On 01/24/2017 10:52 AM, Christoph Hellwig wrote:
> On Tue, Jan 24, 2017 at 10:32:11AM +0100, Matias Bjørling wrote:
>> *(gdb) list *blkdev_direct_IO+0x50c
>> 0xffffffff8142ab8c is in blkdev_direct_IO (fs/block_dev.c:401).
>> 396 submit_bio(bio);
>> 397 bio = bio_alloc(GFP_KERNEL, nr_pages);
>> 398 }
>> 399 blk_finish_plug(&plug);
>> 400
>> 401 if (!dio->is_sync)
>> 402 return -EIOCBQUEUED;
>>
>> It looks like the qc = submit_bio() completes the I/O before the
>> blkdev_direct_IO completes, which then leads to the use after free case,
>> when the private dio struct is accessed.
>
> Ok, the is_sync check here is clearly a bug, we need a local variable
> for is_sync as well for the aio case:
>
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 5db5d13..3c47614 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -331,7 +331,7 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter
> *iter, int nr_pages)
> struct blk_plug plug;
> struct blkdev_dio *dio;
> struct bio *bio;
> - bool is_read = (iov_iter_rw(iter) == READ);
> + bool is_read = (iov_iter_rw(iter) == READ), is_sync;
> loff_t pos = iocb->ki_pos;
> blk_qc_t qc = BLK_QC_T_NONE;
> int ret;
> @@ -344,7 +344,7 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter
> *iter, int nr_pages)
> bio_get(bio); /* extra ref for the completion handler */
>
> dio = container_of(bio, struct blkdev_dio, bio);
> - dio->is_sync = is_sync_kiocb(iocb);
> + dio->is_sync = is_sync = is_sync_kiocb(iocb);
> if (dio->is_sync)
> dio->waiter = current;
> else
> @@ -398,7 +398,7 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter
> *iter, int nr_pages)
> }
> blk_finish_plug(&plug);
>
> - if (!dio->is_sync)
> + if (!is_sync)
> return -EIOCBQUEUED;
>
> for (;;) {
>
Yup. That fixes it. Should I put together the patch, or will you take
care of it?
--
To unsubscribe from this list: send the line "unsubscribe linux-block" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
