xref: /openbmc/qemu/block/nbd.c (revision 8e32ddff)
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 };
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) && !s->info.structured_reply) {
467             nbd_channel_error(s, -EINVAL);
468             return -EINVAL;
469         }
470         ind2 = COOKIE_TO_INDEX(s->reply.cookie);
471         if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].coroutine) {
472             nbd_channel_error(s, -EINVAL);
473             return -EINVAL;
474         }
475         if (s->reply.cookie == cookie) {
476             /* We are done */
477             return 0;
478         }
479         nbd_recv_coroutine_wake_one(&s->requests[ind2]);
480     }
481 }
482 
483 static int coroutine_fn GRAPH_RDLOCK
484 nbd_co_send_request(BlockDriverState *bs, NBDRequest *request,
485                     QEMUIOVector *qiov)
486 {
487     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
488     int rc, i = -1;
489 
490     qemu_mutex_lock(&s->requests_lock);
491     while (s->in_flight == MAX_NBD_REQUESTS ||
492            (s->state != NBD_CLIENT_CONNECTED && s->in_flight > 0)) {
493         qemu_co_queue_wait(&s->free_sema, &s->requests_lock);
494     }
495 
496     s->in_flight++;
497     if (s->state != NBD_CLIENT_CONNECTED) {
498         if (nbd_client_connecting(s)) {
499             nbd_reconnect_attempt(s);
500             qemu_co_queue_restart_all(&s->free_sema);
501         }
502         if (s->state != NBD_CLIENT_CONNECTED) {
503             rc = -EIO;
504             goto err;
505         }
506     }
507 
508     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
509         if (s->requests[i].coroutine == NULL) {
510             break;
511         }
512     }
513 
514     assert(i < MAX_NBD_REQUESTS);
515     s->requests[i].coroutine = qemu_coroutine_self();
516     s->requests[i].offset = request->from;
517     s->requests[i].receiving = false;
518     qemu_mutex_unlock(&s->requests_lock);
519 
520     qemu_co_mutex_lock(&s->send_mutex);
521     request->cookie = INDEX_TO_COOKIE(i);
522 
523     assert(s->ioc);
524 
525     if (qiov) {
526         qio_channel_set_cork(s->ioc, true);
527         rc = nbd_send_request(s->ioc, request);
528         if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
529                                               NULL) < 0) {
530             rc = -EIO;
531         }
532         qio_channel_set_cork(s->ioc, false);
533     } else {
534         rc = nbd_send_request(s->ioc, request);
535     }
536     qemu_co_mutex_unlock(&s->send_mutex);
537 
538     if (rc < 0) {
539         qemu_mutex_lock(&s->requests_lock);
540 err:
541         nbd_channel_error_locked(s, rc);
542         if (i != -1) {
543             s->requests[i].coroutine = NULL;
544         }
545         s->in_flight--;
546         qemu_co_queue_next(&s->free_sema);
547         qemu_mutex_unlock(&s->requests_lock);
548     }
549     return rc;
550 }
551 
552 static inline uint16_t payload_advance16(uint8_t **payload)
553 {
554     *payload += 2;
555     return lduw_be_p(*payload - 2);
556 }
557 
558 static inline uint32_t payload_advance32(uint8_t **payload)
559 {
560     *payload += 4;
561     return ldl_be_p(*payload - 4);
562 }
563 
564 static inline uint64_t payload_advance64(uint8_t **payload)
565 {
566     *payload += 8;
567     return ldq_be_p(*payload - 8);
568 }
569 
570 static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
571                                          NBDStructuredReplyChunk *chunk,
572                                          uint8_t *payload, uint64_t orig_offset,
573                                          QEMUIOVector *qiov, Error **errp)
574 {
575     uint64_t offset;
576     uint32_t hole_size;
577 
578     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
579         error_setg(errp, "Protocol error: invalid payload for "
580                          "NBD_REPLY_TYPE_OFFSET_HOLE");
581         return -EINVAL;
582     }
583 
584     offset = payload_advance64(&payload);
585     hole_size = payload_advance32(&payload);
586 
587     if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
588         offset > orig_offset + qiov->size - hole_size) {
589         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
590                          " region");
591         return -EINVAL;
592     }
593     if (s->info.min_block &&
594         !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
595         trace_nbd_structured_read_compliance("hole");
596     }
597 
598     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
599 
600     return 0;
601 }
602 
603 /*
604  * nbd_parse_blockstatus_payload
605  * Based on our request, we expect only one extent in reply, for the
606  * base:allocation context.
607  */
608 static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
609                                          NBDStructuredReplyChunk *chunk,
610                                          uint8_t *payload, uint64_t orig_length,
611                                          NBDExtent *extent, Error **errp)
612 {
613     uint32_t context_id;
614 
615     /* The server succeeded, so it must have sent [at least] one extent */
616     if (chunk->length < sizeof(context_id) + sizeof(*extent)) {
617         error_setg(errp, "Protocol error: invalid payload for "
618                          "NBD_REPLY_TYPE_BLOCK_STATUS");
619         return -EINVAL;
620     }
621 
622     context_id = payload_advance32(&payload);
623     if (s->info.context_id != context_id) {
624         error_setg(errp, "Protocol error: unexpected context id %d for "
625                          "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
626                          "id is %d", context_id,
627                          s->info.context_id);
628         return -EINVAL;
629     }
630 
631     extent->length = payload_advance32(&payload);
632     extent->flags = payload_advance32(&payload);
633 
634     if (extent->length == 0) {
635         error_setg(errp, "Protocol error: server sent status chunk with "
636                    "zero length");
637         return -EINVAL;
638     }
639 
640     /*
641      * A server sending unaligned block status is in violation of the
642      * protocol, but as qemu-nbd 3.1 is such a server (at least for
643      * POSIX files that are not a multiple of 512 bytes, since qemu
644      * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
645      * still sees an implicit hole beyond the real EOF), it's nicer to
646      * work around the misbehaving server. If the request included
647      * more than the final unaligned block, truncate it back to an
648      * aligned result; if the request was only the final block, round
649      * up to the full block and change the status to fully-allocated
650      * (always a safe status, even if it loses information).
651      */
652     if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
653                                                    s->info.min_block)) {
654         trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
655         if (extent->length > s->info.min_block) {
656             extent->length = QEMU_ALIGN_DOWN(extent->length,
657                                              s->info.min_block);
658         } else {
659             extent->length = s->info.min_block;
660             extent->flags = 0;
661         }
662     }
663 
664     /*
665      * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
666      * sent us any more than one extent, nor should it have included
667      * status beyond our request in that extent. However, it's easy
668      * enough to ignore the server's noncompliance without killing the
669      * connection; just ignore trailing extents, and clamp things to
670      * the length of our request.
671      */
672     if (chunk->length > sizeof(context_id) + sizeof(*extent)) {
673         trace_nbd_parse_blockstatus_compliance("more than one extent");
674     }
675     if (extent->length > orig_length) {
676         extent->length = orig_length;
677         trace_nbd_parse_blockstatus_compliance("extent length too large");
678     }
679 
680     /*
681      * HACK: if we are using x-dirty-bitmaps to access
682      * qemu:allocation-depth, treat all depths > 2 the same as 2,
683      * since nbd_client_co_block_status is only expecting the low two
684      * bits to be set.
685      */
686     if (s->alloc_depth && extent->flags > 2) {
687         extent->flags = 2;
688     }
689 
690     return 0;
691 }
692 
693 /*
694  * nbd_parse_error_payload
695  * on success @errp contains message describing nbd error reply
696  */
697 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
698                                    uint8_t *payload, int *request_ret,
699                                    Error **errp)
700 {
701     uint32_t error;
702     uint16_t message_size;
703 
704     assert(chunk->type & (1 << 15));
705 
706     if (chunk->length < sizeof(error) + sizeof(message_size)) {
707         error_setg(errp,
708                    "Protocol error: invalid payload for structured error");
709         return -EINVAL;
710     }
711 
712     error = nbd_errno_to_system_errno(payload_advance32(&payload));
713     if (error == 0) {
714         error_setg(errp, "Protocol error: server sent structured error chunk "
715                          "with error = 0");
716         return -EINVAL;
717     }
718 
719     *request_ret = -error;
720     message_size = payload_advance16(&payload);
721 
722     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
723         error_setg(errp, "Protocol error: server sent structured error chunk "
724                          "with incorrect message size");
725         return -EINVAL;
726     }
727 
728     /* TODO: Add a trace point to mention the server complaint */
729 
730     /* TODO handle ERROR_OFFSET */
731 
732     return 0;
733 }
734 
735 static int coroutine_fn
736 nbd_co_receive_offset_data_payload(BDRVNBDState *s, uint64_t orig_offset,
737                                    QEMUIOVector *qiov, Error **errp)
738 {
739     QEMUIOVector sub_qiov;
740     uint64_t offset;
741     size_t data_size;
742     int ret;
743     NBDStructuredReplyChunk *chunk = &s->reply.structured;
744 
745     assert(nbd_reply_is_structured(&s->reply));
746 
747     /* The NBD spec requires at least one byte of payload */
748     if (chunk->length <= sizeof(offset)) {
749         error_setg(errp, "Protocol error: invalid payload for "
750                          "NBD_REPLY_TYPE_OFFSET_DATA");
751         return -EINVAL;
752     }
753 
754     if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
755         return -EIO;
756     }
757 
758     data_size = chunk->length - sizeof(offset);
759     assert(data_size);
760     if (offset < orig_offset || data_size > qiov->size ||
761         offset > orig_offset + qiov->size - data_size) {
762         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
763                          " region");
764         return -EINVAL;
765     }
766     if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
767         trace_nbd_structured_read_compliance("data");
768     }
769 
770     qemu_iovec_init(&sub_qiov, qiov->niov);
771     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
772     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
773     qemu_iovec_destroy(&sub_qiov);
774 
775     return ret < 0 ? -EIO : 0;
776 }
777 
778 #define NBD_MAX_MALLOC_PAYLOAD 1000
779 static coroutine_fn int nbd_co_receive_structured_payload(
780         BDRVNBDState *s, void **payload, Error **errp)
781 {
782     int ret;
783     uint32_t len;
784 
785     assert(nbd_reply_is_structured(&s->reply));
786 
787     len = s->reply.structured.length;
788 
789     if (len == 0) {
790         return 0;
791     }
792 
793     if (payload == NULL) {
794         error_setg(errp, "Unexpected structured payload");
795         return -EINVAL;
796     }
797 
798     if (len > NBD_MAX_MALLOC_PAYLOAD) {
799         error_setg(errp, "Payload too large");
800         return -EINVAL;
801     }
802 
803     *payload = g_new(char, len);
804     ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
805     if (ret < 0) {
806         g_free(*payload);
807         *payload = NULL;
808         return ret;
809     }
810 
811     return 0;
812 }
813 
814 /*
815  * nbd_co_do_receive_one_chunk
816  * for simple reply:
817  *   set request_ret to received reply error
818  *   if qiov is not NULL: read payload to @qiov
819  * for structured reply chunk:
820  *   if error chunk: read payload, set @request_ret, do not set @payload
821  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
822  *   else: read payload to @payload
823  *
824  * If function fails, @errp contains corresponding error message, and the
825  * connection with the server is suspect.  If it returns 0, then the
826  * transaction succeeded (although @request_ret may be a negative errno
827  * corresponding to the server's error reply), and errp is unchanged.
828  */
829 static coroutine_fn int nbd_co_do_receive_one_chunk(
830         BDRVNBDState *s, uint64_t cookie, bool only_structured,
831         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
832 {
833     int ret;
834     int i = COOKIE_TO_INDEX(cookie);
835     void *local_payload = NULL;
836     NBDStructuredReplyChunk *chunk;
837 
838     if (payload) {
839         *payload = NULL;
840     }
841     *request_ret = 0;
842 
843     ret = nbd_receive_replies(s, cookie);
844     if (ret < 0) {
845         error_setg(errp, "Connection closed");
846         return -EIO;
847     }
848     assert(s->ioc);
849 
850     assert(s->reply.cookie == cookie);
851 
852     if (nbd_reply_is_simple(&s->reply)) {
853         if (only_structured) {
854             error_setg(errp, "Protocol error: simple reply when structured "
855                              "reply chunk was expected");
856             return -EINVAL;
857         }
858 
859         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
860         if (*request_ret < 0 || !qiov) {
861             return 0;
862         }
863 
864         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
865                                      errp) < 0 ? -EIO : 0;
866     }
867 
868     /* handle structured reply chunk */
869     assert(s->info.structured_reply);
870     chunk = &s->reply.structured;
871 
872     if (chunk->type == NBD_REPLY_TYPE_NONE) {
873         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
874             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
875                        " NBD_REPLY_FLAG_DONE flag set");
876             return -EINVAL;
877         }
878         if (chunk->length) {
879             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
880                        " nonzero length");
881             return -EINVAL;
882         }
883         return 0;
884     }
885 
886     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
887         if (!qiov) {
888             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
889             return -EINVAL;
890         }
891 
892         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
893                                                   qiov, errp);
894     }
895 
896     if (nbd_reply_type_is_error(chunk->type)) {
897         payload = &local_payload;
898     }
899 
900     ret = nbd_co_receive_structured_payload(s, payload, errp);
901     if (ret < 0) {
902         return ret;
903     }
904 
905     if (nbd_reply_type_is_error(chunk->type)) {
906         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
907         g_free(local_payload);
908         return ret;
909     }
910 
911     return 0;
912 }
913 
914 /*
915  * nbd_co_receive_one_chunk
916  * Read reply, wake up connection_co and set s->quit if needed.
917  * Return value is a fatal error code or normal nbd reply error code
918  */
919 static coroutine_fn int nbd_co_receive_one_chunk(
920         BDRVNBDState *s, uint64_t cookie, bool only_structured,
921         int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
922         Error **errp)
923 {
924     int ret = nbd_co_do_receive_one_chunk(s, cookie, only_structured,
925                                           request_ret, qiov, payload, errp);
926 
927     if (ret < 0) {
928         memset(reply, 0, sizeof(*reply));
929         nbd_channel_error(s, ret);
930     } else {
931         /* For assert at loop start in nbd_connection_entry */
932         *reply = s->reply;
933     }
934     s->reply.cookie = 0;
935 
936     nbd_recv_coroutines_wake(s);
937 
938     return ret;
939 }
940 
941 typedef struct NBDReplyChunkIter {
942     int ret;
943     int request_ret;
944     Error *err;
945     bool done, only_structured;
946 } NBDReplyChunkIter;
947 
948 static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
949                                    int ret, Error **local_err)
950 {
951     assert(local_err && *local_err);
952     assert(ret < 0);
953 
954     if (!iter->ret) {
955         iter->ret = ret;
956         error_propagate(&iter->err, *local_err);
957     } else {
958         error_free(*local_err);
959     }
960 
961     *local_err = NULL;
962 }
963 
964 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
965 {
966     assert(ret < 0);
967 
968     if (!iter->request_ret) {
969         iter->request_ret = ret;
970     }
971 }
972 
973 /*
974  * NBD_FOREACH_REPLY_CHUNK
975  * The pointer stored in @payload requires g_free() to free it.
976  */
977 #define NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, structured, \
978                                 qiov, reply, payload) \
979     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
980          nbd_reply_chunk_iter_receive(s, &iter, cookie, qiov, reply, payload);)
981 
982 /*
983  * nbd_reply_chunk_iter_receive
984  * The pointer stored in @payload requires g_free() to free it.
985  */
986 static bool coroutine_fn nbd_reply_chunk_iter_receive(BDRVNBDState *s,
987                                                       NBDReplyChunkIter *iter,
988                                                       uint64_t cookie,
989                                                       QEMUIOVector *qiov,
990                                                       NBDReply *reply,
991                                                       void **payload)
992 {
993     int ret, request_ret;
994     NBDReply local_reply;
995     NBDStructuredReplyChunk *chunk;
996     Error *local_err = NULL;
997 
998     if (iter->done) {
999         /* Previous iteration was last. */
1000         goto break_loop;
1001     }
1002 
1003     if (reply == NULL) {
1004         reply = &local_reply;
1005     }
1006 
1007     ret = nbd_co_receive_one_chunk(s, cookie, iter->only_structured,
1008                                    &request_ret, qiov, reply, payload,
1009                                    &local_err);
1010     if (ret < 0) {
1011         nbd_iter_channel_error(iter, ret, &local_err);
1012     } else if (request_ret < 0) {
1013         nbd_iter_request_error(iter, request_ret);
1014     }
1015 
1016     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1017     if (nbd_reply_is_simple(reply) || iter->ret < 0) {
1018         goto break_loop;
1019     }
1020 
1021     chunk = &reply->structured;
1022     iter->only_structured = true;
1023 
1024     if (chunk->type == NBD_REPLY_TYPE_NONE) {
1025         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1026         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1027         goto break_loop;
1028     }
1029 
1030     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1031         /* This iteration is last. */
1032         iter->done = true;
1033     }
1034 
1035     /* Execute the loop body */
1036     return true;
1037 
1038 break_loop:
1039     qemu_mutex_lock(&s->requests_lock);
1040     s->requests[COOKIE_TO_INDEX(cookie)].coroutine = NULL;
1041     s->in_flight--;
1042     qemu_co_queue_next(&s->free_sema);
1043     qemu_mutex_unlock(&s->requests_lock);
1044 
1045     return false;
1046 }
1047 
1048 static int coroutine_fn
1049 nbd_co_receive_return_code(BDRVNBDState *s, uint64_t cookie,
1050                            int *request_ret, Error **errp)
1051 {
1052     NBDReplyChunkIter iter;
1053 
1054     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, NULL, NULL) {
1055         /* nbd_reply_chunk_iter_receive does all the work */
1056     }
1057 
1058     error_propagate(errp, iter.err);
1059     *request_ret = iter.request_ret;
1060     return iter.ret;
1061 }
1062 
1063 static int coroutine_fn
1064 nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t cookie,
1065                              uint64_t offset, QEMUIOVector *qiov,
1066                              int *request_ret, Error **errp)
1067 {
1068     NBDReplyChunkIter iter;
1069     NBDReply reply;
1070     void *payload = NULL;
1071     Error *local_err = NULL;
1072 
1073     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, s->info.structured_reply,
1074                             qiov, &reply, &payload)
1075     {
1076         int ret;
1077         NBDStructuredReplyChunk *chunk = &reply.structured;
1078 
1079         assert(nbd_reply_is_structured(&reply));
1080 
1081         switch (chunk->type) {
1082         case NBD_REPLY_TYPE_OFFSET_DATA:
1083             /*
1084              * special cased in nbd_co_receive_one_chunk, data is already
1085              * in qiov
1086              */
1087             break;
1088         case NBD_REPLY_TYPE_OFFSET_HOLE:
1089             ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1090                                                 offset, qiov, &local_err);
1091             if (ret < 0) {
1092                 nbd_channel_error(s, ret);
1093                 nbd_iter_channel_error(&iter, ret, &local_err);
1094             }
1095             break;
1096         default:
1097             if (!nbd_reply_type_is_error(chunk->type)) {
1098                 /* not allowed reply type */
1099                 nbd_channel_error(s, -EINVAL);
1100                 error_setg(&local_err,
1101                            "Unexpected reply type: %d (%s) for CMD_READ",
1102                            chunk->type, nbd_reply_type_lookup(chunk->type));
1103                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1104             }
1105         }
1106 
1107         g_free(payload);
1108         payload = NULL;
1109     }
1110 
1111     error_propagate(errp, iter.err);
1112     *request_ret = iter.request_ret;
1113     return iter.ret;
1114 }
1115 
1116 static int coroutine_fn
1117 nbd_co_receive_blockstatus_reply(BDRVNBDState *s, uint64_t cookie,
1118                                  uint64_t length, NBDExtent *extent,
1119                                  int *request_ret, Error **errp)
1120 {
1121     NBDReplyChunkIter iter;
1122     NBDReply reply;
1123     void *payload = NULL;
1124     Error *local_err = NULL;
1125     bool received = false;
1126 
1127     assert(!extent->length);
1128     NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, &reply, &payload) {
1129         int ret;
1130         NBDStructuredReplyChunk *chunk = &reply.structured;
1131 
1132         assert(nbd_reply_is_structured(&reply));
1133 
1134         switch (chunk->type) {
1135         case NBD_REPLY_TYPE_BLOCK_STATUS:
1136             if (received) {
1137                 nbd_channel_error(s, -EINVAL);
1138                 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1139                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1140             }
1141             received = true;
1142 
1143             ret = nbd_parse_blockstatus_payload(s, &reply.structured,
1144                                                 payload, length, extent,
1145                                                 &local_err);
1146             if (ret < 0) {
1147                 nbd_channel_error(s, ret);
1148                 nbd_iter_channel_error(&iter, ret, &local_err);
1149             }
1150             break;
1151         default:
1152             if (!nbd_reply_type_is_error(chunk->type)) {
1153                 nbd_channel_error(s, -EINVAL);
1154                 error_setg(&local_err,
1155                            "Unexpected reply type: %d (%s) "
1156                            "for CMD_BLOCK_STATUS",
1157                            chunk->type, nbd_reply_type_lookup(chunk->type));
1158                 nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1159             }
1160         }
1161 
1162         g_free(payload);
1163         payload = NULL;
1164     }
1165 
1166     if (!extent->length && !iter.request_ret) {
1167         error_setg(&local_err, "Server did not reply with any status extents");
1168         nbd_iter_channel_error(&iter, -EIO, &local_err);
1169     }
1170 
1171     error_propagate(errp, iter.err);
1172     *request_ret = iter.request_ret;
1173     return iter.ret;
1174 }
1175 
1176 static int coroutine_fn GRAPH_RDLOCK
1177 nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1178                QEMUIOVector *write_qiov)
1179 {
1180     int ret, request_ret;
1181     Error *local_err = NULL;
1182     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1183 
1184     assert(request->type != NBD_CMD_READ);
1185     if (write_qiov) {
1186         assert(request->type == NBD_CMD_WRITE);
1187         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1188     } else {
1189         assert(request->type != NBD_CMD_WRITE);
1190     }
1191 
1192     do {
1193         ret = nbd_co_send_request(bs, request, write_qiov);
1194         if (ret < 0) {
1195             continue;
1196         }
1197 
1198         ret = nbd_co_receive_return_code(s, request->cookie,
1199                                          &request_ret, &local_err);
1200         if (local_err) {
1201             trace_nbd_co_request_fail(request->from, request->len,
1202                                       request->cookie, request->flags,
1203                                       request->type,
1204                                       nbd_cmd_lookup(request->type),
1205                                       ret, error_get_pretty(local_err));
1206             error_free(local_err);
1207             local_err = NULL;
1208         }
1209     } while (ret < 0 && nbd_client_will_reconnect(s));
1210 
1211     return ret ? ret : request_ret;
1212 }
1213 
1214 static int coroutine_fn GRAPH_RDLOCK
1215 nbd_client_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1216                      QEMUIOVector *qiov, BdrvRequestFlags flags)
1217 {
1218     int ret, request_ret;
1219     Error *local_err = NULL;
1220     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1221     NBDRequest request = {
1222         .type = NBD_CMD_READ,
1223         .from = offset,
1224         .len = bytes,
1225     };
1226 
1227     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1228 
1229     if (!bytes) {
1230         return 0;
1231     }
1232     /*
1233      * Work around the fact that the block layer doesn't do
1234      * byte-accurate sizing yet - if the read exceeds the server's
1235      * advertised size because the block layer rounded size up, then
1236      * truncate the request to the server and tail-pad with zero.
1237      */
1238     if (offset >= s->info.size) {
1239         assert(bytes < BDRV_SECTOR_SIZE);
1240         qemu_iovec_memset(qiov, 0, 0, bytes);
1241         return 0;
1242     }
1243     if (offset + bytes > s->info.size) {
1244         uint64_t slop = offset + bytes - s->info.size;
1245 
1246         assert(slop < BDRV_SECTOR_SIZE);
1247         qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1248         request.len -= slop;
1249     }
1250 
1251     do {
1252         ret = nbd_co_send_request(bs, &request, NULL);
1253         if (ret < 0) {
1254             continue;
1255         }
1256 
1257         ret = nbd_co_receive_cmdread_reply(s, request.cookie, offset, qiov,
1258                                            &request_ret, &local_err);
1259         if (local_err) {
1260             trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1261                                       request.flags, request.type,
1262                                       nbd_cmd_lookup(request.type),
1263                                       ret, error_get_pretty(local_err));
1264             error_free(local_err);
1265             local_err = NULL;
1266         }
1267     } while (ret < 0 && nbd_client_will_reconnect(s));
1268 
1269     return ret ? ret : request_ret;
1270 }
1271 
1272 static int coroutine_fn GRAPH_RDLOCK
1273 nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
1274                       QEMUIOVector *qiov, BdrvRequestFlags flags)
1275 {
1276     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1277     NBDRequest request = {
1278         .type = NBD_CMD_WRITE,
1279         .from = offset,
1280         .len = bytes,
1281     };
1282 
1283     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1284     if (flags & BDRV_REQ_FUA) {
1285         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1286         request.flags |= NBD_CMD_FLAG_FUA;
1287     }
1288 
1289     assert(bytes <= NBD_MAX_BUFFER_SIZE);
1290 
1291     if (!bytes) {
1292         return 0;
1293     }
1294     return nbd_co_request(bs, &request, qiov);
1295 }
1296 
1297 static int coroutine_fn GRAPH_RDLOCK
1298 nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
1299                             BdrvRequestFlags flags)
1300 {
1301     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1302     NBDRequest request = {
1303         .type = NBD_CMD_WRITE_ZEROES,
1304         .from = offset,
1305         .len = bytes,  /* .len is uint32_t actually */
1306     };
1307 
1308     assert(bytes <= UINT32_MAX); /* rely on max_pwrite_zeroes */
1309 
1310     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1311     if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1312         return -ENOTSUP;
1313     }
1314 
1315     if (flags & BDRV_REQ_FUA) {
1316         assert(s->info.flags & NBD_FLAG_SEND_FUA);
1317         request.flags |= NBD_CMD_FLAG_FUA;
1318     }
1319     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1320         request.flags |= NBD_CMD_FLAG_NO_HOLE;
1321     }
1322     if (flags & BDRV_REQ_NO_FALLBACK) {
1323         assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1324         request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1325     }
1326 
1327     if (!bytes) {
1328         return 0;
1329     }
1330     return nbd_co_request(bs, &request, NULL);
1331 }
1332 
1333 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_flush(BlockDriverState *bs)
1334 {
1335     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1336     NBDRequest request = { .type = NBD_CMD_FLUSH };
1337 
1338     if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1339         return 0;
1340     }
1341 
1342     request.from = 0;
1343     request.len = 0;
1344 
1345     return nbd_co_request(bs, &request, NULL);
1346 }
1347 
1348 static int coroutine_fn GRAPH_RDLOCK
1349 nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
1350 {
1351     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1352     NBDRequest request = {
1353         .type = NBD_CMD_TRIM,
1354         .from = offset,
1355         .len = bytes, /* len is uint32_t */
1356     };
1357 
1358     assert(bytes <= UINT32_MAX); /* rely on max_pdiscard */
1359 
1360     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1361     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1362         return 0;
1363     }
1364 
1365     return nbd_co_request(bs, &request, NULL);
1366 }
1367 
1368 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_block_status(
1369         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1370         int64_t *pnum, int64_t *map, BlockDriverState **file)
1371 {
1372     int ret, request_ret;
1373     NBDExtent extent = { 0 };
1374     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1375     Error *local_err = NULL;
1376 
1377     NBDRequest request = {
1378         .type = NBD_CMD_BLOCK_STATUS,
1379         .from = offset,
1380         .len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1381                    MIN(bytes, s->info.size - offset)),
1382         .flags = NBD_CMD_FLAG_REQ_ONE,
1383     };
1384 
1385     if (!s->info.base_allocation) {
1386         *pnum = bytes;
1387         *map = offset;
1388         *file = bs;
1389         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1390     }
1391 
1392     /*
1393      * Work around the fact that the block layer doesn't do
1394      * byte-accurate sizing yet - if the status request exceeds the
1395      * server's advertised size because the block layer rounded size
1396      * up, we truncated the request to the server (above), or are
1397      * called on just the hole.
1398      */
1399     if (offset >= s->info.size) {
1400         *pnum = bytes;
1401         assert(bytes < BDRV_SECTOR_SIZE);
1402         /* Intentionally don't report offset_valid for the hole */
1403         return BDRV_BLOCK_ZERO;
1404     }
1405 
1406     if (s->info.min_block) {
1407         assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1408     }
1409     do {
1410         ret = nbd_co_send_request(bs, &request, NULL);
1411         if (ret < 0) {
1412             continue;
1413         }
1414 
1415         ret = nbd_co_receive_blockstatus_reply(s, request.cookie, bytes,
1416                                                &extent, &request_ret,
1417                                                &local_err);
1418         if (local_err) {
1419             trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1420                                       request.flags, request.type,
1421                                       nbd_cmd_lookup(request.type),
1422                                       ret, error_get_pretty(local_err));
1423             error_free(local_err);
1424             local_err = NULL;
1425         }
1426     } while (ret < 0 && nbd_client_will_reconnect(s));
1427 
1428     if (ret < 0 || request_ret < 0) {
1429         return ret ? ret : request_ret;
1430     }
1431 
1432     assert(extent.length);
1433     *pnum = extent.length;
1434     *map = offset;
1435     *file = bs;
1436     return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1437         (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1438         BDRV_BLOCK_OFFSET_VALID;
1439 }
1440 
1441 static int nbd_client_reopen_prepare(BDRVReopenState *state,
1442                                      BlockReopenQueue *queue, Error **errp)
1443 {
1444     BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1445 
1446     if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1447         error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1448         return -EACCES;
1449     }
1450     return 0;
1451 }
1452 
1453 static void nbd_yank(void *opaque)
1454 {
1455     BlockDriverState *bs = opaque;
1456     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1457 
1458     QEMU_LOCK_GUARD(&s->requests_lock);
1459     qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1460     s->state = NBD_CLIENT_QUIT;
1461 }
1462 
1463 static void nbd_client_close(BlockDriverState *bs)
1464 {
1465     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1466     NBDRequest request = { .type = NBD_CMD_DISC };
1467 
1468     if (s->ioc) {
1469         nbd_send_request(s->ioc, &request);
1470     }
1471 
1472     nbd_teardown_connection(bs);
1473 }
1474 
1475 
1476 /*
1477  * Parse nbd_open options
1478  */
1479 
1480 static int nbd_parse_uri(const char *filename, QDict *options)
1481 {
1482     URI *uri;
1483     const char *p;
1484     QueryParams *qp = NULL;
1485     int ret = 0;
1486     bool is_unix;
1487 
1488     uri = uri_parse(filename);
1489     if (!uri) {
1490         return -EINVAL;
1491     }
1492 
1493     /* transport */
1494     if (!g_strcmp0(uri->scheme, "nbd")) {
1495         is_unix = false;
1496     } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1497         is_unix = false;
1498     } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1499         is_unix = true;
1500     } else {
1501         ret = -EINVAL;
1502         goto out;
1503     }
1504 
1505     p = uri->path ? uri->path : "";
1506     if (p[0] == '/') {
1507         p++;
1508     }
1509     if (p[0]) {
1510         qdict_put_str(options, "export", p);
1511     }
1512 
1513     qp = query_params_parse(uri->query);
1514     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1515         ret = -EINVAL;
1516         goto out;
1517     }
1518 
1519     if (is_unix) {
1520         /* nbd+unix:///export?socket=path */
1521         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1522             ret = -EINVAL;
1523             goto out;
1524         }
1525         qdict_put_str(options, "server.type", "unix");
1526         qdict_put_str(options, "server.path", qp->p[0].value);
1527     } else {
1528         QString *host;
1529         char *port_str;
1530 
1531         /* nbd[+tcp]://host[:port]/export */
1532         if (!uri->server) {
1533             ret = -EINVAL;
1534             goto out;
1535         }
1536 
1537         /* strip braces from literal IPv6 address */
1538         if (uri->server[0] == '[') {
1539             host = qstring_from_substr(uri->server, 1,
1540                                        strlen(uri->server) - 1);
1541         } else {
1542             host = qstring_from_str(uri->server);
1543         }
1544 
1545         qdict_put_str(options, "server.type", "inet");
1546         qdict_put(options, "server.host", host);
1547 
1548         port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1549         qdict_put_str(options, "server.port", port_str);
1550         g_free(port_str);
1551     }
1552 
1553 out:
1554     if (qp) {
1555         query_params_free(qp);
1556     }
1557     uri_free(uri);
1558     return ret;
1559 }
1560 
1561 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1562 {
1563     const QDictEntry *e;
1564 
1565     for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1566         if (!strcmp(e->key, "host") ||
1567             !strcmp(e->key, "port") ||
1568             !strcmp(e->key, "path") ||
1569             !strcmp(e->key, "export") ||
1570             strstart(e->key, "server.", NULL))
1571         {
1572             error_setg(errp, "Option '%s' cannot be used with a file name",
1573                        e->key);
1574             return true;
1575         }
1576     }
1577 
1578     return false;
1579 }
1580 
1581 static void nbd_parse_filename(const char *filename, QDict *options,
1582                                Error **errp)
1583 {
1584     g_autofree char *file = NULL;
1585     char *export_name;
1586     const char *host_spec;
1587     const char *unixpath;
1588 
1589     if (nbd_has_filename_options_conflict(options, errp)) {
1590         return;
1591     }
1592 
1593     if (strstr(filename, "://")) {
1594         int ret = nbd_parse_uri(filename, options);
1595         if (ret < 0) {
1596             error_setg(errp, "No valid URL specified");
1597         }
1598         return;
1599     }
1600 
1601     file = g_strdup(filename);
1602 
1603     export_name = strstr(file, EN_OPTSTR);
1604     if (export_name) {
1605         if (export_name[strlen(EN_OPTSTR)] == 0) {
1606             return;
1607         }
1608         export_name[0] = 0; /* truncate 'file' */
1609         export_name += strlen(EN_OPTSTR);
1610 
1611         qdict_put_str(options, "export", export_name);
1612     }
1613 
1614     /* extract the host_spec - fail if it's not nbd:... */
1615     if (!strstart(file, "nbd:", &host_spec)) {
1616         error_setg(errp, "File name string for NBD must start with 'nbd:'");
1617         return;
1618     }
1619 
1620     if (!*host_spec) {
1621         return;
1622     }
1623 
1624     /* are we a UNIX or TCP socket? */
1625     if (strstart(host_spec, "unix:", &unixpath)) {
1626         qdict_put_str(options, "server.type", "unix");
1627         qdict_put_str(options, "server.path", unixpath);
1628     } else {
1629         InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1630 
1631         if (inet_parse(addr, host_spec, errp)) {
1632             goto out_inet;
1633         }
1634 
1635         qdict_put_str(options, "server.type", "inet");
1636         qdict_put_str(options, "server.host", addr->host);
1637         qdict_put_str(options, "server.port", addr->port);
1638     out_inet:
1639         qapi_free_InetSocketAddress(addr);
1640     }
1641 }
1642 
1643 static bool nbd_process_legacy_socket_options(QDict *output_options,
1644                                               QemuOpts *legacy_opts,
1645                                               Error **errp)
1646 {
1647     const char *path = qemu_opt_get(legacy_opts, "path");
1648     const char *host = qemu_opt_get(legacy_opts, "host");
1649     const char *port = qemu_opt_get(legacy_opts, "port");
1650     const QDictEntry *e;
1651 
1652     if (!path && !host && !port) {
1653         return true;
1654     }
1655 
1656     for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1657     {
1658         if (strstart(e->key, "server.", NULL)) {
1659             error_setg(errp, "Cannot use 'server' and path/host/port at the "
1660                        "same time");
1661             return false;
1662         }
1663     }
1664 
1665     if (path && host) {
1666         error_setg(errp, "path and host may not be used at the same time");
1667         return false;
1668     } else if (path) {
1669         if (port) {
1670             error_setg(errp, "port may not be used without host");
1671             return false;
1672         }
1673 
1674         qdict_put_str(output_options, "server.type", "unix");
1675         qdict_put_str(output_options, "server.path", path);
1676     } else if (host) {
1677         qdict_put_str(output_options, "server.type", "inet");
1678         qdict_put_str(output_options, "server.host", host);
1679         qdict_put_str(output_options, "server.port",
1680                       port ?: stringify(NBD_DEFAULT_PORT));
1681     }
1682 
1683     return true;
1684 }
1685 
1686 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1687                                  Error **errp)
1688 {
1689     SocketAddress *saddr = NULL;
1690     QDict *addr = NULL;
1691     Visitor *iv = NULL;
1692 
1693     qdict_extract_subqdict(options, &addr, "server.");
1694     if (!qdict_size(addr)) {
1695         error_setg(errp, "NBD server address missing");
1696         goto done;
1697     }
1698 
1699     iv = qobject_input_visitor_new_flat_confused(addr, errp);
1700     if (!iv) {
1701         goto done;
1702     }
1703 
1704     if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1705         goto done;
1706     }
1707 
1708     if (socket_address_parse_named_fd(saddr, errp) < 0) {
1709         qapi_free_SocketAddress(saddr);
1710         saddr = NULL;
1711         goto done;
1712     }
1713 
1714 done:
1715     qobject_unref(addr);
1716     visit_free(iv);
1717     return saddr;
1718 }
1719 
1720 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1721 {
1722     Object *obj;
1723     QCryptoTLSCreds *creds;
1724 
1725     obj = object_resolve_path_component(
1726         object_get_objects_root(), id);
1727     if (!obj) {
1728         error_setg(errp, "No TLS credentials with id '%s'",
1729                    id);
1730         return NULL;
1731     }
1732     creds = (QCryptoTLSCreds *)
1733         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1734     if (!creds) {
1735         error_setg(errp, "Object with id '%s' is not TLS credentials",
1736                    id);
1737         return NULL;
1738     }
1739 
1740     if (!qcrypto_tls_creds_check_endpoint(creds,
1741                                           QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1742                                           errp)) {
1743         return NULL;
1744     }
1745     object_ref(obj);
1746     return creds;
1747 }
1748 
1749 
1750 static QemuOptsList nbd_runtime_opts = {
1751     .name = "nbd",
1752     .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1753     .desc = {
1754         {
1755             .name = "host",
1756             .type = QEMU_OPT_STRING,
1757             .help = "TCP host to connect to",
1758         },
1759         {
1760             .name = "port",
1761             .type = QEMU_OPT_STRING,
1762             .help = "TCP port to connect to",
1763         },
1764         {
1765             .name = "path",
1766             .type = QEMU_OPT_STRING,
1767             .help = "Unix socket path to connect to",
1768         },
1769         {
1770             .name = "export",
1771             .type = QEMU_OPT_STRING,
1772             .help = "Name of the NBD export to open",
1773         },
1774         {
1775             .name = "tls-creds",
1776             .type = QEMU_OPT_STRING,
1777             .help = "ID of the TLS credentials to use",
1778         },
1779         {
1780             .name = "tls-hostname",
1781             .type = QEMU_OPT_STRING,
1782             .help = "Override hostname for validating TLS x509 certificate",
1783         },
1784         {
1785             .name = "x-dirty-bitmap",
1786             .type = QEMU_OPT_STRING,
1787             .help = "experimental: expose named dirty bitmap in place of "
1788                     "block status",
1789         },
1790         {
1791             .name = "reconnect-delay",
1792             .type = QEMU_OPT_NUMBER,
1793             .help = "On an unexpected disconnect, the nbd client tries to "
1794                     "connect again until succeeding or encountering a serious "
1795                     "error.  During the first @reconnect-delay seconds, all "
1796                     "requests are paused and will be rerun on a successful "
1797                     "reconnect. After that time, any delayed requests and all "
1798                     "future requests before a successful reconnect will "
1799                     "immediately fail. Default 0",
1800         },
1801         {
1802             .name = "open-timeout",
1803             .type = QEMU_OPT_NUMBER,
1804             .help = "In seconds. If zero, the nbd driver tries the connection "
1805                     "only once, and fails to open if the connection fails. "
1806                     "If non-zero, the nbd driver will repeat connection "
1807                     "attempts until successful or until @open-timeout seconds "
1808                     "have elapsed. Default 0",
1809         },
1810         { /* end of list */ }
1811     },
1812 };
1813 
1814 static int nbd_process_options(BlockDriverState *bs, QDict *options,
1815                                Error **errp)
1816 {
1817     BDRVNBDState *s = bs->opaque;
1818     QemuOpts *opts;
1819     int ret = -EINVAL;
1820 
1821     opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1822     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1823         goto error;
1824     }
1825 
1826     /* Translate @host, @port, and @path to a SocketAddress */
1827     if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1828         goto error;
1829     }
1830 
1831     /* Pop the config into our state object. Exit if invalid. */
1832     s->saddr = nbd_config(s, options, errp);
1833     if (!s->saddr) {
1834         goto error;
1835     }
1836 
1837     s->export = g_strdup(qemu_opt_get(opts, "export"));
1838     if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1839         error_setg(errp, "export name too long to send to server");
1840         goto error;
1841     }
1842 
1843     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1844     if (s->tlscredsid) {
1845         s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1846         if (!s->tlscreds) {
1847             goto error;
1848         }
1849 
1850         s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname"));
1851         if (!s->tlshostname &&
1852             s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1853             s->tlshostname = g_strdup(s->saddr->u.inet.host);
1854         }
1855     }
1856 
1857     s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1858     if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1859         error_setg(errp, "x-dirty-bitmap query too long to send to server");
1860         goto error;
1861     }
1862 
1863     s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1864     s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1865 
1866     ret = 0;
1867 
1868  error:
1869     qemu_opts_del(opts);
1870     return ret;
1871 }
1872 
1873 static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1874                     Error **errp)
1875 {
1876     int ret;
1877     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1878 
1879     s->bs = bs;
1880     qemu_mutex_init(&s->requests_lock);
1881     qemu_co_queue_init(&s->free_sema);
1882     qemu_co_mutex_init(&s->send_mutex);
1883     qemu_co_mutex_init(&s->receive_mutex);
1884 
1885     if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1886         return -EEXIST;
1887     }
1888 
1889     ret = nbd_process_options(bs, options, errp);
1890     if (ret < 0) {
1891         goto fail;
1892     }
1893 
1894     s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1895                                         s->x_dirty_bitmap, s->tlscreds,
1896                                         s->tlshostname);
1897 
1898     if (s->open_timeout) {
1899         nbd_client_connection_enable_retry(s->conn);
1900         open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1901                         s->open_timeout * NANOSECONDS_PER_SECOND);
1902     }
1903 
1904     s->state = NBD_CLIENT_CONNECTING_WAIT;
1905     ret = nbd_do_establish_connection(bs, true, errp);
1906     if (ret < 0) {
1907         goto fail;
1908     }
1909 
1910     /*
1911      * The connect attempt is done, so we no longer need this timer.
1912      * Delete it, because we do not want it to be around when this node
1913      * is drained or closed.
1914      */
1915     open_timer_del(s);
1916 
1917     nbd_client_connection_enable_retry(s->conn);
1918 
1919     return 0;
1920 
1921 fail:
1922     open_timer_del(s);
1923     nbd_clear_bdrvstate(bs);
1924     return ret;
1925 }
1926 
1927 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1928 {
1929     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1930     uint32_t min = s->info.min_block;
1931     uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1932 
1933     /*
1934      * If the server did not advertise an alignment:
1935      * - a size that is not sector-aligned implies that an alignment
1936      *   of 1 can be used to access those tail bytes
1937      * - advertisement of block status requires an alignment of 1, so
1938      *   that we don't violate block layer constraints that block
1939      *   status is always aligned (as we can't control whether the
1940      *   server will report sub-sector extents, such as a hole at EOF
1941      *   on an unaligned POSIX file)
1942      * - otherwise, assume the server is so old that we are safer avoiding
1943      *   sub-sector requests
1944      */
1945     if (!min) {
1946         min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1947                s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1948     }
1949 
1950     bs->bl.request_alignment = min;
1951     bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1952     bs->bl.max_pwrite_zeroes = max;
1953     bs->bl.max_transfer = max;
1954 
1955     if (s->info.opt_block &&
1956         s->info.opt_block > bs->bl.opt_transfer) {
1957         bs->bl.opt_transfer = s->info.opt_block;
1958     }
1959 }
1960 
1961 static void nbd_close(BlockDriverState *bs)
1962 {
1963     nbd_client_close(bs);
1964     nbd_clear_bdrvstate(bs);
1965 }
1966 
1967 /*
1968  * NBD cannot truncate, but if the caller asks to truncate to the same size, or
1969  * to a smaller size with exact=false, there is no reason to fail the
1970  * operation.
1971  *
1972  * Preallocation mode is ignored since it does not seems useful to fail when
1973  * we never change anything.
1974  */
1975 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
1976                                         bool exact, PreallocMode prealloc,
1977                                         BdrvRequestFlags flags, Error **errp)
1978 {
1979     BDRVNBDState *s = bs->opaque;
1980 
1981     if (offset != s->info.size && exact) {
1982         error_setg(errp, "Cannot resize NBD nodes");
1983         return -ENOTSUP;
1984     }
1985 
1986     if (offset > s->info.size) {
1987         error_setg(errp, "Cannot grow NBD nodes");
1988         return -EINVAL;
1989     }
1990 
1991     return 0;
1992 }
1993 
1994 static int64_t coroutine_fn nbd_co_getlength(BlockDriverState *bs)
1995 {
1996     BDRVNBDState *s = bs->opaque;
1997 
1998     return s->info.size;
1999 }
2000 
2001 static void nbd_refresh_filename(BlockDriverState *bs)
2002 {
2003     BDRVNBDState *s = bs->opaque;
2004     const char *host = NULL, *port = NULL, *path = NULL;
2005     size_t len = 0;
2006 
2007     if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
2008         const InetSocketAddress *inet = &s->saddr->u.inet;
2009         if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
2010             host = inet->host;
2011             port = inet->port;
2012         }
2013     } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2014         path = s->saddr->u.q_unix.path;
2015     } /* else can't represent as pseudo-filename */
2016 
2017     if (path && s->export) {
2018         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2019                        "nbd+unix:///%s?socket=%s", s->export, path);
2020     } else if (path && !s->export) {
2021         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2022                        "nbd+unix://?socket=%s", path);
2023     } else if (host && s->export) {
2024         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2025                        "nbd://%s:%s/%s", host, port, s->export);
2026     } else if (host && !s->export) {
2027         len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2028                        "nbd://%s:%s", host, port);
2029     }
2030     if (len >= sizeof(bs->exact_filename)) {
2031         /* Name is too long to represent exactly, so leave it empty. */
2032         bs->exact_filename[0] = '\0';
2033     }
2034 }
2035 
2036 static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2037 {
2038     /* The generic bdrv_dirname() implementation is able to work out some
2039      * directory name for NBD nodes, but that would be wrong. So far there is no
2040      * specification for how "export paths" would work, so NBD does not have
2041      * directory names. */
2042     error_setg(errp, "Cannot generate a base directory for NBD nodes");
2043     return NULL;
2044 }
2045 
2046 static const char *const nbd_strong_runtime_opts[] = {
2047     "path",
2048     "host",
2049     "port",
2050     "export",
2051     "tls-creds",
2052     "tls-hostname",
2053     "server.",
2054 
2055     NULL
2056 };
2057 
2058 static void nbd_cancel_in_flight(BlockDriverState *bs)
2059 {
2060     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2061 
2062     reconnect_delay_timer_del(s);
2063 
2064     qemu_mutex_lock(&s->requests_lock);
2065     if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2066         s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2067     }
2068     qemu_mutex_unlock(&s->requests_lock);
2069 
2070     nbd_co_establish_connection_cancel(s->conn);
2071 }
2072 
2073 static void nbd_attach_aio_context(BlockDriverState *bs,
2074                                    AioContext *new_context)
2075 {
2076     BDRVNBDState *s = bs->opaque;
2077 
2078     /* The open_timer is used only during nbd_open() */
2079     assert(!s->open_timer);
2080 
2081     /*
2082      * The reconnect_delay_timer is scheduled in I/O paths when the
2083      * connection is lost, to cancel the reconnection attempt after a
2084      * given time.  Once this attempt is done (successfully or not),
2085      * nbd_reconnect_attempt() ensures the timer is deleted before the
2086      * respective I/O request is resumed.
2087      * Since the AioContext can only be changed when a node is drained,
2088      * the reconnect_delay_timer cannot be active here.
2089      */
2090     assert(!s->reconnect_delay_timer);
2091 }
2092 
2093 static void nbd_detach_aio_context(BlockDriverState *bs)
2094 {
2095     BDRVNBDState *s = bs->opaque;
2096 
2097     assert(!s->open_timer);
2098     assert(!s->reconnect_delay_timer);
2099 }
2100 
2101 static BlockDriver bdrv_nbd = {
2102     .format_name                = "nbd",
2103     .protocol_name              = "nbd",
2104     .instance_size              = sizeof(BDRVNBDState),
2105     .bdrv_parse_filename        = nbd_parse_filename,
2106     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2107     .create_opts                = &bdrv_create_opts_simple,
2108     .bdrv_file_open             = nbd_open,
2109     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2110     .bdrv_co_preadv             = nbd_client_co_preadv,
2111     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2112     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2113     .bdrv_close                 = nbd_close,
2114     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2115     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2116     .bdrv_refresh_limits        = nbd_refresh_limits,
2117     .bdrv_co_truncate           = nbd_co_truncate,
2118     .bdrv_co_getlength          = nbd_co_getlength,
2119     .bdrv_refresh_filename      = nbd_refresh_filename,
2120     .bdrv_co_block_status       = nbd_client_co_block_status,
2121     .bdrv_dirname               = nbd_dirname,
2122     .strong_runtime_opts        = nbd_strong_runtime_opts,
2123     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2124 
2125     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2126     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2127 };
2128 
2129 static BlockDriver bdrv_nbd_tcp = {
2130     .format_name                = "nbd",
2131     .protocol_name              = "nbd+tcp",
2132     .instance_size              = sizeof(BDRVNBDState),
2133     .bdrv_parse_filename        = nbd_parse_filename,
2134     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2135     .create_opts                = &bdrv_create_opts_simple,
2136     .bdrv_file_open             = nbd_open,
2137     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2138     .bdrv_co_preadv             = nbd_client_co_preadv,
2139     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2140     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2141     .bdrv_close                 = nbd_close,
2142     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2143     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2144     .bdrv_refresh_limits        = nbd_refresh_limits,
2145     .bdrv_co_truncate           = nbd_co_truncate,
2146     .bdrv_co_getlength          = nbd_co_getlength,
2147     .bdrv_refresh_filename      = nbd_refresh_filename,
2148     .bdrv_co_block_status       = nbd_client_co_block_status,
2149     .bdrv_dirname               = nbd_dirname,
2150     .strong_runtime_opts        = nbd_strong_runtime_opts,
2151     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2152 
2153     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2154     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2155 };
2156 
2157 static BlockDriver bdrv_nbd_unix = {
2158     .format_name                = "nbd",
2159     .protocol_name              = "nbd+unix",
2160     .instance_size              = sizeof(BDRVNBDState),
2161     .bdrv_parse_filename        = nbd_parse_filename,
2162     .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2163     .create_opts                = &bdrv_create_opts_simple,
2164     .bdrv_file_open             = nbd_open,
2165     .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2166     .bdrv_co_preadv             = nbd_client_co_preadv,
2167     .bdrv_co_pwritev            = nbd_client_co_pwritev,
2168     .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2169     .bdrv_close                 = nbd_close,
2170     .bdrv_co_flush_to_os        = nbd_client_co_flush,
2171     .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2172     .bdrv_refresh_limits        = nbd_refresh_limits,
2173     .bdrv_co_truncate           = nbd_co_truncate,
2174     .bdrv_co_getlength          = nbd_co_getlength,
2175     .bdrv_refresh_filename      = nbd_refresh_filename,
2176     .bdrv_co_block_status       = nbd_client_co_block_status,
2177     .bdrv_dirname               = nbd_dirname,
2178     .strong_runtime_opts        = nbd_strong_runtime_opts,
2179     .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2180 
2181     .bdrv_attach_aio_context    = nbd_attach_aio_context,
2182     .bdrv_detach_aio_context    = nbd_detach_aio_context,
2183 };
2184 
2185 static void bdrv_nbd_init(void)
2186 {
2187     bdrv_register(&bdrv_nbd);
2188     bdrv_register(&bdrv_nbd_tcp);
2189     bdrv_register(&bdrv_nbd_unix);
2190 }
2191 
2192 block_init(bdrv_nbd_init);
2193