CVE-2025-68756

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> block: Use RCU in blk_mq_[un]quiesce_tagset() instead of set-&gt;tag_list_lock<br /> <br /> blk_mq_{add,del}_queue_tag_set() functions add and remove queues from<br /> tagset, the functions make sure that tagset and queues are marked as<br /> shared when two or more queues are attached to the same tagset.<br /> Initially a tagset starts as unshared and when the number of added<br /> queues reaches two, blk_mq_add_queue_tag_set() marks it as shared along<br /> with all the queues attached to it. When the number of attached queues<br /> drops to 1 blk_mq_del_queue_tag_set() need to mark both the tagset and<br /> the remaining queues as unshared.<br /> <br /> Both functions need to freeze current queues in tagset before setting on<br /> unsetting BLK_MQ_F_TAG_QUEUE_SHARED flag. While doing so, both functions<br /> hold set-&gt;tag_list_lock mutex, which makes sense as we do not want<br /> queues to be added or deleted in the process. This used to work fine<br /> until commit 98d81f0df70c ("nvme: use blk_mq_[un]quiesce_tagset")<br /> made the nvme driver quiesce tagset instead of quiscing individual<br /> queues. blk_mq_quiesce_tagset() does the job and quiesce the queues in<br /> set-&gt;tag_list while holding set-&gt;tag_list_lock also.<br /> <br /> This results in deadlock between two threads with these stacktraces:<br /> <br /> __schedule+0x47c/0xbb0<br /> ? timerqueue_add+0x66/0xb0<br /> schedule+0x1c/0xa0<br /> schedule_preempt_disabled+0xa/0x10<br /> __mutex_lock.constprop.0+0x271/0x600<br /> blk_mq_quiesce_tagset+0x25/0xc0<br /> nvme_dev_disable+0x9c/0x250<br /> nvme_timeout+0x1fc/0x520<br /> blk_mq_handle_expired+0x5c/0x90<br /> bt_iter+0x7e/0x90<br /> blk_mq_queue_tag_busy_iter+0x27e/0x550<br /> ? __blk_mq_complete_request_remote+0x10/0x10<br /> ? __blk_mq_complete_request_remote+0x10/0x10<br /> ? __call_rcu_common.constprop.0+0x1c0/0x210<br /> blk_mq_timeout_work+0x12d/0x170<br /> process_one_work+0x12e/0x2d0<br /> worker_thread+0x288/0x3a0<br /> ? rescuer_thread+0x480/0x480<br /> kthread+0xb8/0xe0<br /> ? kthread_park+0x80/0x80<br /> ret_from_fork+0x2d/0x50<br /> ? kthread_park+0x80/0x80<br /> ret_from_fork_asm+0x11/0x20<br /> <br /> __schedule+0x47c/0xbb0<br /> ? xas_find+0x161/0x1a0<br /> schedule+0x1c/0xa0<br /> blk_mq_freeze_queue_wait+0x3d/0x70<br /> ? destroy_sched_domains_rcu+0x30/0x30<br /> blk_mq_update_tag_set_shared+0x44/0x80<br /> blk_mq_exit_queue+0x141/0x150<br /> del_gendisk+0x25a/0x2d0<br /> nvme_ns_remove+0xc9/0x170<br /> nvme_remove_namespaces+0xc7/0x100<br /> nvme_remove+0x62/0x150<br /> pci_device_remove+0x23/0x60<br /> device_release_driver_internal+0x159/0x200<br /> unbind_store+0x99/0xa0<br /> kernfs_fop_write_iter+0x112/0x1e0<br /> vfs_write+0x2b1/0x3d0<br /> ksys_write+0x4e/0xb0<br /> do_syscall_64+0x5b/0x160<br /> entry_SYSCALL_64_after_hwframe+0x4b/0x53<br /> <br /> The top stacktrace is showing nvme_timeout() called to handle nvme<br /> command timeout. timeout handler is trying to disable the controller and<br /> as a first step, it needs to blk_mq_quiesce_tagset() to tell blk-mq not<br /> to call queue callback handlers. The thread is stuck waiting for<br /> set-&gt;tag_list_lock as it tries to walk the queues in set-&gt;tag_list.<br /> <br /> The lock is held by the second thread in the bottom stack which is<br /> waiting for one of queues to be frozen. The queue usage counter will<br /> drop to zero after nvme_timeout() finishes, and this will not happen<br /> because the thread will wait for this mutex forever.<br /> <br /> Given that [un]quiescing queue is an operation that does not need to<br /> sleep, update blk_mq_[un]quiesce_tagset() to use RCU instead of taking<br /> set-&gt;tag_list_lock, update blk_mq_{add,del}_queue_tag_set() to use RCU<br /> safe list operations. Also, delete INIT_LIST_HEAD(&amp;q-&gt;tag_set_list)<br /> in blk_mq_del_queue_tag_set() because we can not re-initialize it while<br /> the list is being traversed under RCU. The deleted queue will not be<br /> added/deleted to/from a tagset and it will be freed in blk_free_queue()<br /> after the end of RCU grace period.