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://zfsonlinux.org/>.
  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/cred.h>
  28 #include <sys/vnode.h>
  29 #include <sys/kmem_cache.h>
  30 #include <linux/falloc.h>
  31 #include <linux/file_compat.h>
  32
  33 vnode_t *rootdir = (vnode_t *)0xabcd1234;
  34 EXPORT_SYMBOL(rootdir);
  35
  36 static spl_kmem_cache_t *vn_cache;
  37 static spl_kmem_cache_t *vn_file_cache;
  38
  39 static DEFINE_SPINLOCK(vn_file_lock);
  40 static LIST_HEAD(vn_file_list);
  41
  42 vtype_t
  43 vn_mode_to_vtype(mode_t mode)
  44 {
  45         if (S_ISREG(mode))
  46                 return VREG;
  47
  48         if (S_ISDIR(mode))
  49                 return VDIR;
  50
  51         if (S_ISCHR(mode))
  52                 return VCHR;
  53
  54         if (S_ISBLK(mode))
  55                 return VBLK;
  56
  57         if (S_ISFIFO(mode))
  58                 return VFIFO;
  59
  60         if (S_ISLNK(mode))
  61                 return VLNK;
  62
  63         if (S_ISSOCK(mode))
  64                 return VSOCK;
  65
  66         if (S_ISCHR(mode))
  67                 return VCHR;
  68
  69         return VNON;
  70 } /* vn_mode_to_vtype() */
  71 EXPORT_SYMBOL(vn_mode_to_vtype);
  72
  73 mode_t
  74 vn_vtype_to_mode(vtype_t vtype)
  75 {
  76         if (vtype == VREG)
  77                 return S_IFREG;
  78
  79         if (vtype == VDIR)
  80                 return S_IFDIR;
  81
  82         if (vtype == VCHR)
  83                 return S_IFCHR;
  84
  85         if (vtype == VBLK)
  86                 return S_IFBLK;
  87
  88         if (vtype == VFIFO)
  89                 return S_IFIFO;
  90
  91         if (vtype == VLNK)
  92                 return S_IFLNK;
  93
  94         if (vtype == VSOCK)
  95                 return S_IFSOCK;
  96
  97         return VNON;
  98 } /* vn_vtype_to_mode() */
  99 EXPORT_SYMBOL(vn_vtype_to_mode);
 100
 101 vnode_t *
 102 vn_alloc(int flag)
 103 {
 104         vnode_t *vp;
 105
 106         vp = kmem_cache_alloc(vn_cache, flag);
 107         if (vp != NULL) {
 108                 vp->v_file = NULL;
 109                 vp->v_type = 0;
 110         }
 111
 112         return (vp);
 113 } /* vn_alloc() */
 114 EXPORT_SYMBOL(vn_alloc);
 115
 116 void
 117 vn_free(vnode_t *vp)
 118 {
 119         kmem_cache_free(vn_cache, vp);
 120 } /* vn_free() */
 121 EXPORT_SYMBOL(vn_free);
 122
 123 int
 124 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 125         vnode_t **vpp, int x1, void *x2)
 126 {
 127         struct file *fp;
 128         struct kstat stat;
 129         int rc, saved_umask = 0;
 130         gfp_t saved_gfp;
 131         vnode_t *vp;
 132
 133         ASSERT(flags & (FWRITE | FREAD));
 134         ASSERT(seg == UIO_SYSSPACE);
 135         ASSERT(vpp);
 136         *vpp = NULL;
 137
 138         if (!(flags & FCREAT) && (flags & FWRITE))
 139                 flags |= FEXCL;
 140
 141         /* Note for filp_open() the two low bits must be remapped to mean:
 142          * 01 - read-only  -> 00 read-only
 143          * 10 - write-only -> 01 write-only
 144          * 11 - read-write -> 10 read-write
 145          */
 146         flags--;
 147
 148         if (flags & FCREAT)
 149                 saved_umask = xchg(&current->fs->umask, 0);
 150
 151         fp = filp_open(path, flags, mode);
 152
 153         if (flags & FCREAT)
 154                 (void)xchg(&current->fs->umask, saved_umask);
 155
 156         if (IS_ERR(fp))
 157                 return (-PTR_ERR(fp));
 158
 159 #ifdef HAVE_2ARGS_VFS_GETATTR
 160         rc = vfs_getattr(&fp->f_path, &stat);
 161 #else
 162         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 163 #endif
 164         if (rc) {
 165                 filp_close(fp, 0);
 166                 return (-rc);
 167         }
 168
 169         vp = vn_alloc(KM_SLEEP);
 170         if (!vp) {
 171                 filp_close(fp, 0);
 172                 return (ENOMEM);
 173         }
 174
 175         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 176         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 177
 178         mutex_enter(&vp->v_lock);
 179         vp->v_type = vn_mode_to_vtype(stat.mode);
 180         vp->v_file = fp;
 181         vp->v_gfp_mask = saved_gfp;
 182         *vpp = vp;
 183         mutex_exit(&vp->v_lock);
 184
 185         return (0);
 186 } /* vn_open() */
 187 EXPORT_SYMBOL(vn_open);
 188
 189 int
 190 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 191           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 192 {
 193         char *realpath;
 194         int len, rc;
 195
 196         ASSERT(vp == rootdir);
 197
 198         len = strlen(path) + 2;
 199         realpath = kmalloc(len, kmem_flags_convert(KM_SLEEP));
 200         if (!realpath)
 201                 return (ENOMEM);
 202
 203         (void)snprintf(realpath, len, "/%s", path);
 204         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 205         kfree(realpath);
 206
 207         return (rc);
 208 } /* vn_openat() */
 209 EXPORT_SYMBOL(vn_openat);
 210
 211 int
 212 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 213         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 214 {
 215         loff_t offset;
 216         mm_segment_t saved_fs;
 217         struct file *fp;
 218         int rc;
 219
 220         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 221         ASSERT(vp);
 222         ASSERT(vp->v_file);
 223         ASSERT(seg == UIO_SYSSPACE);
 224         ASSERT((ioflag & ~FAPPEND) == 0);
 225         ASSERT(x2 == RLIM64_INFINITY);
 226
 227         fp = vp->v_file;
 228
 229         offset = off;
 230         if (ioflag & FAPPEND)
 231                 offset = fp->f_pos;
 232
 233         /* Writable user data segment must be briefly increased for this
 234          * process so we can use the user space read call paths to write
 235          * in to memory allocated by the kernel. */
 236         saved_fs = get_fs();
 237         set_fs(get_ds());
 238
 239         if (uio & UIO_WRITE)
 240                 rc = vfs_write(fp, addr, len, &offset);
 241         else
 242                 rc = vfs_read(fp, addr, len, &offset);
 243
 244         set_fs(saved_fs);
 245         fp->f_pos = offset;
 246
 247         if (rc < 0)
 248                 return (-rc);
 249
 250         if (residp) {
 251                 *residp = len - rc;
 252         } else {
 253                 if (rc != len)
 254                         return (EIO);
 255         }
 256
 257         return (0);
 258 } /* vn_rdwr() */
 259 EXPORT_SYMBOL(vn_rdwr);
 260
 261 int
 262 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 263 {
 264         int rc;
 265
 266         ASSERT(vp);
 267         ASSERT(vp->v_file);
 268
 269         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 270         rc = filp_close(vp->v_file, 0);
 271         vn_free(vp);
 272
 273         return (-rc);
 274 } /* vn_close() */
 275 EXPORT_SYMBOL(vn_close);
 276
 277 /* vn_seek() does not actually seek it only performs bounds checking on the
 278  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 279  * anything more serious. */
 280 int
 281 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 282 {
 283         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 284 }
 285 EXPORT_SYMBOL(vn_seek);
 286
 287 /*
 288  * spl_basename() takes a NULL-terminated string s as input containing a path.
 289  * It returns a char pointer to a string and a length that describe the
 290  * basename of the path. If the basename is not "." or "/", it will be an index
 291  * into the string. While the string should be NULL terminated, the section
 292  * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
 293  * CC0. Anyone wishing to reuse it in another codebase may pick either license.
 294  */
 295 static void
 296 spl_basename(const char *s, const char **str, int *len)
 297 {
 298         size_t i, end;
 299
 300         ASSERT(str);
 301         ASSERT(len);
 302
 303         if (!s || !*s) {
 304                 *str = ".";
 305                 *len = 1;
 306                 return;
 307         }
 308
 309         i = strlen(s) - 1;
 310
 311         while (i && s[i--] == '/');
 312
 313         if (i == 0) {
 314                 *str = "/";
 315                 *len = 1;
 316                 return;
 317         }
 318
 319         end = i;
 320
 321         for (end = i; i; i--) {
 322                 if (s[i] == '/') {
 323                         *str = &s[i+1];
 324                         *len = end - i + 1;
 325                         return;
 326                 }
 327         }
 328
 329         *str = s;
 330         *len = end + 1;
 331 }
 332
 333 static struct dentry *
 334 spl_kern_path_locked(const char *name, struct path *path)
 335 {
 336         struct path parent;
 337         struct dentry *dentry;
 338         const char *basename;
 339         int len;
 340         int rc;
 341
 342         ASSERT(name);
 343         ASSERT(path);
 344
 345         spl_basename(name, &basename, &len);
 346
 347         /* We do not accept "." or ".." */
 348         if (len <= 2 && basename[0] == '.')
 349                 if (len == 1 || basename[1] == '.')
 350                         return (ERR_PTR(-EACCES));
 351
 352         rc = kern_path(name, LOOKUP_PARENT, &parent);
 353         if (rc)
 354                 return (ERR_PTR(rc));
 355
 356         /* use I_MUTEX_PARENT because vfs_unlink needs it */
 357         spl_inode_lock_nested(parent.dentry->d_inode, I_MUTEX_PARENT);
 358
 359         dentry = lookup_one_len(basename, parent.dentry, len);
 360         if (IS_ERR(dentry)) {
 361                 spl_inode_unlock(parent.dentry->d_inode);
 362                 path_put(&parent);
 363         } else {
 364                 *path = parent;
 365         }
 366
 367         return (dentry);
 368 }
 369
 370 /* Based on do_unlinkat() from linux/fs/namei.c */
 371 int
 372 vn_remove(const char *path, uio_seg_t seg, int flags)
 373 {
 374         struct dentry *dentry;
 375         struct path parent;
 376         struct inode *inode = NULL;
 377         int rc = 0;
 378
 379         ASSERT(seg == UIO_SYSSPACE);
 380         ASSERT(flags == RMFILE);
 381
 382         dentry = spl_kern_path_locked(path, &parent);
 383         rc = PTR_ERR(dentry);
 384         if (!IS_ERR(dentry)) {
 385                 if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
 386                         rc = 0;
 387                         goto slashes;
 388                 }
 389
 390                 inode = dentry->d_inode;
 391                 if (inode) {
 392                         atomic_inc(&inode->i_count);
 393                 } else {
 394                         rc = 0;
 395                         goto slashes;
 396                 }
 397
 398 #ifdef HAVE_2ARGS_VFS_UNLINK
 399                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 400 #else
 401                 rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
 402 #endif /* HAVE_2ARGS_VFS_UNLINK */
 403 exit1:
 404                 dput(dentry);
 405         } else {
 406                 return (-rc);
 407         }
 408
 409         spl_inode_unlock(parent.dentry->d_inode);
 410         if (inode)
 411                 iput(inode);    /* truncate the inode here */
 412
 413         path_put(&parent);
 414         return (-rc);
 415
 416 slashes:
 417         rc = !dentry->d_inode ? -ENOENT :
 418             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 419         goto exit1;
 420 } /* vn_remove() */
 421 EXPORT_SYMBOL(vn_remove);
 422
 423 /* Based on do_rename() from linux/fs/namei.c */
 424 int
 425 vn_rename(const char *oldname, const char *newname, int x1)
 426 {
 427         struct dentry *old_dir, *new_dir;
 428         struct dentry *old_dentry, *new_dentry;
 429         struct dentry *trap;
 430         struct path old_parent, new_parent;
 431         int rc = 0;
 432
 433         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 434         if (IS_ERR(old_dentry)) {
 435                 rc = PTR_ERR(old_dentry);
 436                 goto exit;
 437         }
 438
 439         spl_inode_unlock(old_parent.dentry->d_inode);
 440
 441         new_dentry = spl_kern_path_locked(newname, &new_parent);
 442         if (IS_ERR(new_dentry)) {
 443                 rc = PTR_ERR(new_dentry);
 444                 goto exit2;
 445         }
 446
 447         spl_inode_unlock(new_parent.dentry->d_inode);
 448
 449         rc = -EXDEV;
 450         if (old_parent.mnt != new_parent.mnt)
 451                 goto exit3;
 452
 453         old_dir = old_parent.dentry;
 454         new_dir = new_parent.dentry;
 455         trap = lock_rename(new_dir, old_dir);
 456
 457         /* source should not be ancestor of target */
 458         rc = -EINVAL;
 459         if (old_dentry == trap)
 460                 goto exit4;
 461
 462         /* target should not be an ancestor of source */
 463         rc = -ENOTEMPTY;
 464         if (new_dentry == trap)
 465                 goto exit4;
 466
 467         /* source must exist */
 468         rc = -ENOENT;
 469         if (!old_dentry->d_inode)
 470                 goto exit4;
 471
 472         /* unless the source is a directory trailing slashes give -ENOTDIR */
 473         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 474                 rc = -ENOTDIR;
 475                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 476                         goto exit4;
 477                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 478                         goto exit4;
 479         }
 480
 481 #if defined(HAVE_4ARGS_VFS_RENAME)
 482         rc = vfs_rename(old_dir->d_inode, old_dentry,
 483             new_dir->d_inode, new_dentry);
 484 #elif defined(HAVE_5ARGS_VFS_RENAME)
 485         rc = vfs_rename(old_dir->d_inode, old_dentry,
 486             new_dir->d_inode, new_dentry, NULL);
 487 #else
 488         rc = vfs_rename(old_dir->d_inode, old_dentry,
 489             new_dir->d_inode, new_dentry, NULL, 0);
 490 #endif
 491 exit4:
 492         unlock_rename(new_dir, old_dir);
 493 exit3:
 494         dput(new_dentry);
 495         path_put(&new_parent);
 496 exit2:
 497         dput(old_dentry);
 498         path_put(&old_parent);
 499 exit:
 500         return (-rc);
 501 }
 502 EXPORT_SYMBOL(vn_rename);
 503
 504 int
 505 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 506 {
 507         struct file *fp;
 508         struct kstat stat;
 509         int rc;
 510
 511         ASSERT(vp);
 512         ASSERT(vp->v_file);
 513         ASSERT(vap);
 514
 515         fp = vp->v_file;
 516
 517 #ifdef HAVE_2ARGS_VFS_GETATTR
 518         rc = vfs_getattr(&fp->f_path, &stat);
 519 #else
 520         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 521 #endif
 522         if (rc)
 523                 return (-rc);
 524
 525         vap->va_type          = vn_mode_to_vtype(stat.mode);
 526         vap->va_mode          = stat.mode;
 527         vap->va_uid           = KUID_TO_SUID(stat.uid);
 528         vap->va_gid           = KGID_TO_SGID(stat.gid);
 529         vap->va_fsid          = 0;
 530         vap->va_nodeid        = stat.ino;
 531         vap->va_nlink         = stat.nlink;
 532         vap->va_size          = stat.size;
 533         vap->va_blksize       = stat.blksize;
 534         vap->va_atime         = stat.atime;
 535         vap->va_mtime         = stat.mtime;
 536         vap->va_ctime         = stat.ctime;
 537         vap->va_rdev          = stat.rdev;
 538         vap->va_nblocks       = stat.blocks;
 539
 540         return (0);
 541 }
 542 EXPORT_SYMBOL(vn_getattr);
 543
 544 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 545 {
 546         int datasync = 0;
 547         int error;
 548         int fstrans;
 549
 550         ASSERT(vp);
 551         ASSERT(vp->v_file);
 552
 553         if (flags & FDSYNC)
 554                 datasync = 1;
 555
 556         /*
 557          * May enter XFS which generates a warning when PF_FSTRANS is set.
 558          * To avoid this the flag is cleared over vfs_sync() and then reset.
 559          */
 560         fstrans = spl_fstrans_check();
 561         if (fstrans)
 562                 current->flags &= ~(PF_FSTRANS);
 563
 564         error = -spl_filp_fsync(vp->v_file, datasync);
 565         if (fstrans)
 566                 current->flags |= PF_FSTRANS;
 567
 568         return (error);
 569 } /* vn_fsync() */
 570 EXPORT_SYMBOL(vn_fsync);
 571
 572 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 573     offset_t offset, void *x6, void *x7)
 574 {
 575         int error = EOPNOTSUPP;
 576
 577         if (cmd != F_FREESP || bfp->l_whence != 0)
 578                 return (EOPNOTSUPP);
 579
 580         ASSERT(vp);
 581         ASSERT(vp->v_file);
 582         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 583
 584 #ifdef FALLOC_FL_PUNCH_HOLE
 585         /*
 586          * When supported by the underlying file system preferentially
 587          * use the fallocate() callback to preallocate the space.
 588          */
 589         error = -spl_filp_fallocate(vp->v_file,
 590             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 591             bfp->l_start, bfp->l_len);
 592         if (error == 0)
 593                 return (0);
 594 #endif
 595
 596 #ifdef HAVE_INODE_TRUNCATE_RANGE
 597         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 598             vp->v_file->f_dentry->d_inode->i_op &&
 599             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 600                 off_t end = bfp->l_start + bfp->l_len;
 601                 /*
 602                  * Judging from the code in shmem_truncate_range(),
 603                  * it seems the kernel expects the end offset to be
 604                  * inclusive and aligned to the end of a page.
 605                  */
 606                 if (end % PAGE_SIZE != 0) {
 607                         end &= ~(off_t)(PAGE_SIZE - 1);
 608                         if (end <= bfp->l_start)
 609                                 return (0);
 610                 }
 611                 --end;
 612
 613                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 614                         vp->v_file->f_dentry->d_inode,
 615                         bfp->l_start, end
 616                 );
 617                 return (0);
 618         }
 619 #endif
 620
 621         return (error);
 622 }
 623 EXPORT_SYMBOL(vn_space);
 624
 625 /* Function must be called while holding the vn_file_lock */
 626 static file_t *
 627 file_find(int fd, struct task_struct *task)
 628 {
 629         file_t *fp;
 630
 631         ASSERT(spin_is_locked(&vn_file_lock));
 632
 633         list_for_each_entry(fp, &vn_file_list,  f_list) {
 634                 if (fd == fp->f_fd && fp->f_task == task) {
 635                         ASSERT(atomic_read(&fp->f_ref) != 0);
 636                         return fp;
 637                 }
 638         }
 639
 640         return NULL;
 641 } /* file_find() */
 642
 643 file_t *
 644 vn_getf(int fd)
 645 {
 646         struct kstat stat;
 647         struct file *lfp;
 648         file_t *fp;
 649         vnode_t *vp;
 650         int rc = 0;
 651
 652         if (fd < 0)
 653                 return (NULL);
 654
 655         /* Already open just take an extra reference */
 656         spin_lock(&vn_file_lock);
 657
 658         fp = file_find(fd, current);
 659         if (fp) {
 660                 atomic_inc(&fp->f_ref);
 661                 spin_unlock(&vn_file_lock);
 662                 return (fp);
 663         }
 664
 665         spin_unlock(&vn_file_lock);
 666
 667         /* File was not yet opened create the object and setup */
 668         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 669         if (fp == NULL)
 670                 goto out;
 671
 672         mutex_enter(&fp->f_lock);
 673
 674         fp->f_fd = fd;
 675         fp->f_task = current;
 676         fp->f_offset = 0;
 677         atomic_inc(&fp->f_ref);
 678
 679         lfp = fget(fd);
 680         if (lfp == NULL)
 681                 goto out_mutex;
 682
 683         vp = vn_alloc(KM_SLEEP);
 684         if (vp == NULL)
 685                 goto out_fget;
 686
 687 #ifdef HAVE_2ARGS_VFS_GETATTR
 688         rc = vfs_getattr(&lfp->f_path, &stat);
 689 #else
 690         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 691 #endif
 692         if (rc)
 693                 goto out_vnode;
 694
 695         mutex_enter(&vp->v_lock);
 696         vp->v_type = vn_mode_to_vtype(stat.mode);
 697         vp->v_file = lfp;
 698         mutex_exit(&vp->v_lock);
 699
 700         fp->f_vnode = vp;
 701         fp->f_file = lfp;
 702
 703         /* Put it on the tracking list */
 704         spin_lock(&vn_file_lock);
 705         list_add(&fp->f_list, &vn_file_list);
 706         spin_unlock(&vn_file_lock);
 707
 708         mutex_exit(&fp->f_lock);
 709         return (fp);
 710
 711 out_vnode:
 712         vn_free(vp);
 713 out_fget:
 714         fput(lfp);
 715 out_mutex:
 716         mutex_exit(&fp->f_lock);
 717         kmem_cache_free(vn_file_cache, fp);
 718 out:
 719         return (NULL);
 720 } /* getf() */
 721 EXPORT_SYMBOL(getf);
 722
 723 static void releasef_locked(file_t *fp)
 724 {
 725         ASSERT(fp->f_file);
 726         ASSERT(fp->f_vnode);
 727
 728         /* Unlinked from list, no refs, safe to free outside mutex */
 729         fput(fp->f_file);
 730         vn_free(fp->f_vnode);
 731
 732         kmem_cache_free(vn_file_cache, fp);
 733 }
 734
 735 void
 736 vn_releasef(int fd)
 737 {
 738         areleasef(fd, P_FINFO(current));
 739 }
 740 EXPORT_SYMBOL(releasef);
 741
 742 void
 743 vn_areleasef(int fd, uf_info_t *fip)
 744 {
 745         file_t *fp;
 746         struct task_struct *task = (struct task_struct *)fip;
 747
 748         if (fd < 0)
 749                 return;
 750
 751         spin_lock(&vn_file_lock);
 752         fp = file_find(fd, task);
 753         if (fp) {
 754                 atomic_dec(&fp->f_ref);
 755                 if (atomic_read(&fp->f_ref) > 0) {
 756                         spin_unlock(&vn_file_lock);
 757                         return;
 758                 }
 759
 760                 list_del(&fp->f_list);
 761                 releasef_locked(fp);
 762         }
 763         spin_unlock(&vn_file_lock);
 764
 765         return;
 766 } /* releasef() */
 767 EXPORT_SYMBOL(areleasef);
 768
 769
 770 static void
 771 #ifdef HAVE_SET_FS_PWD_WITH_CONST
 772 vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
 773 #else
 774 vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
 775 #endif /* HAVE_SET_FS_PWD_WITH_CONST */
 776 {
 777         struct path old_pwd;
 778
 779 #ifdef HAVE_FS_STRUCT_SPINLOCK
 780         spin_lock(&fs->lock);
 781         old_pwd = fs->pwd;
 782         fs->pwd = *path;
 783         path_get(path);
 784         spin_unlock(&fs->lock);
 785 #else
 786         write_lock(&fs->lock);
 787         old_pwd = fs->pwd;
 788         fs->pwd = *path;
 789         path_get(path);
 790         write_unlock(&fs->lock);
 791 #endif /* HAVE_FS_STRUCT_SPINLOCK */
 792
 793         if (old_pwd.dentry)
 794                 path_put(&old_pwd);
 795 }
 796
 797 int
 798 vn_set_pwd(const char *filename)
 799 {
 800         struct path path;
 801         mm_segment_t saved_fs;
 802         int rc;
 803
 804         /*
 805          * user_path_dir() and __user_walk() both expect 'filename' to be
 806          * a user space address so we must briefly increase the data segment
 807          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 808          */
 809         saved_fs = get_fs();
 810         set_fs(get_ds());
 811
 812         rc = user_path_dir(filename, &path);
 813         if (rc)
 814                 goto out;
 815
 816         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 817         if (rc)
 818                 goto dput_and_out;
 819
 820         vn_set_fs_pwd(current->fs, &path);
 821
 822 dput_and_out:
 823         path_put(&path);
 824 out:
 825         set_fs(saved_fs);
 826
 827         return (-rc);
 828 } /* vn_set_pwd() */
 829 EXPORT_SYMBOL(vn_set_pwd);
 830
 831 static int
 832 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 833 {
 834         struct vnode *vp = buf;
 835
 836         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 837
 838         return (0);
 839 } /* vn_cache_constructor() */
 840
 841 static void
 842 vn_cache_destructor(void *buf, void *cdrarg)
 843 {
 844         struct vnode *vp = buf;
 845
 846         mutex_destroy(&vp->v_lock);
 847 } /* vn_cache_destructor() */
 848
 849 static int
 850 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 851 {
 852         file_t *fp = buf;
 853
 854         atomic_set(&fp->f_ref, 0);
 855         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 856         INIT_LIST_HEAD(&fp->f_list);
 857
 858         return (0);
 859 } /* file_cache_constructor() */
 860
 861 static void
 862 vn_file_cache_destructor(void *buf, void *cdrarg)
 863 {
 864         file_t *fp = buf;
 865
 866         mutex_destroy(&fp->f_lock);
 867 } /* vn_file_cache_destructor() */
 868
 869 int
 870 spl_vn_init(void)
 871 {
 872         vn_cache = kmem_cache_create("spl_vn_cache",
 873                                      sizeof(struct vnode), 64,
 874                                      vn_cache_constructor,
 875                                      vn_cache_destructor,
 876                                      NULL, NULL, NULL, 0);
 877
 878         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 879                                           sizeof(file_t), 64,
 880                                           vn_file_cache_constructor,
 881                                           vn_file_cache_destructor,
 882                                           NULL, NULL, NULL, 0);
 883         return (0);
 884 } /* vn_init() */
 885
 886 void
 887 spl_vn_fini(void)
 888 {
 889         file_t *fp, *next_fp;
 890         int leaked = 0;
 891
 892         spin_lock(&vn_file_lock);
 893
 894         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
 895                 list_del(&fp->f_list);
 896                 releasef_locked(fp);
 897                 leaked++;
 898         }
 899
 900         spin_unlock(&vn_file_lock);
 901
 902         if (leaked > 0)
 903                 printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
 904
 905         kmem_cache_destroy(vn_file_cache);
 906         kmem_cache_destroy(vn_cache);
 907
 908         return;
 909 } /* vn_fini() */