]>
Commit | Line | Data |
---|---|---|
75818250 | 1 | /* |
7a5ca864 FB |
2 | * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws> |
3 | * | |
4 | * Network Block Device | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; under version 2 of the License. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
8167ee88 | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
75818250 | 17 | */ |
7a5ca864 | 18 | |
737e150e PB |
19 | #include "block/nbd.h" |
20 | #include "block/block.h" | |
7a5ca864 | 21 | |
737e150e | 22 | #include "block/coroutine.h" |
262db388 | 23 | |
7a5ca864 FB |
24 | #include <errno.h> |
25 | #include <string.h> | |
03ff3ca3 | 26 | #ifndef _WIN32 |
7a5ca864 | 27 | #include <sys/ioctl.h> |
03ff3ca3 | 28 | #endif |
5dc2eec9 | 29 | #if defined(__sun__) || defined(__HAIKU__) |
7e00eb9b AL |
30 | #include <sys/ioccom.h> |
31 | #endif | |
7a5ca864 FB |
32 | #include <ctype.h> |
33 | #include <inttypes.h> | |
75818250 | 34 | |
b90fb4b8 PB |
35 | #ifdef __linux__ |
36 | #include <linux/fs.h> | |
37 | #endif | |
38 | ||
1de7afc9 PB |
39 | #include "qemu/sockets.h" |
40 | #include "qemu/queue.h" | |
6a1751b7 | 41 | #include "qemu/main-loop.h" |
03ff3ca3 AL |
42 | |
43 | //#define DEBUG_NBD | |
44 | ||
45 | #ifdef DEBUG_NBD | |
75818250 | 46 | #define TRACE(msg, ...) do { \ |
03ff3ca3 | 47 | LOG(msg, ## __VA_ARGS__); \ |
75818250 | 48 | } while(0) |
03ff3ca3 AL |
49 | #else |
50 | #define TRACE(msg, ...) \ | |
51 | do { } while (0) | |
52 | #endif | |
7a5ca864 FB |
53 | |
54 | #define LOG(msg, ...) do { \ | |
55 | fprintf(stderr, "%s:%s():L%d: " msg "\n", \ | |
56 | __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \ | |
57 | } while(0) | |
58 | ||
f5076b5a HB |
59 | /* This is all part of the "official" NBD API. |
60 | * | |
61 | * The most up-to-date documentation is available at: | |
62 | * https://github.com/yoe/nbd/blob/master/doc/proto.txt | |
63 | */ | |
7a5ca864 | 64 | |
fa26c26b | 65 | #define NBD_REQUEST_SIZE (4 + 4 + 8 + 8 + 4) |
b2e3d87f | 66 | #define NBD_REPLY_SIZE (4 + 4 + 8) |
7a5ca864 FB |
67 | #define NBD_REQUEST_MAGIC 0x25609513 |
68 | #define NBD_REPLY_MAGIC 0x67446698 | |
fa26c26b PB |
69 | #define NBD_OPTS_MAGIC 0x49484156454F5054LL |
70 | #define NBD_CLIENT_MAGIC 0x0000420281861253LL | |
f5076b5a | 71 | #define NBD_REP_MAGIC 0x3e889045565a9LL |
7a5ca864 FB |
72 | |
73 | #define NBD_SET_SOCK _IO(0xab, 0) | |
74 | #define NBD_SET_BLKSIZE _IO(0xab, 1) | |
75 | #define NBD_SET_SIZE _IO(0xab, 2) | |
76 | #define NBD_DO_IT _IO(0xab, 3) | |
77 | #define NBD_CLEAR_SOCK _IO(0xab, 4) | |
78 | #define NBD_CLEAR_QUE _IO(0xab, 5) | |
b2e3d87f NT |
79 | #define NBD_PRINT_DEBUG _IO(0xab, 6) |
80 | #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7) | |
7a5ca864 | 81 | #define NBD_DISCONNECT _IO(0xab, 8) |
bbb74edd PB |
82 | #define NBD_SET_TIMEOUT _IO(0xab, 9) |
83 | #define NBD_SET_FLAGS _IO(0xab, 10) | |
7a5ca864 | 84 | |
f5076b5a HB |
85 | #define NBD_OPT_EXPORT_NAME (1) |
86 | #define NBD_OPT_ABORT (2) | |
1d45f8b5 | 87 | |
9a304d29 PB |
88 | /* Definitions for opaque data types */ |
89 | ||
90 | typedef struct NBDRequest NBDRequest; | |
91 | ||
92 | struct NBDRequest { | |
93 | QSIMPLEQ_ENTRY(NBDRequest) entry; | |
94 | NBDClient *client; | |
95 | uint8_t *data; | |
96 | }; | |
97 | ||
98 | struct NBDExport { | |
2c8d9f06 | 99 | int refcount; |
0ddf08db PB |
100 | void (*close)(NBDExport *exp); |
101 | ||
9a304d29 | 102 | BlockDriverState *bs; |
ee0a19ec | 103 | char *name; |
9a304d29 PB |
104 | off_t dev_offset; |
105 | off_t size; | |
106 | uint32_t nbdflags; | |
4b9441f6 | 107 | QTAILQ_HEAD(, NBDClient) clients; |
ee0a19ec | 108 | QTAILQ_ENTRY(NBDExport) next; |
9a304d29 PB |
109 | }; |
110 | ||
ee0a19ec PB |
111 | static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); |
112 | ||
9a304d29 PB |
113 | struct NBDClient { |
114 | int refcount; | |
115 | void (*close)(NBDClient *client); | |
116 | ||
117 | NBDExport *exp; | |
118 | int sock; | |
119 | ||
120 | Coroutine *recv_coroutine; | |
121 | ||
122 | CoMutex send_lock; | |
123 | Coroutine *send_coroutine; | |
124 | ||
4b9441f6 | 125 | QTAILQ_ENTRY(NBDClient) next; |
9a304d29 | 126 | int nb_requests; |
ff2b68aa | 127 | bool closing; |
9a304d29 PB |
128 | }; |
129 | ||
7a5ca864 FB |
130 | /* That's all folks */ |
131 | ||
185b4338 | 132 | ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read) |
7a5ca864 FB |
133 | { |
134 | size_t offset = 0; | |
185b4338 | 135 | int err; |
7a5ca864 | 136 | |
ae255e52 PB |
137 | if (qemu_in_coroutine()) { |
138 | if (do_read) { | |
139 | return qemu_co_recv(fd, buffer, size); | |
140 | } else { | |
141 | return qemu_co_send(fd, buffer, size); | |
142 | } | |
143 | } | |
144 | ||
7a5ca864 FB |
145 | while (offset < size) { |
146 | ssize_t len; | |
147 | ||
148 | if (do_read) { | |
00aa0040 | 149 | len = qemu_recv(fd, buffer + offset, size - offset, 0); |
7a5ca864 | 150 | } else { |
03ff3ca3 | 151 | len = send(fd, buffer + offset, size - offset, 0); |
7a5ca864 FB |
152 | } |
153 | ||
fc19f8a0 | 154 | if (len < 0) { |
185b4338 | 155 | err = socket_error(); |
03ff3ca3 | 156 | |
fc19f8a0 | 157 | /* recoverable error */ |
7fe7b68b | 158 | if (err == EINTR || (offset > 0 && err == EAGAIN)) { |
fc19f8a0 PB |
159 | continue; |
160 | } | |
161 | ||
162 | /* unrecoverable error */ | |
185b4338 | 163 | return -err; |
7a5ca864 FB |
164 | } |
165 | ||
166 | /* eof */ | |
167 | if (len == 0) { | |
168 | break; | |
169 | } | |
170 | ||
7a5ca864 FB |
171 | offset += len; |
172 | } | |
173 | ||
174 | return offset; | |
175 | } | |
176 | ||
7fe7b68b PB |
177 | static ssize_t read_sync(int fd, void *buffer, size_t size) |
178 | { | |
179 | /* Sockets are kept in blocking mode in the negotiation phase. After | |
180 | * that, a non-readable socket simply means that another thread stole | |
181 | * our request/reply. Synchronization is done with recv_coroutine, so | |
182 | * that this is coroutine-safe. | |
183 | */ | |
184 | return nbd_wr_sync(fd, buffer, size, true); | |
185 | } | |
186 | ||
187 | static ssize_t write_sync(int fd, void *buffer, size_t size) | |
188 | { | |
189 | int ret; | |
190 | do { | |
191 | /* For writes, we do expect the socket to be writable. */ | |
192 | ret = nbd_wr_sync(fd, buffer, size, false); | |
193 | } while (ret == -EAGAIN); | |
194 | return ret; | |
195 | } | |
196 | ||
6b8c01e7 | 197 | /* Basic flow for negotiation |
7a5ca864 FB |
198 | |
199 | Server Client | |
7a5ca864 | 200 | Negotiate |
6b8c01e7 PB |
201 | |
202 | or | |
203 | ||
204 | Server Client | |
205 | Negotiate #1 | |
206 | Option | |
207 | Negotiate #2 | |
208 | ||
209 | ---- | |
210 | ||
211 | followed by | |
212 | ||
213 | Server Client | |
7a5ca864 FB |
214 | Request |
215 | Response | |
216 | Request | |
217 | Response | |
218 | ... | |
219 | ... | |
220 | Request (type == 2) | |
6b8c01e7 | 221 | |
7a5ca864 FB |
222 | */ |
223 | ||
f5076b5a | 224 | static int nbd_send_rep(int csock, uint32_t type, uint32_t opt) |
6b8c01e7 | 225 | { |
6b8c01e7 | 226 | uint64_t magic; |
f5076b5a | 227 | uint32_t len; |
6b8c01e7 | 228 | |
f5076b5a HB |
229 | magic = cpu_to_be64(NBD_REP_MAGIC); |
230 | if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
231 | LOG("write failed (rep magic)"); | |
232 | return -EINVAL; | |
6b8c01e7 | 233 | } |
f5076b5a HB |
234 | opt = cpu_to_be32(opt); |
235 | if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) { | |
236 | LOG("write failed (rep opt)"); | |
237 | return -EINVAL; | |
6b8c01e7 | 238 | } |
f5076b5a HB |
239 | type = cpu_to_be32(type); |
240 | if (write_sync(csock, &type, sizeof(type)) != sizeof(type)) { | |
241 | LOG("write failed (rep type)"); | |
242 | return -EINVAL; | |
6b8c01e7 | 243 | } |
f5076b5a HB |
244 | len = cpu_to_be32(0); |
245 | if (write_sync(csock, &len, sizeof(len)) != sizeof(len)) { | |
246 | LOG("write failed (rep data length)"); | |
247 | return -EINVAL; | |
6b8c01e7 | 248 | } |
f5076b5a HB |
249 | return 0; |
250 | } | |
6b8c01e7 | 251 | |
f5076b5a HB |
252 | static int nbd_handle_export_name(NBDClient *client, uint32_t length) |
253 | { | |
254 | int rc = -EINVAL, csock = client->sock; | |
255 | char name[256]; | |
6b8c01e7 | 256 | |
f5076b5a HB |
257 | /* Client sends: |
258 | [20 .. xx] export name (length bytes) | |
259 | */ | |
6b8c01e7 | 260 | TRACE("Checking length"); |
6b8c01e7 PB |
261 | if (length > 255) { |
262 | LOG("Bad length received"); | |
263 | goto fail; | |
264 | } | |
265 | if (read_sync(csock, name, length) != length) { | |
266 | LOG("read failed"); | |
267 | goto fail; | |
268 | } | |
269 | name[length] = '\0'; | |
270 | ||
271 | client->exp = nbd_export_find(name); | |
272 | if (!client->exp) { | |
273 | LOG("export not found"); | |
274 | goto fail; | |
275 | } | |
276 | ||
277 | QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); | |
278 | nbd_export_get(client->exp); | |
6b8c01e7 PB |
279 | rc = 0; |
280 | fail: | |
281 | return rc; | |
282 | } | |
283 | ||
f5076b5a HB |
284 | static int nbd_receive_options(NBDClient *client) |
285 | { | |
286 | while (1) { | |
287 | int csock = client->sock; | |
288 | uint32_t tmp, length; | |
289 | uint64_t magic; | |
290 | ||
291 | /* Client sends: | |
292 | [ 0 .. 3] client flags | |
293 | [ 4 .. 11] NBD_OPTS_MAGIC | |
294 | [12 .. 15] NBD option | |
295 | [16 .. 19] length | |
296 | ... Rest of request | |
297 | */ | |
298 | ||
299 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
300 | LOG("read failed"); | |
301 | return -EINVAL; | |
302 | } | |
303 | TRACE("Checking client flags"); | |
304 | tmp = be32_to_cpu(tmp); | |
305 | if (tmp != 0 && tmp != NBD_FLAG_C_FIXED_NEWSTYLE) { | |
306 | LOG("Bad client flags received"); | |
307 | return -EINVAL; | |
308 | } | |
309 | ||
310 | if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
311 | LOG("read failed"); | |
312 | return -EINVAL; | |
313 | } | |
314 | TRACE("Checking opts magic"); | |
315 | if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) { | |
316 | LOG("Bad magic received"); | |
317 | return -EINVAL; | |
318 | } | |
319 | ||
320 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
321 | LOG("read failed"); | |
322 | return -EINVAL; | |
323 | } | |
324 | ||
325 | if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) { | |
326 | LOG("read failed"); | |
327 | return -EINVAL; | |
328 | } | |
329 | length = be32_to_cpu(length); | |
330 | ||
331 | TRACE("Checking option"); | |
332 | switch (be32_to_cpu(tmp)) { | |
333 | case NBD_OPT_ABORT: | |
334 | return -EINVAL; | |
335 | ||
336 | case NBD_OPT_EXPORT_NAME: | |
337 | return nbd_handle_export_name(client, length); | |
338 | ||
339 | default: | |
340 | tmp = be32_to_cpu(tmp); | |
341 | LOG("Unsupported option 0x%x", tmp); | |
342 | nbd_send_rep(client->sock, NBD_REP_ERR_UNSUP, tmp); | |
343 | return -EINVAL; | |
344 | } | |
345 | } | |
346 | } | |
347 | ||
9a304d29 | 348 | static int nbd_send_negotiate(NBDClient *client) |
7a5ca864 | 349 | { |
9a304d29 | 350 | int csock = client->sock; |
b2e3d87f | 351 | char buf[8 + 8 + 8 + 128]; |
185b4338 | 352 | int rc; |
6b8c01e7 PB |
353 | const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM | |
354 | NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA); | |
b2e3d87f | 355 | |
6b8c01e7 PB |
356 | /* Negotiation header without options: |
357 | [ 0 .. 7] passwd ("NBDMAGIC") | |
358 | [ 8 .. 15] magic (NBD_CLIENT_MAGIC) | |
b2e3d87f | 359 | [16 .. 23] size |
6b8c01e7 | 360 | [24 .. 25] server flags (0) |
5672ee54 | 361 | [26 .. 27] export flags |
6b8c01e7 PB |
362 | [28 .. 151] reserved (0) |
363 | ||
364 | Negotiation header with options, part 1: | |
365 | [ 0 .. 7] passwd ("NBDMAGIC") | |
366 | [ 8 .. 15] magic (NBD_OPTS_MAGIC) | |
367 | [16 .. 17] server flags (0) | |
368 | ||
369 | part 2 (after options are sent): | |
370 | [18 .. 25] size | |
371 | [26 .. 27] export flags | |
372 | [28 .. 151] reserved (0) | |
b2e3d87f NT |
373 | */ |
374 | ||
f9e8cacc | 375 | qemu_set_block(csock); |
185b4338 PB |
376 | rc = -EINVAL; |
377 | ||
b2e3d87f | 378 | TRACE("Beginning negotiation."); |
8ffaaba0 | 379 | memset(buf, 0, sizeof(buf)); |
b2e3d87f | 380 | memcpy(buf, "NBDMAGIC", 8); |
6b8c01e7 PB |
381 | if (client->exp) { |
382 | assert ((client->exp->nbdflags & ~65535) == 0); | |
383 | cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC); | |
384 | cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size); | |
385 | cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags); | |
386 | } else { | |
387 | cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC); | |
f5076b5a | 388 | cpu_to_be16w((uint16_t *)(buf + 16), NBD_FLAG_FIXED_NEWSTYLE); |
6b8c01e7 | 389 | } |
b2e3d87f | 390 | |
6b8c01e7 PB |
391 | if (client->exp) { |
392 | if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
393 | LOG("write failed"); | |
394 | goto fail; | |
395 | } | |
396 | } else { | |
397 | if (write_sync(csock, buf, 18) != 18) { | |
398 | LOG("write failed"); | |
399 | goto fail; | |
400 | } | |
401 | rc = nbd_receive_options(client); | |
f5076b5a | 402 | if (rc != 0) { |
6b8c01e7 PB |
403 | LOG("option negotiation failed"); |
404 | goto fail; | |
405 | } | |
406 | ||
407 | assert ((client->exp->nbdflags & ~65535) == 0); | |
408 | cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size); | |
409 | cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags); | |
410 | if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) { | |
411 | LOG("write failed"); | |
412 | goto fail; | |
413 | } | |
b2e3d87f NT |
414 | } |
415 | ||
07f35073 | 416 | TRACE("Negotiation succeeded."); |
185b4338 PB |
417 | rc = 0; |
418 | fail: | |
f9e8cacc | 419 | qemu_set_nonblock(csock); |
185b4338 | 420 | return rc; |
7a5ca864 FB |
421 | } |
422 | ||
1d45f8b5 LV |
423 | int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags, |
424 | off_t *size, size_t *blocksize) | |
7a5ca864 | 425 | { |
b2e3d87f NT |
426 | char buf[256]; |
427 | uint64_t magic, s; | |
428 | uint16_t tmp; | |
185b4338 | 429 | int rc; |
b2e3d87f | 430 | |
07f35073 | 431 | TRACE("Receiving negotiation."); |
b2e3d87f | 432 | |
185b4338 PB |
433 | rc = -EINVAL; |
434 | ||
b2e3d87f NT |
435 | if (read_sync(csock, buf, 8) != 8) { |
436 | LOG("read failed"); | |
185b4338 | 437 | goto fail; |
b2e3d87f NT |
438 | } |
439 | ||
440 | buf[8] = '\0'; | |
441 | if (strlen(buf) == 0) { | |
442 | LOG("server connection closed"); | |
185b4338 | 443 | goto fail; |
b2e3d87f NT |
444 | } |
445 | ||
446 | TRACE("Magic is %c%c%c%c%c%c%c%c", | |
447 | qemu_isprint(buf[0]) ? buf[0] : '.', | |
448 | qemu_isprint(buf[1]) ? buf[1] : '.', | |
449 | qemu_isprint(buf[2]) ? buf[2] : '.', | |
450 | qemu_isprint(buf[3]) ? buf[3] : '.', | |
451 | qemu_isprint(buf[4]) ? buf[4] : '.', | |
452 | qemu_isprint(buf[5]) ? buf[5] : '.', | |
453 | qemu_isprint(buf[6]) ? buf[6] : '.', | |
454 | qemu_isprint(buf[7]) ? buf[7] : '.'); | |
455 | ||
456 | if (memcmp(buf, "NBDMAGIC", 8) != 0) { | |
457 | LOG("Invalid magic received"); | |
185b4338 | 458 | goto fail; |
b2e3d87f NT |
459 | } |
460 | ||
461 | if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
462 | LOG("read failed"); | |
185b4338 | 463 | goto fail; |
b2e3d87f NT |
464 | } |
465 | magic = be64_to_cpu(magic); | |
466 | TRACE("Magic is 0x%" PRIx64, magic); | |
467 | ||
468 | if (name) { | |
469 | uint32_t reserved = 0; | |
470 | uint32_t opt; | |
471 | uint32_t namesize; | |
472 | ||
473 | TRACE("Checking magic (opts_magic)"); | |
fa26c26b | 474 | if (magic != NBD_OPTS_MAGIC) { |
b2e3d87f | 475 | LOG("Bad magic received"); |
185b4338 | 476 | goto fail; |
b2e3d87f NT |
477 | } |
478 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
479 | LOG("flags read failed"); | |
185b4338 | 480 | goto fail; |
b2e3d87f NT |
481 | } |
482 | *flags = be16_to_cpu(tmp) << 16; | |
483 | /* reserved for future use */ | |
484 | if (write_sync(csock, &reserved, sizeof(reserved)) != | |
485 | sizeof(reserved)) { | |
486 | LOG("write failed (reserved)"); | |
185b4338 | 487 | goto fail; |
b2e3d87f NT |
488 | } |
489 | /* write the export name */ | |
490 | magic = cpu_to_be64(magic); | |
491 | if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
492 | LOG("write failed (magic)"); | |
185b4338 | 493 | goto fail; |
b2e3d87f NT |
494 | } |
495 | opt = cpu_to_be32(NBD_OPT_EXPORT_NAME); | |
496 | if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) { | |
497 | LOG("write failed (opt)"); | |
185b4338 | 498 | goto fail; |
b2e3d87f NT |
499 | } |
500 | namesize = cpu_to_be32(strlen(name)); | |
501 | if (write_sync(csock, &namesize, sizeof(namesize)) != | |
502 | sizeof(namesize)) { | |
503 | LOG("write failed (namesize)"); | |
185b4338 | 504 | goto fail; |
b2e3d87f NT |
505 | } |
506 | if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) { | |
507 | LOG("write failed (name)"); | |
185b4338 | 508 | goto fail; |
b2e3d87f NT |
509 | } |
510 | } else { | |
511 | TRACE("Checking magic (cli_magic)"); | |
512 | ||
fa26c26b | 513 | if (magic != NBD_CLIENT_MAGIC) { |
b2e3d87f | 514 | LOG("Bad magic received"); |
185b4338 | 515 | goto fail; |
b2e3d87f NT |
516 | } |
517 | } | |
518 | ||
519 | if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) { | |
520 | LOG("read failed"); | |
185b4338 | 521 | goto fail; |
b2e3d87f NT |
522 | } |
523 | *size = be64_to_cpu(s); | |
524 | *blocksize = 1024; | |
525 | TRACE("Size is %" PRIu64, *size); | |
526 | ||
527 | if (!name) { | |
528 | if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) { | |
529 | LOG("read failed (flags)"); | |
185b4338 | 530 | goto fail; |
b2e3d87f NT |
531 | } |
532 | *flags = be32_to_cpup(flags); | |
533 | } else { | |
534 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
535 | LOG("read failed (tmp)"); | |
185b4338 | 536 | goto fail; |
b2e3d87f NT |
537 | } |
538 | *flags |= be32_to_cpu(tmp); | |
539 | } | |
540 | if (read_sync(csock, &buf, 124) != 124) { | |
541 | LOG("read failed (buf)"); | |
185b4338 | 542 | goto fail; |
b2e3d87f | 543 | } |
185b4338 PB |
544 | rc = 0; |
545 | ||
546 | fail: | |
547 | return rc; | |
cd831bd7 | 548 | } |
7a5ca864 | 549 | |
b90fb4b8 PB |
550 | #ifdef __linux__ |
551 | int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize) | |
cd831bd7 | 552 | { |
3e05c785 CL |
553 | TRACE("Setting NBD socket"); |
554 | ||
fc19f8a0 | 555 | if (ioctl(fd, NBD_SET_SOCK, csock) < 0) { |
3e05c785 CL |
556 | int serrno = errno; |
557 | LOG("Failed to set NBD socket"); | |
185b4338 | 558 | return -serrno; |
3e05c785 CL |
559 | } |
560 | ||
b2e3d87f | 561 | TRACE("Setting block size to %lu", (unsigned long)blocksize); |
7a5ca864 | 562 | |
fc19f8a0 | 563 | if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) { |
b2e3d87f NT |
564 | int serrno = errno; |
565 | LOG("Failed setting NBD block size"); | |
185b4338 | 566 | return -serrno; |
b2e3d87f | 567 | } |
7a5ca864 | 568 | |
0bfcd599 | 569 | TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize)); |
7a5ca864 | 570 | |
fc19f8a0 | 571 | if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) { |
b2e3d87f NT |
572 | int serrno = errno; |
573 | LOG("Failed setting size (in blocks)"); | |
185b4338 | 574 | return -serrno; |
b2e3d87f | 575 | } |
7a5ca864 | 576 | |
c8969ede PB |
577 | if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) { |
578 | if (errno == ENOTTY) { | |
579 | int read_only = (flags & NBD_FLAG_READ_ONLY) != 0; | |
580 | TRACE("Setting readonly attribute"); | |
581 | ||
582 | if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) { | |
583 | int serrno = errno; | |
584 | LOG("Failed setting read-only attribute"); | |
585 | return -serrno; | |
586 | } | |
587 | } else { | |
b90fb4b8 | 588 | int serrno = errno; |
c8969ede | 589 | LOG("Failed setting flags"); |
185b4338 | 590 | return -serrno; |
b90fb4b8 PB |
591 | } |
592 | } | |
593 | ||
b2e3d87f | 594 | TRACE("Negotiation ended"); |
7a5ca864 | 595 | |
b2e3d87f | 596 | return 0; |
7a5ca864 FB |
597 | } |
598 | ||
599 | int nbd_disconnect(int fd) | |
600 | { | |
b2e3d87f NT |
601 | ioctl(fd, NBD_CLEAR_QUE); |
602 | ioctl(fd, NBD_DISCONNECT); | |
603 | ioctl(fd, NBD_CLEAR_SOCK); | |
604 | return 0; | |
7a5ca864 FB |
605 | } |
606 | ||
0a4eb864 | 607 | int nbd_client(int fd) |
7a5ca864 | 608 | { |
b2e3d87f NT |
609 | int ret; |
610 | int serrno; | |
7a5ca864 | 611 | |
b2e3d87f | 612 | TRACE("Doing NBD loop"); |
7a5ca864 | 613 | |
b2e3d87f | 614 | ret = ioctl(fd, NBD_DO_IT); |
fc19f8a0 | 615 | if (ret < 0 && errno == EPIPE) { |
74624688 PB |
616 | /* NBD_DO_IT normally returns EPIPE when someone has disconnected |
617 | * the socket via NBD_DISCONNECT. We do not want to return 1 in | |
618 | * that case. | |
619 | */ | |
620 | ret = 0; | |
621 | } | |
b2e3d87f | 622 | serrno = errno; |
7a5ca864 | 623 | |
b2e3d87f | 624 | TRACE("NBD loop returned %d: %s", ret, strerror(serrno)); |
7a5ca864 | 625 | |
b2e3d87f NT |
626 | TRACE("Clearing NBD queue"); |
627 | ioctl(fd, NBD_CLEAR_QUE); | |
7a5ca864 | 628 | |
b2e3d87f NT |
629 | TRACE("Clearing NBD socket"); |
630 | ioctl(fd, NBD_CLEAR_SOCK); | |
7a5ca864 | 631 | |
b2e3d87f NT |
632 | errno = serrno; |
633 | return ret; | |
7a5ca864 | 634 | } |
03ff3ca3 | 635 | #else |
8e72506e | 636 | int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize) |
03ff3ca3 | 637 | { |
185b4338 | 638 | return -ENOTSUP; |
03ff3ca3 AL |
639 | } |
640 | ||
641 | int nbd_disconnect(int fd) | |
642 | { | |
185b4338 | 643 | return -ENOTSUP; |
03ff3ca3 AL |
644 | } |
645 | ||
0a4eb864 | 646 | int nbd_client(int fd) |
03ff3ca3 | 647 | { |
185b4338 | 648 | return -ENOTSUP; |
03ff3ca3 AL |
649 | } |
650 | #endif | |
7a5ca864 | 651 | |
94e7340b | 652 | ssize_t nbd_send_request(int csock, struct nbd_request *request) |
7a5ca864 | 653 | { |
fa26c26b | 654 | uint8_t buf[NBD_REQUEST_SIZE]; |
185b4338 | 655 | ssize_t ret; |
b2e3d87f NT |
656 | |
657 | cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC); | |
658 | cpu_to_be32w((uint32_t*)(buf + 4), request->type); | |
659 | cpu_to_be64w((uint64_t*)(buf + 8), request->handle); | |
660 | cpu_to_be64w((uint64_t*)(buf + 16), request->from); | |
661 | cpu_to_be32w((uint32_t*)(buf + 24), request->len); | |
75818250 | 662 | |
b2e3d87f NT |
663 | TRACE("Sending request to client: " |
664 | "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}", | |
665 | request->from, request->len, request->handle, request->type); | |
666 | ||
185b4338 PB |
667 | ret = write_sync(csock, buf, sizeof(buf)); |
668 | if (ret < 0) { | |
669 | return ret; | |
670 | } | |
671 | ||
672 | if (ret != sizeof(buf)) { | |
b2e3d87f | 673 | LOG("writing to socket failed"); |
185b4338 | 674 | return -EINVAL; |
b2e3d87f NT |
675 | } |
676 | return 0; | |
677 | } | |
75818250 | 678 | |
94e7340b | 679 | static ssize_t nbd_receive_request(int csock, struct nbd_request *request) |
75818250 | 680 | { |
fa26c26b | 681 | uint8_t buf[NBD_REQUEST_SIZE]; |
b2e3d87f | 682 | uint32_t magic; |
185b4338 | 683 | ssize_t ret; |
b2e3d87f | 684 | |
185b4338 PB |
685 | ret = read_sync(csock, buf, sizeof(buf)); |
686 | if (ret < 0) { | |
687 | return ret; | |
688 | } | |
689 | ||
690 | if (ret != sizeof(buf)) { | |
b2e3d87f | 691 | LOG("read failed"); |
185b4338 | 692 | return -EINVAL; |
b2e3d87f NT |
693 | } |
694 | ||
695 | /* Request | |
696 | [ 0 .. 3] magic (NBD_REQUEST_MAGIC) | |
697 | [ 4 .. 7] type (0 == READ, 1 == WRITE) | |
698 | [ 8 .. 15] handle | |
699 | [16 .. 23] from | |
700 | [24 .. 27] len | |
701 | */ | |
702 | ||
703 | magic = be32_to_cpup((uint32_t*)buf); | |
704 | request->type = be32_to_cpup((uint32_t*)(buf + 4)); | |
705 | request->handle = be64_to_cpup((uint64_t*)(buf + 8)); | |
706 | request->from = be64_to_cpup((uint64_t*)(buf + 16)); | |
707 | request->len = be32_to_cpup((uint32_t*)(buf + 24)); | |
708 | ||
709 | TRACE("Got request: " | |
710 | "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }", | |
711 | magic, request->type, request->from, request->len); | |
712 | ||
713 | if (magic != NBD_REQUEST_MAGIC) { | |
714 | LOG("invalid magic (got 0x%x)", magic); | |
185b4338 | 715 | return -EINVAL; |
b2e3d87f NT |
716 | } |
717 | return 0; | |
75818250 TS |
718 | } |
719 | ||
94e7340b | 720 | ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply) |
75818250 | 721 | { |
b2e3d87f NT |
722 | uint8_t buf[NBD_REPLY_SIZE]; |
723 | uint32_t magic; | |
185b4338 | 724 | ssize_t ret; |
b2e3d87f | 725 | |
185b4338 PB |
726 | ret = read_sync(csock, buf, sizeof(buf)); |
727 | if (ret < 0) { | |
728 | return ret; | |
729 | } | |
730 | ||
731 | if (ret != sizeof(buf)) { | |
b2e3d87f | 732 | LOG("read failed"); |
185b4338 | 733 | return -EINVAL; |
b2e3d87f NT |
734 | } |
735 | ||
736 | /* Reply | |
737 | [ 0 .. 3] magic (NBD_REPLY_MAGIC) | |
738 | [ 4 .. 7] error (0 == no error) | |
739 | [ 7 .. 15] handle | |
740 | */ | |
741 | ||
742 | magic = be32_to_cpup((uint32_t*)buf); | |
743 | reply->error = be32_to_cpup((uint32_t*)(buf + 4)); | |
744 | reply->handle = be64_to_cpup((uint64_t*)(buf + 8)); | |
745 | ||
746 | TRACE("Got reply: " | |
747 | "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }", | |
748 | magic, reply->error, reply->handle); | |
749 | ||
750 | if (magic != NBD_REPLY_MAGIC) { | |
751 | LOG("invalid magic (got 0x%x)", magic); | |
185b4338 | 752 | return -EINVAL; |
b2e3d87f NT |
753 | } |
754 | return 0; | |
75818250 TS |
755 | } |
756 | ||
94e7340b | 757 | static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply) |
75818250 | 758 | { |
fa26c26b | 759 | uint8_t buf[NBD_REPLY_SIZE]; |
185b4338 | 760 | ssize_t ret; |
b2e3d87f NT |
761 | |
762 | /* Reply | |
763 | [ 0 .. 3] magic (NBD_REPLY_MAGIC) | |
764 | [ 4 .. 7] error (0 == no error) | |
765 | [ 7 .. 15] handle | |
766 | */ | |
767 | cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC); | |
768 | cpu_to_be32w((uint32_t*)(buf + 4), reply->error); | |
769 | cpu_to_be64w((uint64_t*)(buf + 8), reply->handle); | |
770 | ||
771 | TRACE("Sending response to client"); | |
772 | ||
185b4338 PB |
773 | ret = write_sync(csock, buf, sizeof(buf)); |
774 | if (ret < 0) { | |
775 | return ret; | |
776 | } | |
777 | ||
778 | if (ret != sizeof(buf)) { | |
b2e3d87f | 779 | LOG("writing to socket failed"); |
185b4338 | 780 | return -EINVAL; |
b2e3d87f NT |
781 | } |
782 | return 0; | |
75818250 | 783 | } |
7a5ca864 | 784 | |
41996e38 PB |
785 | #define MAX_NBD_REQUESTS 16 |
786 | ||
ce33967a | 787 | void nbd_client_get(NBDClient *client) |
1743b515 PB |
788 | { |
789 | client->refcount++; | |
790 | } | |
791 | ||
ce33967a | 792 | void nbd_client_put(NBDClient *client) |
1743b515 PB |
793 | { |
794 | if (--client->refcount == 0) { | |
ff2b68aa PB |
795 | /* The last reference should be dropped by client->close, |
796 | * which is called by nbd_client_close. | |
797 | */ | |
798 | assert(client->closing); | |
799 | ||
800 | qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL); | |
801 | close(client->sock); | |
802 | client->sock = -1; | |
6b8c01e7 PB |
803 | if (client->exp) { |
804 | QTAILQ_REMOVE(&client->exp->clients, client, next); | |
805 | nbd_export_put(client->exp); | |
806 | } | |
1743b515 PB |
807 | g_free(client); |
808 | } | |
809 | } | |
810 | ||
ff2b68aa | 811 | void nbd_client_close(NBDClient *client) |
1743b515 | 812 | { |
ff2b68aa PB |
813 | if (client->closing) { |
814 | return; | |
815 | } | |
816 | ||
817 | client->closing = true; | |
818 | ||
819 | /* Force requests to finish. They will drop their own references, | |
820 | * then we'll close the socket and free the NBDClient. | |
821 | */ | |
822 | shutdown(client->sock, 2); | |
823 | ||
824 | /* Also tell the client, so that they release their reference. */ | |
1743b515 PB |
825 | if (client->close) { |
826 | client->close(client); | |
827 | } | |
1743b515 PB |
828 | } |
829 | ||
72deddc5 | 830 | static NBDRequest *nbd_request_get(NBDClient *client) |
d9a73806 PB |
831 | { |
832 | NBDRequest *req; | |
72deddc5 | 833 | |
41996e38 PB |
834 | assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); |
835 | client->nb_requests++; | |
836 | ||
e1adb27a | 837 | req = g_slice_new0(NBDRequest); |
72deddc5 PB |
838 | nbd_client_get(client); |
839 | req->client = client; | |
d9a73806 PB |
840 | return req; |
841 | } | |
842 | ||
72deddc5 | 843 | static void nbd_request_put(NBDRequest *req) |
d9a73806 | 844 | { |
72deddc5 | 845 | NBDClient *client = req->client; |
e1adb27a | 846 | |
2d821488 SH |
847 | if (req->data) { |
848 | qemu_vfree(req->data); | |
849 | } | |
e1adb27a SH |
850 | g_slice_free(NBDRequest, req); |
851 | ||
41996e38 PB |
852 | if (client->nb_requests-- == MAX_NBD_REQUESTS) { |
853 | qemu_notify_event(); | |
854 | } | |
72deddc5 | 855 | nbd_client_put(client); |
d9a73806 PB |
856 | } |
857 | ||
af49bbbe | 858 | NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, |
0ddf08db PB |
859 | off_t size, uint32_t nbdflags, |
860 | void (*close)(NBDExport *)) | |
af49bbbe PB |
861 | { |
862 | NBDExport *exp = g_malloc0(sizeof(NBDExport)); | |
2c8d9f06 | 863 | exp->refcount = 1; |
4b9441f6 | 864 | QTAILQ_INIT(&exp->clients); |
af49bbbe PB |
865 | exp->bs = bs; |
866 | exp->dev_offset = dev_offset; | |
867 | exp->nbdflags = nbdflags; | |
38ceff04 | 868 | exp->size = size == -1 ? bdrv_getlength(bs) : size; |
0ddf08db | 869 | exp->close = close; |
38b54b6d | 870 | bdrv_ref(bs); |
af49bbbe PB |
871 | return exp; |
872 | } | |
873 | ||
ee0a19ec PB |
874 | NBDExport *nbd_export_find(const char *name) |
875 | { | |
876 | NBDExport *exp; | |
877 | QTAILQ_FOREACH(exp, &exports, next) { | |
878 | if (strcmp(name, exp->name) == 0) { | |
879 | return exp; | |
880 | } | |
881 | } | |
882 | ||
883 | return NULL; | |
884 | } | |
885 | ||
886 | void nbd_export_set_name(NBDExport *exp, const char *name) | |
887 | { | |
888 | if (exp->name == name) { | |
889 | return; | |
890 | } | |
891 | ||
892 | nbd_export_get(exp); | |
893 | if (exp->name != NULL) { | |
894 | g_free(exp->name); | |
895 | exp->name = NULL; | |
896 | QTAILQ_REMOVE(&exports, exp, next); | |
897 | nbd_export_put(exp); | |
898 | } | |
899 | if (name != NULL) { | |
900 | nbd_export_get(exp); | |
901 | exp->name = g_strdup(name); | |
902 | QTAILQ_INSERT_TAIL(&exports, exp, next); | |
903 | } | |
904 | nbd_export_put(exp); | |
905 | } | |
906 | ||
af49bbbe PB |
907 | void nbd_export_close(NBDExport *exp) |
908 | { | |
4b9441f6 | 909 | NBDClient *client, *next; |
2c8d9f06 | 910 | |
4b9441f6 PB |
911 | nbd_export_get(exp); |
912 | QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { | |
913 | nbd_client_close(client); | |
914 | } | |
125afda8 | 915 | nbd_export_set_name(exp, NULL); |
4b9441f6 | 916 | nbd_export_put(exp); |
38b54b6d FZ |
917 | if (exp->bs) { |
918 | bdrv_unref(exp->bs); | |
919 | exp->bs = NULL; | |
920 | } | |
2c8d9f06 PB |
921 | } |
922 | ||
923 | void nbd_export_get(NBDExport *exp) | |
924 | { | |
925 | assert(exp->refcount > 0); | |
926 | exp->refcount++; | |
927 | } | |
928 | ||
929 | void nbd_export_put(NBDExport *exp) | |
930 | { | |
931 | assert(exp->refcount > 0); | |
932 | if (exp->refcount == 1) { | |
933 | nbd_export_close(exp); | |
d9a73806 PB |
934 | } |
935 | ||
2c8d9f06 | 936 | if (--exp->refcount == 0) { |
ee0a19ec PB |
937 | assert(exp->name == NULL); |
938 | ||
0ddf08db PB |
939 | if (exp->close) { |
940 | exp->close(exp); | |
941 | } | |
942 | ||
2c8d9f06 PB |
943 | g_free(exp); |
944 | } | |
af49bbbe PB |
945 | } |
946 | ||
125afda8 PB |
947 | BlockDriverState *nbd_export_get_blockdev(NBDExport *exp) |
948 | { | |
949 | return exp->bs; | |
950 | } | |
951 | ||
ee0a19ec PB |
952 | void nbd_export_close_all(void) |
953 | { | |
954 | NBDExport *exp, *next; | |
955 | ||
956 | QTAILQ_FOREACH_SAFE(exp, &exports, next, next) { | |
957 | nbd_export_close(exp); | |
ee0a19ec PB |
958 | } |
959 | } | |
960 | ||
41996e38 | 961 | static int nbd_can_read(void *opaque); |
262db388 PB |
962 | static void nbd_read(void *opaque); |
963 | static void nbd_restart_write(void *opaque); | |
964 | ||
94e7340b PB |
965 | static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply, |
966 | int len) | |
22045592 | 967 | { |
72deddc5 PB |
968 | NBDClient *client = req->client; |
969 | int csock = client->sock; | |
94e7340b | 970 | ssize_t rc, ret; |
22045592 | 971 | |
262db388 | 972 | qemu_co_mutex_lock(&client->send_lock); |
41996e38 PB |
973 | qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, |
974 | nbd_restart_write, client); | |
262db388 PB |
975 | client->send_coroutine = qemu_coroutine_self(); |
976 | ||
22045592 PB |
977 | if (!len) { |
978 | rc = nbd_send_reply(csock, reply); | |
22045592 PB |
979 | } else { |
980 | socket_set_cork(csock, 1); | |
981 | rc = nbd_send_reply(csock, reply); | |
fc19f8a0 | 982 | if (rc >= 0) { |
262db388 | 983 | ret = qemu_co_send(csock, req->data, len); |
22045592 | 984 | if (ret != len) { |
185b4338 | 985 | rc = -EIO; |
22045592 PB |
986 | } |
987 | } | |
22045592 PB |
988 | socket_set_cork(csock, 0); |
989 | } | |
262db388 PB |
990 | |
991 | client->send_coroutine = NULL; | |
41996e38 | 992 | qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client); |
262db388 | 993 | qemu_co_mutex_unlock(&client->send_lock); |
22045592 PB |
994 | return rc; |
995 | } | |
996 | ||
94e7340b | 997 | static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request) |
a030b347 | 998 | { |
72deddc5 PB |
999 | NBDClient *client = req->client; |
1000 | int csock = client->sock; | |
2d821488 | 1001 | uint32_t command; |
94e7340b | 1002 | ssize_t rc; |
a030b347 | 1003 | |
262db388 | 1004 | client->recv_coroutine = qemu_coroutine_self(); |
7fe7b68b PB |
1005 | rc = nbd_receive_request(csock, request); |
1006 | if (rc < 0) { | |
1007 | if (rc != -EAGAIN) { | |
1008 | rc = -EIO; | |
1009 | } | |
a030b347 PB |
1010 | goto out; |
1011 | } | |
1012 | ||
2d821488 | 1013 | if (request->len > NBD_MAX_BUFFER_SIZE) { |
a030b347 | 1014 | LOG("len (%u) is larger than max len (%u)", |
2d821488 | 1015 | request->len, NBD_MAX_BUFFER_SIZE); |
a030b347 PB |
1016 | rc = -EINVAL; |
1017 | goto out; | |
1018 | } | |
1019 | ||
1020 | if ((request->from + request->len) < request->from) { | |
1021 | LOG("integer overflow detected! " | |
1022 | "you're probably being attacked"); | |
1023 | rc = -EINVAL; | |
1024 | goto out; | |
1025 | } | |
1026 | ||
1027 | TRACE("Decoding type"); | |
1028 | ||
2d821488 SH |
1029 | command = request->type & NBD_CMD_MASK_COMMAND; |
1030 | if (command == NBD_CMD_READ || command == NBD_CMD_WRITE) { | |
1031 | req->data = qemu_blockalign(client->exp->bs, request->len); | |
1032 | } | |
1033 | if (command == NBD_CMD_WRITE) { | |
a030b347 PB |
1034 | TRACE("Reading %u byte(s)", request->len); |
1035 | ||
262db388 | 1036 | if (qemu_co_recv(csock, req->data, request->len) != request->len) { |
a030b347 PB |
1037 | LOG("reading from socket failed"); |
1038 | rc = -EIO; | |
1039 | goto out; | |
1040 | } | |
1041 | } | |
1042 | rc = 0; | |
1043 | ||
1044 | out: | |
262db388 | 1045 | client->recv_coroutine = NULL; |
a030b347 PB |
1046 | return rc; |
1047 | } | |
1048 | ||
262db388 | 1049 | static void nbd_trip(void *opaque) |
75818250 | 1050 | { |
262db388 | 1051 | NBDClient *client = opaque; |
1743b515 | 1052 | NBDExport *exp = client->exp; |
ff2b68aa | 1053 | NBDRequest *req; |
b2e3d87f NT |
1054 | struct nbd_request request; |
1055 | struct nbd_reply reply; | |
94e7340b | 1056 | ssize_t ret; |
8c5d1abb | 1057 | uint32_t command; |
b2e3d87f NT |
1058 | |
1059 | TRACE("Reading request."); | |
ff2b68aa PB |
1060 | if (client->closing) { |
1061 | return; | |
1062 | } | |
b2e3d87f | 1063 | |
ff2b68aa | 1064 | req = nbd_request_get(client); |
262db388 | 1065 | ret = nbd_co_receive_request(req, &request); |
7fe7b68b PB |
1066 | if (ret == -EAGAIN) { |
1067 | goto done; | |
1068 | } | |
a030b347 | 1069 | if (ret == -EIO) { |
d9a73806 | 1070 | goto out; |
a030b347 | 1071 | } |
b2e3d87f | 1072 | |
fae69416 PB |
1073 | reply.handle = request.handle; |
1074 | reply.error = 0; | |
1075 | ||
a030b347 PB |
1076 | if (ret < 0) { |
1077 | reply.error = -ret; | |
1078 | goto error_reply; | |
b2e3d87f | 1079 | } |
8c5d1abb HB |
1080 | command = request.type & NBD_CMD_MASK_COMMAND; |
1081 | if (command != NBD_CMD_DISC && (request.from + request.len) > exp->size) { | |
b2e3d87f NT |
1082 | LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64 |
1083 | ", Offset: %" PRIu64 "\n", | |
af49bbbe | 1084 | request.from, request.len, |
0fee8f34 | 1085 | (uint64_t)exp->size, (uint64_t)exp->dev_offset); |
b2e3d87f | 1086 | LOG("requested operation past EOF--bad client?"); |
fae69416 | 1087 | goto invalid_request; |
b2e3d87f NT |
1088 | } |
1089 | ||
8c5d1abb | 1090 | switch (command) { |
b2e3d87f NT |
1091 | case NBD_CMD_READ: |
1092 | TRACE("Request type is READ"); | |
1093 | ||
e25ceb76 PB |
1094 | if (request.type & NBD_CMD_FLAG_FUA) { |
1095 | ret = bdrv_co_flush(exp->bs); | |
1096 | if (ret < 0) { | |
1097 | LOG("flush failed"); | |
1098 | reply.error = -ret; | |
1099 | goto error_reply; | |
1100 | } | |
1101 | } | |
1102 | ||
af49bbbe | 1103 | ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512, |
d9a73806 | 1104 | req->data, request.len / 512); |
adcf6302 | 1105 | if (ret < 0) { |
b2e3d87f | 1106 | LOG("reading from file failed"); |
adcf6302 | 1107 | reply.error = -ret; |
fae69416 | 1108 | goto error_reply; |
b2e3d87f | 1109 | } |
b2e3d87f NT |
1110 | |
1111 | TRACE("Read %u byte(s)", request.len); | |
262db388 | 1112 | if (nbd_co_send_reply(req, &reply, request.len) < 0) |
d9a73806 | 1113 | goto out; |
b2e3d87f NT |
1114 | break; |
1115 | case NBD_CMD_WRITE: | |
1116 | TRACE("Request type is WRITE"); | |
1117 | ||
af49bbbe | 1118 | if (exp->nbdflags & NBD_FLAG_READ_ONLY) { |
b2e3d87f | 1119 | TRACE("Server is read-only, return error"); |
fae69416 PB |
1120 | reply.error = EROFS; |
1121 | goto error_reply; | |
1122 | } | |
1123 | ||
1124 | TRACE("Writing to device"); | |
1125 | ||
af49bbbe | 1126 | ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512, |
d9a73806 | 1127 | req->data, request.len / 512); |
fae69416 PB |
1128 | if (ret < 0) { |
1129 | LOG("writing to file failed"); | |
1130 | reply.error = -ret; | |
1131 | goto error_reply; | |
1132 | } | |
b2e3d87f | 1133 | |
fae69416 | 1134 | if (request.type & NBD_CMD_FLAG_FUA) { |
262db388 | 1135 | ret = bdrv_co_flush(exp->bs); |
adcf6302 | 1136 | if (ret < 0) { |
fae69416 | 1137 | LOG("flush failed"); |
adcf6302 | 1138 | reply.error = -ret; |
fae69416 | 1139 | goto error_reply; |
2c7989a9 | 1140 | } |
b2e3d87f NT |
1141 | } |
1142 | ||
fc19f8a0 | 1143 | if (nbd_co_send_reply(req, &reply, 0) < 0) { |
d9a73806 | 1144 | goto out; |
fc19f8a0 | 1145 | } |
b2e3d87f NT |
1146 | break; |
1147 | case NBD_CMD_DISC: | |
1148 | TRACE("Request type is DISCONNECT"); | |
1149 | errno = 0; | |
262db388 | 1150 | goto out; |
1486d04a PB |
1151 | case NBD_CMD_FLUSH: |
1152 | TRACE("Request type is FLUSH"); | |
1153 | ||
262db388 | 1154 | ret = bdrv_co_flush(exp->bs); |
1486d04a PB |
1155 | if (ret < 0) { |
1156 | LOG("flush failed"); | |
1157 | reply.error = -ret; | |
1158 | } | |
fc19f8a0 | 1159 | if (nbd_co_send_reply(req, &reply, 0) < 0) { |
d9a73806 | 1160 | goto out; |
fc19f8a0 | 1161 | } |
7a706633 PB |
1162 | break; |
1163 | case NBD_CMD_TRIM: | |
1164 | TRACE("Request type is TRIM"); | |
262db388 PB |
1165 | ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512, |
1166 | request.len / 512); | |
7a706633 PB |
1167 | if (ret < 0) { |
1168 | LOG("discard failed"); | |
1169 | reply.error = -ret; | |
1170 | } | |
fc19f8a0 | 1171 | if (nbd_co_send_reply(req, &reply, 0) < 0) { |
d9a73806 | 1172 | goto out; |
fc19f8a0 | 1173 | } |
1486d04a | 1174 | break; |
b2e3d87f NT |
1175 | default: |
1176 | LOG("invalid request type (%u) received", request.type); | |
fae69416 PB |
1177 | invalid_request: |
1178 | reply.error = -EINVAL; | |
1179 | error_reply: | |
fc19f8a0 | 1180 | if (nbd_co_send_reply(req, &reply, 0) < 0) { |
d9a73806 | 1181 | goto out; |
fc19f8a0 | 1182 | } |
fae69416 | 1183 | break; |
b2e3d87f NT |
1184 | } |
1185 | ||
1186 | TRACE("Request/Reply complete"); | |
1187 | ||
7fe7b68b | 1188 | done: |
262db388 PB |
1189 | nbd_request_put(req); |
1190 | return; | |
1191 | ||
d9a73806 | 1192 | out: |
72deddc5 | 1193 | nbd_request_put(req); |
262db388 | 1194 | nbd_client_close(client); |
7a5ca864 | 1195 | } |
af49bbbe | 1196 | |
41996e38 PB |
1197 | static int nbd_can_read(void *opaque) |
1198 | { | |
1199 | NBDClient *client = opaque; | |
1200 | ||
1201 | return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS; | |
1202 | } | |
1203 | ||
1743b515 PB |
1204 | static void nbd_read(void *opaque) |
1205 | { | |
1206 | NBDClient *client = opaque; | |
1207 | ||
262db388 PB |
1208 | if (client->recv_coroutine) { |
1209 | qemu_coroutine_enter(client->recv_coroutine, NULL); | |
1210 | } else { | |
1211 | qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client); | |
1743b515 | 1212 | } |
1743b515 PB |
1213 | } |
1214 | ||
262db388 PB |
1215 | static void nbd_restart_write(void *opaque) |
1216 | { | |
1217 | NBDClient *client = opaque; | |
1218 | ||
1219 | qemu_coroutine_enter(client->send_coroutine, NULL); | |
1220 | } | |
1221 | ||
1743b515 PB |
1222 | NBDClient *nbd_client_new(NBDExport *exp, int csock, |
1223 | void (*close)(NBDClient *)) | |
af49bbbe | 1224 | { |
1743b515 | 1225 | NBDClient *client; |
1743b515 PB |
1226 | client = g_malloc0(sizeof(NBDClient)); |
1227 | client->refcount = 1; | |
1228 | client->exp = exp; | |
1229 | client->sock = csock; | |
f5076b5a | 1230 | if (nbd_send_negotiate(client)) { |
9a304d29 PB |
1231 | g_free(client); |
1232 | return NULL; | |
1233 | } | |
1743b515 | 1234 | client->close = close; |
262db388 | 1235 | qemu_co_mutex_init(&client->send_lock); |
41996e38 | 1236 | qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client); |
2c8d9f06 | 1237 | |
6b8c01e7 PB |
1238 | if (exp) { |
1239 | QTAILQ_INSERT_TAIL(&exp->clients, client, next); | |
1240 | nbd_export_get(exp); | |
1241 | } | |
1743b515 | 1242 | return client; |
af49bbbe | 1243 | } |