Skip to contentSkip to conversation

bio: fix argument of __bio_add_page() for max_sectors > 0xffff

PATCH In review Akinobu Mita 2 participants 3 messages Nov 18, 2013 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bio: fix argument of __bio_add_page() for max_sectors > 0xffff

The data type of max_sectors and max_hw_sectors in queue settings are unsigned int. But these values are passed to __bio_add_page() as an argument whose data type is unsigned short. In the worst case such as max_sectors is 0x10000, bio_add_page() can't add a page and IOs can't proceed.

Cc Jens Axboe <[email protected]>
Cc Alexander Viro <[email protected]>
---
 fs/bio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/bio.c +1 -1
index 2bdb4e2..33d79a4 100644
@@ -601,7 +601,7 @@ EXPORT_SYMBOL(bio_get_nr_vecs);
 
 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 			  *page, unsigned int len, unsigned int offset,
-			  unsigned short max_sectors)
+			  unsigned int max_sectors)
 {
 	int retried_segments = 0;
 	struct bio_vec *bvec;
Signature · 1 line
1.8.3.2
#1 lore raw
Quoted 5 lines · Akinobu Mita

The data type of max_sectors and max_hw_sectors in queue settings are unsigned int. But these values are passed to __bio_add_page() as an argument whose data type is unsigned short. In the worst case such as max_sectors is 0x10000, bio_add_page() can't add a page and IOs can't proceed.

Thanks, added. Even at 512b sectors, 0xffff sectors would be a huuuuge page. So I don't think we'd ever see this bug hit, iirc huge pages are still broken down for IO. But it's better to be safe...

Signature · 1 line
Jens Axboe
#2 lore raw

2013/11/19 Jens Axboe <[email protected]>:

Quoted 10 lines · Akinobu Mita
Quoted 5 lines

The data type of max_sectors and max_hw_sectors in queue settings are unsigned int. But these values are passed to __bio_add_page() as an argument whose data type is unsigned short. In the worst case such as max_sectors is 0x10000, bio_add_page() can't add a page and IOs can't proceed.

Thanks, added. Even at 512b sectors, 0xffff sectors would be a huuuuge page. So I don't think we'd ever see this bug hit, iirc huge pages are still broken down for IO. But it's better to be safe...

I hit this when I was trying to change the data type of max_sectors in scsi host settings from 16-bit to 32-bit.

BTW, I found that BLKSECTGET only returns 16-bit of max_sectors. Should we need the change like below?

case BLKSECTGET:
- return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
+ max_sectors = min(0xffff,
+ queue_max_sectors(bdev_get_queue(bdev)));
+ return put_ushort(arg, max_sectors);