xref: /openbmc/qemu/block/nbd.c (revision 7c8d2fc4)
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, *tlscredsid;
94     QCryptoTLSCreds *tlscreds;
95     const char *hostname;
96     char *x_dirty_bitmap;
97     bool alloc_depth;
98 
99     NBDClientConnection *conn;
100 } BDRVNBDState;
101 
102 static void nbd_yank(void *opaque);
103 
104 static void nbd_clear_bdrvstate(BlockDriverState *bs)
105 {
106     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
107 
108     nbd_client_connection_release(s->conn);
109     s->conn = NULL;
110 
111     yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name));
112 
113     /* Must not leave timers behind that would access freed data */
114     assert(!s->reconnect_delay_timer);
115     assert(!s->open_timer);
116 
117     object_unref(OBJECT(s->tlscreds));
118     qapi_free_SocketAddress(s->saddr);
119     s->saddr = NULL;
120     g_free(s->export);
121     s->export = NULL;
122     g_free(s->tlscredsid);
123     s->tlscredsid = NULL;
124     g_free(s->x_dirty_bitmap);
125     s->x_dirty_bitmap = NULL;
126 }
127 
128 static bool nbd_client_connected(BDRVNBDState *s)
129 {
130     return qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED;
131 }
132 
133 static bool nbd_recv_coroutine_wake_one(NBDClientRequest *req)
134 {
135     if (req->receiving) {
136         req->receiving = false;
137         aio_co_wake(req->coroutine);
138         return true;
139     }
140 
141     return false;
142 }
143 
144 static void nbd_recv_coroutines_wake(BDRVNBDState *s, bool all)
145 {
146     int i;
147 
148     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
149         if (nbd_recv_coroutine_wake_one(&s->requests[i]) && !all) {
150             return;
151         }
152     }
153 }
154 
155 static void nbd_channel_error(BDRVNBDState *s, int ret)
156 {
157     if (nbd_client_connected(s)) {
158         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
159     }
160 
161     if (ret == -EIO) {
162         if (nbd_client_connected(s)) {
163             s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT :
164                                             NBD_CLIENT_CONNECTING_NOWAIT;
165         }
166     } else {
167         s->state = NBD_CLIENT_QUIT;
168     }
169 
170     nbd_recv_coroutines_wake(s, true);
171 }
172 
173 static void reconnect_delay_timer_del(BDRVNBDState *s)
174 {
175     if (s->reconnect_delay_timer) {
176         timer_free(s->reconnect_delay_timer);
177         s->reconnect_delay_timer = NULL;
178     }
179 }
180 
181 static void reconnect_delay_timer_cb(void *opaque)
182 {
183     BDRVNBDState *s = opaque;
184 
185     if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT) {
186         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
187         nbd_co_establish_connection_cancel(s->conn);
188         while (qemu_co_enter_next(&s->free_sema, NULL)) {
189             /* Resume all queued requests */
190         }
191     }
192 
193     reconnect_delay_timer_del(s);
194 }
195 
196 static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
197 {
198     if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTING_WAIT) {
199         return;
200     }
201 
202     assert(!s->reconnect_delay_timer);
203     s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
204                                              QEMU_CLOCK_REALTIME,
205                                              SCALE_NS,
206                                              reconnect_delay_timer_cb, s);
207     timer_mod(s->reconnect_delay_timer, expire_time_ns);
208 }
209 
210 static void nbd_teardown_connection(BlockDriverState *bs)
211 {
212     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
213 
214     assert(!s->in_flight);
215 
216     if (s->ioc) {
217         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
218         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
219                                  nbd_yank, s->bs);
220         object_unref(OBJECT(s->ioc));
221         s->ioc = NULL;
222     }
223 
224     s->state = NBD_CLIENT_QUIT;
225 }
226 
227 static void open_timer_del(BDRVNBDState *s)
228 {
229     if (s->open_timer) {
230         timer_free(s->open_timer);
231         s->open_timer = NULL;
232     }
233 }
234 
235 static void open_timer_cb(void *opaque)
236 {
237     BDRVNBDState *s = opaque;
238 
239     nbd_co_establish_connection_cancel(s->conn);
240     open_timer_del(s);
241 }
242 
243 static void open_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
244 {
245     assert(!s->open_timer);
246     s->open_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
247                                   QEMU_CLOCK_REALTIME,
248                                   SCALE_NS,
249                                   open_timer_cb, s);
250     timer_mod(s->open_timer, expire_time_ns);
251 }
252 
253 static bool nbd_client_connecting(BDRVNBDState *s)
254 {
255     NBDClientState state = qatomic_load_acquire(&s->state);
256     return state == NBD_CLIENT_CONNECTING_WAIT ||
257         state == NBD_CLIENT_CONNECTING_NOWAIT;
258 }
259 
260 static bool nbd_client_connecting_wait(BDRVNBDState *s)
261 {
262     return qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT;
263 }
264 
265 /*
266  * Update @bs with information learned during a completed negotiation process.
267  * Return failure if the server's advertised options are incompatible with the
268  * client's needs.
269  */
270 static int nbd_handle_updated_info(BlockDriverState *bs, Error **errp)
271 {
272     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
273     int ret;
274 
275     if (s->x_dirty_bitmap) {
276         if (!s->info.base_allocation) {
277             error_setg(errp, "requested x-dirty-bitmap %s not found",
278                        s->x_dirty_bitmap);
279             return -EINVAL;
280         }
281         if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) {
282             s->alloc_depth = true;
283         }
284     }
285 
286     if (s->info.flags & NBD_FLAG_READ_ONLY) {
287         ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp);
288         if (ret < 0) {
289             return ret;
290         }
291     }
292 
293     if (s->info.flags & NBD_FLAG_SEND_FUA) {
294         bs->supported_write_flags = BDRV_REQ_FUA;
295         bs->supported_zero_flags |= BDRV_REQ_FUA;
296     }
297 
298     if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
299         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
300         if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) {
301             bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK;
302         }
303     }
304 
305     trace_nbd_client_handshake_success(s->export);
306 
307     return 0;
308 }
309 
310 int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs,
311                                                 Error **errp)
312 {
313     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
314     int ret;
315     bool blocking = nbd_client_connecting_wait(s);
316 
317     assert(!s->ioc);
318 
319     s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp);
320     if (!s->ioc) {
321         return -ECONNREFUSED;
322     }
323 
324     yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank,
325                            bs);
326 
327     ret = nbd_handle_updated_info(s->bs, NULL);
328     if (ret < 0) {
329         /*
330          * We have connected, but must fail for other reasons.
331          * Send NBD_CMD_DISC as a courtesy to the server.
332          */
333         NBDRequest request = { .type = NBD_CMD_DISC };
334 
335         nbd_send_request(s->ioc, &request);
336 
337         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
338                                  nbd_yank, bs);
339         object_unref(OBJECT(s->ioc));
340         s->ioc = NULL;
341 
342         return ret;
343     }
344 
345     qio_channel_set_blocking(s->ioc, false, NULL);
346     qio_channel_attach_aio_context(s->ioc, bdrv_get_aio_context(bs));
347 
348     /* successfully connected */
349     s->state = NBD_CLIENT_CONNECTED;
350     qemu_co_queue_restart_all(&s->free_sema);
351 
352     return 0;
353 }
354 
355 /* called under s->send_mutex */
356 static coroutine_fn void nbd_reconnect_attempt(BDRVNBDState *s)
357 {
358     assert(nbd_client_connecting(s));
359     assert(s->in_flight == 0);
360 
361     if (nbd_client_connecting_wait(s) && s->reconnect_delay &&
362         !s->reconnect_delay_timer)
363     {
364         /*
365          * It's first reconnect attempt after switching to
366          * NBD_CLIENT_CONNECTING_WAIT
367          */
368         reconnect_delay_timer_init(s,
369             qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
370             s->reconnect_delay * NANOSECONDS_PER_SECOND);
371     }
372 
373     /*
374      * Now we are sure that nobody is accessing the channel, and no one will
375      * try until we set the state to CONNECTED.
376      */
377 
378     /* Finalize previous connection if any */
379     if (s->ioc) {
380         qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc));
381         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
382                                  nbd_yank, s->bs);
383         object_unref(OBJECT(s->ioc));
384         s->ioc = NULL;
385     }
386 
387     nbd_co_do_establish_connection(s->bs, NULL);
388 
389     /*
390      * The reconnect attempt is done (maybe successfully, maybe not), so
391      * we no longer need this timer.  Delete it so it will not outlive
392      * this I/O request (so draining removes all timers).
393      */
394     reconnect_delay_timer_del(s);
395 }
396 
397 static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t handle)
398 {
399     int ret;
400     uint64_t ind = HANDLE_TO_INDEX(s, handle), ind2;
401     QEMU_LOCK_GUARD(&s->receive_mutex);
402 
403     while (true) {
404         if (s->reply.handle == handle) {
405             /* We are done */
406             return 0;
407         }
408 
409         if (!nbd_client_connected(s)) {
410             return -EIO;
411         }
412 
413         if (s->reply.handle != 0) {
414             /*
415              * Some other request is being handled now. It should already be
416              * woken by whoever set s->reply.handle (or never wait in this
417              * yield). So, we should not wake it here.
418              */
419             ind2 = HANDLE_TO_INDEX(s, s->reply.handle);
420             assert(!s->requests[ind2].receiving);
421 
422             s->requests[ind].receiving = true;
423             qemu_co_mutex_unlock(&s->receive_mutex);
424 
425             qemu_coroutine_yield();
426             /*
427              * We may be woken for 3 reasons:
428              * 1. From this function, executing in parallel coroutine, when our
429              *    handle is received.
430              * 2. From nbd_channel_error(), when connection is lost.
431              * 3. From nbd_co_receive_one_chunk(), when previous request is
432              *    finished and s->reply.handle set to 0.
433              * Anyway, it's OK to lock the mutex and go to the next iteration.
434              */
435 
436             qemu_co_mutex_lock(&s->receive_mutex);
437             assert(!s->requests[ind].receiving);
438             continue;
439         }
440 
441         /* We are under mutex and handle is 0. We have to do the dirty work. */
442         assert(s->reply.handle == 0);
443         ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, NULL);
444         if (ret <= 0) {
445             ret = ret ? ret : -EIO;
446             nbd_channel_error(s, ret);
447             return ret;
448         }
449         if (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply) {
450             nbd_channel_error(s, -EINVAL);
451             return -EINVAL;
452         }
453         if (s->reply.handle == handle) {
454             /* We are done */
455             return 0;
456         }
457         ind2 = HANDLE_TO_INDEX(s, s->reply.handle);
458         if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].reply_possible) {
459             nbd_channel_error(s, -EINVAL);
460             return -EINVAL;
461         }
462         nbd_recv_coroutine_wake_one(&s->requests[ind2]);
463     }
464 }
465 
466 static int nbd_co_send_request(BlockDriverState *bs,
467                                NBDRequest *request,
468                                QEMUIOVector *qiov)
469 {
470     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
471     int rc, i = -1;
472 
473     qemu_co_mutex_lock(&s->send_mutex);
474 
475     while (s->in_flight == MAX_NBD_REQUESTS ||
476            (!nbd_client_connected(s) && s->in_flight > 0))
477     {
478         qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
479     }
480 
481     if (nbd_client_connecting(s)) {
482         nbd_reconnect_attempt(s);
483     }
484 
485     if (!nbd_client_connected(s)) {
486         rc = -EIO;
487         goto err;
488     }
489 
490     s->in_flight++;
491 
492     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
493         if (s->requests[i].coroutine == NULL) {
494             break;
495         }
496     }
497 
498     g_assert(qemu_in_coroutine());
499     assert(i < MAX_NBD_REQUESTS);
500 
501     s->requests[i].coroutine = qemu_coroutine_self();
502     s->requests[i].offset = request->from;
503     s->requests[i].receiving = false;
504     s->requests[i].reply_possible = true;
505 
506     request->handle = INDEX_TO_HANDLE(s, i);
507 
508     assert(s->ioc);
509 
510     if (qiov) {
511         qio_channel_set_cork(s->ioc, true);
512         rc = nbd_send_request(s->ioc, request);
513         if (nbd_client_connected(s) && rc >= 0) {
514             if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
515                                        NULL) < 0) {
516                 rc = -EIO;
517             }
518         } else if (rc >= 0) {
519             rc = -EIO;
520         }
521         qio_channel_set_cork(s->ioc, false);
522     } else {
523         rc = nbd_send_request(s->ioc, request);
524     }
525 
526 err:
527     if (rc < 0) {
528         nbd_channel_error(s, rc);
529         if (i != -1) {
530             s->requests[i].coroutine = NULL;
531             s->in_flight--;
532             qemu_co_queue_next(&s->free_sema);
533         }
534     }
535     qemu_co_mutex_unlock(&s->send_mutex);
536     return rc;
537 }
538 
539 static inline uint16_t payload_advance16(uint8_t **payload)
540 {
541     *payload += 2;
542     return lduw_be_p(*payload - 2);
543 }
544 
545 static inline uint32_t payload_advance32(uint8_t **payload)
546 {
547     *payload += 4;
548     return ldl_be_p(*payload - 4);
549 }
550 
551 static inline uint64_t payload_advance64(uint8_t **payload)
552 {
553     *payload += 8;
554     return ldq_be_p(*payload - 8);
555 }
556 
557 static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
558                                          NBDStructuredReplyChunk *chunk,
559                                          uint8_t *payload, uint64_t orig_offset,
560                                          QEMUIOVector *qiov, Error **errp)
561 {
562     uint64_t offset;
563     uint32_t hole_size;
564 
565     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
566         error_setg(errp, "Protocol error: invalid payload for "
567                          "NBD_REPLY_TYPE_OFFSET_HOLE");
568         return -EINVAL;
569     }
570 
571     offset = payload_advance64(&payload);
572     hole_size = payload_advance32(&payload);
573 
574     if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
575         offset > orig_offset + qiov->size - hole_size) {
576         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
577                          " region");
578         return -EINVAL;
579     }
580     if (s->info.min_block &&
581         !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
582         trace_nbd_structured_read_compliance("hole");
583     }
584 
585     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
586 
587     return 0;
588 }
589 
590 /*
591  * nbd_parse_blockstatus_payload
592  * Based on our request, we expect only one extent in reply, for the
593  * base:allocation context.
594  */
595 static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
596                                          NBDStructuredReplyChunk *chunk,
597                                          uint8_t *payload, uint64_t orig_length,
598                                          NBDExtent *extent, Error **errp)
599 {
600     uint32_t context_id;
601 
602     /* The server succeeded, so it must have sent [at least] one extent */
603     if (chunk->length < sizeof(context_id) + sizeof(*extent)) {
604         error_setg(errp, "Protocol error: invalid payload for "
605                          "NBD_REPLY_TYPE_BLOCK_STATUS");
606         return -EINVAL;
607     }
608 
609     context_id = payload_advance32(&payload);
610     if (s->info.context_id != context_id) {
611         error_setg(errp, "Protocol error: unexpected context id %d for "
612                          "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
613                          "id is %d", context_id,
614                          s->info.context_id);
615         return -EINVAL;
616     }
617 
618     extent->length = payload_advance32(&payload);
619     extent->flags = payload_advance32(&payload);
620 
621     if (extent->length == 0) {
622         error_setg(errp, "Protocol error: server sent status chunk with "
623                    "zero length");
624         return -EINVAL;
625     }
626 
627     /*
628      * A server sending unaligned block status is in violation of the
629      * protocol, but as qemu-nbd 3.1 is such a server (at least for
630      * POSIX files that are not a multiple of 512 bytes, since qemu
631      * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
632      * still sees an implicit hole beyond the real EOF), it's nicer to
633      * work around the misbehaving server. If the request included
634      * more than the final unaligned block, truncate it back to an
635      * aligned result; if the request was only the final block, round
636      * up to the full block and change the status to fully-allocated
637      * (always a safe status, even if it loses information).
638      */
639     if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
640                                                    s->info.min_block)) {
641         trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
642         if (extent->length > s->info.min_block) {
643             extent->length = QEMU_ALIGN_DOWN(extent->length,
644                                              s->info.min_block);
645         } else {
646             extent->length = s->info.min_block;
647             extent->flags = 0;
648         }
649     }
650 
651     /*
652      * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
653      * sent us any more than one extent, nor should it have included
654      * status beyond our request in that extent. However, it's easy
655      * enough to ignore the server's noncompliance without killing the
656      * connection; just ignore trailing extents, and clamp things to
657      * the length of our request.
658      */
659     if (chunk->length > sizeof(context_id) + sizeof(*extent)) {
660         trace_nbd_parse_blockstatus_compliance("more than one extent");
661     }
662     if (extent->length > orig_length) {
663         extent->length = orig_length;
664         trace_nbd_parse_blockstatus_compliance("extent length too large");
665     }
666 
667     /*
668      * HACK: if we are using x-dirty-bitmaps to access
669      * qemu:allocation-depth, treat all depths > 2 the same as 2,
670      * since nbd_client_co_block_status is only expecting the low two
671      * bits to be set.
672      */
673     if (s->alloc_depth && extent->flags > 2) {
674         extent->flags = 2;
675     }
676 
677     return 0;
678 }
679 
680 /*
681  * nbd_parse_error_payload
682  * on success @errp contains message describing nbd error reply
683  */
684 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
685                                    uint8_t *payload, int *request_ret,
686                                    Error **errp)
687 {
688     uint32_t error;
689     uint16_t message_size;
690 
691     assert(chunk->type & (1 << 15));
692 
693     if (chunk->length < sizeof(error) + sizeof(message_size)) {
694         error_setg(errp,
695                    "Protocol error: invalid payload for structured error");
696         return -EINVAL;
697     }
698 
699     error = nbd_errno_to_system_errno(payload_advance32(&payload));
700     if (error == 0) {
701         error_setg(errp, "Protocol error: server sent structured error chunk "
702                          "with error = 0");
703         return -EINVAL;
704     }
705 
706     *request_ret = -error;
707     message_size = payload_advance16(&payload);
708 
709     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
710         error_setg(errp, "Protocol error: server sent structured error chunk "
711                          "with incorrect message size");
712         return -EINVAL;
713     }
714 
715     /* TODO: Add a trace point to mention the server complaint */
716 
717     /* TODO handle ERROR_OFFSET */
718 
719     return 0;
720 }
721 
722 static int nbd_co_receive_offset_data_payload(BDRVNBDState *s,
723                                               uint64_t orig_offset,
724                                               QEMUIOVector *qiov, Error **errp)
725 {
726     QEMUIOVector sub_qiov;
727     uint64_t offset;
728     size_t data_size;
729     int ret;
730     NBDStructuredReplyChunk *chunk = &s->reply.structured;
731 
732     assert(nbd_reply_is_structured(&s->reply));
733 
734     /* The NBD spec requires at least one byte of payload */
735     if (chunk->length <= sizeof(offset)) {
736         error_setg(errp, "Protocol error: invalid payload for "
737                          "NBD_REPLY_TYPE_OFFSET_DATA");
738         return -EINVAL;
739     }
740 
741     if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
742         return -EIO;
743     }
744 
745     data_size = chunk->length - sizeof(offset);
746     assert(data_size);
747     if (offset < orig_offset || data_size > qiov->size ||
748         offset > orig_offset + qiov->size - data_size) {
749         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
750                          " region");
751         return -EINVAL;
752     }
753     if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
754         trace_nbd_structured_read_compliance("data");
755     }
756 
757     qemu_iovec_init(&sub_qiov, qiov->niov);
758     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
759     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
760     qemu_iovec_destroy(&sub_qiov);
761 
762     return ret < 0 ? -EIO : 0;
763 }
764 
765 #define NBD_MAX_MALLOC_PAYLOAD 1000
766 static coroutine_fn int nbd_co_receive_structured_payload(
767         BDRVNBDState *s, void **payload, Error **errp)
768 {
769     int ret;
770     uint32_t len;
771 
772     assert(nbd_reply_is_structured(&s->reply));
773 
774     len = s->reply.structured.length;
775 
776     if (len == 0) {
777         return 0;
778     }
779 
780     if (payload == NULL) {
781         error_setg(errp, "Unexpected structured payload");
782         return -EINVAL;
783     }
784 
785     if (len > NBD_MAX_MALLOC_PAYLOAD) {
786         error_setg(errp, "Payload too large");
787         return -EINVAL;
788     }
789 
790     *payload = g_new(char, len);
791     ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
792     if (ret < 0) {
793         g_free(*payload);
794         *payload = NULL;
795         return ret;
796     }
797 
798     return 0;
799 }
800 
801 /*
802  * nbd_co_do_receive_one_chunk
803  * for simple reply:
804  *   set request_ret to received reply error
805  *   if qiov is not NULL: read payload to @qiov
806  * for structured reply chunk:
807  *   if error chunk: read payload, set @request_ret, do not set @payload
808  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
809  *   else: read payload to @payload
810  *
811  * If function fails, @errp contains corresponding error message, and the
812  * connection with the server is suspect.  If it returns 0, then the
813  * transaction succeeded (although @request_ret may be a negative errno
814  * corresponding to the server's error reply), and errp is unchanged.
815  */
816 static coroutine_fn int nbd_co_do_receive_one_chunk(
817         BDRVNBDState *s, uint64_t handle, bool only_structured,
818         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
819 {
820     int ret;
821     int i = HANDLE_TO_INDEX(s, handle);
822     void *local_payload = NULL;
823     NBDStructuredReplyChunk *chunk;
824 
825     if (payload) {
826         *payload = NULL;
827     }
828     *request_ret = 0;
829 
830     nbd_receive_replies(s, handle);
831     if (!nbd_client_connected(s)) {
832         error_setg(errp, "Connection closed");
833         return -EIO;
834     }
835     assert(s->ioc);
836 
837     assert(s->reply.handle == handle);
838 
839     if (nbd_reply_is_simple(&s->reply)) {
840         if (only_structured) {
841             error_setg(errp, "Protocol error: simple reply when structured "
842                              "reply chunk was expected");
843             return -EINVAL;
844         }
845 
846         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
847         if (*request_ret < 0 || !qiov) {
848             return 0;
849         }
850 
851         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
852                                      errp) < 0 ? -EIO : 0;
853     }
854 
855     /* handle structured reply chunk */
856     assert(s->info.structured_reply);
857     chunk = &s->reply.structured;
858 
859     if (chunk->type == NBD_REPLY_TYPE_NONE) {
860         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
861             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
862                        " NBD_REPLY_FLAG_DONE flag set");
863             return -EINVAL;
864         }
865         if (chunk->length) {
866             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
867                        " nonzero length");
868             return -EINVAL;
869         }
870         return 0;
871     }
872 
873     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
874         if (!qiov) {
875             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
876             return -EINVAL;
877         }
878 
879         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
880                                                   qiov, errp);
881     }
882 
883     if (nbd_reply_type_is_error(chunk->type)) {
884         payload = &local_payload;
885     }
886 
887     ret = nbd_co_receive_structured_payload(s, payload, errp);
888     if (ret < 0) {
889         return ret;
890     }
891 
892     if (nbd_reply_type_is_error(chunk->type)) {
893         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
894         g_free(local_payload);
895         return ret;
896     }
897 
898     return 0;
899 }
900 
901 /*
902  * nbd_co_receive_one_chunk
903  * Read reply, wake up connection_co and set s->quit if needed.
904  * Return value is a fatal error code or normal nbd reply error code
905  */
906 static coroutine_fn int nbd_co_receive_one_chunk(
907         BDRVNBDState *s, uint64_t handle, bool only_structured,
908         int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
909         Error **errp)
910 {
911     int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
912                                           request_ret, qiov, payload, errp);
913 
914     if (ret < 0) {
915         memset(reply, 0, sizeof(*reply));
916         nbd_channel_error(s, ret);
917     } else {
918         /* For assert at loop start in nbd_connection_entry */
919         *reply = s->reply;
920     }
921     s->reply.handle = 0;
922 
923     nbd_recv_coroutines_wake(s, false);
924 
925     return ret;
926 }
927 
928 typedef struct NBDReplyChunkIter {
929     int ret;
930     int request_ret;
931     Error *err;
932     bool done, only_structured;
933 } NBDReplyChunkIter;
934 
935 static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
936                                    int ret, Error **local_err)
937 {
938     assert(local_err && *local_err);
939     assert(ret < 0);
940 
941     if (!iter->ret) {
942         iter->ret = ret;
943         error_propagate(&iter->err, *local_err);
944     } else {
945         error_free(*local_err);
946     }
947 
948     *local_err = NULL;
949 }
950 
951 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
952 {
953     assert(ret < 0);
954 
955     if (!iter->request_ret) {
956         iter->request_ret = ret;
957     }
958 }
959 
960 /*
961  * NBD_FOREACH_REPLY_CHUNK
962  * The pointer stored in @payload requires g_free() to free it.
963  */
964 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
965                                 qiov, reply, payload) \
966     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
967          nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
968 
969 /*
970  * nbd_reply_chunk_iter_receive
971  * The pointer stored in @payload requires g_free() to free it.
972  */
973 static bool nbd_reply_chunk_iter_receive(BDRVNBDState *s,
974                                          NBDReplyChunkIter *iter,
975                                          uint64_t handle,
976                                          QEMUIOVector *qiov, NBDReply *reply,
977                                          void **payload)
978 {
979     int ret, request_ret;
980     NBDReply local_reply;
981     NBDStructuredReplyChunk *chunk;
982     Error *local_err = NULL;
983     if (!nbd_client_connected(s)) {
984         error_setg(&local_err, "Connection closed");
985         nbd_iter_channel_error(iter, -EIO, &local_err);
986         goto break_loop;
987     }
988 
989     if (iter->done) {
990         /* Previous iteration was last. */
991         goto break_loop;
992     }
993 
994     if (reply == NULL) {
995         reply = &local_reply;
996     }
997 
998     ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
999                                    &request_ret, qiov, reply, payload,
1000                                    &local_err);
1001     if (ret < 0) {
1002         nbd_iter_channel_error(iter, ret, &local_err);
1003     } else if (request_ret < 0) {
1004         nbd_iter_request_error(iter, request_ret);
1005     }
1006 
1007     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1008     if (nbd_reply_is_simple(reply) || !nbd_client_connected(s)) {
1009         goto break_loop;
1010     }
1011 
1012     chunk = &reply->structured;
1013     iter->only_structured = true;
1014 
1015     if (chunk->type == NBD_REPLY_TYPE_NONE) {
1016         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1017         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1018         goto break_loop;
1019     }
1020 
1021     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1022         /* This iteration is last. */
1023         iter->done = true;
1024     }
1025 
1026     /* Execute the loop body */
1027     return true;
1028 
1029 break_loop:
1030     s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
1031 
1032     qemu_co_mutex_lock(&s->send_mutex);
1033     s->in_flight--;
1034     qemu_co_queue_next(&s->free_sema);
1035     qemu_co_mutex_unlock(&s->send_mutex);
1036 
1037     return false;
1038 }
1039 
1040 static int nbd_co_receive_return_code(BDRVNBDState *s, uint64_t handle,
1041                                       int *request_ret, Error **errp)
1042 {
1043     NBDReplyChunkIter iter;
1044 
1045     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
1046         /* nbd_reply_chunk_iter_receive does all the work */
1047     }
1048 
1049     error_propagate(errp, iter.err);
1050     *request_ret = iter.request_ret;
1051     return iter.ret;
1052 }
1053 
1054 static int nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t handle,
1055                                         uint64_t offset, QEMUIOVector *qiov,
1056                                         int *request_ret, Error **errp)
1057 {
1058     NBDReplyChunkIter iter;
1059     NBDReply reply;
1060     void *payload = NULL;
1061     Error *local_err = NULL;
1062 
1063     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
1064                             qiov, &reply, &payload)
1065     {
1066         int ret;
1067         NBDStructuredReplyChunk *chunk = &reply.structured;
1068 
1069         assert(nbd_reply_is_structured(&reply));
1070 
1071         switch (chunk->type) {
1072         case NBD_REPLY_TYPE_OFFSET_DATA:
1073             /*
1074              * special cased in nbd_co_receive_one_chunk, data is already
1075              * in qiov
1076              */
1077             break;
1078         case NBD_REPLY_TYPE_OFFSET_HOLE:
1079             ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1080                                                 offset, qiov, &local_err);
1081             if (ret < 0) {
1082                 nbd_channel_error(s, ret);
1083                 nbd_iter_channel_error(&iter, ret, &local_err);
1084             }
1085             break;
1086         default:
1087             if (!nbd_reply_type_is_error(chunk->type)) {
1088                 /* not allowed reply type */
1089                 nbd_channel_error(s, -EINVAL);
1090                 error_setg(&local_err,
1091                            "Unexpected reply type: %d (%s) for CMD_READ",
1092                            chunk->type, nbd_reply_type_lookup(chunk->type));
1093                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1094             }
1095         }
1096 
1097         g_free(payload);
1098         payload = NULL;
1099     }
1100 
1101     error_propagate(errp, iter.err);
1102     *request_ret = iter.request_ret;
1103     return iter.ret;
1104 }
1105 
1106 static int nbd_co_receive_blockstatus_reply(BDRVNBDState *s,
1107                                             uint64_t handle, uint64_t length,
1108                                             NBDExtent *extent,
1109                                             int *request_ret, Error **errp)
1110 {
1111     NBDReplyChunkIter iter;
1112     NBDReply reply;
1113     void *payload = NULL;
1114     Error *local_err = NULL;
1115     bool received = false;
1116 
1117     assert(!extent->length);
1118     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) {
1119         int ret;
1120         NBDStructuredReplyChunk *chunk = &reply.structured;
1121 
1122         assert(nbd_reply_is_structured(&reply));
1123 
1124         switch (chunk->type) {
1125         case NBD_REPLY_TYPE_BLOCK_STATUS:
1126             if (received) {
1127                 nbd_channel_error(s, -EINVAL);
1128                 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1129                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1130             }
1131             received = true;
1132 
1133             ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1134                                                 payload, length, extent,
1135                                                 &local_err);
1136             if (ret < 0) {
1137                 nbd_channel_error(s, ret);
1138                 nbd_iter_channel_error(&iter, ret, &local_err);
1139             }
1140             break;
1141         default:
1142             if (!nbd_reply_type_is_error(chunk->type)) {
1143                 nbd_channel_error(s, -EINVAL);
1144                 error_setg(&local_err,
1145                            "Unexpected reply type: %d (%s) "
1146                            "for CMD_BLOCK_STATUS",
1147                            chunk->type, nbd_reply_type_lookup(chunk->type));
1148                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1149             }
1150         }
1151 
1152         g_free(payload);
1153         payload = NULL;
1154     }
1155 
1156     if (!extent->length && !iter.request_ret) {
1157         error_setg(&local_err, "Server did not reply with any status extents");
1158         nbd_iter_channel_error(&iter, -EIO, &local_err);
1159     }
1160 
1161     error_propagate(errp, iter.err);
1162     *request_ret = iter.request_ret;
1163     return iter.ret;
1164 }
1165 
1166 static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1167                           QEMUIOVector *write_qiov)
1168 {
1169     int ret, request_ret;
1170     Error *local_err = NULL;
1171     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1172 
1173     assert(request->type != NBD_CMD_READ);
1174     if (write_qiov) {
1175         assert(request->type == NBD_CMD_WRITE);
1176         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1177     } else {
1178         assert(request->type != NBD_CMD_WRITE);
1179     }
1180 
1181     do {
1182         ret = nbd_co_send_request(bs, request, write_qiov);
1183         if (ret < 0) {
1184             continue;
1185         }
1186 
1187         ret = nbd_co_receive_return_code(s, request->handle,
1188                                          &request_ret, &local_err);
1189         if (local_err) {
1190             trace_nbd_co_request_fail(request->from, request->len,
1191                                       request->handle, request->flags,
1192                                       request->type,
1193                                       nbd_cmd_lookup(request->type),
1194                                       ret, error_get_pretty(local_err));
1195             error_free(local_err);
1196             local_err = NULL;
1197         }
1198     } while (ret < 0 && nbd_client_connecting_wait(s));
1199 
1200     return ret ? ret : request_ret;
1201 }
1202 
1203 static int nbd_client_co_preadv(BlockDriverState *bs, int64_t offset,
1204                                 int64_t bytes, QEMUIOVector *qiov,
1205                                 BdrvRequestFlags flags)
1206 {
1207     int ret, request_ret;
1208     Error *local_err = NULL;
1209     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1210     NBDRequest request = {
1211         .type = NBD_CMD_READ,
1212         .from = offset,
1213         .len = bytes,
1214     };
1215 
1216     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1217     assert(!flags);
1218 
1219     if (!bytes) {
1220         return 0;
1221     }
1222     /*
1223      * Work around the fact that the block layer doesn't do
1224      * byte-accurate sizing yet - if the read exceeds the server's
1225      * advertised size because the block layer rounded size up, then
1226      * truncate the request to the server and tail-pad with zero.
1227      */
1228     if (offset >= s->info.size) {
1229         assert(bytes < BDRV_SECTOR_SIZE);
1230         qemu_iovec_memset(qiov, 0, 0, bytes);
1231         return 0;
1232     }
1233     if (offset + bytes > s->info.size) {
1234         uint64_t slop = offset + bytes - s->info.size;
1235 
1236         assert(slop < BDRV_SECTOR_SIZE);
1237         qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1238         request.len -= slop;
1239     }
1240 
1241     do {
1242         ret = nbd_co_send_request(bs, &request, NULL);
1243         if (ret < 0) {
1244             continue;
1245         }
1246 
1247         ret = nbd_co_receive_cmdread_reply(s, request.handle, offset, qiov,
1248                                            &request_ret, &local_err);
1249         if (local_err) {
1250             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1251                                       request.flags, request.type,
1252                                       nbd_cmd_lookup(request.type),
1253                                       ret, error_get_pretty(local_err));
1254             error_free(local_err);
1255             local_err = NULL;
1256         }
1257     } while (ret < 0 && nbd_client_connecting_wait(s));
1258 
1259     return ret ? ret : request_ret;
1260 }
1261 
1262 static int nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset,
1263                                  int64_t bytes, QEMUIOVector *qiov,
1264                                  BdrvRequestFlags flags)
1265 {
1266     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1267     NBDRequest request = {
1268         .type = NBD_CMD_WRITE,
1269         .from = offset,
1270         .len = bytes,
1271     };
1272 
1273     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1274     if (flags & BDRV_REQ_FUA) {
1275         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1276         request.flags |= NBD_CMD_FLAG_FUA;
1277     }
1278 
1279     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1280 
1281     if (!bytes) {
1282         return 0;
1283     }
1284     return nbd_co_request(bs, &request, qiov);
1285 }
1286 
1287 static int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1288                                        int64_t bytes, BdrvRequestFlags flags)
1289 {
1290     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1291     NBDRequest request = {
1292         .type = NBD_CMD_WRITE_ZEROES,
1293         .from = offset,
1294         .len = bytes,  /* .len is uint32_t actually */
1295     };
1296 
1297     assert(bytes <= UINT32_MAX); /* rely on max_pwrite_zeroes */
1298 
1299     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1300     if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1301         return -ENOTSUP;
1302     }
1303 
1304     if (flags & BDRV_REQ_FUA) {
1305         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1306         request.flags |= NBD_CMD_FLAG_FUA;
1307     }
1308     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1309         request.flags |= NBD_CMD_FLAG_NO_HOLE;
1310     }
1311     if (flags & BDRV_REQ_NO_FALLBACK) {
1312         assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1313         request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1314     }
1315 
1316     if (!bytes) {
1317         return 0;
1318     }
1319     return nbd_co_request(bs, &request, NULL);
1320 }
1321 
1322 static int nbd_client_co_flush(BlockDriverState *bs)
1323 {
1324     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1325     NBDRequest request = { .type = NBD_CMD_FLUSH };
1326 
1327     if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1328         return 0;
1329     }
1330 
1331     request.from = 0;
1332     request.len = 0;
1333 
1334     return nbd_co_request(bs, &request, NULL);
1335 }
1336 
1337 static int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset,
1338                                   int64_t bytes)
1339 {
1340     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1341     NBDRequest request = {
1342         .type = NBD_CMD_TRIM,
1343         .from = offset,
1344         .len = bytes, /* len is uint32_t */
1345     };
1346 
1347     assert(bytes <= UINT32_MAX); /* rely on max_pdiscard */
1348 
1349     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1350     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1351         return 0;
1352     }
1353 
1354     return nbd_co_request(bs, &request, NULL);
1355 }
1356 
1357 static int coroutine_fn nbd_client_co_block_status(
1358         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1359         int64_t *pnum, int64_t *map, BlockDriverState **file)
1360 {
1361     int ret, request_ret;
1362     NBDExtent extent = { 0 };
1363     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1364     Error *local_err = NULL;
1365 
1366     NBDRequest request = {
1367         .type = NBD_CMD_BLOCK_STATUS,
1368         .from = offset,
1369         .len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1370                    MIN(bytes, s->info.size - offset)),
1371         .flags = NBD_CMD_FLAG_REQ_ONE,
1372     };
1373 
1374     if (!s->info.base_allocation) {
1375         *pnum = bytes;
1376         *map = offset;
1377         *file = bs;
1378         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1379     }
1380 
1381     /*
1382      * Work around the fact that the block layer doesn't do
1383      * byte-accurate sizing yet - if the status request exceeds the
1384      * server's advertised size because the block layer rounded size
1385      * up, we truncated the request to the server (above), or are
1386      * called on just the hole.
1387      */
1388     if (offset >= s->info.size) {
1389         *pnum = bytes;
1390         assert(bytes < BDRV_SECTOR_SIZE);
1391         /* Intentionally don't report offset_valid for the hole */
1392         return BDRV_BLOCK_ZERO;
1393     }
1394 
1395     if (s->info.min_block) {
1396         assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1397     }
1398     do {
1399         ret = nbd_co_send_request(bs, &request, NULL);
1400         if (ret < 0) {
1401             continue;
1402         }
1403 
1404         ret = nbd_co_receive_blockstatus_reply(s, request.handle, bytes,
1405                                                &extent, &request_ret,
1406                                                &local_err);
1407         if (local_err) {
1408             trace_nbd_co_request_fail(request.from, request.len, request.handle,
1409                                       request.flags, request.type,
1410                                       nbd_cmd_lookup(request.type),
1411                                       ret, error_get_pretty(local_err));
1412             error_free(local_err);
1413             local_err = NULL;
1414         }
1415     } while (ret < 0 && nbd_client_connecting_wait(s));
1416 
1417     if (ret < 0 || request_ret < 0) {
1418         return ret ? ret : request_ret;
1419     }
1420 
1421     assert(extent.length);
1422     *pnum = extent.length;
1423     *map = offset;
1424     *file = bs;
1425     return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1426         (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1427         BDRV_BLOCK_OFFSET_VALID;
1428 }
1429 
1430 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1431                                      BlockReopenQueue *queue, Error **errp)
1432 {
1433     BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1434 
1435     if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1436         error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1437         return -EACCES;
1438     }
1439     return 0;
1440 }
1441 
1442 static void nbd_yank(void *opaque)
1443 {
1444     BlockDriverState *bs = opaque;
1445     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1446 
1447     qatomic_store_release(&s->state, NBD_CLIENT_QUIT);
1448     qio_channel_shutdown(QIO_CHANNEL(s->ioc), QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1449 }
1450 
1451 static void nbd_client_close(BlockDriverState *bs)
1452 {
1453     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1454     NBDRequest request = { .type = NBD_CMD_DISC };
1455 
1456     if (s->ioc) {
1457         nbd_send_request(s->ioc, &request);
1458     }
1459 
1460     nbd_teardown_connection(bs);
1461 }
1462 
1463 
1464 /*
1465  * Parse nbd_open options
1466  */
1467 
1468 static int nbd_parse_uri(const char *filename, QDict *options)
1469 {
1470     URI *uri;
1471     const char *p;
1472     QueryParams *qp = NULL;
1473     int ret = 0;
1474     bool is_unix;
1475 
1476     uri = uri_parse(filename);
1477     if (!uri) {
1478         return -EINVAL;
1479     }
1480 
1481     /* transport */
1482     if (!g_strcmp0(uri->scheme, "nbd")) {
1483         is_unix = false;
1484     } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1485         is_unix = false;
1486     } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1487         is_unix = true;
1488     } else {
1489         ret = -EINVAL;
1490         goto out;
1491     }
1492 
1493     p = uri->path ? uri->path : "";
1494     if (p[0] == '/') {
1495         p++;
1496     }
1497     if (p[0]) {
1498         qdict_put_str(options, "export", p);
1499     }
1500 
1501     qp = query_params_parse(uri->query);
1502     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1503         ret = -EINVAL;
1504         goto out;
1505     }
1506 
1507     if (is_unix) {
1508         /* nbd+unix:///export?socket=path */
1509         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1510             ret = -EINVAL;
1511             goto out;
1512         }
1513         qdict_put_str(options, "server.type", "unix");
1514         qdict_put_str(options, "server.path", qp->p[0].value);
1515     } else {
1516         QString *host;
1517         char *port_str;
1518 
1519         /* nbd[+tcp]://host[:port]/export */
1520         if (!uri->server) {
1521             ret = -EINVAL;
1522             goto out;
1523         }
1524 
1525         /* strip braces from literal IPv6 address */
1526         if (uri->server[0] == '[') {
1527             host = qstring_from_substr(uri->server, 1,
1528                                        strlen(uri->server) - 1);
1529         } else {
1530             host = qstring_from_str(uri->server);
1531         }
1532 
1533         qdict_put_str(options, "server.type", "inet");
1534         qdict_put(options, "server.host", host);
1535 
1536         port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1537         qdict_put_str(options, "server.port", port_str);
1538         g_free(port_str);
1539     }
1540 
1541 out:
1542     if (qp) {
1543         query_params_free(qp);
1544     }
1545     uri_free(uri);
1546     return ret;
1547 }
1548 
1549 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1550 {
1551     const QDictEntry *e;
1552 
1553     for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1554         if (!strcmp(e->key, "host") ||
1555             !strcmp(e->key, "port") ||
1556             !strcmp(e->key, "path") ||
1557             !strcmp(e->key, "export") ||
1558             strstart(e->key, "server.", NULL))
1559         {
1560             error_setg(errp, "Option '%s' cannot be used with a file name",
1561                        e->key);
1562             return true;
1563         }
1564     }
1565 
1566     return false;
1567 }
1568 
1569 static void nbd_parse_filename(const char *filename, QDict *options,
1570                                Error **errp)
1571 {
1572     g_autofree char *file = NULL;
1573     char *export_name;
1574     const char *host_spec;
1575     const char *unixpath;
1576 
1577     if (nbd_has_filename_options_conflict(options, errp)) {
1578         return;
1579     }
1580 
1581     if (strstr(filename, "://")) {
1582         int ret = nbd_parse_uri(filename, options);
1583         if (ret < 0) {
1584             error_setg(errp, "No valid URL specified");
1585         }
1586         return;
1587     }
1588 
1589     file = g_strdup(filename);
1590 
1591     export_name = strstr(file, EN_OPTSTR);
1592     if (export_name) {
1593         if (export_name[strlen(EN_OPTSTR)] == 0) {
1594             return;
1595         }
1596         export_name[0] = 0; /* truncate 'file' */
1597         export_name += strlen(EN_OPTSTR);
1598 
1599         qdict_put_str(options, "export", export_name);
1600     }
1601 
1602     /* extract the host_spec - fail if it's not nbd:... */
1603     if (!strstart(file, "nbd:", &host_spec)) {
1604         error_setg(errp, "File name string for NBD must start with 'nbd:'");
1605         return;
1606     }
1607 
1608     if (!*host_spec) {
1609         return;
1610     }
1611 
1612     /* are we a UNIX or TCP socket? */
1613     if (strstart(host_spec, "unix:", &unixpath)) {
1614         qdict_put_str(options, "server.type", "unix");
1615         qdict_put_str(options, "server.path", unixpath);
1616     } else {
1617         InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1618 
1619         if (inet_parse(addr, host_spec, errp)) {
1620             goto out_inet;
1621         }
1622 
1623         qdict_put_str(options, "server.type", "inet");
1624         qdict_put_str(options, "server.host", addr->host);
1625         qdict_put_str(options, "server.port", addr->port);
1626     out_inet:
1627         qapi_free_InetSocketAddress(addr);
1628     }
1629 }
1630 
1631 static bool nbd_process_legacy_socket_options(QDict *output_options,
1632                                               QemuOpts *legacy_opts,
1633                                               Error **errp)
1634 {
1635     const char *path = qemu_opt_get(legacy_opts, "path");
1636     const char *host = qemu_opt_get(legacy_opts, "host");
1637     const char *port = qemu_opt_get(legacy_opts, "port");
1638     const QDictEntry *e;
1639 
1640     if (!path && !host && !port) {
1641         return true;
1642     }
1643 
1644     for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1645     {
1646         if (strstart(e->key, "server.", NULL)) {
1647             error_setg(errp, "Cannot use 'server' and path/host/port at the "
1648                        "same time");
1649             return false;
1650         }
1651     }
1652 
1653     if (path && host) {
1654         error_setg(errp, "path and host may not be used at the same time");
1655         return false;
1656     } else if (path) {
1657         if (port) {
1658             error_setg(errp, "port may not be used without host");
1659             return false;
1660         }
1661 
1662         qdict_put_str(output_options, "server.type", "unix");
1663         qdict_put_str(output_options, "server.path", path);
1664     } else if (host) {
1665         qdict_put_str(output_options, "server.type", "inet");
1666         qdict_put_str(output_options, "server.host", host);
1667         qdict_put_str(output_options, "server.port",
1668                       port ?: stringify(NBD_DEFAULT_PORT));
1669     }
1670 
1671     return true;
1672 }
1673 
1674 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1675                                  Error **errp)
1676 {
1677     SocketAddress *saddr = NULL;
1678     QDict *addr = NULL;
1679     Visitor *iv = NULL;
1680 
1681     qdict_extract_subqdict(options, &addr, "server.");
1682     if (!qdict_size(addr)) {
1683         error_setg(errp, "NBD server address missing");
1684         goto done;
1685     }
1686 
1687     iv = qobject_input_visitor_new_flat_confused(addr, errp);
1688     if (!iv) {
1689         goto done;
1690     }
1691 
1692     if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1693         goto done;
1694     }
1695 
1696     if (socket_address_parse_named_fd(saddr, errp) < 0) {
1697         qapi_free_SocketAddress(saddr);
1698         saddr = NULL;
1699         goto done;
1700     }
1701 
1702 done:
1703     qobject_unref(addr);
1704     visit_free(iv);
1705     return saddr;
1706 }
1707 
1708 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1709 {
1710     Object *obj;
1711     QCryptoTLSCreds *creds;
1712 
1713     obj = object_resolve_path_component(
1714         object_get_objects_root(), id);
1715     if (!obj) {
1716         error_setg(errp, "No TLS credentials with id '%s'",
1717                    id);
1718         return NULL;
1719     }
1720     creds = (QCryptoTLSCreds *)
1721         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1722     if (!creds) {
1723         error_setg(errp, "Object with id '%s' is not TLS credentials",
1724                    id);
1725         return NULL;
1726     }
1727 
1728     if (!qcrypto_tls_creds_check_endpoint(creds,
1729                                           QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1730                                           errp)) {
1731         return NULL;
1732     }
1733     object_ref(obj);
1734     return creds;
1735 }
1736 
1737 
1738 static QemuOptsList nbd_runtime_opts = {
1739     .name = "nbd",
1740     .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1741     .desc = {
1742         {
1743             .name = "host",
1744             .type = QEMU_OPT_STRING,
1745             .help = "TCP host to connect to",
1746         },
1747         {
1748             .name = "port",
1749             .type = QEMU_OPT_STRING,
1750             .help = "TCP port to connect to",
1751         },
1752         {
1753             .name = "path",
1754             .type = QEMU_OPT_STRING,
1755             .help = "Unix socket path to connect to",
1756         },
1757         {
1758             .name = "export",
1759             .type = QEMU_OPT_STRING,
1760             .help = "Name of the NBD export to open",
1761         },
1762         {
1763             .name = "tls-creds",
1764             .type = QEMU_OPT_STRING,
1765             .help = "ID of the TLS credentials to use",
1766         },
1767         {
1768             .name = "x-dirty-bitmap",
1769             .type = QEMU_OPT_STRING,
1770             .help = "experimental: expose named dirty bitmap in place of "
1771                     "block status",
1772         },
1773         {
1774             .name = "reconnect-delay",
1775             .type = QEMU_OPT_NUMBER,
1776             .help = "On an unexpected disconnect, the nbd client tries to "
1777                     "connect again until succeeding or encountering a serious "
1778                     "error.  During the first @reconnect-delay seconds, all "
1779                     "requests are paused and will be rerun on a successful "
1780                     "reconnect. After that time, any delayed requests and all "
1781                     "future requests before a successful reconnect will "
1782                     "immediately fail. Default 0",
1783         },
1784         {
1785             .name = "open-timeout",
1786             .type = QEMU_OPT_NUMBER,
1787             .help = "In seconds. If zero, the nbd driver tries the connection "
1788                     "only once, and fails to open if the connection fails. "
1789                     "If non-zero, the nbd driver will repeat connection "
1790                     "attempts until successful or until @open-timeout seconds "
1791                     "have elapsed. Default 0",
1792         },
1793         { /* end of list */ }
1794     },
1795 };
1796 
1797 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1798                                Error **errp)
1799 {
1800     BDRVNBDState *s = bs->opaque;
1801     QemuOpts *opts;
1802     int ret = -EINVAL;
1803 
1804     opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1805     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1806         goto error;
1807     }
1808 
1809     /* Translate @host, @port, and @path to a SocketAddress */
1810     if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1811         goto error;
1812     }
1813 
1814     /* Pop the config into our state object. Exit if invalid. */
1815     s->saddr = nbd_config(s, options, errp);
1816     if (!s->saddr) {
1817         goto error;
1818     }
1819 
1820     s->export = g_strdup(qemu_opt_get(opts, "export"));
1821     if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1822         error_setg(errp, "export name too long to send to server");
1823         goto error;
1824     }
1825 
1826     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1827     if (s->tlscredsid) {
1828         s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1829         if (!s->tlscreds) {
1830             goto error;
1831         }
1832 
1833         /* TODO SOCKET_ADDRESS_KIND_FD where fd has AF_INET or AF_INET6 */
1834         if (s->saddr->type != SOCKET_ADDRESS_TYPE_INET) {
1835             error_setg(errp, "TLS only supported over IP sockets");
1836             goto error;
1837         }
1838         s->hostname = s->saddr->u.inet.host;
1839     }
1840 
1841     s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1842     if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1843         error_setg(errp, "x-dirty-bitmap query too long to send to server");
1844         goto error;
1845     }
1846 
1847     s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1848     s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1849 
1850     ret = 0;
1851 
1852  error:
1853     qemu_opts_del(opts);
1854     return ret;
1855 }
1856 
1857 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1858                     Error **errp)
1859 {
1860     int ret;
1861     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1862 
1863     s->bs = bs;
1864     qemu_co_mutex_init(&s->send_mutex);
1865     qemu_co_queue_init(&s->free_sema);
1866     qemu_co_mutex_init(&s->receive_mutex);
1867 
1868     if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1869         return -EEXIST;
1870     }
1871 
1872     ret = nbd_process_options(bs, options, errp);
1873     if (ret < 0) {
1874         goto fail;
1875     }
1876 
1877     s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1878                                         s->x_dirty_bitmap, s->tlscreds);
1879 
1880     if (s->open_timeout) {
1881         nbd_client_connection_enable_retry(s->conn);
1882         open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1883                         s->open_timeout * NANOSECONDS_PER_SECOND);
1884     }
1885 
1886     s->state = NBD_CLIENT_CONNECTING_WAIT;
1887     ret = nbd_do_establish_connection(bs, errp);
1888     if (ret < 0) {
1889         goto fail;
1890     }
1891 
1892     /*
1893      * The connect attempt is done, so we no longer need this timer.
1894      * Delete it, because we do not want it to be around when this node
1895      * is drained or closed.
1896      */
1897     open_timer_del(s);
1898 
1899     nbd_client_connection_enable_retry(s->conn);
1900 
1901     return 0;
1902 
1903 fail:
1904     open_timer_del(s);
1905     nbd_clear_bdrvstate(bs);
1906     return ret;
1907 }
1908 
1909 static int nbd_co_flush(BlockDriverState *bs)
1910 {
1911     return nbd_client_co_flush(bs);
1912 }
1913 
1914 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1915 {
1916     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1917     uint32_t min = s->info.min_block;
1918     uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1919 
1920     /*
1921      * If the server did not advertise an alignment:
1922      * - a size that is not sector-aligned implies that an alignment
1923      *   of 1 can be used to access those tail bytes
1924      * - advertisement of block status requires an alignment of 1, so
1925      *   that we don't violate block layer constraints that block
1926      *   status is always aligned (as we can't control whether the
1927      *   server will report sub-sector extents, such as a hole at EOF
1928      *   on an unaligned POSIX file)
1929      * - otherwise, assume the server is so old that we are safer avoiding
1930      *   sub-sector requests
1931      */
1932     if (!min) {
1933         min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1934                s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1935     }
1936 
1937     bs->bl.request_alignment = min;
1938     bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1939     bs->bl.max_pwrite_zeroes = max;
1940     bs->bl.max_transfer = max;
1941 
1942     if (s->info.opt_block &&
1943         s->info.opt_block > bs->bl.opt_transfer) {
1944         bs->bl.opt_transfer = s->info.opt_block;
1945     }
1946 }
1947 
1948 static void nbd_close(BlockDriverState *bs)
1949 {
1950     nbd_client_close(bs);
1951     nbd_clear_bdrvstate(bs);
1952 }
1953 
1954 /*
1955  * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1956  * to a smaller size with exact=false, there is no reason to fail the
1957  * operation.
1958  *
1959  * Preallocation mode is ignored since it does not seems useful to fail when
1960  * we never change anything.
1961  */
1962 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
1963                                         bool exact, PreallocMode prealloc,
1964                                         BdrvRequestFlags flags, Error **errp)
1965 {
1966     BDRVNBDState *s = bs->opaque;
1967 
1968     if (offset != s->info.size && exact) {
1969         error_setg(errp, "Cannot resize NBD nodes");
1970         return -ENOTSUP;
1971     }
1972 
1973     if (offset > s->info.size) {
1974         error_setg(errp, "Cannot grow NBD nodes");
1975         return -EINVAL;
1976     }
1977 
1978     return 0;
1979 }
1980 
1981 static int64_t nbd_getlength(BlockDriverState *bs)
1982 {
1983     BDRVNBDState *s = bs->opaque;
1984 
1985     return s->info.size;
1986 }
1987 
1988 static void nbd_refresh_filename(BlockDriverState *bs)
1989 {
1990     BDRVNBDState *s = bs->opaque;
1991     const char *host = NULL, *port = NULL, *path = NULL;
1992     size_t len = 0;
1993 
1994     if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1995         const InetSocketAddress *inet = &s->saddr->u.inet;
1996         if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
1997             host = inet->host;
1998             port = inet->port;
1999         }
2000     } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2001         path = s->saddr->u.q_unix.path;
2002     } /* else can't represent as pseudo-filename */
2003 
2004     if (path && s->export) {
2005         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2006                        "nbd+unix:///%s?socket=%s", s->export, path);
2007     } else if (path && !s->export) {
2008         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2009                        "nbd+unix://?socket=%s", path);
2010     } else if (host && s->export) {
2011         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2012                        "nbd://%s:%s/%s", host, port, s->export);
2013     } else if (host && !s->export) {
2014         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2015                        "nbd://%s:%s", host, port);
2016     }
2017     if (len >= sizeof(bs->exact_filename)) {
2018         /* Name is too long to represent exactly, so leave it empty. */
2019         bs->exact_filename[0] = '\0';
2020     }
2021 }
2022 
2023 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2024 {
2025     /* The generic bdrv_dirname() implementation is able to work out some
2026      * directory name for NBD nodes, but that would be wrong. So far there is no
2027      * specification for how "export paths" would work, so NBD does not have
2028      * directory names. */
2029     error_setg(errp, "Cannot generate a base directory for NBD nodes");
2030     return NULL;
2031 }
2032 
2033 static const char *const nbd_strong_runtime_opts[] = {
2034     "path",
2035     "host",
2036     "port",
2037     "export",
2038     "tls-creds",
2039     "server.",
2040 
2041     NULL
2042 };
2043 
2044 static void nbd_cancel_in_flight(BlockDriverState *bs)
2045 {
2046     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2047 
2048     reconnect_delay_timer_del(s);
2049 
2050     if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2051         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2052         qemu_co_queue_restart_all(&s->free_sema);
2053     }
2054 
2055     nbd_co_establish_connection_cancel(s->conn);
2056 }
2057 
2058 static void nbd_attach_aio_context(BlockDriverState *bs,
2059                                    AioContext *new_context)
2060 {
2061     BDRVNBDState *s = bs->opaque;
2062 
2063     /* The open_timer is used only during nbd_open() */
2064     assert(!s->open_timer);
2065 
2066     /*
2067      * The reconnect_delay_timer is scheduled in I/O paths when the
2068      * connection is lost, to cancel the reconnection attempt after a
2069      * given time.  Once this attempt is done (successfully or not),
2070      * nbd_reconnect_attempt() ensures the timer is deleted before the
2071      * respective I/O request is resumed.
2072      * Since the AioContext can only be changed when a node is drained,
2073      * the reconnect_delay_timer cannot be active here.
2074      */
2075     assert(!s->reconnect_delay_timer);
2076 
2077     if (s->ioc) {
2078         qio_channel_attach_aio_context(s->ioc, new_context);
2079     }
2080 }
2081 
2082 static void nbd_detach_aio_context(BlockDriverState *bs)
2083 {
2084     BDRVNBDState *s = bs->opaque;
2085 
2086     assert(!s->open_timer);
2087     assert(!s->reconnect_delay_timer);
2088 
2089     if (s->ioc) {
2090         qio_channel_detach_aio_context(s->ioc);
2091     }
2092 }
2093 
2094 static BlockDriver bdrv_nbd = {
2095     .format_name                = "nbd",
2096     .protocol_name              = "nbd",
2097     .instance_size              = sizeof(BDRVNBDState),
2098     .bdrv_parse_filename        = nbd_parse_filename,
2099     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2100     .create_opts                = &bdrv_create_opts_simple,
2101     .bdrv_file_open             = nbd_open,
2102     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2103     .bdrv_co_preadv             = nbd_client_co_preadv,
2104     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2105     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2106     .bdrv_close                 = nbd_close,
2107     .bdrv_co_flush_to_os        = nbd_co_flush,
2108     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2109     .bdrv_refresh_limits        = nbd_refresh_limits,
2110     .bdrv_co_truncate           = nbd_co_truncate,
2111     .bdrv_getlength             = nbd_getlength,
2112     .bdrv_refresh_filename      = nbd_refresh_filename,
2113     .bdrv_co_block_status       = nbd_client_co_block_status,
2114     .bdrv_dirname               = nbd_dirname,
2115     .strong_runtime_opts        = nbd_strong_runtime_opts,
2116     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2117 
2118     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2119     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2120 };
2121 
2122 static BlockDriver bdrv_nbd_tcp = {
2123     .format_name                = "nbd",
2124     .protocol_name              = "nbd+tcp",
2125     .instance_size              = sizeof(BDRVNBDState),
2126     .bdrv_parse_filename        = nbd_parse_filename,
2127     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2128     .create_opts                = &bdrv_create_opts_simple,
2129     .bdrv_file_open             = nbd_open,
2130     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2131     .bdrv_co_preadv             = nbd_client_co_preadv,
2132     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2133     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2134     .bdrv_close                 = nbd_close,
2135     .bdrv_co_flush_to_os        = nbd_co_flush,
2136     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2137     .bdrv_refresh_limits        = nbd_refresh_limits,
2138     .bdrv_co_truncate           = nbd_co_truncate,
2139     .bdrv_getlength             = nbd_getlength,
2140     .bdrv_refresh_filename      = nbd_refresh_filename,
2141     .bdrv_co_block_status       = nbd_client_co_block_status,
2142     .bdrv_dirname               = nbd_dirname,
2143     .strong_runtime_opts        = nbd_strong_runtime_opts,
2144     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2145 
2146     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2147     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2148 };
2149 
2150 static BlockDriver bdrv_nbd_unix = {
2151     .format_name                = "nbd",
2152     .protocol_name              = "nbd+unix",
2153     .instance_size              = sizeof(BDRVNBDState),
2154     .bdrv_parse_filename        = nbd_parse_filename,
2155     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2156     .create_opts                = &bdrv_create_opts_simple,
2157     .bdrv_file_open             = nbd_open,
2158     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2159     .bdrv_co_preadv             = nbd_client_co_preadv,
2160     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2161     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2162     .bdrv_close                 = nbd_close,
2163     .bdrv_co_flush_to_os        = nbd_co_flush,
2164     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2165     .bdrv_refresh_limits        = nbd_refresh_limits,
2166     .bdrv_co_truncate           = nbd_co_truncate,
2167     .bdrv_getlength             = nbd_getlength,
2168     .bdrv_refresh_filename      = nbd_refresh_filename,
2169     .bdrv_co_block_status       = nbd_client_co_block_status,
2170     .bdrv_dirname               = nbd_dirname,
2171     .strong_runtime_opts        = nbd_strong_runtime_opts,
2172     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2173 
2174     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2175     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2176 };
2177 
2178 static void bdrv_nbd_init(void)
2179 {
2180     bdrv_register(&bdrv_nbd);
2181     bdrv_register(&bdrv_nbd_tcp);
2182     bdrv_register(&bdrv_nbd_unix);
2183 }
2184 
2185 block_init(bdrv_nbd_init);
2186