hlink.c

   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 version 3 as
  11  * published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with this program; if not, visit the http://fsf.org website.
  20  */
  21
  22 #include "rsync.h"
  23
  24 extern int verbose;
  25 extern int dry_run;
  26 extern int do_xfers;
  27 extern int link_dest;
  28 extern int preserve_acls;
  29 extern int make_backups;
  30 extern int protocol_version;
  31 extern int remove_source_files;
  32 extern int stdout_format_has_i;
  33 extern int maybe_ATTRS_REPORT;
  34 extern char *basis_dir[];
  35 extern struct file_list *cur_flist;
  36 #ifdef ICONV_OPTION
  37 extern int ic_ndx;
  38 #endif
  39
  40 #ifdef SUPPORT_HARD_LINKS
  41
  42 #define HASH_LOAD_LIMIT(size) ((size)*3/4)
  43
  44 struct ihash_table {
  45         int32 size;
  46         int32 entries;
  47         struct idev_node *buckets;
  48 } *dev_tbl;
  49
  50 static struct idev_node *ihash_node(struct ihash_table *tbl, int64 key);
  51
  52 /* Starting with protocol 30, we use a simple hashtable on the sending side
  53  * for hashing the st_dev and st_ino info.  The receiving side gets told
  54  * (via flags and a "group index") which items are hard-linked together, so
  55  * we can avoid the pool of dev+inode data. */
  56
  57 static struct ihash_table *ihash_create(int size)
  58 {
  59         struct ihash_table *tbl;
  60
  61         /* Pick a power of 2 that can hold the requested size. */
  62         if (size & (size-1) || size < 16) {
  63                 int req = size;
  64                 size = 16;
  65                 while (size < req)
  66                         size *= 2;
  67         }
  68
  69         if (!(tbl = new(struct ihash_table))
  70          || !(tbl->buckets = new_array(struct idev_node, size)))
  71                 out_of_memory("ihash_create");
  72         memset(tbl->buckets, 0, size * sizeof tbl->buckets[0]);
  73         tbl->size = size;
  74         tbl->entries = 0;
  75
  76         return tbl;
  77 }
  78
  79 static void ihash_destroy(struct ihash_table *tbl)
  80 {
  81         free(tbl->buckets);
  82         free(tbl);
  83 }
  84
  85 void init_hard_links(void)
  86 {
  87         dev_tbl = ihash_create(16);
  88 }
  89
  90 static void expand_ihash(struct ihash_table *tbl)
  91 {
  92         struct idev_node *old_buckets = tbl->buckets;
  93         int size = tbl->size * 2;
  94         int i;
  95
  96         if (!(tbl->buckets = new_array(struct idev_node, size)))
  97                 out_of_memory("ihash_create");
  98         memset(tbl->buckets, 0, size * sizeof (struct idev_node));
  99         tbl->size = size;
 100         tbl->entries = 0;
 101
 102         for (i = size / 2; i-- > 0; ) {
 103                 int64 key = old_buckets[i].key;
 104                 if (key == 0)
 105                         continue;
 106                 ihash_node(tbl, key)->data = old_buckets[i].data;
 107         }
 108
 109         free(old_buckets);
 110 }
 111
 112 /* This returns the node for the indicated key, either newly created,
 113  * or already existing. */
 114 static struct idev_node *ihash_node(struct ihash_table *tbl, int64 key)
 115 {
 116         uint32 bkt;
 117
 118         if (tbl->entries > HASH_LOAD_LIMIT(tbl->size))
 119                 expand_ihash(tbl);
 120
 121 #if SIZEOF_INT64 < 8
 122         /* Based on Jenkins One-at-a-time hash. */
 123         {
 124                 uchar *keyp = (uchar*)&key;
 125                 int i;
 126
 127                 for (bkt = 0, i = 0; i < SIZEOF_INT64; i++) {
 128                         bkt += keyp[i];
 129                         bkt += (bkt << 10);
 130                         bkt ^= (bkt >> 6);
 131                 }
 132                 bkt += (bkt << 3);
 133                 bkt ^= (bkt >> 11);
 134                 bkt += (bkt << 15);
 135         }
 136 #else
 137 #define rot(x,k) (((x)<<(k)) ^ ((x)>>(32-(k))))
 138         /* Based on Jenkins hashword() from lookup3.c. */
 139         {
 140                 uint32 a, b, c;
 141
 142                 /* Set up the internal state */
 143                 a = b = c = 0xdeadbeef + (8 << 2);
 144
 145                 b += (uint32)(key >> 32);
 146                 a += (uint32)key;
 147                 c ^= b; c -= rot(b, 14);
 148                 a ^= c; a -= rot(c, 11);
 149                 b ^= a; b -= rot(a, 25);
 150                 c ^= b; c -= rot(b, 16);
 151                 a ^= c; a -= rot(c, 4);
 152                 b ^= a; b -= rot(a, 14);
 153                 c ^= b; c -= rot(b, 24);
 154                 bkt = c;
 155         }
 156 #endif
 157
 158         /* If it already exists, return it. */
 159         while (1) {
 160                 bkt &= tbl->size - 1;
 161                 if (tbl->buckets[bkt].key == key)
 162                         return &tbl->buckets[bkt];
 163                 if (tbl->buckets[bkt].key == 0)
 164                         break;
 165                 bkt++;
 166         }
 167
 168         /* Otherwise, take over this empty spot and then return it. */
 169         tbl->buckets[bkt].key = key;
 170         tbl->entries++;
 171         return &tbl->buckets[bkt];
 172 }
 173
 174 struct idev_node *idev_node(int64 dev, int64 ino)
 175 {
 176         static struct idev_node *dev_node = NULL;
 177         struct ihash_table *tbl;
 178
 179         if (!dev_node || dev_node->key != dev) {
 180                 /* We keep a separate hash table of inodes for every device. */
 181                 dev_node = ihash_node(dev_tbl, dev);
 182                 if (!(tbl = dev_node->data))
 183                         tbl = dev_node->data = ihash_create(512);
 184         } else
 185                 tbl = dev_node->data;
 186
 187         return ihash_node(tbl, ino);
 188 }
 189
 190 void idev_destroy(void)
 191 {
 192         int i;
 193
 194         for (i = 0; i < dev_tbl->size; i++) {
 195                 if (dev_tbl->buckets[i].data)
 196                         ihash_destroy(dev_tbl->buckets[i].data);
 197         }
 198
 199         ihash_destroy(dev_tbl);
 200 }
 201
 202 static int hlink_compare_gnum(int *int1, int *int2)
 203 {
 204         struct file_struct *f1 = cur_flist->sorted[*int1];
 205         struct file_struct *f2 = cur_flist->sorted[*int2];
 206         int32 gnum1 = F_HL_GNUM(f1);
 207         int32 gnum2 = F_HL_GNUM(f2);
 208
 209         if (gnum1 != gnum2)
 210                 return gnum1 > gnum2 ? 1 : -1;
 211
 212         return *int1 > *int2 ? 1 : -1;
 213 }
 214
 215 static void match_gnums(int32 *ndx_list, int ndx_count)
 216 {
 217         int32 from, prev;
 218         struct file_struct *file, *file_next;
 219         int32 gnum, gnum_next;
 220
 221         qsort(ndx_list, ndx_count, sizeof ndx_list[0],
 222              (int (*)()) hlink_compare_gnum);
 223
 224         for (from = 0; from < ndx_count; from++) {
 225                 for (file = cur_flist->sorted[ndx_list[from]], gnum = F_HL_GNUM(file), prev = -1;
 226                      from < ndx_count-1;
 227                      file = file_next, gnum = gnum_next, from++) /*SHARED ITERATOR*/
 228                 {
 229                         file_next = cur_flist->sorted[ndx_list[from+1]];
 230                         gnum_next = F_HL_GNUM(file_next);
 231                         if (gnum != gnum_next)
 232                                 break;
 233                         if (prev < 0)
 234                                 file->flags |= FLAG_HLINK_FIRST;
 235                         F_HL_PREV(file) = prev;
 236                         /* The linked list must use raw ndx values. */
 237 #ifdef ICONV_OPTION
 238                         if (ic_ndx)
 239                                 prev = F_NDX(file);
 240                         else
 241 #endif
 242                                 prev = ndx_list[from];
 243                 }
 244                 if (prev < 0)
 245                         file->flags &= ~FLAG_HLINKED;
 246                 else {
 247                         file->flags |= FLAG_HLINK_LAST;
 248                         F_HL_PREV(file) = prev;
 249                 }
 250         }
 251 }
 252
 253 /* Analyze the hard-links in the file-list by creating a list of all the
 254  * items that have hlink data, sorting them, and matching up identical
 255  * values into clusters.  These will be a single linked list from last
 256  * to first when we're done. */
 257 void match_hard_links(void)
 258 {
 259         int i, ndx_count = 0;
 260         int32 *ndx_list;
 261
 262         if (!(ndx_list = new_array(int32, cur_flist->used)))
 263                 out_of_memory("match_hard_links");
 264
 265         for (i = 0; i < cur_flist->used; i++) {
 266                 if (F_IS_HLINKED(cur_flist->sorted[i]))
 267                         ndx_list[ndx_count++] = i;
 268         }
 269
 270         if (ndx_count)
 271                 match_gnums(ndx_list, ndx_count);
 272
 273         free(ndx_list);
 274         if (protocol_version < 30)
 275                 idev_destroy();
 276 }
 277
 278 static int maybe_hard_link(struct file_struct *file, int ndx,
 279                            const char *fname, int statret, statx *sxp,
 280                            const char *oldname, STRUCT_STAT *old_stp,
 281                            const char *realname, int itemizing, enum logcode code)
 282 {
 283         if (statret == 0) {
 284                 if (sxp->st.st_dev == old_stp->st_dev
 285                  && sxp->st.st_ino == old_stp->st_ino) {
 286                         if (itemizing) {
 287                                 itemize(fname, file, ndx, statret, sxp,
 288                                         ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS,
 289                                         0, "");
 290                         }
 291                         if (verbose > 1 && maybe_ATTRS_REPORT)
 292                                 rprintf(FCLIENT, "%s is uptodate\n", fname);
 293                         file->flags |= FLAG_HLINK_DONE;
 294                         return 0;
 295                 }
 296                 if (make_backups > 0) {
 297                         if (!make_backup(fname))
 298                                 return -1;
 299                 } else if (robust_unlink(fname)) {
 300                         rsyserr(FERROR, errno, "unlink %s failed",
 301                                 full_fname(fname));
 302                         return -1;
 303                 }
 304         }
 305
 306         if (hard_link_one(file, fname, oldname, 0)) {
 307                 if (itemizing) {
 308                         itemize(fname, file, ndx, statret, sxp,
 309                                 ITEM_LOCAL_CHANGE | ITEM_XNAME_FOLLOWS, 0,
 310                                 realname);
 311                 }
 312                 if (code != FNONE && verbose)
 313                         rprintf(code, "%s => %s\n", fname, realname);
 314                 return 0;
 315         }
 316         return -1;
 317 }
 318
 319 /* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not.  Returns:
 320  * 0 = process the file, 1 = skip the file, -1 = error occurred. */
 321 int hard_link_check(struct file_struct *file, int ndx, const char *fname,
 322                     int statret, statx *sxp, int itemizing,
 323                     enum logcode code)
 324 {
 325         STRUCT_STAT prev_st;
 326         char prev_name[MAXPATHLEN], altbuf[MAXPATHLEN], *realname;
 327         int alt_dest, prev_ndx = F_HL_PREV(file);
 328         struct file_struct *prev_file = cur_flist->files[prev_ndx];
 329
 330         /* Is the previous link is not complete yet? */
 331         if (!(prev_file->flags & FLAG_HLINK_DONE)) {
 332                 /* Is the previous link being transferred? */
 333                 if (prev_file->flags & FLAG_FILE_SENT) {
 334                         /* Add ourselves to the list of files that will be
 335                          * updated when the transfer completes, and mark
 336                          * ourself as waiting for the transfer. */
 337                         F_HL_PREV(file) = F_HL_PREV(prev_file);
 338                         F_HL_PREV(prev_file) = ndx;
 339                         file->flags |= FLAG_FILE_SENT;
 340                         return 1;
 341                 }
 342                 return 0;
 343         }
 344
 345         /* There is a finished file to link with! */
 346         if (!(prev_file->flags & FLAG_HLINK_FIRST)) {
 347                 /* The previous previous will be marked with FIRST. */
 348                 prev_ndx = F_HL_PREV(prev_file);
 349                 prev_file = cur_flist->files[prev_ndx];
 350                 /* Update our previous pointer to point to the first. */
 351                 F_HL_PREV(file) = prev_ndx;
 352         }
 353         alt_dest = F_HL_PREV(prev_file); /* alternate value when DONE && FIRST */
 354         if (alt_dest >= 0 && dry_run) {
 355                 pathjoin(prev_name, MAXPATHLEN, basis_dir[alt_dest],
 356                          f_name(prev_file, NULL));
 357                 f_name(prev_file, altbuf);
 358                 realname = altbuf;
 359         } else {
 360                 f_name(prev_file, prev_name);
 361                 realname = prev_name;
 362         }
 363
 364         if (link_stat(prev_name, &prev_st, 0) < 0) {
 365                 rsyserr(FERROR, errno, "stat %s failed",
 366                         full_fname(prev_name));
 367                 return -1;
 368         }
 369
 370         if (statret < 0 && basis_dir[0] != NULL) {
 371                 /* If we match an alt-dest item, we don't output this as a change. */
 372                 char cmpbuf[MAXPATHLEN];
 373                 statx alt_sx;
 374                 int j = 0;
 375 #ifdef SUPPORT_ACLS
 376                 alt_sx.acc_acl = alt_sx.def_acl = NULL;
 377 #endif
 378                 do {
 379                         pathjoin(cmpbuf, MAXPATHLEN, basis_dir[j], fname);
 380                         if (link_stat(cmpbuf, &alt_sx.st, 0) < 0)
 381                                 continue;
 382                         if (link_dest) {
 383                                 if (prev_st.st_dev != alt_sx.st.st_dev
 384                                  || prev_st.st_ino != alt_sx.st.st_ino)
 385                                         continue;
 386                                 statret = 1;
 387                                 if (verbose < 2 || !stdout_format_has_i) {
 388                                         itemizing = 0;
 389                                         code = FNONE;
 390                                         if (verbose > 1 && maybe_ATTRS_REPORT)
 391                                                 rprintf(FCLIENT, "%s is uptodate\n", fname);
 392                                 }
 393                                 break;
 394                         }
 395                         if (!unchanged_file(cmpbuf, file, &alt_sx.st))
 396                                 continue;
 397                         statret = 1;
 398                         if (unchanged_attrs(cmpbuf, file, &alt_sx))
 399                                 break;
 400                 } while (basis_dir[++j] != NULL);
 401                 if (statret == 1) {
 402                         sxp->st = alt_sx.st;
 403 #ifdef SUPPORT_ACLS
 404                         if (preserve_acls && !S_ISLNK(file->mode)) {
 405                                 if (!ACL_READY(*sxp))
 406                                         get_acl(cmpbuf, sxp);
 407                                 else {
 408                                         sxp->acc_acl = alt_sx.acc_acl;
 409                                         sxp->def_acl = alt_sx.def_acl;
 410                                 }
 411                         }
 412 #endif
 413                 }
 414 #ifdef SUPPORT_ACLS
 415                 else if (preserve_acls)
 416                         free_acl(&alt_sx);
 417 #endif
 418         }
 419
 420         if (maybe_hard_link(file, ndx, fname, statret, sxp, prev_name, &prev_st,
 421                             realname, itemizing, code) < 0)
 422                 return -1;
 423
 424         if (remove_source_files == 1 && do_xfers)
 425                 send_msg_int(MSG_SUCCESS, ndx);
 426
 427         return 1;
 428 }
 429
 430 int hard_link_one(struct file_struct *file, const char *fname,
 431                   const char *oldname, int terse)
 432 {
 433         if (do_link(oldname, fname) < 0) {
 434                 enum logcode code;
 435                 if (terse) {
 436                         if (!verbose)
 437                                 return -1;
 438                         code = FINFO;
 439                 } else
 440                         code = FERROR;
 441                 rsyserr(code, errno, "link %s => %s failed",
 442                         full_fname(fname), oldname);
 443                 return 0;
 444         }
 445
 446         file->flags |= FLAG_HLINK_DONE;
 447
 448         return 1;
 449 }
 450
 451 void finish_hard_link(struct file_struct *file, const char *fname,
 452                       STRUCT_STAT *stp, int itemizing, enum logcode code,
 453                       int alt_dest)
 454 {
 455         statx prev_sx;
 456         STRUCT_STAT st;
 457         char alt_name[MAXPATHLEN], *prev_name;
 458         const char *our_name;
 459         int prev_statret, ndx, prev_ndx = F_HL_PREV(file);
 460
 461         if (stp == NULL && prev_ndx >= 0) {
 462                 if (link_stat(fname, &st, 0) < 0) {
 463                         rsyserr(FERROR, errno, "stat %s failed",
 464                                 full_fname(fname));
 465                         return;
 466                 }
 467                 stp = &st;
 468         }
 469
 470         /* FIRST combined with DONE means we were the first to get done. */
 471         file->flags |= FLAG_HLINK_FIRST | FLAG_HLINK_DONE;
 472         F_HL_PREV(file) = alt_dest;
 473         if (alt_dest >= 0 && dry_run) {
 474                 pathjoin(alt_name, MAXPATHLEN, basis_dir[alt_dest],
 475                          f_name(file, NULL));
 476                 our_name = alt_name;
 477         } else
 478                 our_name = fname;
 479
 480 #ifdef SUPPORT_ACLS
 481         prev_sx.acc_acl = prev_sx.def_acl = NULL;
 482 #endif
 483
 484         while ((ndx = prev_ndx) >= 0) {
 485                 int val;
 486                 file = cur_flist->files[ndx];
 487                 file->flags = (file->flags & ~FLAG_HLINK_FIRST) | FLAG_HLINK_DONE;
 488                 prev_ndx = F_HL_PREV(file);
 489                 prev_name = f_name(file, NULL);
 490                 prev_statret = link_stat(prev_name, &prev_sx.st, 0);
 491                 val = maybe_hard_link(file, ndx, prev_name, prev_statret, &prev_sx,
 492                                       our_name, stp, fname, itemizing, code);
 493 #ifdef SUPPORT_ACLS
 494                 if (preserve_acls)
 495                         free_acl(&prev_sx);
 496 #endif
 497                 if (val < 0)
 498                         continue;
 499                 if (remove_source_files == 1 && do_xfers)
 500                         send_msg_int(MSG_SUCCESS, ndx);
 501         }
 502 }
 503 #endif