1 /* 2 * User-space Probes (UProbes) for x86 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, 2008-2011 19 * Authors: 20 * Srikar Dronamraju 21 * Jim Keniston 22 */ 23 #include <linux/kernel.h> 24 #include <linux/sched.h> 25 #include <linux/ptrace.h> 26 #include <linux/uprobes.h> 27 #include <linux/uaccess.h> 28 29 #include <linux/kdebug.h> 30 #include <asm/processor.h> 31 #include <asm/insn.h> 32 #include <asm/mmu_context.h> 33 34 /* Post-execution fixups. */ 35 36 /* Adjust IP back to vicinity of actual insn */ 37 #define UPROBE_FIX_IP 0x01 38 39 /* Adjust the return address of a call insn */ 40 #define UPROBE_FIX_CALL 0x02 41 42 /* Instruction will modify TF, don't change it */ 43 #define UPROBE_FIX_SETF 0x04 44 45 #define UPROBE_FIX_RIP_SI 0x08 46 #define UPROBE_FIX_RIP_DI 0x10 47 #define UPROBE_FIX_RIP_BX 0x20 48 #define UPROBE_FIX_RIP_MASK \ 49 (UPROBE_FIX_RIP_SI | UPROBE_FIX_RIP_DI | UPROBE_FIX_RIP_BX) 50 51 #define UPROBE_TRAP_NR UINT_MAX 52 53 /* Adaptations for mhiramat x86 decoder v14. */ 54 #define OPCODE1(insn) ((insn)->opcode.bytes[0]) 55 #define OPCODE2(insn) ((insn)->opcode.bytes[1]) 56 #define OPCODE3(insn) ((insn)->opcode.bytes[2]) 57 #define MODRM_REG(insn) X86_MODRM_REG((insn)->modrm.value) 58 59 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\ 60 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \ 61 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \ 62 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \ 63 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \ 64 << (row % 32)) 65 66 /* 67 * Good-instruction tables for 32-bit apps. This is non-const and volatile 68 * to keep gcc from statically optimizing it out, as variable_test_bit makes 69 * some versions of gcc to think only *(unsigned long*) is used. 70 * 71 * Opcodes we'll probably never support: 72 * 6c-6f - ins,outs. SEGVs if used in userspace 73 * e4-e7 - in,out imm. SEGVs if used in userspace 74 * ec-ef - in,out acc. SEGVs if used in userspace 75 * cc - int3. SIGTRAP if used in userspace 76 * ce - into. Not used in userspace - no kernel support to make it useful. SEGVs 77 * (why we support bound (62) then? it's similar, and similarly unused...) 78 * f1 - int1. SIGTRAP if used in userspace 79 * f4 - hlt. SEGVs if used in userspace 80 * fa - cli. SEGVs if used in userspace 81 * fb - sti. SEGVs if used in userspace 82 * 83 * Opcodes which need some work to be supported: 84 * 07,17,1f - pop es/ss/ds 85 * Normally not used in userspace, but would execute if used. 86 * Can cause GP or stack exception if tries to load wrong segment descriptor. 87 * We hesitate to run them under single step since kernel's handling 88 * of userspace single-stepping (TF flag) is fragile. 89 * We can easily refuse to support push es/cs/ss/ds (06/0e/16/1e) 90 * on the same grounds that they are never used. 91 * cd - int N. 92 * Used by userspace for "int 80" syscall entry. (Other "int N" 93 * cause GP -> SEGV since their IDT gates don't allow calls from CPL 3). 94 * Not supported since kernel's handling of userspace single-stepping 95 * (TF flag) is fragile. 96 * cf - iret. Normally not used in userspace. Doesn't SEGV unless arguments are bad 97 */ 98 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 99 static volatile u32 good_insns_32[256 / 32] = { 100 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 101 /* ---------------------------------------------- */ 102 W(0x00, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 00 */ 103 W(0x10, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0) , /* 10 */ 104 W(0x20, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 20 */ 105 W(0x30, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 30 */ 106 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */ 107 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 50 */ 108 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0) | /* 60 */ 109 W(0x70, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 70 */ 110 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */ 111 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */ 112 W(0xa0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* a0 */ 113 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* b0 */ 114 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0) | /* c0 */ 115 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */ 116 W(0xe0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* e0 */ 117 W(0xf0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1) /* f0 */ 118 /* ---------------------------------------------- */ 119 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 120 }; 121 #else 122 #define good_insns_32 NULL 123 #endif 124 125 /* Good-instruction tables for 64-bit apps. 126 * 127 * Genuinely invalid opcodes: 128 * 06,07 - formerly push/pop es 129 * 0e - formerly push cs 130 * 16,17 - formerly push/pop ss 131 * 1e,1f - formerly push/pop ds 132 * 27,2f,37,3f - formerly daa/das/aaa/aas 133 * 60,61 - formerly pusha/popa 134 * 62 - formerly bound. EVEX prefix for AVX512 (not yet supported) 135 * 82 - formerly redundant encoding of Group1 136 * 9a - formerly call seg:ofs 137 * ce - formerly into 138 * d4,d5 - formerly aam/aad 139 * d6 - formerly undocumented salc 140 * ea - formerly jmp seg:ofs 141 * 142 * Opcodes we'll probably never support: 143 * 6c-6f - ins,outs. SEGVs if used in userspace 144 * e4-e7 - in,out imm. SEGVs if used in userspace 145 * ec-ef - in,out acc. SEGVs if used in userspace 146 * cc - int3. SIGTRAP if used in userspace 147 * f1 - int1. SIGTRAP if used in userspace 148 * f4 - hlt. SEGVs if used in userspace 149 * fa - cli. SEGVs if used in userspace 150 * fb - sti. SEGVs if used in userspace 151 * 152 * Opcodes which need some work to be supported: 153 * cd - int N. 154 * Used by userspace for "int 80" syscall entry. (Other "int N" 155 * cause GP -> SEGV since their IDT gates don't allow calls from CPL 3). 156 * Not supported since kernel's handling of userspace single-stepping 157 * (TF flag) is fragile. 158 * cf - iret. Normally not used in userspace. Doesn't SEGV unless arguments are bad 159 */ 160 #if defined(CONFIG_X86_64) 161 static volatile u32 good_insns_64[256 / 32] = { 162 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 163 /* ---------------------------------------------- */ 164 W(0x00, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 1) | /* 00 */ 165 W(0x10, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0) , /* 10 */ 166 W(0x20, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0) | /* 20 */ 167 W(0x30, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0) , /* 30 */ 168 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */ 169 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 50 */ 170 W(0x60, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0) | /* 60 */ 171 W(0x70, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 70 */ 172 W(0x80, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */ 173 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1) , /* 90 */ 174 W(0xa0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* a0 */ 175 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* b0 */ 176 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0) | /* c0 */ 177 W(0xd0, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */ 178 W(0xe0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0) | /* e0 */ 179 W(0xf0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1) /* f0 */ 180 /* ---------------------------------------------- */ 181 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 182 }; 183 #else 184 #define good_insns_64 NULL 185 #endif 186 187 /* Using this for both 64-bit and 32-bit apps. 188 * Opcodes we don't support: 189 * 0f 00 - SLDT/STR/LLDT/LTR/VERR/VERW/-/- group. System insns 190 * 0f 01 - SGDT/SIDT/LGDT/LIDT/SMSW/-/LMSW/INVLPG group. 191 * Also encodes tons of other system insns if mod=11. 192 * Some are in fact non-system: xend, xtest, rdtscp, maybe more 193 * 0f 05 - syscall 194 * 0f 06 - clts (CPL0 insn) 195 * 0f 07 - sysret 196 * 0f 08 - invd (CPL0 insn) 197 * 0f 09 - wbinvd (CPL0 insn) 198 * 0f 0b - ud2 199 * 0f 30 - wrmsr (CPL0 insn) (then why rdmsr is allowed, it's also CPL0 insn?) 200 * 0f 34 - sysenter 201 * 0f 35 - sysexit 202 * 0f 37 - getsec 203 * 0f 78 - vmread (Intel VMX. CPL0 insn) 204 * 0f 79 - vmwrite (Intel VMX. CPL0 insn) 205 * Note: with prefixes, these two opcodes are 206 * extrq/insertq/AVX512 convert vector ops. 207 * 0f ae - group15: [f]xsave,[f]xrstor,[v]{ld,st}mxcsr,clflush[opt], 208 * {rd,wr}{fs,gs}base,{s,l,m}fence. 209 * Why? They are all user-executable. 210 */ 211 static volatile u32 good_2byte_insns[256 / 32] = { 212 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 213 /* ---------------------------------------------- */ 214 W(0x00, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 1, 1, 1) | /* 00 */ 215 W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 10 */ 216 W(0x20, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 20 */ 217 W(0x30, 0, 1, 1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* 30 */ 218 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */ 219 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 50 */ 220 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 60 */ 221 W(0x70, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* 70 */ 222 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */ 223 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */ 224 W(0xa0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1) | /* a0 */ 225 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* b0 */ 226 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */ 227 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */ 228 W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* e0 */ 229 W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) /* f0 */ 230 /* ---------------------------------------------- */ 231 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 232 }; 233 #undef W 234 235 /* 236 * opcodes we may need to refine support for: 237 * 238 * 0f - 2-byte instructions: For many of these instructions, the validity 239 * depends on the prefix and/or the reg field. On such instructions, we 240 * just consider the opcode combination valid if it corresponds to any 241 * valid instruction. 242 * 243 * 8f - Group 1 - only reg = 0 is OK 244 * c6-c7 - Group 11 - only reg = 0 is OK 245 * d9-df - fpu insns with some illegal encodings 246 * f2, f3 - repnz, repz prefixes. These are also the first byte for 247 * certain floating-point instructions, such as addsd. 248 * 249 * fe - Group 4 - only reg = 0 or 1 is OK 250 * ff - Group 5 - only reg = 0-6 is OK 251 * 252 * others -- Do we need to support these? 253 * 254 * 0f - (floating-point?) prefetch instructions 255 * 07, 17, 1f - pop es, pop ss, pop ds 256 * 26, 2e, 36, 3e - es:, cs:, ss:, ds: segment prefixes -- 257 * but 64 and 65 (fs: and gs:) seem to be used, so we support them 258 * 67 - addr16 prefix 259 * ce - into 260 * f0 - lock prefix 261 */ 262 263 /* 264 * TODO: 265 * - Where necessary, examine the modrm byte and allow only valid instructions 266 * in the different Groups and fpu instructions. 267 */ 268 269 static bool is_prefix_bad(struct insn *insn) 270 { 271 int i; 272 273 for (i = 0; i < insn->prefixes.nbytes; i++) { 274 switch (insn->prefixes.bytes[i]) { 275 case 0x26: /* INAT_PFX_ES */ 276 case 0x2E: /* INAT_PFX_CS */ 277 case 0x36: /* INAT_PFX_DS */ 278 case 0x3E: /* INAT_PFX_SS */ 279 case 0xF0: /* INAT_PFX_LOCK */ 280 return true; 281 } 282 } 283 return false; 284 } 285 286 static int uprobe_init_insn(struct arch_uprobe *auprobe, struct insn *insn, bool x86_64) 287 { 288 u32 volatile *good_insns; 289 290 insn_init(insn, auprobe->insn, sizeof(auprobe->insn), x86_64); 291 /* has the side-effect of processing the entire instruction */ 292 insn_get_length(insn); 293 if (WARN_ON_ONCE(!insn_complete(insn))) 294 return -ENOEXEC; 295 296 if (is_prefix_bad(insn)) 297 return -ENOTSUPP; 298 299 if (x86_64) 300 good_insns = good_insns_64; 301 else 302 good_insns = good_insns_32; 303 304 if (test_bit(OPCODE1(insn), (unsigned long *)good_insns)) 305 return 0; 306 307 if (insn->opcode.nbytes == 2) { 308 if (test_bit(OPCODE2(insn), (unsigned long *)good_2byte_insns)) 309 return 0; 310 } 311 312 return -ENOTSUPP; 313 } 314 315 #ifdef CONFIG_X86_64 316 /* 317 * If arch_uprobe->insn doesn't use rip-relative addressing, return 318 * immediately. Otherwise, rewrite the instruction so that it accesses 319 * its memory operand indirectly through a scratch register. Set 320 * defparam->fixups accordingly. (The contents of the scratch register 321 * will be saved before we single-step the modified instruction, 322 * and restored afterward). 323 * 324 * We do this because a rip-relative instruction can access only a 325 * relatively small area (+/- 2 GB from the instruction), and the XOL 326 * area typically lies beyond that area. At least for instructions 327 * that store to memory, we can't execute the original instruction 328 * and "fix things up" later, because the misdirected store could be 329 * disastrous. 330 * 331 * Some useful facts about rip-relative instructions: 332 * 333 * - There's always a modrm byte with bit layout "00 reg 101". 334 * - There's never a SIB byte. 335 * - The displacement is always 4 bytes. 336 * - REX.B=1 bit in REX prefix, which normally extends r/m field, 337 * has no effect on rip-relative mode. It doesn't make modrm byte 338 * with r/m=101 refer to register 1101 = R13. 339 */ 340 static void riprel_analyze(struct arch_uprobe *auprobe, struct insn *insn) 341 { 342 u8 *cursor; 343 u8 reg; 344 u8 reg2; 345 346 if (!insn_rip_relative(insn)) 347 return; 348 349 /* 350 * insn_rip_relative() would have decoded rex_prefix, vex_prefix, modrm. 351 * Clear REX.b bit (extension of MODRM.rm field): 352 * we want to encode low numbered reg, not r8+. 353 */ 354 if (insn->rex_prefix.nbytes) { 355 cursor = auprobe->insn + insn_offset_rex_prefix(insn); 356 /* REX byte has 0100wrxb layout, clearing REX.b bit */ 357 *cursor &= 0xfe; 358 } 359 /* 360 * Similar treatment for VEX3 prefix. 361 * TODO: add XOP/EVEX treatment when insn decoder supports them 362 */ 363 if (insn->vex_prefix.nbytes == 3) { 364 /* 365 * vex2: c5 rvvvvLpp (has no b bit) 366 * vex3/xop: c4/8f rxbmmmmm wvvvvLpp 367 * evex: 62 rxbR00mm wvvvv1pp zllBVaaa 368 * (evex will need setting of both b and x since 369 * in non-sib encoding evex.x is 4th bit of MODRM.rm) 370 * Setting VEX3.b (setting because it has inverted meaning): 371 */ 372 cursor = auprobe->insn + insn_offset_vex_prefix(insn) + 1; 373 *cursor |= 0x20; 374 } 375 376 /* 377 * Convert from rip-relative addressing to register-relative addressing 378 * via a scratch register. 379 * 380 * This is tricky since there are insns with modrm byte 381 * which also use registers not encoded in modrm byte: 382 * [i]div/[i]mul: implicitly use dx:ax 383 * shift ops: implicitly use cx 384 * cmpxchg: implicitly uses ax 385 * cmpxchg8/16b: implicitly uses dx:ax and bx:cx 386 * Encoding: 0f c7/1 modrm 387 * The code below thinks that reg=1 (cx), chooses si as scratch. 388 * mulx: implicitly uses dx: mulx r/m,r1,r2 does r1:r2 = dx * r/m. 389 * First appeared in Haswell (BMI2 insn). It is vex-encoded. 390 * Example where none of bx,cx,dx can be used as scratch reg: 391 * c4 e2 63 f6 0d disp32 mulx disp32(%rip),%ebx,%ecx 392 * [v]pcmpistri: implicitly uses cx, xmm0 393 * [v]pcmpistrm: implicitly uses xmm0 394 * [v]pcmpestri: implicitly uses ax, dx, cx, xmm0 395 * [v]pcmpestrm: implicitly uses ax, dx, xmm0 396 * Evil SSE4.2 string comparison ops from hell. 397 * maskmovq/[v]maskmovdqu: implicitly uses (ds:rdi) as destination. 398 * Encoding: 0f f7 modrm, 66 0f f7 modrm, vex-encoded: c5 f9 f7 modrm. 399 * Store op1, byte-masked by op2 msb's in each byte, to (ds:rdi). 400 * AMD says it has no 3-operand form (vex.vvvv must be 1111) 401 * and that it can have only register operands, not mem 402 * (its modrm byte must have mode=11). 403 * If these restrictions will ever be lifted, 404 * we'll need code to prevent selection of di as scratch reg! 405 * 406 * Summary: I don't know any insns with modrm byte which 407 * use SI register implicitly. DI register is used only 408 * by one insn (maskmovq) and BX register is used 409 * only by one too (cmpxchg8b). 410 * BP is stack-segment based (may be a problem?). 411 * AX, DX, CX are off-limits (many implicit users). 412 * SP is unusable (it's stack pointer - think about "pop mem"; 413 * also, rsp+disp32 needs sib encoding -> insn length change). 414 */ 415 416 reg = MODRM_REG(insn); /* Fetch modrm.reg */ 417 reg2 = 0xff; /* Fetch vex.vvvv */ 418 if (insn->vex_prefix.nbytes == 2) 419 reg2 = insn->vex_prefix.bytes[1]; 420 else if (insn->vex_prefix.nbytes == 3) 421 reg2 = insn->vex_prefix.bytes[2]; 422 /* 423 * TODO: add XOP, EXEV vvvv reading. 424 * 425 * vex.vvvv field is in bits 6-3, bits are inverted. 426 * But in 32-bit mode, high-order bit may be ignored. 427 * Therefore, let's consider only 3 low-order bits. 428 */ 429 reg2 = ((reg2 >> 3) & 0x7) ^ 0x7; 430 /* 431 * Register numbering is ax,cx,dx,bx, sp,bp,si,di, r8..r15. 432 * 433 * Choose scratch reg. Order is important: must not select bx 434 * if we can use si (cmpxchg8b case!) 435 */ 436 if (reg != 6 && reg2 != 6) { 437 reg2 = 6; 438 auprobe->defparam.fixups |= UPROBE_FIX_RIP_SI; 439 } else if (reg != 7 && reg2 != 7) { 440 reg2 = 7; 441 auprobe->defparam.fixups |= UPROBE_FIX_RIP_DI; 442 /* TODO (paranoia): force maskmovq to not use di */ 443 } else { 444 reg2 = 3; 445 auprobe->defparam.fixups |= UPROBE_FIX_RIP_BX; 446 } 447 /* 448 * Point cursor at the modrm byte. The next 4 bytes are the 449 * displacement. Beyond the displacement, for some instructions, 450 * is the immediate operand. 451 */ 452 cursor = auprobe->insn + insn_offset_modrm(insn); 453 /* 454 * Change modrm from "00 reg 101" to "10 reg reg2". Example: 455 * 89 05 disp32 mov %eax,disp32(%rip) becomes 456 * 89 86 disp32 mov %eax,disp32(%rsi) 457 */ 458 *cursor = 0x80 | (reg << 3) | reg2; 459 } 460 461 static inline unsigned long * 462 scratch_reg(struct arch_uprobe *auprobe, struct pt_regs *regs) 463 { 464 if (auprobe->defparam.fixups & UPROBE_FIX_RIP_SI) 465 return ®s->si; 466 if (auprobe->defparam.fixups & UPROBE_FIX_RIP_DI) 467 return ®s->di; 468 return ®s->bx; 469 } 470 471 /* 472 * If we're emulating a rip-relative instruction, save the contents 473 * of the scratch register and store the target address in that register. 474 */ 475 static void riprel_pre_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 476 { 477 if (auprobe->defparam.fixups & UPROBE_FIX_RIP_MASK) { 478 struct uprobe_task *utask = current->utask; 479 unsigned long *sr = scratch_reg(auprobe, regs); 480 481 utask->autask.saved_scratch_register = *sr; 482 *sr = utask->vaddr + auprobe->defparam.ilen; 483 } 484 } 485 486 static void riprel_post_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 487 { 488 if (auprobe->defparam.fixups & UPROBE_FIX_RIP_MASK) { 489 struct uprobe_task *utask = current->utask; 490 unsigned long *sr = scratch_reg(auprobe, regs); 491 492 *sr = utask->autask.saved_scratch_register; 493 } 494 } 495 #else /* 32-bit: */ 496 /* 497 * No RIP-relative addressing on 32-bit 498 */ 499 static void riprel_analyze(struct arch_uprobe *auprobe, struct insn *insn) 500 { 501 } 502 static void riprel_pre_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 503 { 504 } 505 static void riprel_post_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 506 { 507 } 508 #endif /* CONFIG_X86_64 */ 509 510 struct uprobe_xol_ops { 511 bool (*emulate)(struct arch_uprobe *, struct pt_regs *); 512 int (*pre_xol)(struct arch_uprobe *, struct pt_regs *); 513 int (*post_xol)(struct arch_uprobe *, struct pt_regs *); 514 void (*abort)(struct arch_uprobe *, struct pt_regs *); 515 }; 516 517 static inline int sizeof_long(void) 518 { 519 return is_ia32_task() ? 4 : 8; 520 } 521 522 static int default_pre_xol_op(struct arch_uprobe *auprobe, struct pt_regs *regs) 523 { 524 riprel_pre_xol(auprobe, regs); 525 return 0; 526 } 527 528 static int push_ret_address(struct pt_regs *regs, unsigned long ip) 529 { 530 unsigned long new_sp = regs->sp - sizeof_long(); 531 532 if (copy_to_user((void __user *)new_sp, &ip, sizeof_long())) 533 return -EFAULT; 534 535 regs->sp = new_sp; 536 return 0; 537 } 538 539 /* 540 * We have to fix things up as follows: 541 * 542 * Typically, the new ip is relative to the copied instruction. We need 543 * to make it relative to the original instruction (FIX_IP). Exceptions 544 * are return instructions and absolute or indirect jump or call instructions. 545 * 546 * If the single-stepped instruction was a call, the return address that 547 * is atop the stack is the address following the copied instruction. We 548 * need to make it the address following the original instruction (FIX_CALL). 549 * 550 * If the original instruction was a rip-relative instruction such as 551 * "movl %edx,0xnnnn(%rip)", we have instead executed an equivalent 552 * instruction using a scratch register -- e.g., "movl %edx,0xnnnn(%rsi)". 553 * We need to restore the contents of the scratch register 554 * (FIX_RIP_reg). 555 */ 556 static int default_post_xol_op(struct arch_uprobe *auprobe, struct pt_regs *regs) 557 { 558 struct uprobe_task *utask = current->utask; 559 560 riprel_post_xol(auprobe, regs); 561 if (auprobe->defparam.fixups & UPROBE_FIX_IP) { 562 long correction = utask->vaddr - utask->xol_vaddr; 563 regs->ip += correction; 564 } else if (auprobe->defparam.fixups & UPROBE_FIX_CALL) { 565 regs->sp += sizeof_long(); /* Pop incorrect return address */ 566 if (push_ret_address(regs, utask->vaddr + auprobe->defparam.ilen)) 567 return -ERESTART; 568 } 569 /* popf; tell the caller to not touch TF */ 570 if (auprobe->defparam.fixups & UPROBE_FIX_SETF) 571 utask->autask.saved_tf = true; 572 573 return 0; 574 } 575 576 static void default_abort_op(struct arch_uprobe *auprobe, struct pt_regs *regs) 577 { 578 riprel_post_xol(auprobe, regs); 579 } 580 581 static struct uprobe_xol_ops default_xol_ops = { 582 .pre_xol = default_pre_xol_op, 583 .post_xol = default_post_xol_op, 584 .abort = default_abort_op, 585 }; 586 587 static bool branch_is_call(struct arch_uprobe *auprobe) 588 { 589 return auprobe->branch.opc1 == 0xe8; 590 } 591 592 #define CASE_COND \ 593 COND(70, 71, XF(OF)) \ 594 COND(72, 73, XF(CF)) \ 595 COND(74, 75, XF(ZF)) \ 596 COND(78, 79, XF(SF)) \ 597 COND(7a, 7b, XF(PF)) \ 598 COND(76, 77, XF(CF) || XF(ZF)) \ 599 COND(7c, 7d, XF(SF) != XF(OF)) \ 600 COND(7e, 7f, XF(ZF) || XF(SF) != XF(OF)) 601 602 #define COND(op_y, op_n, expr) \ 603 case 0x ## op_y: DO((expr) != 0) \ 604 case 0x ## op_n: DO((expr) == 0) 605 606 #define XF(xf) (!!(flags & X86_EFLAGS_ ## xf)) 607 608 static bool is_cond_jmp_opcode(u8 opcode) 609 { 610 switch (opcode) { 611 #define DO(expr) \ 612 return true; 613 CASE_COND 614 #undef DO 615 616 default: 617 return false; 618 } 619 } 620 621 static bool check_jmp_cond(struct arch_uprobe *auprobe, struct pt_regs *regs) 622 { 623 unsigned long flags = regs->flags; 624 625 switch (auprobe->branch.opc1) { 626 #define DO(expr) \ 627 return expr; 628 CASE_COND 629 #undef DO 630 631 default: /* not a conditional jmp */ 632 return true; 633 } 634 } 635 636 #undef XF 637 #undef COND 638 #undef CASE_COND 639 640 static bool branch_emulate_op(struct arch_uprobe *auprobe, struct pt_regs *regs) 641 { 642 unsigned long new_ip = regs->ip += auprobe->branch.ilen; 643 unsigned long offs = (long)auprobe->branch.offs; 644 645 if (branch_is_call(auprobe)) { 646 /* 647 * If it fails we execute this (mangled, see the comment in 648 * branch_clear_offset) insn out-of-line. In the likely case 649 * this should trigger the trap, and the probed application 650 * should die or restart the same insn after it handles the 651 * signal, arch_uprobe_post_xol() won't be even called. 652 * 653 * But there is corner case, see the comment in ->post_xol(). 654 */ 655 if (push_ret_address(regs, new_ip)) 656 return false; 657 } else if (!check_jmp_cond(auprobe, regs)) { 658 offs = 0; 659 } 660 661 regs->ip = new_ip + offs; 662 return true; 663 } 664 665 static int branch_post_xol_op(struct arch_uprobe *auprobe, struct pt_regs *regs) 666 { 667 BUG_ON(!branch_is_call(auprobe)); 668 /* 669 * We can only get here if branch_emulate_op() failed to push the ret 670 * address _and_ another thread expanded our stack before the (mangled) 671 * "call" insn was executed out-of-line. Just restore ->sp and restart. 672 * We could also restore ->ip and try to call branch_emulate_op() again. 673 */ 674 regs->sp += sizeof_long(); 675 return -ERESTART; 676 } 677 678 static void branch_clear_offset(struct arch_uprobe *auprobe, struct insn *insn) 679 { 680 /* 681 * Turn this insn into "call 1f; 1:", this is what we will execute 682 * out-of-line if ->emulate() fails. We only need this to generate 683 * a trap, so that the probed task receives the correct signal with 684 * the properly filled siginfo. 685 * 686 * But see the comment in ->post_xol(), in the unlikely case it can 687 * succeed. So we need to ensure that the new ->ip can not fall into 688 * the non-canonical area and trigger #GP. 689 * 690 * We could turn it into (say) "pushf", but then we would need to 691 * divorce ->insn[] and ->ixol[]. We need to preserve the 1st byte 692 * of ->insn[] for set_orig_insn(). 693 */ 694 memset(auprobe->insn + insn_offset_immediate(insn), 695 0, insn->immediate.nbytes); 696 } 697 698 static struct uprobe_xol_ops branch_xol_ops = { 699 .emulate = branch_emulate_op, 700 .post_xol = branch_post_xol_op, 701 }; 702 703 /* Returns -ENOSYS if branch_xol_ops doesn't handle this insn */ 704 static int branch_setup_xol_ops(struct arch_uprobe *auprobe, struct insn *insn) 705 { 706 u8 opc1 = OPCODE1(insn); 707 int i; 708 709 switch (opc1) { 710 case 0xeb: /* jmp 8 */ 711 case 0xe9: /* jmp 32 */ 712 case 0x90: /* prefix* + nop; same as jmp with .offs = 0 */ 713 break; 714 715 case 0xe8: /* call relative */ 716 branch_clear_offset(auprobe, insn); 717 break; 718 719 case 0x0f: 720 if (insn->opcode.nbytes != 2) 721 return -ENOSYS; 722 /* 723 * If it is a "near" conditional jmp, OPCODE2() - 0x10 matches 724 * OPCODE1() of the "short" jmp which checks the same condition. 725 */ 726 opc1 = OPCODE2(insn) - 0x10; 727 default: 728 if (!is_cond_jmp_opcode(opc1)) 729 return -ENOSYS; 730 } 731 732 /* 733 * 16-bit overrides such as CALLW (66 e8 nn nn) are not supported. 734 * Intel and AMD behavior differ in 64-bit mode: Intel ignores 66 prefix. 735 * No one uses these insns, reject any branch insns with such prefix. 736 */ 737 for (i = 0; i < insn->prefixes.nbytes; i++) { 738 if (insn->prefixes.bytes[i] == 0x66) 739 return -ENOTSUPP; 740 } 741 742 auprobe->branch.opc1 = opc1; 743 auprobe->branch.ilen = insn->length; 744 auprobe->branch.offs = insn->immediate.value; 745 746 auprobe->ops = &branch_xol_ops; 747 return 0; 748 } 749 750 /** 751 * arch_uprobe_analyze_insn - instruction analysis including validity and fixups. 752 * @mm: the probed address space. 753 * @arch_uprobe: the probepoint information. 754 * @addr: virtual address at which to install the probepoint 755 * Return 0 on success or a -ve number on error. 756 */ 757 int arch_uprobe_analyze_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long addr) 758 { 759 struct insn insn; 760 u8 fix_ip_or_call = UPROBE_FIX_IP; 761 int ret; 762 763 ret = uprobe_init_insn(auprobe, &insn, is_64bit_mm(mm)); 764 if (ret) 765 return ret; 766 767 ret = branch_setup_xol_ops(auprobe, &insn); 768 if (ret != -ENOSYS) 769 return ret; 770 771 /* 772 * Figure out which fixups default_post_xol_op() will need to perform, 773 * and annotate defparam->fixups accordingly. 774 */ 775 switch (OPCODE1(&insn)) { 776 case 0x9d: /* popf */ 777 auprobe->defparam.fixups |= UPROBE_FIX_SETF; 778 break; 779 case 0xc3: /* ret or lret -- ip is correct */ 780 case 0xcb: 781 case 0xc2: 782 case 0xca: 783 case 0xea: /* jmp absolute -- ip is correct */ 784 fix_ip_or_call = 0; 785 break; 786 case 0x9a: /* call absolute - Fix return addr, not ip */ 787 fix_ip_or_call = UPROBE_FIX_CALL; 788 break; 789 case 0xff: 790 switch (MODRM_REG(&insn)) { 791 case 2: case 3: /* call or lcall, indirect */ 792 fix_ip_or_call = UPROBE_FIX_CALL; 793 break; 794 case 4: case 5: /* jmp or ljmp, indirect */ 795 fix_ip_or_call = 0; 796 break; 797 } 798 /* fall through */ 799 default: 800 riprel_analyze(auprobe, &insn); 801 } 802 803 auprobe->defparam.ilen = insn.length; 804 auprobe->defparam.fixups |= fix_ip_or_call; 805 806 auprobe->ops = &default_xol_ops; 807 return 0; 808 } 809 810 /* 811 * arch_uprobe_pre_xol - prepare to execute out of line. 812 * @auprobe: the probepoint information. 813 * @regs: reflects the saved user state of current task. 814 */ 815 int arch_uprobe_pre_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 816 { 817 struct uprobe_task *utask = current->utask; 818 819 if (auprobe->ops->pre_xol) { 820 int err = auprobe->ops->pre_xol(auprobe, regs); 821 if (err) 822 return err; 823 } 824 825 regs->ip = utask->xol_vaddr; 826 utask->autask.saved_trap_nr = current->thread.trap_nr; 827 current->thread.trap_nr = UPROBE_TRAP_NR; 828 829 utask->autask.saved_tf = !!(regs->flags & X86_EFLAGS_TF); 830 regs->flags |= X86_EFLAGS_TF; 831 if (test_tsk_thread_flag(current, TIF_BLOCKSTEP)) 832 set_task_blockstep(current, false); 833 834 return 0; 835 } 836 837 /* 838 * If xol insn itself traps and generates a signal(Say, 839 * SIGILL/SIGSEGV/etc), then detect the case where a singlestepped 840 * instruction jumps back to its own address. It is assumed that anything 841 * like do_page_fault/do_trap/etc sets thread.trap_nr != -1. 842 * 843 * arch_uprobe_pre_xol/arch_uprobe_post_xol save/restore thread.trap_nr, 844 * arch_uprobe_xol_was_trapped() simply checks that ->trap_nr is not equal to 845 * UPROBE_TRAP_NR == -1 set by arch_uprobe_pre_xol(). 846 */ 847 bool arch_uprobe_xol_was_trapped(struct task_struct *t) 848 { 849 if (t->thread.trap_nr != UPROBE_TRAP_NR) 850 return true; 851 852 return false; 853 } 854 855 /* 856 * Called after single-stepping. To avoid the SMP problems that can 857 * occur when we temporarily put back the original opcode to 858 * single-step, we single-stepped a copy of the instruction. 859 * 860 * This function prepares to resume execution after the single-step. 861 */ 862 int arch_uprobe_post_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 863 { 864 struct uprobe_task *utask = current->utask; 865 bool send_sigtrap = utask->autask.saved_tf; 866 int err = 0; 867 868 WARN_ON_ONCE(current->thread.trap_nr != UPROBE_TRAP_NR); 869 current->thread.trap_nr = utask->autask.saved_trap_nr; 870 871 if (auprobe->ops->post_xol) { 872 err = auprobe->ops->post_xol(auprobe, regs); 873 if (err) { 874 /* 875 * Restore ->ip for restart or post mortem analysis. 876 * ->post_xol() must not return -ERESTART unless this 877 * is really possible. 878 */ 879 regs->ip = utask->vaddr; 880 if (err == -ERESTART) 881 err = 0; 882 send_sigtrap = false; 883 } 884 } 885 /* 886 * arch_uprobe_pre_xol() doesn't save the state of TIF_BLOCKSTEP 887 * so we can get an extra SIGTRAP if we do not clear TF. We need 888 * to examine the opcode to make it right. 889 */ 890 if (send_sigtrap) 891 send_sig(SIGTRAP, current, 0); 892 893 if (!utask->autask.saved_tf) 894 regs->flags &= ~X86_EFLAGS_TF; 895 896 return err; 897 } 898 899 /* callback routine for handling exceptions. */ 900 int arch_uprobe_exception_notify(struct notifier_block *self, unsigned long val, void *data) 901 { 902 struct die_args *args = data; 903 struct pt_regs *regs = args->regs; 904 int ret = NOTIFY_DONE; 905 906 /* We are only interested in userspace traps */ 907 if (regs && !user_mode(regs)) 908 return NOTIFY_DONE; 909 910 switch (val) { 911 case DIE_INT3: 912 if (uprobe_pre_sstep_notifier(regs)) 913 ret = NOTIFY_STOP; 914 915 break; 916 917 case DIE_DEBUG: 918 if (uprobe_post_sstep_notifier(regs)) 919 ret = NOTIFY_STOP; 920 921 default: 922 break; 923 } 924 925 return ret; 926 } 927 928 /* 929 * This function gets called when XOL instruction either gets trapped or 930 * the thread has a fatal signal. Reset the instruction pointer to its 931 * probed address for the potential restart or for post mortem analysis. 932 */ 933 void arch_uprobe_abort_xol(struct arch_uprobe *auprobe, struct pt_regs *regs) 934 { 935 struct uprobe_task *utask = current->utask; 936 937 if (auprobe->ops->abort) 938 auprobe->ops->abort(auprobe, regs); 939 940 current->thread.trap_nr = utask->autask.saved_trap_nr; 941 regs->ip = utask->vaddr; 942 /* clear TF if it was set by us in arch_uprobe_pre_xol() */ 943 if (!utask->autask.saved_tf) 944 regs->flags &= ~X86_EFLAGS_TF; 945 } 946 947 static bool __skip_sstep(struct arch_uprobe *auprobe, struct pt_regs *regs) 948 { 949 if (auprobe->ops->emulate) 950 return auprobe->ops->emulate(auprobe, regs); 951 return false; 952 } 953 954 bool arch_uprobe_skip_sstep(struct arch_uprobe *auprobe, struct pt_regs *regs) 955 { 956 bool ret = __skip_sstep(auprobe, regs); 957 if (ret && (regs->flags & X86_EFLAGS_TF)) 958 send_sig(SIGTRAP, current, 0); 959 return ret; 960 } 961 962 unsigned long 963 arch_uretprobe_hijack_return_addr(unsigned long trampoline_vaddr, struct pt_regs *regs) 964 { 965 int rasize = sizeof_long(), nleft; 966 unsigned long orig_ret_vaddr = 0; /* clear high bits for 32-bit apps */ 967 968 if (copy_from_user(&orig_ret_vaddr, (void __user *)regs->sp, rasize)) 969 return -1; 970 971 /* check whether address has been already hijacked */ 972 if (orig_ret_vaddr == trampoline_vaddr) 973 return orig_ret_vaddr; 974 975 nleft = copy_to_user((void __user *)regs->sp, &trampoline_vaddr, rasize); 976 if (likely(!nleft)) 977 return orig_ret_vaddr; 978 979 if (nleft != rasize) { 980 pr_err("uprobe: return address clobbered: pid=%d, %%sp=%#lx, " 981 "%%ip=%#lx\n", current->pid, regs->sp, regs->ip); 982 983 force_sig_info(SIGSEGV, SEND_SIG_FORCED, current); 984 } 985 986 return -1; 987 } 988 989 bool arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx, 990 struct pt_regs *regs) 991 { 992 if (ctx == RP_CHECK_CALL) /* sp was just decremented by "call" insn */ 993 return regs->sp < ret->stack; 994 else 995 return regs->sp <= ret->stack; 996 } 997