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