ocfs2/trivial: Remove trailing whitespaces
[linux-2.6.git] / fs / ocfs2 / file.c
index 4c850d0..65e9375 100644 (file)
@@ -34,6 +34,8 @@
 #include <linux/splice.h>
 #include <linux/mount.h>
 #include <linux/writeback.h>
+#include <linux/falloc.h>
+#include <linux/quotaops.h>
 
 #define MLOG_MASK_PREFIX ML_INODE
 #include <cluster/masklog.h>
 #include "inode.h"
 #include "ioctl.h"
 #include "journal.h"
+#include "locks.h"
 #include "mmap.h"
 #include "suballoc.h"
 #include "super.h"
+#include "xattr.h"
+#include "acl.h"
+#include "quota.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -62,6 +69,35 @@ static int ocfs2_sync_inode(struct inode *inode)
        return sync_mapping_buffers(inode->i_mapping);
 }
 
+static int ocfs2_init_file_private(struct inode *inode, struct file *file)
+{
+       struct ocfs2_file_private *fp;
+
+       fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
+       if (!fp)
+               return -ENOMEM;
+
+       fp->fp_file = file;
+       mutex_init(&fp->fp_mutex);
+       ocfs2_file_lock_res_init(&fp->fp_flock, fp);
+       file->private_data = fp;
+
+       return 0;
+}
+
+static void ocfs2_free_file_private(struct inode *inode, struct file *file)
+{
+       struct ocfs2_file_private *fp = file->private_data;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+       if (fp) {
+               ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
+               ocfs2_lock_res_free(&fp->fp_flock);
+               kfree(fp);
+               file->private_data = NULL;
+       }
+}
+
 static int ocfs2_file_open(struct inode *inode, struct file *file)
 {
        int status;
@@ -88,7 +124,18 @@ static int ocfs2_file_open(struct inode *inode, struct file *file)
 
        oi->ip_open_count++;
        spin_unlock(&oi->ip_lock);
-       status = 0;
+
+       status = ocfs2_init_file_private(inode, file);
+       if (status) {
+               /*
+                * We want to set open count back if we're failing the
+                * open.
+                */
+               spin_lock(&oi->ip_lock);
+               oi->ip_open_count--;
+               spin_unlock(&oi->ip_lock);
+       }
+
 leave:
        mlog_exit(status);
        return status;
@@ -107,11 +154,24 @@ static int ocfs2_file_release(struct inode *inode, struct file *file)
                oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
        spin_unlock(&oi->ip_lock);
 
+       ocfs2_free_file_private(inode, file);
+
        mlog_exit(0);
 
        return 0;
 }
 
+static int ocfs2_dir_open(struct inode *inode, struct file *file)
+{
+       return ocfs2_init_file_private(inode, file);
+}
+
+static int ocfs2_dir_release(struct inode *inode, struct file *file)
+{
+       ocfs2_free_file_private(inode, file);
+       return 0;
+}
+
 static int ocfs2_sync_file(struct file *file,
                           struct dentry *dentry,
                           int datasync)
@@ -128,8 +188,11 @@ static int ocfs2_sync_file(struct file *file,
        if (err)
                goto bail;
 
+       if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
+               goto bail;
+
        journal = osb->journal->j_journal;
-       err = journal_force_commit(journal);
+       err = jbd2_journal_force_commit(journal);
 
 bail:
        mlog_exit(err);
@@ -186,21 +249,38 @@ int ocfs2_update_inode_atime(struct inode *inode,
        int ret;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        handle_t *handle;
+       struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
 
        mlog_entry_void();
 
        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-       if (handle == NULL) {
-               ret = -ENOMEM;
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
                mlog_errno(ret);
                goto out;
        }
 
+       ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_commit;
+       }
+
+       /*
+        * Don't use ocfs2_mark_inode_dirty() here as we don't always
+        * have i_mutex to guard against concurrent changes to other
+        * inode fields.
+        */
        inode->i_atime = CURRENT_TIME;
-       ret = ocfs2_mark_inode_dirty(handle, inode, bh);
+       di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
+       di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
+
+       ret = ocfs2_journal_dirty(handle, bh);
        if (ret < 0)
                mlog_errno(ret);
 
+out_commit:
        ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 out:
        mlog_exit(ret);
@@ -230,17 +310,17 @@ bail:
        return status;
 }
 
-static int ocfs2_simple_size_update(struct inode *inode,
-                                   struct buffer_head *di_bh,
-                                   u64 new_i_size)
+int ocfs2_simple_size_update(struct inode *inode,
+                            struct buffer_head *di_bh,
+                            u64 new_i_size)
 {
        int ret;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        handle_t *handle = NULL;
 
        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-       if (handle == NULL) {
-               ret = -ENOMEM;
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
                mlog_errno(ret);
                goto out;
        }
@@ -255,6 +335,39 @@ out:
        return ret;
 }
 
+static int ocfs2_cow_file_pos(struct inode *inode,
+                             struct buffer_head *fe_bh,
+                             u64 offset)
+{
+       int status;
+       u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+       unsigned int num_clusters = 0;
+       unsigned int ext_flags = 0;
+
+       /*
+        * If the new offset is aligned to the range of the cluster, there is
+        * no space for ocfs2_zero_range_for_truncate to fill, so no need to
+        * CoW either.
+        */
+       if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
+               return 0;
+
+       status = ocfs2_get_clusters(inode, cpos, &phys,
+                                   &num_clusters, &ext_flags);
+       if (status) {
+               mlog_errno(status);
+               goto out;
+       }
+
+       if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
+               goto out;
+
+       return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
+
+out:
+       return status;
+}
+
 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
                                     struct inode *inode,
                                     struct buffer_head *fe_bh,
@@ -263,9 +376,21 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
        int status;
        handle_t *handle;
        struct ocfs2_dinode *di;
+       u64 cluster_bytes;
 
        mlog_entry_void();
 
+       /*
+        * We need to CoW the cluster contains the offset if it is reflinked
+        * since we will call ocfs2_zero_range_for_truncate later which will
+        * write "0" from offset to the end of the cluster.
+        */
+       status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
+       if (status) {
+               mlog_errno(status);
+               return status;
+       }
+
        /* TODO: This needs to actually orphan the inode in this
         * transaction. */
 
@@ -276,8 +401,8 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
                goto out;
        }
 
-       status = ocfs2_journal_access(handle, inode, fe_bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
+                                        OCFS2_JOURNAL_ACCESS_WRITE);
        if (status < 0) {
                mlog_errno(status);
                goto out_commit;
@@ -286,14 +411,15 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
        /*
         * Do this before setting i_size.
         */
-       status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size);
+       cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
+       status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
+                                              cluster_bytes);
        if (status) {
                mlog_errno(status);
                goto out_commit;
        }
 
        i_size_write(inode, new_i_size);
-       inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 
        di = (struct ocfs2_dinode *) fe_bh->b_data;
@@ -326,12 +452,9 @@ static int ocfs2_truncate_file(struct inode *inode,
                   (unsigned long long)OCFS2_I(inode)->ip_blkno,
                   (unsigned long long)new_i_size);
 
+       /* We trust di_bh because it comes from ocfs2_inode_lock(), which
+        * already validated it */
        fe = (struct ocfs2_dinode *) di_bh->b_data;
-       if (!OCFS2_IS_VALID_DINODE(fe)) {
-               OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
-               status = -EIO;
-               goto bail;
-       }
 
        mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
                        "Inode %llu, inode i_size = %lld != di "
@@ -362,21 +485,25 @@ static int ocfs2_truncate_file(struct inode *inode,
 
        down_write(&OCFS2_I(inode)->ip_alloc_sem);
 
-       /* This forces other nodes to sync and drop their pages. Do
-        * this even if we have a truncate without allocation change -
-        * ocfs2 cluster sizes can be much greater than page size, so
-        * we have to truncate them anyway.  */
-       status = ocfs2_data_lock(inode, 1);
-       if (status < 0) {
-               up_write(&OCFS2_I(inode)->ip_alloc_sem);
-
-               mlog_errno(status);
-               goto bail;
-       }
-
+       /*
+        * The inode lock forced other nodes to sync and drop their
+        * pages, which (correctly) happens even if we have a truncate
+        * without allocation change - ocfs2 cluster sizes can be much
+        * greater than page size, so we have to truncate them
+        * anyway.
+        */
        unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
        truncate_inode_pages(inode->i_mapping, new_i_size);
 
+       if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+               status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
+                                              i_size_read(inode), 1);
+               if (status)
+                       mlog_errno(status);
+
+               goto bail_unlock_sem;
+       }
+
        /* alright, we're going to need to do a full blown alloc size
         * change. Orphan the inode so that recovery can complete the
         * truncate if necessary. This does the task of marking
@@ -384,35 +511,35 @@ static int ocfs2_truncate_file(struct inode *inode,
        status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
        if (status < 0) {
                mlog_errno(status);
-               goto bail_unlock_data;
+               goto bail_unlock_sem;
        }
 
        status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
        if (status < 0) {
                mlog_errno(status);
-               goto bail_unlock_data;
+               goto bail_unlock_sem;
        }
 
        status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
        if (status < 0) {
                mlog_errno(status);
-               goto bail_unlock_data;
+               goto bail_unlock_sem;
        }
 
        /* TODO: orphan dir cleanup here. */
-bail_unlock_data:
-       ocfs2_data_unlock(inode, 1);
-
+bail_unlock_sem:
        up_write(&OCFS2_I(inode)->ip_alloc_sem);
 
 bail:
+       if (!status && OCFS2_I(inode)->ip_clusters == 0)
+               status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 
        mlog_exit(status);
        return status;
 }
 
 /*
- * extend allocation only here.
+ * extend file allocation only here.
  * we'll update all the disk stuff, and oip->alloc_size
  *
  * expect stuff to be locked, a transaction started and enough data /
@@ -421,184 +548,35 @@ bail:
  * Will return -EAGAIN, and a reason if a restart is needed.
  * If passed in, *reason will always be set, even in error.
  */
-int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
-                              struct inode *inode,
-                              u32 *logical_offset,
-                              u32 clusters_to_add,
-                              struct buffer_head *fe_bh,
-                              handle_t *handle,
-                              struct ocfs2_alloc_context *data_ac,
-                              struct ocfs2_alloc_context *meta_ac,
-                              enum ocfs2_alloc_restarted *reason_ret)
-{
-       int status = 0;
-       int free_extents;
-       struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
-       enum ocfs2_alloc_restarted reason = RESTART_NONE;
-       u32 bit_off, num_bits;
-       u64 block;
-
-       BUG_ON(!clusters_to_add);
-
-       free_extents = ocfs2_num_free_extents(osb, inode, fe);
-       if (free_extents < 0) {
-               status = free_extents;
-               mlog_errno(status);
-               goto leave;
-       }
-
-       /* there are two cases which could cause us to EAGAIN in the
-        * we-need-more-metadata case:
-        * 1) we haven't reserved *any*
-        * 2) we are so fragmented, we've needed to add metadata too
-        *    many times. */
-       if (!free_extents && !meta_ac) {
-               mlog(0, "we haven't reserved any metadata!\n");
-               status = -EAGAIN;
-               reason = RESTART_META;
-               goto leave;
-       } else if ((!free_extents)
-                  && (ocfs2_alloc_context_bits_left(meta_ac)
-                      < ocfs2_extend_meta_needed(fe))) {
-               mlog(0, "filesystem is really fragmented...\n");
-               status = -EAGAIN;
-               reason = RESTART_META;
-               goto leave;
-       }
-
-       status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
-                                     &bit_off, &num_bits);
-       if (status < 0) {
-               if (status != -ENOSPC)
-                       mlog_errno(status);
-               goto leave;
-       }
-
-       BUG_ON(num_bits > clusters_to_add);
-
-       /* reserve our write early -- insert_extent may update the inode */
-       status = ocfs2_journal_access(handle, inode, fe_bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
-       if (status < 0) {
-               mlog_errno(status);
-               goto leave;
-       }
-
-       block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
-       mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
-            num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
-       status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
-                                    *logical_offset, block, num_bits,
-                                    meta_ac);
-       if (status < 0) {
-               mlog_errno(status);
-               goto leave;
-       }
-
-       status = ocfs2_journal_dirty(handle, fe_bh);
-       if (status < 0) {
-               mlog_errno(status);
-               goto leave;
-       }
-
-       clusters_to_add -= num_bits;
-       *logical_offset += num_bits;
-
-       if (clusters_to_add) {
-               mlog(0, "need to alloc once more, clusters = %u, wanted = "
-                    "%u\n", fe->i_clusters, clusters_to_add);
-               status = -EAGAIN;
-               reason = RESTART_TRANS;
-       }
-
-leave:
-       mlog_exit(status);
-       if (reason_ret)
-               *reason_ret = reason;
-       return status;
-}
-
-/*
- * For a given allocation, determine which allocators will need to be
- * accessed, and lock them, reserving the appropriate number of bits.
- *
- * Called from ocfs2_extend_allocation() for file systems which don't
- * support holes, and from ocfs2_write() for file systems which
- * understand sparse inodes.
- */
-int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
-                         u32 clusters_to_add,
-                         struct ocfs2_alloc_context **data_ac,
-                         struct ocfs2_alloc_context **meta_ac)
+int ocfs2_add_inode_data(struct ocfs2_super *osb,
+                        struct inode *inode,
+                        u32 *logical_offset,
+                        u32 clusters_to_add,
+                        int mark_unwritten,
+                        struct buffer_head *fe_bh,
+                        handle_t *handle,
+                        struct ocfs2_alloc_context *data_ac,
+                        struct ocfs2_alloc_context *meta_ac,
+                        enum ocfs2_alloc_restarted *reason_ret)
 {
-       int ret, num_free_extents;
-       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-
-       *meta_ac = NULL;
-       *data_ac = NULL;
-
-       mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
-            "clusters_to_add = %u\n",
-            (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
-            le32_to_cpu(di->i_clusters), clusters_to_add);
-
-       num_free_extents = ocfs2_num_free_extents(osb, inode, di);
-       if (num_free_extents < 0) {
-               ret = num_free_extents;
-               mlog_errno(ret);
-               goto out;
-       }
-
-       /*
-        * Sparse allocation file systems need to be more conservative
-        * with reserving room for expansion - the actual allocation
-        * happens while we've got a journal handle open so re-taking
-        * a cluster lock (because we ran out of room for another
-        * extent) will violate ordering rules.
-        *
-        * Most of the time we'll only be seeing this 1 cluster at a time
-        * anyway.
-        */
-       if (!num_free_extents ||
-           (ocfs2_sparse_alloc(osb) && num_free_extents < clusters_to_add)) {
-               ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
-               if (ret < 0) {
-                       if (ret != -ENOSPC)
-                               mlog_errno(ret);
-                       goto out;
-               }
-       }
-
-       ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
-       if (ret < 0) {
-               if (ret != -ENOSPC)
-                       mlog_errno(ret);
-               goto out;
-       }
-
-out:
-       if (ret) {
-               if (*meta_ac) {
-                       ocfs2_free_alloc_context(*meta_ac);
-                       *meta_ac = NULL;
-               }
+       int ret;
+       struct ocfs2_extent_tree et;
 
-               /*
-                * We cannot have an error and a non null *data_ac.
-                */
-       }
+       ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
+       ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
+                                         clusters_to_add, mark_unwritten,
+                                         data_ac, meta_ac, reason_ret);
 
        return ret;
 }
 
-static int ocfs2_extend_allocation(struct inode *inode,
-                                  u32 clusters_to_add)
+static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+                                    u32 clusters_to_add, int mark_unwritten)
 {
        int status = 0;
        int restart_func = 0;
-       int drop_alloc_sem = 0;
        int credits;
-       u32 prev_clusters, logical_start;
+       u32 prev_clusters;
        struct buffer_head *bh = NULL;
        struct ocfs2_dinode *fe = NULL;
        handle_t *handle = NULL;
@@ -606,6 +584,8 @@ static int ocfs2_extend_allocation(struct inode *inode,
        struct ocfs2_alloc_context *meta_ac = NULL;
        enum ocfs2_alloc_restarted why;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       struct ocfs2_extent_tree et;
+       int did_quota = 0;
 
        mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
 
@@ -613,42 +593,33 @@ static int ocfs2_extend_allocation(struct inode *inode,
         * This function only exists for file systems which don't
         * support holes.
         */
-       BUG_ON(ocfs2_sparse_alloc(osb));
+       BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 
-       status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
-                                 OCFS2_BH_CACHED, inode);
+       status = ocfs2_read_inode_block(inode, &bh);
        if (status < 0) {
                mlog_errno(status);
                goto leave;
        }
-
        fe = (struct ocfs2_dinode *) bh->b_data;
-       if (!OCFS2_IS_VALID_DINODE(fe)) {
-               OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
-               status = -EIO;
-               goto leave;
-       }
-
-       logical_start = OCFS2_I(inode)->ip_clusters;
 
 restart_all:
        BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 
-       /* blocks peope in read/write from reading our allocation
-        * until we're done changing it. We depend on i_mutex to block
-        * other extend/truncate calls while we're here. Ordering wrt
-        * start_trans is important here -- always do it before! */
-       down_write(&OCFS2_I(inode)->ip_alloc_sem);
-       drop_alloc_sem = 1;
-
-       status = ocfs2_lock_allocators(inode, fe, clusters_to_add, &data_ac,
-                                      &meta_ac);
+       mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
+            "clusters_to_add = %u\n",
+            (unsigned long long)OCFS2_I(inode)->ip_blkno,
+            (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
+            clusters_to_add);
+       ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
+       status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
+                                      &data_ac, &meta_ac);
        if (status) {
                mlog_errno(status);
                goto leave;
        }
 
-       credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
+       credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
+                                           clusters_to_add);
        handle = ocfs2_start_trans(osb, credits);
        if (IS_ERR(handle)) {
                status = PTR_ERR(handle);
@@ -658,11 +629,18 @@ restart_all:
        }
 
 restarted_transaction:
+       if (vfs_dq_alloc_space_nodirty(inode, ocfs2_clusters_to_bytes(osb->sb,
+           clusters_to_add))) {
+               status = -EDQUOT;
+               goto leave;
+       }
+       did_quota = 1;
+
        /* reserve a write to the file entry early on - that we if we
         * run out of credits in the allocation path, we can still
         * update i_size. */
-       status = ocfs2_journal_access(handle, inode, bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+                                        OCFS2_JOURNAL_ACCESS_WRITE);
        if (status < 0) {
                mlog_errno(status);
                goto leave;
@@ -670,15 +648,16 @@ restarted_transaction:
 
        prev_clusters = OCFS2_I(inode)->ip_clusters;
 
-       status = ocfs2_do_extend_allocation(osb,
-                                           inode,
-                                           &logical_start,
-                                           clusters_to_add,
-                                           bh,
-                                           handle,
-                                           data_ac,
-                                           meta_ac,
-                                           &why);
+       status = ocfs2_add_inode_data(osb,
+                                     inode,
+                                     &logical_start,
+                                     clusters_to_add,
+                                     mark_unwritten,
+                                     bh,
+                                     handle,
+                                     data_ac,
+                                     meta_ac,
+                                     &why);
        if ((status < 0) && (status != -EAGAIN)) {
                if (status != -ENOSPC)
                        mlog_errno(status);
@@ -694,6 +673,10 @@ restarted_transaction:
        spin_lock(&OCFS2_I(inode)->ip_lock);
        clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
        spin_unlock(&OCFS2_I(inode)->ip_lock);
+       /* Release unused quota reservation */
+       vfs_dq_free_space(inode,
+                       ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
+       did_quota = 0;
 
        if (why != RESTART_NONE && clusters_to_add) {
                if (why == RESTART_META) {
@@ -705,7 +688,7 @@ restarted_transaction:
                        mlog(0, "restarting transaction.\n");
                        /* TODO: This can be more intelligent. */
                        credits = ocfs2_calc_extend_credits(osb->sb,
-                                                           fe,
+                                                           &fe->id2.i_list,
                                                            clusters_to_add);
                        status = ocfs2_extend_trans(handle, credits);
                        if (status < 0) {
@@ -723,13 +706,12 @@ restarted_transaction:
             le32_to_cpu(fe->i_clusters),
             (unsigned long long)le64_to_cpu(fe->i_size));
        mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
-            OCFS2_I(inode)->ip_clusters, i_size_read(inode));
+            OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
 
 leave:
-       if (drop_alloc_sem) {
-               up_write(&OCFS2_I(inode)->ip_alloc_sem);
-               drop_alloc_sem = 0;
-       }
+       if (status < 0 && did_quota)
+               vfs_dq_free_space(inode,
+                       ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
        if (handle) {
                ocfs2_commit_trans(osb, handle);
                handle = NULL;
@@ -746,10 +728,8 @@ leave:
                restart_func = 0;
                goto restart_all;
        }
-       if (bh) {
-               brelse(bh);
-               bh = NULL;
-       }
+       brelse(bh);
+       bh = NULL;
 
        mlog_exit(status);
        return status;
@@ -757,8 +737,7 @@ leave:
 
 /* Some parts of this taken from generic_cont_expand, which turned out
  * to be too fragile to do exactly what we need without us having to
- * worry about recursive locking in ->prepare_write() and
- * ->commit_write(). */
+ * worry about recursive locking in ->write_begin() and ->write_end(). */
 static int ocfs2_write_zero_page(struct inode *inode,
                                 u64 size)
 {
@@ -770,7 +749,7 @@ static int ocfs2_write_zero_page(struct inode *inode,
        int ret;
 
        offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
-       /* ugh.  in prepare/commit_write, if from==to==start of block, we 
+       /* ugh.  in prepare/commit_write, if from==to==start of block, we
        ** skip the prepare.  make sure we never send an offset for the start
        ** of a block
        */
@@ -846,25 +825,48 @@ out:
        return ret;
 }
 
-/* 
- * A tail_to_skip value > 0 indicates that we're being called from
- * ocfs2_file_aio_write(). This has the following implications:
- *
- * - we don't want to update i_size
- * - di_bh will be NULL, which is fine because it's only used in the
- *   case where we want to update i_size.
- * - ocfs2_zero_extend() will then only be filling the hole created
- *   between i_size and the start of the write.
- */
+int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
+{
+       int ret;
+       u32 clusters_to_add;
+       struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+       clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
+       if (clusters_to_add < oi->ip_clusters)
+               clusters_to_add = 0;
+       else
+               clusters_to_add -= oi->ip_clusters;
+
+       if (clusters_to_add) {
+               ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
+                                               clusters_to_add, 0);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+       }
+
+       /*
+        * Call this even if we don't add any clusters to the tree. We
+        * still need to zero the area between the old i_size and the
+        * new i_size.
+        */
+       ret = ocfs2_zero_extend(inode, zero_to);
+       if (ret < 0)
+               mlog_errno(ret);
+
+out:
+       return ret;
+}
+
 static int ocfs2_extend_file(struct inode *inode,
                             struct buffer_head *di_bh,
-                            u64 new_i_size,
-                            size_t tail_to_skip)
+                            u64 new_i_size)
 {
        int ret = 0;
-       u32 clusters_to_add = 0;
+       struct ocfs2_inode_info *oi = OCFS2_I(inode);
 
-       BUG_ON(!tail_to_skip && !di_bh);
+       BUG_ON(!di_bh);
 
        /* setattr sometimes calls us like this. */
        if (new_i_size == 0)
@@ -874,58 +876,60 @@ static int ocfs2_extend_file(struct inode *inode,
                goto out;
        BUG_ON(new_i_size < i_size_read(inode));
 
-       if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
-               BUG_ON(tail_to_skip != 0);
+       /*
+        * Fall through for converting inline data, even if the fs
+        * supports sparse files.
+        *
+        * The check for inline data here is legal - nobody can add
+        * the feature since we have i_mutex. We must check it again
+        * after acquiring ip_alloc_sem though, as paths like mmap
+        * might have raced us to converting the inode to extents.
+        */
+       if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+           && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
                goto out_update_size;
-       }
-
-       clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - 
-               OCFS2_I(inode)->ip_clusters;
 
-       /* 
-        * protect the pages that ocfs2_zero_extend is going to be
-        * pulling into the page cache.. we do this before the
-        * metadata extend so that we don't get into the situation
-        * where we've extended the metadata but can't get the data
-        * lock to zero.
+       /*
+        * The alloc sem blocks people in read/write from reading our
+        * allocation until we're done changing it. We depend on
+        * i_mutex to block other extend/truncate calls while we're
+        * here.
         */
-       ret = ocfs2_data_lock(inode, 1);
-       if (ret < 0) {
-               mlog_errno(ret);
-               goto out;
-       }
+       down_write(&oi->ip_alloc_sem);
+
+       if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+               /*
+                * We can optimize small extends by keeping the inodes
+                * inline data.
+                */
+               if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
+                       up_write(&oi->ip_alloc_sem);
+                       goto out_update_size;
+               }
+
+               ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
+               if (ret) {
+                       up_write(&oi->ip_alloc_sem);
 
-       if (clusters_to_add) {
-               ret = ocfs2_extend_allocation(inode, clusters_to_add);
-               if (ret < 0) {
                        mlog_errno(ret);
-                       goto out_unlock;
+                       goto out;
                }
        }
 
-       /*
-        * Call this even if we don't add any clusters to the tree. We
-        * still need to zero the area between the old i_size and the
-        * new i_size.
-        */
-       ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
+       if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+               ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
+
+       up_write(&oi->ip_alloc_sem);
+
        if (ret < 0) {
                mlog_errno(ret);
-               goto out_unlock;
+               goto out;
        }
 
 out_update_size:
-       if (!tail_to_skip) {
-               /* We're being called from ocfs2_setattr() which wants
-                * us to update i_size */
-               ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
-               if (ret < 0)
-                       mlog_errno(ret);
-       }
-
-out_unlock:
-       if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
-               ocfs2_data_unlock(inode, 1);
+       ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+       if (ret < 0)
+               mlog_errno(ret);
 
 out:
        return ret;
@@ -939,10 +943,17 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
        struct ocfs2_super *osb = OCFS2_SB(sb);
        struct buffer_head *bh = NULL;
        handle_t *handle = NULL;
+       int qtype;
+       struct dquot *transfer_from[MAXQUOTAS] = { };
+       struct dquot *transfer_to[MAXQUOTAS] = { };
 
        mlog_entry("(0x%p, '%.*s')\n", dentry,
                   dentry->d_name.len, dentry->d_name.name);
 
+       /* ensuring we don't even attempt to truncate a symlink */
+       if (S_ISLNK(inode->i_mode))
+               attr->ia_valid &= ~ATTR_SIZE;
+
        if (attr->ia_valid & ATTR_MODE)
                mlog(0, "mode change: %d\n", attr->ia_mode);
        if (attr->ia_valid & ATTR_UID)
@@ -974,7 +985,7 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
                }
        }
 
-       status = ocfs2_meta_lock(inode, &bh, 1);
+       status = ocfs2_inode_lock(inode, &bh, 1);
        if (status < 0) {
                if (status != -ENOENT)
                        mlog_errno(status);
@@ -982,10 +993,21 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
        }
 
        if (size_change && attr->ia_size != i_size_read(inode)) {
-               if (i_size_read(inode) > attr->ia_size)
+               if (attr->ia_size > sb->s_maxbytes) {
+                       status = -EFBIG;
+                       goto bail_unlock;
+               }
+
+               if (i_size_read(inode) > attr->ia_size) {
+                       if (ocfs2_should_order_data(inode)) {
+                               status = ocfs2_begin_ordered_truncate(inode,
+                                                                     attr->ia_size);
+                               if (status)
+                                       goto bail_unlock;
+                       }
                        status = ocfs2_truncate_file(inode, bh, attr->ia_size);
-               else
-                       status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
+               } else
+                       status = ocfs2_extend_file(inode, bh, attr->ia_size);
                if (status < 0) {
                        if (status != -ENOSPC)
                                mlog_errno(status);
@@ -994,13 +1016,63 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
                }
        }
 
-       handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-       if (IS_ERR(handle)) {
-               status = PTR_ERR(handle);
-               mlog_errno(status);
-               goto bail_unlock;
+       if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
+           (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
+               /*
+                * Gather pointers to quota structures so that allocation /
+                * freeing of quota structures happens here and not inside
+                * vfs_dq_transfer() where we have problems with lock ordering
+                */
+               if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
+                   && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+                   OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
+                       transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
+                                                     USRQUOTA);
+                       transfer_from[USRQUOTA] = dqget(sb, inode->i_uid,
+                                                       USRQUOTA);
+                       if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) {
+                               status = -ESRCH;
+                               goto bail_unlock;
+                       }
+               }
+               if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
+                   && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+                   OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
+                       transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
+                                                     GRPQUOTA);
+                       transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid,
+                                                       GRPQUOTA);
+                       if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) {
+                               status = -ESRCH;
+                               goto bail_unlock;
+                       }
+               }
+               handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
+                                          2 * ocfs2_quota_trans_credits(sb));
+               if (IS_ERR(handle)) {
+                       status = PTR_ERR(handle);
+                       mlog_errno(status);
+                       goto bail_unlock;
+               }
+               status = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
+               if (status < 0)
+                       goto bail_commit;
+       } else {
+               handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+               if (IS_ERR(handle)) {
+                       status = PTR_ERR(handle);
+                       mlog_errno(status);
+                       goto bail_unlock;
+               }
        }
 
+       /*
+        * This will intentionally not wind up calling vmtruncate(),
+        * since all the work for a size change has been done above.
+        * Otherwise, we could get into problems with truncate as
+        * ip_alloc_sem is used there to protect against i_size
+        * changes.
+        */
        status = inode_setattr(inode, attr);
        if (status < 0) {
                mlog_errno(status);
@@ -1014,13 +1086,24 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
 bail_commit:
        ocfs2_commit_trans(osb, handle);
 bail_unlock:
-       ocfs2_meta_unlock(inode, 1);
+       ocfs2_inode_unlock(inode, 1);
 bail_unlock_rw:
        if (size_change)
                ocfs2_rw_unlock(inode, 1);
 bail:
-       if (bh)
-               brelse(bh);
+       brelse(bh);
+
+       /* Release quota pointers in case we acquired them */
+       for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
+               dqput(transfer_to[qtype]);
+               dqput(transfer_from[qtype]);
+       }
+
+       if (!status && attr->ia_valid & ATTR_MODE) {
+               status = ocfs2_acl_chmod(inode);
+               if (status < 0)
+                       mlog_errno(status);
+       }
 
        mlog_exit(status);
        return status;
@@ -1055,59 +1138,52 @@ bail:
        return err;
 }
 
-int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
+int ocfs2_permission(struct inode *inode, int mask)
 {
        int ret;
 
        mlog_entry_void();
 
-       ret = ocfs2_meta_lock(inode, NULL, 0);
+       ret = ocfs2_inode_lock(inode, NULL, 0);
        if (ret) {
                if (ret != -ENOENT)
                        mlog_errno(ret);
                goto out;
        }
 
-       ret = generic_permission(inode, mask, NULL);
+       ret = generic_permission(inode, mask, ocfs2_check_acl);
 
-       ocfs2_meta_unlock(inode, 0);
+       ocfs2_inode_unlock(inode, 0);
 out:
        mlog_exit(ret);
        return ret;
 }
 
-static int ocfs2_write_remove_suid(struct inode *inode)
+static int __ocfs2_write_remove_suid(struct inode *inode,
+                                    struct buffer_head *bh)
 {
        int ret;
-       struct buffer_head *bh = NULL;
-       struct ocfs2_inode_info *oi = OCFS2_I(inode);
        handle_t *handle;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        struct ocfs2_dinode *di;
 
        mlog_entry("(Inode %llu, mode 0%o)\n",
-                  (unsigned long long)oi->ip_blkno, inode->i_mode);
+                  (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
 
        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
-       if (handle == NULL) {
-               ret = -ENOMEM;
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
                mlog_errno(ret);
                goto out;
        }
 
-       ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
+       ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret < 0) {
                mlog_errno(ret);
                goto out_trans;
        }
 
-       ret = ocfs2_journal_access(handle, inode, bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
-       if (ret < 0) {
-               mlog_errno(ret);
-               goto out_bh;
-       }
-
        inode->i_mode &= ~S_ISUID;
        if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
                inode->i_mode &= ~S_ISGID;
@@ -1118,8 +1194,7 @@ static int ocfs2_write_remove_suid(struct inode *inode)
        ret = ocfs2_journal_dirty(handle, bh);
        if (ret < 0)
                mlog_errno(ret);
-out_bh:
-       brelse(bh);
+
 out_trans:
        ocfs2_commit_trans(osb, handle);
 out:
@@ -1165,283 +1240,664 @@ out:
        return ret;
 }
 
-static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
-                                        loff_t *ppos,
-                                        size_t count,
-                                        int appending,
-                                        int *direct_io)
+static int ocfs2_write_remove_suid(struct inode *inode)
 {
-       int ret = 0, meta_level = appending;
-       struct inode *inode = dentry->d_inode;
-       u32 clusters;
-       loff_t newsize, saved_pos;
+       int ret;
+       struct buffer_head *bh = NULL;
 
-       /* 
-        * We sample i_size under a read level meta lock to see if our write
-        * is extending the file, if it is we back off and get a write level
-        * meta lock.
-        */
-       for(;;) {
-               ret = ocfs2_meta_lock(inode, NULL, meta_level);
-               if (ret < 0) {
-                       meta_level = -1;
+       ret = ocfs2_read_inode_block(inode, &bh);
+       if (ret < 0) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret =  __ocfs2_write_remove_suid(inode, bh);
+out:
+       brelse(bh);
+       return ret;
+}
+
+/*
+ * Allocate enough extents to cover the region starting at byte offset
+ * start for len bytes. Existing extents are skipped, any extents
+ * added are marked as "unwritten".
+ */
+static int ocfs2_allocate_unwritten_extents(struct inode *inode,
+                                           u64 start, u64 len)
+{
+       int ret;
+       u32 cpos, phys_cpos, clusters, alloc_size;
+       u64 end = start + len;
+       struct buffer_head *di_bh = NULL;
+
+       if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+               ret = ocfs2_read_inode_block(inode, &di_bh);
+               if (ret) {
                        mlog_errno(ret);
                        goto out;
                }
 
-               /* Clear suid / sgid if necessary. We do this here
-                * instead of later in the write path because
-                * remove_suid() calls ->setattr without any hint that
-                * we may have already done our cluster locking. Since
-                * ocfs2_setattr() *must* take cluster locks to
-                * proceeed, this will lead us to recursively lock the
-                * inode. There's also the dinode i_size state which
-                * can be lost via setattr during extending writes (we
-                * set inode->i_size at the end of a write. */
-               if (should_remove_suid(dentry)) {
-                       if (meta_level == 0) {
-                               ocfs2_meta_unlock(inode, meta_level);
-                               meta_level = 1;
-                               continue;
-                       }
+               /*
+                * Nothing to do if the requested reservation range
+                * fits within the inode.
+                */
+               if (ocfs2_size_fits_inline_data(di_bh, end))
+                       goto out;
 
-                       ret = ocfs2_write_remove_suid(inode);
-                       if (ret < 0) {
-                               mlog_errno(ret);
-                               goto out_unlock;
-                       }
+               ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
                }
+       }
 
-               /* work on a copy of ppos until we're sure that we won't have
-                * to recalculate it due to relocking. */
-               if (appending) {
-                       saved_pos = i_size_read(inode);
-                       mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
-               } else {
-                       saved_pos = *ppos;
-               }
+       /*
+        * We consider both start and len to be inclusive.
+        */
+       cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+       clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
+       clusters -= cpos;
 
-               if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
-                       loff_t end = saved_pos + count;
+       while (clusters) {
+               ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+                                        &alloc_size, NULL);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
 
-                       /*
-                        * Skip the O_DIRECT checks if we don't need
-                        * them.
-                        */
-                       if (!direct_io || !(*direct_io))
-                               break;
+               /*
+                * Hole or existing extent len can be arbitrary, so
+                * cap it to our own allocation request.
+                */
+               if (alloc_size > clusters)
+                       alloc_size = clusters;
 
+               if (phys_cpos) {
                        /*
-                        * Allowing concurrent direct writes means
-                        * i_size changes wouldn't be synchronized, so
-                        * one node could wind up truncating another
-                        * nodes writes.
+                        * We already have an allocation at this
+                        * region so we can safely skip it.
                         */
-                       if (end > i_size_read(inode)) {
-                               *direct_io = 0;
-                               break;
-                       }
+                       goto next;
+               }
 
-                       /*
-                        * We don't fill holes during direct io, so
-                        * check for them here. If any are found, the
-                        * caller will have to retake some cluster
-                        * locks and initiate the io as buffered.
-                        */
-                       ret = ocfs2_check_range_for_holes(inode, saved_pos,
-                                                         count);
-                       if (ret == 1) {
-                               *direct_io = 0;
-                               ret = 0;
-                       } else if (ret < 0)
+               ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
+               if (ret) {
+                       if (ret != -ENOSPC)
                                mlog_errno(ret);
-                       break;
+                       goto out;
                }
 
+next:
+               cpos += alloc_size;
+               clusters -= alloc_size;
+       }
+
+       ret = 0;
+out:
+
+       brelse(di_bh);
+       return ret;
+}
+
+/*
+ * Truncate a byte range, avoiding pages within partial clusters. This
+ * preserves those pages for the zeroing code to write to.
+ */
+static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
+                                        u64 byte_len)
+{
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       loff_t start, end;
+       struct address_space *mapping = inode->i_mapping;
+
+       start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
+       end = byte_start + byte_len;
+       end = end & ~(osb->s_clustersize - 1);
+
+       if (start < end) {
+               unmap_mapping_range(mapping, start, end - start, 0);
+               truncate_inode_pages_range(mapping, start, end - 1);
+       }
+}
+
+static int ocfs2_zero_partial_clusters(struct inode *inode,
+                                      u64 start, u64 len)
+{
+       int ret = 0;
+       u64 tmpend, end = start + len;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       unsigned int csize = osb->s_clustersize;
+       handle_t *handle;
+
+       /*
+        * The "start" and "end" values are NOT necessarily part of
+        * the range whose allocation is being deleted. Rather, this
+        * is what the user passed in with the request. We must zero
+        * partial clusters here. There's no need to worry about
+        * physical allocation - the zeroing code knows to skip holes.
+        */
+       mlog(0, "byte start: %llu, end: %llu\n",
+            (unsigned long long)start, (unsigned long long)end);
+
+       /*
+        * If both edges are on a cluster boundary then there's no
+        * zeroing required as the region is part of the allocation to
+        * be truncated.
+        */
+       if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
+               goto out;
+
+       handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
+               mlog_errno(ret);
+               goto out;
+       }
+
+       /*
+        * We want to get the byte offset of the end of the 1st cluster.
+        */
+       tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
+       if (tmpend > end)
+               tmpend = end;
+
+       mlog(0, "1st range: start: %llu, tmpend: %llu\n",
+            (unsigned long long)start, (unsigned long long)tmpend);
+
+       ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
+       if (ret)
+               mlog_errno(ret);
+
+       if (tmpend < end) {
                /*
-                * The rest of this loop is concerned with legacy file
-                * systems which don't support sparse files.
+                * This may make start and end equal, but the zeroing
+                * code will skip any work in that case so there's no
+                * need to catch it up here.
                 */
+               start = end & ~(osb->s_clustersize - 1);
 
-               newsize = count + saved_pos;
+               mlog(0, "2nd range: start: %llu, end: %llu\n",
+                    (unsigned long long)start, (unsigned long long)end);
 
-               mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
-                    (long long) saved_pos, (long long) newsize,
-                    (long long) i_size_read(inode));
+               ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
+               if (ret)
+                       mlog_errno(ret);
+       }
 
-               /* No need for a higher level metadata lock if we're
-                * never going past i_size. */
-               if (newsize <= i_size_read(inode))
-                       break;
+       ocfs2_commit_trans(osb, handle);
+out:
+       return ret;
+}
+
+static int ocfs2_remove_inode_range(struct inode *inode,
+                                   struct buffer_head *di_bh, u64 byte_start,
+                                   u64 byte_len)
+{
+       int ret = 0;
+       u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       struct ocfs2_cached_dealloc_ctxt dealloc;
+       struct address_space *mapping = inode->i_mapping;
+       struct ocfs2_extent_tree et;
 
-               if (meta_level == 0) {
-                       ocfs2_meta_unlock(inode, meta_level);
-                       meta_level = 1;
-                       continue;
+       ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
+       ocfs2_init_dealloc_ctxt(&dealloc);
+
+       if (byte_len == 0)
+               return 0;
+
+       if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+               ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
+                                           byte_start + byte_len, 0);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
                }
+               /*
+                * There's no need to get fancy with the page cache
+                * truncate of an inline-data inode. We're talking
+                * about less than a page here, which will be cached
+                * in the dinode buffer anyway.
+                */
+               unmap_mapping_range(mapping, 0, 0, 0);
+               truncate_inode_pages(mapping, 0);
+               goto out;
+       }
 
-               spin_lock(&OCFS2_I(inode)->ip_lock);
-               clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
-                       OCFS2_I(inode)->ip_clusters;
-               spin_unlock(&OCFS2_I(inode)->ip_lock);
+       trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
+       trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
+       if (trunc_len >= trunc_start)
+               trunc_len -= trunc_start;
+       else
+               trunc_len = 0;
 
-               mlog(0, "Writing at EOF, may need more allocation: "
-                    "i_size = %lld, newsize = %lld, need %u clusters\n",
-                    (long long) i_size_read(inode), (long long) newsize,
-                    clusters);
+       mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
+            (unsigned long long)OCFS2_I(inode)->ip_blkno,
+            (unsigned long long)byte_start,
+            (unsigned long long)byte_len, trunc_start, trunc_len);
 
-               /* We only want to continue the rest of this loop if
-                * our extend will actually require more
-                * allocation. */
-               if (!clusters)
-                       break;
+       ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
 
-               ret = ocfs2_extend_file(inode, NULL, newsize, count);
-               if (ret < 0) {
-                       if (ret != -ENOSPC)
+       cpos = trunc_start;
+       while (trunc_len) {
+               ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+                                        &alloc_size, NULL);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               if (alloc_size > trunc_len)
+                       alloc_size = trunc_len;
+
+               /* Only do work for non-holes */
+               if (phys_cpos != 0) {
+                       ret = ocfs2_remove_btree_range(inode, &et, cpos,
+                                                      phys_cpos, alloc_size,
+                                                      &dealloc);
+                       if (ret) {
                                mlog_errno(ret);
-                       goto out_unlock;
+                               goto out;
+                       }
                }
+
+               cpos += alloc_size;
+               trunc_len -= alloc_size;
+       }
+
+       ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
+
+out:
+       ocfs2_schedule_truncate_log_flush(osb, 1);
+       ocfs2_run_deallocs(osb, &dealloc);
+
+       return ret;
+}
+
+/*
+ * Parts of this function taken from xfs_change_file_space()
+ */
+static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
+                                    loff_t f_pos, unsigned int cmd,
+                                    struct ocfs2_space_resv *sr,
+                                    int change_size)
+{
+       int ret;
+       s64 llen;
+       loff_t size;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       struct buffer_head *di_bh = NULL;
+       handle_t *handle;
+       unsigned long long max_off = inode->i_sb->s_maxbytes;
+
+       if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+               return -EROFS;
+
+       mutex_lock(&inode->i_mutex);
+
+       /*
+        * This prevents concurrent writes on other nodes
+        */
+       ret = ocfs2_rw_lock(inode, 1);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_inode_lock(inode, &di_bh, 1);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_rw_unlock;
+       }
+
+       if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
+               ret = -EPERM;
+               goto out_inode_unlock;
+       }
+
+       switch (sr->l_whence) {
+       case 0: /*SEEK_SET*/
+               break;
+       case 1: /*SEEK_CUR*/
+               sr->l_start += f_pos;
+               break;
+       case 2: /*SEEK_END*/
+               sr->l_start += i_size_read(inode);
                break;
+       default:
+               ret = -EINVAL;
+               goto out_inode_unlock;
        }
+       sr->l_whence = 0;
 
-       if (appending)
-               *ppos = saved_pos;
+       llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
 
-out_unlock:
-       ocfs2_meta_unlock(inode, meta_level);
+       if (sr->l_start < 0
+           || sr->l_start > max_off
+           || (sr->l_start + llen) < 0
+           || (sr->l_start + llen) > max_off) {
+               ret = -EINVAL;
+               goto out_inode_unlock;
+       }
+       size = sr->l_start + sr->l_len;
+
+       if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
+               if (sr->l_len <= 0) {
+                       ret = -EINVAL;
+                       goto out_inode_unlock;
+               }
+       }
+
+       if (file && should_remove_suid(file->f_path.dentry)) {
+               ret = __ocfs2_write_remove_suid(inode, di_bh);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out_inode_unlock;
+               }
+       }
+
+       down_write(&OCFS2_I(inode)->ip_alloc_sem);
+       switch (cmd) {
+       case OCFS2_IOC_RESVSP:
+       case OCFS2_IOC_RESVSP64:
+               /*
+                * This takes unsigned offsets, but the signed ones we
+                * pass have been checked against overflow above.
+                */
+               ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
+                                                      sr->l_len);
+               break;
+       case OCFS2_IOC_UNRESVSP:
+       case OCFS2_IOC_UNRESVSP64:
+               ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
+                                              sr->l_len);
+               break;
+       default:
+               ret = -EINVAL;
+       }
+       up_write(&OCFS2_I(inode)->ip_alloc_sem);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_inode_unlock;
+       }
+
+       /*
+        * We update c/mtime for these changes
+        */
+       handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+       if (IS_ERR(handle)) {
+               ret = PTR_ERR(handle);
+               mlog_errno(ret);
+               goto out_inode_unlock;
+       }
+
+       if (change_size && i_size_read(inode) < size)
+               i_size_write(inode, size);
+
+       inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+       ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
+       if (ret < 0)
+               mlog_errno(ret);
+
+       ocfs2_commit_trans(osb, handle);
+
+out_inode_unlock:
+       brelse(di_bh);
+       ocfs2_inode_unlock(inode, 1);
+out_rw_unlock:
+       ocfs2_rw_unlock(inode, 1);
 
 out:
+       mutex_unlock(&inode->i_mutex);
        return ret;
 }
 
-static inline void
-ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
+int ocfs2_change_file_space(struct file *file, unsigned int cmd,
+                           struct ocfs2_space_resv *sr)
 {
-       const struct iovec *iov = *iovp;
-       size_t base = *basep;
+       struct inode *inode = file->f_path.dentry->d_inode;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
-       do {
-               int copy = min(bytes, iov->iov_len - base);
+       if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
+           !ocfs2_writes_unwritten_extents(osb))
+               return -ENOTTY;
+       else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
+                !ocfs2_sparse_alloc(osb))
+               return -ENOTTY;
 
-               bytes -= copy;
-               base += copy;
-               if (iov->iov_len == base) {
-                       iov++;
-                       base = 0;
-               }
-       } while (bytes);
-       *iovp = iov;
-       *basep = base;
+       if (!S_ISREG(inode->i_mode))
+               return -EINVAL;
+
+       if (!(file->f_mode & FMODE_WRITE))
+               return -EBADF;
+
+       return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
 }
 
-static struct page * ocfs2_get_write_source(char **ret_src_buf,
-                                           const struct iovec *cur_iov,
-                                           size_t iov_offset)
+static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
+                           loff_t len)
 {
-       int ret;
-       char *buf = cur_iov->iov_base + iov_offset;
-       struct page *src_page = NULL;
-       unsigned long off;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       struct ocfs2_space_resv sr;
+       int change_size = 1;
 
-       off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
+       if (!ocfs2_writes_unwritten_extents(osb))
+               return -EOPNOTSUPP;
 
-       if (!segment_eq(get_fs(), KERNEL_DS)) {
-               /*
-                * Pull in the user page. We want to do this outside
-                * of the meta data locks in order to preserve locking
-                * order in case of page fault.
-                */
-               ret = get_user_pages(current, current->mm,
-                                    (unsigned long)buf & PAGE_CACHE_MASK, 1,
-                                    0, 0, &src_page, NULL);
-               if (ret == 1)
-                       *ret_src_buf = kmap(src_page) + off;
-               else
-                       src_page = ERR_PTR(-EFAULT);
-       } else {
-               *ret_src_buf = buf;
-       }
+       if (S_ISDIR(inode->i_mode))
+               return -ENODEV;
 
-       return src_page;
+       if (mode & FALLOC_FL_KEEP_SIZE)
+               change_size = 0;
+
+       sr.l_whence = 0;
+       sr.l_start = (s64)offset;
+       sr.l_len = (s64)len;
+
+       return __ocfs2_change_file_space(NULL, inode, offset,
+                                        OCFS2_IOC_RESVSP64, &sr, change_size);
+}
+
+int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
+                                  size_t count)
+{
+       int ret = 0;
+       unsigned int extent_flags;
+       u32 cpos, clusters, extent_len, phys_cpos;
+       struct super_block *sb = inode->i_sb;
+
+       if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
+           !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
+           OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+               return 0;
+
+       cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
+       clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
+
+       while (clusters) {
+               ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
+                                        &extent_flags);
+               if (ret < 0) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
+                       ret = 1;
+                       break;
+               }
+
+               if (extent_len > clusters)
+                       extent_len = clusters;
+
+               clusters -= extent_len;
+               cpos += extent_len;
+       }
+out:
+       return ret;
 }
 
-static void ocfs2_put_write_source(struct page *page)
+static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
+                                           loff_t pos, size_t count,
+                                           int *meta_level)
 {
-       if (page) {
-               kunmap(page);
-               page_cache_release(page);
+       int ret;
+       struct buffer_head *di_bh = NULL;
+       u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+       u32 clusters =
+               ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
+
+       ret = ocfs2_inode_lock(inode, &di_bh, 1);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
        }
+
+       *meta_level = 1;
+
+       ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
+       if (ret)
+               mlog_errno(ret);
+out:
+       brelse(di_bh);
+       return ret;
 }
 
-static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
-                                        const struct iovec *iov,
-                                        unsigned long nr_segs,
+static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
+                                        loff_t *ppos,
                                         size_t count,
-                                        ssize_t o_direct_written)
+                                        int appending,
+                                        int *direct_io,
+                                        int *has_refcount)
 {
-       int ret = 0;
-       ssize_t copied, total = 0;
-       size_t iov_offset = 0, bytes;
-       loff_t pos;
-       const struct iovec *cur_iov = iov;
-       struct page *user_page, *page;
-       char *buf, *dst;
-       void *fsdata;
+       int ret = 0, meta_level = 0;
+       struct inode *inode = dentry->d_inode;
+       loff_t saved_pos, end;
 
        /*
-        * handle partial DIO write.  Adjust cur_iov if needed.
+        * We start with a read level meta lock and only jump to an ex
+        * if we need to make modifications here.
         */
-       ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
+       for(;;) {
+               ret = ocfs2_inode_lock(inode, NULL, meta_level);
+               if (ret < 0) {
+                       meta_level = -1;
+                       mlog_errno(ret);
+                       goto out;
+               }
 
-       do {
-               pos = *ppos;
+               /* Clear suid / sgid if necessary. We do this here
+                * instead of later in the write path because
+                * remove_suid() calls ->setattr without any hint that
+                * we may have already done our cluster locking. Since
+                * ocfs2_setattr() *must* take cluster locks to
+                * proceeed, this will lead us to recursively lock the
+                * inode. There's also the dinode i_size state which
+                * can be lost via setattr during extending writes (we
+                * set inode->i_size at the end of a write. */
+               if (should_remove_suid(dentry)) {
+                       if (meta_level == 0) {
+                               ocfs2_inode_unlock(inode, meta_level);
+                               meta_level = 1;
+                               continue;
+                       }
 
-               user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
-               if (IS_ERR(user_page)) {
-                       ret = PTR_ERR(user_page);
-                       goto out;
+                       ret = ocfs2_write_remove_suid(inode);
+                       if (ret < 0) {
+                               mlog_errno(ret);
+                               goto out_unlock;
+                       }
                }
 
-               /* Stay within our page boundaries */
-               bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
-                           (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
-               /* Stay within the vector boundary */
-               bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
-               /* Stay within count */
-               bytes = min(bytes, count);
-
-               page = NULL;
-               ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
-                                       &page, &fsdata);
-               if (ret) {
+               /* work on a copy of ppos until we're sure that we won't have
+                * to recalculate it due to relocking. */
+               if (appending) {
+                       saved_pos = i_size_read(inode);
+                       mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
+               } else {
+                       saved_pos = *ppos;
+               }
+
+               end = saved_pos + count;
+
+               ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
+               if (ret == 1) {
+                       ocfs2_inode_unlock(inode, meta_level);
+                       meta_level = -1;
+
+                       ret = ocfs2_prepare_inode_for_refcount(inode,
+                                                              saved_pos,
+                                                              count,
+                                                              &meta_level);
+                       if (has_refcount)
+                               *has_refcount = 1;
+               }
+
+               if (ret < 0) {
                        mlog_errno(ret);
-                       goto out;
+                       goto out_unlock;
                }
 
-               dst = kmap_atomic(page, KM_USER0);
-               memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes);
-               kunmap_atomic(dst, KM_USER0);
-               flush_dcache_page(page);
-               ocfs2_put_write_source(user_page);
-
-               copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
-                                        bytes, page, fsdata);
-               if (copied < 0) {
-                       mlog_errno(copied);
-                       ret = copied;
-                       goto out;
+               /*
+                * Skip the O_DIRECT checks if we don't need
+                * them.
+                */
+               if (!direct_io || !(*direct_io))
+                       break;
+
+               /*
+                * There's no sane way to do direct writes to an inode
+                * with inline data.
+                */
+               if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+                       *direct_io = 0;
+                       break;
                }
 
-               total += copied;
-               *ppos = pos + copied;
-               count -= copied;
+               if (has_refcount && *has_refcount == 1) {
+                       *direct_io = 0;
+                       break;
+               }
+               /*
+                * Allowing concurrent direct writes means
+                * i_size changes wouldn't be synchronized, so
+                * one node could wind up truncating another
+                * nodes writes.
+                */
+               if (end > i_size_read(inode)) {
+                       *direct_io = 0;
+                       break;
+               }
 
-               ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
-       } while(count);
+               /*
+                * We don't fill holes during direct io, so
+                * check for them here. If any are found, the
+                * caller will have to retake some cluster
+                * locks and initiate the io as buffered.
+                */
+               ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
+               if (ret == 1) {
+                       *direct_io = 0;
+                       ret = 0;
+               } else if (ret < 0)
+                       mlog_errno(ret);
+               break;
+       }
+
+       if (appending)
+               *ppos = saved_pos;
+
+out_unlock:
+       if (meta_level >= 0)
+               ocfs2_inode_unlock(inode, meta_level);
 
 out:
-       return total ? total : ret;
+       return ret;
 }
 
 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
@@ -1450,13 +1906,15 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
                                    loff_t pos)
 {
        int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
-       int can_do_direct, sync = 0;
+       int can_do_direct, has_refcount = 0;
        ssize_t written = 0;
        size_t ocount;          /* original count */
        size_t count;           /* after file limit checks */
-       loff_t *ppos = &iocb->ki_pos;
+       loff_t old_size, *ppos = &iocb->ki_pos;
+       u32 old_clusters;
        struct file *file = iocb->ki_filp;
        struct inode *inode = file->f_path.dentry->d_inode;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
        mlog_entry("(0x%p, %u, '%.*s')\n", file,
                   (unsigned int)nr_segs,
@@ -1466,12 +1924,6 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
        if (iocb->ki_left == 0)
                return 0;
 
-       ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
-       if (ret)
-               return ret;
-
-       count = ocount;
-
        vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 
        appending = file->f_flags & O_APPEND ? 1 : 0;
@@ -1497,7 +1949,7 @@ relock:
        can_do_direct = direct_io;
        ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
                                            iocb->ki_left, appending,
-                                           &can_do_direct);
+                                           &can_do_direct, &has_refcount);
        if (ret < 0) {
                mlog_errno(ret);
                goto out;
@@ -1515,55 +1967,73 @@ relock:
                rw_level = -1;
 
                direct_io = 0;
-               sync = 1;
                goto relock;
        }
 
-       if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
-               sync = 1;
-
        /*
-        * XXX: Is it ok to execute these checks a second time?
+        * To later detect whether a journal commit for sync writes is
+        * necessary, we sample i_size, and cluster count here.
         */
-       ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
-       if (ret)
-               goto out;
-
-       /*
-        * Set pos so that sync_page_range_nolock() below understands
-        * where to start from. We might've moved it around via the
-        * calls above. The range we want to actually sync starts from
-        * *ppos here.
-        *
-        */
-       pos = *ppos;
+       old_size = i_size_read(inode);
+       old_clusters = OCFS2_I(inode)->ip_clusters;
 
        /* communicate with ocfs2_dio_end_io */
        ocfs2_iocb_set_rw_locked(iocb, rw_level);
 
        if (direct_io) {
+               ret = generic_segment_checks(iov, &nr_segs, &ocount,
+                                            VERIFY_READ);
+               if (ret)
+                       goto out_dio;
+
+               count = ocount;
+               ret = generic_write_checks(file, ppos, &count,
+                                          S_ISBLK(inode->i_mode));
+               if (ret)
+                       goto out_dio;
+
                written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
                                                    ppos, count, ocount);
                if (written < 0) {
+                       /*
+                        * direct write may have instantiated a few
+                        * blocks outside i_size. Trim these off again.
+                        * Don't need i_size_read because we hold i_mutex.
+                        */
+                       if (*ppos + count > inode->i_size)
+                               vmtruncate(inode, inode->i_size);
                        ret = written;
                        goto out_dio;
                }
        } else {
-               written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
-                                                   count, written);
-               if (written < 0) {
-                       ret = written;
-                       if (ret != -EFAULT || ret != -ENOSPC)
-                               mlog_errno(ret);
-                       goto out;
-               }
+               written = __generic_file_aio_write(iocb, iov, nr_segs, ppos);
        }
 
 out_dio:
        /* buffered aio wouldn't have proper lock coverage today */
        BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
 
-       /* 
+       if ((file->f_flags & O_DSYNC && !direct_io) || IS_SYNC(inode) ||
+           (file->f_flags & O_DIRECT && has_refcount)) {
+               ret = filemap_fdatawrite_range(file->f_mapping, pos,
+                                              pos + count - 1);
+               if (ret < 0)
+                       written = ret;
+
+               if (!ret && (old_size != i_size_read(inode) ||
+                   old_clusters != OCFS2_I(inode)->ip_clusters ||
+                   has_refcount)) {
+                       ret = jbd2_journal_force_commit(osb->journal->j_journal);
+                       if (ret < 0)
+                               written = ret;
+               }
+
+               if (!ret)
+                       ret = filemap_fdatawait_range(file->f_mapping, pos,
+                                                     pos + count - 1);
+       }
+
+       /*
         * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
         * function pointer which is called when o_direct io completes so that
         * it can unlock our rw lock.  (it's the clustered equivalent of
@@ -1586,73 +2056,37 @@ out_sems:
        if (have_alloc_sem)
                up_read(&inode->i_alloc_sem);
 
-       if (written > 0 && sync) {
-               ssize_t err;
-
-               err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
-               if (err < 0)
-                       written = err;
-       }
-
        mutex_unlock(&inode->i_mutex);
 
+       if (written)
+               ret = written;
        mlog_exit(ret);
-       return written ? written : ret;
+       return ret;
 }
 
-static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
-                                   struct pipe_buffer *buf,
-                                   struct splice_desc *sd)
+static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
+                               struct file *out,
+                               struct splice_desc *sd)
 {
-       int ret, count;
-       ssize_t copied = 0;
-       struct file *file = sd->u.file;
-       unsigned int offset;
-       struct page *page = NULL;
-       void *fsdata;
-       char *src, *dst;
-
-       ret = buf->ops->confirm(pipe, buf);
-       if (ret)
-               goto out;
-
-       offset = sd->pos & ~PAGE_CACHE_MASK;
-       count = sd->len;
-       if (count + offset > PAGE_CACHE_SIZE)
-               count = PAGE_CACHE_SIZE - offset;
+       int ret;
 
-       ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
-                               &page, &fsdata);
-       if (ret) {
+       ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
+                                           sd->total_len, 0, NULL, NULL);
+       if (ret < 0) {
                mlog_errno(ret);
-               goto out;
-       }
-
-       src = buf->ops->map(pipe, buf, 1);
-       dst = kmap_atomic(page, KM_USER1);
-       memcpy(dst + offset, src + buf->offset, count);
-       kunmap_atomic(page, KM_USER1);
-       buf->ops->unmap(pipe, buf, src);
-
-       copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
-                                page, fsdata);
-       if (copied < 0) {
-               mlog_errno(copied);
-               ret = copied;
-               goto out;
+               return ret;
        }
-out:
 
-       return copied ? copied : ret;
+       return splice_from_pipe_feed(pipe, sd, pipe_to_file);
 }
 
-static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
-                                        struct file *out,
-                                        loff_t *ppos,
-                                        size_t len,
-                                        unsigned int flags)
+static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
+                                      struct file *out,
+                                      loff_t *ppos,
+                                      size_t len,
+                                      unsigned int flags)
 {
-       int ret, err;
+       int ret;
        struct address_space *mapping = out->f_mapping;
        struct inode *inode = mapping->host;
        struct splice_desc sd = {
@@ -1662,57 +2096,52 @@ static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
                .u.file = out,
        };
 
-       ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
-       if (ret > 0) {
-               *ppos += ret;
-
-               if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
-                       err = generic_osync_inode(inode, mapping,
-                                                 OSYNC_METADATA|OSYNC_DATA);
-                       if (err)
-                               ret = err;
-               }
-       }
-
-       return ret;
-}
-
-static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
-                                      struct file *out,
-                                      loff_t *ppos,
-                                      size_t len,
-                                      unsigned int flags)
-{
-       int ret;
-       struct inode *inode = out->f_path.dentry->d_inode;
-
        mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
                   (unsigned int)len,
                   out->f_path.dentry->d_name.len,
                   out->f_path.dentry->d_name.name);
 
-       inode_double_lock(inode, pipe->inode);
+       if (pipe->inode)
+               mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
 
-       ret = ocfs2_rw_lock(inode, 1);
-       if (ret < 0) {
-               mlog_errno(ret);
-               goto out;
-       }
+       splice_from_pipe_begin(&sd);
+       do {
+               ret = splice_from_pipe_next(pipe, &sd);
+               if (ret <= 0)
+                       break;
 
-       ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
-                                           NULL);
-       if (ret < 0) {
-               mlog_errno(ret);
-               goto out_unlock;
-       }
+               mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
+               ret = ocfs2_rw_lock(inode, 1);
+               if (ret < 0)
+                       mlog_errno(ret);
+               else {
+                       ret = ocfs2_splice_to_file(pipe, out, &sd);
+                       ocfs2_rw_unlock(inode, 1);
+               }
+               mutex_unlock(&inode->i_mutex);
+       } while (ret > 0);
+       splice_from_pipe_end(pipe, &sd);
 
-       /* ok, we're done with i_size and alloc work */
-       ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
+       if (pipe->inode)
+               mutex_unlock(&pipe->inode->i_mutex);
 
-out_unlock:
-       ocfs2_rw_unlock(inode, 1);
-out:
-       inode_double_unlock(inode, pipe->inode);
+       if (sd.num_spliced)
+               ret = sd.num_spliced;
+
+       if (ret > 0) {
+               unsigned long nr_pages;
+               int err;
+
+               nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+
+               err = generic_write_sync(out, *ppos, ret);
+               if (err)
+                       ret = err;
+               else
+                       *ppos += ret;
+
+               balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
+       }
 
        mlog_exit(ret);
        return ret;
@@ -1724,7 +2153,7 @@ static ssize_t ocfs2_file_splice_read(struct file *in,
                                      size_t len,
                                      unsigned int flags)
 {
-       int ret = 0;
+       int ret = 0, lock_level = 0;
        struct inode *inode = in->f_path.dentry->d_inode;
 
        mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
@@ -1735,12 +2164,12 @@ static ssize_t ocfs2_file_splice_read(struct file *in,
        /*
         * See the comment in ocfs2_file_aio_read()
         */
-       ret = ocfs2_meta_lock(inode, NULL, 0);
+       ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
        if (ret < 0) {
                mlog_errno(ret);
                goto bail;
        }
-       ocfs2_meta_unlock(inode, 0);
+       ocfs2_inode_unlock(inode, lock_level);
 
        ret = generic_file_splice_read(in, ppos, pipe, len, flags);
 
@@ -1769,7 +2198,7 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
                goto bail;
        }
 
-       /* 
+       /*
         * buffered reads protect themselves in ->readpage().  O_DIRECT reads
         * need locks to protect pending reads from racing with truncate.
         */
@@ -1791,21 +2220,21 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
         * We're fine letting folks race truncates and extending
         * writes with read across the cluster, just like they can
         * locally. Hence no rw_lock during read.
-        * 
+        *
         * Take and drop the meta data lock to update inode fields
         * like i_size. This allows the checks down below
-        * generic_file_aio_read() a chance of actually working. 
+        * generic_file_aio_read() a chance of actually working.
         */
-       ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
+       ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
        if (ret < 0) {
                mlog_errno(ret);
                goto bail;
        }
-       ocfs2_meta_unlock(inode, lock_level);
+       ocfs2_inode_unlock(inode, lock_level);
 
        ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
        if (ret == -EINVAL)
-               mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
+               mlog(0, "generic_file_aio_read returned -EINVAL\n");
 
        /* buffered aio wouldn't have proper lock coverage today */
        BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
@@ -1819,7 +2248,7 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
 bail:
        if (have_alloc_sem)
                up_read(&inode->i_alloc_sem);
-       if (rw_level != -1) 
+       if (rw_level != -1)
                ocfs2_rw_unlock(inode, rw_level);
        mlog_exit(ret);
 
@@ -1830,6 +2259,12 @@ const struct inode_operations ocfs2_file_iops = {
        .setattr        = ocfs2_setattr,
        .getattr        = ocfs2_getattr,
        .permission     = ocfs2_permission,
+       .setxattr       = generic_setxattr,
+       .getxattr       = generic_getxattr,
+       .listxattr      = ocfs2_listxattr,
+       .removexattr    = generic_removexattr,
+       .fallocate      = ocfs2_fallocate,
+       .fiemap         = ocfs2_fiemap,
 };
 
 const struct inode_operations ocfs2_special_file_iops = {
@@ -1838,7 +2273,12 @@ const struct inode_operations ocfs2_special_file_iops = {
        .permission     = ocfs2_permission,
 };
 
+/*
+ * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
+ * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
+ */
 const struct file_operations ocfs2_fops = {
+       .llseek         = generic_file_llseek,
        .read           = do_sync_read,
        .write          = do_sync_write,
        .mmap           = ocfs2_mmap,
@@ -1847,20 +2287,72 @@ const struct file_operations ocfs2_fops = {
        .open           = ocfs2_file_open,
        .aio_read       = ocfs2_file_aio_read,
        .aio_write      = ocfs2_file_aio_write,
-       .ioctl          = ocfs2_ioctl,
+       .unlocked_ioctl = ocfs2_ioctl,
 #ifdef CONFIG_COMPAT
        .compat_ioctl   = ocfs2_compat_ioctl,
 #endif
+       .lock           = ocfs2_lock,
+       .flock          = ocfs2_flock,
        .splice_read    = ocfs2_file_splice_read,
        .splice_write   = ocfs2_file_splice_write,
 };
 
 const struct file_operations ocfs2_dops = {
+       .llseek         = generic_file_llseek,
+       .read           = generic_read_dir,
+       .readdir        = ocfs2_readdir,
+       .fsync          = ocfs2_sync_file,
+       .release        = ocfs2_dir_release,
+       .open           = ocfs2_dir_open,
+       .unlocked_ioctl = ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+       .lock           = ocfs2_lock,
+       .flock          = ocfs2_flock,
+};
+
+/*
+ * POSIX-lockless variants of our file_operations.
+ *
+ * These will be used if the underlying cluster stack does not support
+ * posix file locking, if the user passes the "localflocks" mount
+ * option, or if we have a local-only fs.
+ *
+ * ocfs2_flock is in here because all stacks handle UNIX file locks,
+ * so we still want it in the case of no stack support for
+ * plocks. Internally, it will do the right thing when asked to ignore
+ * the cluster.
+ */
+const struct file_operations ocfs2_fops_no_plocks = {
+       .llseek         = generic_file_llseek,
+       .read           = do_sync_read,
+       .write          = do_sync_write,
+       .mmap           = ocfs2_mmap,
+       .fsync          = ocfs2_sync_file,
+       .release        = ocfs2_file_release,
+       .open           = ocfs2_file_open,
+       .aio_read       = ocfs2_file_aio_read,
+       .aio_write      = ocfs2_file_aio_write,
+       .unlocked_ioctl = ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+       .flock          = ocfs2_flock,
+       .splice_read    = ocfs2_file_splice_read,
+       .splice_write   = ocfs2_file_splice_write,
+};
+
+const struct file_operations ocfs2_dops_no_plocks = {
+       .llseek         = generic_file_llseek,
        .read           = generic_read_dir,
        .readdir        = ocfs2_readdir,
        .fsync          = ocfs2_sync_file,
-       .ioctl          = ocfs2_ioctl,
+       .release        = ocfs2_dir_release,
+       .open           = ocfs2_dir_open,
+       .unlocked_ioctl = ocfs2_ioctl,
 #ifdef CONFIG_COMPAT
        .compat_ioctl   = ocfs2_compat_ioctl,
 #endif
+       .flock          = ocfs2_flock,
 };