xref: /openbmc/qemu/block/nbd.c (revision afb81fe8)
1 /*
2  * QEMU Block driver for NBD
3  *
4  * Copyright (c) 2019 Virtuozzo International GmbH.
5  * Copyright Red Hat
6  * Copyright (C) 2008 Bull S.A.S.
7  *     Author: Laurent Vivier <Laurent.Vivier@bull.net>
8  *
9  * Some parts:
10  *    Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this software and associated documentation files (the "Software"), to deal
14  * in the Software without restriction, including without limitation the rights
15  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
16  * copies of the Software, and to permit persons to whom the Software is
17  * furnished to do so, subject to the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
25  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
27  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
28  * THE SOFTWARE.
29  */
30 
31 #include "qemu/osdep.h"
32 
33 #include "trace.h"
34 #include "qemu/uri.h"
35 #include "qemu/option.h"
36 #include "qemu/cutils.h"
37 #include "qemu/main-loop.h"
38 
39 #include "qapi/qapi-visit-sockets.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi/clone-visitor.h"
42 
43 #include "block/qdict.h"
44 #include "block/nbd.h"
45 #include "block/block_int.h"
46 #include "block/coroutines.h"
47 
48 #include "qemu/yank.h"
49 
50 #define EN_OPTSTR ":exportname="
51 #define MAX_NBD_REQUESTS    16
52 
53 #define COOKIE_TO_INDEX(cookie) ((cookie) - 1)
54 #define INDEX_TO_COOKIE(index)  ((index) + 1)
55 
56 typedef struct {
57     Coroutine *coroutine;
58     uint64_t offset;        /* original offset of the request */
59     bool receiving;         /* sleeping in the yield in nbd_receive_replies */
60 } NBDClientRequest;
61 
62 typedef enum NBDClientState {
63     NBD_CLIENT_CONNECTING_WAIT,
64     NBD_CLIENT_CONNECTING_NOWAIT,
65     NBD_CLIENT_CONNECTED,
66     NBD_CLIENT_QUIT
67 } NBDClientState;
68 
69 typedef struct BDRVNBDState {
70     QIOChannel *ioc; /* The current I/O channel */
71     NBDExportInfo info;
72 
73     /*
74      * Protects state, free_sema, in_flight, requests[].coroutine,
75      * reconnect_delay_timer.
76      */
77     QemuMutex requests_lock;
78     NBDClientState state;
79     CoQueue free_sema;
80     unsigned in_flight;
81     NBDClientRequest requests[MAX_NBD_REQUESTS];
82     QEMUTimer *reconnect_delay_timer;
83 
84     /* Protects sending data on the socket.  */
85     CoMutex send_mutex;
86 
87     /*
88      * Protects receiving reply headers from the socket, as well as the
89      * fields reply and requests[].receiving
90      */
91     CoMutex receive_mutex;
92     NBDReply reply;
93 
94     QEMUTimer *open_timer;
95 
96     BlockDriverState *bs;
97 
98     /* Connection parameters */
99     uint32_t reconnect_delay;
100     uint32_t open_timeout;
101     SocketAddress *saddr;
102     char *export;
103     char *tlscredsid;
104     QCryptoTLSCreds *tlscreds;
105     char *tlshostname;
106     char *x_dirty_bitmap;
107     bool alloc_depth;
108 
109     NBDClientConnection *conn;
110 } BDRVNBDState;
111 
112 static void nbd_yank(void *opaque);
113 
114 static void nbd_clear_bdrvstate(BlockDriverState *bs)
115 {
116     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
117 
118     nbd_client_connection_release(s->conn);
119     s->conn = NULL;
120 
121     yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name));
122 
123     /* Must not leave timers behind that would access freed data */
124     assert(!s->reconnect_delay_timer);
125     assert(!s->open_timer);
126 
127     object_unref(OBJECT(s->tlscreds));
128     qapi_free_SocketAddress(s->saddr);
129     s->saddr = NULL;
130     g_free(s->export);
131     s->export = NULL;
132     g_free(s->tlscredsid);
133     s->tlscredsid = NULL;
134     g_free(s->tlshostname);
135     s->tlshostname = NULL;
136     g_free(s->x_dirty_bitmap);
137     s->x_dirty_bitmap = NULL;
138 }
139 
140 /* Called with s->receive_mutex taken.  */
141 static bool coroutine_fn nbd_recv_coroutine_wake_one(NBDClientRequest *req)
142 {
143     if (req->receiving) {
144         req->receiving = false;
145         aio_co_wake(req->coroutine);
146         return true;
147     }
148 
149     return false;
150 }
151 
152 static void coroutine_fn nbd_recv_coroutines_wake(BDRVNBDState *s)
153 {
154     int i;
155 
156     QEMU_LOCK_GUARD(&s->receive_mutex);
157     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
158         if (nbd_recv_coroutine_wake_one(&s->requests[i])) {
159             return;
160         }
161     }
162 }
163 
164 /* Called with s->requests_lock held.  */
165 static void coroutine_fn nbd_channel_error_locked(BDRVNBDState *s, int ret)
166 {
167     if (s->state == NBD_CLIENT_CONNECTED) {
168         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
169     }
170 
171     if (ret == -EIO) {
172         if (s->state == NBD_CLIENT_CONNECTED) {
173             s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT :
174                                             NBD_CLIENT_CONNECTING_NOWAIT;
175         }
176     } else {
177         s->state = NBD_CLIENT_QUIT;
178     }
179 }
180 
181 static void coroutine_fn nbd_channel_error(BDRVNBDState *s, int ret)
182 {
183     QEMU_LOCK_GUARD(&s->requests_lock);
184     nbd_channel_error_locked(s, ret);
185 }
186 
187 static void reconnect_delay_timer_del(BDRVNBDState *s)
188 {
189     if (s->reconnect_delay_timer) {
190         timer_free(s->reconnect_delay_timer);
191         s->reconnect_delay_timer = NULL;
192     }
193 }
194 
195 static void reconnect_delay_timer_cb(void *opaque)
196 {
197     BDRVNBDState *s = opaque;
198 
199     reconnect_delay_timer_del(s);
200     WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
201         if (s->state != NBD_CLIENT_CONNECTING_WAIT) {
202             return;
203         }
204         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
205     }
206     nbd_co_establish_connection_cancel(s->conn);
207 }
208 
209 static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
210 {
211     assert(!s->reconnect_delay_timer);
212     s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
213                                              QEMU_CLOCK_REALTIME,
214                                              SCALE_NS,
215                                              reconnect_delay_timer_cb, s);
216     timer_mod(s->reconnect_delay_timer, expire_time_ns);
217 }
218 
219 static void nbd_teardown_connection(BlockDriverState *bs)
220 {
221     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
222 
223     assert(!s->in_flight);
224 
225     if (s->ioc) {
226         qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
227         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
228                                  nbd_yank, s->bs);
229         object_unref(OBJECT(s->ioc));
230         s->ioc = NULL;
231     }
232 
233     WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
234         s->state = NBD_CLIENT_QUIT;
235     }
236 }
237 
238 static void open_timer_del(BDRVNBDState *s)
239 {
240     if (s->open_timer) {
241         timer_free(s->open_timer);
242         s->open_timer = NULL;
243     }
244 }
245 
246 static void open_timer_cb(void *opaque)
247 {
248     BDRVNBDState *s = opaque;
249 
250     nbd_co_establish_connection_cancel(s->conn);
251     open_timer_del(s);
252 }
253 
254 static void open_timer_init(BDRVNBDState *s, uint64_t expire_time_ns)
255 {
256     assert(!s->open_timer);
257     s->open_timer = aio_timer_new(bdrv_get_aio_context(s->bs),
258                                   QEMU_CLOCK_REALTIME,
259                                   SCALE_NS,
260                                   open_timer_cb, s);
261     timer_mod(s->open_timer, expire_time_ns);
262 }
263 
264 static bool nbd_client_will_reconnect(BDRVNBDState *s)
265 {
266     /*
267      * Called only after a socket error, so this is not performance sensitive.
268      */
269     QEMU_LOCK_GUARD(&s->requests_lock);
270     return s->state == NBD_CLIENT_CONNECTING_WAIT;
271 }
272 
273 /*
274  * Update @bs with information learned during a completed negotiation process.
275  * Return failure if the server's advertised options are incompatible with the
276  * client's needs.
277  */
278 static int nbd_handle_updated_info(BlockDriverState *bs, Error **errp)
279 {
280     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
281     int ret;
282 
283     if (s->x_dirty_bitmap) {
284         if (!s->info.base_allocation) {
285             error_setg(errp, "requested x-dirty-bitmap %s not found",
286                        s->x_dirty_bitmap);
287             return -EINVAL;
288         }
289         if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) {
290             s->alloc_depth = true;
291         }
292     }
293 
294     if (s->info.flags & NBD_FLAG_READ_ONLY) {
295         ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp);
296         if (ret < 0) {
297             return ret;
298         }
299     }
300 
301     if (s->info.flags & NBD_FLAG_SEND_FUA) {
302         bs->supported_write_flags = BDRV_REQ_FUA;
303         bs->supported_zero_flags |= BDRV_REQ_FUA;
304     }
305 
306     if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
307         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
308         if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) {
309             bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK;
310         }
311     }
312 
313     trace_nbd_client_handshake_success(s->export);
314 
315     return 0;
316 }
317 
318 int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs,
319                                                 bool blocking, Error **errp)
320 {
321     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
322     int ret;
323     IO_CODE();
324 
325     assert_bdrv_graph_readable();
326     assert(!s->ioc);
327 
328     s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp);
329     if (!s->ioc) {
330         return -ECONNREFUSED;
331     }
332 
333     yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank,
334                            bs);
335 
336     ret = nbd_handle_updated_info(s->bs, NULL);
337     if (ret < 0) {
338         /*
339          * We have connected, but must fail for other reasons.
340          * Send NBD_CMD_DISC as a courtesy to the server.
341          */
342         NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode };
343 
344         nbd_send_request(s->ioc, &request);
345 
346         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
347                                  nbd_yank, bs);
348         object_unref(OBJECT(s->ioc));
349         s->ioc = NULL;
350 
351         return ret;
352     }
353 
354     qio_channel_set_blocking(s->ioc, false, NULL);
355     qio_channel_set_follow_coroutine_ctx(s->ioc, true);
356 
357     /* successfully connected */
358     WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
359         s->state = NBD_CLIENT_CONNECTED;
360     }
361 
362     return 0;
363 }
364 
365 /* Called with s->requests_lock held.  */
366 static bool nbd_client_connecting(BDRVNBDState *s)
367 {
368     return s->state == NBD_CLIENT_CONNECTING_WAIT ||
369         s->state == NBD_CLIENT_CONNECTING_NOWAIT;
370 }
371 
372 /* Called with s->requests_lock taken.  */
373 static void coroutine_fn GRAPH_RDLOCK nbd_reconnect_attempt(BDRVNBDState *s)
374 {
375     int ret;
376     bool blocking = s->state == NBD_CLIENT_CONNECTING_WAIT;
377 
378     /*
379      * Now we are sure that nobody is accessing the channel, and no one will
380      * try until we set the state to CONNECTED.
381      */
382     assert(nbd_client_connecting(s));
383     assert(s->in_flight == 1);
384 
385     trace_nbd_reconnect_attempt(s->bs->in_flight);
386 
387     if (blocking && !s->reconnect_delay_timer) {
388         /*
389          * It's the first reconnect attempt after switching to
390          * NBD_CLIENT_CONNECTING_WAIT
391          */
392         g_assert(s->reconnect_delay);
393         reconnect_delay_timer_init(s,
394             qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
395             s->reconnect_delay * NANOSECONDS_PER_SECOND);
396     }
397 
398     /* Finalize previous connection if any */
399     if (s->ioc) {
400         yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
401                                  nbd_yank, s->bs);
402         object_unref(OBJECT(s->ioc));
403         s->ioc = NULL;
404     }
405 
406     qemu_mutex_unlock(&s->requests_lock);
407     ret = nbd_co_do_establish_connection(s->bs, blocking, NULL);
408     trace_nbd_reconnect_attempt_result(ret, s->bs->in_flight);
409     qemu_mutex_lock(&s->requests_lock);
410 
411     /*
412      * The reconnect attempt is done (maybe successfully, maybe not), so
413      * we no longer need this timer.  Delete it so it will not outlive
414      * this I/O request (so draining removes all timers).
415      */
416     reconnect_delay_timer_del(s);
417 }
418 
419 static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t cookie)
420 {
421     int ret;
422     uint64_t ind = COOKIE_TO_INDEX(cookie), ind2;
423     QEMU_LOCK_GUARD(&s->receive_mutex);
424 
425     while (true) {
426         if (s->reply.cookie == cookie) {
427             /* We are done */
428             return 0;
429         }
430 
431         if (s->reply.cookie != 0) {
432             /*
433              * Some other request is being handled now. It should already be
434              * woken by whoever set s->reply.cookie (or never wait in this
435              * yield). So, we should not wake it here.
436              */
437             ind2 = COOKIE_TO_INDEX(s->reply.cookie);
438             assert(!s->requests[ind2].receiving);
439 
440             s->requests[ind].receiving = true;
441             qemu_co_mutex_unlock(&s->receive_mutex);
442 
443             qemu_coroutine_yield();
444             /*
445              * We may be woken for 2 reasons:
446              * 1. From this function, executing in parallel coroutine, when our
447              *    cookie is received.
448              * 2. From nbd_co_receive_one_chunk(), when previous request is
449              *    finished and s->reply.cookie set to 0.
450              * Anyway, it's OK to lock the mutex and go to the next iteration.
451              */
452 
453             qemu_co_mutex_lock(&s->receive_mutex);
454             assert(!s->requests[ind].receiving);
455             continue;
456         }
457 
458         /* We are under mutex and cookie is 0. We have to do the dirty work. */
459         assert(s->reply.cookie == 0);
460         ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, NULL);
461         if (ret <= 0) {
462             ret = ret ? ret : -EIO;
463             nbd_channel_error(s, ret);
464             return ret;
465         }
466         if (nbd_reply_is_structured(&s->reply) &&
467             s->info.mode < NBD_MODE_STRUCTURED) {
468             nbd_channel_error(s, -EINVAL);
469             return -EINVAL;
470         }
471         ind2 = COOKIE_TO_INDEX(s->reply.cookie);
472         if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].coroutine) {
473             nbd_channel_error(s, -EINVAL);
474             return -EINVAL;
475         }
476         if (s->reply.cookie == cookie) {
477             /* We are done */
478             return 0;
479         }
480         nbd_recv_coroutine_wake_one(&s->requests[ind2]);
481     }
482 }
483 
484 static int coroutine_fn GRAPH_RDLOCK
485 nbd_co_send_request(BlockDriverState *bs, NBDRequest *request,
486                     QEMUIOVector *qiov)
487 {
488     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
489     int rc, i = -1;
490 
491     qemu_mutex_lock(&s->requests_lock);
492     while (s->in_flight == MAX_NBD_REQUESTS ||
493            (s->state != NBD_CLIENT_CONNECTED && s->in_flight > 0)) {
494         qemu_co_queue_wait(&s->free_sema, &s->requests_lock);
495     }
496 
497     s->in_flight++;
498     if (s->state != NBD_CLIENT_CONNECTED) {
499         if (nbd_client_connecting(s)) {
500             nbd_reconnect_attempt(s);
501             qemu_co_queue_restart_all(&s->free_sema);
502         }
503         if (s->state != NBD_CLIENT_CONNECTED) {
504             rc = -EIO;
505             goto err;
506         }
507     }
508 
509     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
510         if (s->requests[i].coroutine == NULL) {
511             break;
512         }
513     }
514 
515     assert(i < MAX_NBD_REQUESTS);
516     s->requests[i].coroutine = qemu_coroutine_self();
517     s->requests[i].offset = request->from;
518     s->requests[i].receiving = false;
519     qemu_mutex_unlock(&s->requests_lock);
520 
521     qemu_co_mutex_lock(&s->send_mutex);
522     request->cookie = INDEX_TO_COOKIE(i);
523     request->mode = s->info.mode;
524 
525     assert(s->ioc);
526 
527     if (qiov) {
528         qio_channel_set_cork(s->ioc, true);
529         rc = nbd_send_request(s->ioc, request);
530         if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
531                                               NULL) < 0) {
532             rc = -EIO;
533         }
534         qio_channel_set_cork(s->ioc, false);
535     } else {
536         rc = nbd_send_request(s->ioc, request);
537     }
538     qemu_co_mutex_unlock(&s->send_mutex);
539 
540     if (rc < 0) {
541         qemu_mutex_lock(&s->requests_lock);
542 err:
543         nbd_channel_error_locked(s, rc);
544         if (i != -1) {
545             s->requests[i].coroutine = NULL;
546         }
547         s->in_flight--;
548         qemu_co_queue_next(&s->free_sema);
549         qemu_mutex_unlock(&s->requests_lock);
550     }
551     return rc;
552 }
553 
554 static inline uint16_t payload_advance16(uint8_t **payload)
555 {
556     *payload += 2;
557     return lduw_be_p(*payload - 2);
558 }
559 
560 static inline uint32_t payload_advance32(uint8_t **payload)
561 {
562     *payload += 4;
563     return ldl_be_p(*payload - 4);
564 }
565 
566 static inline uint64_t payload_advance64(uint8_t **payload)
567 {
568     *payload += 8;
569     return ldq_be_p(*payload - 8);
570 }
571 
572 static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
573                                          NBDStructuredReplyChunk *chunk,
574                                          uint8_t *payload, uint64_t orig_offset,
575                                          QEMUIOVector *qiov, Error **errp)
576 {
577     uint64_t offset;
578     uint32_t hole_size;
579 
580     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
581         error_setg(errp, "Protocol error: invalid payload for "
582                          "NBD_REPLY_TYPE_OFFSET_HOLE");
583         return -EINVAL;
584     }
585 
586     offset = payload_advance64(&payload);
587     hole_size = payload_advance32(&payload);
588 
589     if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
590         offset > orig_offset + qiov->size - hole_size) {
591         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
592                          " region");
593         return -EINVAL;
594     }
595     if (s->info.min_block &&
596         !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
597         trace_nbd_structured_read_compliance("hole");
598     }
599 
600     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
601 
602     return 0;
603 }
604 
605 /*
606  * nbd_parse_blockstatus_payload
607  * Based on our request, we expect only one extent in reply, for the
608  * base:allocation context.
609  */
610 static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
611                                          NBDStructuredReplyChunk *chunk,
612                                          uint8_t *payload, uint64_t orig_length,
613                                          NBDExtent32 *extent, Error **errp)
614 {
615     uint32_t context_id;
616 
617     /* The server succeeded, so it must have sent [at least] one extent */
618     if (chunk->length < sizeof(context_id) + sizeof(*extent)) {
619         error_setg(errp, "Protocol error: invalid payload for "
620                          "NBD_REPLY_TYPE_BLOCK_STATUS");
621         return -EINVAL;
622     }
623 
624     context_id = payload_advance32(&payload);
625     if (s->info.context_id != context_id) {
626         error_setg(errp, "Protocol error: unexpected context id %d for "
627                          "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
628                          "id is %d", context_id,
629                          s->info.context_id);
630         return -EINVAL;
631     }
632 
633     extent->length = payload_advance32(&payload);
634     extent->flags = payload_advance32(&payload);
635 
636     if (extent->length == 0) {
637         error_setg(errp, "Protocol error: server sent status chunk with "
638                    "zero length");
639         return -EINVAL;
640     }
641 
642     /*
643      * A server sending unaligned block status is in violation of the
644      * protocol, but as qemu-nbd 3.1 is such a server (at least for
645      * POSIX files that are not a multiple of 512 bytes, since qemu
646      * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
647      * still sees an implicit hole beyond the real EOF), it's nicer to
648      * work around the misbehaving server. If the request included
649      * more than the final unaligned block, truncate it back to an
650      * aligned result; if the request was only the final block, round
651      * up to the full block and change the status to fully-allocated
652      * (always a safe status, even if it loses information).
653      */
654     if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
655                                                    s->info.min_block)) {
656         trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
657         if (extent->length > s->info.min_block) {
658             extent->length = QEMU_ALIGN_DOWN(extent->length,
659                                              s->info.min_block);
660         } else {
661             extent->length = s->info.min_block;
662             extent->flags = 0;
663         }
664     }
665 
666     /*
667      * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
668      * sent us any more than one extent, nor should it have included
669      * status beyond our request in that extent. However, it's easy
670      * enough to ignore the server's noncompliance without killing the
671      * connection; just ignore trailing extents, and clamp things to
672      * the length of our request.
673      */
674     if (chunk->length > sizeof(context_id) + sizeof(*extent)) {
675         trace_nbd_parse_blockstatus_compliance("more than one extent");
676     }
677     if (extent->length > orig_length) {
678         extent->length = orig_length;
679         trace_nbd_parse_blockstatus_compliance("extent length too large");
680     }
681 
682     /*
683      * HACK: if we are using x-dirty-bitmaps to access
684      * qemu:allocation-depth, treat all depths > 2 the same as 2,
685      * since nbd_client_co_block_status is only expecting the low two
686      * bits to be set.
687      */
688     if (s->alloc_depth && extent->flags > 2) {
689         extent->flags = 2;
690     }
691 
692     return 0;
693 }
694 
695 /*
696  * nbd_parse_error_payload
697  * on success @errp contains message describing nbd error reply
698  */
699 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
700                                    uint8_t *payload, int *request_ret,
701                                    Error **errp)
702 {
703     uint32_t error;
704     uint16_t message_size;
705 
706     assert(chunk->type & (1 << 15));
707 
708     if (chunk->length < sizeof(error) + sizeof(message_size)) {
709         error_setg(errp,
710                    "Protocol error: invalid payload for structured error");
711         return -EINVAL;
712     }
713 
714     error = nbd_errno_to_system_errno(payload_advance32(&payload));
715     if (error == 0) {
716         error_setg(errp, "Protocol error: server sent structured error chunk "
717                          "with error = 0");
718         return -EINVAL;
719     }
720 
721     *request_ret = -error;
722     message_size = payload_advance16(&payload);
723 
724     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
725         error_setg(errp, "Protocol error: server sent structured error chunk "
726                          "with incorrect message size");
727         return -EINVAL;
728     }
729 
730     /* TODO: Add a trace point to mention the server complaint */
731 
732     /* TODO handle ERROR_OFFSET */
733 
734     return 0;
735 }
736 
737 static int coroutine_fn
738 nbd_co_receive_offset_data_payload(BDRVNBDState *s, uint64_t orig_offset,
739                                    QEMUIOVector *qiov, Error **errp)
740 {
741     QEMUIOVector sub_qiov;
742     uint64_t offset;
743     size_t data_size;
744     int ret;
745     NBDStructuredReplyChunk *chunk = &s->reply.structured;
746 
747     assert(nbd_reply_is_structured(&s->reply));
748 
749     /* The NBD spec requires at least one byte of payload */
750     if (chunk->length <= sizeof(offset)) {
751         error_setg(errp, "Protocol error: invalid payload for "
752                          "NBD_REPLY_TYPE_OFFSET_DATA");
753         return -EINVAL;
754     }
755 
756     if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
757         return -EIO;
758     }
759 
760     data_size = chunk->length - sizeof(offset);
761     assert(data_size);
762     if (offset < orig_offset || data_size > qiov->size ||
763         offset > orig_offset + qiov->size - data_size) {
764         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
765                          " region");
766         return -EINVAL;
767     }
768     if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
769         trace_nbd_structured_read_compliance("data");
770     }
771 
772     qemu_iovec_init(&sub_qiov, qiov->niov);
773     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
774     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
775     qemu_iovec_destroy(&sub_qiov);
776 
777     return ret < 0 ? -EIO : 0;
778 }
779 
780 #define NBD_MAX_MALLOC_PAYLOAD 1000
781 static coroutine_fn int nbd_co_receive_structured_payload(
782         BDRVNBDState *s, void **payload, Error **errp)
783 {
784     int ret;
785     uint32_t len;
786 
787     assert(nbd_reply_is_structured(&s->reply));
788 
789     len = s->reply.structured.length;
790 
791     if (len == 0) {
792         return 0;
793     }
794 
795     if (payload == NULL) {
796         error_setg(errp, "Unexpected structured payload");
797         return -EINVAL;
798     }
799 
800     if (len > NBD_MAX_MALLOC_PAYLOAD) {
801         error_setg(errp, "Payload too large");
802         return -EINVAL;
803     }
804 
805     *payload = g_new(char, len);
806     ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
807     if (ret < 0) {
808         g_free(*payload);
809         *payload = NULL;
810         return ret;
811     }
812 
813     return 0;
814 }
815 
816 /*
817  * nbd_co_do_receive_one_chunk
818  * for simple reply:
819  *   set request_ret to received reply error
820  *   if qiov is not NULL: read payload to @qiov
821  * for structured reply chunk:
822  *   if error chunk: read payload, set @request_ret, do not set @payload
823  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
824  *   else: read payload to @payload
825  *
826  * If function fails, @errp contains corresponding error message, and the
827  * connection with the server is suspect.  If it returns 0, then the
828  * transaction succeeded (although @request_ret may be a negative errno
829  * corresponding to the server's error reply), and errp is unchanged.
830  */
831 static coroutine_fn int nbd_co_do_receive_one_chunk(
832         BDRVNBDState *s, uint64_t cookie, bool only_structured,
833         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
834 {
835     int ret;
836     int i = COOKIE_TO_INDEX(cookie);
837     void *local_payload = NULL;
838     NBDStructuredReplyChunk *chunk;
839 
840     if (payload) {
841         *payload = NULL;
842     }
843     *request_ret = 0;
844 
845     ret = nbd_receive_replies(s, cookie);
846     if (ret < 0) {
847         error_setg(errp, "Connection closed");
848         return -EIO;
849     }
850     assert(s->ioc);
851 
852     assert(s->reply.cookie == cookie);
853 
854     if (nbd_reply_is_simple(&s->reply)) {
855         if (only_structured) {
856             error_setg(errp, "Protocol error: simple reply when structured "
857                              "reply chunk was expected");
858             return -EINVAL;
859         }
860 
861         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
862         if (*request_ret < 0 || !qiov) {
863             return 0;
864         }
865 
866         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
867                                      errp) < 0 ? -EIO : 0;
868     }
869 
870     /* handle structured reply chunk */
871     assert(s->info.mode >= NBD_MODE_STRUCTURED);
872     chunk = &s->reply.structured;
873 
874     if (chunk->type == NBD_REPLY_TYPE_NONE) {
875         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
876             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
877                        " NBD_REPLY_FLAG_DONE flag set");
878             return -EINVAL;
879         }
880         if (chunk->length) {
881             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
882                        " nonzero length");
883             return -EINVAL;
884         }
885         return 0;
886     }
887 
888     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
889         if (!qiov) {
890             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
891             return -EINVAL;
892         }
893 
894         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
895                                                   qiov, errp);
896     }
897 
898     if (nbd_reply_type_is_error(chunk->type)) {
899         payload = &local_payload;
900     }
901 
902     ret = nbd_co_receive_structured_payload(s, payload, errp);
903     if (ret < 0) {
904         return ret;
905     }
906 
907     if (nbd_reply_type_is_error(chunk->type)) {
908         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
909         g_free(local_payload);
910         return ret;
911     }
912 
913     return 0;
914 }
915 
916 /*
917  * nbd_co_receive_one_chunk
918  * Read reply, wake up connection_co and set s->quit if needed.
919  * Return value is a fatal error code or normal nbd reply error code
920  */
921 static coroutine_fn int nbd_co_receive_one_chunk(
922         BDRVNBDState *s, uint64_t cookie, bool only_structured,
923         int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
924         Error **errp)
925 {
926     int ret = nbd_co_do_receive_one_chunk(s, cookie, only_structured,
927                                           request_ret, qiov, payload, errp);
928 
929     if (ret < 0) {
930         memset(reply, 0, sizeof(*reply));
931         nbd_channel_error(s, ret);
932     } else {
933         /* For assert at loop start in nbd_connection_entry */
934         *reply = s->reply;
935     }
936     s->reply.cookie = 0;
937 
938     nbd_recv_coroutines_wake(s);
939 
940     return ret;
941 }
942 
943 typedef struct NBDReplyChunkIter {
944     int ret;
945     int request_ret;
946     Error *err;
947     bool done, only_structured;
948 } NBDReplyChunkIter;
949 
950 static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
951                                    int ret, Error **local_err)
952 {
953     assert(local_err && *local_err);
954     assert(ret < 0);
955 
956     if (!iter->ret) {
957         iter->ret = ret;
958         error_propagate(&iter->err, *local_err);
959     } else {
960         error_free(*local_err);
961     }
962 
963     *local_err = NULL;
964 }
965 
966 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
967 {
968     assert(ret < 0);
969 
970     if (!iter->request_ret) {
971         iter->request_ret = ret;
972     }
973 }
974 
975 /*
976  * NBD_FOREACH_REPLY_CHUNK
977  * The pointer stored in @payload requires g_free() to free it.
978  */
979 #define NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, structured, \
980                                 qiov, reply, payload) \
981     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
982          nbd_reply_chunk_iter_receive(s, &iter, cookie, qiov, reply, payload);)
983 
984 /*
985  * nbd_reply_chunk_iter_receive
986  * The pointer stored in @payload requires g_free() to free it.
987  */
988 static bool coroutine_fn nbd_reply_chunk_iter_receive(BDRVNBDState *s,
989                                                       NBDReplyChunkIter *iter,
990                                                       uint64_t cookie,
991                                                       QEMUIOVector *qiov,
992                                                       NBDReply *reply,
993                                                       void **payload)
994 {
995     int ret, request_ret;
996     NBDReply local_reply;
997     NBDStructuredReplyChunk *chunk;
998     Error *local_err = NULL;
999 
1000     if (iter->done) {
1001         /* Previous iteration was last. */
1002         goto break_loop;
1003     }
1004 
1005     if (reply == NULL) {
1006         reply = &local_reply;
1007     }
1008 
1009     ret = nbd_co_receive_one_chunk(s, cookie, iter->only_structured,
1010                                    &request_ret, qiov, reply, payload,
1011                                    &local_err);
1012     if (ret < 0) {
1013         nbd_iter_channel_error(iter, ret, &local_err);
1014     } else if (request_ret < 0) {
1015         nbd_iter_request_error(iter, request_ret);
1016     }
1017 
1018     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1019     if (nbd_reply_is_simple(reply) || iter->ret < 0) {
1020         goto break_loop;
1021     }
1022 
1023     chunk = &reply->structured;
1024     iter->only_structured = true;
1025 
1026     if (chunk->type == NBD_REPLY_TYPE_NONE) {
1027         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1028         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1029         goto break_loop;
1030     }
1031 
1032     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1033         /* This iteration is last. */
1034         iter->done = true;
1035     }
1036 
1037     /* Execute the loop body */
1038     return true;
1039 
1040 break_loop:
1041     qemu_mutex_lock(&s->requests_lock);
1042     s->requests[COOKIE_TO_INDEX(cookie)].coroutine = NULL;
1043     s->in_flight--;
1044     qemu_co_queue_next(&s->free_sema);
1045     qemu_mutex_unlock(&s->requests_lock);
1046 
1047     return false;
1048 }
1049 
1050 static int coroutine_fn
1051 nbd_co_receive_return_code(BDRVNBDState *s, uint64_t cookie,
1052                            int *request_ret, Error **errp)
1053 {
1054     NBDReplyChunkIter iter;
1055 
1056     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, NULL, NULL) {
1057         /* nbd_reply_chunk_iter_receive does all the work */
1058     }
1059 
1060     error_propagate(errp, iter.err);
1061     *request_ret = iter.request_ret;
1062     return iter.ret;
1063 }
1064 
1065 static int coroutine_fn
1066 nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t cookie,
1067                              uint64_t offset, QEMUIOVector *qiov,
1068                              int *request_ret, Error **errp)
1069 {
1070     NBDReplyChunkIter iter;
1071     NBDReply reply;
1072     void *payload = NULL;
1073     Error *local_err = NULL;
1074 
1075     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie,
1076                             s->info.mode >= NBD_MODE_STRUCTURED,
1077                             qiov, &reply, &payload)
1078     {
1079         int ret;
1080         NBDStructuredReplyChunk *chunk = &reply.structured;
1081 
1082         assert(nbd_reply_is_structured(&reply));
1083 
1084         switch (chunk->type) {
1085         case NBD_REPLY_TYPE_OFFSET_DATA:
1086             /*
1087              * special cased in nbd_co_receive_one_chunk, data is already
1088              * in qiov
1089              */
1090             break;
1091         case NBD_REPLY_TYPE_OFFSET_HOLE:
1092             ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1093                                                 offset, qiov, &local_err);
1094             if (ret < 0) {
1095                 nbd_channel_error(s, ret);
1096                 nbd_iter_channel_error(&iter, ret, &local_err);
1097             }
1098             break;
1099         default:
1100             if (!nbd_reply_type_is_error(chunk->type)) {
1101                 /* not allowed reply type */
1102                 nbd_channel_error(s, -EINVAL);
1103                 error_setg(&local_err,
1104                            "Unexpected reply type: %d (%s) for CMD_READ",
1105                            chunk->type, nbd_reply_type_lookup(chunk->type));
1106                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1107             }
1108         }
1109 
1110         g_free(payload);
1111         payload = NULL;
1112     }
1113 
1114     error_propagate(errp, iter.err);
1115     *request_ret = iter.request_ret;
1116     return iter.ret;
1117 }
1118 
1119 static int coroutine_fn
1120 nbd_co_receive_blockstatus_reply(BDRVNBDState *s, uint64_t cookie,
1121                                  uint64_t length, NBDExtent32 *extent,
1122                                  int *request_ret, Error **errp)
1123 {
1124     NBDReplyChunkIter iter;
1125     NBDReply reply;
1126     void *payload = NULL;
1127     Error *local_err = NULL;
1128     bool received = false;
1129 
1130     assert(!extent->length);
1131     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, &reply, &payload) {
1132         int ret;
1133         NBDStructuredReplyChunk *chunk = &reply.structured;
1134 
1135         assert(nbd_reply_is_structured(&reply));
1136 
1137         switch (chunk->type) {
1138         case NBD_REPLY_TYPE_BLOCK_STATUS:
1139             if (received) {
1140                 nbd_channel_error(s, -EINVAL);
1141                 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1142                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1143             }
1144             received = true;
1145 
1146             ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1147                                                 payload, length, extent,
1148                                                 &local_err);
1149             if (ret < 0) {
1150                 nbd_channel_error(s, ret);
1151                 nbd_iter_channel_error(&iter, ret, &local_err);
1152             }
1153             break;
1154         default:
1155             if (!nbd_reply_type_is_error(chunk->type)) {
1156                 nbd_channel_error(s, -EINVAL);
1157                 error_setg(&local_err,
1158                            "Unexpected reply type: %d (%s) "
1159                            "for CMD_BLOCK_STATUS",
1160                            chunk->type, nbd_reply_type_lookup(chunk->type));
1161                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1162             }
1163         }
1164 
1165         g_free(payload);
1166         payload = NULL;
1167     }
1168 
1169     if (!extent->length && !iter.request_ret) {
1170         error_setg(&local_err, "Server did not reply with any status extents");
1171         nbd_iter_channel_error(&iter, -EIO, &local_err);
1172     }
1173 
1174     error_propagate(errp, iter.err);
1175     *request_ret = iter.request_ret;
1176     return iter.ret;
1177 }
1178 
1179 static int coroutine_fn GRAPH_RDLOCK
1180 nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1181                QEMUIOVector *write_qiov)
1182 {
1183     int ret, request_ret;
1184     Error *local_err = NULL;
1185     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1186 
1187     assert(request->type != NBD_CMD_READ);
1188     if (write_qiov) {
1189         assert(request->type == NBD_CMD_WRITE);
1190         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1191     } else {
1192         assert(request->type != NBD_CMD_WRITE);
1193     }
1194 
1195     do {
1196         ret = nbd_co_send_request(bs, request, write_qiov);
1197         if (ret < 0) {
1198             continue;
1199         }
1200 
1201         ret = nbd_co_receive_return_code(s, request->cookie,
1202                                          &request_ret, &local_err);
1203         if (local_err) {
1204             trace_nbd_co_request_fail(request->from, request->len,
1205                                       request->cookie, request->flags,
1206                                       request->type,
1207                                       nbd_cmd_lookup(request->type),
1208                                       ret, error_get_pretty(local_err));
1209             error_free(local_err);
1210             local_err = NULL;
1211         }
1212     } while (ret < 0 && nbd_client_will_reconnect(s));
1213 
1214     return ret ? ret : request_ret;
1215 }
1216 
1217 static int coroutine_fn GRAPH_RDLOCK
1218 nbd_client_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1219                      QEMUIOVector *qiov, BdrvRequestFlags flags)
1220 {
1221     int ret, request_ret;
1222     Error *local_err = NULL;
1223     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1224     NBDRequest request = {
1225         .type = NBD_CMD_READ,
1226         .from = offset,
1227         .len = bytes,
1228     };
1229 
1230     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1231 
1232     if (!bytes) {
1233         return 0;
1234     }
1235     /*
1236      * Work around the fact that the block layer doesn't do
1237      * byte-accurate sizing yet - if the read exceeds the server's
1238      * advertised size because the block layer rounded size up, then
1239      * truncate the request to the server and tail-pad with zero.
1240      */
1241     if (offset >= s->info.size) {
1242         assert(bytes < BDRV_SECTOR_SIZE);
1243         qemu_iovec_memset(qiov, 0, 0, bytes);
1244         return 0;
1245     }
1246     if (offset + bytes > s->info.size) {
1247         uint64_t slop = offset + bytes - s->info.size;
1248 
1249         assert(slop < BDRV_SECTOR_SIZE);
1250         qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1251         request.len -= slop;
1252     }
1253 
1254     do {
1255         ret = nbd_co_send_request(bs, &request, NULL);
1256         if (ret < 0) {
1257             continue;
1258         }
1259 
1260         ret = nbd_co_receive_cmdread_reply(s, request.cookie, offset, qiov,
1261                                            &request_ret, &local_err);
1262         if (local_err) {
1263             trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1264                                       request.flags, request.type,
1265                                       nbd_cmd_lookup(request.type),
1266                                       ret, error_get_pretty(local_err));
1267             error_free(local_err);
1268             local_err = NULL;
1269         }
1270     } while (ret < 0 && nbd_client_will_reconnect(s));
1271 
1272     return ret ? ret : request_ret;
1273 }
1274 
1275 static int coroutine_fn GRAPH_RDLOCK
1276 nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
1277                       QEMUIOVector *qiov, BdrvRequestFlags flags)
1278 {
1279     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1280     NBDRequest request = {
1281         .type = NBD_CMD_WRITE,
1282         .from = offset,
1283         .len = bytes,
1284     };
1285 
1286     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1287     if (flags & BDRV_REQ_FUA) {
1288         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1289         request.flags |= NBD_CMD_FLAG_FUA;
1290     }
1291 
1292     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1293 
1294     if (!bytes) {
1295         return 0;
1296     }
1297     return nbd_co_request(bs, &request, qiov);
1298 }
1299 
1300 static int coroutine_fn GRAPH_RDLOCK
1301 nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
1302                             BdrvRequestFlags flags)
1303 {
1304     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1305     NBDRequest request = {
1306         .type = NBD_CMD_WRITE_ZEROES,
1307         .from = offset,
1308         .len = bytes,
1309     };
1310 
1311     /* rely on max_pwrite_zeroes */
1312     assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED);
1313 
1314     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1315     if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1316         return -ENOTSUP;
1317     }
1318 
1319     if (flags & BDRV_REQ_FUA) {
1320         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1321         request.flags |= NBD_CMD_FLAG_FUA;
1322     }
1323     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1324         request.flags |= NBD_CMD_FLAG_NO_HOLE;
1325     }
1326     if (flags & BDRV_REQ_NO_FALLBACK) {
1327         assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1328         request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1329     }
1330 
1331     if (!bytes) {
1332         return 0;
1333     }
1334     return nbd_co_request(bs, &request, NULL);
1335 }
1336 
1337 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_flush(BlockDriverState *bs)
1338 {
1339     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1340     NBDRequest request = { .type = NBD_CMD_FLUSH };
1341 
1342     if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1343         return 0;
1344     }
1345 
1346     request.from = 0;
1347     request.len = 0;
1348 
1349     return nbd_co_request(bs, &request, NULL);
1350 }
1351 
1352 static int coroutine_fn GRAPH_RDLOCK
1353 nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
1354 {
1355     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1356     NBDRequest request = {
1357         .type = NBD_CMD_TRIM,
1358         .from = offset,
1359         .len = bytes,
1360     };
1361 
1362     /* rely on max_pdiscard */
1363     assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED);
1364 
1365     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1366     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1367         return 0;
1368     }
1369 
1370     return nbd_co_request(bs, &request, NULL);
1371 }
1372 
1373 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_block_status(
1374         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1375         int64_t *pnum, int64_t *map, BlockDriverState **file)
1376 {
1377     int ret, request_ret;
1378     NBDExtent32 extent = { 0 };
1379     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1380     Error *local_err = NULL;
1381 
1382     NBDRequest request = {
1383         .type = NBD_CMD_BLOCK_STATUS,
1384         .from = offset,
1385         .len = MIN(bytes, s->info.size - offset),
1386         .flags = NBD_CMD_FLAG_REQ_ONE,
1387     };
1388 
1389     if (!s->info.base_allocation) {
1390         *pnum = bytes;
1391         *map = offset;
1392         *file = bs;
1393         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1394     }
1395     if (s->info.mode < NBD_MODE_EXTENDED) {
1396         request.len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1397                           request.len);
1398     }
1399 
1400     /*
1401      * Work around the fact that the block layer doesn't do
1402      * byte-accurate sizing yet - if the status request exceeds the
1403      * server's advertised size because the block layer rounded size
1404      * up, we truncated the request to the server (above), or are
1405      * called on just the hole.
1406      */
1407     if (offset >= s->info.size) {
1408         *pnum = bytes;
1409         assert(bytes < BDRV_SECTOR_SIZE);
1410         /* Intentionally don't report offset_valid for the hole */
1411         return BDRV_BLOCK_ZERO;
1412     }
1413 
1414     if (s->info.min_block) {
1415         assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1416     }
1417     do {
1418         ret = nbd_co_send_request(bs, &request, NULL);
1419         if (ret < 0) {
1420             continue;
1421         }
1422 
1423         ret = nbd_co_receive_blockstatus_reply(s, request.cookie, bytes,
1424                                                &extent, &request_ret,
1425                                                &local_err);
1426         if (local_err) {
1427             trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1428                                       request.flags, request.type,
1429                                       nbd_cmd_lookup(request.type),
1430                                       ret, error_get_pretty(local_err));
1431             error_free(local_err);
1432             local_err = NULL;
1433         }
1434     } while (ret < 0 && nbd_client_will_reconnect(s));
1435 
1436     if (ret < 0 || request_ret < 0) {
1437         return ret ? ret : request_ret;
1438     }
1439 
1440     assert(extent.length);
1441     *pnum = extent.length;
1442     *map = offset;
1443     *file = bs;
1444     return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1445         (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1446         BDRV_BLOCK_OFFSET_VALID;
1447 }
1448 
1449 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1450                                      BlockReopenQueue *queue, Error **errp)
1451 {
1452     BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1453 
1454     if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1455         error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1456         return -EACCES;
1457     }
1458     return 0;
1459 }
1460 
1461 static void nbd_yank(void *opaque)
1462 {
1463     BlockDriverState *bs = opaque;
1464     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1465 
1466     QEMU_LOCK_GUARD(&s->requests_lock);
1467     qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1468     s->state = NBD_CLIENT_QUIT;
1469 }
1470 
1471 static void nbd_client_close(BlockDriverState *bs)
1472 {
1473     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1474     NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode };
1475 
1476     if (s->ioc) {
1477         nbd_send_request(s->ioc, &request);
1478     }
1479 
1480     nbd_teardown_connection(bs);
1481 }
1482 
1483 
1484 /*
1485  * Parse nbd_open options
1486  */
1487 
1488 static int nbd_parse_uri(const char *filename, QDict *options)
1489 {
1490     URI *uri;
1491     const char *p;
1492     QueryParams *qp = NULL;
1493     int ret = 0;
1494     bool is_unix;
1495 
1496     uri = uri_parse(filename);
1497     if (!uri) {
1498         return -EINVAL;
1499     }
1500 
1501     /* transport */
1502     if (!g_strcmp0(uri->scheme, "nbd")) {
1503         is_unix = false;
1504     } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1505         is_unix = false;
1506     } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1507         is_unix = true;
1508     } else {
1509         ret = -EINVAL;
1510         goto out;
1511     }
1512 
1513     p = uri->path ? uri->path : "";
1514     if (p[0] == '/') {
1515         p++;
1516     }
1517     if (p[0]) {
1518         qdict_put_str(options, "export", p);
1519     }
1520 
1521     qp = query_params_parse(uri->query);
1522     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1523         ret = -EINVAL;
1524         goto out;
1525     }
1526 
1527     if (is_unix) {
1528         /* nbd+unix:///export?socket=path */
1529         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1530             ret = -EINVAL;
1531             goto out;
1532         }
1533         qdict_put_str(options, "server.type", "unix");
1534         qdict_put_str(options, "server.path", qp->p[0].value);
1535     } else {
1536         QString *host;
1537         char *port_str;
1538 
1539         /* nbd[+tcp]://host[:port]/export */
1540         if (!uri->server) {
1541             ret = -EINVAL;
1542             goto out;
1543         }
1544 
1545         /* strip braces from literal IPv6 address */
1546         if (uri->server[0] == '[') {
1547             host = qstring_from_substr(uri->server, 1,
1548                                        strlen(uri->server) - 1);
1549         } else {
1550             host = qstring_from_str(uri->server);
1551         }
1552 
1553         qdict_put_str(options, "server.type", "inet");
1554         qdict_put(options, "server.host", host);
1555 
1556         port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1557         qdict_put_str(options, "server.port", port_str);
1558         g_free(port_str);
1559     }
1560 
1561 out:
1562     if (qp) {
1563         query_params_free(qp);
1564     }
1565     uri_free(uri);
1566     return ret;
1567 }
1568 
1569 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1570 {
1571     const QDictEntry *e;
1572 
1573     for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1574         if (!strcmp(e->key, "host") ||
1575             !strcmp(e->key, "port") ||
1576             !strcmp(e->key, "path") ||
1577             !strcmp(e->key, "export") ||
1578             strstart(e->key, "server.", NULL))
1579         {
1580             error_setg(errp, "Option '%s' cannot be used with a file name",
1581                        e->key);
1582             return true;
1583         }
1584     }
1585 
1586     return false;
1587 }
1588 
1589 static void nbd_parse_filename(const char *filename, QDict *options,
1590                                Error **errp)
1591 {
1592     g_autofree char *file = NULL;
1593     char *export_name;
1594     const char *host_spec;
1595     const char *unixpath;
1596 
1597     if (nbd_has_filename_options_conflict(options, errp)) {
1598         return;
1599     }
1600 
1601     if (strstr(filename, "://")) {
1602         int ret = nbd_parse_uri(filename, options);
1603         if (ret < 0) {
1604             error_setg(errp, "No valid URL specified");
1605         }
1606         return;
1607     }
1608 
1609     file = g_strdup(filename);
1610 
1611     export_name = strstr(file, EN_OPTSTR);
1612     if (export_name) {
1613         if (export_name[strlen(EN_OPTSTR)] == 0) {
1614             return;
1615         }
1616         export_name[0] = 0; /* truncate 'file' */
1617         export_name += strlen(EN_OPTSTR);
1618 
1619         qdict_put_str(options, "export", export_name);
1620     }
1621 
1622     /* extract the host_spec - fail if it's not nbd:... */
1623     if (!strstart(file, "nbd:", &host_spec)) {
1624         error_setg(errp, "File name string for NBD must start with 'nbd:'");
1625         return;
1626     }
1627 
1628     if (!*host_spec) {
1629         return;
1630     }
1631 
1632     /* are we a UNIX or TCP socket? */
1633     if (strstart(host_spec, "unix:", &unixpath)) {
1634         qdict_put_str(options, "server.type", "unix");
1635         qdict_put_str(options, "server.path", unixpath);
1636     } else {
1637         InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1638 
1639         if (inet_parse(addr, host_spec, errp)) {
1640             goto out_inet;
1641         }
1642 
1643         qdict_put_str(options, "server.type", "inet");
1644         qdict_put_str(options, "server.host", addr->host);
1645         qdict_put_str(options, "server.port", addr->port);
1646     out_inet:
1647         qapi_free_InetSocketAddress(addr);
1648     }
1649 }
1650 
1651 static bool nbd_process_legacy_socket_options(QDict *output_options,
1652                                               QemuOpts *legacy_opts,
1653                                               Error **errp)
1654 {
1655     const char *path = qemu_opt_get(legacy_opts, "path");
1656     const char *host = qemu_opt_get(legacy_opts, "host");
1657     const char *port = qemu_opt_get(legacy_opts, "port");
1658     const QDictEntry *e;
1659 
1660     if (!path && !host && !port) {
1661         return true;
1662     }
1663 
1664     for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1665     {
1666         if (strstart(e->key, "server.", NULL)) {
1667             error_setg(errp, "Cannot use 'server' and path/host/port at the "
1668                        "same time");
1669             return false;
1670         }
1671     }
1672 
1673     if (path && host) {
1674         error_setg(errp, "path and host may not be used at the same time");
1675         return false;
1676     } else if (path) {
1677         if (port) {
1678             error_setg(errp, "port may not be used without host");
1679             return false;
1680         }
1681 
1682         qdict_put_str(output_options, "server.type", "unix");
1683         qdict_put_str(output_options, "server.path", path);
1684     } else if (host) {
1685         qdict_put_str(output_options, "server.type", "inet");
1686         qdict_put_str(output_options, "server.host", host);
1687         qdict_put_str(output_options, "server.port",
1688                       port ?: stringify(NBD_DEFAULT_PORT));
1689     }
1690 
1691     return true;
1692 }
1693 
1694 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1695                                  Error **errp)
1696 {
1697     SocketAddress *saddr = NULL;
1698     QDict *addr = NULL;
1699     Visitor *iv = NULL;
1700 
1701     qdict_extract_subqdict(options, &addr, "server.");
1702     if (!qdict_size(addr)) {
1703         error_setg(errp, "NBD server address missing");
1704         goto done;
1705     }
1706 
1707     iv = qobject_input_visitor_new_flat_confused(addr, errp);
1708     if (!iv) {
1709         goto done;
1710     }
1711 
1712     if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1713         goto done;
1714     }
1715 
1716     if (socket_address_parse_named_fd(saddr, errp) < 0) {
1717         qapi_free_SocketAddress(saddr);
1718         saddr = NULL;
1719         goto done;
1720     }
1721 
1722 done:
1723     qobject_unref(addr);
1724     visit_free(iv);
1725     return saddr;
1726 }
1727 
1728 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1729 {
1730     Object *obj;
1731     QCryptoTLSCreds *creds;
1732 
1733     obj = object_resolve_path_component(
1734         object_get_objects_root(), id);
1735     if (!obj) {
1736         error_setg(errp, "No TLS credentials with id '%s'",
1737                    id);
1738         return NULL;
1739     }
1740     creds = (QCryptoTLSCreds *)
1741         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1742     if (!creds) {
1743         error_setg(errp, "Object with id '%s' is not TLS credentials",
1744                    id);
1745         return NULL;
1746     }
1747 
1748     if (!qcrypto_tls_creds_check_endpoint(creds,
1749                                           QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1750                                           errp)) {
1751         return NULL;
1752     }
1753     object_ref(obj);
1754     return creds;
1755 }
1756 
1757 
1758 static QemuOptsList nbd_runtime_opts = {
1759     .name = "nbd",
1760     .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1761     .desc = {
1762         {
1763             .name = "host",
1764             .type = QEMU_OPT_STRING,
1765             .help = "TCP host to connect to",
1766         },
1767         {
1768             .name = "port",
1769             .type = QEMU_OPT_STRING,
1770             .help = "TCP port to connect to",
1771         },
1772         {
1773             .name = "path",
1774             .type = QEMU_OPT_STRING,
1775             .help = "Unix socket path to connect to",
1776         },
1777         {
1778             .name = "export",
1779             .type = QEMU_OPT_STRING,
1780             .help = "Name of the NBD export to open",
1781         },
1782         {
1783             .name = "tls-creds",
1784             .type = QEMU_OPT_STRING,
1785             .help = "ID of the TLS credentials to use",
1786         },
1787         {
1788             .name = "tls-hostname",
1789             .type = QEMU_OPT_STRING,
1790             .help = "Override hostname for validating TLS x509 certificate",
1791         },
1792         {
1793             .name = "x-dirty-bitmap",
1794             .type = QEMU_OPT_STRING,
1795             .help = "experimental: expose named dirty bitmap in place of "
1796                     "block status",
1797         },
1798         {
1799             .name = "reconnect-delay",
1800             .type = QEMU_OPT_NUMBER,
1801             .help = "On an unexpected disconnect, the nbd client tries to "
1802                     "connect again until succeeding or encountering a serious "
1803                     "error.  During the first @reconnect-delay seconds, all "
1804                     "requests are paused and will be rerun on a successful "
1805                     "reconnect. After that time, any delayed requests and all "
1806                     "future requests before a successful reconnect will "
1807                     "immediately fail. Default 0",
1808         },
1809         {
1810             .name = "open-timeout",
1811             .type = QEMU_OPT_NUMBER,
1812             .help = "In seconds. If zero, the nbd driver tries the connection "
1813                     "only once, and fails to open if the connection fails. "
1814                     "If non-zero, the nbd driver will repeat connection "
1815                     "attempts until successful or until @open-timeout seconds "
1816                     "have elapsed. Default 0",
1817         },
1818         { /* end of list */ }
1819     },
1820 };
1821 
1822 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1823                                Error **errp)
1824 {
1825     BDRVNBDState *s = bs->opaque;
1826     QemuOpts *opts;
1827     int ret = -EINVAL;
1828 
1829     opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1830     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1831         goto error;
1832     }
1833 
1834     /* Translate @host, @port, and @path to a SocketAddress */
1835     if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1836         goto error;
1837     }
1838 
1839     /* Pop the config into our state object. Exit if invalid. */
1840     s->saddr = nbd_config(s, options, errp);
1841     if (!s->saddr) {
1842         goto error;
1843     }
1844 
1845     s->export = g_strdup(qemu_opt_get(opts, "export"));
1846     if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1847         error_setg(errp, "export name too long to send to server");
1848         goto error;
1849     }
1850 
1851     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1852     if (s->tlscredsid) {
1853         s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1854         if (!s->tlscreds) {
1855             goto error;
1856         }
1857 
1858         s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname"));
1859         if (!s->tlshostname &&
1860             s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1861             s->tlshostname = g_strdup(s->saddr->u.inet.host);
1862         }
1863     }
1864 
1865     s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1866     if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1867         error_setg(errp, "x-dirty-bitmap query too long to send to server");
1868         goto error;
1869     }
1870 
1871     s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1872     s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1873 
1874     ret = 0;
1875 
1876  error:
1877     qemu_opts_del(opts);
1878     return ret;
1879 }
1880 
1881 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1882                     Error **errp)
1883 {
1884     int ret;
1885     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1886 
1887     s->bs = bs;
1888     qemu_mutex_init(&s->requests_lock);
1889     qemu_co_queue_init(&s->free_sema);
1890     qemu_co_mutex_init(&s->send_mutex);
1891     qemu_co_mutex_init(&s->receive_mutex);
1892 
1893     if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1894         return -EEXIST;
1895     }
1896 
1897     ret = nbd_process_options(bs, options, errp);
1898     if (ret < 0) {
1899         goto fail;
1900     }
1901 
1902     s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1903                                         s->x_dirty_bitmap, s->tlscreds,
1904                                         s->tlshostname);
1905 
1906     if (s->open_timeout) {
1907         nbd_client_connection_enable_retry(s->conn);
1908         open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1909                         s->open_timeout * NANOSECONDS_PER_SECOND);
1910     }
1911 
1912     s->state = NBD_CLIENT_CONNECTING_WAIT;
1913     ret = nbd_do_establish_connection(bs, true, errp);
1914     if (ret < 0) {
1915         goto fail;
1916     }
1917 
1918     /*
1919      * The connect attempt is done, so we no longer need this timer.
1920      * Delete it, because we do not want it to be around when this node
1921      * is drained or closed.
1922      */
1923     open_timer_del(s);
1924 
1925     nbd_client_connection_enable_retry(s->conn);
1926 
1927     return 0;
1928 
1929 fail:
1930     open_timer_del(s);
1931     nbd_clear_bdrvstate(bs);
1932     return ret;
1933 }
1934 
1935 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1936 {
1937     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1938     uint32_t min = s->info.min_block;
1939     uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1940 
1941     /*
1942      * If the server did not advertise an alignment:
1943      * - a size that is not sector-aligned implies that an alignment
1944      *   of 1 can be used to access those tail bytes
1945      * - advertisement of block status requires an alignment of 1, so
1946      *   that we don't violate block layer constraints that block
1947      *   status is always aligned (as we can't control whether the
1948      *   server will report sub-sector extents, such as a hole at EOF
1949      *   on an unaligned POSIX file)
1950      * - otherwise, assume the server is so old that we are safer avoiding
1951      *   sub-sector requests
1952      */
1953     if (!min) {
1954         min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1955                s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1956     }
1957 
1958     bs->bl.request_alignment = min;
1959     bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1960     bs->bl.max_pwrite_zeroes = max;
1961     bs->bl.max_transfer = max;
1962 
1963     /*
1964      * Assume that if the server supports extended headers, it also
1965      * supports unlimited size zero and trim commands.
1966      */
1967     if (s->info.mode >= NBD_MODE_EXTENDED) {
1968         bs->bl.max_pdiscard = bs->bl.max_pwrite_zeroes = 0;
1969     }
1970 
1971     if (s->info.opt_block &&
1972         s->info.opt_block > bs->bl.opt_transfer) {
1973         bs->bl.opt_transfer = s->info.opt_block;
1974     }
1975 }
1976 
1977 static void nbd_close(BlockDriverState *bs)
1978 {
1979     nbd_client_close(bs);
1980     nbd_clear_bdrvstate(bs);
1981 }
1982 
1983 /*
1984  * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1985  * to a smaller size with exact=false, there is no reason to fail the
1986  * operation.
1987  *
1988  * Preallocation mode is ignored since it does not seems useful to fail when
1989  * we never change anything.
1990  */
1991 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
1992                                         bool exact, PreallocMode prealloc,
1993                                         BdrvRequestFlags flags, Error **errp)
1994 {
1995     BDRVNBDState *s = bs->opaque;
1996 
1997     if (offset != s->info.size && exact) {
1998         error_setg(errp, "Cannot resize NBD nodes");
1999         return -ENOTSUP;
2000     }
2001 
2002     if (offset > s->info.size) {
2003         error_setg(errp, "Cannot grow NBD nodes");
2004         return -EINVAL;
2005     }
2006 
2007     return 0;
2008 }
2009 
2010 static int64_t coroutine_fn nbd_co_getlength(BlockDriverState *bs)
2011 {
2012     BDRVNBDState *s = bs->opaque;
2013 
2014     return s->info.size;
2015 }
2016 
2017 static void nbd_refresh_filename(BlockDriverState *bs)
2018 {
2019     BDRVNBDState *s = bs->opaque;
2020     const char *host = NULL, *port = NULL, *path = NULL;
2021     size_t len = 0;
2022 
2023     if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
2024         const InetSocketAddress *inet = &s->saddr->u.inet;
2025         if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
2026             host = inet->host;
2027             port = inet->port;
2028         }
2029     } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2030         path = s->saddr->u.q_unix.path;
2031     } /* else can't represent as pseudo-filename */
2032 
2033     if (path && s->export) {
2034         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2035                        "nbd+unix:///%s?socket=%s", s->export, path);
2036     } else if (path && !s->export) {
2037         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2038                        "nbd+unix://?socket=%s", path);
2039     } else if (host && s->export) {
2040         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2041                        "nbd://%s:%s/%s", host, port, s->export);
2042     } else if (host && !s->export) {
2043         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2044                        "nbd://%s:%s", host, port);
2045     }
2046     if (len >= sizeof(bs->exact_filename)) {
2047         /* Name is too long to represent exactly, so leave it empty. */
2048         bs->exact_filename[0] = '\0';
2049     }
2050 }
2051 
2052 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2053 {
2054     /* The generic bdrv_dirname() implementation is able to work out some
2055      * directory name for NBD nodes, but that would be wrong. So far there is no
2056      * specification for how "export paths" would work, so NBD does not have
2057      * directory names. */
2058     error_setg(errp, "Cannot generate a base directory for NBD nodes");
2059     return NULL;
2060 }
2061 
2062 static const char *const nbd_strong_runtime_opts[] = {
2063     "path",
2064     "host",
2065     "port",
2066     "export",
2067     "tls-creds",
2068     "tls-hostname",
2069     "server.",
2070 
2071     NULL
2072 };
2073 
2074 static void nbd_cancel_in_flight(BlockDriverState *bs)
2075 {
2076     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2077 
2078     reconnect_delay_timer_del(s);
2079 
2080     qemu_mutex_lock(&s->requests_lock);
2081     if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2082         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2083     }
2084     qemu_mutex_unlock(&s->requests_lock);
2085 
2086     nbd_co_establish_connection_cancel(s->conn);
2087 }
2088 
2089 static void nbd_attach_aio_context(BlockDriverState *bs,
2090                                    AioContext *new_context)
2091 {
2092     BDRVNBDState *s = bs->opaque;
2093 
2094     /* The open_timer is used only during nbd_open() */
2095     assert(!s->open_timer);
2096 
2097     /*
2098      * The reconnect_delay_timer is scheduled in I/O paths when the
2099      * connection is lost, to cancel the reconnection attempt after a
2100      * given time.  Once this attempt is done (successfully or not),
2101      * nbd_reconnect_attempt() ensures the timer is deleted before the
2102      * respective I/O request is resumed.
2103      * Since the AioContext can only be changed when a node is drained,
2104      * the reconnect_delay_timer cannot be active here.
2105      */
2106     assert(!s->reconnect_delay_timer);
2107 }
2108 
2109 static void nbd_detach_aio_context(BlockDriverState *bs)
2110 {
2111     BDRVNBDState *s = bs->opaque;
2112 
2113     assert(!s->open_timer);
2114     assert(!s->reconnect_delay_timer);
2115 }
2116 
2117 static BlockDriver bdrv_nbd = {
2118     .format_name                = "nbd",
2119     .protocol_name              = "nbd",
2120     .instance_size              = sizeof(BDRVNBDState),
2121     .bdrv_parse_filename        = nbd_parse_filename,
2122     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2123     .create_opts                = &bdrv_create_opts_simple,
2124     .bdrv_file_open             = nbd_open,
2125     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2126     .bdrv_co_preadv             = nbd_client_co_preadv,
2127     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2128     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2129     .bdrv_close                 = nbd_close,
2130     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2131     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2132     .bdrv_refresh_limits        = nbd_refresh_limits,
2133     .bdrv_co_truncate           = nbd_co_truncate,
2134     .bdrv_co_getlength          = nbd_co_getlength,
2135     .bdrv_refresh_filename      = nbd_refresh_filename,
2136     .bdrv_co_block_status       = nbd_client_co_block_status,
2137     .bdrv_dirname               = nbd_dirname,
2138     .strong_runtime_opts        = nbd_strong_runtime_opts,
2139     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2140 
2141     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2142     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2143 };
2144 
2145 static BlockDriver bdrv_nbd_tcp = {
2146     .format_name                = "nbd",
2147     .protocol_name              = "nbd+tcp",
2148     .instance_size              = sizeof(BDRVNBDState),
2149     .bdrv_parse_filename        = nbd_parse_filename,
2150     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2151     .create_opts                = &bdrv_create_opts_simple,
2152     .bdrv_file_open             = nbd_open,
2153     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2154     .bdrv_co_preadv             = nbd_client_co_preadv,
2155     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2156     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2157     .bdrv_close                 = nbd_close,
2158     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2159     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2160     .bdrv_refresh_limits        = nbd_refresh_limits,
2161     .bdrv_co_truncate           = nbd_co_truncate,
2162     .bdrv_co_getlength          = nbd_co_getlength,
2163     .bdrv_refresh_filename      = nbd_refresh_filename,
2164     .bdrv_co_block_status       = nbd_client_co_block_status,
2165     .bdrv_dirname               = nbd_dirname,
2166     .strong_runtime_opts        = nbd_strong_runtime_opts,
2167     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2168 
2169     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2170     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2171 };
2172 
2173 static BlockDriver bdrv_nbd_unix = {
2174     .format_name                = "nbd",
2175     .protocol_name              = "nbd+unix",
2176     .instance_size              = sizeof(BDRVNBDState),
2177     .bdrv_parse_filename        = nbd_parse_filename,
2178     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2179     .create_opts                = &bdrv_create_opts_simple,
2180     .bdrv_file_open             = nbd_open,
2181     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2182     .bdrv_co_preadv             = nbd_client_co_preadv,
2183     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2184     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2185     .bdrv_close                 = nbd_close,
2186     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2187     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2188     .bdrv_refresh_limits        = nbd_refresh_limits,
2189     .bdrv_co_truncate           = nbd_co_truncate,
2190     .bdrv_co_getlength          = nbd_co_getlength,
2191     .bdrv_refresh_filename      = nbd_refresh_filename,
2192     .bdrv_co_block_status       = nbd_client_co_block_status,
2193     .bdrv_dirname               = nbd_dirname,
2194     .strong_runtime_opts        = nbd_strong_runtime_opts,
2195     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2196 
2197     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2198     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2199 };
2200 
2201 static void bdrv_nbd_init(void)
2202 {
2203     bdrv_register(&bdrv_nbd);
2204     bdrv_register(&bdrv_nbd_tcp);
2205     bdrv_register(&bdrv_nbd_unix);
2206 }
2207 
2208 block_init(bdrv_nbd_init);
2209