nbd-client: avoid spurious qio_channel_yield() re-entry
[qemu/ericb.git] / block / nbd-client.c
blob25bcaa2346c9c4e731e7771fe9750ac4eeee37b0
1 /*
2 * QEMU Block driver for NBD
4 * Copyright (C) 2016 Red Hat, Inc.
5 * Copyright (C) 2008 Bull S.A.S.
6 * Author: Laurent Vivier <Laurent.Vivier@bull.net>
8 * Some parts:
9 * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
11 * Permission is hereby granted, free of charge, to any person obtaining a copy
12 * of this software and associated documentation files (the "Software"), to deal
13 * in the Software without restriction, including without limitation the rights
14 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
15 * copies of the Software, and to permit persons to whom the Software is
16 * furnished to do so, subject to the following conditions:
18 * The above copyright notice and this permission notice shall be included in
19 * all copies or substantial portions of the Software.
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
22 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
23 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
24 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
25 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
26 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 * THE SOFTWARE.
30 #include "qemu/osdep.h"
31 #include "qapi/error.h"
32 #include "nbd-client.h"
34 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
35 #define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs))
37 static void nbd_recv_coroutines_enter_all(NBDClientSession *s)
39 int i;
41 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
42 NBDClientRequest *req = &s->requests[i];
44 if (req->coroutine && req->receiving) {
45 aio_co_wake(req->coroutine);
50 static void nbd_teardown_connection(BlockDriverState *bs)
52 NBDClientSession *client = nbd_get_client_session(bs);
54 if (!client->ioc) { /* Already closed */
55 return;
58 /* finish any pending coroutines */
59 qio_channel_shutdown(client->ioc,
60 QIO_CHANNEL_SHUTDOWN_BOTH,
61 NULL);
62 BDRV_POLL_WHILE(bs, client->read_reply_co);
64 nbd_client_detach_aio_context(bs);
65 object_unref(OBJECT(client->sioc));
66 client->sioc = NULL;
67 object_unref(OBJECT(client->ioc));
68 client->ioc = NULL;
71 static coroutine_fn void nbd_read_reply_entry(void *opaque)
73 NBDClientSession *s = opaque;
74 uint64_t i;
75 int ret = 0;
76 Error *local_err = NULL;
78 while (!s->quit) {
79 assert(s->reply.handle == 0);
80 ret = nbd_receive_reply(s->ioc, &s->reply, &local_err);
81 if (ret < 0) {
82 error_report_err(local_err);
84 if (ret <= 0) {
85 break;
88 /* There's no need for a mutex on the receive side, because the
89 * handler acts as a synchronization point and ensures that only
90 * one coroutine is called until the reply finishes.
92 i = HANDLE_TO_INDEX(s, s->reply.handle);
93 if (i >= MAX_NBD_REQUESTS ||
94 !s->requests[i].coroutine ||
95 !s->requests[i].receiving) {
96 break;
99 /* We're woken up again by the request itself. Note that there
100 * is no race between yielding and reentering read_reply_co. This
101 * is because:
103 * - if the request runs on the same AioContext, it is only
104 * entered after we yield
106 * - if the request runs on a different AioContext, reentering
107 * read_reply_co happens through a bottom half, which can only
108 * run after we yield.
110 aio_co_wake(s->requests[i].coroutine);
111 qemu_coroutine_yield();
114 s->quit = true;
115 nbd_recv_coroutines_enter_all(s);
116 s->read_reply_co = NULL;
119 static int nbd_co_send_request(BlockDriverState *bs,
120 NBDRequest *request,
121 QEMUIOVector *qiov)
123 NBDClientSession *s = nbd_get_client_session(bs);
124 int rc, ret, i;
126 qemu_co_mutex_lock(&s->send_mutex);
127 while (s->in_flight == MAX_NBD_REQUESTS) {
128 qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
130 s->in_flight++;
132 for (i = 0; i < MAX_NBD_REQUESTS; i++) {
133 if (s->requests[i].coroutine == NULL) {
134 break;
138 g_assert(qemu_in_coroutine());
139 assert(i < MAX_NBD_REQUESTS);
141 s->requests[i].coroutine = qemu_coroutine_self();
142 s->requests[i].receiving = false;
144 request->handle = INDEX_TO_HANDLE(s, i);
146 if (s->quit) {
147 qemu_co_mutex_unlock(&s->send_mutex);
148 return -EIO;
150 if (!s->ioc) {
151 qemu_co_mutex_unlock(&s->send_mutex);
152 return -EPIPE;
155 if (qiov) {
156 qio_channel_set_cork(s->ioc, true);
157 rc = nbd_send_request(s->ioc, request);
158 if (rc >= 0 && !s->quit) {
159 ret = nbd_rwv(s->ioc, qiov->iov, qiov->niov, request->len, false,
160 NULL);
161 if (ret != request->len) {
162 rc = -EIO;
165 qio_channel_set_cork(s->ioc, false);
166 } else {
167 rc = nbd_send_request(s->ioc, request);
169 if (rc < 0) {
170 s->quit = true;
172 qemu_co_mutex_unlock(&s->send_mutex);
173 return rc;
176 static void nbd_co_receive_reply(NBDClientSession *s,
177 NBDRequest *request,
178 NBDReply *reply,
179 QEMUIOVector *qiov)
181 int i = HANDLE_TO_INDEX(s, request->handle);
182 int ret;
184 /* Wait until we're woken up by nbd_read_reply_entry. */
185 s->requests[i].receiving = true;
186 qemu_coroutine_yield();
187 s->requests[i].receiving = false;
188 *reply = s->reply;
189 if (reply->handle != request->handle || !s->ioc || s->quit) {
190 reply->error = EIO;
191 } else {
192 if (qiov && reply->error == 0) {
193 ret = nbd_rwv(s->ioc, qiov->iov, qiov->niov, request->len, true,
194 NULL);
195 if (ret != request->len) {
196 reply->error = EIO;
197 s->quit = true;
201 /* Tell the read handler to read another header. */
202 s->reply.handle = 0;
206 static void nbd_coroutine_end(BlockDriverState *bs,
207 NBDRequest *request)
209 NBDClientSession *s = nbd_get_client_session(bs);
210 int i = HANDLE_TO_INDEX(s, request->handle);
212 s->requests[i].coroutine = NULL;
214 /* Kick the read_reply_co to get the next reply. */
215 if (s->read_reply_co) {
216 aio_co_wake(s->read_reply_co);
219 qemu_co_mutex_lock(&s->send_mutex);
220 s->in_flight--;
221 qemu_co_queue_next(&s->free_sema);
222 qemu_co_mutex_unlock(&s->send_mutex);
225 int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
226 uint64_t bytes, QEMUIOVector *qiov, int flags)
228 NBDClientSession *client = nbd_get_client_session(bs);
229 NBDRequest request = {
230 .type = NBD_CMD_READ,
231 .from = offset,
232 .len = bytes,
234 NBDReply reply;
235 ssize_t ret;
237 assert(bytes <= NBD_MAX_BUFFER_SIZE);
238 assert(!flags);
240 ret = nbd_co_send_request(bs, &request, NULL);
241 if (ret < 0) {
242 reply.error = -ret;
243 } else {
244 nbd_co_receive_reply(client, &request, &reply, qiov);
246 nbd_coroutine_end(bs, &request);
247 return -reply.error;
250 int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
251 uint64_t bytes, QEMUIOVector *qiov, int flags)
253 NBDClientSession *client = nbd_get_client_session(bs);
254 NBDRequest request = {
255 .type = NBD_CMD_WRITE,
256 .from = offset,
257 .len = bytes,
259 NBDReply reply;
260 ssize_t ret;
262 if (flags & BDRV_REQ_FUA) {
263 assert(client->info.flags & NBD_FLAG_SEND_FUA);
264 request.flags |= NBD_CMD_FLAG_FUA;
267 assert(bytes <= NBD_MAX_BUFFER_SIZE);
269 ret = nbd_co_send_request(bs, &request, qiov);
270 if (ret < 0) {
271 reply.error = -ret;
272 } else {
273 nbd_co_receive_reply(client, &request, &reply, NULL);
275 nbd_coroutine_end(bs, &request);
276 return -reply.error;
279 int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
280 int bytes, BdrvRequestFlags flags)
282 ssize_t ret;
283 NBDClientSession *client = nbd_get_client_session(bs);
284 NBDRequest request = {
285 .type = NBD_CMD_WRITE_ZEROES,
286 .from = offset,
287 .len = bytes,
289 NBDReply reply;
291 if (!(client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
292 return -ENOTSUP;
295 if (flags & BDRV_REQ_FUA) {
296 assert(client->info.flags & NBD_FLAG_SEND_FUA);
297 request.flags |= NBD_CMD_FLAG_FUA;
299 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
300 request.flags |= NBD_CMD_FLAG_NO_HOLE;
303 ret = nbd_co_send_request(bs, &request, NULL);
304 if (ret < 0) {
305 reply.error = -ret;
306 } else {
307 nbd_co_receive_reply(client, &request, &reply, NULL);
309 nbd_coroutine_end(bs, &request);
310 return -reply.error;
313 int nbd_client_co_flush(BlockDriverState *bs)
315 NBDClientSession *client = nbd_get_client_session(bs);
316 NBDRequest request = { .type = NBD_CMD_FLUSH };
317 NBDReply reply;
318 ssize_t ret;
320 if (!(client->info.flags & NBD_FLAG_SEND_FLUSH)) {
321 return 0;
324 request.from = 0;
325 request.len = 0;
327 ret = nbd_co_send_request(bs, &request, NULL);
328 if (ret < 0) {
329 reply.error = -ret;
330 } else {
331 nbd_co_receive_reply(client, &request, &reply, NULL);
333 nbd_coroutine_end(bs, &request);
334 return -reply.error;
337 int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
339 NBDClientSession *client = nbd_get_client_session(bs);
340 NBDRequest request = {
341 .type = NBD_CMD_TRIM,
342 .from = offset,
343 .len = bytes,
345 NBDReply reply;
346 ssize_t ret;
348 if (!(client->info.flags & NBD_FLAG_SEND_TRIM)) {
349 return 0;
352 ret = nbd_co_send_request(bs, &request, NULL);
353 if (ret < 0) {
354 reply.error = -ret;
355 } else {
356 nbd_co_receive_reply(client, &request, &reply, NULL);
358 nbd_coroutine_end(bs, &request);
359 return -reply.error;
363 void nbd_client_detach_aio_context(BlockDriverState *bs)
365 NBDClientSession *client = nbd_get_client_session(bs);
366 qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
369 void nbd_client_attach_aio_context(BlockDriverState *bs,
370 AioContext *new_context)
372 NBDClientSession *client = nbd_get_client_session(bs);
373 qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc), new_context);
374 aio_co_schedule(new_context, client->read_reply_co);
377 void nbd_client_close(BlockDriverState *bs)
379 NBDClientSession *client = nbd_get_client_session(bs);
380 NBDRequest request = { .type = NBD_CMD_DISC };
382 if (client->ioc == NULL) {
383 return;
386 nbd_send_request(client->ioc, &request);
388 nbd_teardown_connection(bs);
391 int nbd_client_init(BlockDriverState *bs,
392 QIOChannelSocket *sioc,
393 const char *export,
394 QCryptoTLSCreds *tlscreds,
395 const char *hostname,
396 Error **errp)
398 NBDClientSession *client = nbd_get_client_session(bs);
399 int ret;
401 /* NBD handshake */
402 logout("session init %s\n", export);
403 qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
405 client->info.request_sizes = true;
406 ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
407 tlscreds, hostname,
408 &client->ioc, &client->info, errp);
409 if (ret < 0) {
410 logout("Failed to negotiate with the NBD server\n");
411 return ret;
413 if (client->info.flags & NBD_FLAG_SEND_FUA) {
414 bs->supported_write_flags = BDRV_REQ_FUA;
415 bs->supported_zero_flags |= BDRV_REQ_FUA;
417 if (client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
418 bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
420 if (client->info.min_block > bs->bl.request_alignment) {
421 bs->bl.request_alignment = client->info.min_block;
424 qemu_co_mutex_init(&client->send_mutex);
425 qemu_co_queue_init(&client->free_sema);
426 client->sioc = sioc;
427 object_ref(OBJECT(client->sioc));
429 if (!client->ioc) {
430 client->ioc = QIO_CHANNEL(sioc);
431 object_ref(OBJECT(client->ioc));
434 /* Now that we're connected, set the socket to be non-blocking and
435 * kick the reply mechanism. */
436 qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
437 client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
438 nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
440 logout("Established connection with NBD server\n");
441 return 0;