From patchwork Mon Nov 19 19:31:32 2012
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: make offline logfs image and some BUG()
Date: Mon, 19 Nov 2012 09:31:32 -0000
From: tristan.lelong@blunderer.org
X-Patchwork-Id: 200148
Message-Id: <50AA8914.2090009@blunderer.org>
To: linux-mtd@lists.infradead.org

Hi,

I am trying to use logfs for a project I am currently working on. The 
kernel version I used is 2.6.37 with some patch from TI DM8168.

I search on the internet for an offline way to create a logfs image from 
a directory (like mkfs.jffs2), and the only way seems to be using the 
block2mtd kernel module. Does anybody confirm that?

Also I encountered two BUG():
- The first is already documented on some mailing lists: it is the one 
that it appears when doing umount -> call to map_invalidatepage.
I was able to apply the attached patch1 to fix this, but I still see 
call to this function. I guess I might miss something.

- The other one is when mounting a filesystem that detects a previous 
incomplete write: it BUG() because it cannot find any free segment. 
Trying to understand what is happening, I realized that during the 
mount, the function that adds segment to the free list is called after 
the function that check the filesystem. I modified this like in the 
following patch2 and it seems to work fine.

Does anybody have any thoughts about that?

Thanks a lo

---
618FE3EF

Index: super.c
===================================================================
--- super.c	(revision 24035)
+++ super.c	(working copy)
@@ -308,14 +308,13 @@
 	if (err)
 		return err;
 
+	/* Do one GC pass before any data gets dirtied */
+	logfs_gc_pass(sb);
+
 	/* Check areas for trailing unaccounted data */
 	err = logfs_check_areas(sb);
 	if (err)
 		return err;
-
-	/* Do one GC pass before any data gets dirtied */
-	logfs_gc_pass(sb);
-
 	/* after all initializations are done, replay the journal
 	 * for rw-mounts, if necessary */
 	err = logfs_replay_journal(sb);

Index: super.c
===================================================================
--- super.c	(revision 24035)
+++ super.c	(working copy)
@@ -511,6 +510,7 @@
 	/* Alias entries slow down mount, so evict as many as possible */
 	sync_filesystem(sb);
 	logfs_write_anchor(sb);
+	free_areas(sb);
 
 	/*
 	 * From this point on alias entries are simply dropped - and any
Index: segment.c
===================================================================
--- segment.c	(revision 24035)
+++ segment.c	(working copy)
@@ -841,6 +841,16 @@
 	kfree(area);
 }
 
+void free_areas(struct super_block *sb)
+{
+	struct logfs_super *super = logfs_super(sb);
+	int i;
+
+	for_each_area(i)
+		free_area(super->s_area[i]);
+	free_area(super->s_journal_area);
+}
+
 static struct logfs_area *alloc_area(struct super_block *sb)
 {
 	struct logfs_area *area;
@@ -923,10 +933,6 @@
 void logfs_cleanup_areas(struct super_block *sb)
 {
 	struct logfs_super *super = logfs_super(sb);
-	int i;
 
 	btree_grim_visitor128(&super->s_object_alias_tree, 0, kill_alias);
-	for_each_area(i)
-		free_area(super->s_area[i]);
-	free_area(super->s_journal_area);
 }
Index: logfs.h
===================================================================
--- logfs.h	(revision 24035)
+++ logfs.h	(working copy)
@@ -594,6 +594,7 @@
 void logfs_sync_area(struct logfs_area *area);
 void logfs_sync_segments(struct super_block *sb);
 void freeseg(struct super_block *sb, u32 segno);
+void free_areas(struct super_block *sb);
 
 /* area handling */
 int logfs_init_areas(struct super_block *sb);