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 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
  57
  58 const struct file_operations nfs_file_operations = {
  59         .llseek         = nfs_file_llseek,
  60         .read           = do_sync_read,
  61         .write          = do_sync_write,
  62         .aio_read       = nfs_file_read,
  63         .aio_write      = nfs_file_write,
  64         .mmap           = nfs_file_mmap,
  65         .open           = nfs_file_open,
  66         .flush          = nfs_file_flush,
  67         .release        = nfs_file_release,
  68         .fsync          = nfs_fsync,
  69         .lock           = nfs_lock,
  70         .flock          = nfs_flock,
  71         .splice_read    = nfs_file_splice_read,
  72         .check_flags    = nfs_check_flags,
  73         .setlease       = nfs_setlease,
  74 };
  75
  76 const struct inode_operations nfs_file_inode_operations = {
  77         .permission     = nfs_permission,
  78         .getattr        = nfs_getattr,
  79         .setattr        = nfs_setattr,
  80 };
  81
  82 #ifdef CONFIG_NFS_V3
  83 const struct inode_operations nfs3_file_inode_operations = {
  84         .permission     = nfs_permission,
  85         .getattr        = nfs_getattr,
  86         .setattr        = nfs_setattr,
  87         .listxattr      = nfs3_listxattr,
  88         .getxattr       = nfs3_getxattr,
  89         .setxattr       = nfs3_setxattr,
  90         .removexattr    = nfs3_removexattr,
  91 };
  92 #endif  /* CONFIG_NFS_v3 */
  93
  94 /* Hack for future NFS swap support */
  95 #ifndef IS_SWAPFILE
  96 # define IS_SWAPFILE(inode)     (0)
  97 #endif
  98
  99 static int nfs_check_flags(int flags)
 100 {
 101         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 102                 return -EINVAL;
 103
 104         return 0;
 105 }
 106
 107 /*
 108  * Open file
 109  */
 110 static int
 111 nfs_file_open(struct inode *inode, struct file *filp)
 112 {
 113         int res;
 114
 115         res = nfs_check_flags(filp->f_flags);
 116         if (res)
 117                 return res;
 118
 119         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 120         lock_kernel();
 121         res = NFS_PROTO(inode)->file_open(inode, filp);
 122         unlock_kernel();
 123         return res;
 124 }
 125
 126 static int
 127 nfs_file_release(struct inode *inode, struct file *filp)
 128 {
 129         /* Ensure that dirty pages are flushed out with the right creds */
 130         if (filp->f_mode & FMODE_WRITE)
 131                 filemap_fdatawrite(filp->f_mapping);
 132         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 133         return NFS_PROTO(inode)->file_release(inode, filp);
 134 }
 135
 136 /**
 137  * nfs_revalidate_size - Revalidate the file size
 138  * @inode - pointer to inode struct
 139  * @file - pointer to struct file
 140  *
 141  * Revalidates the file length. This is basically a wrapper around
 142  * nfs_revalidate_inode() that takes into account the fact that we may
 143  * have cached writes (in which case we don't care about the server's
 144  * idea of what the file length is), or O_DIRECT (in which case we
 145  * shouldn't trust the cache).
 146  */
 147 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 148 {
 149         struct nfs_server *server = NFS_SERVER(inode);
 150         struct nfs_inode *nfsi = NFS_I(inode);
 151
 152         if (server->flags & NFS_MOUNT_NOAC)
 153                 goto force_reval;
 154         if (filp->f_flags & O_DIRECT)
 155                 goto force_reval;
 156         if (nfsi->npages != 0)
 157                 return 0;
 158         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 159                 return 0;
 160 force_reval:
 161         return __nfs_revalidate_inode(server, inode);
 162 }
 163
 164 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 165 {
 166         /* origin == SEEK_END => we must revalidate the cached file length */
 167         if (origin == SEEK_END) {
 168                 struct inode *inode = filp->f_mapping->host;
 169                 int retval = nfs_revalidate_file_size(inode, filp);
 170                 if (retval < 0)
 171                         return (loff_t)retval;
 172         }
 173         return remote_llseek(filp, offset, origin);
 174 }
 175
 176 /*
 177  * Flush all dirty pages, and check for write errors.
 178  *
 179  */
 180 static int
 181 nfs_file_flush(struct file *file, fl_owner_t id)
 182 {
 183         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 184         struct inode    *inode = file->f_path.dentry->d_inode;
 185         int             status;
 186
 187         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 188
 189         if ((file->f_mode & FMODE_WRITE) == 0)
 190                 return 0;
 191         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 192         lock_kernel();
 193         /* Ensure that data+attribute caches are up to date after close() */
 194         status = nfs_wb_all(inode);
 195         if (!status) {
 196                 status = ctx->error;
 197                 ctx->error = 0;
 198                 if (!status)
 199                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 200         }
 201         unlock_kernel();
 202         return status;
 203 }
 204
 205 static ssize_t
 206 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
 207                 unsigned long nr_segs, loff_t pos)
 208 {
 209         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 210         struct inode * inode = dentry->d_inode;
 211         ssize_t result;
 212         size_t count = iov_length(iov, nr_segs);
 213
 214 #ifdef CONFIG_NFS_DIRECTIO
 215         if (iocb->ki_filp->f_flags & O_DIRECT)
 216                 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
 217 #endif
 218
 219         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 220                 dentry->d_parent->d_name.name, dentry->d_name.name,
 221                 (unsigned long) count, (unsigned long) pos);
 222
 223         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 224         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 225         if (!result)
 226                 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
 227         return result;
 228 }
 229
 230 static ssize_t
 231 nfs_file_splice_read(struct file *filp, loff_t *ppos,
 232                      struct pipe_inode_info *pipe, size_t count,
 233                      unsigned int flags)
 234 {
 235         struct dentry *dentry = filp->f_path.dentry;
 236         struct inode *inode = dentry->d_inode;
 237         ssize_t res;
 238
 239         dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
 240                 dentry->d_parent->d_name.name, dentry->d_name.name,
 241                 (unsigned long) count, (unsigned long long) *ppos);
 242
 243         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 244         if (!res)
 245                 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
 246         return res;
 247 }
 248
 249 static int
 250 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 251 {
 252         struct dentry *dentry = file->f_path.dentry;
 253         struct inode *inode = dentry->d_inode;
 254         int     status;
 255
 256         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 257                 dentry->d_parent->d_name.name, dentry->d_name.name);
 258
 259         status = nfs_revalidate_mapping(inode, file->f_mapping);
 260         if (!status)
 261                 status = generic_file_mmap(file, vma);
 262         return status;
 263 }
 264
 265 /*
 266  * Flush any dirty pages for this process, and check for write errors.
 267  * The return status from this call provides a reliable indication of
 268  * whether any write errors occurred for this process.
 269  */
 270 static int
 271 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 272 {
 273         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 274         struct inode *inode = dentry->d_inode;
 275         int status;
 276
 277         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 278
 279         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 280         lock_kernel();
 281         status = nfs_wb_all(inode);
 282         if (!status) {
 283                 status = ctx->error;
 284                 ctx->error = 0;
 285         }
 286         unlock_kernel();
 287         return status;
 288 }
 289
 290 /*
 291  * This does the "real" work of the write. The generic routine has
 292  * allocated the page, locked it, done all the page alignment stuff
 293  * calculations etc. Now we should just copy the data from user
 294  * space and write it back to the real medium..
 295  *
 296  * If the writer ends up delaying the write, the writer needs to
 297  * increment the page use counts until he is done with the page.
 298  */
 299 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 300 {
 301         return nfs_flush_incompatible(file, page);
 302 }
 303
 304 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 305 {
 306         long status;
 307
 308         lock_kernel();
 309         status = nfs_updatepage(file, page, offset, to-offset);
 310         unlock_kernel();
 311         return status;
 312 }
 313
 314 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 315 {
 316         if (offset != 0)
 317                 return;
 318         /* Cancel any unstarted writes on this page */
 319         nfs_wb_page_cancel(page->mapping->host, page);
 320 }
 321
 322 static int nfs_release_page(struct page *page, gfp_t gfp)
 323 {
 324         /* If PagePrivate() is set, then the page is not freeable */
 325         return 0;
 326 }
 327
 328 static int nfs_launder_page(struct page *page)
 329 {
 330         return nfs_wb_page(page->mapping->host, page);
 331 }
 332
 333 const struct address_space_operations nfs_file_aops = {
 334         .readpage = nfs_readpage,
 335         .readpages = nfs_readpages,
 336         .set_page_dirty = nfs_set_page_dirty,
 337         .writepage = nfs_writepage,
 338         .writepages = nfs_writepages,
 339         .prepare_write = nfs_prepare_write,
 340         .commit_write = nfs_commit_write,
 341         .invalidatepage = nfs_invalidate_page,
 342         .releasepage = nfs_release_page,
 343 #ifdef CONFIG_NFS_DIRECTIO
 344         .direct_IO = nfs_direct_IO,
 345 #endif
 346         .launder_page = nfs_launder_page,
 347 };
 348
 349 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
 350                                 unsigned long nr_segs, loff_t pos)
 351 {
 352         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 353         struct inode * inode = dentry->d_inode;
 354         ssize_t result;
 355         size_t count = iov_length(iov, nr_segs);
 356
 357 #ifdef CONFIG_NFS_DIRECTIO
 358         if (iocb->ki_filp->f_flags & O_DIRECT)
 359                 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
 360 #endif
 361
 362         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
 363                 dentry->d_parent->d_name.name, dentry->d_name.name,
 364                 inode->i_ino, (unsigned long) count, (long long) pos);
 365
 366         result = -EBUSY;
 367         if (IS_SWAPFILE(inode))
 368                 goto out_swapfile;
 369         /*
 370          * O_APPEND implies that we must revalidate the file length.
 371          */
 372         if (iocb->ki_filp->f_flags & O_APPEND) {
 373                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 374                 if (result)
 375                         goto out;
 376         }
 377
 378         result = count;
 379         if (!count)
 380                 goto out;
 381
 382         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 383         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 384         /* Return error values for O_SYNC and IS_SYNC() */
 385         if (result >= 0 && (IS_SYNC(inode) || (iocb->ki_filp->f_flags & O_SYNC))) {
 386                 int err = nfs_fsync(iocb->ki_filp, dentry, 1);
 387                 if (err < 0)
 388                         result = err;
 389         }
 390 out:
 391         return result;
 392
 393 out_swapfile:
 394         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 395         goto out;
 396 }
 397
 398 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 399 {
 400         struct inode *inode = filp->f_mapping->host;
 401         int status = 0;
 402
 403         lock_kernel();
 404         /* Try local locking first */
 405         posix_test_lock(filp, fl);
 406         if (fl->fl_type != F_UNLCK) {
 407                 /* found a conflict */
 408                 goto out;
 409         }
 410
 411         if (nfs_have_delegation(inode, FMODE_READ))
 412                 goto out_noconflict;
 413
 414         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 415                 goto out_noconflict;
 416
 417         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 418 out:
 419         unlock_kernel();
 420         return status;
 421 out_noconflict:
 422         fl->fl_type = F_UNLCK;
 423         goto out;
 424 }
 425
 426 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 427 {
 428         int res = 0;
 429         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 430                 case FL_POSIX:
 431                         res = posix_lock_file_wait(file, fl);
 432                         break;
 433                 case FL_FLOCK:
 434                         res = flock_lock_file_wait(file, fl);
 435                         break;
 436                 default:
 437                         BUG();
 438         }
 439         if (res < 0)
 440                 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
 441                         " - error %d!\n",
 442                                 __FUNCTION__, res);
 443         return res;
 444 }
 445
 446 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 447 {
 448         struct inode *inode = filp->f_mapping->host;
 449         int status;
 450
 451         /*
 452          * Flush all pending writes before doing anything
 453          * with locks..
 454          */
 455         nfs_sync_mapping(filp->f_mapping);
 456
 457         /* NOTE: special case
 458          *      If we're signalled while cleaning up locks on process exit, we
 459          *      still need to complete the unlock.
 460          */
 461         lock_kernel();
 462         /* Use local locking if mounted with "-onolock" */
 463         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 464                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 465         else
 466                 status = do_vfs_lock(filp, fl);
 467         unlock_kernel();
 468         return status;
 469 }
 470
 471 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 472 {
 473         struct inode *inode = filp->f_mapping->host;
 474         int status;
 475
 476         /*
 477          * Flush all pending writes before doing anything
 478          * with locks..
 479          */
 480         status = nfs_sync_mapping(filp->f_mapping);
 481         if (status != 0)
 482                 goto out;
 483
 484         lock_kernel();
 485         /* Use local locking if mounted with "-onolock" */
 486         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 487                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 488                 /* If we were signalled we still need to ensure that
 489                  * we clean up any state on the server. We therefore
 490                  * record the lock call as having succeeded in order to
 491                  * ensure that locks_remove_posix() cleans it out when
 492                  * the process exits.
 493                  */
 494                 if (status == -EINTR || status == -ERESTARTSYS)
 495                         do_vfs_lock(filp, fl);
 496         } else
 497                 status = do_vfs_lock(filp, fl);
 498         unlock_kernel();
 499         if (status < 0)
 500                 goto out;
 501         /*
 502          * Make sure we clear the cache whenever we try to get the lock.
 503          * This makes locking act as a cache coherency point.
 504          */
 505         nfs_sync_mapping(filp->f_mapping);
 506         nfs_zap_caches(inode);
 507 out:
 508         return status;
 509 }
 510
 511 /*
 512  * Lock a (portion of) a file
 513  */
 514 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 515 {
 516         struct inode * inode = filp->f_mapping->host;
 517
 518         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 519                         inode->i_sb->s_id, inode->i_ino,
 520                         fl->fl_type, fl->fl_flags,
 521                         (long long)fl->fl_start, (long long)fl->fl_end);
 522         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 523
 524         /* No mandatory locks over NFS */
 525         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 526             fl->fl_type != F_UNLCK)
 527                 return -ENOLCK;
 528
 529         if (IS_GETLK(cmd))
 530                 return do_getlk(filp, cmd, fl);
 531         if (fl->fl_type == F_UNLCK)
 532                 return do_unlk(filp, cmd, fl);
 533         return do_setlk(filp, cmd, fl);
 534 }
 535
 536 /*
 537  * Lock a (portion of) a file
 538  */
 539 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 540 {
 541         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 542                         filp->f_path.dentry->d_inode->i_sb->s_id,
 543                         filp->f_path.dentry->d_inode->i_ino,
 544                         fl->fl_type, fl->fl_flags);
 545
 546         /*
 547          * No BSD flocks over NFS allowed.
 548          * Note: we could try to fake a POSIX lock request here by
 549          * using ((u32) filp | 0x80000000) or some such as the pid.
 550          * Not sure whether that would be unique, though, or whether
 551          * that would break in other places.
 552          */
 553         if (!(fl->fl_flags & FL_FLOCK))
 554                 return -ENOLCK;
 555
 556         /* We're simulating flock() locks using posix locks on the server */
 557         fl->fl_owner = (fl_owner_t)filp;
 558         fl->fl_start = 0;
 559         fl->fl_end = OFFSET_MAX;
 560
 561         if (fl->fl_type == F_UNLCK)
 562                 return do_unlk(filp, cmd, fl);
 563         return do_setlk(filp, cmd, fl);
 564 }
 565
 566 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
 567 {
 568         /*
 569          * There is no protocol support for leases, so we have no way
 570          * to implement them correctly in the face of opens by other
 571          * clients.
 572          */
 573         return -EINVAL;
 574 }