xref: /openbmc/qemu/nbd/server.c (revision 5d721b78)
1 /*
2  *  Copyright (C) 2016 Red Hat, Inc.
3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
4  *
5  *  Network Block Device Server Side
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; under version 2 of the License.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "trace.h"
23 #include "nbd-internal.h"
24 
25 static int system_errno_to_nbd_errno(int err)
26 {
27     switch (err) {
28     case 0:
29         return NBD_SUCCESS;
30     case EPERM:
31     case EROFS:
32         return NBD_EPERM;
33     case EIO:
34         return NBD_EIO;
35     case ENOMEM:
36         return NBD_ENOMEM;
37 #ifdef EDQUOT
38     case EDQUOT:
39 #endif
40     case EFBIG:
41     case ENOSPC:
42         return NBD_ENOSPC;
43     case ESHUTDOWN:
44         return NBD_ESHUTDOWN;
45     case EINVAL:
46     default:
47         return NBD_EINVAL;
48     }
49 }
50 
51 /* Definitions for opaque data types */
52 
53 typedef struct NBDRequestData NBDRequestData;
54 
55 struct NBDRequestData {
56     QSIMPLEQ_ENTRY(NBDRequestData) entry;
57     NBDClient *client;
58     uint8_t *data;
59     bool complete;
60 };
61 
62 struct NBDExport {
63     int refcount;
64     void (*close)(NBDExport *exp);
65 
66     BlockBackend *blk;
67     char *name;
68     char *description;
69     off_t dev_offset;
70     off_t size;
71     uint16_t nbdflags;
72     QTAILQ_HEAD(, NBDClient) clients;
73     QTAILQ_ENTRY(NBDExport) next;
74 
75     AioContext *ctx;
76 
77     BlockBackend *eject_notifier_blk;
78     Notifier eject_notifier;
79 };
80 
81 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
82 
83 struct NBDClient {
84     int refcount;
85     void (*close_fn)(NBDClient *client, bool negotiated);
86 
87     bool no_zeroes;
88     NBDExport *exp;
89     QCryptoTLSCreds *tlscreds;
90     char *tlsaclname;
91     QIOChannelSocket *sioc; /* The underlying data channel */
92     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
93 
94     Coroutine *recv_coroutine;
95 
96     CoMutex send_lock;
97     Coroutine *send_coroutine;
98 
99     QTAILQ_ENTRY(NBDClient) next;
100     int nb_requests;
101     bool closing;
102 };
103 
104 /* That's all folks */
105 
106 static void nbd_client_receive_next_request(NBDClient *client);
107 
108 /* Basic flow for negotiation
109 
110    Server         Client
111    Negotiate
112 
113    or
114 
115    Server         Client
116    Negotiate #1
117                   Option
118    Negotiate #2
119 
120    ----
121 
122    followed by
123 
124    Server         Client
125                   Request
126    Response
127                   Request
128    Response
129                   ...
130    ...
131                   Request (type == 2)
132 
133 */
134 
135 /* Send a reply header, including length, but no payload.
136  * Return -errno on error, 0 on success. */
137 static int nbd_negotiate_send_rep_len(QIOChannel *ioc, uint32_t type,
138                                       uint32_t opt, uint32_t len, Error **errp)
139 {
140     uint64_t magic;
141 
142     trace_nbd_negotiate_send_rep_len(opt, type, len);
143 
144     magic = cpu_to_be64(NBD_REP_MAGIC);
145     if (nbd_write(ioc, &magic, sizeof(magic), errp) < 0) {
146         error_prepend(errp, "write failed (rep magic): ");
147         return -EINVAL;
148     }
149 
150     opt = cpu_to_be32(opt);
151     if (nbd_write(ioc, &opt, sizeof(opt), errp) < 0) {
152         error_prepend(errp, "write failed (rep opt): ");
153         return -EINVAL;
154     }
155 
156     type = cpu_to_be32(type);
157     if (nbd_write(ioc, &type, sizeof(type), errp) < 0) {
158         error_prepend(errp, "write failed (rep type): ");
159         return -EINVAL;
160     }
161 
162     len = cpu_to_be32(len);
163     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
164         error_prepend(errp, "write failed (rep data length): ");
165         return -EINVAL;
166     }
167     return 0;
168 }
169 
170 /* Send a reply header with default 0 length.
171  * Return -errno on error, 0 on success. */
172 static int nbd_negotiate_send_rep(QIOChannel *ioc, uint32_t type, uint32_t opt,
173                                   Error **errp)
174 {
175     return nbd_negotiate_send_rep_len(ioc, type, opt, 0, errp);
176 }
177 
178 /* Send an error reply.
179  * Return -errno on error, 0 on success. */
180 static int GCC_FMT_ATTR(5, 6)
181 nbd_negotiate_send_rep_err(QIOChannel *ioc, uint32_t type,
182                            uint32_t opt, Error **errp, const char *fmt, ...)
183 {
184     va_list va;
185     char *msg;
186     int ret;
187     size_t len;
188 
189     va_start(va, fmt);
190     msg = g_strdup_vprintf(fmt, va);
191     va_end(va);
192     len = strlen(msg);
193     assert(len < 4096);
194     trace_nbd_negotiate_send_rep_err(msg);
195     ret = nbd_negotiate_send_rep_len(ioc, type, opt, len, errp);
196     if (ret < 0) {
197         goto out;
198     }
199     if (nbd_write(ioc, msg, len, errp) < 0) {
200         error_prepend(errp, "write failed (error message): ");
201         ret = -EIO;
202     } else {
203         ret = 0;
204     }
205 
206 out:
207     g_free(msg);
208     return ret;
209 }
210 
211 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
212  * Return -errno on error, 0 on success. */
213 static int nbd_negotiate_send_rep_list(QIOChannel *ioc, NBDExport *exp,
214                                        Error **errp)
215 {
216     size_t name_len, desc_len;
217     uint32_t len;
218     const char *name = exp->name ? exp->name : "";
219     const char *desc = exp->description ? exp->description : "";
220     int ret;
221 
222     trace_nbd_negotiate_send_rep_list(name, desc);
223     name_len = strlen(name);
224     desc_len = strlen(desc);
225     len = name_len + desc_len + sizeof(len);
226     ret = nbd_negotiate_send_rep_len(ioc, NBD_REP_SERVER, NBD_OPT_LIST, len,
227                                      errp);
228     if (ret < 0) {
229         return ret;
230     }
231 
232     len = cpu_to_be32(name_len);
233     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
234         error_prepend(errp, "write failed (name length): ");
235         return -EINVAL;
236     }
237 
238     if (nbd_write(ioc, name, name_len, errp) < 0) {
239         error_prepend(errp, "write failed (name buffer): ");
240         return -EINVAL;
241     }
242 
243     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
244         error_prepend(errp, "write failed (description buffer): ");
245         return -EINVAL;
246     }
247 
248     return 0;
249 }
250 
251 /* Process the NBD_OPT_LIST command, with a potential series of replies.
252  * Return -errno on error, 0 on success. */
253 static int nbd_negotiate_handle_list(NBDClient *client, uint32_t length,
254                                      Error **errp)
255 {
256     NBDExport *exp;
257 
258     if (length) {
259         if (nbd_drop(client->ioc, length, errp) < 0) {
260             return -EIO;
261         }
262         return nbd_negotiate_send_rep_err(client->ioc,
263                                           NBD_REP_ERR_INVALID, NBD_OPT_LIST,
264                                           errp,
265                                           "OPT_LIST should not have length");
266     }
267 
268     /* For each export, send a NBD_REP_SERVER reply. */
269     QTAILQ_FOREACH(exp, &exports, next) {
270         if (nbd_negotiate_send_rep_list(client->ioc, exp, errp)) {
271             return -EINVAL;
272         }
273     }
274     /* Finish with a NBD_REP_ACK. */
275     return nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, NBD_OPT_LIST, errp);
276 }
277 
278 static int nbd_negotiate_handle_export_name(NBDClient *client, uint32_t length,
279                                             Error **errp)
280 {
281     char name[NBD_MAX_NAME_SIZE + 1];
282 
283     /* Client sends:
284         [20 ..  xx]   export name (length bytes)
285      */
286     trace_nbd_negotiate_handle_export_name();
287     if (length >= sizeof(name)) {
288         error_setg(errp, "Bad length received");
289         return -EINVAL;
290     }
291     if (nbd_read(client->ioc, name, length, errp) < 0) {
292         error_prepend(errp, "read failed: ");
293         return -EINVAL;
294     }
295     name[length] = '\0';
296 
297     trace_nbd_negotiate_handle_export_name_request(name);
298 
299     client->exp = nbd_export_find(name);
300     if (!client->exp) {
301         error_setg(errp, "export not found");
302         return -EINVAL;
303     }
304 
305     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
306     nbd_export_get(client->exp);
307 
308     return 0;
309 }
310 
311 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
312  * new channel for all further (now-encrypted) communication. */
313 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
314                                                  uint32_t length,
315                                                  Error **errp)
316 {
317     QIOChannel *ioc;
318     QIOChannelTLS *tioc;
319     struct NBDTLSHandshakeData data = { 0 };
320 
321     trace_nbd_negotiate_handle_starttls();
322     ioc = client->ioc;
323     if (length) {
324         if (nbd_drop(ioc, length, errp) < 0) {
325             return NULL;
326         }
327         nbd_negotiate_send_rep_err(ioc, NBD_REP_ERR_INVALID, NBD_OPT_STARTTLS,
328                                    errp,
329                                    "OPT_STARTTLS should not have length");
330         return NULL;
331     }
332 
333     if (nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK,
334                                NBD_OPT_STARTTLS, errp) < 0) {
335         return NULL;
336     }
337 
338     tioc = qio_channel_tls_new_server(ioc,
339                                       client->tlscreds,
340                                       client->tlsaclname,
341                                       errp);
342     if (!tioc) {
343         return NULL;
344     }
345 
346     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
347     trace_nbd_negotiate_handle_starttls_handshake();
348     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
349     qio_channel_tls_handshake(tioc,
350                               nbd_tls_handshake,
351                               &data,
352                               NULL);
353 
354     if (!data.complete) {
355         g_main_loop_run(data.loop);
356     }
357     g_main_loop_unref(data.loop);
358     if (data.error) {
359         object_unref(OBJECT(tioc));
360         error_propagate(errp, data.error);
361         return NULL;
362     }
363 
364     return QIO_CHANNEL(tioc);
365 }
366 
367 /* nbd_negotiate_options
368  * Process all NBD_OPT_* client option commands.
369  * Return:
370  * -errno  on error, errp is set
371  * 0       on successful negotiation, errp is not set
372  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
373  *         errp is not set
374  */
375 static int nbd_negotiate_options(NBDClient *client, Error **errp)
376 {
377     uint32_t flags;
378     bool fixedNewstyle = false;
379     Error *local_err = NULL;
380 
381     /* Client sends:
382         [ 0 ..   3]   client flags
383 
384         [ 0 ..   7]   NBD_OPTS_MAGIC
385         [ 8 ..  11]   NBD option
386         [12 ..  15]   Data length
387         ...           Rest of request
388 
389         [ 0 ..   7]   NBD_OPTS_MAGIC
390         [ 8 ..  11]   Second NBD option
391         [12 ..  15]   Data length
392         ...           Rest of request
393     */
394 
395     if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
396         error_prepend(errp, "read failed: ");
397         return -EIO;
398     }
399     trace_nbd_negotiate_options_flags();
400     be32_to_cpus(&flags);
401     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
402         trace_nbd_negotiate_options_newstyle();
403         fixedNewstyle = true;
404         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
405     }
406     if (flags & NBD_FLAG_C_NO_ZEROES) {
407         trace_nbd_negotiate_options_no_zeroes();
408         client->no_zeroes = true;
409         flags &= ~NBD_FLAG_C_NO_ZEROES;
410     }
411     if (flags != 0) {
412         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
413         return -EIO;
414     }
415 
416     while (1) {
417         int ret;
418         uint32_t option, length;
419         uint64_t magic;
420 
421         if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
422             error_prepend(errp, "read failed: ");
423             return -EINVAL;
424         }
425         magic = be64_to_cpu(magic);
426         trace_nbd_negotiate_options_check_magic(magic);
427         if (magic != NBD_OPTS_MAGIC) {
428             error_setg(errp, "Bad magic received");
429             return -EINVAL;
430         }
431 
432         if (nbd_read(client->ioc, &option,
433                      sizeof(option), errp) < 0) {
434             error_prepend(errp, "read failed: ");
435             return -EINVAL;
436         }
437         option = be32_to_cpu(option);
438 
439         if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
440             error_prepend(errp, "read failed: ");
441             return -EINVAL;
442         }
443         length = be32_to_cpu(length);
444 
445         trace_nbd_negotiate_options_check_option(option);
446         if (client->tlscreds &&
447             client->ioc == (QIOChannel *)client->sioc) {
448             QIOChannel *tioc;
449             if (!fixedNewstyle) {
450                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
451                 return -EINVAL;
452             }
453             switch (option) {
454             case NBD_OPT_STARTTLS:
455                 tioc = nbd_negotiate_handle_starttls(client, length, errp);
456                 if (!tioc) {
457                     return -EIO;
458                 }
459                 object_unref(OBJECT(client->ioc));
460                 client->ioc = QIO_CHANNEL(tioc);
461                 break;
462 
463             case NBD_OPT_EXPORT_NAME:
464                 /* No way to return an error to client, so drop connection */
465                 error_setg(errp, "Option 0x%x not permitted before TLS",
466                            option);
467                 return -EINVAL;
468 
469             default:
470                 if (nbd_drop(client->ioc, length, errp) < 0) {
471                     return -EIO;
472                 }
473                 ret = nbd_negotiate_send_rep_err(client->ioc,
474                                                  NBD_REP_ERR_TLS_REQD,
475                                                  option, errp,
476                                                  "Option 0x%" PRIx32
477                                                  "not permitted before TLS",
478                                                  option);
479                 if (ret < 0) {
480                     return ret;
481                 }
482                 /* Let the client keep trying, unless they asked to quit */
483                 if (option == NBD_OPT_ABORT) {
484                     return 1;
485                 }
486                 break;
487             }
488         } else if (fixedNewstyle) {
489             switch (option) {
490             case NBD_OPT_LIST:
491                 ret = nbd_negotiate_handle_list(client, length, errp);
492                 if (ret < 0) {
493                     return ret;
494                 }
495                 break;
496 
497             case NBD_OPT_ABORT:
498                 /* NBD spec says we must try to reply before
499                  * disconnecting, but that we must also tolerate
500                  * guests that don't wait for our reply. */
501                 nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, option,
502                                        &local_err);
503 
504                 if (local_err != NULL) {
505                     const char *error = error_get_pretty(local_err);
506                     trace_nbd_opt_abort_reply_failed(error);
507                     error_free(local_err);
508                 }
509 
510                 return 1;
511 
512             case NBD_OPT_EXPORT_NAME:
513                 return nbd_negotiate_handle_export_name(client, length, errp);
514 
515             case NBD_OPT_STARTTLS:
516                 if (nbd_drop(client->ioc, length, errp) < 0) {
517                     return -EIO;
518                 }
519                 if (client->tlscreds) {
520                     ret = nbd_negotiate_send_rep_err(client->ioc,
521                                                      NBD_REP_ERR_INVALID,
522                                                      option, errp,
523                                                      "TLS already enabled");
524                 } else {
525                     ret = nbd_negotiate_send_rep_err(client->ioc,
526                                                      NBD_REP_ERR_POLICY,
527                                                      option, errp,
528                                                      "TLS not configured");
529                 }
530                 if (ret < 0) {
531                     return ret;
532                 }
533                 break;
534             default:
535                 if (nbd_drop(client->ioc, length, errp) < 0) {
536                     return -EIO;
537                 }
538                 ret = nbd_negotiate_send_rep_err(client->ioc,
539                                                  NBD_REP_ERR_UNSUP,
540                                                  option, errp,
541                                                  "Unsupported option 0x%"
542                                                  PRIx32,
543                                                  option);
544                 if (ret < 0) {
545                     return ret;
546                 }
547                 break;
548             }
549         } else {
550             /*
551              * If broken new-style we should drop the connection
552              * for anything except NBD_OPT_EXPORT_NAME
553              */
554             switch (option) {
555             case NBD_OPT_EXPORT_NAME:
556                 return nbd_negotiate_handle_export_name(client, length, errp);
557 
558             default:
559                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
560                 return -EINVAL;
561             }
562         }
563     }
564 }
565 
566 /* nbd_negotiate
567  * Return:
568  * -errno  on error, errp is set
569  * 0       on successful negotiation, errp is not set
570  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
571  *         errp is not set
572  */
573 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
574 {
575     char buf[8 + 8 + 8 + 128];
576     int ret;
577     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
578                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
579                               NBD_FLAG_SEND_WRITE_ZEROES);
580     bool oldStyle;
581     size_t len;
582 
583     /* Old style negotiation header without options
584         [ 0 ..   7]   passwd       ("NBDMAGIC")
585         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
586         [16 ..  23]   size
587         [24 ..  25]   server flags (0)
588         [26 ..  27]   export flags
589         [28 .. 151]   reserved     (0)
590 
591        New style negotiation header with options
592         [ 0 ..   7]   passwd       ("NBDMAGIC")
593         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
594         [16 ..  17]   server flags (0)
595         ....options sent....
596         [18 ..  25]   size
597         [26 ..  27]   export flags
598         [28 .. 151]   reserved     (0, omit if no_zeroes)
599      */
600 
601     qio_channel_set_blocking(client->ioc, false, NULL);
602 
603     trace_nbd_negotiate_begin();
604     memset(buf, 0, sizeof(buf));
605     memcpy(buf, "NBDMAGIC", 8);
606 
607     oldStyle = client->exp != NULL && !client->tlscreds;
608     if (oldStyle) {
609         trace_nbd_negotiate_old_style(client->exp->size,
610                                       client->exp->nbdflags | myflags);
611         stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
612         stq_be_p(buf + 16, client->exp->size);
613         stw_be_p(buf + 26, client->exp->nbdflags | myflags);
614 
615         if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
616             error_prepend(errp, "write failed: ");
617             return -EINVAL;
618         }
619     } else {
620         stq_be_p(buf + 8, NBD_OPTS_MAGIC);
621         stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
622 
623         if (nbd_write(client->ioc, buf, 18, errp) < 0) {
624             error_prepend(errp, "write failed: ");
625             return -EINVAL;
626         }
627         ret = nbd_negotiate_options(client, errp);
628         if (ret != 0) {
629             if (ret < 0) {
630                 error_prepend(errp, "option negotiation failed: ");
631             }
632             return ret;
633         }
634 
635         trace_nbd_negotiate_new_style_size_flags(
636             client->exp->size, client->exp->nbdflags | myflags);
637         stq_be_p(buf + 18, client->exp->size);
638         stw_be_p(buf + 26, client->exp->nbdflags | myflags);
639         len = client->no_zeroes ? 10 : sizeof(buf) - 18;
640         ret = nbd_write(client->ioc, buf + 18, len, errp);
641         if (ret < 0) {
642             error_prepend(errp, "write failed: ");
643             return ret;
644         }
645     }
646 
647     trace_nbd_negotiate_success();
648 
649     return 0;
650 }
651 
652 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
653                                Error **errp)
654 {
655     uint8_t buf[NBD_REQUEST_SIZE];
656     uint32_t magic;
657     int ret;
658 
659     ret = nbd_read(ioc, buf, sizeof(buf), errp);
660     if (ret < 0) {
661         return ret;
662     }
663 
664     /* Request
665        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
666        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
667        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
668        [ 8 .. 15]   handle
669        [16 .. 23]   from
670        [24 .. 27]   len
671      */
672 
673     magic = ldl_be_p(buf);
674     request->flags  = lduw_be_p(buf + 4);
675     request->type   = lduw_be_p(buf + 6);
676     request->handle = ldq_be_p(buf + 8);
677     request->from   = ldq_be_p(buf + 16);
678     request->len    = ldl_be_p(buf + 24);
679 
680     trace_nbd_receive_request(magic, request->flags, request->type,
681                               request->from, request->len);
682 
683     if (magic != NBD_REQUEST_MAGIC) {
684         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
685         return -EINVAL;
686     }
687     return 0;
688 }
689 
690 static int nbd_send_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
691 {
692     uint8_t buf[NBD_REPLY_SIZE];
693 
694     reply->error = system_errno_to_nbd_errno(reply->error);
695 
696     trace_nbd_send_reply(reply->error, reply->handle);
697 
698     /* Reply
699        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
700        [ 4 ..  7]    error   (0 == no error)
701        [ 7 .. 15]    handle
702      */
703     stl_be_p(buf, NBD_REPLY_MAGIC);
704     stl_be_p(buf + 4, reply->error);
705     stq_be_p(buf + 8, reply->handle);
706 
707     return nbd_write(ioc, buf, sizeof(buf), errp);
708 }
709 
710 #define MAX_NBD_REQUESTS 16
711 
712 void nbd_client_get(NBDClient *client)
713 {
714     client->refcount++;
715 }
716 
717 void nbd_client_put(NBDClient *client)
718 {
719     if (--client->refcount == 0) {
720         /* The last reference should be dropped by client->close,
721          * which is called by client_close.
722          */
723         assert(client->closing);
724 
725         qio_channel_detach_aio_context(client->ioc);
726         object_unref(OBJECT(client->sioc));
727         object_unref(OBJECT(client->ioc));
728         if (client->tlscreds) {
729             object_unref(OBJECT(client->tlscreds));
730         }
731         g_free(client->tlsaclname);
732         if (client->exp) {
733             QTAILQ_REMOVE(&client->exp->clients, client, next);
734             nbd_export_put(client->exp);
735         }
736         g_free(client);
737     }
738 }
739 
740 static void client_close(NBDClient *client, bool negotiated)
741 {
742     if (client->closing) {
743         return;
744     }
745 
746     client->closing = true;
747 
748     /* Force requests to finish.  They will drop their own references,
749      * then we'll close the socket and free the NBDClient.
750      */
751     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
752                          NULL);
753 
754     /* Also tell the client, so that they release their reference.  */
755     if (client->close_fn) {
756         client->close_fn(client, negotiated);
757     }
758 }
759 
760 static NBDRequestData *nbd_request_get(NBDClient *client)
761 {
762     NBDRequestData *req;
763 
764     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
765     client->nb_requests++;
766 
767     req = g_new0(NBDRequestData, 1);
768     nbd_client_get(client);
769     req->client = client;
770     return req;
771 }
772 
773 static void nbd_request_put(NBDRequestData *req)
774 {
775     NBDClient *client = req->client;
776 
777     if (req->data) {
778         qemu_vfree(req->data);
779     }
780     g_free(req);
781 
782     client->nb_requests--;
783     nbd_client_receive_next_request(client);
784 
785     nbd_client_put(client);
786 }
787 
788 static void blk_aio_attached(AioContext *ctx, void *opaque)
789 {
790     NBDExport *exp = opaque;
791     NBDClient *client;
792 
793     trace_nbd_blk_aio_attached(exp->name, ctx);
794 
795     exp->ctx = ctx;
796 
797     QTAILQ_FOREACH(client, &exp->clients, next) {
798         qio_channel_attach_aio_context(client->ioc, ctx);
799         if (client->recv_coroutine) {
800             aio_co_schedule(ctx, client->recv_coroutine);
801         }
802         if (client->send_coroutine) {
803             aio_co_schedule(ctx, client->send_coroutine);
804         }
805     }
806 }
807 
808 static void blk_aio_detach(void *opaque)
809 {
810     NBDExport *exp = opaque;
811     NBDClient *client;
812 
813     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
814 
815     QTAILQ_FOREACH(client, &exp->clients, next) {
816         qio_channel_detach_aio_context(client->ioc);
817     }
818 
819     exp->ctx = NULL;
820 }
821 
822 static void nbd_eject_notifier(Notifier *n, void *data)
823 {
824     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
825     nbd_export_close(exp);
826 }
827 
828 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
829                           uint16_t nbdflags, void (*close)(NBDExport *),
830                           bool writethrough, BlockBackend *on_eject_blk,
831                           Error **errp)
832 {
833     BlockBackend *blk;
834     NBDExport *exp = g_malloc0(sizeof(NBDExport));
835     uint64_t perm;
836     int ret;
837 
838     /* Don't allow resize while the NBD server is running, otherwise we don't
839      * care what happens with the node. */
840     perm = BLK_PERM_CONSISTENT_READ;
841     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
842         perm |= BLK_PERM_WRITE;
843     }
844     blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
845                         BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
846     ret = blk_insert_bs(blk, bs, errp);
847     if (ret < 0) {
848         goto fail;
849     }
850     blk_set_enable_write_cache(blk, !writethrough);
851 
852     exp->refcount = 1;
853     QTAILQ_INIT(&exp->clients);
854     exp->blk = blk;
855     exp->dev_offset = dev_offset;
856     exp->nbdflags = nbdflags;
857     exp->size = size < 0 ? blk_getlength(blk) : size;
858     if (exp->size < 0) {
859         error_setg_errno(errp, -exp->size,
860                          "Failed to determine the NBD export's length");
861         goto fail;
862     }
863     exp->size -= exp->size % BDRV_SECTOR_SIZE;
864 
865     exp->close = close;
866     exp->ctx = blk_get_aio_context(blk);
867     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
868 
869     if (on_eject_blk) {
870         blk_ref(on_eject_blk);
871         exp->eject_notifier_blk = on_eject_blk;
872         exp->eject_notifier.notify = nbd_eject_notifier;
873         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
874     }
875 
876     /*
877      * NBD exports are used for non-shared storage migration.  Make sure
878      * that BDRV_O_INACTIVE is cleared and the image is ready for write
879      * access since the export could be available before migration handover.
880      */
881     aio_context_acquire(exp->ctx);
882     blk_invalidate_cache(blk, NULL);
883     aio_context_release(exp->ctx);
884     return exp;
885 
886 fail:
887     blk_unref(blk);
888     g_free(exp);
889     return NULL;
890 }
891 
892 NBDExport *nbd_export_find(const char *name)
893 {
894     NBDExport *exp;
895     QTAILQ_FOREACH(exp, &exports, next) {
896         if (strcmp(name, exp->name) == 0) {
897             return exp;
898         }
899     }
900 
901     return NULL;
902 }
903 
904 void nbd_export_set_name(NBDExport *exp, const char *name)
905 {
906     if (exp->name == name) {
907         return;
908     }
909 
910     nbd_export_get(exp);
911     if (exp->name != NULL) {
912         g_free(exp->name);
913         exp->name = NULL;
914         QTAILQ_REMOVE(&exports, exp, next);
915         nbd_export_put(exp);
916     }
917     if (name != NULL) {
918         nbd_export_get(exp);
919         exp->name = g_strdup(name);
920         QTAILQ_INSERT_TAIL(&exports, exp, next);
921     }
922     nbd_export_put(exp);
923 }
924 
925 void nbd_export_set_description(NBDExport *exp, const char *description)
926 {
927     g_free(exp->description);
928     exp->description = g_strdup(description);
929 }
930 
931 void nbd_export_close(NBDExport *exp)
932 {
933     NBDClient *client, *next;
934 
935     nbd_export_get(exp);
936     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
937         client_close(client, true);
938     }
939     nbd_export_set_name(exp, NULL);
940     nbd_export_set_description(exp, NULL);
941     nbd_export_put(exp);
942 }
943 
944 void nbd_export_get(NBDExport *exp)
945 {
946     assert(exp->refcount > 0);
947     exp->refcount++;
948 }
949 
950 void nbd_export_put(NBDExport *exp)
951 {
952     assert(exp->refcount > 0);
953     if (exp->refcount == 1) {
954         nbd_export_close(exp);
955     }
956 
957     if (--exp->refcount == 0) {
958         assert(exp->name == NULL);
959         assert(exp->description == NULL);
960 
961         if (exp->close) {
962             exp->close(exp);
963         }
964 
965         if (exp->blk) {
966             if (exp->eject_notifier_blk) {
967                 notifier_remove(&exp->eject_notifier);
968                 blk_unref(exp->eject_notifier_blk);
969             }
970             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
971                                             blk_aio_detach, exp);
972             blk_unref(exp->blk);
973             exp->blk = NULL;
974         }
975 
976         g_free(exp);
977     }
978 }
979 
980 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
981 {
982     return exp->blk;
983 }
984 
985 void nbd_export_close_all(void)
986 {
987     NBDExport *exp, *next;
988 
989     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
990         nbd_export_close(exp);
991     }
992 }
993 
994 static int nbd_co_send_reply(NBDRequestData *req, NBDReply *reply, int len,
995                              Error **errp)
996 {
997     NBDClient *client = req->client;
998     int ret;
999 
1000     g_assert(qemu_in_coroutine());
1001 
1002     trace_nbd_co_send_reply(reply->handle, reply->error, len);
1003 
1004     qemu_co_mutex_lock(&client->send_lock);
1005     client->send_coroutine = qemu_coroutine_self();
1006 
1007     if (!len) {
1008         ret = nbd_send_reply(client->ioc, reply, errp);
1009     } else {
1010         qio_channel_set_cork(client->ioc, true);
1011         ret = nbd_send_reply(client->ioc, reply, errp);
1012         if (ret == 0) {
1013             ret = nbd_write(client->ioc, req->data, len, errp);
1014             if (ret < 0) {
1015                 ret = -EIO;
1016             }
1017         }
1018         qio_channel_set_cork(client->ioc, false);
1019     }
1020 
1021     client->send_coroutine = NULL;
1022     qemu_co_mutex_unlock(&client->send_lock);
1023     return ret;
1024 }
1025 
1026 /* nbd_co_receive_request
1027  * Collect a client request. Return 0 if request looks valid, -EIO to drop
1028  * connection right away, and any other negative value to report an error to
1029  * the client (although the caller may still need to disconnect after reporting
1030  * the error).
1031  */
1032 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
1033                                   Error **errp)
1034 {
1035     NBDClient *client = req->client;
1036 
1037     g_assert(qemu_in_coroutine());
1038     assert(client->recv_coroutine == qemu_coroutine_self());
1039     if (nbd_receive_request(client->ioc, request, errp) < 0) {
1040         return -EIO;
1041     }
1042 
1043     trace_nbd_co_receive_request_decode_type(request->handle, request->type);
1044 
1045     if (request->type != NBD_CMD_WRITE) {
1046         /* No payload, we are ready to read the next request.  */
1047         req->complete = true;
1048     }
1049 
1050     if (request->type == NBD_CMD_DISC) {
1051         /* Special case: we're going to disconnect without a reply,
1052          * whether or not flags, from, or len are bogus */
1053         return -EIO;
1054     }
1055 
1056     /* Check for sanity in the parameters, part 1.  Defer as many
1057      * checks as possible until after reading any NBD_CMD_WRITE
1058      * payload, so we can try and keep the connection alive.  */
1059     if ((request->from + request->len) < request->from) {
1060         error_setg(errp,
1061                    "integer overflow detected, you're probably being attacked");
1062         return -EINVAL;
1063     }
1064 
1065     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) {
1066         if (request->len > NBD_MAX_BUFFER_SIZE) {
1067             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1068                        request->len, NBD_MAX_BUFFER_SIZE);
1069             return -EINVAL;
1070         }
1071 
1072         req->data = blk_try_blockalign(client->exp->blk, request->len);
1073         if (req->data == NULL) {
1074             error_setg(errp, "No memory");
1075             return -ENOMEM;
1076         }
1077     }
1078     if (request->type == NBD_CMD_WRITE) {
1079         if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
1080             error_prepend(errp, "reading from socket failed: ");
1081             return -EIO;
1082         }
1083         req->complete = true;
1084 
1085         trace_nbd_co_receive_request_payload_received(request->handle,
1086                                                       request->len);
1087     }
1088 
1089     /* Sanity checks, part 2. */
1090     if (request->from + request->len > client->exp->size) {
1091         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
1092                    ", Size: %" PRIu64, request->from, request->len,
1093                    (uint64_t)client->exp->size);
1094         return request->type == NBD_CMD_WRITE ? -ENOSPC : -EINVAL;
1095     }
1096     if (request->flags & ~(NBD_CMD_FLAG_FUA | NBD_CMD_FLAG_NO_HOLE)) {
1097         error_setg(errp, "unsupported flags (got 0x%x)", request->flags);
1098         return -EINVAL;
1099     }
1100     if (request->type != NBD_CMD_WRITE_ZEROES &&
1101         (request->flags & NBD_CMD_FLAG_NO_HOLE)) {
1102         error_setg(errp, "unexpected flags (got 0x%x)", request->flags);
1103         return -EINVAL;
1104     }
1105 
1106     return 0;
1107 }
1108 
1109 /* Owns a reference to the NBDClient passed as opaque.  */
1110 static coroutine_fn void nbd_trip(void *opaque)
1111 {
1112     NBDClient *client = opaque;
1113     NBDExport *exp = client->exp;
1114     NBDRequestData *req;
1115     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
1116     NBDReply reply;
1117     int ret;
1118     int flags;
1119     int reply_data_len = 0;
1120     Error *local_err = NULL;
1121 
1122     trace_nbd_trip();
1123     if (client->closing) {
1124         nbd_client_put(client);
1125         return;
1126     }
1127 
1128     req = nbd_request_get(client);
1129     ret = nbd_co_receive_request(req, &request, &local_err);
1130     client->recv_coroutine = NULL;
1131     nbd_client_receive_next_request(client);
1132     if (ret == -EIO) {
1133         goto disconnect;
1134     }
1135 
1136     reply.handle = request.handle;
1137     reply.error = 0;
1138 
1139     if (ret < 0) {
1140         reply.error = -ret;
1141         goto reply;
1142     }
1143 
1144     if (client->closing) {
1145         /*
1146          * The client may be closed when we are blocked in
1147          * nbd_co_receive_request()
1148          */
1149         goto done;
1150     }
1151 
1152     switch (request.type) {
1153     case NBD_CMD_READ:
1154         /* XXX: NBD Protocol only documents use of FUA with WRITE */
1155         if (request.flags & NBD_CMD_FLAG_FUA) {
1156             ret = blk_co_flush(exp->blk);
1157             if (ret < 0) {
1158                 error_setg_errno(&local_err, -ret, "flush failed");
1159                 reply.error = -ret;
1160                 break;
1161             }
1162         }
1163 
1164         ret = blk_pread(exp->blk, request.from + exp->dev_offset,
1165                         req->data, request.len);
1166         if (ret < 0) {
1167             error_setg_errno(&local_err, -ret, "reading from file failed");
1168             reply.error = -ret;
1169             break;
1170         }
1171 
1172         reply_data_len = request.len;
1173 
1174         break;
1175     case NBD_CMD_WRITE:
1176         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1177             reply.error = EROFS;
1178             break;
1179         }
1180 
1181         flags = 0;
1182         if (request.flags & NBD_CMD_FLAG_FUA) {
1183             flags |= BDRV_REQ_FUA;
1184         }
1185         ret = blk_pwrite(exp->blk, request.from + exp->dev_offset,
1186                          req->data, request.len, flags);
1187         if (ret < 0) {
1188             error_setg_errno(&local_err, -ret, "writing to file failed");
1189             reply.error = -ret;
1190         }
1191 
1192         break;
1193     case NBD_CMD_WRITE_ZEROES:
1194         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1195             error_setg(&local_err, "Server is read-only, return error");
1196             reply.error = EROFS;
1197             break;
1198         }
1199 
1200         flags = 0;
1201         if (request.flags & NBD_CMD_FLAG_FUA) {
1202             flags |= BDRV_REQ_FUA;
1203         }
1204         if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) {
1205             flags |= BDRV_REQ_MAY_UNMAP;
1206         }
1207         ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset,
1208                                 request.len, flags);
1209         if (ret < 0) {
1210             error_setg_errno(&local_err, -ret, "writing to file failed");
1211             reply.error = -ret;
1212         }
1213 
1214         break;
1215     case NBD_CMD_DISC:
1216         /* unreachable, thanks to special case in nbd_co_receive_request() */
1217         abort();
1218 
1219     case NBD_CMD_FLUSH:
1220         ret = blk_co_flush(exp->blk);
1221         if (ret < 0) {
1222             error_setg_errno(&local_err, -ret, "flush failed");
1223             reply.error = -ret;
1224         }
1225 
1226         break;
1227     case NBD_CMD_TRIM:
1228         ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset,
1229                               request.len);
1230         if (ret < 0) {
1231             error_setg_errno(&local_err, -ret, "discard failed");
1232             reply.error = -ret;
1233         }
1234 
1235         break;
1236     default:
1237         error_setg(&local_err, "invalid request type (%" PRIu32 ") received",
1238                    request.type);
1239         reply.error = EINVAL;
1240     }
1241 
1242 reply:
1243     if (local_err) {
1244         /* If we are here local_err is not fatal error, already stored in
1245          * reply.error */
1246         error_report_err(local_err);
1247         local_err = NULL;
1248     }
1249 
1250     if (nbd_co_send_reply(req, &reply, reply_data_len, &local_err) < 0) {
1251         error_prepend(&local_err, "Failed to send reply: ");
1252         goto disconnect;
1253     }
1254 
1255     /* We must disconnect after NBD_CMD_WRITE if we did not
1256      * read the payload.
1257      */
1258     if (!req->complete) {
1259         error_setg(&local_err, "Request handling failed in intermediate state");
1260         goto disconnect;
1261     }
1262 
1263 done:
1264     nbd_request_put(req);
1265     nbd_client_put(client);
1266     return;
1267 
1268 disconnect:
1269     if (local_err) {
1270         error_reportf_err(local_err, "Disconnect client, due to: ");
1271     }
1272     nbd_request_put(req);
1273     client_close(client, true);
1274     nbd_client_put(client);
1275 }
1276 
1277 static void nbd_client_receive_next_request(NBDClient *client)
1278 {
1279     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
1280         nbd_client_get(client);
1281         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
1282         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
1283     }
1284 }
1285 
1286 static coroutine_fn void nbd_co_client_start(void *opaque)
1287 {
1288     NBDClient *client = opaque;
1289     NBDExport *exp = client->exp;
1290     Error *local_err = NULL;
1291 
1292     if (exp) {
1293         nbd_export_get(exp);
1294         QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1295     }
1296     qemu_co_mutex_init(&client->send_lock);
1297 
1298     if (nbd_negotiate(client, &local_err)) {
1299         if (local_err) {
1300             error_report_err(local_err);
1301         }
1302         client_close(client, false);
1303         return;
1304     }
1305 
1306     nbd_client_receive_next_request(client);
1307 }
1308 
1309 /*
1310  * Create a new client listener on the given export @exp, using the
1311  * given channel @sioc.  Begin servicing it in a coroutine.  When the
1312  * connection closes, call @close_fn with an indication of whether the
1313  * client completed negotiation.
1314  */
1315 void nbd_client_new(NBDExport *exp,
1316                     QIOChannelSocket *sioc,
1317                     QCryptoTLSCreds *tlscreds,
1318                     const char *tlsaclname,
1319                     void (*close_fn)(NBDClient *, bool))
1320 {
1321     NBDClient *client;
1322     Coroutine *co;
1323 
1324     client = g_malloc0(sizeof(NBDClient));
1325     client->refcount = 1;
1326     client->exp = exp;
1327     client->tlscreds = tlscreds;
1328     if (tlscreds) {
1329         object_ref(OBJECT(client->tlscreds));
1330     }
1331     client->tlsaclname = g_strdup(tlsaclname);
1332     client->sioc = sioc;
1333     object_ref(OBJECT(client->sioc));
1334     client->ioc = QIO_CHANNEL(sioc);
1335     object_ref(OBJECT(client->ioc));
1336     client->close_fn = close_fn;
1337 
1338     co = qemu_coroutine_create(nbd_co_client_start, client);
1339     qemu_coroutine_enter(co);
1340 }
1341