| Message ID | 20230814134147.70289-2-hffilwlqm@gmail.com (mailing list archive) |
|---|---|
| State | RFC |
| Delegated to: | BPF |
| Headers | show |
| Series | bpf, x64: Fix tailcall infinite loop bug | expand |
On Mon, 2023-08-14 at 21:41 +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. > > As a result, a tailcall infinite loop comes up. And the loop halts 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 FENTRY/FEXIT > trampolines. Hi Leon, I'm not familiar with this part of the jit compiler, so decided that taking a look at your series might be a good learning point. I think I got the gist of it, but I don't understand where the initial value of RAX (== 0) is coming from in arch_prepare_bpf_trampoline(), could you please help me out? Also a nitpick: - in arch_prepare_bpf_trampoline() there is a comment detailing the stack layout, it probably should be updated to say that tail call count is stored as well; - before arch_prepare_bpf_trampoline() there is a comment with an example of generated assembly, should it be updated? Thanks, Eduard > > Fixes: ebf7d1f508a7 ("bpf, x64: rework pro/epilogue and tailcall handling in JIT") > Fixes: e411901c0b77 ("bpf: allow for tailcalls in BPF subprograms for x64 JIT") > Signed-off-by: Leon Hwang <hffilwlqm@gmail.com> > --- > arch/x86/net/bpf_jit_comp.c | 23 +++++++++++++++++++---- > include/linux/bpf.h | 6 ++++++ > kernel/bpf/trampoline.c | 5 +++-- > kernel/bpf/verifier.c | 9 +++++++-- > 4 files changed, 35 insertions(+), 8 deletions(-) > > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c > index a5930042139d3..ca5366d97ad04 100644 > --- a/arch/x86/net/bpf_jit_comp.c > +++ b/arch/x86/net/bpf_jit_comp.c > @@ -1018,6 +1018,10 @@ 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 +1627,7 @@ 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); > @@ -2464,6 +2466,8 @@ 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); > > @@ -2516,6 +2520,12 @@ 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. > + */ > + 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 */ > @@ -2569,7 +2579,12 @@ 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. > + */ > + 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..55c72086034ef 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. > */ > @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { > struct module *tgt_mod; > const char *tgt_name; > const struct btf_type *tgt_type; > + bool tail_call_ctx; > }; > > #define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ > diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c > index 78acf28d48732..0fae334e3f7b8 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) { > @@ -783,6 +783,7 @@ struct bpf_trampoline *bpf_trampoline_get(u64 key, > > memcpy(&tr->func.model, &tgt_info->fmodel, sizeof(tgt_info->fmodel)); > tr->func.addr = (void *)tgt_info->tgt_addr; > + tr->flags = (tgt_info->tail_call_ctx ? BPF_TRAMP_F_TAIL_CALL_CTX : 0); > out: > mutex_unlock(&tr->mutex); > return tr; > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > index 4ccca1f6c9981..a78e5a2ae5c72 100644 > --- a/kernel/bpf/verifier.c > +++ b/kernel/bpf/verifier.c > @@ -19400,10 +19400,15 @@ int bpf_check_attach_target(struct bpf_verifier_log *log, > return -EINVAL; > fallthrough; > case BPF_MODIFY_RETURN: > - case BPF_LSM_MAC: > - case BPF_LSM_CGROUP: > case BPF_TRACE_FENTRY: > case BPF_TRACE_FEXIT: > + if (tgt_prog && subprog > 0 && > + tgt_prog->aux->func[subprog]->is_func && > + tgt_prog->aux->tail_call_reachable) > + tgt_info->tail_call_ctx = true; > + fallthrough; > + case BPF_LSM_MAC: > + case BPF_LSM_CGROUP: > if (!btf_type_is_func(t)) { > bpf_log(log, "attach_btf_id %u is not a function\n", > btf_id);
On 15/8/23 08:52, Eduard Zingerman wrote: > On Mon, 2023-08-14 at 21:41 +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. >> >> As a result, a tailcall infinite loop comes up. And the loop halts 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 FENTRY/FEXIT >> trampolines. > > Hi Leon, > > I'm not familiar with this part of the jit compiler, so decided that > taking a look at your series might be a good learning point. > I think I got the gist of it, but I don't understand where > the initial value of RAX (== 0) is coming from in > arch_prepare_bpf_trampoline(), could you please help me out? > > Also a nitpick: > - in arch_prepare_bpf_trampoline() there is a comment detailing > the stack layout, it probably should be updated to say that > tail call count is stored as well; > - before arch_prepare_bpf_trampoline() there is a comment with > an example of generated assembly, should it be updated? > > Thanks, > Eduard > a) Initial value of RAX is in emit_prologue(). if (!ebpf_from_cbpf) { if (tail_call_reachable && !is_subprog) /* When it's the entry of the whole * tailcall context, zeroing the RAX * means init tail_call_cnt. */ EMIT2(0x31, 0xC0); /* xor eax, eax */ else // Keep the same asm layout. EMIT2(0x66, 0x90); /* nop2 */ } I'd like to add this comment to emit_prologue(). b) Good to update the stack layout. I'll do it. c) Its comment will be updated also. Thanks, Leon >> >> Fixes: ebf7d1f508a7 ("bpf, x64: rework pro/epilogue and tailcall handling in JIT") >> Fixes: e411901c0b77 ("bpf: allow for tailcalls in BPF subprograms for x64 JIT") >> Signed-off-by: Leon Hwang <hffilwlqm@gmail.com> >> --- >> arch/x86/net/bpf_jit_comp.c | 23 +++++++++++++++++++---- >> include/linux/bpf.h | 6 ++++++ >> kernel/bpf/trampoline.c | 5 +++-- >> kernel/bpf/verifier.c | 9 +++++++-- >> 4 files changed, 35 insertions(+), 8 deletions(-) >> >> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c >> index a5930042139d3..ca5366d97ad04 100644 >> --- a/arch/x86/net/bpf_jit_comp.c >> +++ b/arch/x86/net/bpf_jit_comp.c >> @@ -1018,6 +1018,10 @@ 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 +1627,7 @@ 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); >> @@ -2464,6 +2466,8 @@ 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); >> >> @@ -2516,6 +2520,12 @@ 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. >> + */ >> + 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 */ >> @@ -2569,7 +2579,12 @@ 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. >> + */ >> + 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..55c72086034ef 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. >> */ >> @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { >> struct module *tgt_mod; >> const char *tgt_name; >> const struct btf_type *tgt_type; >> + bool tail_call_ctx; >> }; >> >> #define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ >> diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c >> index 78acf28d48732..0fae334e3f7b8 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) { >> @@ -783,6 +783,7 @@ struct bpf_trampoline *bpf_trampoline_get(u64 key, >> >> memcpy(&tr->func.model, &tgt_info->fmodel, sizeof(tgt_info->fmodel)); >> tr->func.addr = (void *)tgt_info->tgt_addr; >> + tr->flags = (tgt_info->tail_call_ctx ? BPF_TRAMP_F_TAIL_CALL_CTX : 0); >> out: >> mutex_unlock(&tr->mutex); >> return tr; >> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c >> index 4ccca1f6c9981..a78e5a2ae5c72 100644 >> --- a/kernel/bpf/verifier.c >> +++ b/kernel/bpf/verifier.c >> @@ -19400,10 +19400,15 @@ int bpf_check_attach_target(struct bpf_verifier_log *log, >> return -EINVAL; >> fallthrough; >> case BPF_MODIFY_RETURN: >> - case BPF_LSM_MAC: >> - case BPF_LSM_CGROUP: >> case BPF_TRACE_FENTRY: >> case BPF_TRACE_FEXIT: >> + if (tgt_prog && subprog > 0 && >> + tgt_prog->aux->func[subprog]->is_func && >> + tgt_prog->aux->tail_call_reachable) >> + tgt_info->tail_call_ctx = true; >> + fallthrough; >> + case BPF_LSM_MAC: >> + case BPF_LSM_CGROUP: >> if (!btf_type_is_func(t)) { >> bpf_log(log, "attach_btf_id %u is not a function\n", >> btf_id); >
On Tue, 2023-08-15 at 11:01 +0800, Leon Hwang wrote: [...] > a) Initial value of RAX is in emit_prologue(). > if (!ebpf_from_cbpf) { > if (tail_call_reachable && !is_subprog) > /* When it's the entry of the whole > * tailcall context, zeroing the RAX > * means init tail_call_cnt. > */ > EMIT2(0x31, 0xC0); /* xor eax, eax */ > else > // Keep the same asm layout. > EMIT2(0x66, 0x90); /* nop2 */ > } > I'd like to add this comment to emit_prologue(). Got it, thank you. [...]
On Mon, Aug 14, 2023 at 09:41:46PM +0800, Leon Hwang wrote: > @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { > struct module *tgt_mod; > const char *tgt_name; > const struct btf_type *tgt_type; > + bool tail_call_ctx; Instead of extra flag here can you check tgt_prog->aux->tail_call_reachable in check_attach_btf_id() and set tr->flags there? Other than this the fix makes sense. Please trim your cc list when you respin. Just maintainers, Maciej (author of fixes tag) and bpf@vger is enough.
On 18/8/23 06:31, Alexei Starovoitov wrote: > On Mon, Aug 14, 2023 at 09:41:46PM +0800, Leon Hwang wrote: >> @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { >> struct module *tgt_mod; >> const char *tgt_name; >> const struct btf_type *tgt_type; >> + bool tail_call_ctx; > > Instead of extra flag here can you check tgt_prog->aux->tail_call_reachable in check_attach_btf_id() > and set tr->flags there? Should we check tgt_prog->aux->func[subprog]->is_func? Or, tgt_prog->aux->tail_call_reachable is enough? I think tgt_prog->aux->func[subprog]->is_func is required to check. It's because it's a bug about subprog instead of tgt_prog. In check_attach_btf_id(): bool tail_call_ctx; // ... ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info, &tail_call_ctx); // ... tr->flags = (tail_call_ctx ? BPF_TRAMP_F_TAIL_CALL_CTX : 0); How about changing like this? However, it's bad to change bpf_check_attach_target() declaration. > Other than this the fix makes sense. > Please trim your cc list when you respin.> Just maintainers, Maciej (author of fixes tag) and bpf@vger is enough. I'll trim it. Thanks, Leon
On Thu, Aug 17, 2023 at 7:10 PM Leon Hwang <hffilwlqm@gmail.com> wrote: > > > > On 18/8/23 06:31, Alexei Starovoitov wrote: > > On Mon, Aug 14, 2023 at 09:41:46PM +0800, Leon Hwang wrote: > >> @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { > >> struct module *tgt_mod; > >> const char *tgt_name; > >> const struct btf_type *tgt_type; > >> + bool tail_call_ctx; > > > > Instead of extra flag here can you check tgt_prog->aux->tail_call_reachable in check_attach_btf_id() > > and set tr->flags there? > > Should we check tgt_prog->aux->func[subprog]->is_func? Or, tgt_prog->aux->tail_call_reachable > is enough? Please let the thread continue to a logical conclusion before resending new version. Will reply there.
On 2023/8/19 03:59, Alexei Starovoitov wrote: > On Thu, Aug 17, 2023 at 7:10 PM Leon Hwang <hffilwlqm@gmail.com> wrote: >> >> >> >> On 18/8/23 06:31, Alexei Starovoitov wrote: >>> On Mon, Aug 14, 2023 at 09:41:46PM +0800, Leon Hwang wrote: >>>> @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { >>>> struct module *tgt_mod; >>>> const char *tgt_name; >>>> const struct btf_type *tgt_type; >>>> + bool tail_call_ctx; >>> >>> Instead of extra flag here can you check tgt_prog->aux->tail_call_reachable in check_attach_btf_id() >>> and set tr->flags there? >> >> Should we check tgt_prog->aux->func[subprog]->is_func? Or, tgt_prog->aux->tail_call_reachable >> is enough? > > Please let the thread continue to a logical conclusion before resending > new version. Will reply there. Sorry for the new version without logical conclusion. I'll do it better in the future. Additionally, I'm looking forward to fix it, and then planning to add a feature to trace tailcalls with trampoline. Thanks, Leon
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index a5930042139d3..ca5366d97ad04 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -1018,6 +1018,10 @@ 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 +1627,7 @@ 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); @@ -2464,6 +2466,8 @@ 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); @@ -2516,6 +2520,12 @@ 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. + */ + 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 */ @@ -2569,7 +2579,12 @@ 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. + */ + 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..55c72086034ef 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. */ @@ -1147,6 +1152,7 @@ struct bpf_attach_target_info { struct module *tgt_mod; const char *tgt_name; const struct btf_type *tgt_type; + bool tail_call_ctx; }; #define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c index 78acf28d48732..0fae334e3f7b8 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) { @@ -783,6 +783,7 @@ struct bpf_trampoline *bpf_trampoline_get(u64 key, memcpy(&tr->func.model, &tgt_info->fmodel, sizeof(tgt_info->fmodel)); tr->func.addr = (void *)tgt_info->tgt_addr; + tr->flags = (tgt_info->tail_call_ctx ? BPF_TRAMP_F_TAIL_CALL_CTX : 0); out: mutex_unlock(&tr->mutex); return tr; diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 4ccca1f6c9981..a78e5a2ae5c72 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -19400,10 +19400,15 @@ int bpf_check_attach_target(struct bpf_verifier_log *log, return -EINVAL; fallthrough; case BPF_MODIFY_RETURN: - case BPF_LSM_MAC: - case BPF_LSM_CGROUP: case BPF_TRACE_FENTRY: case BPF_TRACE_FEXIT: + if (tgt_prog && subprog > 0 && + tgt_prog->aux->func[subprog]->is_func && + tgt_prog->aux->tail_call_reachable) + tgt_info->tail_call_ctx = true; + fallthrough; + case BPF_LSM_MAC: + case BPF_LSM_CGROUP: if (!btf_type_is_func(t)) { bpf_log(log, "attach_btf_id %u is not a function\n", btf_id);
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. As a result, a tailcall infinite loop comes up. And the loop halts 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 FENTRY/FEXIT trampolines. Fixes: ebf7d1f508a7 ("bpf, x64: rework pro/epilogue and tailcall handling in JIT") Fixes: e411901c0b77 ("bpf: allow for tailcalls in BPF subprograms for x64 JIT") Signed-off-by: Leon Hwang <hffilwlqm@gmail.com> --- arch/x86/net/bpf_jit_comp.c | 23 +++++++++++++++++++---- include/linux/bpf.h | 6 ++++++ kernel/bpf/trampoline.c | 5 +++-- kernel/bpf/verifier.c | 9 +++++++-- 4 files changed, 35 insertions(+), 8 deletions(-)