{"id":"CVE-2026-98789","published":"2026-10-18T16:17:32.887","lastModified":"2026-10-19T10:19:42.963","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nbpf: Fix objects stuck in free_by_rcu_ttrace\n\ndo_call_rcu_ttrace() returns early when call_rcu_ttrace_in_progress is set\nand leaves the objects in free_by_rcu_ttrace. __free_rcu() frees\nwaiting_for_gp_ttrace only and clears the flag. Hence the objects that\nfree_bulk() or __free_by_rcu() added while RCU tasks trace GP was in flight\nstay in free_by_rcu_ttrace until free_bulk() or alloc_bulk() is called for\nthe same bpf_mem_cache again, which may never happen. The number of such\nobjects is not bounded.\n\nTurn call_rcu_ttrace_in_progress into three states:\n0 - idle\n1 - __free_rcu() is queued\n2 - __free_rcu() is queued and free_by_rcu_ttrace got more objects since\n\ndo_call_rcu_ttrace() sets 2. __free_rcu() does cmpxchg(1 -> 0) and starts\nthe next GP when it fails. It cannot clear the flag first and check\nfree_by_rcu_ttrace later, since bpf_mem_alloc_destroy() frees bpf_mem_cache\nwithout waiting for RCU callbacks when the flag is zero.\n\nNow __free_rcu() queues itself, so the one that didn't see 'draining' may\ndo call_rcu_tasks_trace() after rcu_barrier_tasks_trace() in\nfree_mem_alloc(). Queue it under rcu_read_lock() and do synchronize_rcu()\nbefore the barriers. Calling rcu_barrier_tasks_trace() twice works too, but\ncreating and destroying hash maps in a loop on many cpus slows down to one\nfree_mem_alloc() per GP and kworkers pile up.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/33d0a5d2fee06a51f2f056dd901296515912e954","tags":[]},{"url":"https://git.kernel.org/stable/c/fbd97dd1d3ce32d32fb7be86acc3fdb7c04fa043","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}