Attachment 9840 Details for Bug 10522 – revised patch detect-renamed.diff

[patch] revised patch detect-renamed.diff

detect-renamed.diff (text/plain), 30.16 KB, created by J.R. on 2014-04-10 02:50:13 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: J.R.

Created: 2014-04-10 02:50:13 UTC

Size: 30.16 KB

patch

obsolete

>This patch adds the --detect-renamed option which makes rsync notice files
>that either (1) match in size & modify-time (plus the basename, if possible)
>or (2) match in size & checksum (when --checksum was also specified) and use
>each match as an alternate basis file to speed up the transfer.
>
>The algorithm attempts to scan the receiving-side's files in an efficient
>manner.  If --delete[-before] is enabled, we'll take advantage of the
>pre-transfer delete pass to prepare any alternate-basis-file matches we
>might find.  If --delete-before is not enabled, rsync does the rename scan
>during the regular file-sending scan (scanning each directory right before
>the generator starts updating files from that dir).  In this latter mode,
>rsync might delay the updating of a file (if no alternate-basis match was
>yet found) until the full scan of the receiving side is complete, at which
>point any delayed files are processed.
>
>I chose to hard-link the alternate-basis files into a ".~tmp~" subdir that
>takes advantage of rsync's pre-existing partial-dir logic.  This uses less
>memory than trying to keep track of the matches internally, and also allows
>any deletions or file-updates to occur normally without interfering with
>these alternate-basis discoveries.
>
>To use this patch, run these commands for a successful build:
>
>    patch -p1 <patches/detect-renamed.diff
>    ./configure                                 (optional if already run)
>    make
>
>TODO:
>
>  The routine that makes missing directories for files that get renamed
>  down into a new sub-hierarchy doesn't properly handle the case where some
>  path elements might exist but not be a dir yet.  We need to either change
>  our stash-ahead algorithm (to not require unknown path elements) or we
>  need to create a better path-making routine.
>
>  We need to never return a match from fattr_find() that has a basis
>  file.  This will ensure that we don't try to give a renamed file to
>  a file that can't use it, while missing out on giving it to a file
>  that could use it.
>
>based-on: bc58313bf7f42f0993f7279299e8137e0f14f01c
>diff --git a/compat.c b/compat.c
>--- a/compat.c
>+++ b/compat.c
>@@ -43,6 +43,7 @@ extern int checksum_seed;
> extern int basis_dir_cnt;
> extern int prune_empty_dirs;
> extern int protocol_version;
>+extern int detect_renamed;
> extern int protect_args;
> extern int preserve_uid;
> extern int preserve_gid;
>@@ -123,6 +124,7 @@ void set_allow_inc_recurse(void)
> 		allow_inc_recurse = 0;
> 	else if (!am_sender
> 	 && (delete_before || delete_after
>+	  || detect_renamed
> 	  || delay_updates || prune_empty_dirs))
> 		allow_inc_recurse = 0;
> 	else if (am_server && !local_server
>diff --git a/delete.c b/delete.c
>--- a/delete.c
>+++ b/delete.c
>@@ -25,6 +25,7 @@
> extern int am_root;
> extern int make_backups;
> extern int max_delete;
>+extern int detect_renamed;
> extern char *backup_dir;
> extern char *backup_suffix;
> extern int backup_suffix_len;
>@@ -44,6 +45,8 @@ static inline int is_backup_file(char *fn)
>  * its contents, otherwise just checks for content.  Returns DR_SUCCESS or
>  * DR_NOT_EMPTY.  Note that fname must point to a MAXPATHLEN buffer!  (The
>  * buffer is used for recursion, but returned unchanged.)
>+ *
>+ * Note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
>  */
> static enum delret delete_dir_contents(char *fname, uint16 flags)
> {
>@@ -63,7 +66,9 @@ static enum delret delete_dir_contents(char *fname, uint16 flags)
> 	save_filters = push_local_filters(fname, dlen);
> 
> 	non_perishable_cnt = 0;
>+	file_extra_cnt += SUM_EXTRA_CNT;
> 	dirlist = get_dirlist(fname, dlen, 0);
>+	file_extra_cnt -= SUM_EXTRA_CNT;
> 	ret = non_perishable_cnt ? DR_NOT_EMPTY : DR_SUCCESS;
> 
> 	if (!dirlist->used)
>@@ -103,7 +108,8 @@ static enum delret delete_dir_contents(char *fname, uint16 flags)
> 		if (S_ISDIR(fp->mode)) {
> 			if (delete_dir_contents(fname, flags | DEL_RECURSE) != DR_SUCCESS)
> 				ret = DR_NOT_EMPTY;
>-		}
>+		} else if (detect_renamed && S_ISREG(fp->mode))
>+			look_for_rename(fp, fname);
> 		if (delete_item(fname, fp->mode, flags) != DR_SUCCESS)
> 			ret = DR_NOT_EMPTY;
> 	}
>@@ -126,6 +132,8 @@ static enum delret delete_dir_contents(char *fname, uint16 flags)
>  *
>  * Note that fbuf must point to a MAXPATHLEN buffer if the mode indicates it's
>  * a directory! (The buffer is used for recursion, but returned unchanged.)
>+ *
>+ * Also note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
>  */
> enum delret delete_item(char *fbuf, uint16 mode, uint16 flags)
> {
>@@ -153,6 +161,9 @@
> 		/* OK: try to delete the directory. */
> 	}
> 
>+	if (flags & DEL_NO_DELETIONS)
>+		return DR_SUCCESS;
>+
> 	if (!(flags & DEL_MAKE_ROOM) && max_delete >= 0 && stats.deleted_files >= max_delete) {
> 		skipped_deletes++;
> 		return DR_AT_LIMIT;
>diff --git a/flist.c b/flist.c
>--- a/flist.c
>+++ b/flist.c
>@@ -60,6 +60,7 @@ extern int non_perishable_cnt;
> extern int prune_empty_dirs;
> extern int copy_links;
> extern int copy_unsafe_links;
>+extern int detect_renamed;
> extern int protocol_version;
> extern int sanitize_paths;
> extern int munge_symlinks;
>@@ -125,6 +126,8 @@ static int64 tmp_dev = -1, tmp_ino;
> #endif
> static char tmp_sum[MAX_DIGEST_LEN];
> 
>+struct file_list the_fattr_list;
>+
> static char empty_sum[MAX_DIGEST_LEN];
> static int flist_count_offset; /* for --delete --progress */
> 
>@@ -292,6 +295,45 @@ static int is_excluded(const char *fname, int is_dir, int filter_level)
> 	return 0;
> }
> 
>+static int fattr_compare(struct file_struct **file1, struct file_struct **file2)
>+{
>+	struct file_struct *f1 = *file1;
>+	struct file_struct *f2 = *file2;
>+	int64 len1 = F_LENGTH(f1), len2 = F_LENGTH(f2);
>+	int diff;
>+
>+	if (!f1->basename || !S_ISREG(f1->mode) || !len1) {
>+		if (!f2->basename || !S_ISREG(f2->mode) || !len2)
>+			return 0;
>+		return 1;
>+	}
>+	if (!f2->basename || !S_ISREG(f2->mode) || !len2)
>+		return -1;
>+
>+	/* Don't use diff for values that are longer than an int. */
>+	if (len1 != len2)
>+		return len1 < len2 ? -1 : 1;
>+
>+	if (always_checksum) {
>+		diff = u_memcmp(F_SUM(f1), F_SUM(f2), checksum_len);
>+		if (diff)
>+			return diff;
>+	} else if (f1->modtime != f2->modtime)
>+		return f1->modtime < f2->modtime ? -1 : 1;
>+
>+	diff = u_strcmp(f1->basename, f2->basename);
>+	if (diff)
>+		return diff;
>+
>+	if (f1->dirname == f2->dirname)
>+		return 0;
>+	if (!f1->dirname)
>+		return -1;
>+	if (!f2->dirname)
>+		return 1;
>+	return u_strcmp(f1->dirname, f2->dirname);
>+}
>+
> static void send_directory(int f, struct file_list *flist,
> 			   char *fbuf, int len, int flags);
> 
>@@ -2555,6 +2597,25 @@ struct file_list *recv_file_list(int f)
> 
> 	flist_sort_and_clean(flist, relative_paths);
> 
>+	if (detect_renamed) {
>+		int j = flist->used;
>+		the_fattr_list.used = j;
>+		the_fattr_list.files = new_array(struct file_struct *, j);
>+		if (!the_fattr_list.files)
>+			out_of_memory("recv_file_list");
>+		memcpy(the_fattr_list.files, flist->files,
>+		       j * sizeof (struct file_struct *));
>+		qsort(the_fattr_list.files, j,
>+		      sizeof the_fattr_list.files[0], (int (*)())fattr_compare);
>+		the_fattr_list.low = 0;
>+		while (j-- > 0) {
>+			struct file_struct *fp = the_fattr_list.files[j];
>+			if (fp->basename && S_ISREG(fp->mode) && F_LENGTH(fp))
>+				break;
>+		}
>+		the_fattr_list.high = j;
>+	}
>+
> 	if (protocol_version < 30) {
> 		/* Recv the io_error flag */
> 		int err = read_int(f);
>diff --git a/generator.c b/generator.c
>--- a/generator.c
>+++ b/generator.c
>@@ -78,6 +78,7 @@ extern char *partial_dir;
> extern int compare_dest;
> extern int copy_dest;
> extern int link_dest;
>+extern int detect_renamed;
> extern int whole_file;
> extern int list_only;
> extern int read_batch;
>@@ -96,10 +97,12 @@ extern char *tmpdir;
> extern char *basis_dir[MAX_BASIS_DIRS+1];
> extern struct file_list *cur_flist, *first_flist, *dir_flist;
> extern filter_rule_list filter_list, daemon_filter_list;
>+extern struct file_list the_fattr_list;
> 
> int maybe_ATTRS_REPORT = 0;
> 
> static dev_t dev_zero;
>+static int unexplored_dirs = 1;
> static int deldelay_size = 0, deldelay_cnt = 0;
> static char *deldelay_buf = NULL;
> static int deldelay_fd = -1;
>@@ -179,6 +182,8 @@ static int remember_delete(struct file_struct *file, const char *fname, int flag
> 		if (!flush_delete_delay())
> 			return 0;
> 	}
>+	if (flags & DEL_NO_DELETIONS)
>+		return DR_SUCCESS;
> 
> 	return 1;
> }
>@@ -270,13 +275,18 @@ static void do_delayed_deletions(char *delbuf)
>  * all the --delete-WHEN options.  Note that the fbuf pointer must point to a
>  * MAXPATHLEN buffer with the name of the directory in it (the functions we
>  * call will append names onto the end, but the old dir value will be restored
>- * on exit). */
>-static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
>+ * on exit).
>+ *
>+ * Note:  --detect-rename may use this routine with DEL_NO_DELETIONS set!
>+ */
>+static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev,
>+			  int del_flags)
> {
> 	static int already_warned = 0;
> 	struct file_list *dirlist;
>-	char delbuf[MAXPATHLEN];
>-	int dlen, i;
>+	char *p, delbuf[MAXPATHLEN];
>+	unsigned remainder;
>+	int dlen, i, restore_dot = 0;
> 
> 	if (!fbuf) {
> 		change_local_filter_dir(NULL, 0, 0);
>@@ -290,17 +300,22 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
> 		maybe_send_keepalive(time(NULL), MSK_ALLOW_FLUSH);
> 
> 	if (io_error & IOERR_GENERAL && !ignore_errors) {
>-		if (already_warned)
>+		if (!already_warned) {
>+			rprintf(FINFO,
>+			    "IO error encountered -- skipping file deletion\n");
>+			already_warned = 1;
>+		}
>+		if (!detect_renamed)
> 			return;
>-		rprintf(FINFO,
>-			"IO error encountered -- skipping file deletion\n");
>-		already_warned = 1;
>-		return;
>+		del_flags |= DEL_NO_DELETIONS;
> 	}
> 
> 	dlen = strlen(fbuf);
> 	change_local_filter_dir(fbuf, dlen, F_DEPTH(file));
> 
>+	if (detect_renamed)
>+		unexplored_dirs--;
>+
> 	if (one_file_system) {
> 		if (file->flags & FLAG_TOP_DIR)
> 			filesystem_dev = *fs_dev;
>@@ -310,6 +325,14 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
> 
> 	dirlist = get_dirlist(fbuf, dlen, 0);
> 
>+	p = fbuf + dlen;
>+	if (dlen == 1 && *fbuf == '.') {
>+		restore_dot = 1;
>+		p = fbuf;
>+	} else if (dlen != 1 || *fbuf != '/')
>+		*p++ = '/';
>+	remainder = MAXPATHLEN - (p - fbuf);
>+
> 	/* If an item in dirlist is not found in flist, delete it
> 	 * from the filesystem. */
> 	for (i = dirlist->used; i--; ) {
>@@ -322,6 +345,10 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
> 					f_name(fp, NULL));
> 			continue;
> 		}
>+		if (detect_renamed && S_ISREG(fp->mode)) {
>+			strlcpy(p, fp->basename, remainder);
>+			look_for_rename(fp, fbuf);
>+		}
> 		/* Here we want to match regardless of file type.  Replacement
> 		 * of a file with one of another type is handled separately by
> 		 * a delete_item call with a DEL_MAKE_ROOM flag. */
>@@ -330,14 +357,19 @@ static void delete_in_dir(char *fbuf, struct file_struct *file, dev_t *fs_dev)
> 			if (!(fp->mode & S_IWUSR) && !am_root && fp->flags & FLAG_OWNED_BY_US)
> 				flags |= DEL_NO_UID_WRITE;
> 			f_name(fp, delbuf);
>-			if (delete_during == 2) {
>-				if (!remember_delete(fp, delbuf, flags))
>+			if (delete_during == 2 && !(del_flags & DEL_NO_DELETIONS)) {
>+				if (!remember_delete(fp, delbuf, del_flags | flags))
> 					break;
> 			} else
>-				delete_item(delbuf, fp->mode, flags);
>-		}
>+				delete_item(delbuf, fp->mode, del_flags | flags);
>+		} else if (detect_renamed && S_ISDIR(fp->mode))
>+			unexplored_dirs++;
> 	}
> 
>+	if (restore_dot)
>+		fbuf[0] = '.';
>+	fbuf[dlen] = '\0';
>+
> 	flist_free(dirlist);
> }
> 
>@@ -370,14 +402,122 @@ static void do_delete_pass(void)
> 		 || !S_ISDIR(st.st_mode))
> 			continue;
> 
>-		delete_in_dir(fbuf, file, &st.st_dev);
>+		delete_in_dir(fbuf, file, &st.st_dev, 0);
> 	}
>-	delete_in_dir(NULL, NULL, &dev_zero);
>+	delete_in_dir(NULL, NULL, &dev_zero, 0);
> 
> 	if (INFO_GTE(FLIST, 2) && !am_server)
> 		rprintf(FINFO, "                    \r");
> }
> 
>+/* Search for a regular file that matches either (1) the size & modified
>+ * time (plus the basename, if possible) or (2) the size & checksum.  If
>+ * we find an exact match down to the dirname, return -1 because we found
>+ * an up-to-date file in the transfer, not a renamed file. */
>+static int fattr_find(struct file_struct *f, char *fname)
>+{
>+	int low = the_fattr_list.low, high = the_fattr_list.high;
>+	int mid, ok_match = -1, good_match = -1;
>+	struct file_struct *fmid;
>+	int diff;
>+
>+	while (low <= high) {
>+		mid = (low + high) / 2;
>+		fmid = the_fattr_list.files[mid];
>+		if (F_LENGTH(fmid) != F_LENGTH(f)) {
>+			if (F_LENGTH(fmid) < F_LENGTH(f))
>+				low = mid + 1;
>+			else
>+				high = mid - 1;
>+			continue;
>+		}
>+		if (always_checksum) {
>+			/* We use the FLAG_FILE_SENT flag to indicate when we
>+			 * have computed the checksum for an entry. */
>+			if (!(f->flags & FLAG_FILE_SENT)) {
>+				if (fmid->modtime == f->modtime
>+				 && f_name_cmp(fmid, f) == 0)
>+					return -1; /* assume we can't help */
>+				file_checksum(fname, F_SUM(f), F_LENGTH(f));
>+				f->flags |= FLAG_FILE_SENT;
>+			}
>+			diff = u_memcmp(F_SUM(fmid), F_SUM(f), checksum_len);
>+			if (diff) {
>+				if (diff < 0)
>+					low = mid + 1;
>+				else
>+					high = mid - 1;
>+				continue;
>+			}
>+		} else {
>+			if (fmid->modtime != f->modtime) {
>+				if (fmid->modtime < f->modtime)
>+					low = mid + 1;
>+				else
>+					high = mid - 1;
>+				continue;
>+			}
>+		}
>+		ok_match = mid;
>+		diff = u_strcmp(fmid->basename, f->basename);
>+		if (diff == 0) {
>+			good_match = mid;
>+			if (fmid->dirname == f->dirname)
>+				return -1; /* file is up-to-date */
>+			if (!fmid->dirname) {
>+				low = mid + 1;
>+				continue;
>+			}
>+			if (!f->dirname) {
>+				high = mid - 1;
>+				continue;
>+			}
>+			diff = u_strcmp(fmid->dirname, f->dirname);
>+			if (diff == 0)
>+				return -1; /* file is up-to-date */
>+		}
>+		if (diff < 0)
>+			low = mid + 1;
>+		else
>+			high = mid - 1;
>+	}
>+
>+	return good_match >= 0 ? good_match : ok_match;
>+}
>+
>+void look_for_rename(struct file_struct *file, char *fname)
>+{
>+	struct file_struct *fp;
>+	char *partialptr, *fn;
>+	STRUCT_STAT st;
>+	int ndx;
>+
>+	if (!partial_dir || (ndx = fattr_find(file, fname)) < 0)
>+		return;
>+
>+	fp = the_fattr_list.files[ndx];
>+	fn = f_name(fp, NULL);
>+	/* We don't provide an alternate-basis file if there is a basis file. */
>+	if (link_stat(fn, &st, 0) == 0)
>+		return;
>+
>+	if (!dry_run) {
>+		if ((partialptr = partial_dir_fname(fn)) == NULL
>+		 || !handle_partial_dir(partialptr, PDIR_CREATE))
>+			return;
>+		/* We only use the file if we can hard-link it into our tmp dir. */
>+		if (link(fname, partialptr) != 0) {
>+			if (errno != EEXIST)
>+				handle_partial_dir(partialptr, PDIR_DELETE);
>+			return;
>+		}
>+	}
>+
>+	/* I think this falls into the -vv category with "%s is uptodate", etc. */
>+	if (INFO_GTE(MISC, 2))
>+		rprintf(FINFO, "found renamed: %s => %s\n", fname, fn);
>+}
>+
> static inline int time_differs(struct file_struct *file, stat_x *sxp)
> {
> 	return cmp_time(sxp->st.st_mtime, file->modtime);
>@@ -1141,6 +1281,7 @@ static void list_file_entry(struct file_struct *f)
> 	}
> }
> 
>+static struct bitbag *delayed_bits = NULL;
> static int phase = 0;
> static int dflt_perms;
> 
>@@ -1250,7 +1391,7 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
> 			 && do_stat(dn, &sx.st) < 0) {
> 				if (dry_run)
> 					goto parent_is_dry_missing;
>-				if (make_path(fname, MKP_DROP_NAME | MKP_SKIP_SLASH) < 0) {
>+				if (make_path(fname, ACCESSPERMS, MKP_DROP_NAME | MKP_SKIP_SLASH) < 0) {
> 					rsyserr(FERROR_XFER, errno,
> 						"recv_generator: mkdir %s failed",
> 						full_fname(dn));
>@@ -1401,7 +1542,7 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
> 		}
> 		if (real_ret != 0 && do_mkdir(fname,file->mode|added_perms) < 0 && errno != EEXIST) {
> 			if (!relative_paths || errno != ENOENT
>-			 || make_path(fname, MKP_DROP_NAME | MKP_SKIP_SLASH) < 0
>+			 || make_path(fname, ACCESSPERMS, MKP_DROP_NAME | MKP_SKIP_SLASH) < 0
> 			 || (do_mkdir(fname, file->mode|added_perms) < 0 && errno != EEXIST)) {
> 				rsyserr(FERROR_XFER, errno,
> 					"recv_generator: mkdir %s failed",
>@@ -1450,9 +1591,12 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
> 		}
> 		else if (delete_during && f_out != -1 && !phase
> 		    && !(file->flags & FLAG_MISSING_DIR)) {
>-			if (file->flags & FLAG_CONTENT_DIR)
>-				delete_in_dir(fname, file, &real_sx.st.st_dev);
>-			else
>+			if (file->flags & FLAG_CONTENT_DIR) {
>+				if (detect_renamed && real_ret != 0)
>+					unexplored_dirs++;
>+				delete_in_dir(fname, file, &real_sx.st.st_dev,
>+					      delete_during < 0 ? DEL_NO_DELETIONS : 0);
>+			} else
> 				change_local_filter_dir(fname, strlen(fname), F_DEPTH(file));
> 		}
> 		goto cleanup;
>@@ -1713,8 +1857,14 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx,
> 			goto cleanup;
> 		}
> #endif
>-		if (stat_errno == ENOENT)
>+		if (stat_errno == ENOENT) {
>+			if (detect_renamed && unexplored_dirs > 0
>+			 && F_LENGTH(file)) {
>+				bitbag_set_bit(delayed_bits, ndx);
>+				return;
>+			}
> 			goto notify_others;
>+		}
> 		rsyserr(FERROR_XFER, stat_errno, "recv_generator: failed to stat %s",
> 			full_fname(fname));
> 		goto cleanup;
>@@ -2174,6 +2324,12 @@ void generate_files(int f_out, const char *local_name)
> 	if (DEBUG_GTE(GENR, 1))
> 		rprintf(FINFO, "generator starting pid=%d\n", (int)getpid());
> 
>+	if (detect_renamed) {
>+		delayed_bits = bitbag_create(cur_flist->used);
>+		if (!delete_before && !delete_during)
>+			delete_during = -1;
>+	}
>+
> 	if (delete_before && !solo_file && cur_flist->used > 0)
> 		do_delete_pass();
> 	if (delete_during == 2) {
>@@ -2184,7 +2340,7 @@ void generate_files(int f_out, const char *local_name)
> 	}
> 	info_levels[INFO_FLIST] = info_levels[INFO_PROGRESS] = 0;
> 
>-	if (append_mode > 0 || whole_file < 0)
>+	if (append_mode > 0 || detect_renamed || whole_file < 0)
> 		whole_file = 0;
> 	if (DEBUG_GTE(FLIST, 1)) {
> 		rprintf(FINFO, "delta-transmission %s\n",
>@@ -2220,7 +2376,7 @@ void generate_files(int f_out, const char *local_name)
> 						dirdev = MAKEDEV(DEV_MAJOR(devp), DEV_MINOR(devp));
> 					} else
> 						dirdev = MAKEDEV(0, 0);
>-					delete_in_dir(fbuf, fp, &dirdev);
>+					delete_in_dir(fbuf, fp, &dirdev, 0);
> 				} else
> 					change_local_filter_dir(fbuf, strlen(fbuf), F_DEPTH(fp));
> 			}
>@@ -2267,7 +2423,21 @@ void generate_files(int f_out, const char *local_name)
> 	} while ((cur_flist = cur_flist->next) != NULL);
> 
> 	if (delete_during)
>-		delete_in_dir(NULL, NULL, &dev_zero);
>+		delete_in_dir(NULL, NULL, &dev_zero, 0);
>+	if (detect_renamed) {
>+		if (delete_during < 0)
>+			delete_during = 0;
>+		detect_renamed = 0;
>+
>+		for (i = -1; (i = bitbag_next_bit(delayed_bits, i)) >= 0; ) {
>+			struct file_struct *file = cur_flist->files[i];
>+			if (local_name)
>+				strlcpy(fbuf, local_name, sizeof fbuf);
>+			else
>+				f_name(file, fbuf);
>+			recv_generator(fbuf, file, i, itemizing, code, f_out);
>+		}
>+	}
> 	phase++;
> 	if (DEBUG_GTE(GENR, 1))
> 		rprintf(FINFO, "generate_files phase=%d\n", phase);
>diff --git a/main.c b/main.c
>--- a/main.c
>+++ b/main.c
>@@ -850,7 +850,7 @@ static int do_recv(int f_in, int f_out, char *local_name)
> 	}
> 
> 	if (backup_dir) {
>-		int ret = make_path(backup_dir_buf, MKP_DROP_NAME); /* drops trailing slash */
>+		int ret = make_path(backup_dir_buf, ACCESSPERMS, MKP_DROP_NAME); /* drops trailing slash */
> 		if (ret < 0)
> 			exit_cleanup(RERR_SYNTAX);
> 		if (ret)
>diff --git a/options.c b/options.c
>--- a/options.c
>+++ b/options.c
>@@ -81,6 +81,7 @@ int am_server = 0;
> int am_sender = 0;
> int am_starting_up = 1;
> int relative_paths = -1;
>+int detect_renamed = 0;
> int implied_dirs = 1;
> int missing_args = 0; /* 0 = FERROR_XFER, 1 = ignore, 2 = delete */
> int numeric_ids = 0;
>@@ -758,6 +759,7 @@ void usage(enum logcode F)
>   rprintf(F,"     --modify-window=NUM     compare mod-times with reduced accuracy\n");
>   rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
>   rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
>+  rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
>   rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
>   rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
>   rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
>@@ -961,6 +963,7 @@ static struct poptOption long_options[] = {
>   {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
>   {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
>   {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
>+  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
>   {"fuzzy",           'y', POPT_ARG_NONE,   0, 'y', 0, 0 },
>   {"no-fuzzy",         0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
>   {"no-y",             0,  POPT_ARG_VAL,    &fuzzy_basis, 0, 0, 0 },
>@@ -2228,7 +2231,7 @@ int parse_arguments(int *argc_p, const char ***argv_p)
> 		inplace = 1;
> 	}
> 
>-	if (delay_updates && !partial_dir)
>+	if ((delay_updates || detect_renamed) && !partial_dir)
> 		partial_dir = tmp_partialdir;
> 
> 	if (inplace) {
>@@ -2237,6 +2240,7 @@ int parse_arguments(int *argc_p, const char ***argv_p)
> 			snprintf(err_buf, sizeof err_buf,
> 				 "--%s cannot be used with --%s\n",
> 				 append_mode ? "append" : "inplace",
>+				 detect_renamed ? "detect-renamed" :
> 				 delay_updates ? "delay-updates" : "partial-dir");
> 			return 0;
> 		}
>@@ -2606,6 +2610,8 @@ void server_options(char **args, int *argc_p)
> 			args[ac++] = "--super";
> 		if (size_only)
> 			args[ac++] = "--size-only";
>+		if (detect_renamed)
>+			args[ac++] = "--detect-renamed";
> 		if (do_stats)
> 			args[ac++] = "--stats";
> 	} else {
>diff --git a/receiver.c b/receiver.c
>--- a/receiver.c
>+++ b/receiver.c
>@@ -208,7 +208,7 @@ int open_tmpfile(char *fnametmp, const char *fname, struct file_struct *file)
> 	 * information should have been previously transferred, but that may
> 	 * not be the case with -R */
> 	if (fd == -1 && relative_paths && errno == ENOENT
>-	 && make_path(fnametmp, MKP_SKIP_SLASH | MKP_DROP_NAME) == 0) {
>+	 && make_path(fnametmp, ACCESSPERMS, MKP_SKIP_SLASH | MKP_DROP_NAME) == 0) {
> 		/* Get back to name with XXXXXX in it. */
> 		get_tmpname(fnametmp, fname, False);
> 		fd = do_mkstemp(fnametmp, (file->mode|added_perms) & INITACCESSPERMS);
>diff --git a/rsync.h b/rsync.h
>--- a/rsync.h
>+++ b/rsync.h
>@@ -251,7 +251,7 @@ enum msgcode {
> #define NDX_DEL_STATS -3
> #define NDX_FLIST_OFFSET -101
> 
>-/* For calling delete_item() and delete_dir_contents(). */
>+/* For calling delete_item(), delete_dir_contents(), and delete_in_dir(). */
> #define DEL_NO_UID_WRITE 	(1<<0) /* file/dir has our uid w/o write perm */
> #define DEL_RECURSE		(1<<1) /* if dir, delete all contents */
> #define DEL_DIR_IS_EMPTY	(1<<2) /* internal delete_FUNCTIONS use only */
>@@ -261,6 +261,7 @@ enum msgcode {
> #define DEL_FOR_DEVICE		(1<<6) /* making room for a replacement device */
> #define DEL_FOR_SPECIAL 	(1<<7) /* making room for a replacement special */
> #define DEL_FOR_BACKUP	 	(1<<8) /* the delete is for a backup operation */
>+#define DEL_NO_DELETIONS	(1<<9) /* just check for renames w/o deleting */
> 
> #define DEL_MAKE_ROOM (DEL_FOR_FILE|DEL_FOR_DIR|DEL_FOR_SYMLINK|DEL_FOR_DEVICE|DEL_FOR_SPECIAL)
> 
>diff --git a/rsync.yo b/rsync.yo
>--- a/rsync.yo
>+++ b/rsync.yo
>@@ -413,6 +413,7 @@ to the detailed description below for a complete description.  verb(
>      --modify-window=NUM     compare mod-times with reduced accuracy
>  -T, --temp-dir=DIR          create temporary files in directory DIR
>  -y, --fuzzy                 find similar file for basis if no dest file
>+     --detect-renamed        try to find renamed files to speed the xfer
>      --compare-dest=DIR      also compare received files relative to DIR
>      --copy-dest=DIR         ... and include copies of unchanged files
>      --link-dest=DIR         hardlink to files in DIR when unchanged
>@@ -1768,6 +1769,21 @@ Note that the use of the bf(--delete) option might get rid of any potential
> fuzzy-match files, so either use bf(--delete-after) or specify some
> filename exclusions if you need to prevent this.
> 
>+dit(bf(--detect-renamed)) With this option, for each new source file
>+(call it em(src/S)), rsync looks for a file em(dest/D) anywhere in the
>+destination that passes the quick check with em(src/S).  If such a em(dest/D)
>+is found, rsync uses it as an alternate basis for transferring em(S).  The
>+idea is that if em(src/S) was renamed from em(src/D) (as opposed to em(src/S)
>+passing the quick check with em(dest/D) by coincidence), the delta-transfer
>+algorithm will find that all the data matches between em(src/S) and em(dest/D),
>+and the transfer will be really fast.
>+
>+By default, alternate-basis files are hard-linked into a directory named
>+".~tmp~" in each file's destination directory, but if you've specified
>+the bf(--partial-dir) option, that directory will be used instead.  These
>+potential alternate-basis files will be removed as the transfer progresses.
>+This option conflicts with bf(--inplace) and bf(--append).
>+
> dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
> the destination machine as an additional hierarchy to compare destination
> files against doing transfers (if the files are missing in the destination
>diff --git a/util.c b/util.c
>--- a/util.c
>+++ b/util.c
>@@ -174,7 +174,7 @@ int set_modtime(const char *fname, time_t modtime, uint32 mod_nsec, mode_t mode)
> /* Create any necessary directories in fname.  Any missing directories are
>  * created with default permissions.  Returns < 0 on error, or the number
>  * of directories created. */
>-int make_path(char *fname, int flags)
>+int make_path(char *fname, mode_t mode, int flags)
> {
> 	char *end, *p;
> 	int ret = 0;
>@@ -197,7 +197,7 @@ int make_path(char *fname, int flags)
> 
> 	/* Try to find an existing dir, starting from the deepest dir. */
> 	for (p = end; ; ) {
>-		if (do_mkdir(fname, ACCESSPERMS) == 0) {
>+		if (do_mkdir(fname, mode) == 0) {
> 			ret++;
> 			break;
> 		}
>@@ -229,7 +229,7 @@ int make_path(char *fname, int flags)
> 		p += strlen(p);
> 		if (ret < 0) /* Skip mkdir on error, but keep restoring the path. */
> 			continue;
>-		if (do_mkdir(fname, ACCESSPERMS) < 0)
>+		if (do_mkdir(fname, mode) < 0)
> 			ret = -ret - 1;
> 		else
> 			ret++;
>@@ -1111,6 +1111,32 @@ char *normalize_path(char *path, BOOL force_newbuf, unsigned int *len_ptr)
> 	return path;
> }
> 
>+/* We need to supply our own strcmp function for file list comparisons
>+ * to ensure that signed/unsigned usage is consistent between machines. */
>+int u_strcmp(const char *p1, const char *p2)
>+{
>+        for ( ; *p1; p1++, p2++) {
>+		if (*p1 != *p2)
>+			break;
>+	}
>+
>+	return (int)*(uchar*)p1 - (int)*(uchar*)p2;
>+}
>+
>+/* We need a memcmp function compares unsigned-byte values. */
>+int u_memcmp(const void *p1, const void *p2, size_t len)
>+{
>+	const uchar *u1 = p1;
>+	const uchar *u2 = p2;
>+
>+	while (len--) {
>+		if (*u1 != *u2)
>+			return (int)*u1 - (int)*u2;
>+	}
>+
>+	return 0;
>+}
>+
> /**
>  * Return a quoted string with the full pathname of the indicated filename.
>  * The string " (in MODNAME)" may also be appended.  The returned pointer
>@@ -1204,7 +1230,7 @@ int handle_partial_dir(const char *fname, int create)
> 			}
> 			statret = -1;
> 		}
>-		if (statret < 0 && do_mkdir(dir, 0700) < 0) {
>+		if (statret < 0 && make_path(dir, 0700, 0) < 0) {
> 			*fn = '/';
> 			return 0;
> 		}
>diff -up a/proto.h b/proto.h
>--- a/proto.h
>+++ b/proto.h
>@@ -101,6 +101,7 @@ int f_name_has_prefix(const struct file_
> char *f_name_buf(void);
> char *f_name(const struct file_struct *f, char *fbuf);
> struct file_list *get_dirlist(char *dirname, int dlen, int flags);
>+void look_for_rename(struct file_struct *file, char *fname);
> int unchanged_attrs(const char *fname, struct file_struct *file, stat_x *sxp);
> void itemize(const char *fnamecmp, struct file_struct *file, int ndx, int statret,
> 	     stat_x *sxp, int32 iflags, uchar fnamecmp_type,
>@@ -353,7 +354,7 @@ void set_blocking(int fd);
> int fd_pair(int fd[2]);
> void print_child_argv(const char *prefix, char **cmd);
> int set_modtime(const char *fname, time_t modtime, uint32 mod_nsec, mode_t mode);
>-int make_path(char *fname, int flags);
>+int make_path(char *fname, mode_t mode, int flags);
> int full_write(int desc, const char *ptr, size_t len);
> int copy_file(const char *source, const char *dest, int ofd, mode_t mode);
> int robust_unlink(const char *fname);
>@@ -373,6 +374,8 @@ char *sanitize_path(char *dest, const ch
> 		    int flags);
> int change_dir(const char *dir, int set_path_only);
> char *normalize_path(char *path, BOOL force_newbuf, unsigned int *len_ptr);
>+int u_strcmp(const char *p1, const char *p2);
>+int u_memcmp(const void *p1, const void *p2, size_t len);
> char *full_fname(const char *fn);
> char *partial_dir_fname(const char *fname);
> int handle_partial_dir(const char *fname, int create);
>diff -up a/rsync.1 b/rsync.1
>--- a/rsync.1
>+++ b/rsync.1
>@@ -489,6 +489,7 @@ to the detailed description below for a
>      \-\-modify\-window=NUM     compare mod\-times with reduced accuracy
>  \-T, \-\-temp\-dir=DIR          create temporary files in directory DIR
>  \-y, \-\-fuzzy                 find similar file for basis if no dest file
>+     \-\-detect\-renamed        try to find renamed files to speed the xfer
>      \-\-compare\-dest=DIR      also compare received files relative to DIR
>      \-\-copy\-dest=DIR         ... and include copies of unchanged files
>      \-\-link\-dest=DIR         hardlink to files in DIR when unchanged
>@@ -2020,6 +2021,22 @@ Note that the use of the \fB\-\-delete\f
> fuzzy\-match files, so either use \fB\-\-delete\-after\fP or specify some
> filename exclusions if you need to prevent this.
> .IP 
>+.IP "\fB\-\-detect\-renamed\fP"
>+With this option, for each new source file
>+(call it \fIsrc/S\fP), rsync looks for a file \fIdest/D\fP anywhere in the
>+destination that passes the quick check with \fIsrc/S\fP.  If such a \fIdest/D\fP
>+is found, rsync uses it as an alternate basis for transferring \fIS\fP.  The
>+idea is that if \fIsrc/S\fP was renamed from \fIsrc/D\fP (as opposed to \fIsrc/S\fP
>+passing the quick check with \fIdest/D\fP by coincidence), the delta\-transfer
>+algorithm will find that all the data matches between \fIsrc/S\fP and \fIdest/D\fP,
>+and the transfer will be really fast.
>+.IP 
>+By default, alternate\-basis files are hard\-linked into a directory named
>+\(dq\&.~tmp~\(dq\& in each file\(cq\&s destination directory, but if you\(cq\&ve specified
>+the \fB\-\-partial\-dir\fP option, that directory will be used instead.  These
>+potential alternate\-basis files will be removed as the transfer progresses.
>+This option conflicts with \fB\-\-inplace\fP and \fB\-\-append\fP.
>+.IP 
> .IP "\fB\-\-compare\-dest=DIR\fP"
> This option instructs rsync to use \fIDIR\fP on
> the destination machine as an additional hierarchy to compare destination

Actions: View

Attachments on bug 10522: 9840