xref: /openbmc/qemu/nbd/server.c (revision 74433bf0)
1 /*
2  *  Copyright (C) 2016-2018 Red Hat, Inc.
3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
4  *
5  *  Network Block Device Server Side
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; under version 2 of the License.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "trace.h"
23 #include "nbd-internal.h"
24 
25 #define NBD_META_ID_BASE_ALLOCATION 0
26 #define NBD_META_ID_DIRTY_BITMAP 1
27 
28 /* NBD_MAX_BITMAP_EXTENTS: 1 mb of extents data. An empirical
29  * constant. If an increase is needed, note that the NBD protocol
30  * recommends no larger than 32 mb, so that the client won't consider
31  * the reply as a denial of service attack. */
32 #define NBD_MAX_BITMAP_EXTENTS (0x100000 / 8)
33 
34 static int system_errno_to_nbd_errno(int err)
35 {
36     switch (err) {
37     case 0:
38         return NBD_SUCCESS;
39     case EPERM:
40     case EROFS:
41         return NBD_EPERM;
42     case EIO:
43         return NBD_EIO;
44     case ENOMEM:
45         return NBD_ENOMEM;
46 #ifdef EDQUOT
47     case EDQUOT:
48 #endif
49     case EFBIG:
50     case ENOSPC:
51         return NBD_ENOSPC;
52     case EOVERFLOW:
53         return NBD_EOVERFLOW;
54     case ESHUTDOWN:
55         return NBD_ESHUTDOWN;
56     case EINVAL:
57     default:
58         return NBD_EINVAL;
59     }
60 }
61 
62 /* Definitions for opaque data types */
63 
64 typedef struct NBDRequestData NBDRequestData;
65 
66 struct NBDRequestData {
67     QSIMPLEQ_ENTRY(NBDRequestData) entry;
68     NBDClient *client;
69     uint8_t *data;
70     bool complete;
71 };
72 
73 struct NBDExport {
74     int refcount;
75     void (*close)(NBDExport *exp);
76 
77     BlockBackend *blk;
78     char *name;
79     char *description;
80     uint64_t dev_offset;
81     uint64_t size;
82     uint16_t nbdflags;
83     QTAILQ_HEAD(, NBDClient) clients;
84     QTAILQ_ENTRY(NBDExport) next;
85 
86     AioContext *ctx;
87 
88     BlockBackend *eject_notifier_blk;
89     Notifier eject_notifier;
90 
91     BdrvDirtyBitmap *export_bitmap;
92     char *export_bitmap_context;
93 };
94 
95 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
96 
97 /* NBDExportMetaContexts represents a list of contexts to be exported,
98  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
99  * NBD_OPT_LIST_META_CONTEXT. */
100 typedef struct NBDExportMetaContexts {
101     NBDExport *exp;
102     bool valid; /* means that negotiation of the option finished without
103                    errors */
104     bool base_allocation; /* export base:allocation context (block status) */
105     bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */
106 } NBDExportMetaContexts;
107 
108 struct NBDClient {
109     int refcount;
110     void (*close_fn)(NBDClient *client, bool negotiated);
111 
112     NBDExport *exp;
113     QCryptoTLSCreds *tlscreds;
114     char *tlsauthz;
115     QIOChannelSocket *sioc; /* The underlying data channel */
116     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
117 
118     Coroutine *recv_coroutine;
119 
120     CoMutex send_lock;
121     Coroutine *send_coroutine;
122 
123     QTAILQ_ENTRY(NBDClient) next;
124     int nb_requests;
125     bool closing;
126 
127     uint32_t check_align; /* If non-zero, check for aligned client requests */
128 
129     bool structured_reply;
130     NBDExportMetaContexts export_meta;
131 
132     uint32_t opt; /* Current option being negotiated */
133     uint32_t optlen; /* remaining length of data in ioc for the option being
134                         negotiated now */
135 };
136 
137 static void nbd_client_receive_next_request(NBDClient *client);
138 
139 /* Basic flow for negotiation
140 
141    Server         Client
142    Negotiate
143 
144    or
145 
146    Server         Client
147    Negotiate #1
148                   Option
149    Negotiate #2
150 
151    ----
152 
153    followed by
154 
155    Server         Client
156                   Request
157    Response
158                   Request
159    Response
160                   ...
161    ...
162                   Request (type == 2)
163 
164 */
165 
166 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
167                                      uint32_t type, uint32_t length)
168 {
169     stq_be_p(&rep->magic, NBD_REP_MAGIC);
170     stl_be_p(&rep->option, option);
171     stl_be_p(&rep->type, type);
172     stl_be_p(&rep->length, length);
173 }
174 
175 /* Send a reply header, including length, but no payload.
176  * Return -errno on error, 0 on success. */
177 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
178                                       uint32_t len, Error **errp)
179 {
180     NBDOptionReply rep;
181 
182     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
183                                      type, nbd_rep_lookup(type), len);
184 
185     assert(len < NBD_MAX_BUFFER_SIZE);
186 
187     set_be_option_rep(&rep, client->opt, type, len);
188     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
189 }
190 
191 /* Send a reply header with default 0 length.
192  * Return -errno on error, 0 on success. */
193 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
194                                   Error **errp)
195 {
196     return nbd_negotiate_send_rep_len(client, type, 0, errp);
197 }
198 
199 /* Send an error reply.
200  * Return -errno on error, 0 on success. */
201 static int GCC_FMT_ATTR(4, 0)
202 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
203                             Error **errp, const char *fmt, va_list va)
204 {
205     char *msg;
206     int ret;
207     size_t len;
208 
209     msg = g_strdup_vprintf(fmt, va);
210     len = strlen(msg);
211     assert(len < 4096);
212     trace_nbd_negotiate_send_rep_err(msg);
213     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
214     if (ret < 0) {
215         goto out;
216     }
217     if (nbd_write(client->ioc, msg, len, errp) < 0) {
218         error_prepend(errp, "write failed (error message): ");
219         ret = -EIO;
220     } else {
221         ret = 0;
222     }
223 
224 out:
225     g_free(msg);
226     return ret;
227 }
228 
229 /* Send an error reply.
230  * Return -errno on error, 0 on success. */
231 static int GCC_FMT_ATTR(4, 5)
232 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
233                            Error **errp, const char *fmt, ...)
234 {
235     va_list va;
236     int ret;
237 
238     va_start(va, fmt);
239     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
240     va_end(va);
241     return ret;
242 }
243 
244 /* Drop remainder of the current option, and send a reply with the
245  * given error type and message. Return -errno on read or write
246  * failure; or 0 if connection is still live. */
247 static int GCC_FMT_ATTR(4, 0)
248 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
249               const char *fmt, va_list va)
250 {
251     int ret = nbd_drop(client->ioc, client->optlen, errp);
252 
253     client->optlen = 0;
254     if (!ret) {
255         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
256     }
257     return ret;
258 }
259 
260 static int GCC_FMT_ATTR(4, 5)
261 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
262              const char *fmt, ...)
263 {
264     int ret;
265     va_list va;
266 
267     va_start(va, fmt);
268     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
269     va_end(va);
270 
271     return ret;
272 }
273 
274 static int GCC_FMT_ATTR(3, 4)
275 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
276 {
277     int ret;
278     va_list va;
279 
280     va_start(va, fmt);
281     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
282     va_end(va);
283 
284     return ret;
285 }
286 
287 /* Read size bytes from the unparsed payload of the current option.
288  * Return -errno on I/O error, 0 if option was completely handled by
289  * sending a reply about inconsistent lengths, or 1 on success. */
290 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
291                         Error **errp)
292 {
293     if (size > client->optlen) {
294         return nbd_opt_invalid(client, errp,
295                                "Inconsistent lengths in option %s",
296                                nbd_opt_lookup(client->opt));
297     }
298     client->optlen -= size;
299     return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1;
300 }
301 
302 /* Drop size bytes from the unparsed payload of the current option.
303  * Return -errno on I/O error, 0 if option was completely handled by
304  * sending a reply about inconsistent lengths, or 1 on success. */
305 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
306 {
307     if (size > client->optlen) {
308         return nbd_opt_invalid(client, errp,
309                                "Inconsistent lengths in option %s",
310                                nbd_opt_lookup(client->opt));
311     }
312     client->optlen -= size;
313     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
314 }
315 
316 /* nbd_opt_read_name
317  *
318  * Read a string with the format:
319  *   uint32_t len     (<= NBD_MAX_NAME_SIZE)
320  *   len bytes string (not 0-terminated)
321  *
322  * @name should be enough to store NBD_MAX_NAME_SIZE+1.
323  * If @length is non-null, it will be set to the actual string length.
324  *
325  * Return -errno on I/O error, 0 if option was completely handled by
326  * sending a reply about inconsistent lengths, or 1 on success.
327  */
328 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length,
329                              Error **errp)
330 {
331     int ret;
332     uint32_t len;
333 
334     ret = nbd_opt_read(client, &len, sizeof(len), errp);
335     if (ret <= 0) {
336         return ret;
337     }
338     len = cpu_to_be32(len);
339 
340     if (len > NBD_MAX_NAME_SIZE) {
341         return nbd_opt_invalid(client, errp,
342                                "Invalid name length: %" PRIu32, len);
343     }
344 
345     ret = nbd_opt_read(client, name, len, errp);
346     if (ret <= 0) {
347         return ret;
348     }
349     name[len] = '\0';
350 
351     if (length) {
352         *length = len;
353     }
354 
355     return 1;
356 }
357 
358 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
359  * Return -errno on error, 0 on success. */
360 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
361                                        Error **errp)
362 {
363     size_t name_len, desc_len;
364     uint32_t len;
365     const char *name = exp->name ? exp->name : "";
366     const char *desc = exp->description ? exp->description : "";
367     QIOChannel *ioc = client->ioc;
368     int ret;
369 
370     trace_nbd_negotiate_send_rep_list(name, desc);
371     name_len = strlen(name);
372     desc_len = strlen(desc);
373     len = name_len + desc_len + sizeof(len);
374     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
375     if (ret < 0) {
376         return ret;
377     }
378 
379     len = cpu_to_be32(name_len);
380     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
381         error_prepend(errp, "write failed (name length): ");
382         return -EINVAL;
383     }
384 
385     if (nbd_write(ioc, name, name_len, errp) < 0) {
386         error_prepend(errp, "write failed (name buffer): ");
387         return -EINVAL;
388     }
389 
390     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
391         error_prepend(errp, "write failed (description buffer): ");
392         return -EINVAL;
393     }
394 
395     return 0;
396 }
397 
398 /* Process the NBD_OPT_LIST command, with a potential series of replies.
399  * Return -errno on error, 0 on success. */
400 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
401 {
402     NBDExport *exp;
403     assert(client->opt == NBD_OPT_LIST);
404 
405     /* For each export, send a NBD_REP_SERVER reply. */
406     QTAILQ_FOREACH(exp, &exports, next) {
407         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
408             return -EINVAL;
409         }
410     }
411     /* Finish with a NBD_REP_ACK. */
412     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
413 }
414 
415 static void nbd_check_meta_export(NBDClient *client)
416 {
417     client->export_meta.valid &= client->exp == client->export_meta.exp;
418 }
419 
420 /* Send a reply to NBD_OPT_EXPORT_NAME.
421  * Return -errno on error, 0 on success. */
422 static int nbd_negotiate_handle_export_name(NBDClient *client,
423                                             uint16_t myflags, bool no_zeroes,
424                                             Error **errp)
425 {
426     char name[NBD_MAX_NAME_SIZE + 1];
427     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
428     size_t len;
429     int ret;
430 
431     /* Client sends:
432         [20 ..  xx]   export name (length bytes)
433        Server replies:
434         [ 0 ..   7]   size
435         [ 8 ..   9]   export flags
436         [10 .. 133]   reserved     (0) [unless no_zeroes]
437      */
438     trace_nbd_negotiate_handle_export_name();
439     if (client->optlen >= sizeof(name)) {
440         error_setg(errp, "Bad length received");
441         return -EINVAL;
442     }
443     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
444         return -EIO;
445     }
446     name[client->optlen] = '\0';
447     client->optlen = 0;
448 
449     trace_nbd_negotiate_handle_export_name_request(name);
450 
451     client->exp = nbd_export_find(name);
452     if (!client->exp) {
453         error_setg(errp, "export not found");
454         return -EINVAL;
455     }
456 
457     trace_nbd_negotiate_new_style_size_flags(client->exp->size,
458                                              client->exp->nbdflags | myflags);
459     stq_be_p(buf, client->exp->size);
460     stw_be_p(buf + 8, client->exp->nbdflags | myflags);
461     len = no_zeroes ? 10 : sizeof(buf);
462     ret = nbd_write(client->ioc, buf, len, errp);
463     if (ret < 0) {
464         error_prepend(errp, "write failed: ");
465         return ret;
466     }
467 
468     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
469     nbd_export_get(client->exp);
470     nbd_check_meta_export(client);
471 
472     return 0;
473 }
474 
475 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
476  * The buffer does NOT include the info type prefix.
477  * Return -errno on error, 0 if ready to send more. */
478 static int nbd_negotiate_send_info(NBDClient *client,
479                                    uint16_t info, uint32_t length, void *buf,
480                                    Error **errp)
481 {
482     int rc;
483 
484     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
485     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
486                                     sizeof(info) + length, errp);
487     if (rc < 0) {
488         return rc;
489     }
490     info = cpu_to_be16(info);
491     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
492         return -EIO;
493     }
494     if (nbd_write(client->ioc, buf, length, errp) < 0) {
495         return -EIO;
496     }
497     return 0;
498 }
499 
500 /* nbd_reject_length: Handle any unexpected payload.
501  * @fatal requests that we quit talking to the client, even if we are able
502  * to successfully send an error reply.
503  * Return:
504  * -errno  transmission error occurred or @fatal was requested, errp is set
505  * 0       error message successfully sent to client, errp is not set
506  */
507 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
508 {
509     int ret;
510 
511     assert(client->optlen);
512     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
513                           nbd_opt_lookup(client->opt));
514     if (fatal && !ret) {
515         error_setg(errp, "option '%s' has unexpected length",
516                    nbd_opt_lookup(client->opt));
517         return -EINVAL;
518     }
519     return ret;
520 }
521 
522 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
523  * Return -errno on error, 0 if ready for next option, and 1 to move
524  * into transmission phase.  */
525 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags,
526                                      Error **errp)
527 {
528     int rc;
529     char name[NBD_MAX_NAME_SIZE + 1];
530     NBDExport *exp;
531     uint16_t requests;
532     uint16_t request;
533     uint32_t namelen;
534     bool sendname = false;
535     bool blocksize = false;
536     uint32_t sizes[3];
537     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
538     uint32_t check_align = 0;
539 
540     /* Client sends:
541         4 bytes: L, name length (can be 0)
542         L bytes: export name
543         2 bytes: N, number of requests (can be 0)
544         N * 2 bytes: N requests
545     */
546     rc = nbd_opt_read_name(client, name, &namelen, errp);
547     if (rc <= 0) {
548         return rc;
549     }
550     trace_nbd_negotiate_handle_export_name_request(name);
551 
552     rc = nbd_opt_read(client, &requests, sizeof(requests), errp);
553     if (rc <= 0) {
554         return rc;
555     }
556     requests = be16_to_cpu(requests);
557     trace_nbd_negotiate_handle_info_requests(requests);
558     while (requests--) {
559         rc = nbd_opt_read(client, &request, sizeof(request), errp);
560         if (rc <= 0) {
561             return rc;
562         }
563         request = be16_to_cpu(request);
564         trace_nbd_negotiate_handle_info_request(request,
565                                                 nbd_info_lookup(request));
566         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
567          * everything else is either a request we don't know or
568          * something we send regardless of request */
569         switch (request) {
570         case NBD_INFO_NAME:
571             sendname = true;
572             break;
573         case NBD_INFO_BLOCK_SIZE:
574             blocksize = true;
575             break;
576         }
577     }
578     if (client->optlen) {
579         return nbd_reject_length(client, false, errp);
580     }
581 
582     exp = nbd_export_find(name);
583     if (!exp) {
584         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
585                                           errp, "export '%s' not present",
586                                           name);
587     }
588 
589     /* Don't bother sending NBD_INFO_NAME unless client requested it */
590     if (sendname) {
591         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
592                                      errp);
593         if (rc < 0) {
594             return rc;
595         }
596     }
597 
598     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
599      * client request */
600     if (exp->description) {
601         size_t len = strlen(exp->description);
602 
603         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
604                                      len, exp->description, errp);
605         if (rc < 0) {
606             return rc;
607         }
608     }
609 
610     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
611      * according to whether the client requested it, and according to
612      * whether this is OPT_INFO or OPT_GO. */
613     /* minimum - 1 for back-compat, or actual if client will obey it. */
614     if (client->opt == NBD_OPT_INFO || blocksize) {
615         check_align = sizes[0] = blk_get_request_alignment(exp->blk);
616     } else {
617         sizes[0] = 1;
618     }
619     assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
620     /* preferred - Hard-code to 4096 for now.
621      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
622     sizes[1] = MAX(4096, sizes[0]);
623     /* maximum - At most 32M, but smaller as appropriate. */
624     sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE);
625     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
626     sizes[0] = cpu_to_be32(sizes[0]);
627     sizes[1] = cpu_to_be32(sizes[1]);
628     sizes[2] = cpu_to_be32(sizes[2]);
629     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
630                                  sizeof(sizes), sizes, errp);
631     if (rc < 0) {
632         return rc;
633     }
634 
635     /* Send NBD_INFO_EXPORT always */
636     trace_nbd_negotiate_new_style_size_flags(exp->size,
637                                              exp->nbdflags | myflags);
638     stq_be_p(buf, exp->size);
639     stw_be_p(buf + 8, exp->nbdflags | myflags);
640     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
641                                  sizeof(buf), buf, errp);
642     if (rc < 0) {
643         return rc;
644     }
645 
646     /*
647      * If the client is just asking for NBD_OPT_INFO, but forgot to
648      * request block sizes in a situation that would impact
649      * performance, then return an error. But for NBD_OPT_GO, we
650      * tolerate all clients, regardless of alignments.
651      */
652     if (client->opt == NBD_OPT_INFO && !blocksize &&
653         blk_get_request_alignment(exp->blk) > 1) {
654         return nbd_negotiate_send_rep_err(client,
655                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
656                                           errp,
657                                           "request NBD_INFO_BLOCK_SIZE to "
658                                           "use this export");
659     }
660 
661     /* Final reply */
662     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
663     if (rc < 0) {
664         return rc;
665     }
666 
667     if (client->opt == NBD_OPT_GO) {
668         client->exp = exp;
669         client->check_align = check_align;
670         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
671         nbd_export_get(client->exp);
672         nbd_check_meta_export(client);
673         rc = 1;
674     }
675     return rc;
676 }
677 
678 
679 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
680  * new channel for all further (now-encrypted) communication. */
681 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
682                                                  Error **errp)
683 {
684     QIOChannel *ioc;
685     QIOChannelTLS *tioc;
686     struct NBDTLSHandshakeData data = { 0 };
687 
688     assert(client->opt == NBD_OPT_STARTTLS);
689 
690     trace_nbd_negotiate_handle_starttls();
691     ioc = client->ioc;
692 
693     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
694         return NULL;
695     }
696 
697     tioc = qio_channel_tls_new_server(ioc,
698                                       client->tlscreds,
699                                       client->tlsauthz,
700                                       errp);
701     if (!tioc) {
702         return NULL;
703     }
704 
705     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
706     trace_nbd_negotiate_handle_starttls_handshake();
707     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
708     qio_channel_tls_handshake(tioc,
709                               nbd_tls_handshake,
710                               &data,
711                               NULL,
712                               NULL);
713 
714     if (!data.complete) {
715         g_main_loop_run(data.loop);
716     }
717     g_main_loop_unref(data.loop);
718     if (data.error) {
719         object_unref(OBJECT(tioc));
720         error_propagate(errp, data.error);
721         return NULL;
722     }
723 
724     return QIO_CHANNEL(tioc);
725 }
726 
727 /* nbd_negotiate_send_meta_context
728  *
729  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
730  *
731  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
732  */
733 static int nbd_negotiate_send_meta_context(NBDClient *client,
734                                            const char *context,
735                                            uint32_t context_id,
736                                            Error **errp)
737 {
738     NBDOptionReplyMetaContext opt;
739     struct iovec iov[] = {
740         {.iov_base = &opt, .iov_len = sizeof(opt)},
741         {.iov_base = (void *)context, .iov_len = strlen(context)}
742     };
743 
744     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
745         context_id = 0;
746     }
747 
748     trace_nbd_negotiate_meta_query_reply(context, context_id);
749     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
750                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
751     stl_be_p(&opt.context_id, context_id);
752 
753     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
754 }
755 
756 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern.
757  * @match is never set to false.
758  *
759  * Return -errno on I/O error, 0 if option was completely handled by
760  * sending a reply about inconsistent lengths, or 1 on success.
761  *
762  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
763  * It only means that there are no errors.
764  */
765 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match,
766                             Error **errp)
767 {
768     int ret;
769     char *query;
770     size_t len = strlen(pattern);
771 
772     assert(len);
773 
774     query = g_malloc(len);
775     ret = nbd_opt_read(client, query, len, errp);
776     if (ret <= 0) {
777         g_free(query);
778         return ret;
779     }
780 
781     if (strncmp(query, pattern, len) == 0) {
782         trace_nbd_negotiate_meta_query_parse(pattern);
783         *match = true;
784     } else {
785         trace_nbd_negotiate_meta_query_skip("pattern not matched");
786     }
787     g_free(query);
788 
789     return 1;
790 }
791 
792 /*
793  * Read @len bytes, and set @match to true if they match @pattern, or if @len
794  * is 0 and the client is performing _LIST_. @match is never set to false.
795  *
796  * Return -errno on I/O error, 0 if option was completely handled by
797  * sending a reply about inconsistent lengths, or 1 on success.
798  *
799  * Note: return code = 1 doesn't mean that we've read exactly @pattern.
800  * It only means that there are no errors.
801  */
802 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
803                                      uint32_t len, bool *match, Error **errp)
804 {
805     if (len == 0) {
806         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
807             *match = true;
808         }
809         trace_nbd_negotiate_meta_query_parse("empty");
810         return 1;
811     }
812 
813     if (len != strlen(pattern)) {
814         trace_nbd_negotiate_meta_query_skip("different lengths");
815         return nbd_opt_skip(client, len, errp);
816     }
817 
818     return nbd_meta_pattern(client, pattern, match, errp);
819 }
820 
821 /* nbd_meta_base_query
822  *
823  * Handle queries to 'base' namespace. For now, only the base:allocation
824  * context is available.  'len' is the amount of text remaining to be read from
825  * the current name, after the 'base:' portion has been stripped.
826  *
827  * Return -errno on I/O error, 0 if option was completely handled by
828  * sending a reply about inconsistent lengths, or 1 on success.
829  */
830 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta,
831                                uint32_t len, Error **errp)
832 {
833     return nbd_meta_empty_or_pattern(client, "allocation", len,
834                                      &meta->base_allocation, errp);
835 }
836 
837 /* nbd_meta_bitmap_query
838  *
839  * Handle query to 'qemu:' namespace.
840  * @len is the amount of text remaining to be read from the current name, after
841  * the 'qemu:' portion has been stripped.
842  *
843  * Return -errno on I/O error, 0 if option was completely handled by
844  * sending a reply about inconsistent lengths, or 1 on success. */
845 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta,
846                                uint32_t len, Error **errp)
847 {
848     bool dirty_bitmap = false;
849     size_t dirty_bitmap_len = strlen("dirty-bitmap:");
850     int ret;
851 
852     if (!meta->exp->export_bitmap) {
853         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported");
854         return nbd_opt_skip(client, len, errp);
855     }
856 
857     if (len == 0) {
858         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
859             meta->bitmap = true;
860         }
861         trace_nbd_negotiate_meta_query_parse("empty");
862         return 1;
863     }
864 
865     if (len < dirty_bitmap_len) {
866         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
867         return nbd_opt_skip(client, len, errp);
868     }
869 
870     len -= dirty_bitmap_len;
871     ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp);
872     if (ret <= 0) {
873         return ret;
874     }
875     if (!dirty_bitmap) {
876         trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:");
877         return nbd_opt_skip(client, len, errp);
878     }
879 
880     trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
881 
882     return nbd_meta_empty_or_pattern(
883             client, meta->exp->export_bitmap_context +
884             strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp);
885 }
886 
887 /* nbd_negotiate_meta_query
888  *
889  * Parse namespace name and call corresponding function to parse body of the
890  * query.
891  *
892  * The only supported namespace now is 'base'.
893  *
894  * The function aims not wasting time and memory to read long unknown namespace
895  * names.
896  *
897  * Return -errno on I/O error, 0 if option was completely handled by
898  * sending a reply about inconsistent lengths, or 1 on success. */
899 static int nbd_negotiate_meta_query(NBDClient *client,
900                                     NBDExportMetaContexts *meta, Error **errp)
901 {
902     /*
903      * Both 'qemu' and 'base' namespaces have length = 5 including a
904      * colon. If another length namespace is later introduced, this
905      * should certainly be refactored.
906      */
907     int ret;
908     size_t ns_len = 5;
909     char ns[5];
910     uint32_t len;
911 
912     ret = nbd_opt_read(client, &len, sizeof(len), errp);
913     if (ret <= 0) {
914         return ret;
915     }
916     len = cpu_to_be32(len);
917 
918     if (len < ns_len) {
919         trace_nbd_negotiate_meta_query_skip("length too short");
920         return nbd_opt_skip(client, len, errp);
921     }
922 
923     len -= ns_len;
924     ret = nbd_opt_read(client, ns, ns_len, errp);
925     if (ret <= 0) {
926         return ret;
927     }
928 
929     if (!strncmp(ns, "base:", ns_len)) {
930         trace_nbd_negotiate_meta_query_parse("base:");
931         return nbd_meta_base_query(client, meta, len, errp);
932     } else if (!strncmp(ns, "qemu:", ns_len)) {
933         trace_nbd_negotiate_meta_query_parse("qemu:");
934         return nbd_meta_qemu_query(client, meta, len, errp);
935     }
936 
937     trace_nbd_negotiate_meta_query_skip("unknown namespace");
938     return nbd_opt_skip(client, len, errp);
939 }
940 
941 /* nbd_negotiate_meta_queries
942  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
943  *
944  * Return -errno on I/O error, or 0 if option was completely handled. */
945 static int nbd_negotiate_meta_queries(NBDClient *client,
946                                       NBDExportMetaContexts *meta, Error **errp)
947 {
948     int ret;
949     char export_name[NBD_MAX_NAME_SIZE + 1];
950     NBDExportMetaContexts local_meta;
951     uint32_t nb_queries;
952     int i;
953 
954     if (!client->structured_reply) {
955         return nbd_opt_invalid(client, errp,
956                                "request option '%s' when structured reply "
957                                "is not negotiated",
958                                nbd_opt_lookup(client->opt));
959     }
960 
961     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
962         /* Only change the caller's meta on SET. */
963         meta = &local_meta;
964     }
965 
966     memset(meta, 0, sizeof(*meta));
967 
968     ret = nbd_opt_read_name(client, export_name, NULL, errp);
969     if (ret <= 0) {
970         return ret;
971     }
972 
973     meta->exp = nbd_export_find(export_name);
974     if (meta->exp == NULL) {
975         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
976                             "export '%s' not present", export_name);
977     }
978 
979     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp);
980     if (ret <= 0) {
981         return ret;
982     }
983     nb_queries = cpu_to_be32(nb_queries);
984     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
985                                      export_name, nb_queries);
986 
987     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
988         /* enable all known contexts */
989         meta->base_allocation = true;
990         meta->bitmap = !!meta->exp->export_bitmap;
991     } else {
992         for (i = 0; i < nb_queries; ++i) {
993             ret = nbd_negotiate_meta_query(client, meta, errp);
994             if (ret <= 0) {
995                 return ret;
996             }
997         }
998     }
999 
1000     if (meta->base_allocation) {
1001         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1002                                               NBD_META_ID_BASE_ALLOCATION,
1003                                               errp);
1004         if (ret < 0) {
1005             return ret;
1006         }
1007     }
1008 
1009     if (meta->bitmap) {
1010         ret = nbd_negotiate_send_meta_context(client,
1011                                               meta->exp->export_bitmap_context,
1012                                               NBD_META_ID_DIRTY_BITMAP,
1013                                               errp);
1014         if (ret < 0) {
1015             return ret;
1016         }
1017     }
1018 
1019     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1020     if (ret == 0) {
1021         meta->valid = true;
1022     }
1023 
1024     return ret;
1025 }
1026 
1027 /* nbd_negotiate_options
1028  * Process all NBD_OPT_* client option commands, during fixed newstyle
1029  * negotiation.
1030  * Return:
1031  * -errno  on error, errp is set
1032  * 0       on successful negotiation, errp is not set
1033  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1034  *         errp is not set
1035  */
1036 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags,
1037                                  Error **errp)
1038 {
1039     uint32_t flags;
1040     bool fixedNewstyle = false;
1041     bool no_zeroes = false;
1042 
1043     /* Client sends:
1044         [ 0 ..   3]   client flags
1045 
1046        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1047         [ 0 ..   7]   NBD_OPTS_MAGIC
1048         [ 8 ..  11]   NBD option
1049         [12 ..  15]   Data length
1050         ...           Rest of request
1051 
1052         [ 0 ..   7]   NBD_OPTS_MAGIC
1053         [ 8 ..  11]   Second NBD option
1054         [12 ..  15]   Data length
1055         ...           Rest of request
1056     */
1057 
1058     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1059         return -EIO;
1060     }
1061     trace_nbd_negotiate_options_flags(flags);
1062     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1063         fixedNewstyle = true;
1064         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1065     }
1066     if (flags & NBD_FLAG_C_NO_ZEROES) {
1067         no_zeroes = true;
1068         flags &= ~NBD_FLAG_C_NO_ZEROES;
1069     }
1070     if (flags != 0) {
1071         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1072         return -EINVAL;
1073     }
1074 
1075     while (1) {
1076         int ret;
1077         uint32_t option, length;
1078         uint64_t magic;
1079 
1080         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1081             return -EINVAL;
1082         }
1083         trace_nbd_negotiate_options_check_magic(magic);
1084         if (magic != NBD_OPTS_MAGIC) {
1085             error_setg(errp, "Bad magic received");
1086             return -EINVAL;
1087         }
1088 
1089         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1090             return -EINVAL;
1091         }
1092         client->opt = option;
1093 
1094         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1095             return -EINVAL;
1096         }
1097         assert(!client->optlen);
1098         client->optlen = length;
1099 
1100         if (length > NBD_MAX_BUFFER_SIZE) {
1101             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
1102                        length, NBD_MAX_BUFFER_SIZE);
1103             return -EINVAL;
1104         }
1105 
1106         trace_nbd_negotiate_options_check_option(option,
1107                                                  nbd_opt_lookup(option));
1108         if (client->tlscreds &&
1109             client->ioc == (QIOChannel *)client->sioc) {
1110             QIOChannel *tioc;
1111             if (!fixedNewstyle) {
1112                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1113                 return -EINVAL;
1114             }
1115             switch (option) {
1116             case NBD_OPT_STARTTLS:
1117                 if (length) {
1118                     /* Unconditionally drop the connection if the client
1119                      * can't start a TLS negotiation correctly */
1120                     return nbd_reject_length(client, true, errp);
1121                 }
1122                 tioc = nbd_negotiate_handle_starttls(client, errp);
1123                 if (!tioc) {
1124                     return -EIO;
1125                 }
1126                 ret = 0;
1127                 object_unref(OBJECT(client->ioc));
1128                 client->ioc = QIO_CHANNEL(tioc);
1129                 break;
1130 
1131             case NBD_OPT_EXPORT_NAME:
1132                 /* No way to return an error to client, so drop connection */
1133                 error_setg(errp, "Option 0x%x not permitted before TLS",
1134                            option);
1135                 return -EINVAL;
1136 
1137             default:
1138                 /* Let the client keep trying, unless they asked to
1139                  * quit. Always try to give an error back to the
1140                  * client; but when replying to OPT_ABORT, be aware
1141                  * that the client may hang up before receiving the
1142                  * error, in which case we are fine ignoring the
1143                  * resulting EPIPE. */
1144                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1145                                    option == NBD_OPT_ABORT ? NULL : errp,
1146                                    "Option 0x%" PRIx32
1147                                    " not permitted before TLS", option);
1148                 if (option == NBD_OPT_ABORT) {
1149                     return 1;
1150                 }
1151                 break;
1152             }
1153         } else if (fixedNewstyle) {
1154             switch (option) {
1155             case NBD_OPT_LIST:
1156                 if (length) {
1157                     ret = nbd_reject_length(client, false, errp);
1158                 } else {
1159                     ret = nbd_negotiate_handle_list(client, errp);
1160                 }
1161                 break;
1162 
1163             case NBD_OPT_ABORT:
1164                 /* NBD spec says we must try to reply before
1165                  * disconnecting, but that we must also tolerate
1166                  * guests that don't wait for our reply. */
1167                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1168                 return 1;
1169 
1170             case NBD_OPT_EXPORT_NAME:
1171                 return nbd_negotiate_handle_export_name(client,
1172                                                         myflags, no_zeroes,
1173                                                         errp);
1174 
1175             case NBD_OPT_INFO:
1176             case NBD_OPT_GO:
1177                 ret = nbd_negotiate_handle_info(client, myflags, errp);
1178                 if (ret == 1) {
1179                     assert(option == NBD_OPT_GO);
1180                     return 0;
1181                 }
1182                 break;
1183 
1184             case NBD_OPT_STARTTLS:
1185                 if (length) {
1186                     ret = nbd_reject_length(client, false, errp);
1187                 } else if (client->tlscreds) {
1188                     ret = nbd_negotiate_send_rep_err(client,
1189                                                      NBD_REP_ERR_INVALID, errp,
1190                                                      "TLS already enabled");
1191                 } else {
1192                     ret = nbd_negotiate_send_rep_err(client,
1193                                                      NBD_REP_ERR_POLICY, errp,
1194                                                      "TLS not configured");
1195                 }
1196                 break;
1197 
1198             case NBD_OPT_STRUCTURED_REPLY:
1199                 if (length) {
1200                     ret = nbd_reject_length(client, false, errp);
1201                 } else if (client->structured_reply) {
1202                     ret = nbd_negotiate_send_rep_err(
1203                         client, NBD_REP_ERR_INVALID, errp,
1204                         "structured reply already negotiated");
1205                 } else {
1206                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1207                     client->structured_reply = true;
1208                     myflags |= NBD_FLAG_SEND_DF;
1209                 }
1210                 break;
1211 
1212             case NBD_OPT_LIST_META_CONTEXT:
1213             case NBD_OPT_SET_META_CONTEXT:
1214                 ret = nbd_negotiate_meta_queries(client, &client->export_meta,
1215                                                  errp);
1216                 break;
1217 
1218             default:
1219                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1220                                    "Unsupported option %" PRIu32 " (%s)",
1221                                    option, nbd_opt_lookup(option));
1222                 break;
1223             }
1224         } else {
1225             /*
1226              * If broken new-style we should drop the connection
1227              * for anything except NBD_OPT_EXPORT_NAME
1228              */
1229             switch (option) {
1230             case NBD_OPT_EXPORT_NAME:
1231                 return nbd_negotiate_handle_export_name(client,
1232                                                         myflags, no_zeroes,
1233                                                         errp);
1234 
1235             default:
1236                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1237                            option, nbd_opt_lookup(option));
1238                 return -EINVAL;
1239             }
1240         }
1241         if (ret < 0) {
1242             return ret;
1243         }
1244     }
1245 }
1246 
1247 /* nbd_negotiate
1248  * Return:
1249  * -errno  on error, errp is set
1250  * 0       on successful negotiation, errp is not set
1251  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1252  *         errp is not set
1253  */
1254 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1255 {
1256     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1257     int ret;
1258     const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
1259                               NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA |
1260                               NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE);
1261 
1262     /* Old style negotiation header, no room for options
1263         [ 0 ..   7]   passwd       ("NBDMAGIC")
1264         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1265         [16 ..  23]   size
1266         [24 ..  27]   export flags (zero-extended)
1267         [28 .. 151]   reserved     (0)
1268 
1269        New style negotiation header, client can send options
1270         [ 0 ..   7]   passwd       ("NBDMAGIC")
1271         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1272         [16 ..  17]   server flags (0)
1273         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1274      */
1275 
1276     qio_channel_set_blocking(client->ioc, false, NULL);
1277 
1278     trace_nbd_negotiate_begin();
1279     memcpy(buf, "NBDMAGIC", 8);
1280 
1281     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1282     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1283 
1284     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1285         error_prepend(errp, "write failed: ");
1286         return -EINVAL;
1287     }
1288     ret = nbd_negotiate_options(client, myflags, errp);
1289     if (ret != 0) {
1290         if (ret < 0) {
1291             error_prepend(errp, "option negotiation failed: ");
1292         }
1293         return ret;
1294     }
1295 
1296     assert(!client->optlen);
1297     trace_nbd_negotiate_success();
1298 
1299     return 0;
1300 }
1301 
1302 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request,
1303                                Error **errp)
1304 {
1305     uint8_t buf[NBD_REQUEST_SIZE];
1306     uint32_t magic;
1307     int ret;
1308 
1309     ret = nbd_read(ioc, buf, sizeof(buf), "request", errp);
1310     if (ret < 0) {
1311         return ret;
1312     }
1313 
1314     /* Request
1315        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1316        [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1317        [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1318        [ 8 .. 15]   handle
1319        [16 .. 23]   from
1320        [24 .. 27]   len
1321      */
1322 
1323     magic = ldl_be_p(buf);
1324     request->flags  = lduw_be_p(buf + 4);
1325     request->type   = lduw_be_p(buf + 6);
1326     request->handle = ldq_be_p(buf + 8);
1327     request->from   = ldq_be_p(buf + 16);
1328     request->len    = ldl_be_p(buf + 24);
1329 
1330     trace_nbd_receive_request(magic, request->flags, request->type,
1331                               request->from, request->len);
1332 
1333     if (magic != NBD_REQUEST_MAGIC) {
1334         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
1335         return -EINVAL;
1336     }
1337     return 0;
1338 }
1339 
1340 #define MAX_NBD_REQUESTS 16
1341 
1342 void nbd_client_get(NBDClient *client)
1343 {
1344     client->refcount++;
1345 }
1346 
1347 void nbd_client_put(NBDClient *client)
1348 {
1349     if (--client->refcount == 0) {
1350         /* The last reference should be dropped by client->close,
1351          * which is called by client_close.
1352          */
1353         assert(client->closing);
1354 
1355         qio_channel_detach_aio_context(client->ioc);
1356         object_unref(OBJECT(client->sioc));
1357         object_unref(OBJECT(client->ioc));
1358         if (client->tlscreds) {
1359             object_unref(OBJECT(client->tlscreds));
1360         }
1361         g_free(client->tlsauthz);
1362         if (client->exp) {
1363             QTAILQ_REMOVE(&client->exp->clients, client, next);
1364             nbd_export_put(client->exp);
1365         }
1366         g_free(client);
1367     }
1368 }
1369 
1370 static void client_close(NBDClient *client, bool negotiated)
1371 {
1372     if (client->closing) {
1373         return;
1374     }
1375 
1376     client->closing = true;
1377 
1378     /* Force requests to finish.  They will drop their own references,
1379      * then we'll close the socket and free the NBDClient.
1380      */
1381     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1382                          NULL);
1383 
1384     /* Also tell the client, so that they release their reference.  */
1385     if (client->close_fn) {
1386         client->close_fn(client, negotiated);
1387     }
1388 }
1389 
1390 static NBDRequestData *nbd_request_get(NBDClient *client)
1391 {
1392     NBDRequestData *req;
1393 
1394     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1395     client->nb_requests++;
1396 
1397     req = g_new0(NBDRequestData, 1);
1398     nbd_client_get(client);
1399     req->client = client;
1400     return req;
1401 }
1402 
1403 static void nbd_request_put(NBDRequestData *req)
1404 {
1405     NBDClient *client = req->client;
1406 
1407     if (req->data) {
1408         qemu_vfree(req->data);
1409     }
1410     g_free(req);
1411 
1412     client->nb_requests--;
1413     nbd_client_receive_next_request(client);
1414 
1415     nbd_client_put(client);
1416 }
1417 
1418 static void blk_aio_attached(AioContext *ctx, void *opaque)
1419 {
1420     NBDExport *exp = opaque;
1421     NBDClient *client;
1422 
1423     trace_nbd_blk_aio_attached(exp->name, ctx);
1424 
1425     exp->ctx = ctx;
1426 
1427     QTAILQ_FOREACH(client, &exp->clients, next) {
1428         qio_channel_attach_aio_context(client->ioc, ctx);
1429         if (client->recv_coroutine) {
1430             aio_co_schedule(ctx, client->recv_coroutine);
1431         }
1432         if (client->send_coroutine) {
1433             aio_co_schedule(ctx, client->send_coroutine);
1434         }
1435     }
1436 }
1437 
1438 static void blk_aio_detach(void *opaque)
1439 {
1440     NBDExport *exp = opaque;
1441     NBDClient *client;
1442 
1443     trace_nbd_blk_aio_detach(exp->name, exp->ctx);
1444 
1445     QTAILQ_FOREACH(client, &exp->clients, next) {
1446         qio_channel_detach_aio_context(client->ioc);
1447     }
1448 
1449     exp->ctx = NULL;
1450 }
1451 
1452 static void nbd_eject_notifier(Notifier *n, void *data)
1453 {
1454     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1455     nbd_export_close(exp);
1456 }
1457 
1458 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
1459                           uint64_t size, const char *name, const char *desc,
1460                           const char *bitmap, uint16_t nbdflags,
1461                           void (*close)(NBDExport *), bool writethrough,
1462                           BlockBackend *on_eject_blk, Error **errp)
1463 {
1464     AioContext *ctx;
1465     BlockBackend *blk;
1466     NBDExport *exp = g_new0(NBDExport, 1);
1467     uint64_t perm;
1468     int ret;
1469 
1470     /*
1471      * NBD exports are used for non-shared storage migration.  Make sure
1472      * that BDRV_O_INACTIVE is cleared and the image is ready for write
1473      * access since the export could be available before migration handover.
1474      */
1475     assert(name);
1476     ctx = bdrv_get_aio_context(bs);
1477     aio_context_acquire(ctx);
1478     bdrv_invalidate_cache(bs, NULL);
1479     aio_context_release(ctx);
1480 
1481     /* Don't allow resize while the NBD server is running, otherwise we don't
1482      * care what happens with the node. */
1483     perm = BLK_PERM_CONSISTENT_READ;
1484     if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) {
1485         perm |= BLK_PERM_WRITE;
1486     }
1487     blk = blk_new(bdrv_get_aio_context(bs), perm,
1488                   BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
1489                   BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
1490     ret = blk_insert_bs(blk, bs, errp);
1491     if (ret < 0) {
1492         goto fail;
1493     }
1494     blk_set_enable_write_cache(blk, !writethrough);
1495     blk_set_allow_aio_context_change(blk, true);
1496 
1497     exp->refcount = 1;
1498     QTAILQ_INIT(&exp->clients);
1499     exp->blk = blk;
1500     assert(dev_offset <= INT64_MAX);
1501     exp->dev_offset = dev_offset;
1502     exp->name = g_strdup(name);
1503     exp->description = g_strdup(desc);
1504     exp->nbdflags = nbdflags;
1505     assert(size <= INT64_MAX - dev_offset);
1506     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1507 
1508     if (bitmap) {
1509         BdrvDirtyBitmap *bm = NULL;
1510 
1511         while (true) {
1512             bm = bdrv_find_dirty_bitmap(bs, bitmap);
1513             if (bm != NULL || bs->backing == NULL) {
1514                 break;
1515             }
1516 
1517             bs = bs->backing->bs;
1518         }
1519 
1520         if (bm == NULL) {
1521             error_setg(errp, "Bitmap '%s' is not found", bitmap);
1522             goto fail;
1523         }
1524 
1525         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1526             goto fail;
1527         }
1528 
1529         if ((nbdflags & NBD_FLAG_READ_ONLY) && bdrv_is_writable(bs) &&
1530             bdrv_dirty_bitmap_enabled(bm)) {
1531             error_setg(errp,
1532                        "Enabled bitmap '%s' incompatible with readonly export",
1533                        bitmap);
1534             goto fail;
1535         }
1536 
1537         bdrv_dirty_bitmap_set_busy(bm, true);
1538         exp->export_bitmap = bm;
1539         exp->export_bitmap_context = g_strdup_printf("qemu:dirty-bitmap:%s",
1540                                                      bitmap);
1541     }
1542 
1543     exp->close = close;
1544     exp->ctx = blk_get_aio_context(blk);
1545     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1546 
1547     if (on_eject_blk) {
1548         blk_ref(on_eject_blk);
1549         exp->eject_notifier_blk = on_eject_blk;
1550         exp->eject_notifier.notify = nbd_eject_notifier;
1551         blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier);
1552     }
1553     QTAILQ_INSERT_TAIL(&exports, exp, next);
1554     nbd_export_get(exp);
1555     return exp;
1556 
1557 fail:
1558     blk_unref(blk);
1559     g_free(exp->name);
1560     g_free(exp->description);
1561     g_free(exp);
1562     return NULL;
1563 }
1564 
1565 NBDExport *nbd_export_find(const char *name)
1566 {
1567     NBDExport *exp;
1568     QTAILQ_FOREACH(exp, &exports, next) {
1569         if (strcmp(name, exp->name) == 0) {
1570             return exp;
1571         }
1572     }
1573 
1574     return NULL;
1575 }
1576 
1577 void nbd_export_close(NBDExport *exp)
1578 {
1579     NBDClient *client, *next;
1580 
1581     nbd_export_get(exp);
1582     /*
1583      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1584      * close mode that stops advertising the export to new clients but
1585      * still permits existing clients to run to completion? Because of
1586      * that possibility, nbd_export_close() can be called more than
1587      * once on an export.
1588      */
1589     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1590         client_close(client, true);
1591     }
1592     if (exp->name) {
1593         nbd_export_put(exp);
1594         g_free(exp->name);
1595         exp->name = NULL;
1596         QTAILQ_REMOVE(&exports, exp, next);
1597     }
1598     g_free(exp->description);
1599     exp->description = NULL;
1600     nbd_export_put(exp);
1601 }
1602 
1603 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp)
1604 {
1605     if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) {
1606         nbd_export_close(exp);
1607         return;
1608     }
1609 
1610     assert(mode == NBD_SERVER_REMOVE_MODE_SAFE);
1611 
1612     error_setg(errp, "export '%s' still in use", exp->name);
1613     error_append_hint(errp, "Use mode='hard' to force client disconnect\n");
1614 }
1615 
1616 void nbd_export_get(NBDExport *exp)
1617 {
1618     assert(exp->refcount > 0);
1619     exp->refcount++;
1620 }
1621 
1622 void nbd_export_put(NBDExport *exp)
1623 {
1624     assert(exp->refcount > 0);
1625     if (exp->refcount == 1) {
1626         nbd_export_close(exp);
1627     }
1628 
1629     /* nbd_export_close() may theoretically reduce refcount to 0. It may happen
1630      * if someone calls nbd_export_put() on named export not through
1631      * nbd_export_set_name() when refcount is 1. So, let's assert that
1632      * it is > 0.
1633      */
1634     assert(exp->refcount > 0);
1635     if (--exp->refcount == 0) {
1636         assert(exp->name == NULL);
1637         assert(exp->description == NULL);
1638 
1639         if (exp->close) {
1640             exp->close(exp);
1641         }
1642 
1643         if (exp->blk) {
1644             if (exp->eject_notifier_blk) {
1645                 notifier_remove(&exp->eject_notifier);
1646                 blk_unref(exp->eject_notifier_blk);
1647             }
1648             blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
1649                                             blk_aio_detach, exp);
1650             blk_unref(exp->blk);
1651             exp->blk = NULL;
1652         }
1653 
1654         if (exp->export_bitmap) {
1655             bdrv_dirty_bitmap_set_busy(exp->export_bitmap, false);
1656             g_free(exp->export_bitmap_context);
1657         }
1658 
1659         g_free(exp);
1660     }
1661 }
1662 
1663 BlockBackend *nbd_export_get_blockdev(NBDExport *exp)
1664 {
1665     return exp->blk;
1666 }
1667 
1668 void nbd_export_close_all(void)
1669 {
1670     NBDExport *exp, *next;
1671 
1672     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
1673         nbd_export_close(exp);
1674     }
1675 }
1676 
1677 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1678                                         unsigned niov, Error **errp)
1679 {
1680     int ret;
1681 
1682     g_assert(qemu_in_coroutine());
1683     qemu_co_mutex_lock(&client->send_lock);
1684     client->send_coroutine = qemu_coroutine_self();
1685 
1686     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1687 
1688     client->send_coroutine = NULL;
1689     qemu_co_mutex_unlock(&client->send_lock);
1690 
1691     return ret;
1692 }
1693 
1694 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1695                                        uint64_t handle)
1696 {
1697     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1698     stl_be_p(&reply->error, error);
1699     stq_be_p(&reply->handle, handle);
1700 }
1701 
1702 static int nbd_co_send_simple_reply(NBDClient *client,
1703                                     uint64_t handle,
1704                                     uint32_t error,
1705                                     void *data,
1706                                     size_t len,
1707                                     Error **errp)
1708 {
1709     NBDSimpleReply reply;
1710     int nbd_err = system_errno_to_nbd_errno(error);
1711     struct iovec iov[] = {
1712         {.iov_base = &reply, .iov_len = sizeof(reply)},
1713         {.iov_base = data, .iov_len = len}
1714     };
1715 
1716     trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err),
1717                                    len);
1718     set_be_simple_reply(&reply, nbd_err, handle);
1719 
1720     return nbd_co_send_iov(client, iov, len ? 2 : 1, errp);
1721 }
1722 
1723 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags,
1724                                 uint16_t type, uint64_t handle, uint32_t length)
1725 {
1726     stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1727     stw_be_p(&chunk->flags, flags);
1728     stw_be_p(&chunk->type, type);
1729     stq_be_p(&chunk->handle, handle);
1730     stl_be_p(&chunk->length, length);
1731 }
1732 
1733 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client,
1734                                                     uint64_t handle,
1735                                                     Error **errp)
1736 {
1737     NBDStructuredReplyChunk chunk;
1738     struct iovec iov[] = {
1739         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1740     };
1741 
1742     trace_nbd_co_send_structured_done(handle);
1743     set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0);
1744 
1745     return nbd_co_send_iov(client, iov, 1, errp);
1746 }
1747 
1748 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client,
1749                                                     uint64_t handle,
1750                                                     uint64_t offset,
1751                                                     void *data,
1752                                                     size_t size,
1753                                                     bool final,
1754                                                     Error **errp)
1755 {
1756     NBDStructuredReadData chunk;
1757     struct iovec iov[] = {
1758         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1759         {.iov_base = data, .iov_len = size}
1760     };
1761 
1762     assert(size);
1763     trace_nbd_co_send_structured_read(handle, offset, data, size);
1764     set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1765                  NBD_REPLY_TYPE_OFFSET_DATA, handle,
1766                  sizeof(chunk) - sizeof(chunk.h) + size);
1767     stq_be_p(&chunk.offset, offset);
1768 
1769     return nbd_co_send_iov(client, iov, 2, errp);
1770 }
1771 
1772 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client,
1773                                                      uint64_t handle,
1774                                                      uint32_t error,
1775                                                      const char *msg,
1776                                                      Error **errp)
1777 {
1778     NBDStructuredError chunk;
1779     int nbd_err = system_errno_to_nbd_errno(error);
1780     struct iovec iov[] = {
1781         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1782         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
1783     };
1784 
1785     assert(nbd_err);
1786     trace_nbd_co_send_structured_error(handle, nbd_err,
1787                                        nbd_err_lookup(nbd_err), msg ? msg : "");
1788     set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle,
1789                  sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1790     stl_be_p(&chunk.error, nbd_err);
1791     stw_be_p(&chunk.message_length, iov[1].iov_len);
1792 
1793     return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp);
1794 }
1795 
1796 /* Do a sparse read and send the structured reply to the client.
1797  * Returns -errno if sending fails. bdrv_block_status_above() failure is
1798  * reported to the client, at which point this function succeeds.
1799  */
1800 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
1801                                                 uint64_t handle,
1802                                                 uint64_t offset,
1803                                                 uint8_t *data,
1804                                                 size_t size,
1805                                                 Error **errp)
1806 {
1807     int ret = 0;
1808     NBDExport *exp = client->exp;
1809     size_t progress = 0;
1810 
1811     while (progress < size) {
1812         int64_t pnum;
1813         int status = bdrv_block_status_above(blk_bs(exp->blk), NULL,
1814                                              offset + progress,
1815                                              size - progress, &pnum, NULL,
1816                                              NULL);
1817         bool final;
1818 
1819         if (status < 0) {
1820             char *msg = g_strdup_printf("unable to check for holes: %s",
1821                                         strerror(-status));
1822 
1823             ret = nbd_co_send_structured_error(client, handle, -status, msg,
1824                                                errp);
1825             g_free(msg);
1826             return ret;
1827         }
1828         assert(pnum && pnum <= size - progress);
1829         final = progress + pnum == size;
1830         if (status & BDRV_BLOCK_ZERO) {
1831             NBDStructuredReadHole chunk;
1832             struct iovec iov[] = {
1833                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1834             };
1835 
1836             trace_nbd_co_send_structured_read_hole(handle, offset + progress,
1837                                                    pnum);
1838             set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0,
1839                          NBD_REPLY_TYPE_OFFSET_HOLE,
1840                          handle, sizeof(chunk) - sizeof(chunk.h));
1841             stq_be_p(&chunk.offset, offset + progress);
1842             stl_be_p(&chunk.length, pnum);
1843             ret = nbd_co_send_iov(client, iov, 1, errp);
1844         } else {
1845             ret = blk_pread(exp->blk, offset + progress + exp->dev_offset,
1846                             data + progress, pnum);
1847             if (ret < 0) {
1848                 error_setg_errno(errp, -ret, "reading from file failed");
1849                 break;
1850             }
1851             ret = nbd_co_send_structured_read(client, handle, offset + progress,
1852                                               data + progress, pnum, final,
1853                                               errp);
1854         }
1855 
1856         if (ret < 0) {
1857             break;
1858         }
1859         progress += pnum;
1860     }
1861     return ret;
1862 }
1863 
1864 /*
1865  * Populate @extents from block status. Update @bytes to be the actual
1866  * length encoded (which may be smaller than the original), and update
1867  * @nb_extents to the number of extents used.
1868  *
1869  * Returns zero on success and -errno on bdrv_block_status_above failure.
1870  */
1871 static int blockstatus_to_extents(BlockDriverState *bs, uint64_t offset,
1872                                   uint64_t *bytes, NBDExtent *extents,
1873                                   unsigned int *nb_extents)
1874 {
1875     uint64_t remaining_bytes = *bytes;
1876     NBDExtent *extent = extents, *extents_end = extents + *nb_extents;
1877     bool first_extent = true;
1878 
1879     assert(*nb_extents);
1880     while (remaining_bytes) {
1881         uint32_t flags;
1882         int64_t num;
1883         int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes,
1884                                           &num, NULL, NULL);
1885 
1886         if (ret < 0) {
1887             return ret;
1888         }
1889 
1890         flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) |
1891                 (ret & BDRV_BLOCK_ZERO      ? NBD_STATE_ZERO : 0);
1892 
1893         if (first_extent) {
1894             extent->flags = flags;
1895             extent->length = num;
1896             first_extent = false;
1897         } else if (flags == extent->flags) {
1898             /* extend current extent */
1899             extent->length += num;
1900         } else {
1901             if (extent + 1 == extents_end) {
1902                 break;
1903             }
1904 
1905             /* start new extent */
1906             extent++;
1907             extent->flags = flags;
1908             extent->length = num;
1909         }
1910         offset += num;
1911         remaining_bytes -= num;
1912     }
1913 
1914     extents_end = extent + 1;
1915 
1916     for (extent = extents; extent < extents_end; extent++) {
1917         extent->flags = cpu_to_be32(extent->flags);
1918         extent->length = cpu_to_be32(extent->length);
1919     }
1920 
1921     *bytes -= remaining_bytes;
1922     *nb_extents = extents_end - extents;
1923 
1924     return 0;
1925 }
1926 
1927 /* nbd_co_send_extents
1928  *
1929  * @length is only for tracing purposes (and may be smaller or larger
1930  * than the client's original request). @last controls whether
1931  * NBD_REPLY_FLAG_DONE is sent. @extents should already be in
1932  * big-endian format.
1933  */
1934 static int nbd_co_send_extents(NBDClient *client, uint64_t handle,
1935                                NBDExtent *extents, unsigned int nb_extents,
1936                                uint64_t length, bool last,
1937                                uint32_t context_id, Error **errp)
1938 {
1939     NBDStructuredMeta chunk;
1940 
1941     struct iovec iov[] = {
1942         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
1943         {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])}
1944     };
1945 
1946     trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last);
1947     set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0,
1948                  NBD_REPLY_TYPE_BLOCK_STATUS,
1949                  handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len);
1950     stl_be_p(&chunk.context_id, context_id);
1951 
1952     return nbd_co_send_iov(client, iov, 2, errp);
1953 }
1954 
1955 /* Get block status from the exported device and send it to the client */
1956 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle,
1957                                     BlockDriverState *bs, uint64_t offset,
1958                                     uint32_t length, bool dont_fragment,
1959                                     bool last, uint32_t context_id,
1960                                     Error **errp)
1961 {
1962     int ret;
1963     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
1964     NBDExtent *extents = g_new(NBDExtent, nb_extents);
1965     uint64_t final_length = length;
1966 
1967     ret = blockstatus_to_extents(bs, offset, &final_length, extents,
1968                                  &nb_extents);
1969     if (ret < 0) {
1970         g_free(extents);
1971         return nbd_co_send_structured_error(
1972                 client, handle, -ret, "can't get block status", errp);
1973     }
1974 
1975     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
1976                               final_length, last, context_id, errp);
1977 
1978     g_free(extents);
1979 
1980     return ret;
1981 }
1982 
1983 /*
1984  * Populate @extents from a dirty bitmap. Unless @dont_fragment, the
1985  * final extent may exceed the original @length. Store in @length the
1986  * byte length encoded (which may be smaller or larger than the
1987  * original), and return the number of extents used.
1988  */
1989 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset,
1990                                       uint64_t *length, NBDExtent *extents,
1991                                       unsigned int nb_extents,
1992                                       bool dont_fragment)
1993 {
1994     uint64_t begin = offset, end = offset;
1995     uint64_t overall_end = offset + *length;
1996     unsigned int i = 0;
1997     BdrvDirtyBitmapIter *it;
1998     bool dirty;
1999 
2000     bdrv_dirty_bitmap_lock(bitmap);
2001 
2002     it = bdrv_dirty_iter_new(bitmap);
2003     dirty = bdrv_get_dirty_locked(NULL, bitmap, offset);
2004 
2005     assert(begin < overall_end && nb_extents);
2006     while (begin < overall_end && i < nb_extents) {
2007         bool next_dirty = !dirty;
2008 
2009         if (dirty) {
2010             end = bdrv_dirty_bitmap_next_zero(bitmap, begin, UINT64_MAX);
2011         } else {
2012             bdrv_set_dirty_iter(it, begin);
2013             end = bdrv_dirty_iter_next(it);
2014         }
2015         if (end == -1 || end - begin > UINT32_MAX) {
2016             /* Cap to an aligned value < 4G beyond begin. */
2017             end = MIN(bdrv_dirty_bitmap_size(bitmap),
2018                       begin + UINT32_MAX + 1 -
2019                       bdrv_dirty_bitmap_granularity(bitmap));
2020             next_dirty = dirty;
2021         }
2022         if (dont_fragment && end > overall_end) {
2023             end = overall_end;
2024         }
2025 
2026         extents[i].length = cpu_to_be32(end - begin);
2027         extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0);
2028         i++;
2029         begin = end;
2030         dirty = next_dirty;
2031     }
2032 
2033     bdrv_dirty_iter_free(it);
2034 
2035     bdrv_dirty_bitmap_unlock(bitmap);
2036 
2037     assert(offset < end);
2038     *length = end - offset;
2039     return i;
2040 }
2041 
2042 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle,
2043                               BdrvDirtyBitmap *bitmap, uint64_t offset,
2044                               uint32_t length, bool dont_fragment, bool last,
2045                               uint32_t context_id, Error **errp)
2046 {
2047     int ret;
2048     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS;
2049     NBDExtent *extents = g_new(NBDExtent, nb_extents);
2050     uint64_t final_length = length;
2051 
2052     nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents,
2053                                    nb_extents, dont_fragment);
2054 
2055     ret = nbd_co_send_extents(client, handle, extents, nb_extents,
2056                               final_length, last, context_id, errp);
2057 
2058     g_free(extents);
2059 
2060     return ret;
2061 }
2062 
2063 /* nbd_co_receive_request
2064  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2065  * connection right away, and any other negative value to report an error to
2066  * the client (although the caller may still need to disconnect after reporting
2067  * the error).
2068  */
2069 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request,
2070                                   Error **errp)
2071 {
2072     NBDClient *client = req->client;
2073     int valid_flags;
2074 
2075     g_assert(qemu_in_coroutine());
2076     assert(client->recv_coroutine == qemu_coroutine_self());
2077     if (nbd_receive_request(client->ioc, request, errp) < 0) {
2078         return -EIO;
2079     }
2080 
2081     trace_nbd_co_receive_request_decode_type(request->handle, request->type,
2082                                              nbd_cmd_lookup(request->type));
2083 
2084     if (request->type != NBD_CMD_WRITE) {
2085         /* No payload, we are ready to read the next request.  */
2086         req->complete = true;
2087     }
2088 
2089     if (request->type == NBD_CMD_DISC) {
2090         /* Special case: we're going to disconnect without a reply,
2091          * whether or not flags, from, or len are bogus */
2092         return -EIO;
2093     }
2094 
2095     if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE ||
2096         request->type == NBD_CMD_CACHE)
2097     {
2098         if (request->len > NBD_MAX_BUFFER_SIZE) {
2099             error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)",
2100                        request->len, NBD_MAX_BUFFER_SIZE);
2101             return -EINVAL;
2102         }
2103 
2104         req->data = blk_try_blockalign(client->exp->blk, request->len);
2105         if (req->data == NULL) {
2106             error_setg(errp, "No memory");
2107             return -ENOMEM;
2108         }
2109     }
2110     if (request->type == NBD_CMD_WRITE) {
2111         if (nbd_read(client->ioc, req->data, request->len, "CMD_WRITE data",
2112                      errp) < 0)
2113         {
2114             return -EIO;
2115         }
2116         req->complete = true;
2117 
2118         trace_nbd_co_receive_request_payload_received(request->handle,
2119                                                       request->len);
2120     }
2121 
2122     /* Sanity checks. */
2123     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY &&
2124         (request->type == NBD_CMD_WRITE ||
2125          request->type == NBD_CMD_WRITE_ZEROES ||
2126          request->type == NBD_CMD_TRIM)) {
2127         error_setg(errp, "Export is read-only");
2128         return -EROFS;
2129     }
2130     if (request->from > client->exp->size ||
2131         request->len > client->exp->size - request->from) {
2132         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32
2133                    ", Size: %" PRIu64, request->from, request->len,
2134                    client->exp->size);
2135         return (request->type == NBD_CMD_WRITE ||
2136                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2137     }
2138     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2139                                                 client->check_align)) {
2140         /*
2141          * The block layer gracefully handles unaligned requests, but
2142          * it's still worth tracing client non-compliance
2143          */
2144         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2145                                               request->from,
2146                                               request->len,
2147                                               client->check_align);
2148     }
2149     valid_flags = NBD_CMD_FLAG_FUA;
2150     if (request->type == NBD_CMD_READ && client->structured_reply) {
2151         valid_flags |= NBD_CMD_FLAG_DF;
2152     } else if (request->type == NBD_CMD_WRITE_ZEROES) {
2153         valid_flags |= NBD_CMD_FLAG_NO_HOLE;
2154     } else if (request->type == NBD_CMD_BLOCK_STATUS) {
2155         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2156     }
2157     if (request->flags & ~valid_flags) {
2158         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2159                    nbd_cmd_lookup(request->type), request->flags);
2160         return -EINVAL;
2161     }
2162 
2163     return 0;
2164 }
2165 
2166 /* Send simple reply without a payload, or a structured error
2167  * @error_msg is ignored if @ret >= 0
2168  * Returns 0 if connection is still live, -errno on failure to talk to client
2169  */
2170 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2171                                                uint64_t handle,
2172                                                int ret,
2173                                                const char *error_msg,
2174                                                Error **errp)
2175 {
2176     if (client->structured_reply && ret < 0) {
2177         return nbd_co_send_structured_error(client, handle, -ret, error_msg,
2178                                             errp);
2179     } else {
2180         return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0,
2181                                         NULL, 0, errp);
2182     }
2183 }
2184 
2185 /* Handle NBD_CMD_READ request.
2186  * Return -errno if sending fails. Other errors are reported directly to the
2187  * client as an error reply. */
2188 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2189                                         uint8_t *data, Error **errp)
2190 {
2191     int ret;
2192     NBDExport *exp = client->exp;
2193 
2194     assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE);
2195 
2196     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2197     if (request->flags & NBD_CMD_FLAG_FUA) {
2198         ret = blk_co_flush(exp->blk);
2199         if (ret < 0) {
2200             return nbd_send_generic_reply(client, request->handle, ret,
2201                                           "flush failed", errp);
2202         }
2203     }
2204 
2205     if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) &&
2206         request->len && request->type != NBD_CMD_CACHE)
2207     {
2208         return nbd_co_send_sparse_read(client, request->handle, request->from,
2209                                        data, request->len, errp);
2210     }
2211 
2212     ret = blk_pread(exp->blk, request->from + exp->dev_offset, data,
2213                     request->len);
2214     if (ret < 0 || request->type == NBD_CMD_CACHE) {
2215         return nbd_send_generic_reply(client, request->handle, ret,
2216                                       "reading from file failed", errp);
2217     }
2218 
2219     if (client->structured_reply) {
2220         if (request->len) {
2221             return nbd_co_send_structured_read(client, request->handle,
2222                                                request->from, data,
2223                                                request->len, true, errp);
2224         } else {
2225             return nbd_co_send_structured_done(client, request->handle, errp);
2226         }
2227     } else {
2228         return nbd_co_send_simple_reply(client, request->handle, 0,
2229                                         data, request->len, errp);
2230     }
2231 }
2232 
2233 /* Handle NBD request.
2234  * Return -errno if sending fails. Other errors are reported directly to the
2235  * client as an error reply. */
2236 static coroutine_fn int nbd_handle_request(NBDClient *client,
2237                                            NBDRequest *request,
2238                                            uint8_t *data, Error **errp)
2239 {
2240     int ret;
2241     int flags;
2242     NBDExport *exp = client->exp;
2243     char *msg;
2244 
2245     switch (request->type) {
2246     case NBD_CMD_READ:
2247     case NBD_CMD_CACHE:
2248         return nbd_do_cmd_read(client, request, data, errp);
2249 
2250     case NBD_CMD_WRITE:
2251         flags = 0;
2252         if (request->flags & NBD_CMD_FLAG_FUA) {
2253             flags |= BDRV_REQ_FUA;
2254         }
2255         ret = blk_pwrite(exp->blk, request->from + exp->dev_offset,
2256                          data, request->len, flags);
2257         return nbd_send_generic_reply(client, request->handle, ret,
2258                                       "writing to file failed", errp);
2259 
2260     case NBD_CMD_WRITE_ZEROES:
2261         flags = 0;
2262         if (request->flags & NBD_CMD_FLAG_FUA) {
2263             flags |= BDRV_REQ_FUA;
2264         }
2265         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2266             flags |= BDRV_REQ_MAY_UNMAP;
2267         }
2268         ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset,
2269                                 request->len, flags);
2270         return nbd_send_generic_reply(client, request->handle, ret,
2271                                       "writing to file failed", errp);
2272 
2273     case NBD_CMD_DISC:
2274         /* unreachable, thanks to special case in nbd_co_receive_request() */
2275         abort();
2276 
2277     case NBD_CMD_FLUSH:
2278         ret = blk_co_flush(exp->blk);
2279         return nbd_send_generic_reply(client, request->handle, ret,
2280                                       "flush failed", errp);
2281 
2282     case NBD_CMD_TRIM:
2283         ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset,
2284                               request->len);
2285         if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) {
2286             ret = blk_co_flush(exp->blk);
2287         }
2288         return nbd_send_generic_reply(client, request->handle, ret,
2289                                       "discard failed", errp);
2290 
2291     case NBD_CMD_BLOCK_STATUS:
2292         if (!request->len) {
2293             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2294                                           "need non-zero length", errp);
2295         }
2296         if (client->export_meta.valid &&
2297             (client->export_meta.base_allocation ||
2298              client->export_meta.bitmap))
2299         {
2300             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2301 
2302             if (client->export_meta.base_allocation) {
2303                 ret = nbd_co_send_block_status(client, request->handle,
2304                                                blk_bs(exp->blk), request->from,
2305                                                request->len, dont_fragment,
2306                                                !client->export_meta.bitmap,
2307                                                NBD_META_ID_BASE_ALLOCATION,
2308                                                errp);
2309                 if (ret < 0) {
2310                     return ret;
2311                 }
2312             }
2313 
2314             if (client->export_meta.bitmap) {
2315                 ret = nbd_co_send_bitmap(client, request->handle,
2316                                          client->exp->export_bitmap,
2317                                          request->from, request->len,
2318                                          dont_fragment,
2319                                          true, NBD_META_ID_DIRTY_BITMAP, errp);
2320                 if (ret < 0) {
2321                     return ret;
2322                 }
2323             }
2324 
2325             return ret;
2326         } else {
2327             return nbd_send_generic_reply(client, request->handle, -EINVAL,
2328                                           "CMD_BLOCK_STATUS not negotiated",
2329                                           errp);
2330         }
2331 
2332     default:
2333         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2334                               request->type);
2335         ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg,
2336                                      errp);
2337         g_free(msg);
2338         return ret;
2339     }
2340 }
2341 
2342 /* Owns a reference to the NBDClient passed as opaque.  */
2343 static coroutine_fn void nbd_trip(void *opaque)
2344 {
2345     NBDClient *client = opaque;
2346     NBDRequestData *req;
2347     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2348     int ret;
2349     Error *local_err = NULL;
2350 
2351     trace_nbd_trip();
2352     if (client->closing) {
2353         nbd_client_put(client);
2354         return;
2355     }
2356 
2357     req = nbd_request_get(client);
2358     ret = nbd_co_receive_request(req, &request, &local_err);
2359     client->recv_coroutine = NULL;
2360 
2361     if (client->closing) {
2362         /*
2363          * The client may be closed when we are blocked in
2364          * nbd_co_receive_request()
2365          */
2366         goto done;
2367     }
2368 
2369     nbd_client_receive_next_request(client);
2370     if (ret == -EIO) {
2371         goto disconnect;
2372     }
2373 
2374     if (ret < 0) {
2375         /* It wans't -EIO, so, according to nbd_co_receive_request()
2376          * semantics, we should return the error to the client. */
2377         Error *export_err = local_err;
2378 
2379         local_err = NULL;
2380         ret = nbd_send_generic_reply(client, request.handle, -EINVAL,
2381                                      error_get_pretty(export_err), &local_err);
2382         error_free(export_err);
2383     } else {
2384         ret = nbd_handle_request(client, &request, req->data, &local_err);
2385     }
2386     if (ret < 0) {
2387         error_prepend(&local_err, "Failed to send reply: ");
2388         goto disconnect;
2389     }
2390 
2391     /* We must disconnect after NBD_CMD_WRITE if we did not
2392      * read the payload.
2393      */
2394     if (!req->complete) {
2395         error_setg(&local_err, "Request handling failed in intermediate state");
2396         goto disconnect;
2397     }
2398 
2399 done:
2400     nbd_request_put(req);
2401     nbd_client_put(client);
2402     return;
2403 
2404 disconnect:
2405     if (local_err) {
2406         error_reportf_err(local_err, "Disconnect client, due to: ");
2407     }
2408     nbd_request_put(req);
2409     client_close(client, true);
2410     nbd_client_put(client);
2411 }
2412 
2413 static void nbd_client_receive_next_request(NBDClient *client)
2414 {
2415     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
2416         nbd_client_get(client);
2417         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
2418         aio_co_schedule(client->exp->ctx, client->recv_coroutine);
2419     }
2420 }
2421 
2422 static coroutine_fn void nbd_co_client_start(void *opaque)
2423 {
2424     NBDClient *client = opaque;
2425     Error *local_err = NULL;
2426 
2427     qemu_co_mutex_init(&client->send_lock);
2428 
2429     if (nbd_negotiate(client, &local_err)) {
2430         if (local_err) {
2431             error_report_err(local_err);
2432         }
2433         client_close(client, false);
2434         return;
2435     }
2436 
2437     nbd_client_receive_next_request(client);
2438 }
2439 
2440 /*
2441  * Create a new client listener using the given channel @sioc.
2442  * Begin servicing it in a coroutine.  When the connection closes, call
2443  * @close_fn with an indication of whether the client completed negotiation.
2444  */
2445 void nbd_client_new(QIOChannelSocket *sioc,
2446                     QCryptoTLSCreds *tlscreds,
2447                     const char *tlsauthz,
2448                     void (*close_fn)(NBDClient *, bool))
2449 {
2450     NBDClient *client;
2451     Coroutine *co;
2452 
2453     client = g_new0(NBDClient, 1);
2454     client->refcount = 1;
2455     client->tlscreds = tlscreds;
2456     if (tlscreds) {
2457         object_ref(OBJECT(client->tlscreds));
2458     }
2459     client->tlsauthz = g_strdup(tlsauthz);
2460     client->sioc = sioc;
2461     object_ref(OBJECT(client->sioc));
2462     client->ioc = QIO_CHANNEL(sioc);
2463     object_ref(OBJECT(client->ioc));
2464     client->close_fn = close_fn;
2465 
2466     co = qemu_coroutine_create(nbd_co_client_start, client);
2467     qemu_coroutine_enter(co);
2468 }
2469