[PATCH] zram: unsafe zram_get_element call in zram_read_page()

[PATCH] zram: unsafe zram_get_element call in zram_read_page()

[Vasily Averin]

Taken lock is required to access the content of zram entry.

Fixes: 8e19d540d107 ("zram: extend zero pages to same element pages")
Signed-off-by: Vasily Averin <vasily.averin@linux.dev>
---
 drivers/block/zram/zram_drv.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index d77d3664ca08..9ac3d4e51d26 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -1362,14 +1362,14 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
 		ret = zram_read_from_zspool(zram, page, index);
 		zram_slot_unlock(zram, index);
 	} else {
+		unsigned long entry = zram_get_element(zram, index);
 		/*
 		 * The slot should be unlocked before reading from the backing
 		 * device.
 		 */
 		zram_slot_unlock(zram, index);
 
-		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
-				     parent);
+		ret = read_from_bdev(zram, page, entry, parent);
 	}
 
 	/* Should NEVER happen. Return bio error if it does. */
-- 
2.34.1

Re: [PATCH] zram: unsafe zram_get_element call in zram_read_page()

[Sergey Senozhatsky]

On (23/11/06 22:54), Vasily Averin wrote:
> @@ -1362,14 +1362,14 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
>  		ret = zram_read_from_zspool(zram, page, index);
>  		zram_slot_unlock(zram, index);
>  	} else {
> +		unsigned long entry = zram_get_element(zram, index);
>  		/*
>  		 * The slot should be unlocked before reading from the backing
>  		 * device.
>  		 */
>  		zram_slot_unlock(zram, index);
>  
> -		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
> -				     parent);
> +		ret = read_from_bdev(zram, page, entry, parent);

Hmmm,
We may want to do more here. Basically, we probably need to re-confirm
after read_from_bdev() that the entry at index still has ZRAM_WB set
and, if so, that it points to the same blk_idx. IOW, check that it has
not been free-ed and re-used under us.

Minchan, what do you think? Is that scenario possible?

Re: [PATCH] zram: unsafe zram_get_element call in zram_read_page()

[Sergey Senozhatsky]

On (23/11/07 16:39), Sergey Senozhatsky wrote:
> On (23/11/06 22:54), Vasily Averin wrote:
> > @@ -1362,14 +1362,14 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
> >  		ret = zram_read_from_zspool(zram, page, index);
> >  		zram_slot_unlock(zram, index);
> >  	} else {
> > +		unsigned long entry = zram_get_element(zram, index);
> >  		/*
> >  		 * The slot should be unlocked before reading from the backing
> >  		 * device.
> >  		 */
> >  		zram_slot_unlock(zram, index);
> >  
> > -		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
> > -				     parent);
> > +		ret = read_from_bdev(zram, page, entry, parent);
> 
> Hmmm,
> We may want to do more here. Basically, we probably need to re-confirm
> after read_from_bdev() that the entry at index still has ZRAM_WB set
> and, if so, that it points to the same blk_idx. IOW, check that it has
> not been free-ed and re-used under us.
> 
> Minchan, what do you think? Is that scenario possible?

Basically

---

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index f5e3756fc21a..2d69f6efcee3 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -1364,14 +1364,21 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
 		ret = zram_read_from_zspool(zram, page, index);
 		zram_slot_unlock(zram, index);
 	} else {
+		unsigned long idx = zram_get_element(zram, index);
 		/*
 		 * The slot should be unlocked before reading from the backing
 		 * device.
 		 */
 		zram_slot_unlock(zram, index);
 
-		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
-				     parent);
+		ret = read_from_bdev(zram, page, idx, parent);
+		if (ret == 0) {
+			zram_slot_lock(zram, index);
+			if (!zram_test_flag(zram, index, ZRAM_WB) ||
+			    idx != zram_get_element(zram, index))
+				ret = -EINVAL;
+			zram_slot_unlock(zram, index);
+		}
 	}
 
 	/* Should NEVER happen. Return bio error if it does. */

Re: [PATCH] zram: unsafe zram_get_element call in zram_read_page()

[Vasily Averin]

On 11/7/23 13:40, Sergey Senozhatsky wrote:
> On (23/11/07 16:39), Sergey Senozhatsky wrote:
>> Hmmm,
>> We may want to do more here. Basically, we probably need to re-confirm
>> after read_from_bdev() that the entry at index still has ZRAM_WB set
>> and, if so, that it points to the same blk_idx. IOW, check that it has
>> not been free-ed and re-used under us.
> --- a/drivers/block/zram/zram_drv.c
> +++ b/drivers/block/zram/zram_drv.c
> @@ -1364,14 +1364,21 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
>  		ret = zram_read_from_zspool(zram, page, index);
>  		zram_slot_unlock(zram, index);
>  	} else {
> +		unsigned long idx = zram_get_element(zram, index);
>  		/*
>  		 * The slot should be unlocked before reading from the backing
>  		 * device.
>  		 */
>  		zram_slot_unlock(zram, index);
>  
> -		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
> -				     parent);
> +		ret = read_from_bdev(zram, page, idx, parent);
> +		if (ret == 0) {
> +			zram_slot_lock(zram, index);
> +			if (!zram_test_flag(zram, index, ZRAM_WB) ||
> +			    idx != zram_get_element(zram, index))
> +				ret = -EINVAL;
> +			zram_slot_unlock(zram, index);
> +		}

Why overwritten page can not be pushed to WB to the same blk_idx? 
However I'm agree that this is VERY unlikely case, and this check is better than nothing.
 

Re: [PATCH] zram: unsafe zram_get_element call in zram_read_page()

[Sergey Senozhatsky]

On (23/11/07 21:19), Vasily Averin wrote:
> On 11/7/23 13:40, Sergey Senozhatsky wrote:
> > On (23/11/07 16:39), Sergey Senozhatsky wrote:
> >> Hmmm,
> >> We may want to do more here. Basically, we probably need to re-confirm
> >> after read_from_bdev() that the entry at index still has ZRAM_WB set
> >> and, if so, that it points to the same blk_idx. IOW, check that it has
> >> not been free-ed and re-used under us.
> > --- a/drivers/block/zram/zram_drv.c
> > +++ b/drivers/block/zram/zram_drv.c
> > @@ -1364,14 +1364,21 @@ static int zram_read_page(struct zram *zram, struct page *page, u32 index,
> >  		ret = zram_read_from_zspool(zram, page, index);
> >  		zram_slot_unlock(zram, index);
> >  	} else {
> > +		unsigned long idx = zram_get_element(zram, index);
> >  		/*
> >  		 * The slot should be unlocked before reading from the backing
> >  		 * device.
> >  		 */
> >  		zram_slot_unlock(zram, index);
> >  
> > -		ret = read_from_bdev(zram, page, zram_get_element(zram, index),
> > -				     parent);
> > +		ret = read_from_bdev(zram, page, idx, parent);
> > +		if (ret == 0) {
> > +			zram_slot_lock(zram, index);
> > +			if (!zram_test_flag(zram, index, ZRAM_WB) ||
> > +			    idx != zram_get_element(zram, index))
> > +				ret = -EINVAL;
> > +			zram_slot_unlock(zram, index);
> > +		}
> 
> Why overwritten page can not be pushed to WB to the same blk_idx? 

Yeah, so I thought about it too but didn't want to go too deep into it.
We probably can only address it if we synchronize free_page (?), read_page()
and writeback(), so that we never have concurrent bitmap modifications when
one of the operations is in progress.