2 * QEMU Block driver for NBD
4 * Copyright (C) 2008 Bull S.A.S.
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
29 #include "qemu-common.h"
30 #include "block/nbd.h"
32 #include "block/block_int.h"
33 #include "qemu/module.h"
34 #include "qemu/sockets.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qint.h"
38 #include <sys/types.h>
41 #define EN_OPTSTR ":exportname="
43 /* #define DEBUG_NBD */
45 #if defined(DEBUG_NBD)
46 #define logout(fmt, ...) \
47 fprintf(stderr, "nbd\t%-24s" fmt, __func__, ##__VA_ARGS__)
49 #define logout(fmt, ...) ((void)0)
52 #define MAX_NBD_REQUESTS 16
53 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
54 #define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs))
56 typedef struct BDRVNBDState {
64 Coroutine *send_coroutine;
67 Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
68 struct nbd_reply reply;
71 QemuOpts *socket_opts;
73 char *export_name; /* An NBD server may export several devices */
76 static int nbd_parse_uri(const char *filename, QDict *options)
80 QueryParams *qp = NULL;
84 uri = uri_parse(filename);
90 if (!strcmp(uri->scheme, "nbd")) {
92 } else if (!strcmp(uri->scheme, "nbd+tcp")) {
94 } else if (!strcmp(uri->scheme, "nbd+unix")) {
101 p = uri->path ? uri->path : "/";
104 qdict_put(options, "export", qstring_from_str(p));
107 qp = query_params_parse(uri->query);
108 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
114 /* nbd+unix:///export?socket=path */
115 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
119 qdict_put(options, "path", qstring_from_str(qp->p[0].value));
121 /* nbd[+tcp]://host[:port]/export */
127 qdict_put(options, "host", qstring_from_str(uri->server));
129 char* port_str = g_strdup_printf("%d", uri->port);
130 qdict_put(options, "port", qstring_from_str(port_str));
137 query_params_free(qp);
143 static void nbd_parse_filename(const char *filename, QDict *options,
148 const char *host_spec;
149 const char *unixpath;
151 if (qdict_haskey(options, "host")
152 || qdict_haskey(options, "port")
153 || qdict_haskey(options, "path"))
155 error_setg(errp, "host/port/path and a file name may not be specified "
160 if (strstr(filename, "://")) {
161 int ret = nbd_parse_uri(filename, options);
163 error_setg(errp, "No valid URL specified");
168 file = g_strdup(filename);
170 export_name = strstr(file, EN_OPTSTR);
172 if (export_name[strlen(EN_OPTSTR)] == 0) {
175 export_name[0] = 0; /* truncate 'file' */
176 export_name += strlen(EN_OPTSTR);
178 qdict_put(options, "export", qstring_from_str(export_name));
181 /* extract the host_spec - fail if it's not nbd:... */
182 if (!strstart(file, "nbd:", &host_spec)) {
183 error_setg(errp, "File name string for NBD must start with 'nbd:'");
191 /* are we a UNIX or TCP socket? */
192 if (strstart(host_spec, "unix:", &unixpath)) {
193 qdict_put(options, "path", qstring_from_str(unixpath));
195 InetSocketAddress *addr = NULL;
197 addr = inet_parse(host_spec, errp);
198 if (error_is_set(errp)) {
202 qdict_put(options, "host", qstring_from_str(addr->host));
203 qdict_put(options, "port", qstring_from_str(addr->port));
204 qapi_free_InetSocketAddress(addr);
211 static int nbd_config(BDRVNBDState *s, QDict *options)
213 Error *local_err = NULL;
215 if (qdict_haskey(options, "path")) {
216 if (qdict_haskey(options, "host")) {
217 qerror_report(ERROR_CLASS_GENERIC_ERROR, "path and host may not "
218 "be used at the same time.");
222 } else if (qdict_haskey(options, "host")) {
228 s->socket_opts = qemu_opts_create_nofail(&socket_optslist);
230 qemu_opts_absorb_qdict(s->socket_opts, options, &local_err);
231 if (error_is_set(&local_err)) {
232 qerror_report_err(local_err);
233 error_free(local_err);
237 if (!qemu_opt_get(s->socket_opts, "port")) {
238 qemu_opt_set_number(s->socket_opts, "port", NBD_DEFAULT_PORT);
241 s->export_name = g_strdup(qdict_get_try_str(options, "export"));
242 if (s->export_name) {
243 qdict_del(options, "export");
250 static void nbd_coroutine_start(BDRVNBDState *s, struct nbd_request *request)
254 /* Poor man semaphore. The free_sema is locked when no other request
255 * can be accepted, and unlocked after receiving one reply. */
256 if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
257 qemu_co_mutex_lock(&s->free_sema);
258 assert(s->in_flight < MAX_NBD_REQUESTS);
262 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
263 if (s->recv_coroutine[i] == NULL) {
264 s->recv_coroutine[i] = qemu_coroutine_self();
269 assert(i < MAX_NBD_REQUESTS);
270 request->handle = INDEX_TO_HANDLE(s, i);
273 static int nbd_have_request(void *opaque)
275 BDRVNBDState *s = opaque;
277 return s->in_flight > 0;
280 static void nbd_reply_ready(void *opaque)
282 BDRVNBDState *s = opaque;
286 if (s->reply.handle == 0) {
287 /* No reply already in flight. Fetch a header. It is possible
288 * that another thread has done the same thing in parallel, so
289 * the socket is not readable anymore.
291 ret = nbd_receive_reply(s->sock, &s->reply);
292 if (ret == -EAGAIN) {
301 /* There's no need for a mutex on the receive side, because the
302 * handler acts as a synchronization point and ensures that only
303 * one coroutine is called until the reply finishes. */
304 i = HANDLE_TO_INDEX(s, s->reply.handle);
305 if (i >= MAX_NBD_REQUESTS) {
309 if (s->recv_coroutine[i]) {
310 qemu_coroutine_enter(s->recv_coroutine[i], NULL);
315 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
316 if (s->recv_coroutine[i]) {
317 qemu_coroutine_enter(s->recv_coroutine[i], NULL);
322 static void nbd_restart_write(void *opaque)
324 BDRVNBDState *s = opaque;
325 qemu_coroutine_enter(s->send_coroutine, NULL);
328 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
329 QEMUIOVector *qiov, int offset)
333 qemu_co_mutex_lock(&s->send_mutex);
334 s->send_coroutine = qemu_coroutine_self();
335 qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
336 nbd_have_request, s);
339 socket_set_cork(s->sock, 1);
341 rc = nbd_send_request(s->sock, request);
343 ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
344 offset, request->len);
345 if (ret != request->len) {
350 socket_set_cork(s->sock, 0);
353 rc = nbd_send_request(s->sock, request);
355 qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
356 nbd_have_request, s);
357 s->send_coroutine = NULL;
358 qemu_co_mutex_unlock(&s->send_mutex);
362 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
363 struct nbd_reply *reply,
364 QEMUIOVector *qiov, int offset)
368 /* Wait until we're woken up by the read handler. TODO: perhaps
369 * peek at the next reply and avoid yielding if it's ours? */
370 qemu_coroutine_yield();
372 if (reply->handle != request->handle) {
375 if (qiov && reply->error == 0) {
376 ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
377 offset, request->len);
378 if (ret != request->len) {
383 /* Tell the read handler to read another header. */
388 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
390 int i = HANDLE_TO_INDEX(s, request->handle);
391 s->recv_coroutine[i] = NULL;
392 if (s->in_flight-- == MAX_NBD_REQUESTS) {
393 qemu_co_mutex_unlock(&s->free_sema);
397 static int nbd_establish_connection(BlockDriverState *bs)
399 BDRVNBDState *s = bs->opaque;
406 sock = unix_socket_outgoing(qemu_opt_get(s->socket_opts, "path"));
408 sock = tcp_socket_outgoing_opts(s->socket_opts);
410 socket_set_nodelay(sock);
414 /* Failed to establish connection */
416 logout("Failed to establish connection to NBD server\n");
421 ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
424 logout("Failed to negotiate with the NBD server\n");
429 /* Now that we're connected, set the socket to be non-blocking and
430 * kick the reply mechanism. */
431 qemu_set_nonblock(sock);
432 qemu_aio_set_fd_handler(sock, nbd_reply_ready, NULL,
433 nbd_have_request, s);
437 s->blocksize = blocksize;
439 logout("Established connection with NBD server\n");
443 static void nbd_teardown_connection(BlockDriverState *bs)
445 BDRVNBDState *s = bs->opaque;
446 struct nbd_request request;
448 request.type = NBD_CMD_DISC;
451 nbd_send_request(s->sock, &request);
453 qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
454 closesocket(s->sock);
457 static int nbd_open(BlockDriverState *bs, QDict *options, int flags)
459 BDRVNBDState *s = bs->opaque;
462 qemu_co_mutex_init(&s->send_mutex);
463 qemu_co_mutex_init(&s->free_sema);
465 /* Pop the config into our state object. Exit if invalid. */
466 result = nbd_config(s, options);
471 /* establish TCP connection, return error if it fails
472 * TODO: Configurable retry-until-timeout behaviour.
474 result = nbd_establish_connection(bs);
479 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
480 int nb_sectors, QEMUIOVector *qiov,
483 BDRVNBDState *s = bs->opaque;
484 struct nbd_request request;
485 struct nbd_reply reply;
488 request.type = NBD_CMD_READ;
489 request.from = sector_num * 512;
490 request.len = nb_sectors * 512;
492 nbd_coroutine_start(s, &request);
493 ret = nbd_co_send_request(s, &request, NULL, 0);
497 nbd_co_receive_reply(s, &request, &reply, qiov, offset);
499 nbd_coroutine_end(s, &request);
504 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
505 int nb_sectors, QEMUIOVector *qiov,
508 BDRVNBDState *s = bs->opaque;
509 struct nbd_request request;
510 struct nbd_reply reply;
513 request.type = NBD_CMD_WRITE;
514 if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
515 request.type |= NBD_CMD_FLAG_FUA;
518 request.from = sector_num * 512;
519 request.len = nb_sectors * 512;
521 nbd_coroutine_start(s, &request);
522 ret = nbd_co_send_request(s, &request, qiov, offset);
526 nbd_co_receive_reply(s, &request, &reply, NULL, 0);
528 nbd_coroutine_end(s, &request);
532 /* qemu-nbd has a limit of slightly less than 1M per request. Try to
533 * remain aligned to 4K. */
534 #define NBD_MAX_SECTORS 2040
536 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
537 int nb_sectors, QEMUIOVector *qiov)
541 while (nb_sectors > NBD_MAX_SECTORS) {
542 ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
546 offset += NBD_MAX_SECTORS * 512;
547 sector_num += NBD_MAX_SECTORS;
548 nb_sectors -= NBD_MAX_SECTORS;
550 return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
553 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
554 int nb_sectors, QEMUIOVector *qiov)
558 while (nb_sectors > NBD_MAX_SECTORS) {
559 ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
563 offset += NBD_MAX_SECTORS * 512;
564 sector_num += NBD_MAX_SECTORS;
565 nb_sectors -= NBD_MAX_SECTORS;
567 return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
570 static int nbd_co_flush(BlockDriverState *bs)
572 BDRVNBDState *s = bs->opaque;
573 struct nbd_request request;
574 struct nbd_reply reply;
577 if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
581 request.type = NBD_CMD_FLUSH;
582 if (s->nbdflags & NBD_FLAG_SEND_FUA) {
583 request.type |= NBD_CMD_FLAG_FUA;
589 nbd_coroutine_start(s, &request);
590 ret = nbd_co_send_request(s, &request, NULL, 0);
594 nbd_co_receive_reply(s, &request, &reply, NULL, 0);
596 nbd_coroutine_end(s, &request);
600 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
603 BDRVNBDState *s = bs->opaque;
604 struct nbd_request request;
605 struct nbd_reply reply;
608 if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
611 request.type = NBD_CMD_TRIM;
612 request.from = sector_num * 512;;
613 request.len = nb_sectors * 512;
615 nbd_coroutine_start(s, &request);
616 ret = nbd_co_send_request(s, &request, NULL, 0);
620 nbd_co_receive_reply(s, &request, &reply, NULL, 0);
622 nbd_coroutine_end(s, &request);
626 static void nbd_close(BlockDriverState *bs)
628 BDRVNBDState *s = bs->opaque;
629 g_free(s->export_name);
630 qemu_opts_del(s->socket_opts);
632 nbd_teardown_connection(bs);
635 static int64_t nbd_getlength(BlockDriverState *bs)
637 BDRVNBDState *s = bs->opaque;
642 static BlockDriver bdrv_nbd = {
643 .format_name = "nbd",
644 .protocol_name = "nbd",
645 .instance_size = sizeof(BDRVNBDState),
646 .bdrv_parse_filename = nbd_parse_filename,
647 .bdrv_file_open = nbd_open,
648 .bdrv_co_readv = nbd_co_readv,
649 .bdrv_co_writev = nbd_co_writev,
650 .bdrv_close = nbd_close,
651 .bdrv_co_flush_to_os = nbd_co_flush,
652 .bdrv_co_discard = nbd_co_discard,
653 .bdrv_getlength = nbd_getlength,
656 static BlockDriver bdrv_nbd_tcp = {
657 .format_name = "nbd",
658 .protocol_name = "nbd+tcp",
659 .instance_size = sizeof(BDRVNBDState),
660 .bdrv_parse_filename = nbd_parse_filename,
661 .bdrv_file_open = nbd_open,
662 .bdrv_co_readv = nbd_co_readv,
663 .bdrv_co_writev = nbd_co_writev,
664 .bdrv_close = nbd_close,
665 .bdrv_co_flush_to_os = nbd_co_flush,
666 .bdrv_co_discard = nbd_co_discard,
667 .bdrv_getlength = nbd_getlength,
670 static BlockDriver bdrv_nbd_unix = {
671 .format_name = "nbd",
672 .protocol_name = "nbd+unix",
673 .instance_size = sizeof(BDRVNBDState),
674 .bdrv_parse_filename = nbd_parse_filename,
675 .bdrv_file_open = nbd_open,
676 .bdrv_co_readv = nbd_co_readv,
677 .bdrv_co_writev = nbd_co_writev,
678 .bdrv_close = nbd_close,
679 .bdrv_co_flush_to_os = nbd_co_flush,
680 .bdrv_co_discard = nbd_co_discard,
681 .bdrv_getlength = nbd_getlength,
684 static void bdrv_nbd_init(void)
686 bdrv_register(&bdrv_nbd);
687 bdrv_register(&bdrv_nbd_tcp);
688 bdrv_register(&bdrv_nbd_unix);
691 block_init(bdrv_nbd_init);