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
  31 #include <asm/uaccess.h>
  32 #include <asm/system.h>
  33
  34 #include "delegation.h"
  35 #include "iostat.h"
  36
  37 #define NFSDBG_FACILITY         NFSDBG_FILE
  38
  39 static int nfs_file_open(struct inode *, struct file *);
  40 static int nfs_file_release(struct inode *, struct file *);
  41 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  42 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  43 static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
  44 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
  45 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
  46 static int  nfs_file_flush(struct file *, fl_owner_t id);
  47 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  48 static int nfs_check_flags(int flags);
  49 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  50 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  51
  52 const struct file_operations nfs_file_operations = {
  53         .llseek         = nfs_file_llseek,
  54         .read           = do_sync_read,
  55         .write          = do_sync_write,
  56         .aio_read               = nfs_file_read,
  57         .aio_write              = nfs_file_write,
  58         .mmap           = nfs_file_mmap,
  59         .open           = nfs_file_open,
  60         .flush          = nfs_file_flush,
  61         .release        = nfs_file_release,
  62         .fsync          = nfs_fsync,
  63         .lock           = nfs_lock,
  64         .flock          = nfs_flock,
  65         .sendfile       = nfs_file_sendfile,
  66         .check_flags    = nfs_check_flags,
  67 };
  68
  69 struct inode_operations nfs_file_inode_operations = {
  70         .permission     = nfs_permission,
  71         .getattr        = nfs_getattr,
  72         .setattr        = nfs_setattr,
  73 };
  74
  75 #ifdef CONFIG_NFS_V3
  76 struct inode_operations nfs3_file_inode_operations = {
  77         .permission     = nfs_permission,
  78         .getattr        = nfs_getattr,
  79         .setattr        = nfs_setattr,
  80         .listxattr      = nfs3_listxattr,
  81         .getxattr       = nfs3_getxattr,
  82         .setxattr       = nfs3_setxattr,
  83         .removexattr    = nfs3_removexattr,
  84 };
  85 #endif  /* CONFIG_NFS_v3 */
  86
  87 /* Hack for future NFS swap support */
  88 #ifndef IS_SWAPFILE
  89 # define IS_SWAPFILE(inode)     (0)
  90 #endif
  91
  92 static int nfs_check_flags(int flags)
  93 {
  94         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  95                 return -EINVAL;
  96
  97         return 0;
  98 }
  99
 100 /*
 101  * Open file
 102  */
 103 static int
 104 nfs_file_open(struct inode *inode, struct file *filp)
 105 {
 106         int res;
 107
 108         res = nfs_check_flags(filp->f_flags);
 109         if (res)
 110                 return res;
 111
 112         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 113         lock_kernel();
 114         res = NFS_SERVER(inode)->rpc_ops->file_open(inode, filp);
 115         unlock_kernel();
 116         return res;
 117 }
 118
 119 static int
 120 nfs_file_release(struct inode *inode, struct file *filp)
 121 {
 122         /* Ensure that dirty pages are flushed out with the right creds */
 123         if (filp->f_mode & FMODE_WRITE)
 124                 filemap_fdatawrite(filp->f_mapping);
 125         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 126         return NFS_PROTO(inode)->file_release(inode, filp);
 127 }
 128
 129 /**
 130  * nfs_revalidate_size - Revalidate the file size
 131  * @inode - pointer to inode struct
 132  * @file - pointer to struct file
 133  *
 134  * Revalidates the file length. This is basically a wrapper around
 135  * nfs_revalidate_inode() that takes into account the fact that we may
 136  * have cached writes (in which case we don't care about the server's
 137  * idea of what the file length is), or O_DIRECT (in which case we
 138  * shouldn't trust the cache).
 139  */
 140 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 141 {
 142         struct nfs_server *server = NFS_SERVER(inode);
 143         struct nfs_inode *nfsi = NFS_I(inode);
 144
 145         if (server->flags & NFS_MOUNT_NOAC)
 146                 goto force_reval;
 147         if (filp->f_flags & O_DIRECT)
 148                 goto force_reval;
 149         if (nfsi->npages != 0)
 150                 return 0;
 151         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 152                 return 0;
 153 force_reval:
 154         return __nfs_revalidate_inode(server, inode);
 155 }
 156
 157 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 158 {
 159         /* origin == SEEK_END => we must revalidate the cached file length */
 160         if (origin == 2) {
 161                 struct inode *inode = filp->f_mapping->host;
 162                 int retval = nfs_revalidate_file_size(inode, filp);
 163                 if (retval < 0)
 164                         return (loff_t)retval;
 165         }
 166         return remote_llseek(filp, offset, origin);
 167 }
 168
 169 /*
 170  * Flush all dirty pages, and check for write errors.
 171  *
 172  */
 173 static int
 174 nfs_file_flush(struct file *file, fl_owner_t id)
 175 {
 176         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 177         struct inode    *inode = file->f_dentry->d_inode;
 178         int             status;
 179
 180         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 181
 182         if ((file->f_mode & FMODE_WRITE) == 0)
 183                 return 0;
 184         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 185         lock_kernel();
 186         /* Ensure that data+attribute caches are up to date after close() */
 187         status = nfs_wb_all(inode);
 188         if (!status) {
 189                 status = ctx->error;
 190                 ctx->error = 0;
 191                 if (!status)
 192                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 193         }
 194         unlock_kernel();
 195         return status;
 196 }
 197
 198 static ssize_t
 199 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
 200 {
 201         struct dentry * dentry = iocb->ki_filp->f_dentry;
 202         struct inode * inode = dentry->d_inode;
 203         ssize_t result;
 204
 205 #ifdef CONFIG_NFS_DIRECTIO
 206         if (iocb->ki_filp->f_flags & O_DIRECT)
 207                 return nfs_file_direct_read(iocb, buf, count, pos);
 208 #endif
 209
 210         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 211                 dentry->d_parent->d_name.name, dentry->d_name.name,
 212                 (unsigned long) count, (unsigned long) pos);
 213
 214         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 215         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 216         if (!result)
 217                 result = generic_file_aio_read(iocb, buf, count, pos);
 218         return result;
 219 }
 220
 221 static ssize_t
 222 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 223                 read_actor_t actor, void *target)
 224 {
 225         struct dentry *dentry = filp->f_dentry;
 226         struct inode *inode = dentry->d_inode;
 227         ssize_t res;
 228
 229         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 230                 dentry->d_parent->d_name.name, dentry->d_name.name,
 231                 (unsigned long) count, (unsigned long long) *ppos);
 232
 233         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 234         if (!res)
 235                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 236         return res;
 237 }
 238
 239 static int
 240 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 241 {
 242         struct dentry *dentry = file->f_dentry;
 243         struct inode *inode = dentry->d_inode;
 244         int     status;
 245
 246         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 247                 dentry->d_parent->d_name.name, dentry->d_name.name);
 248
 249         status = nfs_revalidate_mapping(inode, file->f_mapping);
 250         if (!status)
 251                 status = generic_file_mmap(file, vma);
 252         return status;
 253 }
 254
 255 /*
 256  * Flush any dirty pages for this process, and check for write errors.
 257  * The return status from this call provides a reliable indication of
 258  * whether any write errors occurred for this process.
 259  */
 260 static int
 261 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 262 {
 263         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 264         struct inode *inode = dentry->d_inode;
 265         int status;
 266
 267         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 268
 269         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 270         lock_kernel();
 271         status = nfs_wb_all(inode);
 272         if (!status) {
 273                 status = ctx->error;
 274                 ctx->error = 0;
 275         }
 276         unlock_kernel();
 277         return status;
 278 }
 279
 280 /*
 281  * This does the "real" work of the write. The generic routine has
 282  * allocated the page, locked it, done all the page alignment stuff
 283  * calculations etc. Now we should just copy the data from user
 284  * space and write it back to the real medium..
 285  *
 286  * If the writer ends up delaying the write, the writer needs to
 287  * increment the page use counts until he is done with the page.
 288  */
 289 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 290 {
 291         return nfs_flush_incompatible(file, page);
 292 }
 293
 294 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 295 {
 296         long status;
 297
 298         lock_kernel();
 299         status = nfs_updatepage(file, page, offset, to-offset);
 300         unlock_kernel();
 301         return status;
 302 }
 303
 304 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 305 {
 306         struct inode *inode = page->mapping->host;
 307
 308         /* Cancel any unstarted writes on this page */
 309         if (offset == 0)
 310                 nfs_sync_inode_wait(inode, page->index, 1, FLUSH_INVALIDATE);
 311 }
 312
 313 static int nfs_release_page(struct page *page, gfp_t gfp)
 314 {
 315         return !nfs_wb_page(page->mapping->host, page);
 316 }
 317
 318 struct address_space_operations nfs_file_aops = {
 319         .readpage = nfs_readpage,
 320         .readpages = nfs_readpages,
 321         .set_page_dirty = __set_page_dirty_nobuffers,
 322         .writepage = nfs_writepage,
 323         .writepages = nfs_writepages,
 324         .prepare_write = nfs_prepare_write,
 325         .commit_write = nfs_commit_write,
 326         .invalidatepage = nfs_invalidate_page,
 327         .releasepage = nfs_release_page,
 328 #ifdef CONFIG_NFS_DIRECTIO
 329         .direct_IO = nfs_direct_IO,
 330 #endif
 331 };
 332
 333 /*
 334  * Write to a file (through the page cache).
 335  */
 336 static ssize_t
 337 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 338 {
 339         struct dentry * dentry = iocb->ki_filp->f_dentry;
 340         struct inode * inode = dentry->d_inode;
 341         ssize_t result;
 342
 343 #ifdef CONFIG_NFS_DIRECTIO
 344         if (iocb->ki_filp->f_flags & O_DIRECT)
 345                 return nfs_file_direct_write(iocb, buf, count, pos);
 346 #endif
 347
 348         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 349                 dentry->d_parent->d_name.name, dentry->d_name.name,
 350                 inode->i_ino, (unsigned long) count, (unsigned long) pos);
 351
 352         result = -EBUSY;
 353         if (IS_SWAPFILE(inode))
 354                 goto out_swapfile;
 355         /*
 356          * O_APPEND implies that we must revalidate the file length.
 357          */
 358         if (iocb->ki_filp->f_flags & O_APPEND) {
 359                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 360                 if (result)
 361                         goto out;
 362         }
 363
 364         result = count;
 365         if (!count)
 366                 goto out;
 367
 368         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 369         result = generic_file_aio_write(iocb, buf, count, pos);
 370 out:
 371         return result;
 372
 373 out_swapfile:
 374         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 375         goto out;
 376 }
 377
 378 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 379 {
 380         struct file_lock cfl;
 381         struct inode *inode = filp->f_mapping->host;
 382         int status = 0;
 383
 384         lock_kernel();
 385         /* Try local locking first */
 386         if (posix_test_lock(filp, fl, &cfl)) {
 387                 fl->fl_start = cfl.fl_start;
 388                 fl->fl_end = cfl.fl_end;
 389                 fl->fl_type = cfl.fl_type;
 390                 fl->fl_pid = cfl.fl_pid;
 391                 goto out;
 392         }
 393
 394         if (nfs_have_delegation(inode, FMODE_READ))
 395                 goto out_noconflict;
 396
 397         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 398                 goto out_noconflict;
 399
 400         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 401 out:
 402         unlock_kernel();
 403         return status;
 404 out_noconflict:
 405         fl->fl_type = F_UNLCK;
 406         goto out;
 407 }
 408
 409 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 410 {
 411         int res = 0;
 412         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 413                 case FL_POSIX:
 414                         res = posix_lock_file_wait(file, fl);
 415                         break;
 416                 case FL_FLOCK:
 417                         res = flock_lock_file_wait(file, fl);
 418                         break;
 419                 default:
 420                         BUG();
 421         }
 422         if (res < 0)
 423                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 424                                 __FUNCTION__);
 425         return res;
 426 }
 427
 428 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 429 {
 430         struct inode *inode = filp->f_mapping->host;
 431         int status;
 432
 433         /*
 434          * Flush all pending writes before doing anything
 435          * with locks..
 436          */
 437         nfs_sync_mapping(filp->f_mapping);
 438
 439         /* NOTE: special case
 440          *      If we're signalled while cleaning up locks on process exit, we
 441          *      still need to complete the unlock.
 442          */
 443         lock_kernel();
 444         /* Use local locking if mounted with "-onolock" */
 445         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 446                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 447         else
 448                 status = do_vfs_lock(filp, fl);
 449         unlock_kernel();
 450         return status;
 451 }
 452
 453 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 454 {
 455         struct inode *inode = filp->f_mapping->host;
 456         int status;
 457
 458         /*
 459          * Flush all pending writes before doing anything
 460          * with locks..
 461          */
 462         status = nfs_sync_mapping(filp->f_mapping);
 463         if (status != 0)
 464                 goto out;
 465
 466         lock_kernel();
 467         /* Use local locking if mounted with "-onolock" */
 468         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 469                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 470                 /* If we were signalled we still need to ensure that
 471                  * we clean up any state on the server. We therefore
 472                  * record the lock call as having succeeded in order to
 473                  * ensure that locks_remove_posix() cleans it out when
 474                  * the process exits.
 475                  */
 476                 if (status == -EINTR || status == -ERESTARTSYS)
 477                         do_vfs_lock(filp, fl);
 478         } else
 479                 status = do_vfs_lock(filp, fl);
 480         unlock_kernel();
 481         if (status < 0)
 482                 goto out;
 483         /*
 484          * Make sure we clear the cache whenever we try to get the lock.
 485          * This makes locking act as a cache coherency point.
 486          */
 487         nfs_sync_mapping(filp->f_mapping);
 488         nfs_zap_caches(inode);
 489 out:
 490         return status;
 491 }
 492
 493 /*
 494  * Lock a (portion of) a file
 495  */
 496 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 497 {
 498         struct inode * inode = filp->f_mapping->host;
 499
 500         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 501                         inode->i_sb->s_id, inode->i_ino,
 502                         fl->fl_type, fl->fl_flags,
 503                         (long long)fl->fl_start, (long long)fl->fl_end);
 504         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 505
 506         /* No mandatory locks over NFS */
 507         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 508             fl->fl_type != F_UNLCK)
 509                 return -ENOLCK;
 510
 511         if (IS_GETLK(cmd))
 512                 return do_getlk(filp, cmd, fl);
 513         if (fl->fl_type == F_UNLCK)
 514                 return do_unlk(filp, cmd, fl);
 515         return do_setlk(filp, cmd, fl);
 516 }
 517
 518 /*
 519  * Lock a (portion of) a file
 520  */
 521 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 522 {
 523         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 524                         filp->f_dentry->d_inode->i_sb->s_id,
 525                         filp->f_dentry->d_inode->i_ino,
 526                         fl->fl_type, fl->fl_flags);
 527
 528         /*
 529          * No BSD flocks over NFS allowed.
 530          * Note: we could try to fake a POSIX lock request here by
 531          * using ((u32) filp | 0x80000000) or some such as the pid.
 532          * Not sure whether that would be unique, though, or whether
 533          * that would break in other places.
 534          */
 535         if (!(fl->fl_flags & FL_FLOCK))
 536                 return -ENOLCK;
 537
 538         /* We're simulating flock() locks using posix locks on the server */
 539         fl->fl_owner = (fl_owner_t)filp;
 540         fl->fl_start = 0;
 541         fl->fl_end = OFFSET_MAX;
 542
 543         if (fl->fl_type == F_UNLCK)
 544                 return do_unlk(filp, cmd, fl);
 545         return do_setlk(filp, cmd, fl);
 546 }