Commit | Line | Data |
---|---|---|
953192ba MD |
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 | ||
01a204f0 CB |
41 | static inline int fls(unsigned int x) |
42 | { | |
43 | int r = 32; | |
44 | ||
45 | if (!x) | |
46 | return 0; | |
47 | if (!(x & 0xFFFF0000U)) { | |
48 | x <<= 16; | |
49 | r -= 16; | |
50 | } | |
51 | if (!(x & 0xFF000000U)) { | |
52 | x <<= 8; | |
53 | r -= 8; | |
54 | } | |
55 | if (!(x & 0xF0000000U)) { | |
56 | x <<= 4; | |
57 | r -= 4; | |
58 | } | |
59 | if (!(x & 0xC0000000U)) { | |
60 | x <<= 2; | |
61 | r -= 2; | |
62 | } | |
63 | if (!(x & 0x80000000U)) { | |
64 | x <<= 1; | |
65 | r -= 1; | |
66 | } | |
67 | return r; | |
68 | } | |
69 | ||
70 | static inline int get_count_order(unsigned int count) | |
71 | { | |
72 | int order; | |
73 | ||
74 | order = fls(count) - 1; | |
75 | if (count & (count - 1)) | |
76 | order++; | |
77 | return order; | |
78 | } | |
79 | ||
953192ba | 80 | static |
53a80697 | 81 | int bytecode_init(struct lttng_filter_bytecode_alloc **fb) |
953192ba | 82 | { |
53a80697 | 83 | *fb = calloc(sizeof(struct lttng_filter_bytecode_alloc) + INIT_ALLOC_SIZE, 1); |
953192ba MD |
84 | if (!*fb) { |
85 | return -ENOMEM; | |
86 | } else { | |
87 | (*fb)->alloc_len = INIT_ALLOC_SIZE; | |
88 | return 0; | |
89 | } | |
90 | } | |
91 | ||
92 | static | |
53a80697 | 93 | int32_t bytecode_reserve(struct lttng_filter_bytecode_alloc **fb, uint32_t align, uint32_t len) |
953192ba MD |
94 | { |
95 | int32_t ret; | |
96 | uint32_t padding = offset_align((*fb)->b.len, align); | |
97 | ||
5ddb0a08 CB |
98 | if ((*fb)->b.len + padding + len > LTTNG_FILTER_MAX_LEN) |
99 | return -EINVAL; | |
100 | ||
953192ba MD |
101 | if ((*fb)->b.len + padding + len > (*fb)->alloc_len) { |
102 | uint32_t new_len = | |
01a204f0 | 103 | max_t(uint32_t, 1U << get_count_order((*fb)->b.len + padding + len), |
953192ba MD |
104 | (*fb)->alloc_len << 1); |
105 | uint32_t old_len = (*fb)->alloc_len; | |
106 | ||
53a80697 | 107 | *fb = realloc(*fb, sizeof(struct lttng_filter_bytecode_alloc) + new_len); |
953192ba MD |
108 | if (!*fb) |
109 | return -ENOMEM; | |
110 | memset(&(*fb)->b.data[old_len], 0, new_len - old_len); | |
111 | (*fb)->alloc_len = new_len; | |
112 | } | |
113 | (*fb)->b.len += padding; | |
114 | ret = (*fb)->b.len; | |
115 | (*fb)->b.len += len; | |
116 | return ret; | |
117 | } | |
118 | ||
119 | static | |
53a80697 | 120 | int bytecode_push(struct lttng_filter_bytecode_alloc **fb, const void *data, |
953192ba MD |
121 | uint32_t align, uint32_t len) |
122 | { | |
123 | int32_t offset; | |
124 | ||
125 | offset = bytecode_reserve(fb, align, len); | |
126 | if (offset < 0) | |
127 | return offset; | |
128 | memcpy(&(*fb)->b.data[offset], data, len); | |
129 | return 0; | |
130 | } | |
131 | ||
132 | static | |
53a80697 | 133 | int bytecode_push_logical(struct lttng_filter_bytecode_alloc **fb, |
953192ba MD |
134 | struct logical_op *data, |
135 | uint32_t align, uint32_t len, | |
136 | uint16_t *skip_offset) | |
137 | { | |
138 | int32_t offset; | |
139 | ||
140 | offset = bytecode_reserve(fb, align, len); | |
141 | if (offset < 0) | |
142 | return offset; | |
143 | memcpy(&(*fb)->b.data[offset], data, len); | |
144 | *skip_offset = | |
145 | (void *) &((struct logical_op *) &(*fb)->b.data[offset])->skip_offset | |
146 | - (void *) &(*fb)->b.data[0]; | |
147 | return 0; | |
148 | } | |
149 | ||
150 | static | |
53a80697 | 151 | int bytecode_patch(struct lttng_filter_bytecode_alloc **fb, |
953192ba MD |
152 | const void *data, |
153 | uint16_t offset, | |
154 | uint32_t len) | |
155 | { | |
156 | if (offset >= (*fb)->b.len) { | |
157 | return -EINVAL; | |
158 | } | |
159 | memcpy(&(*fb)->b.data[offset], data, len); | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static | |
164 | int visit_node_root(struct filter_parser_ctx *ctx, struct ir_op *node) | |
165 | { | |
166 | int ret; | |
167 | struct return_op insn; | |
168 | ||
169 | /* Visit child */ | |
170 | ret = recursive_visit_gen_bytecode(ctx, node->u.root.child); | |
171 | if (ret) | |
172 | return ret; | |
173 | ||
174 | /* Generate end of bytecode instruction */ | |
175 | insn.op = FILTER_OP_RETURN; | |
176 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); | |
177 | } | |
178 | ||
953192ba MD |
179 | static |
180 | int visit_node_load(struct filter_parser_ctx *ctx, struct ir_op *node) | |
181 | { | |
182 | int ret; | |
183 | ||
184 | switch (node->data_type) { | |
185 | case IR_DATA_UNKNOWN: | |
186 | default: | |
187 | fprintf(stderr, "[error] Unknown data type in %s\n", | |
188 | __func__); | |
189 | return -EINVAL; | |
190 | ||
191 | case IR_DATA_STRING: | |
192 | { | |
193 | struct load_op *insn; | |
194 | uint32_t insn_len = sizeof(struct load_op) | |
195 | + strlen(node->u.load.u.string) + 1; | |
196 | ||
197 | insn = calloc(insn_len, 1); | |
198 | if (!insn) | |
199 | return -ENOMEM; | |
200 | insn->op = FILTER_OP_LOAD_STRING; | |
953192ba MD |
201 | strcpy(insn->data, node->u.load.u.string); |
202 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
203 | free(insn); | |
204 | return ret; | |
205 | } | |
206 | case IR_DATA_NUMERIC: | |
207 | { | |
208 | struct load_op *insn; | |
209 | uint32_t insn_len = sizeof(struct load_op) | |
210 | + sizeof(struct literal_numeric); | |
211 | ||
212 | insn = calloc(insn_len, 1); | |
213 | if (!insn) | |
214 | return -ENOMEM; | |
215 | insn->op = FILTER_OP_LOAD_S64; | |
953192ba MD |
216 | *(int64_t *) insn->data = node->u.load.u.num; |
217 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
218 | free(insn); | |
219 | return ret; | |
220 | } | |
e90d8561 MD |
221 | case IR_DATA_FLOAT: |
222 | { | |
223 | struct load_op *insn; | |
224 | uint32_t insn_len = sizeof(struct load_op) | |
225 | + sizeof(struct literal_double); | |
226 | ||
227 | insn = calloc(insn_len, 1); | |
228 | if (!insn) | |
229 | return -ENOMEM; | |
230 | insn->op = FILTER_OP_LOAD_DOUBLE; | |
e90d8561 MD |
231 | *(double *) insn->data = node->u.load.u.flt; |
232 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
233 | free(insn); | |
234 | return ret; | |
235 | } | |
953192ba MD |
236 | case IR_DATA_FIELD_REF: |
237 | { | |
238 | struct load_op *insn; | |
239 | uint32_t insn_len = sizeof(struct load_op) | |
240 | + sizeof(struct field_ref); | |
241 | struct field_ref ref_offset; | |
d93c4f1f | 242 | uint32_t reloc_offset; |
953192ba MD |
243 | |
244 | insn = calloc(insn_len, 1); | |
245 | if (!insn) | |
246 | return -ENOMEM; | |
247 | insn->op = FILTER_OP_LOAD_FIELD_REF; | |
953192ba MD |
248 | ref_offset.offset = (uint16_t) -1U; |
249 | memcpy(insn->data, &ref_offset, sizeof(ref_offset)); | |
65775683 | 250 | /* reloc_offset points to struct load_op */ |
953192ba | 251 | reloc_offset = bytecode_get_len(&ctx->bytecode->b); |
953192ba MD |
252 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
253 | if (ret) { | |
254 | free(insn); | |
255 | return ret; | |
256 | } | |
257 | /* append reloc */ | |
258 | ret = bytecode_push(&ctx->bytecode_reloc, &reloc_offset, | |
259 | 1, sizeof(reloc_offset)); | |
260 | if (ret) { | |
261 | free(insn); | |
262 | return ret; | |
263 | } | |
264 | ret = bytecode_push(&ctx->bytecode_reloc, node->u.load.u.ref, | |
265 | 1, strlen(node->u.load.u.ref) + 1); | |
266 | free(insn); | |
267 | return ret; | |
268 | } | |
269 | } | |
270 | } | |
271 | ||
272 | static | |
273 | int visit_node_unary(struct filter_parser_ctx *ctx, struct ir_op *node) | |
274 | { | |
275 | int ret; | |
276 | struct unary_op insn; | |
277 | ||
278 | /* Visit child */ | |
279 | ret = recursive_visit_gen_bytecode(ctx, node->u.unary.child); | |
280 | if (ret) | |
281 | return ret; | |
282 | ||
283 | /* Generate end of bytecode instruction */ | |
284 | switch (node->u.unary.type) { | |
285 | case AST_UNARY_UNKNOWN: | |
286 | default: | |
287 | fprintf(stderr, "[error] Unknown unary node type in %s\n", | |
288 | __func__); | |
289 | return -EINVAL; | |
290 | case AST_UNARY_PLUS: | |
291 | /* Nothing to do. */ | |
292 | return 0; | |
293 | case AST_UNARY_MINUS: | |
294 | insn.op = FILTER_OP_UNARY_MINUS; | |
953192ba MD |
295 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
296 | case AST_UNARY_NOT: | |
297 | insn.op = FILTER_OP_UNARY_NOT; | |
953192ba MD |
298 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
299 | } | |
300 | } | |
301 | ||
302 | /* | |
303 | * Binary comparator nesting is disallowed. This allows fitting into | |
304 | * only 2 registers. | |
305 | */ | |
306 | static | |
307 | int visit_node_binary(struct filter_parser_ctx *ctx, struct ir_op *node) | |
308 | { | |
309 | int ret; | |
310 | struct binary_op insn; | |
311 | ||
312 | /* Visit child */ | |
313 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); | |
314 | if (ret) | |
315 | return ret; | |
316 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); | |
317 | if (ret) | |
318 | return ret; | |
319 | ||
320 | switch (node->u.binary.type) { | |
321 | case AST_OP_UNKNOWN: | |
322 | default: | |
323 | fprintf(stderr, "[error] Unknown unary node type in %s\n", | |
324 | __func__); | |
325 | return -EINVAL; | |
326 | ||
327 | case AST_OP_AND: | |
328 | case AST_OP_OR: | |
329 | fprintf(stderr, "[error] Unexpected logical node type in %s\n", | |
330 | __func__); | |
331 | return -EINVAL; | |
332 | ||
333 | case AST_OP_MUL: | |
334 | insn.op = FILTER_OP_MUL; | |
335 | break; | |
336 | case AST_OP_DIV: | |
337 | insn.op = FILTER_OP_DIV; | |
338 | break; | |
339 | case AST_OP_MOD: | |
340 | insn.op = FILTER_OP_MOD; | |
341 | break; | |
342 | case AST_OP_PLUS: | |
343 | insn.op = FILTER_OP_PLUS; | |
344 | break; | |
345 | case AST_OP_MINUS: | |
346 | insn.op = FILTER_OP_MINUS; | |
347 | break; | |
348 | case AST_OP_RSHIFT: | |
349 | insn.op = FILTER_OP_RSHIFT; | |
350 | break; | |
351 | case AST_OP_LSHIFT: | |
352 | insn.op = FILTER_OP_LSHIFT; | |
353 | break; | |
354 | case AST_OP_BIN_AND: | |
355 | insn.op = FILTER_OP_BIN_AND; | |
356 | break; | |
357 | case AST_OP_BIN_OR: | |
358 | insn.op = FILTER_OP_BIN_OR; | |
359 | break; | |
360 | case AST_OP_BIN_XOR: | |
361 | insn.op = FILTER_OP_BIN_XOR; | |
362 | break; | |
363 | ||
364 | case AST_OP_EQ: | |
365 | insn.op = FILTER_OP_EQ; | |
366 | break; | |
367 | case AST_OP_NE: | |
368 | insn.op = FILTER_OP_NE; | |
369 | break; | |
370 | case AST_OP_GT: | |
371 | insn.op = FILTER_OP_GT; | |
372 | break; | |
373 | case AST_OP_LT: | |
374 | insn.op = FILTER_OP_LT; | |
375 | break; | |
376 | case AST_OP_GE: | |
377 | insn.op = FILTER_OP_GE; | |
378 | break; | |
379 | case AST_OP_LE: | |
380 | insn.op = FILTER_OP_LE; | |
381 | break; | |
382 | } | |
383 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); | |
384 | } | |
385 | ||
8cf9540a MD |
386 | /* |
387 | * A logical op always return a s64 (1 or 0). | |
388 | */ | |
953192ba MD |
389 | static |
390 | int visit_node_logical(struct filter_parser_ctx *ctx, struct ir_op *node) | |
391 | { | |
392 | int ret; | |
393 | struct logical_op insn; | |
394 | uint16_t skip_offset_loc; | |
395 | uint16_t target_loc; | |
396 | ||
397 | /* Visit left child */ | |
398 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); | |
399 | if (ret) | |
400 | return ret; | |
8cf9540a MD |
401 | /* Cast to s64 if float or field ref */ |
402 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF | |
403 | || node->u.binary.left->data_type == IR_DATA_FLOAT) { | |
404 | struct cast_op cast_insn; | |
405 | ||
29fefef8 MD |
406 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF) { |
407 | cast_insn.op = FILTER_OP_CAST_TO_S64; | |
408 | } else { | |
409 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; | |
410 | } | |
8cf9540a MD |
411 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
412 | 1, sizeof(cast_insn)); | |
413 | if (ret) | |
414 | return ret; | |
415 | } | |
953192ba MD |
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; | |
8cf9540a MD |
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_FLOAT) { | |
441 | struct cast_op cast_insn; | |
442 | ||
29fefef8 MD |
443 | if (node->u.binary.right->data_type == IR_DATA_FIELD_REF) { |
444 | cast_insn.op = FILTER_OP_CAST_TO_S64; | |
445 | } else { | |
446 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; | |
447 | } | |
8cf9540a MD |
448 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
449 | 1, sizeof(cast_insn)); | |
450 | if (ret) | |
451 | return ret; | |
452 | } | |
953192ba MD |
453 | /* We now know where the logical op can skip. */ |
454 | target_loc = (uint16_t) bytecode_get_len(&ctx->bytecode->b); | |
455 | ret = bytecode_patch(&ctx->bytecode, | |
456 | &target_loc, /* Offset to jump to */ | |
457 | skip_offset_loc, /* Where to patch */ | |
458 | sizeof(uint16_t)); | |
459 | return ret; | |
460 | } | |
461 | ||
462 | /* | |
463 | * Postorder traversal of the tree. We need the children result before | |
464 | * we can evaluate the parent. | |
465 | */ | |
466 | static | |
467 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, | |
468 | struct ir_op *node) | |
469 | { | |
470 | switch (node->op) { | |
471 | case IR_OP_UNKNOWN: | |
472 | default: | |
473 | fprintf(stderr, "[error] Unknown node type in %s\n", | |
474 | __func__); | |
475 | return -EINVAL; | |
476 | ||
477 | case IR_OP_ROOT: | |
478 | return visit_node_root(ctx, node); | |
479 | case IR_OP_LOAD: | |
480 | return visit_node_load(ctx, node); | |
481 | case IR_OP_UNARY: | |
482 | return visit_node_unary(ctx, node); | |
483 | case IR_OP_BINARY: | |
484 | return visit_node_binary(ctx, node); | |
485 | case IR_OP_LOGICAL: | |
486 | return visit_node_logical(ctx, node); | |
487 | } | |
488 | } | |
489 | ||
d00c599e | 490 | __attribute__((visibility("hidden"))) |
953192ba MD |
491 | void filter_bytecode_free(struct filter_parser_ctx *ctx) |
492 | { | |
493 | free(ctx->bytecode); | |
494 | ctx->bytecode = NULL; | |
495 | free(ctx->bytecode_reloc); | |
496 | ctx->bytecode_reloc = NULL; | |
497 | } | |
498 | ||
d00c599e | 499 | __attribute__((visibility("hidden"))) |
953192ba MD |
500 | int filter_visitor_bytecode_generate(struct filter_parser_ctx *ctx) |
501 | { | |
502 | int ret; | |
503 | ||
504 | ret = bytecode_init(&ctx->bytecode); | |
505 | if (ret) | |
506 | return ret; | |
507 | ret = bytecode_init(&ctx->bytecode_reloc); | |
508 | if (ret) | |
509 | goto error; | |
510 | ret = recursive_visit_gen_bytecode(ctx, ctx->ir_root); | |
511 | if (ret) | |
512 | goto error; | |
513 | ||
514 | /* Finally, append symbol table to bytecode */ | |
515 | ctx->bytecode->b.reloc_table_offset = bytecode_get_len(&ctx->bytecode->b); | |
516 | return bytecode_push(&ctx->bytecode, ctx->bytecode_reloc->b.data, | |
517 | 1, bytecode_get_len(&ctx->bytecode_reloc->b)); | |
518 | ||
519 | error: | |
520 | filter_bytecode_free(ctx); | |
521 | return ret; | |
522 | } |