fs/nfs/read.c

   1 /*
   2  * linux/fs/nfs/read.c
   3  *
   4  * Block I/O for NFS
   5  *
   6  * Partial copy of Linus' read cache modifications to fs/nfs/file.c
   7  * modified for async RPC by okir@monad.swb.de
   8  */
   9
  10 #include <linux/time.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/fcntl.h>
  14 #include <linux/stat.h>
  15 #include <linux/mm.h>
  16 #include <linux/slab.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/sunrpc/clnt.h>
  19 #include <linux/nfs_fs.h>
  20 #include <linux/nfs_page.h>
  21 #include <linux/module.h>
  22
  23 #include "nfs4_fs.h"
  24 #include "internal.h"
  25 #include "iostat.h"
  26 #include "fscache.h"
  27
  28 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
  29
  30 static const struct nfs_pageio_ops nfs_pageio_read_ops;
  31 static const struct rpc_call_ops nfs_read_common_ops;
  32 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
  33
  34 static struct kmem_cache *nfs_rdata_cachep;
  35
  36 struct nfs_read_header *nfs_readhdr_alloc(void)
  37 {
  38         struct nfs_read_header *rhdr;
  39
  40         rhdr = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
  41         if (rhdr) {
  42                 struct nfs_pgio_header *hdr = &rhdr->header;
  43
  44                 INIT_LIST_HEAD(&hdr->pages);
  45                 INIT_LIST_HEAD(&hdr->rpc_list);
  46                 spin_lock_init(&hdr->lock);
  47                 atomic_set(&hdr->refcnt, 0);
  48         }
  49         return rhdr;
  50 }
  51 EXPORT_SYMBOL_GPL(nfs_readhdr_alloc);
  52
  53 static struct nfs_read_data *nfs_readdata_alloc(struct nfs_pgio_header *hdr,
  54                                                 unsigned int pagecount)
  55 {
  56         struct nfs_read_data *data, *prealloc;
  57
  58         prealloc = &container_of(hdr, struct nfs_read_header, header)->rpc_data;
  59         if (prealloc->header == NULL)
  60                 data = prealloc;
  61         else
  62                 data = kzalloc(sizeof(*data), GFP_KERNEL);
  63         if (!data)
  64                 goto out;
  65
  66         if (nfs_pgarray_set(&data->pages, pagecount)) {
  67                 data->header = hdr;
  68                 atomic_inc(&hdr->refcnt);
  69         } else {
  70                 if (data != prealloc)
  71                         kfree(data);
  72                 data = NULL;
  73         }
  74 out:
  75         return data;
  76 }
  77
  78 void nfs_readhdr_free(struct nfs_pgio_header *hdr)
  79 {
  80         struct nfs_read_header *rhdr = container_of(hdr, struct nfs_read_header, header);
  81
  82         kmem_cache_free(nfs_rdata_cachep, rhdr);
  83 }
  84 EXPORT_SYMBOL_GPL(nfs_readhdr_free);
  85
  86 void nfs_readdata_release(struct nfs_read_data *rdata)
  87 {
  88         struct nfs_pgio_header *hdr = rdata->header;
  89         struct nfs_read_header *read_header = container_of(hdr, struct nfs_read_header, header);
  90
  91         put_nfs_open_context(rdata->args.context);
  92         if (rdata->pages.pagevec != rdata->pages.page_array)
  93                 kfree(rdata->pages.pagevec);
  94         if (rdata == &read_header->rpc_data) {
  95                 rdata->header = NULL;
  96                 rdata = NULL;
  97         }
  98         if (atomic_dec_and_test(&hdr->refcnt))
  99                 hdr->completion_ops->completion(hdr);
 100         /* Note: we only free the rpc_task after callbacks are done.
 101          * See the comment in rpc_free_task() for why
 102          */
 103         kfree(rdata);
 104 }
 105 EXPORT_SYMBOL_GPL(nfs_readdata_release);
 106
 107 static
 108 int nfs_return_empty_page(struct page *page)
 109 {
 110         zero_user(page, 0, PAGE_CACHE_SIZE);
 111         SetPageUptodate(page);
 112         unlock_page(page);
 113         return 0;
 114 }
 115
 116 void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
 117                               struct inode *inode,
 118                               const struct nfs_pgio_completion_ops *compl_ops)
 119 {
 120         nfs_pageio_init(pgio, inode, &nfs_pageio_read_ops, compl_ops,
 121                         NFS_SERVER(inode)->rsize, 0);
 122 }
 123 EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
 124
 125 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
 126 {
 127         pgio->pg_ops = &nfs_pageio_read_ops;
 128         pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
 129 }
 130 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
 131
 132 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
 133                        struct page *page)
 134 {
 135         struct nfs_page *new;
 136         unsigned int len;
 137         struct nfs_pageio_descriptor pgio;
 138
 139         len = nfs_page_length(page);
 140         if (len == 0)
 141                 return nfs_return_empty_page(page);
 142         new = nfs_create_request(ctx, inode, page, 0, len);
 143         if (IS_ERR(new)) {
 144                 unlock_page(page);
 145                 return PTR_ERR(new);
 146         }
 147         if (len < PAGE_CACHE_SIZE)
 148                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 149
 150         NFS_PROTO(inode)->read_pageio_init(&pgio, inode, &nfs_async_read_completion_ops);
 151         nfs_pageio_add_request(&pgio, new);
 152         nfs_pageio_complete(&pgio);
 153         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 154         return 0;
 155 }
 156
 157 static void nfs_readpage_release(struct nfs_page *req)
 158 {
 159         struct inode *d_inode = req->wb_context->dentry->d_inode;
 160
 161         if (PageUptodate(req->wb_page))
 162                 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
 163
 164         unlock_page(req->wb_page);
 165
 166         dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
 167                         req->wb_context->dentry->d_inode->i_sb->s_id,
 168                         (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
 169                         req->wb_bytes,
 170                         (long long)req_offset(req));
 171         nfs_release_request(req);
 172 }
 173
 174 /* Note io was page aligned */
 175 static void nfs_read_completion(struct nfs_pgio_header *hdr)
 176 {
 177         unsigned long bytes = 0;
 178
 179         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
 180                 goto out;
 181         while (!list_empty(&hdr->pages)) {
 182                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
 183                 struct page *page = req->wb_page;
 184
 185                 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
 186                         if (bytes > hdr->good_bytes)
 187                                 zero_user(page, 0, PAGE_SIZE);
 188                         else if (hdr->good_bytes - bytes < PAGE_SIZE)
 189                                 zero_user_segment(page,
 190                                         hdr->good_bytes & ~PAGE_MASK,
 191                                         PAGE_SIZE);
 192                 }
 193                 bytes += req->wb_bytes;
 194                 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
 195                         if (bytes <= hdr->good_bytes)
 196                                 SetPageUptodate(page);
 197                 } else
 198                         SetPageUptodate(page);
 199                 nfs_list_remove_request(req);
 200                 nfs_readpage_release(req);
 201         }
 202 out:
 203         hdr->release(hdr);
 204 }
 205
 206 int nfs_initiate_read(struct rpc_clnt *clnt,
 207                       struct nfs_read_data *data,
 208                       const struct rpc_call_ops *call_ops, int flags)
 209 {
 210         struct inode *inode = data->header->inode;
 211         int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
 212         struct rpc_task *task;
 213         struct rpc_message msg = {
 214                 .rpc_argp = &data->args,
 215                 .rpc_resp = &data->res,
 216                 .rpc_cred = data->header->cred,
 217         };
 218         struct rpc_task_setup task_setup_data = {
 219                 .task = &data->task,
 220                 .rpc_client = clnt,
 221                 .rpc_message = &msg,
 222                 .callback_ops = call_ops,
 223                 .callback_data = data,
 224                 .workqueue = nfsiod_workqueue,
 225                 .flags = RPC_TASK_ASYNC | swap_flags | flags,
 226         };
 227
 228         /* Set up the initial task struct. */
 229         NFS_PROTO(inode)->read_setup(data, &msg);
 230
 231         dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
 232                         "offset %llu)\n",
 233                         data->task.tk_pid,
 234                         inode->i_sb->s_id,
 235                         (long long)NFS_FILEID(inode),
 236                         data->args.count,
 237                         (unsigned long long)data->args.offset);
 238
 239         task = rpc_run_task(&task_setup_data);
 240         if (IS_ERR(task))
 241                 return PTR_ERR(task);
 242         rpc_put_task(task);
 243         return 0;
 244 }
 245 EXPORT_SYMBOL_GPL(nfs_initiate_read);
 246
 247 /*
 248  * Set up the NFS read request struct
 249  */
 250 static void nfs_read_rpcsetup(struct nfs_read_data *data,
 251                 unsigned int count, unsigned int offset)
 252 {
 253         struct nfs_page *req = data->header->req;
 254
 255         data->args.fh     = NFS_FH(data->header->inode);
 256         data->args.offset = req_offset(req) + offset;
 257         data->args.pgbase = req->wb_pgbase + offset;
 258         data->args.pages  = data->pages.pagevec;
 259         data->args.count  = count;
 260         data->args.context = get_nfs_open_context(req->wb_context);
 261         data->args.lock_context = req->wb_lock_context;
 262
 263         data->res.fattr   = &data->fattr;
 264         data->res.count   = count;
 265         data->res.eof     = 0;
 266         nfs_fattr_init(&data->fattr);
 267 }
 268
 269 static int nfs_do_read(struct nfs_read_data *data,
 270                 const struct rpc_call_ops *call_ops)
 271 {
 272         struct inode *inode = data->header->inode;
 273
 274         return nfs_initiate_read(NFS_CLIENT(inode), data, call_ops, 0);
 275 }
 276
 277 static int
 278 nfs_do_multiple_reads(struct list_head *head,
 279                 const struct rpc_call_ops *call_ops)
 280 {
 281         struct nfs_read_data *data;
 282         int ret = 0;
 283
 284         while (!list_empty(head)) {
 285                 int ret2;
 286
 287                 data = list_first_entry(head, struct nfs_read_data, list);
 288                 list_del_init(&data->list);
 289
 290                 ret2 = nfs_do_read(data, call_ops);
 291                 if (ret == 0)
 292                         ret = ret2;
 293         }
 294         return ret;
 295 }
 296
 297 static void
 298 nfs_async_read_error(struct list_head *head)
 299 {
 300         struct nfs_page *req;
 301
 302         while (!list_empty(head)) {
 303                 req = nfs_list_entry(head->next);
 304                 nfs_list_remove_request(req);
 305                 nfs_readpage_release(req);
 306         }
 307 }
 308
 309 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
 310         .error_cleanup = nfs_async_read_error,
 311         .completion = nfs_read_completion,
 312 };
 313
 314 static void nfs_pagein_error(struct nfs_pageio_descriptor *desc,
 315                 struct nfs_pgio_header *hdr)
 316 {
 317         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 318         while (!list_empty(&hdr->rpc_list)) {
 319                 struct nfs_read_data *data = list_first_entry(&hdr->rpc_list,
 320                                 struct nfs_read_data, list);
 321                 list_del(&data->list);
 322                 nfs_readdata_release(data);
 323         }
 324         desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 325 }
 326
 327 /*
 328  * Generate multiple requests to fill a single page.
 329  *
 330  * We optimize to reduce the number of read operations on the wire.  If we
 331  * detect that we're reading a page, or an area of a page, that is past the
 332  * end of file, we do not generate NFS read operations but just clear the
 333  * parts of the page that would have come back zero from the server anyway.
 334  *
 335  * We rely on the cached value of i_size to make this determination; another
 336  * client can fill pages on the server past our cached end-of-file, but we
 337  * won't see the new data until our attribute cache is updated.  This is more
 338  * or less conventional NFS client behavior.
 339  */
 340 static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc,
 341                             struct nfs_pgio_header *hdr)
 342 {
 343         struct nfs_page *req = hdr->req;
 344         struct page *page = req->wb_page;
 345         struct nfs_read_data *data;
 346         size_t rsize = desc->pg_bsize, nbytes;
 347         unsigned int offset;
 348
 349         offset = 0;
 350         nbytes = desc->pg_count;
 351         do {
 352                 size_t len = min(nbytes,rsize);
 353
 354                 data = nfs_readdata_alloc(hdr, 1);
 355                 if (!data) {
 356                         nfs_pagein_error(desc, hdr);
 357                         return -ENOMEM;
 358                 }
 359                 data->pages.pagevec[0] = page;
 360                 nfs_read_rpcsetup(data, len, offset);
 361                 list_add(&data->list, &hdr->rpc_list);
 362                 nbytes -= len;
 363                 offset += len;
 364         } while (nbytes != 0);
 365
 366         nfs_list_remove_request(req);
 367         nfs_list_add_request(req, &hdr->pages);
 368         desc->pg_rpc_callops = &nfs_read_common_ops;
 369         return 0;
 370 }
 371
 372 static int nfs_pagein_one(struct nfs_pageio_descriptor *desc,
 373                           struct nfs_pgio_header *hdr)
 374 {
 375         struct nfs_page         *req;
 376         struct page             **pages;
 377         struct nfs_read_data    *data;
 378         struct list_head *head = &desc->pg_list;
 379
 380         data = nfs_readdata_alloc(hdr, nfs_page_array_len(desc->pg_base,
 381                                                           desc->pg_count));
 382         if (!data) {
 383                 nfs_pagein_error(desc, hdr);
 384                 return -ENOMEM;
 385         }
 386
 387         pages = data->pages.pagevec;
 388         while (!list_empty(head)) {
 389                 req = nfs_list_entry(head->next);
 390                 nfs_list_remove_request(req);
 391                 nfs_list_add_request(req, &hdr->pages);
 392                 *pages++ = req->wb_page;
 393         }
 394
 395         nfs_read_rpcsetup(data, desc->pg_count, 0);
 396         list_add(&data->list, &hdr->rpc_list);
 397         desc->pg_rpc_callops = &nfs_read_common_ops;
 398         return 0;
 399 }
 400
 401 int nfs_generic_pagein(struct nfs_pageio_descriptor *desc,
 402                        struct nfs_pgio_header *hdr)
 403 {
 404         if (desc->pg_bsize < PAGE_CACHE_SIZE)
 405                 return nfs_pagein_multi(desc, hdr);
 406         return nfs_pagein_one(desc, hdr);
 407 }
 408 EXPORT_SYMBOL_GPL(nfs_generic_pagein);
 409
 410 static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
 411 {
 412         struct nfs_read_header *rhdr;
 413         struct nfs_pgio_header *hdr;
 414         int ret;
 415
 416         rhdr = nfs_readhdr_alloc();
 417         if (!rhdr) {
 418                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 419                 return -ENOMEM;
 420         }
 421         hdr = &rhdr->header;
 422         nfs_pgheader_init(desc, hdr, nfs_readhdr_free);
 423         atomic_inc(&hdr->refcnt);
 424         ret = nfs_generic_pagein(desc, hdr);
 425         if (ret == 0)
 426                 ret = nfs_do_multiple_reads(&hdr->rpc_list,
 427                                             desc->pg_rpc_callops);
 428         if (atomic_dec_and_test(&hdr->refcnt))
 429                 hdr->completion_ops->completion(hdr);
 430         return ret;
 431 }
 432
 433 static const struct nfs_pageio_ops nfs_pageio_read_ops = {
 434         .pg_test = nfs_generic_pg_test,
 435         .pg_doio = nfs_generic_pg_readpages,
 436 };
 437
 438 /*
 439  * This is the callback from RPC telling us whether a reply was
 440  * received or some error occurred (timeout or socket shutdown).
 441  */
 442 int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
 443 {
 444         struct inode *inode = data->header->inode;
 445         int status;
 446
 447         dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
 448                         task->tk_status);
 449
 450         status = NFS_PROTO(inode)->read_done(task, data);
 451         if (status != 0)
 452                 return status;
 453
 454         nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, data->res.count);
 455
 456         if (task->tk_status == -ESTALE) {
 457                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 458                 nfs_mark_for_revalidate(inode);
 459         }
 460         return 0;
 461 }
 462
 463 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
 464 {
 465         struct nfs_readargs *argp = &data->args;
 466         struct nfs_readres *resp = &data->res;
 467
 468         /* This is a short read! */
 469         nfs_inc_stats(data->header->inode, NFSIOS_SHORTREAD);
 470         /* Has the server at least made some progress? */
 471         if (resp->count == 0) {
 472                 nfs_set_pgio_error(data->header, -EIO, argp->offset);
 473                 return;
 474         }
 475         /* Yes, so retry the read at the end of the data */
 476         data->mds_offset += resp->count;
 477         argp->offset += resp->count;
 478         argp->pgbase += resp->count;
 479         argp->count -= resp->count;
 480         rpc_restart_call_prepare(task);
 481 }
 482
 483 static void nfs_readpage_result_common(struct rpc_task *task, void *calldata)
 484 {
 485         struct nfs_read_data *data = calldata;
 486         struct nfs_pgio_header *hdr = data->header;
 487
 488         /* Note the only returns of nfs_readpage_result are 0 and -EAGAIN */
 489         if (nfs_readpage_result(task, data) != 0)
 490                 return;
 491         if (task->tk_status < 0)
 492                 nfs_set_pgio_error(hdr, task->tk_status, data->args.offset);
 493         else if (data->res.eof) {
 494                 loff_t bound;
 495
 496                 bound = data->args.offset + data->res.count;
 497                 spin_lock(&hdr->lock);
 498                 if (bound < hdr->io_start + hdr->good_bytes) {
 499                         set_bit(NFS_IOHDR_EOF, &hdr->flags);
 500                         clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
 501                         hdr->good_bytes = bound - hdr->io_start;
 502                 }
 503                 spin_unlock(&hdr->lock);
 504         } else if (data->res.count != data->args.count)
 505                 nfs_readpage_retry(task, data);
 506 }
 507
 508 static void nfs_readpage_release_common(void *calldata)
 509 {
 510         nfs_readdata_release(calldata);
 511 }
 512
 513 void nfs_read_prepare(struct rpc_task *task, void *calldata)
 514 {
 515         struct nfs_read_data *data = calldata;
 516         NFS_PROTO(data->header->inode)->read_rpc_prepare(task, data);
 517 }
 518
 519 static const struct rpc_call_ops nfs_read_common_ops = {
 520         .rpc_call_prepare = nfs_read_prepare,
 521         .rpc_call_done = nfs_readpage_result_common,
 522         .rpc_release = nfs_readpage_release_common,
 523 };
 524
 525 /*
 526  * Read a page over NFS.
 527  * We read the page synchronously in the following case:
 528  *  -   The error flag is set for this page. This happens only when a
 529  *      previous async read operation failed.
 530  */
 531 int nfs_readpage(struct file *file, struct page *page)
 532 {
 533         struct nfs_open_context *ctx;
 534         struct inode *inode = page_file_mapping(page)->host;
 535         int             error;
 536
 537         dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
 538                 page, PAGE_CACHE_SIZE, page_file_index(page));
 539         nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
 540         nfs_add_stats(inode, NFSIOS_READPAGES, 1);
 541
 542         /*
 543          * Try to flush any pending writes to the file..
 544          *
 545          * NOTE! Because we own the page lock, there cannot
 546          * be any new pending writes generated at this point
 547          * for this page (other pages can be written to).
 548          */
 549         error = nfs_wb_page(inode, page);
 550         if (error)
 551                 goto out_unlock;
 552         if (PageUptodate(page))
 553                 goto out_unlock;
 554
 555         error = -ESTALE;
 556         if (NFS_STALE(inode))
 557                 goto out_unlock;
 558
 559         if (file == NULL) {
 560                 error = -EBADF;
 561                 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 562                 if (ctx == NULL)
 563                         goto out_unlock;
 564         } else
 565                 ctx = get_nfs_open_context(nfs_file_open_context(file));
 566
 567         if (!IS_SYNC(inode)) {
 568                 error = nfs_readpage_from_fscache(ctx, inode, page);
 569                 if (error == 0)
 570                         goto out;
 571         }
 572
 573         error = nfs_readpage_async(ctx, inode, page);
 574
 575 out:
 576         put_nfs_open_context(ctx);
 577         return error;
 578 out_unlock:
 579         unlock_page(page);
 580         return error;
 581 }
 582
 583 struct nfs_readdesc {
 584         struct nfs_pageio_descriptor *pgio;
 585         struct nfs_open_context *ctx;
 586 };
 587
 588 static int
 589 readpage_async_filler(void *data, struct page *page)
 590 {
 591         struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
 592         struct inode *inode = page_file_mapping(page)->host;
 593         struct nfs_page *new;
 594         unsigned int len;
 595         int error;
 596
 597         len = nfs_page_length(page);
 598         if (len == 0)
 599                 return nfs_return_empty_page(page);
 600
 601         new = nfs_create_request(desc->ctx, inode, page, 0, len);
 602         if (IS_ERR(new))
 603                 goto out_error;
 604
 605         if (len < PAGE_CACHE_SIZE)
 606                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 607         if (!nfs_pageio_add_request(desc->pgio, new)) {
 608                 error = desc->pgio->pg_error;
 609                 goto out_unlock;
 610         }
 611         return 0;
 612 out_error:
 613         error = PTR_ERR(new);
 614 out_unlock:
 615         unlock_page(page);
 616         return error;
 617 }
 618
 619 int nfs_readpages(struct file *filp, struct address_space *mapping,
 620                 struct list_head *pages, unsigned nr_pages)
 621 {
 622         struct nfs_pageio_descriptor pgio;
 623         struct nfs_readdesc desc = {
 624                 .pgio = &pgio,
 625         };
 626         struct inode *inode = mapping->host;
 627         unsigned long npages;
 628         int ret = -ESTALE;
 629
 630         dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
 631                         inode->i_sb->s_id,
 632                         (long long)NFS_FILEID(inode),
 633                         nr_pages);
 634         nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
 635
 636         if (NFS_STALE(inode))
 637                 goto out;
 638
 639         if (filp == NULL) {
 640                 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 641                 if (desc.ctx == NULL)
 642                         return -EBADF;
 643         } else
 644                 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
 645
 646         /* attempt to read as many of the pages as possible from the cache
 647          * - this returns -ENOBUFS immediately if the cookie is negative
 648          */
 649         ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
 650                                          pages, &nr_pages);
 651         if (ret == 0)
 652                 goto read_complete; /* all pages were read */
 653
 654         NFS_PROTO(inode)->read_pageio_init(&pgio, inode, &nfs_async_read_completion_ops);
 655
 656         ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
 657
 658         nfs_pageio_complete(&pgio);
 659         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 660         npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 661         nfs_add_stats(inode, NFSIOS_READPAGES, npages);
 662 read_complete:
 663         put_nfs_open_context(desc.ctx);
 664 out:
 665         return ret;
 666 }
 667
 668 int __init nfs_init_readpagecache(void)
 669 {
 670         nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
 671                                              sizeof(struct nfs_read_header),
 672                                              0, SLAB_HWCACHE_ALIGN,
 673                                              NULL);
 674         if (nfs_rdata_cachep == NULL)
 675                 return -ENOMEM;
 676
 677         return 0;
 678 }
 679
 680 void nfs_destroy_readpagecache(void)
 681 {
 682         kmem_cache_destroy(nfs_rdata_cachep);
 683 }