fs/nfs/direct.c

   1 /*
   2  * linux/fs/nfs/direct.c
   3  *
   4  * Copyright (C) 2003 by Chuck Lever <cel@netapp.com>
   5  *
   6  * High-performance uncached I/O for the Linux NFS client
   7  *
   8  * There are important applications whose performance or correctness
   9  * depends on uncached access to file data.  Database clusters
  10  * (multiple copies of the same instance running on separate hosts)
  11  * implement their own cache coherency protocol that subsumes file
  12  * system cache protocols.  Applications that process datasets
  13  * considerably larger than the client's memory do not always benefit
  14  * from a local cache.  A streaming video server, for instance, has no
  15  * need to cache the contents of a file.
  16  *
  17  * When an application requests uncached I/O, all read and write requests
  18  * are made directly to the server; data stored or fetched via these
  19  * requests is not cached in the Linux page cache.  The client does not
  20  * correct unaligned requests from applications.  All requested bytes are
  21  * held on permanent storage before a direct write system call returns to
  22  * an application.
  23  *
  24  * Solaris implements an uncached I/O facility called directio() that
  25  * is used for backups and sequential I/O to very large files.  Solaris
  26  * also supports uncaching whole NFS partitions with "-o forcedirectio,"
  27  * an undocumented mount option.
  28  *
  29  * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with
  30  * help from Andrew Morton.
  31  *
  32  * 18 Dec 2001  Initial implementation for 2.4  --cel
  33  * 08 Jul 2002  Version for 2.4.19, with bug fixes --trondmy
  34  * 08 Jun 2003  Port to 2.5 APIs  --cel
  35  * 31 Mar 2004  Handle direct I/O without VFS support  --cel
  36  * 15 Sep 2004  Parallel async reads  --cel
  37  * 04 May 2005  support O_DIRECT with aio  --cel
  38  *
  39  */
  40
  41 #include <linux/errno.h>
  42 #include <linux/sched.h>
  43 #include <linux/kernel.h>
  44 #include <linux/smp_lock.h>
  45 #include <linux/file.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/kref.h>
  48
  49 #include <linux/nfs_fs.h>
  50 #include <linux/nfs_page.h>
  51 #include <linux/sunrpc/clnt.h>
  52
  53 #include <asm/system.h>
  54 #include <asm/uaccess.h>
  55 #include <asm/atomic.h>
  56
  57 #include "iostat.h"
  58
  59 #define NFSDBG_FACILITY         NFSDBG_VFS
  60
  61 static kmem_cache_t *nfs_direct_cachep;
  62
  63 /*
  64  * This represents a set of asynchronous requests that we're waiting on
  65  */
  66 struct nfs_direct_req {
  67         struct kref             kref;           /* release manager */
  68
  69         /* I/O parameters */
  70         struct nfs_open_context *ctx;           /* file open context info */
  71         struct kiocb *          iocb;           /* controlling i/o request */
  72         struct inode *          inode;          /* target file of i/o */
  73
  74         /* completion state */
  75         atomic_t                io_count;       /* i/os we're waiting for */
  76         spinlock_t              lock;           /* protect completion state */
  77         ssize_t                 count,          /* bytes actually processed */
  78                                 error;          /* any reported error */
  79         struct completion       completion;     /* wait for i/o completion */
  80
  81         /* commit state */
  82         struct list_head        rewrite_list;   /* saved nfs_write_data structs */
  83         struct nfs_write_data * commit_data;    /* special write_data for commits */
  84         int                     flags;
  85 #define NFS_ODIRECT_DO_COMMIT           (1)     /* an unstable reply was received */
  86 #define NFS_ODIRECT_RESCHED_WRITES      (2)     /* write verification failed */
  87         struct nfs_writeverf    verf;           /* unstable write verifier */
  88 };
  89
  90 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
  91 static const struct rpc_call_ops nfs_write_direct_ops;
  92
  93 static inline void get_dreq(struct nfs_direct_req *dreq)
  94 {
  95         atomic_inc(&dreq->io_count);
  96 }
  97
  98 static inline int put_dreq(struct nfs_direct_req *dreq)
  99 {
 100         return atomic_dec_and_test(&dreq->io_count);
 101 }
 102
 103 /*
 104  * "size" is never larger than rsize or wsize.
 105  */
 106 static inline int nfs_direct_count_pages(unsigned long user_addr, size_t size)
 107 {
 108         int page_count;
 109
 110         page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 111         page_count -= user_addr >> PAGE_SHIFT;
 112         BUG_ON(page_count < 0);
 113
 114         return page_count;
 115 }
 116
 117 static inline unsigned int nfs_max_pages(unsigned int size)
 118 {
 119         return (size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 120 }
 121
 122 /**
 123  * nfs_direct_IO - NFS address space operation for direct I/O
 124  * @rw: direction (read or write)
 125  * @iocb: target I/O control block
 126  * @iov: array of vectors that define I/O buffer
 127  * @pos: offset in file to begin the operation
 128  * @nr_segs: size of iovec array
 129  *
 130  * The presence of this routine in the address space ops vector means
 131  * the NFS client supports direct I/O.  However, we shunt off direct
 132  * read and write requests before the VFS gets them, so this method
 133  * should never be called.
 134  */
 135 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
 136 {
 137         dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
 138                         iocb->ki_filp->f_dentry->d_name.name,
 139                         (long long) pos, nr_segs);
 140
 141         return -EINVAL;
 142 }
 143
 144 static void nfs_direct_dirty_pages(struct page **pages, int npages)
 145 {
 146         int i;
 147         for (i = 0; i < npages; i++) {
 148                 struct page *page = pages[i];
 149                 if (!PageCompound(page))
 150                         set_page_dirty_lock(page);
 151         }
 152 }
 153
 154 static void nfs_direct_release_pages(struct page **pages, int npages)
 155 {
 156         int i;
 157         for (i = 0; i < npages; i++)
 158                 page_cache_release(pages[i]);
 159 }
 160
 161 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
 162 {
 163         struct nfs_direct_req *dreq;
 164
 165         dreq = kmem_cache_alloc(nfs_direct_cachep, SLAB_KERNEL);
 166         if (!dreq)
 167                 return NULL;
 168
 169         kref_init(&dreq->kref);
 170         kref_get(&dreq->kref);
 171         init_completion(&dreq->completion);
 172         INIT_LIST_HEAD(&dreq->rewrite_list);
 173         dreq->iocb = NULL;
 174         dreq->ctx = NULL;
 175         spin_lock_init(&dreq->lock);
 176         atomic_set(&dreq->io_count, 0);
 177         dreq->count = 0;
 178         dreq->error = 0;
 179         dreq->flags = 0;
 180
 181         return dreq;
 182 }
 183
 184 static void nfs_direct_req_release(struct kref *kref)
 185 {
 186         struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
 187
 188         if (dreq->ctx != NULL)
 189                 put_nfs_open_context(dreq->ctx);
 190         kmem_cache_free(nfs_direct_cachep, dreq);
 191 }
 192
 193 /*
 194  * Collects and returns the final error value/byte-count.
 195  */
 196 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
 197 {
 198         ssize_t result = -EIOCBQUEUED;
 199
 200         /* Async requests don't wait here */
 201         if (dreq->iocb)
 202                 goto out;
 203
 204         result = wait_for_completion_interruptible(&dreq->completion);
 205
 206         if (!result)
 207                 result = dreq->error;
 208         if (!result)
 209                 result = dreq->count;
 210
 211 out:
 212         kref_put(&dreq->kref, nfs_direct_req_release);
 213         return (ssize_t) result;
 214 }
 215
 216 /*
 217  * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
 218  * the iocb is still valid here if this is a synchronous request.
 219  */
 220 static void nfs_direct_complete(struct nfs_direct_req *dreq)
 221 {
 222         if (dreq->iocb) {
 223                 long res = (long) dreq->error;
 224                 if (!res)
 225                         res = (long) dreq->count;
 226                 aio_complete(dreq->iocb, res, 0);
 227         }
 228         complete_all(&dreq->completion);
 229
 230         kref_put(&dreq->kref, nfs_direct_req_release);
 231 }
 232
 233 /*
 234  * We must hold a reference to all the pages in this direct read request
 235  * until the RPCs complete.  This could be long *after* we are woken up in
 236  * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
 237  */
 238 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 239 {
 240         struct nfs_read_data *data = calldata;
 241         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 242
 243         if (nfs_readpage_result(task, data) != 0)
 244                 return;
 245
 246         nfs_direct_dirty_pages(data->pagevec, data->npages);
 247         nfs_direct_release_pages(data->pagevec, data->npages);
 248
 249         spin_lock(&dreq->lock);
 250
 251         if (likely(task->tk_status >= 0))
 252                 dreq->count += data->res.count;
 253         else
 254                 dreq->error = task->tk_status;
 255
 256         spin_unlock(&dreq->lock);
 257
 258         if (put_dreq(dreq))
 259                 nfs_direct_complete(dreq);
 260 }
 261
 262 static const struct rpc_call_ops nfs_read_direct_ops = {
 263         .rpc_call_done = nfs_direct_read_result,
 264         .rpc_release = nfs_readdata_release,
 265 };
 266
 267 /*
 268  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 269  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 270  * bail and stop sending more reads.  Read length accounting is
 271  * handled automatically by nfs_direct_read_result().  Otherwise, if
 272  * no requests have been sent, just return an error.
 273  */
 274 static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos)
 275 {
 276         struct nfs_open_context *ctx = dreq->ctx;
 277         struct inode *inode = ctx->dentry->d_inode;
 278         size_t rsize = NFS_SERVER(inode)->rsize;
 279         unsigned int rpages = nfs_max_pages(rsize);
 280         unsigned int pgbase;
 281         int result;
 282         ssize_t started = 0;
 283
 284         get_dreq(dreq);
 285
 286         pgbase = user_addr & ~PAGE_MASK;
 287         do {
 288                 struct nfs_read_data *data;
 289                 size_t bytes;
 290
 291                 result = -ENOMEM;
 292                 data = nfs_readdata_alloc(rpages);
 293                 if (unlikely(!data))
 294                         break;
 295
 296                 bytes = rsize;
 297                 if (count < rsize)
 298                         bytes = count;
 299
 300                 data->npages = nfs_direct_count_pages(user_addr, bytes);
 301                 down_read(&current->mm->mmap_sem);
 302                 result = get_user_pages(current, current->mm, user_addr,
 303                                         data->npages, 1, 0, data->pagevec, NULL);
 304                 up_read(&current->mm->mmap_sem);
 305                 if (unlikely(result < data->npages)) {
 306                         if (result > 0)
 307                                 nfs_direct_release_pages(data->pagevec, result);
 308                         nfs_readdata_release(data);
 309                         break;
 310                 }
 311
 312                 get_dreq(dreq);
 313
 314                 data->req = (struct nfs_page *) dreq;
 315                 data->inode = inode;
 316                 data->cred = ctx->cred;
 317                 data->args.fh = NFS_FH(inode);
 318                 data->args.context = ctx;
 319                 data->args.offset = pos;
 320                 data->args.pgbase = pgbase;
 321                 data->args.pages = data->pagevec;
 322                 data->args.count = bytes;
 323                 data->res.fattr = &data->fattr;
 324                 data->res.eof = 0;
 325                 data->res.count = bytes;
 326
 327                 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
 328                                 &nfs_read_direct_ops, data);
 329                 NFS_PROTO(inode)->read_setup(data);
 330
 331                 data->task.tk_cookie = (unsigned long) inode;
 332
 333                 lock_kernel();
 334                 rpc_execute(&data->task);
 335                 unlock_kernel();
 336
 337                 dfprintk(VFS, "NFS: %5u initiated direct read call (req %s/%Ld, %zu bytes @ offset %Lu)\n",
 338                                 data->task.tk_pid,
 339                                 inode->i_sb->s_id,
 340                                 (long long)NFS_FILEID(inode),
 341                                 bytes,
 342                                 (unsigned long long)data->args.offset);
 343
 344                 started += bytes;
 345                 user_addr += bytes;
 346                 pos += bytes;
 347                 pgbase += bytes;
 348                 pgbase &= ~PAGE_MASK;
 349
 350                 count -= bytes;
 351         } while (count != 0);
 352
 353         if (put_dreq(dreq))
 354                 nfs_direct_complete(dreq);
 355
 356         if (started)
 357                 return 0;
 358         return result < 0 ? (ssize_t) result : -EFAULT;
 359 }
 360
 361 static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
 362 {
 363         ssize_t result = 0;
 364         sigset_t oldset;
 365         struct inode *inode = iocb->ki_filp->f_mapping->host;
 366         struct rpc_clnt *clnt = NFS_CLIENT(inode);
 367         struct nfs_direct_req *dreq;
 368
 369         dreq = nfs_direct_req_alloc();
 370         if (!dreq)
 371                 return -ENOMEM;
 372
 373         dreq->inode = inode;
 374         dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
 375         if (!is_sync_kiocb(iocb))
 376                 dreq->iocb = iocb;
 377
 378         nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
 379         rpc_clnt_sigmask(clnt, &oldset);
 380         result = nfs_direct_read_schedule(dreq, user_addr, count, pos);
 381         if (!result)
 382                 result = nfs_direct_wait(dreq);
 383         rpc_clnt_sigunmask(clnt, &oldset);
 384
 385         return result;
 386 }
 387
 388 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 389 {
 390         while (!list_empty(&dreq->rewrite_list)) {
 391                 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 392                 list_del(&data->pages);
 393                 nfs_direct_release_pages(data->pagevec, data->npages);
 394                 nfs_writedata_release(data);
 395         }
 396 }
 397
 398 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 399 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 400 {
 401         struct inode *inode = dreq->inode;
 402         struct list_head *p;
 403         struct nfs_write_data *data;
 404
 405         dreq->count = 0;
 406         get_dreq(dreq);
 407
 408         list_for_each(p, &dreq->rewrite_list) {
 409                 data = list_entry(p, struct nfs_write_data, pages);
 410
 411                 get_dreq(dreq);
 412
 413                 /*
 414                  * Reset data->res.
 415                  */
 416                 nfs_fattr_init(&data->fattr);
 417                 data->res.count = data->args.count;
 418                 memset(&data->verf, 0, sizeof(data->verf));
 419
 420                 /*
 421                  * Reuse data->task; data->args should not have changed
 422                  * since the original request was sent.
 423                  */
 424                 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
 425                                 &nfs_write_direct_ops, data);
 426                 NFS_PROTO(inode)->write_setup(data, FLUSH_STABLE);
 427
 428                 data->task.tk_priority = RPC_PRIORITY_NORMAL;
 429                 data->task.tk_cookie = (unsigned long) inode;
 430
 431                 /*
 432                  * We're called via an RPC callback, so BKL is already held.
 433                  */
 434                 rpc_execute(&data->task);
 435
 436                 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 437                                 data->task.tk_pid,
 438                                 inode->i_sb->s_id,
 439                                 (long long)NFS_FILEID(inode),
 440                                 data->args.count,
 441                                 (unsigned long long)data->args.offset);
 442         }
 443
 444         if (put_dreq(dreq))
 445                 nfs_direct_write_complete(dreq, inode);
 446 }
 447
 448 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
 449 {
 450         struct nfs_write_data *data = calldata;
 451         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 452
 453         /* Call the NFS version-specific code */
 454         if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
 455                 return;
 456         if (unlikely(task->tk_status < 0)) {
 457                 dreq->error = task->tk_status;
 458                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 459         }
 460         if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
 461                 dprintk("NFS: %5u commit verify failed\n", task->tk_pid);
 462                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 463         }
 464
 465         dprintk("NFS: %5u commit returned %d\n", task->tk_pid, task->tk_status);
 466         nfs_direct_write_complete(dreq, data->inode);
 467 }
 468
 469 static const struct rpc_call_ops nfs_commit_direct_ops = {
 470         .rpc_call_done = nfs_direct_commit_result,
 471         .rpc_release = nfs_commit_release,
 472 };
 473
 474 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 475 {
 476         struct nfs_write_data *data = dreq->commit_data;
 477
 478         data->inode = dreq->inode;
 479         data->cred = dreq->ctx->cred;
 480
 481         data->args.fh = NFS_FH(data->inode);
 482         data->args.offset = 0;
 483         data->args.count = 0;
 484         data->res.count = 0;
 485         data->res.fattr = &data->fattr;
 486         data->res.verf = &data->verf;
 487
 488         rpc_init_task(&data->task, NFS_CLIENT(dreq->inode), RPC_TASK_ASYNC,
 489                                 &nfs_commit_direct_ops, data);
 490         NFS_PROTO(data->inode)->commit_setup(data, 0);
 491
 492         data->task.tk_priority = RPC_PRIORITY_NORMAL;
 493         data->task.tk_cookie = (unsigned long)data->inode;
 494         /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
 495         dreq->commit_data = NULL;
 496
 497         dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
 498
 499         lock_kernel();
 500         rpc_execute(&data->task);
 501         unlock_kernel();
 502 }
 503
 504 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 505 {
 506         int flags = dreq->flags;
 507
 508         dreq->flags = 0;
 509         switch (flags) {
 510                 case NFS_ODIRECT_DO_COMMIT:
 511                         nfs_direct_commit_schedule(dreq);
 512                         break;
 513                 case NFS_ODIRECT_RESCHED_WRITES:
 514                         nfs_direct_write_reschedule(dreq);
 515                         break;
 516                 default:
 517                         nfs_end_data_update(inode);
 518                         if (dreq->commit_data != NULL)
 519                                 nfs_commit_free(dreq->commit_data);
 520                         nfs_direct_free_writedata(dreq);
 521                         nfs_direct_complete(dreq);
 522         }
 523 }
 524
 525 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 526 {
 527         dreq->commit_data = nfs_commit_alloc(0);
 528         if (dreq->commit_data != NULL)
 529                 dreq->commit_data->req = (struct nfs_page *) dreq;
 530 }
 531 #else
 532 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 533 {
 534         dreq->commit_data = NULL;
 535 }
 536
 537 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 538 {
 539         nfs_end_data_update(inode);
 540         nfs_direct_free_writedata(dreq);
 541         nfs_direct_complete(dreq);
 542 }
 543 #endif
 544
 545 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 546 {
 547         struct nfs_write_data *data = calldata;
 548         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 549         int status = task->tk_status;
 550
 551         if (nfs_writeback_done(task, data) != 0)
 552                 return;
 553
 554         spin_lock(&dreq->lock);
 555
 556         if (likely(status >= 0))
 557                 dreq->count += data->res.count;
 558         else
 559                 dreq->error = task->tk_status;
 560
 561         if (data->res.verf->committed != NFS_FILE_SYNC) {
 562                 switch (dreq->flags) {
 563                         case 0:
 564                                 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
 565                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
 566                                 break;
 567                         case NFS_ODIRECT_DO_COMMIT:
 568                                 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
 569                                         dprintk("NFS: %5u write verify failed\n", task->tk_pid);
 570                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 571                                 }
 572                 }
 573         }
 574
 575         spin_unlock(&dreq->lock);
 576 }
 577
 578 /*
 579  * NB: Return the value of the first error return code.  Subsequent
 580  *     errors after the first one are ignored.
 581  */
 582 static void nfs_direct_write_release(void *calldata)
 583 {
 584         struct nfs_write_data *data = calldata;
 585         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 586
 587         if (put_dreq(dreq))
 588                 nfs_direct_write_complete(dreq, data->inode);
 589 }
 590
 591 static const struct rpc_call_ops nfs_write_direct_ops = {
 592         .rpc_call_done = nfs_direct_write_result,
 593         .rpc_release = nfs_direct_write_release,
 594 };
 595
 596 /*
 597  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 598  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 599  * bail and stop sending more writes.  Write length accounting is
 600  * handled automatically by nfs_direct_write_result().  Otherwise, if
 601  * no requests have been sent, just return an error.
 602  */
 603 static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos, int sync)
 604 {
 605         struct nfs_open_context *ctx = dreq->ctx;
 606         struct inode *inode = ctx->dentry->d_inode;
 607         size_t wsize = NFS_SERVER(inode)->wsize;
 608         unsigned int wpages = nfs_max_pages(wsize);
 609         unsigned int pgbase;
 610         int result;
 611         ssize_t started = 0;
 612
 613         get_dreq(dreq);
 614
 615         pgbase = user_addr & ~PAGE_MASK;
 616         do {
 617                 struct nfs_write_data *data;
 618                 size_t bytes;
 619
 620                 result = -ENOMEM;
 621                 data = nfs_writedata_alloc(wpages);
 622                 if (unlikely(!data))
 623                         break;
 624
 625                 bytes = wsize;
 626                 if (count < wsize)
 627                         bytes = count;
 628
 629                 data->npages = nfs_direct_count_pages(user_addr, bytes);
 630                 down_read(&current->mm->mmap_sem);
 631                 result = get_user_pages(current, current->mm, user_addr,
 632                                         data->npages, 0, 0, data->pagevec, NULL);
 633                 up_read(&current->mm->mmap_sem);
 634                 if (unlikely(result < data->npages)) {
 635                         if (result > 0)
 636                                 nfs_direct_release_pages(data->pagevec, result);
 637                         nfs_writedata_release(data);
 638                         break;
 639                 }
 640
 641                 get_dreq(dreq);
 642
 643                 list_move_tail(&data->pages, &dreq->rewrite_list);
 644
 645                 data->req = (struct nfs_page *) dreq;
 646                 data->inode = inode;
 647                 data->cred = ctx->cred;
 648                 data->args.fh = NFS_FH(inode);
 649                 data->args.context = ctx;
 650                 data->args.offset = pos;
 651                 data->args.pgbase = pgbase;
 652                 data->args.pages = data->pagevec;
 653                 data->args.count = bytes;
 654                 data->res.fattr = &data->fattr;
 655                 data->res.count = bytes;
 656                 data->res.verf = &data->verf;
 657
 658                 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
 659                                 &nfs_write_direct_ops, data);
 660                 NFS_PROTO(inode)->write_setup(data, sync);
 661
 662                 data->task.tk_priority = RPC_PRIORITY_NORMAL;
 663                 data->task.tk_cookie = (unsigned long) inode;
 664
 665                 lock_kernel();
 666                 rpc_execute(&data->task);
 667                 unlock_kernel();
 668
 669                 dfprintk(VFS, "NFS: %5u initiated direct write call (req %s/%Ld, %zu bytes @ offset %Lu)\n",
 670                                 data->task.tk_pid,
 671                                 inode->i_sb->s_id,
 672                                 (long long)NFS_FILEID(inode),
 673                                 bytes,
 674                                 (unsigned long long)data->args.offset);
 675
 676                 started += bytes;
 677                 user_addr += bytes;
 678                 pos += bytes;
 679                 pgbase += bytes;
 680                 pgbase &= ~PAGE_MASK;
 681
 682                 count -= bytes;
 683         } while (count != 0);
 684
 685         if (put_dreq(dreq))
 686                 nfs_direct_write_complete(dreq, inode);
 687
 688         if (started)
 689                 return 0;
 690         return result < 0 ? (ssize_t) result : -EFAULT;
 691 }
 692
 693 static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
 694 {
 695         ssize_t result = 0;
 696         sigset_t oldset;
 697         struct inode *inode = iocb->ki_filp->f_mapping->host;
 698         struct rpc_clnt *clnt = NFS_CLIENT(inode);
 699         struct nfs_direct_req *dreq;
 700         size_t wsize = NFS_SERVER(inode)->wsize;
 701         int sync = 0;
 702
 703         dreq = nfs_direct_req_alloc();
 704         if (!dreq)
 705                 return -ENOMEM;
 706         nfs_alloc_commit_data(dreq);
 707
 708         if (dreq->commit_data == NULL || count < wsize)
 709                 sync = FLUSH_STABLE;
 710
 711         dreq->inode = inode;
 712         dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
 713         if (!is_sync_kiocb(iocb))
 714                 dreq->iocb = iocb;
 715
 716         nfs_add_stats(inode, NFSIOS_DIRECTWRITTENBYTES, count);
 717
 718         nfs_begin_data_update(inode);
 719
 720         rpc_clnt_sigmask(clnt, &oldset);
 721         result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync);
 722         if (!result)
 723                 result = nfs_direct_wait(dreq);
 724         rpc_clnt_sigunmask(clnt, &oldset);
 725
 726         return result;
 727 }
 728
 729 /**
 730  * nfs_file_direct_read - file direct read operation for NFS files
 731  * @iocb: target I/O control block
 732  * @buf: user's buffer into which to read data
 733  * @count: number of bytes to read
 734  * @pos: byte offset in file where reading starts
 735  *
 736  * We use this function for direct reads instead of calling
 737  * generic_file_aio_read() in order to avoid gfar's check to see if
 738  * the request starts before the end of the file.  For that check
 739  * to work, we must generate a GETATTR before each direct read, and
 740  * even then there is a window between the GETATTR and the subsequent
 741  * READ where the file size could change.  Our preference is simply
 742  * to do all reads the application wants, and the server will take
 743  * care of managing the end of file boundary.
 744  *
 745  * This function also eliminates unnecessarily updating the file's
 746  * atime locally, as the NFS server sets the file's atime, and this
 747  * client must read the updated atime from the server back into its
 748  * cache.
 749  */
 750 ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
 751 {
 752         ssize_t retval = -EINVAL;
 753         struct file *file = iocb->ki_filp;
 754         struct address_space *mapping = file->f_mapping;
 755
 756         dprintk("nfs: direct read(%s/%s, %lu@%Ld)\n",
 757                 file->f_dentry->d_parent->d_name.name,
 758                 file->f_dentry->d_name.name,
 759                 (unsigned long) count, (long long) pos);
 760
 761         if (count < 0)
 762                 goto out;
 763         retval = -EFAULT;
 764         if (!access_ok(VERIFY_WRITE, buf, count))
 765                 goto out;
 766         retval = 0;
 767         if (!count)
 768                 goto out;
 769
 770         retval = nfs_sync_mapping(mapping);
 771         if (retval)
 772                 goto out;
 773
 774         retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos);
 775         if (retval > 0)
 776                 iocb->ki_pos = pos + retval;
 777
 778 out:
 779         return retval;
 780 }
 781
 782 /**
 783  * nfs_file_direct_write - file direct write operation for NFS files
 784  * @iocb: target I/O control block
 785  * @buf: user's buffer from which to write data
 786  * @count: number of bytes to write
 787  * @pos: byte offset in file where writing starts
 788  *
 789  * We use this function for direct writes instead of calling
 790  * generic_file_aio_write() in order to avoid taking the inode
 791  * semaphore and updating the i_size.  The NFS server will set
 792  * the new i_size and this client must read the updated size
 793  * back into its cache.  We let the server do generic write
 794  * parameter checking and report problems.
 795  *
 796  * We also avoid an unnecessary invocation of generic_osync_inode(),
 797  * as it is fairly meaningless to sync the metadata of an NFS file.
 798  *
 799  * We eliminate local atime updates, see direct read above.
 800  *
 801  * We avoid unnecessary page cache invalidations for normal cached
 802  * readers of this file.
 803  *
 804  * Note that O_APPEND is not supported for NFS direct writes, as there
 805  * is no atomic O_APPEND write facility in the NFS protocol.
 806  */
 807 ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 808 {
 809         ssize_t retval;
 810         struct file *file = iocb->ki_filp;
 811         struct address_space *mapping = file->f_mapping;
 812
 813         dfprintk(VFS, "nfs: direct write(%s/%s, %lu@%Ld)\n",
 814                 file->f_dentry->d_parent->d_name.name,
 815                 file->f_dentry->d_name.name,
 816                 (unsigned long) count, (long long) pos);
 817
 818         retval = generic_write_checks(file, &pos, &count, 0);
 819         if (retval)
 820                 goto out;
 821
 822         retval = -EINVAL;
 823         if ((ssize_t) count < 0)
 824                 goto out;
 825         retval = 0;
 826         if (!count)
 827                 goto out;
 828
 829         retval = -EFAULT;
 830         if (!access_ok(VERIFY_READ, buf, count))
 831                 goto out;
 832
 833         retval = nfs_sync_mapping(mapping);
 834         if (retval)
 835                 goto out;
 836
 837         retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos);
 838
 839         /*
 840          * XXX: nfs_end_data_update() already ensures this file's
 841          *      cached data is subsequently invalidated.  Do we really
 842          *      need to call invalidate_inode_pages2() again here?
 843          *
 844          *      For aio writes, this invalidation will almost certainly
 845          *      occur before the writes complete.  Kind of racey.
 846          */
 847         if (mapping->nrpages)
 848                 invalidate_inode_pages2(mapping);
 849
 850         if (retval > 0)
 851                 iocb->ki_pos = pos + retval;
 852
 853 out:
 854         return retval;
 855 }
 856
 857 /**
 858  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 859  *
 860  */
 861 int __init nfs_init_directcache(void)
 862 {
 863         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
 864                                                 sizeof(struct nfs_direct_req),
 865                                                 0, (SLAB_RECLAIM_ACCOUNT|
 866                                                         SLAB_MEM_SPREAD),
 867                                                 NULL, NULL);
 868         if (nfs_direct_cachep == NULL)
 869                 return -ENOMEM;
 870
 871         return 0;
 872 }
 873
 874 /**
 875  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
 876  *
 877  */
 878 void nfs_destroy_directcache(void)
 879 {
 880         if (kmem_cache_destroy(nfs_direct_cachep))
 881                 printk(KERN_INFO "nfs_direct_cache: not all structures were freed\n");
 882 }