]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blobdiff - spl/module/spl/spl-vnode.c
UBUNTU: SAUCE: (noup) Update spl to 0.6.5.9-1, zfs to 0.6.5.9-2
[mirror_ubuntu-artful-kernel.git] / spl / module / spl / spl-vnode.c
diff --git a/spl/module/spl/spl-vnode.c b/spl/module/spl/spl-vnode.c
new file mode 100644 (file)
index 0000000..f1ebcd2
--- /dev/null
@@ -0,0 +1,909 @@
+/*****************************************************************************\
+ *  Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
+ *  Copyright (C) 2007 The Regents of the University of California.
+ *  Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
+ *  Written by Brian Behlendorf <behlendorf1@llnl.gov>.
+ *  UCRL-CODE-235197
+ *
+ *  This file is part of the SPL, Solaris Porting Layer.
+ *  For details, see <http://zfsonlinux.org/>.
+ *
+ *  The SPL is free software; you can redistribute it and/or modify it
+ *  under the terms of the GNU General Public License as published by the
+ *  Free Software Foundation; either version 2 of the License, or (at your
+ *  option) any later version.
+ *
+ *  The SPL is distributed in the hope that it will be useful, but WITHOUT
+ *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ *  for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along
+ *  with the SPL.  If not, see <http://www.gnu.org/licenses/>.
+ *****************************************************************************
+ *  Solaris Porting Layer (SPL) Vnode Implementation.
+\*****************************************************************************/
+
+#include <sys/cred.h>
+#include <sys/vnode.h>
+#include <sys/kmem_cache.h>
+#include <linux/falloc.h>
+#include <linux/file_compat.h>
+
+vnode_t *rootdir = (vnode_t *)0xabcd1234;
+EXPORT_SYMBOL(rootdir);
+
+static spl_kmem_cache_t *vn_cache;
+static spl_kmem_cache_t *vn_file_cache;
+
+static DEFINE_SPINLOCK(vn_file_lock);
+static LIST_HEAD(vn_file_list);
+
+vtype_t
+vn_mode_to_vtype(mode_t mode)
+{
+       if (S_ISREG(mode))
+               return VREG;
+
+       if (S_ISDIR(mode))
+               return VDIR;
+
+       if (S_ISCHR(mode))
+               return VCHR;
+
+       if (S_ISBLK(mode))
+               return VBLK;
+
+       if (S_ISFIFO(mode))
+               return VFIFO;
+
+       if (S_ISLNK(mode))
+               return VLNK;
+
+       if (S_ISSOCK(mode))
+               return VSOCK;
+
+       if (S_ISCHR(mode))
+               return VCHR;
+
+       return VNON;
+} /* vn_mode_to_vtype() */
+EXPORT_SYMBOL(vn_mode_to_vtype);
+
+mode_t
+vn_vtype_to_mode(vtype_t vtype)
+{
+       if (vtype == VREG)
+               return S_IFREG;
+
+       if (vtype == VDIR)
+               return S_IFDIR;
+
+       if (vtype == VCHR)
+               return S_IFCHR;
+
+       if (vtype == VBLK)
+               return S_IFBLK;
+
+       if (vtype == VFIFO)
+               return S_IFIFO;
+
+       if (vtype == VLNK)
+               return S_IFLNK;
+
+       if (vtype == VSOCK)
+               return S_IFSOCK;
+
+       return VNON;
+} /* vn_vtype_to_mode() */
+EXPORT_SYMBOL(vn_vtype_to_mode);
+
+vnode_t *
+vn_alloc(int flag)
+{
+       vnode_t *vp;
+
+       vp = kmem_cache_alloc(vn_cache, flag);
+       if (vp != NULL) {
+               vp->v_file = NULL;
+               vp->v_type = 0;
+       }
+
+       return (vp);
+} /* vn_alloc() */
+EXPORT_SYMBOL(vn_alloc);
+
+void
+vn_free(vnode_t *vp)
+{
+       kmem_cache_free(vn_cache, vp);
+} /* vn_free() */
+EXPORT_SYMBOL(vn_free);
+
+int
+vn_open(const char *path, uio_seg_t seg, int flags, int mode,
+       vnode_t **vpp, int x1, void *x2)
+{
+       struct file *fp;
+       struct kstat stat;
+       int rc, saved_umask = 0;
+       gfp_t saved_gfp;
+       vnode_t *vp;
+
+       ASSERT(flags & (FWRITE | FREAD));
+       ASSERT(seg == UIO_SYSSPACE);
+       ASSERT(vpp);
+       *vpp = NULL;
+
+       if (!(flags & FCREAT) && (flags & FWRITE))
+               flags |= FEXCL;
+
+       /* Note for filp_open() the two low bits must be remapped to mean:
+        * 01 - read-only  -> 00 read-only
+        * 10 - write-only -> 01 write-only
+        * 11 - read-write -> 10 read-write
+        */
+       flags--;
+
+       if (flags & FCREAT)
+               saved_umask = xchg(&current->fs->umask, 0);
+
+       fp = filp_open(path, flags, mode);
+
+       if (flags & FCREAT)
+               (void)xchg(&current->fs->umask, saved_umask);
+
+       if (IS_ERR(fp))
+               return (-PTR_ERR(fp));
+
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&fp->f_path, &stat);
+#else
+       rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
+#endif
+       if (rc) {
+               filp_close(fp, 0);
+               return (-rc);
+       }
+
+       vp = vn_alloc(KM_SLEEP);
+       if (!vp) {
+               filp_close(fp, 0);
+               return (ENOMEM);
+       }
+
+       saved_gfp = mapping_gfp_mask(fp->f_mapping);
+       mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
+
+       mutex_enter(&vp->v_lock);
+       vp->v_type = vn_mode_to_vtype(stat.mode);
+       vp->v_file = fp;
+       vp->v_gfp_mask = saved_gfp;
+       *vpp = vp;
+       mutex_exit(&vp->v_lock);
+
+       return (0);
+} /* vn_open() */
+EXPORT_SYMBOL(vn_open);
+
+int
+vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
+         vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
+{
+       char *realpath;
+       int len, rc;
+
+       ASSERT(vp == rootdir);
+
+       len = strlen(path) + 2;
+       realpath = kmalloc(len, kmem_flags_convert(KM_SLEEP));
+       if (!realpath)
+               return (ENOMEM);
+
+       (void)snprintf(realpath, len, "/%s", path);
+       rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
+       kfree(realpath);
+
+       return (rc);
+} /* vn_openat() */
+EXPORT_SYMBOL(vn_openat);
+
+int
+vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
+       uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
+{
+       loff_t offset;
+       mm_segment_t saved_fs;
+       struct file *fp;
+       int rc;
+
+       ASSERT(uio == UIO_WRITE || uio == UIO_READ);
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+       ASSERT(seg == UIO_SYSSPACE);
+       ASSERT((ioflag & ~FAPPEND) == 0);
+       ASSERT(x2 == RLIM64_INFINITY);
+
+       fp = vp->v_file;
+
+       offset = off;
+       if (ioflag & FAPPEND)
+               offset = fp->f_pos;
+
+       /* Writable user data segment must be briefly increased for this
+        * process so we can use the user space read call paths to write
+        * in to memory allocated by the kernel. */
+       saved_fs = get_fs();
+        set_fs(get_ds());
+
+       if (uio & UIO_WRITE)
+               rc = vfs_write(fp, addr, len, &offset);
+       else
+               rc = vfs_read(fp, addr, len, &offset);
+
+       set_fs(saved_fs);
+       fp->f_pos = offset;
+
+       if (rc < 0)
+               return (-rc);
+
+       if (residp) {
+               *residp = len - rc;
+       } else {
+               if (rc != len)
+                       return (EIO);
+       }
+
+       return (0);
+} /* vn_rdwr() */
+EXPORT_SYMBOL(vn_rdwr);
+
+int
+vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
+{
+       int rc;
+
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+
+       mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
+       rc = filp_close(vp->v_file, 0);
+       vn_free(vp);
+
+       return (-rc);
+} /* vn_close() */
+EXPORT_SYMBOL(vn_close);
+
+/* vn_seek() does not actually seek it only performs bounds checking on the
+ * proposed seek.  We perform minimal checking and allow vn_rdwr() to catch
+ * anything more serious. */
+int
+vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
+{
+       return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
+}
+EXPORT_SYMBOL(vn_seek);
+
+/*
+ * spl_basename() takes a NULL-terminated string s as input containing a path.
+ * It returns a char pointer to a string and a length that describe the
+ * basename of the path. If the basename is not "." or "/", it will be an index
+ * into the string. While the string should be NULL terminated, the section
+ * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
+ * CC0. Anyone wishing to reuse it in another codebase may pick either license.
+ */
+static void
+spl_basename(const char *s, const char **str, int *len)
+{
+       size_t i, end;
+
+       ASSERT(str);
+       ASSERT(len);
+
+       if (!s || !*s) {
+               *str = ".";
+               *len = 1;
+               return;
+       }
+
+       i = strlen(s) - 1;
+
+       while (i && s[i--] == '/');
+
+       if (i == 0) {
+               *str = "/";
+               *len = 1;
+               return;
+       }
+
+       end = i;
+
+       for (end = i; i; i--) {
+               if (s[i] == '/') {
+                       *str = &s[i+1];
+                       *len = end - i + 1;
+                       return;
+               }
+       }
+
+       *str = s;
+       *len = end + 1;
+}
+
+static struct dentry *
+spl_kern_path_locked(const char *name, struct path *path)
+{
+       struct path parent;
+       struct dentry *dentry;
+       const char *basename;
+       int len;
+       int rc;
+
+       ASSERT(name);
+       ASSERT(path);
+
+       spl_basename(name, &basename, &len);
+
+       /* We do not accept "." or ".." */
+       if (len <= 2 && basename[0] == '.')
+               if (len == 1 || basename[1] == '.')
+                       return (ERR_PTR(-EACCES));
+
+       rc = kern_path(name, LOOKUP_PARENT, &parent);
+       if (rc)
+               return (ERR_PTR(rc));
+
+       /* use I_MUTEX_PARENT because vfs_unlink needs it */
+       spl_inode_lock_nested(parent.dentry->d_inode, I_MUTEX_PARENT);
+
+       dentry = lookup_one_len(basename, parent.dentry, len);
+       if (IS_ERR(dentry)) {
+               spl_inode_unlock(parent.dentry->d_inode);
+               path_put(&parent);
+       } else {
+               *path = parent;
+       }
+
+       return (dentry);
+}
+
+/* Based on do_unlinkat() from linux/fs/namei.c */
+int
+vn_remove(const char *path, uio_seg_t seg, int flags)
+{
+       struct dentry *dentry;
+       struct path parent;
+       struct inode *inode = NULL;
+       int rc = 0;
+
+       ASSERT(seg == UIO_SYSSPACE);
+       ASSERT(flags == RMFILE);
+
+       dentry = spl_kern_path_locked(path, &parent);
+       rc = PTR_ERR(dentry);
+       if (!IS_ERR(dentry)) {
+               if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
+                       rc = 0;
+                       goto slashes;
+               }
+
+               inode = dentry->d_inode;
+               if (inode) {
+                       atomic_inc(&inode->i_count);
+               } else {
+                       rc = 0;
+                       goto slashes;
+               }
+
+#ifdef HAVE_2ARGS_VFS_UNLINK
+               rc = vfs_unlink(parent.dentry->d_inode, dentry);
+#else
+               rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
+#endif /* HAVE_2ARGS_VFS_UNLINK */
+exit1:
+               dput(dentry);
+       } else {
+               return (-rc);
+       }
+
+       spl_inode_unlock(parent.dentry->d_inode);
+       if (inode)
+               iput(inode);    /* truncate the inode here */
+
+       path_put(&parent);
+       return (-rc);
+
+slashes:
+       rc = !dentry->d_inode ? -ENOENT :
+           S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
+       goto exit1;
+} /* vn_remove() */
+EXPORT_SYMBOL(vn_remove);
+
+/* Based on do_rename() from linux/fs/namei.c */
+int
+vn_rename(const char *oldname, const char *newname, int x1)
+{
+       struct dentry *old_dir, *new_dir;
+       struct dentry *old_dentry, *new_dentry;
+       struct dentry *trap;
+       struct path old_parent, new_parent;
+       int rc = 0;
+
+       old_dentry = spl_kern_path_locked(oldname, &old_parent);
+       if (IS_ERR(old_dentry)) {
+               rc = PTR_ERR(old_dentry);
+               goto exit;
+       }
+
+       spl_inode_unlock(old_parent.dentry->d_inode);
+
+       new_dentry = spl_kern_path_locked(newname, &new_parent);
+       if (IS_ERR(new_dentry)) {
+               rc = PTR_ERR(new_dentry);
+               goto exit2;
+       }
+
+       spl_inode_unlock(new_parent.dentry->d_inode);
+
+       rc = -EXDEV;
+       if (old_parent.mnt != new_parent.mnt)
+               goto exit3;
+
+       old_dir = old_parent.dentry;
+       new_dir = new_parent.dentry;
+       trap = lock_rename(new_dir, old_dir);
+
+       /* source should not be ancestor of target */
+       rc = -EINVAL;
+       if (old_dentry == trap)
+               goto exit4;
+
+       /* target should not be an ancestor of source */
+       rc = -ENOTEMPTY;
+       if (new_dentry == trap)
+               goto exit4;
+
+       /* source must exist */
+       rc = -ENOENT;
+       if (!old_dentry->d_inode)
+               goto exit4;
+
+       /* unless the source is a directory trailing slashes give -ENOTDIR */
+       if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
+               rc = -ENOTDIR;
+               if (old_dentry->d_name.name[old_dentry->d_name.len])
+                       goto exit4;
+               if (new_dentry->d_name.name[new_dentry->d_name.len])
+                       goto exit4;
+       }
+
+#if defined(HAVE_4ARGS_VFS_RENAME)
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry);
+#elif defined(HAVE_5ARGS_VFS_RENAME)
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry, NULL);
+#else
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry, NULL, 0);
+#endif
+exit4:
+       unlock_rename(new_dir, old_dir);
+exit3:
+       dput(new_dentry);
+       path_put(&new_parent);
+exit2:
+       dput(old_dentry);
+       path_put(&old_parent);
+exit:
+       return (-rc);
+}
+EXPORT_SYMBOL(vn_rename);
+
+int
+vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
+{
+       struct file *fp;
+       struct kstat stat;
+       int rc;
+
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+       ASSERT(vap);
+
+       fp = vp->v_file;
+
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&fp->f_path, &stat);
+#else
+       rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
+#endif
+       if (rc)
+               return (-rc);
+
+       vap->va_type          = vn_mode_to_vtype(stat.mode);
+       vap->va_mode          = stat.mode;
+       vap->va_uid           = KUID_TO_SUID(stat.uid);
+       vap->va_gid           = KGID_TO_SGID(stat.gid);
+       vap->va_fsid          = 0;
+       vap->va_nodeid        = stat.ino;
+       vap->va_nlink         = stat.nlink;
+        vap->va_size          = stat.size;
+       vap->va_blksize       = stat.blksize;
+       vap->va_atime         = stat.atime;
+       vap->va_mtime         = stat.mtime;
+       vap->va_ctime         = stat.ctime;
+       vap->va_rdev          = stat.rdev;
+       vap->va_nblocks       = stat.blocks;
+
+       return (0);
+}
+EXPORT_SYMBOL(vn_getattr);
+
+int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
+{
+       int datasync = 0;
+       int error;
+       int fstrans;
+
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+
+       if (flags & FDSYNC)
+               datasync = 1;
+
+       /*
+        * May enter XFS which generates a warning when PF_FSTRANS is set.
+        * To avoid this the flag is cleared over vfs_sync() and then reset.
+        */
+       fstrans = spl_fstrans_check();
+       if (fstrans)
+               current->flags &= ~(PF_FSTRANS);
+
+       error = -spl_filp_fsync(vp->v_file, datasync);
+       if (fstrans)
+               current->flags |= PF_FSTRANS;
+
+       return (error);
+} /* vn_fsync() */
+EXPORT_SYMBOL(vn_fsync);
+
+int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
+    offset_t offset, void *x6, void *x7)
+{
+       int error = EOPNOTSUPP;
+
+       if (cmd != F_FREESP || bfp->l_whence != 0)
+               return (EOPNOTSUPP);
+
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+       ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
+
+#ifdef FALLOC_FL_PUNCH_HOLE
+       /*
+        * When supported by the underlying file system preferentially
+        * use the fallocate() callback to preallocate the space.
+        */
+       error = -spl_filp_fallocate(vp->v_file,
+           FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
+           bfp->l_start, bfp->l_len);
+       if (error == 0)
+               return (0);
+#endif
+
+#ifdef HAVE_INODE_TRUNCATE_RANGE
+       if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
+           vp->v_file->f_dentry->d_inode->i_op &&
+           vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
+               off_t end = bfp->l_start + bfp->l_len;
+               /*
+                * Judging from the code in shmem_truncate_range(),
+                * it seems the kernel expects the end offset to be
+                * inclusive and aligned to the end of a page.
+                */
+               if (end % PAGE_SIZE != 0) {
+                       end &= ~(off_t)(PAGE_SIZE - 1);
+                       if (end <= bfp->l_start)
+                               return (0);
+               }
+               --end;
+
+               vp->v_file->f_dentry->d_inode->i_op->truncate_range(
+                       vp->v_file->f_dentry->d_inode,
+                       bfp->l_start, end
+               );
+               return (0);
+       }
+#endif
+
+       return (error);
+}
+EXPORT_SYMBOL(vn_space);
+
+/* Function must be called while holding the vn_file_lock */
+static file_t *
+file_find(int fd, struct task_struct *task)
+{
+        file_t *fp;
+
+       ASSERT(spin_is_locked(&vn_file_lock));
+
+        list_for_each_entry(fp, &vn_file_list,  f_list) {
+               if (fd == fp->f_fd && fp->f_task == task) {
+                       ASSERT(atomic_read(&fp->f_ref) != 0);
+                        return fp;
+               }
+       }
+
+        return NULL;
+} /* file_find() */
+
+file_t *
+vn_getf(int fd)
+{
+        struct kstat stat;
+       struct file *lfp;
+       file_t *fp;
+       vnode_t *vp;
+       int rc = 0;
+
+       if (fd < 0)
+               return (NULL);
+
+       /* Already open just take an extra reference */
+       spin_lock(&vn_file_lock);
+
+       fp = file_find(fd, current);
+       if (fp) {
+               atomic_inc(&fp->f_ref);
+               spin_unlock(&vn_file_lock);
+               return (fp);
+       }
+
+       spin_unlock(&vn_file_lock);
+
+       /* File was not yet opened create the object and setup */
+       fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
+       if (fp == NULL)
+               goto out;
+
+       mutex_enter(&fp->f_lock);
+
+       fp->f_fd = fd;
+       fp->f_task = current;
+       fp->f_offset = 0;
+       atomic_inc(&fp->f_ref);
+
+       lfp = fget(fd);
+       if (lfp == NULL)
+               goto out_mutex;
+
+       vp = vn_alloc(KM_SLEEP);
+       if (vp == NULL)
+               goto out_fget;
+
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&lfp->f_path, &stat);
+#else
+       rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
+#endif
+        if (rc)
+               goto out_vnode;
+
+       mutex_enter(&vp->v_lock);
+       vp->v_type = vn_mode_to_vtype(stat.mode);
+       vp->v_file = lfp;
+       mutex_exit(&vp->v_lock);
+
+       fp->f_vnode = vp;
+       fp->f_file = lfp;
+
+       /* Put it on the tracking list */
+       spin_lock(&vn_file_lock);
+       list_add(&fp->f_list, &vn_file_list);
+       spin_unlock(&vn_file_lock);
+
+       mutex_exit(&fp->f_lock);
+       return (fp);
+
+out_vnode:
+       vn_free(vp);
+out_fget:
+       fput(lfp);
+out_mutex:
+       mutex_exit(&fp->f_lock);
+       kmem_cache_free(vn_file_cache, fp);
+out:
+        return (NULL);
+} /* getf() */
+EXPORT_SYMBOL(getf);
+
+static void releasef_locked(file_t *fp)
+{
+       ASSERT(fp->f_file);
+       ASSERT(fp->f_vnode);
+
+       /* Unlinked from list, no refs, safe to free outside mutex */
+       fput(fp->f_file);
+       vn_free(fp->f_vnode);
+
+       kmem_cache_free(vn_file_cache, fp);
+}
+
+void
+vn_releasef(int fd)
+{
+       areleasef(fd, P_FINFO(current));
+}
+EXPORT_SYMBOL(releasef);
+
+void
+vn_areleasef(int fd, uf_info_t *fip)
+{
+       file_t *fp;
+       struct task_struct *task = (struct task_struct *)fip;
+
+       if (fd < 0)
+               return;
+
+       spin_lock(&vn_file_lock);
+       fp = file_find(fd, task);
+       if (fp) {
+               atomic_dec(&fp->f_ref);
+               if (atomic_read(&fp->f_ref) > 0) {
+                       spin_unlock(&vn_file_lock);
+                       return;
+               }
+
+               list_del(&fp->f_list);
+               releasef_locked(fp);
+       }
+       spin_unlock(&vn_file_lock);
+
+       return;
+} /* releasef() */
+EXPORT_SYMBOL(areleasef);
+
+
+static void
+#ifdef HAVE_SET_FS_PWD_WITH_CONST
+vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
+#else
+vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
+#endif /* HAVE_SET_FS_PWD_WITH_CONST */
+{
+       struct path old_pwd;
+
+#ifdef HAVE_FS_STRUCT_SPINLOCK
+       spin_lock(&fs->lock);
+       old_pwd = fs->pwd;
+       fs->pwd = *path;
+       path_get(path);
+       spin_unlock(&fs->lock);
+#else
+       write_lock(&fs->lock);
+       old_pwd = fs->pwd;
+       fs->pwd = *path;
+       path_get(path);
+       write_unlock(&fs->lock);
+#endif /* HAVE_FS_STRUCT_SPINLOCK */
+
+       if (old_pwd.dentry)
+               path_put(&old_pwd);
+}
+
+int
+vn_set_pwd(const char *filename)
+{
+        struct path path;
+        mm_segment_t saved_fs;
+        int rc;
+
+        /*
+         * user_path_dir() and __user_walk() both expect 'filename' to be
+         * a user space address so we must briefly increase the data segment
+         * size to ensure strncpy_from_user() does not fail with -EFAULT.
+         */
+        saved_fs = get_fs();
+        set_fs(get_ds());
+
+        rc = user_path_dir(filename, &path);
+        if (rc)
+               goto out;
+
+        rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
+        if (rc)
+               goto dput_and_out;
+
+        vn_set_fs_pwd(current->fs, &path);
+
+dput_and_out:
+        path_put(&path);
+out:
+       set_fs(saved_fs);
+
+        return (-rc);
+} /* vn_set_pwd() */
+EXPORT_SYMBOL(vn_set_pwd);
+
+static int
+vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
+{
+       struct vnode *vp = buf;
+
+       mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
+
+       return (0);
+} /* vn_cache_constructor() */
+
+static void
+vn_cache_destructor(void *buf, void *cdrarg)
+{
+       struct vnode *vp = buf;
+
+       mutex_destroy(&vp->v_lock);
+} /* vn_cache_destructor() */
+
+static int
+vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
+{
+       file_t *fp = buf;
+
+       atomic_set(&fp->f_ref, 0);
+        mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
+       INIT_LIST_HEAD(&fp->f_list);
+
+        return (0);
+} /* file_cache_constructor() */
+
+static void
+vn_file_cache_destructor(void *buf, void *cdrarg)
+{
+       file_t *fp = buf;
+
+       mutex_destroy(&fp->f_lock);
+} /* vn_file_cache_destructor() */
+
+int
+spl_vn_init(void)
+{
+       vn_cache = kmem_cache_create("spl_vn_cache",
+                                    sizeof(struct vnode), 64,
+                                    vn_cache_constructor,
+                                    vn_cache_destructor,
+                                    NULL, NULL, NULL, 0);
+
+       vn_file_cache = kmem_cache_create("spl_vn_file_cache",
+                                         sizeof(file_t), 64,
+                                         vn_file_cache_constructor,
+                                         vn_file_cache_destructor,
+                                         NULL, NULL, NULL, 0);
+       return (0);
+} /* vn_init() */
+
+void
+spl_vn_fini(void)
+{
+        file_t *fp, *next_fp;
+       int leaked = 0;
+
+       spin_lock(&vn_file_lock);
+
+        list_for_each_entry_safe(fp, next_fp, &vn_file_list,  f_list) {
+               list_del(&fp->f_list);
+               releasef_locked(fp);
+               leaked++;
+       }
+
+       spin_unlock(&vn_file_lock);
+
+       if (leaked > 0)
+               printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
+
+       kmem_cache_destroy(vn_file_cache);
+       kmem_cache_destroy(vn_cache);
+
+       return;
+} /* vn_fini() */