release/src-rt-6.x/linux/linux-2.6/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/smp_lock.h>
  30 #include <linux/aio.h>
  31
  32 #include <asm/uaccess.h>
  33 #include <asm/system.h>
  34
  35 #include "delegation.h"
  36 #include "iostat.h"
  37
  38 #define NFSDBG_FACILITY         NFSDBG_FILE
  39
  40 static int nfs_file_open(struct inode *, struct file *);
  41 static int nfs_file_release(struct inode *, struct file *);
  42 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  43 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  44 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
  45                                         struct pipe_inode_info *pipe,
  46                                         size_t count, unsigned int flags);
  47 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
  48                                 unsigned long nr_segs, loff_t pos);
  49 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
  50                                 unsigned long nr_segs, loff_t pos);
  51 static int  nfs_file_flush(struct file *, fl_owner_t id);
  52 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  53 static int nfs_check_flags(int flags);
  54 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  55 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  56
  57 const struct file_operations nfs_file_operations = {
  58         .llseek         = nfs_file_llseek,
  59         .read           = do_sync_read,
  60         .write          = do_sync_write,
  61         .aio_read       = nfs_file_read,
  62         .aio_write      = nfs_file_write,
  63         .mmap           = nfs_file_mmap,
  64         .open           = nfs_file_open,
  65         .flush          = nfs_file_flush,
  66         .release        = nfs_file_release,
  67         .fsync          = nfs_fsync,
  68         .lock           = nfs_lock,
  69         .flock          = nfs_flock,
  70         .splice_read    = nfs_file_splice_read,
  71         .check_flags    = nfs_check_flags,
  72 };
  73
  74 const struct inode_operations nfs_file_inode_operations = {
  75         .permission     = nfs_permission,
  76         .getattr        = nfs_getattr,
  77         .setattr        = nfs_setattr,
  78 };
  79
  80 #ifdef CONFIG_NFS_V3
  81 const struct inode_operations nfs3_file_inode_operations = {
  82         .permission     = nfs_permission,
  83         .getattr        = nfs_getattr,
  84         .setattr        = nfs_setattr,
  85         .listxattr      = nfs3_listxattr,
  86         .getxattr       = nfs3_getxattr,
  87         .setxattr       = nfs3_setxattr,
  88         .removexattr    = nfs3_removexattr,
  89 };
  90 #endif  /* CONFIG_NFS_v3 */
  91
  92 /* Hack for future NFS swap support */
  93 #ifndef IS_SWAPFILE
  94 # define IS_SWAPFILE(inode)     (0)
  95 #endif
  96
  97 static int nfs_check_flags(int flags)
  98 {
  99         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 100                 return -EINVAL;
 101
 102         return 0;
 103 }
 104
 105 /*
 106  * Open file
 107  */
 108 static int
 109 nfs_file_open(struct inode *inode, struct file *filp)
 110 {
 111         int res;
 112
 113         res = nfs_check_flags(filp->f_flags);
 114         if (res)
 115                 return res;
 116
 117         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 118         lock_kernel();
 119         res = NFS_PROTO(inode)->file_open(inode, filp);
 120         unlock_kernel();
 121         return res;
 122 }
 123
 124 static int
 125 nfs_file_release(struct inode *inode, struct file *filp)
 126 {
 127         /* Ensure that dirty pages are flushed out with the right creds */
 128         if (filp->f_mode & FMODE_WRITE)
 129                 filemap_fdatawrite(filp->f_mapping);
 130         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 131         return NFS_PROTO(inode)->file_release(inode, filp);
 132 }
 133
 134 /**
 135  * nfs_revalidate_size - Revalidate the file size
 136  * @inode - pointer to inode struct
 137  * @file - pointer to struct file
 138  *
 139  * Revalidates the file length. This is basically a wrapper around
 140  * nfs_revalidate_inode() that takes into account the fact that we may
 141  * have cached writes (in which case we don't care about the server's
 142  * idea of what the file length is), or O_DIRECT (in which case we
 143  * shouldn't trust the cache).
 144  */
 145 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 146 {
 147         struct nfs_server *server = NFS_SERVER(inode);
 148         struct nfs_inode *nfsi = NFS_I(inode);
 149
 150         if (server->flags & NFS_MOUNT_NOAC)
 151                 goto force_reval;
 152         if (filp->f_flags & O_DIRECT)
 153                 goto force_reval;
 154         if (nfsi->npages != 0)
 155                 return 0;
 156         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 157                 return 0;
 158 force_reval:
 159         return __nfs_revalidate_inode(server, inode);
 160 }
 161
 162 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 163 {
 164         /* origin == SEEK_END => we must revalidate the cached file length */
 165         if (origin == SEEK_END) {
 166                 struct inode *inode = filp->f_mapping->host;
 167                 int retval = nfs_revalidate_file_size(inode, filp);
 168                 if (retval < 0)
 169                         return (loff_t)retval;
 170         }
 171         return remote_llseek(filp, offset, origin);
 172 }
 173
 174 /*
 175  * Flush all dirty pages, and check for write errors.
 176  *
 177  */
 178 static int
 179 nfs_file_flush(struct file *file, fl_owner_t id)
 180 {
 181         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 182         struct inode    *inode = file->f_path.dentry->d_inode;
 183         int             status;
 184
 185         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 186
 187         if ((file->f_mode & FMODE_WRITE) == 0)
 188                 return 0;
 189         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 190         lock_kernel();
 191         /* Ensure that data+attribute caches are up to date after close() */
 192         status = nfs_wb_all(inode);
 193         if (!status) {
 194                 status = ctx->error;
 195                 ctx->error = 0;
 196                 if (!status)
 197                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 198         }
 199         unlock_kernel();
 200         return status;
 201 }
 202
 203 static ssize_t
 204 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
 205                 unsigned long nr_segs, loff_t pos)
 206 {
 207         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 208         struct inode * inode = dentry->d_inode;
 209         ssize_t result;
 210         size_t count = iov_length(iov, nr_segs);
 211
 212 #ifdef CONFIG_NFS_DIRECTIO
 213         if (iocb->ki_filp->f_flags & O_DIRECT)
 214                 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
 215 #endif
 216
 217         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 218                 dentry->d_parent->d_name.name, dentry->d_name.name,
 219                 (unsigned long) count, (unsigned long) pos);
 220
 221         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 222         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 223         if (!result)
 224                 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
 225         return result;
 226 }
 227
 228 static ssize_t
 229 nfs_file_splice_read(struct file *filp, loff_t *ppos,
 230                      struct pipe_inode_info *pipe, size_t count,
 231                      unsigned int flags)
 232 {
 233         struct dentry *dentry = filp->f_path.dentry;
 234         struct inode *inode = dentry->d_inode;
 235         ssize_t res;
 236
 237         dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
 238                 dentry->d_parent->d_name.name, dentry->d_name.name,
 239                 (unsigned long) count, (unsigned long long) *ppos);
 240
 241         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 242         if (!res)
 243                 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
 244         return res;
 245 }
 246
 247 static int
 248 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 249 {
 250         struct dentry *dentry = file->f_path.dentry;
 251         struct inode *inode = dentry->d_inode;
 252         int     status;
 253
 254         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 255                 dentry->d_parent->d_name.name, dentry->d_name.name);
 256
 257         status = nfs_revalidate_mapping(inode, file->f_mapping);
 258         if (!status)
 259                 status = generic_file_mmap(file, vma);
 260         return status;
 261 }
 262
 263 /*
 264  * Flush any dirty pages for this process, and check for write errors.
 265  * The return status from this call provides a reliable indication of
 266  * whether any write errors occurred for this process.
 267  */
 268 static int
 269 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 270 {
 271         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 272         struct inode *inode = dentry->d_inode;
 273         int status;
 274
 275         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 276
 277         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 278         lock_kernel();
 279         status = nfs_wb_all(inode);
 280         if (!status) {
 281                 status = ctx->error;
 282                 ctx->error = 0;
 283         }
 284         unlock_kernel();
 285         return status;
 286 }
 287
 288 /*
 289  * This does the "real" work of the write. The generic routine has
 290  * allocated the page, locked it, done all the page alignment stuff
 291  * calculations etc. Now we should just copy the data from user
 292  * space and write it back to the real medium..
 293  *
 294  * If the writer ends up delaying the write, the writer needs to
 295  * increment the page use counts until he is done with the page.
 296  */
 297 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 298 {
 299         return nfs_flush_incompatible(file, page);
 300 }
 301
 302 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 303 {
 304         long status;
 305
 306         lock_kernel();
 307         status = nfs_updatepage(file, page, offset, to-offset);
 308         unlock_kernel();
 309         return status;
 310 }
 311
 312 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 313 {
 314         if (offset != 0)
 315                 return;
 316         /* Cancel any unstarted writes on this page */
 317         nfs_wb_page_priority(page->mapping->host, page, FLUSH_INVALIDATE);
 318 }
 319
 320 static int nfs_release_page(struct page *page, gfp_t gfp)
 321 {
 322         /* If PagePrivate() is set, then the page is not freeable */
 323         return 0;
 324 }
 325
 326 static int nfs_launder_page(struct page *page)
 327 {
 328         return nfs_wb_page(page->mapping->host, page);
 329 }
 330
 331 const struct address_space_operations nfs_file_aops = {
 332         .readpage = nfs_readpage,
 333         .readpages = nfs_readpages,
 334         .set_page_dirty = nfs_set_page_dirty,
 335         .writepage = nfs_writepage,
 336         .writepages = nfs_writepages,
 337         .prepare_write = nfs_prepare_write,
 338         .commit_write = nfs_commit_write,
 339         .invalidatepage = nfs_invalidate_page,
 340         .releasepage = nfs_release_page,
 341 #ifdef CONFIG_NFS_DIRECTIO
 342         .direct_IO = nfs_direct_IO,
 343 #endif
 344         .launder_page = nfs_launder_page,
 345 };
 346
 347 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
 348                                 unsigned long nr_segs, loff_t pos)
 349 {
 350         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 351         struct inode * inode = dentry->d_inode;
 352         ssize_t result;
 353         size_t count = iov_length(iov, nr_segs);
 354
 355 #ifdef CONFIG_NFS_DIRECTIO
 356         if (iocb->ki_filp->f_flags & O_DIRECT)
 357                 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
 358 #endif
 359
 360         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
 361                 dentry->d_parent->d_name.name, dentry->d_name.name,
 362                 inode->i_ino, (unsigned long) count, (long long) pos);
 363
 364         result = -EBUSY;
 365         if (IS_SWAPFILE(inode))
 366                 goto out_swapfile;
 367         /*
 368          * O_APPEND implies that we must revalidate the file length.
 369          */
 370         if (iocb->ki_filp->f_flags & O_APPEND) {
 371                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 372                 if (result)
 373                         goto out;
 374         }
 375
 376         result = count;
 377         if (!count)
 378                 goto out;
 379
 380         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 381         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 382         /* Return error values for O_SYNC and IS_SYNC() */
 383         if (result >= 0 && (IS_SYNC(inode) || (iocb->ki_filp->f_flags & O_SYNC))) {
 384                 int err = nfs_fsync(iocb->ki_filp, dentry, 1);
 385                 if (err < 0)
 386                         result = err;
 387         }
 388 out:
 389         return result;
 390
 391 out_swapfile:
 392         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 393         goto out;
 394 }
 395
 396 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 397 {
 398         struct inode *inode = filp->f_mapping->host;
 399         int status = 0;
 400
 401         lock_kernel();
 402         /* Try local locking first */
 403         if (posix_test_lock(filp, fl)) {
 404                 goto out;
 405         }
 406
 407         if (nfs_have_delegation(inode, FMODE_READ))
 408                 goto out_noconflict;
 409
 410         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 411                 goto out_noconflict;
 412
 413         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 414 out:
 415         unlock_kernel();
 416         return status;
 417 out_noconflict:
 418         fl->fl_type = F_UNLCK;
 419         goto out;
 420 }
 421
 422 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 423 {
 424         int res = 0;
 425         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 426                 case FL_POSIX:
 427                         res = posix_lock_file_wait(file, fl);
 428                         break;
 429                 case FL_FLOCK:
 430                         res = flock_lock_file_wait(file, fl);
 431                         break;
 432                 default:
 433                         BUG();
 434         }
 435         if (res < 0)
 436                 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
 437                         " - error %d!\n",
 438                                 __FUNCTION__, res);
 439         return res;
 440 }
 441
 442 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 443 {
 444         struct inode *inode = filp->f_mapping->host;
 445         int status;
 446
 447         /*
 448          * Flush all pending writes before doing anything
 449          * with locks..
 450          */
 451         nfs_sync_mapping(filp->f_mapping);
 452
 453         /* NOTE: special case
 454          *      If we're signalled while cleaning up locks on process exit, we
 455          *      still need to complete the unlock.
 456          */
 457         lock_kernel();
 458         /* Use local locking if mounted with "-onolock" */
 459         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 460                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 461         else
 462                 status = do_vfs_lock(filp, fl);
 463         unlock_kernel();
 464         return status;
 465 }
 466
 467 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 468 {
 469         struct inode *inode = filp->f_mapping->host;
 470         int status;
 471
 472         /*
 473          * Flush all pending writes before doing anything
 474          * with locks..
 475          */
 476         status = nfs_sync_mapping(filp->f_mapping);
 477         if (status != 0)
 478                 goto out;
 479
 480         lock_kernel();
 481         /* Use local locking if mounted with "-onolock" */
 482         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 483                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 484                 /* If we were signalled we still need to ensure that
 485                  * we clean up any state on the server. We therefore
 486                  * record the lock call as having succeeded in order to
 487                  * ensure that locks_remove_posix() cleans it out when
 488                  * the process exits.
 489                  */
 490                 if (status == -EINTR || status == -ERESTARTSYS)
 491                         do_vfs_lock(filp, fl);
 492         } else
 493                 status = do_vfs_lock(filp, fl);
 494         unlock_kernel();
 495         if (status < 0)
 496                 goto out;
 497         /*
 498          * Make sure we clear the cache whenever we try to get the lock.
 499          * This makes locking act as a cache coherency point.
 500          */
 501         nfs_sync_mapping(filp->f_mapping);
 502         nfs_zap_caches(inode);
 503 out:
 504         return status;
 505 }
 506
 507 /*
 508  * Lock a (portion of) a file
 509  */
 510 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 511 {
 512         struct inode * inode = filp->f_mapping->host;
 513
 514         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 515                         inode->i_sb->s_id, inode->i_ino,
 516                         fl->fl_type, fl->fl_flags,
 517                         (long long)fl->fl_start, (long long)fl->fl_end);
 518         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 519
 520         /* No mandatory locks over NFS */
 521         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 522             fl->fl_type != F_UNLCK)
 523                 return -ENOLCK;
 524
 525         if (IS_GETLK(cmd))
 526                 return do_getlk(filp, cmd, fl);
 527         if (fl->fl_type == F_UNLCK)
 528                 return do_unlk(filp, cmd, fl);
 529         return do_setlk(filp, cmd, fl);
 530 }
 531
 532 /*
 533  * Lock a (portion of) a file
 534  */
 535 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 536 {
 537         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 538                         filp->f_path.dentry->d_inode->i_sb->s_id,
 539                         filp->f_path.dentry->d_inode->i_ino,
 540                         fl->fl_type, fl->fl_flags);
 541
 542         /*
 543          * No BSD flocks over NFS allowed.
 544          * Note: we could try to fake a POSIX lock request here by
 545          * using ((u32) filp | 0x80000000) or some such as the pid.
 546          * Not sure whether that would be unique, though, or whether
 547          * that would break in other places.
 548          */
 549         if (!(fl->fl_flags & FL_FLOCK))
 550                 return -ENOLCK;
 551
 552         /* We're simulating flock() locks using posix locks on the server */
 553         fl->fl_owner = (fl_owner_t)filp;
 554         fl->fl_start = 0;
 555         fl->fl_end = OFFSET_MAX;
 556
 557         if (fl->fl_type == F_UNLCK)
 558                 return do_unlk(filp, cmd, fl);
 559         return do_setlk(filp, cmd, fl);
 560 }