xref: /openbmc/qemu/nbd/server.c (revision fd38b162)
1 /*
2  *  Copyright (C) 2016-2018 Red Hat, Inc.
3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
4  *
5  *  Network Block Device Server Side
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; under version 2 of the License.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "qemu/queue.h"
23 #include "trace.h"
24 #include "nbd-internal.h"
25 #include "qemu/units.h"
26 
27 #define NBD_META_ID_BASE_ALLOCATION 0
28 #define NBD_META_ID_DIRTY_BITMAP 1
29 
30 /*
31  * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
32  * constant. If an increase is needed, note that the NBD protocol
33  * recommends no larger than 32 mb, so that the client won't consider
34  * the reply as a denial of service attack.
35  */
36 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
37 
38 static int system_errno_to_nbd_errno(int err)
39 {
40     switch (err) {
41     case 0:
42         return NBD_SUCCESS;
43     case EPERM:
44     case EROFS:
45         return NBD_EPERM;
46     case EIO:
47         return NBD_EIO;
48     case ENOMEM:
49         return NBD_ENOMEM;
50 #ifdef EDQUOT
51     case EDQUOT:
52 #endif
53     case EFBIG:
54     case ENOSPC:
55         return NBD_ENOSPC;
56     case EOVERFLOW:
57         return NBD_EOVERFLOW;
58     case ESHUTDOWN:
59         return NBD_ESHUTDOWN;
60     case EINVAL:
61     default:
62         return NBD_EINVAL;
63     }
64 }
65 
66 /* Definitions for opaque data types */
67 
68 typedef struct NBDRequestData NBDRequestData;
69 
70 struct NBDRequestData {
71     QSIMPLEQ_ENTRY(NBDRequestData) entry;
72     NBDClient *client;
73     uint8_t *data;
74     bool complete;
75 };
76 
77 struct NBDExport {
78     int refcount;
79     void (*close)(NBDExport *exp);
80 
81     BlockBackend *blk;
82     char *name;
83     char *description;
84     uint64_t dev_offset;
85     uint64_t size;
86     uint16_t nbdflags;
87     QTAILQ_HEAD(, NBDClient) clients;
88     QTAILQ_ENTRY(NBDExport) next;
89 
90     AioContext *ctx;
91 
92     BlockBackend *eject_notifier_blk;
93     Notifier eject_notifier;
94 
95     BdrvDirtyBitmap *export_bitmap;
96     char *export_bitmap_context;
97 };
98 
99 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
100 
101 /* NBDExportMetaContexts represents a list of contexts to be exported,
102  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
103  * NBD_OPT_LIST_META_CONTEXT. */
104 typedef struct NBDExportMetaContexts {
105     NBDExport *exp;
106     bool valid; /* means that negotiation of the option finished without
107                    errors */
108     bool base_allocation; /* export base:allocation context (block status) */
109     bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
110 } NBDExportMetaContexts;
111 
112 struct NBDClient {
113     int refcount;
114     void (*close_fn)(NBDClient *client, bool negotiated);
115 
116     NBDExport *exp;
117     QCryptoTLSCreds *tlscreds;
118     char *tlsauthz;
119     QIOChannelSocket *sioc; /* The underlying data channel */
120     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
121 
122     Coroutine *recv_coroutine;
123 
124     CoMutex send_lock;
125     Coroutine *send_coroutine;
126 
127     QTAILQ_ENTRY(NBDClient) next;
128     int nb_requests;
129     bool closing;
130 
131     uint32_t check_align; /* If non-zero, check for aligned client requests */
132 
133     bool structured_reply;
134     NBDExportMetaContexts export_meta;
135 
136     uint32_t opt; /* Current option being negotiated */
137     uint32_t optlen; /* remaining length of data in ioc for the option being
138                         negotiated now */
139 };
140 
141 static void nbd_client_receive_next_request(NBDClient *client);
142 
143 /* Basic flow for negotiation
144 
145    Server         Client
146    Negotiate
147 
148    or
149 
150    Server         Client
151    Negotiate #1
152                   Option
153    Negotiate #2
154 
155    ----
156 
157    followed by
158 
159    Server         Client
160                   Request
161    Response
162                   Request
163    Response
164                   ...
165    ...
166                   Request (type == 2)
167 
168 */
169 
170 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
171                                      uint32_t type, uint32_t length)
172 {
173     stq_be_p(&rep->magic, NBD_REP_MAGIC);
174     stl_be_p(&rep->option, option);
175     stl_be_p(&rep->type, type);
176     stl_be_p(&rep->length, length);
177 }
178 
179 /* Send a reply header, including length, but no payload.
180  * Return -errno on error, 0 on success. */
181 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
182                                       uint32_t len, Error **errp)
183 {
184     NBDOptionReply rep;
185 
186     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
187                                      type, nbd_rep_lookup(type), len);
188 
189     assert(len < NBD_MAX_BUFFER_SIZE);
190 
191     set_be_option_rep(&rep, client->opt, type, len);
192     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
193 }
194 
195 /* Send a reply header with default 0 length.
196  * Return -errno on error, 0 on success. */
197 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
198                                   Error **errp)
199 {
200     return nbd_negotiate_send_rep_len(client, type, 0, errp);
201 }
202 
203 /* Send an error reply.
204  * Return -errno on error, 0 on success. */
205 static int GCC_FMT_ATTR(4, 0)
206 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
207                             Error **errp, const char *fmt, va_list va)
208 {
209     char *msg;
210     int ret;
211     size_t len;
212 
213     msg = g_strdup_vprintf(fmt, va);
214     len = strlen(msg);
215     assert(len < 4096);
216     trace_nbd_negotiate_send_rep_err(msg);
217     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
218     if (ret < 0) {
219         goto out;
220     }
221     if (nbd_write(client->ioc, msg, len, errp) < 0) {
222         error_prepend(errp, "write failed (error message): ");
223         ret = -EIO;
224     } else {
225         ret = 0;
226     }
227 
228 out:
229     g_free(msg);
230     return ret;
231 }
232 
233 /* Send an error reply.
234  * Return -errno on error, 0 on success. */
235 static int GCC_FMT_ATTR(4, 5)
236 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
237                            Error **errp, const char *fmt, ...)
238 {
239     va_list va;
240     int ret;
241 
242     va_start(va, fmt);
243     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
244     va_end(va);
245     return ret;
246 }
247 
248 /* Drop remainder of the current option, and send a reply with the
249  * given error type and message. Return -errno on read or write
250  * failure; or 0 if connection is still live. */
251 static int GCC_FMT_ATTR(4, 0)
252 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
253               const char *fmt, va_list va)
254 {
255     int ret = nbd_drop(client->ioc, client->optlen, errp);
256 
257     client->optlen = 0;
258     if (!ret) {
259         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
260     }
261     return ret;
262 }
263 
264 static int GCC_FMT_ATTR(4, 5)
265 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
266              const char *fmt, ...)
267 {
268     int ret;
269     va_list va;
270 
271     va_start(va, fmt);
272     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
273     va_end(va);
274 
275     return ret;
276 }
277 
278 static int GCC_FMT_ATTR(3, 4)
279 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
280 {
281     int ret;
282     va_list va;
283 
284     va_start(va, fmt);
285     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
286     va_end(va);
287 
288     return ret;
289 }
290 
291 /* Read size bytes from the unparsed payload of the current option.
292  * Return -errno on I/O error, 0 if option was completely handled by
293  * sending a reply about inconsistent lengths, or 1 on success. */
294 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
295                         Error **errp)
296 {
297     if (size > client->optlen) {
298         return nbd_opt_invalid(client, errp,
299                                "Inconsistent lengths in option %s",
300                                nbd_opt_lookup(client->opt));
301     }
302     client->optlen -= size;
303     return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
304 }
305 
306 /* Drop size bytes from the unparsed payload of the current option.
307  * Return -errno on I/O error, 0 if option was completely handled by
308  * sending a reply about inconsistent lengths, or 1 on success. */
309 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
310 {
311     if (size > client->optlen) {
312         return nbd_opt_invalid(client, errp,
313                                "Inconsistent lengths in option %s",
314                                nbd_opt_lookup(client->opt));
315     }
316     client->optlen -= size;
317     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
318 }
319 
320 /* nbd_opt_read_name
321  *
322  * Read a string with the format:
323  *   uint32_t len     (<= NBD_MAX_NAME_SIZE)
324  *   len bytes string (not 0-terminated)
325  *
326  * @name should be enough to store NBD_MAX_NAME_SIZE+1.
327  * If @length is non-null, it will be set to the actual string length.
328  *
329  * Return -errno on I/O error, 0 if option was completely handled by
330  * sending a reply about inconsistent lengths, or 1 on success.
331  */
332 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
333                              Error **errp)
334 {
335     int ret;
336     uint32_t len;
337 
338     ret = nbd_opt_read(client, &len, sizeof(len), errp);
339     if (ret <= 0) {
340         return ret;
341     }
342     len = cpu_to_be32(len);
343 
344     if (len > NBD_MAX_NAME_SIZE) {
345         return nbd_opt_invalid(client, errp,
346                                "Invalid name length: %" PRIu32, len);
347     }
348 
349     ret = nbd_opt_read(client, name, len, errp);
350     if (ret <= 0) {
351         return ret;
352     }
353     name[len] = '\0';
354 
355     if (length) {
356         *length = len;
357     }
358 
359     return 1;
360 }
361 
362 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
363  * Return -errno on error, 0 on success. */
364 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
365                                        Error **errp)
366 {
367     size_t name_len, desc_len;
368     uint32_t len;
369     const char *name = exp->name ? exp->name : "";
370     const char *desc = exp->description ? exp->description : "";
371     QIOChannel *ioc = client->ioc;
372     int ret;
373 
374     trace_nbd_negotiate_send_rep_list(name, desc);
375     name_len = strlen(name);
376     desc_len = strlen(desc);
377     len = name_len + desc_len + sizeof(len);
378     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
379     if (ret < 0) {
380         return ret;
381     }
382 
383     len = cpu_to_be32(name_len);
384     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
385         error_prepend(errp, "write failed (name length): ");
386         return -EINVAL;
387     }
388 
389     if (nbd_write(ioc, name, name_len, errp) < 0) {
390         error_prepend(errp, "write failed (name buffer): ");
391         return -EINVAL;
392     }
393 
394     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
395         error_prepend(errp, "write failed (description buffer): ");
396         return -EINVAL;
397     }
398 
399     return 0;
400 }
401 
402 /* Process the NBD_OPT_LIST command, with a potential series of replies.
403  * Return -errno on error, 0 on success. */
404 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
405 {
406     NBDExport *exp;
407     assert(client->opt == NBD_OPT_LIST);
408 
409     /* For each export, send a NBD_REP_SERVER reply. */
410     QTAILQ_FOREACH(exp, &exports, next) {
411         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
412             return -EINVAL;
413         }
414     }
415     /* Finish with a NBD_REP_ACK. */
416     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
417 }
418 
419 static void nbd_check_meta_export(NBDClient *client)
420 {
421     client->export_meta.valid &= client->exp == client->export_meta.exp;
422 }
423 
424 /* Send a reply to NBD_OPT_EXPORT_NAME.
425  * Return -errno on error, 0 on success. */
426 static int nbd_negotiate_handle_export_name(NBDClient *client,
427                                             uint16_t myflags, bool no_zeroes,
428                                             Error **errp)
429 {
430     char name[NBD_MAX_NAME_SIZE + 1];
431     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
432     size_t len;
433     int ret;
434 
435     /* Client sends:
436         [20 ..  xx]   export name (length bytes)
437        Server replies:
438         [ 0 ..   7]   size
439         [ 8 ..   9]   export flags
440         [10 .. 133]   reserved     (0) [unless no_zeroes]
441      */
442     trace_nbd_negotiate_handle_export_name();
443     if (client->optlen >= sizeof(name)) {
444         error_setg(errp, "Bad length received");
445         return -EINVAL;
446     }
447     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
448         return -EIO;
449     }
450     name[client->optlen] = '\0';
451     client->optlen = 0;
452 
453     trace_nbd_negotiate_handle_export_name_request(name);
454 
455     client->exp = nbd_export_find(name);
456     if (!client->exp) {
457         error_setg(errp, "export not found");
458         return -EINVAL;
459     }
460 
461     trace_nbd_negotiate_new_style_size_flags(client->exp->size,
462                                              client->exp->nbdflags | myflags);
463     stq_be_p(buf, client->exp->size);
464     stw_be_p(buf + 8, client->exp->nbdflags | myflags);
465     len = no_zeroes ? 10 : sizeof(buf);
466     ret = nbd_write(client->ioc, buf, len, errp);
467     if (ret < 0) {
468         error_prepend(errp, "write failed: ");
469         return ret;
470     }
471 
472     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
473     nbd_export_get(client->exp);
474     nbd_check_meta_export(client);
475 
476     return 0;
477 }
478 
479 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
480  * The buffer does NOT include the info type prefix.
481  * Return -errno on error, 0 if ready to send more. */
482 static int nbd_negotiate_send_info(NBDClient *client,
483                                    uint16_t info, uint32_t length, void *buf,
484                                    Error **errp)
485 {
486     int rc;
487 
488     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
489     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
490                                     sizeof(info) + length, errp);
491     if (rc < 0) {
492         return rc;
493     }
494     info = cpu_to_be16(info);
495     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
496         return -EIO;
497     }
498     if (nbd_write(client->ioc, buf, length, errp) < 0) {
499         return -EIO;
500     }
501     return 0;
502 }
503 
504 /* nbd_reject_length: Handle any unexpected payload.
505  * @fatal requests that we quit talking to the client, even if we are able
506  * to successfully send an error reply.
507  * Return:
508  * -errno  transmission error occurred or @fatal was requested, errp is set
509  * 0       error message successfully sent to client, errp is not set
510  */
511 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
512 {
513     int ret;
514 
515     assert(client->optlen);
516     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
517                           nbd_opt_lookup(client->opt));
518     if (fatal && !ret) {
519         error_setg(errp, "option '%s' has unexpected length",
520                    nbd_opt_lookup(client->opt));
521         return -EINVAL;
522     }
523     return ret;
524 }
525 
526 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
527  * Return -errno on error, 0 if ready for next option, and 1 to move
528  * into transmission phase.  */
529 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
530                                      Error **errp)
531 {
532     int rc;
533     char name[NBD_MAX_NAME_SIZE + 1];
534     NBDExport *exp;
535     uint16_t requests;
536     uint16_t request;
537     uint32_t namelen;
538     bool sendname = false;
539     bool blocksize = false;
540     uint32_t sizes[3];
541     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
542     uint32_t check_align = 0;
543 
544     /* Client sends:
545         4 bytes: L, name length (can be 0)
546         L bytes: export name
547         2 bytes: N, number of requests (can be 0)
548         N * 2 bytes: N requests
549     */
550     rc = nbd_opt_read_name(client, name, &namelen, errp);
551     if (rc <= 0) {
552         return rc;
553     }
554     trace_nbd_negotiate_handle_export_name_request(name);
555 
556     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
557     if (rc <= 0) {
558         return rc;
559     }
560     requests = be16_to_cpu(requests);
561     trace_nbd_negotiate_handle_info_requests(requests);
562     while (requests--) {
563         rc = nbd_opt_read(client, &request, sizeof(request), errp);
564         if (rc <= 0) {
565             return rc;
566         }
567         request = be16_to_cpu(request);
568         trace_nbd_negotiate_handle_info_request(request,
569                                                 nbd_info_lookup(request));
570         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
571          * everything else is either a request we don't know or
572          * something we send regardless of request */
573         switch (request) {
574         case NBD_INFO_NAME:
575             sendname = true;
576             break;
577         case NBD_INFO_BLOCK_SIZE:
578             blocksize = true;
579             break;
580         }
581     }
582     if (client->optlen) {
583         return nbd_reject_length(client, false, errp);
584     }
585 
586     exp = nbd_export_find(name);
587     if (!exp) {
588         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
589                                           errp, "export '%s' not present",
590                                           name);
591     }
592 
593     /* Don't bother sending NBD_INFO_NAME unless client requested it */
594     if (sendname) {
595         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
596                                      errp);
597         if (rc < 0) {
598             return rc;
599         }
600     }
601 
602     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
603      * client request */
604     if (exp->description) {
605         size_t len = strlen(exp->description);
606 
607         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
608                                      len, exp->description, errp);
609         if (rc < 0) {
610             return rc;
611         }
612     }
613 
614     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
615      * according to whether the client requested it, and according to
616      * whether this is OPT_INFO or OPT_GO. */
617     /* minimum - 1 for back-compat, or actual if client will obey it. */
618     if (client->opt == NBD_OPT_INFO || blocksize) {
619         check_align = sizes[0] = blk_get_request_alignment(exp->blk);
620     } else {
621         sizes[0] = 1;
622     }
623     assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
624     /* preferred - Hard-code to 4096 for now.
625      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
626     sizes[1] = MAX(4096, sizes[0]);
627     /* maximum - At most 32M, but smaller as appropriate. */
628     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
629     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
630     sizes[0] = cpu_to_be32(sizes[0]);
631     sizes[1] = cpu_to_be32(sizes[1]);
632     sizes[2] = cpu_to_be32(sizes[2]);
633     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
634                                  sizeof(sizes), sizes, errp);
635     if (rc < 0) {
636         return rc;
637     }
638 
639     /* Send NBD_INFO_EXPORT always */
640     trace_nbd_negotiate_new_style_size_flags(exp->size,
641                                              exp->nbdflags | myflags);
642     stq_be_p(buf, exp->size);
643     stw_be_p(buf + 8, exp->nbdflags | myflags);
644     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
645                                  sizeof(buf), buf, errp);
646     if (rc < 0) {
647         return rc;
648     }
649 
650     /*
651      * If the client is just asking for NBD_OPT_INFO, but forgot to
652      * request block sizes in a situation that would impact
653      * performance, then return an error. But for NBD_OPT_GO, we
654      * tolerate all clients, regardless of alignments.
655      */
656     if (client->opt == NBD_OPT_INFO && !blocksize &&
657         blk_get_request_alignment(exp->blk) > 1) {
658         return nbd_negotiate_send_rep_err(client,
659                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
660                                           errp,
661                                           "request NBD_INFO_BLOCK_SIZE to "
662                                           "use this export");
663     }
664 
665     /* Final reply */
666     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
667     if (rc < 0) {
668         return rc;
669     }
670 
671     if (client->opt == NBD_OPT_GO) {
672         client->exp = exp;
673         client->check_align = check_align;
674         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
675         nbd_export_get(client->exp);
676         nbd_check_meta_export(client);
677         rc = 1;
678     }
679     return rc;
680 }
681 
682 
683 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
684  * new channel for all further (now-encrypted) communication. */
685 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
686                                                  Error **errp)
687 {
688     QIOChannel *ioc;
689     QIOChannelTLS *tioc;
690     struct NBDTLSHandshakeData data = { 0 };
691 
692     assert(client->opt == NBD_OPT_STARTTLS);
693 
694     trace_nbd_negotiate_handle_starttls();
695     ioc = client->ioc;
696 
697     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
698         return NULL;
699     }
700 
701     tioc = qio_channel_tls_new_server(ioc,
702                                       client->tlscreds,
703                                       client->tlsauthz,
704                                       errp);
705     if (!tioc) {
706         return NULL;
707     }
708 
709     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
710     trace_nbd_negotiate_handle_starttls_handshake();
711     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
712     qio_channel_tls_handshake(tioc,
713                               nbd_tls_handshake,
714                               &data,
715                               NULL,
716                               NULL);
717 
718     if (!data.complete) {
719         g_main_loop_run(data.loop);
720     }
721     g_main_loop_unref(data.loop);
722     if (data.error) {
723         object_unref(OBJECT(tioc));
724         error_propagate(errp, data.error);
725         return NULL;
726     }
727 
728     return QIO_CHANNEL(tioc);
729 }
730 
731 /* nbd_negotiate_send_meta_context
732  *
733  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
734  *
735  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
736  */
737 static int nbd_negotiate_send_meta_context(NBDClient *client,
738                                            const char *context,
739                                            uint32_t context_id,
740                                            Error **errp)
741 {
742     NBDOptionReplyMetaContext opt;
743     struct iovec iov[] = {
744         {.iov_base = &opt, .iov_len = sizeof(opt)},
745         {.iov_base = (void *)context, .iov_len = strlen(context)}
746     };
747 
748     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
749         context_id = 0;
750     }
751 
752     trace_nbd_negotiate_meta_query_reply(context, context_id);
753     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
754                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
755     stl_be_p(&opt.context_id, context_id);
756 
757     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
758 }
759 
760 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
761  * @match is never set to false.
762  *
763  * Return -errno on I/O error, 0 if option was completely handled by
764  * sending a reply about inconsistent lengths, or 1 on success.
765  *
766  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
767  * It only means that there are no errors.
768  */
769 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
770                             Error **errp)
771 {
772     int ret;
773     char *query;
774     size_t len = strlen(pattern);
775 
776     assert(len);
777 
778     query = g_malloc(len);
779     ret = nbd_opt_read(client, query, len, errp);
780     if (ret <= 0) {
781         g_free(query);
782         return ret;
783     }
784 
785     if (strncmp(query, pattern, len) == 0) {
786         trace_nbd_negotiate_meta_query_parse(pattern);
787         *match = true;
788     } else {
789         trace_nbd_negotiate_meta_query_skip("pattern not matched");
790     }
791     g_free(query);
792 
793     return 1;
794 }
795 
796 /*
797  * Read @len bytes, and set @match to true if they match @pattern, or if @len
798  * is 0 and the client is performing _LIST_. @match is never set to false.
799  *
800  * Return -errno on I/O error, 0 if option was completely handled by
801  * sending a reply about inconsistent lengths, or 1 on success.
802  *
803  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
804  * It only means that there are no errors.
805  */
806 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
807                                      uint32_t len, bool *match, Error **errp)
808 {
809     if (len == 0) {
810         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
811             *match = true;
812         }
813         trace_nbd_negotiate_meta_query_parse("empty");
814         return 1;
815     }
816 
817     if (len != strlen(pattern)) {
818         trace_nbd_negotiate_meta_query_skip("different lengths");
819         return nbd_opt_skip(client, len, errp);
820     }
821 
822     return nbd_meta_pattern(client, pattern, match, errp);
823 }
824 
825 /* nbd_meta_base_query
826  *
827  * Handle queries to 'base' namespace. For now, only the base:allocation
828  * context is available.  'len' is the amount of text remaining to be read from
829  * the current name, after the 'base:' portion has been stripped.
830  *
831  * Return -errno on I/O error, 0 if option was completely handled by
832  * sending a reply about inconsistent lengths, or 1 on success.
833  */
834 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
835                                uint32_t len, Error **errp)
836 {
837     return nbd_meta_empty_or_pattern(client, "allocation", len,
838                                      &meta->base_allocation, errp);
839 }
840 
841 /* nbd_meta_bitmap_query
842  *
843  * Handle query to 'qemu:' namespace.
844  * @len is the amount of text remaining to be read from the current name, after
845  * the 'qemu:' portion has been stripped.
846  *
847  * Return -errno on I/O error, 0 if option was completely handled by
848  * sending a reply about inconsistent lengths, or 1 on success. */
849 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
850                                uint32_t len, Error **errp)
851 {
852     bool dirty_bitmap = false;
853     size_t dirty_bitmap_len = strlen("dirty-bitmap:");
854     int ret;
855 
856     if (!meta->exp->export_bitmap) {
857         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
858         return nbd_opt_skip(client, len, errp);
859     }
860 
861     if (len == 0) {
862         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
863             meta->bitmap = true;
864         }
865         trace_nbd_negotiate_meta_query_parse("empty");
866         return 1;
867     }
868 
869     if (len < dirty_bitmap_len) {
870         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
871         return nbd_opt_skip(client, len, errp);
872     }
873 
874     len -= dirty_bitmap_len;
875     ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
876     if (ret <= 0) {
877         return ret;
878     }
879     if (!dirty_bitmap) {
880         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
881         return nbd_opt_skip(client, len, errp);
882     }
883 
884     trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
885 
886     return nbd_meta_empty_or_pattern(
887             client, meta->exp->export_bitmap_context +
888             strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
889 }
890 
891 /* nbd_negotiate_meta_query
892  *
893  * Parse namespace name and call corresponding function to parse body of the
894  * query.
895  *
896  * The only supported namespace now is 'base'.
897  *
898  * The function aims not wasting time and memory to read long unknown namespace
899  * names.
900  *
901  * Return -errno on I/O error, 0 if option was completely handled by
902  * sending a reply about inconsistent lengths, or 1 on success. */
903 static int nbd_negotiate_meta_query(NBDClient *client,
904                                     NBDExportMetaContexts *meta, Error **errp)
905 {
906     /*
907      * Both 'qemu' and 'base' namespaces have length = 5 including a
908      * colon. If another length namespace is later introduced, this
909      * should certainly be refactored.
910      */
911     int ret;
912     size_t ns_len = 5;
913     char ns[5];
914     uint32_t len;
915 
916     ret = nbd_opt_read(client, &len, sizeof(len), errp);
917     if (ret <= 0) {
918         return ret;
919     }
920     len = cpu_to_be32(len);
921 
922     if (len < ns_len) {
923         trace_nbd_negotiate_meta_query_skip("length too short");
924         return nbd_opt_skip(client, len, errp);
925     }
926 
927     len -= ns_len;
928     ret = nbd_opt_read(client, ns, ns_len, errp);
929     if (ret <= 0) {
930         return ret;
931     }
932 
933     if (!strncmp(ns, "base:", ns_len)) {
934         trace_nbd_negotiate_meta_query_parse("base:");
935         return nbd_meta_base_query(client, meta, len, errp);
936     } else if (!strncmp(ns, "qemu:", ns_len)) {
937         trace_nbd_negotiate_meta_query_parse("qemu:");
938         return nbd_meta_qemu_query(client, meta, len, errp);
939     }
940 
941     trace_nbd_negotiate_meta_query_skip("unknown namespace");
942     return nbd_opt_skip(client, len, errp);
943 }
944 
945 /* nbd_negotiate_meta_queries
946  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
947  *
948  * Return -errno on I/O error, or 0 if option was completely handled. */
949 static int nbd_negotiate_meta_queries(NBDClient *client,
950                                       NBDExportMetaContexts *meta, Error **errp)
951 {
952     int ret;
953     char export_name[NBD_MAX_NAME_SIZE + 1];
954     NBDExportMetaContexts local_meta;
955     uint32_t nb_queries;
956     int i;
957 
958     if (!client->structured_reply) {
959         return nbd_opt_invalid(client, errp,
960                                "request option '%s' when structured reply "
961                                "is not negotiated",
962                                nbd_opt_lookup(client->opt));
963     }
964 
965     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
966         /* Only change the caller's meta on SET. */
967         meta = &local_meta;
968     }
969 
970     memset(meta, 0, sizeof(*meta));
971 
972     ret = nbd_opt_read_name(client, export_name, NULL, errp);
973     if (ret <= 0) {
974         return ret;
975     }
976 
977     meta->exp = nbd_export_find(export_name);
978     if (meta->exp == NULL) {
979         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
980                             "export '%s' not present", export_name);
981     }
982 
983     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
984     if (ret <= 0) {
985         return ret;
986     }
987     nb_queries = cpu_to_be32(nb_queries);
988     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
989                                      export_name, nb_queries);
990 
991     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
992         /* enable all known contexts */
993         meta->base_allocation = true;
994         meta->bitmap = !!meta->exp->export_bitmap;
995     } else {
996         for (i = 0; i < nb_queries; ++i) {
997             ret = nbd_negotiate_meta_query(client, meta, errp);
998             if (ret <= 0) {
999                 return ret;
1000             }
1001         }
1002     }
1003 
1004     if (meta->base_allocation) {
1005         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1006                                               NBD_META_ID_BASE_ALLOCATION,
1007                                               errp);
1008         if (ret < 0) {
1009             return ret;
1010         }
1011     }
1012 
1013     if (meta->bitmap) {
1014         ret = nbd_negotiate_send_meta_context(client,
1015                                               meta->exp->export_bitmap_context,
1016                                               NBD_META_ID_DIRTY_BITMAP,
1017                                               errp);
1018         if (ret < 0) {
1019             return ret;
1020         }
1021     }
1022 
1023     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1024     if (ret == 0) {
1025         meta->valid = true;
1026     }
1027 
1028     return ret;
1029 }
1030 
1031 /* nbd_negotiate_options
1032  * Process all NBD_OPT_* client option commands, during fixed newstyle
1033  * negotiation.
1034  * Return:
1035  * -errno  on error, errp is set
1036  * 0       on successful negotiation, errp is not set
1037  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1038  *         errp is not set
1039  */
1040 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1041                                  Error **errp)
1042 {
1043     uint32_t flags;
1044     bool fixedNewstyle = false;
1045     bool no_zeroes = false;
1046 
1047     /* Client sends:
1048         [ 0 ..   3]   client flags
1049 
1050        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1051         [ 0 ..   7]   NBD_OPTS_MAGIC
1052         [ 8 ..  11]   NBD option
1053         [12 ..  15]   Data length
1054         ...           Rest of request
1055 
1056         [ 0 ..   7]   NBD_OPTS_MAGIC
1057         [ 8 ..  11]   Second NBD option
1058         [12 ..  15]   Data length
1059         ...           Rest of request
1060     */
1061 
1062     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1063         return -EIO;
1064     }
1065     trace_nbd_negotiate_options_flags(flags);
1066     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1067         fixedNewstyle = true;
1068         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1069     }
1070     if (flags & NBD_FLAG_C_NO_ZEROES) {
1071         no_zeroes = true;
1072         flags &= ~NBD_FLAG_C_NO_ZEROES;
1073     }
1074     if (flags != 0) {
1075         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1076         return -EINVAL;
1077     }
1078 
1079     while (1) {
1080         int ret;
1081         uint32_t option, length;
1082         uint64_t magic;
1083 
1084         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1085             return -EINVAL;
1086         }
1087         trace_nbd_negotiate_options_check_magic(magic);
1088         if (magic != NBD_OPTS_MAGIC) {
1089             error_setg(errp, "Bad magic received");
1090             return -EINVAL;
1091         }
1092 
1093         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1094             return -EINVAL;
1095         }
1096         client->opt = option;
1097 
1098         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1099             return -EINVAL;
1100         }
1101         assert(!client->optlen);
1102         client->optlen = length;
1103 
1104         if (length > NBD_MAX_BUFFER_SIZE) {
1105             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1106                        length, NBD_MAX_BUFFER_SIZE);
1107             return -EINVAL;
1108         }
1109 
1110         trace_nbd_negotiate_options_check_option(option,
1111                                                  nbd_opt_lookup(option));
1112         if (client->tlscreds &&
1113             client->ioc == (QIOChannel *)client->sioc) {
1114             QIOChannel *tioc;
1115             if (!fixedNewstyle) {
1116                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1117                 return -EINVAL;
1118             }
1119             switch (option) {
1120             case NBD_OPT_STARTTLS:
1121                 if (length) {
1122                     /* Unconditionally drop the connection if the client
1123                      * can't start a TLS negotiation correctly */
1124                     return nbd_reject_length(client, true, errp);
1125                 }
1126                 tioc = nbd_negotiate_handle_starttls(client, errp);
1127                 if (!tioc) {
1128                     return -EIO;
1129                 }
1130                 ret = 0;
1131                 object_unref(OBJECT(client->ioc));
1132                 client->ioc = QIO_CHANNEL(tioc);
1133                 break;
1134 
1135             case NBD_OPT_EXPORT_NAME:
1136                 /* No way to return an error to client, so drop connection */
1137                 error_setg(errp, "Option 0x%x not permitted before TLS",
1138                            option);
1139                 return -EINVAL;
1140 
1141             default:
1142                 /* Let the client keep trying, unless they asked to
1143                  * quit. Always try to give an error back to the
1144                  * client; but when replying to OPT_ABORT, be aware
1145                  * that the client may hang up before receiving the
1146                  * error, in which case we are fine ignoring the
1147                  * resulting EPIPE. */
1148                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1149                                    option == NBD_OPT_ABORT ? NULL : errp,
1150                                    "Option 0x%" PRIx32
1151                                    " not permitted before TLS", option);
1152                 if (option == NBD_OPT_ABORT) {
1153                     return 1;
1154                 }
1155                 break;
1156             }
1157         } else if (fixedNewstyle) {
1158             switch (option) {
1159             case NBD_OPT_LIST:
1160                 if (length) {
1161                     ret = nbd_reject_length(client, false, errp);
1162                 } else {
1163                     ret = nbd_negotiate_handle_list(client, errp);
1164                 }
1165                 break;
1166 
1167             case NBD_OPT_ABORT:
1168                 /* NBD spec says we must try to reply before
1169                  * disconnecting, but that we must also tolerate
1170                  * guests that don't wait for our reply. */
1171                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1172                 return 1;
1173 
1174             case NBD_OPT_EXPORT_NAME:
1175                 return nbd_negotiate_handle_export_name(client,
1176                                                         myflags, no_zeroes,
1177                                                         errp);
1178 
1179             case NBD_OPT_INFO:
1180             case NBD_OPT_GO:
1181                 ret = nbd_negotiate_handle_info(client, myflags, errp);
1182                 if (ret == 1) {
1183                     assert(option == NBD_OPT_GO);
1184                     return 0;
1185                 }
1186                 break;
1187 
1188             case NBD_OPT_STARTTLS:
1189                 if (length) {
1190                     ret = nbd_reject_length(client, false, errp);
1191                 } else if (client->tlscreds) {
1192                     ret = nbd_negotiate_send_rep_err(client,
1193                                                      NBD_REP_ERR_INVALID, errp,
1194                                                      "TLS already enabled");
1195                 } else {
1196                     ret = nbd_negotiate_send_rep_err(client,
1197                                                      NBD_REP_ERR_POLICY, errp,
1198                                                      "TLS not configured");
1199                 }
1200                 break;
1201 
1202             case NBD_OPT_STRUCTURED_REPLY:
1203                 if (length) {
1204                     ret = nbd_reject_length(client, false, errp);
1205                 } else if (client->structured_reply) {
1206                     ret = nbd_negotiate_send_rep_err(
1207                         client, NBD_REP_ERR_INVALID, errp,
1208                         "structured reply already negotiated");
1209                 } else {
1210                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1211                     client->structured_reply = true;
1212                     myflags |= NBD_FLAG_SEND_DF;
1213                 }
1214                 break;
1215 
1216             case NBD_OPT_LIST_META_CONTEXT:
1217             case NBD_OPT_SET_META_CONTEXT:
1218                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1219                                                  errp);
1220                 break;
1221 
1222             default:
1223                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1224                                    "Unsupported option %" PRIu32 " (%s)",
1225                                    option, nbd_opt_lookup(option));
1226                 break;
1227             }
1228         } else {
1229             /*
1230              * If broken new-style we should drop the connection
1231              * for anything except NBD_OPT_EXPORT_NAME
1232              */
1233             switch (option) {
1234             case NBD_OPT_EXPORT_NAME:
1235                 return nbd_negotiate_handle_export_name(client,
1236                                                         myflags, no_zeroes,
1237                                                         errp);
1238 
1239             default:
1240                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1241                            option, nbd_opt_lookup(option));
1242                 return -EINVAL;
1243             }
1244         }
1245         if (ret < 0) {
1246             return ret;
1247         }
1248     }
1249 }
1250 
1251 /* nbd_negotiate
1252  * Return:
1253  * -errno  on error, errp is set
1254  * 0       on successful negotiation, errp is not set
1255  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1256  *         errp is not set
1257  */
1258 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1259 {
1260     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1261     int ret;
1262     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1263                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1264                               NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1265 
1266     /* Old style negotiation header, no room for options
1267         [ 0 ..   7]   passwd       ("NBDMAGIC")
1268         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1269         [16 ..  23]   size
1270         [24 ..  27]   export flags (zero-extended)
1271         [28 .. 151]   reserved     (0)
1272 
1273        New style negotiation header, client can send options
1274         [ 0 ..   7]   passwd       ("NBDMAGIC")
1275         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1276         [16 ..  17]   server flags (0)
1277         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1278      */
1279 
1280     qio_channel_set_blocking(client->ioc, false, NULL);
1281 
1282     trace_nbd_negotiate_begin();
1283     memcpy(buf, "NBDMAGIC", 8);
1284 
1285     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1286     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1287 
1288     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1289         error_prepend(errp, "write failed: ");
1290         return -EINVAL;
1291     }
1292     ret = nbd_negotiate_options(client, myflags, errp);
1293     if (ret != 0) {
1294         if (ret < 0) {
1295             error_prepend(errp, "option negotiation failed: ");
1296         }
1297         return ret;
1298     }
1299 
1300     assert(!client->optlen);
1301     trace_nbd_negotiate_success();
1302 
1303     return 0;
1304 }
1305 
1306 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1307                                Error **errp)
1308 {
1309     uint8_t buf[NBD_REQUEST_SIZE];
1310     uint32_t magic;
1311     int ret;
1312 
1313     ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1314     if (ret < 0) {
1315         return ret;
1316     }
1317 
1318     /* Request
1319        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1320        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1321        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1322        [ 8 .. 15]   handle
1323        [16 .. 23]   from
1324        [24 .. 27]   len
1325      */
1326 
1327     magic = ldl_be_p(buf);
1328     request->flags  = lduw_be_p(buf + 4);
1329     request->type   = lduw_be_p(buf + 6);
1330     request->handle = ldq_be_p(buf + 8);
1331     request->from   = ldq_be_p(buf + 16);
1332     request->len    = ldl_be_p(buf + 24);
1333 
1334     trace_nbd_receive_request(magic, request->flags, request->type,
1335                               request->from, request->len);
1336 
1337     if (magic != NBD_REQUEST_MAGIC) {
1338         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1339         return -EINVAL;
1340     }
1341     return 0;
1342 }
1343 
1344 #define MAX_NBD_REQUESTS 16
1345 
1346 void nbd_client_get(NBDClient *client)
1347 {
1348     client->refcount++;
1349 }
1350 
1351 void nbd_client_put(NBDClient *client)
1352 {
1353     if (--client->refcount == 0) {
1354         /* The last reference should be dropped by client->close,
1355          * which is called by client_close.
1356          */
1357         assert(client->closing);
1358 
1359         qio_channel_detach_aio_context(client->ioc);
1360         object_unref(OBJECT(client->sioc));
1361         object_unref(OBJECT(client->ioc));
1362         if (client->tlscreds) {
1363             object_unref(OBJECT(client->tlscreds));
1364         }
1365         g_free(client->tlsauthz);
1366         if (client->exp) {
1367             QTAILQ_REMOVE(&client->exp->clients, client, next);
1368             nbd_export_put(client->exp);
1369         }
1370         g_free(client);
1371     }
1372 }
1373 
1374 static void client_close(NBDClient *client, bool negotiated)
1375 {
1376     if (client->closing) {
1377         return;
1378     }
1379 
1380     client->closing = true;
1381 
1382     /* Force requests to finish.  They will drop their own references,
1383      * then we'll close the socket and free the NBDClient.
1384      */
1385     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1386                          NULL);
1387 
1388     /* Also tell the client, so that they release their reference.  */
1389     if (client->close_fn) {
1390         client->close_fn(client, negotiated);
1391     }
1392 }
1393 
1394 static NBDRequestData *nbd_request_get(NBDClient *client)
1395 {
1396     NBDRequestData *req;
1397 
1398     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1399     client->nb_requests++;
1400 
1401     req = g_new0(NBDRequestData, 1);
1402     nbd_client_get(client);
1403     req->client = client;
1404     return req;
1405 }
1406 
1407 static void nbd_request_put(NBDRequestData *req)
1408 {
1409     NBDClient *client = req->client;
1410 
1411     if (req->data) {
1412         qemu_vfree(req->data);
1413     }
1414     g_free(req);
1415 
1416     client->nb_requests--;
1417     nbd_client_receive_next_request(client);
1418 
1419     nbd_client_put(client);
1420 }
1421 
1422 static void blk_aio_attached(AioContext *ctx, void *opaque)
1423 {
1424     NBDExport *exp = opaque;
1425     NBDClient *client;
1426 
1427     trace_nbd_blk_aio_attached(exp->name, ctx);
1428 
1429     exp->ctx = ctx;
1430 
1431     QTAILQ_FOREACH(client, &exp->clients, next) {
1432         qio_channel_attach_aio_context(client->ioc, ctx);
1433         if (client->recv_coroutine) {
1434             aio_co_schedule(ctx, client->recv_coroutine);
1435         }
1436         if (client->send_coroutine) {
1437             aio_co_schedule(ctx, client->send_coroutine);
1438         }
1439     }
1440 }
1441 
1442 static void blk_aio_detach(void *opaque)
1443 {
1444     NBDExport *exp = opaque;
1445     NBDClient *client;
1446 
1447     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1448 
1449     QTAILQ_FOREACH(client, &exp->clients, next) {
1450         qio_channel_detach_aio_context(client->ioc);
1451     }
1452 
1453     exp->ctx = NULL;
1454 }
1455 
1456 static void nbd_eject_notifier(Notifier *n, void *data)
1457 {
1458     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1459     nbd_export_close(exp);
1460 }
1461 
1462 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1463                           uint64_t size, const char *name, const char *desc,
1464                           const char *bitmap, uint16_t nbdflags,
1465                           void (*close)(NBDExport *), bool writethrough,
1466                           BlockBackend *on_eject_blk, Error **errp)
1467 {
1468     AioContext *ctx;
1469     BlockBackend *blk;
1470     NBDExport *exp = g_new0(NBDExport, 1);
1471     uint64_t perm;
1472     int ret;
1473 
1474     /*
1475      * NBD exports are used for non-shared storage migration.  Make sure
1476      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1477      * access since the export could be available before migration handover.
1478      */
1479     assert(name);
1480     ctx = bdrv_get_aio_context(bs);
1481     aio_context_acquire(ctx);
1482     bdrv_invalidate_cache(bs, NULL);
1483     aio_context_release(ctx);
1484 
1485     /* Don't allow resize while the NBD server is running, otherwise we don't
1486      * care what happens with the node. */
1487     perm = BLK_PERM_CONSISTENT_READ;
1488     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1489         perm |= BLK_PERM_WRITE;
1490     }
1491     blk = blk_new(bdrv_get_aio_context(bs), perm,
1492                   BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1493                   BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1494     ret = blk_insert_bs(blk, bs, errp);
1495     if (ret < 0) {
1496         goto fail;
1497     }
1498     blk_set_enable_write_cache(blk, !writethrough);
1499     blk_set_allow_aio_context_change(blk, true);
1500 
1501     exp->refcount = 1;
1502     QTAILQ_INIT(&exp->clients);
1503     exp->blk = blk;
1504     assert(dev_offset <= INT64_MAX);
1505     exp->dev_offset = dev_offset;
1506     exp->name = g_strdup(name);
1507     exp->description = g_strdup(desc);
1508     exp->nbdflags = nbdflags;
1509     assert(size <= INT64_MAX - dev_offset);
1510     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1511 
1512     if (bitmap) {
1513         BdrvDirtyBitmap *bm = NULL;
1514 
1515         while (true) {
1516             bm = bdrv_find_dirty_bitmap(bs, bitmap);
1517             if (bm != NULL || bs->backing == NULL) {
1518                 break;
1519             }
1520 
1521             bs = bs->backing->bs;
1522         }
1523 
1524         if (bm == NULL) {
1525             error_setg(errp, "Bitmap '%s' is not found", bitmap);
1526             goto fail;
1527         }
1528 
1529         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1530             goto fail;
1531         }
1532 
1533         if ((nbdflags & NBD_FLAG_READ_ONLY) && bdrv_is_writable(bs) &&
1534             bdrv_dirty_bitmap_enabled(bm)) {
1535             error_setg(errp,
1536                        "Enabled bitmap '%s' incompatible with readonly export",
1537                        bitmap);
1538             goto fail;
1539         }
1540 
1541         bdrv_dirty_bitmap_set_busy(bm, true);
1542         exp->export_bitmap = bm;
1543         exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1544                                                      bitmap);
1545     }
1546 
1547     exp->close = close;
1548     exp->ctx = blk_get_aio_context(blk);
1549     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1550 
1551     if (on_eject_blk) {
1552         blk_ref(on_eject_blk);
1553         exp->eject_notifier_blk = on_eject_blk;
1554         exp->eject_notifier.notify = nbd_eject_notifier;
1555         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1556     }
1557     QTAILQ_INSERT_TAIL(&exports, exp, next);
1558     nbd_export_get(exp);
1559     return exp;
1560 
1561 fail:
1562     blk_unref(blk);
1563     g_free(exp->name);
1564     g_free(exp->description);
1565     g_free(exp);
1566     return NULL;
1567 }
1568 
1569 NBDExport *nbd_export_find(const char *name)
1570 {
1571     NBDExport *exp;
1572     QTAILQ_FOREACH(exp, &exports, next) {
1573         if (strcmp(name, exp->name) == 0) {
1574             return exp;
1575         }
1576     }
1577 
1578     return NULL;
1579 }
1580 
1581 void nbd_export_close(NBDExport *exp)
1582 {
1583     NBDClient *client, *next;
1584 
1585     nbd_export_get(exp);
1586     /*
1587      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1588      * close mode that stops advertising the export to new clients but
1589      * still permits existing clients to run to completion? Because of
1590      * that possibility, nbd_export_close() can be called more than
1591      * once on an export.
1592      */
1593     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1594         client_close(client, true);
1595     }
1596     if (exp->name) {
1597         nbd_export_put(exp);
1598         g_free(exp->name);
1599         exp->name = NULL;
1600         QTAILQ_REMOVE(&exports, exp, next);
1601     }
1602     g_free(exp->description);
1603     exp->description = NULL;
1604     nbd_export_put(exp);
1605 }
1606 
1607 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1608 {
1609     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1610         nbd_export_close(exp);
1611         return;
1612     }
1613 
1614     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1615 
1616     error_setg(errp, "export '%s' still in use", exp->name);
1617     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1618 }
1619 
1620 void nbd_export_get(NBDExport *exp)
1621 {
1622     assert(exp->refcount > 0);
1623     exp->refcount++;
1624 }
1625 
1626 void nbd_export_put(NBDExport *exp)
1627 {
1628     assert(exp->refcount > 0);
1629     if (exp->refcount == 1) {
1630         nbd_export_close(exp);
1631     }
1632 
1633     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1634      * if someone calls nbd_export_put() on named export not through
1635      * nbd_export_set_name() when refcount is 1. So, let's assert that
1636      * it is > 0.
1637      */
1638     assert(exp->refcount > 0);
1639     if (--exp->refcount == 0) {
1640         assert(exp->name == NULL);
1641         assert(exp->description == NULL);
1642 
1643         if (exp->close) {
1644             exp->close(exp);
1645         }
1646 
1647         if (exp->blk) {
1648             if (exp->eject_notifier_blk) {
1649                 notifier_remove(&exp->eject_notifier);
1650                 blk_unref(exp->eject_notifier_blk);
1651             }
1652             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1653                                             blk_aio_detach, exp);
1654             blk_unref(exp->blk);
1655             exp->blk = NULL;
1656         }
1657 
1658         if (exp->export_bitmap) {
1659             bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1660             g_free(exp->export_bitmap_context);
1661         }
1662 
1663         g_free(exp);
1664     }
1665 }
1666 
1667 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1668 {
1669     return exp->blk;
1670 }
1671 
1672 void nbd_export_close_all(void)
1673 {
1674     NBDExport *exp, *next;
1675 
1676     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1677         nbd_export_close(exp);
1678     }
1679 }
1680 
1681 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1682                                         unsigned niov, Error **errp)
1683 {
1684     int ret;
1685 
1686     g_assert(qemu_in_coroutine());
1687     qemu_co_mutex_lock(&client->send_lock);
1688     client->send_coroutine = qemu_coroutine_self();
1689 
1690     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1691 
1692     client->send_coroutine = NULL;
1693     qemu_co_mutex_unlock(&client->send_lock);
1694 
1695     return ret;
1696 }
1697 
1698 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1699                                        uint64_t handle)
1700 {
1701     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1702     stl_be_p(&reply->error, error);
1703     stq_be_p(&reply->handle, handle);
1704 }
1705 
1706 static int nbd_co_send_simple_reply(NBDClient *client,
1707                                     uint64_t handle,
1708                                     uint32_t error,
1709                                     void *data,
1710                                     size_t len,
1711                                     Error **errp)
1712 {
1713     NBDSimpleReply reply;
1714     int nbd_err = system_errno_to_nbd_errno(error);
1715     struct iovec iov[] = {
1716         {.iov_base = &reply, .iov_len = sizeof(reply)},
1717         {.iov_base = data, .iov_len = len}
1718     };
1719 
1720     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1721                                    len);
1722     set_be_simple_reply(&reply, nbd_err, handle);
1723 
1724     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1725 }
1726 
1727 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1728                                 uint16_t type, uint64_t handle, uint32_t length)
1729 {
1730     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1731     stw_be_p(&chunk->flags, flags);
1732     stw_be_p(&chunk->type, type);
1733     stq_be_p(&chunk->handle, handle);
1734     stl_be_p(&chunk->length, length);
1735 }
1736 
1737 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1738                                                     uint64_t handle,
1739                                                     Error **errp)
1740 {
1741     NBDStructuredReplyChunk chunk;
1742     struct iovec iov[] = {
1743         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1744     };
1745 
1746     trace_nbd_co_send_structured_done(handle);
1747     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1748 
1749     return nbd_co_send_iov(client, iov, 1, errp);
1750 }
1751 
1752 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1753                                                     uint64_t handle,
1754                                                     uint64_t offset,
1755                                                     void *data,
1756                                                     size_t size,
1757                                                     bool final,
1758                                                     Error **errp)
1759 {
1760     NBDStructuredReadData chunk;
1761     struct iovec iov[] = {
1762         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1763         {.iov_base = data, .iov_len = size}
1764     };
1765 
1766     assert(size);
1767     trace_nbd_co_send_structured_read(handle, offset, data, size);
1768     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1769                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1770                  sizeof(chunk) - sizeof(chunk.h) + size);
1771     stq_be_p(&chunk.offset, offset);
1772 
1773     return nbd_co_send_iov(client, iov, 2, errp);
1774 }
1775 
1776 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1777                                                      uint64_t handle,
1778                                                      uint32_t error,
1779                                                      const char *msg,
1780                                                      Error **errp)
1781 {
1782     NBDStructuredError chunk;
1783     int nbd_err = system_errno_to_nbd_errno(error);
1784     struct iovec iov[] = {
1785         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1786         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1787     };
1788 
1789     assert(nbd_err);
1790     trace_nbd_co_send_structured_error(handle, nbd_err,
1791                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1792     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1793                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1794     stl_be_p(&chunk.error, nbd_err);
1795     stw_be_p(&chunk.message_length, iov[1].iov_len);
1796 
1797     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1798 }
1799 
1800 /* Do a sparse read and send the structured reply to the client.
1801  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1802  * reported to the client, at which point this function succeeds.
1803  */
1804 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1805                                                 uint64_t handle,
1806                                                 uint64_t offset,
1807                                                 uint8_t *data,
1808                                                 size_t size,
1809                                                 Error **errp)
1810 {
1811     int ret = 0;
1812     NBDExport *exp = client->exp;
1813     size_t progress = 0;
1814 
1815     while (progress < size) {
1816         int64_t pnum;
1817         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1818                                              offset + progress,
1819                                              size - progress, &pnum, NULL,
1820                                              NULL);
1821         bool final;
1822 
1823         if (status < 0) {
1824             char *msg = g_strdup_printf("unable to check for holes: %s",
1825                                         strerror(-status));
1826 
1827             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1828                                                errp);
1829             g_free(msg);
1830             return ret;
1831         }
1832         assert(pnum && pnum <= size - progress);
1833         final = progress + pnum == size;
1834         if (status & BDRV_BLOCK_ZERO) {
1835             NBDStructuredReadHole chunk;
1836             struct iovec iov[] = {
1837                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1838             };
1839 
1840             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1841                                                    pnum);
1842             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1843                          NBD_REPLY_TYPE_OFFSET_HOLE,
1844                          handle, sizeof(chunk) - sizeof(chunk.h));
1845             stq_be_p(&chunk.offset, offset + progress);
1846             stl_be_p(&chunk.length, pnum);
1847             ret = nbd_co_send_iov(client, iov, 1, errp);
1848         } else {
1849             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1850                             data + progress, pnum);
1851             if (ret < 0) {
1852                 error_setg_errno(errp, -ret, "reading from file failed");
1853                 break;
1854             }
1855             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1856                                               data + progress, pnum, final,
1857                                               errp);
1858         }
1859 
1860         if (ret < 0) {
1861             break;
1862         }
1863         progress += pnum;
1864     }
1865     return ret;
1866 }
1867 
1868 /*
1869  * Populate @extents from block status. Update @bytes to be the actual
1870  * length encoded (which may be smaller than the original), and update
1871  * @nb_extents to the number of extents used.
1872  *
1873  * Returns zero on success and -errno on bdrv_block_status_above failure.
1874  */
1875 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1876                                   uint64_t *bytes, NBDExtent *extents,
1877                                   unsigned int *nb_extents)
1878 {
1879     uint64_t remaining_bytes = *bytes;
1880     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1881     bool first_extent = true;
1882 
1883     assert(*nb_extents);
1884     while (remaining_bytes) {
1885         uint32_t flags;
1886         int64_t num;
1887         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1888                                           &num, NULL, NULL);
1889 
1890         if (ret < 0) {
1891             return ret;
1892         }
1893 
1894         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1895                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1896 
1897         if (first_extent) {
1898             extent->flags = flags;
1899             extent->length = num;
1900             first_extent = false;
1901         } else if (flags == extent->flags) {
1902             /* extend current extent */
1903             extent->length += num;
1904         } else {
1905             if (extent + 1 == extents_end) {
1906                 break;
1907             }
1908 
1909             /* start new extent */
1910             extent++;
1911             extent->flags = flags;
1912             extent->length = num;
1913         }
1914         offset += num;
1915         remaining_bytes -= num;
1916     }
1917 
1918     extents_end = extent + 1;
1919 
1920     for (extent = extents; extent < extents_end; extent++) {
1921         extent->flags = cpu_to_be32(extent->flags);
1922         extent->length = cpu_to_be32(extent->length);
1923     }
1924 
1925     *bytes -= remaining_bytes;
1926     *nb_extents = extents_end - extents;
1927 
1928     return 0;
1929 }
1930 
1931 /* nbd_co_send_extents
1932  *
1933  * @length is only for tracing purposes (and may be smaller or larger
1934  * than the client's original request). @last controls whether
1935  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1936  * big-endian format.
1937  */
1938 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1939                                NBDExtent *extents, unsigned int nb_extents,
1940                                uint64_t length, bool last,
1941                                uint32_t context_id, Error **errp)
1942 {
1943     NBDStructuredMeta chunk;
1944 
1945     struct iovec iov[] = {
1946         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1947         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1948     };
1949 
1950     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1951     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1952                  NBD_REPLY_TYPE_BLOCK_STATUS,
1953                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1954     stl_be_p(&chunk.context_id, context_id);
1955 
1956     return nbd_co_send_iov(client, iov, 2, errp);
1957 }
1958 
1959 /* Get block status from the exported device and send it to the client */
1960 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1961                                     BlockDriverState *bs, uint64_t offset,
1962                                     uint32_t length, bool dont_fragment,
1963                                     bool last, uint32_t context_id,
1964                                     Error **errp)
1965 {
1966     int ret;
1967     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
1968     NBDExtent *extents = g_new(NBDExtent, nb_extents);
1969     uint64_t final_length = length;
1970 
1971     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1972                                  &nb_extents);
1973     if (ret < 0) {
1974         g_free(extents);
1975         return nbd_co_send_structured_error(
1976                 client, handle, -ret, "can't get block status", errp);
1977     }
1978 
1979     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1980                               final_length, last, context_id, errp);
1981 
1982     g_free(extents);
1983 
1984     return ret;
1985 }
1986 
1987 /*
1988  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1989  * final extent may exceed the original @length. Store in @length the
1990  * byte length encoded (which may be smaller or larger than the
1991  * original), and return the number of extents used.
1992  */
1993 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1994                                       uint64_t *length, NBDExtent *extents,
1995                                       unsigned int nb_extents,
1996                                       bool dont_fragment)
1997 {
1998     uint64_t begin = offset, end = offset;
1999     uint64_t overall_end = offset + *length;
2000     unsigned int i = 0;
2001     BdrvDirtyBitmapIter *it;
2002     bool dirty;
2003 
2004     bdrv_dirty_bitmap_lock(bitmap);
2005 
2006     it = bdrv_dirty_iter_new(bitmap);
2007     dirty = bdrv_dirty_bitmap_get_locked(bitmap, offset);
2008 
2009     assert(begin < overall_end && nb_extents);
2010     while (begin < overall_end && i < nb_extents) {
2011         bool next_dirty = !dirty;
2012 
2013         if (dirty) {
2014             end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2015         } else {
2016             bdrv_set_dirty_iter(it, begin);
2017             end = bdrv_dirty_iter_next(it);
2018         }
2019         if (end == -1 || end - begin > UINT32_MAX) {
2020             /* Cap to an aligned value < 4G beyond begin. */
2021             end = MIN(bdrv_dirty_bitmap_size(bitmap),
2022                       begin + UINT32_MAX + 1 -
2023                       bdrv_dirty_bitmap_granularity(bitmap));
2024             next_dirty = dirty;
2025         }
2026         if (dont_fragment && end > overall_end) {
2027             end = overall_end;
2028         }
2029 
2030         extents[i].length = cpu_to_be32(end - begin);
2031         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2032         i++;
2033         begin = end;
2034         dirty = next_dirty;
2035     }
2036 
2037     bdrv_dirty_iter_free(it);
2038 
2039     bdrv_dirty_bitmap_unlock(bitmap);
2040 
2041     assert(offset < end);
2042     *length = end - offset;
2043     return i;
2044 }
2045 
2046 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2047                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2048                               uint32_t length, bool dont_fragment, bool last,
2049                               uint32_t context_id, Error **errp)
2050 {
2051     int ret;
2052     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2053     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2054     uint64_t final_length = length;
2055 
2056     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2057                                    nb_extents, dont_fragment);
2058 
2059     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2060                               final_length, last, context_id, errp);
2061 
2062     g_free(extents);
2063 
2064     return ret;
2065 }
2066 
2067 /* nbd_co_receive_request
2068  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2069  * connection right away, and any other negative value to report an error to
2070  * the client (although the caller may still need to disconnect after reporting
2071  * the error).
2072  */
2073 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2074                                   Error **errp)
2075 {
2076     NBDClient *client = req->client;
2077     int valid_flags;
2078 
2079     g_assert(qemu_in_coroutine());
2080     assert(client->recv_coroutine == qemu_coroutine_self());
2081     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2082         return -EIO;
2083     }
2084 
2085     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2086                                              nbd_cmd_lookup(request->type));
2087 
2088     if (request->type != NBD_CMD_WRITE) {
2089         /* No payload, we are ready to read the next request.  */
2090         req->complete = true;
2091     }
2092 
2093     if (request->type == NBD_CMD_DISC) {
2094         /* Special case: we're going to disconnect without a reply,
2095          * whether or not flags, from, or len are bogus */
2096         return -EIO;
2097     }
2098 
2099     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2100         request->type == NBD_CMD_CACHE)
2101     {
2102         if (request->len > NBD_MAX_BUFFER_SIZE) {
2103             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2104                        request->len, NBD_MAX_BUFFER_SIZE);
2105             return -EINVAL;
2106         }
2107 
2108         if (request->type != NBD_CMD_CACHE) {
2109             req->data = blk_try_blockalign(client->exp->blk, request->len);
2110             if (req->data == NULL) {
2111                 error_setg(errp, "No memory");
2112                 return -ENOMEM;
2113             }
2114         }
2115     }
2116 
2117     if (request->type == NBD_CMD_WRITE) {
2118         if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2119                      errp) < 0)
2120         {
2121             return -EIO;
2122         }
2123         req->complete = true;
2124 
2125         trace_nbd_co_receive_request_payload_received(request->handle,
2126                                                       request->len);
2127     }
2128 
2129     /* Sanity checks. */
2130     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2131         (request->type == NBD_CMD_WRITE ||
2132          request->type == NBD_CMD_WRITE_ZEROES ||
2133          request->type == NBD_CMD_TRIM)) {
2134         error_setg(errp, "Export is read-only");
2135         return -EROFS;
2136     }
2137     if (request->from > client->exp->size ||
2138         request->len > client->exp->size - request->from) {
2139         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2140                    ", Size: %" PRIu64, request->from, request->len,
2141                    client->exp->size);
2142         return (request->type == NBD_CMD_WRITE ||
2143                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2144     }
2145     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2146                                                 client->check_align)) {
2147         /*
2148          * The block layer gracefully handles unaligned requests, but
2149          * it's still worth tracing client non-compliance
2150          */
2151         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2152                                               request->from,
2153                                               request->len,
2154                                               client->check_align);
2155     }
2156     valid_flags = NBD_CMD_FLAG_FUA;
2157     if (request->type == NBD_CMD_READ && client->structured_reply) {
2158         valid_flags |= NBD_CMD_FLAG_DF;
2159     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2160         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2161     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2162         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2163     }
2164     if (request->flags & ~valid_flags) {
2165         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2166                    nbd_cmd_lookup(request->type), request->flags);
2167         return -EINVAL;
2168     }
2169 
2170     return 0;
2171 }
2172 
2173 /* Send simple reply without a payload, or a structured error
2174  * @error_msg is ignored if @ret >= 0
2175  * Returns 0 if connection is still live, -errno on failure to talk to client
2176  */
2177 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2178                                                uint64_t handle,
2179                                                int ret,
2180                                                const char *error_msg,
2181                                                Error **errp)
2182 {
2183     if (client->structured_reply && ret < 0) {
2184         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2185                                             errp);
2186     } else {
2187         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2188                                         NULL, 0, errp);
2189     }
2190 }
2191 
2192 /* Handle NBD_CMD_READ request.
2193  * Return -errno if sending fails. Other errors are reported directly to the
2194  * client as an error reply. */
2195 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2196                                         uint8_t *data, Error **errp)
2197 {
2198     int ret;
2199     NBDExport *exp = client->exp;
2200 
2201     assert(request->type == NBD_CMD_READ);
2202 
2203     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2204     if (request->flags & NBD_CMD_FLAG_FUA) {
2205         ret = blk_co_flush(exp->blk);
2206         if (ret < 0) {
2207             return nbd_send_generic_reply(client, request->handle, ret,
2208                                           "flush failed", errp);
2209         }
2210     }
2211 
2212     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2213         request->len)
2214     {
2215         return nbd_co_send_sparse_read(client, request->handle, request->from,
2216                                        data, request->len, errp);
2217     }
2218 
2219     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2220                     request->len);
2221     if (ret < 0) {
2222         return nbd_send_generic_reply(client, request->handle, ret,
2223                                       "reading from file failed", errp);
2224     }
2225 
2226     if (client->structured_reply) {
2227         if (request->len) {
2228             return nbd_co_send_structured_read(client, request->handle,
2229                                                request->from, data,
2230                                                request->len, true, errp);
2231         } else {
2232             return nbd_co_send_structured_done(client, request->handle, errp);
2233         }
2234     } else {
2235         return nbd_co_send_simple_reply(client, request->handle, 0,
2236                                         data, request->len, errp);
2237     }
2238 }
2239 
2240 /*
2241  * nbd_do_cmd_cache
2242  *
2243  * Handle NBD_CMD_CACHE request.
2244  * Return -errno if sending fails. Other errors are reported directly to the
2245  * client as an error reply.
2246  */
2247 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2248                                          Error **errp)
2249 {
2250     int ret;
2251     NBDExport *exp = client->exp;
2252 
2253     assert(request->type == NBD_CMD_CACHE);
2254 
2255     ret = blk_co_preadv(exp->blk, request->from + exp->dev_offset, request->len,
2256                         NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2257 
2258     return nbd_send_generic_reply(client, request->handle, ret,
2259                                   "caching data failed", errp);
2260 }
2261 
2262 /* Handle NBD request.
2263  * Return -errno if sending fails. Other errors are reported directly to the
2264  * client as an error reply. */
2265 static coroutine_fn int nbd_handle_request(NBDClient *client,
2266                                            NBDRequest *request,
2267                                            uint8_t *data, Error **errp)
2268 {
2269     int ret;
2270     int flags;
2271     NBDExport *exp = client->exp;
2272     char *msg;
2273 
2274     switch (request->type) {
2275     case NBD_CMD_CACHE:
2276         return nbd_do_cmd_cache(client, request, errp);
2277 
2278     case NBD_CMD_READ:
2279         return nbd_do_cmd_read(client, request, data, errp);
2280 
2281     case NBD_CMD_WRITE:
2282         flags = 0;
2283         if (request->flags & NBD_CMD_FLAG_FUA) {
2284             flags |= BDRV_REQ_FUA;
2285         }
2286         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2287                          data, request->len, flags);
2288         return nbd_send_generic_reply(client, request->handle, ret,
2289                                       "writing to file failed", errp);
2290 
2291     case NBD_CMD_WRITE_ZEROES:
2292         flags = 0;
2293         if (request->flags & NBD_CMD_FLAG_FUA) {
2294             flags |= BDRV_REQ_FUA;
2295         }
2296         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2297             flags |= BDRV_REQ_MAY_UNMAP;
2298         }
2299         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2300                                 request->len, flags);
2301         return nbd_send_generic_reply(client, request->handle, ret,
2302                                       "writing to file failed", errp);
2303 
2304     case NBD_CMD_DISC:
2305         /* unreachable, thanks to special case in nbd_co_receive_request() */
2306         abort();
2307 
2308     case NBD_CMD_FLUSH:
2309         ret = blk_co_flush(exp->blk);
2310         return nbd_send_generic_reply(client, request->handle, ret,
2311                                       "flush failed", errp);
2312 
2313     case NBD_CMD_TRIM:
2314         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2315                               request->len);
2316         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2317             ret = blk_co_flush(exp->blk);
2318         }
2319         return nbd_send_generic_reply(client, request->handle, ret,
2320                                       "discard failed", errp);
2321 
2322     case NBD_CMD_BLOCK_STATUS:
2323         if (!request->len) {
2324             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2325                                           "need non-zero length", errp);
2326         }
2327         if (client->export_meta.valid &&
2328             (client->export_meta.base_allocation ||
2329              client->export_meta.bitmap))
2330         {
2331             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2332 
2333             if (client->export_meta.base_allocation) {
2334                 ret = nbd_co_send_block_status(client, request->handle,
2335                                                blk_bs(exp->blk), request->from,
2336                                                request->len, dont_fragment,
2337                                                !client->export_meta.bitmap,
2338                                                NBD_META_ID_BASE_ALLOCATION,
2339                                                errp);
2340                 if (ret < 0) {
2341                     return ret;
2342                 }
2343             }
2344 
2345             if (client->export_meta.bitmap) {
2346                 ret = nbd_co_send_bitmap(client, request->handle,
2347                                          client->exp->export_bitmap,
2348                                          request->from, request->len,
2349                                          dont_fragment,
2350                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2351                 if (ret < 0) {
2352                     return ret;
2353                 }
2354             }
2355 
2356             return ret;
2357         } else {
2358             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2359                                           "CMD_BLOCK_STATUS not negotiated",
2360                                           errp);
2361         }
2362 
2363     default:
2364         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2365                               request->type);
2366         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2367                                      errp);
2368         g_free(msg);
2369         return ret;
2370     }
2371 }
2372 
2373 /* Owns a reference to the NBDClient passed as opaque.  */
2374 static coroutine_fn void nbd_trip(void *opaque)
2375 {
2376     NBDClient *client = opaque;
2377     NBDRequestData *req;
2378     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2379     int ret;
2380     Error *local_err = NULL;
2381 
2382     trace_nbd_trip();
2383     if (client->closing) {
2384         nbd_client_put(client);
2385         return;
2386     }
2387 
2388     req = nbd_request_get(client);
2389     ret = nbd_co_receive_request(req, &request, &local_err);
2390     client->recv_coroutine = NULL;
2391 
2392     if (client->closing) {
2393         /*
2394          * The client may be closed when we are blocked in
2395          * nbd_co_receive_request()
2396          */
2397         goto done;
2398     }
2399 
2400     nbd_client_receive_next_request(client);
2401     if (ret == -EIO) {
2402         goto disconnect;
2403     }
2404 
2405     if (ret < 0) {
2406         /* It wans't -EIO, so, according to nbd_co_receive_request()
2407          * semantics, we should return the error to the client. */
2408         Error *export_err = local_err;
2409 
2410         local_err = NULL;
2411         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2412                                      error_get_pretty(export_err), &local_err);
2413         error_free(export_err);
2414     } else {
2415         ret = nbd_handle_request(client, &request, req->data, &local_err);
2416     }
2417     if (ret < 0) {
2418         error_prepend(&local_err, "Failed to send reply: ");
2419         goto disconnect;
2420     }
2421 
2422     /* We must disconnect after NBD_CMD_WRITE if we did not
2423      * read the payload.
2424      */
2425     if (!req->complete) {
2426         error_setg(&local_err, "Request handling failed in intermediate state");
2427         goto disconnect;
2428     }
2429 
2430 done:
2431     nbd_request_put(req);
2432     nbd_client_put(client);
2433     return;
2434 
2435 disconnect:
2436     if (local_err) {
2437         error_reportf_err(local_err, "Disconnect client, due to: ");
2438     }
2439     nbd_request_put(req);
2440     client_close(client, true);
2441     nbd_client_put(client);
2442 }
2443 
2444 static void nbd_client_receive_next_request(NBDClient *client)
2445 {
2446     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2447         nbd_client_get(client);
2448         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2449         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2450     }
2451 }
2452 
2453 static coroutine_fn void nbd_co_client_start(void *opaque)
2454 {
2455     NBDClient *client = opaque;
2456     Error *local_err = NULL;
2457 
2458     qemu_co_mutex_init(&client->send_lock);
2459 
2460     if (nbd_negotiate(client, &local_err)) {
2461         if (local_err) {
2462             error_report_err(local_err);
2463         }
2464         client_close(client, false);
2465         return;
2466     }
2467 
2468     nbd_client_receive_next_request(client);
2469 }
2470 
2471 /*
2472  * Create a new client listener using the given channel @sioc.
2473  * Begin servicing it in a coroutine.  When the connection closes, call
2474  * @close_fn with an indication of whether the client completed negotiation.
2475  */
2476 void nbd_client_new(QIOChannelSocket *sioc,
2477                     QCryptoTLSCreds *tlscreds,
2478                     const char *tlsauthz,
2479                     void (*close_fn)(NBDClient *, bool))
2480 {
2481     NBDClient *client;
2482     Coroutine *co;
2483 
2484     client = g_new0(NBDClient, 1);
2485     client->refcount = 1;
2486     client->tlscreds = tlscreds;
2487     if (tlscreds) {
2488         object_ref(OBJECT(client->tlscreds));
2489     }
2490     client->tlsauthz = g_strdup(tlsauthz);
2491     client->sioc = sioc;
2492     object_ref(OBJECT(client->sioc));
2493     client->ioc = QIO_CHANNEL(sioc);
2494     object_ref(OBJECT(client->ioc));
2495     client->close_fn = close_fn;
2496 
2497     co = qemu_coroutine_create(nbd_co_client_start, client);
2498     qemu_coroutine_enter(co);
2499 }
2500