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 *, const struct iovec *iov,
  45                                 unsigned long nr_segs, loff_t pos);
  46 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
  47                                 unsigned long nr_segs, loff_t pos);
  48 static int  nfs_file_flush(struct file *, fl_owner_t id);
  49 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  50 static int nfs_check_flags(int flags);
  51 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  52 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  53
  54 const struct file_operations nfs_file_operations = {
  55         .llseek         = nfs_file_llseek,
  56         .read           = do_sync_read,
  57         .write          = do_sync_write,
  58         .aio_read       = nfs_file_read,
  59         .aio_write      = nfs_file_write,
  60         .mmap           = nfs_file_mmap,
  61         .open           = nfs_file_open,
  62         .flush          = nfs_file_flush,
  63         .release        = nfs_file_release,
  64         .fsync          = nfs_fsync,
  65         .lock           = nfs_lock,
  66         .flock          = nfs_flock,
  67         .sendfile       = nfs_file_sendfile,
  68         .check_flags    = nfs_check_flags,
  69 };
  70
  71 struct inode_operations nfs_file_inode_operations = {
  72         .permission     = nfs_permission,
  73         .getattr        = nfs_getattr,
  74         .setattr        = nfs_setattr,
  75 };
  76
  77 #ifdef CONFIG_NFS_V3
  78 struct inode_operations nfs3_file_inode_operations = {
  79         .permission     = nfs_permission,
  80         .getattr        = nfs_getattr,
  81         .setattr        = nfs_setattr,
  82         .listxattr      = nfs3_listxattr,
  83         .getxattr       = nfs3_getxattr,
  84         .setxattr       = nfs3_setxattr,
  85         .removexattr    = nfs3_removexattr,
  86 };
  87 #endif  /* CONFIG_NFS_v3 */
  88
  89 /* Hack for future NFS swap support */
  90 #ifndef IS_SWAPFILE
  91 # define IS_SWAPFILE(inode)     (0)
  92 #endif
  93
  94 static int nfs_check_flags(int flags)
  95 {
  96         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  97                 return -EINVAL;
  98
  99         return 0;
 100 }
 101
 102 /*
 103  * Open file
 104  */
 105 static int
 106 nfs_file_open(struct inode *inode, struct file *filp)
 107 {
 108         int res;
 109
 110         res = nfs_check_flags(filp->f_flags);
 111         if (res)
 112                 return res;
 113
 114         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 115         lock_kernel();
 116         res = NFS_PROTO(inode)->file_open(inode, filp);
 117         unlock_kernel();
 118         return res;
 119 }
 120
 121 static int
 122 nfs_file_release(struct inode *inode, struct file *filp)
 123 {
 124         /* Ensure that dirty pages are flushed out with the right creds */
 125         if (filp->f_mode & FMODE_WRITE)
 126                 filemap_fdatawrite(filp->f_mapping);
 127         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 128         return NFS_PROTO(inode)->file_release(inode, filp);
 129 }
 130
 131 /**
 132  * nfs_revalidate_size - Revalidate the file size
 133  * @inode - pointer to inode struct
 134  * @file - pointer to struct file
 135  *
 136  * Revalidates the file length. This is basically a wrapper around
 137  * nfs_revalidate_inode() that takes into account the fact that we may
 138  * have cached writes (in which case we don't care about the server's
 139  * idea of what the file length is), or O_DIRECT (in which case we
 140  * shouldn't trust the cache).
 141  */
 142 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 143 {
 144         struct nfs_server *server = NFS_SERVER(inode);
 145         struct nfs_inode *nfsi = NFS_I(inode);
 146
 147         if (server->flags & NFS_MOUNT_NOAC)
 148                 goto force_reval;
 149         if (filp->f_flags & O_DIRECT)
 150                 goto force_reval;
 151         if (nfsi->npages != 0)
 152                 return 0;
 153         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 154                 return 0;
 155 force_reval:
 156         return __nfs_revalidate_inode(server, inode);
 157 }
 158
 159 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 160 {
 161         /* origin == SEEK_END => we must revalidate the cached file length */
 162         if (origin == SEEK_END) {
 163                 struct inode *inode = filp->f_mapping->host;
 164                 int retval = nfs_revalidate_file_size(inode, filp);
 165                 if (retval < 0)
 166                         return (loff_t)retval;
 167         }
 168         return remote_llseek(filp, offset, origin);
 169 }
 170
 171 /*
 172  * Flush all dirty pages, and check for write errors.
 173  *
 174  */
 175 static int
 176 nfs_file_flush(struct file *file, fl_owner_t id)
 177 {
 178         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 179         struct inode    *inode = file->f_path.dentry->d_inode;
 180         int             status;
 181
 182         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 183
 184         if ((file->f_mode & FMODE_WRITE) == 0)
 185                 return 0;
 186         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 187         lock_kernel();
 188         /* Ensure that data+attribute caches are up to date after close() */
 189         status = nfs_wb_all(inode);
 190         if (!status) {
 191                 status = ctx->error;
 192                 ctx->error = 0;
 193                 if (!status)
 194                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 195         }
 196         unlock_kernel();
 197         return status;
 198 }
 199
 200 static ssize_t
 201 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
 202                 unsigned long nr_segs, loff_t pos)
 203 {
 204         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 205         struct inode * inode = dentry->d_inode;
 206         ssize_t result;
 207         size_t count = iov_length(iov, nr_segs);
 208
 209 #ifdef CONFIG_NFS_DIRECTIO
 210         if (iocb->ki_filp->f_flags & O_DIRECT)
 211                 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
 212 #endif
 213
 214         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 215                 dentry->d_parent->d_name.name, dentry->d_name.name,
 216                 (unsigned long) count, (unsigned long) pos);
 217
 218         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 219         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 220         if (!result)
 221                 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
 222         return result;
 223 }
 224
 225 static ssize_t
 226 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 227                 read_actor_t actor, void *target)
 228 {
 229         struct dentry *dentry = filp->f_path.dentry;
 230         struct inode *inode = dentry->d_inode;
 231         ssize_t res;
 232
 233         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 234                 dentry->d_parent->d_name.name, dentry->d_name.name,
 235                 (unsigned long) count, (unsigned long long) *ppos);
 236
 237         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 238         if (!res)
 239                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 240         return res;
 241 }
 242
 243 static int
 244 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 245 {
 246         struct dentry *dentry = file->f_path.dentry;
 247         struct inode *inode = dentry->d_inode;
 248         int     status;
 249
 250         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 251                 dentry->d_parent->d_name.name, dentry->d_name.name);
 252
 253         status = nfs_revalidate_mapping(inode, file->f_mapping);
 254         if (!status)
 255                 status = generic_file_mmap(file, vma);
 256         return status;
 257 }
 258
 259 /*
 260  * Flush any dirty pages for this process, and check for write errors.
 261  * The return status from this call provides a reliable indication of
 262  * whether any write errors occurred for this process.
 263  */
 264 static int
 265 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 266 {
 267         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 268         struct inode *inode = dentry->d_inode;
 269         int status;
 270
 271         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 272
 273         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 274         lock_kernel();
 275         status = nfs_wb_all(inode);
 276         if (!status) {
 277                 status = ctx->error;
 278                 ctx->error = 0;
 279         }
 280         unlock_kernel();
 281         return status;
 282 }
 283
 284 /*
 285  * This does the "real" work of the write. The generic routine has
 286  * allocated the page, locked it, done all the page alignment stuff
 287  * calculations etc. Now we should just copy the data from user
 288  * space and write it back to the real medium..
 289  *
 290  * If the writer ends up delaying the write, the writer needs to
 291  * increment the page use counts until he is done with the page.
 292  */
 293 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 294 {
 295         return nfs_flush_incompatible(file, page);
 296 }
 297
 298 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 299 {
 300         long status;
 301
 302         lock_kernel();
 303         status = nfs_updatepage(file, page, offset, to-offset);
 304         unlock_kernel();
 305         return status;
 306 }
 307
 308 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 309 {
 310         if (offset != 0)
 311                 return;
 312         /* Cancel any unstarted writes on this page */
 313         nfs_wb_page_priority(page->mapping->host, page, FLUSH_INVALIDATE);
 314 }
 315
 316 static int nfs_release_page(struct page *page, gfp_t gfp)
 317 {
 318         /*
 319          * Avoid deadlock on nfs_wait_on_request().
 320          */
 321         if (!(gfp & __GFP_FS))
 322                 return 0;
 323         /* Hack... Force nfs_wb_page() to write out the page */
 324         SetPageDirty(page);
 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 };
 342
 343 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
 344                                 unsigned long nr_segs, loff_t pos)
 345 {
 346         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 347         struct inode * inode = dentry->d_inode;
 348         ssize_t result;
 349         size_t count = iov_length(iov, nr_segs);
 350
 351 #ifdef CONFIG_NFS_DIRECTIO
 352         if (iocb->ki_filp->f_flags & O_DIRECT)
 353                 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
 354 #endif
 355
 356         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
 357                 dentry->d_parent->d_name.name, dentry->d_name.name,
 358                 inode->i_ino, (unsigned long) count, (long long) pos);
 359
 360         result = -EBUSY;
 361         if (IS_SWAPFILE(inode))
 362                 goto out_swapfile;
 363         /*
 364          * O_APPEND implies that we must revalidate the file length.
 365          */
 366         if (iocb->ki_filp->f_flags & O_APPEND) {
 367                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 368                 if (result)
 369                         goto out;
 370         }
 371
 372         result = count;
 373         if (!count)
 374                 goto out;
 375
 376         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 377         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 378         /* Return error values for O_SYNC and IS_SYNC() */
 379         if (result >= 0 && (IS_SYNC(inode) || (iocb->ki_filp->f_flags & O_SYNC))) {
 380                 int err = nfs_fsync(iocb->ki_filp, dentry, 1);
 381                 if (err < 0)
 382                         result = err;
 383         }
 384 out:
 385         return result;
 386
 387 out_swapfile:
 388         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 389         goto out;
 390 }
 391
 392 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 393 {
 394         struct file_lock cfl;
 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, &cfl)) {
 401                 fl->fl_start = cfl.fl_start;
 402                 fl->fl_end = cfl.fl_end;
 403                 fl->fl_type = cfl.fl_type;
 404                 fl->fl_pid = cfl.fl_pid;
 405                 goto out;
 406         }
 407
 408         if (nfs_have_delegation(inode, FMODE_READ))
 409                 goto out_noconflict;
 410
 411         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 412                 goto out_noconflict;
 413
 414         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 415 out:
 416         unlock_kernel();
 417         return status;
 418 out_noconflict:
 419         fl->fl_type = F_UNLCK;
 420         goto out;
 421 }
 422
 423 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 424 {
 425         int res = 0;
 426         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 427                 case FL_POSIX:
 428                         res = posix_lock_file_wait(file, fl);
 429                         break;
 430                 case FL_FLOCK:
 431                         res = flock_lock_file_wait(file, fl);
 432                         break;
 433                 default:
 434                         BUG();
 435         }
 436         if (res < 0)
 437                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 438                                 __FUNCTION__);
 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 }