xref: /openbmc/qemu/nbd/server.c (revision 7200fb21)
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     GLOBAL_STATE_CODE();
1693     assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD);
1694 
1695     if (!nbd_server_is_running()) {
1696         error_setg(errp, "NBD server not running");
1697         return -EINVAL;
1698     }
1699 
1700     if (strlen(name) > NBD_MAX_STRING_SIZE) {
1701         error_setg(errp, "export name '%s' too long", name);
1702         return -EINVAL;
1703     }
1704 
1705     if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) {
1706         error_setg(errp, "description '%s' too long", arg->description);
1707         return -EINVAL;
1708     }
1709 
1710     if (nbd_export_find(name)) {
1711         error_setg(errp, "NBD server already has export named '%s'", name);
1712         return -EEXIST;
1713     }
1714 
1715     size = blk_getlength(blk);
1716     if (size < 0) {
1717         error_setg_errno(errp, -size,
1718                          "Failed to determine the NBD export's length");
1719         return size;
1720     }
1721 
1722     /* Don't allow resize while the NBD server is running, otherwise we don't
1723      * care what happens with the node. */
1724     blk_get_perm(blk, &perm, &shared_perm);
1725     ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp);
1726     if (ret < 0) {
1727         return ret;
1728     }
1729 
1730     QTAILQ_INIT(&exp->clients);
1731     exp->name = g_strdup(name);
1732     exp->description = g_strdup(arg->description);
1733     exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH |
1734                      NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE);
1735 
1736     if (nbd_server_max_connections() != 1) {
1737         exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN;
1738     }
1739     if (readonly) {
1740         exp->nbdflags |= NBD_FLAG_READ_ONLY;
1741     } else {
1742         exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES |
1743                           NBD_FLAG_SEND_FAST_ZERO);
1744     }
1745     exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE);
1746 
1747     bdrv_graph_rdlock_main_loop();
1748 
1749     for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) {
1750         exp->nr_export_bitmaps++;
1751     }
1752     exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps);
1753     for (i = 0, bitmaps = arg->bitmaps; bitmaps;
1754          i++, bitmaps = bitmaps->next)
1755     {
1756         const char *bitmap;
1757         BlockDriverState *bs = blk_bs(blk);
1758         BdrvDirtyBitmap *bm = NULL;
1759 
1760         switch (bitmaps->value->type) {
1761         case QTYPE_QSTRING:
1762             bitmap = bitmaps->value->u.local;
1763             while (bs) {
1764                 bm = bdrv_find_dirty_bitmap(bs, bitmap);
1765                 if (bm != NULL) {
1766                     break;
1767                 }
1768 
1769                 bs = bdrv_filter_or_cow_bs(bs);
1770             }
1771 
1772             if (bm == NULL) {
1773                 ret = -ENOENT;
1774                 error_setg(errp, "Bitmap '%s' is not found",
1775                            bitmaps->value->u.local);
1776                 goto fail;
1777             }
1778 
1779             if (readonly && bdrv_is_writable(bs) &&
1780                 bdrv_dirty_bitmap_enabled(bm)) {
1781                 ret = -EINVAL;
1782                 error_setg(errp, "Enabled bitmap '%s' incompatible with "
1783                            "readonly export", bitmap);
1784                 goto fail;
1785             }
1786             break;
1787         case QTYPE_QDICT:
1788             bitmap = bitmaps->value->u.external.name;
1789             bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node,
1790                                            bitmap, NULL, errp);
1791             if (!bm) {
1792                 ret = -ENOENT;
1793                 goto fail;
1794             }
1795             break;
1796         default:
1797             abort();
1798         }
1799 
1800         assert(bm);
1801 
1802         if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) {
1803             ret = -EINVAL;
1804             goto fail;
1805         }
1806 
1807         exp->export_bitmaps[i] = bm;
1808         assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE);
1809     }
1810 
1811     /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */
1812     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1813         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true);
1814     }
1815 
1816     exp->allocation_depth = arg->allocation_depth;
1817 
1818     /*
1819      * We need to inhibit request queuing in the block layer to ensure we can
1820      * be properly quiesced when entering a drained section, as our coroutines
1821      * servicing pending requests might enter blk_pread().
1822      */
1823     blk_set_disable_request_queuing(blk, true);
1824 
1825     blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp);
1826 
1827     blk_set_dev_ops(blk, &nbd_block_ops, exp);
1828 
1829     QTAILQ_INSERT_TAIL(&exports, exp, next);
1830 
1831     bdrv_graph_rdunlock_main_loop();
1832 
1833     return 0;
1834 
1835 fail:
1836     bdrv_graph_rdunlock_main_loop();
1837     g_free(exp->export_bitmaps);
1838     g_free(exp->name);
1839     g_free(exp->description);
1840     return ret;
1841 }
1842 
1843 NBDExport *nbd_export_find(const char *name)
1844 {
1845     NBDExport *exp;
1846     QTAILQ_FOREACH(exp, &exports, next) {
1847         if (strcmp(name, exp->name) == 0) {
1848             return exp;
1849         }
1850     }
1851 
1852     return NULL;
1853 }
1854 
1855 AioContext *
1856 nbd_export_aio_context(NBDExport *exp)
1857 {
1858     return exp->common.ctx;
1859 }
1860 
1861 static void nbd_export_request_shutdown(BlockExport *blk_exp)
1862 {
1863     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1864     NBDClient *client, *next;
1865 
1866     blk_exp_ref(&exp->common);
1867     /*
1868      * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a
1869      * close mode that stops advertising the export to new clients but
1870      * still permits existing clients to run to completion? Because of
1871      * that possibility, nbd_export_close() can be called more than
1872      * once on an export.
1873      */
1874     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
1875         client_close(client, true);
1876     }
1877     if (exp->name) {
1878         g_free(exp->name);
1879         exp->name = NULL;
1880         QTAILQ_REMOVE(&exports, exp, next);
1881     }
1882     blk_exp_unref(&exp->common);
1883 }
1884 
1885 static void nbd_export_delete(BlockExport *blk_exp)
1886 {
1887     size_t i;
1888     NBDExport *exp = container_of(blk_exp, NBDExport, common);
1889 
1890     assert(exp->name == NULL);
1891     assert(QTAILQ_EMPTY(&exp->clients));
1892 
1893     g_free(exp->description);
1894     exp->description = NULL;
1895 
1896     if (exp->eject_notifier_blk) {
1897         notifier_remove(&exp->eject_notifier);
1898         blk_unref(exp->eject_notifier_blk);
1899     }
1900     blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached,
1901                                     blk_aio_detach, exp);
1902     blk_set_disable_request_queuing(exp->common.blk, false);
1903 
1904     for (i = 0; i < exp->nr_export_bitmaps; i++) {
1905         bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false);
1906     }
1907 }
1908 
1909 const BlockExportDriver blk_exp_nbd = {
1910     .type               = BLOCK_EXPORT_TYPE_NBD,
1911     .instance_size      = sizeof(NBDExport),
1912     .create             = nbd_export_create,
1913     .delete             = nbd_export_delete,
1914     .request_shutdown   = nbd_export_request_shutdown,
1915 };
1916 
1917 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov,
1918                                         unsigned niov, Error **errp)
1919 {
1920     int ret;
1921 
1922     g_assert(qemu_in_coroutine());
1923     qemu_co_mutex_lock(&client->send_lock);
1924     client->send_coroutine = qemu_coroutine_self();
1925 
1926     ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0;
1927 
1928     client->send_coroutine = NULL;
1929     qemu_co_mutex_unlock(&client->send_lock);
1930 
1931     return ret;
1932 }
1933 
1934 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error,
1935                                        uint64_t cookie)
1936 {
1937     stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC);
1938     stl_be_p(&reply->error, error);
1939     stq_be_p(&reply->cookie, cookie);
1940 }
1941 
1942 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client,
1943                                                  NBDRequest *request,
1944                                                  uint32_t error,
1945                                                  void *data,
1946                                                  uint64_t len,
1947                                                  Error **errp)
1948 {
1949     NBDSimpleReply reply;
1950     int nbd_err = system_errno_to_nbd_errno(error);
1951     struct iovec iov[] = {
1952         {.iov_base = &reply, .iov_len = sizeof(reply)},
1953         {.iov_base = data, .iov_len = len}
1954     };
1955 
1956     assert(!len || !nbd_err);
1957     assert(len <= NBD_MAX_BUFFER_SIZE);
1958     assert(client->mode < NBD_MODE_STRUCTURED ||
1959            (client->mode == NBD_MODE_STRUCTURED &&
1960             request->type != NBD_CMD_READ));
1961     trace_nbd_co_send_simple_reply(request->cookie, nbd_err,
1962                                    nbd_err_lookup(nbd_err), len);
1963     set_be_simple_reply(&reply, nbd_err, request->cookie);
1964 
1965     return nbd_co_send_iov(client, iov, 2, errp);
1966 }
1967 
1968 /*
1969  * Prepare the header of a reply chunk for network transmission.
1970  *
1971  * On input, @iov is partially initialized: iov[0].iov_base must point
1972  * to an uninitialized NBDReply, while the remaining @niov elements
1973  * (if any) must be ready for transmission.  This function then
1974  * populates iov[0] for transmission.
1975  */
1976 static inline void set_be_chunk(NBDClient *client, struct iovec *iov,
1977                                 size_t niov, uint16_t flags, uint16_t type,
1978                                 NBDRequest *request)
1979 {
1980     size_t i, length = 0;
1981 
1982     for (i = 1; i < niov; i++) {
1983         length += iov[i].iov_len;
1984     }
1985     assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData));
1986 
1987     if (client->mode >= NBD_MODE_EXTENDED) {
1988         NBDExtendedReplyChunk *chunk = iov->iov_base;
1989 
1990         iov[0].iov_len = sizeof(*chunk);
1991         stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC);
1992         stw_be_p(&chunk->flags, flags);
1993         stw_be_p(&chunk->type, type);
1994         stq_be_p(&chunk->cookie, request->cookie);
1995         stq_be_p(&chunk->offset, request->from);
1996         stq_be_p(&chunk->length, length);
1997     } else {
1998         NBDStructuredReplyChunk *chunk = iov->iov_base;
1999 
2000         iov[0].iov_len = sizeof(*chunk);
2001         stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC);
2002         stw_be_p(&chunk->flags, flags);
2003         stw_be_p(&chunk->type, type);
2004         stq_be_p(&chunk->cookie, request->cookie);
2005         stl_be_p(&chunk->length, length);
2006     }
2007 }
2008 
2009 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client,
2010                                                NBDRequest *request,
2011                                                Error **errp)
2012 {
2013     NBDReply hdr;
2014     struct iovec iov[] = {
2015         {.iov_base = &hdr},
2016     };
2017 
2018     trace_nbd_co_send_chunk_done(request->cookie);
2019     set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE,
2020                  NBD_REPLY_TYPE_NONE, request);
2021     return nbd_co_send_iov(client, iov, 1, errp);
2022 }
2023 
2024 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client,
2025                                                NBDRequest *request,
2026                                                uint64_t offset,
2027                                                void *data,
2028                                                uint64_t size,
2029                                                bool final,
2030                                                Error **errp)
2031 {
2032     NBDReply hdr;
2033     NBDStructuredReadData chunk;
2034     struct iovec iov[] = {
2035         {.iov_base = &hdr},
2036         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2037         {.iov_base = data, .iov_len = size}
2038     };
2039 
2040     assert(size && size <= NBD_MAX_BUFFER_SIZE);
2041     trace_nbd_co_send_chunk_read(request->cookie, offset, data, size);
2042     set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0,
2043                  NBD_REPLY_TYPE_OFFSET_DATA, request);
2044     stq_be_p(&chunk.offset, offset);
2045 
2046     return nbd_co_send_iov(client, iov, 3, errp);
2047 }
2048 
2049 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client,
2050                                                 NBDRequest *request,
2051                                                 uint32_t error,
2052                                                 const char *msg,
2053                                                 Error **errp)
2054 {
2055     NBDReply hdr;
2056     NBDStructuredError chunk;
2057     int nbd_err = system_errno_to_nbd_errno(error);
2058     struct iovec iov[] = {
2059         {.iov_base = &hdr},
2060         {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2061         {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0},
2062     };
2063 
2064     assert(nbd_err);
2065     trace_nbd_co_send_chunk_error(request->cookie, nbd_err,
2066                                   nbd_err_lookup(nbd_err), msg ? msg : "");
2067     set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE,
2068                  NBD_REPLY_TYPE_ERROR, request);
2069     stl_be_p(&chunk.error, nbd_err);
2070     stw_be_p(&chunk.message_length, iov[2].iov_len);
2071 
2072     return nbd_co_send_iov(client, iov, 3, errp);
2073 }
2074 
2075 /* Do a sparse read and send the structured reply to the client.
2076  * Returns -errno if sending fails. blk_co_block_status_above() failure is
2077  * reported to the client, at which point this function succeeds.
2078  */
2079 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client,
2080                                                 NBDRequest *request,
2081                                                 uint64_t offset,
2082                                                 uint8_t *data,
2083                                                 uint64_t size,
2084                                                 Error **errp)
2085 {
2086     int ret = 0;
2087     NBDExport *exp = client->exp;
2088     size_t progress = 0;
2089 
2090     assert(size <= NBD_MAX_BUFFER_SIZE);
2091     while (progress < size) {
2092         int64_t pnum;
2093         int status = blk_co_block_status_above(exp->common.blk, NULL,
2094                                                offset + progress,
2095                                                size - progress, &pnum, NULL,
2096                                                NULL);
2097         bool final;
2098 
2099         if (status < 0) {
2100             char *msg = g_strdup_printf("unable to check for holes: %s",
2101                                         strerror(-status));
2102 
2103             ret = nbd_co_send_chunk_error(client, request, -status, msg, errp);
2104             g_free(msg);
2105             return ret;
2106         }
2107         assert(pnum && pnum <= size - progress);
2108         final = progress + pnum == size;
2109         if (status & BDRV_BLOCK_ZERO) {
2110             NBDReply hdr;
2111             NBDStructuredReadHole chunk;
2112             struct iovec iov[] = {
2113                 {.iov_base = &hdr},
2114                 {.iov_base = &chunk, .iov_len = sizeof(chunk)},
2115             };
2116 
2117             trace_nbd_co_send_chunk_read_hole(request->cookie,
2118                                               offset + progress, pnum);
2119             set_be_chunk(client, iov, 2,
2120                          final ? NBD_REPLY_FLAG_DONE : 0,
2121                          NBD_REPLY_TYPE_OFFSET_HOLE, request);
2122             stq_be_p(&chunk.offset, offset + progress);
2123             stl_be_p(&chunk.length, pnum);
2124             ret = nbd_co_send_iov(client, iov, 2, errp);
2125         } else {
2126             ret = blk_co_pread(exp->common.blk, offset + progress, pnum,
2127                                data + progress, 0);
2128             if (ret < 0) {
2129                 error_setg_errno(errp, -ret, "reading from file failed");
2130                 break;
2131             }
2132             ret = nbd_co_send_chunk_read(client, request, offset + progress,
2133                                          data + progress, pnum, final, errp);
2134         }
2135 
2136         if (ret < 0) {
2137             break;
2138         }
2139         progress += pnum;
2140     }
2141     return ret;
2142 }
2143 
2144 typedef struct NBDExtentArray {
2145     NBDExtent64 *extents;
2146     unsigned int nb_alloc;
2147     unsigned int count;
2148     uint64_t total_length;
2149     bool extended;
2150     bool can_add;
2151     bool converted_to_be;
2152 } NBDExtentArray;
2153 
2154 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc,
2155                                             NBDMode mode)
2156 {
2157     NBDExtentArray *ea = g_new0(NBDExtentArray, 1);
2158 
2159     assert(mode >= NBD_MODE_STRUCTURED);
2160     ea->nb_alloc = nb_alloc;
2161     ea->extents = g_new(NBDExtent64, nb_alloc);
2162     ea->extended = mode >= NBD_MODE_EXTENDED;
2163     ea->can_add = true;
2164 
2165     return ea;
2166 }
2167 
2168 static void nbd_extent_array_free(NBDExtentArray *ea)
2169 {
2170     g_free(ea->extents);
2171     g_free(ea);
2172 }
2173 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free)
2174 
2175 /* Further modifications of the array after conversion are abandoned */
2176 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea)
2177 {
2178     int i;
2179 
2180     assert(!ea->converted_to_be);
2181     assert(ea->extended);
2182     ea->can_add = false;
2183     ea->converted_to_be = true;
2184 
2185     for (i = 0; i < ea->count; i++) {
2186         ea->extents[i].length = cpu_to_be64(ea->extents[i].length);
2187         ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags);
2188     }
2189 }
2190 
2191 /* Further modifications of the array after conversion are abandoned */
2192 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea)
2193 {
2194     int i;
2195     NBDExtent32 *extents = g_new(NBDExtent32, ea->count);
2196 
2197     assert(!ea->converted_to_be);
2198     assert(!ea->extended);
2199     ea->can_add = false;
2200     ea->converted_to_be = true;
2201 
2202     for (i = 0; i < ea->count; i++) {
2203         assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX);
2204         extents[i].length = cpu_to_be32(ea->extents[i].length);
2205         extents[i].flags = cpu_to_be32(ea->extents[i].flags);
2206     }
2207 
2208     return extents;
2209 }
2210 
2211 /*
2212  * Add extent to NBDExtentArray. If extent can't be added (no available space),
2213  * return -1.
2214  * For safety, when returning -1 for the first time, .can_add is set to false,
2215  * and further calls to nbd_extent_array_add() will crash.
2216  * (this avoids the situation where a caller ignores failure to add one extent,
2217  * where adding another extent that would squash into the last array entry
2218  * would result in an incorrect range reported to the client)
2219  */
2220 static int nbd_extent_array_add(NBDExtentArray *ea,
2221                                 uint64_t length, uint32_t flags)
2222 {
2223     assert(ea->can_add);
2224 
2225     if (!length) {
2226         return 0;
2227     }
2228     if (!ea->extended) {
2229         assert(length <= UINT32_MAX);
2230     }
2231 
2232     /* Extend previous extent if flags are the same */
2233     if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) {
2234         uint64_t sum = length + ea->extents[ea->count - 1].length;
2235 
2236         /*
2237          * sum cannot overflow: the block layer bounds image size at
2238          * 2^63, and ea->extents[].length comes from the block layer.
2239          */
2240         assert(sum >= length);
2241         if (sum <= UINT32_MAX || ea->extended) {
2242             ea->extents[ea->count - 1].length = sum;
2243             ea->total_length += length;
2244             return 0;
2245         }
2246     }
2247 
2248     if (ea->count >= ea->nb_alloc) {
2249         ea->can_add = false;
2250         return -1;
2251     }
2252 
2253     ea->total_length += length;
2254     ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags};
2255     ea->count++;
2256 
2257     return 0;
2258 }
2259 
2260 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk,
2261                                                uint64_t offset, uint64_t bytes,
2262                                                NBDExtentArray *ea)
2263 {
2264     while (bytes) {
2265         uint32_t flags;
2266         int64_t num;
2267         int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num,
2268                                             NULL, NULL);
2269 
2270         if (ret < 0) {
2271             return ret;
2272         }
2273 
2274         flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) |
2275                 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0);
2276 
2277         if (nbd_extent_array_add(ea, num, flags) < 0) {
2278             return 0;
2279         }
2280 
2281         offset += num;
2282         bytes -= num;
2283     }
2284 
2285     return 0;
2286 }
2287 
2288 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk,
2289                                               uint64_t offset, uint64_t bytes,
2290                                               NBDExtentArray *ea)
2291 {
2292     while (bytes) {
2293         int64_t num;
2294         int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes,
2295                                             &num);
2296 
2297         if (ret < 0) {
2298             return ret;
2299         }
2300 
2301         if (nbd_extent_array_add(ea, num, ret) < 0) {
2302             return 0;
2303         }
2304 
2305         offset += num;
2306         bytes -= num;
2307     }
2308 
2309     return 0;
2310 }
2311 
2312 /*
2313  * nbd_co_send_extents
2314  *
2315  * @ea is converted to BE by the function
2316  * @last controls whether NBD_REPLY_FLAG_DONE is sent.
2317  */
2318 static int coroutine_fn
2319 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea,
2320                     bool last, uint32_t context_id, Error **errp)
2321 {
2322     NBDReply hdr;
2323     NBDStructuredMeta meta;
2324     NBDExtendedMeta meta_ext;
2325     g_autofree NBDExtent32 *extents = NULL;
2326     uint16_t type;
2327     struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} };
2328 
2329     if (client->mode >= NBD_MODE_EXTENDED) {
2330         type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT;
2331 
2332         iov[1].iov_base = &meta_ext;
2333         iov[1].iov_len = sizeof(meta_ext);
2334         stl_be_p(&meta_ext.context_id, context_id);
2335         stl_be_p(&meta_ext.count, ea->count);
2336 
2337         nbd_extent_array_convert_to_be(ea);
2338         iov[2].iov_base = ea->extents;
2339         iov[2].iov_len = ea->count * sizeof(ea->extents[0]);
2340     } else {
2341         type = NBD_REPLY_TYPE_BLOCK_STATUS;
2342 
2343         iov[1].iov_base = &meta;
2344         iov[1].iov_len = sizeof(meta);
2345         stl_be_p(&meta.context_id, context_id);
2346 
2347         extents = nbd_extent_array_convert_to_narrow(ea);
2348         iov[2].iov_base = extents;
2349         iov[2].iov_len = ea->count * sizeof(extents[0]);
2350     }
2351 
2352     trace_nbd_co_send_extents(request->cookie, ea->count, context_id,
2353                               ea->total_length, last);
2354     set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type,
2355                  request);
2356 
2357     return nbd_co_send_iov(client, iov, 3, errp);
2358 }
2359 
2360 /* Get block status from the exported device and send it to the client */
2361 static int
2362 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request,
2363                                       BlockBackend *blk, uint64_t offset,
2364                                       uint64_t length, bool dont_fragment,
2365                                       bool last, uint32_t context_id,
2366                                       Error **errp)
2367 {
2368     int ret;
2369     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2370     g_autoptr(NBDExtentArray) ea =
2371         nbd_extent_array_new(nb_extents, client->mode);
2372 
2373     if (context_id == NBD_META_ID_BASE_ALLOCATION) {
2374         ret = blockstatus_to_extents(blk, offset, length, ea);
2375     } else {
2376         ret = blockalloc_to_extents(blk, offset, length, ea);
2377     }
2378     if (ret < 0) {
2379         return nbd_co_send_chunk_error(client, request, -ret,
2380                                        "can't get block status", errp);
2381     }
2382 
2383     return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2384 }
2385 
2386 /* Populate @ea from a dirty bitmap. */
2387 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap,
2388                               uint64_t offset, uint64_t length,
2389                               NBDExtentArray *es)
2390 {
2391     int64_t start, dirty_start, dirty_count;
2392     int64_t end = offset + length;
2393     bool full = false;
2394     int64_t bound = es->extended ? INT64_MAX : INT32_MAX;
2395 
2396     bdrv_dirty_bitmap_lock(bitmap);
2397 
2398     for (start = offset;
2399          bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound,
2400                                            &dirty_start, &dirty_count);
2401          start = dirty_start + dirty_count)
2402     {
2403         if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) ||
2404             (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0))
2405         {
2406             full = true;
2407             break;
2408         }
2409     }
2410 
2411     if (!full) {
2412         /* last non dirty extent, nothing to do if array is now full */
2413         (void) nbd_extent_array_add(es, end - start, 0);
2414     }
2415 
2416     bdrv_dirty_bitmap_unlock(bitmap);
2417 }
2418 
2419 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client,
2420                                            NBDRequest *request,
2421                                            BdrvDirtyBitmap *bitmap,
2422                                            uint64_t offset,
2423                                            uint64_t length, bool dont_fragment,
2424                                            bool last, uint32_t context_id,
2425                                            Error **errp)
2426 {
2427     unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS;
2428     g_autoptr(NBDExtentArray) ea =
2429         nbd_extent_array_new(nb_extents, client->mode);
2430 
2431     bitmap_to_extents(bitmap, offset, length, ea);
2432 
2433     return nbd_co_send_extents(client, request, ea, last, context_id, errp);
2434 }
2435 
2436 /*
2437  * nbd_co_block_status_payload_read
2438  * Called when a client wants a subset of negotiated contexts via a
2439  * BLOCK_STATUS payload.  Check the payload for valid length and
2440  * contents.  On success, return 0 with request updated to effective
2441  * length.  If request was invalid but all payload consumed, return 0
2442  * with request->len and request->contexts->count set to 0 (which will
2443  * trigger an appropriate NBD_EINVAL response later on).  Return
2444  * negative errno if the payload was not fully consumed.
2445  */
2446 static int
2447 nbd_co_block_status_payload_read(NBDClient *client, NBDRequest *request,
2448                                  Error **errp)
2449 {
2450     uint64_t payload_len = request->len;
2451     g_autofree char *buf = NULL;
2452     size_t count, i, nr_bitmaps;
2453     uint32_t id;
2454 
2455     if (payload_len > NBD_MAX_BUFFER_SIZE) {
2456         error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2457                    request->len, NBD_MAX_BUFFER_SIZE);
2458         return -EINVAL;
2459     }
2460 
2461     assert(client->contexts.exp == client->exp);
2462     nr_bitmaps = client->exp->nr_export_bitmaps;
2463     request->contexts = g_new0(NBDMetaContexts, 1);
2464     request->contexts->exp = client->exp;
2465 
2466     if (payload_len % sizeof(uint32_t) ||
2467         payload_len < sizeof(NBDBlockStatusPayload) ||
2468         payload_len > (sizeof(NBDBlockStatusPayload) +
2469                        sizeof(id) * client->contexts.count)) {
2470         goto skip;
2471     }
2472 
2473     buf = g_malloc(payload_len);
2474     if (nbd_read(client->ioc, buf, payload_len,
2475                  "CMD_BLOCK_STATUS data", errp) < 0) {
2476         return -EIO;
2477     }
2478     trace_nbd_co_receive_request_payload_received(request->cookie,
2479                                                   payload_len);
2480     request->contexts->bitmaps = g_new0(bool, nr_bitmaps);
2481     count = (payload_len - sizeof(NBDBlockStatusPayload)) / sizeof(id);
2482     payload_len = 0;
2483 
2484     for (i = 0; i < count; i++) {
2485         id = ldl_be_p(buf + sizeof(NBDBlockStatusPayload) + sizeof(id) * i);
2486         if (id == NBD_META_ID_BASE_ALLOCATION) {
2487             if (!client->contexts.base_allocation ||
2488                 request->contexts->base_allocation) {
2489                 goto skip;
2490             }
2491             request->contexts->base_allocation = true;
2492         } else if (id == NBD_META_ID_ALLOCATION_DEPTH) {
2493             if (!client->contexts.allocation_depth ||
2494                 request->contexts->allocation_depth) {
2495                 goto skip;
2496             }
2497             request->contexts->allocation_depth = true;
2498         } else {
2499             unsigned idx = id - NBD_META_ID_DIRTY_BITMAP;
2500 
2501             if (idx >= nr_bitmaps || !client->contexts.bitmaps[idx] ||
2502                 request->contexts->bitmaps[idx]) {
2503                 goto skip;
2504             }
2505             request->contexts->bitmaps[idx] = true;
2506         }
2507     }
2508 
2509     request->len = ldq_be_p(buf);
2510     request->contexts->count = count;
2511     return 0;
2512 
2513  skip:
2514     trace_nbd_co_receive_block_status_payload_compliance(request->from,
2515                                                          request->len);
2516     request->len = request->contexts->count = 0;
2517     return nbd_drop(client->ioc, payload_len, errp);
2518 }
2519 
2520 /* nbd_co_receive_request
2521  * Collect a client request. Return 0 if request looks valid, -EIO to drop
2522  * connection right away, -EAGAIN to indicate we were interrupted and the
2523  * channel should be quiesced, and any other negative value to report an error
2524  * to the client (although the caller may still need to disconnect after
2525  * reporting the error).
2526  */
2527 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req,
2528                                                NBDRequest *request,
2529                                                Error **errp)
2530 {
2531     NBDClient *client = req->client;
2532     bool extended_with_payload;
2533     bool check_length = false;
2534     bool check_rofs = false;
2535     bool allocate_buffer = false;
2536     bool payload_okay = false;
2537     uint64_t payload_len = 0;
2538     int valid_flags = NBD_CMD_FLAG_FUA;
2539     int ret;
2540 
2541     g_assert(qemu_in_coroutine());
2542     assert(client->recv_coroutine == qemu_coroutine_self());
2543     ret = nbd_receive_request(client, request, errp);
2544     if (ret < 0) {
2545         return ret;
2546     }
2547 
2548     trace_nbd_co_receive_request_decode_type(request->cookie, request->type,
2549                                              nbd_cmd_lookup(request->type));
2550     extended_with_payload = client->mode >= NBD_MODE_EXTENDED &&
2551         request->flags & NBD_CMD_FLAG_PAYLOAD_LEN;
2552     if (extended_with_payload) {
2553         payload_len = request->len;
2554         check_length = true;
2555     }
2556 
2557     switch (request->type) {
2558     case NBD_CMD_DISC:
2559         /* Special case: we're going to disconnect without a reply,
2560          * whether or not flags, from, or len are bogus */
2561         req->complete = true;
2562         return -EIO;
2563 
2564     case NBD_CMD_READ:
2565         if (client->mode >= NBD_MODE_STRUCTURED) {
2566             valid_flags |= NBD_CMD_FLAG_DF;
2567         }
2568         check_length = true;
2569         allocate_buffer = true;
2570         break;
2571 
2572     case NBD_CMD_WRITE:
2573         if (client->mode >= NBD_MODE_EXTENDED) {
2574             if (!extended_with_payload) {
2575                 /* The client is noncompliant. Trace it, but proceed. */
2576                 trace_nbd_co_receive_ext_payload_compliance(request->from,
2577                                                             request->len);
2578             }
2579             valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2580         }
2581         payload_okay = true;
2582         payload_len = request->len;
2583         check_length = true;
2584         allocate_buffer = true;
2585         check_rofs = true;
2586         break;
2587 
2588     case NBD_CMD_FLUSH:
2589         break;
2590 
2591     case NBD_CMD_TRIM:
2592         check_rofs = true;
2593         break;
2594 
2595     case NBD_CMD_CACHE:
2596         check_length = true;
2597         break;
2598 
2599     case NBD_CMD_WRITE_ZEROES:
2600         valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO;
2601         check_rofs = true;
2602         break;
2603 
2604     case NBD_CMD_BLOCK_STATUS:
2605         if (extended_with_payload) {
2606             ret = nbd_co_block_status_payload_read(client, request, errp);
2607             if (ret < 0) {
2608                 return ret;
2609             }
2610             /* payload now consumed */
2611             check_length = false;
2612             payload_len = 0;
2613             valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN;
2614         } else {
2615             request->contexts = &client->contexts;
2616         }
2617         valid_flags |= NBD_CMD_FLAG_REQ_ONE;
2618         break;
2619 
2620     default:
2621         /* Unrecognized, will fail later */
2622         ;
2623     }
2624 
2625     /* Payload and buffer handling. */
2626     if (!payload_len) {
2627         req->complete = true;
2628     }
2629     if (check_length && request->len > NBD_MAX_BUFFER_SIZE) {
2630         /* READ, WRITE, CACHE */
2631         error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)",
2632                    request->len, NBD_MAX_BUFFER_SIZE);
2633         return -EINVAL;
2634     }
2635     if (payload_len && !payload_okay) {
2636         /*
2637          * For now, we don't support payloads on other commands; but
2638          * we can keep the connection alive by ignoring the payload.
2639          * We will fail the command later with NBD_EINVAL for the use
2640          * of an unsupported flag (and not for access beyond bounds).
2641          */
2642         assert(request->type != NBD_CMD_WRITE);
2643         request->len = 0;
2644     }
2645     if (allocate_buffer) {
2646         /* READ, WRITE */
2647         req->data = blk_try_blockalign(client->exp->common.blk,
2648                                        request->len);
2649         if (req->data == NULL) {
2650             error_setg(errp, "No memory");
2651             return -ENOMEM;
2652         }
2653     }
2654     if (payload_len) {
2655         if (payload_okay) {
2656             /* WRITE */
2657             assert(req->data);
2658             ret = nbd_read(client->ioc, req->data, payload_len,
2659                            "CMD_WRITE data", errp);
2660         } else {
2661             ret = nbd_drop(client->ioc, payload_len, errp);
2662         }
2663         if (ret < 0) {
2664             return -EIO;
2665         }
2666         req->complete = true;
2667         trace_nbd_co_receive_request_payload_received(request->cookie,
2668                                                       payload_len);
2669     }
2670 
2671     /* Sanity checks. */
2672     if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) {
2673         /* WRITE, TRIM, WRITE_ZEROES */
2674         error_setg(errp, "Export is read-only");
2675         return -EROFS;
2676     }
2677     if (request->from > client->exp->size ||
2678         request->len > client->exp->size - request->from) {
2679         error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64
2680                    ", Size: %" PRIu64, request->from, request->len,
2681                    client->exp->size);
2682         return (request->type == NBD_CMD_WRITE ||
2683                 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL;
2684     }
2685     if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len,
2686                                                 client->check_align)) {
2687         /*
2688          * The block layer gracefully handles unaligned requests, but
2689          * it's still worth tracing client non-compliance
2690          */
2691         trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type),
2692                                               request->from,
2693                                               request->len,
2694                                               client->check_align);
2695     }
2696     if (request->flags & ~valid_flags) {
2697         error_setg(errp, "unsupported flags for command %s (got 0x%x)",
2698                    nbd_cmd_lookup(request->type), request->flags);
2699         return -EINVAL;
2700     }
2701 
2702     return 0;
2703 }
2704 
2705 /* Send simple reply without a payload, or a structured error
2706  * @error_msg is ignored if @ret >= 0
2707  * Returns 0 if connection is still live, -errno on failure to talk to client
2708  */
2709 static coroutine_fn int nbd_send_generic_reply(NBDClient *client,
2710                                                NBDRequest *request,
2711                                                int ret,
2712                                                const char *error_msg,
2713                                                Error **errp)
2714 {
2715     if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) {
2716         return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp);
2717     } else if (client->mode >= NBD_MODE_EXTENDED) {
2718         return nbd_co_send_chunk_done(client, request, errp);
2719     } else {
2720         return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0,
2721                                         NULL, 0, errp);
2722     }
2723 }
2724 
2725 /* Handle NBD_CMD_READ request.
2726  * Return -errno if sending fails. Other errors are reported directly to the
2727  * client as an error reply. */
2728 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request,
2729                                         uint8_t *data, Error **errp)
2730 {
2731     int ret;
2732     NBDExport *exp = client->exp;
2733 
2734     assert(request->type == NBD_CMD_READ);
2735     assert(request->len <= NBD_MAX_BUFFER_SIZE);
2736 
2737     /* XXX: NBD Protocol only documents use of FUA with WRITE */
2738     if (request->flags & NBD_CMD_FLAG_FUA) {
2739         ret = blk_co_flush(exp->common.blk);
2740         if (ret < 0) {
2741             return nbd_send_generic_reply(client, request, ret,
2742                                           "flush failed", errp);
2743         }
2744     }
2745 
2746     if (client->mode >= NBD_MODE_STRUCTURED &&
2747         !(request->flags & NBD_CMD_FLAG_DF) && request->len)
2748     {
2749         return nbd_co_send_sparse_read(client, request, request->from,
2750                                        data, request->len, errp);
2751     }
2752 
2753     ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0);
2754     if (ret < 0) {
2755         return nbd_send_generic_reply(client, request, ret,
2756                                       "reading from file failed", errp);
2757     }
2758 
2759     if (client->mode >= NBD_MODE_STRUCTURED) {
2760         if (request->len) {
2761             return nbd_co_send_chunk_read(client, request, request->from, data,
2762                                           request->len, true, errp);
2763         } else {
2764             return nbd_co_send_chunk_done(client, request, errp);
2765         }
2766     } else {
2767         return nbd_co_send_simple_reply(client, request, 0,
2768                                         data, request->len, errp);
2769     }
2770 }
2771 
2772 /*
2773  * nbd_do_cmd_cache
2774  *
2775  * Handle NBD_CMD_CACHE request.
2776  * Return -errno if sending fails. Other errors are reported directly to the
2777  * client as an error reply.
2778  */
2779 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request,
2780                                          Error **errp)
2781 {
2782     int ret;
2783     NBDExport *exp = client->exp;
2784 
2785     assert(request->type == NBD_CMD_CACHE);
2786     assert(request->len <= NBD_MAX_BUFFER_SIZE);
2787 
2788     ret = blk_co_preadv(exp->common.blk, request->from, request->len,
2789                         NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
2790 
2791     return nbd_send_generic_reply(client, request, ret,
2792                                   "caching data failed", errp);
2793 }
2794 
2795 /* Handle NBD request.
2796  * Return -errno if sending fails. Other errors are reported directly to the
2797  * client as an error reply. */
2798 static coroutine_fn int nbd_handle_request(NBDClient *client,
2799                                            NBDRequest *request,
2800                                            uint8_t *data, Error **errp)
2801 {
2802     int ret;
2803     int flags;
2804     NBDExport *exp = client->exp;
2805     char *msg;
2806     size_t i;
2807 
2808     switch (request->type) {
2809     case NBD_CMD_CACHE:
2810         return nbd_do_cmd_cache(client, request, errp);
2811 
2812     case NBD_CMD_READ:
2813         return nbd_do_cmd_read(client, request, data, errp);
2814 
2815     case NBD_CMD_WRITE:
2816         flags = 0;
2817         if (request->flags & NBD_CMD_FLAG_FUA) {
2818             flags |= BDRV_REQ_FUA;
2819         }
2820         assert(request->len <= NBD_MAX_BUFFER_SIZE);
2821         ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data,
2822                             flags);
2823         return nbd_send_generic_reply(client, request, ret,
2824                                       "writing to file failed", errp);
2825 
2826     case NBD_CMD_WRITE_ZEROES:
2827         flags = 0;
2828         if (request->flags & NBD_CMD_FLAG_FUA) {
2829             flags |= BDRV_REQ_FUA;
2830         }
2831         if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) {
2832             flags |= BDRV_REQ_MAY_UNMAP;
2833         }
2834         if (request->flags & NBD_CMD_FLAG_FAST_ZERO) {
2835             flags |= BDRV_REQ_NO_FALLBACK;
2836         }
2837         ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len,
2838                                    flags);
2839         return nbd_send_generic_reply(client, request, ret,
2840                                       "writing to file failed", errp);
2841 
2842     case NBD_CMD_DISC:
2843         /* unreachable, thanks to special case in nbd_co_receive_request() */
2844         abort();
2845 
2846     case NBD_CMD_FLUSH:
2847         ret = blk_co_flush(exp->common.blk);
2848         return nbd_send_generic_reply(client, request, ret,
2849                                       "flush failed", errp);
2850 
2851     case NBD_CMD_TRIM:
2852         ret = blk_co_pdiscard(exp->common.blk, request->from, request->len);
2853         if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) {
2854             ret = blk_co_flush(exp->common.blk);
2855         }
2856         return nbd_send_generic_reply(client, request, ret,
2857                                       "discard failed", errp);
2858 
2859     case NBD_CMD_BLOCK_STATUS:
2860         assert(request->contexts);
2861         assert(client->mode >= NBD_MODE_EXTENDED ||
2862                request->len <= UINT32_MAX);
2863         if (request->contexts->count) {
2864             bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE;
2865             int contexts_remaining = request->contexts->count;
2866 
2867             if (!request->len) {
2868                 return nbd_send_generic_reply(client, request, -EINVAL,
2869                                               "need non-zero length", errp);
2870             }
2871             if (request->contexts->base_allocation) {
2872                 ret = nbd_co_send_block_status(client, request,
2873                                                exp->common.blk,
2874                                                request->from,
2875                                                request->len, dont_fragment,
2876                                                !--contexts_remaining,
2877                                                NBD_META_ID_BASE_ALLOCATION,
2878                                                errp);
2879                 if (ret < 0) {
2880                     return ret;
2881                 }
2882             }
2883 
2884             if (request->contexts->allocation_depth) {
2885                 ret = nbd_co_send_block_status(client, request,
2886                                                exp->common.blk,
2887                                                request->from, request->len,
2888                                                dont_fragment,
2889                                                !--contexts_remaining,
2890                                                NBD_META_ID_ALLOCATION_DEPTH,
2891                                                errp);
2892                 if (ret < 0) {
2893                     return ret;
2894                 }
2895             }
2896 
2897             assert(request->contexts->exp == client->exp);
2898             for (i = 0; i < client->exp->nr_export_bitmaps; i++) {
2899                 if (!request->contexts->bitmaps[i]) {
2900                     continue;
2901                 }
2902                 ret = nbd_co_send_bitmap(client, request,
2903                                          client->exp->export_bitmaps[i],
2904                                          request->from, request->len,
2905                                          dont_fragment, !--contexts_remaining,
2906                                          NBD_META_ID_DIRTY_BITMAP + i, errp);
2907                 if (ret < 0) {
2908                     return ret;
2909                 }
2910             }
2911 
2912             assert(!contexts_remaining);
2913 
2914             return 0;
2915         } else if (client->contexts.count) {
2916             return nbd_send_generic_reply(client, request, -EINVAL,
2917                                           "CMD_BLOCK_STATUS payload not valid",
2918                                           errp);
2919         } else {
2920             return nbd_send_generic_reply(client, request, -EINVAL,
2921                                           "CMD_BLOCK_STATUS not negotiated",
2922                                           errp);
2923         }
2924 
2925     default:
2926         msg = g_strdup_printf("invalid request type (%" PRIu32 ") received",
2927                               request->type);
2928         ret = nbd_send_generic_reply(client, request, -EINVAL, msg,
2929                                      errp);
2930         g_free(msg);
2931         return ret;
2932     }
2933 }
2934 
2935 /* Owns a reference to the NBDClient passed as opaque.  */
2936 static coroutine_fn void nbd_trip(void *opaque)
2937 {
2938     NBDClient *client = opaque;
2939     NBDRequestData *req;
2940     NBDRequest request = { 0 };    /* GCC thinks it can be used uninitialized */
2941     int ret;
2942     Error *local_err = NULL;
2943 
2944     trace_nbd_trip();
2945     if (client->closing) {
2946         nbd_client_put(client);
2947         return;
2948     }
2949 
2950     if (client->quiescing) {
2951         /*
2952          * We're switching between AIO contexts. Don't attempt to receive a new
2953          * request and kick the main context which may be waiting for us.
2954          */
2955         nbd_client_put(client);
2956         client->recv_coroutine = NULL;
2957         aio_wait_kick();
2958         return;
2959     }
2960 
2961     req = nbd_request_get(client);
2962     ret = nbd_co_receive_request(req, &request, &local_err);
2963     client->recv_coroutine = NULL;
2964 
2965     if (client->closing) {
2966         /*
2967          * The client may be closed when we are blocked in
2968          * nbd_co_receive_request()
2969          */
2970         goto done;
2971     }
2972 
2973     if (ret == -EAGAIN) {
2974         assert(client->quiescing);
2975         goto done;
2976     }
2977 
2978     nbd_client_receive_next_request(client);
2979     if (ret == -EIO) {
2980         goto disconnect;
2981     }
2982 
2983     qio_channel_set_cork(client->ioc, true);
2984 
2985     if (ret < 0) {
2986         /* It wasn't -EIO, so, according to nbd_co_receive_request()
2987          * semantics, we should return the error to the client. */
2988         Error *export_err = local_err;
2989 
2990         local_err = NULL;
2991         ret = nbd_send_generic_reply(client, &request, -EINVAL,
2992                                      error_get_pretty(export_err), &local_err);
2993         error_free(export_err);
2994     } else {
2995         ret = nbd_handle_request(client, &request, req->data, &local_err);
2996     }
2997     if (request.contexts && request.contexts != &client->contexts) {
2998         assert(request.type == NBD_CMD_BLOCK_STATUS);
2999         g_free(request.contexts->bitmaps);
3000         g_free(request.contexts);
3001     }
3002     if (ret < 0) {
3003         error_prepend(&local_err, "Failed to send reply: ");
3004         goto disconnect;
3005     }
3006 
3007     /*
3008      * We must disconnect after NBD_CMD_WRITE or BLOCK_STATUS with
3009      * payload if we did not read the payload.
3010      */
3011     if (!req->complete) {
3012         error_setg(&local_err, "Request handling failed in intermediate state");
3013         goto disconnect;
3014     }
3015 
3016     qio_channel_set_cork(client->ioc, false);
3017 done:
3018     nbd_request_put(req);
3019     nbd_client_put(client);
3020     return;
3021 
3022 disconnect:
3023     if (local_err) {
3024         error_reportf_err(local_err, "Disconnect client, due to: ");
3025     }
3026     nbd_request_put(req);
3027     client_close(client, true);
3028     nbd_client_put(client);
3029 }
3030 
3031 static void nbd_client_receive_next_request(NBDClient *client)
3032 {
3033     if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS &&
3034         !client->quiescing) {
3035         nbd_client_get(client);
3036         client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
3037         aio_co_schedule(client->exp->common.ctx, client->recv_coroutine);
3038     }
3039 }
3040 
3041 static coroutine_fn void nbd_co_client_start(void *opaque)
3042 {
3043     NBDClient *client = opaque;
3044     Error *local_err = NULL;
3045 
3046     qemu_co_mutex_init(&client->send_lock);
3047 
3048     if (nbd_negotiate(client, &local_err)) {
3049         if (local_err) {
3050             error_report_err(local_err);
3051         }
3052         client_close(client, false);
3053         return;
3054     }
3055 
3056     nbd_client_receive_next_request(client);
3057 }
3058 
3059 /*
3060  * Create a new client listener using the given channel @sioc.
3061  * Begin servicing it in a coroutine.  When the connection closes, call
3062  * @close_fn with an indication of whether the client completed negotiation.
3063  */
3064 void nbd_client_new(QIOChannelSocket *sioc,
3065                     QCryptoTLSCreds *tlscreds,
3066                     const char *tlsauthz,
3067                     void (*close_fn)(NBDClient *, bool))
3068 {
3069     NBDClient *client;
3070     Coroutine *co;
3071 
3072     client = g_new0(NBDClient, 1);
3073     client->refcount = 1;
3074     client->tlscreds = tlscreds;
3075     if (tlscreds) {
3076         object_ref(OBJECT(client->tlscreds));
3077     }
3078     client->tlsauthz = g_strdup(tlsauthz);
3079     client->sioc = sioc;
3080     qio_channel_set_delay(QIO_CHANNEL(sioc), false);
3081     object_ref(OBJECT(client->sioc));
3082     client->ioc = QIO_CHANNEL(sioc);
3083     object_ref(OBJECT(client->ioc));
3084     client->close_fn = close_fn;
3085 
3086     co = qemu_coroutine_create(nbd_co_client_start, client);
3087     qemu_coroutine_enter(co);
3088 }
3089