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 #ifdef HAVE_KERN_PATH_PARENT_HEADER
  48 #ifndef HAVE_KERN_PATH_PARENT_SYMBOL
  49 kern_path_parent_t kern_path_parent_fn = SYMBOL_POISON;
  50 EXPORT_SYMBOL(kern_path_parent_fn);
  51 #endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
  52 #endif /* HAVE_KERN_PATH_PARENT_HEADER */
  53
  54 #ifdef HAVE_KERN_PATH_LOCKED
  55 kern_path_locked_t kern_path_locked_fn = SYMBOL_POISON;
  56 #endif /* HAVE_KERN_PATH_LOCKED */
  57
  58 vtype_t
  59 vn_mode_to_vtype(mode_t mode)
  60 {
  61         if (S_ISREG(mode))
  62                 return VREG;
  63
  64         if (S_ISDIR(mode))
  65                 return VDIR;
  66
  67         if (S_ISCHR(mode))
  68                 return VCHR;
  69
  70         if (S_ISBLK(mode))
  71                 return VBLK;
  72
  73         if (S_ISFIFO(mode))
  74                 return VFIFO;
  75
  76         if (S_ISLNK(mode))
  77                 return VLNK;
  78
  79         if (S_ISSOCK(mode))
  80                 return VSOCK;
  81
  82         if (S_ISCHR(mode))
  83                 return VCHR;
  84
  85         return VNON;
  86 } /* vn_mode_to_vtype() */
  87 EXPORT_SYMBOL(vn_mode_to_vtype);
  88
  89 mode_t
  90 vn_vtype_to_mode(vtype_t vtype)
  91 {
  92         if (vtype == VREG)
  93                 return S_IFREG;
  94
  95         if (vtype == VDIR)
  96                 return S_IFDIR;
  97
  98         if (vtype == VCHR)
  99                 return S_IFCHR;
 100
 101         if (vtype == VBLK)
 102                 return S_IFBLK;
 103
 104         if (vtype == VFIFO)
 105                 return S_IFIFO;
 106
 107         if (vtype == VLNK)
 108                 return S_IFLNK;
 109
 110         if (vtype == VSOCK)
 111                 return S_IFSOCK;
 112
 113         return VNON;
 114 } /* vn_vtype_to_mode() */
 115 EXPORT_SYMBOL(vn_vtype_to_mode);
 116
 117 vnode_t *
 118 vn_alloc(int flag)
 119 {
 120         vnode_t *vp;
 121         SENTRY;
 122
 123         vp = kmem_cache_alloc(vn_cache, flag);
 124         if (vp != NULL) {
 125                 vp->v_file = NULL;
 126                 vp->v_type = 0;
 127         }
 128
 129         SRETURN(vp);
 130 } /* vn_alloc() */
 131 EXPORT_SYMBOL(vn_alloc);
 132
 133 void
 134 vn_free(vnode_t *vp)
 135 {
 136         SENTRY;
 137         kmem_cache_free(vn_cache, vp);
 138         SEXIT;
 139 } /* vn_free() */
 140 EXPORT_SYMBOL(vn_free);
 141
 142 int
 143 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 144         vnode_t **vpp, int x1, void *x2)
 145 {
 146         struct file *fp;
 147         struct kstat stat;
 148         int rc, saved_umask = 0;
 149         gfp_t saved_gfp;
 150         vnode_t *vp;
 151         SENTRY;
 152
 153         ASSERT(flags & (FWRITE | FREAD));
 154         ASSERT(seg == UIO_SYSSPACE);
 155         ASSERT(vpp);
 156         *vpp = NULL;
 157
 158         if (!(flags & FCREAT) && (flags & FWRITE))
 159                 flags |= FEXCL;
 160
 161         /* Note for filp_open() the two low bits must be remapped to mean:
 162          * 01 - read-only  -> 00 read-only
 163          * 10 - write-only -> 01 write-only
 164          * 11 - read-write -> 10 read-write
 165          */
 166         flags--;
 167
 168         if (flags & FCREAT)
 169                 saved_umask = xchg(&current->fs->umask, 0);
 170
 171         fp = filp_open(path, flags, mode);
 172
 173         if (flags & FCREAT)
 174                 (void)xchg(&current->fs->umask, saved_umask);
 175
 176         if (IS_ERR(fp))
 177                 SRETURN(-PTR_ERR(fp));
 178
 179 #ifdef HAVE_2ARGS_VFS_GETATTR
 180         rc = vfs_getattr(&fp->f_path, &stat);
 181 #else
 182         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 183 #endif
 184         if (rc) {
 185                 filp_close(fp, 0);
 186                 SRETURN(-rc);
 187         }
 188
 189         vp = vn_alloc(KM_SLEEP);
 190         if (!vp) {
 191                 filp_close(fp, 0);
 192                 SRETURN(ENOMEM);
 193         }
 194
 195         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 196         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 197
 198         mutex_enter(&vp->v_lock);
 199         vp->v_type = vn_mode_to_vtype(stat.mode);
 200         vp->v_file = fp;
 201         vp->v_gfp_mask = saved_gfp;
 202         *vpp = vp;
 203         mutex_exit(&vp->v_lock);
 204
 205         SRETURN(0);
 206 } /* vn_open() */
 207 EXPORT_SYMBOL(vn_open);
 208
 209 int
 210 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 211           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 212 {
 213         char *realpath;
 214         int len, rc;
 215         SENTRY;
 216
 217         ASSERT(vp == rootdir);
 218
 219         len = strlen(path) + 2;
 220         realpath = kmalloc(len, GFP_KERNEL);
 221         if (!realpath)
 222                 SRETURN(ENOMEM);
 223
 224         (void)snprintf(realpath, len, "/%s", path);
 225         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 226         kfree(realpath);
 227
 228         SRETURN(rc);
 229 } /* vn_openat() */
 230 EXPORT_SYMBOL(vn_openat);
 231
 232 int
 233 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 234         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 235 {
 236         loff_t offset;
 237         mm_segment_t saved_fs;
 238         struct file *fp;
 239         int rc;
 240         SENTRY;
 241
 242         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 243         ASSERT(vp);
 244         ASSERT(vp->v_file);
 245         ASSERT(seg == UIO_SYSSPACE);
 246         ASSERT((ioflag & ~FAPPEND) == 0);
 247         ASSERT(x2 == RLIM64_INFINITY);
 248
 249         fp = vp->v_file;
 250
 251         offset = off;
 252         if (ioflag & FAPPEND)
 253                 offset = fp->f_pos;
 254
 255         /* Writable user data segment must be briefly increased for this
 256          * process so we can use the user space read call paths to write
 257          * in to memory allocated by the kernel. */
 258         saved_fs = get_fs();
 259         set_fs(get_ds());
 260
 261         if (uio & UIO_WRITE)
 262                 rc = vfs_write(fp, addr, len, &offset);
 263         else
 264                 rc = vfs_read(fp, addr, len, &offset);
 265
 266         set_fs(saved_fs);
 267         fp->f_pos = offset;
 268
 269         if (rc < 0)
 270                 SRETURN(-rc);
 271
 272         if (residp) {
 273                 *residp = len - rc;
 274         } else {
 275                 if (rc != len)
 276                         SRETURN(EIO);
 277         }
 278
 279         SRETURN(0);
 280 } /* vn_rdwr() */
 281 EXPORT_SYMBOL(vn_rdwr);
 282
 283 int
 284 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 285 {
 286         int rc;
 287         SENTRY;
 288
 289         ASSERT(vp);
 290         ASSERT(vp->v_file);
 291
 292         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 293         rc = filp_close(vp->v_file, 0);
 294         vn_free(vp);
 295
 296         SRETURN(-rc);
 297 } /* vn_close() */
 298 EXPORT_SYMBOL(vn_close);
 299
 300 /* vn_seek() does not actually seek it only performs bounds checking on the
 301  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 302  * anything more serious. */
 303 int
 304 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 305 {
 306         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 307 }
 308 EXPORT_SYMBOL(vn_seek);
 309
 310 #ifdef HAVE_KERN_PATH_LOCKED
 311 /* Based on do_unlinkat() from linux/fs/namei.c */
 312 int
 313 vn_remove(const char *path, uio_seg_t seg, int flags)
 314 {
 315         struct dentry *dentry;
 316         struct path parent;
 317         struct inode *inode = NULL;
 318         int rc = 0;
 319         SENTRY;
 320
 321         ASSERT(seg == UIO_SYSSPACE);
 322         ASSERT(flags == RMFILE);
 323
 324         dentry = spl_kern_path_locked(path, &parent);
 325         rc = PTR_ERR(dentry);
 326         if (!IS_ERR(dentry)) {
 327                 if (parent.dentry->d_name.name[parent.dentry->d_name.len])
 328                         SGOTO(slashes, rc = 0);
 329
 330                 inode = dentry->d_inode;
 331                 if (!inode)
 332                         SGOTO(slashes, rc = 0);
 333
 334                 if (inode)
 335                         ihold(inode);
 336
 337                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 338 exit1:
 339                 dput(dentry);
 340         } else {
 341                 return (-rc);
 342         }
 343
 344         spl_inode_unlock(parent.dentry->d_inode);
 345         if (inode)
 346                 iput(inode);    /* truncate the inode here */
 347
 348         path_put(&parent);
 349         SRETURN(-rc);
 350
 351 slashes:
 352         rc = !dentry->d_inode ? -ENOENT :
 353             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 354         SGOTO(exit1, rc);
 355 } /* vn_remove() */
 356 EXPORT_SYMBOL(vn_remove);
 357
 358 /* Based on do_rename() from linux/fs/namei.c */
 359 int
 360 vn_rename(const char *oldname, const char *newname, int x1)
 361 {
 362         struct dentry *old_dir, *new_dir;
 363         struct dentry *old_dentry, *new_dentry;
 364         struct dentry *trap;
 365         struct path old_parent, new_parent;
 366         int rc = 0;
 367         SENTRY;
 368
 369         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 370         if (IS_ERR(old_dentry))
 371                 SGOTO(exit, rc = PTR_ERR(old_dentry));
 372
 373         spl_inode_unlock(old_parent.dentry->d_inode);
 374
 375         new_dentry = spl_kern_path_locked(newname, &new_parent);
 376         if (IS_ERR(new_dentry))
 377                 SGOTO(exit2, rc = PTR_ERR(new_dentry));
 378
 379         spl_inode_unlock(new_parent.dentry->d_inode);
 380
 381         rc = -EXDEV;
 382         if (old_parent.mnt != new_parent.mnt)
 383                 SGOTO(exit3, rc);
 384
 385         old_dir = old_parent.dentry;
 386         new_dir = new_parent.dentry;
 387         trap = lock_rename(new_dir, old_dir);
 388
 389         /* source should not be ancestor of target */
 390         rc = -EINVAL;
 391         if (old_dentry == trap)
 392                 SGOTO(exit4, rc);
 393
 394         /* target should not be an ancestor of source */
 395         rc = -ENOTEMPTY;
 396         if (new_dentry == trap)
 397                 SGOTO(exit4, rc);
 398
 399         /* source must exist */
 400         rc = -ENOENT;
 401         if (!old_dentry->d_inode)
 402                 SGOTO(exit4, rc);
 403
 404         /* unless the source is a directory trailing slashes give -ENOTDIR */
 405         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 406                 rc = -ENOTDIR;
 407                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 408                         SGOTO(exit4, rc);
 409                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 410                         SGOTO(exit4, rc);
 411         }
 412
 413 #ifdef HAVE_4ARGS_VFS_RENAME
 414         rc = vfs_rename(old_dir->d_inode, old_dentry,
 415                         new_dir->d_inode, new_dentry);
 416 #else
 417         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 418                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 419 #endif /* HAVE_4ARGS_VFS_RENAME */
 420 exit4:
 421         unlock_rename(new_dir, old_dir);
 422 exit3:
 423         dput(new_dentry);
 424         path_put(&new_parent);
 425 exit2:
 426         dput(old_dentry);
 427         path_put(&old_parent);
 428 exit:
 429         SRETURN(-rc);
 430 }
 431 EXPORT_SYMBOL(vn_rename);
 432
 433 #else
 434 static struct dentry *
 435 vn_lookup_hash(struct nameidata *nd)
 436 {
 437         return lookup_one_len((const char *)nd->last.name,
 438                               nd->nd_dentry, nd->last.len);
 439 } /* lookup_hash() */
 440
 441 static void
 442 vn_path_release(struct nameidata *nd)
 443 {
 444         dput(nd->nd_dentry);
 445         mntput(nd->nd_mnt);
 446 }
 447
 448 /* Modified do_unlinkat() from linux/fs/namei.c, only uses exported symbols */
 449 int
 450 vn_remove(const char *path, uio_seg_t seg, int flags)
 451 {
 452         struct dentry *dentry;
 453         struct nameidata nd;
 454         struct inode *inode = NULL;
 455         int rc = 0;
 456         SENTRY;
 457
 458         ASSERT(seg == UIO_SYSSPACE);
 459         ASSERT(flags == RMFILE);
 460
 461         rc = spl_kern_path_parent(path, &nd);
 462         if (rc)
 463                 SGOTO(exit, rc);
 464
 465         rc = -EISDIR;
 466         if (nd.last_type != LAST_NORM)
 467                 SGOTO(exit1, rc);
 468
 469         spl_inode_lock_nested(nd.nd_dentry->d_inode, I_MUTEX_PARENT);
 470         dentry = vn_lookup_hash(&nd);
 471         rc = PTR_ERR(dentry);
 472         if (!IS_ERR(dentry)) {
 473                 /* Why not before? Because we want correct rc value */
 474                 if (nd.last.name[nd.last.len])
 475                         SGOTO(slashes, rc);
 476
 477                 inode = dentry->d_inode;
 478                 if (inode)
 479                         atomic_inc(&inode->i_count);
 480 #ifdef HAVE_2ARGS_VFS_UNLINK
 481                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry);
 482 #else
 483                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry, nd.nd_mnt);
 484 #endif /* HAVE_2ARGS_VFS_UNLINK */
 485 exit2:
 486                 dput(dentry);
 487         }
 488
 489         spl_inode_unlock(nd.nd_dentry->d_inode);
 490         if (inode)
 491                 iput(inode);    /* truncate the inode here */
 492 exit1:
 493         vn_path_release(&nd);
 494 exit:
 495         SRETURN(-rc);
 496
 497 slashes:
 498         rc = !dentry->d_inode ? -ENOENT :
 499                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 500         SGOTO(exit2, rc);
 501 } /* vn_remove() */
 502 EXPORT_SYMBOL(vn_remove);
 503
 504 /* Modified do_rename() from linux/fs/namei.c, only uses exported symbols */
 505 int
 506 vn_rename(const char *oldname, const char *newname, int x1)
 507 {
 508         struct dentry *old_dir, *new_dir;
 509         struct dentry *old_dentry, *new_dentry;
 510         struct dentry *trap;
 511         struct nameidata oldnd, newnd;
 512         int rc = 0;
 513         SENTRY;
 514
 515         rc = spl_kern_path_parent(oldname, &oldnd);
 516         if (rc)
 517                 SGOTO(exit, rc);
 518
 519         rc = spl_kern_path_parent(newname, &newnd);
 520         if (rc)
 521                 SGOTO(exit1, rc);
 522
 523         rc = -EXDEV;
 524         if (oldnd.nd_mnt != newnd.nd_mnt)
 525                 SGOTO(exit2, rc);
 526
 527         old_dir = oldnd.nd_dentry;
 528         rc = -EBUSY;
 529         if (oldnd.last_type != LAST_NORM)
 530                 SGOTO(exit2, rc);
 531
 532         new_dir = newnd.nd_dentry;
 533         if (newnd.last_type != LAST_NORM)
 534                 SGOTO(exit2, rc);
 535
 536         trap = lock_rename(new_dir, old_dir);
 537
 538         old_dentry = vn_lookup_hash(&oldnd);
 539
 540         rc = PTR_ERR(old_dentry);
 541         if (IS_ERR(old_dentry))
 542                 SGOTO(exit3, rc);
 543
 544         /* source must exist */
 545         rc = -ENOENT;
 546         if (!old_dentry->d_inode)
 547                 SGOTO(exit4, rc);
 548
 549         /* unless the source is a directory trailing slashes give -ENOTDIR */
 550         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 551                 rc = -ENOTDIR;
 552                 if (oldnd.last.name[oldnd.last.len])
 553                         SGOTO(exit4, rc);
 554                 if (newnd.last.name[newnd.last.len])
 555                         SGOTO(exit4, rc);
 556         }
 557
 558         /* source should not be ancestor of target */
 559         rc = -EINVAL;
 560         if (old_dentry == trap)
 561                 SGOTO(exit4, rc);
 562
 563         new_dentry = vn_lookup_hash(&newnd);
 564         rc = PTR_ERR(new_dentry);
 565         if (IS_ERR(new_dentry))
 566                 SGOTO(exit4, rc);
 567
 568         /* target should not be an ancestor of source */
 569         rc = -ENOTEMPTY;
 570         if (new_dentry == trap)
 571                 SGOTO(exit5, rc);
 572
 573 #ifdef HAVE_4ARGS_VFS_RENAME
 574         rc = vfs_rename(old_dir->d_inode, old_dentry,
 575                         new_dir->d_inode, new_dentry);
 576 #else
 577         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 578                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 579 #endif /* HAVE_4ARGS_VFS_RENAME */
 580 exit5:
 581         dput(new_dentry);
 582 exit4:
 583         dput(old_dentry);
 584 exit3:
 585         unlock_rename(new_dir, old_dir);
 586 exit2:
 587         vn_path_release(&newnd);
 588 exit1:
 589         vn_path_release(&oldnd);
 590 exit:
 591         SRETURN(-rc);
 592 }
 593 EXPORT_SYMBOL(vn_rename);
 594 #endif /* HAVE_KERN_PATH_LOCKED */
 595
 596 int
 597 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 598 {
 599         struct file *fp;
 600         struct kstat stat;
 601         int rc;
 602         SENTRY;
 603
 604         ASSERT(vp);
 605         ASSERT(vp->v_file);
 606         ASSERT(vap);
 607
 608         fp = vp->v_file;
 609
 610 #ifdef HAVE_2ARGS_VFS_GETATTR
 611         rc = vfs_getattr(&fp->f_path, &stat);
 612 #else
 613         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 614 #endif
 615         if (rc)
 616                 SRETURN(-rc);
 617
 618         vap->va_type          = vn_mode_to_vtype(stat.mode);
 619         vap->va_mode          = stat.mode;
 620         vap->va_uid           = KUID_TO_SUID(stat.uid);
 621         vap->va_gid           = KGID_TO_SGID(stat.gid);
 622         vap->va_fsid          = 0;
 623         vap->va_nodeid        = stat.ino;
 624         vap->va_nlink         = stat.nlink;
 625         vap->va_size          = stat.size;
 626         vap->va_blksize       = stat.blksize;
 627         vap->va_atime         = stat.atime;
 628         vap->va_mtime         = stat.mtime;
 629         vap->va_ctime         = stat.ctime;
 630         vap->va_rdev          = stat.rdev;
 631         vap->va_nblocks       = stat.blocks;
 632
 633         SRETURN(0);
 634 }
 635 EXPORT_SYMBOL(vn_getattr);
 636
 637 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 638 {
 639         int datasync = 0;
 640         SENTRY;
 641
 642         ASSERT(vp);
 643         ASSERT(vp->v_file);
 644
 645         if (flags & FDSYNC)
 646                 datasync = 1;
 647
 648         SRETURN(-spl_filp_fsync(vp->v_file, datasync));
 649 } /* vn_fsync() */
 650 EXPORT_SYMBOL(vn_fsync);
 651
 652 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 653     offset_t offset, void *x6, void *x7)
 654 {
 655         int error = EOPNOTSUPP;
 656         SENTRY;
 657
 658         if (cmd != F_FREESP || bfp->l_whence != 0)
 659                 SRETURN(EOPNOTSUPP);
 660
 661         ASSERT(vp);
 662         ASSERT(vp->v_file);
 663         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 664
 665 #ifdef FALLOC_FL_PUNCH_HOLE
 666         /*
 667          * When supported by the underlying file system preferentially
 668          * use the fallocate() callback to preallocate the space.
 669          */
 670         error = -spl_filp_fallocate(vp->v_file,
 671             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 672             bfp->l_start, bfp->l_len);
 673         if (error == 0)
 674                 SRETURN(0);
 675 #endif
 676
 677 #ifdef HAVE_INODE_TRUNCATE_RANGE
 678         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 679             vp->v_file->f_dentry->d_inode->i_op &&
 680             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 681                 off_t end = bfp->l_start + bfp->l_len;
 682                 /*
 683                  * Judging from the code in shmem_truncate_range(),
 684                  * it seems the kernel expects the end offset to be
 685                  * inclusive and aligned to the end of a page.
 686                  */
 687                 if (end % PAGE_SIZE != 0) {
 688                         end &= ~(off_t)(PAGE_SIZE - 1);
 689                         if (end <= bfp->l_start)
 690                                 SRETURN(0);
 691                 }
 692                 --end;
 693
 694                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 695                         vp->v_file->f_dentry->d_inode,
 696                         bfp->l_start, end
 697                 );
 698                 SRETURN(0);
 699         }
 700 #endif
 701
 702         SRETURN(error);
 703 }
 704 EXPORT_SYMBOL(vn_space);
 705
 706 /* Function must be called while holding the vn_file_lock */
 707 static file_t *
 708 file_find(int fd)
 709 {
 710         file_t *fp;
 711
 712         ASSERT(spin_is_locked(&vn_file_lock));
 713
 714         list_for_each_entry(fp, &vn_file_list,  f_list) {
 715                 if (fd == fp->f_fd && fp->f_task == current) {
 716                         ASSERT(atomic_read(&fp->f_ref) != 0);
 717                         return fp;
 718                 }
 719         }
 720
 721         return NULL;
 722 } /* file_find() */
 723
 724 file_t *
 725 vn_getf(int fd)
 726 {
 727         struct kstat stat;
 728         struct file *lfp;
 729         file_t *fp;
 730         vnode_t *vp;
 731         int rc = 0;
 732         SENTRY;
 733
 734         /* Already open just take an extra reference */
 735         spin_lock(&vn_file_lock);
 736
 737         fp = file_find(fd);
 738         if (fp) {
 739                 atomic_inc(&fp->f_ref);
 740                 spin_unlock(&vn_file_lock);
 741                 SRETURN(fp);
 742         }
 743
 744         spin_unlock(&vn_file_lock);
 745
 746         /* File was not yet opened create the object and setup */
 747         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 748         if (fp == NULL)
 749                 SGOTO(out, rc);
 750
 751         mutex_enter(&fp->f_lock);
 752
 753         fp->f_fd = fd;
 754         fp->f_task = current;
 755         fp->f_offset = 0;
 756         atomic_inc(&fp->f_ref);
 757
 758         lfp = fget(fd);
 759         if (lfp == NULL)
 760                 SGOTO(out_mutex, rc);
 761
 762         vp = vn_alloc(KM_SLEEP);
 763         if (vp == NULL)
 764                 SGOTO(out_fget, rc);
 765
 766 #ifdef HAVE_2ARGS_VFS_GETATTR
 767         rc = vfs_getattr(&lfp->f_path, &stat);
 768 #else
 769         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 770 #endif
 771         if (rc)
 772                 SGOTO(out_vnode, rc);
 773
 774         mutex_enter(&vp->v_lock);
 775         vp->v_type = vn_mode_to_vtype(stat.mode);
 776         vp->v_file = lfp;
 777         mutex_exit(&vp->v_lock);
 778
 779         fp->f_vnode = vp;
 780         fp->f_file = lfp;
 781
 782         /* Put it on the tracking list */
 783         spin_lock(&vn_file_lock);
 784         list_add(&fp->f_list, &vn_file_list);
 785         spin_unlock(&vn_file_lock);
 786
 787         mutex_exit(&fp->f_lock);
 788         SRETURN(fp);
 789
 790 out_vnode:
 791         vn_free(vp);
 792 out_fget:
 793         fput(lfp);
 794 out_mutex:
 795         mutex_exit(&fp->f_lock);
 796         kmem_cache_free(vn_file_cache, fp);
 797 out:
 798         SRETURN(NULL);
 799 } /* getf() */
 800 EXPORT_SYMBOL(getf);
 801
 802 static void releasef_locked(file_t *fp)
 803 {
 804         ASSERT(fp->f_file);
 805         ASSERT(fp->f_vnode);
 806
 807         /* Unlinked from list, no refs, safe to free outside mutex */
 808         fput(fp->f_file);
 809         vn_free(fp->f_vnode);
 810
 811         kmem_cache_free(vn_file_cache, fp);
 812 }
 813
 814 void
 815 vn_releasef(int fd)
 816 {
 817         file_t *fp;
 818         SENTRY;
 819
 820         spin_lock(&vn_file_lock);
 821         fp = file_find(fd);
 822         if (fp) {
 823                 atomic_dec(&fp->f_ref);
 824                 if (atomic_read(&fp->f_ref) > 0) {
 825                         spin_unlock(&vn_file_lock);
 826                         SEXIT;
 827                         return;
 828                 }
 829
 830                 list_del(&fp->f_list);
 831                 releasef_locked(fp);
 832         }
 833         spin_unlock(&vn_file_lock);
 834
 835         SEXIT;
 836         return;
 837 } /* releasef() */
 838 EXPORT_SYMBOL(releasef);
 839
 840 #ifndef HAVE_SET_FS_PWD
 841 void
 842 #  ifdef HAVE_SET_FS_PWD_WITH_CONST
 843 set_fs_pwd(struct fs_struct *fs, const struct path *path)
 844 #  else
 845 set_fs_pwd(struct fs_struct *fs, struct path *path)
 846 #  endif
 847 {
 848         struct path old_pwd;
 849
 850 #  ifdef HAVE_FS_STRUCT_SPINLOCK
 851         spin_lock(&fs->lock);
 852         old_pwd = fs->pwd;
 853         fs->pwd = *path;
 854         path_get(path);
 855         spin_unlock(&fs->lock);
 856 #  else
 857         write_lock(&fs->lock);
 858         old_pwd = fs->pwd;
 859         fs->pwd = *path;
 860         path_get(path);
 861         write_unlock(&fs->lock);
 862 #  endif /* HAVE_FS_STRUCT_SPINLOCK */
 863
 864         if (old_pwd.dentry)
 865                 path_put(&old_pwd);
 866 }
 867 #endif /* HAVE_SET_FS_PWD */
 868
 869 int
 870 vn_set_pwd(const char *filename)
 871 {
 872 #ifdef HAVE_USER_PATH_DIR
 873         struct path path;
 874 #else
 875         struct nameidata nd;
 876 #endif /* HAVE_USER_PATH_DIR */
 877         mm_segment_t saved_fs;
 878         int rc;
 879         SENTRY;
 880
 881         /*
 882          * user_path_dir() and __user_walk() both expect 'filename' to be
 883          * a user space address so we must briefly increase the data segment
 884          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 885          */
 886         saved_fs = get_fs();
 887         set_fs(get_ds());
 888
 889 # ifdef HAVE_USER_PATH_DIR
 890         rc = user_path_dir(filename, &path);
 891         if (rc)
 892                 SGOTO(out, rc);
 893
 894         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 895         if (rc)
 896                 SGOTO(dput_and_out, rc);
 897
 898         set_fs_pwd(current->fs, &path);
 899
 900 dput_and_out:
 901         path_put(&path);
 902 # else
 903         rc = __user_walk(filename,
 904                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 905         if (rc)
 906                 SGOTO(out, rc);
 907
 908         rc = vfs_permission(&nd, MAY_EXEC);
 909         if (rc)
 910                 SGOTO(dput_and_out, rc);
 911
 912         set_fs_pwd(current->fs, &nd.path);
 913
 914 dput_and_out:
 915         path_put(&nd.path);
 916 # endif /* HAVE_USER_PATH_DIR */
 917 out:
 918         set_fs(saved_fs);
 919
 920         SRETURN(-rc);
 921 } /* vn_set_pwd() */
 922 EXPORT_SYMBOL(vn_set_pwd);
 923
 924 static int
 925 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 926 {
 927         struct vnode *vp = buf;
 928
 929         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 930
 931         return (0);
 932 } /* vn_cache_constructor() */
 933
 934 static void
 935 vn_cache_destructor(void *buf, void *cdrarg)
 936 {
 937         struct vnode *vp = buf;
 938
 939         mutex_destroy(&vp->v_lock);
 940 } /* vn_cache_destructor() */
 941
 942 static int
 943 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 944 {
 945         file_t *fp = buf;
 946
 947         atomic_set(&fp->f_ref, 0);
 948         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 949         INIT_LIST_HEAD(&fp->f_list);
 950
 951         return (0);
 952 } /* file_cache_constructor() */
 953
 954 static void
 955 vn_file_cache_destructor(void *buf, void *cdrarg)
 956 {
 957         file_t *fp = buf;
 958
 959         mutex_destroy(&fp->f_lock);
 960 } /* vn_file_cache_destructor() */
 961
 962 int spl_vn_init_kallsyms_lookup(void)
 963 {
 964 #ifdef HAVE_KERN_PATH_PARENT_HEADER
 965 #ifndef HAVE_KERN_PATH_PARENT_SYMBOL
 966         kern_path_parent_fn = (kern_path_parent_t)
 967                 spl_kallsyms_lookup_name("kern_path_parent");
 968         if (!kern_path_parent_fn) {
 969                 printk(KERN_ERR "Error: Unknown symbol kern_path_parent\n");
 970                 return -EFAULT;
 971         }
 972 #endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
 973 #endif /* HAVE_KERN_PATH_PARENT_HEADER */
 974
 975 #ifdef HAVE_KERN_PATH_LOCKED
 976         kern_path_locked_fn = (kern_path_locked_t)
 977                 spl_kallsyms_lookup_name("kern_path_locked");
 978         if (!kern_path_locked_fn) {
 979                 printk(KERN_ERR "Error: Unknown symbol kern_path_locked\n");
 980                 return -EFAULT;
 981         }
 982 #endif
 983
 984         return (0);
 985 }
 986
 987 int
 988 spl_vn_init(void)
 989 {
 990         SENTRY;
 991         vn_cache = kmem_cache_create("spl_vn_cache",
 992                                      sizeof(struct vnode), 64,
 993                                      vn_cache_constructor,
 994                                      vn_cache_destructor,
 995                                      NULL, NULL, NULL, KMC_KMEM);
 996
 997         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 998                                           sizeof(file_t), 64,
 999                                           vn_file_cache_constructor,
1000                                           vn_file_cache_destructor,
1001                                           NULL, NULL, NULL, KMC_KMEM);
1002         SRETURN(0);
1003 } /* vn_init() */
1004
1005 void
1006 spl_vn_fini(void)
1007 {
1008         file_t *fp, *next_fp;
1009         int leaked = 0;
1010         SENTRY;
1011
1012         spin_lock(&vn_file_lock);
1013
1014         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
1015                 list_del(&fp->f_list);
1016                 releasef_locked(fp);
1017                 leaked++;
1018         }
1019
1020         spin_unlock(&vn_file_lock);
1021
1022         if (leaked > 0)
1023                 SWARN("Warning %d files leaked\n", leaked);
1024
1025         kmem_cache_destroy(vn_file_cache);
1026         kmem_cache_destroy(vn_cache);
1027
1028         SEXIT;
1029         return;
1030 } /* vn_fini() */