detect-renamed-lax.diff

   1 This patch adds the options --detect-renamed-lax and --detect-moved.
   2 These modify the --detect-renamed algorithm to adopt a matching file
   3 without verifying that the content is as expected.  The former blindly
   4 accepts a file that matches in size and modified time.  The latter
   5 requires that the filename also match (ignoring any renamed files).
   6
   7 This patch is EXPERIMENTAL, though it did work correctly in my light
   8 testing.
   9
  10 To use this patch, run these commands for a successful build:
  11
  12     patch -p1 <patches/detect-renamed.diff
  13     patch -p1 <patches/detect-renamed-lax.diff
  14     ./configure                                 (optional if already run)
  15     make
  16
  17 FIXME: If a run with --detect-renamed-lax stages a different-basename
  18 destination file and then gets interrupted, a subsequent run that
  19 switches to --detect-moved blindly accepts the staged file.
  20
  21 -- Matt McCutchen <hashproduct+rsync@gmail.com>
  22
  23 based-on: patch/master/detect-renamed
  24 diff --git a/generator.c b/generator.c
  25 --- a/generator.c
  26 +++ b/generator.c
  27 @@ -467,7 +467,9 @@ static int fattr_find(struct file_struct *f, char *fname)
  28                                 continue;
  29                         }
  30                 }
  31 -               ok_match = mid;
  32 +               /* --detect-moved doesn't allow non-basename matches */
  33 +               if (detect_renamed != 3)
  34 +                       ok_match = mid;
  35                 diff = u_strcmp(fmid->basename, f->basename);
  36                 if (diff == 0) {
  37                         good_match = mid;
  38 @@ -1991,6 +1993,21 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
  39                 fnamecmp = partialptr;
  40                 fnamecmp_type = FNAMECMP_PARTIAL_DIR;
  41                 statret = 0;
  42 +               if (detect_renamed > 1 && quick_check_ok(FT_REG, fnamecmp, file, &sx.st)) {
  43 +                       /* Adopt the partial file. */
  44 +                       finish_transfer(fname, fnamecmp, NULL, NULL, file, 1, 1);
  45 +                       handle_partial_dir(partialptr, PDIR_DELETE);
  46 +                       if (itemizing)
  47 +                               itemize(fnamecmp, file, ndx, -1, &sx,
  48 +                                               ITEM_LOCAL_CHANGE, fnamecmp_type, NULL);
  49 +#ifdef SUPPORT_HARD_LINKS
  50 +                       if (preserve_hard_links && F_IS_HLINKED(file))
  51 +                               finish_hard_link(file, fname, ndx, &sx.st, itemizing, code, -1);
  52 +#endif
  53 +                       if (remove_source_files == 1)
  54 +                               goto return_with_success;
  55 +                       goto cleanup;
  56 +               }
  57         }
  58
  59         if (!do_xfers)
  60 diff --git a/options.c b/options.c
  61 --- a/options.c
  62 +++ b/options.c
  63 @@ -744,7 +744,9 @@ static struct poptOption long_options[] = {
  64    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
  65    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
  66    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
  67 -  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
  68 +  {"detect-renamed",   0,  POPT_ARG_VAL,    &detect_renamed, 1, 0, 0 },
  69 +  {"detect-renamed-lax",0, POPT_ARG_VAL,    &detect_renamed, 2, 0, 0 },
  70 +  {"detect-moved",     0,  POPT_ARG_VAL,    &detect_renamed, 3, 0, 0 },
  71    {"fuzzy",           'y', POPT_ARG_NONE,   0, 'y', 0, 0 },
  72    {"no-fuzzy",         0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
  73    {"no-y",             0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
  74 @@ -2838,8 +2840,14 @@ void server_options(char **args, int *argc_p)
  75                         args[ac++] = "--super";
  76                 if (size_only)
  77                         args[ac++] = "--size-only";
  78 -               if (detect_renamed)
  79 -                       args[ac++] = "--detect-renamed";
  80 +               if (detect_renamed) {
  81 +                       if (detect_renamed == 1)
  82 +                               args[ac++] = "--detect-renamed";
  83 +                       else if (detect_renamed == 2)
  84 +                               args[ac++] = "--detect-renamed-lax";
  85 +                       else
  86 +                               args[ac++] = "--detect-moved";
  87 +               }
  88                 if (do_stats)
  89                         args[ac++] = "--stats";
  90         } else {
  91 diff --git a/rsync.1.md b/rsync.1.md
  92 --- a/rsync.1.md
  93 +++ b/rsync.1.md
  94 @@ -508,6 +508,8 @@ has its own detailed description later in this manpage.
  95  --temp-dir=DIR, -T       create temporary files in directory DIR
  96  --fuzzy, -y              find similar file for basis if no dest file
  97  --detect-renamed         try to find renamed files to speed the xfer
  98 +--detect-renamed-lax     ...& assume identical to src files (risky!)
  99 +--detect-moved           ... only if basenames match (less risky)
 100  --compare-dest=DIR       also compare destination files relative to DIR
 101  --copy-dest=DIR          ... and include copies of unchanged files
 102  --link-dest=DIR          hardlink to files in DIR when unchanged
 103 @@ -2652,6 +2654,20 @@ expand it.
 104      otential alternate-basis files will be removed as the transfer progresses.
 105      This option conflicts with [`--inplace`](#opt) and [`--append`](#opt).
 106
 107 +0.  `--detect-renamed-lax`
 108 +
 109 +    This version of [`--detect-renamed`](#opt) makes rsync hard-link `dest/D`
 110 +    to `dest/S` without verifying that `src/S` and `dest/S` have the same data.
 111 +    This poses a significant risk of corrupting the destination by representing
 112 +    a new source file by an unrelated destination file that coincidentally
 113 +    passes the quick check with the source file.  Use this option only if you
 114 +    accept the risk and disk I/O is a bottleneck.
 115 +
 116 +0.  `--detect-moved`
 117 +
 118 +    A less risky variant of [`--detect-renamed-lax`](#opt) that only uses a
 119 +    destination file that has the same basename as the new source file.
 120 +
 121  0.  `--compare-dest=DIR`
 122
 123      This option instructs rsync to use _DIR_ on the destination machine as an