module/spl/spl-vnode.c

   1 /*****************************************************************************\
   2  *  Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
   3  *  Copyright (C) 2007 The Regents of the University of California.
   4  *  Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
   5  *  Written by Brian Behlendorf <behlendorf1@llnl.gov>.
   6  *  UCRL-CODE-235197
   7  *
   8  *  This file is part of the SPL, Solaris Porting Layer.
   9  *  For details, see <http://zfsonlinux.org/>.
  10  *
  11  *  The SPL is free software; you can redistribute it and/or modify it
  12  *  under the terms of the GNU General Public License as published by the
  13  *  Free Software Foundation; either version 2 of the License, or (at your
  14  *  option) any later version.
  15  *
  16  *  The SPL is distributed in the hope that it will be useful, but WITHOUT
  17  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  18  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  19  *  for more details.
  20  *
  21  *  You should have received a copy of the GNU General Public License along
  22  *  with the SPL.  If not, see <http://www.gnu.org/licenses/>.
  23  *****************************************************************************
  24  *  Solaris Porting Layer (SPL) Vnode Implementation.
  25 \*****************************************************************************/
  26
  27 #include <sys/cred.h>
  28 #include <sys/vnode.h>
  29 #include <sys/kmem_cache.h>
  30 #include <linux/falloc.h>
  31 #include <linux/file_compat.h>
  32
  33 vnode_t *rootdir = (vnode_t *)0xabcd1234;
  34 EXPORT_SYMBOL(rootdir);
  35
  36 static spl_kmem_cache_t *vn_cache;
  37 static spl_kmem_cache_t *vn_file_cache;
  38
  39 static DEFINE_SPINLOCK(vn_file_lock);
  40 static LIST_HEAD(vn_file_list);
  41
  42 vtype_t
  43 vn_mode_to_vtype(mode_t mode)
  44 {
  45         if (S_ISREG(mode))
  46                 return VREG;
  47
  48         if (S_ISDIR(mode))
  49                 return VDIR;
  50
  51         if (S_ISCHR(mode))
  52                 return VCHR;
  53
  54         if (S_ISBLK(mode))
  55                 return VBLK;
  56
  57         if (S_ISFIFO(mode))
  58                 return VFIFO;
  59
  60         if (S_ISLNK(mode))
  61                 return VLNK;
  62
  63         if (S_ISSOCK(mode))
  64                 return VSOCK;
  65
  66         return VNON;
  67 } /* vn_mode_to_vtype() */
  68 EXPORT_SYMBOL(vn_mode_to_vtype);
  69
  70 mode_t
  71 vn_vtype_to_mode(vtype_t vtype)
  72 {
  73         if (vtype == VREG)
  74                 return S_IFREG;
  75
  76         if (vtype == VDIR)
  77                 return S_IFDIR;
  78
  79         if (vtype == VCHR)
  80                 return S_IFCHR;
  81
  82         if (vtype == VBLK)
  83                 return S_IFBLK;
  84
  85         if (vtype == VFIFO)
  86                 return S_IFIFO;
  87
  88         if (vtype == VLNK)
  89                 return S_IFLNK;
  90
  91         if (vtype == VSOCK)
  92                 return S_IFSOCK;
  93
  94         return VNON;
  95 } /* vn_vtype_to_mode() */
  96 EXPORT_SYMBOL(vn_vtype_to_mode);
  97
  98 vnode_t *
  99 vn_alloc(int flag)
 100 {
 101         vnode_t *vp;
 102
 103         vp = kmem_cache_alloc(vn_cache, flag);
 104         if (vp != NULL) {
 105                 vp->v_file = NULL;
 106                 vp->v_type = 0;
 107         }
 108
 109         return (vp);
 110 } /* vn_alloc() */
 111 EXPORT_SYMBOL(vn_alloc);
 112
 113 void
 114 vn_free(vnode_t *vp)
 115 {
 116         kmem_cache_free(vn_cache, vp);
 117 } /* vn_free() */
 118 EXPORT_SYMBOL(vn_free);
 119
 120 int
 121 vn_open(const char *path, uio_seg_t seg, int flags, int mode,
 122         vnode_t **vpp, int x1, void *x2)
 123 {
 124         struct file *fp;
 125         struct kstat stat;
 126         int rc, saved_umask = 0;
 127         gfp_t saved_gfp;
 128         vnode_t *vp;
 129
 130         ASSERT(flags & (FWRITE | FREAD));
 131         ASSERT(seg == UIO_SYSSPACE);
 132         ASSERT(vpp);
 133         *vpp = NULL;
 134
 135         if (!(flags & FCREAT) && (flags & FWRITE))
 136                 flags |= FEXCL;
 137
 138         /* Note for filp_open() the two low bits must be remapped to mean:
 139          * 01 - read-only  -> 00 read-only
 140          * 10 - write-only -> 01 write-only
 141          * 11 - read-write -> 10 read-write
 142          */
 143         flags--;
 144
 145         if (flags & FCREAT)
 146                 saved_umask = xchg(&current->fs->umask, 0);
 147
 148         fp = filp_open(path, flags, mode);
 149
 150         if (flags & FCREAT)
 151                 (void)xchg(&current->fs->umask, saved_umask);
 152
 153         if (IS_ERR(fp))
 154                 return (-PTR_ERR(fp));
 155
 156 #if defined(HAVE_4ARGS_VFS_GETATTR)
 157         rc = vfs_getattr(&fp->f_path, &stat, STATX_TYPE, AT_STATX_SYNC_AS_STAT);
 158 #elif defined(HAVE_2ARGS_VFS_GETATTR)
 159         rc = vfs_getattr(&fp->f_path, &stat);
 160 #else
 161         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 162 #endif
 163         if (rc) {
 164                 filp_close(fp, 0);
 165                 return (-rc);
 166         }
 167
 168         vp = vn_alloc(KM_SLEEP);
 169         if (!vp) {
 170                 filp_close(fp, 0);
 171                 return (ENOMEM);
 172         }
 173
 174         saved_gfp = mapping_gfp_mask(fp->f_mapping);
 175         mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
 176
 177         mutex_enter(&vp->v_lock);
 178         vp->v_type = vn_mode_to_vtype(stat.mode);
 179         vp->v_file = fp;
 180         vp->v_gfp_mask = saved_gfp;
 181         *vpp = vp;
 182         mutex_exit(&vp->v_lock);
 183
 184         return (0);
 185 } /* vn_open() */
 186 EXPORT_SYMBOL(vn_open);
 187
 188 int
 189 vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
 190           vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
 191 {
 192         char *realpath;
 193         int len, rc;
 194
 195         ASSERT(vp == rootdir);
 196
 197         len = strlen(path) + 2;
 198         realpath = kmalloc(len, kmem_flags_convert(KM_SLEEP));
 199         if (!realpath)
 200                 return (ENOMEM);
 201
 202         (void)snprintf(realpath, len, "/%s", path);
 203         rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
 204         kfree(realpath);
 205
 206         return (rc);
 207 } /* vn_openat() */
 208 EXPORT_SYMBOL(vn_openat);
 209
 210 int
 211 vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
 212         uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
 213 {
 214         struct file *fp = vp->v_file;
 215         loff_t offset = off;
 216         int rc;
 217
 218         ASSERT(uio == UIO_WRITE || uio == UIO_READ);
 219         ASSERT(seg == UIO_SYSSPACE);
 220         ASSERT((ioflag & ~FAPPEND) == 0);
 221
 222         if (ioflag & FAPPEND)
 223                 offset = fp->f_pos;
 224
 225         if (uio & UIO_WRITE)
 226                 rc = spl_kernel_write(fp, addr, len, &offset);
 227         else
 228                 rc = spl_kernel_read(fp, addr, len, &offset);
 229
 230         fp->f_pos = offset;
 231
 232         if (rc < 0)
 233                 return (-rc);
 234
 235         if (residp) {
 236                 *residp = len - rc;
 237         } else {
 238                 if (rc != len)
 239                         return (EIO);
 240         }
 241
 242         return (0);
 243 } /* vn_rdwr() */
 244 EXPORT_SYMBOL(vn_rdwr);
 245
 246 int
 247 vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
 248 {
 249         int rc;
 250
 251         ASSERT(vp);
 252         ASSERT(vp->v_file);
 253
 254         mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
 255         rc = filp_close(vp->v_file, 0);
 256         vn_free(vp);
 257
 258         return (-rc);
 259 } /* vn_close() */
 260 EXPORT_SYMBOL(vn_close);
 261
 262 /* vn_seek() does not actually seek it only performs bounds checking on the
 263  * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
 264  * anything more serious. */
 265 int
 266 vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 267 {
 268         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
 269 }
 270 EXPORT_SYMBOL(vn_seek);
 271
 272 /*
 273  * spl_basename() takes a NULL-terminated string s as input containing a path.
 274  * It returns a char pointer to a string and a length that describe the
 275  * basename of the path. If the basename is not "." or "/", it will be an index
 276  * into the string. While the string should be NULL terminated, the section
 277  * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
 278  * CC0. Anyone wishing to reuse it in another codebase may pick either license.
 279  */
 280 static void
 281 spl_basename(const char *s, const char **str, int *len)
 282 {
 283         size_t i, end;
 284
 285         ASSERT(str);
 286         ASSERT(len);
 287
 288         if (!s || !*s) {
 289                 *str = ".";
 290                 *len = 1;
 291                 return;
 292         }
 293
 294         i = strlen(s) - 1;
 295
 296         while (i && s[i--] == '/');
 297
 298         if (i == 0) {
 299                 *str = "/";
 300                 *len = 1;
 301                 return;
 302         }
 303
 304         end = i;
 305
 306         for (end = i; i; i--) {
 307                 if (s[i] == '/') {
 308                         *str = &s[i+1];
 309                         *len = end - i + 1;
 310                         return;
 311                 }
 312         }
 313
 314         *str = s;
 315         *len = end + 1;
 316 }
 317
 318 static struct dentry *
 319 spl_kern_path_locked(const char *name, struct path *path)
 320 {
 321         struct path parent;
 322         struct dentry *dentry;
 323         const char *basename;
 324         int len;
 325         int rc;
 326
 327         ASSERT(name);
 328         ASSERT(path);
 329
 330         spl_basename(name, &basename, &len);
 331
 332         /* We do not accept "." or ".." */
 333         if (len <= 2 && basename[0] == '.')
 334                 if (len == 1 || basename[1] == '.')
 335                         return (ERR_PTR(-EACCES));
 336
 337         rc = kern_path(name, LOOKUP_PARENT, &parent);
 338         if (rc)
 339                 return (ERR_PTR(rc));
 340
 341         /* use I_MUTEX_PARENT because vfs_unlink needs it */
 342         spl_inode_lock_nested(parent.dentry->d_inode, I_MUTEX_PARENT);
 343
 344         dentry = lookup_one_len(basename, parent.dentry, len);
 345         if (IS_ERR(dentry)) {
 346                 spl_inode_unlock(parent.dentry->d_inode);
 347                 path_put(&parent);
 348         } else {
 349                 *path = parent;
 350         }
 351
 352         return (dentry);
 353 }
 354
 355 /* Based on do_unlinkat() from linux/fs/namei.c */
 356 int
 357 vn_remove(const char *path, uio_seg_t seg, int flags)
 358 {
 359         struct dentry *dentry;
 360         struct path parent;
 361         struct inode *inode = NULL;
 362         int rc = 0;
 363
 364         ASSERT(seg == UIO_SYSSPACE);
 365         ASSERT(flags == RMFILE);
 366
 367         dentry = spl_kern_path_locked(path, &parent);
 368         rc = PTR_ERR(dentry);
 369         if (!IS_ERR(dentry)) {
 370                 if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
 371                         rc = 0;
 372                         goto slashes;
 373                 }
 374
 375                 inode = dentry->d_inode;
 376                 if (inode) {
 377                         atomic_inc(&inode->i_count);
 378                 } else {
 379                         rc = 0;
 380                         goto slashes;
 381                 }
 382
 383 #ifdef HAVE_2ARGS_VFS_UNLINK
 384                 rc = vfs_unlink(parent.dentry->d_inode, dentry);
 385 #else
 386                 rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
 387 #endif /* HAVE_2ARGS_VFS_UNLINK */
 388 exit1:
 389                 dput(dentry);
 390         } else {
 391                 return (-rc);
 392         }
 393
 394         spl_inode_unlock(parent.dentry->d_inode);
 395         if (inode)
 396                 iput(inode);    /* truncate the inode here */
 397
 398         path_put(&parent);
 399         return (-rc);
 400
 401 slashes:
 402         rc = !dentry->d_inode ? -ENOENT :
 403             S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
 404         goto exit1;
 405 } /* vn_remove() */
 406 EXPORT_SYMBOL(vn_remove);
 407
 408 /* Based on do_rename() from linux/fs/namei.c */
 409 int
 410 vn_rename(const char *oldname, const char *newname, int x1)
 411 {
 412         struct dentry *old_dir, *new_dir;
 413         struct dentry *old_dentry, *new_dentry;
 414         struct dentry *trap;
 415         struct path old_parent, new_parent;
 416         int rc = 0;
 417
 418         old_dentry = spl_kern_path_locked(oldname, &old_parent);
 419         if (IS_ERR(old_dentry)) {
 420                 rc = PTR_ERR(old_dentry);
 421                 goto exit;
 422         }
 423
 424         spl_inode_unlock(old_parent.dentry->d_inode);
 425
 426         new_dentry = spl_kern_path_locked(newname, &new_parent);
 427         if (IS_ERR(new_dentry)) {
 428                 rc = PTR_ERR(new_dentry);
 429                 goto exit2;
 430         }
 431
 432         spl_inode_unlock(new_parent.dentry->d_inode);
 433
 434         rc = -EXDEV;
 435         if (old_parent.mnt != new_parent.mnt)
 436                 goto exit3;
 437
 438         old_dir = old_parent.dentry;
 439         new_dir = new_parent.dentry;
 440         trap = lock_rename(new_dir, old_dir);
 441
 442         /* source should not be ancestor of target */
 443         rc = -EINVAL;
 444         if (old_dentry == trap)
 445                 goto exit4;
 446
 447         /* target should not be an ancestor of source */
 448         rc = -ENOTEMPTY;
 449         if (new_dentry == trap)
 450                 goto exit4;
 451
 452         /* source must exist */
 453         rc = -ENOENT;
 454         if (!old_dentry->d_inode)
 455                 goto exit4;
 456
 457         /* unless the source is a directory trailing slashes give -ENOTDIR */
 458         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
 459                 rc = -ENOTDIR;
 460                 if (old_dentry->d_name.name[old_dentry->d_name.len])
 461                         goto exit4;
 462                 if (new_dentry->d_name.name[new_dentry->d_name.len])
 463                         goto exit4;
 464         }
 465
 466 #if defined(HAVE_4ARGS_VFS_RENAME)
 467         rc = vfs_rename(old_dir->d_inode, old_dentry,
 468             new_dir->d_inode, new_dentry);
 469 #elif defined(HAVE_5ARGS_VFS_RENAME)
 470         rc = vfs_rename(old_dir->d_inode, old_dentry,
 471             new_dir->d_inode, new_dentry, NULL);
 472 #else
 473         rc = vfs_rename(old_dir->d_inode, old_dentry,
 474             new_dir->d_inode, new_dentry, NULL, 0);
 475 #endif
 476 exit4:
 477         unlock_rename(new_dir, old_dir);
 478 exit3:
 479         dput(new_dentry);
 480         path_put(&new_parent);
 481 exit2:
 482         dput(old_dentry);
 483         path_put(&old_parent);
 484 exit:
 485         return (-rc);
 486 }
 487 EXPORT_SYMBOL(vn_rename);
 488
 489 int
 490 vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 491 {
 492         struct file *fp;
 493         struct kstat stat;
 494         int rc;
 495
 496         ASSERT(vp);
 497         ASSERT(vp->v_file);
 498         ASSERT(vap);
 499
 500         fp = vp->v_file;
 501
 502 #if defined(HAVE_4ARGS_VFS_GETATTR)
 503         rc = vfs_getattr(&fp->f_path, &stat, STATX_BASIC_STATS,
 504             AT_STATX_SYNC_AS_STAT);
 505 #elif defined(HAVE_2ARGS_VFS_GETATTR)
 506         rc = vfs_getattr(&fp->f_path, &stat);
 507 #else
 508         rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
 509 #endif
 510         if (rc)
 511                 return (-rc);
 512
 513         vap->va_type          = vn_mode_to_vtype(stat.mode);
 514         vap->va_mode          = stat.mode;
 515         vap->va_uid           = KUID_TO_SUID(stat.uid);
 516         vap->va_gid           = KGID_TO_SGID(stat.gid);
 517         vap->va_fsid          = 0;
 518         vap->va_nodeid        = stat.ino;
 519         vap->va_nlink         = stat.nlink;
 520         vap->va_size          = stat.size;
 521         vap->va_blksize       = stat.blksize;
 522         vap->va_atime         = stat.atime;
 523         vap->va_mtime         = stat.mtime;
 524         vap->va_ctime         = stat.ctime;
 525         vap->va_rdev          = stat.rdev;
 526         vap->va_nblocks       = stat.blocks;
 527
 528         return (0);
 529 }
 530 EXPORT_SYMBOL(vn_getattr);
 531
 532 int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 533 {
 534         int datasync = 0;
 535         int error;
 536         int fstrans;
 537
 538         ASSERT(vp);
 539         ASSERT(vp->v_file);
 540
 541         if (flags & FDSYNC)
 542                 datasync = 1;
 543
 544         /*
 545          * May enter XFS which generates a warning when PF_FSTRANS is set.
 546          * To avoid this the flag is cleared over vfs_sync() and then reset.
 547          */
 548         fstrans = __spl_pf_fstrans_check();
 549         if (fstrans)
 550                 current->flags &= ~(__SPL_PF_FSTRANS);
 551
 552         error = -spl_filp_fsync(vp->v_file, datasync);
 553         if (fstrans)
 554                 current->flags |= __SPL_PF_FSTRANS;
 555
 556         return (error);
 557 } /* vn_fsync() */
 558 EXPORT_SYMBOL(vn_fsync);
 559
 560 int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
 561     offset_t offset, void *x6, void *x7)
 562 {
 563         int error = EOPNOTSUPP;
 564 #ifdef FALLOC_FL_PUNCH_HOLE
 565         int fstrans;
 566 #endif
 567
 568         if (cmd != F_FREESP || bfp->l_whence != 0)
 569                 return (EOPNOTSUPP);
 570
 571         ASSERT(vp);
 572         ASSERT(vp->v_file);
 573         ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
 574
 575 #ifdef FALLOC_FL_PUNCH_HOLE
 576         /*
 577          * May enter XFS which generates a warning when PF_FSTRANS is set.
 578          * To avoid this the flag is cleared over vfs_sync() and then reset.
 579          */
 580         fstrans = __spl_pf_fstrans_check();
 581         if (fstrans)
 582                 current->flags &= ~(__SPL_PF_FSTRANS);
 583
 584         /*
 585          * When supported by the underlying file system preferentially
 586          * use the fallocate() callback to preallocate the space.
 587          */
 588         error = -spl_filp_fallocate(vp->v_file,
 589             FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 590             bfp->l_start, bfp->l_len);
 591
 592         if (fstrans)
 593                 current->flags |= __SPL_PF_FSTRANS;
 594
 595         if (error == 0)
 596                 return (0);
 597 #endif
 598
 599 #ifdef HAVE_INODE_TRUNCATE_RANGE
 600         if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
 601             vp->v_file->f_dentry->d_inode->i_op &&
 602             vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
 603                 off_t end = bfp->l_start + bfp->l_len;
 604                 /*
 605                  * Judging from the code in shmem_truncate_range(),
 606                  * it seems the kernel expects the end offset to be
 607                  * inclusive and aligned to the end of a page.
 608                  */
 609                 if (end % PAGE_SIZE != 0) {
 610                         end &= ~(off_t)(PAGE_SIZE - 1);
 611                         if (end <= bfp->l_start)
 612                                 return (0);
 613                 }
 614                 --end;
 615
 616                 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
 617                         vp->v_file->f_dentry->d_inode,
 618                         bfp->l_start, end
 619                 );
 620                 return (0);
 621         }
 622 #endif
 623
 624         return (error);
 625 }
 626 EXPORT_SYMBOL(vn_space);
 627
 628 /* Function must be called while holding the vn_file_lock */
 629 static file_t *
 630 file_find(int fd, struct task_struct *task)
 631 {
 632         file_t *fp;
 633
 634         ASSERT(spin_is_locked(&vn_file_lock));
 635
 636         list_for_each_entry(fp, &vn_file_list,  f_list) {
 637                 if (fd == fp->f_fd && fp->f_task == task) {
 638                         ASSERT(atomic_read(&fp->f_ref) != 0);
 639                         return fp;
 640                 }
 641         }
 642
 643         return NULL;
 644 } /* file_find() */
 645
 646 file_t *
 647 vn_getf(int fd)
 648 {
 649         struct kstat stat;
 650         struct file *lfp;
 651         file_t *fp;
 652         vnode_t *vp;
 653         int rc = 0;
 654
 655         if (fd < 0)
 656                 return (NULL);
 657
 658         /* Already open just take an extra reference */
 659         spin_lock(&vn_file_lock);
 660
 661         fp = file_find(fd, current);
 662         if (fp) {
 663                 lfp = fget(fd);
 664                 fput(fp->f_file);
 665                 /*
 666                  * areleasef() can cause us to see a stale reference when
 667                  * userspace has reused a file descriptor before areleasef()
 668                  * has run. fput() the stale reference and replace it. We
 669                  * retain the original reference count such that the concurrent
 670                  * areleasef() will decrement its reference and terminate.
 671                  */
 672                 if (lfp != fp->f_file) {
 673                         fp->f_file = lfp;
 674                         fp->f_vnode->v_file = lfp;
 675                 }
 676                 atomic_inc(&fp->f_ref);
 677                 spin_unlock(&vn_file_lock);
 678                 return (fp);
 679         }
 680
 681         spin_unlock(&vn_file_lock);
 682
 683         /* File was not yet opened create the object and setup */
 684         fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
 685         if (fp == NULL)
 686                 goto out;
 687
 688         mutex_enter(&fp->f_lock);
 689
 690         fp->f_fd = fd;
 691         fp->f_task = current;
 692         fp->f_offset = 0;
 693         atomic_inc(&fp->f_ref);
 694
 695         lfp = fget(fd);
 696         if (lfp == NULL)
 697                 goto out_mutex;
 698
 699         vp = vn_alloc(KM_SLEEP);
 700         if (vp == NULL)
 701                 goto out_fget;
 702
 703 #if defined(HAVE_4ARGS_VFS_GETATTR)
 704         rc = vfs_getattr(&lfp->f_path, &stat, STATX_TYPE, AT_STATX_SYNC_AS_STAT);
 705 #elif defined(HAVE_2ARGS_VFS_GETATTR)
 706         rc = vfs_getattr(&lfp->f_path, &stat);
 707 #else
 708         rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
 709 #endif
 710         if (rc)
 711                 goto out_vnode;
 712
 713         mutex_enter(&vp->v_lock);
 714         vp->v_type = vn_mode_to_vtype(stat.mode);
 715         vp->v_file = lfp;
 716         mutex_exit(&vp->v_lock);
 717
 718         fp->f_vnode = vp;
 719         fp->f_file = lfp;
 720
 721         /* Put it on the tracking list */
 722         spin_lock(&vn_file_lock);
 723         list_add(&fp->f_list, &vn_file_list);
 724         spin_unlock(&vn_file_lock);
 725
 726         mutex_exit(&fp->f_lock);
 727         return (fp);
 728
 729 out_vnode:
 730         vn_free(vp);
 731 out_fget:
 732         fput(lfp);
 733 out_mutex:
 734         mutex_exit(&fp->f_lock);
 735         kmem_cache_free(vn_file_cache, fp);
 736 out:
 737         return (NULL);
 738 } /* getf() */
 739 EXPORT_SYMBOL(getf);
 740
 741 static void releasef_locked(file_t *fp)
 742 {
 743         ASSERT(fp->f_file);
 744         ASSERT(fp->f_vnode);
 745
 746         /* Unlinked from list, no refs, safe to free outside mutex */
 747         fput(fp->f_file);
 748         vn_free(fp->f_vnode);
 749
 750         kmem_cache_free(vn_file_cache, fp);
 751 }
 752
 753 void
 754 vn_releasef(int fd)
 755 {
 756         areleasef(fd, P_FINFO(current));
 757 }
 758 EXPORT_SYMBOL(releasef);
 759
 760 void
 761 vn_areleasef(int fd, uf_info_t *fip)
 762 {
 763         file_t *fp;
 764         struct task_struct *task = (struct task_struct *)fip;
 765
 766         if (fd < 0)
 767                 return;
 768
 769         spin_lock(&vn_file_lock);
 770         fp = file_find(fd, task);
 771         if (fp) {
 772                 atomic_dec(&fp->f_ref);
 773                 if (atomic_read(&fp->f_ref) > 0) {
 774                         spin_unlock(&vn_file_lock);
 775                         return;
 776                 }
 777
 778                 list_del(&fp->f_list);
 779                 releasef_locked(fp);
 780         }
 781         spin_unlock(&vn_file_lock);
 782
 783         return;
 784 } /* releasef() */
 785 EXPORT_SYMBOL(areleasef);
 786
 787
 788 static void
 789 #ifdef HAVE_SET_FS_PWD_WITH_CONST
 790 vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
 791 #else
 792 vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
 793 #endif /* HAVE_SET_FS_PWD_WITH_CONST */
 794 {
 795         struct path old_pwd;
 796
 797 #ifdef HAVE_FS_STRUCT_SPINLOCK
 798         spin_lock(&fs->lock);
 799         old_pwd = fs->pwd;
 800         fs->pwd = *path;
 801         path_get(path);
 802         spin_unlock(&fs->lock);
 803 #else
 804         write_lock(&fs->lock);
 805         old_pwd = fs->pwd;
 806         fs->pwd = *path;
 807         path_get(path);
 808         write_unlock(&fs->lock);
 809 #endif /* HAVE_FS_STRUCT_SPINLOCK */
 810
 811         if (old_pwd.dentry)
 812                 path_put(&old_pwd);
 813 }
 814
 815 int
 816 vn_set_pwd(const char *filename)
 817 {
 818         struct path path;
 819         mm_segment_t saved_fs;
 820         int rc;
 821
 822         /*
 823          * user_path_dir() and __user_walk() both expect 'filename' to be
 824          * a user space address so we must briefly increase the data segment
 825          * size to ensure strncpy_from_user() does not fail with -EFAULT.
 826          */
 827         saved_fs = get_fs();
 828         set_fs(get_ds());
 829
 830         rc = user_path_dir(filename, &path);
 831         if (rc)
 832                 goto out;
 833
 834         rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 835         if (rc)
 836                 goto dput_and_out;
 837
 838         vn_set_fs_pwd(current->fs, &path);
 839
 840 dput_and_out:
 841         path_put(&path);
 842 out:
 843         set_fs(saved_fs);
 844
 845         return (-rc);
 846 } /* vn_set_pwd() */
 847 EXPORT_SYMBOL(vn_set_pwd);
 848
 849 static int
 850 vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 851 {
 852         struct vnode *vp = buf;
 853
 854         mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
 855
 856         return (0);
 857 } /* vn_cache_constructor() */
 858
 859 static void
 860 vn_cache_destructor(void *buf, void *cdrarg)
 861 {
 862         struct vnode *vp = buf;
 863
 864         mutex_destroy(&vp->v_lock);
 865 } /* vn_cache_destructor() */
 866
 867 static int
 868 vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
 869 {
 870         file_t *fp = buf;
 871
 872         atomic_set(&fp->f_ref, 0);
 873         mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
 874         INIT_LIST_HEAD(&fp->f_list);
 875
 876         return (0);
 877 } /* file_cache_constructor() */
 878
 879 static void
 880 vn_file_cache_destructor(void *buf, void *cdrarg)
 881 {
 882         file_t *fp = buf;
 883
 884         mutex_destroy(&fp->f_lock);
 885 } /* vn_file_cache_destructor() */
 886
 887 int
 888 spl_vn_init(void)
 889 {
 890         vn_cache = kmem_cache_create("spl_vn_cache",
 891                                      sizeof(struct vnode), 64,
 892                                      vn_cache_constructor,
 893                                      vn_cache_destructor,
 894                                      NULL, NULL, NULL, 0);
 895
 896         vn_file_cache = kmem_cache_create("spl_vn_file_cache",
 897                                           sizeof(file_t), 64,
 898                                           vn_file_cache_constructor,
 899                                           vn_file_cache_destructor,
 900                                           NULL, NULL, NULL, 0);
 901         return (0);
 902 } /* vn_init() */
 903
 904 void
 905 spl_vn_fini(void)
 906 {
 907         file_t *fp, *next_fp;
 908         int leaked = 0;
 909
 910         spin_lock(&vn_file_lock);
 911
 912         list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
 913                 list_del(&fp->f_list);
 914                 releasef_locked(fp);
 915                 leaked++;
 916         }
 917
 918         spin_unlock(&vn_file_lock);
 919
 920         if (leaked > 0)
 921                 printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
 922
 923         kmem_cache_destroy(vn_file_cache);
 924         kmem_cache_destroy(vn_cache);
 925
 926         return;
 927 } /* vn_fini() */