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 <nfsc/libnfs.h>
  39
  40 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
  41 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
  42
  43 typedef struct NFSClient {
  44     struct nfs_context *context;
  45     struct nfsfh *fh;
  46     int events;
  47     bool has_zero_init;
  48     AioContext *aio_context;
  49     blkcnt_t st_blocks;
  50     bool cache_used;
  51 } NFSClient;
  52
  53 typedef struct NFSRPC {
  54     int ret;
  55     int complete;
  56     QEMUIOVector *iov;
  57     struct stat *st;
  58     Coroutine *co;
  59     QEMUBH *bh;
  60     NFSClient *client;
  61 } NFSRPC;
  62
  63 static void nfs_process_read(void *arg);
  64 static void nfs_process_write(void *arg);
  65
  66 static void nfs_set_events(NFSClient *client)
  67 {
  68     int ev = nfs_which_events(client->context);
  69     if (ev != client->events) {
  70         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
  71                            false,
  72                            (ev & POLLIN) ? nfs_process_read : NULL,
  73                            (ev & POLLOUT) ? nfs_process_write : NULL, client);
  74
  75     }
  76     client->events = ev;
  77 }
  78
  79 static void nfs_process_read(void *arg)
  80 {
  81     NFSClient *client = arg;
  82     nfs_service(client->context, POLLIN);
  83     nfs_set_events(client);
  84 }
  85
  86 static void nfs_process_write(void *arg)
  87 {
  88     NFSClient *client = arg;
  89     nfs_service(client->context, POLLOUT);
  90     nfs_set_events(client);
  91 }
  92
  93 static void nfs_co_init_task(NFSClient *client, NFSRPC *task)
  94 {
  95     *task = (NFSRPC) {
  96         .co             = qemu_coroutine_self(),
  97         .client         = client,
  98     };
  99 }
 100
 101 static void nfs_co_generic_bh_cb(void *opaque)
 102 {
 103     NFSRPC *task = opaque;
 104     task->complete = 1;
 105     qemu_bh_delete(task->bh);
 106     qemu_coroutine_enter(task->co, NULL);
 107 }
 108
 109 static void
 110 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
 111                   void *private_data)
 112 {
 113     NFSRPC *task = private_data;
 114     task->ret = ret;
 115     if (task->ret > 0 && task->iov) {
 116         if (task->ret <= task->iov->size) {
 117             qemu_iovec_from_buf(task->iov, 0, data, task->ret);
 118         } else {
 119             task->ret = -EIO;
 120         }
 121     }
 122     if (task->ret == 0 && task->st) {
 123         memcpy(task->st, data, sizeof(struct stat));
 124     }
 125     if (task->ret < 0) {
 126         error_report("NFS Error: %s", nfs_get_error(nfs));
 127     }
 128     if (task->co) {
 129         task->bh = aio_bh_new(task->client->aio_context,
 130                               nfs_co_generic_bh_cb, task);
 131         qemu_bh_schedule(task->bh);
 132     } else {
 133         task->complete = 1;
 134     }
 135 }
 136
 137 static int coroutine_fn nfs_co_readv(BlockDriverState *bs,
 138                                      int64_t sector_num, int nb_sectors,
 139                                      QEMUIOVector *iov)
 140 {
 141     NFSClient *client = bs->opaque;
 142     NFSRPC task;
 143
 144     nfs_co_init_task(client, &task);
 145     task.iov = iov;
 146
 147     if (nfs_pread_async(client->context, client->fh,
 148                         sector_num * BDRV_SECTOR_SIZE,
 149                         nb_sectors * BDRV_SECTOR_SIZE,
 150                         nfs_co_generic_cb, &task) != 0) {
 151         return -ENOMEM;
 152     }
 153
 154     while (!task.complete) {
 155         nfs_set_events(client);
 156         qemu_coroutine_yield();
 157     }
 158
 159     if (task.ret < 0) {
 160         return task.ret;
 161     }
 162
 163     /* zero pad short reads */
 164     if (task.ret < iov->size) {
 165         qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
 166     }
 167
 168     return 0;
 169 }
 170
 171 static int coroutine_fn nfs_co_writev(BlockDriverState *bs,
 172                                         int64_t sector_num, int nb_sectors,
 173                                         QEMUIOVector *iov)
 174 {
 175     NFSClient *client = bs->opaque;
 176     NFSRPC task;
 177     char *buf = NULL;
 178
 179     nfs_co_init_task(client, &task);
 180
 181     buf = g_try_malloc(nb_sectors * BDRV_SECTOR_SIZE);
 182     if (nb_sectors && buf == NULL) {
 183         return -ENOMEM;
 184     }
 185
 186     qemu_iovec_to_buf(iov, 0, buf, nb_sectors * BDRV_SECTOR_SIZE);
 187
 188     if (nfs_pwrite_async(client->context, client->fh,
 189                          sector_num * BDRV_SECTOR_SIZE,
 190                          nb_sectors * BDRV_SECTOR_SIZE,
 191                          buf, nfs_co_generic_cb, &task) != 0) {
 192         g_free(buf);
 193         return -ENOMEM;
 194     }
 195
 196     while (!task.complete) {
 197         nfs_set_events(client);
 198         qemu_coroutine_yield();
 199     }
 200
 201     g_free(buf);
 202
 203     if (task.ret != nb_sectors * BDRV_SECTOR_SIZE) {
 204         return task.ret < 0 ? task.ret : -EIO;
 205     }
 206
 207     return 0;
 208 }
 209
 210 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
 211 {
 212     NFSClient *client = bs->opaque;
 213     NFSRPC task;
 214
 215     nfs_co_init_task(client, &task);
 216
 217     if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
 218                         &task) != 0) {
 219         return -ENOMEM;
 220     }
 221
 222     while (!task.complete) {
 223         nfs_set_events(client);
 224         qemu_coroutine_yield();
 225     }
 226
 227     return task.ret;
 228 }
 229
 230 /* TODO Convert to fine grained options */
 231 static QemuOptsList runtime_opts = {
 232     .name = "nfs",
 233     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 234     .desc = {
 235         {
 236             .name = "filename",
 237             .type = QEMU_OPT_STRING,
 238             .help = "URL to the NFS file",
 239         },
 240         { /* end of list */ }
 241     },
 242 };
 243
 244 static void nfs_detach_aio_context(BlockDriverState *bs)
 245 {
 246     NFSClient *client = bs->opaque;
 247
 248     aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 249                        false, NULL, NULL, NULL);
 250     client->events = 0;
 251 }
 252
 253 static void nfs_attach_aio_context(BlockDriverState *bs,
 254                                    AioContext *new_context)
 255 {
 256     NFSClient *client = bs->opaque;
 257
 258     client->aio_context = new_context;
 259     nfs_set_events(client);
 260 }
 261
 262 static void nfs_client_close(NFSClient *client)
 263 {
 264     if (client->context) {
 265         if (client->fh) {
 266             nfs_close(client->context, client->fh);
 267         }
 268         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 269                            false, NULL, NULL, NULL);
 270         nfs_destroy_context(client->context);
 271     }
 272     memset(client, 0, sizeof(NFSClient));
 273 }
 274
 275 static void nfs_file_close(BlockDriverState *bs)
 276 {
 277     NFSClient *client = bs->opaque;
 278     nfs_client_close(client);
 279 }
 280
 281 static int64_t nfs_client_open(NFSClient *client, const char *filename,
 282                                int flags, Error **errp, int open_flags)
 283 {
 284     int ret = -EINVAL, i;
 285     struct stat st;
 286     URI *uri;
 287     QueryParams *qp = NULL;
 288     char *file = NULL, *strp = NULL;
 289
 290     uri = uri_parse(filename);
 291     if (!uri) {
 292         error_setg(errp, "Invalid URL specified");
 293         goto fail;
 294     }
 295     if (!uri->server) {
 296         error_setg(errp, "Invalid URL specified");
 297         goto fail;
 298     }
 299     strp = strrchr(uri->path, '/');
 300     if (strp == NULL) {
 301         error_setg(errp, "Invalid URL specified");
 302         goto fail;
 303     }
 304     file = g_strdup(strp);
 305     *strp = 0;
 306
 307     client->context = nfs_init_context();
 308     if (client->context == NULL) {
 309         error_setg(errp, "Failed to init NFS context");
 310         goto fail;
 311     }
 312
 313     qp = query_params_parse(uri->query);
 314     for (i = 0; i < qp->n; i++) {
 315         unsigned long long val;
 316         if (!qp->p[i].value) {
 317             error_setg(errp, "Value for NFS parameter expected: %s",
 318                        qp->p[i].name);
 319             goto fail;
 320         }
 321         if (parse_uint_full(qp->p[i].value, &val, 0)) {
 322             error_setg(errp, "Illegal value for NFS parameter: %s",
 323                        qp->p[i].name);
 324             goto fail;
 325         }
 326         if (!strcmp(qp->p[i].name, "uid")) {
 327             nfs_set_uid(client->context, val);
 328         } else if (!strcmp(qp->p[i].name, "gid")) {
 329             nfs_set_gid(client->context, val);
 330         } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
 331             nfs_set_tcp_syncnt(client->context, val);
 332 #ifdef LIBNFS_FEATURE_READAHEAD
 333         } else if (!strcmp(qp->p[i].name, "readahead")) {
 334             if (open_flags & BDRV_O_NOCACHE) {
 335                 error_setg(errp, "Cannot enable NFS readahead "
 336                                  "if cache.direct = on");
 337                 goto fail;
 338             }
 339             if (val > QEMU_NFS_MAX_READAHEAD_SIZE) {
 340                 error_report("NFS Warning: Truncating NFS readahead"
 341                              " size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
 342                 val = QEMU_NFS_MAX_READAHEAD_SIZE;
 343             }
 344             nfs_set_readahead(client->context, val);
 345             client->cache_used = true;
 346 #endif
 347 #ifdef LIBNFS_FEATURE_DEBUG
 348         } else if (!strcmp(qp->p[i].name, "debug")) {
 349             /* limit the maximum debug level to avoid potential flooding
 350              * of our log files. */
 351             if (val > QEMU_NFS_MAX_DEBUG_LEVEL) {
 352                 error_report("NFS Warning: Limiting NFS debug level"
 353                              " to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
 354                 val = QEMU_NFS_MAX_DEBUG_LEVEL;
 355             }
 356             nfs_set_debug(client->context, val);
 357 #endif
 358         } else {
 359             error_setg(errp, "Unknown NFS parameter name: %s",
 360                        qp->p[i].name);
 361             goto fail;
 362         }
 363     }
 364
 365     ret = nfs_mount(client->context, uri->server, uri->path);
 366     if (ret < 0) {
 367         error_setg(errp, "Failed to mount nfs share: %s",
 368                    nfs_get_error(client->context));
 369         goto fail;
 370     }
 371
 372     if (flags & O_CREAT) {
 373         ret = nfs_creat(client->context, file, 0600, &client->fh);
 374         if (ret < 0) {
 375             error_setg(errp, "Failed to create file: %s",
 376                        nfs_get_error(client->context));
 377             goto fail;
 378         }
 379     } else {
 380         ret = nfs_open(client->context, file, flags, &client->fh);
 381         if (ret < 0) {
 382             error_setg(errp, "Failed to open file : %s",
 383                        nfs_get_error(client->context));
 384             goto fail;
 385         }
 386     }
 387
 388     ret = nfs_fstat(client->context, client->fh, &st);
 389     if (ret < 0) {
 390         error_setg(errp, "Failed to fstat file: %s",
 391                    nfs_get_error(client->context));
 392         goto fail;
 393     }
 394
 395     ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
 396     client->st_blocks = st.st_blocks;
 397     client->has_zero_init = S_ISREG(st.st_mode);
 398     goto out;
 399 fail:
 400     nfs_client_close(client);
 401 out:
 402     if (qp) {
 403         query_params_free(qp);
 404     }
 405     uri_free(uri);
 406     g_free(file);
 407     return ret;
 408 }
 409
 410 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
 411                          Error **errp) {
 412     NFSClient *client = bs->opaque;
 413     int64_t ret;
 414     QemuOpts *opts;
 415     Error *local_err = NULL;
 416
 417     client->aio_context = bdrv_get_aio_context(bs);
 418
 419     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 420     qemu_opts_absorb_qdict(opts, options, &local_err);
 421     if (local_err) {
 422         error_propagate(errp, local_err);
 423         ret = -EINVAL;
 424         goto out;
 425     }
 426     ret = nfs_client_open(client, qemu_opt_get(opts, "filename"),
 427                           (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
 428                           errp, bs->open_flags);
 429     if (ret < 0) {
 430         goto out;
 431     }
 432     bs->total_sectors = ret;
 433     ret = 0;
 434 out:
 435     qemu_opts_del(opts);
 436     return ret;
 437 }
 438
 439 static QemuOptsList nfs_create_opts = {
 440     .name = "nfs-create-opts",
 441     .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
 442     .desc = {
 443         {
 444             .name = BLOCK_OPT_SIZE,
 445             .type = QEMU_OPT_SIZE,
 446             .help = "Virtual disk size"
 447         },
 448         { /* end of list */ }
 449     }
 450 };
 451
 452 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
 453 {
 454     int ret = 0;
 455     int64_t total_size = 0;
 456     NFSClient *client = g_new0(NFSClient, 1);
 457
 458     client->aio_context = qemu_get_aio_context();
 459
 460     /* Read out options */
 461     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 462                           BDRV_SECTOR_SIZE);
 463
 464     ret = nfs_client_open(client, url, O_CREAT, errp, 0);
 465     if (ret < 0) {
 466         goto out;
 467     }
 468     ret = nfs_ftruncate(client->context, client->fh, total_size);
 469     nfs_client_close(client);
 470 out:
 471     g_free(client);
 472     return ret;
 473 }
 474
 475 static int nfs_has_zero_init(BlockDriverState *bs)
 476 {
 477     NFSClient *client = bs->opaque;
 478     return client->has_zero_init;
 479 }
 480
 481 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
 482 {
 483     NFSClient *client = bs->opaque;
 484     NFSRPC task = {0};
 485     struct stat st;
 486
 487     if (bdrv_is_read_only(bs) &&
 488         !(bs->open_flags & BDRV_O_NOCACHE)) {
 489         return client->st_blocks * 512;
 490     }
 491
 492     task.st = &st;
 493     if (nfs_fstat_async(client->context, client->fh, nfs_co_generic_cb,
 494                         &task) != 0) {
 495         return -ENOMEM;
 496     }
 497
 498     while (!task.complete) {
 499         nfs_set_events(client);
 500         aio_poll(client->aio_context, true);
 501     }
 502
 503     return (task.ret < 0 ? task.ret : st.st_blocks * 512);
 504 }
 505
 506 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
 507 {
 508     NFSClient *client = bs->opaque;
 509     return nfs_ftruncate(client->context, client->fh, offset);
 510 }
 511
 512 /* Note that this will not re-establish a connection with the NFS server
 513  * - it is effectively a NOP.  */
 514 static int nfs_reopen_prepare(BDRVReopenState *state,
 515                               BlockReopenQueue *queue, Error **errp)
 516 {
 517     NFSClient *client = state->bs->opaque;
 518     struct stat st;
 519     int ret = 0;
 520
 521     if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
 522         error_setg(errp, "Cannot open a read-only mount as read-write");
 523         return -EACCES;
 524     }
 525
 526     if ((state->flags & BDRV_O_NOCACHE) && client->cache_used) {
 527         error_setg(errp, "Cannot disable cache if libnfs readahead is enabled");
 528         return -EINVAL;
 529     }
 530
 531     /* Update cache for read-only reopens */
 532     if (!(state->flags & BDRV_O_RDWR)) {
 533         ret = nfs_fstat(client->context, client->fh, &st);
 534         if (ret < 0) {
 535             error_setg(errp, "Failed to fstat file: %s",
 536                        nfs_get_error(client->context));
 537             return ret;
 538         }
 539         client->st_blocks = st.st_blocks;
 540     }
 541
 542     return 0;
 543 }
 544
 545 static BlockDriver bdrv_nfs = {
 546     .format_name                    = "nfs",
 547     .protocol_name                  = "nfs",
 548
 549     .instance_size                  = sizeof(NFSClient),
 550     .bdrv_needs_filename            = true,
 551     .create_opts                    = &nfs_create_opts,
 552
 553     .bdrv_has_zero_init             = nfs_has_zero_init,
 554     .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
 555     .bdrv_truncate                  = nfs_file_truncate,
 556
 557     .bdrv_file_open                 = nfs_file_open,
 558     .bdrv_close                     = nfs_file_close,
 559     .bdrv_create                    = nfs_file_create,
 560     .bdrv_reopen_prepare            = nfs_reopen_prepare,
 561
 562     .bdrv_co_readv                  = nfs_co_readv,
 563     .bdrv_co_writev                 = nfs_co_writev,
 564     .bdrv_co_flush_to_disk          = nfs_co_flush,
 565
 566     .bdrv_detach_aio_context        = nfs_detach_aio_context,
 567     .bdrv_attach_aio_context        = nfs_attach_aio_context,
 568 };
 569
 570 static void nfs_block_init(void)
 571 {
 572     bdrv_register(&bdrv_nfs);
 573 }
 574
 575 block_init(nfs_block_init);