Re: [PATCH v2 1/5] bpf: Add new bpf map type to store the pointer to struct perf_event

["Wangnan (F)" <wangnan0@huawei.com>]

On 2015/7/23 8:48, Alexei Starovoitov wrote:
> On 7/22/15 1:09 AM, Kaixu Xia wrote:
>> Introduce a new bpf map type 'BPF_MAP_TYPE_PERF_EVENT_ARRAY'.
>> This map will only store the pointer to struct perf_event.
>>
>> Signed-off-by: Kaixu Xia <xiakaixu@huawei.com>
>> ---
>>   include/linux/bpf.h      |  2 ++
>>   include/uapi/linux/bpf.h |  1 +
>>   kernel/bpf/arraymap.c    | 40 ++++++++++++++++++++++++++++++++++++++++
>>   3 files changed, 43 insertions(+)
>>
>> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
>> index 4383476..f6a2442 100644
>> --- a/include/linux/bpf.h
>> +++ b/include/linux/bpf.h
>> @@ -11,6 +11,8 @@
>>   #include <linux/workqueue.h>
>>   #include <linux/file.h>
>>
>> +#define MAX_PERF_EVENT_ARRAY_ENTRY    (2*NR_CPUS)
>
> why this artificial limit?
> Just drop it.
>

Then we should find another way to prevent user create a large map but 
only use
a small portion of it, since normal array map can't report whether a 
slot is used
or not. When releasing the map, we have to release each perf event. 
Currently the only
way is to check map value in each slot.

>> +static struct bpf_map *perf_event_array_map_alloc(union bpf_attr *attr)
>> +{
>> +    /* only the pointer to struct perf_event can be stored in
>> +     * perf_event_array map
>> +     */
>> +    if (attr->value_size != sizeof(void *))
>> +        return ERR_PTR(-EINVAL);
>
> hmm. that's odd. why reinvent things? please do the same as
> prog_array does. Namely below:
>
>> +static const struct bpf_map_ops perf_event_array_ops = {
>> +    .map_alloc = perf_event_array_map_alloc,
>> +    .map_free = array_map_free,
>> +    .map_get_next_key = perf_event_array_map_get_next_key,
>> +    .map_lookup_elem = array_map_lookup_elem,
>> +    .map_delete_elem = array_map_delete_elem,
>
> this is broken. you don't want programs to manipulate
> 'struct perf_event *' pointers.
> lookup/update/delete helpers shouldn't be accessible from the programs
> then update/delete can be cleanly implemented and called via syscall.
> See how prog_array does it.
>
> Also please collapse patches 1-3 into one. Their logically one piece.
> I'll comment on them as well, but it would have been easier for me
> and you if their were part of one email thread.
>