xref: /openbmc/qemu/nbd/server.c (revision 83a71719)
1 /*
2  *  Copyright (C) 2016-2018 Red Hat, Inc.
3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
4  *
5  *  Network Block Device Server Side
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; under version 2 of the License.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "trace.h"
23 #include "nbd-internal.h"
24 
25 #define NBD_META_ID_BASE_ALLOCATION 0
26 #define NBD_META_ID_DIRTY_BITMAP 1
27 
28 /* NBD_MAX_BITMAP_EXTENTS: 1 mb of extents data. An empirical
29  * constant. If an increase is needed, note that the NBD protocol
30  * recommends no larger than 32 mb, so that the client won't consider
31  * the reply as a denial of service attack. */
32 #define NBD_MAX_BITMAP_EXTENTS (0x100000 / 8)
33 
34 static int system_errno_to_nbd_errno(int err)
35 {
36     switch (err) {
37     case 0:
38         return NBD_SUCCESS;
39     case EPERM:
40     case EROFS:
41         return NBD_EPERM;
42     case EIO:
43         return NBD_EIO;
44     case ENOMEM:
45         return NBD_ENOMEM;
46 #ifdef EDQUOT
47     case EDQUOT:
48 #endif
49     case EFBIG:
50     case ENOSPC:
51         return NBD_ENOSPC;
52     case EOVERFLOW:
53         return NBD_EOVERFLOW;
54     case ESHUTDOWN:
55         return NBD_ESHUTDOWN;
56     case EINVAL:
57     default:
58         return NBD_EINVAL;
59     }
60 }
61 
62 /* Definitions for opaque data types */
63 
64 typedef struct NBDRequestData NBDRequestData;
65 
66 struct NBDRequestData {
67     QSIMPLEQ_ENTRY(NBDRequestData) entry;
68     NBDClient *client;
69     uint8_t *data;
70     bool complete;
71 };
72 
73 struct NBDExport {
74     int refcount;
75     void (*close)(NBDExport *exp);
76 
77     BlockBackend *blk;
78     char *name;
79     char *description;
80     uint64_t dev_offset;
81     uint64_t size;
82     uint16_t nbdflags;
83     QTAILQ_HEAD(, NBDClient) clients;
84     QTAILQ_ENTRY(NBDExport) next;
85 
86     AioContext *ctx;
87 
88     BlockBackend *eject_notifier_blk;
89     Notifier eject_notifier;
90 
91     BdrvDirtyBitmap *export_bitmap;
92     char *export_bitmap_context;
93 };
94 
95 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
96 
97 /* NBDExportMetaContexts represents a list of contexts to be exported,
98  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
99  * NBD_OPT_LIST_META_CONTEXT. */
100 typedef struct NBDExportMetaContexts {
101     NBDExport *exp;
102     bool valid; /* means that negotiation of the option finished without
103                    errors */
104     bool base_allocation; /* export base:allocation context (block status) */
105     bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
106 } NBDExportMetaContexts;
107 
108 struct NBDClient {
109     int refcount;
110     void (*close_fn)(NBDClient *client, bool negotiated);
111 
112     NBDExport *exp;
113     QCryptoTLSCreds *tlscreds;
114     char *tlsaclname;
115     QIOChannelSocket *sioc; /* The underlying data channel */
116     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
117 
118     Coroutine *recv_coroutine;
119 
120     CoMutex send_lock;
121     Coroutine *send_coroutine;
122 
123     QTAILQ_ENTRY(NBDClient) next;
124     int nb_requests;
125     bool closing;
126 
127     bool structured_reply;
128     NBDExportMetaContexts export_meta;
129 
130     uint32_t opt; /* Current option being negotiated */
131     uint32_t optlen; /* remaining length of data in ioc for the option being
132                         negotiated now */
133 };
134 
135 static void nbd_client_receive_next_request(NBDClient *client);
136 
137 /* Basic flow for negotiation
138 
139    Server         Client
140    Negotiate
141 
142    or
143 
144    Server         Client
145    Negotiate #1
146                   Option
147    Negotiate #2
148 
149    ----
150 
151    followed by
152 
153    Server         Client
154                   Request
155    Response
156                   Request
157    Response
158                   ...
159    ...
160                   Request (type == 2)
161 
162 */
163 
164 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
165                                      uint32_t type, uint32_t length)
166 {
167     stq_be_p(&rep->magic, NBD_REP_MAGIC);
168     stl_be_p(&rep->option, option);
169     stl_be_p(&rep->type, type);
170     stl_be_p(&rep->length, length);
171 }
172 
173 /* Send a reply header, including length, but no payload.
174  * Return -errno on error, 0 on success. */
175 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
176                                       uint32_t len, Error **errp)
177 {
178     NBDOptionReply rep;
179 
180     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
181                                      type, nbd_rep_lookup(type), len);
182 
183     assert(len < NBD_MAX_BUFFER_SIZE);
184 
185     set_be_option_rep(&rep, client->opt, type, len);
186     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
187 }
188 
189 /* Send a reply header with default 0 length.
190  * Return -errno on error, 0 on success. */
191 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
192                                   Error **errp)
193 {
194     return nbd_negotiate_send_rep_len(client, type, 0, errp);
195 }
196 
197 /* Send an error reply.
198  * Return -errno on error, 0 on success. */
199 static int GCC_FMT_ATTR(4, 0)
200 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
201                             Error **errp, const char *fmt, va_list va)
202 {
203     char *msg;
204     int ret;
205     size_t len;
206 
207     msg = g_strdup_vprintf(fmt, va);
208     len = strlen(msg);
209     assert(len < 4096);
210     trace_nbd_negotiate_send_rep_err(msg);
211     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
212     if (ret < 0) {
213         goto out;
214     }
215     if (nbd_write(client->ioc, msg, len, errp) < 0) {
216         error_prepend(errp, "write failed (error message): ");
217         ret = -EIO;
218     } else {
219         ret = 0;
220     }
221 
222 out:
223     g_free(msg);
224     return ret;
225 }
226 
227 /* Send an error reply.
228  * Return -errno on error, 0 on success. */
229 static int GCC_FMT_ATTR(4, 5)
230 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
231                            Error **errp, const char *fmt, ...)
232 {
233     va_list va;
234     int ret;
235 
236     va_start(va, fmt);
237     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
238     va_end(va);
239     return ret;
240 }
241 
242 /* Drop remainder of the current option, and send a reply with the
243  * given error type and message. Return -errno on read or write
244  * failure; or 0 if connection is still live. */
245 static int GCC_FMT_ATTR(4, 0)
246 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
247               const char *fmt, va_list va)
248 {
249     int ret = nbd_drop(client->ioc, client->optlen, errp);
250 
251     client->optlen = 0;
252     if (!ret) {
253         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
254     }
255     return ret;
256 }
257 
258 static int GCC_FMT_ATTR(4, 5)
259 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
260              const char *fmt, ...)
261 {
262     int ret;
263     va_list va;
264 
265     va_start(va, fmt);
266     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
267     va_end(va);
268 
269     return ret;
270 }
271 
272 static int GCC_FMT_ATTR(3, 4)
273 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
274 {
275     int ret;
276     va_list va;
277 
278     va_start(va, fmt);
279     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
280     va_end(va);
281 
282     return ret;
283 }
284 
285 /* Read size bytes from the unparsed payload of the current option.
286  * Return -errno on I/O error, 0 if option was completely handled by
287  * sending a reply about inconsistent lengths, or 1 on success. */
288 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
289                         Error **errp)
290 {
291     if (size > client->optlen) {
292         return nbd_opt_invalid(client, errp,
293                                "Inconsistent lengths in option %s",
294                                nbd_opt_lookup(client->opt));
295     }
296     client->optlen -= size;
297     return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
298 }
299 
300 /* Drop size bytes from the unparsed payload of the current option.
301  * Return -errno on I/O error, 0 if option was completely handled by
302  * sending a reply about inconsistent lengths, or 1 on success. */
303 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
304 {
305     if (size > client->optlen) {
306         return nbd_opt_invalid(client, errp,
307                                "Inconsistent lengths in option %s",
308                                nbd_opt_lookup(client->opt));
309     }
310     client->optlen -= size;
311     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
312 }
313 
314 /* nbd_opt_read_name
315  *
316  * Read a string with the format:
317  *   uint32_t len     (<= NBD_MAX_NAME_SIZE)
318  *   len bytes string (not 0-terminated)
319  *
320  * @name should be enough to store NBD_MAX_NAME_SIZE+1.
321  * If @length is non-null, it will be set to the actual string length.
322  *
323  * Return -errno on I/O error, 0 if option was completely handled by
324  * sending a reply about inconsistent lengths, or 1 on success.
325  */
326 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
327                              Error **errp)
328 {
329     int ret;
330     uint32_t len;
331 
332     ret = nbd_opt_read(client, &len, sizeof(len), errp);
333     if (ret <= 0) {
334         return ret;
335     }
336     len = cpu_to_be32(len);
337 
338     if (len > NBD_MAX_NAME_SIZE) {
339         return nbd_opt_invalid(client, errp,
340                                "Invalid name length: %" PRIu32, len);
341     }
342 
343     ret = nbd_opt_read(client, name, len, errp);
344     if (ret <= 0) {
345         return ret;
346     }
347     name[len] = '\0';
348 
349     if (length) {
350         *length = len;
351     }
352 
353     return 1;
354 }
355 
356 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
357  * Return -errno on error, 0 on success. */
358 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
359                                        Error **errp)
360 {
361     size_t name_len, desc_len;
362     uint32_t len;
363     const char *name = exp->name ? exp->name : "";
364     const char *desc = exp->description ? exp->description : "";
365     QIOChannel *ioc = client->ioc;
366     int ret;
367 
368     trace_nbd_negotiate_send_rep_list(name, desc);
369     name_len = strlen(name);
370     desc_len = strlen(desc);
371     len = name_len + desc_len + sizeof(len);
372     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
373     if (ret < 0) {
374         return ret;
375     }
376 
377     len = cpu_to_be32(name_len);
378     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
379         error_prepend(errp, "write failed (name length): ");
380         return -EINVAL;
381     }
382 
383     if (nbd_write(ioc, name, name_len, errp) < 0) {
384         error_prepend(errp, "write failed (name buffer): ");
385         return -EINVAL;
386     }
387 
388     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
389         error_prepend(errp, "write failed (description buffer): ");
390         return -EINVAL;
391     }
392 
393     return 0;
394 }
395 
396 /* Process the NBD_OPT_LIST command, with a potential series of replies.
397  * Return -errno on error, 0 on success. */
398 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
399 {
400     NBDExport *exp;
401     assert(client->opt == NBD_OPT_LIST);
402 
403     /* For each export, send a NBD_REP_SERVER reply. */
404     QTAILQ_FOREACH(exp, &exports, next) {
405         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
406             return -EINVAL;
407         }
408     }
409     /* Finish with a NBD_REP_ACK. */
410     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
411 }
412 
413 static void nbd_check_meta_export(NBDClient *client)
414 {
415     client->export_meta.valid &= client->exp == client->export_meta.exp;
416 }
417 
418 /* Send a reply to NBD_OPT_EXPORT_NAME.
419  * Return -errno on error, 0 on success. */
420 static int nbd_negotiate_handle_export_name(NBDClient *client,
421                                             uint16_t myflags, bool no_zeroes,
422                                             Error **errp)
423 {
424     char name[NBD_MAX_NAME_SIZE + 1];
425     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
426     size_t len;
427     int ret;
428 
429     /* Client sends:
430         [20 ..  xx]   export name (length bytes)
431        Server replies:
432         [ 0 ..   7]   size
433         [ 8 ..   9]   export flags
434         [10 .. 133]   reserved     (0) [unless no_zeroes]
435      */
436     trace_nbd_negotiate_handle_export_name();
437     if (client->optlen >= sizeof(name)) {
438         error_setg(errp, "Bad length received");
439         return -EINVAL;
440     }
441     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
442         return -EIO;
443     }
444     name[client->optlen] = '\0';
445     client->optlen = 0;
446 
447     trace_nbd_negotiate_handle_export_name_request(name);
448 
449     client->exp = nbd_export_find(name);
450     if (!client->exp) {
451         error_setg(errp, "export not found");
452         return -EINVAL;
453     }
454 
455     trace_nbd_negotiate_new_style_size_flags(client->exp->size,
456                                              client->exp->nbdflags | myflags);
457     stq_be_p(buf, client->exp->size);
458     stw_be_p(buf + 8, client->exp->nbdflags | myflags);
459     len = no_zeroes ? 10 : sizeof(buf);
460     ret = nbd_write(client->ioc, buf, len, errp);
461     if (ret < 0) {
462         error_prepend(errp, "write failed: ");
463         return ret;
464     }
465 
466     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
467     nbd_export_get(client->exp);
468     nbd_check_meta_export(client);
469 
470     return 0;
471 }
472 
473 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
474  * The buffer does NOT include the info type prefix.
475  * Return -errno on error, 0 if ready to send more. */
476 static int nbd_negotiate_send_info(NBDClient *client,
477                                    uint16_t info, uint32_t length, void *buf,
478                                    Error **errp)
479 {
480     int rc;
481 
482     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
483     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
484                                     sizeof(info) + length, errp);
485     if (rc < 0) {
486         return rc;
487     }
488     info = cpu_to_be16(info);
489     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
490         return -EIO;
491     }
492     if (nbd_write(client->ioc, buf, length, errp) < 0) {
493         return -EIO;
494     }
495     return 0;
496 }
497 
498 /* nbd_reject_length: Handle any unexpected payload.
499  * @fatal requests that we quit talking to the client, even if we are able
500  * to successfully send an error reply.
501  * Return:
502  * -errno  transmission error occurred or @fatal was requested, errp is set
503  * 0       error message successfully sent to client, errp is not set
504  */
505 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
506 {
507     int ret;
508 
509     assert(client->optlen);
510     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
511                           nbd_opt_lookup(client->opt));
512     if (fatal && !ret) {
513         error_setg(errp, "option '%s' has unexpected length",
514                    nbd_opt_lookup(client->opt));
515         return -EINVAL;
516     }
517     return ret;
518 }
519 
520 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
521  * Return -errno on error, 0 if ready for next option, and 1 to move
522  * into transmission phase.  */
523 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
524                                      Error **errp)
525 {
526     int rc;
527     char name[NBD_MAX_NAME_SIZE + 1];
528     NBDExport *exp;
529     uint16_t requests;
530     uint16_t request;
531     uint32_t namelen;
532     bool sendname = false;
533     bool blocksize = false;
534     uint32_t sizes[3];
535     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
536 
537     /* Client sends:
538         4 bytes: L, name length (can be 0)
539         L bytes: export name
540         2 bytes: N, number of requests (can be 0)
541         N * 2 bytes: N requests
542     */
543     rc = nbd_opt_read_name(client, name, &namelen, errp);
544     if (rc <= 0) {
545         return rc;
546     }
547     trace_nbd_negotiate_handle_export_name_request(name);
548 
549     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
550     if (rc <= 0) {
551         return rc;
552     }
553     requests = be16_to_cpu(requests);
554     trace_nbd_negotiate_handle_info_requests(requests);
555     while (requests--) {
556         rc = nbd_opt_read(client, &request, sizeof(request), errp);
557         if (rc <= 0) {
558             return rc;
559         }
560         request = be16_to_cpu(request);
561         trace_nbd_negotiate_handle_info_request(request,
562                                                 nbd_info_lookup(request));
563         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
564          * everything else is either a request we don't know or
565          * something we send regardless of request */
566         switch (request) {
567         case NBD_INFO_NAME:
568             sendname = true;
569             break;
570         case NBD_INFO_BLOCK_SIZE:
571             blocksize = true;
572             break;
573         }
574     }
575     if (client->optlen) {
576         return nbd_reject_length(client, false, errp);
577     }
578 
579     exp = nbd_export_find(name);
580     if (!exp) {
581         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
582                                           errp, "export '%s' not present",
583                                           name);
584     }
585 
586     /* Don't bother sending NBD_INFO_NAME unless client requested it */
587     if (sendname) {
588         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
589                                      errp);
590         if (rc < 0) {
591             return rc;
592         }
593     }
594 
595     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
596      * client request */
597     if (exp->description) {
598         size_t len = strlen(exp->description);
599 
600         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
601                                      len, exp->description, errp);
602         if (rc < 0) {
603             return rc;
604         }
605     }
606 
607     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
608      * according to whether the client requested it, and according to
609      * whether this is OPT_INFO or OPT_GO. */
610     /* minimum - 1 for back-compat, or 512 if client is new enough.
611      * TODO: consult blk_bs(blk)->bl.request_alignment? */
612     sizes[0] =
613             (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
614     /* preferred - Hard-code to 4096 for now.
615      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
616     sizes[1] = 4096;
617     /* maximum - At most 32M, but smaller as appropriate. */
618     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
619     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
620     sizes[0] = cpu_to_be32(sizes[0]);
621     sizes[1] = cpu_to_be32(sizes[1]);
622     sizes[2] = cpu_to_be32(sizes[2]);
623     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
624                                  sizeof(sizes), sizes, errp);
625     if (rc < 0) {
626         return rc;
627     }
628 
629     /* Send NBD_INFO_EXPORT always */
630     trace_nbd_negotiate_new_style_size_flags(exp->size,
631                                              exp->nbdflags | myflags);
632     stq_be_p(buf, exp->size);
633     stw_be_p(buf + 8, exp->nbdflags | myflags);
634     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
635                                  sizeof(buf), buf, errp);
636     if (rc < 0) {
637         return rc;
638     }
639 
640     /* If the client is just asking for NBD_OPT_INFO, but forgot to
641      * request block sizes, return an error.
642      * TODO: consult blk_bs(blk)->request_align, and only error if it
643      * is not 1? */
644     if (client->opt == NBD_OPT_INFO && !blocksize) {
645         return nbd_negotiate_send_rep_err(client,
646                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
647                                           errp,
648                                           "request NBD_INFO_BLOCK_SIZE to "
649                                           "use this export");
650     }
651 
652     /* Final reply */
653     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
654     if (rc < 0) {
655         return rc;
656     }
657 
658     if (client->opt == NBD_OPT_GO) {
659         client->exp = exp;
660         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
661         nbd_export_get(client->exp);
662         nbd_check_meta_export(client);
663         rc = 1;
664     }
665     return rc;
666 }
667 
668 
669 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
670  * new channel for all further (now-encrypted) communication. */
671 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
672                                                  Error **errp)
673 {
674     QIOChannel *ioc;
675     QIOChannelTLS *tioc;
676     struct NBDTLSHandshakeData data = { 0 };
677 
678     assert(client->opt == NBD_OPT_STARTTLS);
679 
680     trace_nbd_negotiate_handle_starttls();
681     ioc = client->ioc;
682 
683     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
684         return NULL;
685     }
686 
687     tioc = qio_channel_tls_new_server(ioc,
688                                       client->tlscreds,
689                                       client->tlsaclname,
690                                       errp);
691     if (!tioc) {
692         return NULL;
693     }
694 
695     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
696     trace_nbd_negotiate_handle_starttls_handshake();
697     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
698     qio_channel_tls_handshake(tioc,
699                               nbd_tls_handshake,
700                               &data,
701                               NULL,
702                               NULL);
703 
704     if (!data.complete) {
705         g_main_loop_run(data.loop);
706     }
707     g_main_loop_unref(data.loop);
708     if (data.error) {
709         object_unref(OBJECT(tioc));
710         error_propagate(errp, data.error);
711         return NULL;
712     }
713 
714     return QIO_CHANNEL(tioc);
715 }
716 
717 /* nbd_negotiate_send_meta_context
718  *
719  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
720  *
721  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
722  */
723 static int nbd_negotiate_send_meta_context(NBDClient *client,
724                                            const char *context,
725                                            uint32_t context_id,
726                                            Error **errp)
727 {
728     NBDOptionReplyMetaContext opt;
729     struct iovec iov[] = {
730         {.iov_base = &opt, .iov_len = sizeof(opt)},
731         {.iov_base = (void *)context, .iov_len = strlen(context)}
732     };
733 
734     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
735         context_id = 0;
736     }
737 
738     trace_nbd_negotiate_meta_query_reply(context, context_id);
739     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
740                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
741     stl_be_p(&opt.context_id, context_id);
742 
743     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
744 }
745 
746 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
747  * @match is never set to false.
748  *
749  * Return -errno on I/O error, 0 if option was completely handled by
750  * sending a reply about inconsistent lengths, or 1 on success.
751  *
752  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
753  * It only means that there are no errors.
754  */
755 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
756                             Error **errp)
757 {
758     int ret;
759     char *query;
760     size_t len = strlen(pattern);
761 
762     assert(len);
763 
764     query = g_malloc(len);
765     ret = nbd_opt_read(client, query, len, errp);
766     if (ret <= 0) {
767         g_free(query);
768         return ret;
769     }
770 
771     if (strncmp(query, pattern, len) == 0) {
772         trace_nbd_negotiate_meta_query_parse(pattern);
773         *match = true;
774     } else {
775         trace_nbd_negotiate_meta_query_skip("pattern not matched");
776     }
777     g_free(query);
778 
779     return 1;
780 }
781 
782 /*
783  * Read @len bytes, and set @match to true if they match @pattern, or if @len
784  * is 0 and the client is performing _LIST_. @match is never set to false.
785  *
786  * Return -errno on I/O error, 0 if option was completely handled by
787  * sending a reply about inconsistent lengths, or 1 on success.
788  *
789  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
790  * It only means that there are no errors.
791  */
792 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
793                                      uint32_t len, bool *match, Error **errp)
794 {
795     if (len == 0) {
796         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
797             *match = true;
798         }
799         trace_nbd_negotiate_meta_query_parse("empty");
800         return 1;
801     }
802 
803     if (len != strlen(pattern)) {
804         trace_nbd_negotiate_meta_query_skip("different lengths");
805         return nbd_opt_skip(client, len, errp);
806     }
807 
808     return nbd_meta_pattern(client, pattern, match, errp);
809 }
810 
811 /* nbd_meta_base_query
812  *
813  * Handle queries to 'base' namespace. For now, only the base:allocation
814  * context is available.  'len' is the amount of text remaining to be read from
815  * the current name, after the 'base:' portion has been stripped.
816  *
817  * Return -errno on I/O error, 0 if option was completely handled by
818  * sending a reply about inconsistent lengths, or 1 on success.
819  */
820 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
821                                uint32_t len, Error **errp)
822 {
823     return nbd_meta_empty_or_pattern(client, "allocation", len,
824                                      &meta->base_allocation, errp);
825 }
826 
827 /* nbd_meta_bitmap_query
828  *
829  * Handle query to 'qemu:' namespace.
830  * @len is the amount of text remaining to be read from the current name, after
831  * the 'qemu:' portion has been stripped.
832  *
833  * Return -errno on I/O error, 0 if option was completely handled by
834  * sending a reply about inconsistent lengths, or 1 on success. */
835 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
836                                uint32_t len, Error **errp)
837 {
838     bool dirty_bitmap = false;
839     size_t dirty_bitmap_len = strlen("dirty-bitmap:");
840     int ret;
841 
842     if (!meta->exp->export_bitmap) {
843         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
844         return nbd_opt_skip(client, len, errp);
845     }
846 
847     if (len == 0) {
848         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
849             meta->bitmap = true;
850         }
851         trace_nbd_negotiate_meta_query_parse("empty");
852         return 1;
853     }
854 
855     if (len < dirty_bitmap_len) {
856         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
857         return nbd_opt_skip(client, len, errp);
858     }
859 
860     len -= dirty_bitmap_len;
861     ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
862     if (ret <= 0) {
863         return ret;
864     }
865     if (!dirty_bitmap) {
866         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
867         return nbd_opt_skip(client, len, errp);
868     }
869 
870     trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
871 
872     return nbd_meta_empty_or_pattern(
873             client, meta->exp->export_bitmap_context +
874             strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
875 }
876 
877 /* nbd_negotiate_meta_query
878  *
879  * Parse namespace name and call corresponding function to parse body of the
880  * query.
881  *
882  * The only supported namespace now is 'base'.
883  *
884  * The function aims not wasting time and memory to read long unknown namespace
885  * names.
886  *
887  * Return -errno on I/O error, 0 if option was completely handled by
888  * sending a reply about inconsistent lengths, or 1 on success. */
889 static int nbd_negotiate_meta_query(NBDClient *client,
890                                     NBDExportMetaContexts *meta, Error **errp)
891 {
892     /*
893      * Both 'qemu' and 'base' namespaces have length = 5 including a
894      * colon. If another length namespace is later introduced, this
895      * should certainly be refactored.
896      */
897     int ret;
898     size_t ns_len = 5;
899     char ns[5];
900     uint32_t len;
901 
902     ret = nbd_opt_read(client, &len, sizeof(len), errp);
903     if (ret <= 0) {
904         return ret;
905     }
906     len = cpu_to_be32(len);
907 
908     if (len < ns_len) {
909         trace_nbd_negotiate_meta_query_skip("length too short");
910         return nbd_opt_skip(client, len, errp);
911     }
912 
913     len -= ns_len;
914     ret = nbd_opt_read(client, ns, ns_len, errp);
915     if (ret <= 0) {
916         return ret;
917     }
918 
919     if (!strncmp(ns, "base:", ns_len)) {
920         trace_nbd_negotiate_meta_query_parse("base:");
921         return nbd_meta_base_query(client, meta, len, errp);
922     } else if (!strncmp(ns, "qemu:", ns_len)) {
923         trace_nbd_negotiate_meta_query_parse("qemu:");
924         return nbd_meta_qemu_query(client, meta, len, errp);
925     }
926 
927     trace_nbd_negotiate_meta_query_skip("unknown namespace");
928     return nbd_opt_skip(client, len, errp);
929 }
930 
931 /* nbd_negotiate_meta_queries
932  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
933  *
934  * Return -errno on I/O error, or 0 if option was completely handled. */
935 static int nbd_negotiate_meta_queries(NBDClient *client,
936                                       NBDExportMetaContexts *meta, Error **errp)
937 {
938     int ret;
939     char export_name[NBD_MAX_NAME_SIZE + 1];
940     NBDExportMetaContexts local_meta;
941     uint32_t nb_queries;
942     int i;
943 
944     if (!client->structured_reply) {
945         return nbd_opt_invalid(client, errp,
946                                "request option '%s' when structured reply "
947                                "is not negotiated",
948                                nbd_opt_lookup(client->opt));
949     }
950 
951     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
952         /* Only change the caller's meta on SET. */
953         meta = &local_meta;
954     }
955 
956     memset(meta, 0, sizeof(*meta));
957 
958     ret = nbd_opt_read_name(client, export_name, NULL, errp);
959     if (ret <= 0) {
960         return ret;
961     }
962 
963     meta->exp = nbd_export_find(export_name);
964     if (meta->exp == NULL) {
965         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
966                             "export '%s' not present", export_name);
967     }
968 
969     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
970     if (ret <= 0) {
971         return ret;
972     }
973     nb_queries = cpu_to_be32(nb_queries);
974     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
975                                      export_name, nb_queries);
976 
977     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
978         /* enable all known contexts */
979         meta->base_allocation = true;
980         meta->bitmap = !!meta->exp->export_bitmap;
981     } else {
982         for (i = 0; i < nb_queries; ++i) {
983             ret = nbd_negotiate_meta_query(client, meta, errp);
984             if (ret <= 0) {
985                 return ret;
986             }
987         }
988     }
989 
990     if (meta->base_allocation) {
991         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
992                                               NBD_META_ID_BASE_ALLOCATION,
993                                               errp);
994         if (ret < 0) {
995             return ret;
996         }
997     }
998 
999     if (meta->bitmap) {
1000         ret = nbd_negotiate_send_meta_context(client,
1001                                               meta->exp->export_bitmap_context,
1002                                               NBD_META_ID_DIRTY_BITMAP,
1003                                               errp);
1004         if (ret < 0) {
1005             return ret;
1006         }
1007     }
1008 
1009     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1010     if (ret == 0) {
1011         meta->valid = true;
1012     }
1013 
1014     return ret;
1015 }
1016 
1017 /* nbd_negotiate_options
1018  * Process all NBD_OPT_* client option commands, during fixed newstyle
1019  * negotiation.
1020  * Return:
1021  * -errno  on error, errp is set
1022  * 0       on successful negotiation, errp is not set
1023  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1024  *         errp is not set
1025  */
1026 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1027                                  Error **errp)
1028 {
1029     uint32_t flags;
1030     bool fixedNewstyle = false;
1031     bool no_zeroes = false;
1032 
1033     /* Client sends:
1034         [ 0 ..   3]   client flags
1035 
1036        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1037         [ 0 ..   7]   NBD_OPTS_MAGIC
1038         [ 8 ..  11]   NBD option
1039         [12 ..  15]   Data length
1040         ...           Rest of request
1041 
1042         [ 0 ..   7]   NBD_OPTS_MAGIC
1043         [ 8 ..  11]   Second NBD option
1044         [12 ..  15]   Data length
1045         ...           Rest of request
1046     */
1047 
1048     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1049         return -EIO;
1050     }
1051     trace_nbd_negotiate_options_flags(flags);
1052     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1053         fixedNewstyle = true;
1054         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1055     }
1056     if (flags & NBD_FLAG_C_NO_ZEROES) {
1057         no_zeroes = true;
1058         flags &= ~NBD_FLAG_C_NO_ZEROES;
1059     }
1060     if (flags != 0) {
1061         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1062         return -EINVAL;
1063     }
1064 
1065     while (1) {
1066         int ret;
1067         uint32_t option, length;
1068         uint64_t magic;
1069 
1070         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1071             return -EINVAL;
1072         }
1073         trace_nbd_negotiate_options_check_magic(magic);
1074         if (magic != NBD_OPTS_MAGIC) {
1075             error_setg(errp, "Bad magic received");
1076             return -EINVAL;
1077         }
1078 
1079         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1080             return -EINVAL;
1081         }
1082         client->opt = option;
1083 
1084         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1085             return -EINVAL;
1086         }
1087         assert(!client->optlen);
1088         client->optlen = length;
1089 
1090         if (length > NBD_MAX_BUFFER_SIZE) {
1091             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1092                        length, NBD_MAX_BUFFER_SIZE);
1093             return -EINVAL;
1094         }
1095 
1096         trace_nbd_negotiate_options_check_option(option,
1097                                                  nbd_opt_lookup(option));
1098         if (client->tlscreds &&
1099             client->ioc == (QIOChannel *)client->sioc) {
1100             QIOChannel *tioc;
1101             if (!fixedNewstyle) {
1102                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1103                 return -EINVAL;
1104             }
1105             switch (option) {
1106             case NBD_OPT_STARTTLS:
1107                 if (length) {
1108                     /* Unconditionally drop the connection if the client
1109                      * can't start a TLS negotiation correctly */
1110                     return nbd_reject_length(client, true, errp);
1111                 }
1112                 tioc = nbd_negotiate_handle_starttls(client, errp);
1113                 if (!tioc) {
1114                     return -EIO;
1115                 }
1116                 ret = 0;
1117                 object_unref(OBJECT(client->ioc));
1118                 client->ioc = QIO_CHANNEL(tioc);
1119                 break;
1120 
1121             case NBD_OPT_EXPORT_NAME:
1122                 /* No way to return an error to client, so drop connection */
1123                 error_setg(errp, "Option 0x%x not permitted before TLS",
1124                            option);
1125                 return -EINVAL;
1126 
1127             default:
1128                 /* Let the client keep trying, unless they asked to
1129                  * quit. Always try to give an error back to the
1130                  * client; but when replying to OPT_ABORT, be aware
1131                  * that the client may hang up before receiving the
1132                  * error, in which case we are fine ignoring the
1133                  * resulting EPIPE. */
1134                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1135                                    option == NBD_OPT_ABORT ? NULL : errp,
1136                                    "Option 0x%" PRIx32
1137                                    " not permitted before TLS", option);
1138                 if (option == NBD_OPT_ABORT) {
1139                     return 1;
1140                 }
1141                 break;
1142             }
1143         } else if (fixedNewstyle) {
1144             switch (option) {
1145             case NBD_OPT_LIST:
1146                 if (length) {
1147                     ret = nbd_reject_length(client, false, errp);
1148                 } else {
1149                     ret = nbd_negotiate_handle_list(client, errp);
1150                 }
1151                 break;
1152 
1153             case NBD_OPT_ABORT:
1154                 /* NBD spec says we must try to reply before
1155                  * disconnecting, but that we must also tolerate
1156                  * guests that don't wait for our reply. */
1157                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1158                 return 1;
1159 
1160             case NBD_OPT_EXPORT_NAME:
1161                 return nbd_negotiate_handle_export_name(client,
1162                                                         myflags, no_zeroes,
1163                                                         errp);
1164 
1165             case NBD_OPT_INFO:
1166             case NBD_OPT_GO:
1167                 ret = nbd_negotiate_handle_info(client, myflags, errp);
1168                 if (ret == 1) {
1169                     assert(option == NBD_OPT_GO);
1170                     return 0;
1171                 }
1172                 break;
1173 
1174             case NBD_OPT_STARTTLS:
1175                 if (length) {
1176                     ret = nbd_reject_length(client, false, errp);
1177                 } else if (client->tlscreds) {
1178                     ret = nbd_negotiate_send_rep_err(client,
1179                                                      NBD_REP_ERR_INVALID, errp,
1180                                                      "TLS already enabled");
1181                 } else {
1182                     ret = nbd_negotiate_send_rep_err(client,
1183                                                      NBD_REP_ERR_POLICY, errp,
1184                                                      "TLS not configured");
1185                 }
1186                 break;
1187 
1188             case NBD_OPT_STRUCTURED_REPLY:
1189                 if (length) {
1190                     ret = nbd_reject_length(client, false, errp);
1191                 } else if (client->structured_reply) {
1192                     ret = nbd_negotiate_send_rep_err(
1193                         client, NBD_REP_ERR_INVALID, errp,
1194                         "structured reply already negotiated");
1195                 } else {
1196                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1197                     client->structured_reply = true;
1198                     myflags |= NBD_FLAG_SEND_DF;
1199                 }
1200                 break;
1201 
1202             case NBD_OPT_LIST_META_CONTEXT:
1203             case NBD_OPT_SET_META_CONTEXT:
1204                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1205                                                  errp);
1206                 break;
1207 
1208             default:
1209                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1210                                    "Unsupported option %" PRIu32 " (%s)",
1211                                    option, nbd_opt_lookup(option));
1212                 break;
1213             }
1214         } else {
1215             /*
1216              * If broken new-style we should drop the connection
1217              * for anything except NBD_OPT_EXPORT_NAME
1218              */
1219             switch (option) {
1220             case NBD_OPT_EXPORT_NAME:
1221                 return nbd_negotiate_handle_export_name(client,
1222                                                         myflags, no_zeroes,
1223                                                         errp);
1224 
1225             default:
1226                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1227                            option, nbd_opt_lookup(option));
1228                 return -EINVAL;
1229             }
1230         }
1231         if (ret < 0) {
1232             return ret;
1233         }
1234     }
1235 }
1236 
1237 /* nbd_negotiate
1238  * Return:
1239  * -errno  on error, errp is set
1240  * 0       on successful negotiation, errp is not set
1241  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1242  *         errp is not set
1243  */
1244 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1245 {
1246     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1247     int ret;
1248     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1249                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1250                               NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1251 
1252     /* Old style negotiation header, no room for options
1253         [ 0 ..   7]   passwd       ("NBDMAGIC")
1254         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1255         [16 ..  23]   size
1256         [24 ..  27]   export flags (zero-extended)
1257         [28 .. 151]   reserved     (0)
1258 
1259        New style negotiation header, client can send options
1260         [ 0 ..   7]   passwd       ("NBDMAGIC")
1261         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1262         [16 ..  17]   server flags (0)
1263         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1264      */
1265 
1266     qio_channel_set_blocking(client->ioc, false, NULL);
1267 
1268     trace_nbd_negotiate_begin();
1269     memcpy(buf, "NBDMAGIC", 8);
1270 
1271     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1272     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1273 
1274     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1275         error_prepend(errp, "write failed: ");
1276         return -EINVAL;
1277     }
1278     ret = nbd_negotiate_options(client, myflags, errp);
1279     if (ret != 0) {
1280         if (ret < 0) {
1281             error_prepend(errp, "option negotiation failed: ");
1282         }
1283         return ret;
1284     }
1285 
1286     assert(!client->optlen);
1287     trace_nbd_negotiate_success();
1288 
1289     return 0;
1290 }
1291 
1292 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1293                                Error **errp)
1294 {
1295     uint8_t buf[NBD_REQUEST_SIZE];
1296     uint32_t magic;
1297     int ret;
1298 
1299     ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1300     if (ret < 0) {
1301         return ret;
1302     }
1303 
1304     /* Request
1305        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1306        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1307        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1308        [ 8 .. 15]   handle
1309        [16 .. 23]   from
1310        [24 .. 27]   len
1311      */
1312 
1313     magic = ldl_be_p(buf);
1314     request->flags  = lduw_be_p(buf + 4);
1315     request->type   = lduw_be_p(buf + 6);
1316     request->handle = ldq_be_p(buf + 8);
1317     request->from   = ldq_be_p(buf + 16);
1318     request->len    = ldl_be_p(buf + 24);
1319 
1320     trace_nbd_receive_request(magic, request->flags, request->type,
1321                               request->from, request->len);
1322 
1323     if (magic != NBD_REQUEST_MAGIC) {
1324         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1325         return -EINVAL;
1326     }
1327     return 0;
1328 }
1329 
1330 #define MAX_NBD_REQUESTS 16
1331 
1332 void nbd_client_get(NBDClient *client)
1333 {
1334     client->refcount++;
1335 }
1336 
1337 void nbd_client_put(NBDClient *client)
1338 {
1339     if (--client->refcount == 0) {
1340         /* The last reference should be dropped by client->close,
1341          * which is called by client_close.
1342          */
1343         assert(client->closing);
1344 
1345         qio_channel_detach_aio_context(client->ioc);
1346         object_unref(OBJECT(client->sioc));
1347         object_unref(OBJECT(client->ioc));
1348         if (client->tlscreds) {
1349             object_unref(OBJECT(client->tlscreds));
1350         }
1351         g_free(client->tlsaclname);
1352         if (client->exp) {
1353             QTAILQ_REMOVE(&client->exp->clients, client, next);
1354             nbd_export_put(client->exp);
1355         }
1356         g_free(client);
1357     }
1358 }
1359 
1360 static void client_close(NBDClient *client, bool negotiated)
1361 {
1362     if (client->closing) {
1363         return;
1364     }
1365 
1366     client->closing = true;
1367 
1368     /* Force requests to finish.  They will drop their own references,
1369      * then we'll close the socket and free the NBDClient.
1370      */
1371     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1372                          NULL);
1373 
1374     /* Also tell the client, so that they release their reference.  */
1375     if (client->close_fn) {
1376         client->close_fn(client, negotiated);
1377     }
1378 }
1379 
1380 static NBDRequestData *nbd_request_get(NBDClient *client)
1381 {
1382     NBDRequestData *req;
1383 
1384     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1385     client->nb_requests++;
1386 
1387     req = g_new0(NBDRequestData, 1);
1388     nbd_client_get(client);
1389     req->client = client;
1390     return req;
1391 }
1392 
1393 static void nbd_request_put(NBDRequestData *req)
1394 {
1395     NBDClient *client = req->client;
1396 
1397     if (req->data) {
1398         qemu_vfree(req->data);
1399     }
1400     g_free(req);
1401 
1402     client->nb_requests--;
1403     nbd_client_receive_next_request(client);
1404 
1405     nbd_client_put(client);
1406 }
1407 
1408 static void blk_aio_attached(AioContext *ctx, void *opaque)
1409 {
1410     NBDExport *exp = opaque;
1411     NBDClient *client;
1412 
1413     trace_nbd_blk_aio_attached(exp->name, ctx);
1414 
1415     exp->ctx = ctx;
1416 
1417     QTAILQ_FOREACH(client, &exp->clients, next) {
1418         qio_channel_attach_aio_context(client->ioc, ctx);
1419         if (client->recv_coroutine) {
1420             aio_co_schedule(ctx, client->recv_coroutine);
1421         }
1422         if (client->send_coroutine) {
1423             aio_co_schedule(ctx, client->send_coroutine);
1424         }
1425     }
1426 }
1427 
1428 static void blk_aio_detach(void *opaque)
1429 {
1430     NBDExport *exp = opaque;
1431     NBDClient *client;
1432 
1433     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1434 
1435     QTAILQ_FOREACH(client, &exp->clients, next) {
1436         qio_channel_detach_aio_context(client->ioc);
1437     }
1438 
1439     exp->ctx = NULL;
1440 }
1441 
1442 static void nbd_eject_notifier(Notifier *n, void *data)
1443 {
1444     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1445     nbd_export_close(exp);
1446 }
1447 
1448 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1449                           uint64_t size, const char *name, const char *desc,
1450                           const char *bitmap, uint16_t nbdflags,
1451                           void (*close)(NBDExport *), bool writethrough,
1452                           BlockBackend *on_eject_blk, Error **errp)
1453 {
1454     AioContext *ctx;
1455     BlockBackend *blk;
1456     NBDExport *exp = g_new0(NBDExport, 1);
1457     uint64_t perm;
1458     int ret;
1459 
1460     /*
1461      * NBD exports are used for non-shared storage migration.  Make sure
1462      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1463      * access since the export could be available before migration handover.
1464      */
1465     assert(name);
1466     ctx = bdrv_get_aio_context(bs);
1467     aio_context_acquire(ctx);
1468     bdrv_invalidate_cache(bs, NULL);
1469     aio_context_release(ctx);
1470 
1471     /* Don't allow resize while the NBD server is running, otherwise we don't
1472      * care what happens with the node. */
1473     perm = BLK_PERM_CONSISTENT_READ;
1474     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1475         perm |= BLK_PERM_WRITE;
1476     }
1477     blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1478                         BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1479     ret = blk_insert_bs(blk, bs, errp);
1480     if (ret < 0) {
1481         goto fail;
1482     }
1483     blk_set_enable_write_cache(blk, !writethrough);
1484 
1485     exp->refcount = 1;
1486     QTAILQ_INIT(&exp->clients);
1487     exp->blk = blk;
1488     assert(dev_offset <= INT64_MAX);
1489     exp->dev_offset = dev_offset;
1490     exp->name = g_strdup(name);
1491     exp->description = g_strdup(desc);
1492     exp->nbdflags = nbdflags;
1493     assert(size <= INT64_MAX - dev_offset);
1494     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1495 
1496     if (bitmap) {
1497         BdrvDirtyBitmap *bm = NULL;
1498         BlockDriverState *bs = blk_bs(blk);
1499 
1500         while (true) {
1501             bm = bdrv_find_dirty_bitmap(bs, bitmap);
1502             if (bm != NULL || bs->backing == NULL) {
1503                 break;
1504             }
1505 
1506             bs = bs->backing->bs;
1507         }
1508 
1509         if (bm == NULL) {
1510             error_setg(errp, "Bitmap '%s' is not found", bitmap);
1511             goto fail;
1512         }
1513 
1514         if ((nbdflags & NBD_FLAG_READ_ONLY) && bdrv_is_writable(bs) &&
1515             bdrv_dirty_bitmap_enabled(bm)) {
1516             error_setg(errp,
1517                        "Enabled bitmap '%s' incompatible with readonly export",
1518                        bitmap);
1519             goto fail;
1520         }
1521 
1522         if (bdrv_dirty_bitmap_user_locked(bm)) {
1523             error_setg(errp, "Bitmap '%s' is in use", bitmap);
1524             goto fail;
1525         }
1526 
1527         bdrv_dirty_bitmap_set_qmp_locked(bm, true);
1528         exp->export_bitmap = bm;
1529         exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1530                                                      bitmap);
1531     }
1532 
1533     exp->close = close;
1534     exp->ctx = blk_get_aio_context(blk);
1535     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1536 
1537     if (on_eject_blk) {
1538         blk_ref(on_eject_blk);
1539         exp->eject_notifier_blk = on_eject_blk;
1540         exp->eject_notifier.notify = nbd_eject_notifier;
1541         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1542     }
1543     QTAILQ_INSERT_TAIL(&exports, exp, next);
1544     nbd_export_get(exp);
1545     return exp;
1546 
1547 fail:
1548     blk_unref(blk);
1549     g_free(exp->name);
1550     g_free(exp->description);
1551     g_free(exp);
1552     return NULL;
1553 }
1554 
1555 NBDExport *nbd_export_find(const char *name)
1556 {
1557     NBDExport *exp;
1558     QTAILQ_FOREACH(exp, &exports, next) {
1559         if (strcmp(name, exp->name) == 0) {
1560             return exp;
1561         }
1562     }
1563 
1564     return NULL;
1565 }
1566 
1567 void nbd_export_close(NBDExport *exp)
1568 {
1569     NBDClient *client, *next;
1570 
1571     nbd_export_get(exp);
1572     /*
1573      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1574      * close mode that stops advertising the export to new clients but
1575      * still permits existing clients to run to completion? Because of
1576      * that possibility, nbd_export_close() can be called more than
1577      * once on an export.
1578      */
1579     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1580         client_close(client, true);
1581     }
1582     if (exp->name) {
1583         nbd_export_put(exp);
1584         g_free(exp->name);
1585         exp->name = NULL;
1586         QTAILQ_REMOVE(&exports, exp, next);
1587     }
1588     g_free(exp->description);
1589     exp->description = NULL;
1590     nbd_export_put(exp);
1591 }
1592 
1593 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1594 {
1595     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1596         nbd_export_close(exp);
1597         return;
1598     }
1599 
1600     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1601 
1602     error_setg(errp, "export '%s' still in use", exp->name);
1603     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1604 }
1605 
1606 void nbd_export_get(NBDExport *exp)
1607 {
1608     assert(exp->refcount > 0);
1609     exp->refcount++;
1610 }
1611 
1612 void nbd_export_put(NBDExport *exp)
1613 {
1614     assert(exp->refcount > 0);
1615     if (exp->refcount == 1) {
1616         nbd_export_close(exp);
1617     }
1618 
1619     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1620      * if someone calls nbd_export_put() on named export not through
1621      * nbd_export_set_name() when refcount is 1. So, let's assert that
1622      * it is > 0.
1623      */
1624     assert(exp->refcount > 0);
1625     if (--exp->refcount == 0) {
1626         assert(exp->name == NULL);
1627         assert(exp->description == NULL);
1628 
1629         if (exp->close) {
1630             exp->close(exp);
1631         }
1632 
1633         if (exp->blk) {
1634             if (exp->eject_notifier_blk) {
1635                 notifier_remove(&exp->eject_notifier);
1636                 blk_unref(exp->eject_notifier_blk);
1637             }
1638             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1639                                             blk_aio_detach, exp);
1640             blk_unref(exp->blk);
1641             exp->blk = NULL;
1642         }
1643 
1644         if (exp->export_bitmap) {
1645             bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1646             g_free(exp->export_bitmap_context);
1647         }
1648 
1649         g_free(exp);
1650     }
1651 }
1652 
1653 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1654 {
1655     return exp->blk;
1656 }
1657 
1658 void nbd_export_close_all(void)
1659 {
1660     NBDExport *exp, *next;
1661 
1662     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1663         nbd_export_close(exp);
1664     }
1665 }
1666 
1667 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1668                                         unsigned niov, Error **errp)
1669 {
1670     int ret;
1671 
1672     g_assert(qemu_in_coroutine());
1673     qemu_co_mutex_lock(&client->send_lock);
1674     client->send_coroutine = qemu_coroutine_self();
1675 
1676     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1677 
1678     client->send_coroutine = NULL;
1679     qemu_co_mutex_unlock(&client->send_lock);
1680 
1681     return ret;
1682 }
1683 
1684 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1685                                        uint64_t handle)
1686 {
1687     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1688     stl_be_p(&reply->error, error);
1689     stq_be_p(&reply->handle, handle);
1690 }
1691 
1692 static int nbd_co_send_simple_reply(NBDClient *client,
1693                                     uint64_t handle,
1694                                     uint32_t error,
1695                                     void *data,
1696                                     size_t len,
1697                                     Error **errp)
1698 {
1699     NBDSimpleReply reply;
1700     int nbd_err = system_errno_to_nbd_errno(error);
1701     struct iovec iov[] = {
1702         {.iov_base = &reply, .iov_len = sizeof(reply)},
1703         {.iov_base = data, .iov_len = len}
1704     };
1705 
1706     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1707                                    len);
1708     set_be_simple_reply(&reply, nbd_err, handle);
1709 
1710     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1711 }
1712 
1713 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1714                                 uint16_t type, uint64_t handle, uint32_t length)
1715 {
1716     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1717     stw_be_p(&chunk->flags, flags);
1718     stw_be_p(&chunk->type, type);
1719     stq_be_p(&chunk->handle, handle);
1720     stl_be_p(&chunk->length, length);
1721 }
1722 
1723 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1724                                                     uint64_t handle,
1725                                                     Error **errp)
1726 {
1727     NBDStructuredReplyChunk chunk;
1728     struct iovec iov[] = {
1729         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1730     };
1731 
1732     trace_nbd_co_send_structured_done(handle);
1733     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1734 
1735     return nbd_co_send_iov(client, iov, 1, errp);
1736 }
1737 
1738 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1739                                                     uint64_t handle,
1740                                                     uint64_t offset,
1741                                                     void *data,
1742                                                     size_t size,
1743                                                     bool final,
1744                                                     Error **errp)
1745 {
1746     NBDStructuredReadData chunk;
1747     struct iovec iov[] = {
1748         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1749         {.iov_base = data, .iov_len = size}
1750     };
1751 
1752     assert(size);
1753     trace_nbd_co_send_structured_read(handle, offset, data, size);
1754     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1755                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1756                  sizeof(chunk) - sizeof(chunk.h) + size);
1757     stq_be_p(&chunk.offset, offset);
1758 
1759     return nbd_co_send_iov(client, iov, 2, errp);
1760 }
1761 
1762 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1763                                                      uint64_t handle,
1764                                                      uint32_t error,
1765                                                      const char *msg,
1766                                                      Error **errp)
1767 {
1768     NBDStructuredError chunk;
1769     int nbd_err = system_errno_to_nbd_errno(error);
1770     struct iovec iov[] = {
1771         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1772         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1773     };
1774 
1775     assert(nbd_err);
1776     trace_nbd_co_send_structured_error(handle, nbd_err,
1777                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1778     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1779                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1780     stl_be_p(&chunk.error, nbd_err);
1781     stw_be_p(&chunk.message_length, iov[1].iov_len);
1782 
1783     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1784 }
1785 
1786 /* Do a sparse read and send the structured reply to the client.
1787  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1788  * reported to the client, at which point this function succeeds.
1789  */
1790 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1791                                                 uint64_t handle,
1792                                                 uint64_t offset,
1793                                                 uint8_t *data,
1794                                                 size_t size,
1795                                                 Error **errp)
1796 {
1797     int ret = 0;
1798     NBDExport *exp = client->exp;
1799     size_t progress = 0;
1800 
1801     while (progress < size) {
1802         int64_t pnum;
1803         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1804                                              offset + progress,
1805                                              size - progress, &pnum, NULL,
1806                                              NULL);
1807         bool final;
1808 
1809         if (status < 0) {
1810             char *msg = g_strdup_printf("unable to check for holes: %s",
1811                                         strerror(-status));
1812 
1813             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1814                                                errp);
1815             g_free(msg);
1816             return ret;
1817         }
1818         assert(pnum && pnum <= size - progress);
1819         final = progress + pnum == size;
1820         if (status & BDRV_BLOCK_ZERO) {
1821             NBDStructuredReadHole chunk;
1822             struct iovec iov[] = {
1823                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1824             };
1825 
1826             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1827                                                    pnum);
1828             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1829                          NBD_REPLY_TYPE_OFFSET_HOLE,
1830                          handle, sizeof(chunk) - sizeof(chunk.h));
1831             stq_be_p(&chunk.offset, offset + progress);
1832             stl_be_p(&chunk.length, pnum);
1833             ret = nbd_co_send_iov(client, iov, 1, errp);
1834         } else {
1835             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1836                             data + progress, pnum);
1837             if (ret < 0) {
1838                 error_setg_errno(errp, -ret, "reading from file failed");
1839                 break;
1840             }
1841             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1842                                               data + progress, pnum, final,
1843                                               errp);
1844         }
1845 
1846         if (ret < 0) {
1847             break;
1848         }
1849         progress += pnum;
1850     }
1851     return ret;
1852 }
1853 
1854 /*
1855  * Populate @extents from block status. Update @bytes to be the actual
1856  * length encoded (which may be smaller than the original), and update
1857  * @nb_extents to the number of extents used.
1858  *
1859  * Returns zero on success and -errno on bdrv_block_status_above failure.
1860  */
1861 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1862                                   uint64_t *bytes, NBDExtent *extents,
1863                                   unsigned int *nb_extents)
1864 {
1865     uint64_t remaining_bytes = *bytes;
1866     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1867     bool first_extent = true;
1868 
1869     assert(*nb_extents);
1870     while (remaining_bytes) {
1871         uint32_t flags;
1872         int64_t num;
1873         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1874                                           &num, NULL, NULL);
1875 
1876         if (ret < 0) {
1877             return ret;
1878         }
1879 
1880         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1881                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1882         offset += num;
1883         remaining_bytes -= num;
1884 
1885         if (first_extent) {
1886             extent->flags = flags;
1887             extent->length = num;
1888             first_extent = false;
1889             continue;
1890         }
1891 
1892         if (flags == extent->flags) {
1893             /* extend current extent */
1894             extent->length += num;
1895         } else {
1896             if (extent + 1 == extents_end) {
1897                 break;
1898             }
1899 
1900             /* start new extent */
1901             extent++;
1902             extent->flags = flags;
1903             extent->length = num;
1904         }
1905     }
1906 
1907     extents_end = extent + 1;
1908 
1909     for (extent = extents; extent < extents_end; extent++) {
1910         extent->flags = cpu_to_be32(extent->flags);
1911         extent->length = cpu_to_be32(extent->length);
1912     }
1913 
1914     *bytes -= remaining_bytes;
1915     *nb_extents = extents_end - extents;
1916 
1917     return 0;
1918 }
1919 
1920 /* nbd_co_send_extents
1921  *
1922  * @length is only for tracing purposes (and may be smaller or larger
1923  * than the client's original request). @last controls whether
1924  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1925  * big-endian format.
1926  */
1927 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1928                                NBDExtent *extents, unsigned int nb_extents,
1929                                uint64_t length, bool last,
1930                                uint32_t context_id, Error **errp)
1931 {
1932     NBDStructuredMeta chunk;
1933 
1934     struct iovec iov[] = {
1935         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1936         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1937     };
1938 
1939     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1940     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1941                  NBD_REPLY_TYPE_BLOCK_STATUS,
1942                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1943     stl_be_p(&chunk.context_id, context_id);
1944 
1945     return nbd_co_send_iov(client, iov, 2, errp);
1946 }
1947 
1948 /* Get block status from the exported device and send it to the client */
1949 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1950                                     BlockDriverState *bs, uint64_t offset,
1951                                     uint32_t length, bool dont_fragment,
1952                                     bool last, uint32_t context_id,
1953                                     Error **errp)
1954 {
1955     int ret;
1956     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1957     NBDExtent *extents = g_new(NBDExtent, nb_extents);
1958     uint64_t final_length = length;
1959 
1960     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1961                                  &nb_extents);
1962     if (ret < 0) {
1963         g_free(extents);
1964         return nbd_co_send_structured_error(
1965                 client, handle, -ret, "can't get block status", errp);
1966     }
1967 
1968     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1969                               final_length, last, context_id, errp);
1970 
1971     g_free(extents);
1972 
1973     return ret;
1974 }
1975 
1976 /*
1977  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1978  * final extent may exceed the original @length. Store in @length the
1979  * byte length encoded (which may be smaller or larger than the
1980  * original), and return the number of extents used.
1981  */
1982 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1983                                       uint64_t *length, NBDExtent *extents,
1984                                       unsigned int nb_extents,
1985                                       bool dont_fragment)
1986 {
1987     uint64_t begin = offset, end = offset;
1988     uint64_t overall_end = offset + *length;
1989     unsigned int i = 0;
1990     BdrvDirtyBitmapIter *it;
1991     bool dirty;
1992 
1993     bdrv_dirty_bitmap_lock(bitmap);
1994 
1995     it = bdrv_dirty_iter_new(bitmap);
1996     dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1997 
1998     assert(begin < overall_end && nb_extents);
1999     while (begin < overall_end && i < nb_extents) {
2000         bool next_dirty = !dirty;
2001 
2002         if (dirty) {
2003             end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2004         } else {
2005             bdrv_set_dirty_iter(it, begin);
2006             end = bdrv_dirty_iter_next(it);
2007         }
2008         if (end == -1 || end - begin > UINT32_MAX) {
2009             /* Cap to an aligned value < 4G beyond begin. */
2010             end = MIN(bdrv_dirty_bitmap_size(bitmap),
2011                       begin + UINT32_MAX + 1 -
2012                       bdrv_dirty_bitmap_granularity(bitmap));
2013             next_dirty = dirty;
2014         }
2015         if (dont_fragment && end > overall_end) {
2016             end = overall_end;
2017         }
2018 
2019         extents[i].length = cpu_to_be32(end - begin);
2020         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2021         i++;
2022         begin = end;
2023         dirty = next_dirty;
2024     }
2025 
2026     bdrv_dirty_iter_free(it);
2027 
2028     bdrv_dirty_bitmap_unlock(bitmap);
2029 
2030     assert(offset < end);
2031     *length = end - offset;
2032     return i;
2033 }
2034 
2035 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2036                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2037                               uint32_t length, bool dont_fragment, bool last,
2038                               uint32_t context_id, Error **errp)
2039 {
2040     int ret;
2041     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2042     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2043     uint64_t final_length = length;
2044 
2045     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2046                                    nb_extents, dont_fragment);
2047 
2048     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2049                               final_length, last, context_id, errp);
2050 
2051     g_free(extents);
2052 
2053     return ret;
2054 }
2055 
2056 /* nbd_co_receive_request
2057  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2058  * connection right away, and any other negative value to report an error to
2059  * the client (although the caller may still need to disconnect after reporting
2060  * the error).
2061  */
2062 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2063                                   Error **errp)
2064 {
2065     NBDClient *client = req->client;
2066     int valid_flags;
2067 
2068     g_assert(qemu_in_coroutine());
2069     assert(client->recv_coroutine == qemu_coroutine_self());
2070     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2071         return -EIO;
2072     }
2073 
2074     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2075                                              nbd_cmd_lookup(request->type));
2076 
2077     if (request->type != NBD_CMD_WRITE) {
2078         /* No payload, we are ready to read the next request.  */
2079         req->complete = true;
2080     }
2081 
2082     if (request->type == NBD_CMD_DISC) {
2083         /* Special case: we're going to disconnect without a reply,
2084          * whether or not flags, from, or len are bogus */
2085         return -EIO;
2086     }
2087 
2088     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2089         request->type == NBD_CMD_CACHE)
2090     {
2091         if (request->len > NBD_MAX_BUFFER_SIZE) {
2092             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2093                        request->len, NBD_MAX_BUFFER_SIZE);
2094             return -EINVAL;
2095         }
2096 
2097         req->data = blk_try_blockalign(client->exp->blk, request->len);
2098         if (req->data == NULL) {
2099             error_setg(errp, "No memory");
2100             return -ENOMEM;
2101         }
2102     }
2103     if (request->type == NBD_CMD_WRITE) {
2104         if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2105                      errp) < 0)
2106         {
2107             return -EIO;
2108         }
2109         req->complete = true;
2110 
2111         trace_nbd_co_receive_request_payload_received(request->handle,
2112                                                       request->len);
2113     }
2114 
2115     /* Sanity checks. */
2116     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2117         (request->type == NBD_CMD_WRITE ||
2118          request->type == NBD_CMD_WRITE_ZEROES ||
2119          request->type == NBD_CMD_TRIM)) {
2120         error_setg(errp, "Export is read-only");
2121         return -EROFS;
2122     }
2123     if (request->from > client->exp->size ||
2124         request->len > client->exp->size - request->from) {
2125         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2126                    ", Size: %" PRIu64, request->from, request->len,
2127                    client->exp->size);
2128         return (request->type == NBD_CMD_WRITE ||
2129                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2130     }
2131     valid_flags = NBD_CMD_FLAG_FUA;
2132     if (request->type == NBD_CMD_READ && client->structured_reply) {
2133         valid_flags |= NBD_CMD_FLAG_DF;
2134     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2135         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2136     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2137         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2138     }
2139     if (request->flags & ~valid_flags) {
2140         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2141                    nbd_cmd_lookup(request->type), request->flags);
2142         return -EINVAL;
2143     }
2144 
2145     return 0;
2146 }
2147 
2148 /* Send simple reply without a payload, or a structured error
2149  * @error_msg is ignored if @ret >= 0
2150  * Returns 0 if connection is still live, -errno on failure to talk to client
2151  */
2152 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2153                                                uint64_t handle,
2154                                                int ret,
2155                                                const char *error_msg,
2156                                                Error **errp)
2157 {
2158     if (client->structured_reply && ret < 0) {
2159         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2160                                             errp);
2161     } else {
2162         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2163                                         NULL, 0, errp);
2164     }
2165 }
2166 
2167 /* Handle NBD_CMD_READ request.
2168  * Return -errno if sending fails. Other errors are reported directly to the
2169  * client as an error reply. */
2170 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2171                                         uint8_t *data, Error **errp)
2172 {
2173     int ret;
2174     NBDExport *exp = client->exp;
2175 
2176     assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2177 
2178     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2179     if (request->flags & NBD_CMD_FLAG_FUA) {
2180         ret = blk_co_flush(exp->blk);
2181         if (ret < 0) {
2182             return nbd_send_generic_reply(client, request->handle, ret,
2183                                           "flush failed", errp);
2184         }
2185     }
2186 
2187     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2188         request->len && request->type != NBD_CMD_CACHE)
2189     {
2190         return nbd_co_send_sparse_read(client, request->handle, request->from,
2191                                        data, request->len, errp);
2192     }
2193 
2194     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2195                     request->len);
2196     if (ret < 0 || request->type == NBD_CMD_CACHE) {
2197         return nbd_send_generic_reply(client, request->handle, ret,
2198                                       "reading from file failed", errp);
2199     }
2200 
2201     if (client->structured_reply) {
2202         if (request->len) {
2203             return nbd_co_send_structured_read(client, request->handle,
2204                                                request->from, data,
2205                                                request->len, true, errp);
2206         } else {
2207             return nbd_co_send_structured_done(client, request->handle, errp);
2208         }
2209     } else {
2210         return nbd_co_send_simple_reply(client, request->handle, 0,
2211                                         data, request->len, errp);
2212     }
2213 }
2214 
2215 /* Handle NBD request.
2216  * Return -errno if sending fails. Other errors are reported directly to the
2217  * client as an error reply. */
2218 static coroutine_fn int nbd_handle_request(NBDClient *client,
2219                                            NBDRequest *request,
2220                                            uint8_t *data, Error **errp)
2221 {
2222     int ret;
2223     int flags;
2224     NBDExport *exp = client->exp;
2225     char *msg;
2226 
2227     switch (request->type) {
2228     case NBD_CMD_READ:
2229     case NBD_CMD_CACHE:
2230         return nbd_do_cmd_read(client, request, data, errp);
2231 
2232     case NBD_CMD_WRITE:
2233         flags = 0;
2234         if (request->flags & NBD_CMD_FLAG_FUA) {
2235             flags |= BDRV_REQ_FUA;
2236         }
2237         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2238                          data, request->len, flags);
2239         return nbd_send_generic_reply(client, request->handle, ret,
2240                                       "writing to file failed", errp);
2241 
2242     case NBD_CMD_WRITE_ZEROES:
2243         flags = 0;
2244         if (request->flags & NBD_CMD_FLAG_FUA) {
2245             flags |= BDRV_REQ_FUA;
2246         }
2247         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2248             flags |= BDRV_REQ_MAY_UNMAP;
2249         }
2250         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2251                                 request->len, flags);
2252         return nbd_send_generic_reply(client, request->handle, ret,
2253                                       "writing to file failed", errp);
2254 
2255     case NBD_CMD_DISC:
2256         /* unreachable, thanks to special case in nbd_co_receive_request() */
2257         abort();
2258 
2259     case NBD_CMD_FLUSH:
2260         ret = blk_co_flush(exp->blk);
2261         return nbd_send_generic_reply(client, request->handle, ret,
2262                                       "flush failed", errp);
2263 
2264     case NBD_CMD_TRIM:
2265         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2266                               request->len);
2267         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2268             ret = blk_co_flush(exp->blk);
2269         }
2270         return nbd_send_generic_reply(client, request->handle, ret,
2271                                       "discard failed", errp);
2272 
2273     case NBD_CMD_BLOCK_STATUS:
2274         if (!request->len) {
2275             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2276                                           "need non-zero length", errp);
2277         }
2278         if (client->export_meta.valid &&
2279             (client->export_meta.base_allocation ||
2280              client->export_meta.bitmap))
2281         {
2282             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2283 
2284             if (client->export_meta.base_allocation) {
2285                 ret = nbd_co_send_block_status(client, request->handle,
2286                                                blk_bs(exp->blk), request->from,
2287                                                request->len, dont_fragment,
2288                                                !client->export_meta.bitmap,
2289                                                NBD_META_ID_BASE_ALLOCATION,
2290                                                errp);
2291                 if (ret < 0) {
2292                     return ret;
2293                 }
2294             }
2295 
2296             if (client->export_meta.bitmap) {
2297                 ret = nbd_co_send_bitmap(client, request->handle,
2298                                          client->exp->export_bitmap,
2299                                          request->from, request->len,
2300                                          dont_fragment,
2301                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2302                 if (ret < 0) {
2303                     return ret;
2304                 }
2305             }
2306 
2307             return ret;
2308         } else {
2309             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2310                                           "CMD_BLOCK_STATUS not negotiated",
2311                                           errp);
2312         }
2313 
2314     default:
2315         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2316                               request->type);
2317         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2318                                      errp);
2319         g_free(msg);
2320         return ret;
2321     }
2322 }
2323 
2324 /* Owns a reference to the NBDClient passed as opaque.  */
2325 static coroutine_fn void nbd_trip(void *opaque)
2326 {
2327     NBDClient *client = opaque;
2328     NBDRequestData *req;
2329     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2330     int ret;
2331     Error *local_err = NULL;
2332 
2333     trace_nbd_trip();
2334     if (client->closing) {
2335         nbd_client_put(client);
2336         return;
2337     }
2338 
2339     req = nbd_request_get(client);
2340     ret = nbd_co_receive_request(req, &request, &local_err);
2341     client->recv_coroutine = NULL;
2342 
2343     if (client->closing) {
2344         /*
2345          * The client may be closed when we are blocked in
2346          * nbd_co_receive_request()
2347          */
2348         goto done;
2349     }
2350 
2351     nbd_client_receive_next_request(client);
2352     if (ret == -EIO) {
2353         goto disconnect;
2354     }
2355 
2356     if (ret < 0) {
2357         /* It wans't -EIO, so, according to nbd_co_receive_request()
2358          * semantics, we should return the error to the client. */
2359         Error *export_err = local_err;
2360 
2361         local_err = NULL;
2362         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2363                                      error_get_pretty(export_err), &local_err);
2364         error_free(export_err);
2365     } else {
2366         ret = nbd_handle_request(client, &request, req->data, &local_err);
2367     }
2368     if (ret < 0) {
2369         error_prepend(&local_err, "Failed to send reply: ");
2370         goto disconnect;
2371     }
2372 
2373     /* We must disconnect after NBD_CMD_WRITE if we did not
2374      * read the payload.
2375      */
2376     if (!req->complete) {
2377         error_setg(&local_err, "Request handling failed in intermediate state");
2378         goto disconnect;
2379     }
2380 
2381 done:
2382     nbd_request_put(req);
2383     nbd_client_put(client);
2384     return;
2385 
2386 disconnect:
2387     if (local_err) {
2388         error_reportf_err(local_err, "Disconnect client, due to: ");
2389     }
2390     nbd_request_put(req);
2391     client_close(client, true);
2392     nbd_client_put(client);
2393 }
2394 
2395 static void nbd_client_receive_next_request(NBDClient *client)
2396 {
2397     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2398         nbd_client_get(client);
2399         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2400         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2401     }
2402 }
2403 
2404 static coroutine_fn void nbd_co_client_start(void *opaque)
2405 {
2406     NBDClient *client = opaque;
2407     Error *local_err = NULL;
2408 
2409     qemu_co_mutex_init(&client->send_lock);
2410 
2411     if (nbd_negotiate(client, &local_err)) {
2412         if (local_err) {
2413             error_report_err(local_err);
2414         }
2415         client_close(client, false);
2416         return;
2417     }
2418 
2419     nbd_client_receive_next_request(client);
2420 }
2421 
2422 /*
2423  * Create a new client listener using the given channel @sioc.
2424  * Begin servicing it in a coroutine.  When the connection closes, call
2425  * @close_fn with an indication of whether the client completed negotiation.
2426  */
2427 void nbd_client_new(QIOChannelSocket *sioc,
2428                     QCryptoTLSCreds *tlscreds,
2429                     const char *tlsaclname,
2430                     void (*close_fn)(NBDClient *, bool))
2431 {
2432     NBDClient *client;
2433     Coroutine *co;
2434 
2435     client = g_new0(NBDClient, 1);
2436     client->refcount = 1;
2437     client->tlscreds = tlscreds;
2438     if (tlscreds) {
2439         object_ref(OBJECT(client->tlscreds));
2440     }
2441     client->tlsaclname = g_strdup(tlsaclname);
2442     client->sioc = sioc;
2443     object_ref(OBJECT(client->sioc));
2444     client->ioc = QIO_CHANNEL(sioc);
2445     object_ref(OBJECT(client->ioc));
2446     client->close_fn = close_fn;
2447 
2448     co = qemu_coroutine_create(nbd_co_client_start, client);
2449     qemu_coroutine_enter(co);
2450 }
2451