2 * QEMU Block driver for native access to files on NFS shares
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
25 #include "qemu/osdep.h"
28 #include "qemu/config-file.h"
29 #include "qemu/error-report.h"
30 #include "qapi/error.h"
31 #include "block/block_int.h"
32 #include "block/qdict.h"
35 #include "qemu/option.h"
37 #include "qemu/cutils.h"
38 #include "sysemu/sysemu.h"
39 #include "qapi/qapi-visit-block-core.h"
40 #include "qapi/qmp/qdict.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/qobject-input-visitor.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include <nfsc/libnfs.h>
47 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
48 #define QEMU_NFS_MAX_PAGECACHE_SIZE (8388608 / NFS_BLKSIZE)
49 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
51 typedef struct NFSClient {
52 struct nfs_context *context;
56 AioContext *aio_context;
62 int64_t uid, gid, tcp_syncnt, readahead, pagecache, debug;
65 typedef struct NFSRPC {
75 static int nfs_parse_uri(const char *filename, QDict *options, Error **errp)
78 QueryParams *qp = NULL;
81 uri = uri_parse(filename);
83 error_setg(errp, "Invalid URI specified");
86 if (g_strcmp0(uri->scheme, "nfs") != 0) {
87 error_setg(errp, "URI scheme must be 'nfs'");
92 error_setg(errp, "missing hostname in URI");
97 error_setg(errp, "missing file path in URI");
101 qp = query_params_parse(uri->query);
103 error_setg(errp, "could not parse query parameters");
107 qdict_put_str(options, "server.host", uri->server);
108 qdict_put_str(options, "server.type", "inet");
109 qdict_put_str(options, "path", uri->path);
111 for (i = 0; i < qp->n; i++) {
112 unsigned long long val;
113 if (!qp->p[i].value) {
114 error_setg(errp, "Value for NFS parameter expected: %s",
118 if (parse_uint_full(qp->p[i].value, &val, 0)) {
119 error_setg(errp, "Illegal value for NFS parameter: %s",
123 if (!strcmp(qp->p[i].name, "uid")) {
124 qdict_put_str(options, "user", qp->p[i].value);
125 } else if (!strcmp(qp->p[i].name, "gid")) {
126 qdict_put_str(options, "group", qp->p[i].value);
127 } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
128 qdict_put_str(options, "tcp-syn-count", qp->p[i].value);
129 } else if (!strcmp(qp->p[i].name, "readahead")) {
130 qdict_put_str(options, "readahead-size", qp->p[i].value);
131 } else if (!strcmp(qp->p[i].name, "pagecache")) {
132 qdict_put_str(options, "page-cache-size", qp->p[i].value);
133 } else if (!strcmp(qp->p[i].name, "debug")) {
134 qdict_put_str(options, "debug", qp->p[i].value);
136 error_setg(errp, "Unknown NFS parameter name: %s",
144 query_params_free(qp);
152 static bool nfs_has_filename_options_conflict(QDict *options, Error **errp)
154 const QDictEntry *qe;
156 for (qe = qdict_first(options); qe; qe = qdict_next(options, qe)) {
157 if (!strcmp(qe->key, "host") ||
158 !strcmp(qe->key, "path") ||
159 !strcmp(qe->key, "user") ||
160 !strcmp(qe->key, "group") ||
161 !strcmp(qe->key, "tcp-syn-count") ||
162 !strcmp(qe->key, "readahead-size") ||
163 !strcmp(qe->key, "page-cache-size") ||
164 !strcmp(qe->key, "debug") ||
165 strstart(qe->key, "server.", NULL))
167 error_setg(errp, "Option %s cannot be used with a filename",
176 static void nfs_parse_filename(const char *filename, QDict *options,
179 if (nfs_has_filename_options_conflict(options, errp)) {
183 nfs_parse_uri(filename, options, errp);
186 static void nfs_process_read(void *arg);
187 static void nfs_process_write(void *arg);
189 /* Called with QemuMutex held. */
190 static void nfs_set_events(NFSClient *client)
192 int ev = nfs_which_events(client->context);
193 if (ev != client->events) {
194 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
196 (ev & POLLIN) ? nfs_process_read : NULL,
197 (ev & POLLOUT) ? nfs_process_write : NULL,
204 static void nfs_process_read(void *arg)
206 NFSClient *client = arg;
208 qemu_mutex_lock(&client->mutex);
209 nfs_service(client->context, POLLIN);
210 nfs_set_events(client);
211 qemu_mutex_unlock(&client->mutex);
214 static void nfs_process_write(void *arg)
216 NFSClient *client = arg;
218 qemu_mutex_lock(&client->mutex);
219 nfs_service(client->context, POLLOUT);
220 nfs_set_events(client);
221 qemu_mutex_unlock(&client->mutex);
224 static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
227 .co = qemu_coroutine_self(),
229 .client = bs->opaque,
233 static void nfs_co_generic_bh_cb(void *opaque)
235 NFSRPC *task = opaque;
238 aio_co_wake(task->co);
241 /* Called (via nfs_service) with QemuMutex held. */
243 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
246 NFSRPC *task = private_data;
249 if (task->ret > 0 && task->iov) {
250 if (task->ret <= task->iov->size) {
251 qemu_iovec_from_buf(task->iov, 0, data, task->ret);
257 error_report("NFS Error: %s", nfs_get_error(nfs));
259 aio_bh_schedule_oneshot(task->client->aio_context,
260 nfs_co_generic_bh_cb, task);
263 static int coroutine_fn nfs_co_preadv(BlockDriverState *bs, uint64_t offset,
264 uint64_t bytes, QEMUIOVector *iov,
267 NFSClient *client = bs->opaque;
270 nfs_co_init_task(bs, &task);
273 qemu_mutex_lock(&client->mutex);
274 if (nfs_pread_async(client->context, client->fh,
275 offset, bytes, nfs_co_generic_cb, &task) != 0) {
276 qemu_mutex_unlock(&client->mutex);
280 nfs_set_events(client);
281 qemu_mutex_unlock(&client->mutex);
282 while (!task.complete) {
283 qemu_coroutine_yield();
290 /* zero pad short reads */
291 if (task.ret < iov->size) {
292 qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
298 static int coroutine_fn nfs_co_pwritev(BlockDriverState *bs, uint64_t offset,
299 uint64_t bytes, QEMUIOVector *iov,
302 NFSClient *client = bs->opaque;
305 bool my_buffer = false;
307 nfs_co_init_task(bs, &task);
309 if (iov->niov != 1) {
310 buf = g_try_malloc(bytes);
311 if (bytes && buf == NULL) {
314 qemu_iovec_to_buf(iov, 0, buf, bytes);
317 buf = iov->iov[0].iov_base;
320 qemu_mutex_lock(&client->mutex);
321 if (nfs_pwrite_async(client->context, client->fh,
323 nfs_co_generic_cb, &task) != 0) {
324 qemu_mutex_unlock(&client->mutex);
331 nfs_set_events(client);
332 qemu_mutex_unlock(&client->mutex);
333 while (!task.complete) {
334 qemu_coroutine_yield();
341 if (task.ret != bytes) {
342 return task.ret < 0 ? task.ret : -EIO;
348 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
350 NFSClient *client = bs->opaque;
353 nfs_co_init_task(bs, &task);
355 qemu_mutex_lock(&client->mutex);
356 if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
358 qemu_mutex_unlock(&client->mutex);
362 nfs_set_events(client);
363 qemu_mutex_unlock(&client->mutex);
364 while (!task.complete) {
365 qemu_coroutine_yield();
371 static void nfs_detach_aio_context(BlockDriverState *bs)
373 NFSClient *client = bs->opaque;
375 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
376 false, NULL, NULL, NULL, NULL);
380 static void nfs_attach_aio_context(BlockDriverState *bs,
381 AioContext *new_context)
383 NFSClient *client = bs->opaque;
385 client->aio_context = new_context;
386 nfs_set_events(client);
389 static void nfs_client_close(NFSClient *client)
391 if (client->context) {
393 nfs_close(client->context, client->fh);
396 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
397 false, NULL, NULL, NULL, NULL);
398 nfs_destroy_context(client->context);
399 client->context = NULL;
401 g_free(client->path);
402 qemu_mutex_destroy(&client->mutex);
403 qapi_free_NFSServer(client->server);
404 client->server = NULL;
407 static void nfs_file_close(BlockDriverState *bs)
409 NFSClient *client = bs->opaque;
410 nfs_client_close(client);
413 static int64_t nfs_client_open(NFSClient *client, BlockdevOptionsNfs *opts,
414 int flags, int open_flags, Error **errp)
416 int64_t ret = -EINVAL;
418 char *file = NULL, *strp = NULL;
420 qemu_mutex_init(&client->mutex);
422 client->path = g_strdup(opts->path);
424 strp = strrchr(client->path, '/');
426 error_setg(errp, "Invalid URL specified");
429 file = g_strdup(strp);
432 /* Steal the NFSServer object from opts; set the original pointer to NULL
433 * to avoid use after free and double free. */
434 client->server = opts->server;
437 client->context = nfs_init_context();
438 if (client->context == NULL) {
439 error_setg(errp, "Failed to init NFS context");
443 if (opts->has_user) {
444 client->uid = opts->user;
445 nfs_set_uid(client->context, client->uid);
448 if (opts->has_group) {
449 client->gid = opts->group;
450 nfs_set_gid(client->context, client->gid);
453 if (opts->has_tcp_syn_count) {
454 client->tcp_syncnt = opts->tcp_syn_count;
455 nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
458 #ifdef LIBNFS_FEATURE_READAHEAD
459 if (opts->has_readahead_size) {
460 if (open_flags & BDRV_O_NOCACHE) {
461 error_setg(errp, "Cannot enable NFS readahead "
462 "if cache.direct = on");
465 client->readahead = opts->readahead_size;
466 if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
467 warn_report("Truncating NFS readahead size to %d",
468 QEMU_NFS_MAX_READAHEAD_SIZE);
469 client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
471 nfs_set_readahead(client->context, client->readahead);
472 #ifdef LIBNFS_FEATURE_PAGECACHE
473 nfs_set_pagecache_ttl(client->context, 0);
475 client->cache_used = true;
479 #ifdef LIBNFS_FEATURE_PAGECACHE
480 if (opts->has_page_cache_size) {
481 if (open_flags & BDRV_O_NOCACHE) {
482 error_setg(errp, "Cannot enable NFS pagecache "
483 "if cache.direct = on");
486 client->pagecache = opts->page_cache_size;
487 if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
488 warn_report("Truncating NFS pagecache size to %d pages",
489 QEMU_NFS_MAX_PAGECACHE_SIZE);
490 client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
492 nfs_set_pagecache(client->context, client->pagecache);
493 nfs_set_pagecache_ttl(client->context, 0);
494 client->cache_used = true;
498 #ifdef LIBNFS_FEATURE_DEBUG
499 if (opts->has_debug) {
500 client->debug = opts->debug;
501 /* limit the maximum debug level to avoid potential flooding
502 * of our log files. */
503 if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
504 warn_report("Limiting NFS debug level to %d",
505 QEMU_NFS_MAX_DEBUG_LEVEL);
506 client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
508 nfs_set_debug(client->context, client->debug);
512 ret = nfs_mount(client->context, client->server->host, client->path);
514 error_setg(errp, "Failed to mount nfs share: %s",
515 nfs_get_error(client->context));
519 if (flags & O_CREAT) {
520 ret = nfs_creat(client->context, file, 0600, &client->fh);
522 error_setg(errp, "Failed to create file: %s",
523 nfs_get_error(client->context));
527 ret = nfs_open(client->context, file, flags, &client->fh);
529 error_setg(errp, "Failed to open file : %s",
530 nfs_get_error(client->context));
535 ret = nfs_fstat(client->context, client->fh, &st);
537 error_setg(errp, "Failed to fstat file: %s",
538 nfs_get_error(client->context));
542 ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
543 client->st_blocks = st.st_blocks;
544 client->has_zero_init = S_ISREG(st.st_mode);
549 nfs_client_close(client);
555 static BlockdevOptionsNfs *nfs_options_qdict_to_qapi(QDict *options,
558 BlockdevOptionsNfs *opts = NULL;
561 Error *local_err = NULL;
563 v = qobject_input_visitor_new_flat_confused(options, errp);
568 visit_type_BlockdevOptionsNfs(v, NULL, &opts, &local_err);
572 error_propagate(errp, local_err);
576 /* Remove the processed options from the QDict (the visitor processes
577 * _all_ options in the QDict) */
578 while ((e = qdict_first(options))) {
579 qdict_del(options, e->key);
585 static int64_t nfs_client_open_qdict(NFSClient *client, QDict *options,
586 int flags, int open_flags, Error **errp)
588 BlockdevOptionsNfs *opts;
591 opts = nfs_options_qdict_to_qapi(options, errp);
597 ret = nfs_client_open(client, opts, flags, open_flags, errp);
599 qapi_free_BlockdevOptionsNfs(opts);
603 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
605 NFSClient *client = bs->opaque;
608 client->aio_context = bdrv_get_aio_context(bs);
610 ret = nfs_client_open_qdict(client, options,
611 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
612 bs->open_flags, errp);
617 bs->total_sectors = ret;
622 static QemuOptsList nfs_create_opts = {
623 .name = "nfs-create-opts",
624 .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
627 .name = BLOCK_OPT_SIZE,
628 .type = QEMU_OPT_SIZE,
629 .help = "Virtual disk size"
631 { /* end of list */ }
635 static int nfs_file_co_create(BlockdevCreateOptions *options, Error **errp)
637 BlockdevCreateOptionsNfs *opts = &options->u.nfs;
638 NFSClient *client = g_new0(NFSClient, 1);
641 assert(options->driver == BLOCKDEV_DRIVER_NFS);
643 client->aio_context = qemu_get_aio_context();
645 ret = nfs_client_open(client, opts->location, O_CREAT, 0, errp);
649 ret = nfs_ftruncate(client->context, client->fh, opts->size);
650 nfs_client_close(client);
657 static int coroutine_fn nfs_file_co_create_opts(const char *url, QemuOpts *opts,
660 BlockdevCreateOptions *create_options;
661 BlockdevCreateOptionsNfs *nfs_opts;
665 create_options = g_new0(BlockdevCreateOptions, 1);
666 create_options->driver = BLOCKDEV_DRIVER_NFS;
667 nfs_opts = &create_options->u.nfs;
669 /* Read out options */
670 nfs_opts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
673 options = qdict_new();
674 ret = nfs_parse_uri(url, options, errp);
679 nfs_opts->location = nfs_options_qdict_to_qapi(options, errp);
680 if (nfs_opts->location == NULL) {
685 ret = nfs_file_co_create(create_options, errp);
692 qobject_unref(options);
693 qapi_free_BlockdevCreateOptions(create_options);
697 static int nfs_has_zero_init(BlockDriverState *bs)
699 NFSClient *client = bs->opaque;
700 return client->has_zero_init;
703 /* Called (via nfs_service) with QemuMutex held. */
705 nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
708 NFSRPC *task = private_data;
710 if (task->ret == 0) {
711 memcpy(task->st, data, sizeof(struct stat));
714 error_report("NFS Error: %s", nfs_get_error(nfs));
717 /* Set task->complete before reading bs->wakeup. */
718 atomic_mb_set(&task->complete, 1);
719 bdrv_wakeup(task->bs);
722 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
724 NFSClient *client = bs->opaque;
728 if (bdrv_is_read_only(bs) &&
729 !(bs->open_flags & BDRV_O_NOCACHE)) {
730 return client->st_blocks * 512;
735 if (nfs_fstat_async(client->context, client->fh, nfs_get_allocated_file_size_cb,
740 nfs_set_events(client);
741 BDRV_POLL_WHILE(bs, !task.complete);
743 return (task.ret < 0 ? task.ret : st.st_blocks * 512);
746 static int coroutine_fn
747 nfs_file_co_truncate(BlockDriverState *bs, int64_t offset,
748 PreallocMode prealloc, Error **errp)
750 NFSClient *client = bs->opaque;
753 if (prealloc != PREALLOC_MODE_OFF) {
754 error_setg(errp, "Unsupported preallocation mode '%s'",
755 PreallocMode_str(prealloc));
759 ret = nfs_ftruncate(client->context, client->fh, offset);
761 error_setg_errno(errp, -ret, "Failed to truncate file");
768 /* Note that this will not re-establish a connection with the NFS server
769 * - it is effectively a NOP. */
770 static int nfs_reopen_prepare(BDRVReopenState *state,
771 BlockReopenQueue *queue, Error **errp)
773 NFSClient *client = state->bs->opaque;
777 if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
778 error_setg(errp, "Cannot open a read-only mount as read-write");
782 if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
783 error_setg(errp, "Cannot disable cache if libnfs readahead or"
784 " pagecache is enabled");
788 /* Update cache for read-only reopens */
789 if (!(state->flags & BDRV_O_RDWR)) {
790 ret = nfs_fstat(client->context, client->fh, &st);
792 error_setg(errp, "Failed to fstat file: %s",
793 nfs_get_error(client->context));
796 client->st_blocks = st.st_blocks;
802 static void nfs_refresh_filename(BlockDriverState *bs, QDict *options)
804 NFSClient *client = bs->opaque;
805 QDict *opts = qdict_new();
806 QObject *server_qdict;
809 qdict_put_str(opts, "driver", "nfs");
811 if (client->uid && !client->gid) {
812 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
813 "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
815 } else if (!client->uid && client->gid) {
816 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
817 "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
819 } else if (client->uid && client->gid) {
820 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
821 "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
822 client->server->host, client->path, client->uid, client->gid);
824 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
825 "nfs://%s%s", client->server->host, client->path);
828 ov = qobject_output_visitor_new(&server_qdict);
829 visit_type_NFSServer(ov, NULL, &client->server, &error_abort);
830 visit_complete(ov, &server_qdict);
831 qdict_put_obj(opts, "server", server_qdict);
832 qdict_put_str(opts, "path", client->path);
835 qdict_put_int(opts, "user", client->uid);
838 qdict_put_int(opts, "group", client->gid);
840 if (client->tcp_syncnt) {
841 qdict_put_int(opts, "tcp-syn-cnt", client->tcp_syncnt);
843 if (client->readahead) {
844 qdict_put_int(opts, "readahead-size", client->readahead);
846 if (client->pagecache) {
847 qdict_put_int(opts, "page-cache-size", client->pagecache);
850 qdict_put_int(opts, "debug", client->debug);
855 bs->full_open_options = opts;
858 #ifdef LIBNFS_FEATURE_PAGECACHE
859 static void coroutine_fn nfs_co_invalidate_cache(BlockDriverState *bs,
862 NFSClient *client = bs->opaque;
863 nfs_pagecache_invalidate(client->context, client->fh);
867 static BlockDriver bdrv_nfs = {
868 .format_name = "nfs",
869 .protocol_name = "nfs",
871 .instance_size = sizeof(NFSClient),
872 .bdrv_parse_filename = nfs_parse_filename,
873 .create_opts = &nfs_create_opts,
875 .bdrv_has_zero_init = nfs_has_zero_init,
876 .bdrv_get_allocated_file_size = nfs_get_allocated_file_size,
877 .bdrv_co_truncate = nfs_file_co_truncate,
879 .bdrv_file_open = nfs_file_open,
880 .bdrv_close = nfs_file_close,
881 .bdrv_co_create = nfs_file_co_create,
882 .bdrv_co_create_opts = nfs_file_co_create_opts,
883 .bdrv_reopen_prepare = nfs_reopen_prepare,
885 .bdrv_co_preadv = nfs_co_preadv,
886 .bdrv_co_pwritev = nfs_co_pwritev,
887 .bdrv_co_flush_to_disk = nfs_co_flush,
889 .bdrv_detach_aio_context = nfs_detach_aio_context,
890 .bdrv_attach_aio_context = nfs_attach_aio_context,
891 .bdrv_refresh_filename = nfs_refresh_filename,
893 #ifdef LIBNFS_FEATURE_PAGECACHE
894 .bdrv_co_invalidate_cache = nfs_co_invalidate_cache,
898 static void nfs_block_init(void)
900 bdrv_register(&bdrv_nfs);
903 block_init(nfs_block_init);