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 *);
  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 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_file - Revalidate the page cache & related metadata
 131  * @inode - pointer to inode struct
 132  * @file - pointer to file
 133  */
 134 static int nfs_revalidate_file(struct inode *inode, struct file *filp)
 135 {
 136         struct nfs_inode *nfsi = NFS_I(inode);
 137         int retval = 0;
 138
 139         if ((nfsi->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATTR))
 140                         || nfs_attribute_timeout(inode))
 141                 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 142         nfs_revalidate_mapping(inode, filp->f_mapping);
 143         return 0;
 144 }
 145
 146 /**
 147  * nfs_revalidate_size - Revalidate the file size
 148  * @inode - pointer to inode struct
 149  * @file - pointer to struct file
 150  *
 151  * Revalidates the file length. This is basically a wrapper around
 152  * nfs_revalidate_inode() that takes into account the fact that we may
 153  * have cached writes (in which case we don't care about the server's
 154  * idea of what the file length is), or O_DIRECT (in which case we
 155  * shouldn't trust the cache).
 156  */
 157 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 158 {
 159         struct nfs_server *server = NFS_SERVER(inode);
 160         struct nfs_inode *nfsi = NFS_I(inode);
 161
 162         if (server->flags & NFS_MOUNT_NOAC)
 163                 goto force_reval;
 164         if (filp->f_flags & O_DIRECT)
 165                 goto force_reval;
 166         if (nfsi->npages != 0)
 167                 return 0;
 168         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 169                 return 0;
 170 force_reval:
 171         return __nfs_revalidate_inode(server, inode);
 172 }
 173
 174 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 175 {
 176         /* origin == SEEK_END => we must revalidate the cached file length */
 177         if (origin == 2) {
 178                 struct inode *inode = filp->f_mapping->host;
 179                 int retval = nfs_revalidate_file_size(inode, filp);
 180                 if (retval < 0)
 181                         return (loff_t)retval;
 182         }
 183         return remote_llseek(filp, offset, origin);
 184 }
 185
 186 /*
 187  * Flush all dirty pages, and check for write errors.
 188  *
 189  */
 190 static int
 191 nfs_file_flush(struct file *file)
 192 {
 193         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 194         struct inode    *inode = file->f_dentry->d_inode;
 195         int             status;
 196
 197         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 198
 199         if ((file->f_mode & FMODE_WRITE) == 0)
 200                 return 0;
 201         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 202         lock_kernel();
 203         /* Ensure that data+attribute caches are up to date after close() */
 204         status = nfs_wb_all(inode);
 205         if (!status) {
 206                 status = ctx->error;
 207                 ctx->error = 0;
 208                 if (!status)
 209                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 210         }
 211         unlock_kernel();
 212         return status;
 213 }
 214
 215 static ssize_t
 216 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
 217 {
 218         struct dentry * dentry = iocb->ki_filp->f_dentry;
 219         struct inode * inode = dentry->d_inode;
 220         ssize_t result;
 221
 222 #ifdef CONFIG_NFS_DIRECTIO
 223         if (iocb->ki_filp->f_flags & O_DIRECT)
 224                 return nfs_file_direct_read(iocb, buf, count, pos);
 225 #endif
 226
 227         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 228                 dentry->d_parent->d_name.name, dentry->d_name.name,
 229                 (unsigned long) count, (unsigned long) pos);
 230
 231         result = nfs_revalidate_file(inode, iocb->ki_filp);
 232         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 233         if (!result)
 234                 result = generic_file_aio_read(iocb, buf, count, pos);
 235         return result;
 236 }
 237
 238 static ssize_t
 239 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 240                 read_actor_t actor, void *target)
 241 {
 242         struct dentry *dentry = filp->f_dentry;
 243         struct inode *inode = dentry->d_inode;
 244         ssize_t res;
 245
 246         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 247                 dentry->d_parent->d_name.name, dentry->d_name.name,
 248                 (unsigned long) count, (unsigned long long) *ppos);
 249
 250         res = nfs_revalidate_file(inode, filp);
 251         if (!res)
 252                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 253         return res;
 254 }
 255
 256 static int
 257 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 258 {
 259         struct dentry *dentry = file->f_dentry;
 260         struct inode *inode = dentry->d_inode;
 261         int     status;
 262
 263         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 264                 dentry->d_parent->d_name.name, dentry->d_name.name);
 265
 266         status = nfs_revalidate_file(inode, file);
 267         if (!status)
 268                 status = generic_file_mmap(file, vma);
 269         return status;
 270 }
 271
 272 /*
 273  * Flush any dirty pages for this process, and check for write errors.
 274  * The return status from this call provides a reliable indication of
 275  * whether any write errors occurred for this process.
 276  */
 277 static int
 278 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 279 {
 280         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 281         struct inode *inode = dentry->d_inode;
 282         int status;
 283
 284         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 285
 286         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 287         lock_kernel();
 288         status = nfs_wb_all(inode);
 289         if (!status) {
 290                 status = ctx->error;
 291                 ctx->error = 0;
 292         }
 293         unlock_kernel();
 294         return status;
 295 }
 296
 297 /*
 298  * This does the "real" work of the write. The generic routine has
 299  * allocated the page, locked it, done all the page alignment stuff
 300  * calculations etc. Now we should just copy the data from user
 301  * space and write it back to the real medium..
 302  *
 303  * If the writer ends up delaying the write, the writer needs to
 304  * increment the page use counts until he is done with the page.
 305  */
 306 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 307 {
 308         return nfs_flush_incompatible(file, page);
 309 }
 310
 311 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 312 {
 313         long status;
 314
 315         lock_kernel();
 316         status = nfs_updatepage(file, page, offset, to-offset);
 317         unlock_kernel();
 318         return status;
 319 }
 320
 321 static int nfs_invalidate_page(struct page *page, unsigned long offset)
 322 {
 323         /* FIXME: we really should cancel any unstarted writes on this page */
 324         return 1;
 325 }
 326
 327 static int nfs_release_page(struct page *page, gfp_t gfp)
 328 {
 329         return !nfs_wb_page(page->mapping->host, page);
 330 }
 331
 332 struct address_space_operations nfs_file_aops = {
 333         .readpage = nfs_readpage,
 334         .readpages = nfs_readpages,
 335         .set_page_dirty = __set_page_dirty_nobuffers,
 336         .writepage = nfs_writepage,
 337         .writepages = nfs_writepages,
 338         .prepare_write = nfs_prepare_write,
 339         .commit_write = nfs_commit_write,
 340         .invalidatepage = nfs_invalidate_page,
 341         .releasepage = nfs_release_page,
 342 #ifdef CONFIG_NFS_DIRECTIO
 343         .direct_IO = nfs_direct_IO,
 344 #endif
 345 };
 346
 347 /*
 348  * Write to a file (through the page cache).
 349  */
 350 static ssize_t
 351 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 352 {
 353         struct dentry * dentry = iocb->ki_filp->f_dentry;
 354         struct inode * inode = dentry->d_inode;
 355         ssize_t result;
 356
 357 #ifdef CONFIG_NFS_DIRECTIO
 358         if (iocb->ki_filp->f_flags & O_DIRECT)
 359                 return nfs_file_direct_write(iocb, buf, count, pos);
 360 #endif
 361
 362         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 363                 dentry->d_parent->d_name.name, dentry->d_name.name,
 364                 inode->i_ino, (unsigned long) count, (unsigned 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         nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 378
 379         result = count;
 380         if (!count)
 381                 goto out;
 382
 383         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 384         result = generic_file_aio_write(iocb, buf, count, pos);
 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 file_lock *cfl;
 396         struct inode *inode = filp->f_mapping->host;
 397         int status = 0;
 398
 399         lock_kernel();
 400         /* Try local locking first */
 401         cfl = posix_test_lock(filp, fl);
 402         if (cfl != NULL) {
 403                 locks_copy_lock(fl, cfl);
 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                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 437                                 __FUNCTION__);
 438         return res;
 439 }
 440
 441 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 442 {
 443         struct inode *inode = filp->f_mapping->host;
 444         sigset_t oldset;
 445         int status;
 446
 447         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 448         /*
 449          * Flush all pending writes before doing anything
 450          * with locks..
 451          */
 452         nfs_sync_mapping(filp->f_mapping);
 453
 454         /* NOTE: special case
 455          *      If we're signalled while cleaning up locks on process exit, we
 456          *      still need to complete the unlock.
 457          */
 458         lock_kernel();
 459         /* Use local locking if mounted with "-onolock" */
 460         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 461                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 462         else
 463                 status = do_vfs_lock(filp, fl);
 464         unlock_kernel();
 465         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 466         return status;
 467 }
 468
 469 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 470 {
 471         struct inode *inode = filp->f_mapping->host;
 472         sigset_t oldset;
 473         int status;
 474
 475         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 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         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 509         return status;
 510 }
 511
 512 /*
 513  * Lock a (portion of) a file
 514  */
 515 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 516 {
 517         struct inode * inode = filp->f_mapping->host;
 518
 519         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 520                         inode->i_sb->s_id, inode->i_ino,
 521                         fl->fl_type, fl->fl_flags,
 522                         (long long)fl->fl_start, (long long)fl->fl_end);
 523         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 524
 525         /* No mandatory locks over NFS */
 526         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 527             fl->fl_type != F_UNLCK)
 528                 return -ENOLCK;
 529
 530         if (IS_GETLK(cmd))
 531                 return do_getlk(filp, cmd, fl);
 532         if (fl->fl_type == F_UNLCK)
 533                 return do_unlk(filp, cmd, fl);
 534         return do_setlk(filp, cmd, fl);
 535 }
 536
 537 /*
 538  * Lock a (portion of) a file
 539  */
 540 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 541 {
 542         struct inode * inode = filp->f_mapping->host;
 543
 544         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 545                         inode->i_sb->s_id, inode->i_ino,
 546                         fl->fl_type, fl->fl_flags);
 547
 548         /*
 549          * No BSD flocks over NFS allowed.
 550          * Note: we could try to fake a POSIX lock request here by
 551          * using ((u32) filp | 0x80000000) or some such as the pid.
 552          * Not sure whether that would be unique, though, or whether
 553          * that would break in other places.
 554          */
 555         if (!(fl->fl_flags & FL_FLOCK))
 556                 return -ENOLCK;
 557
 558         /* We're simulating flock() locks using posix locks on the server */
 559         fl->fl_owner = (fl_owner_t)filp;
 560         fl->fl_start = 0;
 561         fl->fl_end = OFFSET_MAX;
 562
 563         if (fl->fl_type == F_UNLCK)
 564                 return do_unlk(filp, cmd, fl);
 565         return do_setlk(filp, cmd, fl);
 566 }