]>
Commit | Line | Data |
---|---|---|
2302c1ca MAL |
1 | /* |
2 | * QEMU Block driver for NBD | |
3 | * | |
4 | * Copyright (C) 2008 Bull S.A.S. | |
5 | * Author: Laurent Vivier <[email protected]> | |
6 | * | |
7 | * Some parts: | |
8 | * Copyright (C) 2007 Anthony Liguori <[email protected]> | |
9 | * | |
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: | |
16 | * | |
17 | * The above copyright notice and this permission notice shall be included in | |
18 | * all copies or substantial portions of the Software. | |
19 | * | |
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 | |
26 | * THE SOFTWARE. | |
27 | */ | |
28 | ||
29 | #include "nbd-client.h" | |
30 | #include "qemu/sockets.h" | |
31 | ||
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)) | |
34 | ||
69152c09 MAL |
35 | static void nbd_recv_coroutines_enter_all(NbdClientSession *s) |
36 | { | |
37 | int i; | |
38 | ||
39 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
40 | if (s->recv_coroutine[i]) { | |
41 | qemu_coroutine_enter(s->recv_coroutine[i], NULL); | |
42 | } | |
43 | } | |
44 | } | |
45 | ||
4a41a2d6 SH |
46 | static void nbd_teardown_connection(NbdClientSession *client) |
47 | { | |
48 | /* finish any pending coroutines */ | |
49 | shutdown(client->sock, 2); | |
50 | nbd_recv_coroutines_enter_all(client); | |
51 | ||
69447cd8 | 52 | nbd_client_session_detach_aio_context(client); |
4a41a2d6 SH |
53 | closesocket(client->sock); |
54 | client->sock = -1; | |
55 | } | |
56 | ||
2302c1ca MAL |
57 | static void nbd_reply_ready(void *opaque) |
58 | { | |
59 | NbdClientSession *s = opaque; | |
60 | uint64_t i; | |
61 | int ret; | |
62 | ||
63 | if (s->reply.handle == 0) { | |
64 | /* No reply already in flight. Fetch a header. It is possible | |
65 | * that another thread has done the same thing in parallel, so | |
66 | * the socket is not readable anymore. | |
67 | */ | |
68 | ret = nbd_receive_reply(s->sock, &s->reply); | |
69 | if (ret == -EAGAIN) { | |
70 | return; | |
71 | } | |
72 | if (ret < 0) { | |
73 | s->reply.handle = 0; | |
74 | goto fail; | |
75 | } | |
76 | } | |
77 | ||
78 | /* There's no need for a mutex on the receive side, because the | |
79 | * handler acts as a synchronization point and ensures that only | |
80 | * one coroutine is called until the reply finishes. */ | |
81 | i = HANDLE_TO_INDEX(s, s->reply.handle); | |
82 | if (i >= MAX_NBD_REQUESTS) { | |
83 | goto fail; | |
84 | } | |
85 | ||
86 | if (s->recv_coroutine[i]) { | |
87 | qemu_coroutine_enter(s->recv_coroutine[i], NULL); | |
88 | return; | |
89 | } | |
90 | ||
91 | fail: | |
4a41a2d6 | 92 | nbd_teardown_connection(s); |
2302c1ca MAL |
93 | } |
94 | ||
95 | static void nbd_restart_write(void *opaque) | |
96 | { | |
97 | NbdClientSession *s = opaque; | |
98 | ||
99 | qemu_coroutine_enter(s->send_coroutine, NULL); | |
100 | } | |
101 | ||
102 | static int nbd_co_send_request(NbdClientSession *s, | |
103 | struct nbd_request *request, | |
104 | QEMUIOVector *qiov, int offset) | |
105 | { | |
69447cd8 | 106 | AioContext *aio_context; |
2302c1ca MAL |
107 | int rc, ret; |
108 | ||
109 | qemu_co_mutex_lock(&s->send_mutex); | |
110 | s->send_coroutine = qemu_coroutine_self(); | |
69447cd8 SH |
111 | aio_context = bdrv_get_aio_context(s->bs); |
112 | aio_set_fd_handler(aio_context, s->sock, | |
113 | nbd_reply_ready, nbd_restart_write, s); | |
2302c1ca MAL |
114 | if (qiov) { |
115 | if (!s->is_unix) { | |
116 | socket_set_cork(s->sock, 1); | |
117 | } | |
118 | rc = nbd_send_request(s->sock, request); | |
119 | if (rc >= 0) { | |
120 | ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov, | |
121 | offset, request->len); | |
122 | if (ret != request->len) { | |
123 | rc = -EIO; | |
124 | } | |
125 | } | |
126 | if (!s->is_unix) { | |
127 | socket_set_cork(s->sock, 0); | |
128 | } | |
129 | } else { | |
130 | rc = nbd_send_request(s->sock, request); | |
131 | } | |
69447cd8 | 132 | aio_set_fd_handler(aio_context, s->sock, nbd_reply_ready, NULL, s); |
2302c1ca MAL |
133 | s->send_coroutine = NULL; |
134 | qemu_co_mutex_unlock(&s->send_mutex); | |
135 | return rc; | |
136 | } | |
137 | ||
138 | static void nbd_co_receive_reply(NbdClientSession *s, | |
139 | struct nbd_request *request, struct nbd_reply *reply, | |
140 | QEMUIOVector *qiov, int offset) | |
141 | { | |
142 | int ret; | |
143 | ||
144 | /* Wait until we're woken up by the read handler. TODO: perhaps | |
145 | * peek at the next reply and avoid yielding if it's ours? */ | |
146 | qemu_coroutine_yield(); | |
147 | *reply = s->reply; | |
148 | if (reply->handle != request->handle) { | |
149 | reply->error = EIO; | |
150 | } else { | |
151 | if (qiov && reply->error == 0) { | |
152 | ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov, | |
153 | offset, request->len); | |
154 | if (ret != request->len) { | |
155 | reply->error = EIO; | |
156 | } | |
157 | } | |
158 | ||
159 | /* Tell the read handler to read another header. */ | |
160 | s->reply.handle = 0; | |
161 | } | |
162 | } | |
163 | ||
164 | static void nbd_coroutine_start(NbdClientSession *s, | |
165 | struct nbd_request *request) | |
166 | { | |
167 | int i; | |
168 | ||
169 | /* Poor man semaphore. The free_sema is locked when no other request | |
170 | * can be accepted, and unlocked after receiving one reply. */ | |
171 | if (s->in_flight >= MAX_NBD_REQUESTS - 1) { | |
172 | qemu_co_mutex_lock(&s->free_sema); | |
173 | assert(s->in_flight < MAX_NBD_REQUESTS); | |
174 | } | |
175 | s->in_flight++; | |
176 | ||
177 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
178 | if (s->recv_coroutine[i] == NULL) { | |
179 | s->recv_coroutine[i] = qemu_coroutine_self(); | |
180 | break; | |
181 | } | |
182 | } | |
183 | ||
184 | assert(i < MAX_NBD_REQUESTS); | |
185 | request->handle = INDEX_TO_HANDLE(s, i); | |
186 | } | |
187 | ||
188 | static void nbd_coroutine_end(NbdClientSession *s, | |
189 | struct nbd_request *request) | |
190 | { | |
191 | int i = HANDLE_TO_INDEX(s, request->handle); | |
192 | s->recv_coroutine[i] = NULL; | |
193 | if (s->in_flight-- == MAX_NBD_REQUESTS) { | |
194 | qemu_co_mutex_unlock(&s->free_sema); | |
195 | } | |
196 | } | |
197 | ||
198 | static int nbd_co_readv_1(NbdClientSession *client, int64_t sector_num, | |
199 | int nb_sectors, QEMUIOVector *qiov, | |
200 | int offset) | |
201 | { | |
b1b27b64 | 202 | struct nbd_request request = { .type = NBD_CMD_READ }; |
2302c1ca MAL |
203 | struct nbd_reply reply; |
204 | ssize_t ret; | |
205 | ||
2302c1ca MAL |
206 | request.from = sector_num * 512; |
207 | request.len = nb_sectors * 512; | |
208 | ||
209 | nbd_coroutine_start(client, &request); | |
210 | ret = nbd_co_send_request(client, &request, NULL, 0); | |
211 | if (ret < 0) { | |
212 | reply.error = -ret; | |
213 | } else { | |
214 | nbd_co_receive_reply(client, &request, &reply, qiov, offset); | |
215 | } | |
216 | nbd_coroutine_end(client, &request); | |
217 | return -reply.error; | |
218 | ||
219 | } | |
220 | ||
221 | static int nbd_co_writev_1(NbdClientSession *client, int64_t sector_num, | |
222 | int nb_sectors, QEMUIOVector *qiov, | |
223 | int offset) | |
224 | { | |
b1b27b64 | 225 | struct nbd_request request = { .type = NBD_CMD_WRITE }; |
2302c1ca MAL |
226 | struct nbd_reply reply; |
227 | ssize_t ret; | |
228 | ||
2302c1ca MAL |
229 | if (!bdrv_enable_write_cache(client->bs) && |
230 | (client->nbdflags & NBD_FLAG_SEND_FUA)) { | |
231 | request.type |= NBD_CMD_FLAG_FUA; | |
232 | } | |
233 | ||
234 | request.from = sector_num * 512; | |
235 | request.len = nb_sectors * 512; | |
236 | ||
237 | nbd_coroutine_start(client, &request); | |
238 | ret = nbd_co_send_request(client, &request, qiov, offset); | |
239 | if (ret < 0) { | |
240 | reply.error = -ret; | |
241 | } else { | |
242 | nbd_co_receive_reply(client, &request, &reply, NULL, 0); | |
243 | } | |
244 | nbd_coroutine_end(client, &request); | |
245 | return -reply.error; | |
246 | } | |
247 | ||
248 | /* qemu-nbd has a limit of slightly less than 1M per request. Try to | |
249 | * remain aligned to 4K. */ | |
250 | #define NBD_MAX_SECTORS 2040 | |
251 | ||
252 | int nbd_client_session_co_readv(NbdClientSession *client, int64_t sector_num, | |
253 | int nb_sectors, QEMUIOVector *qiov) | |
254 | { | |
255 | int offset = 0; | |
256 | int ret; | |
257 | while (nb_sectors > NBD_MAX_SECTORS) { | |
258 | ret = nbd_co_readv_1(client, sector_num, | |
259 | NBD_MAX_SECTORS, qiov, offset); | |
260 | if (ret < 0) { | |
261 | return ret; | |
262 | } | |
263 | offset += NBD_MAX_SECTORS * 512; | |
264 | sector_num += NBD_MAX_SECTORS; | |
265 | nb_sectors -= NBD_MAX_SECTORS; | |
266 | } | |
267 | return nbd_co_readv_1(client, sector_num, nb_sectors, qiov, offset); | |
268 | } | |
269 | ||
270 | int nbd_client_session_co_writev(NbdClientSession *client, int64_t sector_num, | |
271 | int nb_sectors, QEMUIOVector *qiov) | |
272 | { | |
273 | int offset = 0; | |
274 | int ret; | |
275 | while (nb_sectors > NBD_MAX_SECTORS) { | |
276 | ret = nbd_co_writev_1(client, sector_num, | |
277 | NBD_MAX_SECTORS, qiov, offset); | |
278 | if (ret < 0) { | |
279 | return ret; | |
280 | } | |
281 | offset += NBD_MAX_SECTORS * 512; | |
282 | sector_num += NBD_MAX_SECTORS; | |
283 | nb_sectors -= NBD_MAX_SECTORS; | |
284 | } | |
285 | return nbd_co_writev_1(client, sector_num, nb_sectors, qiov, offset); | |
286 | } | |
287 | ||
288 | int nbd_client_session_co_flush(NbdClientSession *client) | |
289 | { | |
b1b27b64 | 290 | struct nbd_request request = { .type = NBD_CMD_FLUSH }; |
2302c1ca MAL |
291 | struct nbd_reply reply; |
292 | ssize_t ret; | |
293 | ||
294 | if (!(client->nbdflags & NBD_FLAG_SEND_FLUSH)) { | |
295 | return 0; | |
296 | } | |
297 | ||
2302c1ca MAL |
298 | if (client->nbdflags & NBD_FLAG_SEND_FUA) { |
299 | request.type |= NBD_CMD_FLAG_FUA; | |
300 | } | |
301 | ||
302 | request.from = 0; | |
303 | request.len = 0; | |
304 | ||
305 | nbd_coroutine_start(client, &request); | |
306 | ret = nbd_co_send_request(client, &request, NULL, 0); | |
307 | if (ret < 0) { | |
308 | reply.error = -ret; | |
309 | } else { | |
310 | nbd_co_receive_reply(client, &request, &reply, NULL, 0); | |
311 | } | |
312 | nbd_coroutine_end(client, &request); | |
313 | return -reply.error; | |
314 | } | |
315 | ||
316 | int nbd_client_session_co_discard(NbdClientSession *client, int64_t sector_num, | |
317 | int nb_sectors) | |
318 | { | |
b1b27b64 | 319 | struct nbd_request request = { .type = NBD_CMD_TRIM }; |
2302c1ca MAL |
320 | struct nbd_reply reply; |
321 | ssize_t ret; | |
322 | ||
323 | if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) { | |
324 | return 0; | |
325 | } | |
2302c1ca MAL |
326 | request.from = sector_num * 512; |
327 | request.len = nb_sectors * 512; | |
328 | ||
329 | nbd_coroutine_start(client, &request); | |
330 | ret = nbd_co_send_request(client, &request, NULL, 0); | |
331 | if (ret < 0) { | |
332 | reply.error = -ret; | |
333 | } else { | |
334 | nbd_co_receive_reply(client, &request, &reply, NULL, 0); | |
335 | } | |
336 | nbd_coroutine_end(client, &request); | |
337 | return -reply.error; | |
338 | ||
339 | } | |
340 | ||
69447cd8 SH |
341 | void nbd_client_session_detach_aio_context(NbdClientSession *client) |
342 | { | |
343 | aio_set_fd_handler(bdrv_get_aio_context(client->bs), client->sock, | |
344 | NULL, NULL, NULL); | |
345 | } | |
346 | ||
347 | void nbd_client_session_attach_aio_context(NbdClientSession *client, | |
348 | AioContext *new_context) | |
349 | { | |
350 | aio_set_fd_handler(new_context, client->sock, | |
351 | nbd_reply_ready, NULL, client); | |
352 | } | |
353 | ||
4a41a2d6 | 354 | void nbd_client_session_close(NbdClientSession *client) |
2302c1ca | 355 | { |
b1b27b64 MAL |
356 | struct nbd_request request = { |
357 | .type = NBD_CMD_DISC, | |
358 | .from = 0, | |
359 | .len = 0 | |
360 | }; | |
2302c1ca | 361 | |
5ad283eb MAL |
362 | if (!client->bs) { |
363 | return; | |
364 | } | |
4a41a2d6 SH |
365 | if (client->sock == -1) { |
366 | return; | |
367 | } | |
368 | ||
369 | nbd_send_request(client->sock, &request); | |
5ad283eb | 370 | |
2302c1ca | 371 | nbd_teardown_connection(client); |
5ad283eb | 372 | client->bs = NULL; |
2302c1ca MAL |
373 | } |
374 | ||
e2bc625f MAL |
375 | int nbd_client_session_init(NbdClientSession *client, BlockDriverState *bs, |
376 | int sock, const char *export) | |
2302c1ca MAL |
377 | { |
378 | int ret; | |
379 | ||
380 | /* NBD handshake */ | |
e2bc625f | 381 | logout("session init %s\n", export); |
e53a18e4 | 382 | qemu_set_block(sock); |
e2bc625f | 383 | ret = nbd_receive_negotiate(sock, export, |
2302c1ca MAL |
384 | &client->nbdflags, &client->size, |
385 | &client->blocksize); | |
386 | if (ret < 0) { | |
387 | logout("Failed to negotiate with the NBD server\n"); | |
388 | closesocket(sock); | |
389 | return ret; | |
390 | } | |
391 | ||
392 | qemu_co_mutex_init(&client->send_mutex); | |
393 | qemu_co_mutex_init(&client->free_sema); | |
394 | client->bs = bs; | |
395 | client->sock = sock; | |
396 | ||
397 | /* Now that we're connected, set the socket to be non-blocking and | |
398 | * kick the reply mechanism. */ | |
399 | qemu_set_nonblock(sock); | |
69447cd8 | 400 | nbd_client_session_attach_aio_context(client, bdrv_get_aio_context(bs)); |
2302c1ca MAL |
401 | |
402 | logout("Established connection with NBD server\n"); | |
403 | return 0; | |
404 | } |