block/nfs.c

   1 /*
   2  * QEMU Block driver for native access to files on NFS shares
   3  *
   4  * Copyright (c) 2014-2017 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 #if !defined(_WIN32)
  28 #include <poll.h>
  29 #endif
  30 #include "qemu/config-file.h"
  31 #include "qemu/error-report.h"
  32 #include "qapi/error.h"
  33 #include "block/block_int.h"
  34 #include "block/qdict.h"
  35 #include "trace.h"
  36 #include "qemu/iov.h"
  37 #include "qemu/main-loop.h"
  38 #include "qemu/module.h"
  39 #include "qemu/option.h"
  40 #include "qemu/uri.h"
  41 #include "qemu/cutils.h"
  42 #include "sysemu/replay.h"
  43 #include "qapi/qapi-visit-block-core.h"
  44 #include "qapi/qmp/qdict.h"
  45 #include "qapi/qmp/qstring.h"
  46 #include "qapi/qobject-input-visitor.h"
  47 #include "qapi/qobject-output-visitor.h"
  48 #include <nfsc/libnfs.h>
  49
  50
  51 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
  52 #define QEMU_NFS_MAX_PAGECACHE_SIZE (8388608 / NFS_BLKSIZE)
  53 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
  54
  55 typedef struct NFSClient {
  56     struct nfs_context *context;
  57     struct nfsfh *fh;
  58     int events;
  59     bool has_zero_init;
  60     AioContext *aio_context;
  61     QemuMutex mutex;
  62     uint64_t st_blocks;
  63     bool cache_used;
  64     NFSServer *server;
  65     char *path;
  66     int64_t uid, gid, tcp_syncnt, readahead, pagecache, debug;
  67 } NFSClient;
  68
  69 typedef struct NFSRPC {
  70     BlockDriverState *bs;
  71     int ret;
  72     int complete;
  73     QEMUIOVector *iov;
  74     struct stat *st;
  75     Coroutine *co;
  76     NFSClient *client;
  77 } NFSRPC;
  78
  79 static int nfs_parse_uri(const char *filename, QDict *options, Error **errp)
  80 {
  81     URI *uri = NULL;
  82     QueryParams *qp = NULL;
  83     int ret = -EINVAL, i;
  84
  85     uri = uri_parse(filename);
  86     if (!uri) {
  87         error_setg(errp, "Invalid URI specified");
  88         goto out;
  89     }
  90     if (g_strcmp0(uri->scheme, "nfs") != 0) {
  91         error_setg(errp, "URI scheme must be 'nfs'");
  92         goto out;
  93     }
  94
  95     if (!uri->server) {
  96         error_setg(errp, "missing hostname in URI");
  97         goto out;
  98     }
  99
 100     if (!uri->path) {
 101         error_setg(errp, "missing file path in URI");
 102         goto out;
 103     }
 104
 105     qp = query_params_parse(uri->query);
 106     if (!qp) {
 107         error_setg(errp, "could not parse query parameters");
 108         goto out;
 109     }
 110
 111     qdict_put_str(options, "server.host", uri->server);
 112     qdict_put_str(options, "server.type", "inet");
 113     qdict_put_str(options, "path", uri->path);
 114
 115     for (i = 0; i < qp->n; i++) {
 116         unsigned long long val;
 117         if (!qp->p[i].value) {
 118             error_setg(errp, "Value for NFS parameter expected: %s",
 119                        qp->p[i].name);
 120             goto out;
 121         }
 122         if (parse_uint_full(qp->p[i].value, &val, 0)) {
 123             error_setg(errp, "Illegal value for NFS parameter: %s",
 124                        qp->p[i].name);
 125             goto out;
 126         }
 127         if (!strcmp(qp->p[i].name, "uid")) {
 128             qdict_put_str(options, "user", qp->p[i].value);
 129         } else if (!strcmp(qp->p[i].name, "gid")) {
 130             qdict_put_str(options, "group", qp->p[i].value);
 131         } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
 132             qdict_put_str(options, "tcp-syn-count", qp->p[i].value);
 133         } else if (!strcmp(qp->p[i].name, "readahead")) {
 134             qdict_put_str(options, "readahead-size", qp->p[i].value);
 135         } else if (!strcmp(qp->p[i].name, "pagecache")) {
 136             qdict_put_str(options, "page-cache-size", qp->p[i].value);
 137         } else if (!strcmp(qp->p[i].name, "debug")) {
 138             qdict_put_str(options, "debug", qp->p[i].value);
 139         } else {
 140             error_setg(errp, "Unknown NFS parameter name: %s",
 141                        qp->p[i].name);
 142             goto out;
 143         }
 144     }
 145     ret = 0;
 146 out:
 147     if (qp) {
 148         query_params_free(qp);
 149     }
 150     if (uri) {
 151         uri_free(uri);
 152     }
 153     return ret;
 154 }
 155
 156 static bool nfs_has_filename_options_conflict(QDict *options, Error **errp)
 157 {
 158     const QDictEntry *qe;
 159
 160     for (qe = qdict_first(options); qe; qe = qdict_next(options, qe)) {
 161         if (!strcmp(qe->key, "host") ||
 162             !strcmp(qe->key, "path") ||
 163             !strcmp(qe->key, "user") ||
 164             !strcmp(qe->key, "group") ||
 165             !strcmp(qe->key, "tcp-syn-count") ||
 166             !strcmp(qe->key, "readahead-size") ||
 167             !strcmp(qe->key, "page-cache-size") ||
 168             !strcmp(qe->key, "debug") ||
 169             strstart(qe->key, "server.", NULL))
 170         {
 171             error_setg(errp, "Option %s cannot be used with a filename",
 172                        qe->key);
 173             return true;
 174         }
 175     }
 176
 177     return false;
 178 }
 179
 180 static void nfs_parse_filename(const char *filename, QDict *options,
 181                                Error **errp)
 182 {
 183     if (nfs_has_filename_options_conflict(options, errp)) {
 184         return;
 185     }
 186
 187     nfs_parse_uri(filename, options, errp);
 188 }
 189
 190 static void nfs_process_read(void *arg);
 191 static void nfs_process_write(void *arg);
 192
 193 /* Called with QemuMutex held.  */
 194 static void nfs_set_events(NFSClient *client)
 195 {
 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),
 199                            false,
 200                            (ev & POLLIN) ? nfs_process_read : NULL,
 201                            (ev & POLLOUT) ? nfs_process_write : NULL,
 202                            NULL, client);
 203
 204     }
 205     client->events = ev;
 206 }
 207
 208 static void nfs_process_read(void *arg)
 209 {
 210     NFSClient *client = arg;
 211
 212     qemu_mutex_lock(&client->mutex);
 213     nfs_service(client->context, POLLIN);
 214     nfs_set_events(client);
 215     qemu_mutex_unlock(&client->mutex);
 216 }
 217
 218 static void nfs_process_write(void *arg)
 219 {
 220     NFSClient *client = arg;
 221
 222     qemu_mutex_lock(&client->mutex);
 223     nfs_service(client->context, POLLOUT);
 224     nfs_set_events(client);
 225     qemu_mutex_unlock(&client->mutex);
 226 }
 227
 228 static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
 229 {
 230     *task = (NFSRPC) {
 231         .co             = qemu_coroutine_self(),
 232         .bs             = bs,
 233         .client         = bs->opaque,
 234     };
 235 }
 236
 237 static void nfs_co_generic_bh_cb(void *opaque)
 238 {
 239     NFSRPC *task = opaque;
 240
 241     task->complete = 1;
 242     aio_co_wake(task->co);
 243 }
 244
 245 /* Called (via nfs_service) with QemuMutex held.  */
 246 static void
 247 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
 248                   void *private_data)
 249 {
 250     NFSRPC *task = private_data;
 251     task->ret = ret;
 252     assert(!task->st);
 253     if (task->ret > 0 && task->iov) {
 254         if (task->ret <= task->iov->size) {
 255             qemu_iovec_from_buf(task->iov, 0, data, task->ret);
 256         } else {
 257             task->ret = -EIO;
 258         }
 259     }
 260     if (task->ret < 0) {
 261         error_report("NFS Error: %s", nfs_get_error(nfs));
 262     }
 263     replay_bh_schedule_oneshot_event(task->client->aio_context,
 264                                      nfs_co_generic_bh_cb, task);
 265 }
 266
 267 static int coroutine_fn nfs_co_preadv(BlockDriverState *bs, uint64_t offset,
 268                                       uint64_t bytes, QEMUIOVector *iov,
 269                                       int flags)
 270 {
 271     NFSClient *client = bs->opaque;
 272     NFSRPC task;
 273
 274     nfs_co_init_task(bs, &task);
 275     task.iov = iov;
 276
 277     WITH_QEMU_LOCK_GUARD(&client->mutex) {
 278         if (nfs_pread_async(client->context, client->fh,
 279                             offset, bytes, nfs_co_generic_cb, &task) != 0) {
 280             return -ENOMEM;
 281         }
 282
 283         nfs_set_events(client);
 284     }
 285     while (!task.complete) {
 286         qemu_coroutine_yield();
 287     }
 288
 289     if (task.ret < 0) {
 290         return task.ret;
 291     }
 292
 293     /* zero pad short reads */
 294     if (task.ret < iov->size) {
 295         qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
 296     }
 297
 298     return 0;
 299 }
 300
 301 static int coroutine_fn nfs_co_pwritev(BlockDriverState *bs, uint64_t offset,
 302                                        uint64_t bytes, QEMUIOVector *iov,
 303                                        int flags)
 304 {
 305     NFSClient *client = bs->opaque;
 306     NFSRPC task;
 307     char *buf = NULL;
 308     bool my_buffer = false;
 309
 310     nfs_co_init_task(bs, &task);
 311
 312     if (iov->niov != 1) {
 313         buf = g_try_malloc(bytes);
 314         if (bytes && buf == NULL) {
 315             return -ENOMEM;
 316         }
 317         qemu_iovec_to_buf(iov, 0, buf, bytes);
 318         my_buffer = true;
 319     } else {
 320         buf = iov->iov[0].iov_base;
 321     }
 322
 323     WITH_QEMU_LOCK_GUARD(&client->mutex) {
 324         if (nfs_pwrite_async(client->context, client->fh,
 325                              offset, bytes, buf,
 326                              nfs_co_generic_cb, &task) != 0) {
 327             if (my_buffer) {
 328                 g_free(buf);
 329             }
 330             return -ENOMEM;
 331         }
 332
 333         nfs_set_events(client);
 334     }
 335     while (!task.complete) {
 336         qemu_coroutine_yield();
 337     }
 338
 339     if (my_buffer) {
 340         g_free(buf);
 341     }
 342
 343     if (task.ret != bytes) {
 344         return task.ret < 0 ? task.ret : -EIO;
 345     }
 346
 347     return 0;
 348 }
 349
 350 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
 351 {
 352     NFSClient *client = bs->opaque;
 353     NFSRPC task;
 354
 355     nfs_co_init_task(bs, &task);
 356
 357     WITH_QEMU_LOCK_GUARD(&client->mutex) {
 358         if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
 359                             &task) != 0) {
 360             return -ENOMEM;
 361         }
 362
 363         nfs_set_events(client);
 364     }
 365     while (!task.complete) {
 366         qemu_coroutine_yield();
 367     }
 368
 369     return task.ret;
 370 }
 371
 372 static void nfs_detach_aio_context(BlockDriverState *bs)
 373 {
 374     NFSClient *client = bs->opaque;
 375
 376     aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 377                        false, NULL, NULL, NULL, NULL);
 378     client->events = 0;
 379 }
 380
 381 static void nfs_attach_aio_context(BlockDriverState *bs,
 382                                    AioContext *new_context)
 383 {
 384     NFSClient *client = bs->opaque;
 385
 386     client->aio_context = new_context;
 387     nfs_set_events(client);
 388 }
 389
 390 static void nfs_client_close(NFSClient *client)
 391 {
 392     if (client->context) {
 393         qemu_mutex_lock(&client->mutex);
 394         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 395                            false, NULL, NULL, NULL, NULL);
 396         qemu_mutex_unlock(&client->mutex);
 397         if (client->fh) {
 398             nfs_close(client->context, client->fh);
 399             client->fh = NULL;
 400         }
 401 #ifdef LIBNFS_FEATURE_UMOUNT
 402         nfs_umount(client->context);
 403 #endif
 404         nfs_destroy_context(client->context);
 405         client->context = NULL;
 406     }
 407     g_free(client->path);
 408     qemu_mutex_destroy(&client->mutex);
 409     qapi_free_NFSServer(client->server);
 410     client->server = NULL;
 411 }
 412
 413 static void nfs_file_close(BlockDriverState *bs)
 414 {
 415     NFSClient *client = bs->opaque;
 416     nfs_client_close(client);
 417 }
 418
 419 static int64_t nfs_client_open(NFSClient *client, BlockdevOptionsNfs *opts,
 420                                int flags, int open_flags, Error **errp)
 421 {
 422     int64_t ret = -EINVAL;
 423     struct stat st;
 424     char *file = NULL, *strp = NULL;
 425
 426     qemu_mutex_init(&client->mutex);
 427
 428     client->path = g_strdup(opts->path);
 429
 430     strp = strrchr(client->path, '/');
 431     if (strp == NULL) {
 432         error_setg(errp, "Invalid URL specified");
 433         goto fail;
 434     }
 435     file = g_strdup(strp);
 436     *strp = 0;
 437
 438     /* Steal the NFSServer object from opts; set the original pointer to NULL
 439      * to avoid use after free and double free. */
 440     client->server = opts->server;
 441     opts->server = NULL;
 442
 443     client->context = nfs_init_context();
 444     if (client->context == NULL) {
 445         error_setg(errp, "Failed to init NFS context");
 446         goto fail;
 447     }
 448
 449     if (opts->has_user) {
 450         client->uid = opts->user;
 451         nfs_set_uid(client->context, client->uid);
 452     }
 453
 454     if (opts->has_group) {
 455         client->gid = opts->group;
 456         nfs_set_gid(client->context, client->gid);
 457     }
 458
 459     if (opts->has_tcp_syn_count) {
 460         client->tcp_syncnt = opts->tcp_syn_count;
 461         nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
 462     }
 463
 464 #ifdef LIBNFS_FEATURE_READAHEAD
 465     if (opts->has_readahead_size) {
 466         if (open_flags & BDRV_O_NOCACHE) {
 467             error_setg(errp, "Cannot enable NFS readahead "
 468                              "if cache.direct = on");
 469             goto fail;
 470         }
 471         client->readahead = opts->readahead_size;
 472         if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
 473             warn_report("Truncating NFS readahead size to %d",
 474                         QEMU_NFS_MAX_READAHEAD_SIZE);
 475             client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
 476         }
 477         nfs_set_readahead(client->context, client->readahead);
 478 #ifdef LIBNFS_FEATURE_PAGECACHE
 479         nfs_set_pagecache_ttl(client->context, 0);
 480 #endif
 481         client->cache_used = true;
 482     }
 483 #endif
 484
 485 #ifdef LIBNFS_FEATURE_PAGECACHE
 486     if (opts->has_page_cache_size) {
 487         if (open_flags & BDRV_O_NOCACHE) {
 488             error_setg(errp, "Cannot enable NFS pagecache "
 489                              "if cache.direct = on");
 490             goto fail;
 491         }
 492         client->pagecache = opts->page_cache_size;
 493         if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
 494             warn_report("Truncating NFS pagecache size to %d pages",
 495                         QEMU_NFS_MAX_PAGECACHE_SIZE);
 496             client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
 497         }
 498         nfs_set_pagecache(client->context, client->pagecache);
 499         nfs_set_pagecache_ttl(client->context, 0);
 500         client->cache_used = true;
 501     }
 502 #endif
 503
 504 #ifdef LIBNFS_FEATURE_DEBUG
 505     if (opts->has_debug) {
 506         client->debug = opts->debug;
 507         /* limit the maximum debug level to avoid potential flooding
 508          * of our log files. */
 509         if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
 510             warn_report("Limiting NFS debug level to %d",
 511                         QEMU_NFS_MAX_DEBUG_LEVEL);
 512             client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
 513         }
 514         nfs_set_debug(client->context, client->debug);
 515     }
 516 #endif
 517
 518     ret = nfs_mount(client->context, client->server->host, client->path);
 519     if (ret < 0) {
 520         error_setg(errp, "Failed to mount nfs share: %s",
 521                    nfs_get_error(client->context));
 522         goto fail;
 523     }
 524
 525     if (flags & O_CREAT) {
 526         ret = nfs_creat(client->context, file, 0600, &client->fh);
 527         if (ret < 0) {
 528             error_setg(errp, "Failed to create file: %s",
 529                        nfs_get_error(client->context));
 530             goto fail;
 531         }
 532     } else {
 533         ret = nfs_open(client->context, file, flags, &client->fh);
 534         if (ret < 0) {
 535             error_setg(errp, "Failed to open file : %s",
 536                        nfs_get_error(client->context));
 537             goto fail;
 538         }
 539     }
 540
 541     ret = nfs_fstat(client->context, client->fh, &st);
 542     if (ret < 0) {
 543         error_setg(errp, "Failed to fstat file: %s",
 544                    nfs_get_error(client->context));
 545         goto fail;
 546     }
 547
 548     ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
 549 #if !defined(_WIN32)
 550     client->st_blocks = st.st_blocks;
 551 #endif
 552     client->has_zero_init = S_ISREG(st.st_mode);
 553     *strp = '/';
 554     goto out;
 555
 556 fail:
 557     nfs_client_close(client);
 558 out:
 559     g_free(file);
 560     return ret;
 561 }
 562
 563 static BlockdevOptionsNfs *nfs_options_qdict_to_qapi(QDict *options,
 564                                                      Error **errp)
 565 {
 566     BlockdevOptionsNfs *opts = NULL;
 567     Visitor *v;
 568     const QDictEntry *e;
 569
 570     v = qobject_input_visitor_new_flat_confused(options, errp);
 571     if (!v) {
 572         return NULL;
 573     }
 574
 575     visit_type_BlockdevOptionsNfs(v, NULL, &opts, errp);
 576     visit_free(v);
 577     if (!opts) {
 578         return NULL;
 579     }
 580
 581     /* Remove the processed options from the QDict (the visitor processes
 582      * _all_ options in the QDict) */
 583     while ((e = qdict_first(options))) {
 584         qdict_del(options, e->key);
 585     }
 586
 587     return opts;
 588 }
 589
 590 static int64_t nfs_client_open_qdict(NFSClient *client, QDict *options,
 591                                      int flags, int open_flags, Error **errp)
 592 {
 593     BlockdevOptionsNfs *opts;
 594     int64_t ret;
 595
 596     opts = nfs_options_qdict_to_qapi(options, errp);
 597     if (opts == NULL) {
 598         ret = -EINVAL;
 599         goto fail;
 600     }
 601
 602     ret = nfs_client_open(client, opts, flags, open_flags, errp);
 603 fail:
 604     qapi_free_BlockdevOptionsNfs(opts);
 605     return ret;
 606 }
 607
 608 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
 609                          Error **errp) {
 610     NFSClient *client = bs->opaque;
 611     int64_t ret;
 612
 613     client->aio_context = bdrv_get_aio_context(bs);
 614
 615     ret = nfs_client_open_qdict(client, options,
 616                                 (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
 617                                 bs->open_flags, errp);
 618     if (ret < 0) {
 619         return ret;
 620     }
 621
 622     bs->total_sectors = ret;
 623     if (client->has_zero_init) {
 624         bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
 625     }
 626     return 0;
 627 }
 628
 629 static QemuOptsList nfs_create_opts = {
 630     .name = "nfs-create-opts",
 631     .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
 632     .desc = {
 633         {
 634             .name = BLOCK_OPT_SIZE,
 635             .type = QEMU_OPT_SIZE,
 636             .help = "Virtual disk size"
 637         },
 638         { /* end of list */ }
 639     }
 640 };
 641
 642 static int nfs_file_co_create(BlockdevCreateOptions *options, Error **errp)
 643 {
 644     BlockdevCreateOptionsNfs *opts = &options->u.nfs;
 645     NFSClient *client = g_new0(NFSClient, 1);
 646     int ret;
 647
 648     assert(options->driver == BLOCKDEV_DRIVER_NFS);
 649
 650     client->aio_context = qemu_get_aio_context();
 651
 652     ret = nfs_client_open(client, opts->location, O_CREAT, 0, errp);
 653     if (ret < 0) {
 654         goto out;
 655     }
 656     ret = nfs_ftruncate(client->context, client->fh, opts->size);
 657     nfs_client_close(client);
 658
 659 out:
 660     g_free(client);
 661     return ret;
 662 }
 663
 664 static int coroutine_fn nfs_file_co_create_opts(BlockDriver *drv,
 665                                                 const char *url,
 666                                                 QemuOpts *opts,
 667                                                 Error **errp)
 668 {
 669     BlockdevCreateOptions *create_options;
 670     BlockdevCreateOptionsNfs *nfs_opts;
 671     QDict *options;
 672     int ret;
 673
 674     create_options = g_new0(BlockdevCreateOptions, 1);
 675     create_options->driver = BLOCKDEV_DRIVER_NFS;
 676     nfs_opts = &create_options->u.nfs;
 677
 678     /* Read out options */
 679     nfs_opts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 680                               BDRV_SECTOR_SIZE);
 681
 682     options = qdict_new();
 683     ret = nfs_parse_uri(url, options, errp);
 684     if (ret < 0) {
 685         goto out;
 686     }
 687
 688     nfs_opts->location = nfs_options_qdict_to_qapi(options, errp);
 689     if (nfs_opts->location == NULL) {
 690         ret = -EINVAL;
 691         goto out;
 692     }
 693
 694     ret = nfs_file_co_create(create_options, errp);
 695     if (ret < 0) {
 696         goto out;
 697     }
 698
 699     ret = 0;
 700 out:
 701     qobject_unref(options);
 702     qapi_free_BlockdevCreateOptions(create_options);
 703     return ret;
 704 }
 705
 706 static int nfs_has_zero_init(BlockDriverState *bs)
 707 {
 708     NFSClient *client = bs->opaque;
 709     return client->has_zero_init;
 710 }
 711
 712 #if !defined(_WIN32)
 713 /* Called (via nfs_service) with QemuMutex held.  */
 714 static void
 715 nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
 716                                void *private_data)
 717 {
 718     NFSRPC *task = private_data;
 719     task->ret = ret;
 720     if (task->ret == 0) {
 721         memcpy(task->st, data, sizeof(struct stat));
 722     }
 723     if (task->ret < 0) {
 724         error_report("NFS Error: %s", nfs_get_error(nfs));
 725     }
 726
 727     /* Set task->complete before reading bs->wakeup.  */
 728     qatomic_mb_set(&task->complete, 1);
 729     bdrv_wakeup(task->bs);
 730 }
 731
 732 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
 733 {
 734     NFSClient *client = bs->opaque;
 735     NFSRPC task = {0};
 736     struct stat st;
 737
 738     if (bdrv_is_read_only(bs) &&
 739         !(bs->open_flags & BDRV_O_NOCACHE)) {
 740         return client->st_blocks * 512;
 741     }
 742
 743     task.bs = bs;
 744     task.st = &st;
 745     if (nfs_fstat_async(client->context, client->fh, nfs_get_allocated_file_size_cb,
 746                         &task) != 0) {
 747         return -ENOMEM;
 748     }
 749
 750     nfs_set_events(client);
 751     BDRV_POLL_WHILE(bs, !task.complete);
 752
 753     return (task.ret < 0 ? task.ret : st.st_blocks * 512);
 754 }
 755 #endif
 756
 757 static int coroutine_fn
 758 nfs_file_co_truncate(BlockDriverState *bs, int64_t offset, bool exact,
 759                      PreallocMode prealloc, BdrvRequestFlags flags,
 760                      Error **errp)
 761 {
 762     NFSClient *client = bs->opaque;
 763     int ret;
 764
 765     if (prealloc != PREALLOC_MODE_OFF) {
 766         error_setg(errp, "Unsupported preallocation mode '%s'",
 767                    PreallocMode_str(prealloc));
 768         return -ENOTSUP;
 769     }
 770
 771     ret = nfs_ftruncate(client->context, client->fh, offset);
 772     if (ret < 0) {
 773         error_setg_errno(errp, -ret, "Failed to truncate file");
 774         return ret;
 775     }
 776
 777     return 0;
 778 }
 779
 780 /* Note that this will not re-establish a connection with the NFS server
 781  * - it is effectively a NOP.  */
 782 static int nfs_reopen_prepare(BDRVReopenState *state,
 783                               BlockReopenQueue *queue, Error **errp)
 784 {
 785     NFSClient *client = state->bs->opaque;
 786     struct stat st;
 787     int ret = 0;
 788
 789     if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
 790         error_setg(errp, "Cannot open a read-only mount as read-write");
 791         return -EACCES;
 792     }
 793
 794     if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
 795         error_setg(errp, "Cannot disable cache if libnfs readahead or"
 796                          " pagecache is enabled");
 797         return -EINVAL;
 798     }
 799
 800     /* Update cache for read-only reopens */
 801     if (!(state->flags & BDRV_O_RDWR)) {
 802         ret = nfs_fstat(client->context, client->fh, &st);
 803         if (ret < 0) {
 804             error_setg(errp, "Failed to fstat file: %s",
 805                        nfs_get_error(client->context));
 806             return ret;
 807         }
 808 #if !defined(_WIN32)
 809         client->st_blocks = st.st_blocks;
 810 #endif
 811     }
 812
 813     return 0;
 814 }
 815
 816 static void nfs_refresh_filename(BlockDriverState *bs)
 817 {
 818     NFSClient *client = bs->opaque;
 819
 820     if (client->uid && !client->gid) {
 821         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 822                  "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
 823                  client->uid);
 824     } else if (!client->uid && client->gid) {
 825         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 826                  "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
 827                  client->gid);
 828     } else if (client->uid && client->gid) {
 829         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 830                  "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
 831                  client->server->host, client->path, client->uid, client->gid);
 832     } else {
 833         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 834                  "nfs://%s%s", client->server->host, client->path);
 835     }
 836 }
 837
 838 static char *nfs_dirname(BlockDriverState *bs, Error **errp)
 839 {
 840     NFSClient *client = bs->opaque;
 841
 842     if (client->uid || client->gid) {
 843         bdrv_refresh_filename(bs);
 844         error_setg(errp, "Cannot generate a base directory for NFS node '%s'",
 845                    bs->filename);
 846         return NULL;
 847     }
 848
 849     return g_strdup_printf("nfs://%s%s/", client->server->host, client->path);
 850 }
 851
 852 #ifdef LIBNFS_FEATURE_PAGECACHE
 853 static void coroutine_fn nfs_co_invalidate_cache(BlockDriverState *bs,
 854                                                  Error **errp)
 855 {
 856     NFSClient *client = bs->opaque;
 857     nfs_pagecache_invalidate(client->context, client->fh);
 858 }
 859 #endif
 860
 861 static const char *nfs_strong_runtime_opts[] = {
 862     "path",
 863     "user",
 864     "group",
 865     "server.",
 866
 867     NULL
 868 };
 869
 870 static BlockDriver bdrv_nfs = {
 871     .format_name                    = "nfs",
 872     .protocol_name                  = "nfs",
 873
 874     .instance_size                  = sizeof(NFSClient),
 875     .bdrv_parse_filename            = nfs_parse_filename,
 876     .create_opts                    = &nfs_create_opts,
 877
 878     .bdrv_has_zero_init             = nfs_has_zero_init,
 879 /* libnfs does not provide the allocated filesize of a file on win32. */
 880 #if !defined(_WIN32)
 881     .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
 882 #endif
 883     .bdrv_co_truncate               = nfs_file_co_truncate,
 884
 885     .bdrv_file_open                 = nfs_file_open,
 886     .bdrv_close                     = nfs_file_close,
 887     .bdrv_co_create                 = nfs_file_co_create,
 888     .bdrv_co_create_opts            = nfs_file_co_create_opts,
 889     .bdrv_reopen_prepare            = nfs_reopen_prepare,
 890
 891     .bdrv_co_preadv                 = nfs_co_preadv,
 892     .bdrv_co_pwritev                = nfs_co_pwritev,
 893     .bdrv_co_flush_to_disk          = nfs_co_flush,
 894
 895     .bdrv_detach_aio_context        = nfs_detach_aio_context,
 896     .bdrv_attach_aio_context        = nfs_attach_aio_context,
 897     .bdrv_refresh_filename          = nfs_refresh_filename,
 898     .bdrv_dirname                   = nfs_dirname,
 899
 900     .strong_runtime_opts            = nfs_strong_runtime_opts,
 901
 902 #ifdef LIBNFS_FEATURE_PAGECACHE
 903     .bdrv_co_invalidate_cache       = nfs_co_invalidate_cache,
 904 #endif
 905 };
 906
 907 static void nfs_block_init(void)
 908 {
 909     bdrv_register(&bdrv_nfs);
 910 }
 911
 912 block_init(nfs_block_init);