]> git.proxmox.com Git - mirror_qemu.git/blob - block/nfs.c
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
[mirror_qemu.git] / block / nfs.c
1 /*
2 * QEMU Block driver for native access to files on NFS shares
3 *
4 * Copyright (c) 2014-2016 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 "qemu/osdep.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 "qapi/error.h"
32 #include "block/block_int.h"
33 #include "trace.h"
34 #include "qemu/iov.h"
35 #include "qemu/uri.h"
36 #include "qemu/cutils.h"
37 #include "sysemu/sysemu.h"
38 #include <nfsc/libnfs.h>
39
40 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
41 #define QEMU_NFS_MAX_PAGECACHE_SIZE (8388608 / NFS_BLKSIZE)
42 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
43
44 typedef struct NFSClient {
45 struct nfs_context *context;
46 struct nfsfh *fh;
47 int events;
48 bool has_zero_init;
49 AioContext *aio_context;
50 blkcnt_t st_blocks;
51 bool cache_used;
52 } NFSClient;
53
54 typedef struct NFSRPC {
55 int ret;
56 int complete;
57 QEMUIOVector *iov;
58 struct stat *st;
59 Coroutine *co;
60 NFSClient *client;
61 } NFSRPC;
62
63 static void nfs_process_read(void *arg);
64 static void nfs_process_write(void *arg);
65
66 static void nfs_set_events(NFSClient *client)
67 {
68 int ev = nfs_which_events(client->context);
69 if (ev != client->events) {
70 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
71 false,
72 (ev & POLLIN) ? nfs_process_read : NULL,
73 (ev & POLLOUT) ? nfs_process_write : NULL, client);
74
75 }
76 client->events = ev;
77 }
78
79 static void nfs_process_read(void *arg)
80 {
81 NFSClient *client = arg;
82 nfs_service(client->context, POLLIN);
83 nfs_set_events(client);
84 }
85
86 static void nfs_process_write(void *arg)
87 {
88 NFSClient *client = arg;
89 nfs_service(client->context, POLLOUT);
90 nfs_set_events(client);
91 }
92
93 static void nfs_co_init_task(NFSClient *client, NFSRPC *task)
94 {
95 *task = (NFSRPC) {
96 .co = qemu_coroutine_self(),
97 .client = client,
98 };
99 }
100
101 static void nfs_co_generic_bh_cb(void *opaque)
102 {
103 NFSRPC *task = opaque;
104 task->complete = 1;
105 qemu_coroutine_enter(task->co);
106 }
107
108 static void
109 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
110 void *private_data)
111 {
112 NFSRPC *task = private_data;
113 task->ret = ret;
114 if (task->ret > 0 && task->iov) {
115 if (task->ret <= task->iov->size) {
116 qemu_iovec_from_buf(task->iov, 0, data, task->ret);
117 } else {
118 task->ret = -EIO;
119 }
120 }
121 if (task->ret == 0 && task->st) {
122 memcpy(task->st, data, sizeof(struct stat));
123 }
124 if (task->ret < 0) {
125 error_report("NFS Error: %s", nfs_get_error(nfs));
126 }
127 if (task->co) {
128 aio_bh_schedule_oneshot(task->client->aio_context,
129 nfs_co_generic_bh_cb, task);
130 } else {
131 task->complete = 1;
132 }
133 }
134
135 static int coroutine_fn nfs_co_readv(BlockDriverState *bs,
136 int64_t sector_num, int nb_sectors,
137 QEMUIOVector *iov)
138 {
139 NFSClient *client = bs->opaque;
140 NFSRPC task;
141
142 nfs_co_init_task(client, &task);
143 task.iov = iov;
144
145 if (nfs_pread_async(client->context, client->fh,
146 sector_num * BDRV_SECTOR_SIZE,
147 nb_sectors * BDRV_SECTOR_SIZE,
148 nfs_co_generic_cb, &task) != 0) {
149 return -ENOMEM;
150 }
151
152 while (!task.complete) {
153 nfs_set_events(client);
154 qemu_coroutine_yield();
155 }
156
157 if (task.ret < 0) {
158 return task.ret;
159 }
160
161 /* zero pad short reads */
162 if (task.ret < iov->size) {
163 qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
164 }
165
166 return 0;
167 }
168
169 static int coroutine_fn nfs_co_writev(BlockDriverState *bs,
170 int64_t sector_num, int nb_sectors,
171 QEMUIOVector *iov)
172 {
173 NFSClient *client = bs->opaque;
174 NFSRPC task;
175 char *buf = NULL;
176
177 nfs_co_init_task(client, &task);
178
179 buf = g_try_malloc(nb_sectors * BDRV_SECTOR_SIZE);
180 if (nb_sectors && buf == NULL) {
181 return -ENOMEM;
182 }
183
184 qemu_iovec_to_buf(iov, 0, buf, nb_sectors * BDRV_SECTOR_SIZE);
185
186 if (nfs_pwrite_async(client->context, client->fh,
187 sector_num * BDRV_SECTOR_SIZE,
188 nb_sectors * BDRV_SECTOR_SIZE,
189 buf, nfs_co_generic_cb, &task) != 0) {
190 g_free(buf);
191 return -ENOMEM;
192 }
193
194 while (!task.complete) {
195 nfs_set_events(client);
196 qemu_coroutine_yield();
197 }
198
199 g_free(buf);
200
201 if (task.ret != nb_sectors * BDRV_SECTOR_SIZE) {
202 return task.ret < 0 ? task.ret : -EIO;
203 }
204
205 return 0;
206 }
207
208 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
209 {
210 NFSClient *client = bs->opaque;
211 NFSRPC task;
212
213 nfs_co_init_task(client, &task);
214
215 if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
216 &task) != 0) {
217 return -ENOMEM;
218 }
219
220 while (!task.complete) {
221 nfs_set_events(client);
222 qemu_coroutine_yield();
223 }
224
225 return task.ret;
226 }
227
228 /* TODO Convert to fine grained options */
229 static QemuOptsList runtime_opts = {
230 .name = "nfs",
231 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
232 .desc = {
233 {
234 .name = "filename",
235 .type = QEMU_OPT_STRING,
236 .help = "URL to the NFS file",
237 },
238 { /* end of list */ }
239 },
240 };
241
242 static void nfs_detach_aio_context(BlockDriverState *bs)
243 {
244 NFSClient *client = bs->opaque;
245
246 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
247 false, 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, nfs_get_fd(client->context),
267 false, NULL, NULL, NULL);
268 nfs_destroy_context(client->context);
269 }
270 memset(client, 0, sizeof(NFSClient));
271 }
272
273 static void nfs_file_close(BlockDriverState *bs)
274 {
275 NFSClient *client = bs->opaque;
276 nfs_client_close(client);
277 }
278
279 static int64_t nfs_client_open(NFSClient *client, const char *filename,
280 int flags, Error **errp, int open_flags)
281 {
282 int ret = -EINVAL, i;
283 struct stat st;
284 URI *uri;
285 QueryParams *qp = NULL;
286 char *file = NULL, *strp = NULL;
287
288 uri = uri_parse(filename);
289 if (!uri) {
290 error_setg(errp, "Invalid URL specified");
291 goto fail;
292 }
293 if (!uri->server) {
294 error_setg(errp, "Invalid URL specified");
295 goto fail;
296 }
297 strp = strrchr(uri->path, '/');
298 if (strp == NULL) {
299 error_setg(errp, "Invalid URL specified");
300 goto fail;
301 }
302 file = g_strdup(strp);
303 *strp = 0;
304
305 client->context = nfs_init_context();
306 if (client->context == NULL) {
307 error_setg(errp, "Failed to init NFS context");
308 goto fail;
309 }
310
311 qp = query_params_parse(uri->query);
312 for (i = 0; i < qp->n; i++) {
313 unsigned long long val;
314 if (!qp->p[i].value) {
315 error_setg(errp, "Value for NFS parameter expected: %s",
316 qp->p[i].name);
317 goto fail;
318 }
319 if (parse_uint_full(qp->p[i].value, &val, 0)) {
320 error_setg(errp, "Illegal value for NFS parameter: %s",
321 qp->p[i].name);
322 goto fail;
323 }
324 if (!strcmp(qp->p[i].name, "uid")) {
325 nfs_set_uid(client->context, val);
326 } else if (!strcmp(qp->p[i].name, "gid")) {
327 nfs_set_gid(client->context, val);
328 } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
329 nfs_set_tcp_syncnt(client->context, val);
330 #ifdef LIBNFS_FEATURE_READAHEAD
331 } else if (!strcmp(qp->p[i].name, "readahead")) {
332 if (open_flags & BDRV_O_NOCACHE) {
333 error_setg(errp, "Cannot enable NFS readahead "
334 "if cache.direct = on");
335 goto fail;
336 }
337 if (val > QEMU_NFS_MAX_READAHEAD_SIZE) {
338 error_report("NFS Warning: Truncating NFS readahead"
339 " size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
340 val = QEMU_NFS_MAX_READAHEAD_SIZE;
341 }
342 nfs_set_readahead(client->context, val);
343 #ifdef LIBNFS_FEATURE_PAGECACHE
344 nfs_set_pagecache_ttl(client->context, 0);
345 #endif
346 client->cache_used = true;
347 #endif
348 #ifdef LIBNFS_FEATURE_PAGECACHE
349 nfs_set_pagecache_ttl(client->context, 0);
350 } else if (!strcmp(qp->p[i].name, "pagecache")) {
351 if (open_flags & BDRV_O_NOCACHE) {
352 error_setg(errp, "Cannot enable NFS pagecache "
353 "if cache.direct = on");
354 goto fail;
355 }
356 if (val > QEMU_NFS_MAX_PAGECACHE_SIZE) {
357 error_report("NFS Warning: Truncating NFS pagecache"
358 " size to %d pages", QEMU_NFS_MAX_PAGECACHE_SIZE);
359 val = QEMU_NFS_MAX_PAGECACHE_SIZE;
360 }
361 nfs_set_pagecache(client->context, val);
362 nfs_set_pagecache_ttl(client->context, 0);
363 client->cache_used = true;
364 #endif
365 #ifdef LIBNFS_FEATURE_DEBUG
366 } else if (!strcmp(qp->p[i].name, "debug")) {
367 /* limit the maximum debug level to avoid potential flooding
368 * of our log files. */
369 if (val > QEMU_NFS_MAX_DEBUG_LEVEL) {
370 error_report("NFS Warning: Limiting NFS debug level"
371 " to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
372 val = QEMU_NFS_MAX_DEBUG_LEVEL;
373 }
374 nfs_set_debug(client->context, val);
375 #endif
376 } else {
377 error_setg(errp, "Unknown NFS parameter name: %s",
378 qp->p[i].name);
379 goto fail;
380 }
381 }
382
383 ret = nfs_mount(client->context, uri->server, uri->path);
384 if (ret < 0) {
385 error_setg(errp, "Failed to mount nfs share: %s",
386 nfs_get_error(client->context));
387 goto fail;
388 }
389
390 if (flags & O_CREAT) {
391 ret = nfs_creat(client->context, file, 0600, &client->fh);
392 if (ret < 0) {
393 error_setg(errp, "Failed to create file: %s",
394 nfs_get_error(client->context));
395 goto fail;
396 }
397 } else {
398 ret = nfs_open(client->context, file, flags, &client->fh);
399 if (ret < 0) {
400 error_setg(errp, "Failed to open file : %s",
401 nfs_get_error(client->context));
402 goto fail;
403 }
404 }
405
406 ret = nfs_fstat(client->context, client->fh, &st);
407 if (ret < 0) {
408 error_setg(errp, "Failed to fstat file: %s",
409 nfs_get_error(client->context));
410 goto fail;
411 }
412
413 ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
414 client->st_blocks = st.st_blocks;
415 client->has_zero_init = S_ISREG(st.st_mode);
416 goto out;
417 fail:
418 nfs_client_close(client);
419 out:
420 if (qp) {
421 query_params_free(qp);
422 }
423 uri_free(uri);
424 g_free(file);
425 return ret;
426 }
427
428 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
429 Error **errp) {
430 NFSClient *client = bs->opaque;
431 int64_t ret;
432 QemuOpts *opts;
433 Error *local_err = NULL;
434
435 client->aio_context = bdrv_get_aio_context(bs);
436
437 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
438 qemu_opts_absorb_qdict(opts, options, &local_err);
439 if (local_err) {
440 error_propagate(errp, local_err);
441 ret = -EINVAL;
442 goto out;
443 }
444 ret = nfs_client_open(client, qemu_opt_get(opts, "filename"),
445 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
446 errp, bs->open_flags);
447 if (ret < 0) {
448 goto out;
449 }
450 bs->total_sectors = ret;
451 ret = 0;
452 out:
453 qemu_opts_del(opts);
454 return ret;
455 }
456
457 static QemuOptsList nfs_create_opts = {
458 .name = "nfs-create-opts",
459 .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
460 .desc = {
461 {
462 .name = BLOCK_OPT_SIZE,
463 .type = QEMU_OPT_SIZE,
464 .help = "Virtual disk size"
465 },
466 { /* end of list */ }
467 }
468 };
469
470 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
471 {
472 int ret = 0;
473 int64_t total_size = 0;
474 NFSClient *client = g_new0(NFSClient, 1);
475
476 client->aio_context = qemu_get_aio_context();
477
478 /* Read out options */
479 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
480 BDRV_SECTOR_SIZE);
481
482 ret = nfs_client_open(client, url, O_CREAT, errp, 0);
483 if (ret < 0) {
484 goto out;
485 }
486 ret = nfs_ftruncate(client->context, client->fh, total_size);
487 nfs_client_close(client);
488 out:
489 g_free(client);
490 return ret;
491 }
492
493 static int nfs_has_zero_init(BlockDriverState *bs)
494 {
495 NFSClient *client = bs->opaque;
496 return client->has_zero_init;
497 }
498
499 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
500 {
501 NFSClient *client = bs->opaque;
502 NFSRPC task = {0};
503 struct stat st;
504
505 if (bdrv_is_read_only(bs) &&
506 !(bs->open_flags & BDRV_O_NOCACHE)) {
507 return client->st_blocks * 512;
508 }
509
510 task.st = &st;
511 if (nfs_fstat_async(client->context, client->fh, nfs_co_generic_cb,
512 &task) != 0) {
513 return -ENOMEM;
514 }
515
516 while (!task.complete) {
517 nfs_set_events(client);
518 aio_poll(client->aio_context, true);
519 }
520
521 return (task.ret < 0 ? task.ret : st.st_blocks * 512);
522 }
523
524 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
525 {
526 NFSClient *client = bs->opaque;
527 return nfs_ftruncate(client->context, client->fh, offset);
528 }
529
530 /* Note that this will not re-establish a connection with the NFS server
531 * - it is effectively a NOP. */
532 static int nfs_reopen_prepare(BDRVReopenState *state,
533 BlockReopenQueue *queue, Error **errp)
534 {
535 NFSClient *client = state->bs->opaque;
536 struct stat st;
537 int ret = 0;
538
539 if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
540 error_setg(errp, "Cannot open a read-only mount as read-write");
541 return -EACCES;
542 }
543
544 if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
545 error_setg(errp, "Cannot disable cache if libnfs readahead or"
546 " pagecache is enabled");
547 return -EINVAL;
548 }
549
550 /* Update cache for read-only reopens */
551 if (!(state->flags & BDRV_O_RDWR)) {
552 ret = nfs_fstat(client->context, client->fh, &st);
553 if (ret < 0) {
554 error_setg(errp, "Failed to fstat file: %s",
555 nfs_get_error(client->context));
556 return ret;
557 }
558 client->st_blocks = st.st_blocks;
559 }
560
561 return 0;
562 }
563
564 #ifdef LIBNFS_FEATURE_PAGECACHE
565 static void nfs_invalidate_cache(BlockDriverState *bs,
566 Error **errp)
567 {
568 NFSClient *client = bs->opaque;
569 nfs_pagecache_invalidate(client->context, client->fh);
570 }
571 #endif
572
573 static BlockDriver bdrv_nfs = {
574 .format_name = "nfs",
575 .protocol_name = "nfs",
576
577 .instance_size = sizeof(NFSClient),
578 .bdrv_needs_filename = true,
579 .create_opts = &nfs_create_opts,
580
581 .bdrv_has_zero_init = nfs_has_zero_init,
582 .bdrv_get_allocated_file_size = nfs_get_allocated_file_size,
583 .bdrv_truncate = nfs_file_truncate,
584
585 .bdrv_file_open = nfs_file_open,
586 .bdrv_close = nfs_file_close,
587 .bdrv_create = nfs_file_create,
588 .bdrv_reopen_prepare = nfs_reopen_prepare,
589
590 .bdrv_co_readv = nfs_co_readv,
591 .bdrv_co_writev = nfs_co_writev,
592 .bdrv_co_flush_to_disk = nfs_co_flush,
593
594 .bdrv_detach_aio_context = nfs_detach_aio_context,
595 .bdrv_attach_aio_context = nfs_attach_aio_context,
596
597 #ifdef LIBNFS_FEATURE_PAGECACHE
598 .bdrv_invalidate_cache = nfs_invalidate_cache,
599 #endif
600 };
601
602 static void nfs_block_init(void)
603 {
604 bdrv_register(&bdrv_nfs);
605 }
606
607 block_init(nfs_block_init);