| 1 | /* |
| 2 | * trie.c: implementation of trie.h. |
| 3 | */ |
| 4 | |
| 5 | #include "agedu.h" |
| 6 | #include "alloc.h" |
| 7 | #include "trie.h" |
| 8 | |
| 9 | #define alignof(typ) ( offsetof(struct { char c; typ t; }, t) ) |
| 10 | |
| 11 | /* |
| 12 | * Compare functions for pathnames. Returns the relative order of |
| 13 | * the names, like strcmp; also passes back the offset of the |
| 14 | * first differing character if desired. |
| 15 | */ |
| 16 | static int trieccmp(unsigned char a, unsigned char b) |
| 17 | { |
| 18 | int ia = (a == '\0' ? '\0' : a == pathsep ? '\1' : a+1); |
| 19 | int ib = (b == '\0' ? '\0' : b == pathsep ? '\1' : b+1); |
| 20 | return ia - ib; |
| 21 | } |
| 22 | |
| 23 | static int triencmp(const char *a, size_t alen, |
| 24 | const char *b, size_t blen, int *offset) |
| 25 | { |
| 26 | int off = 0; |
| 27 | while (off < alen && off < blen && a[off] == b[off]) |
| 28 | off++; |
| 29 | if (offset) |
| 30 | *offset = off; |
| 31 | if (off == alen || off == blen) return (off == blen) - (off == alen); |
| 32 | return trieccmp(a[off], b[off]); |
| 33 | } |
| 34 | |
| 35 | static int triecmp(const char *a, const char *b, int *offset) |
| 36 | { |
| 37 | return triencmp(a, strlen(a), b, strlen(b), offset); |
| 38 | } |
| 39 | |
| 40 | /* ---------------------------------------------------------------------- |
| 41 | * Trie node structures. |
| 42 | * |
| 43 | * The trie format stored in the file consists of three distinct |
| 44 | * node types, each with a distinguishing type field at the start. |
| 45 | * |
| 46 | * TRIE_LEAF is a leaf node; it contains an actual trie_file |
| 47 | * structure, and indicates that when you're searching down the |
| 48 | * trie with a string, you should now expect to encounter |
| 49 | * end-of-string. |
| 50 | * |
| 51 | * TRIE_SWITCH indicates that the set of strings in the trie |
| 52 | * include strings with more than one distinct character after the |
| 53 | * prefix leading up to this point. Hence, it stores multiple |
| 54 | * subnode pointers and a different single character for each one. |
| 55 | * |
| 56 | * TRIE_STRING indicates that at this point everything in the trie |
| 57 | * has the same next few characters; it stores a single mandatory |
| 58 | * string fragment and exactly one subnode pointer. |
| 59 | */ |
| 60 | enum { |
| 61 | TRIE_LEAF = 0x7fffe000, |
| 62 | TRIE_SWITCH, |
| 63 | TRIE_STRING |
| 64 | }; |
| 65 | |
| 66 | struct trie_common { |
| 67 | int type; |
| 68 | }; |
| 69 | |
| 70 | struct trie_switchentry { |
| 71 | off_t subnode; |
| 72 | int subcount; |
| 73 | }; |
| 74 | |
| 75 | struct trie_leaf { |
| 76 | struct trie_common c; |
| 77 | struct trie_file file; |
| 78 | }; |
| 79 | |
| 80 | struct trie_switch { |
| 81 | struct trie_common c; |
| 82 | /* |
| 83 | * sw[0] to sw[len-1] give the subnode pointers and element |
| 84 | * counts. At &sw[len] is stored len single bytes which are |
| 85 | * the characters corresponding to each subnode. |
| 86 | */ |
| 87 | int len; |
| 88 | struct trie_switchentry sw[]; |
| 89 | }; |
| 90 | |
| 91 | struct trie_string { |
| 92 | struct trie_common c; |
| 93 | int stringlen; |
| 94 | off_t subnode; |
| 95 | char string[]; |
| 96 | }; |
| 97 | |
| 98 | static const char magic_ident_string[16] = "agedu index file"; |
| 99 | struct trie_magic { |
| 100 | /* |
| 101 | * 'Magic numbers' to go at the start of an agedu index file. |
| 102 | * These are checked (using trie_check_magic) by every agedu mode |
| 103 | * which reads a pre-existing index. |
| 104 | * |
| 105 | * As well as identifying an agedu file from any other kind of |
| 106 | * file, this magic-number structure is also intended to detect |
| 107 | * agedu files which were written on the wrong platform and hence |
| 108 | * whose structure layouts are incompatible. To make that as |
| 109 | * reliable as possible, I design the structure of magic numbers |
| 110 | * as follows: it contains one of each integer type I might use, |
| 111 | * each containing a different magic number, and each followed by |
| 112 | * a char to indicate where it ends in the file. One byte is set |
| 113 | * to the length of the magic-number structure itself, which means |
| 114 | * that no two structures of different lengths can possibly |
| 115 | * compare equal even if by some chance they match up to the |
| 116 | * length of the shorter one. Finally, the whole magic number |
| 117 | * structure is memset to another random byte value before |
| 118 | * initialising any of these fields, so that padding in between |
| 119 | * can be readily identified. |
| 120 | */ |
| 121 | char ident[16]; /* human-readable string */ |
| 122 | |
| 123 | unsigned char magic_len; |
| 124 | |
| 125 | unsigned long long longlong_magic; |
| 126 | unsigned char postlonglong_char_magic; |
| 127 | |
| 128 | off_t offset_magic; |
| 129 | unsigned char postoffset_char_magic; |
| 130 | |
| 131 | size_t size_magic; |
| 132 | unsigned char postsize_char_magic; |
| 133 | |
| 134 | void *null_pointer; |
| 135 | unsigned char postptr_char_magic; |
| 136 | |
| 137 | unsigned long long_magic; |
| 138 | unsigned char postlong_char_magic; |
| 139 | |
| 140 | unsigned short short_magic; |
| 141 | unsigned char postshort_char_magic; |
| 142 | }; |
| 143 | |
| 144 | struct trie_header { |
| 145 | struct trie_magic magic; |
| 146 | off_t root, indexroot; |
| 147 | int count; |
| 148 | size_t maxpathlen; |
| 149 | int pathsep; |
| 150 | }; |
| 151 | |
| 152 | /* Union only used for computing alignment */ |
| 153 | union trie_node { |
| 154 | struct trie_leaf leaf; |
| 155 | struct { /* fake trie_switch with indeterminate array length filled in */ |
| 156 | struct trie_common c; |
| 157 | int len; |
| 158 | struct trie_switchentry sw[1]; |
| 159 | } sw; |
| 160 | struct { /* fake trie_string with indeterminate array length filled in */ |
| 161 | struct trie_common c; |
| 162 | int stringlen; |
| 163 | off_t subnode; |
| 164 | char string[1]; |
| 165 | } str; |
| 166 | }; |
| 167 | #define TRIE_ALIGN alignof(union trie_node) |
| 168 | |
| 169 | static void setup_magic(struct trie_magic *th) |
| 170 | { |
| 171 | /* |
| 172 | * Magic values are chosen so that every byte value used is |
| 173 | * distinct (so that we can't fail to spot endianness issues), and |
| 174 | * we cast 64 bits of data into each integer type just in case |
| 175 | * the platform makes it longer than we expect it to be. |
| 176 | */ |
| 177 | |
| 178 | memset(th, 0xCDU, sizeof(*th)); |
| 179 | |
| 180 | th->magic_len = sizeof(*th); |
| 181 | |
| 182 | memcpy(th->ident, magic_ident_string, 16); |
| 183 | |
| 184 | th->longlong_magic = 0x5583F34D5D84F73CULL; |
| 185 | th->postlonglong_char_magic = 0xDDU; |
| 186 | |
| 187 | th->offset_magic = (off_t)0xB39BF9AD56D48E0BULL; |
| 188 | th->postoffset_char_magic = 0x95U; |
| 189 | |
| 190 | th->size_magic = (size_t)0x6EC752B0EEAEBAC1ULL; |
| 191 | th->postsize_char_magic = 0x77U; |
| 192 | |
| 193 | th->null_pointer = NULL; |
| 194 | th->postptr_char_magic = 0x71U; |
| 195 | |
| 196 | th->long_magic = (unsigned long)0xA81A5E1F44334716ULL; |
| 197 | th->postlong_char_magic = 0x99U; |
| 198 | |
| 199 | th->short_magic = (unsigned short)0x0C8BD7984B68D9FCULL; |
| 200 | th->postshort_char_magic = 0x35U; |
| 201 | } |
| 202 | |
| 203 | /* ---------------------------------------------------------------------- |
| 204 | * Trie-building functions. |
| 205 | */ |
| 206 | |
| 207 | struct tbswitch { |
| 208 | int len; |
| 209 | char c[256]; |
| 210 | off_t off[256]; |
| 211 | int count[256]; |
| 212 | }; |
| 213 | |
| 214 | struct triebuild { |
| 215 | int fd; |
| 216 | off_t offset; |
| 217 | char *lastpath; |
| 218 | int lastlen, lastsize; |
| 219 | off_t lastoff; |
| 220 | struct tbswitch *switches; |
| 221 | int switchsize; |
| 222 | size_t maxpathlen; |
| 223 | }; |
| 224 | |
| 225 | static void tb_seek(triebuild *tb, off_t off) |
| 226 | { |
| 227 | tb->offset = off; |
| 228 | if (lseek(tb->fd, off, SEEK_SET) < 0) { |
| 229 | fprintf(stderr, PNAME ": lseek: %s\n", strerror(errno)); |
| 230 | exit(1); |
| 231 | } |
| 232 | } |
| 233 | |
| 234 | static void tb_write(triebuild *tb, const void *buf, size_t len) |
| 235 | { |
| 236 | tb->offset += len; |
| 237 | while (len > 0) { |
| 238 | int ret = write(tb->fd, buf, len); |
| 239 | if (ret < 0) { |
| 240 | fprintf(stderr, PNAME ": write: %s\n", strerror(errno)); |
| 241 | exit(1); |
| 242 | } |
| 243 | len -= ret; |
| 244 | buf = (const void *)((const char *)buf + ret); |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | static char trie_align_zeroes[TRIE_ALIGN]; |
| 249 | |
| 250 | static void tb_align(triebuild *tb) |
| 251 | { |
| 252 | int off = (TRIE_ALIGN - ((tb)->offset % TRIE_ALIGN)) % TRIE_ALIGN; |
| 253 | tb_write(tb, trie_align_zeroes, off); |
| 254 | } |
| 255 | |
| 256 | triebuild *triebuild_new(int fd) |
| 257 | { |
| 258 | triebuild *tb = snew(triebuild); |
| 259 | struct trie_header th; |
| 260 | |
| 261 | tb->fd = fd; |
| 262 | tb->lastpath = NULL; |
| 263 | tb->lastlen = tb->lastsize = 0; |
| 264 | tb->lastoff = 0; |
| 265 | tb->switches = NULL; |
| 266 | tb->switchsize = 0; |
| 267 | tb->maxpathlen = 0; |
| 268 | |
| 269 | setup_magic(&th.magic); |
| 270 | th.root = th.count = 0; |
| 271 | th.indexroot = 0; |
| 272 | th.maxpathlen = 0; |
| 273 | th.pathsep = (unsigned char)pathsep; |
| 274 | |
| 275 | tb_seek(tb, 0); |
| 276 | tb_write(tb, &th, sizeof(th)); |
| 277 | |
| 278 | return tb; |
| 279 | } |
| 280 | |
| 281 | static off_t triebuild_unwind(triebuild *tb, int targetdepth, int *outcount) |
| 282 | { |
| 283 | off_t offset; |
| 284 | int count, depth; |
| 285 | |
| 286 | if (tb->lastoff == 0) { |
| 287 | *outcount = 0; |
| 288 | return 0; |
| 289 | } |
| 290 | |
| 291 | offset = tb->lastoff; |
| 292 | count = 1; |
| 293 | depth = tb->lastlen + 1; |
| 294 | |
| 295 | assert(depth >= targetdepth); |
| 296 | |
| 297 | while (depth > targetdepth) { |
| 298 | int odepth = depth; |
| 299 | while (depth > targetdepth && |
| 300 | (depth-1 >= tb->switchsize || !tb->switches || |
| 301 | tb->switches[depth-1].len == 0)) |
| 302 | depth--; |
| 303 | if (odepth > depth) { |
| 304 | /* |
| 305 | * Write out a string node. |
| 306 | */ |
| 307 | size_t nodesize = sizeof(struct trie_string) + odepth - depth; |
| 308 | struct trie_string *st = (struct trie_string *)smalloc(nodesize); |
| 309 | st->c.type = TRIE_STRING; |
| 310 | st->stringlen = odepth - depth; |
| 311 | st->subnode = offset; |
| 312 | memcpy(st->string, tb->lastpath + depth, odepth - depth); |
| 313 | tb_align(tb); |
| 314 | offset = tb->offset; |
| 315 | tb_write(tb, st, nodesize); |
| 316 | sfree(st); |
| 317 | } |
| 318 | |
| 319 | assert(depth >= targetdepth); |
| 320 | if (depth <= targetdepth) |
| 321 | break; |
| 322 | |
| 323 | /* |
| 324 | * Now we expect to be sitting just below a switch node. |
| 325 | * Add our final entry to it and write it out. |
| 326 | */ |
| 327 | depth--; |
| 328 | { |
| 329 | struct trie_switch *sw; |
| 330 | char *chars; |
| 331 | size_t nodesize; |
| 332 | int swlen = tb->switches[depth].len; |
| 333 | int i; |
| 334 | |
| 335 | assert(swlen > 0); |
| 336 | |
| 337 | tb->switches[depth].c[swlen] = tb->lastpath[depth]; |
| 338 | tb->switches[depth].off[swlen] = offset; |
| 339 | tb->switches[depth].count[swlen] = count; |
| 340 | swlen++; |
| 341 | |
| 342 | nodesize = sizeof(struct trie_switch) + |
| 343 | swlen * sizeof(struct trie_switchentry) + swlen; |
| 344 | sw = (struct trie_switch *)smalloc(nodesize); |
| 345 | chars = (char *)&sw->sw[swlen]; |
| 346 | |
| 347 | sw->c.type = TRIE_SWITCH; |
| 348 | sw->len = swlen; |
| 349 | count = 0; |
| 350 | for (i = 0; i < swlen; i++) { |
| 351 | sw->sw[i].subnode = tb->switches[depth].off[i]; |
| 352 | sw->sw[i].subcount = tb->switches[depth].count[i]; |
| 353 | chars[i] = tb->switches[depth].c[i]; |
| 354 | |
| 355 | count += tb->switches[depth].count[i]; |
| 356 | } |
| 357 | |
| 358 | tb_align(tb); |
| 359 | offset = tb->offset; |
| 360 | tb_write(tb, sw, nodesize); |
| 361 | sfree(sw); |
| 362 | |
| 363 | tb->switches[depth].len = 0; /* clear this node */ |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | *outcount = count; |
| 368 | return offset; |
| 369 | } |
| 370 | |
| 371 | void triebuild_add(triebuild *tb, const char *pathname, |
| 372 | const struct trie_file *file) |
| 373 | { |
| 374 | int pathlen = strlen(pathname); |
| 375 | int depth; |
| 376 | |
| 377 | if (tb->maxpathlen < pathlen+1) |
| 378 | tb->maxpathlen = pathlen+1; |
| 379 | |
| 380 | if (tb->lastpath) { |
| 381 | off_t offset; |
| 382 | int count; |
| 383 | |
| 384 | /* |
| 385 | * Find the first differing character between this pathname |
| 386 | * and the previous one. |
| 387 | */ |
| 388 | int ret = triecmp(tb->lastpath, pathname, &depth); |
| 389 | assert(ret < 0); |
| 390 | |
| 391 | /* |
| 392 | * Finalise all nodes above this depth. |
| 393 | */ |
| 394 | offset = triebuild_unwind(tb, depth+1, &count); |
| 395 | |
| 396 | /* |
| 397 | * Add the final node we just acquired to the switch node |
| 398 | * at our chosen depth, creating it if it isn't already |
| 399 | * there. |
| 400 | */ |
| 401 | if (tb->switchsize <= depth) { |
| 402 | int oldsize = tb->switchsize; |
| 403 | tb->switchsize = depth * 3 / 2 + 64; |
| 404 | tb->switches = sresize(tb->switches, tb->switchsize, |
| 405 | struct tbswitch); |
| 406 | while (oldsize < tb->switchsize) |
| 407 | tb->switches[oldsize++].len = 0; |
| 408 | } |
| 409 | |
| 410 | tb->switches[depth].c[tb->switches[depth].len] = tb->lastpath[depth]; |
| 411 | tb->switches[depth].off[tb->switches[depth].len] = offset; |
| 412 | tb->switches[depth].count[tb->switches[depth].len] = count; |
| 413 | tb->switches[depth].len++; |
| 414 | } |
| 415 | |
| 416 | /* |
| 417 | * Write out a leaf node for the new file, and remember its |
| 418 | * file offset. |
| 419 | */ |
| 420 | { |
| 421 | struct trie_leaf leaf; |
| 422 | |
| 423 | leaf.c.type = TRIE_LEAF; |
| 424 | leaf.file = *file; /* structure copy */ |
| 425 | |
| 426 | tb_align(tb); |
| 427 | tb->lastoff = tb->offset; |
| 428 | tb_write(tb, &leaf, sizeof(leaf)); |
| 429 | } |
| 430 | |
| 431 | /* |
| 432 | * Store this pathname for comparison with the next one. |
| 433 | */ |
| 434 | if (tb->lastsize < pathlen+1) { |
| 435 | tb->lastsize = pathlen * 3 / 2 + 64; |
| 436 | tb->lastpath = sresize(tb->lastpath, tb->lastsize, char); |
| 437 | } |
| 438 | strcpy(tb->lastpath, pathname); |
| 439 | tb->lastlen = pathlen; |
| 440 | } |
| 441 | |
| 442 | int triebuild_finish(triebuild *tb) |
| 443 | { |
| 444 | struct trie_header th; |
| 445 | |
| 446 | setup_magic(&th.magic); |
| 447 | th.root = triebuild_unwind(tb, 0, &th.count); |
| 448 | th.indexroot = 0; |
| 449 | th.maxpathlen = tb->maxpathlen; |
| 450 | th.pathsep = (unsigned char)pathsep; |
| 451 | |
| 452 | tb_seek(tb, 0); |
| 453 | tb_write(tb, &th, sizeof(th)); |
| 454 | |
| 455 | return th.count; |
| 456 | } |
| 457 | |
| 458 | void triebuild_free(triebuild *tb) |
| 459 | { |
| 460 | sfree(tb->switches); |
| 461 | sfree(tb->lastpath); |
| 462 | sfree(tb); |
| 463 | } |
| 464 | |
| 465 | /* ---------------------------------------------------------------------- |
| 466 | * Memory-mapped trie modification. |
| 467 | */ |
| 468 | |
| 469 | #define MNODE(t, off, type) \ |
| 470 | ((struct type *)((char *)(t) + (off))) |
| 471 | |
| 472 | static unsigned long long fake_atime_recurse(void *t, struct trie_common *node, |
| 473 | int last_seen_pathsep) |
| 474 | { |
| 475 | while (node->type == TRIE_STRING) { |
| 476 | struct trie_string *st = (struct trie_string *)node; |
| 477 | last_seen_pathsep = (st->string[st->stringlen-1] == pathsep); |
| 478 | node = MNODE(t, st->subnode, trie_common); |
| 479 | } |
| 480 | |
| 481 | if (node->type == TRIE_LEAF) { |
| 482 | struct trie_leaf *leaf = (struct trie_leaf *)node; |
| 483 | return leaf->file.atime; |
| 484 | } else if (assert(node->type == TRIE_SWITCH), 1) { |
| 485 | struct trie_switch *sw = (struct trie_switch *)node; |
| 486 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 487 | unsigned long long max = 0, subdir, ret; |
| 488 | int i; |
| 489 | int slashindex = -1, bareindex = -1; |
| 490 | |
| 491 | /* |
| 492 | * First, process all the children of this node whose |
| 493 | * switch characters are not \0 or pathsep. We do this in |
| 494 | * reverse order so as to maintain best cache locality |
| 495 | * (tracking generally backwards through the file), though |
| 496 | * it doesn't matter semantically. |
| 497 | * |
| 498 | * For each of these children, we're just recursing into |
| 499 | * it to do any fixups required below it, and amalgamating |
| 500 | * the max atimes we get back. |
| 501 | */ |
| 502 | for (i = sw->len; i-- > 0 ;) { |
| 503 | if (chars[i] == '\0') { |
| 504 | bareindex = i; |
| 505 | } else if (chars[i] == pathsep) { |
| 506 | slashindex = i; |
| 507 | } else { |
| 508 | ret = fake_atime_recurse(t, MNODE(t, sw->sw[i].subnode, |
| 509 | trie_common), 0); |
| 510 | if (max < ret) |
| 511 | max = ret; |
| 512 | } |
| 513 | } |
| 514 | |
| 515 | /* |
| 516 | * Now we have at most two child nodes left to deal with: |
| 517 | * one with a slash (or general pathsep) and one with \0. |
| 518 | * |
| 519 | * If there's a slash node and a bare node, then the slash |
| 520 | * node contains precisely everything inside the directory |
| 521 | * described by the bare node; so we must retrieve the max |
| 522 | * atime for the slash node and use it to fix up the bare |
| 523 | * node. |
| 524 | * |
| 525 | * If there's only a bare node but the pathname leading up |
| 526 | * to this point ends in a slash, then _all_ of the child |
| 527 | * nodes of this node contain stuff inside the directory |
| 528 | * described by the bare node; so we use the whole of the |
| 529 | * maximum value we've computed so far to update the bare |
| 530 | * node. |
| 531 | */ |
| 532 | if (slashindex >= 0) { |
| 533 | ret = fake_atime_recurse(t, MNODE(t, sw->sw[slashindex].subnode, |
| 534 | trie_common), 1); |
| 535 | if (max < ret) |
| 536 | max = ret; |
| 537 | |
| 538 | subdir = ret; |
| 539 | } else if (last_seen_pathsep) { |
| 540 | subdir = max; |
| 541 | } else { |
| 542 | /* Don't update the bare subnode at all. */ |
| 543 | subdir = 0; |
| 544 | } |
| 545 | |
| 546 | if (bareindex >= 0) { |
| 547 | struct trie_leaf *leaf; |
| 548 | |
| 549 | leaf = MNODE(t, sw->sw[bareindex].subnode, trie_leaf); |
| 550 | |
| 551 | if (leaf && leaf->c.type == TRIE_LEAF) { |
| 552 | if (leaf->file.atime < subdir) |
| 553 | leaf->file.atime = subdir; |
| 554 | ret = leaf->file.atime; |
| 555 | } else { |
| 556 | /* Shouldn't really happen, but be cautious anyway */ |
| 557 | ret = fake_atime_recurse(t, &leaf->c, 0); |
| 558 | } |
| 559 | |
| 560 | if (max < ret) |
| 561 | max = ret; |
| 562 | } |
| 563 | |
| 564 | return max; |
| 565 | } |
| 566 | return 0; /* placate lint */ |
| 567 | } |
| 568 | |
| 569 | void trie_fake_dir_atimes(void *t) |
| 570 | { |
| 571 | struct trie_header *hdr = MNODE(t, 0, trie_header); |
| 572 | struct trie_common *node = MNODE(t, hdr->root, trie_common); |
| 573 | |
| 574 | fake_atime_recurse(t, node, 1); |
| 575 | } |
| 576 | |
| 577 | /* ---------------------------------------------------------------------- |
| 578 | * Querying functions. |
| 579 | */ |
| 580 | |
| 581 | #define NODE(t, off, type) \ |
| 582 | ((const struct type *)((const char *)(t) + (off))) |
| 583 | |
| 584 | int trie_check_magic(const void *t) |
| 585 | { |
| 586 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 587 | struct trie_magic magic; |
| 588 | |
| 589 | setup_magic(&magic); |
| 590 | return !memcmp(&magic, &hdr->magic, sizeof(magic)); |
| 591 | } |
| 592 | |
| 593 | size_t trie_maxpathlen(const void *t) |
| 594 | { |
| 595 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 596 | return hdr->maxpathlen; |
| 597 | } |
| 598 | |
| 599 | unsigned long trie_before(const void *t, const char *pathname) |
| 600 | { |
| 601 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 602 | int ret = 0, lastcount = hdr->count; |
| 603 | int len = 1 + strlen(pathname), depth = 0; |
| 604 | off_t off = hdr->root; |
| 605 | |
| 606 | while (1) { |
| 607 | const struct trie_common *node = NODE(t, off, trie_common); |
| 608 | if (node->type == TRIE_LEAF) { |
| 609 | if (depth < len) |
| 610 | ret += lastcount; /* _shouldn't_ happen, but in principle */ |
| 611 | return ret; |
| 612 | } else if (node->type == TRIE_STRING) { |
| 613 | const struct trie_string *st = NODE(t, off, trie_string); |
| 614 | |
| 615 | int offset; |
| 616 | int cmp = triencmp(st->string, st->stringlen, |
| 617 | pathname + depth, len-depth, &offset); |
| 618 | |
| 619 | if (offset < st->stringlen) { |
| 620 | if (cmp < 0) |
| 621 | ret += lastcount; |
| 622 | return ret; |
| 623 | } |
| 624 | |
| 625 | depth += st->stringlen; |
| 626 | off = st->subnode; |
| 627 | } else if (node->type == TRIE_SWITCH) { |
| 628 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
| 629 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 630 | int i; |
| 631 | |
| 632 | for (i = 0; i < sw->len; i++) { |
| 633 | int c = chars[i]; |
| 634 | int cmp = trieccmp(pathname[depth], c); |
| 635 | if (cmp > 0) |
| 636 | ret += sw->sw[i].subcount; |
| 637 | else if (cmp < 0) |
| 638 | return ret; |
| 639 | else { |
| 640 | off = sw->sw[i].subnode; |
| 641 | lastcount = sw->sw[i].subcount; |
| 642 | depth++; |
| 643 | break; |
| 644 | } |
| 645 | } |
| 646 | if (i == sw->len) |
| 647 | return ret; |
| 648 | } |
| 649 | } |
| 650 | } |
| 651 | |
| 652 | void trie_getpath(const void *t, unsigned long n, char *buf) |
| 653 | { |
| 654 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 655 | int depth = 0; |
| 656 | off_t off = hdr->root; |
| 657 | |
| 658 | while (1) { |
| 659 | const struct trie_common *node = NODE(t, off, trie_common); |
| 660 | if (node->type == TRIE_LEAF) { |
| 661 | assert(depth > 0 && buf[depth-1] == '\0'); |
| 662 | return; |
| 663 | } else if (node->type == TRIE_STRING) { |
| 664 | const struct trie_string *st = NODE(t, off, trie_string); |
| 665 | |
| 666 | memcpy(buf + depth, st->string, st->stringlen); |
| 667 | depth += st->stringlen; |
| 668 | off = st->subnode; |
| 669 | } else if (node->type == TRIE_SWITCH) { |
| 670 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
| 671 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 672 | int i; |
| 673 | |
| 674 | for (i = 0; i < sw->len; i++) { |
| 675 | if (n < sw->sw[i].subcount) { |
| 676 | buf[depth++] = chars[i]; |
| 677 | off = sw->sw[i].subnode; |
| 678 | break; |
| 679 | } else |
| 680 | n -= sw->sw[i].subcount; |
| 681 | } |
| 682 | assert(i < sw->len); |
| 683 | } |
| 684 | } |
| 685 | } |
| 686 | |
| 687 | const struct trie_file *trie_getfile(const void *t, unsigned long n) |
| 688 | { |
| 689 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 690 | off_t off = hdr->root; |
| 691 | |
| 692 | while (1) { |
| 693 | const struct trie_common *node = NODE(t, off, trie_common); |
| 694 | if (node->type == TRIE_LEAF) { |
| 695 | const struct trie_leaf *leaf = NODE(t, off, trie_leaf); |
| 696 | return &leaf->file; |
| 697 | } else if (node->type == TRIE_STRING) { |
| 698 | const struct trie_string *st = NODE(t, off, trie_string); |
| 699 | off = st->subnode; |
| 700 | } else if (node->type == TRIE_SWITCH) { |
| 701 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
| 702 | int i; |
| 703 | |
| 704 | for (i = 0; i < sw->len; i++) { |
| 705 | if (n < sw->sw[i].subcount) { |
| 706 | off = sw->sw[i].subnode; |
| 707 | break; |
| 708 | } else |
| 709 | n -= sw->sw[i].subcount; |
| 710 | } |
| 711 | assert(i < sw->len); |
| 712 | } |
| 713 | } |
| 714 | } |
| 715 | |
| 716 | unsigned long trie_count(const void *t) |
| 717 | { |
| 718 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 719 | return hdr->count; |
| 720 | } |
| 721 | |
| 722 | char trie_pathsep(const void *t) |
| 723 | { |
| 724 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 725 | return (char)hdr->pathsep; |
| 726 | } |
| 727 | |
| 728 | struct triewalk_switch { |
| 729 | const struct trie_switch *sw; |
| 730 | int pos, depth, count; |
| 731 | }; |
| 732 | struct triewalk { |
| 733 | const void *t; |
| 734 | struct triewalk_switch *switches; |
| 735 | int nswitches, switchsize; |
| 736 | int count; |
| 737 | }; |
| 738 | triewalk *triewalk_new(const void *vt) |
| 739 | { |
| 740 | triewalk *tw = snew(triewalk); |
| 741 | |
| 742 | tw->t = (const char *)vt; |
| 743 | tw->switches = NULL; |
| 744 | tw->switchsize = 0; |
| 745 | tw->nswitches = -1; |
| 746 | tw->count = 0; |
| 747 | |
| 748 | return tw; |
| 749 | } |
| 750 | |
| 751 | void triewalk_rebase(triewalk *tw, const void *t) |
| 752 | { |
| 753 | ptrdiff_t diff = ((const unsigned char *)t - (const unsigned char *)(tw->t)); |
| 754 | int i; |
| 755 | |
| 756 | tw->t = t; |
| 757 | |
| 758 | for (i = 0; i < tw->nswitches; i++) |
| 759 | tw->switches[i].sw = (const struct trie_switch *) |
| 760 | ((const unsigned char *)(tw->switches[i].sw) + diff); |
| 761 | } |
| 762 | |
| 763 | const struct trie_file *triewalk_next(triewalk *tw, char *buf) |
| 764 | { |
| 765 | off_t off; |
| 766 | int depth; |
| 767 | |
| 768 | if (tw->nswitches < 0) { |
| 769 | const struct trie_header *hdr = NODE(tw->t, 0, trie_header); |
| 770 | off = hdr->root; |
| 771 | depth = 0; |
| 772 | tw->nswitches = 0; |
| 773 | } else { |
| 774 | while (1) { |
| 775 | int swpos; |
| 776 | const struct trie_switch *sw; |
| 777 | const char *chars; |
| 778 | |
| 779 | if (tw->nswitches == 0) { |
| 780 | assert(tw->count == NODE(tw->t, 0, trie_header)->count); |
| 781 | return NULL; /* run out of trie */ |
| 782 | } |
| 783 | |
| 784 | swpos = tw->switches[tw->nswitches-1].pos; |
| 785 | sw = tw->switches[tw->nswitches-1].sw; |
| 786 | chars = (const char *)&sw->sw[sw->len]; |
| 787 | |
| 788 | if (swpos < sw->len) { |
| 789 | depth = tw->switches[tw->nswitches-1].depth; |
| 790 | off = sw->sw[swpos].subnode; |
| 791 | if (buf) |
| 792 | buf[depth++] = chars[swpos]; |
| 793 | assert(tw->count == tw->switches[tw->nswitches-1].count); |
| 794 | tw->switches[tw->nswitches-1].count += sw->sw[swpos].subcount; |
| 795 | tw->switches[tw->nswitches-1].pos++; |
| 796 | break; |
| 797 | } |
| 798 | |
| 799 | tw->nswitches--; |
| 800 | } |
| 801 | } |
| 802 | |
| 803 | while (1) { |
| 804 | const struct trie_common *node = NODE(tw->t, off, trie_common); |
| 805 | if (node->type == TRIE_LEAF) { |
| 806 | const struct trie_leaf *lf = NODE(tw->t, off, trie_leaf); |
| 807 | if (buf) |
| 808 | assert(depth > 0 && buf[depth-1] == '\0'); |
| 809 | tw->count++; |
| 810 | return &lf->file; |
| 811 | } else if (node->type == TRIE_STRING) { |
| 812 | const struct trie_string *st = NODE(tw->t, off, trie_string); |
| 813 | |
| 814 | if (buf) |
| 815 | memcpy(buf + depth, st->string, st->stringlen); |
| 816 | depth += st->stringlen; |
| 817 | off = st->subnode; |
| 818 | } else if (node->type == TRIE_SWITCH) { |
| 819 | const struct trie_switch *sw = NODE(tw->t, off, trie_switch); |
| 820 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 821 | |
| 822 | if (tw->nswitches >= tw->switchsize) { |
| 823 | tw->switchsize = tw->nswitches * 3 / 2 + 32; |
| 824 | tw->switches = sresize(tw->switches, tw->switchsize, |
| 825 | struct triewalk_switch); |
| 826 | } |
| 827 | |
| 828 | tw->switches[tw->nswitches].sw = sw; |
| 829 | tw->switches[tw->nswitches].pos = 1; |
| 830 | tw->switches[tw->nswitches].depth = depth; |
| 831 | tw->switches[tw->nswitches].count = tw->count + sw->sw[0].subcount; |
| 832 | off = sw->sw[0].subnode; |
| 833 | if (buf) |
| 834 | buf[depth++] = chars[0]; |
| 835 | tw->nswitches++; |
| 836 | } |
| 837 | } |
| 838 | } |
| 839 | void triewalk_free(triewalk *tw) |
| 840 | { |
| 841 | sfree(tw->switches); |
| 842 | sfree(tw); |
| 843 | } |
| 844 | |
| 845 | void trie_set_index_offset(void *t, off_t ptr) |
| 846 | { |
| 847 | ((struct trie_header *)t)->indexroot = ptr; |
| 848 | } |
| 849 | off_t trie_get_index_offset(const void *t) |
| 850 | { |
| 851 | return ((const struct trie_header *)t)->indexroot; |
| 852 | } |
| 853 | |
| 854 | void make_successor(char *pathbuf) |
| 855 | { |
| 856 | int len = strlen(pathbuf); |
| 857 | if (len > 0 && pathbuf[len-1] == pathsep) |
| 858 | len--; |
| 859 | pathbuf[len] = '\001'; |
| 860 | pathbuf[len+1] = '\0'; |
| 861 | } |