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