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 "pnfs.h"
  24
  25 #include "nfs4_fs.h"
  26 #include "internal.h"
  27 #include "iostat.h"
  28 #include "fscache.h"
  29
  30 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
  31
  32 static const struct nfs_pageio_ops nfs_pageio_read_ops;
  33 static const struct rpc_call_ops nfs_read_common_ops;
  34 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
  35
  36 static struct kmem_cache *nfs_rdata_cachep;
  37
  38 struct nfs_read_header *nfs_readhdr_alloc()
  39 {
  40         struct nfs_read_header *rhdr;
  41
  42         rhdr = kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
  43         if (rhdr) {
  44                 struct nfs_pgio_header *hdr = &rhdr->header;
  45
  46                 INIT_LIST_HEAD(&hdr->pages);
  47                 INIT_LIST_HEAD(&hdr->rpc_list);
  48                 spin_lock_init(&hdr->lock);
  49                 atomic_set(&hdr->refcnt, 0);
  50         }
  51         return rhdr;
  52 }
  53
  54 struct nfs_read_data *nfs_readdata_alloc(struct nfs_pgio_header *hdr,
  55                                          unsigned int pagecount)
  56 {
  57         struct nfs_read_data *data, *prealloc;
  58
  59         prealloc = &container_of(hdr, struct nfs_read_header, header)->rpc_data;
  60         if (prealloc->header == NULL)
  61                 data = prealloc;
  62         else
  63                 data = kzalloc(sizeof(*data), GFP_KERNEL);
  64         if (!data)
  65                 goto out;
  66
  67         if (nfs_pgarray_set(&data->pages, pagecount)) {
  68                 data->header = hdr;
  69                 atomic_inc(&hdr->refcnt);
  70         } else {
  71                 if (data != prealloc)
  72                         kfree(data);
  73                 data = NULL;
  74         }
  75 out:
  76         return data;
  77 }
  78
  79 void nfs_readhdr_free(struct nfs_pgio_header *hdr)
  80 {
  81         struct nfs_read_header *rhdr = container_of(hdr, struct nfs_read_header, header);
  82
  83         kmem_cache_free(nfs_rdata_cachep, rhdr);
  84 }
  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                 kfree(rdata);
  96         else
  97                 rdata->header = NULL;
  98         if (atomic_dec_and_test(&hdr->refcnt))
  99                 hdr->completion_ops->completion(hdr);
 100 }
 101
 102 static
 103 int nfs_return_empty_page(struct page *page)
 104 {
 105         zero_user(page, 0, PAGE_CACHE_SIZE);
 106         SetPageUptodate(page);
 107         unlock_page(page);
 108         return 0;
 109 }
 110
 111 void nfs_pageio_init_read_mds(struct nfs_pageio_descriptor *pgio,
 112                               struct inode *inode,
 113                               const struct nfs_pgio_completion_ops *compl_ops)
 114 {
 115         nfs_pageio_init(pgio, inode, &nfs_pageio_read_ops, compl_ops,
 116                         NFS_SERVER(inode)->rsize, 0);
 117 }
 118
 119 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
 120 {
 121         pgio->pg_ops = &nfs_pageio_read_ops;
 122         pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
 123 }
 124 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
 125
 126 static void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
 127                                 struct inode *inode,
 128                                 const struct nfs_pgio_completion_ops *compl_ops)
 129 {
 130         if (!pnfs_pageio_init_read(pgio, inode, compl_ops))
 131                 nfs_pageio_init_read_mds(pgio, inode, compl_ops);
 132 }
 133
 134 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
 135                        struct page *page)
 136 {
 137         struct nfs_page *new;
 138         unsigned int len;
 139         struct nfs_pageio_descriptor pgio;
 140
 141         len = nfs_page_length(page);
 142         if (len == 0)
 143                 return nfs_return_empty_page(page);
 144         new = nfs_create_request(ctx, inode, page, 0, len);
 145         if (IS_ERR(new)) {
 146                 unlock_page(page);
 147                 return PTR_ERR(new);
 148         }
 149         if (len < PAGE_CACHE_SIZE)
 150                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 151
 152         nfs_pageio_init_read(&pgio, inode, &nfs_async_read_completion_ops);
 153         nfs_pageio_add_request(&pgio, new);
 154         nfs_pageio_complete(&pgio);
 155         return 0;
 156 }
 157
 158 static void nfs_readpage_release(struct nfs_page *req)
 159 {
 160         struct inode *d_inode = req->wb_context->dentry->d_inode;
 161
 162         if (PageUptodate(req->wb_page))
 163                 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
 164
 165         unlock_page(req->wb_page);
 166
 167         dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
 168                         req->wb_context->dentry->d_inode->i_sb->s_id,
 169                         (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
 170                         req->wb_bytes,
 171                         (long long)req_offset(req));
 172         nfs_release_request(req);
 173 }
 174
 175 /* Note io was page aligned */
 176 static void nfs_read_completion(struct nfs_pgio_header *hdr)
 177 {
 178         unsigned long bytes = 0;
 179
 180         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
 181                 goto out;
 182         if (!test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
 183                 while (!list_empty(&hdr->pages)) {
 184                         struct nfs_page *req = nfs_list_entry(hdr->pages.next);
 185                         struct page *page = req->wb_page;
 186
 187                         if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
 188                                 if (bytes > hdr->good_bytes)
 189                                         zero_user(page, 0, PAGE_SIZE);
 190                                 else if (hdr->good_bytes - bytes < PAGE_SIZE)
 191                                         zero_user_segment(page,
 192                                                 hdr->good_bytes & ~PAGE_MASK,
 193                                                 PAGE_SIZE);
 194                         }
 195                         SetPageUptodate(page);
 196                         nfs_list_remove_request(req);
 197                         nfs_readpage_release(req);
 198                         bytes += PAGE_SIZE;
 199                 }
 200         } else {
 201                 while (!list_empty(&hdr->pages)) {
 202                         struct nfs_page *req = nfs_list_entry(hdr->pages.next);
 203
 204                         bytes += req->wb_bytes;
 205                         if (bytes <= hdr->good_bytes)
 206                                 SetPageUptodate(req->wb_page);
 207                         nfs_list_remove_request(req);
 208                         nfs_readpage_release(req);
 209                 }
 210         }
 211 out:
 212         hdr->release(hdr);
 213 }
 214
 215 int nfs_initiate_read(struct rpc_clnt *clnt,
 216                       struct nfs_read_data *data,
 217                       const struct rpc_call_ops *call_ops)
 218 {
 219         struct inode *inode = data->header->inode;
 220         int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
 221         struct rpc_task *task;
 222         struct rpc_message msg = {
 223                 .rpc_argp = &data->args,
 224                 .rpc_resp = &data->res,
 225                 .rpc_cred = data->header->cred,
 226         };
 227         struct rpc_task_setup task_setup_data = {
 228                 .task = &data->task,
 229                 .rpc_client = clnt,
 230                 .rpc_message = &msg,
 231                 .callback_ops = call_ops,
 232                 .callback_data = data,
 233                 .workqueue = nfsiod_workqueue,
 234                 .flags = RPC_TASK_ASYNC | swap_flags,
 235         };
 236
 237         /* Set up the initial task struct. */
 238         NFS_PROTO(inode)->read_setup(data, &msg);
 239
 240         dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ "
 241                         "offset %llu)\n",
 242                         data->task.tk_pid,
 243                         inode->i_sb->s_id,
 244                         (long long)NFS_FILEID(inode),
 245                         data->args.count,
 246                         (unsigned long long)data->args.offset);
 247
 248         task = rpc_run_task(&task_setup_data);
 249         if (IS_ERR(task))
 250                 return PTR_ERR(task);
 251         rpc_put_task(task);
 252         return 0;
 253 }
 254 EXPORT_SYMBOL_GPL(nfs_initiate_read);
 255
 256 /*
 257  * Set up the NFS read request struct
 258  */
 259 static void nfs_read_rpcsetup(struct nfs_read_data *data,
 260                 unsigned int count, unsigned int offset)
 261 {
 262         struct nfs_page *req = data->header->req;
 263
 264         data->args.fh     = NFS_FH(data->header->inode);
 265         data->args.offset = req_offset(req) + offset;
 266         data->args.pgbase = req->wb_pgbase + offset;
 267         data->args.pages  = data->pages.pagevec;
 268         data->args.count  = count;
 269         data->args.context = get_nfs_open_context(req->wb_context);
 270         data->args.lock_context = req->wb_lock_context;
 271
 272         data->res.fattr   = &data->fattr;
 273         data->res.count   = count;
 274         data->res.eof     = 0;
 275         nfs_fattr_init(&data->fattr);
 276 }
 277
 278 static int nfs_do_read(struct nfs_read_data *data,
 279                 const struct rpc_call_ops *call_ops)
 280 {
 281         struct inode *inode = data->header->inode;
 282
 283         return nfs_initiate_read(NFS_CLIENT(inode), data, call_ops);
 284 }
 285
 286 static int
 287 nfs_do_multiple_reads(struct list_head *head,
 288                 const struct rpc_call_ops *call_ops)
 289 {
 290         struct nfs_read_data *data;
 291         int ret = 0;
 292
 293         while (!list_empty(head)) {
 294                 int ret2;
 295
 296                 data = list_first_entry(head, struct nfs_read_data, list);
 297                 list_del_init(&data->list);
 298
 299                 ret2 = nfs_do_read(data, call_ops);
 300                 if (ret == 0)
 301                         ret = ret2;
 302         }
 303         return ret;
 304 }
 305
 306 static void
 307 nfs_async_read_error(struct list_head *head)
 308 {
 309         struct nfs_page *req;
 310
 311         while (!list_empty(head)) {
 312                 req = nfs_list_entry(head->next);
 313                 nfs_list_remove_request(req);
 314                 nfs_readpage_release(req);
 315         }
 316 }
 317
 318 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
 319         .error_cleanup = nfs_async_read_error,
 320         .completion = nfs_read_completion,
 321 };
 322
 323 /*
 324  * Generate multiple requests to fill a single page.
 325  *
 326  * We optimize to reduce the number of read operations on the wire.  If we
 327  * detect that we're reading a page, or an area of a page, that is past the
 328  * end of file, we do not generate NFS read operations but just clear the
 329  * parts of the page that would have come back zero from the server anyway.
 330  *
 331  * We rely on the cached value of i_size to make this determination; another
 332  * client can fill pages on the server past our cached end-of-file, but we
 333  * won't see the new data until our attribute cache is updated.  This is more
 334  * or less conventional NFS client behavior.
 335  */
 336 static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc,
 337                             struct nfs_pgio_header *hdr)
 338 {
 339         struct nfs_page *req = hdr->req;
 340         struct page *page = req->wb_page;
 341         struct nfs_read_data *data;
 342         size_t rsize = desc->pg_bsize, nbytes;
 343         unsigned int offset;
 344         int requests = 0;
 345         int ret = 0;
 346
 347         nfs_list_remove_request(req);
 348         nfs_list_add_request(req, &hdr->pages);
 349
 350         offset = 0;
 351         nbytes = desc->pg_count;
 352         do {
 353                 size_t len = min(nbytes,rsize);
 354
 355                 data = nfs_readdata_alloc(hdr, 1);
 356                 if (!data)
 357                         goto out_bad;
 358                 data->pages.pagevec[0] = page;
 359                 nfs_read_rpcsetup(data, len, offset);
 360                 list_add(&data->list, &hdr->rpc_list);
 361                 requests++;
 362                 nbytes -= len;
 363                 offset += len;
 364         } while(nbytes != 0);
 365         desc->pg_rpc_callops = &nfs_read_common_ops;
 366         return ret;
 367 out_bad:
 368         while (!list_empty(&hdr->rpc_list)) {
 369                 data = list_first_entry(&hdr->rpc_list, struct nfs_read_data, list);
 370                 list_del(&data->list);
 371                 nfs_readdata_release(data);
 372         }
 373         desc->pg_completion_ops->error_cleanup(&hdr->pages);
 374         return -ENOMEM;
 375 }
 376
 377 static int nfs_pagein_one(struct nfs_pageio_descriptor *desc,
 378                           struct nfs_pgio_header *hdr)
 379 {
 380         struct nfs_page         *req;
 381         struct page             **pages;
 382         struct nfs_read_data    *data;
 383         struct list_head *head = &desc->pg_list;
 384         int ret = 0;
 385
 386         data = nfs_readdata_alloc(hdr, nfs_page_array_len(desc->pg_base,
 387                                                           desc->pg_count));
 388         if (!data) {
 389                 desc->pg_completion_ops->error_cleanup(head);
 390                 ret = -ENOMEM;
 391                 goto out;
 392         }
 393
 394         pages = data->pages.pagevec;
 395         while (!list_empty(head)) {
 396                 req = nfs_list_entry(head->next);
 397                 nfs_list_remove_request(req);
 398                 nfs_list_add_request(req, &hdr->pages);
 399                 *pages++ = req->wb_page;
 400         }
 401
 402         nfs_read_rpcsetup(data, desc->pg_count, 0);
 403         list_add(&data->list, &hdr->rpc_list);
 404         desc->pg_rpc_callops = &nfs_read_common_ops;
 405 out:
 406         return ret;
 407 }
 408
 409 int nfs_generic_pagein(struct nfs_pageio_descriptor *desc,
 410                        struct nfs_pgio_header *hdr)
 411 {
 412         if (desc->pg_bsize < PAGE_CACHE_SIZE)
 413                 return nfs_pagein_multi(desc, hdr);
 414         return nfs_pagein_one(desc, hdr);
 415 }
 416
 417 static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
 418 {
 419         struct nfs_read_header *rhdr;
 420         struct nfs_pgio_header *hdr;
 421         int ret;
 422
 423         rhdr = nfs_readhdr_alloc();
 424         if (!rhdr) {
 425                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 426                 return -ENOMEM;
 427         }
 428         hdr = &rhdr->header;
 429         nfs_pgheader_init(desc, hdr, nfs_readhdr_free);
 430         atomic_inc(&hdr->refcnt);
 431         ret = nfs_generic_pagein(desc, hdr);
 432         if (ret == 0)
 433                 ret = nfs_do_multiple_reads(&hdr->rpc_list,
 434                                             desc->pg_rpc_callops);
 435         else
 436                 set_bit(NFS_IOHDR_REDO, &hdr->flags);
 437         if (atomic_dec_and_test(&hdr->refcnt))
 438                 hdr->completion_ops->completion(hdr);
 439         return ret;
 440 }
 441
 442 static const struct nfs_pageio_ops nfs_pageio_read_ops = {
 443         .pg_test = nfs_generic_pg_test,
 444         .pg_doio = nfs_generic_pg_readpages,
 445 };
 446
 447 /*
 448  * This is the callback from RPC telling us whether a reply was
 449  * received or some error occurred (timeout or socket shutdown).
 450  */
 451 int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
 452 {
 453         struct inode *inode = data->header->inode;
 454         int status;
 455
 456         dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
 457                         task->tk_status);
 458
 459         status = NFS_PROTO(inode)->read_done(task, data);
 460         if (status != 0)
 461                 return status;
 462
 463         nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, data->res.count);
 464
 465         if (task->tk_status == -ESTALE) {
 466                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 467                 nfs_mark_for_revalidate(inode);
 468         }
 469         return 0;
 470 }
 471
 472 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
 473 {
 474         struct nfs_readargs *argp = &data->args;
 475         struct nfs_readres *resp = &data->res;
 476
 477         /* This is a short read! */
 478         nfs_inc_stats(data->header->inode, NFSIOS_SHORTREAD);
 479         /* Has the server at least made some progress? */
 480         if (resp->count == 0) {
 481                 nfs_set_pgio_error(data->header, -EIO, argp->offset);
 482                 return;
 483         }
 484         /* Yes, so retry the read at the end of the data */
 485         data->mds_offset += resp->count;
 486         argp->offset += resp->count;
 487         argp->pgbase += resp->count;
 488         argp->count -= resp->count;
 489         rpc_restart_call_prepare(task);
 490 }
 491
 492 static void nfs_readpage_result_common(struct rpc_task *task, void *calldata)
 493 {
 494         struct nfs_read_data *data = calldata;
 495         struct nfs_pgio_header *hdr = data->header;
 496
 497         /* Note the only returns of nfs_readpage_result are 0 and -EAGAIN */
 498         if (nfs_readpage_result(task, data) != 0)
 499                 return;
 500         if (task->tk_status < 0)
 501                 nfs_set_pgio_error(hdr, task->tk_status, data->args.offset);
 502         else if (data->res.eof) {
 503                 loff_t bound;
 504
 505                 bound = data->args.offset + data->res.count;
 506                 spin_lock(&hdr->lock);
 507                 if (bound < hdr->io_start + hdr->good_bytes) {
 508                         set_bit(NFS_IOHDR_EOF, &hdr->flags);
 509                         clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
 510                         hdr->good_bytes = bound - hdr->io_start;
 511                 }
 512                 spin_unlock(&hdr->lock);
 513         } else if (data->res.count != data->args.count)
 514                 nfs_readpage_retry(task, data);
 515 }
 516
 517 static void nfs_readpage_release_common(void *calldata)
 518 {
 519         nfs_readdata_release(calldata);
 520 }
 521
 522 void nfs_read_prepare(struct rpc_task *task, void *calldata)
 523 {
 524         struct nfs_read_data *data = calldata;
 525         NFS_PROTO(data->header->inode)->read_rpc_prepare(task, data);
 526 }
 527
 528 static const struct rpc_call_ops nfs_read_common_ops = {
 529         .rpc_call_prepare = nfs_read_prepare,
 530         .rpc_call_done = nfs_readpage_result_common,
 531         .rpc_release = nfs_readpage_release_common,
 532 };
 533
 534 /*
 535  * Read a page over NFS.
 536  * We read the page synchronously in the following case:
 537  *  -   The error flag is set for this page. This happens only when a
 538  *      previous async read operation failed.
 539  */
 540 int nfs_readpage(struct file *file, struct page *page)
 541 {
 542         struct nfs_open_context *ctx;
 543         struct inode *inode = page->mapping->host;
 544         int             error;
 545
 546         dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
 547                 page, PAGE_CACHE_SIZE, page->index);
 548         nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
 549         nfs_add_stats(inode, NFSIOS_READPAGES, 1);
 550
 551         /*
 552          * Try to flush any pending writes to the file..
 553          *
 554          * NOTE! Because we own the page lock, there cannot
 555          * be any new pending writes generated at this point
 556          * for this page (other pages can be written to).
 557          */
 558         error = nfs_wb_page(inode, page);
 559         if (error)
 560                 goto out_unlock;
 561         if (PageUptodate(page))
 562                 goto out_unlock;
 563
 564         error = -ESTALE;
 565         if (NFS_STALE(inode))
 566                 goto out_unlock;
 567
 568         if (file == NULL) {
 569                 error = -EBADF;
 570                 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 571                 if (ctx == NULL)
 572                         goto out_unlock;
 573         } else
 574                 ctx = get_nfs_open_context(nfs_file_open_context(file));
 575
 576         if (!IS_SYNC(inode)) {
 577                 error = nfs_readpage_from_fscache(ctx, inode, page);
 578                 if (error == 0)
 579                         goto out;
 580         }
 581
 582         error = nfs_readpage_async(ctx, inode, page);
 583
 584 out:
 585         put_nfs_open_context(ctx);
 586         return error;
 587 out_unlock:
 588         unlock_page(page);
 589         return error;
 590 }
 591
 592 struct nfs_readdesc {
 593         struct nfs_pageio_descriptor *pgio;
 594         struct nfs_open_context *ctx;
 595 };
 596
 597 static int
 598 readpage_async_filler(void *data, struct page *page)
 599 {
 600         struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
 601         struct inode *inode = page->mapping->host;
 602         struct nfs_page *new;
 603         unsigned int len;
 604         int error;
 605
 606         len = nfs_page_length(page);
 607         if (len == 0)
 608                 return nfs_return_empty_page(page);
 609
 610         new = nfs_create_request(desc->ctx, inode, page, 0, len);
 611         if (IS_ERR(new))
 612                 goto out_error;
 613
 614         if (len < PAGE_CACHE_SIZE)
 615                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 616         if (!nfs_pageio_add_request(desc->pgio, new)) {
 617                 error = desc->pgio->pg_error;
 618                 goto out_unlock;
 619         }
 620         return 0;
 621 out_error:
 622         error = PTR_ERR(new);
 623 out_unlock:
 624         unlock_page(page);
 625         return error;
 626 }
 627
 628 int nfs_readpages(struct file *filp, struct address_space *mapping,
 629                 struct list_head *pages, unsigned nr_pages)
 630 {
 631         struct nfs_pageio_descriptor pgio;
 632         struct nfs_readdesc desc = {
 633                 .pgio = &pgio,
 634         };
 635         struct inode *inode = mapping->host;
 636         unsigned long npages;
 637         int ret = -ESTALE;
 638
 639         dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
 640                         inode->i_sb->s_id,
 641                         (long long)NFS_FILEID(inode),
 642                         nr_pages);
 643         nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
 644
 645         if (NFS_STALE(inode))
 646                 goto out;
 647
 648         if (filp == NULL) {
 649                 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 650                 if (desc.ctx == NULL)
 651                         return -EBADF;
 652         } else
 653                 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
 654
 655         /* attempt to read as many of the pages as possible from the cache
 656          * - this returns -ENOBUFS immediately if the cookie is negative
 657          */
 658         ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
 659                                          pages, &nr_pages);
 660         if (ret == 0)
 661                 goto read_complete; /* all pages were read */
 662
 663         nfs_pageio_init_read(&pgio, inode, &nfs_async_read_completion_ops);
 664
 665         ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
 666
 667         nfs_pageio_complete(&pgio);
 668         npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 669         nfs_add_stats(inode, NFSIOS_READPAGES, npages);
 670 read_complete:
 671         put_nfs_open_context(desc.ctx);
 672 out:
 673         return ret;
 674 }
 675
 676 int __init nfs_init_readpagecache(void)
 677 {
 678         nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
 679                                              sizeof(struct nfs_read_header),
 680                                              0, SLAB_HWCACHE_ALIGN,
 681                                              NULL);
 682         if (nfs_rdata_cachep == NULL)
 683                 return -ENOMEM;
 684
 685         return 0;
 686 }
 687
 688 void nfs_destroy_readpagecache(void)
 689 {
 690         kmem_cache_destroy(nfs_rdata_cachep);
 691 }