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