CVE-2025-21850

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> nvmet: Fix crash when a namespace is disabled<br /> <br /> The namespace percpu counter protects pending I/O, and we can<br /> only safely diable the namespace once the counter drop to zero.<br /> Otherwise we end up with a crash when running blktests/nvme/058<br /> (eg for loop transport):<br /> <br /> [ 2352.930426] [ T53909] Oops: general protection fault, probably for non-canonical address 0xdffffc0000000005: 0000 [#1] PREEMPT SMP KASAN PTI<br /> [ 2352.930431] [ T53909] KASAN: null-ptr-deref in range [0x0000000000000028-0x000000000000002f]<br /> [ 2352.930434] [ T53909] CPU: 3 UID: 0 PID: 53909 Comm: kworker/u16:5 Tainted: G W 6.13.0-rc6 #232<br /> [ 2352.930438] [ T53909] Tainted: [W]=WARN<br /> [ 2352.930440] [ T53909] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-3.fc41 04/01/2014<br /> [ 2352.930443] [ T53909] Workqueue: nvmet-wq nvme_loop_execute_work [nvme_loop]<br /> [ 2352.930449] [ T53909] RIP: 0010:blkcg_set_ioprio+0x44/0x180<br /> <br /> as the queue is already torn down when calling submit_bio();<br /> <br /> So we need to init the percpu counter in nvmet_ns_enable(), and<br /> wait for it to drop to zero in nvmet_ns_disable() to avoid having<br /> I/O pending after the namespace has been disabled.