1 /* 2 * Kernel Probes Jump Optimization (Optprobes) 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write to the Free Software 16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 17 * 18 * Copyright (C) IBM Corporation, 2002, 2004 19 * Copyright (C) Hitachi Ltd., 2012 20 */ 21 #include <linux/kprobes.h> 22 #include <linux/ptrace.h> 23 #include <linux/string.h> 24 #include <linux/slab.h> 25 #include <linux/hardirq.h> 26 #include <linux/preempt.h> 27 #include <linux/module.h> 28 #include <linux/kdebug.h> 29 #include <linux/kallsyms.h> 30 #include <linux/ftrace.h> 31 32 #include <asm/cacheflush.h> 33 #include <asm/desc.h> 34 #include <asm/pgtable.h> 35 #include <asm/uaccess.h> 36 #include <asm/alternative.h> 37 #include <asm/insn.h> 38 #include <asm/debugreg.h> 39 40 #include "common.h" 41 42 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr) 43 { 44 struct optimized_kprobe *op; 45 struct kprobe *kp; 46 long offs; 47 int i; 48 49 for (i = 0; i < RELATIVEJUMP_SIZE; i++) { 50 kp = get_kprobe((void *)addr - i); 51 /* This function only handles jump-optimized kprobe */ 52 if (kp && kprobe_optimized(kp)) { 53 op = container_of(kp, struct optimized_kprobe, kp); 54 /* If op->list is not empty, op is under optimizing */ 55 if (list_empty(&op->list)) 56 goto found; 57 } 58 } 59 60 return addr; 61 found: 62 /* 63 * If the kprobe can be optimized, original bytes which can be 64 * overwritten by jump destination address. In this case, original 65 * bytes must be recovered from op->optinsn.copied_insn buffer. 66 */ 67 memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t)); 68 if (addr == (unsigned long)kp->addr) { 69 buf[0] = kp->opcode; 70 memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE); 71 } else { 72 offs = addr - (unsigned long)kp->addr - 1; 73 memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs); 74 } 75 76 return (unsigned long)buf; 77 } 78 79 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */ 80 static void __kprobes synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val) 81 { 82 #ifdef CONFIG_X86_64 83 *addr++ = 0x48; 84 *addr++ = 0xbf; 85 #else 86 *addr++ = 0xb8; 87 #endif 88 *(unsigned long *)addr = val; 89 } 90 91 asm ( 92 ".global optprobe_template_entry\n" 93 "optprobe_template_entry:\n" 94 #ifdef CONFIG_X86_64 95 /* We don't bother saving the ss register */ 96 " pushq %rsp\n" 97 " pushfq\n" 98 SAVE_REGS_STRING 99 " movq %rsp, %rsi\n" 100 ".global optprobe_template_val\n" 101 "optprobe_template_val:\n" 102 ASM_NOP5 103 ASM_NOP5 104 ".global optprobe_template_call\n" 105 "optprobe_template_call:\n" 106 ASM_NOP5 107 /* Move flags to rsp */ 108 " movq 144(%rsp), %rdx\n" 109 " movq %rdx, 152(%rsp)\n" 110 RESTORE_REGS_STRING 111 /* Skip flags entry */ 112 " addq $8, %rsp\n" 113 " popfq\n" 114 #else /* CONFIG_X86_32 */ 115 " pushf\n" 116 SAVE_REGS_STRING 117 " movl %esp, %edx\n" 118 ".global optprobe_template_val\n" 119 "optprobe_template_val:\n" 120 ASM_NOP5 121 ".global optprobe_template_call\n" 122 "optprobe_template_call:\n" 123 ASM_NOP5 124 RESTORE_REGS_STRING 125 " addl $4, %esp\n" /* skip cs */ 126 " popf\n" 127 #endif 128 ".global optprobe_template_end\n" 129 "optprobe_template_end:\n"); 130 131 #define TMPL_MOVE_IDX \ 132 ((long)&optprobe_template_val - (long)&optprobe_template_entry) 133 #define TMPL_CALL_IDX \ 134 ((long)&optprobe_template_call - (long)&optprobe_template_entry) 135 #define TMPL_END_IDX \ 136 ((long)&optprobe_template_end - (long)&optprobe_template_entry) 137 138 #define INT3_SIZE sizeof(kprobe_opcode_t) 139 140 /* Optimized kprobe call back function: called from optinsn */ 141 static void __kprobes optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs) 142 { 143 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 144 unsigned long flags; 145 146 /* This is possible if op is under delayed unoptimizing */ 147 if (kprobe_disabled(&op->kp)) 148 return; 149 150 local_irq_save(flags); 151 if (kprobe_running()) { 152 kprobes_inc_nmissed_count(&op->kp); 153 } else { 154 /* Save skipped registers */ 155 #ifdef CONFIG_X86_64 156 regs->cs = __KERNEL_CS; 157 #else 158 regs->cs = __KERNEL_CS | get_kernel_rpl(); 159 regs->gs = 0; 160 #endif 161 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE; 162 regs->orig_ax = ~0UL; 163 164 __this_cpu_write(current_kprobe, &op->kp); 165 kcb->kprobe_status = KPROBE_HIT_ACTIVE; 166 opt_pre_handler(&op->kp, regs); 167 __this_cpu_write(current_kprobe, NULL); 168 } 169 local_irq_restore(flags); 170 } 171 172 static int __kprobes copy_optimized_instructions(u8 *dest, u8 *src) 173 { 174 int len = 0, ret; 175 176 while (len < RELATIVEJUMP_SIZE) { 177 ret = __copy_instruction(dest + len, src + len); 178 if (!ret || !can_boost(dest + len)) 179 return -EINVAL; 180 len += ret; 181 } 182 /* Check whether the address range is reserved */ 183 if (ftrace_text_reserved(src, src + len - 1) || 184 alternatives_text_reserved(src, src + len - 1) || 185 jump_label_text_reserved(src, src + len - 1)) 186 return -EBUSY; 187 188 return len; 189 } 190 191 /* Check whether insn is indirect jump */ 192 static int __kprobes insn_is_indirect_jump(struct insn *insn) 193 { 194 return ((insn->opcode.bytes[0] == 0xff && 195 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */ 196 insn->opcode.bytes[0] == 0xea); /* Segment based jump */ 197 } 198 199 /* Check whether insn jumps into specified address range */ 200 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len) 201 { 202 unsigned long target = 0; 203 204 switch (insn->opcode.bytes[0]) { 205 case 0xe0: /* loopne */ 206 case 0xe1: /* loope */ 207 case 0xe2: /* loop */ 208 case 0xe3: /* jcxz */ 209 case 0xe9: /* near relative jump */ 210 case 0xeb: /* short relative jump */ 211 break; 212 case 0x0f: 213 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */ 214 break; 215 return 0; 216 default: 217 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */ 218 break; 219 return 0; 220 } 221 target = (unsigned long)insn->next_byte + insn->immediate.value; 222 223 return (start <= target && target <= start + len); 224 } 225 226 /* Decode whole function to ensure any instructions don't jump into target */ 227 static int __kprobes can_optimize(unsigned long paddr) 228 { 229 unsigned long addr, size = 0, offset = 0; 230 struct insn insn; 231 kprobe_opcode_t buf[MAX_INSN_SIZE]; 232 233 /* Lookup symbol including addr */ 234 if (!kallsyms_lookup_size_offset(paddr, &size, &offset)) 235 return 0; 236 237 /* 238 * Do not optimize in the entry code due to the unstable 239 * stack handling. 240 */ 241 if ((paddr >= (unsigned long)__entry_text_start) && 242 (paddr < (unsigned long)__entry_text_end)) 243 return 0; 244 245 /* Check there is enough space for a relative jump. */ 246 if (size - offset < RELATIVEJUMP_SIZE) 247 return 0; 248 249 /* Decode instructions */ 250 addr = paddr - offset; 251 while (addr < paddr - offset + size) { /* Decode until function end */ 252 if (search_exception_tables(addr)) 253 /* 254 * Since some fixup code will jumps into this function, 255 * we can't optimize kprobe in this function. 256 */ 257 return 0; 258 kernel_insn_init(&insn, (void *)recover_probed_instruction(buf, addr)); 259 insn_get_length(&insn); 260 /* Another subsystem puts a breakpoint */ 261 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) 262 return 0; 263 /* Recover address */ 264 insn.kaddr = (void *)addr; 265 insn.next_byte = (void *)(addr + insn.length); 266 /* Check any instructions don't jump into target */ 267 if (insn_is_indirect_jump(&insn) || 268 insn_jump_into_range(&insn, paddr + INT3_SIZE, 269 RELATIVE_ADDR_SIZE)) 270 return 0; 271 addr += insn.length; 272 } 273 274 return 1; 275 } 276 277 /* Check optimized_kprobe can actually be optimized. */ 278 int __kprobes arch_check_optimized_kprobe(struct optimized_kprobe *op) 279 { 280 int i; 281 struct kprobe *p; 282 283 for (i = 1; i < op->optinsn.size; i++) { 284 p = get_kprobe(op->kp.addr + i); 285 if (p && !kprobe_disabled(p)) 286 return -EEXIST; 287 } 288 289 return 0; 290 } 291 292 /* Check the addr is within the optimized instructions. */ 293 int __kprobes 294 arch_within_optimized_kprobe(struct optimized_kprobe *op, unsigned long addr) 295 { 296 return ((unsigned long)op->kp.addr <= addr && 297 (unsigned long)op->kp.addr + op->optinsn.size > addr); 298 } 299 300 /* Free optimized instruction slot */ 301 static __kprobes 302 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty) 303 { 304 if (op->optinsn.insn) { 305 free_optinsn_slot(op->optinsn.insn, dirty); 306 op->optinsn.insn = NULL; 307 op->optinsn.size = 0; 308 } 309 } 310 311 void __kprobes arch_remove_optimized_kprobe(struct optimized_kprobe *op) 312 { 313 __arch_remove_optimized_kprobe(op, 1); 314 } 315 316 /* 317 * Copy replacing target instructions 318 * Target instructions MUST be relocatable (checked inside) 319 * This is called when new aggr(opt)probe is allocated or reused. 320 */ 321 int __kprobes arch_prepare_optimized_kprobe(struct optimized_kprobe *op) 322 { 323 u8 *buf; 324 int ret; 325 long rel; 326 327 if (!can_optimize((unsigned long)op->kp.addr)) 328 return -EILSEQ; 329 330 op->optinsn.insn = get_optinsn_slot(); 331 if (!op->optinsn.insn) 332 return -ENOMEM; 333 334 /* 335 * Verify if the address gap is in 2GB range, because this uses 336 * a relative jump. 337 */ 338 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE; 339 if (abs(rel) > 0x7fffffff) 340 return -ERANGE; 341 342 buf = (u8 *)op->optinsn.insn; 343 344 /* Copy instructions into the out-of-line buffer */ 345 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr); 346 if (ret < 0) { 347 __arch_remove_optimized_kprobe(op, 0); 348 return ret; 349 } 350 op->optinsn.size = ret; 351 352 /* Copy arch-dep-instance from template */ 353 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX); 354 355 /* Set probe information */ 356 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op); 357 358 /* Set probe function call */ 359 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback); 360 361 /* Set returning jmp instruction at the tail of out-of-line buffer */ 362 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size, 363 (u8 *)op->kp.addr + op->optinsn.size); 364 365 flush_icache_range((unsigned long) buf, 366 (unsigned long) buf + TMPL_END_IDX + 367 op->optinsn.size + RELATIVEJUMP_SIZE); 368 return 0; 369 } 370 371 /* 372 * Replace breakpoints (int3) with relative jumps. 373 * Caller must call with locking kprobe_mutex and text_mutex. 374 */ 375 void __kprobes arch_optimize_kprobes(struct list_head *oplist) 376 { 377 struct optimized_kprobe *op, *tmp; 378 u8 insn_buf[RELATIVEJUMP_SIZE]; 379 380 list_for_each_entry_safe(op, tmp, oplist, list) { 381 s32 rel = (s32)((long)op->optinsn.insn - 382 ((long)op->kp.addr + RELATIVEJUMP_SIZE)); 383 384 WARN_ON(kprobe_disabled(&op->kp)); 385 386 /* Backup instructions which will be replaced by jump address */ 387 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE, 388 RELATIVE_ADDR_SIZE); 389 390 insn_buf[0] = RELATIVEJUMP_OPCODE; 391 *(s32 *)(&insn_buf[1]) = rel; 392 393 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE, 394 op->optinsn.insn); 395 396 list_del_init(&op->list); 397 } 398 } 399 400 /* Replace a relative jump with a breakpoint (int3). */ 401 void __kprobes arch_unoptimize_kprobe(struct optimized_kprobe *op) 402 { 403 u8 insn_buf[RELATIVEJUMP_SIZE]; 404 405 /* Set int3 to first byte for kprobes */ 406 insn_buf[0] = BREAKPOINT_INSTRUCTION; 407 memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE); 408 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE, 409 op->optinsn.insn); 410 } 411 412 /* 413 * Recover original instructions and breakpoints from relative jumps. 414 * Caller must call with locking kprobe_mutex. 415 */ 416 extern void arch_unoptimize_kprobes(struct list_head *oplist, 417 struct list_head *done_list) 418 { 419 struct optimized_kprobe *op, *tmp; 420 421 list_for_each_entry_safe(op, tmp, oplist, list) { 422 arch_unoptimize_kprobe(op); 423 list_move(&op->list, done_list); 424 } 425 } 426 427 int __kprobes 428 setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter) 429 { 430 struct optimized_kprobe *op; 431 432 if (p->flags & KPROBE_FLAG_OPTIMIZED) { 433 /* This kprobe is really able to run optimized path. */ 434 op = container_of(p, struct optimized_kprobe, kp); 435 /* Detour through copied instructions */ 436 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX; 437 if (!reenter) 438 reset_current_kprobe(); 439 preempt_enable_no_resched(); 440 return 1; 441 } 442 return 0; 443 } 444