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://github.com/behlendorf/spl/>.
  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/vnode.h>
  28 #include <linux/falloc.h>
  29 #include <spl-debug.h>
  30
  31 #ifdef SS_DEBUG_SUBSYS
  32 #undef SS_DEBUG_SUBSYS
  33 #endif
  34
  35 #define SS_DEBUG_SUBSYS SS_VNODE
  36
  37 vnode_t *rootdir = (vnode_t *)0xabcd1234;
  38 EXPORT_SYMBOL(rootdir);
  39
  40 static spl_kmem_cache_t *vn_cache;
  41 static spl_kmem_cache_t *vn_file_cache;
  42
  43 static DEFINE_SPINLOCK(vn_file_lock);
  44 static LIST_HEAD(vn_file_list);
  45
  46 #ifdef HAVE_KERN_PATH_PARENT_HEADER
  47 #ifndef HAVE_KERN_PATH_PARENT_SYMBOL
  48 kern_path_parent_t kern_path_parent_fn = SYMBOL_POISON;
  49 EXPORT_SYMBOL(kern_path_parent_fn);
  50 #endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
  51 #endif /* HAVE_KERN_PATH_PARENT_HEADER */
  52
  53 #ifdef HAVE_KERN_PATH_LOCKED
  54 kern_path_locked_t kern_path_locked_fn = SYMBOL_POISON;
  55 #endif /* HAVE_KERN_PATH_LOCKED */
  56
  57 vtype_t
  58 vn_mode_to_vtype(mode_t mode)
  59 {
  60         if (S_ISREG(mode))
  61                 return VREG;
  62
  63         if (S_ISDIR(mode))
  64                 return VDIR;
  65
  66         if (S_ISCHR(mode))
  67                 return VCHR;
  68
  69         if (S_ISBLK(mode))
  70                 return VBLK;
  71
  72         if (S_ISFIFO(mode))
  73                 return VFIFO;
  74
  75         if (S_ISLNK(mode))
  76                 return VLNK;
  77
  78         if (S_ISSOCK(mode))
  79                 return VSOCK;
  80
  81         if (S_ISCHR(mode))
  82                 return VCHR;
  83
  84         return VNON;
  85 } /* vn_mode_to_vtype() */
  86 EXPORT_SYMBOL(vn_mode_to_vtype);
  87
  88 mode_t
  89 vn_vtype_to_mode(vtype_t vtype)
  90 {
  91         if (vtype == VREG)
  92                 return S_IFREG;
  93
  94         if (vtype == VDIR)
  95                 return S_IFDIR;
  96
  97         if (vtype == VCHR)
  98                 return S_IFCHR;
  99
 100         if (vtype == VBLK)
 101                 return S_IFBLK;
 102
 103         if (vtype == VFIFO)
 104                 return S_IFIFO;
 105
 106         if (vtype == VLNK)
 107                 return S_IFLNK;
 108
 109         if (vtype == VSOCK)
 110                 return S_IFSOCK;
 111
 112         return VNON;
 113 } /* vn_vtype_to_mode() */
 114 EXPORT_SYMBOL(vn_vtype_to_mode);
 115
 116 vnode_t *
 117 vn_alloc(int flag)
 118 {
 119         vnode_t *vp;
 120         SENTRY;
 121
 122         vp = kmem_cache_alloc(vn_cache, flag);
 123         if (vp != NULL) {
 124                 vp->v_file = NULL;
 125                 vp->v_type = 0;
 126         }
 127
 128         SRETURN(vp);
 129 } /* vn_alloc() */
 130 EXPORT_SYMBOL(vn_alloc);
 131
 132 void
 133 vn_free(vnode_t *vp)
 134 {
 135         SENTRY;
 136         kmem_cache_free(vn_cache, vp);
 137         SEXIT;
 138 } /* vn_free() */
 139 EXPORT_SYMBOL(vn_free);
 140
 141 int
 142 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 143         vnode_t **vpp, int x1, void *x2)
 144 {
 145         struct file *fp;
 146         struct kstat stat;
 147         int rc, saved_umask = 0;
 148         gfp_t saved_gfp;
 149         vnode_t *vp;
 150         SENTRY;
 151
 152         ASSERT(flags & (FWRITE | FREAD));
 153         ASSERT(seg == UIO_SYSSPACE);
 154         ASSERT(vpp);
 155         *vpp = NULL;
 156
 157         if (!(flags & FCREAT) && (flags & FWRITE))
 158                 flags |= FEXCL;
 159
 160         /* Note for filp_open() the two low bits must be remapped to mean:
 161          * 01 - read-only  -> 00 read-only
 162          * 10 - write-only -> 01 write-only
 163          * 11 - read-write -> 10 read-write
 164          */
 165         flags--;
 166
 167         if (flags & FCREAT)
 168                 saved_umask = xchg(&current->fs->umask, 0);
 169
 170         fp = filp_open(path, flags, mode);
 171
 172         if (flags & FCREAT)
 173                 (void)xchg(&current->fs->umask, saved_umask);
 174
 175         if (IS_ERR(fp))
 176                 SRETURN(-PTR_ERR(fp));
 177
 178         rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
 179         if (rc) {
 180                 filp_close(fp, 0);
 181                 SRETURN(-rc);
 182         }
 183
 184         vp = vn_alloc(KM_SLEEP);
 185         if (!vp) {
 186                 filp_close(fp, 0);
 187                 SRETURN(ENOMEM);
 188         }
 189
 190         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 191         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 192
 193         mutex_enter(&vp->v_lock);
 194         vp->v_type = vn_mode_to_vtype(stat.mode);
 195         vp->v_file = fp;
 196         vp->v_gfp_mask = saved_gfp;
 197         *vpp = vp;
 198         mutex_exit(&vp->v_lock);
 199
 200         SRETURN(0);
 201 } /* vn_open() */
 202 EXPORT_SYMBOL(vn_open);
 203
 204 int
 205 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 206           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 207 {
 208         char *realpath;
 209         int len, rc;
 210         SENTRY;
 211
 212         ASSERT(vp == rootdir);
 213
 214         len = strlen(path) + 2;
 215         realpath = kmalloc(len, GFP_KERNEL);
 216         if (!realpath)
 217                 SRETURN(ENOMEM);
 218
 219         (void)snprintf(realpath, len, "/%s", path);
 220         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 221         kfree(realpath);
 222
 223         SRETURN(rc);
 224 } /* vn_openat() */
 225 EXPORT_SYMBOL(vn_openat);
 226
 227 int
 228 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 229         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 230 {
 231         loff_t offset;
 232         mm_segment_t saved_fs;
 233         struct file *fp;
 234         int rc;
 235         SENTRY;
 236
 237         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 238         ASSERT(vp);
 239         ASSERT(vp->v_file);
 240         ASSERT(seg == UIO_SYSSPACE);
 241         ASSERT((ioflag & ~FAPPEND) == 0);
 242         ASSERT(x2 == RLIM64_INFINITY);
 243
 244         fp = vp->v_file;
 245
 246         offset = off;
 247         if (ioflag & FAPPEND)
 248                 offset = fp->f_pos;
 249
 250         /* Writable user data segment must be briefly increased for this
 251          * process so we can use the user space read call paths to write
 252          * in to memory allocated by the kernel. */
 253         saved_fs = get_fs();
 254         set_fs(get_ds());
 255
 256         if (uio & UIO_WRITE)
 257                 rc = vfs_write(fp, addr, len, &offset);
 258         else
 259                 rc = vfs_read(fp, addr, len, &offset);
 260
 261         set_fs(saved_fs);
 262         fp->f_pos = offset;
 263
 264         if (rc < 0)
 265                 SRETURN(-rc);
 266
 267         if (residp) {
 268                 *residp = len - rc;
 269         } else {
 270                 if (rc != len)
 271                         SRETURN(EIO);
 272         }
 273
 274         SRETURN(0);
 275 } /* vn_rdwr() */
 276 EXPORT_SYMBOL(vn_rdwr);
 277
 278 int
 279 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 280 {
 281         int rc;
 282         SENTRY;
 283
 284         ASSERT(vp);
 285         ASSERT(vp->v_file);
 286
 287         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 288         rc = filp_close(vp->v_file, 0);
 289         vn_free(vp);
 290
 291         SRETURN(-rc);
 292 } /* vn_close() */
 293 EXPORT_SYMBOL(vn_close);
 294
 295 /* vn_seek() does not actually seek it only performs bounds checking on the
 296  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 297  * anything more serious. */
 298 int
 299 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 300 {
 301         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 302 }
 303 EXPORT_SYMBOL(vn_seek);
 304
 305 #ifdef HAVE_KERN_PATH_LOCKED
 306 /* Based on do_unlinkat() from linux/fs/namei.c */
 307 int
 308 vn_remove(const char *path, uio_seg_t seg, int flags)
 309 {
 310         struct dentry *dentry;
 311         struct path parent;
 312         struct inode *inode = NULL;
 313         int rc = 0;
 314         SENTRY;
 315
 316         ASSERT(seg == UIO_SYSSPACE);
 317         ASSERT(flags == RMFILE);
 318
 319         dentry = spl_kern_path_locked(path, &parent);
 320         rc = PTR_ERR(dentry);
 321         if (!IS_ERR(dentry)) {
 322                 if (parent.dentry->d_name.name[parent.dentry->d_name.len])
 323                         SGOTO(slashes, rc = 0);
 324
 325                 inode = dentry->d_inode;
 326                 if (!inode)
 327                         SGOTO(slashes, rc = 0);
 328
 329                 if (inode)
 330                         ihold(inode);
 331
 332                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 333 exit1:
 334                 dput(dentry);
 335         }
 336
 337         spl_inode_unlock(parent.dentry->d_inode);
 338         if (inode)
 339                 iput(inode);    /* truncate the inode here */
 340
 341         path_put(&parent);
 342         SRETURN(-rc);
 343
 344 slashes:
 345         rc = !dentry->d_inode ? -ENOENT :
 346             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 347         SGOTO(exit1, rc);
 348 } /* vn_remove() */
 349 EXPORT_SYMBOL(vn_remove);
 350
 351 /* Based on do_rename() from linux/fs/namei.c */
 352 int
 353 vn_rename(const char *oldname, const char *newname, int x1)
 354 {
 355         struct dentry *old_dir, *new_dir;
 356         struct dentry *old_dentry, *new_dentry;
 357         struct dentry *trap;
 358         struct path old_parent, new_parent;
 359         int rc = 0;
 360         SENTRY;
 361
 362         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 363         if (IS_ERR(old_dentry))
 364                 SGOTO(exit, rc = PTR_ERR(old_dentry));
 365
 366         spl_inode_unlock(old_parent.dentry->d_inode);
 367
 368         new_dentry = spl_kern_path_locked(newname, &new_parent);
 369         if (IS_ERR(new_dentry))
 370                 SGOTO(exit2, rc = PTR_ERR(new_dentry));
 371
 372         spl_inode_unlock(new_parent.dentry->d_inode);
 373
 374         rc = -EXDEV;
 375         if (old_parent.mnt != new_parent.mnt)
 376                 SGOTO(exit3, rc);
 377
 378         old_dir = old_parent.dentry;
 379         new_dir = new_parent.dentry;
 380         trap = lock_rename(new_dir, old_dir);
 381
 382         /* source should not be ancestor of target */
 383         rc = -EINVAL;
 384         if (old_dentry == trap)
 385                 SGOTO(exit4, rc);
 386
 387         /* target should not be an ancestor of source */
 388         rc = -ENOTEMPTY;
 389         if (new_dentry == trap)
 390                 SGOTO(exit4, rc);
 391
 392         /* source must exist */
 393         rc = -ENOENT;
 394         if (!old_dentry->d_inode)
 395                 SGOTO(exit4, rc);
 396
 397         /* unless the source is a directory trailing slashes give -ENOTDIR */
 398         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 399                 rc = -ENOTDIR;
 400                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 401                         SGOTO(exit4, rc);
 402                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 403                         SGOTO(exit4, rc);
 404         }
 405
 406 #ifdef HAVE_4ARGS_VFS_RENAME
 407         rc = vfs_rename(old_dir->d_inode, old_dentry,
 408                         new_dir->d_inode, new_dentry);
 409 #else
 410         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 411                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 412 #endif /* HAVE_4ARGS_VFS_RENAME */
 413 exit4:
 414         unlock_rename(new_dir, old_dir);
 415 exit3:
 416         dput(new_dentry);
 417         path_put(&new_parent);
 418 exit2:
 419         dput(old_dentry);
 420         path_put(&old_parent);
 421 exit:
 422         SRETURN(-rc);
 423 }
 424 EXPORT_SYMBOL(vn_rename);
 425
 426 #else
 427 static struct dentry *
 428 vn_lookup_hash(struct nameidata *nd)
 429 {
 430         return lookup_one_len((const char *)nd->last.name,
 431                               nd->nd_dentry, nd->last.len);
 432 } /* lookup_hash() */
 433
 434 static void
 435 vn_path_release(struct nameidata *nd)
 436 {
 437         dput(nd->nd_dentry);
 438         mntput(nd->nd_mnt);
 439 }
 440
 441 /* Modified do_unlinkat() from linux/fs/namei.c, only uses exported symbols */
 442 int
 443 vn_remove(const char *path, uio_seg_t seg, int flags)
 444 {
 445         struct dentry *dentry;
 446         struct nameidata nd;
 447         struct inode *inode = NULL;
 448         int rc = 0;
 449         SENTRY;
 450
 451         ASSERT(seg == UIO_SYSSPACE);
 452         ASSERT(flags == RMFILE);
 453
 454         rc = spl_kern_path_parent(path, &nd);
 455         if (rc)
 456                 SGOTO(exit, rc);
 457
 458         rc = -EISDIR;
 459         if (nd.last_type != LAST_NORM)
 460                 SGOTO(exit1, rc);
 461
 462         spl_inode_lock_nested(nd.nd_dentry->d_inode, I_MUTEX_PARENT);
 463         dentry = vn_lookup_hash(&nd);
 464         rc = PTR_ERR(dentry);
 465         if (!IS_ERR(dentry)) {
 466                 /* Why not before? Because we want correct rc value */
 467                 if (nd.last.name[nd.last.len])
 468                         SGOTO(slashes, rc);
 469
 470                 inode = dentry->d_inode;
 471                 if (inode)
 472                         atomic_inc(&inode->i_count);
 473 #ifdef HAVE_2ARGS_VFS_UNLINK
 474                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry);
 475 #else
 476                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry, nd.nd_mnt);
 477 #endif /* HAVE_2ARGS_VFS_UNLINK */
 478 exit2:
 479                 dput(dentry);
 480         }
 481
 482         spl_inode_unlock(nd.nd_dentry->d_inode);
 483         if (inode)
 484                 iput(inode);    /* truncate the inode here */
 485 exit1:
 486         vn_path_release(&nd);
 487 exit:
 488         SRETURN(-rc);
 489
 490 slashes:
 491         rc = !dentry->d_inode ? -ENOENT :
 492                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 493         SGOTO(exit2, rc);
 494 } /* vn_remove() */
 495 EXPORT_SYMBOL(vn_remove);
 496
 497 /* Modified do_rename() from linux/fs/namei.c, only uses exported symbols */
 498 int
 499 vn_rename(const char *oldname, const char *newname, int x1)
 500 {
 501         struct dentry *old_dir, *new_dir;
 502         struct dentry *old_dentry, *new_dentry;
 503         struct dentry *trap;
 504         struct nameidata oldnd, newnd;
 505         int rc = 0;
 506         SENTRY;
 507
 508         rc = spl_kern_path_parent(oldname, &oldnd);
 509         if (rc)
 510                 SGOTO(exit, rc);
 511
 512         rc = spl_kern_path_parent(newname, &newnd);
 513         if (rc)
 514                 SGOTO(exit1, rc);
 515
 516         rc = -EXDEV;
 517         if (oldnd.nd_mnt != newnd.nd_mnt)
 518                 SGOTO(exit2, rc);
 519
 520         old_dir = oldnd.nd_dentry;
 521         rc = -EBUSY;
 522         if (oldnd.last_type != LAST_NORM)
 523                 SGOTO(exit2, rc);
 524
 525         new_dir = newnd.nd_dentry;
 526         if (newnd.last_type != LAST_NORM)
 527                 SGOTO(exit2, rc);
 528
 529         trap = lock_rename(new_dir, old_dir);
 530
 531         old_dentry = vn_lookup_hash(&oldnd);
 532
 533         rc = PTR_ERR(old_dentry);
 534         if (IS_ERR(old_dentry))
 535                 SGOTO(exit3, rc);
 536
 537         /* source must exist */
 538         rc = -ENOENT;
 539         if (!old_dentry->d_inode)
 540                 SGOTO(exit4, rc);
 541
 542         /* unless the source is a directory trailing slashes give -ENOTDIR */
 543         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 544                 rc = -ENOTDIR;
 545                 if (oldnd.last.name[oldnd.last.len])
 546                         SGOTO(exit4, rc);
 547                 if (newnd.last.name[newnd.last.len])
 548                         SGOTO(exit4, rc);
 549         }
 550
 551         /* source should not be ancestor of target */
 552         rc = -EINVAL;
 553         if (old_dentry == trap)
 554                 SGOTO(exit4, rc);
 555
 556         new_dentry = vn_lookup_hash(&newnd);
 557         rc = PTR_ERR(new_dentry);
 558         if (IS_ERR(new_dentry))
 559                 SGOTO(exit4, rc);
 560
 561         /* target should not be an ancestor of source */
 562         rc = -ENOTEMPTY;
 563         if (new_dentry == trap)
 564                 SGOTO(exit5, rc);
 565
 566 #ifdef HAVE_4ARGS_VFS_RENAME
 567         rc = vfs_rename(old_dir->d_inode, old_dentry,
 568                         new_dir->d_inode, new_dentry);
 569 #else
 570         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 571                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 572 #endif /* HAVE_4ARGS_VFS_RENAME */
 573 exit5:
 574         dput(new_dentry);
 575 exit4:
 576         dput(old_dentry);
 577 exit3:
 578         unlock_rename(new_dir, old_dir);
 579 exit2:
 580         vn_path_release(&newnd);
 581 exit1:
 582         vn_path_release(&oldnd);
 583 exit:
 584         SRETURN(-rc);
 585 }
 586 EXPORT_SYMBOL(vn_rename);
 587 #endif /* HAVE_KERN_PATH_LOCKED */
 588
 589 int
 590 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 591 {
 592         struct file *fp;
 593         struct kstat stat;
 594         int rc;
 595         SENTRY;
 596
 597         ASSERT(vp);
 598         ASSERT(vp->v_file);
 599         ASSERT(vap);
 600
 601         fp = vp->v_file;
 602
 603         rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
 604         if (rc)
 605                 SRETURN(-rc);
 606
 607         vap->va_type          = vn_mode_to_vtype(stat.mode);
 608         vap->va_mode          = stat.mode;
 609         vap->va_uid           = stat.uid;
 610         vap->va_gid           = stat.gid;
 611         vap->va_fsid          = 0;
 612         vap->va_nodeid        = stat.ino;
 613         vap->va_nlink         = stat.nlink;
 614         vap->va_size          = stat.size;
 615         vap->va_blksize       = stat.blksize;
 616         vap->va_atime         = stat.atime;
 617         vap->va_mtime         = stat.mtime;
 618         vap->va_ctime         = stat.ctime;
 619         vap->va_rdev          = stat.rdev;
 620         vap->va_nblocks       = stat.blocks;
 621
 622         SRETURN(0);
 623 }
 624 EXPORT_SYMBOL(vn_getattr);
 625
 626 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 627 {
 628         int datasync = 0;
 629         SENTRY;
 630
 631         ASSERT(vp);
 632         ASSERT(vp->v_file);
 633
 634         if (flags & FDSYNC)
 635                 datasync = 1;
 636
 637         SRETURN(-spl_filp_fsync(vp->v_file, datasync));
 638 } /* vn_fsync() */
 639 EXPORT_SYMBOL(vn_fsync);
 640
 641 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 642     offset_t offset, void *x6, void *x7)
 643 {
 644         int error = EOPNOTSUPP;
 645         SENTRY;
 646
 647         if (cmd != F_FREESP || bfp->l_whence != 0)
 648                 SRETURN(EOPNOTSUPP);
 649
 650         ASSERT(vp);
 651         ASSERT(vp->v_file);
 652         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 653
 654 #ifdef FALLOC_FL_PUNCH_HOLE
 655         if (vp->v_file->f_op->fallocate) {
 656                 error = -vp->v_file->f_op->fallocate(vp->v_file,
 657                     FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 658                     bfp->l_start, bfp->l_len);
 659                 if (!error)
 660                         SRETURN(0);
 661         }
 662 #endif
 663
 664 #ifdef HAVE_INODE_TRUNCATE_RANGE
 665         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 666             vp->v_file->f_dentry->d_inode->i_op &&
 667             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 668                 off_t end = bfp->l_start + bfp->l_len;
 669                 /*
 670                  * Judging from the code in shmem_truncate_range(),
 671                  * it seems the kernel expects the end offset to be
 672                  * inclusive and aligned to the end of a page.
 673                  */
 674                 if (end % PAGE_SIZE != 0) {
 675                         end &= ~(off_t)(PAGE_SIZE - 1);
 676                         if (end <= bfp->l_start)
 677                                 SRETURN(0);
 678                 }
 679                 --end;
 680
 681                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 682                         vp->v_file->f_dentry->d_inode,
 683                         bfp->l_start, end
 684                 );
 685                 SRETURN(0);
 686         }
 687 #endif
 688
 689         SRETURN(error);
 690 }
 691 EXPORT_SYMBOL(vn_space);
 692
 693 /* Function must be called while holding the vn_file_lock */
 694 static file_t *
 695 file_find(int fd)
 696 {
 697         file_t *fp;
 698
 699         ASSERT(spin_is_locked(&vn_file_lock));
 700
 701         list_for_each_entry(fp, &vn_file_list,  f_list) {
 702                 if (fd == fp->f_fd && fp->f_task == current) {
 703                         ASSERT(atomic_read(&fp->f_ref) != 0);
 704                         return fp;
 705                 }
 706         }
 707
 708         return NULL;
 709 } /* file_find() */
 710
 711 file_t *
 712 vn_getf(int fd)
 713 {
 714         struct kstat stat;
 715         struct file *lfp;
 716         file_t *fp;
 717         vnode_t *vp;
 718         int rc = 0;
 719         SENTRY;
 720
 721         /* Already open just take an extra reference */
 722         spin_lock(&vn_file_lock);
 723
 724         fp = file_find(fd);
 725         if (fp) {
 726                 atomic_inc(&fp->f_ref);
 727                 spin_unlock(&vn_file_lock);
 728                 SRETURN(fp);
 729         }
 730
 731         spin_unlock(&vn_file_lock);
 732
 733         /* File was not yet opened create the object and setup */
 734         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 735         if (fp == NULL)
 736                 SGOTO(out, rc);
 737
 738         mutex_enter(&fp->f_lock);
 739
 740         fp->f_fd = fd;
 741         fp->f_task = current;
 742         fp->f_offset = 0;
 743         atomic_inc(&fp->f_ref);
 744
 745         lfp = fget(fd);
 746         if (lfp == NULL)
 747                 SGOTO(out_mutex, rc);
 748
 749         vp = vn_alloc(KM_SLEEP);
 750         if (vp == NULL)
 751                 SGOTO(out_fget, rc);
 752
 753         if (vfs_getattr(lfp->f_vfsmnt, lfp->f_dentry, &stat))
 754                 SGOTO(out_vnode, rc);
 755
 756         mutex_enter(&vp->v_lock);
 757         vp->v_type = vn_mode_to_vtype(stat.mode);
 758         vp->v_file = lfp;
 759         mutex_exit(&vp->v_lock);
 760
 761         fp->f_vnode = vp;
 762         fp->f_file = lfp;
 763
 764         /* Put it on the tracking list */
 765         spin_lock(&vn_file_lock);
 766         list_add(&fp->f_list, &vn_file_list);
 767         spin_unlock(&vn_file_lock);
 768
 769         mutex_exit(&fp->f_lock);
 770         SRETURN(fp);
 771
 772 out_vnode:
 773         vn_free(vp);
 774 out_fget:
 775         fput(lfp);
 776 out_mutex:
 777         mutex_exit(&fp->f_lock);
 778         kmem_cache_free(vn_file_cache, fp);
 779 out:
 780         SRETURN(NULL);
 781 } /* getf() */
 782 EXPORT_SYMBOL(getf);
 783
 784 static void releasef_locked(file_t *fp)
 785 {
 786         ASSERT(fp->f_file);
 787         ASSERT(fp->f_vnode);
 788
 789         /* Unlinked from list, no refs, safe to free outside mutex */
 790         fput(fp->f_file);
 791         vn_free(fp->f_vnode);
 792
 793         kmem_cache_free(vn_file_cache, fp);
 794 }
 795
 796 void
 797 vn_releasef(int fd)
 798 {
 799         file_t *fp;
 800         SENTRY;
 801
 802         spin_lock(&vn_file_lock);
 803         fp = file_find(fd);
 804         if (fp) {
 805                 atomic_dec(&fp->f_ref);
 806                 if (atomic_read(&fp->f_ref) > 0) {
 807                         spin_unlock(&vn_file_lock);
 808                         SEXIT;
 809                         return;
 810                 }
 811
 812                 list_del(&fp->f_list);
 813                 releasef_locked(fp);
 814         }
 815         spin_unlock(&vn_file_lock);
 816
 817         SEXIT;
 818         return;
 819 } /* releasef() */
 820 EXPORT_SYMBOL(releasef);
 821
 822 #ifndef HAVE_SET_FS_PWD
 823 # ifdef HAVE_2ARGS_SET_FS_PWD
 824 /* Used from 2.6.25 - 2.6.31+ */
 825 void
 826 set_fs_pwd(struct fs_struct *fs, struct path *path)
 827 {
 828         struct path old_pwd;
 829
 830 #  ifdef HAVE_FS_STRUCT_SPINLOCK
 831         spin_lock(&fs->lock);
 832         old_pwd = fs->pwd;
 833         fs->pwd = *path;
 834         path_get(path);
 835         spin_unlock(&fs->lock);
 836 #  else
 837         write_lock(&fs->lock);
 838         old_pwd = fs->pwd;
 839         fs->pwd = *path;
 840         path_get(path);
 841         write_unlock(&fs->lock);
 842 #  endif /* HAVE_FS_STRUCT_SPINLOCK */
 843
 844         if (old_pwd.dentry)
 845                 path_put(&old_pwd);
 846 }
 847 # else
 848 /* Used from 2.6.11 - 2.6.24 */
 849 void
 850 set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, struct dentry *dentry)
 851 {
 852         struct dentry *old_pwd;
 853         struct vfsmount *old_pwdmnt;
 854
 855         write_lock(&fs->lock);
 856         old_pwd = fs->pwd;
 857         old_pwdmnt = fs->pwdmnt;
 858         fs->pwdmnt = mntget(mnt);
 859         fs->pwd = dget(dentry);
 860         write_unlock(&fs->lock);
 861
 862         if (old_pwd) {
 863                 dput(old_pwd);
 864                 mntput(old_pwdmnt);
 865         }
 866 }
 867 # endif /* HAVE_2ARGS_SET_FS_PWD */
 868 #endif /* HAVE_SET_FS_PWD */
 869
 870 int
 871 vn_set_pwd(const char *filename)
 872 {
 873 #if defined(HAVE_2ARGS_SET_FS_PWD) && defined(HAVE_USER_PATH_DIR)
 874         struct path path;
 875 #else
 876         struct nameidata nd;
 877 #endif /* HAVE_2ARGS_SET_FS_PWD */
 878         mm_segment_t saved_fs;
 879         int rc;
 880         SENTRY;
 881
 882         /*
 883          * user_path_dir() and __user_walk() both expect 'filename' to be
 884          * a user space address so we must briefly increase the data segment
 885          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 886          */
 887         saved_fs = get_fs();
 888         set_fs(get_ds());
 889
 890 #ifdef HAVE_2ARGS_SET_FS_PWD
 891 # ifdef HAVE_USER_PATH_DIR
 892         rc = user_path_dir(filename, &path);
 893         if (rc)
 894                 SGOTO(out, rc);
 895
 896         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 897         if (rc)
 898                 SGOTO(dput_and_out, rc);
 899
 900         set_fs_pwd(current->fs, &path);
 901
 902 dput_and_out:
 903         path_put(&path);
 904 # else
 905         rc = __user_walk(filename,
 906                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 907         if (rc)
 908                 SGOTO(out, rc);
 909
 910         rc = vfs_permission(&nd, MAY_EXEC);
 911         if (rc)
 912                 SGOTO(dput_and_out, rc);
 913
 914         set_fs_pwd(current->fs, &nd.path);
 915
 916 dput_and_out:
 917         path_put(&nd.path);
 918 # endif /* HAVE_USER_PATH_DIR */
 919 #else
 920         rc = __user_walk(filename,
 921                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 922         if (rc)
 923                 SGOTO(out, rc);
 924
 925         rc = vfs_permission(&nd, MAY_EXEC);
 926         if (rc)
 927                 SGOTO(dput_and_out, rc);
 928
 929         set_fs_pwd(current->fs, nd.nd_mnt, nd.nd_dentry);
 930
 931 dput_and_out:
 932         vn_path_release(&nd);
 933 #endif /* HAVE_2ARGS_SET_FS_PWD */
 934 out:
 935         set_fs(saved_fs);
 936
 937         SRETURN(-rc);
 938 } /* vn_set_pwd() */
 939 EXPORT_SYMBOL(vn_set_pwd);
 940
 941 static int
 942 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 943 {
 944         struct vnode *vp = buf;
 945
 946         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 947
 948         return (0);
 949 } /* vn_cache_constructor() */
 950
 951 static void
 952 vn_cache_destructor(void *buf, void *cdrarg)
 953 {
 954         struct vnode *vp = buf;
 955
 956         mutex_destroy(&vp->v_lock);
 957 } /* vn_cache_destructor() */
 958
 959 static int
 960 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 961 {
 962         file_t *fp = buf;
 963
 964         atomic_set(&fp->f_ref, 0);
 965         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 966         INIT_LIST_HEAD(&fp->f_list);
 967
 968         return (0);
 969 } /* file_cache_constructor() */
 970
 971 static void
 972 vn_file_cache_destructor(void *buf, void *cdrarg)
 973 {
 974         file_t *fp = buf;
 975
 976         mutex_destroy(&fp->f_lock);
 977 } /* vn_file_cache_destructor() */
 978
 979 int spl_vn_init_kallsyms_lookup(void)
 980 {
 981 #ifdef HAVE_KERN_PATH_PARENT_HEADER
 982 #ifndef HAVE_KERN_PATH_PARENT_SYMBOL
 983         kern_path_parent_fn = (kern_path_parent_t)
 984                 spl_kallsyms_lookup_name("kern_path_parent");
 985         if (!kern_path_parent_fn) {
 986                 printk(KERN_ERR "Error: Unknown symbol kern_path_parent\n");
 987                 return -EFAULT;
 988         }
 989 #endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
 990 #endif /* HAVE_KERN_PATH_PARENT_HEADER */
 991
 992 #ifdef HAVE_KERN_PATH_LOCKED
 993         kern_path_locked_fn = (kern_path_locked_t)
 994                 spl_kallsyms_lookup_name("kern_path_locked");
 995         if (!kern_path_locked_fn) {
 996                 printk(KERN_ERR "Error: Unknown symbol kern_path_locked\n");
 997                 return -EFAULT;
 998         }
 999 #endif
1000
1001         return (0);
1002 }
1003
1004 int
1005 spl_vn_init(void)
1006 {
1007         SENTRY;
1008         vn_cache = kmem_cache_create("spl_vn_cache",
1009                                      sizeof(struct vnode), 64,
1010                                      vn_cache_constructor,
1011                                      vn_cache_destructor,
1012                                      NULL, NULL, NULL, KMC_KMEM);
1013
1014         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
1015                                           sizeof(file_t), 64,
1016                                           vn_file_cache_constructor,
1017                                           vn_file_cache_destructor,
1018                                           NULL, NULL, NULL, KMC_KMEM);
1019         SRETURN(0);
1020 } /* vn_init() */
1021
1022 void
1023 spl_vn_fini(void)
1024 {
1025         file_t *fp, *next_fp;
1026         int leaked = 0;
1027         SENTRY;
1028
1029         spin_lock(&vn_file_lock);
1030
1031         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
1032                 list_del(&fp->f_list);
1033                 releasef_locked(fp);
1034                 leaked++;
1035         }
1036
1037         spin_unlock(&vn_file_lock);
1038
1039         if (leaked > 0)
1040                 SWARN("Warning %d files leaked\n", leaked);
1041
1042         kmem_cache_destroy(vn_file_cache);
1043         kmem_cache_destroy(vn_cache);
1044
1045         SEXIT;
1046         return;
1047 } /* vn_fini() */