On 2023/8/26 01:58, Maciej Fijalkowski wrote:
> On Fri, Aug 25, 2023 at 10:52:15PM +0800, Leon Hwang wrote:
>> From commit ebf7d1f508a73871 ("bpf, x64: rework pro/epilogue and tailcall
>> handling in JIT"), the tailcall on x64 works better than before.
>>
>> From commit e411901c0b775a3a ("bpf: allow for tailcalls in BPF subprograms
>> for x64 JIT"), tailcall is able to run in BPF subprograms on x64.
>>
>> From commit 5b92a28aae4dd0f8 ("bpf: Support attaching tracing BPF program
>> to other BPF programs"), BPF program is able to trace other BPF programs.
>>
>> How about combining them all together?
>>
>> 1. FENTRY/FEXIT on a BPF subprogram.
>> 2. A tailcall runs in the BPF subprogram.
>> 3. The tailcall calls itself.
>
> I would be interested in seeing broken asm code TBH :)
>
>>
>> As a result, a tailcall infinite loop comes up. And the loop would halt
>> the machine.
>>
>> As we know, in tail call context, the tail_call_cnt propagates by stack
>> and rax register between BPF subprograms. So do it in trampolines.
>>
>> Signed-off-by: Leon Hwang <hffilwlqm@xxxxxxxxx>
>> ---
>> arch/x86/net/bpf_jit_comp.c | 32 ++++++++++++++++++++++++++------
>> include/linux/bpf.h | 5 +++++
>> kernel/bpf/trampoline.c | 4 ++--
>> kernel/bpf/verifier.c | 30 +++++++++++++++++++++++-------
>> 4 files changed, 56 insertions(+), 15 deletions(-)
>>
>> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
>> index a5930042139d3..2846c21d75bfa 100644
>> --- a/arch/x86/net/bpf_jit_comp.c
>> +++ b/arch/x86/net/bpf_jit_comp.c
>> @@ -303,8 +303,12 @@ static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf,
>> prog += X86_PATCH_SIZE;
>> if (!ebpf_from_cbpf) {
>> if (tail_call_reachable && !is_subprog)
>> + /* When it's the entry of the whole tailcall context,
>> + * zeroing rax means initialising tail_call_cnt.
>> + */
>> EMIT2(0x31, 0xC0); /* xor eax, eax */
>> else
>> + /* Keep the same instruction layout. */
>
> While these comments are helpful I have mixed feelings about them residing
> in this patch - rule of thumb to me is to keep the fixes as small as
> possible.
Got it. I'll separate them into another patch.
Thanks for your rule of thumb.
>
>> EMIT2(0x66, 0x90); /* nop2 */
>> }
>> EMIT1(0x55); /* push rbp */
>> @@ -1018,6 +1022,10 @@ static void emit_shiftx(u8 **pprog, u32 dst_reg, u8 src_reg, bool is64, u8 op)
>>
>> #define INSN_SZ_DIFF (((addrs[i] - addrs[i - 1]) - (prog - temp)))
>>
>> +/* mov rax, qword ptr [rbp - rounded_stack_depth - 8] */
>> +#define RESTORE_TAIL_CALL_CNT(stack) \
>> + EMIT3_off32(0x48, 0x8B, 0x85, -round_up(stack, 8) - 8)
>> +
>> static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image, u8 *rw_image,
>> int oldproglen, struct jit_context *ctx, bool jmp_padding)
>> {
>> @@ -1623,9 +1631,7 @@ st: if (is_imm8(insn->off))
>>
>> func = (u8 *) __bpf_call_base + imm32;
>> if (tail_call_reachable) {
>> - /* mov rax, qword ptr [rbp - rounded_stack_depth - 8] */
>> - EMIT3_off32(0x48, 0x8B, 0x85,
>> - -round_up(bpf_prog->aux->stack_depth, 8) - 8);
>> + RESTORE_TAIL_CALL_CNT(bpf_prog->aux->stack_depth);
>> if (!imm32)
>> return -EINVAL;
>> offs = 7 + x86_call_depth_emit_accounting(&prog, func);
>> @@ -2400,6 +2406,7 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>> * [ ... ]
>> * [ stack_arg2 ]
>> * RBP - arg_stack_off [ stack_arg1 ]
>> + * RSP [ tail_call_cnt ] BPF_TRAMP_F_TAIL_CALL_CTX
>> */
>>
>> /* room for return value of orig_call or fentry prog */
>> @@ -2464,6 +2471,8 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>> else
>> /* sub rsp, stack_size */
>> EMIT4(0x48, 0x83, 0xEC, stack_size);
>> + if (flags & BPF_TRAMP_F_TAIL_CALL_CTX)
>> + EMIT1(0x50); /* push rax */
>> /* mov QWORD PTR [rbp - rbx_off], rbx */
>> emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_6, -rbx_off);
>>
>> @@ -2516,9 +2525,15 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>> restore_regs(m, &prog, regs_off);
>> save_args(m, &prog, arg_stack_off, true);
>>
>> + if (flags & BPF_TRAMP_F_TAIL_CALL_CTX)
>> + /* Before calling the original function, restore the
>> + * tail_call_cnt from stack to rax.
>> + */
>> + RESTORE_TAIL_CALL_CNT(stack_size);
>> +
>> if (flags & BPF_TRAMP_F_ORIG_STACK) {
>> - emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, 8);
>> - EMIT2(0xff, 0xd0); /* call *rax */
>> + emit_ldx(&prog, BPF_DW, BPF_REG_6, BPF_REG_FP, 8);
>> + EMIT2(0xff, 0xd3); /* call *rbx */
>> } else {
>> /* call original function */
>> if (emit_rsb_call(&prog, orig_call, prog)) {
>> @@ -2569,7 +2584,12 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>> ret = -EINVAL;
>> goto cleanup;
>> }
>> - }
>> + } else if (flags & BPF_TRAMP_F_TAIL_CALL_CTX)
>> + /* Before running the original function, restore the
>> + * tail_call_cnt from stack to rax.
>> + */
>> + RESTORE_TAIL_CALL_CNT(stack_size);
>> +
>> /* restore return value of orig_call or fentry prog back into RAX */
>> if (save_ret)
>> emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, -8);
>> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
>> index cfabbcf47bdb8..c8df257ea435d 100644
>> --- a/include/linux/bpf.h
>> +++ b/include/linux/bpf.h
>> @@ -1028,6 +1028,11 @@ struct btf_func_model {
>> */
>> #define BPF_TRAMP_F_SHARE_IPMODIFY BIT(6)
>>
>> +/* Indicate that current trampoline is in a tail call context. Then, it has to
>> + * cache and restore tail_call_cnt to avoid infinite tail call loop.
>> + */
>> +#define BPF_TRAMP_F_TAIL_CALL_CTX BIT(7)
>> +
>> /* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50
>> * bytes on x86.
>> */
>> diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
>> index 78acf28d48732..16ab5da7161f2 100644
>> --- a/kernel/bpf/trampoline.c
>> +++ b/kernel/bpf/trampoline.c
>> @@ -415,8 +415,8 @@ static int bpf_trampoline_update(struct bpf_trampoline *tr, bool lock_direct_mut
>> goto out;
>> }
>>
>> - /* clear all bits except SHARE_IPMODIFY */
>> - tr->flags &= BPF_TRAMP_F_SHARE_IPMODIFY;
>> + /* clear all bits except SHARE_IPMODIFY and TAIL_CALL_CTX */
>> + tr->flags &= (BPF_TRAMP_F_SHARE_IPMODIFY | BPF_TRAMP_F_TAIL_CALL_CTX);
>>
>> if (tlinks[BPF_TRAMP_FEXIT].nr_links ||
>> tlinks[BPF_TRAMP_MODIFY_RETURN].nr_links) {
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index 4ccca1f6c9981..6f290bc6f5f19 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
>> @@ -19246,6 +19246,21 @@ static int check_non_sleepable_error_inject(u32 btf_id)
>> return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id);
>> }
>>
>> +static inline int find_subprog_index(const struct bpf_prog *prog,
>
> FWIW please no inlines in source files, but I don't currently follow the
> need for that routine.
Got it. It's unnecessary to inline it.
>
>> + u32 btf_id)
>> +{
>> + struct bpf_prog_aux *aux = prog->aux;
>> + int i, subprog = -1;
>> +
>> + for (i = 0; i < aux->func_info_cnt; i++)
>> + if (aux->func_info[i].type_id == btf_id) {
>> + subprog = i;
>> + break;
>> + }
>> +
>> + return subprog;
>> +}
>> +
>> int bpf_check_attach_target(struct bpf_verifier_log *log,
>> const struct bpf_prog *prog,
>> const struct bpf_prog *tgt_prog,
>> @@ -19254,9 +19269,9 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
>> {
>> bool prog_extension = prog->type == BPF_PROG_TYPE_EXT;
>> const char prefix[] = "btf_trace_";
>> - int ret = 0, subprog = -1, i;
>> const struct btf_type *t;
>> bool conservative = true;
>> + int ret = 0, subprog;
>> const char *tname;
>> struct btf *btf;
>> long addr = 0;
>> @@ -19291,11 +19306,7 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
>> return -EINVAL;
>> }
>>
>> - for (i = 0; i < aux->func_info_cnt; i++)
>> - if (aux->func_info[i].type_id == btf_id) {
>> - subprog = i;
>> - break;
>> - }
>> + subprog = find_subprog_index(tgt_prog, btf_id);
>> if (subprog == -1) {
>> bpf_log(log, "Subprog %s doesn't exist\n", tname);
>> return -EINVAL;
>> @@ -19559,7 +19570,7 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
>> struct bpf_attach_target_info tgt_info = {};
>> u32 btf_id = prog->aux->attach_btf_id;
>> struct bpf_trampoline *tr;
>> - int ret;
>> + int ret, subprog;
>> u64 key;
>>
>> if (prog->type == BPF_PROG_TYPE_SYSCALL) {
>> @@ -19629,6 +19640,11 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
>> if (!tr)
>> return -ENOMEM;
>>
>> + if (tgt_prog && tgt_prog->aux->tail_call_reachable) {
>> + subprog = find_subprog_index(tgt_prog, btf_id);
>> + tr->flags = subprog > 0 ? BPF_TRAMP_F_TAIL_CALL_CTX : 0;
>> + }
>
> I kinda forgot trampoline internals so please bear with me.
> Here you are checking actually...what? That current program is a subprog
> of tgt prog? My knee jerk reaction would be to propagate the
> BPF_TRAMP_F_TAIL_CALL_CTX based on just tail_call_reachable, but I need
> some more time to get my head around it again, sorry :<
Yeah, that current program must be a subprog of tgt prog.
For example:
tailcall_subprog() {
bpf_tail_call_static(&jmp_table, 0);
}
tailcall_prog() {
tailcall_subprog();
}
prog() {
bpf_tail_call_static(&jmp_table, 0);
}
jmp_table populates with tailcall_prog().
When do fentry on prog(), there's no tail_call_cnt for fentry to
propagate. As we can see in emit_prologue(), fentry runs before
initialising tail_call_cnt.
When do fentry on tailcall_prog()? NO, it's impossible to do fentry on
tailcall_prog(). Because the tailcall 'jmp' skips the fentry on
tailcall_prog().
And, when do fentry on tailcall_subprog(), fentry has to propagate
tail_call_cnt properly.
In conclusion, that current program must be a subprog of tgt prog.
Thanks,
Leon
