xref: /openbmc/qemu/block/nbd.c (revision 1f9d714e9a92d382f5f7c054de647d3dde4ba803)
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 coroutine_fn GRAPH_RDLOCK
279  nbd_handle_updated_info(BlockDriverState *bs, Error **errp)
280  {
281      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
282      int ret;
283  
284      if (s->x_dirty_bitmap) {
285          if (!s->info.base_allocation) {
286              error_setg(errp, "requested x-dirty-bitmap %s not found",
287                         s->x_dirty_bitmap);
288              return -EINVAL;
289          }
290          if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) {
291              s->alloc_depth = true;
292          }
293      }
294  
295      if (s->info.flags & NBD_FLAG_READ_ONLY) {
296          ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp);
297          if (ret < 0) {
298              return ret;
299          }
300      }
301  
302      if (s->info.flags & NBD_FLAG_SEND_FUA) {
303          bs->supported_write_flags = BDRV_REQ_FUA;
304          bs->supported_zero_flags |= BDRV_REQ_FUA;
305      }
306  
307      if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
308          bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
309          if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) {
310              bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK;
311          }
312      }
313  
314      trace_nbd_client_handshake_success(s->export);
315  
316      return 0;
317  }
318  
319  int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs,
320                                                  bool blocking, Error **errp)
321  {
322      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
323      int ret;
324      IO_CODE();
325  
326      assert_bdrv_graph_readable();
327      assert(!s->ioc);
328  
329      s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp);
330      if (!s->ioc) {
331          return -ECONNREFUSED;
332      }
333  
334      yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank,
335                             bs);
336  
337      ret = nbd_handle_updated_info(s->bs, NULL);
338      if (ret < 0) {
339          /*
340           * We have connected, but must fail for other reasons.
341           * Send NBD_CMD_DISC as a courtesy to the server.
342           */
343          NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode };
344  
345          nbd_send_request(s->ioc, &request);
346  
347          yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
348                                   nbd_yank, bs);
349          object_unref(OBJECT(s->ioc));
350          s->ioc = NULL;
351  
352          return ret;
353      }
354  
355      qio_channel_set_blocking(s->ioc, false, NULL);
356      qio_channel_set_follow_coroutine_ctx(s->ioc, true);
357  
358      /* successfully connected */
359      WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
360          s->state = NBD_CLIENT_CONNECTED;
361      }
362  
363      return 0;
364  }
365  
366  /* Called with s->requests_lock held.  */
367  static bool nbd_client_connecting(BDRVNBDState *s)
368  {
369      return s->state == NBD_CLIENT_CONNECTING_WAIT ||
370          s->state == NBD_CLIENT_CONNECTING_NOWAIT;
371  }
372  
373  /* Called with s->requests_lock taken.  */
374  static void coroutine_fn GRAPH_RDLOCK nbd_reconnect_attempt(BDRVNBDState *s)
375  {
376      int ret;
377      bool blocking = s->state == NBD_CLIENT_CONNECTING_WAIT;
378  
379      /*
380       * Now we are sure that nobody is accessing the channel, and no one will
381       * try until we set the state to CONNECTED.
382       */
383      assert(nbd_client_connecting(s));
384      assert(s->in_flight == 1);
385  
386      trace_nbd_reconnect_attempt(s->bs->in_flight);
387  
388      if (blocking && !s->reconnect_delay_timer) {
389          /*
390           * It's the first reconnect attempt after switching to
391           * NBD_CLIENT_CONNECTING_WAIT
392           */
393          g_assert(s->reconnect_delay);
394          reconnect_delay_timer_init(s,
395              qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
396              s->reconnect_delay * NANOSECONDS_PER_SECOND);
397      }
398  
399      /* Finalize previous connection if any */
400      if (s->ioc) {
401          yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name),
402                                   nbd_yank, s->bs);
403          object_unref(OBJECT(s->ioc));
404          s->ioc = NULL;
405      }
406  
407      qemu_mutex_unlock(&s->requests_lock);
408      ret = nbd_co_do_establish_connection(s->bs, blocking, NULL);
409      trace_nbd_reconnect_attempt_result(ret, s->bs->in_flight);
410      qemu_mutex_lock(&s->requests_lock);
411  
412      /*
413       * The reconnect attempt is done (maybe successfully, maybe not), so
414       * we no longer need this timer.  Delete it so it will not outlive
415       * this I/O request (so draining removes all timers).
416       */
417      reconnect_delay_timer_del(s);
418  }
419  
420  static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t cookie,
421                                              Error **errp)
422  {
423      int ret;
424      uint64_t ind = COOKIE_TO_INDEX(cookie), ind2;
425      QEMU_LOCK_GUARD(&s->receive_mutex);
426  
427      while (true) {
428          if (s->reply.cookie == cookie) {
429              /* We are done */
430              return 0;
431          }
432  
433          if (s->reply.cookie != 0) {
434              /*
435               * Some other request is being handled now. It should already be
436               * woken by whoever set s->reply.cookie (or never wait in this
437               * yield). So, we should not wake it here.
438               */
439              ind2 = COOKIE_TO_INDEX(s->reply.cookie);
440              assert(!s->requests[ind2].receiving);
441  
442              s->requests[ind].receiving = true;
443              qemu_co_mutex_unlock(&s->receive_mutex);
444  
445              qemu_coroutine_yield();
446              /*
447               * We may be woken for 2 reasons:
448               * 1. From this function, executing in parallel coroutine, when our
449               *    cookie is received.
450               * 2. From nbd_co_receive_one_chunk(), when previous request is
451               *    finished and s->reply.cookie set to 0.
452               * Anyway, it's OK to lock the mutex and go to the next iteration.
453               */
454  
455              qemu_co_mutex_lock(&s->receive_mutex);
456              assert(!s->requests[ind].receiving);
457              continue;
458          }
459  
460          /* We are under mutex and cookie is 0. We have to do the dirty work. */
461          assert(s->reply.cookie == 0);
462          ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, s->info.mode, errp);
463          if (ret == 0) {
464              ret = -EIO;
465              error_setg(errp, "server dropped connection");
466          }
467          if (ret < 0) {
468              nbd_channel_error(s, ret);
469              return ret;
470          }
471          if (nbd_reply_is_structured(&s->reply) &&
472              s->info.mode < NBD_MODE_STRUCTURED) {
473              nbd_channel_error(s, -EINVAL);
474              error_setg(errp, "unexpected structured reply");
475              return -EINVAL;
476          }
477          ind2 = COOKIE_TO_INDEX(s->reply.cookie);
478          if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].coroutine) {
479              nbd_channel_error(s, -EINVAL);
480              error_setg(errp, "unexpected cookie value");
481              return -EINVAL;
482          }
483          if (s->reply.cookie == cookie) {
484              /* We are done */
485              return 0;
486          }
487          nbd_recv_coroutine_wake_one(&s->requests[ind2]);
488      }
489  }
490  
491  static int coroutine_fn GRAPH_RDLOCK
492  nbd_co_send_request(BlockDriverState *bs, NBDRequest *request,
493                      QEMUIOVector *qiov)
494  {
495      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
496      int rc, i = -1;
497  
498      qemu_mutex_lock(&s->requests_lock);
499      while (s->in_flight == MAX_NBD_REQUESTS ||
500             (s->state != NBD_CLIENT_CONNECTED && s->in_flight > 0)) {
501          qemu_co_queue_wait(&s->free_sema, &s->requests_lock);
502      }
503  
504      s->in_flight++;
505      if (s->state != NBD_CLIENT_CONNECTED) {
506          if (nbd_client_connecting(s)) {
507              nbd_reconnect_attempt(s);
508              qemu_co_queue_restart_all(&s->free_sema);
509          }
510          if (s->state != NBD_CLIENT_CONNECTED) {
511              rc = -EIO;
512              goto err;
513          }
514      }
515  
516      for (i = 0; i < MAX_NBD_REQUESTS; i++) {
517          if (s->requests[i].coroutine == NULL) {
518              break;
519          }
520      }
521  
522      assert(i < MAX_NBD_REQUESTS);
523      s->requests[i].coroutine = qemu_coroutine_self();
524      s->requests[i].offset = request->from;
525      s->requests[i].receiving = false;
526      qemu_mutex_unlock(&s->requests_lock);
527  
528      qemu_co_mutex_lock(&s->send_mutex);
529      request->cookie = INDEX_TO_COOKIE(i);
530      request->mode = s->info.mode;
531  
532      assert(s->ioc);
533  
534      if (qiov) {
535          qio_channel_set_cork(s->ioc, true);
536          rc = nbd_send_request(s->ioc, request);
537          if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
538                                                NULL) < 0) {
539              rc = -EIO;
540          }
541          qio_channel_set_cork(s->ioc, false);
542      } else {
543          rc = nbd_send_request(s->ioc, request);
544      }
545      qemu_co_mutex_unlock(&s->send_mutex);
546  
547      if (rc < 0) {
548          qemu_mutex_lock(&s->requests_lock);
549  err:
550          nbd_channel_error_locked(s, rc);
551          if (i != -1) {
552              s->requests[i].coroutine = NULL;
553          }
554          s->in_flight--;
555          qemu_co_queue_next(&s->free_sema);
556          qemu_mutex_unlock(&s->requests_lock);
557      }
558      return rc;
559  }
560  
561  static inline uint16_t payload_advance16(uint8_t **payload)
562  {
563      *payload += 2;
564      return lduw_be_p(*payload - 2);
565  }
566  
567  static inline uint32_t payload_advance32(uint8_t **payload)
568  {
569      *payload += 4;
570      return ldl_be_p(*payload - 4);
571  }
572  
573  static inline uint64_t payload_advance64(uint8_t **payload)
574  {
575      *payload += 8;
576      return ldq_be_p(*payload - 8);
577  }
578  
579  static int nbd_parse_offset_hole_payload(BDRVNBDState *s,
580                                           NBDStructuredReplyChunk *chunk,
581                                           uint8_t *payload, uint64_t orig_offset,
582                                           QEMUIOVector *qiov, Error **errp)
583  {
584      uint64_t offset;
585      uint32_t hole_size;
586  
587      if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
588          error_setg(errp, "Protocol error: invalid payload for "
589                           "NBD_REPLY_TYPE_OFFSET_HOLE");
590          return -EINVAL;
591      }
592  
593      offset = payload_advance64(&payload);
594      hole_size = payload_advance32(&payload);
595  
596      if (!hole_size || offset < orig_offset || hole_size > qiov->size ||
597          offset > orig_offset + qiov->size - hole_size) {
598          error_setg(errp, "Protocol error: server sent chunk exceeding requested"
599                           " region");
600          return -EINVAL;
601      }
602      if (s->info.min_block &&
603          !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) {
604          trace_nbd_structured_read_compliance("hole");
605      }
606  
607      qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
608  
609      return 0;
610  }
611  
612  /*
613   * nbd_parse_blockstatus_payload
614   * Based on our request, we expect only one extent in reply, for the
615   * base:allocation context.
616   */
617  static int nbd_parse_blockstatus_payload(BDRVNBDState *s,
618                                           NBDStructuredReplyChunk *chunk,
619                                           uint8_t *payload, bool wide,
620                                           uint64_t orig_length,
621                                           NBDExtent64 *extent, Error **errp)
622  {
623      uint32_t context_id;
624      uint32_t count;
625      size_t ext_len = wide ? sizeof(*extent) : sizeof(NBDExtent32);
626      size_t pay_len = sizeof(context_id) + wide * sizeof(count) + ext_len;
627  
628      /* The server succeeded, so it must have sent [at least] one extent */
629      if (chunk->length < pay_len) {
630          error_setg(errp, "Protocol error: invalid payload for "
631                           "NBD_REPLY_TYPE_BLOCK_STATUS");
632          return -EINVAL;
633      }
634  
635      context_id = payload_advance32(&payload);
636      if (s->info.context_id != context_id) {
637          error_setg(errp, "Protocol error: unexpected context id %d for "
638                           "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context "
639                           "id is %d", context_id,
640                           s->info.context_id);
641          return -EINVAL;
642      }
643  
644      if (wide) {
645          count = payload_advance32(&payload);
646          extent->length = payload_advance64(&payload);
647          extent->flags = payload_advance64(&payload);
648      } else {
649          count = 0;
650          extent->length = payload_advance32(&payload);
651          extent->flags = payload_advance32(&payload);
652      }
653  
654      if (extent->length == 0) {
655          error_setg(errp, "Protocol error: server sent status chunk with "
656                     "zero length");
657          return -EINVAL;
658      }
659  
660      /*
661       * A server sending unaligned block status is in violation of the
662       * protocol, but as qemu-nbd 3.1 is such a server (at least for
663       * POSIX files that are not a multiple of 512 bytes, since qemu
664       * rounds files up to 512-byte multiples but lseek(SEEK_HOLE)
665       * still sees an implicit hole beyond the real EOF), it's nicer to
666       * work around the misbehaving server. If the request included
667       * more than the final unaligned block, truncate it back to an
668       * aligned result; if the request was only the final block, round
669       * up to the full block and change the status to fully-allocated
670       * (always a safe status, even if it loses information).
671       */
672      if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length,
673                                                s->info.min_block)) {
674          trace_nbd_parse_blockstatus_compliance("extent length is unaligned");
675          if (extent->length > s->info.min_block) {
676              extent->length = QEMU_ALIGN_DOWN(extent->length,
677                                               s->info.min_block);
678          } else {
679              extent->length = s->info.min_block;
680              extent->flags = 0;
681          }
682      }
683  
684      /*
685       * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have
686       * sent us any more than one extent, nor should it have included
687       * status beyond our request in that extent. Furthermore, a wide
688       * server should have replied with an accurate count (we left
689       * count at 0 for a narrow server).  However, it's easy enough to
690       * ignore the server's noncompliance without killing the
691       * connection; just ignore trailing extents, and clamp things to
692       * the length of our request.
693       */
694      if (count != wide || chunk->length > pay_len) {
695          trace_nbd_parse_blockstatus_compliance("unexpected extent count");
696      }
697      if (extent->length > orig_length) {
698          extent->length = orig_length;
699          trace_nbd_parse_blockstatus_compliance("extent length too large");
700      }
701  
702      /*
703       * HACK: if we are using x-dirty-bitmaps to access
704       * qemu:allocation-depth, treat all depths > 2 the same as 2,
705       * since nbd_client_co_block_status is only expecting the low two
706       * bits to be set.
707       */
708      if (s->alloc_depth && extent->flags > 2) {
709          extent->flags = 2;
710      }
711  
712      return 0;
713  }
714  
715  /*
716   * nbd_parse_error_payload
717   * on success @errp contains message describing nbd error reply
718   */
719  static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
720                                     uint8_t *payload, int *request_ret,
721                                     Error **errp)
722  {
723      uint32_t error;
724      uint16_t message_size;
725  
726      assert(chunk->type & (1 << 15));
727  
728      if (chunk->length < sizeof(error) + sizeof(message_size)) {
729          error_setg(errp,
730                     "Protocol error: invalid payload for structured error");
731          return -EINVAL;
732      }
733  
734      error = nbd_errno_to_system_errno(payload_advance32(&payload));
735      if (error == 0) {
736          error_setg(errp, "Protocol error: server sent structured error chunk "
737                           "with error = 0");
738          return -EINVAL;
739      }
740  
741      *request_ret = -error;
742      message_size = payload_advance16(&payload);
743  
744      if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
745          error_setg(errp, "Protocol error: server sent structured error chunk "
746                           "with incorrect message size");
747          return -EINVAL;
748      }
749  
750      /* TODO: Add a trace point to mention the server complaint */
751  
752      /* TODO handle ERROR_OFFSET */
753  
754      return 0;
755  }
756  
757  static int coroutine_fn
758  nbd_co_receive_offset_data_payload(BDRVNBDState *s, uint64_t orig_offset,
759                                     QEMUIOVector *qiov, Error **errp)
760  {
761      QEMUIOVector sub_qiov;
762      uint64_t offset;
763      size_t data_size;
764      int ret;
765      NBDStructuredReplyChunk *chunk = &s->reply.structured;
766  
767      assert(nbd_reply_is_structured(&s->reply));
768  
769      /* The NBD spec requires at least one byte of payload */
770      if (chunk->length <= sizeof(offset)) {
771          error_setg(errp, "Protocol error: invalid payload for "
772                           "NBD_REPLY_TYPE_OFFSET_DATA");
773          return -EINVAL;
774      }
775  
776      if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) {
777          return -EIO;
778      }
779  
780      data_size = chunk->length - sizeof(offset);
781      assert(data_size);
782      if (offset < orig_offset || data_size > qiov->size ||
783          offset > orig_offset + qiov->size - data_size) {
784          error_setg(errp, "Protocol error: server sent chunk exceeding requested"
785                           " region");
786          return -EINVAL;
787      }
788      if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) {
789          trace_nbd_structured_read_compliance("data");
790      }
791  
792      qemu_iovec_init(&sub_qiov, qiov->niov);
793      qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
794      ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
795      qemu_iovec_destroy(&sub_qiov);
796  
797      return ret < 0 ? -EIO : 0;
798  }
799  
800  #define NBD_MAX_MALLOC_PAYLOAD 1000
801  static coroutine_fn int nbd_co_receive_structured_payload(
802          BDRVNBDState *s, void **payload, Error **errp)
803  {
804      int ret;
805      uint32_t len;
806  
807      assert(nbd_reply_is_structured(&s->reply));
808  
809      len = s->reply.structured.length;
810  
811      if (len == 0) {
812          return 0;
813      }
814  
815      if (payload == NULL) {
816          error_setg(errp, "Unexpected structured payload");
817          return -EINVAL;
818      }
819  
820      if (len > NBD_MAX_MALLOC_PAYLOAD) {
821          error_setg(errp, "Payload too large");
822          return -EINVAL;
823      }
824  
825      *payload = g_new(char, len);
826      ret = nbd_read(s->ioc, *payload, len, "structured payload", errp);
827      if (ret < 0) {
828          g_free(*payload);
829          *payload = NULL;
830          return ret;
831      }
832  
833      return 0;
834  }
835  
836  /*
837   * nbd_co_do_receive_one_chunk
838   * for simple reply:
839   *   set request_ret to received reply error
840   *   if qiov is not NULL: read payload to @qiov
841   * for structured reply chunk:
842   *   if error chunk: read payload, set @request_ret, do not set @payload
843   *   else if offset_data chunk: read payload data to @qiov, do not set @payload
844   *   else: read payload to @payload
845   *
846   * If function fails, @errp contains corresponding error message, and the
847   * connection with the server is suspect.  If it returns 0, then the
848   * transaction succeeded (although @request_ret may be a negative errno
849   * corresponding to the server's error reply), and errp is unchanged.
850   */
851  static coroutine_fn int nbd_co_do_receive_one_chunk(
852          BDRVNBDState *s, uint64_t cookie, bool only_structured,
853          int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
854  {
855      int ret;
856      int i = COOKIE_TO_INDEX(cookie);
857      void *local_payload = NULL;
858      NBDStructuredReplyChunk *chunk;
859  
860      if (payload) {
861          *payload = NULL;
862      }
863      *request_ret = 0;
864  
865      ret = nbd_receive_replies(s, cookie, errp);
866      if (ret < 0) {
867          error_prepend(errp, "Connection closed: ");
868          return -EIO;
869      }
870      assert(s->ioc);
871  
872      assert(s->reply.cookie == cookie);
873  
874      if (nbd_reply_is_simple(&s->reply)) {
875          if (only_structured) {
876              error_setg(errp, "Protocol error: simple reply when structured "
877                               "reply chunk was expected");
878              return -EINVAL;
879          }
880  
881          *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
882          if (*request_ret < 0 || !qiov) {
883              return 0;
884          }
885  
886          return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
887                                       errp) < 0 ? -EIO : 0;
888      }
889  
890      /* handle structured reply chunk */
891      assert(s->info.mode >= NBD_MODE_STRUCTURED);
892      chunk = &s->reply.structured;
893  
894      if (chunk->type == NBD_REPLY_TYPE_NONE) {
895          if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
896              error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
897                         " NBD_REPLY_FLAG_DONE flag set");
898              return -EINVAL;
899          }
900          if (chunk->length) {
901              error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
902                         " nonzero length");
903              return -EINVAL;
904          }
905          return 0;
906      }
907  
908      if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
909          if (!qiov) {
910              error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
911              return -EINVAL;
912          }
913  
914          return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
915                                                    qiov, errp);
916      }
917  
918      if (nbd_reply_type_is_error(chunk->type)) {
919          payload = &local_payload;
920      }
921  
922      ret = nbd_co_receive_structured_payload(s, payload, errp);
923      if (ret < 0) {
924          return ret;
925      }
926  
927      if (nbd_reply_type_is_error(chunk->type)) {
928          ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
929          g_free(local_payload);
930          return ret;
931      }
932  
933      return 0;
934  }
935  
936  /*
937   * nbd_co_receive_one_chunk
938   * Read reply, wake up connection_co and set s->quit if needed.
939   * Return value is a fatal error code or normal nbd reply error code
940   */
941  static coroutine_fn int nbd_co_receive_one_chunk(
942          BDRVNBDState *s, uint64_t cookie, bool only_structured,
943          int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload,
944          Error **errp)
945  {
946      int ret = nbd_co_do_receive_one_chunk(s, cookie, only_structured,
947                                            request_ret, qiov, payload, errp);
948  
949      if (ret < 0) {
950          memset(reply, 0, sizeof(*reply));
951          nbd_channel_error(s, ret);
952      } else {
953          /* For assert at loop start in nbd_connection_entry */
954          *reply = s->reply;
955      }
956      s->reply.cookie = 0;
957  
958      nbd_recv_coroutines_wake(s);
959  
960      return ret;
961  }
962  
963  typedef struct NBDReplyChunkIter {
964      int ret;
965      int request_ret;
966      Error *err;
967      bool done, only_structured;
968  } NBDReplyChunkIter;
969  
970  static void nbd_iter_channel_error(NBDReplyChunkIter *iter,
971                                     int ret, Error **local_err)
972  {
973      assert(local_err && *local_err);
974      assert(ret < 0);
975  
976      if (!iter->ret) {
977          iter->ret = ret;
978          error_propagate(&iter->err, *local_err);
979      } else {
980          error_free(*local_err);
981      }
982  
983      *local_err = NULL;
984  }
985  
986  static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret)
987  {
988      assert(ret < 0);
989  
990      if (!iter->request_ret) {
991          iter->request_ret = ret;
992      }
993  }
994  
995  /*
996   * NBD_FOREACH_REPLY_CHUNK
997   * The pointer stored in @payload requires g_free() to free it.
998   */
999  #define NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, structured, \
1000                                  qiov, reply, payload) \
1001      for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
1002           nbd_reply_chunk_iter_receive(s, &iter, cookie, qiov, reply, payload);)
1003  
1004  /*
1005   * nbd_reply_chunk_iter_receive
1006   * The pointer stored in @payload requires g_free() to free it.
1007   */
1008  static bool coroutine_fn nbd_reply_chunk_iter_receive(BDRVNBDState *s,
1009                                                        NBDReplyChunkIter *iter,
1010                                                        uint64_t cookie,
1011                                                        QEMUIOVector *qiov,
1012                                                        NBDReply *reply,
1013                                                        void **payload)
1014  {
1015      int ret, request_ret;
1016      NBDReply local_reply;
1017      NBDStructuredReplyChunk *chunk;
1018      Error *local_err = NULL;
1019  
1020      if (iter->done) {
1021          /* Previous iteration was last. */
1022          goto break_loop;
1023      }
1024  
1025      if (reply == NULL) {
1026          reply = &local_reply;
1027      }
1028  
1029      ret = nbd_co_receive_one_chunk(s, cookie, iter->only_structured,
1030                                     &request_ret, qiov, reply, payload,
1031                                     &local_err);
1032      if (ret < 0) {
1033          nbd_iter_channel_error(iter, ret, &local_err);
1034      } else if (request_ret < 0) {
1035          nbd_iter_request_error(iter, request_ret);
1036      }
1037  
1038      /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
1039      if (nbd_reply_is_simple(reply) || iter->ret < 0) {
1040          goto break_loop;
1041      }
1042  
1043      chunk = &reply->structured;
1044      iter->only_structured = true;
1045  
1046      if (chunk->type == NBD_REPLY_TYPE_NONE) {
1047          /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
1048          assert(chunk->flags & NBD_REPLY_FLAG_DONE);
1049          goto break_loop;
1050      }
1051  
1052      if (chunk->flags & NBD_REPLY_FLAG_DONE) {
1053          /* This iteration is last. */
1054          iter->done = true;
1055      }
1056  
1057      /* Execute the loop body */
1058      return true;
1059  
1060  break_loop:
1061      qemu_mutex_lock(&s->requests_lock);
1062      s->requests[COOKIE_TO_INDEX(cookie)].coroutine = NULL;
1063      s->in_flight--;
1064      qemu_co_queue_next(&s->free_sema);
1065      qemu_mutex_unlock(&s->requests_lock);
1066  
1067      return false;
1068  }
1069  
1070  static int coroutine_fn
1071  nbd_co_receive_return_code(BDRVNBDState *s, uint64_t cookie,
1072                             int *request_ret, Error **errp)
1073  {
1074      NBDReplyChunkIter iter;
1075  
1076      NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, NULL, NULL) {
1077          /* nbd_reply_chunk_iter_receive does all the work */
1078      }
1079  
1080      error_propagate(errp, iter.err);
1081      *request_ret = iter.request_ret;
1082      return iter.ret;
1083  }
1084  
1085  static int coroutine_fn
1086  nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t cookie,
1087                               uint64_t offset, QEMUIOVector *qiov,
1088                               int *request_ret, Error **errp)
1089  {
1090      NBDReplyChunkIter iter;
1091      NBDReply reply;
1092      void *payload = NULL;
1093      Error *local_err = NULL;
1094  
1095      NBD_FOREACH_REPLY_CHUNK(s, iter, cookie,
1096                              s->info.mode >= NBD_MODE_STRUCTURED,
1097                              qiov, &reply, &payload)
1098      {
1099          int ret;
1100          NBDStructuredReplyChunk *chunk = &reply.structured;
1101  
1102          assert(nbd_reply_is_structured(&reply));
1103  
1104          switch (chunk->type) {
1105          case NBD_REPLY_TYPE_OFFSET_DATA:
1106              /*
1107               * special cased in nbd_co_receive_one_chunk, data is already
1108               * in qiov
1109               */
1110              break;
1111          case NBD_REPLY_TYPE_OFFSET_HOLE:
1112              ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload,
1113                                                  offset, qiov, &local_err);
1114              if (ret < 0) {
1115                  nbd_channel_error(s, ret);
1116                  nbd_iter_channel_error(&iter, ret, &local_err);
1117              }
1118              break;
1119          default:
1120              if (!nbd_reply_type_is_error(chunk->type)) {
1121                  /* not allowed reply type */
1122                  nbd_channel_error(s, -EINVAL);
1123                  error_setg(&local_err,
1124                             "Unexpected reply type: %d (%s) for CMD_READ",
1125                             chunk->type, nbd_reply_type_lookup(chunk->type));
1126                  nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1127              }
1128          }
1129  
1130          g_free(payload);
1131          payload = NULL;
1132      }
1133  
1134      error_propagate(errp, iter.err);
1135      *request_ret = iter.request_ret;
1136      return iter.ret;
1137  }
1138  
1139  static int coroutine_fn
1140  nbd_co_receive_blockstatus_reply(BDRVNBDState *s, uint64_t cookie,
1141                                   uint64_t length, NBDExtent64 *extent,
1142                                   int *request_ret, Error **errp)
1143  {
1144      NBDReplyChunkIter iter;
1145      NBDReply reply;
1146      void *payload = NULL;
1147      Error *local_err = NULL;
1148      bool received = false;
1149  
1150      assert(!extent->length);
1151      NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, &reply, &payload) {
1152          int ret;
1153          NBDStructuredReplyChunk *chunk = &reply.structured;
1154          bool wide;
1155  
1156          assert(nbd_reply_is_structured(&reply));
1157  
1158          switch (chunk->type) {
1159          case NBD_REPLY_TYPE_BLOCK_STATUS_EXT:
1160          case NBD_REPLY_TYPE_BLOCK_STATUS:
1161              wide = chunk->type == NBD_REPLY_TYPE_BLOCK_STATUS_EXT;
1162              if ((s->info.mode >= NBD_MODE_EXTENDED) != wide) {
1163                  trace_nbd_extended_headers_compliance("block_status");
1164              }
1165              if (received) {
1166                  nbd_channel_error(s, -EINVAL);
1167                  error_setg(&local_err, "Several BLOCK_STATUS chunks in reply");
1168                  nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1169              }
1170              received = true;
1171  
1172              ret = nbd_parse_blockstatus_payload(
1173                  s, &reply.structured, payload, wide,
1174                  length, extent, &local_err);
1175              if (ret < 0) {
1176                  nbd_channel_error(s, ret);
1177                  nbd_iter_channel_error(&iter, ret, &local_err);
1178              }
1179              break;
1180          default:
1181              if (!nbd_reply_type_is_error(chunk->type)) {
1182                  nbd_channel_error(s, -EINVAL);
1183                  error_setg(&local_err,
1184                             "Unexpected reply type: %d (%s) "
1185                             "for CMD_BLOCK_STATUS",
1186                             chunk->type, nbd_reply_type_lookup(chunk->type));
1187                  nbd_iter_channel_error(&iter, -EINVAL, &local_err);
1188              }
1189          }
1190  
1191          g_free(payload);
1192          payload = NULL;
1193      }
1194  
1195      if (!extent->length && !iter.request_ret) {
1196          error_setg(&local_err, "Server did not reply with any status extents");
1197          nbd_iter_channel_error(&iter, -EIO, &local_err);
1198      }
1199  
1200      error_propagate(errp, iter.err);
1201      *request_ret = iter.request_ret;
1202      return iter.ret;
1203  }
1204  
1205  static int coroutine_fn GRAPH_RDLOCK
1206  nbd_co_request(BlockDriverState *bs, NBDRequest *request,
1207                 QEMUIOVector *write_qiov)
1208  {
1209      int ret, request_ret;
1210      Error *local_err = NULL;
1211      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1212  
1213      assert(request->type != NBD_CMD_READ);
1214      if (write_qiov) {
1215          assert(request->type == NBD_CMD_WRITE);
1216          assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
1217      } else {
1218          assert(request->type != NBD_CMD_WRITE);
1219      }
1220  
1221      do {
1222          ret = nbd_co_send_request(bs, request, write_qiov);
1223          if (ret < 0) {
1224              continue;
1225          }
1226  
1227          ret = nbd_co_receive_return_code(s, request->cookie,
1228                                           &request_ret, &local_err);
1229          if (local_err) {
1230              trace_nbd_co_request_fail(request->from, request->len,
1231                                        request->cookie, request->flags,
1232                                        request->type,
1233                                        nbd_cmd_lookup(request->type),
1234                                        ret, error_get_pretty(local_err));
1235              error_free(local_err);
1236              local_err = NULL;
1237          }
1238      } while (ret < 0 && nbd_client_will_reconnect(s));
1239  
1240      return ret ? ret : request_ret;
1241  }
1242  
1243  static int coroutine_fn GRAPH_RDLOCK
1244  nbd_client_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1245                       QEMUIOVector *qiov, BdrvRequestFlags flags)
1246  {
1247      int ret, request_ret;
1248      Error *local_err = NULL;
1249      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1250      NBDRequest request = {
1251          .type = NBD_CMD_READ,
1252          .from = offset,
1253          .len = bytes,
1254      };
1255  
1256      assert(bytes <= NBD_MAX_BUFFER_SIZE);
1257  
1258      if (!bytes) {
1259          return 0;
1260      }
1261      /*
1262       * Work around the fact that the block layer doesn't do
1263       * byte-accurate sizing yet - if the read exceeds the server's
1264       * advertised size because the block layer rounded size up, then
1265       * truncate the request to the server and tail-pad with zero.
1266       */
1267      if (offset >= s->info.size) {
1268          assert(bytes < BDRV_SECTOR_SIZE);
1269          qemu_iovec_memset(qiov, 0, 0, bytes);
1270          return 0;
1271      }
1272      if (offset + bytes > s->info.size) {
1273          uint64_t slop = offset + bytes - s->info.size;
1274  
1275          assert(slop < BDRV_SECTOR_SIZE);
1276          qemu_iovec_memset(qiov, bytes - slop, 0, slop);
1277          request.len -= slop;
1278      }
1279  
1280      do {
1281          ret = nbd_co_send_request(bs, &request, NULL);
1282          if (ret < 0) {
1283              continue;
1284          }
1285  
1286          ret = nbd_co_receive_cmdread_reply(s, request.cookie, offset, qiov,
1287                                             &request_ret, &local_err);
1288          if (local_err) {
1289              trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1290                                        request.flags, request.type,
1291                                        nbd_cmd_lookup(request.type),
1292                                        ret, error_get_pretty(local_err));
1293              error_free(local_err);
1294              local_err = NULL;
1295          }
1296      } while (ret < 0 && nbd_client_will_reconnect(s));
1297  
1298      return ret ? ret : request_ret;
1299  }
1300  
1301  static int coroutine_fn GRAPH_RDLOCK
1302  nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
1303                        QEMUIOVector *qiov, BdrvRequestFlags flags)
1304  {
1305      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1306      NBDRequest request = {
1307          .type = NBD_CMD_WRITE,
1308          .from = offset,
1309          .len = bytes,
1310      };
1311  
1312      assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1313      if (flags & BDRV_REQ_FUA) {
1314          assert(s->info.flags & NBD_FLAG_SEND_FUA);
1315          request.flags |= NBD_CMD_FLAG_FUA;
1316      }
1317  
1318      assert(bytes <= NBD_MAX_BUFFER_SIZE);
1319  
1320      if (!bytes) {
1321          return 0;
1322      }
1323      return nbd_co_request(bs, &request, qiov);
1324  }
1325  
1326  static int coroutine_fn GRAPH_RDLOCK
1327  nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
1328                              BdrvRequestFlags flags)
1329  {
1330      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1331      NBDRequest request = {
1332          .type = NBD_CMD_WRITE_ZEROES,
1333          .from = offset,
1334          .len = bytes,
1335      };
1336  
1337      /* rely on max_pwrite_zeroes */
1338      assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED);
1339  
1340      assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1341      if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
1342          return -ENOTSUP;
1343      }
1344  
1345      if (flags & BDRV_REQ_FUA) {
1346          assert(s->info.flags & NBD_FLAG_SEND_FUA);
1347          request.flags |= NBD_CMD_FLAG_FUA;
1348      }
1349      if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1350          request.flags |= NBD_CMD_FLAG_NO_HOLE;
1351      }
1352      if (flags & BDRV_REQ_NO_FALLBACK) {
1353          assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO);
1354          request.flags |= NBD_CMD_FLAG_FAST_ZERO;
1355      }
1356  
1357      if (!bytes) {
1358          return 0;
1359      }
1360      return nbd_co_request(bs, &request, NULL);
1361  }
1362  
1363  static int coroutine_fn GRAPH_RDLOCK nbd_client_co_flush(BlockDriverState *bs)
1364  {
1365      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1366      NBDRequest request = { .type = NBD_CMD_FLUSH };
1367  
1368      if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) {
1369          return 0;
1370      }
1371  
1372      request.from = 0;
1373      request.len = 0;
1374  
1375      return nbd_co_request(bs, &request, NULL);
1376  }
1377  
1378  static int coroutine_fn GRAPH_RDLOCK
1379  nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
1380  {
1381      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1382      NBDRequest request = {
1383          .type = NBD_CMD_TRIM,
1384          .from = offset,
1385          .len = bytes,
1386      };
1387  
1388      /* rely on max_pdiscard */
1389      assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED);
1390  
1391      assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
1392      if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
1393          return 0;
1394      }
1395  
1396      return nbd_co_request(bs, &request, NULL);
1397  }
1398  
1399  static int coroutine_fn GRAPH_RDLOCK nbd_client_co_block_status(
1400          BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
1401          int64_t *pnum, int64_t *map, BlockDriverState **file)
1402  {
1403      int ret, request_ret;
1404      NBDExtent64 extent = { 0 };
1405      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1406      Error *local_err = NULL;
1407  
1408      NBDRequest request = {
1409          .type = NBD_CMD_BLOCK_STATUS,
1410          .from = offset,
1411          .len = MIN(bytes, s->info.size - offset),
1412          .flags = NBD_CMD_FLAG_REQ_ONE,
1413      };
1414  
1415      if (!s->info.base_allocation) {
1416          *pnum = bytes;
1417          *map = offset;
1418          *file = bs;
1419          return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1420      }
1421      if (s->info.mode < NBD_MODE_EXTENDED) {
1422          request.len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment),
1423                            request.len);
1424      }
1425  
1426      /*
1427       * Work around the fact that the block layer doesn't do
1428       * byte-accurate sizing yet - if the status request exceeds the
1429       * server's advertised size because the block layer rounded size
1430       * up, we truncated the request to the server (above), or are
1431       * called on just the hole.
1432       */
1433      if (offset >= s->info.size) {
1434          *pnum = bytes;
1435          assert(bytes < BDRV_SECTOR_SIZE);
1436          /* Intentionally don't report offset_valid for the hole */
1437          return BDRV_BLOCK_ZERO;
1438      }
1439  
1440      if (s->info.min_block) {
1441          assert(QEMU_IS_ALIGNED(request.len, s->info.min_block));
1442      }
1443      do {
1444          ret = nbd_co_send_request(bs, &request, NULL);
1445          if (ret < 0) {
1446              continue;
1447          }
1448  
1449          ret = nbd_co_receive_blockstatus_reply(s, request.cookie, bytes,
1450                                                 &extent, &request_ret,
1451                                                 &local_err);
1452          if (local_err) {
1453              trace_nbd_co_request_fail(request.from, request.len, request.cookie,
1454                                        request.flags, request.type,
1455                                        nbd_cmd_lookup(request.type),
1456                                        ret, error_get_pretty(local_err));
1457              error_free(local_err);
1458              local_err = NULL;
1459          }
1460      } while (ret < 0 && nbd_client_will_reconnect(s));
1461  
1462      if (ret < 0 || request_ret < 0) {
1463          return ret ? ret : request_ret;
1464      }
1465  
1466      assert(extent.length);
1467      *pnum = extent.length;
1468      *map = offset;
1469      *file = bs;
1470      return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) |
1471          (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) |
1472          BDRV_BLOCK_OFFSET_VALID;
1473  }
1474  
1475  static int nbd_client_reopen_prepare(BDRVReopenState *state,
1476                                       BlockReopenQueue *queue, Error **errp)
1477  {
1478      BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque;
1479  
1480      if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) {
1481          error_setg(errp, "Can't reopen read-only NBD mount as read/write");
1482          return -EACCES;
1483      }
1484      return 0;
1485  }
1486  
1487  static void nbd_yank(void *opaque)
1488  {
1489      BlockDriverState *bs = opaque;
1490      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1491  
1492      QEMU_LOCK_GUARD(&s->requests_lock);
1493      qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
1494      s->state = NBD_CLIENT_QUIT;
1495  }
1496  
1497  static void nbd_client_close(BlockDriverState *bs)
1498  {
1499      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1500      NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode };
1501  
1502      if (s->ioc) {
1503          nbd_send_request(s->ioc, &request);
1504      }
1505  
1506      nbd_teardown_connection(bs);
1507  }
1508  
1509  
1510  /*
1511   * Parse nbd_open options
1512   */
1513  
1514  static int nbd_parse_uri(const char *filename, QDict *options)
1515  {
1516      URI *uri;
1517      const char *p;
1518      QueryParams *qp = NULL;
1519      int ret = 0;
1520      bool is_unix;
1521  
1522      uri = uri_parse(filename);
1523      if (!uri) {
1524          return -EINVAL;
1525      }
1526  
1527      /* transport */
1528      if (!g_strcmp0(uri->scheme, "nbd")) {
1529          is_unix = false;
1530      } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) {
1531          is_unix = false;
1532      } else if (!g_strcmp0(uri->scheme, "nbd+unix")) {
1533          is_unix = true;
1534      } else {
1535          ret = -EINVAL;
1536          goto out;
1537      }
1538  
1539      p = uri->path ? uri->path : "";
1540      if (p[0] == '/') {
1541          p++;
1542      }
1543      if (p[0]) {
1544          qdict_put_str(options, "export", p);
1545      }
1546  
1547      qp = query_params_parse(uri->query);
1548      if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
1549          ret = -EINVAL;
1550          goto out;
1551      }
1552  
1553      if (is_unix) {
1554          /* nbd+unix:///export?socket=path */
1555          if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
1556              ret = -EINVAL;
1557              goto out;
1558          }
1559          qdict_put_str(options, "server.type", "unix");
1560          qdict_put_str(options, "server.path", qp->p[0].value);
1561      } else {
1562          QString *host;
1563          char *port_str;
1564  
1565          /* nbd[+tcp]://host[:port]/export */
1566          if (!uri->server) {
1567              ret = -EINVAL;
1568              goto out;
1569          }
1570  
1571          /* strip braces from literal IPv6 address */
1572          if (uri->server[0] == '[') {
1573              host = qstring_from_substr(uri->server, 1,
1574                                         strlen(uri->server) - 1);
1575          } else {
1576              host = qstring_from_str(uri->server);
1577          }
1578  
1579          qdict_put_str(options, "server.type", "inet");
1580          qdict_put(options, "server.host", host);
1581  
1582          port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT);
1583          qdict_put_str(options, "server.port", port_str);
1584          g_free(port_str);
1585      }
1586  
1587  out:
1588      if (qp) {
1589          query_params_free(qp);
1590      }
1591      uri_free(uri);
1592      return ret;
1593  }
1594  
1595  static bool nbd_has_filename_options_conflict(QDict *options, Error **errp)
1596  {
1597      const QDictEntry *e;
1598  
1599      for (e = qdict_first(options); e; e = qdict_next(options, e)) {
1600          if (!strcmp(e->key, "host") ||
1601              !strcmp(e->key, "port") ||
1602              !strcmp(e->key, "path") ||
1603              !strcmp(e->key, "export") ||
1604              strstart(e->key, "server.", NULL))
1605          {
1606              error_setg(errp, "Option '%s' cannot be used with a file name",
1607                         e->key);
1608              return true;
1609          }
1610      }
1611  
1612      return false;
1613  }
1614  
1615  static void nbd_parse_filename(const char *filename, QDict *options,
1616                                 Error **errp)
1617  {
1618      g_autofree char *file = NULL;
1619      char *export_name;
1620      const char *host_spec;
1621      const char *unixpath;
1622  
1623      if (nbd_has_filename_options_conflict(options, errp)) {
1624          return;
1625      }
1626  
1627      if (strstr(filename, "://")) {
1628          int ret = nbd_parse_uri(filename, options);
1629          if (ret < 0) {
1630              error_setg(errp, "No valid URL specified");
1631          }
1632          return;
1633      }
1634  
1635      file = g_strdup(filename);
1636  
1637      export_name = strstr(file, EN_OPTSTR);
1638      if (export_name) {
1639          if (export_name[strlen(EN_OPTSTR)] == 0) {
1640              return;
1641          }
1642          export_name[0] = 0; /* truncate 'file' */
1643          export_name += strlen(EN_OPTSTR);
1644  
1645          qdict_put_str(options, "export", export_name);
1646      }
1647  
1648      /* extract the host_spec - fail if it's not nbd:... */
1649      if (!strstart(file, "nbd:", &host_spec)) {
1650          error_setg(errp, "File name string for NBD must start with 'nbd:'");
1651          return;
1652      }
1653  
1654      if (!*host_spec) {
1655          return;
1656      }
1657  
1658      /* are we a UNIX or TCP socket? */
1659      if (strstart(host_spec, "unix:", &unixpath)) {
1660          qdict_put_str(options, "server.type", "unix");
1661          qdict_put_str(options, "server.path", unixpath);
1662      } else {
1663          InetSocketAddress *addr = g_new(InetSocketAddress, 1);
1664  
1665          if (inet_parse(addr, host_spec, errp)) {
1666              goto out_inet;
1667          }
1668  
1669          qdict_put_str(options, "server.type", "inet");
1670          qdict_put_str(options, "server.host", addr->host);
1671          qdict_put_str(options, "server.port", addr->port);
1672      out_inet:
1673          qapi_free_InetSocketAddress(addr);
1674      }
1675  }
1676  
1677  static bool nbd_process_legacy_socket_options(QDict *output_options,
1678                                                QemuOpts *legacy_opts,
1679                                                Error **errp)
1680  {
1681      const char *path = qemu_opt_get(legacy_opts, "path");
1682      const char *host = qemu_opt_get(legacy_opts, "host");
1683      const char *port = qemu_opt_get(legacy_opts, "port");
1684      const QDictEntry *e;
1685  
1686      if (!path && !host && !port) {
1687          return true;
1688      }
1689  
1690      for (e = qdict_first(output_options); e; e = qdict_next(output_options, e))
1691      {
1692          if (strstart(e->key, "server.", NULL)) {
1693              error_setg(errp, "Cannot use 'server' and path/host/port at the "
1694                         "same time");
1695              return false;
1696          }
1697      }
1698  
1699      if (path && host) {
1700          error_setg(errp, "path and host may not be used at the same time");
1701          return false;
1702      } else if (path) {
1703          if (port) {
1704              error_setg(errp, "port may not be used without host");
1705              return false;
1706          }
1707  
1708          qdict_put_str(output_options, "server.type", "unix");
1709          qdict_put_str(output_options, "server.path", path);
1710      } else if (host) {
1711          qdict_put_str(output_options, "server.type", "inet");
1712          qdict_put_str(output_options, "server.host", host);
1713          qdict_put_str(output_options, "server.port",
1714                        port ?: stringify(NBD_DEFAULT_PORT));
1715      }
1716  
1717      return true;
1718  }
1719  
1720  static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options,
1721                                   Error **errp)
1722  {
1723      SocketAddress *saddr = NULL;
1724      QDict *addr = NULL;
1725      Visitor *iv = NULL;
1726  
1727      qdict_extract_subqdict(options, &addr, "server.");
1728      if (!qdict_size(addr)) {
1729          error_setg(errp, "NBD server address missing");
1730          goto done;
1731      }
1732  
1733      iv = qobject_input_visitor_new_flat_confused(addr, errp);
1734      if (!iv) {
1735          goto done;
1736      }
1737  
1738      if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
1739          goto done;
1740      }
1741  
1742      if (socket_address_parse_named_fd(saddr, errp) < 0) {
1743          qapi_free_SocketAddress(saddr);
1744          saddr = NULL;
1745          goto done;
1746      }
1747  
1748  done:
1749      qobject_unref(addr);
1750      visit_free(iv);
1751      return saddr;
1752  }
1753  
1754  static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp)
1755  {
1756      Object *obj;
1757      QCryptoTLSCreds *creds;
1758  
1759      obj = object_resolve_path_component(
1760          object_get_objects_root(), id);
1761      if (!obj) {
1762          error_setg(errp, "No TLS credentials with id '%s'",
1763                     id);
1764          return NULL;
1765      }
1766      creds = (QCryptoTLSCreds *)
1767          object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS);
1768      if (!creds) {
1769          error_setg(errp, "Object with id '%s' is not TLS credentials",
1770                     id);
1771          return NULL;
1772      }
1773  
1774      if (!qcrypto_tls_creds_check_endpoint(creds,
1775                                            QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT,
1776                                            errp)) {
1777          return NULL;
1778      }
1779      object_ref(obj);
1780      return creds;
1781  }
1782  
1783  
1784  static QemuOptsList nbd_runtime_opts = {
1785      .name = "nbd",
1786      .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head),
1787      .desc = {
1788          {
1789              .name = "host",
1790              .type = QEMU_OPT_STRING,
1791              .help = "TCP host to connect to",
1792          },
1793          {
1794              .name = "port",
1795              .type = QEMU_OPT_STRING,
1796              .help = "TCP port to connect to",
1797          },
1798          {
1799              .name = "path",
1800              .type = QEMU_OPT_STRING,
1801              .help = "Unix socket path to connect to",
1802          },
1803          {
1804              .name = "export",
1805              .type = QEMU_OPT_STRING,
1806              .help = "Name of the NBD export to open",
1807          },
1808          {
1809              .name = "tls-creds",
1810              .type = QEMU_OPT_STRING,
1811              .help = "ID of the TLS credentials to use",
1812          },
1813          {
1814              .name = "tls-hostname",
1815              .type = QEMU_OPT_STRING,
1816              .help = "Override hostname for validating TLS x509 certificate",
1817          },
1818          {
1819              .name = "x-dirty-bitmap",
1820              .type = QEMU_OPT_STRING,
1821              .help = "experimental: expose named dirty bitmap in place of "
1822                      "block status",
1823          },
1824          {
1825              .name = "reconnect-delay",
1826              .type = QEMU_OPT_NUMBER,
1827              .help = "On an unexpected disconnect, the nbd client tries to "
1828                      "connect again until succeeding or encountering a serious "
1829                      "error.  During the first @reconnect-delay seconds, all "
1830                      "requests are paused and will be rerun on a successful "
1831                      "reconnect. After that time, any delayed requests and all "
1832                      "future requests before a successful reconnect will "
1833                      "immediately fail. Default 0",
1834          },
1835          {
1836              .name = "open-timeout",
1837              .type = QEMU_OPT_NUMBER,
1838              .help = "In seconds. If zero, the nbd driver tries the connection "
1839                      "only once, and fails to open if the connection fails. "
1840                      "If non-zero, the nbd driver will repeat connection "
1841                      "attempts until successful or until @open-timeout seconds "
1842                      "have elapsed. Default 0",
1843          },
1844          { /* end of list */ }
1845      },
1846  };
1847  
1848  static int nbd_process_options(BlockDriverState *bs, QDict *options,
1849                                 Error **errp)
1850  {
1851      BDRVNBDState *s = bs->opaque;
1852      QemuOpts *opts;
1853      int ret = -EINVAL;
1854  
1855      opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort);
1856      if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1857          goto error;
1858      }
1859  
1860      /* Translate @host, @port, and @path to a SocketAddress */
1861      if (!nbd_process_legacy_socket_options(options, opts, errp)) {
1862          goto error;
1863      }
1864  
1865      /* Pop the config into our state object. Exit if invalid. */
1866      s->saddr = nbd_config(s, options, errp);
1867      if (!s->saddr) {
1868          goto error;
1869      }
1870  
1871      s->export = g_strdup(qemu_opt_get(opts, "export"));
1872      if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) {
1873          error_setg(errp, "export name too long to send to server");
1874          goto error;
1875      }
1876  
1877      s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
1878      if (s->tlscredsid) {
1879          s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp);
1880          if (!s->tlscreds) {
1881              goto error;
1882          }
1883  
1884          s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname"));
1885          if (!s->tlshostname &&
1886              s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
1887              s->tlshostname = g_strdup(s->saddr->u.inet.host);
1888          }
1889      }
1890  
1891      s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap"));
1892      if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) {
1893          error_setg(errp, "x-dirty-bitmap query too long to send to server");
1894          goto error;
1895      }
1896  
1897      s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0);
1898      s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0);
1899  
1900      ret = 0;
1901  
1902   error:
1903      qemu_opts_del(opts);
1904      return ret;
1905  }
1906  
1907  static int nbd_open(BlockDriverState *bs, QDict *options, int flags,
1908                      Error **errp)
1909  {
1910      int ret;
1911      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1912  
1913      s->bs = bs;
1914      qemu_mutex_init(&s->requests_lock);
1915      qemu_co_queue_init(&s->free_sema);
1916      qemu_co_mutex_init(&s->send_mutex);
1917      qemu_co_mutex_init(&s->receive_mutex);
1918  
1919      if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) {
1920          return -EEXIST;
1921      }
1922  
1923      ret = nbd_process_options(bs, options, errp);
1924      if (ret < 0) {
1925          goto fail;
1926      }
1927  
1928      s->conn = nbd_client_connection_new(s->saddr, true, s->export,
1929                                          s->x_dirty_bitmap, s->tlscreds,
1930                                          s->tlshostname);
1931  
1932      if (s->open_timeout) {
1933          nbd_client_connection_enable_retry(s->conn);
1934          open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) +
1935                          s->open_timeout * NANOSECONDS_PER_SECOND);
1936      }
1937  
1938      s->state = NBD_CLIENT_CONNECTING_WAIT;
1939      ret = nbd_do_establish_connection(bs, true, errp);
1940      if (ret < 0) {
1941          goto fail;
1942      }
1943  
1944      /*
1945       * The connect attempt is done, so we no longer need this timer.
1946       * Delete it, because we do not want it to be around when this node
1947       * is drained or closed.
1948       */
1949      open_timer_del(s);
1950  
1951      nbd_client_connection_enable_retry(s->conn);
1952  
1953      return 0;
1954  
1955  fail:
1956      open_timer_del(s);
1957      nbd_clear_bdrvstate(bs);
1958      return ret;
1959  }
1960  
1961  static void nbd_refresh_limits(BlockDriverState *bs, Error **errp)
1962  {
1963      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
1964      uint32_t min = s->info.min_block;
1965      uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block);
1966  
1967      /*
1968       * If the server did not advertise an alignment:
1969       * - a size that is not sector-aligned implies that an alignment
1970       *   of 1 can be used to access those tail bytes
1971       * - advertisement of block status requires an alignment of 1, so
1972       *   that we don't violate block layer constraints that block
1973       *   status is always aligned (as we can't control whether the
1974       *   server will report sub-sector extents, such as a hole at EOF
1975       *   on an unaligned POSIX file)
1976       * - otherwise, assume the server is so old that we are safer avoiding
1977       *   sub-sector requests
1978       */
1979      if (!min) {
1980          min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) ||
1981                 s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE;
1982      }
1983  
1984      bs->bl.request_alignment = min;
1985      bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min);
1986      bs->bl.max_pwrite_zeroes = max;
1987      bs->bl.max_transfer = max;
1988  
1989      /*
1990       * Assume that if the server supports extended headers, it also
1991       * supports unlimited size zero and trim commands.
1992       */
1993      if (s->info.mode >= NBD_MODE_EXTENDED) {
1994          bs->bl.max_pdiscard = bs->bl.max_pwrite_zeroes = 0;
1995      }
1996  
1997      if (s->info.opt_block &&
1998          s->info.opt_block > bs->bl.opt_transfer) {
1999          bs->bl.opt_transfer = s->info.opt_block;
2000      }
2001  }
2002  
2003  static void nbd_close(BlockDriverState *bs)
2004  {
2005      nbd_client_close(bs);
2006      nbd_clear_bdrvstate(bs);
2007  }
2008  
2009  /*
2010   * NBD cannot truncate, but if the caller asks to truncate to the same size, or
2011   * to a smaller size with exact=false, there is no reason to fail the
2012   * operation.
2013   *
2014   * Preallocation mode is ignored since it does not seems useful to fail when
2015   * we never change anything.
2016   */
2017  static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset,
2018                                          bool exact, PreallocMode prealloc,
2019                                          BdrvRequestFlags flags, Error **errp)
2020  {
2021      BDRVNBDState *s = bs->opaque;
2022  
2023      if (offset != s->info.size && exact) {
2024          error_setg(errp, "Cannot resize NBD nodes");
2025          return -ENOTSUP;
2026      }
2027  
2028      if (offset > s->info.size) {
2029          error_setg(errp, "Cannot grow NBD nodes");
2030          return -EINVAL;
2031      }
2032  
2033      return 0;
2034  }
2035  
2036  static int64_t coroutine_fn nbd_co_getlength(BlockDriverState *bs)
2037  {
2038      BDRVNBDState *s = bs->opaque;
2039  
2040      return s->info.size;
2041  }
2042  
2043  static void nbd_refresh_filename(BlockDriverState *bs)
2044  {
2045      BDRVNBDState *s = bs->opaque;
2046      const char *host = NULL, *port = NULL, *path = NULL;
2047      size_t len = 0;
2048  
2049      if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) {
2050          const InetSocketAddress *inet = &s->saddr->u.inet;
2051          if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) {
2052              host = inet->host;
2053              port = inet->port;
2054          }
2055      } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) {
2056          path = s->saddr->u.q_unix.path;
2057      } /* else can't represent as pseudo-filename */
2058  
2059      if (path && s->export) {
2060          len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2061                         "nbd+unix:///%s?socket=%s", s->export, path);
2062      } else if (path && !s->export) {
2063          len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2064                         "nbd+unix://?socket=%s", path);
2065      } else if (host && s->export) {
2066          len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2067                         "nbd://%s:%s/%s", host, port, s->export);
2068      } else if (host && !s->export) {
2069          len = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
2070                         "nbd://%s:%s", host, port);
2071      }
2072      if (len >= sizeof(bs->exact_filename)) {
2073          /* Name is too long to represent exactly, so leave it empty. */
2074          bs->exact_filename[0] = '\0';
2075      }
2076  }
2077  
2078  static char *nbd_dirname(BlockDriverState *bs, Error **errp)
2079  {
2080      /* The generic bdrv_dirname() implementation is able to work out some
2081       * directory name for NBD nodes, but that would be wrong. So far there is no
2082       * specification for how "export paths" would work, so NBD does not have
2083       * directory names. */
2084      error_setg(errp, "Cannot generate a base directory for NBD nodes");
2085      return NULL;
2086  }
2087  
2088  static const char *const nbd_strong_runtime_opts[] = {
2089      "path",
2090      "host",
2091      "port",
2092      "export",
2093      "tls-creds",
2094      "tls-hostname",
2095      "server.",
2096  
2097      NULL
2098  };
2099  
2100  static void nbd_cancel_in_flight(BlockDriverState *bs)
2101  {
2102      BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
2103  
2104      reconnect_delay_timer_del(s);
2105  
2106      qemu_mutex_lock(&s->requests_lock);
2107      if (s->state == NBD_CLIENT_CONNECTING_WAIT) {
2108          s->state = NBD_CLIENT_CONNECTING_NOWAIT;
2109      }
2110      qemu_mutex_unlock(&s->requests_lock);
2111  
2112      nbd_co_establish_connection_cancel(s->conn);
2113  }
2114  
2115  static void nbd_attach_aio_context(BlockDriverState *bs,
2116                                     AioContext *new_context)
2117  {
2118      BDRVNBDState *s = bs->opaque;
2119  
2120      /* The open_timer is used only during nbd_open() */
2121      assert(!s->open_timer);
2122  
2123      /*
2124       * The reconnect_delay_timer is scheduled in I/O paths when the
2125       * connection is lost, to cancel the reconnection attempt after a
2126       * given time.  Once this attempt is done (successfully or not),
2127       * nbd_reconnect_attempt() ensures the timer is deleted before the
2128       * respective I/O request is resumed.
2129       * Since the AioContext can only be changed when a node is drained,
2130       * the reconnect_delay_timer cannot be active here.
2131       */
2132      assert(!s->reconnect_delay_timer);
2133  }
2134  
2135  static void nbd_detach_aio_context(BlockDriverState *bs)
2136  {
2137      BDRVNBDState *s = bs->opaque;
2138  
2139      assert(!s->open_timer);
2140      assert(!s->reconnect_delay_timer);
2141  }
2142  
2143  static BlockDriver bdrv_nbd = {
2144      .format_name                = "nbd",
2145      .protocol_name              = "nbd",
2146      .instance_size              = sizeof(BDRVNBDState),
2147      .bdrv_parse_filename        = nbd_parse_filename,
2148      .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2149      .create_opts                = &bdrv_create_opts_simple,
2150      .bdrv_file_open             = nbd_open,
2151      .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2152      .bdrv_co_preadv             = nbd_client_co_preadv,
2153      .bdrv_co_pwritev            = nbd_client_co_pwritev,
2154      .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2155      .bdrv_close                 = nbd_close,
2156      .bdrv_co_flush_to_os        = nbd_client_co_flush,
2157      .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2158      .bdrv_refresh_limits        = nbd_refresh_limits,
2159      .bdrv_co_truncate           = nbd_co_truncate,
2160      .bdrv_co_getlength          = nbd_co_getlength,
2161      .bdrv_refresh_filename      = nbd_refresh_filename,
2162      .bdrv_co_block_status       = nbd_client_co_block_status,
2163      .bdrv_dirname               = nbd_dirname,
2164      .strong_runtime_opts        = nbd_strong_runtime_opts,
2165      .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2166  
2167      .bdrv_attach_aio_context    = nbd_attach_aio_context,
2168      .bdrv_detach_aio_context    = nbd_detach_aio_context,
2169  };
2170  
2171  static BlockDriver bdrv_nbd_tcp = {
2172      .format_name                = "nbd",
2173      .protocol_name              = "nbd+tcp",
2174      .instance_size              = sizeof(BDRVNBDState),
2175      .bdrv_parse_filename        = nbd_parse_filename,
2176      .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2177      .create_opts                = &bdrv_create_opts_simple,
2178      .bdrv_file_open             = nbd_open,
2179      .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2180      .bdrv_co_preadv             = nbd_client_co_preadv,
2181      .bdrv_co_pwritev            = nbd_client_co_pwritev,
2182      .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2183      .bdrv_close                 = nbd_close,
2184      .bdrv_co_flush_to_os        = nbd_client_co_flush,
2185      .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2186      .bdrv_refresh_limits        = nbd_refresh_limits,
2187      .bdrv_co_truncate           = nbd_co_truncate,
2188      .bdrv_co_getlength          = nbd_co_getlength,
2189      .bdrv_refresh_filename      = nbd_refresh_filename,
2190      .bdrv_co_block_status       = nbd_client_co_block_status,
2191      .bdrv_dirname               = nbd_dirname,
2192      .strong_runtime_opts        = nbd_strong_runtime_opts,
2193      .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2194  
2195      .bdrv_attach_aio_context    = nbd_attach_aio_context,
2196      .bdrv_detach_aio_context    = nbd_detach_aio_context,
2197  };
2198  
2199  static BlockDriver bdrv_nbd_unix = {
2200      .format_name                = "nbd",
2201      .protocol_name              = "nbd+unix",
2202      .instance_size              = sizeof(BDRVNBDState),
2203      .bdrv_parse_filename        = nbd_parse_filename,
2204      .bdrv_co_create_opts        = bdrv_co_create_opts_simple,
2205      .create_opts                = &bdrv_create_opts_simple,
2206      .bdrv_file_open             = nbd_open,
2207      .bdrv_reopen_prepare        = nbd_client_reopen_prepare,
2208      .bdrv_co_preadv             = nbd_client_co_preadv,
2209      .bdrv_co_pwritev            = nbd_client_co_pwritev,
2210      .bdrv_co_pwrite_zeroes      = nbd_client_co_pwrite_zeroes,
2211      .bdrv_close                 = nbd_close,
2212      .bdrv_co_flush_to_os        = nbd_client_co_flush,
2213      .bdrv_co_pdiscard           = nbd_client_co_pdiscard,
2214      .bdrv_refresh_limits        = nbd_refresh_limits,
2215      .bdrv_co_truncate           = nbd_co_truncate,
2216      .bdrv_co_getlength          = nbd_co_getlength,
2217      .bdrv_refresh_filename      = nbd_refresh_filename,
2218      .bdrv_co_block_status       = nbd_client_co_block_status,
2219      .bdrv_dirname               = nbd_dirname,
2220      .strong_runtime_opts        = nbd_strong_runtime_opts,
2221      .bdrv_cancel_in_flight      = nbd_cancel_in_flight,
2222  
2223      .bdrv_attach_aio_context    = nbd_attach_aio_context,
2224      .bdrv_detach_aio_context    = nbd_detach_aio_context,
2225  };
2226  
2227  static void bdrv_nbd_init(void)
2228  {
2229      bdrv_register(&bdrv_nbd);
2230      bdrv_register(&bdrv_nbd_tcp);
2231      bdrv_register(&bdrv_nbd_unix);
2232  }
2233  
2234  block_init(bdrv_nbd_init);
2235