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
  49 #include <linux/nfs_fs.h>
  50 #include <linux/nfs_page.h>
  51 #include <linux/sunrpc/clnt.h>
  52
  53 #include <asm/system.h>
  54 #include <asm/uaccess.h>
  55 #include <asm/atomic.h>
  56
  57 #include "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 kiocb *          iocb;           /* controlling i/o request */
  73         struct inode *          inode;          /* target file of i/o */
  74
  75         /* completion state */
  76         atomic_t                io_count;       /* i/os we're waiting for */
  77         spinlock_t              lock;           /* protect completion state */
  78         ssize_t                 count,          /* bytes actually processed */
  79                                 error;          /* any reported error */
  80         struct completion       completion;     /* wait for i/o completion */
  81
  82         /* commit state */
  83         struct list_head        rewrite_list;   /* saved nfs_write_data structs */
  84         struct nfs_write_data * commit_data;    /* special write_data for commits */
  85         int                     flags;
  86 #define NFS_ODIRECT_DO_COMMIT           (1)     /* an unstable reply was received */
  87 #define NFS_ODIRECT_RESCHED_WRITES      (2)     /* write verification failed */
  88         struct nfs_writeverf    verf;           /* unstable write verifier */
  89 };
  90
  91 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
  92 static const struct rpc_call_ops nfs_write_direct_ops;
  93
  94 static inline void get_dreq(struct nfs_direct_req *dreq)
  95 {
  96         atomic_inc(&dreq->io_count);
  97 }
  98
  99 static inline int put_dreq(struct nfs_direct_req *dreq)
 100 {
 101         return atomic_dec_and_test(&dreq->io_count);
 102 }
 103
 104 /**
 105  * nfs_direct_IO - NFS address space operation for direct I/O
 106  * @rw: direction (read or write)
 107  * @iocb: target I/O control block
 108  * @iov: array of vectors that define I/O buffer
 109  * @pos: offset in file to begin the operation
 110  * @nr_segs: size of iovec array
 111  *
 112  * The presence of this routine in the address space ops vector means
 113  * the NFS client supports direct I/O.  However, we shunt off direct
 114  * read and write requests before the VFS gets them, so this method
 115  * should never be called.
 116  */
 117 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
 118 {
 119         dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
 120                         iocb->ki_filp->f_path.dentry->d_name.name,
 121                         (long long) pos, nr_segs);
 122
 123         return -EINVAL;
 124 }
 125
 126 static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
 127 {
 128         unsigned int npages;
 129         unsigned int i;
 130
 131         if (count == 0)
 132                 return;
 133         pages += (pgbase >> PAGE_SHIFT);
 134         npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 135         for (i = 0; i < npages; i++) {
 136                 struct page *page = pages[i];
 137                 if (!PageCompound(page))
 138                         set_page_dirty(page);
 139         }
 140 }
 141
 142 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
 143 {
 144         unsigned int i;
 145         for (i = 0; i < npages; i++)
 146                 page_cache_release(pages[i]);
 147 }
 148
 149 static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
 150 {
 151         struct nfs_direct_req *dreq;
 152
 153         dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
 154         if (!dreq)
 155                 return NULL;
 156
 157         kref_init(&dreq->kref);
 158         kref_get(&dreq->kref);
 159         init_completion(&dreq->completion);
 160         INIT_LIST_HEAD(&dreq->rewrite_list);
 161         dreq->iocb = NULL;
 162         dreq->ctx = NULL;
 163         spin_lock_init(&dreq->lock);
 164         atomic_set(&dreq->io_count, 0);
 165         dreq->count = 0;
 166         dreq->error = 0;
 167         dreq->flags = 0;
 168
 169         return dreq;
 170 }
 171
 172 static void nfs_direct_req_free(struct kref *kref)
 173 {
 174         struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
 175
 176         if (dreq->ctx != NULL)
 177                 put_nfs_open_context(dreq->ctx);
 178         kmem_cache_free(nfs_direct_cachep, dreq);
 179 }
 180
 181 static void nfs_direct_req_release(struct nfs_direct_req *dreq)
 182 {
 183         kref_put(&dreq->kref, nfs_direct_req_free);
 184 }
 185
 186 /*
 187  * Collects and returns the final error value/byte-count.
 188  */
 189 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
 190 {
 191         ssize_t result = -EIOCBQUEUED;
 192
 193         /* Async requests don't wait here */
 194         if (dreq->iocb)
 195                 goto out;
 196
 197         result = wait_for_completion_killable(&dreq->completion);
 198
 199         if (!result)
 200                 result = dreq->error;
 201         if (!result)
 202                 result = dreq->count;
 203
 204 out:
 205         return (ssize_t) result;
 206 }
 207
 208 /*
 209  * Synchronous I/O uses a stack-allocated iocb.  Thus we can't trust
 210  * the iocb is still valid here if this is a synchronous request.
 211  */
 212 static void nfs_direct_complete(struct nfs_direct_req *dreq)
 213 {
 214         if (dreq->iocb) {
 215                 long res = (long) dreq->error;
 216                 if (!res)
 217                         res = (long) dreq->count;
 218                 aio_complete(dreq->iocb, res, 0);
 219         }
 220         complete_all(&dreq->completion);
 221
 222         nfs_direct_req_release(dreq);
 223 }
 224
 225 /*
 226  * We must hold a reference to all the pages in this direct read request
 227  * until the RPCs complete.  This could be long *after* we are woken up in
 228  * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
 229  */
 230 static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
 231 {
 232         struct nfs_read_data *data = calldata;
 233
 234         nfs_readpage_result(task, data);
 235 }
 236
 237 static void nfs_direct_read_release(void *calldata)
 238 {
 239
 240         struct nfs_read_data *data = calldata;
 241         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 242         int status = data->task.tk_status;
 243
 244         spin_lock(&dreq->lock);
 245         if (unlikely(status < 0)) {
 246                 dreq->error = status;
 247                 spin_unlock(&dreq->lock);
 248         } else {
 249                 dreq->count += data->res.count;
 250                 spin_unlock(&dreq->lock);
 251                 nfs_direct_dirty_pages(data->pagevec,
 252                                 data->args.pgbase,
 253                                 data->res.count);
 254         }
 255         nfs_direct_release_pages(data->pagevec, data->npages);
 256
 257         if (put_dreq(dreq))
 258                 nfs_direct_complete(dreq);
 259         nfs_readdata_free(data);
 260 }
 261
 262 static const struct rpc_call_ops nfs_read_direct_ops = {
 263 #if defined(CONFIG_NFS_V4_1)
 264         .rpc_call_prepare = nfs_read_prepare,
 265 #endif /* CONFIG_NFS_V4_1 */
 266         .rpc_call_done = nfs_direct_read_result,
 267         .rpc_release = nfs_direct_read_release,
 268 };
 269
 270 /*
 271  * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
 272  * operation.  If nfs_readdata_alloc() or get_user_pages() fails,
 273  * bail and stop sending more reads.  Read length accounting is
 274  * handled automatically by nfs_direct_read_result().  Otherwise, if
 275  * no requests have been sent, just return an error.
 276  */
 277 static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
 278                                                 const struct iovec *iov,
 279                                                 loff_t pos)
 280 {
 281         struct nfs_open_context *ctx = dreq->ctx;
 282         struct inode *inode = ctx->path.dentry->d_inode;
 283         unsigned long user_addr = (unsigned long)iov->iov_base;
 284         size_t count = iov->iov_len;
 285         size_t rsize = NFS_SERVER(inode)->rsize;
 286         struct rpc_task *task;
 287         struct rpc_message msg = {
 288                 .rpc_cred = ctx->cred,
 289         };
 290         struct rpc_task_setup task_setup_data = {
 291                 .rpc_client = NFS_CLIENT(inode),
 292                 .rpc_message = &msg,
 293                 .callback_ops = &nfs_read_direct_ops,
 294                 .workqueue = nfsiod_workqueue,
 295                 .flags = RPC_TASK_ASYNC,
 296         };
 297         unsigned int pgbase;
 298         int result;
 299         ssize_t started = 0;
 300
 301         do {
 302                 struct nfs_read_data *data;
 303                 size_t bytes;
 304
 305                 pgbase = user_addr & ~PAGE_MASK;
 306                 bytes = min(rsize,count);
 307
 308                 result = -ENOMEM;
 309                 data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
 310                 if (unlikely(!data))
 311                         break;
 312
 313                 down_read(&current->mm->mmap_sem);
 314                 result = get_user_pages(current, current->mm, user_addr,
 315                                         data->npages, 1, 0, data->pagevec, NULL);
 316                 up_read(&current->mm->mmap_sem);
 317                 if (result < 0) {
 318                         nfs_readdata_free(data);
 319                         break;
 320                 }
 321                 if ((unsigned)result < data->npages) {
 322                         bytes = result * PAGE_SIZE;
 323                         if (bytes <= pgbase) {
 324                                 nfs_direct_release_pages(data->pagevec, result);
 325                                 nfs_readdata_free(data);
 326                                 break;
 327                         }
 328                         bytes -= pgbase;
 329                         data->npages = result;
 330                 }
 331
 332                 get_dreq(dreq);
 333
 334                 data->req = (struct nfs_page *) dreq;
 335                 data->inode = inode;
 336                 data->cred = msg.rpc_cred;
 337                 data->args.fh = NFS_FH(inode);
 338                 data->args.context = ctx;
 339                 data->args.offset = pos;
 340                 data->args.pgbase = pgbase;
 341                 data->args.pages = data->pagevec;
 342                 data->args.count = bytes;
 343                 data->res.fattr = &data->fattr;
 344                 data->res.eof = 0;
 345                 data->res.count = bytes;
 346                 nfs_fattr_init(&data->fattr);
 347                 msg.rpc_argp = &data->args;
 348                 msg.rpc_resp = &data->res;
 349
 350                 task_setup_data.task = &data->task;
 351                 task_setup_data.callback_data = data;
 352                 NFS_PROTO(inode)->read_setup(data, &msg);
 353
 354                 task = rpc_run_task(&task_setup_data);
 355                 if (IS_ERR(task))
 356                         break;
 357                 rpc_put_task(task);
 358
 359                 dprintk("NFS: %5u initiated direct read call "
 360                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 361                                 data->task.tk_pid,
 362                                 inode->i_sb->s_id,
 363                                 (long long)NFS_FILEID(inode),
 364                                 bytes,
 365                                 (unsigned long long)data->args.offset);
 366
 367                 started += bytes;
 368                 user_addr += bytes;
 369                 pos += bytes;
 370                 /* FIXME: Remove this unnecessary math from final patch */
 371                 pgbase += bytes;
 372                 pgbase &= ~PAGE_MASK;
 373                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 374
 375                 count -= bytes;
 376         } while (count != 0);
 377
 378         if (started)
 379                 return started;
 380         return result < 0 ? (ssize_t) result : -EFAULT;
 381 }
 382
 383 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
 384                                               const struct iovec *iov,
 385                                               unsigned long nr_segs,
 386                                               loff_t pos)
 387 {
 388         ssize_t result = -EINVAL;
 389         size_t requested_bytes = 0;
 390         unsigned long seg;
 391
 392         get_dreq(dreq);
 393
 394         for (seg = 0; seg < nr_segs; seg++) {
 395                 const struct iovec *vec = &iov[seg];
 396                 result = nfs_direct_read_schedule_segment(dreq, vec, pos);
 397                 if (result < 0)
 398                         break;
 399                 requested_bytes += result;
 400                 if ((size_t)result < vec->iov_len)
 401                         break;
 402                 pos += vec->iov_len;
 403         }
 404
 405         if (put_dreq(dreq))
 406                 nfs_direct_complete(dreq);
 407
 408         if (requested_bytes != 0)
 409                 return 0;
 410
 411         if (result < 0)
 412                 return result;
 413         return -EIO;
 414 }
 415
 416 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
 417                                unsigned long nr_segs, loff_t pos)
 418 {
 419         ssize_t result = 0;
 420         struct inode *inode = iocb->ki_filp->f_mapping->host;
 421         struct nfs_direct_req *dreq;
 422
 423         dreq = nfs_direct_req_alloc();
 424         if (!dreq)
 425                 return -ENOMEM;
 426
 427         dreq->inode = inode;
 428         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 429         if (!is_sync_kiocb(iocb))
 430                 dreq->iocb = iocb;
 431
 432         result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
 433         if (!result)
 434                 result = nfs_direct_wait(dreq);
 435         nfs_direct_req_release(dreq);
 436
 437         return result;
 438 }
 439
 440 static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
 441 {
 442         while (!list_empty(&dreq->rewrite_list)) {
 443                 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
 444                 list_del(&data->pages);
 445                 nfs_direct_release_pages(data->pagevec, data->npages);
 446                 nfs_writedata_free(data);
 447         }
 448 }
 449
 450 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 451 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
 452 {
 453         struct inode *inode = dreq->inode;
 454         struct list_head *p;
 455         struct nfs_write_data *data;
 456         struct rpc_task *task;
 457         struct rpc_message msg = {
 458                 .rpc_cred = dreq->ctx->cred,
 459         };
 460         struct rpc_task_setup task_setup_data = {
 461                 .rpc_client = NFS_CLIENT(inode),
 462                 .rpc_message = &msg,
 463                 .callback_ops = &nfs_write_direct_ops,
 464                 .workqueue = nfsiod_workqueue,
 465                 .flags = RPC_TASK_ASYNC,
 466         };
 467
 468         dreq->count = 0;
 469         get_dreq(dreq);
 470
 471         list_for_each(p, &dreq->rewrite_list) {
 472                 data = list_entry(p, struct nfs_write_data, pages);
 473
 474                 get_dreq(dreq);
 475
 476                 /* Use stable writes */
 477                 data->args.stable = NFS_FILE_SYNC;
 478
 479                 /*
 480                  * Reset data->res.
 481                  */
 482                 nfs_fattr_init(&data->fattr);
 483                 data->res.count = data->args.count;
 484                 memset(&data->verf, 0, sizeof(data->verf));
 485
 486                 /*
 487                  * Reuse data->task; data->args should not have changed
 488                  * since the original request was sent.
 489                  */
 490                 task_setup_data.task = &data->task;
 491                 task_setup_data.callback_data = data;
 492                 msg.rpc_argp = &data->args;
 493                 msg.rpc_resp = &data->res;
 494                 NFS_PROTO(inode)->write_setup(data, &msg);
 495
 496                 /*
 497                  * We're called via an RPC callback, so BKL is already held.
 498                  */
 499                 task = rpc_run_task(&task_setup_data);
 500                 if (!IS_ERR(task))
 501                         rpc_put_task(task);
 502
 503                 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
 504                                 data->task.tk_pid,
 505                                 inode->i_sb->s_id,
 506                                 (long long)NFS_FILEID(inode),
 507                                 data->args.count,
 508                                 (unsigned long long)data->args.offset);
 509         }
 510
 511         if (put_dreq(dreq))
 512                 nfs_direct_write_complete(dreq, inode);
 513 }
 514
 515 static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
 516 {
 517         struct nfs_write_data *data = calldata;
 518
 519         /* Call the NFS version-specific code */
 520         NFS_PROTO(data->inode)->commit_done(task, data);
 521 }
 522
 523 static void nfs_direct_commit_release(void *calldata)
 524 {
 525         struct nfs_write_data *data = calldata;
 526         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 527         int status = data->task.tk_status;
 528
 529         if (status < 0) {
 530                 dprintk("NFS: %5u commit failed with error %d.\n",
 531                                 data->task.tk_pid, status);
 532                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 533         } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
 534                 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid);
 535                 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 536         }
 537
 538         dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status);
 539         nfs_direct_write_complete(dreq, data->inode);
 540         nfs_commit_free(data);
 541 }
 542
 543 static const struct rpc_call_ops nfs_commit_direct_ops = {
 544 #if defined(CONFIG_NFS_V4_1)
 545         .rpc_call_prepare = nfs_write_prepare,
 546 #endif /* CONFIG_NFS_V4_1 */
 547         .rpc_call_done = nfs_direct_commit_result,
 548         .rpc_release = nfs_direct_commit_release,
 549 };
 550
 551 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
 552 {
 553         struct nfs_write_data *data = dreq->commit_data;
 554         struct rpc_task *task;
 555         struct rpc_message msg = {
 556                 .rpc_argp = &data->args,
 557                 .rpc_resp = &data->res,
 558                 .rpc_cred = dreq->ctx->cred,
 559         };
 560         struct rpc_task_setup task_setup_data = {
 561                 .task = &data->task,
 562                 .rpc_client = NFS_CLIENT(dreq->inode),
 563                 .rpc_message = &msg,
 564                 .callback_ops = &nfs_commit_direct_ops,
 565                 .callback_data = data,
 566                 .workqueue = nfsiod_workqueue,
 567                 .flags = RPC_TASK_ASYNC,
 568         };
 569
 570         data->inode = dreq->inode;
 571         data->cred = msg.rpc_cred;
 572
 573         data->args.fh = NFS_FH(data->inode);
 574         data->args.offset = 0;
 575         data->args.count = 0;
 576         data->args.context = dreq->ctx;
 577         data->res.count = 0;
 578         data->res.fattr = &data->fattr;
 579         data->res.verf = &data->verf;
 580         nfs_fattr_init(&data->fattr);
 581
 582         NFS_PROTO(data->inode)->commit_setup(data, &msg);
 583
 584         /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
 585         dreq->commit_data = NULL;
 586
 587         dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
 588
 589         task = rpc_run_task(&task_setup_data);
 590         if (!IS_ERR(task))
 591                 rpc_put_task(task);
 592 }
 593
 594 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 595 {
 596         int flags = dreq->flags;
 597
 598         dreq->flags = 0;
 599         switch (flags) {
 600                 case NFS_ODIRECT_DO_COMMIT:
 601                         nfs_direct_commit_schedule(dreq);
 602                         break;
 603                 case NFS_ODIRECT_RESCHED_WRITES:
 604                         nfs_direct_write_reschedule(dreq);
 605                         break;
 606                 default:
 607                         if (dreq->commit_data != NULL)
 608                                 nfs_commit_free(dreq->commit_data);
 609                         nfs_direct_free_writedata(dreq);
 610                         nfs_zap_mapping(inode, inode->i_mapping);
 611                         nfs_direct_complete(dreq);
 612         }
 613 }
 614
 615 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 616 {
 617         dreq->commit_data = nfs_commitdata_alloc();
 618         if (dreq->commit_data != NULL)
 619                 dreq->commit_data->req = (struct nfs_page *) dreq;
 620 }
 621 #else
 622 static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 623 {
 624         dreq->commit_data = NULL;
 625 }
 626
 627 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 628 {
 629         nfs_direct_free_writedata(dreq);
 630         nfs_zap_mapping(inode, inode->i_mapping);
 631         nfs_direct_complete(dreq);
 632 }
 633 #endif
 634
 635 static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
 636 {
 637         struct nfs_write_data *data = calldata;
 638
 639         if (nfs_writeback_done(task, data) != 0)
 640                 return;
 641 }
 642
 643 /*
 644  * NB: Return the value of the first error return code.  Subsequent
 645  *     errors after the first one are ignored.
 646  */
 647 static void nfs_direct_write_release(void *calldata)
 648 {
 649         struct nfs_write_data *data = calldata;
 650         struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
 651         int status = data->task.tk_status;
 652
 653         spin_lock(&dreq->lock);
 654
 655         if (unlikely(status < 0)) {
 656                 /* An error has occurred, so we should not commit */
 657                 dreq->flags = 0;
 658                 dreq->error = status;
 659         }
 660         if (unlikely(dreq->error != 0))
 661                 goto out_unlock;
 662
 663         dreq->count += data->res.count;
 664
 665         if (data->res.verf->committed != NFS_FILE_SYNC) {
 666                 switch (dreq->flags) {
 667                         case 0:
 668                                 memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
 669                                 dreq->flags = NFS_ODIRECT_DO_COMMIT;
 670                                 break;
 671                         case NFS_ODIRECT_DO_COMMIT:
 672                                 if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
 673                                         dprintk("NFS: %5u write verify failed\n", data->task.tk_pid);
 674                                         dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
 675                                 }
 676                 }
 677         }
 678 out_unlock:
 679         spin_unlock(&dreq->lock);
 680
 681         if (put_dreq(dreq))
 682                 nfs_direct_write_complete(dreq, data->inode);
 683 }
 684
 685 static const struct rpc_call_ops nfs_write_direct_ops = {
 686 #if defined(CONFIG_NFS_V4_1)
 687         .rpc_call_prepare = nfs_write_prepare,
 688 #endif /* CONFIG_NFS_V4_1 */
 689         .rpc_call_done = nfs_direct_write_result,
 690         .rpc_release = nfs_direct_write_release,
 691 };
 692
 693 /*
 694  * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
 695  * operation.  If nfs_writedata_alloc() or get_user_pages() fails,
 696  * bail and stop sending more writes.  Write length accounting is
 697  * handled automatically by nfs_direct_write_result().  Otherwise, if
 698  * no requests have been sent, just return an error.
 699  */
 700 static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
 701                                                  const struct iovec *iov,
 702                                                  loff_t pos, int sync)
 703 {
 704         struct nfs_open_context *ctx = dreq->ctx;
 705         struct inode *inode = ctx->path.dentry->d_inode;
 706         unsigned long user_addr = (unsigned long)iov->iov_base;
 707         size_t count = iov->iov_len;
 708         struct rpc_task *task;
 709         struct rpc_message msg = {
 710                 .rpc_cred = ctx->cred,
 711         };
 712         struct rpc_task_setup task_setup_data = {
 713                 .rpc_client = NFS_CLIENT(inode),
 714                 .rpc_message = &msg,
 715                 .callback_ops = &nfs_write_direct_ops,
 716                 .workqueue = nfsiod_workqueue,
 717                 .flags = RPC_TASK_ASYNC,
 718         };
 719         size_t wsize = NFS_SERVER(inode)->wsize;
 720         unsigned int pgbase;
 721         int result;
 722         ssize_t started = 0;
 723
 724         do {
 725                 struct nfs_write_data *data;
 726                 size_t bytes;
 727
 728                 pgbase = user_addr & ~PAGE_MASK;
 729                 bytes = min(wsize,count);
 730
 731                 result = -ENOMEM;
 732                 data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
 733                 if (unlikely(!data))
 734                         break;
 735
 736                 down_read(&current->mm->mmap_sem);
 737                 result = get_user_pages(current, current->mm, user_addr,
 738                                         data->npages, 0, 0, data->pagevec, NULL);
 739                 up_read(&current->mm->mmap_sem);
 740                 if (result < 0) {
 741                         nfs_writedata_free(data);
 742                         break;
 743                 }
 744                 if ((unsigned)result < data->npages) {
 745                         bytes = result * PAGE_SIZE;
 746                         if (bytes <= pgbase) {
 747                                 nfs_direct_release_pages(data->pagevec, result);
 748                                 nfs_writedata_free(data);
 749                                 break;
 750                         }
 751                         bytes -= pgbase;
 752                         data->npages = result;
 753                 }
 754
 755                 get_dreq(dreq);
 756
 757                 list_move_tail(&data->pages, &dreq->rewrite_list);
 758
 759                 data->req = (struct nfs_page *) dreq;
 760                 data->inode = inode;
 761                 data->cred = msg.rpc_cred;
 762                 data->args.fh = NFS_FH(inode);
 763                 data->args.context = ctx;
 764                 data->args.offset = pos;
 765                 data->args.pgbase = pgbase;
 766                 data->args.pages = data->pagevec;
 767                 data->args.count = bytes;
 768                 data->args.stable = sync;
 769                 data->res.fattr = &data->fattr;
 770                 data->res.count = bytes;
 771                 data->res.verf = &data->verf;
 772                 nfs_fattr_init(&data->fattr);
 773
 774                 task_setup_data.task = &data->task;
 775                 task_setup_data.callback_data = data;
 776                 msg.rpc_argp = &data->args;
 777                 msg.rpc_resp = &data->res;
 778                 NFS_PROTO(inode)->write_setup(data, &msg);
 779
 780                 task = rpc_run_task(&task_setup_data);
 781                 if (IS_ERR(task))
 782                         break;
 783                 rpc_put_task(task);
 784
 785                 dprintk("NFS: %5u initiated direct write call "
 786                         "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
 787                                 data->task.tk_pid,
 788                                 inode->i_sb->s_id,
 789                                 (long long)NFS_FILEID(inode),
 790                                 bytes,
 791                                 (unsigned long long)data->args.offset);
 792
 793                 started += bytes;
 794                 user_addr += bytes;
 795                 pos += bytes;
 796
 797                 /* FIXME: Remove this useless math from the final patch */
 798                 pgbase += bytes;
 799                 pgbase &= ~PAGE_MASK;
 800                 BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 801
 802                 count -= bytes;
 803         } while (count != 0);
 804
 805         if (started)
 806                 return started;
 807         return result < 0 ? (ssize_t) result : -EFAULT;
 808 }
 809
 810 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
 811                                                const struct iovec *iov,
 812                                                unsigned long nr_segs,
 813                                                loff_t pos, int sync)
 814 {
 815         ssize_t result = 0;
 816         size_t requested_bytes = 0;
 817         unsigned long seg;
 818
 819         get_dreq(dreq);
 820
 821         for (seg = 0; seg < nr_segs; seg++) {
 822                 const struct iovec *vec = &iov[seg];
 823                 result = nfs_direct_write_schedule_segment(dreq, vec,
 824                                                            pos, sync);
 825                 if (result < 0)
 826                         break;
 827                 requested_bytes += result;
 828                 if ((size_t)result < vec->iov_len)
 829                         break;
 830                 pos += vec->iov_len;
 831         }
 832
 833         if (put_dreq(dreq))
 834                 nfs_direct_write_complete(dreq, dreq->inode);
 835
 836         if (requested_bytes != 0)
 837                 return 0;
 838
 839         if (result < 0)
 840                 return result;
 841         return -EIO;
 842 }
 843
 844 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
 845                                 unsigned long nr_segs, loff_t pos,
 846                                 size_t count)
 847 {
 848         ssize_t result = 0;
 849         struct inode *inode = iocb->ki_filp->f_mapping->host;
 850         struct nfs_direct_req *dreq;
 851         size_t wsize = NFS_SERVER(inode)->wsize;
 852         int sync = NFS_UNSTABLE;
 853
 854         dreq = nfs_direct_req_alloc();
 855         if (!dreq)
 856                 return -ENOMEM;
 857         nfs_alloc_commit_data(dreq);
 858
 859         if (dreq->commit_data == NULL || count < wsize)
 860                 sync = NFS_FILE_SYNC;
 861
 862         dreq->inode = inode;
 863         dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
 864         if (!is_sync_kiocb(iocb))
 865                 dreq->iocb = iocb;
 866
 867         result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
 868         if (!result)
 869                 result = nfs_direct_wait(dreq);
 870         nfs_direct_req_release(dreq);
 871
 872         return result;
 873 }
 874
 875 /**
 876  * nfs_file_direct_read - file direct read operation for NFS files
 877  * @iocb: target I/O control block
 878  * @iov: vector of user buffers into which to read data
 879  * @nr_segs: size of iov vector
 880  * @pos: byte offset in file where reading starts
 881  *
 882  * We use this function for direct reads instead of calling
 883  * generic_file_aio_read() in order to avoid gfar's check to see if
 884  * the request starts before the end of the file.  For that check
 885  * to work, we must generate a GETATTR before each direct read, and
 886  * even then there is a window between the GETATTR and the subsequent
 887  * READ where the file size could change.  Our preference is simply
 888  * to do all reads the application wants, and the server will take
 889  * care of managing the end of file boundary.
 890  *
 891  * This function also eliminates unnecessarily updating the file's
 892  * atime locally, as the NFS server sets the file's atime, and this
 893  * client must read the updated atime from the server back into its
 894  * cache.
 895  */
 896 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
 897                                 unsigned long nr_segs, loff_t pos)
 898 {
 899         ssize_t retval = -EINVAL;
 900         struct file *file = iocb->ki_filp;
 901         struct address_space *mapping = file->f_mapping;
 902         size_t count;
 903
 904         count = iov_length(iov, nr_segs);
 905         nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
 906
 907         dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n",
 908                 file->f_path.dentry->d_parent->d_name.name,
 909                 file->f_path.dentry->d_name.name,
 910                 count, (long long) pos);
 911
 912         retval = 0;
 913         if (!count)
 914                 goto out;
 915
 916         retval = nfs_sync_mapping(mapping);
 917         if (retval)
 918                 goto out;
 919
 920         retval = nfs_direct_read(iocb, iov, nr_segs, pos);
 921         if (retval > 0)
 922                 iocb->ki_pos = pos + retval;
 923
 924 out:
 925         return retval;
 926 }
 927
 928 /**
 929  * nfs_file_direct_write - file direct write operation for NFS files
 930  * @iocb: target I/O control block
 931  * @iov: vector of user buffers from which to write data
 932  * @nr_segs: size of iov vector
 933  * @pos: byte offset in file where writing starts
 934  *
 935  * We use this function for direct writes instead of calling
 936  * generic_file_aio_write() in order to avoid taking the inode
 937  * semaphore and updating the i_size.  The NFS server will set
 938  * the new i_size and this client must read the updated size
 939  * back into its cache.  We let the server do generic write
 940  * parameter checking and report problems.
 941  *
 942  * We eliminate local atime updates, see direct read above.
 943  *
 944  * We avoid unnecessary page cache invalidations for normal cached
 945  * readers of this file.
 946  *
 947  * Note that O_APPEND is not supported for NFS direct writes, as there
 948  * is no atomic O_APPEND write facility in the NFS protocol.
 949  */
 950 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
 951                                 unsigned long nr_segs, loff_t pos)
 952 {
 953         ssize_t retval = -EINVAL;
 954         struct file *file = iocb->ki_filp;
 955         struct address_space *mapping = file->f_mapping;
 956         size_t count;
 957
 958         count = iov_length(iov, nr_segs);
 959         nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
 960
 961         dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n",
 962                 file->f_path.dentry->d_parent->d_name.name,
 963                 file->f_path.dentry->d_name.name,
 964                 count, (long long) pos);
 965
 966         retval = generic_write_checks(file, &pos, &count, 0);
 967         if (retval)
 968                 goto out;
 969
 970         retval = -EINVAL;
 971         if ((ssize_t) count < 0)
 972                 goto out;
 973         retval = 0;
 974         if (!count)
 975                 goto out;
 976
 977         retval = nfs_sync_mapping(mapping);
 978         if (retval)
 979                 goto out;
 980
 981         retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
 982
 983         if (retval > 0)
 984                 iocb->ki_pos = pos + retval;
 985
 986 out:
 987         return retval;
 988 }
 989
 990 /**
 991  * nfs_init_directcache - create a slab cache for nfs_direct_req structures
 992  *
 993  */
 994 int __init nfs_init_directcache(void)
 995 {
 996         nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
 997                                                 sizeof(struct nfs_direct_req),
 998                                                 0, (SLAB_RECLAIM_ACCOUNT|
 999                                                         SLAB_MEM_SPREAD),
1000                                                 NULL);
1001         if (nfs_direct_cachep == NULL)
1002                 return -ENOMEM;
1003
1004         return 0;
1005 }
1006
1007 /**
1008  * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
1009  *
1010  */
1011 void nfs_destroy_directcache(void)
1012 {
1013         kmem_cache_destroy(nfs_direct_cachep);
1014 }