| 1 | /* |
| 2 | * filter-visitor-generate-bytecode.c |
| 3 | * |
| 4 | * LTTng filter bytecode generation |
| 5 | * |
| 6 | * Copyright 2012 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 7 | * |
| 8 | * This library is free software; you can redistribute it and/or modify it |
| 9 | * under the terms of the GNU Lesser General Public License, version 2.1 only, |
| 10 | * as published by the Free Software Foundation. |
| 11 | * |
| 12 | * This library is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * Lesser General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU Lesser General Public License |
| 18 | * along with this library; if not, write to the Free Software Foundation, |
| 19 | * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ |
| 21 | |
| 22 | #include <stdlib.h> |
| 23 | #include <string.h> |
| 24 | #include <errno.h> |
| 25 | #include <common/align.h> |
| 26 | #include <common/compat/string.h> |
| 27 | |
| 28 | #include "filter-bytecode.h" |
| 29 | #include "filter-ir.h" |
| 30 | #include "filter-ast.h" |
| 31 | |
| 32 | #include <common/macros.h> |
| 33 | |
| 34 | #ifndef max_t |
| 35 | #define max_t(type, a, b) ((type) ((a) > (b) ? (a) : (b))) |
| 36 | #endif |
| 37 | |
| 38 | #define INIT_ALLOC_SIZE 4 |
| 39 | |
| 40 | static |
| 41 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, |
| 42 | struct ir_op *node); |
| 43 | |
| 44 | static inline int get_count_order(unsigned int count) |
| 45 | { |
| 46 | int order; |
| 47 | |
| 48 | order = lttng_fls(count) - 1; |
| 49 | if (count & (count - 1)) |
| 50 | order++; |
| 51 | return order; |
| 52 | } |
| 53 | |
| 54 | static |
| 55 | int bytecode_init(struct lttng_filter_bytecode_alloc **fb) |
| 56 | { |
| 57 | uint32_t alloc_len; |
| 58 | |
| 59 | alloc_len = sizeof(struct lttng_filter_bytecode_alloc) + INIT_ALLOC_SIZE; |
| 60 | *fb = calloc(alloc_len, 1); |
| 61 | if (!*fb) { |
| 62 | return -ENOMEM; |
| 63 | } else { |
| 64 | (*fb)->alloc_len = alloc_len; |
| 65 | return 0; |
| 66 | } |
| 67 | } |
| 68 | |
| 69 | static |
| 70 | int32_t bytecode_reserve(struct lttng_filter_bytecode_alloc **fb, uint32_t align, uint32_t len) |
| 71 | { |
| 72 | int32_t ret; |
| 73 | uint32_t padding = offset_align((*fb)->b.len, align); |
| 74 | uint32_t new_len = (*fb)->b.len + padding + len; |
| 75 | uint32_t new_alloc_len = sizeof(struct lttng_filter_bytecode_alloc) + new_len; |
| 76 | uint32_t old_alloc_len = (*fb)->alloc_len; |
| 77 | |
| 78 | if (new_len > LTTNG_FILTER_MAX_LEN) |
| 79 | return -EINVAL; |
| 80 | |
| 81 | if (new_alloc_len > old_alloc_len) { |
| 82 | struct lttng_filter_bytecode_alloc *newptr; |
| 83 | |
| 84 | new_alloc_len = |
| 85 | max_t(uint32_t, 1U << get_count_order(new_alloc_len), old_alloc_len << 1); |
| 86 | newptr = realloc(*fb, new_alloc_len); |
| 87 | if (!newptr) |
| 88 | return -ENOMEM; |
| 89 | *fb = newptr; |
| 90 | /* We zero directly the memory from start of allocation. */ |
| 91 | memset(&((char *) *fb)[old_alloc_len], 0, new_alloc_len - old_alloc_len); |
| 92 | (*fb)->alloc_len = new_alloc_len; |
| 93 | } |
| 94 | (*fb)->b.len += padding; |
| 95 | ret = (*fb)->b.len; |
| 96 | (*fb)->b.len += len; |
| 97 | return ret; |
| 98 | } |
| 99 | |
| 100 | static |
| 101 | int bytecode_push(struct lttng_filter_bytecode_alloc **fb, const void *data, |
| 102 | uint32_t align, uint32_t len) |
| 103 | { |
| 104 | int32_t offset; |
| 105 | |
| 106 | offset = bytecode_reserve(fb, align, len); |
| 107 | if (offset < 0) |
| 108 | return offset; |
| 109 | memcpy(&(*fb)->b.data[offset], data, len); |
| 110 | return 0; |
| 111 | } |
| 112 | |
| 113 | static |
| 114 | int bytecode_push_logical(struct lttng_filter_bytecode_alloc **fb, |
| 115 | struct logical_op *data, |
| 116 | uint32_t align, uint32_t len, |
| 117 | uint16_t *skip_offset) |
| 118 | { |
| 119 | int32_t offset; |
| 120 | |
| 121 | offset = bytecode_reserve(fb, align, len); |
| 122 | if (offset < 0) |
| 123 | return offset; |
| 124 | memcpy(&(*fb)->b.data[offset], data, len); |
| 125 | *skip_offset = |
| 126 | (void *) &((struct logical_op *) &(*fb)->b.data[offset])->skip_offset |
| 127 | - (void *) &(*fb)->b.data[0]; |
| 128 | return 0; |
| 129 | } |
| 130 | |
| 131 | static |
| 132 | int bytecode_patch(struct lttng_filter_bytecode_alloc **fb, |
| 133 | const void *data, |
| 134 | uint16_t offset, |
| 135 | uint32_t len) |
| 136 | { |
| 137 | if (offset >= (*fb)->b.len) { |
| 138 | return -EINVAL; |
| 139 | } |
| 140 | memcpy(&(*fb)->b.data[offset], data, len); |
| 141 | return 0; |
| 142 | } |
| 143 | |
| 144 | static |
| 145 | int visit_node_root(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 146 | { |
| 147 | int ret; |
| 148 | struct return_op insn; |
| 149 | |
| 150 | /* Visit child */ |
| 151 | ret = recursive_visit_gen_bytecode(ctx, node->u.root.child); |
| 152 | if (ret) |
| 153 | return ret; |
| 154 | |
| 155 | /* Generate end of bytecode instruction */ |
| 156 | insn.op = FILTER_OP_RETURN; |
| 157 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 158 | } |
| 159 | |
| 160 | static |
| 161 | int visit_node_load(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 162 | { |
| 163 | int ret; |
| 164 | |
| 165 | switch (node->data_type) { |
| 166 | case IR_DATA_UNKNOWN: |
| 167 | default: |
| 168 | fprintf(stderr, "[error] Unknown data type in %s\n", |
| 169 | __func__); |
| 170 | return -EINVAL; |
| 171 | |
| 172 | case IR_DATA_STRING: |
| 173 | { |
| 174 | struct load_op *insn; |
| 175 | uint32_t insn_len = sizeof(struct load_op) |
| 176 | + strlen(node->u.load.u.string) + 1; |
| 177 | |
| 178 | insn = calloc(insn_len, 1); |
| 179 | if (!insn) |
| 180 | return -ENOMEM; |
| 181 | insn->op = FILTER_OP_LOAD_STRING; |
| 182 | strcpy(insn->data, node->u.load.u.string); |
| 183 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 184 | free(insn); |
| 185 | return ret; |
| 186 | } |
| 187 | case IR_DATA_NUMERIC: |
| 188 | { |
| 189 | struct load_op *insn; |
| 190 | uint32_t insn_len = sizeof(struct load_op) |
| 191 | + sizeof(struct literal_numeric); |
| 192 | |
| 193 | insn = calloc(insn_len, 1); |
| 194 | if (!insn) |
| 195 | return -ENOMEM; |
| 196 | insn->op = FILTER_OP_LOAD_S64; |
| 197 | memcpy(insn->data, &node->u.load.u.num, sizeof(int64_t)); |
| 198 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 199 | free(insn); |
| 200 | return ret; |
| 201 | } |
| 202 | case IR_DATA_FLOAT: |
| 203 | { |
| 204 | struct load_op *insn; |
| 205 | uint32_t insn_len = sizeof(struct load_op) |
| 206 | + sizeof(struct literal_double); |
| 207 | |
| 208 | insn = calloc(insn_len, 1); |
| 209 | if (!insn) |
| 210 | return -ENOMEM; |
| 211 | insn->op = FILTER_OP_LOAD_DOUBLE; |
| 212 | memcpy(insn->data, &node->u.load.u.flt, sizeof(double)); |
| 213 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 214 | free(insn); |
| 215 | return ret; |
| 216 | } |
| 217 | case IR_DATA_FIELD_REF: /* fall-through */ |
| 218 | case IR_DATA_GET_CONTEXT_REF: |
| 219 | { |
| 220 | struct load_op *insn; |
| 221 | uint32_t insn_len = sizeof(struct load_op) |
| 222 | + sizeof(struct field_ref); |
| 223 | struct field_ref ref_offset; |
| 224 | uint32_t reloc_offset_u32; |
| 225 | uint16_t reloc_offset; |
| 226 | |
| 227 | insn = calloc(insn_len, 1); |
| 228 | if (!insn) |
| 229 | return -ENOMEM; |
| 230 | switch(node->data_type) { |
| 231 | case IR_DATA_FIELD_REF: |
| 232 | insn->op = FILTER_OP_LOAD_FIELD_REF; |
| 233 | break; |
| 234 | case IR_DATA_GET_CONTEXT_REF: |
| 235 | insn->op = FILTER_OP_GET_CONTEXT_REF; |
| 236 | break; |
| 237 | default: |
| 238 | free(insn); |
| 239 | return -EINVAL; |
| 240 | } |
| 241 | ref_offset.offset = (uint16_t) -1U; |
| 242 | memcpy(insn->data, &ref_offset, sizeof(ref_offset)); |
| 243 | /* reloc_offset points to struct load_op */ |
| 244 | reloc_offset_u32 = bytecode_get_len(&ctx->bytecode->b); |
| 245 | if (reloc_offset_u32 > LTTNG_FILTER_MAX_LEN - 1) { |
| 246 | free(insn); |
| 247 | return -EINVAL; |
| 248 | } |
| 249 | reloc_offset = (uint16_t) reloc_offset_u32; |
| 250 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 251 | if (ret) { |
| 252 | free(insn); |
| 253 | return ret; |
| 254 | } |
| 255 | /* append reloc */ |
| 256 | ret = bytecode_push(&ctx->bytecode_reloc, &reloc_offset, |
| 257 | 1, sizeof(reloc_offset)); |
| 258 | if (ret) { |
| 259 | free(insn); |
| 260 | return ret; |
| 261 | } |
| 262 | ret = bytecode_push(&ctx->bytecode_reloc, node->u.load.u.ref, |
| 263 | 1, strlen(node->u.load.u.ref) + 1); |
| 264 | free(insn); |
| 265 | return ret; |
| 266 | } |
| 267 | } |
| 268 | } |
| 269 | |
| 270 | static |
| 271 | int visit_node_unary(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 272 | { |
| 273 | int ret; |
| 274 | struct unary_op insn; |
| 275 | |
| 276 | /* Visit child */ |
| 277 | ret = recursive_visit_gen_bytecode(ctx, node->u.unary.child); |
| 278 | if (ret) |
| 279 | return ret; |
| 280 | |
| 281 | /* Generate end of bytecode instruction */ |
| 282 | switch (node->u.unary.type) { |
| 283 | case AST_UNARY_UNKNOWN: |
| 284 | default: |
| 285 | fprintf(stderr, "[error] Unknown unary node type in %s\n", |
| 286 | __func__); |
| 287 | return -EINVAL; |
| 288 | case AST_UNARY_PLUS: |
| 289 | /* Nothing to do. */ |
| 290 | return 0; |
| 291 | case AST_UNARY_MINUS: |
| 292 | insn.op = FILTER_OP_UNARY_MINUS; |
| 293 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 294 | case AST_UNARY_NOT: |
| 295 | insn.op = FILTER_OP_UNARY_NOT; |
| 296 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | /* |
| 301 | * Binary comparator nesting is disallowed. This allows fitting into |
| 302 | * only 2 registers. |
| 303 | */ |
| 304 | static |
| 305 | int visit_node_binary(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 306 | { |
| 307 | int ret; |
| 308 | struct binary_op insn; |
| 309 | |
| 310 | /* Visit child */ |
| 311 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); |
| 312 | if (ret) |
| 313 | return ret; |
| 314 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); |
| 315 | if (ret) |
| 316 | return ret; |
| 317 | |
| 318 | switch (node->u.binary.type) { |
| 319 | case AST_OP_UNKNOWN: |
| 320 | default: |
| 321 | fprintf(stderr, "[error] Unknown unary node type in %s\n", |
| 322 | __func__); |
| 323 | return -EINVAL; |
| 324 | |
| 325 | case AST_OP_AND: |
| 326 | case AST_OP_OR: |
| 327 | fprintf(stderr, "[error] Unexpected logical node type in %s\n", |
| 328 | __func__); |
| 329 | return -EINVAL; |
| 330 | |
| 331 | case AST_OP_MUL: |
| 332 | insn.op = FILTER_OP_MUL; |
| 333 | break; |
| 334 | case AST_OP_DIV: |
| 335 | insn.op = FILTER_OP_DIV; |
| 336 | break; |
| 337 | case AST_OP_MOD: |
| 338 | insn.op = FILTER_OP_MOD; |
| 339 | break; |
| 340 | case AST_OP_PLUS: |
| 341 | insn.op = FILTER_OP_PLUS; |
| 342 | break; |
| 343 | case AST_OP_MINUS: |
| 344 | insn.op = FILTER_OP_MINUS; |
| 345 | break; |
| 346 | case AST_OP_RSHIFT: |
| 347 | insn.op = FILTER_OP_RSHIFT; |
| 348 | break; |
| 349 | case AST_OP_LSHIFT: |
| 350 | insn.op = FILTER_OP_LSHIFT; |
| 351 | break; |
| 352 | case AST_OP_BIN_AND: |
| 353 | insn.op = FILTER_OP_BIN_AND; |
| 354 | break; |
| 355 | case AST_OP_BIN_OR: |
| 356 | insn.op = FILTER_OP_BIN_OR; |
| 357 | break; |
| 358 | case AST_OP_BIN_XOR: |
| 359 | insn.op = FILTER_OP_BIN_XOR; |
| 360 | break; |
| 361 | |
| 362 | case AST_OP_EQ: |
| 363 | insn.op = FILTER_OP_EQ; |
| 364 | break; |
| 365 | case AST_OP_NE: |
| 366 | insn.op = FILTER_OP_NE; |
| 367 | break; |
| 368 | case AST_OP_GT: |
| 369 | insn.op = FILTER_OP_GT; |
| 370 | break; |
| 371 | case AST_OP_LT: |
| 372 | insn.op = FILTER_OP_LT; |
| 373 | break; |
| 374 | case AST_OP_GE: |
| 375 | insn.op = FILTER_OP_GE; |
| 376 | break; |
| 377 | case AST_OP_LE: |
| 378 | insn.op = FILTER_OP_LE; |
| 379 | break; |
| 380 | } |
| 381 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 382 | } |
| 383 | |
| 384 | /* |
| 385 | * A logical op always return a s64 (1 or 0). |
| 386 | */ |
| 387 | static |
| 388 | int visit_node_logical(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 389 | { |
| 390 | int ret; |
| 391 | struct logical_op insn; |
| 392 | uint16_t skip_offset_loc; |
| 393 | uint16_t target_loc; |
| 394 | |
| 395 | /* Visit left child */ |
| 396 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); |
| 397 | if (ret) |
| 398 | return ret; |
| 399 | /* Cast to s64 if float or field ref */ |
| 400 | if ((node->u.binary.left->data_type == IR_DATA_FIELD_REF |
| 401 | || node->u.binary.left->data_type == IR_DATA_GET_CONTEXT_REF) |
| 402 | || node->u.binary.left->data_type == IR_DATA_FLOAT) { |
| 403 | struct cast_op cast_insn; |
| 404 | |
| 405 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF |
| 406 | || node->u.binary.left->data_type == IR_DATA_GET_CONTEXT_REF) { |
| 407 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
| 408 | } else { |
| 409 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; |
| 410 | } |
| 411 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
| 412 | 1, sizeof(cast_insn)); |
| 413 | if (ret) |
| 414 | return ret; |
| 415 | } |
| 416 | switch (node->u.logical.type) { |
| 417 | default: |
| 418 | fprintf(stderr, "[error] Unknown node type in %s\n", |
| 419 | __func__); |
| 420 | return -EINVAL; |
| 421 | |
| 422 | case AST_OP_AND: |
| 423 | insn.op = FILTER_OP_AND; |
| 424 | break; |
| 425 | case AST_OP_OR: |
| 426 | insn.op = FILTER_OP_OR; |
| 427 | break; |
| 428 | } |
| 429 | insn.skip_offset = (uint16_t) -1UL; /* Temporary */ |
| 430 | ret = bytecode_push_logical(&ctx->bytecode, &insn, 1, sizeof(insn), |
| 431 | &skip_offset_loc); |
| 432 | if (ret) |
| 433 | return ret; |
| 434 | /* Visit right child */ |
| 435 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); |
| 436 | if (ret) |
| 437 | return ret; |
| 438 | /* Cast to s64 if float or field ref */ |
| 439 | if ((node->u.binary.right->data_type == IR_DATA_FIELD_REF |
| 440 | || node->u.binary.right->data_type == IR_DATA_GET_CONTEXT_REF) |
| 441 | || node->u.binary.right->data_type == IR_DATA_FLOAT) { |
| 442 | struct cast_op cast_insn; |
| 443 | |
| 444 | if (node->u.binary.right->data_type == IR_DATA_FIELD_REF |
| 445 | || node->u.binary.right->data_type == IR_DATA_GET_CONTEXT_REF) { |
| 446 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
| 447 | } else { |
| 448 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; |
| 449 | } |
| 450 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
| 451 | 1, sizeof(cast_insn)); |
| 452 | if (ret) |
| 453 | return ret; |
| 454 | } |
| 455 | /* We now know where the logical op can skip. */ |
| 456 | target_loc = (uint16_t) bytecode_get_len(&ctx->bytecode->b); |
| 457 | ret = bytecode_patch(&ctx->bytecode, |
| 458 | &target_loc, /* Offset to jump to */ |
| 459 | skip_offset_loc, /* Where to patch */ |
| 460 | sizeof(uint16_t)); |
| 461 | return ret; |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * Postorder traversal of the tree. We need the children result before |
| 466 | * we can evaluate the parent. |
| 467 | */ |
| 468 | static |
| 469 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, |
| 470 | struct ir_op *node) |
| 471 | { |
| 472 | switch (node->op) { |
| 473 | case IR_OP_UNKNOWN: |
| 474 | default: |
| 475 | fprintf(stderr, "[error] Unknown node type in %s\n", |
| 476 | __func__); |
| 477 | return -EINVAL; |
| 478 | |
| 479 | case IR_OP_ROOT: |
| 480 | return visit_node_root(ctx, node); |
| 481 | case IR_OP_LOAD: |
| 482 | return visit_node_load(ctx, node); |
| 483 | case IR_OP_UNARY: |
| 484 | return visit_node_unary(ctx, node); |
| 485 | case IR_OP_BINARY: |
| 486 | return visit_node_binary(ctx, node); |
| 487 | case IR_OP_LOGICAL: |
| 488 | return visit_node_logical(ctx, node); |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | LTTNG_HIDDEN |
| 493 | void filter_bytecode_free(struct filter_parser_ctx *ctx) |
| 494 | { |
| 495 | if (!ctx) { |
| 496 | return; |
| 497 | } |
| 498 | |
| 499 | if (ctx->bytecode) { |
| 500 | free(ctx->bytecode); |
| 501 | ctx->bytecode = NULL; |
| 502 | } |
| 503 | |
| 504 | if (ctx->bytecode_reloc) { |
| 505 | free(ctx->bytecode_reloc); |
| 506 | ctx->bytecode_reloc = NULL; |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | LTTNG_HIDDEN |
| 511 | int filter_visitor_bytecode_generate(struct filter_parser_ctx *ctx) |
| 512 | { |
| 513 | int ret; |
| 514 | |
| 515 | ret = bytecode_init(&ctx->bytecode); |
| 516 | if (ret) |
| 517 | return ret; |
| 518 | ret = bytecode_init(&ctx->bytecode_reloc); |
| 519 | if (ret) |
| 520 | goto error; |
| 521 | ret = recursive_visit_gen_bytecode(ctx, ctx->ir_root); |
| 522 | if (ret) |
| 523 | goto error; |
| 524 | |
| 525 | /* Finally, append symbol table to bytecode */ |
| 526 | ctx->bytecode->b.reloc_table_offset = bytecode_get_len(&ctx->bytecode->b); |
| 527 | return bytecode_push(&ctx->bytecode, ctx->bytecode_reloc->b.data, |
| 528 | 1, bytecode_get_len(&ctx->bytecode_reloc->b)); |
| 529 | |
| 530 | error: |
| 531 | filter_bytecode_free(ctx); |
| 532 | return ret; |
| 533 | } |