[GFS2] Add generation number
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / gfs2 / ops_file.c
index 1e6f51b74f4474702a60509ea3f659fb4feb35b4..26f1d3249b0fce0803a77a67337d27f1128d8dd2 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
- * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+ * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
  *
  * This copyrighted material is made available to anyone wishing to use,
  * modify, copy, or redistribute it subject to the terms and conditions
 #include <linux/blkdev.h>
 #include <linux/mm.h>
 #include <linux/smp_lock.h>
-#include <linux/gfs2_ioctl.h>
-#include <asm/semaphore.h>
+#include <linux/fs.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/ext2_fs.h>
+#include <linux/crc32.h>
+#include <linux/iflags.h>
 #include <asm/uaccess.h>
 
 #include "gfs2.h"
+#include "lm_interface.h"
+#include "incore.h"
 #include "bmap.h"
 #include "dir.h"
 #include "glock.h"
 #include "glops.h"
 #include "inode.h"
-#include "jdata.h"
 #include "lm.h"
 #include "log.h"
 #include "meta_io.h"
@@ -36,6 +40,8 @@
 #include "quota.h"
 #include "rgrp.h"
 #include "trans.h"
+#include "util.h"
+#include "eaops.h"
 
 /* "bad" is for NFS support */
 struct filldir_bad_entry {
@@ -67,10 +73,49 @@ struct filldir_reg {
        void *fdr_opaque;
 };
 
-typedef ssize_t(*do_rw_t) (struct file *file,
-                  char __user *buf,
-                  size_t size, loff_t *offset,
-                  unsigned int num_gh, struct gfs2_holder *ghs);
+/*
+ * Most fields left uninitialised to catch anybody who tries to
+ * use them. f_flags set to prevent file_accessed() from touching
+ * any other part of this. Its use is purely as a flag so that we
+ * know (in readpage()) whether or not do to locking.
+ */
+struct file gfs2_internal_file_sentinal = {
+       .f_flags = O_NOATIME|O_RDONLY,
+};
+
+static int gfs2_read_actor(read_descriptor_t *desc, struct page *page,
+                          unsigned long offset, unsigned long size)
+{
+       char *kaddr;
+       unsigned long count = desc->count;
+
+       if (size > count)
+               size = count;
+
+       kaddr = kmap(page);
+       memcpy(desc->arg.buf, kaddr + offset, size);
+        kunmap(page);
+
+        desc->count = count - size;
+        desc->written += size;
+        desc->arg.buf += size;
+        return size;
+}
+
+int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
+                      char *buf, loff_t *pos, unsigned size)
+{
+       struct inode *inode = &ip->i_inode;
+       read_descriptor_t desc;
+       desc.written = 0;
+       desc.arg.buf = buf;
+       desc.count = size;
+       desc.error = 0;
+       do_generic_mapping_read(inode->i_mapping, ra_state,
+                               &gfs2_internal_file_sentinal, pos, &desc,
+                               gfs2_read_actor);
+       return desc.written ? desc.written : desc.error;
+}
 
 /**
  * gfs2_llseek - seek to a location in a file
@@ -86,12 +131,10 @@ typedef ssize_t(*do_rw_t) (struct file *file,
 
 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
+       struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
        struct gfs2_holder i_gh;
        loff_t error;
 
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
        if (origin == 2) {
                error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
                                           &i_gh);
@@ -105,247 +148,119 @@ static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
        return error;
 }
 
-static inline unsigned int vma2state(struct vm_area_struct *vma)
-{
-       if ((vma->vm_flags & (VM_MAYWRITE | VM_MAYSHARE)) ==
-           (VM_MAYWRITE | VM_MAYSHARE))
-               return LM_ST_EXCLUSIVE;
-       return LM_ST_SHARED;
-}
 
-static ssize_t walk_vm_hard(struct file *file, const char __user *buf, size_t size,
-                   loff_t *offset, do_rw_t operation)
+static ssize_t gfs2_direct_IO_read(struct kiocb *iocb, const struct iovec *iov,
+                                  loff_t offset, unsigned long nr_segs)
 {
-       struct gfs2_holder *ghs;
-       unsigned int num_gh = 0;
-       ssize_t count;
-       struct super_block *sb = file->f_dentry->d_inode->i_sb;
-       struct mm_struct *mm = current->mm;
-       struct vm_area_struct *vma;
-       unsigned long start = (unsigned long)buf;
-       unsigned long end = start + size;
-       int dumping = (current->flags & PF_DUMPCORE);
-       unsigned int x = 0;
-
-       for (vma = find_vma(mm, start); vma; vma = vma->vm_next) {
-               if (end <= vma->vm_start)
-                       break;
-               if (vma->vm_file &&
-                   vma->vm_file->f_dentry->d_inode->i_sb == sb) {
-                       num_gh++;
-               }
-       }
-
-       ghs = kcalloc((num_gh + 1), sizeof(struct gfs2_holder), GFP_KERNEL);
-       if (!ghs) {
-               if (!dumping)
-                       up_read(&mm->mmap_sem);
-               return -ENOMEM;
-       }
+       struct file *file = iocb->ki_filp;
+       struct address_space *mapping = file->f_mapping;
+       ssize_t retval;
 
-       for (vma = find_vma(mm, start); vma; vma = vma->vm_next) {
-               if (end <= vma->vm_start)
-                       break;
-               if (vma->vm_file) {
-                       struct inode *inode = vma->vm_file->f_dentry->d_inode;
-                       if (inode->i_sb == sb)
-                               gfs2_holder_init(get_v2ip(inode)->i_gl,
-                                                vma2state(vma), 0, &ghs[x++]);
-               }
+       retval = filemap_write_and_wait(mapping);
+       if (retval == 0) {
+               retval = mapping->a_ops->direct_IO(READ, iocb, iov, offset,
+                                                  nr_segs);
        }
-
-       if (!dumping)
-               up_read(&mm->mmap_sem);
-
-       gfs2_assert(get_v2sdp(sb), x == num_gh);
-
-       count = operation(file, buf, size, offset, num_gh, ghs);
-
-       while (num_gh--)
-               gfs2_holder_uninit(&ghs[num_gh]);
-       kfree(ghs);
-
-       return count;
+       return retval;
 }
 
 /**
- * walk_vm - Walk the vmas associated with a buffer for read or write.
- *    If any of them are gfs2, pass the gfs2 inode down to the read/write
- *    worker function so that locks can be acquired in the correct order.
- * @file: The file to read/write from/to
- * @buf: The buffer to copy to/from
- * @size: The amount of data requested
- * @offset: The current file offset
- * @operation: The read or write worker function
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
+ * __gfs2_file_aio_read - The main GFS2 read function
+ * 
+ * N.B. This is almost, but not quite the same as __generic_file_aio_read()
+ * the important subtle different being that inode->i_size isn't valid
+ * unless we are holding a lock, and we do this _only_ on the O_DIRECT
+ * path since otherwise locking is done entirely at the page cache
+ * layer.
  */
-
-static ssize_t walk_vm(struct file *file, const char __user *buf, size_t size,
-              loff_t *offset, do_rw_t operation)
+static ssize_t __gfs2_file_aio_read(struct kiocb *iocb,
+                                   const struct iovec *iov,
+                                   unsigned long nr_segs, loff_t *ppos)
 {
+       struct file *filp = iocb->ki_filp;
+       struct gfs2_inode *ip = GFS2_I(filp->f_mapping->host);
        struct gfs2_holder gh;
-
-       if (current->mm) {
-               struct super_block *sb = file->f_dentry->d_inode->i_sb;
-               struct mm_struct *mm = current->mm;
-               struct vm_area_struct *vma;
-               unsigned long start = (unsigned long)buf;
-               unsigned long end = start + size;
-               int dumping = (current->flags & PF_DUMPCORE);
-
-               if (!dumping)
-                       down_read(&mm->mmap_sem);
-
-               for (vma = find_vma(mm, start); vma; vma = vma->vm_next) {
-                       if (end <= vma->vm_start)
-                               break;
-                       if (vma->vm_file &&
-                           vma->vm_file->f_dentry->d_inode->i_sb == sb)
-                               goto do_locks;
+       ssize_t retval;
+       unsigned long seg;
+       size_t count;
+
+       count = 0;
+       for (seg = 0; seg < nr_segs; seg++) {
+               const struct iovec *iv = &iov[seg];
+
+               /*
+                * If any segment has a negative length, or the cumulative
+                * length ever wraps negative then return -EINVAL.
+                */
+               count += iv->iov_len;
+               if (unlikely((ssize_t)(count|iv->iov_len) < 0))
+                       return -EINVAL;
+               if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
+                       continue;
+               if (seg == 0)
+                       return -EFAULT;
+               nr_segs = seg;
+               count -= iv->iov_len;   /* This segment is no good */
+               break;
+       }
+
+       /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
+       if (filp->f_flags & O_DIRECT) {
+               loff_t pos = *ppos, size;
+               struct address_space *mapping;
+               struct inode *inode;
+
+               mapping = filp->f_mapping;
+               inode = mapping->host;
+               retval = 0;
+               if (!count)
+                       goto out; /* skip atime */
+
+               gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
+               retval = gfs2_glock_nq_m_atime(1, &gh);
+               if (retval)
+                       goto out;
+               if (gfs2_is_stuffed(ip)) {
+                       gfs2_glock_dq_m(1, &gh);
+                       gfs2_holder_uninit(&gh);
+                       goto fallback_to_normal;
                }
-
-               if (!dumping)
-                       up_read(&mm->mmap_sem);
-       }
-
-       return operation(file, buf, size, offset, 0, &gh);
-
-do_locks:
-       return walk_vm_hard(file, buf, size, offset, operation);
-}
-
-static ssize_t do_jdata_read(struct file *file, char __user *buf, size_t size,
-                            loff_t *offset)
-{
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       ssize_t count = 0;
-
-       if (*offset < 0)
-               return -EINVAL;
-       if (!access_ok(VERIFY_WRITE, buf, size))
-               return -EFAULT;
-
-       if (!(file->f_flags & O_LARGEFILE)) {
-               if (*offset >= MAX_NON_LFS)
-                       return -EFBIG;
-               if (*offset + size > MAX_NON_LFS)
-                       size = MAX_NON_LFS - *offset;
-       }
-
-       count = gfs2_jdata_read(ip, buf, *offset, size, gfs2_copy2user);
-
-       if (count > 0)
-               *offset += count;
-
-       return count;
-}
-
-/**
- * do_read_direct - Read bytes from a file
- * @file: The file to read from
- * @buf: The buffer to copy into
- * @size: The amount of data requested
- * @offset: The current file offset
- * @num_gh: The number of other locks we need to do the read
- * @ghs: the locks we need plus one for our lock
- *
- * Outputs: Offset - updated according to number of bytes read
- *
- * Returns: The number of bytes read, errno on failure
- */
-
-static ssize_t do_read_direct(struct file *file, char __user *buf, size_t size,
-                             loff_t *offset, unsigned int num_gh,
-                             struct gfs2_holder *ghs)
-{
-       struct inode *inode = file->f_mapping->host;
-       struct gfs2_inode *ip = get_v2ip(inode);
-       unsigned int state = LM_ST_DEFERRED;
-       int flags = 0;
-       unsigned int x;
-       ssize_t count = 0;
-       int error;
-
-       for (x = 0; x < num_gh; x++)
-               if (ghs[x].gh_gl == ip->i_gl) {
-                       state = LM_ST_SHARED;
-                       flags |= GL_LOCAL_EXCL;
-                       break;
+               size = i_size_read(inode);
+               if (pos < size) {
+                       retval = gfs2_direct_IO_read(iocb, iov, pos, nr_segs);
+                       if (retval > 0 && !is_sync_kiocb(iocb))
+                               retval = -EIOCBQUEUED;
+                       if (retval > 0)
+                               *ppos = pos + retval;
                }
-
-       gfs2_holder_init(ip->i_gl, state, flags, &ghs[num_gh]);
-
-       error = gfs2_glock_nq_m(num_gh + 1, ghs);
-       if (error)
-               goto out;
-
-       error = -EINVAL;
-       if (gfs2_is_jdata(ip))
-               goto out_gunlock;
-
-       if (gfs2_is_stuffed(ip)) {
-               size_t mask = bdev_hardsect_size(inode->i_sb->s_bdev) - 1;
-
-               if (((*offset) & mask) || (((unsigned long)buf) & mask))
-                       goto out_gunlock;
-
-               count = do_jdata_read(file, buf, size & ~mask, offset);
-       } else
-               count = generic_file_read(file, buf, size, offset);
-
-       error = 0;
-
- out_gunlock:
-       gfs2_glock_dq_m(num_gh + 1, ghs);
-
- out:
-       gfs2_holder_uninit(&ghs[num_gh]);
-
-       return (count) ? count : error;
-}
-
-/**
- * do_read_buf - Read bytes from a file
- * @file: The file to read from
- * @buf: The buffer to copy into
- * @size: The amount of data requested
- * @offset: The current file offset
- * @num_gh: The number of other locks we need to do the read
- * @ghs: the locks we need plus one for our lock
- *
- * Outputs: Offset - updated according to number of bytes read
- *
- * Returns: The number of bytes read, errno on failure
- */
-
-static ssize_t do_read_buf(struct file *file, char __user *buf, size_t size,
-                          loff_t *offset, unsigned int num_gh,
-                          struct gfs2_holder *ghs)
-{
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       ssize_t count = 0;
-       int error;
-
-       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &ghs[num_gh]);
-
-       error = gfs2_glock_nq_m_atime(num_gh + 1, ghs);
-       if (error)
+               file_accessed(filp);
+               gfs2_glock_dq_m(1, &gh);
+               gfs2_holder_uninit(&gh);
                goto out;
+       }
 
-       if (gfs2_is_jdata(ip))
-               count = do_jdata_read(file, buf, size, offset);
-       else
-               count = generic_file_read(file, buf, size, offset);
-
-       gfs2_glock_dq_m(num_gh + 1, ghs);
-
- out:
-       gfs2_holder_uninit(&ghs[num_gh]);
-
-       return (count) ? count : error;
+fallback_to_normal:
+       retval = 0;
+       if (count) {
+               for (seg = 0; seg < nr_segs; seg++) {
+                       read_descriptor_t desc;
+
+                       desc.written = 0;
+                       desc.arg.buf = iov[seg].iov_base;
+                       desc.count = iov[seg].iov_len;
+                       if (desc.count == 0)
+                               continue;
+                       desc.error = 0;
+                       do_generic_file_read(filp,ppos,&desc,file_read_actor);
+                       retval += desc.written;
+                       if (desc.error) {
+                               retval = retval ?: desc.error;
+                                break;
+                       }
+               }
+       }
+out:
+       return retval;
 }
 
 /**
@@ -360,550 +275,42 @@ static ssize_t do_read_buf(struct file *file, char __user *buf, size_t size,
  * Returns: The number of bytes read, errno on failure
  */
 
-static ssize_t gfs2_read(struct file *file, char __user *buf, size_t size,
+static ssize_t gfs2_read(struct file *filp, char __user *buf, size_t size,
                         loff_t *offset)
 {
-       atomic_inc(&get_v2sdp(file->f_mapping->host->i_sb)->sd_ops_file);
-
-       if (file->f_flags & O_DIRECT)
-               return walk_vm(file, buf, size, offset, do_read_direct);
-       else
-               return walk_vm(file, buf, size, offset, do_read_buf);
-}
-
-/**
- * grope_mapping - feel up a mapping that needs to be written
- * @buf: the start of the memory to be written
- * @size: the size of the memory to be written
- *
- * We do this after acquiring the locks on the mapping,
- * but before starting the write transaction.  We need to make
- * sure that we don't cause recursive transactions if blocks
- * need to be allocated to the file backing the mapping.
- *
- * Returns: errno
- */
-
-static int grope_mapping(const char __user *buf, size_t size)
-{
-       const char __user *stop = buf + size;
-       char c;
-
-       while (buf < stop) {
-               if (copy_from_user(&c, buf, 1))
-                       return -EFAULT;
-               buf += PAGE_CACHE_SIZE;
-               buf = (const char __user *)PAGE_ALIGN((unsigned long)buf);
-       }
-
-       return 0;
-}
-
-/**
- * do_write_direct_alloc - Write bytes to a file
- * @file: The file to write to
- * @buf: The buffer to copy from
- * @size: The amount of data requested
- * @offset: The current file offset
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
- */
-
-static ssize_t do_write_direct_alloc(struct file *file, const char __user *buf, size_t size,
-                                    loff_t *offset)
-{
-       struct inode *inode = file->f_mapping->host;
-       struct gfs2_inode *ip = get_v2ip(inode);
-       struct gfs2_sbd *sdp = ip->i_sbd;
-       struct gfs2_alloc *al = NULL;
        struct iovec local_iov = { .iov_base = buf, .iov_len = size };
-       struct buffer_head *dibh;
-       unsigned int data_blocks, ind_blocks;
-       ssize_t count;
-       int error;
-
-       gfs2_write_calc_reserv(ip, size, &data_blocks, &ind_blocks);
-
-       al = gfs2_alloc_get(ip);
-
-       error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
-       if (error)
-               goto fail;
-
-       error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
-       if (error)
-               goto fail_gunlock_q;
-
-       al->al_requested = data_blocks + ind_blocks;
-
-       error = gfs2_inplace_reserve(ip);
-       if (error)
-               goto fail_gunlock_q;
-
-       error = gfs2_trans_begin(sdp,
-                                al->al_rgd->rd_ri.ri_length + ind_blocks +
-                                RES_DINODE + RES_STATFS + RES_QUOTA, 0);
-       if (error)
-               goto fail_ipres;
-
-       if ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID)) {
-               error = gfs2_meta_inode_buffer(ip, &dibh);
-               if (error)
-                       goto fail_end_trans;
-
-               ip->i_di.di_mode &= (ip->i_di.di_mode & S_IXGRP) ?
-                       (~(S_ISUID | S_ISGID)) : (~S_ISUID);
-
-               gfs2_trans_add_bh(ip->i_gl, dibh);
-               gfs2_dinode_out(&ip->i_di, dibh->b_data);
-               brelse(dibh);
-       }
-
-       if (gfs2_is_stuffed(ip)) {
-               error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_sync, NULL);
-               if (error)
-                       goto fail_end_trans;
-       }
-
-       count = generic_file_write_nolock(file, &local_iov, 1, offset);
-       if (count < 0) {
-               error = count;
-               goto fail_end_trans;
-       }
-
-       error = gfs2_meta_inode_buffer(ip, &dibh);
-       if (error)
-               goto fail_end_trans;
-
-       if (ip->i_di.di_size < inode->i_size)
-               ip->i_di.di_size = inode->i_size;
-       ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
-
-       gfs2_trans_add_bh(ip->i_gl, dibh);
-       gfs2_dinode_out(&ip->i_di, dibh->b_data);
-       brelse(dibh);
-
-       gfs2_trans_end(sdp);
-
-       if (file->f_flags & O_SYNC)
-               gfs2_log_flush_glock(ip->i_gl);
-
-       gfs2_inplace_release(ip);
-       gfs2_quota_unlock(ip);
-       gfs2_alloc_put(ip);
-
-       if (file->f_mapping->nrpages) {
-               error = filemap_fdatawrite(file->f_mapping);
-               if (!error)
-                       error = filemap_fdatawait(file->f_mapping);
-       }
-       if (error)
-               return error;
-
-       return count;
-
- fail_end_trans:
-       gfs2_trans_end(sdp);
-
- fail_ipres:
-       gfs2_inplace_release(ip);
-
- fail_gunlock_q:
-       gfs2_quota_unlock(ip);
-
- fail:
-       gfs2_alloc_put(ip);
-
-       return error;
+       struct kiocb kiocb;
+       ssize_t ret;
+
+       init_sync_kiocb(&kiocb, filp);
+       ret = __gfs2_file_aio_read(&kiocb, &local_iov, 1, offset);
+       if (-EIOCBQUEUED == ret)
+               ret = wait_on_sync_kiocb(&kiocb);
+       return ret;
 }
 
-/**
- * do_write_direct - Write bytes to a file
- * @file: The file to write to
- * @buf: The buffer to copy from
- * @size: The amount of data requested
- * @offset: The current file offset
- * @num_gh: The number of other locks we need to do the read
- * @gh: the locks we need plus one for our lock
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
- */
-
-static ssize_t do_write_direct(struct file *file, const char __user *buf, size_t size,
-                              loff_t *offset, unsigned int num_gh,
-                              struct gfs2_holder *ghs)
+static ssize_t gfs2_file_readv(struct file *filp, const struct iovec *iov,
+                              unsigned long nr_segs, loff_t *ppos)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = ip->i_sbd;
-       struct gfs2_file *fp = get_v2fp(file);
-       unsigned int state = LM_ST_DEFERRED;
-       int alloc_required;
-       unsigned int x;
-       size_t s;
-       ssize_t count = 0;
-       int error;
-
-       if (test_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags))
-               state = LM_ST_EXCLUSIVE;
-       else
-               for (x = 0; x < num_gh; x++)
-                       if (ghs[x].gh_gl == ip->i_gl) {
-                               state = LM_ST_EXCLUSIVE;
-                               break;
-                       }
-
- restart:
-       gfs2_holder_init(ip->i_gl, state, 0, &ghs[num_gh]);
-
-       error = gfs2_glock_nq_m(num_gh + 1, ghs);
-       if (error)
-               goto out;
-
-       error = -EINVAL;
-       if (gfs2_is_jdata(ip))
-               goto out_gunlock;
-
-       if (num_gh) {
-               error = grope_mapping(buf, size);
-               if (error)
-                       goto out_gunlock;
-       }
-
-       if (file->f_flags & O_APPEND)
-               *offset = ip->i_di.di_size;
-
-       if (!(file->f_flags & O_LARGEFILE)) {
-               error = -EFBIG;
-               if (*offset >= MAX_NON_LFS)
-                       goto out_gunlock;
-               if (*offset + size > MAX_NON_LFS)
-                       size = MAX_NON_LFS - *offset;
-       }
-
-       if (gfs2_is_stuffed(ip) ||
-           *offset + size > ip->i_di.di_size ||
-           ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID)))
-               alloc_required = 1;
-       else {
-               error = gfs2_write_alloc_required(ip, *offset, size,
-                                                &alloc_required);
-               if (error)
-                       goto out_gunlock;
-       }
-
-       if (alloc_required && state != LM_ST_EXCLUSIVE) {
-               gfs2_glock_dq_m(num_gh + 1, ghs);
-               gfs2_holder_uninit(&ghs[num_gh]);
-               state = LM_ST_EXCLUSIVE;
-               goto restart;
-       }
-
-       if (alloc_required) {
-               set_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags);
-
-               /* split large writes into smaller atomic transactions */
-               while (size) {
-                       s = gfs2_tune_get(sdp, gt_max_atomic_write);
-                       if (s > size)
-                               s = size;
-
-                       error = do_write_direct_alloc(file, buf, s, offset);
-                       if (error < 0)
-                               goto out_gunlock;
-
-                       buf += error;
-                       size -= error;
-                       count += error;
-               }
-       } else {
-               struct iovec local_iov = { .iov_base = buf, .iov_len = size };
-               struct gfs2_holder t_gh;
-
-               clear_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags);
-
-               error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
-                                          GL_NEVER_RECURSE, &t_gh);
-               if (error)
-                       goto out_gunlock;
-
-               count = generic_file_write_nolock(file, &local_iov, 1, offset);
-
-               gfs2_glock_dq_uninit(&t_gh);
-       }
-
-       error = 0;
-
- out_gunlock:
-       gfs2_glock_dq_m(num_gh + 1, ghs);
-
- out:
-       gfs2_holder_uninit(&ghs[num_gh]);
-
-       return (count) ? count : error;
-}
-
-/**
- * do_do_write_buf - Write bytes to a file
- * @file: The file to write to
- * @buf: The buffer to copy from
- * @size: The amount of data requested
- * @offset: The current file offset
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
- */
-
-static ssize_t do_do_write_buf(struct file *file, const char __user *buf, size_t size,
-                              loff_t *offset)
-{
-       struct inode *inode = file->f_mapping->host;
-       struct gfs2_inode *ip = get_v2ip(inode);
-       struct gfs2_sbd *sdp = ip->i_sbd;
-       struct gfs2_alloc *al = NULL;
-       struct buffer_head *dibh;
-       unsigned int data_blocks, ind_blocks;
-       int alloc_required, journaled;
-       ssize_t count;
-       int error;
-
-       journaled = gfs2_is_jdata(ip);
-
-       gfs2_write_calc_reserv(ip, size, &data_blocks, &ind_blocks);
-
-       error = gfs2_write_alloc_required(ip, *offset, size, &alloc_required);
-       if (error)
-               return error;
-
-       if (alloc_required) {
-               al = gfs2_alloc_get(ip);
-
-               error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
-               if (error)
-                       goto fail;
-
-               error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
-               if (error)
-                       goto fail_gunlock_q;
-
-               al->al_requested = data_blocks + ind_blocks;
-
-               error = gfs2_inplace_reserve(ip);
-               if (error)
-                       goto fail_gunlock_q;
-
-               error = gfs2_trans_begin(sdp,
-                                        al->al_rgd->rd_ri.ri_length +
-                                        ind_blocks +
-                                        ((journaled) ? data_blocks : 0) +
-                                        RES_DINODE + RES_STATFS + RES_QUOTA,
-                                        0);
-               if (error)
-                       goto fail_ipres;
-       } else {
-               error = gfs2_trans_begin(sdp,
-                                       ((journaled) ? data_blocks : 0) +
-                                       RES_DINODE,
-                                       0);
-               if (error)
-                       goto fail_ipres;
-       }
-
-       if ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID)) {
-               error = gfs2_meta_inode_buffer(ip, &dibh);
-               if (error)
-                       goto fail_end_trans;
-
-               ip->i_di.di_mode &= (ip->i_di.di_mode & S_IXGRP) ?
-                                         (~(S_ISUID | S_ISGID)) : (~S_ISUID);
-
-               gfs2_trans_add_bh(ip->i_gl, dibh);
-               gfs2_dinode_out(&ip->i_di, dibh->b_data);
-               brelse(dibh);
-       }
-
-       if (journaled) {
-               count = gfs2_jdata_write(ip, buf, *offset, size,
-                                        gfs2_copy_from_user);
-               if (count < 0) {
-                       error = count;
-                       goto fail_end_trans;
-               }
-
-               *offset += count;
-       } else {
-               struct iovec local_iov = { .iov_base = buf, .iov_len = size };
-
-               count = generic_file_write_nolock(file, &local_iov, 1, offset);
-               if (count < 0) {
-                       error = count;
-                       goto fail_end_trans;
-               }
-
-               error = gfs2_meta_inode_buffer(ip, &dibh);
-               if (error)
-                       goto fail_end_trans;
-
-               if (ip->i_di.di_size < inode->i_size)
-                       ip->i_di.di_size = inode->i_size;
-               ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
-
-               gfs2_trans_add_bh(ip->i_gl, dibh);
-               gfs2_dinode_out(&ip->i_di, dibh->b_data);
-               brelse(dibh);
-       }
-
-       gfs2_trans_end(sdp);
-
-       if (file->f_flags & O_SYNC || IS_SYNC(inode)) {
-               gfs2_log_flush_glock(ip->i_gl);
-               error = filemap_fdatawrite(file->f_mapping);
-               if (error == 0)
-                       error = filemap_fdatawait(file->f_mapping);
-               if (error)
-                       goto fail_ipres;
-       }
-
-       if (alloc_required) {
-               gfs2_assert_warn(sdp, count != size ||
-                                al->al_alloced);
-               gfs2_inplace_release(ip);
-               gfs2_quota_unlock(ip);
-               gfs2_alloc_put(ip);
-       }
-
-       return count;
-
- fail_end_trans:
-       gfs2_trans_end(sdp);
-
- fail_ipres:
-       if (alloc_required)
-               gfs2_inplace_release(ip);
-
- fail_gunlock_q:
-       if (alloc_required)
-               gfs2_quota_unlock(ip);
-
- fail:
-       if (alloc_required)
-               gfs2_alloc_put(ip);
-
-       return error;
+       struct kiocb kiocb;
+       ssize_t ret;
+
+       init_sync_kiocb(&kiocb, filp);
+       ret = __gfs2_file_aio_read(&kiocb, iov, nr_segs, ppos);
+       if (-EIOCBQUEUED == ret)
+               ret = wait_on_sync_kiocb(&kiocb);
+       return ret;
 }
 
-/**
- * do_write_buf - Write bytes to a file
- * @file: The file to write to
- * @buf: The buffer to copy from
- * @size: The amount of data requested
- * @offset: The current file offset
- * @num_gh: The number of other locks we need to do the read
- * @gh: the locks we need plus one for our lock
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
- */
-
-static ssize_t do_write_buf(struct file *file, const char __user *buf, size_t size,
-                           loff_t *offset, unsigned int num_gh,
-                           struct gfs2_holder *ghs)
+static ssize_t gfs2_file_aio_read(struct kiocb *iocb, char __user *buf,
+                                 size_t count, loff_t pos)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = ip->i_sbd;
-       size_t s;
-       ssize_t count = 0;
-       int error;
-
-       gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ghs[num_gh]);
-
-       error = gfs2_glock_nq_m(num_gh + 1, ghs);
-       if (error)
-               goto out;
-
-       if (num_gh) {
-               error = grope_mapping(buf, size);
-               if (error)
-                       goto out_gunlock;
-       }
-
-       if (file->f_flags & O_APPEND)
-               *offset = ip->i_di.di_size;
-
-       if (!(file->f_flags & O_LARGEFILE)) {
-               error = -EFBIG;
-               if (*offset >= MAX_NON_LFS)
-                       goto out_gunlock;
-               if (*offset + size > MAX_NON_LFS)
-                       size = MAX_NON_LFS - *offset;
-       }
-
-       /* split large writes into smaller atomic transactions */
-       while (size) {
-               s = gfs2_tune_get(sdp, gt_max_atomic_write);
-               if (s > size)
-                       s = size;
-
-               error = do_do_write_buf(file, buf, s, offset);
-               if (error < 0)
-                       goto out_gunlock;
-
-               buf += error;
-               size -= error;
-               count += error;
-       }
-
-       error = 0;
-
- out_gunlock:
-       gfs2_glock_dq_m(num_gh + 1, ghs);
+        struct iovec local_iov = { .iov_base = buf, .iov_len = count };
 
- out:
-       gfs2_holder_uninit(&ghs[num_gh]);
-
-       return (count) ? count : error;
+        BUG_ON(iocb->ki_pos != pos);
+        return __gfs2_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
 }
 
-/**
- * gfs2_write - Write bytes to a file
- * @file: The file to write to
- * @buf: The buffer to copy from
- * @size: The amount of data requested
- * @offset: The current file offset
- *
- * Outputs: Offset - updated according to number of bytes written
- *
- * Returns: The number of bytes written, errno on failure
- */
-
-static ssize_t gfs2_write(struct file *file, const char __user *buf,
-                         size_t size, loff_t *offset)
-{
-       struct inode *inode = file->f_mapping->host;
-       ssize_t count;
-
-       atomic_inc(&get_v2sdp(inode->i_sb)->sd_ops_file);
-
-       if (*offset < 0)
-               return -EINVAL;
-       if (!access_ok(VERIFY_READ, buf, size))
-               return -EFAULT;
-
-       mutex_lock(&inode->i_mutex);
-       if (file->f_flags & O_DIRECT)
-               count = walk_vm(file, buf, size, offset,
-                               do_write_direct);
-       else
-               count = walk_vm(file, buf, size, offset, do_write_buf);
-       mutex_unlock(&inode->i_mutex);
-
-       return count;
-}
 
 /**
  * filldir_reg_func - Report a directory entry to the caller of gfs2_dir_read()
@@ -926,7 +333,7 @@ static int filldir_reg_func(void *opaque, const char *name, unsigned int length,
        int error;
 
        error = fdr->fdr_filldir(fdr->fdr_opaque, name, length, offset,
-                                inum->no_formal_ino, type);
+                                inum->no_addr, type);
        if (error)
                return 1;
 
@@ -953,13 +360,14 @@ static int filldir_reg_func(void *opaque, const char *name, unsigned int length,
 
 static int readdir_reg(struct file *file, void *dirent, filldir_t filldir)
 {
-       struct gfs2_inode *dip = get_v2ip(file->f_mapping->host);
+       struct inode *dir = file->f_mapping->host;
+       struct gfs2_inode *dip = GFS2_I(dir);
        struct filldir_reg fdr;
        struct gfs2_holder d_gh;
        uint64_t offset = file->f_pos;
        int error;
 
-       fdr.fdr_sbd = dip->i_sbd;
+       fdr.fdr_sbd = GFS2_SB(dir);
        fdr.fdr_prefetch = 1;
        fdr.fdr_filldir = filldir;
        fdr.fdr_opaque = dirent;
@@ -971,7 +379,7 @@ static int readdir_reg(struct file *file, void *dirent, filldir_t filldir)
                return error;
        }
 
-       error = gfs2_dir_read(dip, &offset, &fdr, filldir_reg_func);
+       error = gfs2_dir_read(dir, &offset, &fdr, filldir_reg_func);
 
        gfs2_glock_dq_uninit(&d_gh);
 
@@ -1042,8 +450,9 @@ static int filldir_bad_func(void *opaque, const char *name, unsigned int length,
 
 static int readdir_bad(struct file *file, void *dirent, filldir_t filldir)
 {
-       struct gfs2_inode *dip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = dip->i_sbd;
+       struct inode *dir = file->f_mapping->host;
+       struct gfs2_inode *dip = GFS2_I(dir);
+       struct gfs2_sbd *sdp = GFS2_SB(dir);
        struct filldir_reg fdr;
        unsigned int entries, size;
        struct filldir_bad *fdb;
@@ -1075,7 +484,7 @@ static int readdir_bad(struct file *file, void *dirent, filldir_t filldir)
                goto out;
        }
 
-       error = gfs2_dir_read(dip, &offset, fdb, filldir_bad_func);
+       error = gfs2_dir_read(dir, &offset, fdb, filldir_bad_func);
 
        gfs2_glock_dq_uninit(&d_gh);
 
@@ -1119,8 +528,6 @@ static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
 {
        int error;
 
-       atomic_inc(&get_v2sdp(file->f_mapping->host->i_sb)->sd_ops_file);
-
        if (strcmp(current->comm, "nfsd") != 0)
                error = readdir_reg(file, dirent, filldir);
        else
@@ -1129,107 +536,153 @@ static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
        return error;
 }
 
-static int gfs2_ioctl_flags(struct gfs2_inode *ip, unsigned int cmd, unsigned long arg)
+static const u32 iflags_to_gfs2[32] = {
+       [iflag_Sync] = GFS2_DIF_SYNC,
+       [iflag_Immutable] = GFS2_DIF_IMMUTABLE,
+       [iflag_Append] = GFS2_DIF_APPENDONLY,
+       [iflag_NoAtime] = GFS2_DIF_NOATIME,
+       [iflag_Index] = GFS2_DIF_EXHASH,
+       [iflag_JournalData] = GFS2_DIF_JDATA,
+       [iflag_DirectIO] = GFS2_DIF_DIRECTIO,
+};
+
+static const u32 gfs2_to_iflags[32] = {
+       [gfs2fl_Sync] = IFLAG_SYNC,
+       [gfs2fl_Immutable] = IFLAG_IMMUTABLE,
+       [gfs2fl_AppendOnly] = IFLAG_APPEND,
+       [gfs2fl_NoAtime] = IFLAG_NOATIME,
+       [gfs2fl_ExHash] = IFLAG_INDEX,
+       [gfs2fl_Jdata] = IFLAG_JOURNAL_DATA,
+       [gfs2fl_Directio] = IFLAG_DIRECTIO,
+       [gfs2fl_InheritDirectio] = IFLAG_DIRECTIO,
+       [gfs2fl_InheritJdata] = IFLAG_JOURNAL_DATA,
+};
+
+static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 {
-       unsigned int lmode = (cmd == GFS2_IOCTL_SETFLAGS) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
-       struct buffer_head *dibh;
-       struct gfs2_holder i_gh;
+       struct inode *inode = filp->f_dentry->d_inode;
+       struct gfs2_inode *ip = GFS2_I(inode);
+       struct gfs2_holder gh;
        int error;
-       __u32 flags = 0, change;
+       u32 iflags;
 
-       if (cmd == GFS2_IOCTL_SETFLAGS) {
-               error = get_user(flags, (__u32 __user *)arg);
-               if (error)
-                       return -EFAULT;
-       }
-
-       error = gfs2_glock_nq_init(ip->i_gl, lmode, 0, &i_gh);
+       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
+       error = gfs2_glock_nq_m_atime(1, &gh);
        if (error)
                return error;
+       
+       iflags = iflags_cvt(gfs2_to_iflags, ip->i_di.di_flags);
+       if (put_user(iflags, ptr))
+               error = -EFAULT;
 
-       if (cmd == GFS2_IOCTL_SETFLAGS) {
-               change = flags ^ ip->i_di.di_flags;
-               error = -EPERM;
-               if (change & (GFS2_DIF_IMMUTABLE|GFS2_DIF_APPENDONLY)) {
-                       if (!capable(CAP_LINUX_IMMUTABLE))
-                               goto out;
-               }
-               error = -EINVAL;
-               if (flags & (GFS2_DIF_JDATA|GFS2_DIF_DIRECTIO)) {
-                       if (!S_ISREG(ip->i_di.di_mode))
-                               goto out;
-                       /* FIXME: Would be nice not to require the following test */
-                       if ((flags & GFS2_DIF_JDATA) && ip->i_di.di_size)
-                               goto out;
-               }
-               if (flags & (GFS2_DIF_INHERIT_JDATA|GFS2_DIF_INHERIT_DIRECTIO)) {
-                       if (!S_ISDIR(ip->i_di.di_mode))
-                               goto out;
-               }
+       gfs2_glock_dq_m(1, &gh);
+       gfs2_holder_uninit(&gh);
+       return error;
+}
 
-               error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
-               if (error)
-                       goto out;
+/* Flags that can be set by user space */
+#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|                   \
+                            GFS2_DIF_DIRECTIO|                 \
+                            GFS2_DIF_IMMUTABLE|                \
+                            GFS2_DIF_APPENDONLY|               \
+                            GFS2_DIF_NOATIME|                  \
+                            GFS2_DIF_SYNC|                     \
+                            GFS2_DIF_SYSTEM|                   \
+                            GFS2_DIF_INHERIT_DIRECTIO|         \
+                            GFS2_DIF_INHERIT_JDATA)
 
-               error = gfs2_meta_inode_buffer(ip, &dibh);
-               if (error)
-                       goto out_trans_end;
+/**
+ * gfs2_set_flags - set flags on an inode
+ * @inode: The inode
+ * @flags: The flags to set
+ * @mask: Indicates which flags are valid
+ *
+ */
+static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
+{
+       struct inode *inode = filp->f_dentry->d_inode;
+       struct gfs2_inode *ip = GFS2_I(inode);
+       struct gfs2_sbd *sdp = GFS2_SB(inode);
+       struct buffer_head *bh;
+       struct gfs2_holder gh;
+       int error;
+       u32 new_flags, flags;
+
+       gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       if (error) {
+               gfs2_holder_uninit(&gh);
+               return error;
+       }
 
-               ip->i_di.di_flags = flags;
+       flags = ip->i_di.di_flags;
+       new_flags = (flags & ~mask) | (reqflags & mask);
+       if ((new_flags ^ flags) == 0)
+               goto out;
 
-               gfs2_trans_add_bh(ip->i_gl, dibh);
-               gfs2_dinode_out(&ip->i_di, dibh->b_data);
+       if (S_ISDIR(inode->i_mode)) {
+               if ((new_flags ^ flags) & GFS2_DIF_JDATA)
+                       new_flags ^= (GFS2_DIF_JDATA|GFS2_DIF_INHERIT_JDATA);
+               if ((new_flags ^ flags) & GFS2_DIF_DIRECTIO)
+                       new_flags ^= (GFS2_DIF_DIRECTIO|GFS2_DIF_INHERIT_DIRECTIO);
+       }
 
-               brelse(dibh);
+       error = -EINVAL;
+       if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
+               goto out;
 
-out_trans_end:
-               gfs2_trans_end(ip->i_sbd);
-       } else {
-               flags = ip->i_di.di_flags;
+       error = -EPERM;
+       if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
+               goto out;
+       if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
+               goto out;
+       if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 
+           !capable(CAP_LINUX_IMMUTABLE))
+               goto out;
+       if (!IS_IMMUTABLE(inode)) {
+               error = permission(inode, MAY_WRITE, NULL);
+               if (error)
+                       goto out;
        }
+
+       error = gfs2_trans_begin(sdp, RES_DINODE, 0);
+       if (error)
+               goto out;
+       error = gfs2_meta_inode_buffer(ip, &bh);
+       if (error)
+               goto out_trans_end;
+       gfs2_trans_add_bh(ip->i_gl, bh, 1);
+       ip->i_di.di_flags = new_flags;
+       gfs2_dinode_out(&ip->i_di, bh->b_data);
+       brelse(bh);
+out_trans_end:
+       gfs2_trans_end(sdp);
 out:
-       gfs2_glock_dq_uninit(&i_gh);
-       if (cmd == GFS2_IOCTL_GETFLAGS) {
-               if (put_user(flags, (__u32 __user *)arg))
-                       return -EFAULT;
-       }
+       gfs2_glock_dq_uninit(&gh);
        return error;
 }
 
-/**
- * gfs2_ioctl - do an ioctl on a file
- * @inode: the inode
- * @file: the file pointer
- * @cmd: the ioctl command
- * @arg: the argument
- *
- * Returns: errno
- */
-
-static int gfs2_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
-                     unsigned long arg)
+static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 {
-       struct gfs2_inode *ip = get_v2ip(inode);
-
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
-       switch (cmd) {
-       case GFS2_IOCTL_IDENTIFY: {
-               unsigned int x = GFS2_MAGIC;
-               if (copy_to_user((unsigned int __user *)arg, &x, sizeof(unsigned int)))
-                       return -EFAULT;
-               return 0;
-
-       case GFS2_IOCTL_SETFLAGS:
-       case GFS2_IOCTL_GETFLAGS:
-               return gfs2_ioctl_flags(ip, cmd, arg);
-       }
+       u32 iflags, gfsflags;
+       if (get_user(iflags, ptr))
+               return -EFAULT;
+       gfsflags = iflags_cvt(iflags_to_gfs2, iflags);
+       return do_gfs2_set_flags(filp, gfsflags, ~0);
+}
 
-       default:
-               return -ENOTTY;
+static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+       switch(cmd) {
+       case IFLAGS_GET_IOC:
+               return gfs2_get_flags(filp, (u32 __user *)arg);
+       case IFLAGS_SET_IOC:
+               return gfs2_set_flags(filp, (u32 __user *)arg);
        }
+       return -ENOTTY;
 }
 
+
 /**
  * gfs2_mmap -
  * @file: The file to map
@@ -1240,12 +693,10 @@ static int gfs2_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
 
 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
+       struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
        struct gfs2_holder i_gh;
        int error;
 
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
        gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
        error = gfs2_glock_nq_atime(&i_gh);
        if (error) {
@@ -1253,21 +704,14 @@ static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
                return error;
        }
 
-       if (gfs2_is_jdata(ip)) {
-               if (vma->vm_flags & VM_MAYSHARE)
-                       error = -EOPNOTSUPP;
-               else
-                       vma->vm_ops = &gfs2_vm_ops_private;
-       } else {
-               /* This is VM_MAYWRITE instead of VM_WRITE because a call
-                  to mprotect() can turn on VM_WRITE later. */
-
-               if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) ==
-                   (VM_MAYSHARE | VM_MAYWRITE))
-                       vma->vm_ops = &gfs2_vm_ops_sharewrite;
-               else
-                       vma->vm_ops = &gfs2_vm_ops_private;
-       }
+       /* This is VM_MAYWRITE instead of VM_WRITE because a call
+          to mprotect() can turn on VM_WRITE later. */
+
+       if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) ==
+           (VM_MAYSHARE | VM_MAYWRITE))
+               vma->vm_ops = &gfs2_vm_ops_sharewrite;
+       else
+               vma->vm_ops = &gfs2_vm_ops_private;
 
        gfs2_glock_dq_uninit(&i_gh);
 
@@ -1284,24 +728,19 @@ static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 
 static int gfs2_open(struct inode *inode, struct file *file)
 {
-       struct gfs2_inode *ip = get_v2ip(inode);
+       struct gfs2_inode *ip = GFS2_I(inode);
        struct gfs2_holder i_gh;
        struct gfs2_file *fp;
        int error;
 
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
        fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
        if (!fp)
                return -ENOMEM;
 
-       init_MUTEX(&fp->f_fl_mutex);
-
-       fp->f_inode = ip;
-       fp->f_vfile = file;
+       mutex_init(&fp->f_fl_mutex);
 
-       gfs2_assert_warn(ip->i_sbd, !get_v2fp(file));
-       set_v2fp(file, fp);
+       gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
+       file->private_data = fp;
 
        if (S_ISREG(ip->i_di.di_mode)) {
                error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
@@ -1320,13 +759,6 @@ static int gfs2_open(struct inode *inode, struct file *file)
                if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
                        file->f_flags |= O_DIRECT;
 
-               /* Don't let the user open O_DIRECT on a jdata file */
-
-               if ((file->f_flags & O_DIRECT) && gfs2_is_jdata(ip)) {
-                       error = -EINVAL;
-                       goto fail_gunlock;
-               }
-
                gfs2_glock_dq_uninit(&i_gh);
        }
 
@@ -1336,7 +768,7 @@ static int gfs2_open(struct inode *inode, struct file *file)
        gfs2_glock_dq_uninit(&i_gh);
 
  fail:
-       set_v2fp(file, NULL);
+       file->private_data = NULL;
        kfree(fp);
 
        return error;
@@ -1352,13 +784,11 @@ static int gfs2_open(struct inode *inode, struct file *file)
 
 static int gfs2_close(struct inode *inode, struct file *file)
 {
-       struct gfs2_sbd *sdp = get_v2sdp(inode->i_sb);
+       struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
        struct gfs2_file *fp;
 
-       atomic_inc(&sdp->sd_ops_file);
-
-       fp = get_v2fp(file);
-       set_v2fp(file, NULL);
+       fp = file->private_data;
+       file->private_data = NULL;
 
        if (gfs2_assert_warn(sdp, fp))
                return -EIO;
@@ -1378,10 +808,9 @@ static int gfs2_close(struct inode *inode, struct file *file)
 
 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
 {
-       struct gfs2_inode *ip = get_v2ip(dentry->d_inode);
+       struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
 
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-       gfs2_log_flush_glock(ip->i_gl);
+       gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
 
        return 0;
 }
@@ -1397,14 +826,12 @@ static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
 
 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = ip->i_sbd;
+       struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
+       struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
        struct lm_lockname name =
                { .ln_number = ip->i_num.no_addr,
                  .ln_type = LM_TYPE_PLOCK };
 
-       atomic_inc(&sdp->sd_ops_file);
-
        if (!(fl->fl_flags & FL_POSIX))
                return -ENOLCK;
        if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
@@ -1412,20 +839,15 @@ static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
 
        if (sdp->sd_args.ar_localflocks) {
                if (IS_GETLK(cmd)) {
-                       struct file_lock *tmp;
-                       lock_kernel();
-                       tmp = posix_test_lock(file, fl);
+                       struct file_lock tmp;
+                       int ret;
+                       ret = posix_test_lock(file, fl, &tmp);
                        fl->fl_type = F_UNLCK;
-                       if (tmp)
-                               memcpy(fl, tmp, sizeof(struct file_lock));
-                       unlock_kernel();
+                       if (ret)
+                               memcpy(fl, &tmp, sizeof(struct file_lock));
                        return 0;
                } else {
-                       int error;
-                       lock_kernel();
-                       error = posix_lock_file_wait(file, fl);
-                       unlock_kernel();
-                       return error;
+                       return posix_lock_file_wait(file, fl);
                }
        }
 
@@ -1437,52 +859,11 @@ static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
                return gfs2_lm_plock(sdp, &name, file, cmd, fl);
 }
 
-/**
- * gfs2_sendfile - Send bytes to a file or socket
- * @in_file: The file to read from
- * @out_file: The file to write to
- * @count: The amount of data
- * @offset: The beginning file offset
- *
- * Outputs: offset - updated according to number of bytes read
- *
- * Returns: The number of bytes sent, errno on failure
- */
-
-static ssize_t gfs2_sendfile(struct file *in_file, loff_t *offset, size_t count,
-                            read_actor_t actor, void *target)
-{
-       struct gfs2_inode *ip = get_v2ip(in_file->f_mapping->host);
-       struct gfs2_holder gh;
-       ssize_t retval;
-
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
-       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
-
-       retval = gfs2_glock_nq_atime(&gh);
-       if (retval)
-               goto out;
-
-       if (gfs2_is_jdata(ip))
-               retval = -EOPNOTSUPP;
-       else
-               retval = generic_file_sendfile(in_file, offset, count, actor,
-                                              target);
-
-       gfs2_glock_dq(&gh);
-
- out:
-       gfs2_holder_uninit(&gh);
-
-       return retval;
-}
-
 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 {
-       struct gfs2_file *fp = get_v2fp(file);
+       struct gfs2_file *fp = file->private_data;
        struct gfs2_holder *fl_gh = &fp->f_fl_gh;
-       struct gfs2_inode *ip = fp->f_inode;
+       struct gfs2_inode *ip = GFS2_I(file->f_dentry->d_inode);
        struct gfs2_glock *gl;
        unsigned int state;
        int flags;
@@ -1491,7 +872,7 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
        state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
        flags = ((IS_SETLKW(cmd)) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
 
-       down(&fp->f_fl_mutex);
+       mutex_lock(&fp->f_fl_mutex);
 
        gl = fl_gh->gh_gl;
        if (gl) {
@@ -1502,7 +883,7 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
                                     &(struct file_lock){.fl_type = F_UNLCK});          
                gfs2_glock_dq_uninit(fl_gh);
        } else {
-               error = gfs2_glock_get(ip->i_sbd,
+               error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
                                      ip->i_num.no_addr, &gfs2_flock_glops,
                                      CREATE, &gl);
                if (error)
@@ -1519,25 +900,25 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
                        error = -EAGAIN;
        } else {
                error = flock_lock_file_wait(file, fl);
-               gfs2_assert_warn(ip->i_sbd, !error);
+               gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
        }
 
  out:
-       up(&fp->f_fl_mutex);
+       mutex_unlock(&fp->f_fl_mutex);
 
        return error;
 }
 
 static void do_unflock(struct file *file, struct file_lock *fl)
 {
-       struct gfs2_file *fp = get_v2fp(file);
+       struct gfs2_file *fp = file->private_data;
        struct gfs2_holder *fl_gh = &fp->f_fl_gh;
 
-       down(&fp->f_fl_mutex);
+       mutex_lock(&fp->f_fl_mutex);
        flock_lock_file_wait(file, fl);
        if (fl_gh->gh_gl)
                gfs2_glock_dq_uninit(fl_gh);
-       up(&fp->f_fl_mutex);
+       mutex_unlock(&fp->f_fl_mutex);
 }
 
 /**
@@ -1551,10 +932,8 @@ static void do_unflock(struct file *file, struct file_lock *fl)
 
 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
 {
-       struct gfs2_inode *ip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = ip->i_sbd;
-
-       atomic_inc(&ip->i_sbd->sd_ops_file);
+       struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
+       struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
 
        if (!(fl->fl_flags & FL_FLOCK))
                return -ENOLCK;
@@ -1571,23 +950,29 @@ static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
                return do_flock(file, cmd, fl);
 }
 
-struct file_operations gfs2_file_fops = {
+const struct file_operations gfs2_file_fops = {
        .llseek = gfs2_llseek,
        .read = gfs2_read,
-       .write = gfs2_write,
-       .ioctl = gfs2_ioctl,
+       .readv = gfs2_file_readv,
+       .aio_read = gfs2_file_aio_read,
+       .write = generic_file_write,
+       .writev = generic_file_writev,
+       .aio_write = generic_file_aio_write,
+       .unlocked_ioctl = gfs2_ioctl,
        .mmap = gfs2_mmap,
        .open = gfs2_open,
        .release = gfs2_close,
        .fsync = gfs2_fsync,
        .lock = gfs2_lock,
-       .sendfile = gfs2_sendfile,
+       .sendfile = generic_file_sendfile,
        .flock = gfs2_flock,
+       .splice_read = generic_file_splice_read,
+       .splice_write = generic_file_splice_write,
 };
 
-struct file_operations gfs2_dir_fops = {
+const struct file_operations gfs2_dir_fops = {
        .readdir = gfs2_readdir,
-       .ioctl = gfs2_ioctl,
+       .unlocked_ioctl = gfs2_ioctl,
        .open = gfs2_open,
        .release = gfs2_close,
        .fsync = gfs2_fsync,