xref: /openbmc/qemu/block/nbd.c (revision 5242ef88)
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 #include "qemu/atomic.h"
39 
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/clone-visitor.h"
43 
44 #include "block/qdict.h"
45 #include "block/nbd.h"
46 #include "block/block_int.h"
47 #include "block/coroutines.h"
48 
49 #include "qemu/yank.h"
50 
51 #define EN_OPTSTR ":exportname="
52 #define MAX_NBD_REQUESTS    16
53 
54 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
55 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ (uint64_t)(intptr_t)(bs))
56 
57 typedef struct {
58     Coroutine *coroutine;
59     uint64_t offset;        /* original offset of the request */
60     bool receiving;         /* sleeping in the yield in nbd_receive_replies */
61     bool reply_possible;    /* reply header not yet received */
62 } NBDClientRequest;
63 
64 typedef enum NBDClientState {
65     NBD_CLIENT_CONNECTING_WAIT,
66     NBD_CLIENT_CONNECTING_NOWAIT,
67     NBD_CLIENT_CONNECTED,
68     NBD_CLIENT_QUIT
69 } NBDClientState;
70 
71 typedef struct BDRVNBDState {
72     QIOChannel *ioc; /* The current I/O channel */
73     NBDExportInfo info;
74 
75     CoMutex send_mutex;
76     CoQueue free_sema;
77 
78     CoMutex receive_mutex;
79     int in_flight;
80     NBDClientState state;
81 
82     QEMUTimer *reconnect_delay_timer;
83     QEMUTimer *open_timer;
84 
85     NBDClientRequest requests[MAX_NBD_REQUESTS];
86     NBDReply reply;
87     BlockDriverState *bs;
88 
89     /* Connection parameters */
90     uint32_t reconnect_delay;
91     uint32_t open_timeout;
92     SocketAddress *saddr;
93     char *export;
94     char *tlscredsid;
95     QCryptoTLSCreds *tlscreds;
96     char *tlshostname;
97     char *x_dirty_bitmap;
98     bool alloc_depth;
99 
100     NBDClientConnection *conn;
101 } BDRVNBDState;
102 
103 static void nbd_yank(void *opaque);
104 
105 static void nbd_clear_bdrvstate(BlockDriverState *bs)
106 {
107     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
108 
109     nbd_client_connection_release(s->conn);
110     s->conn = NULL;
111 
112     yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name));
113 
114     /* Must not leave timers behind that would access freed data */
115     assert(!s->reconnect_delay_timer);
116     assert(!s->open_timer);
117 
118     object_unref(OBJECT(s->tlscreds));
119     qapi_free_SocketAddress(s->saddr);
120     s->saddr = NULL;
121     g_free(s->export);
122     s->export = NULL;
123     g_free(s->tlscredsid);
124     s->tlscredsid = NULL;
125     g_free(s->tlshostname);
126     s->tlshostname = NULL;
127     g_free(s->x_dirty_bitmap);
128     s->x_dirty_bitmap = NULL;
129 }
130 
131 static bool nbd_client_connected(BDRVNBDState *s)
132 {
133     return qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED;
134 }
135 
136 static bool nbd_recv_coroutine_wake_one(NBDClientRequest *req)
137 {
138     if (req->receiving) {
139         req->receiving = false;
140         aio_co_wake(req->coroutine);
141         return true;
142     }
143 
144     return false;
145 }
146 
147 static void nbd_recv_coroutines_wake(BDRVNBDState *s, bool all)
148 {
149     int i;
150 
151     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
152         if (nbd_recv_coroutine_wake_one(&s->requests[i]) && !all) {
153             return;
154         }
155     }
156 }
157 
158 static void nbd_channel_error(BDRVNBDState *s, int ret)
159 {
160     if (nbd_client_connected(s)) {
161         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
162     }
163 
164     if (ret == -EIO) {
165         if (nbd_client_connected(s)) {
166             s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT :
167                                             NBD_CLIENT_CONNECTING_NOWAIT;
168         }
169     } else {
170         s->state = NBD_CLIENT_QUIT;
171     }
172 
173     nbd_recv_coroutines_wake(s, true);
174 }
175 
176 static void reconnect_delay_timer_del(BDRVNBDState *s)
177 {
178     if (s->reconnect_delay_timer) {
179         timer_free(s->reconnect_delay_timer);
180         s->reconnect_delay_timer = NULL;
181     }
182 }
183 
184 static void reconnect_delay_timer_cb(void *opaque)
185 {
186     BDRVNBDState *s = opaque;
187 
188     if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT) {
189         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
190         nbd_co_establish_connection_cancel(s->conn);
191         while (qemu_co_enter_next(&s->free_sema, NULL)) {
192             /* Resume all queued requests */
193         }
194     }
195 
196     reconnect_delay_timer_del(s);
197 }
198 
199 static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
200 {
201     if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTING_WAIT) {
202         return;
203     }
204 
205     assert(!s->reconnect_delay_timer);
206     s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
207                                              QEMU_CLOCK_REALTIME,
208                                              SCALE_NS,
209                                              reconnect_delay_timer_cb, s);
210     timer_mod(s->reconnect_delay_timer, expire_time_ns);
211 }
212 
213 static void nbd_teardown_connection(BlockDriverState *bs)
214 {
215     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
216 
217     assert(!s->in_flight);
218 
219     if (s->ioc) {
220         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
221         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
222                                  nbd_yank, s->bs);
223         object_unref(OBJECT(s->ioc));
224         s->ioc = NULL;
225     }
226 
227     s->state = NBD_CLIENT_QUIT;
228 }
229 
230 static void open_timer_del(BDRVNBDState *s)
231 {
232     if (s->open_timer) {
233         timer_free(s->open_timer);
234         s->open_timer = NULL;
235     }
236 }
237 
238 static void open_timer_cb(void *opaque)
239 {
240     BDRVNBDState *s = opaque;
241 
242     nbd_co_establish_connection_cancel(s->conn);
243     open_timer_del(s);
244 }
245 
246 static void open_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
247 {
248     assert(!s->open_timer);
249     s->open_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
250                                   QEMU_CLOCK_REALTIME,
251                                   SCALE_NS,
252                                   open_timer_cb, s);
253     timer_mod(s->open_timer, expire_time_ns);
254 }
255 
256 static bool nbd_client_connecting(BDRVNBDState *s)
257 {
258     NBDClientState state = qatomic_load_acquire(&s->state);
259     return state == NBD_CLIENT_CONNECTING_WAIT ||
260         state == NBD_CLIENT_CONNECTING_NOWAIT;
261 }
262 
263 static bool nbd_client_connecting_wait(BDRVNBDState *s)
264 {
265     return qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT;
266 }
267 
268 /*
269  * Update @bs with information learned during a completed negotiation process.
270  * Return failure if the server's advertised options are incompatible with the
271  * client's needs.
272  */
273 static int nbd_handle_updated_info(BlockDriverState *bs, Error **errp)
274 {
275     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
276     int ret;
277 
278     if (s->x_dirty_bitmap) {
279         if (!s->info.base_allocation) {
280             error_setg(errp, "requested x-dirty-bitmap %s not found",
281                        s->x_dirty_bitmap);
282             return -EINVAL;
283         }
284         if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) {
285             s->alloc_depth = true;
286         }
287     }
288 
289     if (s->info.flags & NBD_FLAG_READ_ONLY) {
290         ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp);
291         if (ret < 0) {
292             return ret;
293         }
294     }
295 
296     if (s->info.flags & NBD_FLAG_SEND_FUA) {
297         bs->supported_write_flags = BDRV_REQ_FUA;
298         bs->supported_zero_flags |= BDRV_REQ_FUA;
299     }
300 
301     if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
302         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
303         if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) {
304             bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK;
305         }
306     }
307 
308     trace_nbd_client_handshake_success(s->export);
309 
310     return 0;
311 }
312 
313 int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs,
314                                                 Error **errp)
315 {
316     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
317     int ret;
318     bool blocking = nbd_client_connecting_wait(s);
319     IO_CODE();
320 
321     assert(!s->ioc);
322 
323     s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp);
324     if (!s->ioc) {
325         return -ECONNREFUSED;
326     }
327 
328     yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank,
329                            bs);
330 
331     ret = nbd_handle_updated_info(s->bs, NULL);
332     if (ret < 0) {
333         /*
334          * We have connected, but must fail for other reasons.
335          * Send NBD_CMD_DISC as a courtesy to the server.
336          */
337         NBDRequest request = { .type = NBD_CMD_DISC };
338 
339         nbd_send_request(s->ioc, &request);
340 
341         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
342                                  nbd_yank, bs);
343         object_unref(OBJECT(s->ioc));
344         s->ioc = NULL;
345 
346         return ret;
347     }
348 
349     qio_channel_set_blocking(s->ioc, false, NULL);
350     qio_channel_attach_aio_context(s->ioc, bdrv_get_aio_context(bs));
351 
352     /* successfully connected */
353     s->state = NBD_CLIENT_CONNECTED;
354     qemu_co_queue_restart_all(&s->free_sema);
355 
356     return 0;
357 }
358 
359 /* called under s->send_mutex */
360 static coroutine_fn void nbd_reconnect_attempt(BDRVNBDState *s)
361 {
362     assert(nbd_client_connecting(s));
363     assert(s->in_flight == 0);
364 
365     if (nbd_client_connecting_wait(s) && s->reconnect_delay &&
366         !s->reconnect_delay_timer)
367     {
368         /*
369          * It's first reconnect attempt after switching to
370          * NBD_CLIENT_CONNECTING_WAIT
371          */
372         reconnect_delay_timer_init(s,
373             qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
374             s->reconnect_delay * NANOSECONDS_PER_SECOND);
375     }
376 
377     /*
378      * Now we are sure that nobody is accessing the channel, and no one will
379      * try until we set the state to CONNECTED.
380      */
381 
382     /* Finalize previous connection if any */
383     if (s->ioc) {
384         qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc));
385         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
386                                  nbd_yank, s->bs);
387         object_unref(OBJECT(s->ioc));
388         s->ioc = NULL;
389     }
390 
391     nbd_co_do_establish_connection(s->bs, NULL);
392 
393     /*
394      * The reconnect attempt is done (maybe successfully, maybe not), so
395      * we no longer need this timer.  Delete it so it will not outlive
396      * this I/O request (so draining removes all timers).
397      */
398     reconnect_delay_timer_del(s);
399 }
400 
401 static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t handle)
402 {
403     int ret;
404     uint64_t ind = HANDLE_TO_INDEX(s, handle), ind2;
405     QEMU_LOCK_GUARD(&s->receive_mutex);
406 
407     while (true) {
408         if (s->reply.handle == handle) {
409             /* We are done */
410             return 0;
411         }
412 
413         if (!nbd_client_connected(s)) {
414             return -EIO;
415         }
416 
417         if (s->reply.handle != 0) {
418             /*
419              * Some other request is being handled now. It should already be
420              * woken by whoever set s->reply.handle (or never wait in this
421              * yield). So, we should not wake it here.
422              */
423             ind2 = HANDLE_TO_INDEX(s, s->reply.handle);
424             assert(!s->requests[ind2].receiving);
425 
426             s->requests[ind].receiving = true;
427             qemu_co_mutex_unlock(&s->receive_mutex);
428 
429             qemu_coroutine_yield();
430             /*
431              * We may be woken for 3 reasons:
432              * 1. From this function, executing in parallel coroutine, when our
433              *    handle is received.
434              * 2. From nbd_channel_error(), when connection is lost.
435              * 3. From nbd_co_receive_one_chunk(), when previous request is
436              *    finished and s->reply.handle set to 0.
437              * Anyway, it's OK to lock the mutex and go to the next iteration.
438              */
439 
440             qemu_co_mutex_lock(&s->receive_mutex);
441             assert(!s->requests[ind].receiving);
442             continue;
443         }
444 
445         /* We are under mutex and handle is 0. We have to do the dirty work. */
446         assert(s->reply.handle == 0);
447         ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, NULL);
448         if (ret <= 0) {
449             ret = ret ? ret : -EIO;
450             nbd_channel_error(s, ret);
451             return ret;
452         }
453         if (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply) {
454             nbd_channel_error(s, -EINVAL);
455             return -EINVAL;
456         }
457         if (s->reply.handle == handle) {
458             /* We are done */
459             return 0;
460         }
461         ind2 = HANDLE_TO_INDEX(s, s->reply.handle);
462         if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].reply_possible) {
463             nbd_channel_error(s, -EINVAL);
464             return -EINVAL;
465         }
466         nbd_recv_coroutine_wake_one(&s->requests[ind2]);
467     }
468 }
469 
470 static int nbd_co_send_request(BlockDriverState *bs,
471                                NBDRequest *request,
472                                QEMUIOVector *qiov)
473 {
474     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
475     int rc, i = -1;
476 
477     qemu_co_mutex_lock(&s->send_mutex);
478 
479     while (s->in_flight == MAX_NBD_REQUESTS ||
480            (!nbd_client_connected(s) && s->in_flight > 0))
481     {
482         qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
483     }
484 
485     if (nbd_client_connecting(s)) {
486         nbd_reconnect_attempt(s);
487     }
488 
489     if (!nbd_client_connected(s)) {
490         rc = -EIO;
491         goto err;
492     }
493 
494     s->in_flight++;
495 
496     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
497         if (s->requests[i].coroutine == NULL) {
498             break;
499         }
500     }
501 
502     g_assert(qemu_in_coroutine());
503     assert(i < MAX_NBD_REQUESTS);
504 
505     s->requests[i].coroutine = qemu_coroutine_self();
506     s->requests[i].offset = request->from;
507     s->requests[i].receiving = false;
508     s->requests[i].reply_possible = true;
509 
510     request->handle = INDEX_TO_HANDLE(s, i);
511 
512     assert(s->ioc);
513 
514     if (qiov) {
515         qio_channel_set_cork(s->ioc, true);
516         rc = nbd_send_request(s->ioc, request);
517         if (nbd_client_connected(s) && rc >= 0) {
518             if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
519                                        NULL) < 0) {
520                 rc = -EIO;
521             }
522         } else if (rc >= 0) {
523             rc = -EIO;
524         }
525         qio_channel_set_cork(s->ioc, false);
526     } else {
527         rc = nbd_send_request(s->ioc, request);
528     }
529 
530 err:
531     if (rc < 0) {
532         nbd_channel_error(s, rc);
533         if (i != -1) {
534             s->requests[i].coroutine = NULL;
535             s->in_flight--;
536         }
537         qemu_co_queue_next(&s->free_sema);
538     }
539     qemu_co_mutex_unlock(&s->send_mutex);
540     return rc;
541 }
542 
543 static inline uint16_t payload_advance16(uint8_t **payload)
544 {
545     *payload += 2;
546     return lduw_be_p(*payload - 2);
547 }
548 
549 static inline uint32_t payload_advance32(uint8_t **payload)
550 {
551     *payload += 4;
552     return ldl_be_p(*payload - 4);
553 }
554 
555 static inline uint64_t payload_advance64(uint8_t **payload)
556 {
557     *payload += 8;
558     return ldq_be_p(*payload - 8);
559 }
560 
561 static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
562                                          NBDStructuredReplyChunk *chunk,
563                                          uint8_t *payload, uint64_t orig_offset,
564                                          QEMUIOVector *qiov, Error **errp)
565 {
566     uint64_t offset;
567     uint32_t hole_size;
568 
569     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
570         error_setg(errp, "Protocol error: invalid payload for "
571                          "NBD_REPLY_TYPE_OFFSET_HOLE");
572         return -EINVAL;
573     }
574 
575     offset = payload_advance64(&payload);
576     hole_size = payload_advance32(&payload);
577 
578     if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
579         offset > orig_offset + qiov->size - hole_size) {
580         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
581                          " region");
582         return -EINVAL;
583     }
584     if (s->info.min_block &&
585         !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
586         trace_nbd_structured_read_compliance("hole");
587     }
588 
589     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
590 
591     return 0;
592 }
593 
594 /*
595  * nbd_parse_blockstatus_payload
596  * Based on our request, we expect only one extent in reply, for the
597  * base:allocation context.
598  */
599 static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
600                                          NBDStructuredReplyChunk *chunk,
601                                          uint8_t *payload, uint64_t orig_length,
602                                          NBDExtent *extent, Error **errp)
603 {
604     uint32_t context_id;
605 
606     /* The server succeeded, so it must have sent [at least] one extent */
607     if (chunk->length < sizeof(context_id) + sizeof(*extent)) {
608         error_setg(errp, "Protocol error: invalid payload for "
609                          "NBD_REPLY_TYPE_BLOCK_STATUS");
610         return -EINVAL;
611     }
612 
613     context_id = payload_advance32(&payload);
614     if (s->info.context_id != context_id) {
615         error_setg(errp, "Protocol error: unexpected context id %d for "
616                          "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
617                          "id is %d", context_id,
618                          s->info.context_id);
619         return -EINVAL;
620     }
621 
622     extent->length = payload_advance32(&payload);
623     extent->flags = payload_advance32(&payload);
624 
625     if (extent->length == 0) {
626         error_setg(errp, "Protocol error: server sent status chunk with "
627                    "zero length");
628         return -EINVAL;
629     }
630 
631     /*
632      * A server sending unaligned block status is in violation of the
633      * protocol, but as qemu-nbd 3.1 is such a server (at least for
634      * POSIX files that are not a multiple of 512 bytes, since qemu
635      * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
636      * still sees an implicit hole beyond the real EOF), it's nicer to
637      * work around the misbehaving server. If the request included
638      * more than the final unaligned block, truncate it back to an
639      * aligned result; if the request was only the final block, round
640      * up to the full block and change the status to fully-allocated
641      * (always a safe status, even if it loses information).
642      */
643     if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
644                                                    s->info.min_block)) {
645         trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
646         if (extent->length > s->info.min_block) {
647             extent->length = QEMU_ALIGN_DOWN(extent->length,
648                                              s->info.min_block);
649         } else {
650             extent->length = s->info.min_block;
651             extent->flags = 0;
652         }
653     }
654 
655     /*
656      * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
657      * sent us any more than one extent, nor should it have included
658      * status beyond our request in that extent. However, it's easy
659      * enough to ignore the server's noncompliance without killing the
660      * connection; just ignore trailing extents, and clamp things to
661      * the length of our request.
662      */
663     if (chunk->length > sizeof(context_id) + sizeof(*extent)) {
664         trace_nbd_parse_blockstatus_compliance("more than one extent");
665     }
666     if (extent->length > orig_length) {
667         extent->length = orig_length;
668         trace_nbd_parse_blockstatus_compliance("extent length too large");
669     }
670 
671     /*
672      * HACK: if we are using x-dirty-bitmaps to access
673      * qemu:allocation-depth, treat all depths > 2 the same as 2,
674      * since nbd_client_co_block_status is only expecting the low two
675      * bits to be set.
676      */
677     if (s->alloc_depth && extent->flags > 2) {
678         extent->flags = 2;
679     }
680 
681     return 0;
682 }
683 
684 /*
685  * nbd_parse_error_payload
686  * on success @errp contains message describing nbd error reply
687  */
688 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
689                                    uint8_t *payload, int *request_ret,
690                                    Error **errp)
691 {
692     uint32_t error;
693     uint16_t message_size;
694 
695     assert(chunk->type & (1 << 15));
696 
697     if (chunk->length < sizeof(error) + sizeof(message_size)) {
698         error_setg(errp,
699                    "Protocol error: invalid payload for structured error");
700         return -EINVAL;
701     }
702 
703     error = nbd_errno_to_system_errno(payload_advance32(&payload));
704     if (error == 0) {
705         error_setg(errp, "Protocol error: server sent structured error chunk "
706                          "with error = 0");
707         return -EINVAL;
708     }
709 
710     *request_ret = -error;
711     message_size = payload_advance16(&payload);
712 
713     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
714         error_setg(errp, "Protocol error: server sent structured error chunk "
715                          "with incorrect message size");
716         return -EINVAL;
717     }
718 
719     /* TODO: Add a trace point to mention the server complaint */
720 
721     /* TODO handle ERROR_OFFSET */
722 
723     return 0;
724 }
725 
726 static int nbd_co_receive_offset_data_payload(BDRVNBDState *s,
727                                               uint64_t orig_offset,
728                                               QEMUIOVector *qiov, Error **errp)
729 {
730     QEMUIOVector sub_qiov;
731     uint64_t offset;
732     size_t data_size;
733     int ret;
734     NBDStructuredReplyChunk *chunk = &s->reply.structured;
735 
736     assert(nbd_reply_is_structured(&s->reply));
737 
738     /* The NBD spec requires at least one byte of payload */
739     if (chunk->length <= sizeof(offset)) {
740         error_setg(errp, "Protocol error: invalid payload for "
741                          "NBD_REPLY_TYPE_OFFSET_DATA");
742         return -EINVAL;
743     }
744 
745     if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
746         return -EIO;
747     }
748 
749     data_size = chunk->length - sizeof(offset);
750     assert(data_size);
751     if (offset < orig_offset || data_size > qiov->size ||
752         offset > orig_offset + qiov->size - data_size) {
753         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
754                          " region");
755         return -EINVAL;
756     }
757     if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
758         trace_nbd_structured_read_compliance("data");
759     }
760 
761     qemu_iovec_init(&sub_qiov, qiov->niov);
762     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
763     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
764     qemu_iovec_destroy(&sub_qiov);
765 
766     return ret < 0 ? -EIO : 0;
767 }
768 
769 #define NBD_MAX_MALLOC_PAYLOAD 1000
770 static coroutine_fn int nbd_co_receive_structured_payload(
771         BDRVNBDState *s, void **payload, Error **errp)
772 {
773     int ret;
774     uint32_t len;
775 
776     assert(nbd_reply_is_structured(&s->reply));
777 
778     len = s->reply.structured.length;
779 
780     if (len == 0) {
781         return 0;
782     }
783 
784     if (payload == NULL) {
785         error_setg(errp, "Unexpected structured payload");
786         return -EINVAL;
787     }
788 
789     if (len > NBD_MAX_MALLOC_PAYLOAD) {
790         error_setg(errp, "Payload too large");
791         return -EINVAL;
792     }
793 
794     *payload = g_new(char, len);
795     ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
796     if (ret < 0) {
797         g_free(*payload);
798         *payload = NULL;
799         return ret;
800     }
801 
802     return 0;
803 }
804 
805 /*
806  * nbd_co_do_receive_one_chunk
807  * for simple reply:
808  *   set request_ret to received reply error
809  *   if qiov is not NULL: read payload to @qiov
810  * for structured reply chunk:
811  *   if error chunk: read payload, set @request_ret, do not set @payload
812  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
813  *   else: read payload to @payload
814  *
815  * If function fails, @errp contains corresponding error message, and the
816  * connection with the server is suspect.  If it returns 0, then the
817  * transaction succeeded (although @request_ret may be a negative errno
818  * corresponding to the server's error reply), and errp is unchanged.
819  */
820 static coroutine_fn int nbd_co_do_receive_one_chunk(
821         BDRVNBDState *s, uint64_t handle, bool only_structured,
822         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
823 {
824     int ret;
825     int i = HANDLE_TO_INDEX(s, handle);
826     void *local_payload = NULL;
827     NBDStructuredReplyChunk *chunk;
828 
829     if (payload) {
830         *payload = NULL;
831     }
832     *request_ret = 0;
833 
834     nbd_receive_replies(s, handle);
835     if (!nbd_client_connected(s)) {
836         error_setg(errp, "Connection closed");
837         return -EIO;
838     }
839     assert(s->ioc);
840 
841     assert(s->reply.handle == handle);
842 
843     if (nbd_reply_is_simple(&s->reply)) {
844         if (only_structured) {
845             error_setg(errp, "Protocol error: simple reply when structured "
846                              "reply chunk was expected");
847             return -EINVAL;
848         }
849 
850         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
851         if (*request_ret < 0 || !qiov) {
852             return 0;
853         }
854 
855         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
856                                      errp) < 0 ? -EIO : 0;
857     }
858 
859     /* handle structured reply chunk */
860     assert(s->info.structured_reply);
861     chunk = &s->reply.structured;
862 
863     if (chunk->type == NBD_REPLY_TYPE_NONE) {
864         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
865             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
866                        " NBD_REPLY_FLAG_DONE flag set");
867             return -EINVAL;
868         }
869         if (chunk->length) {
870             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
871                        " nonzero length");
872             return -EINVAL;
873         }
874         return 0;
875     }
876 
877     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
878         if (!qiov) {
879             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
880             return -EINVAL;
881         }
882 
883         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
884                                                   qiov, errp);
885     }
886 
887     if (nbd_reply_type_is_error(chunk->type)) {
888         payload = &local_payload;
889     }
890 
891     ret = nbd_co_receive_structured_payload(s, payload, errp);
892     if (ret < 0) {
893         return ret;
894     }
895 
896     if (nbd_reply_type_is_error(chunk->type)) {
897         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
898         g_free(local_payload);
899         return ret;
900     }
901 
902     return 0;
903 }
904 
905 /*
906  * nbd_co_receive_one_chunk
907  * Read reply, wake up connection_co and set s->quit if needed.
908  * Return value is a fatal error code or normal nbd reply error code
909  */
910 static coroutine_fn int nbd_co_receive_one_chunk(
911         BDRVNBDState *s, uint64_t handle, bool only_structured,
912         int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
913         Error **errp)
914 {
915     int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
916                                           request_ret, qiov, payload, errp);
917 
918     if (ret < 0) {
919         memset(reply, 0, sizeof(*reply));
920         nbd_channel_error(s, ret);
921     } else {
922         /* For assert at loop start in nbd_connection_entry */
923         *reply = s->reply;
924     }
925     s->reply.handle = 0;
926 
927     nbd_recv_coroutines_wake(s, false);
928 
929     return ret;
930 }
931 
932 typedef struct NBDReplyChunkIter {
933     int ret;
934     int request_ret;
935     Error *err;
936     bool done, only_structured;
937 } NBDReplyChunkIter;
938 
939 static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
940                                    int ret, Error **local_err)
941 {
942     assert(local_err && *local_err);
943     assert(ret < 0);
944 
945     if (!iter->ret) {
946         iter->ret = ret;
947         error_propagate(&iter->err, *local_err);
948     } else {
949         error_free(*local_err);
950     }
951 
952     *local_err = NULL;
953 }
954 
955 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
956 {
957     assert(ret < 0);
958 
959     if (!iter->request_ret) {
960         iter->request_ret = ret;
961     }
962 }
963 
964 /*
965  * NBD_FOREACH_REPLY_CHUNK
966  * The pointer stored in @payload requires g_free() to free it.
967  */
968 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
969                                 qiov, reply, payload) \
970     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
971          nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
972 
973 /*
974  * nbd_reply_chunk_iter_receive
975  * The pointer stored in @payload requires g_free() to free it.
976  */
977 static bool nbd_reply_chunk_iter_receive(BDRVNBDState *s,
978                                          NBDReplyChunkIter *iter,
979                                          uint64_t handle,
980                                          QEMUIOVector *qiov, NBDReply *reply,
981                                          void **payload)
982 {
983     int ret, request_ret;
984     NBDReply local_reply;
985     NBDStructuredReplyChunk *chunk;
986     Error *local_err = NULL;
987     if (!nbd_client_connected(s)) {
988         error_setg(&local_err, "Connection closed");
989         nbd_iter_channel_error(iter, -EIO, &local_err);
990         goto break_loop;
991     }
992 
993     if (iter->done) {
994         /* Previous iteration was last. */
995         goto break_loop;
996     }
997 
998     if (reply == NULL) {
999         reply = &local_reply;
1000     }
1001 
1002     ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
1003                                    &request_ret, qiov, reply, payload,
1004                                    &local_err);
1005     if (ret < 0) {
1006         nbd_iter_channel_error(iter, ret, &local_err);
1007     } else if (request_ret < 0) {
1008         nbd_iter_request_error(iter, request_ret);
1009     }
1010 
1011     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1012     if (nbd_reply_is_simple(reply) || !nbd_client_connected(s)) {
1013         goto break_loop;
1014     }
1015 
1016     chunk = &reply->structured;
1017     iter->only_structured = true;
1018 
1019     if (chunk->type == NBD_REPLY_TYPE_NONE) {
1020         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1021         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1022         goto break_loop;
1023     }
1024 
1025     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1026         /* This iteration is last. */
1027         iter->done = true;
1028     }
1029 
1030     /* Execute the loop body */
1031     return true;
1032 
1033 break_loop:
1034     s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
1035 
1036     qemu_co_mutex_lock(&s->send_mutex);
1037     s->in_flight--;
1038     qemu_co_queue_next(&s->free_sema);
1039     qemu_co_mutex_unlock(&s->send_mutex);
1040 
1041     return false;
1042 }
1043 
1044 static int nbd_co_receive_return_code(BDRVNBDState *s, uint64_t handle,
1045                                       int *request_ret, Error **errp)
1046 {
1047     NBDReplyChunkIter iter;
1048 
1049     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
1050         /* nbd_reply_chunk_iter_receive does all the work */
1051     }
1052 
1053     error_propagate(errp, iter.err);
1054     *request_ret = iter.request_ret;
1055     return iter.ret;
1056 }
1057 
1058 static int nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t handle,
1059                                         uint64_t offset, QEMUIOVector *qiov,
1060                                         int *request_ret, Error **errp)
1061 {
1062     NBDReplyChunkIter iter;
1063     NBDReply reply;
1064     void *payload = NULL;
1065     Error *local_err = NULL;
1066 
1067     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
1068                             qiov, &reply, &payload)
1069     {
1070         int ret;
1071         NBDStructuredReplyChunk *chunk = &reply.structured;
1072 
1073         assert(nbd_reply_is_structured(&reply));
1074 
1075         switch (chunk->type) {
1076         case NBD_REPLY_TYPE_OFFSET_DATA:
1077             /*
1078              * special cased in nbd_co_receive_one_chunk, data is already
1079              * in qiov
1080              */
1081             break;
1082         case NBD_REPLY_TYPE_OFFSET_HOLE:
1083             ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1084                                                 offset, qiov, &local_err);
1085             if (ret < 0) {
1086                 nbd_channel_error(s, ret);
1087                 nbd_iter_channel_error(&iter, ret, &local_err);
1088             }
1089             break;
1090         default:
1091             if (!nbd_reply_type_is_error(chunk->type)) {
1092                 /* not allowed reply type */
1093                 nbd_channel_error(s, -EINVAL);
1094                 error_setg(&local_err,
1095                            "Unexpected reply type: %d (%s) for CMD_READ",
1096                            chunk->type, nbd_reply_type_lookup(chunk->type));
1097                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1098             }
1099         }
1100 
1101         g_free(payload);
1102         payload = NULL;
1103     }
1104 
1105     error_propagate(errp, iter.err);
1106     *request_ret = iter.request_ret;
1107     return iter.ret;
1108 }
1109 
1110 static int nbd_co_receive_blockstatus_reply(BDRVNBDState *s,
1111                                             uint64_t handle, uint64_t length,
1112                                             NBDExtent *extent,
1113                                             int *request_ret, Error **errp)
1114 {
1115     NBDReplyChunkIter iter;
1116     NBDReply reply;
1117     void *payload = NULL;
1118     Error *local_err = NULL;
1119     bool received = false;
1120 
1121     assert(!extent->length);
1122     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) {
1123         int ret;
1124         NBDStructuredReplyChunk *chunk = &reply.structured;
1125 
1126         assert(nbd_reply_is_structured(&reply));
1127 
1128         switch (chunk->type) {
1129         case NBD_REPLY_TYPE_BLOCK_STATUS:
1130             if (received) {
1131                 nbd_channel_error(s, -EINVAL);
1132                 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1133                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1134             }
1135             received = true;
1136 
1137             ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1138                                                 payload, length, extent,
1139                                                 &local_err);
1140             if (ret < 0) {
1141                 nbd_channel_error(s, ret);
1142                 nbd_iter_channel_error(&iter, ret, &local_err);
1143             }
1144             break;
1145         default:
1146             if (!nbd_reply_type_is_error(chunk->type)) {
1147                 nbd_channel_error(s, -EINVAL);
1148                 error_setg(&local_err,
1149                            "Unexpected reply type: %d (%s) "
1150                            "for CMD_BLOCK_STATUS",
1151                            chunk->type, nbd_reply_type_lookup(chunk->type));
1152                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1153             }
1154         }
1155 
1156         g_free(payload);
1157         payload = NULL;
1158     }
1159 
1160     if (!extent->length && !iter.request_ret) {
1161         error_setg(&local_err, "Server did not reply with any status extents");
1162         nbd_iter_channel_error(&iter, -EIO, &local_err);
1163     }
1164 
1165     error_propagate(errp, iter.err);
1166     *request_ret = iter.request_ret;
1167     return iter.ret;
1168 }
1169 
1170 static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1171                           QEMUIOVector *write_qiov)
1172 {
1173     int ret, request_ret;
1174     Error *local_err = NULL;
1175     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1176 
1177     assert(request->type != NBD_CMD_READ);
1178     if (write_qiov) {
1179         assert(request->type == NBD_CMD_WRITE);
1180         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1181     } else {
1182         assert(request->type != NBD_CMD_WRITE);
1183     }
1184 
1185     do {
1186         ret = nbd_co_send_request(bs, request, write_qiov);
1187         if (ret < 0) {
1188             continue;
1189         }
1190 
1191         ret = nbd_co_receive_return_code(s, request->handle,
1192                                          &request_ret, &local_err);
1193         if (local_err) {
1194             trace_nbd_co_request_fail(request->from, request->len,
1195                                       request->handle, request->flags,
1196                                       request->type,
1197                                       nbd_cmd_lookup(request->type),
1198                                       ret, error_get_pretty(local_err));
1199             error_free(local_err);
1200             local_err = NULL;
1201         }
1202     } while (ret < 0 && nbd_client_connecting_wait(s));
1203 
1204     return ret ? ret : request_ret;
1205 }
1206 
1207 static int nbd_client_co_preadv(BlockDriverState *bs, int64_t offset,
1208                                 int64_t bytes, QEMUIOVector *qiov,
1209                                 BdrvRequestFlags flags)
1210 {
1211     int ret, request_ret;
1212     Error *local_err = NULL;
1213     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1214     NBDRequest request = {
1215         .type = NBD_CMD_READ,
1216         .from = offset,
1217         .len = bytes,
1218     };
1219 
1220     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1221     assert(!flags);
1222 
1223     if (!bytes) {
1224         return 0;
1225     }
1226     /*
1227      * Work around the fact that the block layer doesn't do
1228      * byte-accurate sizing yet - if the read exceeds the server's
1229      * advertised size because the block layer rounded size up, then
1230      * truncate the request to the server and tail-pad with zero.
1231      */
1232     if (offset >= s->info.size) {
1233         assert(bytes < BDRV_SECTOR_SIZE);
1234         qemu_iovec_memset(qiov, 0, 0, bytes);
1235         return 0;
1236     }
1237     if (offset + bytes > s->info.size) {
1238         uint64_t slop = offset + bytes - s->info.size;
1239 
1240         assert(slop < BDRV_SECTOR_SIZE);
1241         qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1242         request.len -= slop;
1243     }
1244 
1245     do {
1246         ret = nbd_co_send_request(bs, &request, NULL);
1247         if (ret < 0) {
1248             continue;
1249         }
1250 
1251         ret = nbd_co_receive_cmdread_reply(s, request.handle, offset, qiov,
1252                                            &request_ret, &local_err);
1253         if (local_err) {
1254             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1255                                       request.flags, request.type,
1256                                       nbd_cmd_lookup(request.type),
1257                                       ret, error_get_pretty(local_err));
1258             error_free(local_err);
1259             local_err = NULL;
1260         }
1261     } while (ret < 0 && nbd_client_connecting_wait(s));
1262 
1263     return ret ? ret : request_ret;
1264 }
1265 
1266 static int nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset,
1267                                  int64_t bytes, QEMUIOVector *qiov,
1268                                  BdrvRequestFlags flags)
1269 {
1270     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1271     NBDRequest request = {
1272         .type = NBD_CMD_WRITE,
1273         .from = offset,
1274         .len = bytes,
1275     };
1276 
1277     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1278     if (flags & BDRV_REQ_FUA) {
1279         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1280         request.flags |= NBD_CMD_FLAG_FUA;
1281     }
1282 
1283     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1284 
1285     if (!bytes) {
1286         return 0;
1287     }
1288     return nbd_co_request(bs, &request, qiov);
1289 }
1290 
1291 static int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1292                                        int64_t bytes, BdrvRequestFlags flags)
1293 {
1294     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1295     NBDRequest request = {
1296         .type = NBD_CMD_WRITE_ZEROES,
1297         .from = offset,
1298         .len = bytes,  /* .len is uint32_t actually */
1299     };
1300 
1301     assert(bytes <= UINT32_MAX); /* rely on max_pwrite_zeroes */
1302 
1303     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1304     if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1305         return -ENOTSUP;
1306     }
1307 
1308     if (flags & BDRV_REQ_FUA) {
1309         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1310         request.flags |= NBD_CMD_FLAG_FUA;
1311     }
1312     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1313         request.flags |= NBD_CMD_FLAG_NO_HOLE;
1314     }
1315     if (flags & BDRV_REQ_NO_FALLBACK) {
1316         assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1317         request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1318     }
1319 
1320     if (!bytes) {
1321         return 0;
1322     }
1323     return nbd_co_request(bs, &request, NULL);
1324 }
1325 
1326 static int nbd_client_co_flush(BlockDriverState *bs)
1327 {
1328     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1329     NBDRequest request = { .type = NBD_CMD_FLUSH };
1330 
1331     if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1332         return 0;
1333     }
1334 
1335     request.from = 0;
1336     request.len = 0;
1337 
1338     return nbd_co_request(bs, &request, NULL);
1339 }
1340 
1341 static int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset,
1342                                   int64_t bytes)
1343 {
1344     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1345     NBDRequest request = {
1346         .type = NBD_CMD_TRIM,
1347         .from = offset,
1348         .len = bytes, /* len is uint32_t */
1349     };
1350 
1351     assert(bytes <= UINT32_MAX); /* rely on max_pdiscard */
1352 
1353     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1354     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1355         return 0;
1356     }
1357 
1358     return nbd_co_request(bs, &request, NULL);
1359 }
1360 
1361 static int coroutine_fn nbd_client_co_block_status(
1362         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1363         int64_t *pnum, int64_t *map, BlockDriverState **file)
1364 {
1365     int ret, request_ret;
1366     NBDExtent extent = { 0 };
1367     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1368     Error *local_err = NULL;
1369 
1370     NBDRequest request = {
1371         .type = NBD_CMD_BLOCK_STATUS,
1372         .from = offset,
1373         .len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1374                    MIN(bytes, s->info.size - offset)),
1375         .flags = NBD_CMD_FLAG_REQ_ONE,
1376     };
1377 
1378     if (!s->info.base_allocation) {
1379         *pnum = bytes;
1380         *map = offset;
1381         *file = bs;
1382         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1383     }
1384 
1385     /*
1386      * Work around the fact that the block layer doesn't do
1387      * byte-accurate sizing yet - if the status request exceeds the
1388      * server's advertised size because the block layer rounded size
1389      * up, we truncated the request to the server (above), or are
1390      * called on just the hole.
1391      */
1392     if (offset >= s->info.size) {
1393         *pnum = bytes;
1394         assert(bytes < BDRV_SECTOR_SIZE);
1395         /* Intentionally don't report offset_valid for the hole */
1396         return BDRV_BLOCK_ZERO;
1397     }
1398 
1399     if (s->info.min_block) {
1400         assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1401     }
1402     do {
1403         ret = nbd_co_send_request(bs, &request, NULL);
1404         if (ret < 0) {
1405             continue;
1406         }
1407 
1408         ret = nbd_co_receive_blockstatus_reply(s, request.handle, bytes,
1409                                                &extent, &request_ret,
1410                                                &local_err);
1411         if (local_err) {
1412             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1413                                       request.flags, request.type,
1414                                       nbd_cmd_lookup(request.type),
1415                                       ret, error_get_pretty(local_err));
1416             error_free(local_err);
1417             local_err = NULL;
1418         }
1419     } while (ret < 0 && nbd_client_connecting_wait(s));
1420 
1421     if (ret < 0 || request_ret < 0) {
1422         return ret ? ret : request_ret;
1423     }
1424 
1425     assert(extent.length);
1426     *pnum = extent.length;
1427     *map = offset;
1428     *file = bs;
1429     return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1430         (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1431         BDRV_BLOCK_OFFSET_VALID;
1432 }
1433 
1434 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1435                                      BlockReopenQueue *queue, Error **errp)
1436 {
1437     BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1438 
1439     if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1440         error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1441         return -EACCES;
1442     }
1443     return 0;
1444 }
1445 
1446 static void nbd_yank(void *opaque)
1447 {
1448     BlockDriverState *bs = opaque;
1449     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1450 
1451     qatomic_store_release(&s->state, NBD_CLIENT_QUIT);
1452     qio_channel_shutdown(QIO_CHANNEL(s->ioc), QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1453 }
1454 
1455 static void nbd_client_close(BlockDriverState *bs)
1456 {
1457     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1458     NBDRequest request = { .type = NBD_CMD_DISC };
1459 
1460     if (s->ioc) {
1461         nbd_send_request(s->ioc, &request);
1462     }
1463 
1464     nbd_teardown_connection(bs);
1465 }
1466 
1467 
1468 /*
1469  * Parse nbd_open options
1470  */
1471 
1472 static int nbd_parse_uri(const char *filename, QDict *options)
1473 {
1474     URI *uri;
1475     const char *p;
1476     QueryParams *qp = NULL;
1477     int ret = 0;
1478     bool is_unix;
1479 
1480     uri = uri_parse(filename);
1481     if (!uri) {
1482         return -EINVAL;
1483     }
1484 
1485     /* transport */
1486     if (!g_strcmp0(uri->scheme, "nbd")) {
1487         is_unix = false;
1488     } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1489         is_unix = false;
1490     } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1491         is_unix = true;
1492     } else {
1493         ret = -EINVAL;
1494         goto out;
1495     }
1496 
1497     p = uri->path ? uri->path : "";
1498     if (p[0] == '/') {
1499         p++;
1500     }
1501     if (p[0]) {
1502         qdict_put_str(options, "export", p);
1503     }
1504 
1505     qp = query_params_parse(uri->query);
1506     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1507         ret = -EINVAL;
1508         goto out;
1509     }
1510 
1511     if (is_unix) {
1512         /* nbd+unix:///export?socket=path */
1513         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1514             ret = -EINVAL;
1515             goto out;
1516         }
1517         qdict_put_str(options, "server.type", "unix");
1518         qdict_put_str(options, "server.path", qp->p[0].value);
1519     } else {
1520         QString *host;
1521         char *port_str;
1522 
1523         /* nbd[+tcp]://host[:port]/export */
1524         if (!uri->server) {
1525             ret = -EINVAL;
1526             goto out;
1527         }
1528 
1529         /* strip braces from literal IPv6 address */
1530         if (uri->server[0] == '[') {
1531             host = qstring_from_substr(uri->server, 1,
1532                                        strlen(uri->server) - 1);
1533         } else {
1534             host = qstring_from_str(uri->server);
1535         }
1536 
1537         qdict_put_str(options, "server.type", "inet");
1538         qdict_put(options, "server.host", host);
1539 
1540         port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1541         qdict_put_str(options, "server.port", port_str);
1542         g_free(port_str);
1543     }
1544 
1545 out:
1546     if (qp) {
1547         query_params_free(qp);
1548     }
1549     uri_free(uri);
1550     return ret;
1551 }
1552 
1553 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1554 {
1555     const QDictEntry *e;
1556 
1557     for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1558         if (!strcmp(e->key, "host") ||
1559             !strcmp(e->key, "port") ||
1560             !strcmp(e->key, "path") ||
1561             !strcmp(e->key, "export") ||
1562             strstart(e->key, "server.", NULL))
1563         {
1564             error_setg(errp, "Option '%s' cannot be used with a file name",
1565                        e->key);
1566             return true;
1567         }
1568     }
1569 
1570     return false;
1571 }
1572 
1573 static void nbd_parse_filename(const char *filename, QDict *options,
1574                                Error **errp)
1575 {
1576     g_autofree char *file = NULL;
1577     char *export_name;
1578     const char *host_spec;
1579     const char *unixpath;
1580 
1581     if (nbd_has_filename_options_conflict(options, errp)) {
1582         return;
1583     }
1584 
1585     if (strstr(filename, "://")) {
1586         int ret = nbd_parse_uri(filename, options);
1587         if (ret < 0) {
1588             error_setg(errp, "No valid URL specified");
1589         }
1590         return;
1591     }
1592 
1593     file = g_strdup(filename);
1594 
1595     export_name = strstr(file, EN_OPTSTR);
1596     if (export_name) {
1597         if (export_name[strlen(EN_OPTSTR)] == 0) {
1598             return;
1599         }
1600         export_name[0] = 0; /* truncate 'file' */
1601         export_name += strlen(EN_OPTSTR);
1602 
1603         qdict_put_str(options, "export", export_name);
1604     }
1605 
1606     /* extract the host_spec - fail if it's not nbd:... */
1607     if (!strstart(file, "nbd:", &host_spec)) {
1608         error_setg(errp, "File name string for NBD must start with 'nbd:'");
1609         return;
1610     }
1611 
1612     if (!*host_spec) {
1613         return;
1614     }
1615 
1616     /* are we a UNIX or TCP socket? */
1617     if (strstart(host_spec, "unix:", &unixpath)) {
1618         qdict_put_str(options, "server.type", "unix");
1619         qdict_put_str(options, "server.path", unixpath);
1620     } else {
1621         InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1622 
1623         if (inet_parse(addr, host_spec, errp)) {
1624             goto out_inet;
1625         }
1626 
1627         qdict_put_str(options, "server.type", "inet");
1628         qdict_put_str(options, "server.host", addr->host);
1629         qdict_put_str(options, "server.port", addr->port);
1630     out_inet:
1631         qapi_free_InetSocketAddress(addr);
1632     }
1633 }
1634 
1635 static bool nbd_process_legacy_socket_options(QDict *output_options,
1636                                               QemuOpts *legacy_opts,
1637                                               Error **errp)
1638 {
1639     const char *path = qemu_opt_get(legacy_opts, "path");
1640     const char *host = qemu_opt_get(legacy_opts, "host");
1641     const char *port = qemu_opt_get(legacy_opts, "port");
1642     const QDictEntry *e;
1643 
1644     if (!path && !host && !port) {
1645         return true;
1646     }
1647 
1648     for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1649     {
1650         if (strstart(e->key, "server.", NULL)) {
1651             error_setg(errp, "Cannot use 'server' and path/host/port at the "
1652                        "same time");
1653             return false;
1654         }
1655     }
1656 
1657     if (path && host) {
1658         error_setg(errp, "path and host may not be used at the same time");
1659         return false;
1660     } else if (path) {
1661         if (port) {
1662             error_setg(errp, "port may not be used without host");
1663             return false;
1664         }
1665 
1666         qdict_put_str(output_options, "server.type", "unix");
1667         qdict_put_str(output_options, "server.path", path);
1668     } else if (host) {
1669         qdict_put_str(output_options, "server.type", "inet");
1670         qdict_put_str(output_options, "server.host", host);
1671         qdict_put_str(output_options, "server.port",
1672                       port ?: stringify(NBD_DEFAULT_PORT));
1673     }
1674 
1675     return true;
1676 }
1677 
1678 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1679                                  Error **errp)
1680 {
1681     SocketAddress *saddr = NULL;
1682     QDict *addr = NULL;
1683     Visitor *iv = NULL;
1684 
1685     qdict_extract_subqdict(options, &addr, "server.");
1686     if (!qdict_size(addr)) {
1687         error_setg(errp, "NBD server address missing");
1688         goto done;
1689     }
1690 
1691     iv = qobject_input_visitor_new_flat_confused(addr, errp);
1692     if (!iv) {
1693         goto done;
1694     }
1695 
1696     if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1697         goto done;
1698     }
1699 
1700     if (socket_address_parse_named_fd(saddr, errp) < 0) {
1701         qapi_free_SocketAddress(saddr);
1702         saddr = NULL;
1703         goto done;
1704     }
1705 
1706 done:
1707     qobject_unref(addr);
1708     visit_free(iv);
1709     return saddr;
1710 }
1711 
1712 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1713 {
1714     Object *obj;
1715     QCryptoTLSCreds *creds;
1716 
1717     obj = object_resolve_path_component(
1718         object_get_objects_root(), id);
1719     if (!obj) {
1720         error_setg(errp, "No TLS credentials with id '%s'",
1721                    id);
1722         return NULL;
1723     }
1724     creds = (QCryptoTLSCreds *)
1725         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1726     if (!creds) {
1727         error_setg(errp, "Object with id '%s' is not TLS credentials",
1728                    id);
1729         return NULL;
1730     }
1731 
1732     if (!qcrypto_tls_creds_check_endpoint(creds,
1733                                           QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1734                                           errp)) {
1735         return NULL;
1736     }
1737     object_ref(obj);
1738     return creds;
1739 }
1740 
1741 
1742 static QemuOptsList nbd_runtime_opts = {
1743     .name = "nbd",
1744     .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1745     .desc = {
1746         {
1747             .name = "host",
1748             .type = QEMU_OPT_STRING,
1749             .help = "TCP host to connect to",
1750         },
1751         {
1752             .name = "port",
1753             .type = QEMU_OPT_STRING,
1754             .help = "TCP port to connect to",
1755         },
1756         {
1757             .name = "path",
1758             .type = QEMU_OPT_STRING,
1759             .help = "Unix socket path to connect to",
1760         },
1761         {
1762             .name = "export",
1763             .type = QEMU_OPT_STRING,
1764             .help = "Name of the NBD export to open",
1765         },
1766         {
1767             .name = "tls-creds",
1768             .type = QEMU_OPT_STRING,
1769             .help = "ID of the TLS credentials to use",
1770         },
1771         {
1772             .name = "tls-hostname",
1773             .type = QEMU_OPT_STRING,
1774             .help = "Override hostname for validating TLS x509 certificate",
1775         },
1776         {
1777             .name = "x-dirty-bitmap",
1778             .type = QEMU_OPT_STRING,
1779             .help = "experimental: expose named dirty bitmap in place of "
1780                     "block status",
1781         },
1782         {
1783             .name = "reconnect-delay",
1784             .type = QEMU_OPT_NUMBER,
1785             .help = "On an unexpected disconnect, the nbd client tries to "
1786                     "connect again until succeeding or encountering a serious "
1787                     "error.  During the first @reconnect-delay seconds, all "
1788                     "requests are paused and will be rerun on a successful "
1789                     "reconnect. After that time, any delayed requests and all "
1790                     "future requests before a successful reconnect will "
1791                     "immediately fail. Default 0",
1792         },
1793         {
1794             .name = "open-timeout",
1795             .type = QEMU_OPT_NUMBER,
1796             .help = "In seconds. If zero, the nbd driver tries the connection "
1797                     "only once, and fails to open if the connection fails. "
1798                     "If non-zero, the nbd driver will repeat connection "
1799                     "attempts until successful or until @open-timeout seconds "
1800                     "have elapsed. Default 0",
1801         },
1802         { /* end of list */ }
1803     },
1804 };
1805 
1806 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1807                                Error **errp)
1808 {
1809     BDRVNBDState *s = bs->opaque;
1810     QemuOpts *opts;
1811     int ret = -EINVAL;
1812 
1813     opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1814     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1815         goto error;
1816     }
1817 
1818     /* Translate @host, @port, and @path to a SocketAddress */
1819     if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1820         goto error;
1821     }
1822 
1823     /* Pop the config into our state object. Exit if invalid. */
1824     s->saddr = nbd_config(s, options, errp);
1825     if (!s->saddr) {
1826         goto error;
1827     }
1828 
1829     s->export = g_strdup(qemu_opt_get(opts, "export"));
1830     if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1831         error_setg(errp, "export name too long to send to server");
1832         goto error;
1833     }
1834 
1835     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1836     if (s->tlscredsid) {
1837         s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1838         if (!s->tlscreds) {
1839             goto error;
1840         }
1841 
1842         s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname"));
1843         if (!s->tlshostname &&
1844             s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1845             s->tlshostname = g_strdup(s->saddr->u.inet.host);
1846         }
1847     }
1848 
1849     s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1850     if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1851         error_setg(errp, "x-dirty-bitmap query too long to send to server");
1852         goto error;
1853     }
1854 
1855     s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1856     s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1857 
1858     ret = 0;
1859 
1860  error:
1861     qemu_opts_del(opts);
1862     return ret;
1863 }
1864 
1865 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1866                     Error **errp)
1867 {
1868     int ret;
1869     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1870 
1871     s->bs = bs;
1872     qemu_co_mutex_init(&s->send_mutex);
1873     qemu_co_queue_init(&s->free_sema);
1874     qemu_co_mutex_init(&s->receive_mutex);
1875 
1876     if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1877         return -EEXIST;
1878     }
1879 
1880     ret = nbd_process_options(bs, options, errp);
1881     if (ret < 0) {
1882         goto fail;
1883     }
1884 
1885     s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1886                                         s->x_dirty_bitmap, s->tlscreds,
1887                                         s->tlshostname);
1888 
1889     if (s->open_timeout) {
1890         nbd_client_connection_enable_retry(s->conn);
1891         open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1892                         s->open_timeout * NANOSECONDS_PER_SECOND);
1893     }
1894 
1895     s->state = NBD_CLIENT_CONNECTING_WAIT;
1896     ret = nbd_do_establish_connection(bs, errp);
1897     if (ret < 0) {
1898         goto fail;
1899     }
1900 
1901     /*
1902      * The connect attempt is done, so we no longer need this timer.
1903      * Delete it, because we do not want it to be around when this node
1904      * is drained or closed.
1905      */
1906     open_timer_del(s);
1907 
1908     nbd_client_connection_enable_retry(s->conn);
1909 
1910     return 0;
1911 
1912 fail:
1913     open_timer_del(s);
1914     nbd_clear_bdrvstate(bs);
1915     return ret;
1916 }
1917 
1918 static int nbd_co_flush(BlockDriverState *bs)
1919 {
1920     return nbd_client_co_flush(bs);
1921 }
1922 
1923 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1924 {
1925     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1926     uint32_t min = s->info.min_block;
1927     uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1928 
1929     /*
1930      * If the server did not advertise an alignment:
1931      * - a size that is not sector-aligned implies that an alignment
1932      *   of 1 can be used to access those tail bytes
1933      * - advertisement of block status requires an alignment of 1, so
1934      *   that we don't violate block layer constraints that block
1935      *   status is always aligned (as we can't control whether the
1936      *   server will report sub-sector extents, such as a hole at EOF
1937      *   on an unaligned POSIX file)
1938      * - otherwise, assume the server is so old that we are safer avoiding
1939      *   sub-sector requests
1940      */
1941     if (!min) {
1942         min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1943                s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1944     }
1945 
1946     bs->bl.request_alignment = min;
1947     bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1948     bs->bl.max_pwrite_zeroes = max;
1949     bs->bl.max_transfer = max;
1950 
1951     if (s->info.opt_block &&
1952         s->info.opt_block > bs->bl.opt_transfer) {
1953         bs->bl.opt_transfer = s->info.opt_block;
1954     }
1955 }
1956 
1957 static void nbd_close(BlockDriverState *bs)
1958 {
1959     nbd_client_close(bs);
1960     nbd_clear_bdrvstate(bs);
1961 }
1962 
1963 /*
1964  * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1965  * to a smaller size with exact=false, there is no reason to fail the
1966  * operation.
1967  *
1968  * Preallocation mode is ignored since it does not seems useful to fail when
1969  * we never change anything.
1970  */
1971 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
1972                                         bool exact, PreallocMode prealloc,
1973                                         BdrvRequestFlags flags, Error **errp)
1974 {
1975     BDRVNBDState *s = bs->opaque;
1976 
1977     if (offset != s->info.size && exact) {
1978         error_setg(errp, "Cannot resize NBD nodes");
1979         return -ENOTSUP;
1980     }
1981 
1982     if (offset > s->info.size) {
1983         error_setg(errp, "Cannot grow NBD nodes");
1984         return -EINVAL;
1985     }
1986 
1987     return 0;
1988 }
1989 
1990 static int64_t nbd_getlength(BlockDriverState *bs)
1991 {
1992     BDRVNBDState *s = bs->opaque;
1993 
1994     return s->info.size;
1995 }
1996 
1997 static void nbd_refresh_filename(BlockDriverState *bs)
1998 {
1999     BDRVNBDState *s = bs->opaque;
2000     const char *host = NULL, *port = NULL, *path = NULL;
2001     size_t len = 0;
2002 
2003     if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
2004         const InetSocketAddress *inet = &s->saddr->u.inet;
2005         if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
2006             host = inet->host;
2007             port = inet->port;
2008         }
2009     } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2010         path = s->saddr->u.q_unix.path;
2011     } /* else can't represent as pseudo-filename */
2012 
2013     if (path && s->export) {
2014         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2015                        "nbd+unix:///%s?socket=%s", s->export, path);
2016     } else if (path && !s->export) {
2017         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2018                        "nbd+unix://?socket=%s", path);
2019     } else if (host && s->export) {
2020         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2021                        "nbd://%s:%s/%s", host, port, s->export);
2022     } else if (host && !s->export) {
2023         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2024                        "nbd://%s:%s", host, port);
2025     }
2026     if (len >= sizeof(bs->exact_filename)) {
2027         /* Name is too long to represent exactly, so leave it empty. */
2028         bs->exact_filename[0] = '\0';
2029     }
2030 }
2031 
2032 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2033 {
2034     /* The generic bdrv_dirname() implementation is able to work out some
2035      * directory name for NBD nodes, but that would be wrong. So far there is no
2036      * specification for how "export paths" would work, so NBD does not have
2037      * directory names. */
2038     error_setg(errp, "Cannot generate a base directory for NBD nodes");
2039     return NULL;
2040 }
2041 
2042 static const char *const nbd_strong_runtime_opts[] = {
2043     "path",
2044     "host",
2045     "port",
2046     "export",
2047     "tls-creds",
2048     "tls-hostname",
2049     "server.",
2050 
2051     NULL
2052 };
2053 
2054 static void nbd_cancel_in_flight(BlockDriverState *bs)
2055 {
2056     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2057 
2058     reconnect_delay_timer_del(s);
2059 
2060     if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2061         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2062         qemu_co_queue_restart_all(&s->free_sema);
2063     }
2064 
2065     nbd_co_establish_connection_cancel(s->conn);
2066 }
2067 
2068 static void nbd_attach_aio_context(BlockDriverState *bs,
2069                                    AioContext *new_context)
2070 {
2071     BDRVNBDState *s = bs->opaque;
2072 
2073     /* The open_timer is used only during nbd_open() */
2074     assert(!s->open_timer);
2075 
2076     /*
2077      * The reconnect_delay_timer is scheduled in I/O paths when the
2078      * connection is lost, to cancel the reconnection attempt after a
2079      * given time.  Once this attempt is done (successfully or not),
2080      * nbd_reconnect_attempt() ensures the timer is deleted before the
2081      * respective I/O request is resumed.
2082      * Since the AioContext can only be changed when a node is drained,
2083      * the reconnect_delay_timer cannot be active here.
2084      */
2085     assert(!s->reconnect_delay_timer);
2086 
2087     if (s->ioc) {
2088         qio_channel_attach_aio_context(s->ioc, new_context);
2089     }
2090 }
2091 
2092 static void nbd_detach_aio_context(BlockDriverState *bs)
2093 {
2094     BDRVNBDState *s = bs->opaque;
2095 
2096     assert(!s->open_timer);
2097     assert(!s->reconnect_delay_timer);
2098 
2099     if (s->ioc) {
2100         qio_channel_detach_aio_context(s->ioc);
2101     }
2102 }
2103 
2104 static BlockDriver bdrv_nbd = {
2105     .format_name                = "nbd",
2106     .protocol_name              = "nbd",
2107     .instance_size              = sizeof(BDRVNBDState),
2108     .bdrv_parse_filename        = nbd_parse_filename,
2109     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2110     .create_opts                = &bdrv_create_opts_simple,
2111     .bdrv_file_open             = nbd_open,
2112     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2113     .bdrv_co_preadv             = nbd_client_co_preadv,
2114     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2115     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2116     .bdrv_close                 = nbd_close,
2117     .bdrv_co_flush_to_os        = nbd_co_flush,
2118     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2119     .bdrv_refresh_limits        = nbd_refresh_limits,
2120     .bdrv_co_truncate           = nbd_co_truncate,
2121     .bdrv_getlength             = nbd_getlength,
2122     .bdrv_refresh_filename      = nbd_refresh_filename,
2123     .bdrv_co_block_status       = nbd_client_co_block_status,
2124     .bdrv_dirname               = nbd_dirname,
2125     .strong_runtime_opts        = nbd_strong_runtime_opts,
2126     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2127 
2128     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2129     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2130 };
2131 
2132 static BlockDriver bdrv_nbd_tcp = {
2133     .format_name                = "nbd",
2134     .protocol_name              = "nbd+tcp",
2135     .instance_size              = sizeof(BDRVNBDState),
2136     .bdrv_parse_filename        = nbd_parse_filename,
2137     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2138     .create_opts                = &bdrv_create_opts_simple,
2139     .bdrv_file_open             = nbd_open,
2140     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2141     .bdrv_co_preadv             = nbd_client_co_preadv,
2142     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2143     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2144     .bdrv_close                 = nbd_close,
2145     .bdrv_co_flush_to_os        = nbd_co_flush,
2146     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2147     .bdrv_refresh_limits        = nbd_refresh_limits,
2148     .bdrv_co_truncate           = nbd_co_truncate,
2149     .bdrv_getlength             = nbd_getlength,
2150     .bdrv_refresh_filename      = nbd_refresh_filename,
2151     .bdrv_co_block_status       = nbd_client_co_block_status,
2152     .bdrv_dirname               = nbd_dirname,
2153     .strong_runtime_opts        = nbd_strong_runtime_opts,
2154     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2155 
2156     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2157     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2158 };
2159 
2160 static BlockDriver bdrv_nbd_unix = {
2161     .format_name                = "nbd",
2162     .protocol_name              = "nbd+unix",
2163     .instance_size              = sizeof(BDRVNBDState),
2164     .bdrv_parse_filename        = nbd_parse_filename,
2165     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2166     .create_opts                = &bdrv_create_opts_simple,
2167     .bdrv_file_open             = nbd_open,
2168     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2169     .bdrv_co_preadv             = nbd_client_co_preadv,
2170     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2171     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2172     .bdrv_close                 = nbd_close,
2173     .bdrv_co_flush_to_os        = nbd_co_flush,
2174     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2175     .bdrv_refresh_limits        = nbd_refresh_limits,
2176     .bdrv_co_truncate           = nbd_co_truncate,
2177     .bdrv_getlength             = nbd_getlength,
2178     .bdrv_refresh_filename      = nbd_refresh_filename,
2179     .bdrv_co_block_status       = nbd_client_co_block_status,
2180     .bdrv_dirname               = nbd_dirname,
2181     .strong_runtime_opts        = nbd_strong_runtime_opts,
2182     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2183 
2184     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2185     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2186 };
2187 
2188 static void bdrv_nbd_init(void)
2189 {
2190     bdrv_register(&bdrv_nbd);
2191     bdrv_register(&bdrv_nbd_tcp);
2192     bdrv_register(&bdrv_nbd_unix);
2193 }
2194 
2195 block_init(bdrv_nbd_init);
2196