mm: fault feedback #1
[safe/jmp/linux-2.6] / fs / gfs2 / ops_file.c
index cf2e26e..1a5e8e8 100644 (file)
@@ -1,13 +1,12 @@
 /*
  * 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
- * of the GNU General Public License v.2.
+ * of the GNU General Public License version 2.
  */
 
-#include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/completion.h>
 #include <linux/uio.h>
 #include <linux/blkdev.h>
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
-#include <linux/gfs2_ioctl.h>
 #include <linux/fs.h>
-#include <asm/semaphore.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/ext2_fs.h>
+#include <linux/crc32.h>
+#include <linux/lm_interface.h>
+#include <linux/writeback.h>
 #include <asm/uaccess.h>
 
 #include "gfs2.h"
+#include "incore.h"
 #include "bmap.h"
 #include "dir.h"
 #include "glock.h"
 #include "quota.h"
 #include "rgrp.h"
 #include "trans.h"
-
-/* "bad" is for NFS support */
-struct filldir_bad_entry {
-       char *fbe_name;
-       unsigned int fbe_length;
-       uint64_t fbe_offset;
-       struct gfs2_inum fbe_inum;
-       unsigned int fbe_type;
-};
-
-struct filldir_bad {
-       struct gfs2_sbd *fdb_sbd;
-
-       struct filldir_bad_entry *fdb_entry;
-       unsigned int fdb_entry_num;
-       unsigned int fdb_entry_off;
-
-       char *fdb_name;
-       unsigned int fdb_name_size;
-       unsigned int fdb_name_off;
-};
-
-/* For regular, non-NFS */
-struct filldir_reg {
-       struct gfs2_sbd *fdr_sbd;
-       int fdr_prefetch;
-
-       filldir_t fdr_filldir;
-       void *fdr_opaque;
-};
+#include "util.h"
+#include "eaops.h"
 
 /*
  * Most fields left uninitialised to catch anybody who tries to
@@ -73,7 +47,7 @@ struct filldir_reg {
  * 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 = {
+struct file gfs2_internal_file_sentinel = {
        .f_flags = O_NOATIME|O_RDONLY,
 };
 
@@ -87,26 +61,26 @@ static int gfs2_read_actor(read_descriptor_t *desc, struct page *page,
                size = count;
 
        kaddr = kmap(page);
-       memcpy(desc->arg.buf, kaddr + offset, size);
-        kunmap(page);
+       memcpy(desc->arg.data, kaddr + offset, size);
+       kunmap(page);
 
-        desc->count = count - size;
-        desc->written += size;
-        desc->arg.buf += size;
-        return size;
+       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_vnode;
+       struct inode *inode = &ip->i_inode;
        read_descriptor_t desc;
        desc.written = 0;
-       desc.arg.buf = buf;
+       desc.arg.data = buf;
        desc.count = size;
        desc.error = 0;
        do_generic_mapping_read(inode->i_mapping, ra_state,
-                               &gfs2_internal_file_sentinal, pos, &desc,
+                               &gfs2_internal_file_sentinel, pos, &desc,
                                gfs2_read_actor);
        return desc.written ? desc.written : desc.error;
 }
@@ -125,12 +99,10 @@ int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
 
 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);
@@ -144,216 +116,8 @@ static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
        return error;
 }
 
-
-static ssize_t gfs2_direct_IO_read(struct kiocb *iocb, const struct iovec *iov,
-                                  loff_t offset, unsigned long nr_segs)
-{
-       struct file *file = iocb->ki_filp;
-       struct address_space *mapping = file->f_mapping;
-       ssize_t retval;
-
-       retval = filemap_write_and_wait(mapping);
-       if (retval == 0) {
-               retval = mapping->a_ops->direct_IO(READ, iocb, iov, offset,
-                                                  nr_segs);
-       }
-       return retval;
-}
-
-/**
- * __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 __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 = get_v2ip(filp->f_mapping->host);
-       struct gfs2_holder gh;
-       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;
-               }
-               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;
-               }
-               file_accessed(filp);
-               gfs2_glock_dq_m(1, &gh);
-               gfs2_holder_uninit(&gh);
-               goto out;
-       }
-
-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;
-}
-
-/**
- * gfs2_read - 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
- *
- * Outputs: Offset - updated according to number of bytes read
- *
- * Returns: The number of bytes read, errno on failure
- */
-
-static ssize_t gfs2_read(struct file *filp, char __user *buf, size_t size,
-                        loff_t *offset)
-{
-       struct iovec local_iov = { .iov_base = buf, .iov_len = size };
-       struct kiocb kiocb;
-       ssize_t ret;
-
-       atomic_inc(&get_v2sdp(filp->f_mapping->host->i_sb)->sd_ops_file);
-
-       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;
-}
-
-static ssize_t gfs2_file_readv(struct file *filp, const struct iovec *iov,
-                              unsigned long nr_segs, loff_t *ppos)
-{
-       struct kiocb kiocb;
-       ssize_t ret;
-
-       atomic_inc(&get_v2sdp(filp->f_mapping->host->i_sb)->sd_ops_file);
-
-       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;
-}
-
-static ssize_t gfs2_file_aio_read(struct kiocb *iocb, char __user *buf,
-                                 size_t count, loff_t pos)
-{
-       struct file *filp = iocb->ki_filp;
-        struct iovec local_iov = { .iov_base = buf, .iov_len = count };
-
-       atomic_inc(&get_v2sdp(filp->f_mapping->host->i_sb)->sd_ops_file);
-
-        BUG_ON(iocb->ki_pos != pos);
-        return __gfs2_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
-}
-
-
-/**
- * filldir_reg_func - Report a directory entry to the caller of gfs2_dir_read()
- * @opaque: opaque data used by the function
- * @name: the name of the directory entry
- * @length: the length of the name
- * @offset: the entry's offset in the directory
- * @inum: the inode number the entry points to
- * @type: the type of inode the entry points to
- *
- * Returns: 0 on success, 1 if buffer full
- */
-
-static int filldir_reg_func(void *opaque, const char *name, unsigned int length,
-                           uint64_t offset, struct gfs2_inum *inum,
-                           unsigned int type)
-{
-       struct filldir_reg *fdr = (struct filldir_reg *)opaque;
-       struct gfs2_sbd *sdp = fdr->fdr_sbd;
-       int error;
-
-       error = fdr->fdr_filldir(fdr->fdr_opaque, name, length, offset,
-                                inum->no_formal_ino, type);
-       if (error)
-               return 1;
-
-       if (fdr->fdr_prefetch && !(length == 1 && *name == '.')) {
-               gfs2_glock_prefetch_num(sdp,
-                                      inum->no_addr, &gfs2_inode_glops,
-                                      LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY);
-               gfs2_glock_prefetch_num(sdp,
-                                      inum->no_addr, &gfs2_iopen_glops,
-                                      LM_ST_SHARED, LM_FLAG_TRY);
-       }
-
-       return 0;
-}
-
 /**
- * readdir_reg - Read directory entries from a directory
+ * gfs2_readdir - Read directory entries from a directory
  * @file: The directory to read from
  * @dirent: Buffer for dirents
  * @filldir: Function used to do the copying
@@ -361,19 +125,14 @@ static int filldir_reg_func(void *opaque, const char *name, unsigned int length,
  * Returns: errno
  */
 
-static int readdir_reg(struct file *file, void *dirent, filldir_t filldir)
+static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
 {
-       struct gfs2_inode *dip = get_v2ip(file->f_mapping->host);
-       struct filldir_reg fdr;
+       struct inode *dir = file->f_mapping->host;
+       struct gfs2_inode *dip = GFS2_I(dir);
        struct gfs2_holder d_gh;
-       uint64_t offset = file->f_pos;
+       u64 offset = file->f_pos;
        int error;
 
-       fdr.fdr_sbd = dip->i_sbd;
-       fdr.fdr_prefetch = 1;
-       fdr.fdr_filldir = filldir;
-       fdr.fdr_opaque = dirent;
-
        gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
        error = gfs2_glock_nq_atime(&d_gh);
        if (error) {
@@ -381,7 +140,7 @@ 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, dirent, filldir);
 
        gfs2_glock_dq_uninit(&d_gh);
 
@@ -391,245 +150,190 @@ static int readdir_reg(struct file *file, void *dirent, filldir_t filldir)
 }
 
 /**
- * filldir_bad_func - Report a directory entry to the caller of gfs2_dir_read()
- * @opaque: opaque data used by the function
- * @name: the name of the directory entry
- * @length: the length of the name
- * @offset: the entry's offset in the directory
- * @inum: the inode number the entry points to
- * @type: the type of inode the entry points to
+ * fsflags_cvt
+ * @table: A table of 32 u32 flags
+ * @val: a 32 bit value to convert
  *
- * For supporting NFS.
+ * This function can be used to convert between fsflags values and
+ * GFS2's own flags values.
  *
- * Returns: 0 on success, 1 if buffer full
+ * Returns: the converted flags
  */
-
-static int filldir_bad_func(void *opaque, const char *name, unsigned int length,
-                           uint64_t offset, struct gfs2_inum *inum,
-                           unsigned int type)
+static u32 fsflags_cvt(const u32 *table, u32 val)
 {
-       struct filldir_bad *fdb = (struct filldir_bad *)opaque;
-       struct gfs2_sbd *sdp = fdb->fdb_sbd;
-       struct filldir_bad_entry *fbe;
-
-       if (fdb->fdb_entry_off == fdb->fdb_entry_num ||
-           fdb->fdb_name_off + length > fdb->fdb_name_size)
-               return 1;
-
-       fbe = &fdb->fdb_entry[fdb->fdb_entry_off];
-       fbe->fbe_name = fdb->fdb_name + fdb->fdb_name_off;
-       memcpy(fbe->fbe_name, name, length);
-       fbe->fbe_length = length;
-       fbe->fbe_offset = offset;
-       fbe->fbe_inum = *inum;
-       fbe->fbe_type = type;
-
-       fdb->fdb_entry_off++;
-       fdb->fdb_name_off += length;
-
-       if (!(length == 1 && *name == '.')) {
-               gfs2_glock_prefetch_num(sdp,
-                                      inum->no_addr, &gfs2_inode_glops,
-                                      LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY);
-               gfs2_glock_prefetch_num(sdp,
-                                      inum->no_addr, &gfs2_iopen_glops,
-                                      LM_ST_SHARED, LM_FLAG_TRY);
+       u32 res = 0;
+       while(val) {
+               if (val & 1)
+                       res |= *table;
+               table++;
+               val >>= 1;
        }
-
-       return 0;
+       return res;
 }
 
-/**
- * readdir_bad - Read directory entries from a directory
- * @file: The directory to read from
- * @dirent: Buffer for dirents
- * @filldir: Function used to do the copying
- *
- * For supporting NFS.
- *
- * Returns: errno
- */
+static const u32 fsflags_to_gfs2[32] = {
+       [3] = GFS2_DIF_SYNC,
+       [4] = GFS2_DIF_IMMUTABLE,
+       [5] = GFS2_DIF_APPENDONLY,
+       [7] = GFS2_DIF_NOATIME,
+       [12] = GFS2_DIF_EXHASH,
+       [14] = GFS2_DIF_JDATA,
+       [20] = GFS2_DIF_DIRECTIO,
+};
+
+static const u32 gfs2_to_fsflags[32] = {
+       [gfs2fl_Sync] = FS_SYNC_FL,
+       [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
+       [gfs2fl_AppendOnly] = FS_APPEND_FL,
+       [gfs2fl_NoAtime] = FS_NOATIME_FL,
+       [gfs2fl_ExHash] = FS_INDEX_FL,
+       [gfs2fl_Jdata] = FS_JOURNAL_DATA_FL,
+       [gfs2fl_Directio] = FS_DIRECTIO_FL,
+       [gfs2fl_InheritDirectio] = FS_DIRECTIO_FL,
+       [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
+};
 
-static int readdir_bad(struct file *file, void *dirent, filldir_t filldir)
+static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 {
-       struct gfs2_inode *dip = get_v2ip(file->f_mapping->host);
-       struct gfs2_sbd *sdp = dip->i_sbd;
-       struct filldir_reg fdr;
-       unsigned int entries, size;
-       struct filldir_bad *fdb;
-       struct gfs2_holder d_gh;
-       uint64_t offset = file->f_pos;
-       unsigned int x;
-       struct filldir_bad_entry *fbe;
+       struct inode *inode = filp->f_path.dentry->d_inode;
+       struct gfs2_inode *ip = GFS2_I(inode);
+       struct gfs2_holder gh;
        int error;
+       u32 fsflags;
 
-       entries = gfs2_tune_get(sdp, gt_entries_per_readdir);
-       size = sizeof(struct filldir_bad) +
-           entries * (sizeof(struct filldir_bad_entry) + GFS2_FAST_NAME_SIZE);
-
-       fdb = kzalloc(size, GFP_KERNEL);
-       if (!fdb)
-               return -ENOMEM;
-
-       fdb->fdb_sbd = sdp;
-       fdb->fdb_entry = (struct filldir_bad_entry *)(fdb + 1);
-       fdb->fdb_entry_num = entries;
-       fdb->fdb_name = ((char *)fdb) + sizeof(struct filldir_bad) +
-               entries * sizeof(struct filldir_bad_entry);
-       fdb->fdb_name_size = entries * GFS2_FAST_NAME_SIZE;
-
-       gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
-       error = gfs2_glock_nq_atime(&d_gh);
-       if (error) {
-               gfs2_holder_uninit(&d_gh);
-               goto out;
-       }
-
-       error = gfs2_dir_read(dip, &offset, fdb, filldir_bad_func);
-
-       gfs2_glock_dq_uninit(&d_gh);
-
-       fdr.fdr_sbd = sdp;
-       fdr.fdr_prefetch = 0;
-       fdr.fdr_filldir = filldir;
-       fdr.fdr_opaque = dirent;
-
-       for (x = 0; x < fdb->fdb_entry_off; x++) {
-               fbe = &fdb->fdb_entry[x];
-
-               error = filldir_reg_func(&fdr,
-                                        fbe->fbe_name, fbe->fbe_length,
-                                        fbe->fbe_offset,
-                                        &fbe->fbe_inum, fbe->fbe_type);
-               if (error) {
-                       file->f_pos = fbe->fbe_offset;
-                       error = 0;
-                       goto out;
-               }
-       }
-
-       file->f_pos = offset;
+       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
+       error = gfs2_glock_nq_atime(&gh);
+       if (error)
+               return error;
 
- out:
-       kfree(fdb);
+       fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_di.di_flags);
+       if (put_user(fsflags, ptr))
+               error = -EFAULT;
 
+       gfs2_glock_dq_m(1, &gh);
+       gfs2_holder_uninit(&gh);
        return error;
 }
 
+void gfs2_set_inode_flags(struct inode *inode)
+{
+       struct gfs2_inode *ip = GFS2_I(inode);
+       struct gfs2_dinode_host *di = &ip->i_di;
+       unsigned int flags = inode->i_flags;
+
+       flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
+       if (di->di_flags & GFS2_DIF_IMMUTABLE)
+               flags |= S_IMMUTABLE;
+       if (di->di_flags & GFS2_DIF_APPENDONLY)
+               flags |= S_APPEND;
+       if (di->di_flags & GFS2_DIF_NOATIME)
+               flags |= S_NOATIME;
+       if (di->di_flags & GFS2_DIF_SYNC)
+               flags |= S_SYNC;
+       inode->i_flags = flags;
+}
+
+/* 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)
+
 /**
- * gfs2_readdir - Read directory entries from a directory
- * @file: The directory to read from
- * @dirent: Buffer for dirents
- * @filldir: Function used to do the copying
+ * gfs2_set_flags - set flags on an inode
+ * @inode: The inode
+ * @flags: The flags to set
+ * @mask: Indicates which flags are valid
  *
- * Returns: errno
  */
-
-static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
+static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 {
+       struct inode *inode = filp->f_path.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;
 
-       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
-               error = readdir_bad(file, dirent, filldir);
-
-       return error;
-}
+       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       if (error)
+               return error;
 
-static int gfs2_ioctl_flags(struct gfs2_inode *ip, unsigned int cmd, unsigned long arg)
-{
-       unsigned int lmode = (cmd == GFS2_IOCTL_SETFLAGS) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
-       struct buffer_head *dibh;
-       struct gfs2_holder i_gh;
-       int error;
-       __u32 flags = 0, change;
+       flags = ip->i_di.di_flags;
+       new_flags = (flags & ~mask) | (reqflags & mask);
+       if ((new_flags ^ flags) == 0)
+               goto out;
 
-       if (cmd == GFS2_IOCTL_SETFLAGS) {
-               error = get_user(flags, (__u32 __user *)arg);
-               if (error)
-                       return -EFAULT;
+       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);
        }
 
-       error = gfs2_glock_nq_init(ip->i_gl, lmode, 0, &i_gh);
-       if (error)
-               return error;
-
-       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;
-               }
-               if (flags & (GFS2_DIF_INHERIT_JDATA|GFS2_DIF_INHERIT_DIRECTIO)) {
-                       if (!S_ISDIR(ip->i_di.di_mode))
-                               goto out;
-               }
+       error = -EINVAL;
+       if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
+               goto out;
 
-               error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
+       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_meta_inode_buffer(ip, &dibh);
-               if (error)
-                       goto out_trans_end;
-
-               ip->i_di.di_flags = flags;
-
-               gfs2_trans_add_bh(ip->i_gl, dibh, 1);
-               gfs2_dinode_out(&ip->i_di, dibh->b_data);
-
-               brelse(dibh);
-
+       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, bh->b_data);
+       brelse(bh);
+       gfs2_set_inode_flags(inode);
 out_trans_end:
-               gfs2_trans_end(ip->i_sbd);
-       } else {
-               flags = ip->i_di.di_flags;
-       }
+       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_SETFLAGS:
-       case GFS2_IOCTL_GETFLAGS:
-               return gfs2_ioctl_flags(ip, cmd, arg);
+       u32 fsflags, gfsflags;
+       if (get_user(fsflags, ptr))
+               return -EFAULT;
+       gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
+       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 FS_IOC_GETFLAGS:
+               return gfs2_get_flags(filp, (u32 __user *)arg);
+       case FS_IOC_SETFLAGS:
+               return gfs2_set_flags(filp, (u32 __user *)arg);
        }
+       return -ENOTTY;
 }
 
+
 /**
  * gfs2_mmap -
  * @file: The file to map
@@ -640,12 +344,10 @@ 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) {
@@ -677,26 +379,21 @@ 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;
 
        mutex_init(&fp->f_fl_mutex);
 
-       fp->f_inode = ip;
-       fp->f_vfile = file;
-
-       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)) {
+       if (S_ISREG(ip->i_inode.i_mode)) {
                error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
                                           &i_gh);
                if (error)
@@ -718,13 +415,11 @@ static int gfs2_open(struct inode *inode, struct file *file)
 
        return 0;
 
- fail_gunlock:
+fail_gunlock:
        gfs2_glock_dq_uninit(&i_gh);
-
- fail:
-       set_v2fp(file, NULL);
+fail:
+       file->private_data = NULL;
        kfree(fp);
-
        return error;
 }
 
@@ -738,13 +433,11 @@ 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;
@@ -759,17 +452,63 @@ static int gfs2_close(struct inode *inode, struct file *file)
  * @file: the file that points to the dentry (we ignore this)
  * @dentry: the dentry that points to the inode to sync
  *
+ * The VFS will flush "normal" data for us. We only need to worry
+ * about metadata here. For journaled data, we just do a log flush
+ * as we can't avoid it. Otherwise we can just bale out if datasync
+ * is set. For stuffed inodes we must flush the log in order to
+ * ensure that all data is on disk.
+ *
+ * The call to write_inode_now() is there to write back metadata and
+ * the inode itself. It does also try and write the data, but thats
+ * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
+ * for us.
+ *
  * Returns: errno
  */
 
 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
 {
-       struct gfs2_inode *ip = get_v2ip(dentry->d_inode);
+       struct inode *inode = dentry->d_inode;
+       int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
+       int ret = 0;
 
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-       gfs2_log_flush_glock(ip->i_gl);
+       if (gfs2_is_jdata(GFS2_I(inode))) {
+               gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
+               return 0;
+       }
 
-       return 0;
+       if (sync_state != 0) {
+               if (!datasync)
+                       ret = write_inode_now(inode, 0);
+
+               if (gfs2_is_stuffed(GFS2_I(inode)))
+                       gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
+       }
+
+       return ret;
+}
+
+/**
+ * gfs2_setlease - acquire/release a file lease
+ * @file: the file pointer
+ * @arg: lease type
+ * @fl: file lock
+ *
+ * Returns: errno
+ */
+
+static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
+{
+       struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
+
+       /*
+        * We don't currently have a way to enforce a lease across the whole
+        * cluster; until we do, disable leases (by just returning -EINVAL),
+        * unless the administrator has requested purely local locking.
+        */
+       if (!sdp->sd_args.ar_localflocks)
+               return -EINVAL;
+       return setlease(file, arg, fl);
 }
 
 /**
@@ -783,38 +522,31 @@ 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_number = ip->i_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)
+       if ((ip->i_inode.i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
                return -ENOLCK;
 
        if (sdp->sd_args.ar_localflocks) {
                if (IS_GETLK(cmd)) {
-                       struct file_lock *tmp;
-                       lock_kernel();
-                       tmp = posix_test_lock(file, fl);
-                       fl->fl_type = F_UNLCK;
-                       if (tmp)
-                               memcpy(fl, tmp, sizeof(struct file_lock));
-                       unlock_kernel();
+                       posix_test_lock(file, fl);
                        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);
                }
        }
 
+       if (cmd == F_CANCELLK) {
+               /* Hack: */
+               cmd = F_SETLK;
+               fl->fl_type = F_UNLCK;
+       }
        if (IS_GETLK(cmd))
                return gfs2_lm_plock_get(sdp, &name, file, fl);
        else if (fl->fl_type == F_UNLCK)
@@ -823,40 +555,18 @@ 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);
-
-       atomic_inc(&ip->i_sbd->sd_ops_file);
-
-       return generic_file_sendfile(in_file, offset, count, actor, target);
-}
-
 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_path.dentry->d_inode);
        struct gfs2_glock *gl;
        unsigned int state;
        int flags;
        int error = 0;
 
        state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
-       flags = ((IS_SETLKW(cmd)) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
+       flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
 
        mutex_lock(&fp->f_fl_mutex);
 
@@ -866,11 +576,11 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
                        goto out;
                gfs2_glock_hold(gl);
                flock_lock_file_wait(file,
-                                    &(struct file_lock){.fl_type = F_UNLCK});          
+                                    &(struct file_lock){.fl_type = F_UNLCK});
                gfs2_glock_dq_uninit(fl_gh);
        } else {
-               error = gfs2_glock_get(ip->i_sbd,
-                                     ip->i_num.no_addr, &gfs2_flock_glops,
+               error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
+                                     ip->i_no_addr, &gfs2_flock_glops,
                                      CREATE, &gl);
                if (error)
                        goto out;
@@ -886,18 +596,17 @@ 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:
+out:
        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;
 
        mutex_lock(&fp->f_fl_mutex);
@@ -918,14 +627,12 @@ 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;
-       if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
+       if ((ip->i_inode.i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
                return -ENOLCK;
 
        if (sdp->sd_args.ar_localflocks)
@@ -934,35 +641,36 @@ static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
        if (fl->fl_type == F_UNLCK) {
                do_unflock(file, fl);
                return 0;
-       } else
+       } else {
                return do_flock(file, cmd, fl);
+       }
 }
 
-struct file_operations gfs2_file_fops = {
-       .llseek = gfs2_llseek,
-       .read = gfs2_read,
-       .readv = gfs2_file_readv,
-       .aio_read = gfs2_file_aio_read,
-       .write = generic_file_write,
-       .writev = generic_file_writev,
-       .aio_write = generic_file_aio_write,
-       .ioctl = gfs2_ioctl,
-       .mmap = gfs2_mmap,
-       .open = gfs2_open,
-       .release = gfs2_close,
-       .fsync = gfs2_fsync,
-       .lock = gfs2_lock,
-       .sendfile = gfs2_sendfile,
-       .flock = gfs2_flock,
+const struct file_operations gfs2_file_fops = {
+       .llseek         = gfs2_llseek,
+       .read           = do_sync_read,
+       .aio_read       = generic_file_aio_read,
+       .write          = do_sync_write,
+       .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,
+       .flock          = gfs2_flock,
+       .splice_read    = generic_file_splice_read,
+       .splice_write   = generic_file_splice_write,
+       .setlease       = gfs2_setlease,
 };
 
-struct file_operations gfs2_dir_fops = {
-       .readdir = gfs2_readdir,
-       .ioctl = gfs2_ioctl,
-       .open = gfs2_open,
-       .release = gfs2_close,
-       .fsync = gfs2_fsync,
-       .lock = gfs2_lock,
-       .flock = gfs2_flock,
+const struct file_operations gfs2_dir_fops = {
+       .readdir        = gfs2_readdir,
+       .unlocked_ioctl = gfs2_ioctl,
+       .open           = gfs2_open,
+       .release        = gfs2_close,
+       .fsync          = gfs2_fsync,
+       .lock           = gfs2_lock,
+       .flock          = gfs2_flock,
 };