git.oblomov.eu Git - linux-2.6/blob - 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/file.h>
  45 #include <linux/pagemap.h>
  46 #include <linux/kref.h>
  47
  48 #include <linux/nfs_fs.h>
  49 #include <linux/nfs_page.h>
  50 #include <linux/sunrpc/clnt.h>
  51
  52 #include <asm/system.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/atomic.h>
  55
  56 #include "internal.h"
  57 #include "iostat.h"
  58
  59 #define NFSDBG_FACILITY         NFSDBG_VFS
  60
  61 static struct kmem_cache *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  * nfs_direct_IO - NFS address space operation for direct I/O
 105  * @rw: direction (read or write)
 106  * @iocb: target I/O control block
 107  * @iov: array of vectors that define I/O buffer
 108  * @pos: offset in file to begin the operation
 109  * @nr_segs: size of iovec array
 110  *
 111  * The presence of this routine in the address space ops vector means
 112  * the NFS client supports direct I/O.  However, we shunt off direct
 113  * read and write requests before the VFS gets them, so this method
 114  * should never be called.
 115  */
 116 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
 117 {
 118         dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
 119                         iocb->ki_filp->f_path.dentry->d_name.name,
 120                         (long long) pos, nr_segs);
 121
 122         return -EINVAL;
 123 }
 124
 125 static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
 126 {
 127         unsigned int npages;
 128         unsigned int i;
 129
 130         if (count == 0)
 131                 return;
 132         pages += (pgbase >> PAGE_SHIFT);
 133         npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 134         for (i = 0; i < npages; i++) {
 135                 struct page *page = pages[i];
 136                 if (!PageCompound(page))
 137                         set_page_dirty(page);
 138         }
 139 }
 140
 141 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
 142 {
 143         unsigned int i;
 144         for (i = 0; i < npages; i++)
 145                 page_cache_release(pages[i]);
 146 }
 147
 148 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
 149 {
 150         struct nfs_direct_req *dreq;
 151
 152         dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
 153         if (!dreq)
 154                 return NULL;
 155
 156         kref_init(&dreq->kref);
 157         kref_get(&dreq->kref);
 158         init_completion(&dreq->completion);
 159         INIT_LIST_HEAD(&dreq->rewrite_list);
 160         dreq->iocb = NULL;
 161         dreq->ctx = NULL;
 162         spin_lock_init(&dreq->lock);
 163         atomic_set(&dreq->io_count, 0);
 164         dreq->count = 0;
 165         dreq->error = 0;
 166         dreq->flags = 0;
 167
 168         return dreq;
 169 }
 170
 171 static void nfs_direct_req_free(struct kref *kref)
 172 {
 173         struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
 174
 175         if (dreq->ctx != NULL)
 176                 put_nfs_open_context(dreq->ctx);
 177         kmem_cache_free(nfs_direct_cachep, dreq);
 178 }
 179
 180 static void nfs_direct_req_release(struct nfs_direct_req *dreq)
 181 {
 182         kref_put(&dreq->kref, nfs_direct_req_free);
 183 }
 184
 185 /*
 186  * Collects and returns the final error value/byte-count.
 187  */
 188 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
 189 {
 190         ssize_t result = -EIOCBQUEUED;
 191
 192         /* Async requests don't wait here */
 193         if (dreq->iocb)
 194                 goto out;
 195
 196         result = wait_for_completion_killable(&dreq->completion);
 197
 198         if (!result)
 199                 result = dreq->error;
 200         if (!result)
 201                 result = dreq->count;
 202
 203 out:
 204         return (ssize_t) result;
 205 }
 206
 207 /*
 208  * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
 209  * the iocb is still valid here if this is a synchronous request.
 210  */
 211 static void nfs_direct_complete(struct nfs_direct_req *dreq)
 212 {
 213         if (dreq->iocb) {
 214                 long res = (long) dreq->error;
 215                 if (!res)
 216                         res = (long) dreq->count;
 217                 aio_complete(dreq->iocb, res, 0);
 218         }
 219         complete_all(&dreq->completion);
 220
 221         nfs_direct_req_release(dreq);
 222 }
 223
 224 /*
 225  * We must hold a reference to all the pages in this direct read request
 226  * until the RPCs complete.  This could be long *after* we are woken up in
 227  * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
 228  */
 229 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 230 {
 231         struct nfs_read_data *data = calldata;
 232
 233         nfs_readpage_result(task, data);
 234 }
 235
 236 static void nfs_direct_read_release(void *calldata)
 237 {
 238
 239         struct nfs_read_data *data = calldata;
 240         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 241         int status = data->task.tk_status;
 242
 243         spin_lock(&dreq->lock);
 244         if (unlikely(status < 0)) {
 245                 dreq->error = status;
 246                 spin_unlock(&dreq->lock);
 247         } else {
 248                 dreq->count += data->res.count;
 249                 spin_unlock(&dreq->lock);
 250                 nfs_direct_dirty_pages(data->pagevec,
 251                                 data->args.pgbase,
 252                                 data->res.count);
 253         }
 254         nfs_direct_release_pages(data->pagevec, data->npages);
 255
 256         if (put_dreq(dreq))
 257                 nfs_direct_complete(dreq);
 258         nfs_readdata_release(calldata);
 259 }
 260
 261 static const struct rpc_call_ops nfs_read_direct_ops = {
 262         .rpc_call_done = nfs_direct_read_result,
 263         .rpc_release = nfs_direct_read_release,
 264 };
 265
 266 /*
 267  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 268  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 269  * bail and stop sending more reads.  Read length accounting is
 270  * handled automatically by nfs_direct_read_result().  Otherwise, if
 271  * no requests have been sent, just return an error.
 272  */
 273 static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
 274                                                 const struct iovec *iov,
 275                                                 loff_t pos)
 276 {
 277         struct nfs_open_context *ctx = dreq->ctx;
 278         struct inode *inode = ctx->path.dentry->d_inode;
 279         unsigned long user_addr = (unsigned long)iov->iov_base;
 280         size_t count = iov->iov_len;
 281         size_t rsize = NFS_SERVER(inode)->rsize;
 282         struct rpc_task *task;
 283         struct rpc_message msg = {
 284                 .rpc_cred = ctx->cred,
 285         };
 286         struct rpc_task_setup task_setup_data = {
 287                 .rpc_client = NFS_CLIENT(inode),
 288                 .rpc_message = &msg,
 289                 .callback_ops = &nfs_read_direct_ops,
 290                 .workqueue = nfsiod_workqueue,
 291                 .flags = RPC_TASK_ASYNC,
 292         };
 293         unsigned int pgbase;
 294         int result;
 295         ssize_t started = 0;
 296
 297         do {
 298                 struct nfs_read_data *data;
 299                 size_t bytes;
 300
 301                 pgbase = user_addr & ~PAGE_MASK;
 302                 bytes = min(rsize,count);
 303
 304                 result = -ENOMEM;
 305                 data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
 306                 if (unlikely(!data))
 307                         break;
 308
 309                 down_read(&current->mm->mmap_sem);
 310                 result = get_user_pages(current, current->mm, user_addr,
 311                                         data->npages, 1, 0, data->pagevec, NULL);
 312                 up_read(&current->mm->mmap_sem);
 313                 if (result < 0) {
 314                         nfs_readdata_release(data);
 315                         break;
 316                 }
 317                 if ((unsigned)result < data->npages) {
 318                         bytes = result * PAGE_SIZE;
 319                         if (bytes <= pgbase) {
 320                                 nfs_direct_release_pages(data->pagevec, result);
 321                                 nfs_readdata_release(data);
 322                                 break;
 323                         }
 324                         bytes -= pgbase;
 325                         data->npages = result;
 326                 }
 327
 328                 get_dreq(dreq);
 329
 330                 data->req = (struct nfs_page *) dreq;
 331                 data->inode = inode;
 332                 data->cred = msg.rpc_cred;
 333                 data->args.fh = NFS_FH(inode);
 334                 data->args.context = get_nfs_open_context(ctx);
 335                 data->args.offset = pos;
 336                 data->args.pgbase = pgbase;
 337                 data->args.pages = data->pagevec;
 338                 data->args.count = bytes;
 339                 data->res.fattr = &data->fattr;
 340                 data->res.eof = 0;
 341                 data->res.count = bytes;
 342                 msg.rpc_argp = &data->args;
 343                 msg.rpc_resp = &data->res;
 344
 345                 task_setup_data.task = &data->task;
 346                 task_setup_data.callback_data = data;
 347                 NFS_PROTO(inode)->read_setup(data, &msg);
 348
 349                 task = rpc_run_task(&task_setup_data);
 350                 if (IS_ERR(task))
 351                         break;
 352                 rpc_put_task(task);
 353
 354                 dprintk("NFS: %5u initiated direct read call "
 355                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 356                                 data->task.tk_pid,
 357                                 inode->i_sb->s_id,
 358                                 (long long)NFS_FILEID(inode),
 359                                 bytes,
 360                                 (unsigned long long)data->args.offset);
 361
 362                 started += bytes;
 363                 user_addr += bytes;
 364                 pos += bytes;
 365                 /* FIXME: Remove this unnecessary math from final patch */
 366                 pgbase += bytes;
 367                 pgbase &= ~PAGE_MASK;
 368                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 369
 370                 count -= bytes;
 371         } while (count != 0);
 372
 373         if (started)
 374                 return started;
 375         return result < 0 ? (ssize_t) result : -EFAULT;
 376 }
 377
 378 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
 379                                               const struct iovec *iov,
 380                                               unsigned long nr_segs,
 381                                               loff_t pos)
 382 {
 383         ssize_t result = -EINVAL;
 384         size_t requested_bytes = 0;
 385         unsigned long seg;
 386
 387         get_dreq(dreq);
 388
 389         for (seg = 0; seg < nr_segs; seg++) {
 390                 const struct iovec *vec = &iov[seg];
 391                 result = nfs_direct_read_schedule_segment(dreq, vec, pos);
 392                 if (result < 0)
 393                         break;
 394                 requested_bytes += result;
 395                 if ((size_t)result < vec->iov_len)
 396                         break;
 397                 pos += vec->iov_len;
 398         }
 399
 400         if (put_dreq(dreq))
 401                 nfs_direct_complete(dreq);
 402
 403         if (requested_bytes != 0)
 404                 return 0;
 405
 406         if (result < 0)
 407                 return result;
 408         return -EIO;
 409 }
 410
 411 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
 412                                unsigned long nr_segs, loff_t pos)
 413 {
 414         ssize_t result = 0;
 415         struct inode *inode = iocb->ki_filp->f_mapping->host;
 416         struct nfs_direct_req *dreq;
 417
 418         dreq = nfs_direct_req_alloc();
 419         if (!dreq)
 420                 return -ENOMEM;
 421
 422         dreq->inode = inode;
 423         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 424         if (!is_sync_kiocb(iocb))
 425                 dreq->iocb = iocb;
 426
 427         result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
 428         if (!result)
 429                 result = nfs_direct_wait(dreq);
 430         nfs_direct_req_release(dreq);
 431
 432         return result;
 433 }
 434
 435 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 436 {
 437         while (!list_empty(&dreq->rewrite_list)) {
 438                 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 439                 list_del(&data->pages);
 440                 nfs_direct_release_pages(data->pagevec, data->npages);
 441                 nfs_writedata_release(data);
 442         }
 443 }
 444
 445 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 446 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 447 {
 448         struct inode *inode = dreq->inode;
 449         struct list_head *p;
 450         struct nfs_write_data *data;
 451         struct rpc_task *task;
 452         struct rpc_message msg = {
 453                 .rpc_cred = dreq->ctx->cred,
 454         };
 455         struct rpc_task_setup task_setup_data = {
 456                 .rpc_client = NFS_CLIENT(inode),
 457                 .callback_ops = &nfs_write_direct_ops,
 458                 .workqueue = nfsiod_workqueue,
 459                 .flags = RPC_TASK_ASYNC,
 460         };
 461
 462         dreq->count = 0;
 463         get_dreq(dreq);
 464
 465         list_for_each(p, &dreq->rewrite_list) {
 466                 data = list_entry(p, struct nfs_write_data, pages);
 467
 468                 get_dreq(dreq);
 469
 470                 /* Use stable writes */
 471                 data->args.stable = NFS_FILE_SYNC;
 472
 473                 /*
 474                  * Reset data->res.
 475                  */
 476                 nfs_fattr_init(&data->fattr);
 477                 data->res.count = data->args.count;
 478                 memset(&data->verf, 0, sizeof(data->verf));
 479
 480                 /*
 481                  * Reuse data->task; data->args should not have changed
 482                  * since the original request was sent.
 483                  */
 484                 task_setup_data.task = &data->task;
 485                 task_setup_data.callback_data = data;
 486                 msg.rpc_argp = &data->args;
 487                 msg.rpc_resp = &data->res;
 488                 NFS_PROTO(inode)->write_setup(data, &msg);
 489
 490                 /*
 491                  * We're called via an RPC callback, so BKL is already held.
 492                  */
 493                 task = rpc_run_task(&task_setup_data);
 494                 if (!IS_ERR(task))
 495                         rpc_put_task(task);
 496
 497                 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 498                                 data->task.tk_pid,
 499                                 inode->i_sb->s_id,
 500                                 (long long)NFS_FILEID(inode),
 501                                 data->args.count,
 502                                 (unsigned long long)data->args.offset);
 503         }
 504
 505         if (put_dreq(dreq))
 506                 nfs_direct_write_complete(dreq, inode);
 507 }
 508
 509 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
 510 {
 511         struct nfs_write_data *data = calldata;
 512
 513         /* Call the NFS version-specific code */
 514         NFS_PROTO(data->inode)->commit_done(task, data);
 515 }
 516
 517 static void nfs_direct_commit_release(void *calldata)
 518 {
 519         struct nfs_write_data *data = calldata;
 520         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 521         int status = data->task.tk_status;
 522
 523         if (status < 0) {
 524                 dprintk("NFS: %5u commit failed with error %d.\n",
 525                                 data->task.tk_pid, status);
 526                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 527         } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
 528                 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
 529                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 530         }
 531
 532         dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
 533         nfs_direct_write_complete(dreq, data->inode);
 534         nfs_commitdata_release(calldata);
 535 }
 536
 537 static const struct rpc_call_ops nfs_commit_direct_ops = {
 538         .rpc_call_done = nfs_direct_commit_result,
 539         .rpc_release = nfs_direct_commit_release,
 540 };
 541
 542 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 543 {
 544         struct nfs_write_data *data = dreq->commit_data;
 545         struct rpc_task *task;
 546         struct rpc_message msg = {
 547                 .rpc_argp = &data->args,
 548                 .rpc_resp = &data->res,
 549                 .rpc_cred = dreq->ctx->cred,
 550         };
 551         struct rpc_task_setup task_setup_data = {
 552                 .task = &data->task,
 553                 .rpc_client = NFS_CLIENT(dreq->inode),
 554                 .rpc_message = &msg,
 555                 .callback_ops = &nfs_commit_direct_ops,
 556                 .callback_data = data,
 557                 .workqueue = nfsiod_workqueue,
 558                 .flags = RPC_TASK_ASYNC,
 559         };
 560
 561         data->inode = dreq->inode;
 562         data->cred = msg.rpc_cred;
 563
 564         data->args.fh = NFS_FH(data->inode);
 565         data->args.offset = 0;
 566         data->args.count = 0;
 567         data->args.context = get_nfs_open_context(dreq->ctx);
 568         data->res.count = 0;
 569         data->res.fattr = &data->fattr;
 570         data->res.verf = &data->verf;
 571
 572         NFS_PROTO(data->inode)->commit_setup(data, &msg);
 573
 574         /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
 575         dreq->commit_data = NULL;
 576
 577         dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
 578
 579         task = rpc_run_task(&task_setup_data);
 580         if (!IS_ERR(task))
 581                 rpc_put_task(task);
 582 }
 583
 584 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 585 {
 586         int flags = dreq->flags;
 587
 588         dreq->flags = 0;
 589         switch (flags) {
 590                 case NFS_ODIRECT_DO_COMMIT:
 591                         nfs_direct_commit_schedule(dreq);
 592                         break;
 593                 case NFS_ODIRECT_RESCHED_WRITES:
 594                         nfs_direct_write_reschedule(dreq);
 595                         break;
 596                 default:
 597                         if (dreq->commit_data != NULL)
 598                                 nfs_commit_free(dreq->commit_data);
 599                         nfs_direct_free_writedata(dreq);
 600                         nfs_zap_mapping(inode, inode->i_mapping);
 601                         nfs_direct_complete(dreq);
 602         }
 603 }
 604
 605 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 606 {
 607         dreq->commit_data = nfs_commitdata_alloc();
 608         if (dreq->commit_data != NULL)
 609                 dreq->commit_data->req = (struct nfs_page *) dreq;
 610 }
 611 #else
 612 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 613 {
 614         dreq->commit_data = NULL;
 615 }
 616
 617 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 618 {
 619         nfs_direct_free_writedata(dreq);
 620         nfs_zap_mapping(inode, inode->i_mapping);
 621         nfs_direct_complete(dreq);
 622 }
 623 #endif
 624
 625 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 626 {
 627         struct nfs_write_data *data = calldata;
 628
 629         if (nfs_writeback_done(task, data) != 0)
 630                 return;
 631 }
 632
 633 /*
 634  * NB: Return the value of the first error return code.  Subsequent
 635  *     errors after the first one are ignored.
 636  */
 637 static void nfs_direct_write_release(void *calldata)
 638 {
 639         struct nfs_write_data *data = calldata;
 640         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 641         int status = data->task.tk_status;
 642
 643         spin_lock(&dreq->lock);
 644
 645         if (unlikely(status < 0)) {
 646                 /* An error has occurred, so we should not commit */
 647                 dreq->flags = 0;
 648                 dreq->error = status;
 649         }
 650         if (unlikely(dreq->error != 0))
 651                 goto out_unlock;
 652
 653         dreq->count += data->res.count;
 654
 655         if (data->res.verf->committed != NFS_FILE_SYNC) {
 656                 switch (dreq->flags) {
 657                         case 0:
 658                                 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
 659                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
 660                                 break;
 661                         case NFS_ODIRECT_DO_COMMIT:
 662                                 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
 663                                         dprintk("NFS: %5u write verify failed\n", data->task.tk_pid);
 664                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 665                                 }
 666                 }
 667         }
 668 out_unlock:
 669         spin_unlock(&dreq->lock);
 670
 671         if (put_dreq(dreq))
 672                 nfs_direct_write_complete(dreq, data->inode);
 673 }
 674
 675 static const struct rpc_call_ops nfs_write_direct_ops = {
 676         .rpc_call_done = nfs_direct_write_result,
 677         .rpc_release = nfs_direct_write_release,
 678 };
 679
 680 /*
 681  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 682  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 683  * bail and stop sending more writes.  Write length accounting is
 684  * handled automatically by nfs_direct_write_result().  Otherwise, if
 685  * no requests have been sent, just return an error.
 686  */
 687 static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
 688                                                  const struct iovec *iov,
 689                                                  loff_t pos, int sync)
 690 {
 691         struct nfs_open_context *ctx = dreq->ctx;
 692         struct inode *inode = ctx->path.dentry->d_inode;
 693         unsigned long user_addr = (unsigned long)iov->iov_base;
 694         size_t count = iov->iov_len;
 695         struct rpc_task *task;
 696         struct rpc_message msg = {
 697                 .rpc_cred = ctx->cred,
 698         };
 699         struct rpc_task_setup task_setup_data = {
 700                 .rpc_client = NFS_CLIENT(inode),
 701                 .rpc_message = &msg,
 702                 .callback_ops = &nfs_write_direct_ops,
 703                 .workqueue = nfsiod_workqueue,
 704                 .flags = RPC_TASK_ASYNC,
 705         };
 706         size_t wsize = NFS_SERVER(inode)->wsize;
 707         unsigned int pgbase;
 708         int result;
 709         ssize_t started = 0;
 710
 711         do {
 712                 struct nfs_write_data *data;
 713                 size_t bytes;
 714
 715                 pgbase = user_addr & ~PAGE_MASK;
 716                 bytes = min(wsize,count);
 717
 718                 result = -ENOMEM;
 719                 data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
 720                 if (unlikely(!data))
 721                         break;
 722
 723                 down_read(&current->mm->mmap_sem);
 724                 result = get_user_pages(current, current->mm, user_addr,
 725                                         data->npages, 0, 0, data->pagevec, NULL);
 726                 up_read(&current->mm->mmap_sem);
 727                 if (result < 0) {
 728                         nfs_writedata_release(data);
 729                         break;
 730                 }
 731                 if ((unsigned)result < data->npages) {
 732                         bytes = result * PAGE_SIZE;
 733                         if (bytes <= pgbase) {
 734                                 nfs_direct_release_pages(data->pagevec, result);
 735                                 nfs_writedata_release(data);
 736                                 break;
 737                         }
 738                         bytes -= pgbase;
 739                         data->npages = result;
 740                 }
 741
 742                 get_dreq(dreq);
 743
 744                 list_move_tail(&data->pages, &dreq->rewrite_list);
 745
 746                 data->req = (struct nfs_page *) dreq;
 747                 data->inode = inode;
 748                 data->cred = msg.rpc_cred;
 749                 data->args.fh = NFS_FH(inode);
 750                 data->args.context = get_nfs_open_context(ctx);
 751                 data->args.offset = pos;
 752                 data->args.pgbase = pgbase;
 753                 data->args.pages = data->pagevec;
 754                 data->args.count = bytes;
 755                 data->args.stable = sync;
 756                 data->res.fattr = &data->fattr;
 757                 data->res.count = bytes;
 758                 data->res.verf = &data->verf;
 759
 760                 task_setup_data.task = &data->task;
 761                 task_setup_data.callback_data = data;
 762                 msg.rpc_argp = &data->args;
 763                 msg.rpc_resp = &data->res;
 764                 NFS_PROTO(inode)->write_setup(data, &msg);
 765
 766                 task = rpc_run_task(&task_setup_data);
 767                 if (IS_ERR(task))
 768                         break;
 769                 rpc_put_task(task);
 770
 771                 dprintk("NFS: %5u initiated direct write call "
 772                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 773                                 data->task.tk_pid,
 774                                 inode->i_sb->s_id,
 775                                 (long long)NFS_FILEID(inode),
 776                                 bytes,
 777                                 (unsigned long long)data->args.offset);
 778
 779                 started += bytes;
 780                 user_addr += bytes;
 781                 pos += bytes;
 782
 783                 /* FIXME: Remove this useless math from the final patch */
 784                 pgbase += bytes;
 785                 pgbase &= ~PAGE_MASK;
 786                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 787
 788                 count -= bytes;
 789         } while (count != 0);
 790
 791         if (started)
 792                 return started;
 793         return result < 0 ? (ssize_t) result : -EFAULT;
 794 }
 795
 796 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 797                                                const struct iovec *iov,
 798                                                unsigned long nr_segs,
 799                                                loff_t pos, int sync)
 800 {
 801         ssize_t result = 0;
 802         size_t requested_bytes = 0;
 803         unsigned long seg;
 804
 805         get_dreq(dreq);
 806
 807         for (seg = 0; seg < nr_segs; seg++) {
 808                 const struct iovec *vec = &iov[seg];
 809                 result = nfs_direct_write_schedule_segment(dreq, vec,
 810                                                            pos, sync);
 811                 if (result < 0)
 812                         break;
 813                 requested_bytes += result;
 814                 if ((size_t)result < vec->iov_len)
 815                         break;
 816                 pos += vec->iov_len;
 817         }
 818
 819         if (put_dreq(dreq))
 820                 nfs_direct_write_complete(dreq, dreq->inode);
 821
 822         if (requested_bytes != 0)
 823                 return 0;
 824
 825         if (result < 0)
 826                 return result;
 827         return -EIO;
 828 }
 829
 830 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
 831                                 unsigned long nr_segs, loff_t pos,
 832                                 size_t count)
 833 {
 834         ssize_t result = 0;
 835         struct inode *inode = iocb->ki_filp->f_mapping->host;
 836         struct nfs_direct_req *dreq;
 837         size_t wsize = NFS_SERVER(inode)->wsize;
 838         int sync = NFS_UNSTABLE;
 839
 840         dreq = nfs_direct_req_alloc();
 841         if (!dreq)
 842                 return -ENOMEM;
 843         nfs_alloc_commit_data(dreq);
 844
 845         if (dreq->commit_data == NULL || count < wsize)
 846                 sync = NFS_FILE_SYNC;
 847
 848         dreq->inode = inode;
 849         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 850         if (!is_sync_kiocb(iocb))
 851                 dreq->iocb = iocb;
 852
 853         result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
 854         if (!result)
 855                 result = nfs_direct_wait(dreq);
 856         nfs_direct_req_release(dreq);
 857
 858         return result;
 859 }
 860
 861 /**
 862  * nfs_file_direct_read - file direct read operation for NFS files
 863  * @iocb: target I/O control block
 864  * @iov: vector of user buffers into which to read data
 865  * @nr_segs: size of iov vector
 866  * @pos: byte offset in file where reading starts
 867  *
 868  * We use this function for direct reads instead of calling
 869  * generic_file_aio_read() in order to avoid gfar's check to see if
 870  * the request starts before the end of the file.  For that check
 871  * to work, we must generate a GETATTR before each direct read, and
 872  * even then there is a window between the GETATTR and the subsequent
 873  * READ where the file size could change.  Our preference is simply
 874  * to do all reads the application wants, and the server will take
 875  * care of managing the end of file boundary.
 876  *
 877  * This function also eliminates unnecessarily updating the file's
 878  * atime locally, as the NFS server sets the file's atime, and this
 879  * client must read the updated atime from the server back into its
 880  * cache.
 881  */
 882 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
 883                                 unsigned long nr_segs, loff_t pos)
 884 {
 885         ssize_t retval = -EINVAL;
 886         struct file *file = iocb->ki_filp;
 887         struct address_space *mapping = file->f_mapping;
 888         size_t count;
 889
 890         count = iov_length(iov, nr_segs);
 891         nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
 892
 893         dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
 894                 file->f_path.dentry->d_parent->d_name.name,
 895                 file->f_path.dentry->d_name.name,
 896                 count, (long long) pos);
 897
 898         retval = 0;
 899         if (!count)
 900                 goto out;
 901
 902         retval = nfs_sync_mapping(mapping);
 903         if (retval)
 904                 goto out;
 905
 906         retval = nfs_direct_read(iocb, iov, nr_segs, pos);
 907         if (retval > 0)
 908                 iocb->ki_pos = pos + retval;
 909
 910 out:
 911         return retval;
 912 }
 913
 914 /**
 915  * nfs_file_direct_write - file direct write operation for NFS files
 916  * @iocb: target I/O control block
 917  * @iov: vector of user buffers from which to write data
 918  * @nr_segs: size of iov vector
 919  * @pos: byte offset in file where writing starts
 920  *
 921  * We use this function for direct writes instead of calling
 922  * generic_file_aio_write() in order to avoid taking the inode
 923  * semaphore and updating the i_size.  The NFS server will set
 924  * the new i_size and this client must read the updated size
 925  * back into its cache.  We let the server do generic write
 926  * parameter checking and report problems.
 927  *
 928  * We also avoid an unnecessary invocation of generic_osync_inode(),
 929  * as it is fairly meaningless to sync the metadata of an NFS file.
 930  *
 931  * We eliminate local atime updates, see direct read above.
 932  *
 933  * We avoid unnecessary page cache invalidations for normal cached
 934  * readers of this file.
 935  *
 936  * Note that O_APPEND is not supported for NFS direct writes, as there
 937  * is no atomic O_APPEND write facility in the NFS protocol.
 938  */
 939 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
 940                                 unsigned long nr_segs, loff_t pos)
 941 {
 942         ssize_t retval = -EINVAL;
 943         struct file *file = iocb->ki_filp;
 944         struct address_space *mapping = file->f_mapping;
 945         size_t count;
 946
 947         count = iov_length(iov, nr_segs);
 948         nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
 949
 950         dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
 951                 file->f_path.dentry->d_parent->d_name.name,
 952                 file->f_path.dentry->d_name.name,
 953                 count, (long long) pos);
 954
 955         retval = generic_write_checks(file, &pos, &count, 0);
 956         if (retval)
 957                 goto out;
 958
 959         retval = -EINVAL;
 960         if ((ssize_t) count < 0)
 961                 goto out;
 962         retval = 0;
 963         if (!count)
 964                 goto out;
 965
 966         retval = nfs_sync_mapping(mapping);
 967         if (retval)
 968                 goto out;
 969
 970         retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
 971
 972         if (retval > 0)
 973                 iocb->ki_pos = pos + retval;
 974
 975 out:
 976         return retval;
 977 }
 978
 979 /**
 980  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 981  *
 982  */
 983 int __init nfs_init_directcache(void)
 984 {
 985         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
 986                                                 sizeof(struct nfs_direct_req),
 987                                                 0, (SLAB_RECLAIM_ACCOUNT|
 988                                                         SLAB_MEM_SPREAD),
 989                                                 NULL);
 990         if (nfs_direct_cachep == NULL)
 991                 return -ENOMEM;
 992
 993         return 0;
 994 }
 995
 996 /**
 997  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
 998  *
 999  */
1000 void nfs_destroy_directcache(void)
1001 {
1002         kmem_cache_destroy(nfs_direct_cachep);
1003 }