blkdev_pr_read_keys() takes num_keys from userspace and uses it to
calculate the allocation size for keys_info via struct_size(). While
there is a check for SIZE_MAX (integer overflow), there is no upper
bound validation on the allocation size itself.

A malicious or buggy userspace can pass a large num_keys value that
doesn't trigger overflow but still results in an excessive allocation
attempt, causing a warning in the page allocator when the order exceeds
MAX_PAGE_ORDER.

Fix this by introducing PR_KEYS_MAX to limit the number of keys to
a sane value. This makes the SIZE_MAX check redundant, so remove it.
Also switch to kvzalloc/kvfree to handle larger allocations gracefully.

Fixes: 22a1ffea5f80 ("block: add IOC_PR_READ_KEYS ioctl")
Tested-by: syzbot+660d079d90f8a1baf54d@syzkaller.appspotmail.com
Reported-by: syzbot+660d079d90f8a1baf54d@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=660d079d90f8a1baf54d
Link: https://lore.kernel.org/all/20251212013510.3576091-1-kartikey406@gmail.com/T/ [v1]
Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
---
v3:
  - Renamed PR_KEYS_MAX_NUM to PR_KEYS_MAX
  - Moved define to include/uapi/linux/pr.h
v2:
  - Added PR_KEYS_MAX_NUM (64K) limit instead of checking KMALLOC_MAX_SIZE
  - Removed redundant SIZE_MAX check
  - Switched to kvzalloc/kvfree
---
 block/ioctl.c           | 9 +++++----
 include/uapi/linux/pr.h | 2 ++
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/block/ioctl.c b/block/ioctl.c
index 61feed686418..344478348a54 100644
--- a/block/ioctl.c
+++ b/block/ioctl.c
@@ -442,11 +442,12 @@ static int blkdev_pr_read_keys(struct block_device *bdev, blk_mode_t mode,
 	if (copy_from_user(&read_keys, arg, sizeof(read_keys)))
 		return -EFAULT;
 
-	keys_info_len = struct_size(keys_info, keys, read_keys.num_keys);
-	if (keys_info_len == SIZE_MAX)
+	if (read_keys.num_keys > PR_KEYS_MAX)
 		return -EINVAL;
 
-	keys_info = kzalloc(keys_info_len, GFP_KERNEL);
+	keys_info_len = struct_size(keys_info, keys, read_keys.num_keys);
+
+	keys_info = kvzalloc(keys_info_len, GFP_KERNEL);
 	if (!keys_info)
 		return -ENOMEM;
 
@@ -473,7 +474,7 @@ static int blkdev_pr_read_keys(struct block_device *bdev, blk_mode_t mode,
 	if (copy_to_user(arg, &read_keys, sizeof(read_keys)))
 		ret = -EFAULT;
 out:
-	kfree(keys_info);
+	kvfree(keys_info);
 	return ret;
 }
 
diff --git a/include/uapi/linux/pr.h b/include/uapi/linux/pr.h
index 847f3051057a..f0ecb1677317 100644
--- a/include/uapi/linux/pr.h
+++ b/include/uapi/linux/pr.h
@@ -79,4 +79,6 @@ struct pr_read_reservation {
 #define IOC_PR_READ_KEYS	_IOWR('p', 206, struct pr_read_keys)
 #define IOC_PR_READ_RESERVATION	_IOR('p', 207, struct pr_read_reservation)
 
+#define PR_KEYS_MAX		(1u << 16)
+
 #endif /* _UAPI_PR_H */
-- 
2.43.0

On Wed, 17 Dec 2025 07:17:12 +0530, Deepanshu Kartikey wrote:
> blkdev_pr_read_keys() takes num_keys from userspace and uses it to
> calculate the allocation size for keys_info via struct_size(). While
> there is a check for SIZE_MAX (integer overflow), there is no upper
> bound validation on the allocation size itself.
> 
> A malicious or buggy userspace can pass a large num_keys value that
> doesn't trigger overflow but still results in an excessive allocation
> attempt, causing a warning in the page allocator when the order exceeds
> MAX_PAGE_ORDER.
> 
> [...]

Applied, thanks!

[1/1] block: add allocation size check in blkdev_pr_read_keys()
      commit: a58383fa45c706bda3bf4a1955c3a0327dbec7e7

Best regards,
-- 
Jens Axboe

Deepanshu,

> blkdev_pr_read_keys() takes num_keys from userspace and uses it to
> calculate the allocation size for keys_info via struct_size(). While
> there is a check for SIZE_MAX (integer overflow), there is no upper
> bound validation on the allocation size itself.

LGTM.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

On Wed, Dec 17, 2025 at 07:17:12AM +0530, Deepanshu Kartikey wrote:
> blkdev_pr_read_keys() takes num_keys from userspace and uses it to
> calculate the allocation size for keys_info via struct_size(). While
> there is a check for SIZE_MAX (integer overflow), there is no upper
> bound validation on the allocation size itself.
> 
> A malicious or buggy userspace can pass a large num_keys value that
> doesn't trigger overflow but still results in an excessive allocation
> attempt, causing a warning in the page allocator when the order exceeds
> MAX_PAGE_ORDER.
> 
> Fix this by introducing PR_KEYS_MAX to limit the number of keys to
> a sane value. This makes the SIZE_MAX check redundant, so remove it.
> Also switch to kvzalloc/kvfree to handle larger allocations gracefully.
> 
> Fixes: 22a1ffea5f80 ("block: add IOC_PR_READ_KEYS ioctl")
> Tested-by: syzbot+660d079d90f8a1baf54d@syzkaller.appspotmail.com
> Reported-by: syzbot+660d079d90f8a1baf54d@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=660d079d90f8a1baf54d
> Link: https://lore.kernel.org/all/20251212013510.3576091-1-kartikey406@gmail.com/T/ [v1]
> Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
> ---
> v3:
>   - Renamed PR_KEYS_MAX_NUM to PR_KEYS_MAX
>   - Moved define to include/uapi/linux/pr.h
> v2:
>   - Added PR_KEYS_MAX_NUM (64K) limit instead of checking KMALLOC_MAX_SIZE
>   - Removed redundant SIZE_MAX check
>   - Switched to kvzalloc/kvfree
> ---
>  block/ioctl.c           | 9 +++++----
>  include/uapi/linux/pr.h | 2 ++
>  2 files changed, 7 insertions(+), 4 deletions(-)

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>