"value":"In the Linux kernel, the following vulnerability has been resolved:\n\nbinder: fix OOB in binder_add_freeze_work()\n\nIn binder_add_freeze_work() we iterate over the proc->nodes with the\nproc->inner_lock held. However, this lock is temporarily dropped to\nacquire the node->lock first (lock nesting order). This can race with\nbinder_deferred_release() which removes the nodes from the proc->nodes\nrbtree and adds them into binder_dead_nodes list. This leads to a broken\niteration in binder_add_freeze_work() as rb_next() will use data from\nbinder_dead_nodes, triggering an out-of-bounds access:\n\n ==================================================================\n BUG: KASAN: global-out-of-bounds in rb_next+0xfc/0x124\n Read of size 8 at addr ffffcb84285f7170 by task freeze/660\n\n CPU: 8 UID: 0 PID: 660 Comm: freeze Not tainted 6.11.0-07343-ga727812a8d45 #18\n Hardware name: linux,dummy-virt (DT)\n Call trace:\n rb_next+0xfc/0x124\n binder_add_freeze_work+0x344/0x534\n binder_ioctl+0x1e70/0x25ac\n __arm64_sys_ioctl+0x124/0x190\n\n The buggy address belongs to the variable:\n binder_dead_nodes+0x10/0x40\n [...]\n ==================================================================\n\nThis is possible because proc->nodes (rbtree) and binder_dead_nodes\n(list) share entries in binder_node through a union:\n\n\tstruct binder_node {\n\t[...]\n\t\tunion {\n\t\t\tstruct rb_node rb_node;\n\t\t\tstruct hlist_node dead_node;\n\t\t};\n\nFix the race by checking that the proc is still alive. If not, simply\nbreak out of the iteration."
