xfs: fix dentry aliasing issues in open_by_handle
authorChristoph Hellwig <hch@lst.de>
Mon, 19 Jan 2009 01:02:57 +0000 (02:02 +0100)
committerChristoph Hellwig <hch@brick.lst.de>
Mon, 19 Jan 2009 01:02:57 +0000 (02:02 +0100)
Open by handle just grabs an inode by handle and then creates itself
a dentry for it.  While this works for regular files it is horribly
broken for directories, where the VFS locking relies on the fact that
there is only just one single dentry for a given inode, and that
these are always connected to the root of the filesystem so that
it's locking algorithms work (see Documentations/filesystems/Locking)

Remove all the existing open by handle code and replace it with a small
wrapper around the exportfs code which deals with all these issues.
At the same time we also make the checks for a valid handle strict
enough to reject all not perfectly well formed handles - given that
we never hand out others that's okay and simplifies the code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <david@fromorbit.com>
fs/xfs/Kconfig
fs/xfs/linux-2.6/xfs_ioctl.c
fs/xfs/linux-2.6/xfs_ioctl.h
fs/xfs/linux-2.6/xfs_ioctl32.c

index 3f53dd1..29228f5 100644 (file)
@@ -1,6 +1,7 @@
 config XFS_FS
        tristate "XFS filesystem support"
        depends on BLOCK
+       select EXPORTFS
        help
          XFS is a high performance journaling filesystem which originated
          on the SGI IRIX platform.  It is completely multi-threaded, can
index e5be1e0..4bd1123 100644 (file)
 #include "xfs_vnodeops.h"
 #include "xfs_quota.h"
 #include "xfs_inode_item.h"
+#include "xfs_export.h"
 
 #include <linux/capability.h>
 #include <linux/dcache.h>
 #include <linux/mount.h>
 #include <linux/namei.h>
 #include <linux/pagemap.h>
+#include <linux/exportfs.h>
 
 /*
  * xfs_find_handle maps from userspace xfs_fsop_handlereq structure to
@@ -164,97 +166,69 @@ xfs_find_handle(
        return 0;
 }
 
-
 /*
- * Convert userspace handle data into inode.
- *
- * We use the fact that all the fsop_handlereq ioctl calls have a data
- * structure argument whose first component is always a xfs_fsop_handlereq_t,
- * so we can pass that sub structure into this handy, shared routine.
- *
- * If no error, caller must always iput the returned inode.
+ * No need to do permission checks on the various pathname components
+ * as the handle operations are privileged.
  */
 STATIC int
-xfs_vget_fsop_handlereq(
-       xfs_mount_t             *mp,
-       struct inode            *parinode,      /* parent inode pointer    */
-       xfs_fsop_handlereq_t    *hreq,
-       struct inode            **inode)
+xfs_handle_acceptable(
+       void                    *context,
+       struct dentry           *dentry)
+{
+       return 1;
+}
+
+/*
+ * Convert userspace handle data into a dentry.
+ */
+struct dentry *
+xfs_handle_to_dentry(
+       struct file             *parfilp,
+       void __user             *uhandle,
+       u32                     hlen)
 {
-       void                    __user *hanp;
-       size_t                  hlen;
-       xfs_fid_t               *xfid;
-       xfs_handle_t            *handlep;
        xfs_handle_t            handle;
-       xfs_inode_t             *ip;
-       xfs_ino_t               ino;
-       __u32                   igen;
-       int                     error;
+       struct xfs_fid64        fid;
 
        /*
         * Only allow handle opens under a directory.
         */
-       if (!S_ISDIR(parinode->i_mode))
-               return XFS_ERROR(ENOTDIR);
-
-       hanp = hreq->ihandle;
-       hlen = hreq->ihandlen;
-       handlep = &handle;
-
-       if (hlen < sizeof(handlep->ha_fsid) || hlen > sizeof(*handlep))
-               return XFS_ERROR(EINVAL);
-       if (copy_from_user(handlep, hanp, hlen))
-               return XFS_ERROR(EFAULT);
-       if (hlen < sizeof(*handlep))
-               memset(((char *)handlep) + hlen, 0, sizeof(*handlep) - hlen);
-       if (hlen > sizeof(handlep->ha_fsid)) {
-               if (handlep->ha_fid.fid_len !=
-                   (hlen - sizeof(handlep->ha_fsid) -
-                           sizeof(handlep->ha_fid.fid_len)) ||
-                   handlep->ha_fid.fid_pad)
-                       return XFS_ERROR(EINVAL);
-       }
-
-       /*
-        * Crack the handle, obtain the inode # & generation #
-        */
-       xfid = (struct xfs_fid *)&handlep->ha_fid;
-       if (xfid->fid_len == sizeof(*xfid) - sizeof(xfid->fid_len)) {
-               ino  = xfid->fid_ino;
-               igen = xfid->fid_gen;
-       } else {
-               return XFS_ERROR(EINVAL);
-       }
-
-       /*
-        * Get the XFS inode, building a Linux inode to go with it.
-        */
-       error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
-       if (error)
-               return error;
-       if (ip == NULL)
-               return XFS_ERROR(EIO);
-       if (ip->i_d.di_gen != igen) {
-               xfs_iput_new(ip, XFS_ILOCK_SHARED);
-               return XFS_ERROR(ENOENT);
-       }
-
-       xfs_iunlock(ip, XFS_ILOCK_SHARED);
+       if (!S_ISDIR(parfilp->f_path.dentry->d_inode->i_mode))
+               return ERR_PTR(-ENOTDIR);
+
+       if (hlen != sizeof(xfs_handle_t))
+               return ERR_PTR(-EINVAL);
+       if (copy_from_user(&handle, uhandle, hlen))
+               return ERR_PTR(-EFAULT);
+       if (handle.ha_fid.fid_len !=
+           sizeof(handle.ha_fid) - sizeof(handle.ha_fid.fid_len))
+               return ERR_PTR(-EINVAL);
+
+       memset(&fid, 0, sizeof(struct fid));
+       fid.ino = handle.ha_fid.fid_ino;
+       fid.gen = handle.ha_fid.fid_gen;
+
+       return exportfs_decode_fh(parfilp->f_path.mnt, (struct fid *)&fid, 3,
+                       FILEID_INO32_GEN | XFS_FILEID_TYPE_64FLAG,
+                       xfs_handle_acceptable, NULL);
+}
 
-       *inode = VFS_I(ip);
-       return 0;
+STATIC struct dentry *
+xfs_handlereq_to_dentry(
+       struct file             *parfilp,
+       xfs_fsop_handlereq_t    *hreq)
+{
+       return xfs_handle_to_dentry(parfilp, hreq->ihandle, hreq->ihandlen);
 }
 
 int
 xfs_open_by_handle(
-       xfs_mount_t             *mp,
-       xfs_fsop_handlereq_t    *hreq,
        struct file             *parfilp,
-       struct inode            *parinode)
+       xfs_fsop_handlereq_t    *hreq)
 {
        const struct cred       *cred = current_cred();
        int                     error;
-       int                     new_fd;
+       int                     fd;
        int                     permflag;
        struct file             *filp;
        struct inode            *inode;
@@ -263,19 +237,21 @@ xfs_open_by_handle(
        if (!capable(CAP_SYS_ADMIN))
                return -XFS_ERROR(EPERM);
 
-       error = xfs_vget_fsop_handlereq(mp, parinode, hreq, &inode);
-       if (error)
-               return -error;
+       dentry = xfs_handlereq_to_dentry(parfilp, hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
+       inode = dentry->d_inode;
 
        /* Restrict xfs_open_by_handle to directories & regular files. */
        if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) {
-               iput(inode);
-               return -XFS_ERROR(EINVAL);
+               error = -XFS_ERROR(EPERM);
+               goto out_dput;
        }
 
 #if BITS_PER_LONG != 32
        hreq->oflags |= O_LARGEFILE;
 #endif
+
        /* Put open permission in namei format. */
        permflag = hreq->oflags;
        if ((permflag+1) & O_ACCMODE)
@@ -285,50 +261,45 @@ xfs_open_by_handle(
 
        if ((!(permflag & O_APPEND) || (permflag & O_TRUNC)) &&
            (permflag & FMODE_WRITE) && IS_APPEND(inode)) {
-               iput(inode);
-               return -XFS_ERROR(EPERM);
+               error = -XFS_ERROR(EPERM);
+               goto out_dput;
        }
 
        if ((permflag & FMODE_WRITE) && IS_IMMUTABLE(inode)) {
-               iput(inode);
-               return -XFS_ERROR(EACCES);
+               error = -XFS_ERROR(EACCES);
+               goto out_dput;
        }
 
        /* Can't write directories. */
-       if ( S_ISDIR(inode->i_mode) && (permflag & FMODE_WRITE)) {
-               iput(inode);
-               return -XFS_ERROR(EISDIR);
+       if (S_ISDIR(inode->i_mode) && (permflag & FMODE_WRITE)) {
+               error = -XFS_ERROR(EISDIR);
+               goto out_dput;
        }
 
-       if ((new_fd = get_unused_fd()) < 0) {
-               iput(inode);
-               return new_fd;
+       fd = get_unused_fd();
+       if (fd < 0) {
+               error = fd;
+               goto out_dput;
        }
 
-       dentry = d_obtain_alias(inode);
-       if (IS_ERR(dentry)) {
-               put_unused_fd(new_fd);
-               return PTR_ERR(dentry);
-       }
-
-       /* Ensure umount returns EBUSY on umounts while this file is open. */
-       mntget(parfilp->f_path.mnt);
-
-       /* Create file pointer. */
-       filp = dentry_open(dentry, parfilp->f_path.mnt, hreq->oflags, cred);
+       filp = dentry_open(dentry, mntget(parfilp->f_path.mnt),
+                          hreq->oflags, cred);
        if (IS_ERR(filp)) {
-               put_unused_fd(new_fd);
-               return -XFS_ERROR(-PTR_ERR(filp));
+               put_unused_fd(fd);
+               return PTR_ERR(filp);
        }
 
        if (inode->i_mode & S_IFREG) {
-               /* invisible operation should not change atime */
                filp->f_flags |= O_NOATIME;
                filp->f_mode |= FMODE_NOCMTIME;
        }
 
-       fd_install(new_fd, filp);
-       return new_fd;
+       fd_install(fd, filp);
+       return fd;
+
+ out_dput:
+       dput(dentry);
+       return error;
 }
 
 /*
@@ -359,11 +330,10 @@ do_readlink(
 
 int
 xfs_readlink_by_handle(
-       xfs_mount_t             *mp,
-       xfs_fsop_handlereq_t    *hreq,
-       struct inode            *parinode)
+       struct file             *parfilp,
+       xfs_fsop_handlereq_t    *hreq)
 {
-       struct inode            *inode;
+       struct dentry           *dentry;
        __u32                   olen;
        void                    *link;
        int                     error;
@@ -371,26 +341,28 @@ xfs_readlink_by_handle(
        if (!capable(CAP_SYS_ADMIN))
                return -XFS_ERROR(EPERM);
 
-       error = xfs_vget_fsop_handlereq(mp, parinode, hreq, &inode);
-       if (error)
-               return -error;
+       dentry = xfs_handlereq_to_dentry(parfilp, hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
        /* Restrict this handle operation to symlinks only. */
-       if (!S_ISLNK(inode->i_mode)) {
+       if (!S_ISLNK(dentry->d_inode->i_mode)) {
                error = -XFS_ERROR(EINVAL);
-               goto out_iput;
+               goto out_dput;
        }
 
        if (copy_from_user(&olen, hreq->ohandlen, sizeof(__u32))) {
                error = -XFS_ERROR(EFAULT);
-               goto out_iput;
+               goto out_dput;
        }
 
        link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
-       if (!link)
-               goto out_iput;
+       if (!link) {
+               error = -XFS_ERROR(ENOMEM);
+               goto out_dput;
+       }
 
-       error = -xfs_readlink(XFS_I(inode), link);
+       error = -xfs_readlink(XFS_I(dentry->d_inode), link);
        if (error)
                goto out_kfree;
        error = do_readlink(hreq->ohandle, olen, link);
@@ -399,32 +371,31 @@ xfs_readlink_by_handle(
 
  out_kfree:
        kfree(link);
- out_iput:
-       iput(inode);
+ out_dput:
+       dput(dentry);
        return error;
 }
 
 STATIC int
 xfs_fssetdm_by_handle(
-       xfs_mount_t             *mp,
-       void                    __user *arg,
-       struct inode            *parinode)
+       struct file             *parfilp,
+       void                    __user *arg)
 {
        int                     error;
        struct fsdmidata        fsd;
        xfs_fsop_setdm_handlereq_t dmhreq;
-       struct inode            *inode;
+       struct dentry           *dentry;
 
        if (!capable(CAP_MKNOD))
                return -XFS_ERROR(EPERM);
        if (copy_from_user(&dmhreq, arg, sizeof(xfs_fsop_setdm_handlereq_t)))
                return -XFS_ERROR(EFAULT);
 
-       error = xfs_vget_fsop_handlereq(mp, parinode, &dmhreq.hreq, &inode);
-       if (error)
-               return -error;
+       dentry = xfs_handlereq_to_dentry(parfilp, &dmhreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
-       if (IS_IMMUTABLE(inode) || IS_APPEND(inode)) {
+       if (IS_IMMUTABLE(dentry->d_inode) || IS_APPEND(dentry->d_inode)) {
                error = -XFS_ERROR(EPERM);
                goto out;
        }
@@ -434,24 +405,23 @@ xfs_fssetdm_by_handle(
                goto out;
        }
 
-       error = -xfs_set_dmattrs(XFS_I(inode), fsd.fsd_dmevmask,
+       error = -xfs_set_dmattrs(XFS_I(dentry->d_inode), fsd.fsd_dmevmask,
                                 fsd.fsd_dmstate);
 
  out:
-       iput(inode);
+       dput(dentry);
        return error;
 }
 
 STATIC int
 xfs_attrlist_by_handle(
-       xfs_mount_t             *mp,
-       void                    __user *arg,
-       struct inode            *parinode)
+       struct file             *parfilp,
+       void                    __user *arg)
 {
-       int                     error;
+       int                     error = -ENOMEM;
        attrlist_cursor_kern_t  *cursor;
        xfs_fsop_attrlist_handlereq_t al_hreq;
-       struct inode            *inode;
+       struct dentry           *dentry;
        char                    *kbuf;
 
        if (!capable(CAP_SYS_ADMIN))
@@ -467,16 +437,16 @@ xfs_attrlist_by_handle(
        if (al_hreq.flags & ~(ATTR_ROOT | ATTR_SECURE))
                return -XFS_ERROR(EINVAL);
 
-       error = xfs_vget_fsop_handlereq(mp, parinode, &al_hreq.hreq, &inode);
-       if (error)
-               goto out;
+       dentry = xfs_handlereq_to_dentry(parfilp, &al_hreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
        kbuf = kmalloc(al_hreq.buflen, GFP_KERNEL);
        if (!kbuf)
-               goto out_vn_rele;
+               goto out_dput;
 
        cursor = (attrlist_cursor_kern_t *)&al_hreq.pos;
-       error = xfs_attr_list(XFS_I(inode), kbuf, al_hreq.buflen,
+       error = -xfs_attr_list(XFS_I(dentry->d_inode), kbuf, al_hreq.buflen,
                                        al_hreq.flags, cursor);
        if (error)
                goto out_kfree;
@@ -486,10 +456,9 @@ xfs_attrlist_by_handle(
 
  out_kfree:
        kfree(kbuf);
- out_vn_rele:
-       iput(inode);
- out:
-       return -error;
+ out_dput:
+       dput(dentry);
+       return error;
 }
 
 int
@@ -564,15 +533,13 @@ xfs_attrmulti_attr_remove(
 
 STATIC int
 xfs_attrmulti_by_handle(
-       xfs_mount_t             *mp,
-       void                    __user *arg,
        struct file             *parfilp,
-       struct inode            *parinode)
+       void                    __user *arg)
 {
        int                     error;
        xfs_attr_multiop_t      *ops;
        xfs_fsop_attrmulti_handlereq_t am_hreq;
-       struct inode            *inode;
+       struct dentry           *dentry;
        unsigned int            i, size;
        char                    *attr_name;
 
@@ -581,19 +548,19 @@ xfs_attrmulti_by_handle(
        if (copy_from_user(&am_hreq, arg, sizeof(xfs_fsop_attrmulti_handlereq_t)))
                return -XFS_ERROR(EFAULT);
 
-       error = xfs_vget_fsop_handlereq(mp, parinode, &am_hreq.hreq, &inode);
-       if (error)
-               goto out;
+       dentry = xfs_handlereq_to_dentry(parfilp, &am_hreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
        error = E2BIG;
        size = am_hreq.opcount * sizeof(xfs_attr_multiop_t);
        if (!size || size > 16 * PAGE_SIZE)
-               goto out_vn_rele;
+               goto out_dput;
 
        error = ENOMEM;
        ops = kmalloc(size, GFP_KERNEL);
        if (!ops)
-               goto out_vn_rele;
+               goto out_dput;
 
        error = EFAULT;
        if (copy_from_user(ops, am_hreq.ops, size))
@@ -615,25 +582,28 @@ xfs_attrmulti_by_handle(
 
                switch (ops[i].am_opcode) {
                case ATTR_OP_GET:
-                       ops[i].am_error = xfs_attrmulti_attr_get(inode,
-                                       attr_name, ops[i].am_attrvalue,
-                                       &ops[i].am_length, ops[i].am_flags);
+                       ops[i].am_error = xfs_attrmulti_attr_get(
+                                       dentry->d_inode, attr_name,
+                                       ops[i].am_attrvalue, &ops[i].am_length,
+                                       ops[i].am_flags);
                        break;
                case ATTR_OP_SET:
                        ops[i].am_error = mnt_want_write(parfilp->f_path.mnt);
                        if (ops[i].am_error)
                                break;
-                       ops[i].am_error = xfs_attrmulti_attr_set(inode,
-                                       attr_name, ops[i].am_attrvalue,
-                                       ops[i].am_length, ops[i].am_flags);
+                       ops[i].am_error = xfs_attrmulti_attr_set(
+                                       dentry->d_inode, attr_name,
+                                       ops[i].am_attrvalue, ops[i].am_length,
+                                       ops[i].am_flags);
                        mnt_drop_write(parfilp->f_path.mnt);
                        break;
                case ATTR_OP_REMOVE:
                        ops[i].am_error = mnt_want_write(parfilp->f_path.mnt);
                        if (ops[i].am_error)
                                break;
-                       ops[i].am_error = xfs_attrmulti_attr_remove(inode,
-                                       attr_name, ops[i].am_flags);
+                       ops[i].am_error = xfs_attrmulti_attr_remove(
+                                       dentry->d_inode, attr_name,
+                                       ops[i].am_flags);
                        mnt_drop_write(parfilp->f_path.mnt);
                        break;
                default:
@@ -647,9 +617,8 @@ xfs_attrmulti_by_handle(
        kfree(attr_name);
  out_kfree_ops:
        kfree(ops);
- out_vn_rele:
-       iput(inode);
- out:
+ out_dput:
+       dput(dentry);
        return -error;
 }
 
@@ -1440,23 +1409,23 @@ xfs_file_ioctl(
 
                if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
                        return -XFS_ERROR(EFAULT);
-               return xfs_open_by_handle(mp, &hreq, filp, inode);
+               return xfs_open_by_handle(filp, &hreq);
        }
        case XFS_IOC_FSSETDM_BY_HANDLE:
-               return xfs_fssetdm_by_handle(mp, arg, inode);
+               return xfs_fssetdm_by_handle(filp, arg);
 
        case XFS_IOC_READLINK_BY_HANDLE: {
                xfs_fsop_handlereq_t    hreq;
 
                if (copy_from_user(&hreq, arg, sizeof(xfs_fsop_handlereq_t)))
                        return -XFS_ERROR(EFAULT);
-               return xfs_readlink_by_handle(mp, &hreq, inode);
+               return xfs_readlink_by_handle(filp, &hreq);
        }
        case XFS_IOC_ATTRLIST_BY_HANDLE:
-               return xfs_attrlist_by_handle(mp, arg, inode);
+               return xfs_attrlist_by_handle(filp, arg);
 
        case XFS_IOC_ATTRMULTI_BY_HANDLE:
-               return xfs_attrmulti_by_handle(mp, arg, filp, inode);
+               return xfs_attrmulti_by_handle(filp, arg);
 
        case XFS_IOC_SWAPEXT: {
                struct xfs_swapext      sxp;
index 8c16bf2..7bd7c6a 100644 (file)
@@ -34,16 +34,13 @@ xfs_find_handle(
 
 extern int
 xfs_open_by_handle(
-       xfs_mount_t             *mp,
-       xfs_fsop_handlereq_t    *hreq,
        struct file             *parfilp,
-       struct inode            *parinode);
+       xfs_fsop_handlereq_t    *hreq);
 
 extern int
 xfs_readlink_by_handle(
-       xfs_mount_t             *mp,
-       xfs_fsop_handlereq_t    *hreq,
-       struct inode            *parinode);
+       struct file             *parfilp,
+       xfs_fsop_handlereq_t    *hreq);
 
 extern int
 xfs_attrmulti_attr_get(
@@ -67,6 +64,12 @@ xfs_attrmulti_attr_remove(
        char                    *name,
        __uint32_t              flags);
 
+extern struct dentry *
+xfs_handle_to_dentry(
+       struct file             *parfilp,
+       void __user             *uhandle,
+       u32                     hlen);
+
 extern long
 xfs_file_ioctl(
        struct file             *filp,
index 50903ad..fd43620 100644 (file)
@@ -340,96 +340,24 @@ xfs_compat_handlereq_copyin(
        return 0;
 }
 
-/*
- * Convert userspace handle data into inode.
- *
- * We use the fact that all the fsop_handlereq ioctl calls have a data
- * structure argument whose first component is always a xfs_fsop_handlereq_t,
- * so we can pass that sub structure into this handy, shared routine.
- *
- * If no error, caller must always iput the returned inode.
- */
-STATIC int
-xfs_vget_fsop_handlereq_compat(
-       xfs_mount_t             *mp,
-       struct inode            *parinode,      /* parent inode pointer    */
-       compat_xfs_fsop_handlereq_t     *hreq,
-       struct inode            **inode)
+STATIC struct dentry *
+xfs_compat_handlereq_to_dentry(
+       struct file             *parfilp,
+       compat_xfs_fsop_handlereq_t *hreq)
 {
-       void                    __user *hanp;
-       size_t                  hlen;
-       xfs_fid_t               *xfid;
-       xfs_handle_t            *handlep;
-       xfs_handle_t            handle;
-       xfs_inode_t             *ip;
-       xfs_ino_t               ino;
-       __u32                   igen;
-       int                     error;
-
-       /*
-        * Only allow handle opens under a directory.
-        */
-       if (!S_ISDIR(parinode->i_mode))
-               return XFS_ERROR(ENOTDIR);
-
-       hanp = compat_ptr(hreq->ihandle);
-       hlen = hreq->ihandlen;
-       handlep = &handle;
-
-       if (hlen < sizeof(handlep->ha_fsid) || hlen > sizeof(*handlep))
-               return XFS_ERROR(EINVAL);
-       if (copy_from_user(handlep, hanp, hlen))
-               return XFS_ERROR(EFAULT);
-       if (hlen < sizeof(*handlep))
-               memset(((char *)handlep) + hlen, 0, sizeof(*handlep) - hlen);
-       if (hlen > sizeof(handlep->ha_fsid)) {
-               if (handlep->ha_fid.fid_len !=
-                   (hlen - sizeof(handlep->ha_fsid) -
-                           sizeof(handlep->ha_fid.fid_len)) ||
-                   handlep->ha_fid.fid_pad)
-                       return XFS_ERROR(EINVAL);
-       }
-
-       /*
-        * Crack the handle, obtain the inode # & generation #
-        */
-       xfid = (struct xfs_fid *)&handlep->ha_fid;
-       if (xfid->fid_len == sizeof(*xfid) - sizeof(xfid->fid_len)) {
-               ino  = xfid->fid_ino;
-               igen = xfid->fid_gen;
-       } else {
-               return XFS_ERROR(EINVAL);
-       }
-
-       /*
-        * Get the XFS inode, building a Linux inode to go with it.
-        */
-       error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
-       if (error)
-               return error;
-       if (ip == NULL)
-               return XFS_ERROR(EIO);
-       if (ip->i_d.di_gen != igen) {
-               xfs_iput_new(ip, XFS_ILOCK_SHARED);
-               return XFS_ERROR(ENOENT);
-       }
-
-       xfs_iunlock(ip, XFS_ILOCK_SHARED);
-
-       *inode = VFS_I(ip);
-       return 0;
+       return xfs_handle_to_dentry(parfilp,
+                       compat_ptr(hreq->ihandle), hreq->ihandlen);
 }
 
 STATIC int
 xfs_compat_attrlist_by_handle(
-       xfs_mount_t             *mp,
-       void                    __user *arg,
-       struct inode            *parinode)
+       struct file             *parfilp,
+       void                    __user *arg)
 {
        int                     error;
        attrlist_cursor_kern_t  *cursor;
        compat_xfs_fsop_attrlist_handlereq_t al_hreq;
-       struct inode            *inode;
+       struct dentry           *dentry;
        char                    *kbuf;
 
        if (!capable(CAP_SYS_ADMIN))
@@ -446,17 +374,17 @@ xfs_compat_attrlist_by_handle(
        if (al_hreq.flags & ~(ATTR_ROOT | ATTR_SECURE))
                return -XFS_ERROR(EINVAL);
 
-       error = xfs_vget_fsop_handlereq_compat(mp, parinode, &al_hreq.hreq,
-                                              &inode);
-       if (error)
-               goto out;
+       dentry = xfs_compat_handlereq_to_dentry(parfilp, &al_hreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
+       error = -ENOMEM;
        kbuf = kmalloc(al_hreq.buflen, GFP_KERNEL);
        if (!kbuf)
-               goto out_vn_rele;
+               goto out_dput;
 
        cursor = (attrlist_cursor_kern_t *)&al_hreq.pos;
-       error = xfs_attr_list(XFS_I(inode), kbuf, al_hreq.buflen,
+       error = -xfs_attr_list(XFS_I(dentry->d_inode), kbuf, al_hreq.buflen,
                                        al_hreq.flags, cursor);
        if (error)
                goto out_kfree;
@@ -466,22 +394,20 @@ xfs_compat_attrlist_by_handle(
 
  out_kfree:
        kfree(kbuf);
- out_vn_rele:
-       iput(inode);
- out:
-       return -error;
+ out_dput:
+       dput(dentry);
+       return error;
 }
 
 STATIC int
 xfs_compat_attrmulti_by_handle(
-       xfs_mount_t                             *mp,
-       void                                    __user *arg,
-       struct inode                            *parinode)
+       struct file                             *parfilp,
+       void                                    __user *arg)
 {
        int                                     error;
        compat_xfs_attr_multiop_t               *ops;
        compat_xfs_fsop_attrmulti_handlereq_t   am_hreq;
-       struct inode                            *inode;
+       struct dentry                           *dentry;
        unsigned int                            i, size;
        char                                    *attr_name;
 
@@ -491,20 +417,19 @@ xfs_compat_attrmulti_by_handle(
                           sizeof(compat_xfs_fsop_attrmulti_handlereq_t)))
                return -XFS_ERROR(EFAULT);
 
-       error = xfs_vget_fsop_handlereq_compat(mp, parinode, &am_hreq.hreq,
-                                              &inode);
-       if (error)
-               goto out;
+       dentry = xfs_compat_handlereq_to_dentry(parfilp, &am_hreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
        error = E2BIG;
        size = am_hreq.opcount * sizeof(compat_xfs_attr_multiop_t);
        if (!size || size > 16 * PAGE_SIZE)
-               goto out_vn_rele;
+               goto out_dput;
 
        error = ENOMEM;
        ops = kmalloc(size, GFP_KERNEL);
        if (!ops)
-               goto out_vn_rele;
+               goto out_dput;
 
        error = EFAULT;
        if (copy_from_user(ops, compat_ptr(am_hreq.ops), size))
@@ -527,20 +452,21 @@ xfs_compat_attrmulti_by_handle(
 
                switch (ops[i].am_opcode) {
                case ATTR_OP_GET:
-                       ops[i].am_error = xfs_attrmulti_attr_get(inode,
-                                       attr_name,
+                       ops[i].am_error = xfs_attrmulti_attr_get(
+                                       dentry->d_inode, attr_name,
                                        compat_ptr(ops[i].am_attrvalue),
                                        &ops[i].am_length, ops[i].am_flags);
                        break;
                case ATTR_OP_SET:
-                       ops[i].am_error = xfs_attrmulti_attr_set(inode,
-                                       attr_name,
+                       ops[i].am_error = xfs_attrmulti_attr_set(
+                                       dentry->d_inode, attr_name,
                                        compat_ptr(ops[i].am_attrvalue),
                                        ops[i].am_length, ops[i].am_flags);
                        break;
                case ATTR_OP_REMOVE:
-                       ops[i].am_error = xfs_attrmulti_attr_remove(inode,
-                                       attr_name, ops[i].am_flags);
+                       ops[i].am_error = xfs_attrmulti_attr_remove(
+                                       dentry->d_inode, attr_name,
+                                       ops[i].am_flags);
                        break;
                default:
                        ops[i].am_error = EINVAL;
@@ -553,22 +479,20 @@ xfs_compat_attrmulti_by_handle(
        kfree(attr_name);
  out_kfree_ops:
        kfree(ops);
- out_vn_rele:
-       iput(inode);
- out:
+ out_dput:
+       dput(dentry);
        return -error;
 }
 
 STATIC int
 xfs_compat_fssetdm_by_handle(
-       xfs_mount_t             *mp,
-       void                    __user *arg,
-       struct inode            *parinode)
+       struct file             *parfilp,
+       void                    __user *arg)
 {
        int                     error;
        struct fsdmidata        fsd;
        compat_xfs_fsop_setdm_handlereq_t dmhreq;
-       struct inode            *inode;
+       struct dentry           *dentry;
 
        if (!capable(CAP_MKNOD))
                return -XFS_ERROR(EPERM);
@@ -576,12 +500,11 @@ xfs_compat_fssetdm_by_handle(
                           sizeof(compat_xfs_fsop_setdm_handlereq_t)))
                return -XFS_ERROR(EFAULT);
 
-       error = xfs_vget_fsop_handlereq_compat(mp, parinode, &dmhreq.hreq,
-                                              &inode);
-       if (error)
-               return -error;
+       dentry = xfs_compat_handlereq_to_dentry(parfilp, &dmhreq.hreq);
+       if (IS_ERR(dentry))
+               return PTR_ERR(dentry);
 
-       if (IS_IMMUTABLE(inode) || IS_APPEND(inode)) {
+       if (IS_IMMUTABLE(dentry->d_inode) || IS_APPEND(dentry->d_inode)) {
                error = -XFS_ERROR(EPERM);
                goto out;
        }
@@ -591,11 +514,11 @@ xfs_compat_fssetdm_by_handle(
                goto out;
        }
 
-       error = -xfs_set_dmattrs(XFS_I(inode), fsd.fsd_dmevmask,
+       error = -xfs_set_dmattrs(XFS_I(dentry->d_inode), fsd.fsd_dmevmask,
                                 fsd.fsd_dmstate);
 
 out:
-       iput(inode);
+       dput(dentry);
        return error;
 }
 
@@ -722,21 +645,21 @@ xfs_file_compat_ioctl(
 
                if (xfs_compat_handlereq_copyin(&hreq, arg))
                        return -XFS_ERROR(EFAULT);
-               return xfs_open_by_handle(mp, &hreq, filp, inode);
+               return xfs_open_by_handle(filp, &hreq);
        }
        case XFS_IOC_READLINK_BY_HANDLE_32: {
                struct xfs_fsop_handlereq       hreq;
 
                if (xfs_compat_handlereq_copyin(&hreq, arg))
                        return -XFS_ERROR(EFAULT);
-               return xfs_readlink_by_handle(mp, &hreq, inode);
+               return xfs_readlink_by_handle(filp, &hreq);
        }
        case XFS_IOC_ATTRLIST_BY_HANDLE_32:
-               return xfs_compat_attrlist_by_handle(mp, arg, inode);
+               return xfs_compat_attrlist_by_handle(filp, arg);
        case XFS_IOC_ATTRMULTI_BY_HANDLE_32:
-               return xfs_compat_attrmulti_by_handle(mp, arg, inode);
+               return xfs_compat_attrmulti_by_handle(filp, arg);
        case XFS_IOC_FSSETDM_BY_HANDLE_32:
-               return xfs_compat_fssetdm_by_handle(mp, arg, inode);
+               return xfs_compat_fssetdm_by_handle(filp, arg);
        default:
                return -XFS_ERROR(ENOIOCTLCMD);
        }