block/nfs.c

   1 /*
   2  * QEMU Block driver for native access to files on NFS shares
   3  *
   4  * Copyright (c) 2014-2016 Peter Lieven <pl@kamp.de>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu/osdep.h"
  26
  27 #include <poll.h>
  28 #include "qemu-common.h"
  29 #include "qemu/config-file.h"
  30 #include "qemu/error-report.h"
  31 #include "qapi/error.h"
  32 #include "block/block_int.h"
  33 #include "trace.h"
  34 #include "qemu/iov.h"
  35 #include "qemu/uri.h"
  36 #include "qemu/cutils.h"
  37 #include "sysemu/sysemu.h"
  38 #include "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>
  45
  46
  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
  50
  51 typedef struct NFSClient {
  52     struct nfs_context *context;
  53     struct nfsfh *fh;
  54     int events;
  55     bool has_zero_init;
  56     AioContext *aio_context;
  57     QemuMutex mutex;
  58     blkcnt_t st_blocks;
  59     bool cache_used;
  60     NFSServer *server;
  61     char *path;
  62     int64_t uid, gid, tcp_syncnt, readahead, pagecache, debug;
  63 } NFSClient;
  64
  65 typedef struct NFSRPC {
  66     BlockDriverState *bs;
  67     int ret;
  68     int complete;
  69     QEMUIOVector *iov;
  70     struct stat *st;
  71     Coroutine *co;
  72     NFSClient *client;
  73 } NFSRPC;
  74
  75 static int nfs_parse_uri(const char *filename, QDict *options, Error **errp)
  76 {
  77     URI *uri = NULL;
  78     QueryParams *qp = NULL;
  79     int ret = -EINVAL, i;
  80
  81     uri = uri_parse(filename);
  82     if (!uri) {
  83         error_setg(errp, "Invalid URI specified");
  84         goto out;
  85     }
  86     if (strcmp(uri->scheme, "nfs") != 0) {
  87         error_setg(errp, "URI scheme must be 'nfs'");
  88         goto out;
  89     }
  90
  91     if (!uri->server) {
  92         error_setg(errp, "missing hostname in URI");
  93         goto out;
  94     }
  95
  96     if (!uri->path) {
  97         error_setg(errp, "missing file path in URI");
  98         goto out;
  99     }
 100
 101     qp = query_params_parse(uri->query);
 102     if (!qp) {
 103         error_setg(errp, "could not parse query parameters");
 104         goto out;
 105     }
 106
 107     qdict_put(options, "server.host", qstring_from_str(uri->server));
 108     qdict_put(options, "server.type", qstring_from_str("inet"));
 109     qdict_put(options, "path", qstring_from_str(uri->path));
 110
 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",
 115                        qp->p[i].name);
 116             goto out;
 117         }
 118         if (parse_uint_full(qp->p[i].value, &val, 0)) {
 119             error_setg(errp, "Illegal value for NFS parameter: %s",
 120                        qp->p[i].name);
 121             goto out;
 122         }
 123         if (!strcmp(qp->p[i].name, "uid")) {
 124             qdict_put(options, "user",
 125                       qstring_from_str(qp->p[i].value));
 126         } else if (!strcmp(qp->p[i].name, "gid")) {
 127             qdict_put(options, "group",
 128                       qstring_from_str(qp->p[i].value));
 129         } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
 130             qdict_put(options, "tcp-syn-count",
 131                       qstring_from_str(qp->p[i].value));
 132         } else if (!strcmp(qp->p[i].name, "readahead")) {
 133             qdict_put(options, "readahead-size",
 134                       qstring_from_str(qp->p[i].value));
 135         } else if (!strcmp(qp->p[i].name, "pagecache")) {
 136             qdict_put(options, "page-cache-size",
 137                       qstring_from_str(qp->p[i].value));
 138         } else if (!strcmp(qp->p[i].name, "debug")) {
 139             qdict_put(options, "debug",
 140                       qstring_from_str(qp->p[i].value));
 141         } else {
 142             error_setg(errp, "Unknown NFS parameter name: %s",
 143                        qp->p[i].name);
 144             goto out;
 145         }
 146     }
 147     ret = 0;
 148 out:
 149     if (qp) {
 150         query_params_free(qp);
 151     }
 152     if (uri) {
 153         uri_free(uri);
 154     }
 155     return ret;
 156 }
 157
 158 static bool nfs_has_filename_options_conflict(QDict *options, Error **errp)
 159 {
 160     const QDictEntry *qe;
 161
 162     for (qe = qdict_first(options); qe; qe = qdict_next(options, qe)) {
 163         if (!strcmp(qe->key, "host") ||
 164             !strcmp(qe->key, "path") ||
 165             !strcmp(qe->key, "user") ||
 166             !strcmp(qe->key, "group") ||
 167             !strcmp(qe->key, "tcp-syn-count") ||
 168             !strcmp(qe->key, "readahead-size") ||
 169             !strcmp(qe->key, "page-cache-size") ||
 170             !strcmp(qe->key, "debug") ||
 171             strstart(qe->key, "server.", NULL))
 172         {
 173             error_setg(errp, "Option %s cannot be used with a filename",
 174                        qe->key);
 175             return true;
 176         }
 177     }
 178
 179     return false;
 180 }
 181
 182 static void nfs_parse_filename(const char *filename, QDict *options,
 183                                Error **errp)
 184 {
 185     if (nfs_has_filename_options_conflict(options, errp)) {
 186         return;
 187     }
 188
 189     nfs_parse_uri(filename, options, errp);
 190 }
 191
 192 static void nfs_process_read(void *arg);
 193 static void nfs_process_write(void *arg);
 194
 195 /* Called with QemuMutex held.  */
 196 static void nfs_set_events(NFSClient *client)
 197 {
 198     int ev = nfs_which_events(client->context);
 199     if (ev != client->events) {
 200         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 201                            false,
 202                            (ev & POLLIN) ? nfs_process_read : NULL,
 203                            (ev & POLLOUT) ? nfs_process_write : NULL,
 204                            NULL, client);
 205
 206     }
 207     client->events = ev;
 208 }
 209
 210 static void nfs_process_read(void *arg)
 211 {
 212     NFSClient *client = arg;
 213
 214     qemu_mutex_lock(&client->mutex);
 215     nfs_service(client->context, POLLIN);
 216     nfs_set_events(client);
 217     qemu_mutex_unlock(&client->mutex);
 218 }
 219
 220 static void nfs_process_write(void *arg)
 221 {
 222     NFSClient *client = arg;
 223
 224     qemu_mutex_lock(&client->mutex);
 225     nfs_service(client->context, POLLOUT);
 226     nfs_set_events(client);
 227     qemu_mutex_unlock(&client->mutex);
 228 }
 229
 230 static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
 231 {
 232     *task = (NFSRPC) {
 233         .co             = qemu_coroutine_self(),
 234         .bs             = bs,
 235         .client         = bs->opaque,
 236     };
 237 }
 238
 239 static void nfs_co_generic_bh_cb(void *opaque)
 240 {
 241     NFSRPC *task = opaque;
 242
 243     task->complete = 1;
 244     aio_co_wake(task->co);
 245 }
 246
 247 /* Called (via nfs_service) with QemuMutex held.  */
 248 static void
 249 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
 250                   void *private_data)
 251 {
 252     NFSRPC *task = private_data;
 253     task->ret = ret;
 254     assert(!task->st);
 255     if (task->ret > 0 && task->iov) {
 256         if (task->ret <= task->iov->size) {
 257             qemu_iovec_from_buf(task->iov, 0, data, task->ret);
 258         } else {
 259             task->ret = -EIO;
 260         }
 261     }
 262     if (task->ret < 0) {
 263         error_report("NFS Error: %s", nfs_get_error(nfs));
 264     }
 265     aio_bh_schedule_oneshot(task->client->aio_context,
 266                             nfs_co_generic_bh_cb, task);
 267 }
 268
 269 static int coroutine_fn nfs_co_preadv(BlockDriverState *bs, uint64_t offset,
 270                                       uint64_t bytes, QEMUIOVector *iov,
 271                                       int flags)
 272 {
 273     NFSClient *client = bs->opaque;
 274     NFSRPC task;
 275
 276     nfs_co_init_task(bs, &task);
 277     task.iov = iov;
 278
 279     qemu_mutex_lock(&client->mutex);
 280     if (nfs_pread_async(client->context, client->fh,
 281                         offset, bytes, nfs_co_generic_cb, &task) != 0) {
 282         qemu_mutex_unlock(&client->mutex);
 283         return -ENOMEM;
 284     }
 285
 286     nfs_set_events(client);
 287     qemu_mutex_unlock(&client->mutex);
 288     while (!task.complete) {
 289         qemu_coroutine_yield();
 290     }
 291
 292     if (task.ret < 0) {
 293         return task.ret;
 294     }
 295
 296     /* zero pad short reads */
 297     if (task.ret < iov->size) {
 298         qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
 299     }
 300
 301     return 0;
 302 }
 303
 304 static int coroutine_fn nfs_co_pwritev(BlockDriverState *bs, uint64_t offset,
 305                                        uint64_t bytes, QEMUIOVector *iov,
 306                                        int flags)
 307 {
 308     NFSClient *client = bs->opaque;
 309     NFSRPC task;
 310     char *buf = NULL;
 311     bool my_buffer = false;
 312
 313     nfs_co_init_task(bs, &task);
 314
 315     if (iov->niov != 1) {
 316         buf = g_try_malloc(bytes);
 317         if (bytes && buf == NULL) {
 318             return -ENOMEM;
 319         }
 320         qemu_iovec_to_buf(iov, 0, buf, bytes);
 321         my_buffer = true;
 322     } else {
 323         buf = iov->iov[0].iov_base;
 324     }
 325
 326     qemu_mutex_lock(&client->mutex);
 327     if (nfs_pwrite_async(client->context, client->fh,
 328                          offset, bytes, buf,
 329                          nfs_co_generic_cb, &task) != 0) {
 330         qemu_mutex_unlock(&client->mutex);
 331         if (my_buffer) {
 332             g_free(buf);
 333         }
 334         return -ENOMEM;
 335     }
 336
 337     nfs_set_events(client);
 338     qemu_mutex_unlock(&client->mutex);
 339     while (!task.complete) {
 340         qemu_coroutine_yield();
 341     }
 342
 343     if (my_buffer) {
 344         g_free(buf);
 345     }
 346
 347     if (task.ret != bytes) {
 348         return task.ret < 0 ? task.ret : -EIO;
 349     }
 350
 351     return 0;
 352 }
 353
 354 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
 355 {
 356     NFSClient *client = bs->opaque;
 357     NFSRPC task;
 358
 359     nfs_co_init_task(bs, &task);
 360
 361     qemu_mutex_lock(&client->mutex);
 362     if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
 363                         &task) != 0) {
 364         qemu_mutex_unlock(&client->mutex);
 365         return -ENOMEM;
 366     }
 367
 368     nfs_set_events(client);
 369     qemu_mutex_unlock(&client->mutex);
 370     while (!task.complete) {
 371         qemu_coroutine_yield();
 372     }
 373
 374     return task.ret;
 375 }
 376
 377 static QemuOptsList runtime_opts = {
 378     .name = "nfs",
 379     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 380     .desc = {
 381         {
 382             .name = "path",
 383             .type = QEMU_OPT_STRING,
 384             .help = "Path of the image on the host",
 385         },
 386         {
 387             .name = "user",
 388             .type = QEMU_OPT_NUMBER,
 389             .help = "UID value to use when talking to the server",
 390         },
 391         {
 392             .name = "group",
 393             .type = QEMU_OPT_NUMBER,
 394             .help = "GID value to use when talking to the server",
 395         },
 396         {
 397             .name = "tcp-syn-count",
 398             .type = QEMU_OPT_NUMBER,
 399             .help = "Number of SYNs to send during the session establish",
 400         },
 401         {
 402             .name = "readahead-size",
 403             .type = QEMU_OPT_NUMBER,
 404             .help = "Set the readahead size in bytes",
 405         },
 406         {
 407             .name = "page-cache-size",
 408             .type = QEMU_OPT_NUMBER,
 409             .help = "Set the pagecache size in bytes",
 410         },
 411         {
 412             .name = "debug",
 413             .type = QEMU_OPT_NUMBER,
 414             .help = "Set the NFS debug level (max 2)",
 415         },
 416         { /* end of list */ }
 417     },
 418 };
 419
 420 static void nfs_detach_aio_context(BlockDriverState *bs)
 421 {
 422     NFSClient *client = bs->opaque;
 423
 424     aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 425                        false, NULL, NULL, NULL, NULL);
 426     client->events = 0;
 427 }
 428
 429 static void nfs_attach_aio_context(BlockDriverState *bs,
 430                                    AioContext *new_context)
 431 {
 432     NFSClient *client = bs->opaque;
 433
 434     client->aio_context = new_context;
 435     nfs_set_events(client);
 436 }
 437
 438 static void nfs_client_close(NFSClient *client)
 439 {
 440     if (client->context) {
 441         if (client->fh) {
 442             nfs_close(client->context, client->fh);
 443         }
 444         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 445                            false, NULL, NULL, NULL, NULL);
 446         nfs_destroy_context(client->context);
 447     }
 448     memset(client, 0, sizeof(NFSClient));
 449 }
 450
 451 static void nfs_file_close(BlockDriverState *bs)
 452 {
 453     NFSClient *client = bs->opaque;
 454     nfs_client_close(client);
 455     qemu_mutex_destroy(&client->mutex);
 456 }
 457
 458 static NFSServer *nfs_config(QDict *options, Error **errp)
 459 {
 460     NFSServer *server = NULL;
 461     QDict *addr = NULL;
 462     QObject *crumpled_addr = NULL;
 463     Visitor *iv = NULL;
 464     Error *local_error = NULL;
 465
 466     qdict_extract_subqdict(options, &addr, "server.");
 467     if (!qdict_size(addr)) {
 468         error_setg(errp, "NFS server address missing");
 469         goto out;
 470     }
 471
 472     crumpled_addr = qdict_crumple(addr, errp);
 473     if (!crumpled_addr) {
 474         goto out;
 475     }
 476
 477     /*
 478      * Caution: this works only because all scalar members of
 479      * NFSServer are QString in @crumpled_addr.  The visitor expects
 480      * @crumpled_addr to be typed according to the QAPI schema.  It
 481      * is when @options come from -blockdev or blockdev_add.  But when
 482      * they come from -drive, they're all QString.
 483      */
 484     iv = qobject_input_visitor_new(crumpled_addr);
 485     visit_type_NFSServer(iv, NULL, &server, &local_error);
 486     if (local_error) {
 487         error_propagate(errp, local_error);
 488         goto out;
 489     }
 490
 491 out:
 492     QDECREF(addr);
 493     qobject_decref(crumpled_addr);
 494     visit_free(iv);
 495     return server;
 496 }
 497
 498
 499 static int64_t nfs_client_open(NFSClient *client, QDict *options,
 500                                int flags, int open_flags, Error **errp)
 501 {
 502     int ret = -EINVAL;
 503     QemuOpts *opts = NULL;
 504     Error *local_err = NULL;
 505     struct stat st;
 506     char *file = NULL, *strp = NULL;
 507
 508     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 509     qemu_opts_absorb_qdict(opts, options, &local_err);
 510     if (local_err) {
 511         error_propagate(errp, local_err);
 512         ret = -EINVAL;
 513         goto fail;
 514     }
 515
 516     client->path = g_strdup(qemu_opt_get(opts, "path"));
 517     if (!client->path) {
 518         ret = -EINVAL;
 519         error_setg(errp, "No path was specified");
 520         goto fail;
 521     }
 522
 523     strp = strrchr(client->path, '/');
 524     if (strp == NULL) {
 525         error_setg(errp, "Invalid URL specified");
 526         goto fail;
 527     }
 528     file = g_strdup(strp);
 529     *strp = 0;
 530
 531     /* Pop the config into our state object, Exit if invalid */
 532     client->server = nfs_config(options, errp);
 533     if (!client->server) {
 534         ret = -EINVAL;
 535         goto fail;
 536     }
 537
 538     client->context = nfs_init_context();
 539     if (client->context == NULL) {
 540         error_setg(errp, "Failed to init NFS context");
 541         goto fail;
 542     }
 543
 544     if (qemu_opt_get(opts, "user")) {
 545         client->uid = qemu_opt_get_number(opts, "user", 0);
 546         nfs_set_uid(client->context, client->uid);
 547     }
 548
 549     if (qemu_opt_get(opts, "group")) {
 550         client->gid = qemu_opt_get_number(opts, "group", 0);
 551         nfs_set_gid(client->context, client->gid);
 552     }
 553
 554     if (qemu_opt_get(opts, "tcp-syn-count")) {
 555         client->tcp_syncnt = qemu_opt_get_number(opts, "tcp-syn-count", 0);
 556         nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
 557     }
 558
 559 #ifdef LIBNFS_FEATURE_READAHEAD
 560     if (qemu_opt_get(opts, "readahead-size")) {
 561         if (open_flags & BDRV_O_NOCACHE) {
 562             error_setg(errp, "Cannot enable NFS readahead "
 563                              "if cache.direct = on");
 564             goto fail;
 565         }
 566         client->readahead = qemu_opt_get_number(opts, "readahead-size", 0);
 567         if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
 568             error_report("NFS Warning: Truncating NFS readahead "
 569                          "size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
 570             client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
 571         }
 572         nfs_set_readahead(client->context, client->readahead);
 573 #ifdef LIBNFS_FEATURE_PAGECACHE
 574         nfs_set_pagecache_ttl(client->context, 0);
 575 #endif
 576         client->cache_used = true;
 577     }
 578 #endif
 579
 580 #ifdef LIBNFS_FEATURE_PAGECACHE
 581     if (qemu_opt_get(opts, "page-cache-size")) {
 582         if (open_flags & BDRV_O_NOCACHE) {
 583             error_setg(errp, "Cannot enable NFS pagecache "
 584                              "if cache.direct = on");
 585             goto fail;
 586         }
 587         client->pagecache = qemu_opt_get_number(opts, "page-cache-size", 0);
 588         if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
 589             error_report("NFS Warning: Truncating NFS pagecache "
 590                          "size to %d pages", QEMU_NFS_MAX_PAGECACHE_SIZE);
 591             client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
 592         }
 593         nfs_set_pagecache(client->context, client->pagecache);
 594         nfs_set_pagecache_ttl(client->context, 0);
 595         client->cache_used = true;
 596     }
 597 #endif
 598
 599 #ifdef LIBNFS_FEATURE_DEBUG
 600     if (qemu_opt_get(opts, "debug")) {
 601         client->debug = qemu_opt_get_number(opts, "debug", 0);
 602         /* limit the maximum debug level to avoid potential flooding
 603          * of our log files. */
 604         if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
 605             error_report("NFS Warning: Limiting NFS debug level "
 606                          "to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
 607             client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
 608         }
 609         nfs_set_debug(client->context, client->debug);
 610     }
 611 #endif
 612
 613     ret = nfs_mount(client->context, client->server->host, client->path);
 614     if (ret < 0) {
 615         error_setg(errp, "Failed to mount nfs share: %s",
 616                    nfs_get_error(client->context));
 617         goto fail;
 618     }
 619
 620     if (flags & O_CREAT) {
 621         ret = nfs_creat(client->context, file, 0600, &client->fh);
 622         if (ret < 0) {
 623             error_setg(errp, "Failed to create file: %s",
 624                        nfs_get_error(client->context));
 625             goto fail;
 626         }
 627     } else {
 628         ret = nfs_open(client->context, file, flags, &client->fh);
 629         if (ret < 0) {
 630             error_setg(errp, "Failed to open file : %s",
 631                        nfs_get_error(client->context));
 632             goto fail;
 633         }
 634     }
 635
 636     ret = nfs_fstat(client->context, client->fh, &st);
 637     if (ret < 0) {
 638         error_setg(errp, "Failed to fstat file: %s",
 639                    nfs_get_error(client->context));
 640         goto fail;
 641     }
 642
 643     ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
 644     client->st_blocks = st.st_blocks;
 645     client->has_zero_init = S_ISREG(st.st_mode);
 646     *strp = '/';
 647     goto out;
 648
 649 fail:
 650     nfs_client_close(client);
 651 out:
 652     qemu_opts_del(opts);
 653     g_free(file);
 654     return ret;
 655 }
 656
 657 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
 658                          Error **errp) {
 659     NFSClient *client = bs->opaque;
 660     int64_t ret;
 661
 662     client->aio_context = bdrv_get_aio_context(bs);
 663
 664     ret = nfs_client_open(client, options,
 665                           (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
 666                           bs->open_flags, errp);
 667     if (ret < 0) {
 668         return ret;
 669     }
 670     qemu_mutex_init(&client->mutex);
 671     bs->total_sectors = ret;
 672     ret = 0;
 673     return ret;
 674 }
 675
 676 static QemuOptsList nfs_create_opts = {
 677     .name = "nfs-create-opts",
 678     .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
 679     .desc = {
 680         {
 681             .name = BLOCK_OPT_SIZE,
 682             .type = QEMU_OPT_SIZE,
 683             .help = "Virtual disk size"
 684         },
 685         { /* end of list */ }
 686     }
 687 };
 688
 689 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
 690 {
 691     int ret = 0;
 692     int64_t total_size = 0;
 693     NFSClient *client = g_new0(NFSClient, 1);
 694     QDict *options = NULL;
 695
 696     client->aio_context = qemu_get_aio_context();
 697
 698     /* Read out options */
 699     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 700                           BDRV_SECTOR_SIZE);
 701
 702     options = qdict_new();
 703     ret = nfs_parse_uri(url, options, errp);
 704     if (ret < 0) {
 705         goto out;
 706     }
 707
 708     ret = nfs_client_open(client, options, O_CREAT, 0, errp);
 709     if (ret < 0) {
 710         goto out;
 711     }
 712     ret = nfs_ftruncate(client->context, client->fh, total_size);
 713     nfs_client_close(client);
 714 out:
 715     QDECREF(options);
 716     g_free(client);
 717     return ret;
 718 }
 719
 720 static int nfs_has_zero_init(BlockDriverState *bs)
 721 {
 722     NFSClient *client = bs->opaque;
 723     return client->has_zero_init;
 724 }
 725
 726 /* Called (via nfs_service) with QemuMutex held.  */
 727 static void
 728 nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
 729                                void *private_data)
 730 {
 731     NFSRPC *task = private_data;
 732     task->ret = ret;
 733     if (task->ret == 0) {
 734         memcpy(task->st, data, sizeof(struct stat));
 735     }
 736     if (task->ret < 0) {
 737         error_report("NFS Error: %s", nfs_get_error(nfs));
 738     }
 739     task->complete = 1;
 740     bdrv_wakeup(task->bs);
 741 }
 742
 743 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
 744 {
 745     NFSClient *client = bs->opaque;
 746     NFSRPC task = {0};
 747     struct stat st;
 748
 749     if (bdrv_is_read_only(bs) &&
 750         !(bs->open_flags & BDRV_O_NOCACHE)) {
 751         return client->st_blocks * 512;
 752     }
 753
 754     task.bs = bs;
 755     task.st = &st;
 756     if (nfs_fstat_async(client->context, client->fh, nfs_get_allocated_file_size_cb,
 757                         &task) != 0) {
 758         return -ENOMEM;
 759     }
 760
 761     nfs_set_events(client);
 762     BDRV_POLL_WHILE(bs, !task.complete);
 763
 764     return (task.ret < 0 ? task.ret : st.st_blocks * 512);
 765 }
 766
 767 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
 768 {
 769     NFSClient *client = bs->opaque;
 770     return nfs_ftruncate(client->context, client->fh, offset);
 771 }
 772
 773 /* Note that this will not re-establish a connection with the NFS server
 774  * - it is effectively a NOP.  */
 775 static int nfs_reopen_prepare(BDRVReopenState *state,
 776                               BlockReopenQueue *queue, Error **errp)
 777 {
 778     NFSClient *client = state->bs->opaque;
 779     struct stat st;
 780     int ret = 0;
 781
 782     if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
 783         error_setg(errp, "Cannot open a read-only mount as read-write");
 784         return -EACCES;
 785     }
 786
 787     if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
 788         error_setg(errp, "Cannot disable cache if libnfs readahead or"
 789                          " pagecache is enabled");
 790         return -EINVAL;
 791     }
 792
 793     /* Update cache for read-only reopens */
 794     if (!(state->flags & BDRV_O_RDWR)) {
 795         ret = nfs_fstat(client->context, client->fh, &st);
 796         if (ret < 0) {
 797             error_setg(errp, "Failed to fstat file: %s",
 798                        nfs_get_error(client->context));
 799             return ret;
 800         }
 801         client->st_blocks = st.st_blocks;
 802     }
 803
 804     return 0;
 805 }
 806
 807 static void nfs_refresh_filename(BlockDriverState *bs, QDict *options)
 808 {
 809     NFSClient *client = bs->opaque;
 810     QDict *opts = qdict_new();
 811     QObject *server_qdict;
 812     Visitor *ov;
 813
 814     qdict_put(opts, "driver", qstring_from_str("nfs"));
 815
 816     if (client->uid && !client->gid) {
 817         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 818                  "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
 819                  client->uid);
 820     } else if (!client->uid && client->gid) {
 821         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 822                  "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
 823                  client->gid);
 824     } else if (client->uid && client->gid) {
 825         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 826                  "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
 827                  client->server->host, client->path, client->uid, client->gid);
 828     } else {
 829         snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 830                  "nfs://%s%s", client->server->host, client->path);
 831     }
 832
 833     ov = qobject_output_visitor_new(&server_qdict);
 834     visit_type_NFSServer(ov, NULL, &client->server, &error_abort);
 835     visit_complete(ov, &server_qdict);
 836     qdict_put_obj(opts, "server", server_qdict);
 837     qdict_put(opts, "path", qstring_from_str(client->path));
 838
 839     if (client->uid) {
 840         qdict_put(opts, "user", qint_from_int(client->uid));
 841     }
 842     if (client->gid) {
 843         qdict_put(opts, "group", qint_from_int(client->gid));
 844     }
 845     if (client->tcp_syncnt) {
 846         qdict_put(opts, "tcp-syn-cnt",
 847                   qint_from_int(client->tcp_syncnt));
 848     }
 849     if (client->readahead) {
 850         qdict_put(opts, "readahead-size",
 851                   qint_from_int(client->readahead));
 852     }
 853     if (client->pagecache) {
 854         qdict_put(opts, "page-cache-size",
 855                   qint_from_int(client->pagecache));
 856     }
 857     if (client->debug) {
 858         qdict_put(opts, "debug", qint_from_int(client->debug));
 859     }
 860
 861     visit_free(ov);
 862     qdict_flatten(opts);
 863     bs->full_open_options = opts;
 864 }
 865
 866 #ifdef LIBNFS_FEATURE_PAGECACHE
 867 static void nfs_invalidate_cache(BlockDriverState *bs,
 868                                  Error **errp)
 869 {
 870     NFSClient *client = bs->opaque;
 871     nfs_pagecache_invalidate(client->context, client->fh);
 872 }
 873 #endif
 874
 875 static BlockDriver bdrv_nfs = {
 876     .format_name                    = "nfs",
 877     .protocol_name                  = "nfs",
 878
 879     .instance_size                  = sizeof(NFSClient),
 880     .bdrv_parse_filename            = nfs_parse_filename,
 881     .create_opts                    = &nfs_create_opts,
 882
 883     .bdrv_has_zero_init             = nfs_has_zero_init,
 884     .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
 885     .bdrv_truncate                  = nfs_file_truncate,
 886
 887     .bdrv_file_open                 = nfs_file_open,
 888     .bdrv_close                     = nfs_file_close,
 889     .bdrv_create                    = nfs_file_create,
 890     .bdrv_reopen_prepare            = nfs_reopen_prepare,
 891
 892     .bdrv_co_preadv                 = nfs_co_preadv,
 893     .bdrv_co_pwritev                = nfs_co_pwritev,
 894     .bdrv_co_flush_to_disk          = nfs_co_flush,
 895
 896     .bdrv_detach_aio_context        = nfs_detach_aio_context,
 897     .bdrv_attach_aio_context        = nfs_attach_aio_context,
 898     .bdrv_refresh_filename          = nfs_refresh_filename,
 899
 900 #ifdef LIBNFS_FEATURE_PAGECACHE
 901     .bdrv_invalidate_cache          = nfs_invalidate_cache,
 902 #endif
 903 };
 904
 905 static void nfs_block_init(void)
 906 {
 907     bdrv_register(&bdrv_nfs);
 908 }
 909
 910 block_init(nfs_block_init);