detect-renamed.diff

   1 This patch adds the --detect-renamed option which makes rsync notice files
   2 that either (1) match in size & modify-time (plus the basename, if possible)
   3 or (2) match in size & checksum (when --checksum was also specified) and use
   4 each match as an alternate basis file to speed up the transfer.
   5
   6 The algorithm attempts to scan the receiving-side's files in an efficient
   7 manner.  If --delete[-before] is enabled, we'll take advantage of the
   8 pre-transfer delete pass to prepare any alternate-basis-file matches we
   9 might find.  If --delete-before is not enabled, rsync does the rename scan
  10 during the regular file-sending scan (scanning each directory right before
  11 the generator starts updating files from that dir).  In this latter mode,
  12 rsync might delay the updating of a file (if no alternate-basis match was
  13 yet found) until the full scan of the receiving side is complete, at which
  14 point any delayed files are processed.
  15
  16 I chose to hard-link the alternate-basis files into a ".~tmp~" subdir that
  17 takes advantage of rsync's pre-existing partial-dir logic.  This uses less
  18 memory than trying to keep track of the matches internally, and also allows
  19 any deletions or file-updates to occur normally without interfering with
  20 these alternate-basis discoveries.
  21
  22 To use this patch, run these commands for a successful build:
  23
  24     patch -p1 <patches/detect-renamed.diff
  25     ./configure                                 (optional if already run)
  26     make
  27
  28 TODO:
  29
  30   We need to never return a match from fattr_find() that has a basis
  31   file.  This will ensure that we don't try to give a renamed file to
  32   a file that can't use it, while missing out on giving it to a file
  33   that could use it.
  34
  35 diff --git a/compat.c b/compat.c
  36 --- a/compat.c
  37 +++ b/compat.c
  38 @@ -41,6 +41,7 @@ extern int checksum_seed;
  39  extern int basis_dir_cnt;
  40  extern int prune_empty_dirs;
  41  extern int protocol_version;
  42 +extern int detect_renamed;
  43  extern int protect_args;
  44  extern int preserve_uid;
  45  extern int preserve_gid;
  46 @@ -107,6 +108,7 @@ void set_allow_inc_recurse(void)
  47                 allow_inc_recurse = 0;
  48         else if (!am_sender
  49          && (delete_before || delete_after
  50 +         || detect_renamed
  51           || delay_updates || prune_empty_dirs))
  52                 allow_inc_recurse = 0;
  53         else if (am_server && !local_server
  54 diff --git a/flist.c b/flist.c
  55 --- a/flist.c
  56 +++ b/flist.c
  57 @@ -61,6 +61,7 @@ extern int non_perishable_cnt;
  58  extern int prune_empty_dirs;
  59  extern int copy_links;
  60  extern int copy_unsafe_links;
  61 +extern int detect_renamed;
  62  extern int protocol_version;
  63  extern int sanitize_paths;
  64  extern int munge_symlinks;
  65 @@ -118,6 +119,8 @@ static int64 tmp_dev, tmp_ino;
  66  #endif
  67  static char tmp_sum[MAX_DIGEST_LEN];
  68
  69 +struct file_list the_fattr_list;
  70 +
  71  static char empty_sum[MAX_DIGEST_LEN];
  72  static int flist_count_offset; /* for --delete --progress */
  73  static int dir_count = 0;
  74 @@ -262,6 +265,45 @@ static int is_excluded(char *fname, int is_dir, int filter_level)
  75         return 0;
  76  }
  77
  78 +static int fattr_compare(struct file_struct **file1, struct file_struct **file2)
  79 +{
  80 +       struct file_struct *f1 = *file1;
  81 +       struct file_struct *f2 = *file2;
  82 +       int64 len1 = F_LENGTH(f1), len2 = F_LENGTH(f2);
  83 +       int diff;
  84 +
  85 +       if (!f1->basename || !S_ISREG(f1->mode) || !len1) {
  86 +               if (!f2->basename || !S_ISREG(f2->mode) || !len2)
  87 +                       return 0;
  88 +               return 1;
  89 +       }
  90 +       if (!f2->basename || !S_ISREG(f2->mode) || !len2)
  91 +               return -1;
  92 +
  93 +       /* Don't use diff for values that are longer than an int. */
  94 +       if (len1 != len2)
  95 +               return len1 < len2 ? -1 : 1;
  96 +
  97 +       if (always_checksum) {
  98 +               diff = u_memcmp(F_SUM(f1), F_SUM(f2), checksum_len);
  99 +               if (diff)
 100 +                       return diff;
 101 +       } else if (f1->modtime != f2->modtime)
 102 +               return f1->modtime < f2->modtime ? -1 : 1;
 103 +
 104 +       diff = u_strcmp(f1->basename, f2->basename);
 105 +       if (diff)
 106 +               return diff;
 107 +
 108 +       if (f1->dirname == f2->dirname)
 109 +               return 0;
 110 +       if (!f1->dirname)
 111 +               return -1;
 112 +       if (!f2->dirname)
 113 +               return 1;
 114 +       return u_strcmp(f1->dirname, f2->dirname);
 115 +}
 116 +
 117  static void send_directory(int f, struct file_list *flist,
 118                            char *fbuf, int len, int flags);
 119
 120 @@ -2183,6 +2225,25 @@ struct file_list *recv_file_list(int f)
 121
 122         clean_flist(flist, relative_paths);
 123
 124 +       if (detect_renamed) {
 125 +               int j = flist->used;
 126 +               the_fattr_list.used = j;
 127 +               the_fattr_list.files = new_array(struct file_struct *, j);
 128 +               if (!the_fattr_list.files)
 129 +                       out_of_memory("recv_file_list");
 130 +               memcpy(the_fattr_list.files, flist->files,
 131 +                      j * sizeof (struct file_struct *));
 132 +               qsort(the_fattr_list.files, j,
 133 +                     sizeof the_fattr_list.files[0], (int (*)())fattr_compare);
 134 +               the_fattr_list.low = 0;
 135 +               while (j-- > 0) {
 136 +                       struct file_struct *fp = the_fattr_list.files[j];
 137 +                       if (fp->basename && S_ISREG(fp->mode) && F_LENGTH(fp))
 138 +                               break;
 139 +               }
 140 +               the_fattr_list.high = j;
 141 +       }
 142 +
 143         if (protocol_version < 30) {
 144                 /* Recv the io_error flag */
 145                 if (ignore_errors)
 146 diff --git a/generator.c b/generator.c
 147 --- a/generator.c
 148 +++ b/generator.c
 149 @@ -79,6 +79,7 @@ extern char *basis_dir[];
 150  extern int compare_dest;
 151  extern int copy_dest;
 152  extern int link_dest;
 153 +extern int detect_renamed;
 154  extern int whole_file;
 155  extern int list_only;
 156  extern int read_batch;
 157 @@ -96,6 +97,7 @@ extern char *backup_suffix;
 158  extern int backup_suffix_len;
 159  extern struct file_list *cur_flist, *first_flist, *dir_flist;
 160  extern struct filter_list_struct server_filter_list;
 161 +extern struct file_list the_fattr_list;
 162
 163  int ignore_perishable = 0;
 164  int non_perishable_cnt = 0;
 165 @@ -103,6 +105,7 @@ int maybe_ATTRS_REPORT = 0;
 166
 167  static dev_t dev_zero;
 168  static int deletion_count = 0; /* used to implement --max-delete */
 169 +static int unexplored_dirs = 1;
 170  static int deldelay_size = 0, deldelay_cnt = 0;
 171  static char *deldelay_buf = NULL;
 172  static int deldelay_fd = -1;
 173 @@ -112,7 +115,8 @@ static int need_retouch_dir_times;
 174  static int need_retouch_dir_perms;
 175  static const char *solo_file = NULL;
 176
 177 -/* For calling delete_item() and delete_dir_contents(). */
 178 +/* For calling delete_item(), delete_dir_contents(), and delete_in_dir(). */
 179 +#define DEL_NO_DELETIONS       (1<<0)
 180  #define DEL_RECURSE            (1<<1) /* recurse */
 181  #define DEL_DIR_IS_EMPTY       (1<<2) /* internal delete_FUNCTIONS use only */
 182
 183 @@ -134,11 +138,121 @@ static int is_backup_file(char *fn)
 184         return k > 0 && strcmp(fn+k, backup_suffix) == 0;
 185  }
 186
 187 +/* Search for a regular file that matches either (1) the size & modified
 188 + * time (plus the basename, if possible) or (2) the size & checksum.  If
 189 + * we find an exact match down to the dirname, return -1 because we found
 190 + * an up-to-date file in the transfer, not a renamed file. */
 191 +static int fattr_find(struct file_struct *f, char *fname)
 192 +{
 193 +       int low = the_fattr_list.low, high = the_fattr_list.high;
 194 +       int mid, ok_match = -1, good_match = -1;
 195 +       struct file_struct *fmid;
 196 +       int diff;
 197 +
 198 +       while (low <= high) {
 199 +               mid = (low + high) / 2;
 200 +               fmid = the_fattr_list.files[mid];
 201 +               if (F_LENGTH(fmid) != F_LENGTH(f)) {
 202 +                       if (F_LENGTH(fmid) < F_LENGTH(f))
 203 +                               low = mid + 1;
 204 +                       else
 205 +                               high = mid - 1;
 206 +                       continue;
 207 +               }
 208 +               if (always_checksum) {
 209 +                       /* We use the FLAG_FILE_SENT flag to indicate when we
 210 +                        * have computed the checksum for an entry. */
 211 +                       if (!(f->flags & FLAG_FILE_SENT)) {
 212 +                               if (fmid->modtime == f->modtime
 213 +                                && f_name_cmp(fmid, f) == 0)
 214 +                                       return -1; /* assume we can't help */
 215 +                               file_checksum(fname, F_SUM(f), F_LENGTH(f));
 216 +                               f->flags |= FLAG_FILE_SENT;
 217 +                       }
 218 +                       diff = u_memcmp(F_SUM(fmid), F_SUM(f), checksum_len);
 219 +                       if (diff) {
 220 +                               if (diff < 0)
 221 +                                       low = mid + 1;
 222 +                               else
 223 +                                       high = mid - 1;
 224 +                               continue;
 225 +                       }
 226 +               } else {
 227 +                       if (fmid->modtime != f->modtime) {
 228 +                               if (fmid->modtime < f->modtime)
 229 +                                       low = mid + 1;
 230 +                               else
 231 +                                       high = mid - 1;
 232 +                               continue;
 233 +                       }
 234 +               }
 235 +               ok_match = mid;
 236 +               diff = u_strcmp(fmid->basename, f->basename);
 237 +               if (diff == 0) {
 238 +                       good_match = mid;
 239 +                       if (fmid->dirname == f->dirname)
 240 +                               return -1; /* file is up-to-date */
 241 +                       if (!fmid->dirname) {
 242 +                               low = mid + 1;
 243 +                               continue;
 244 +                       }
 245 +                       if (!f->dirname) {
 246 +                               high = mid - 1;
 247 +                               continue;
 248 +                       }
 249 +                       diff = u_strcmp(fmid->dirname, f->dirname);
 250 +                       if (diff == 0)
 251 +                               return -1; /* file is up-to-date */
 252 +               }
 253 +               if (diff < 0)
 254 +                       low = mid + 1;
 255 +               else
 256 +                       high = mid - 1;
 257 +       }
 258 +
 259 +       return good_match >= 0 ? good_match : ok_match;
 260 +}
 261 +
 262 +static void look_for_rename(struct file_struct *file, char *fname)
 263 +{
 264 +       struct file_struct *fp;
 265 +       char *partialptr, *fn;
 266 +       STRUCT_STAT st;
 267 +       int ndx;
 268 +
 269 +       if (!partial_dir || (ndx = fattr_find(file, fname)) < 0)
 270 +               return;
 271 +
 272 +       fp = the_fattr_list.files[ndx];
 273 +       fn = f_name(fp, NULL);
 274 +       /* We don't provide an alternate-basis file if there is a basis file. */
 275 +       if (link_stat(fn, &st, 0) == 0)
 276 +               return;
 277 +
 278 +       if (!dry_run) {
 279 +               if ((partialptr = partial_dir_fname(fn)) == NULL
 280 +                || !handle_partial_dir(partialptr, PDIR_CREATE))
 281 +                       return;
 282 +               /* We only use the file if we can hard-link it into our tmp dir. */
 283 +               if (link(fname, partialptr) != 0) {
 284 +                       if (errno != EEXIST)
 285 +                               handle_partial_dir(partialptr, PDIR_DELETE);
 286 +                       return;
 287 +               }
 288 +       }
 289 +
 290 +       /* I think this falls into the -vv category with "%s is uptodate", etc. */
 291 +       if (verbose > 1)
 292 +               rprintf(FINFO, "found renamed: %s => %s\n", fname, fn);
 293 +}
 294 +
 295  /* Delete a file or directory.  If DEL_RECURSE is set in the flags, this will
 296   * delete recursively.
 297   *
 298   * Note that fbuf must point to a MAXPATHLEN buffer if the mode indicates it's
 299   * a directory! (The buffer is used for recursion, but returned unchanged.)
 300 + *
 301 + * Also note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 302   */
 303  static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 304  {
 305 @@ -160,6 +274,8 @@ static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 306                         goto check_ret;
 307                 /* OK: try to delete the directory. */
 308         }
 309 +       if (flags & DEL_NO_DELETIONS)
 310 +               return DR_SUCCESS;
 311
 312         if (!replace && max_delete >= 0 && ++deletion_count > max_delete)
 313                 return DR_AT_LIMIT;
 314 @@ -206,6 +322,8 @@ static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 315   * its contents, otherwise just checks for content.  Returns DR_SUCCESS or
 316   * DR_NOT_EMPTY.  Note that fname must point to a MAXPATHLEN buffer!  (The
 317   * buffer is used for recursion, but returned unchanged.)
 318 + *
 319 + * Note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 320   */
 321  static enum delret delete_dir_contents(char *fname, int flags)
 322  {
 323 @@ -225,7 +343,9 @@ static enum delret delete_dir_contents(char *fname, int flags)
 324         save_filters = push_local_filters(fname, dlen);
 325
 326         non_perishable_cnt = 0;
 327 +       file_extra_cnt += SUM_EXTRA_CNT;
 328         dirlist = get_dirlist(fname, dlen, 0);
 329 +       file_extra_cnt -= SUM_EXTRA_CNT;
 330         ret = non_perishable_cnt ? DR_NOT_EMPTY : DR_SUCCESS;
 331
 332         if (!dirlist->used)
 333 @@ -262,6 +382,8 @@ static enum delret delete_dir_contents(char *fname, int flags)
 334                 if (S_ISDIR(fp->mode)
 335                  && delete_dir_contents(fname, flags | DEL_RECURSE) != DR_SUCCESS)
 336                         ret = DR_NOT_EMPTY;
 337 +               if (detect_renamed && S_ISREG(fp->mode))
 338 +                       look_for_rename(fp, fname);
 339                 if (delete_item(fname, fp->mode, NULL, flags) != DR_SUCCESS)
 340                         ret = DR_NOT_EMPTY;
 341         }
 342 @@ -414,13 +536,18 @@ static void do_delayed_deletions(char *delbuf)
 343   * all the --delete-WHEN options.  Note that the fbuf pointer must point to a
 344   * MAXPATHLEN buffer with the name of the directory in it (the functions we
 345   * call will append names onto the end, but the old dir value will be restored
 346 - * on exit). */
 347 -static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
 348 + * on exit).
 349 + *
 350 + * Note:  --detect-rename may use this routine with DEL_NO_DELETIONS set!
 351 + */
 352 +static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev,
 353 +                         int flags)
 354  {
 355         static int already_warned = 0;
 356         struct file_list *dirlist;
 357 -       char delbuf[MAXPATHLEN];
 358 -       int dlen, i;
 359 +       char *p, delbuf[MAXPATHLEN];
 360 +       unsigned remainder;
 361 +       int dlen, i, restore_dot = 0;
 362
 363         if (!fbuf) {
 364                 change_local_filter_dir(NULL, 0, 0);
 365 @@ -430,21 +557,28 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
 366         if (verbose > 2)
 367                 rprintf(FINFO, "delete_in_dir(%s)\n", fbuf);
 368
 369 +       flags |= DEL_RECURSE;
 370 +
 371         if (allowed_lull)
 372                 maybe_send_keepalive();
 373
 374         if (io_error && !ignore_errors) {
 375 -               if (already_warned)
 376 +               if (!already_warned) {
 377 +                       rprintf(FINFO,
 378 +                           "IO error encountered -- skipping file deletion\n");
 379 +                       already_warned = 1;
 380 +               }
 381 +               if (!detect_renamed)
 382                         return;
 383 -               rprintf(FINFO,
 384 -                       "IO error encountered -- skipping file deletion\n");
 385 -               already_warned = 1;
 386 -               return;
 387 +               flags |= DEL_NO_DELETIONS;
 388         }
 389
 390         dlen = strlen(fbuf);
 391         change_local_filter_dir(fbuf, dlen, F_DEPTH(file));
 392
 393 +       if (detect_renamed)
 394 +               unexplored_dirs--;
 395 +
 396         if (one_file_system) {
 397                 if (file->flags & FLAG_TOP_DIR)
 398                         filesystem_dev = *fs_dev;
 399 @@ -454,6 +588,14 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
 400
 401         dirlist = get_dirlist(fbuf, dlen, 0);
 402
 403 +       p = fbuf + dlen;
 404 +       if (dlen == 1 && *fbuf == '.') {
 405 +               restore_dot = 1;
 406 +               p = fbuf;
 407 +       } else if (dlen != 1 || *fbuf != '/')
 408 +               *p++ = '/';
 409 +       remainder = MAXPATHLEN - (p - fbuf);
 410 +
 411         /* If an item in dirlist is not found in flist, delete it
 412          * from the filesystem. */
 413         for (i = dirlist->used; i--; ) {
 414 @@ -466,16 +608,25 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
 415                                         f_name(fp, NULL));
 416                         continue;
 417                 }
 418 +               if (detect_renamed && S_ISREG(fp->mode)) {
 419 +                       strlcpy(p, fp->basename, remainder);
 420 +                       look_for_rename(fp, fbuf);
 421 +               }
 422                 if (flist_find(cur_flist, fp) < 0) {
 423                         f_name(fp, delbuf);
 424 -                       if (delete_during == 2) {
 425 +                       if (delete_during == 2 && !(flags & DEL_NO_DELETIONS)) {
 426                                 if (!remember_delete(fp, delbuf))
 427                                         break;
 428                         } else
 429 -                               delete_item(delbuf, fp->mode, NULL, DEL_RECURSE);
 430 -               }
 431 +                               delete_item(delbuf, fp->mode, NULL, flags);
 432 +               } else if (detect_renamed && S_ISDIR(fp->mode))
 433 +                       unexplored_dirs++;
 434         }
 435
 436 +       if (restore_dot)
 437 +               fbuf[0] = '.';
 438 +       fbuf[dlen] = '\0';
 439 +
 440         flist_free(dirlist);
 441  }
 442
 443 @@ -505,9 +656,9 @@ static void do_delete_pass(void)
 444                  || !S_ISDIR(st.st_mode))
 445                         continue;
 446
 447 -               delete_in_dir(fbuf, file, &st.st_dev);
 448 +               delete_in_dir(fbuf, file, &st.st_dev, 0);
 449         }
 450 -       delete_in_dir(NULL, NULL, &dev_zero);
 451 +       delete_in_dir(NULL, NULL, &dev_zero, 0);
 452
 453         if (do_progress && !am_server)
 454                 rprintf(FINFO, "                    \r");
 455 @@ -1139,6 +1290,7 @@ static void list_file_entry(struct file_struct *f)
 456         }
 457  }
 458
 459 +static struct bitbag *delayed_bits = NULL;
 460  static int phase = 0;
 461  static int dflt_perms;
 462
 463 @@ -1383,8 +1535,12 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
 464                         }
 465                 }
 466                 else if (delete_during && f_out != -1 && !phase && dry_run < 2
 467 -                   && (file->flags & FLAG_CONTENT_DIR))
 468 -                       delete_in_dir(fname, file, &real_sx.st.st_dev);
 469 +                   && (file->flags & FLAG_CONTENT_DIR)) {
 470 +                       if (detect_renamed && real_ret != 0)
 471 +                               unexplored_dirs++;
 472 +                       delete_in_dir(fname, file, &real_sx.st.st_dev,
 473 +                                     delete_during < 0 ? DEL_NO_DELETIONS : 0);
 474 +               }
 475                 goto cleanup;
 476         }
 477
 478 @@ -1662,8 +1818,14 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
 479                         goto cleanup;
 480                 }
 481  #endif
 482 -               if (stat_errno == ENOENT)
 483 +               if (stat_errno == ENOENT) {
 484 +                       if (detect_renamed && unexplored_dirs > 0
 485 +                        && F_LENGTH(file)) {
 486 +                               bitbag_set_bit(delayed_bits, ndx);
 487 +                               return;
 488 +                       }
 489                         goto notify_others;
 490 +               }
 491                 rsyserr(FERROR_XFER, stat_errno, "recv_generator: failed to stat %s",
 492                         full_fname(fname));
 493                 goto cleanup;
 494 @@ -2000,6 +2162,12 @@ void generate_files(int f_out, const char *local_name)
 495         if (verbose > 2)
 496                 rprintf(FINFO, "generator starting pid=%ld\n", (long)getpid());
 497
 498 +       if (detect_renamed) {
 499 +               delayed_bits = bitbag_create(cur_flist->used);
 500 +               if (!delete_before && !delete_during)
 501 +                       delete_during = -1;
 502 +       }
 503 +
 504         if (delete_before && !solo_file && cur_flist->used > 0)
 505                 do_delete_pass();
 506         if (delete_during == 2) {
 507 @@ -2010,7 +2178,7 @@ void generate_files(int f_out, const char *local_name)
 508         }
 509         do_progress = 0;
 510
 511 -       if (append_mode > 0 || whole_file < 0)
 512 +       if (append_mode > 0 || detect_renamed || whole_file < 0)
 513                 whole_file = 0;
 514         if (verbose >= 2) {
 515                 rprintf(FINFO, "delta-transmission %s\n",
 516 @@ -2048,7 +2216,7 @@ void generate_files(int f_out, const char *local_name)
 517                                                 dirdev = MAKEDEV(DEV_MAJOR(devp), DEV_MINOR(devp));
 518                                         } else
 519                                                 dirdev = MAKEDEV(0, 0);
 520 -                                       delete_in_dir(f_name(fp, fbuf), fp, &dirdev);
 521 +                                       delete_in_dir(f_name(fp, fbuf), fp, &dirdev, 0);
 522                                 }
 523                         }
 524                 }
 525 @@ -2091,7 +2259,21 @@ void generate_files(int f_out, const char *local_name)
 526         } while ((cur_flist = cur_flist->next) != NULL);
 527
 528         if (delete_during)
 529 -               delete_in_dir(NULL, NULL, &dev_zero);
 530 +               delete_in_dir(NULL, NULL, &dev_zero, 0);
 531 +       if (detect_renamed) {
 532 +               if (delete_during < 0)
 533 +                       delete_during = 0;
 534 +               detect_renamed = 0;
 535 +
 536 +               for (i = -1; (i = bitbag_next_bit(delayed_bits, i)) >= 0; ) {
 537 +                       struct file_struct *file = cur_flist->files[i];
 538 +                       if (local_name)
 539 +                               strlcpy(fbuf, local_name, sizeof fbuf);
 540 +                       else
 541 +                               f_name(file, fbuf);
 542 +                       recv_generator(fbuf, file, i, itemizing, code, f_out);
 543 +               }
 544 +       }
 545         phase++;
 546         if (verbose > 2)
 547                 rprintf(FINFO, "generate_files phase=%d\n", phase);
 548 diff --git a/options.c b/options.c
 549 --- a/options.c
 550 +++ b/options.c
 551 @@ -80,6 +80,7 @@ int am_generator = 0;
 552  int am_starting_up = 1;
 553  int relative_paths = -1;
 554  int implied_dirs = 1;
 555 +int detect_renamed = 0;
 556  int numeric_ids = 0;
 557  int allow_8bit_chars = 0;
 558  int force_delete = 0;
 559 @@ -385,6 +386,7 @@ void usage(enum logcode F)
 560    rprintf(F,"     --modify-window=NUM     compare mod-times with reduced accuracy\n");
 561    rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
 562    rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
 563 +  rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
 564    rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
 565    rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
 566    rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
 567 @@ -563,6 +565,7 @@ static struct poptOption long_options[] = {
 568    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
 569    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
 570    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
 571 +  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
 572    {"fuzzy",           'y', POPT_ARG_NONE,   &fuzzy_basis, 0, 0, 0 },
 573    {"compress",        'z', POPT_ARG_NONE,   0, 'z', 0, 0 },
 574    {"no-compress",      0,  POPT_ARG_VAL,    &do_compression, 0, 0, 0 },
 575 @@ -1535,7 +1538,7 @@ int parse_arguments(int *argc_p, const char ***argv_p, int frommain)
 576                 inplace = 1;
 577         }
 578
 579 -       if (delay_updates && !partial_dir)
 580 +       if ((delay_updates || detect_renamed) && !partial_dir)
 581                 partial_dir = tmp_partialdir;
 582
 583         if (inplace) {
 584 @@ -1544,6 +1547,7 @@ int parse_arguments(int *argc_p, const char ***argv_p, int frommain)
 585                         snprintf(err_buf, sizeof err_buf,
 586                                  "--%s cannot be used with --%s\n",
 587                                  append_mode ? "append" : "inplace",
 588 +                                detect_renamed ? "detect-renamed" :
 589                                  delay_updates ? "delay-updates" : "partial-dir");
 590                         return 0;
 591                 }
 592 @@ -1888,6 +1892,8 @@ void server_options(char **args, int *argc_p)
 593                         args[ac++] = "--super";
 594                 if (size_only)
 595                         args[ac++] = "--size-only";
 596 +               if (detect_renamed)
 597 +                       args[ac++] = "--detect-renamed";
 598         } else {
 599                 if (skip_compress) {
 600                         if (asprintf(&arg, "--skip-compress=%s", skip_compress) < 0)
 601 diff --git a/rsync.yo b/rsync.yo
 602 --- a/rsync.yo
 603 +++ b/rsync.yo
 604 @@ -385,6 +385,7 @@ to the detailed description below for a complete description.  verb(
 605       --modify-window=NUM     compare mod-times with reduced accuracy
 606   -T, --temp-dir=DIR          create temporary files in directory DIR
 607   -y, --fuzzy                 find similar file for basis if no dest file
 608 +     --detect-renamed        try to find renamed files to speed the xfer
 609       --compare-dest=DIR      also compare received files relative to DIR
 610       --copy-dest=DIR         ... and include copies of unchanged files
 611       --link-dest=DIR         hardlink to files in DIR when unchanged
 612 @@ -1451,6 +1452,21 @@ Note that the use of the bf(--delete) option might get rid of any potential
 613  fuzzy-match files, so either use bf(--delete-after) or specify some
 614  filename exclusions if you need to prevent this.
 615
 616 +dit(bf(--detect-renamed)) With this option, for each new source file
 617 +(call it em(src/S)), rsync looks for a file em(dest/D) anywhere in the
 618 +destination that passes the quick check with em(src/S).  If such a em(dest/D)
 619 +is found, rsync uses it as an alternate basis for transferring em(S).  The
 620 +idea is that if em(src/S) was renamed from em(src/D) (as opposed to em(src/S)
 621 +passing the quick check with em(dest/D) by coincidence), the delta-transfer
 622 +algorithm will find that all the data matches between em(src/S) and em(dest/D),
 623 +and the transfer will be really fast.
 624 +
 625 +By default, alternate-basis files are hard-linked into a directory named
 626 +".~tmp~" in each file's destination directory, but if you've specified
 627 +the bf(--partial-dir) option, that directory will be used instead.  These
 628 +potential alternate-basis files will be removed as the transfer progresses.
 629 +This option conflicts with bf(--inplace) and bf(--append).
 630 +
 631  dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
 632  the destination machine as an additional hierarchy to compare destination
 633  files against doing transfers (if the files are missing in the destination
 634 diff --git a/util.c b/util.c
 635 --- a/util.c
 636 +++ b/util.c
 637 @@ -1019,6 +1019,32 @@ int handle_partial_dir(const char *fname, int create)
 638         return 1;
 639  }
 640
 641 +/* We need to supply our own strcmp function for file list comparisons
 642 + * to ensure that signed/unsigned usage is consistent between machines. */
 643 +int u_strcmp(const char *p1, const char *p2)
 644 +{
 645 +        for ( ; *p1; p1++, p2++) {
 646 +               if (*p1 != *p2)
 647 +                       break;
 648 +       }
 649 +
 650 +       return (int)*(uchar*)p1 - (int)*(uchar*)p2;
 651 +}
 652 +
 653 +/* We need a memcmp function compares unsigned-byte values. */
 654 +int u_memcmp(const void *p1, const void *p2, size_t len)
 655 +{
 656 +       const uchar *u1 = p1;
 657 +       const uchar *u2 = p2;
 658 +
 659 +       while (len--) {
 660 +               if (*u1 != *u2)
 661 +                       return (int)*u1 - (int)*u2;
 662 +       }
 663 +
 664 +       return 0;
 665 +}
 666 +
 667  /**
 668   * Determine if a symlink points outside the current directory tree.
 669   * This is considered "unsafe" because e.g. when mirroring somebody