fs/nfs/file.c

   1 /*
   2  *  linux/fs/nfs/file.c
   3  *
   4  *  Copyright (C) 1992  Rick Sladkey
   5  *
   6  *  Changes Copyright (C) 1994 by Florian La Roche
   7  *   - Do not copy data too often around in the kernel.
   8  *   - In nfs_file_read the return value of kmalloc wasn't checked.
   9  *   - Put in a better version of read look-ahead buffering. Original idea
  10  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
  11  *
  12  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
  13  *
  14  *  Total rewrite of read side for new NFS buffer cache.. Linus.
  15  *
  16  *  nfs regular file handling functions
  17  */
  18
  19 #include <linux/time.h>
  20 #include <linux/kernel.h>
  21 #include <linux/errno.h>
  22 #include <linux/fcntl.h>
  23 #include <linux/stat.h>
  24 #include <linux/nfs_fs.h>
  25 #include <linux/nfs_mount.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/aio.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/system.h>
  33
  34 #include "delegation.h"
  35 #include "internal.h"
  36 #include "iostat.h"
  37 #include "fscache.h"
  38
  39 #define NFSDBG_FACILITY         NFSDBG_FILE
  40
  41 static int nfs_file_open(struct inode *, struct file *);
  42 static int nfs_file_release(struct inode *, struct file *);
  43 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  44 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  45 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
  46                                         struct pipe_inode_info *pipe,
  47                                         size_t count, unsigned int flags);
  48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
  49                                 unsigned long nr_segs, loff_t pos);
  50 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
  51                                         struct file *filp, loff_t *ppos,
  52                                         size_t count, unsigned int flags);
  53 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
  54                                 unsigned long nr_segs, loff_t pos);
  55 static int  nfs_file_flush(struct file *, fl_owner_t id);
  56 static int  nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
  57 static int nfs_check_flags(int flags);
  58 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  59 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  60 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
  61
  62 static const struct vm_operations_struct nfs_file_vm_ops;
  63
  64 const struct file_operations nfs_file_operations = {
  65         .llseek         = nfs_file_llseek,
  66         .read           = do_sync_read,
  67         .write          = do_sync_write,
  68         .aio_read       = nfs_file_read,
  69         .aio_write      = nfs_file_write,
  70         .mmap           = nfs_file_mmap,
  71         .open           = nfs_file_open,
  72         .flush          = nfs_file_flush,
  73         .release        = nfs_file_release,
  74         .fsync          = nfs_file_fsync,
  75         .lock           = nfs_lock,
  76         .flock          = nfs_flock,
  77         .splice_read    = nfs_file_splice_read,
  78         .splice_write   = nfs_file_splice_write,
  79         .check_flags    = nfs_check_flags,
  80         .setlease       = nfs_setlease,
  81 };
  82
  83 const struct inode_operations nfs_file_inode_operations = {
  84         .permission     = nfs_permission,
  85         .getattr        = nfs_getattr,
  86         .setattr        = nfs_setattr,
  87 };
  88
  89 #ifdef CONFIG_NFS_V3
  90 const struct inode_operations nfs3_file_inode_operations = {
  91         .permission     = nfs_permission,
  92         .getattr        = nfs_getattr,
  93         .setattr        = nfs_setattr,
  94         .listxattr      = nfs3_listxattr,
  95         .getxattr       = nfs3_getxattr,
  96         .setxattr       = nfs3_setxattr,
  97         .removexattr    = nfs3_removexattr,
  98 };
  99 #endif  /* CONFIG_NFS_v3 */
 100
 101 /* Hack for future NFS swap support */
 102 #ifndef IS_SWAPFILE
 103 # define IS_SWAPFILE(inode)     (0)
 104 #endif
 105
 106 static int nfs_check_flags(int flags)
 107 {
 108         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 109                 return -EINVAL;
 110
 111         return 0;
 112 }
 113
 114 /*
 115  * Open file
 116  */
 117 static int
 118 nfs_file_open(struct inode *inode, struct file *filp)
 119 {
 120         int res;
 121
 122         dprintk("NFS: open file(%s/%s)\n",
 123                         filp->f_path.dentry->d_parent->d_name.name,
 124                         filp->f_path.dentry->d_name.name);
 125
 126         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 127         res = nfs_check_flags(filp->f_flags);
 128         if (res)
 129                 return res;
 130
 131         res = nfs_open(inode, filp);
 132         return res;
 133 }
 134
 135 static int
 136 nfs_file_release(struct inode *inode, struct file *filp)
 137 {
 138         struct dentry *dentry = filp->f_path.dentry;
 139
 140         dprintk("NFS: release(%s/%s)\n",
 141                         dentry->d_parent->d_name.name,
 142                         dentry->d_name.name);
 143
 144         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 145         return nfs_release(inode, filp);
 146 }
 147
 148 /**
 149  * nfs_revalidate_size - Revalidate the file size
 150  * @inode - pointer to inode struct
 151  * @file - pointer to struct file
 152  *
 153  * Revalidates the file length. This is basically a wrapper around
 154  * nfs_revalidate_inode() that takes into account the fact that we may
 155  * have cached writes (in which case we don't care about the server's
 156  * idea of what the file length is), or O_DIRECT (in which case we
 157  * shouldn't trust the cache).
 158  */
 159 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 160 {
 161         struct nfs_server *server = NFS_SERVER(inode);
 162         struct nfs_inode *nfsi = NFS_I(inode);
 163
 164         if (server->flags & NFS_MOUNT_NOAC)
 165                 goto force_reval;
 166         if (filp->f_flags & O_DIRECT)
 167                 goto force_reval;
 168         if (nfsi->npages != 0)
 169                 return 0;
 170         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 171                 return 0;
 172 force_reval:
 173         return __nfs_revalidate_inode(server, inode);
 174 }
 175
 176 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 177 {
 178         loff_t loff;
 179
 180         dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
 181                         filp->f_path.dentry->d_parent->d_name.name,
 182                         filp->f_path.dentry->d_name.name,
 183                         offset, origin);
 184
 185         /* origin == SEEK_END => we must revalidate the cached file length */
 186         if (origin == SEEK_END) {
 187                 struct inode *inode = filp->f_mapping->host;
 188
 189                 int retval = nfs_revalidate_file_size(inode, filp);
 190                 if (retval < 0)
 191                         return (loff_t)retval;
 192
 193                 spin_lock(&inode->i_lock);
 194                 loff = generic_file_llseek_unlocked(filp, offset, origin);
 195                 spin_unlock(&inode->i_lock);
 196         } else
 197                 loff = generic_file_llseek_unlocked(filp, offset, origin);
 198         return loff;
 199 }
 200
 201 /*
 202  * Helper for nfs_file_flush() and nfs_file_fsync()
 203  *
 204  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
 205  * disk, but it retrieves and clears ctx->error after synching, despite
 206  * the two being set at the same time in nfs_context_set_write_error().
 207  * This is because the former is used to notify the _next_ call to
 208  * nfs_file_write() that a write error occured, and hence cause it to
 209  * fall back to doing a synchronous write.
 210  */
 211 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
 212 {
 213         int have_error, status;
 214         int ret = 0;
 215
 216         have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
 217         status = nfs_wb_all(inode);
 218         have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
 219         if (have_error)
 220                 ret = xchg(&ctx->error, 0);
 221         if (!ret)
 222                 ret = status;
 223         return ret;
 224 }
 225
 226 /*
 227  * Flush all dirty pages, and check for write errors.
 228  */
 229 static int
 230 nfs_file_flush(struct file *file, fl_owner_t id)
 231 {
 232         struct nfs_open_context *ctx = nfs_file_open_context(file);
 233         struct dentry   *dentry = file->f_path.dentry;
 234         struct inode    *inode = dentry->d_inode;
 235
 236         dprintk("NFS: flush(%s/%s)\n",
 237                         dentry->d_parent->d_name.name,
 238                         dentry->d_name.name);
 239
 240         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 241         if ((file->f_mode & FMODE_WRITE) == 0)
 242                 return 0;
 243
 244         /* Flush writes to the server and return any errors */
 245         return nfs_do_fsync(ctx, inode);
 246 }
 247
 248 static ssize_t
 249 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
 250                 unsigned long nr_segs, loff_t pos)
 251 {
 252         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 253         struct inode * inode = dentry->d_inode;
 254         ssize_t result;
 255         size_t count = iov_length(iov, nr_segs);
 256
 257         if (iocb->ki_filp->f_flags & O_DIRECT)
 258                 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
 259
 260         dprintk("NFS: read(%s/%s, %lu@%lu)\n",
 261                 dentry->d_parent->d_name.name, dentry->d_name.name,
 262                 (unsigned long) count, (unsigned long) pos);
 263
 264         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 265         if (!result) {
 266                 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
 267                 if (result > 0)
 268                         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
 269         }
 270         return result;
 271 }
 272
 273 static ssize_t
 274 nfs_file_splice_read(struct file *filp, loff_t *ppos,
 275                      struct pipe_inode_info *pipe, size_t count,
 276                      unsigned int flags)
 277 {
 278         struct dentry *dentry = filp->f_path.dentry;
 279         struct inode *inode = dentry->d_inode;
 280         ssize_t res;
 281
 282         dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
 283                 dentry->d_parent->d_name.name, dentry->d_name.name,
 284                 (unsigned long) count, (unsigned long long) *ppos);
 285
 286         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 287         if (!res) {
 288                 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
 289                 if (res > 0)
 290                         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
 291         }
 292         return res;
 293 }
 294
 295 static int
 296 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 297 {
 298         struct dentry *dentry = file->f_path.dentry;
 299         struct inode *inode = dentry->d_inode;
 300         int     status;
 301
 302         dprintk("NFS: mmap(%s/%s)\n",
 303                 dentry->d_parent->d_name.name, dentry->d_name.name);
 304
 305         /* Note: generic_file_mmap() returns ENOSYS on nommu systems
 306          *       so we call that before revalidating the mapping
 307          */
 308         status = generic_file_mmap(file, vma);
 309         if (!status) {
 310                 vma->vm_ops = &nfs_file_vm_ops;
 311                 status = nfs_revalidate_mapping(inode, file->f_mapping);
 312         }
 313         return status;
 314 }
 315
 316 /*
 317  * Flush any dirty pages for this process, and check for write errors.
 318  * The return status from this call provides a reliable indication of
 319  * whether any write errors occurred for this process.
 320  */
 321 static int
 322 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
 323 {
 324         struct nfs_open_context *ctx = nfs_file_open_context(file);
 325         struct inode *inode = dentry->d_inode;
 326
 327         dprintk("NFS: fsync file(%s/%s) datasync %d\n",
 328                         dentry->d_parent->d_name.name, dentry->d_name.name,
 329                         datasync);
 330
 331         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 332         return nfs_do_fsync(ctx, inode);
 333 }
 334
 335 /*
 336  * Decide whether a read/modify/write cycle may be more efficient
 337  * then a modify/write/read cycle when writing to a page in the
 338  * page cache.
 339  *
 340  * The modify/write/read cycle may occur if a page is read before
 341  * being completely filled by the writer.  In this situation, the
 342  * page must be completely written to stable storage on the server
 343  * before it can be refilled by reading in the page from the server.
 344  * This can lead to expensive, small, FILE_SYNC mode writes being
 345  * done.
 346  *
 347  * It may be more efficient to read the page first if the file is
 348  * open for reading in addition to writing, the page is not marked
 349  * as Uptodate, it is not dirty or waiting to be committed,
 350  * indicating that it was previously allocated and then modified,
 351  * that there were valid bytes of data in that range of the file,
 352  * and that the new data won't completely replace the old data in
 353  * that range of the file.
 354  */
 355 static int nfs_want_read_modify_write(struct file *file, struct page *page,
 356                         loff_t pos, unsigned len)
 357 {
 358         unsigned int pglen = nfs_page_length(page);
 359         unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
 360         unsigned int end = offset + len;
 361
 362         if ((file->f_mode & FMODE_READ) &&      /* open for read? */
 363             !PageUptodate(page) &&              /* Uptodate? */
 364             !PagePrivate(page) &&               /* i/o request already? */
 365             pglen &&                            /* valid bytes of file? */
 366             (end < pglen || offset))            /* replace all valid bytes? */
 367                 return 1;
 368         return 0;
 369 }
 370
 371 /*
 372  * This does the "real" work of the write. We must allocate and lock the
 373  * page to be sent back to the generic routine, which then copies the
 374  * data from user space.
 375  *
 376  * If the writer ends up delaying the write, the writer needs to
 377  * increment the page use counts until he is done with the page.
 378  */
 379 static int nfs_write_begin(struct file *file, struct address_space *mapping,
 380                         loff_t pos, unsigned len, unsigned flags,
 381                         struct page **pagep, void **fsdata)
 382 {
 383         int ret;
 384         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 385         struct page *page;
 386         int once_thru = 0;
 387
 388         dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
 389                 file->f_path.dentry->d_parent->d_name.name,
 390                 file->f_path.dentry->d_name.name,
 391                 mapping->host->i_ino, len, (long long) pos);
 392
 393 start:
 394         /*
 395          * Prevent starvation issues if someone is doing a consistency
 396          * sync-to-disk
 397          */
 398         ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
 399                         nfs_wait_bit_killable, TASK_KILLABLE);
 400         if (ret)
 401                 return ret;
 402
 403         page = grab_cache_page_write_begin(mapping, index, flags);
 404         if (!page)
 405                 return -ENOMEM;
 406         *pagep = page;
 407
 408         ret = nfs_flush_incompatible(file, page);
 409         if (ret) {
 410                 unlock_page(page);
 411                 page_cache_release(page);
 412         } else if (!once_thru &&
 413                    nfs_want_read_modify_write(file, page, pos, len)) {
 414                 once_thru = 1;
 415                 ret = nfs_readpage(file, page);
 416                 page_cache_release(page);
 417                 if (!ret)
 418                         goto start;
 419         }
 420         return ret;
 421 }
 422
 423 static int nfs_write_end(struct file *file, struct address_space *mapping,
 424                         loff_t pos, unsigned len, unsigned copied,
 425                         struct page *page, void *fsdata)
 426 {
 427         unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
 428         int status;
 429
 430         dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
 431                 file->f_path.dentry->d_parent->d_name.name,
 432                 file->f_path.dentry->d_name.name,
 433                 mapping->host->i_ino, len, (long long) pos);
 434
 435         /*
 436          * Zero any uninitialised parts of the page, and then mark the page
 437          * as up to date if it turns out that we're extending the file.
 438          */
 439         if (!PageUptodate(page)) {
 440                 unsigned pglen = nfs_page_length(page);
 441                 unsigned end = offset + len;
 442
 443                 if (pglen == 0) {
 444                         zero_user_segments(page, 0, offset,
 445                                         end, PAGE_CACHE_SIZE);
 446                         SetPageUptodate(page);
 447                 } else if (end >= pglen) {
 448                         zero_user_segment(page, end, PAGE_CACHE_SIZE);
 449                         if (offset == 0)
 450                                 SetPageUptodate(page);
 451                 } else
 452                         zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
 453         }
 454
 455         status = nfs_updatepage(file, page, offset, copied);
 456
 457         unlock_page(page);
 458         page_cache_release(page);
 459
 460         if (status < 0)
 461                 return status;
 462         return copied;
 463 }
 464
 465 /*
 466  * Partially or wholly invalidate a page
 467  * - Release the private state associated with a page if undergoing complete
 468  *   page invalidation
 469  * - Called if either PG_private or PG_fscache is set on the page
 470  * - Caller holds page lock
 471  */
 472 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 473 {
 474         dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
 475
 476         if (offset != 0)
 477                 return;
 478         /* Cancel any unstarted writes on this page */
 479         nfs_wb_page_cancel(page->mapping->host, page);
 480
 481         nfs_fscache_invalidate_page(page, page->mapping->host);
 482 }
 483
 484 /*
 485  * Attempt to release the private state associated with a page
 486  * - Called if either PG_private or PG_fscache is set on the page
 487  * - Caller holds page lock
 488  * - Return true (may release page) or false (may not)
 489  */
 490 static int nfs_release_page(struct page *page, gfp_t gfp)
 491 {
 492         dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
 493
 494         if (gfp & __GFP_WAIT)
 495                 nfs_wb_page(page->mapping->host, page);
 496         /* If PagePrivate() is set, then the page is not freeable */
 497         if (PagePrivate(page))
 498                 return 0;
 499         return nfs_fscache_release_page(page, gfp);
 500 }
 501
 502 /*
 503  * Attempt to clear the private state associated with a page when an error
 504  * occurs that requires the cached contents of an inode to be written back or
 505  * destroyed
 506  * - Called if either PG_private or fscache is set on the page
 507  * - Caller holds page lock
 508  * - Return 0 if successful, -error otherwise
 509  */
 510 static int nfs_launder_page(struct page *page)
 511 {
 512         struct inode *inode = page->mapping->host;
 513         struct nfs_inode *nfsi = NFS_I(inode);
 514
 515         dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
 516                 inode->i_ino, (long long)page_offset(page));
 517
 518         nfs_fscache_wait_on_page_write(nfsi, page);
 519         return nfs_wb_page(inode, page);
 520 }
 521
 522 const struct address_space_operations nfs_file_aops = {
 523         .readpage = nfs_readpage,
 524         .readpages = nfs_readpages,
 525         .set_page_dirty = __set_page_dirty_nobuffers,
 526         .writepage = nfs_writepage,
 527         .writepages = nfs_writepages,
 528         .write_begin = nfs_write_begin,
 529         .write_end = nfs_write_end,
 530         .invalidatepage = nfs_invalidate_page,
 531         .releasepage = nfs_release_page,
 532         .direct_IO = nfs_direct_IO,
 533         .migratepage = nfs_migrate_page,
 534         .launder_page = nfs_launder_page,
 535         .error_remove_page = generic_error_remove_page,
 536 };
 537
 538 /*
 539  * Notification that a PTE pointing to an NFS page is about to be made
 540  * writable, implying that someone is about to modify the page through a
 541  * shared-writable mapping
 542  */
 543 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 544 {
 545         struct page *page = vmf->page;
 546         struct file *filp = vma->vm_file;
 547         struct dentry *dentry = filp->f_path.dentry;
 548         unsigned pagelen;
 549         int ret = -EINVAL;
 550         struct address_space *mapping;
 551
 552         dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
 553                 dentry->d_parent->d_name.name, dentry->d_name.name,
 554                 filp->f_mapping->host->i_ino,
 555                 (long long)page_offset(page));
 556
 557         /* make sure the cache has finished storing the page */
 558         nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
 559
 560         lock_page(page);
 561         mapping = page->mapping;
 562         if (mapping != dentry->d_inode->i_mapping)
 563                 goto out_unlock;
 564
 565         ret = 0;
 566         pagelen = nfs_page_length(page);
 567         if (pagelen == 0)
 568                 goto out_unlock;
 569
 570         ret = nfs_flush_incompatible(filp, page);
 571         if (ret != 0)
 572                 goto out_unlock;
 573
 574         ret = nfs_updatepage(filp, page, 0, pagelen);
 575 out_unlock:
 576         if (!ret)
 577                 return VM_FAULT_LOCKED;
 578         unlock_page(page);
 579         return VM_FAULT_SIGBUS;
 580 }
 581
 582 static const struct vm_operations_struct nfs_file_vm_ops = {
 583         .fault = filemap_fault,
 584         .page_mkwrite = nfs_vm_page_mkwrite,
 585 };
 586
 587 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
 588 {
 589         struct nfs_open_context *ctx;
 590
 591         if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
 592                 return 1;
 593         ctx = nfs_file_open_context(filp);
 594         if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
 595                 return 1;
 596         return 0;
 597 }
 598
 599 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
 600                                 unsigned long nr_segs, loff_t pos)
 601 {
 602         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 603         struct inode * inode = dentry->d_inode;
 604         unsigned long written = 0;
 605         ssize_t result;
 606         size_t count = iov_length(iov, nr_segs);
 607
 608         if (iocb->ki_filp->f_flags & O_DIRECT)
 609                 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
 610
 611         dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
 612                 dentry->d_parent->d_name.name, dentry->d_name.name,
 613                 (unsigned long) count, (long long) pos);
 614
 615         result = -EBUSY;
 616         if (IS_SWAPFILE(inode))
 617                 goto out_swapfile;
 618         /*
 619          * O_APPEND implies that we must revalidate the file length.
 620          */
 621         if (iocb->ki_filp->f_flags & O_APPEND) {
 622                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 623                 if (result)
 624                         goto out;
 625         }
 626
 627         result = count;
 628         if (!count)
 629                 goto out;
 630
 631         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 632         if (result > 0)
 633                 written = result;
 634
 635         /* Return error values for O_DSYNC and IS_SYNC() */
 636         if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
 637                 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
 638                 if (err < 0)
 639                         result = err;
 640         }
 641         if (result > 0)
 642                 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
 643 out:
 644         return result;
 645
 646 out_swapfile:
 647         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 648         goto out;
 649 }
 650
 651 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
 652                                      struct file *filp, loff_t *ppos,
 653                                      size_t count, unsigned int flags)
 654 {
 655         struct dentry *dentry = filp->f_path.dentry;
 656         struct inode *inode = dentry->d_inode;
 657         unsigned long written = 0;
 658         ssize_t ret;
 659
 660         dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
 661                 dentry->d_parent->d_name.name, dentry->d_name.name,
 662                 (unsigned long) count, (unsigned long long) *ppos);
 663
 664         /*
 665          * The combination of splice and an O_APPEND destination is disallowed.
 666          */
 667
 668         ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
 669         if (ret > 0)
 670                 written = ret;
 671
 672         if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
 673                 int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
 674                 if (err < 0)
 675                         ret = err;
 676         }
 677         if (ret > 0)
 678                 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
 679         return ret;
 680 }
 681
 682 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 683 {
 684         struct inode *inode = filp->f_mapping->host;
 685         int status = 0;
 686
 687         /* Try local locking first */
 688         posix_test_lock(filp, fl);
 689         if (fl->fl_type != F_UNLCK) {
 690                 /* found a conflict */
 691                 goto out;
 692         }
 693
 694         if (nfs_have_delegation(inode, FMODE_READ))
 695                 goto out_noconflict;
 696
 697         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 698                 goto out_noconflict;
 699
 700         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 701 out:
 702         return status;
 703 out_noconflict:
 704         fl->fl_type = F_UNLCK;
 705         goto out;
 706 }
 707
 708 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 709 {
 710         int res = 0;
 711         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 712                 case FL_POSIX:
 713                         res = posix_lock_file_wait(file, fl);
 714                         break;
 715                 case FL_FLOCK:
 716                         res = flock_lock_file_wait(file, fl);
 717                         break;
 718                 default:
 719                         BUG();
 720         }
 721         if (res < 0)
 722                 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
 723                         " - error %d!\n",
 724                                 __func__, res);
 725         return res;
 726 }
 727
 728 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 729 {
 730         struct inode *inode = filp->f_mapping->host;
 731         int status;
 732
 733         /*
 734          * Flush all pending writes before doing anything
 735          * with locks..
 736          */
 737         nfs_sync_mapping(filp->f_mapping);
 738
 739         /* NOTE: special case
 740          *      If we're signalled while cleaning up locks on process exit, we
 741          *      still need to complete the unlock.
 742          */
 743         /* Use local locking if mounted with "-onolock" */
 744         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 745                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 746         else
 747                 status = do_vfs_lock(filp, fl);
 748         return status;
 749 }
 750
 751 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 752 {
 753         struct inode *inode = filp->f_mapping->host;
 754         int status;
 755
 756         /*
 757          * Flush all pending writes before doing anything
 758          * with locks..
 759          */
 760         status = nfs_sync_mapping(filp->f_mapping);
 761         if (status != 0)
 762                 goto out;
 763
 764         /* Use local locking if mounted with "-onolock" */
 765         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 766                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 767         else
 768                 status = do_vfs_lock(filp, fl);
 769         if (status < 0)
 770                 goto out;
 771         /*
 772          * Make sure we clear the cache whenever we try to get the lock.
 773          * This makes locking act as a cache coherency point.
 774          */
 775         nfs_sync_mapping(filp->f_mapping);
 776         if (!nfs_have_delegation(inode, FMODE_READ))
 777                 nfs_zap_caches(inode);
 778 out:
 779         return status;
 780 }
 781
 782 /*
 783  * Lock a (portion of) a file
 784  */
 785 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 786 {
 787         struct inode *inode = filp->f_mapping->host;
 788         int ret = -ENOLCK;
 789
 790         dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
 791                         filp->f_path.dentry->d_parent->d_name.name,
 792                         filp->f_path.dentry->d_name.name,
 793                         fl->fl_type, fl->fl_flags,
 794                         (long long)fl->fl_start, (long long)fl->fl_end);
 795
 796         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 797
 798         /* No mandatory locks over NFS */
 799         if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
 800                 goto out_err;
 801
 802         if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
 803                 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
 804                 if (ret < 0)
 805                         goto out_err;
 806         }
 807
 808         if (IS_GETLK(cmd))
 809                 ret = do_getlk(filp, cmd, fl);
 810         else if (fl->fl_type == F_UNLCK)
 811                 ret = do_unlk(filp, cmd, fl);
 812         else
 813                 ret = do_setlk(filp, cmd, fl);
 814 out_err:
 815         return ret;
 816 }
 817
 818 /*
 819  * Lock a (portion of) a file
 820  */
 821 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 822 {
 823         dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
 824                         filp->f_path.dentry->d_parent->d_name.name,
 825                         filp->f_path.dentry->d_name.name,
 826                         fl->fl_type, fl->fl_flags);
 827
 828         if (!(fl->fl_flags & FL_FLOCK))
 829                 return -ENOLCK;
 830
 831         /* We're simulating flock() locks using posix locks on the server */
 832         fl->fl_owner = (fl_owner_t)filp;
 833         fl->fl_start = 0;
 834         fl->fl_end = OFFSET_MAX;
 835
 836         if (fl->fl_type == F_UNLCK)
 837                 return do_unlk(filp, cmd, fl);
 838         return do_setlk(filp, cmd, fl);
 839 }
 840
 841 /*
 842  * There is no protocol support for leases, so we have no way to implement
 843  * them correctly in the face of opens by other clients.
 844  */
 845 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
 846 {
 847         dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
 848                         file->f_path.dentry->d_parent->d_name.name,
 849                         file->f_path.dentry->d_name.name, arg);
 850
 851         return -EINVAL;
 852 }