xref: /openbmc/qemu/nbd/server.c (revision 4e6b1384)
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 "qemu/queue.h"
23 #include "trace.h"
24 #include "nbd-internal.h"
25 #include "qemu/units.h"
26 
27 #define NBD_META_ID_BASE_ALLOCATION 0
28 #define NBD_META_ID_DIRTY_BITMAP 1
29 
30 /*
31  * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
32  * constant. If an increase is needed, note that the NBD protocol
33  * recommends no larger than 32 mb, so that the client won't consider
34  * the reply as a denial of service attack.
35  */
36 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
37 
38 static int system_errno_to_nbd_errno(int err)
39 {
40     switch (err) {
41     case 0:
42         return NBD_SUCCESS;
43     case EPERM:
44     case EROFS:
45         return NBD_EPERM;
46     case EIO:
47         return NBD_EIO;
48     case ENOMEM:
49         return NBD_ENOMEM;
50 #ifdef EDQUOT
51     case EDQUOT:
52 #endif
53     case EFBIG:
54     case ENOSPC:
55         return NBD_ENOSPC;
56     case EOVERFLOW:
57         return NBD_EOVERFLOW;
58     case ENOTSUP:
59 #if ENOTSUP != EOPNOTSUPP
60     case EOPNOTSUPP:
61 #endif
62         return NBD_ENOTSUP;
63     case ESHUTDOWN:
64         return NBD_ESHUTDOWN;
65     case EINVAL:
66     default:
67         return NBD_EINVAL;
68     }
69 }
70 
71 /* Definitions for opaque data types */
72 
73 typedef struct NBDRequestData NBDRequestData;
74 
75 struct NBDRequestData {
76     QSIMPLEQ_ENTRY(NBDRequestData) entry;
77     NBDClient *client;
78     uint8_t *data;
79     bool complete;
80 };
81 
82 struct NBDExport {
83     int refcount;
84     void (*close)(NBDExport *exp);
85 
86     BlockBackend *blk;
87     char *name;
88     char *description;
89     uint64_t dev_offset;
90     uint64_t size;
91     uint16_t nbdflags;
92     QTAILQ_HEAD(, NBDClient) clients;
93     QTAILQ_ENTRY(NBDExport) next;
94 
95     AioContext *ctx;
96 
97     BlockBackend *eject_notifier_blk;
98     Notifier eject_notifier;
99 
100     BdrvDirtyBitmap *export_bitmap;
101     char *export_bitmap_context;
102 };
103 
104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
105 
106 /* NBDExportMetaContexts represents a list of contexts to be exported,
107  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
108  * NBD_OPT_LIST_META_CONTEXT. */
109 typedef struct NBDExportMetaContexts {
110     NBDExport *exp;
111     bool valid; /* means that negotiation of the option finished without
112                    errors */
113     bool base_allocation; /* export base:allocation context (block status) */
114     bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
115 } NBDExportMetaContexts;
116 
117 struct NBDClient {
118     int refcount;
119     void (*close_fn)(NBDClient *client, bool negotiated);
120 
121     NBDExport *exp;
122     QCryptoTLSCreds *tlscreds;
123     char *tlsauthz;
124     QIOChannelSocket *sioc; /* The underlying data channel */
125     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
126 
127     Coroutine *recv_coroutine;
128 
129     CoMutex send_lock;
130     Coroutine *send_coroutine;
131 
132     QTAILQ_ENTRY(NBDClient) next;
133     int nb_requests;
134     bool closing;
135 
136     uint32_t check_align; /* If non-zero, check for aligned client requests */
137 
138     bool structured_reply;
139     NBDExportMetaContexts export_meta;
140 
141     uint32_t opt; /* Current option being negotiated */
142     uint32_t optlen; /* remaining length of data in ioc for the option being
143                         negotiated now */
144 };
145 
146 static void nbd_client_receive_next_request(NBDClient *client);
147 
148 /* Basic flow for negotiation
149 
150    Server         Client
151    Negotiate
152 
153    or
154 
155    Server         Client
156    Negotiate #1
157                   Option
158    Negotiate #2
159 
160    ----
161 
162    followed by
163 
164    Server         Client
165                   Request
166    Response
167                   Request
168    Response
169                   ...
170    ...
171                   Request (type == 2)
172 
173 */
174 
175 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
176                                      uint32_t type, uint32_t length)
177 {
178     stq_be_p(&rep->magic, NBD_REP_MAGIC);
179     stl_be_p(&rep->option, option);
180     stl_be_p(&rep->type, type);
181     stl_be_p(&rep->length, length);
182 }
183 
184 /* Send a reply header, including length, but no payload.
185  * Return -errno on error, 0 on success. */
186 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
187                                       uint32_t len, Error **errp)
188 {
189     NBDOptionReply rep;
190 
191     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
192                                      type, nbd_rep_lookup(type), len);
193 
194     assert(len < NBD_MAX_BUFFER_SIZE);
195 
196     set_be_option_rep(&rep, client->opt, type, len);
197     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
198 }
199 
200 /* Send a reply header with default 0 length.
201  * Return -errno on error, 0 on success. */
202 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
203                                   Error **errp)
204 {
205     return nbd_negotiate_send_rep_len(client, type, 0, errp);
206 }
207 
208 /* Send an error reply.
209  * Return -errno on error, 0 on success. */
210 static int GCC_FMT_ATTR(4, 0)
211 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
212                             Error **errp, const char *fmt, va_list va)
213 {
214     g_autofree char *msg = NULL;
215     int ret;
216     size_t len;
217 
218     msg = g_strdup_vprintf(fmt, va);
219     len = strlen(msg);
220     assert(len < 4096);
221     trace_nbd_negotiate_send_rep_err(msg);
222     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
223     if (ret < 0) {
224         return ret;
225     }
226     if (nbd_write(client->ioc, msg, len, errp) < 0) {
227         error_prepend(errp, "write failed (error message): ");
228         return -EIO;
229     }
230 
231     return 0;
232 }
233 
234 /* Send an error reply.
235  * Return -errno on error, 0 on success. */
236 static int GCC_FMT_ATTR(4, 5)
237 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
238                            Error **errp, const char *fmt, ...)
239 {
240     va_list va;
241     int ret;
242 
243     va_start(va, fmt);
244     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
245     va_end(va);
246     return ret;
247 }
248 
249 /* Drop remainder of the current option, and send a reply with the
250  * given error type and message. Return -errno on read or write
251  * failure; or 0 if connection is still live. */
252 static int GCC_FMT_ATTR(4, 0)
253 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
254               const char *fmt, va_list va)
255 {
256     int ret = nbd_drop(client->ioc, client->optlen, errp);
257 
258     client->optlen = 0;
259     if (!ret) {
260         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
261     }
262     return ret;
263 }
264 
265 static int GCC_FMT_ATTR(4, 5)
266 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
267              const char *fmt, ...)
268 {
269     int ret;
270     va_list va;
271 
272     va_start(va, fmt);
273     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
274     va_end(va);
275 
276     return ret;
277 }
278 
279 static int GCC_FMT_ATTR(3, 4)
280 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
281 {
282     int ret;
283     va_list va;
284 
285     va_start(va, fmt);
286     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
287     va_end(va);
288 
289     return ret;
290 }
291 
292 /* Read size bytes from the unparsed payload of the current option.
293  * Return -errno on I/O error, 0 if option was completely handled by
294  * sending a reply about inconsistent lengths, or 1 on success. */
295 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
296                         Error **errp)
297 {
298     if (size > client->optlen) {
299         return nbd_opt_invalid(client, errp,
300                                "Inconsistent lengths in option %s",
301                                nbd_opt_lookup(client->opt));
302     }
303     client->optlen -= size;
304     return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
305 }
306 
307 /* Drop size bytes from the unparsed payload of the current option.
308  * Return -errno on I/O error, 0 if option was completely handled by
309  * sending a reply about inconsistent lengths, or 1 on success. */
310 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
311 {
312     if (size > client->optlen) {
313         return nbd_opt_invalid(client, errp,
314                                "Inconsistent lengths in option %s",
315                                nbd_opt_lookup(client->opt));
316     }
317     client->optlen -= size;
318     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
319 }
320 
321 /* nbd_opt_read_name
322  *
323  * Read a string with the format:
324  *   uint32_t len     (<= NBD_MAX_STRING_SIZE)
325  *   len bytes string (not 0-terminated)
326  *
327  * On success, @name will be allocated.
328  * If @length is non-null, it will be set to the actual string length.
329  *
330  * Return -errno on I/O error, 0 if option was completely handled by
331  * sending a reply about inconsistent lengths, or 1 on success.
332  */
333 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
334                              Error **errp)
335 {
336     int ret;
337     uint32_t len;
338     g_autofree char *local_name = NULL;
339 
340     *name = NULL;
341     ret = nbd_opt_read(client, &len, sizeof(len), errp);
342     if (ret <= 0) {
343         return ret;
344     }
345     len = cpu_to_be32(len);
346 
347     if (len > NBD_MAX_STRING_SIZE) {
348         return nbd_opt_invalid(client, errp,
349                                "Invalid name length: %" PRIu32, len);
350     }
351 
352     local_name = g_malloc(len + 1);
353     ret = nbd_opt_read(client, local_name, len, errp);
354     if (ret <= 0) {
355         return ret;
356     }
357     local_name[len] = '\0';
358 
359     if (length) {
360         *length = len;
361     }
362     *name = g_steal_pointer(&local_name);
363 
364     return 1;
365 }
366 
367 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
368  * Return -errno on error, 0 on success. */
369 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
370                                        Error **errp)
371 {
372     size_t name_len, desc_len;
373     uint32_t len;
374     const char *name = exp->name ? exp->name : "";
375     const char *desc = exp->description ? exp->description : "";
376     QIOChannel *ioc = client->ioc;
377     int ret;
378 
379     trace_nbd_negotiate_send_rep_list(name, desc);
380     name_len = strlen(name);
381     desc_len = strlen(desc);
382     assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
383     len = name_len + desc_len + sizeof(len);
384     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
385     if (ret < 0) {
386         return ret;
387     }
388 
389     len = cpu_to_be32(name_len);
390     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
391         error_prepend(errp, "write failed (name length): ");
392         return -EINVAL;
393     }
394 
395     if (nbd_write(ioc, name, name_len, errp) < 0) {
396         error_prepend(errp, "write failed (name buffer): ");
397         return -EINVAL;
398     }
399 
400     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
401         error_prepend(errp, "write failed (description buffer): ");
402         return -EINVAL;
403     }
404 
405     return 0;
406 }
407 
408 /* Process the NBD_OPT_LIST command, with a potential series of replies.
409  * Return -errno on error, 0 on success. */
410 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
411 {
412     NBDExport *exp;
413     assert(client->opt == NBD_OPT_LIST);
414 
415     /* For each export, send a NBD_REP_SERVER reply. */
416     QTAILQ_FOREACH(exp, &exports, next) {
417         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
418             return -EINVAL;
419         }
420     }
421     /* Finish with a NBD_REP_ACK. */
422     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
423 }
424 
425 static void nbd_check_meta_export(NBDClient *client)
426 {
427     client->export_meta.valid &= client->exp == client->export_meta.exp;
428 }
429 
430 /* Send a reply to NBD_OPT_EXPORT_NAME.
431  * Return -errno on error, 0 on success. */
432 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
433                                             Error **errp)
434 {
435     g_autofree char *name = NULL;
436     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
437     size_t len;
438     int ret;
439     uint16_t myflags;
440 
441     /* Client sends:
442         [20 ..  xx]   export name (length bytes)
443        Server replies:
444         [ 0 ..   7]   size
445         [ 8 ..   9]   export flags
446         [10 .. 133]   reserved     (0) [unless no_zeroes]
447      */
448     trace_nbd_negotiate_handle_export_name();
449     if (client->optlen > NBD_MAX_STRING_SIZE) {
450         error_setg(errp, "Bad length received");
451         return -EINVAL;
452     }
453     name = g_malloc(client->optlen + 1);
454     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
455         return -EIO;
456     }
457     name[client->optlen] = '\0';
458     client->optlen = 0;
459 
460     trace_nbd_negotiate_handle_export_name_request(name);
461 
462     client->exp = nbd_export_find(name);
463     if (!client->exp) {
464         error_setg(errp, "export not found");
465         return -EINVAL;
466     }
467 
468     myflags = client->exp->nbdflags;
469     if (client->structured_reply) {
470         myflags |= NBD_FLAG_SEND_DF;
471     }
472     trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
473     stq_be_p(buf, client->exp->size);
474     stw_be_p(buf + 8, myflags);
475     len = no_zeroes ? 10 : sizeof(buf);
476     ret = nbd_write(client->ioc, buf, len, errp);
477     if (ret < 0) {
478         error_prepend(errp, "write failed: ");
479         return ret;
480     }
481 
482     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
483     nbd_export_get(client->exp);
484     nbd_check_meta_export(client);
485 
486     return 0;
487 }
488 
489 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
490  * The buffer does NOT include the info type prefix.
491  * Return -errno on error, 0 if ready to send more. */
492 static int nbd_negotiate_send_info(NBDClient *client,
493                                    uint16_t info, uint32_t length, void *buf,
494                                    Error **errp)
495 {
496     int rc;
497 
498     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
499     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
500                                     sizeof(info) + length, errp);
501     if (rc < 0) {
502         return rc;
503     }
504     info = cpu_to_be16(info);
505     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
506         return -EIO;
507     }
508     if (nbd_write(client->ioc, buf, length, errp) < 0) {
509         return -EIO;
510     }
511     return 0;
512 }
513 
514 /* nbd_reject_length: Handle any unexpected payload.
515  * @fatal requests that we quit talking to the client, even if we are able
516  * to successfully send an error reply.
517  * Return:
518  * -errno  transmission error occurred or @fatal was requested, errp is set
519  * 0       error message successfully sent to client, errp is not set
520  */
521 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
522 {
523     int ret;
524 
525     assert(client->optlen);
526     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
527                           nbd_opt_lookup(client->opt));
528     if (fatal && !ret) {
529         error_setg(errp, "option '%s' has unexpected length",
530                    nbd_opt_lookup(client->opt));
531         return -EINVAL;
532     }
533     return ret;
534 }
535 
536 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
537  * Return -errno on error, 0 if ready for next option, and 1 to move
538  * into transmission phase.  */
539 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
540 {
541     int rc;
542     g_autofree char *name = NULL;
543     NBDExport *exp;
544     uint16_t requests;
545     uint16_t request;
546     uint32_t namelen;
547     bool sendname = false;
548     bool blocksize = false;
549     uint32_t sizes[3];
550     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
551     uint32_t check_align = 0;
552     uint16_t myflags;
553 
554     /* Client sends:
555         4 bytes: L, name length (can be 0)
556         L bytes: export name
557         2 bytes: N, number of requests (can be 0)
558         N * 2 bytes: N requests
559     */
560     rc = nbd_opt_read_name(client, &name, &namelen, errp);
561     if (rc <= 0) {
562         return rc;
563     }
564     trace_nbd_negotiate_handle_export_name_request(name);
565 
566     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
567     if (rc <= 0) {
568         return rc;
569     }
570     requests = be16_to_cpu(requests);
571     trace_nbd_negotiate_handle_info_requests(requests);
572     while (requests--) {
573         rc = nbd_opt_read(client, &request, sizeof(request), errp);
574         if (rc <= 0) {
575             return rc;
576         }
577         request = be16_to_cpu(request);
578         trace_nbd_negotiate_handle_info_request(request,
579                                                 nbd_info_lookup(request));
580         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
581          * everything else is either a request we don't know or
582          * something we send regardless of request */
583         switch (request) {
584         case NBD_INFO_NAME:
585             sendname = true;
586             break;
587         case NBD_INFO_BLOCK_SIZE:
588             blocksize = true;
589             break;
590         }
591     }
592     if (client->optlen) {
593         return nbd_reject_length(client, false, errp);
594     }
595 
596     exp = nbd_export_find(name);
597     if (!exp) {
598         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
599                                           errp, "export '%s' not present",
600                                           name);
601     }
602 
603     /* Don't bother sending NBD_INFO_NAME unless client requested it */
604     if (sendname) {
605         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
606                                      errp);
607         if (rc < 0) {
608             return rc;
609         }
610     }
611 
612     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
613      * client request */
614     if (exp->description) {
615         size_t len = strlen(exp->description);
616 
617         assert(len <= NBD_MAX_STRING_SIZE);
618         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
619                                      len, exp->description, errp);
620         if (rc < 0) {
621             return rc;
622         }
623     }
624 
625     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
626      * according to whether the client requested it, and according to
627      * whether this is OPT_INFO or OPT_GO. */
628     /* minimum - 1 for back-compat, or actual if client will obey it. */
629     if (client->opt == NBD_OPT_INFO || blocksize) {
630         check_align = sizes[0] = blk_get_request_alignment(exp->blk);
631     } else {
632         sizes[0] = 1;
633     }
634     assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
635     /* preferred - Hard-code to 4096 for now.
636      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
637     sizes[1] = MAX(4096, sizes[0]);
638     /* maximum - At most 32M, but smaller as appropriate. */
639     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
640     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
641     sizes[0] = cpu_to_be32(sizes[0]);
642     sizes[1] = cpu_to_be32(sizes[1]);
643     sizes[2] = cpu_to_be32(sizes[2]);
644     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
645                                  sizeof(sizes), sizes, errp);
646     if (rc < 0) {
647         return rc;
648     }
649 
650     /* Send NBD_INFO_EXPORT always */
651     myflags = exp->nbdflags;
652     if (client->structured_reply) {
653         myflags |= NBD_FLAG_SEND_DF;
654     }
655     trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
656     stq_be_p(buf, exp->size);
657     stw_be_p(buf + 8, myflags);
658     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
659                                  sizeof(buf), buf, errp);
660     if (rc < 0) {
661         return rc;
662     }
663 
664     /*
665      * If the client is just asking for NBD_OPT_INFO, but forgot to
666      * request block sizes in a situation that would impact
667      * performance, then return an error. But for NBD_OPT_GO, we
668      * tolerate all clients, regardless of alignments.
669      */
670     if (client->opt == NBD_OPT_INFO && !blocksize &&
671         blk_get_request_alignment(exp->blk) > 1) {
672         return nbd_negotiate_send_rep_err(client,
673                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
674                                           errp,
675                                           "request NBD_INFO_BLOCK_SIZE to "
676                                           "use this export");
677     }
678 
679     /* Final reply */
680     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
681     if (rc < 0) {
682         return rc;
683     }
684 
685     if (client->opt == NBD_OPT_GO) {
686         client->exp = exp;
687         client->check_align = check_align;
688         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
689         nbd_export_get(client->exp);
690         nbd_check_meta_export(client);
691         rc = 1;
692     }
693     return rc;
694 }
695 
696 
697 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
698  * new channel for all further (now-encrypted) communication. */
699 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
700                                                  Error **errp)
701 {
702     QIOChannel *ioc;
703     QIOChannelTLS *tioc;
704     struct NBDTLSHandshakeData data = { 0 };
705 
706     assert(client->opt == NBD_OPT_STARTTLS);
707 
708     trace_nbd_negotiate_handle_starttls();
709     ioc = client->ioc;
710 
711     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
712         return NULL;
713     }
714 
715     tioc = qio_channel_tls_new_server(ioc,
716                                       client->tlscreds,
717                                       client->tlsauthz,
718                                       errp);
719     if (!tioc) {
720         return NULL;
721     }
722 
723     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
724     trace_nbd_negotiate_handle_starttls_handshake();
725     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
726     qio_channel_tls_handshake(tioc,
727                               nbd_tls_handshake,
728                               &data,
729                               NULL,
730                               NULL);
731 
732     if (!data.complete) {
733         g_main_loop_run(data.loop);
734     }
735     g_main_loop_unref(data.loop);
736     if (data.error) {
737         object_unref(OBJECT(tioc));
738         error_propagate(errp, data.error);
739         return NULL;
740     }
741 
742     return QIO_CHANNEL(tioc);
743 }
744 
745 /* nbd_negotiate_send_meta_context
746  *
747  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
748  *
749  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
750  */
751 static int nbd_negotiate_send_meta_context(NBDClient *client,
752                                            const char *context,
753                                            uint32_t context_id,
754                                            Error **errp)
755 {
756     NBDOptionReplyMetaContext opt;
757     struct iovec iov[] = {
758         {.iov_base = &opt, .iov_len = sizeof(opt)},
759         {.iov_base = (void *)context, .iov_len = strlen(context)}
760     };
761 
762     assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
763     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
764         context_id = 0;
765     }
766 
767     trace_nbd_negotiate_meta_query_reply(context, context_id);
768     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
769                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
770     stl_be_p(&opt.context_id, context_id);
771 
772     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
773 }
774 
775 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
776  * @match is never set to false.
777  *
778  * Return -errno on I/O error, 0 if option was completely handled by
779  * sending a reply about inconsistent lengths, or 1 on success.
780  *
781  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
782  * It only means that there are no errors.
783  */
784 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
785                             Error **errp)
786 {
787     int ret;
788     char *query;
789     size_t len = strlen(pattern);
790 
791     assert(len);
792 
793     query = g_malloc(len);
794     ret = nbd_opt_read(client, query, len, errp);
795     if (ret <= 0) {
796         g_free(query);
797         return ret;
798     }
799 
800     if (strncmp(query, pattern, len) == 0) {
801         trace_nbd_negotiate_meta_query_parse(pattern);
802         *match = true;
803     } else {
804         trace_nbd_negotiate_meta_query_skip("pattern not matched");
805     }
806     g_free(query);
807 
808     return 1;
809 }
810 
811 /*
812  * Read @len bytes, and set @match to true if they match @pattern, or if @len
813  * is 0 and the client is performing _LIST_. @match is never set to false.
814  *
815  * Return -errno on I/O error, 0 if option was completely handled by
816  * sending a reply about inconsistent lengths, or 1 on success.
817  *
818  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
819  * It only means that there are no errors.
820  */
821 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
822                                      uint32_t len, bool *match, Error **errp)
823 {
824     if (len == 0) {
825         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
826             *match = true;
827         }
828         trace_nbd_negotiate_meta_query_parse("empty");
829         return 1;
830     }
831 
832     if (len != strlen(pattern)) {
833         trace_nbd_negotiate_meta_query_skip("different lengths");
834         return nbd_opt_skip(client, len, errp);
835     }
836 
837     return nbd_meta_pattern(client, pattern, match, errp);
838 }
839 
840 /* nbd_meta_base_query
841  *
842  * Handle queries to 'base' namespace. For now, only the base:allocation
843  * context is available.  'len' is the amount of text remaining to be read from
844  * the current name, after the 'base:' portion has been stripped.
845  *
846  * Return -errno on I/O error, 0 if option was completely handled by
847  * sending a reply about inconsistent lengths, or 1 on success.
848  */
849 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
850                                uint32_t len, Error **errp)
851 {
852     return nbd_meta_empty_or_pattern(client, "allocation", len,
853                                      &meta->base_allocation, errp);
854 }
855 
856 /* nbd_meta_bitmap_query
857  *
858  * Handle query to 'qemu:' namespace.
859  * @len is the amount of text remaining to be read from the current name, after
860  * the 'qemu:' portion has been stripped.
861  *
862  * Return -errno on I/O error, 0 if option was completely handled by
863  * sending a reply about inconsistent lengths, or 1 on success. */
864 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
865                                uint32_t len, Error **errp)
866 {
867     bool dirty_bitmap = false;
868     size_t dirty_bitmap_len = strlen("dirty-bitmap:");
869     int ret;
870 
871     if (!meta->exp->export_bitmap) {
872         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
873         return nbd_opt_skip(client, len, errp);
874     }
875 
876     if (len == 0) {
877         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
878             meta->bitmap = true;
879         }
880         trace_nbd_negotiate_meta_query_parse("empty");
881         return 1;
882     }
883 
884     if (len < dirty_bitmap_len) {
885         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
886         return nbd_opt_skip(client, len, errp);
887     }
888 
889     len -= dirty_bitmap_len;
890     ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
891     if (ret <= 0) {
892         return ret;
893     }
894     if (!dirty_bitmap) {
895         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
896         return nbd_opt_skip(client, len, errp);
897     }
898 
899     trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
900 
901     return nbd_meta_empty_or_pattern(
902             client, meta->exp->export_bitmap_context +
903             strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
904 }
905 
906 /* nbd_negotiate_meta_query
907  *
908  * Parse namespace name and call corresponding function to parse body of the
909  * query.
910  *
911  * The only supported namespaces are 'base' and 'qemu'.
912  *
913  * The function aims not wasting time and memory to read long unknown namespace
914  * names.
915  *
916  * Return -errno on I/O error, 0 if option was completely handled by
917  * sending a reply about inconsistent lengths, or 1 on success. */
918 static int nbd_negotiate_meta_query(NBDClient *client,
919                                     NBDExportMetaContexts *meta, Error **errp)
920 {
921     /*
922      * Both 'qemu' and 'base' namespaces have length = 5 including a
923      * colon. If another length namespace is later introduced, this
924      * should certainly be refactored.
925      */
926     int ret;
927     size_t ns_len = 5;
928     char ns[5];
929     uint32_t len;
930 
931     ret = nbd_opt_read(client, &len, sizeof(len), errp);
932     if (ret <= 0) {
933         return ret;
934     }
935     len = cpu_to_be32(len);
936 
937     if (len > NBD_MAX_STRING_SIZE) {
938         trace_nbd_negotiate_meta_query_skip("length too long");
939         return nbd_opt_skip(client, len, errp);
940     }
941     if (len < ns_len) {
942         trace_nbd_negotiate_meta_query_skip("length too short");
943         return nbd_opt_skip(client, len, errp);
944     }
945 
946     len -= ns_len;
947     ret = nbd_opt_read(client, ns, ns_len, errp);
948     if (ret <= 0) {
949         return ret;
950     }
951 
952     if (!strncmp(ns, "base:", ns_len)) {
953         trace_nbd_negotiate_meta_query_parse("base:");
954         return nbd_meta_base_query(client, meta, len, errp);
955     } else if (!strncmp(ns, "qemu:", ns_len)) {
956         trace_nbd_negotiate_meta_query_parse("qemu:");
957         return nbd_meta_qemu_query(client, meta, len, errp);
958     }
959 
960     trace_nbd_negotiate_meta_query_skip("unknown namespace");
961     return nbd_opt_skip(client, len, errp);
962 }
963 
964 /* nbd_negotiate_meta_queries
965  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
966  *
967  * Return -errno on I/O error, or 0 if option was completely handled. */
968 static int nbd_negotiate_meta_queries(NBDClient *client,
969                                       NBDExportMetaContexts *meta, Error **errp)
970 {
971     int ret;
972     g_autofree char *export_name = NULL;
973     NBDExportMetaContexts local_meta;
974     uint32_t nb_queries;
975     int i;
976 
977     if (!client->structured_reply) {
978         return nbd_opt_invalid(client, errp,
979                                "request option '%s' when structured reply "
980                                "is not negotiated",
981                                nbd_opt_lookup(client->opt));
982     }
983 
984     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
985         /* Only change the caller's meta on SET. */
986         meta = &local_meta;
987     }
988 
989     memset(meta, 0, sizeof(*meta));
990 
991     ret = nbd_opt_read_name(client, &export_name, NULL, errp);
992     if (ret <= 0) {
993         return ret;
994     }
995 
996     meta->exp = nbd_export_find(export_name);
997     if (meta->exp == NULL) {
998         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
999                             "export '%s' not present", export_name);
1000     }
1001 
1002     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
1003     if (ret <= 0) {
1004         return ret;
1005     }
1006     nb_queries = cpu_to_be32(nb_queries);
1007     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1008                                      export_name, nb_queries);
1009 
1010     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1011         /* enable all known contexts */
1012         meta->base_allocation = true;
1013         meta->bitmap = !!meta->exp->export_bitmap;
1014     } else {
1015         for (i = 0; i < nb_queries; ++i) {
1016             ret = nbd_negotiate_meta_query(client, meta, errp);
1017             if (ret <= 0) {
1018                 return ret;
1019             }
1020         }
1021     }
1022 
1023     if (meta->base_allocation) {
1024         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1025                                               NBD_META_ID_BASE_ALLOCATION,
1026                                               errp);
1027         if (ret < 0) {
1028             return ret;
1029         }
1030     }
1031 
1032     if (meta->bitmap) {
1033         ret = nbd_negotiate_send_meta_context(client,
1034                                               meta->exp->export_bitmap_context,
1035                                               NBD_META_ID_DIRTY_BITMAP,
1036                                               errp);
1037         if (ret < 0) {
1038             return ret;
1039         }
1040     }
1041 
1042     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1043     if (ret == 0) {
1044         meta->valid = true;
1045     }
1046 
1047     return ret;
1048 }
1049 
1050 /* nbd_negotiate_options
1051  * Process all NBD_OPT_* client option commands, during fixed newstyle
1052  * negotiation.
1053  * Return:
1054  * -errno  on error, errp is set
1055  * 0       on successful negotiation, errp is not set
1056  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1057  *         errp is not set
1058  */
1059 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1060 {
1061     uint32_t flags;
1062     bool fixedNewstyle = false;
1063     bool no_zeroes = false;
1064 
1065     /* Client sends:
1066         [ 0 ..   3]   client flags
1067 
1068        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1069         [ 0 ..   7]   NBD_OPTS_MAGIC
1070         [ 8 ..  11]   NBD option
1071         [12 ..  15]   Data length
1072         ...           Rest of request
1073 
1074         [ 0 ..   7]   NBD_OPTS_MAGIC
1075         [ 8 ..  11]   Second NBD option
1076         [12 ..  15]   Data length
1077         ...           Rest of request
1078     */
1079 
1080     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1081         return -EIO;
1082     }
1083     trace_nbd_negotiate_options_flags(flags);
1084     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1085         fixedNewstyle = true;
1086         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1087     }
1088     if (flags & NBD_FLAG_C_NO_ZEROES) {
1089         no_zeroes = true;
1090         flags &= ~NBD_FLAG_C_NO_ZEROES;
1091     }
1092     if (flags != 0) {
1093         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1094         return -EINVAL;
1095     }
1096 
1097     while (1) {
1098         int ret;
1099         uint32_t option, length;
1100         uint64_t magic;
1101 
1102         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1103             return -EINVAL;
1104         }
1105         trace_nbd_negotiate_options_check_magic(magic);
1106         if (magic != NBD_OPTS_MAGIC) {
1107             error_setg(errp, "Bad magic received");
1108             return -EINVAL;
1109         }
1110 
1111         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1112             return -EINVAL;
1113         }
1114         client->opt = option;
1115 
1116         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1117             return -EINVAL;
1118         }
1119         assert(!client->optlen);
1120         client->optlen = length;
1121 
1122         if (length > NBD_MAX_BUFFER_SIZE) {
1123             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1124                        length, NBD_MAX_BUFFER_SIZE);
1125             return -EINVAL;
1126         }
1127 
1128         trace_nbd_negotiate_options_check_option(option,
1129                                                  nbd_opt_lookup(option));
1130         if (client->tlscreds &&
1131             client->ioc == (QIOChannel *)client->sioc) {
1132             QIOChannel *tioc;
1133             if (!fixedNewstyle) {
1134                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1135                 return -EINVAL;
1136             }
1137             switch (option) {
1138             case NBD_OPT_STARTTLS:
1139                 if (length) {
1140                     /* Unconditionally drop the connection if the client
1141                      * can't start a TLS negotiation correctly */
1142                     return nbd_reject_length(client, true, errp);
1143                 }
1144                 tioc = nbd_negotiate_handle_starttls(client, errp);
1145                 if (!tioc) {
1146                     return -EIO;
1147                 }
1148                 ret = 0;
1149                 object_unref(OBJECT(client->ioc));
1150                 client->ioc = QIO_CHANNEL(tioc);
1151                 break;
1152 
1153             case NBD_OPT_EXPORT_NAME:
1154                 /* No way to return an error to client, so drop connection */
1155                 error_setg(errp, "Option 0x%x not permitted before TLS",
1156                            option);
1157                 return -EINVAL;
1158 
1159             default:
1160                 /* Let the client keep trying, unless they asked to
1161                  * quit. Always try to give an error back to the
1162                  * client; but when replying to OPT_ABORT, be aware
1163                  * that the client may hang up before receiving the
1164                  * error, in which case we are fine ignoring the
1165                  * resulting EPIPE. */
1166                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1167                                    option == NBD_OPT_ABORT ? NULL : errp,
1168                                    "Option 0x%" PRIx32
1169                                    " not permitted before TLS", option);
1170                 if (option == NBD_OPT_ABORT) {
1171                     return 1;
1172                 }
1173                 break;
1174             }
1175         } else if (fixedNewstyle) {
1176             switch (option) {
1177             case NBD_OPT_LIST:
1178                 if (length) {
1179                     ret = nbd_reject_length(client, false, errp);
1180                 } else {
1181                     ret = nbd_negotiate_handle_list(client, errp);
1182                 }
1183                 break;
1184 
1185             case NBD_OPT_ABORT:
1186                 /* NBD spec says we must try to reply before
1187                  * disconnecting, but that we must also tolerate
1188                  * guests that don't wait for our reply. */
1189                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1190                 return 1;
1191 
1192             case NBD_OPT_EXPORT_NAME:
1193                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1194                                                         errp);
1195 
1196             case NBD_OPT_INFO:
1197             case NBD_OPT_GO:
1198                 ret = nbd_negotiate_handle_info(client, errp);
1199                 if (ret == 1) {
1200                     assert(option == NBD_OPT_GO);
1201                     return 0;
1202                 }
1203                 break;
1204 
1205             case NBD_OPT_STARTTLS:
1206                 if (length) {
1207                     ret = nbd_reject_length(client, false, errp);
1208                 } else if (client->tlscreds) {
1209                     ret = nbd_negotiate_send_rep_err(client,
1210                                                      NBD_REP_ERR_INVALID, errp,
1211                                                      "TLS already enabled");
1212                 } else {
1213                     ret = nbd_negotiate_send_rep_err(client,
1214                                                      NBD_REP_ERR_POLICY, errp,
1215                                                      "TLS not configured");
1216                 }
1217                 break;
1218 
1219             case NBD_OPT_STRUCTURED_REPLY:
1220                 if (length) {
1221                     ret = nbd_reject_length(client, false, errp);
1222                 } else if (client->structured_reply) {
1223                     ret = nbd_negotiate_send_rep_err(
1224                         client, NBD_REP_ERR_INVALID, errp,
1225                         "structured reply already negotiated");
1226                 } else {
1227                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1228                     client->structured_reply = true;
1229                 }
1230                 break;
1231 
1232             case NBD_OPT_LIST_META_CONTEXT:
1233             case NBD_OPT_SET_META_CONTEXT:
1234                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1235                                                  errp);
1236                 break;
1237 
1238             default:
1239                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1240                                    "Unsupported option %" PRIu32 " (%s)",
1241                                    option, nbd_opt_lookup(option));
1242                 break;
1243             }
1244         } else {
1245             /*
1246              * If broken new-style we should drop the connection
1247              * for anything except NBD_OPT_EXPORT_NAME
1248              */
1249             switch (option) {
1250             case NBD_OPT_EXPORT_NAME:
1251                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1252                                                         errp);
1253 
1254             default:
1255                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1256                            option, nbd_opt_lookup(option));
1257                 return -EINVAL;
1258             }
1259         }
1260         if (ret < 0) {
1261             return ret;
1262         }
1263     }
1264 }
1265 
1266 /* nbd_negotiate
1267  * Return:
1268  * -errno  on error, errp is set
1269  * 0       on successful negotiation, errp is not set
1270  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1271  *         errp is not set
1272  */
1273 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1274 {
1275     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1276     int ret;
1277 
1278     /* Old style negotiation header, no room for options
1279         [ 0 ..   7]   passwd       ("NBDMAGIC")
1280         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1281         [16 ..  23]   size
1282         [24 ..  27]   export flags (zero-extended)
1283         [28 .. 151]   reserved     (0)
1284 
1285        New style negotiation header, client can send options
1286         [ 0 ..   7]   passwd       ("NBDMAGIC")
1287         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1288         [16 ..  17]   server flags (0)
1289         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1290      */
1291 
1292     qio_channel_set_blocking(client->ioc, false, NULL);
1293 
1294     trace_nbd_negotiate_begin();
1295     memcpy(buf, "NBDMAGIC", 8);
1296 
1297     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1298     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1299 
1300     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1301         error_prepend(errp, "write failed: ");
1302         return -EINVAL;
1303     }
1304     ret = nbd_negotiate_options(client, errp);
1305     if (ret != 0) {
1306         if (ret < 0) {
1307             error_prepend(errp, "option negotiation failed: ");
1308         }
1309         return ret;
1310     }
1311 
1312     /* Attach the channel to the same AioContext as the export */
1313     if (client->exp && client->exp->ctx) {
1314         qio_channel_attach_aio_context(client->ioc, client->exp->ctx);
1315     }
1316 
1317     assert(!client->optlen);
1318     trace_nbd_negotiate_success();
1319 
1320     return 0;
1321 }
1322 
1323 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1324                                Error **errp)
1325 {
1326     uint8_t buf[NBD_REQUEST_SIZE];
1327     uint32_t magic;
1328     int ret;
1329 
1330     ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1331     if (ret < 0) {
1332         return ret;
1333     }
1334 
1335     /* Request
1336        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1337        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1338        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1339        [ 8 .. 15]   handle
1340        [16 .. 23]   from
1341        [24 .. 27]   len
1342      */
1343 
1344     magic = ldl_be_p(buf);
1345     request->flags  = lduw_be_p(buf + 4);
1346     request->type   = lduw_be_p(buf + 6);
1347     request->handle = ldq_be_p(buf + 8);
1348     request->from   = ldq_be_p(buf + 16);
1349     request->len    = ldl_be_p(buf + 24);
1350 
1351     trace_nbd_receive_request(magic, request->flags, request->type,
1352                               request->from, request->len);
1353 
1354     if (magic != NBD_REQUEST_MAGIC) {
1355         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1356         return -EINVAL;
1357     }
1358     return 0;
1359 }
1360 
1361 #define MAX_NBD_REQUESTS 16
1362 
1363 void nbd_client_get(NBDClient *client)
1364 {
1365     client->refcount++;
1366 }
1367 
1368 void nbd_client_put(NBDClient *client)
1369 {
1370     if (--client->refcount == 0) {
1371         /* The last reference should be dropped by client->close,
1372          * which is called by client_close.
1373          */
1374         assert(client->closing);
1375 
1376         qio_channel_detach_aio_context(client->ioc);
1377         object_unref(OBJECT(client->sioc));
1378         object_unref(OBJECT(client->ioc));
1379         if (client->tlscreds) {
1380             object_unref(OBJECT(client->tlscreds));
1381         }
1382         g_free(client->tlsauthz);
1383         if (client->exp) {
1384             QTAILQ_REMOVE(&client->exp->clients, client, next);
1385             nbd_export_put(client->exp);
1386         }
1387         g_free(client);
1388     }
1389 }
1390 
1391 static void client_close(NBDClient *client, bool negotiated)
1392 {
1393     if (client->closing) {
1394         return;
1395     }
1396 
1397     client->closing = true;
1398 
1399     /* Force requests to finish.  They will drop their own references,
1400      * then we'll close the socket and free the NBDClient.
1401      */
1402     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1403                          NULL);
1404 
1405     /* Also tell the client, so that they release their reference.  */
1406     if (client->close_fn) {
1407         client->close_fn(client, negotiated);
1408     }
1409 }
1410 
1411 static NBDRequestData *nbd_request_get(NBDClient *client)
1412 {
1413     NBDRequestData *req;
1414 
1415     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1416     client->nb_requests++;
1417 
1418     req = g_new0(NBDRequestData, 1);
1419     nbd_client_get(client);
1420     req->client = client;
1421     return req;
1422 }
1423 
1424 static void nbd_request_put(NBDRequestData *req)
1425 {
1426     NBDClient *client = req->client;
1427 
1428     if (req->data) {
1429         qemu_vfree(req->data);
1430     }
1431     g_free(req);
1432 
1433     client->nb_requests--;
1434     nbd_client_receive_next_request(client);
1435 
1436     nbd_client_put(client);
1437 }
1438 
1439 static void blk_aio_attached(AioContext *ctx, void *opaque)
1440 {
1441     NBDExport *exp = opaque;
1442     NBDClient *client;
1443 
1444     trace_nbd_blk_aio_attached(exp->name, ctx);
1445 
1446     exp->ctx = ctx;
1447 
1448     QTAILQ_FOREACH(client, &exp->clients, next) {
1449         qio_channel_attach_aio_context(client->ioc, ctx);
1450         if (client->recv_coroutine) {
1451             aio_co_schedule(ctx, client->recv_coroutine);
1452         }
1453         if (client->send_coroutine) {
1454             aio_co_schedule(ctx, client->send_coroutine);
1455         }
1456     }
1457 }
1458 
1459 static void blk_aio_detach(void *opaque)
1460 {
1461     NBDExport *exp = opaque;
1462     NBDClient *client;
1463 
1464     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1465 
1466     QTAILQ_FOREACH(client, &exp->clients, next) {
1467         qio_channel_detach_aio_context(client->ioc);
1468     }
1469 
1470     exp->ctx = NULL;
1471 }
1472 
1473 static void nbd_eject_notifier(Notifier *n, void *data)
1474 {
1475     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1476     AioContext *aio_context;
1477 
1478     aio_context = exp->ctx;
1479     aio_context_acquire(aio_context);
1480     nbd_export_close(exp);
1481     aio_context_release(aio_context);
1482 }
1483 
1484 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1485                           uint64_t size, const char *name, const char *desc,
1486                           const char *bitmap, bool readonly, bool shared,
1487                           void (*close)(NBDExport *), bool writethrough,
1488                           BlockBackend *on_eject_blk, Error **errp)
1489 {
1490     AioContext *ctx;
1491     BlockBackend *blk;
1492     NBDExport *exp = g_new0(NBDExport, 1);
1493     uint64_t perm;
1494     int ret;
1495 
1496     /*
1497      * NBD exports are used for non-shared storage migration.  Make sure
1498      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1499      * access since the export could be available before migration handover.
1500      * ctx was acquired in the caller.
1501      */
1502     assert(name && strlen(name) <= NBD_MAX_STRING_SIZE);
1503     ctx = bdrv_get_aio_context(bs);
1504     bdrv_invalidate_cache(bs, NULL);
1505 
1506     /* Don't allow resize while the NBD server is running, otherwise we don't
1507      * care what happens with the node. */
1508     perm = BLK_PERM_CONSISTENT_READ;
1509     if (!readonly) {
1510         perm |= BLK_PERM_WRITE;
1511     }
1512     blk = blk_new(ctx, perm,
1513                   BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1514                   BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1515     ret = blk_insert_bs(blk, bs, errp);
1516     if (ret < 0) {
1517         goto fail;
1518     }
1519     blk_set_enable_write_cache(blk, !writethrough);
1520     blk_set_allow_aio_context_change(blk, true);
1521 
1522     exp->refcount = 1;
1523     QTAILQ_INIT(&exp->clients);
1524     exp->blk = blk;
1525     assert(dev_offset <= INT64_MAX);
1526     exp->dev_offset = dev_offset;
1527     exp->name = g_strdup(name);
1528     assert(!desc || strlen(desc) <= NBD_MAX_STRING_SIZE);
1529     exp->description = g_strdup(desc);
1530     exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1531                      NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1532     if (readonly) {
1533         exp->nbdflags |= NBD_FLAG_READ_ONLY;
1534         if (shared) {
1535             exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1536         }
1537     } else {
1538         exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1539                           NBD_FLAG_SEND_FAST_ZERO);
1540     }
1541     assert(size <= INT64_MAX - dev_offset);
1542     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1543 
1544     if (bitmap) {
1545         BdrvDirtyBitmap *bm = NULL;
1546 
1547         while (true) {
1548             bm = bdrv_find_dirty_bitmap(bs, bitmap);
1549             if (bm != NULL || bs->backing == NULL) {
1550                 break;
1551             }
1552 
1553             bs = bs->backing->bs;
1554         }
1555 
1556         if (bm == NULL) {
1557             error_setg(errp, "Bitmap '%s' is not found", bitmap);
1558             goto fail;
1559         }
1560 
1561         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1562             goto fail;
1563         }
1564 
1565         if (readonly && bdrv_is_writable(bs) &&
1566             bdrv_dirty_bitmap_enabled(bm)) {
1567             error_setg(errp,
1568                        "Enabled bitmap '%s' incompatible with readonly export",
1569                        bitmap);
1570             goto fail;
1571         }
1572 
1573         bdrv_dirty_bitmap_set_busy(bm, true);
1574         exp->export_bitmap = bm;
1575         assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1576         exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1577                                                      bitmap);
1578         assert(strlen(exp->export_bitmap_context) < NBD_MAX_STRING_SIZE);
1579     }
1580 
1581     exp->close = close;
1582     exp->ctx = ctx;
1583     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1584 
1585     if (on_eject_blk) {
1586         blk_ref(on_eject_blk);
1587         exp->eject_notifier_blk = on_eject_blk;
1588         exp->eject_notifier.notify = nbd_eject_notifier;
1589         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1590     }
1591     QTAILQ_INSERT_TAIL(&exports, exp, next);
1592     nbd_export_get(exp);
1593     return exp;
1594 
1595 fail:
1596     blk_unref(blk);
1597     g_free(exp->name);
1598     g_free(exp->description);
1599     g_free(exp);
1600     return NULL;
1601 }
1602 
1603 NBDExport *nbd_export_find(const char *name)
1604 {
1605     NBDExport *exp;
1606     QTAILQ_FOREACH(exp, &exports, next) {
1607         if (strcmp(name, exp->name) == 0) {
1608             return exp;
1609         }
1610     }
1611 
1612     return NULL;
1613 }
1614 
1615 AioContext *
1616 nbd_export_aio_context(NBDExport *exp)
1617 {
1618     return exp->ctx;
1619 }
1620 
1621 void nbd_export_close(NBDExport *exp)
1622 {
1623     NBDClient *client, *next;
1624 
1625     nbd_export_get(exp);
1626     /*
1627      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1628      * close mode that stops advertising the export to new clients but
1629      * still permits existing clients to run to completion? Because of
1630      * that possibility, nbd_export_close() can be called more than
1631      * once on an export.
1632      */
1633     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1634         client_close(client, true);
1635     }
1636     if (exp->name) {
1637         nbd_export_put(exp);
1638         g_free(exp->name);
1639         exp->name = NULL;
1640         QTAILQ_REMOVE(&exports, exp, next);
1641     }
1642     g_free(exp->description);
1643     exp->description = NULL;
1644     nbd_export_put(exp);
1645 }
1646 
1647 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1648 {
1649     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1650         nbd_export_close(exp);
1651         return;
1652     }
1653 
1654     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1655 
1656     error_setg(errp, "export '%s' still in use", exp->name);
1657     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1658 }
1659 
1660 void nbd_export_get(NBDExport *exp)
1661 {
1662     assert(exp->refcount > 0);
1663     exp->refcount++;
1664 }
1665 
1666 void nbd_export_put(NBDExport *exp)
1667 {
1668     assert(exp->refcount > 0);
1669     if (exp->refcount == 1) {
1670         nbd_export_close(exp);
1671     }
1672 
1673     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1674      * if someone calls nbd_export_put() on named export not through
1675      * nbd_export_set_name() when refcount is 1. So, let's assert that
1676      * it is > 0.
1677      */
1678     assert(exp->refcount > 0);
1679     if (--exp->refcount == 0) {
1680         assert(exp->name == NULL);
1681         assert(exp->description == NULL);
1682 
1683         if (exp->close) {
1684             exp->close(exp);
1685         }
1686 
1687         if (exp->blk) {
1688             if (exp->eject_notifier_blk) {
1689                 notifier_remove(&exp->eject_notifier);
1690                 blk_unref(exp->eject_notifier_blk);
1691             }
1692             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1693                                             blk_aio_detach, exp);
1694             blk_unref(exp->blk);
1695             exp->blk = NULL;
1696         }
1697 
1698         if (exp->export_bitmap) {
1699             bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1700             g_free(exp->export_bitmap_context);
1701         }
1702 
1703         g_free(exp);
1704     }
1705 }
1706 
1707 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1708 {
1709     return exp->blk;
1710 }
1711 
1712 void nbd_export_close_all(void)
1713 {
1714     NBDExport *exp, *next;
1715     AioContext *aio_context;
1716 
1717     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1718         aio_context = exp->ctx;
1719         aio_context_acquire(aio_context);
1720         nbd_export_close(exp);
1721         aio_context_release(aio_context);
1722     }
1723 }
1724 
1725 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1726                                         unsigned niov, Error **errp)
1727 {
1728     int ret;
1729 
1730     g_assert(qemu_in_coroutine());
1731     qemu_co_mutex_lock(&client->send_lock);
1732     client->send_coroutine = qemu_coroutine_self();
1733 
1734     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1735 
1736     client->send_coroutine = NULL;
1737     qemu_co_mutex_unlock(&client->send_lock);
1738 
1739     return ret;
1740 }
1741 
1742 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1743                                        uint64_t handle)
1744 {
1745     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1746     stl_be_p(&reply->error, error);
1747     stq_be_p(&reply->handle, handle);
1748 }
1749 
1750 static int nbd_co_send_simple_reply(NBDClient *client,
1751                                     uint64_t handle,
1752                                     uint32_t error,
1753                                     void *data,
1754                                     size_t len,
1755                                     Error **errp)
1756 {
1757     NBDSimpleReply reply;
1758     int nbd_err = system_errno_to_nbd_errno(error);
1759     struct iovec iov[] = {
1760         {.iov_base = &reply, .iov_len = sizeof(reply)},
1761         {.iov_base = data, .iov_len = len}
1762     };
1763 
1764     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1765                                    len);
1766     set_be_simple_reply(&reply, nbd_err, handle);
1767 
1768     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1769 }
1770 
1771 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1772                                 uint16_t type, uint64_t handle, uint32_t length)
1773 {
1774     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1775     stw_be_p(&chunk->flags, flags);
1776     stw_be_p(&chunk->type, type);
1777     stq_be_p(&chunk->handle, handle);
1778     stl_be_p(&chunk->length, length);
1779 }
1780 
1781 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1782                                                     uint64_t handle,
1783                                                     Error **errp)
1784 {
1785     NBDStructuredReplyChunk chunk;
1786     struct iovec iov[] = {
1787         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1788     };
1789 
1790     trace_nbd_co_send_structured_done(handle);
1791     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1792 
1793     return nbd_co_send_iov(client, iov, 1, errp);
1794 }
1795 
1796 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1797                                                     uint64_t handle,
1798                                                     uint64_t offset,
1799                                                     void *data,
1800                                                     size_t size,
1801                                                     bool final,
1802                                                     Error **errp)
1803 {
1804     NBDStructuredReadData chunk;
1805     struct iovec iov[] = {
1806         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1807         {.iov_base = data, .iov_len = size}
1808     };
1809 
1810     assert(size);
1811     trace_nbd_co_send_structured_read(handle, offset, data, size);
1812     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1813                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1814                  sizeof(chunk) - sizeof(chunk.h) + size);
1815     stq_be_p(&chunk.offset, offset);
1816 
1817     return nbd_co_send_iov(client, iov, 2, errp);
1818 }
1819 
1820 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1821                                                      uint64_t handle,
1822                                                      uint32_t error,
1823                                                      const char *msg,
1824                                                      Error **errp)
1825 {
1826     NBDStructuredError chunk;
1827     int nbd_err = system_errno_to_nbd_errno(error);
1828     struct iovec iov[] = {
1829         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1830         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1831     };
1832 
1833     assert(nbd_err);
1834     trace_nbd_co_send_structured_error(handle, nbd_err,
1835                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1836     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1837                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1838     stl_be_p(&chunk.error, nbd_err);
1839     stw_be_p(&chunk.message_length, iov[1].iov_len);
1840 
1841     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1842 }
1843 
1844 /* Do a sparse read and send the structured reply to the client.
1845  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1846  * reported to the client, at which point this function succeeds.
1847  */
1848 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1849                                                 uint64_t handle,
1850                                                 uint64_t offset,
1851                                                 uint8_t *data,
1852                                                 size_t size,
1853                                                 Error **errp)
1854 {
1855     int ret = 0;
1856     NBDExport *exp = client->exp;
1857     size_t progress = 0;
1858 
1859     while (progress < size) {
1860         int64_t pnum;
1861         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1862                                              offset + progress,
1863                                              size - progress, &pnum, NULL,
1864                                              NULL);
1865         bool final;
1866 
1867         if (status < 0) {
1868             char *msg = g_strdup_printf("unable to check for holes: %s",
1869                                         strerror(-status));
1870 
1871             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1872                                                errp);
1873             g_free(msg);
1874             return ret;
1875         }
1876         assert(pnum && pnum <= size - progress);
1877         final = progress + pnum == size;
1878         if (status & BDRV_BLOCK_ZERO) {
1879             NBDStructuredReadHole chunk;
1880             struct iovec iov[] = {
1881                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1882             };
1883 
1884             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1885                                                    pnum);
1886             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1887                          NBD_REPLY_TYPE_OFFSET_HOLE,
1888                          handle, sizeof(chunk) - sizeof(chunk.h));
1889             stq_be_p(&chunk.offset, offset + progress);
1890             stl_be_p(&chunk.length, pnum);
1891             ret = nbd_co_send_iov(client, iov, 1, errp);
1892         } else {
1893             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1894                             data + progress, pnum);
1895             if (ret < 0) {
1896                 error_setg_errno(errp, -ret, "reading from file failed");
1897                 break;
1898             }
1899             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1900                                               data + progress, pnum, final,
1901                                               errp);
1902         }
1903 
1904         if (ret < 0) {
1905             break;
1906         }
1907         progress += pnum;
1908     }
1909     return ret;
1910 }
1911 
1912 /*
1913  * Populate @extents from block status. Update @bytes to be the actual
1914  * length encoded (which may be smaller than the original), and update
1915  * @nb_extents to the number of extents used.
1916  *
1917  * Returns zero on success and -errno on bdrv_block_status_above failure.
1918  */
1919 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1920                                   uint64_t *bytes, NBDExtent *extents,
1921                                   unsigned int *nb_extents)
1922 {
1923     uint64_t remaining_bytes = *bytes;
1924     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1925     bool first_extent = true;
1926 
1927     assert(*nb_extents);
1928     while (remaining_bytes) {
1929         uint32_t flags;
1930         int64_t num;
1931         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1932                                           &num, NULL, NULL);
1933 
1934         if (ret < 0) {
1935             return ret;
1936         }
1937 
1938         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1939                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1940 
1941         if (first_extent) {
1942             extent->flags = flags;
1943             extent->length = num;
1944             first_extent = false;
1945         } else if (flags == extent->flags) {
1946             /* extend current extent */
1947             extent->length += num;
1948         } else {
1949             if (extent + 1 == extents_end) {
1950                 break;
1951             }
1952 
1953             /* start new extent */
1954             extent++;
1955             extent->flags = flags;
1956             extent->length = num;
1957         }
1958         offset += num;
1959         remaining_bytes -= num;
1960     }
1961 
1962     extents_end = extent + 1;
1963 
1964     for (extent = extents; extent < extents_end; extent++) {
1965         extent->flags = cpu_to_be32(extent->flags);
1966         extent->length = cpu_to_be32(extent->length);
1967     }
1968 
1969     *bytes -= remaining_bytes;
1970     *nb_extents = extents_end - extents;
1971 
1972     return 0;
1973 }
1974 
1975 /* nbd_co_send_extents
1976  *
1977  * @length is only for tracing purposes (and may be smaller or larger
1978  * than the client's original request). @last controls whether
1979  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1980  * big-endian format.
1981  */
1982 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1983                                NBDExtent *extents, unsigned int nb_extents,
1984                                uint64_t length, bool last,
1985                                uint32_t context_id, Error **errp)
1986 {
1987     NBDStructuredMeta chunk;
1988 
1989     struct iovec iov[] = {
1990         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1991         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1992     };
1993 
1994     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1995     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1996                  NBD_REPLY_TYPE_BLOCK_STATUS,
1997                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1998     stl_be_p(&chunk.context_id, context_id);
1999 
2000     return nbd_co_send_iov(client, iov, 2, errp);
2001 }
2002 
2003 /* Get block status from the exported device and send it to the client */
2004 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2005                                     BlockDriverState *bs, uint64_t offset,
2006                                     uint32_t length, bool dont_fragment,
2007                                     bool last, uint32_t context_id,
2008                                     Error **errp)
2009 {
2010     int ret;
2011     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2012     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2013     uint64_t final_length = length;
2014 
2015     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
2016                                  &nb_extents);
2017     if (ret < 0) {
2018         g_free(extents);
2019         return nbd_co_send_structured_error(
2020                 client, handle, -ret, "can't get block status", errp);
2021     }
2022 
2023     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2024                               final_length, last, context_id, errp);
2025 
2026     g_free(extents);
2027 
2028     return ret;
2029 }
2030 
2031 /*
2032  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
2033  * final extent may exceed the original @length. Store in @length the
2034  * byte length encoded (which may be smaller or larger than the
2035  * original), and return the number of extents used.
2036  */
2037 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
2038                                       uint64_t *length, NBDExtent *extents,
2039                                       unsigned int nb_extents,
2040                                       bool dont_fragment)
2041 {
2042     uint64_t begin = offset, end = offset;
2043     uint64_t overall_end = offset + *length;
2044     unsigned int i = 0;
2045     BdrvDirtyBitmapIter *it;
2046     bool dirty;
2047 
2048     bdrv_dirty_bitmap_lock(bitmap);
2049 
2050     it = bdrv_dirty_iter_new(bitmap);
2051     dirty = bdrv_dirty_bitmap_get_locked(bitmap, offset);
2052 
2053     assert(begin < overall_end && nb_extents);
2054     while (begin < overall_end && i < nb_extents) {
2055         bool next_dirty = !dirty;
2056 
2057         if (dirty) {
2058             end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2059         } else {
2060             bdrv_set_dirty_iter(it, begin);
2061             end = bdrv_dirty_iter_next(it);
2062         }
2063         if (end == -1 || end - begin > UINT32_MAX) {
2064             /* Cap to an aligned value < 4G beyond begin. */
2065             end = MIN(bdrv_dirty_bitmap_size(bitmap),
2066                       begin + UINT32_MAX + 1 -
2067                       bdrv_dirty_bitmap_granularity(bitmap));
2068             next_dirty = dirty;
2069         }
2070         if (dont_fragment && end > overall_end) {
2071             end = overall_end;
2072         }
2073 
2074         extents[i].length = cpu_to_be32(end - begin);
2075         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2076         i++;
2077         begin = end;
2078         dirty = next_dirty;
2079     }
2080 
2081     bdrv_dirty_iter_free(it);
2082 
2083     bdrv_dirty_bitmap_unlock(bitmap);
2084 
2085     assert(offset < end);
2086     *length = end - offset;
2087     return i;
2088 }
2089 
2090 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2091                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2092                               uint32_t length, bool dont_fragment, bool last,
2093                               uint32_t context_id, Error **errp)
2094 {
2095     int ret;
2096     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2097     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2098     uint64_t final_length = length;
2099 
2100     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2101                                    nb_extents, dont_fragment);
2102 
2103     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2104                               final_length, last, context_id, errp);
2105 
2106     g_free(extents);
2107 
2108     return ret;
2109 }
2110 
2111 /* nbd_co_receive_request
2112  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2113  * connection right away, and any other negative value to report an error to
2114  * the client (although the caller may still need to disconnect after reporting
2115  * the error).
2116  */
2117 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2118                                   Error **errp)
2119 {
2120     NBDClient *client = req->client;
2121     int valid_flags;
2122 
2123     g_assert(qemu_in_coroutine());
2124     assert(client->recv_coroutine == qemu_coroutine_self());
2125     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2126         return -EIO;
2127     }
2128 
2129     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2130                                              nbd_cmd_lookup(request->type));
2131 
2132     if (request->type != NBD_CMD_WRITE) {
2133         /* No payload, we are ready to read the next request.  */
2134         req->complete = true;
2135     }
2136 
2137     if (request->type == NBD_CMD_DISC) {
2138         /* Special case: we're going to disconnect without a reply,
2139          * whether or not flags, from, or len are bogus */
2140         return -EIO;
2141     }
2142 
2143     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2144         request->type == NBD_CMD_CACHE)
2145     {
2146         if (request->len > NBD_MAX_BUFFER_SIZE) {
2147             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2148                        request->len, NBD_MAX_BUFFER_SIZE);
2149             return -EINVAL;
2150         }
2151 
2152         if (request->type != NBD_CMD_CACHE) {
2153             req->data = blk_try_blockalign(client->exp->blk, request->len);
2154             if (req->data == NULL) {
2155                 error_setg(errp, "No memory");
2156                 return -ENOMEM;
2157             }
2158         }
2159     }
2160 
2161     if (request->type == NBD_CMD_WRITE) {
2162         if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2163                      errp) < 0)
2164         {
2165             return -EIO;
2166         }
2167         req->complete = true;
2168 
2169         trace_nbd_co_receive_request_payload_received(request->handle,
2170                                                       request->len);
2171     }
2172 
2173     /* Sanity checks. */
2174     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2175         (request->type == NBD_CMD_WRITE ||
2176          request->type == NBD_CMD_WRITE_ZEROES ||
2177          request->type == NBD_CMD_TRIM)) {
2178         error_setg(errp, "Export is read-only");
2179         return -EROFS;
2180     }
2181     if (request->from > client->exp->size ||
2182         request->len > client->exp->size - request->from) {
2183         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2184                    ", Size: %" PRIu64, request->from, request->len,
2185                    client->exp->size);
2186         return (request->type == NBD_CMD_WRITE ||
2187                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2188     }
2189     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2190                                                 client->check_align)) {
2191         /*
2192          * The block layer gracefully handles unaligned requests, but
2193          * it's still worth tracing client non-compliance
2194          */
2195         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2196                                               request->from,
2197                                               request->len,
2198                                               client->check_align);
2199     }
2200     valid_flags = NBD_CMD_FLAG_FUA;
2201     if (request->type == NBD_CMD_READ && client->structured_reply) {
2202         valid_flags |= NBD_CMD_FLAG_DF;
2203     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2204         valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2205     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2206         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2207     }
2208     if (request->flags & ~valid_flags) {
2209         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2210                    nbd_cmd_lookup(request->type), request->flags);
2211         return -EINVAL;
2212     }
2213 
2214     return 0;
2215 }
2216 
2217 /* Send simple reply without a payload, or a structured error
2218  * @error_msg is ignored if @ret >= 0
2219  * Returns 0 if connection is still live, -errno on failure to talk to client
2220  */
2221 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2222                                                uint64_t handle,
2223                                                int ret,
2224                                                const char *error_msg,
2225                                                Error **errp)
2226 {
2227     if (client->structured_reply && ret < 0) {
2228         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2229                                             errp);
2230     } else {
2231         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2232                                         NULL, 0, errp);
2233     }
2234 }
2235 
2236 /* Handle NBD_CMD_READ request.
2237  * Return -errno if sending fails. Other errors are reported directly to the
2238  * client as an error reply. */
2239 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2240                                         uint8_t *data, Error **errp)
2241 {
2242     int ret;
2243     NBDExport *exp = client->exp;
2244 
2245     assert(request->type == NBD_CMD_READ);
2246 
2247     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2248     if (request->flags & NBD_CMD_FLAG_FUA) {
2249         ret = blk_co_flush(exp->blk);
2250         if (ret < 0) {
2251             return nbd_send_generic_reply(client, request->handle, ret,
2252                                           "flush failed", errp);
2253         }
2254     }
2255 
2256     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2257         request->len)
2258     {
2259         return nbd_co_send_sparse_read(client, request->handle, request->from,
2260                                        data, request->len, errp);
2261     }
2262 
2263     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2264                     request->len);
2265     if (ret < 0) {
2266         return nbd_send_generic_reply(client, request->handle, ret,
2267                                       "reading from file failed", errp);
2268     }
2269 
2270     if (client->structured_reply) {
2271         if (request->len) {
2272             return nbd_co_send_structured_read(client, request->handle,
2273                                                request->from, data,
2274                                                request->len, true, errp);
2275         } else {
2276             return nbd_co_send_structured_done(client, request->handle, errp);
2277         }
2278     } else {
2279         return nbd_co_send_simple_reply(client, request->handle, 0,
2280                                         data, request->len, errp);
2281     }
2282 }
2283 
2284 /*
2285  * nbd_do_cmd_cache
2286  *
2287  * Handle NBD_CMD_CACHE request.
2288  * Return -errno if sending fails. Other errors are reported directly to the
2289  * client as an error reply.
2290  */
2291 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2292                                          Error **errp)
2293 {
2294     int ret;
2295     NBDExport *exp = client->exp;
2296 
2297     assert(request->type == NBD_CMD_CACHE);
2298 
2299     ret = blk_co_preadv(exp->blk, request->from + exp->dev_offset, request->len,
2300                         NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2301 
2302     return nbd_send_generic_reply(client, request->handle, ret,
2303                                   "caching data failed", errp);
2304 }
2305 
2306 /* Handle NBD request.
2307  * Return -errno if sending fails. Other errors are reported directly to the
2308  * client as an error reply. */
2309 static coroutine_fn int nbd_handle_request(NBDClient *client,
2310                                            NBDRequest *request,
2311                                            uint8_t *data, Error **errp)
2312 {
2313     int ret;
2314     int flags;
2315     NBDExport *exp = client->exp;
2316     char *msg;
2317 
2318     switch (request->type) {
2319     case NBD_CMD_CACHE:
2320         return nbd_do_cmd_cache(client, request, errp);
2321 
2322     case NBD_CMD_READ:
2323         return nbd_do_cmd_read(client, request, data, errp);
2324 
2325     case NBD_CMD_WRITE:
2326         flags = 0;
2327         if (request->flags & NBD_CMD_FLAG_FUA) {
2328             flags |= BDRV_REQ_FUA;
2329         }
2330         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2331                          data, request->len, flags);
2332         return nbd_send_generic_reply(client, request->handle, ret,
2333                                       "writing to file failed", errp);
2334 
2335     case NBD_CMD_WRITE_ZEROES:
2336         flags = 0;
2337         if (request->flags & NBD_CMD_FLAG_FUA) {
2338             flags |= BDRV_REQ_FUA;
2339         }
2340         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2341             flags |= BDRV_REQ_MAY_UNMAP;
2342         }
2343         if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2344             flags |= BDRV_REQ_NO_FALLBACK;
2345         }
2346         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2347                                 request->len, flags);
2348         return nbd_send_generic_reply(client, request->handle, ret,
2349                                       "writing to file failed", errp);
2350 
2351     case NBD_CMD_DISC:
2352         /* unreachable, thanks to special case in nbd_co_receive_request() */
2353         abort();
2354 
2355     case NBD_CMD_FLUSH:
2356         ret = blk_co_flush(exp->blk);
2357         return nbd_send_generic_reply(client, request->handle, ret,
2358                                       "flush failed", errp);
2359 
2360     case NBD_CMD_TRIM:
2361         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2362                               request->len);
2363         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2364             ret = blk_co_flush(exp->blk);
2365         }
2366         return nbd_send_generic_reply(client, request->handle, ret,
2367                                       "discard failed", errp);
2368 
2369     case NBD_CMD_BLOCK_STATUS:
2370         if (!request->len) {
2371             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2372                                           "need non-zero length", errp);
2373         }
2374         if (client->export_meta.valid &&
2375             (client->export_meta.base_allocation ||
2376              client->export_meta.bitmap))
2377         {
2378             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2379 
2380             if (client->export_meta.base_allocation) {
2381                 ret = nbd_co_send_block_status(client, request->handle,
2382                                                blk_bs(exp->blk), request->from,
2383                                                request->len, dont_fragment,
2384                                                !client->export_meta.bitmap,
2385                                                NBD_META_ID_BASE_ALLOCATION,
2386                                                errp);
2387             } else {              /* client->export_meta.bitmap */
2388                 ret = nbd_co_send_bitmap(client, request->handle,
2389                                          client->exp->export_bitmap,
2390                                          request->from, request->len,
2391                                          dont_fragment,
2392                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2393             }
2394 
2395             return ret;
2396         } else {
2397             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2398                                           "CMD_BLOCK_STATUS not negotiated",
2399                                           errp);
2400         }
2401 
2402     default:
2403         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2404                               request->type);
2405         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2406                                      errp);
2407         g_free(msg);
2408         return ret;
2409     }
2410 }
2411 
2412 /* Owns a reference to the NBDClient passed as opaque.  */
2413 static coroutine_fn void nbd_trip(void *opaque)
2414 {
2415     NBDClient *client = opaque;
2416     NBDRequestData *req;
2417     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2418     int ret;
2419     Error *local_err = NULL;
2420 
2421     trace_nbd_trip();
2422     if (client->closing) {
2423         nbd_client_put(client);
2424         return;
2425     }
2426 
2427     req = nbd_request_get(client);
2428     ret = nbd_co_receive_request(req, &request, &local_err);
2429     client->recv_coroutine = NULL;
2430 
2431     if (client->closing) {
2432         /*
2433          * The client may be closed when we are blocked in
2434          * nbd_co_receive_request()
2435          */
2436         goto done;
2437     }
2438 
2439     nbd_client_receive_next_request(client);
2440     if (ret == -EIO) {
2441         goto disconnect;
2442     }
2443 
2444     if (ret < 0) {
2445         /* It wans't -EIO, so, according to nbd_co_receive_request()
2446          * semantics, we should return the error to the client. */
2447         Error *export_err = local_err;
2448 
2449         local_err = NULL;
2450         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2451                                      error_get_pretty(export_err), &local_err);
2452         error_free(export_err);
2453     } else {
2454         ret = nbd_handle_request(client, &request, req->data, &local_err);
2455     }
2456     if (ret < 0) {
2457         error_prepend(&local_err, "Failed to send reply: ");
2458         goto disconnect;
2459     }
2460 
2461     /* We must disconnect after NBD_CMD_WRITE if we did not
2462      * read the payload.
2463      */
2464     if (!req->complete) {
2465         error_setg(&local_err, "Request handling failed in intermediate state");
2466         goto disconnect;
2467     }
2468 
2469 done:
2470     nbd_request_put(req);
2471     nbd_client_put(client);
2472     return;
2473 
2474 disconnect:
2475     if (local_err) {
2476         error_reportf_err(local_err, "Disconnect client, due to: ");
2477     }
2478     nbd_request_put(req);
2479     client_close(client, true);
2480     nbd_client_put(client);
2481 }
2482 
2483 static void nbd_client_receive_next_request(NBDClient *client)
2484 {
2485     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2486         nbd_client_get(client);
2487         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2488         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2489     }
2490 }
2491 
2492 static coroutine_fn void nbd_co_client_start(void *opaque)
2493 {
2494     NBDClient *client = opaque;
2495     Error *local_err = NULL;
2496 
2497     qemu_co_mutex_init(&client->send_lock);
2498 
2499     if (nbd_negotiate(client, &local_err)) {
2500         if (local_err) {
2501             error_report_err(local_err);
2502         }
2503         client_close(client, false);
2504         return;
2505     }
2506 
2507     nbd_client_receive_next_request(client);
2508 }
2509 
2510 /*
2511  * Create a new client listener using the given channel @sioc.
2512  * Begin servicing it in a coroutine.  When the connection closes, call
2513  * @close_fn with an indication of whether the client completed negotiation.
2514  */
2515 void nbd_client_new(QIOChannelSocket *sioc,
2516                     QCryptoTLSCreds *tlscreds,
2517                     const char *tlsauthz,
2518                     void (*close_fn)(NBDClient *, bool))
2519 {
2520     NBDClient *client;
2521     Coroutine *co;
2522 
2523     client = g_new0(NBDClient, 1);
2524     client->refcount = 1;
2525     client->tlscreds = tlscreds;
2526     if (tlscreds) {
2527         object_ref(OBJECT(client->tlscreds));
2528     }
2529     client->tlsauthz = g_strdup(tlsauthz);
2530     client->sioc = sioc;
2531     object_ref(OBJECT(client->sioc));
2532     client->ioc = QIO_CHANNEL(sioc);
2533     object_ref(OBJECT(client->ioc));
2534     client->close_fn = close_fn;
2535 
2536     co = qemu_coroutine_create(nbd_co_client_start, client);
2537     qemu_coroutine_enter(co);
2538 }
2539