xref: /openbmc/qemu/nbd/server.c (revision 9543fdaf)
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     off_t dev_offset;
81     off_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, errp) < 0) {
442         error_prepend(errp, "read failed: ");
443         return -EIO;
444     }
445     name[client->optlen] = '\0';
446     client->optlen = 0;
447 
448     trace_nbd_negotiate_handle_export_name_request(name);
449 
450     client->exp = nbd_export_find(name);
451     if (!client->exp) {
452         error_setg(errp, "export not found");
453         return -EINVAL;
454     }
455 
456     trace_nbd_negotiate_new_style_size_flags(client->exp->size,
457                                              client->exp->nbdflags | myflags);
458     stq_be_p(buf, client->exp->size);
459     stw_be_p(buf + 8, client->exp->nbdflags | myflags);
460     len = no_zeroes ? 10 : sizeof(buf);
461     ret = nbd_write(client->ioc, buf, len, errp);
462     if (ret < 0) {
463         error_prepend(errp, "write failed: ");
464         return ret;
465     }
466 
467     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
468     nbd_export_get(client->exp);
469     nbd_check_meta_export(client);
470 
471     return 0;
472 }
473 
474 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
475  * The buffer does NOT include the info type prefix.
476  * Return -errno on error, 0 if ready to send more. */
477 static int nbd_negotiate_send_info(NBDClient *client,
478                                    uint16_t info, uint32_t length, void *buf,
479                                    Error **errp)
480 {
481     int rc;
482 
483     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
484     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
485                                     sizeof(info) + length, errp);
486     if (rc < 0) {
487         return rc;
488     }
489     info = cpu_to_be16(info);
490     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
491         return -EIO;
492     }
493     if (nbd_write(client->ioc, buf, length, errp) < 0) {
494         return -EIO;
495     }
496     return 0;
497 }
498 
499 /* nbd_reject_length: Handle any unexpected payload.
500  * @fatal requests that we quit talking to the client, even if we are able
501  * to successfully send an error reply.
502  * Return:
503  * -errno  transmission error occurred or @fatal was requested, errp is set
504  * 0       error message successfully sent to client, errp is not set
505  */
506 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
507 {
508     int ret;
509 
510     assert(client->optlen);
511     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
512                           nbd_opt_lookup(client->opt));
513     if (fatal && !ret) {
514         error_setg(errp, "option '%s' has unexpected length",
515                    nbd_opt_lookup(client->opt));
516         return -EINVAL;
517     }
518     return ret;
519 }
520 
521 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
522  * Return -errno on error, 0 if ready for next option, and 1 to move
523  * into transmission phase.  */
524 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
525                                      Error **errp)
526 {
527     int rc;
528     char name[NBD_MAX_NAME_SIZE + 1];
529     NBDExport *exp;
530     uint16_t requests;
531     uint16_t request;
532     uint32_t namelen;
533     bool sendname = false;
534     bool blocksize = false;
535     uint32_t sizes[3];
536     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
537 
538     /* Client sends:
539         4 bytes: L, name length (can be 0)
540         L bytes: export name
541         2 bytes: N, number of requests (can be 0)
542         N * 2 bytes: N requests
543     */
544     rc = nbd_opt_read_name(client, name, &namelen, errp);
545     if (rc <= 0) {
546         return rc;
547     }
548     trace_nbd_negotiate_handle_export_name_request(name);
549 
550     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
551     if (rc <= 0) {
552         return rc;
553     }
554     requests = be16_to_cpu(requests);
555     trace_nbd_negotiate_handle_info_requests(requests);
556     while (requests--) {
557         rc = nbd_opt_read(client, &request, sizeof(request), errp);
558         if (rc <= 0) {
559             return rc;
560         }
561         request = be16_to_cpu(request);
562         trace_nbd_negotiate_handle_info_request(request,
563                                                 nbd_info_lookup(request));
564         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
565          * everything else is either a request we don't know or
566          * something we send regardless of request */
567         switch (request) {
568         case NBD_INFO_NAME:
569             sendname = true;
570             break;
571         case NBD_INFO_BLOCK_SIZE:
572             blocksize = true;
573             break;
574         }
575     }
576     if (client->optlen) {
577         return nbd_reject_length(client, false, errp);
578     }
579 
580     exp = nbd_export_find(name);
581     if (!exp) {
582         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
583                                           errp, "export '%s' not present",
584                                           name);
585     }
586 
587     /* Don't bother sending NBD_INFO_NAME unless client requested it */
588     if (sendname) {
589         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
590                                      errp);
591         if (rc < 0) {
592             return rc;
593         }
594     }
595 
596     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
597      * client request */
598     if (exp->description) {
599         size_t len = strlen(exp->description);
600 
601         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
602                                      len, exp->description, errp);
603         if (rc < 0) {
604             return rc;
605         }
606     }
607 
608     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
609      * according to whether the client requested it, and according to
610      * whether this is OPT_INFO or OPT_GO. */
611     /* minimum - 1 for back-compat, or 512 if client is new enough.
612      * TODO: consult blk_bs(blk)->bl.request_alignment? */
613     sizes[0] =
614             (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1;
615     /* preferred - Hard-code to 4096 for now.
616      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
617     sizes[1] = 4096;
618     /* maximum - At most 32M, but smaller as appropriate. */
619     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
620     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
621     sizes[0] = cpu_to_be32(sizes[0]);
622     sizes[1] = cpu_to_be32(sizes[1]);
623     sizes[2] = cpu_to_be32(sizes[2]);
624     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
625                                  sizeof(sizes), sizes, errp);
626     if (rc < 0) {
627         return rc;
628     }
629 
630     /* Send NBD_INFO_EXPORT always */
631     trace_nbd_negotiate_new_style_size_flags(exp->size,
632                                              exp->nbdflags | myflags);
633     stq_be_p(buf, exp->size);
634     stw_be_p(buf + 8, exp->nbdflags | myflags);
635     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
636                                  sizeof(buf), buf, errp);
637     if (rc < 0) {
638         return rc;
639     }
640 
641     /* If the client is just asking for NBD_OPT_INFO, but forgot to
642      * request block sizes, return an error.
643      * TODO: consult blk_bs(blk)->request_align, and only error if it
644      * is not 1? */
645     if (client->opt == NBD_OPT_INFO && !blocksize) {
646         return nbd_negotiate_send_rep_err(client,
647                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
648                                           errp,
649                                           "request NBD_INFO_BLOCK_SIZE to "
650                                           "use this export");
651     }
652 
653     /* Final reply */
654     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
655     if (rc < 0) {
656         return rc;
657     }
658 
659     if (client->opt == NBD_OPT_GO) {
660         client->exp = exp;
661         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
662         nbd_export_get(client->exp);
663         nbd_check_meta_export(client);
664         rc = 1;
665     }
666     return rc;
667 }
668 
669 
670 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
671  * new channel for all further (now-encrypted) communication. */
672 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
673                                                  Error **errp)
674 {
675     QIOChannel *ioc;
676     QIOChannelTLS *tioc;
677     struct NBDTLSHandshakeData data = { 0 };
678 
679     assert(client->opt == NBD_OPT_STARTTLS);
680 
681     trace_nbd_negotiate_handle_starttls();
682     ioc = client->ioc;
683 
684     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
685         return NULL;
686     }
687 
688     tioc = qio_channel_tls_new_server(ioc,
689                                       client->tlscreds,
690                                       client->tlsaclname,
691                                       errp);
692     if (!tioc) {
693         return NULL;
694     }
695 
696     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
697     trace_nbd_negotiate_handle_starttls_handshake();
698     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
699     qio_channel_tls_handshake(tioc,
700                               nbd_tls_handshake,
701                               &data,
702                               NULL,
703                               NULL);
704 
705     if (!data.complete) {
706         g_main_loop_run(data.loop);
707     }
708     g_main_loop_unref(data.loop);
709     if (data.error) {
710         object_unref(OBJECT(tioc));
711         error_propagate(errp, data.error);
712         return NULL;
713     }
714 
715     return QIO_CHANNEL(tioc);
716 }
717 
718 /* nbd_negotiate_send_meta_context
719  *
720  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
721  *
722  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
723  */
724 static int nbd_negotiate_send_meta_context(NBDClient *client,
725                                            const char *context,
726                                            uint32_t context_id,
727                                            Error **errp)
728 {
729     NBDOptionReplyMetaContext opt;
730     struct iovec iov[] = {
731         {.iov_base = &opt, .iov_len = sizeof(opt)},
732         {.iov_base = (void *)context, .iov_len = strlen(context)}
733     };
734 
735     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
736         context_id = 0;
737     }
738 
739     trace_nbd_negotiate_meta_query_reply(context, context_id);
740     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
741                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
742     stl_be_p(&opt.context_id, context_id);
743 
744     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
745 }
746 
747 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
748  * @match is never set to false.
749  *
750  * Return -errno on I/O error, 0 if option was completely handled by
751  * sending a reply about inconsistent lengths, or 1 on success.
752  *
753  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
754  * It only means that there are no errors.
755  */
756 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
757                             Error **errp)
758 {
759     int ret;
760     char *query;
761     size_t len = strlen(pattern);
762 
763     assert(len);
764 
765     query = g_malloc(len);
766     ret = nbd_opt_read(client, query, len, errp);
767     if (ret <= 0) {
768         g_free(query);
769         return ret;
770     }
771 
772     if (strncmp(query, pattern, len) == 0) {
773         trace_nbd_negotiate_meta_query_parse(pattern);
774         *match = true;
775     } else {
776         trace_nbd_negotiate_meta_query_skip("pattern not matched");
777     }
778     g_free(query);
779 
780     return 1;
781 }
782 
783 /*
784  * Read @len bytes, and set @match to true if they match @pattern, or if @len
785  * is 0 and the client is performing _LIST_. @match is never set to false.
786  *
787  * Return -errno on I/O error, 0 if option was completely handled by
788  * sending a reply about inconsistent lengths, or 1 on success.
789  *
790  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
791  * It only means that there are no errors.
792  */
793 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
794                                      uint32_t len, bool *match, Error **errp)
795 {
796     if (len == 0) {
797         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
798             *match = true;
799         }
800         trace_nbd_negotiate_meta_query_parse("empty");
801         return 1;
802     }
803 
804     if (len != strlen(pattern)) {
805         trace_nbd_negotiate_meta_query_skip("different lengths");
806         return nbd_opt_skip(client, len, errp);
807     }
808 
809     return nbd_meta_pattern(client, pattern, match, errp);
810 }
811 
812 /* nbd_meta_base_query
813  *
814  * Handle queries to 'base' namespace. For now, only the base:allocation
815  * context is available.  'len' is the amount of text remaining to be read from
816  * the current name, after the 'base:' portion has been stripped.
817  *
818  * Return -errno on I/O error, 0 if option was completely handled by
819  * sending a reply about inconsistent lengths, or 1 on success.
820  */
821 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
822                                uint32_t len, Error **errp)
823 {
824     return nbd_meta_empty_or_pattern(client, "allocation", len,
825                                      &meta->base_allocation, errp);
826 }
827 
828 /* nbd_meta_bitmap_query
829  *
830  * Handle query to 'qemu:' namespace.
831  * @len is the amount of text remaining to be read from the current name, after
832  * the 'qemu:' portion has been stripped.
833  *
834  * Return -errno on I/O error, 0 if option was completely handled by
835  * sending a reply about inconsistent lengths, or 1 on success. */
836 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
837                                uint32_t len, Error **errp)
838 {
839     bool dirty_bitmap = false;
840     size_t dirty_bitmap_len = strlen("dirty-bitmap:");
841     int ret;
842 
843     if (!meta->exp->export_bitmap) {
844         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
845         return nbd_opt_skip(client, len, errp);
846     }
847 
848     if (len == 0) {
849         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
850             meta->bitmap = true;
851         }
852         trace_nbd_negotiate_meta_query_parse("empty");
853         return 1;
854     }
855 
856     if (len < dirty_bitmap_len) {
857         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
858         return nbd_opt_skip(client, len, errp);
859     }
860 
861     len -= dirty_bitmap_len;
862     ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
863     if (ret <= 0) {
864         return ret;
865     }
866     if (!dirty_bitmap) {
867         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
868         return nbd_opt_skip(client, len, errp);
869     }
870 
871     trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
872 
873     return nbd_meta_empty_or_pattern(
874             client, meta->exp->export_bitmap_context +
875             strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
876 }
877 
878 /* nbd_negotiate_meta_query
879  *
880  * Parse namespace name and call corresponding function to parse body of the
881  * query.
882  *
883  * The only supported namespace now is 'base'.
884  *
885  * The function aims not wasting time and memory to read long unknown namespace
886  * names.
887  *
888  * Return -errno on I/O error, 0 if option was completely handled by
889  * sending a reply about inconsistent lengths, or 1 on success. */
890 static int nbd_negotiate_meta_query(NBDClient *client,
891                                     NBDExportMetaContexts *meta, Error **errp)
892 {
893     /*
894      * Both 'qemu' and 'base' namespaces have length = 5 including a
895      * colon. If another length namespace is later introduced, this
896      * should certainly be refactored.
897      */
898     int ret;
899     size_t ns_len = 5;
900     char ns[5];
901     uint32_t len;
902 
903     ret = nbd_opt_read(client, &len, sizeof(len), errp);
904     if (ret <= 0) {
905         return ret;
906     }
907     len = cpu_to_be32(len);
908 
909     if (len < ns_len) {
910         trace_nbd_negotiate_meta_query_skip("length too short");
911         return nbd_opt_skip(client, len, errp);
912     }
913 
914     len -= ns_len;
915     ret = nbd_opt_read(client, ns, ns_len, errp);
916     if (ret <= 0) {
917         return ret;
918     }
919 
920     if (!strncmp(ns, "base:", ns_len)) {
921         trace_nbd_negotiate_meta_query_parse("base:");
922         return nbd_meta_base_query(client, meta, len, errp);
923     } else if (!strncmp(ns, "qemu:", ns_len)) {
924         trace_nbd_negotiate_meta_query_parse("qemu:");
925         return nbd_meta_qemu_query(client, meta, len, errp);
926     }
927 
928     trace_nbd_negotiate_meta_query_skip("unknown namespace");
929     return nbd_opt_skip(client, len, errp);
930 }
931 
932 /* nbd_negotiate_meta_queries
933  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
934  *
935  * Return -errno on I/O error, or 0 if option was completely handled. */
936 static int nbd_negotiate_meta_queries(NBDClient *client,
937                                       NBDExportMetaContexts *meta, Error **errp)
938 {
939     int ret;
940     char export_name[NBD_MAX_NAME_SIZE + 1];
941     NBDExportMetaContexts local_meta;
942     uint32_t nb_queries;
943     int i;
944 
945     if (!client->structured_reply) {
946         return nbd_opt_invalid(client, errp,
947                                "request option '%s' when structured reply "
948                                "is not negotiated",
949                                nbd_opt_lookup(client->opt));
950     }
951 
952     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
953         /* Only change the caller's meta on SET. */
954         meta = &local_meta;
955     }
956 
957     memset(meta, 0, sizeof(*meta));
958 
959     ret = nbd_opt_read_name(client, export_name, NULL, errp);
960     if (ret <= 0) {
961         return ret;
962     }
963 
964     meta->exp = nbd_export_find(export_name);
965     if (meta->exp == NULL) {
966         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
967                             "export '%s' not present", export_name);
968     }
969 
970     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
971     if (ret <= 0) {
972         return ret;
973     }
974     nb_queries = cpu_to_be32(nb_queries);
975     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
976                                      export_name, nb_queries);
977 
978     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
979         /* enable all known contexts */
980         meta->base_allocation = true;
981         meta->bitmap = !!meta->exp->export_bitmap;
982     } else {
983         for (i = 0; i < nb_queries; ++i) {
984             ret = nbd_negotiate_meta_query(client, meta, errp);
985             if (ret <= 0) {
986                 return ret;
987             }
988         }
989     }
990 
991     if (meta->base_allocation) {
992         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
993                                               NBD_META_ID_BASE_ALLOCATION,
994                                               errp);
995         if (ret < 0) {
996             return ret;
997         }
998     }
999 
1000     if (meta->bitmap) {
1001         ret = nbd_negotiate_send_meta_context(client,
1002                                               meta->exp->export_bitmap_context,
1003                                               NBD_META_ID_DIRTY_BITMAP,
1004                                               errp);
1005         if (ret < 0) {
1006             return ret;
1007         }
1008     }
1009 
1010     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1011     if (ret == 0) {
1012         meta->valid = true;
1013     }
1014 
1015     return ret;
1016 }
1017 
1018 /* nbd_negotiate_options
1019  * Process all NBD_OPT_* client option commands, during fixed newstyle
1020  * negotiation.
1021  * Return:
1022  * -errno  on error, errp is set
1023  * 0       on successful negotiation, errp is not set
1024  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1025  *         errp is not set
1026  */
1027 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1028                                  Error **errp)
1029 {
1030     uint32_t flags;
1031     bool fixedNewstyle = false;
1032     bool no_zeroes = false;
1033 
1034     /* Client sends:
1035         [ 0 ..   3]   client flags
1036 
1037        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1038         [ 0 ..   7]   NBD_OPTS_MAGIC
1039         [ 8 ..  11]   NBD option
1040         [12 ..  15]   Data length
1041         ...           Rest of request
1042 
1043         [ 0 ..   7]   NBD_OPTS_MAGIC
1044         [ 8 ..  11]   Second NBD option
1045         [12 ..  15]   Data length
1046         ...           Rest of request
1047     */
1048 
1049     if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
1050         error_prepend(errp, "read failed: ");
1051         return -EIO;
1052     }
1053     flags = be32_to_cpu(flags);
1054     trace_nbd_negotiate_options_flags(flags);
1055     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1056         fixedNewstyle = true;
1057         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1058     }
1059     if (flags & NBD_FLAG_C_NO_ZEROES) {
1060         no_zeroes = true;
1061         flags &= ~NBD_FLAG_C_NO_ZEROES;
1062     }
1063     if (flags != 0) {
1064         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1065         return -EINVAL;
1066     }
1067 
1068     while (1) {
1069         int ret;
1070         uint32_t option, length;
1071         uint64_t magic;
1072 
1073         if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
1074             error_prepend(errp, "read failed: ");
1075             return -EINVAL;
1076         }
1077         magic = be64_to_cpu(magic);
1078         trace_nbd_negotiate_options_check_magic(magic);
1079         if (magic != NBD_OPTS_MAGIC) {
1080             error_setg(errp, "Bad magic received");
1081             return -EINVAL;
1082         }
1083 
1084         if (nbd_read(client->ioc, &option,
1085                      sizeof(option), errp) < 0) {
1086             error_prepend(errp, "read failed: ");
1087             return -EINVAL;
1088         }
1089         option = be32_to_cpu(option);
1090         client->opt = option;
1091 
1092         if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
1093             error_prepend(errp, "read failed: ");
1094             return -EINVAL;
1095         }
1096         length = be32_to_cpu(length);
1097         assert(!client->optlen);
1098         client->optlen = length;
1099 
1100         if (length > NBD_MAX_BUFFER_SIZE) {
1101             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1102                        length, NBD_MAX_BUFFER_SIZE);
1103             return -EINVAL;
1104         }
1105 
1106         trace_nbd_negotiate_options_check_option(option,
1107                                                  nbd_opt_lookup(option));
1108         if (client->tlscreds &&
1109             client->ioc == (QIOChannel *)client->sioc) {
1110             QIOChannel *tioc;
1111             if (!fixedNewstyle) {
1112                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1113                 return -EINVAL;
1114             }
1115             switch (option) {
1116             case NBD_OPT_STARTTLS:
1117                 if (length) {
1118                     /* Unconditionally drop the connection if the client
1119                      * can't start a TLS negotiation correctly */
1120                     return nbd_reject_length(client, true, errp);
1121                 }
1122                 tioc = nbd_negotiate_handle_starttls(client, errp);
1123                 if (!tioc) {
1124                     return -EIO;
1125                 }
1126                 ret = 0;
1127                 object_unref(OBJECT(client->ioc));
1128                 client->ioc = QIO_CHANNEL(tioc);
1129                 break;
1130 
1131             case NBD_OPT_EXPORT_NAME:
1132                 /* No way to return an error to client, so drop connection */
1133                 error_setg(errp, "Option 0x%x not permitted before TLS",
1134                            option);
1135                 return -EINVAL;
1136 
1137             default:
1138                 /* Let the client keep trying, unless they asked to
1139                  * quit. Always try to give an error back to the
1140                  * client; but when replying to OPT_ABORT, be aware
1141                  * that the client may hang up before receiving the
1142                  * error, in which case we are fine ignoring the
1143                  * resulting EPIPE. */
1144                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1145                                    option == NBD_OPT_ABORT ? NULL : errp,
1146                                    "Option 0x%" PRIx32
1147                                    " not permitted before TLS", option);
1148                 if (option == NBD_OPT_ABORT) {
1149                     return 1;
1150                 }
1151                 break;
1152             }
1153         } else if (fixedNewstyle) {
1154             switch (option) {
1155             case NBD_OPT_LIST:
1156                 if (length) {
1157                     ret = nbd_reject_length(client, false, errp);
1158                 } else {
1159                     ret = nbd_negotiate_handle_list(client, errp);
1160                 }
1161                 break;
1162 
1163             case NBD_OPT_ABORT:
1164                 /* NBD spec says we must try to reply before
1165                  * disconnecting, but that we must also tolerate
1166                  * guests that don't wait for our reply. */
1167                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1168                 return 1;
1169 
1170             case NBD_OPT_EXPORT_NAME:
1171                 return nbd_negotiate_handle_export_name(client,
1172                                                         myflags, no_zeroes,
1173                                                         errp);
1174 
1175             case NBD_OPT_INFO:
1176             case NBD_OPT_GO:
1177                 ret = nbd_negotiate_handle_info(client, myflags, errp);
1178                 if (ret == 1) {
1179                     assert(option == NBD_OPT_GO);
1180                     return 0;
1181                 }
1182                 break;
1183 
1184             case NBD_OPT_STARTTLS:
1185                 if (length) {
1186                     ret = nbd_reject_length(client, false, errp);
1187                 } else if (client->tlscreds) {
1188                     ret = nbd_negotiate_send_rep_err(client,
1189                                                      NBD_REP_ERR_INVALID, errp,
1190                                                      "TLS already enabled");
1191                 } else {
1192                     ret = nbd_negotiate_send_rep_err(client,
1193                                                      NBD_REP_ERR_POLICY, errp,
1194                                                      "TLS not configured");
1195                 }
1196                 break;
1197 
1198             case NBD_OPT_STRUCTURED_REPLY:
1199                 if (length) {
1200                     ret = nbd_reject_length(client, false, errp);
1201                 } else if (client->structured_reply) {
1202                     ret = nbd_negotiate_send_rep_err(
1203                         client, NBD_REP_ERR_INVALID, errp,
1204                         "structured reply already negotiated");
1205                 } else {
1206                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1207                     client->structured_reply = true;
1208                     myflags |= NBD_FLAG_SEND_DF;
1209                 }
1210                 break;
1211 
1212             case NBD_OPT_LIST_META_CONTEXT:
1213             case NBD_OPT_SET_META_CONTEXT:
1214                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1215                                                  errp);
1216                 break;
1217 
1218             default:
1219                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1220                                    "Unsupported option %" PRIu32 " (%s)",
1221                                    option, nbd_opt_lookup(option));
1222                 break;
1223             }
1224         } else {
1225             /*
1226              * If broken new-style we should drop the connection
1227              * for anything except NBD_OPT_EXPORT_NAME
1228              */
1229             switch (option) {
1230             case NBD_OPT_EXPORT_NAME:
1231                 return nbd_negotiate_handle_export_name(client,
1232                                                         myflags, no_zeroes,
1233                                                         errp);
1234 
1235             default:
1236                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1237                            option, nbd_opt_lookup(option));
1238                 return -EINVAL;
1239             }
1240         }
1241         if (ret < 0) {
1242             return ret;
1243         }
1244     }
1245 }
1246 
1247 /* nbd_negotiate
1248  * Return:
1249  * -errno  on error, errp is set
1250  * 0       on successful negotiation, errp is not set
1251  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1252  *         errp is not set
1253  */
1254 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1255 {
1256     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1257     int ret;
1258     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1259                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1260                               NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1261 
1262     /* Old style negotiation header, no room for options
1263         [ 0 ..   7]   passwd       ("NBDMAGIC")
1264         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1265         [16 ..  23]   size
1266         [24 ..  27]   export flags (zero-extended)
1267         [28 .. 151]   reserved     (0)
1268 
1269        New style negotiation header, client can send options
1270         [ 0 ..   7]   passwd       ("NBDMAGIC")
1271         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1272         [16 ..  17]   server flags (0)
1273         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1274      */
1275 
1276     qio_channel_set_blocking(client->ioc, false, NULL);
1277 
1278     trace_nbd_negotiate_begin();
1279     memcpy(buf, "NBDMAGIC", 8);
1280 
1281     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1282     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1283 
1284     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1285         error_prepend(errp, "write failed: ");
1286         return -EINVAL;
1287     }
1288     ret = nbd_negotiate_options(client, myflags, errp);
1289     if (ret != 0) {
1290         if (ret < 0) {
1291             error_prepend(errp, "option negotiation failed: ");
1292         }
1293         return ret;
1294     }
1295 
1296     assert(!client->optlen);
1297     trace_nbd_negotiate_success();
1298 
1299     return 0;
1300 }
1301 
1302 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1303                                Error **errp)
1304 {
1305     uint8_t buf[NBD_REQUEST_SIZE];
1306     uint32_t magic;
1307     int ret;
1308 
1309     ret = nbd_read(ioc, buf, sizeof(buf), errp);
1310     if (ret < 0) {
1311         return ret;
1312     }
1313 
1314     /* Request
1315        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1316        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1317        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1318        [ 8 .. 15]   handle
1319        [16 .. 23]   from
1320        [24 .. 27]   len
1321      */
1322 
1323     magic = ldl_be_p(buf);
1324     request->flags  = lduw_be_p(buf + 4);
1325     request->type   = lduw_be_p(buf + 6);
1326     request->handle = ldq_be_p(buf + 8);
1327     request->from   = ldq_be_p(buf + 16);
1328     request->len    = ldl_be_p(buf + 24);
1329 
1330     trace_nbd_receive_request(magic, request->flags, request->type,
1331                               request->from, request->len);
1332 
1333     if (magic != NBD_REQUEST_MAGIC) {
1334         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1335         return -EINVAL;
1336     }
1337     return 0;
1338 }
1339 
1340 #define MAX_NBD_REQUESTS 16
1341 
1342 void nbd_client_get(NBDClient *client)
1343 {
1344     client->refcount++;
1345 }
1346 
1347 void nbd_client_put(NBDClient *client)
1348 {
1349     if (--client->refcount == 0) {
1350         /* The last reference should be dropped by client->close,
1351          * which is called by client_close.
1352          */
1353         assert(client->closing);
1354 
1355         qio_channel_detach_aio_context(client->ioc);
1356         object_unref(OBJECT(client->sioc));
1357         object_unref(OBJECT(client->ioc));
1358         if (client->tlscreds) {
1359             object_unref(OBJECT(client->tlscreds));
1360         }
1361         g_free(client->tlsaclname);
1362         if (client->exp) {
1363             QTAILQ_REMOVE(&client->exp->clients, client, next);
1364             nbd_export_put(client->exp);
1365         }
1366         g_free(client);
1367     }
1368 }
1369 
1370 static void client_close(NBDClient *client, bool negotiated)
1371 {
1372     if (client->closing) {
1373         return;
1374     }
1375 
1376     client->closing = true;
1377 
1378     /* Force requests to finish.  They will drop their own references,
1379      * then we'll close the socket and free the NBDClient.
1380      */
1381     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1382                          NULL);
1383 
1384     /* Also tell the client, so that they release their reference.  */
1385     if (client->close_fn) {
1386         client->close_fn(client, negotiated);
1387     }
1388 }
1389 
1390 static NBDRequestData *nbd_request_get(NBDClient *client)
1391 {
1392     NBDRequestData *req;
1393 
1394     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1395     client->nb_requests++;
1396 
1397     req = g_new0(NBDRequestData, 1);
1398     nbd_client_get(client);
1399     req->client = client;
1400     return req;
1401 }
1402 
1403 static void nbd_request_put(NBDRequestData *req)
1404 {
1405     NBDClient *client = req->client;
1406 
1407     if (req->data) {
1408         qemu_vfree(req->data);
1409     }
1410     g_free(req);
1411 
1412     client->nb_requests--;
1413     nbd_client_receive_next_request(client);
1414 
1415     nbd_client_put(client);
1416 }
1417 
1418 static void blk_aio_attached(AioContext *ctx, void *opaque)
1419 {
1420     NBDExport *exp = opaque;
1421     NBDClient *client;
1422 
1423     trace_nbd_blk_aio_attached(exp->name, ctx);
1424 
1425     exp->ctx = ctx;
1426 
1427     QTAILQ_FOREACH(client, &exp->clients, next) {
1428         qio_channel_attach_aio_context(client->ioc, ctx);
1429         if (client->recv_coroutine) {
1430             aio_co_schedule(ctx, client->recv_coroutine);
1431         }
1432         if (client->send_coroutine) {
1433             aio_co_schedule(ctx, client->send_coroutine);
1434         }
1435     }
1436 }
1437 
1438 static void blk_aio_detach(void *opaque)
1439 {
1440     NBDExport *exp = opaque;
1441     NBDClient *client;
1442 
1443     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1444 
1445     QTAILQ_FOREACH(client, &exp->clients, next) {
1446         qio_channel_detach_aio_context(client->ioc);
1447     }
1448 
1449     exp->ctx = NULL;
1450 }
1451 
1452 static void nbd_eject_notifier(Notifier *n, void *data)
1453 {
1454     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1455     nbd_export_close(exp);
1456 }
1457 
1458 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1459                           uint16_t nbdflags, void (*close)(NBDExport *),
1460                           bool writethrough, BlockBackend *on_eject_blk,
1461                           Error **errp)
1462 {
1463     AioContext *ctx;
1464     BlockBackend *blk;
1465     NBDExport *exp = g_new0(NBDExport, 1);
1466     uint64_t perm;
1467     int ret;
1468 
1469     /*
1470      * NBD exports are used for non-shared storage migration.  Make sure
1471      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1472      * access since the export could be available before migration handover.
1473      */
1474     ctx = bdrv_get_aio_context(bs);
1475     aio_context_acquire(ctx);
1476     bdrv_invalidate_cache(bs, NULL);
1477     aio_context_release(ctx);
1478 
1479     /* Don't allow resize while the NBD server is running, otherwise we don't
1480      * care what happens with the node. */
1481     perm = BLK_PERM_CONSISTENT_READ;
1482     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1483         perm |= BLK_PERM_WRITE;
1484     }
1485     blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1486                         BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1487     ret = blk_insert_bs(blk, bs, errp);
1488     if (ret < 0) {
1489         goto fail;
1490     }
1491     blk_set_enable_write_cache(blk, !writethrough);
1492 
1493     exp->refcount = 1;
1494     QTAILQ_INIT(&exp->clients);
1495     exp->blk = blk;
1496     exp->dev_offset = dev_offset;
1497     exp->nbdflags = nbdflags;
1498     exp->size = size < 0 ? blk_getlength(blk) : size;
1499     if (exp->size < 0) {
1500         error_setg_errno(errp, -exp->size,
1501                          "Failed to determine the NBD export's length");
1502         goto fail;
1503     }
1504     exp->size -= exp->size % BDRV_SECTOR_SIZE;
1505 
1506     exp->close = close;
1507     exp->ctx = blk_get_aio_context(blk);
1508     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1509 
1510     if (on_eject_blk) {
1511         blk_ref(on_eject_blk);
1512         exp->eject_notifier_blk = on_eject_blk;
1513         exp->eject_notifier.notify = nbd_eject_notifier;
1514         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1515     }
1516     return exp;
1517 
1518 fail:
1519     blk_unref(blk);
1520     g_free(exp);
1521     return NULL;
1522 }
1523 
1524 NBDExport *nbd_export_find(const char *name)
1525 {
1526     NBDExport *exp;
1527     QTAILQ_FOREACH(exp, &exports, next) {
1528         if (strcmp(name, exp->name) == 0) {
1529             return exp;
1530         }
1531     }
1532 
1533     return NULL;
1534 }
1535 
1536 void nbd_export_set_name(NBDExport *exp, const char *name)
1537 {
1538     if (exp->name == name) {
1539         return;
1540     }
1541 
1542     nbd_export_get(exp);
1543     if (exp->name != NULL) {
1544         g_free(exp->name);
1545         exp->name = NULL;
1546         QTAILQ_REMOVE(&exports, exp, next);
1547         nbd_export_put(exp);
1548     }
1549     if (name != NULL) {
1550         nbd_export_get(exp);
1551         exp->name = g_strdup(name);
1552         QTAILQ_INSERT_TAIL(&exports, exp, next);
1553     }
1554     nbd_export_put(exp);
1555 }
1556 
1557 void nbd_export_set_description(NBDExport *exp, const char *description)
1558 {
1559     g_free(exp->description);
1560     exp->description = g_strdup(description);
1561 }
1562 
1563 void nbd_export_close(NBDExport *exp)
1564 {
1565     NBDClient *client, *next;
1566 
1567     nbd_export_get(exp);
1568     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1569         client_close(client, true);
1570     }
1571     nbd_export_set_name(exp, NULL);
1572     nbd_export_set_description(exp, NULL);
1573     nbd_export_put(exp);
1574 }
1575 
1576 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1577 {
1578     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1579         nbd_export_close(exp);
1580         return;
1581     }
1582 
1583     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1584 
1585     error_setg(errp, "export '%s' still in use", exp->name);
1586     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1587 }
1588 
1589 void nbd_export_get(NBDExport *exp)
1590 {
1591     assert(exp->refcount > 0);
1592     exp->refcount++;
1593 }
1594 
1595 void nbd_export_put(NBDExport *exp)
1596 {
1597     assert(exp->refcount > 0);
1598     if (exp->refcount == 1) {
1599         nbd_export_close(exp);
1600     }
1601 
1602     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1603      * if someone calls nbd_export_put() on named export not through
1604      * nbd_export_set_name() when refcount is 1. So, let's assert that
1605      * it is > 0.
1606      */
1607     assert(exp->refcount > 0);
1608     if (--exp->refcount == 0) {
1609         assert(exp->name == NULL);
1610         assert(exp->description == NULL);
1611 
1612         if (exp->close) {
1613             exp->close(exp);
1614         }
1615 
1616         if (exp->blk) {
1617             if (exp->eject_notifier_blk) {
1618                 notifier_remove(&exp->eject_notifier);
1619                 blk_unref(exp->eject_notifier_blk);
1620             }
1621             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1622                                             blk_aio_detach, exp);
1623             blk_unref(exp->blk);
1624             exp->blk = NULL;
1625         }
1626 
1627         if (exp->export_bitmap) {
1628             bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1629             g_free(exp->export_bitmap_context);
1630         }
1631 
1632         g_free(exp);
1633     }
1634 }
1635 
1636 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1637 {
1638     return exp->blk;
1639 }
1640 
1641 void nbd_export_close_all(void)
1642 {
1643     NBDExport *exp, *next;
1644 
1645     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1646         nbd_export_close(exp);
1647     }
1648 }
1649 
1650 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1651                                         unsigned niov, Error **errp)
1652 {
1653     int ret;
1654 
1655     g_assert(qemu_in_coroutine());
1656     qemu_co_mutex_lock(&client->send_lock);
1657     client->send_coroutine = qemu_coroutine_self();
1658 
1659     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1660 
1661     client->send_coroutine = NULL;
1662     qemu_co_mutex_unlock(&client->send_lock);
1663 
1664     return ret;
1665 }
1666 
1667 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1668                                        uint64_t handle)
1669 {
1670     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1671     stl_be_p(&reply->error, error);
1672     stq_be_p(&reply->handle, handle);
1673 }
1674 
1675 static int nbd_co_send_simple_reply(NBDClient *client,
1676                                     uint64_t handle,
1677                                     uint32_t error,
1678                                     void *data,
1679                                     size_t len,
1680                                     Error **errp)
1681 {
1682     NBDSimpleReply reply;
1683     int nbd_err = system_errno_to_nbd_errno(error);
1684     struct iovec iov[] = {
1685         {.iov_base = &reply, .iov_len = sizeof(reply)},
1686         {.iov_base = data, .iov_len = len}
1687     };
1688 
1689     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1690                                    len);
1691     set_be_simple_reply(&reply, nbd_err, handle);
1692 
1693     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1694 }
1695 
1696 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1697                                 uint16_t type, uint64_t handle, uint32_t length)
1698 {
1699     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1700     stw_be_p(&chunk->flags, flags);
1701     stw_be_p(&chunk->type, type);
1702     stq_be_p(&chunk->handle, handle);
1703     stl_be_p(&chunk->length, length);
1704 }
1705 
1706 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1707                                                     uint64_t handle,
1708                                                     Error **errp)
1709 {
1710     NBDStructuredReplyChunk chunk;
1711     struct iovec iov[] = {
1712         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1713     };
1714 
1715     trace_nbd_co_send_structured_done(handle);
1716     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1717 
1718     return nbd_co_send_iov(client, iov, 1, errp);
1719 }
1720 
1721 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1722                                                     uint64_t handle,
1723                                                     uint64_t offset,
1724                                                     void *data,
1725                                                     size_t size,
1726                                                     bool final,
1727                                                     Error **errp)
1728 {
1729     NBDStructuredReadData chunk;
1730     struct iovec iov[] = {
1731         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1732         {.iov_base = data, .iov_len = size}
1733     };
1734 
1735     assert(size);
1736     trace_nbd_co_send_structured_read(handle, offset, data, size);
1737     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1738                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1739                  sizeof(chunk) - sizeof(chunk.h) + size);
1740     stq_be_p(&chunk.offset, offset);
1741 
1742     return nbd_co_send_iov(client, iov, 2, errp);
1743 }
1744 
1745 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1746                                                      uint64_t handle,
1747                                                      uint32_t error,
1748                                                      const char *msg,
1749                                                      Error **errp)
1750 {
1751     NBDStructuredError chunk;
1752     int nbd_err = system_errno_to_nbd_errno(error);
1753     struct iovec iov[] = {
1754         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1755         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1756     };
1757 
1758     assert(nbd_err);
1759     trace_nbd_co_send_structured_error(handle, nbd_err,
1760                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1761     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1762                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1763     stl_be_p(&chunk.error, nbd_err);
1764     stw_be_p(&chunk.message_length, iov[1].iov_len);
1765 
1766     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1767 }
1768 
1769 /* Do a sparse read and send the structured reply to the client.
1770  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1771  * reported to the client, at which point this function succeeds.
1772  */
1773 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1774                                                 uint64_t handle,
1775                                                 uint64_t offset,
1776                                                 uint8_t *data,
1777                                                 size_t size,
1778                                                 Error **errp)
1779 {
1780     int ret = 0;
1781     NBDExport *exp = client->exp;
1782     size_t progress = 0;
1783 
1784     while (progress < size) {
1785         int64_t pnum;
1786         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1787                                              offset + progress,
1788                                              size - progress, &pnum, NULL,
1789                                              NULL);
1790         bool final;
1791 
1792         if (status < 0) {
1793             char *msg = g_strdup_printf("unable to check for holes: %s",
1794                                         strerror(-status));
1795 
1796             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1797                                                errp);
1798             g_free(msg);
1799             return ret;
1800         }
1801         assert(pnum && pnum <= size - progress);
1802         final = progress + pnum == size;
1803         if (status & BDRV_BLOCK_ZERO) {
1804             NBDStructuredReadHole chunk;
1805             struct iovec iov[] = {
1806                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1807             };
1808 
1809             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1810                                                    pnum);
1811             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1812                          NBD_REPLY_TYPE_OFFSET_HOLE,
1813                          handle, sizeof(chunk) - sizeof(chunk.h));
1814             stq_be_p(&chunk.offset, offset + progress);
1815             stl_be_p(&chunk.length, pnum);
1816             ret = nbd_co_send_iov(client, iov, 1, errp);
1817         } else {
1818             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1819                             data + progress, pnum);
1820             if (ret < 0) {
1821                 error_setg_errno(errp, -ret, "reading from file failed");
1822                 break;
1823             }
1824             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1825                                               data + progress, pnum, final,
1826                                               errp);
1827         }
1828 
1829         if (ret < 0) {
1830             break;
1831         }
1832         progress += pnum;
1833     }
1834     return ret;
1835 }
1836 
1837 /*
1838  * Populate @extents from block status. Update @bytes to be the actual
1839  * length encoded (which may be smaller than the original), and update
1840  * @nb_extents to the number of extents used.
1841  *
1842  * Returns zero on success and -errno on bdrv_block_status_above failure.
1843  */
1844 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1845                                   uint64_t *bytes, NBDExtent *extents,
1846                                   unsigned int *nb_extents)
1847 {
1848     uint64_t remaining_bytes = *bytes;
1849     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1850     bool first_extent = true;
1851 
1852     assert(*nb_extents);
1853     while (remaining_bytes) {
1854         uint32_t flags;
1855         int64_t num;
1856         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1857                                           &num, NULL, NULL);
1858 
1859         if (ret < 0) {
1860             return ret;
1861         }
1862 
1863         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1864                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1865         offset += num;
1866         remaining_bytes -= num;
1867 
1868         if (first_extent) {
1869             extent->flags = flags;
1870             extent->length = num;
1871             first_extent = false;
1872             continue;
1873         }
1874 
1875         if (flags == extent->flags) {
1876             /* extend current extent */
1877             extent->length += num;
1878         } else {
1879             if (extent + 1 == extents_end) {
1880                 break;
1881             }
1882 
1883             /* start new extent */
1884             extent++;
1885             extent->flags = flags;
1886             extent->length = num;
1887         }
1888     }
1889 
1890     extents_end = extent + 1;
1891 
1892     for (extent = extents; extent < extents_end; extent++) {
1893         extent->flags = cpu_to_be32(extent->flags);
1894         extent->length = cpu_to_be32(extent->length);
1895     }
1896 
1897     *bytes -= remaining_bytes;
1898     *nb_extents = extents_end - extents;
1899 
1900     return 0;
1901 }
1902 
1903 /* nbd_co_send_extents
1904  *
1905  * @length is only for tracing purposes (and may be smaller or larger
1906  * than the client's original request). @last controls whether
1907  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1908  * big-endian format.
1909  */
1910 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1911                                NBDExtent *extents, unsigned int nb_extents,
1912                                uint64_t length, bool last,
1913                                uint32_t context_id, Error **errp)
1914 {
1915     NBDStructuredMeta chunk;
1916 
1917     struct iovec iov[] = {
1918         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1919         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1920     };
1921 
1922     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1923     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1924                  NBD_REPLY_TYPE_BLOCK_STATUS,
1925                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1926     stl_be_p(&chunk.context_id, context_id);
1927 
1928     return nbd_co_send_iov(client, iov, 2, errp);
1929 }
1930 
1931 /* Get block status from the exported device and send it to the client */
1932 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1933                                     BlockDriverState *bs, uint64_t offset,
1934                                     uint32_t length, bool dont_fragment,
1935                                     bool last, uint32_t context_id,
1936                                     Error **errp)
1937 {
1938     int ret;
1939     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1940     NBDExtent *extents = g_new(NBDExtent, nb_extents);
1941     uint64_t final_length = length;
1942 
1943     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1944                                  &nb_extents);
1945     if (ret < 0) {
1946         g_free(extents);
1947         return nbd_co_send_structured_error(
1948                 client, handle, -ret, "can't get block status", errp);
1949     }
1950 
1951     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1952                               final_length, last, context_id, errp);
1953 
1954     g_free(extents);
1955 
1956     return ret;
1957 }
1958 
1959 /*
1960  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1961  * final extent may exceed the original @length. Store in @length the
1962  * byte length encoded (which may be smaller or larger than the
1963  * original), and return the number of extents used.
1964  */
1965 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1966                                       uint64_t *length, NBDExtent *extents,
1967                                       unsigned int nb_extents,
1968                                       bool dont_fragment)
1969 {
1970     uint64_t begin = offset, end = offset;
1971     uint64_t overall_end = offset + *length;
1972     unsigned int i = 0;
1973     BdrvDirtyBitmapIter *it;
1974     bool dirty;
1975 
1976     bdrv_dirty_bitmap_lock(bitmap);
1977 
1978     it = bdrv_dirty_iter_new(bitmap);
1979     dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1980 
1981     assert(begin < overall_end && nb_extents);
1982     while (begin < overall_end && i < nb_extents) {
1983         bool next_dirty = !dirty;
1984 
1985         if (dirty) {
1986             end = bdrv_dirty_bitmap_next_zero(bitmap, begin);
1987         } else {
1988             bdrv_set_dirty_iter(it, begin);
1989             end = bdrv_dirty_iter_next(it);
1990         }
1991         if (end == -1 || end - begin > UINT32_MAX) {
1992             /* Cap to an aligned value < 4G beyond begin. */
1993             end = MIN(bdrv_dirty_bitmap_size(bitmap),
1994                       begin + UINT32_MAX + 1 -
1995                       bdrv_dirty_bitmap_granularity(bitmap));
1996             next_dirty = dirty;
1997         }
1998         if (dont_fragment && end > overall_end) {
1999             end = overall_end;
2000         }
2001 
2002         extents[i].length = cpu_to_be32(end - begin);
2003         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2004         i++;
2005         begin = end;
2006         dirty = next_dirty;
2007     }
2008 
2009     bdrv_dirty_iter_free(it);
2010 
2011     bdrv_dirty_bitmap_unlock(bitmap);
2012 
2013     assert(offset < end);
2014     *length = end - offset;
2015     return i;
2016 }
2017 
2018 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2019                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2020                               uint32_t length, bool dont_fragment, bool last,
2021                               uint32_t context_id, Error **errp)
2022 {
2023     int ret;
2024     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2025     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2026     uint64_t final_length = length;
2027 
2028     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2029                                    nb_extents, dont_fragment);
2030 
2031     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2032                               final_length, last, context_id, errp);
2033 
2034     g_free(extents);
2035 
2036     return ret;
2037 }
2038 
2039 /* nbd_co_receive_request
2040  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2041  * connection right away, and any other negative value to report an error to
2042  * the client (although the caller may still need to disconnect after reporting
2043  * the error).
2044  */
2045 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2046                                   Error **errp)
2047 {
2048     NBDClient *client = req->client;
2049     int valid_flags;
2050 
2051     g_assert(qemu_in_coroutine());
2052     assert(client->recv_coroutine == qemu_coroutine_self());
2053     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2054         return -EIO;
2055     }
2056 
2057     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2058                                              nbd_cmd_lookup(request->type));
2059 
2060     if (request->type != NBD_CMD_WRITE) {
2061         /* No payload, we are ready to read the next request.  */
2062         req->complete = true;
2063     }
2064 
2065     if (request->type == NBD_CMD_DISC) {
2066         /* Special case: we're going to disconnect without a reply,
2067          * whether or not flags, from, or len are bogus */
2068         return -EIO;
2069     }
2070 
2071     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2072         request->type == NBD_CMD_CACHE)
2073     {
2074         if (request->len > NBD_MAX_BUFFER_SIZE) {
2075             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2076                        request->len, NBD_MAX_BUFFER_SIZE);
2077             return -EINVAL;
2078         }
2079 
2080         req->data = blk_try_blockalign(client->exp->blk, request->len);
2081         if (req->data == NULL) {
2082             error_setg(errp, "No memory");
2083             return -ENOMEM;
2084         }
2085     }
2086     if (request->type == NBD_CMD_WRITE) {
2087         if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
2088             error_prepend(errp, "reading from socket failed: ");
2089             return -EIO;
2090         }
2091         req->complete = true;
2092 
2093         trace_nbd_co_receive_request_payload_received(request->handle,
2094                                                       request->len);
2095     }
2096 
2097     /* Sanity checks. */
2098     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2099         (request->type == NBD_CMD_WRITE ||
2100          request->type == NBD_CMD_WRITE_ZEROES ||
2101          request->type == NBD_CMD_TRIM)) {
2102         error_setg(errp, "Export is read-only");
2103         return -EROFS;
2104     }
2105     if (request->from > client->exp->size ||
2106         request->from + request->len > client->exp->size) {
2107         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2108                    ", Size: %" PRIu64, request->from, request->len,
2109                    (uint64_t)client->exp->size);
2110         return (request->type == NBD_CMD_WRITE ||
2111                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2112     }
2113     valid_flags = NBD_CMD_FLAG_FUA;
2114     if (request->type == NBD_CMD_READ && client->structured_reply) {
2115         valid_flags |= NBD_CMD_FLAG_DF;
2116     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2117         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2118     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2119         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2120     }
2121     if (request->flags & ~valid_flags) {
2122         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2123                    nbd_cmd_lookup(request->type), request->flags);
2124         return -EINVAL;
2125     }
2126 
2127     return 0;
2128 }
2129 
2130 /* Send simple reply without a payload, or a structured error
2131  * @error_msg is ignored if @ret >= 0
2132  * Returns 0 if connection is still live, -errno on failure to talk to client
2133  */
2134 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2135                                                uint64_t handle,
2136                                                int ret,
2137                                                const char *error_msg,
2138                                                Error **errp)
2139 {
2140     if (client->structured_reply && ret < 0) {
2141         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2142                                             errp);
2143     } else {
2144         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2145                                         NULL, 0, errp);
2146     }
2147 }
2148 
2149 /* Handle NBD_CMD_READ request.
2150  * Return -errno if sending fails. Other errors are reported directly to the
2151  * client as an error reply. */
2152 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2153                                         uint8_t *data, Error **errp)
2154 {
2155     int ret;
2156     NBDExport *exp = client->exp;
2157 
2158     assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2159 
2160     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2161     if (request->flags & NBD_CMD_FLAG_FUA) {
2162         ret = blk_co_flush(exp->blk);
2163         if (ret < 0) {
2164             return nbd_send_generic_reply(client, request->handle, ret,
2165                                           "flush failed", errp);
2166         }
2167     }
2168 
2169     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2170         request->len && request->type != NBD_CMD_CACHE)
2171     {
2172         return nbd_co_send_sparse_read(client, request->handle, request->from,
2173                                        data, request->len, errp);
2174     }
2175 
2176     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2177                     request->len);
2178     if (ret < 0 || request->type == NBD_CMD_CACHE) {
2179         return nbd_send_generic_reply(client, request->handle, ret,
2180                                       "reading from file failed", errp);
2181     }
2182 
2183     if (client->structured_reply) {
2184         if (request->len) {
2185             return nbd_co_send_structured_read(client, request->handle,
2186                                                request->from, data,
2187                                                request->len, true, errp);
2188         } else {
2189             return nbd_co_send_structured_done(client, request->handle, errp);
2190         }
2191     } else {
2192         return nbd_co_send_simple_reply(client, request->handle, 0,
2193                                         data, request->len, errp);
2194     }
2195 }
2196 
2197 /* Handle NBD request.
2198  * Return -errno if sending fails. Other errors are reported directly to the
2199  * client as an error reply. */
2200 static coroutine_fn int nbd_handle_request(NBDClient *client,
2201                                            NBDRequest *request,
2202                                            uint8_t *data, Error **errp)
2203 {
2204     int ret;
2205     int flags;
2206     NBDExport *exp = client->exp;
2207     char *msg;
2208 
2209     switch (request->type) {
2210     case NBD_CMD_READ:
2211     case NBD_CMD_CACHE:
2212         return nbd_do_cmd_read(client, request, data, errp);
2213 
2214     case NBD_CMD_WRITE:
2215         flags = 0;
2216         if (request->flags & NBD_CMD_FLAG_FUA) {
2217             flags |= BDRV_REQ_FUA;
2218         }
2219         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2220                          data, request->len, flags);
2221         return nbd_send_generic_reply(client, request->handle, ret,
2222                                       "writing to file failed", errp);
2223 
2224     case NBD_CMD_WRITE_ZEROES:
2225         flags = 0;
2226         if (request->flags & NBD_CMD_FLAG_FUA) {
2227             flags |= BDRV_REQ_FUA;
2228         }
2229         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2230             flags |= BDRV_REQ_MAY_UNMAP;
2231         }
2232         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2233                                 request->len, flags);
2234         return nbd_send_generic_reply(client, request->handle, ret,
2235                                       "writing to file failed", errp);
2236 
2237     case NBD_CMD_DISC:
2238         /* unreachable, thanks to special case in nbd_co_receive_request() */
2239         abort();
2240 
2241     case NBD_CMD_FLUSH:
2242         ret = blk_co_flush(exp->blk);
2243         return nbd_send_generic_reply(client, request->handle, ret,
2244                                       "flush failed", errp);
2245 
2246     case NBD_CMD_TRIM:
2247         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2248                               request->len);
2249         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2250             ret = blk_co_flush(exp->blk);
2251         }
2252         return nbd_send_generic_reply(client, request->handle, ret,
2253                                       "discard failed", errp);
2254 
2255     case NBD_CMD_BLOCK_STATUS:
2256         if (!request->len) {
2257             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2258                                           "need non-zero length", errp);
2259         }
2260         if (client->export_meta.valid &&
2261             (client->export_meta.base_allocation ||
2262              client->export_meta.bitmap))
2263         {
2264             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2265 
2266             if (client->export_meta.base_allocation) {
2267                 ret = nbd_co_send_block_status(client, request->handle,
2268                                                blk_bs(exp->blk), request->from,
2269                                                request->len, dont_fragment,
2270                                                !client->export_meta.bitmap,
2271                                                NBD_META_ID_BASE_ALLOCATION,
2272                                                errp);
2273                 if (ret < 0) {
2274                     return ret;
2275                 }
2276             }
2277 
2278             if (client->export_meta.bitmap) {
2279                 ret = nbd_co_send_bitmap(client, request->handle,
2280                                          client->exp->export_bitmap,
2281                                          request->from, request->len,
2282                                          dont_fragment,
2283                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2284                 if (ret < 0) {
2285                     return ret;
2286                 }
2287             }
2288 
2289             return ret;
2290         } else {
2291             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2292                                           "CMD_BLOCK_STATUS not negotiated",
2293                                           errp);
2294         }
2295 
2296     default:
2297         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2298                               request->type);
2299         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2300                                      errp);
2301         g_free(msg);
2302         return ret;
2303     }
2304 }
2305 
2306 /* Owns a reference to the NBDClient passed as opaque.  */
2307 static coroutine_fn void nbd_trip(void *opaque)
2308 {
2309     NBDClient *client = opaque;
2310     NBDRequestData *req;
2311     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2312     int ret;
2313     Error *local_err = NULL;
2314 
2315     trace_nbd_trip();
2316     if (client->closing) {
2317         nbd_client_put(client);
2318         return;
2319     }
2320 
2321     req = nbd_request_get(client);
2322     ret = nbd_co_receive_request(req, &request, &local_err);
2323     client->recv_coroutine = NULL;
2324 
2325     if (client->closing) {
2326         /*
2327          * The client may be closed when we are blocked in
2328          * nbd_co_receive_request()
2329          */
2330         goto done;
2331     }
2332 
2333     nbd_client_receive_next_request(client);
2334     if (ret == -EIO) {
2335         goto disconnect;
2336     }
2337 
2338     if (ret < 0) {
2339         /* It wans't -EIO, so, according to nbd_co_receive_request()
2340          * semantics, we should return the error to the client. */
2341         Error *export_err = local_err;
2342 
2343         local_err = NULL;
2344         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2345                                      error_get_pretty(export_err), &local_err);
2346         error_free(export_err);
2347     } else {
2348         ret = nbd_handle_request(client, &request, req->data, &local_err);
2349     }
2350     if (ret < 0) {
2351         error_prepend(&local_err, "Failed to send reply: ");
2352         goto disconnect;
2353     }
2354 
2355     /* We must disconnect after NBD_CMD_WRITE if we did not
2356      * read the payload.
2357      */
2358     if (!req->complete) {
2359         error_setg(&local_err, "Request handling failed in intermediate state");
2360         goto disconnect;
2361     }
2362 
2363 done:
2364     nbd_request_put(req);
2365     nbd_client_put(client);
2366     return;
2367 
2368 disconnect:
2369     if (local_err) {
2370         error_reportf_err(local_err, "Disconnect client, due to: ");
2371     }
2372     nbd_request_put(req);
2373     client_close(client, true);
2374     nbd_client_put(client);
2375 }
2376 
2377 static void nbd_client_receive_next_request(NBDClient *client)
2378 {
2379     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2380         nbd_client_get(client);
2381         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2382         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2383     }
2384 }
2385 
2386 static coroutine_fn void nbd_co_client_start(void *opaque)
2387 {
2388     NBDClient *client = opaque;
2389     Error *local_err = NULL;
2390 
2391     qemu_co_mutex_init(&client->send_lock);
2392 
2393     if (nbd_negotiate(client, &local_err)) {
2394         if (local_err) {
2395             error_report_err(local_err);
2396         }
2397         client_close(client, false);
2398         return;
2399     }
2400 
2401     nbd_client_receive_next_request(client);
2402 }
2403 
2404 /*
2405  * Create a new client listener using the given channel @sioc.
2406  * Begin servicing it in a coroutine.  When the connection closes, call
2407  * @close_fn with an indication of whether the client completed negotiation.
2408  */
2409 void nbd_client_new(QIOChannelSocket *sioc,
2410                     QCryptoTLSCreds *tlscreds,
2411                     const char *tlsaclname,
2412                     void (*close_fn)(NBDClient *, bool))
2413 {
2414     NBDClient *client;
2415     Coroutine *co;
2416 
2417     client = g_new0(NBDClient, 1);
2418     client->refcount = 1;
2419     client->tlscreds = tlscreds;
2420     if (tlscreds) {
2421         object_ref(OBJECT(client->tlscreds));
2422     }
2423     client->tlsaclname = g_strdup(tlsaclname);
2424     client->sioc = sioc;
2425     object_ref(OBJECT(client->sioc));
2426     client->ioc = QIO_CHANNEL(sioc);
2427     object_ref(OBJECT(client->ioc));
2428     client->close_fn = close_fn;
2429 
2430     co = qemu_coroutine_create(nbd_co_client_start, client);
2431     qemu_coroutine_enter(co);
2432 }
2433 
2434 void nbd_export_bitmap(NBDExport *exp, const char *bitmap,
2435                        const char *bitmap_export_name, Error **errp)
2436 {
2437     BdrvDirtyBitmap *bm = NULL;
2438     BlockDriverState *bs = blk_bs(exp->blk);
2439 
2440     if (exp->export_bitmap) {
2441         error_setg(errp, "Export bitmap is already set");
2442         return;
2443     }
2444 
2445     while (true) {
2446         bm = bdrv_find_dirty_bitmap(bs, bitmap);
2447         if (bm != NULL || bs->backing == NULL) {
2448             break;
2449         }
2450 
2451         bs = bs->backing->bs;
2452     }
2453 
2454     if (bm == NULL) {
2455         error_setg(errp, "Bitmap '%s' is not found", bitmap);
2456         return;
2457     }
2458 
2459     if (bdrv_dirty_bitmap_enabled(bm)) {
2460         error_setg(errp, "Bitmap '%s' is enabled", bitmap);
2461         return;
2462     }
2463 
2464     if (bdrv_dirty_bitmap_user_locked(bm)) {
2465         error_setg(errp, "Bitmap '%s' is in use", bitmap);
2466         return;
2467     }
2468 
2469     bdrv_dirty_bitmap_set_qmp_locked(bm, true);
2470     exp->export_bitmap = bm;
2471     exp->export_bitmap_context =
2472             g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name);
2473 }
2474