xref: /openbmc/qemu/block/nfs.c (revision 61a67f71)
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 "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi-visit.h"
41 #include "qapi/qobject-input-visitor.h"
42 #include "qapi/qobject-output-visitor.h"
43 #include <nfsc/libnfs.h>
44 
45 
46 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
47 #define QEMU_NFS_MAX_PAGECACHE_SIZE (8388608 / NFS_BLKSIZE)
48 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
49 
50 typedef struct NFSClient {
51     struct nfs_context *context;
52     struct nfsfh *fh;
53     int events;
54     bool has_zero_init;
55     AioContext *aio_context;
56     QemuMutex mutex;
57     blkcnt_t st_blocks;
58     bool cache_used;
59     NFSServer *server;
60     char *path;
61     int64_t uid, gid, tcp_syncnt, readahead, pagecache, debug;
62 } NFSClient;
63 
64 typedef struct NFSRPC {
65     BlockDriverState *bs;
66     int ret;
67     int complete;
68     QEMUIOVector *iov;
69     struct stat *st;
70     Coroutine *co;
71     NFSClient *client;
72 } NFSRPC;
73 
74 static int nfs_parse_uri(const char *filename, QDict *options, Error **errp)
75 {
76     URI *uri = NULL;
77     QueryParams *qp = NULL;
78     int ret = -EINVAL, i;
79 
80     uri = uri_parse(filename);
81     if (!uri) {
82         error_setg(errp, "Invalid URI specified");
83         goto out;
84     }
85     if (g_strcmp0(uri->scheme, "nfs") != 0) {
86         error_setg(errp, "URI scheme must be 'nfs'");
87         goto out;
88     }
89 
90     if (!uri->server) {
91         error_setg(errp, "missing hostname in URI");
92         goto out;
93     }
94 
95     if (!uri->path) {
96         error_setg(errp, "missing file path in URI");
97         goto out;
98     }
99 
100     qp = query_params_parse(uri->query);
101     if (!qp) {
102         error_setg(errp, "could not parse query parameters");
103         goto out;
104     }
105 
106     qdict_put_str(options, "server.host", uri->server);
107     qdict_put_str(options, "server.type", "inet");
108     qdict_put_str(options, "path", uri->path);
109 
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",
114                        qp->p[i].name);
115             goto out;
116         }
117         if (parse_uint_full(qp->p[i].value, &val, 0)) {
118             error_setg(errp, "Illegal value for NFS parameter: %s",
119                        qp->p[i].name);
120             goto out;
121         }
122         if (!strcmp(qp->p[i].name, "uid")) {
123             qdict_put_str(options, "user", qp->p[i].value);
124         } else if (!strcmp(qp->p[i].name, "gid")) {
125             qdict_put_str(options, "group", qp->p[i].value);
126         } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
127             qdict_put_str(options, "tcp-syn-count", qp->p[i].value);
128         } else if (!strcmp(qp->p[i].name, "readahead")) {
129             qdict_put_str(options, "readahead-size", qp->p[i].value);
130         } else if (!strcmp(qp->p[i].name, "pagecache")) {
131             qdict_put_str(options, "page-cache-size", qp->p[i].value);
132         } else if (!strcmp(qp->p[i].name, "debug")) {
133             qdict_put_str(options, "debug", qp->p[i].value);
134         } else {
135             error_setg(errp, "Unknown NFS parameter name: %s",
136                        qp->p[i].name);
137             goto out;
138         }
139     }
140     ret = 0;
141 out:
142     if (qp) {
143         query_params_free(qp);
144     }
145     if (uri) {
146         uri_free(uri);
147     }
148     return ret;
149 }
150 
151 static bool nfs_has_filename_options_conflict(QDict *options, Error **errp)
152 {
153     const QDictEntry *qe;
154 
155     for (qe = qdict_first(options); qe; qe = qdict_next(options, qe)) {
156         if (!strcmp(qe->key, "host") ||
157             !strcmp(qe->key, "path") ||
158             !strcmp(qe->key, "user") ||
159             !strcmp(qe->key, "group") ||
160             !strcmp(qe->key, "tcp-syn-count") ||
161             !strcmp(qe->key, "readahead-size") ||
162             !strcmp(qe->key, "page-cache-size") ||
163             !strcmp(qe->key, "debug") ||
164             strstart(qe->key, "server.", NULL))
165         {
166             error_setg(errp, "Option %s cannot be used with a filename",
167                        qe->key);
168             return true;
169         }
170     }
171 
172     return false;
173 }
174 
175 static void nfs_parse_filename(const char *filename, QDict *options,
176                                Error **errp)
177 {
178     if (nfs_has_filename_options_conflict(options, errp)) {
179         return;
180     }
181 
182     nfs_parse_uri(filename, options, errp);
183 }
184 
185 static void nfs_process_read(void *arg);
186 static void nfs_process_write(void *arg);
187 
188 /* Called with QemuMutex held.  */
189 static void nfs_set_events(NFSClient *client)
190 {
191     int ev = nfs_which_events(client->context);
192     if (ev != client->events) {
193         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
194                            false,
195                            (ev & POLLIN) ? nfs_process_read : NULL,
196                            (ev & POLLOUT) ? nfs_process_write : NULL,
197                            NULL, client);
198 
199     }
200     client->events = ev;
201 }
202 
203 static void nfs_process_read(void *arg)
204 {
205     NFSClient *client = arg;
206 
207     qemu_mutex_lock(&client->mutex);
208     nfs_service(client->context, POLLIN);
209     nfs_set_events(client);
210     qemu_mutex_unlock(&client->mutex);
211 }
212 
213 static void nfs_process_write(void *arg)
214 {
215     NFSClient *client = arg;
216 
217     qemu_mutex_lock(&client->mutex);
218     nfs_service(client->context, POLLOUT);
219     nfs_set_events(client);
220     qemu_mutex_unlock(&client->mutex);
221 }
222 
223 static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
224 {
225     *task = (NFSRPC) {
226         .co             = qemu_coroutine_self(),
227         .bs             = bs,
228         .client         = bs->opaque,
229     };
230 }
231 
232 static void nfs_co_generic_bh_cb(void *opaque)
233 {
234     NFSRPC *task = opaque;
235 
236     task->complete = 1;
237     aio_co_wake(task->co);
238 }
239 
240 /* Called (via nfs_service) with QemuMutex held.  */
241 static void
242 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
243                   void *private_data)
244 {
245     NFSRPC *task = private_data;
246     task->ret = ret;
247     assert(!task->st);
248     if (task->ret > 0 && task->iov) {
249         if (task->ret <= task->iov->size) {
250             qemu_iovec_from_buf(task->iov, 0, data, task->ret);
251         } else {
252             task->ret = -EIO;
253         }
254     }
255     if (task->ret < 0) {
256         error_report("NFS Error: %s", nfs_get_error(nfs));
257     }
258     aio_bh_schedule_oneshot(task->client->aio_context,
259                             nfs_co_generic_bh_cb, task);
260 }
261 
262 static int coroutine_fn nfs_co_preadv(BlockDriverState *bs, uint64_t offset,
263                                       uint64_t bytes, QEMUIOVector *iov,
264                                       int flags)
265 {
266     NFSClient *client = bs->opaque;
267     NFSRPC task;
268 
269     nfs_co_init_task(bs, &task);
270     task.iov = iov;
271 
272     qemu_mutex_lock(&client->mutex);
273     if (nfs_pread_async(client->context, client->fh,
274                         offset, bytes, nfs_co_generic_cb, &task) != 0) {
275         qemu_mutex_unlock(&client->mutex);
276         return -ENOMEM;
277     }
278 
279     nfs_set_events(client);
280     qemu_mutex_unlock(&client->mutex);
281     while (!task.complete) {
282         qemu_coroutine_yield();
283     }
284 
285     if (task.ret < 0) {
286         return task.ret;
287     }
288 
289     /* zero pad short reads */
290     if (task.ret < iov->size) {
291         qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
292     }
293 
294     return 0;
295 }
296 
297 static int coroutine_fn nfs_co_pwritev(BlockDriverState *bs, uint64_t offset,
298                                        uint64_t bytes, QEMUIOVector *iov,
299                                        int flags)
300 {
301     NFSClient *client = bs->opaque;
302     NFSRPC task;
303     char *buf = NULL;
304     bool my_buffer = false;
305 
306     nfs_co_init_task(bs, &task);
307 
308     if (iov->niov != 1) {
309         buf = g_try_malloc(bytes);
310         if (bytes && buf == NULL) {
311             return -ENOMEM;
312         }
313         qemu_iovec_to_buf(iov, 0, buf, bytes);
314         my_buffer = true;
315     } else {
316         buf = iov->iov[0].iov_base;
317     }
318 
319     qemu_mutex_lock(&client->mutex);
320     if (nfs_pwrite_async(client->context, client->fh,
321                          offset, bytes, buf,
322                          nfs_co_generic_cb, &task) != 0) {
323         qemu_mutex_unlock(&client->mutex);
324         if (my_buffer) {
325             g_free(buf);
326         }
327         return -ENOMEM;
328     }
329 
330     nfs_set_events(client);
331     qemu_mutex_unlock(&client->mutex);
332     while (!task.complete) {
333         qemu_coroutine_yield();
334     }
335 
336     if (my_buffer) {
337         g_free(buf);
338     }
339 
340     if (task.ret != bytes) {
341         return task.ret < 0 ? task.ret : -EIO;
342     }
343 
344     return 0;
345 }
346 
347 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
348 {
349     NFSClient *client = bs->opaque;
350     NFSRPC task;
351 
352     nfs_co_init_task(bs, &task);
353 
354     qemu_mutex_lock(&client->mutex);
355     if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
356                         &task) != 0) {
357         qemu_mutex_unlock(&client->mutex);
358         return -ENOMEM;
359     }
360 
361     nfs_set_events(client);
362     qemu_mutex_unlock(&client->mutex);
363     while (!task.complete) {
364         qemu_coroutine_yield();
365     }
366 
367     return task.ret;
368 }
369 
370 static QemuOptsList runtime_opts = {
371     .name = "nfs",
372     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
373     .desc = {
374         {
375             .name = "path",
376             .type = QEMU_OPT_STRING,
377             .help = "Path of the image on the host",
378         },
379         {
380             .name = "user",
381             .type = QEMU_OPT_NUMBER,
382             .help = "UID value to use when talking to the server",
383         },
384         {
385             .name = "group",
386             .type = QEMU_OPT_NUMBER,
387             .help = "GID value to use when talking to the server",
388         },
389         {
390             .name = "tcp-syn-count",
391             .type = QEMU_OPT_NUMBER,
392             .help = "Number of SYNs to send during the session establish",
393         },
394         {
395             .name = "readahead-size",
396             .type = QEMU_OPT_NUMBER,
397             .help = "Set the readahead size in bytes",
398         },
399         {
400             .name = "page-cache-size",
401             .type = QEMU_OPT_NUMBER,
402             .help = "Set the pagecache size in bytes",
403         },
404         {
405             .name = "debug",
406             .type = QEMU_OPT_NUMBER,
407             .help = "Set the NFS debug level (max 2)",
408         },
409         { /* end of list */ }
410     },
411 };
412 
413 static void nfs_detach_aio_context(BlockDriverState *bs)
414 {
415     NFSClient *client = bs->opaque;
416 
417     aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
418                        false, NULL, NULL, NULL, NULL);
419     client->events = 0;
420 }
421 
422 static void nfs_attach_aio_context(BlockDriverState *bs,
423                                    AioContext *new_context)
424 {
425     NFSClient *client = bs->opaque;
426 
427     client->aio_context = new_context;
428     nfs_set_events(client);
429 }
430 
431 static void nfs_client_close(NFSClient *client)
432 {
433     if (client->context) {
434         if (client->fh) {
435             nfs_close(client->context, client->fh);
436         }
437         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
438                            false, NULL, NULL, NULL, NULL);
439         nfs_destroy_context(client->context);
440     }
441     memset(client, 0, sizeof(NFSClient));
442 }
443 
444 static void nfs_file_close(BlockDriverState *bs)
445 {
446     NFSClient *client = bs->opaque;
447     nfs_client_close(client);
448     qemu_mutex_destroy(&client->mutex);
449 }
450 
451 static NFSServer *nfs_config(QDict *options, Error **errp)
452 {
453     NFSServer *server = NULL;
454     QDict *addr = NULL;
455     QObject *crumpled_addr = NULL;
456     Visitor *iv = NULL;
457     Error *local_error = NULL;
458 
459     qdict_extract_subqdict(options, &addr, "server.");
460     if (!qdict_size(addr)) {
461         error_setg(errp, "NFS server address missing");
462         goto out;
463     }
464 
465     crumpled_addr = qdict_crumple(addr, errp);
466     if (!crumpled_addr) {
467         goto out;
468     }
469 
470     /*
471      * Caution: this works only because all scalar members of
472      * NFSServer are QString in @crumpled_addr.  The visitor expects
473      * @crumpled_addr to be typed according to the QAPI schema.  It
474      * is when @options come from -blockdev or blockdev_add.  But when
475      * they come from -drive, they're all QString.
476      */
477     iv = qobject_input_visitor_new(crumpled_addr);
478     visit_type_NFSServer(iv, NULL, &server, &local_error);
479     if (local_error) {
480         error_propagate(errp, local_error);
481         goto out;
482     }
483 
484 out:
485     QDECREF(addr);
486     qobject_decref(crumpled_addr);
487     visit_free(iv);
488     return server;
489 }
490 
491 
492 static int64_t nfs_client_open(NFSClient *client, QDict *options,
493                                int flags, int open_flags, Error **errp)
494 {
495     int ret = -EINVAL;
496     QemuOpts *opts = NULL;
497     Error *local_err = NULL;
498     struct stat st;
499     char *file = NULL, *strp = NULL;
500 
501     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
502     qemu_opts_absorb_qdict(opts, options, &local_err);
503     if (local_err) {
504         error_propagate(errp, local_err);
505         ret = -EINVAL;
506         goto fail;
507     }
508 
509     client->path = g_strdup(qemu_opt_get(opts, "path"));
510     if (!client->path) {
511         ret = -EINVAL;
512         error_setg(errp, "No path was specified");
513         goto fail;
514     }
515 
516     strp = strrchr(client->path, '/');
517     if (strp == NULL) {
518         error_setg(errp, "Invalid URL specified");
519         goto fail;
520     }
521     file = g_strdup(strp);
522     *strp = 0;
523 
524     /* Pop the config into our state object, Exit if invalid */
525     client->server = nfs_config(options, errp);
526     if (!client->server) {
527         ret = -EINVAL;
528         goto fail;
529     }
530 
531     client->context = nfs_init_context();
532     if (client->context == NULL) {
533         error_setg(errp, "Failed to init NFS context");
534         goto fail;
535     }
536 
537     if (qemu_opt_get(opts, "user")) {
538         client->uid = qemu_opt_get_number(opts, "user", 0);
539         nfs_set_uid(client->context, client->uid);
540     }
541 
542     if (qemu_opt_get(opts, "group")) {
543         client->gid = qemu_opt_get_number(opts, "group", 0);
544         nfs_set_gid(client->context, client->gid);
545     }
546 
547     if (qemu_opt_get(opts, "tcp-syn-count")) {
548         client->tcp_syncnt = qemu_opt_get_number(opts, "tcp-syn-count", 0);
549         nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
550     }
551 
552 #ifdef LIBNFS_FEATURE_READAHEAD
553     if (qemu_opt_get(opts, "readahead-size")) {
554         if (open_flags & BDRV_O_NOCACHE) {
555             error_setg(errp, "Cannot enable NFS readahead "
556                              "if cache.direct = on");
557             goto fail;
558         }
559         client->readahead = qemu_opt_get_number(opts, "readahead-size", 0);
560         if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
561             warn_report("Truncating NFS readahead size to %d",
562                         QEMU_NFS_MAX_READAHEAD_SIZE);
563             client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
564         }
565         nfs_set_readahead(client->context, client->readahead);
566 #ifdef LIBNFS_FEATURE_PAGECACHE
567         nfs_set_pagecache_ttl(client->context, 0);
568 #endif
569         client->cache_used = true;
570     }
571 #endif
572 
573 #ifdef LIBNFS_FEATURE_PAGECACHE
574     if (qemu_opt_get(opts, "page-cache-size")) {
575         if (open_flags & BDRV_O_NOCACHE) {
576             error_setg(errp, "Cannot enable NFS pagecache "
577                              "if cache.direct = on");
578             goto fail;
579         }
580         client->pagecache = qemu_opt_get_number(opts, "page-cache-size", 0);
581         if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
582             warn_report("Truncating NFS pagecache size to %d pages",
583                         QEMU_NFS_MAX_PAGECACHE_SIZE);
584             client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
585         }
586         nfs_set_pagecache(client->context, client->pagecache);
587         nfs_set_pagecache_ttl(client->context, 0);
588         client->cache_used = true;
589     }
590 #endif
591 
592 #ifdef LIBNFS_FEATURE_DEBUG
593     if (qemu_opt_get(opts, "debug")) {
594         client->debug = qemu_opt_get_number(opts, "debug", 0);
595         /* limit the maximum debug level to avoid potential flooding
596          * of our log files. */
597         if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
598             warn_report("Limiting NFS debug level to %d",
599                         QEMU_NFS_MAX_DEBUG_LEVEL);
600             client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
601         }
602         nfs_set_debug(client->context, client->debug);
603     }
604 #endif
605 
606     ret = nfs_mount(client->context, client->server->host, client->path);
607     if (ret < 0) {
608         error_setg(errp, "Failed to mount nfs share: %s",
609                    nfs_get_error(client->context));
610         goto fail;
611     }
612 
613     if (flags & O_CREAT) {
614         ret = nfs_creat(client->context, file, 0600, &client->fh);
615         if (ret < 0) {
616             error_setg(errp, "Failed to create file: %s",
617                        nfs_get_error(client->context));
618             goto fail;
619         }
620     } else {
621         ret = nfs_open(client->context, file, flags, &client->fh);
622         if (ret < 0) {
623             error_setg(errp, "Failed to open file : %s",
624                        nfs_get_error(client->context));
625             goto fail;
626         }
627     }
628 
629     ret = nfs_fstat(client->context, client->fh, &st);
630     if (ret < 0) {
631         error_setg(errp, "Failed to fstat file: %s",
632                    nfs_get_error(client->context));
633         goto fail;
634     }
635 
636     ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
637     client->st_blocks = st.st_blocks;
638     client->has_zero_init = S_ISREG(st.st_mode);
639     *strp = '/';
640     goto out;
641 
642 fail:
643     nfs_client_close(client);
644 out:
645     qemu_opts_del(opts);
646     g_free(file);
647     return ret;
648 }
649 
650 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
651                          Error **errp) {
652     NFSClient *client = bs->opaque;
653     int64_t ret;
654 
655     client->aio_context = bdrv_get_aio_context(bs);
656 
657     ret = nfs_client_open(client, options,
658                           (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
659                           bs->open_flags, errp);
660     if (ret < 0) {
661         return ret;
662     }
663     qemu_mutex_init(&client->mutex);
664     bs->total_sectors = ret;
665     ret = 0;
666     return ret;
667 }
668 
669 static QemuOptsList nfs_create_opts = {
670     .name = "nfs-create-opts",
671     .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
672     .desc = {
673         {
674             .name = BLOCK_OPT_SIZE,
675             .type = QEMU_OPT_SIZE,
676             .help = "Virtual disk size"
677         },
678         { /* end of list */ }
679     }
680 };
681 
682 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
683 {
684     int ret = 0;
685     int64_t total_size = 0;
686     NFSClient *client = g_new0(NFSClient, 1);
687     QDict *options = NULL;
688 
689     client->aio_context = qemu_get_aio_context();
690 
691     /* Read out options */
692     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
693                           BDRV_SECTOR_SIZE);
694 
695     options = qdict_new();
696     ret = nfs_parse_uri(url, options, errp);
697     if (ret < 0) {
698         goto out;
699     }
700 
701     ret = nfs_client_open(client, options, O_CREAT, 0, errp);
702     if (ret < 0) {
703         goto out;
704     }
705     ret = nfs_ftruncate(client->context, client->fh, total_size);
706     nfs_client_close(client);
707 out:
708     QDECREF(options);
709     g_free(client);
710     return ret;
711 }
712 
713 static int nfs_has_zero_init(BlockDriverState *bs)
714 {
715     NFSClient *client = bs->opaque;
716     return client->has_zero_init;
717 }
718 
719 /* Called (via nfs_service) with QemuMutex held.  */
720 static void
721 nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
722                                void *private_data)
723 {
724     NFSRPC *task = private_data;
725     task->ret = ret;
726     if (task->ret == 0) {
727         memcpy(task->st, data, sizeof(struct stat));
728     }
729     if (task->ret < 0) {
730         error_report("NFS Error: %s", nfs_get_error(nfs));
731     }
732 
733     /* Set task->complete before reading bs->wakeup.  */
734     atomic_mb_set(&task->complete, 1);
735     bdrv_wakeup(task->bs);
736 }
737 
738 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
739 {
740     NFSClient *client = bs->opaque;
741     NFSRPC task = {0};
742     struct stat st;
743 
744     if (bdrv_is_read_only(bs) &&
745         !(bs->open_flags & BDRV_O_NOCACHE)) {
746         return client->st_blocks * 512;
747     }
748 
749     task.bs = bs;
750     task.st = &st;
751     if (nfs_fstat_async(client->context, client->fh, nfs_get_allocated_file_size_cb,
752                         &task) != 0) {
753         return -ENOMEM;
754     }
755 
756     nfs_set_events(client);
757     BDRV_POLL_WHILE(bs, !task.complete);
758 
759     return (task.ret < 0 ? task.ret : st.st_blocks * 512);
760 }
761 
762 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset,
763                              PreallocMode prealloc, Error **errp)
764 {
765     NFSClient *client = bs->opaque;
766     int ret;
767 
768     if (prealloc != PREALLOC_MODE_OFF) {
769         error_setg(errp, "Unsupported preallocation mode '%s'",
770                    PreallocMode_lookup[prealloc]);
771         return -ENOTSUP;
772     }
773 
774     ret = nfs_ftruncate(client->context, client->fh, offset);
775     if (ret < 0) {
776         error_setg_errno(errp, -ret, "Failed to truncate file");
777         return ret;
778     }
779 
780     return 0;
781 }
782 
783 /* Note that this will not re-establish a connection with the NFS server
784  * - it is effectively a NOP.  */
785 static int nfs_reopen_prepare(BDRVReopenState *state,
786                               BlockReopenQueue *queue, Error **errp)
787 {
788     NFSClient *client = state->bs->opaque;
789     struct stat st;
790     int ret = 0;
791 
792     if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
793         error_setg(errp, "Cannot open a read-only mount as read-write");
794         return -EACCES;
795     }
796 
797     if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
798         error_setg(errp, "Cannot disable cache if libnfs readahead or"
799                          " pagecache is enabled");
800         return -EINVAL;
801     }
802 
803     /* Update cache for read-only reopens */
804     if (!(state->flags & BDRV_O_RDWR)) {
805         ret = nfs_fstat(client->context, client->fh, &st);
806         if (ret < 0) {
807             error_setg(errp, "Failed to fstat file: %s",
808                        nfs_get_error(client->context));
809             return ret;
810         }
811         client->st_blocks = st.st_blocks;
812     }
813 
814     return 0;
815 }
816 
817 static void nfs_refresh_filename(BlockDriverState *bs, QDict *options)
818 {
819     NFSClient *client = bs->opaque;
820     QDict *opts = qdict_new();
821     QObject *server_qdict;
822     Visitor *ov;
823 
824     qdict_put_str(opts, "driver", "nfs");
825 
826     if (client->uid && !client->gid) {
827         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
828                  "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
829                  client->uid);
830     } else if (!client->uid && client->gid) {
831         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
832                  "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
833                  client->gid);
834     } else if (client->uid && client->gid) {
835         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
836                  "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
837                  client->server->host, client->path, client->uid, client->gid);
838     } else {
839         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
840                  "nfs://%s%s", client->server->host, client->path);
841     }
842 
843     ov = qobject_output_visitor_new(&server_qdict);
844     visit_type_NFSServer(ov, NULL, &client->server, &error_abort);
845     visit_complete(ov, &server_qdict);
846     qdict_put_obj(opts, "server", server_qdict);
847     qdict_put_str(opts, "path", client->path);
848 
849     if (client->uid) {
850         qdict_put_int(opts, "user", client->uid);
851     }
852     if (client->gid) {
853         qdict_put_int(opts, "group", client->gid);
854     }
855     if (client->tcp_syncnt) {
856         qdict_put_int(opts, "tcp-syn-cnt", client->tcp_syncnt);
857     }
858     if (client->readahead) {
859         qdict_put_int(opts, "readahead-size", client->readahead);
860     }
861     if (client->pagecache) {
862         qdict_put_int(opts, "page-cache-size", client->pagecache);
863     }
864     if (client->debug) {
865         qdict_put_int(opts, "debug", client->debug);
866     }
867 
868     visit_free(ov);
869     qdict_flatten(opts);
870     bs->full_open_options = opts;
871 }
872 
873 #ifdef LIBNFS_FEATURE_PAGECACHE
874 static void nfs_invalidate_cache(BlockDriverState *bs,
875                                  Error **errp)
876 {
877     NFSClient *client = bs->opaque;
878     nfs_pagecache_invalidate(client->context, client->fh);
879 }
880 #endif
881 
882 static BlockDriver bdrv_nfs = {
883     .format_name                    = "nfs",
884     .protocol_name                  = "nfs",
885 
886     .instance_size                  = sizeof(NFSClient),
887     .bdrv_parse_filename            = nfs_parse_filename,
888     .create_opts                    = &nfs_create_opts,
889 
890     .bdrv_has_zero_init             = nfs_has_zero_init,
891     .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
892     .bdrv_truncate                  = nfs_file_truncate,
893 
894     .bdrv_file_open                 = nfs_file_open,
895     .bdrv_close                     = nfs_file_close,
896     .bdrv_create                    = nfs_file_create,
897     .bdrv_reopen_prepare            = nfs_reopen_prepare,
898 
899     .bdrv_co_preadv                 = nfs_co_preadv,
900     .bdrv_co_pwritev                = nfs_co_pwritev,
901     .bdrv_co_flush_to_disk          = nfs_co_flush,
902 
903     .bdrv_detach_aio_context        = nfs_detach_aio_context,
904     .bdrv_attach_aio_context        = nfs_attach_aio_context,
905     .bdrv_refresh_filename          = nfs_refresh_filename,
906 
907 #ifdef LIBNFS_FEATURE_PAGECACHE
908     .bdrv_invalidate_cache          = nfs_invalidate_cache,
909 #endif
910 };
911 
912 static void nfs_block_init(void)
913 {
914     bdrv_register(&bdrv_nfs);
915 }
916 
917 block_init(nfs_block_init);
918