Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / ext3 / fsync.c
diff --git a/kernel/fs/ext3/fsync.c b/kernel/fs/ext3/fsync.c
new file mode 100644 (file)
index 0000000..1cb9c7e
--- /dev/null
@@ -0,0 +1,109 @@
+/*
+ *  linux/fs/ext3/fsync.c
+ *
+ *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
+ *  from
+ *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
+ *                      Laboratoire MASI - Institut Blaise Pascal
+ *                      Universite Pierre et Marie Curie (Paris VI)
+ *  from
+ *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
+ *
+ *  ext3fs fsync primitive
+ *
+ *  Big-endian to little-endian byte-swapping/bitmaps by
+ *        David S. Miller (davem@caip.rutgers.edu), 1995
+ *
+ *  Removed unnecessary code duplication for little endian machines
+ *  and excessive __inline__s.
+ *        Andi Kleen, 1997
+ *
+ * Major simplications and cleanup - we only need to do the metadata, because
+ * we can depend on generic_block_fdatasync() to sync the data blocks.
+ */
+
+#include <linux/blkdev.h>
+#include <linux/writeback.h>
+#include "ext3.h"
+
+/*
+ * akpm: A new design for ext3_sync_file().
+ *
+ * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
+ * There cannot be a transaction open by this task.
+ * Another task could have dirtied this inode.  Its data can be in any
+ * state in the journalling system.
+ *
+ * What we do is just kick off a commit and wait on it.  This will snapshot the
+ * inode to disk.
+ */
+
+int ext3_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
+{
+       struct inode *inode = file->f_mapping->host;
+       struct ext3_inode_info *ei = EXT3_I(inode);
+       journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
+       int ret, needs_barrier = 0;
+       tid_t commit_tid;
+
+       trace_ext3_sync_file_enter(file, datasync);
+
+       if (inode->i_sb->s_flags & MS_RDONLY) {
+               /* Make sure that we read updated state */
+               smp_rmb();
+               if (EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS)
+                       return -EROFS;
+               return 0;
+       }
+       ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
+       if (ret)
+               goto out;
+
+       J_ASSERT(ext3_journal_current_handle() == NULL);
+
+       /*
+        * data=writeback,ordered:
+        *  The caller's filemap_fdatawrite()/wait will sync the data.
+        *  Metadata is in the journal, we wait for a proper transaction
+        *  to commit here.
+        *
+        * data=journal:
+        *  filemap_fdatawrite won't do anything (the buffers are clean).
+        *  ext3_force_commit will write the file data into the journal and
+        *  will wait on that.
+        *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
+        *  (they were dirtied by commit).  But that's OK - the blocks are
+        *  safe in-journal, which is all fsync() needs to ensure.
+        */
+       if (ext3_should_journal_data(inode)) {
+               ret = ext3_force_commit(inode->i_sb);
+               goto out;
+       }
+
+       if (datasync)
+               commit_tid = atomic_read(&ei->i_datasync_tid);
+       else
+               commit_tid = atomic_read(&ei->i_sync_tid);
+
+       if (test_opt(inode->i_sb, BARRIER) &&
+           !journal_trans_will_send_data_barrier(journal, commit_tid))
+               needs_barrier = 1;
+       log_start_commit(journal, commit_tid);
+       ret = log_wait_commit(journal, commit_tid);
+
+       /*
+        * In case we didn't commit a transaction, we have to flush
+        * disk caches manually so that data really is on persistent
+        * storage
+        */
+       if (needs_barrier) {
+               int err;
+
+               err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
+               if (!ret)
+                       ret = err;
+       }
+out:
+       trace_ext3_sync_file_exit(inode, ret);
+       return ret;
+}