Colin Cross | 64ceac3 | 2010-01-13 21:19:52 -0800 | [diff] [blame] | 1 | /* $OpenBSD: fts.c,v 1.43 2009/08/27 16:19:27 millert Exp $ */ |
| 2 | |
| 3 | /*- |
| 4 | * Copyright (c) 1990, 1993, 1994 |
| 5 | * The Regents of the University of California. All rights reserved. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * 1. Redistributions of source code must retain the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer in the |
| 14 | * documentation and/or other materials provided with the distribution. |
| 15 | * 3. Neither the name of the University nor the names of its contributors |
| 16 | * may be used to endorse or promote products derived from this software |
| 17 | * without specific prior written permission. |
| 18 | * |
| 19 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 20 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 21 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 22 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 23 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 24 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 25 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 26 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 27 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 28 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 29 | * SUCH DAMAGE. |
| 30 | */ |
| 31 | |
| 32 | #include <sys/param.h> |
| 33 | #include <sys/stat.h> |
| 34 | |
| 35 | #include <dirent.h> |
| 36 | #include <errno.h> |
| 37 | #include <fcntl.h> |
| 38 | #include <fts.h> |
| 39 | #include <stdlib.h> |
| 40 | #include <string.h> |
| 41 | #include <unistd.h> |
| 42 | |
| 43 | #define MAX(a,b) ((a)>(b)?(a):(b)) |
| 44 | |
| 45 | static FTSENT *fts_alloc(FTS *, char *, size_t); |
| 46 | static FTSENT *fts_build(FTS *, int); |
| 47 | static void fts_lfree(FTSENT *); |
| 48 | static void fts_load(FTS *, FTSENT *); |
| 49 | static size_t fts_maxarglen(char * const *); |
| 50 | static void fts_padjust(FTS *, FTSENT *); |
| 51 | static int fts_palloc(FTS *, size_t); |
| 52 | static FTSENT *fts_sort(FTS *, FTSENT *, int); |
| 53 | static u_short fts_stat(FTS *, FTSENT *, int); |
| 54 | static int fts_safe_changedir(FTS *, FTSENT *, int, char *); |
| 55 | |
| 56 | #define ISDOT(a) (a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2]))) |
| 57 | |
| 58 | #define CLR(opt) (sp->fts_options &= ~(opt)) |
| 59 | #define ISSET(opt) (sp->fts_options & (opt)) |
| 60 | #define SET(opt) (sp->fts_options |= (opt)) |
| 61 | |
| 62 | #define FCHDIR(sp, fd) (!ISSET(FTS_NOCHDIR) && fchdir(fd)) |
| 63 | |
| 64 | /* fts_build flags */ |
| 65 | #define BCHILD 1 /* fts_children */ |
| 66 | #define BNAMES 2 /* fts_children, names only */ |
| 67 | #define BREAD 3 /* fts_read */ |
| 68 | |
| 69 | FTS * |
| 70 | fts_open(char * const *argv, int options, |
| 71 | int (*compar)(const FTSENT **, const FTSENT **)) |
| 72 | { |
| 73 | FTS *sp; |
| 74 | FTSENT *p, *root; |
| 75 | int nitems; |
| 76 | FTSENT *parent, *tmp; |
| 77 | size_t len; |
| 78 | |
| 79 | /* Options check. */ |
| 80 | if (options & ~FTS_OPTIONMASK) { |
| 81 | errno = EINVAL; |
| 82 | return (NULL); |
| 83 | } |
| 84 | |
| 85 | /* Allocate/initialize the stream */ |
| 86 | if ((sp = calloc(1, sizeof(FTS))) == NULL) |
| 87 | return (NULL); |
| 88 | sp->fts_compar = compar; |
| 89 | sp->fts_options = options; |
| 90 | |
| 91 | /* Logical walks turn on NOCHDIR; symbolic links are too hard. */ |
| 92 | if (ISSET(FTS_LOGICAL)) |
| 93 | SET(FTS_NOCHDIR); |
| 94 | |
| 95 | /* |
| 96 | * Start out with 1K of path space, and enough, in any case, |
| 97 | * to hold the user's paths. |
| 98 | */ |
| 99 | if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN))) |
| 100 | goto mem1; |
| 101 | |
| 102 | /* Allocate/initialize root's parent. */ |
| 103 | if ((parent = fts_alloc(sp, "", 0)) == NULL) |
| 104 | goto mem2; |
| 105 | parent->fts_level = FTS_ROOTPARENTLEVEL; |
| 106 | |
| 107 | /* Allocate/initialize root(s). */ |
| 108 | for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) { |
| 109 | /* Don't allow zero-length paths. */ |
| 110 | if ((len = strlen(*argv)) == 0) { |
| 111 | errno = ENOENT; |
| 112 | goto mem3; |
| 113 | } |
| 114 | |
| 115 | if ((p = fts_alloc(sp, *argv, len)) == NULL) |
| 116 | goto mem3; |
| 117 | p->fts_level = FTS_ROOTLEVEL; |
| 118 | p->fts_parent = parent; |
| 119 | p->fts_accpath = p->fts_name; |
| 120 | p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW)); |
| 121 | |
| 122 | /* Command-line "." and ".." are real directories. */ |
| 123 | if (p->fts_info == FTS_DOT) |
| 124 | p->fts_info = FTS_D; |
| 125 | |
| 126 | /* |
| 127 | * If comparison routine supplied, traverse in sorted |
| 128 | * order; otherwise traverse in the order specified. |
| 129 | */ |
| 130 | if (compar) { |
| 131 | p->fts_link = root; |
| 132 | root = p; |
| 133 | } else { |
| 134 | p->fts_link = NULL; |
| 135 | if (root == NULL) |
| 136 | tmp = root = p; |
| 137 | else { |
| 138 | tmp->fts_link = p; |
| 139 | tmp = p; |
| 140 | } |
| 141 | } |
| 142 | } |
| 143 | if (compar && nitems > 1) |
| 144 | root = fts_sort(sp, root, nitems); |
| 145 | |
| 146 | /* |
| 147 | * Allocate a dummy pointer and make fts_read think that we've just |
| 148 | * finished the node before the root(s); set p->fts_info to FTS_INIT |
| 149 | * so that everything about the "current" node is ignored. |
| 150 | */ |
| 151 | if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL) |
| 152 | goto mem3; |
| 153 | sp->fts_cur->fts_link = root; |
| 154 | sp->fts_cur->fts_info = FTS_INIT; |
| 155 | |
| 156 | /* |
| 157 | * If using chdir(2), grab a file descriptor pointing to dot to ensure |
| 158 | * that we can get back here; this could be avoided for some paths, |
| 159 | * but almost certainly not worth the effort. Slashes, symbolic links, |
| 160 | * and ".." are all fairly nasty problems. Note, if we can't get the |
| 161 | * descriptor we run anyway, just more slowly. |
| 162 | */ |
| 163 | if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = open(".", O_RDONLY, 0)) < 0) |
| 164 | SET(FTS_NOCHDIR); |
| 165 | |
| 166 | if (nitems == 0) |
| 167 | free(parent); |
| 168 | |
| 169 | return (sp); |
| 170 | |
| 171 | mem3: fts_lfree(root); |
| 172 | free(parent); |
| 173 | mem2: free(sp->fts_path); |
| 174 | mem1: free(sp); |
| 175 | return (NULL); |
| 176 | } |
| 177 | |
| 178 | static void |
| 179 | fts_load(FTS *sp, FTSENT *p) |
| 180 | { |
| 181 | size_t len; |
| 182 | char *cp; |
| 183 | |
| 184 | /* |
| 185 | * Load the stream structure for the next traversal. Since we don't |
| 186 | * actually enter the directory until after the preorder visit, set |
| 187 | * the fts_accpath field specially so the chdir gets done to the right |
| 188 | * place and the user can access the first node. From fts_open it's |
| 189 | * known that the path will fit. |
| 190 | */ |
| 191 | len = p->fts_pathlen = p->fts_namelen; |
| 192 | memmove(sp->fts_path, p->fts_name, len + 1); |
| 193 | if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) { |
| 194 | len = strlen(++cp); |
| 195 | memmove(p->fts_name, cp, len + 1); |
| 196 | p->fts_namelen = len; |
| 197 | } |
| 198 | p->fts_accpath = p->fts_path = sp->fts_path; |
| 199 | sp->fts_dev = p->fts_dev; |
| 200 | } |
| 201 | |
| 202 | int |
| 203 | fts_close(FTS *sp) |
| 204 | { |
| 205 | FTSENT *freep, *p; |
| 206 | int rfd, error = 0; |
| 207 | |
| 208 | /* |
| 209 | * This still works if we haven't read anything -- the dummy structure |
| 210 | * points to the root list, so we step through to the end of the root |
| 211 | * list which has a valid parent pointer. |
| 212 | */ |
| 213 | if (sp->fts_cur) { |
| 214 | for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) { |
| 215 | freep = p; |
| 216 | p = p->fts_link ? p->fts_link : p->fts_parent; |
| 217 | free(freep); |
| 218 | } |
| 219 | free(p); |
| 220 | } |
| 221 | |
| 222 | /* Stash the original directory fd if needed. */ |
| 223 | rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd; |
| 224 | |
| 225 | /* Free up child linked list, sort array, path buffer, stream ptr.*/ |
| 226 | if (sp->fts_child) |
| 227 | fts_lfree(sp->fts_child); |
| 228 | if (sp->fts_array) |
| 229 | free(sp->fts_array); |
| 230 | free(sp->fts_path); |
| 231 | free(sp); |
| 232 | |
| 233 | /* Return to original directory, checking for error. */ |
| 234 | if (rfd != -1) { |
| 235 | int saved_errno; |
| 236 | error = fchdir(rfd); |
| 237 | saved_errno = errno; |
| 238 | (void)close(rfd); |
| 239 | errno = saved_errno; |
| 240 | } |
| 241 | |
| 242 | return (error); |
| 243 | } |
| 244 | |
| 245 | /* |
| 246 | * Special case of "/" at the end of the path so that slashes aren't |
| 247 | * appended which would cause paths to be written as "....//foo". |
| 248 | */ |
| 249 | #define NAPPEND(p) \ |
| 250 | (p->fts_path[p->fts_pathlen - 1] == '/' \ |
| 251 | ? p->fts_pathlen - 1 : p->fts_pathlen) |
| 252 | |
| 253 | FTSENT * |
| 254 | fts_read(FTS *sp) |
| 255 | { |
| 256 | FTSENT *p, *tmp; |
| 257 | int instr; |
| 258 | char *t; |
| 259 | int saved_errno; |
| 260 | |
| 261 | /* If finished or unrecoverable error, return NULL. */ |
| 262 | if (sp->fts_cur == NULL || ISSET(FTS_STOP)) |
| 263 | return (NULL); |
| 264 | |
| 265 | /* Set current node pointer. */ |
| 266 | p = sp->fts_cur; |
| 267 | |
| 268 | /* Save and zero out user instructions. */ |
| 269 | instr = p->fts_instr; |
| 270 | p->fts_instr = FTS_NOINSTR; |
| 271 | |
| 272 | /* Any type of file may be re-visited; re-stat and re-turn. */ |
| 273 | if (instr == FTS_AGAIN) { |
| 274 | p->fts_info = fts_stat(sp, p, 0); |
| 275 | return (p); |
| 276 | } |
| 277 | |
| 278 | /* |
| 279 | * Following a symlink -- SLNONE test allows application to see |
| 280 | * SLNONE and recover. If indirecting through a symlink, have |
| 281 | * keep a pointer to current location. If unable to get that |
| 282 | * pointer, follow fails. |
| 283 | */ |
| 284 | if (instr == FTS_FOLLOW && |
| 285 | (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) { |
| 286 | p->fts_info = fts_stat(sp, p, 1); |
| 287 | if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) { |
| 288 | if ((p->fts_symfd = open(".", O_RDONLY, 0)) < 0) { |
| 289 | p->fts_errno = errno; |
| 290 | p->fts_info = FTS_ERR; |
| 291 | } else |
| 292 | p->fts_flags |= FTS_SYMFOLLOW; |
| 293 | } |
| 294 | return (p); |
| 295 | } |
| 296 | |
| 297 | /* Directory in pre-order. */ |
| 298 | if (p->fts_info == FTS_D) { |
| 299 | /* If skipped or crossed mount point, do post-order visit. */ |
| 300 | if (instr == FTS_SKIP || |
| 301 | (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) { |
| 302 | if (p->fts_flags & FTS_SYMFOLLOW) |
| 303 | (void)close(p->fts_symfd); |
| 304 | if (sp->fts_child) { |
| 305 | fts_lfree(sp->fts_child); |
| 306 | sp->fts_child = NULL; |
| 307 | } |
| 308 | p->fts_info = FTS_DP; |
| 309 | return (p); |
| 310 | } |
| 311 | |
| 312 | /* Rebuild if only read the names and now traversing. */ |
| 313 | if (sp->fts_child && ISSET(FTS_NAMEONLY)) { |
| 314 | CLR(FTS_NAMEONLY); |
| 315 | fts_lfree(sp->fts_child); |
| 316 | sp->fts_child = NULL; |
| 317 | } |
| 318 | |
| 319 | /* |
| 320 | * Cd to the subdirectory. |
| 321 | * |
| 322 | * If have already read and now fail to chdir, whack the list |
| 323 | * to make the names come out right, and set the parent errno |
| 324 | * so the application will eventually get an error condition. |
| 325 | * Set the FTS_DONTCHDIR flag so that when we logically change |
| 326 | * directories back to the parent we don't do a chdir. |
| 327 | * |
| 328 | * If haven't read do so. If the read fails, fts_build sets |
| 329 | * FTS_STOP or the fts_info field of the node. |
| 330 | */ |
| 331 | if (sp->fts_child) { |
| 332 | if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) { |
| 333 | p->fts_errno = errno; |
| 334 | p->fts_flags |= FTS_DONTCHDIR; |
| 335 | for (p = sp->fts_child; p; p = p->fts_link) |
| 336 | p->fts_accpath = |
| 337 | p->fts_parent->fts_accpath; |
| 338 | } |
| 339 | } else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) { |
| 340 | if (ISSET(FTS_STOP)) |
| 341 | return (NULL); |
| 342 | return (p); |
| 343 | } |
| 344 | p = sp->fts_child; |
| 345 | sp->fts_child = NULL; |
| 346 | goto name; |
| 347 | } |
| 348 | |
| 349 | /* Move to the next node on this level. */ |
| 350 | next: tmp = p; |
| 351 | if ((p = p->fts_link)) { |
| 352 | free(tmp); |
| 353 | |
| 354 | /* |
| 355 | * If reached the top, return to the original directory (or |
| 356 | * the root of the tree), and load the paths for the next root. |
| 357 | */ |
| 358 | if (p->fts_level == FTS_ROOTLEVEL) { |
| 359 | if (FCHDIR(sp, sp->fts_rfd)) { |
| 360 | SET(FTS_STOP); |
| 361 | return (NULL); |
| 362 | } |
| 363 | fts_load(sp, p); |
| 364 | return (sp->fts_cur = p); |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * User may have called fts_set on the node. If skipped, |
| 369 | * ignore. If followed, get a file descriptor so we can |
| 370 | * get back if necessary. |
| 371 | */ |
| 372 | if (p->fts_instr == FTS_SKIP) |
| 373 | goto next; |
| 374 | if (p->fts_instr == FTS_FOLLOW) { |
| 375 | p->fts_info = fts_stat(sp, p, 1); |
| 376 | if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) { |
| 377 | if ((p->fts_symfd = |
| 378 | open(".", O_RDONLY, 0)) < 0) { |
| 379 | p->fts_errno = errno; |
| 380 | p->fts_info = FTS_ERR; |
| 381 | } else |
| 382 | p->fts_flags |= FTS_SYMFOLLOW; |
| 383 | } |
| 384 | p->fts_instr = FTS_NOINSTR; |
| 385 | } |
| 386 | |
| 387 | name: t = sp->fts_path + NAPPEND(p->fts_parent); |
| 388 | *t++ = '/'; |
| 389 | memmove(t, p->fts_name, p->fts_namelen + 1); |
| 390 | return (sp->fts_cur = p); |
| 391 | } |
| 392 | |
| 393 | /* Move up to the parent node. */ |
| 394 | p = tmp->fts_parent; |
| 395 | free(tmp); |
| 396 | |
| 397 | if (p->fts_level == FTS_ROOTPARENTLEVEL) { |
| 398 | /* |
| 399 | * Done; free everything up and set errno to 0 so the user |
| 400 | * can distinguish between error and EOF. |
| 401 | */ |
| 402 | free(p); |
| 403 | errno = 0; |
| 404 | return (sp->fts_cur = NULL); |
| 405 | } |
| 406 | |
| 407 | /* NUL terminate the pathname. */ |
| 408 | sp->fts_path[p->fts_pathlen] = '\0'; |
| 409 | |
| 410 | /* |
| 411 | * Return to the parent directory. If at a root node or came through |
| 412 | * a symlink, go back through the file descriptor. Otherwise, cd up |
| 413 | * one directory. |
| 414 | */ |
| 415 | if (p->fts_level == FTS_ROOTLEVEL) { |
| 416 | if (FCHDIR(sp, sp->fts_rfd)) { |
| 417 | SET(FTS_STOP); |
| 418 | sp->fts_cur = p; |
| 419 | return (NULL); |
| 420 | } |
| 421 | } else if (p->fts_flags & FTS_SYMFOLLOW) { |
| 422 | if (FCHDIR(sp, p->fts_symfd)) { |
| 423 | saved_errno = errno; |
| 424 | (void)close(p->fts_symfd); |
| 425 | errno = saved_errno; |
| 426 | SET(FTS_STOP); |
| 427 | sp->fts_cur = p; |
| 428 | return (NULL); |
| 429 | } |
| 430 | (void)close(p->fts_symfd); |
| 431 | } else if (!(p->fts_flags & FTS_DONTCHDIR) && |
| 432 | fts_safe_changedir(sp, p->fts_parent, -1, "..")) { |
| 433 | SET(FTS_STOP); |
| 434 | sp->fts_cur = p; |
| 435 | return (NULL); |
| 436 | } |
| 437 | p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP; |
| 438 | return (sp->fts_cur = p); |
| 439 | } |
| 440 | |
| 441 | /* |
| 442 | * Fts_set takes the stream as an argument although it's not used in this |
| 443 | * implementation; it would be necessary if anyone wanted to add global |
| 444 | * semantics to fts using fts_set. An error return is allowed for similar |
| 445 | * reasons. |
| 446 | */ |
| 447 | /* ARGSUSED */ |
| 448 | int |
| 449 | fts_set(FTS *sp, FTSENT *p, int instr) |
| 450 | { |
| 451 | if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW && |
| 452 | instr != FTS_NOINSTR && instr != FTS_SKIP) { |
| 453 | errno = EINVAL; |
| 454 | return (1); |
| 455 | } |
| 456 | p->fts_instr = instr; |
| 457 | return (0); |
| 458 | } |
| 459 | |
| 460 | FTSENT * |
| 461 | fts_children(FTS *sp, int instr) |
| 462 | { |
| 463 | FTSENT *p; |
| 464 | int fd; |
| 465 | |
| 466 | if (instr && instr != FTS_NAMEONLY) { |
| 467 | errno = EINVAL; |
| 468 | return (NULL); |
| 469 | } |
| 470 | |
| 471 | /* Set current node pointer. */ |
| 472 | p = sp->fts_cur; |
| 473 | |
| 474 | /* |
| 475 | * Errno set to 0 so user can distinguish empty directory from |
| 476 | * an error. |
| 477 | */ |
| 478 | errno = 0; |
| 479 | |
| 480 | /* Fatal errors stop here. */ |
| 481 | if (ISSET(FTS_STOP)) |
| 482 | return (NULL); |
| 483 | |
| 484 | /* Return logical hierarchy of user's arguments. */ |
| 485 | if (p->fts_info == FTS_INIT) |
| 486 | return (p->fts_link); |
| 487 | |
| 488 | /* |
| 489 | * If not a directory being visited in pre-order, stop here. Could |
| 490 | * allow FTS_DNR, assuming the user has fixed the problem, but the |
| 491 | * same effect is available with FTS_AGAIN. |
| 492 | */ |
| 493 | if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */) |
| 494 | return (NULL); |
| 495 | |
| 496 | /* Free up any previous child list. */ |
| 497 | if (sp->fts_child) |
| 498 | fts_lfree(sp->fts_child); |
| 499 | |
| 500 | if (instr == FTS_NAMEONLY) { |
| 501 | SET(FTS_NAMEONLY); |
| 502 | instr = BNAMES; |
| 503 | } else |
| 504 | instr = BCHILD; |
| 505 | |
| 506 | /* |
| 507 | * If using chdir on a relative path and called BEFORE fts_read does |
| 508 | * its chdir to the root of a traversal, we can lose -- we need to |
| 509 | * chdir into the subdirectory, and we don't know where the current |
| 510 | * directory is, so we can't get back so that the upcoming chdir by |
| 511 | * fts_read will work. |
| 512 | */ |
| 513 | if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' || |
| 514 | ISSET(FTS_NOCHDIR)) |
| 515 | return (sp->fts_child = fts_build(sp, instr)); |
| 516 | |
| 517 | if ((fd = open(".", O_RDONLY, 0)) < 0) |
| 518 | return (NULL); |
| 519 | sp->fts_child = fts_build(sp, instr); |
| 520 | if (fchdir(fd)) { |
| 521 | (void)close(fd); |
| 522 | return (NULL); |
| 523 | } |
| 524 | (void)close(fd); |
| 525 | return (sp->fts_child); |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * This is the tricky part -- do not casually change *anything* in here. The |
| 530 | * idea is to build the linked list of entries that are used by fts_children |
| 531 | * and fts_read. There are lots of special cases. |
| 532 | * |
| 533 | * The real slowdown in walking the tree is the stat calls. If FTS_NOSTAT is |
| 534 | * set and it's a physical walk (so that symbolic links can't be directories), |
| 535 | * we can do things quickly. First, if it's a 4.4BSD file system, the type |
| 536 | * of the file is in the directory entry. Otherwise, we assume that the number |
| 537 | * of subdirectories in a node is equal to the number of links to the parent. |
| 538 | * The former skips all stat calls. The latter skips stat calls in any leaf |
| 539 | * directories and for any files after the subdirectories in the directory have |
| 540 | * been found, cutting the stat calls by about 2/3. |
| 541 | */ |
| 542 | static FTSENT * |
| 543 | fts_build(FTS *sp, int type) |
| 544 | { |
| 545 | struct dirent *dp; |
| 546 | FTSENT *p, *head; |
| 547 | FTSENT *cur, *tail; |
| 548 | DIR *dirp; |
| 549 | void *oldaddr; |
| 550 | size_t len, maxlen; |
David 'Digit' Turner | 50ace4f | 2010-06-16 16:36:41 -0700 | [diff] [blame] | 551 | int nitems, cderrno, descend, level, nlinks, nostat = 0, doadjust; |
Colin Cross | 64ceac3 | 2010-01-13 21:19:52 -0800 | [diff] [blame] | 552 | int saved_errno; |
David 'Digit' Turner | 50ace4f | 2010-06-16 16:36:41 -0700 | [diff] [blame] | 553 | char *cp = NULL; |
Colin Cross | 64ceac3 | 2010-01-13 21:19:52 -0800 | [diff] [blame] | 554 | |
| 555 | /* Set current node pointer. */ |
| 556 | cur = sp->fts_cur; |
| 557 | |
| 558 | /* |
| 559 | * Open the directory for reading. If this fails, we're done. |
| 560 | * If being called from fts_read, set the fts_info field. |
| 561 | */ |
| 562 | if ((dirp = opendir(cur->fts_accpath)) == NULL) { |
| 563 | if (type == BREAD) { |
| 564 | cur->fts_info = FTS_DNR; |
| 565 | cur->fts_errno = errno; |
| 566 | } |
| 567 | return (NULL); |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * Nlinks is the number of possible entries of type directory in the |
| 572 | * directory if we're cheating on stat calls, 0 if we're not doing |
| 573 | * any stat calls at all, -1 if we're doing stats on everything. |
| 574 | */ |
| 575 | if (type == BNAMES) |
| 576 | nlinks = 0; |
| 577 | else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) { |
| 578 | nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2); |
| 579 | nostat = 1; |
| 580 | } else { |
| 581 | nlinks = -1; |
| 582 | nostat = 0; |
| 583 | } |
| 584 | |
| 585 | #ifdef notdef |
| 586 | (void)printf("nlinks == %d (cur: %u)\n", nlinks, cur->fts_nlink); |
| 587 | (void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n", |
| 588 | ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT)); |
| 589 | #endif |
| 590 | /* |
| 591 | * If we're going to need to stat anything or we want to descend |
| 592 | * and stay in the directory, chdir. If this fails we keep going, |
| 593 | * but set a flag so we don't chdir after the post-order visit. |
| 594 | * We won't be able to stat anything, but we can still return the |
| 595 | * names themselves. Note, that since fts_read won't be able to |
| 596 | * chdir into the directory, it will have to return different path |
| 597 | * names than before, i.e. "a/b" instead of "b". Since the node |
| 598 | * has already been visited in pre-order, have to wait until the |
| 599 | * post-order visit to return the error. There is a special case |
| 600 | * here, if there was nothing to stat then it's not an error to |
| 601 | * not be able to stat. This is all fairly nasty. If a program |
| 602 | * needed sorted entries or stat information, they had better be |
| 603 | * checking FTS_NS on the returned nodes. |
| 604 | */ |
| 605 | cderrno = 0; |
| 606 | if (nlinks || type == BREAD) { |
| 607 | if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) { |
| 608 | if (nlinks && type == BREAD) |
| 609 | cur->fts_errno = errno; |
| 610 | cur->fts_flags |= FTS_DONTCHDIR; |
| 611 | descend = 0; |
| 612 | cderrno = errno; |
| 613 | (void)closedir(dirp); |
| 614 | dirp = NULL; |
| 615 | } else |
| 616 | descend = 1; |
| 617 | } else |
| 618 | descend = 0; |
| 619 | |
| 620 | /* |
| 621 | * Figure out the max file name length that can be stored in the |
| 622 | * current path -- the inner loop allocates more path as necessary. |
| 623 | * We really wouldn't have to do the maxlen calculations here, we |
| 624 | * could do them in fts_read before returning the path, but it's a |
| 625 | * lot easier here since the length is part of the dirent structure. |
| 626 | * |
| 627 | * If not changing directories set a pointer so that can just append |
| 628 | * each new name into the path. |
| 629 | */ |
| 630 | len = NAPPEND(cur); |
| 631 | if (ISSET(FTS_NOCHDIR)) { |
| 632 | cp = sp->fts_path + len; |
| 633 | *cp++ = '/'; |
| 634 | } |
| 635 | len++; |
| 636 | maxlen = sp->fts_pathlen - len; |
| 637 | |
| 638 | /* |
| 639 | * fts_level is a short so we must prevent it from wrapping |
| 640 | * around to FTS_ROOTLEVEL and FTS_ROOTPARENTLEVEL. |
| 641 | */ |
| 642 | level = cur->fts_level; |
| 643 | if (level < FTS_MAXLEVEL) |
| 644 | level++; |
| 645 | |
| 646 | /* Read the directory, attaching each entry to the `link' pointer. */ |
| 647 | doadjust = 0; |
| 648 | for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) { |
| 649 | if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name)) |
| 650 | continue; |
| 651 | |
| 652 | if (!(p = fts_alloc(sp, dp->d_name, strlen(dp->d_name)))) |
| 653 | goto mem1; |
| 654 | if (strlen(dp->d_name) >= maxlen) { /* include space for NUL */ |
| 655 | oldaddr = sp->fts_path; |
| 656 | if (fts_palloc(sp, strlen(dp->d_name) +len + 1)) { |
| 657 | /* |
| 658 | * No more memory for path or structures. Save |
| 659 | * errno, free up the current structure and the |
| 660 | * structures already allocated. |
| 661 | */ |
| 662 | mem1: saved_errno = errno; |
| 663 | if (p) |
| 664 | free(p); |
| 665 | fts_lfree(head); |
| 666 | (void)closedir(dirp); |
| 667 | cur->fts_info = FTS_ERR; |
| 668 | SET(FTS_STOP); |
| 669 | errno = saved_errno; |
| 670 | return (NULL); |
| 671 | } |
| 672 | /* Did realloc() change the pointer? */ |
| 673 | if (oldaddr != sp->fts_path) { |
| 674 | doadjust = 1; |
| 675 | if (ISSET(FTS_NOCHDIR)) |
| 676 | cp = sp->fts_path + len; |
| 677 | } |
| 678 | maxlen = sp->fts_pathlen - len; |
| 679 | } |
| 680 | |
| 681 | p->fts_level = level; |
| 682 | p->fts_parent = sp->fts_cur; |
| 683 | p->fts_pathlen = len + strlen(dp->d_name); |
| 684 | if (p->fts_pathlen < len) { |
| 685 | /* |
| 686 | * If we wrap, free up the current structure and |
| 687 | * the structures already allocated, then error |
| 688 | * out with ENAMETOOLONG. |
| 689 | */ |
| 690 | free(p); |
| 691 | fts_lfree(head); |
| 692 | (void)closedir(dirp); |
| 693 | cur->fts_info = FTS_ERR; |
| 694 | SET(FTS_STOP); |
| 695 | errno = ENAMETOOLONG; |
| 696 | return (NULL); |
| 697 | } |
| 698 | |
| 699 | if (cderrno) { |
| 700 | if (nlinks) { |
| 701 | p->fts_info = FTS_NS; |
| 702 | p->fts_errno = cderrno; |
| 703 | } else |
| 704 | p->fts_info = FTS_NSOK; |
| 705 | p->fts_accpath = cur->fts_accpath; |
| 706 | } else if (nlinks == 0 |
| 707 | #ifdef DT_DIR |
| 708 | || (nostat && |
| 709 | dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) |
| 710 | #endif |
| 711 | ) { |
| 712 | p->fts_accpath = |
| 713 | ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name; |
| 714 | p->fts_info = FTS_NSOK; |
| 715 | } else { |
| 716 | /* Build a file name for fts_stat to stat. */ |
| 717 | if (ISSET(FTS_NOCHDIR)) { |
| 718 | p->fts_accpath = p->fts_path; |
| 719 | memmove(cp, p->fts_name, p->fts_namelen + 1); |
| 720 | } else |
| 721 | p->fts_accpath = p->fts_name; |
| 722 | /* Stat it. */ |
| 723 | p->fts_info = fts_stat(sp, p, 0); |
| 724 | |
| 725 | /* Decrement link count if applicable. */ |
| 726 | if (nlinks > 0 && (p->fts_info == FTS_D || |
| 727 | p->fts_info == FTS_DC || p->fts_info == FTS_DOT)) |
| 728 | --nlinks; |
| 729 | } |
| 730 | |
| 731 | /* We walk in directory order so "ls -f" doesn't get upset. */ |
| 732 | p->fts_link = NULL; |
| 733 | if (head == NULL) |
| 734 | head = tail = p; |
| 735 | else { |
| 736 | tail->fts_link = p; |
| 737 | tail = p; |
| 738 | } |
| 739 | ++nitems; |
| 740 | } |
| 741 | if (dirp) |
| 742 | (void)closedir(dirp); |
| 743 | |
| 744 | /* |
| 745 | * If realloc() changed the address of the path, adjust the |
| 746 | * addresses for the rest of the tree and the dir list. |
| 747 | */ |
| 748 | if (doadjust) |
| 749 | fts_padjust(sp, head); |
| 750 | |
| 751 | /* |
| 752 | * If not changing directories, reset the path back to original |
| 753 | * state. |
| 754 | */ |
| 755 | if (ISSET(FTS_NOCHDIR)) { |
| 756 | if (len == sp->fts_pathlen || nitems == 0) |
| 757 | --cp; |
| 758 | *cp = '\0'; |
| 759 | } |
| 760 | |
| 761 | /* |
| 762 | * If descended after called from fts_children or after called from |
| 763 | * fts_read and nothing found, get back. At the root level we use |
| 764 | * the saved fd; if one of fts_open()'s arguments is a relative path |
| 765 | * to an empty directory, we wind up here with no other way back. If |
| 766 | * can't get back, we're done. |
| 767 | */ |
| 768 | if (descend && (type == BCHILD || !nitems) && |
| 769 | (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) : |
| 770 | fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) { |
| 771 | cur->fts_info = FTS_ERR; |
| 772 | SET(FTS_STOP); |
| 773 | return (NULL); |
| 774 | } |
| 775 | |
| 776 | /* If didn't find anything, return NULL. */ |
| 777 | if (!nitems) { |
| 778 | if (type == BREAD) |
| 779 | cur->fts_info = FTS_DP; |
| 780 | return (NULL); |
| 781 | } |
| 782 | |
| 783 | /* Sort the entries. */ |
| 784 | if (sp->fts_compar && nitems > 1) |
| 785 | head = fts_sort(sp, head, nitems); |
| 786 | return (head); |
| 787 | } |
| 788 | |
| 789 | static u_short |
| 790 | fts_stat(FTS *sp, FTSENT *p, int follow) |
| 791 | { |
| 792 | FTSENT *t; |
| 793 | dev_t dev; |
| 794 | ino_t ino; |
| 795 | struct stat *sbp, sb; |
| 796 | int saved_errno; |
| 797 | |
| 798 | /* If user needs stat info, stat buffer already allocated. */ |
| 799 | sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp; |
| 800 | |
| 801 | /* |
| 802 | * If doing a logical walk, or application requested FTS_FOLLOW, do |
| 803 | * a stat(2). If that fails, check for a non-existent symlink. If |
| 804 | * fail, set the errno from the stat call. |
| 805 | */ |
| 806 | if (ISSET(FTS_LOGICAL) || follow) { |
| 807 | if (stat(p->fts_accpath, sbp)) { |
| 808 | saved_errno = errno; |
| 809 | if (!lstat(p->fts_accpath, sbp)) { |
| 810 | errno = 0; |
| 811 | return (FTS_SLNONE); |
| 812 | } |
| 813 | p->fts_errno = saved_errno; |
| 814 | goto err; |
| 815 | } |
| 816 | } else if (lstat(p->fts_accpath, sbp)) { |
| 817 | p->fts_errno = errno; |
| 818 | err: memset(sbp, 0, sizeof(struct stat)); |
| 819 | return (FTS_NS); |
| 820 | } |
| 821 | |
| 822 | if (S_ISDIR(sbp->st_mode)) { |
| 823 | /* |
| 824 | * Set the device/inode. Used to find cycles and check for |
| 825 | * crossing mount points. Also remember the link count, used |
| 826 | * in fts_build to limit the number of stat calls. It is |
| 827 | * understood that these fields are only referenced if fts_info |
| 828 | * is set to FTS_D. |
| 829 | */ |
| 830 | dev = p->fts_dev = sbp->st_dev; |
| 831 | ino = p->fts_ino = sbp->st_ino; |
| 832 | p->fts_nlink = sbp->st_nlink; |
| 833 | |
| 834 | if (ISDOT(p->fts_name)) |
| 835 | return (FTS_DOT); |
| 836 | |
| 837 | /* |
| 838 | * Cycle detection is done by brute force when the directory |
| 839 | * is first encountered. If the tree gets deep enough or the |
| 840 | * number of symbolic links to directories is high enough, |
| 841 | * something faster might be worthwhile. |
| 842 | */ |
| 843 | for (t = p->fts_parent; |
| 844 | t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent) |
| 845 | if (ino == t->fts_ino && dev == t->fts_dev) { |
| 846 | p->fts_cycle = t; |
| 847 | return (FTS_DC); |
| 848 | } |
| 849 | return (FTS_D); |
| 850 | } |
| 851 | if (S_ISLNK(sbp->st_mode)) |
| 852 | return (FTS_SL); |
| 853 | if (S_ISREG(sbp->st_mode)) |
| 854 | return (FTS_F); |
| 855 | return (FTS_DEFAULT); |
| 856 | } |
| 857 | |
| 858 | static FTSENT * |
| 859 | fts_sort(FTS *sp, FTSENT *head, int nitems) |
| 860 | { |
| 861 | FTSENT **ap, *p; |
| 862 | |
| 863 | /* |
| 864 | * Construct an array of pointers to the structures and call qsort(3). |
| 865 | * Reassemble the array in the order returned by qsort. If unable to |
| 866 | * sort for memory reasons, return the directory entries in their |
| 867 | * current order. Allocate enough space for the current needs plus |
| 868 | * 40 so don't realloc one entry at a time. |
| 869 | */ |
| 870 | if (nitems > sp->fts_nitems) { |
| 871 | struct _ftsent **a; |
| 872 | |
| 873 | sp->fts_nitems = nitems + 40; |
| 874 | if ((a = realloc(sp->fts_array, |
| 875 | sp->fts_nitems * sizeof(FTSENT *))) == NULL) { |
| 876 | if (sp->fts_array) |
| 877 | free(sp->fts_array); |
| 878 | sp->fts_array = NULL; |
| 879 | sp->fts_nitems = 0; |
| 880 | return (head); |
| 881 | } |
| 882 | sp->fts_array = a; |
| 883 | } |
| 884 | for (ap = sp->fts_array, p = head; p; p = p->fts_link) |
| 885 | *ap++ = p; |
| 886 | qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar); |
| 887 | for (head = *(ap = sp->fts_array); --nitems; ++ap) |
| 888 | ap[0]->fts_link = ap[1]; |
| 889 | ap[0]->fts_link = NULL; |
| 890 | return (head); |
| 891 | } |
| 892 | |
| 893 | static FTSENT * |
| 894 | fts_alloc(FTS *sp, char *name, size_t namelen) |
| 895 | { |
| 896 | FTSENT *p; |
| 897 | size_t len; |
| 898 | |
| 899 | /* |
| 900 | * The file name is a variable length array and no stat structure is |
| 901 | * necessary if the user has set the nostat bit. Allocate the FTSENT |
| 902 | * structure, the file name and the stat structure in one chunk, but |
| 903 | * be careful that the stat structure is reasonably aligned. Since the |
| 904 | * fts_name field is declared to be of size 1, the fts_name pointer is |
| 905 | * namelen + 2 before the first possible address of the stat structure. |
| 906 | */ |
| 907 | len = sizeof(FTSENT) + namelen; |
| 908 | if (!ISSET(FTS_NOSTAT)) |
| 909 | len += sizeof(struct stat) + ALIGNBYTES; |
| 910 | if ((p = malloc(len)) == NULL) |
| 911 | return (NULL); |
| 912 | |
| 913 | memset(p, 0, len); |
| 914 | p->fts_path = sp->fts_path; |
| 915 | p->fts_namelen = namelen; |
| 916 | p->fts_instr = FTS_NOINSTR; |
| 917 | if (!ISSET(FTS_NOSTAT)) |
| 918 | p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2); |
| 919 | memcpy(p->fts_name, name, namelen); |
| 920 | |
| 921 | return (p); |
| 922 | } |
| 923 | |
| 924 | static void |
| 925 | fts_lfree(FTSENT *head) |
| 926 | { |
| 927 | FTSENT *p; |
| 928 | |
| 929 | /* Free a linked list of structures. */ |
| 930 | while ((p = head)) { |
| 931 | head = head->fts_link; |
| 932 | free(p); |
| 933 | } |
| 934 | } |
| 935 | |
| 936 | /* |
| 937 | * Allow essentially unlimited paths; find, rm, ls should all work on any tree. |
| 938 | * Most systems will allow creation of paths much longer than MAXPATHLEN, even |
| 939 | * though the kernel won't resolve them. Add the size (not just what's needed) |
| 940 | * plus 256 bytes so don't realloc the path 2 bytes at a time. |
| 941 | */ |
| 942 | static int |
| 943 | fts_palloc(FTS *sp, size_t more) |
| 944 | { |
| 945 | char *p; |
| 946 | |
| 947 | /* |
| 948 | * Check for possible wraparound. |
| 949 | */ |
| 950 | more += 256; |
| 951 | if (sp->fts_pathlen + more < sp->fts_pathlen) { |
| 952 | if (sp->fts_path) |
| 953 | free(sp->fts_path); |
| 954 | sp->fts_path = NULL; |
| 955 | errno = ENAMETOOLONG; |
| 956 | return (1); |
| 957 | } |
| 958 | sp->fts_pathlen += more; |
| 959 | p = realloc(sp->fts_path, sp->fts_pathlen); |
| 960 | if (p == NULL) { |
| 961 | if (sp->fts_path) |
| 962 | free(sp->fts_path); |
| 963 | sp->fts_path = NULL; |
| 964 | return (1); |
| 965 | } |
| 966 | sp->fts_path = p; |
| 967 | return (0); |
| 968 | } |
| 969 | |
| 970 | /* |
| 971 | * When the path is realloc'd, have to fix all of the pointers in structures |
| 972 | * already returned. |
| 973 | */ |
| 974 | static void |
| 975 | fts_padjust(FTS *sp, FTSENT *head) |
| 976 | { |
| 977 | FTSENT *p; |
| 978 | char *addr = sp->fts_path; |
| 979 | |
| 980 | #define ADJUST(p) { \ |
| 981 | if ((p)->fts_accpath != (p)->fts_name) { \ |
| 982 | (p)->fts_accpath = \ |
| 983 | (char *)addr + ((p)->fts_accpath - (p)->fts_path); \ |
| 984 | } \ |
| 985 | (p)->fts_path = addr; \ |
| 986 | } |
| 987 | /* Adjust the current set of children. */ |
| 988 | for (p = sp->fts_child; p; p = p->fts_link) |
| 989 | ADJUST(p); |
| 990 | |
| 991 | /* Adjust the rest of the tree, including the current level. */ |
| 992 | for (p = head; p->fts_level >= FTS_ROOTLEVEL;) { |
| 993 | ADJUST(p); |
| 994 | p = p->fts_link ? p->fts_link : p->fts_parent; |
| 995 | } |
| 996 | } |
| 997 | |
| 998 | static size_t |
| 999 | fts_maxarglen(char * const *argv) |
| 1000 | { |
| 1001 | size_t len, max; |
| 1002 | |
| 1003 | for (max = 0; *argv; ++argv) |
| 1004 | if ((len = strlen(*argv)) > max) |
| 1005 | max = len; |
| 1006 | return (max + 1); |
| 1007 | } |
| 1008 | |
| 1009 | /* |
| 1010 | * Change to dir specified by fd or p->fts_accpath without getting |
| 1011 | * tricked by someone changing the world out from underneath us. |
| 1012 | * Assumes p->fts_dev and p->fts_ino are filled in. |
| 1013 | */ |
| 1014 | static int |
| 1015 | fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path) |
| 1016 | { |
| 1017 | int ret, oerrno, newfd; |
| 1018 | struct stat sb; |
| 1019 | |
| 1020 | newfd = fd; |
| 1021 | if (ISSET(FTS_NOCHDIR)) |
| 1022 | return (0); |
| 1023 | if (fd < 0 && (newfd = open(path, O_RDONLY, 0)) < 0) |
| 1024 | return (-1); |
| 1025 | if (fstat(newfd, &sb)) { |
| 1026 | ret = -1; |
| 1027 | goto bail; |
| 1028 | } |
| 1029 | if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) { |
| 1030 | errno = ENOENT; /* disinformation */ |
| 1031 | ret = -1; |
| 1032 | goto bail; |
| 1033 | } |
| 1034 | ret = fchdir(newfd); |
| 1035 | bail: |
| 1036 | oerrno = errno; |
| 1037 | if (fd < 0) |
| 1038 | (void)close(newfd); |
| 1039 | errno = oerrno; |
| 1040 | return (ret); |
| 1041 | } |