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
  36 #define NFSDBG_FACILITY         NFSDBG_FILE
  37
  38 static int nfs_file_open(struct inode *, struct file *);
  39 static int nfs_file_release(struct inode *, struct file *);
  40 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  41 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  42 static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
  43 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
  44 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
  45 static int  nfs_file_flush(struct file *);
  46 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  47 static int nfs_check_flags(int flags);
  48 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  49 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  50
  51 struct file_operations nfs_file_operations = {
  52         .llseek         = nfs_file_llseek,
  53         .read           = do_sync_read,
  54         .write          = do_sync_write,
  55         .aio_read               = nfs_file_read,
  56         .aio_write              = nfs_file_write,
  57         .mmap           = nfs_file_mmap,
  58         .open           = nfs_file_open,
  59         .flush          = nfs_file_flush,
  60         .release        = nfs_file_release,
  61         .fsync          = nfs_fsync,
  62         .lock           = nfs_lock,
  63         .flock          = nfs_flock,
  64         .sendfile       = nfs_file_sendfile,
  65         .check_flags    = nfs_check_flags,
  66 };
  67
  68 struct inode_operations nfs_file_inode_operations = {
  69         .permission     = nfs_permission,
  70         .getattr        = nfs_getattr,
  71         .setattr        = nfs_setattr,
  72 };
  73
  74 #ifdef CONFIG_NFS_V3
  75 struct inode_operations nfs3_file_inode_operations = {
  76         .permission     = nfs_permission,
  77         .getattr        = nfs_getattr,
  78         .setattr        = nfs_setattr,
  79         .listxattr      = nfs3_listxattr,
  80         .getxattr       = nfs3_getxattr,
  81         .setxattr       = nfs3_setxattr,
  82         .removexattr    = nfs3_removexattr,
  83 };
  84 #endif  /* CONFIG_NFS_v3 */
  85
  86 /* Hack for future NFS swap support */
  87 #ifndef IS_SWAPFILE
  88 # define IS_SWAPFILE(inode)     (0)
  89 #endif
  90
  91 static int nfs_check_flags(int flags)
  92 {
  93         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  94                 return -EINVAL;
  95
  96         return 0;
  97 }
  98
  99 /*
 100  * Open file
 101  */
 102 static int
 103 nfs_file_open(struct inode *inode, struct file *filp)
 104 {
 105         struct nfs_server *server = NFS_SERVER(inode);
 106         int (*open)(struct inode *, struct file *);
 107         int res;
 108
 109         res = nfs_check_flags(filp->f_flags);
 110         if (res)
 111                 return res;
 112
 113         lock_kernel();
 114         /* Do NFSv4 open() call */
 115         if ((open = server->rpc_ops->file_open) != NULL)
 116                 res = 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         return NFS_PROTO(inode)->file_release(inode, filp);
 128 }
 129
 130 /**
 131  * nfs_revalidate_file - Revalidate the page cache & related metadata
 132  * @inode - pointer to inode struct
 133  * @file - pointer to file
 134  */
 135 static int nfs_revalidate_file(struct inode *inode, struct file *filp)
 136 {
 137         struct nfs_inode *nfsi = NFS_I(inode);
 138         int retval = 0;
 139
 140         if ((nfsi->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATTR))
 141                         || nfs_attribute_timeout(inode))
 142                 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 143         nfs_revalidate_mapping(inode, filp->f_mapping);
 144         return 0;
 145 }
 146
 147 /**
 148  * nfs_revalidate_size - Revalidate the file size
 149  * @inode - pointer to inode struct
 150  * @file - pointer to struct file
 151  *
 152  * Revalidates the file length. This is basically a wrapper around
 153  * nfs_revalidate_inode() that takes into account the fact that we may
 154  * have cached writes (in which case we don't care about the server's
 155  * idea of what the file length is), or O_DIRECT (in which case we
 156  * shouldn't trust the cache).
 157  */
 158 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 159 {
 160         struct nfs_server *server = NFS_SERVER(inode);
 161         struct nfs_inode *nfsi = NFS_I(inode);
 162
 163         if (server->flags & NFS_MOUNT_NOAC)
 164                 goto force_reval;
 165         if (filp->f_flags & O_DIRECT)
 166                 goto force_reval;
 167         if (nfsi->npages != 0)
 168                 return 0;
 169         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 170                 return 0;
 171 force_reval:
 172         return __nfs_revalidate_inode(server, inode);
 173 }
 174
 175 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 176 {
 177         /* origin == SEEK_END => we must revalidate the cached file length */
 178         if (origin == 2) {
 179                 struct inode *inode = filp->f_mapping->host;
 180                 int retval = nfs_revalidate_file_size(inode, filp);
 181                 if (retval < 0)
 182                         return (loff_t)retval;
 183         }
 184         return remote_llseek(filp, offset, origin);
 185 }
 186
 187 /*
 188  * Flush all dirty pages, and check for write errors.
 189  *
 190  */
 191 static int
 192 nfs_file_flush(struct file *file)
 193 {
 194         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 195         struct inode    *inode = file->f_dentry->d_inode;
 196         int             status;
 197
 198         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 199
 200         if ((file->f_mode & FMODE_WRITE) == 0)
 201                 return 0;
 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         if (!result)
 233                 result = generic_file_aio_read(iocb, buf, count, pos);
 234         return result;
 235 }
 236
 237 static ssize_t
 238 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 239                 read_actor_t actor, void *target)
 240 {
 241         struct dentry *dentry = filp->f_dentry;
 242         struct inode *inode = dentry->d_inode;
 243         ssize_t res;
 244
 245         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 246                 dentry->d_parent->d_name.name, dentry->d_name.name,
 247                 (unsigned long) count, (unsigned long long) *ppos);
 248
 249         res = nfs_revalidate_file(inode, filp);
 250         if (!res)
 251                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 252         return res;
 253 }
 254
 255 static int
 256 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 257 {
 258         struct dentry *dentry = file->f_dentry;
 259         struct inode *inode = dentry->d_inode;
 260         int     status;
 261
 262         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 263                 dentry->d_parent->d_name.name, dentry->d_name.name);
 264
 265         status = nfs_revalidate_file(inode, file);
 266         if (!status)
 267                 status = generic_file_mmap(file, vma);
 268         return status;
 269 }
 270
 271 /*
 272  * Flush any dirty pages for this process, and check for write errors.
 273  * The return status from this call provides a reliable indication of
 274  * whether any write errors occurred for this process.
 275  */
 276 static int
 277 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 278 {
 279         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 280         struct inode *inode = dentry->d_inode;
 281         int status;
 282
 283         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 284
 285         lock_kernel();
 286         status = nfs_wb_all(inode);
 287         if (!status) {
 288                 status = ctx->error;
 289                 ctx->error = 0;
 290         }
 291         unlock_kernel();
 292         return status;
 293 }
 294
 295 /*
 296  * This does the "real" work of the write. The generic routine has
 297  * allocated the page, locked it, done all the page alignment stuff
 298  * calculations etc. Now we should just copy the data from user
 299  * space and write it back to the real medium..
 300  *
 301  * If the writer ends up delaying the write, the writer needs to
 302  * increment the page use counts until he is done with the page.
 303  */
 304 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 305 {
 306         return nfs_flush_incompatible(file, page);
 307 }
 308
 309 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 310 {
 311         long status;
 312
 313         lock_kernel();
 314         status = nfs_updatepage(file, page, offset, to-offset);
 315         unlock_kernel();
 316         return status;
 317 }
 318
 319 struct address_space_operations nfs_file_aops = {
 320         .readpage = nfs_readpage,
 321         .readpages = nfs_readpages,
 322         .set_page_dirty = __set_page_dirty_nobuffers,
 323         .writepage = nfs_writepage,
 324         .writepages = nfs_writepages,
 325         .prepare_write = nfs_prepare_write,
 326         .commit_write = nfs_commit_write,
 327 #ifdef CONFIG_NFS_DIRECTIO
 328         .direct_IO = nfs_direct_IO,
 329 #endif
 330 };
 331
 332 /*
 333  * Write to a file (through the page cache).
 334  */
 335 static ssize_t
 336 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 337 {
 338         struct dentry * dentry = iocb->ki_filp->f_dentry;
 339         struct inode * inode = dentry->d_inode;
 340         ssize_t result;
 341
 342 #ifdef CONFIG_NFS_DIRECTIO
 343         if (iocb->ki_filp->f_flags & O_DIRECT)
 344                 return nfs_file_direct_write(iocb, buf, count, pos);
 345 #endif
 346
 347         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 348                 dentry->d_parent->d_name.name, dentry->d_name.name,
 349                 inode->i_ino, (unsigned long) count, (unsigned long) pos);
 350
 351         result = -EBUSY;
 352         if (IS_SWAPFILE(inode))
 353                 goto out_swapfile;
 354         /*
 355          * O_APPEND implies that we must revalidate the file length.
 356          */
 357         if (iocb->ki_filp->f_flags & O_APPEND) {
 358                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 359                 if (result)
 360                         goto out;
 361         }
 362         nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 363
 364         result = count;
 365         if (!count)
 366                 goto out;
 367
 368         result = generic_file_aio_write(iocb, buf, count, pos);
 369 out:
 370         return result;
 371
 372 out_swapfile:
 373         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 374         goto out;
 375 }
 376
 377 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 378 {
 379         struct file_lock *cfl;
 380         struct inode *inode = filp->f_mapping->host;
 381         int status = 0;
 382
 383         lock_kernel();
 384         /* Try local locking first */
 385         cfl = posix_test_lock(filp, fl);
 386         if (cfl != NULL) {
 387                 locks_copy_lock(fl, cfl);
 388                 goto out;
 389         }
 390
 391         if (nfs_have_delegation(inode, FMODE_READ))
 392                 goto out_noconflict;
 393
 394         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 395                 goto out_noconflict;
 396
 397         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 398 out:
 399         unlock_kernel();
 400         return status;
 401 out_noconflict:
 402         fl->fl_type = F_UNLCK;
 403         goto out;
 404 }
 405
 406 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 407 {
 408         int res = 0;
 409         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 410                 case FL_POSIX:
 411                         res = posix_lock_file_wait(file, fl);
 412                         break;
 413                 case FL_FLOCK:
 414                         res = flock_lock_file_wait(file, fl);
 415                         break;
 416                 default:
 417                         BUG();
 418         }
 419         if (res < 0)
 420                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 421                                 __FUNCTION__);
 422         return res;
 423 }
 424
 425 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 426 {
 427         struct inode *inode = filp->f_mapping->host;
 428         sigset_t oldset;
 429         int status;
 430
 431         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 432         /*
 433          * Flush all pending writes before doing anything
 434          * with locks..
 435          */
 436         nfs_sync_mapping(filp->f_mapping);
 437
 438         /* NOTE: special case
 439          *      If we're signalled while cleaning up locks on process exit, we
 440          *      still need to complete the unlock.
 441          */
 442         lock_kernel();
 443         /* Use local locking if mounted with "-onolock" */
 444         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 445                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 446         else
 447                 status = do_vfs_lock(filp, fl);
 448         unlock_kernel();
 449         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 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         sigset_t oldset;
 457         int status;
 458
 459         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 460         /*
 461          * Flush all pending writes before doing anything
 462          * with locks..
 463          */
 464         status = nfs_sync_mapping(filp->f_mapping);
 465         if (status != 0)
 466                 goto out;
 467
 468         lock_kernel();
 469         /* Use local locking if mounted with "-onolock" */
 470         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 471                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 472                 /* If we were signalled we still need to ensure that
 473                  * we clean up any state on the server. We therefore
 474                  * record the lock call as having succeeded in order to
 475                  * ensure that locks_remove_posix() cleans it out when
 476                  * the process exits.
 477                  */
 478                 if (status == -EINTR || status == -ERESTARTSYS)
 479                         do_vfs_lock(filp, fl);
 480         } else
 481                 status = do_vfs_lock(filp, fl);
 482         unlock_kernel();
 483         if (status < 0)
 484                 goto out;
 485         /*
 486          * Make sure we clear the cache whenever we try to get the lock.
 487          * This makes locking act as a cache coherency point.
 488          */
 489         nfs_sync_mapping(filp->f_mapping);
 490         nfs_zap_caches(inode);
 491 out:
 492         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 493         return status;
 494 }
 495
 496 /*
 497  * Lock a (portion of) a file
 498  */
 499 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 500 {
 501         struct inode * inode = filp->f_mapping->host;
 502
 503         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 504                         inode->i_sb->s_id, inode->i_ino,
 505                         fl->fl_type, fl->fl_flags,
 506                         (long long)fl->fl_start, (long long)fl->fl_end);
 507
 508         if (!inode)
 509                 return -EINVAL;
 510
 511         /* No mandatory locks over NFS */
 512         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 513             fl->fl_type != F_UNLCK)
 514                 return -ENOLCK;
 515
 516         if (IS_GETLK(cmd))
 517                 return do_getlk(filp, cmd, fl);
 518         if (fl->fl_type == F_UNLCK)
 519                 return do_unlk(filp, cmd, fl);
 520         return do_setlk(filp, cmd, fl);
 521 }
 522
 523 /*
 524  * Lock a (portion of) a file
 525  */
 526 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 527 {
 528         struct inode * inode = filp->f_mapping->host;
 529
 530         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 531                         inode->i_sb->s_id, inode->i_ino,
 532                         fl->fl_type, fl->fl_flags);
 533
 534         if (!inode)
 535                 return -EINVAL;
 536
 537         /*
 538          * No BSD flocks over NFS allowed.
 539          * Note: we could try to fake a POSIX lock request here by
 540          * using ((u32) filp | 0x80000000) or some such as the pid.
 541          * Not sure whether that would be unique, though, or whether
 542          * that would break in other places.
 543          */
 544         if (!(fl->fl_flags & FL_FLOCK))
 545                 return -ENOLCK;
 546
 547         /* We're simulating flock() locks using posix locks on the server */
 548         fl->fl_owner = (fl_owner_t)filp;
 549         fl->fl_start = 0;
 550         fl->fl_end = OFFSET_MAX;
 551
 552         if (fl->fl_type == F_UNLCK)
 553                 return do_unlk(filp, cmd, fl);
 554         return do_setlk(filp, cmd, fl);
 555 }