From 38f51c07054ff4796e473dba3bff2e648378002c Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Thu, 3 Oct 2019 01:45:11 +0200 Subject: bpf, x86: Small optimization in comparing against imm0 Replace 'cmp reg, 0' with 'test reg, reg' for comparisons against zero. Saves 1 byte of instruction encoding per occurrence. The flag results of test 'reg, reg' are identical to 'cmp reg, 0' in all cases except for AF which we don't use/care about. In terms of macro-fusibility in combination with a subsequent conditional jump instruction, both have the same properties for the jumps used in the JIT translation. For example, same JITed Cilium program can shrink a bit from e.g. 12,455 to 12,317 bytes as tests with 0 are used quite frequently. Signed-off-by: Daniel Borkmann Signed-off-by: Alexei Starovoitov Acked-by: Song Liu Acked-by: John Fastabend --- arch/x86/net/bpf_jit_comp.c | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 991549a1c5f3..3ad2ba1ad855 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -909,6 +909,16 @@ xadd: if (is_imm8(insn->off)) case BPF_JMP32 | BPF_JSLT | BPF_K: case BPF_JMP32 | BPF_JSGE | BPF_K: case BPF_JMP32 | BPF_JSLE | BPF_K: + /* test dst_reg, dst_reg to save one extra byte */ + if (imm32 == 0) { + if (BPF_CLASS(insn->code) == BPF_JMP) + EMIT1(add_2mod(0x48, dst_reg, dst_reg)); + else if (is_ereg(dst_reg)) + EMIT1(add_2mod(0x40, dst_reg, dst_reg)); + EMIT2(0x85, add_2reg(0xC0, dst_reg, dst_reg)); + goto emit_cond_jmp; + } + /* cmp dst_reg, imm8/32 */ if (BPF_CLASS(insn->code) == BPF_JMP) EMIT1(add_1mod(0x48, dst_reg)); -- cgit v1.2.3 From 3dec541b2e632d630fe7142ed44f0b3702ef1f8c Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Tue, 15 Oct 2019 20:25:03 -0700 Subject: bpf: Add support for BTF pointers to x86 JIT Pointer to BTF object is a pointer to kernel object or NULL. Such pointers can only be used by BPF_LDX instructions. The verifier changed their opcode from LDX|MEM|size to LDX|PROBE_MEM|size to make JITing easier. The number of entries in extable is the number of BPF_LDX insns that access kernel memory via "pointer to BTF type". Only these load instructions can fault. Since x86 extable is relative it has to be allocated in the same memory region as JITed code. Allocate it prior to last pass of JITing and let the last pass populate it. Pointer to extable in bpf_prog_aux is necessary to make page fault handling fast. Page fault handling is done in two steps: 1. bpf_prog_kallsyms_find() finds BPF program that page faulted. It's done by walking rb tree. 2. then extable for given bpf program is binary searched. This process is similar to how page faulting is done for kernel modules. The exception handler skips over faulting x86 instruction and initializes destination register with zero. This mimics exact behavior of bpf_probe_read (when probe_kernel_read faults dest is zeroed). JITs for other architectures can add support in similar way. Until then they will reject unknown opcode and fallback to interpreter. Since extable should be aligned and placed near JITed code make bpf_jit_binary_alloc() return 4 byte aligned image offset, so that extable aligning formula in bpf_int_jit_compile() doesn't need to rely on internal implementation of bpf_jit_binary_alloc(). On x86 gcc defaults to 16-byte alignment for regular kernel functions due to better performance. JITed code may be aligned to 16 in the future, but it will use 4 in the meantime. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Andrii Nakryiko Acked-by: Martin KaFai Lau Link: https://lore.kernel.org/bpf/20191016032505.2089704-10-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 97 +++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 93 insertions(+), 4 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 3ad2ba1ad855..8cd23d8309bf 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -9,7 +9,7 @@ #include #include #include - +#include #include #include @@ -123,6 +123,19 @@ static const int reg2hex[] = { [AUX_REG] = 3, /* R11 temp register */ }; +static const int reg2pt_regs[] = { + [BPF_REG_0] = offsetof(struct pt_regs, ax), + [BPF_REG_1] = offsetof(struct pt_regs, di), + [BPF_REG_2] = offsetof(struct pt_regs, si), + [BPF_REG_3] = offsetof(struct pt_regs, dx), + [BPF_REG_4] = offsetof(struct pt_regs, cx), + [BPF_REG_5] = offsetof(struct pt_regs, r8), + [BPF_REG_6] = offsetof(struct pt_regs, bx), + [BPF_REG_7] = offsetof(struct pt_regs, r13), + [BPF_REG_8] = offsetof(struct pt_regs, r14), + [BPF_REG_9] = offsetof(struct pt_regs, r15), +}; + /* * is_ereg() == true if BPF register 'reg' maps to x86-64 r8..r15 * which need extra byte of encoding. @@ -377,6 +390,19 @@ static void emit_mov_reg(u8 **pprog, bool is64, u32 dst_reg, u32 src_reg) *pprog = prog; } + +static bool ex_handler_bpf(const struct exception_table_entry *x, + struct pt_regs *regs, int trapnr, + unsigned long error_code, unsigned long fault_addr) +{ + u32 reg = x->fixup >> 8; + + /* jump over faulting load and clear dest register */ + *(unsigned long *)((void *)regs + reg) = 0; + regs->ip += x->fixup & 0xff; + return true; +} + static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image, int oldproglen, struct jit_context *ctx) { @@ -384,7 +410,7 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image, int insn_cnt = bpf_prog->len; bool seen_exit = false; u8 temp[BPF_MAX_INSN_SIZE + BPF_INSN_SAFETY]; - int i, cnt = 0; + int i, cnt = 0, excnt = 0; int proglen = 0; u8 *prog = temp; @@ -778,14 +804,17 @@ stx: if (is_imm8(insn->off)) /* LDX: dst_reg = *(u8*)(src_reg + off) */ case BPF_LDX | BPF_MEM | BPF_B: + case BPF_LDX | BPF_PROBE_MEM | BPF_B: /* Emit 'movzx rax, byte ptr [rax + off]' */ EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB6); goto ldx; case BPF_LDX | BPF_MEM | BPF_H: + case BPF_LDX | BPF_PROBE_MEM | BPF_H: /* Emit 'movzx rax, word ptr [rax + off]' */ EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB7); goto ldx; case BPF_LDX | BPF_MEM | BPF_W: + case BPF_LDX | BPF_PROBE_MEM | BPF_W: /* Emit 'mov eax, dword ptr [rax+0x14]' */ if (is_ereg(dst_reg) || is_ereg(src_reg)) EMIT2(add_2mod(0x40, src_reg, dst_reg), 0x8B); @@ -793,6 +822,7 @@ stx: if (is_imm8(insn->off)) EMIT1(0x8B); goto ldx; case BPF_LDX | BPF_MEM | BPF_DW: + case BPF_LDX | BPF_PROBE_MEM | BPF_DW: /* Emit 'mov rax, qword ptr [rax+0x14]' */ EMIT2(add_2mod(0x48, src_reg, dst_reg), 0x8B); ldx: /* @@ -805,6 +835,48 @@ ldx: /* else EMIT1_off32(add_2reg(0x80, src_reg, dst_reg), insn->off); + if (BPF_MODE(insn->code) == BPF_PROBE_MEM) { + struct exception_table_entry *ex; + u8 *_insn = image + proglen; + s64 delta; + + if (!bpf_prog->aux->extable) + break; + + if (excnt >= bpf_prog->aux->num_exentries) { + pr_err("ex gen bug\n"); + return -EFAULT; + } + ex = &bpf_prog->aux->extable[excnt++]; + + delta = _insn - (u8 *)&ex->insn; + if (!is_simm32(delta)) { + pr_err("extable->insn doesn't fit into 32-bit\n"); + return -EFAULT; + } + ex->insn = delta; + + delta = (u8 *)ex_handler_bpf - (u8 *)&ex->handler; + if (!is_simm32(delta)) { + pr_err("extable->handler doesn't fit into 32-bit\n"); + return -EFAULT; + } + ex->handler = delta; + + if (dst_reg > BPF_REG_9) { + pr_err("verifier error\n"); + return -EFAULT; + } + /* + * Compute size of x86 insn and its target dest x86 register. + * ex_handler_bpf() will use lower 8 bits to adjust + * pt_regs->ip to jump over this x86 instruction + * and upper bits to figure out which pt_regs to zero out. + * End result: x86 insn "mov rbx, qword ptr [rax+0x14]" + * of 4 bytes will be ignored and rbx will be zero inited. + */ + ex->fixup = (prog - temp) | (reg2pt_regs[dst_reg] << 8); + } break; /* STX XADD: lock *(u32*)(dst_reg + off) += src_reg */ @@ -1058,6 +1130,11 @@ emit_jmp: addrs[i] = proglen; prog = temp; } + + if (image && excnt != bpf_prog->aux->num_exentries) { + pr_err("extable is not populated\n"); + return -EFAULT; + } return proglen; } @@ -1158,12 +1235,24 @@ out_image: break; } if (proglen == oldproglen) { - header = bpf_jit_binary_alloc(proglen, &image, - 1, jit_fill_hole); + /* + * The number of entries in extable is the number of BPF_LDX + * insns that access kernel memory via "pointer to BTF type". + * The verifier changed their opcode from LDX|MEM|size + * to LDX|PROBE_MEM|size to make JITing easier. + */ + u32 align = __alignof__(struct exception_table_entry); + u32 extable_size = prog->aux->num_exentries * + sizeof(struct exception_table_entry); + + /* allocate module memory for x86 insns and extable */ + header = bpf_jit_binary_alloc(roundup(proglen, align) + extable_size, + &image, align, jit_fill_hole); if (!header) { prog = orig_prog; goto out_addrs; } + prog->aux->extable = (void *) image + roundup(proglen, align); } oldproglen = proglen; cond_resched(); -- cgit v1.2.3 From 3b2744e665206ea82ce7673cb3ec889b2898a267 Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Thu, 14 Nov 2019 10:57:02 -0800 Subject: bpf: Refactor x86 JIT into helpers Refactor x86 JITing of LDX, STX, CALL instructions into separate helper functions. No functional changes in LDX and STX helpers. There is a minor change in CALL helper. It will populate target address correctly on the first pass of JIT instead of second pass. That won't reduce total number of JIT passes though. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Song Liu Acked-by: Andrii Nakryiko Link: https://lore.kernel.org/bpf/20191114185720.1641606-3-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 152 ++++++++++++++++++++++++++++---------------- 1 file changed, 98 insertions(+), 54 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 8cd23d8309bf..fb99d976ad6e 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -198,6 +198,8 @@ struct jit_context { /* Maximum number of bytes emitted while JITing one eBPF insn */ #define BPF_MAX_INSN_SIZE 128 #define BPF_INSN_SAFETY 64 +/* number of bytes emit_call() needs to generate call instruction */ +#define X86_CALL_SIZE 5 #define PROLOGUE_SIZE 20 @@ -390,6 +392,99 @@ static void emit_mov_reg(u8 **pprog, bool is64, u32 dst_reg, u32 src_reg) *pprog = prog; } +/* LDX: dst_reg = *(u8*)(src_reg + off) */ +static void emit_ldx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off) +{ + u8 *prog = *pprog; + int cnt = 0; + + switch (size) { + case BPF_B: + /* Emit 'movzx rax, byte ptr [rax + off]' */ + EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB6); + break; + case BPF_H: + /* Emit 'movzx rax, word ptr [rax + off]' */ + EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB7); + break; + case BPF_W: + /* Emit 'mov eax, dword ptr [rax+0x14]' */ + if (is_ereg(dst_reg) || is_ereg(src_reg)) + EMIT2(add_2mod(0x40, src_reg, dst_reg), 0x8B); + else + EMIT1(0x8B); + break; + case BPF_DW: + /* Emit 'mov rax, qword ptr [rax+0x14]' */ + EMIT2(add_2mod(0x48, src_reg, dst_reg), 0x8B); + break; + } + /* + * If insn->off == 0 we can save one extra byte, but + * special case of x86 R13 which always needs an offset + * is not worth the hassle + */ + if (is_imm8(off)) + EMIT2(add_2reg(0x40, src_reg, dst_reg), off); + else + EMIT1_off32(add_2reg(0x80, src_reg, dst_reg), off); + *pprog = prog; +} + +/* STX: *(u8*)(dst_reg + off) = src_reg */ +static void emit_stx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off) +{ + u8 *prog = *pprog; + int cnt = 0; + + switch (size) { + case BPF_B: + /* Emit 'mov byte ptr [rax + off], al' */ + if (is_ereg(dst_reg) || is_ereg(src_reg) || + /* We have to add extra byte for x86 SIL, DIL regs */ + src_reg == BPF_REG_1 || src_reg == BPF_REG_2) + EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x88); + else + EMIT1(0x88); + break; + case BPF_H: + if (is_ereg(dst_reg) || is_ereg(src_reg)) + EMIT3(0x66, add_2mod(0x40, dst_reg, src_reg), 0x89); + else + EMIT2(0x66, 0x89); + break; + case BPF_W: + if (is_ereg(dst_reg) || is_ereg(src_reg)) + EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x89); + else + EMIT1(0x89); + break; + case BPF_DW: + EMIT2(add_2mod(0x48, dst_reg, src_reg), 0x89); + break; + } + if (is_imm8(off)) + EMIT2(add_2reg(0x40, dst_reg, src_reg), off); + else + EMIT1_off32(add_2reg(0x80, dst_reg, src_reg), off); + *pprog = prog; +} + +static int emit_call(u8 **pprog, void *func, void *ip) +{ + u8 *prog = *pprog; + int cnt = 0; + s64 offset; + + offset = func - (ip + X86_CALL_SIZE); + if (!is_simm32(offset)) { + pr_err("Target call %p is out of range\n", func); + return -EINVAL; + } + EMIT1_off32(0xE8, offset); + *pprog = prog; + return 0; +} static bool ex_handler_bpf(const struct exception_table_entry *x, struct pt_regs *regs, int trapnr, @@ -773,68 +868,22 @@ st: if (is_imm8(insn->off)) /* STX: *(u8*)(dst_reg + off) = src_reg */ case BPF_STX | BPF_MEM | BPF_B: - /* Emit 'mov byte ptr [rax + off], al' */ - if (is_ereg(dst_reg) || is_ereg(src_reg) || - /* We have to add extra byte for x86 SIL, DIL regs */ - src_reg == BPF_REG_1 || src_reg == BPF_REG_2) - EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x88); - else - EMIT1(0x88); - goto stx; case BPF_STX | BPF_MEM | BPF_H: - if (is_ereg(dst_reg) || is_ereg(src_reg)) - EMIT3(0x66, add_2mod(0x40, dst_reg, src_reg), 0x89); - else - EMIT2(0x66, 0x89); - goto stx; case BPF_STX | BPF_MEM | BPF_W: - if (is_ereg(dst_reg) || is_ereg(src_reg)) - EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x89); - else - EMIT1(0x89); - goto stx; case BPF_STX | BPF_MEM | BPF_DW: - EMIT2(add_2mod(0x48, dst_reg, src_reg), 0x89); -stx: if (is_imm8(insn->off)) - EMIT2(add_2reg(0x40, dst_reg, src_reg), insn->off); - else - EMIT1_off32(add_2reg(0x80, dst_reg, src_reg), - insn->off); + emit_stx(&prog, BPF_SIZE(insn->code), dst_reg, src_reg, insn->off); break; /* LDX: dst_reg = *(u8*)(src_reg + off) */ case BPF_LDX | BPF_MEM | BPF_B: case BPF_LDX | BPF_PROBE_MEM | BPF_B: - /* Emit 'movzx rax, byte ptr [rax + off]' */ - EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB6); - goto ldx; case BPF_LDX | BPF_MEM | BPF_H: case BPF_LDX | BPF_PROBE_MEM | BPF_H: - /* Emit 'movzx rax, word ptr [rax + off]' */ - EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB7); - goto ldx; case BPF_LDX | BPF_MEM | BPF_W: case BPF_LDX | BPF_PROBE_MEM | BPF_W: - /* Emit 'mov eax, dword ptr [rax+0x14]' */ - if (is_ereg(dst_reg) || is_ereg(src_reg)) - EMIT2(add_2mod(0x40, src_reg, dst_reg), 0x8B); - else - EMIT1(0x8B); - goto ldx; case BPF_LDX | BPF_MEM | BPF_DW: case BPF_LDX | BPF_PROBE_MEM | BPF_DW: - /* Emit 'mov rax, qword ptr [rax+0x14]' */ - EMIT2(add_2mod(0x48, src_reg, dst_reg), 0x8B); -ldx: /* - * If insn->off == 0 we can save one extra byte, but - * special case of x86 R13 which always needs an offset - * is not worth the hassle - */ - if (is_imm8(insn->off)) - EMIT2(add_2reg(0x40, src_reg, dst_reg), insn->off); - else - EMIT1_off32(add_2reg(0x80, src_reg, dst_reg), - insn->off); + emit_ldx(&prog, BPF_SIZE(insn->code), dst_reg, src_reg, insn->off); if (BPF_MODE(insn->code) == BPF_PROBE_MEM) { struct exception_table_entry *ex; u8 *_insn = image + proglen; @@ -899,13 +948,8 @@ xadd: if (is_imm8(insn->off)) /* call */ case BPF_JMP | BPF_CALL: func = (u8 *) __bpf_call_base + imm32; - jmp_offset = func - (image + addrs[i]); - if (!imm32 || !is_simm32(jmp_offset)) { - pr_err("unsupported BPF func %d addr %p image %p\n", - imm32, func, image); + if (!imm32 || emit_call(&prog, func, image + addrs[i - 1])) return -EINVAL; - } - EMIT1_off32(0xE8, jmp_offset); break; case BPF_JMP | BPF_TAIL_CALL: -- cgit v1.2.3 From 5964b2000f283ff5df366f718e0f083ebbaae977 Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Thu, 14 Nov 2019 10:57:03 -0800 Subject: bpf: Add bpf_arch_text_poke() helper Add bpf_arch_text_poke() helper that is used by BPF trampoline logic to patch nops/calls in kernel text into calls into BPF trampoline and to patch calls/nops inside BPF programs too. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Song Liu Acked-by: Andrii Nakryiko Link: https://lore.kernel.org/bpf/20191114185720.1641606-4-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 51 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index fb99d976ad6e..254b2889e881 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -9,9 +9,11 @@ #include #include #include +#include #include #include #include +#include static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len) { @@ -486,6 +488,55 @@ static int emit_call(u8 **pprog, void *func, void *ip) return 0; } +int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, + void *old_addr, void *new_addr) +{ + u8 old_insn[X86_CALL_SIZE] = {}; + u8 new_insn[X86_CALL_SIZE] = {}; + u8 *prog; + int ret; + + if (!is_kernel_text((long)ip)) + /* BPF trampoline in modules is not supported */ + return -EINVAL; + + if (old_addr) { + prog = old_insn; + ret = emit_call(&prog, old_addr, (void *)ip); + if (ret) + return ret; + } + if (new_addr) { + prog = new_insn; + ret = emit_call(&prog, new_addr, (void *)ip); + if (ret) + return ret; + } + ret = -EBUSY; + mutex_lock(&text_mutex); + switch (t) { + case BPF_MOD_NOP_TO_CALL: + if (memcmp(ip, ideal_nops[NOP_ATOMIC5], X86_CALL_SIZE)) + goto out; + text_poke_bp(ip, new_insn, X86_CALL_SIZE, NULL); + break; + case BPF_MOD_CALL_TO_CALL: + if (memcmp(ip, old_insn, X86_CALL_SIZE)) + goto out; + text_poke_bp(ip, new_insn, X86_CALL_SIZE, NULL); + break; + case BPF_MOD_CALL_TO_NOP: + if (memcmp(ip, old_insn, X86_CALL_SIZE)) + goto out; + text_poke_bp(ip, ideal_nops[NOP_ATOMIC5], X86_CALL_SIZE, NULL); + break; + } + ret = 0; +out: + mutex_unlock(&text_mutex); + return ret; +} + static bool ex_handler_bpf(const struct exception_table_entry *x, struct pt_regs *regs, int trapnr, unsigned long error_code, unsigned long fault_addr) -- cgit v1.2.3 From fec56f5890d93fc2ed74166c397dc186b1c25951 Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Thu, 14 Nov 2019 10:57:04 -0800 Subject: bpf: Introduce BPF trampoline Introduce BPF trampoline concept to allow kernel code to call into BPF programs with practically zero overhead. The trampoline generation logic is architecture dependent. It's converting native calling convention into BPF calling convention. BPF ISA is 64-bit (even on 32-bit architectures). The registers R1 to R5 are used to pass arguments into BPF functions. The main BPF program accepts only single argument "ctx" in R1. Whereas CPU native calling convention is different. x86-64 is passing first 6 arguments in registers and the rest on the stack. x86-32 is passing first 3 arguments in registers. sparc64 is passing first 6 in registers. And so on. The trampolines between BPF and kernel already exist. BPF_CALL_x macros in include/linux/filter.h statically compile trampolines from BPF into kernel helpers. They convert up to five u64 arguments into kernel C pointers and integers. On 64-bit architectures this BPF_to_kernel trampolines are nops. On 32-bit architecture they're meaningful. The opposite job kernel_to_BPF trampolines is done by CAST_TO_U64 macros and __bpf_trace_##call() shim functions in include/trace/bpf_probe.h. They convert kernel function arguments into array of u64s that BPF program consumes via R1=ctx pointer. This patch set is doing the same job as __bpf_trace_##call() static trampolines, but dynamically for any kernel function. There are ~22k global kernel functions that are attachable via nop at function entry. The function arguments and types are described in BTF. The job of btf_distill_func_proto() function is to extract useful information from BTF into "function model" that architecture dependent trampoline generators will use to generate assembly code to cast kernel function arguments into array of u64s. For example the kernel function eth_type_trans has two pointers. They will be casted to u64 and stored into stack of generated trampoline. The pointer to that stack space will be passed into BPF program in R1. On x86-64 such generated trampoline will consume 16 bytes of stack and two stores of %rdi and %rsi into stack. The verifier will make sure that only two u64 are accessed read-only by BPF program. The verifier will also recognize the precise type of the pointers being accessed and will not allow typecasting of the pointer to a different type within BPF program. The tracing use case in the datacenter demonstrated that certain key kernel functions have (like tcp_retransmit_skb) have 2 or more kprobes that are always active. Other functions have both kprobe and kretprobe. So it is essential to keep both kernel code and BPF programs executing at maximum speed. Hence generated BPF trampoline is re-generated every time new program is attached or detached to maintain maximum performance. To avoid the high cost of retpoline the attached BPF programs are called directly. __bpf_prog_enter/exit() are used to support per-program execution stats. In the future this logic will be optimized further by adding support for bpf_stats_enabled_key inside generated assembly code. Introduction of preemptible and sleepable BPF programs will completely remove the need to call to __bpf_prog_enter/exit(). Detach of a BPF program from the trampoline should not fail. To avoid memory allocation in detach path the half of the page is used as a reserve and flipped after each attach/detach. 2k bytes is enough to call 40+ BPF programs directly which is enough for BPF tracing use cases. This limit can be increased in the future. BPF_TRACE_FENTRY programs have access to raw kernel function arguments while BPF_TRACE_FEXIT programs have access to kernel return value as well. Often kprobe BPF program remembers function arguments in a map while kretprobe fetches arguments from a map and analyzes them together with return value. BPF_TRACE_FEXIT accelerates this typical use case. Recursion prevention for kprobe BPF programs is done via per-cpu bpf_prog_active counter. In practice that turned out to be a mistake. It caused programs to randomly skip execution. The tracing tools missed results they were looking for. Hence BPF trampoline doesn't provide builtin recursion prevention. It's a job of BPF program itself and will be addressed in the follow up patches. BPF trampoline is intended to be used beyond tracing and fentry/fexit use cases in the future. For example to remove retpoline cost from XDP programs. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Andrii Nakryiko Acked-by: Song Liu Link: https://lore.kernel.org/bpf/20191114185720.1641606-5-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 211 +++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 209 insertions(+), 2 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 254b2889e881..be2b43a894f6 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -98,6 +98,7 @@ static int bpf_size_to_x86_bytes(int bpf_size) /* Pick a register outside of BPF range for JIT internal work */ #define AUX_REG (MAX_BPF_JIT_REG + 1) +#define X86_REG_R9 (MAX_BPF_JIT_REG + 2) /* * The following table maps BPF registers to x86-64 registers. @@ -106,8 +107,8 @@ static int bpf_size_to_x86_bytes(int bpf_size) * register in load/store instructions, it always needs an * extra byte of encoding and is callee saved. * - * Also x86-64 register R9 is unused. x86-64 register R10 is - * used for blinding (if enabled). + * x86-64 register R9 is not used by BPF programs, but can be used by BPF + * trampoline. x86-64 register R10 is used for blinding (if enabled). */ static const int reg2hex[] = { [BPF_REG_0] = 0, /* RAX */ @@ -123,6 +124,7 @@ static const int reg2hex[] = { [BPF_REG_FP] = 5, /* RBP readonly */ [BPF_REG_AX] = 2, /* R10 temp register */ [AUX_REG] = 3, /* R11 temp register */ + [X86_REG_R9] = 1, /* R9 register, 6th function argument */ }; static const int reg2pt_regs[] = { @@ -150,6 +152,7 @@ static bool is_ereg(u32 reg) BIT(BPF_REG_7) | BIT(BPF_REG_8) | BIT(BPF_REG_9) | + BIT(X86_REG_R9) | BIT(BPF_REG_AX)); } @@ -1233,6 +1236,210 @@ emit_jmp: return proglen; } +static void save_regs(struct btf_func_model *m, u8 **prog, int nr_args, + int stack_size) +{ + int i; + /* Store function arguments to stack. + * For a function that accepts two pointers the sequence will be: + * mov QWORD PTR [rbp-0x10],rdi + * mov QWORD PTR [rbp-0x8],rsi + */ + for (i = 0; i < min(nr_args, 6); i++) + emit_stx(prog, bytes_to_bpf_size(m->arg_size[i]), + BPF_REG_FP, + i == 5 ? X86_REG_R9 : BPF_REG_1 + i, + -(stack_size - i * 8)); +} + +static void restore_regs(struct btf_func_model *m, u8 **prog, int nr_args, + int stack_size) +{ + int i; + + /* Restore function arguments from stack. + * For a function that accepts two pointers the sequence will be: + * EMIT4(0x48, 0x8B, 0x7D, 0xF0); mov rdi,QWORD PTR [rbp-0x10] + * EMIT4(0x48, 0x8B, 0x75, 0xF8); mov rsi,QWORD PTR [rbp-0x8] + */ + for (i = 0; i < min(nr_args, 6); i++) + emit_ldx(prog, bytes_to_bpf_size(m->arg_size[i]), + i == 5 ? X86_REG_R9 : BPF_REG_1 + i, + BPF_REG_FP, + -(stack_size - i * 8)); +} + +static int invoke_bpf(struct btf_func_model *m, u8 **pprog, + struct bpf_prog **progs, int prog_cnt, int stack_size) +{ + u8 *prog = *pprog; + int cnt = 0, i; + + for (i = 0; i < prog_cnt; i++) { + if (emit_call(&prog, __bpf_prog_enter, prog)) + return -EINVAL; + /* remember prog start time returned by __bpf_prog_enter */ + emit_mov_reg(&prog, true, BPF_REG_6, BPF_REG_0); + + /* arg1: lea rdi, [rbp - stack_size] */ + EMIT4(0x48, 0x8D, 0x7D, -stack_size); + /* arg2: progs[i]->insnsi for interpreter */ + if (!progs[i]->jited) + emit_mov_imm64(&prog, BPF_REG_2, + (long) progs[i]->insnsi >> 32, + (u32) (long) progs[i]->insnsi); + /* call JITed bpf program or interpreter */ + if (emit_call(&prog, progs[i]->bpf_func, prog)) + return -EINVAL; + + /* arg1: mov rdi, progs[i] */ + emit_mov_imm64(&prog, BPF_REG_1, (long) progs[i] >> 32, + (u32) (long) progs[i]); + /* arg2: mov rsi, rbx <- start time in nsec */ + emit_mov_reg(&prog, true, BPF_REG_2, BPF_REG_6); + if (emit_call(&prog, __bpf_prog_exit, prog)) + return -EINVAL; + } + *pprog = prog; + return 0; +} + +/* Example: + * __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev); + * its 'struct btf_func_model' will be nr_args=2 + * The assembly code when eth_type_trans is executing after trampoline: + * + * push rbp + * mov rbp, rsp + * sub rsp, 16 // space for skb and dev + * push rbx // temp regs to pass start time + * mov qword ptr [rbp - 16], rdi // save skb pointer to stack + * mov qword ptr [rbp - 8], rsi // save dev pointer to stack + * call __bpf_prog_enter // rcu_read_lock and preempt_disable + * mov rbx, rax // remember start time in bpf stats are enabled + * lea rdi, [rbp - 16] // R1==ctx of bpf prog + * call addr_of_jited_FENTRY_prog + * movabsq rdi, 64bit_addr_of_struct_bpf_prog // unused if bpf stats are off + * mov rsi, rbx // prog start time + * call __bpf_prog_exit // rcu_read_unlock, preempt_enable and stats math + * mov rdi, qword ptr [rbp - 16] // restore skb pointer from stack + * mov rsi, qword ptr [rbp - 8] // restore dev pointer from stack + * pop rbx + * leave + * ret + * + * eth_type_trans has 5 byte nop at the beginning. These 5 bytes will be + * replaced with 'call generated_bpf_trampoline'. When it returns + * eth_type_trans will continue executing with original skb and dev pointers. + * + * The assembly code when eth_type_trans is called from trampoline: + * + * push rbp + * mov rbp, rsp + * sub rsp, 24 // space for skb, dev, return value + * push rbx // temp regs to pass start time + * mov qword ptr [rbp - 24], rdi // save skb pointer to stack + * mov qword ptr [rbp - 16], rsi // save dev pointer to stack + * call __bpf_prog_enter // rcu_read_lock and preempt_disable + * mov rbx, rax // remember start time if bpf stats are enabled + * lea rdi, [rbp - 24] // R1==ctx of bpf prog + * call addr_of_jited_FENTRY_prog // bpf prog can access skb and dev + * movabsq rdi, 64bit_addr_of_struct_bpf_prog // unused if bpf stats are off + * mov rsi, rbx // prog start time + * call __bpf_prog_exit // rcu_read_unlock, preempt_enable and stats math + * mov rdi, qword ptr [rbp - 24] // restore skb pointer from stack + * mov rsi, qword ptr [rbp - 16] // restore dev pointer from stack + * call eth_type_trans+5 // execute body of eth_type_trans + * mov qword ptr [rbp - 8], rax // save return value + * call __bpf_prog_enter // rcu_read_lock and preempt_disable + * mov rbx, rax // remember start time in bpf stats are enabled + * lea rdi, [rbp - 24] // R1==ctx of bpf prog + * call addr_of_jited_FEXIT_prog // bpf prog can access skb, dev, return value + * movabsq rdi, 64bit_addr_of_struct_bpf_prog // unused if bpf stats are off + * mov rsi, rbx // prog start time + * call __bpf_prog_exit // rcu_read_unlock, preempt_enable and stats math + * mov rax, qword ptr [rbp - 8] // restore eth_type_trans's return value + * pop rbx + * leave + * add rsp, 8 // skip eth_type_trans's frame + * ret // return to its caller + */ +int arch_prepare_bpf_trampoline(void *image, struct btf_func_model *m, u32 flags, + struct bpf_prog **fentry_progs, int fentry_cnt, + struct bpf_prog **fexit_progs, int fexit_cnt, + void *orig_call) +{ + int cnt = 0, nr_args = m->nr_args; + int stack_size = nr_args * 8; + u8 *prog; + + /* x86-64 supports up to 6 arguments. 7+ can be added in the future */ + if (nr_args > 6) + return -ENOTSUPP; + + if ((flags & BPF_TRAMP_F_RESTORE_REGS) && + (flags & BPF_TRAMP_F_SKIP_FRAME)) + return -EINVAL; + + if (flags & BPF_TRAMP_F_CALL_ORIG) + stack_size += 8; /* room for return value of orig_call */ + + if (flags & BPF_TRAMP_F_SKIP_FRAME) + /* skip patched call instruction and point orig_call to actual + * body of the kernel function. + */ + orig_call += X86_CALL_SIZE; + + prog = image; + + EMIT1(0x55); /* push rbp */ + EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */ + EMIT4(0x48, 0x83, 0xEC, stack_size); /* sub rsp, stack_size */ + EMIT1(0x53); /* push rbx */ + + save_regs(m, &prog, nr_args, stack_size); + + if (fentry_cnt) + if (invoke_bpf(m, &prog, fentry_progs, fentry_cnt, stack_size)) + return -EINVAL; + + if (flags & BPF_TRAMP_F_CALL_ORIG) { + if (fentry_cnt) + restore_regs(m, &prog, nr_args, stack_size); + + /* call original function */ + if (emit_call(&prog, orig_call, prog)) + return -EINVAL; + /* remember return value in a stack for bpf prog to access */ + emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8); + } + + if (fexit_cnt) + if (invoke_bpf(m, &prog, fexit_progs, fexit_cnt, stack_size)) + return -EINVAL; + + if (flags & BPF_TRAMP_F_RESTORE_REGS) + restore_regs(m, &prog, nr_args, stack_size); + + if (flags & BPF_TRAMP_F_CALL_ORIG) + /* restore original return value back into RAX */ + emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, -8); + + EMIT1(0x5B); /* pop rbx */ + EMIT1(0xC9); /* leave */ + if (flags & BPF_TRAMP_F_SKIP_FRAME) + /* skip our return address and return to parent */ + EMIT4(0x48, 0x83, 0xC4, 8); /* add rsp, 8 */ + EMIT1(0xC3); /* ret */ + /* One half of the page has active running trampoline. + * Another half is an area for next trampoline. + * Make sure the trampoline generation logic doesn't overflow. + */ + if (WARN_ON_ONCE(prog - (u8 *)image > PAGE_SIZE / 2 - BPF_INSN_SAFETY)) + return -EFAULT; + return 0; +} + struct x64_jit_data { struct bpf_binary_header *header; int *addrs; -- cgit v1.2.3 From 9fd4a39dc7fe734d26eb89ea97e8c91331c6378c Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Thu, 14 Nov 2019 10:57:13 -0800 Subject: bpf: Reserve space for BPF trampoline in BPF programs BPF trampoline can be made to work with existing 5 bytes of BPF program prologue, but let's add 5 bytes of NOPs to the beginning of every JITed BPF program to make BPF trampoline job easier. They can be removed in the future. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Andrii Nakryiko Acked-by: Song Liu Link: https://lore.kernel.org/bpf/20191114185720.1641606-14-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index be2b43a894f6..c06096df9118 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -206,7 +206,7 @@ struct jit_context { /* number of bytes emit_call() needs to generate call instruction */ #define X86_CALL_SIZE 5 -#define PROLOGUE_SIZE 20 +#define PROLOGUE_SIZE 25 /* * Emit x86-64 prologue code for BPF program and check its size. @@ -215,8 +215,13 @@ struct jit_context { static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf) { u8 *prog = *pprog; - int cnt = 0; + int cnt = X86_CALL_SIZE; + /* BPF trampoline can be made to work without these nops, + * but let's waste 5 bytes for now and optimize later + */ + memcpy(prog, ideal_nops[NOP_ATOMIC5], cnt); + prog += cnt; EMIT1(0x55); /* push rbp */ EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */ /* sub rsp, rounded_stack_depth */ -- cgit v1.2.3 From 5b92a28aae4dd0f88778d540ecfdcdaec5a41723 Mon Sep 17 00:00:00 2001 From: Alexei Starovoitov Date: Thu, 14 Nov 2019 10:57:17 -0800 Subject: bpf: Support attaching tracing BPF program to other BPF programs Allow FENTRY/FEXIT BPF programs to attach to other BPF programs of any type including their subprograms. This feature allows snooping on input and output packets in XDP, TC programs including their return values. In order to do that the verifier needs to track types not only of vmlinux, but types of other BPF programs as well. The verifier also needs to translate uapi/linux/bpf.h types used by networking programs into kernel internal BTF types used by FENTRY/FEXIT BPF programs. In some cases LLVM optimizations can remove arguments from BPF subprograms without adjusting BTF info that LLVM backend knows. When BTF info disagrees with actual types that the verifiers sees the BPF trampoline has to fallback to conservative and treat all arguments as u64. The FENTRY/FEXIT program can still attach to such subprograms, but it won't be able to recognize pointer types like 'struct sk_buff *' and it won't be able to pass them to bpf_skb_output() for dumping packets to user space. The FENTRY/FEXIT program would need to use bpf_probe_read_kernel() instead. The BPF_PROG_LOAD command is extended with attach_prog_fd field. When it's set to zero the attach_btf_id is one vmlinux BTF type ids. When attach_prog_fd points to previously loaded BPF program the attach_btf_id is BTF type id of main function or one of its subprograms. Signed-off-by: Alexei Starovoitov Signed-off-by: Daniel Borkmann Acked-by: Song Liu Link: https://lore.kernel.org/bpf/20191114185720.1641606-18-ast@kernel.org --- arch/x86/net/bpf_jit_comp.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index c06096df9118..2e586f579945 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -504,7 +504,8 @@ int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, u8 *prog; int ret; - if (!is_kernel_text((long)ip)) + if (!is_kernel_text((long)ip) && + !is_bpf_text_address((long)ip)) /* BPF trampoline in modules is not supported */ return -EINVAL; -- cgit v1.2.3 From 4b3da77b72ad6b3c48c6fe4a395ace7db39a12c5 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Fri, 22 Nov 2019 21:07:54 +0100 Subject: bpf, x86: Generalize and extend bpf_arch_text_poke for direct jumps Add BPF_MOD_{NOP_TO_JUMP,JUMP_TO_JUMP,JUMP_TO_NOP} patching for x86 JIT in order to be able to patch direct jumps or nop them out. We need this facility in order to patch tail call jumps and in later work also BPF static keys. Signed-off-by: Daniel Borkmann Signed-off-by: Alexei Starovoitov Acked-by: Andrii Nakryiko Link: https://lore.kernel.org/bpf/aa4784196a8e5e985af4b30a4fe5336bce6e9643.1574452833.git.daniel@iogearbox.net --- arch/x86/net/bpf_jit_comp.c | 64 ++++++++++++++++++++++++++++++++------------- 1 file changed, 46 insertions(+), 18 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 2e586f579945..f438bd3b7689 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -203,8 +203,9 @@ struct jit_context { /* Maximum number of bytes emitted while JITing one eBPF insn */ #define BPF_MAX_INSN_SIZE 128 #define BPF_INSN_SAFETY 64 -/* number of bytes emit_call() needs to generate call instruction */ -#define X86_CALL_SIZE 5 + +/* Number of bytes emit_patch() needs to generate instructions */ +#define X86_PATCH_SIZE 5 #define PROLOGUE_SIZE 25 @@ -215,7 +216,7 @@ struct jit_context { static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf) { u8 *prog = *pprog; - int cnt = X86_CALL_SIZE; + int cnt = X86_PATCH_SIZE; /* BPF trampoline can be made to work without these nops, * but let's waste 5 bytes for now and optimize later @@ -480,64 +481,91 @@ static void emit_stx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off) *pprog = prog; } -static int emit_call(u8 **pprog, void *func, void *ip) +static int emit_patch(u8 **pprog, void *func, void *ip, u8 opcode) { u8 *prog = *pprog; int cnt = 0; s64 offset; - offset = func - (ip + X86_CALL_SIZE); + offset = func - (ip + X86_PATCH_SIZE); if (!is_simm32(offset)) { pr_err("Target call %p is out of range\n", func); return -EINVAL; } - EMIT1_off32(0xE8, offset); + EMIT1_off32(opcode, offset); *pprog = prog; return 0; } +static int emit_call(u8 **pprog, void *func, void *ip) +{ + return emit_patch(pprog, func, ip, 0xE8); +} + +static int emit_jump(u8 **pprog, void *func, void *ip) +{ + return emit_patch(pprog, func, ip, 0xE9); +} + int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, void *old_addr, void *new_addr) { - u8 old_insn[X86_CALL_SIZE] = {}; - u8 new_insn[X86_CALL_SIZE] = {}; + int (*emit_patch_fn)(u8 **pprog, void *func, void *ip); + u8 old_insn[X86_PATCH_SIZE] = {}; + u8 new_insn[X86_PATCH_SIZE] = {}; u8 *prog; int ret; if (!is_kernel_text((long)ip) && !is_bpf_text_address((long)ip)) - /* BPF trampoline in modules is not supported */ + /* BPF poking in modules is not supported */ return -EINVAL; + switch (t) { + case BPF_MOD_NOP_TO_CALL ... BPF_MOD_CALL_TO_NOP: + emit_patch_fn = emit_call; + break; + case BPF_MOD_NOP_TO_JUMP ... BPF_MOD_JUMP_TO_NOP: + emit_patch_fn = emit_jump; + break; + default: + return -ENOTSUPP; + } + if (old_addr) { prog = old_insn; - ret = emit_call(&prog, old_addr, (void *)ip); + ret = emit_patch_fn(&prog, old_addr, (void *)ip); if (ret) return ret; } if (new_addr) { prog = new_insn; - ret = emit_call(&prog, new_addr, (void *)ip); + ret = emit_patch_fn(&prog, new_addr, (void *)ip); if (ret) return ret; } + ret = -EBUSY; mutex_lock(&text_mutex); switch (t) { case BPF_MOD_NOP_TO_CALL: - if (memcmp(ip, ideal_nops[NOP_ATOMIC5], X86_CALL_SIZE)) + case BPF_MOD_NOP_TO_JUMP: + if (memcmp(ip, ideal_nops[NOP_ATOMIC5], X86_PATCH_SIZE)) goto out; - text_poke_bp(ip, new_insn, X86_CALL_SIZE, NULL); + text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); break; case BPF_MOD_CALL_TO_CALL: - if (memcmp(ip, old_insn, X86_CALL_SIZE)) + case BPF_MOD_JUMP_TO_JUMP: + if (memcmp(ip, old_insn, X86_PATCH_SIZE)) goto out; - text_poke_bp(ip, new_insn, X86_CALL_SIZE, NULL); + text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); break; case BPF_MOD_CALL_TO_NOP: - if (memcmp(ip, old_insn, X86_CALL_SIZE)) + case BPF_MOD_JUMP_TO_NOP: + if (memcmp(ip, old_insn, X86_PATCH_SIZE)) goto out; - text_poke_bp(ip, ideal_nops[NOP_ATOMIC5], X86_CALL_SIZE, NULL); + text_poke_bp(ip, ideal_nops[NOP_ATOMIC5], X86_PATCH_SIZE, + NULL); break; } ret = 0; @@ -1394,7 +1422,7 @@ int arch_prepare_bpf_trampoline(void *image, struct btf_func_model *m, u32 flags /* skip patched call instruction and point orig_call to actual * body of the kernel function. */ - orig_call += X86_CALL_SIZE; + orig_call += X86_PATCH_SIZE; prog = image; -- cgit v1.2.3 From 428d5df1fa4f28daf622c48dd19da35585c9053c Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Fri, 22 Nov 2019 21:08:00 +0100 Subject: bpf, x86: Emit patchable direct jump as tail call Add initial code emission for *direct* jumps for tail call maps in order to avoid the retpoline overhead from a493a87f38cf ("bpf, x64: implement retpoline for tail call") for situations that allow for it, meaning, for known constant keys at verification time which are used as index into the tail call map. In case of Cilium which makes heavy use of tail calls, constant keys are used in the vast majority, only for a single occurrence we use a dynamic key. High level outline is that if the target prog is NULL in the map, we emit a 5-byte nop for the fall-through case and if not, we emit a 5-byte direct relative jmp to the target bpf_func + skipped prologue offset. Later during runtime, we patch these 5-byte nop/jmps upon tail call map update or deletions dynamically. Note that on x86-64 the direct jmp works as we reuse the same stack frame and skip prologue (as opposed to some other JIT implementations). One of the issues is that the tail call map slots can change at any given time even during JITing. Therefore, we have two passes: i) emit nops for all patchable locations during main JITing phase until we declare prog->jited = 1 eventually. At this point the image is stable, not public yet and with all jmps disabled. While JITing, we collect additional info like poke->ip in order to remember the patch location for later modifications. In ii) bpf_tail_call_direct_fixup() walks over the progs poke_tab, locks the tail call maps poke_mutex to prevent from parallel updates and patches in the right locations via __bpf_arch_text_poke(). Note, the main bpf_arch_text_poke() cannot be used at this point since we're not yet exposed to kallsyms. For the update we use plain memcpy() since the image is not public and still in read-write mode. After patching, we activate that poke entry through poke->ip_stable. Meaning, at this point any tail call map updates/deletions are not going to ignore that poke entry anymore. Then, bpf_arch_text_poke() might still occur on the read-write image until we finally locked it as read-only. Both modifications on the given image are under text_mutex to avoid interference with each other when update requests come in in parallel for different tail call maps (current one we have locked in JIT and different one where poke->ip_stable was already set). Example prog: # ./bpftool p d x i 1655 0: (b7) r3 = 0 1: (18) r2 = map[id:526] 3: (85) call bpf_tail_call#12 4: (b7) r0 = 1 5: (95) exit Before: # ./bpftool p d j i 1655 0xffffffffc076e55c: 0: nopl 0x0(%rax,%rax,1) 5: push %rbp 6: mov %rsp,%rbp 9: sub $0x200,%rsp 10: push %rbx 11: push %r13 13: push %r14 15: push %r15 17: pushq $0x0 _ 19: xor %edx,%edx |_ index (arg 3) 1b: movabs $0xffff88d95cc82600,%rsi |_ map (arg 2) 25: mov %edx,%edx | index >= array->map.max_entries 27: cmp %edx,0x24(%rsi) | 2a: jbe 0x0000000000000066 |_ 2c: mov -0x224(%rbp),%eax | tail call limit check 32: cmp $0x20,%eax | 35: ja 0x0000000000000066 | 37: add $0x1,%eax | 3a: mov %eax,-0x224(%rbp) |_ 40: mov 0xd0(%rsi,%rdx,8),%rax |_ prog = array->ptrs[index] 48: test %rax,%rax | prog == NULL check 4b: je 0x0000000000000066 |_ 4d: mov 0x30(%rax),%rax | goto *(prog->bpf_func + prologue_size) 51: add $0x19,%rax | 55: callq 0x0000000000000061 | retpoline for indirect jump 5a: pause | 5c: lfence | 5f: jmp 0x000000000000005a | 61: mov %rax,(%rsp) | 65: retq |_ 66: mov $0x1,%eax 6b: pop %rbx 6c: pop %r15 6e: pop %r14 70: pop %r13 72: pop %rbx 73: leaveq 74: retq After; state after JIT: # ./bpftool p d j i 1655 0xffffffffc08e8930: 0: nopl 0x0(%rax,%rax,1) 5: push %rbp 6: mov %rsp,%rbp 9: sub $0x200,%rsp 10: push %rbx 11: push %r13 13: push %r14 15: push %r15 17: pushq $0x0 _ 19: xor %edx,%edx |_ index (arg 3) 1b: movabs $0xffff9d8afd74c000,%rsi |_ map (arg 2) 25: mov -0x224(%rbp),%eax | tail call limit check 2b: cmp $0x20,%eax | 2e: ja 0x000000000000003e | 30: add $0x1,%eax | 33: mov %eax,-0x224(%rbp) |_ 39: jmpq 0xfffffffffffd1785 |_ [direct] goto *(prog->bpf_func + prologue_size) 3e: mov $0x1,%eax 43: pop %rbx 44: pop %r15 46: pop %r14 48: pop %r13 4a: pop %rbx 4b: leaveq 4c: retq After; state after map update (target prog): # ./bpftool p d j i 1655 0xffffffffc08e8930: 0: nopl 0x0(%rax,%rax,1) 5: push %rbp 6: mov %rsp,%rbp 9: sub $0x200,%rsp 10: push %rbx 11: push %r13 13: push %r14 15: push %r15 17: pushq $0x0 19: xor %edx,%edx 1b: movabs $0xffff9d8afd74c000,%rsi 25: mov -0x224(%rbp),%eax 2b: cmp $0x20,%eax . 2e: ja 0x000000000000003e . 30: add $0x1,%eax . 33: mov %eax,-0x224(%rbp) |_ 39: jmpq 0xffffffffffb09f55 |_ goto *(prog->bpf_func + prologue_size) 3e: mov $0x1,%eax 43: pop %rbx 44: pop %r15 46: pop %r14 48: pop %r13 4a: pop %rbx 4b: leaveq 4c: retq After; state after map update (no prog): # ./bpftool p d j i 1655 0xffffffffc08e8930: 0: nopl 0x0(%rax,%rax,1) 5: push %rbp 6: mov %rsp,%rbp 9: sub $0x200,%rsp 10: push %rbx 11: push %r13 13: push %r14 15: push %r15 17: pushq $0x0 19: xor %edx,%edx 1b: movabs $0xffff9d8afd74c000,%rsi 25: mov -0x224(%rbp),%eax 2b: cmp $0x20,%eax . 2e: ja 0x000000000000003e . 30: add $0x1,%eax . 33: mov %eax,-0x224(%rbp) |_ 39: nopl 0x0(%rax,%rax,1) |_ fall-through nop 3e: mov $0x1,%eax 43: pop %rbx 44: pop %r15 46: pop %r14 48: pop %r13 4a: pop %rbx 4b: leaveq 4c: retq Nice bonus is that this also shrinks the code emission quite a bit for every tail call invocation. Signed-off-by: Daniel Borkmann Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/6ada4c1c9d35eeb5f4ecfab94593dafa6b5c4b09.1574452833.git.daniel@iogearbox.net --- arch/x86/net/bpf_jit_comp.c | 282 +++++++++++++++++++++++++++++--------------- 1 file changed, 187 insertions(+), 95 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index f438bd3b7689..15615c94804f 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -239,6 +239,123 @@ static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf) *pprog = prog; } +static int emit_patch(u8 **pprog, void *func, void *ip, u8 opcode) +{ + u8 *prog = *pprog; + int cnt = 0; + s64 offset; + + offset = func - (ip + X86_PATCH_SIZE); + if (!is_simm32(offset)) { + pr_err("Target call %p is out of range\n", func); + return -ERANGE; + } + EMIT1_off32(opcode, offset); + *pprog = prog; + return 0; +} + +static int emit_call(u8 **pprog, void *func, void *ip) +{ + return emit_patch(pprog, func, ip, 0xE8); +} + +static int emit_jump(u8 **pprog, void *func, void *ip) +{ + return emit_patch(pprog, func, ip, 0xE9); +} + +static int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, + void *old_addr, void *new_addr, + const bool text_live) +{ + int (*emit_patch_fn)(u8 **pprog, void *func, void *ip); + const u8 *nop_insn = ideal_nops[NOP_ATOMIC5]; + u8 old_insn[X86_PATCH_SIZE] = {}; + u8 new_insn[X86_PATCH_SIZE] = {}; + u8 *prog; + int ret; + + switch (t) { + case BPF_MOD_NOP_TO_CALL ... BPF_MOD_CALL_TO_NOP: + emit_patch_fn = emit_call; + break; + case BPF_MOD_NOP_TO_JUMP ... BPF_MOD_JUMP_TO_NOP: + emit_patch_fn = emit_jump; + break; + default: + return -ENOTSUPP; + } + + switch (t) { + case BPF_MOD_NOP_TO_CALL: + case BPF_MOD_NOP_TO_JUMP: + if (!old_addr && new_addr) { + memcpy(old_insn, nop_insn, X86_PATCH_SIZE); + + prog = new_insn; + ret = emit_patch_fn(&prog, new_addr, ip); + if (ret) + return ret; + break; + } + return -ENXIO; + case BPF_MOD_CALL_TO_CALL: + case BPF_MOD_JUMP_TO_JUMP: + if (old_addr && new_addr) { + prog = old_insn; + ret = emit_patch_fn(&prog, old_addr, ip); + if (ret) + return ret; + + prog = new_insn; + ret = emit_patch_fn(&prog, new_addr, ip); + if (ret) + return ret; + break; + } + return -ENXIO; + case BPF_MOD_CALL_TO_NOP: + case BPF_MOD_JUMP_TO_NOP: + if (old_addr && !new_addr) { + memcpy(new_insn, nop_insn, X86_PATCH_SIZE); + + prog = old_insn; + ret = emit_patch_fn(&prog, old_addr, ip); + if (ret) + return ret; + break; + } + return -ENXIO; + default: + return -ENOTSUPP; + } + + ret = -EBUSY; + mutex_lock(&text_mutex); + if (memcmp(ip, old_insn, X86_PATCH_SIZE)) + goto out; + if (text_live) + text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); + else + memcpy(ip, new_insn, X86_PATCH_SIZE); + ret = 0; +out: + mutex_unlock(&text_mutex); + return ret; +} + +int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, + void *old_addr, void *new_addr) +{ + if (!is_kernel_text((long)ip) && + !is_bpf_text_address((long)ip)) + /* BPF poking in modules is not supported */ + return -EINVAL; + + return __bpf_arch_text_poke(ip, t, old_addr, new_addr, true); +} + /* * Generate the following code: * @@ -253,7 +370,7 @@ static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf) * goto *(prog->bpf_func + prologue_size); * out: */ -static void emit_bpf_tail_call(u8 **pprog) +static void emit_bpf_tail_call_indirect(u8 **pprog) { u8 *prog = *pprog; int label1, label2, label3; @@ -320,6 +437,69 @@ static void emit_bpf_tail_call(u8 **pprog) *pprog = prog; } +static void emit_bpf_tail_call_direct(struct bpf_jit_poke_descriptor *poke, + u8 **pprog, int addr, u8 *image) +{ + u8 *prog = *pprog; + int cnt = 0; + + /* + * if (tail_call_cnt > MAX_TAIL_CALL_CNT) + * goto out; + */ + EMIT2_off32(0x8B, 0x85, -36 - MAX_BPF_STACK); /* mov eax, dword ptr [rbp - 548] */ + EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT); /* cmp eax, MAX_TAIL_CALL_CNT */ + EMIT2(X86_JA, 14); /* ja out */ + EMIT3(0x83, 0xC0, 0x01); /* add eax, 1 */ + EMIT2_off32(0x89, 0x85, -36 - MAX_BPF_STACK); /* mov dword ptr [rbp -548], eax */ + + poke->ip = image + (addr - X86_PATCH_SIZE); + poke->adj_off = PROLOGUE_SIZE; + + memcpy(prog, ideal_nops[NOP_ATOMIC5], X86_PATCH_SIZE); + prog += X86_PATCH_SIZE; + /* out: */ + + *pprog = prog; +} + +static void bpf_tail_call_direct_fixup(struct bpf_prog *prog) +{ + static const enum bpf_text_poke_type type = BPF_MOD_NOP_TO_JUMP; + struct bpf_jit_poke_descriptor *poke; + struct bpf_array *array; + struct bpf_prog *target; + int i, ret; + + for (i = 0; i < prog->aux->size_poke_tab; i++) { + poke = &prog->aux->poke_tab[i]; + WARN_ON_ONCE(READ_ONCE(poke->ip_stable)); + + if (poke->reason != BPF_POKE_REASON_TAIL_CALL) + continue; + + array = container_of(poke->tail_call.map, struct bpf_array, map); + mutex_lock(&array->aux->poke_mutex); + target = array->ptrs[poke->tail_call.key]; + if (target) { + /* Plain memcpy is used when image is not live yet + * and still not locked as read-only. Once poke + * location is active (poke->ip_stable), any parallel + * bpf_arch_text_poke() might occur still on the + * read-write image until we finally locked it as + * read-only. Both modifications on the given image + * are under text_mutex to avoid interference. + */ + ret = __bpf_arch_text_poke(poke->ip, type, NULL, + (u8 *)target->bpf_func + + poke->adj_off, false); + BUG_ON(ret < 0); + } + WRITE_ONCE(poke->ip_stable, true); + mutex_unlock(&array->aux->poke_mutex); + } +} + static void emit_mov_imm32(u8 **pprog, bool sign_propagate, u32 dst_reg, const u32 imm32) { @@ -481,99 +661,6 @@ static void emit_stx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off) *pprog = prog; } -static int emit_patch(u8 **pprog, void *func, void *ip, u8 opcode) -{ - u8 *prog = *pprog; - int cnt = 0; - s64 offset; - - offset = func - (ip + X86_PATCH_SIZE); - if (!is_simm32(offset)) { - pr_err("Target call %p is out of range\n", func); - return -EINVAL; - } - EMIT1_off32(opcode, offset); - *pprog = prog; - return 0; -} - -static int emit_call(u8 **pprog, void *func, void *ip) -{ - return emit_patch(pprog, func, ip, 0xE8); -} - -static int emit_jump(u8 **pprog, void *func, void *ip) -{ - return emit_patch(pprog, func, ip, 0xE9); -} - -int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, - void *old_addr, void *new_addr) -{ - int (*emit_patch_fn)(u8 **pprog, void *func, void *ip); - u8 old_insn[X86_PATCH_SIZE] = {}; - u8 new_insn[X86_PATCH_SIZE] = {}; - u8 *prog; - int ret; - - if (!is_kernel_text((long)ip) && - !is_bpf_text_address((long)ip)) - /* BPF poking in modules is not supported */ - return -EINVAL; - - switch (t) { - case BPF_MOD_NOP_TO_CALL ... BPF_MOD_CALL_TO_NOP: - emit_patch_fn = emit_call; - break; - case BPF_MOD_NOP_TO_JUMP ... BPF_MOD_JUMP_TO_NOP: - emit_patch_fn = emit_jump; - break; - default: - return -ENOTSUPP; - } - - if (old_addr) { - prog = old_insn; - ret = emit_patch_fn(&prog, old_addr, (void *)ip); - if (ret) - return ret; - } - if (new_addr) { - prog = new_insn; - ret = emit_patch_fn(&prog, new_addr, (void *)ip); - if (ret) - return ret; - } - - ret = -EBUSY; - mutex_lock(&text_mutex); - switch (t) { - case BPF_MOD_NOP_TO_CALL: - case BPF_MOD_NOP_TO_JUMP: - if (memcmp(ip, ideal_nops[NOP_ATOMIC5], X86_PATCH_SIZE)) - goto out; - text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); - break; - case BPF_MOD_CALL_TO_CALL: - case BPF_MOD_JUMP_TO_JUMP: - if (memcmp(ip, old_insn, X86_PATCH_SIZE)) - goto out; - text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); - break; - case BPF_MOD_CALL_TO_NOP: - case BPF_MOD_JUMP_TO_NOP: - if (memcmp(ip, old_insn, X86_PATCH_SIZE)) - goto out; - text_poke_bp(ip, ideal_nops[NOP_ATOMIC5], X86_PATCH_SIZE, - NULL); - break; - } - ret = 0; -out: - mutex_unlock(&text_mutex); - return ret; -} - static bool ex_handler_bpf(const struct exception_table_entry *x, struct pt_regs *regs, int trapnr, unsigned long error_code, unsigned long fault_addr) @@ -1041,7 +1128,11 @@ xadd: if (is_imm8(insn->off)) break; case BPF_JMP | BPF_TAIL_CALL: - emit_bpf_tail_call(&prog); + if (imm32) + emit_bpf_tail_call_direct(&bpf_prog->aux->poke_tab[imm32 - 1], + &prog, addrs[i], image); + else + emit_bpf_tail_call_indirect(&prog); break; /* cond jump */ @@ -1599,6 +1690,7 @@ out_image: if (image) { if (!prog->is_func || extra_pass) { + bpf_tail_call_direct_fixup(prog); bpf_jit_binary_lock_ro(header); } else { jit_data->addrs = addrs; -- cgit v1.2.3 From b553a6ec570044fc1ae300c6fb24f9ce204c5894 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Sun, 24 Nov 2019 01:39:42 +0100 Subject: bpf: Simplify __bpf_arch_text_poke poke type handling Given that we have BPF_MOD_NOP_TO_{CALL,JUMP}, BPF_MOD_{CALL,JUMP}_TO_NOP and BPF_MOD_{CALL,JUMP}_TO_{CALL,JUMP} poke types and that we also pass in old_addr as well as new_addr, it's a bit redundant and unnecessarily complicates __bpf_arch_text_poke() itself since we can derive the same from the *_addr that were passed in. Hence simplify and use BPF_MOD_{CALL,JUMP} as types which also allows to clean up call-sites. In addition to that, __bpf_arch_text_poke() currently verifies that text matches expected old_insn before we invoke text_poke_bp(). Also add a check on new_insn and skip rewrite if it already matches. Reason why this is rather useful is that it avoids making any special casing in prog_array_map_poke_run() when old and new prog were NULL and has the benefit that also for this case we perform a check on text whether it really matches our expectations. Suggested-by: Andrii Nakryiko Signed-off-by: Daniel Borkmann Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/fcb00a2b0b288d6c73de4ef58116a821c8fe8f2f.1574555798.git.daniel@iogearbox.net --- arch/x86/net/bpf_jit_comp.c | 85 +++++++++++++-------------------------------- 1 file changed, 25 insertions(+), 60 deletions(-) (limited to 'arch/x86/net/bpf_jit_comp.c') diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 15615c94804f..b8be18427277 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -269,76 +269,42 @@ static int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, void *old_addr, void *new_addr, const bool text_live) { - int (*emit_patch_fn)(u8 **pprog, void *func, void *ip); const u8 *nop_insn = ideal_nops[NOP_ATOMIC5]; - u8 old_insn[X86_PATCH_SIZE] = {}; - u8 new_insn[X86_PATCH_SIZE] = {}; + u8 old_insn[X86_PATCH_SIZE]; + u8 new_insn[X86_PATCH_SIZE]; u8 *prog; int ret; - switch (t) { - case BPF_MOD_NOP_TO_CALL ... BPF_MOD_CALL_TO_NOP: - emit_patch_fn = emit_call; - break; - case BPF_MOD_NOP_TO_JUMP ... BPF_MOD_JUMP_TO_NOP: - emit_patch_fn = emit_jump; - break; - default: - return -ENOTSUPP; + memcpy(old_insn, nop_insn, X86_PATCH_SIZE); + if (old_addr) { + prog = old_insn; + ret = t == BPF_MOD_CALL ? + emit_call(&prog, old_addr, ip) : + emit_jump(&prog, old_addr, ip); + if (ret) + return ret; } - switch (t) { - case BPF_MOD_NOP_TO_CALL: - case BPF_MOD_NOP_TO_JUMP: - if (!old_addr && new_addr) { - memcpy(old_insn, nop_insn, X86_PATCH_SIZE); - - prog = new_insn; - ret = emit_patch_fn(&prog, new_addr, ip); - if (ret) - return ret; - break; - } - return -ENXIO; - case BPF_MOD_CALL_TO_CALL: - case BPF_MOD_JUMP_TO_JUMP: - if (old_addr && new_addr) { - prog = old_insn; - ret = emit_patch_fn(&prog, old_addr, ip); - if (ret) - return ret; - - prog = new_insn; - ret = emit_patch_fn(&prog, new_addr, ip); - if (ret) - return ret; - break; - } - return -ENXIO; - case BPF_MOD_CALL_TO_NOP: - case BPF_MOD_JUMP_TO_NOP: - if (old_addr && !new_addr) { - memcpy(new_insn, nop_insn, X86_PATCH_SIZE); - - prog = old_insn; - ret = emit_patch_fn(&prog, old_addr, ip); - if (ret) - return ret; - break; - } - return -ENXIO; - default: - return -ENOTSUPP; + memcpy(new_insn, nop_insn, X86_PATCH_SIZE); + if (new_addr) { + prog = new_insn; + ret = t == BPF_MOD_CALL ? + emit_call(&prog, new_addr, ip) : + emit_jump(&prog, new_addr, ip); + if (ret) + return ret; } ret = -EBUSY; mutex_lock(&text_mutex); if (memcmp(ip, old_insn, X86_PATCH_SIZE)) goto out; - if (text_live) - text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); - else - memcpy(ip, new_insn, X86_PATCH_SIZE); + if (memcmp(ip, new_insn, X86_PATCH_SIZE)) { + if (text_live) + text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL); + else + memcpy(ip, new_insn, X86_PATCH_SIZE); + } ret = 0; out: mutex_unlock(&text_mutex); @@ -465,7 +431,6 @@ static void emit_bpf_tail_call_direct(struct bpf_jit_poke_descriptor *poke, static void bpf_tail_call_direct_fixup(struct bpf_prog *prog) { - static const enum bpf_text_poke_type type = BPF_MOD_NOP_TO_JUMP; struct bpf_jit_poke_descriptor *poke; struct bpf_array *array; struct bpf_prog *target; @@ -490,7 +455,7 @@ static void bpf_tail_call_direct_fixup(struct bpf_prog *prog) * read-only. Both modifications on the given image * are under text_mutex to avoid interference. */ - ret = __bpf_arch_text_poke(poke->ip, type, NULL, + ret = __bpf_arch_text_poke(poke->ip, BPF_MOD_JUMP, NULL, (u8 *)target->bpf_func + poke->adj_off, false); BUG_ON(ret < 0); -- cgit v1.2.3