On Fri, Nov 22, 2019 at 12:27 PM <glider@xxxxxxxxxx> wrote:
>
> Some functions are called from handwritten assembly, and therefore don't
> have their arguments' metadata fully set up by the instrumentation code.
> Mark them with __no_sanitize_memory to avoid false positives from
> spreading further.
> Certain functions perform task switching, so that the value of |current|
> is different as they proceed. Because KMSAN state pointer is only read
> once at the beginning of the function, touching it after |current| has
> changed may be dangerous.
>
> Signed-off-by: Alexander Potapenko <glider@xxxxxxxxxx>
> To: Alexander Potapenko <glider@xxxxxxxxxx>
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Vegard Nossum <vegard.nossum@xxxxxxxxxx>
> Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> Cc: linux-mm@xxxxxxxxx
> ---
> v3:
> - removed TODOs from comments
>
> Change-Id: I684d23dac5a22eb0a4cea71993cb934302b17cea
> ---
> arch/x86/entry/common.c | 1 +
> arch/x86/include/asm/irq_regs.h | 1 +
> arch/x86/include/asm/syscall_wrapper.h | 1 +
> arch/x86/kernel/apic/apic.c | 1 +
> arch/x86/kernel/dumpstack_64.c | 1 +
> arch/x86/kernel/process_64.c | 5 +++++
> arch/x86/kernel/traps.c | 12 ++++++++++--
> arch/x86/kernel/uprobes.c | 7 ++++++-
> kernel/profile.c | 1 +
> kernel/sched/core.c | 6 ++++++
> 10 files changed, 33 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
> index 3f8e22615812..0dd5b2acb355 100644
> --- a/arch/x86/entry/common.c
> +++ b/arch/x86/entry/common.c
> @@ -275,6 +275,7 @@ __visible inline void syscall_return_slowpath(struct pt_regs *regs)
> }
>
> #ifdef CONFIG_X86_64
> +__no_sanitize_memory
> __visible void do_syscall_64(unsigned long nr, struct pt_regs *regs)
> {
> struct thread_info *ti;
> diff --git a/arch/x86/include/asm/irq_regs.h b/arch/x86/include/asm/irq_regs.h
> index 187ce59aea28..d65a00bd6f02 100644
> --- a/arch/x86/include/asm/irq_regs.h
> +++ b/arch/x86/include/asm/irq_regs.h
> @@ -14,6 +14,7 @@
>
> DECLARE_PER_CPU(struct pt_regs *, irq_regs);
>
> +__no_sanitize_memory
> static inline struct pt_regs *get_irq_regs(void)
> {
> return __this_cpu_read(irq_regs);
> diff --git a/arch/x86/include/asm/syscall_wrapper.h b/arch/x86/include/asm/syscall_wrapper.h
> index e046a405743d..43910ce1b53b 100644
> --- a/arch/x86/include/asm/syscall_wrapper.h
> +++ b/arch/x86/include/asm/syscall_wrapper.h
> @@ -159,6 +159,7 @@
> ALLOW_ERROR_INJECTION(__x64_sys##name, ERRNO); \
> static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
> static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
> + __no_sanitize_memory \
> asmlinkage long __x64_sys##name(const struct pt_regs *regs) \
> { \
> return __se_sys##name(SC_X86_64_REGS_TO_ARGS(x,__VA_ARGS__));\
> diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
> index 9e2dd2b296cd..7b24bda22c38 100644
> --- a/arch/x86/kernel/apic/apic.c
> +++ b/arch/x86/kernel/apic/apic.c
> @@ -1118,6 +1118,7 @@ static void local_apic_timer_interrupt(void)
> * [ if a single-CPU system runs an SMP kernel then we call the local
> * interrupt as well. Thus we cannot inline the local irq ... ]
> */
> +__no_sanitize_memory /* |regs| may be uninitialized */
The comment style around __no_sanitize_memory looks very different for
different call sites, perhaps it makes sense to unify it somehow.
> __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
> {
> struct pt_regs *old_regs = set_irq_regs(regs);
> diff --git a/arch/x86/kernel/dumpstack_64.c b/arch/x86/kernel/dumpstack_64.c
> index 753b8cfe8b8a..ba883d282a43 100644
> --- a/arch/x86/kernel/dumpstack_64.c
> +++ b/arch/x86/kernel/dumpstack_64.c
> @@ -143,6 +143,7 @@ static bool in_irq_stack(unsigned long *stack, struct stack_info *info)
> return true;
> }
>
> +__no_sanitize_memory
> int get_stack_info(unsigned long *stack, struct task_struct *task,
> struct stack_info *info, unsigned long *visit_mask)
> {
> diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
> index af64519b2695..70e33150a83a 100644
> --- a/arch/x86/kernel/process_64.c
> +++ b/arch/x86/kernel/process_64.c
> @@ -500,6 +500,11 @@ void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp)
> * Kprobes not supported here. Set the probe on schedule instead.
> * Function graph tracer not supported too.
> */
> +/*
> + * Avoid touching KMSAN state or reporting anything here, as __switch_to() does
> + * weird things with tasks.
> + */
> +__no_sanitize_memory
> __visible __notrace_funcgraph struct task_struct *
> __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
> {
> diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
> index 4bb0f8447112..a94282d1f60b 100644
> --- a/arch/x86/kernel/traps.c
> +++ b/arch/x86/kernel/traps.c
> @@ -618,7 +618,10 @@ NOKPROBE_SYMBOL(do_int3);
> * Help handler running on a per-cpu (IST or entry trampoline) stack
> * to switch to the normal thread stack if the interrupted code was in
> * user mode. The actual stack switch is done in entry_64.S
> + *
> + * This function switches the registers - don't instrument it with KMSAN!
> */
> +__no_sanitize_memory
> asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
> {
> struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
> @@ -634,6 +637,11 @@ struct bad_iret_stack {
> };
>
> asmlinkage __visible notrace
> +/*
> + * Dark magic happening here, let's not instrument this function.
> + * Also avoid copying any metadata by using raw __memmove().
> + */
> +__no_sanitize_memory
> struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
> {
> /*
> @@ -648,10 +656,10 @@ struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
> (struct bad_iret_stack *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
>
> /* Copy the IRET target to the new stack. */
> - memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
> + __memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
>
> /* Copy the remainder of the stack from the current stack. */
> - memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
> + __memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
Looks like this change should go into a separate patch.
>
> BUG_ON(!user_mode(&new_stack->regs));
> return new_stack;
> diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
> index 8cd745ef8c7b..bcd4bf5a909f 100644
> --- a/arch/x86/kernel/uprobes.c
> +++ b/arch/x86/kernel/uprobes.c
> @@ -8,6 +8,7 @@
> * Jim Keniston
> */
> #include <linux/kernel.h>
> +#include <linux/kmsan-checks.h>
> #include <linux/sched.h>
> #include <linux/ptrace.h>
> #include <linux/uprobes.h>
> @@ -997,9 +998,13 @@ int arch_uprobe_post_xol(struct arch_uprobe *auprobe, struct pt_regs *regs)
> int arch_uprobe_exception_notify(struct notifier_block *self, unsigned long val, void *data)
> {
> struct die_args *args = data;
> - struct pt_regs *regs = args->regs;
> + struct pt_regs *regs;
> int ret = NOTIFY_DONE;
>
> + kmsan_unpoison_shadow(args, sizeof(*args));
> + regs = args->regs;
> + if (regs)
> + kmsan_unpoison_shadow(regs, sizeof(*regs));
This one as well.
> /* We are only interested in userspace traps */
> if (regs && !user_mode(regs))
> return NOTIFY_DONE;
> diff --git a/kernel/profile.c b/kernel/profile.c
> index af7c94bf5fa1..835a5b66d1a4 100644
> --- a/kernel/profile.c
> +++ b/kernel/profile.c
> @@ -399,6 +399,7 @@ void profile_hits(int type, void *__pc, unsigned int nr_hits)
> }
> EXPORT_SYMBOL_GPL(profile_hits);
>
> +__no_sanitize_memory
> void profile_tick(int type)
> {
> struct pt_regs *regs = get_irq_regs();
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index dd05a378631a..674d36fe9d44 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -475,6 +475,7 @@ void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task)
> put_task_struct(task);
> }
>
> +__no_sanitize_memory /* context switching here */
> void wake_up_q(struct wake_q_head *head)
> {
> struct wake_q_node *node = head->first;
> @@ -3180,6 +3181,7 @@ prepare_task_switch(struct rq *rq, struct task_struct *prev,
> * past. prev == current is still correct but we need to recalculate this_rq
> * because prev may have moved to another CPU.
> */
> +__no_sanitize_memory /* |current| changes here */
> static struct rq *finish_task_switch(struct task_struct *prev)
> __releases(rq->lock)
> {
> @@ -3986,6 +3988,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
> *
> * WARNING: must be called with preemption disabled!
> */
> +__no_sanitize_memory /* |current| changes here */
> static void __sched notrace __schedule(bool preempt)
> {
> struct task_struct *prev, *next;
> @@ -4605,6 +4608,7 @@ int task_prio(const struct task_struct *p)
> *
> * Return: 1 if the CPU is currently idle. 0 otherwise.
> */
> +__no_sanitize_memory /* nothing to report here */
> int idle_cpu(int cpu)
> {
> struct rq *rq = cpu_rq(cpu);
> @@ -6544,6 +6548,7 @@ static struct kmem_cache *task_group_cache __read_mostly;
> DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
> DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
>
> +__no_sanitize_memory
> void __init sched_init(void)
> {
> unsigned long ptr = 0;
> @@ -6716,6 +6721,7 @@ static inline int preempt_count_equals(int preempt_offset)
> return (nested == preempt_offset);
> }
>
> +__no_sanitize_memory /* expect the arguments to be initialized */
> void __might_sleep(const char *file, int line, int preempt_offset)
> {
> /*
> --
> 2.24.0.432.g9d3f5f5b63-goog
>
