move atexit(term_exit) and O_NONBLOCK to qemu_chr_open_stdio
[qemu-kvm.git] / nbd.c
blobd8ebc4298fc0f5bd85b7ea1ff7eafd1cdb723af8
1 /*
2 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
4 * Network Block Device
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.
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.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include "nbd.h"
21 #include <errno.h>
22 #include <string.h>
23 #ifndef _WIN32
24 #include <sys/ioctl.h>
25 #endif
26 #if defined(__sun__) || defined(__HAIKU__)
27 #include <sys/ioccom.h>
28 #endif
29 #include <ctype.h>
30 #include <inttypes.h>
32 #include "qemu_socket.h"
34 //#define DEBUG_NBD
36 #ifdef DEBUG_NBD
37 #define TRACE(msg, ...) do { \
38 LOG(msg, ## __VA_ARGS__); \
39 } while(0)
40 #else
41 #define TRACE(msg, ...) \
42 do { } while (0)
43 #endif
45 #define LOG(msg, ...) do { \
46 fprintf(stderr, "%s:%s():L%d: " msg "\n", \
47 __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
48 } while(0)
50 /* This is all part of the "official" NBD API */
52 #define NBD_REPLY_SIZE (4 + 4 + 8)
53 #define NBD_REQUEST_MAGIC 0x25609513
54 #define NBD_REPLY_MAGIC 0x67446698
56 #define NBD_SET_SOCK _IO(0xab, 0)
57 #define NBD_SET_BLKSIZE _IO(0xab, 1)
58 #define NBD_SET_SIZE _IO(0xab, 2)
59 #define NBD_DO_IT _IO(0xab, 3)
60 #define NBD_CLEAR_SOCK _IO(0xab, 4)
61 #define NBD_CLEAR_QUE _IO(0xab, 5)
62 #define NBD_PRINT_DEBUG _IO(0xab, 6)
63 #define NBD_SET_SIZE_BLOCKS _IO(0xab, 7)
64 #define NBD_DISCONNECT _IO(0xab, 8)
66 #define NBD_OPT_EXPORT_NAME (1 << 0)
68 /* That's all folks */
70 #define read_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, true)
71 #define write_sync(fd, buffer, size) nbd_wr_sync(fd, buffer, size, false)
73 size_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
75 size_t offset = 0;
77 while (offset < size) {
78 ssize_t len;
80 if (do_read) {
81 len = recv(fd, buffer + offset, size - offset, 0);
82 } else {
83 len = send(fd, buffer + offset, size - offset, 0);
86 if (len == -1)
87 errno = socket_error();
89 /* recoverable error */
90 if (len == -1 && (errno == EAGAIN || errno == EINTR)) {
91 continue;
94 /* eof */
95 if (len == 0) {
96 break;
99 /* unrecoverable error */
100 if (len == -1) {
101 return 0;
104 offset += len;
107 return offset;
110 int tcp_socket_outgoing(const char *address, uint16_t port)
112 int s;
113 struct in_addr in;
114 struct sockaddr_in addr;
116 s = socket(PF_INET, SOCK_STREAM, 0);
117 if (s == -1) {
118 return -1;
121 if (inet_aton(address, &in) == 0) {
122 struct hostent *ent;
124 ent = gethostbyname(address);
125 if (ent == NULL) {
126 goto error;
129 memcpy(&in, ent->h_addr, sizeof(in));
132 addr.sin_family = AF_INET;
133 addr.sin_port = htons(port);
134 memcpy(&addr.sin_addr.s_addr, &in, sizeof(in));
136 if (connect(s, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
137 goto error;
140 return s;
141 error:
142 closesocket(s);
143 return -1;
146 int tcp_socket_incoming(const char *address, uint16_t port)
148 int s;
149 struct in_addr in;
150 struct sockaddr_in addr;
151 int opt;
153 s = socket(PF_INET, SOCK_STREAM, 0);
154 if (s == -1) {
155 return -1;
158 if (inet_aton(address, &in) == 0) {
159 struct hostent *ent;
161 ent = gethostbyname(address);
162 if (ent == NULL) {
163 goto error;
166 memcpy(&in, ent->h_addr, sizeof(in));
169 addr.sin_family = AF_INET;
170 addr.sin_port = htons(port);
171 memcpy(&addr.sin_addr.s_addr, &in, sizeof(in));
173 opt = 1;
174 if (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,
175 (const void *) &opt, sizeof(opt)) == -1) {
176 goto error;
179 if (bind(s, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
180 goto error;
183 if (listen(s, 128) == -1) {
184 goto error;
187 return s;
188 error:
189 closesocket(s);
190 return -1;
193 #ifndef _WIN32
194 int unix_socket_incoming(const char *path)
196 int s;
197 struct sockaddr_un addr;
199 s = socket(PF_UNIX, SOCK_STREAM, 0);
200 if (s == -1) {
201 return -1;
204 memset(&addr, 0, sizeof(addr));
205 addr.sun_family = AF_UNIX;
206 pstrcpy(addr.sun_path, sizeof(addr.sun_path), path);
208 if (bind(s, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
209 goto error;
212 if (listen(s, 128) == -1) {
213 goto error;
216 return s;
217 error:
218 closesocket(s);
219 return -1;
222 int unix_socket_outgoing(const char *path)
224 int s;
225 struct sockaddr_un addr;
227 s = socket(PF_UNIX, SOCK_STREAM, 0);
228 if (s == -1) {
229 return -1;
232 memset(&addr, 0, sizeof(addr));
233 addr.sun_family = AF_UNIX;
234 pstrcpy(addr.sun_path, sizeof(addr.sun_path), path);
236 if (connect(s, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
237 goto error;
240 return s;
241 error:
242 closesocket(s);
243 return -1;
245 #else
246 int unix_socket_incoming(const char *path)
248 errno = ENOTSUP;
249 return -1;
252 int unix_socket_outgoing(const char *path)
254 errno = ENOTSUP;
255 return -1;
257 #endif
260 /* Basic flow
262 Server Client
264 Negotiate
265 Request
266 Response
267 Request
268 Response
271 Request (type == 2)
274 int nbd_negotiate(int csock, off_t size)
276 char buf[8 + 8 + 8 + 128];
278 /* Negotiate
279 [ 0 .. 7] passwd ("NBDMAGIC")
280 [ 8 .. 15] magic (0x00420281861253)
281 [16 .. 23] size
282 [24 .. 151] reserved (0)
285 TRACE("Beginning negotiation.");
286 memcpy(buf, "NBDMAGIC", 8);
287 cpu_to_be64w((uint64_t*)(buf + 8), 0x00420281861253LL);
288 cpu_to_be64w((uint64_t*)(buf + 16), size);
289 memset(buf + 24, 0, 128);
291 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
292 LOG("write failed");
293 errno = EINVAL;
294 return -1;
297 TRACE("Negotation succeeded.");
299 return 0;
302 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
303 off_t *size, size_t *blocksize)
305 char buf[256];
306 uint64_t magic, s;
307 uint16_t tmp;
309 TRACE("Receiving negotation.");
311 if (read_sync(csock, buf, 8) != 8) {
312 LOG("read failed");
313 errno = EINVAL;
314 return -1;
317 buf[8] = '\0';
318 if (strlen(buf) == 0) {
319 LOG("server connection closed");
320 errno = EINVAL;
321 return -1;
324 TRACE("Magic is %c%c%c%c%c%c%c%c",
325 qemu_isprint(buf[0]) ? buf[0] : '.',
326 qemu_isprint(buf[1]) ? buf[1] : '.',
327 qemu_isprint(buf[2]) ? buf[2] : '.',
328 qemu_isprint(buf[3]) ? buf[3] : '.',
329 qemu_isprint(buf[4]) ? buf[4] : '.',
330 qemu_isprint(buf[5]) ? buf[5] : '.',
331 qemu_isprint(buf[6]) ? buf[6] : '.',
332 qemu_isprint(buf[7]) ? buf[7] : '.');
334 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
335 LOG("Invalid magic received");
336 errno = EINVAL;
337 return -1;
340 if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
341 LOG("read failed");
342 errno = EINVAL;
343 return -1;
345 magic = be64_to_cpu(magic);
346 TRACE("Magic is 0x%" PRIx64, magic);
348 if (name) {
349 uint32_t reserved = 0;
350 uint32_t opt;
351 uint32_t namesize;
353 TRACE("Checking magic (opts_magic)");
354 if (magic != 0x49484156454F5054LL) {
355 LOG("Bad magic received");
356 errno = EINVAL;
357 return -1;
359 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
360 LOG("flags read failed");
361 errno = EINVAL;
362 return -1;
364 *flags = be16_to_cpu(tmp) << 16;
365 /* reserved for future use */
366 if (write_sync(csock, &reserved, sizeof(reserved)) !=
367 sizeof(reserved)) {
368 LOG("write failed (reserved)");
369 errno = EINVAL;
370 return -1;
372 /* write the export name */
373 magic = cpu_to_be64(magic);
374 if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
375 LOG("write failed (magic)");
376 errno = EINVAL;
377 return -1;
379 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
380 if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
381 LOG("write failed (opt)");
382 errno = EINVAL;
383 return -1;
385 namesize = cpu_to_be32(strlen(name));
386 if (write_sync(csock, &namesize, sizeof(namesize)) !=
387 sizeof(namesize)) {
388 LOG("write failed (namesize)");
389 errno = EINVAL;
390 return -1;
392 if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
393 LOG("write failed (name)");
394 errno = EINVAL;
395 return -1;
397 } else {
398 TRACE("Checking magic (cli_magic)");
400 if (magic != 0x00420281861253LL) {
401 LOG("Bad magic received");
402 errno = EINVAL;
403 return -1;
407 if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
408 LOG("read failed");
409 errno = EINVAL;
410 return -1;
412 *size = be64_to_cpu(s);
413 *blocksize = 1024;
414 TRACE("Size is %" PRIu64, *size);
416 if (!name) {
417 if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
418 LOG("read failed (flags)");
419 errno = EINVAL;
420 return -1;
422 *flags = be32_to_cpup(flags);
423 } else {
424 if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
425 LOG("read failed (tmp)");
426 errno = EINVAL;
427 return -1;
429 *flags |= be32_to_cpu(tmp);
431 if (read_sync(csock, &buf, 124) != 124) {
432 LOG("read failed (buf)");
433 errno = EINVAL;
434 return -1;
436 return 0;
439 #ifndef _WIN32
440 int nbd_init(int fd, int csock, off_t size, size_t blocksize)
442 TRACE("Setting block size to %lu", (unsigned long)blocksize);
444 if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) == -1) {
445 int serrno = errno;
446 LOG("Failed setting NBD block size");
447 errno = serrno;
448 return -1;
451 TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
453 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) == -1) {
454 int serrno = errno;
455 LOG("Failed setting size (in blocks)");
456 errno = serrno;
457 return -1;
460 TRACE("Clearing NBD socket");
462 if (ioctl(fd, NBD_CLEAR_SOCK) == -1) {
463 int serrno = errno;
464 LOG("Failed clearing NBD socket");
465 errno = serrno;
466 return -1;
469 TRACE("Setting NBD socket");
471 if (ioctl(fd, NBD_SET_SOCK, csock) == -1) {
472 int serrno = errno;
473 LOG("Failed to set NBD socket");
474 errno = serrno;
475 return -1;
478 TRACE("Negotiation ended");
480 return 0;
483 int nbd_disconnect(int fd)
485 ioctl(fd, NBD_CLEAR_QUE);
486 ioctl(fd, NBD_DISCONNECT);
487 ioctl(fd, NBD_CLEAR_SOCK);
488 return 0;
491 int nbd_client(int fd)
493 int ret;
494 int serrno;
496 TRACE("Doing NBD loop");
498 ret = ioctl(fd, NBD_DO_IT);
499 serrno = errno;
501 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
503 TRACE("Clearing NBD queue");
504 ioctl(fd, NBD_CLEAR_QUE);
506 TRACE("Clearing NBD socket");
507 ioctl(fd, NBD_CLEAR_SOCK);
509 errno = serrno;
510 return ret;
512 #else
513 int nbd_init(int fd, int csock, off_t size, size_t blocksize)
515 errno = ENOTSUP;
516 return -1;
519 int nbd_disconnect(int fd)
521 errno = ENOTSUP;
522 return -1;
525 int nbd_client(int fd)
527 errno = ENOTSUP;
528 return -1;
530 #endif
532 int nbd_send_request(int csock, struct nbd_request *request)
534 uint8_t buf[4 + 4 + 8 + 8 + 4];
536 cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
537 cpu_to_be32w((uint32_t*)(buf + 4), request->type);
538 cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
539 cpu_to_be64w((uint64_t*)(buf + 16), request->from);
540 cpu_to_be32w((uint32_t*)(buf + 24), request->len);
542 TRACE("Sending request to client");
544 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
545 LOG("writing to socket failed");
546 errno = EINVAL;
547 return -1;
549 return 0;
553 static int nbd_receive_request(int csock, struct nbd_request *request)
555 uint8_t buf[4 + 4 + 8 + 8 + 4];
556 uint32_t magic;
558 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
559 LOG("read failed");
560 errno = EINVAL;
561 return -1;
564 /* Request
565 [ 0 .. 3] magic (NBD_REQUEST_MAGIC)
566 [ 4 .. 7] type (0 == READ, 1 == WRITE)
567 [ 8 .. 15] handle
568 [16 .. 23] from
569 [24 .. 27] len
572 magic = be32_to_cpup((uint32_t*)buf);
573 request->type = be32_to_cpup((uint32_t*)(buf + 4));
574 request->handle = be64_to_cpup((uint64_t*)(buf + 8));
575 request->from = be64_to_cpup((uint64_t*)(buf + 16));
576 request->len = be32_to_cpup((uint32_t*)(buf + 24));
578 TRACE("Got request: "
579 "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
580 magic, request->type, request->from, request->len);
582 if (magic != NBD_REQUEST_MAGIC) {
583 LOG("invalid magic (got 0x%x)", magic);
584 errno = EINVAL;
585 return -1;
587 return 0;
590 int nbd_receive_reply(int csock, struct nbd_reply *reply)
592 uint8_t buf[NBD_REPLY_SIZE];
593 uint32_t magic;
595 memset(buf, 0xAA, sizeof(buf));
597 if (read_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
598 LOG("read failed");
599 errno = EINVAL;
600 return -1;
603 /* Reply
604 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
605 [ 4 .. 7] error (0 == no error)
606 [ 7 .. 15] handle
609 magic = be32_to_cpup((uint32_t*)buf);
610 reply->error = be32_to_cpup((uint32_t*)(buf + 4));
611 reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
613 TRACE("Got reply: "
614 "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
615 magic, reply->error, reply->handle);
617 if (magic != NBD_REPLY_MAGIC) {
618 LOG("invalid magic (got 0x%x)", magic);
619 errno = EINVAL;
620 return -1;
622 return 0;
625 static int nbd_send_reply(int csock, struct nbd_reply *reply)
627 uint8_t buf[4 + 4 + 8];
629 /* Reply
630 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
631 [ 4 .. 7] error (0 == no error)
632 [ 7 .. 15] handle
634 cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
635 cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
636 cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
638 TRACE("Sending response to client");
640 if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
641 LOG("writing to socket failed");
642 errno = EINVAL;
643 return -1;
645 return 0;
648 int nbd_trip(BlockDriverState *bs, int csock, off_t size, uint64_t dev_offset,
649 off_t *offset, bool readonly, uint8_t *data, int data_size)
651 struct nbd_request request;
652 struct nbd_reply reply;
654 TRACE("Reading request.");
656 if (nbd_receive_request(csock, &request) == -1)
657 return -1;
659 if (request.len + NBD_REPLY_SIZE > data_size) {
660 LOG("len (%u) is larger than max len (%u)",
661 request.len + NBD_REPLY_SIZE, data_size);
662 errno = EINVAL;
663 return -1;
666 if ((request.from + request.len) < request.from) {
667 LOG("integer overflow detected! "
668 "you're probably being attacked");
669 errno = EINVAL;
670 return -1;
673 if ((request.from + request.len) > size) {
674 LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
675 ", Offset: %" PRIu64 "\n",
676 request.from, request.len, (uint64_t)size, dev_offset);
677 LOG("requested operation past EOF--bad client?");
678 errno = EINVAL;
679 return -1;
682 TRACE("Decoding type");
684 reply.handle = request.handle;
685 reply.error = 0;
687 switch (request.type) {
688 case NBD_CMD_READ:
689 TRACE("Request type is READ");
691 if (bdrv_read(bs, (request.from + dev_offset) / 512,
692 data + NBD_REPLY_SIZE,
693 request.len / 512) == -1) {
694 LOG("reading from file failed");
695 errno = EINVAL;
696 return -1;
698 *offset += request.len;
700 TRACE("Read %u byte(s)", request.len);
702 /* Reply
703 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
704 [ 4 .. 7] error (0 == no error)
705 [ 7 .. 15] handle
708 cpu_to_be32w((uint32_t*)data, NBD_REPLY_MAGIC);
709 cpu_to_be32w((uint32_t*)(data + 4), reply.error);
710 cpu_to_be64w((uint64_t*)(data + 8), reply.handle);
712 TRACE("Sending data to client");
714 if (write_sync(csock, data,
715 request.len + NBD_REPLY_SIZE) !=
716 request.len + NBD_REPLY_SIZE) {
717 LOG("writing to socket failed");
718 errno = EINVAL;
719 return -1;
721 break;
722 case NBD_CMD_WRITE:
723 TRACE("Request type is WRITE");
725 TRACE("Reading %u byte(s)", request.len);
727 if (read_sync(csock, data, request.len) != request.len) {
728 LOG("reading from socket failed");
729 errno = EINVAL;
730 return -1;
733 if (readonly) {
734 TRACE("Server is read-only, return error");
735 reply.error = 1;
736 } else {
737 TRACE("Writing to device");
739 if (bdrv_write(bs, (request.from + dev_offset) / 512,
740 data, request.len / 512) == -1) {
741 LOG("writing to file failed");
742 errno = EINVAL;
743 return -1;
746 *offset += request.len;
749 if (nbd_send_reply(csock, &reply) == -1)
750 return -1;
751 break;
752 case NBD_CMD_DISC:
753 TRACE("Request type is DISCONNECT");
754 errno = 0;
755 return 1;
756 default:
757 LOG("invalid request type (%u) received", request.type);
758 errno = EINVAL;
759 return -1;
762 TRACE("Request/Reply complete");
764 return 0;