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 "nbd-client.h"
30 #include "qemu/sockets.h"
32 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
33 #define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs))
35 static void nbd_recv_coroutines_enter_all(NbdClientSession *s)
39 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
40 if (s->recv_coroutine[i]) {
41 qemu_coroutine_enter(s->recv_coroutine[i], NULL);
46 static void nbd_teardown_connection(BlockDriverState *bs)
48 NbdClientSession *client = nbd_get_client_session(bs);
50 /* finish any pending coroutines */
51 shutdown(client->sock, 2);
52 nbd_recv_coroutines_enter_all(client);
54 nbd_client_detach_aio_context(bs);
55 closesocket(client->sock);
59 static void nbd_reply_ready(void *opaque)
61 BlockDriverState *bs = opaque;
62 NbdClientSession *s = nbd_get_client_session(bs);
66 if (s->reply.handle == 0) {
67 /* No reply already in flight. Fetch a header. It is possible
68 * that another thread has done the same thing in parallel, so
69 * the socket is not readable anymore.
71 ret = nbd_receive_reply(s->sock, &s->reply);
81 /* There's no need for a mutex on the receive side, because the
82 * handler acts as a synchronization point and ensures that only
83 * one coroutine is called until the reply finishes. */
84 i = HANDLE_TO_INDEX(s, s->reply.handle);
85 if (i >= MAX_NBD_REQUESTS) {
89 if (s->recv_coroutine[i]) {
90 qemu_coroutine_enter(s->recv_coroutine[i], NULL);
95 nbd_teardown_connection(bs);
98 static void nbd_restart_write(void *opaque)
100 BlockDriverState *bs = opaque;
102 qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine, NULL);
105 static int nbd_co_send_request(BlockDriverState *bs,
106 struct nbd_request *request,
107 QEMUIOVector *qiov, int offset)
109 NbdClientSession *s = nbd_get_client_session(bs);
110 AioContext *aio_context;
113 qemu_co_mutex_lock(&s->send_mutex);
114 s->send_coroutine = qemu_coroutine_self();
115 aio_context = bdrv_get_aio_context(bs);
116 aio_set_fd_handler(aio_context, s->sock,
117 nbd_reply_ready, nbd_restart_write, bs);
120 socket_set_cork(s->sock, 1);
122 rc = nbd_send_request(s->sock, request);
124 ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
125 offset, request->len);
126 if (ret != request->len) {
131 socket_set_cork(s->sock, 0);
134 rc = nbd_send_request(s->sock, request);
136 aio_set_fd_handler(aio_context, s->sock, nbd_reply_ready, NULL, bs);
137 s->send_coroutine = NULL;
138 qemu_co_mutex_unlock(&s->send_mutex);
142 static void nbd_co_receive_reply(NbdClientSession *s,
143 struct nbd_request *request, struct nbd_reply *reply,
144 QEMUIOVector *qiov, int offset)
148 /* Wait until we're woken up by the read handler. TODO: perhaps
149 * peek at the next reply and avoid yielding if it's ours? */
150 qemu_coroutine_yield();
152 if (reply->handle != request->handle) {
155 if (qiov && reply->error == 0) {
156 ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
157 offset, request->len);
158 if (ret != request->len) {
163 /* Tell the read handler to read another header. */
168 static void nbd_coroutine_start(NbdClientSession *s,
169 struct nbd_request *request)
173 /* Poor man semaphore. The free_sema is locked when no other request
174 * can be accepted, and unlocked after receiving one reply. */
175 if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
176 qemu_co_mutex_lock(&s->free_sema);
177 assert(s->in_flight < MAX_NBD_REQUESTS);
181 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
182 if (s->recv_coroutine[i] == NULL) {
183 s->recv_coroutine[i] = qemu_coroutine_self();
188 assert(i < MAX_NBD_REQUESTS);
189 request->handle = INDEX_TO_HANDLE(s, i);
192 static void nbd_coroutine_end(NbdClientSession *s,
193 struct nbd_request *request)
195 int i = HANDLE_TO_INDEX(s, request->handle);
196 s->recv_coroutine[i] = NULL;
197 if (s->in_flight-- == MAX_NBD_REQUESTS) {
198 qemu_co_mutex_unlock(&s->free_sema);
202 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
203 int nb_sectors, QEMUIOVector *qiov,
206 NbdClientSession *client = nbd_get_client_session(bs);
207 struct nbd_request request = { .type = NBD_CMD_READ };
208 struct nbd_reply reply;
211 request.from = sector_num * 512;
212 request.len = nb_sectors * 512;
214 nbd_coroutine_start(client, &request);
215 ret = nbd_co_send_request(bs, &request, NULL, 0);
219 nbd_co_receive_reply(client, &request, &reply, qiov, offset);
221 nbd_coroutine_end(client, &request);
226 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
227 int nb_sectors, QEMUIOVector *qiov,
230 NbdClientSession *client = nbd_get_client_session(bs);
231 struct nbd_request request = { .type = NBD_CMD_WRITE };
232 struct nbd_reply reply;
235 if (!bdrv_enable_write_cache(bs) &&
236 (client->nbdflags & NBD_FLAG_SEND_FUA)) {
237 request.type |= NBD_CMD_FLAG_FUA;
240 request.from = sector_num * 512;
241 request.len = nb_sectors * 512;
243 nbd_coroutine_start(client, &request);
244 ret = nbd_co_send_request(bs, &request, qiov, offset);
248 nbd_co_receive_reply(client, &request, &reply, NULL, 0);
250 nbd_coroutine_end(client, &request);
254 /* qemu-nbd has a limit of slightly less than 1M per request. Try to
255 * remain aligned to 4K. */
256 #define NBD_MAX_SECTORS 2040
258 int nbd_client_co_readv(BlockDriverState *bs, int64_t sector_num,
259 int nb_sectors, QEMUIOVector *qiov)
263 while (nb_sectors > NBD_MAX_SECTORS) {
264 ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
268 offset += NBD_MAX_SECTORS * 512;
269 sector_num += NBD_MAX_SECTORS;
270 nb_sectors -= NBD_MAX_SECTORS;
272 return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
275 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
276 int nb_sectors, QEMUIOVector *qiov)
280 while (nb_sectors > NBD_MAX_SECTORS) {
281 ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
285 offset += NBD_MAX_SECTORS * 512;
286 sector_num += NBD_MAX_SECTORS;
287 nb_sectors -= NBD_MAX_SECTORS;
289 return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
292 int nbd_client_co_flush(BlockDriverState *bs)
294 NbdClientSession *client = nbd_get_client_session(bs);
295 struct nbd_request request = { .type = NBD_CMD_FLUSH };
296 struct nbd_reply reply;
299 if (!(client->nbdflags & NBD_FLAG_SEND_FLUSH)) {
303 if (client->nbdflags & NBD_FLAG_SEND_FUA) {
304 request.type |= NBD_CMD_FLAG_FUA;
310 nbd_coroutine_start(client, &request);
311 ret = nbd_co_send_request(bs, &request, NULL, 0);
315 nbd_co_receive_reply(client, &request, &reply, NULL, 0);
317 nbd_coroutine_end(client, &request);
321 int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num,
324 NbdClientSession *client = nbd_get_client_session(bs);
325 struct nbd_request request = { .type = NBD_CMD_TRIM };
326 struct nbd_reply reply;
329 if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) {
332 request.from = sector_num * 512;
333 request.len = nb_sectors * 512;
335 nbd_coroutine_start(client, &request);
336 ret = nbd_co_send_request(bs, &request, NULL, 0);
340 nbd_co_receive_reply(client, &request, &reply, NULL, 0);
342 nbd_coroutine_end(client, &request);
347 void nbd_client_detach_aio_context(BlockDriverState *bs)
349 aio_set_fd_handler(bdrv_get_aio_context(bs),
350 nbd_get_client_session(bs)->sock, NULL, NULL, NULL);
353 void nbd_client_attach_aio_context(BlockDriverState *bs,
354 AioContext *new_context)
356 aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sock,
357 nbd_reply_ready, NULL, bs);
360 void nbd_client_close(BlockDriverState *bs)
362 NbdClientSession *client = nbd_get_client_session(bs);
363 struct nbd_request request = {
364 .type = NBD_CMD_DISC,
369 if (client->sock == -1) {
373 nbd_send_request(client->sock, &request);
375 nbd_teardown_connection(bs);
378 int nbd_client_init(BlockDriverState *bs, int sock, const char *export,
381 NbdClientSession *client = nbd_get_client_session(bs);
385 logout("session init %s\n", export);
386 qemu_set_block(sock);
387 ret = nbd_receive_negotiate(sock, export,
388 &client->nbdflags, &client->size,
389 &client->blocksize, errp);
391 logout("Failed to negotiate with the NBD server\n");
396 qemu_co_mutex_init(&client->send_mutex);
397 qemu_co_mutex_init(&client->free_sema);
400 /* Now that we're connected, set the socket to be non-blocking and
401 * kick the reply mechanism. */
402 qemu_set_nonblock(sock);
403 nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
405 logout("Established connection with NBD server\n");