{"id":"CVE-2026-97603","published":"2026-09-25T11:17:13.393","lastModified":"2026-09-25T11:17:13.393","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nidpf: disable DIM work before freeing q_vectors\n\nidpf never drains the Tx/Rx DIM works before freeing the memory they\nlive in.  tx_dim and rx_dim are embedded in struct idpf_q_vector, they\nare queued from the NAPI poll via net_dim(), and idpf_vport_intr_rel()\nends with kfree(rsrc->q_vectors).  Nothing in the driver cancels them.\n\nidpf_tx_dim_work() and idpf_rx_dim_work() then run on freed memory:\nidpf_vport_intr_write_itr() writes the ITR register through\nq_vector->intr_reg.tx_itr / rx_itr, void __iomem pointers loaded out of\nthe freed q_vector.  No configuration is needed to get there --\nIDPF_ITR_IS_DYNAMIC() is defined as (itr_mode) and idpf_vport_alloc()\ninitialises both modes to IDPF_ITR_DYNAMIC.\n\nDraining after idpf_vport_intr_napi_dis_all() is not enough on its own.\nidpf_net_dim() is called from inside the\n\"if (napi_complete_done(napi, work_done))\" branch of the poll, and\nnapi_complete_done() has already cleared NAPIF_STATE_SCHED by then.\nnapi_disable_locked() waits only while (val & (NAPIF_STATE_SCHED |\nNAPIF_STATE_NPSVC)), so napi_disable() can return while the poll tail is\nstill queueing the work, and a plain cancel_work_sync() would be\nre-armed behind the drain.\n\nUse disable_work_sync(): schedule_work() on a work with a non-zero\ndisable count is dropped by clear_pending_if_disabled() before\n__queue_work() is reached.\n\nMove idpf_init_dim() to idpf_vport_intr_alloc() so the works are\ninitialised on every path that can reach the drain -- the three\n\"goto intr_deinit\" sites between idpf_vport_intr_init() and\nidpf_vport_intr_ena() get there without the enable side having run.\nNothing re-enables them: rsrc->q_vectors is freed on every exit from\nidpf_vport_open() and on every idpf_vport_stop(), so the count dies with\nthe object.\n\nIt is a race, not a deterministic failure -- net_dim() only schedules\nonce DIM_NEVENTS events have accumulated and the profile index changes.\nA KASAN ifup/ifdown loop under load is the way to see it.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/7dd4c829bac2916be98a3e34b41daaba7f42b4c4","tags":[]},{"url":"https://git.kernel.org/stable/c/7e4312953788244d47074987d40d76702cf929f1","tags":[]},{"url":"https://git.kernel.org/stable/c/cd7a1598645b6917a8676b7e2efe8d97ce68a952","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}