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 diff --git a/generator.c b/generator.c
  24 --- a/generator.c
  25 +++ b/generator.c
  26 @@ -200,7 +200,9 @@ static int fattr_find(struct file_struct *f, char *fname)
  27                                 continue;
  28                         }
  29                 }
  30 -               ok_match = mid;
  31 +               /* --detect-moved doesn't allow non-basename matches */
  32 +               if (detect_renamed != 3)
  33 +                       ok_match = mid;
  34                 diff = u_strcmp(fmid->basename, f->basename);
  35                 if (diff == 0) {
  36                         good_match = mid;
  37 @@ -1967,6 +1969,21 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
  38                 fnamecmp = partialptr;
  39                 fnamecmp_type = FNAMECMP_PARTIAL_DIR;
  40                 statret = 0;
  41 +               if (detect_renamed > 1 && unchanged_file(fnamecmp, file, &sx.st)) {
  42 +                       /* Adopt the partial file. */
  43 +                       finish_transfer(fname, fnamecmp, NULL, NULL, file, 1, 1);
  44 +                       handle_partial_dir(partialptr, PDIR_DELETE);
  45 +                       if (itemizing)
  46 +                               itemize(fnamecmp, file, ndx, -1, &sx,
  47 +                                               ITEM_LOCAL_CHANGE, fnamecmp_type, NULL);
  48 +#ifdef SUPPORT_HARD_LINKS
  49 +                       if (preserve_hard_links && F_IS_HLINKED(file))
  50 +                               finish_hard_link(file, fname, ndx, &sx.st, itemizing, code, -1);
  51 +#endif
  52 +                       if (remove_source_files == 1)
  53 +                               goto return_with_success;
  54 +                       goto cleanup;
  55 +               }
  56         }
  57
  58         if (!do_xfers)
  59 diff --git a/options.c b/options.c
  60 --- a/options.c
  61 +++ b/options.c
  62 @@ -393,6 +393,8 @@ void usage(enum logcode F)
  63    rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
  64    rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
  65    rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
  66 +  rprintf(F,"     --detect-renamed-lax    ... and assume identical to source files (risky!)\n");
  67 +  rprintf(F,"     --detect-moved          ... only if basenames match (less risky)\n");
  68    rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
  69    rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
  70    rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
  71 @@ -580,7 +582,9 @@ static struct poptOption long_options[] = {
  72    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
  73    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
  74    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
  75 -  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
  76 +  {"detect-renamed",   0,  POPT_ARG_VAL,    &detect_renamed, 1, 0, 0 },
  77 +  {"detect-renamed-lax",0, POPT_ARG_VAL,    &detect_renamed, 2, 0, 0 },
  78 +  {"detect-moved",     0,  POPT_ARG_VAL,    &detect_renamed, 3, 0, 0 },
  79    {"fuzzy",           'y', POPT_ARG_VAL,    &fuzzy_basis, 1, 0, 0 },
  80    {"no-fuzzy",         0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
  81    {"no-y",             0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
  82 @@ -1949,8 +1953,14 @@ void server_options(char **args, int *argc_p)
  83                         args[ac++] = "--super";
  84                 if (size_only)
  85                         args[ac++] = "--size-only";
  86 -               if (detect_renamed)
  87 -                       args[ac++] = "--detect-renamed";
  88 +               if (detect_renamed) {
  89 +                       if (detect_renamed == 1)
  90 +                               args[ac++] = "--detect-renamed";
  91 +                       else if (detect_renamed == 2)
  92 +                               args[ac++] = "--detect-renamed-lax";
  93 +                       else
  94 +                               args[ac++] = "--detect-moved";
  95 +               }
  96         } else {
  97                 if (skip_compress) {
  98                         if (asprintf(&arg, "--skip-compress=%s", skip_compress) < 0)
  99 diff --git a/rsync.yo b/rsync.yo
 100 --- a/rsync.yo
 101 +++ b/rsync.yo
 102 @@ -386,6 +386,8 @@ to the detailed description below for a complete description.  verb(
 103   -T, --temp-dir=DIR          create temporary files in directory DIR
 104   -y, --fuzzy                 find similar file for basis if no dest file
 105       --detect-renamed        try to find renamed files to speed the xfer
 106 +     --detect-renamed-lax    ...& assume identical to src files (risky!)
 107 +     --detect-moved          ... only if basenames match (less risky)
 108       --compare-dest=DIR      also compare received files relative to DIR
 109       --copy-dest=DIR         ... and include copies of unchanged files
 110       --link-dest=DIR         hardlink to files in DIR when unchanged
 111 @@ -1490,6 +1492,17 @@ the bf(--partial-dir) option, that directory will be used instead.  These
 112  potential alternate-basis files will be removed as the transfer progresses.
 113  This option conflicts with bf(--inplace) and bf(--append).
 114
 115 +dit(bf(--detect-renamed-lax)) This version of bf(--detect-renamed)
 116 +makes rsync hard-link em(dest/D) to em(dest/S) without verifying that
 117 +em(src/S) and em(dest/S) have the same data.  This poses a significant risk
 118 +of corrupting the destination by representing a new source file by an
 119 +unrelated destination file that coincidentally passes the quick check with the
 120 +source file.  Use this option only if you accept the risk and disk I/O is a
 121 +bottleneck.
 122 +
 123 +dit(bf(--detect-moved)) A less risky variant of bf(--detect-renamed-lax) that
 124 +only uses a destination file that has the same basename as the new source file.
 125 +
 126  dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
 127  the destination machine as an additional hierarchy to compare destination
 128  files against doing transfers (if the files are missing in the destination