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 #include <linux/slab.h>
  48 #include <linux/task_io_accounting_ops.h>
  49
  50 #include <linux/nfs_fs.h>
  51 #include <linux/nfs_page.h>
  52 #include <linux/sunrpc/clnt.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <linux/atomic.h>
  56
  57 #include "internal.h"
  58 #include "iostat.h"
  59
  60 #define NFSDBG_FACILITY         NFSDBG_VFS
  61
  62 static struct kmem_cache *nfs_direct_cachep;
  63
  64 /*
  65  * This represents a set of asynchronous requests that we're waiting on
  66  */
  67 struct nfs_direct_req {
  68         struct kref             kref;           /* release manager */
  69
  70         /* I/O parameters */
  71         struct nfs_open_context *ctx;           /* file open context info */
  72         struct nfs_lock_context *l_ctx;         /* Lock context info */
  73         struct kiocb *          iocb;           /* controlling i/o request */
  74         struct inode *          inode;          /* target file of i/o */
  75
  76         /* completion state */
  77         atomic_t                io_count;       /* i/os we're waiting for */
  78         spinlock_t              lock;           /* protect completion state */
  79         ssize_t                 count,          /* bytes actually processed */
  80                                 error;          /* any reported error */
  81         struct completion       completion;     /* wait for i/o completion */
  82
  83         /* commit state */
  84         struct list_head        rewrite_list;   /* saved nfs_write_data structs */
  85         struct nfs_commit_data *commit_data;    /* special write_data for commits */
  86         int                     flags;
  87 #define NFS_ODIRECT_DO_COMMIT           (1)     /* an unstable reply was received */
  88 #define NFS_ODIRECT_RESCHED_WRITES      (2)     /* write verification failed */
  89         struct nfs_writeverf    verf;           /* unstable write verifier */
  90 };
  91
  92 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
  93 static const struct rpc_call_ops nfs_write_direct_ops;
  94
  95 static inline void get_dreq(struct nfs_direct_req *dreq)
  96 {
  97         atomic_inc(&dreq->io_count);
  98 }
  99
 100 static inline int put_dreq(struct nfs_direct_req *dreq)
 101 {
 102         return atomic_dec_and_test(&dreq->io_count);
 103 }
 104
 105 /**
 106  * nfs_direct_IO - NFS address space operation for direct I/O
 107  * @rw: direction (read or write)
 108  * @iocb: target I/O control block
 109  * @iov: array of vectors that define I/O buffer
 110  * @pos: offset in file to begin the operation
 111  * @nr_segs: size of iovec array
 112  *
 113  * The presence of this routine in the address space ops vector means
 114  * the NFS client supports direct I/O.  However, we shunt off direct
 115  * read and write requests before the VFS gets them, so this method
 116  * should never be called.
 117  */
 118 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
 119 {
 120         dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
 121                         iocb->ki_filp->f_path.dentry->d_name.name,
 122                         (long long) pos, nr_segs);
 123
 124         return -EINVAL;
 125 }
 126
 127 static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
 128 {
 129         unsigned int npages;
 130         unsigned int i;
 131
 132         if (count == 0)
 133                 return;
 134         pages += (pgbase >> PAGE_SHIFT);
 135         npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 136         for (i = 0; i < npages; i++) {
 137                 struct page *page = pages[i];
 138                 if (!PageCompound(page))
 139                         set_page_dirty(page);
 140         }
 141 }
 142
 143 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
 144 {
 145         unsigned int i;
 146         for (i = 0; i < npages; i++)
 147                 page_cache_release(pages[i]);
 148 }
 149
 150 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
 151 {
 152         struct nfs_direct_req *dreq;
 153
 154         dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
 155         if (!dreq)
 156                 return NULL;
 157
 158         kref_init(&dreq->kref);
 159         kref_get(&dreq->kref);
 160         init_completion(&dreq->completion);
 161         INIT_LIST_HEAD(&dreq->rewrite_list);
 162         dreq->iocb = NULL;
 163         dreq->ctx = NULL;
 164         dreq->l_ctx = NULL;
 165         spin_lock_init(&dreq->lock);
 166         atomic_set(&dreq->io_count, 0);
 167         dreq->count = 0;
 168         dreq->error = 0;
 169         dreq->flags = 0;
 170
 171         return dreq;
 172 }
 173
 174 static void nfs_direct_req_free(struct kref *kref)
 175 {
 176         struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
 177
 178         if (dreq->l_ctx != NULL)
 179                 nfs_put_lock_context(dreq->l_ctx);
 180         if (dreq->ctx != NULL)
 181                 put_nfs_open_context(dreq->ctx);
 182         kmem_cache_free(nfs_direct_cachep, dreq);
 183 }
 184
 185 static void nfs_direct_req_release(struct nfs_direct_req *dreq)
 186 {
 187         kref_put(&dreq->kref, nfs_direct_req_free);
 188 }
 189
 190 /*
 191  * Collects and returns the final error value/byte-count.
 192  */
 193 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
 194 {
 195         ssize_t result = -EIOCBQUEUED;
 196
 197         /* Async requests don't wait here */
 198         if (dreq->iocb)
 199                 goto out;
 200
 201         result = wait_for_completion_killable(&dreq->completion);
 202
 203         if (!result)
 204                 result = dreq->error;
 205         if (!result)
 206                 result = dreq->count;
 207
 208 out:
 209         return (ssize_t) result;
 210 }
 211
 212 /*
 213  * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
 214  * the iocb is still valid here if this is a synchronous request.
 215  */
 216 static void nfs_direct_complete(struct nfs_direct_req *dreq)
 217 {
 218         if (dreq->iocb) {
 219                 long res = (long) dreq->error;
 220                 if (!res)
 221                         res = (long) dreq->count;
 222                 aio_complete(dreq->iocb, res, 0);
 223         }
 224         complete_all(&dreq->completion);
 225
 226         nfs_direct_req_release(dreq);
 227 }
 228
 229 /*
 230  * We must hold a reference to all the pages in this direct read request
 231  * until the RPCs complete.  This could be long *after* we are woken up in
 232  * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
 233  */
 234 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 235 {
 236         struct nfs_read_data *data = calldata;
 237
 238         nfs_readpage_result(task, data);
 239 }
 240
 241 static void nfs_direct_read_release(void *calldata)
 242 {
 243
 244         struct nfs_read_data *data = calldata;
 245         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 246         int status = data->task.tk_status;
 247
 248         spin_lock(&dreq->lock);
 249         if (unlikely(status < 0)) {
 250                 dreq->error = status;
 251                 spin_unlock(&dreq->lock);
 252         } else {
 253                 dreq->count += data->res.count;
 254                 spin_unlock(&dreq->lock);
 255                 nfs_direct_dirty_pages(data->pagevec,
 256                                 data->args.pgbase,
 257                                 data->res.count);
 258         }
 259         nfs_direct_release_pages(data->pagevec, data->npages);
 260
 261         if (put_dreq(dreq))
 262                 nfs_direct_complete(dreq);
 263         nfs_readdata_release(data);
 264 }
 265
 266 static const struct rpc_call_ops nfs_read_direct_ops = {
 267         .rpc_call_prepare = nfs_read_prepare,
 268         .rpc_call_done = nfs_direct_read_result,
 269         .rpc_release = nfs_direct_read_release,
 270 };
 271
 272 /*
 273  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 274  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 275  * bail and stop sending more reads.  Read length accounting is
 276  * handled automatically by nfs_direct_read_result().  Otherwise, if
 277  * no requests have been sent, just return an error.
 278  */
 279 static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
 280                                                 const struct iovec *iov,
 281                                                 loff_t pos)
 282 {
 283         struct nfs_open_context *ctx = dreq->ctx;
 284         struct inode *inode = ctx->dentry->d_inode;
 285         unsigned long user_addr = (unsigned long)iov->iov_base;
 286         size_t count = iov->iov_len;
 287         size_t rsize = NFS_SERVER(inode)->rsize;
 288         struct rpc_task *task;
 289         struct rpc_message msg = {
 290                 .rpc_cred = ctx->cred,
 291         };
 292         struct rpc_task_setup task_setup_data = {
 293                 .rpc_client = NFS_CLIENT(inode),
 294                 .rpc_message = &msg,
 295                 .callback_ops = &nfs_read_direct_ops,
 296                 .workqueue = nfsiod_workqueue,
 297                 .flags = RPC_TASK_ASYNC,
 298         };
 299         unsigned int pgbase;
 300         int result;
 301         ssize_t started = 0;
 302
 303         do {
 304                 struct nfs_read_data *data;
 305                 size_t bytes;
 306
 307                 pgbase = user_addr & ~PAGE_MASK;
 308                 bytes = min(rsize,count);
 309
 310                 result = -ENOMEM;
 311                 data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
 312                 if (unlikely(!data))
 313                         break;
 314
 315                 down_read(&current->mm->mmap_sem);
 316                 result = get_user_pages(current, current->mm, user_addr,
 317                                         data->npages, 1, 0, data->pagevec, NULL);
 318                 up_read(&current->mm->mmap_sem);
 319                 if (result < 0) {
 320                         nfs_readdata_free(data);
 321                         break;
 322                 }
 323                 if ((unsigned)result < data->npages) {
 324                         bytes = result * PAGE_SIZE;
 325                         if (bytes <= pgbase) {
 326                                 nfs_direct_release_pages(data->pagevec, result);
 327                                 nfs_readdata_free(data);
 328                                 break;
 329                         }
 330                         bytes -= pgbase;
 331                         data->npages = result;
 332                 }
 333
 334                 get_dreq(dreq);
 335
 336                 data->req = (struct nfs_page *) dreq;
 337                 data->inode = inode;
 338                 data->cred = msg.rpc_cred;
 339                 data->args.fh = NFS_FH(inode);
 340                 data->args.context = get_nfs_open_context(ctx);
 341                 data->args.lock_context = dreq->l_ctx;
 342                 data->args.offset = pos;
 343                 data->args.pgbase = pgbase;
 344                 data->args.pages = data->pagevec;
 345                 data->args.count = bytes;
 346                 data->res.fattr = &data->fattr;
 347                 data->res.eof = 0;
 348                 data->res.count = bytes;
 349                 nfs_fattr_init(&data->fattr);
 350                 msg.rpc_argp = &data->args;
 351                 msg.rpc_resp = &data->res;
 352
 353                 task_setup_data.task = &data->task;
 354                 task_setup_data.callback_data = data;
 355                 NFS_PROTO(inode)->read_setup(data, &msg);
 356
 357                 task = rpc_run_task(&task_setup_data);
 358                 if (IS_ERR(task))
 359                         break;
 360                 rpc_put_task(task);
 361
 362                 dprintk("NFS: %5u initiated direct read call "
 363                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 364                                 data->task.tk_pid,
 365                                 inode->i_sb->s_id,
 366                                 (long long)NFS_FILEID(inode),
 367                                 bytes,
 368                                 (unsigned long long)data->args.offset);
 369
 370                 started += bytes;
 371                 user_addr += bytes;
 372                 pos += bytes;
 373                 /* FIXME: Remove this unnecessary math from final patch */
 374                 pgbase += bytes;
 375                 pgbase &= ~PAGE_MASK;
 376                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 377
 378                 count -= bytes;
 379         } while (count != 0);
 380
 381         if (started)
 382                 return started;
 383         return result < 0 ? (ssize_t) result : -EFAULT;
 384 }
 385
 386 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
 387                                               const struct iovec *iov,
 388                                               unsigned long nr_segs,
 389                                               loff_t pos)
 390 {
 391         ssize_t result = -EINVAL;
 392         size_t requested_bytes = 0;
 393         unsigned long seg;
 394
 395         get_dreq(dreq);
 396
 397         for (seg = 0; seg < nr_segs; seg++) {
 398                 const struct iovec *vec = &iov[seg];
 399                 result = nfs_direct_read_schedule_segment(dreq, vec, pos);
 400                 if (result < 0)
 401                         break;
 402                 requested_bytes += result;
 403                 if ((size_t)result < vec->iov_len)
 404                         break;
 405                 pos += vec->iov_len;
 406         }
 407
 408         /*
 409          * If no bytes were started, return the error, and let the
 410          * generic layer handle the completion.
 411          */
 412         if (requested_bytes == 0) {
 413                 nfs_direct_req_release(dreq);
 414                 return result < 0 ? result : -EIO;
 415         }
 416
 417         if (put_dreq(dreq))
 418                 nfs_direct_complete(dreq);
 419         return 0;
 420 }
 421
 422 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
 423                                unsigned long nr_segs, loff_t pos)
 424 {
 425         ssize_t result = -ENOMEM;
 426         struct inode *inode = iocb->ki_filp->f_mapping->host;
 427         struct nfs_direct_req *dreq;
 428
 429         dreq = nfs_direct_req_alloc();
 430         if (dreq == NULL)
 431                 goto out;
 432
 433         dreq->inode = inode;
 434         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 435         dreq->l_ctx = nfs_get_lock_context(dreq->ctx);
 436         if (dreq->l_ctx == NULL)
 437                 goto out_release;
 438         if (!is_sync_kiocb(iocb))
 439                 dreq->iocb = iocb;
 440
 441         result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
 442         if (!result)
 443                 result = nfs_direct_wait(dreq);
 444 out_release:
 445         nfs_direct_req_release(dreq);
 446 out:
 447         return result;
 448 }
 449
 450 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 451 {
 452         while (!list_empty(&dreq->rewrite_list)) {
 453                 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 454                 list_del(&data->pages);
 455                 nfs_direct_release_pages(data->pagevec, data->npages);
 456                 nfs_writedata_free(data);
 457         }
 458 }
 459
 460 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 461 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 462 {
 463         struct inode *inode = dreq->inode;
 464         struct list_head *p;
 465         struct nfs_write_data *data;
 466         struct rpc_task *task;
 467         struct rpc_message msg = {
 468                 .rpc_cred = dreq->ctx->cred,
 469         };
 470         struct rpc_task_setup task_setup_data = {
 471                 .rpc_client = NFS_CLIENT(inode),
 472                 .rpc_message = &msg,
 473                 .callback_ops = &nfs_write_direct_ops,
 474                 .workqueue = nfsiod_workqueue,
 475                 .flags = RPC_TASK_ASYNC,
 476         };
 477
 478         dreq->count = 0;
 479         get_dreq(dreq);
 480
 481         list_for_each(p, &dreq->rewrite_list) {
 482                 data = list_entry(p, struct nfs_write_data, pages);
 483
 484                 get_dreq(dreq);
 485
 486                 /* Use stable writes */
 487                 data->args.stable = NFS_FILE_SYNC;
 488
 489                 /*
 490                  * Reset data->res.
 491                  */
 492                 nfs_fattr_init(&data->fattr);
 493                 data->res.count = data->args.count;
 494                 memset(&data->verf, 0, sizeof(data->verf));
 495
 496                 /*
 497                  * Reuse data->task; data->args should not have changed
 498                  * since the original request was sent.
 499                  */
 500                 task_setup_data.task = &data->task;
 501                 task_setup_data.callback_data = data;
 502                 msg.rpc_argp = &data->args;
 503                 msg.rpc_resp = &data->res;
 504                 NFS_PROTO(inode)->write_setup(data, &msg);
 505
 506                 /*
 507                  * We're called via an RPC callback, so BKL is already held.
 508                  */
 509                 task = rpc_run_task(&task_setup_data);
 510                 if (!IS_ERR(task))
 511                         rpc_put_task(task);
 512
 513                 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 514                                 data->task.tk_pid,
 515                                 inode->i_sb->s_id,
 516                                 (long long)NFS_FILEID(inode),
 517                                 data->args.count,
 518                                 (unsigned long long)data->args.offset);
 519         }
 520
 521         if (put_dreq(dreq))
 522                 nfs_direct_write_complete(dreq, inode);
 523 }
 524
 525 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
 526 {
 527         struct nfs_commit_data *data = calldata;
 528
 529         /* Call the NFS version-specific code */
 530         NFS_PROTO(data->inode)->commit_done(task, data);
 531 }
 532
 533 static void nfs_direct_commit_release(void *calldata)
 534 {
 535         struct nfs_commit_data *data = calldata;
 536         struct nfs_direct_req *dreq = data->dreq;
 537         int status = data->task.tk_status;
 538
 539         if (status < 0) {
 540                 dprintk("NFS: %5u commit failed with error %d.\n",
 541                                 data->task.tk_pid, status);
 542                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 543         } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
 544                 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
 545                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 546         }
 547
 548         dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
 549         nfs_direct_write_complete(dreq, data->inode);
 550         nfs_commit_free(data);
 551 }
 552
 553 static const struct rpc_call_ops nfs_commit_direct_ops = {
 554         .rpc_call_prepare = nfs_commit_prepare,
 555         .rpc_call_done = nfs_direct_commit_result,
 556         .rpc_release = nfs_direct_commit_release,
 557 };
 558
 559 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 560 {
 561         struct nfs_commit_data *data = dreq->commit_data;
 562         struct rpc_task *task;
 563         struct rpc_message msg = {
 564                 .rpc_argp = &data->args,
 565                 .rpc_resp = &data->res,
 566                 .rpc_cred = dreq->ctx->cred,
 567         };
 568         struct rpc_task_setup task_setup_data = {
 569                 .task = &data->task,
 570                 .rpc_client = NFS_CLIENT(dreq->inode),
 571                 .rpc_message = &msg,
 572                 .callback_ops = &nfs_commit_direct_ops,
 573                 .callback_data = data,
 574                 .workqueue = nfsiod_workqueue,
 575                 .flags = RPC_TASK_ASYNC,
 576         };
 577
 578         data->inode = dreq->inode;
 579         data->cred = msg.rpc_cred;
 580
 581         data->args.fh = NFS_FH(data->inode);
 582         data->args.offset = 0;
 583         data->args.count = 0;
 584         data->res.fattr = &data->fattr;
 585         data->res.verf = &data->verf;
 586         nfs_fattr_init(&data->fattr);
 587
 588         NFS_PROTO(data->inode)->commit_setup(data, &msg);
 589
 590         /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
 591         dreq->commit_data = NULL;
 592
 593         dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
 594
 595         task = rpc_run_task(&task_setup_data);
 596         if (!IS_ERR(task))
 597                 rpc_put_task(task);
 598 }
 599
 600 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 601 {
 602         int flags = dreq->flags;
 603
 604         dreq->flags = 0;
 605         switch (flags) {
 606                 case NFS_ODIRECT_DO_COMMIT:
 607                         nfs_direct_commit_schedule(dreq);
 608                         break;
 609                 case NFS_ODIRECT_RESCHED_WRITES:
 610                         nfs_direct_write_reschedule(dreq);
 611                         break;
 612                 default:
 613                         if (dreq->commit_data != NULL)
 614                                 nfs_commit_free(dreq->commit_data);
 615                         nfs_direct_free_writedata(dreq);
 616                         nfs_zap_mapping(inode, inode->i_mapping);
 617                         nfs_direct_complete(dreq);
 618         }
 619 }
 620
 621 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 622 {
 623         dreq->commit_data = nfs_commitdata_alloc();
 624         if (dreq->commit_data != NULL)
 625                 dreq->commit_data->dreq = dreq;
 626 }
 627 #else
 628 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 629 {
 630         dreq->commit_data = NULL;
 631 }
 632
 633 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 634 {
 635         nfs_direct_free_writedata(dreq);
 636         nfs_zap_mapping(inode, inode->i_mapping);
 637         nfs_direct_complete(dreq);
 638 }
 639 #endif
 640
 641 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 642 {
 643         struct nfs_write_data *data = calldata;
 644
 645         nfs_writeback_done(task, data);
 646 }
 647
 648 /*
 649  * NB: Return the value of the first error return code.  Subsequent
 650  *     errors after the first one are ignored.
 651  */
 652 static void nfs_direct_write_release(void *calldata)
 653 {
 654         struct nfs_write_data *data = calldata;
 655         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 656         int status = data->task.tk_status;
 657
 658         spin_lock(&dreq->lock);
 659
 660         if (unlikely(status < 0)) {
 661                 /* An error has occurred, so we should not commit */
 662                 dreq->flags = 0;
 663                 dreq->error = status;
 664         }
 665         if (unlikely(dreq->error != 0))
 666                 goto out_unlock;
 667
 668         dreq->count += data->res.count;
 669
 670         if (data->res.verf->committed != NFS_FILE_SYNC) {
 671                 switch (dreq->flags) {
 672                         case 0:
 673                                 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
 674                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
 675                                 break;
 676                         case NFS_ODIRECT_DO_COMMIT:
 677                                 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
 678                                         dprintk("NFS: %5u write verify failed\n", data->task.tk_pid);
 679                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 680                                 }
 681                 }
 682         }
 683 out_unlock:
 684         spin_unlock(&dreq->lock);
 685
 686         if (put_dreq(dreq))
 687                 nfs_direct_write_complete(dreq, data->inode);
 688 }
 689
 690 static const struct rpc_call_ops nfs_write_direct_ops = {
 691         .rpc_call_prepare = nfs_write_prepare,
 692         .rpc_call_done = nfs_direct_write_result,
 693         .rpc_release = nfs_direct_write_release,
 694 };
 695
 696 /*
 697  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 698  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 699  * bail and stop sending more writes.  Write length accounting is
 700  * handled automatically by nfs_direct_write_result().  Otherwise, if
 701  * no requests have been sent, just return an error.
 702  */
 703 static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
 704                                                  const struct iovec *iov,
 705                                                  loff_t pos, int sync)
 706 {
 707         struct nfs_open_context *ctx = dreq->ctx;
 708         struct inode *inode = ctx->dentry->d_inode;
 709         unsigned long user_addr = (unsigned long)iov->iov_base;
 710         size_t count = iov->iov_len;
 711         struct rpc_task *task;
 712         struct rpc_message msg = {
 713                 .rpc_cred = ctx->cred,
 714         };
 715         struct rpc_task_setup task_setup_data = {
 716                 .rpc_client = NFS_CLIENT(inode),
 717                 .rpc_message = &msg,
 718                 .callback_ops = &nfs_write_direct_ops,
 719                 .workqueue = nfsiod_workqueue,
 720                 .flags = RPC_TASK_ASYNC,
 721         };
 722         size_t wsize = NFS_SERVER(inode)->wsize;
 723         unsigned int pgbase;
 724         int result;
 725         ssize_t started = 0;
 726
 727         do {
 728                 struct nfs_write_data *data;
 729                 size_t bytes;
 730
 731                 pgbase = user_addr & ~PAGE_MASK;
 732                 bytes = min(wsize,count);
 733
 734                 result = -ENOMEM;
 735                 data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
 736                 if (unlikely(!data))
 737                         break;
 738
 739                 down_read(&current->mm->mmap_sem);
 740                 result = get_user_pages(current, current->mm, user_addr,
 741                                         data->npages, 0, 0, data->pagevec, NULL);
 742                 up_read(&current->mm->mmap_sem);
 743                 if (result < 0) {
 744                         nfs_writedata_free(data);
 745                         break;
 746                 }
 747                 if ((unsigned)result < data->npages) {
 748                         bytes = result * PAGE_SIZE;
 749                         if (bytes <= pgbase) {
 750                                 nfs_direct_release_pages(data->pagevec, result);
 751                                 nfs_writedata_free(data);
 752                                 break;
 753                         }
 754                         bytes -= pgbase;
 755                         data->npages = result;
 756                 }
 757
 758                 get_dreq(dreq);
 759
 760                 list_move_tail(&data->pages, &dreq->rewrite_list);
 761
 762                 data->req = (struct nfs_page *) dreq;
 763                 data->inode = inode;
 764                 data->cred = msg.rpc_cred;
 765                 data->args.fh = NFS_FH(inode);
 766                 data->args.context = ctx;
 767                 data->args.lock_context = dreq->l_ctx;
 768                 data->args.offset = pos;
 769                 data->args.pgbase = pgbase;
 770                 data->args.pages = data->pagevec;
 771                 data->args.count = bytes;
 772                 data->args.stable = sync;
 773                 data->res.fattr = &data->fattr;
 774                 data->res.count = bytes;
 775                 data->res.verf = &data->verf;
 776                 nfs_fattr_init(&data->fattr);
 777
 778                 task_setup_data.task = &data->task;
 779                 task_setup_data.callback_data = data;
 780                 msg.rpc_argp = &data->args;
 781                 msg.rpc_resp = &data->res;
 782                 NFS_PROTO(inode)->write_setup(data, &msg);
 783
 784                 task = rpc_run_task(&task_setup_data);
 785                 if (IS_ERR(task))
 786                         break;
 787                 rpc_put_task(task);
 788
 789                 dprintk("NFS: %5u initiated direct write call "
 790                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 791                                 data->task.tk_pid,
 792                                 inode->i_sb->s_id,
 793                                 (long long)NFS_FILEID(inode),
 794                                 bytes,
 795                                 (unsigned long long)data->args.offset);
 796
 797                 started += bytes;
 798                 user_addr += bytes;
 799                 pos += bytes;
 800
 801                 /* FIXME: Remove this useless math from the final patch */
 802                 pgbase += bytes;
 803                 pgbase &= ~PAGE_MASK;
 804                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 805
 806                 count -= bytes;
 807         } while (count != 0);
 808
 809         if (started)
 810                 return started;
 811         return result < 0 ? (ssize_t) result : -EFAULT;
 812 }
 813
 814 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 815                                                const struct iovec *iov,
 816                                                unsigned long nr_segs,
 817                                                loff_t pos, int sync)
 818 {
 819         ssize_t result = 0;
 820         size_t requested_bytes = 0;
 821         unsigned long seg;
 822
 823         get_dreq(dreq);
 824
 825         for (seg = 0; seg < nr_segs; seg++) {
 826                 const struct iovec *vec = &iov[seg];
 827                 result = nfs_direct_write_schedule_segment(dreq, vec,
 828                                                            pos, sync);
 829                 if (result < 0)
 830                         break;
 831                 requested_bytes += result;
 832                 if ((size_t)result < vec->iov_len)
 833                         break;
 834                 pos += vec->iov_len;
 835         }
 836
 837         /*
 838          * If no bytes were started, return the error, and let the
 839          * generic layer handle the completion.
 840          */
 841         if (requested_bytes == 0) {
 842                 nfs_direct_req_release(dreq);
 843                 return result < 0 ? result : -EIO;
 844         }
 845
 846         if (put_dreq(dreq))
 847                 nfs_direct_write_complete(dreq, dreq->inode);
 848         return 0;
 849 }
 850
 851 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
 852                                 unsigned long nr_segs, loff_t pos,
 853                                 size_t count)
 854 {
 855         ssize_t result = -ENOMEM;
 856         struct inode *inode = iocb->ki_filp->f_mapping->host;
 857         struct nfs_direct_req *dreq;
 858         size_t wsize = NFS_SERVER(inode)->wsize;
 859         int sync = NFS_UNSTABLE;
 860
 861         dreq = nfs_direct_req_alloc();
 862         if (!dreq)
 863                 goto out;
 864         nfs_alloc_commit_data(dreq);
 865
 866         if (dreq->commit_data == NULL || count <= wsize)
 867                 sync = NFS_FILE_SYNC;
 868
 869         dreq->inode = inode;
 870         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 871         dreq->l_ctx = nfs_get_lock_context(dreq->ctx);
 872         if (dreq->l_ctx == NULL)
 873                 goto out_release;
 874         if (!is_sync_kiocb(iocb))
 875                 dreq->iocb = iocb;
 876
 877         result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
 878         if (!result)
 879                 result = nfs_direct_wait(dreq);
 880 out_release:
 881         nfs_direct_req_release(dreq);
 882 out:
 883         return result;
 884 }
 885
 886 /**
 887  * nfs_file_direct_read - file direct read operation for NFS files
 888  * @iocb: target I/O control block
 889  * @iov: vector of user buffers into which to read data
 890  * @nr_segs: size of iov vector
 891  * @pos: byte offset in file where reading starts
 892  *
 893  * We use this function for direct reads instead of calling
 894  * generic_file_aio_read() in order to avoid gfar's check to see if
 895  * the request starts before the end of the file.  For that check
 896  * to work, we must generate a GETATTR before each direct read, and
 897  * even then there is a window between the GETATTR and the subsequent
 898  * READ where the file size could change.  Our preference is simply
 899  * to do all reads the application wants, and the server will take
 900  * care of managing the end of file boundary.
 901  *
 902  * This function also eliminates unnecessarily updating the file's
 903  * atime locally, as the NFS server sets the file's atime, and this
 904  * client must read the updated atime from the server back into its
 905  * cache.
 906  */
 907 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
 908                                 unsigned long nr_segs, loff_t pos)
 909 {
 910         ssize_t retval = -EINVAL;
 911         struct file *file = iocb->ki_filp;
 912         struct address_space *mapping = file->f_mapping;
 913         size_t count;
 914
 915         count = iov_length(iov, nr_segs);
 916         nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
 917
 918         dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
 919                 file->f_path.dentry->d_parent->d_name.name,
 920                 file->f_path.dentry->d_name.name,
 921                 count, (long long) pos);
 922
 923         retval = 0;
 924         if (!count)
 925                 goto out;
 926
 927         retval = nfs_sync_mapping(mapping);
 928         if (retval)
 929                 goto out;
 930
 931         task_io_account_read(count);
 932
 933         retval = nfs_direct_read(iocb, iov, nr_segs, pos);
 934         if (retval > 0)
 935                 iocb->ki_pos = pos + retval;
 936
 937 out:
 938         return retval;
 939 }
 940
 941 /**
 942  * nfs_file_direct_write - file direct write operation for NFS files
 943  * @iocb: target I/O control block
 944  * @iov: vector of user buffers from which to write data
 945  * @nr_segs: size of iov vector
 946  * @pos: byte offset in file where writing starts
 947  *
 948  * We use this function for direct writes instead of calling
 949  * generic_file_aio_write() in order to avoid taking the inode
 950  * semaphore and updating the i_size.  The NFS server will set
 951  * the new i_size and this client must read the updated size
 952  * back into its cache.  We let the server do generic write
 953  * parameter checking and report problems.
 954  *
 955  * We eliminate local atime updates, see direct read above.
 956  *
 957  * We avoid unnecessary page cache invalidations for normal cached
 958  * readers of this file.
 959  *
 960  * Note that O_APPEND is not supported for NFS direct writes, as there
 961  * is no atomic O_APPEND write facility in the NFS protocol.
 962  */
 963 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
 964                                 unsigned long nr_segs, loff_t pos)
 965 {
 966         ssize_t retval = -EINVAL;
 967         struct file *file = iocb->ki_filp;
 968         struct address_space *mapping = file->f_mapping;
 969         size_t count;
 970
 971         count = iov_length(iov, nr_segs);
 972         nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
 973
 974         dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
 975                 file->f_path.dentry->d_parent->d_name.name,
 976                 file->f_path.dentry->d_name.name,
 977                 count, (long long) pos);
 978
 979         retval = generic_write_checks(file, &pos, &count, 0);
 980         if (retval)
 981                 goto out;
 982
 983         retval = -EINVAL;
 984         if ((ssize_t) count < 0)
 985                 goto out;
 986         retval = 0;
 987         if (!count)
 988                 goto out;
 989
 990         retval = nfs_sync_mapping(mapping);
 991         if (retval)
 992                 goto out;
 993
 994         task_io_account_write(count);
 995
 996         retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
 997
 998         if (retval > 0)
 999                 iocb->ki_pos = pos + retval;
1000
1001 out:
1002         return retval;
1003 }
1004
1005 /**
1006  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
1007  *
1008  */
1009 int __init nfs_init_directcache(void)
1010 {
1011         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
1012                                                 sizeof(struct nfs_direct_req),
1013                                                 0, (SLAB_RECLAIM_ACCOUNT|
1014                                                         SLAB_MEM_SPREAD),
1015                                                 NULL);
1016         if (nfs_direct_cachep == NULL)
1017                 return -ENOMEM;
1018
1019         return 0;
1020 }
1021
1022 /**
1023  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1024  *
1025  */
1026 void nfs_destroy_directcache(void)
1027 {
1028         kmem_cache_destroy(nfs_direct_cachep);
1029 }