SAFE public projects git trees. - safe/jmp/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_free(data);
 259 }
 260
 261 static const struct rpc_call_ops nfs_read_direct_ops = {
 262 #if defined(CONFIG_NFS_V4_1)
 263         .rpc_call_prepare = nfs_read_prepare,
 264 #endif /* CONFIG_NFS_V4_1 */
 265         .rpc_call_done = nfs_direct_read_result,
 266         .rpc_release = nfs_direct_read_release,
 267 };
 268
 269 /*
 270  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 271  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 272  * bail and stop sending more reads.  Read length accounting is
 273  * handled automatically by nfs_direct_read_result().  Otherwise, if
 274  * no requests have been sent, just return an error.
 275  */
 276 static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
 277                                                 const struct iovec *iov,
 278                                                 loff_t pos)
 279 {
 280         struct nfs_open_context *ctx = dreq->ctx;
 281         struct inode *inode = ctx->path.dentry->d_inode;
 282         unsigned long user_addr = (unsigned long)iov->iov_base;
 283         size_t count = iov->iov_len;
 284         size_t rsize = NFS_SERVER(inode)->rsize;
 285         struct rpc_task *task;
 286         struct rpc_message msg = {
 287                 .rpc_cred = ctx->cred,
 288         };
 289         struct rpc_task_setup task_setup_data = {
 290                 .rpc_client = NFS_CLIENT(inode),
 291                 .rpc_message = &msg,
 292                 .callback_ops = &nfs_read_direct_ops,
 293                 .workqueue = nfsiod_workqueue,
 294                 .flags = RPC_TASK_ASYNC,
 295         };
 296         unsigned int pgbase;
 297         int result;
 298         ssize_t started = 0;
 299
 300         do {
 301                 struct nfs_read_data *data;
 302                 size_t bytes;
 303
 304                 pgbase = user_addr & ~PAGE_MASK;
 305                 bytes = min(rsize,count);
 306
 307                 result = -ENOMEM;
 308                 data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
 309                 if (unlikely(!data))
 310                         break;
 311
 312                 down_read(&current->mm->mmap_sem);
 313                 result = get_user_pages(current, current->mm, user_addr,
 314                                         data->npages, 1, 0, data->pagevec, NULL);
 315                 up_read(&current->mm->mmap_sem);
 316                 if (result < 0) {
 317                         nfs_readdata_free(data);
 318                         break;
 319                 }
 320                 if ((unsigned)result < data->npages) {
 321                         bytes = result * PAGE_SIZE;
 322                         if (bytes <= pgbase) {
 323                                 nfs_direct_release_pages(data->pagevec, result);
 324                                 nfs_readdata_free(data);
 325                                 break;
 326                         }
 327                         bytes -= pgbase;
 328                         data->npages = result;
 329                 }
 330
 331                 get_dreq(dreq);
 332
 333                 data->req = (struct nfs_page *) dreq;
 334                 data->inode = inode;
 335                 data->cred = msg.rpc_cred;
 336                 data->args.fh = NFS_FH(inode);
 337                 data->args.context = ctx;
 338                 data->args.offset = pos;
 339                 data->args.pgbase = pgbase;
 340                 data->args.pages = data->pagevec;
 341                 data->args.count = bytes;
 342                 data->res.fattr = &data->fattr;
 343                 data->res.eof = 0;
 344                 data->res.count = bytes;
 345                 nfs_fattr_init(&data->fattr);
 346                 msg.rpc_argp = &data->args;
 347                 msg.rpc_resp = &data->res;
 348
 349                 task_setup_data.task = &data->task;
 350                 task_setup_data.callback_data = data;
 351                 NFS_PROTO(inode)->read_setup(data, &msg);
 352
 353                 task = rpc_run_task(&task_setup_data);
 354                 if (IS_ERR(task))
 355                         break;
 356                 rpc_put_task(task);
 357
 358                 dprintk("NFS: %5u initiated direct read call "
 359                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 360                                 data->task.tk_pid,
 361                                 inode->i_sb->s_id,
 362                                 (long long)NFS_FILEID(inode),
 363                                 bytes,
 364                                 (unsigned long long)data->args.offset);
 365
 366                 started += bytes;
 367                 user_addr += bytes;
 368                 pos += bytes;
 369                 /* FIXME: Remove this unnecessary math from final patch */
 370                 pgbase += bytes;
 371                 pgbase &= ~PAGE_MASK;
 372                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 373
 374                 count -= bytes;
 375         } while (count != 0);
 376
 377         if (started)
 378                 return started;
 379         return result < 0 ? (ssize_t) result : -EFAULT;
 380 }
 381
 382 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
 383                                               const struct iovec *iov,
 384                                               unsigned long nr_segs,
 385                                               loff_t pos)
 386 {
 387         ssize_t result = -EINVAL;
 388         size_t requested_bytes = 0;
 389         unsigned long seg;
 390
 391         get_dreq(dreq);
 392
 393         for (seg = 0; seg < nr_segs; seg++) {
 394                 const struct iovec *vec = &iov[seg];
 395                 result = nfs_direct_read_schedule_segment(dreq, vec, pos);
 396                 if (result < 0)
 397                         break;
 398                 requested_bytes += result;
 399                 if ((size_t)result < vec->iov_len)
 400                         break;
 401                 pos += vec->iov_len;
 402         }
 403
 404         if (put_dreq(dreq))
 405                 nfs_direct_complete(dreq);
 406
 407         if (requested_bytes != 0)
 408                 return 0;
 409
 410         if (result < 0)
 411                 return result;
 412         return -EIO;
 413 }
 414
 415 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
 416                                unsigned long nr_segs, loff_t pos)
 417 {
 418         ssize_t result = 0;
 419         struct inode *inode = iocb->ki_filp->f_mapping->host;
 420         struct nfs_direct_req *dreq;
 421
 422         dreq = nfs_direct_req_alloc();
 423         if (!dreq)
 424                 return -ENOMEM;
 425
 426         dreq->inode = inode;
 427         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 428         if (!is_sync_kiocb(iocb))
 429                 dreq->iocb = iocb;
 430
 431         result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
 432         if (!result)
 433                 result = nfs_direct_wait(dreq);
 434         nfs_direct_req_release(dreq);
 435
 436         return result;
 437 }
 438
 439 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 440 {
 441         while (!list_empty(&dreq->rewrite_list)) {
 442                 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 443                 list_del(&data->pages);
 444                 nfs_direct_release_pages(data->pagevec, data->npages);
 445                 nfs_writedata_free(data);
 446         }
 447 }
 448
 449 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 450 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 451 {
 452         struct inode *inode = dreq->inode;
 453         struct list_head *p;
 454         struct nfs_write_data *data;
 455         struct rpc_task *task;
 456         struct rpc_message msg = {
 457                 .rpc_cred = dreq->ctx->cred,
 458         };
 459         struct rpc_task_setup task_setup_data = {
 460                 .rpc_client = NFS_CLIENT(inode),
 461                 .rpc_message = &msg,
 462                 .callback_ops = &nfs_write_direct_ops,
 463                 .workqueue = nfsiod_workqueue,
 464                 .flags = RPC_TASK_ASYNC,
 465         };
 466
 467         dreq->count = 0;
 468         get_dreq(dreq);
 469
 470         list_for_each(p, &dreq->rewrite_list) {
 471                 data = list_entry(p, struct nfs_write_data, pages);
 472
 473                 get_dreq(dreq);
 474
 475                 /* Use stable writes */
 476                 data->args.stable = NFS_FILE_SYNC;
 477
 478                 /*
 479                  * Reset data->res.
 480                  */
 481                 nfs_fattr_init(&data->fattr);
 482                 data->res.count = data->args.count;
 483                 memset(&data->verf, 0, sizeof(data->verf));
 484
 485                 /*
 486                  * Reuse data->task; data->args should not have changed
 487                  * since the original request was sent.
 488                  */
 489                 task_setup_data.task = &data->task;
 490                 task_setup_data.callback_data = data;
 491                 msg.rpc_argp = &data->args;
 492                 msg.rpc_resp = &data->res;
 493                 NFS_PROTO(inode)->write_setup(data, &msg);
 494
 495                 /*
 496                  * We're called via an RPC callback, so BKL is already held.
 497                  */
 498                 task = rpc_run_task(&task_setup_data);
 499                 if (!IS_ERR(task))
 500                         rpc_put_task(task);
 501
 502                 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 503                                 data->task.tk_pid,
 504                                 inode->i_sb->s_id,
 505                                 (long long)NFS_FILEID(inode),
 506                                 data->args.count,
 507                                 (unsigned long long)data->args.offset);
 508         }
 509
 510         if (put_dreq(dreq))
 511                 nfs_direct_write_complete(dreq, inode);
 512 }
 513
 514 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
 515 {
 516         struct nfs_write_data *data = calldata;
 517
 518         /* Call the NFS version-specific code */
 519         NFS_PROTO(data->inode)->commit_done(task, data);
 520 }
 521
 522 static void nfs_direct_commit_release(void *calldata)
 523 {
 524         struct nfs_write_data *data = calldata;
 525         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 526         int status = data->task.tk_status;
 527
 528         if (status < 0) {
 529                 dprintk("NFS: %5u commit failed with error %d.\n",
 530                                 data->task.tk_pid, status);
 531                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 532         } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
 533                 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
 534                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 535         }
 536
 537         dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
 538         nfs_direct_write_complete(dreq, data->inode);
 539         nfs_commit_free(data);
 540 }
 541
 542 static const struct rpc_call_ops nfs_commit_direct_ops = {
 543 #if defined(CONFIG_NFS_V4_1)
 544         .rpc_call_prepare = nfs_write_prepare,
 545 #endif /* CONFIG_NFS_V4_1 */
 546         .rpc_call_done = nfs_direct_commit_result,
 547         .rpc_release = nfs_direct_commit_release,
 548 };
 549
 550 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 551 {
 552         struct nfs_write_data *data = dreq->commit_data;
 553         struct rpc_task *task;
 554         struct rpc_message msg = {
 555                 .rpc_argp = &data->args,
 556                 .rpc_resp = &data->res,
 557                 .rpc_cred = dreq->ctx->cred,
 558         };
 559         struct rpc_task_setup task_setup_data = {
 560                 .task = &data->task,
 561                 .rpc_client = NFS_CLIENT(dreq->inode),
 562                 .rpc_message = &msg,
 563                 .callback_ops = &nfs_commit_direct_ops,
 564                 .callback_data = data,
 565                 .workqueue = nfsiod_workqueue,
 566                 .flags = RPC_TASK_ASYNC,
 567         };
 568
 569         data->inode = dreq->inode;
 570         data->cred = msg.rpc_cred;
 571
 572         data->args.fh = NFS_FH(data->inode);
 573         data->args.offset = 0;
 574         data->args.count = 0;
 575         data->args.context = dreq->ctx;
 576         data->res.count = 0;
 577         data->res.fattr = &data->fattr;
 578         data->res.verf = &data->verf;
 579         nfs_fattr_init(&data->fattr);
 580
 581         NFS_PROTO(data->inode)->commit_setup(data, &msg);
 582
 583         /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
 584         dreq->commit_data = NULL;
 585
 586         dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
 587
 588         task = rpc_run_task(&task_setup_data);
 589         if (!IS_ERR(task))
 590                 rpc_put_task(task);
 591 }
 592
 593 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 594 {
 595         int flags = dreq->flags;
 596
 597         dreq->flags = 0;
 598         switch (flags) {
 599                 case NFS_ODIRECT_DO_COMMIT:
 600                         nfs_direct_commit_schedule(dreq);
 601                         break;
 602                 case NFS_ODIRECT_RESCHED_WRITES:
 603                         nfs_direct_write_reschedule(dreq);
 604                         break;
 605                 default:
 606                         if (dreq->commit_data != NULL)
 607                                 nfs_commit_free(dreq->commit_data);
 608                         nfs_direct_free_writedata(dreq);
 609                         nfs_zap_mapping(inode, inode->i_mapping);
 610                         nfs_direct_complete(dreq);
 611         }
 612 }
 613
 614 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 615 {
 616         dreq->commit_data = nfs_commitdata_alloc();
 617         if (dreq->commit_data != NULL)
 618                 dreq->commit_data->req = (struct nfs_page *) dreq;
 619 }
 620 #else
 621 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 622 {
 623         dreq->commit_data = NULL;
 624 }
 625
 626 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 627 {
 628         nfs_direct_free_writedata(dreq);
 629         nfs_zap_mapping(inode, inode->i_mapping);
 630         nfs_direct_complete(dreq);
 631 }
 632 #endif
 633
 634 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 635 {
 636         struct nfs_write_data *data = calldata;
 637
 638         if (nfs_writeback_done(task, data) != 0)
 639                 return;
 640 }
 641
 642 /*
 643  * NB: Return the value of the first error return code.  Subsequent
 644  *     errors after the first one are ignored.
 645  */
 646 static void nfs_direct_write_release(void *calldata)
 647 {
 648         struct nfs_write_data *data = calldata;
 649         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 650         int status = data->task.tk_status;
 651
 652         spin_lock(&dreq->lock);
 653
 654         if (unlikely(status < 0)) {
 655                 /* An error has occurred, so we should not commit */
 656                 dreq->flags = 0;
 657                 dreq->error = status;
 658         }
 659         if (unlikely(dreq->error != 0))
 660                 goto out_unlock;
 661
 662         dreq->count += data->res.count;
 663
 664         if (data->res.verf->committed != NFS_FILE_SYNC) {
 665                 switch (dreq->flags) {
 666                         case 0:
 667                                 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
 668                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
 669                                 break;
 670                         case NFS_ODIRECT_DO_COMMIT:
 671                                 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
 672                                         dprintk("NFS: %5u write verify failed\n", data->task.tk_pid);
 673                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 674                                 }
 675                 }
 676         }
 677 out_unlock:
 678         spin_unlock(&dreq->lock);
 679
 680         if (put_dreq(dreq))
 681                 nfs_direct_write_complete(dreq, data->inode);
 682 }
 683
 684 static const struct rpc_call_ops nfs_write_direct_ops = {
 685 #if defined(CONFIG_NFS_V4_1)
 686         .rpc_call_prepare = nfs_write_prepare,
 687 #endif /* CONFIG_NFS_V4_1 */
 688         .rpc_call_done = nfs_direct_write_result,
 689         .rpc_release = nfs_direct_write_release,
 690 };
 691
 692 /*
 693  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 694  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 695  * bail and stop sending more writes.  Write length accounting is
 696  * handled automatically by nfs_direct_write_result().  Otherwise, if
 697  * no requests have been sent, just return an error.
 698  */
 699 static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
 700                                                  const struct iovec *iov,
 701                                                  loff_t pos, int sync)
 702 {
 703         struct nfs_open_context *ctx = dreq->ctx;
 704         struct inode *inode = ctx->path.dentry->d_inode;
 705         unsigned long user_addr = (unsigned long)iov->iov_base;
 706         size_t count = iov->iov_len;
 707         struct rpc_task *task;
 708         struct rpc_message msg = {
 709                 .rpc_cred = ctx->cred,
 710         };
 711         struct rpc_task_setup task_setup_data = {
 712                 .rpc_client = NFS_CLIENT(inode),
 713                 .rpc_message = &msg,
 714                 .callback_ops = &nfs_write_direct_ops,
 715                 .workqueue = nfsiod_workqueue,
 716                 .flags = RPC_TASK_ASYNC,
 717         };
 718         size_t wsize = NFS_SERVER(inode)->wsize;
 719         unsigned int pgbase;
 720         int result;
 721         ssize_t started = 0;
 722
 723         do {
 724                 struct nfs_write_data *data;
 725                 size_t bytes;
 726
 727                 pgbase = user_addr & ~PAGE_MASK;
 728                 bytes = min(wsize,count);
 729
 730                 result = -ENOMEM;
 731                 data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
 732                 if (unlikely(!data))
 733                         break;
 734
 735                 down_read(&current->mm->mmap_sem);
 736                 result = get_user_pages(current, current->mm, user_addr,
 737                                         data->npages, 0, 0, data->pagevec, NULL);
 738                 up_read(&current->mm->mmap_sem);
 739                 if (result < 0) {
 740                         nfs_writedata_free(data);
 741                         break;
 742                 }
 743                 if ((unsigned)result < data->npages) {
 744                         bytes = result * PAGE_SIZE;
 745                         if (bytes <= pgbase) {
 746                                 nfs_direct_release_pages(data->pagevec, result);
 747                                 nfs_writedata_free(data);
 748                                 break;
 749                         }
 750                         bytes -= pgbase;
 751                         data->npages = result;
 752                 }
 753
 754                 get_dreq(dreq);
 755
 756                 list_move_tail(&data->pages, &dreq->rewrite_list);
 757
 758                 data->req = (struct nfs_page *) dreq;
 759                 data->inode = inode;
 760                 data->cred = msg.rpc_cred;
 761                 data->args.fh = NFS_FH(inode);
 762                 data->args.context = ctx;
 763                 data->args.offset = pos;
 764                 data->args.pgbase = pgbase;
 765                 data->args.pages = data->pagevec;
 766                 data->args.count = bytes;
 767                 data->args.stable = sync;
 768                 data->res.fattr = &data->fattr;
 769                 data->res.count = bytes;
 770                 data->res.verf = &data->verf;
 771                 nfs_fattr_init(&data->fattr);
 772
 773                 task_setup_data.task = &data->task;
 774                 task_setup_data.callback_data = data;
 775                 msg.rpc_argp = &data->args;
 776                 msg.rpc_resp = &data->res;
 777                 NFS_PROTO(inode)->write_setup(data, &msg);
 778
 779                 task = rpc_run_task(&task_setup_data);
 780                 if (IS_ERR(task))
 781                         break;
 782                 rpc_put_task(task);
 783
 784                 dprintk("NFS: %5u initiated direct write call "
 785                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 786                                 data->task.tk_pid,
 787                                 inode->i_sb->s_id,
 788                                 (long long)NFS_FILEID(inode),
 789                                 bytes,
 790                                 (unsigned long long)data->args.offset);
 791
 792                 started += bytes;
 793                 user_addr += bytes;
 794                 pos += bytes;
 795
 796                 /* FIXME: Remove this useless math from the final patch */
 797                 pgbase += bytes;
 798                 pgbase &= ~PAGE_MASK;
 799                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 800
 801                 count -= bytes;
 802         } while (count != 0);
 803
 804         if (started)
 805                 return started;
 806         return result < 0 ? (ssize_t) result : -EFAULT;
 807 }
 808
 809 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 810                                                const struct iovec *iov,
 811                                                unsigned long nr_segs,
 812                                                loff_t pos, int sync)
 813 {
 814         ssize_t result = 0;
 815         size_t requested_bytes = 0;
 816         unsigned long seg;
 817
 818         get_dreq(dreq);
 819
 820         for (seg = 0; seg < nr_segs; seg++) {
 821                 const struct iovec *vec = &iov[seg];
 822                 result = nfs_direct_write_schedule_segment(dreq, vec,
 823                                                            pos, sync);
 824                 if (result < 0)
 825                         break;
 826                 requested_bytes += result;
 827                 if ((size_t)result < vec->iov_len)
 828                         break;
 829                 pos += vec->iov_len;
 830         }
 831
 832         if (put_dreq(dreq))
 833                 nfs_direct_write_complete(dreq, dreq->inode);
 834
 835         if (requested_bytes != 0)
 836                 return 0;
 837
 838         if (result < 0)
 839                 return result;
 840         return -EIO;
 841 }
 842
 843 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
 844                                 unsigned long nr_segs, loff_t pos,
 845                                 size_t count)
 846 {
 847         ssize_t result = 0;
 848         struct inode *inode = iocb->ki_filp->f_mapping->host;
 849         struct nfs_direct_req *dreq;
 850         size_t wsize = NFS_SERVER(inode)->wsize;
 851         int sync = NFS_UNSTABLE;
 852
 853         dreq = nfs_direct_req_alloc();
 854         if (!dreq)
 855                 return -ENOMEM;
 856         nfs_alloc_commit_data(dreq);
 857
 858         if (dreq->commit_data == NULL || count < wsize)
 859                 sync = NFS_FILE_SYNC;
 860
 861         dreq->inode = inode;
 862         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 863         if (!is_sync_kiocb(iocb))
 864                 dreq->iocb = iocb;
 865
 866         result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
 867         if (!result)
 868                 result = nfs_direct_wait(dreq);
 869         nfs_direct_req_release(dreq);
 870
 871         return result;
 872 }
 873
 874 /**
 875  * nfs_file_direct_read - file direct read operation for NFS files
 876  * @iocb: target I/O control block
 877  * @iov: vector of user buffers into which to read data
 878  * @nr_segs: size of iov vector
 879  * @pos: byte offset in file where reading starts
 880  *
 881  * We use this function for direct reads instead of calling
 882  * generic_file_aio_read() in order to avoid gfar's check to see if
 883  * the request starts before the end of the file.  For that check
 884  * to work, we must generate a GETATTR before each direct read, and
 885  * even then there is a window between the GETATTR and the subsequent
 886  * READ where the file size could change.  Our preference is simply
 887  * to do all reads the application wants, and the server will take
 888  * care of managing the end of file boundary.
 889  *
 890  * This function also eliminates unnecessarily updating the file's
 891  * atime locally, as the NFS server sets the file's atime, and this
 892  * client must read the updated atime from the server back into its
 893  * cache.
 894  */
 895 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
 896                                 unsigned long nr_segs, loff_t pos)
 897 {
 898         ssize_t retval = -EINVAL;
 899         struct file *file = iocb->ki_filp;
 900         struct address_space *mapping = file->f_mapping;
 901         size_t count;
 902
 903         count = iov_length(iov, nr_segs);
 904         nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
 905
 906         dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
 907                 file->f_path.dentry->d_parent->d_name.name,
 908                 file->f_path.dentry->d_name.name,
 909                 count, (long long) pos);
 910
 911         retval = 0;
 912         if (!count)
 913                 goto out;
 914
 915         retval = nfs_sync_mapping(mapping);
 916         if (retval)
 917                 goto out;
 918
 919         retval = nfs_direct_read(iocb, iov, nr_segs, pos);
 920         if (retval > 0)
 921                 iocb->ki_pos = pos + retval;
 922
 923 out:
 924         return retval;
 925 }
 926
 927 /**
 928  * nfs_file_direct_write - file direct write operation for NFS files
 929  * @iocb: target I/O control block
 930  * @iov: vector of user buffers from which to write data
 931  * @nr_segs: size of iov vector
 932  * @pos: byte offset in file where writing starts
 933  *
 934  * We use this function for direct writes instead of calling
 935  * generic_file_aio_write() in order to avoid taking the inode
 936  * semaphore and updating the i_size.  The NFS server will set
 937  * the new i_size and this client must read the updated size
 938  * back into its cache.  We let the server do generic write
 939  * parameter checking and report problems.
 940  *
 941  * We eliminate local atime updates, see direct read above.
 942  *
 943  * We avoid unnecessary page cache invalidations for normal cached
 944  * readers of this file.
 945  *
 946  * Note that O_APPEND is not supported for NFS direct writes, as there
 947  * is no atomic O_APPEND write facility in the NFS protocol.
 948  */
 949 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
 950                                 unsigned long nr_segs, loff_t pos)
 951 {
 952         ssize_t retval = -EINVAL;
 953         struct file *file = iocb->ki_filp;
 954         struct address_space *mapping = file->f_mapping;
 955         size_t count;
 956
 957         count = iov_length(iov, nr_segs);
 958         nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
 959
 960         dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
 961                 file->f_path.dentry->d_parent->d_name.name,
 962                 file->f_path.dentry->d_name.name,
 963                 count, (long long) pos);
 964
 965         retval = generic_write_checks(file, &pos, &count, 0);
 966         if (retval)
 967                 goto out;
 968
 969         retval = -EINVAL;
 970         if ((ssize_t) count < 0)
 971                 goto out;
 972         retval = 0;
 973         if (!count)
 974                 goto out;
 975
 976         retval = nfs_sync_mapping(mapping);
 977         if (retval)
 978                 goto out;
 979
 980         retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
 981
 982         if (retval > 0)
 983                 iocb->ki_pos = pos + retval;
 984
 985 out:
 986         return retval;
 987 }
 988
 989 /**
 990  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 991  *
 992  */
 993 int __init nfs_init_directcache(void)
 994 {
 995         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
 996                                                 sizeof(struct nfs_direct_req),
 997                                                 0, (SLAB_RECLAIM_ACCOUNT|
 998                                                         SLAB_MEM_SPREAD),
 999                                                 NULL);
1000         if (nfs_direct_cachep == NULL)
1001                 return -ENOMEM;
1002
1003         return 0;
1004 }
1005
1006 /**
1007  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1008  *
1009  */
1010 void nfs_destroy_directcache(void)
1011 {
1012         kmem_cache_destroy(nfs_direct_cachep);
1013 }