]> git.proxmox.com Git - mirror_qemu.git/blob - block/nfs.c
Merge remote-tracking branch 'remotes/xtensa/tags/20160714-xtensa' 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 QEMUBH *bh;
61 NFSClient *client;
62 } NFSRPC;
63
64 static void nfs_process_read(void *arg);
65 static void nfs_process_write(void *arg);
66
67 static void nfs_set_events(NFSClient *client)
68 {
69 int ev = nfs_which_events(client->context);
70 if (ev != client->events) {
71 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
72 false,
73 (ev & POLLIN) ? nfs_process_read : NULL,
74 (ev & POLLOUT) ? nfs_process_write : NULL, client);
75
76 }
77 client->events = ev;
78 }
79
80 static void nfs_process_read(void *arg)
81 {
82 NFSClient *client = arg;
83 nfs_service(client->context, POLLIN);
84 nfs_set_events(client);
85 }
86
87 static void nfs_process_write(void *arg)
88 {
89 NFSClient *client = arg;
90 nfs_service(client->context, POLLOUT);
91 nfs_set_events(client);
92 }
93
94 static void nfs_co_init_task(NFSClient *client, NFSRPC *task)
95 {
96 *task = (NFSRPC) {
97 .co = qemu_coroutine_self(),
98 .client = client,
99 };
100 }
101
102 static void nfs_co_generic_bh_cb(void *opaque)
103 {
104 NFSRPC *task = opaque;
105 task->complete = 1;
106 qemu_bh_delete(task->bh);
107 qemu_coroutine_enter(task->co);
108 }
109
110 static void
111 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
112 void *private_data)
113 {
114 NFSRPC *task = private_data;
115 task->ret = ret;
116 if (task->ret > 0 && task->iov) {
117 if (task->ret <= task->iov->size) {
118 qemu_iovec_from_buf(task->iov, 0, data, task->ret);
119 } else {
120 task->ret = -EIO;
121 }
122 }
123 if (task->ret == 0 && task->st) {
124 memcpy(task->st, data, sizeof(struct stat));
125 }
126 if (task->ret < 0) {
127 error_report("NFS Error: %s", nfs_get_error(nfs));
128 }
129 if (task->co) {
130 task->bh = aio_bh_new(task->client->aio_context,
131 nfs_co_generic_bh_cb, task);
132 qemu_bh_schedule(task->bh);
133 } else {
134 task->complete = 1;
135 }
136 }
137
138 static int coroutine_fn nfs_co_readv(BlockDriverState *bs,
139 int64_t sector_num, int nb_sectors,
140 QEMUIOVector *iov)
141 {
142 NFSClient *client = bs->opaque;
143 NFSRPC task;
144
145 nfs_co_init_task(client, &task);
146 task.iov = iov;
147
148 if (nfs_pread_async(client->context, client->fh,
149 sector_num * BDRV_SECTOR_SIZE,
150 nb_sectors * BDRV_SECTOR_SIZE,
151 nfs_co_generic_cb, &task) != 0) {
152 return -ENOMEM;
153 }
154
155 while (!task.complete) {
156 nfs_set_events(client);
157 qemu_coroutine_yield();
158 }
159
160 if (task.ret < 0) {
161 return task.ret;
162 }
163
164 /* zero pad short reads */
165 if (task.ret < iov->size) {
166 qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
167 }
168
169 return 0;
170 }
171
172 static int coroutine_fn nfs_co_writev(BlockDriverState *bs,
173 int64_t sector_num, int nb_sectors,
174 QEMUIOVector *iov)
175 {
176 NFSClient *client = bs->opaque;
177 NFSRPC task;
178 char *buf = NULL;
179
180 nfs_co_init_task(client, &task);
181
182 buf = g_try_malloc(nb_sectors * BDRV_SECTOR_SIZE);
183 if (nb_sectors && buf == NULL) {
184 return -ENOMEM;
185 }
186
187 qemu_iovec_to_buf(iov, 0, buf, nb_sectors * BDRV_SECTOR_SIZE);
188
189 if (nfs_pwrite_async(client->context, client->fh,
190 sector_num * BDRV_SECTOR_SIZE,
191 nb_sectors * BDRV_SECTOR_SIZE,
192 buf, nfs_co_generic_cb, &task) != 0) {
193 g_free(buf);
194 return -ENOMEM;
195 }
196
197 while (!task.complete) {
198 nfs_set_events(client);
199 qemu_coroutine_yield();
200 }
201
202 g_free(buf);
203
204 if (task.ret != nb_sectors * BDRV_SECTOR_SIZE) {
205 return task.ret < 0 ? task.ret : -EIO;
206 }
207
208 return 0;
209 }
210
211 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
212 {
213 NFSClient *client = bs->opaque;
214 NFSRPC task;
215
216 nfs_co_init_task(client, &task);
217
218 if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
219 &task) != 0) {
220 return -ENOMEM;
221 }
222
223 while (!task.complete) {
224 nfs_set_events(client);
225 qemu_coroutine_yield();
226 }
227
228 return task.ret;
229 }
230
231 /* TODO Convert to fine grained options */
232 static QemuOptsList runtime_opts = {
233 .name = "nfs",
234 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
235 .desc = {
236 {
237 .name = "filename",
238 .type = QEMU_OPT_STRING,
239 .help = "URL to the NFS file",
240 },
241 { /* end of list */ }
242 },
243 };
244
245 static void nfs_detach_aio_context(BlockDriverState *bs)
246 {
247 NFSClient *client = bs->opaque;
248
249 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
250 false, NULL, NULL, NULL);
251 client->events = 0;
252 }
253
254 static void nfs_attach_aio_context(BlockDriverState *bs,
255 AioContext *new_context)
256 {
257 NFSClient *client = bs->opaque;
258
259 client->aio_context = new_context;
260 nfs_set_events(client);
261 }
262
263 static void nfs_client_close(NFSClient *client)
264 {
265 if (client->context) {
266 if (client->fh) {
267 nfs_close(client->context, client->fh);
268 }
269 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
270 false, NULL, NULL, NULL);
271 nfs_destroy_context(client->context);
272 }
273 memset(client, 0, sizeof(NFSClient));
274 }
275
276 static void nfs_file_close(BlockDriverState *bs)
277 {
278 NFSClient *client = bs->opaque;
279 nfs_client_close(client);
280 }
281
282 static int64_t nfs_client_open(NFSClient *client, const char *filename,
283 int flags, Error **errp, int open_flags)
284 {
285 int ret = -EINVAL, i;
286 struct stat st;
287 URI *uri;
288 QueryParams *qp = NULL;
289 char *file = NULL, *strp = NULL;
290
291 uri = uri_parse(filename);
292 if (!uri) {
293 error_setg(errp, "Invalid URL specified");
294 goto fail;
295 }
296 if (!uri->server) {
297 error_setg(errp, "Invalid URL specified");
298 goto fail;
299 }
300 strp = strrchr(uri->path, '/');
301 if (strp == NULL) {
302 error_setg(errp, "Invalid URL specified");
303 goto fail;
304 }
305 file = g_strdup(strp);
306 *strp = 0;
307
308 client->context = nfs_init_context();
309 if (client->context == NULL) {
310 error_setg(errp, "Failed to init NFS context");
311 goto fail;
312 }
313
314 qp = query_params_parse(uri->query);
315 for (i = 0; i < qp->n; i++) {
316 unsigned long long val;
317 if (!qp->p[i].value) {
318 error_setg(errp, "Value for NFS parameter expected: %s",
319 qp->p[i].name);
320 goto fail;
321 }
322 if (parse_uint_full(qp->p[i].value, &val, 0)) {
323 error_setg(errp, "Illegal value for NFS parameter: %s",
324 qp->p[i].name);
325 goto fail;
326 }
327 if (!strcmp(qp->p[i].name, "uid")) {
328 nfs_set_uid(client->context, val);
329 } else if (!strcmp(qp->p[i].name, "gid")) {
330 nfs_set_gid(client->context, val);
331 } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
332 nfs_set_tcp_syncnt(client->context, val);
333 #ifdef LIBNFS_FEATURE_READAHEAD
334 } else if (!strcmp(qp->p[i].name, "readahead")) {
335 if (open_flags & BDRV_O_NOCACHE) {
336 error_setg(errp, "Cannot enable NFS readahead "
337 "if cache.direct = on");
338 goto fail;
339 }
340 if (val > QEMU_NFS_MAX_READAHEAD_SIZE) {
341 error_report("NFS Warning: Truncating NFS readahead"
342 " size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
343 val = QEMU_NFS_MAX_READAHEAD_SIZE;
344 }
345 nfs_set_readahead(client->context, val);
346 #ifdef LIBNFS_FEATURE_PAGECACHE
347 nfs_set_pagecache_ttl(client->context, 0);
348 #endif
349 client->cache_used = true;
350 #endif
351 #ifdef LIBNFS_FEATURE_PAGECACHE
352 nfs_set_pagecache_ttl(client->context, 0);
353 } else if (!strcmp(qp->p[i].name, "pagecache")) {
354 if (open_flags & BDRV_O_NOCACHE) {
355 error_setg(errp, "Cannot enable NFS pagecache "
356 "if cache.direct = on");
357 goto fail;
358 }
359 if (val > QEMU_NFS_MAX_PAGECACHE_SIZE) {
360 error_report("NFS Warning: Truncating NFS pagecache"
361 " size to %d pages", QEMU_NFS_MAX_PAGECACHE_SIZE);
362 val = QEMU_NFS_MAX_PAGECACHE_SIZE;
363 }
364 nfs_set_pagecache(client->context, val);
365 nfs_set_pagecache_ttl(client->context, 0);
366 client->cache_used = true;
367 #endif
368 #ifdef LIBNFS_FEATURE_DEBUG
369 } else if (!strcmp(qp->p[i].name, "debug")) {
370 /* limit the maximum debug level to avoid potential flooding
371 * of our log files. */
372 if (val > QEMU_NFS_MAX_DEBUG_LEVEL) {
373 error_report("NFS Warning: Limiting NFS debug level"
374 " to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
375 val = QEMU_NFS_MAX_DEBUG_LEVEL;
376 }
377 nfs_set_debug(client->context, val);
378 #endif
379 } else {
380 error_setg(errp, "Unknown NFS parameter name: %s",
381 qp->p[i].name);
382 goto fail;
383 }
384 }
385
386 ret = nfs_mount(client->context, uri->server, uri->path);
387 if (ret < 0) {
388 error_setg(errp, "Failed to mount nfs share: %s",
389 nfs_get_error(client->context));
390 goto fail;
391 }
392
393 if (flags & O_CREAT) {
394 ret = nfs_creat(client->context, file, 0600, &client->fh);
395 if (ret < 0) {
396 error_setg(errp, "Failed to create file: %s",
397 nfs_get_error(client->context));
398 goto fail;
399 }
400 } else {
401 ret = nfs_open(client->context, file, flags, &client->fh);
402 if (ret < 0) {
403 error_setg(errp, "Failed to open file : %s",
404 nfs_get_error(client->context));
405 goto fail;
406 }
407 }
408
409 ret = nfs_fstat(client->context, client->fh, &st);
410 if (ret < 0) {
411 error_setg(errp, "Failed to fstat file: %s",
412 nfs_get_error(client->context));
413 goto fail;
414 }
415
416 ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
417 client->st_blocks = st.st_blocks;
418 client->has_zero_init = S_ISREG(st.st_mode);
419 goto out;
420 fail:
421 nfs_client_close(client);
422 out:
423 if (qp) {
424 query_params_free(qp);
425 }
426 uri_free(uri);
427 g_free(file);
428 return ret;
429 }
430
431 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
432 Error **errp) {
433 NFSClient *client = bs->opaque;
434 int64_t ret;
435 QemuOpts *opts;
436 Error *local_err = NULL;
437
438 client->aio_context = bdrv_get_aio_context(bs);
439
440 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
441 qemu_opts_absorb_qdict(opts, options, &local_err);
442 if (local_err) {
443 error_propagate(errp, local_err);
444 ret = -EINVAL;
445 goto out;
446 }
447 ret = nfs_client_open(client, qemu_opt_get(opts, "filename"),
448 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
449 errp, bs->open_flags);
450 if (ret < 0) {
451 goto out;
452 }
453 bs->total_sectors = ret;
454 ret = 0;
455 out:
456 qemu_opts_del(opts);
457 return ret;
458 }
459
460 static QemuOptsList nfs_create_opts = {
461 .name = "nfs-create-opts",
462 .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
463 .desc = {
464 {
465 .name = BLOCK_OPT_SIZE,
466 .type = QEMU_OPT_SIZE,
467 .help = "Virtual disk size"
468 },
469 { /* end of list */ }
470 }
471 };
472
473 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
474 {
475 int ret = 0;
476 int64_t total_size = 0;
477 NFSClient *client = g_new0(NFSClient, 1);
478
479 client->aio_context = qemu_get_aio_context();
480
481 /* Read out options */
482 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
483 BDRV_SECTOR_SIZE);
484
485 ret = nfs_client_open(client, url, O_CREAT, errp, 0);
486 if (ret < 0) {
487 goto out;
488 }
489 ret = nfs_ftruncate(client->context, client->fh, total_size);
490 nfs_client_close(client);
491 out:
492 g_free(client);
493 return ret;
494 }
495
496 static int nfs_has_zero_init(BlockDriverState *bs)
497 {
498 NFSClient *client = bs->opaque;
499 return client->has_zero_init;
500 }
501
502 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
503 {
504 NFSClient *client = bs->opaque;
505 NFSRPC task = {0};
506 struct stat st;
507
508 if (bdrv_is_read_only(bs) &&
509 !(bs->open_flags & BDRV_O_NOCACHE)) {
510 return client->st_blocks * 512;
511 }
512
513 task.st = &st;
514 if (nfs_fstat_async(client->context, client->fh, nfs_co_generic_cb,
515 &task) != 0) {
516 return -ENOMEM;
517 }
518
519 while (!task.complete) {
520 nfs_set_events(client);
521 aio_poll(client->aio_context, true);
522 }
523
524 return (task.ret < 0 ? task.ret : st.st_blocks * 512);
525 }
526
527 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
528 {
529 NFSClient *client = bs->opaque;
530 return nfs_ftruncate(client->context, client->fh, offset);
531 }
532
533 /* Note that this will not re-establish a connection with the NFS server
534 * - it is effectively a NOP. */
535 static int nfs_reopen_prepare(BDRVReopenState *state,
536 BlockReopenQueue *queue, Error **errp)
537 {
538 NFSClient *client = state->bs->opaque;
539 struct stat st;
540 int ret = 0;
541
542 if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
543 error_setg(errp, "Cannot open a read-only mount as read-write");
544 return -EACCES;
545 }
546
547 if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
548 error_setg(errp, "Cannot disable cache if libnfs readahead or"
549 " pagecache is enabled");
550 return -EINVAL;
551 }
552
553 /* Update cache for read-only reopens */
554 if (!(state->flags & BDRV_O_RDWR)) {
555 ret = nfs_fstat(client->context, client->fh, &st);
556 if (ret < 0) {
557 error_setg(errp, "Failed to fstat file: %s",
558 nfs_get_error(client->context));
559 return ret;
560 }
561 client->st_blocks = st.st_blocks;
562 }
563
564 return 0;
565 }
566
567 #ifdef LIBNFS_FEATURE_PAGECACHE
568 static void nfs_invalidate_cache(BlockDriverState *bs,
569 Error **errp)
570 {
571 NFSClient *client = bs->opaque;
572 nfs_pagecache_invalidate(client->context, client->fh);
573 }
574 #endif
575
576 static BlockDriver bdrv_nfs = {
577 .format_name = "nfs",
578 .protocol_name = "nfs",
579
580 .instance_size = sizeof(NFSClient),
581 .bdrv_needs_filename = true,
582 .create_opts = &nfs_create_opts,
583
584 .bdrv_has_zero_init = nfs_has_zero_init,
585 .bdrv_get_allocated_file_size = nfs_get_allocated_file_size,
586 .bdrv_truncate = nfs_file_truncate,
587
588 .bdrv_file_open = nfs_file_open,
589 .bdrv_close = nfs_file_close,
590 .bdrv_create = nfs_file_create,
591 .bdrv_reopen_prepare = nfs_reopen_prepare,
592
593 .bdrv_co_readv = nfs_co_readv,
594 .bdrv_co_writev = nfs_co_writev,
595 .bdrv_co_flush_to_disk = nfs_co_flush,
596
597 .bdrv_detach_aio_context = nfs_detach_aio_context,
598 .bdrv_attach_aio_context = nfs_attach_aio_context,
599
600 #ifdef LIBNFS_FEATURE_PAGECACHE
601 .bdrv_invalidate_cache = nfs_invalidate_cache,
602 #endif
603 };
604
605 static void nfs_block_init(void)
606 {
607 bdrv_register(&bdrv_nfs);
608 }
609
610 block_init(nfs_block_init);