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         return VNON;
  67 } /* vn_mode_to_vtype() */
  68 EXPORT_SYMBOL(vn_mode_to_vtype);
  69
  70 mode_t
  71 vn_vtype_to_mode(vtype_t vtype)
  72 {
  73         if (vtype == VREG)
  74                 return S_IFREG;
  75
  76         if (vtype == VDIR)
  77                 return S_IFDIR;
  78
  79         if (vtype == VCHR)
  80                 return S_IFCHR;
  81
  82         if (vtype == VBLK)
  83                 return S_IFBLK;
  84
  85         if (vtype == VFIFO)
  86                 return S_IFIFO;
  87
  88         if (vtype == VLNK)
  89                 return S_IFLNK;
  90
  91         if (vtype == VSOCK)
  92                 return S_IFSOCK;
  93
  94         return VNON;
  95 } /* vn_vtype_to_mode() */
  96 EXPORT_SYMBOL(vn_vtype_to_mode);
  97
  98 vnode_t *
  99 vn_alloc(int flag)
 100 {
 101         vnode_t *vp;
 102
 103         vp = kmem_cache_alloc(vn_cache, flag);
 104         if (vp != NULL) {
 105                 vp->v_file = NULL;
 106                 vp->v_type = 0;
 107         }
 108
 109         return (vp);
 110 } /* vn_alloc() */
 111 EXPORT_SYMBOL(vn_alloc);
 112
 113 void
 114 vn_free(vnode_t *vp)
 115 {
 116         kmem_cache_free(vn_cache, vp);
 117 } /* vn_free() */
 118 EXPORT_SYMBOL(vn_free);
 119
 120 int
 121 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 122         vnode_t **vpp, int x1, void *x2)
 123 {
 124         struct file *fp;
 125         struct kstat stat;
 126         int rc, saved_umask = 0;
 127         gfp_t saved_gfp;
 128         vnode_t *vp;
 129
 130         ASSERT(flags & (FWRITE | FREAD));
 131         ASSERT(seg == UIO_SYSSPACE);
 132         ASSERT(vpp);
 133         *vpp = NULL;
 134
 135         if (!(flags & FCREAT) && (flags & FWRITE))
 136                 flags |= FEXCL;
 137
 138         /* Note for filp_open() the two low bits must be remapped to mean:
 139          * 01 - read-only  -> 00 read-only
 140          * 10 - write-only -> 01 write-only
 141          * 11 - read-write -> 10 read-write
 142          */
 143         flags--;
 144
 145         if (flags & FCREAT)
 146                 saved_umask = xchg(&current->fs->umask, 0);
 147
 148         fp = filp_open(path, flags, mode);
 149
 150         if (flags & FCREAT)
 151                 (void)xchg(&current->fs->umask, saved_umask);
 152
 153         if (IS_ERR(fp))
 154                 return (-PTR_ERR(fp));
 155
 156 #ifdef HAVE_2ARGS_VFS_GETATTR
 157         rc = vfs_getattr(&fp->f_path, &stat);
 158 #else
 159         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 160 #endif
 161         if (rc) {
 162                 filp_close(fp, 0);
 163                 return (-rc);
 164         }
 165
 166         vp = vn_alloc(KM_SLEEP);
 167         if (!vp) {
 168                 filp_close(fp, 0);
 169                 return (ENOMEM);
 170         }
 171
 172         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 173         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 174
 175         mutex_enter(&vp->v_lock);
 176         vp->v_type = vn_mode_to_vtype(stat.mode);
 177         vp->v_file = fp;
 178         vp->v_gfp_mask = saved_gfp;
 179         *vpp = vp;
 180         mutex_exit(&vp->v_lock);
 181
 182         return (0);
 183 } /* vn_open() */
 184 EXPORT_SYMBOL(vn_open);
 185
 186 int
 187 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 188           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 189 {
 190         char *realpath;
 191         int len, rc;
 192
 193         ASSERT(vp == rootdir);
 194
 195         len = strlen(path) + 2;
 196         realpath = kmalloc(len, kmem_flags_convert(KM_SLEEP));
 197         if (!realpath)
 198                 return (ENOMEM);
 199
 200         (void)snprintf(realpath, len, "/%s", path);
 201         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 202         kfree(realpath);
 203
 204         return (rc);
 205 } /* vn_openat() */
 206 EXPORT_SYMBOL(vn_openat);
 207
 208 int
 209 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 210         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 211 {
 212         loff_t offset;
 213         mm_segment_t saved_fs;
 214         struct file *fp;
 215         int rc;
 216
 217         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 218         ASSERT(vp);
 219         ASSERT(vp->v_file);
 220         ASSERT(seg == UIO_SYSSPACE);
 221         ASSERT((ioflag & ~FAPPEND) == 0);
 222
 223         fp = vp->v_file;
 224
 225         offset = off;
 226         if (ioflag & FAPPEND)
 227                 offset = fp->f_pos;
 228
 229         /* Writable user data segment must be briefly increased for this
 230          * process so we can use the user space read call paths to write
 231          * in to memory allocated by the kernel. */
 232         saved_fs = get_fs();
 233         set_fs(get_ds());
 234
 235         if (uio & UIO_WRITE)
 236                 rc = vfs_write(fp, addr, len, &offset);
 237         else
 238                 rc = vfs_read(fp, addr, len, &offset);
 239
 240         set_fs(saved_fs);
 241         fp->f_pos = offset;
 242
 243         if (rc < 0)
 244                 return (-rc);
 245
 246         if (residp) {
 247                 *residp = len - rc;
 248         } else {
 249                 if (rc != len)
 250                         return (EIO);
 251         }
 252
 253         return (0);
 254 } /* vn_rdwr() */
 255 EXPORT_SYMBOL(vn_rdwr);
 256
 257 int
 258 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 259 {
 260         int rc;
 261
 262         ASSERT(vp);
 263         ASSERT(vp->v_file);
 264
 265         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 266         rc = filp_close(vp->v_file, 0);
 267         vn_free(vp);
 268
 269         return (-rc);
 270 } /* vn_close() */
 271 EXPORT_SYMBOL(vn_close);
 272
 273 /* vn_seek() does not actually seek it only performs bounds checking on the
 274  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 275  * anything more serious. */
 276 int
 277 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 278 {
 279         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 280 }
 281 EXPORT_SYMBOL(vn_seek);
 282
 283 /*
 284  * spl_basename() takes a NULL-terminated string s as input containing a path.
 285  * It returns a char pointer to a string and a length that describe the
 286  * basename of the path. If the basename is not "." or "/", it will be an index
 287  * into the string. While the string should be NULL terminated, the section
 288  * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
 289  * CC0. Anyone wishing to reuse it in another codebase may pick either license.
 290  */
 291 static void
 292 spl_basename(const char *s, const char **str, int *len)
 293 {
 294         size_t i, end;
 295
 296         ASSERT(str);
 297         ASSERT(len);
 298
 299         if (!s || !*s) {
 300                 *str = ".";
 301                 *len = 1;
 302                 return;
 303         }
 304
 305         i = strlen(s) - 1;
 306
 307         while (i && s[i--] == '/');
 308
 309         if (i == 0) {
 310                 *str = "/";
 311                 *len = 1;
 312                 return;
 313         }
 314
 315         end = i;
 316
 317         for (end = i; i; i--) {
 318                 if (s[i] == '/') {
 319                         *str = &s[i+1];
 320                         *len = end - i + 1;
 321                         return;
 322                 }
 323         }
 324
 325         *str = s;
 326         *len = end + 1;
 327 }
 328
 329 static struct dentry *
 330 spl_kern_path_locked(const char *name, struct path *path)
 331 {
 332         struct path parent;
 333         struct dentry *dentry;
 334         const char *basename;
 335         int len;
 336         int rc;
 337
 338         ASSERT(name);
 339         ASSERT(path);
 340
 341         spl_basename(name, &basename, &len);
 342
 343         /* We do not accept "." or ".." */
 344         if (len <= 2 && basename[0] == '.')
 345                 if (len == 1 || basename[1] == '.')
 346                         return (ERR_PTR(-EACCES));
 347
 348         rc = kern_path(name, LOOKUP_PARENT, &parent);
 349         if (rc)
 350                 return (ERR_PTR(rc));
 351
 352         /* use I_MUTEX_PARENT because vfs_unlink needs it */
 353         spl_inode_lock_nested(parent.dentry->d_inode, I_MUTEX_PARENT);
 354
 355         dentry = lookup_one_len(basename, parent.dentry, len);
 356         if (IS_ERR(dentry)) {
 357                 spl_inode_unlock(parent.dentry->d_inode);
 358                 path_put(&parent);
 359         } else {
 360                 *path = parent;
 361         }
 362
 363         return (dentry);
 364 }
 365
 366 /* Based on do_unlinkat() from linux/fs/namei.c */
 367 int
 368 vn_remove(const char *path, uio_seg_t seg, int flags)
 369 {
 370         struct dentry *dentry;
 371         struct path parent;
 372         struct inode *inode = NULL;
 373         int rc = 0;
 374
 375         ASSERT(seg == UIO_SYSSPACE);
 376         ASSERT(flags == RMFILE);
 377
 378         dentry = spl_kern_path_locked(path, &parent);
 379         rc = PTR_ERR(dentry);
 380         if (!IS_ERR(dentry)) {
 381                 if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
 382                         rc = 0;
 383                         goto slashes;
 384                 }
 385
 386                 inode = dentry->d_inode;
 387                 if (inode) {
 388                         atomic_inc(&inode->i_count);
 389                 } else {
 390                         rc = 0;
 391                         goto slashes;
 392                 }
 393
 394 #ifdef HAVE_2ARGS_VFS_UNLINK
 395                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 396 #else
 397                 rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
 398 #endif /* HAVE_2ARGS_VFS_UNLINK */
 399 exit1:
 400                 dput(dentry);
 401         } else {
 402                 return (-rc);
 403         }
 404
 405         spl_inode_unlock(parent.dentry->d_inode);
 406         if (inode)
 407                 iput(inode);    /* truncate the inode here */
 408
 409         path_put(&parent);
 410         return (-rc);
 411
 412 slashes:
 413         rc = !dentry->d_inode ? -ENOENT :
 414             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 415         goto exit1;
 416 } /* vn_remove() */
 417 EXPORT_SYMBOL(vn_remove);
 418
 419 /* Based on do_rename() from linux/fs/namei.c */
 420 int
 421 vn_rename(const char *oldname, const char *newname, int x1)
 422 {
 423         struct dentry *old_dir, *new_dir;
 424         struct dentry *old_dentry, *new_dentry;
 425         struct dentry *trap;
 426         struct path old_parent, new_parent;
 427         int rc = 0;
 428
 429         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 430         if (IS_ERR(old_dentry)) {
 431                 rc = PTR_ERR(old_dentry);
 432                 goto exit;
 433         }
 434
 435         spl_inode_unlock(old_parent.dentry->d_inode);
 436
 437         new_dentry = spl_kern_path_locked(newname, &new_parent);
 438         if (IS_ERR(new_dentry)) {
 439                 rc = PTR_ERR(new_dentry);
 440                 goto exit2;
 441         }
 442
 443         spl_inode_unlock(new_parent.dentry->d_inode);
 444
 445         rc = -EXDEV;
 446         if (old_parent.mnt != new_parent.mnt)
 447                 goto exit3;
 448
 449         old_dir = old_parent.dentry;
 450         new_dir = new_parent.dentry;
 451         trap = lock_rename(new_dir, old_dir);
 452
 453         /* source should not be ancestor of target */
 454         rc = -EINVAL;
 455         if (old_dentry == trap)
 456                 goto exit4;
 457
 458         /* target should not be an ancestor of source */
 459         rc = -ENOTEMPTY;
 460         if (new_dentry == trap)
 461                 goto exit4;
 462
 463         /* source must exist */
 464         rc = -ENOENT;
 465         if (!old_dentry->d_inode)
 466                 goto exit4;
 467
 468         /* unless the source is a directory trailing slashes give -ENOTDIR */
 469         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 470                 rc = -ENOTDIR;
 471                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 472                         goto exit4;
 473                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 474                         goto exit4;
 475         }
 476
 477 #if defined(HAVE_4ARGS_VFS_RENAME)
 478         rc = vfs_rename(old_dir->d_inode, old_dentry,
 479             new_dir->d_inode, new_dentry);
 480 #elif defined(HAVE_5ARGS_VFS_RENAME)
 481         rc = vfs_rename(old_dir->d_inode, old_dentry,
 482             new_dir->d_inode, new_dentry, NULL);
 483 #else
 484         rc = vfs_rename(old_dir->d_inode, old_dentry,
 485             new_dir->d_inode, new_dentry, NULL, 0);
 486 #endif
 487 exit4:
 488         unlock_rename(new_dir, old_dir);
 489 exit3:
 490         dput(new_dentry);
 491         path_put(&new_parent);
 492 exit2:
 493         dput(old_dentry);
 494         path_put(&old_parent);
 495 exit:
 496         return (-rc);
 497 }
 498 EXPORT_SYMBOL(vn_rename);
 499
 500 int
 501 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 502 {
 503         struct file *fp;
 504         struct kstat stat;
 505         int rc;
 506
 507         ASSERT(vp);
 508         ASSERT(vp->v_file);
 509         ASSERT(vap);
 510
 511         fp = vp->v_file;
 512
 513 #ifdef HAVE_2ARGS_VFS_GETATTR
 514         rc = vfs_getattr(&fp->f_path, &stat);
 515 #else
 516         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 517 #endif
 518         if (rc)
 519                 return (-rc);
 520
 521         vap->va_type          = vn_mode_to_vtype(stat.mode);
 522         vap->va_mode          = stat.mode;
 523         vap->va_uid           = KUID_TO_SUID(stat.uid);
 524         vap->va_gid           = KGID_TO_SGID(stat.gid);
 525         vap->va_fsid          = 0;
 526         vap->va_nodeid        = stat.ino;
 527         vap->va_nlink         = stat.nlink;
 528         vap->va_size          = stat.size;
 529         vap->va_blksize       = stat.blksize;
 530         vap->va_atime         = stat.atime;
 531         vap->va_mtime         = stat.mtime;
 532         vap->va_ctime         = stat.ctime;
 533         vap->va_rdev          = stat.rdev;
 534         vap->va_nblocks       = stat.blocks;
 535
 536         return (0);
 537 }
 538 EXPORT_SYMBOL(vn_getattr);
 539
 540 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 541 {
 542         int datasync = 0;
 543         int error;
 544         int fstrans;
 545
 546         ASSERT(vp);
 547         ASSERT(vp->v_file);
 548
 549         if (flags & FDSYNC)
 550                 datasync = 1;
 551
 552         /*
 553          * May enter XFS which generates a warning when PF_FSTRANS is set.
 554          * To avoid this the flag is cleared over vfs_sync() and then reset.
 555          */
 556         fstrans = spl_fstrans_check();
 557         if (fstrans)
 558                 current->flags &= ~(PF_FSTRANS);
 559
 560         error = -spl_filp_fsync(vp->v_file, datasync);
 561         if (fstrans)
 562                 current->flags |= PF_FSTRANS;
 563
 564         return (error);
 565 } /* vn_fsync() */
 566 EXPORT_SYMBOL(vn_fsync);
 567
 568 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 569     offset_t offset, void *x6, void *x7)
 570 {
 571         int error = EOPNOTSUPP;
 572 #ifdef FALLOC_FL_PUNCH_HOLE
 573         int fstrans;
 574 #endif
 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          * May enter XFS which generates a warning when PF_FSTRANS is set.
 586          * To avoid this the flag is cleared over vfs_sync() and then reset.
 587          */
 588         fstrans = spl_fstrans_check();
 589         if (fstrans)
 590                 current->flags &= ~(PF_FSTRANS);
 591
 592         /*
 593          * When supported by the underlying file system preferentially
 594          * use the fallocate() callback to preallocate the space.
 595          */
 596         error = -spl_filp_fallocate(vp->v_file,
 597             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 598             bfp->l_start, bfp->l_len);
 599
 600         if (fstrans)
 601                 current->flags |= PF_FSTRANS;
 602
 603         if (error == 0)
 604                 return (0);
 605 #endif
 606
 607 #ifdef HAVE_INODE_TRUNCATE_RANGE
 608         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 609             vp->v_file->f_dentry->d_inode->i_op &&
 610             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 611                 off_t end = bfp->l_start + bfp->l_len;
 612                 /*
 613                  * Judging from the code in shmem_truncate_range(),
 614                  * it seems the kernel expects the end offset to be
 615                  * inclusive and aligned to the end of a page.
 616                  */
 617                 if (end % PAGE_SIZE != 0) {
 618                         end &= ~(off_t)(PAGE_SIZE - 1);
 619                         if (end <= bfp->l_start)
 620                                 return (0);
 621                 }
 622                 --end;
 623
 624                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 625                         vp->v_file->f_dentry->d_inode,
 626                         bfp->l_start, end
 627                 );
 628                 return (0);
 629         }
 630 #endif
 631
 632         return (error);
 633 }
 634 EXPORT_SYMBOL(vn_space);
 635
 636 /* Function must be called while holding the vn_file_lock */
 637 static file_t *
 638 file_find(int fd, struct task_struct *task)
 639 {
 640         file_t *fp;
 641
 642         ASSERT(spin_is_locked(&vn_file_lock));
 643
 644         list_for_each_entry(fp, &vn_file_list,  f_list) {
 645                 if (fd == fp->f_fd && fp->f_task == task) {
 646                         ASSERT(atomic_read(&fp->f_ref) != 0);
 647                         return fp;
 648                 }
 649         }
 650
 651         return NULL;
 652 } /* file_find() */
 653
 654 file_t *
 655 vn_getf(int fd)
 656 {
 657         struct kstat stat;
 658         struct file *lfp;
 659         file_t *fp;
 660         vnode_t *vp;
 661         int rc = 0;
 662
 663         if (fd < 0)
 664                 return (NULL);
 665
 666         /* Already open just take an extra reference */
 667         spin_lock(&vn_file_lock);
 668
 669         fp = file_find(fd, current);
 670         if (fp) {
 671                 lfp = fget(fd);
 672                 fput(fp->f_file);
 673                 /*
 674                  * areleasef() can cause us to see a stale reference when
 675                  * userspace has reused a file descriptor before areleasef()
 676                  * has run. fput() the stale reference and replace it. We
 677                  * retain the original reference count such that the concurrent
 678                  * areleasef() will decrement its reference and terminate.
 679                  */
 680                 if (lfp != fp->f_file) {
 681                         fp->f_file = lfp;
 682                         fp->f_vnode->v_file = lfp;
 683                 }
 684                 atomic_inc(&fp->f_ref);
 685                 spin_unlock(&vn_file_lock);
 686                 return (fp);
 687         }
 688
 689         spin_unlock(&vn_file_lock);
 690
 691         /* File was not yet opened create the object and setup */
 692         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 693         if (fp == NULL)
 694                 goto out;
 695
 696         mutex_enter(&fp->f_lock);
 697
 698         fp->f_fd = fd;
 699         fp->f_task = current;
 700         fp->f_offset = 0;
 701         atomic_inc(&fp->f_ref);
 702
 703         lfp = fget(fd);
 704         if (lfp == NULL)
 705                 goto out_mutex;
 706
 707         vp = vn_alloc(KM_SLEEP);
 708         if (vp == NULL)
 709                 goto out_fget;
 710
 711 #ifdef HAVE_2ARGS_VFS_GETATTR
 712         rc = vfs_getattr(&lfp->f_path, &stat);
 713 #else
 714         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 715 #endif
 716         if (rc)
 717                 goto out_vnode;
 718
 719         mutex_enter(&vp->v_lock);
 720         vp->v_type = vn_mode_to_vtype(stat.mode);
 721         vp->v_file = lfp;
 722         mutex_exit(&vp->v_lock);
 723
 724         fp->f_vnode = vp;
 725         fp->f_file = lfp;
 726
 727         /* Put it on the tracking list */
 728         spin_lock(&vn_file_lock);
 729         list_add(&fp->f_list, &vn_file_list);
 730         spin_unlock(&vn_file_lock);
 731
 732         mutex_exit(&fp->f_lock);
 733         return (fp);
 734
 735 out_vnode:
 736         vn_free(vp);
 737 out_fget:
 738         fput(lfp);
 739 out_mutex:
 740         mutex_exit(&fp->f_lock);
 741         kmem_cache_free(vn_file_cache, fp);
 742 out:
 743         return (NULL);
 744 } /* getf() */
 745 EXPORT_SYMBOL(getf);
 746
 747 static void releasef_locked(file_t *fp)
 748 {
 749         ASSERT(fp->f_file);
 750         ASSERT(fp->f_vnode);
 751
 752         /* Unlinked from list, no refs, safe to free outside mutex */
 753         fput(fp->f_file);
 754         vn_free(fp->f_vnode);
 755
 756         kmem_cache_free(vn_file_cache, fp);
 757 }
 758
 759 void
 760 vn_releasef(int fd)
 761 {
 762         areleasef(fd, P_FINFO(current));
 763 }
 764 EXPORT_SYMBOL(releasef);
 765
 766 void
 767 vn_areleasef(int fd, uf_info_t *fip)
 768 {
 769         file_t *fp;
 770         struct task_struct *task = (struct task_struct *)fip;
 771
 772         if (fd < 0)
 773                 return;
 774
 775         spin_lock(&vn_file_lock);
 776         fp = file_find(fd, task);
 777         if (fp) {
 778                 atomic_dec(&fp->f_ref);
 779                 if (atomic_read(&fp->f_ref) > 0) {
 780                         spin_unlock(&vn_file_lock);
 781                         return;
 782                 }
 783
 784                 list_del(&fp->f_list);
 785                 releasef_locked(fp);
 786         }
 787         spin_unlock(&vn_file_lock);
 788
 789         return;
 790 } /* releasef() */
 791 EXPORT_SYMBOL(areleasef);
 792
 793
 794 static void
 795 #ifdef HAVE_SET_FS_PWD_WITH_CONST
 796 vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
 797 #else
 798 vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
 799 #endif /* HAVE_SET_FS_PWD_WITH_CONST */
 800 {
 801         struct path old_pwd;
 802
 803 #ifdef HAVE_FS_STRUCT_SPINLOCK
 804         spin_lock(&fs->lock);
 805         old_pwd = fs->pwd;
 806         fs->pwd = *path;
 807         path_get(path);
 808         spin_unlock(&fs->lock);
 809 #else
 810         write_lock(&fs->lock);
 811         old_pwd = fs->pwd;
 812         fs->pwd = *path;
 813         path_get(path);
 814         write_unlock(&fs->lock);
 815 #endif /* HAVE_FS_STRUCT_SPINLOCK */
 816
 817         if (old_pwd.dentry)
 818                 path_put(&old_pwd);
 819 }
 820
 821 int
 822 vn_set_pwd(const char *filename)
 823 {
 824         struct path path;
 825         mm_segment_t saved_fs;
 826         int rc;
 827
 828         /*
 829          * user_path_dir() and __user_walk() both expect 'filename' to be
 830          * a user space address so we must briefly increase the data segment
 831          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 832          */
 833         saved_fs = get_fs();
 834         set_fs(get_ds());
 835
 836         rc = user_path_dir(filename, &path);
 837         if (rc)
 838                 goto out;
 839
 840         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 841         if (rc)
 842                 goto dput_and_out;
 843
 844         vn_set_fs_pwd(current->fs, &path);
 845
 846 dput_and_out:
 847         path_put(&path);
 848 out:
 849         set_fs(saved_fs);
 850
 851         return (-rc);
 852 } /* vn_set_pwd() */
 853 EXPORT_SYMBOL(vn_set_pwd);
 854
 855 static int
 856 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 857 {
 858         struct vnode *vp = buf;
 859
 860         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 861
 862         return (0);
 863 } /* vn_cache_constructor() */
 864
 865 static void
 866 vn_cache_destructor(void *buf, void *cdrarg)
 867 {
 868         struct vnode *vp = buf;
 869
 870         mutex_destroy(&vp->v_lock);
 871 } /* vn_cache_destructor() */
 872
 873 static int
 874 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 875 {
 876         file_t *fp = buf;
 877
 878         atomic_set(&fp->f_ref, 0);
 879         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 880         INIT_LIST_HEAD(&fp->f_list);
 881
 882         return (0);
 883 } /* file_cache_constructor() */
 884
 885 static void
 886 vn_file_cache_destructor(void *buf, void *cdrarg)
 887 {
 888         file_t *fp = buf;
 889
 890         mutex_destroy(&fp->f_lock);
 891 } /* vn_file_cache_destructor() */
 892
 893 int
 894 spl_vn_init(void)
 895 {
 896         vn_cache = kmem_cache_create("spl_vn_cache",
 897                                      sizeof(struct vnode), 64,
 898                                      vn_cache_constructor,
 899                                      vn_cache_destructor,
 900                                      NULL, NULL, NULL, 0);
 901
 902         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 903                                           sizeof(file_t), 64,
 904                                           vn_file_cache_constructor,
 905                                           vn_file_cache_destructor,
 906                                           NULL, NULL, NULL, 0);
 907         return (0);
 908 } /* vn_init() */
 909
 910 void
 911 spl_vn_fini(void)
 912 {
 913         file_t *fp, *next_fp;
 914         int leaked = 0;
 915
 916         spin_lock(&vn_file_lock);
 917
 918         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
 919                 list_del(&fp->f_list);
 920                 releasef_locked(fp);
 921                 leaked++;
 922         }
 923
 924         spin_unlock(&vn_file_lock);
 925
 926         if (leaked > 0)
 927                 printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
 928
 929         kmem_cache_destroy(vn_file_cache);
 930         kmem_cache_destroy(vn_cache);
 931
 932         return;
 933 } /* vn_fini() */