"value":"In the Linux kernel, the following vulnerability has been resolved:\n\nnet/sched: taprio: avoid disabling offload when it was never enabled\n\nIn an incredibly strange API design decision, qdisc->destroy() gets\ncalled even if qdisc->init() never succeeded, not exclusively since\ncommit 87b60cfacf9f (\"net_sched: fix error recovery at qdisc creation\"),\nbut apparently also earlier (in the case of qdisc_create_dflt()).\n\nThe taprio qdisc does not fully acknowledge this when it attempts full\noffload, because it starts off with q->flags = TAPRIO_FLAGS_INVALID in\ntaprio_init(), then it replaces q->flags with TCA_TAPRIO_ATTR_FLAGS\nparsed from netlink (in taprio_change(), tail called from taprio_init()).\n\nBut in taprio_destroy(), we call taprio_disable_offload(), and this\ndetermines what to do based on FULL_OFFLOAD_IS_ENABLED(q->flags).\n\nBut looking at the implementation of FULL_OFFLOAD_IS_ENABLED()\n(a bitwise check of bit 1 in q->flags), it is invalid to call this macro\non q->flags when it contains TAPRIO_FLAGS_INVALID, because that is set\nto U32_MAX, and therefore FULL_OFFLOAD_IS_ENABLED() will return true on\nan invalid set of flags.\n\nAs a result, it is possible to crash the kernel if user space forces an\nerror between setting q->flags = TAPRIO_FLAGS_INVALID, and the calling\nof taprio_enable_offload(). This is because drivers do not expect the\noffload to be disabled when it was never enabled.\n\nThe error that we force here is to attach taprio as a non-root qdisc,\nbut instead as child of an mqprio root qdisc:\n\n$ tc qdisc add dev swp0 root handle 1: \\\n\tmqprio num_tc 8 map 0 1 2 3 4 5 6 7 \\\n\tqueues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 hw 0\n$ tc qdisc replace dev swp0 parent 1:1 \\\n\ttaprio num_tc 8 map 0 1 2 3 4 5 6 7 \\\n\tqueues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 base-time 0 \\\n\tsched-entry S 0x7f 990000 sched-entry S 0x80 100000 \\\n\tflags 0x0 clockid CLOCK_TAI\nUnable to handle kernel paging request at virtual address fffffffffffffff8\n[fffffffffffffff8] pgd=0000000000000000, p4d=0000000000000000\nInternal error: Oops: 96000004 [#1] PREEMPT SMP\nCall trace:\n taprio_dump+0x27c/0x310\n vsc9959_port_setup_tc+0x1f4/0x460\n felix_port_setup_tc+0x24/0x3c\n dsa_slave_setup_tc+0x54/0x27c\n taprio_disable_offload.isra.0+0x58/0xe0\n taprio_destroy+0x80/0x104\n qdisc_create+0x240/0x470\n tc_modify_qdisc+0x1fc/0x6b0\n rtnetlink_rcv_msg+0x12c/0x390\n netlink_rcv_skb+0x5c/0x130\n rtnetlink_rcv+0x1c/0x2c\n\nFix this by keeping track of the operations we made, and undo the\noffload only if we actually did it.\n\nI've added \"bool offloaded\" inside a 4 byte hole between \"int clockid\"\nand \"atomic64_t picos_per_byte\". Now the first cache line looks like\nbelow:\n\n$ pahole -C taprio_sched net/sched/sch_taprio.o\nstruct taprio_sched {\n struct Qdisc * * qdiscs; /* 0 8 */\n struct Qdisc * root; /* 8 8 */\n u32 flags; /* 16 4 */\n enum tk_offsets tk_offset; /* 20 4 */\n int clockid; /* 24 4 */\n bool offloaded; /* 28 1 */\n\n /* XXX 3 bytes hole, try to pack */\n\n atomic64_t picos_per_byte; /* 32 0 */\n\n /* XXX 8 bytes hole, try to pack */\n\n spinlock_t current_entry_lock; /* 40 0 */\n\n /* XXX 8 bytes hole, try to pack */\n\n struct sched_entry * current_entry; /* 48 8 */\n struct sched_gate_list * oper_sched; /* 56 8 */\n /* --- cacheline 1 boundary (64 bytes) --- */"
