xref: /openbmc/qemu/nbd/server.c (revision b3c818a4)
1 /*
2  *  Copyright (C) 2016-2020 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 
22 #include "block/export.h"
23 #include "qapi/error.h"
24 #include "qemu/queue.h"
25 #include "trace.h"
26 #include "nbd-internal.h"
27 #include "qemu/units.h"
28 
29 #define NBD_META_ID_BASE_ALLOCATION 0
30 #define NBD_META_ID_ALLOCATION_DEPTH 1
31 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
32 #define NBD_META_ID_DIRTY_BITMAP 2
33 
34 /*
35  * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
36  * constant. If an increase is needed, note that the NBD protocol
37  * recommends no larger than 32 mb, so that the client won't consider
38  * the reply as a denial of service attack.
39  */
40 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
41 
42 static int system_errno_to_nbd_errno(int err)
43 {
44     switch (err) {
45     case 0:
46         return NBD_SUCCESS;
47     case EPERM:
48     case EROFS:
49         return NBD_EPERM;
50     case EIO:
51         return NBD_EIO;
52     case ENOMEM:
53         return NBD_ENOMEM;
54 #ifdef EDQUOT
55     case EDQUOT:
56 #endif
57     case EFBIG:
58     case ENOSPC:
59         return NBD_ENOSPC;
60     case EOVERFLOW:
61         return NBD_EOVERFLOW;
62     case ENOTSUP:
63 #if ENOTSUP != EOPNOTSUPP
64     case EOPNOTSUPP:
65 #endif
66         return NBD_ENOTSUP;
67     case ESHUTDOWN:
68         return NBD_ESHUTDOWN;
69     case EINVAL:
70     default:
71         return NBD_EINVAL;
72     }
73 }
74 
75 /* Definitions for opaque data types */
76 
77 typedef struct NBDRequestData NBDRequestData;
78 
79 struct NBDRequestData {
80     QSIMPLEQ_ENTRY(NBDRequestData) entry;
81     NBDClient *client;
82     uint8_t *data;
83     bool complete;
84 };
85 
86 struct NBDExport {
87     BlockExport common;
88 
89     char *name;
90     char *description;
91     uint64_t size;
92     uint16_t nbdflags;
93     QTAILQ_HEAD(, NBDClient) clients;
94     QTAILQ_ENTRY(NBDExport) next;
95 
96     BlockBackend *eject_notifier_blk;
97     Notifier eject_notifier;
98 
99     bool allocation_depth;
100     BdrvDirtyBitmap **export_bitmaps;
101     size_t nr_export_bitmaps;
102 };
103 
104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
105 
106 /* NBDExportMetaContexts represents a list of contexts to be exported,
107  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
108  * NBD_OPT_LIST_META_CONTEXT. */
109 typedef struct NBDExportMetaContexts {
110     NBDExport *exp;
111     size_t count; /* number of negotiated contexts */
112     bool base_allocation; /* export base:allocation context (block status) */
113     bool allocation_depth; /* export qemu:allocation-depth */
114     bool *bitmaps; /*
115                     * export qemu:dirty-bitmap:<export bitmap name>,
116                     * sized by exp->nr_export_bitmaps
117                     */
118 } NBDExportMetaContexts;
119 
120 struct NBDClient {
121     int refcount;
122     void (*close_fn)(NBDClient *client, bool negotiated);
123 
124     NBDExport *exp;
125     QCryptoTLSCreds *tlscreds;
126     char *tlsauthz;
127     QIOChannelSocket *sioc; /* The underlying data channel */
128     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
129 
130     Coroutine *recv_coroutine;
131 
132     CoMutex send_lock;
133     Coroutine *send_coroutine;
134 
135     QTAILQ_ENTRY(NBDClient) next;
136     int nb_requests;
137     bool closing;
138 
139     uint32_t check_align; /* If non-zero, check for aligned client requests */
140 
141     bool structured_reply;
142     NBDExportMetaContexts export_meta;
143 
144     uint32_t opt; /* Current option being negotiated */
145     uint32_t optlen; /* remaining length of data in ioc for the option being
146                         negotiated now */
147 };
148 
149 static void nbd_client_receive_next_request(NBDClient *client);
150 
151 /* Basic flow for negotiation
152 
153    Server         Client
154    Negotiate
155 
156    or
157 
158    Server         Client
159    Negotiate #1
160                   Option
161    Negotiate #2
162 
163    ----
164 
165    followed by
166 
167    Server         Client
168                   Request
169    Response
170                   Request
171    Response
172                   ...
173    ...
174                   Request (type == 2)
175 
176 */
177 
178 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
179                                      uint32_t type, uint32_t length)
180 {
181     stq_be_p(&rep->magic, NBD_REP_MAGIC);
182     stl_be_p(&rep->option, option);
183     stl_be_p(&rep->type, type);
184     stl_be_p(&rep->length, length);
185 }
186 
187 /* Send a reply header, including length, but no payload.
188  * Return -errno on error, 0 on success. */
189 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
190                                       uint32_t len, Error **errp)
191 {
192     NBDOptionReply rep;
193 
194     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
195                                      type, nbd_rep_lookup(type), len);
196 
197     assert(len < NBD_MAX_BUFFER_SIZE);
198 
199     set_be_option_rep(&rep, client->opt, type, len);
200     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
201 }
202 
203 /* Send a reply header with default 0 length.
204  * Return -errno on error, 0 on success. */
205 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
206                                   Error **errp)
207 {
208     return nbd_negotiate_send_rep_len(client, type, 0, errp);
209 }
210 
211 /* Send an error reply.
212  * Return -errno on error, 0 on success. */
213 static int GCC_FMT_ATTR(4, 0)
214 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
215                             Error **errp, const char *fmt, va_list va)
216 {
217     ERRP_GUARD();
218     g_autofree char *msg = NULL;
219     int ret;
220     size_t len;
221 
222     msg = g_strdup_vprintf(fmt, va);
223     len = strlen(msg);
224     assert(len < NBD_MAX_STRING_SIZE);
225     trace_nbd_negotiate_send_rep_err(msg);
226     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
227     if (ret < 0) {
228         return ret;
229     }
230     if (nbd_write(client->ioc, msg, len, errp) < 0) {
231         error_prepend(errp, "write failed (error message): ");
232         return -EIO;
233     }
234 
235     return 0;
236 }
237 
238 /*
239  * Return a malloc'd copy of @name suitable for use in an error reply.
240  */
241 static char *
242 nbd_sanitize_name(const char *name)
243 {
244     if (strnlen(name, 80) < 80) {
245         return g_strdup(name);
246     }
247     /* XXX Should we also try to sanitize any control characters? */
248     return g_strdup_printf("%.80s...", name);
249 }
250 
251 /* Send an error reply.
252  * Return -errno on error, 0 on success. */
253 static int GCC_FMT_ATTR(4, 5)
254 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
255                            Error **errp, const char *fmt, ...)
256 {
257     va_list va;
258     int ret;
259 
260     va_start(va, fmt);
261     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
262     va_end(va);
263     return ret;
264 }
265 
266 /* Drop remainder of the current option, and send a reply with the
267  * given error type and message. Return -errno on read or write
268  * failure; or 0 if connection is still live. */
269 static int GCC_FMT_ATTR(4, 0)
270 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
271               const char *fmt, va_list va)
272 {
273     int ret = nbd_drop(client->ioc, client->optlen, errp);
274 
275     client->optlen = 0;
276     if (!ret) {
277         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
278     }
279     return ret;
280 }
281 
282 static int GCC_FMT_ATTR(4, 5)
283 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
284              const char *fmt, ...)
285 {
286     int ret;
287     va_list va;
288 
289     va_start(va, fmt);
290     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
291     va_end(va);
292 
293     return ret;
294 }
295 
296 static int GCC_FMT_ATTR(3, 4)
297 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
298 {
299     int ret;
300     va_list va;
301 
302     va_start(va, fmt);
303     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
304     va_end(va);
305 
306     return ret;
307 }
308 
309 /* Read size bytes from the unparsed payload of the current option.
310  * If @check_nul, require that no NUL bytes appear in buffer.
311  * Return -errno on I/O error, 0 if option was completely handled by
312  * sending a reply about inconsistent lengths, or 1 on success. */
313 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
314                         bool check_nul, Error **errp)
315 {
316     if (size > client->optlen) {
317         return nbd_opt_invalid(client, errp,
318                                "Inconsistent lengths in option %s",
319                                nbd_opt_lookup(client->opt));
320     }
321     client->optlen -= size;
322     if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
323         return -EIO;
324     }
325 
326     if (check_nul && strnlen(buffer, size) != size) {
327         return nbd_opt_invalid(client, errp,
328                                "Unexpected embedded NUL in option %s",
329                                nbd_opt_lookup(client->opt));
330     }
331     return 1;
332 }
333 
334 /* Drop size bytes from the unparsed payload of the current option.
335  * Return -errno on I/O error, 0 if option was completely handled by
336  * sending a reply about inconsistent lengths, or 1 on success. */
337 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
338 {
339     if (size > client->optlen) {
340         return nbd_opt_invalid(client, errp,
341                                "Inconsistent lengths in option %s",
342                                nbd_opt_lookup(client->opt));
343     }
344     client->optlen -= size;
345     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
346 }
347 
348 /* nbd_opt_read_name
349  *
350  * Read a string with the format:
351  *   uint32_t len     (<= NBD_MAX_STRING_SIZE)
352  *   len bytes string (not 0-terminated)
353  *
354  * On success, @name will be allocated.
355  * If @length is non-null, it will be set to the actual string length.
356  *
357  * Return -errno on I/O error, 0 if option was completely handled by
358  * sending a reply about inconsistent lengths, or 1 on success.
359  */
360 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
361                              Error **errp)
362 {
363     int ret;
364     uint32_t len;
365     g_autofree char *local_name = NULL;
366 
367     *name = NULL;
368     ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
369     if (ret <= 0) {
370         return ret;
371     }
372     len = cpu_to_be32(len);
373 
374     if (len > NBD_MAX_STRING_SIZE) {
375         return nbd_opt_invalid(client, errp,
376                                "Invalid name length: %" PRIu32, len);
377     }
378 
379     local_name = g_malloc(len + 1);
380     ret = nbd_opt_read(client, local_name, len, true, errp);
381     if (ret <= 0) {
382         return ret;
383     }
384     local_name[len] = '\0';
385 
386     if (length) {
387         *length = len;
388     }
389     *name = g_steal_pointer(&local_name);
390 
391     return 1;
392 }
393 
394 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
395  * Return -errno on error, 0 on success. */
396 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
397                                        Error **errp)
398 {
399     ERRP_GUARD();
400     size_t name_len, desc_len;
401     uint32_t len;
402     const char *name = exp->name ? exp->name : "";
403     const char *desc = exp->description ? exp->description : "";
404     QIOChannel *ioc = client->ioc;
405     int ret;
406 
407     trace_nbd_negotiate_send_rep_list(name, desc);
408     name_len = strlen(name);
409     desc_len = strlen(desc);
410     assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
411     len = name_len + desc_len + sizeof(len);
412     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
413     if (ret < 0) {
414         return ret;
415     }
416 
417     len = cpu_to_be32(name_len);
418     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
419         error_prepend(errp, "write failed (name length): ");
420         return -EINVAL;
421     }
422 
423     if (nbd_write(ioc, name, name_len, errp) < 0) {
424         error_prepend(errp, "write failed (name buffer): ");
425         return -EINVAL;
426     }
427 
428     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
429         error_prepend(errp, "write failed (description buffer): ");
430         return -EINVAL;
431     }
432 
433     return 0;
434 }
435 
436 /* Process the NBD_OPT_LIST command, with a potential series of replies.
437  * Return -errno on error, 0 on success. */
438 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
439 {
440     NBDExport *exp;
441     assert(client->opt == NBD_OPT_LIST);
442 
443     /* For each export, send a NBD_REP_SERVER reply. */
444     QTAILQ_FOREACH(exp, &exports, next) {
445         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
446             return -EINVAL;
447         }
448     }
449     /* Finish with a NBD_REP_ACK. */
450     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
451 }
452 
453 static void nbd_check_meta_export(NBDClient *client)
454 {
455     if (client->exp != client->export_meta.exp) {
456         client->export_meta.count = 0;
457     }
458 }
459 
460 /* Send a reply to NBD_OPT_EXPORT_NAME.
461  * Return -errno on error, 0 on success. */
462 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
463                                             Error **errp)
464 {
465     ERRP_GUARD();
466     g_autofree char *name = NULL;
467     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
468     size_t len;
469     int ret;
470     uint16_t myflags;
471 
472     /* Client sends:
473         [20 ..  xx]   export name (length bytes)
474        Server replies:
475         [ 0 ..   7]   size
476         [ 8 ..   9]   export flags
477         [10 .. 133]   reserved     (0) [unless no_zeroes]
478      */
479     trace_nbd_negotiate_handle_export_name();
480     if (client->optlen > NBD_MAX_STRING_SIZE) {
481         error_setg(errp, "Bad length received");
482         return -EINVAL;
483     }
484     name = g_malloc(client->optlen + 1);
485     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
486         return -EIO;
487     }
488     name[client->optlen] = '\0';
489     client->optlen = 0;
490 
491     trace_nbd_negotiate_handle_export_name_request(name);
492 
493     client->exp = nbd_export_find(name);
494     if (!client->exp) {
495         error_setg(errp, "export not found");
496         return -EINVAL;
497     }
498 
499     myflags = client->exp->nbdflags;
500     if (client->structured_reply) {
501         myflags |= NBD_FLAG_SEND_DF;
502     }
503     trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
504     stq_be_p(buf, client->exp->size);
505     stw_be_p(buf + 8, myflags);
506     len = no_zeroes ? 10 : sizeof(buf);
507     ret = nbd_write(client->ioc, buf, len, errp);
508     if (ret < 0) {
509         error_prepend(errp, "write failed: ");
510         return ret;
511     }
512 
513     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
514     blk_exp_ref(&client->exp->common);
515     nbd_check_meta_export(client);
516 
517     return 0;
518 }
519 
520 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
521  * The buffer does NOT include the info type prefix.
522  * Return -errno on error, 0 if ready to send more. */
523 static int nbd_negotiate_send_info(NBDClient *client,
524                                    uint16_t info, uint32_t length, void *buf,
525                                    Error **errp)
526 {
527     int rc;
528 
529     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
530     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
531                                     sizeof(info) + length, errp);
532     if (rc < 0) {
533         return rc;
534     }
535     info = cpu_to_be16(info);
536     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
537         return -EIO;
538     }
539     if (nbd_write(client->ioc, buf, length, errp) < 0) {
540         return -EIO;
541     }
542     return 0;
543 }
544 
545 /* nbd_reject_length: Handle any unexpected payload.
546  * @fatal requests that we quit talking to the client, even if we are able
547  * to successfully send an error reply.
548  * Return:
549  * -errno  transmission error occurred or @fatal was requested, errp is set
550  * 0       error message successfully sent to client, errp is not set
551  */
552 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
553 {
554     int ret;
555 
556     assert(client->optlen);
557     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
558                           nbd_opt_lookup(client->opt));
559     if (fatal && !ret) {
560         error_setg(errp, "option '%s' has unexpected length",
561                    nbd_opt_lookup(client->opt));
562         return -EINVAL;
563     }
564     return ret;
565 }
566 
567 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
568  * Return -errno on error, 0 if ready for next option, and 1 to move
569  * into transmission phase.  */
570 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
571 {
572     int rc;
573     g_autofree char *name = NULL;
574     NBDExport *exp;
575     uint16_t requests;
576     uint16_t request;
577     uint32_t namelen = 0;
578     bool sendname = false;
579     bool blocksize = false;
580     uint32_t sizes[3];
581     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
582     uint32_t check_align = 0;
583     uint16_t myflags;
584 
585     /* Client sends:
586         4 bytes: L, name length (can be 0)
587         L bytes: export name
588         2 bytes: N, number of requests (can be 0)
589         N * 2 bytes: N requests
590     */
591     rc = nbd_opt_read_name(client, &name, &namelen, errp);
592     if (rc <= 0) {
593         return rc;
594     }
595     trace_nbd_negotiate_handle_export_name_request(name);
596 
597     rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
598     if (rc <= 0) {
599         return rc;
600     }
601     requests = be16_to_cpu(requests);
602     trace_nbd_negotiate_handle_info_requests(requests);
603     while (requests--) {
604         rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
605         if (rc <= 0) {
606             return rc;
607         }
608         request = be16_to_cpu(request);
609         trace_nbd_negotiate_handle_info_request(request,
610                                                 nbd_info_lookup(request));
611         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
612          * everything else is either a request we don't know or
613          * something we send regardless of request */
614         switch (request) {
615         case NBD_INFO_NAME:
616             sendname = true;
617             break;
618         case NBD_INFO_BLOCK_SIZE:
619             blocksize = true;
620             break;
621         }
622     }
623     if (client->optlen) {
624         return nbd_reject_length(client, false, errp);
625     }
626 
627     exp = nbd_export_find(name);
628     if (!exp) {
629         g_autofree char *sane_name = nbd_sanitize_name(name);
630 
631         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
632                                           errp, "export '%s' not present",
633                                           sane_name);
634     }
635 
636     /* Don't bother sending NBD_INFO_NAME unless client requested it */
637     if (sendname) {
638         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
639                                      errp);
640         if (rc < 0) {
641             return rc;
642         }
643     }
644 
645     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
646      * client request */
647     if (exp->description) {
648         size_t len = strlen(exp->description);
649 
650         assert(len <= NBD_MAX_STRING_SIZE);
651         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
652                                      len, exp->description, errp);
653         if (rc < 0) {
654             return rc;
655         }
656     }
657 
658     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
659      * according to whether the client requested it, and according to
660      * whether this is OPT_INFO or OPT_GO. */
661     /* minimum - 1 for back-compat, or actual if client will obey it. */
662     if (client->opt == NBD_OPT_INFO || blocksize) {
663         check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
664     } else {
665         sizes[0] = 1;
666     }
667     assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
668     /* preferred - Hard-code to 4096 for now.
669      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
670     sizes[1] = MAX(4096, sizes[0]);
671     /* maximum - At most 32M, but smaller as appropriate. */
672     sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
673     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
674     sizes[0] = cpu_to_be32(sizes[0]);
675     sizes[1] = cpu_to_be32(sizes[1]);
676     sizes[2] = cpu_to_be32(sizes[2]);
677     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
678                                  sizeof(sizes), sizes, errp);
679     if (rc < 0) {
680         return rc;
681     }
682 
683     /* Send NBD_INFO_EXPORT always */
684     myflags = exp->nbdflags;
685     if (client->structured_reply) {
686         myflags |= NBD_FLAG_SEND_DF;
687     }
688     trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
689     stq_be_p(buf, exp->size);
690     stw_be_p(buf + 8, myflags);
691     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
692                                  sizeof(buf), buf, errp);
693     if (rc < 0) {
694         return rc;
695     }
696 
697     /*
698      * If the client is just asking for NBD_OPT_INFO, but forgot to
699      * request block sizes in a situation that would impact
700      * performance, then return an error. But for NBD_OPT_GO, we
701      * tolerate all clients, regardless of alignments.
702      */
703     if (client->opt == NBD_OPT_INFO && !blocksize &&
704         blk_get_request_alignment(exp->common.blk) > 1) {
705         return nbd_negotiate_send_rep_err(client,
706                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
707                                           errp,
708                                           "request NBD_INFO_BLOCK_SIZE to "
709                                           "use this export");
710     }
711 
712     /* Final reply */
713     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
714     if (rc < 0) {
715         return rc;
716     }
717 
718     if (client->opt == NBD_OPT_GO) {
719         client->exp = exp;
720         client->check_align = check_align;
721         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
722         blk_exp_ref(&client->exp->common);
723         nbd_check_meta_export(client);
724         rc = 1;
725     }
726     return rc;
727 }
728 
729 
730 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
731  * new channel for all further (now-encrypted) communication. */
732 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
733                                                  Error **errp)
734 {
735     QIOChannel *ioc;
736     QIOChannelTLS *tioc;
737     struct NBDTLSHandshakeData data = { 0 };
738 
739     assert(client->opt == NBD_OPT_STARTTLS);
740 
741     trace_nbd_negotiate_handle_starttls();
742     ioc = client->ioc;
743 
744     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
745         return NULL;
746     }
747 
748     tioc = qio_channel_tls_new_server(ioc,
749                                       client->tlscreds,
750                                       client->tlsauthz,
751                                       errp);
752     if (!tioc) {
753         return NULL;
754     }
755 
756     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
757     trace_nbd_negotiate_handle_starttls_handshake();
758     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
759     qio_channel_tls_handshake(tioc,
760                               nbd_tls_handshake,
761                               &data,
762                               NULL,
763                               NULL);
764 
765     if (!data.complete) {
766         g_main_loop_run(data.loop);
767     }
768     g_main_loop_unref(data.loop);
769     if (data.error) {
770         object_unref(OBJECT(tioc));
771         error_propagate(errp, data.error);
772         return NULL;
773     }
774 
775     return QIO_CHANNEL(tioc);
776 }
777 
778 /* nbd_negotiate_send_meta_context
779  *
780  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
781  *
782  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
783  */
784 static int nbd_negotiate_send_meta_context(NBDClient *client,
785                                            const char *context,
786                                            uint32_t context_id,
787                                            Error **errp)
788 {
789     NBDOptionReplyMetaContext opt;
790     struct iovec iov[] = {
791         {.iov_base = &opt, .iov_len = sizeof(opt)},
792         {.iov_base = (void *)context, .iov_len = strlen(context)}
793     };
794 
795     assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
796     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
797         context_id = 0;
798     }
799 
800     trace_nbd_negotiate_meta_query_reply(context, context_id);
801     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
802                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
803     stl_be_p(&opt.context_id, context_id);
804 
805     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
806 }
807 
808 /*
809  * Return true if @query matches @pattern, or if @query is empty when
810  * the @client is performing _LIST_.
811  */
812 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
813                                       const char *query)
814 {
815     if (!*query) {
816         trace_nbd_negotiate_meta_query_parse("empty");
817         return client->opt == NBD_OPT_LIST_META_CONTEXT;
818     }
819     if (strcmp(query, pattern) == 0) {
820         trace_nbd_negotiate_meta_query_parse(pattern);
821         return true;
822     }
823     trace_nbd_negotiate_meta_query_skip("pattern not matched");
824     return false;
825 }
826 
827 /*
828  * Return true and adjust @str in place if it begins with @prefix.
829  */
830 static bool nbd_strshift(const char **str, const char *prefix)
831 {
832     size_t len = strlen(prefix);
833 
834     if (strncmp(*str, prefix, len) == 0) {
835         *str += len;
836         return true;
837     }
838     return false;
839 }
840 
841 /* nbd_meta_base_query
842  *
843  * Handle queries to 'base' namespace. For now, only the base:allocation
844  * context is available.  Return true if @query has been handled.
845  */
846 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
847                                 const char *query)
848 {
849     if (!nbd_strshift(&query, "base:")) {
850         return false;
851     }
852     trace_nbd_negotiate_meta_query_parse("base:");
853 
854     if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
855         meta->base_allocation = true;
856     }
857     return true;
858 }
859 
860 /* nbd_meta_qemu_query
861  *
862  * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
863  * and qemu:allocation-depth contexts are available.  Return true if @query
864  * has been handled.
865  */
866 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
867                                 const char *query)
868 {
869     size_t i;
870 
871     if (!nbd_strshift(&query, "qemu:")) {
872         return false;
873     }
874     trace_nbd_negotiate_meta_query_parse("qemu:");
875 
876     if (!*query) {
877         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
878             meta->allocation_depth = meta->exp->allocation_depth;
879             memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
880         }
881         trace_nbd_negotiate_meta_query_parse("empty");
882         return true;
883     }
884 
885     if (strcmp(query, "allocation-depth") == 0) {
886         trace_nbd_negotiate_meta_query_parse("allocation-depth");
887         meta->allocation_depth = meta->exp->allocation_depth;
888         return true;
889     }
890 
891     if (nbd_strshift(&query, "dirty-bitmap:")) {
892         trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
893         if (!*query) {
894             if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
895                 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
896             }
897             trace_nbd_negotiate_meta_query_parse("empty");
898             return true;
899         }
900 
901         for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
902             const char *bm_name;
903 
904             bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
905             if (strcmp(bm_name, query) == 0) {
906                 meta->bitmaps[i] = true;
907                 trace_nbd_negotiate_meta_query_parse(query);
908                 return true;
909             }
910         }
911         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
912         return true;
913     }
914 
915     trace_nbd_negotiate_meta_query_skip("unknown qemu context");
916     return true;
917 }
918 
919 /* nbd_negotiate_meta_query
920  *
921  * Parse namespace name and call corresponding function to parse body of the
922  * query.
923  *
924  * The only supported namespaces are 'base' and 'qemu'.
925  *
926  * Return -errno on I/O error, 0 if option was completely handled by
927  * sending a reply about inconsistent lengths, or 1 on success. */
928 static int nbd_negotiate_meta_query(NBDClient *client,
929                                     NBDExportMetaContexts *meta, Error **errp)
930 {
931     int ret;
932     g_autofree char *query = NULL;
933     uint32_t len;
934 
935     ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
936     if (ret <= 0) {
937         return ret;
938     }
939     len = cpu_to_be32(len);
940 
941     if (len > NBD_MAX_STRING_SIZE) {
942         trace_nbd_negotiate_meta_query_skip("length too long");
943         return nbd_opt_skip(client, len, errp);
944     }
945 
946     query = g_malloc(len + 1);
947     ret = nbd_opt_read(client, query, len, true, errp);
948     if (ret <= 0) {
949         return ret;
950     }
951     query[len] = '\0';
952 
953     if (nbd_meta_base_query(client, meta, query)) {
954         return 1;
955     }
956     if (nbd_meta_qemu_query(client, meta, query)) {
957         return 1;
958     }
959 
960     trace_nbd_negotiate_meta_query_skip("unknown namespace");
961     return 1;
962 }
963 
964 /* nbd_negotiate_meta_queries
965  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
966  *
967  * Return -errno on I/O error, or 0 if option was completely handled. */
968 static int nbd_negotiate_meta_queries(NBDClient *client,
969                                       NBDExportMetaContexts *meta, Error **errp)
970 {
971     int ret;
972     g_autofree char *export_name = NULL;
973     g_autofree bool *bitmaps = NULL;
974     NBDExportMetaContexts local_meta = {0};
975     uint32_t nb_queries;
976     size_t i;
977     size_t count = 0;
978 
979     if (!client->structured_reply) {
980         return nbd_opt_invalid(client, errp,
981                                "request option '%s' when structured reply "
982                                "is not negotiated",
983                                nbd_opt_lookup(client->opt));
984     }
985 
986     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
987         /* Only change the caller's meta on SET. */
988         meta = &local_meta;
989     }
990 
991     g_free(meta->bitmaps);
992     memset(meta, 0, sizeof(*meta));
993 
994     ret = nbd_opt_read_name(client, &export_name, NULL, errp);
995     if (ret <= 0) {
996         return ret;
997     }
998 
999     meta->exp = nbd_export_find(export_name);
1000     if (meta->exp == NULL) {
1001         g_autofree char *sane_name = nbd_sanitize_name(export_name);
1002 
1003         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1004                             "export '%s' not present", sane_name);
1005     }
1006     meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1007     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1008         bitmaps = meta->bitmaps;
1009     }
1010 
1011     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1012     if (ret <= 0) {
1013         return ret;
1014     }
1015     nb_queries = cpu_to_be32(nb_queries);
1016     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1017                                      export_name, nb_queries);
1018 
1019     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1020         /* enable all known contexts */
1021         meta->base_allocation = true;
1022         meta->allocation_depth = meta->exp->allocation_depth;
1023         memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1024     } else {
1025         for (i = 0; i < nb_queries; ++i) {
1026             ret = nbd_negotiate_meta_query(client, meta, errp);
1027             if (ret <= 0) {
1028                 return ret;
1029             }
1030         }
1031     }
1032 
1033     if (meta->base_allocation) {
1034         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1035                                               NBD_META_ID_BASE_ALLOCATION,
1036                                               errp);
1037         if (ret < 0) {
1038             return ret;
1039         }
1040         count++;
1041     }
1042 
1043     if (meta->allocation_depth) {
1044         ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1045                                               NBD_META_ID_ALLOCATION_DEPTH,
1046                                               errp);
1047         if (ret < 0) {
1048             return ret;
1049         }
1050         count++;
1051     }
1052 
1053     for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1054         const char *bm_name;
1055         g_autofree char *context = NULL;
1056 
1057         if (!meta->bitmaps[i]) {
1058             continue;
1059         }
1060 
1061         bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1062         context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1063 
1064         ret = nbd_negotiate_send_meta_context(client, context,
1065                                               NBD_META_ID_DIRTY_BITMAP + i,
1066                                               errp);
1067         if (ret < 0) {
1068             return ret;
1069         }
1070         count++;
1071     }
1072 
1073     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1074     if (ret == 0) {
1075         meta->count = count;
1076     }
1077 
1078     return ret;
1079 }
1080 
1081 /* nbd_negotiate_options
1082  * Process all NBD_OPT_* client option commands, during fixed newstyle
1083  * negotiation.
1084  * Return:
1085  * -errno  on error, errp is set
1086  * 0       on successful negotiation, errp is not set
1087  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1088  *         errp is not set
1089  */
1090 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1091 {
1092     uint32_t flags;
1093     bool fixedNewstyle = false;
1094     bool no_zeroes = false;
1095 
1096     /* Client sends:
1097         [ 0 ..   3]   client flags
1098 
1099        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1100         [ 0 ..   7]   NBD_OPTS_MAGIC
1101         [ 8 ..  11]   NBD option
1102         [12 ..  15]   Data length
1103         ...           Rest of request
1104 
1105         [ 0 ..   7]   NBD_OPTS_MAGIC
1106         [ 8 ..  11]   Second NBD option
1107         [12 ..  15]   Data length
1108         ...           Rest of request
1109     */
1110 
1111     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1112         return -EIO;
1113     }
1114     trace_nbd_negotiate_options_flags(flags);
1115     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1116         fixedNewstyle = true;
1117         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1118     }
1119     if (flags & NBD_FLAG_C_NO_ZEROES) {
1120         no_zeroes = true;
1121         flags &= ~NBD_FLAG_C_NO_ZEROES;
1122     }
1123     if (flags != 0) {
1124         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1125         return -EINVAL;
1126     }
1127 
1128     while (1) {
1129         int ret;
1130         uint32_t option, length;
1131         uint64_t magic;
1132 
1133         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1134             return -EINVAL;
1135         }
1136         trace_nbd_negotiate_options_check_magic(magic);
1137         if (magic != NBD_OPTS_MAGIC) {
1138             error_setg(errp, "Bad magic received");
1139             return -EINVAL;
1140         }
1141 
1142         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1143             return -EINVAL;
1144         }
1145         client->opt = option;
1146 
1147         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1148             return -EINVAL;
1149         }
1150         assert(!client->optlen);
1151         client->optlen = length;
1152 
1153         if (length > NBD_MAX_BUFFER_SIZE) {
1154             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1155                        length, NBD_MAX_BUFFER_SIZE);
1156             return -EINVAL;
1157         }
1158 
1159         trace_nbd_negotiate_options_check_option(option,
1160                                                  nbd_opt_lookup(option));
1161         if (client->tlscreds &&
1162             client->ioc == (QIOChannel *)client->sioc) {
1163             QIOChannel *tioc;
1164             if (!fixedNewstyle) {
1165                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1166                 return -EINVAL;
1167             }
1168             switch (option) {
1169             case NBD_OPT_STARTTLS:
1170                 if (length) {
1171                     /* Unconditionally drop the connection if the client
1172                      * can't start a TLS negotiation correctly */
1173                     return nbd_reject_length(client, true, errp);
1174                 }
1175                 tioc = nbd_negotiate_handle_starttls(client, errp);
1176                 if (!tioc) {
1177                     return -EIO;
1178                 }
1179                 ret = 0;
1180                 object_unref(OBJECT(client->ioc));
1181                 client->ioc = QIO_CHANNEL(tioc);
1182                 break;
1183 
1184             case NBD_OPT_EXPORT_NAME:
1185                 /* No way to return an error to client, so drop connection */
1186                 error_setg(errp, "Option 0x%x not permitted before TLS",
1187                            option);
1188                 return -EINVAL;
1189 
1190             default:
1191                 /* Let the client keep trying, unless they asked to
1192                  * quit. Always try to give an error back to the
1193                  * client; but when replying to OPT_ABORT, be aware
1194                  * that the client may hang up before receiving the
1195                  * error, in which case we are fine ignoring the
1196                  * resulting EPIPE. */
1197                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1198                                    option == NBD_OPT_ABORT ? NULL : errp,
1199                                    "Option 0x%" PRIx32
1200                                    " not permitted before TLS", option);
1201                 if (option == NBD_OPT_ABORT) {
1202                     return 1;
1203                 }
1204                 break;
1205             }
1206         } else if (fixedNewstyle) {
1207             switch (option) {
1208             case NBD_OPT_LIST:
1209                 if (length) {
1210                     ret = nbd_reject_length(client, false, errp);
1211                 } else {
1212                     ret = nbd_negotiate_handle_list(client, errp);
1213                 }
1214                 break;
1215 
1216             case NBD_OPT_ABORT:
1217                 /* NBD spec says we must try to reply before
1218                  * disconnecting, but that we must also tolerate
1219                  * guests that don't wait for our reply. */
1220                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1221                 return 1;
1222 
1223             case NBD_OPT_EXPORT_NAME:
1224                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1225                                                         errp);
1226 
1227             case NBD_OPT_INFO:
1228             case NBD_OPT_GO:
1229                 ret = nbd_negotiate_handle_info(client, errp);
1230                 if (ret == 1) {
1231                     assert(option == NBD_OPT_GO);
1232                     return 0;
1233                 }
1234                 break;
1235 
1236             case NBD_OPT_STARTTLS:
1237                 if (length) {
1238                     ret = nbd_reject_length(client, false, errp);
1239                 } else if (client->tlscreds) {
1240                     ret = nbd_negotiate_send_rep_err(client,
1241                                                      NBD_REP_ERR_INVALID, errp,
1242                                                      "TLS already enabled");
1243                 } else {
1244                     ret = nbd_negotiate_send_rep_err(client,
1245                                                      NBD_REP_ERR_POLICY, errp,
1246                                                      "TLS not configured");
1247                 }
1248                 break;
1249 
1250             case NBD_OPT_STRUCTURED_REPLY:
1251                 if (length) {
1252                     ret = nbd_reject_length(client, false, errp);
1253                 } else if (client->structured_reply) {
1254                     ret = nbd_negotiate_send_rep_err(
1255                         client, NBD_REP_ERR_INVALID, errp,
1256                         "structured reply already negotiated");
1257                 } else {
1258                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1259                     client->structured_reply = true;
1260                 }
1261                 break;
1262 
1263             case NBD_OPT_LIST_META_CONTEXT:
1264             case NBD_OPT_SET_META_CONTEXT:
1265                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1266                                                  errp);
1267                 break;
1268 
1269             default:
1270                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1271                                    "Unsupported option %" PRIu32 " (%s)",
1272                                    option, nbd_opt_lookup(option));
1273                 break;
1274             }
1275         } else {
1276             /*
1277              * If broken new-style we should drop the connection
1278              * for anything except NBD_OPT_EXPORT_NAME
1279              */
1280             switch (option) {
1281             case NBD_OPT_EXPORT_NAME:
1282                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1283                                                         errp);
1284 
1285             default:
1286                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1287                            option, nbd_opt_lookup(option));
1288                 return -EINVAL;
1289             }
1290         }
1291         if (ret < 0) {
1292             return ret;
1293         }
1294     }
1295 }
1296 
1297 /* nbd_negotiate
1298  * Return:
1299  * -errno  on error, errp is set
1300  * 0       on successful negotiation, errp is not set
1301  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1302  *         errp is not set
1303  */
1304 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1305 {
1306     ERRP_GUARD();
1307     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1308     int ret;
1309 
1310     /* Old style negotiation header, no room for options
1311         [ 0 ..   7]   passwd       ("NBDMAGIC")
1312         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1313         [16 ..  23]   size
1314         [24 ..  27]   export flags (zero-extended)
1315         [28 .. 151]   reserved     (0)
1316 
1317        New style negotiation header, client can send options
1318         [ 0 ..   7]   passwd       ("NBDMAGIC")
1319         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1320         [16 ..  17]   server flags (0)
1321         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1322      */
1323 
1324     qio_channel_set_blocking(client->ioc, false, NULL);
1325 
1326     trace_nbd_negotiate_begin();
1327     memcpy(buf, "NBDMAGIC", 8);
1328 
1329     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1330     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1331 
1332     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1333         error_prepend(errp, "write failed: ");
1334         return -EINVAL;
1335     }
1336     ret = nbd_negotiate_options(client, errp);
1337     if (ret != 0) {
1338         if (ret < 0) {
1339             error_prepend(errp, "option negotiation failed: ");
1340         }
1341         return ret;
1342     }
1343 
1344     /* Attach the channel to the same AioContext as the export */
1345     if (client->exp && client->exp->common.ctx) {
1346         qio_channel_attach_aio_context(client->ioc, client->exp->common.ctx);
1347     }
1348 
1349     assert(!client->optlen);
1350     trace_nbd_negotiate_success();
1351 
1352     return 0;
1353 }
1354 
1355 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1356                                Error **errp)
1357 {
1358     uint8_t buf[NBD_REQUEST_SIZE];
1359     uint32_t magic;
1360     int ret;
1361 
1362     ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1363     if (ret < 0) {
1364         return ret;
1365     }
1366 
1367     /* Request
1368        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1369        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1370        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1371        [ 8 .. 15]   handle
1372        [16 .. 23]   from
1373        [24 .. 27]   len
1374      */
1375 
1376     magic = ldl_be_p(buf);
1377     request->flags  = lduw_be_p(buf + 4);
1378     request->type   = lduw_be_p(buf + 6);
1379     request->handle = ldq_be_p(buf + 8);
1380     request->from   = ldq_be_p(buf + 16);
1381     request->len    = ldl_be_p(buf + 24);
1382 
1383     trace_nbd_receive_request(magic, request->flags, request->type,
1384                               request->from, request->len);
1385 
1386     if (magic != NBD_REQUEST_MAGIC) {
1387         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1388         return -EINVAL;
1389     }
1390     return 0;
1391 }
1392 
1393 #define MAX_NBD_REQUESTS 16
1394 
1395 void nbd_client_get(NBDClient *client)
1396 {
1397     client->refcount++;
1398 }
1399 
1400 void nbd_client_put(NBDClient *client)
1401 {
1402     if (--client->refcount == 0) {
1403         /* The last reference should be dropped by client->close,
1404          * which is called by client_close.
1405          */
1406         assert(client->closing);
1407 
1408         qio_channel_detach_aio_context(client->ioc);
1409         object_unref(OBJECT(client->sioc));
1410         object_unref(OBJECT(client->ioc));
1411         if (client->tlscreds) {
1412             object_unref(OBJECT(client->tlscreds));
1413         }
1414         g_free(client->tlsauthz);
1415         if (client->exp) {
1416             QTAILQ_REMOVE(&client->exp->clients, client, next);
1417             blk_exp_unref(&client->exp->common);
1418         }
1419         g_free(client->export_meta.bitmaps);
1420         g_free(client);
1421     }
1422 }
1423 
1424 static void client_close(NBDClient *client, bool negotiated)
1425 {
1426     if (client->closing) {
1427         return;
1428     }
1429 
1430     client->closing = true;
1431 
1432     /* Force requests to finish.  They will drop their own references,
1433      * then we'll close the socket and free the NBDClient.
1434      */
1435     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1436                          NULL);
1437 
1438     /* Also tell the client, so that they release their reference.  */
1439     if (client->close_fn) {
1440         client->close_fn(client, negotiated);
1441     }
1442 }
1443 
1444 static NBDRequestData *nbd_request_get(NBDClient *client)
1445 {
1446     NBDRequestData *req;
1447 
1448     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1449     client->nb_requests++;
1450 
1451     req = g_new0(NBDRequestData, 1);
1452     nbd_client_get(client);
1453     req->client = client;
1454     return req;
1455 }
1456 
1457 static void nbd_request_put(NBDRequestData *req)
1458 {
1459     NBDClient *client = req->client;
1460 
1461     if (req->data) {
1462         qemu_vfree(req->data);
1463     }
1464     g_free(req);
1465 
1466     client->nb_requests--;
1467     nbd_client_receive_next_request(client);
1468 
1469     nbd_client_put(client);
1470 }
1471 
1472 static void blk_aio_attached(AioContext *ctx, void *opaque)
1473 {
1474     NBDExport *exp = opaque;
1475     NBDClient *client;
1476 
1477     trace_nbd_blk_aio_attached(exp->name, ctx);
1478 
1479     exp->common.ctx = ctx;
1480 
1481     QTAILQ_FOREACH(client, &exp->clients, next) {
1482         qio_channel_attach_aio_context(client->ioc, ctx);
1483         if (client->recv_coroutine) {
1484             aio_co_schedule(ctx, client->recv_coroutine);
1485         }
1486         if (client->send_coroutine) {
1487             aio_co_schedule(ctx, client->send_coroutine);
1488         }
1489     }
1490 }
1491 
1492 static void blk_aio_detach(void *opaque)
1493 {
1494     NBDExport *exp = opaque;
1495     NBDClient *client;
1496 
1497     trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1498 
1499     QTAILQ_FOREACH(client, &exp->clients, next) {
1500         qio_channel_detach_aio_context(client->ioc);
1501     }
1502 
1503     exp->common.ctx = NULL;
1504 }
1505 
1506 static void nbd_eject_notifier(Notifier *n, void *data)
1507 {
1508     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1509 
1510     blk_exp_request_shutdown(&exp->common);
1511 }
1512 
1513 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1514 {
1515     NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1516     assert(exp->drv == &blk_exp_nbd);
1517     assert(nbd_exp->eject_notifier_blk == NULL);
1518 
1519     blk_ref(blk);
1520     nbd_exp->eject_notifier_blk = blk;
1521     nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1522     blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1523 }
1524 
1525 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1526                              Error **errp)
1527 {
1528     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1529     BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1530     BlockBackend *blk = blk_exp->blk;
1531     int64_t size;
1532     uint64_t perm, shared_perm;
1533     bool readonly = !exp_args->writable;
1534     bool shared = !exp_args->writable;
1535     strList *bitmaps;
1536     size_t i;
1537     int ret;
1538 
1539     assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1540 
1541     if (!nbd_server_is_running()) {
1542         error_setg(errp, "NBD server not running");
1543         return -EINVAL;
1544     }
1545 
1546     if (!arg->has_name) {
1547         arg->name = exp_args->node_name;
1548     }
1549 
1550     if (strlen(arg->name) > NBD_MAX_STRING_SIZE) {
1551         error_setg(errp, "export name '%s' too long", arg->name);
1552         return -EINVAL;
1553     }
1554 
1555     if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1556         error_setg(errp, "description '%s' too long", arg->description);
1557         return -EINVAL;
1558     }
1559 
1560     if (nbd_export_find(arg->name)) {
1561         error_setg(errp, "NBD server already has export named '%s'", arg->name);
1562         return -EEXIST;
1563     }
1564 
1565     size = blk_getlength(blk);
1566     if (size < 0) {
1567         error_setg_errno(errp, -size,
1568                          "Failed to determine the NBD export's length");
1569         return size;
1570     }
1571 
1572     /* Don't allow resize while the NBD server is running, otherwise we don't
1573      * care what happens with the node. */
1574     blk_get_perm(blk, &perm, &shared_perm);
1575     ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1576     if (ret < 0) {
1577         return ret;
1578     }
1579 
1580     QTAILQ_INIT(&exp->clients);
1581     exp->name = g_strdup(arg->name);
1582     exp->description = g_strdup(arg->description);
1583     exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1584                      NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1585     if (readonly) {
1586         exp->nbdflags |= NBD_FLAG_READ_ONLY;
1587         if (shared) {
1588             exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1589         }
1590     } else {
1591         exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1592                           NBD_FLAG_SEND_FAST_ZERO);
1593     }
1594     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1595 
1596     for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1597         exp->nr_export_bitmaps++;
1598     }
1599     exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1600     for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1601          i++, bitmaps = bitmaps->next) {
1602         const char *bitmap = bitmaps->value;
1603         BlockDriverState *bs = blk_bs(blk);
1604         BdrvDirtyBitmap *bm = NULL;
1605 
1606         while (bs) {
1607             bm = bdrv_find_dirty_bitmap(bs, bitmap);
1608             if (bm != NULL) {
1609                 break;
1610             }
1611 
1612             bs = bdrv_filter_or_cow_bs(bs);
1613         }
1614 
1615         if (bm == NULL) {
1616             ret = -ENOENT;
1617             error_setg(errp, "Bitmap '%s' is not found", bitmap);
1618             goto fail;
1619         }
1620 
1621         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1622             ret = -EINVAL;
1623             goto fail;
1624         }
1625 
1626         if (readonly && bdrv_is_writable(bs) &&
1627             bdrv_dirty_bitmap_enabled(bm)) {
1628             ret = -EINVAL;
1629             error_setg(errp,
1630                        "Enabled bitmap '%s' incompatible with readonly export",
1631                        bitmap);
1632             goto fail;
1633         }
1634 
1635         exp->export_bitmaps[i] = bm;
1636         assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1637     }
1638 
1639     /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1640     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1641         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1642     }
1643 
1644     exp->allocation_depth = arg->allocation_depth;
1645 
1646     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1647 
1648     QTAILQ_INSERT_TAIL(&exports, exp, next);
1649 
1650     return 0;
1651 
1652 fail:
1653     g_free(exp->export_bitmaps);
1654     g_free(exp->name);
1655     g_free(exp->description);
1656     return ret;
1657 }
1658 
1659 NBDExport *nbd_export_find(const char *name)
1660 {
1661     NBDExport *exp;
1662     QTAILQ_FOREACH(exp, &exports, next) {
1663         if (strcmp(name, exp->name) == 0) {
1664             return exp;
1665         }
1666     }
1667 
1668     return NULL;
1669 }
1670 
1671 AioContext *
1672 nbd_export_aio_context(NBDExport *exp)
1673 {
1674     return exp->common.ctx;
1675 }
1676 
1677 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1678 {
1679     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1680     NBDClient *client, *next;
1681 
1682     blk_exp_ref(&exp->common);
1683     /*
1684      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1685      * close mode that stops advertising the export to new clients but
1686      * still permits existing clients to run to completion? Because of
1687      * that possibility, nbd_export_close() can be called more than
1688      * once on an export.
1689      */
1690     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1691         client_close(client, true);
1692     }
1693     if (exp->name) {
1694         g_free(exp->name);
1695         exp->name = NULL;
1696         QTAILQ_REMOVE(&exports, exp, next);
1697     }
1698     blk_exp_unref(&exp->common);
1699 }
1700 
1701 static void nbd_export_delete(BlockExport *blk_exp)
1702 {
1703     size_t i;
1704     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1705 
1706     assert(exp->name == NULL);
1707     assert(QTAILQ_EMPTY(&exp->clients));
1708 
1709     g_free(exp->description);
1710     exp->description = NULL;
1711 
1712     if (exp->common.blk) {
1713         if (exp->eject_notifier_blk) {
1714             notifier_remove(&exp->eject_notifier);
1715             blk_unref(exp->eject_notifier_blk);
1716         }
1717         blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1718                                         blk_aio_detach, exp);
1719     }
1720 
1721     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1722         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1723     }
1724 }
1725 
1726 const BlockExportDriver blk_exp_nbd = {
1727     .type               = BLOCK_EXPORT_TYPE_NBD,
1728     .instance_size      = sizeof(NBDExport),
1729     .create             = nbd_export_create,
1730     .delete             = nbd_export_delete,
1731     .request_shutdown   = nbd_export_request_shutdown,
1732 };
1733 
1734 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1735                                         unsigned niov, Error **errp)
1736 {
1737     int ret;
1738 
1739     g_assert(qemu_in_coroutine());
1740     qemu_co_mutex_lock(&client->send_lock);
1741     client->send_coroutine = qemu_coroutine_self();
1742 
1743     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1744 
1745     client->send_coroutine = NULL;
1746     qemu_co_mutex_unlock(&client->send_lock);
1747 
1748     return ret;
1749 }
1750 
1751 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1752                                        uint64_t handle)
1753 {
1754     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1755     stl_be_p(&reply->error, error);
1756     stq_be_p(&reply->handle, handle);
1757 }
1758 
1759 static int nbd_co_send_simple_reply(NBDClient *client,
1760                                     uint64_t handle,
1761                                     uint32_t error,
1762                                     void *data,
1763                                     size_t len,
1764                                     Error **errp)
1765 {
1766     NBDSimpleReply reply;
1767     int nbd_err = system_errno_to_nbd_errno(error);
1768     struct iovec iov[] = {
1769         {.iov_base = &reply, .iov_len = sizeof(reply)},
1770         {.iov_base = data, .iov_len = len}
1771     };
1772 
1773     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1774                                    len);
1775     set_be_simple_reply(&reply, nbd_err, handle);
1776 
1777     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1778 }
1779 
1780 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1781                                 uint16_t type, uint64_t handle, uint32_t length)
1782 {
1783     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1784     stw_be_p(&chunk->flags, flags);
1785     stw_be_p(&chunk->type, type);
1786     stq_be_p(&chunk->handle, handle);
1787     stl_be_p(&chunk->length, length);
1788 }
1789 
1790 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1791                                                     uint64_t handle,
1792                                                     Error **errp)
1793 {
1794     NBDStructuredReplyChunk chunk;
1795     struct iovec iov[] = {
1796         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1797     };
1798 
1799     trace_nbd_co_send_structured_done(handle);
1800     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1801 
1802     return nbd_co_send_iov(client, iov, 1, errp);
1803 }
1804 
1805 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1806                                                     uint64_t handle,
1807                                                     uint64_t offset,
1808                                                     void *data,
1809                                                     size_t size,
1810                                                     bool final,
1811                                                     Error **errp)
1812 {
1813     NBDStructuredReadData chunk;
1814     struct iovec iov[] = {
1815         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1816         {.iov_base = data, .iov_len = size}
1817     };
1818 
1819     assert(size);
1820     trace_nbd_co_send_structured_read(handle, offset, data, size);
1821     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1822                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1823                  sizeof(chunk) - sizeof(chunk.h) + size);
1824     stq_be_p(&chunk.offset, offset);
1825 
1826     return nbd_co_send_iov(client, iov, 2, errp);
1827 }
1828 
1829 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1830                                                      uint64_t handle,
1831                                                      uint32_t error,
1832                                                      const char *msg,
1833                                                      Error **errp)
1834 {
1835     NBDStructuredError chunk;
1836     int nbd_err = system_errno_to_nbd_errno(error);
1837     struct iovec iov[] = {
1838         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1839         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1840     };
1841 
1842     assert(nbd_err);
1843     trace_nbd_co_send_structured_error(handle, nbd_err,
1844                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1845     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1846                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1847     stl_be_p(&chunk.error, nbd_err);
1848     stw_be_p(&chunk.message_length, iov[1].iov_len);
1849 
1850     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1851 }
1852 
1853 /* Do a sparse read and send the structured reply to the client.
1854  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1855  * reported to the client, at which point this function succeeds.
1856  */
1857 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1858                                                 uint64_t handle,
1859                                                 uint64_t offset,
1860                                                 uint8_t *data,
1861                                                 size_t size,
1862                                                 Error **errp)
1863 {
1864     int ret = 0;
1865     NBDExport *exp = client->exp;
1866     size_t progress = 0;
1867 
1868     while (progress < size) {
1869         int64_t pnum;
1870         int status = bdrv_block_status_above(blk_bs(exp->common.blk), NULL,
1871                                              offset + progress,
1872                                              size - progress, &pnum, NULL,
1873                                              NULL);
1874         bool final;
1875 
1876         if (status < 0) {
1877             char *msg = g_strdup_printf("unable to check for holes: %s",
1878                                         strerror(-status));
1879 
1880             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1881                                                errp);
1882             g_free(msg);
1883             return ret;
1884         }
1885         assert(pnum && pnum <= size - progress);
1886         final = progress + pnum == size;
1887         if (status & BDRV_BLOCK_ZERO) {
1888             NBDStructuredReadHole chunk;
1889             struct iovec iov[] = {
1890                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1891             };
1892 
1893             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1894                                                    pnum);
1895             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1896                          NBD_REPLY_TYPE_OFFSET_HOLE,
1897                          handle, sizeof(chunk) - sizeof(chunk.h));
1898             stq_be_p(&chunk.offset, offset + progress);
1899             stl_be_p(&chunk.length, pnum);
1900             ret = nbd_co_send_iov(client, iov, 1, errp);
1901         } else {
1902             ret = blk_pread(exp->common.blk, offset + progress,
1903                             data + progress, pnum);
1904             if (ret < 0) {
1905                 error_setg_errno(errp, -ret, "reading from file failed");
1906                 break;
1907             }
1908             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1909                                               data + progress, pnum, final,
1910                                               errp);
1911         }
1912 
1913         if (ret < 0) {
1914             break;
1915         }
1916         progress += pnum;
1917     }
1918     return ret;
1919 }
1920 
1921 typedef struct NBDExtentArray {
1922     NBDExtent *extents;
1923     unsigned int nb_alloc;
1924     unsigned int count;
1925     uint64_t total_length;
1926     bool can_add;
1927     bool converted_to_be;
1928 } NBDExtentArray;
1929 
1930 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc)
1931 {
1932     NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
1933 
1934     ea->nb_alloc = nb_alloc;
1935     ea->extents = g_new(NBDExtent, nb_alloc);
1936     ea->can_add = true;
1937 
1938     return ea;
1939 }
1940 
1941 static void nbd_extent_array_free(NBDExtentArray *ea)
1942 {
1943     g_free(ea->extents);
1944     g_free(ea);
1945 }
1946 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free);
1947 
1948 /* Further modifications of the array after conversion are abandoned */
1949 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
1950 {
1951     int i;
1952 
1953     assert(!ea->converted_to_be);
1954     ea->can_add = false;
1955     ea->converted_to_be = true;
1956 
1957     for (i = 0; i < ea->count; i++) {
1958         ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags);
1959         ea->extents[i].length = cpu_to_be32(ea->extents[i].length);
1960     }
1961 }
1962 
1963 /*
1964  * Add extent to NBDExtentArray. If extent can't be added (no available space),
1965  * return -1.
1966  * For safety, when returning -1 for the first time, .can_add is set to false,
1967  * further call to nbd_extent_array_add() will crash.
1968  * (to avoid the situation, when after failing to add an extent (returned -1),
1969  * user miss this failure and add another extent, which is successfully added
1970  * (array is full, but new extent may be squashed into the last one), then we
1971  * have invalid array with skipped extent)
1972  */
1973 static int nbd_extent_array_add(NBDExtentArray *ea,
1974                                 uint32_t length, uint32_t flags)
1975 {
1976     assert(ea->can_add);
1977 
1978     if (!length) {
1979         return 0;
1980     }
1981 
1982     /* Extend previous extent if flags are the same */
1983     if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
1984         uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length;
1985 
1986         if (sum <= UINT32_MAX) {
1987             ea->extents[ea->count - 1].length = sum;
1988             ea->total_length += length;
1989             return 0;
1990         }
1991     }
1992 
1993     if (ea->count >= ea->nb_alloc) {
1994         ea->can_add = false;
1995         return -1;
1996     }
1997 
1998     ea->total_length += length;
1999     ea->extents[ea->count] = (NBDExtent) {.length = length, .flags = flags};
2000     ea->count++;
2001 
2002     return 0;
2003 }
2004 
2005 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
2006                                   uint64_t bytes, NBDExtentArray *ea)
2007 {
2008     while (bytes) {
2009         uint32_t flags;
2010         int64_t num;
2011         int ret = bdrv_block_status_above(bs, NULL, offset, bytes, &num,
2012                                           NULL, NULL);
2013 
2014         if (ret < 0) {
2015             return ret;
2016         }
2017 
2018         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
2019                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
2020 
2021         if (nbd_extent_array_add(ea, num, flags) < 0) {
2022             return 0;
2023         }
2024 
2025         offset += num;
2026         bytes -= num;
2027     }
2028 
2029     return 0;
2030 }
2031 
2032 static int blockalloc_to_extents(BlockDriverState *bs, uint64_t offset,
2033                                  uint64_t bytes, NBDExtentArray *ea)
2034 {
2035     while (bytes) {
2036         int64_t num;
2037         int ret = bdrv_is_allocated_above(bs, NULL, false, offset, bytes,
2038                                           &num);
2039 
2040         if (ret < 0) {
2041             return ret;
2042         }
2043 
2044         if (nbd_extent_array_add(ea, num, ret) < 0) {
2045             return 0;
2046         }
2047 
2048         offset += num;
2049         bytes -= num;
2050     }
2051 
2052     return 0;
2053 }
2054 
2055 /*
2056  * nbd_co_send_extents
2057  *
2058  * @ea is converted to BE by the function
2059  * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2060  */
2061 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
2062                                NBDExtentArray *ea,
2063                                bool last, uint32_t context_id, Error **errp)
2064 {
2065     NBDStructuredMeta chunk;
2066     struct iovec iov[] = {
2067         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2068         {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])}
2069     };
2070 
2071     nbd_extent_array_convert_to_be(ea);
2072 
2073     trace_nbd_co_send_extents(handle, ea->count, context_id, ea->total_length,
2074                               last);
2075     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
2076                  NBD_REPLY_TYPE_BLOCK_STATUS,
2077                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
2078     stl_be_p(&chunk.context_id, context_id);
2079 
2080     return nbd_co_send_iov(client, iov, 2, errp);
2081 }
2082 
2083 /* Get block status from the exported device and send it to the client */
2084 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
2085                                     BlockDriverState *bs, uint64_t offset,
2086                                     uint32_t length, bool dont_fragment,
2087                                     bool last, uint32_t context_id,
2088                                     Error **errp)
2089 {
2090     int ret;
2091     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2092     g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2093 
2094     if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2095         ret = blockstatus_to_extents(bs, offset, length, ea);
2096     } else {
2097         ret = blockalloc_to_extents(bs, offset, length, ea);
2098     }
2099     if (ret < 0) {
2100         return nbd_co_send_structured_error(
2101                 client, handle, -ret, "can't get block status", errp);
2102     }
2103 
2104     return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2105 }
2106 
2107 /* Populate @ea from a dirty bitmap. */
2108 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2109                               uint64_t offset, uint64_t length,
2110                               NBDExtentArray *es)
2111 {
2112     int64_t start, dirty_start, dirty_count;
2113     int64_t end = offset + length;
2114     bool full = false;
2115 
2116     bdrv_dirty_bitmap_lock(bitmap);
2117 
2118     for (start = offset;
2119          bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX,
2120                                            &dirty_start, &dirty_count);
2121          start = dirty_start + dirty_count)
2122     {
2123         if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2124             (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2125         {
2126             full = true;
2127             break;
2128         }
2129     }
2130 
2131     if (!full) {
2132         /* last non dirty extent, nothing to do if array is now full */
2133         (void) nbd_extent_array_add(es, end - start, 0);
2134     }
2135 
2136     bdrv_dirty_bitmap_unlock(bitmap);
2137 }
2138 
2139 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2140                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2141                               uint32_t length, bool dont_fragment, bool last,
2142                               uint32_t context_id, Error **errp)
2143 {
2144     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2145     g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents);
2146 
2147     bitmap_to_extents(bitmap, offset, length, ea);
2148 
2149     return nbd_co_send_extents(client, handle, ea, last, context_id, errp);
2150 }
2151 
2152 /* nbd_co_receive_request
2153  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2154  * connection right away, and any other negative value to report an error to
2155  * the client (although the caller may still need to disconnect after reporting
2156  * the error).
2157  */
2158 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2159                                   Error **errp)
2160 {
2161     NBDClient *client = req->client;
2162     int valid_flags;
2163 
2164     g_assert(qemu_in_coroutine());
2165     assert(client->recv_coroutine == qemu_coroutine_self());
2166     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2167         return -EIO;
2168     }
2169 
2170     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2171                                              nbd_cmd_lookup(request->type));
2172 
2173     if (request->type != NBD_CMD_WRITE) {
2174         /* No payload, we are ready to read the next request.  */
2175         req->complete = true;
2176     }
2177 
2178     if (request->type == NBD_CMD_DISC) {
2179         /* Special case: we're going to disconnect without a reply,
2180          * whether or not flags, from, or len are bogus */
2181         return -EIO;
2182     }
2183 
2184     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2185         request->type == NBD_CMD_CACHE)
2186     {
2187         if (request->len > NBD_MAX_BUFFER_SIZE) {
2188             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2189                        request->len, NBD_MAX_BUFFER_SIZE);
2190             return -EINVAL;
2191         }
2192 
2193         if (request->type != NBD_CMD_CACHE) {
2194             req->data = blk_try_blockalign(client->exp->common.blk,
2195                                            request->len);
2196             if (req->data == NULL) {
2197                 error_setg(errp, "No memory");
2198                 return -ENOMEM;
2199             }
2200         }
2201     }
2202 
2203     if (request->type == NBD_CMD_WRITE) {
2204         if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2205                      errp) < 0)
2206         {
2207             return -EIO;
2208         }
2209         req->complete = true;
2210 
2211         trace_nbd_co_receive_request_payload_received(request->handle,
2212                                                       request->len);
2213     }
2214 
2215     /* Sanity checks. */
2216     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2217         (request->type == NBD_CMD_WRITE ||
2218          request->type == NBD_CMD_WRITE_ZEROES ||
2219          request->type == NBD_CMD_TRIM)) {
2220         error_setg(errp, "Export is read-only");
2221         return -EROFS;
2222     }
2223     if (request->from > client->exp->size ||
2224         request->len > client->exp->size - request->from) {
2225         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2226                    ", Size: %" PRIu64, request->from, request->len,
2227                    client->exp->size);
2228         return (request->type == NBD_CMD_WRITE ||
2229                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2230     }
2231     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2232                                                 client->check_align)) {
2233         /*
2234          * The block layer gracefully handles unaligned requests, but
2235          * it's still worth tracing client non-compliance
2236          */
2237         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2238                                               request->from,
2239                                               request->len,
2240                                               client->check_align);
2241     }
2242     valid_flags = NBD_CMD_FLAG_FUA;
2243     if (request->type == NBD_CMD_READ && client->structured_reply) {
2244         valid_flags |= NBD_CMD_FLAG_DF;
2245     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2246         valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2247     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2248         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2249     }
2250     if (request->flags & ~valid_flags) {
2251         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2252                    nbd_cmd_lookup(request->type), request->flags);
2253         return -EINVAL;
2254     }
2255 
2256     return 0;
2257 }
2258 
2259 /* Send simple reply without a payload, or a structured error
2260  * @error_msg is ignored if @ret >= 0
2261  * Returns 0 if connection is still live, -errno on failure to talk to client
2262  */
2263 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2264                                                uint64_t handle,
2265                                                int ret,
2266                                                const char *error_msg,
2267                                                Error **errp)
2268 {
2269     if (client->structured_reply && ret < 0) {
2270         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2271                                             errp);
2272     } else {
2273         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2274                                         NULL, 0, errp);
2275     }
2276 }
2277 
2278 /* Handle NBD_CMD_READ request.
2279  * Return -errno if sending fails. Other errors are reported directly to the
2280  * client as an error reply. */
2281 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2282                                         uint8_t *data, Error **errp)
2283 {
2284     int ret;
2285     NBDExport *exp = client->exp;
2286 
2287     assert(request->type == NBD_CMD_READ);
2288 
2289     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2290     if (request->flags & NBD_CMD_FLAG_FUA) {
2291         ret = blk_co_flush(exp->common.blk);
2292         if (ret < 0) {
2293             return nbd_send_generic_reply(client, request->handle, ret,
2294                                           "flush failed", errp);
2295         }
2296     }
2297 
2298     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2299         request->len)
2300     {
2301         return nbd_co_send_sparse_read(client, request->handle, request->from,
2302                                        data, request->len, errp);
2303     }
2304 
2305     ret = blk_pread(exp->common.blk, request->from, data, request->len);
2306     if (ret < 0) {
2307         return nbd_send_generic_reply(client, request->handle, ret,
2308                                       "reading from file failed", errp);
2309     }
2310 
2311     if (client->structured_reply) {
2312         if (request->len) {
2313             return nbd_co_send_structured_read(client, request->handle,
2314                                                request->from, data,
2315                                                request->len, true, errp);
2316         } else {
2317             return nbd_co_send_structured_done(client, request->handle, errp);
2318         }
2319     } else {
2320         return nbd_co_send_simple_reply(client, request->handle, 0,
2321                                         data, request->len, errp);
2322     }
2323 }
2324 
2325 /*
2326  * nbd_do_cmd_cache
2327  *
2328  * Handle NBD_CMD_CACHE request.
2329  * Return -errno if sending fails. Other errors are reported directly to the
2330  * client as an error reply.
2331  */
2332 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2333                                          Error **errp)
2334 {
2335     int ret;
2336     NBDExport *exp = client->exp;
2337 
2338     assert(request->type == NBD_CMD_CACHE);
2339 
2340     ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2341                         NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2342 
2343     return nbd_send_generic_reply(client, request->handle, ret,
2344                                   "caching data failed", errp);
2345 }
2346 
2347 /* Handle NBD request.
2348  * Return -errno if sending fails. Other errors are reported directly to the
2349  * client as an error reply. */
2350 static coroutine_fn int nbd_handle_request(NBDClient *client,
2351                                            NBDRequest *request,
2352                                            uint8_t *data, Error **errp)
2353 {
2354     int ret;
2355     int flags;
2356     NBDExport *exp = client->exp;
2357     char *msg;
2358     size_t i;
2359 
2360     switch (request->type) {
2361     case NBD_CMD_CACHE:
2362         return nbd_do_cmd_cache(client, request, errp);
2363 
2364     case NBD_CMD_READ:
2365         return nbd_do_cmd_read(client, request, data, errp);
2366 
2367     case NBD_CMD_WRITE:
2368         flags = 0;
2369         if (request->flags & NBD_CMD_FLAG_FUA) {
2370             flags |= BDRV_REQ_FUA;
2371         }
2372         ret = blk_pwrite(exp->common.blk, request->from, data, request->len,
2373                          flags);
2374         return nbd_send_generic_reply(client, request->handle, ret,
2375                                       "writing to file failed", errp);
2376 
2377     case NBD_CMD_WRITE_ZEROES:
2378         flags = 0;
2379         if (request->flags & NBD_CMD_FLAG_FUA) {
2380             flags |= BDRV_REQ_FUA;
2381         }
2382         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2383             flags |= BDRV_REQ_MAY_UNMAP;
2384         }
2385         if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2386             flags |= BDRV_REQ_NO_FALLBACK;
2387         }
2388         ret = 0;
2389         /* FIXME simplify this when blk_pwrite_zeroes switches to 64-bit */
2390         while (ret >= 0 && request->len) {
2391             int align = client->check_align ?: 1;
2392             int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2393                                                         align));
2394             ret = blk_pwrite_zeroes(exp->common.blk, request->from, len, flags);
2395             request->len -= len;
2396             request->from += len;
2397         }
2398         return nbd_send_generic_reply(client, request->handle, ret,
2399                                       "writing to file failed", errp);
2400 
2401     case NBD_CMD_DISC:
2402         /* unreachable, thanks to special case in nbd_co_receive_request() */
2403         abort();
2404 
2405     case NBD_CMD_FLUSH:
2406         ret = blk_co_flush(exp->common.blk);
2407         return nbd_send_generic_reply(client, request->handle, ret,
2408                                       "flush failed", errp);
2409 
2410     case NBD_CMD_TRIM:
2411         ret = 0;
2412         /* FIXME simplify this when blk_co_pdiscard switches to 64-bit */
2413         while (ret >= 0 && request->len) {
2414             int align = client->check_align ?: 1;
2415             int len = MIN(request->len, QEMU_ALIGN_DOWN(BDRV_REQUEST_MAX_BYTES,
2416                                                         align));
2417             ret = blk_co_pdiscard(exp->common.blk, request->from, len);
2418             request->len -= len;
2419             request->from += len;
2420         }
2421         if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2422             ret = blk_co_flush(exp->common.blk);
2423         }
2424         return nbd_send_generic_reply(client, request->handle, ret,
2425                                       "discard failed", errp);
2426 
2427     case NBD_CMD_BLOCK_STATUS:
2428         if (!request->len) {
2429             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2430                                           "need non-zero length", errp);
2431         }
2432         if (client->export_meta.count) {
2433             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2434             int contexts_remaining = client->export_meta.count;
2435 
2436             if (client->export_meta.base_allocation) {
2437                 ret = nbd_co_send_block_status(client, request->handle,
2438                                                blk_bs(exp->common.blk),
2439                                                request->from,
2440                                                request->len, dont_fragment,
2441                                                !--contexts_remaining,
2442                                                NBD_META_ID_BASE_ALLOCATION,
2443                                                errp);
2444                 if (ret < 0) {
2445                     return ret;
2446                 }
2447             }
2448 
2449             if (client->export_meta.allocation_depth) {
2450                 ret = nbd_co_send_block_status(client, request->handle,
2451                                                blk_bs(exp->common.blk),
2452                                                request->from, request->len,
2453                                                dont_fragment,
2454                                                !--contexts_remaining,
2455                                                NBD_META_ID_ALLOCATION_DEPTH,
2456                                                errp);
2457                 if (ret < 0) {
2458                     return ret;
2459                 }
2460             }
2461 
2462             for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2463                 if (!client->export_meta.bitmaps[i]) {
2464                     continue;
2465                 }
2466                 ret = nbd_co_send_bitmap(client, request->handle,
2467                                          client->exp->export_bitmaps[i],
2468                                          request->from, request->len,
2469                                          dont_fragment, !--contexts_remaining,
2470                                          NBD_META_ID_DIRTY_BITMAP + i, errp);
2471                 if (ret < 0) {
2472                     return ret;
2473                 }
2474             }
2475 
2476             assert(!contexts_remaining);
2477 
2478             return 0;
2479         } else {
2480             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2481                                           "CMD_BLOCK_STATUS not negotiated",
2482                                           errp);
2483         }
2484 
2485     default:
2486         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2487                               request->type);
2488         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2489                                      errp);
2490         g_free(msg);
2491         return ret;
2492     }
2493 }
2494 
2495 /* Owns a reference to the NBDClient passed as opaque.  */
2496 static coroutine_fn void nbd_trip(void *opaque)
2497 {
2498     NBDClient *client = opaque;
2499     NBDRequestData *req;
2500     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2501     int ret;
2502     Error *local_err = NULL;
2503 
2504     trace_nbd_trip();
2505     if (client->closing) {
2506         nbd_client_put(client);
2507         return;
2508     }
2509 
2510     req = nbd_request_get(client);
2511     ret = nbd_co_receive_request(req, &request, &local_err);
2512     client->recv_coroutine = NULL;
2513 
2514     if (client->closing) {
2515         /*
2516          * The client may be closed when we are blocked in
2517          * nbd_co_receive_request()
2518          */
2519         goto done;
2520     }
2521 
2522     nbd_client_receive_next_request(client);
2523     if (ret == -EIO) {
2524         goto disconnect;
2525     }
2526 
2527     if (ret < 0) {
2528         /* It wans't -EIO, so, according to nbd_co_receive_request()
2529          * semantics, we should return the error to the client. */
2530         Error *export_err = local_err;
2531 
2532         local_err = NULL;
2533         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2534                                      error_get_pretty(export_err), &local_err);
2535         error_free(export_err);
2536     } else {
2537         ret = nbd_handle_request(client, &request, req->data, &local_err);
2538     }
2539     if (ret < 0) {
2540         error_prepend(&local_err, "Failed to send reply: ");
2541         goto disconnect;
2542     }
2543 
2544     /* We must disconnect after NBD_CMD_WRITE if we did not
2545      * read the payload.
2546      */
2547     if (!req->complete) {
2548         error_setg(&local_err, "Request handling failed in intermediate state");
2549         goto disconnect;
2550     }
2551 
2552 done:
2553     nbd_request_put(req);
2554     nbd_client_put(client);
2555     return;
2556 
2557 disconnect:
2558     if (local_err) {
2559         error_reportf_err(local_err, "Disconnect client, due to: ");
2560     }
2561     nbd_request_put(req);
2562     client_close(client, true);
2563     nbd_client_put(client);
2564 }
2565 
2566 static void nbd_client_receive_next_request(NBDClient *client)
2567 {
2568     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2569         nbd_client_get(client);
2570         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2571         aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
2572     }
2573 }
2574 
2575 static coroutine_fn void nbd_co_client_start(void *opaque)
2576 {
2577     NBDClient *client = opaque;
2578     Error *local_err = NULL;
2579 
2580     qemu_co_mutex_init(&client->send_lock);
2581 
2582     if (nbd_negotiate(client, &local_err)) {
2583         if (local_err) {
2584             error_report_err(local_err);
2585         }
2586         client_close(client, false);
2587         return;
2588     }
2589 
2590     nbd_client_receive_next_request(client);
2591 }
2592 
2593 /*
2594  * Create a new client listener using the given channel @sioc.
2595  * Begin servicing it in a coroutine.  When the connection closes, call
2596  * @close_fn with an indication of whether the client completed negotiation.
2597  */
2598 void nbd_client_new(QIOChannelSocket *sioc,
2599                     QCryptoTLSCreds *tlscreds,
2600                     const char *tlsauthz,
2601                     void (*close_fn)(NBDClient *, bool))
2602 {
2603     NBDClient *client;
2604     Coroutine *co;
2605 
2606     client = g_new0(NBDClient, 1);
2607     client->refcount = 1;
2608     client->tlscreds = tlscreds;
2609     if (tlscreds) {
2610         object_ref(OBJECT(client->tlscreds));
2611     }
2612     client->tlsauthz = g_strdup(tlsauthz);
2613     client->sioc = sioc;
2614     object_ref(OBJECT(client->sioc));
2615     client->ioc = QIO_CHANNEL(sioc);
2616     object_ref(OBJECT(client->ioc));
2617     client->close_fn = close_fn;
2618 
2619     co = qemu_coroutine_create(nbd_co_client_start, client);
2620     qemu_coroutine_enter(co);
2621 }
2622