Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[pandora-kernel.git] / fs / ext4 / inode.c
index 7a83c27..bdbe699 100644 (file)
@@ -53,6 +53,7 @@
 static inline int ext4_begin_ordered_truncate(struct inode *inode,
                                              loff_t new_size)
 {
+       trace_ext4_begin_ordered_truncate(inode, new_size);
        return jbd2_journal_begin_ordered_truncate(
                                        EXT4_SB(inode->i_sb)->s_journal,
                                        &EXT4_I(inode)->jinode,
@@ -178,6 +179,7 @@ void ext4_evict_inode(struct inode *inode)
        handle_t *handle;
        int err;
 
+       trace_ext4_evict_inode(inode);
        if (inode->i_nlink) {
                truncate_inode_pages(&inode->i_data, 0);
                goto no_delete;
@@ -1551,10 +1553,10 @@ static int do_journal_get_write_access(handle_t *handle,
        if (!buffer_mapped(bh) || buffer_freed(bh))
                return 0;
        /*
-        * __block_prepare_write() could have dirtied some buffers. Clean
+        * __block_write_begin() could have dirtied some buffers. Clean
         * the dirty bit as jbd2_journal_get_write_access() could complain
         * otherwise about fs integrity issues. Setting of the dirty bit
-        * by __block_prepare_write() isn't a real problem here as we clear
+        * by __block_write_begin() isn't a real problem here as we clear
         * the bit before releasing a page lock and thus writeback cannot
         * ever write the buffer.
         */
@@ -2065,11 +2067,11 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd,
                        /*
                         * If the page does not have buffers (for
                         * whatever reason), try to create them using
-                        * block_prepare_write.  If this fails,
+                        * __block_write_begin.  If this fails,
                         * redirty the page and move on.
                         */
                        if (!page_has_buffers(page)) {
-                               if (block_prepare_write(page, 0, len,
+                               if (__block_write_begin(page, 0, len,
                                                noalloc_get_block_write)) {
                                redirty_page:
                                        redirty_page_for_writepage(mpd->wbc,
@@ -2550,8 +2552,7 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
                if (buffer_delay(bh))
                        return 0; /* Not sure this could or should happen */
                /*
-                * XXX: __block_prepare_write() unmaps passed block,
-                * is it OK?
+                * XXX: __block_write_begin() unmaps passed block, is it OK?
                 */
                ret = ext4_da_reserve_space(inode, iblock);
                if (ret)
@@ -2583,7 +2584,7 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
 /*
  * This function is used as a standard get_block_t calback function
  * when there is no desire to allocate any blocks.  It is used as a
- * callback function for block_prepare_write() and block_write_full_page().
+ * callback function for block_write_begin() and block_write_full_page().
  * These functions should only try to map a single block at a time.
  *
  * Since this function doesn't do block allocations even if the caller
@@ -2716,11 +2717,11 @@ static int ext4_writepage(struct page *page,
 
        /*
         * If the page does not have buffers (for whatever reason),
-        * try to create them using block_prepare_write.  If this
+        * try to create them using __block_write_begin.  If this
         * fails, redirty the page and move on.
         */
-       if (!page_buffers(page)) {
-               if (block_prepare_write(page, 0, len,
+       if (!page_has_buffers(page)) {
+               if (__block_write_begin(page, 0, len,
                                        noalloc_get_block_write)) {
                redirty_page:
                        redirty_page_for_writepage(wbc, page);
@@ -2733,12 +2734,10 @@ static int ext4_writepage(struct page *page,
        if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
                              ext4_bh_delay_or_unwritten)) {
                /*
-                * We don't want to do block allocation So redirty the
-                * page and return We may reach here when we do a
-                * journal commit via
-                * journal_submit_inode_data_buffers.  If we don't
-                * have mapping block we just ignore them. We can also
-                * reach here via shrink_page_list
+                * We don't want to do block allocation, so redirty
+                * the page and return.  We may reach here when we do
+                * a journal commit via journal_submit_inode_data_buffers.
+                * We can also reach here via shrink_page_list
                 */
                goto redirty_page;
        }
@@ -3626,89 +3625,6 @@ static int ext4_get_block_write(struct inode *inode, sector_t iblock,
                               EXT4_GET_BLOCKS_IO_CREATE_EXT);
 }
 
-static void dump_completed_IO(struct inode * inode)
-{
-#ifdef EXT4_DEBUG
-       struct list_head *cur, *before, *after;
-       ext4_io_end_t *io, *io0, *io1;
-       unsigned long flags;
-
-       if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
-               ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
-               return;
-       }
-
-       ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
-       spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
-       list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
-               cur = &io->list;
-               before = cur->prev;
-               io0 = container_of(before, ext4_io_end_t, list);
-               after = cur->next;
-               io1 = container_of(after, ext4_io_end_t, list);
-
-               ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
-                           io, inode->i_ino, io0, io1);
-       }
-       spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
-#endif
-}
-
-/*
- * This function is called from ext4_sync_file().
- *
- * When IO is completed, the work to convert unwritten extents to
- * written is queued on workqueue but may not get immediately
- * scheduled. When fsync is called, we need to ensure the
- * conversion is complete before fsync returns.
- * The inode keeps track of a list of pending/completed IO that
- * might needs to do the conversion. This function walks through
- * the list and convert the related unwritten extents for completed IO
- * to written.
- * The function return the number of pending IOs on success.
- */
-int flush_completed_IO(struct inode *inode)
-{
-       ext4_io_end_t *io;
-       struct ext4_inode_info *ei = EXT4_I(inode);
-       unsigned long flags;
-       int ret = 0;
-       int ret2 = 0;
-
-       if (list_empty(&ei->i_completed_io_list))
-               return ret;
-
-       dump_completed_IO(inode);
-       spin_lock_irqsave(&ei->i_completed_io_lock, flags);
-       while (!list_empty(&ei->i_completed_io_list)){
-               io = list_entry(ei->i_completed_io_list.next,
-                               ext4_io_end_t, list);
-               /*
-                * Calling ext4_end_io_nolock() to convert completed
-                * IO to written.
-                *
-                * When ext4_sync_file() is called, run_queue() may already
-                * about to flush the work corresponding to this io structure.
-                * It will be upset if it founds the io structure related
-                * to the work-to-be schedule is freed.
-                *
-                * Thus we need to keep the io structure still valid here after
-                * convertion finished. The io structure has a flag to
-                * avoid double converting from both fsync and background work
-                * queue work.
-                */
-               spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
-               ret = ext4_end_io_nolock(io);
-               spin_lock_irqsave(&ei->i_completed_io_lock, flags);
-               if (ret < 0)
-                       ret2 = ret;
-               else
-                       list_del_init(&io->list);
-       }
-       spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
-       return (ret2 < 0) ? ret2 : 0;
-}
-
 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
                            ssize_t size, void *private, int ret,
                            bool is_async)
@@ -5364,6 +5280,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
 {
        struct inode *inode = dentry->d_inode;
        int error, rc = 0;
+       int orphan = 0;
        const unsigned int ia_valid = attr->ia_valid;
 
        error = inode_change_ok(inode, attr);
@@ -5419,8 +5336,10 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
                        error = PTR_ERR(handle);
                        goto err_out;
                }
-
-               error = ext4_orphan_add(handle, inode);
+               if (ext4_handle_valid(handle)) {
+                       error = ext4_orphan_add(handle, inode);
+                       orphan = 1;
+               }
                EXT4_I(inode)->i_disksize = attr->ia_size;
                rc = ext4_mark_inode_dirty(handle, inode);
                if (!error)
@@ -5438,6 +5357,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
                                        goto err_out;
                                }
                                ext4_orphan_del(handle, inode);
+                               orphan = 0;
                                ext4_journal_stop(handle);
                                goto err_out;
                        }
@@ -5460,7 +5380,7 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
         * If the call to ext4_truncate failed to get a transaction handle at
         * all, we need to clean up the in-core orphan list manually.
         */
-       if (inode->i_nlink)
+       if (orphan && inode->i_nlink)
                ext4_orphan_del(NULL, inode);
 
        if (!rc && (ia_valid & ATTR_MODE))
@@ -5492,9 +5412,7 @@ int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
         * will return the blocks that include the delayed allocation
         * blocks for this file.
         */
-       spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
        delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
-       spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 
        stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
        return 0;
@@ -5731,6 +5649,7 @@ int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
        int err, ret;
 
        might_sleep();
+       trace_ext4_mark_inode_dirty(inode, _RET_IP_);
        err = ext4_reserve_inode_write(handle, inode, &iloc);
        if (ext4_handle_valid(handle) &&
            EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&