include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[safe/jmp/linux-2.6] / fs / xfs / linux-2.6 / xfs_iops.c
index 4da034f..e65a793 100644 (file)
@@ -17,6 +17,7 @@
  */
 #include "xfs.h"
 #include "xfs_fs.h"
+#include "xfs_acl.h"
 #include "xfs_bit.h"
 #include "xfs_log.h"
 #include "xfs_inum.h"
 #include "xfs_error.h"
 #include "xfs_itable.h"
 #include "xfs_rw.h"
-#include "xfs_acl.h"
 #include "xfs_attr.h"
 #include "xfs_buf_item.h"
 #include "xfs_utils.h"
 #include "xfs_vnodeops.h"
+#include "xfs_trace.h"
 
 #include <linux/capability.h>
 #include <linux/xattr.h>
 #include <linux/namei.h>
+#include <linux/posix_acl.h>
 #include <linux/security.h>
+#include <linux/falloc.h>
+#include <linux/fiemap.h>
+#include <linux/slab.h>
 
 /*
- * Bring the atime in the XFS inode uptodate.
- * Used before logging the inode to disk or when the Linux inode goes away.
+ * Bring the timestamps in the XFS inode uptodate.
+ *
+ * Used before writing the inode to disk.
  */
 void
-xfs_synchronize_atime(
+xfs_synchronize_times(
        xfs_inode_t     *ip)
 {
-       bhv_vnode_t     *vp;
-
-       vp = XFS_ITOV_NULL(ip);
-       if (vp) {
-               struct inode *inode = &vp->v_inode;
-               ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
-               ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
-       }
+       struct inode    *inode = VFS_I(ip);
+
+       ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
+       ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
+       ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec;
+       ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec;
+       ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec;
+       ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec;
 }
 
 /*
- * Change the requested timestamp in the given inode.
- * We don't lock across timestamp updates, and we don't log them but
- * we do record the fact that there is dirty information in core.
- *
- * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
- *             with XFS_ICHGTIME_ACC to be sure that access time
- *             update will take.  Calling first with XFS_ICHGTIME_ACC
- *             and then XFS_ICHGTIME_MOD may fail to modify the access
- *             timestamp if the filesystem is mounted noacctm.
+ * If the linux inode is valid, mark it dirty.
+ * Used when commiting a dirty inode into a transaction so that
+ * the inode will get written back by the linux code
  */
 void
-xfs_ichgtime(
-       xfs_inode_t     *ip,
-       int             flags)
+xfs_mark_inode_dirty_sync(
+       xfs_inode_t     *ip)
 {
-       struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
-       timespec_t      tv;
+       struct inode    *inode = VFS_I(ip);
 
-       nanotime(&tv);
-       if (flags & XFS_ICHGTIME_MOD) {
-               inode->i_mtime = tv;
-               ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
-               ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
-       }
-       if (flags & XFS_ICHGTIME_ACC) {
-               inode->i_atime = tv;
-               ip->i_d.di_atime.t_sec = (__int32_t)tv.tv_sec;
-               ip->i_d.di_atime.t_nsec = (__int32_t)tv.tv_nsec;
-       }
-       if (flags & XFS_ICHGTIME_CHG) {
-               inode->i_ctime = tv;
-               ip->i_d.di_ctime.t_sec = (__int32_t)tv.tv_sec;
-               ip->i_d.di_ctime.t_nsec = (__int32_t)tv.tv_nsec;
-       }
-
-       /*
-        * We update the i_update_core field _after_ changing
-        * the timestamps in order to coordinate properly with
-        * xfs_iflush() so that we don't lose timestamp updates.
-        * This keeps us from having to hold the inode lock
-        * while doing this.  We use the SYNCHRONIZE macro to
-        * ensure that the compiler does not reorder the update
-        * of i_update_core above the timestamp updates above.
-        */
-       SYNCHRONIZE();
-       ip->i_update_core = 1;
-       if (!(inode->i_state & I_LOCK))
+       if (!(inode->i_state & (I_WILL_FREE|I_FREEING|I_CLEAR)))
                mark_inode_dirty_sync(inode);
 }
 
+void
+xfs_mark_inode_dirty(
+       xfs_inode_t     *ip)
+{
+       struct inode    *inode = VFS_I(ip);
+
+       if (!(inode->i_state & (I_WILL_FREE|I_FREEING|I_CLEAR)))
+               mark_inode_dirty(inode);
+}
+
 /*
- * Variant on the above which avoids querying the system clock
- * in situations where we know the Linux inode timestamps have
- * just been updated (and so we can update our inode cheaply).
+ * Change the requested timestamp in the given inode.
+ * We don't lock across timestamp updates, and we don't log them but
+ * we do record the fact that there is dirty information in core.
  */
 void
-xfs_ichgtime_fast(
+xfs_ichgtime(
        xfs_inode_t     *ip,
-       struct inode    *inode,
        int             flags)
 {
-       timespec_t      *tvp;
-
-       /*
-        * Atime updates for read() & friends are handled lazily now, and
-        * explicit updates must go through xfs_ichgtime()
-        */
-       ASSERT((flags & XFS_ICHGTIME_ACC) == 0);
+       struct inode    *inode = VFS_I(ip);
+       timespec_t      tv;
+       int             sync_it = 0;
 
-       /*
-        * We're not supposed to change timestamps in readonly-mounted
-        * filesystems.  Throw it away if anyone asks us.
-        */
-       if (unlikely(IS_RDONLY(inode)))
-               return;
+       tv = current_fs_time(inode->i_sb);
 
-       if (flags & XFS_ICHGTIME_MOD) {
-               tvp = &inode->i_mtime;
-               ip->i_d.di_mtime.t_sec = (__int32_t)tvp->tv_sec;
-               ip->i_d.di_mtime.t_nsec = (__int32_t)tvp->tv_nsec;
+       if ((flags & XFS_ICHGTIME_MOD) &&
+           !timespec_equal(&inode->i_mtime, &tv)) {
+               inode->i_mtime = tv;
+               sync_it = 1;
        }
-       if (flags & XFS_ICHGTIME_CHG) {
-               tvp = &inode->i_ctime;
-               ip->i_d.di_ctime.t_sec = (__int32_t)tvp->tv_sec;
-               ip->i_d.di_ctime.t_nsec = (__int32_t)tvp->tv_nsec;
+       if ((flags & XFS_ICHGTIME_CHG) &&
+           !timespec_equal(&inode->i_ctime, &tv)) {
+               inode->i_ctime = tv;
+               sync_it = 1;
        }
 
        /*
-        * We update the i_update_core field _after_ changing
-        * the timestamps in order to coordinate properly with
-        * xfs_iflush() so that we don't lose timestamp updates.
-        * This keeps us from having to hold the inode lock
-        * while doing this.  We use the SYNCHRONIZE macro to
-        * ensure that the compiler does not reorder the update
-        * of i_update_core above the timestamp updates above.
+        * Update complete - now make sure everyone knows that the inode
+        * is dirty.
         */
-       SYNCHRONIZE();
-       ip->i_update_core = 1;
-       if (!(inode->i_state & I_LOCK))
-               mark_inode_dirty_sync(inode);
-}
-
-
-/*
- * Pull the link count and size up from the xfs inode to the linux inode
- */
-STATIC void
-xfs_validate_fields(
-       struct inode    *ip,
-       bhv_vattr_t     *vattr)
-{
-       vattr->va_mask = XFS_AT_NLINK|XFS_AT_SIZE|XFS_AT_NBLOCKS;
-       if (!xfs_getattr(XFS_I(ip), vattr, ATTR_LAZY)) {
-               ip->i_nlink = vattr->va_nlink;
-               ip->i_blocks = vattr->va_nblocks;
-
-               /* we're under i_sem so i_size can't change under us */
-               if (i_size_read(ip) != vattr->va_size)
-                       i_size_write(ip, vattr->va_size);
-       }
+       if (sync_it)
+               xfs_mark_inode_dirty_sync(ip);
 }
 
 /*
@@ -202,66 +145,56 @@ xfs_validate_fields(
  */
 STATIC int
 xfs_init_security(
-       bhv_vnode_t     *vp,
+       struct inode    *inode,
        struct inode    *dir)
 {
-       struct inode    *ip = vn_to_inode(vp);
+       struct xfs_inode *ip = XFS_I(inode);
        size_t          length;
        void            *value;
-       char            *name;
+       unsigned char   *name;
        int             error;
 
-       error = security_inode_init_security(ip, dir, &name, &value, &length);
+       error = security_inode_init_security(inode, dir, (char **)&name,
+                                            &value, &length);
        if (error) {
                if (error == -EOPNOTSUPP)
                        return 0;
                return -error;
        }
 
-       error = xfs_attr_set(XFS_I(ip), name, value,
-                       length, ATTR_SECURE);
-       if (!error)
-               VMODIFY(vp);
+       error = xfs_attr_set(ip, name, value, length, ATTR_SECURE);
 
        kfree(name);
        kfree(value);
        return error;
 }
 
-/*
- * Determine whether a process has a valid fs_struct (kernel daemons
- * like knfsd don't have an fs_struct).
- *
- * XXX(hch):  nfsd is broken, better fix it instead.
- */
-STATIC_INLINE int
-xfs_has_fs_struct(struct task_struct *task)
+static void
+xfs_dentry_to_name(
+       struct xfs_name *namep,
+       struct dentry   *dentry)
 {
-       return (task->fs != init_task.fs);
+       namep->name = dentry->d_name.name;
+       namep->len = dentry->d_name.len;
 }
 
 STATIC void
 xfs_cleanup_inode(
        struct inode    *dir,
-       bhv_vnode_t     *vp,
-       struct dentry   *dentry,
-       int             mode)
+       struct inode    *inode,
+       struct dentry   *dentry)
 {
-       struct dentry   teardown = {};
+       struct xfs_name teardown;
 
        /* Oh, the horror.
         * If we can't add the ACL or we fail in
         * xfs_init_security we must back out.
         * ENOSPC can hit here, among other things.
         */
-       teardown.d_inode = vn_to_inode(vp);
-       teardown.d_name = dentry->d_name;
+       xfs_dentry_to_name(&teardown, dentry);
 
-       if (S_ISDIR(mode))
-               xfs_rmdir(XFS_I(dir), &teardown);
-       else
-               xfs_remove(XFS_I(dir), &teardown);
-       VN_RELE(vp);
+       xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
+       iput(inode);
 }
 
 STATIC int
@@ -271,80 +204,59 @@ xfs_vn_mknod(
        int             mode,
        dev_t           rdev)
 {
-       struct inode    *ip;
-       bhv_vattr_t     vattr = { 0 };
-       bhv_vnode_t     *vp = NULL, *dvp = vn_from_inode(dir);
-       xfs_acl_t       *default_acl = NULL;
-       attrexists_t    test_default_acl = _ACL_DEFAULT_EXISTS;
+       struct inode    *inode;
+       struct xfs_inode *ip = NULL;
+       struct posix_acl *default_acl = NULL;
+       struct xfs_name name;
        int             error;
 
        /*
         * Irix uses Missed'em'V split, but doesn't want to see
         * the upper 5 bits of (14bit) major.
         */
-       if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
-               return -EINVAL;
-
-       if (unlikely(test_default_acl && test_default_acl(dvp))) {
-               if (!_ACL_ALLOC(default_acl)) {
-                       return -ENOMEM;
-               }
-               if (!_ACL_GET_DEFAULT(dvp, default_acl)) {
-                       _ACL_FREE(default_acl);
-                       default_acl = NULL;
-               }
+       if (S_ISCHR(mode) || S_ISBLK(mode)) {
+               if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
+                       return -EINVAL;
+               rdev = sysv_encode_dev(rdev);
+       } else {
+               rdev = 0;
        }
 
-       if (IS_POSIXACL(dir) && !default_acl && xfs_has_fs_struct(current))
-               mode &= ~current->fs->umask;
-
-       vattr.va_mask = XFS_AT_TYPE|XFS_AT_MODE;
-       vattr.va_mode = mode;
+       if (IS_POSIXACL(dir)) {
+               default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
+               if (IS_ERR(default_acl))
+                       return -PTR_ERR(default_acl);
 
-       switch (mode & S_IFMT) {
-       case S_IFCHR: case S_IFBLK: case S_IFIFO: case S_IFSOCK:
-               vattr.va_rdev = sysv_encode_dev(rdev);
-               vattr.va_mask |= XFS_AT_RDEV;
-               /*FALLTHROUGH*/
-       case S_IFREG:
-               error = xfs_create(XFS_I(dir), dentry, &vattr, &vp, NULL);
-               break;
-       case S_IFDIR:
-               error = xfs_mkdir(XFS_I(dir), dentry, &vattr, &vp, NULL);
-               break;
-       default:
-               error = EINVAL;
-               break;
+               if (!default_acl)
+                       mode &= ~current_umask();
        }
 
-       if (unlikely(!error)) {
-               error = xfs_init_security(vp, dir);
-               if (error)
-                       xfs_cleanup_inode(dir, vp, dentry, mode);
-       }
+       xfs_dentry_to_name(&name, dentry);
+       error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip, NULL);
+       if (unlikely(error))
+               goto out_free_acl;
 
-       if (unlikely(default_acl)) {
-               if (!error) {
-                       error = _ACL_INHERIT(vp, &vattr, default_acl);
-                       if (!error)
-                               VMODIFY(vp);
-                       else
-                               xfs_cleanup_inode(dir, vp, dentry, mode);
-               }
-               _ACL_FREE(default_acl);
-       }
+       inode = VFS_I(ip);
 
-       if (likely(!error)) {
-               ASSERT(vp);
-               ip = vn_to_inode(vp);
+       error = xfs_init_security(inode, dir);
+       if (unlikely(error))
+               goto out_cleanup_inode;
 
-               if (S_ISCHR(mode) || S_ISBLK(mode))
-                       ip->i_rdev = rdev;
-               else if (S_ISDIR(mode))
-                       xfs_validate_fields(ip, &vattr);
-               d_instantiate(dentry, ip);
-               xfs_validate_fields(dir, &vattr);
+       if (default_acl) {
+               error = -xfs_inherit_acl(inode, default_acl);
+               if (unlikely(error))
+                       goto out_cleanup_inode;
+               posix_acl_release(default_acl);
        }
+
+
+       d_instantiate(dentry, inode);
+       return -error;
+
+ out_cleanup_inode:
+       xfs_cleanup_inode(dir, inode, dentry);
+ out_free_acl:
+       posix_acl_release(default_acl);
        return -error;
 }
 
@@ -373,13 +285,15 @@ xfs_vn_lookup(
        struct dentry   *dentry,
        struct nameidata *nd)
 {
-       bhv_vnode_t     *cvp;
+       struct xfs_inode *cip;
+       struct xfs_name name;
        int             error;
 
        if (dentry->d_name.len >= MAXNAMELEN)
                return ERR_PTR(-ENAMETOOLONG);
 
-       error = xfs_lookup(XFS_I(dir), dentry, &cvp);
+       xfs_dentry_to_name(&name, dentry);
+       error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
        if (unlikely(error)) {
                if (unlikely(error != ENOENT))
                        return ERR_PTR(-error);
@@ -387,7 +301,47 @@ xfs_vn_lookup(
                return NULL;
        }
 
-       return d_splice_alias(vn_to_inode(cvp), dentry);
+       return d_splice_alias(VFS_I(cip), dentry);
+}
+
+STATIC struct dentry *
+xfs_vn_ci_lookup(
+       struct inode    *dir,
+       struct dentry   *dentry,
+       struct nameidata *nd)
+{
+       struct xfs_inode *ip;
+       struct xfs_name xname;
+       struct xfs_name ci_name;
+       struct qstr     dname;
+       int             error;
+
+       if (dentry->d_name.len >= MAXNAMELEN)
+               return ERR_PTR(-ENAMETOOLONG);
+
+       xfs_dentry_to_name(&xname, dentry);
+       error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
+       if (unlikely(error)) {
+               if (unlikely(error != ENOENT))
+                       return ERR_PTR(-error);
+               /*
+                * call d_add(dentry, NULL) here when d_drop_negative_children
+                * is called in xfs_vn_mknod (ie. allow negative dentries
+                * with CI filesystems).
+                */
+               return NULL;
+       }
+
+       /* if exact match, just splice and exit */
+       if (!ci_name.name)
+               return d_splice_alias(VFS_I(ip), dentry);
+
+       /* else case-insensitive match... */
+       dname.name = ci_name.name;
+       dname.len = ci_name.len;
+       dentry = d_add_ci(dentry, VFS_I(ip), &dname);
+       kmem_free(ci_name.name);
+       return dentry;
 }
 
 STATIC int
@@ -396,24 +350,19 @@ xfs_vn_link(
        struct inode    *dir,
        struct dentry   *dentry)
 {
-       struct inode    *ip;    /* inode of guy being linked to */
-       bhv_vnode_t     *vp;    /* vp of name being linked */
-       bhv_vattr_t     vattr;
+       struct inode    *inode = old_dentry->d_inode;
+       struct xfs_name name;
        int             error;
 
-       ip = old_dentry->d_inode;       /* inode being linked to */
-       vp = vn_from_inode(ip);
+       xfs_dentry_to_name(&name, dentry);
 
-       VN_HOLD(vp);
-       error = xfs_link(XFS_I(dir), vp, dentry);
-       if (unlikely(error)) {
-               VN_RELE(vp);
-       } else {
-               VMODIFY(vn_from_inode(dir));
-               xfs_validate_fields(ip, &vattr);
-               d_instantiate(dentry, ip);
-       }
-       return -error;
+       error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
+       if (unlikely(error))
+               return -error;
+
+       atomic_inc(&inode->i_count);
+       d_instantiate(dentry, inode);
+       return 0;
 }
 
 STATIC int
@@ -421,18 +370,23 @@ xfs_vn_unlink(
        struct inode    *dir,
        struct dentry   *dentry)
 {
-       struct inode    *inode;
-       bhv_vattr_t     vattr;
+       struct xfs_name name;
        int             error;
 
-       inode = dentry->d_inode;
+       xfs_dentry_to_name(&name, dentry);
 
-       error = xfs_remove(XFS_I(dir), dentry);
-       if (likely(!error)) {
-               xfs_validate_fields(dir, &vattr);       /* size needs update */
-               xfs_validate_fields(inode, &vattr);
-       }
-       return -error;
+       error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
+       if (error)
+               return error;
+
+       /*
+        * With unlink, the VFS makes the dentry "negative": no inode,
+        * but still hashed. This is incompatible with case-insensitive
+        * mode, so invalidate (unhash) the dentry in CI-mode.
+        */
+       if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
+               d_invalidate(dentry);
+       return 0;
 }
 
 STATIC int
@@ -441,47 +395,32 @@ xfs_vn_symlink(
        struct dentry   *dentry,
        const char      *symname)
 {
-       struct inode    *ip;
-       bhv_vattr_t     va = { 0 };
-       bhv_vnode_t     *cvp;   /* used to lookup symlink to put in dentry */
+       struct inode    *inode;
+       struct xfs_inode *cip = NULL;
+       struct xfs_name name;
        int             error;
+       mode_t          mode;
 
-       cvp = NULL;
-
-       va.va_mode = S_IFLNK |
-               (irix_symlink_mode ? 0777 & ~current->fs->umask : S_IRWXUGO);
-       va.va_mask = XFS_AT_TYPE|XFS_AT_MODE;
-
-       error = xfs_symlink(XFS_I(dir), dentry, &va,
-                       (char *)symname, &cvp, NULL);
-       if (likely(!error && cvp)) {
-               error = xfs_init_security(cvp, dir);
-               if (likely(!error)) {
-                       ip = vn_to_inode(cvp);
-                       d_instantiate(dentry, ip);
-                       xfs_validate_fields(dir, &va);
-                       xfs_validate_fields(ip, &va);
-               } else {
-                       xfs_cleanup_inode(dir, cvp, dentry, 0);
-               }
-       }
-       return -error;
-}
+       mode = S_IFLNK |
+               (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
+       xfs_dentry_to_name(&name, dentry);
 
-STATIC int
-xfs_vn_rmdir(
-       struct inode    *dir,
-       struct dentry   *dentry)
-{
-       struct inode    *inode = dentry->d_inode;
-       bhv_vattr_t     vattr;
-       int             error;
+       error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip, NULL);
+       if (unlikely(error))
+               goto out;
 
-       error = xfs_rmdir(XFS_I(dir), dentry);
-       if (likely(!error)) {
-               xfs_validate_fields(inode, &vattr);
-               xfs_validate_fields(dir, &vattr);
-       }
+       inode = VFS_I(cip);
+
+       error = xfs_init_security(inode, dir);
+       if (unlikely(error))
+               goto out_cleanup_inode;
+
+       d_instantiate(dentry, inode);
+       return 0;
+
+ out_cleanup_inode:
+       xfs_cleanup_inode(dir, inode, dentry);
+ out:
        return -error;
 }
 
@@ -493,21 +432,15 @@ xfs_vn_rename(
        struct dentry   *ndentry)
 {
        struct inode    *new_inode = ndentry->d_inode;
-       bhv_vnode_t     *tvp;   /* target directory */
-       bhv_vattr_t     vattr;
-       int             error;
+       struct xfs_name oname;
+       struct xfs_name nname;
 
-       tvp = vn_from_inode(ndir);
+       xfs_dentry_to_name(&oname, odentry);
+       xfs_dentry_to_name(&nname, ndentry);
 
-       error = xfs_rename(XFS_I(odir), odentry, tvp, ndentry);
-       if (likely(!error)) {
-               if (new_inode)
-                       xfs_validate_fields(new_inode, &vattr);
-               xfs_validate_fields(odir, &vattr);
-               if (ndir != odir)
-                       xfs_validate_fields(ndir, &vattr);
-       }
-       return -error;
+       return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
+                          XFS_I(ndir), &nname, new_inode ?
+                                               XFS_I(new_inode) : NULL);
 }
 
 /*
@@ -553,256 +486,381 @@ xfs_vn_put_link(
                kfree(s);
 }
 
-#ifdef CONFIG_XFS_POSIX_ACL
-STATIC int
-xfs_vn_permission(
-       struct inode    *inode,
-       int             mode,
-       struct nameidata *nd)
-{
-       return -xfs_access(XFS_I(inode), mode << 6, NULL);
-}
-#else
-#define xfs_vn_permission NULL
-#endif
-
 STATIC int
 xfs_vn_getattr(
-       struct vfsmount *mnt,
-       struct dentry   *dentry,
-       struct kstat    *stat)
+       struct vfsmount         *mnt,
+       struct dentry           *dentry,
+       struct kstat            *stat)
 {
-       struct inode    *inode = dentry->d_inode;
-       bhv_vattr_t     vattr = { .va_mask = XFS_AT_STAT };
-       int             error;
-
-       error = xfs_getattr(XFS_I(inode), &vattr, ATTR_LAZY);
-       if (likely(!error)) {
-               stat->size = i_size_read(inode);
-               stat->dev = inode->i_sb->s_dev;
-               stat->rdev = (vattr.va_rdev == 0) ? 0 :
-                               MKDEV(sysv_major(vattr.va_rdev) & 0x1ff,
-                                     sysv_minor(vattr.va_rdev));
-               stat->mode = vattr.va_mode;
-               stat->nlink = vattr.va_nlink;
-               stat->uid = vattr.va_uid;
-               stat->gid = vattr.va_gid;
-               stat->ino = vattr.va_nodeid;
-               stat->atime = vattr.va_atime;
-               stat->mtime = vattr.va_mtime;
-               stat->ctime = vattr.va_ctime;
-               stat->blocks = vattr.va_nblocks;
-               stat->blksize = vattr.va_blocksize;
+       struct inode            *inode = dentry->d_inode;
+       struct xfs_inode        *ip = XFS_I(inode);
+       struct xfs_mount        *mp = ip->i_mount;
+
+       xfs_itrace_entry(ip);
+
+       if (XFS_FORCED_SHUTDOWN(mp))
+               return XFS_ERROR(EIO);
+
+       stat->size = XFS_ISIZE(ip);
+       stat->dev = inode->i_sb->s_dev;
+       stat->mode = ip->i_d.di_mode;
+       stat->nlink = ip->i_d.di_nlink;
+       stat->uid = ip->i_d.di_uid;
+       stat->gid = ip->i_d.di_gid;
+       stat->ino = ip->i_ino;
+       stat->atime = inode->i_atime;
+       stat->mtime = inode->i_mtime;
+       stat->ctime = inode->i_ctime;
+       stat->blocks =
+               XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
+
+
+       switch (inode->i_mode & S_IFMT) {
+       case S_IFBLK:
+       case S_IFCHR:
+               stat->blksize = BLKDEV_IOSIZE;
+               stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
+                                  sysv_minor(ip->i_df.if_u2.if_rdev));
+               break;
+       default:
+               if (XFS_IS_REALTIME_INODE(ip)) {
+                       /*
+                        * If the file blocks are being allocated from a
+                        * realtime volume, then return the inode's realtime
+                        * extent size or the realtime volume's extent size.
+                        */
+                       stat->blksize =
+                               xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
+               } else
+                       stat->blksize = xfs_preferred_iosize(mp);
+               stat->rdev = 0;
+               break;
        }
-       return -error;
+
+       return 0;
 }
 
 STATIC int
 xfs_vn_setattr(
        struct dentry   *dentry,
-       struct iattr    *attr)
+       struct iattr    *iattr)
 {
-       struct inode    *inode = dentry->d_inode;
-       unsigned int    ia_valid = attr->ia_valid;
-       bhv_vattr_t     vattr = { 0 };
-       int             flags = 0;
-       int             error;
-
-       if (ia_valid & ATTR_UID) {
-               vattr.va_mask |= XFS_AT_UID;
-               vattr.va_uid = attr->ia_uid;
-       }
-       if (ia_valid & ATTR_GID) {
-               vattr.va_mask |= XFS_AT_GID;
-               vattr.va_gid = attr->ia_gid;
-       }
-       if (ia_valid & ATTR_SIZE) {
-               vattr.va_mask |= XFS_AT_SIZE;
-               vattr.va_size = attr->ia_size;
-       }
-       if (ia_valid & ATTR_ATIME) {
-               vattr.va_mask |= XFS_AT_ATIME;
-               vattr.va_atime = attr->ia_atime;
-               inode->i_atime = attr->ia_atime;
-       }
-       if (ia_valid & ATTR_MTIME) {
-               vattr.va_mask |= XFS_AT_MTIME;
-               vattr.va_mtime = attr->ia_mtime;
-       }
-       if (ia_valid & ATTR_CTIME) {
-               vattr.va_mask |= XFS_AT_CTIME;
-               vattr.va_ctime = attr->ia_ctime;
-       }
-       if (ia_valid & ATTR_MODE) {
-               vattr.va_mask |= XFS_AT_MODE;
-               vattr.va_mode = attr->ia_mode;
-               if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
-                       inode->i_mode &= ~S_ISGID;
-       }
-
-       if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))
-               flags |= ATTR_UTIME;
-#ifdef ATTR_NO_BLOCK
-       if ((ia_valid & ATTR_NO_BLOCK))
-               flags |= ATTR_NONBLOCK;
-#endif
-
-       error = xfs_setattr(XFS_I(inode), &vattr, flags, NULL);
-       if (likely(!error))
-               __vn_revalidate(vn_from_inode(inode), &vattr);
-       return -error;
+       return -xfs_setattr(XFS_I(dentry->d_inode), iattr, 0);
 }
 
+/*
+ * block_truncate_page can return an error, but we can't propagate it
+ * at all here. Leave a complaint + stack trace in the syslog because
+ * this could be bad. If it is bad, we need to propagate the error further.
+ */
 STATIC void
 xfs_vn_truncate(
        struct inode    *inode)
 {
-       block_truncate_page(inode->i_mapping, inode->i_size, xfs_get_blocks);
+       int     error;
+       error = block_truncate_page(inode->i_mapping, inode->i_size,
+                                                       xfs_get_blocks);
+       WARN_ON(error);
 }
 
-STATIC int
-xfs_vn_setxattr(
-       struct dentry   *dentry,
-       const char      *name,
-       const void      *data,
-       size_t          size,
-       int             flags)
+STATIC long
+xfs_vn_fallocate(
+       struct inode    *inode,
+       int             mode,
+       loff_t          offset,
+       loff_t          len)
 {
-       bhv_vnode_t     *vp = vn_from_inode(dentry->d_inode);
-       char            *attr = (char *)name;
-       attrnames_t     *namesp;
-       int             xflags = 0;
-       int             error;
-
-       namesp = attr_lookup_namespace(attr, attr_namespaces, ATTR_NAMECOUNT);
-       if (!namesp)
-               return -EOPNOTSUPP;
-       attr += namesp->attr_namelen;
-       error = namesp->attr_capable(vp, NULL);
-       if (error)
-               return error;
+       long            error;
+       loff_t          new_size = 0;
+       xfs_flock64_t   bf;
+       xfs_inode_t     *ip = XFS_I(inode);
+
+       /* preallocation on directories not yet supported */
+       error = -ENODEV;
+       if (S_ISDIR(inode->i_mode))
+               goto out_error;
+
+       bf.l_whence = 0;
+       bf.l_start = offset;
+       bf.l_len = len;
+
+       xfs_ilock(ip, XFS_IOLOCK_EXCL);
+       error = -xfs_change_file_space(ip, XFS_IOC_RESVSP, &bf,
+                                      0, XFS_ATTR_NOLOCK);
+       if (!error && !(mode & FALLOC_FL_KEEP_SIZE) &&
+           offset + len > i_size_read(inode))
+               new_size = offset + len;
+
+       /* Change file size if needed */
+       if (new_size) {
+               struct iattr iattr;
+
+               iattr.ia_valid = ATTR_SIZE;
+               iattr.ia_size = new_size;
+               error = -xfs_setattr(ip, &iattr, XFS_ATTR_NOLOCK);
+       }
 
-       /* Convert Linux syscall to XFS internal ATTR flags */
-       if (flags & XATTR_CREATE)
-               xflags |= ATTR_CREATE;
-       if (flags & XATTR_REPLACE)
-               xflags |= ATTR_REPLACE;
-       xflags |= namesp->attr_flag;
-       return namesp->attr_set(vp, attr, (void *)data, size, xflags);
+       xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+out_error:
+       return error;
 }
 
-STATIC ssize_t
-xfs_vn_getxattr(
-       struct dentry   *dentry,
-       const char      *name,
-       void            *data,
-       size_t          size)
-{
-       bhv_vnode_t     *vp = vn_from_inode(dentry->d_inode);
-       char            *attr = (char *)name;
-       attrnames_t     *namesp;
-       int             xflags = 0;
-       ssize_t         error;
-
-       namesp = attr_lookup_namespace(attr, attr_namespaces, ATTR_NAMECOUNT);
-       if (!namesp)
-               return -EOPNOTSUPP;
-       attr += namesp->attr_namelen;
-       error = namesp->attr_capable(vp, NULL);
-       if (error)
-               return error;
+#define XFS_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
 
-       /* Convert Linux syscall to XFS internal ATTR flags */
-       if (!size) {
-               xflags |= ATTR_KERNOVAL;
-               data = NULL;
-       }
-       xflags |= namesp->attr_flag;
-       return namesp->attr_get(vp, attr, (void *)data, size, xflags);
-}
-
-STATIC ssize_t
-xfs_vn_listxattr(
-       struct dentry           *dentry,
-       char                    *data,
-       size_t                  size)
+/*
+ * Call fiemap helper to fill in user data.
+ * Returns positive errors to xfs_getbmap.
+ */
+STATIC int
+xfs_fiemap_format(
+       void                    **arg,
+       struct getbmapx         *bmv,
+       int                     *full)
 {
-       bhv_vnode_t             *vp = vn_from_inode(dentry->d_inode);
-       int                     error, xflags = ATTR_KERNAMELS;
-       ssize_t                 result;
-
-       if (!size)
-               xflags |= ATTR_KERNOVAL;
-       xflags |= capable(CAP_SYS_ADMIN) ? ATTR_KERNFULLS : ATTR_KERNORMALS;
+       int                     error;
+       struct fiemap_extent_info *fieinfo = *arg;
+       u32                     fiemap_flags = 0;
+       u64                     logical, physical, length;
+
+       /* Do nothing for a hole */
+       if (bmv->bmv_block == -1LL)
+               return 0;
+
+       logical = BBTOB(bmv->bmv_offset);
+       physical = BBTOB(bmv->bmv_block);
+       length = BBTOB(bmv->bmv_length);
+
+       if (bmv->bmv_oflags & BMV_OF_PREALLOC)
+               fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
+       else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
+               fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
+               physical = 0;   /* no block yet */
+       }
+       if (bmv->bmv_oflags & BMV_OF_LAST)
+               fiemap_flags |= FIEMAP_EXTENT_LAST;
+
+       error = fiemap_fill_next_extent(fieinfo, logical, physical,
+                                       length, fiemap_flags);
+       if (error > 0) {
+               error = 0;
+               *full = 1;      /* user array now full */
+       }
 
-       error = attr_generic_list(vp, data, size, xflags, &result);
-       if (error < 0)
-               return error;
-       return result;
+       return -error;
 }
 
 STATIC int
-xfs_vn_removexattr(
-       struct dentry   *dentry,
-       const char      *name)
+xfs_vn_fiemap(
+       struct inode            *inode,
+       struct fiemap_extent_info *fieinfo,
+       u64                     start,
+       u64                     length)
 {
-       bhv_vnode_t     *vp = vn_from_inode(dentry->d_inode);
-       char            *attr = (char *)name;
-       attrnames_t     *namesp;
-       int             xflags = 0;
-       int             error;
+       xfs_inode_t             *ip = XFS_I(inode);
+       struct getbmapx         bm;
+       int                     error;
 
-       namesp = attr_lookup_namespace(attr, attr_namespaces, ATTR_NAMECOUNT);
-       if (!namesp)
-               return -EOPNOTSUPP;
-       attr += namesp->attr_namelen;
-       error = namesp->attr_capable(vp, NULL);
+       error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
        if (error)
                return error;
-       xflags |= namesp->attr_flag;
-       return namesp->attr_remove(vp, attr, xflags);
-}
 
+       /* Set up bmap header for xfs internal routine */
+       bm.bmv_offset = BTOBB(start);
+       /* Special case for whole file */
+       if (length == FIEMAP_MAX_OFFSET)
+               bm.bmv_length = -1LL;
+       else
+               bm.bmv_length = BTOBB(length);
+
+       /* We add one because in getbmap world count includes the header */
+       bm.bmv_count = fieinfo->fi_extents_max + 1;
+       bm.bmv_iflags = BMV_IF_PREALLOC;
+       if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
+               bm.bmv_iflags |= BMV_IF_ATTRFORK;
+       if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
+               bm.bmv_iflags |= BMV_IF_DELALLOC;
 
-const struct inode_operations xfs_inode_operations = {
-       .permission             = xfs_vn_permission,
+       error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
+       if (error)
+               return -error;
+
+       return 0;
+}
+
+static const struct inode_operations xfs_inode_operations = {
+       .check_acl              = xfs_check_acl,
        .truncate               = xfs_vn_truncate,
        .getattr                = xfs_vn_getattr,
        .setattr                = xfs_vn_setattr,
-       .setxattr               = xfs_vn_setxattr,
-       .getxattr               = xfs_vn_getxattr,
+       .setxattr               = generic_setxattr,
+       .getxattr               = generic_getxattr,
+       .removexattr            = generic_removexattr,
        .listxattr              = xfs_vn_listxattr,
-       .removexattr            = xfs_vn_removexattr,
+       .fallocate              = xfs_vn_fallocate,
+       .fiemap                 = xfs_vn_fiemap,
 };
 
-const struct inode_operations xfs_dir_inode_operations = {
+static const struct inode_operations xfs_dir_inode_operations = {
        .create                 = xfs_vn_create,
        .lookup                 = xfs_vn_lookup,
        .link                   = xfs_vn_link,
        .unlink                 = xfs_vn_unlink,
        .symlink                = xfs_vn_symlink,
        .mkdir                  = xfs_vn_mkdir,
-       .rmdir                  = xfs_vn_rmdir,
+       /*
+        * Yes, XFS uses the same method for rmdir and unlink.
+        *
+        * There are some subtile differences deeper in the code,
+        * but we use S_ISDIR to check for those.
+        */
+       .rmdir                  = xfs_vn_unlink,
+       .mknod                  = xfs_vn_mknod,
+       .rename                 = xfs_vn_rename,
+       .check_acl              = xfs_check_acl,
+       .getattr                = xfs_vn_getattr,
+       .setattr                = xfs_vn_setattr,
+       .setxattr               = generic_setxattr,
+       .getxattr               = generic_getxattr,
+       .removexattr            = generic_removexattr,
+       .listxattr              = xfs_vn_listxattr,
+};
+
+static const struct inode_operations xfs_dir_ci_inode_operations = {
+       .create                 = xfs_vn_create,
+       .lookup                 = xfs_vn_ci_lookup,
+       .link                   = xfs_vn_link,
+       .unlink                 = xfs_vn_unlink,
+       .symlink                = xfs_vn_symlink,
+       .mkdir                  = xfs_vn_mkdir,
+       /*
+        * Yes, XFS uses the same method for rmdir and unlink.
+        *
+        * There are some subtile differences deeper in the code,
+        * but we use S_ISDIR to check for those.
+        */
+       .rmdir                  = xfs_vn_unlink,
        .mknod                  = xfs_vn_mknod,
        .rename                 = xfs_vn_rename,
-       .permission             = xfs_vn_permission,
+       .check_acl              = xfs_check_acl,
        .getattr                = xfs_vn_getattr,
        .setattr                = xfs_vn_setattr,
-       .setxattr               = xfs_vn_setxattr,
-       .getxattr               = xfs_vn_getxattr,
+       .setxattr               = generic_setxattr,
+       .getxattr               = generic_getxattr,
+       .removexattr            = generic_removexattr,
        .listxattr              = xfs_vn_listxattr,
-       .removexattr            = xfs_vn_removexattr,
 };
 
-const struct inode_operations xfs_symlink_inode_operations = {
+static const struct inode_operations xfs_symlink_inode_operations = {
        .readlink               = generic_readlink,
        .follow_link            = xfs_vn_follow_link,
        .put_link               = xfs_vn_put_link,
-       .permission             = xfs_vn_permission,
+       .check_acl              = xfs_check_acl,
        .getattr                = xfs_vn_getattr,
        .setattr                = xfs_vn_setattr,
-       .setxattr               = xfs_vn_setxattr,
-       .getxattr               = xfs_vn_getxattr,
+       .setxattr               = generic_setxattr,
+       .getxattr               = generic_getxattr,
+       .removexattr            = generic_removexattr,
        .listxattr              = xfs_vn_listxattr,
-       .removexattr            = xfs_vn_removexattr,
 };
+
+STATIC void
+xfs_diflags_to_iflags(
+       struct inode            *inode,
+       struct xfs_inode        *ip)
+{
+       if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
+               inode->i_flags |= S_IMMUTABLE;
+       else
+               inode->i_flags &= ~S_IMMUTABLE;
+       if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
+               inode->i_flags |= S_APPEND;
+       else
+               inode->i_flags &= ~S_APPEND;
+       if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
+               inode->i_flags |= S_SYNC;
+       else
+               inode->i_flags &= ~S_SYNC;
+       if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
+               inode->i_flags |= S_NOATIME;
+       else
+               inode->i_flags &= ~S_NOATIME;
+}
+
+/*
+ * Initialize the Linux inode, set up the operation vectors and
+ * unlock the inode.
+ *
+ * When reading existing inodes from disk this is called directly
+ * from xfs_iget, when creating a new inode it is called from
+ * xfs_ialloc after setting up the inode.
+ *
+ * We are always called with an uninitialised linux inode here.
+ * We need to initialise the necessary fields and take a reference
+ * on it.
+ */
+void
+xfs_setup_inode(
+       struct xfs_inode        *ip)
+{
+       struct inode            *inode = &ip->i_vnode;
+
+       inode->i_ino = ip->i_ino;
+       inode->i_state = I_NEW;
+       inode_add_to_lists(ip->i_mount->m_super, inode);
+
+       inode->i_mode   = ip->i_d.di_mode;
+       inode->i_nlink  = ip->i_d.di_nlink;
+       inode->i_uid    = ip->i_d.di_uid;
+       inode->i_gid    = ip->i_d.di_gid;
+
+       switch (inode->i_mode & S_IFMT) {
+       case S_IFBLK:
+       case S_IFCHR:
+               inode->i_rdev =
+                       MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
+                             sysv_minor(ip->i_df.if_u2.if_rdev));
+               break;
+       default:
+               inode->i_rdev = 0;
+               break;
+       }
+
+       inode->i_generation = ip->i_d.di_gen;
+       i_size_write(inode, ip->i_d.di_size);
+       inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
+       inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
+       inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
+       inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
+       inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
+       inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
+       xfs_diflags_to_iflags(inode, ip);
+
+       switch (inode->i_mode & S_IFMT) {
+       case S_IFREG:
+               inode->i_op = &xfs_inode_operations;
+               inode->i_fop = &xfs_file_operations;
+               inode->i_mapping->a_ops = &xfs_address_space_operations;
+               break;
+       case S_IFDIR:
+               if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
+                       inode->i_op = &xfs_dir_ci_inode_operations;
+               else
+                       inode->i_op = &xfs_dir_inode_operations;
+               inode->i_fop = &xfs_dir_file_operations;
+               break;
+       case S_IFLNK:
+               inode->i_op = &xfs_symlink_inode_operations;
+               if (!(ip->i_df.if_flags & XFS_IFINLINE))
+                       inode->i_mapping->a_ops = &xfs_address_space_operations;
+               break;
+       default:
+               inode->i_op = &xfs_inode_operations;
+               init_special_inode(inode, inode->i_mode, inode->i_rdev);
+               break;
+       }
+
+       xfs_iflags_clear(ip, XFS_INEW);
+       barrier();
+
+       unlock_new_inode(inode);
+}