]> nv-tegra.nvidia Code Review - linux-2.6.git/blobdiff - fs/gfs2/ops_file.c
GFS2: Fix up jdata writepage/delete_inode
[linux-2.6.git] / fs / gfs2 / ops_file.c
index 9dcf72aa9f6604367af395a281e6ed3513f1c4ac..fcfaaefc92fbc2b8fd4a19b7bf65e8513f3411ac 100644 (file)
@@ -7,7 +7,6 @@
  * 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/mount.h>
 #include <linux/fs.h>
 #include <linux/gfs2_ondisk.h>
 #include <linux/ext2_fs.h>
 #include <linux/crc32.h>
-#include <linux/iflags.h>
+#include <linux/lm_interface.h>
+#include <linux/writeback.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 "lm.h"
 #include "log.h"
 #include "meta_io.h"
-#include "ops_file.h"
-#include "ops_vm.h"
 #include "quota.h"
 #include "rgrp.h"
 #include "trans.h"
 #include "util.h"
 #include "eaops.h"
-
-/* "bad" is for NFS support */
-struct filldir_bad_entry {
-       char *fbe_name;
-       unsigned int fbe_length;
-       u64 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;
-};
-
-/*
- * 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_sentinel = {
-       .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_sentinel, pos, &desc,
-                               gfs2_read_actor);
-       return desc.written ? desc.written : desc.error;
-}
+#include "ops_address.h"
 
 /**
  * gfs2_llseek - seek to a location in a file
@@ -139,54 +62,17 @@ static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
                error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
                                           &i_gh);
                if (!error) {
-                       error = remote_llseek(file, offset, origin);
+                       error = generic_file_llseek_unlocked(file, offset, origin);
                        gfs2_glock_dq_uninit(&i_gh);
                }
        } else
-               error = remote_llseek(file, offset, origin);
+               error = generic_file_llseek_unlocked(file, offset, origin);
 
        return error;
 }
 
 /**
- * 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,
-                           u64 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_addr, 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
@@ -194,28 +80,22 @@ 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 inode *dir = file->f_mapping->host;
        struct gfs2_inode *dip = GFS2_I(dir);
-       struct filldir_reg fdr;
        struct gfs2_holder d_gh;
        u64 offset = file->f_pos;
        int error;
 
-       fdr.fdr_sbd = GFS2_SB(dir);
-       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);
+       gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
+       error = gfs2_glock_nq(&d_gh);
        if (error) {
                gfs2_holder_uninit(&d_gh);
                return error;
        }
 
-       error = gfs2_dir_read(dir, &offset, &fdr, filldir_reg_func);
+       error = gfs2_dir_read(dir, &offset, dirent, filldir);
 
        gfs2_glock_dq_uninit(&d_gh);
 
@@ -225,205 +105,94 @@ 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
- *
- * For supporting NFS.
- *
- * Returns: 0 on success, 1 if buffer full
- */
-
-static int filldir_bad_func(void *opaque, const char *name, unsigned int length,
-                           u64 offset, struct gfs2_inum *inum,
-                           unsigned int type)
-{
-       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);
-       }
-
-       return 0;
-}
-
-/**
- * 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
+ * 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: errno
+ * Returns: the converted flags
  */
-
-static int readdir_bad(struct file *file, void *dirent, filldir_t filldir)
+static u32 fsflags_cvt(const u32 *table, u32 val)
 {
-       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;
-       struct gfs2_holder d_gh;
-       u64 offset = file->f_pos;
-       unsigned int x;
-       struct filldir_bad_entry *fbe;
-       int error;
-
-       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(dir, &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;
-               }
+       u32 res = 0;
+       while(val) {
+               if (val & 1)
+                       res |= *table;
+               table++;
+               val >>= 1;
        }
-
-       file->f_pos = offset;
-
-out:
-       kfree(fdb);
-       return error;
-}
-
-/**
- * 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
- *
- * Returns: errno
- */
-
-static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
-{
-       int error;
-
-       if (strcmp(current->comm, "nfsd") != 0)
-               error = readdir_reg(file, dirent, filldir);
-       else
-               error = readdir_bad(file, dirent, filldir);
-
-       return error;
+       return res;
 }
 
-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 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_INHERIT_JDATA,
 };
 
-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 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_InheritJdata] = FS_JOURNAL_DATA_FL,
 };
 
 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
+       struct inode *inode = filp->f_path.dentry->d_inode;
        struct gfs2_inode *ip = GFS2_I(inode);
        struct gfs2_holder gh;
        int error;
-       u32 iflags;
+       u32 fsflags;
 
-       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
-       error = gfs2_glock_nq_m_atime(1, &gh);
+       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
+       error = gfs2_glock_nq(&gh);
        if (error)
                return error;
-       
-       iflags = iflags_cvt(gfs2_to_iflags, ip->i_di.di_flags);
-       if (put_user(iflags, ptr))
+
+       fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_di.di_flags);
+       if (!S_ISDIR(inode->i_mode) && ip->i_di.di_flags & GFS2_DIF_JDATA)
+               fsflags |= FS_JOURNAL_DATA_FL;
+       if (put_user(fsflags, ptr))
                error = -EFAULT;
 
-       gfs2_glock_dq_m(1, &gh);
+       gfs2_glock_dq(&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)
 
 /**
@@ -435,7 +204,7 @@ static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
  */
 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
+       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;
@@ -443,22 +212,19 @@ static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
        int error;
        u32 new_flags, flags;
 
-       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       error = mnt_want_write(filp->f_path.mnt);
        if (error)
                return error;
 
+       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       if (error)
+               goto out_drop_write;
+
        flags = ip->i_di.di_flags;
        new_flags = (flags & ~mask) | (reqflags & mask);
        if ((new_flags ^ flags) == 0)
                goto out;
 
-       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 = -EINVAL;
        if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
                goto out;
@@ -468,15 +234,24 @@ static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
                goto out;
        if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
                goto out;
-       if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 
+       if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
            !capable(CAP_LINUX_IMMUTABLE))
                goto out;
        if (!IS_IMMUTABLE(inode)) {
-               error = permission(inode, MAY_WRITE, NULL);
+               error = gfs2_permission(inode, MAY_WRITE);
+               if (error)
+                       goto out;
+       }
+       if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
+               if (flags & GFS2_DIF_JDATA)
+                       gfs2_log_flush(sdp, ip->i_gl);
+               error = filemap_fdatawrite(inode->i_mapping);
+               if (error)
+                       goto out;
+               error = filemap_fdatawait(inode->i_mapping);
                if (error)
                        goto out;
        }
-
        error = gfs2_trans_begin(sdp, RES_DINODE, 0);
        if (error)
                goto out;
@@ -485,35 +260,164 @@ static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
                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);
+       gfs2_dinode_out(ip, bh->b_data);
        brelse(bh);
+       gfs2_set_inode_flags(inode);
+       gfs2_set_aops(inode);
 out_trans_end:
        gfs2_trans_end(sdp);
 out:
        gfs2_glock_dq_uninit(&gh);
+out_drop_write:
+       mnt_drop_write(filp->f_path.mnt);
        return error;
 }
 
 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 {
-       u32 iflags, gfsflags;
-       if (get_user(iflags, ptr))
+       struct inode *inode = filp->f_path.dentry->d_inode;
+       u32 fsflags, gfsflags;
+       if (get_user(fsflags, ptr))
                return -EFAULT;
-       gfsflags = iflags_cvt(iflags_to_gfs2, iflags);
-       return do_gfs2_set_flags(filp, gfsflags, ~0);
+       gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
+       if (!S_ISDIR(inode->i_mode)) {
+               if (gfsflags & GFS2_DIF_INHERIT_JDATA)
+                       gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
+               return do_gfs2_set_flags(filp, gfsflags, ~0);
+       }
+       return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
 }
 
 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
        switch(cmd) {
-       case IFLAGS_GET_IOC:
+       case FS_IOC_GETFLAGS:
                return gfs2_get_flags(filp, (u32 __user *)arg);
-       case IFLAGS_SET_IOC:
+       case FS_IOC_SETFLAGS:
                return gfs2_set_flags(filp, (u32 __user *)arg);
        }
        return -ENOTTY;
 }
 
+/**
+ * gfs2_allocate_page_backing - Use bmap to allocate blocks
+ * @page: The (locked) page to allocate backing for
+ *
+ * We try to allocate all the blocks required for the page in
+ * one go. This might fail for various reasons, so we keep
+ * trying until all the blocks to back this page are allocated.
+ * If some of the blocks are already allocated, thats ok too.
+ */
+
+static int gfs2_allocate_page_backing(struct page *page)
+{
+       struct inode *inode = page->mapping->host;
+       struct buffer_head bh;
+       unsigned long size = PAGE_CACHE_SIZE;
+       u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
+
+       do {
+               bh.b_state = 0;
+               bh.b_size = size;
+               gfs2_block_map(inode, lblock, &bh, 1);
+               if (!buffer_mapped(&bh))
+                       return -EIO;
+               size -= bh.b_size;
+               lblock += (bh.b_size >> inode->i_blkbits);
+       } while(size > 0);
+       return 0;
+}
+
+/**
+ * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
+ * @vma: The virtual memory area
+ * @page: The page which is about to become writable
+ *
+ * When the page becomes writable, we need to ensure that we have
+ * blocks allocated on disk to back that page.
+ */
+
+static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page)
+{
+       struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
+       struct gfs2_inode *ip = GFS2_I(inode);
+       struct gfs2_sbd *sdp = GFS2_SB(inode);
+       unsigned long last_index;
+       u64 pos = page->index << (PAGE_CACHE_SIZE - inode->i_blkbits);
+       unsigned int data_blocks, ind_blocks, rblocks;
+       int alloc_required = 0;
+       struct gfs2_holder gh;
+       struct gfs2_alloc *al;
+       int ret;
+
+       gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
+       ret = gfs2_glock_nq(&gh);
+       if (ret)
+               goto out;
+
+       set_bit(GIF_SW_PAGED, &ip->i_flags);
+       gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
+       ret = gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE, &alloc_required);
+       if (ret || !alloc_required)
+               goto out_unlock;
+       ret = -ENOMEM;
+       al = gfs2_alloc_get(ip);
+       if (al == NULL)
+               goto out_unlock;
+
+       ret = gfs2_quota_lock_check(ip);
+       if (ret)
+               goto out_alloc_put;
+       al->al_requested = data_blocks + ind_blocks;
+       ret = gfs2_inplace_reserve(ip);
+       if (ret)
+               goto out_quota_unlock;
+
+       rblocks = RES_DINODE + ind_blocks;
+       if (gfs2_is_jdata(ip))
+               rblocks += data_blocks ? data_blocks : 1;
+       if (ind_blocks || data_blocks)
+               rblocks += RES_STATFS + RES_QUOTA;
+       ret = gfs2_trans_begin(sdp, rblocks, 0);
+       if (ret)
+               goto out_trans_fail;
+
+       lock_page(page);
+       ret = -EINVAL;
+       last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
+       if (page->index > last_index)
+               goto out_unlock_page;
+       ret = 0;
+       if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
+               goto out_unlock_page;
+       if (gfs2_is_stuffed(ip)) {
+               ret = gfs2_unstuff_dinode(ip, page);
+               if (ret)
+                       goto out_unlock_page;
+       }
+       ret = gfs2_allocate_page_backing(page);
+
+out_unlock_page:
+       unlock_page(page);
+       gfs2_trans_end(sdp);
+out_trans_fail:
+       gfs2_inplace_release(ip);
+out_quota_unlock:
+       gfs2_quota_unlock(ip);
+out_alloc_put:
+       gfs2_alloc_put(ip);
+out_unlock:
+       gfs2_glock_dq(&gh);
+out:
+       gfs2_holder_uninit(&gh);
+       return ret;
+}
+
+static struct vm_operations_struct gfs2_vm_ops = {
+       .fault = filemap_fault,
+       .page_mkwrite = gfs2_page_mkwrite,
+};
+
 
 /**
  * gfs2_mmap -
@@ -529,21 +433,14 @@ static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
        struct gfs2_holder i_gh;
        int error;
 
-       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
-       error = gfs2_glock_nq_atime(&i_gh);
+       gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
+       error = gfs2_glock_nq(&i_gh);
        if (error) {
                gfs2_holder_uninit(&i_gh);
                return error;
        }
 
-       /* 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;
+       vma->vm_ops = &gfs2_vm_ops;
 
        gfs2_glock_dq_uninit(&i_gh);
 
@@ -574,7 +471,7 @@ static int gfs2_open(struct inode *inode, struct file *file)
        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)
@@ -582,15 +479,10 @@ static int gfs2_open(struct inode *inode, struct file *file)
 
                if (!(file->f_flags & O_LARGEFILE) &&
                    ip->i_di.di_size > MAX_NON_LFS) {
-                       error = -EFBIG;
+                       error = -EOVERFLOW;
                        goto fail_gunlock;
                }
 
-               /* Listen to the Direct I/O flag */
-
-               if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
-                       file->f_flags |= O_DIRECT;
-
                gfs2_glock_dq_uninit(&i_gh);
        }
 
@@ -633,16 +525,93 @@ 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 = GFS2_I(dentry->d_inode);
+       struct inode *inode = dentry->d_inode;
+       int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
+       int ret = 0;
+
+       if (gfs2_is_jdata(GFS2_I(inode))) {
+               gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
+               return 0;
+       }
 
-       gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
+       if (sync_state != 0) {
+               if (!datasync)
+                       ret = write_inode_now(inode, 0);
 
-       return 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 generic_setlease(file, arg, fl);
+}
+
+static int gfs2_lm_plock_get(struct gfs2_sbd *sdp, struct lm_lockname *name,
+                     struct file *file, struct file_lock *fl)
+{
+       int error = -EIO;
+       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               error = sdp->sd_lockstruct.ls_ops->lm_plock_get(
+                               sdp->sd_lockstruct.ls_lockspace, name, file, fl);
+       return error;
+}
+
+static int gfs2_lm_plock(struct gfs2_sbd *sdp, struct lm_lockname *name,
+                 struct file *file, int cmd, struct file_lock *fl)
+{
+       int error = -EIO;
+       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               error = sdp->sd_lockstruct.ls_ops->lm_plock(
+                               sdp->sd_lockstruct.ls_lockspace, name, file, cmd, fl);
+       return error;
+}
+
+static int gfs2_lm_punlock(struct gfs2_sbd *sdp, struct lm_lockname *name,
+                   struct file *file, struct file_lock *fl)
+{
+       int error = -EIO;
+       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               error = sdp->sd_lockstruct.ls_ops->lm_punlock(
+                               sdp->sd_lockstruct.ls_lockspace, name, file, fl);
+       return error;
 }
 
 /**
@@ -659,28 +628,19 @@ static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
        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 };
 
        if (!(fl->fl_flags & FL_POSIX))
                return -ENOLCK;
-       if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
+       if (__mandatory_lock(&ip->i_inode))
                return -ENOLCK;
 
-       if (sdp->sd_args.ar_localflocks) {
-               if (IS_GETLK(cmd)) {
-                       struct file_lock tmp;
-                       int ret;
-                       ret = posix_test_lock(file, fl, &tmp);
-                       fl->fl_type = F_UNLCK;
-                       if (ret)
-                               memcpy(fl, &tmp, sizeof(struct file_lock));
-                       return 0;
-               } else {
-                       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)
@@ -693,14 +653,14 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
 {
        struct gfs2_file *fp = file->private_data;
        struct gfs2_holder *fl_gh = &fp->f_fl_gh;
-       struct gfs2_inode *ip = GFS2_I(file->f_dentry->d_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);
 
@@ -708,21 +668,18 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
        if (gl) {
                if (fl_gh->gh_state == state)
                        goto out;
-               gfs2_glock_hold(gl);
                flock_lock_file_wait(file,
-                                    &(struct file_lock){.fl_type = F_UNLCK});          
-               gfs2_glock_dq_uninit(fl_gh);
+                                    &(struct file_lock){.fl_type = F_UNLCK});
+               gfs2_glock_dq_wait(fl_gh);
+               gfs2_holder_reinit(state, flags, fl_gh);
        } else {
-               error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
-                                     ip->i_num.no_addr, &gfs2_flock_glops,
-                                     CREATE, &gl);
+               error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
+                                      &gfs2_flock_glops, CREATE, &gl);
                if (error)
                        goto out;
+               gfs2_holder_init(gl, state, flags, fl_gh);
+               gfs2_glock_put(gl);
        }
-
-       gfs2_holder_init(gl, state, flags, fl_gh);
-       gfs2_glock_put(gl);
-
        error = gfs2_glock_nq(fl_gh);
        if (error) {
                gfs2_holder_uninit(fl_gh);
@@ -762,50 +719,69 @@ 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 = 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 (__mandatory_lock(&ip->i_inode))
                return -ENOLCK;
 
-       if (sdp->sd_args.ar_localflocks)
-               return flock_lock_file_wait(file, fl);
-
        if (fl->fl_type == F_UNLCK) {
                do_unflock(file, fl);
                return 0;
-       } else
+       } else {
                return do_flock(file, cmd, fl);
+       }
 }
 
 const struct file_operations gfs2_file_fops = {
-       .llseek = gfs2_llseek,
-       .read = generic_file_read,
-       .readv = generic_file_readv,
-       .aio_read = generic_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 = generic_file_sendfile,
-       .flock = gfs2_flock,
-       .splice_read = generic_file_splice_read,
-       .splice_write = generic_file_splice_write,
+       .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,
 };
 
 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,
+       .readdir        = gfs2_readdir,
+       .unlocked_ioctl = gfs2_ioctl,
+       .open           = gfs2_open,
+       .release        = gfs2_close,
+       .fsync          = gfs2_fsync,
+       .lock           = gfs2_lock,
+       .flock          = gfs2_flock,
+};
+
+const struct file_operations gfs2_file_fops_nolock = {
+       .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,
+       .splice_read    = generic_file_splice_read,
+       .splice_write   = generic_file_splice_write,
+       .setlease       = gfs2_setlease,
+};
+
+const struct file_operations gfs2_dir_fops_nolock = {
+       .readdir        = gfs2_readdir,
+       .unlocked_ioctl = gfs2_ioctl,
+       .open           = gfs2_open,
+       .release        = gfs2_close,
+       .fsync          = gfs2_fsync,
 };