btrfs: Fix slab-use-after-free Read in add_ra_bio_pages

PATCH Superseded Pei Li 2 participants 2 messages Jul 11, 2024 Btrfs file system

lore

Share & tools

lore

Pei Li

author

2024-07-11 03:42 UTC #0 lore raw

PATCH 1/1 btrfs: Fix slab-use-after-free Read in add_ra_bio_pages

We are accessing the start and len field in em after it is free'd.

This patch stores the values that we are going to access from em before it was free'd so we won't access free'd memory.

Reported-by [email protected]

Closes https://syzkaller.appspot.com/bug?extid=853d80cba98ce1157ae6

Signed-off-by Pei Li <[email protected]>

---
Syzbot reported the following error:

BUG: KASAN: slab-use-after-free in add_ra_bio_pages.constprop.0.isra.0+0xf03/0xfb0 fs/btrfs/compression.c:529

This is because we are reading the values from em right after freeing it before through free_extent_map(em).

This patch stores the values that we are going to access from em before it was free'd so we won't access free'd memory.

---
 fs/btrfs/compression.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

fs/btrfs/compression.c +8 -2

index 6441e47d8a5e..42b528aee63b 100644
@@ -449,6 +449,7 @@ static noinline int add_ra_bio_pages(struct inode *inode,
 		u64 page_end;
 		u64 pg_index = cur >> PAGE_SHIFT;
 		u32 add_size;
+		u64 start = 0, len = 0;
 
 		if (pg_index > end_index)
 			break;

@@ -500,12 +501,17 @@ static noinline int add_ra_bio_pages(struct inode *inode,
 		em = lookup_extent_mapping(em_tree, cur, page_end + 1 - cur);
 		read_unlock(&em_tree->lock);
 
+		if (em) {
+			start = em->start;
+			len = em->len;
+		}
+
 		/*
 		 * At this point, we have a locked page in the page cache for
 		 * these bytes in the file.  But, we have to make sure they map
 		 * to this compressed extent on disk.
 		 */
-		if (!em || cur < em->start ||
+		if (!em || cur < start ||
 		    (cur + fs_info->sectorsize > extent_map_end(em)) ||
 		    (em->block_start >> SECTOR_SHIFT) != orig_bio->bi_iter.bi_sector) {
 			free_extent_map(em);

@@ -526,7 +532,7 @@ static noinline int add_ra_bio_pages(struct inode *inode,
 			}
 		}
 
-		add_size = min(em->start + em->len, page_end + 1) - cur;
+		add_size = min(start + len, page_end + 1) - cur;
 		ret = bio_add_page(orig_bio, page, add_size, offset_in_page(cur));
 		if (ret != add_size) {
 			unlock_extent(tree, cur, page_end, NULL);

---
base-commit: 563a50672d8a86ec4b114a4a2f44d6e7ff855f5b
change-id: 20240710-bug11-a8ac18afb724

Best regards,

Signature · 1 line

Pei Li <[email protected]>

Qu Wenruo

2024-07-11 03:57 UTC #1 lore raw

Quoted 8 lines · Pei Li

We are accessing the start and len field in em after it is free'd.
This patch stores the values that we are going to access from em before it was free'd so we won't access free'd memory.
Reported-by [email protected]
Closes https://syzkaller.appspot.com/bug?extid=853d80cba98ce1157ae6
Signed-off-by Pei Li <[email protected]>

And missing the fixes tag:

Fixes 6a4049102055 ("btrfs: subpage: make add_ra_bio_pages() compatible")

And it is better to Cc to stable kernel for v5.16 and newer kernels.

Quoted 9 lines

---
Syzbot reported the following error:
BUG: KASAN: slab-use-after-free in add_ra_bio_pages.constprop.0.isra.0+0xf03/0xfb0 fs/btrfs/compression.c:529
This is because we are reading the values from em right after freeing it before through free_extent_map(em).
This patch stores the values that we are going to access from em before it was free'd so we won't access free'd memory.

Indeed it's a use after free.

But the only usage is in this line:

	add_size = min(em->start + em->len, page_end + 1) - cur;

So why not just move that line before "free_extent_map(em);"? It should be much easier to read than the current change.

Thanks,
Qu

Quoted 50 lines

---
  fs/btrfs/compression.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

fs/btrfs/compression.c +8 -3

index 6441e47d8a5e..42b528aee63b 100644
@@ -449,6 +449,7 @@ static noinline int add_ra_bio_pages(struct inode *inode,
  		u64 page_end;
  		u64 pg_index = cur >> PAGE_SHIFT;
  		u32 add_size;
+		u64 start = 0, len = 0;
 
  		if (pg_index > end_index)
  			break;

@@ -500,12 +501,17 @@ static noinline int add_ra_bio_pages(struct inode *inode,
  		em = lookup_extent_mapping(em_tree, cur, page_end + 1 - cur);
  		read_unlock(&em_tree->lock);
 
+		if (em) {
+			start = em->start;
+			len = em->len;
+		}
+
  		/*
  		 * At this point, we have a locked page in the page cache for
  		 * these bytes in the file.  But, we have to make sure they map
  		 * to this compressed extent on disk.
  		 */
-		if (!em || cur < em->start ||
+		if (!em || cur < start ||
  		    (cur + fs_info->sectorsize > extent_map_end(em)) ||
  		    (em->block_start >> SECTOR_SHIFT) != orig_bio->bi_iter.bi_sector) {
  			free_extent_map(em);

@@ -526,7 +532,7 @@ static noinline int add_ra_bio_pages(struct inode *inode,
  			}
  		}
 
-		add_size = min(em->start + em->len, page_end + 1) - cur;
+		add_size = min(start + len, page_end + 1) - cur;
  		ret = bio_add_page(orig_bio, page, add_size, offset_in_page(cur));
  		if (ret != add_size) {
  			unlock_extent(tree, cur, page_end, NULL);
 
---

base-commit: 563a50672d8a86ec4b114a4a2f44d6e7ff855f5b
change-id: 20240710-bug11-a8ac18afb724

Best regards,