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/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 #ifdef HAVE_2ARGS_VFS_GETATTR
 179         rc = vfs_getattr(&fp->f_path, &stat);
 180 #else
 181         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 182 #endif
 183         if (rc) {
 184                 filp_close(fp, 0);
 185                 SRETURN(-rc);
 186         }
 187
 188         vp = vn_alloc(KM_SLEEP);
 189         if (!vp) {
 190                 filp_close(fp, 0);
 191                 SRETURN(ENOMEM);
 192         }
 193
 194         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 195         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 196
 197         mutex_enter(&vp->v_lock);
 198         vp->v_type = vn_mode_to_vtype(stat.mode);
 199         vp->v_file = fp;
 200         vp->v_gfp_mask = saved_gfp;
 201         *vpp = vp;
 202         mutex_exit(&vp->v_lock);
 203
 204         SRETURN(0);
 205 } /* vn_open() */
 206 EXPORT_SYMBOL(vn_open);
 207
 208 int
 209 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 210           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 211 {
 212         char *realpath;
 213         int len, rc;
 214         SENTRY;
 215
 216         ASSERT(vp == rootdir);
 217
 218         len = strlen(path) + 2;
 219         realpath = kmalloc(len, GFP_KERNEL);
 220         if (!realpath)
 221                 SRETURN(ENOMEM);
 222
 223         (void)snprintf(realpath, len, "/%s", path);
 224         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 225         kfree(realpath);
 226
 227         SRETURN(rc);
 228 } /* vn_openat() */
 229 EXPORT_SYMBOL(vn_openat);
 230
 231 int
 232 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 233         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 234 {
 235         loff_t offset;
 236         mm_segment_t saved_fs;
 237         struct file *fp;
 238         int rc;
 239         SENTRY;
 240
 241         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 242         ASSERT(vp);
 243         ASSERT(vp->v_file);
 244         ASSERT(seg == UIO_SYSSPACE);
 245         ASSERT((ioflag & ~FAPPEND) == 0);
 246         ASSERT(x2 == RLIM64_INFINITY);
 247
 248         fp = vp->v_file;
 249
 250         offset = off;
 251         if (ioflag & FAPPEND)
 252                 offset = fp->f_pos;
 253
 254         /* Writable user data segment must be briefly increased for this
 255          * process so we can use the user space read call paths to write
 256          * in to memory allocated by the kernel. */
 257         saved_fs = get_fs();
 258         set_fs(get_ds());
 259
 260         if (uio & UIO_WRITE)
 261                 rc = vfs_write(fp, addr, len, &offset);
 262         else
 263                 rc = vfs_read(fp, addr, len, &offset);
 264
 265         set_fs(saved_fs);
 266         fp->f_pos = offset;
 267
 268         if (rc < 0)
 269                 SRETURN(-rc);
 270
 271         if (residp) {
 272                 *residp = len - rc;
 273         } else {
 274                 if (rc != len)
 275                         SRETURN(EIO);
 276         }
 277
 278         SRETURN(0);
 279 } /* vn_rdwr() */
 280 EXPORT_SYMBOL(vn_rdwr);
 281
 282 int
 283 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 284 {
 285         int rc;
 286         SENTRY;
 287
 288         ASSERT(vp);
 289         ASSERT(vp->v_file);
 290
 291         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 292         rc = filp_close(vp->v_file, 0);
 293         vn_free(vp);
 294
 295         SRETURN(-rc);
 296 } /* vn_close() */
 297 EXPORT_SYMBOL(vn_close);
 298
 299 /* vn_seek() does not actually seek it only performs bounds checking on the
 300  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 301  * anything more serious. */
 302 int
 303 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 304 {
 305         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 306 }
 307 EXPORT_SYMBOL(vn_seek);
 308
 309 #ifdef HAVE_KERN_PATH_LOCKED
 310 /* Based on do_unlinkat() from linux/fs/namei.c */
 311 int
 312 vn_remove(const char *path, uio_seg_t seg, int flags)
 313 {
 314         struct dentry *dentry;
 315         struct path parent;
 316         struct inode *inode = NULL;
 317         int rc = 0;
 318         SENTRY;
 319
 320         ASSERT(seg == UIO_SYSSPACE);
 321         ASSERT(flags == RMFILE);
 322
 323         dentry = spl_kern_path_locked(path, &parent);
 324         rc = PTR_ERR(dentry);
 325         if (!IS_ERR(dentry)) {
 326                 if (parent.dentry->d_name.name[parent.dentry->d_name.len])
 327                         SGOTO(slashes, rc = 0);
 328
 329                 inode = dentry->d_inode;
 330                 if (!inode)
 331                         SGOTO(slashes, rc = 0);
 332
 333                 if (inode)
 334                         ihold(inode);
 335
 336                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 337 exit1:
 338                 dput(dentry);
 339         } else {
 340                 return (-rc);
 341         }
 342
 343         spl_inode_unlock(parent.dentry->d_inode);
 344         if (inode)
 345                 iput(inode);    /* truncate the inode here */
 346
 347         path_put(&parent);
 348         SRETURN(-rc);
 349
 350 slashes:
 351         rc = !dentry->d_inode ? -ENOENT :
 352             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 353         SGOTO(exit1, rc);
 354 } /* vn_remove() */
 355 EXPORT_SYMBOL(vn_remove);
 356
 357 /* Based on do_rename() from linux/fs/namei.c */
 358 int
 359 vn_rename(const char *oldname, const char *newname, int x1)
 360 {
 361         struct dentry *old_dir, *new_dir;
 362         struct dentry *old_dentry, *new_dentry;
 363         struct dentry *trap;
 364         struct path old_parent, new_parent;
 365         int rc = 0;
 366         SENTRY;
 367
 368         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 369         if (IS_ERR(old_dentry))
 370                 SGOTO(exit, rc = PTR_ERR(old_dentry));
 371
 372         spl_inode_unlock(old_parent.dentry->d_inode);
 373
 374         new_dentry = spl_kern_path_locked(newname, &new_parent);
 375         if (IS_ERR(new_dentry))
 376                 SGOTO(exit2, rc = PTR_ERR(new_dentry));
 377
 378         spl_inode_unlock(new_parent.dentry->d_inode);
 379
 380         rc = -EXDEV;
 381         if (old_parent.mnt != new_parent.mnt)
 382                 SGOTO(exit3, rc);
 383
 384         old_dir = old_parent.dentry;
 385         new_dir = new_parent.dentry;
 386         trap = lock_rename(new_dir, old_dir);
 387
 388         /* source should not be ancestor of target */
 389         rc = -EINVAL;
 390         if (old_dentry == trap)
 391                 SGOTO(exit4, rc);
 392
 393         /* target should not be an ancestor of source */
 394         rc = -ENOTEMPTY;
 395         if (new_dentry == trap)
 396                 SGOTO(exit4, rc);
 397
 398         /* source must exist */
 399         rc = -ENOENT;
 400         if (!old_dentry->d_inode)
 401                 SGOTO(exit4, rc);
 402
 403         /* unless the source is a directory trailing slashes give -ENOTDIR */
 404         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 405                 rc = -ENOTDIR;
 406                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 407                         SGOTO(exit4, rc);
 408                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 409                         SGOTO(exit4, rc);
 410         }
 411
 412 #ifdef HAVE_4ARGS_VFS_RENAME
 413         rc = vfs_rename(old_dir->d_inode, old_dentry,
 414                         new_dir->d_inode, new_dentry);
 415 #else
 416         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 417                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 418 #endif /* HAVE_4ARGS_VFS_RENAME */
 419 exit4:
 420         unlock_rename(new_dir, old_dir);
 421 exit3:
 422         dput(new_dentry);
 423         path_put(&new_parent);
 424 exit2:
 425         dput(old_dentry);
 426         path_put(&old_parent);
 427 exit:
 428         SRETURN(-rc);
 429 }
 430 EXPORT_SYMBOL(vn_rename);
 431
 432 #else
 433 static struct dentry *
 434 vn_lookup_hash(struct nameidata *nd)
 435 {
 436         return lookup_one_len((const char *)nd->last.name,
 437                               nd->nd_dentry, nd->last.len);
 438 } /* lookup_hash() */
 439
 440 static void
 441 vn_path_release(struct nameidata *nd)
 442 {
 443         dput(nd->nd_dentry);
 444         mntput(nd->nd_mnt);
 445 }
 446
 447 /* Modified do_unlinkat() from linux/fs/namei.c, only uses exported symbols */
 448 int
 449 vn_remove(const char *path, uio_seg_t seg, int flags)
 450 {
 451         struct dentry *dentry;
 452         struct nameidata nd;
 453         struct inode *inode = NULL;
 454         int rc = 0;
 455         SENTRY;
 456
 457         ASSERT(seg == UIO_SYSSPACE);
 458         ASSERT(flags == RMFILE);
 459
 460         rc = spl_kern_path_parent(path, &nd);
 461         if (rc)
 462                 SGOTO(exit, rc);
 463
 464         rc = -EISDIR;
 465         if (nd.last_type != LAST_NORM)
 466                 SGOTO(exit1, rc);
 467
 468         spl_inode_lock_nested(nd.nd_dentry->d_inode, I_MUTEX_PARENT);
 469         dentry = vn_lookup_hash(&nd);
 470         rc = PTR_ERR(dentry);
 471         if (!IS_ERR(dentry)) {
 472                 /* Why not before? Because we want correct rc value */
 473                 if (nd.last.name[nd.last.len])
 474                         SGOTO(slashes, rc);
 475
 476                 inode = dentry->d_inode;
 477                 if (inode)
 478                         atomic_inc(&inode->i_count);
 479 #ifdef HAVE_2ARGS_VFS_UNLINK
 480                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry);
 481 #else
 482                 rc = vfs_unlink(nd.nd_dentry->d_inode, dentry, nd.nd_mnt);
 483 #endif /* HAVE_2ARGS_VFS_UNLINK */
 484 exit2:
 485                 dput(dentry);
 486         }
 487
 488         spl_inode_unlock(nd.nd_dentry->d_inode);
 489         if (inode)
 490                 iput(inode);    /* truncate the inode here */
 491 exit1:
 492         vn_path_release(&nd);
 493 exit:
 494         SRETURN(-rc);
 495
 496 slashes:
 497         rc = !dentry->d_inode ? -ENOENT :
 498                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 499         SGOTO(exit2, rc);
 500 } /* vn_remove() */
 501 EXPORT_SYMBOL(vn_remove);
 502
 503 /* Modified do_rename() from linux/fs/namei.c, only uses exported symbols */
 504 int
 505 vn_rename(const char *oldname, const char *newname, int x1)
 506 {
 507         struct dentry *old_dir, *new_dir;
 508         struct dentry *old_dentry, *new_dentry;
 509         struct dentry *trap;
 510         struct nameidata oldnd, newnd;
 511         int rc = 0;
 512         SENTRY;
 513
 514         rc = spl_kern_path_parent(oldname, &oldnd);
 515         if (rc)
 516                 SGOTO(exit, rc);
 517
 518         rc = spl_kern_path_parent(newname, &newnd);
 519         if (rc)
 520                 SGOTO(exit1, rc);
 521
 522         rc = -EXDEV;
 523         if (oldnd.nd_mnt != newnd.nd_mnt)
 524                 SGOTO(exit2, rc);
 525
 526         old_dir = oldnd.nd_dentry;
 527         rc = -EBUSY;
 528         if (oldnd.last_type != LAST_NORM)
 529                 SGOTO(exit2, rc);
 530
 531         new_dir = newnd.nd_dentry;
 532         if (newnd.last_type != LAST_NORM)
 533                 SGOTO(exit2, rc);
 534
 535         trap = lock_rename(new_dir, old_dir);
 536
 537         old_dentry = vn_lookup_hash(&oldnd);
 538
 539         rc = PTR_ERR(old_dentry);
 540         if (IS_ERR(old_dentry))
 541                 SGOTO(exit3, rc);
 542
 543         /* source must exist */
 544         rc = -ENOENT;
 545         if (!old_dentry->d_inode)
 546                 SGOTO(exit4, rc);
 547
 548         /* unless the source is a directory trailing slashes give -ENOTDIR */
 549         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 550                 rc = -ENOTDIR;
 551                 if (oldnd.last.name[oldnd.last.len])
 552                         SGOTO(exit4, rc);
 553                 if (newnd.last.name[newnd.last.len])
 554                         SGOTO(exit4, rc);
 555         }
 556
 557         /* source should not be ancestor of target */
 558         rc = -EINVAL;
 559         if (old_dentry == trap)
 560                 SGOTO(exit4, rc);
 561
 562         new_dentry = vn_lookup_hash(&newnd);
 563         rc = PTR_ERR(new_dentry);
 564         if (IS_ERR(new_dentry))
 565                 SGOTO(exit4, rc);
 566
 567         /* target should not be an ancestor of source */
 568         rc = -ENOTEMPTY;
 569         if (new_dentry == trap)
 570                 SGOTO(exit5, rc);
 571
 572 #ifdef HAVE_4ARGS_VFS_RENAME
 573         rc = vfs_rename(old_dir->d_inode, old_dentry,
 574                         new_dir->d_inode, new_dentry);
 575 #else
 576         rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
 577                         new_dir->d_inode, new_dentry, newnd.nd_mnt);
 578 #endif /* HAVE_4ARGS_VFS_RENAME */
 579 exit5:
 580         dput(new_dentry);
 581 exit4:
 582         dput(old_dentry);
 583 exit3:
 584         unlock_rename(new_dir, old_dir);
 585 exit2:
 586         vn_path_release(&newnd);
 587 exit1:
 588         vn_path_release(&oldnd);
 589 exit:
 590         SRETURN(-rc);
 591 }
 592 EXPORT_SYMBOL(vn_rename);
 593 #endif /* HAVE_KERN_PATH_LOCKED */
 594
 595 int
 596 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 597 {
 598         struct file *fp;
 599         struct kstat stat;
 600         int rc;
 601         SENTRY;
 602
 603         ASSERT(vp);
 604         ASSERT(vp->v_file);
 605         ASSERT(vap);
 606
 607         fp = vp->v_file;
 608
 609 #ifdef HAVE_2ARGS_VFS_GETATTR
 610         rc = vfs_getattr(&fp->f_path, &stat);
 611 #else
 612         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 613 #endif
 614         if (rc)
 615                 SRETURN(-rc);
 616
 617         vap->va_type          = vn_mode_to_vtype(stat.mode);
 618         vap->va_mode          = stat.mode;
 619         vap->va_uid           = stat.uid;
 620         vap->va_gid           = stat.gid;
 621         vap->va_fsid          = 0;
 622         vap->va_nodeid        = stat.ino;
 623         vap->va_nlink         = stat.nlink;
 624         vap->va_size          = stat.size;
 625         vap->va_blksize       = stat.blksize;
 626         vap->va_atime         = stat.atime;
 627         vap->va_mtime         = stat.mtime;
 628         vap->va_ctime         = stat.ctime;
 629         vap->va_rdev          = stat.rdev;
 630         vap->va_nblocks       = stat.blocks;
 631
 632         SRETURN(0);
 633 }
 634 EXPORT_SYMBOL(vn_getattr);
 635
 636 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 637 {
 638         int datasync = 0;
 639         SENTRY;
 640
 641         ASSERT(vp);
 642         ASSERT(vp->v_file);
 643
 644         if (flags & FDSYNC)
 645                 datasync = 1;
 646
 647         SRETURN(-spl_filp_fsync(vp->v_file, datasync));
 648 } /* vn_fsync() */
 649 EXPORT_SYMBOL(vn_fsync);
 650
 651 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 652     offset_t offset, void *x6, void *x7)
 653 {
 654         int error = EOPNOTSUPP;
 655         SENTRY;
 656
 657         if (cmd != F_FREESP || bfp->l_whence != 0)
 658                 SRETURN(EOPNOTSUPP);
 659
 660         ASSERT(vp);
 661         ASSERT(vp->v_file);
 662         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 663
 664 #ifdef FALLOC_FL_PUNCH_HOLE
 665         /*
 666          * When supported by the underlying file system preferentially
 667          * use the fallocate() callback to preallocate the space.
 668          */
 669         error = -spl_filp_fallocate(vp->v_file,
 670             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 671             bfp->l_start, bfp->l_len);
 672         if (error == 0)
 673                 SRETURN(0);
 674 #endif
 675
 676 #ifdef HAVE_INODE_TRUNCATE_RANGE
 677         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 678             vp->v_file->f_dentry->d_inode->i_op &&
 679             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 680                 off_t end = bfp->l_start + bfp->l_len;
 681                 /*
 682                  * Judging from the code in shmem_truncate_range(),
 683                  * it seems the kernel expects the end offset to be
 684                  * inclusive and aligned to the end of a page.
 685                  */
 686                 if (end % PAGE_SIZE != 0) {
 687                         end &= ~(off_t)(PAGE_SIZE - 1);
 688                         if (end <= bfp->l_start)
 689                                 SRETURN(0);
 690                 }
 691                 --end;
 692
 693                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 694                         vp->v_file->f_dentry->d_inode,
 695                         bfp->l_start, end
 696                 );
 697                 SRETURN(0);
 698         }
 699 #endif
 700
 701         SRETURN(error);
 702 }
 703 EXPORT_SYMBOL(vn_space);
 704
 705 /* Function must be called while holding the vn_file_lock */
 706 static file_t *
 707 file_find(int fd)
 708 {
 709         file_t *fp;
 710
 711         ASSERT(spin_is_locked(&vn_file_lock));
 712
 713         list_for_each_entry(fp, &vn_file_list,  f_list) {
 714                 if (fd == fp->f_fd && fp->f_task == current) {
 715                         ASSERT(atomic_read(&fp->f_ref) != 0);
 716                         return fp;
 717                 }
 718         }
 719
 720         return NULL;
 721 } /* file_find() */
 722
 723 file_t *
 724 vn_getf(int fd)
 725 {
 726         struct kstat stat;
 727         struct file *lfp;
 728         file_t *fp;
 729         vnode_t *vp;
 730         int rc = 0;
 731         SENTRY;
 732
 733         /* Already open just take an extra reference */
 734         spin_lock(&vn_file_lock);
 735
 736         fp = file_find(fd);
 737         if (fp) {
 738                 atomic_inc(&fp->f_ref);
 739                 spin_unlock(&vn_file_lock);
 740                 SRETURN(fp);
 741         }
 742
 743         spin_unlock(&vn_file_lock);
 744
 745         /* File was not yet opened create the object and setup */
 746         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 747         if (fp == NULL)
 748                 SGOTO(out, rc);
 749
 750         mutex_enter(&fp->f_lock);
 751
 752         fp->f_fd = fd;
 753         fp->f_task = current;
 754         fp->f_offset = 0;
 755         atomic_inc(&fp->f_ref);
 756
 757         lfp = fget(fd);
 758         if (lfp == NULL)
 759                 SGOTO(out_mutex, rc);
 760
 761         vp = vn_alloc(KM_SLEEP);
 762         if (vp == NULL)
 763                 SGOTO(out_fget, rc);
 764
 765 #ifdef HAVE_2ARGS_VFS_GETATTR
 766         rc = vfs_getattr(&lfp->f_path, &stat);
 767 #else
 768         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 769 #endif
 770         if (rc)
 771                 SGOTO(out_vnode, rc);
 772
 773         mutex_enter(&vp->v_lock);
 774         vp->v_type = vn_mode_to_vtype(stat.mode);
 775         vp->v_file = lfp;
 776         mutex_exit(&vp->v_lock);
 777
 778         fp->f_vnode = vp;
 779         fp->f_file = lfp;
 780
 781         /* Put it on the tracking list */
 782         spin_lock(&vn_file_lock);
 783         list_add(&fp->f_list, &vn_file_list);
 784         spin_unlock(&vn_file_lock);
 785
 786         mutex_exit(&fp->f_lock);
 787         SRETURN(fp);
 788
 789 out_vnode:
 790         vn_free(vp);
 791 out_fget:
 792         fput(lfp);
 793 out_mutex:
 794         mutex_exit(&fp->f_lock);
 795         kmem_cache_free(vn_file_cache, fp);
 796 out:
 797         SRETURN(NULL);
 798 } /* getf() */
 799 EXPORT_SYMBOL(getf);
 800
 801 static void releasef_locked(file_t *fp)
 802 {
 803         ASSERT(fp->f_file);
 804         ASSERT(fp->f_vnode);
 805
 806         /* Unlinked from list, no refs, safe to free outside mutex */
 807         fput(fp->f_file);
 808         vn_free(fp->f_vnode);
 809
 810         kmem_cache_free(vn_file_cache, fp);
 811 }
 812
 813 void
 814 vn_releasef(int fd)
 815 {
 816         file_t *fp;
 817         SENTRY;
 818
 819         spin_lock(&vn_file_lock);
 820         fp = file_find(fd);
 821         if (fp) {
 822                 atomic_dec(&fp->f_ref);
 823                 if (atomic_read(&fp->f_ref) > 0) {
 824                         spin_unlock(&vn_file_lock);
 825                         SEXIT;
 826                         return;
 827                 }
 828
 829                 list_del(&fp->f_list);
 830                 releasef_locked(fp);
 831         }
 832         spin_unlock(&vn_file_lock);
 833
 834         SEXIT;
 835         return;
 836 } /* releasef() */
 837 EXPORT_SYMBOL(releasef);
 838
 839 #ifndef HAVE_SET_FS_PWD
 840 # ifdef HAVE_2ARGS_SET_FS_PWD
 841 /* Used from 2.6.25 - 2.6.31+ */
 842 void
 843 #  ifdef HAVE_SET_FS_PWD_WITH_CONST
 844 set_fs_pwd(struct fs_struct *fs, const struct path *path)
 845 #  else
 846 set_fs_pwd(struct fs_struct *fs, struct path *path)
 847 #  endif
 848 {
 849         struct path old_pwd;
 850
 851 #  ifdef HAVE_FS_STRUCT_SPINLOCK
 852         spin_lock(&fs->lock);
 853         old_pwd = fs->pwd;
 854         fs->pwd = *path;
 855         path_get(path);
 856         spin_unlock(&fs->lock);
 857 #  else
 858         write_lock(&fs->lock);
 859         old_pwd = fs->pwd;
 860         fs->pwd = *path;
 861         path_get(path);
 862         write_unlock(&fs->lock);
 863 #  endif /* HAVE_FS_STRUCT_SPINLOCK */
 864
 865         if (old_pwd.dentry)
 866                 path_put(&old_pwd);
 867 }
 868 # else
 869 /* Used from 2.6.11 - 2.6.24 */
 870 void
 871 set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, struct dentry *dentry)
 872 {
 873         struct dentry *old_pwd;
 874         struct vfsmount *old_pwdmnt;
 875
 876         write_lock(&fs->lock);
 877         old_pwd = fs->pwd;
 878         old_pwdmnt = fs->pwdmnt;
 879         fs->pwdmnt = mntget(mnt);
 880         fs->pwd = dget(dentry);
 881         write_unlock(&fs->lock);
 882
 883         if (old_pwd) {
 884                 dput(old_pwd);
 885                 mntput(old_pwdmnt);
 886         }
 887 }
 888 # endif /* HAVE_2ARGS_SET_FS_PWD */
 889 #endif /* HAVE_SET_FS_PWD */
 890
 891 int
 892 vn_set_pwd(const char *filename)
 893 {
 894 #if defined(HAVE_2ARGS_SET_FS_PWD) && defined(HAVE_USER_PATH_DIR)
 895         struct path path;
 896 #else
 897         struct nameidata nd;
 898 #endif /* HAVE_2ARGS_SET_FS_PWD */
 899         mm_segment_t saved_fs;
 900         int rc;
 901         SENTRY;
 902
 903         /*
 904          * user_path_dir() and __user_walk() both expect 'filename' to be
 905          * a user space address so we must briefly increase the data segment
 906          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 907          */
 908         saved_fs = get_fs();
 909         set_fs(get_ds());
 910
 911 #ifdef HAVE_2ARGS_SET_FS_PWD
 912 # ifdef HAVE_USER_PATH_DIR
 913         rc = user_path_dir(filename, &path);
 914         if (rc)
 915                 SGOTO(out, rc);
 916
 917         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 918         if (rc)
 919                 SGOTO(dput_and_out, rc);
 920
 921         set_fs_pwd(current->fs, &path);
 922
 923 dput_and_out:
 924         path_put(&path);
 925 # else
 926         rc = __user_walk(filename,
 927                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 928         if (rc)
 929                 SGOTO(out, rc);
 930
 931         rc = vfs_permission(&nd, MAY_EXEC);
 932         if (rc)
 933                 SGOTO(dput_and_out, rc);
 934
 935         set_fs_pwd(current->fs, &nd.path);
 936
 937 dput_and_out:
 938         path_put(&nd.path);
 939 # endif /* HAVE_USER_PATH_DIR */
 940 #else
 941         rc = __user_walk(filename,
 942                          LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 943         if (rc)
 944                 SGOTO(out, rc);
 945
 946         rc = vfs_permission(&nd, MAY_EXEC);
 947         if (rc)
 948                 SGOTO(dput_and_out, rc);
 949
 950         set_fs_pwd(current->fs, nd.nd_mnt, nd.nd_dentry);
 951
 952 dput_and_out:
 953         vn_path_release(&nd);
 954 #endif /* HAVE_2ARGS_SET_FS_PWD */
 955 out:
 956         set_fs(saved_fs);
 957
 958         SRETURN(-rc);
 959 } /* vn_set_pwd() */
 960 EXPORT_SYMBOL(vn_set_pwd);
 961
 962 static int
 963 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 964 {
 965         struct vnode *vp = buf;
 966
 967         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 968
 969         return (0);
 970 } /* vn_cache_constructor() */
 971
 972 static void
 973 vn_cache_destructor(void *buf, void *cdrarg)
 974 {
 975         struct vnode *vp = buf;
 976
 977         mutex_destroy(&vp->v_lock);
 978 } /* vn_cache_destructor() */
 979
 980 static int
 981 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 982 {
 983         file_t *fp = buf;
 984
 985         atomic_set(&fp->f_ref, 0);
 986         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 987         INIT_LIST_HEAD(&fp->f_list);
 988
 989         return (0);
 990 } /* file_cache_constructor() */
 991
 992 static void
 993 vn_file_cache_destructor(void *buf, void *cdrarg)
 994 {
 995         file_t *fp = buf;
 996
 997         mutex_destroy(&fp->f_lock);
 998 } /* vn_file_cache_destructor() */
 999
1000 int spl_vn_init_kallsyms_lookup(void)
1001 {
1002 #ifdef HAVE_KERN_PATH_PARENT_HEADER
1003 #ifndef HAVE_KERN_PATH_PARENT_SYMBOL
1004         kern_path_parent_fn = (kern_path_parent_t)
1005                 spl_kallsyms_lookup_name("kern_path_parent");
1006         if (!kern_path_parent_fn) {
1007                 printk(KERN_ERR "Error: Unknown symbol kern_path_parent\n");
1008                 return -EFAULT;
1009         }
1010 #endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
1011 #endif /* HAVE_KERN_PATH_PARENT_HEADER */
1012
1013 #ifdef HAVE_KERN_PATH_LOCKED
1014         kern_path_locked_fn = (kern_path_locked_t)
1015                 spl_kallsyms_lookup_name("kern_path_locked");
1016         if (!kern_path_locked_fn) {
1017                 printk(KERN_ERR "Error: Unknown symbol kern_path_locked\n");
1018                 return -EFAULT;
1019         }
1020 #endif
1021
1022         return (0);
1023 }
1024
1025 int
1026 spl_vn_init(void)
1027 {
1028         SENTRY;
1029         vn_cache = kmem_cache_create("spl_vn_cache",
1030                                      sizeof(struct vnode), 64,
1031                                      vn_cache_constructor,
1032                                      vn_cache_destructor,
1033                                      NULL, NULL, NULL, KMC_KMEM);
1034
1035         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
1036                                           sizeof(file_t), 64,
1037                                           vn_file_cache_constructor,
1038                                           vn_file_cache_destructor,
1039                                           NULL, NULL, NULL, KMC_KMEM);
1040         SRETURN(0);
1041 } /* vn_init() */
1042
1043 void
1044 spl_vn_fini(void)
1045 {
1046         file_t *fp, *next_fp;
1047         int leaked = 0;
1048         SENTRY;
1049
1050         spin_lock(&vn_file_lock);
1051
1052         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
1053                 list_del(&fp->f_list);
1054                 releasef_locked(fp);
1055                 leaked++;
1056         }
1057
1058         spin_unlock(&vn_file_lock);
1059
1060         if (leaked > 0)
1061                 SWARN("Warning %d files leaked\n", leaked);
1062
1063         kmem_cache_destroy(vn_file_cache);
1064         kmem_cache_destroy(vn_cache);
1065
1066         SEXIT;
1067         return;
1068 } /* vn_fini() */