xref: /openbmc/qemu/nbd/server.c (revision 259ebed4)
1 /*
2  *  Copyright Red Hat
3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
4  *
5  *  Network Block Device Server Side
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; under version 2 of the License.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 
22 #include "block/block_int.h"
23 #include "block/export.h"
24 #include "block/dirty-bitmap.h"
25 #include "qapi/error.h"
26 #include "qemu/queue.h"
27 #include "trace.h"
28 #include "nbd-internal.h"
29 #include "qemu/units.h"
30 #include "qemu/memalign.h"
31 
32 #define NBD_META_ID_BASE_ALLOCATION 0
33 #define NBD_META_ID_ALLOCATION_DEPTH 1
34 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */
35 #define NBD_META_ID_DIRTY_BITMAP 2
36 
37 /*
38  * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical
39  * constant. If an increase is needed, note that the NBD protocol
40  * recommends no larger than 32 mb, so that the client won't consider
41  * the reply as a denial of service attack.
42  */
43 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8)
44 
45 static int system_errno_to_nbd_errno(int err)
46 {
47     switch (err) {
48     case 0:
49         return NBD_SUCCESS;
50     case EPERM:
51     case EROFS:
52         return NBD_EPERM;
53     case EIO:
54         return NBD_EIO;
55     case ENOMEM:
56         return NBD_ENOMEM;
57 #ifdef EDQUOT
58     case EDQUOT:
59 #endif
60     case EFBIG:
61     case ENOSPC:
62         return NBD_ENOSPC;
63     case EOVERFLOW:
64         return NBD_EOVERFLOW;
65     case ENOTSUP:
66 #if ENOTSUP != EOPNOTSUPP
67     case EOPNOTSUPP:
68 #endif
69         return NBD_ENOTSUP;
70     case ESHUTDOWN:
71         return NBD_ESHUTDOWN;
72     case EINVAL:
73     default:
74         return NBD_EINVAL;
75     }
76 }
77 
78 /* Definitions for opaque data types */
79 
80 typedef struct NBDRequestData NBDRequestData;
81 
82 struct NBDRequestData {
83     NBDClient *client;
84     uint8_t *data;
85     bool complete;
86 };
87 
88 struct NBDExport {
89     BlockExport common;
90 
91     char *name;
92     char *description;
93     uint64_t size;
94     uint16_t nbdflags;
95     QTAILQ_HEAD(, NBDClient) clients;
96     QTAILQ_ENTRY(NBDExport) next;
97 
98     BlockBackend *eject_notifier_blk;
99     Notifier eject_notifier;
100 
101     bool allocation_depth;
102     BdrvDirtyBitmap **export_bitmaps;
103     size_t nr_export_bitmaps;
104 };
105 
106 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
107 
108 /*
109  * NBDMetaContexts represents a list of meta contexts in use,
110  * as selected by NBD_OPT_SET_META_CONTEXT. Also used for
111  * NBD_OPT_LIST_META_CONTEXT.
112  */
113 struct NBDMetaContexts {
114     const NBDExport *exp; /* associated export */
115     size_t count; /* number of negotiated contexts */
116     bool base_allocation; /* export base:allocation context (block status) */
117     bool allocation_depth; /* export qemu:allocation-depth */
118     bool *bitmaps; /*
119                     * export qemu:dirty-bitmap:<export bitmap name>,
120                     * sized by exp->nr_export_bitmaps
121                     */
122 };
123 
124 struct NBDClient {
125     int refcount;
126     void (*close_fn)(NBDClient *client, bool negotiated);
127 
128     NBDExport *exp;
129     QCryptoTLSCreds *tlscreds;
130     char *tlsauthz;
131     QIOChannelSocket *sioc; /* The underlying data channel */
132     QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */
133 
134     Coroutine *recv_coroutine;
135 
136     CoMutex send_lock;
137     Coroutine *send_coroutine;
138 
139     bool read_yielding;
140     bool quiescing;
141 
142     QTAILQ_ENTRY(NBDClient) next;
143     int nb_requests;
144     bool closing;
145 
146     uint32_t check_align; /* If non-zero, check for aligned client requests */
147 
148     NBDMode mode;
149     NBDMetaContexts contexts; /* Negotiated meta contexts */
150 
151     uint32_t opt; /* Current option being negotiated */
152     uint32_t optlen; /* remaining length of data in ioc for the option being
153                         negotiated now */
154 };
155 
156 static void nbd_client_receive_next_request(NBDClient *client);
157 
158 /* Basic flow for negotiation
159 
160    Server         Client
161    Negotiate
162 
163    or
164 
165    Server         Client
166    Negotiate #1
167                   Option
168    Negotiate #2
169 
170    ----
171 
172    followed by
173 
174    Server         Client
175                   Request
176    Response
177                   Request
178    Response
179                   ...
180    ...
181                   Request (type == 2)
182 
183 */
184 
185 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option,
186                                      uint32_t type, uint32_t length)
187 {
188     stq_be_p(&rep->magic, NBD_REP_MAGIC);
189     stl_be_p(&rep->option, option);
190     stl_be_p(&rep->type, type);
191     stl_be_p(&rep->length, length);
192 }
193 
194 /* Send a reply header, including length, but no payload.
195  * Return -errno on error, 0 on success. */
196 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type,
197                                       uint32_t len, Error **errp)
198 {
199     NBDOptionReply rep;
200 
201     trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt),
202                                      type, nbd_rep_lookup(type), len);
203 
204     assert(len < NBD_MAX_BUFFER_SIZE);
205 
206     set_be_option_rep(&rep, client->opt, type, len);
207     return nbd_write(client->ioc, &rep, sizeof(rep), errp);
208 }
209 
210 /* Send a reply header with default 0 length.
211  * Return -errno on error, 0 on success. */
212 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type,
213                                   Error **errp)
214 {
215     return nbd_negotiate_send_rep_len(client, type, 0, errp);
216 }
217 
218 /* Send an error reply.
219  * Return -errno on error, 0 on success. */
220 static int G_GNUC_PRINTF(4, 0)
221 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type,
222                             Error **errp, const char *fmt, va_list va)
223 {
224     ERRP_GUARD();
225     g_autofree char *msg = NULL;
226     int ret;
227     size_t len;
228 
229     msg = g_strdup_vprintf(fmt, va);
230     len = strlen(msg);
231     assert(len < NBD_MAX_STRING_SIZE);
232     trace_nbd_negotiate_send_rep_err(msg);
233     ret = nbd_negotiate_send_rep_len(client, type, len, errp);
234     if (ret < 0) {
235         return ret;
236     }
237     if (nbd_write(client->ioc, msg, len, errp) < 0) {
238         error_prepend(errp, "write failed (error message): ");
239         return -EIO;
240     }
241 
242     return 0;
243 }
244 
245 /*
246  * Return a malloc'd copy of @name suitable for use in an error reply.
247  */
248 static char *
249 nbd_sanitize_name(const char *name)
250 {
251     if (strnlen(name, 80) < 80) {
252         return g_strdup(name);
253     }
254     /* XXX Should we also try to sanitize any control characters? */
255     return g_strdup_printf("%.80s...", name);
256 }
257 
258 /* Send an error reply.
259  * Return -errno on error, 0 on success. */
260 static int G_GNUC_PRINTF(4, 5)
261 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type,
262                            Error **errp, const char *fmt, ...)
263 {
264     va_list va;
265     int ret;
266 
267     va_start(va, fmt);
268     ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
269     va_end(va);
270     return ret;
271 }
272 
273 /* Drop remainder of the current option, and send a reply with the
274  * given error type and message. Return -errno on read or write
275  * failure; or 0 if connection is still live. */
276 static int G_GNUC_PRINTF(4, 0)
277 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp,
278               const char *fmt, va_list va)
279 {
280     int ret = nbd_drop(client->ioc, client->optlen, errp);
281 
282     client->optlen = 0;
283     if (!ret) {
284         ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va);
285     }
286     return ret;
287 }
288 
289 static int G_GNUC_PRINTF(4, 5)
290 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp,
291              const char *fmt, ...)
292 {
293     int ret;
294     va_list va;
295 
296     va_start(va, fmt);
297     ret = nbd_opt_vdrop(client, type, errp, fmt, va);
298     va_end(va);
299 
300     return ret;
301 }
302 
303 static int G_GNUC_PRINTF(3, 4)
304 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...)
305 {
306     int ret;
307     va_list va;
308 
309     va_start(va, fmt);
310     ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va);
311     va_end(va);
312 
313     return ret;
314 }
315 
316 /* Read size bytes from the unparsed payload of the current option.
317  * If @check_nul, require that no NUL bytes appear in buffer.
318  * Return -errno on I/O error, 0 if option was completely handled by
319  * sending a reply about inconsistent lengths, or 1 on success. */
320 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size,
321                         bool check_nul, Error **errp)
322 {
323     if (size > client->optlen) {
324         return nbd_opt_invalid(client, errp,
325                                "Inconsistent lengths in option %s",
326                                nbd_opt_lookup(client->opt));
327     }
328     client->optlen -= size;
329     if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) {
330         return -EIO;
331     }
332 
333     if (check_nul && strnlen(buffer, size) != size) {
334         return nbd_opt_invalid(client, errp,
335                                "Unexpected embedded NUL in option %s",
336                                nbd_opt_lookup(client->opt));
337     }
338     return 1;
339 }
340 
341 /* Drop size bytes from the unparsed payload of the current option.
342  * Return -errno on I/O error, 0 if option was completely handled by
343  * sending a reply about inconsistent lengths, or 1 on success. */
344 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp)
345 {
346     if (size > client->optlen) {
347         return nbd_opt_invalid(client, errp,
348                                "Inconsistent lengths in option %s",
349                                nbd_opt_lookup(client->opt));
350     }
351     client->optlen -= size;
352     return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1;
353 }
354 
355 /* nbd_opt_read_name
356  *
357  * Read a string with the format:
358  *   uint32_t len     (<= NBD_MAX_STRING_SIZE)
359  *   len bytes string (not 0-terminated)
360  *
361  * On success, @name will be allocated.
362  * If @length is non-null, it will be set to the actual string length.
363  *
364  * Return -errno on I/O error, 0 if option was completely handled by
365  * sending a reply about inconsistent lengths, or 1 on success.
366  */
367 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length,
368                              Error **errp)
369 {
370     int ret;
371     uint32_t len;
372     g_autofree char *local_name = NULL;
373 
374     *name = NULL;
375     ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
376     if (ret <= 0) {
377         return ret;
378     }
379     len = cpu_to_be32(len);
380 
381     if (len > NBD_MAX_STRING_SIZE) {
382         return nbd_opt_invalid(client, errp,
383                                "Invalid name length: %" PRIu32, len);
384     }
385 
386     local_name = g_malloc(len + 1);
387     ret = nbd_opt_read(client, local_name, len, true, errp);
388     if (ret <= 0) {
389         return ret;
390     }
391     local_name[len] = '\0';
392 
393     if (length) {
394         *length = len;
395     }
396     *name = g_steal_pointer(&local_name);
397 
398     return 1;
399 }
400 
401 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload.
402  * Return -errno on error, 0 on success. */
403 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp,
404                                        Error **errp)
405 {
406     ERRP_GUARD();
407     size_t name_len, desc_len;
408     uint32_t len;
409     const char *name = exp->name ? exp->name : "";
410     const char *desc = exp->description ? exp->description : "";
411     QIOChannel *ioc = client->ioc;
412     int ret;
413 
414     trace_nbd_negotiate_send_rep_list(name, desc);
415     name_len = strlen(name);
416     desc_len = strlen(desc);
417     assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE);
418     len = name_len + desc_len + sizeof(len);
419     ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp);
420     if (ret < 0) {
421         return ret;
422     }
423 
424     len = cpu_to_be32(name_len);
425     if (nbd_write(ioc, &len, sizeof(len), errp) < 0) {
426         error_prepend(errp, "write failed (name length): ");
427         return -EINVAL;
428     }
429 
430     if (nbd_write(ioc, name, name_len, errp) < 0) {
431         error_prepend(errp, "write failed (name buffer): ");
432         return -EINVAL;
433     }
434 
435     if (nbd_write(ioc, desc, desc_len, errp) < 0) {
436         error_prepend(errp, "write failed (description buffer): ");
437         return -EINVAL;
438     }
439 
440     return 0;
441 }
442 
443 /* Process the NBD_OPT_LIST command, with a potential series of replies.
444  * Return -errno on error, 0 on success. */
445 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp)
446 {
447     NBDExport *exp;
448     assert(client->opt == NBD_OPT_LIST);
449 
450     /* For each export, send a NBD_REP_SERVER reply. */
451     QTAILQ_FOREACH(exp, &exports, next) {
452         if (nbd_negotiate_send_rep_list(client, exp, errp)) {
453             return -EINVAL;
454         }
455     }
456     /* Finish with a NBD_REP_ACK. */
457     return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
458 }
459 
460 static void nbd_check_meta_export(NBDClient *client, NBDExport *exp)
461 {
462     if (exp != client->contexts.exp) {
463         client->contexts.count = 0;
464     }
465 }
466 
467 /* Send a reply to NBD_OPT_EXPORT_NAME.
468  * Return -errno on error, 0 on success. */
469 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes,
470                                             Error **errp)
471 {
472     ERRP_GUARD();
473     g_autofree char *name = NULL;
474     char buf[NBD_REPLY_EXPORT_NAME_SIZE] = "";
475     size_t len;
476     int ret;
477     uint16_t myflags;
478 
479     /* Client sends:
480         [20 ..  xx]   export name (length bytes)
481        Server replies:
482         [ 0 ..   7]   size
483         [ 8 ..   9]   export flags
484         [10 .. 133]   reserved     (0) [unless no_zeroes]
485      */
486     trace_nbd_negotiate_handle_export_name();
487     if (client->mode >= NBD_MODE_EXTENDED) {
488         error_setg(errp, "Extended headers already negotiated");
489         return -EINVAL;
490     }
491     if (client->optlen > NBD_MAX_STRING_SIZE) {
492         error_setg(errp, "Bad length received");
493         return -EINVAL;
494     }
495     name = g_malloc(client->optlen + 1);
496     if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) {
497         return -EIO;
498     }
499     name[client->optlen] = '\0';
500     client->optlen = 0;
501 
502     trace_nbd_negotiate_handle_export_name_request(name);
503 
504     client->exp = nbd_export_find(name);
505     if (!client->exp) {
506         error_setg(errp, "export not found");
507         return -EINVAL;
508     }
509     nbd_check_meta_export(client, client->exp);
510 
511     myflags = client->exp->nbdflags;
512     if (client->mode >= NBD_MODE_STRUCTURED) {
513         myflags |= NBD_FLAG_SEND_DF;
514     }
515     if (client->mode >= NBD_MODE_EXTENDED && client->contexts.count) {
516         myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD;
517     }
518     trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags);
519     stq_be_p(buf, client->exp->size);
520     stw_be_p(buf + 8, myflags);
521     len = no_zeroes ? 10 : sizeof(buf);
522     ret = nbd_write(client->ioc, buf, len, errp);
523     if (ret < 0) {
524         error_prepend(errp, "write failed: ");
525         return ret;
526     }
527 
528     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
529     blk_exp_ref(&client->exp->common);
530 
531     return 0;
532 }
533 
534 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes.
535  * The buffer does NOT include the info type prefix.
536  * Return -errno on error, 0 if ready to send more. */
537 static int nbd_negotiate_send_info(NBDClient *client,
538                                    uint16_t info, uint32_t length, void *buf,
539                                    Error **errp)
540 {
541     int rc;
542 
543     trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length);
544     rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO,
545                                     sizeof(info) + length, errp);
546     if (rc < 0) {
547         return rc;
548     }
549     info = cpu_to_be16(info);
550     if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) {
551         return -EIO;
552     }
553     if (nbd_write(client->ioc, buf, length, errp) < 0) {
554         return -EIO;
555     }
556     return 0;
557 }
558 
559 /* nbd_reject_length: Handle any unexpected payload.
560  * @fatal requests that we quit talking to the client, even if we are able
561  * to successfully send an error reply.
562  * Return:
563  * -errno  transmission error occurred or @fatal was requested, errp is set
564  * 0       error message successfully sent to client, errp is not set
565  */
566 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp)
567 {
568     int ret;
569 
570     assert(client->optlen);
571     ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length",
572                           nbd_opt_lookup(client->opt));
573     if (fatal && !ret) {
574         error_setg(errp, "option '%s' has unexpected length",
575                    nbd_opt_lookup(client->opt));
576         return -EINVAL;
577     }
578     return ret;
579 }
580 
581 /* Handle NBD_OPT_INFO and NBD_OPT_GO.
582  * Return -errno on error, 0 if ready for next option, and 1 to move
583  * into transmission phase.  */
584 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp)
585 {
586     int rc;
587     g_autofree char *name = NULL;
588     NBDExport *exp;
589     uint16_t requests;
590     uint16_t request;
591     uint32_t namelen = 0;
592     bool sendname = false;
593     bool blocksize = false;
594     uint32_t sizes[3];
595     char buf[sizeof(uint64_t) + sizeof(uint16_t)];
596     uint32_t check_align = 0;
597     uint16_t myflags;
598 
599     /* Client sends:
600         4 bytes: L, name length (can be 0)
601         L bytes: export name
602         2 bytes: N, number of requests (can be 0)
603         N * 2 bytes: N requests
604     */
605     rc = nbd_opt_read_name(client, &name, &namelen, errp);
606     if (rc <= 0) {
607         return rc;
608     }
609     trace_nbd_negotiate_handle_export_name_request(name);
610 
611     rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp);
612     if (rc <= 0) {
613         return rc;
614     }
615     requests = be16_to_cpu(requests);
616     trace_nbd_negotiate_handle_info_requests(requests);
617     while (requests--) {
618         rc = nbd_opt_read(client, &request, sizeof(request), false, errp);
619         if (rc <= 0) {
620             return rc;
621         }
622         request = be16_to_cpu(request);
623         trace_nbd_negotiate_handle_info_request(request,
624                                                 nbd_info_lookup(request));
625         /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE;
626          * everything else is either a request we don't know or
627          * something we send regardless of request */
628         switch (request) {
629         case NBD_INFO_NAME:
630             sendname = true;
631             break;
632         case NBD_INFO_BLOCK_SIZE:
633             blocksize = true;
634             break;
635         }
636     }
637     if (client->optlen) {
638         return nbd_reject_length(client, false, errp);
639     }
640 
641     exp = nbd_export_find(name);
642     if (!exp) {
643         g_autofree char *sane_name = nbd_sanitize_name(name);
644 
645         return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN,
646                                           errp, "export '%s' not present",
647                                           sane_name);
648     }
649     if (client->opt == NBD_OPT_GO) {
650         nbd_check_meta_export(client, exp);
651     }
652 
653     /* Don't bother sending NBD_INFO_NAME unless client requested it */
654     if (sendname) {
655         rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
656                                      errp);
657         if (rc < 0) {
658             return rc;
659         }
660     }
661 
662     /* Send NBD_INFO_DESCRIPTION only if available, regardless of
663      * client request */
664     if (exp->description) {
665         size_t len = strlen(exp->description);
666 
667         assert(len <= NBD_MAX_STRING_SIZE);
668         rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION,
669                                      len, exp->description, errp);
670         if (rc < 0) {
671             return rc;
672         }
673     }
674 
675     /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size
676      * according to whether the client requested it, and according to
677      * whether this is OPT_INFO or OPT_GO. */
678     /* minimum - 1 for back-compat, or actual if client will obey it. */
679     if (client->opt == NBD_OPT_INFO || blocksize) {
680         check_align = sizes[0] = blk_get_request_alignment(exp->common.blk);
681     } else {
682         sizes[0] = 1;
683     }
684     assert(sizes[0] <= NBD_MAX_BUFFER_SIZE);
685     /* preferred - Hard-code to 4096 for now.
686      * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */
687     sizes[1] = MAX(4096, sizes[0]);
688     /* maximum - At most 32M, but smaller as appropriate. */
689     sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE);
690     trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]);
691     sizes[0] = cpu_to_be32(sizes[0]);
692     sizes[1] = cpu_to_be32(sizes[1]);
693     sizes[2] = cpu_to_be32(sizes[2]);
694     rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE,
695                                  sizeof(sizes), sizes, errp);
696     if (rc < 0) {
697         return rc;
698     }
699 
700     /* Send NBD_INFO_EXPORT always */
701     myflags = exp->nbdflags;
702     if (client->mode >= NBD_MODE_STRUCTURED) {
703         myflags |= NBD_FLAG_SEND_DF;
704     }
705     if (client->mode >= NBD_MODE_EXTENDED &&
706         (client->contexts.count || client->opt == NBD_OPT_INFO)) {
707         myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD;
708     }
709     trace_nbd_negotiate_new_style_size_flags(exp->size, myflags);
710     stq_be_p(buf, exp->size);
711     stw_be_p(buf + 8, myflags);
712     rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT,
713                                  sizeof(buf), buf, errp);
714     if (rc < 0) {
715         return rc;
716     }
717 
718     /*
719      * If the client is just asking for NBD_OPT_INFO, but forgot to
720      * request block sizes in a situation that would impact
721      * performance, then return an error. But for NBD_OPT_GO, we
722      * tolerate all clients, regardless of alignments.
723      */
724     if (client->opt == NBD_OPT_INFO && !blocksize &&
725         blk_get_request_alignment(exp->common.blk) > 1) {
726         return nbd_negotiate_send_rep_err(client,
727                                           NBD_REP_ERR_BLOCK_SIZE_REQD,
728                                           errp,
729                                           "request NBD_INFO_BLOCK_SIZE to "
730                                           "use this export");
731     }
732 
733     /* Final reply */
734     rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
735     if (rc < 0) {
736         return rc;
737     }
738 
739     if (client->opt == NBD_OPT_GO) {
740         client->exp = exp;
741         client->check_align = check_align;
742         QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
743         blk_exp_ref(&client->exp->common);
744         rc = 1;
745     }
746     return rc;
747 }
748 
749 
750 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the
751  * new channel for all further (now-encrypted) communication. */
752 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client,
753                                                  Error **errp)
754 {
755     QIOChannel *ioc;
756     QIOChannelTLS *tioc;
757     struct NBDTLSHandshakeData data = { 0 };
758 
759     assert(client->opt == NBD_OPT_STARTTLS);
760 
761     trace_nbd_negotiate_handle_starttls();
762     ioc = client->ioc;
763 
764     if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) {
765         return NULL;
766     }
767 
768     tioc = qio_channel_tls_new_server(ioc,
769                                       client->tlscreds,
770                                       client->tlsauthz,
771                                       errp);
772     if (!tioc) {
773         return NULL;
774     }
775 
776     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls");
777     trace_nbd_negotiate_handle_starttls_handshake();
778     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
779     qio_channel_tls_handshake(tioc,
780                               nbd_tls_handshake,
781                               &data,
782                               NULL,
783                               NULL);
784 
785     if (!data.complete) {
786         g_main_loop_run(data.loop);
787     }
788     g_main_loop_unref(data.loop);
789     if (data.error) {
790         object_unref(OBJECT(tioc));
791         error_propagate(errp, data.error);
792         return NULL;
793     }
794 
795     return QIO_CHANNEL(tioc);
796 }
797 
798 /* nbd_negotiate_send_meta_context
799  *
800  * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT
801  *
802  * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead.
803  */
804 static int nbd_negotiate_send_meta_context(NBDClient *client,
805                                            const char *context,
806                                            uint32_t context_id,
807                                            Error **errp)
808 {
809     NBDOptionReplyMetaContext opt;
810     struct iovec iov[] = {
811         {.iov_base = &opt, .iov_len = sizeof(opt)},
812         {.iov_base = (void *)context, .iov_len = strlen(context)}
813     };
814 
815     assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE);
816     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
817         context_id = 0;
818     }
819 
820     trace_nbd_negotiate_meta_query_reply(context, context_id);
821     set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT,
822                       sizeof(opt) - sizeof(opt.h) + iov[1].iov_len);
823     stl_be_p(&opt.context_id, context_id);
824 
825     return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0;
826 }
827 
828 /*
829  * Return true if @query matches @pattern, or if @query is empty when
830  * the @client is performing _LIST_.
831  */
832 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern,
833                                       const char *query)
834 {
835     if (!*query) {
836         trace_nbd_negotiate_meta_query_parse("empty");
837         return client->opt == NBD_OPT_LIST_META_CONTEXT;
838     }
839     if (strcmp(query, pattern) == 0) {
840         trace_nbd_negotiate_meta_query_parse(pattern);
841         return true;
842     }
843     trace_nbd_negotiate_meta_query_skip("pattern not matched");
844     return false;
845 }
846 
847 /*
848  * Return true and adjust @str in place if it begins with @prefix.
849  */
850 static bool nbd_strshift(const char **str, const char *prefix)
851 {
852     size_t len = strlen(prefix);
853 
854     if (strncmp(*str, prefix, len) == 0) {
855         *str += len;
856         return true;
857     }
858     return false;
859 }
860 
861 /* nbd_meta_base_query
862  *
863  * Handle queries to 'base' namespace. For now, only the base:allocation
864  * context is available.  Return true if @query has been handled.
865  */
866 static bool nbd_meta_base_query(NBDClient *client, NBDMetaContexts *meta,
867                                 const char *query)
868 {
869     if (!nbd_strshift(&query, "base:")) {
870         return false;
871     }
872     trace_nbd_negotiate_meta_query_parse("base:");
873 
874     if (nbd_meta_empty_or_pattern(client, "allocation", query)) {
875         meta->base_allocation = true;
876     }
877     return true;
878 }
879 
880 /* nbd_meta_qemu_query
881  *
882  * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap:
883  * and qemu:allocation-depth contexts are available.  Return true if @query
884  * has been handled.
885  */
886 static bool nbd_meta_qemu_query(NBDClient *client, NBDMetaContexts *meta,
887                                 const char *query)
888 {
889     size_t i;
890 
891     if (!nbd_strshift(&query, "qemu:")) {
892         return false;
893     }
894     trace_nbd_negotiate_meta_query_parse("qemu:");
895 
896     if (!*query) {
897         if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
898             meta->allocation_depth = meta->exp->allocation_depth;
899             if (meta->exp->nr_export_bitmaps) {
900                 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
901             }
902         }
903         trace_nbd_negotiate_meta_query_parse("empty");
904         return true;
905     }
906 
907     if (strcmp(query, "allocation-depth") == 0) {
908         trace_nbd_negotiate_meta_query_parse("allocation-depth");
909         meta->allocation_depth = meta->exp->allocation_depth;
910         return true;
911     }
912 
913     if (nbd_strshift(&query, "dirty-bitmap:")) {
914         trace_nbd_negotiate_meta_query_parse("dirty-bitmap:");
915         if (!*query) {
916             if (client->opt == NBD_OPT_LIST_META_CONTEXT &&
917                 meta->exp->nr_export_bitmaps) {
918                 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
919             }
920             trace_nbd_negotiate_meta_query_parse("empty");
921             return true;
922         }
923 
924         for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
925             const char *bm_name;
926 
927             bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
928             if (strcmp(bm_name, query) == 0) {
929                 meta->bitmaps[i] = true;
930                 trace_nbd_negotiate_meta_query_parse(query);
931                 return true;
932             }
933         }
934         trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match");
935         return true;
936     }
937 
938     trace_nbd_negotiate_meta_query_skip("unknown qemu context");
939     return true;
940 }
941 
942 /* nbd_negotiate_meta_query
943  *
944  * Parse namespace name and call corresponding function to parse body of the
945  * query.
946  *
947  * The only supported namespaces are 'base' and 'qemu'.
948  *
949  * Return -errno on I/O error, 0 if option was completely handled by
950  * sending a reply about inconsistent lengths, or 1 on success. */
951 static int nbd_negotiate_meta_query(NBDClient *client,
952                                     NBDMetaContexts *meta, Error **errp)
953 {
954     int ret;
955     g_autofree char *query = NULL;
956     uint32_t len;
957 
958     ret = nbd_opt_read(client, &len, sizeof(len), false, errp);
959     if (ret <= 0) {
960         return ret;
961     }
962     len = cpu_to_be32(len);
963 
964     if (len > NBD_MAX_STRING_SIZE) {
965         trace_nbd_negotiate_meta_query_skip("length too long");
966         return nbd_opt_skip(client, len, errp);
967     }
968 
969     query = g_malloc(len + 1);
970     ret = nbd_opt_read(client, query, len, true, errp);
971     if (ret <= 0) {
972         return ret;
973     }
974     query[len] = '\0';
975 
976     if (nbd_meta_base_query(client, meta, query)) {
977         return 1;
978     }
979     if (nbd_meta_qemu_query(client, meta, query)) {
980         return 1;
981     }
982 
983     trace_nbd_negotiate_meta_query_skip("unknown namespace");
984     return 1;
985 }
986 
987 /* nbd_negotiate_meta_queries
988  * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT
989  *
990  * Return -errno on I/O error, or 0 if option was completely handled. */
991 static int nbd_negotiate_meta_queries(NBDClient *client, Error **errp)
992 {
993     int ret;
994     g_autofree char *export_name = NULL;
995     /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */
996     g_autofree G_GNUC_UNUSED bool *bitmaps = NULL;
997     NBDMetaContexts local_meta = {0};
998     NBDMetaContexts *meta;
999     uint32_t nb_queries;
1000     size_t i;
1001     size_t count = 0;
1002 
1003     if (client->opt == NBD_OPT_SET_META_CONTEXT &&
1004         client->mode < NBD_MODE_STRUCTURED) {
1005         return nbd_opt_invalid(client, errp,
1006                                "request option '%s' when structured reply "
1007                                "is not negotiated",
1008                                nbd_opt_lookup(client->opt));
1009     }
1010 
1011     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1012         /* Only change the caller's meta on SET. */
1013         meta = &local_meta;
1014     } else {
1015         meta = &client->contexts;
1016     }
1017 
1018     g_free(meta->bitmaps);
1019     memset(meta, 0, sizeof(*meta));
1020 
1021     ret = nbd_opt_read_name(client, &export_name, NULL, errp);
1022     if (ret <= 0) {
1023         return ret;
1024     }
1025 
1026     meta->exp = nbd_export_find(export_name);
1027     if (meta->exp == NULL) {
1028         g_autofree char *sane_name = nbd_sanitize_name(export_name);
1029 
1030         return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp,
1031                             "export '%s' not present", sane_name);
1032     }
1033     meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps);
1034     if (client->opt == NBD_OPT_LIST_META_CONTEXT) {
1035         bitmaps = meta->bitmaps;
1036     }
1037 
1038     ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp);
1039     if (ret <= 0) {
1040         return ret;
1041     }
1042     nb_queries = cpu_to_be32(nb_queries);
1043     trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt),
1044                                      export_name, nb_queries);
1045 
1046     if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) {
1047         /* enable all known contexts */
1048         meta->base_allocation = true;
1049         meta->allocation_depth = meta->exp->allocation_depth;
1050         if (meta->exp->nr_export_bitmaps) {
1051             memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps);
1052         }
1053     } else {
1054         for (i = 0; i < nb_queries; ++i) {
1055             ret = nbd_negotiate_meta_query(client, meta, errp);
1056             if (ret <= 0) {
1057                 return ret;
1058             }
1059         }
1060     }
1061 
1062     if (meta->base_allocation) {
1063         ret = nbd_negotiate_send_meta_context(client, "base:allocation",
1064                                               NBD_META_ID_BASE_ALLOCATION,
1065                                               errp);
1066         if (ret < 0) {
1067             return ret;
1068         }
1069         count++;
1070     }
1071 
1072     if (meta->allocation_depth) {
1073         ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth",
1074                                               NBD_META_ID_ALLOCATION_DEPTH,
1075                                               errp);
1076         if (ret < 0) {
1077             return ret;
1078         }
1079         count++;
1080     }
1081 
1082     for (i = 0; i < meta->exp->nr_export_bitmaps; i++) {
1083         const char *bm_name;
1084         g_autofree char *context = NULL;
1085 
1086         if (!meta->bitmaps[i]) {
1087             continue;
1088         }
1089 
1090         bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]);
1091         context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name);
1092 
1093         ret = nbd_negotiate_send_meta_context(client, context,
1094                                               NBD_META_ID_DIRTY_BITMAP + i,
1095                                               errp);
1096         if (ret < 0) {
1097             return ret;
1098         }
1099         count++;
1100     }
1101 
1102     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1103     if (ret == 0) {
1104         meta->count = count;
1105     }
1106 
1107     return ret;
1108 }
1109 
1110 /* nbd_negotiate_options
1111  * Process all NBD_OPT_* client option commands, during fixed newstyle
1112  * negotiation.
1113  * Return:
1114  * -errno  on error, errp is set
1115  * 0       on successful negotiation, errp is not set
1116  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1117  *         errp is not set
1118  */
1119 static int nbd_negotiate_options(NBDClient *client, Error **errp)
1120 {
1121     uint32_t flags;
1122     bool fixedNewstyle = false;
1123     bool no_zeroes = false;
1124 
1125     /* Client sends:
1126         [ 0 ..   3]   client flags
1127 
1128        Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO:
1129         [ 0 ..   7]   NBD_OPTS_MAGIC
1130         [ 8 ..  11]   NBD option
1131         [12 ..  15]   Data length
1132         ...           Rest of request
1133 
1134         [ 0 ..   7]   NBD_OPTS_MAGIC
1135         [ 8 ..  11]   Second NBD option
1136         [12 ..  15]   Data length
1137         ...           Rest of request
1138     */
1139 
1140     if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) {
1141         return -EIO;
1142     }
1143     client->mode = NBD_MODE_EXPORT_NAME;
1144     trace_nbd_negotiate_options_flags(flags);
1145     if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) {
1146         fixedNewstyle = true;
1147         flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE;
1148         client->mode = NBD_MODE_SIMPLE;
1149     }
1150     if (flags & NBD_FLAG_C_NO_ZEROES) {
1151         no_zeroes = true;
1152         flags &= ~NBD_FLAG_C_NO_ZEROES;
1153     }
1154     if (flags != 0) {
1155         error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags);
1156         return -EINVAL;
1157     }
1158 
1159     while (1) {
1160         int ret;
1161         uint32_t option, length;
1162         uint64_t magic;
1163 
1164         if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) {
1165             return -EINVAL;
1166         }
1167         trace_nbd_negotiate_options_check_magic(magic);
1168         if (magic != NBD_OPTS_MAGIC) {
1169             error_setg(errp, "Bad magic received");
1170             return -EINVAL;
1171         }
1172 
1173         if (nbd_read32(client->ioc, &option, "option", errp) < 0) {
1174             return -EINVAL;
1175         }
1176         client->opt = option;
1177 
1178         if (nbd_read32(client->ioc, &length, "option length", errp) < 0) {
1179             return -EINVAL;
1180         }
1181         assert(!client->optlen);
1182         client->optlen = length;
1183 
1184         if (length > NBD_MAX_BUFFER_SIZE) {
1185             error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)",
1186                        length, NBD_MAX_BUFFER_SIZE);
1187             return -EINVAL;
1188         }
1189 
1190         trace_nbd_negotiate_options_check_option(option,
1191                                                  nbd_opt_lookup(option));
1192         if (client->tlscreds &&
1193             client->ioc == (QIOChannel *)client->sioc) {
1194             QIOChannel *tioc;
1195             if (!fixedNewstyle) {
1196                 error_setg(errp, "Unsupported option 0x%" PRIx32, option);
1197                 return -EINVAL;
1198             }
1199             switch (option) {
1200             case NBD_OPT_STARTTLS:
1201                 if (length) {
1202                     /* Unconditionally drop the connection if the client
1203                      * can't start a TLS negotiation correctly */
1204                     return nbd_reject_length(client, true, errp);
1205                 }
1206                 tioc = nbd_negotiate_handle_starttls(client, errp);
1207                 if (!tioc) {
1208                     return -EIO;
1209                 }
1210                 ret = 0;
1211                 object_unref(OBJECT(client->ioc));
1212                 client->ioc = tioc;
1213                 break;
1214 
1215             case NBD_OPT_EXPORT_NAME:
1216                 /* No way to return an error to client, so drop connection */
1217                 error_setg(errp, "Option 0x%x not permitted before TLS",
1218                            option);
1219                 return -EINVAL;
1220 
1221             default:
1222                 /* Let the client keep trying, unless they asked to
1223                  * quit. Always try to give an error back to the
1224                  * client; but when replying to OPT_ABORT, be aware
1225                  * that the client may hang up before receiving the
1226                  * error, in which case we are fine ignoring the
1227                  * resulting EPIPE. */
1228                 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD,
1229                                    option == NBD_OPT_ABORT ? NULL : errp,
1230                                    "Option 0x%" PRIx32
1231                                    " not permitted before TLS", option);
1232                 if (option == NBD_OPT_ABORT) {
1233                     return 1;
1234                 }
1235                 break;
1236             }
1237         } else if (fixedNewstyle) {
1238             switch (option) {
1239             case NBD_OPT_LIST:
1240                 if (length) {
1241                     ret = nbd_reject_length(client, false, errp);
1242                 } else {
1243                     ret = nbd_negotiate_handle_list(client, errp);
1244                 }
1245                 break;
1246 
1247             case NBD_OPT_ABORT:
1248                 /* NBD spec says we must try to reply before
1249                  * disconnecting, but that we must also tolerate
1250                  * guests that don't wait for our reply. */
1251                 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL);
1252                 return 1;
1253 
1254             case NBD_OPT_EXPORT_NAME:
1255                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1256                                                         errp);
1257 
1258             case NBD_OPT_INFO:
1259             case NBD_OPT_GO:
1260                 ret = nbd_negotiate_handle_info(client, errp);
1261                 if (ret == 1) {
1262                     assert(option == NBD_OPT_GO);
1263                     return 0;
1264                 }
1265                 break;
1266 
1267             case NBD_OPT_STARTTLS:
1268                 if (length) {
1269                     ret = nbd_reject_length(client, false, errp);
1270                 } else if (client->tlscreds) {
1271                     ret = nbd_negotiate_send_rep_err(client,
1272                                                      NBD_REP_ERR_INVALID, errp,
1273                                                      "TLS already enabled");
1274                 } else {
1275                     ret = nbd_negotiate_send_rep_err(client,
1276                                                      NBD_REP_ERR_POLICY, errp,
1277                                                      "TLS not configured");
1278                 }
1279                 break;
1280 
1281             case NBD_OPT_STRUCTURED_REPLY:
1282                 if (length) {
1283                     ret = nbd_reject_length(client, false, errp);
1284                 } else if (client->mode >= NBD_MODE_EXTENDED) {
1285                     ret = nbd_negotiate_send_rep_err(
1286                         client, NBD_REP_ERR_EXT_HEADER_REQD, errp,
1287                         "extended headers already negotiated");
1288                 } else if (client->mode >= NBD_MODE_STRUCTURED) {
1289                     ret = nbd_negotiate_send_rep_err(
1290                         client, NBD_REP_ERR_INVALID, errp,
1291                         "structured reply already negotiated");
1292                 } else {
1293                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1294                     client->mode = NBD_MODE_STRUCTURED;
1295                 }
1296                 break;
1297 
1298             case NBD_OPT_LIST_META_CONTEXT:
1299             case NBD_OPT_SET_META_CONTEXT:
1300                 ret = nbd_negotiate_meta_queries(client, errp);
1301                 break;
1302 
1303             case NBD_OPT_EXTENDED_HEADERS:
1304                 if (length) {
1305                     ret = nbd_reject_length(client, false, errp);
1306                 } else if (client->mode >= NBD_MODE_EXTENDED) {
1307                     ret = nbd_negotiate_send_rep_err(
1308                         client, NBD_REP_ERR_INVALID, errp,
1309                         "extended headers already negotiated");
1310                 } else {
1311                     ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp);
1312                     client->mode = NBD_MODE_EXTENDED;
1313                 }
1314                 break;
1315 
1316             default:
1317                 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp,
1318                                    "Unsupported option %" PRIu32 " (%s)",
1319                                    option, nbd_opt_lookup(option));
1320                 break;
1321             }
1322         } else {
1323             /*
1324              * If broken new-style we should drop the connection
1325              * for anything except NBD_OPT_EXPORT_NAME
1326              */
1327             switch (option) {
1328             case NBD_OPT_EXPORT_NAME:
1329                 return nbd_negotiate_handle_export_name(client, no_zeroes,
1330                                                         errp);
1331 
1332             default:
1333                 error_setg(errp, "Unsupported option %" PRIu32 " (%s)",
1334                            option, nbd_opt_lookup(option));
1335                 return -EINVAL;
1336             }
1337         }
1338         if (ret < 0) {
1339             return ret;
1340         }
1341     }
1342 }
1343 
1344 /* nbd_negotiate
1345  * Return:
1346  * -errno  on error, errp is set
1347  * 0       on successful negotiation, errp is not set
1348  * 1       if client sent NBD_OPT_ABORT, i.e. on valid disconnect,
1349  *         errp is not set
1350  */
1351 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
1352 {
1353     ERRP_GUARD();
1354     char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = "";
1355     int ret;
1356 
1357     /* Old style negotiation header, no room for options
1358         [ 0 ..   7]   passwd       ("NBDMAGIC")
1359         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
1360         [16 ..  23]   size
1361         [24 ..  27]   export flags (zero-extended)
1362         [28 .. 151]   reserved     (0)
1363 
1364        New style negotiation header, client can send options
1365         [ 0 ..   7]   passwd       ("NBDMAGIC")
1366         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
1367         [16 ..  17]   server flags (0)
1368         ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO....
1369      */
1370 
1371     qio_channel_set_blocking(client->ioc, false, NULL);
1372     qio_channel_set_follow_coroutine_ctx(client->ioc, true);
1373 
1374     trace_nbd_negotiate_begin();
1375     memcpy(buf, "NBDMAGIC", 8);
1376 
1377     stq_be_p(buf + 8, NBD_OPTS_MAGIC);
1378     stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES);
1379 
1380     if (nbd_write(client->ioc, buf, 18, errp) < 0) {
1381         error_prepend(errp, "write failed: ");
1382         return -EINVAL;
1383     }
1384     ret = nbd_negotiate_options(client, errp);
1385     if (ret != 0) {
1386         if (ret < 0) {
1387             error_prepend(errp, "option negotiation failed: ");
1388         }
1389         return ret;
1390     }
1391 
1392     assert(!client->optlen);
1393     trace_nbd_negotiate_success();
1394 
1395     return 0;
1396 }
1397 
1398 /* nbd_read_eof
1399  * Tries to read @size bytes from @ioc. This is a local implementation of
1400  * qio_channel_readv_all_eof. We have it here because we need it to be
1401  * interruptible and to know when the coroutine is yielding.
1402  * Returns 1 on success
1403  *         0 on eof, when no data was read (errp is not set)
1404  *         negative errno on failure (errp is set)
1405  */
1406 static inline int coroutine_fn
1407 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp)
1408 {
1409     bool partial = false;
1410 
1411     assert(size);
1412     while (size > 0) {
1413         struct iovec iov = { .iov_base = buffer, .iov_len = size };
1414         ssize_t len;
1415 
1416         len = qio_channel_readv(client->ioc, &iov, 1, errp);
1417         if (len == QIO_CHANNEL_ERR_BLOCK) {
1418             client->read_yielding = true;
1419             qio_channel_yield(client->ioc, G_IO_IN);
1420             client->read_yielding = false;
1421             if (client->quiescing) {
1422                 return -EAGAIN;
1423             }
1424             continue;
1425         } else if (len < 0) {
1426             return -EIO;
1427         } else if (len == 0) {
1428             if (partial) {
1429                 error_setg(errp,
1430                            "Unexpected end-of-file before all bytes were read");
1431                 return -EIO;
1432             } else {
1433                 return 0;
1434             }
1435         }
1436 
1437         partial = true;
1438         size -= len;
1439         buffer = (uint8_t *) buffer + len;
1440     }
1441     return 1;
1442 }
1443 
1444 static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request,
1445                                             Error **errp)
1446 {
1447     uint8_t buf[NBD_EXTENDED_REQUEST_SIZE];
1448     uint32_t magic, expect;
1449     int ret;
1450     size_t size = client->mode >= NBD_MODE_EXTENDED ?
1451         NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE;
1452 
1453     ret = nbd_read_eof(client, buf, size, errp);
1454     if (ret < 0) {
1455         return ret;
1456     }
1457     if (ret == 0) {
1458         return -EIO;
1459     }
1460 
1461     /*
1462      * Compact request
1463      *  [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
1464      *  [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, ...)
1465      *  [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1466      *  [ 8 .. 15]   cookie
1467      *  [16 .. 23]   from
1468      *  [24 .. 27]   len
1469      * Extended request
1470      *  [ 0 ..  3]   magic   (NBD_EXTENDED_REQUEST_MAGIC)
1471      *  [ 4 ..  5]   flags   (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...)
1472      *  [ 6 ..  7]   type    (NBD_CMD_READ, ...)
1473      *  [ 8 .. 15]   cookie
1474      *  [16 .. 23]   from
1475      *  [24 .. 31]   len
1476      */
1477 
1478     magic = ldl_be_p(buf);
1479     request->flags  = lduw_be_p(buf + 4);
1480     request->type   = lduw_be_p(buf + 6);
1481     request->cookie = ldq_be_p(buf + 8);
1482     request->from   = ldq_be_p(buf + 16);
1483     if (client->mode >= NBD_MODE_EXTENDED) {
1484         request->len = ldq_be_p(buf + 24);
1485         expect = NBD_EXTENDED_REQUEST_MAGIC;
1486     } else {
1487         request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */
1488         expect = NBD_REQUEST_MAGIC;
1489     }
1490 
1491     trace_nbd_receive_request(magic, request->flags, request->type,
1492                               request->from, request->len);
1493 
1494     if (magic != expect) {
1495         error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%"
1496                    PRIx32 ")", magic, expect);
1497         return -EINVAL;
1498     }
1499     return 0;
1500 }
1501 
1502 #define MAX_NBD_REQUESTS 16
1503 
1504 void nbd_client_get(NBDClient *client)
1505 {
1506     client->refcount++;
1507 }
1508 
1509 void nbd_client_put(NBDClient *client)
1510 {
1511     if (--client->refcount == 0) {
1512         /* The last reference should be dropped by client->close,
1513          * which is called by client_close.
1514          */
1515         assert(client->closing);
1516 
1517         object_unref(OBJECT(client->sioc));
1518         object_unref(OBJECT(client->ioc));
1519         if (client->tlscreds) {
1520             object_unref(OBJECT(client->tlscreds));
1521         }
1522         g_free(client->tlsauthz);
1523         if (client->exp) {
1524             QTAILQ_REMOVE(&client->exp->clients, client, next);
1525             blk_exp_unref(&client->exp->common);
1526         }
1527         g_free(client->contexts.bitmaps);
1528         g_free(client);
1529     }
1530 }
1531 
1532 static void client_close(NBDClient *client, bool negotiated)
1533 {
1534     if (client->closing) {
1535         return;
1536     }
1537 
1538     client->closing = true;
1539 
1540     /* Force requests to finish.  They will drop their own references,
1541      * then we'll close the socket and free the NBDClient.
1542      */
1543     qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH,
1544                          NULL);
1545 
1546     /* Also tell the client, so that they release their reference.  */
1547     if (client->close_fn) {
1548         client->close_fn(client, negotiated);
1549     }
1550 }
1551 
1552 static NBDRequestData *nbd_request_get(NBDClient *client)
1553 {
1554     NBDRequestData *req;
1555 
1556     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
1557     client->nb_requests++;
1558 
1559     req = g_new0(NBDRequestData, 1);
1560     nbd_client_get(client);
1561     req->client = client;
1562     return req;
1563 }
1564 
1565 static void nbd_request_put(NBDRequestData *req)
1566 {
1567     NBDClient *client = req->client;
1568 
1569     if (req->data) {
1570         qemu_vfree(req->data);
1571     }
1572     g_free(req);
1573 
1574     client->nb_requests--;
1575 
1576     if (client->quiescing && client->nb_requests == 0) {
1577         aio_wait_kick();
1578     }
1579 
1580     nbd_client_receive_next_request(client);
1581 
1582     nbd_client_put(client);
1583 }
1584 
1585 static void blk_aio_attached(AioContext *ctx, void *opaque)
1586 {
1587     NBDExport *exp = opaque;
1588     NBDClient *client;
1589 
1590     trace_nbd_blk_aio_attached(exp->name, ctx);
1591 
1592     exp->common.ctx = ctx;
1593 
1594     QTAILQ_FOREACH(client, &exp->clients, next) {
1595         assert(client->nb_requests == 0);
1596         assert(client->recv_coroutine == NULL);
1597         assert(client->send_coroutine == NULL);
1598     }
1599 }
1600 
1601 static void blk_aio_detach(void *opaque)
1602 {
1603     NBDExport *exp = opaque;
1604 
1605     trace_nbd_blk_aio_detach(exp->name, exp->common.ctx);
1606 
1607     exp->common.ctx = NULL;
1608 }
1609 
1610 static void nbd_drained_begin(void *opaque)
1611 {
1612     NBDExport *exp = opaque;
1613     NBDClient *client;
1614 
1615     QTAILQ_FOREACH(client, &exp->clients, next) {
1616         client->quiescing = true;
1617     }
1618 }
1619 
1620 static void nbd_drained_end(void *opaque)
1621 {
1622     NBDExport *exp = opaque;
1623     NBDClient *client;
1624 
1625     QTAILQ_FOREACH(client, &exp->clients, next) {
1626         client->quiescing = false;
1627         nbd_client_receive_next_request(client);
1628     }
1629 }
1630 
1631 static bool nbd_drained_poll(void *opaque)
1632 {
1633     NBDExport *exp = opaque;
1634     NBDClient *client;
1635 
1636     QTAILQ_FOREACH(client, &exp->clients, next) {
1637         if (client->nb_requests != 0) {
1638             /*
1639              * If there's a coroutine waiting for a request on nbd_read_eof()
1640              * enter it here so we don't depend on the client to wake it up.
1641              */
1642             if (client->recv_coroutine != NULL && client->read_yielding) {
1643                 qio_channel_wake_read(client->ioc);
1644             }
1645 
1646             return true;
1647         }
1648     }
1649 
1650     return false;
1651 }
1652 
1653 static void nbd_eject_notifier(Notifier *n, void *data)
1654 {
1655     NBDExport *exp = container_of(n, NBDExport, eject_notifier);
1656 
1657     blk_exp_request_shutdown(&exp->common);
1658 }
1659 
1660 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk)
1661 {
1662     NBDExport *nbd_exp = container_of(exp, NBDExport, common);
1663     assert(exp->drv == &blk_exp_nbd);
1664     assert(nbd_exp->eject_notifier_blk == NULL);
1665 
1666     blk_ref(blk);
1667     nbd_exp->eject_notifier_blk = blk;
1668     nbd_exp->eject_notifier.notify = nbd_eject_notifier;
1669     blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier);
1670 }
1671 
1672 static const BlockDevOps nbd_block_ops = {
1673     .drained_begin = nbd_drained_begin,
1674     .drained_end = nbd_drained_end,
1675     .drained_poll = nbd_drained_poll,
1676 };
1677 
1678 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args,
1679                              Error **errp)
1680 {
1681     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1682     BlockExportOptionsNbd *arg = &exp_args->u.nbd;
1683     const char *name = arg->name ?: exp_args->node_name;
1684     BlockBackend *blk = blk_exp->blk;
1685     int64_t size;
1686     uint64_t perm, shared_perm;
1687     bool readonly = !exp_args->writable;
1688     BlockDirtyBitmapOrStrList *bitmaps;
1689     size_t i;
1690     int ret;
1691 
1692     assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1693 
1694     if (!nbd_server_is_running()) {
1695         error_setg(errp, "NBD server not running");
1696         return -EINVAL;
1697     }
1698 
1699     if (strlen(name) > NBD_MAX_STRING_SIZE) {
1700         error_setg(errp, "export name '%s' too long", name);
1701         return -EINVAL;
1702     }
1703 
1704     if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1705         error_setg(errp, "description '%s' too long", arg->description);
1706         return -EINVAL;
1707     }
1708 
1709     if (nbd_export_find(name)) {
1710         error_setg(errp, "NBD server already has export named '%s'", name);
1711         return -EEXIST;
1712     }
1713 
1714     size = blk_getlength(blk);
1715     if (size < 0) {
1716         error_setg_errno(errp, -size,
1717                          "Failed to determine the NBD export's length");
1718         return size;
1719     }
1720 
1721     /* Don't allow resize while the NBD server is running, otherwise we don't
1722      * care what happens with the node. */
1723     blk_get_perm(blk, &perm, &shared_perm);
1724     ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1725     if (ret < 0) {
1726         return ret;
1727     }
1728 
1729     QTAILQ_INIT(&exp->clients);
1730     exp->name = g_strdup(name);
1731     exp->description = g_strdup(arg->description);
1732     exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1733                      NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1734 
1735     if (nbd_server_max_connections() != 1) {
1736         exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1737     }
1738     if (readonly) {
1739         exp->nbdflags |= NBD_FLAG_READ_ONLY;
1740     } else {
1741         exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1742                           NBD_FLAG_SEND_FAST_ZERO);
1743     }
1744     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1745 
1746     for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1747         exp->nr_export_bitmaps++;
1748     }
1749     exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1750     for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1751          i++, bitmaps = bitmaps->next)
1752     {
1753         const char *bitmap;
1754         BlockDriverState *bs = blk_bs(blk);
1755         BdrvDirtyBitmap *bm = NULL;
1756 
1757         switch (bitmaps->value->type) {
1758         case QTYPE_QSTRING:
1759             bitmap = bitmaps->value->u.local;
1760             while (bs) {
1761                 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1762                 if (bm != NULL) {
1763                     break;
1764                 }
1765 
1766                 bs = bdrv_filter_or_cow_bs(bs);
1767             }
1768 
1769             if (bm == NULL) {
1770                 ret = -ENOENT;
1771                 error_setg(errp, "Bitmap '%s' is not found",
1772                            bitmaps->value->u.local);
1773                 goto fail;
1774             }
1775 
1776             if (readonly && bdrv_is_writable(bs) &&
1777                 bdrv_dirty_bitmap_enabled(bm)) {
1778                 ret = -EINVAL;
1779                 error_setg(errp, "Enabled bitmap '%s' incompatible with "
1780                            "readonly export", bitmap);
1781                 goto fail;
1782             }
1783             break;
1784         case QTYPE_QDICT:
1785             bitmap = bitmaps->value->u.external.name;
1786             bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node,
1787                                            bitmap, NULL, errp);
1788             if (!bm) {
1789                 ret = -ENOENT;
1790                 goto fail;
1791             }
1792             break;
1793         default:
1794             abort();
1795         }
1796 
1797         assert(bm);
1798 
1799         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1800             ret = -EINVAL;
1801             goto fail;
1802         }
1803 
1804         exp->export_bitmaps[i] = bm;
1805         assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1806     }
1807 
1808     /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1809     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1810         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1811     }
1812 
1813     exp->allocation_depth = arg->allocation_depth;
1814 
1815     /*
1816      * We need to inhibit request queuing in the block layer to ensure we can
1817      * be properly quiesced when entering a drained section, as our coroutines
1818      * servicing pending requests might enter blk_pread().
1819      */
1820     blk_set_disable_request_queuing(blk, true);
1821 
1822     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1823 
1824     blk_set_dev_ops(blk, &nbd_block_ops, exp);
1825 
1826     QTAILQ_INSERT_TAIL(&exports, exp, next);
1827 
1828     return 0;
1829 
1830 fail:
1831     g_free(exp->export_bitmaps);
1832     g_free(exp->name);
1833     g_free(exp->description);
1834     return ret;
1835 }
1836 
1837 NBDExport *nbd_export_find(const char *name)
1838 {
1839     NBDExport *exp;
1840     QTAILQ_FOREACH(exp, &exports, next) {
1841         if (strcmp(name, exp->name) == 0) {
1842             return exp;
1843         }
1844     }
1845 
1846     return NULL;
1847 }
1848 
1849 AioContext *
1850 nbd_export_aio_context(NBDExport *exp)
1851 {
1852     return exp->common.ctx;
1853 }
1854 
1855 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1856 {
1857     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1858     NBDClient *client, *next;
1859 
1860     blk_exp_ref(&exp->common);
1861     /*
1862      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1863      * close mode that stops advertising the export to new clients but
1864      * still permits existing clients to run to completion? Because of
1865      * that possibility, nbd_export_close() can be called more than
1866      * once on an export.
1867      */
1868     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1869         client_close(client, true);
1870     }
1871     if (exp->name) {
1872         g_free(exp->name);
1873         exp->name = NULL;
1874         QTAILQ_REMOVE(&exports, exp, next);
1875     }
1876     blk_exp_unref(&exp->common);
1877 }
1878 
1879 static void nbd_export_delete(BlockExport *blk_exp)
1880 {
1881     size_t i;
1882     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1883 
1884     assert(exp->name == NULL);
1885     assert(QTAILQ_EMPTY(&exp->clients));
1886 
1887     g_free(exp->description);
1888     exp->description = NULL;
1889 
1890     if (exp->eject_notifier_blk) {
1891         notifier_remove(&exp->eject_notifier);
1892         blk_unref(exp->eject_notifier_blk);
1893     }
1894     blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1895                                     blk_aio_detach, exp);
1896     blk_set_disable_request_queuing(exp->common.blk, false);
1897 
1898     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1899         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1900     }
1901 }
1902 
1903 const BlockExportDriver blk_exp_nbd = {
1904     .type               = BLOCK_EXPORT_TYPE_NBD,
1905     .instance_size      = sizeof(NBDExport),
1906     .create             = nbd_export_create,
1907     .delete             = nbd_export_delete,
1908     .request_shutdown   = nbd_export_request_shutdown,
1909 };
1910 
1911 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1912                                         unsigned niov, Error **errp)
1913 {
1914     int ret;
1915 
1916     g_assert(qemu_in_coroutine());
1917     qemu_co_mutex_lock(&client->send_lock);
1918     client->send_coroutine = qemu_coroutine_self();
1919 
1920     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1921 
1922     client->send_coroutine = NULL;
1923     qemu_co_mutex_unlock(&client->send_lock);
1924 
1925     return ret;
1926 }
1927 
1928 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1929                                        uint64_t cookie)
1930 {
1931     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1932     stl_be_p(&reply->error, error);
1933     stq_be_p(&reply->cookie, cookie);
1934 }
1935 
1936 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client,
1937                                                  NBDRequest *request,
1938                                                  uint32_t error,
1939                                                  void *data,
1940                                                  uint64_t len,
1941                                                  Error **errp)
1942 {
1943     NBDSimpleReply reply;
1944     int nbd_err = system_errno_to_nbd_errno(error);
1945     struct iovec iov[] = {
1946         {.iov_base = &reply, .iov_len = sizeof(reply)},
1947         {.iov_base = data, .iov_len = len}
1948     };
1949 
1950     assert(!len || !nbd_err);
1951     assert(len <= NBD_MAX_BUFFER_SIZE);
1952     assert(client->mode < NBD_MODE_STRUCTURED ||
1953            (client->mode == NBD_MODE_STRUCTURED &&
1954             request->type != NBD_CMD_READ));
1955     trace_nbd_co_send_simple_reply(request->cookie, nbd_err,
1956                                    nbd_err_lookup(nbd_err), len);
1957     set_be_simple_reply(&reply, nbd_err, request->cookie);
1958 
1959     return nbd_co_send_iov(client, iov, 2, errp);
1960 }
1961 
1962 /*
1963  * Prepare the header of a reply chunk for network transmission.
1964  *
1965  * On input, @iov is partially initialized: iov[0].iov_base must point
1966  * to an uninitialized NBDReply, while the remaining @niov elements
1967  * (if any) must be ready for transmission.  This function then
1968  * populates iov[0] for transmission.
1969  */
1970 static inline void set_be_chunk(NBDClient *client, struct iovec *iov,
1971                                 size_t niov, uint16_t flags, uint16_t type,
1972                                 NBDRequest *request)
1973 {
1974     size_t i, length = 0;
1975 
1976     for (i = 1; i < niov; i++) {
1977         length += iov[i].iov_len;
1978     }
1979     assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData));
1980 
1981     if (client->mode >= NBD_MODE_EXTENDED) {
1982         NBDExtendedReplyChunk *chunk = iov->iov_base;
1983 
1984         iov[0].iov_len = sizeof(*chunk);
1985         stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC);
1986         stw_be_p(&chunk->flags, flags);
1987         stw_be_p(&chunk->type, type);
1988         stq_be_p(&chunk->cookie, request->cookie);
1989         stq_be_p(&chunk->offset, request->from);
1990         stq_be_p(&chunk->length, length);
1991     } else {
1992         NBDStructuredReplyChunk *chunk = iov->iov_base;
1993 
1994         iov[0].iov_len = sizeof(*chunk);
1995         stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
1996         stw_be_p(&chunk->flags, flags);
1997         stw_be_p(&chunk->type, type);
1998         stq_be_p(&chunk->cookie, request->cookie);
1999         stl_be_p(&chunk->length, length);
2000     }
2001 }
2002 
2003 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client,
2004                                                NBDRequest *request,
2005                                                Error **errp)
2006 {
2007     NBDReply hdr;
2008     struct iovec iov[] = {
2009         {.iov_base = &hdr},
2010     };
2011 
2012     trace_nbd_co_send_chunk_done(request->cookie);
2013     set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE,
2014                  NBD_REPLY_TYPE_NONE, request);
2015     return nbd_co_send_iov(client, iov, 1, errp);
2016 }
2017 
2018 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client,
2019                                                NBDRequest *request,
2020                                                uint64_t offset,
2021                                                void *data,
2022                                                uint64_t size,
2023                                                bool final,
2024                                                Error **errp)
2025 {
2026     NBDReply hdr;
2027     NBDStructuredReadData chunk;
2028     struct iovec iov[] = {
2029         {.iov_base = &hdr},
2030         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2031         {.iov_base = data, .iov_len = size}
2032     };
2033 
2034     assert(size && size <= NBD_MAX_BUFFER_SIZE);
2035     trace_nbd_co_send_chunk_read(request->cookie, offset, data, size);
2036     set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0,
2037                  NBD_REPLY_TYPE_OFFSET_DATA, request);
2038     stq_be_p(&chunk.offset, offset);
2039 
2040     return nbd_co_send_iov(client, iov, 3, errp);
2041 }
2042 
2043 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client,
2044                                                 NBDRequest *request,
2045                                                 uint32_t error,
2046                                                 const char *msg,
2047                                                 Error **errp)
2048 {
2049     NBDReply hdr;
2050     NBDStructuredError chunk;
2051     int nbd_err = system_errno_to_nbd_errno(error);
2052     struct iovec iov[] = {
2053         {.iov_base = &hdr},
2054         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2055         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
2056     };
2057 
2058     assert(nbd_err);
2059     trace_nbd_co_send_chunk_error(request->cookie, nbd_err,
2060                                   nbd_err_lookup(nbd_err), msg ? msg : "");
2061     set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE,
2062                  NBD_REPLY_TYPE_ERROR, request);
2063     stl_be_p(&chunk.error, nbd_err);
2064     stw_be_p(&chunk.message_length, iov[2].iov_len);
2065 
2066     return nbd_co_send_iov(client, iov, 3, errp);
2067 }
2068 
2069 /* Do a sparse read and send the structured reply to the client.
2070  * Returns -errno if sending fails. blk_co_block_status_above() failure is
2071  * reported to the client, at which point this function succeeds.
2072  */
2073 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
2074                                                 NBDRequest *request,
2075                                                 uint64_t offset,
2076                                                 uint8_t *data,
2077                                                 uint64_t size,
2078                                                 Error **errp)
2079 {
2080     int ret = 0;
2081     NBDExport *exp = client->exp;
2082     size_t progress = 0;
2083 
2084     assert(size <= NBD_MAX_BUFFER_SIZE);
2085     while (progress < size) {
2086         int64_t pnum;
2087         int status = blk_co_block_status_above(exp->common.blk, NULL,
2088                                                offset + progress,
2089                                                size - progress, &pnum, NULL,
2090                                                NULL);
2091         bool final;
2092 
2093         if (status < 0) {
2094             char *msg = g_strdup_printf("unable to check for holes: %s",
2095                                         strerror(-status));
2096 
2097             ret = nbd_co_send_chunk_error(client, request, -status, msg, errp);
2098             g_free(msg);
2099             return ret;
2100         }
2101         assert(pnum && pnum <= size - progress);
2102         final = progress + pnum == size;
2103         if (status & BDRV_BLOCK_ZERO) {
2104             NBDReply hdr;
2105             NBDStructuredReadHole chunk;
2106             struct iovec iov[] = {
2107                 {.iov_base = &hdr},
2108                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2109             };
2110 
2111             trace_nbd_co_send_chunk_read_hole(request->cookie,
2112                                               offset + progress, pnum);
2113             set_be_chunk(client, iov, 2,
2114                          final ? NBD_REPLY_FLAG_DONE : 0,
2115                          NBD_REPLY_TYPE_OFFSET_HOLE, request);
2116             stq_be_p(&chunk.offset, offset + progress);
2117             stl_be_p(&chunk.length, pnum);
2118             ret = nbd_co_send_iov(client, iov, 2, errp);
2119         } else {
2120             ret = blk_co_pread(exp->common.blk, offset + progress, pnum,
2121                                data + progress, 0);
2122             if (ret < 0) {
2123                 error_setg_errno(errp, -ret, "reading from file failed");
2124                 break;
2125             }
2126             ret = nbd_co_send_chunk_read(client, request, offset + progress,
2127                                          data + progress, pnum, final, errp);
2128         }
2129 
2130         if (ret < 0) {
2131             break;
2132         }
2133         progress += pnum;
2134     }
2135     return ret;
2136 }
2137 
2138 typedef struct NBDExtentArray {
2139     NBDExtent64 *extents;
2140     unsigned int nb_alloc;
2141     unsigned int count;
2142     uint64_t total_length;
2143     bool extended;
2144     bool can_add;
2145     bool converted_to_be;
2146 } NBDExtentArray;
2147 
2148 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc,
2149                                             NBDMode mode)
2150 {
2151     NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2152 
2153     assert(mode >= NBD_MODE_STRUCTURED);
2154     ea->nb_alloc = nb_alloc;
2155     ea->extents = g_new(NBDExtent64, nb_alloc);
2156     ea->extended = mode >= NBD_MODE_EXTENDED;
2157     ea->can_add = true;
2158 
2159     return ea;
2160 }
2161 
2162 static void nbd_extent_array_free(NBDExtentArray *ea)
2163 {
2164     g_free(ea->extents);
2165     g_free(ea);
2166 }
2167 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free)
2168 
2169 /* Further modifications of the array after conversion are abandoned */
2170 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2171 {
2172     int i;
2173 
2174     assert(!ea->converted_to_be);
2175     assert(ea->extended);
2176     ea->can_add = false;
2177     ea->converted_to_be = true;
2178 
2179     for (i = 0; i < ea->count; i++) {
2180         ea->extents[i].length = cpu_to_be64(ea->extents[i].length);
2181         ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags);
2182     }
2183 }
2184 
2185 /* Further modifications of the array after conversion are abandoned */
2186 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea)
2187 {
2188     int i;
2189     NBDExtent32 *extents = g_new(NBDExtent32, ea->count);
2190 
2191     assert(!ea->converted_to_be);
2192     assert(!ea->extended);
2193     ea->can_add = false;
2194     ea->converted_to_be = true;
2195 
2196     for (i = 0; i < ea->count; i++) {
2197         assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX);
2198         extents[i].length = cpu_to_be32(ea->extents[i].length);
2199         extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2200     }
2201 
2202     return extents;
2203 }
2204 
2205 /*
2206  * Add extent to NBDExtentArray. If extent can't be added (no available space),
2207  * return -1.
2208  * For safety, when returning -1 for the first time, .can_add is set to false,
2209  * and further calls to nbd_extent_array_add() will crash.
2210  * (this avoids the situation where a caller ignores failure to add one extent,
2211  * where adding another extent that would squash into the last array entry
2212  * would result in an incorrect range reported to the client)
2213  */
2214 static int nbd_extent_array_add(NBDExtentArray *ea,
2215                                 uint64_t length, uint32_t flags)
2216 {
2217     assert(ea->can_add);
2218 
2219     if (!length) {
2220         return 0;
2221     }
2222     if (!ea->extended) {
2223         assert(length <= UINT32_MAX);
2224     }
2225 
2226     /* Extend previous extent if flags are the same */
2227     if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2228         uint64_t sum = length + ea->extents[ea->count - 1].length;
2229 
2230         /*
2231          * sum cannot overflow: the block layer bounds image size at
2232          * 2^63, and ea->extents[].length comes from the block layer.
2233          */
2234         assert(sum >= length);
2235         if (sum <= UINT32_MAX || ea->extended) {
2236             ea->extents[ea->count - 1].length = sum;
2237             ea->total_length += length;
2238             return 0;
2239         }
2240     }
2241 
2242     if (ea->count >= ea->nb_alloc) {
2243         ea->can_add = false;
2244         return -1;
2245     }
2246 
2247     ea->total_length += length;
2248     ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags};
2249     ea->count++;
2250 
2251     return 0;
2252 }
2253 
2254 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk,
2255                                                uint64_t offset, uint64_t bytes,
2256                                                NBDExtentArray *ea)
2257 {
2258     while (bytes) {
2259         uint32_t flags;
2260         int64_t num;
2261         int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num,
2262                                             NULL, NULL);
2263 
2264         if (ret < 0) {
2265             return ret;
2266         }
2267 
2268         flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2269                 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2270 
2271         if (nbd_extent_array_add(ea, num, flags) < 0) {
2272             return 0;
2273         }
2274 
2275         offset += num;
2276         bytes -= num;
2277     }
2278 
2279     return 0;
2280 }
2281 
2282 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk,
2283                                               uint64_t offset, uint64_t bytes,
2284                                               NBDExtentArray *ea)
2285 {
2286     while (bytes) {
2287         int64_t num;
2288         int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes,
2289                                             &num);
2290 
2291         if (ret < 0) {
2292             return ret;
2293         }
2294 
2295         if (nbd_extent_array_add(ea, num, ret) < 0) {
2296             return 0;
2297         }
2298 
2299         offset += num;
2300         bytes -= num;
2301     }
2302 
2303     return 0;
2304 }
2305 
2306 /*
2307  * nbd_co_send_extents
2308  *
2309  * @ea is converted to BE by the function
2310  * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2311  */
2312 static int coroutine_fn
2313 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea,
2314                     bool last, uint32_t context_id, Error **errp)
2315 {
2316     NBDReply hdr;
2317     NBDStructuredMeta meta;
2318     NBDExtendedMeta meta_ext;
2319     g_autofree NBDExtent32 *extents = NULL;
2320     uint16_t type;
2321     struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} };
2322 
2323     if (client->mode >= NBD_MODE_EXTENDED) {
2324         type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT;
2325 
2326         iov[1].iov_base = &meta_ext;
2327         iov[1].iov_len = sizeof(meta_ext);
2328         stl_be_p(&meta_ext.context_id, context_id);
2329         stl_be_p(&meta_ext.count, ea->count);
2330 
2331         nbd_extent_array_convert_to_be(ea);
2332         iov[2].iov_base = ea->extents;
2333         iov[2].iov_len = ea->count * sizeof(ea->extents[0]);
2334     } else {
2335         type = NBD_REPLY_TYPE_BLOCK_STATUS;
2336 
2337         iov[1].iov_base = &meta;
2338         iov[1].iov_len = sizeof(meta);
2339         stl_be_p(&meta.context_id, context_id);
2340 
2341         extents = nbd_extent_array_convert_to_narrow(ea);
2342         iov[2].iov_base = extents;
2343         iov[2].iov_len = ea->count * sizeof(extents[0]);
2344     }
2345 
2346     trace_nbd_co_send_extents(request->cookie, ea->count, context_id,
2347                               ea->total_length, last);
2348     set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type,
2349                  request);
2350 
2351     return nbd_co_send_iov(client, iov, 3, errp);
2352 }
2353 
2354 /* Get block status from the exported device and send it to the client */
2355 static int
2356 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request,
2357                                       BlockBackend *blk, uint64_t offset,
2358                                       uint64_t length, bool dont_fragment,
2359                                       bool last, uint32_t context_id,
2360                                       Error **errp)
2361 {
2362     int ret;
2363     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2364     g_autoptr(NBDExtentArray) ea =
2365         nbd_extent_array_new(nb_extents, client->mode);
2366 
2367     if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2368         ret = blockstatus_to_extents(blk, offset, length, ea);
2369     } else {
2370         ret = blockalloc_to_extents(blk, offset, length, ea);
2371     }
2372     if (ret < 0) {
2373         return nbd_co_send_chunk_error(client, request, -ret,
2374                                        "can't get block status", errp);
2375     }
2376 
2377     return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2378 }
2379 
2380 /* Populate @ea from a dirty bitmap. */
2381 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2382                               uint64_t offset, uint64_t length,
2383                               NBDExtentArray *es)
2384 {
2385     int64_t start, dirty_start, dirty_count;
2386     int64_t end = offset + length;
2387     bool full = false;
2388     int64_t bound = es->extended ? INT64_MAX : INT32_MAX;
2389 
2390     bdrv_dirty_bitmap_lock(bitmap);
2391 
2392     for (start = offset;
2393          bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound,
2394                                            &dirty_start, &dirty_count);
2395          start = dirty_start + dirty_count)
2396     {
2397         if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2398             (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2399         {
2400             full = true;
2401             break;
2402         }
2403     }
2404 
2405     if (!full) {
2406         /* last non dirty extent, nothing to do if array is now full */
2407         (void) nbd_extent_array_add(es, end - start, 0);
2408     }
2409 
2410     bdrv_dirty_bitmap_unlock(bitmap);
2411 }
2412 
2413 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client,
2414                                            NBDRequest *request,
2415                                            BdrvDirtyBitmap *bitmap,
2416                                            uint64_t offset,
2417                                            uint64_t length, bool dont_fragment,
2418                                            bool last, uint32_t context_id,
2419                                            Error **errp)
2420 {
2421     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2422     g_autoptr(NBDExtentArray) ea =
2423         nbd_extent_array_new(nb_extents, client->mode);
2424 
2425     bitmap_to_extents(bitmap, offset, length, ea);
2426 
2427     return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2428 }
2429 
2430 /*
2431  * nbd_co_block_status_payload_read
2432  * Called when a client wants a subset of negotiated contexts via a
2433  * BLOCK_STATUS payload.  Check the payload for valid length and
2434  * contents.  On success, return 0 with request updated to effective
2435  * length.  If request was invalid but all payload consumed, return 0
2436  * with request->len and request->contexts->count set to 0 (which will
2437  * trigger an appropriate NBD_EINVAL response later on).  Return
2438  * negative errno if the payload was not fully consumed.
2439  */
2440 static int
2441 nbd_co_block_status_payload_read(NBDClient *client, NBDRequest *request,
2442                                  Error **errp)
2443 {
2444     uint64_t payload_len = request->len;
2445     g_autofree char *buf = NULL;
2446     size_t count, i, nr_bitmaps;
2447     uint32_t id;
2448 
2449     if (payload_len > NBD_MAX_BUFFER_SIZE) {
2450         error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2451                    request->len, NBD_MAX_BUFFER_SIZE);
2452         return -EINVAL;
2453     }
2454 
2455     assert(client->contexts.exp == client->exp);
2456     nr_bitmaps = client->exp->nr_export_bitmaps;
2457     request->contexts = g_new0(NBDMetaContexts, 1);
2458     request->contexts->exp = client->exp;
2459 
2460     if (payload_len % sizeof(uint32_t) ||
2461         payload_len < sizeof(NBDBlockStatusPayload) ||
2462         payload_len > (sizeof(NBDBlockStatusPayload) +
2463                        sizeof(id) * client->contexts.count)) {
2464         goto skip;
2465     }
2466 
2467     buf = g_malloc(payload_len);
2468     if (nbd_read(client->ioc, buf, payload_len,
2469                  "CMD_BLOCK_STATUS data", errp) < 0) {
2470         return -EIO;
2471     }
2472     trace_nbd_co_receive_request_payload_received(request->cookie,
2473                                                   payload_len);
2474     request->contexts->bitmaps = g_new0(bool, nr_bitmaps);
2475     count = (payload_len - sizeof(NBDBlockStatusPayload)) / sizeof(id);
2476     payload_len = 0;
2477 
2478     for (i = 0; i < count; i++) {
2479         id = ldl_be_p(buf + sizeof(NBDBlockStatusPayload) + sizeof(id) * i);
2480         if (id == NBD_META_ID_BASE_ALLOCATION) {
2481             if (!client->contexts.base_allocation ||
2482                 request->contexts->base_allocation) {
2483                 goto skip;
2484             }
2485             request->contexts->base_allocation = true;
2486         } else if (id == NBD_META_ID_ALLOCATION_DEPTH) {
2487             if (!client->contexts.allocation_depth ||
2488                 request->contexts->allocation_depth) {
2489                 goto skip;
2490             }
2491             request->contexts->allocation_depth = true;
2492         } else {
2493             unsigned idx = id - NBD_META_ID_DIRTY_BITMAP;
2494 
2495             if (idx >= nr_bitmaps || !client->contexts.bitmaps[idx] ||
2496                 request->contexts->bitmaps[idx]) {
2497                 goto skip;
2498             }
2499             request->contexts->bitmaps[idx] = true;
2500         }
2501     }
2502 
2503     request->len = ldq_be_p(buf);
2504     request->contexts->count = count;
2505     return 0;
2506 
2507  skip:
2508     trace_nbd_co_receive_block_status_payload_compliance(request->from,
2509                                                          request->len);
2510     request->len = request->contexts->count = 0;
2511     return nbd_drop(client->ioc, payload_len, errp);
2512 }
2513 
2514 /* nbd_co_receive_request
2515  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2516  * connection right away, -EAGAIN to indicate we were interrupted and the
2517  * channel should be quiesced, and any other negative value to report an error
2518  * to the client (although the caller may still need to disconnect after
2519  * reporting the error).
2520  */
2521 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req,
2522                                                NBDRequest *request,
2523                                                Error **errp)
2524 {
2525     NBDClient *client = req->client;
2526     bool extended_with_payload;
2527     bool check_length = false;
2528     bool check_rofs = false;
2529     bool allocate_buffer = false;
2530     bool payload_okay = false;
2531     uint64_t payload_len = 0;
2532     int valid_flags = NBD_CMD_FLAG_FUA;
2533     int ret;
2534 
2535     g_assert(qemu_in_coroutine());
2536     assert(client->recv_coroutine == qemu_coroutine_self());
2537     ret = nbd_receive_request(client, request, errp);
2538     if (ret < 0) {
2539         return ret;
2540     }
2541 
2542     trace_nbd_co_receive_request_decode_type(request->cookie, request->type,
2543                                              nbd_cmd_lookup(request->type));
2544     extended_with_payload = client->mode >= NBD_MODE_EXTENDED &&
2545         request->flags & NBD_CMD_FLAG_PAYLOAD_LEN;
2546     if (extended_with_payload) {
2547         payload_len = request->len;
2548         check_length = true;
2549     }
2550 
2551     switch (request->type) {
2552     case NBD_CMD_DISC:
2553         /* Special case: we're going to disconnect without a reply,
2554          * whether or not flags, from, or len are bogus */
2555         req->complete = true;
2556         return -EIO;
2557 
2558     case NBD_CMD_READ:
2559         if (client->mode >= NBD_MODE_STRUCTURED) {
2560             valid_flags |= NBD_CMD_FLAG_DF;
2561         }
2562         check_length = true;
2563         allocate_buffer = true;
2564         break;
2565 
2566     case NBD_CMD_WRITE:
2567         if (client->mode >= NBD_MODE_EXTENDED) {
2568             if (!extended_with_payload) {
2569                 /* The client is noncompliant. Trace it, but proceed. */
2570                 trace_nbd_co_receive_ext_payload_compliance(request->from,
2571                                                             request->len);
2572             }
2573             valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2574         }
2575         payload_okay = true;
2576         payload_len = request->len;
2577         check_length = true;
2578         allocate_buffer = true;
2579         check_rofs = true;
2580         break;
2581 
2582     case NBD_CMD_FLUSH:
2583         break;
2584 
2585     case NBD_CMD_TRIM:
2586         check_rofs = true;
2587         break;
2588 
2589     case NBD_CMD_CACHE:
2590         check_length = true;
2591         break;
2592 
2593     case NBD_CMD_WRITE_ZEROES:
2594         valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2595         check_rofs = true;
2596         break;
2597 
2598     case NBD_CMD_BLOCK_STATUS:
2599         if (extended_with_payload) {
2600             ret = nbd_co_block_status_payload_read(client, request, errp);
2601             if (ret < 0) {
2602                 return ret;
2603             }
2604             /* payload now consumed */
2605             check_length = false;
2606             payload_len = 0;
2607             valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2608         } else {
2609             request->contexts = &client->contexts;
2610         }
2611         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2612         break;
2613 
2614     default:
2615         /* Unrecognized, will fail later */
2616         ;
2617     }
2618 
2619     /* Payload and buffer handling. */
2620     if (!payload_len) {
2621         req->complete = true;
2622     }
2623     if (check_length && request->len > NBD_MAX_BUFFER_SIZE) {
2624         /* READ, WRITE, CACHE */
2625         error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2626                    request->len, NBD_MAX_BUFFER_SIZE);
2627         return -EINVAL;
2628     }
2629     if (payload_len && !payload_okay) {
2630         /*
2631          * For now, we don't support payloads on other commands; but
2632          * we can keep the connection alive by ignoring the payload.
2633          * We will fail the command later with NBD_EINVAL for the use
2634          * of an unsupported flag (and not for access beyond bounds).
2635          */
2636         assert(request->type != NBD_CMD_WRITE);
2637         request->len = 0;
2638     }
2639     if (allocate_buffer) {
2640         /* READ, WRITE */
2641         req->data = blk_try_blockalign(client->exp->common.blk,
2642                                        request->len);
2643         if (req->data == NULL) {
2644             error_setg(errp, "No memory");
2645             return -ENOMEM;
2646         }
2647     }
2648     if (payload_len) {
2649         if (payload_okay) {
2650             /* WRITE */
2651             assert(req->data);
2652             ret = nbd_read(client->ioc, req->data, payload_len,
2653                            "CMD_WRITE data", errp);
2654         } else {
2655             ret = nbd_drop(client->ioc, payload_len, errp);
2656         }
2657         if (ret < 0) {
2658             return -EIO;
2659         }
2660         req->complete = true;
2661         trace_nbd_co_receive_request_payload_received(request->cookie,
2662                                                       payload_len);
2663     }
2664 
2665     /* Sanity checks. */
2666     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) {
2667         /* WRITE, TRIM, WRITE_ZEROES */
2668         error_setg(errp, "Export is read-only");
2669         return -EROFS;
2670     }
2671     if (request->from > client->exp->size ||
2672         request->len > client->exp->size - request->from) {
2673         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64
2674                    ", Size: %" PRIu64, request->from, request->len,
2675                    client->exp->size);
2676         return (request->type == NBD_CMD_WRITE ||
2677                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2678     }
2679     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2680                                                 client->check_align)) {
2681         /*
2682          * The block layer gracefully handles unaligned requests, but
2683          * it's still worth tracing client non-compliance
2684          */
2685         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2686                                               request->from,
2687                                               request->len,
2688                                               client->check_align);
2689     }
2690     if (request->flags & ~valid_flags) {
2691         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2692                    nbd_cmd_lookup(request->type), request->flags);
2693         return -EINVAL;
2694     }
2695 
2696     return 0;
2697 }
2698 
2699 /* Send simple reply without a payload, or a structured error
2700  * @error_msg is ignored if @ret >= 0
2701  * Returns 0 if connection is still live, -errno on failure to talk to client
2702  */
2703 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2704                                                NBDRequest *request,
2705                                                int ret,
2706                                                const char *error_msg,
2707                                                Error **errp)
2708 {
2709     if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) {
2710         return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp);
2711     } else if (client->mode >= NBD_MODE_EXTENDED) {
2712         return nbd_co_send_chunk_done(client, request, errp);
2713     } else {
2714         return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0,
2715                                         NULL, 0, errp);
2716     }
2717 }
2718 
2719 /* Handle NBD_CMD_READ request.
2720  * Return -errno if sending fails. Other errors are reported directly to the
2721  * client as an error reply. */
2722 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2723                                         uint8_t *data, Error **errp)
2724 {
2725     int ret;
2726     NBDExport *exp = client->exp;
2727 
2728     assert(request->type == NBD_CMD_READ);
2729     assert(request->len <= NBD_MAX_BUFFER_SIZE);
2730 
2731     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2732     if (request->flags & NBD_CMD_FLAG_FUA) {
2733         ret = blk_co_flush(exp->common.blk);
2734         if (ret < 0) {
2735             return nbd_send_generic_reply(client, request, ret,
2736                                           "flush failed", errp);
2737         }
2738     }
2739 
2740     if (client->mode >= NBD_MODE_STRUCTURED &&
2741         !(request->flags & NBD_CMD_FLAG_DF) && request->len)
2742     {
2743         return nbd_co_send_sparse_read(client, request, request->from,
2744                                        data, request->len, errp);
2745     }
2746 
2747     ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0);
2748     if (ret < 0) {
2749         return nbd_send_generic_reply(client, request, ret,
2750                                       "reading from file failed", errp);
2751     }
2752 
2753     if (client->mode >= NBD_MODE_STRUCTURED) {
2754         if (request->len) {
2755             return nbd_co_send_chunk_read(client, request, request->from, data,
2756                                           request->len, true, errp);
2757         } else {
2758             return nbd_co_send_chunk_done(client, request, errp);
2759         }
2760     } else {
2761         return nbd_co_send_simple_reply(client, request, 0,
2762                                         data, request->len, errp);
2763     }
2764 }
2765 
2766 /*
2767  * nbd_do_cmd_cache
2768  *
2769  * Handle NBD_CMD_CACHE request.
2770  * Return -errno if sending fails. Other errors are reported directly to the
2771  * client as an error reply.
2772  */
2773 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2774                                          Error **errp)
2775 {
2776     int ret;
2777     NBDExport *exp = client->exp;
2778 
2779     assert(request->type == NBD_CMD_CACHE);
2780     assert(request->len <= NBD_MAX_BUFFER_SIZE);
2781 
2782     ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2783                         NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2784 
2785     return nbd_send_generic_reply(client, request, ret,
2786                                   "caching data failed", errp);
2787 }
2788 
2789 /* Handle NBD request.
2790  * Return -errno if sending fails. Other errors are reported directly to the
2791  * client as an error reply. */
2792 static coroutine_fn int nbd_handle_request(NBDClient *client,
2793                                            NBDRequest *request,
2794                                            uint8_t *data, Error **errp)
2795 {
2796     int ret;
2797     int flags;
2798     NBDExport *exp = client->exp;
2799     char *msg;
2800     size_t i;
2801 
2802     switch (request->type) {
2803     case NBD_CMD_CACHE:
2804         return nbd_do_cmd_cache(client, request, errp);
2805 
2806     case NBD_CMD_READ:
2807         return nbd_do_cmd_read(client, request, data, errp);
2808 
2809     case NBD_CMD_WRITE:
2810         flags = 0;
2811         if (request->flags & NBD_CMD_FLAG_FUA) {
2812             flags |= BDRV_REQ_FUA;
2813         }
2814         assert(request->len <= NBD_MAX_BUFFER_SIZE);
2815         ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data,
2816                             flags);
2817         return nbd_send_generic_reply(client, request, ret,
2818                                       "writing to file failed", errp);
2819 
2820     case NBD_CMD_WRITE_ZEROES:
2821         flags = 0;
2822         if (request->flags & NBD_CMD_FLAG_FUA) {
2823             flags |= BDRV_REQ_FUA;
2824         }
2825         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2826             flags |= BDRV_REQ_MAY_UNMAP;
2827         }
2828         if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2829             flags |= BDRV_REQ_NO_FALLBACK;
2830         }
2831         ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len,
2832                                    flags);
2833         return nbd_send_generic_reply(client, request, ret,
2834                                       "writing to file failed", errp);
2835 
2836     case NBD_CMD_DISC:
2837         /* unreachable, thanks to special case in nbd_co_receive_request() */
2838         abort();
2839 
2840     case NBD_CMD_FLUSH:
2841         ret = blk_co_flush(exp->common.blk);
2842         return nbd_send_generic_reply(client, request, ret,
2843                                       "flush failed", errp);
2844 
2845     case NBD_CMD_TRIM:
2846         ret = blk_co_pdiscard(exp->common.blk, request->from, request->len);
2847         if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2848             ret = blk_co_flush(exp->common.blk);
2849         }
2850         return nbd_send_generic_reply(client, request, ret,
2851                                       "discard failed", errp);
2852 
2853     case NBD_CMD_BLOCK_STATUS:
2854         assert(request->contexts);
2855         assert(client->mode >= NBD_MODE_EXTENDED ||
2856                request->len <= UINT32_MAX);
2857         if (request->contexts->count) {
2858             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2859             int contexts_remaining = request->contexts->count;
2860 
2861             if (!request->len) {
2862                 return nbd_send_generic_reply(client, request, -EINVAL,
2863                                               "need non-zero length", errp);
2864             }
2865             if (request->contexts->base_allocation) {
2866                 ret = nbd_co_send_block_status(client, request,
2867                                                exp->common.blk,
2868                                                request->from,
2869                                                request->len, dont_fragment,
2870                                                !--contexts_remaining,
2871                                                NBD_META_ID_BASE_ALLOCATION,
2872                                                errp);
2873                 if (ret < 0) {
2874                     return ret;
2875                 }
2876             }
2877 
2878             if (request->contexts->allocation_depth) {
2879                 ret = nbd_co_send_block_status(client, request,
2880                                                exp->common.blk,
2881                                                request->from, request->len,
2882                                                dont_fragment,
2883                                                !--contexts_remaining,
2884                                                NBD_META_ID_ALLOCATION_DEPTH,
2885                                                errp);
2886                 if (ret < 0) {
2887                     return ret;
2888                 }
2889             }
2890 
2891             assert(request->contexts->exp == client->exp);
2892             for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2893                 if (!request->contexts->bitmaps[i]) {
2894                     continue;
2895                 }
2896                 ret = nbd_co_send_bitmap(client, request,
2897                                          client->exp->export_bitmaps[i],
2898                                          request->from, request->len,
2899                                          dont_fragment, !--contexts_remaining,
2900                                          NBD_META_ID_DIRTY_BITMAP + i, errp);
2901                 if (ret < 0) {
2902                     return ret;
2903                 }
2904             }
2905 
2906             assert(!contexts_remaining);
2907 
2908             return 0;
2909         } else if (client->contexts.count) {
2910             return nbd_send_generic_reply(client, request, -EINVAL,
2911                                           "CMD_BLOCK_STATUS payload not valid",
2912                                           errp);
2913         } else {
2914             return nbd_send_generic_reply(client, request, -EINVAL,
2915                                           "CMD_BLOCK_STATUS not negotiated",
2916                                           errp);
2917         }
2918 
2919     default:
2920         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2921                               request->type);
2922         ret = nbd_send_generic_reply(client, request, -EINVAL, msg,
2923                                      errp);
2924         g_free(msg);
2925         return ret;
2926     }
2927 }
2928 
2929 /* Owns a reference to the NBDClient passed as opaque.  */
2930 static coroutine_fn void nbd_trip(void *opaque)
2931 {
2932     NBDClient *client = opaque;
2933     NBDRequestData *req;
2934     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2935     int ret;
2936     Error *local_err = NULL;
2937 
2938     trace_nbd_trip();
2939     if (client->closing) {
2940         nbd_client_put(client);
2941         return;
2942     }
2943 
2944     if (client->quiescing) {
2945         /*
2946          * We're switching between AIO contexts. Don't attempt to receive a new
2947          * request and kick the main context which may be waiting for us.
2948          */
2949         nbd_client_put(client);
2950         client->recv_coroutine = NULL;
2951         aio_wait_kick();
2952         return;
2953     }
2954 
2955     req = nbd_request_get(client);
2956     ret = nbd_co_receive_request(req, &request, &local_err);
2957     client->recv_coroutine = NULL;
2958 
2959     if (client->closing) {
2960         /*
2961          * The client may be closed when we are blocked in
2962          * nbd_co_receive_request()
2963          */
2964         goto done;
2965     }
2966 
2967     if (ret == -EAGAIN) {
2968         assert(client->quiescing);
2969         goto done;
2970     }
2971 
2972     nbd_client_receive_next_request(client);
2973     if (ret == -EIO) {
2974         goto disconnect;
2975     }
2976 
2977     qio_channel_set_cork(client->ioc, true);
2978 
2979     if (ret < 0) {
2980         /* It wasn't -EIO, so, according to nbd_co_receive_request()
2981          * semantics, we should return the error to the client. */
2982         Error *export_err = local_err;
2983 
2984         local_err = NULL;
2985         ret = nbd_send_generic_reply(client, &request, -EINVAL,
2986                                      error_get_pretty(export_err), &local_err);
2987         error_free(export_err);
2988     } else {
2989         ret = nbd_handle_request(client, &request, req->data, &local_err);
2990     }
2991     if (request.contexts && request.contexts != &client->contexts) {
2992         assert(request.type == NBD_CMD_BLOCK_STATUS);
2993         g_free(request.contexts->bitmaps);
2994         g_free(request.contexts);
2995     }
2996     if (ret < 0) {
2997         error_prepend(&local_err, "Failed to send reply: ");
2998         goto disconnect;
2999     }
3000 
3001     /*
3002      * We must disconnect after NBD_CMD_WRITE or BLOCK_STATUS with
3003      * payload if we did not read the payload.
3004      */
3005     if (!req->complete) {
3006         error_setg(&local_err, "Request handling failed in intermediate state");
3007         goto disconnect;
3008     }
3009 
3010     qio_channel_set_cork(client->ioc, false);
3011 done:
3012     nbd_request_put(req);
3013     nbd_client_put(client);
3014     return;
3015 
3016 disconnect:
3017     if (local_err) {
3018         error_reportf_err(local_err, "Disconnect client, due to: ");
3019     }
3020     nbd_request_put(req);
3021     client_close(client, true);
3022     nbd_client_put(client);
3023 }
3024 
3025 static void nbd_client_receive_next_request(NBDClient *client)
3026 {
3027     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
3028         !client->quiescing) {
3029         nbd_client_get(client);
3030         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
3031         aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
3032     }
3033 }
3034 
3035 static coroutine_fn void nbd_co_client_start(void *opaque)
3036 {
3037     NBDClient *client = opaque;
3038     Error *local_err = NULL;
3039 
3040     qemu_co_mutex_init(&client->send_lock);
3041 
3042     if (nbd_negotiate(client, &local_err)) {
3043         if (local_err) {
3044             error_report_err(local_err);
3045         }
3046         client_close(client, false);
3047         return;
3048     }
3049 
3050     nbd_client_receive_next_request(client);
3051 }
3052 
3053 /*
3054  * Create a new client listener using the given channel @sioc.
3055  * Begin servicing it in a coroutine.  When the connection closes, call
3056  * @close_fn with an indication of whether the client completed negotiation.
3057  */
3058 void nbd_client_new(QIOChannelSocket *sioc,
3059                     QCryptoTLSCreds *tlscreds,
3060                     const char *tlsauthz,
3061                     void (*close_fn)(NBDClient *, bool))
3062 {
3063     NBDClient *client;
3064     Coroutine *co;
3065 
3066     client = g_new0(NBDClient, 1);
3067     client->refcount = 1;
3068     client->tlscreds = tlscreds;
3069     if (tlscreds) {
3070         object_ref(OBJECT(client->tlscreds));
3071     }
3072     client->tlsauthz = g_strdup(tlsauthz);
3073     client->sioc = sioc;
3074     qio_channel_set_delay(QIO_CHANNEL(sioc), false);
3075     object_ref(OBJECT(client->sioc));
3076     client->ioc = QIO_CHANNEL(sioc);
3077     object_ref(OBJECT(client->ioc));
3078     client->close_fn = close_fn;
3079 
3080     co = qemu_coroutine_create(nbd_co_client_start, client);
3081     qemu_coroutine_enter(co);
3082 }
3083