module/spl/spl-vnode.c

   1 /*****************************************************************************\
   2  *  Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
   3  *  Copyright (C) 2007 The Regents of the University of California.
   4  *  Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
   5  *  Written by Brian Behlendorf <behlendorf1@llnl.gov>.
   6  *  UCRL-CODE-235197
   7  *
   8  *  This file is part of the SPL, Solaris Porting Layer.
   9  *  For details, see <http://github.com/behlendorf/spl/>.
  10  *
  11  *  The SPL is free software; you can redistribute it and/or modify it
  12  *  under the terms of the GNU General Public License as published by the
  13  *  Free Software Foundation; either version 2 of the License, or (at your
  14  *  option) any later version.
  15  *
  16  *  The SPL is distributed in the hope that it will be useful, but WITHOUT
  17  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  18  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  19  *  for more details.
  20  *
  21  *  You should have received a copy of the GNU General Public License along
  22  *  with the SPL.  If not, see <http://www.gnu.org/licenses/>.
  23  *****************************************************************************
  24  *  Solaris Porting Layer (SPL) Vnode Implementation.
  25 \*****************************************************************************/
  26
  27 #include <sys/vnode.h>
  28 #include <spl-debug.h>
  29
  30 #ifdef SS_DEBUG_SUBSYS
  31 #undef SS_DEBUG_SUBSYS
  32 #endif
  33
  34 #define SS_DEBUG_SUBSYS SS_VNODE
  35
  36 vnode_t *rootdir = (vnode_t *)0xabcd1234;
  37 EXPORT_SYMBOL(rootdir);
  38
  39 static spl_kmem_cache_t *vn_cache;
  40 static spl_kmem_cache_t *vn_file_cache;
  41
  42 static DEFINE_SPINLOCK(vn_file_lock);
  43 static LIST_HEAD(vn_file_list);
  44
  45 vtype_t
  46 vn_mode_to_vtype(mode_t mode)
  47 {
  48         if (S_ISREG(mode))
  49                 return VREG;
  50
  51         if (S_ISDIR(mode))
  52                 return VDIR;
  53
  54         if (S_ISCHR(mode))
  55                 return VCHR;
  56
  57         if (S_ISBLK(mode))
  58                 return VBLK;
  59
  60         if (S_ISFIFO(mode))
  61                 return VFIFO;
  62
  63         if (S_ISLNK(mode))
  64                 return VLNK;
  65
  66         if (S_ISSOCK(mode))
  67                 return VSOCK;
  68
  69         if (S_ISCHR(mode))
  70                 return VCHR;
  71
  72         return VNON;
  73 } /* vn_mode_to_vtype() */
  74 EXPORT_SYMBOL(vn_mode_to_vtype);
  75
  76 mode_t
  77 vn_vtype_to_mode(vtype_t vtype)
  78 {
  79         if (vtype == VREG)
  80                 return S_IFREG;
  81
  82         if (vtype == VDIR)
  83                 return S_IFDIR;
  84
  85         if (vtype == VCHR)
  86                 return S_IFCHR;
  87
  88         if (vtype == VBLK)
  89                 return S_IFBLK;
  90
  91         if (vtype == VFIFO)
  92                 return S_IFIFO;
  93
  94         if (vtype == VLNK)
  95                 return S_IFLNK;
  96
  97         if (vtype == VSOCK)
  98                 return S_IFSOCK;
  99
 100         return VNON;
 101 } /* vn_vtype_to_mode() */
 102 EXPORT_SYMBOL(vn_vtype_to_mode);
 103
 104 vnode_t *
 105 vn_alloc(int flag)
 106 {
 107         vnode_t *vp;
 108         SENTRY;
 109
 110         vp = kmem_cache_alloc(vn_cache, flag);
 111         if (vp != NULL) {
 112                 vp->v_file = NULL;
 113                 vp->v_type = 0;
 114         }
 115
 116         SRETURN(vp);
 117 } /* vn_alloc() */
 118 EXPORT_SYMBOL(vn_alloc);
 119
 120 void
 121 vn_free(vnode_t *vp)
 122 {
 123         SENTRY;
 124         kmem_cache_free(vn_cache, vp);
 125         SEXIT;
 126 } /* vn_free() */
 127 EXPORT_SYMBOL(vn_free);
 128
 129 int
 130 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 131         vnode_t **vpp, int x1, void *x2)
 132 {
 133         struct file *fp;
 134         struct kstat stat;
 135         int rc, saved_umask = 0;
 136         gfp_t saved_gfp;
 137         vnode_t *vp;
 138         SENTRY;
 139
 140         ASSERT(flags & (FWRITE | FREAD));
 141         ASSERT(seg == UIO_SYSSPACE);
 142         ASSERT(vpp);
 143         *vpp = NULL;
 144
 145         if (!(flags & FCREAT) && (flags & FWRITE))
 146                 flags |= FEXCL;
 147
 148         /* Note for filp_open() the two low bits must be remapped to mean:
 149          * 01 - read-only  -> 00 read-only
 150          * 10 - write-only -> 01 write-only
 151          * 11 - read-write -> 10 read-write
 152          */
 153         flags--;
 154
 155         if (flags & FCREAT)
 156                 saved_umask = xchg(&current->fs->umask, 0);
 157
 158         fp = filp_open(path, flags, mode);
 159
 160         if (flags & FCREAT)
 161                 (void)xchg(&current->fs->umask, saved_umask);
 162
 163         if (IS_ERR(fp))
 164                 SRETURN(-PTR_ERR(fp));
 165
 166         rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
 167         if (rc) {
 168                 filp_close(fp, 0);
 169                 SRETURN(-rc);
 170         }
 171
 172         vp = vn_alloc(KM_SLEEP);
 173         if (!vp) {
 174                 filp_close(fp, 0);
 175                 SRETURN(ENOMEM);
 176         }
 177
 178         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 179         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 180
 181         mutex_enter(&vp->v_lock);
 182         vp->v_type = vn_mode_to_vtype(stat.mode);
 183         vp->v_file = fp;
 184         vp->v_gfp_mask = saved_gfp;
 185         *vpp = vp;
 186         mutex_exit(&vp->v_lock);
 187
 188         SRETURN(0);
 189 } /* vn_open() */
 190 EXPORT_SYMBOL(vn_open);
 191
 192 int
 193 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 194           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 195 {
 196         char *realpath;
 197         int len, rc;
 198         SENTRY;
 199
 200         ASSERT(vp == rootdir);
 201
 202         len = strlen(path) + 2;
 203         realpath = kmalloc(len, GFP_KERNEL);
 204         if (!realpath)
 205                 SRETURN(ENOMEM);
 206
 207         (void)snprintf(realpath, len, "/%s", path);
 208         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 209         kfree(realpath);
 210
 211         SRETURN(rc);
 212 } /* vn_openat() */
 213 EXPORT_SYMBOL(vn_openat);
 214
 215 int
 216 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 217         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 218 {
 219         loff_t offset;
 220         mm_segment_t saved_fs;
 221         struct file *fp;
 222         int rc;
 223         SENTRY;
 224
 225         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 226         ASSERT(vp);
 227         ASSERT(vp->v_file);
 228         ASSERT(seg == UIO_SYSSPACE);
 229         ASSERT((ioflag & ~FAPPEND) == 0);
 230         ASSERT(x2 == RLIM64_INFINITY);
 231
 232         fp = vp->v_file;
 233
 234         offset = off;
 235         if (ioflag & FAPPEND)
 236                 offset = fp->f_pos;
 237
 238         /* Writable user data segment must be briefly increased for this
 239          * process so we can use the user space read call paths to write
 240          * in to memory allocated by the kernel. */
 241         saved_fs = get_fs();
 242         set_fs(get_ds());
 243
 244         if (uio & UIO_WRITE)
 245                 rc = vfs_write(fp, addr, len, &offset);
 246         else
 247                 rc = vfs_read(fp, addr, len, &offset);
 248
 249         set_fs(saved_fs);
 250
 251         if (rc < 0)
 252                 SRETURN(-rc);
 253
 254         if (residp) {
 255                 *residp = len - rc;
 256         } else {
 257                 if (rc != len)
 258                         SRETURN(EIO);
 259         }
 260
 261         SRETURN(0);
 262 } /* vn_rdwr() */
 263 EXPORT_SYMBOL(vn_rdwr);
 264
 265 int
 266 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 267 {
 268         int rc;
 269         SENTRY;
 270
 271         ASSERT(vp);
 272         ASSERT(vp->v_file);
 273
 274         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 275         rc = filp_close(vp->v_file, 0);
 276         vn_free(vp);
 277
 278         SRETURN(-rc);
 279 } /* vn_close() */
 280 EXPORT_SYMBOL(vn_close);
 281
 282 /* vn_seek() does not actually seek it only performs bounds checking on the
 283  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 284  * anything more serious. */
 285 int
 286 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 287 {
 288         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 289 }
 290 EXPORT_SYMBOL(vn_seek);
 291
 292 static struct dentry *
 293 vn_lookup_hash(struct nameidata *nd)
 294 {
 295         return lookup_one_len((const char *)nd->last.name,
 296                               nd->nd_dentry, nd->last.len);
 297 } /* lookup_hash() */
 298
 299 static void
 300 vn_path_release(struct nameidata *nd)
 301 {
 302         dput(nd->nd_dentry);
 303         mntput(nd->nd_mnt);
 304 }
 305
 306 /* Modified do_unlinkat() from linux/fs/namei.c, only uses exported symbols */
 307 int
 308 vn_remove(const char *path, uio_seg_t seg, int flags)
 309 {
 310         struct dentry *dentry;
 311         struct nameidata nd;
 312         struct inode *inode = NULL;
 313         int rc = 0;
 314         SENTRY;
 315
 316         ASSERT(seg == UIO_SYSSPACE);
 317         ASSERT(flags == RMFILE);
 318
 319         rc = path_lookup(path, LOOKUP_PARENT, &nd);
 320         if (rc)
 321                 SGOTO(exit, rc);
 322
 323         rc = -EISDIR;
 324         if (nd.last_type != LAST_NORM)
 325                 SGOTO(exit1, rc);
 326
 327         spl_inode_lock_nested(nd.nd_dentry->d_inode, I_MUTEX_PARENT);
 328         dentry = vn_lookup_hash(&nd);
 329         rc = PTR_ERR(dentry);
 330         if (!IS_ERR(dentry)) {
 331                 /* Why not before? Because we want correct rc value */
 332                 if (nd.last.name[nd.last.len])
 333                         SGOTO(slashes, rc);
 334
 335                 inode = dentry->d_inode;
 336                 if (inode)
 337                         atomic_inc(&inode->i_count);
 338 #ifdef HAVE_2ARGS_VFS_UNLINK
 339                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry);
 340 #else
 341                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry, nd.nd_mnt);
 342 #endif /* HAVE_2ARGS_VFS_UNLINK */
 343 exit2:
 344                 dput(dentry);
 345         }
 346
 347         spl_inode_unlock(nd.nd_dentry->d_inode);
 348         if (inode)
 349                 iput(inode);    /* truncate the inode here */
 350 exit1:
 351         vn_path_release(&nd);
 352 exit:
 353         SRETURN(-rc);
 354
 355 slashes:
 356         rc = !dentry->d_inode ? -ENOENT :
 357                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 358         SGOTO(exit2, rc);
 359 } /* vn_remove() */
 360 EXPORT_SYMBOL(vn_remove);
 361
 362 /* Modified do_rename() from linux/fs/namei.c, only uses exported symbols */
 363 int
 364 vn_rename(const char *oldname, const char *newname, int x1)
 365 {
 366         struct dentry *old_dir, *new_dir;
 367         struct dentry *old_dentry, *new_dentry;
 368         struct dentry *trap;
 369         struct nameidata oldnd, newnd;
 370         int rc = 0;
 371         SENTRY;
 372
 373         rc = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
 374         if (rc)
 375                 SGOTO(exit, rc);
 376
 377         rc = path_lookup(newname, LOOKUP_PARENT, &newnd);
 378         if (rc)
 379                 SGOTO(exit1, rc);
 380
 381         rc = -EXDEV;
 382         if (oldnd.nd_mnt != newnd.nd_mnt)
 383                 SGOTO(exit2, rc);
 384
 385         old_dir = oldnd.nd_dentry;
 386         rc = -EBUSY;
 387         if (oldnd.last_type != LAST_NORM)
 388                 SGOTO(exit2, rc);
 389
 390         new_dir = newnd.nd_dentry;
 391         if (newnd.last_type != LAST_NORM)
 392                 SGOTO(exit2, rc);
 393
 394         trap = lock_rename(new_dir, old_dir);
 395
 396         old_dentry = vn_lookup_hash(&oldnd);
 397
 398         rc = PTR_ERR(old_dentry);
 399         if (IS_ERR(old_dentry))
 400                 SGOTO(exit3, rc);
 401
 402         /* source must exist */
 403         rc = -ENOENT;
 404         if (!old_dentry->d_inode)
 405                 SGOTO(exit4, rc);
 406
 407         /* unless the source is a directory trailing slashes give -ENOTDIR */
 408         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 409                 rc = -ENOTDIR;
 410                 if (oldnd.last.name[oldnd.last.len])
 411                         SGOTO(exit4, rc);
 412                 if (newnd.last.name[newnd.last.len])
 413                         SGOTO(exit4, rc);
 414         }
 415
 416         /* source should not be ancestor of target */
 417         rc = -EINVAL;
 418         if (old_dentry == trap)
 419                 SGOTO(exit4, rc);
 420
 421         new_dentry = vn_lookup_hash(&newnd);
 422         rc = PTR_ERR(new_dentry);
 423         if (IS_ERR(new_dentry))
 424                 SGOTO(exit4, rc);
 425
 426         /* target should not be an ancestor of source */
 427         rc = -ENOTEMPTY;
 428         if (new_dentry == trap)
 429                 SGOTO(exit5, rc);
 430
 431 #ifdef HAVE_4ARGS_VFS_RENAME
 432         rc = vfs_rename(old_dir->d_inode, old_dentry,
 433                         new_dir->d_inode, new_dentry);
 434 #else
 435         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 436                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 437 #endif /* HAVE_4ARGS_VFS_RENAME */
 438 exit5:
 439         dput(new_dentry);
 440 exit4:
 441         dput(old_dentry);
 442 exit3:
 443         unlock_rename(new_dir, old_dir);
 444 exit2:
 445         vn_path_release(&newnd);
 446 exit1:
 447         vn_path_release(&oldnd);
 448 exit:
 449         SRETURN(-rc);
 450 }
 451 EXPORT_SYMBOL(vn_rename);
 452
 453 int
 454 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 455 {
 456         struct file *fp;
 457         struct kstat stat;
 458         int rc;
 459         SENTRY;
 460
 461         ASSERT(vp);
 462         ASSERT(vp->v_file);
 463         ASSERT(vap);
 464
 465         fp = vp->v_file;
 466
 467         rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
 468         if (rc)
 469                 SRETURN(-rc);
 470
 471         vap->va_type          = vn_mode_to_vtype(stat.mode);
 472         vap->va_mode          = stat.mode;
 473         vap->va_uid           = stat.uid;
 474         vap->va_gid           = stat.gid;
 475         vap->va_fsid          = 0;
 476         vap->va_nodeid        = stat.ino;
 477         vap->va_nlink         = stat.nlink;
 478         vap->va_size          = stat.size;
 479         vap->va_blksize       = stat.blksize;
 480         vap->va_atime         = stat.atime;
 481         vap->va_mtime         = stat.mtime;
 482         vap->va_ctime         = stat.ctime;
 483         vap->va_rdev          = stat.rdev;
 484         vap->va_nblocks       = stat.blocks;
 485
 486         SRETURN(0);
 487 }
 488 EXPORT_SYMBOL(vn_getattr);
 489
 490 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 491 {
 492         int datasync = 0;
 493         SENTRY;
 494
 495         ASSERT(vp);
 496         ASSERT(vp->v_file);
 497
 498         if (flags & FDSYNC)
 499                 datasync = 1;
 500
 501         SRETURN(-spl_filp_fsync(vp->v_file, datasync));
 502 } /* vn_fsync() */
 503 EXPORT_SYMBOL(vn_fsync);
 504
 505 /* Function must be called while holding the vn_file_lock */
 506 static file_t *
 507 file_find(int fd)
 508 {
 509         file_t *fp;
 510
 511         ASSERT(spin_is_locked(&vn_file_lock));
 512
 513         list_for_each_entry(fp, &vn_file_list,  f_list) {
 514                 if (fd == fp->f_fd) {
 515                         ASSERT(atomic_read(&fp->f_ref) != 0);
 516                         return fp;
 517                 }
 518         }
 519
 520         return NULL;
 521 } /* file_find() */
 522
 523 file_t *
 524 vn_getf(int fd)
 525 {
 526         struct kstat stat;
 527         struct file *lfp;
 528         file_t *fp;
 529         vnode_t *vp;
 530         int rc = 0;
 531         SENTRY;
 532
 533         /* Already open just take an extra reference */
 534         spin_lock(&vn_file_lock);
 535
 536         fp = file_find(fd);
 537         if (fp) {
 538                 atomic_inc(&fp->f_ref);
 539                 spin_unlock(&vn_file_lock);
 540                 SRETURN(fp);
 541         }
 542
 543         spin_unlock(&vn_file_lock);
 544
 545         /* File was not yet opened create the object and setup */
 546         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 547         if (fp == NULL)
 548                 SGOTO(out, rc);
 549
 550         mutex_enter(&fp->f_lock);
 551
 552         fp->f_fd = fd;
 553         fp->f_offset = 0;
 554         atomic_inc(&fp->f_ref);
 555
 556         lfp = fget(fd);
 557         if (lfp == NULL)
 558                 SGOTO(out_mutex, rc);
 559
 560         vp = vn_alloc(KM_SLEEP);
 561         if (vp == NULL)
 562                 SGOTO(out_fget, rc);
 563
 564         if (vfs_getattr(lfp->f_vfsmnt, lfp->f_dentry, &stat))
 565                 SGOTO(out_vnode, rc);
 566
 567         mutex_enter(&vp->v_lock);
 568         vp->v_type = vn_mode_to_vtype(stat.mode);
 569         vp->v_file = lfp;
 570         mutex_exit(&vp->v_lock);
 571
 572         fp->f_vnode = vp;
 573         fp->f_file = lfp;
 574
 575         /* Put it on the tracking list */
 576         spin_lock(&vn_file_lock);
 577         list_add(&fp->f_list, &vn_file_list);
 578         spin_unlock(&vn_file_lock);
 579
 580         mutex_exit(&fp->f_lock);
 581         SRETURN(fp);
 582
 583 out_vnode:
 584         vn_free(vp);
 585 out_fget:
 586         fput(lfp);
 587 out_mutex:
 588         mutex_exit(&fp->f_lock);
 589         kmem_cache_free(vn_file_cache, fp);
 590 out:
 591         SRETURN(NULL);
 592 } /* getf() */
 593 EXPORT_SYMBOL(getf);
 594
 595 static void releasef_locked(file_t *fp)
 596 {
 597         ASSERT(fp->f_file);
 598         ASSERT(fp->f_vnode);
 599
 600         /* Unlinked from list, no refs, safe to free outside mutex */
 601         fput(fp->f_file);
 602         vn_free(fp->f_vnode);
 603
 604         kmem_cache_free(vn_file_cache, fp);
 605 }
 606
 607 void
 608 vn_releasef(int fd)
 609 {
 610         file_t *fp;
 611         SENTRY;
 612
 613         spin_lock(&vn_file_lock);
 614         fp = file_find(fd);
 615         if (fp) {
 616                 atomic_dec(&fp->f_ref);
 617                 if (atomic_read(&fp->f_ref) > 0) {
 618                         spin_unlock(&vn_file_lock);
 619                         SEXIT;
 620                         return;
 621                 }
 622
 623                 list_del(&fp->f_list);
 624                 releasef_locked(fp);
 625         }
 626         spin_unlock(&vn_file_lock);
 627
 628         SEXIT;
 629         return;
 630 } /* releasef() */
 631 EXPORT_SYMBOL(releasef);
 632
 633 #ifndef HAVE_SET_FS_PWD
 634 # ifdef HAVE_2ARGS_SET_FS_PWD
 635 /* Used from 2.6.25 - 2.6.31+ */
 636 void
 637 set_fs_pwd(struct fs_struct *fs, struct path *path)
 638 {
 639         struct path old_pwd;
 640
 641 #  ifdef HAVE_FS_STRUCT_SPINLOCK
 642         spin_lock(&fs->lock);
 643         old_pwd = fs->pwd;
 644         fs->pwd = *path;
 645         path_get(path);
 646         spin_unlock(&fs->lock);
 647 #  else
 648         write_lock(&fs->lock);
 649         old_pwd = fs->pwd;
 650         fs->pwd = *path;
 651         path_get(path);
 652         write_unlock(&fs->lock);
 653 #  endif /* HAVE_FS_STRUCT_SPINLOCK */
 654
 655         if (old_pwd.dentry)
 656                 path_put(&old_pwd);
 657 }
 658 # else
 659 /* Used from 2.6.11 - 2.6.24 */
 660 void
 661 set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, struct dentry *dentry)
 662 {
 663         struct dentry *old_pwd;
 664         struct vfsmount *old_pwdmnt;
 665
 666         write_lock(&fs->lock);
 667         old_pwd = fs->pwd;
 668         old_pwdmnt = fs->pwdmnt;
 669         fs->pwdmnt = mntget(mnt);
 670         fs->pwd = dget(dentry);
 671         write_unlock(&fs->lock);
 672
 673         if (old_pwd) {
 674                 dput(old_pwd);
 675                 mntput(old_pwdmnt);
 676         }
 677 }
 678 # endif /* HAVE_2ARGS_SET_FS_PWD */
 679 #endif /* HAVE_SET_FS_PWD */
 680
 681 int
 682 vn_set_pwd(const char *filename)
 683 {
 684 #if defined(HAVE_2ARGS_SET_FS_PWD) && defined(HAVE_USER_PATH_DIR)
 685         struct path path;
 686 #else
 687         struct nameidata nd;
 688 #endif /* HAVE_2ARGS_SET_FS_PWD */
 689         mm_segment_t saved_fs;
 690         int rc;
 691         SENTRY;
 692
 693         /*
 694          * user_path_dir() and __user_walk() both expect 'filename' to be
 695          * a user space address so we must briefly increase the data segment
 696          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 697          */
 698         saved_fs = get_fs();
 699         set_fs(get_ds());
 700
 701 #ifdef HAVE_2ARGS_SET_FS_PWD
 702 # ifdef HAVE_USER_PATH_DIR
 703         rc = user_path_dir(filename, &path);
 704         if (rc)
 705                 SGOTO(out, rc);
 706
 707         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 708         if (rc)
 709                 SGOTO(dput_and_out, rc);
 710
 711         set_fs_pwd(current->fs, &path);
 712
 713 dput_and_out:
 714         path_put(&path);
 715 # else
 716         rc = __user_walk(filename,
 717                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 718         if (rc)
 719                 SGOTO(out, rc);
 720
 721         rc = vfs_permission(&nd, MAY_EXEC);
 722         if (rc)
 723                 SGOTO(dput_and_out, rc);
 724
 725         set_fs_pwd(current->fs, &nd.path);
 726
 727 dput_and_out:
 728         path_put(&nd.path);
 729 # endif /* HAVE_USER_PATH_DIR */
 730 #else
 731         rc = __user_walk(filename,
 732                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 733         if (rc)
 734                 SGOTO(out, rc);
 735
 736         rc = vfs_permission(&nd, MAY_EXEC);
 737         if (rc)
 738                 SGOTO(dput_and_out, rc);
 739
 740         set_fs_pwd(current->fs, nd.nd_mnt, nd.nd_dentry);
 741
 742 dput_and_out:
 743         vn_path_release(&nd);
 744 #endif /* HAVE_2ARGS_SET_FS_PWD */
 745 out:
 746         set_fs(saved_fs);
 747
 748         SRETURN(-rc);
 749 } /* vn_set_pwd() */
 750 EXPORT_SYMBOL(vn_set_pwd);
 751
 752 static int
 753 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 754 {
 755         struct vnode *vp = buf;
 756
 757         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 758
 759         return (0);
 760 } /* vn_cache_constructor() */
 761
 762 static void
 763 vn_cache_destructor(void *buf, void *cdrarg)
 764 {
 765         struct vnode *vp = buf;
 766
 767         mutex_destroy(&vp->v_lock);
 768 } /* vn_cache_destructor() */
 769
 770 static int
 771 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 772 {
 773         file_t *fp = buf;
 774
 775         atomic_set(&fp->f_ref, 0);
 776         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 777         INIT_LIST_HEAD(&fp->f_list);
 778
 779         return (0);
 780 } /* file_cache_constructor() */
 781
 782 static void
 783 vn_file_cache_destructor(void *buf, void *cdrarg)
 784 {
 785         file_t *fp = buf;
 786
 787         mutex_destroy(&fp->f_lock);
 788 } /* vn_file_cache_destructor() */
 789
 790 int
 791 vn_init(void)
 792 {
 793         SENTRY;
 794         vn_cache = kmem_cache_create("spl_vn_cache",
 795                                      sizeof(struct vnode), 64,
 796                                      vn_cache_constructor,
 797                                      vn_cache_destructor,
 798                                      NULL, NULL, NULL, KMC_KMEM);
 799
 800         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 801                                           sizeof(file_t), 64,
 802                                           vn_file_cache_constructor,
 803                                           vn_file_cache_destructor,
 804                                           NULL, NULL, NULL, KMC_KMEM);
 805         SRETURN(0);
 806 } /* vn_init() */
 807
 808 void
 809 vn_fini(void)
 810 {
 811         file_t *fp, *next_fp;
 812         int leaked = 0;
 813         SENTRY;
 814
 815         spin_lock(&vn_file_lock);
 816
 817         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
 818                 list_del(&fp->f_list);
 819                 releasef_locked(fp);
 820                 leaked++;
 821         }
 822
 823         kmem_cache_destroy(vn_file_cache);
 824         vn_file_cache = NULL;
 825         spin_unlock(&vn_file_lock);
 826
 827         if (leaked > 0)
 828                 SWARN("Warning %d files leaked\n", leaked);
 829
 830         kmem_cache_destroy(vn_cache);
 831
 832         SEXIT;
 833         return;
 834 } /* vn_fini() */