Message-ID: <Zj3kowGa9XzJ0yak@visitorckw-System-Product-Name>
Date: Fri, 10 May 2024 17:10:59 +0800
From: Kuan-Wei Chiu <visitorckw@...il.com>
To: Matthew Mirvish <matthew@...2.xyz>
Cc: Kent Overstreet <kent.overstreet@...ux.dev>, Coly Li <colyli@...e.de>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Linux Next Mailing List <linux-next@...r.kernel.org>
Subject: Re: linux-next: manual merge of the refactor-heap tree with the
 block tree

On Thu, May 09, 2024 at 11:46:18PM -0400, Matthew Mirvish wrote:
> On Fri, May 10, 2024 at 11:07:11AM +0800, Kuan-Wei Chiu wrote:
> > On Thu, May 09, 2024 at 07:16:31PM -0400, Kent Overstreet wrote:
> > > On Fri, May 10, 2024 at 06:44:29AM +0800, Kuan-Wei Chiu wrote:
> > > > On Thu, May 09, 2024 at 03:58:57PM -0400, Kent Overstreet wrote:
> > > > > On Thu, May 09, 2024 at 03:27:45PM +1000, Stephen Rothwell wrote:
> > > > > > Hi all,
> > > > > > 
> > > > > > Today's linux-next merge of the refactor-heap tree got conflicts in:
> > > > > > 
> > > > > >   drivers/md/bcache/bset.c
> > > > > >   drivers/md/bcache/bset.h
> > > > > >   drivers/md/bcache/btree.c
> > > > > >   drivers/md/bcache/writeback.c
> > > > > > 
> > > > > > between commit:
> > > > > > 
> > > > > >   3a861560ccb3 ("bcache: fix variable length array abuse in btree_iter")
> > > > > > 
> > > > > > from the block tree and commit:
> > > > > > 
> > > > > >   afa5721abaaa ("bcache: Remove heap-related macros and switch to generic min_heap")
> > > > > > 
> > > > > > from the refactor-heap tree.
> > > > > > 
> > > > > > Ok, these conflicts are too extensive, so I am dropping the refactor-heap
> > > > > > tree for today.  I suggest you all get together and sort something out.
> > > > > 
> > > > > Coli and Kuan, you guys will need to get this sorted out quick if we
> > > > > want refactor-heap to make the merge window
> > > > 
> > > > Hi Coli and Kent,
> > > > 
> > > > If I understand correctly, the reported bug is because we attempted to
> > > > point (heap)->data to a dynamically allocated memory , but at that time
> > > > (heap)->data was not a regular pointer but a fixed size array with a
> > > > length of MAX_BSETS.
> > > > 
> > > > In my refactor heap patch series, I introduced a preallocated array and
> > > > decided in min_heap_init() whether the data pointer should point to an
> > > > incoming pointer or to the preallocated array. Therefore, I am
> > > > wondering if my patch might have unintentionally fixed this bug?
> > > > 
> > > > I am unsure how to reproduce the reported issue. Could you assist me in
> > > > verifying whether my assumption is correct?
> > > 
> > > This is a merge conflict, not a runtime. Can you rebase onto Coli's
> > > tree? We'll have to retest.
> > 
> > Oh, sorry for the misunderstanding I caused. When I mentioned "bug" [1]
> > earlier, I was referring to the bug addressed in
> > 3a861560ccb3 ("bcache: fix variable length array abuse in btree_iter"),
> > not a merge conflict.
> > 
> > Here are the results after the rebase:
> > https://github.com/visitorckw/linux.git refactor-heap
> > 
> > [1]: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2039368
> 
> The ubuntu kernels build with UBSAN now, and the bug reported is just a
> UBSAN warning. The original implementation's iterator has a fixed size
> sets array that is indexed out of bounds when the iterator is allocated
> on the heap with more space -- the patch restructures it a bit to have a
> single iterator type with a flexible array and then a larger "stack"
> type which embeds the iterator along with the preallocated region.
> 
> I took a brief look at the refactor-heap branch but I'm not entirely
> sure what's going on with the new min heaps: in the one place where the
> larger iterators are used (in bch_btree_node_read_done) it doesn't look
> like the heap is ever initialized (perhaps since the old iter_init
> wasn't used here because of the special case it got missed in the
> refactor?) With the new heaps it should be fairly easy to fix though;
> just change the fill_iter mempool to be allocating only the minheap data
> arrays and setup iter->heap.data properly with that instead.

Thank you, Matthew.
Not initializing the heap's data pointer was indeed my mistake.
Following your advice, I made the following modifications to the code
on the refactor-heap branch in my github repo. I hope this time it
works well.

Regards,
Kuan-Wei

diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index a2bb86d52ad4..ce9d729bc8ff 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -149,19 +149,19 @@ void bch_btree_node_read_done(struct btree *b)
 {
 	const char *err = "bad btree header";
 	struct bset *i = btree_bset_first(b);
-	struct btree_iter *iter;
+	struct btree_iter iter;

 	/*
 	 * c->fill_iter can allocate an iterator with more memory space
 	 * than static MAX_BSETS.
 	 * See the comment arount cache_set->fill_iter.
 	 */
-	iter = mempool_alloc(&b->c->fill_iter, GFP_NOIO);
-	iter->heap.size = b->c->cache->sb.bucket_size / b->c->cache->sb.block_size;
-	iter->heap.nr = 0;
+	iter.heap.data = mempool_alloc(&b->c->fill_iter, GFP_NOIO);
+	iter.heap.size = b->c->cache->sb.bucket_size / b->c->cache->sb.block_size;
+	iter.heap.nr = 0;

 #ifdef CONFIG_BCACHE_DEBUG
-	iter->b = &b->keys;
+	iter.b = &b->keys;
 #endif

 	if (!i->seq)
@@ -199,7 +199,7 @@ void bch_btree_node_read_done(struct btree *b)
 		if (i != b->keys.set[0].data && !i->keys)
 			goto err;

-		bch_btree_iter_push(iter, i->start, bset_bkey_last(i));
+		bch_btree_iter_push(&iter, i->start, bset_bkey_last(i));

 		b->written += set_blocks(i, block_bytes(b->c->cache));
 	}
@@ -211,7 +211,7 @@ void bch_btree_node_read_done(struct btree *b)
 		if (i->seq == b->keys.set[0].data->seq)
 			goto err;

-	bch_btree_sort_and_fix_extents(&b->keys, iter, &b->c->sort);
+	bch_btree_sort_and_fix_extents(&b->keys, &iter, &b->c->sort);

 	i = b->keys.set[0].data;
 	err = "short btree key";
@@ -223,7 +223,7 @@ void bch_btree_node_read_done(struct btree *b)
 		bch_bset_init_next(&b->keys, write_block(b),
 				   bset_magic(&b->c->cache->sb));
 out:
-	mempool_free(iter, &b->c->fill_iter);
+	mempool_free(iter.heap.data, &b->c->fill_iter);
 	return;
 err:
 	set_btree_node_io_error(b);
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index cba09660148a..c6f5592996a8 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c
@@ -1914,8 +1914,7 @@ struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
 	INIT_LIST_HEAD(&c->btree_cache_freed);
 	INIT_LIST_HEAD(&c->data_buckets);

-	iter_size = sizeof(struct btree_iter) +
-		    ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size) *
+	iter_size = ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size) *
 			    sizeof(struct btree_iter_set);

 	c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL);

Powered by blists - more mailing lists