xref: /openbmc/qemu/nbd/server.c (revision 9af23989)
1 /*
2  *  Copyright (C) 2016-2018 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 EOVERFLOW:
44         return NBD_EOVERFLOW;
45     case ESHUTDOWN:
46         return NBD_ESHUTDOWN;
47     case EINVAL:
48     default:
49         return NBD_EINVAL;
50     }
51 }
52 
53 /* Definitions for opaque data types */
54 
55 typedef struct NBDRequestData NBDRequestData;
56 
57 struct NBDRequestData {
58     QSIMPLEQ_ENTRY(NBDRequestData) entry;
59     NBDClient *client;
60     uint8_t *data;
61     bool complete;
62 };
63 
64 struct NBDExport {
65     int refcount;
66     void (*close)(NBDExport *exp);
67 
68     BlockBackend *blk;
69     char *name;
70     char *description;
71     off_t dev_offset;
72     off_t size;
73     uint16_t nbdflags;
74     QTAILQ_HEAD(, NBDClient) clients;
75     QTAILQ_ENTRY(NBDExport) next;
76 
77     AioContext *ctx;
78 
79     BlockBackend *eject_notifier_blk;
80     Notifier eject_notifier;
81 };
82 
83 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
84 
85 struct NBDClient {
86     int refcount;
87     void (*close_fn)(NBDClient *client, bool negotiated);
88 
89     NBDExport *exp;
90     QCryptoTLSCreds *tlscreds;
91     char *tlsaclname;
92     QIOChannelSocket *sioc; /* The underlying data channel */
93     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
94 
95     Coroutine *recv_coroutine;
96 
97     CoMutex send_lock;
98     Coroutine *send_coroutine;
99 
100     QTAILQ_ENTRY(NBDClient) next;
101     int nb_requests;
102     bool closing;
103 
104     bool structured_reply;
105 
106     uint32_t opt; /* Current option being negotiated */
107     uint32_t optlen; /* remaining length of data in ioc for the option being
108                         negotiated now */
109 };
110 
111 static void nbd_client_receive_next_request(NBDClient *client);
112 
113 /* Basic flow for negotiation
114 
115    Server         Client
116    Negotiate
117 
118    or
119 
120    Server         Client
121    Negotiate #1
122                   Option
123    Negotiate #2
124 
125    ----
126 
127    followed by
128 
129    Server         Client
130                   Request
131    Response
132                   Request
133    Response
134                   ...
135    ...
136                   Request (type == 2)
137 
138 */
139 
140 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
141                                      uint32_t type, uint32_t length)
142 {
143     stq_be_p(&rep->magic, NBD_REP_MAGIC);
144     stl_be_p(&rep->option, option);
145     stl_be_p(&rep->type, type);
146     stl_be_p(&rep->length, length);
147 }
148 
149 /* Send a reply header, including length, but no payload.
150  * Return -errno on error, 0 on success. */
151 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
152                                       uint32_t len, Error **errp)
153 {
154     NBDOptionReply rep;
155 
156     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
157                                      type, nbd_rep_lookup(type), len);
158 
159     assert(len < NBD_MAX_BUFFER_SIZE);
160 
161     set_be_option_rep(&rep, client->opt, type, len);
162     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
163 }
164 
165 /* Send a reply header with default 0 length.
166  * Return -errno on error, 0 on success. */
167 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
168                                   Error **errp)
169 {
170     return nbd_negotiate_send_rep_len(client, type, 0, errp);
171 }
172 
173 /* Send an error reply.
174  * Return -errno on error, 0 on success. */
175 static int GCC_FMT_ATTR(4, 0)
176 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
177                             Error **errp, const char *fmt, va_list va)
178 {
179     char *msg;
180     int ret;
181     size_t len;
182 
183     msg = g_strdup_vprintf(fmt, va);
184     len = strlen(msg);
185     assert(len < 4096);
186     trace_nbd_negotiate_send_rep_err(msg);
187     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
188     if (ret < 0) {
189         goto out;
190     }
191     if (nbd_write(client->ioc, msg, len, errp) < 0) {
192         error_prepend(errp, "write failed (error message): ");
193         ret = -EIO;
194     } else {
195         ret = 0;
196     }
197 
198 out:
199     g_free(msg);
200     return ret;
201 }
202 
203 /* Send an error reply.
204  * Return -errno on error, 0 on success. */
205 static int GCC_FMT_ATTR(4, 5)
206 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
207                            Error **errp, const char *fmt, ...)
208 {
209     va_list va;
210     int ret;
211 
212     va_start(va, fmt);
213     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
214     va_end(va);
215     return ret;
216 }
217 
218 /* Drop remainder of the current option, and send a reply with the
219  * given error type and message. Return -errno on read or write
220  * failure; or 0 if connection is still live. */
221 static int GCC_FMT_ATTR(4, 5)
222 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
223              const char *fmt, ...)
224 {
225     int ret = nbd_drop(client->ioc, client->optlen, errp);
226     va_list va;
227 
228     client->optlen = 0;
229     if (!ret) {
230         va_start(va, fmt);
231         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
232         va_end(va);
233     }
234     return ret;
235 }
236 
237 /* Read size bytes from the unparsed payload of the current option.
238  * Return -errno on I/O error, 0 if option was completely handled by
239  * sending a reply about inconsistent lengths, or 1 on success. */
240 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
241                         Error **errp)
242 {
243     if (size > client->optlen) {
244         return nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp,
245                             "Inconsistent lengths in option %s",
246                             nbd_opt_lookup(client->opt));
247     }
248     client->optlen -= size;
249     return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
250 }
251 
252 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
253  * Return -errno on error, 0 on success. */
254 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
255                                        Error **errp)
256 {
257     size_t name_len, desc_len;
258     uint32_t len;
259     const char *name = exp->name ? exp->name : "";
260     const char *desc = exp->description ? exp->description : "";
261     QIOChannel *ioc = client->ioc;
262     int ret;
263 
264     trace_nbd_negotiate_send_rep_list(name, desc);
265     name_len = strlen(name);
266     desc_len = strlen(desc);
267     len = name_len + desc_len + sizeof(len);
268     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
269     if (ret < 0) {
270         return ret;
271     }
272 
273     len = cpu_to_be32(name_len);
274     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
275         error_prepend(errp, "write failed (name length): ");
276         return -EINVAL;
277     }
278 
279     if (nbd_write(ioc, name, name_len, errp) < 0) {
280         error_prepend(errp, "write failed (name buffer): ");
281         return -EINVAL;
282     }
283 
284     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
285         error_prepend(errp, "write failed (description buffer): ");
286         return -EINVAL;
287     }
288 
289     return 0;
290 }
291 
292 /* Process the NBD_OPT_LIST command, with a potential series of replies.
293  * Return -errno on error, 0 on success. */
294 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
295 {
296     NBDExport *exp;
297     assert(client->opt == NBD_OPT_LIST);
298 
299     /* For each export, send a NBD_REP_SERVER reply. */
300     QTAILQ_FOREACH(exp, &exports, next) {
301         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
302             return -EINVAL;
303         }
304     }
305     /* Finish with a NBD_REP_ACK. */
306     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
307 }
308 
309 /* Send a reply to NBD_OPT_EXPORT_NAME.
310  * Return -errno on error, 0 on success. */
311 static int nbd_negotiate_handle_export_name(NBDClient *client,
312                                             uint16_t myflags, bool no_zeroes,
313                                             Error **errp)
314 {
315     char name[NBD_MAX_NAME_SIZE + 1];
316     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
317     size_t len;
318     int ret;
319 
320     /* Client sends:
321         [20 ..  xx]   export name (length bytes)
322        Server replies:
323         [ 0 ..   7]   size
324         [ 8 ..   9]   export flags
325         [10 .. 133]   reserved     (0) [unless no_zeroes]
326      */
327     trace_nbd_negotiate_handle_export_name();
328     if (client->optlen >= sizeof(name)) {
329         error_setg(errp, "Bad length received");
330         return -EINVAL;
331     }
332     if (nbd_read(client->ioc, name, client->optlen, errp) < 0) {
333         error_prepend(errp, "read failed: ");
334         return -EIO;
335     }
336     name[client->optlen] = '\0';
337     client->optlen = 0;
338 
339     trace_nbd_negotiate_handle_export_name_request(name);
340 
341     client->exp = nbd_export_find(name);
342     if (!client->exp) {
343         error_setg(errp, "export not found");
344         return -EINVAL;
345     }
346 
347     trace_nbd_negotiate_new_style_size_flags(client->exp->size,
348                                              client->exp->nbdflags | myflags);
349     stq_be_p(buf, client->exp->size);
350     stw_be_p(buf + 8, client->exp->nbdflags | myflags);
351     len = no_zeroes ? 10 : sizeof(buf);
352     ret = nbd_write(client->ioc, buf, len, errp);
353     if (ret < 0) {
354         error_prepend(errp, "write failed: ");
355         return ret;
356     }
357 
358     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
359     nbd_export_get(client->exp);
360 
361     return 0;
362 }
363 
364 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
365  * The buffer does NOT include the info type prefix.
366  * Return -errno on error, 0 if ready to send more. */
367 static int nbd_negotiate_send_info(NBDClient *client,
368                                    uint16_t info, uint32_t length, void *buf,
369                                    Error **errp)
370 {
371     int rc;
372 
373     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
374     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
375                                     sizeof(info) + length, errp);
376     if (rc < 0) {
377         return rc;
378     }
379     cpu_to_be16s(&info);
380     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
381         return -EIO;
382     }
383     if (nbd_write(client->ioc, buf, length, errp) < 0) {
384         return -EIO;
385     }
386     return 0;
387 }
388 
389 /* nbd_reject_length: Handle any unexpected payload.
390  * @fatal requests that we quit talking to the client, even if we are able
391  * to successfully send an error reply.
392  * Return:
393  * -errno  transmission error occurred or @fatal was requested, errp is set
394  * 0       error message successfully sent to client, errp is not set
395  */
396 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
397 {
398     int ret;
399 
400     assert(client->optlen);
401     ret = nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp,
402                        "option '%s' has unexpected length",
403                        nbd_opt_lookup(client->opt));
404     if (fatal && !ret) {
405         error_setg(errp, "option '%s' has unexpected length",
406                    nbd_opt_lookup(client->opt));
407         return -EINVAL;
408     }
409     return ret;
410 }
411 
412 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
413  * Return -errno on error, 0 if ready for next option, and 1 to move
414  * into transmission phase.  */
415 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
416                                      Error **errp)
417 {
418     int rc;
419     char name[NBD_MAX_NAME_SIZE + 1];
420     NBDExport *exp;
421     uint16_t requests;
422     uint16_t request;
423     uint32_t namelen;
424     bool sendname = false;
425     bool blocksize = false;
426     uint32_t sizes[3];
427     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
428 
429     /* Client sends:
430         4 bytes: L, name length (can be 0)
431         L bytes: export name
432         2 bytes: N, number of requests (can be 0)
433         N * 2 bytes: N requests
434     */
435     rc = nbd_opt_read(client, &namelen, sizeof(namelen), errp);
436     if (rc <= 0) {
437         return rc;
438     }
439     be32_to_cpus(&namelen);
440     if (namelen >= sizeof(name)) {
441         return nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp,
442                             "name too long for qemu");
443     }
444     rc = nbd_opt_read(client, name, namelen, errp);
445     if (rc <= 0) {
446         return rc;
447     }
448     name[namelen] = '\0';
449     trace_nbd_negotiate_handle_export_name_request(name);
450 
451     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
452     if (rc <= 0) {
453         return rc;
454     }
455     be16_to_cpus(&requests);
456     trace_nbd_negotiate_handle_info_requests(requests);
457     while (requests--) {
458         rc = nbd_opt_read(client, &request, sizeof(request), errp);
459         if (rc <= 0) {
460             return rc;
461         }
462         be16_to_cpus(&request);
463         trace_nbd_negotiate_handle_info_request(request,
464                                                 nbd_info_lookup(request));
465         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
466          * everything else is either a request we don't know or
467          * something we send regardless of request */
468         switch (request) {
469         case NBD_INFO_NAME:
470             sendname = true;
471             break;
472         case NBD_INFO_BLOCK_SIZE:
473             blocksize = true;
474             break;
475         }
476     }
477     if (client->optlen) {
478         return nbd_reject_length(client, false, errp);
479     }
480 
481     exp = nbd_export_find(name);
482     if (!exp) {
483         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
484                                           errp, "export '%s' not present",
485                                           name);
486     }
487 
488     /* Don't bother sending NBD_INFO_NAME unless client requested it */
489     if (sendname) {
490         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
491                                      errp);
492         if (rc < 0) {
493             return rc;
494         }
495     }
496 
497     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
498      * client request */
499     if (exp->description) {
500         size_t len = strlen(exp->description);
501 
502         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
503                                      len, exp->description, errp);
504         if (rc < 0) {
505             return rc;
506         }
507     }
508 
509     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
510      * according to whether the client requested it, and according to
511      * whether this is OPT_INFO or OPT_GO. */
512     /* minimum - 1 for back-compat, or 512 if client is new enough.
513      * TODO: consult blk_bs(blk)->bl.request_alignment? */
514     sizes[0] =
515             (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
516     /* preferred - Hard-code to 4096 for now.
517      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
518     sizes[1] = 4096;
519     /* maximum - At most 32M, but smaller as appropriate. */
520     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
521     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
522     cpu_to_be32s(&sizes[0]);
523     cpu_to_be32s(&sizes[1]);
524     cpu_to_be32s(&sizes[2]);
525     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
526                                  sizeof(sizes), sizes, errp);
527     if (rc < 0) {
528         return rc;
529     }
530 
531     /* Send NBD_INFO_EXPORT always */
532     trace_nbd_negotiate_new_style_size_flags(exp->size,
533                                              exp->nbdflags | myflags);
534     stq_be_p(buf, exp->size);
535     stw_be_p(buf + 8, exp->nbdflags | myflags);
536     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
537                                  sizeof(buf), buf, errp);
538     if (rc < 0) {
539         return rc;
540     }
541 
542     /* If the client is just asking for NBD_OPT_INFO, but forgot to
543      * request block sizes, return an error.
544      * TODO: consult blk_bs(blk)->request_align, and only error if it
545      * is not 1? */
546     if (client->opt == NBD_OPT_INFO && !blocksize) {
547         return nbd_negotiate_send_rep_err(client,
548                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
549                                           errp,
550                                           "request NBD_INFO_BLOCK_SIZE to "
551                                           "use this export");
552     }
553 
554     /* Final reply */
555     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
556     if (rc < 0) {
557         return rc;
558     }
559 
560     if (client->opt == NBD_OPT_GO) {
561         client->exp = exp;
562         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
563         nbd_export_get(client->exp);
564         rc = 1;
565     }
566     return rc;
567 }
568 
569 
570 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
571  * new channel for all further (now-encrypted) communication. */
572 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
573                                                  Error **errp)
574 {
575     QIOChannel *ioc;
576     QIOChannelTLS *tioc;
577     struct NBDTLSHandshakeData data = { 0 };
578 
579     assert(client->opt == NBD_OPT_STARTTLS);
580 
581     trace_nbd_negotiate_handle_starttls();
582     ioc = client->ioc;
583 
584     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
585         return NULL;
586     }
587 
588     tioc = qio_channel_tls_new_server(ioc,
589                                       client->tlscreds,
590                                       client->tlsaclname,
591                                       errp);
592     if (!tioc) {
593         return NULL;
594     }
595 
596     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
597     trace_nbd_negotiate_handle_starttls_handshake();
598     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
599     qio_channel_tls_handshake(tioc,
600                               nbd_tls_handshake,
601                               &data,
602                               NULL);
603 
604     if (!data.complete) {
605         g_main_loop_run(data.loop);
606     }
607     g_main_loop_unref(data.loop);
608     if (data.error) {
609         object_unref(OBJECT(tioc));
610         error_propagate(errp, data.error);
611         return NULL;
612     }
613 
614     return QIO_CHANNEL(tioc);
615 }
616 
617 /* nbd_negotiate_options
618  * Process all NBD_OPT_* client option commands, during fixed newstyle
619  * negotiation.
620  * Return:
621  * -errno  on error, errp is set
622  * 0       on successful negotiation, errp is not set
623  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
624  *         errp is not set
625  */
626 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
627                                  Error **errp)
628 {
629     uint32_t flags;
630     bool fixedNewstyle = false;
631     bool no_zeroes = false;
632 
633     /* Client sends:
634         [ 0 ..   3]   client flags
635 
636        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
637         [ 0 ..   7]   NBD_OPTS_MAGIC
638         [ 8 ..  11]   NBD option
639         [12 ..  15]   Data length
640         ...           Rest of request
641 
642         [ 0 ..   7]   NBD_OPTS_MAGIC
643         [ 8 ..  11]   Second NBD option
644         [12 ..  15]   Data length
645         ...           Rest of request
646     */
647 
648     if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
649         error_prepend(errp, "read failed: ");
650         return -EIO;
651     }
652     be32_to_cpus(&flags);
653     trace_nbd_negotiate_options_flags(flags);
654     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
655         fixedNewstyle = true;
656         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
657     }
658     if (flags & NBD_FLAG_C_NO_ZEROES) {
659         no_zeroes = true;
660         flags &= ~NBD_FLAG_C_NO_ZEROES;
661     }
662     if (flags != 0) {
663         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
664         return -EINVAL;
665     }
666 
667     while (1) {
668         int ret;
669         uint32_t option, length;
670         uint64_t magic;
671 
672         if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
673             error_prepend(errp, "read failed: ");
674             return -EINVAL;
675         }
676         magic = be64_to_cpu(magic);
677         trace_nbd_negotiate_options_check_magic(magic);
678         if (magic != NBD_OPTS_MAGIC) {
679             error_setg(errp, "Bad magic received");
680             return -EINVAL;
681         }
682 
683         if (nbd_read(client->ioc, &option,
684                      sizeof(option), errp) < 0) {
685             error_prepend(errp, "read failed: ");
686             return -EINVAL;
687         }
688         option = be32_to_cpu(option);
689         client->opt = option;
690 
691         if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
692             error_prepend(errp, "read failed: ");
693             return -EINVAL;
694         }
695         length = be32_to_cpu(length);
696         assert(!client->optlen);
697         client->optlen = length;
698 
699         if (length > NBD_MAX_BUFFER_SIZE) {
700             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
701                        length, NBD_MAX_BUFFER_SIZE);
702             return -EINVAL;
703         }
704 
705         trace_nbd_negotiate_options_check_option(option,
706                                                  nbd_opt_lookup(option));
707         if (client->tlscreds &&
708             client->ioc == (QIOChannel *)client->sioc) {
709             QIOChannel *tioc;
710             if (!fixedNewstyle) {
711                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
712                 return -EINVAL;
713             }
714             switch (option) {
715             case NBD_OPT_STARTTLS:
716                 if (length) {
717                     /* Unconditionally drop the connection if the client
718                      * can't start a TLS negotiation correctly */
719                     return nbd_reject_length(client, true, errp);
720                 }
721                 tioc = nbd_negotiate_handle_starttls(client, errp);
722                 if (!tioc) {
723                     return -EIO;
724                 }
725                 ret = 0;
726                 object_unref(OBJECT(client->ioc));
727                 client->ioc = QIO_CHANNEL(tioc);
728                 break;
729 
730             case NBD_OPT_EXPORT_NAME:
731                 /* No way to return an error to client, so drop connection */
732                 error_setg(errp, "Option 0x%x not permitted before TLS",
733                            option);
734                 return -EINVAL;
735 
736             default:
737                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, errp,
738                                    "Option 0x%" PRIx32
739                                    "not permitted before TLS", option);
740                 /* Let the client keep trying, unless they asked to
741                  * quit. In this mode, we've already sent an error, so
742                  * we can't ack the abort.  */
743                 if (option == NBD_OPT_ABORT) {
744                     return 1;
745                 }
746                 break;
747             }
748         } else if (fixedNewstyle) {
749             switch (option) {
750             case NBD_OPT_LIST:
751                 if (length) {
752                     ret = nbd_reject_length(client, false, errp);
753                 } else {
754                     ret = nbd_negotiate_handle_list(client, errp);
755                 }
756                 break;
757 
758             case NBD_OPT_ABORT:
759                 /* NBD spec says we must try to reply before
760                  * disconnecting, but that we must also tolerate
761                  * guests that don't wait for our reply. */
762                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
763                 return 1;
764 
765             case NBD_OPT_EXPORT_NAME:
766                 return nbd_negotiate_handle_export_name(client,
767                                                         myflags, no_zeroes,
768                                                         errp);
769 
770             case NBD_OPT_INFO:
771             case NBD_OPT_GO:
772                 ret = nbd_negotiate_handle_info(client, myflags, errp);
773                 if (ret == 1) {
774                     assert(option == NBD_OPT_GO);
775                     return 0;
776                 }
777                 break;
778 
779             case NBD_OPT_STARTTLS:
780                 if (length) {
781                     ret = nbd_reject_length(client, false, errp);
782                 } else if (client->tlscreds) {
783                     ret = nbd_negotiate_send_rep_err(client,
784                                                      NBD_REP_ERR_INVALID, errp,
785                                                      "TLS already enabled");
786                 } else {
787                     ret = nbd_negotiate_send_rep_err(client,
788                                                      NBD_REP_ERR_POLICY, errp,
789                                                      "TLS not configured");
790                 }
791                 break;
792 
793             case NBD_OPT_STRUCTURED_REPLY:
794                 if (length) {
795                     ret = nbd_reject_length(client, false, errp);
796                 } else if (client->structured_reply) {
797                     ret = nbd_negotiate_send_rep_err(
798                         client, NBD_REP_ERR_INVALID, errp,
799                         "structured reply already negotiated");
800                 } else {
801                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
802                     client->structured_reply = true;
803                     myflags |= NBD_FLAG_SEND_DF;
804                 }
805                 break;
806 
807             default:
808                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
809                                    "Unsupported option %" PRIu32 " (%s)",
810                                    option, nbd_opt_lookup(option));
811                 break;
812             }
813         } else {
814             /*
815              * If broken new-style we should drop the connection
816              * for anything except NBD_OPT_EXPORT_NAME
817              */
818             switch (option) {
819             case NBD_OPT_EXPORT_NAME:
820                 return nbd_negotiate_handle_export_name(client,
821                                                         myflags, no_zeroes,
822                                                         errp);
823 
824             default:
825                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
826                            option, nbd_opt_lookup(option));
827                 return -EINVAL;
828             }
829         }
830         if (ret < 0) {
831             return ret;
832         }
833     }
834 }
835 
836 /* nbd_negotiate
837  * Return:
838  * -errno  on error, errp is set
839  * 0       on successful negotiation, errp is not set
840  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
841  *         errp is not set
842  */
843 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
844 {
845     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
846     int ret;
847     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
848                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
849                               NBD_FLAG_SEND_WRITE_ZEROES);
850     bool oldStyle;
851 
852     /* Old style negotiation header, no room for options
853         [ 0 ..   7]   passwd       ("NBDMAGIC")
854         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
855         [16 ..  23]   size
856         [24 ..  27]   export flags (zero-extended)
857         [28 .. 151]   reserved     (0)
858 
859        New style negotiation header, client can send options
860         [ 0 ..   7]   passwd       ("NBDMAGIC")
861         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
862         [16 ..  17]   server flags (0)
863         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
864      */
865 
866     qio_channel_set_blocking(client->ioc, false, NULL);
867 
868     trace_nbd_negotiate_begin();
869     memcpy(buf, "NBDMAGIC", 8);
870 
871     oldStyle = client->exp != NULL && !client->tlscreds;
872     if (oldStyle) {
873         trace_nbd_negotiate_old_style(client->exp->size,
874                                       client->exp->nbdflags | myflags);
875         stq_be_p(buf + 8, NBD_CLIENT_MAGIC);
876         stq_be_p(buf + 16, client->exp->size);
877         stl_be_p(buf + 24, client->exp->nbdflags | myflags);
878 
879         if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) {
880             error_prepend(errp, "write failed: ");
881             return -EINVAL;
882         }
883     } else {
884         stq_be_p(buf + 8, NBD_OPTS_MAGIC);
885         stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
886 
887         if (nbd_write(client->ioc, buf, 18, errp) < 0) {
888             error_prepend(errp, "write failed: ");
889             return -EINVAL;
890         }
891         ret = nbd_negotiate_options(client, myflags, errp);
892         if (ret != 0) {
893             if (ret < 0) {
894                 error_prepend(errp, "option negotiation failed: ");
895             }
896             return ret;
897         }
898     }
899 
900     assert(!client->optlen);
901     trace_nbd_negotiate_success();
902 
903     return 0;
904 }
905 
906 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
907                                Error **errp)
908 {
909     uint8_t buf[NBD_REQUEST_SIZE];
910     uint32_t magic;
911     int ret;
912 
913     ret = nbd_read(ioc, buf, sizeof(buf), errp);
914     if (ret < 0) {
915         return ret;
916     }
917 
918     /* Request
919        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
920        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
921        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
922        [ 8 .. 15]   handle
923        [16 .. 23]   from
924        [24 .. 27]   len
925      */
926 
927     magic = ldl_be_p(buf);
928     request->flags  = lduw_be_p(buf + 4);
929     request->type   = lduw_be_p(buf + 6);
930     request->handle = ldq_be_p(buf + 8);
931     request->from   = ldq_be_p(buf + 16);
932     request->len    = ldl_be_p(buf + 24);
933 
934     trace_nbd_receive_request(magic, request->flags, request->type,
935                               request->from, request->len);
936 
937     if (magic != NBD_REQUEST_MAGIC) {
938         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
939         return -EINVAL;
940     }
941     return 0;
942 }
943 
944 #define MAX_NBD_REQUESTS 16
945 
946 void nbd_client_get(NBDClient *client)
947 {
948     client->refcount++;
949 }
950 
951 void nbd_client_put(NBDClient *client)
952 {
953     if (--client->refcount == 0) {
954         /* The last reference should be dropped by client->close,
955          * which is called by client_close.
956          */
957         assert(client->closing);
958 
959         qio_channel_detach_aio_context(client->ioc);
960         object_unref(OBJECT(client->sioc));
961         object_unref(OBJECT(client->ioc));
962         if (client->tlscreds) {
963             object_unref(OBJECT(client->tlscreds));
964         }
965         g_free(client->tlsaclname);
966         if (client->exp) {
967             QTAILQ_REMOVE(&client->exp->clients, client, next);
968             nbd_export_put(client->exp);
969         }
970         g_free(client);
971     }
972 }
973 
974 static void client_close(NBDClient *client, bool negotiated)
975 {
976     if (client->closing) {
977         return;
978     }
979 
980     client->closing = true;
981 
982     /* Force requests to finish.  They will drop their own references,
983      * then we'll close the socket and free the NBDClient.
984      */
985     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
986                          NULL);
987 
988     /* Also tell the client, so that they release their reference.  */
989     if (client->close_fn) {
990         client->close_fn(client, negotiated);
991     }
992 }
993 
994 static NBDRequestData *nbd_request_get(NBDClient *client)
995 {
996     NBDRequestData *req;
997 
998     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
999     client->nb_requests++;
1000 
1001     req = g_new0(NBDRequestData, 1);
1002     nbd_client_get(client);
1003     req->client = client;
1004     return req;
1005 }
1006 
1007 static void nbd_request_put(NBDRequestData *req)
1008 {
1009     NBDClient *client = req->client;
1010 
1011     if (req->data) {
1012         qemu_vfree(req->data);
1013     }
1014     g_free(req);
1015 
1016     client->nb_requests--;
1017     nbd_client_receive_next_request(client);
1018 
1019     nbd_client_put(client);
1020 }
1021 
1022 static void blk_aio_attached(AioContext *ctx, void *opaque)
1023 {
1024     NBDExport *exp = opaque;
1025     NBDClient *client;
1026 
1027     trace_nbd_blk_aio_attached(exp->name, ctx);
1028 
1029     exp->ctx = ctx;
1030 
1031     QTAILQ_FOREACH(client, &exp->clients, next) {
1032         qio_channel_attach_aio_context(client->ioc, ctx);
1033         if (client->recv_coroutine) {
1034             aio_co_schedule(ctx, client->recv_coroutine);
1035         }
1036         if (client->send_coroutine) {
1037             aio_co_schedule(ctx, client->send_coroutine);
1038         }
1039     }
1040 }
1041 
1042 static void blk_aio_detach(void *opaque)
1043 {
1044     NBDExport *exp = opaque;
1045     NBDClient *client;
1046 
1047     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1048 
1049     QTAILQ_FOREACH(client, &exp->clients, next) {
1050         qio_channel_detach_aio_context(client->ioc);
1051     }
1052 
1053     exp->ctx = NULL;
1054 }
1055 
1056 static void nbd_eject_notifier(Notifier *n, void *data)
1057 {
1058     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1059     nbd_export_close(exp);
1060 }
1061 
1062 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1063                           uint16_t nbdflags, void (*close)(NBDExport *),
1064                           bool writethrough, BlockBackend *on_eject_blk,
1065                           Error **errp)
1066 {
1067     AioContext *ctx;
1068     BlockBackend *blk;
1069     NBDExport *exp = g_new0(NBDExport, 1);
1070     uint64_t perm;
1071     int ret;
1072 
1073     /*
1074      * NBD exports are used for non-shared storage migration.  Make sure
1075      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1076      * access since the export could be available before migration handover.
1077      */
1078     ctx = bdrv_get_aio_context(bs);
1079     aio_context_acquire(ctx);
1080     bdrv_invalidate_cache(bs, NULL);
1081     aio_context_release(ctx);
1082 
1083     /* Don't allow resize while the NBD server is running, otherwise we don't
1084      * care what happens with the node. */
1085     perm = BLK_PERM_CONSISTENT_READ;
1086     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1087         perm |= BLK_PERM_WRITE;
1088     }
1089     blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1090                         BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1091     ret = blk_insert_bs(blk, bs, errp);
1092     if (ret < 0) {
1093         goto fail;
1094     }
1095     blk_set_enable_write_cache(blk, !writethrough);
1096 
1097     exp->refcount = 1;
1098     QTAILQ_INIT(&exp->clients);
1099     exp->blk = blk;
1100     exp->dev_offset = dev_offset;
1101     exp->nbdflags = nbdflags;
1102     exp->size = size < 0 ? blk_getlength(blk) : size;
1103     if (exp->size < 0) {
1104         error_setg_errno(errp, -exp->size,
1105                          "Failed to determine the NBD export's length");
1106         goto fail;
1107     }
1108     exp->size -= exp->size % BDRV_SECTOR_SIZE;
1109 
1110     exp->close = close;
1111     exp->ctx = blk_get_aio_context(blk);
1112     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1113 
1114     if (on_eject_blk) {
1115         blk_ref(on_eject_blk);
1116         exp->eject_notifier_blk = on_eject_blk;
1117         exp->eject_notifier.notify = nbd_eject_notifier;
1118         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1119     }
1120     return exp;
1121 
1122 fail:
1123     blk_unref(blk);
1124     g_free(exp);
1125     return NULL;
1126 }
1127 
1128 NBDExport *nbd_export_find(const char *name)
1129 {
1130     NBDExport *exp;
1131     QTAILQ_FOREACH(exp, &exports, next) {
1132         if (strcmp(name, exp->name) == 0) {
1133             return exp;
1134         }
1135     }
1136 
1137     return NULL;
1138 }
1139 
1140 void nbd_export_set_name(NBDExport *exp, const char *name)
1141 {
1142     if (exp->name == name) {
1143         return;
1144     }
1145 
1146     nbd_export_get(exp);
1147     if (exp->name != NULL) {
1148         g_free(exp->name);
1149         exp->name = NULL;
1150         QTAILQ_REMOVE(&exports, exp, next);
1151         nbd_export_put(exp);
1152     }
1153     if (name != NULL) {
1154         nbd_export_get(exp);
1155         exp->name = g_strdup(name);
1156         QTAILQ_INSERT_TAIL(&exports, exp, next);
1157     }
1158     nbd_export_put(exp);
1159 }
1160 
1161 void nbd_export_set_description(NBDExport *exp, const char *description)
1162 {
1163     g_free(exp->description);
1164     exp->description = g_strdup(description);
1165 }
1166 
1167 void nbd_export_close(NBDExport *exp)
1168 {
1169     NBDClient *client, *next;
1170 
1171     nbd_export_get(exp);
1172     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1173         client_close(client, true);
1174     }
1175     nbd_export_set_name(exp, NULL);
1176     nbd_export_set_description(exp, NULL);
1177     nbd_export_put(exp);
1178 }
1179 
1180 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1181 {
1182     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1183         nbd_export_close(exp);
1184         return;
1185     }
1186 
1187     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1188 
1189     error_setg(errp, "export '%s' still in use", exp->name);
1190     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1191 }
1192 
1193 void nbd_export_get(NBDExport *exp)
1194 {
1195     assert(exp->refcount > 0);
1196     exp->refcount++;
1197 }
1198 
1199 void nbd_export_put(NBDExport *exp)
1200 {
1201     assert(exp->refcount > 0);
1202     if (exp->refcount == 1) {
1203         nbd_export_close(exp);
1204     }
1205 
1206     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1207      * if someone calls nbd_export_put() on named export not through
1208      * nbd_export_set_name() when refcount is 1. So, let's assert that
1209      * it is > 0.
1210      */
1211     assert(exp->refcount > 0);
1212     if (--exp->refcount == 0) {
1213         assert(exp->name == NULL);
1214         assert(exp->description == NULL);
1215 
1216         if (exp->close) {
1217             exp->close(exp);
1218         }
1219 
1220         if (exp->blk) {
1221             if (exp->eject_notifier_blk) {
1222                 notifier_remove(&exp->eject_notifier);
1223                 blk_unref(exp->eject_notifier_blk);
1224             }
1225             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1226                                             blk_aio_detach, exp);
1227             blk_unref(exp->blk);
1228             exp->blk = NULL;
1229         }
1230 
1231         g_free(exp);
1232     }
1233 }
1234 
1235 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1236 {
1237     return exp->blk;
1238 }
1239 
1240 void nbd_export_close_all(void)
1241 {
1242     NBDExport *exp, *next;
1243 
1244     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1245         nbd_export_close(exp);
1246     }
1247 }
1248 
1249 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1250                                         unsigned niov, Error **errp)
1251 {
1252     int ret;
1253 
1254     g_assert(qemu_in_coroutine());
1255     qemu_co_mutex_lock(&client->send_lock);
1256     client->send_coroutine = qemu_coroutine_self();
1257 
1258     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1259 
1260     client->send_coroutine = NULL;
1261     qemu_co_mutex_unlock(&client->send_lock);
1262 
1263     return ret;
1264 }
1265 
1266 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1267                                        uint64_t handle)
1268 {
1269     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1270     stl_be_p(&reply->error, error);
1271     stq_be_p(&reply->handle, handle);
1272 }
1273 
1274 static int nbd_co_send_simple_reply(NBDClient *client,
1275                                     uint64_t handle,
1276                                     uint32_t error,
1277                                     void *data,
1278                                     size_t len,
1279                                     Error **errp)
1280 {
1281     NBDSimpleReply reply;
1282     int nbd_err = system_errno_to_nbd_errno(error);
1283     struct iovec iov[] = {
1284         {.iov_base = &reply, .iov_len = sizeof(reply)},
1285         {.iov_base = data, .iov_len = len}
1286     };
1287 
1288     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1289                                    len);
1290     set_be_simple_reply(&reply, nbd_err, handle);
1291 
1292     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1293 }
1294 
1295 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1296                                 uint16_t type, uint64_t handle, uint32_t length)
1297 {
1298     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1299     stw_be_p(&chunk->flags, flags);
1300     stw_be_p(&chunk->type, type);
1301     stq_be_p(&chunk->handle, handle);
1302     stl_be_p(&chunk->length, length);
1303 }
1304 
1305 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1306                                                     uint64_t handle,
1307                                                     Error **errp)
1308 {
1309     NBDStructuredReplyChunk chunk;
1310     struct iovec iov[] = {
1311         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1312     };
1313 
1314     trace_nbd_co_send_structured_done(handle);
1315     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1316 
1317     return nbd_co_send_iov(client, iov, 1, errp);
1318 }
1319 
1320 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1321                                                     uint64_t handle,
1322                                                     uint64_t offset,
1323                                                     void *data,
1324                                                     size_t size,
1325                                                     bool final,
1326                                                     Error **errp)
1327 {
1328     NBDStructuredReadData chunk;
1329     struct iovec iov[] = {
1330         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1331         {.iov_base = data, .iov_len = size}
1332     };
1333 
1334     assert(size);
1335     trace_nbd_co_send_structured_read(handle, offset, data, size);
1336     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1337                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1338                  sizeof(chunk) - sizeof(chunk.h) + size);
1339     stq_be_p(&chunk.offset, offset);
1340 
1341     return nbd_co_send_iov(client, iov, 2, errp);
1342 }
1343 
1344 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1345                                                 uint64_t handle,
1346                                                 uint64_t offset,
1347                                                 uint8_t *data,
1348                                                 size_t size,
1349                                                 Error **errp)
1350 {
1351     int ret = 0;
1352     NBDExport *exp = client->exp;
1353     size_t progress = 0;
1354 
1355     while (progress < size) {
1356         int64_t pnum;
1357         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1358                                              offset + progress,
1359                                              size - progress, &pnum, NULL,
1360                                              NULL);
1361         bool final;
1362 
1363         if (status < 0) {
1364             error_setg_errno(errp, -status, "unable to check for holes");
1365             return status;
1366         }
1367         assert(pnum && pnum <= size - progress);
1368         final = progress + pnum == size;
1369         if (status & BDRV_BLOCK_ZERO) {
1370             NBDStructuredReadHole chunk;
1371             struct iovec iov[] = {
1372                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1373             };
1374 
1375             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1376                                                    pnum);
1377             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1378                          NBD_REPLY_TYPE_OFFSET_HOLE,
1379                          handle, sizeof(chunk) - sizeof(chunk.h));
1380             stq_be_p(&chunk.offset, offset + progress);
1381             stl_be_p(&chunk.length, pnum);
1382             ret = nbd_co_send_iov(client, iov, 1, errp);
1383         } else {
1384             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1385                             data + progress, pnum);
1386             if (ret < 0) {
1387                 error_setg_errno(errp, -ret, "reading from file failed");
1388                 break;
1389             }
1390             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1391                                               data + progress, pnum, final,
1392                                               errp);
1393         }
1394 
1395         if (ret < 0) {
1396             break;
1397         }
1398         progress += pnum;
1399     }
1400     return ret;
1401 }
1402 
1403 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1404                                                      uint64_t handle,
1405                                                      uint32_t error,
1406                                                      const char *msg,
1407                                                      Error **errp)
1408 {
1409     NBDStructuredError chunk;
1410     int nbd_err = system_errno_to_nbd_errno(error);
1411     struct iovec iov[] = {
1412         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1413         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1414     };
1415 
1416     assert(nbd_err);
1417     trace_nbd_co_send_structured_error(handle, nbd_err,
1418                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1419     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1420                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1421     stl_be_p(&chunk.error, nbd_err);
1422     stw_be_p(&chunk.message_length, iov[1].iov_len);
1423 
1424     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1425 }
1426 
1427 /* nbd_co_receive_request
1428  * Collect a client request. Return 0 if request looks valid, -EIO to drop
1429  * connection right away, and any other negative value to report an error to
1430  * the client (although the caller may still need to disconnect after reporting
1431  * the error).
1432  */
1433 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
1434                                   Error **errp)
1435 {
1436     NBDClient *client = req->client;
1437     int valid_flags;
1438 
1439     g_assert(qemu_in_coroutine());
1440     assert(client->recv_coroutine == qemu_coroutine_self());
1441     if (nbd_receive_request(client->ioc, request, errp) < 0) {
1442         return -EIO;
1443     }
1444 
1445     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
1446                                              nbd_cmd_lookup(request->type));
1447 
1448     if (request->type != NBD_CMD_WRITE) {
1449         /* No payload, we are ready to read the next request.  */
1450         req->complete = true;
1451     }
1452 
1453     if (request->type == NBD_CMD_DISC) {
1454         /* Special case: we're going to disconnect without a reply,
1455          * whether or not flags, from, or len are bogus */
1456         return -EIO;
1457     }
1458 
1459     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) {
1460         if (request->len > NBD_MAX_BUFFER_SIZE) {
1461             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1462                        request->len, NBD_MAX_BUFFER_SIZE);
1463             return -EINVAL;
1464         }
1465 
1466         req->data = blk_try_blockalign(client->exp->blk, request->len);
1467         if (req->data == NULL) {
1468             error_setg(errp, "No memory");
1469             return -ENOMEM;
1470         }
1471     }
1472     if (request->type == NBD_CMD_WRITE) {
1473         if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
1474             error_prepend(errp, "reading from socket failed: ");
1475             return -EIO;
1476         }
1477         req->complete = true;
1478 
1479         trace_nbd_co_receive_request_payload_received(request->handle,
1480                                                       request->len);
1481     }
1482 
1483     /* Sanity checks. */
1484     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
1485         (request->type == NBD_CMD_WRITE ||
1486          request->type == NBD_CMD_WRITE_ZEROES ||
1487          request->type == NBD_CMD_TRIM)) {
1488         error_setg(errp, "Export is read-only");
1489         return -EROFS;
1490     }
1491     if (request->from > client->exp->size ||
1492         request->from + request->len > client->exp->size) {
1493         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
1494                    ", Size: %" PRIu64, request->from, request->len,
1495                    (uint64_t)client->exp->size);
1496         return (request->type == NBD_CMD_WRITE ||
1497                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
1498     }
1499     valid_flags = NBD_CMD_FLAG_FUA;
1500     if (request->type == NBD_CMD_READ && client->structured_reply) {
1501         valid_flags |= NBD_CMD_FLAG_DF;
1502     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
1503         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
1504     }
1505     if (request->flags & ~valid_flags) {
1506         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
1507                    nbd_cmd_lookup(request->type), request->flags);
1508         return -EINVAL;
1509     }
1510 
1511     return 0;
1512 }
1513 
1514 /* Owns a reference to the NBDClient passed as opaque.  */
1515 static coroutine_fn void nbd_trip(void *opaque)
1516 {
1517     NBDClient *client = opaque;
1518     NBDExport *exp = client->exp;
1519     NBDRequestData *req;
1520     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
1521     int ret;
1522     int flags;
1523     int reply_data_len = 0;
1524     Error *local_err = NULL;
1525     char *msg = NULL;
1526 
1527     trace_nbd_trip();
1528     if (client->closing) {
1529         nbd_client_put(client);
1530         return;
1531     }
1532 
1533     req = nbd_request_get(client);
1534     ret = nbd_co_receive_request(req, &request, &local_err);
1535     client->recv_coroutine = NULL;
1536     nbd_client_receive_next_request(client);
1537     if (ret == -EIO) {
1538         goto disconnect;
1539     }
1540 
1541     if (ret < 0) {
1542         goto reply;
1543     }
1544 
1545     if (client->closing) {
1546         /*
1547          * The client may be closed when we are blocked in
1548          * nbd_co_receive_request()
1549          */
1550         goto done;
1551     }
1552 
1553     switch (request.type) {
1554     case NBD_CMD_READ:
1555         /* XXX: NBD Protocol only documents use of FUA with WRITE */
1556         if (request.flags & NBD_CMD_FLAG_FUA) {
1557             ret = blk_co_flush(exp->blk);
1558             if (ret < 0) {
1559                 error_setg_errno(&local_err, -ret, "flush failed");
1560                 break;
1561             }
1562         }
1563 
1564         if (client->structured_reply && !(request.flags & NBD_CMD_FLAG_DF) &&
1565             request.len) {
1566             ret = nbd_co_send_sparse_read(req->client, request.handle,
1567                                           request.from, req->data, request.len,
1568                                           &local_err);
1569             if (ret < 0) {
1570                 goto reply;
1571             }
1572             goto done;
1573         }
1574 
1575         ret = blk_pread(exp->blk, request.from + exp->dev_offset,
1576                         req->data, request.len);
1577         if (ret < 0) {
1578             error_setg_errno(&local_err, -ret, "reading from file failed");
1579             break;
1580         }
1581 
1582         reply_data_len = request.len;
1583 
1584         break;
1585     case NBD_CMD_WRITE:
1586         flags = 0;
1587         if (request.flags & NBD_CMD_FLAG_FUA) {
1588             flags |= BDRV_REQ_FUA;
1589         }
1590         ret = blk_pwrite(exp->blk, request.from + exp->dev_offset,
1591                          req->data, request.len, flags);
1592         if (ret < 0) {
1593             error_setg_errno(&local_err, -ret, "writing to file failed");
1594         }
1595 
1596         break;
1597     case NBD_CMD_WRITE_ZEROES:
1598         flags = 0;
1599         if (request.flags & NBD_CMD_FLAG_FUA) {
1600             flags |= BDRV_REQ_FUA;
1601         }
1602         if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) {
1603             flags |= BDRV_REQ_MAY_UNMAP;
1604         }
1605         ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset,
1606                                 request.len, flags);
1607         if (ret < 0) {
1608             error_setg_errno(&local_err, -ret, "writing to file failed");
1609         }
1610 
1611         break;
1612     case NBD_CMD_DISC:
1613         /* unreachable, thanks to special case in nbd_co_receive_request() */
1614         abort();
1615 
1616     case NBD_CMD_FLUSH:
1617         ret = blk_co_flush(exp->blk);
1618         if (ret < 0) {
1619             error_setg_errno(&local_err, -ret, "flush failed");
1620         }
1621 
1622         break;
1623     case NBD_CMD_TRIM:
1624         ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset,
1625                               request.len);
1626         if (ret < 0) {
1627             error_setg_errno(&local_err, -ret, "discard failed");
1628         }
1629 
1630         break;
1631     default:
1632         error_setg(&local_err, "invalid request type (%" PRIu32 ") received",
1633                    request.type);
1634         ret = -EINVAL;
1635     }
1636 
1637 reply:
1638     if (local_err) {
1639         /* If we get here, local_err was not a fatal error, and should be sent
1640          * to the client. */
1641         assert(ret < 0);
1642         msg = g_strdup(error_get_pretty(local_err));
1643         error_report_err(local_err);
1644         local_err = NULL;
1645     }
1646 
1647     if (client->structured_reply &&
1648         (ret < 0 || request.type == NBD_CMD_READ)) {
1649         if (ret < 0) {
1650             ret = nbd_co_send_structured_error(req->client, request.handle,
1651                                                -ret, msg, &local_err);
1652         } else if (reply_data_len) {
1653             ret = nbd_co_send_structured_read(req->client, request.handle,
1654                                               request.from, req->data,
1655                                               reply_data_len, true,
1656                                               &local_err);
1657         } else {
1658             ret = nbd_co_send_structured_done(req->client, request.handle,
1659                                               &local_err);
1660         }
1661     } else {
1662         ret = nbd_co_send_simple_reply(req->client, request.handle,
1663                                        ret < 0 ? -ret : 0,
1664                                        req->data, reply_data_len, &local_err);
1665     }
1666     g_free(msg);
1667     if (ret < 0) {
1668         error_prepend(&local_err, "Failed to send reply: ");
1669         goto disconnect;
1670     }
1671 
1672     /* We must disconnect after NBD_CMD_WRITE if we did not
1673      * read the payload.
1674      */
1675     if (!req->complete) {
1676         error_setg(&local_err, "Request handling failed in intermediate state");
1677         goto disconnect;
1678     }
1679 
1680 done:
1681     nbd_request_put(req);
1682     nbd_client_put(client);
1683     return;
1684 
1685 disconnect:
1686     if (local_err) {
1687         error_reportf_err(local_err, "Disconnect client, due to: ");
1688     }
1689     nbd_request_put(req);
1690     client_close(client, true);
1691     nbd_client_put(client);
1692 }
1693 
1694 static void nbd_client_receive_next_request(NBDClient *client)
1695 {
1696     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
1697         nbd_client_get(client);
1698         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
1699         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
1700     }
1701 }
1702 
1703 static coroutine_fn void nbd_co_client_start(void *opaque)
1704 {
1705     NBDClient *client = opaque;
1706     NBDExport *exp = client->exp;
1707     Error *local_err = NULL;
1708 
1709     if (exp) {
1710         nbd_export_get(exp);
1711         QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1712     }
1713     qemu_co_mutex_init(&client->send_lock);
1714 
1715     if (nbd_negotiate(client, &local_err)) {
1716         if (local_err) {
1717             error_report_err(local_err);
1718         }
1719         client_close(client, false);
1720         return;
1721     }
1722 
1723     nbd_client_receive_next_request(client);
1724 }
1725 
1726 /*
1727  * Create a new client listener on the given export @exp, using the
1728  * given channel @sioc.  Begin servicing it in a coroutine.  When the
1729  * connection closes, call @close_fn with an indication of whether the
1730  * client completed negotiation.
1731  */
1732 void nbd_client_new(NBDExport *exp,
1733                     QIOChannelSocket *sioc,
1734                     QCryptoTLSCreds *tlscreds,
1735                     const char *tlsaclname,
1736                     void (*close_fn)(NBDClient *, bool))
1737 {
1738     NBDClient *client;
1739     Coroutine *co;
1740 
1741     client = g_new0(NBDClient, 1);
1742     client->refcount = 1;
1743     client->exp = exp;
1744     client->tlscreds = tlscreds;
1745     if (tlscreds) {
1746         object_ref(OBJECT(client->tlscreds));
1747     }
1748     client->tlsaclname = g_strdup(tlsaclname);
1749     client->sioc = sioc;
1750     object_ref(OBJECT(client->sioc));
1751     client->ioc = QIO_CHANNEL(sioc);
1752     object_ref(OBJECT(client->ioc));
1753     client->close_fn = close_fn;
1754 
1755     co = qemu_coroutine_create(nbd_co_client_start, client);
1756     qemu_coroutine_enter(co);
1757 }
1758