| 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 "align.h" |
| 26 | #include "filter-bytecode.h" |
| 27 | #include "filter-ir.h" |
| 28 | #include "filter-ast.h" |
| 29 | |
| 30 | #ifndef max_t |
| 31 | #define max_t(type, a, b) ((type) ((a) > (b) ? (a) : (b))) |
| 32 | #endif |
| 33 | |
| 34 | //#define INIT_ALLOC_SIZE PAGE_SIZE |
| 35 | #define INIT_ALLOC_SIZE 4 |
| 36 | |
| 37 | static |
| 38 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, |
| 39 | struct ir_op *node); |
| 40 | |
| 41 | static |
| 42 | int bytecode_init(struct lttng_filter_bytecode_alloc **fb) |
| 43 | { |
| 44 | *fb = calloc(sizeof(struct lttng_filter_bytecode_alloc) + INIT_ALLOC_SIZE, 1); |
| 45 | if (!*fb) { |
| 46 | return -ENOMEM; |
| 47 | } else { |
| 48 | (*fb)->alloc_len = INIT_ALLOC_SIZE; |
| 49 | return 0; |
| 50 | } |
| 51 | } |
| 52 | |
| 53 | static |
| 54 | int32_t bytecode_reserve(struct lttng_filter_bytecode_alloc **fb, uint32_t align, uint32_t len) |
| 55 | { |
| 56 | int32_t ret; |
| 57 | uint32_t padding = offset_align((*fb)->b.len, align); |
| 58 | |
| 59 | if ((*fb)->b.len + padding + len > (*fb)->alloc_len) { |
| 60 | uint32_t new_len = |
| 61 | max_t(uint32_t, (*fb)->b.len + padding + len, |
| 62 | (*fb)->alloc_len << 1); |
| 63 | uint32_t old_len = (*fb)->alloc_len; |
| 64 | |
| 65 | if (new_len > 0xFFFF) |
| 66 | return -EINVAL; |
| 67 | *fb = realloc(*fb, sizeof(struct lttng_filter_bytecode_alloc) + new_len); |
| 68 | if (!*fb) |
| 69 | return -ENOMEM; |
| 70 | memset(&(*fb)->b.data[old_len], 0, new_len - old_len); |
| 71 | (*fb)->alloc_len = new_len; |
| 72 | } |
| 73 | (*fb)->b.len += padding; |
| 74 | ret = (*fb)->b.len; |
| 75 | (*fb)->b.len += len; |
| 76 | return ret; |
| 77 | } |
| 78 | |
| 79 | static |
| 80 | int bytecode_push(struct lttng_filter_bytecode_alloc **fb, const void *data, |
| 81 | uint32_t align, uint32_t len) |
| 82 | { |
| 83 | int32_t offset; |
| 84 | |
| 85 | offset = bytecode_reserve(fb, align, len); |
| 86 | if (offset < 0) |
| 87 | return offset; |
| 88 | memcpy(&(*fb)->b.data[offset], data, len); |
| 89 | return 0; |
| 90 | } |
| 91 | |
| 92 | static |
| 93 | int bytecode_push_logical(struct lttng_filter_bytecode_alloc **fb, |
| 94 | struct logical_op *data, |
| 95 | uint32_t align, uint32_t len, |
| 96 | uint16_t *skip_offset) |
| 97 | { |
| 98 | int32_t offset; |
| 99 | |
| 100 | offset = bytecode_reserve(fb, align, len); |
| 101 | if (offset < 0) |
| 102 | return offset; |
| 103 | memcpy(&(*fb)->b.data[offset], data, len); |
| 104 | *skip_offset = |
| 105 | (void *) &((struct logical_op *) &(*fb)->b.data[offset])->skip_offset |
| 106 | - (void *) &(*fb)->b.data[0]; |
| 107 | return 0; |
| 108 | } |
| 109 | |
| 110 | static |
| 111 | int bytecode_patch(struct lttng_filter_bytecode_alloc **fb, |
| 112 | const void *data, |
| 113 | uint16_t offset, |
| 114 | uint32_t len) |
| 115 | { |
| 116 | if (offset >= (*fb)->b.len) { |
| 117 | return -EINVAL; |
| 118 | } |
| 119 | memcpy(&(*fb)->b.data[offset], data, len); |
| 120 | return 0; |
| 121 | } |
| 122 | |
| 123 | static |
| 124 | int visit_node_root(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 125 | { |
| 126 | int ret; |
| 127 | struct return_op insn; |
| 128 | |
| 129 | /* Visit child */ |
| 130 | ret = recursive_visit_gen_bytecode(ctx, node->u.root.child); |
| 131 | if (ret) |
| 132 | return ret; |
| 133 | |
| 134 | /* Generate end of bytecode instruction */ |
| 135 | insn.op = FILTER_OP_RETURN; |
| 136 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 137 | } |
| 138 | |
| 139 | static |
| 140 | int visit_node_load(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 141 | { |
| 142 | int ret; |
| 143 | |
| 144 | switch (node->data_type) { |
| 145 | case IR_DATA_UNKNOWN: |
| 146 | default: |
| 147 | fprintf(stderr, "[error] Unknown data type in %s\n", |
| 148 | __func__); |
| 149 | return -EINVAL; |
| 150 | |
| 151 | case IR_DATA_STRING: |
| 152 | { |
| 153 | struct load_op *insn; |
| 154 | uint32_t insn_len = sizeof(struct load_op) |
| 155 | + strlen(node->u.load.u.string) + 1; |
| 156 | |
| 157 | insn = calloc(insn_len, 1); |
| 158 | if (!insn) |
| 159 | return -ENOMEM; |
| 160 | insn->op = FILTER_OP_LOAD_STRING; |
| 161 | strcpy(insn->data, node->u.load.u.string); |
| 162 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 163 | free(insn); |
| 164 | return ret; |
| 165 | } |
| 166 | case IR_DATA_NUMERIC: |
| 167 | { |
| 168 | struct load_op *insn; |
| 169 | uint32_t insn_len = sizeof(struct load_op) |
| 170 | + sizeof(struct literal_numeric); |
| 171 | |
| 172 | insn = calloc(insn_len, 1); |
| 173 | if (!insn) |
| 174 | return -ENOMEM; |
| 175 | insn->op = FILTER_OP_LOAD_S64; |
| 176 | *(int64_t *) insn->data = node->u.load.u.num; |
| 177 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 178 | free(insn); |
| 179 | return ret; |
| 180 | } |
| 181 | case IR_DATA_FLOAT: |
| 182 | { |
| 183 | struct load_op *insn; |
| 184 | uint32_t insn_len = sizeof(struct load_op) |
| 185 | + sizeof(struct literal_double); |
| 186 | |
| 187 | insn = calloc(insn_len, 1); |
| 188 | if (!insn) |
| 189 | return -ENOMEM; |
| 190 | insn->op = FILTER_OP_LOAD_DOUBLE; |
| 191 | *(double *) insn->data = node->u.load.u.flt; |
| 192 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 193 | free(insn); |
| 194 | return ret; |
| 195 | } |
| 196 | case IR_DATA_FIELD_REF: |
| 197 | { |
| 198 | struct load_op *insn; |
| 199 | uint32_t insn_len = sizeof(struct load_op) |
| 200 | + sizeof(struct field_ref); |
| 201 | struct field_ref ref_offset; |
| 202 | uint16_t reloc_offset; |
| 203 | |
| 204 | insn = calloc(insn_len, 1); |
| 205 | if (!insn) |
| 206 | return -ENOMEM; |
| 207 | insn->op = FILTER_OP_LOAD_FIELD_REF; |
| 208 | ref_offset.offset = (uint16_t) -1U; |
| 209 | memcpy(insn->data, &ref_offset, sizeof(ref_offset)); |
| 210 | /* reloc_offset points to struct load_op */ |
| 211 | reloc_offset = bytecode_get_len(&ctx->bytecode->b); |
| 212 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
| 213 | if (ret) { |
| 214 | free(insn); |
| 215 | return ret; |
| 216 | } |
| 217 | /* append reloc */ |
| 218 | ret = bytecode_push(&ctx->bytecode_reloc, &reloc_offset, |
| 219 | 1, sizeof(reloc_offset)); |
| 220 | if (ret) { |
| 221 | free(insn); |
| 222 | return ret; |
| 223 | } |
| 224 | ret = bytecode_push(&ctx->bytecode_reloc, node->u.load.u.ref, |
| 225 | 1, strlen(node->u.load.u.ref) + 1); |
| 226 | free(insn); |
| 227 | return ret; |
| 228 | } |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | static |
| 233 | int visit_node_unary(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 234 | { |
| 235 | int ret; |
| 236 | struct unary_op insn; |
| 237 | |
| 238 | /* Visit child */ |
| 239 | ret = recursive_visit_gen_bytecode(ctx, node->u.unary.child); |
| 240 | if (ret) |
| 241 | return ret; |
| 242 | |
| 243 | /* Generate end of bytecode instruction */ |
| 244 | switch (node->u.unary.type) { |
| 245 | case AST_UNARY_UNKNOWN: |
| 246 | default: |
| 247 | fprintf(stderr, "[error] Unknown unary node type in %s\n", |
| 248 | __func__); |
| 249 | return -EINVAL; |
| 250 | case AST_UNARY_PLUS: |
| 251 | /* Nothing to do. */ |
| 252 | return 0; |
| 253 | case AST_UNARY_MINUS: |
| 254 | insn.op = FILTER_OP_UNARY_MINUS; |
| 255 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 256 | case AST_UNARY_NOT: |
| 257 | insn.op = FILTER_OP_UNARY_NOT; |
| 258 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | /* |
| 263 | * Binary comparator nesting is disallowed. This allows fitting into |
| 264 | * only 2 registers. |
| 265 | */ |
| 266 | static |
| 267 | int visit_node_binary(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 268 | { |
| 269 | int ret; |
| 270 | struct binary_op insn; |
| 271 | |
| 272 | /* Visit child */ |
| 273 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); |
| 274 | if (ret) |
| 275 | return ret; |
| 276 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); |
| 277 | if (ret) |
| 278 | return ret; |
| 279 | |
| 280 | switch (node->u.binary.type) { |
| 281 | case AST_OP_UNKNOWN: |
| 282 | default: |
| 283 | fprintf(stderr, "[error] Unknown unary node type in %s\n", |
| 284 | __func__); |
| 285 | return -EINVAL; |
| 286 | |
| 287 | case AST_OP_AND: |
| 288 | case AST_OP_OR: |
| 289 | fprintf(stderr, "[error] Unexpected logical node type in %s\n", |
| 290 | __func__); |
| 291 | return -EINVAL; |
| 292 | |
| 293 | case AST_OP_MUL: |
| 294 | insn.op = FILTER_OP_MUL; |
| 295 | break; |
| 296 | case AST_OP_DIV: |
| 297 | insn.op = FILTER_OP_DIV; |
| 298 | break; |
| 299 | case AST_OP_MOD: |
| 300 | insn.op = FILTER_OP_MOD; |
| 301 | break; |
| 302 | case AST_OP_PLUS: |
| 303 | insn.op = FILTER_OP_PLUS; |
| 304 | break; |
| 305 | case AST_OP_MINUS: |
| 306 | insn.op = FILTER_OP_MINUS; |
| 307 | break; |
| 308 | case AST_OP_RSHIFT: |
| 309 | insn.op = FILTER_OP_RSHIFT; |
| 310 | break; |
| 311 | case AST_OP_LSHIFT: |
| 312 | insn.op = FILTER_OP_LSHIFT; |
| 313 | break; |
| 314 | case AST_OP_BIN_AND: |
| 315 | insn.op = FILTER_OP_BIN_AND; |
| 316 | break; |
| 317 | case AST_OP_BIN_OR: |
| 318 | insn.op = FILTER_OP_BIN_OR; |
| 319 | break; |
| 320 | case AST_OP_BIN_XOR: |
| 321 | insn.op = FILTER_OP_BIN_XOR; |
| 322 | break; |
| 323 | |
| 324 | case AST_OP_EQ: |
| 325 | insn.op = FILTER_OP_EQ; |
| 326 | break; |
| 327 | case AST_OP_NE: |
| 328 | insn.op = FILTER_OP_NE; |
| 329 | break; |
| 330 | case AST_OP_GT: |
| 331 | insn.op = FILTER_OP_GT; |
| 332 | break; |
| 333 | case AST_OP_LT: |
| 334 | insn.op = FILTER_OP_LT; |
| 335 | break; |
| 336 | case AST_OP_GE: |
| 337 | insn.op = FILTER_OP_GE; |
| 338 | break; |
| 339 | case AST_OP_LE: |
| 340 | insn.op = FILTER_OP_LE; |
| 341 | break; |
| 342 | } |
| 343 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
| 344 | } |
| 345 | |
| 346 | /* |
| 347 | * A logical op always return a s64 (1 or 0). |
| 348 | */ |
| 349 | static |
| 350 | int visit_node_logical(struct filter_parser_ctx *ctx, struct ir_op *node) |
| 351 | { |
| 352 | int ret; |
| 353 | struct logical_op insn; |
| 354 | uint16_t skip_offset_loc; |
| 355 | uint16_t target_loc; |
| 356 | |
| 357 | /* Visit left child */ |
| 358 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); |
| 359 | if (ret) |
| 360 | return ret; |
| 361 | /* Cast to s64 if float or field ref */ |
| 362 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF |
| 363 | || node->u.binary.left->data_type == IR_DATA_FLOAT) { |
| 364 | struct cast_op cast_insn; |
| 365 | |
| 366 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF) { |
| 367 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
| 368 | } else { |
| 369 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; |
| 370 | } |
| 371 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
| 372 | 1, sizeof(cast_insn)); |
| 373 | if (ret) |
| 374 | return ret; |
| 375 | } |
| 376 | switch (node->u.logical.type) { |
| 377 | default: |
| 378 | fprintf(stderr, "[error] Unknown node type in %s\n", |
| 379 | __func__); |
| 380 | return -EINVAL; |
| 381 | |
| 382 | case AST_OP_AND: |
| 383 | insn.op = FILTER_OP_AND; |
| 384 | break; |
| 385 | case AST_OP_OR: |
| 386 | insn.op = FILTER_OP_OR; |
| 387 | break; |
| 388 | } |
| 389 | insn.skip_offset = (uint16_t) -1UL; /* Temporary */ |
| 390 | ret = bytecode_push_logical(&ctx->bytecode, &insn, 1, sizeof(insn), |
| 391 | &skip_offset_loc); |
| 392 | if (ret) |
| 393 | return ret; |
| 394 | /* Visit right child */ |
| 395 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); |
| 396 | if (ret) |
| 397 | return ret; |
| 398 | /* Cast to s64 if float or field ref */ |
| 399 | if (node->u.binary.right->data_type == IR_DATA_FIELD_REF |
| 400 | || node->u.binary.right->data_type == IR_DATA_FLOAT) { |
| 401 | struct cast_op cast_insn; |
| 402 | |
| 403 | if (node->u.binary.right->data_type == IR_DATA_FIELD_REF) { |
| 404 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
| 405 | } else { |
| 406 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; |
| 407 | } |
| 408 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
| 409 | 1, sizeof(cast_insn)); |
| 410 | if (ret) |
| 411 | return ret; |
| 412 | } |
| 413 | /* We now know where the logical op can skip. */ |
| 414 | target_loc = (uint16_t) bytecode_get_len(&ctx->bytecode->b); |
| 415 | ret = bytecode_patch(&ctx->bytecode, |
| 416 | &target_loc, /* Offset to jump to */ |
| 417 | skip_offset_loc, /* Where to patch */ |
| 418 | sizeof(uint16_t)); |
| 419 | return ret; |
| 420 | } |
| 421 | |
| 422 | /* |
| 423 | * Postorder traversal of the tree. We need the children result before |
| 424 | * we can evaluate the parent. |
| 425 | */ |
| 426 | static |
| 427 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, |
| 428 | struct ir_op *node) |
| 429 | { |
| 430 | switch (node->op) { |
| 431 | case IR_OP_UNKNOWN: |
| 432 | default: |
| 433 | fprintf(stderr, "[error] Unknown node type in %s\n", |
| 434 | __func__); |
| 435 | return -EINVAL; |
| 436 | |
| 437 | case IR_OP_ROOT: |
| 438 | return visit_node_root(ctx, node); |
| 439 | case IR_OP_LOAD: |
| 440 | return visit_node_load(ctx, node); |
| 441 | case IR_OP_UNARY: |
| 442 | return visit_node_unary(ctx, node); |
| 443 | case IR_OP_BINARY: |
| 444 | return visit_node_binary(ctx, node); |
| 445 | case IR_OP_LOGICAL: |
| 446 | return visit_node_logical(ctx, node); |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | void filter_bytecode_free(struct filter_parser_ctx *ctx) |
| 451 | { |
| 452 | free(ctx->bytecode); |
| 453 | ctx->bytecode = NULL; |
| 454 | free(ctx->bytecode_reloc); |
| 455 | ctx->bytecode_reloc = NULL; |
| 456 | } |
| 457 | |
| 458 | int filter_visitor_bytecode_generate(struct filter_parser_ctx *ctx) |
| 459 | { |
| 460 | int ret; |
| 461 | |
| 462 | ret = bytecode_init(&ctx->bytecode); |
| 463 | if (ret) |
| 464 | return ret; |
| 465 | ret = bytecode_init(&ctx->bytecode_reloc); |
| 466 | if (ret) |
| 467 | goto error; |
| 468 | ret = recursive_visit_gen_bytecode(ctx, ctx->ir_root); |
| 469 | if (ret) |
| 470 | goto error; |
| 471 | |
| 472 | /* Finally, append symbol table to bytecode */ |
| 473 | ctx->bytecode->b.reloc_table_offset = bytecode_get_len(&ctx->bytecode->b); |
| 474 | return bytecode_push(&ctx->bytecode, ctx->bytecode_reloc->b.data, |
| 475 | 1, bytecode_get_len(&ctx->bytecode_reloc->b)); |
| 476 | |
| 477 | error: |
| 478 | filter_bytecode_free(ctx); |
| 479 | return ret; |
| 480 | } |