xref: /openbmc/qemu/nbd/server.c (revision 4a9b31b8)
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     } 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_read(client->ioc, &flags, sizeof(flags), errp) < 0) {
1049         error_prepend(errp, "read failed: ");
1050         return -EIO;
1051     }
1052     flags = be32_to_cpu(flags);
1053     trace_nbd_negotiate_options_flags(flags);
1054     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1055         fixedNewstyle = true;
1056         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1057     }
1058     if (flags & NBD_FLAG_C_NO_ZEROES) {
1059         no_zeroes = true;
1060         flags &= ~NBD_FLAG_C_NO_ZEROES;
1061     }
1062     if (flags != 0) {
1063         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1064         return -EINVAL;
1065     }
1066 
1067     while (1) {
1068         int ret;
1069         uint32_t option, length;
1070         uint64_t magic;
1071 
1072         if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) {
1073             error_prepend(errp, "read failed: ");
1074             return -EINVAL;
1075         }
1076         magic = be64_to_cpu(magic);
1077         trace_nbd_negotiate_options_check_magic(magic);
1078         if (magic != NBD_OPTS_MAGIC) {
1079             error_setg(errp, "Bad magic received");
1080             return -EINVAL;
1081         }
1082 
1083         if (nbd_read(client->ioc, &option,
1084                      sizeof(option), errp) < 0) {
1085             error_prepend(errp, "read failed: ");
1086             return -EINVAL;
1087         }
1088         option = be32_to_cpu(option);
1089         client->opt = option;
1090 
1091         if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) {
1092             error_prepend(errp, "read failed: ");
1093             return -EINVAL;
1094         }
1095         length = be32_to_cpu(length);
1096         assert(!client->optlen);
1097         client->optlen = length;
1098 
1099         if (length > NBD_MAX_BUFFER_SIZE) {
1100             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1101                        length, NBD_MAX_BUFFER_SIZE);
1102             return -EINVAL;
1103         }
1104 
1105         trace_nbd_negotiate_options_check_option(option,
1106                                                  nbd_opt_lookup(option));
1107         if (client->tlscreds &&
1108             client->ioc == (QIOChannel *)client->sioc) {
1109             QIOChannel *tioc;
1110             if (!fixedNewstyle) {
1111                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1112                 return -EINVAL;
1113             }
1114             switch (option) {
1115             case NBD_OPT_STARTTLS:
1116                 if (length) {
1117                     /* Unconditionally drop the connection if the client
1118                      * can't start a TLS negotiation correctly */
1119                     return nbd_reject_length(client, true, errp);
1120                 }
1121                 tioc = nbd_negotiate_handle_starttls(client, errp);
1122                 if (!tioc) {
1123                     return -EIO;
1124                 }
1125                 ret = 0;
1126                 object_unref(OBJECT(client->ioc));
1127                 client->ioc = QIO_CHANNEL(tioc);
1128                 break;
1129 
1130             case NBD_OPT_EXPORT_NAME:
1131                 /* No way to return an error to client, so drop connection */
1132                 error_setg(errp, "Option 0x%x not permitted before TLS",
1133                            option);
1134                 return -EINVAL;
1135 
1136             default:
1137                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, errp,
1138                                    "Option 0x%" PRIx32
1139                                    "not permitted before TLS", option);
1140                 /* Let the client keep trying, unless they asked to
1141                  * quit. In this mode, we've already sent an error, so
1142                  * we can't ack the abort.  */
1143                 if (option == NBD_OPT_ABORT) {
1144                     return 1;
1145                 }
1146                 break;
1147             }
1148         } else if (fixedNewstyle) {
1149             switch (option) {
1150             case NBD_OPT_LIST:
1151                 if (length) {
1152                     ret = nbd_reject_length(client, false, errp);
1153                 } else {
1154                     ret = nbd_negotiate_handle_list(client, errp);
1155                 }
1156                 break;
1157 
1158             case NBD_OPT_ABORT:
1159                 /* NBD spec says we must try to reply before
1160                  * disconnecting, but that we must also tolerate
1161                  * guests that don't wait for our reply. */
1162                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1163                 return 1;
1164 
1165             case NBD_OPT_EXPORT_NAME:
1166                 return nbd_negotiate_handle_export_name(client,
1167                                                         myflags, no_zeroes,
1168                                                         errp);
1169 
1170             case NBD_OPT_INFO:
1171             case NBD_OPT_GO:
1172                 ret = nbd_negotiate_handle_info(client, myflags, errp);
1173                 if (ret == 1) {
1174                     assert(option == NBD_OPT_GO);
1175                     return 0;
1176                 }
1177                 break;
1178 
1179             case NBD_OPT_STARTTLS:
1180                 if (length) {
1181                     ret = nbd_reject_length(client, false, errp);
1182                 } else if (client->tlscreds) {
1183                     ret = nbd_negotiate_send_rep_err(client,
1184                                                      NBD_REP_ERR_INVALID, errp,
1185                                                      "TLS already enabled");
1186                 } else {
1187                     ret = nbd_negotiate_send_rep_err(client,
1188                                                      NBD_REP_ERR_POLICY, errp,
1189                                                      "TLS not configured");
1190                 }
1191                 break;
1192 
1193             case NBD_OPT_STRUCTURED_REPLY:
1194                 if (length) {
1195                     ret = nbd_reject_length(client, false, errp);
1196                 } else if (client->structured_reply) {
1197                     ret = nbd_negotiate_send_rep_err(
1198                         client, NBD_REP_ERR_INVALID, errp,
1199                         "structured reply already negotiated");
1200                 } else {
1201                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1202                     client->structured_reply = true;
1203                     myflags |= NBD_FLAG_SEND_DF;
1204                 }
1205                 break;
1206 
1207             case NBD_OPT_LIST_META_CONTEXT:
1208             case NBD_OPT_SET_META_CONTEXT:
1209                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1210                                                  errp);
1211                 break;
1212 
1213             default:
1214                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1215                                    "Unsupported option %" PRIu32 " (%s)",
1216                                    option, nbd_opt_lookup(option));
1217                 break;
1218             }
1219         } else {
1220             /*
1221              * If broken new-style we should drop the connection
1222              * for anything except NBD_OPT_EXPORT_NAME
1223              */
1224             switch (option) {
1225             case NBD_OPT_EXPORT_NAME:
1226                 return nbd_negotiate_handle_export_name(client,
1227                                                         myflags, no_zeroes,
1228                                                         errp);
1229 
1230             default:
1231                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1232                            option, nbd_opt_lookup(option));
1233                 return -EINVAL;
1234             }
1235         }
1236         if (ret < 0) {
1237             return ret;
1238         }
1239     }
1240 }
1241 
1242 /* nbd_negotiate
1243  * Return:
1244  * -errno  on error, errp is set
1245  * 0       on successful negotiation, errp is not set
1246  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1247  *         errp is not set
1248  */
1249 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1250 {
1251     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1252     int ret;
1253     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1254                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1255                               NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1256 
1257     /* Old style negotiation header, no room for options
1258         [ 0 ..   7]   passwd       ("NBDMAGIC")
1259         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1260         [16 ..  23]   size
1261         [24 ..  27]   export flags (zero-extended)
1262         [28 .. 151]   reserved     (0)
1263 
1264        New style negotiation header, client can send options
1265         [ 0 ..   7]   passwd       ("NBDMAGIC")
1266         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1267         [16 ..  17]   server flags (0)
1268         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1269      */
1270 
1271     qio_channel_set_blocking(client->ioc, false, NULL);
1272 
1273     trace_nbd_negotiate_begin();
1274     memcpy(buf, "NBDMAGIC", 8);
1275 
1276     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1277     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1278 
1279     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1280         error_prepend(errp, "write failed: ");
1281         return -EINVAL;
1282     }
1283     ret = nbd_negotiate_options(client, myflags, errp);
1284     if (ret != 0) {
1285         if (ret < 0) {
1286             error_prepend(errp, "option negotiation failed: ");
1287         }
1288         return ret;
1289     }
1290 
1291     assert(!client->optlen);
1292     trace_nbd_negotiate_success();
1293 
1294     return 0;
1295 }
1296 
1297 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1298                                Error **errp)
1299 {
1300     uint8_t buf[NBD_REQUEST_SIZE];
1301     uint32_t magic;
1302     int ret;
1303 
1304     ret = nbd_read(ioc, buf, sizeof(buf), errp);
1305     if (ret < 0) {
1306         return ret;
1307     }
1308 
1309     /* Request
1310        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1311        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1312        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1313        [ 8 .. 15]   handle
1314        [16 .. 23]   from
1315        [24 .. 27]   len
1316      */
1317 
1318     magic = ldl_be_p(buf);
1319     request->flags  = lduw_be_p(buf + 4);
1320     request->type   = lduw_be_p(buf + 6);
1321     request->handle = ldq_be_p(buf + 8);
1322     request->from   = ldq_be_p(buf + 16);
1323     request->len    = ldl_be_p(buf + 24);
1324 
1325     trace_nbd_receive_request(magic, request->flags, request->type,
1326                               request->from, request->len);
1327 
1328     if (magic != NBD_REQUEST_MAGIC) {
1329         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1330         return -EINVAL;
1331     }
1332     return 0;
1333 }
1334 
1335 #define MAX_NBD_REQUESTS 16
1336 
1337 void nbd_client_get(NBDClient *client)
1338 {
1339     client->refcount++;
1340 }
1341 
1342 void nbd_client_put(NBDClient *client)
1343 {
1344     if (--client->refcount == 0) {
1345         /* The last reference should be dropped by client->close,
1346          * which is called by client_close.
1347          */
1348         assert(client->closing);
1349 
1350         qio_channel_detach_aio_context(client->ioc);
1351         object_unref(OBJECT(client->sioc));
1352         object_unref(OBJECT(client->ioc));
1353         if (client->tlscreds) {
1354             object_unref(OBJECT(client->tlscreds));
1355         }
1356         g_free(client->tlsaclname);
1357         if (client->exp) {
1358             QTAILQ_REMOVE(&client->exp->clients, client, next);
1359             nbd_export_put(client->exp);
1360         }
1361         g_free(client);
1362     }
1363 }
1364 
1365 static void client_close(NBDClient *client, bool negotiated)
1366 {
1367     if (client->closing) {
1368         return;
1369     }
1370 
1371     client->closing = true;
1372 
1373     /* Force requests to finish.  They will drop their own references,
1374      * then we'll close the socket and free the NBDClient.
1375      */
1376     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1377                          NULL);
1378 
1379     /* Also tell the client, so that they release their reference.  */
1380     if (client->close_fn) {
1381         client->close_fn(client, negotiated);
1382     }
1383 }
1384 
1385 static NBDRequestData *nbd_request_get(NBDClient *client)
1386 {
1387     NBDRequestData *req;
1388 
1389     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1390     client->nb_requests++;
1391 
1392     req = g_new0(NBDRequestData, 1);
1393     nbd_client_get(client);
1394     req->client = client;
1395     return req;
1396 }
1397 
1398 static void nbd_request_put(NBDRequestData *req)
1399 {
1400     NBDClient *client = req->client;
1401 
1402     if (req->data) {
1403         qemu_vfree(req->data);
1404     }
1405     g_free(req);
1406 
1407     client->nb_requests--;
1408     nbd_client_receive_next_request(client);
1409 
1410     nbd_client_put(client);
1411 }
1412 
1413 static void blk_aio_attached(AioContext *ctx, void *opaque)
1414 {
1415     NBDExport *exp = opaque;
1416     NBDClient *client;
1417 
1418     trace_nbd_blk_aio_attached(exp->name, ctx);
1419 
1420     exp->ctx = ctx;
1421 
1422     QTAILQ_FOREACH(client, &exp->clients, next) {
1423         qio_channel_attach_aio_context(client->ioc, ctx);
1424         if (client->recv_coroutine) {
1425             aio_co_schedule(ctx, client->recv_coroutine);
1426         }
1427         if (client->send_coroutine) {
1428             aio_co_schedule(ctx, client->send_coroutine);
1429         }
1430     }
1431 }
1432 
1433 static void blk_aio_detach(void *opaque)
1434 {
1435     NBDExport *exp = opaque;
1436     NBDClient *client;
1437 
1438     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1439 
1440     QTAILQ_FOREACH(client, &exp->clients, next) {
1441         qio_channel_detach_aio_context(client->ioc);
1442     }
1443 
1444     exp->ctx = NULL;
1445 }
1446 
1447 static void nbd_eject_notifier(Notifier *n, void *data)
1448 {
1449     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1450     nbd_export_close(exp);
1451 }
1452 
1453 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size,
1454                           uint16_t nbdflags, void (*close)(NBDExport *),
1455                           bool writethrough, BlockBackend *on_eject_blk,
1456                           Error **errp)
1457 {
1458     AioContext *ctx;
1459     BlockBackend *blk;
1460     NBDExport *exp = g_new0(NBDExport, 1);
1461     uint64_t perm;
1462     int ret;
1463 
1464     /*
1465      * NBD exports are used for non-shared storage migration.  Make sure
1466      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1467      * access since the export could be available before migration handover.
1468      */
1469     ctx = bdrv_get_aio_context(bs);
1470     aio_context_acquire(ctx);
1471     bdrv_invalidate_cache(bs, NULL);
1472     aio_context_release(ctx);
1473 
1474     /* Don't allow resize while the NBD server is running, otherwise we don't
1475      * care what happens with the node. */
1476     perm = BLK_PERM_CONSISTENT_READ;
1477     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1478         perm |= BLK_PERM_WRITE;
1479     }
1480     blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1481                         BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1482     ret = blk_insert_bs(blk, bs, errp);
1483     if (ret < 0) {
1484         goto fail;
1485     }
1486     blk_set_enable_write_cache(blk, !writethrough);
1487 
1488     exp->refcount = 1;
1489     QTAILQ_INIT(&exp->clients);
1490     exp->blk = blk;
1491     exp->dev_offset = dev_offset;
1492     exp->nbdflags = nbdflags;
1493     exp->size = size < 0 ? blk_getlength(blk) : size;
1494     if (exp->size < 0) {
1495         error_setg_errno(errp, -exp->size,
1496                          "Failed to determine the NBD export's length");
1497         goto fail;
1498     }
1499     exp->size -= exp->size % BDRV_SECTOR_SIZE;
1500 
1501     exp->close = close;
1502     exp->ctx = blk_get_aio_context(blk);
1503     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1504 
1505     if (on_eject_blk) {
1506         blk_ref(on_eject_blk);
1507         exp->eject_notifier_blk = on_eject_blk;
1508         exp->eject_notifier.notify = nbd_eject_notifier;
1509         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1510     }
1511     return exp;
1512 
1513 fail:
1514     blk_unref(blk);
1515     g_free(exp);
1516     return NULL;
1517 }
1518 
1519 NBDExport *nbd_export_find(const char *name)
1520 {
1521     NBDExport *exp;
1522     QTAILQ_FOREACH(exp, &exports, next) {
1523         if (strcmp(name, exp->name) == 0) {
1524             return exp;
1525         }
1526     }
1527 
1528     return NULL;
1529 }
1530 
1531 void nbd_export_set_name(NBDExport *exp, const char *name)
1532 {
1533     if (exp->name == name) {
1534         return;
1535     }
1536 
1537     nbd_export_get(exp);
1538     if (exp->name != NULL) {
1539         g_free(exp->name);
1540         exp->name = NULL;
1541         QTAILQ_REMOVE(&exports, exp, next);
1542         nbd_export_put(exp);
1543     }
1544     if (name != NULL) {
1545         nbd_export_get(exp);
1546         exp->name = g_strdup(name);
1547         QTAILQ_INSERT_TAIL(&exports, exp, next);
1548     }
1549     nbd_export_put(exp);
1550 }
1551 
1552 void nbd_export_set_description(NBDExport *exp, const char *description)
1553 {
1554     g_free(exp->description);
1555     exp->description = g_strdup(description);
1556 }
1557 
1558 void nbd_export_close(NBDExport *exp)
1559 {
1560     NBDClient *client, *next;
1561 
1562     nbd_export_get(exp);
1563     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1564         client_close(client, true);
1565     }
1566     nbd_export_set_name(exp, NULL);
1567     nbd_export_set_description(exp, NULL);
1568     nbd_export_put(exp);
1569 }
1570 
1571 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1572 {
1573     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1574         nbd_export_close(exp);
1575         return;
1576     }
1577 
1578     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1579 
1580     error_setg(errp, "export '%s' still in use", exp->name);
1581     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1582 }
1583 
1584 void nbd_export_get(NBDExport *exp)
1585 {
1586     assert(exp->refcount > 0);
1587     exp->refcount++;
1588 }
1589 
1590 void nbd_export_put(NBDExport *exp)
1591 {
1592     assert(exp->refcount > 0);
1593     if (exp->refcount == 1) {
1594         nbd_export_close(exp);
1595     }
1596 
1597     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1598      * if someone calls nbd_export_put() on named export not through
1599      * nbd_export_set_name() when refcount is 1. So, let's assert that
1600      * it is > 0.
1601      */
1602     assert(exp->refcount > 0);
1603     if (--exp->refcount == 0) {
1604         assert(exp->name == NULL);
1605         assert(exp->description == NULL);
1606 
1607         if (exp->close) {
1608             exp->close(exp);
1609         }
1610 
1611         if (exp->blk) {
1612             if (exp->eject_notifier_blk) {
1613                 notifier_remove(&exp->eject_notifier);
1614                 blk_unref(exp->eject_notifier_blk);
1615             }
1616             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1617                                             blk_aio_detach, exp);
1618             blk_unref(exp->blk);
1619             exp->blk = NULL;
1620         }
1621 
1622         if (exp->export_bitmap) {
1623             bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false);
1624             g_free(exp->export_bitmap_context);
1625         }
1626 
1627         g_free(exp);
1628     }
1629 }
1630 
1631 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1632 {
1633     return exp->blk;
1634 }
1635 
1636 void nbd_export_close_all(void)
1637 {
1638     NBDExport *exp, *next;
1639 
1640     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1641         nbd_export_close(exp);
1642     }
1643 }
1644 
1645 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1646                                         unsigned niov, Error **errp)
1647 {
1648     int ret;
1649 
1650     g_assert(qemu_in_coroutine());
1651     qemu_co_mutex_lock(&client->send_lock);
1652     client->send_coroutine = qemu_coroutine_self();
1653 
1654     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1655 
1656     client->send_coroutine = NULL;
1657     qemu_co_mutex_unlock(&client->send_lock);
1658 
1659     return ret;
1660 }
1661 
1662 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1663                                        uint64_t handle)
1664 {
1665     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1666     stl_be_p(&reply->error, error);
1667     stq_be_p(&reply->handle, handle);
1668 }
1669 
1670 static int nbd_co_send_simple_reply(NBDClient *client,
1671                                     uint64_t handle,
1672                                     uint32_t error,
1673                                     void *data,
1674                                     size_t len,
1675                                     Error **errp)
1676 {
1677     NBDSimpleReply reply;
1678     int nbd_err = system_errno_to_nbd_errno(error);
1679     struct iovec iov[] = {
1680         {.iov_base = &reply, .iov_len = sizeof(reply)},
1681         {.iov_base = data, .iov_len = len}
1682     };
1683 
1684     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1685                                    len);
1686     set_be_simple_reply(&reply, nbd_err, handle);
1687 
1688     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1689 }
1690 
1691 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1692                                 uint16_t type, uint64_t handle, uint32_t length)
1693 {
1694     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1695     stw_be_p(&chunk->flags, flags);
1696     stw_be_p(&chunk->type, type);
1697     stq_be_p(&chunk->handle, handle);
1698     stl_be_p(&chunk->length, length);
1699 }
1700 
1701 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1702                                                     uint64_t handle,
1703                                                     Error **errp)
1704 {
1705     NBDStructuredReplyChunk chunk;
1706     struct iovec iov[] = {
1707         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1708     };
1709 
1710     trace_nbd_co_send_structured_done(handle);
1711     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1712 
1713     return nbd_co_send_iov(client, iov, 1, errp);
1714 }
1715 
1716 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1717                                                     uint64_t handle,
1718                                                     uint64_t offset,
1719                                                     void *data,
1720                                                     size_t size,
1721                                                     bool final,
1722                                                     Error **errp)
1723 {
1724     NBDStructuredReadData chunk;
1725     struct iovec iov[] = {
1726         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1727         {.iov_base = data, .iov_len = size}
1728     };
1729 
1730     assert(size);
1731     trace_nbd_co_send_structured_read(handle, offset, data, size);
1732     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1733                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1734                  sizeof(chunk) - sizeof(chunk.h) + size);
1735     stq_be_p(&chunk.offset, offset);
1736 
1737     return nbd_co_send_iov(client, iov, 2, errp);
1738 }
1739 
1740 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1741                                                      uint64_t handle,
1742                                                      uint32_t error,
1743                                                      const char *msg,
1744                                                      Error **errp)
1745 {
1746     NBDStructuredError chunk;
1747     int nbd_err = system_errno_to_nbd_errno(error);
1748     struct iovec iov[] = {
1749         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1750         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1751     };
1752 
1753     assert(nbd_err);
1754     trace_nbd_co_send_structured_error(handle, nbd_err,
1755                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1756     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1757                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1758     stl_be_p(&chunk.error, nbd_err);
1759     stw_be_p(&chunk.message_length, iov[1].iov_len);
1760 
1761     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1762 }
1763 
1764 /* Do a sparse read and send the structured reply to the client.
1765  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1766  * reported to the client, at which point this function succeeds.
1767  */
1768 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1769                                                 uint64_t handle,
1770                                                 uint64_t offset,
1771                                                 uint8_t *data,
1772                                                 size_t size,
1773                                                 Error **errp)
1774 {
1775     int ret = 0;
1776     NBDExport *exp = client->exp;
1777     size_t progress = 0;
1778 
1779     while (progress < size) {
1780         int64_t pnum;
1781         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1782                                              offset + progress,
1783                                              size - progress, &pnum, NULL,
1784                                              NULL);
1785         bool final;
1786 
1787         if (status < 0) {
1788             char *msg = g_strdup_printf("unable to check for holes: %s",
1789                                         strerror(-status));
1790 
1791             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1792                                                errp);
1793             g_free(msg);
1794             return ret;
1795         }
1796         assert(pnum && pnum <= size - progress);
1797         final = progress + pnum == size;
1798         if (status & BDRV_BLOCK_ZERO) {
1799             NBDStructuredReadHole chunk;
1800             struct iovec iov[] = {
1801                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1802             };
1803 
1804             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1805                                                    pnum);
1806             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1807                          NBD_REPLY_TYPE_OFFSET_HOLE,
1808                          handle, sizeof(chunk) - sizeof(chunk.h));
1809             stq_be_p(&chunk.offset, offset + progress);
1810             stl_be_p(&chunk.length, pnum);
1811             ret = nbd_co_send_iov(client, iov, 1, errp);
1812         } else {
1813             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1814                             data + progress, pnum);
1815             if (ret < 0) {
1816                 error_setg_errno(errp, -ret, "reading from file failed");
1817                 break;
1818             }
1819             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1820                                               data + progress, pnum, final,
1821                                               errp);
1822         }
1823 
1824         if (ret < 0) {
1825             break;
1826         }
1827         progress += pnum;
1828     }
1829     return ret;
1830 }
1831 
1832 /*
1833  * Populate @extents from block status. Update @bytes to be the actual
1834  * length encoded (which may be smaller than the original), and update
1835  * @nb_extents to the number of extents used.
1836  *
1837  * Returns zero on success and -errno on bdrv_block_status_above failure.
1838  */
1839 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1840                                   uint64_t *bytes, NBDExtent *extents,
1841                                   unsigned int *nb_extents)
1842 {
1843     uint64_t remaining_bytes = *bytes;
1844     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1845     bool first_extent = true;
1846 
1847     assert(*nb_extents);
1848     while (remaining_bytes) {
1849         uint32_t flags;
1850         int64_t num;
1851         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1852                                           &num, NULL, NULL);
1853 
1854         if (ret < 0) {
1855             return ret;
1856         }
1857 
1858         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1859                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1860         offset += num;
1861         remaining_bytes -= num;
1862 
1863         if (first_extent) {
1864             extent->flags = flags;
1865             extent->length = num;
1866             first_extent = false;
1867             continue;
1868         }
1869 
1870         if (flags == extent->flags) {
1871             /* extend current extent */
1872             extent->length += num;
1873         } else {
1874             if (extent + 1 == extents_end) {
1875                 break;
1876             }
1877 
1878             /* start new extent */
1879             extent++;
1880             extent->flags = flags;
1881             extent->length = num;
1882         }
1883     }
1884 
1885     extents_end = extent + 1;
1886 
1887     for (extent = extents; extent < extents_end; extent++) {
1888         extent->flags = cpu_to_be32(extent->flags);
1889         extent->length = cpu_to_be32(extent->length);
1890     }
1891 
1892     *bytes -= remaining_bytes;
1893     *nb_extents = extents_end - extents;
1894 
1895     return 0;
1896 }
1897 
1898 /* nbd_co_send_extents
1899  *
1900  * @length is only for tracing purposes (and may be smaller or larger
1901  * than the client's original request). @last controls whether
1902  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1903  * big-endian format.
1904  */
1905 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1906                                NBDExtent *extents, unsigned int nb_extents,
1907                                uint64_t length, bool last,
1908                                uint32_t context_id, Error **errp)
1909 {
1910     NBDStructuredMeta chunk;
1911 
1912     struct iovec iov[] = {
1913         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1914         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1915     };
1916 
1917     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1918     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1919                  NBD_REPLY_TYPE_BLOCK_STATUS,
1920                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1921     stl_be_p(&chunk.context_id, context_id);
1922 
1923     return nbd_co_send_iov(client, iov, 2, errp);
1924 }
1925 
1926 /* Get block status from the exported device and send it to the client */
1927 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1928                                     BlockDriverState *bs, uint64_t offset,
1929                                     uint32_t length, bool dont_fragment,
1930                                     bool last, uint32_t context_id,
1931                                     Error **errp)
1932 {
1933     int ret;
1934     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1935     NBDExtent *extents = g_new(NBDExtent, nb_extents);
1936     uint64_t final_length = length;
1937 
1938     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1939                                  &nb_extents);
1940     if (ret < 0) {
1941         g_free(extents);
1942         return nbd_co_send_structured_error(
1943                 client, handle, -ret, "can't get block status", errp);
1944     }
1945 
1946     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1947                               final_length, last, context_id, errp);
1948 
1949     g_free(extents);
1950 
1951     return ret;
1952 }
1953 
1954 /*
1955  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1956  * final extent may exceed the original @length. Store in @length the
1957  * byte length encoded (which may be smaller or larger than the
1958  * original), and return the number of extents used.
1959  */
1960 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1961                                       uint64_t *length, NBDExtent *extents,
1962                                       unsigned int nb_extents,
1963                                       bool dont_fragment)
1964 {
1965     uint64_t begin = offset, end = offset;
1966     uint64_t overall_end = offset + *length;
1967     unsigned int i = 0;
1968     BdrvDirtyBitmapIter *it;
1969     bool dirty;
1970 
1971     bdrv_dirty_bitmap_lock(bitmap);
1972 
1973     it = bdrv_dirty_iter_new(bitmap);
1974     dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
1975 
1976     assert(begin < overall_end && nb_extents);
1977     while (begin < overall_end && i < nb_extents) {
1978         bool next_dirty = !dirty;
1979 
1980         if (dirty) {
1981             end = bdrv_dirty_bitmap_next_zero(bitmap, begin);
1982         } else {
1983             bdrv_set_dirty_iter(it, begin);
1984             end = bdrv_dirty_iter_next(it);
1985         }
1986         if (end == -1 || end - begin > UINT32_MAX) {
1987             /* Cap to an aligned value < 4G beyond begin. */
1988             end = MIN(bdrv_dirty_bitmap_size(bitmap),
1989                       begin + UINT32_MAX + 1 -
1990                       bdrv_dirty_bitmap_granularity(bitmap));
1991             next_dirty = dirty;
1992         }
1993         if (dont_fragment && end > overall_end) {
1994             end = overall_end;
1995         }
1996 
1997         extents[i].length = cpu_to_be32(end - begin);
1998         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
1999         i++;
2000         begin = end;
2001         dirty = next_dirty;
2002     }
2003 
2004     bdrv_dirty_iter_free(it);
2005 
2006     bdrv_dirty_bitmap_unlock(bitmap);
2007 
2008     assert(offset < end);
2009     *length = end - offset;
2010     return i;
2011 }
2012 
2013 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2014                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2015                               uint32_t length, bool dont_fragment, bool last,
2016                               uint32_t context_id, Error **errp)
2017 {
2018     int ret;
2019     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2020     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2021     uint64_t final_length = length;
2022 
2023     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2024                                    nb_extents, dont_fragment);
2025 
2026     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2027                               final_length, last, context_id, errp);
2028 
2029     g_free(extents);
2030 
2031     return ret;
2032 }
2033 
2034 /* nbd_co_receive_request
2035  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2036  * connection right away, and any other negative value to report an error to
2037  * the client (although the caller may still need to disconnect after reporting
2038  * the error).
2039  */
2040 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2041                                   Error **errp)
2042 {
2043     NBDClient *client = req->client;
2044     int valid_flags;
2045 
2046     g_assert(qemu_in_coroutine());
2047     assert(client->recv_coroutine == qemu_coroutine_self());
2048     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2049         return -EIO;
2050     }
2051 
2052     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2053                                              nbd_cmd_lookup(request->type));
2054 
2055     if (request->type != NBD_CMD_WRITE) {
2056         /* No payload, we are ready to read the next request.  */
2057         req->complete = true;
2058     }
2059 
2060     if (request->type == NBD_CMD_DISC) {
2061         /* Special case: we're going to disconnect without a reply,
2062          * whether or not flags, from, or len are bogus */
2063         return -EIO;
2064     }
2065 
2066     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2067         request->type == NBD_CMD_CACHE)
2068     {
2069         if (request->len > NBD_MAX_BUFFER_SIZE) {
2070             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2071                        request->len, NBD_MAX_BUFFER_SIZE);
2072             return -EINVAL;
2073         }
2074 
2075         req->data = blk_try_blockalign(client->exp->blk, request->len);
2076         if (req->data == NULL) {
2077             error_setg(errp, "No memory");
2078             return -ENOMEM;
2079         }
2080     }
2081     if (request->type == NBD_CMD_WRITE) {
2082         if (nbd_read(client->ioc, req->data, request->len, errp) < 0) {
2083             error_prepend(errp, "reading from socket failed: ");
2084             return -EIO;
2085         }
2086         req->complete = true;
2087 
2088         trace_nbd_co_receive_request_payload_received(request->handle,
2089                                                       request->len);
2090     }
2091 
2092     /* Sanity checks. */
2093     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2094         (request->type == NBD_CMD_WRITE ||
2095          request->type == NBD_CMD_WRITE_ZEROES ||
2096          request->type == NBD_CMD_TRIM)) {
2097         error_setg(errp, "Export is read-only");
2098         return -EROFS;
2099     }
2100     if (request->from > client->exp->size ||
2101         request->from + request->len > client->exp->size) {
2102         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2103                    ", Size: %" PRIu64, request->from, request->len,
2104                    (uint64_t)client->exp->size);
2105         return (request->type == NBD_CMD_WRITE ||
2106                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2107     }
2108     valid_flags = NBD_CMD_FLAG_FUA;
2109     if (request->type == NBD_CMD_READ && client->structured_reply) {
2110         valid_flags |= NBD_CMD_FLAG_DF;
2111     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2112         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2113     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2114         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2115     }
2116     if (request->flags & ~valid_flags) {
2117         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2118                    nbd_cmd_lookup(request->type), request->flags);
2119         return -EINVAL;
2120     }
2121 
2122     return 0;
2123 }
2124 
2125 /* Send simple reply without a payload, or a structured error
2126  * @error_msg is ignored if @ret >= 0
2127  * Returns 0 if connection is still live, -errno on failure to talk to client
2128  */
2129 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2130                                                uint64_t handle,
2131                                                int ret,
2132                                                const char *error_msg,
2133                                                Error **errp)
2134 {
2135     if (client->structured_reply && ret < 0) {
2136         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2137                                             errp);
2138     } else {
2139         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2140                                         NULL, 0, errp);
2141     }
2142 }
2143 
2144 /* Handle NBD_CMD_READ request.
2145  * Return -errno if sending fails. Other errors are reported directly to the
2146  * client as an error reply. */
2147 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2148                                         uint8_t *data, Error **errp)
2149 {
2150     int ret;
2151     NBDExport *exp = client->exp;
2152 
2153     assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2154 
2155     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2156     if (request->flags & NBD_CMD_FLAG_FUA) {
2157         ret = blk_co_flush(exp->blk);
2158         if (ret < 0) {
2159             return nbd_send_generic_reply(client, request->handle, ret,
2160                                           "flush failed", errp);
2161         }
2162     }
2163 
2164     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2165         request->len && request->type != NBD_CMD_CACHE)
2166     {
2167         return nbd_co_send_sparse_read(client, request->handle, request->from,
2168                                        data, request->len, errp);
2169     }
2170 
2171     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2172                     request->len);
2173     if (ret < 0 || request->type == NBD_CMD_CACHE) {
2174         return nbd_send_generic_reply(client, request->handle, ret,
2175                                       "reading from file failed", errp);
2176     }
2177 
2178     if (client->structured_reply) {
2179         if (request->len) {
2180             return nbd_co_send_structured_read(client, request->handle,
2181                                                request->from, data,
2182                                                request->len, true, errp);
2183         } else {
2184             return nbd_co_send_structured_done(client, request->handle, errp);
2185         }
2186     } else {
2187         return nbd_co_send_simple_reply(client, request->handle, 0,
2188                                         data, request->len, errp);
2189     }
2190 }
2191 
2192 /* Handle NBD request.
2193  * Return -errno if sending fails. Other errors are reported directly to the
2194  * client as an error reply. */
2195 static coroutine_fn int nbd_handle_request(NBDClient *client,
2196                                            NBDRequest *request,
2197                                            uint8_t *data, Error **errp)
2198 {
2199     int ret;
2200     int flags;
2201     NBDExport *exp = client->exp;
2202     char *msg;
2203 
2204     switch (request->type) {
2205     case NBD_CMD_READ:
2206     case NBD_CMD_CACHE:
2207         return nbd_do_cmd_read(client, request, data, errp);
2208 
2209     case NBD_CMD_WRITE:
2210         flags = 0;
2211         if (request->flags & NBD_CMD_FLAG_FUA) {
2212             flags |= BDRV_REQ_FUA;
2213         }
2214         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2215                          data, request->len, flags);
2216         return nbd_send_generic_reply(client, request->handle, ret,
2217                                       "writing to file failed", errp);
2218 
2219     case NBD_CMD_WRITE_ZEROES:
2220         flags = 0;
2221         if (request->flags & NBD_CMD_FLAG_FUA) {
2222             flags |= BDRV_REQ_FUA;
2223         }
2224         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2225             flags |= BDRV_REQ_MAY_UNMAP;
2226         }
2227         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2228                                 request->len, flags);
2229         return nbd_send_generic_reply(client, request->handle, ret,
2230                                       "writing to file failed", errp);
2231 
2232     case NBD_CMD_DISC:
2233         /* unreachable, thanks to special case in nbd_co_receive_request() */
2234         abort();
2235 
2236     case NBD_CMD_FLUSH:
2237         ret = blk_co_flush(exp->blk);
2238         return nbd_send_generic_reply(client, request->handle, ret,
2239                                       "flush failed", errp);
2240 
2241     case NBD_CMD_TRIM:
2242         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2243                               request->len);
2244         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2245             ret = blk_co_flush(exp->blk);
2246         }
2247         return nbd_send_generic_reply(client, request->handle, ret,
2248                                       "discard failed", errp);
2249 
2250     case NBD_CMD_BLOCK_STATUS:
2251         if (!request->len) {
2252             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2253                                           "need non-zero length", errp);
2254         }
2255         if (client->export_meta.valid &&
2256             (client->export_meta.base_allocation ||
2257              client->export_meta.bitmap))
2258         {
2259             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2260 
2261             if (client->export_meta.base_allocation) {
2262                 ret = nbd_co_send_block_status(client, request->handle,
2263                                                blk_bs(exp->blk), request->from,
2264                                                request->len, dont_fragment,
2265                                                !client->export_meta.bitmap,
2266                                                NBD_META_ID_BASE_ALLOCATION,
2267                                                errp);
2268                 if (ret < 0) {
2269                     return ret;
2270                 }
2271             }
2272 
2273             if (client->export_meta.bitmap) {
2274                 ret = nbd_co_send_bitmap(client, request->handle,
2275                                          client->exp->export_bitmap,
2276                                          request->from, request->len,
2277                                          dont_fragment,
2278                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2279                 if (ret < 0) {
2280                     return ret;
2281                 }
2282             }
2283 
2284             return ret;
2285         } else {
2286             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2287                                           "CMD_BLOCK_STATUS not negotiated",
2288                                           errp);
2289         }
2290 
2291     default:
2292         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2293                               request->type);
2294         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2295                                      errp);
2296         g_free(msg);
2297         return ret;
2298     }
2299 }
2300 
2301 /* Owns a reference to the NBDClient passed as opaque.  */
2302 static coroutine_fn void nbd_trip(void *opaque)
2303 {
2304     NBDClient *client = opaque;
2305     NBDRequestData *req;
2306     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2307     int ret;
2308     Error *local_err = NULL;
2309 
2310     trace_nbd_trip();
2311     if (client->closing) {
2312         nbd_client_put(client);
2313         return;
2314     }
2315 
2316     req = nbd_request_get(client);
2317     ret = nbd_co_receive_request(req, &request, &local_err);
2318     client->recv_coroutine = NULL;
2319 
2320     if (client->closing) {
2321         /*
2322          * The client may be closed when we are blocked in
2323          * nbd_co_receive_request()
2324          */
2325         goto done;
2326     }
2327 
2328     nbd_client_receive_next_request(client);
2329     if (ret == -EIO) {
2330         goto disconnect;
2331     }
2332 
2333     if (ret < 0) {
2334         /* It wans't -EIO, so, according to nbd_co_receive_request()
2335          * semantics, we should return the error to the client. */
2336         Error *export_err = local_err;
2337 
2338         local_err = NULL;
2339         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2340                                      error_get_pretty(export_err), &local_err);
2341         error_free(export_err);
2342     } else {
2343         ret = nbd_handle_request(client, &request, req->data, &local_err);
2344     }
2345     if (ret < 0) {
2346         error_prepend(&local_err, "Failed to send reply: ");
2347         goto disconnect;
2348     }
2349 
2350     /* We must disconnect after NBD_CMD_WRITE if we did not
2351      * read the payload.
2352      */
2353     if (!req->complete) {
2354         error_setg(&local_err, "Request handling failed in intermediate state");
2355         goto disconnect;
2356     }
2357 
2358 done:
2359     nbd_request_put(req);
2360     nbd_client_put(client);
2361     return;
2362 
2363 disconnect:
2364     if (local_err) {
2365         error_reportf_err(local_err, "Disconnect client, due to: ");
2366     }
2367     nbd_request_put(req);
2368     client_close(client, true);
2369     nbd_client_put(client);
2370 }
2371 
2372 static void nbd_client_receive_next_request(NBDClient *client)
2373 {
2374     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2375         nbd_client_get(client);
2376         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2377         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2378     }
2379 }
2380 
2381 static coroutine_fn void nbd_co_client_start(void *opaque)
2382 {
2383     NBDClient *client = opaque;
2384     Error *local_err = NULL;
2385 
2386     qemu_co_mutex_init(&client->send_lock);
2387 
2388     if (nbd_negotiate(client, &local_err)) {
2389         if (local_err) {
2390             error_report_err(local_err);
2391         }
2392         client_close(client, false);
2393         return;
2394     }
2395 
2396     nbd_client_receive_next_request(client);
2397 }
2398 
2399 /*
2400  * Create a new client listener using the given channel @sioc.
2401  * Begin servicing it in a coroutine.  When the connection closes, call
2402  * @close_fn with an indication of whether the client completed negotiation.
2403  */
2404 void nbd_client_new(QIOChannelSocket *sioc,
2405                     QCryptoTLSCreds *tlscreds,
2406                     const char *tlsaclname,
2407                     void (*close_fn)(NBDClient *, bool))
2408 {
2409     NBDClient *client;
2410     Coroutine *co;
2411 
2412     client = g_new0(NBDClient, 1);
2413     client->refcount = 1;
2414     client->tlscreds = tlscreds;
2415     if (tlscreds) {
2416         object_ref(OBJECT(client->tlscreds));
2417     }
2418     client->tlsaclname = g_strdup(tlsaclname);
2419     client->sioc = sioc;
2420     object_ref(OBJECT(client->sioc));
2421     client->ioc = QIO_CHANNEL(sioc);
2422     object_ref(OBJECT(client->ioc));
2423     client->close_fn = close_fn;
2424 
2425     co = qemu_coroutine_create(nbd_co_client_start, client);
2426     qemu_coroutine_enter(co);
2427 }
2428 
2429 void nbd_export_bitmap(NBDExport *exp, const char *bitmap,
2430                        const char *bitmap_export_name, Error **errp)
2431 {
2432     BdrvDirtyBitmap *bm = NULL;
2433     BlockDriverState *bs = blk_bs(exp->blk);
2434 
2435     if (exp->export_bitmap) {
2436         error_setg(errp, "Export bitmap is already set");
2437         return;
2438     }
2439 
2440     while (true) {
2441         bm = bdrv_find_dirty_bitmap(bs, bitmap);
2442         if (bm != NULL || bs->backing == NULL) {
2443             break;
2444         }
2445 
2446         bs = bs->backing->bs;
2447     }
2448 
2449     if (bm == NULL) {
2450         error_setg(errp, "Bitmap '%s' is not found", bitmap);
2451         return;
2452     }
2453 
2454     if (bdrv_dirty_bitmap_enabled(bm)) {
2455         error_setg(errp, "Bitmap '%s' is enabled", bitmap);
2456         return;
2457     }
2458 
2459     if (bdrv_dirty_bitmap_user_locked(bm)) {
2460         error_setg(errp, "Bitmap '%s' is in use", bitmap);
2461         return;
2462     }
2463 
2464     bdrv_dirty_bitmap_set_qmp_locked(bm, true);
2465     exp->export_bitmap = bm;
2466     exp->export_bitmap_context =
2467             g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name);
2468 }
2469