]> git.baikalelectronics.ru Git - kernel.git/commitdiff
blk-mq: grab rq->refcount before calling ->fn in blk_mq_tagset_busy_iter
authorMing Lei <ming.lei@redhat.com>
Tue, 11 May 2021 15:22:34 +0000 (23:22 +0800)
committerJens Axboe <axboe@kernel.dk>
Mon, 24 May 2021 12:47:22 +0000 (06:47 -0600)
Grab rq->refcount before calling ->fn in blk_mq_tagset_busy_iter(), and
this way will prevent the request from being re-used when ->fn is
running. The approach is same as what we do during handling timeout.

Fix request use-after-free(UAF) related with completion race or queue
releasing:

- If one rq is referred before rq->q is frozen, then queue won't be
frozen before the request is released during iteration.

- If one rq is referred after rq->q is frozen, refcount_inc_not_zero()
will return false, and we won't iterate over this request.

However, still one request UAF not covered: refcount_inc_not_zero() may
read one freed request, and it will be handled in next patch.

Tested-by: John Garry <john.garry@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20210511152236.763464-3-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-mq-tag.c
block/blk-mq.c
block/blk-mq.h

index 2a37731e8244b8a54facc9f5cdfecdc54ef36edb..544edf2c56a5a1492277afc84a6c67a0521f8143 100644 (file)
@@ -199,6 +199,16 @@ struct bt_iter_data {
        bool reserved;
 };
 
+static struct request *blk_mq_find_and_get_req(struct blk_mq_tags *tags,
+               unsigned int bitnr)
+{
+       struct request *rq = tags->rqs[bitnr];
+
+       if (!rq || !refcount_inc_not_zero(&rq->ref))
+               return NULL;
+       return rq;
+}
+
 static bool bt_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
 {
        struct bt_iter_data *iter_data = data;
@@ -206,18 +216,22 @@ static bool bt_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
        struct blk_mq_tags *tags = hctx->tags;
        bool reserved = iter_data->reserved;
        struct request *rq;
+       bool ret = true;
 
        if (!reserved)
                bitnr += tags->nr_reserved_tags;
-       rq = tags->rqs[bitnr];
-
        /*
         * We can hit rq == NULL here, because the tagging functions
         * test and set the bit before assigning ->rqs[].
         */
-       if (rq && rq->q == hctx->queue && rq->mq_hctx == hctx)
-               return iter_data->fn(hctx, rq, iter_data->data, reserved);
-       return true;
+       rq = blk_mq_find_and_get_req(tags, bitnr);
+       if (!rq)
+               return true;
+
+       if (rq->q == hctx->queue && rq->mq_hctx == hctx)
+               ret = iter_data->fn(hctx, rq, iter_data->data, reserved);
+       blk_mq_put_rq_ref(rq);
+       return ret;
 }
 
 /**
@@ -264,6 +278,8 @@ static bool bt_tags_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
        struct blk_mq_tags *tags = iter_data->tags;
        bool reserved = iter_data->flags & BT_TAG_ITER_RESERVED;
        struct request *rq;
+       bool ret = true;
+       bool iter_static_rqs = !!(iter_data->flags & BT_TAG_ITER_STATIC_RQS);
 
        if (!reserved)
                bitnr += tags->nr_reserved_tags;
@@ -272,16 +288,19 @@ static bool bt_tags_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
         * We can hit rq == NULL here, because the tagging functions
         * test and set the bit before assigning ->rqs[].
         */
-       if (iter_data->flags & BT_TAG_ITER_STATIC_RQS)
+       if (iter_static_rqs)
                rq = tags->static_rqs[bitnr];
        else
-               rq = tags->rqs[bitnr];
+               rq = blk_mq_find_and_get_req(tags, bitnr);
        if (!rq)
                return true;
-       if ((iter_data->flags & BT_TAG_ITER_STARTED) &&
-           !blk_mq_request_started(rq))
-               return true;
-       return iter_data->fn(rq, iter_data->data, reserved);
+
+       if (!(iter_data->flags & BT_TAG_ITER_STARTED) ||
+           blk_mq_request_started(rq))
+               ret = iter_data->fn(rq, iter_data->data, reserved);
+       if (!iter_static_rqs)
+               blk_mq_put_rq_ref(rq);
+       return ret;
 }
 
 /**
@@ -348,6 +367,9 @@ void blk_mq_all_tag_iter(struct blk_mq_tags *tags, busy_tag_iter_fn *fn,
  *             indicates whether or not @rq is a reserved request. Return
  *             true to continue iterating tags, false to stop.
  * @priv:      Will be passed as second argument to @fn.
+ *
+ * We grab one request reference before calling @fn and release it after
+ * @fn returns.
  */
 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
                busy_tag_iter_fn *fn, void *priv)
index c86c01bfecdbe51e67128bc952c2780c4a9eecab..debfa5cd8025e75647eabfa4da7dd603cd5b0780 100644 (file)
@@ -909,6 +909,14 @@ static bool blk_mq_req_expired(struct request *rq, unsigned long *next)
        return false;
 }
 
+void blk_mq_put_rq_ref(struct request *rq)
+{
+       if (is_flush_rq(rq, rq->mq_hctx))
+               rq->end_io(rq, 0);
+       else if (refcount_dec_and_test(&rq->ref))
+               __blk_mq_free_request(rq);
+}
+
 static bool blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
                struct request *rq, void *priv, bool reserved)
 {
@@ -942,11 +950,7 @@ static bool blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
        if (blk_mq_req_expired(rq, next))
                blk_mq_rq_timed_out(rq, reserved);
 
-       if (is_flush_rq(rq, hctx))
-               rq->end_io(rq, 0);
-       else if (refcount_dec_and_test(&rq->ref))
-               __blk_mq_free_request(rq);
-
+       blk_mq_put_rq_ref(rq);
        return true;
 }
 
index 9ce64bc4a6c8f2a38205aa20bc943108bd479141..556368d2c5b692fa0647be9ddcc588ef3a06ea4e 100644 (file)
@@ -47,6 +47,7 @@ void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list);
 struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
                                        struct blk_mq_ctx *start);
+void blk_mq_put_rq_ref(struct request *rq);
 
 /*
  * Internal helpers for allocating/freeing the request map