| 1 | /* |
| 2 | * Routines to support hard-linking. |
| 3 | * |
| 4 | * Copyright (C) 1996 Andrew Tridgell |
| 5 | * Copyright (C) 1996 Paul Mackerras |
| 6 | * Copyright (C) 2002 Martin Pool <mbp@samba.org> |
| 7 | * Copyright (C) 2004-2007 Wayne Davison |
| 8 | * |
| 9 | * This program is free software; you can redistribute it and/or modify |
| 10 | * it under the terms of the GNU General Public License as published by |
| 11 | * the Free Software Foundation; either version 3 of the License, or |
| 12 | * (at your option) any later version. |
| 13 | * |
| 14 | * This program is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | * GNU General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU General Public License along |
| 20 | * with this program; if not, visit the http://fsf.org website. |
| 21 | */ |
| 22 | |
| 23 | #include "rsync.h" |
| 24 | |
| 25 | extern int verbose; |
| 26 | extern int dry_run; |
| 27 | extern int am_sender; |
| 28 | extern int inc_recurse; |
| 29 | extern int do_xfers; |
| 30 | extern int link_dest; |
| 31 | extern int preserve_acls; |
| 32 | extern int make_backups; |
| 33 | extern int protocol_version; |
| 34 | extern int remove_source_files; |
| 35 | extern int stdout_format_has_i; |
| 36 | extern int maybe_ATTRS_REPORT; |
| 37 | extern char *basis_dir[]; |
| 38 | extern struct file_list *cur_flist; |
| 39 | #ifdef ICONV_OPTION |
| 40 | extern int ic_ndx; |
| 41 | #endif |
| 42 | |
| 43 | #ifdef SUPPORT_HARD_LINKS |
| 44 | |
| 45 | /* Starting with protocol 30, we use a simple hashtable on the sending side |
| 46 | * for hashing the st_dev and st_ino info. The receiving side gets told |
| 47 | * (via flags and a "group index") which items are hard-linked together, so |
| 48 | * we can avoid the pool of dev+inode data. For incremental recursion mode, |
| 49 | * the receiver will use a ndx hash to remember old pathnames. */ |
| 50 | |
| 51 | static struct hashtable *dev_tbl; |
| 52 | |
| 53 | static struct hashtable *prior_hlinks; |
| 54 | |
| 55 | static struct file_list *hlink_flist; |
| 56 | |
| 57 | void init_hard_links(void) |
| 58 | { |
| 59 | if (am_sender || protocol_version < 30) |
| 60 | dev_tbl = hashtable_create(16, SIZEOF_INT64 == 8); |
| 61 | else if (inc_recurse) |
| 62 | prior_hlinks = hashtable_create(1024, 0); |
| 63 | } |
| 64 | |
| 65 | struct ht_int64_node *idev_find(int64 dev, int64 ino) |
| 66 | { |
| 67 | static struct ht_int64_node *dev_node = NULL; |
| 68 | struct hashtable *tbl; |
| 69 | |
| 70 | if (!dev_node || dev_node->key != dev) { |
| 71 | /* We keep a separate hash table of inodes for every device. */ |
| 72 | dev_node = hashtable_find(dev_tbl, dev, 1); |
| 73 | if (!(tbl = dev_node->data)) |
| 74 | tbl = dev_node->data = hashtable_create(512, SIZEOF_INT64 == 8); |
| 75 | } else |
| 76 | tbl = dev_node->data; |
| 77 | |
| 78 | return hashtable_find(tbl, ino, 1); |
| 79 | } |
| 80 | |
| 81 | void idev_destroy(void) |
| 82 | { |
| 83 | int i; |
| 84 | |
| 85 | for (i = 0; i < dev_tbl->size; i++) { |
| 86 | struct ht_int32_node *node = HT_NODE(dev_tbl, dev_tbl->nodes, i); |
| 87 | if (node->data) |
| 88 | hashtable_destroy(node->data); |
| 89 | } |
| 90 | |
| 91 | hashtable_destroy(dev_tbl); |
| 92 | } |
| 93 | |
| 94 | static int hlink_compare_gnum(int *int1, int *int2) |
| 95 | { |
| 96 | struct file_struct *f1 = hlink_flist->sorted[*int1]; |
| 97 | struct file_struct *f2 = hlink_flist->sorted[*int2]; |
| 98 | int32 gnum1 = F_HL_GNUM(f1); |
| 99 | int32 gnum2 = F_HL_GNUM(f2); |
| 100 | |
| 101 | if (gnum1 != gnum2) |
| 102 | return gnum1 > gnum2 ? 1 : -1; |
| 103 | |
| 104 | return *int1 > *int2 ? 1 : -1; |
| 105 | } |
| 106 | |
| 107 | static void match_gnums(int32 *ndx_list, int ndx_count) |
| 108 | { |
| 109 | int32 from, prev; |
| 110 | struct file_struct *file, *file_next; |
| 111 | struct ht_int32_node *node = NULL; |
| 112 | int32 gnum, gnum_next; |
| 113 | |
| 114 | qsort(ndx_list, ndx_count, sizeof ndx_list[0], |
| 115 | (int (*)()) hlink_compare_gnum); |
| 116 | |
| 117 | for (from = 0; from < ndx_count; from++) { |
| 118 | file = hlink_flist->sorted[ndx_list[from]]; |
| 119 | gnum = F_HL_GNUM(file); |
| 120 | if (inc_recurse) { |
| 121 | node = hashtable_find(prior_hlinks, gnum, 1); |
| 122 | if (!node->data) { |
| 123 | node->data = new_array0(char, 5); |
| 124 | assert(gnum >= hlink_flist->ndx_start); |
| 125 | file->flags |= FLAG_HLINK_FIRST; |
| 126 | prev = -1; |
| 127 | } else if (CVAL(node->data, 0) == 0) { |
| 128 | struct file_list *flist; |
| 129 | struct file_struct *fp; |
| 130 | prev = IVAL(node->data, 1); |
| 131 | flist = flist_for_ndx(prev); |
| 132 | assert(flist != NULL); |
| 133 | fp = flist->files[prev - flist->ndx_start]; |
| 134 | fp->flags &= ~FLAG_HLINK_LAST; |
| 135 | } else |
| 136 | prev = -1; |
| 137 | } else { |
| 138 | file->flags |= FLAG_HLINK_FIRST; |
| 139 | prev = -1; |
| 140 | } |
| 141 | for ( ; from < ndx_count-1; file = file_next, gnum = gnum_next, from++) { /*SHARED ITERATOR*/ |
| 142 | file_next = hlink_flist->sorted[ndx_list[from+1]]; |
| 143 | gnum_next = F_HL_GNUM(file_next); |
| 144 | if (gnum != gnum_next) |
| 145 | break; |
| 146 | F_HL_PREV(file) = prev; |
| 147 | /* The linked list must use raw ndx values. */ |
| 148 | #ifdef ICONV_OPTION |
| 149 | if (ic_ndx) |
| 150 | prev = F_NDX(file); |
| 151 | else |
| 152 | #endif |
| 153 | prev = ndx_list[from] + hlink_flist->ndx_start; |
| 154 | } |
| 155 | if (prev < 0 && !inc_recurse) { |
| 156 | /* Disable hard-link bit and set DONE so that |
| 157 | * HLINK_BUMP()-dependent values are unaffected. */ |
| 158 | file->flags &= ~(FLAG_HLINKED | FLAG_HLINK_FIRST); |
| 159 | file->flags |= FLAG_HLINK_DONE; |
| 160 | continue; |
| 161 | } |
| 162 | |
| 163 | file->flags |= FLAG_HLINK_LAST; |
| 164 | F_HL_PREV(file) = prev; |
| 165 | if (inc_recurse && CVAL(node->data, 0) == 0) { |
| 166 | #ifdef ICONV_OPTION |
| 167 | if (ic_ndx) |
| 168 | prev = F_NDX(file); |
| 169 | else |
| 170 | #endif |
| 171 | prev = ndx_list[from] + hlink_flist->ndx_start; |
| 172 | SIVAL(node->data, 1, prev); |
| 173 | } |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | /* Analyze the hard-links in the file-list by creating a list of all the |
| 178 | * items that have hlink data, sorting them, and matching up identical |
| 179 | * values into clusters. These will be a single linked list from last |
| 180 | * to first when we're done. */ |
| 181 | void match_hard_links(struct file_list *flist) |
| 182 | { |
| 183 | int i, ndx_count = 0; |
| 184 | int32 *ndx_list; |
| 185 | |
| 186 | if (!(ndx_list = new_array(int32, flist->used))) |
| 187 | out_of_memory("match_hard_links"); |
| 188 | |
| 189 | for (i = 0; i < flist->used; i++) { |
| 190 | if (F_IS_HLINKED(flist->sorted[i])) |
| 191 | ndx_list[ndx_count++] = i; |
| 192 | } |
| 193 | |
| 194 | hlink_flist = flist; |
| 195 | |
| 196 | if (ndx_count) |
| 197 | match_gnums(ndx_list, ndx_count); |
| 198 | |
| 199 | free(ndx_list); |
| 200 | if (protocol_version < 30) |
| 201 | idev_destroy(); |
| 202 | } |
| 203 | |
| 204 | static int maybe_hard_link(struct file_struct *file, int ndx, |
| 205 | const char *fname, int statret, statx *sxp, |
| 206 | const char *oldname, STRUCT_STAT *old_stp, |
| 207 | const char *realname, int itemizing, enum logcode code) |
| 208 | { |
| 209 | if (statret == 0) { |
| 210 | if (sxp->st.st_dev == old_stp->st_dev |
| 211 | && sxp->st.st_ino == old_stp->st_ino) { |
| 212 | if (itemizing) { |
| 213 | itemize(fname, file, ndx, statret, sxp, |
| 214 | ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, |
| 215 | 0, ""); |
| 216 | } |
| 217 | if (verbose > 1 && maybe_ATTRS_REPORT) |
| 218 | rprintf(FCLIENT, "%s is uptodate\n", fname); |
| 219 | file->flags |= FLAG_HLINK_DONE; |
| 220 | return 0; |
| 221 | } |
| 222 | if (make_backups > 0) { |
| 223 | if (!make_backup(fname)) |
| 224 | return -1; |
| 225 | } else if (robust_unlink(fname)) { |
| 226 | rsyserr(FERROR, errno, "unlink %s failed", |
| 227 | full_fname(fname)); |
| 228 | return -1; |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | if (hard_link_one(file, fname, oldname, 0)) { |
| 233 | if (itemizing) { |
| 234 | itemize(fname, file, ndx, statret, sxp, |
| 235 | ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0, |
| 236 | realname); |
| 237 | } |
| 238 | if (code != FNONE && verbose) |
| 239 | rprintf(code, "%s => %s\n", fname, realname); |
| 240 | return 0; |
| 241 | } |
| 242 | return -1; |
| 243 | } |
| 244 | |
| 245 | /* Figure out if a prior entry is still there or if we just have a |
| 246 | * cached name for it. Never called with a FLAG_HLINK_FIRST entry. */ |
| 247 | static char *check_prior(int prev_ndx, int gnum, struct file_list **flist_p) |
| 248 | { |
| 249 | struct file_list *flist = flist_for_ndx(prev_ndx); |
| 250 | struct ht_int32_node *node; |
| 251 | |
| 252 | if (flist) { |
| 253 | *flist_p = flist; |
| 254 | return NULL; |
| 255 | } |
| 256 | |
| 257 | node = hashtable_find(prior_hlinks, gnum, 0); |
| 258 | assert(node != NULL && node->data); |
| 259 | assert(CVAL(node->data, 0) != 0); |
| 260 | return node->data; |
| 261 | } |
| 262 | |
| 263 | /* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns: |
| 264 | * 0 = process the file, 1 = skip the file, -1 = error occurred. */ |
| 265 | int hard_link_check(struct file_struct *file, int ndx, const char *fname, |
| 266 | int statret, statx *sxp, int itemizing, |
| 267 | enum logcode code) |
| 268 | { |
| 269 | STRUCT_STAT prev_st; |
| 270 | char namebuf[MAXPATHLEN], altbuf[MAXPATHLEN]; |
| 271 | char *realname, *prev_name; |
| 272 | struct file_list *flist; |
| 273 | int gnum = inc_recurse ? F_HL_GNUM(file) : -1; |
| 274 | int prev_ndx = F_HL_PREV(file); |
| 275 | |
| 276 | prev_name = realname = check_prior(prev_ndx, gnum, &flist); |
| 277 | |
| 278 | if (!prev_name) { |
| 279 | struct file_struct *prev_file = flist->files[prev_ndx - flist->ndx_start]; |
| 280 | |
| 281 | /* Is the previous link is not complete yet? */ |
| 282 | if (!(prev_file->flags & FLAG_HLINK_DONE)) { |
| 283 | /* Is the previous link being transferred? */ |
| 284 | if (prev_file->flags & FLAG_FILE_SENT) { |
| 285 | /* Add ourselves to the list of files that will be |
| 286 | * updated when the transfer completes, and mark |
| 287 | * ourself as waiting for the transfer. */ |
| 288 | F_HL_PREV(file) = F_HL_PREV(prev_file); |
| 289 | F_HL_PREV(prev_file) = ndx; |
| 290 | file->flags |= FLAG_FILE_SENT; |
| 291 | cur_flist->in_progress++; |
| 292 | return 1; |
| 293 | } |
| 294 | return 0; |
| 295 | } |
| 296 | |
| 297 | /* There is a finished file to link with! */ |
| 298 | if (!(prev_file->flags & FLAG_HLINK_FIRST)) { |
| 299 | /* The previous previous is FIRST when prev is not. */ |
| 300 | prev_ndx = F_HL_PREV(prev_file); |
| 301 | prev_name = realname = check_prior(prev_ndx, gnum, &flist); |
| 302 | /* Update our previous pointer to point to the FIRST. */ |
| 303 | F_HL_PREV(file) = prev_ndx; |
| 304 | } |
| 305 | |
| 306 | if (!prev_name) { |
| 307 | int alt_dest; |
| 308 | |
| 309 | prev_file = flist->files[prev_ndx - flist->ndx_start]; |
| 310 | /* F_HL_PREV() is alt_dest value when DONE && FIRST. */ |
| 311 | alt_dest = F_HL_PREV(prev_file); |
| 312 | |
| 313 | if (alt_dest >= 0 && dry_run) { |
| 314 | pathjoin(namebuf, MAXPATHLEN, basis_dir[alt_dest], |
| 315 | f_name(prev_file, NULL)); |
| 316 | prev_name = namebuf; |
| 317 | realname = f_name(prev_file, altbuf); |
| 318 | } else { |
| 319 | prev_name = f_name(prev_file, namebuf); |
| 320 | realname = prev_name; |
| 321 | } |
| 322 | } |
| 323 | } |
| 324 | |
| 325 | if (link_stat(prev_name, &prev_st, 0) < 0) { |
| 326 | rsyserr(FERROR, errno, "stat %s failed", |
| 327 | full_fname(prev_name)); |
| 328 | return -1; |
| 329 | } |
| 330 | |
| 331 | if (statret < 0 && basis_dir[0] != NULL) { |
| 332 | /* If we match an alt-dest item, we don't output this as a change. */ |
| 333 | char cmpbuf[MAXPATHLEN]; |
| 334 | statx alt_sx; |
| 335 | int j = 0; |
| 336 | #ifdef SUPPORT_ACLS |
| 337 | alt_sx.acc_acl = alt_sx.def_acl = NULL; |
| 338 | #endif |
| 339 | do { |
| 340 | pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname); |
| 341 | if (link_stat(cmpbuf, &alt_sx.st, 0) < 0) |
| 342 | continue; |
| 343 | if (link_dest) { |
| 344 | if (prev_st.st_dev != alt_sx.st.st_dev |
| 345 | || prev_st.st_ino != alt_sx.st.st_ino) |
| 346 | continue; |
| 347 | statret = 1; |
| 348 | if (verbose < 2 || !stdout_format_has_i) { |
| 349 | itemizing = 0; |
| 350 | code = FNONE; |
| 351 | if (verbose > 1 && maybe_ATTRS_REPORT) |
| 352 | rprintf(FCLIENT, "%s is uptodate\n", fname); |
| 353 | } |
| 354 | break; |
| 355 | } |
| 356 | if (!unchanged_file(cmpbuf, file, &alt_sx.st)) |
| 357 | continue; |
| 358 | statret = 1; |
| 359 | if (unchanged_attrs(cmpbuf, file, &alt_sx)) |
| 360 | break; |
| 361 | } while (basis_dir[++j] != NULL); |
| 362 | if (statret == 1) { |
| 363 | sxp->st = alt_sx.st; |
| 364 | #ifdef SUPPORT_ACLS |
| 365 | if (preserve_acls && !S_ISLNK(file->mode)) { |
| 366 | if (!ACL_READY(*sxp)) |
| 367 | get_acl(cmpbuf, sxp); |
| 368 | else { |
| 369 | sxp->acc_acl = alt_sx.acc_acl; |
| 370 | sxp->def_acl = alt_sx.def_acl; |
| 371 | } |
| 372 | } |
| 373 | #endif |
| 374 | } |
| 375 | #ifdef SUPPORT_ACLS |
| 376 | else if (preserve_acls) |
| 377 | free_acl(&alt_sx); |
| 378 | #endif |
| 379 | } |
| 380 | |
| 381 | if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st, |
| 382 | realname, itemizing, code) < 0) |
| 383 | return -1; |
| 384 | |
| 385 | if (remove_source_files == 1 && do_xfers) |
| 386 | send_msg_int(MSG_SUCCESS, ndx); |
| 387 | |
| 388 | return 1; |
| 389 | } |
| 390 | |
| 391 | int hard_link_one(struct file_struct *file, const char *fname, |
| 392 | const char *oldname, int terse) |
| 393 | { |
| 394 | if (do_link(oldname, fname) < 0) { |
| 395 | enum logcode code; |
| 396 | if (terse) { |
| 397 | if (!verbose) |
| 398 | return -1; |
| 399 | code = FINFO; |
| 400 | } else |
| 401 | code = FERROR; |
| 402 | rsyserr(code, errno, "link %s => %s failed", |
| 403 | full_fname(fname), oldname); |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | file->flags |= FLAG_HLINK_DONE; |
| 408 | |
| 409 | return 1; |
| 410 | } |
| 411 | |
| 412 | void finish_hard_link(struct file_struct *file, const char *fname, int fin_ndx, |
| 413 | STRUCT_STAT *stp, int itemizing, enum logcode code, |
| 414 | int alt_dest) |
| 415 | { |
| 416 | statx prev_sx; |
| 417 | STRUCT_STAT st; |
| 418 | char alt_name[MAXPATHLEN], *prev_name; |
| 419 | const char *our_name; |
| 420 | struct file_list *flist; |
| 421 | int prev_statret, ndx, prev_ndx = F_HL_PREV(file); |
| 422 | |
| 423 | if (stp == NULL && prev_ndx >= 0) { |
| 424 | if (link_stat(fname, &st, 0) < 0) { |
| 425 | rsyserr(FERROR, errno, "stat %s failed", |
| 426 | full_fname(fname)); |
| 427 | return; |
| 428 | } |
| 429 | stp = &st; |
| 430 | } |
| 431 | |
| 432 | /* FIRST combined with DONE means we were the first to get done. */ |
| 433 | file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE; |
| 434 | F_HL_PREV(file) = alt_dest; |
| 435 | if (alt_dest >= 0 && dry_run) { |
| 436 | pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest], |
| 437 | f_name(file, NULL)); |
| 438 | our_name = alt_name; |
| 439 | } else |
| 440 | our_name = fname; |
| 441 | |
| 442 | #ifdef SUPPORT_ACLS |
| 443 | prev_sx.acc_acl = prev_sx.def_acl = NULL; |
| 444 | #endif |
| 445 | |
| 446 | while ((ndx = prev_ndx) >= 0) { |
| 447 | int val; |
| 448 | flist = flist_for_ndx(ndx); |
| 449 | assert(flist != NULL); |
| 450 | file = flist->files[ndx - flist->ndx_start]; |
| 451 | file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE; |
| 452 | prev_ndx = F_HL_PREV(file); |
| 453 | F_HL_PREV(file) = fin_ndx; |
| 454 | prev_name = f_name(file, NULL); |
| 455 | prev_statret = link_stat(prev_name, &prev_sx.st, 0); |
| 456 | val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx, |
| 457 | our_name, stp, fname, itemizing, code); |
| 458 | flist->in_progress--; |
| 459 | #ifdef SUPPORT_ACLS |
| 460 | if (preserve_acls) |
| 461 | free_acl(&prev_sx); |
| 462 | #endif |
| 463 | if (val < 0) |
| 464 | continue; |
| 465 | if (remove_source_files == 1 && do_xfers) |
| 466 | send_msg_int(MSG_SUCCESS, ndx); |
| 467 | } |
| 468 | |
| 469 | if (inc_recurse) { |
| 470 | int gnum = F_HL_GNUM(file); |
| 471 | struct ht_int32_node *node = hashtable_find(prior_hlinks, gnum, 0); |
| 472 | assert(node != NULL && node->data != NULL); |
| 473 | assert(CVAL(node->data, 0) == 0); |
| 474 | free(node->data); |
| 475 | if (!(node->data = strdup(our_name))) |
| 476 | out_of_memory("finish_hard_link"); |
| 477 | } |
| 478 | } |
| 479 | #endif |