CVE-2025-68178

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> blk-cgroup: fix possible deadlock while configuring policy<br /> <br /> Following deadlock can be triggered easily by lockdep:<br /> <br /> WARNING: possible circular locking dependency detected<br /> 6.17.0-rc3-00124-ga12c2658ced0 #1665 Not tainted<br /> ------------------------------------------------------<br /> check/1334 is trying to acquire lock:<br /> ff1100011d9d0678 (&amp;q-&gt;sysfs_lock){+.+.}-{4:4}, at: blk_unregister_queue+0x53/0x180<br /> <br /> but task is already holding lock:<br /> ff1100011d9d00e0 (&amp;q-&gt;q_usage_counter(queue)#3){++++}-{0:0}, at: del_gendisk+0xba/0x110<br /> <br /> which lock already depends on the new lock.<br /> <br /> the existing dependency chain (in reverse order) is:<br /> <br /> -&gt; #2 (&amp;q-&gt;q_usage_counter(queue)#3){++++}-{0:0}:<br /> blk_queue_enter+0x40b/0x470<br /> blkg_conf_prep+0x7b/0x3c0<br /> tg_set_limit+0x10a/0x3e0<br /> cgroup_file_write+0xc6/0x420<br /> kernfs_fop_write_iter+0x189/0x280<br /> vfs_write+0x256/0x490<br /> ksys_write+0x83/0x190<br /> __x64_sys_write+0x21/0x30<br /> x64_sys_call+0x4608/0x4630<br /> do_syscall_64+0xdb/0x6b0<br /> entry_SYSCALL_64_after_hwframe+0x76/0x7e<br /> <br /> -&gt; #1 (&amp;q-&gt;rq_qos_mutex){+.+.}-{4:4}:<br /> __mutex_lock+0xd8/0xf50<br /> mutex_lock_nested+0x2b/0x40<br /> wbt_init+0x17e/0x280<br /> wbt_enable_default+0xe9/0x140<br /> blk_register_queue+0x1da/0x2e0<br /> __add_disk+0x38c/0x5d0<br /> add_disk_fwnode+0x89/0x250<br /> device_add_disk+0x18/0x30<br /> virtblk_probe+0x13a3/0x1800<br /> virtio_dev_probe+0x389/0x610<br /> really_probe+0x136/0x620<br /> __driver_probe_device+0xb3/0x230<br /> driver_probe_device+0x2f/0xe0<br /> __driver_attach+0x158/0x250<br /> bus_for_each_dev+0xa9/0x130<br /> driver_attach+0x26/0x40<br /> bus_add_driver+0x178/0x3d0<br /> driver_register+0x7d/0x1c0<br /> __register_virtio_driver+0x2c/0x60<br /> virtio_blk_init+0x6f/0xe0<br /> do_one_initcall+0x94/0x540<br /> kernel_init_freeable+0x56a/0x7b0<br /> kernel_init+0x2b/0x270<br /> ret_from_fork+0x268/0x4c0<br /> ret_from_fork_asm+0x1a/0x30<br /> <br /> -&gt; #0 (&amp;q-&gt;sysfs_lock){+.+.}-{4:4}:<br /> __lock_acquire+0x1835/0x2940<br /> lock_acquire+0xf9/0x450<br /> __mutex_lock+0xd8/0xf50<br /> mutex_lock_nested+0x2b/0x40<br /> blk_unregister_queue+0x53/0x180<br /> __del_gendisk+0x226/0x690<br /> del_gendisk+0xba/0x110<br /> sd_remove+0x49/0xb0 [sd_mod]<br /> device_remove+0x87/0xb0<br /> device_release_driver_internal+0x11e/0x230<br /> device_release_driver+0x1a/0x30<br /> bus_remove_device+0x14d/0x220<br /> device_del+0x1e1/0x5a0<br /> __scsi_remove_device+0x1ff/0x2f0<br /> scsi_remove_device+0x37/0x60<br /> sdev_store_delete+0x77/0x100<br /> dev_attr_store+0x1f/0x40<br /> sysfs_kf_write+0x65/0x90<br /> kernfs_fop_write_iter+0x189/0x280<br /> vfs_write+0x256/0x490<br /> ksys_write+0x83/0x190<br /> __x64_sys_write+0x21/0x30<br /> x64_sys_call+0x4608/0x4630<br /> do_syscall_64+0xdb/0x6b0<br /> entry_SYSCALL_64_after_hwframe+0x76/0x7e<br /> <br /> other info that might help us debug this:<br /> <br /> Chain exists of:<br /> &amp;q-&gt;sysfs_lock --&gt; &amp;q-&gt;rq_qos_mutex --&gt; &amp;q-&gt;q_usage_counter(queue)#3<br /> <br /> Possible unsafe locking scenario:<br /> <br /> CPU0 CPU1<br /> ---- ----<br /> lock(&amp;q-&gt;q_usage_counter(queue)#3);<br /> lock(&amp;q-&gt;rq_qos_mutex);<br /> lock(&amp;q-&gt;q_usage_counter(queue)#3);<br /> lock(&amp;q-&gt;sysfs_lock);<br /> <br /> Root cause is that queue_usage_counter is grabbed with rq_qos_mutex<br /> held in blkg_conf_prep(), while queue should be freezed before<br /> rq_qos_mutex from other context.<br /> <br /> The blk_queue_enter() from blkg_conf_prep() is used to protect against<br /> policy deactivation, which is already protected with blkcg_mutex, hence<br /> convert blk_queue_enter() to blkcg_mutex to fix this problem. Meanwhile,<br /> consider that blkcg_mutex is held after queue is freezed from policy<br /> deactivation, also convert blkg_alloc() to use GFP_NOIO.