qemu/block/nfs.c
<<
>>
Prefs
   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
  50typedef 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
  64typedef 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
  74static 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;
 141out:
 142    if (qp) {
 143        query_params_free(qp);
 144    }
 145    if (uri) {
 146        uri_free(uri);
 147    }
 148    return ret;
 149}
 150
 151static 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
 175static 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
 185static void nfs_process_read(void *arg);
 186static void nfs_process_write(void *arg);
 187
 188/* Called with QemuMutex held.  */
 189static 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
 203static 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
 213static 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
 223static 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
 232static 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.  */
 241static void
 242nfs_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
 262static 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
 297static 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
 347static 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
 370static 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
 413static 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
 422static 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
 431static void nfs_client_close(NFSClient *client)
 432{
 433    if (client->context) {
 434        if (client->fh) {
 435            nfs_close(client->context, client->fh);
 436            client->fh = NULL;
 437        }
 438        aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 439                           false, NULL, NULL, NULL, NULL);
 440        nfs_destroy_context(client->context);
 441        client->context = NULL;
 442    }
 443    g_free(client->path);
 444    qemu_mutex_destroy(&client->mutex);
 445    qapi_free_NFSServer(client->server);
 446    client->server = NULL;
 447}
 448
 449static void nfs_file_close(BlockDriverState *bs)
 450{
 451    NFSClient *client = bs->opaque;
 452    nfs_client_close(client);
 453}
 454
 455static NFSServer *nfs_config(QDict *options, Error **errp)
 456{
 457    NFSServer *server = NULL;
 458    QDict *addr = NULL;
 459    QObject *crumpled_addr = NULL;
 460    Visitor *iv = NULL;
 461    Error *local_error = NULL;
 462
 463    qdict_extract_subqdict(options, &addr, "server.");
 464    if (!qdict_size(addr)) {
 465        error_setg(errp, "NFS server address missing");
 466        goto out;
 467    }
 468
 469    crumpled_addr = qdict_crumple(addr, errp);
 470    if (!crumpled_addr) {
 471        goto out;
 472    }
 473
 474    /*
 475     * Caution: this works only because all scalar members of
 476     * NFSServer are QString in @crumpled_addr.  The visitor expects
 477     * @crumpled_addr to be typed according to the QAPI schema.  It
 478     * is when @options come from -blockdev or blockdev_add.  But when
 479     * they come from -drive, they're all QString.
 480     */
 481    iv = qobject_input_visitor_new(crumpled_addr);
 482    visit_type_NFSServer(iv, NULL, &server, &local_error);
 483    if (local_error) {
 484        error_propagate(errp, local_error);
 485        goto out;
 486    }
 487
 488out:
 489    QDECREF(addr);
 490    qobject_decref(crumpled_addr);
 491    visit_free(iv);
 492    return server;
 493}
 494
 495
 496static int64_t nfs_client_open(NFSClient *client, QDict *options,
 497                               int flags, int open_flags, Error **errp)
 498{
 499    int ret = -EINVAL;
 500    QemuOpts *opts = NULL;
 501    Error *local_err = NULL;
 502    struct stat st;
 503    char *file = NULL, *strp = NULL;
 504
 505    qemu_mutex_init(&client->mutex);
 506    opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 507    qemu_opts_absorb_qdict(opts, options, &local_err);
 508    if (local_err) {
 509        error_propagate(errp, local_err);
 510        ret = -EINVAL;
 511        goto fail;
 512    }
 513
 514    client->path = g_strdup(qemu_opt_get(opts, "path"));
 515    if (!client->path) {
 516        ret = -EINVAL;
 517        error_setg(errp, "No path was specified");
 518        goto fail;
 519    }
 520
 521    strp = strrchr(client->path, '/');
 522    if (strp == NULL) {
 523        error_setg(errp, "Invalid URL specified");
 524        goto fail;
 525    }
 526    file = g_strdup(strp);
 527    *strp = 0;
 528
 529    /* Pop the config into our state object, Exit if invalid */
 530    client->server = nfs_config(options, errp);
 531    if (!client->server) {
 532        ret = -EINVAL;
 533        goto fail;
 534    }
 535
 536    client->context = nfs_init_context();
 537    if (client->context == NULL) {
 538        error_setg(errp, "Failed to init NFS context");
 539        goto fail;
 540    }
 541
 542    if (qemu_opt_get(opts, "user")) {
 543        client->uid = qemu_opt_get_number(opts, "user", 0);
 544        nfs_set_uid(client->context, client->uid);
 545    }
 546
 547    if (qemu_opt_get(opts, "group")) {
 548        client->gid = qemu_opt_get_number(opts, "group", 0);
 549        nfs_set_gid(client->context, client->gid);
 550    }
 551
 552    if (qemu_opt_get(opts, "tcp-syn-count")) {
 553        client->tcp_syncnt = qemu_opt_get_number(opts, "tcp-syn-count", 0);
 554        nfs_set_tcp_syncnt(client->context, client->tcp_syncnt);
 555    }
 556
 557#ifdef LIBNFS_FEATURE_READAHEAD
 558    if (qemu_opt_get(opts, "readahead-size")) {
 559        if (open_flags & BDRV_O_NOCACHE) {
 560            error_setg(errp, "Cannot enable NFS readahead "
 561                             "if cache.direct = on");
 562            goto fail;
 563        }
 564        client->readahead = qemu_opt_get_number(opts, "readahead-size", 0);
 565        if (client->readahead > QEMU_NFS_MAX_READAHEAD_SIZE) {
 566            warn_report("Truncating NFS readahead size to %d",
 567                        QEMU_NFS_MAX_READAHEAD_SIZE);
 568            client->readahead = QEMU_NFS_MAX_READAHEAD_SIZE;
 569        }
 570        nfs_set_readahead(client->context, client->readahead);
 571#ifdef LIBNFS_FEATURE_PAGECACHE
 572        nfs_set_pagecache_ttl(client->context, 0);
 573#endif
 574        client->cache_used = true;
 575    }
 576#endif
 577
 578#ifdef LIBNFS_FEATURE_PAGECACHE
 579    if (qemu_opt_get(opts, "page-cache-size")) {
 580        if (open_flags & BDRV_O_NOCACHE) {
 581            error_setg(errp, "Cannot enable NFS pagecache "
 582                             "if cache.direct = on");
 583            goto fail;
 584        }
 585        client->pagecache = qemu_opt_get_number(opts, "page-cache-size", 0);
 586        if (client->pagecache > QEMU_NFS_MAX_PAGECACHE_SIZE) {
 587            warn_report("Truncating NFS pagecache size to %d pages",
 588                        QEMU_NFS_MAX_PAGECACHE_SIZE);
 589            client->pagecache = QEMU_NFS_MAX_PAGECACHE_SIZE;
 590        }
 591        nfs_set_pagecache(client->context, client->pagecache);
 592        nfs_set_pagecache_ttl(client->context, 0);
 593        client->cache_used = true;
 594    }
 595#endif
 596
 597#ifdef LIBNFS_FEATURE_DEBUG
 598    if (qemu_opt_get(opts, "debug")) {
 599        client->debug = qemu_opt_get_number(opts, "debug", 0);
 600        /* limit the maximum debug level to avoid potential flooding
 601         * of our log files. */
 602        if (client->debug > QEMU_NFS_MAX_DEBUG_LEVEL) {
 603            warn_report("Limiting NFS debug level to %d",
 604                        QEMU_NFS_MAX_DEBUG_LEVEL);
 605            client->debug = QEMU_NFS_MAX_DEBUG_LEVEL;
 606        }
 607        nfs_set_debug(client->context, client->debug);
 608    }
 609#endif
 610
 611    ret = nfs_mount(client->context, client->server->host, client->path);
 612    if (ret < 0) {
 613        error_setg(errp, "Failed to mount nfs share: %s",
 614                   nfs_get_error(client->context));
 615        goto fail;
 616    }
 617
 618    if (flags & O_CREAT) {
 619        ret = nfs_creat(client->context, file, 0600, &client->fh);
 620        if (ret < 0) {
 621            error_setg(errp, "Failed to create file: %s",
 622                       nfs_get_error(client->context));
 623            goto fail;
 624        }
 625    } else {
 626        ret = nfs_open(client->context, file, flags, &client->fh);
 627        if (ret < 0) {
 628            error_setg(errp, "Failed to open file : %s",
 629                       nfs_get_error(client->context));
 630            goto fail;
 631        }
 632    }
 633
 634    ret = nfs_fstat(client->context, client->fh, &st);
 635    if (ret < 0) {
 636        error_setg(errp, "Failed to fstat file: %s",
 637                   nfs_get_error(client->context));
 638        goto fail;
 639    }
 640
 641    ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
 642    client->st_blocks = st.st_blocks;
 643    client->has_zero_init = S_ISREG(st.st_mode);
 644    *strp = '/';
 645    goto out;
 646
 647fail:
 648    nfs_client_close(client);
 649out:
 650    qemu_opts_del(opts);
 651    g_free(file);
 652    return ret;
 653}
 654
 655static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
 656                         Error **errp) {
 657    NFSClient *client = bs->opaque;
 658    int64_t ret;
 659
 660    client->aio_context = bdrv_get_aio_context(bs);
 661
 662    ret = nfs_client_open(client, options,
 663                          (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
 664                          bs->open_flags, errp);
 665    if (ret < 0) {
 666        return ret;
 667    }
 668
 669    bs->total_sectors = ret;
 670    ret = 0;
 671    return ret;
 672}
 673
 674static QemuOptsList nfs_create_opts = {
 675    .name = "nfs-create-opts",
 676    .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
 677    .desc = {
 678        {
 679            .name = BLOCK_OPT_SIZE,
 680            .type = QEMU_OPT_SIZE,
 681            .help = "Virtual disk size"
 682        },
 683        { /* end of list */ }
 684    }
 685};
 686
 687static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
 688{
 689    int ret = 0;
 690    int64_t total_size = 0;
 691    NFSClient *client = g_new0(NFSClient, 1);
 692    QDict *options = NULL;
 693
 694    client->aio_context = qemu_get_aio_context();
 695
 696    /* Read out options */
 697    total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 698                          BDRV_SECTOR_SIZE);
 699
 700    options = qdict_new();
 701    ret = nfs_parse_uri(url, options, errp);
 702    if (ret < 0) {
 703        goto out;
 704    }
 705
 706    ret = nfs_client_open(client, options, O_CREAT, 0, errp);
 707    if (ret < 0) {
 708        goto out;
 709    }
 710    ret = nfs_ftruncate(client->context, client->fh, total_size);
 711    nfs_client_close(client);
 712out:
 713    QDECREF(options);
 714    g_free(client);
 715    return ret;
 716}
 717
 718static int nfs_has_zero_init(BlockDriverState *bs)
 719{
 720    NFSClient *client = bs->opaque;
 721    return client->has_zero_init;
 722}
 723
 724/* Called (via nfs_service) with QemuMutex held.  */
 725static void
 726nfs_get_allocated_file_size_cb(int ret, struct nfs_context *nfs, void *data,
 727                               void *private_data)
 728{
 729    NFSRPC *task = private_data;
 730    task->ret = ret;
 731    if (task->ret == 0) {
 732        memcpy(task->st, data, sizeof(struct stat));
 733    }
 734    if (task->ret < 0) {
 735        error_report("NFS Error: %s", nfs_get_error(nfs));
 736    }
 737
 738    /* Set task->complete before reading bs->wakeup.  */
 739    atomic_mb_set(&task->complete, 1);
 740    bdrv_wakeup(task->bs);
 741}
 742
 743static 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
 767static int nfs_file_truncate(BlockDriverState *bs, int64_t offset,
 768                             PreallocMode prealloc, Error **errp)
 769{
 770    NFSClient *client = bs->opaque;
 771    int ret;
 772
 773    if (prealloc != PREALLOC_MODE_OFF) {
 774        error_setg(errp, "Unsupported preallocation mode '%s'",
 775                   PreallocMode_lookup[prealloc]);
 776        return -ENOTSUP;
 777    }
 778
 779    ret = nfs_ftruncate(client->context, client->fh, offset);
 780    if (ret < 0) {
 781        error_setg_errno(errp, -ret, "Failed to truncate file");
 782        return ret;
 783    }
 784
 785    return 0;
 786}
 787
 788/* Note that this will not re-establish a connection with the NFS server
 789 * - it is effectively a NOP.  */
 790static int nfs_reopen_prepare(BDRVReopenState *state,
 791                              BlockReopenQueue *queue, Error **errp)
 792{
 793    NFSClient *client = state->bs->opaque;
 794    struct stat st;
 795    int ret = 0;
 796
 797    if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
 798        error_setg(errp, "Cannot open a read-only mount as read-write");
 799        return -EACCES;
 800    }
 801
 802    if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
 803        error_setg(errp, "Cannot disable cache if libnfs readahead or"
 804                         " pagecache is enabled");
 805        return -EINVAL;
 806    }
 807
 808    /* Update cache for read-only reopens */
 809    if (!(state->flags & BDRV_O_RDWR)) {
 810        ret = nfs_fstat(client->context, client->fh, &st);
 811        if (ret < 0) {
 812            error_setg(errp, "Failed to fstat file: %s",
 813                       nfs_get_error(client->context));
 814            return ret;
 815        }
 816        client->st_blocks = st.st_blocks;
 817    }
 818
 819    return 0;
 820}
 821
 822static void nfs_refresh_filename(BlockDriverState *bs, QDict *options)
 823{
 824    NFSClient *client = bs->opaque;
 825    QDict *opts = qdict_new();
 826    QObject *server_qdict;
 827    Visitor *ov;
 828
 829    qdict_put_str(opts, "driver", "nfs");
 830
 831    if (client->uid && !client->gid) {
 832        snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 833                 "nfs://%s%s?uid=%" PRId64, client->server->host, client->path,
 834                 client->uid);
 835    } else if (!client->uid && client->gid) {
 836        snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 837                 "nfs://%s%s?gid=%" PRId64, client->server->host, client->path,
 838                 client->gid);
 839    } else if (client->uid && client->gid) {
 840        snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 841                 "nfs://%s%s?uid=%" PRId64 "&gid=%" PRId64,
 842                 client->server->host, client->path, client->uid, client->gid);
 843    } else {
 844        snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 845                 "nfs://%s%s", client->server->host, client->path);
 846    }
 847
 848    ov = qobject_output_visitor_new(&server_qdict);
 849    visit_type_NFSServer(ov, NULL, &client->server, &error_abort);
 850    visit_complete(ov, &server_qdict);
 851    qdict_put_obj(opts, "server", server_qdict);
 852    qdict_put_str(opts, "path", client->path);
 853
 854    if (client->uid) {
 855        qdict_put_int(opts, "user", client->uid);
 856    }
 857    if (client->gid) {
 858        qdict_put_int(opts, "group", client->gid);
 859    }
 860    if (client->tcp_syncnt) {
 861        qdict_put_int(opts, "tcp-syn-cnt", client->tcp_syncnt);
 862    }
 863    if (client->readahead) {
 864        qdict_put_int(opts, "readahead-size", client->readahead);
 865    }
 866    if (client->pagecache) {
 867        qdict_put_int(opts, "page-cache-size", client->pagecache);
 868    }
 869    if (client->debug) {
 870        qdict_put_int(opts, "debug", client->debug);
 871    }
 872
 873    visit_free(ov);
 874    qdict_flatten(opts);
 875    bs->full_open_options = opts;
 876}
 877
 878#ifdef LIBNFS_FEATURE_PAGECACHE
 879static void nfs_invalidate_cache(BlockDriverState *bs,
 880                                 Error **errp)
 881{
 882    NFSClient *client = bs->opaque;
 883    nfs_pagecache_invalidate(client->context, client->fh);
 884}
 885#endif
 886
 887static BlockDriver bdrv_nfs = {
 888    .format_name                    = "nfs",
 889    .protocol_name                  = "nfs",
 890
 891    .instance_size                  = sizeof(NFSClient),
 892    .bdrv_parse_filename            = nfs_parse_filename,
 893    .create_opts                    = &nfs_create_opts,
 894
 895    .bdrv_has_zero_init             = nfs_has_zero_init,
 896    .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
 897    .bdrv_truncate                  = nfs_file_truncate,
 898
 899    .bdrv_file_open                 = nfs_file_open,
 900    .bdrv_close                     = nfs_file_close,
 901    .bdrv_create                    = nfs_file_create,
 902    .bdrv_reopen_prepare            = nfs_reopen_prepare,
 903
 904    .bdrv_co_preadv                 = nfs_co_preadv,
 905    .bdrv_co_pwritev                = nfs_co_pwritev,
 906    .bdrv_co_flush_to_disk          = nfs_co_flush,
 907
 908    .bdrv_detach_aio_context        = nfs_detach_aio_context,
 909    .bdrv_attach_aio_context        = nfs_attach_aio_context,
 910    .bdrv_refresh_filename          = nfs_refresh_filename,
 911
 912#ifdef LIBNFS_FEATURE_PAGECACHE
 913    .bdrv_invalidate_cache          = nfs_invalidate_cache,
 914#endif
 915};
 916
 917static void nfs_block_init(void)
 918{
 919    bdrv_register(&bdrv_nfs);
 920}
 921
 922block_init(nfs_block_init);
 923