]> git.proxmox.com Git - mirror_qemu.git/blob - block/nfs.c
block/nfs: cache allocated filesize for read-only files
[mirror_qemu.git] / block / nfs.c
1 /*
2 * QEMU Block driver for native access to files on NFS shares
3 *
4 * Copyright (c) 2014 Peter Lieven <pl@kamp.de>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "config-host.h"
26
27 #include <poll.h>
28 #include "qemu-common.h"
29 #include "qemu/config-file.h"
30 #include "qemu/error-report.h"
31 #include "block/block_int.h"
32 #include "trace.h"
33 #include "qemu/iov.h"
34 #include "qemu/uri.h"
35 #include "sysemu/sysemu.h"
36 #include <nfsc/libnfs.h>
37
38 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
39
40 typedef struct NFSClient {
41 struct nfs_context *context;
42 struct nfsfh *fh;
43 int events;
44 bool has_zero_init;
45 AioContext *aio_context;
46 blkcnt_t st_blocks;
47 } NFSClient;
48
49 typedef struct NFSRPC {
50 int ret;
51 int complete;
52 QEMUIOVector *iov;
53 struct stat *st;
54 Coroutine *co;
55 QEMUBH *bh;
56 NFSClient *client;
57 } NFSRPC;
58
59 static void nfs_process_read(void *arg);
60 static void nfs_process_write(void *arg);
61
62 static void nfs_set_events(NFSClient *client)
63 {
64 int ev = nfs_which_events(client->context);
65 if (ev != client->events) {
66 aio_set_fd_handler(client->aio_context,
67 nfs_get_fd(client->context),
68 (ev & POLLIN) ? nfs_process_read : NULL,
69 (ev & POLLOUT) ? nfs_process_write : NULL,
70 client);
71
72 }
73 client->events = ev;
74 }
75
76 static void nfs_process_read(void *arg)
77 {
78 NFSClient *client = arg;
79 nfs_service(client->context, POLLIN);
80 nfs_set_events(client);
81 }
82
83 static void nfs_process_write(void *arg)
84 {
85 NFSClient *client = arg;
86 nfs_service(client->context, POLLOUT);
87 nfs_set_events(client);
88 }
89
90 static void nfs_co_init_task(NFSClient *client, NFSRPC *task)
91 {
92 *task = (NFSRPC) {
93 .co = qemu_coroutine_self(),
94 .client = client,
95 };
96 }
97
98 static void nfs_co_generic_bh_cb(void *opaque)
99 {
100 NFSRPC *task = opaque;
101 task->complete = 1;
102 qemu_bh_delete(task->bh);
103 qemu_coroutine_enter(task->co, NULL);
104 }
105
106 static void
107 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
108 void *private_data)
109 {
110 NFSRPC *task = private_data;
111 task->ret = ret;
112 if (task->ret > 0 && task->iov) {
113 if (task->ret <= task->iov->size) {
114 qemu_iovec_from_buf(task->iov, 0, data, task->ret);
115 } else {
116 task->ret = -EIO;
117 }
118 }
119 if (task->ret == 0 && task->st) {
120 memcpy(task->st, data, sizeof(struct stat));
121 }
122 if (task->ret < 0) {
123 error_report("NFS Error: %s", nfs_get_error(nfs));
124 }
125 if (task->co) {
126 task->bh = aio_bh_new(task->client->aio_context,
127 nfs_co_generic_bh_cb, task);
128 qemu_bh_schedule(task->bh);
129 } else {
130 task->complete = 1;
131 }
132 }
133
134 static int coroutine_fn nfs_co_readv(BlockDriverState *bs,
135 int64_t sector_num, int nb_sectors,
136 QEMUIOVector *iov)
137 {
138 NFSClient *client = bs->opaque;
139 NFSRPC task;
140
141 nfs_co_init_task(client, &task);
142 task.iov = iov;
143
144 if (nfs_pread_async(client->context, client->fh,
145 sector_num * BDRV_SECTOR_SIZE,
146 nb_sectors * BDRV_SECTOR_SIZE,
147 nfs_co_generic_cb, &task) != 0) {
148 return -ENOMEM;
149 }
150
151 while (!task.complete) {
152 nfs_set_events(client);
153 qemu_coroutine_yield();
154 }
155
156 if (task.ret < 0) {
157 return task.ret;
158 }
159
160 /* zero pad short reads */
161 if (task.ret < iov->size) {
162 qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
163 }
164
165 return 0;
166 }
167
168 static int coroutine_fn nfs_co_writev(BlockDriverState *bs,
169 int64_t sector_num, int nb_sectors,
170 QEMUIOVector *iov)
171 {
172 NFSClient *client = bs->opaque;
173 NFSRPC task;
174 char *buf = NULL;
175
176 nfs_co_init_task(client, &task);
177
178 buf = g_try_malloc(nb_sectors * BDRV_SECTOR_SIZE);
179 if (nb_sectors && buf == NULL) {
180 return -ENOMEM;
181 }
182
183 qemu_iovec_to_buf(iov, 0, buf, nb_sectors * BDRV_SECTOR_SIZE);
184
185 if (nfs_pwrite_async(client->context, client->fh,
186 sector_num * BDRV_SECTOR_SIZE,
187 nb_sectors * BDRV_SECTOR_SIZE,
188 buf, nfs_co_generic_cb, &task) != 0) {
189 g_free(buf);
190 return -ENOMEM;
191 }
192
193 while (!task.complete) {
194 nfs_set_events(client);
195 qemu_coroutine_yield();
196 }
197
198 g_free(buf);
199
200 if (task.ret != nb_sectors * BDRV_SECTOR_SIZE) {
201 return task.ret < 0 ? task.ret : -EIO;
202 }
203
204 return 0;
205 }
206
207 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
208 {
209 NFSClient *client = bs->opaque;
210 NFSRPC task;
211
212 nfs_co_init_task(client, &task);
213
214 if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
215 &task) != 0) {
216 return -ENOMEM;
217 }
218
219 while (!task.complete) {
220 nfs_set_events(client);
221 qemu_coroutine_yield();
222 }
223
224 return task.ret;
225 }
226
227 /* TODO Convert to fine grained options */
228 static QemuOptsList runtime_opts = {
229 .name = "nfs",
230 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
231 .desc = {
232 {
233 .name = "filename",
234 .type = QEMU_OPT_STRING,
235 .help = "URL to the NFS file",
236 },
237 { /* end of list */ }
238 },
239 };
240
241 static void nfs_detach_aio_context(BlockDriverState *bs)
242 {
243 NFSClient *client = bs->opaque;
244
245 aio_set_fd_handler(client->aio_context,
246 nfs_get_fd(client->context),
247 NULL, NULL, NULL);
248 client->events = 0;
249 }
250
251 static void nfs_attach_aio_context(BlockDriverState *bs,
252 AioContext *new_context)
253 {
254 NFSClient *client = bs->opaque;
255
256 client->aio_context = new_context;
257 nfs_set_events(client);
258 }
259
260 static void nfs_client_close(NFSClient *client)
261 {
262 if (client->context) {
263 if (client->fh) {
264 nfs_close(client->context, client->fh);
265 }
266 aio_set_fd_handler(client->aio_context,
267 nfs_get_fd(client->context),
268 NULL, NULL, NULL);
269 nfs_destroy_context(client->context);
270 }
271 memset(client, 0, sizeof(NFSClient));
272 }
273
274 static void nfs_file_close(BlockDriverState *bs)
275 {
276 NFSClient *client = bs->opaque;
277 nfs_client_close(client);
278 }
279
280 static int64_t nfs_client_open(NFSClient *client, const char *filename,
281 int flags, Error **errp)
282 {
283 int ret = -EINVAL, i;
284 struct stat st;
285 URI *uri;
286 QueryParams *qp = NULL;
287 char *file = NULL, *strp = NULL;
288
289 uri = uri_parse(filename);
290 if (!uri) {
291 error_setg(errp, "Invalid URL specified");
292 goto fail;
293 }
294 if (!uri->server) {
295 error_setg(errp, "Invalid URL specified");
296 goto fail;
297 }
298 strp = strrchr(uri->path, '/');
299 if (strp == NULL) {
300 error_setg(errp, "Invalid URL specified");
301 goto fail;
302 }
303 file = g_strdup(strp);
304 *strp = 0;
305
306 client->context = nfs_init_context();
307 if (client->context == NULL) {
308 error_setg(errp, "Failed to init NFS context");
309 goto fail;
310 }
311
312 qp = query_params_parse(uri->query);
313 for (i = 0; i < qp->n; i++) {
314 unsigned long long val;
315 if (!qp->p[i].value) {
316 error_setg(errp, "Value for NFS parameter expected: %s",
317 qp->p[i].name);
318 goto fail;
319 }
320 if (parse_uint_full(qp->p[i].value, &val, 0)) {
321 error_setg(errp, "Illegal value for NFS parameter: %s",
322 qp->p[i].name);
323 goto fail;
324 }
325 if (!strcmp(qp->p[i].name, "uid")) {
326 nfs_set_uid(client->context, val);
327 } else if (!strcmp(qp->p[i].name, "gid")) {
328 nfs_set_gid(client->context, val);
329 } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
330 nfs_set_tcp_syncnt(client->context, val);
331 #ifdef LIBNFS_FEATURE_READAHEAD
332 } else if (!strcmp(qp->p[i].name, "readahead")) {
333 if (val > QEMU_NFS_MAX_READAHEAD_SIZE) {
334 error_report("NFS Warning: Truncating NFS readahead"
335 " size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
336 val = QEMU_NFS_MAX_READAHEAD_SIZE;
337 }
338 nfs_set_readahead(client->context, val);
339 #endif
340 } else {
341 error_setg(errp, "Unknown NFS parameter name: %s",
342 qp->p[i].name);
343 goto fail;
344 }
345 }
346
347 ret = nfs_mount(client->context, uri->server, uri->path);
348 if (ret < 0) {
349 error_setg(errp, "Failed to mount nfs share: %s",
350 nfs_get_error(client->context));
351 goto fail;
352 }
353
354 if (flags & O_CREAT) {
355 ret = nfs_creat(client->context, file, 0600, &client->fh);
356 if (ret < 0) {
357 error_setg(errp, "Failed to create file: %s",
358 nfs_get_error(client->context));
359 goto fail;
360 }
361 } else {
362 ret = nfs_open(client->context, file, flags, &client->fh);
363 if (ret < 0) {
364 error_setg(errp, "Failed to open file : %s",
365 nfs_get_error(client->context));
366 goto fail;
367 }
368 }
369
370 ret = nfs_fstat(client->context, client->fh, &st);
371 if (ret < 0) {
372 error_setg(errp, "Failed to fstat file: %s",
373 nfs_get_error(client->context));
374 goto fail;
375 }
376
377 ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
378 client->st_blocks = st.st_blocks;
379 client->has_zero_init = S_ISREG(st.st_mode);
380 goto out;
381 fail:
382 nfs_client_close(client);
383 out:
384 if (qp) {
385 query_params_free(qp);
386 }
387 uri_free(uri);
388 g_free(file);
389 return ret;
390 }
391
392 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
393 Error **errp) {
394 NFSClient *client = bs->opaque;
395 int64_t ret;
396 QemuOpts *opts;
397 Error *local_err = NULL;
398
399 client->aio_context = bdrv_get_aio_context(bs);
400
401 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
402 qemu_opts_absorb_qdict(opts, options, &local_err);
403 if (local_err) {
404 error_propagate(errp, local_err);
405 ret = -EINVAL;
406 goto out;
407 }
408 ret = nfs_client_open(client, qemu_opt_get(opts, "filename"),
409 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
410 errp);
411 if (ret < 0) {
412 goto out;
413 }
414 bs->total_sectors = ret;
415 ret = 0;
416 out:
417 qemu_opts_del(opts);
418 return ret;
419 }
420
421 static QemuOptsList nfs_create_opts = {
422 .name = "nfs-create-opts",
423 .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
424 .desc = {
425 {
426 .name = BLOCK_OPT_SIZE,
427 .type = QEMU_OPT_SIZE,
428 .help = "Virtual disk size"
429 },
430 { /* end of list */ }
431 }
432 };
433
434 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
435 {
436 int ret = 0;
437 int64_t total_size = 0;
438 NFSClient *client = g_new0(NFSClient, 1);
439
440 client->aio_context = qemu_get_aio_context();
441
442 /* Read out options */
443 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
444 BDRV_SECTOR_SIZE);
445
446 ret = nfs_client_open(client, url, O_CREAT, errp);
447 if (ret < 0) {
448 goto out;
449 }
450 ret = nfs_ftruncate(client->context, client->fh, total_size);
451 nfs_client_close(client);
452 out:
453 g_free(client);
454 return ret;
455 }
456
457 static int nfs_has_zero_init(BlockDriverState *bs)
458 {
459 NFSClient *client = bs->opaque;
460 return client->has_zero_init;
461 }
462
463 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
464 {
465 NFSClient *client = bs->opaque;
466 NFSRPC task = {0};
467 struct stat st;
468
469 if (bdrv_is_read_only(bs) &&
470 !(bs->open_flags & BDRV_O_NOCACHE)) {
471 return client->st_blocks * 512;
472 }
473
474 task.st = &st;
475 if (nfs_fstat_async(client->context, client->fh, nfs_co_generic_cb,
476 &task) != 0) {
477 return -ENOMEM;
478 }
479
480 while (!task.complete) {
481 nfs_set_events(client);
482 aio_poll(client->aio_context, true);
483 }
484
485 return (task.ret < 0 ? task.ret : st.st_blocks * 512);
486 }
487
488 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
489 {
490 NFSClient *client = bs->opaque;
491 return nfs_ftruncate(client->context, client->fh, offset);
492 }
493
494 /* Note that this will not re-establish a connection with the NFS server
495 * - it is effectively a NOP. */
496 static int nfs_reopen_prepare(BDRVReopenState *state,
497 BlockReopenQueue *queue, Error **errp)
498 {
499 NFSClient *client = state->bs->opaque;
500 struct stat st;
501 int ret = 0;
502
503 if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
504 error_setg(errp, "Cannot open a read-only mount as read-write");
505 return -EACCES;
506 }
507
508 /* Update cache for read-only reopens */
509 if (!(state->flags & BDRV_O_RDWR)) {
510 ret = nfs_fstat(client->context, client->fh, &st);
511 if (ret < 0) {
512 error_setg(errp, "Failed to fstat file: %s",
513 nfs_get_error(client->context));
514 return ret;
515 }
516 client->st_blocks = st.st_blocks;
517 }
518
519 return 0;
520 }
521
522 static BlockDriver bdrv_nfs = {
523 .format_name = "nfs",
524 .protocol_name = "nfs",
525
526 .instance_size = sizeof(NFSClient),
527 .bdrv_needs_filename = true,
528 .create_opts = &nfs_create_opts,
529
530 .bdrv_has_zero_init = nfs_has_zero_init,
531 .bdrv_get_allocated_file_size = nfs_get_allocated_file_size,
532 .bdrv_truncate = nfs_file_truncate,
533
534 .bdrv_file_open = nfs_file_open,
535 .bdrv_close = nfs_file_close,
536 .bdrv_create = nfs_file_create,
537 .bdrv_reopen_prepare = nfs_reopen_prepare,
538
539 .bdrv_co_readv = nfs_co_readv,
540 .bdrv_co_writev = nfs_co_writev,
541 .bdrv_co_flush_to_disk = nfs_co_flush,
542
543 .bdrv_detach_aio_context = nfs_detach_aio_context,
544 .bdrv_attach_aio_context = nfs_attach_aio_context,
545 };
546
547 static void nfs_block_init(void)
548 {
549 bdrv_register(&bdrv_nfs);
550 }
551
552 block_init(nfs_block_init);