Re: Should block-number-in-file vars be using blkcnt_t?

Takashi Sato wrote:

>> Can anyone summarize the status of the various patchsets which have been
>> submitted for 2^32-1 -block filesystems?  What can I do to help get this
>> finished up and committed?

> AFAIK, the status is as follows.
> - Basically, >8TB filesystem became available with Mingming's patches(*1)
>  which add ext3_fsblk_t and ext3_grpblk_t on 2.6.18-rc1.
>  However, the overflow problems on full 2^32-1 blocks filesystem
>  remained.  (Eric caught these bugs.)

Looking over Takashi's patches, I think that this is all that remains to fix up the
32-bit containers in kernelspace.  If this patch (based on Takashi's patches)
looks acceptable, I'll submit it upstream.

A few comments/questions:

As Andreas noted, the changes to balloc.c aren't really necessary; underflow will
save us.  But I like the notion of not overflowing/underflowing if we can help it.
:)

I changed Takashi's resize checks to not cast up to 64 bits, I think the tests
below will catch overflows w/o the casting, and will work when we go to 64-bit
block containers.  The warning message is a little cryptic, but it's simple.  :)
Is "EPERM" what we want to return when the resize request is too large?

Signed-off-by: Eric Sandeen <[email protected]>

Index: linux-2.6.17/fs/ext3/balloc.c
===================================================================
--- linux-2.6.17.orig/fs/ext3/balloc.c
+++ linux-2.6.17/fs/ext3/balloc.c
@@ -168,7 +168,7 @@ goal_in_my_reservation(struct ext3_reser
 	ext3_fsblk_t group_first_block, group_last_block;
 
 	group_first_block = ext3_group_first_block_no(sb, group);
-	group_last_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
+	group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
 
 	if ((rsv->_rsv_start > group_last_block) ||
 	    (rsv->_rsv_end < group_first_block))
@@ -897,7 +897,7 @@ static int alloc_new_reservation(struct 
 	spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
 
 	group_first_block = ext3_group_first_block_no(sb, group);
-	group_end_block = group_first_block + EXT3_BLOCKS_PER_GROUP(sb) - 1;
+	group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
 
 	if (grp_goal < 0)
 		start_block = group_first_block;
@@ -1132,7 +1132,7 @@ ext3_try_to_allocate_with_rsv(struct sup
 			try_to_extend_reservation(my_rsv, sb,
 					*count-my_rsv->rsv_end + grp_goal - 1);
 
-		if ((my_rsv->rsv_start >= group_first_block + EXT3_BLOCKS_PER_GROUP(sb))
+		if ((my_rsv->rsv_start > group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1))
 		    || (my_rsv->rsv_end < group_first_block))
 			BUG();
 		ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh, grp_goal,
Index: linux-2.6.17/fs/ext3/ialloc.c
===================================================================
--- linux-2.6.17.orig/fs/ext3/ialloc.c
+++ linux-2.6.17/fs/ext3/ialloc.c
@@ -202,7 +202,7 @@ error_return:
 static int find_group_dir(struct super_block *sb, struct inode *parent)
 {
 	int ngroups = EXT3_SB(sb)->s_groups_count;
-	int freei, avefreei;
+	unsigned long freei, avefreei;
 	struct ext3_group_desc *desc, *best_desc = NULL;
 	struct buffer_head *bh;
 	int group, best_group = -1;
@@ -261,10 +261,10 @@ static int find_group_orlov(struct super
 	struct ext3_super_block *es = sbi->s_es;
 	int ngroups = sbi->s_groups_count;
 	int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
-	int freei, avefreei;
+	unsigned long freei, avefreei;
 	ext3_fsblk_t freeb, avefreeb;
 	ext3_fsblk_t blocks_per_dir;
-	int ndirs;
+	unsigned long ndirs;
 	int max_debt, max_dirs, min_inodes;
 	ext3_grpblk_t min_blocks;
 	int group = -1, i;
Index: linux-2.6.17/fs/ext3/resize.c
===================================================================
--- linux-2.6.17.orig/fs/ext3/resize.c
+++ linux-2.6.17/fs/ext3/resize.c
@@ -730,6 +730,16 @@ int ext3_group_add(struct super_block *s
 		return -EPERM;
 	}
 
+	if (es->s_blocks_count + input->blocks_count < es->s_blocks_count) {
+		ext3_warning(sb, __FUNCTION__, "blocks_count overflow\n");
+		return -EPERM;
+	}
+
+	if (es->s_inodes_count+EXT3_INODES_PER_GROUP(sb) < es->s_inodes_count) {
+		ext3_warning(sb, __FUNCTION__, "inodes_count overflow\n");
+		return -EPERM;
+	}
+
 	if (reserved_gdb || gdb_off == 0) {
 		if (!EXT3_HAS_COMPAT_FEATURE(sb,
 					     EXT3_FEATURE_COMPAT_RESIZE_INODE)){
@@ -958,6 +968,11 @@ int ext3_group_extend(struct super_block
 
 	add = EXT3_BLOCKS_PER_GROUP(sb) - last;
 
+	if (o_blocks_count + add < o_blocks_count) {
+		ext3_warning(sb, __FUNCTION__, "blocks_count overflow");
+		return -EPERM;
+	}
+
 	if (o_blocks_count + add > n_blocks_count)
 		add = n_blocks_count - o_blocks_count;
 


> Cheers, Takashi


Thanks!

-Eric


-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642