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-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"
36 #include "qemu/cutils.h"
37 #include "sysemu/sysemu.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qint.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi-visit.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;
61 int64_t uid, gid, tcp_syncnt, readahead, pagecache, debug;
64 typedef struct NFSRPC {
74 static int nfs_parse_uri(const char *filename, QDict *options, Error **errp)
77 QueryParams *qp = NULL;
80 uri = uri_parse(filename);
82 error_setg(errp, "Invalid URI specified");
85 if (strcmp(uri->scheme, "nfs") != 0) {
86 error_setg(errp, "URI scheme must be 'nfs'");
91 error_setg(errp, "missing hostname in URI");
96 error_setg(errp, "missing file path in URI");
100 qp = query_params_parse(uri->query);
102 error_setg(errp, "could not parse query parameters");
106 qdict_put(options, "server.host", qstring_from_str(uri->server));
107 qdict_put(options, "server.type", qstring_from_str("inet"));
108 qdict_put(options, "path", qstring_from_str(uri->path));
110 for (i = 0; i < qp->n; i++) {
111 unsigned long long val;
112 if (!qp->p[i].value) {
113 error_setg(errp, "Value for NFS parameter expected: %s",
117 if (parse_uint_full(qp->p[i].value, &val, 0)) {
118 error_setg(errp, "Illegal value for NFS parameter: %s",
122 if (!strcmp(qp->p[i].name, "uid")) {
123 qdict_put(options, "user",
124 qstring_from_str(qp->p[i].value));
125 } else if (!strcmp(qp->p[i].name, "gid")) {
126 qdict_put(options, "group",
127 qstring_from_str(qp->p[i].value));
128 } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
129 qdict_put(options, "tcp-syn-count",
130 qstring_from_str(qp->p[i].value));
131 } else if (!strcmp(qp->p[i].name, "readahead")) {
132 qdict_put(options, "readahead-size",
133 qstring_from_str(qp->p[i].value));
134 } else if (!strcmp(qp->p[i].name, "pagecache")) {
135 qdict_put(options, "page-cache-size",
136 qstring_from_str(qp->p[i].value));
137 } else if (!strcmp(qp->p[i].name, "debug")) {
138 qdict_put(options, "debug",
139 qstring_from_str(qp->p[i].value));
141 error_setg(errp, "Unknown NFS parameter name: %s",
149 query_params_free(qp);
157 static bool nfs_has_filename_options_conflict(QDict *options, Error **errp)
159 const QDictEntry *qe;
161 for (qe = qdict_first(options); qe; qe = qdict_next(options, qe)) {
162 if (!strcmp(qe->key, "host") ||
163 !strcmp(qe->key, "path") ||
164 !strcmp(qe->key, "user") ||
165 !strcmp(qe->key, "group") ||
166 !strcmp(qe->key, "tcp-syn-count") ||
167 !strcmp(qe->key, "readahead-size") ||
168 !strcmp(qe->key, "page-cache-size") ||
169 !strcmp(qe->key, "debug") ||
170 strstart(qe->key, "server.", NULL))
172 error_setg(errp, "Option %s cannot be used with a filename",
181 static void nfs_parse_filename(const char *filename, QDict *options,
184 if (nfs_has_filename_options_conflict(options, errp)) {
188 nfs_parse_uri(filename, options, errp);
191 static void nfs_process_read(void *arg);
192 static void nfs_process_write(void *arg);
194 static void nfs_set_events(NFSClient *client)
196 int ev = nfs_which_events(client->context);
197 if (ev != client->events) {
198 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
200 (ev & POLLIN) ? nfs_process_read : NULL,
201 (ev & POLLOUT) ? nfs_process_write : NULL,
208 static void nfs_process_read(void *arg)
210 NFSClient *client = arg;
212 aio_context_acquire(client->aio_context);
213 nfs_service(client->context, POLLIN);
214 nfs_set_events(client);
215 aio_context_release(client->aio_context);
218 static void nfs_process_write(void *arg)
220 NFSClient *client = arg;
222 aio_context_acquire(client->aio_context);
223 nfs_service(client->context, POLLOUT);
224 nfs_set_events(client);
225 aio_context_release(client->aio_context);
228 static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
231 .co = qemu_coroutine_self(),
233 .client = bs->opaque,
237 static void nfs_co_generic_bh_cb(void *opaque)
239 NFSRPC *task = opaque;
242 aio_co_wake(task->co);
246 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
249 NFSRPC *task = private_data;
252 if (task->ret > 0 && task->iov) {
253 if (task->ret <= task->iov->size) {
254 qemu_iovec_from_buf(task->iov, 0, data, task->ret);
260 error_report("NFS Error: %s", nfs_get_error(nfs));
262 aio_bh_schedule_oneshot(task->client->aio_context,
263 nfs_co_generic_bh_cb, task);
266 static int coroutine_fn nfs_co_preadv(BlockDriverState *bs, uint64_t offset,
267 uint64_t bytes, QEMUIOVector *iov,
270 NFSClient *client = bs->opaque;
273 nfs_co_init_task(bs, &task);
276 if (nfs_pread_async(client->context, client->fh,
277 offset, bytes, nfs_co_generic_cb, &task) != 0) {
281 nfs_set_events(client);
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 if (nfs_pwrite_async(client->context, client->fh,
322 nfs_co_generic_cb, &task) != 0) {
329 nfs_set_events(client);
330 while (!task.complete) {
331 qemu_coroutine_yield();
338 if (task.ret != bytes) {
339 return task.ret < 0 ? task.ret : -EIO;
345 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
347 NFSClient *client = bs->opaque;
350 nfs_co_init_task(bs, &task);
352 if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
357 nfs_set_events(client);
358 while (!task.complete) {
359 qemu_coroutine_yield();
365 static QemuOptsList runtime_opts = {
367 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
371 .type = QEMU_OPT_STRING,
372 .help = "Path of the image on the host",
376 .type = QEMU_OPT_NUMBER,
377 .help = "UID value to use when talking to the server",
381 .type = QEMU_OPT_NUMBER,
382 .help = "GID value to use when talking to the server",
385 .name = "tcp-syn-count",
386 .type = QEMU_OPT_NUMBER,
387 .help = "Number of SYNs to send during the session establish",
390 .name = "readahead-size",
391 .type = QEMU_OPT_NUMBER,
392 .help = "Set the readahead size in bytes",
395 .name = "page-cache-size",
396 .type = QEMU_OPT_NUMBER,
397 .help = "Set the pagecache size in bytes",
401 .type = QEMU_OPT_NUMBER,
402 .help = "Set the NFS debug level (max 2)",
404 { /* end of list */ }
408 static void nfs_detach_aio_context(BlockDriverState *bs)
410 NFSClient *client = bs->opaque;
412 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
413 false, NULL, NULL, NULL, NULL);
417 static void nfs_attach_aio_context(BlockDriverState *bs,
418 AioContext *new_context)
420 NFSClient *client = bs->opaque;
422 client->aio_context = new_context;
423 nfs_set_events(client);
426 static void nfs_client_close(NFSClient *client)
428 if (client->context) {
430 nfs_close(client->context, client->fh);
432 aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
433 false, NULL, NULL, NULL, NULL);
434 nfs_destroy_context(client->context);
436 memset(client, 0, sizeof(NFSClient));
439 static void nfs_file_close(BlockDriverState *bs)
441 NFSClient *client = bs->opaque;
442 nfs_client_close(client);
445 static NFSServer *nfs_config(QDict *options, Error **errp)
447 NFSServer *server = NULL;
449 QObject *crumpled_addr = NULL;
451 Error *local_error = NULL;
453 qdict_extract_subqdict(options, &addr, "server.");
454 if (!qdict_size(addr)) {
455 error_setg(errp, "NFS server address missing");
459 crumpled_addr = qdict_crumple(addr, errp);
460 if (!crumpled_addr) {
464 iv = qobject_input_visitor_new(crumpled_addr, true);
465 visit_type_NFSServer(iv, NULL, &server, &local_error);
467 error_propagate(errp, local_error);
473 qobject_decref(crumpled_addr);
479 static int64_t nfs_client_open(NFSClient *client, QDict *options,
480 int flags, Error **errp, int open_flags)
483 QemuOpts *opts = NULL;
484 Error *local_err = NULL;
486 char *file = NULL, *strp = NULL;
488 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
489 qemu_opts_absorb_qdict(opts, options, &local_err);
491 error_propagate(errp, local_err);
496 client->path = g_strdup(qemu_opt_get(opts, "path"));
499 error_setg(errp, "No path was specified");
503 strp = strrchr(client->path, '/');
505 error_setg(errp, "Invalid URL specified");
508 file = g_strdup(strp);
511 /* Pop the config into our state object, Exit if invalid */
512 client->server = nfs_config(options, errp);
513 if (!client->server) {
518 client->context = nfs_init_context();
519 if (client->context == NULL) {
520 error_setg(errp, "Failed to init NFS context");
524 if (qemu_opt_get(opts, "user")) {
525 client->uid = qemu_opt_get_number(opts, "user", 0);
526 nfs_set_uid(client->context, client->uid);
529 if (qemu_opt_get(opts, "group")) {
530 client->gid = qemu_opt_get_number(opts, "group", 0);
531 nfs_set_gid(client->context, client->gid);
534 if (qemu_opt_get(opts, "tcp-syn-count")) {
535 client->tcp_syncnt = qemu_opt_get_number(opts, "tcp-syn-count", 0);
536 nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
539 #ifdef LIBNFS_FEATURE_READAHEAD
540 if (qemu_opt_get(opts, "readahead-size")) {
541 if (open_flags & BDRV_O_NOCACHE) {
542 error_setg(errp, "Cannot enable NFS readahead "
543 "if cache.direct = on");
546 client->readahead = qemu_opt_get_number(opts, "readahead-size", 0);
547 if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
548 error_report("NFS Warning: Truncating NFS readahead "
549 "size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
550 client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
552 nfs_set_readahead(client->context, client->readahead);
553 #ifdef LIBNFS_FEATURE_PAGECACHE
554 nfs_set_pagecache_ttl(client->context, 0);
556 client->cache_used = true;
560 #ifdef LIBNFS_FEATURE_PAGECACHE
561 if (qemu_opt_get(opts, "page-cache-size")) {
562 if (open_flags & BDRV_O_NOCACHE) {
563 error_setg(errp, "Cannot enable NFS pagecache "
564 "if cache.direct = on");
567 client->pagecache = qemu_opt_get_number(opts, "page-cache-size", 0);
568 if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
569 error_report("NFS Warning: Truncating NFS pagecache "
570 "size to %d pages", QEMU_NFS_MAX_PAGECACHE_SIZE);
571 client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
573 nfs_set_pagecache(client->context, client->pagecache);
574 nfs_set_pagecache_ttl(client->context, 0);
575 client->cache_used = true;
579 #ifdef LIBNFS_FEATURE_DEBUG
580 if (qemu_opt_get(opts, "debug")) {
581 client->debug = qemu_opt_get_number(opts, "debug", 0);
582 /* limit the maximum debug level to avoid potential flooding
583 * of our log files. */
584 if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
585 error_report("NFS Warning: Limiting NFS debug level "
586 "to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
587 client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
589 nfs_set_debug(client->context, client->debug);
593 ret = nfs_mount(client->context, client->server->host, client->path);
595 error_setg(errp, "Failed to mount nfs share: %s",
596 nfs_get_error(client->context));
600 if (flags & O_CREAT) {
601 ret = nfs_creat(client->context, file, 0600, &client->fh);
603 error_setg(errp, "Failed to create file: %s",
604 nfs_get_error(client->context));
608 ret = nfs_open(client->context, file, flags, &client->fh);
610 error_setg(errp, "Failed to open file : %s",
611 nfs_get_error(client->context));
616 ret = nfs_fstat(client->context, client->fh, &st);
618 error_setg(errp, "Failed to fstat file: %s",
619 nfs_get_error(client->context));
623 ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
624 client->st_blocks = st.st_blocks;
625 client->has_zero_init = S_ISREG(st.st_mode);
630 nfs_client_close(client);
637 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
639 NFSClient *client = bs->opaque;
642 client->aio_context = bdrv_get_aio_context(bs);
644 ret = nfs_client_open(client, options,
645 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
646 errp, bs->open_flags);
650 bs->total_sectors = ret;
655 static QemuOptsList nfs_create_opts = {
656 .name = "nfs-create-opts",
657 .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
660 .name = BLOCK_OPT_SIZE,
661 .type = QEMU_OPT_SIZE,
662 .help = "Virtual disk size"
664 { /* end of list */ }
668 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
671 int64_t total_size = 0;
672 NFSClient *client = g_new0(NFSClient, 1);
673 QDict *options = NULL;
675 client->aio_context = qemu_get_aio_context();
677 /* Read out options */
678 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
681 options = qdict_new();
682 ret = nfs_parse_uri(url, options, errp);
687 ret = nfs_client_open(client, options, O_CREAT, errp, 0);
691 ret = nfs_ftruncate(client->context, client->fh, total_size);
692 nfs_client_close(client);
699 static int nfs_has_zero_init(BlockDriverState *bs)
701 NFSClient *client = bs->opaque;
702 return client->has_zero_init;
706 nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
709 NFSRPC *task = private_data;
711 if (task->ret == 0) {
712 memcpy(task->st, data, sizeof(struct stat));
715 error_report("NFS Error: %s", nfs_get_error(nfs));
718 bdrv_wakeup(task->bs);
721 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
723 NFSClient *client = bs->opaque;
727 if (bdrv_is_read_only(bs) &&
728 !(bs->open_flags & BDRV_O_NOCACHE)) {
729 return client->st_blocks * 512;
734 if (nfs_fstat_async(client->context, client->fh, nfs_get_allocated_file_size_cb,
739 nfs_set_events(client);
740 BDRV_POLL_WHILE(bs, !task.complete);
742 return (task.ret < 0 ? task.ret : st.st_blocks * 512);
745 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
747 NFSClient *client = bs->opaque;
748 return nfs_ftruncate(client->context, client->fh, offset);
751 /* Note that this will not re-establish a connection with the NFS server
752 * - it is effectively a NOP. */
753 static int nfs_reopen_prepare(BDRVReopenState *state,
754 BlockReopenQueue *queue, Error **errp)
756 NFSClient *client = state->bs->opaque;
760 if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
761 error_setg(errp, "Cannot open a read-only mount as read-write");
765 if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
766 error_setg(errp, "Cannot disable cache if libnfs readahead or"
767 " pagecache is enabled");
771 /* Update cache for read-only reopens */
772 if (!(state->flags & BDRV_O_RDWR)) {
773 ret = nfs_fstat(client->context, client->fh, &st);
775 error_setg(errp, "Failed to fstat file: %s",
776 nfs_get_error(client->context));
779 client->st_blocks = st.st_blocks;
785 static void nfs_refresh_filename(BlockDriverState *bs, QDict *options)
787 NFSClient *client = bs->opaque;
788 QDict *opts = qdict_new();
789 QObject *server_qdict;
792 qdict_put(opts, "driver", qstring_from_str("nfs"));
794 if (client->uid && !client->gid) {
795 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
796 "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
798 } else if (!client->uid && client->gid) {
799 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
800 "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
802 } else if (client->uid && client->gid) {
803 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
804 "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
805 client->server->host, client->path, client->uid, client->gid);
807 snprintf(bs->exact_filename, sizeof(bs->exact_filename),
808 "nfs://%s%s", client->server->host, client->path);
811 ov = qobject_output_visitor_new(&server_qdict);
812 visit_type_NFSServer(ov, NULL, &client->server, &error_abort);
813 visit_complete(ov, &server_qdict);
814 qdict_put_obj(opts, "server", server_qdict);
815 qdict_put(opts, "path", qstring_from_str(client->path));
818 qdict_put(opts, "user", qint_from_int(client->uid));
821 qdict_put(opts, "group", qint_from_int(client->gid));
823 if (client->tcp_syncnt) {
824 qdict_put(opts, "tcp-syn-cnt",
825 qint_from_int(client->tcp_syncnt));
827 if (client->readahead) {
828 qdict_put(opts, "readahead-size",
829 qint_from_int(client->readahead));
831 if (client->pagecache) {
832 qdict_put(opts, "page-cache-size",
833 qint_from_int(client->pagecache));
836 qdict_put(opts, "debug", qint_from_int(client->debug));
841 bs->full_open_options = opts;
844 #ifdef LIBNFS_FEATURE_PAGECACHE
845 static void nfs_invalidate_cache(BlockDriverState *bs,
848 NFSClient *client = bs->opaque;
849 nfs_pagecache_invalidate(client->context, client->fh);
853 static BlockDriver bdrv_nfs = {
854 .format_name = "nfs",
855 .protocol_name = "nfs",
857 .instance_size = sizeof(NFSClient),
858 .bdrv_parse_filename = nfs_parse_filename,
859 .create_opts = &nfs_create_opts,
861 .bdrv_has_zero_init = nfs_has_zero_init,
862 .bdrv_get_allocated_file_size = nfs_get_allocated_file_size,
863 .bdrv_truncate = nfs_file_truncate,
865 .bdrv_file_open = nfs_file_open,
866 .bdrv_close = nfs_file_close,
867 .bdrv_create = nfs_file_create,
868 .bdrv_reopen_prepare = nfs_reopen_prepare,
870 .bdrv_co_preadv = nfs_co_preadv,
871 .bdrv_co_pwritev = nfs_co_pwritev,
872 .bdrv_co_flush_to_disk = nfs_co_flush,
874 .bdrv_detach_aio_context = nfs_detach_aio_context,
875 .bdrv_attach_aio_context = nfs_attach_aio_context,
876 .bdrv_refresh_filename = nfs_refresh_filename,
878 #ifdef LIBNFS_FEATURE_PAGECACHE
879 .bdrv_invalidate_cache = nfs_invalidate_cache,
883 static void nfs_block_init(void)
885 bdrv_register(&bdrv_nfs);
888 block_init(nfs_block_init);