nbd: Treat flags vs. command type as separate fields
[qemu/ar7.git] / nbd / client.c
blob4620e8dcbaec8ded335ddc5586650a7c6ccf549c
1 /*
2 * Copyright (C) 2016 Red Hat, Inc.
3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws>
5 * Network Block Device Client Side
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.
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.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "qapi/error.h"
22 #include "nbd-internal.h"
24 static int nbd_errno_to_system_errno(int err)
26 switch (err) {
27 case NBD_SUCCESS:
28 return 0;
29 case NBD_EPERM:
30 return EPERM;
31 case NBD_EIO:
32 return EIO;
33 case NBD_ENOMEM:
34 return ENOMEM;
35 case NBD_ENOSPC:
36 return ENOSPC;
37 default:
38 TRACE("Squashing unexpected error %d to EINVAL", err);
39 /* fallthrough */
40 case NBD_EINVAL:
41 return EINVAL;
45 /* Definitions for opaque data types */
47 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
49 /* That's all folks */
51 /* Basic flow for negotiation
53 Server Client
54 Negotiate
58 Server Client
59 Negotiate #1
60 Option
61 Negotiate #2
63 ----
65 followed by
67 Server Client
68 Request
69 Response
70 Request
71 Response
72 ...
73 ...
74 Request (type == 2)
79 /* If type represents success, return 1 without further action.
80 * If type represents an error reply, consume the rest of the packet on ioc.
81 * Then return 0 for unsupported (so the client can fall back to
82 * other approaches), or -1 with errp set for other errors.
84 static int nbd_handle_reply_err(QIOChannel *ioc, uint32_t opt, uint32_t type,
85 Error **errp)
87 uint32_t len;
88 char *msg = NULL;
89 int result = -1;
91 if (!(type & (1 << 31))) {
92 return 1;
95 if (read_sync(ioc, &len, sizeof(len)) != sizeof(len)) {
96 error_setg(errp, "failed to read option length");
97 return -1;
99 len = be32_to_cpu(len);
100 if (len) {
101 if (len > NBD_MAX_BUFFER_SIZE) {
102 error_setg(errp, "server's error message is too long");
103 goto cleanup;
105 msg = g_malloc(len + 1);
106 if (read_sync(ioc, msg, len) != len) {
107 error_setg(errp, "failed to read option error message");
108 goto cleanup;
110 msg[len] = '\0';
113 switch (type) {
114 case NBD_REP_ERR_UNSUP:
115 TRACE("server doesn't understand request %" PRIx32
116 ", attempting fallback", opt);
117 result = 0;
118 goto cleanup;
120 case NBD_REP_ERR_POLICY:
121 error_setg(errp, "Denied by server for option %" PRIx32, opt);
122 break;
124 case NBD_REP_ERR_INVALID:
125 error_setg(errp, "Invalid data length for option %" PRIx32, opt);
126 break;
128 case NBD_REP_ERR_TLS_REQD:
129 error_setg(errp, "TLS negotiation required before option %" PRIx32,
130 opt);
131 break;
133 default:
134 error_setg(errp, "Unknown error code when asking for option %" PRIx32,
135 opt);
136 break;
139 if (msg) {
140 error_append_hint(errp, "%s\n", msg);
143 cleanup:
144 g_free(msg);
145 return result;
148 static int nbd_receive_list(QIOChannel *ioc, char **name, Error **errp)
150 uint64_t magic;
151 uint32_t opt;
152 uint32_t type;
153 uint32_t len;
154 uint32_t namelen;
155 int error;
157 *name = NULL;
158 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
159 error_setg(errp, "failed to read list option magic");
160 return -1;
162 magic = be64_to_cpu(magic);
163 if (magic != NBD_REP_MAGIC) {
164 error_setg(errp, "Unexpected option list magic");
165 return -1;
167 if (read_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
168 error_setg(errp, "failed to read list option");
169 return -1;
171 opt = be32_to_cpu(opt);
172 if (opt != NBD_OPT_LIST) {
173 error_setg(errp, "Unexpected option type %" PRIx32 " expected %x",
174 opt, NBD_OPT_LIST);
175 return -1;
178 if (read_sync(ioc, &type, sizeof(type)) != sizeof(type)) {
179 error_setg(errp, "failed to read list option type");
180 return -1;
182 type = be32_to_cpu(type);
183 error = nbd_handle_reply_err(ioc, opt, type, errp);
184 if (error <= 0) {
185 return error;
188 if (read_sync(ioc, &len, sizeof(len)) != sizeof(len)) {
189 error_setg(errp, "failed to read option length");
190 return -1;
192 len = be32_to_cpu(len);
194 if (type == NBD_REP_ACK) {
195 if (len != 0) {
196 error_setg(errp, "length too long for option end");
197 return -1;
199 } else if (type == NBD_REP_SERVER) {
200 if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
201 error_setg(errp, "incorrect option length");
202 return -1;
204 if (read_sync(ioc, &namelen, sizeof(namelen)) != sizeof(namelen)) {
205 error_setg(errp, "failed to read option name length");
206 return -1;
208 namelen = be32_to_cpu(namelen);
209 len -= sizeof(namelen);
210 if (len < namelen) {
211 error_setg(errp, "incorrect option name length");
212 return -1;
214 if (namelen > NBD_MAX_NAME_SIZE) {
215 error_setg(errp, "export name length too long %" PRIu32, namelen);
216 return -1;
219 *name = g_new0(char, namelen + 1);
220 if (read_sync(ioc, *name, namelen) != namelen) {
221 error_setg(errp, "failed to read export name");
222 g_free(*name);
223 *name = NULL;
224 return -1;
226 (*name)[namelen] = '\0';
227 len -= namelen;
228 if (len) {
229 char *buf = g_malloc(len + 1);
230 if (read_sync(ioc, buf, len) != len) {
231 error_setg(errp, "failed to read export description");
232 g_free(*name);
233 g_free(buf);
234 *name = NULL;
235 return -1;
237 buf[len] = '\0';
238 TRACE("Ignoring export description: %s", buf);
239 g_free(buf);
241 } else {
242 error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
243 type, NBD_REP_SERVER);
244 return -1;
246 return 1;
250 static int nbd_receive_query_exports(QIOChannel *ioc,
251 const char *wantname,
252 Error **errp)
254 uint64_t magic = cpu_to_be64(NBD_OPTS_MAGIC);
255 uint32_t opt = cpu_to_be32(NBD_OPT_LIST);
256 uint32_t length = 0;
257 bool foundExport = false;
259 TRACE("Querying export list");
260 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
261 error_setg(errp, "Failed to send list option magic");
262 return -1;
265 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
266 error_setg(errp, "Failed to send list option number");
267 return -1;
270 if (write_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
271 error_setg(errp, "Failed to send list option length");
272 return -1;
275 TRACE("Reading available export names");
276 while (1) {
277 char *name = NULL;
278 int ret = nbd_receive_list(ioc, &name, errp);
280 if (ret < 0) {
281 g_free(name);
282 name = NULL;
283 return -1;
285 if (ret == 0) {
286 /* Server doesn't support export listing, so
287 * we will just assume an export with our
288 * wanted name exists */
289 foundExport = true;
290 break;
292 if (name == NULL) {
293 TRACE("End of export name list");
294 break;
296 if (g_str_equal(name, wantname)) {
297 foundExport = true;
298 TRACE("Found desired export name '%s'", name);
299 } else {
300 TRACE("Ignored export name '%s'", name);
302 g_free(name);
305 if (!foundExport) {
306 error_setg(errp, "No export with name '%s' available", wantname);
307 return -1;
310 return 0;
313 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
314 QCryptoTLSCreds *tlscreds,
315 const char *hostname, Error **errp)
317 uint64_t magic = cpu_to_be64(NBD_OPTS_MAGIC);
318 uint32_t opt = cpu_to_be32(NBD_OPT_STARTTLS);
319 uint32_t length = 0;
320 uint32_t type;
321 QIOChannelTLS *tioc;
322 struct NBDTLSHandshakeData data = { 0 };
324 TRACE("Requesting TLS from server");
325 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
326 error_setg(errp, "Failed to send option magic");
327 return NULL;
330 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
331 error_setg(errp, "Failed to send option number");
332 return NULL;
335 if (write_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
336 error_setg(errp, "Failed to send option length");
337 return NULL;
340 TRACE("Getting TLS reply from server1");
341 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
342 error_setg(errp, "failed to read option magic");
343 return NULL;
345 magic = be64_to_cpu(magic);
346 if (magic != NBD_REP_MAGIC) {
347 error_setg(errp, "Unexpected option magic");
348 return NULL;
350 TRACE("Getting TLS reply from server2");
351 if (read_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
352 error_setg(errp, "failed to read option");
353 return NULL;
355 opt = be32_to_cpu(opt);
356 if (opt != NBD_OPT_STARTTLS) {
357 error_setg(errp, "Unexpected option type %" PRIx32 " expected %x",
358 opt, NBD_OPT_STARTTLS);
359 return NULL;
362 TRACE("Getting TLS reply from server");
363 if (read_sync(ioc, &type, sizeof(type)) != sizeof(type)) {
364 error_setg(errp, "failed to read option type");
365 return NULL;
367 type = be32_to_cpu(type);
368 if (type != NBD_REP_ACK) {
369 error_setg(errp, "Server rejected request to start TLS %" PRIx32,
370 type);
371 return NULL;
374 TRACE("Getting TLS reply from server");
375 if (read_sync(ioc, &length, sizeof(length)) != sizeof(length)) {
376 error_setg(errp, "failed to read option length");
377 return NULL;
379 length = be32_to_cpu(length);
380 if (length != 0) {
381 error_setg(errp, "Start TLS response was not zero %" PRIu32,
382 length);
383 return NULL;
386 TRACE("TLS request approved, setting up TLS");
387 tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
388 if (!tioc) {
389 return NULL;
391 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
392 data.loop = g_main_loop_new(g_main_context_default(), FALSE);
393 TRACE("Starting TLS handshake");
394 qio_channel_tls_handshake(tioc,
395 nbd_tls_handshake,
396 &data,
397 NULL);
399 if (!data.complete) {
400 g_main_loop_run(data.loop);
402 g_main_loop_unref(data.loop);
403 if (data.error) {
404 error_propagate(errp, data.error);
405 object_unref(OBJECT(tioc));
406 return NULL;
409 return QIO_CHANNEL(tioc);
413 int nbd_receive_negotiate(QIOChannel *ioc, const char *name, uint16_t *flags,
414 QCryptoTLSCreds *tlscreds, const char *hostname,
415 QIOChannel **outioc,
416 off_t *size, Error **errp)
418 char buf[256];
419 uint64_t magic, s;
420 int rc;
422 TRACE("Receiving negotiation tlscreds=%p hostname=%s.",
423 tlscreds, hostname ? hostname : "<null>");
425 rc = -EINVAL;
427 if (outioc) {
428 *outioc = NULL;
430 if (tlscreds && !outioc) {
431 error_setg(errp, "Output I/O channel required for TLS");
432 goto fail;
435 if (read_sync(ioc, buf, 8) != 8) {
436 error_setg(errp, "Failed to read data");
437 goto fail;
440 buf[8] = '\0';
441 if (strlen(buf) == 0) {
442 error_setg(errp, "Server connection closed unexpectedly");
443 goto fail;
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] : '.');
456 if (memcmp(buf, "NBDMAGIC", 8) != 0) {
457 error_setg(errp, "Invalid magic received");
458 goto fail;
461 if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
462 error_setg(errp, "Failed to read magic");
463 goto fail;
465 magic = be64_to_cpu(magic);
466 TRACE("Magic is 0x%" PRIx64, magic);
468 if (magic == NBD_OPTS_MAGIC) {
469 uint32_t clientflags = 0;
470 uint32_t opt;
471 uint32_t namesize;
472 uint16_t globalflags;
473 bool fixedNewStyle = false;
475 if (read_sync(ioc, &globalflags, sizeof(globalflags)) !=
476 sizeof(globalflags)) {
477 error_setg(errp, "Failed to read server flags");
478 goto fail;
480 globalflags = be16_to_cpu(globalflags);
481 TRACE("Global flags are %" PRIx32, globalflags);
482 if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
483 fixedNewStyle = true;
484 TRACE("Server supports fixed new style");
485 clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
487 /* client requested flags */
488 clientflags = cpu_to_be32(clientflags);
489 if (write_sync(ioc, &clientflags, sizeof(clientflags)) !=
490 sizeof(clientflags)) {
491 error_setg(errp, "Failed to send clientflags field");
492 goto fail;
494 if (tlscreds) {
495 if (fixedNewStyle) {
496 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
497 if (!*outioc) {
498 goto fail;
500 ioc = *outioc;
501 } else {
502 error_setg(errp, "Server does not support STARTTLS");
503 goto fail;
506 if (!name) {
507 TRACE("Using default NBD export name \"\"");
508 name = "";
510 if (fixedNewStyle) {
511 /* Check our desired export is present in the
512 * server export list. Since NBD_OPT_EXPORT_NAME
513 * cannot return an error message, running this
514 * query gives us good error reporting if the
515 * server required TLS
517 if (nbd_receive_query_exports(ioc, name, errp) < 0) {
518 goto fail;
521 /* write the export name */
522 magic = cpu_to_be64(magic);
523 if (write_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
524 error_setg(errp, "Failed to send export name magic");
525 goto fail;
527 opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
528 if (write_sync(ioc, &opt, sizeof(opt)) != sizeof(opt)) {
529 error_setg(errp, "Failed to send export name option number");
530 goto fail;
532 namesize = cpu_to_be32(strlen(name));
533 if (write_sync(ioc, &namesize, sizeof(namesize)) !=
534 sizeof(namesize)) {
535 error_setg(errp, "Failed to send export name length");
536 goto fail;
538 if (write_sync(ioc, (char *)name, strlen(name)) != strlen(name)) {
539 error_setg(errp, "Failed to send export name");
540 goto fail;
543 if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
544 error_setg(errp, "Failed to read export length");
545 goto fail;
547 *size = be64_to_cpu(s);
549 if (read_sync(ioc, flags, sizeof(*flags)) != sizeof(*flags)) {
550 error_setg(errp, "Failed to read export flags");
551 goto fail;
553 be16_to_cpus(flags);
554 } else if (magic == NBD_CLIENT_MAGIC) {
555 uint32_t oldflags;
557 if (name) {
558 error_setg(errp, "Server does not support export names");
559 goto fail;
561 if (tlscreds) {
562 error_setg(errp, "Server does not support STARTTLS");
563 goto fail;
566 if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
567 error_setg(errp, "Failed to read export length");
568 goto fail;
570 *size = be64_to_cpu(s);
571 TRACE("Size is %" PRIu64, *size);
573 if (read_sync(ioc, &oldflags, sizeof(oldflags)) != sizeof(oldflags)) {
574 error_setg(errp, "Failed to read export flags");
575 goto fail;
577 be32_to_cpus(&oldflags);
578 if (oldflags & ~0xffff) {
579 error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
580 goto fail;
582 *flags = oldflags;
583 } else {
584 error_setg(errp, "Bad magic received");
585 goto fail;
588 TRACE("Size is %" PRIu64 ", export flags %" PRIx16, *size, *flags);
589 if (read_sync(ioc, &buf, 124) != 124) {
590 error_setg(errp, "Failed to read reserved block");
591 goto fail;
593 rc = 0;
595 fail:
596 return rc;
599 #ifdef __linux__
600 int nbd_init(int fd, QIOChannelSocket *sioc, uint16_t flags, off_t size)
602 unsigned long sectors = size / BDRV_SECTOR_SIZE;
603 if (size / BDRV_SECTOR_SIZE != sectors) {
604 LOG("Export size %lld too large for 32-bit kernel", (long long) size);
605 return -E2BIG;
608 TRACE("Setting NBD socket");
610 if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
611 int serrno = errno;
612 LOG("Failed to set NBD socket");
613 return -serrno;
616 TRACE("Setting block size to %lu", (unsigned long)BDRV_SECTOR_SIZE);
618 if (ioctl(fd, NBD_SET_BLKSIZE, (unsigned long)BDRV_SECTOR_SIZE) < 0) {
619 int serrno = errno;
620 LOG("Failed setting NBD block size");
621 return -serrno;
624 TRACE("Setting size to %lu block(s)", sectors);
625 if (size % BDRV_SECTOR_SIZE) {
626 TRACE("Ignoring trailing %d bytes of export",
627 (int) (size % BDRV_SECTOR_SIZE));
630 if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
631 int serrno = errno;
632 LOG("Failed setting size (in blocks)");
633 return -serrno;
636 if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) flags) < 0) {
637 if (errno == ENOTTY) {
638 int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
639 TRACE("Setting readonly attribute");
641 if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
642 int serrno = errno;
643 LOG("Failed setting read-only attribute");
644 return -serrno;
646 } else {
647 int serrno = errno;
648 LOG("Failed setting flags");
649 return -serrno;
653 TRACE("Negotiation ended");
655 return 0;
658 int nbd_client(int fd)
660 int ret;
661 int serrno;
663 TRACE("Doing NBD loop");
665 ret = ioctl(fd, NBD_DO_IT);
666 if (ret < 0 && errno == EPIPE) {
667 /* NBD_DO_IT normally returns EPIPE when someone has disconnected
668 * the socket via NBD_DISCONNECT. We do not want to return 1 in
669 * that case.
671 ret = 0;
673 serrno = errno;
675 TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
677 TRACE("Clearing NBD queue");
678 ioctl(fd, NBD_CLEAR_QUE);
680 TRACE("Clearing NBD socket");
681 ioctl(fd, NBD_CLEAR_SOCK);
683 errno = serrno;
684 return ret;
687 int nbd_disconnect(int fd)
689 ioctl(fd, NBD_CLEAR_QUE);
690 ioctl(fd, NBD_DISCONNECT);
691 ioctl(fd, NBD_CLEAR_SOCK);
692 return 0;
695 #else
696 int nbd_init(int fd, QIOChannelSocket *ioc, uint16_t flags, off_t size)
698 return -ENOTSUP;
701 int nbd_client(int fd)
703 return -ENOTSUP;
705 int nbd_disconnect(int fd)
707 return -ENOTSUP;
709 #endif
711 ssize_t nbd_send_request(QIOChannel *ioc, struct nbd_request *request)
713 uint8_t buf[NBD_REQUEST_SIZE];
714 ssize_t ret;
716 TRACE("Sending request to server: "
717 "{ .from = %" PRIu64", .len = %" PRIu32 ", .handle = %" PRIu64
718 ", .flags = %" PRIx16 ", .type = %" PRIu16 " }",
719 request->from, request->len, request->handle,
720 request->flags, request->type);
722 stl_be_p(buf, NBD_REQUEST_MAGIC);
723 stw_be_p(buf + 4, request->flags);
724 stw_be_p(buf + 6, request->type);
725 stq_be_p(buf + 8, request->handle);
726 stq_be_p(buf + 16, request->from);
727 stl_be_p(buf + 24, request->len);
729 ret = write_sync(ioc, buf, sizeof(buf));
730 if (ret < 0) {
731 return ret;
734 if (ret != sizeof(buf)) {
735 LOG("writing to socket failed");
736 return -EINVAL;
738 return 0;
741 ssize_t nbd_receive_reply(QIOChannel *ioc, struct nbd_reply *reply)
743 uint8_t buf[NBD_REPLY_SIZE];
744 uint32_t magic;
745 ssize_t ret;
747 ret = read_sync(ioc, buf, sizeof(buf));
748 if (ret < 0) {
749 return ret;
752 if (ret != sizeof(buf)) {
753 LOG("read failed");
754 return -EINVAL;
757 /* Reply
758 [ 0 .. 3] magic (NBD_REPLY_MAGIC)
759 [ 4 .. 7] error (0 == no error)
760 [ 7 .. 15] handle
763 magic = ldl_be_p(buf);
764 reply->error = ldl_be_p(buf + 4);
765 reply->handle = ldq_be_p(buf + 8);
767 reply->error = nbd_errno_to_system_errno(reply->error);
769 TRACE("Got reply: { magic = 0x%" PRIx32 ", .error = % " PRId32
770 ", handle = %" PRIu64" }",
771 magic, reply->error, reply->handle);
773 if (magic != NBD_REPLY_MAGIC) {
774 LOG("invalid magic (got 0x%" PRIx32 ")", magic);
775 return -EINVAL;
777 return 0;