Date:   Tue, 28 Nov 2023 01:36:54 +0800
From:   Juntong Deng <juntong.deng@...look.com>
To:     Dmitry Vyukov <dvyukov@...gle.com>
Cc:     ryabinin.a.a@...il.com, glider@...gle.com, andreyknvl@...il.com,
        vincenzo.frascino@....com, akpm@...ux-foundation.org,
        kasan-dev@...glegroups.com, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org,
        linux-kernel-mentees@...ts.linuxfoundation.org
Subject: Re: [PATCH v2] kasan: Record and report more information

On 2023/11/27 17:38, Dmitry Vyukov wrote:
> On Mon, 27 Nov 2023 at 10:35, Juntong Deng <juntong.deng@...look.com> wrote:
>>
>> On 2023/11/27 12:34, Dmitry Vyukov wrote:
>>> On Sun, 26 Nov 2023 at 23:25, Juntong Deng <juntong.deng@...look.com> wrote:
>>>>
>>>> Record and report more information to help us find the cause of the
>>>> bug and to help us correlate the error with other system events.
>>>>
>>>> This patch adds recording and showing CPU number and timestamp at
>>>> allocation and free (controlled by CONFIG_KASAN_EXTRA_INFO). The
>>>> timestamps in the report use the same format and source as printk.
>>>>
>>>> Error occurrence timestamp is already implicit in the printk log,
>>>> and CPU number is already shown by dump_stack_lvl, so there is no
>>>> need to add it.
>>>>
>>>> In order to record CPU number and timestamp at allocation and free,
>>>> corresponding members need to be added to the relevant data structures,
>>>> which will lead to increased memory consumption.
>>>>
>>>> In Generic KASAN, members are added to struct kasan_track. Since in
>>>> most cases, alloc meta is stored in the redzone and free meta is
>>>> stored in the object or the redzone, memory consumption will not
>>>> increase much.
>>>>
>>>> In SW_TAGS KASAN and HW_TAGS KASAN, members are added to
>>>> struct kasan_stack_ring_entry. Memory consumption increases as the
>>>> size of struct kasan_stack_ring_entry increases (this part of the
>>>> memory is allocated by memblock), but since this is configurable,
>>>> it is up to the user to choose.
>>>>
>>>> Signed-off-by: Juntong Deng <juntong.deng@...look.com>
>>>> ---
>>>> V1 -> V2: Use bit field to reduce memory consumption. Add more detailed
>>>> config help. Cancel printing of redundant error occurrence timestamp.
>>>>
>>>>    lib/Kconfig.kasan      | 21 +++++++++++++++++++++
>>>>    mm/kasan/common.c      | 10 ++++++++++
>>>>    mm/kasan/kasan.h       | 10 ++++++++++
>>>>    mm/kasan/report.c      |  6 ++++++
>>>>    mm/kasan/report_tags.c | 16 ++++++++++++++++
>>>>    mm/kasan/tags.c        | 17 +++++++++++++++++
>>>>    6 files changed, 80 insertions(+)
>>>>
>>>> diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
>>>> index 935eda08b1e1..8653f5c38be7 100644
>>>> --- a/lib/Kconfig.kasan
>>>> +++ b/lib/Kconfig.kasan
>>>> @@ -207,4 +207,25 @@ config KASAN_MODULE_TEST
>>>>             A part of the KASAN test suite that is not integrated with KUnit.
>>>>             Incompatible with Hardware Tag-Based KASAN.
>>>>
>>>> +config KASAN_EXTRA_INFO
>>>> +       bool "Record and report more information"
>>>> +       depends on KASAN
>>>> +       help
>>>> +         Record and report more information to help us find the cause of the
>>>> +         bug and to help us correlate the error with other system events.
>>>> +
>>>> +         Currently, the CPU number and timestamp are additionally
>>>> +         recorded for each heap block at allocation and free time, and
>>>> +         8 bytes will be added to each metadata structure that records
>>>> +         allocation or free information.
>>>> +
>>>> +         In Generic KASAN, each kmalloc-8 and kmalloc-16 object will add
>>>> +         16 bytes of additional memory consumption, and each kmalloc-32
>>>> +         object will add 8 bytes of additional memory consumption, not
>>>> +         affecting other larger objects.
>>>> +
>>>> +         In SW_TAGS KASAN and HW_TAGS KASAN, depending on the stack_ring_size
>>>> +         boot parameter, it will add 8 * stack_ring_size bytes of additional
>>>> +         memory consumption.
>>>> +
>>>>    endif # KASAN
>>>> diff --git a/mm/kasan/common.c b/mm/kasan/common.c
>>>> index b5d8bd26fced..2f0884c762b7 100644
>>>> --- a/mm/kasan/common.c
>>>> +++ b/mm/kasan/common.c
>>>> @@ -20,6 +20,7 @@
>>>>    #include <linux/module.h>
>>>>    #include <linux/printk.h>
>>>>    #include <linux/sched.h>
>>>> +#include <linux/sched/clock.h>
>>>>    #include <linux/sched/task_stack.h>
>>>>    #include <linux/slab.h>
>>>>    #include <linux/stackdepot.h>
>>>> @@ -49,6 +50,15 @@ depot_stack_handle_t kasan_save_stack(gfp_t flags, depot_flags_t depot_flags)
>>>>
>>>>    void kasan_set_track(struct kasan_track *track, gfp_t flags)
>>>>    {
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +       u32 cpu = raw_smp_processor_id();
>>>> +       u64 ts_nsec = local_clock();
>>>> +       unsigned long rem_usec = do_div(ts_nsec, NSEC_PER_SEC) / 1000;
>>>> +
>>>> +       track->cpu = cpu;
>>>> +       track->ts_sec = ts_nsec;
>>>> +       track->ts_usec = rem_usec;
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>           track->pid = current->pid;
>>>>           track->stack = kasan_save_stack(flags,
>>>>                           STACK_DEPOT_FLAG_CAN_ALLOC | STACK_DEPOT_FLAG_GET);
>>>> diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
>>>> index b29d46b83d1f..2a37baa4ce2f 100644
>>>> --- a/mm/kasan/kasan.h
>>>> +++ b/mm/kasan/kasan.h
>>>> @@ -187,6 +187,11 @@ static inline bool kasan_requires_meta(void)
>>>>    struct kasan_track {
>>>>           u32 pid;
>>>>           depot_stack_handle_t stack;
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +       u64 cpu:20;
>>>> +       u64 ts_sec:22;
>>>> +       u64 ts_usec:22;
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>    };
>>>>
>>>>    enum kasan_report_type {
>>>> @@ -278,6 +283,11 @@ struct kasan_stack_ring_entry {
>>>>           u32 pid;
>>>>           depot_stack_handle_t stack;
>>>>           bool is_free;
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +       u64 cpu:20;
>>>> +       u64 ts_sec:22;
>>>> +       u64 ts_usec:22;
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>    };
>>>>
>>>>    struct kasan_stack_ring {
>>>> diff --git a/mm/kasan/report.c b/mm/kasan/report.c
>>>> index e77facb62900..8cd8f6e5cf24 100644
>>>> --- a/mm/kasan/report.c
>>>> +++ b/mm/kasan/report.c
>>>> @@ -262,7 +262,13 @@ static void print_error_description(struct kasan_report_info *info)
>>>>
>>>>    static void print_track(struct kasan_track *track, const char *prefix)
>>>>    {
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +       pr_err("%s by task %u on cpu %d at %u.%06us:\n",
>>>> +                       prefix, track->pid, track->cpu,
>>>> +                       track->ts_sec, track->ts_usec);
>>>> +#else
>>>>           pr_err("%s by task %u:\n", prefix, track->pid);
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>           if (track->stack)
>>>>                   stack_depot_print(track->stack);
>>>>           else
>>>> diff --git a/mm/kasan/report_tags.c b/mm/kasan/report_tags.c
>>>> index 55154743f915..bf895b1d2dc2 100644
>>>> --- a/mm/kasan/report_tags.c
>>>> +++ b/mm/kasan/report_tags.c
>>>> @@ -27,6 +27,16 @@ static const char *get_common_bug_type(struct kasan_report_info *info)
>>>>           return "invalid-access";
>>>>    }
>>>>
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +static void kasan_complete_extra_report_info(struct kasan_track *track,
>>>> +                                        struct kasan_stack_ring_entry *entry)
>>>> +{
>>>> +       track->cpu = entry->cpu;
>>>> +       track->ts_sec = entry->ts_sec;
>>>> +       track->ts_usec = entry->ts_usec;
>>>> +}
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>> +
>>>>    void kasan_complete_mode_report_info(struct kasan_report_info *info)
>>>>    {
>>>>           unsigned long flags;
>>>> @@ -73,6 +83,9 @@ void kasan_complete_mode_report_info(struct kasan_report_info *info)
>>>>
>>>>                           info->free_track.pid = entry->pid;
>>>>                           info->free_track.stack = entry->stack;
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +                       kasan_complete_extra_report_info(&info->free_track, entry);
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>                           free_found = true;
>>>>
>>>>                           /*
>>>> @@ -88,6 +101,9 @@ void kasan_complete_mode_report_info(struct kasan_report_info *info)
>>>>
>>>>                           info->alloc_track.pid = entry->pid;
>>>>                           info->alloc_track.stack = entry->stack;
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +                       kasan_complete_extra_report_info(&info->alloc_track, entry);
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>                           alloc_found = true;
>>>>
>>>>                           /*
>>>> diff --git a/mm/kasan/tags.c b/mm/kasan/tags.c
>>>> index 739ae997463d..c172e115b9bb 100644
>>>> --- a/mm/kasan/tags.c
>>>> +++ b/mm/kasan/tags.c
>>>> @@ -13,6 +13,7 @@
>>>>    #include <linux/memblock.h>
>>>>    #include <linux/memory.h>
>>>>    #include <linux/mm.h>
>>>> +#include <linux/sched/clock.h>
>>>>    #include <linux/stackdepot.h>
>>>>    #include <linux/static_key.h>
>>>>    #include <linux/string.h>
>>>> @@ -93,6 +94,19 @@ void __init kasan_init_tags(void)
>>>>           }
>>>>    }
>>>>
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +static void save_extra_info(struct kasan_stack_ring_entry *entry)
>>>> +{
>>>> +       u32 cpu = raw_smp_processor_id();
>>>> +       u64 ts_nsec = local_clock();
>>>> +       unsigned long rem_usec = do_div(ts_nsec, NSEC_PER_SEC) / 1000;
>>>> +
>>>> +       entry->cpu = cpu;
>>>> +       entry->ts_sec = ts_nsec;
>>>> +       entry->ts_usec = rem_usec;
>>>
>>> I would timestamp as a single field in all structs and convert it to
>>> sec/usec only when we print it. It would make all initialization and
>>> copying shorter. E.g. this function can be just:
>>>
>>>          entry->cpu = raw_smp_processor_id();
>>>          entry->timestamp = local_clock() / 1024;
>>>
>>> Dividing by 1024 is much faster and gives roughly the same precision.
>>> This can be unscaled during reporting:
>>>
>>>          u64 sec = entry->timestamp * 1024;
>>>          unsigned long usec = do_div(sec, NSEC_PER_SEC) / 1000;
>>>
>>> But otherwise the patch looks good to me.
>>>
>>> Reviewed-by: Dmitry Vyukov <dvyukov@...gle.com>
>>>
>>
>>
>> I think it would be better to use left shift and right shift because
>> dropping the last 3 bits would not affect the microsecond part and
>> would not affect the precision at all.
>>
>> In addition, 44 bits are enough to store the maximum value of the
>> displayable time 99999.999999 (5-bit seconds + 6-bit microseconds).
>>
>> 010110101111001100010000011110100011111111111111 (99999.999999) >> 3
>> = 10110101111001100010000011110100011111111111 (44 bits)
>>
>> I will send the V3 patch.
> 
> Agree.
> Modern compilers are smart enough to turn division/multiplication by
> pow-2 const into necessary shift, so we may not obfuscate the code.
> 


In my actual tests, right/left shifting by 3 bits gives a different
result than dividing/multiplying by 1024.

Right/left shifting by 3 bits did not cause loss of precision,
but dividing/multiplying by 1024 did.

I think the compiler did not convert that part of the code very well.

I think using bit shift would be a better option.


> 
>>>> +}
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>> +
>>>>    static void save_stack_info(struct kmem_cache *cache, void *object,
>>>>                           gfp_t gfp_flags, bool is_free)
>>>>    {
>>>> @@ -128,6 +142,9 @@ static void save_stack_info(struct kmem_cache *cache, void *object,
>>>>           entry->pid = current->pid;
>>>>           entry->stack = stack;
>>>>           entry->is_free = is_free;
>>>> +#ifdef CONFIG_KASAN_EXTRA_INFO
>>>> +       save_extra_info(entry);
>>>> +#endif /* CONFIG_KASAN_EXTRA_INFO */
>>>>
>>>>           entry->ptr = object;
>>>>
>>>> --
>>>> 2.39.2
>>>>
>>

Powered by blists - more mailing lists