CVE-2024-35884

The Linux kernel contains an ABBA deadlock between the mmap_lock and disassociation_lock in the RDMA/uverbs subsystem. A thread performing mmap/fork acquires locks in the order mmap_lock → disassociation_lock, while the thread executing uverbs_user_mmap_disassociate() acquires them in reverse order, leading to a potential deadlock. The issue was resolved by removing disassociation_lock and using the existing hw_destroy_rwsem instead.

In the Linux kernel, the RDMA/efa management IRQ handler accesses both the admin completion queue and the async event queue. The driver registered the IRQ before constructing these queues and destroyed them before freeing the IRQ, so the handler's lifetime was not contained by the resources it accesses. The fix initializes the queues with interrupts masked, requests the IRQ, then switches to interrupt mode, and on removal resets the device and frees the IRQ before destroying the queues.

In the Linux kernel, the RDMA/efa completion IRQ handler accesses the EQ state and DMA buffer. Its IRQ was registered before that state was initialized, while teardown released the buffer before free_irq() synchronized the handler. The fix initializes the EQ without arming it, registers the IRQ, then arms it, and reverses the resource order during teardown by freeing the IRQ before destroying the EQ.

In the Linux kernel, the USB 6fire driver uses URBs embedded in different structs for PCM, MIDI and communication, which is a buggy implementation nowadays; since a URB is managed with a refcount, this may lead to a UAF when the URB is released asynchronously. The fix converts those embedded URBs to ones properly allocated via usb_alloc_urb(), changing each embedded struct urb to a pointer and updating callers; resources are released in common destructors called at both the init error path and disconnect.

In the Linux kernel, the CoDel scheduler's dropping loop in codel_dequeue() could fail to advance with very small interval values, draining the entire backlog under the qdisc lock in a single call. An unprivileged user could trigger a soft lockup this way. The fix ensures the control-law step is at least 1 tick and caps the dropping loop at 256 drops per codel_dequeue() call.

In the Linux kernel, the mwifiex wifi driver's mwifiex_pcie_request_irq() registers each MSI-X vector with a per-index dev_id (&card->msix_ctx[i]). On a request_irq() failure the cleanup loop "for (j = 0; j < i; j++)" frees msix_entries[j].vector but passes the failed index's &card->msix_ctx[i] as the dev_id. Since free_irq() matches on (irq, dev_id), it fails to find the action registered with &card->msix_ctx[j]: the already-requested IRQ j is not freed (leaked) and free_irq() warns about freeing a non-existent IRQ. The fix uses &card->msix_ctx[j].

In the Linux kernel, ieee80211_request_ibss_scan() in mac80211 warns when the regulatory domain leaves no allowed channel to scan. This can happen because the regdomain can change while IBSS is operating, and IBSS can continue to operate briefly during the 60s grace period until it is shut down. The fix removes the warning in this case.

In the Linux kernel, cfg80211_switch_netns() in cfg80211 moves all interfaces, then moves the wiphy itself by setting its network namespace and renaming it, which makes sysfs move it. The rename can fail (only on allocation failures), leaving things mixed up and hitting a warning. The fix undoes the move and lets the change fail, and if undo fails it WARNs.

In the Linux kernel, mac80211_hwsim checks ->started for frames coming from wmediumd, but the radio can be stopped after the check and before frame delivery, causing mac80211 to hit WARN_ON(!local->started). The fix expands the mutex for this case and synchronizes against it when the radio is stopped, and drops an error print that complicated the error path.

In the Linux kernel, mac80211 AP_VLAN interfaces get crypto_tx_tailroom_needed_cnt from the AP interface on ifup, but it is never decremented unless the AP is also brought down. Bringing the same AP_VLAN up again increments the counter again and eventually hits the WARN_ON_ONCE sanity check. The fix resets it on ifdown.