]>
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 FB |
18 | |
19 | #include "nbd.h" | |
ab359cd1 | 20 | #include "block.h" |
af49bbbe | 21 | #include "block_int.h" |
7a5ca864 FB |
22 | |
23 | #include <errno.h> | |
24 | #include <string.h> | |
03ff3ca3 | 25 | #ifndef _WIN32 |
7a5ca864 | 26 | #include <sys/ioctl.h> |
03ff3ca3 | 27 | #endif |
5dc2eec9 | 28 | #if defined(__sun__) || defined(__HAIKU__) |
7e00eb9b AL |
29 | #include <sys/ioccom.h> |
30 | #endif | |
7a5ca864 FB |
31 | #include <ctype.h> |
32 | #include <inttypes.h> | |
75818250 | 33 | |
b90fb4b8 PB |
34 | #ifdef __linux__ |
35 | #include <linux/fs.h> | |
36 | #endif | |
37 | ||
03ff3ca3 AL |
38 | #include "qemu_socket.h" |
39 | ||
40 | //#define DEBUG_NBD | |
41 | ||
42 | #ifdef DEBUG_NBD | |
75818250 | 43 | #define TRACE(msg, ...) do { \ |
03ff3ca3 | 44 | LOG(msg, ## __VA_ARGS__); \ |
75818250 | 45 | } while(0) |
03ff3ca3 AL |
46 | #else |
47 | #define TRACE(msg, ...) \ | |
48 | do { } while (0) | |
49 | #endif | |
7a5ca864 FB |
50 | |
51 | #define LOG(msg, ...) do { \ | |
52 | fprintf(stderr, "%s:%s():L%d: " msg "\n", \ | |
53 | __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \ | |
54 | } while(0) | |
55 | ||
7a5ca864 FB |
56 | /* This is all part of the "official" NBD API */ |
57 | ||
b2e3d87f | 58 | #define NBD_REPLY_SIZE (4 + 4 + 8) |
7a5ca864 FB |
59 | #define NBD_REQUEST_MAGIC 0x25609513 |
60 | #define NBD_REPLY_MAGIC 0x67446698 | |
61 | ||
62 | #define NBD_SET_SOCK _IO(0xab, 0) | |
63 | #define NBD_SET_BLKSIZE _IO(0xab, 1) | |
64 | #define NBD_SET_SIZE _IO(0xab, 2) | |
65 | #define NBD_DO_IT _IO(0xab, 3) | |
66 | #define NBD_CLEAR_SOCK _IO(0xab, 4) | |
67 | #define NBD_CLEAR_QUE _IO(0xab, 5) | |
b2e3d87f NT |
68 | #define NBD_PRINT_DEBUG _IO(0xab, 6) |
69 | #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7) | |
7a5ca864 | 70 | #define NBD_DISCONNECT _IO(0xab, 8) |
bbb74edd PB |
71 | #define NBD_SET_TIMEOUT _IO(0xab, 9) |
72 | #define NBD_SET_FLAGS _IO(0xab, 10) | |
7a5ca864 | 73 | |
b2e3d87f | 74 | #define NBD_OPT_EXPORT_NAME (1 << 0) |
1d45f8b5 | 75 | |
7a5ca864 FB |
76 | /* That's all folks */ |
77 | ||
75818250 TS |
78 | #define read_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, true) |
79 | #define write_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, false) | |
7a5ca864 | 80 | |
75818250 | 81 | size_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read) |
7a5ca864 FB |
82 | { |
83 | size_t offset = 0; | |
84 | ||
ae255e52 PB |
85 | if (qemu_in_coroutine()) { |
86 | if (do_read) { | |
87 | return qemu_co_recv(fd, buffer, size); | |
88 | } else { | |
89 | return qemu_co_send(fd, buffer, size); | |
90 | } | |
91 | } | |
92 | ||
7a5ca864 FB |
93 | while (offset < size) { |
94 | ssize_t len; | |
95 | ||
96 | if (do_read) { | |
00aa0040 | 97 | len = qemu_recv(fd, buffer + offset, size - offset, 0); |
7a5ca864 | 98 | } else { |
03ff3ca3 | 99 | len = send(fd, buffer + offset, size - offset, 0); |
7a5ca864 FB |
100 | } |
101 | ||
03ff3ca3 AL |
102 | if (len == -1) |
103 | errno = socket_error(); | |
104 | ||
7a5ca864 | 105 | /* recoverable error */ |
75818250 | 106 | if (len == -1 && (errno == EAGAIN || errno == EINTR)) { |
7a5ca864 FB |
107 | continue; |
108 | } | |
109 | ||
110 | /* eof */ | |
111 | if (len == 0) { | |
112 | break; | |
113 | } | |
114 | ||
115 | /* unrecoverable error */ | |
116 | if (len == -1) { | |
117 | return 0; | |
118 | } | |
119 | ||
120 | offset += len; | |
121 | } | |
122 | ||
123 | return offset; | |
124 | } | |
125 | ||
c12504ce NT |
126 | static void combine_addr(char *buf, size_t len, const char* address, |
127 | uint16_t port) | |
7a5ca864 | 128 | { |
c12504ce NT |
129 | /* If the address-part contains a colon, it's an IPv6 IP so needs [] */ |
130 | if (strstr(address, ":")) { | |
131 | snprintf(buf, len, "[%s]:%u", address, port); | |
132 | } else { | |
133 | snprintf(buf, len, "%s:%u", address, port); | |
7a5ca864 | 134 | } |
7a5ca864 FB |
135 | } |
136 | ||
c12504ce | 137 | int tcp_socket_outgoing(const char *address, uint16_t port) |
7a5ca864 | 138 | { |
c12504ce NT |
139 | char address_and_port[128]; |
140 | combine_addr(address_and_port, 128, address, port); | |
141 | return tcp_socket_outgoing_spec(address_and_port); | |
7a5ca864 FB |
142 | } |
143 | ||
c12504ce | 144 | int tcp_socket_outgoing_spec(const char *address_and_port) |
cd831bd7 | 145 | { |
c12504ce | 146 | return inet_connect(address_and_port, SOCK_STREAM); |
cd831bd7 TS |
147 | } |
148 | ||
c12504ce | 149 | int tcp_socket_incoming(const char *address, uint16_t port) |
cd831bd7 | 150 | { |
c12504ce NT |
151 | char address_and_port[128]; |
152 | combine_addr(address_and_port, 128, address, port); | |
153 | return tcp_socket_incoming_spec(address_and_port); | |
154 | } | |
cd831bd7 | 155 | |
c12504ce NT |
156 | int tcp_socket_incoming_spec(const char *address_and_port) |
157 | { | |
158 | char *ostr = NULL; | |
159 | int olen = 0; | |
160 | return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0); | |
03ff3ca3 | 161 | } |
c12504ce | 162 | |
03ff3ca3 AL |
163 | int unix_socket_incoming(const char *path) |
164 | { | |
c12504ce NT |
165 | char *ostr = NULL; |
166 | int olen = 0; | |
167 | ||
168 | return unix_listen(path, ostr, olen); | |
cd831bd7 TS |
169 | } |
170 | ||
03ff3ca3 AL |
171 | int unix_socket_outgoing(const char *path) |
172 | { | |
c12504ce | 173 | return unix_connect(path); |
03ff3ca3 | 174 | } |
cd831bd7 | 175 | |
7a5ca864 FB |
176 | /* Basic flow |
177 | ||
178 | Server Client | |
179 | ||
180 | Negotiate | |
181 | Request | |
182 | Response | |
183 | Request | |
184 | Response | |
185 | ... | |
186 | ... | |
187 | Request (type == 2) | |
188 | */ | |
189 | ||
af49bbbe | 190 | static int nbd_send_negotiate(int csock, off_t size, uint32_t flags) |
7a5ca864 | 191 | { |
b2e3d87f NT |
192 | char buf[8 + 8 + 8 + 128]; |
193 | ||
194 | /* Negotiate | |
195 | [ 0 .. 7] passwd ("NBDMAGIC") | |
196 | [ 8 .. 15] magic (0x00420281861253) | |
197 | [16 .. 23] size | |
b90fb4b8 PB |
198 | [24 .. 27] flags |
199 | [28 .. 151] reserved (0) | |
b2e3d87f NT |
200 | */ |
201 | ||
202 | TRACE("Beginning negotiation."); | |
203 | memcpy(buf, "NBDMAGIC", 8); | |
204 | cpu_to_be64w((uint64_t*)(buf + 8), 0x00420281861253LL); | |
205 | cpu_to_be64w((uint64_t*)(buf + 16), size); | |
2c7989a9 | 206 | cpu_to_be32w((uint32_t*)(buf + 24), |
7a706633 PB |
207 | flags | NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM | |
208 | NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA); | |
b90fb4b8 | 209 | memset(buf + 28, 0, 124); |
b2e3d87f NT |
210 | |
211 | if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
212 | LOG("write failed"); | |
213 | errno = EINVAL; | |
214 | return -1; | |
215 | } | |
216 | ||
07f35073 | 217 | TRACE("Negotiation succeeded."); |
b2e3d87f NT |
218 | |
219 | return 0; | |
7a5ca864 FB |
220 | } |
221 | ||
1d45f8b5 LV |
222 | int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags, |
223 | off_t *size, size_t *blocksize) | |
7a5ca864 | 224 | { |
b2e3d87f NT |
225 | char buf[256]; |
226 | uint64_t magic, s; | |
227 | uint16_t tmp; | |
228 | ||
07f35073 | 229 | TRACE("Receiving negotiation."); |
b2e3d87f NT |
230 | |
231 | if (read_sync(csock, buf, 8) != 8) { | |
232 | LOG("read failed"); | |
233 | errno = EINVAL; | |
234 | return -1; | |
235 | } | |
236 | ||
237 | buf[8] = '\0'; | |
238 | if (strlen(buf) == 0) { | |
239 | LOG("server connection closed"); | |
240 | errno = EINVAL; | |
241 | return -1; | |
242 | } | |
243 | ||
244 | TRACE("Magic is %c%c%c%c%c%c%c%c", | |
245 | qemu_isprint(buf[0]) ? buf[0] : '.', | |
246 | qemu_isprint(buf[1]) ? buf[1] : '.', | |
247 | qemu_isprint(buf[2]) ? buf[2] : '.', | |
248 | qemu_isprint(buf[3]) ? buf[3] : '.', | |
249 | qemu_isprint(buf[4]) ? buf[4] : '.', | |
250 | qemu_isprint(buf[5]) ? buf[5] : '.', | |
251 | qemu_isprint(buf[6]) ? buf[6] : '.', | |
252 | qemu_isprint(buf[7]) ? buf[7] : '.'); | |
253 | ||
254 | if (memcmp(buf, "NBDMAGIC", 8) != 0) { | |
255 | LOG("Invalid magic received"); | |
256 | errno = EINVAL; | |
257 | return -1; | |
258 | } | |
259 | ||
260 | if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
261 | LOG("read failed"); | |
262 | errno = EINVAL; | |
263 | return -1; | |
264 | } | |
265 | magic = be64_to_cpu(magic); | |
266 | TRACE("Magic is 0x%" PRIx64, magic); | |
267 | ||
268 | if (name) { | |
269 | uint32_t reserved = 0; | |
270 | uint32_t opt; | |
271 | uint32_t namesize; | |
272 | ||
273 | TRACE("Checking magic (opts_magic)"); | |
274 | if (magic != 0x49484156454F5054LL) { | |
275 | LOG("Bad magic received"); | |
276 | errno = EINVAL; | |
277 | return -1; | |
278 | } | |
279 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
280 | LOG("flags read failed"); | |
281 | errno = EINVAL; | |
282 | return -1; | |
283 | } | |
284 | *flags = be16_to_cpu(tmp) << 16; | |
285 | /* reserved for future use */ | |
286 | if (write_sync(csock, &reserved, sizeof(reserved)) != | |
287 | sizeof(reserved)) { | |
288 | LOG("write failed (reserved)"); | |
289 | errno = EINVAL; | |
290 | return -1; | |
291 | } | |
292 | /* write the export name */ | |
293 | magic = cpu_to_be64(magic); | |
294 | if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) { | |
295 | LOG("write failed (magic)"); | |
296 | errno = EINVAL; | |
297 | return -1; | |
298 | } | |
299 | opt = cpu_to_be32(NBD_OPT_EXPORT_NAME); | |
300 | if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) { | |
301 | LOG("write failed (opt)"); | |
302 | errno = EINVAL; | |
303 | return -1; | |
304 | } | |
305 | namesize = cpu_to_be32(strlen(name)); | |
306 | if (write_sync(csock, &namesize, sizeof(namesize)) != | |
307 | sizeof(namesize)) { | |
308 | LOG("write failed (namesize)"); | |
309 | errno = EINVAL; | |
310 | return -1; | |
311 | } | |
312 | if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) { | |
313 | LOG("write failed (name)"); | |
314 | errno = EINVAL; | |
315 | return -1; | |
316 | } | |
317 | } else { | |
318 | TRACE("Checking magic (cli_magic)"); | |
319 | ||
320 | if (magic != 0x00420281861253LL) { | |
321 | LOG("Bad magic received"); | |
322 | errno = EINVAL; | |
323 | return -1; | |
324 | } | |
325 | } | |
326 | ||
327 | if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) { | |
328 | LOG("read failed"); | |
329 | errno = EINVAL; | |
330 | return -1; | |
331 | } | |
332 | *size = be64_to_cpu(s); | |
333 | *blocksize = 1024; | |
334 | TRACE("Size is %" PRIu64, *size); | |
335 | ||
336 | if (!name) { | |
337 | if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) { | |
338 | LOG("read failed (flags)"); | |
339 | errno = EINVAL; | |
340 | return -1; | |
341 | } | |
342 | *flags = be32_to_cpup(flags); | |
343 | } else { | |
344 | if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) { | |
345 | LOG("read failed (tmp)"); | |
346 | errno = EINVAL; | |
347 | return -1; | |
348 | } | |
349 | *flags |= be32_to_cpu(tmp); | |
350 | } | |
351 | if (read_sync(csock, &buf, 124) != 124) { | |
352 | LOG("read failed (buf)"); | |
353 | errno = EINVAL; | |
354 | return -1; | |
355 | } | |
cd831bd7 TS |
356 | return 0; |
357 | } | |
7a5ca864 | 358 | |
b90fb4b8 PB |
359 | #ifdef __linux__ |
360 | int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize) | |
cd831bd7 | 361 | { |
3e05c785 CL |
362 | TRACE("Setting NBD socket"); |
363 | ||
364 | if (ioctl(fd, NBD_SET_SOCK, csock) == -1) { | |
365 | int serrno = errno; | |
366 | LOG("Failed to set NBD socket"); | |
367 | errno = serrno; | |
368 | return -1; | |
369 | } | |
370 | ||
b2e3d87f | 371 | TRACE("Setting block size to %lu", (unsigned long)blocksize); |
7a5ca864 | 372 | |
b2e3d87f NT |
373 | if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) == -1) { |
374 | int serrno = errno; | |
375 | LOG("Failed setting NBD block size"); | |
376 | errno = serrno; | |
377 | return -1; | |
378 | } | |
7a5ca864 | 379 | |
0bfcd599 | 380 | TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize)); |
7a5ca864 | 381 | |
b2e3d87f NT |
382 | if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) == -1) { |
383 | int serrno = errno; | |
384 | LOG("Failed setting size (in blocks)"); | |
385 | errno = serrno; | |
386 | return -1; | |
387 | } | |
7a5ca864 | 388 | |
b90fb4b8 PB |
389 | if (flags & NBD_FLAG_READ_ONLY) { |
390 | int read_only = 1; | |
391 | TRACE("Setting readonly attribute"); | |
392 | ||
393 | if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) { | |
394 | int serrno = errno; | |
395 | LOG("Failed setting read-only attribute"); | |
396 | errno = serrno; | |
397 | return -1; | |
398 | } | |
399 | } | |
400 | ||
973b3d0a PB |
401 | if (ioctl(fd, NBD_SET_FLAGS, flags) < 0 |
402 | && errno != ENOTTY) { | |
403 | int serrno = errno; | |
404 | LOG("Failed setting flags"); | |
405 | errno = serrno; | |
406 | return -1; | |
407 | } | |
408 | ||
b2e3d87f | 409 | TRACE("Negotiation ended"); |
7a5ca864 | 410 | |
b2e3d87f | 411 | return 0; |
7a5ca864 FB |
412 | } |
413 | ||
414 | int nbd_disconnect(int fd) | |
415 | { | |
b2e3d87f NT |
416 | ioctl(fd, NBD_CLEAR_QUE); |
417 | ioctl(fd, NBD_DISCONNECT); | |
418 | ioctl(fd, NBD_CLEAR_SOCK); | |
419 | return 0; | |
7a5ca864 FB |
420 | } |
421 | ||
0a4eb864 | 422 | int nbd_client(int fd) |
7a5ca864 | 423 | { |
b2e3d87f NT |
424 | int ret; |
425 | int serrno; | |
7a5ca864 | 426 | |
b2e3d87f | 427 | TRACE("Doing NBD loop"); |
7a5ca864 | 428 | |
b2e3d87f | 429 | ret = ioctl(fd, NBD_DO_IT); |
74624688 PB |
430 | if (ret == -1 && errno == EPIPE) { |
431 | /* NBD_DO_IT normally returns EPIPE when someone has disconnected | |
432 | * the socket via NBD_DISCONNECT. We do not want to return 1 in | |
433 | * that case. | |
434 | */ | |
435 | ret = 0; | |
436 | } | |
b2e3d87f | 437 | serrno = errno; |
7a5ca864 | 438 | |
b2e3d87f | 439 | TRACE("NBD loop returned %d: %s", ret, strerror(serrno)); |
7a5ca864 | 440 | |
b2e3d87f NT |
441 | TRACE("Clearing NBD queue"); |
442 | ioctl(fd, NBD_CLEAR_QUE); | |
7a5ca864 | 443 | |
b2e3d87f NT |
444 | TRACE("Clearing NBD socket"); |
445 | ioctl(fd, NBD_CLEAR_SOCK); | |
7a5ca864 | 446 | |
b2e3d87f NT |
447 | errno = serrno; |
448 | return ret; | |
7a5ca864 | 449 | } |
03ff3ca3 | 450 | #else |
8e72506e | 451 | int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize) |
03ff3ca3 AL |
452 | { |
453 | errno = ENOTSUP; | |
454 | return -1; | |
455 | } | |
456 | ||
457 | int nbd_disconnect(int fd) | |
458 | { | |
459 | errno = ENOTSUP; | |
460 | return -1; | |
461 | } | |
462 | ||
0a4eb864 | 463 | int nbd_client(int fd) |
03ff3ca3 AL |
464 | { |
465 | errno = ENOTSUP; | |
466 | return -1; | |
467 | } | |
468 | #endif | |
7a5ca864 | 469 | |
75818250 | 470 | int nbd_send_request(int csock, struct nbd_request *request) |
7a5ca864 | 471 | { |
b2e3d87f NT |
472 | uint8_t buf[4 + 4 + 8 + 8 + 4]; |
473 | ||
474 | cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC); | |
475 | cpu_to_be32w((uint32_t*)(buf + 4), request->type); | |
476 | cpu_to_be64w((uint64_t*)(buf + 8), request->handle); | |
477 | cpu_to_be64w((uint64_t*)(buf + 16), request->from); | |
478 | cpu_to_be32w((uint32_t*)(buf + 24), request->len); | |
75818250 | 479 | |
b2e3d87f NT |
480 | TRACE("Sending request to client: " |
481 | "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}", | |
482 | request->from, request->len, request->handle, request->type); | |
483 | ||
484 | if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
485 | LOG("writing to socket failed"); | |
486 | errno = EINVAL; | |
487 | return -1; | |
488 | } | |
489 | return 0; | |
490 | } | |
75818250 TS |
491 | |
492 | static int nbd_receive_request(int csock, struct nbd_request *request) | |
493 | { | |
b2e3d87f NT |
494 | uint8_t buf[4 + 4 + 8 + 8 + 4]; |
495 | uint32_t magic; | |
496 | ||
497 | if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
498 | LOG("read failed"); | |
499 | errno = EINVAL; | |
500 | return -1; | |
501 | } | |
502 | ||
503 | /* Request | |
504 | [ 0 .. 3] magic (NBD_REQUEST_MAGIC) | |
505 | [ 4 .. 7] type (0 == READ, 1 == WRITE) | |
506 | [ 8 .. 15] handle | |
507 | [16 .. 23] from | |
508 | [24 .. 27] len | |
509 | */ | |
510 | ||
511 | magic = be32_to_cpup((uint32_t*)buf); | |
512 | request->type = be32_to_cpup((uint32_t*)(buf + 4)); | |
513 | request->handle = be64_to_cpup((uint64_t*)(buf + 8)); | |
514 | request->from = be64_to_cpup((uint64_t*)(buf + 16)); | |
515 | request->len = be32_to_cpup((uint32_t*)(buf + 24)); | |
516 | ||
517 | TRACE("Got request: " | |
518 | "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }", | |
519 | magic, request->type, request->from, request->len); | |
520 | ||
521 | if (magic != NBD_REQUEST_MAGIC) { | |
522 | LOG("invalid magic (got 0x%x)", magic); | |
523 | errno = EINVAL; | |
524 | return -1; | |
525 | } | |
526 | return 0; | |
75818250 TS |
527 | } |
528 | ||
529 | int nbd_receive_reply(int csock, struct nbd_reply *reply) | |
530 | { | |
b2e3d87f NT |
531 | uint8_t buf[NBD_REPLY_SIZE]; |
532 | uint32_t magic; | |
533 | ||
534 | memset(buf, 0xAA, sizeof(buf)); | |
535 | ||
536 | if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
537 | LOG("read failed"); | |
538 | errno = EINVAL; | |
539 | return -1; | |
540 | } | |
541 | ||
542 | /* Reply | |
543 | [ 0 .. 3] magic (NBD_REPLY_MAGIC) | |
544 | [ 4 .. 7] error (0 == no error) | |
545 | [ 7 .. 15] handle | |
546 | */ | |
547 | ||
548 | magic = be32_to_cpup((uint32_t*)buf); | |
549 | reply->error = be32_to_cpup((uint32_t*)(buf + 4)); | |
550 | reply->handle = be64_to_cpup((uint64_t*)(buf + 8)); | |
551 | ||
552 | TRACE("Got reply: " | |
553 | "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }", | |
554 | magic, reply->error, reply->handle); | |
555 | ||
556 | if (magic != NBD_REPLY_MAGIC) { | |
557 | LOG("invalid magic (got 0x%x)", magic); | |
558 | errno = EINVAL; | |
559 | return -1; | |
560 | } | |
561 | return 0; | |
75818250 TS |
562 | } |
563 | ||
564 | static int nbd_send_reply(int csock, struct nbd_reply *reply) | |
565 | { | |
b2e3d87f NT |
566 | uint8_t buf[4 + 4 + 8]; |
567 | ||
568 | /* Reply | |
569 | [ 0 .. 3] magic (NBD_REPLY_MAGIC) | |
570 | [ 4 .. 7] error (0 == no error) | |
571 | [ 7 .. 15] handle | |
572 | */ | |
573 | cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC); | |
574 | cpu_to_be32w((uint32_t*)(buf + 4), reply->error); | |
575 | cpu_to_be64w((uint64_t*)(buf + 8), reply->handle); | |
576 | ||
577 | TRACE("Sending response to client"); | |
578 | ||
579 | if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) { | |
580 | LOG("writing to socket failed"); | |
581 | errno = EINVAL; | |
582 | return -1; | |
583 | } | |
584 | return 0; | |
75818250 | 585 | } |
7a5ca864 | 586 | |
af49bbbe PB |
587 | struct NBDExport { |
588 | BlockDriverState *bs; | |
589 | off_t dev_offset; | |
590 | off_t size; | |
591 | uint8_t *data; | |
592 | uint32_t nbdflags; | |
593 | }; | |
594 | ||
595 | NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, | |
596 | off_t size, uint32_t nbdflags) | |
597 | { | |
598 | NBDExport *exp = g_malloc0(sizeof(NBDExport)); | |
599 | exp->bs = bs; | |
600 | exp->dev_offset = dev_offset; | |
601 | exp->nbdflags = nbdflags; | |
602 | exp->size = size == -1 ? exp->bs->total_sectors * 512 : size; | |
603 | exp->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE); | |
604 | return exp; | |
605 | } | |
606 | ||
607 | void nbd_export_close(NBDExport *exp) | |
608 | { | |
609 | qemu_vfree(exp->data); | |
610 | bdrv_close(exp->bs); | |
611 | g_free(exp); | |
612 | } | |
613 | ||
22045592 PB |
614 | static int nbd_do_send_reply(int csock, struct nbd_reply *reply, |
615 | uint8_t *data, int len) | |
616 | { | |
617 | int rc, ret; | |
618 | ||
619 | if (!len) { | |
620 | rc = nbd_send_reply(csock, reply); | |
621 | if (rc == -1) { | |
622 | rc = -errno; | |
623 | } | |
624 | } else { | |
625 | socket_set_cork(csock, 1); | |
626 | rc = nbd_send_reply(csock, reply); | |
627 | if (rc != -1) { | |
628 | ret = write_sync(csock, data, len); | |
629 | if (ret != len) { | |
630 | errno = EIO; | |
631 | rc = -1; | |
632 | } | |
633 | } | |
634 | if (rc == -1) { | |
635 | rc = -errno; | |
636 | } | |
637 | socket_set_cork(csock, 0); | |
638 | } | |
639 | return rc; | |
640 | } | |
641 | ||
a030b347 PB |
642 | static int nbd_do_receive_request(int csock, struct nbd_request *request, |
643 | uint8_t *data) | |
644 | { | |
645 | int rc; | |
646 | ||
647 | if (nbd_receive_request(csock, request) == -1) { | |
648 | rc = -EIO; | |
649 | goto out; | |
650 | } | |
651 | ||
652 | if (request->len > NBD_BUFFER_SIZE) { | |
653 | LOG("len (%u) is larger than max len (%u)", | |
654 | request->len, NBD_BUFFER_SIZE); | |
655 | rc = -EINVAL; | |
656 | goto out; | |
657 | } | |
658 | ||
659 | if ((request->from + request->len) < request->from) { | |
660 | LOG("integer overflow detected! " | |
661 | "you're probably being attacked"); | |
662 | rc = -EINVAL; | |
663 | goto out; | |
664 | } | |
665 | ||
666 | TRACE("Decoding type"); | |
667 | ||
668 | if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) { | |
669 | TRACE("Reading %u byte(s)", request->len); | |
670 | ||
671 | if (read_sync(csock, data, request->len) != request->len) { | |
672 | LOG("reading from socket failed"); | |
673 | rc = -EIO; | |
674 | goto out; | |
675 | } | |
676 | } | |
677 | rc = 0; | |
678 | ||
679 | out: | |
680 | return rc; | |
681 | } | |
682 | ||
af49bbbe | 683 | int nbd_trip(NBDExport *exp, int csock) |
75818250 | 684 | { |
b2e3d87f NT |
685 | struct nbd_request request; |
686 | struct nbd_reply reply; | |
adcf6302 | 687 | int ret; |
b2e3d87f NT |
688 | |
689 | TRACE("Reading request."); | |
690 | ||
af49bbbe | 691 | ret = nbd_do_receive_request(csock, &request, exp->data); |
a030b347 | 692 | if (ret == -EIO) { |
b2e3d87f | 693 | return -1; |
a030b347 | 694 | } |
b2e3d87f | 695 | |
fae69416 PB |
696 | reply.handle = request.handle; |
697 | reply.error = 0; | |
698 | ||
a030b347 PB |
699 | if (ret < 0) { |
700 | reply.error = -ret; | |
701 | goto error_reply; | |
b2e3d87f NT |
702 | } |
703 | ||
af49bbbe | 704 | if ((request.from + request.len) > exp->size) { |
b2e3d87f NT |
705 | LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64 |
706 | ", Offset: %" PRIu64 "\n", | |
af49bbbe PB |
707 | request.from, request.len, |
708 | (uint64_t)exp->size, exp->dev_offset); | |
b2e3d87f | 709 | LOG("requested operation past EOF--bad client?"); |
fae69416 | 710 | goto invalid_request; |
b2e3d87f NT |
711 | } |
712 | ||
2c7989a9 | 713 | switch (request.type & NBD_CMD_MASK_COMMAND) { |
b2e3d87f NT |
714 | case NBD_CMD_READ: |
715 | TRACE("Request type is READ"); | |
716 | ||
af49bbbe PB |
717 | ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512, |
718 | exp->data, request.len / 512); | |
adcf6302 | 719 | if (ret < 0) { |
b2e3d87f | 720 | LOG("reading from file failed"); |
adcf6302 | 721 | reply.error = -ret; |
fae69416 | 722 | goto error_reply; |
b2e3d87f | 723 | } |
b2e3d87f NT |
724 | |
725 | TRACE("Read %u byte(s)", request.len); | |
af49bbbe | 726 | if (nbd_do_send_reply(csock, &reply, exp->data, request.len) < 0) |
a478f6e5 | 727 | return -1; |
b2e3d87f NT |
728 | break; |
729 | case NBD_CMD_WRITE: | |
730 | TRACE("Request type is WRITE"); | |
731 | ||
af49bbbe | 732 | if (exp->nbdflags & NBD_FLAG_READ_ONLY) { |
b2e3d87f | 733 | TRACE("Server is read-only, return error"); |
fae69416 PB |
734 | reply.error = EROFS; |
735 | goto error_reply; | |
736 | } | |
737 | ||
738 | TRACE("Writing to device"); | |
739 | ||
af49bbbe PB |
740 | ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512, |
741 | exp->data, request.len / 512); | |
fae69416 PB |
742 | if (ret < 0) { |
743 | LOG("writing to file failed"); | |
744 | reply.error = -ret; | |
745 | goto error_reply; | |
746 | } | |
b2e3d87f | 747 | |
fae69416 | 748 | if (request.type & NBD_CMD_FLAG_FUA) { |
af49bbbe | 749 | ret = bdrv_flush(exp->bs); |
adcf6302 | 750 | if (ret < 0) { |
fae69416 | 751 | LOG("flush failed"); |
adcf6302 | 752 | reply.error = -ret; |
fae69416 | 753 | goto error_reply; |
2c7989a9 | 754 | } |
b2e3d87f NT |
755 | } |
756 | ||
22045592 | 757 | if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0) |
b2e3d87f NT |
758 | return -1; |
759 | break; | |
760 | case NBD_CMD_DISC: | |
761 | TRACE("Request type is DISCONNECT"); | |
762 | errno = 0; | |
763 | return 1; | |
1486d04a PB |
764 | case NBD_CMD_FLUSH: |
765 | TRACE("Request type is FLUSH"); | |
766 | ||
af49bbbe | 767 | ret = bdrv_flush(exp->bs); |
1486d04a PB |
768 | if (ret < 0) { |
769 | LOG("flush failed"); | |
770 | reply.error = -ret; | |
771 | } | |
772 | ||
22045592 | 773 | if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0) |
7a706633 PB |
774 | return -1; |
775 | break; | |
776 | case NBD_CMD_TRIM: | |
777 | TRACE("Request type is TRIM"); | |
af49bbbe | 778 | ret = bdrv_discard(exp->bs, (request.from + exp->dev_offset) / 512, |
7a706633 PB |
779 | request.len / 512); |
780 | if (ret < 0) { | |
781 | LOG("discard failed"); | |
782 | reply.error = -ret; | |
783 | } | |
22045592 | 784 | if (nbd_do_send_reply(csock, &reply, NULL, 0) < 0) |
1486d04a PB |
785 | return -1; |
786 | break; | |
b2e3d87f NT |
787 | default: |
788 | LOG("invalid request type (%u) received", request.type); | |
fae69416 PB |
789 | invalid_request: |
790 | reply.error = -EINVAL; | |
791 | error_reply: | |
792 | if (nbd_do_send_reply(csock, &reply, NULL, 0) == -1) | |
793 | return -1; | |
794 | break; | |
b2e3d87f NT |
795 | } |
796 | ||
797 | TRACE("Request/Reply complete"); | |
798 | ||
799 | return 0; | |
7a5ca864 | 800 | } |
af49bbbe PB |
801 | |
802 | int nbd_negotiate(NBDExport *exp, int csock) | |
803 | { | |
804 | return nbd_send_negotiate(csock, exp->size, exp->nbdflags); | |
805 | } |