lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Fri, 15 Jan 2021 23:34:33 +0000
From:   Song Liu <songliubraving@...com>
To:     KP Singh <kpsingh@...nel.org>
CC:     Yonghong Song <yhs@...com>, Martin Lau <kafai@...com>,
        bpf <bpf@...r.kernel.org>, Networking <netdev@...r.kernel.org>,
        open list <linux-kernel@...r.kernel.org>,
        "mingo@...hat.com" <mingo@...hat.com>,
        "Peter Zijlstra" <peterz@...radead.org>,
        Alexei Starovoitov <ast@...nel.org>,
        "Daniel Borkmann" <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        "John Fastabend" <john.fastabend@...il.com>,
        Kernel Team <Kernel-team@...com>,
        "Hao Luo" <haoluo@...gle.com>, kernel test robot <lkp@...el.com>
Subject: Re: [PATCH bpf-next 1/4] bpf: enable task local storage for tracing
 programs



> On Jan 12, 2021, at 8:53 AM, KP Singh <kpsingh@...nel.org> wrote:
> 
> On Tue, Jan 12, 2021 at 5:32 PM Yonghong Song <yhs@...com> wrote:
>> 
>> 
>> 
>> On 1/11/21 3:45 PM, Song Liu wrote:
>>> 
>>> 
>>>> On Jan 11, 2021, at 1:58 PM, Martin Lau <kafai@...com> wrote:
>>>> 
>>>> On Mon, Jan 11, 2021 at 10:35:43PM +0100, KP Singh wrote:
>>>>> On Mon, Jan 11, 2021 at 7:57 PM Martin KaFai Lau <kafai@...com> wrote:
>>>>>> 
>>>>>> On Fri, Jan 08, 2021 at 03:19:47PM -0800, Song Liu wrote:
>>>>>> 
>>>>>> [ ... ]
>>>>>> 
>>>>>>> diff --git a/kernel/bpf/bpf_local_storage.c b/kernel/bpf/bpf_local_storage.c
>>>>>>> index dd5aedee99e73..9bd47ad2b26f1 100644
>>>>>>> --- a/kernel/bpf/bpf_local_storage.c
>>>>>>> +++ b/kernel/bpf/bpf_local_storage.c
> 
> [...]
> 
>>>>>>> +#include <linux/bpf.h>
>>>>>>> 
>>>>>>> #include <asm/pgalloc.h>
>>>>>>> #include <linux/uaccess.h>
>>>>>>> @@ -734,6 +735,7 @@ void __put_task_struct(struct task_struct *tsk)
>>>>>>>      cgroup_free(tsk);
>>>>>>>      task_numa_free(tsk, true);
>>>>>>>      security_task_free(tsk);
>>>>>>> +     bpf_task_storage_free(tsk);
>>>>>>>      exit_creds(tsk);
>>>>>> If exit_creds() is traced by a bpf and this bpf is doing
>>>>>> bpf_task_storage_get(..., BPF_LOCAL_STORAGE_GET_F_CREATE),
>>>>>> new task storage will be created after bpf_task_storage_free().
>>>>>> 
>>>>>> I recalled there was an earlier discussion with KP and KP mentioned
>>>>>> BPF_LSM will not be called with a task that is going away.
>>>>>> It seems enabling bpf task storage in bpf tracing will break
>>>>>> this assumption and needs to be addressed?
>>>>> 
>>>>> For tracing programs, I think we will need an allow list where
>>>>> task local storage can be used.
>>>> Instead of whitelist, can refcount_inc_not_zero(&tsk->usage) be used?
>>> 
>>> I think we can put refcount_inc_not_zero() in bpf_task_storage_get, like:
>>> 
>>> diff --git i/kernel/bpf/bpf_task_storage.c w/kernel/bpf/bpf_task_storage.c
>>> index f654b56907b69..93d01b0a010e6 100644
>>> --- i/kernel/bpf/bpf_task_storage.c
>>> +++ w/kernel/bpf/bpf_task_storage.c
>>> @@ -216,6 +216,9 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
>>>          * by an RCU read-side critical section.
>>>          */
>>>         if (flags & BPF_LOCAL_STORAGE_GET_F_CREATE) {
>>> +               if (!refcount_inc_not_zero(&task->usage))
>>> +                       return -EBUSY;
>>> +
>>>                 sdata = bpf_local_storage_update(
>>>                         task, (struct bpf_local_storage_map *)map, value,
>>>                         BPF_NOEXIST);
>>> 
>>> But where shall we add the refcount_dec()? IIUC, we cannot add it to
>>> __put_task_struct().
>> 
>> Maybe put_task_struct()?
> 
> Yeah, something like, or if you find a more elegant alternative :)
> 
> --- a/include/linux/sched/task.h
> +++ b/include/linux/sched/task.h
> @@ -107,13 +107,20 @@ extern void __put_task_struct(struct task_struct *t);
> 
> static inline void put_task_struct(struct task_struct *t)
> {
> -       if (refcount_dec_and_test(&t->usage))
> +
> +       if (rcu_access_pointer(t->bpf_storage)) {
> +               if (refcount_sub_and_test(2, &t->usage))
> +                       __put_task_struct(t);
> +       } else if (refcount_dec_and_test(&t->usage))
>                __put_task_struct(t);
> }
> 
> static inline void put_task_struct_many(struct task_struct *t, int nr)
> {
> -       if (refcount_sub_and_test(nr, &t->usage))
> +       if (rcu_access_pointer(t->bpf_storage)) {
> +               if (refcount_sub_and_test(nr + 1, &t->usage))
> +                       __put_task_struct(t);
> +       } else if (refcount_sub_and_test(nr, &t->usage))
>                __put_task_struct(t);
> }

It is not ideal to leak bpf_storage here. How about we only add the
following:

diff --git i/kernel/bpf/bpf_task_storage.c w/kernel/bpf/bpf_task_storage.c
index f654b56907b69..2811b9fc47233 100644
--- i/kernel/bpf/bpf_task_storage.c
+++ w/kernel/bpf/bpf_task_storage.c
@@ -216,6 +216,10 @@ BPF_CALL_4(bpf_task_storage_get, struct bpf_map *, map, struct task_struct *,
         * by an RCU read-side critical section.
         */
        if (flags & BPF_LOCAL_STORAGE_GET_F_CREATE) {
+               /* the task_struct is being freed, fail over*/
+               if (!refcount_read(&task->usage))
+                       return -EBUSY;
+
                sdata = bpf_local_storage_update(
                        task, (struct bpf_local_storage_map *)map, value,
                        BPF_NOEXIST);

> 
> 
> I may be missing something but shouldn't bpf_storage be an __rcu
> member like we have for sk_bpf_storage?

Good catch! I will fix this in v2. 

Thanks,
Song

Powered by blists - more mailing lists