context(bs), s->connection_co);
1903 
1904     return 0;
1905 }
1906 
1907 static int nbd_co_flush(BlockDriverState *bs)
1908 {
1909     return nbd_client_co_flush(bs);
1910 }
1911 
1912 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1913 {
1914     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1915     uint32_t min = s->info.min_block;
1916     uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1917 
1918     /*
1919      * If the server did not advertise an alignment:
1920      * - a size that is not sector-aligned implies that an alignment
1921      *   of 1 can be used to access those tail bytes
1922      * - advertisement of block status requires an alignment of 1, so
1923      *   that we don't violate block layer constraints that block
1924      *   status is always aligned (as we can't control whether the
1925      *   server will report sub-sector extents, such as a hole at EOF
1926      *   on an unaligned POSIX file)
1927      * - otherwise, assume the server is so old that we are safer avoiding
1928      *   sub-sector requests
1929      */
1930     if (!min) {
1931         min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1932                s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1933     }
1934 
1935     bs->bl.request_alignment = min;
1936     bs->bl.max_pdiscard = max;
1937     bs->bl.max_pwrite_zeroes = max;
1938     bs->bl.max_transfer = max;
1939 
1940     if (s->info.opt_block &&
1941         s->info.opt_block > bs->bl.opt_transfer) {
1942         bs->bl.opt_transfer = s->info.opt_block;
1943     }
1944 }
1945 
1946 static void nbd_close(BlockDriverState *bs)
1947 {
1948     BDRVNBDState *s = bs->opaque;
1949 
1950     nbd_client_close(bs);
1951 
1952     object_unref(OBJECT(s->tlscreds));
1953     qapi_free_SocketAddress(s->saddr);
1954     g_free(s->export);
1955     g_free(s->tlscredsid);
1956     g_free(s->x_dirty_bitmap);
1957 }
1958 
1959 static int64_t nbd_getlength(BlockDriverState *bs)
1960 {
1961     BDRVNBDState *s = bs->opaque;
1962 
1963     return s->info.size;
1964 }
1965 
1966 static void nbd_refresh_filename(BlockDriverState *bs)
1967 {
1968     BDRVNBDState *s = bs->opaque;
1969     const char *host = NULL, *port = NULL, *path = NULL;
1970 
1971     if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1972         const InetSocketAddress *inet = &s->saddr->u.inet;
1973         if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
1974             host = inet->host;
1975             port = inet->port;
1976         }
1977     } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
1978         path = s->saddr->u.q_unix.path;
1979     } /* else can't represent as pseudo-filename */
1980 
1981     if (path && s->export) {
1982         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1983                  "nbd+unix:///%s?socket=%s", s->export, path);
1984     } else if (path && !s->export) {
1985         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1986                  "nbd+unix://?socket=%s", path);
1987     } else if (host && s->export) {
1988         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1989                  "nbd://%s:%s/%s", host, port, s->export);
1990     } else if (host && !s->export) {
1991         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1992                  "nbd://%s:%s", host, port);
1993     }
1994 }
1995 
1996 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
1997 {
1998     /* The generic bdrv_dirname() implementation is able to work out some
1999      * directory name for NBD nodes, but that would be wrong. So far there is no
2000      * specification for how "export paths" would work, so NBD does not have
2001      * directory names. */
2002     error_setg(errp, "Cannot generate a base directory for NBD nodes");
2003     return NULL;
2004 }
2005 
2006 static const char *const nbd_strong_runtime_opts[] = {
2007     "path",
2008     "host",
2009     "port",
2010     "export",
2011     "tls-creds",
2012     "server.",
2013 
2014     NULL
2015 };
2016 
2017 static BlockDriver bdrv_nbd = {
2018     .format_name                = "nbd",
2019     .protocol_name              = "nbd",
2020     .instance_size              = sizeof(BDRVNBDState),
2021     .bdrv_parse_filename        = nbd_parse_filename,
2022     .bdrv_file_open             = nbd_open,
2023     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2024     .bdrv_co_preadv             = nbd_client_co_preadv,
2025     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2026     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2027     .bdrv_close                 = nbd_close,
2028     .bdrv_co_flush_to_os        = nbd_co_flush,
2029     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2030     .bdrv_refresh_limits        = nbd_refresh_limits,
2031     .bdrv_getlength             = nbd_getlength,
2032     .bdrv_detach_aio_context    = nbd_client_detach_aio_context,
2033     .bdrv_attach_aio_context    = nbd_client_attach_aio_context,
2034     .bdrv_co_drain_begin        = nbd_client_co_drain_begin,
2035     .bdrv_co_drain_end          = nbd_client_co_drain_end,
2036     .bdrv_refresh_filename      = nbd_refresh_filename,
2037     .bdrv_co_block_status       = nbd_client_co_block_status,
2038     .bdrv_dirname               = nbd_dirname,
2039     .strong_runtime_opts        = nbd_strong_runtime_opts,
2040 };
2041 
2042 static BlockDriver bdrv_nbd_tcp = {
2043     .format_name                = "nbd",
2044     .protocol_name              = "nbd+tcp",
2045     .instance_size              = sizeof(BDRVNBDState),
2046     .bdrv_parse_filename        = nbd_parse_filename,
2047     .bdrv_file_open             = nbd_open,
2048     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2049     .bdrv_co_preadv             = nbd_client_co_preadv,
2050     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2051     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2052     .bdrv_close                 = nbd_close,
2053     .bdrv_co_flush_to_os        = nbd_co_flush,
2054     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2055     .bdrv_refresh_limits        = nbd_refresh_limits,
2056     .bdrv_getlength             = nbd_getlength,
2057     .bdrv_detach_aio_context    = nbd_client_detach_aio_context,
2058     .bdrv_attach_aio_context    = nbd_client_attach_aio_context,
2059     .bdrv_co_drain_begin        = nbd_client_co_drain_begin,
2060     .bdrv_co_drain_end          = nbd_client_co_drain_end,
2061     .bdrv_refresh_filename      = nbd_refresh_filename,
2062     .bdrv_co_block_status       = nbd_client_co_block_status,
2063     .bdrv_dirname               = nbd_dirname,
2064     .strong_runtime_opts        = nbd_strong_runtime_opts,
2065 };
2066 
2067 static BlockDriver bdrv_nbd_unix = {
2068     .format_name                = "nbd",
2069     .protocol_name              = "nbd+unix",
2070     .instance_size              = sizeof(BDRVNBDState),
2071     .bdrv_parse_filename        = nbd_parse_filename,
2072     .bdrv_file_open             = nbd_open,
2073     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2074     .bdrv_co_preadv             = nbd_client_co_preadv,
2075     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2076     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2077     .bdrv_close                 = nbd_close,
2078     .bdrv_co_flush_to_os        = nbd_co_flush,
2079     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2080     .bdrv_refresh_limits        = nbd_refresh_limits,
2081     .bdrv_getlength             = nbd_getlength,
2082     .bdrv_detach_aio_context    = nbd_client_detach_aio_context,
2083     .bdrv_attach_aio_context    = nbd_client_attach_aio_context,
2084     .bdrv_co_drain_begin        = nbd_client_co_drain_begin,
2085     .bdrv_co_drain_end          = nbd_client_co_drain_end,
2086     .bdrv_refresh_filename      = nbd_refresh_filename,
2087     .bdrv_co_block_status       = nbd_client_co_block_status,
2088     .bdrv_dirname               = nbd_dirname,
2089     .strong_runtime_opts        = nbd_strong_runtime_opts,
2090 };
2091 
2092 static void bdrv_nbd_init(void)
2093 {
2094     bdrv_register(&bdrv_nbd);
2095     bdrv_register(&bdrv_nbd_tcp);
2096     bdrv_register(&bdrv_nbd_unix);
2097 }
2098 
2099 block_init(bdrv_nbd_init);
2100