Fix a problem with ext2fs so that filesystems mounted at reboot don't

keep ahold of buffers, and therefore leave filesystems dirty.  I haven't
been able to test, but the code compiles.  Those who run -current, please
test and report back!!!  (Sorry :-)).

PR:		kern/3571
Submitted by:	Dirk Keunecke <dk@panda.rhein-main.de>

8 files changed, 76 insertions, 38 deletions

diff --git a/sys/gnu/ext2fs/ext2_linux_balloc.c b/sys/gnu/ext2fs/ext2_linux_balloc.c
index e0c955a..aa17650 100644
--- a/sys/gnu/ext2fs/ext2_linux_balloc.c
+++ b/sys/gnu/ext2fs/ext2_linux_balloc.c
@@ -69,6 +69,7 @@ static void read_block_bitmap (struct mount * mp,
 			    block_group, (unsigned long) gdp->bg_block_bitmap);
 	sb->s_block_bitmap_number[bitmap_nr] = block_group;
 	sb->s_block_bitmap[bitmap_nr] = bh;
+	LCK_BUF(bh)
 }
 
 /*
@@ -129,7 +130,8 @@ static int load__block_bitmap (struct mount * mp,
 		if (sb->s_loaded_block_bitmaps < EXT2_MAX_GROUP_LOADED)
 			sb->s_loaded_block_bitmaps++;
 		else
-			brelse (sb->s_block_bitmap[EXT2_MAX_GROUP_LOADED - 1]);
+			ULCK_BUF(sb->s_block_bitmap[EXT2_MAX_GROUP_LOADED - 1])
+
 		for (j = sb->s_loaded_block_bitmaps - 1; j > 0;  j--) {
 			sb->s_block_bitmap_number[j] =
 				sb->s_block_bitmap_number[j - 1];
diff --git a/sys/gnu/ext2fs/ext2_linux_ialloc.c b/sys/gnu/ext2fs/ext2_linux_ialloc.c
index c93c20e..9f2a967 100644
--- a/sys/gnu/ext2fs/ext2_linux_ialloc.c
+++ b/sys/gnu/ext2fs/ext2_linux_ialloc.c
@@ -127,6 +127,7 @@ static void read_inode_bitmap (struct mount * mp,
 			    block_group, (unsigned long) gdp->bg_inode_bitmap);
 	sb->s_inode_bitmap_number[bitmap_nr] = block_group;
 	sb->s_inode_bitmap[bitmap_nr] = bh;
+	LCK_BUF(bh)
 }
 
 /*
@@ -190,7 +191,7 @@ static int load_inode_bitmap (struct mount * mp,
 		if (sb->s_loaded_inode_bitmaps < EXT2_MAX_GROUP_LOADED)
 			sb->s_loaded_inode_bitmaps++;
 		else
-			brelse (sb->s_inode_bitmap[EXT2_MAX_GROUP_LOADED - 1]);
+			ULCK_BUF(sb->s_inode_bitmap[EXT2_MAX_GROUP_LOADED - 1])
 		for (j = sb->s_loaded_inode_bitmaps - 1; j > 0; j--) {
 			sb->s_inode_bitmap_number[j] =
 				sb->s_inode_bitmap_number[j - 1];
diff --git a/sys/gnu/ext2fs/ext2_vfsops.c b/sys/gnu/ext2fs/ext2_vfsops.c
index b56ed97..253d0e5 100644
--- a/sys/gnu/ext2fs/ext2_vfsops.c
+++ b/sys/gnu/ext2fs/ext2_vfsops.c
@@ -412,10 +412,12 @@ static int compute_sb_data(devvp, es, fs)
 	    printf("EXT2-fs: unable to read group descriptors (%d)\n", error);
 	    return EIO;
 	}
+	/* Set the B_LOCKED flag on the buffer, then brelse() it */
+	LCK_BUF(fs->s_group_desc[i])
     }
     if(!ext2_check_descriptors(fs)) {
 	    for (j = 0; j < db_count; j++)
-		brelse(fs->s_group_desc[j]);
+		    ULCK_BUF(fs->s_group_desc[j])
 	    bsd_free(fs->s_group_desc, M_UFSMNT);
 	    printf("EXT2-fs: (ext2_check_descriptors failure) "
 		   "unable to read group descriptors\n");
@@ -691,16 +693,19 @@ ext2_unmount(mp, mntflags, p)
 		fs->s_es->s_state |= EXT2_VALID_FS;	/* was fs_clean = 1 */
 		ext2_sbupdate(ump, MNT_WAIT);
 	}
+
 	/* release buffers containing group descriptors */
 	for(i = 0; i < fs->s_db_per_group; i++) 
-		brelse(fs->s_group_desc[i]);
+		ULCK_BUF(fs->s_group_desc[i])
+
 	/* release cached inode/block bitmaps */
         for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
                 if (fs->s_inode_bitmap[i])
-                        brelse (fs->s_inode_bitmap[i]);
+			ULCK_BUF(fs->s_inode_bitmap[i])
+
         for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
                 if (fs->s_block_bitmap[i])
-                        brelse (fs->s_block_bitmap[i]);
+			ULCK_BUF(fs->s_block_bitmap[i])
 
 	ump->um_devvp->v_specflags &= ~SI_MOUNTEDON;
 	error = VOP_CLOSE(ump->um_devvp, ronly ? FREAD : FREAD|FWRITE,
@@ -1106,20 +1111,11 @@ printf("\nupdating superblock, waitfor=%s\n", waitfor == MNT_WAIT ? "yes":"no");
 	else
 		bawrite(bp);
 
-	/* write group descriptors back on disk */
-	for(i = 0; i < fs->s_db_per_group; i++) 
-		/* Godmar thinks: we must avoid using any of the b*write
-		 * functions here: we want to keep the buffer locked
-		 * so we use my 'housemade' write routine:
+	/*
+	 * The buffers for group descriptors, inode bitmaps and block bitmaps
+	 * are not busy at this point and are (hopefully) written by the
+	 * usual sync mechanism. No need to write them here
 		 */
-		error |= ll_w_block(fs->s_group_desc[i], waitfor == MNT_WAIT);
-
-        for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
-                if (fs->s_inode_bitmap[i])
-                        ll_w_block (fs->s_inode_bitmap[i], 1);
-        for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
-                if (fs->s_block_bitmap[i])
-                        ll_w_block (fs->s_block_bitmap[i], 1);
 
 	return (error);
 }
diff --git a/sys/gnu/ext2fs/fs.h b/sys/gnu/ext2fs/fs.h
index 03f137c..3cb4fc0 100644
--- a/sys/gnu/ext2fs/fs.h
+++ b/sys/gnu/ext2fs/fs.h
@@ -155,3 +155,23 @@ extern u_char *fragtbl[];
 #define  lock_super(devvp)   	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, curproc)
 #define  unlock_super(devvp) 	VOP_UNLOCK(devvp, 0, curproc)
 
+/*
+ * To lock a buffer, set the B_LOCKED flag and then brelse() it. To unlock,
+ * reset the B_LOCKED flag and brelse() the buffer back on the LRU list
+ */
+#define LCK_BUF(bp) { \
+	int s; \
+	s = splbio(); \
+	(bp)->b_flags |= B_LOCKED; \
+	splx(s); \
+	brelse(bp); \
+}
+
+#define ULCK_BUF(bp) { \
+	int s; \
+	s = splbio(); \
+	(bp)->b_flags &= ~B_LOCKED; \
+	splx(s); \
+	bremfree(bp); \
+	brelse(bp); \
+}
diff --git a/sys/gnu/fs/ext2fs/ext2_linux_balloc.c b/sys/gnu/fs/ext2fs/ext2_linux_balloc.c
index e0c955a..aa17650 100644
--- a/sys/gnu/fs/ext2fs/ext2_linux_balloc.c
+++ b/sys/gnu/fs/ext2fs/ext2_linux_balloc.c
@@ -69,6 +69,7 @@ static void read_block_bitmap (struct mount * mp,
 			    block_group, (unsigned long) gdp->bg_block_bitmap);
 	sb->s_block_bitmap_number[bitmap_nr] = block_group;
 	sb->s_block_bitmap[bitmap_nr] = bh;
+	LCK_BUF(bh)
 }
 
 /*
@@ -129,7 +130,8 @@ static int load__block_bitmap (struct mount * mp,
 		if (sb->s_loaded_block_bitmaps < EXT2_MAX_GROUP_LOADED)
 			sb->s_loaded_block_bitmaps++;
 		else
-			brelse (sb->s_block_bitmap[EXT2_MAX_GROUP_LOADED - 1]);
+			ULCK_BUF(sb->s_block_bitmap[EXT2_MAX_GROUP_LOADED - 1])
+
 		for (j = sb->s_loaded_block_bitmaps - 1; j > 0;  j--) {
 			sb->s_block_bitmap_number[j] =
 				sb->s_block_bitmap_number[j - 1];
diff --git a/sys/gnu/fs/ext2fs/ext2_linux_ialloc.c b/sys/gnu/fs/ext2fs/ext2_linux_ialloc.c
index c93c20e..9f2a967 100644
--- a/sys/gnu/fs/ext2fs/ext2_linux_ialloc.c
+++ b/sys/gnu/fs/ext2fs/ext2_linux_ialloc.c
@@ -127,6 +127,7 @@ static void read_inode_bitmap (struct mount * mp,
 			    block_group, (unsigned long) gdp->bg_inode_bitmap);
 	sb->s_inode_bitmap_number[bitmap_nr] = block_group;
 	sb->s_inode_bitmap[bitmap_nr] = bh;
+	LCK_BUF(bh)
 }
 
 /*
@@ -190,7 +191,7 @@ static int load_inode_bitmap (struct mount * mp,
 		if (sb->s_loaded_inode_bitmaps < EXT2_MAX_GROUP_LOADED)
 			sb->s_loaded_inode_bitmaps++;
 		else
-			brelse (sb->s_inode_bitmap[EXT2_MAX_GROUP_LOADED - 1]);
+			ULCK_BUF(sb->s_inode_bitmap[EXT2_MAX_GROUP_LOADED - 1])
 		for (j = sb->s_loaded_inode_bitmaps - 1; j > 0; j--) {
 			sb->s_inode_bitmap_number[j] =
 				sb->s_inode_bitmap_number[j - 1];
diff --git a/sys/gnu/fs/ext2fs/ext2_vfsops.c b/sys/gnu/fs/ext2fs/ext2_vfsops.c
index b56ed97..253d0e5 100644
--- a/sys/gnu/fs/ext2fs/ext2_vfsops.c
+++ b/sys/gnu/fs/ext2fs/ext2_vfsops.c
@@ -412,10 +412,12 @@ static int compute_sb_data(devvp, es, fs)
 	    printf("EXT2-fs: unable to read group descriptors (%d)\n", error);
 	    return EIO;
 	}
+	/* Set the B_LOCKED flag on the buffer, then brelse() it */
+	LCK_BUF(fs->s_group_desc[i])
     }
     if(!ext2_check_descriptors(fs)) {
 	    for (j = 0; j < db_count; j++)
-		brelse(fs->s_group_desc[j]);
+		    ULCK_BUF(fs->s_group_desc[j])
 	    bsd_free(fs->s_group_desc, M_UFSMNT);
 	    printf("EXT2-fs: (ext2_check_descriptors failure) "
 		   "unable to read group descriptors\n");
@@ -691,16 +693,19 @@ ext2_unmount(mp, mntflags, p)
 		fs->s_es->s_state |= EXT2_VALID_FS;	/* was fs_clean = 1 */
 		ext2_sbupdate(ump, MNT_WAIT);
 	}
+
 	/* release buffers containing group descriptors */
 	for(i = 0; i < fs->s_db_per_group; i++) 
-		brelse(fs->s_group_desc[i]);
+		ULCK_BUF(fs->s_group_desc[i])
+
 	/* release cached inode/block bitmaps */
         for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
                 if (fs->s_inode_bitmap[i])
-                        brelse (fs->s_inode_bitmap[i]);
+			ULCK_BUF(fs->s_inode_bitmap[i])
+
         for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
                 if (fs->s_block_bitmap[i])
-                        brelse (fs->s_block_bitmap[i]);
+			ULCK_BUF(fs->s_block_bitmap[i])
 
 	ump->um_devvp->v_specflags &= ~SI_MOUNTEDON;
 	error = VOP_CLOSE(ump->um_devvp, ronly ? FREAD : FREAD|FWRITE,
@@ -1106,20 +1111,11 @@ printf("\nupdating superblock, waitfor=%s\n", waitfor == MNT_WAIT ? "yes":"no");
 	else
 		bawrite(bp);
 
-	/* write group descriptors back on disk */
-	for(i = 0; i < fs->s_db_per_group; i++) 
-		/* Godmar thinks: we must avoid using any of the b*write
-		 * functions here: we want to keep the buffer locked
-		 * so we use my 'housemade' write routine:
+	/*
+	 * The buffers for group descriptors, inode bitmaps and block bitmaps
+	 * are not busy at this point and are (hopefully) written by the
+	 * usual sync mechanism. No need to write them here
 		 */
-		error |= ll_w_block(fs->s_group_desc[i], waitfor == MNT_WAIT);
-
-        for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
-                if (fs->s_inode_bitmap[i])
-                        ll_w_block (fs->s_inode_bitmap[i], 1);
-        for (i = 0; i < EXT2_MAX_GROUP_LOADED; i++)
-                if (fs->s_block_bitmap[i])
-                        ll_w_block (fs->s_block_bitmap[i], 1);
 
 	return (error);
 }
diff --git a/sys/gnu/fs/ext2fs/fs.h b/sys/gnu/fs/ext2fs/fs.h
index 03f137c..3cb4fc0 100644
--- a/sys/gnu/fs/ext2fs/fs.h
+++ b/sys/gnu/fs/ext2fs/fs.h
@@ -155,3 +155,23 @@ extern u_char *fragtbl[];
 #define  lock_super(devvp)   	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, curproc)
 #define  unlock_super(devvp) 	VOP_UNLOCK(devvp, 0, curproc)
 
+/*
+ * To lock a buffer, set the B_LOCKED flag and then brelse() it. To unlock,
+ * reset the B_LOCKED flag and brelse() the buffer back on the LRU list
+ */
+#define LCK_BUF(bp) { \
+	int s; \
+	s = splbio(); \
+	(bp)->b_flags |= B_LOCKED; \
+	splx(s); \
+	brelse(bp); \
+}
+
+#define ULCK_BUF(bp) { \
+	int s; \
+	s = splbio(); \
+	(bp)->b_flags &= ~B_LOCKED; \
+	splx(s); \
+	bremfree(bp); \
+	brelse(bp); \
+}