From: Yonghong Song <yhs@fb.com>
To: Kumar Kartikeya Dwivedi <memxor@gmail.com>,
Alexei Starovoitov <ast@fb.com>
Cc: Dave Marchevsky <davemarchevsky@fb.com>,
bpf@vger.kernel.org, Alexei Starovoitov <ast@kernel.org>,
Daniel Borkmann <daniel@iogearbox.net>,
Andrii Nakryiko <andrii@kernel.org>,
Kernel Team <kernel-team@fb.com>, Tejun Heo <tj@kernel.org>
Subject: Re: [RFC PATCH bpf-next 05/11] bpf: Add bpf_spin_lock member to rbtree
Date: Sun, 14 Aug 2022 22:33:59 -0700 [thread overview]
Message-ID: <18a8e565-95d0-80e1-b596-95babf279912@fb.com> (raw)
In-Reply-To: <CAP01T75nt69=jgGPGXYXHSGc5EDHejgLQpyY8TMeUy2U4Prxvg@mail.gmail.com>
On 8/10/22 2:46 PM, Kumar Kartikeya Dwivedi wrote:
> On Tue, 2 Aug 2022 at 00:23, Alexei Starovoitov <ast@fb.com> wrote:
>>
>> On 7/22/22 11:34 AM, Dave Marchevsky wrote:
>>> This patch adds a struct bpf_spin_lock *lock member to bpf_rbtree, as
>>> well as a bpf_rbtree_get_lock helper which allows bpf programs to access
>>> the lock.
>>>
>>> Ideally the bpf_spin_lock would be created independently oustide of the
>>> tree and associated with it before the tree is used, either as part of
>>> map definition or via some call like rbtree_init(&rbtree, &lock). Doing
>>> this in an ergonomic way is proving harder than expected, so for now use
>>> this workaround.
>>>
>>> Why is creating the bpf_spin_lock independently and associating it with
>>> the tree preferable? Because we want to be able to transfer nodes
>>> between trees atomically, and for this to work need same lock associated
>>> with 2 trees.
>>
>> Right. We need one lock to protect multiple rbtrees.
>> Since add/find/remove helpers will look into rbtree->lock
>> the two different rbtree (== map) have to point to the same lock.
>> Other than rbtree_init(&rbtree, &lock) what would be an alternative ?
>>
>>>
>>> Further locking-related patches will make it possible for the lock to be
>>> used in BPF programs and add code which enforces that the lock is held
>>> when doing any operation on the tree.
>>>
>>> Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
>>> ---
>>> include/uapi/linux/bpf.h | 7 +++++++
>>> kernel/bpf/helpers.c | 3 +++
>>> kernel/bpf/rbtree.c | 24 ++++++++++++++++++++++++
>>> tools/include/uapi/linux/bpf.h | 7 +++++++
>>> 4 files changed, 41 insertions(+)
>>>
>>> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
>>> index 4688ce88caf4..c677d92de3bc 100644
>>> --- a/include/uapi/linux/bpf.h
>>> +++ b/include/uapi/linux/bpf.h
>>> @@ -5385,6 +5385,12 @@ union bpf_attr {
>>> * Return
>>> * 0
>>> *
>>> + * void *bpf_rbtree_get_lock(struct bpf_map *map)
>>> + * Description
>>> + * Return the bpf_spin_lock associated with the rbtree
>>> + *
>>> + * Return
>>> + * Ptr to lock
>>> */
>>> #define __BPF_FUNC_MAPPER(FN) \
>>> FN(unspec), \
>>> @@ -5600,6 +5606,7 @@ union bpf_attr {
>>> FN(rbtree_find), \
>>> FN(rbtree_remove), \
>>> FN(rbtree_free_node), \
>>> + FN(rbtree_get_lock), \
>>> /* */
>>>
>>> /* integer value in 'imm' field of BPF_CALL instruction selects which helper
>>> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
>>> index 35eb66d11bf6..257a808bb767 100644
>>> --- a/kernel/bpf/helpers.c
>>> +++ b/kernel/bpf/helpers.c
>>> @@ -1587,6 +1587,7 @@ const struct bpf_func_proto bpf_rbtree_add_proto __weak;
>>> const struct bpf_func_proto bpf_rbtree_find_proto __weak;
>>> const struct bpf_func_proto bpf_rbtree_remove_proto __weak;
>>> const struct bpf_func_proto bpf_rbtree_free_node_proto __weak;
>>> +const struct bpf_func_proto bpf_rbtree_get_lock_proto __weak;
>>>
>>> const struct bpf_func_proto *
>>> bpf_base_func_proto(enum bpf_func_id func_id)
>>> @@ -1686,6 +1687,8 @@ bpf_base_func_proto(enum bpf_func_id func_id)
>>> return &bpf_rbtree_remove_proto;
>>> case BPF_FUNC_rbtree_free_node:
>>> return &bpf_rbtree_free_node_proto;
>>> + case BPF_FUNC_rbtree_get_lock:
>>> + return &bpf_rbtree_get_lock_proto;
>>> default:
>>> break;
>>> }
>>> diff --git a/kernel/bpf/rbtree.c b/kernel/bpf/rbtree.c
>>> index 250d62210804..c6f0a2a083f6 100644
>>> --- a/kernel/bpf/rbtree.c
>>> +++ b/kernel/bpf/rbtree.c
>>> @@ -9,6 +9,7 @@
>>> struct bpf_rbtree {
>>> struct bpf_map map;
>>> struct rb_root_cached root;
>>> + struct bpf_spin_lock *lock;
>>> };
>>>
>>> BTF_ID_LIST_SINGLE(bpf_rbtree_btf_ids, struct, rb_node);
>>> @@ -39,6 +40,14 @@ static struct bpf_map *rbtree_map_alloc(union bpf_attr *attr)
>>>
>>> tree->root = RB_ROOT_CACHED;
>>> bpf_map_init_from_attr(&tree->map, attr);
>>> +
>>> + tree->lock = bpf_map_kzalloc(&tree->map, sizeof(struct bpf_spin_lock),
>>> + GFP_KERNEL | __GFP_NOWARN);
>>> + if (!tree->lock) {
>>> + bpf_map_area_free(tree);
>>> + return ERR_PTR(-ENOMEM);
>>> + }
>>> +
>>> return &tree->map;
>>> }
>>>
>>> @@ -139,6 +148,7 @@ static void rbtree_map_free(struct bpf_map *map)
>>>
>>> bpf_rbtree_postorder_for_each_entry_safe(pos, n, &tree->root.rb_root)
>>> kfree(pos);
>>> + kfree(tree->lock);
>>> bpf_map_area_free(tree);
>>> }
>>>
>>> @@ -238,6 +248,20 @@ static int rbtree_map_get_next_key(struct bpf_map *map, void *key,
>>> return -ENOTSUPP;
>>> }
>>>
>>> +BPF_CALL_1(bpf_rbtree_get_lock, struct bpf_map *, map)
>>> +{
>>> + struct bpf_rbtree *tree = container_of(map, struct bpf_rbtree, map);
>>> +
>>> + return (u64)tree->lock;
>>> +}
>>> +
>>> +const struct bpf_func_proto bpf_rbtree_get_lock_proto = {
>>> + .func = bpf_rbtree_get_lock,
>>> + .gpl_only = true,
>>> + .ret_type = RET_PTR_TO_MAP_VALUE,
>>
>> This hack and
>>
>> +const struct bpf_func_proto bpf_rbtree_unlock_proto = {
>> + .func = bpf_rbtree_unlock,
>> + .gpl_only = true,
>> + .ret_type = RET_INTEGER,
>> + .arg1_type = ARG_PTR_TO_SPIN_LOCK,
>>
>> may be too much arm twisting to reuse bpf_spin_lock.
>>
>> May be return ptr_to_btf_id here and bpf_rbtree_lock
>> should match the type?
>> It could be new 'struct bpf_lock' in kernel's BTF.
>>
>> Let's figure out how to associate locks with rbtrees.
>>
>> Reiterating my proposal that was done earlier in the context
>> of Delyan's irq_work, but for different type:
>> How about:
>> struct bpf_lock *l;
>> l = bpf_mem_alloc(allocator, bpf_core_type_id_kernel(*l));
>>
>> that would allocate ptr_to_btf_id object from kernel's btf.
>> The bpf_lock would have constructor and destructor provided by the
>> kernel code.
>> constructor will set bpf_lock's refcnt to 1.
>> then bpf_rbtree_init(&map, lock) will bump the refcnt.
>> and dtor will eventually free it when all rbtrees are freed.
>> That would be similar to kptr's logic with kptr_get and dtor's
>> associated with kernel's btf_id-s.
>
> Just to continue brainstorming: Comments on this?
>
> Instead of a rbtree map, you have a struct bpf_rbtree global variable
> which works like a rbtree. To associate a lock with multiple
> bpf_rbtree, you do clang style thread safety annotation in the bpf
> program:
>
> #define __guarded_by(lock) __attribute__((btf_type_tag("guarded_by:" #lock))
>
> struct bpf_spin_lock shared_lock;
> struct bpf_rbtree rbtree1 __guarded_by(shared_lock);
> struct bpf_rbtree rbtree2 __guarded_by(shared_lock);
For the above __guarded_by macro, we should use
btf_decl_tag instead of btf_type_tag
#define __guarded_by(lock) __attribute__((btf_decl_tag("guarded_by:" #lock))
Currently, in llvm implementation, btf_type_tag only applies
to pointee type's. btf_decl_tag can apply to global variable,
function argument, function return value and struct/union members.
So btf_decl_tag shoul work for the above global variable case or
below struct rbtree_set member case.
>
> guarded_by tag is mandatory for the rbtree. Verifier ensures
> shared_lock spin lock is held whenever rbtree1 or rbtree2 is being
> accessed, and whitelists add/remove helpers inside the critical
> section.
>
> I don't think associating locks to rbtree dynamically is a hard
> requirement for your use case? But if you need that, you may probably
> also allocate sets of rbtree that are part of the same lock "class"
> dynamically using bpf_kptr_alloc, and do similar annotation for the
> struct being allocated.
> struct rbtree_set {
> struct bpf_spin_lock lock;
> struct bpf_rbtree rbtree1 __guarded_by(lock);
> struct bpf_rbtree rbtree2 __guarded_by(lock);
> };
> struct rbtree_set *s = bpf_kptr_alloc(sizeof(*s), btf_local_type_id(*s));
> // Just stash the pointer somewhere with kptr_xchg
> On bpf_kptr_free, the verifier knows this is not a "trivial" struct,
> so inserts destructor calls for bpf_rbtree fields during program
> fixup.
>
> The main insight is that lock and rbtree are part of the same
> allocation (map value for global case, ptr_to_btf_id for dynamic case)
> so the locked state can be bound to the reg state in the verifier.
> Then we can also make this new rbtree API use kfuncs instead of UAPI
> helpers, to get some field experience before baking it in.
>
> Any opinions? Any brainos or deficiencies in the scheme above?
>
> Background: I have been thinking about how I can bind kptr and normal
> data synchronization without having unneeded atomic xchg cost when
> lock is already protecting kptr. In my tests this guarded_by
> annotation has been proving very useful (you mark data and kptr
> protected by lock as guarded_by some spin lock member in same map
> value, verifier ensures lock is held during access, and kptr_xchg for
> guarded kptr is lowered to non-atomic load/store assuming no
> concurrent access, and kptr_xchg from outside the lock section is
> rejected).
next prev parent reply other threads:[~2022-08-15 5:34 UTC|newest]
Thread overview: 37+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-22 18:34 [RFC PATCH bpf-next 00/11] bpf: Introduce rbtree map Dave Marchevsky
2022-07-22 18:34 ` [RFC PATCH bpf-next 01/11] bpf: Pull repeated reg access bounds check into helper fn Dave Marchevsky
2022-07-22 18:34 ` [RFC PATCH bpf-next 02/11] bpf: Add verifier support for custom callback return range Dave Marchevsky
2022-07-22 18:34 ` [RFC PATCH bpf-next 03/11] bpf: Add rb_node_off to bpf_map Dave Marchevsky
2022-08-01 22:19 ` Alexei Starovoitov
2022-07-22 18:34 ` [RFC PATCH bpf-next 04/11] bpf: Add rbtree map Dave Marchevsky
2022-08-01 21:49 ` Alexei Starovoitov
2022-07-22 18:34 ` [RFC PATCH bpf-next 05/11] bpf: Add bpf_spin_lock member to rbtree Dave Marchevsky
2022-08-01 22:17 ` Alexei Starovoitov
2022-08-02 13:59 ` Kumar Kartikeya Dwivedi
2022-08-02 15:30 ` Alexei Starovoitov
2022-08-10 21:46 ` Kumar Kartikeya Dwivedi
2022-08-10 22:06 ` Alexei Starovoitov
2022-08-10 23:16 ` Kumar Kartikeya Dwivedi
2022-08-15 5:33 ` Yonghong Song [this message]
2022-08-15 5:37 ` Kumar Kartikeya Dwivedi
2022-07-22 18:34 ` [RFC PATCH bpf-next 06/11] bpf: Add bpf_rbtree_{lock,unlock} helpers Dave Marchevsky
2022-08-01 21:58 ` Alexei Starovoitov
2022-07-22 18:34 ` [RFC PATCH bpf-next 07/11] bpf: Enforce spinlock hold for bpf_rbtree_{add,remove,find} Dave Marchevsky
2022-07-22 18:34 ` [RFC PATCH bpf-next 08/11] bpf: Add OBJ_NON_OWNING_REF type flag Dave Marchevsky
2022-08-01 22:41 ` Alexei Starovoitov
2022-07-22 18:34 ` [RFC PATCH bpf-next 09/11] bpf: Add CONDITIONAL_RELEASE " Dave Marchevsky
2022-08-01 22:23 ` Alexei Starovoitov
2022-07-22 18:34 ` [RFC PATCH bpf-next 10/11] bpf: Introduce PTR_ITER and PTR_ITER_END type flags Dave Marchevsky
2022-07-29 16:31 ` Tejun Heo
2022-08-01 22:44 ` Alexei Starovoitov
2022-08-02 13:05 ` Kumar Kartikeya Dwivedi
2022-08-02 15:10 ` Alexei Starovoitov
2022-08-10 17:56 ` Dave Marchevsky
2022-07-22 18:34 ` [RFC PATCH bpf-next 11/11] selftests/bpf: Add rbtree map tests Dave Marchevsky
2022-07-28 7:18 ` Yonghong Song
2022-08-10 17:48 ` Dave Marchevsky
2022-07-28 7:04 ` [RFC PATCH bpf-next 00/11] bpf: Introduce rbtree map Yonghong Song
2022-08-10 17:54 ` Dave Marchevsky
2022-08-01 21:27 ` Alexei Starovoitov
2022-08-10 18:11 ` Dave Marchevsky
2022-08-02 22:02 ` Andrii Nakryiko
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=18a8e565-95d0-80e1-b596-95babf279912@fb.com \
--to=yhs@fb.com \
--cc=andrii@kernel.org \
--cc=ast@fb.com \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=davemarchevsky@fb.com \
--cc=kernel-team@fb.com \
--cc=memxor@gmail.com \
--cc=tj@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).