]>
Commit | Line | Data |
---|---|---|
75818250 | 1 | /* |
a7c8ed36 | 2 | * Copyright Red Hat |
7a5ca864 FB |
3 | * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws> |
4 | * | |
798bfe00 | 5 | * Network Block Device Server Side |
7a5ca864 FB |
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 | |
8167ee88 | 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
75818250 | 18 | */ |
7a5ca864 | 19 | |
d38ea87a | 20 | #include "qemu/osdep.h" |
56ee8626 | 21 | |
e2c1c34f | 22 | #include "block/block_int.h" |
56ee8626 | 23 | #include "block/export.h" |
e2c1c34f | 24 | #include "block/dirty-bitmap.h" |
da34e65c | 25 | #include "qapi/error.h" |
dc5e9ac7 | 26 | #include "qemu/queue.h" |
9588463e | 27 | #include "trace.h" |
798bfe00 | 28 | #include "nbd-internal.h" |
416e34bd | 29 | #include "qemu/units.h" |
5df022cf | 30 | #include "qemu/memalign.h" |
ca441480 | 31 | |
e7b1948d | 32 | #define NBD_META_ID_BASE_ALLOCATION 0 |
71719cd5 | 33 | #define NBD_META_ID_ALLOCATION_DEPTH 1 |
3b1f244c | 34 | /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */ |
71719cd5 | 35 | #define NBD_META_ID_DIRTY_BITMAP 2 |
3d068aff | 36 | |
416e34bd EB |
37 | /* |
38 | * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical | |
3d068aff VSO |
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 | |
416e34bd EB |
41 | * the reply as a denial of service attack. |
42 | */ | |
43 | #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8) | |
e7b1948d | 44 | |
ca441480 PB |
45 | static int system_errno_to_nbd_errno(int err) |
46 | { | |
47 | switch (err) { | |
48 | case 0: | |
49 | return NBD_SUCCESS; | |
50 | case EPERM: | |
c0301fcc | 51 | case EROFS: |
ca441480 PB |
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; | |
bae245d1 EB |
63 | case EOVERFLOW: |
64 | return NBD_EOVERFLOW; | |
0a479545 EB |
65 | case ENOTSUP: |
66 | #if ENOTSUP != EOPNOTSUPP | |
67 | case EOPNOTSUPP: | |
68 | #endif | |
69 | return NBD_ENOTSUP; | |
b6f5d3b5 EB |
70 | case ESHUTDOWN: |
71 | return NBD_ESHUTDOWN; | |
ca441480 PB |
72 | case EINVAL: |
73 | default: | |
74 | return NBD_EINVAL; | |
75 | } | |
76 | } | |
77 | ||
9a304d29 PB |
78 | /* Definitions for opaque data types */ |
79 | ||
315f78ab | 80 | typedef struct NBDRequestData NBDRequestData; |
9a304d29 | 81 | |
315f78ab | 82 | struct NBDRequestData { |
9a304d29 PB |
83 | NBDClient *client; |
84 | uint8_t *data; | |
29b6c3b3 | 85 | bool complete; |
9a304d29 PB |
86 | }; |
87 | ||
88 | struct NBDExport { | |
56ee8626 | 89 | BlockExport common; |
0ddf08db | 90 | |
ee0a19ec | 91 | char *name; |
b1a75b33 | 92 | char *description; |
9d26dfcb | 93 | uint64_t size; |
7423f417 | 94 | uint16_t nbdflags; |
4b9441f6 | 95 | QTAILQ_HEAD(, NBDClient) clients; |
ee0a19ec | 96 | QTAILQ_ENTRY(NBDExport) next; |
958c717d | 97 | |
cd7fca95 | 98 | BlockBackend *eject_notifier_blk; |
741cc431 | 99 | Notifier eject_notifier; |
3d068aff | 100 | |
71719cd5 | 101 | bool allocation_depth; |
3b1f244c EB |
102 | BdrvDirtyBitmap **export_bitmaps; |
103 | size_t nr_export_bitmaps; | |
9a304d29 PB |
104 | }; |
105 | ||
ee0a19ec PB |
106 | static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); |
107 | ||
e7b1948d VSO |
108 | /* NBDExportMetaContexts represents a list of contexts to be exported, |
109 | * as selected by NBD_OPT_SET_META_CONTEXT. Also used for | |
110 | * NBD_OPT_LIST_META_CONTEXT. */ | |
111 | typedef struct NBDExportMetaContexts { | |
af736e54 | 112 | NBDExport *exp; |
47ec485e | 113 | size_t count; /* number of negotiated contexts */ |
e7b1948d | 114 | bool base_allocation; /* export base:allocation context (block status) */ |
71719cd5 | 115 | bool allocation_depth; /* export qemu:allocation-depth */ |
3b1f244c EB |
116 | bool *bitmaps; /* |
117 | * export qemu:dirty-bitmap:<export bitmap name>, | |
118 | * sized by exp->nr_export_bitmaps | |
119 | */ | |
e7b1948d VSO |
120 | } NBDExportMetaContexts; |
121 | ||
9a304d29 PB |
122 | struct NBDClient { |
123 | int refcount; | |
0c9390d9 | 124 | void (*close_fn)(NBDClient *client, bool negotiated); |
9a304d29 PB |
125 | |
126 | NBDExport *exp; | |
f95910fe | 127 | QCryptoTLSCreds *tlscreds; |
b25e12da | 128 | char *tlsauthz; |
1c778ef7 DB |
129 | QIOChannelSocket *sioc; /* The underlying data channel */ |
130 | QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ | |
9a304d29 PB |
131 | |
132 | Coroutine *recv_coroutine; | |
133 | ||
134 | CoMutex send_lock; | |
135 | Coroutine *send_coroutine; | |
136 | ||
f148ae7d SL |
137 | bool read_yielding; |
138 | bool quiescing; | |
139 | ||
4b9441f6 | 140 | QTAILQ_ENTRY(NBDClient) next; |
9a304d29 | 141 | int nb_requests; |
ff2b68aa | 142 | bool closing; |
5c54e7fa | 143 | |
6e280648 EB |
144 | uint32_t check_align; /* If non-zero, check for aligned client requests */ |
145 | ||
ac132d05 | 146 | NBDMode mode; |
e7b1948d | 147 | NBDExportMetaContexts export_meta; |
9a304d29 | 148 | |
0cfae925 VSO |
149 | uint32_t opt; /* Current option being negotiated */ |
150 | uint32_t optlen; /* remaining length of data in ioc for the option being | |
151 | negotiated now */ | |
152 | }; | |
7a5ca864 | 153 | |
ff82911c | 154 | static void nbd_client_receive_next_request(NBDClient *client); |
958c717d | 155 | |
6b8c01e7 | 156 | /* Basic flow for negotiation |
7a5ca864 FB |
157 | |
158 | Server Client | |
7a5ca864 | 159 | Negotiate |
6b8c01e7 PB |
160 | |
161 | or | |
162 | ||
163 | Server Client | |
164 | Negotiate #1 | |
165 | Option | |
166 | Negotiate #2 | |
167 | ||
168 | ---- | |
169 | ||
170 | followed by | |
171 | ||
172 | Server Client | |
7a5ca864 FB |
173 | Request |
174 | Response | |
175 | Request | |
176 | Response | |
177 | ... | |
178 | ... | |
179 | Request (type == 2) | |
6b8c01e7 | 180 | |
7a5ca864 FB |
181 | */ |
182 | ||
1d17922a VSO |
183 | static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, |
184 | uint32_t type, uint32_t length) | |
185 | { | |
186 | stq_be_p(&rep->magic, NBD_REP_MAGIC); | |
187 | stl_be_p(&rep->option, option); | |
188 | stl_be_p(&rep->type, type); | |
189 | stl_be_p(&rep->length, length); | |
190 | } | |
191 | ||
526e5c65 EB |
192 | /* Send a reply header, including length, but no payload. |
193 | * Return -errno on error, 0 on success. */ | |
0cfae925 VSO |
194 | static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, |
195 | uint32_t len, Error **errp) | |
6b8c01e7 | 196 | { |
1d17922a | 197 | NBDOptionReply rep; |
6b8c01e7 | 198 | |
1d17922a | 199 | trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), |
3736cc5b | 200 | type, nbd_rep_lookup(type), len); |
f95910fe | 201 | |
f37708f6 | 202 | assert(len < NBD_MAX_BUFFER_SIZE); |
2fd2c840 | 203 | |
1d17922a VSO |
204 | set_be_option_rep(&rep, client->opt, type, len); |
205 | return nbd_write(client->ioc, &rep, sizeof(rep), errp); | |
f5076b5a | 206 | } |
6b8c01e7 | 207 | |
526e5c65 EB |
208 | /* Send a reply header with default 0 length. |
209 | * Return -errno on error, 0 on success. */ | |
0cfae925 | 210 | static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type, |
2fd2c840 | 211 | Error **errp) |
526e5c65 | 212 | { |
0cfae925 | 213 | return nbd_negotiate_send_rep_len(client, type, 0, errp); |
526e5c65 EB |
214 | } |
215 | ||
36683283 EB |
216 | /* Send an error reply. |
217 | * Return -errno on error, 0 on success. */ | |
9edc6313 | 218 | static int G_GNUC_PRINTF(4, 0) |
41f5dfaf EB |
219 | nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, |
220 | Error **errp, const char *fmt, va_list va) | |
36683283 | 221 | { |
795d946d | 222 | ERRP_GUARD(); |
df18c04e | 223 | g_autofree char *msg = NULL; |
36683283 EB |
224 | int ret; |
225 | size_t len; | |
226 | ||
36683283 | 227 | msg = g_strdup_vprintf(fmt, va); |
36683283 | 228 | len = strlen(msg); |
5c4fe018 | 229 | assert(len < NBD_MAX_STRING_SIZE); |
9588463e | 230 | trace_nbd_negotiate_send_rep_err(msg); |
0cfae925 | 231 | ret = nbd_negotiate_send_rep_len(client, type, len, errp); |
36683283 | 232 | if (ret < 0) { |
df18c04e | 233 | return ret; |
36683283 | 234 | } |
0cfae925 | 235 | if (nbd_write(client->ioc, msg, len, errp) < 0) { |
2fd2c840 | 236 | error_prepend(errp, "write failed (error message): "); |
df18c04e | 237 | return -EIO; |
36683283 | 238 | } |
2fd2c840 | 239 | |
df18c04e | 240 | return 0; |
36683283 EB |
241 | } |
242 | ||
5c4fe018 EB |
243 | /* |
244 | * Return a malloc'd copy of @name suitable for use in an error reply. | |
245 | */ | |
246 | static char * | |
247 | nbd_sanitize_name(const char *name) | |
248 | { | |
249 | if (strnlen(name, 80) < 80) { | |
250 | return g_strdup(name); | |
251 | } | |
252 | /* XXX Should we also try to sanitize any control characters? */ | |
253 | return g_strdup_printf("%.80s...", name); | |
254 | } | |
255 | ||
41f5dfaf EB |
256 | /* Send an error reply. |
257 | * Return -errno on error, 0 on success. */ | |
9edc6313 | 258 | static int G_GNUC_PRINTF(4, 5) |
41f5dfaf EB |
259 | nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, |
260 | Error **errp, const char *fmt, ...) | |
261 | { | |
262 | va_list va; | |
263 | int ret; | |
264 | ||
265 | va_start(va, fmt); | |
266 | ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); | |
267 | va_end(va); | |
268 | return ret; | |
269 | } | |
270 | ||
894e0280 EB |
271 | /* Drop remainder of the current option, and send a reply with the |
272 | * given error type and message. Return -errno on read or write | |
273 | * failure; or 0 if connection is still live. */ | |
9edc6313 | 274 | static int G_GNUC_PRINTF(4, 0) |
2e425fd5 VSO |
275 | nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp, |
276 | const char *fmt, va_list va) | |
894e0280 EB |
277 | { |
278 | int ret = nbd_drop(client->ioc, client->optlen, errp); | |
894e0280 EB |
279 | |
280 | client->optlen = 0; | |
281 | if (!ret) { | |
894e0280 | 282 | ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); |
894e0280 EB |
283 | } |
284 | return ret; | |
285 | } | |
286 | ||
9edc6313 | 287 | static int G_GNUC_PRINTF(4, 5) |
2e425fd5 VSO |
288 | nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, |
289 | const char *fmt, ...) | |
290 | { | |
291 | int ret; | |
292 | va_list va; | |
293 | ||
294 | va_start(va, fmt); | |
295 | ret = nbd_opt_vdrop(client, type, errp, fmt, va); | |
296 | va_end(va); | |
297 | ||
298 | return ret; | |
299 | } | |
300 | ||
9edc6313 | 301 | static int G_GNUC_PRINTF(3, 4) |
2e425fd5 VSO |
302 | nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...) |
303 | { | |
304 | int ret; | |
305 | va_list va; | |
306 | ||
307 | va_start(va, fmt); | |
308 | ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va); | |
309 | va_end(va); | |
310 | ||
311 | return ret; | |
312 | } | |
313 | ||
894e0280 | 314 | /* Read size bytes from the unparsed payload of the current option. |
d1e2c3e7 | 315 | * If @check_nul, require that no NUL bytes appear in buffer. |
894e0280 EB |
316 | * Return -errno on I/O error, 0 if option was completely handled by |
317 | * sending a reply about inconsistent lengths, or 1 on success. */ | |
318 | static int nbd_opt_read(NBDClient *client, void *buffer, size_t size, | |
d1e2c3e7 | 319 | bool check_nul, Error **errp) |
894e0280 EB |
320 | { |
321 | if (size > client->optlen) { | |
2e425fd5 VSO |
322 | return nbd_opt_invalid(client, errp, |
323 | "Inconsistent lengths in option %s", | |
324 | nbd_opt_lookup(client->opt)); | |
894e0280 EB |
325 | } |
326 | client->optlen -= size; | |
d1e2c3e7 EB |
327 | if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) { |
328 | return -EIO; | |
329 | } | |
330 | ||
331 | if (check_nul && strnlen(buffer, size) != size) { | |
332 | return nbd_opt_invalid(client, errp, | |
333 | "Unexpected embedded NUL in option %s", | |
334 | nbd_opt_lookup(client->opt)); | |
335 | } | |
336 | return 1; | |
894e0280 EB |
337 | } |
338 | ||
e7b1948d VSO |
339 | /* Drop size bytes from the unparsed payload of the current option. |
340 | * Return -errno on I/O error, 0 if option was completely handled by | |
341 | * sending a reply about inconsistent lengths, or 1 on success. */ | |
342 | static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp) | |
343 | { | |
344 | if (size > client->optlen) { | |
345 | return nbd_opt_invalid(client, errp, | |
346 | "Inconsistent lengths in option %s", | |
347 | nbd_opt_lookup(client->opt)); | |
348 | } | |
349 | client->optlen -= size; | |
350 | return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1; | |
351 | } | |
352 | ||
12296459 VSO |
353 | /* nbd_opt_read_name |
354 | * | |
355 | * Read a string with the format: | |
93676c88 | 356 | * uint32_t len (<= NBD_MAX_STRING_SIZE) |
12296459 VSO |
357 | * len bytes string (not 0-terminated) |
358 | * | |
9d7ab222 | 359 | * On success, @name will be allocated. |
12296459 VSO |
360 | * If @length is non-null, it will be set to the actual string length. |
361 | * | |
362 | * Return -errno on I/O error, 0 if option was completely handled by | |
363 | * sending a reply about inconsistent lengths, or 1 on success. | |
364 | */ | |
9d7ab222 | 365 | static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length, |
12296459 VSO |
366 | Error **errp) |
367 | { | |
368 | int ret; | |
369 | uint32_t len; | |
9d7ab222 | 370 | g_autofree char *local_name = NULL; |
12296459 | 371 | |
9d7ab222 | 372 | *name = NULL; |
d1e2c3e7 | 373 | ret = nbd_opt_read(client, &len, sizeof(len), false, errp); |
12296459 VSO |
374 | if (ret <= 0) { |
375 | return ret; | |
376 | } | |
80c7c2b0 | 377 | len = cpu_to_be32(len); |
12296459 | 378 | |
93676c88 | 379 | if (len > NBD_MAX_STRING_SIZE) { |
12296459 VSO |
380 | return nbd_opt_invalid(client, errp, |
381 | "Invalid name length: %" PRIu32, len); | |
382 | } | |
383 | ||
9d7ab222 | 384 | local_name = g_malloc(len + 1); |
d1e2c3e7 | 385 | ret = nbd_opt_read(client, local_name, len, true, errp); |
12296459 VSO |
386 | if (ret <= 0) { |
387 | return ret; | |
388 | } | |
9d7ab222 | 389 | local_name[len] = '\0'; |
12296459 VSO |
390 | |
391 | if (length) { | |
392 | *length = len; | |
393 | } | |
9d7ab222 | 394 | *name = g_steal_pointer(&local_name); |
12296459 VSO |
395 | |
396 | return 1; | |
397 | } | |
398 | ||
526e5c65 EB |
399 | /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. |
400 | * Return -errno on error, 0 on success. */ | |
0cfae925 | 401 | static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, |
2fd2c840 | 402 | Error **errp) |
32d7d2e0 | 403 | { |
795d946d | 404 | ERRP_GUARD(); |
b1a75b33 | 405 | size_t name_len, desc_len; |
526e5c65 | 406 | uint32_t len; |
b1a75b33 EB |
407 | const char *name = exp->name ? exp->name : ""; |
408 | const char *desc = exp->description ? exp->description : ""; | |
0cfae925 | 409 | QIOChannel *ioc = client->ioc; |
2e5c9ad6 | 410 | int ret; |
32d7d2e0 | 411 | |
9588463e | 412 | trace_nbd_negotiate_send_rep_list(name, desc); |
b1a75b33 EB |
413 | name_len = strlen(name); |
414 | desc_len = strlen(desc); | |
93676c88 | 415 | assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE); |
526e5c65 | 416 | len = name_len + desc_len + sizeof(len); |
0cfae925 | 417 | ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); |
2e5c9ad6 VSO |
418 | if (ret < 0) { |
419 | return ret; | |
32d7d2e0 | 420 | } |
526e5c65 | 421 | |
32d7d2e0 | 422 | len = cpu_to_be32(name_len); |
2fd2c840 VSO |
423 | if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { |
424 | error_prepend(errp, "write failed (name length): "); | |
b1a75b33 EB |
425 | return -EINVAL; |
426 | } | |
2fd2c840 VSO |
427 | |
428 | if (nbd_write(ioc, name, name_len, errp) < 0) { | |
429 | error_prepend(errp, "write failed (name buffer): "); | |
32d7d2e0 HB |
430 | return -EINVAL; |
431 | } | |
2fd2c840 VSO |
432 | |
433 | if (nbd_write(ioc, desc, desc_len, errp) < 0) { | |
434 | error_prepend(errp, "write failed (description buffer): "); | |
32d7d2e0 HB |
435 | return -EINVAL; |
436 | } | |
2fd2c840 | 437 | |
32d7d2e0 HB |
438 | return 0; |
439 | } | |
440 | ||
526e5c65 EB |
441 | /* Process the NBD_OPT_LIST command, with a potential series of replies. |
442 | * Return -errno on error, 0 on success. */ | |
e68c35cf | 443 | static int nbd_negotiate_handle_list(NBDClient *client, Error **errp) |
32d7d2e0 | 444 | { |
32d7d2e0 | 445 | NBDExport *exp; |
0cfae925 | 446 | assert(client->opt == NBD_OPT_LIST); |
32d7d2e0 | 447 | |
32d7d2e0 HB |
448 | /* For each export, send a NBD_REP_SERVER reply. */ |
449 | QTAILQ_FOREACH(exp, &exports, next) { | |
0cfae925 | 450 | if (nbd_negotiate_send_rep_list(client, exp, errp)) { |
32d7d2e0 HB |
451 | return -EINVAL; |
452 | } | |
453 | } | |
454 | /* Finish with a NBD_REP_ACK. */ | |
0cfae925 | 455 | return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); |
32d7d2e0 HB |
456 | } |
457 | ||
af736e54 | 458 | static void nbd_check_meta_export(NBDClient *client) |
e7b1948d | 459 | { |
47ec485e EB |
460 | if (client->exp != client->export_meta.exp) { |
461 | client->export_meta.count = 0; | |
462 | } | |
e7b1948d VSO |
463 | } |
464 | ||
f37708f6 EB |
465 | /* Send a reply to NBD_OPT_EXPORT_NAME. |
466 | * Return -errno on error, 0 on success. */ | |
dbb38caa | 467 | static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes, |
2fd2c840 | 468 | Error **errp) |
f5076b5a | 469 | { |
795d946d | 470 | ERRP_GUARD(); |
9d7ab222 | 471 | g_autofree char *name = NULL; |
5f66d060 | 472 | char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; |
23e099c3 EB |
473 | size_t len; |
474 | int ret; | |
dbb38caa | 475 | uint16_t myflags; |
6b8c01e7 | 476 | |
f5076b5a HB |
477 | /* Client sends: |
478 | [20 .. xx] export name (length bytes) | |
5f66d060 EB |
479 | Server replies: |
480 | [ 0 .. 7] size | |
481 | [ 8 .. 9] export flags | |
482 | [10 .. 133] reserved (0) [unless no_zeroes] | |
f5076b5a | 483 | */ |
9588463e | 484 | trace_nbd_negotiate_handle_export_name(); |
93676c88 | 485 | if (client->optlen > NBD_MAX_STRING_SIZE) { |
2fd2c840 | 486 | error_setg(errp, "Bad length received"); |
d9faeed8 | 487 | return -EINVAL; |
6b8c01e7 | 488 | } |
9d7ab222 | 489 | name = g_malloc(client->optlen + 1); |
e6798f06 | 490 | if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) { |
32f158a6 | 491 | return -EIO; |
6b8c01e7 | 492 | } |
0cfae925 VSO |
493 | name[client->optlen] = '\0'; |
494 | client->optlen = 0; | |
6b8c01e7 | 495 | |
9588463e | 496 | trace_nbd_negotiate_handle_export_name_request(name); |
9344e5f5 | 497 | |
6b8c01e7 PB |
498 | client->exp = nbd_export_find(name); |
499 | if (!client->exp) { | |
2fd2c840 | 500 | error_setg(errp, "export not found"); |
d9faeed8 | 501 | return -EINVAL; |
6b8c01e7 PB |
502 | } |
503 | ||
dbb38caa | 504 | myflags = client->exp->nbdflags; |
ac132d05 | 505 | if (client->mode >= NBD_MODE_STRUCTURED) { |
dbb38caa EB |
506 | myflags |= NBD_FLAG_SEND_DF; |
507 | } | |
508 | trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags); | |
23e099c3 | 509 | stq_be_p(buf, client->exp->size); |
dbb38caa | 510 | stw_be_p(buf + 8, myflags); |
23e099c3 EB |
511 | len = no_zeroes ? 10 : sizeof(buf); |
512 | ret = nbd_write(client->ioc, buf, len, errp); | |
513 | if (ret < 0) { | |
514 | error_prepend(errp, "write failed: "); | |
515 | return ret; | |
516 | } | |
517 | ||
6b8c01e7 | 518 | QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); |
c69de1be | 519 | blk_exp_ref(&client->exp->common); |
af736e54 | 520 | nbd_check_meta_export(client); |
d9faeed8 VSO |
521 | |
522 | return 0; | |
6b8c01e7 PB |
523 | } |
524 | ||
f37708f6 EB |
525 | /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. |
526 | * The buffer does NOT include the info type prefix. | |
527 | * Return -errno on error, 0 if ready to send more. */ | |
0cfae925 | 528 | static int nbd_negotiate_send_info(NBDClient *client, |
f37708f6 EB |
529 | uint16_t info, uint32_t length, void *buf, |
530 | Error **errp) | |
531 | { | |
532 | int rc; | |
533 | ||
534 | trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); | |
0cfae925 | 535 | rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, |
f37708f6 EB |
536 | sizeof(info) + length, errp); |
537 | if (rc < 0) { | |
538 | return rc; | |
539 | } | |
80c7c2b0 | 540 | info = cpu_to_be16(info); |
f37708f6 EB |
541 | if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { |
542 | return -EIO; | |
543 | } | |
544 | if (nbd_write(client->ioc, buf, length, errp) < 0) { | |
545 | return -EIO; | |
546 | } | |
547 | return 0; | |
548 | } | |
549 | ||
a16a7907 EB |
550 | /* nbd_reject_length: Handle any unexpected payload. |
551 | * @fatal requests that we quit talking to the client, even if we are able | |
552 | * to successfully send an error reply. | |
553 | * Return: | |
554 | * -errno transmission error occurred or @fatal was requested, errp is set | |
555 | * 0 error message successfully sent to client, errp is not set | |
556 | */ | |
0cfae925 | 557 | static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp) |
a16a7907 EB |
558 | { |
559 | int ret; | |
560 | ||
0cfae925 | 561 | assert(client->optlen); |
2e425fd5 VSO |
562 | ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length", |
563 | nbd_opt_lookup(client->opt)); | |
a16a7907 | 564 | if (fatal && !ret) { |
894e0280 | 565 | error_setg(errp, "option '%s' has unexpected length", |
0cfae925 | 566 | nbd_opt_lookup(client->opt)); |
a16a7907 EB |
567 | return -EINVAL; |
568 | } | |
569 | return ret; | |
570 | } | |
571 | ||
f37708f6 EB |
572 | /* Handle NBD_OPT_INFO and NBD_OPT_GO. |
573 | * Return -errno on error, 0 if ready for next option, and 1 to move | |
574 | * into transmission phase. */ | |
dbb38caa | 575 | static int nbd_negotiate_handle_info(NBDClient *client, Error **errp) |
f37708f6 EB |
576 | { |
577 | int rc; | |
9d7ab222 | 578 | g_autofree char *name = NULL; |
f37708f6 EB |
579 | NBDExport *exp; |
580 | uint16_t requests; | |
581 | uint16_t request; | |
bbc35fc2 | 582 | uint32_t namelen = 0; |
f37708f6 | 583 | bool sendname = false; |
0c1d50bd EB |
584 | bool blocksize = false; |
585 | uint32_t sizes[3]; | |
f37708f6 | 586 | char buf[sizeof(uint64_t) + sizeof(uint16_t)]; |
6e280648 | 587 | uint32_t check_align = 0; |
dbb38caa | 588 | uint16_t myflags; |
f37708f6 EB |
589 | |
590 | /* Client sends: | |
591 | 4 bytes: L, name length (can be 0) | |
592 | L bytes: export name | |
593 | 2 bytes: N, number of requests (can be 0) | |
594 | N * 2 bytes: N requests | |
595 | */ | |
9d7ab222 | 596 | rc = nbd_opt_read_name(client, &name, &namelen, errp); |
894e0280 EB |
597 | if (rc <= 0) { |
598 | return rc; | |
f37708f6 | 599 | } |
f37708f6 EB |
600 | trace_nbd_negotiate_handle_export_name_request(name); |
601 | ||
d1e2c3e7 | 602 | rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp); |
894e0280 EB |
603 | if (rc <= 0) { |
604 | return rc; | |
f37708f6 | 605 | } |
80c7c2b0 | 606 | requests = be16_to_cpu(requests); |
f37708f6 | 607 | trace_nbd_negotiate_handle_info_requests(requests); |
f37708f6 | 608 | while (requests--) { |
d1e2c3e7 | 609 | rc = nbd_opt_read(client, &request, sizeof(request), false, errp); |
894e0280 EB |
610 | if (rc <= 0) { |
611 | return rc; | |
f37708f6 | 612 | } |
80c7c2b0 | 613 | request = be16_to_cpu(request); |
f37708f6 EB |
614 | trace_nbd_negotiate_handle_info_request(request, |
615 | nbd_info_lookup(request)); | |
0c1d50bd EB |
616 | /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; |
617 | * everything else is either a request we don't know or | |
618 | * something we send regardless of request */ | |
619 | switch (request) { | |
620 | case NBD_INFO_NAME: | |
f37708f6 | 621 | sendname = true; |
0c1d50bd EB |
622 | break; |
623 | case NBD_INFO_BLOCK_SIZE: | |
624 | blocksize = true; | |
625 | break; | |
f37708f6 EB |
626 | } |
627 | } | |
894e0280 EB |
628 | if (client->optlen) { |
629 | return nbd_reject_length(client, false, errp); | |
630 | } | |
f37708f6 EB |
631 | |
632 | exp = nbd_export_find(name); | |
633 | if (!exp) { | |
5c4fe018 EB |
634 | g_autofree char *sane_name = nbd_sanitize_name(name); |
635 | ||
0cfae925 VSO |
636 | return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, |
637 | errp, "export '%s' not present", | |
5c4fe018 | 638 | sane_name); |
f37708f6 EB |
639 | } |
640 | ||
641 | /* Don't bother sending NBD_INFO_NAME unless client requested it */ | |
642 | if (sendname) { | |
0cfae925 | 643 | rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, |
f37708f6 EB |
644 | errp); |
645 | if (rc < 0) { | |
646 | return rc; | |
647 | } | |
648 | } | |
649 | ||
650 | /* Send NBD_INFO_DESCRIPTION only if available, regardless of | |
651 | * client request */ | |
652 | if (exp->description) { | |
653 | size_t len = strlen(exp->description); | |
654 | ||
93676c88 | 655 | assert(len <= NBD_MAX_STRING_SIZE); |
0cfae925 | 656 | rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, |
f37708f6 EB |
657 | len, exp->description, errp); |
658 | if (rc < 0) { | |
659 | return rc; | |
660 | } | |
661 | } | |
662 | ||
0c1d50bd EB |
663 | /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size |
664 | * according to whether the client requested it, and according to | |
665 | * whether this is OPT_INFO or OPT_GO. */ | |
b0245d64 EB |
666 | /* minimum - 1 for back-compat, or actual if client will obey it. */ |
667 | if (client->opt == NBD_OPT_INFO || blocksize) { | |
37a4f70c | 668 | check_align = sizes[0] = blk_get_request_alignment(exp->common.blk); |
b0245d64 EB |
669 | } else { |
670 | sizes[0] = 1; | |
671 | } | |
672 | assert(sizes[0] <= NBD_MAX_BUFFER_SIZE); | |
0c1d50bd EB |
673 | /* preferred - Hard-code to 4096 for now. |
674 | * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ | |
b0245d64 | 675 | sizes[1] = MAX(4096, sizes[0]); |
0c1d50bd | 676 | /* maximum - At most 32M, but smaller as appropriate. */ |
37a4f70c | 677 | sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE); |
0c1d50bd | 678 | trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); |
80c7c2b0 PM |
679 | sizes[0] = cpu_to_be32(sizes[0]); |
680 | sizes[1] = cpu_to_be32(sizes[1]); | |
681 | sizes[2] = cpu_to_be32(sizes[2]); | |
0cfae925 | 682 | rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, |
0c1d50bd EB |
683 | sizeof(sizes), sizes, errp); |
684 | if (rc < 0) { | |
685 | return rc; | |
686 | } | |
687 | ||
f37708f6 | 688 | /* Send NBD_INFO_EXPORT always */ |
dbb38caa | 689 | myflags = exp->nbdflags; |
ac132d05 | 690 | if (client->mode >= NBD_MODE_STRUCTURED) { |
dbb38caa EB |
691 | myflags |= NBD_FLAG_SEND_DF; |
692 | } | |
693 | trace_nbd_negotiate_new_style_size_flags(exp->size, myflags); | |
f37708f6 | 694 | stq_be_p(buf, exp->size); |
dbb38caa | 695 | stw_be_p(buf + 8, myflags); |
0cfae925 | 696 | rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, |
f37708f6 EB |
697 | sizeof(buf), buf, errp); |
698 | if (rc < 0) { | |
699 | return rc; | |
700 | } | |
701 | ||
099fbcd6 EB |
702 | /* |
703 | * If the client is just asking for NBD_OPT_INFO, but forgot to | |
704 | * request block sizes in a situation that would impact | |
705 | * performance, then return an error. But for NBD_OPT_GO, we | |
706 | * tolerate all clients, regardless of alignments. | |
707 | */ | |
708 | if (client->opt == NBD_OPT_INFO && !blocksize && | |
37a4f70c | 709 | blk_get_request_alignment(exp->common.blk) > 1) { |
0cfae925 VSO |
710 | return nbd_negotiate_send_rep_err(client, |
711 | NBD_REP_ERR_BLOCK_SIZE_REQD, | |
0c1d50bd EB |
712 | errp, |
713 | "request NBD_INFO_BLOCK_SIZE to " | |
714 | "use this export"); | |
715 | } | |
716 | ||
f37708f6 | 717 | /* Final reply */ |
0cfae925 | 718 | rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); |
f37708f6 EB |
719 | if (rc < 0) { |
720 | return rc; | |
721 | } | |
722 | ||
0cfae925 | 723 | if (client->opt == NBD_OPT_GO) { |
f37708f6 | 724 | client->exp = exp; |
6e280648 | 725 | client->check_align = check_align; |
f37708f6 | 726 | QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); |
c69de1be | 727 | blk_exp_ref(&client->exp->common); |
af736e54 | 728 | nbd_check_meta_export(client); |
f37708f6 EB |
729 | rc = 1; |
730 | } | |
731 | return rc; | |
f37708f6 EB |
732 | } |
733 | ||
734 | ||
36683283 EB |
735 | /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the |
736 | * new channel for all further (now-encrypted) communication. */ | |
f95910fe | 737 | static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client, |
2fd2c840 | 738 | Error **errp) |
f95910fe DB |
739 | { |
740 | QIOChannel *ioc; | |
741 | QIOChannelTLS *tioc; | |
742 | struct NBDTLSHandshakeData data = { 0 }; | |
743 | ||
0cfae925 VSO |
744 | assert(client->opt == NBD_OPT_STARTTLS); |
745 | ||
9588463e | 746 | trace_nbd_negotiate_handle_starttls(); |
f95910fe | 747 | ioc = client->ioc; |
f95910fe | 748 | |
0cfae925 | 749 | if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { |
63d5ef86 EB |
750 | return NULL; |
751 | } | |
f95910fe DB |
752 | |
753 | tioc = qio_channel_tls_new_server(ioc, | |
754 | client->tlscreds, | |
b25e12da | 755 | client->tlsauthz, |
2fd2c840 | 756 | errp); |
f95910fe DB |
757 | if (!tioc) { |
758 | return NULL; | |
759 | } | |
760 | ||
0d73f725 | 761 | qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); |
9588463e | 762 | trace_nbd_negotiate_handle_starttls_handshake(); |
f95910fe DB |
763 | data.loop = g_main_loop_new(g_main_context_default(), FALSE); |
764 | qio_channel_tls_handshake(tioc, | |
765 | nbd_tls_handshake, | |
766 | &data, | |
1939ccda | 767 | NULL, |
f95910fe DB |
768 | NULL); |
769 | ||
770 | if (!data.complete) { | |
771 | g_main_loop_run(data.loop); | |
772 | } | |
773 | g_main_loop_unref(data.loop); | |
774 | if (data.error) { | |
775 | object_unref(OBJECT(tioc)); | |
2fd2c840 | 776 | error_propagate(errp, data.error); |
f95910fe DB |
777 | return NULL; |
778 | } | |
779 | ||
780 | return QIO_CHANNEL(tioc); | |
781 | } | |
782 | ||
e7b1948d VSO |
783 | /* nbd_negotiate_send_meta_context |
784 | * | |
785 | * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT | |
786 | * | |
787 | * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead. | |
788 | */ | |
789 | static int nbd_negotiate_send_meta_context(NBDClient *client, | |
790 | const char *context, | |
791 | uint32_t context_id, | |
792 | Error **errp) | |
793 | { | |
794 | NBDOptionReplyMetaContext opt; | |
795 | struct iovec iov[] = { | |
796 | {.iov_base = &opt, .iov_len = sizeof(opt)}, | |
797 | {.iov_base = (void *)context, .iov_len = strlen(context)} | |
798 | }; | |
799 | ||
93676c88 | 800 | assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE); |
e7b1948d VSO |
801 | if (client->opt == NBD_OPT_LIST_META_CONTEXT) { |
802 | context_id = 0; | |
803 | } | |
804 | ||
2b53af25 | 805 | trace_nbd_negotiate_meta_query_reply(context, context_id); |
e7b1948d VSO |
806 | set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT, |
807 | sizeof(opt) - sizeof(opt.h) + iov[1].iov_len); | |
808 | stl_be_p(&opt.context_id, context_id); | |
809 | ||
810 | return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0; | |
811 | } | |
812 | ||
ebd57062 EB |
813 | /* |
814 | * Return true if @query matches @pattern, or if @query is empty when | |
815 | * the @client is performing _LIST_. | |
dbb8b396 | 816 | */ |
ebd57062 EB |
817 | static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern, |
818 | const char *query) | |
e7b1948d | 819 | { |
ebd57062 EB |
820 | if (!*query) { |
821 | trace_nbd_negotiate_meta_query_parse("empty"); | |
822 | return client->opt == NBD_OPT_LIST_META_CONTEXT; | |
e7b1948d | 823 | } |
ebd57062 | 824 | if (strcmp(query, pattern) == 0) { |
b0769d8f | 825 | trace_nbd_negotiate_meta_query_parse(pattern); |
ebd57062 | 826 | return true; |
e7b1948d | 827 | } |
ebd57062 EB |
828 | trace_nbd_negotiate_meta_query_skip("pattern not matched"); |
829 | return false; | |
e7b1948d VSO |
830 | } |
831 | ||
b0769d8f | 832 | /* |
ebd57062 | 833 | * Return true and adjust @str in place if it begins with @prefix. |
b0769d8f | 834 | */ |
ebd57062 | 835 | static bool nbd_strshift(const char **str, const char *prefix) |
b0769d8f | 836 | { |
ebd57062 | 837 | size_t len = strlen(prefix); |
b0769d8f | 838 | |
ebd57062 EB |
839 | if (strncmp(*str, prefix, len) == 0) { |
840 | *str += len; | |
841 | return true; | |
b0769d8f | 842 | } |
ebd57062 | 843 | return false; |
b0769d8f VSO |
844 | } |
845 | ||
846 | /* nbd_meta_base_query | |
847 | * | |
848 | * Handle queries to 'base' namespace. For now, only the base:allocation | |
ebd57062 | 849 | * context is available. Return true if @query has been handled. |
b0769d8f | 850 | */ |
ebd57062 EB |
851 | static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta, |
852 | const char *query) | |
b0769d8f | 853 | { |
ebd57062 EB |
854 | if (!nbd_strshift(&query, "base:")) { |
855 | return false; | |
856 | } | |
857 | trace_nbd_negotiate_meta_query_parse("base:"); | |
858 | ||
859 | if (nbd_meta_empty_or_pattern(client, "allocation", query)) { | |
860 | meta->base_allocation = true; | |
861 | } | |
862 | return true; | |
b0769d8f VSO |
863 | } |
864 | ||
ebd57062 | 865 | /* nbd_meta_qemu_query |
3d068aff | 866 | * |
ebd57062 | 867 | * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap: |
71719cd5 EB |
868 | * and qemu:allocation-depth contexts are available. Return true if @query |
869 | * has been handled. | |
ebd57062 EB |
870 | */ |
871 | static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta, | |
872 | const char *query) | |
3d068aff | 873 | { |
3b1f244c EB |
874 | size_t i; |
875 | ||
ebd57062 EB |
876 | if (!nbd_strshift(&query, "qemu:")) { |
877 | return false; | |
3d068aff | 878 | } |
ebd57062 | 879 | trace_nbd_negotiate_meta_query_parse("qemu:"); |
3d068aff | 880 | |
ebd57062 | 881 | if (!*query) { |
3d068aff | 882 | if (client->opt == NBD_OPT_LIST_META_CONTEXT) { |
71719cd5 | 883 | meta->allocation_depth = meta->exp->allocation_depth; |
76df2b8d EB |
884 | if (meta->exp->nr_export_bitmaps) { |
885 | memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); | |
886 | } | |
3d068aff VSO |
887 | } |
888 | trace_nbd_negotiate_meta_query_parse("empty"); | |
ebd57062 | 889 | return true; |
3d068aff VSO |
890 | } |
891 | ||
71719cd5 EB |
892 | if (strcmp(query, "allocation-depth") == 0) { |
893 | trace_nbd_negotiate_meta_query_parse("allocation-depth"); | |
894 | meta->allocation_depth = meta->exp->allocation_depth; | |
895 | return true; | |
896 | } | |
897 | ||
ebd57062 EB |
898 | if (nbd_strshift(&query, "dirty-bitmap:")) { |
899 | trace_nbd_negotiate_meta_query_parse("dirty-bitmap:"); | |
3b1f244c | 900 | if (!*query) { |
76df2b8d EB |
901 | if (client->opt == NBD_OPT_LIST_META_CONTEXT && |
902 | meta->exp->nr_export_bitmaps) { | |
3b1f244c EB |
903 | memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); |
904 | } | |
905 | trace_nbd_negotiate_meta_query_parse("empty"); | |
ebd57062 EB |
906 | return true; |
907 | } | |
3b1f244c EB |
908 | |
909 | for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { | |
910 | const char *bm_name; | |
911 | ||
912 | bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); | |
913 | if (strcmp(bm_name, query) == 0) { | |
914 | meta->bitmaps[i] = true; | |
915 | trace_nbd_negotiate_meta_query_parse(query); | |
916 | return true; | |
917 | } | |
ebd57062 | 918 | } |
3b1f244c | 919 | trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match"); |
ebd57062 | 920 | return true; |
3d068aff VSO |
921 | } |
922 | ||
71719cd5 | 923 | trace_nbd_negotiate_meta_query_skip("unknown qemu context"); |
ebd57062 | 924 | return true; |
3d068aff VSO |
925 | } |
926 | ||
e7b1948d VSO |
927 | /* nbd_negotiate_meta_query |
928 | * | |
929 | * Parse namespace name and call corresponding function to parse body of the | |
930 | * query. | |
931 | * | |
93676c88 | 932 | * The only supported namespaces are 'base' and 'qemu'. |
e7b1948d | 933 | * |
e7b1948d VSO |
934 | * Return -errno on I/O error, 0 if option was completely handled by |
935 | * sending a reply about inconsistent lengths, or 1 on success. */ | |
936 | static int nbd_negotiate_meta_query(NBDClient *client, | |
937 | NBDExportMetaContexts *meta, Error **errp) | |
938 | { | |
939 | int ret; | |
ebd57062 | 940 | g_autofree char *query = NULL; |
e7b1948d VSO |
941 | uint32_t len; |
942 | ||
d1e2c3e7 | 943 | ret = nbd_opt_read(client, &len, sizeof(len), false, errp); |
e7b1948d VSO |
944 | if (ret <= 0) { |
945 | return ret; | |
946 | } | |
80c7c2b0 | 947 | len = cpu_to_be32(len); |
e7b1948d | 948 | |
93676c88 EB |
949 | if (len > NBD_MAX_STRING_SIZE) { |
950 | trace_nbd_negotiate_meta_query_skip("length too long"); | |
951 | return nbd_opt_skip(client, len, errp); | |
952 | } | |
e7b1948d | 953 | |
ebd57062 EB |
954 | query = g_malloc(len + 1); |
955 | ret = nbd_opt_read(client, query, len, true, errp); | |
e7b1948d VSO |
956 | if (ret <= 0) { |
957 | return ret; | |
958 | } | |
ebd57062 | 959 | query[len] = '\0'; |
3d068aff | 960 | |
ebd57062 EB |
961 | if (nbd_meta_base_query(client, meta, query)) { |
962 | return 1; | |
963 | } | |
964 | if (nbd_meta_qemu_query(client, meta, query)) { | |
965 | return 1; | |
e7b1948d VSO |
966 | } |
967 | ||
3d068aff | 968 | trace_nbd_negotiate_meta_query_skip("unknown namespace"); |
ebd57062 | 969 | return 1; |
e7b1948d VSO |
970 | } |
971 | ||
972 | /* nbd_negotiate_meta_queries | |
973 | * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT | |
974 | * | |
975 | * Return -errno on I/O error, or 0 if option was completely handled. */ | |
976 | static int nbd_negotiate_meta_queries(NBDClient *client, | |
977 | NBDExportMetaContexts *meta, Error **errp) | |
978 | { | |
979 | int ret; | |
9d7ab222 | 980 | g_autofree char *export_name = NULL; |
cd1675f8 RH |
981 | /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */ |
982 | g_autofree G_GNUC_UNUSED bool *bitmaps = NULL; | |
3b1f244c | 983 | NBDExportMetaContexts local_meta = {0}; |
e7b1948d | 984 | uint32_t nb_queries; |
3b1f244c | 985 | size_t i; |
47ec485e | 986 | size_t count = 0; |
e7b1948d | 987 | |
ac132d05 EB |
988 | if (client->opt == NBD_OPT_SET_META_CONTEXT && |
989 | client->mode < NBD_MODE_STRUCTURED) { | |
e7b1948d VSO |
990 | return nbd_opt_invalid(client, errp, |
991 | "request option '%s' when structured reply " | |
992 | "is not negotiated", | |
993 | nbd_opt_lookup(client->opt)); | |
994 | } | |
995 | ||
996 | if (client->opt == NBD_OPT_LIST_META_CONTEXT) { | |
997 | /* Only change the caller's meta on SET. */ | |
998 | meta = &local_meta; | |
999 | } | |
1000 | ||
3b1f244c | 1001 | g_free(meta->bitmaps); |
e7b1948d VSO |
1002 | memset(meta, 0, sizeof(*meta)); |
1003 | ||
9d7ab222 | 1004 | ret = nbd_opt_read_name(client, &export_name, NULL, errp); |
e7b1948d VSO |
1005 | if (ret <= 0) { |
1006 | return ret; | |
1007 | } | |
1008 | ||
af736e54 VSO |
1009 | meta->exp = nbd_export_find(export_name); |
1010 | if (meta->exp == NULL) { | |
5c4fe018 EB |
1011 | g_autofree char *sane_name = nbd_sanitize_name(export_name); |
1012 | ||
e7b1948d | 1013 | return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp, |
5c4fe018 | 1014 | "export '%s' not present", sane_name); |
e7b1948d | 1015 | } |
3b1f244c EB |
1016 | meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps); |
1017 | if (client->opt == NBD_OPT_LIST_META_CONTEXT) { | |
1018 | bitmaps = meta->bitmaps; | |
1019 | } | |
e7b1948d | 1020 | |
d1e2c3e7 | 1021 | ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp); |
e7b1948d VSO |
1022 | if (ret <= 0) { |
1023 | return ret; | |
1024 | } | |
80c7c2b0 | 1025 | nb_queries = cpu_to_be32(nb_queries); |
2b53af25 | 1026 | trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt), |
af736e54 | 1027 | export_name, nb_queries); |
e7b1948d VSO |
1028 | |
1029 | if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) { | |
1030 | /* enable all known contexts */ | |
1031 | meta->base_allocation = true; | |
71719cd5 | 1032 | meta->allocation_depth = meta->exp->allocation_depth; |
76df2b8d EB |
1033 | if (meta->exp->nr_export_bitmaps) { |
1034 | memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); | |
1035 | } | |
e7b1948d VSO |
1036 | } else { |
1037 | for (i = 0; i < nb_queries; ++i) { | |
1038 | ret = nbd_negotiate_meta_query(client, meta, errp); | |
1039 | if (ret <= 0) { | |
1040 | return ret; | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | if (meta->base_allocation) { | |
1046 | ret = nbd_negotiate_send_meta_context(client, "base:allocation", | |
1047 | NBD_META_ID_BASE_ALLOCATION, | |
1048 | errp); | |
1049 | if (ret < 0) { | |
1050 | return ret; | |
1051 | } | |
47ec485e | 1052 | count++; |
e7b1948d VSO |
1053 | } |
1054 | ||
71719cd5 EB |
1055 | if (meta->allocation_depth) { |
1056 | ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth", | |
1057 | NBD_META_ID_ALLOCATION_DEPTH, | |
1058 | errp); | |
1059 | if (ret < 0) { | |
1060 | return ret; | |
1061 | } | |
1062 | count++; | |
1063 | } | |
1064 | ||
3b1f244c EB |
1065 | for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { |
1066 | const char *bm_name; | |
1067 | g_autofree char *context = NULL; | |
1068 | ||
1069 | if (!meta->bitmaps[i]) { | |
1070 | continue; | |
1071 | } | |
1072 | ||
1073 | bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); | |
1074 | context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name); | |
02e87e3b EB |
1075 | |
1076 | ret = nbd_negotiate_send_meta_context(client, context, | |
3b1f244c | 1077 | NBD_META_ID_DIRTY_BITMAP + i, |
3d068aff VSO |
1078 | errp); |
1079 | if (ret < 0) { | |
1080 | return ret; | |
1081 | } | |
47ec485e | 1082 | count++; |
3d068aff VSO |
1083 | } |
1084 | ||
e7b1948d VSO |
1085 | ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); |
1086 | if (ret == 0) { | |
47ec485e | 1087 | meta->count = count; |
e7b1948d VSO |
1088 | } |
1089 | ||
1090 | return ret; | |
1091 | } | |
1092 | ||
1e120ffe | 1093 | /* nbd_negotiate_options |
f37708f6 EB |
1094 | * Process all NBD_OPT_* client option commands, during fixed newstyle |
1095 | * negotiation. | |
1e120ffe | 1096 | * Return: |
2fd2c840 VSO |
1097 | * -errno on error, errp is set |
1098 | * 0 on successful negotiation, errp is not set | |
1099 | * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, | |
1100 | * errp is not set | |
1e120ffe | 1101 | */ |
dbb38caa | 1102 | static int nbd_negotiate_options(NBDClient *client, Error **errp) |
f5076b5a | 1103 | { |
9c122ada | 1104 | uint32_t flags; |
26afa868 | 1105 | bool fixedNewstyle = false; |
23e099c3 | 1106 | bool no_zeroes = false; |
9c122ada HR |
1107 | |
1108 | /* Client sends: | |
1109 | [ 0 .. 3] client flags | |
1110 | ||
f37708f6 | 1111 | Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: |
9c122ada HR |
1112 | [ 0 .. 7] NBD_OPTS_MAGIC |
1113 | [ 8 .. 11] NBD option | |
1114 | [12 .. 15] Data length | |
1115 | ... Rest of request | |
1116 | ||
1117 | [ 0 .. 7] NBD_OPTS_MAGIC | |
1118 | [ 8 .. 11] Second NBD option | |
1119 | [12 .. 15] Data length | |
1120 | ... Rest of request | |
1121 | */ | |
1122 | ||
e6798f06 | 1123 | if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) { |
9c122ada HR |
1124 | return -EIO; |
1125 | } | |
ac132d05 | 1126 | client->mode = NBD_MODE_EXPORT_NAME; |
621c4f4e | 1127 | trace_nbd_negotiate_options_flags(flags); |
26afa868 | 1128 | if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { |
26afa868 DB |
1129 | fixedNewstyle = true; |
1130 | flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; | |
ac132d05 | 1131 | client->mode = NBD_MODE_SIMPLE; |
26afa868 | 1132 | } |
c203c59a | 1133 | if (flags & NBD_FLAG_C_NO_ZEROES) { |
23e099c3 | 1134 | no_zeroes = true; |
c203c59a EB |
1135 | flags &= ~NBD_FLAG_C_NO_ZEROES; |
1136 | } | |
26afa868 | 1137 | if (flags != 0) { |
2fd2c840 | 1138 | error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); |
621c4f4e | 1139 | return -EINVAL; |
9c122ada HR |
1140 | } |
1141 | ||
f5076b5a | 1142 | while (1) { |
9c122ada | 1143 | int ret; |
7f9039cd | 1144 | uint32_t option, length; |
f5076b5a HB |
1145 | uint64_t magic; |
1146 | ||
e6798f06 | 1147 | if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) { |
f5076b5a HB |
1148 | return -EINVAL; |
1149 | } | |
9588463e VSO |
1150 | trace_nbd_negotiate_options_check_magic(magic); |
1151 | if (magic != NBD_OPTS_MAGIC) { | |
2fd2c840 | 1152 | error_setg(errp, "Bad magic received"); |
f5076b5a HB |
1153 | return -EINVAL; |
1154 | } | |
1155 | ||
e6798f06 | 1156 | if (nbd_read32(client->ioc, &option, "option", errp) < 0) { |
f5076b5a HB |
1157 | return -EINVAL; |
1158 | } | |
0cfae925 | 1159 | client->opt = option; |
f5076b5a | 1160 | |
e6798f06 | 1161 | if (nbd_read32(client->ioc, &length, "option length", errp) < 0) { |
f5076b5a HB |
1162 | return -EINVAL; |
1163 | } | |
894e0280 | 1164 | assert(!client->optlen); |
0cfae925 | 1165 | client->optlen = length; |
f5076b5a | 1166 | |
fdad35ef | 1167 | if (length > NBD_MAX_BUFFER_SIZE) { |
b2578459 | 1168 | error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)", |
fdad35ef EB |
1169 | length, NBD_MAX_BUFFER_SIZE); |
1170 | return -EINVAL; | |
1171 | } | |
1172 | ||
3736cc5b EB |
1173 | trace_nbd_negotiate_options_check_option(option, |
1174 | nbd_opt_lookup(option)); | |
f95910fe DB |
1175 | if (client->tlscreds && |
1176 | client->ioc == (QIOChannel *)client->sioc) { | |
1177 | QIOChannel *tioc; | |
1178 | if (!fixedNewstyle) { | |
7f9039cd | 1179 | error_setg(errp, "Unsupported option 0x%" PRIx32, option); |
f95910fe DB |
1180 | return -EINVAL; |
1181 | } | |
7f9039cd | 1182 | switch (option) { |
f95910fe | 1183 | case NBD_OPT_STARTTLS: |
e68c35cf EB |
1184 | if (length) { |
1185 | /* Unconditionally drop the connection if the client | |
1186 | * can't start a TLS negotiation correctly */ | |
0cfae925 | 1187 | return nbd_reject_length(client, true, errp); |
e68c35cf EB |
1188 | } |
1189 | tioc = nbd_negotiate_handle_starttls(client, errp); | |
f95910fe DB |
1190 | if (!tioc) { |
1191 | return -EIO; | |
1192 | } | |
8cbee49e | 1193 | ret = 0; |
f95910fe | 1194 | object_unref(OBJECT(client->ioc)); |
7d5b0d68 | 1195 | client->ioc = tioc; |
f95910fe DB |
1196 | break; |
1197 | ||
d1129a8a EB |
1198 | case NBD_OPT_EXPORT_NAME: |
1199 | /* No way to return an error to client, so drop connection */ | |
2fd2c840 | 1200 | error_setg(errp, "Option 0x%x not permitted before TLS", |
7f9039cd | 1201 | option); |
d1129a8a EB |
1202 | return -EINVAL; |
1203 | ||
f95910fe | 1204 | default: |
3e99ebb9 EB |
1205 | /* Let the client keep trying, unless they asked to |
1206 | * quit. Always try to give an error back to the | |
1207 | * client; but when replying to OPT_ABORT, be aware | |
1208 | * that the client may hang up before receiving the | |
1209 | * error, in which case we are fine ignoring the | |
1210 | * resulting EPIPE. */ | |
1211 | ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, | |
1212 | option == NBD_OPT_ABORT ? NULL : errp, | |
894e0280 | 1213 | "Option 0x%" PRIx32 |
0b0bb124 | 1214 | " not permitted before TLS", option); |
7f9039cd | 1215 | if (option == NBD_OPT_ABORT) { |
1e120ffe | 1216 | return 1; |
b6f5d3b5 | 1217 | } |
d1129a8a | 1218 | break; |
f95910fe DB |
1219 | } |
1220 | } else if (fixedNewstyle) { | |
7f9039cd | 1221 | switch (option) { |
26afa868 | 1222 | case NBD_OPT_LIST: |
e68c35cf | 1223 | if (length) { |
0cfae925 | 1224 | ret = nbd_reject_length(client, false, errp); |
e68c35cf EB |
1225 | } else { |
1226 | ret = nbd_negotiate_handle_list(client, errp); | |
1227 | } | |
26afa868 DB |
1228 | break; |
1229 | ||
1230 | case NBD_OPT_ABORT: | |
b6f5d3b5 EB |
1231 | /* NBD spec says we must try to reply before |
1232 | * disconnecting, but that we must also tolerate | |
1233 | * guests that don't wait for our reply. */ | |
0cfae925 | 1234 | nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); |
1e120ffe | 1235 | return 1; |
26afa868 DB |
1236 | |
1237 | case NBD_OPT_EXPORT_NAME: | |
dbb38caa | 1238 | return nbd_negotiate_handle_export_name(client, no_zeroes, |
23e099c3 | 1239 | errp); |
26afa868 | 1240 | |
f37708f6 EB |
1241 | case NBD_OPT_INFO: |
1242 | case NBD_OPT_GO: | |
dbb38caa | 1243 | ret = nbd_negotiate_handle_info(client, errp); |
f37708f6 EB |
1244 | if (ret == 1) { |
1245 | assert(option == NBD_OPT_GO); | |
1246 | return 0; | |
1247 | } | |
f37708f6 EB |
1248 | break; |
1249 | ||
f95910fe | 1250 | case NBD_OPT_STARTTLS: |
e68c35cf | 1251 | if (length) { |
0cfae925 | 1252 | ret = nbd_reject_length(client, false, errp); |
e68c35cf | 1253 | } else if (client->tlscreds) { |
0cfae925 VSO |
1254 | ret = nbd_negotiate_send_rep_err(client, |
1255 | NBD_REP_ERR_INVALID, errp, | |
36683283 | 1256 | "TLS already enabled"); |
f95910fe | 1257 | } else { |
0cfae925 VSO |
1258 | ret = nbd_negotiate_send_rep_err(client, |
1259 | NBD_REP_ERR_POLICY, errp, | |
36683283 | 1260 | "TLS not configured"); |
63d5ef86 | 1261 | } |
d1129a8a | 1262 | break; |
5c54e7fa VSO |
1263 | |
1264 | case NBD_OPT_STRUCTURED_REPLY: | |
1265 | if (length) { | |
0cfae925 | 1266 | ret = nbd_reject_length(client, false, errp); |
ac132d05 | 1267 | } else if (client->mode >= NBD_MODE_STRUCTURED) { |
5c54e7fa | 1268 | ret = nbd_negotiate_send_rep_err( |
0cfae925 | 1269 | client, NBD_REP_ERR_INVALID, errp, |
5c54e7fa VSO |
1270 | "structured reply already negotiated"); |
1271 | } else { | |
0cfae925 | 1272 | ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); |
ac132d05 | 1273 | client->mode = NBD_MODE_STRUCTURED; |
5c54e7fa VSO |
1274 | } |
1275 | break; | |
1276 | ||
e7b1948d VSO |
1277 | case NBD_OPT_LIST_META_CONTEXT: |
1278 | case NBD_OPT_SET_META_CONTEXT: | |
1279 | ret = nbd_negotiate_meta_queries(client, &client->export_meta, | |
1280 | errp); | |
1281 | break; | |
1282 | ||
26afa868 | 1283 | default: |
894e0280 | 1284 | ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, |
28fb494f | 1285 | "Unsupported option %" PRIu32 " (%s)", |
894e0280 | 1286 | option, nbd_opt_lookup(option)); |
156f6a10 | 1287 | break; |
26afa868 DB |
1288 | } |
1289 | } else { | |
1290 | /* | |
1291 | * If broken new-style we should drop the connection | |
1292 | * for anything except NBD_OPT_EXPORT_NAME | |
1293 | */ | |
7f9039cd | 1294 | switch (option) { |
26afa868 | 1295 | case NBD_OPT_EXPORT_NAME: |
dbb38caa | 1296 | return nbd_negotiate_handle_export_name(client, no_zeroes, |
23e099c3 | 1297 | errp); |
26afa868 DB |
1298 | |
1299 | default: | |
28fb494f | 1300 | error_setg(errp, "Unsupported option %" PRIu32 " (%s)", |
3736cc5b | 1301 | option, nbd_opt_lookup(option)); |
26afa868 | 1302 | return -EINVAL; |
32d7d2e0 | 1303 | } |
f5076b5a | 1304 | } |
8cbee49e EB |
1305 | if (ret < 0) { |
1306 | return ret; | |
1307 | } | |
f5076b5a HB |
1308 | } |
1309 | } | |
1310 | ||
1e120ffe VSO |
1311 | /* nbd_negotiate |
1312 | * Return: | |
2fd2c840 VSO |
1313 | * -errno on error, errp is set |
1314 | * 0 on successful negotiation, errp is not set | |
1315 | * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, | |
1316 | * errp is not set | |
1e120ffe | 1317 | */ |
2fd2c840 | 1318 | static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) |
7a5ca864 | 1319 | { |
795d946d | 1320 | ERRP_GUARD(); |
5f66d060 | 1321 | char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; |
2e5c9ad6 | 1322 | int ret; |
b2e3d87f | 1323 | |
5f66d060 | 1324 | /* Old style negotiation header, no room for options |
6b8c01e7 PB |
1325 | [ 0 .. 7] passwd ("NBDMAGIC") |
1326 | [ 8 .. 15] magic (NBD_CLIENT_MAGIC) | |
b2e3d87f | 1327 | [16 .. 23] size |
5f66d060 | 1328 | [24 .. 27] export flags (zero-extended) |
6b8c01e7 PB |
1329 | [28 .. 151] reserved (0) |
1330 | ||
5f66d060 | 1331 | New style negotiation header, client can send options |
6b8c01e7 PB |
1332 | [ 0 .. 7] passwd ("NBDMAGIC") |
1333 | [ 8 .. 15] magic (NBD_OPTS_MAGIC) | |
1334 | [16 .. 17] server flags (0) | |
f37708f6 | 1335 | ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... |
b2e3d87f NT |
1336 | */ |
1337 | ||
1c778ef7 | 1338 | qio_channel_set_blocking(client->ioc, false, NULL); |
06e0f098 | 1339 | qio_channel_set_follow_coroutine_ctx(client->ioc, true); |
185b4338 | 1340 | |
9588463e | 1341 | trace_nbd_negotiate_begin(); |
b2e3d87f | 1342 | memcpy(buf, "NBDMAGIC", 8); |
f95910fe | 1343 | |
7f7dfe2a VSO |
1344 | stq_be_p(buf + 8, NBD_OPTS_MAGIC); |
1345 | stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); | |
b2e3d87f | 1346 | |
7f7dfe2a VSO |
1347 | if (nbd_write(client->ioc, buf, 18, errp) < 0) { |
1348 | error_prepend(errp, "write failed: "); | |
1349 | return -EINVAL; | |
1350 | } | |
dbb38caa | 1351 | ret = nbd_negotiate_options(client, errp); |
7f7dfe2a VSO |
1352 | if (ret != 0) { |
1353 | if (ret < 0) { | |
1354 | error_prepend(errp, "option negotiation failed: "); | |
6b8c01e7 | 1355 | } |
7f7dfe2a | 1356 | return ret; |
b2e3d87f NT |
1357 | } |
1358 | ||
0cfae925 | 1359 | assert(!client->optlen); |
9588463e | 1360 | trace_nbd_negotiate_success(); |
d9faeed8 VSO |
1361 | |
1362 | return 0; | |
7a5ca864 FB |
1363 | } |
1364 | ||
f148ae7d SL |
1365 | /* nbd_read_eof |
1366 | * Tries to read @size bytes from @ioc. This is a local implementation of | |
1367 | * qio_channel_readv_all_eof. We have it here because we need it to be | |
1368 | * interruptible and to know when the coroutine is yielding. | |
1369 | * Returns 1 on success | |
1370 | * 0 on eof, when no data was read (errp is not set) | |
1371 | * negative errno on failure (errp is set) | |
1372 | */ | |
1373 | static inline int coroutine_fn | |
1374 | nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp) | |
1375 | { | |
1376 | bool partial = false; | |
1377 | ||
1378 | assert(size); | |
1379 | while (size > 0) { | |
1380 | struct iovec iov = { .iov_base = buffer, .iov_len = size }; | |
1381 | ssize_t len; | |
1382 | ||
1383 | len = qio_channel_readv(client->ioc, &iov, 1, errp); | |
1384 | if (len == QIO_CHANNEL_ERR_BLOCK) { | |
1385 | client->read_yielding = true; | |
1386 | qio_channel_yield(client->ioc, G_IO_IN); | |
1387 | client->read_yielding = false; | |
1388 | if (client->quiescing) { | |
1389 | return -EAGAIN; | |
1390 | } | |
1391 | continue; | |
1392 | } else if (len < 0) { | |
1393 | return -EIO; | |
1394 | } else if (len == 0) { | |
1395 | if (partial) { | |
1396 | error_setg(errp, | |
1397 | "Unexpected end-of-file before all bytes were read"); | |
1398 | return -EIO; | |
1399 | } else { | |
1400 | return 0; | |
1401 | } | |
1402 | } | |
1403 | ||
1404 | partial = true; | |
1405 | size -= len; | |
1406 | buffer = (uint8_t *) buffer + len; | |
1407 | } | |
1408 | return 1; | |
1409 | } | |
1410 | ||
d2223cdd PB |
1411 | static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request, |
1412 | Error **errp) | |
75818250 | 1413 | { |
c8720ca0 EB |
1414 | uint8_t buf[NBD_EXTENDED_REQUEST_SIZE]; |
1415 | uint32_t magic, expect; | |
a0dc63a6 | 1416 | int ret; |
c8720ca0 EB |
1417 | size_t size = client->mode >= NBD_MODE_EXTENDED ? |
1418 | NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE; | |
b2e3d87f | 1419 | |
c8720ca0 | 1420 | ret = nbd_read_eof(client, buf, size, errp); |
185b4338 PB |
1421 | if (ret < 0) { |
1422 | return ret; | |
1423 | } | |
1644ccce EB |
1424 | if (ret == 0) { |
1425 | return -EIO; | |
1426 | } | |
185b4338 | 1427 | |
c8720ca0 EB |
1428 | /* |
1429 | * Compact request | |
1430 | * [ 0 .. 3] magic (NBD_REQUEST_MAGIC) | |
1431 | * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) | |
1432 | * [ 6 .. 7] type (NBD_CMD_READ, ...) | |
1433 | * [ 8 .. 15] cookie | |
1434 | * [16 .. 23] from | |
1435 | * [24 .. 27] len | |
1436 | * Extended request | |
1437 | * [ 0 .. 3] magic (NBD_EXTENDED_REQUEST_MAGIC) | |
1438 | * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...) | |
1439 | * [ 6 .. 7] type (NBD_CMD_READ, ...) | |
1440 | * [ 8 .. 15] cookie | |
1441 | * [16 .. 23] from | |
1442 | * [24 .. 31] len | |
b2e3d87f NT |
1443 | */ |
1444 | ||
773dce3c | 1445 | magic = ldl_be_p(buf); |
b626b51a EB |
1446 | request->flags = lduw_be_p(buf + 4); |
1447 | request->type = lduw_be_p(buf + 6); | |
22efd811 | 1448 | request->cookie = ldq_be_p(buf + 8); |
773dce3c | 1449 | request->from = ldq_be_p(buf + 16); |
c8720ca0 EB |
1450 | if (client->mode >= NBD_MODE_EXTENDED) { |
1451 | request->len = ldq_be_p(buf + 24); | |
1452 | expect = NBD_EXTENDED_REQUEST_MAGIC; | |
1453 | } else { | |
1454 | request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */ | |
1455 | expect = NBD_REQUEST_MAGIC; | |
1456 | } | |
b2e3d87f | 1457 | |
9588463e VSO |
1458 | trace_nbd_receive_request(magic, request->flags, request->type, |
1459 | request->from, request->len); | |
b2e3d87f | 1460 | |
c8720ca0 EB |
1461 | if (magic != expect) { |
1462 | error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%" | |
1463 | PRIx32 ")", magic, expect); | |
185b4338 | 1464 | return -EINVAL; |
b2e3d87f NT |
1465 | } |
1466 | return 0; | |
75818250 TS |
1467 | } |
1468 | ||
41996e38 PB |
1469 | #define MAX_NBD_REQUESTS 16 |
1470 | ||
ce33967a | 1471 | void nbd_client_get(NBDClient *client) |
1743b515 PB |
1472 | { |
1473 | client->refcount++; | |
1474 | } | |
1475 | ||
ce33967a | 1476 | void nbd_client_put(NBDClient *client) |
1743b515 PB |
1477 | { |
1478 | if (--client->refcount == 0) { | |
ff2b68aa | 1479 | /* The last reference should be dropped by client->close, |
f53a829b | 1480 | * which is called by client_close. |
ff2b68aa PB |
1481 | */ |
1482 | assert(client->closing); | |
1483 | ||
1c778ef7 DB |
1484 | object_unref(OBJECT(client->sioc)); |
1485 | object_unref(OBJECT(client->ioc)); | |
f95910fe DB |
1486 | if (client->tlscreds) { |
1487 | object_unref(OBJECT(client->tlscreds)); | |
1488 | } | |
b25e12da | 1489 | g_free(client->tlsauthz); |
6b8c01e7 PB |
1490 | if (client->exp) { |
1491 | QTAILQ_REMOVE(&client->exp->clients, client, next); | |
c69de1be | 1492 | blk_exp_unref(&client->exp->common); |
6b8c01e7 | 1493 | } |
3b1f244c | 1494 | g_free(client->export_meta.bitmaps); |
1743b515 PB |
1495 | g_free(client); |
1496 | } | |
1497 | } | |
1498 | ||
0c9390d9 | 1499 | static void client_close(NBDClient *client, bool negotiated) |
1743b515 | 1500 | { |
ff2b68aa PB |
1501 | if (client->closing) { |
1502 | return; | |
1503 | } | |
1504 | ||
1505 | client->closing = true; | |
1506 | ||
1507 | /* Force requests to finish. They will drop their own references, | |
1508 | * then we'll close the socket and free the NBDClient. | |
1509 | */ | |
1c778ef7 DB |
1510 | qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, |
1511 | NULL); | |
ff2b68aa PB |
1512 | |
1513 | /* Also tell the client, so that they release their reference. */ | |
0c9390d9 EB |
1514 | if (client->close_fn) { |
1515 | client->close_fn(client, negotiated); | |
1743b515 | 1516 | } |
1743b515 PB |
1517 | } |
1518 | ||
315f78ab | 1519 | static NBDRequestData *nbd_request_get(NBDClient *client) |
d9a73806 | 1520 | { |
315f78ab | 1521 | NBDRequestData *req; |
72deddc5 | 1522 | |
41996e38 PB |
1523 | assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); |
1524 | client->nb_requests++; | |
1525 | ||
315f78ab | 1526 | req = g_new0(NBDRequestData, 1); |
72deddc5 PB |
1527 | nbd_client_get(client); |
1528 | req->client = client; | |
d9a73806 PB |
1529 | return req; |
1530 | } | |
1531 | ||
315f78ab | 1532 | static void nbd_request_put(NBDRequestData *req) |
d9a73806 | 1533 | { |
72deddc5 | 1534 | NBDClient *client = req->client; |
e1adb27a | 1535 | |
2d821488 SH |
1536 | if (req->data) { |
1537 | qemu_vfree(req->data); | |
1538 | } | |
1729404c | 1539 | g_free(req); |
e1adb27a | 1540 | |
958c717d | 1541 | client->nb_requests--; |
fd6afc50 SL |
1542 | |
1543 | if (client->quiescing && client->nb_requests == 0) { | |
1544 | aio_wait_kick(); | |
1545 | } | |
1546 | ||
ff82911c PB |
1547 | nbd_client_receive_next_request(client); |
1548 | ||
72deddc5 | 1549 | nbd_client_put(client); |
d9a73806 PB |
1550 | } |
1551 | ||
aadf99a7 | 1552 | static void blk_aio_attached(AioContext *ctx, void *opaque) |
f2149281 HR |
1553 | { |
1554 | NBDExport *exp = opaque; | |
1555 | NBDClient *client; | |
1556 | ||
9588463e | 1557 | trace_nbd_blk_aio_attached(exp->name, ctx); |
f2149281 | 1558 | |
8612c686 | 1559 | exp->common.ctx = ctx; |
f2149281 HR |
1560 | |
1561 | QTAILQ_FOREACH(client, &exp->clients, next) { | |
fd6afc50 | 1562 | assert(client->nb_requests == 0); |
f148ae7d SL |
1563 | assert(client->recv_coroutine == NULL); |
1564 | assert(client->send_coroutine == NULL); | |
f148ae7d SL |
1565 | } |
1566 | } | |
1567 | ||
fd6afc50 | 1568 | static void blk_aio_detach(void *opaque) |
f148ae7d SL |
1569 | { |
1570 | NBDExport *exp = opaque; | |
f148ae7d | 1571 | |
fd6afc50 SL |
1572 | trace_nbd_blk_aio_detach(exp->name, exp->common.ctx); |
1573 | ||
fd6afc50 SL |
1574 | exp->common.ctx = NULL; |
1575 | } | |
1576 | ||
1577 | static void nbd_drained_begin(void *opaque) | |
1578 | { | |
1579 | NBDExport *exp = opaque; | |
1580 | NBDClient *client; | |
1581 | ||
1582 | QTAILQ_FOREACH(client, &exp->clients, next) { | |
f148ae7d | 1583 | client->quiescing = true; |
fd6afc50 SL |
1584 | } |
1585 | } | |
f148ae7d | 1586 | |
fd6afc50 SL |
1587 | static void nbd_drained_end(void *opaque) |
1588 | { | |
1589 | NBDExport *exp = opaque; | |
1590 | NBDClient *client; | |
f148ae7d | 1591 | |
fd6afc50 SL |
1592 | QTAILQ_FOREACH(client, &exp->clients, next) { |
1593 | client->quiescing = false; | |
1594 | nbd_client_receive_next_request(client); | |
f2149281 HR |
1595 | } |
1596 | } | |
1597 | ||
fd6afc50 | 1598 | static bool nbd_drained_poll(void *opaque) |
f2149281 HR |
1599 | { |
1600 | NBDExport *exp = opaque; | |
fd6afc50 | 1601 | NBDClient *client; |
f2149281 | 1602 | |
fd6afc50 SL |
1603 | QTAILQ_FOREACH(client, &exp->clients, next) { |
1604 | if (client->nb_requests != 0) { | |
1605 | /* | |
1606 | * If there's a coroutine waiting for a request on nbd_read_eof() | |
1607 | * enter it here so we don't depend on the client to wake it up. | |
1608 | */ | |
1609 | if (client->recv_coroutine != NULL && client->read_yielding) { | |
7c1f51bf | 1610 | qio_channel_wake_read(client->ioc); |
fd6afc50 | 1611 | } |
f2149281 | 1612 | |
fd6afc50 SL |
1613 | return true; |
1614 | } | |
1615 | } | |
f2149281 | 1616 | |
fd6afc50 | 1617 | return false; |
f2149281 HR |
1618 | } |
1619 | ||
741cc431 HR |
1620 | static void nbd_eject_notifier(Notifier *n, void *data) |
1621 | { | |
1622 | NBDExport *exp = container_of(n, NBDExport, eject_notifier); | |
61bc846d | 1623 | |
bc4ee65b | 1624 | blk_exp_request_shutdown(&exp->common); |
741cc431 HR |
1625 | } |
1626 | ||
9b562c64 KW |
1627 | void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk) |
1628 | { | |
1629 | NBDExport *nbd_exp = container_of(exp, NBDExport, common); | |
1630 | assert(exp->drv == &blk_exp_nbd); | |
1631 | assert(nbd_exp->eject_notifier_blk == NULL); | |
1632 | ||
1633 | blk_ref(blk); | |
1634 | nbd_exp->eject_notifier_blk = blk; | |
1635 | nbd_exp->eject_notifier.notify = nbd_eject_notifier; | |
1636 | blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier); | |
1637 | } | |
1638 | ||
fd6afc50 SL |
1639 | static const BlockDevOps nbd_block_ops = { |
1640 | .drained_begin = nbd_drained_begin, | |
1641 | .drained_end = nbd_drained_end, | |
1642 | .drained_poll = nbd_drained_poll, | |
1643 | }; | |
1644 | ||
5b1cb497 KW |
1645 | static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args, |
1646 | Error **errp) | |
af49bbbe | 1647 | { |
a6ff7989 | 1648 | NBDExport *exp = container_of(blk_exp, NBDExport, common); |
5b1cb497 | 1649 | BlockExportOptionsNbd *arg = &exp_args->u.nbd; |
8461b4d6 | 1650 | const char *name = arg->name ?: exp_args->node_name; |
331170e0 | 1651 | BlockBackend *blk = blk_exp->blk; |
b57e4de0 | 1652 | int64_t size; |
331170e0 | 1653 | uint64_t perm, shared_perm; |
5b1cb497 | 1654 | bool readonly = !exp_args->writable; |
e5fb29d5 | 1655 | BlockDirtyBitmapOrStrList *bitmaps; |
3b1f244c | 1656 | size_t i; |
d7086422 | 1657 | int ret; |
cd7fca95 | 1658 | |
5b1cb497 KW |
1659 | assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD); |
1660 | ||
1661 | if (!nbd_server_is_running()) { | |
1662 | error_setg(errp, "NBD server not running"); | |
1663 | return -EINVAL; | |
1664 | } | |
1665 | ||
8461b4d6 MA |
1666 | if (strlen(name) > NBD_MAX_STRING_SIZE) { |
1667 | error_setg(errp, "export name '%s' too long", name); | |
5b1cb497 KW |
1668 | return -EINVAL; |
1669 | } | |
1670 | ||
1671 | if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) { | |
1672 | error_setg(errp, "description '%s' too long", arg->description); | |
1673 | return -EINVAL; | |
1674 | } | |
1675 | ||
8461b4d6 MA |
1676 | if (nbd_export_find(name)) { |
1677 | error_setg(errp, "NBD server already has export named '%s'", name); | |
5b1cb497 KW |
1678 | return -EEXIST; |
1679 | } | |
1680 | ||
331170e0 | 1681 | size = blk_getlength(blk); |
b57e4de0 KW |
1682 | if (size < 0) { |
1683 | error_setg_errno(errp, -size, | |
1684 | "Failed to determine the NBD export's length"); | |
a6ff7989 | 1685 | return size; |
b57e4de0 KW |
1686 | } |
1687 | ||
8a7ce4f9 KW |
1688 | /* Don't allow resize while the NBD server is running, otherwise we don't |
1689 | * care what happens with the node. */ | |
331170e0 | 1690 | blk_get_perm(blk, &perm, &shared_perm); |
331170e0 | 1691 | ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp); |
d7086422 | 1692 | if (ret < 0) { |
331170e0 | 1693 | return ret; |
d7086422 | 1694 | } |
331170e0 | 1695 | |
4b9441f6 | 1696 | QTAILQ_INIT(&exp->clients); |
8461b4d6 | 1697 | exp->name = g_strdup(name); |
5b1cb497 | 1698 | exp->description = g_strdup(arg->description); |
dbb38caa EB |
1699 | exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH | |
1700 | NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE); | |
58a6fdcc EB |
1701 | |
1702 | if (nbd_server_max_connections() != 1) { | |
1703 | exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN; | |
1704 | } | |
dbb38caa EB |
1705 | if (readonly) { |
1706 | exp->nbdflags |= NBD_FLAG_READ_ONLY; | |
dbb38caa | 1707 | } else { |
b491dbb7 EB |
1708 | exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES | |
1709 | NBD_FLAG_SEND_FAST_ZERO); | |
dbb38caa | 1710 | } |
7596bbb3 | 1711 | exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE); |
98f44bbe | 1712 | |
cbad81ce | 1713 | for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) { |
3b1f244c EB |
1714 | exp->nr_export_bitmaps++; |
1715 | } | |
1716 | exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps); | |
1717 | for (i = 0, bitmaps = arg->bitmaps; bitmaps; | |
e5fb29d5 VSO |
1718 | i++, bitmaps = bitmaps->next) |
1719 | { | |
1720 | const char *bitmap; | |
331170e0 | 1721 | BlockDriverState *bs = blk_bs(blk); |
678ba275 | 1722 | BdrvDirtyBitmap *bm = NULL; |
678ba275 | 1723 | |
e5fb29d5 VSO |
1724 | switch (bitmaps->value->type) { |
1725 | case QTYPE_QSTRING: | |
1726 | bitmap = bitmaps->value->u.local; | |
1727 | while (bs) { | |
1728 | bm = bdrv_find_dirty_bitmap(bs, bitmap); | |
1729 | if (bm != NULL) { | |
1730 | break; | |
1731 | } | |
1732 | ||
1733 | bs = bdrv_filter_or_cow_bs(bs); | |
678ba275 EB |
1734 | } |
1735 | ||
e5fb29d5 VSO |
1736 | if (bm == NULL) { |
1737 | ret = -ENOENT; | |
1738 | error_setg(errp, "Bitmap '%s' is not found", | |
1739 | bitmaps->value->u.local); | |
1740 | goto fail; | |
1741 | } | |
678ba275 | 1742 | |
e5fb29d5 VSO |
1743 | if (readonly && bdrv_is_writable(bs) && |
1744 | bdrv_dirty_bitmap_enabled(bm)) { | |
1745 | ret = -EINVAL; | |
1746 | error_setg(errp, "Enabled bitmap '%s' incompatible with " | |
1747 | "readonly export", bitmap); | |
1748 | goto fail; | |
1749 | } | |
1750 | break; | |
1751 | case QTYPE_QDICT: | |
1752 | bitmap = bitmaps->value->u.external.name; | |
1753 | bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node, | |
1754 | bitmap, NULL, errp); | |
1755 | if (!bm) { | |
1756 | ret = -ENOENT; | |
1757 | goto fail; | |
1758 | } | |
1759 | break; | |
1760 | default: | |
1761 | abort(); | |
678ba275 EB |
1762 | } |
1763 | ||
e5fb29d5 | 1764 | assert(bm); |
3b78a927 | 1765 | |
e5fb29d5 | 1766 | if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) { |
a6ff7989 | 1767 | ret = -EINVAL; |
678ba275 EB |
1768 | goto fail; |
1769 | } | |
1770 | ||
3b1f244c | 1771 | exp->export_bitmaps[i] = bm; |
cbad81ce | 1772 | assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE); |
678ba275 EB |
1773 | } |
1774 | ||
3b1f244c EB |
1775 | /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */ |
1776 | for (i = 0; i < exp->nr_export_bitmaps; i++) { | |
1777 | bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true); | |
1778 | } | |
1779 | ||
dbc7b014 EB |
1780 | exp->allocation_depth = arg->allocation_depth; |
1781 | ||
fd6afc50 SL |
1782 | /* |
1783 | * We need to inhibit request queuing in the block layer to ensure we can | |
1784 | * be properly quiesced when entering a drained section, as our coroutines | |
1785 | * servicing pending requests might enter blk_pread(). | |
1786 | */ | |
1787 | blk_set_disable_request_queuing(blk, true); | |
1788 | ||
aadf99a7 | 1789 | blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); |
741cc431 | 1790 | |
fd6afc50 SL |
1791 | blk_set_dev_ops(blk, &nbd_block_ops, exp); |
1792 | ||
3fa4c765 | 1793 | QTAILQ_INSERT_TAIL(&exports, exp, next); |
c69de1be | 1794 | |
a6ff7989 | 1795 | return 0; |
98f44bbe HR |
1796 | |
1797 | fail: | |
3b1f244c | 1798 | g_free(exp->export_bitmaps); |
3fa4c765 EB |
1799 | g_free(exp->name); |
1800 | g_free(exp->description); | |
a6ff7989 | 1801 | return ret; |
af49bbbe PB |
1802 | } |
1803 | ||
ee0a19ec PB |
1804 | NBDExport *nbd_export_find(const char *name) |
1805 | { | |
1806 | NBDExport *exp; | |
1807 | QTAILQ_FOREACH(exp, &exports, next) { | |
1808 | if (strcmp(name, exp->name) == 0) { | |
1809 | return exp; | |
1810 | } | |
1811 | } | |
1812 | ||
1813 | return NULL; | |
1814 | } | |
1815 | ||
61bc846d EB |
1816 | AioContext * |
1817 | nbd_export_aio_context(NBDExport *exp) | |
1818 | { | |
8612c686 | 1819 | return exp->common.ctx; |
61bc846d EB |
1820 | } |
1821 | ||
bc4ee65b | 1822 | static void nbd_export_request_shutdown(BlockExport *blk_exp) |
af49bbbe | 1823 | { |
bc4ee65b | 1824 | NBDExport *exp = container_of(blk_exp, NBDExport, common); |
4b9441f6 | 1825 | NBDClient *client, *next; |
2c8d9f06 | 1826 | |
c69de1be | 1827 | blk_exp_ref(&exp->common); |
3fa4c765 EB |
1828 | /* |
1829 | * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a | |
1830 | * close mode that stops advertising the export to new clients but | |
1831 | * still permits existing clients to run to completion? Because of | |
1832 | * that possibility, nbd_export_close() can be called more than | |
1833 | * once on an export. | |
1834 | */ | |
4b9441f6 | 1835 | QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { |
0c9390d9 | 1836 | client_close(client, true); |
4b9441f6 | 1837 | } |
3fa4c765 | 1838 | if (exp->name) { |
3fa4c765 EB |
1839 | g_free(exp->name); |
1840 | exp->name = NULL; | |
1841 | QTAILQ_REMOVE(&exports, exp, next); | |
1842 | } | |
c69de1be | 1843 | blk_exp_unref(&exp->common); |
2c8d9f06 PB |
1844 | } |
1845 | ||
c69de1be | 1846 | static void nbd_export_delete(BlockExport *blk_exp) |
2c8d9f06 | 1847 | { |
3b1f244c | 1848 | size_t i; |
c69de1be | 1849 | NBDExport *exp = container_of(blk_exp, NBDExport, common); |
2c8d9f06 | 1850 | |
c69de1be KW |
1851 | assert(exp->name == NULL); |
1852 | assert(QTAILQ_EMPTY(&exp->clients)); | |
d6268348 | 1853 | |
c69de1be KW |
1854 | g_free(exp->description); |
1855 | exp->description = NULL; | |
1856 | ||
dd5b6780 PB |
1857 | if (exp->eject_notifier_blk) { |
1858 | notifier_remove(&exp->eject_notifier); | |
1859 | blk_unref(exp->eject_notifier_blk); | |
c69de1be | 1860 | } |
dd5b6780 PB |
1861 | blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached, |
1862 | blk_aio_detach, exp); | |
1863 | blk_set_disable_request_queuing(exp->common.blk, false); | |
3d068aff | 1864 | |
3b1f244c EB |
1865 | for (i = 0; i < exp->nr_export_bitmaps; i++) { |
1866 | bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false); | |
2c8d9f06 | 1867 | } |
af49bbbe PB |
1868 | } |
1869 | ||
56ee8626 KW |
1870 | const BlockExportDriver blk_exp_nbd = { |
1871 | .type = BLOCK_EXPORT_TYPE_NBD, | |
a6ff7989 | 1872 | .instance_size = sizeof(NBDExport), |
56ee8626 | 1873 | .create = nbd_export_create, |
c69de1be | 1874 | .delete = nbd_export_delete, |
bc4ee65b | 1875 | .request_shutdown = nbd_export_request_shutdown, |
56ee8626 KW |
1876 | }; |
1877 | ||
de79bfc3 VSO |
1878 | static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, |
1879 | unsigned niov, Error **errp) | |
1880 | { | |
1881 | int ret; | |
1882 | ||
1883 | g_assert(qemu_in_coroutine()); | |
1884 | qemu_co_mutex_lock(&client->send_lock); | |
1885 | client->send_coroutine = qemu_coroutine_self(); | |
1886 | ||
1887 | ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; | |
1888 | ||
1889 | client->send_coroutine = NULL; | |
1890 | qemu_co_mutex_unlock(&client->send_lock); | |
1891 | ||
1892 | return ret; | |
1893 | } | |
1894 | ||
caad5384 | 1895 | static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, |
22efd811 | 1896 | uint64_t cookie) |
caad5384 VSO |
1897 | { |
1898 | stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); | |
1899 | stl_be_p(&reply->error, error); | |
22efd811 | 1900 | stq_be_p(&reply->cookie, cookie); |
caad5384 VSO |
1901 | } |
1902 | ||
d2223cdd | 1903 | static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client, |
66d4f4fe | 1904 | NBDRequest *request, |
d2223cdd PB |
1905 | uint32_t error, |
1906 | void *data, | |
b2578459 | 1907 | uint64_t len, |
d2223cdd | 1908 | Error **errp) |
22045592 | 1909 | { |
de79bfc3 | 1910 | NBDSimpleReply reply; |
14cea41d | 1911 | int nbd_err = system_errno_to_nbd_errno(error); |
de79bfc3 VSO |
1912 | struct iovec iov[] = { |
1913 | {.iov_base = &reply, .iov_len = sizeof(reply)}, | |
1914 | {.iov_base = data, .iov_len = len} | |
1915 | }; | |
6fb2b972 | 1916 | |
a7c8ed36 | 1917 | assert(!len || !nbd_err); |
b2578459 | 1918 | assert(len <= NBD_MAX_BUFFER_SIZE); |
ac132d05 EB |
1919 | assert(client->mode < NBD_MODE_STRUCTURED || |
1920 | (client->mode == NBD_MODE_STRUCTURED && | |
1921 | request->type != NBD_CMD_READ)); | |
22efd811 | 1922 | trace_nbd_co_send_simple_reply(request->cookie, nbd_err, |
66d4f4fe | 1923 | nbd_err_lookup(nbd_err), len); |
22efd811 | 1924 | set_be_simple_reply(&reply, nbd_err, request->cookie); |
262db388 | 1925 | |
a7c8ed36 | 1926 | return nbd_co_send_iov(client, iov, 2, errp); |
22045592 PB |
1927 | } |
1928 | ||
a7c8ed36 EB |
1929 | /* |
1930 | * Prepare the header of a reply chunk for network transmission. | |
1931 | * | |
1932 | * On input, @iov is partially initialized: iov[0].iov_base must point | |
1933 | * to an uninitialized NBDReply, while the remaining @niov elements | |
1934 | * (if any) must be ready for transmission. This function then | |
1935 | * populates iov[0] for transmission. | |
1936 | */ | |
1937 | static inline void set_be_chunk(NBDClient *client, struct iovec *iov, | |
1938 | size_t niov, uint16_t flags, uint16_t type, | |
66d4f4fe | 1939 | NBDRequest *request) |
5c54e7fa | 1940 | { |
a7c8ed36 EB |
1941 | /* TODO - handle structured vs. extended replies */ |
1942 | NBDStructuredReplyChunk *chunk = iov->iov_base; | |
1943 | size_t i, length = 0; | |
1944 | ||
1945 | for (i = 1; i < niov; i++) { | |
1946 | length += iov[i].iov_len; | |
1947 | } | |
1948 | assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData)); | |
1949 | ||
1950 | iov[0].iov_len = sizeof(*chunk); | |
5c54e7fa VSO |
1951 | stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC); |
1952 | stw_be_p(&chunk->flags, flags); | |
1953 | stw_be_p(&chunk->type, type); | |
22efd811 | 1954 | stq_be_p(&chunk->cookie, request->cookie); |
5c54e7fa VSO |
1955 | stl_be_p(&chunk->length, length); |
1956 | } | |
1957 | ||
a7c8ed36 | 1958 | static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client, |
66d4f4fe | 1959 | NBDRequest *request, |
a7c8ed36 | 1960 | Error **errp) |
ef8c887e | 1961 | { |
a7c8ed36 | 1962 | NBDReply hdr; |
ef8c887e | 1963 | struct iovec iov[] = { |
a7c8ed36 | 1964 | {.iov_base = &hdr}, |
ef8c887e EB |
1965 | }; |
1966 | ||
22efd811 | 1967 | trace_nbd_co_send_chunk_done(request->cookie); |
a7c8ed36 | 1968 | set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE, |
66d4f4fe | 1969 | NBD_REPLY_TYPE_NONE, request); |
ef8c887e EB |
1970 | return nbd_co_send_iov(client, iov, 1, errp); |
1971 | } | |
1972 | ||
a7c8ed36 | 1973 | static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client, |
66d4f4fe | 1974 | NBDRequest *request, |
a7c8ed36 EB |
1975 | uint64_t offset, |
1976 | void *data, | |
b2578459 | 1977 | uint64_t size, |
a7c8ed36 EB |
1978 | bool final, |
1979 | Error **errp) | |
5c54e7fa | 1980 | { |
a7c8ed36 | 1981 | NBDReply hdr; |
efdc0c10 | 1982 | NBDStructuredReadData chunk; |
5c54e7fa | 1983 | struct iovec iov[] = { |
a7c8ed36 | 1984 | {.iov_base = &hdr}, |
5c54e7fa VSO |
1985 | {.iov_base = &chunk, .iov_len = sizeof(chunk)}, |
1986 | {.iov_base = data, .iov_len = size} | |
1987 | }; | |
1988 | ||
b2578459 | 1989 | assert(size && size <= NBD_MAX_BUFFER_SIZE); |
22efd811 | 1990 | trace_nbd_co_send_chunk_read(request->cookie, offset, data, size); |
a7c8ed36 | 1991 | set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0, |
66d4f4fe | 1992 | NBD_REPLY_TYPE_OFFSET_DATA, request); |
5c54e7fa VSO |
1993 | stq_be_p(&chunk.offset, offset); |
1994 | ||
a7c8ed36 | 1995 | return nbd_co_send_iov(client, iov, 3, errp); |
5c54e7fa | 1996 | } |
ac132d05 | 1997 | |
a7c8ed36 | 1998 | static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client, |
66d4f4fe | 1999 | NBDRequest *request, |
a7c8ed36 EB |
2000 | uint32_t error, |
2001 | const char *msg, | |
2002 | Error **errp) | |
60ace2ba | 2003 | { |
a7c8ed36 | 2004 | NBDReply hdr; |
60ace2ba VSO |
2005 | NBDStructuredError chunk; |
2006 | int nbd_err = system_errno_to_nbd_errno(error); | |
2007 | struct iovec iov[] = { | |
a7c8ed36 | 2008 | {.iov_base = &hdr}, |
60ace2ba VSO |
2009 | {.iov_base = &chunk, .iov_len = sizeof(chunk)}, |
2010 | {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, | |
2011 | }; | |
2012 | ||
2013 | assert(nbd_err); | |
22efd811 | 2014 | trace_nbd_co_send_chunk_error(request->cookie, nbd_err, |
a7c8ed36 EB |
2015 | nbd_err_lookup(nbd_err), msg ? msg : ""); |
2016 | set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE, | |
66d4f4fe | 2017 | NBD_REPLY_TYPE_ERROR, request); |
60ace2ba | 2018 | stl_be_p(&chunk.error, nbd_err); |
a7c8ed36 | 2019 | stw_be_p(&chunk.message_length, iov[2].iov_len); |
60ace2ba | 2020 | |
a7c8ed36 | 2021 | return nbd_co_send_iov(client, iov, 3, errp); |
60ace2ba VSO |
2022 | } |
2023 | ||
37e02aeb | 2024 | /* Do a sparse read and send the structured reply to the client. |
ff7e261b | 2025 | * Returns -errno if sending fails. blk_co_block_status_above() failure is |
37e02aeb VSO |
2026 | * reported to the client, at which point this function succeeds. |
2027 | */ | |
418638d3 | 2028 | static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, |
66d4f4fe | 2029 | NBDRequest *request, |
418638d3 EB |
2030 | uint64_t offset, |
2031 | uint8_t *data, | |
b2578459 | 2032 | uint64_t size, |
418638d3 EB |
2033 | Error **errp) |
2034 | { | |
2035 | int ret = 0; | |
2036 | NBDExport *exp = client->exp; | |
2037 | size_t progress = 0; | |
2038 | ||
b2578459 | 2039 | assert(size <= NBD_MAX_BUFFER_SIZE); |
418638d3 EB |
2040 | while (progress < size) { |
2041 | int64_t pnum; | |
ff7e261b EGE |
2042 | int status = blk_co_block_status_above(exp->common.blk, NULL, |
2043 | offset + progress, | |
2044 | size - progress, &pnum, NULL, | |
2045 | NULL); | |
e2de3256 | 2046 | bool final; |
418638d3 EB |
2047 | |
2048 | if (status < 0) { | |
37e02aeb VSO |
2049 | char *msg = g_strdup_printf("unable to check for holes: %s", |
2050 | strerror(-status)); | |
2051 | ||
66d4f4fe | 2052 | ret = nbd_co_send_chunk_error(client, request, -status, msg, errp); |
37e02aeb VSO |
2053 | g_free(msg); |
2054 | return ret; | |
418638d3 EB |
2055 | } |
2056 | assert(pnum && pnum <= size - progress); | |
e2de3256 | 2057 | final = progress + pnum == size; |
418638d3 | 2058 | if (status & BDRV_BLOCK_ZERO) { |
a7c8ed36 | 2059 | NBDReply hdr; |
418638d3 EB |
2060 | NBDStructuredReadHole chunk; |
2061 | struct iovec iov[] = { | |
a7c8ed36 | 2062 | {.iov_base = &hdr}, |
418638d3 EB |
2063 | {.iov_base = &chunk, .iov_len = sizeof(chunk)}, |
2064 | }; | |
2065 | ||
22efd811 | 2066 | trace_nbd_co_send_chunk_read_hole(request->cookie, |
66d4f4fe | 2067 | offset + progress, pnum); |
a7c8ed36 EB |
2068 | set_be_chunk(client, iov, 2, |
2069 | final ? NBD_REPLY_FLAG_DONE : 0, | |
66d4f4fe | 2070 | NBD_REPLY_TYPE_OFFSET_HOLE, request); |
418638d3 EB |
2071 | stq_be_p(&chunk.offset, offset + progress); |
2072 | stl_be_p(&chunk.length, pnum); | |
a7c8ed36 | 2073 | ret = nbd_co_send_iov(client, iov, 2, errp); |
418638d3 | 2074 | } else { |
d2223cdd PB |
2075 | ret = blk_co_pread(exp->common.blk, offset + progress, pnum, |
2076 | data + progress, 0); | |
418638d3 EB |
2077 | if (ret < 0) { |
2078 | error_setg_errno(errp, -ret, "reading from file failed"); | |
2079 | break; | |
2080 | } | |
66d4f4fe | 2081 | ret = nbd_co_send_chunk_read(client, request, offset + progress, |
a7c8ed36 | 2082 | data + progress, pnum, final, errp); |
418638d3 EB |
2083 | } |
2084 | ||
2085 | if (ret < 0) { | |
2086 | break; | |
2087 | } | |
2088 | progress += pnum; | |
2089 | } | |
418638d3 EB |
2090 | return ret; |
2091 | } | |
2092 | ||
89cbc7e3 | 2093 | typedef struct NBDExtentArray { |
d95ffb6f | 2094 | NBDExtent32 *extents; |
89cbc7e3 VSO |
2095 | unsigned int nb_alloc; |
2096 | unsigned int count; | |
2097 | uint64_t total_length; | |
2098 | bool can_add; | |
2099 | bool converted_to_be; | |
2100 | } NBDExtentArray; | |
2101 | ||
2102 | static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc) | |
2103 | { | |
2104 | NBDExtentArray *ea = g_new0(NBDExtentArray, 1); | |
2105 | ||
2106 | ea->nb_alloc = nb_alloc; | |
d95ffb6f | 2107 | ea->extents = g_new(NBDExtent32, nb_alloc); |
89cbc7e3 VSO |
2108 | ea->can_add = true; |
2109 | ||
2110 | return ea; | |
2111 | } | |
2112 | ||
2113 | static void nbd_extent_array_free(NBDExtentArray *ea) | |
2114 | { | |
2115 | g_free(ea->extents); | |
2116 | g_free(ea); | |
2117 | } | |
e0e7fe07 | 2118 | G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free) |
89cbc7e3 VSO |
2119 | |
2120 | /* Further modifications of the array after conversion are abandoned */ | |
2121 | static void nbd_extent_array_convert_to_be(NBDExtentArray *ea) | |
2122 | { | |
2123 | int i; | |
2124 | ||
2125 | assert(!ea->converted_to_be); | |
2126 | ea->can_add = false; | |
2127 | ea->converted_to_be = true; | |
2128 | ||
2129 | for (i = 0; i < ea->count; i++) { | |
2130 | ea->extents[i].flags = cpu_to_be32(ea->extents[i].flags); | |
2131 | ea->extents[i].length = cpu_to_be32(ea->extents[i].length); | |
2132 | } | |
2133 | } | |
2134 | ||
fb7afc79 | 2135 | /* |
89cbc7e3 VSO |
2136 | * Add extent to NBDExtentArray. If extent can't be added (no available space), |
2137 | * return -1. | |
2138 | * For safety, when returning -1 for the first time, .can_add is set to false, | |
314b9026 EB |
2139 | * and further calls to nbd_extent_array_add() will crash. |
2140 | * (this avoids the situation where a caller ignores failure to add one extent, | |
2141 | * where adding another extent that would squash into the last array entry | |
2142 | * would result in an incorrect range reported to the client) | |
fb7afc79 | 2143 | */ |
89cbc7e3 VSO |
2144 | static int nbd_extent_array_add(NBDExtentArray *ea, |
2145 | uint32_t length, uint32_t flags) | |
e7b1948d | 2146 | { |
89cbc7e3 VSO |
2147 | assert(ea->can_add); |
2148 | ||
2149 | if (!length) { | |
2150 | return 0; | |
2151 | } | |
2152 | ||
2153 | /* Extend previous extent if flags are the same */ | |
2154 | if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) { | |
2155 | uint64_t sum = (uint64_t)length + ea->extents[ea->count - 1].length; | |
2156 | ||
2157 | if (sum <= UINT32_MAX) { | |
2158 | ea->extents[ea->count - 1].length = sum; | |
2159 | ea->total_length += length; | |
2160 | return 0; | |
2161 | } | |
2162 | } | |
2163 | ||
2164 | if (ea->count >= ea->nb_alloc) { | |
2165 | ea->can_add = false; | |
2166 | return -1; | |
2167 | } | |
2168 | ||
2169 | ea->total_length += length; | |
d95ffb6f | 2170 | ea->extents[ea->count] = (NBDExtent32) {.length = length, .flags = flags}; |
89cbc7e3 | 2171 | ea->count++; |
e7b1948d | 2172 | |
89cbc7e3 VSO |
2173 | return 0; |
2174 | } | |
2175 | ||
ff7e261b | 2176 | static int coroutine_fn blockstatus_to_extents(BlockBackend *blk, |
6f58ac55 EGE |
2177 | uint64_t offset, uint64_t bytes, |
2178 | NBDExtentArray *ea) | |
89cbc7e3 VSO |
2179 | { |
2180 | while (bytes) { | |
e7b1948d VSO |
2181 | uint32_t flags; |
2182 | int64_t num; | |
ff7e261b EGE |
2183 | int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num, |
2184 | NULL, NULL); | |
fb7afc79 | 2185 | |
e7b1948d VSO |
2186 | if (ret < 0) { |
2187 | return ret; | |
2188 | } | |
2189 | ||
0da98568 NS |
2190 | flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) | |
2191 | (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0); | |
e7b1948d | 2192 | |
89cbc7e3 VSO |
2193 | if (nbd_extent_array_add(ea, num, flags) < 0) { |
2194 | return 0; | |
e7b1948d | 2195 | } |
fb7afc79 | 2196 | |
89cbc7e3 VSO |
2197 | offset += num; |
2198 | bytes -= num; | |
e7b1948d VSO |
2199 | } |
2200 | ||
e7b1948d VSO |
2201 | return 0; |
2202 | } | |
2203 | ||
ff7e261b | 2204 | static int coroutine_fn blockalloc_to_extents(BlockBackend *blk, |
6f58ac55 EGE |
2205 | uint64_t offset, uint64_t bytes, |
2206 | NBDExtentArray *ea) | |
71719cd5 EB |
2207 | { |
2208 | while (bytes) { | |
2209 | int64_t num; | |
ff7e261b EGE |
2210 | int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes, |
2211 | &num); | |
71719cd5 EB |
2212 | |
2213 | if (ret < 0) { | |
2214 | return ret; | |
2215 | } | |
2216 | ||
2217 | if (nbd_extent_array_add(ea, num, ret) < 0) { | |
2218 | return 0; | |
2219 | } | |
2220 | ||
2221 | offset += num; | |
2222 | bytes -= num; | |
2223 | } | |
2224 | ||
2225 | return 0; | |
2226 | } | |
2227 | ||
89cbc7e3 VSO |
2228 | /* |
2229 | * nbd_co_send_extents | |
3d068aff | 2230 | * |
89cbc7e3 VSO |
2231 | * @ea is converted to BE by the function |
2232 | * @last controls whether NBD_REPLY_FLAG_DONE is sent. | |
3d068aff | 2233 | */ |
d2223cdd | 2234 | static int coroutine_fn |
66d4f4fe | 2235 | nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea, |
d2223cdd | 2236 | bool last, uint32_t context_id, Error **errp) |
e7b1948d | 2237 | { |
a7c8ed36 | 2238 | NBDReply hdr; |
e7b1948d | 2239 | NBDStructuredMeta chunk; |
e7b1948d | 2240 | struct iovec iov[] = { |
a7c8ed36 | 2241 | {.iov_base = &hdr}, |
e7b1948d | 2242 | {.iov_base = &chunk, .iov_len = sizeof(chunk)}, |
89cbc7e3 | 2243 | {.iov_base = ea->extents, .iov_len = ea->count * sizeof(ea->extents[0])} |
e7b1948d VSO |
2244 | }; |
2245 | ||
89cbc7e3 VSO |
2246 | nbd_extent_array_convert_to_be(ea); |
2247 | ||
22efd811 | 2248 | trace_nbd_co_send_extents(request->cookie, ea->count, context_id, |
66d4f4fe | 2249 | ea->total_length, last); |
a7c8ed36 | 2250 | set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, |
66d4f4fe | 2251 | NBD_REPLY_TYPE_BLOCK_STATUS, request); |
e7b1948d VSO |
2252 | stl_be_p(&chunk.context_id, context_id); |
2253 | ||
a7c8ed36 | 2254 | return nbd_co_send_iov(client, iov, 3, errp); |
e7b1948d VSO |
2255 | } |
2256 | ||
2257 | /* Get block status from the exported device and send it to the client */ | |
6f58ac55 | 2258 | static int |
66d4f4fe | 2259 | coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request, |
ff7e261b | 2260 | BlockBackend *blk, uint64_t offset, |
6f58ac55 EGE |
2261 | uint32_t length, bool dont_fragment, |
2262 | bool last, uint32_t context_id, | |
2263 | Error **errp) | |
e7b1948d VSO |
2264 | { |
2265 | int ret; | |
416e34bd | 2266 | unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; |
89cbc7e3 | 2267 | g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents); |
e7b1948d | 2268 | |
71719cd5 | 2269 | if (context_id == NBD_META_ID_BASE_ALLOCATION) { |
ff7e261b | 2270 | ret = blockstatus_to_extents(blk, offset, length, ea); |
71719cd5 | 2271 | } else { |
ff7e261b | 2272 | ret = blockalloc_to_extents(blk, offset, length, ea); |
71719cd5 | 2273 | } |
e7b1948d | 2274 | if (ret < 0) { |
66d4f4fe | 2275 | return nbd_co_send_chunk_error(client, request, -ret, |
a7c8ed36 | 2276 | "can't get block status", errp); |
e7b1948d VSO |
2277 | } |
2278 | ||
66d4f4fe | 2279 | return nbd_co_send_extents(client, request, ea, last, context_id, errp); |
3d068aff VSO |
2280 | } |
2281 | ||
dacbb6eb | 2282 | /* Populate @ea from a dirty bitmap. */ |
89cbc7e3 VSO |
2283 | static void bitmap_to_extents(BdrvDirtyBitmap *bitmap, |
2284 | uint64_t offset, uint64_t length, | |
dacbb6eb | 2285 | NBDExtentArray *es) |
3d068aff | 2286 | { |
dacbb6eb VSO |
2287 | int64_t start, dirty_start, dirty_count; |
2288 | int64_t end = offset + length; | |
2289 | bool full = false; | |
3d068aff VSO |
2290 | |
2291 | bdrv_dirty_bitmap_lock(bitmap); | |
2292 | ||
dacbb6eb VSO |
2293 | for (start = offset; |
2294 | bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, INT32_MAX, | |
2295 | &dirty_start, &dirty_count); | |
2296 | start = dirty_start + dirty_count) | |
2297 | { | |
2298 | if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) || | |
2299 | (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0)) | |
2300 | { | |
2301 | full = true; | |
89cbc7e3 VSO |
2302 | break; |
2303 | } | |
3d068aff VSO |
2304 | } |
2305 | ||
dacbb6eb | 2306 | if (!full) { |
c0b21f2e EB |
2307 | /* last non dirty extent, nothing to do if array is now full */ |
2308 | (void) nbd_extent_array_add(es, end - start, 0); | |
dacbb6eb | 2309 | } |
3d068aff VSO |
2310 | |
2311 | bdrv_dirty_bitmap_unlock(bitmap); | |
3d068aff VSO |
2312 | } |
2313 | ||
66d4f4fe EB |
2314 | static int coroutine_fn nbd_co_send_bitmap(NBDClient *client, |
2315 | NBDRequest *request, | |
2316 | BdrvDirtyBitmap *bitmap, | |
2317 | uint64_t offset, | |
2318 | uint32_t length, bool dont_fragment, | |
2319 | bool last, uint32_t context_id, | |
2320 | Error **errp) | |
3d068aff | 2321 | { |
416e34bd | 2322 | unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; |
89cbc7e3 | 2323 | g_autoptr(NBDExtentArray) ea = nbd_extent_array_new(nb_extents); |
3d068aff | 2324 | |
dacbb6eb | 2325 | bitmap_to_extents(bitmap, offset, length, ea); |
3d068aff | 2326 | |
66d4f4fe | 2327 | return nbd_co_send_extents(client, request, ea, last, context_id, errp); |
e7b1948d VSO |
2328 | } |
2329 | ||
2a6e128b VSO |
2330 | /* nbd_co_receive_request |
2331 | * Collect a client request. Return 0 if request looks valid, -EIO to drop | |
f148ae7d SL |
2332 | * connection right away, -EAGAIN to indicate we were interrupted and the |
2333 | * channel should be quiesced, and any other negative value to report an error | |
2334 | * to the client (although the caller may still need to disconnect after | |
2335 | * reporting the error). | |
2a6e128b | 2336 | */ |
8db7e2d6 EB |
2337 | static int coroutine_fn nbd_co_receive_request(NBDRequestData *req, |
2338 | NBDRequest *request, | |
d2223cdd | 2339 | Error **errp) |
a030b347 | 2340 | { |
72deddc5 | 2341 | NBDClient *client = req->client; |
009cd866 | 2342 | bool extended_with_payload; |
8db7e2d6 EB |
2343 | bool check_length = false; |
2344 | bool check_rofs = false; | |
2345 | bool allocate_buffer = false; | |
009cd866 EB |
2346 | bool payload_okay = false; |
2347 | uint64_t payload_len = 0; | |
8db7e2d6 | 2348 | int valid_flags = NBD_CMD_FLAG_FUA; |
f148ae7d | 2349 | int ret; |
a030b347 | 2350 | |
1c778ef7 | 2351 | g_assert(qemu_in_coroutine()); |
ff82911c | 2352 | assert(client->recv_coroutine == qemu_coroutine_self()); |
f148ae7d SL |
2353 | ret = nbd_receive_request(client, request, errp); |
2354 | if (ret < 0) { | |
314b9026 | 2355 | return ret; |
a030b347 PB |
2356 | } |
2357 | ||
22efd811 | 2358 | trace_nbd_co_receive_request_decode_type(request->cookie, request->type, |
3736cc5b | 2359 | nbd_cmd_lookup(request->type)); |
009cd866 EB |
2360 | extended_with_payload = client->mode >= NBD_MODE_EXTENDED && |
2361 | request->flags & NBD_CMD_FLAG_PAYLOAD_LEN; | |
2362 | if (extended_with_payload) { | |
2363 | payload_len = request->len; | |
2364 | check_length = true; | |
2365 | } | |
2366 | ||
8db7e2d6 EB |
2367 | switch (request->type) { |
2368 | case NBD_CMD_DISC: | |
29b6c3b3 EB |
2369 | /* Special case: we're going to disconnect without a reply, |
2370 | * whether or not flags, from, or len are bogus */ | |
8db7e2d6 | 2371 | req->complete = true; |
ee898b87 | 2372 | return -EIO; |
29b6c3b3 | 2373 | |
8db7e2d6 EB |
2374 | case NBD_CMD_READ: |
2375 | if (client->mode >= NBD_MODE_STRUCTURED) { | |
2376 | valid_flags |= NBD_CMD_FLAG_DF; | |
eb38c3b6 | 2377 | } |
8db7e2d6 EB |
2378 | check_length = true; |
2379 | allocate_buffer = true; | |
2380 | break; | |
eb38c3b6 | 2381 | |
8db7e2d6 | 2382 | case NBD_CMD_WRITE: |
009cd866 EB |
2383 | if (client->mode >= NBD_MODE_EXTENDED) { |
2384 | if (!extended_with_payload) { | |
2385 | /* The client is noncompliant. Trace it, but proceed. */ | |
2386 | trace_nbd_co_receive_ext_payload_compliance(request->from, | |
2387 | request->len); | |
2388 | } | |
2389 | valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; | |
2390 | } | |
2391 | payload_okay = true; | |
8db7e2d6 EB |
2392 | payload_len = request->len; |
2393 | check_length = true; | |
2394 | allocate_buffer = true; | |
2395 | check_rofs = true; | |
2396 | break; | |
2397 | ||
2398 | case NBD_CMD_FLUSH: | |
2399 | break; | |
2400 | ||
2401 | case NBD_CMD_TRIM: | |
2402 | check_rofs = true; | |
2403 | break; | |
2404 | ||
2405 | case NBD_CMD_CACHE: | |
2406 | check_length = true; | |
2407 | break; | |
2408 | ||
2409 | case NBD_CMD_WRITE_ZEROES: | |
2410 | valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO; | |
2411 | check_rofs = true; | |
2412 | break; | |
2413 | ||
2414 | case NBD_CMD_BLOCK_STATUS: | |
2415 | valid_flags |= NBD_CMD_FLAG_REQ_ONE; | |
2416 | break; | |
2417 | ||
2418 | default: | |
2419 | /* Unrecognized, will fail later */ | |
2420 | ; | |
2d821488 | 2421 | } |
7fa5c565 | 2422 | |
8db7e2d6 EB |
2423 | /* Payload and buffer handling. */ |
2424 | if (!payload_len) { | |
2425 | req->complete = true; | |
2426 | } | |
2427 | if (check_length && request->len > NBD_MAX_BUFFER_SIZE) { | |
2428 | /* READ, WRITE, CACHE */ | |
2429 | error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", | |
2430 | request->len, NBD_MAX_BUFFER_SIZE); | |
2431 | return -EINVAL; | |
2432 | } | |
009cd866 EB |
2433 | if (payload_len && !payload_okay) { |
2434 | /* | |
2435 | * For now, we don't support payloads on other commands; but | |
2436 | * we can keep the connection alive by ignoring the payload. | |
2437 | * We will fail the command later with NBD_EINVAL for the use | |
2438 | * of an unsupported flag (and not for access beyond bounds). | |
2439 | */ | |
2440 | assert(request->type != NBD_CMD_WRITE); | |
2441 | request->len = 0; | |
2442 | } | |
8db7e2d6 EB |
2443 | if (allocate_buffer) { |
2444 | /* READ, WRITE */ | |
2445 | req->data = blk_try_blockalign(client->exp->common.blk, | |
2446 | request->len); | |
2447 | if (req->data == NULL) { | |
2448 | error_setg(errp, "No memory"); | |
2449 | return -ENOMEM; | |
2450 | } | |
2451 | } | |
2452 | if (payload_len) { | |
009cd866 EB |
2453 | if (payload_okay) { |
2454 | /* WRITE */ | |
2455 | assert(req->data); | |
2456 | ret = nbd_read(client->ioc, req->data, payload_len, | |
2457 | "CMD_WRITE data", errp); | |
2458 | } else { | |
2459 | ret = nbd_drop(client->ioc, payload_len, errp); | |
2460 | } | |
8db7e2d6 | 2461 | if (ret < 0) { |
ee898b87 | 2462 | return -EIO; |
a030b347 | 2463 | } |
29b6c3b3 | 2464 | req->complete = true; |
22efd811 | 2465 | trace_nbd_co_receive_request_payload_received(request->cookie, |
8db7e2d6 | 2466 | payload_len); |
a030b347 | 2467 | } |
29b6c3b3 | 2468 | |
fed5f8f8 | 2469 | /* Sanity checks. */ |
8db7e2d6 EB |
2470 | if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) { |
2471 | /* WRITE, TRIM, WRITE_ZEROES */ | |
fed5f8f8 EB |
2472 | error_setg(errp, "Export is read-only"); |
2473 | return -EROFS; | |
2474 | } | |
2475 | if (request->from > client->exp->size || | |
9d26dfcb | 2476 | request->len > client->exp->size - request->from) { |
b2578459 | 2477 | error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64 |
2fd2c840 | 2478 | ", Size: %" PRIu64, request->from, request->len, |
9d26dfcb | 2479 | client->exp->size); |
fed5f8f8 EB |
2480 | return (request->type == NBD_CMD_WRITE || |
2481 | request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; | |
29b6c3b3 | 2482 | } |
6e280648 EB |
2483 | if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len, |
2484 | client->check_align)) { | |
2485 | /* | |
2486 | * The block layer gracefully handles unaligned requests, but | |
2487 | * it's still worth tracing client non-compliance | |
2488 | */ | |
2489 | trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type), | |
2490 | request->from, | |
2491 | request->len, | |
2492 | client->check_align); | |
2493 | } | |
5c54e7fa VSO |
2494 | if (request->flags & ~valid_flags) { |
2495 | error_setg(errp, "unsupported flags for command %s (got 0x%x)", | |
2496 | nbd_cmd_lookup(request->type), request->flags); | |
ee898b87 | 2497 | return -EINVAL; |
1f4d6d18 | 2498 | } |
29b6c3b3 | 2499 | |
ee898b87 | 2500 | return 0; |
a030b347 PB |
2501 | } |
2502 | ||
6a417599 VSO |
2503 | /* Send simple reply without a payload, or a structured error |
2504 | * @error_msg is ignored if @ret >= 0 | |
2505 | * Returns 0 if connection is still live, -errno on failure to talk to client | |
2506 | */ | |
2507 | static coroutine_fn int nbd_send_generic_reply(NBDClient *client, | |
66d4f4fe | 2508 | NBDRequest *request, |
6a417599 VSO |
2509 | int ret, |
2510 | const char *error_msg, | |
2511 | Error **errp) | |
2512 | { | |
ac132d05 | 2513 | if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) { |
66d4f4fe | 2514 | return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp); |
6a417599 | 2515 | } else { |
66d4f4fe | 2516 | return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0, |
6a417599 VSO |
2517 | NULL, 0, errp); |
2518 | } | |
2519 | } | |
2520 | ||
2521 | /* Handle NBD_CMD_READ request. | |
2522 | * Return -errno if sending fails. Other errors are reported directly to the | |
2523 | * client as an error reply. */ | |
2524 | static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request, | |
2525 | uint8_t *data, Error **errp) | |
2526 | { | |
2527 | int ret; | |
2528 | NBDExport *exp = client->exp; | |
2529 | ||
7fa5c565 | 2530 | assert(request->type == NBD_CMD_READ); |
b2578459 | 2531 | assert(request->len <= NBD_MAX_BUFFER_SIZE); |
6a417599 VSO |
2532 | |
2533 | /* XXX: NBD Protocol only documents use of FUA with WRITE */ | |
2534 | if (request->flags & NBD_CMD_FLAG_FUA) { | |
37a4f70c | 2535 | ret = blk_co_flush(exp->common.blk); |
6a417599 | 2536 | if (ret < 0) { |
66d4f4fe | 2537 | return nbd_send_generic_reply(client, request, ret, |
6a417599 VSO |
2538 | "flush failed", errp); |
2539 | } | |
2540 | } | |
2541 | ||
ac132d05 EB |
2542 | if (client->mode >= NBD_MODE_STRUCTURED && |
2543 | !(request->flags & NBD_CMD_FLAG_DF) && request->len) | |
2f454def | 2544 | { |
66d4f4fe | 2545 | return nbd_co_send_sparse_read(client, request, request->from, |
6a417599 VSO |
2546 | data, request->len, errp); |
2547 | } | |
2548 | ||
d2223cdd | 2549 | ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0); |
7fa5c565 | 2550 | if (ret < 0) { |
66d4f4fe | 2551 | return nbd_send_generic_reply(client, request, ret, |
6a417599 VSO |
2552 | "reading from file failed", errp); |
2553 | } | |
2554 | ||
ac132d05 | 2555 | if (client->mode >= NBD_MODE_STRUCTURED) { |
6a417599 | 2556 | if (request->len) { |
66d4f4fe | 2557 | return nbd_co_send_chunk_read(client, request, request->from, data, |
a7c8ed36 | 2558 | request->len, true, errp); |
6a417599 | 2559 | } else { |
66d4f4fe | 2560 | return nbd_co_send_chunk_done(client, request, errp); |
6a417599 VSO |
2561 | } |
2562 | } else { | |
66d4f4fe | 2563 | return nbd_co_send_simple_reply(client, request, 0, |
6a417599 VSO |
2564 | data, request->len, errp); |
2565 | } | |
2566 | } | |
2567 | ||
7fa5c565 VSO |
2568 | /* |
2569 | * nbd_do_cmd_cache | |
2570 | * | |
2571 | * Handle NBD_CMD_CACHE request. | |
2572 | * Return -errno if sending fails. Other errors are reported directly to the | |
2573 | * client as an error reply. | |
2574 | */ | |
2575 | static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request, | |
2576 | Error **errp) | |
2577 | { | |
2578 | int ret; | |
2579 | NBDExport *exp = client->exp; | |
2580 | ||
2581 | assert(request->type == NBD_CMD_CACHE); | |
b2578459 | 2582 | assert(request->len <= NBD_MAX_BUFFER_SIZE); |
7fa5c565 | 2583 | |
37a4f70c | 2584 | ret = blk_co_preadv(exp->common.blk, request->from, request->len, |
7fa5c565 VSO |
2585 | NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH); |
2586 | ||
66d4f4fe | 2587 | return nbd_send_generic_reply(client, request, ret, |
7fa5c565 VSO |
2588 | "caching data failed", errp); |
2589 | } | |
2590 | ||
6f302e60 VSO |
2591 | /* Handle NBD request. |
2592 | * Return -errno if sending fails. Other errors are reported directly to the | |
2593 | * client as an error reply. */ | |
2594 | static coroutine_fn int nbd_handle_request(NBDClient *client, | |
2595 | NBDRequest *request, | |
2596 | uint8_t *data, Error **errp) | |
2597 | { | |
2598 | int ret; | |
2599 | int flags; | |
2600 | NBDExport *exp = client->exp; | |
2601 | char *msg; | |
3b1f244c | 2602 | size_t i; |
6f302e60 VSO |
2603 | |
2604 | switch (request->type) { | |
bc37b06a | 2605 | case NBD_CMD_CACHE: |
7fa5c565 VSO |
2606 | return nbd_do_cmd_cache(client, request, errp); |
2607 | ||
2608 | case NBD_CMD_READ: | |
6f302e60 VSO |
2609 | return nbd_do_cmd_read(client, request, data, errp); |
2610 | ||
2611 | case NBD_CMD_WRITE: | |
2612 | flags = 0; | |
2613 | if (request->flags & NBD_CMD_FLAG_FUA) { | |
2614 | flags |= BDRV_REQ_FUA; | |
2615 | } | |
b2578459 | 2616 | assert(request->len <= NBD_MAX_BUFFER_SIZE); |
d2223cdd PB |
2617 | ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data, |
2618 | flags); | |
66d4f4fe | 2619 | return nbd_send_generic_reply(client, request, ret, |
6f302e60 VSO |
2620 | "writing to file failed", errp); |
2621 | ||
2622 | case NBD_CMD_WRITE_ZEROES: | |
2623 | flags = 0; | |
2624 | if (request->flags & NBD_CMD_FLAG_FUA) { | |
2625 | flags |= BDRV_REQ_FUA; | |
2626 | } | |
2627 | if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) { | |
2628 | flags |= BDRV_REQ_MAY_UNMAP; | |
2629 | } | |
b491dbb7 EB |
2630 | if (request->flags & NBD_CMD_FLAG_FAST_ZERO) { |
2631 | flags |= BDRV_REQ_NO_FALLBACK; | |
2632 | } | |
d2223cdd PB |
2633 | ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len, |
2634 | flags); | |
66d4f4fe | 2635 | return nbd_send_generic_reply(client, request, ret, |
6f302e60 VSO |
2636 | "writing to file failed", errp); |
2637 | ||
2638 | case NBD_CMD_DISC: | |
2639 | /* unreachable, thanks to special case in nbd_co_receive_request() */ | |
2640 | abort(); | |
2641 | ||
2642 | case NBD_CMD_FLUSH: | |
37a4f70c | 2643 | ret = blk_co_flush(exp->common.blk); |
66d4f4fe | 2644 | return nbd_send_generic_reply(client, request, ret, |
6f302e60 VSO |
2645 | "flush failed", errp); |
2646 | ||
2647 | case NBD_CMD_TRIM: | |
e3557422 | 2648 | ret = blk_co_pdiscard(exp->common.blk, request->from, request->len); |
890cbccb | 2649 | if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) { |
37a4f70c | 2650 | ret = blk_co_flush(exp->common.blk); |
65529782 | 2651 | } |
66d4f4fe | 2652 | return nbd_send_generic_reply(client, request, ret, |
6f302e60 VSO |
2653 | "discard failed", errp); |
2654 | ||
e7b1948d | 2655 | case NBD_CMD_BLOCK_STATUS: |
d8b20291 | 2656 | if (!request->len) { |
66d4f4fe | 2657 | return nbd_send_generic_reply(client, request, -EINVAL, |
d8b20291 EB |
2658 | "need non-zero length", errp); |
2659 | } | |
b2578459 | 2660 | assert(request->len <= UINT32_MAX); |
47ec485e | 2661 | if (client->export_meta.count) { |
fb7afc79 | 2662 | bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE; |
47ec485e | 2663 | int contexts_remaining = client->export_meta.count; |
fb7afc79 | 2664 | |
3d068aff | 2665 | if (client->export_meta.base_allocation) { |
66d4f4fe | 2666 | ret = nbd_co_send_block_status(client, request, |
ff7e261b | 2667 | exp->common.blk, |
37a4f70c | 2668 | request->from, |
fb7afc79 | 2669 | request->len, dont_fragment, |
47ec485e | 2670 | !--contexts_remaining, |
3d068aff VSO |
2671 | NBD_META_ID_BASE_ALLOCATION, |
2672 | errp); | |
73e064cc EB |
2673 | if (ret < 0) { |
2674 | return ret; | |
2675 | } | |
2676 | } | |
2677 | ||
71719cd5 | 2678 | if (client->export_meta.allocation_depth) { |
66d4f4fe | 2679 | ret = nbd_co_send_block_status(client, request, |
ff7e261b | 2680 | exp->common.blk, |
71719cd5 EB |
2681 | request->from, request->len, |
2682 | dont_fragment, | |
2683 | !--contexts_remaining, | |
2684 | NBD_META_ID_ALLOCATION_DEPTH, | |
2685 | errp); | |
2686 | if (ret < 0) { | |
2687 | return ret; | |
2688 | } | |
2689 | } | |
2690 | ||
3b1f244c EB |
2691 | for (i = 0; i < client->exp->nr_export_bitmaps; i++) { |
2692 | if (!client->export_meta.bitmaps[i]) { | |
2693 | continue; | |
2694 | } | |
66d4f4fe | 2695 | ret = nbd_co_send_bitmap(client, request, |
3b1f244c | 2696 | client->exp->export_bitmaps[i], |
3d068aff | 2697 | request->from, request->len, |
47ec485e | 2698 | dont_fragment, !--contexts_remaining, |
3b1f244c | 2699 | NBD_META_ID_DIRTY_BITMAP + i, errp); |
73e064cc EB |
2700 | if (ret < 0) { |
2701 | return ret; | |
2702 | } | |
3d068aff VSO |
2703 | } |
2704 | ||
47ec485e EB |
2705 | assert(!contexts_remaining); |
2706 | ||
73e064cc | 2707 | return 0; |
e7b1948d | 2708 | } else { |
66d4f4fe | 2709 | return nbd_send_generic_reply(client, request, -EINVAL, |
e7b1948d VSO |
2710 | "CMD_BLOCK_STATUS not negotiated", |
2711 | errp); | |
2712 | } | |
2713 | ||
6f302e60 VSO |
2714 | default: |
2715 | msg = g_strdup_printf("invalid request type (%" PRIu32 ") received", | |
2716 | request->type); | |
66d4f4fe | 2717 | ret = nbd_send_generic_reply(client, request, -EINVAL, msg, |
6f302e60 VSO |
2718 | errp); |
2719 | g_free(msg); | |
2720 | return ret; | |
2721 | } | |
2722 | } | |
2723 | ||
ff82911c PB |
2724 | /* Owns a reference to the NBDClient passed as opaque. */ |
2725 | static coroutine_fn void nbd_trip(void *opaque) | |
75818250 | 2726 | { |
262db388 | 2727 | NBDClient *client = opaque; |
315f78ab | 2728 | NBDRequestData *req; |
ff82911c | 2729 | NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ |
a0dc63a6 | 2730 | int ret; |
2fd2c840 | 2731 | Error *local_err = NULL; |
b2e3d87f | 2732 | |
9588463e | 2733 | trace_nbd_trip(); |
ff2b68aa | 2734 | if (client->closing) { |
ff82911c | 2735 | nbd_client_put(client); |
ff2b68aa PB |
2736 | return; |
2737 | } | |
b2e3d87f | 2738 | |
f148ae7d SL |
2739 | if (client->quiescing) { |
2740 | /* | |
2741 | * We're switching between AIO contexts. Don't attempt to receive a new | |
2742 | * request and kick the main context which may be waiting for us. | |
2743 | */ | |
2744 | nbd_client_put(client); | |
2745 | client->recv_coroutine = NULL; | |
2746 | aio_wait_kick(); | |
2747 | return; | |
2748 | } | |
2749 | ||
ff2b68aa | 2750 | req = nbd_request_get(client); |
2fd2c840 | 2751 | ret = nbd_co_receive_request(req, &request, &local_err); |
ee898b87 | 2752 | client->recv_coroutine = NULL; |
b2e3d87f | 2753 | |
d6268348 WC |
2754 | if (client->closing) { |
2755 | /* | |
2756 | * The client may be closed when we are blocked in | |
2757 | * nbd_co_receive_request() | |
2758 | */ | |
2759 | goto done; | |
2760 | } | |
2761 | ||
f148ae7d SL |
2762 | if (ret == -EAGAIN) { |
2763 | assert(client->quiescing); | |
2764 | goto done; | |
2765 | } | |
2766 | ||
a0d7ce20 VSO |
2767 | nbd_client_receive_next_request(client); |
2768 | if (ret == -EIO) { | |
2769 | goto disconnect; | |
2770 | } | |
2771 | ||
bd2cd4a4 FW |
2772 | qio_channel_set_cork(client->ioc, true); |
2773 | ||
a0d7ce20 | 2774 | if (ret < 0) { |
314b9026 | 2775 | /* It wasn't -EIO, so, according to nbd_co_receive_request() |
6a417599 VSO |
2776 | * semantics, we should return the error to the client. */ |
2777 | Error *export_err = local_err; | |
2778 | ||
2779 | local_err = NULL; | |
66d4f4fe | 2780 | ret = nbd_send_generic_reply(client, &request, -EINVAL, |
6a417599 VSO |
2781 | error_get_pretty(export_err), &local_err); |
2782 | error_free(export_err); | |
6f302e60 VSO |
2783 | } else { |
2784 | ret = nbd_handle_request(client, &request, req->data, &local_err); | |
5c54e7fa VSO |
2785 | } |
2786 | if (ret < 0) { | |
c7b97282 | 2787 | error_prepend(&local_err, "Failed to send reply: "); |
2fd2c840 VSO |
2788 | goto disconnect; |
2789 | } | |
2790 | ||
8c372a02 VSO |
2791 | /* We must disconnect after NBD_CMD_WRITE if we did not |
2792 | * read the payload. | |
2793 | */ | |
2fd2c840 VSO |
2794 | if (!req->complete) { |
2795 | error_setg(&local_err, "Request handling failed in intermediate state"); | |
8c372a02 | 2796 | goto disconnect; |
b2e3d87f NT |
2797 | } |
2798 | ||
bd2cd4a4 | 2799 | qio_channel_set_cork(client->ioc, false); |
7fe7b68b | 2800 | done: |
262db388 | 2801 | nbd_request_put(req); |
ff82911c | 2802 | nbd_client_put(client); |
262db388 PB |
2803 | return; |
2804 | ||
8c372a02 | 2805 | disconnect: |
2fd2c840 VSO |
2806 | if (local_err) { |
2807 | error_reportf_err(local_err, "Disconnect client, due to: "); | |
2808 | } | |
72deddc5 | 2809 | nbd_request_put(req); |
0c9390d9 | 2810 | client_close(client, true); |
ff82911c | 2811 | nbd_client_put(client); |
7a5ca864 | 2812 | } |
af49bbbe | 2813 | |
ff82911c | 2814 | static void nbd_client_receive_next_request(NBDClient *client) |
958c717d | 2815 | { |
f148ae7d SL |
2816 | if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS && |
2817 | !client->quiescing) { | |
ff82911c PB |
2818 | nbd_client_get(client); |
2819 | client->recv_coroutine = qemu_coroutine_create(nbd_trip, client); | |
8612c686 | 2820 | aio_co_schedule(client->exp->common.ctx, client->recv_coroutine); |
958c717d HR |
2821 | } |
2822 | } | |
2823 | ||
1a6245a5 FZ |
2824 | static coroutine_fn void nbd_co_client_start(void *opaque) |
2825 | { | |
c84087f2 | 2826 | NBDClient *client = opaque; |
2fd2c840 | 2827 | Error *local_err = NULL; |
1a6245a5 | 2828 | |
df8ad9f1 EB |
2829 | qemu_co_mutex_init(&client->send_lock); |
2830 | ||
2fd2c840 VSO |
2831 | if (nbd_negotiate(client, &local_err)) { |
2832 | if (local_err) { | |
2833 | error_report_err(local_err); | |
2834 | } | |
0c9390d9 | 2835 | client_close(client, false); |
c84087f2 | 2836 | return; |
1a6245a5 | 2837 | } |
ff82911c PB |
2838 | |
2839 | nbd_client_receive_next_request(client); | |
1a6245a5 FZ |
2840 | } |
2841 | ||
0c9390d9 | 2842 | /* |
7f7dfe2a VSO |
2843 | * Create a new client listener using the given channel @sioc. |
2844 | * Begin servicing it in a coroutine. When the connection closes, call | |
2845 | * @close_fn with an indication of whether the client completed negotiation. | |
0c9390d9 | 2846 | */ |
7f7dfe2a | 2847 | void nbd_client_new(QIOChannelSocket *sioc, |
f95910fe | 2848 | QCryptoTLSCreds *tlscreds, |
b25e12da | 2849 | const char *tlsauthz, |
0c9390d9 | 2850 | void (*close_fn)(NBDClient *, bool)) |
af49bbbe | 2851 | { |
1743b515 | 2852 | NBDClient *client; |
c84087f2 | 2853 | Coroutine *co; |
1a6245a5 | 2854 | |
e8d3eb74 | 2855 | client = g_new0(NBDClient, 1); |
1743b515 | 2856 | client->refcount = 1; |
f95910fe DB |
2857 | client->tlscreds = tlscreds; |
2858 | if (tlscreds) { | |
2859 | object_ref(OBJECT(client->tlscreds)); | |
2860 | } | |
b25e12da | 2861 | client->tlsauthz = g_strdup(tlsauthz); |
1c778ef7 | 2862 | client->sioc = sioc; |
f1426881 | 2863 | qio_channel_set_delay(QIO_CHANNEL(sioc), false); |
1c778ef7 DB |
2864 | object_ref(OBJECT(client->sioc)); |
2865 | client->ioc = QIO_CHANNEL(sioc); | |
2866 | object_ref(OBJECT(client->ioc)); | |
0c9390d9 | 2867 | client->close_fn = close_fn; |
2c8d9f06 | 2868 | |
c84087f2 VSO |
2869 | co = qemu_coroutine_create(nbd_co_client_start, client); |
2870 | qemu_coroutine_enter(co); | |
af49bbbe | 2871 | } |