{"id":"CVE-2026-90001","aliases":[],"url":"https://o3.security/vulnerability/CVE-2026-90001","summary":"In the Linux kernel, the following vulnerability has been resolved:\n\nHID: bpf: serialize device reference release in struct_ops destroy path\n\n__hid_bpf_ops_destroy_device() and hid_bpf_unreg()…","details":"In the Linux kernel, the following vulnerability has been resolved:\n\nHID: bpf: serialize device reference release in struct_ops destroy path\n\n__hid_bpf_ops_destroy_device() and hid_bpf_unreg() can race on the\nsame registration reference, double-putting struct hid_device and\nfreeing it while hid_destroy_device() still uses it.  Serialize the\nremove/NULL decision under hdev->bpf.prog_list_lock so exactly one\npath releases each registration reference: unreg re-checks ops->hdev\nunder the lock and returns without putting when the destroy path\nalready cleared it; all put_device() calls happen after the lock is\ndropped, which is safe because a concurrent unreg then observes\nops->hdev == NULL under the lock.\n\nBackground: each successful attach (hid_bpf_ops_reg) acquires one\ndevice reference (hid_get_device()).  Two paths can release it:\n\n- device destruction: hid_destroy_device() -> hid_bpf_destroy_device()\n  -> __hid_bpf_ops_destroy_device(), which walks hdev->bpf.prog_list\n  under rcu_read_lock() and drops one reference per attached program;\n- BPF link release: bpf map delete (no BPF_F_LINK) synchronously calls\n  st_ops->unreg() -> hid_bpf_unreg(), which drops the reference for\n  its own registration.\n\nThe coordination handshake (e->hdev = NULL on the destroy side vs\n\"if (!hdev) return\" on the unreg side) is a TOCTOU check: the two\npaths run under different lock domains (rcu_read_lock vs\nprog_list_lock), so a concurrent unreg can read ops->hdev as\nnon-NULL, block on prog_list_lock, and then proceed while the\ndestroy traversal executes - both paths then drop the same\nreference.  The refcount reaches zero legitimately (each decrement\nis individually valid), so no refcount_t saturation fires: the\ndevice is simply freed while the transport is still inside\nhid_destroy_device(), and subsequent teardown touches freed memory.\n\nThe fix serializes the remove/NULL decision under prog_list_lock on\nboth sides and moves the destroy-side puts outside the lock.  With\nthe lock held, plain reads/writes of ops->hdev are sufficient; no\nREAD_ONCE/WRITE_ONCE are added, keeping the patch minimal.\n\nUnlocked-read safety: the unlocked read of ops->hdev at the top of\nhid_bpf_unreg() cannot touch a freed device, because the unreg path\nitself still holds this registration's reference (released only by\nits own hid_put_device() after the lock is dropped), and a destroy\ntraversal that already cleared ops->hdev makes the lock-internal\nre-check return early without any put.  At most one of the two\npaths releases each registration reference.","published":"2026-09-16T11:17:11.700","modified":"2026-09-16T11:17:11.700","cvss":null,"epss":null,"cisaKev":null,"exploitsKnown":null,"affectedPackages":[],"fix":null,"references":[{"type":"WEB","url":"https://git.kernel.org/stable/c/401359684620145be710de97b87e1a47abfe1459"},{"type":"WEB","url":"https://git.kernel.org/stable/c/9cdc7e6dc7a99ad7311ad5e7c145f2b9ce4e24b0"},{"type":"WEB","url":"https://git.kernel.org/stable/c/bfb7939788f3c8dd080a4dd81e38d625b35d194e"},{"type":"WEB","url":"https://git.kernel.org/stable/c/c7f927aa8b55008ed5ea0814313d5dad771dcf3c"}],"provenance":{"sources":["OSV.dev","NVD","FIRST.org (EPSS)"],"lastVerified":"2026-09-16T11:17:11.700"}}