Re: [PATCH] ext4: convert pa_count from atomic_t to refcount_t

On Sun 19-07-26 20:44:41, rafad900 wrote:
> The pa_count field in struct ext4_prealloc_space is
> reference counter for preallocation descriptors.
> Converting it from atomic_t to refcount_t provides
> underflow and overflow protection.
> 
> This was found using Coccinelle script
> atomic_as_refcounter.cocci
> 
> Testing was done on a QEMU x86 VM revealing a
> pre-existing race condition leading to a
> use-after-free.
> 
> Signed-off-by: rafad900 <[email protected]>

This is a bit problematic as for pa_count 0->1 transition is fine (as long
as pa_deleted isn't set) so this conversion will trigger false warnings as
the 0-day report shows.

								Honza

> ---
>  fs/ext4/mballoc.c | 22 +++++++++++-----------
>  fs/ext4/mballoc.h |  2 +-
>  2 files changed, 12 insertions(+), 12 deletions(-)
> 
> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
> index ed1bd00e11cd..59b3d4abaab0 100644
> --- a/fs/ext4/mballoc.c
> +++ b/fs/ext4/mballoc.c
> @@ -4823,7 +4823,7 @@ ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
>  	ext4_fsblk_t cur_distance, new_distance;
>  
>  	if (cpa == NULL) {
> -		atomic_inc(&pa->pa_count);
> +		refcount_inc(&pa->pa_count);
>  		return pa;
>  	}
>  	cur_distance = abs(goal_block - cpa->pa_pstart);
> @@ -4833,8 +4833,8 @@ ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
>  		return cpa;
>  
>  	/* drop the previous reference */
> -	atomic_dec(&cpa->pa_count);
> -	atomic_inc(&pa->pa_count);
> +	refcount_dec(&cpa->pa_count);
> +	refcount_inc(&pa->pa_count);
>  	return pa;
>  }
>  
> @@ -4993,7 +4993,7 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
>  	}
>  
>  	if (tmp_pa->pa_free && likely(ext4_mb_pa_goal_check(ac, tmp_pa))) {
> -		atomic_inc(&tmp_pa->pa_count);
> +		refcount_inc(&tmp_pa->pa_count);
>  		ext4_mb_use_inode_pa(ac, tmp_pa);
>  		spin_unlock(&tmp_pa->pa_lock);
>  		read_unlock(&ei->i_prealloc_lock);
> @@ -5139,7 +5139,7 @@ static void ext4_mb_mark_pa_deleted(struct super_block *sb,
>  static inline void ext4_mb_pa_free(struct ext4_prealloc_space *pa)
>  {
>  	BUG_ON(!pa);
> -	BUG_ON(atomic_read(&pa->pa_count));
> +	BUG_ON(refcount_read(&pa->pa_count));
>  	BUG_ON(pa->pa_deleted == 0);
>  	kmem_cache_free(ext4_pspace_cachep, pa);
>  }
> @@ -5165,7 +5165,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac,
>  
>  	/* in this short window concurrent discard can set pa_deleted */
>  	spin_lock(&pa->pa_lock);
> -	if (!atomic_dec_and_test(&pa->pa_count) || pa->pa_free != 0) {
> +	if (!refcount_dec_and_test(&pa->pa_count) || pa->pa_free != 0) {
>  		spin_unlock(&pa->pa_lock);
>  		return;
>  	}
> @@ -5535,7 +5535,7 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
>  	list_for_each_entry_safe(pa, tmp,
>  				&grp->bb_prealloc_list, pa_group_list) {
>  		spin_lock(&pa->pa_lock);
> -		if (atomic_read(&pa->pa_count)) {
> +		if (refcount_read(&pa->pa_count)) {
>  			spin_unlock(&pa->pa_lock);
>  			*busy = 1;
>  			continue;
> @@ -5637,7 +5637,7 @@ void ext4_discard_preallocations(struct inode *inode)
>  		BUG_ON(pa->pa_node_lock.inode_lock != &ei->i_prealloc_lock);
>  
>  		spin_lock(&pa->pa_lock);
> -		if (atomic_read(&pa->pa_count)) {
> +		if (refcount_read(&pa->pa_count)) {
>  			/* this shouldn't happen often - nobody should
>  			 * use preallocation while we're discarding it */
>  			spin_unlock(&pa->pa_lock);
> @@ -5720,7 +5720,7 @@ static int ext4_mb_pa_alloc(struct ext4_allocation_context *ac)
>  	pa = kmem_cache_zalloc(ext4_pspace_cachep, GFP_NOFS);
>  	if (!pa)
>  		return -ENOMEM;
> -	atomic_set(&pa->pa_count, 1);
> +	refcount_set(&pa->pa_count, 1);
>  	ac->ac_pa = pa;
>  	return 0;
>  }
> @@ -5731,7 +5731,7 @@ static void ext4_mb_pa_put_free(struct ext4_allocation_context *ac)
>  
>  	BUG_ON(!pa);
>  	ac->ac_pa = NULL;
> -	WARN_ON(!atomic_dec_and_test(&pa->pa_count));
> +	WARN_ON(!refcount_dec_and_test(&pa->pa_count));
>  	/*
>  	 * current function is only called due to an error or due to
>  	 * len of found blocks < len of requested blocks hence the PA has not
> @@ -5948,7 +5948,7 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
>  				pa_node.lg_list,
>  				lockdep_is_held(&lg->lg_prealloc_lock)) {
>  		spin_lock(&pa->pa_lock);
> -		if (atomic_read(&pa->pa_count)) {
> +		if (refcount_read(&pa->pa_count)) {
>  			/*
>  			 * This is the pa that we just used
>  			 * for block allocation. So don't
> diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
> index 39333ce72cbd..f708d303228e 100644
> --- a/fs/ext4/mballoc.h
> +++ b/fs/ext4/mballoc.h
> @@ -126,7 +126,7 @@ struct ext4_prealloc_space {
>  		struct rcu_head	pa_rcu;
>  	} u;
>  	spinlock_t		pa_lock;
> -	atomic_t		pa_count;
> +	refcount_t		pa_count;
>  	unsigned		pa_deleted;
>  	ext4_fsblk_t		pa_pstart;	/* phys. block */
>  	ext4_lblk_t		pa_lstart;	/* log. block */
> -- 
> 2.43.0
> 
-- 
Jan Kara <[email protected]>
SUSE Labs, CR