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