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         spl_inode_lock(parent.dentry->d_inode);
 357
 358         dentry = lookup_one_len(basename, parent.dentry, len);
 359         if (IS_ERR(dentry)) {
 360                 spl_inode_unlock(parent.dentry->d_inode);
 361                 path_put(&parent);
 362         } else {
 363                 *path = parent;
 364         }
 365
 366         return (dentry);
 367 }
 368
 369 /* Based on do_unlinkat() from linux/fs/namei.c */
 370 int
 371 vn_remove(const char *path, uio_seg_t seg, int flags)
 372 {
 373         struct dentry *dentry;
 374         struct path parent;
 375         struct inode *inode = NULL;
 376         int rc = 0;
 377
 378         ASSERT(seg == UIO_SYSSPACE);
 379         ASSERT(flags == RMFILE);
 380
 381         dentry = spl_kern_path_locked(path, &parent);
 382         rc = PTR_ERR(dentry);
 383         if (!IS_ERR(dentry)) {
 384                 if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
 385                         rc = 0;
 386                         goto slashes;
 387                 }
 388
 389                 inode = dentry->d_inode;
 390                 if (inode) {
 391                         atomic_inc(&inode->i_count);
 392                 } else {
 393                         rc = 0;
 394                         goto slashes;
 395                 }
 396
 397 #ifdef HAVE_2ARGS_VFS_UNLINK
 398                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 399 #else
 400                 rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
 401 #endif /* HAVE_2ARGS_VFS_UNLINK */
 402 exit1:
 403                 dput(dentry);
 404         } else {
 405                 return (-rc);
 406         }
 407
 408         spl_inode_unlock(parent.dentry->d_inode);
 409         if (inode)
 410                 iput(inode);    /* truncate the inode here */
 411
 412         path_put(&parent);
 413         return (-rc);
 414
 415 slashes:
 416         rc = !dentry->d_inode ? -ENOENT :
 417             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 418         goto exit1;
 419 } /* vn_remove() */
 420 EXPORT_SYMBOL(vn_remove);
 421
 422 /* Based on do_rename() from linux/fs/namei.c */
 423 int
 424 vn_rename(const char *oldname, const char *newname, int x1)
 425 {
 426         struct dentry *old_dir, *new_dir;
 427         struct dentry *old_dentry, *new_dentry;
 428         struct dentry *trap;
 429         struct path old_parent, new_parent;
 430         int rc = 0;
 431
 432         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 433         if (IS_ERR(old_dentry)) {
 434                 rc = PTR_ERR(old_dentry);
 435                 goto exit;
 436         }
 437
 438         spl_inode_unlock(old_parent.dentry->d_inode);
 439
 440         new_dentry = spl_kern_path_locked(newname, &new_parent);
 441         if (IS_ERR(new_dentry)) {
 442                 rc = PTR_ERR(new_dentry);
 443                 goto exit2;
 444         }
 445
 446         spl_inode_unlock(new_parent.dentry->d_inode);
 447
 448         rc = -EXDEV;
 449         if (old_parent.mnt != new_parent.mnt)
 450                 goto exit3;
 451
 452         old_dir = old_parent.dentry;
 453         new_dir = new_parent.dentry;
 454         trap = lock_rename(new_dir, old_dir);
 455
 456         /* source should not be ancestor of target */
 457         rc = -EINVAL;
 458         if (old_dentry == trap)
 459                 goto exit4;
 460
 461         /* target should not be an ancestor of source */
 462         rc = -ENOTEMPTY;
 463         if (new_dentry == trap)
 464                 goto exit4;
 465
 466         /* source must exist */
 467         rc = -ENOENT;
 468         if (!old_dentry->d_inode)
 469                 goto exit4;
 470
 471         /* unless the source is a directory trailing slashes give -ENOTDIR */
 472         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 473                 rc = -ENOTDIR;
 474                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 475                         goto exit4;
 476                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 477                         goto exit4;
 478         }
 479
 480 #if defined(HAVE_4ARGS_VFS_RENAME)
 481         rc = vfs_rename(old_dir->d_inode, old_dentry,
 482             new_dir->d_inode, new_dentry);
 483 #elif defined(HAVE_5ARGS_VFS_RENAME)
 484         rc = vfs_rename(old_dir->d_inode, old_dentry,
 485             new_dir->d_inode, new_dentry, NULL);
 486 #else
 487         rc = vfs_rename(old_dir->d_inode, old_dentry,
 488             new_dir->d_inode, new_dentry, NULL, 0);
 489 #endif
 490 exit4:
 491         unlock_rename(new_dir, old_dir);
 492 exit3:
 493         dput(new_dentry);
 494         path_put(&new_parent);
 495 exit2:
 496         dput(old_dentry);
 497         path_put(&old_parent);
 498 exit:
 499         return (-rc);
 500 }
 501 EXPORT_SYMBOL(vn_rename);
 502
 503 int
 504 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 505 {
 506         struct file *fp;
 507         struct kstat stat;
 508         int rc;
 509
 510         ASSERT(vp);
 511         ASSERT(vp->v_file);
 512         ASSERT(vap);
 513
 514         fp = vp->v_file;
 515
 516 #ifdef HAVE_2ARGS_VFS_GETATTR
 517         rc = vfs_getattr(&fp->f_path, &stat);
 518 #else
 519         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 520 #endif
 521         if (rc)
 522                 return (-rc);
 523
 524         vap->va_type          = vn_mode_to_vtype(stat.mode);
 525         vap->va_mode          = stat.mode;
 526         vap->va_uid           = KUID_TO_SUID(stat.uid);
 527         vap->va_gid           = KGID_TO_SGID(stat.gid);
 528         vap->va_fsid          = 0;
 529         vap->va_nodeid        = stat.ino;
 530         vap->va_nlink         = stat.nlink;
 531         vap->va_size          = stat.size;
 532         vap->va_blksize       = stat.blksize;
 533         vap->va_atime         = stat.atime;
 534         vap->va_mtime         = stat.mtime;
 535         vap->va_ctime         = stat.ctime;
 536         vap->va_rdev          = stat.rdev;
 537         vap->va_nblocks       = stat.blocks;
 538
 539         return (0);
 540 }
 541 EXPORT_SYMBOL(vn_getattr);
 542
 543 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 544 {
 545         int datasync = 0;
 546         int error;
 547         int fstrans;
 548
 549         ASSERT(vp);
 550         ASSERT(vp->v_file);
 551
 552         if (flags & FDSYNC)
 553                 datasync = 1;
 554
 555         /*
 556          * May enter XFS which generates a warning when PF_FSTRANS is set.
 557          * To avoid this the flag is cleared over vfs_sync() and then reset.
 558          */
 559         fstrans = spl_fstrans_check();
 560         if (fstrans)
 561                 current->flags &= ~(PF_FSTRANS);
 562
 563         error = -spl_filp_fsync(vp->v_file, datasync);
 564         if (fstrans)
 565                 current->flags |= PF_FSTRANS;
 566
 567         return (error);
 568 } /* vn_fsync() */
 569 EXPORT_SYMBOL(vn_fsync);
 570
 571 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 572     offset_t offset, void *x6, void *x7)
 573 {
 574         int error = EOPNOTSUPP;
 575
 576         if (cmd != F_FREESP || bfp->l_whence != 0)
 577                 return (EOPNOTSUPP);
 578
 579         ASSERT(vp);
 580         ASSERT(vp->v_file);
 581         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 582
 583 #ifdef FALLOC_FL_PUNCH_HOLE
 584         /*
 585          * When supported by the underlying file system preferentially
 586          * use the fallocate() callback to preallocate the space.
 587          */
 588         error = -spl_filp_fallocate(vp->v_file,
 589             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 590             bfp->l_start, bfp->l_len);
 591         if (error == 0)
 592                 return (0);
 593 #endif
 594
 595 #ifdef HAVE_INODE_TRUNCATE_RANGE
 596         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 597             vp->v_file->f_dentry->d_inode->i_op &&
 598             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 599                 off_t end = bfp->l_start + bfp->l_len;
 600                 /*
 601                  * Judging from the code in shmem_truncate_range(),
 602                  * it seems the kernel expects the end offset to be
 603                  * inclusive and aligned to the end of a page.
 604                  */
 605                 if (end % PAGE_SIZE != 0) {
 606                         end &= ~(off_t)(PAGE_SIZE - 1);
 607                         if (end <= bfp->l_start)
 608                                 return (0);
 609                 }
 610                 --end;
 611
 612                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 613                         vp->v_file->f_dentry->d_inode,
 614                         bfp->l_start, end
 615                 );
 616                 return (0);
 617         }
 618 #endif
 619
 620         return (error);
 621 }
 622 EXPORT_SYMBOL(vn_space);
 623
 624 /* Function must be called while holding the vn_file_lock */
 625 static file_t *
 626 file_find(int fd)
 627 {
 628         file_t *fp;
 629
 630         ASSERT(spin_is_locked(&vn_file_lock));
 631
 632         list_for_each_entry(fp, &vn_file_list,  f_list) {
 633                 if (fd == fp->f_fd && fp->f_task == current) {
 634                         ASSERT(atomic_read(&fp->f_ref) != 0);
 635                         return fp;
 636                 }
 637         }
 638
 639         return NULL;
 640 } /* file_find() */
 641
 642 file_t *
 643 vn_getf(int fd)
 644 {
 645         struct kstat stat;
 646         struct file *lfp;
 647         file_t *fp;
 648         vnode_t *vp;
 649         int rc = 0;
 650
 651         if (fd < 0)
 652                 return (NULL);
 653
 654         /* Already open just take an extra reference */
 655         spin_lock(&vn_file_lock);
 656
 657         fp = file_find(fd);
 658         if (fp) {
 659                 atomic_inc(&fp->f_ref);
 660                 spin_unlock(&vn_file_lock);
 661                 return (fp);
 662         }
 663
 664         spin_unlock(&vn_file_lock);
 665
 666         /* File was not yet opened create the object and setup */
 667         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 668         if (fp == NULL)
 669                 goto out;
 670
 671         mutex_enter(&fp->f_lock);
 672
 673         fp->f_fd = fd;
 674         fp->f_task = current;
 675         fp->f_offset = 0;
 676         atomic_inc(&fp->f_ref);
 677
 678         lfp = fget(fd);
 679         if (lfp == NULL)
 680                 goto out_mutex;
 681
 682         vp = vn_alloc(KM_SLEEP);
 683         if (vp == NULL)
 684                 goto out_fget;
 685
 686 #ifdef HAVE_2ARGS_VFS_GETATTR
 687         rc = vfs_getattr(&lfp->f_path, &stat);
 688 #else
 689         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 690 #endif
 691         if (rc)
 692                 goto out_vnode;
 693
 694         mutex_enter(&vp->v_lock);
 695         vp->v_type = vn_mode_to_vtype(stat.mode);
 696         vp->v_file = lfp;
 697         mutex_exit(&vp->v_lock);
 698
 699         fp->f_vnode = vp;
 700         fp->f_file = lfp;
 701
 702         /* Put it on the tracking list */
 703         spin_lock(&vn_file_lock);
 704         list_add(&fp->f_list, &vn_file_list);
 705         spin_unlock(&vn_file_lock);
 706
 707         mutex_exit(&fp->f_lock);
 708         return (fp);
 709
 710 out_vnode:
 711         vn_free(vp);
 712 out_fget:
 713         fput(lfp);
 714 out_mutex:
 715         mutex_exit(&fp->f_lock);
 716         kmem_cache_free(vn_file_cache, fp);
 717 out:
 718         return (NULL);
 719 } /* getf() */
 720 EXPORT_SYMBOL(getf);
 721
 722 static void releasef_locked(file_t *fp)
 723 {
 724         ASSERT(fp->f_file);
 725         ASSERT(fp->f_vnode);
 726
 727         /* Unlinked from list, no refs, safe to free outside mutex */
 728         fput(fp->f_file);
 729         vn_free(fp->f_vnode);
 730
 731         kmem_cache_free(vn_file_cache, fp);
 732 }
 733
 734 void
 735 vn_releasef(int fd)
 736 {
 737         file_t *fp;
 738
 739         if (fd < 0)
 740                 return;
 741
 742         spin_lock(&vn_file_lock);
 743         fp = file_find(fd);
 744         if (fp) {
 745                 atomic_dec(&fp->f_ref);
 746                 if (atomic_read(&fp->f_ref) > 0) {
 747                         spin_unlock(&vn_file_lock);
 748                         return;
 749                 }
 750
 751                 list_del(&fp->f_list);
 752                 releasef_locked(fp);
 753         }
 754         spin_unlock(&vn_file_lock);
 755
 756         return;
 757 } /* releasef() */
 758 EXPORT_SYMBOL(releasef);
 759
 760 static void
 761 #ifdef HAVE_SET_FS_PWD_WITH_CONST
 762 vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
 763 #else
 764 vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
 765 #endif /* HAVE_SET_FS_PWD_WITH_CONST */
 766 {
 767         struct path old_pwd;
 768
 769 #ifdef HAVE_FS_STRUCT_SPINLOCK
 770         spin_lock(&fs->lock);
 771         old_pwd = fs->pwd;
 772         fs->pwd = *path;
 773         path_get(path);
 774         spin_unlock(&fs->lock);
 775 #else
 776         write_lock(&fs->lock);
 777         old_pwd = fs->pwd;
 778         fs->pwd = *path;
 779         path_get(path);
 780         write_unlock(&fs->lock);
 781 #endif /* HAVE_FS_STRUCT_SPINLOCK */
 782
 783         if (old_pwd.dentry)
 784                 path_put(&old_pwd);
 785 }
 786
 787 int
 788 vn_set_pwd(const char *filename)
 789 {
 790         struct path path;
 791         mm_segment_t saved_fs;
 792         int rc;
 793
 794         /*
 795          * user_path_dir() and __user_walk() both expect 'filename' to be
 796          * a user space address so we must briefly increase the data segment
 797          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 798          */
 799         saved_fs = get_fs();
 800         set_fs(get_ds());
 801
 802         rc = user_path_dir(filename, &path);
 803         if (rc)
 804                 goto out;
 805
 806         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 807         if (rc)
 808                 goto dput_and_out;
 809
 810         vn_set_fs_pwd(current->fs, &path);
 811
 812 dput_and_out:
 813         path_put(&path);
 814 out:
 815         set_fs(saved_fs);
 816
 817         return (-rc);
 818 } /* vn_set_pwd() */
 819 EXPORT_SYMBOL(vn_set_pwd);
 820
 821 static int
 822 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 823 {
 824         struct vnode *vp = buf;
 825
 826         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 827
 828         return (0);
 829 } /* vn_cache_constructor() */
 830
 831 static void
 832 vn_cache_destructor(void *buf, void *cdrarg)
 833 {
 834         struct vnode *vp = buf;
 835
 836         mutex_destroy(&vp->v_lock);
 837 } /* vn_cache_destructor() */
 838
 839 static int
 840 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 841 {
 842         file_t *fp = buf;
 843
 844         atomic_set(&fp->f_ref, 0);
 845         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 846         INIT_LIST_HEAD(&fp->f_list);
 847
 848         return (0);
 849 } /* file_cache_constructor() */
 850
 851 static void
 852 vn_file_cache_destructor(void *buf, void *cdrarg)
 853 {
 854         file_t *fp = buf;
 855
 856         mutex_destroy(&fp->f_lock);
 857 } /* vn_file_cache_destructor() */
 858
 859 int
 860 spl_vn_init(void)
 861 {
 862         vn_cache = kmem_cache_create("spl_vn_cache",
 863                                      sizeof(struct vnode), 64,
 864                                      vn_cache_constructor,
 865                                      vn_cache_destructor,
 866                                      NULL, NULL, NULL, KMC_KMEM);
 867
 868         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 869                                           sizeof(file_t), 64,
 870                                           vn_file_cache_constructor,
 871                                           vn_file_cache_destructor,
 872                                           NULL, NULL, NULL, KMC_KMEM);
 873         return (0);
 874 } /* vn_init() */
 875
 876 void
 877 spl_vn_fini(void)
 878 {
 879         file_t *fp, *next_fp;
 880         int leaked = 0;
 881
 882         spin_lock(&vn_file_lock);
 883
 884         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
 885                 list_del(&fp->f_list);
 886                 releasef_locked(fp);
 887                 leaked++;
 888         }
 889
 890         spin_unlock(&vn_file_lock);
 891
 892         if (leaked > 0)
 893                 printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
 894
 895         kmem_cache_destroy(vn_file_cache);
 896         kmem_cache_destroy(vn_cache);
 897
 898         return;
 899 } /* vn_fini() */