block/nfs.c

   1 /*
   2  * QEMU Block driver for native access to files on NFS shares
   3  *
   4  * Copyright (c) 2014 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 "block/block_int.h"
  32 #include "trace.h"
  33 #include "qemu/iov.h"
  34 #include "qemu/uri.h"
  35 #include "sysemu/sysemu.h"
  36 #include <nfsc/libnfs.h>
  37
  38 #define QEMU_NFS_MAX_READAHEAD_SIZE 1048576
  39 #define QEMU_NFS_MAX_DEBUG_LEVEL 2
  40
  41 typedef struct NFSClient {
  42     struct nfs_context *context;
  43     struct nfsfh *fh;
  44     int events;
  45     bool has_zero_init;
  46     AioContext *aio_context;
  47     blkcnt_t st_blocks;
  48 } NFSClient;
  49
  50 typedef struct NFSRPC {
  51     int ret;
  52     int complete;
  53     QEMUIOVector *iov;
  54     struct stat *st;
  55     Coroutine *co;
  56     QEMUBH *bh;
  57     NFSClient *client;
  58 } NFSRPC;
  59
  60 static void nfs_process_read(void *arg);
  61 static void nfs_process_write(void *arg);
  62
  63 static void nfs_set_events(NFSClient *client)
  64 {
  65     int ev = nfs_which_events(client->context);
  66     if (ev != client->events) {
  67         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
  68                            false,
  69                            (ev & POLLIN) ? nfs_process_read : NULL,
  70                            (ev & POLLOUT) ? nfs_process_write : NULL, client);
  71
  72     }
  73     client->events = ev;
  74 }
  75
  76 static void nfs_process_read(void *arg)
  77 {
  78     NFSClient *client = arg;
  79     nfs_service(client->context, POLLIN);
  80     nfs_set_events(client);
  81 }
  82
  83 static void nfs_process_write(void *arg)
  84 {
  85     NFSClient *client = arg;
  86     nfs_service(client->context, POLLOUT);
  87     nfs_set_events(client);
  88 }
  89
  90 static void nfs_co_init_task(NFSClient *client, NFSRPC *task)
  91 {
  92     *task = (NFSRPC) {
  93         .co             = qemu_coroutine_self(),
  94         .client         = client,
  95     };
  96 }
  97
  98 static void nfs_co_generic_bh_cb(void *opaque)
  99 {
 100     NFSRPC *task = opaque;
 101     task->complete = 1;
 102     qemu_bh_delete(task->bh);
 103     qemu_coroutine_enter(task->co, NULL);
 104 }
 105
 106 static void
 107 nfs_co_generic_cb(int ret, struct nfs_context *nfs, void *data,
 108                   void *private_data)
 109 {
 110     NFSRPC *task = private_data;
 111     task->ret = ret;
 112     if (task->ret > 0 && task->iov) {
 113         if (task->ret <= task->iov->size) {
 114             qemu_iovec_from_buf(task->iov, 0, data, task->ret);
 115         } else {
 116             task->ret = -EIO;
 117         }
 118     }
 119     if (task->ret == 0 && task->st) {
 120         memcpy(task->st, data, sizeof(struct stat));
 121     }
 122     if (task->ret < 0) {
 123         error_report("NFS Error: %s", nfs_get_error(nfs));
 124     }
 125     if (task->co) {
 126         task->bh = aio_bh_new(task->client->aio_context,
 127                               nfs_co_generic_bh_cb, task);
 128         qemu_bh_schedule(task->bh);
 129     } else {
 130         task->complete = 1;
 131     }
 132 }
 133
 134 static int coroutine_fn nfs_co_readv(BlockDriverState *bs,
 135                                      int64_t sector_num, int nb_sectors,
 136                                      QEMUIOVector *iov)
 137 {
 138     NFSClient *client = bs->opaque;
 139     NFSRPC task;
 140
 141     nfs_co_init_task(client, &task);
 142     task.iov = iov;
 143
 144     if (nfs_pread_async(client->context, client->fh,
 145                         sector_num * BDRV_SECTOR_SIZE,
 146                         nb_sectors * BDRV_SECTOR_SIZE,
 147                         nfs_co_generic_cb, &task) != 0) {
 148         return -ENOMEM;
 149     }
 150
 151     while (!task.complete) {
 152         nfs_set_events(client);
 153         qemu_coroutine_yield();
 154     }
 155
 156     if (task.ret < 0) {
 157         return task.ret;
 158     }
 159
 160     /* zero pad short reads */
 161     if (task.ret < iov->size) {
 162         qemu_iovec_memset(iov, task.ret, 0, iov->size - task.ret);
 163     }
 164
 165     return 0;
 166 }
 167
 168 static int coroutine_fn nfs_co_writev(BlockDriverState *bs,
 169                                         int64_t sector_num, int nb_sectors,
 170                                         QEMUIOVector *iov)
 171 {
 172     NFSClient *client = bs->opaque;
 173     NFSRPC task;
 174     char *buf = NULL;
 175
 176     nfs_co_init_task(client, &task);
 177
 178     buf = g_try_malloc(nb_sectors * BDRV_SECTOR_SIZE);
 179     if (nb_sectors && buf == NULL) {
 180         return -ENOMEM;
 181     }
 182
 183     qemu_iovec_to_buf(iov, 0, buf, nb_sectors * BDRV_SECTOR_SIZE);
 184
 185     if (nfs_pwrite_async(client->context, client->fh,
 186                          sector_num * BDRV_SECTOR_SIZE,
 187                          nb_sectors * BDRV_SECTOR_SIZE,
 188                          buf, nfs_co_generic_cb, &task) != 0) {
 189         g_free(buf);
 190         return -ENOMEM;
 191     }
 192
 193     while (!task.complete) {
 194         nfs_set_events(client);
 195         qemu_coroutine_yield();
 196     }
 197
 198     g_free(buf);
 199
 200     if (task.ret != nb_sectors * BDRV_SECTOR_SIZE) {
 201         return task.ret < 0 ? task.ret : -EIO;
 202     }
 203
 204     return 0;
 205 }
 206
 207 static int coroutine_fn nfs_co_flush(BlockDriverState *bs)
 208 {
 209     NFSClient *client = bs->opaque;
 210     NFSRPC task;
 211
 212     nfs_co_init_task(client, &task);
 213
 214     if (nfs_fsync_async(client->context, client->fh, nfs_co_generic_cb,
 215                         &task) != 0) {
 216         return -ENOMEM;
 217     }
 218
 219     while (!task.complete) {
 220         nfs_set_events(client);
 221         qemu_coroutine_yield();
 222     }
 223
 224     return task.ret;
 225 }
 226
 227 /* TODO Convert to fine grained options */
 228 static QemuOptsList runtime_opts = {
 229     .name = "nfs",
 230     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 231     .desc = {
 232         {
 233             .name = "filename",
 234             .type = QEMU_OPT_STRING,
 235             .help = "URL to the NFS file",
 236         },
 237         { /* end of list */ }
 238     },
 239 };
 240
 241 static void nfs_detach_aio_context(BlockDriverState *bs)
 242 {
 243     NFSClient *client = bs->opaque;
 244
 245     aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 246                        false, NULL, NULL, NULL);
 247     client->events = 0;
 248 }
 249
 250 static void nfs_attach_aio_context(BlockDriverState *bs,
 251                                    AioContext *new_context)
 252 {
 253     NFSClient *client = bs->opaque;
 254
 255     client->aio_context = new_context;
 256     nfs_set_events(client);
 257 }
 258
 259 static void nfs_client_close(NFSClient *client)
 260 {
 261     if (client->context) {
 262         if (client->fh) {
 263             nfs_close(client->context, client->fh);
 264         }
 265         aio_set_fd_handler(client->aio_context, nfs_get_fd(client->context),
 266                            false, NULL, NULL, NULL);
 267         nfs_destroy_context(client->context);
 268     }
 269     memset(client, 0, sizeof(NFSClient));
 270 }
 271
 272 static void nfs_file_close(BlockDriverState *bs)
 273 {
 274     NFSClient *client = bs->opaque;
 275     nfs_client_close(client);
 276 }
 277
 278 static int64_t nfs_client_open(NFSClient *client, const char *filename,
 279                                int flags, Error **errp)
 280 {
 281     int ret = -EINVAL, i;
 282     struct stat st;
 283     URI *uri;
 284     QueryParams *qp = NULL;
 285     char *file = NULL, *strp = NULL;
 286
 287     uri = uri_parse(filename);
 288     if (!uri) {
 289         error_setg(errp, "Invalid URL specified");
 290         goto fail;
 291     }
 292     if (!uri->server) {
 293         error_setg(errp, "Invalid URL specified");
 294         goto fail;
 295     }
 296     strp = strrchr(uri->path, '/');
 297     if (strp == NULL) {
 298         error_setg(errp, "Invalid URL specified");
 299         goto fail;
 300     }
 301     file = g_strdup(strp);
 302     *strp = 0;
 303
 304     client->context = nfs_init_context();
 305     if (client->context == NULL) {
 306         error_setg(errp, "Failed to init NFS context");
 307         goto fail;
 308     }
 309
 310     qp = query_params_parse(uri->query);
 311     for (i = 0; i < qp->n; i++) {
 312         unsigned long long val;
 313         if (!qp->p[i].value) {
 314             error_setg(errp, "Value for NFS parameter expected: %s",
 315                        qp->p[i].name);
 316             goto fail;
 317         }
 318         if (parse_uint_full(qp->p[i].value, &val, 0)) {
 319             error_setg(errp, "Illegal value for NFS parameter: %s",
 320                        qp->p[i].name);
 321             goto fail;
 322         }
 323         if (!strcmp(qp->p[i].name, "uid")) {
 324             nfs_set_uid(client->context, val);
 325         } else if (!strcmp(qp->p[i].name, "gid")) {
 326             nfs_set_gid(client->context, val);
 327         } else if (!strcmp(qp->p[i].name, "tcp-syncnt")) {
 328             nfs_set_tcp_syncnt(client->context, val);
 329 #ifdef LIBNFS_FEATURE_READAHEAD
 330         } else if (!strcmp(qp->p[i].name, "readahead")) {
 331             if (val > QEMU_NFS_MAX_READAHEAD_SIZE) {
 332                 error_report("NFS Warning: Truncating NFS readahead"
 333                              " size to %d", QEMU_NFS_MAX_READAHEAD_SIZE);
 334                 val = QEMU_NFS_MAX_READAHEAD_SIZE;
 335             }
 336             nfs_set_readahead(client->context, val);
 337 #endif
 338 #ifdef LIBNFS_FEATURE_DEBUG
 339         } else if (!strcmp(qp->p[i].name, "debug")) {
 340             /* limit the maximum debug level to avoid potential flooding
 341              * of our log files. */
 342             if (val > QEMU_NFS_MAX_DEBUG_LEVEL) {
 343                 error_report("NFS Warning: Limiting NFS debug level"
 344                              " to %d", QEMU_NFS_MAX_DEBUG_LEVEL);
 345                 val = QEMU_NFS_MAX_DEBUG_LEVEL;
 346             }
 347             nfs_set_debug(client->context, val);
 348 #endif
 349         } else {
 350             error_setg(errp, "Unknown NFS parameter name: %s",
 351                        qp->p[i].name);
 352             goto fail;
 353         }
 354     }
 355
 356     ret = nfs_mount(client->context, uri->server, uri->path);
 357     if (ret < 0) {
 358         error_setg(errp, "Failed to mount nfs share: %s",
 359                    nfs_get_error(client->context));
 360         goto fail;
 361     }
 362
 363     if (flags & O_CREAT) {
 364         ret = nfs_creat(client->context, file, 0600, &client->fh);
 365         if (ret < 0) {
 366             error_setg(errp, "Failed to create file: %s",
 367                        nfs_get_error(client->context));
 368             goto fail;
 369         }
 370     } else {
 371         ret = nfs_open(client->context, file, flags, &client->fh);
 372         if (ret < 0) {
 373             error_setg(errp, "Failed to open file : %s",
 374                        nfs_get_error(client->context));
 375             goto fail;
 376         }
 377     }
 378
 379     ret = nfs_fstat(client->context, client->fh, &st);
 380     if (ret < 0) {
 381         error_setg(errp, "Failed to fstat file: %s",
 382                    nfs_get_error(client->context));
 383         goto fail;
 384     }
 385
 386     ret = DIV_ROUND_UP(st.st_size, BDRV_SECTOR_SIZE);
 387     client->st_blocks = st.st_blocks;
 388     client->has_zero_init = S_ISREG(st.st_mode);
 389     goto out;
 390 fail:
 391     nfs_client_close(client);
 392 out:
 393     if (qp) {
 394         query_params_free(qp);
 395     }
 396     uri_free(uri);
 397     g_free(file);
 398     return ret;
 399 }
 400
 401 static int nfs_file_open(BlockDriverState *bs, QDict *options, int flags,
 402                          Error **errp) {
 403     NFSClient *client = bs->opaque;
 404     int64_t ret;
 405     QemuOpts *opts;
 406     Error *local_err = NULL;
 407
 408     client->aio_context = bdrv_get_aio_context(bs);
 409
 410     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 411     qemu_opts_absorb_qdict(opts, options, &local_err);
 412     if (local_err) {
 413         error_propagate(errp, local_err);
 414         ret = -EINVAL;
 415         goto out;
 416     }
 417     ret = nfs_client_open(client, qemu_opt_get(opts, "filename"),
 418                           (flags & BDRV_O_RDWR) ? O_RDWR : O_RDONLY,
 419                           errp);
 420     if (ret < 0) {
 421         goto out;
 422     }
 423     bs->total_sectors = ret;
 424     ret = 0;
 425 out:
 426     qemu_opts_del(opts);
 427     return ret;
 428 }
 429
 430 static QemuOptsList nfs_create_opts = {
 431     .name = "nfs-create-opts",
 432     .head = QTAILQ_HEAD_INITIALIZER(nfs_create_opts.head),
 433     .desc = {
 434         {
 435             .name = BLOCK_OPT_SIZE,
 436             .type = QEMU_OPT_SIZE,
 437             .help = "Virtual disk size"
 438         },
 439         { /* end of list */ }
 440     }
 441 };
 442
 443 static int nfs_file_create(const char *url, QemuOpts *opts, Error **errp)
 444 {
 445     int ret = 0;
 446     int64_t total_size = 0;
 447     NFSClient *client = g_new0(NFSClient, 1);
 448
 449     client->aio_context = qemu_get_aio_context();
 450
 451     /* Read out options */
 452     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 453                           BDRV_SECTOR_SIZE);
 454
 455     ret = nfs_client_open(client, url, O_CREAT, errp);
 456     if (ret < 0) {
 457         goto out;
 458     }
 459     ret = nfs_ftruncate(client->context, client->fh, total_size);
 460     nfs_client_close(client);
 461 out:
 462     g_free(client);
 463     return ret;
 464 }
 465
 466 static int nfs_has_zero_init(BlockDriverState *bs)
 467 {
 468     NFSClient *client = bs->opaque;
 469     return client->has_zero_init;
 470 }
 471
 472 static int64_t nfs_get_allocated_file_size(BlockDriverState *bs)
 473 {
 474     NFSClient *client = bs->opaque;
 475     NFSRPC task = {0};
 476     struct stat st;
 477
 478     if (bdrv_is_read_only(bs) &&
 479         !(bs->open_flags & BDRV_O_NOCACHE)) {
 480         return client->st_blocks * 512;
 481     }
 482
 483     task.st = &st;
 484     if (nfs_fstat_async(client->context, client->fh, nfs_co_generic_cb,
 485                         &task) != 0) {
 486         return -ENOMEM;
 487     }
 488
 489     while (!task.complete) {
 490         nfs_set_events(client);
 491         aio_poll(client->aio_context, true);
 492     }
 493
 494     return (task.ret < 0 ? task.ret : st.st_blocks * 512);
 495 }
 496
 497 static int nfs_file_truncate(BlockDriverState *bs, int64_t offset)
 498 {
 499     NFSClient *client = bs->opaque;
 500     return nfs_ftruncate(client->context, client->fh, offset);
 501 }
 502
 503 /* Note that this will not re-establish a connection with the NFS server
 504  * - it is effectively a NOP.  */
 505 static int nfs_reopen_prepare(BDRVReopenState *state,
 506                               BlockReopenQueue *queue, Error **errp)
 507 {
 508     NFSClient *client = state->bs->opaque;
 509     struct stat st;
 510     int ret = 0;
 511
 512     if (state->flags & BDRV_O_RDWR && bdrv_is_read_only(state->bs)) {
 513         error_setg(errp, "Cannot open a read-only mount as read-write");
 514         return -EACCES;
 515     }
 516
 517     /* Update cache for read-only reopens */
 518     if (!(state->flags & BDRV_O_RDWR)) {
 519         ret = nfs_fstat(client->context, client->fh, &st);
 520         if (ret < 0) {
 521             error_setg(errp, "Failed to fstat file: %s",
 522                        nfs_get_error(client->context));
 523             return ret;
 524         }
 525         client->st_blocks = st.st_blocks;
 526     }
 527
 528     return 0;
 529 }
 530
 531 static BlockDriver bdrv_nfs = {
 532     .format_name                    = "nfs",
 533     .protocol_name                  = "nfs",
 534
 535     .instance_size                  = sizeof(NFSClient),
 536     .bdrv_needs_filename            = true,
 537     .create_opts                    = &nfs_create_opts,
 538
 539     .bdrv_has_zero_init             = nfs_has_zero_init,
 540     .bdrv_get_allocated_file_size   = nfs_get_allocated_file_size,
 541     .bdrv_truncate                  = nfs_file_truncate,
 542
 543     .bdrv_file_open                 = nfs_file_open,
 544     .bdrv_close                     = nfs_file_close,
 545     .bdrv_create                    = nfs_file_create,
 546     .bdrv_reopen_prepare            = nfs_reopen_prepare,
 547
 548     .bdrv_co_readv                  = nfs_co_readv,
 549     .bdrv_co_writev                 = nfs_co_writev,
 550     .bdrv_co_flush_to_disk          = nfs_co_flush,
 551
 552     .bdrv_detach_aio_context        = nfs_detach_aio_context,
 553     .bdrv_attach_aio_context        = nfs_attach_aio_context,
 554 };
 555
 556 static void nfs_block_init(void)
 557 {
 558     bdrv_register(&bdrv_nfs);
 559 }
 560
 561 block_init(nfs_block_init);