1 This builds on the sender optimization feature of the checksum4mirrors
2 patch and adds the ability to create and/or updates the .rsyncsums files
3 when --checksum-updating (or "checksum updating = true") is specified.
5 To use this patch, run these commands for a successful build:
7 patch -p1 <patches/checksum4mirrors.diff
8 patch -p1 <patches/checksum-updating.diff
9 ./configure (optional if already run)
12 --- old/clientserver.c
13 +++ new/clientserver.c
14 @@ -37,6 +37,7 @@ extern int sanitize_paths;
15 extern int filesfrom_fd;
16 extern int remote_protocol;
17 extern int protocol_version;
18 +extern int checksum_updating;
19 extern int io_timeout;
21 extern int default_af_hint;
22 @@ -687,6 +688,8 @@ static int rsync_module(int f_in, int f_
23 else if (am_root < 0) /* Treat --fake-super from client as --super. */
26 + checksum_updating = lp_checksum_updating(i);
28 if (filesfrom_fd == 0)
41 @@ -58,6 +59,7 @@ extern int implied_dirs;
42 extern int file_extra_cnt;
43 extern int ignore_perishable;
44 extern int non_perishable_cnt;
45 +extern int checksum_updating;
46 extern int prune_empty_dirs;
47 extern int copy_links;
48 extern int copy_unsafe_links;
49 @@ -82,6 +84,9 @@ extern iconv_t ic_send, ic_recv;
51 #define PTR_SIZE (sizeof (struct file_struct *))
53 +#define FLAG_SUM_MISSING (1<<1) /* F_SUM() data is undefined */
54 +#define FLAG_SUM_KEEP (1<<2) /* keep entry when rewriting */
58 dev_t filesystem_dev; /* used to implement -x */
59 @@ -117,6 +122,9 @@ static char empty_sum[MAX_DIGEST_LEN];
60 static int flist_count_offset; /* for --delete --progress */
61 static int dir_count = 0;
62 static int high_hlink_ndx;
63 +static int checksum_matches = 0;
64 +static int checksum_updates = 0;
65 +static int regular_skipped = 0;
66 static struct file_list *checksum_flist = NULL;
68 static void clean_flist(struct file_list *flist, int strip_root);
69 @@ -307,7 +315,8 @@ static void flist_done_allocating(struct
71 /* The len count is the length of the basename + 1 for the null. */
72 static int add_checksum(const char *dirname, const char *basename, int len,
73 - OFF_T file_length, time_t mtime, const char *sum)
74 + OFF_T file_length, time_t mtime, int32 ctime, int32 inode,
75 + const char *sum, const char *alt_sum, int flags)
77 struct file_struct *file;
78 int alloc_len, extra_len;
79 @@ -318,13 +327,14 @@ static int add_checksum(const char *dirn
83 - extra_len = (file_extra_cnt + (file_length > 0xFFFFFFFFu) + SUM_EXTRA_CNT)
84 + /* "2" is for a 32-bit ctime num and an 32-bit inode num. */
85 + extra_len = (file_extra_cnt + (file_length > 0xFFFFFFFFu) + SUM_EXTRA_CNT + 2)
87 #if EXTRA_ROUNDING > 0
88 if (extra_len & (EXTRA_ROUNDING * EXTRA_LEN))
89 extra_len = (extra_len | (EXTRA_ROUNDING * EXTRA_LEN)) + EXTRA_LEN;
91 - alloc_len = FILE_STRUCT_LEN + extra_len + len;
92 + alloc_len = FILE_STRUCT_LEN + extra_len + len + checksum_len*2 + 1;
93 bp = pool_alloc(checksum_flist->file_pool, alloc_len, "add_checksum");
95 memset(bp, 0, extra_len + FILE_STRUCT_LEN);
96 @@ -333,7 +343,14 @@ static int add_checksum(const char *dirn
97 bp += FILE_STRUCT_LEN;
99 memcpy(bp, basename, len);
101 + strlcpy(bp+len, alt_sum, checksum_len*2 + 1);
103 + memset(bp+len, '=', checksum_len*2);
104 + bp[len+checksum_len*2] = '\0';
107 + file->flags = flags;
108 file->mode = S_IFREG;
109 file->modtime = mtime;
110 file->len32 = (uint32)file_length;
111 @@ -344,6 +361,8 @@ static int add_checksum(const char *dirn
112 file->dirname = dirname;
114 memcpy(bp, sum, checksum_len);
115 + F_CTIME(file) = ctime;
116 + F_INODE(file) = inode;
118 flist_expand(checksum_flist, 1);
119 checksum_flist->files[checksum_flist->used++] = file;
120 @@ -353,17 +372,104 @@ static int add_checksum(const char *dirn
124 +static void write_checksums(const char *next_dirname, int whole_dir)
126 + static const char *dirname_save;
127 + char fbuf[MAXPATHLEN];
128 + const char *dirname;
129 + int used, new_entries, counts_match, no_skipped;
133 + dirname = dirname_save;
134 + dirname_save = next_dirname;
139 + used = checksum_flist->used;
140 + new_entries = checksum_updates != 0;
141 + counts_match = used == checksum_matches;
142 + no_skipped = whole_dir && regular_skipped == 0;
144 + clean_flist(checksum_flist, 0);
146 + checksum_flist->used = 0;
147 + checksum_matches = 0;
148 + checksum_updates = 0;
149 + regular_skipped = 0;
155 + if (pathjoin(fbuf, sizeof fbuf, dirname, ".rsyncsums") >= sizeof fbuf)
158 + strlcpy(fbuf, ".rsyncsums", sizeof fbuf);
160 + if (checksum_flist->high - checksum_flist->low < 0 && no_skipped) {
165 + if (!new_entries && (counts_match || !whole_dir))
168 + if (!(out_fp = fopen(fbuf, "w")))
172 + for (i = checksum_flist->low; i <= checksum_flist->high; i++) {
173 + struct file_struct *file = checksum_flist->sorted[i];
174 + const char *cp = F_SUM(file);
175 + const char *end = cp + checksum_len;
176 + const char *alt_sum = file->basename + strlen(file->basename) + 1;
177 + int32 ctime, inode;
178 + if (whole_dir && !(file->flags & FLAG_SUM_KEEP))
180 + ctime = F_CTIME(file);
181 + inode = F_INODE(file);
182 + if (protocol_version >= 30)
183 + fprintf(out_fp, "%s ", alt_sum);
184 + if (file->flags & FLAG_SUM_MISSING) {
187 + fprintf(out_fp, "==");
188 + } while (++cp != end);
191 + fprintf(out_fp, "%02x", (int)CVAL(cp, 0));
192 + } while (++cp != end);
194 + if (protocol_version < 30)
195 + fprintf(out_fp, " %s", alt_sum);
196 + if (*alt_sum == '=')
198 + fprintf(out_fp, " %10.0f %10.0f %10lu %10lu %s\n",
199 + (double)F_LENGTH(file), (double)file->modtime,
200 + (long)ctime, (long)inode, file->basename);
206 /* The direname value must remain unchanged during the lifespan of the
207 * created checksum_flist object because we use it directly. */
208 static void read_checksums(const char *dirname)
210 char line[MAXPATHLEN+1024], fbuf[MAXPATHLEN], sum[MAX_DIGEST_LEN];
211 + const char *alt_sum = NULL;
215 + int32 ctime, inode;
216 + int len, dlen, i, flags;
220 + if (checksum_updating)
221 + write_checksums(dirname, 0);
223 if (checksum_flist) {
224 /* Reset the pool memory and empty the file-list array. */
225 pool_free_old(checksum_flist->file_pool,
226 @@ -374,6 +480,9 @@ static void read_checksums(const char *d
228 checksum_flist->low = 0;
229 checksum_flist->high = -1;
230 + checksum_matches = 0;
231 + checksum_updates = 0;
232 + regular_skipped = 0;
236 @@ -392,7 +501,7 @@ static void read_checksums(const char *d
237 while (fgets(line, sizeof line, fp)) {
239 if (protocol_version >= 30) {
240 - char *alt_sum = cp;
243 while (*++cp == '=') {}
245 @@ -403,7 +512,14 @@ static void read_checksums(const char *d
250 + for (i = 0; i < checksum_len*2; i++, cp++) {
256 + memset(sum, 0, checksum_len);
257 + flags = FLAG_SUM_MISSING;
259 for (i = 0; i < checksum_len*2; i++, cp++) {
261 @@ -421,13 +537,14 @@ static void read_checksums(const char *d
269 while (*++cp == ' ') {}
271 if (protocol_version < 30) {
272 - char *alt_sum = cp;
275 while (*++cp == '=') {}
277 @@ -451,16 +568,16 @@ static void read_checksums(const char *d
279 while (*++cp == ' ') {}
281 - /* Ignore ctime. */
285 + ctime = ctime * 10 + *cp++ - '0';
288 while (*++cp == ' ') {}
290 - /* Ignore inode. */
294 + inode = inode * 10 + *cp++ - '0';
297 while (*++cp == ' ') {}
298 @@ -477,8 +594,13 @@ static void read_checksums(const char *d
301 strlcpy(fbuf+dlen, cp, sizeof fbuf - dlen);
302 + if (is_excluded(fbuf, 0, ALL_FILTERS)) {
303 + flags |= FLAG_SUM_KEEP;
304 + checksum_matches++;
307 - add_checksum(dirname, cp, len, file_length, mtime, sum);
308 + add_checksum(dirname, cp, len, file_length, mtime, ctime, inode,
309 + sum, alt_sum, flags);
313 @@ -1260,6 +1382,8 @@ struct file_struct *make_file(const char
314 if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level)) {
315 if (ignore_perishable)
316 non_perishable_cnt++;
317 + if (S_ISREG(st.st_mode))
322 @@ -1391,13 +1515,36 @@ struct file_struct *make_file(const char
324 if (flist && (j = flist_find(checksum_flist, file)) >= 0) {
325 struct file_struct *fp = checksum_flist->sorted[j];
326 + int32 ctime = F_CTIME(fp);
327 + int32 inode = F_INODE(fp);
328 if (F_LENGTH(fp) == st.st_size
329 - && fp->modtime == st.st_mtime)
330 - memcpy(tmp_sum, F_SUM(fp), MAX_DIGEST_LEN);
332 - file_checksum(thisname, tmp_sum, st.st_size);
334 + && fp->modtime == st.st_mtime
335 + && ctime == (int32)st.st_ctime
336 + && inode == (int32)st.st_ino) {
337 + if (fp->flags & FLAG_SUM_MISSING) {
338 + fp->flags &= ~FLAG_SUM_MISSING;
339 + checksum_updates++;
340 + file_checksum(thisname, tmp_sum, st.st_size);
341 + memcpy(F_SUM(fp), tmp_sum, MAX_DIGEST_LEN);
343 + checksum_matches++;
344 + memcpy(tmp_sum, F_SUM(fp), MAX_DIGEST_LEN);
346 + fp->flags |= FLAG_SUM_KEEP;
349 + goto compute_new_checksum;
352 + compute_new_checksum:
353 file_checksum(thisname, tmp_sum, st.st_size);
354 + if (checksum_updating && flist) {
355 + checksum_updates +=
356 + add_checksum(file->dirname, basename, basename_len,
357 + st.st_size, st.st_mtime, st.st_ctime,
358 + st.st_ino, tmp_sum, NULL, FLAG_SUM_KEEP);
363 /* This code is only used by the receiver when it is building
364 @@ -1688,6 +1835,9 @@ static void send_directory(int f, struct
368 + if (checksum_updating && always_checksum && am_sender && f >= 0)
369 + write_checksums(NULL, 1);
371 if (f >= 0 && recurse && !divert_dirs) {
372 int i, end = flist->used - 1;
373 /* send_if_directory() bumps flist->used, so use "end". */
374 @@ -2253,7 +2403,7 @@ struct file_list *send_file_list(int f,
377 if (checksum_updating && always_checksum && flist_eof)
378 - read_checksums(NULL);
379 + read_checksums(NULL); /* writes any last updates */
385 @@ -153,6 +153,7 @@ typedef struct
389 + BOOL checksum_updating;
392 BOOL ignore_nonreadable;
393 @@ -201,6 +202,7 @@ static service sDefault =
394 /* syslog_facility; */ LOG_DAEMON,
397 + /* checksum_updating; */ False,
398 /* fake_super; */ False,
399 /* ignore_errors; */ False,
400 /* ignore_nonreadable; */ False,
401 @@ -317,6 +319,7 @@ static struct parm_struct parm_table[] =
402 {"lock file", P_STRING, P_LOCAL, &sDefault.lock_file, NULL,0},
403 {"log file", P_STRING, P_LOCAL, &sDefault.log_file, NULL,0},
404 {"log format", P_STRING, P_LOCAL, &sDefault.log_format, NULL,0},
405 + {"checksum updating", P_BOOL, P_LOCAL, &sDefault.checksum_updating, NULL,0},
406 {"max connections", P_INTEGER,P_LOCAL, &sDefault.max_connections, NULL,0},
407 {"max verbosity", P_INTEGER,P_LOCAL, &sDefault.max_verbosity, NULL,0},
408 {"name", P_STRING, P_LOCAL, &sDefault.name, NULL,0},
409 @@ -422,6 +425,7 @@ FN_LOCAL_BOOL(lp_fake_super, fake_super)
410 FN_LOCAL_BOOL(lp_ignore_errors, ignore_errors)
411 FN_LOCAL_BOOL(lp_ignore_nonreadable, ignore_nonreadable)
412 FN_LOCAL_BOOL(lp_list, list)
413 +FN_LOCAL_BOOL(lp_checksum_updating, checksum_updating)
414 FN_LOCAL_BOOL(lp_read_only, read_only)
415 FN_LOCAL_BOOL(lp_strict_modes, strict_modes)
416 FN_LOCAL_BOOL(lp_transfer_logging, transfer_logging)
419 @@ -110,6 +110,7 @@ size_t bwlimit_writemax = 0;
420 int ignore_existing = 0;
421 int ignore_non_existing = 0;
422 int need_messages_from_generator = 0;
423 +int checksum_updating = 0;
424 int max_delete = INT_MIN;
427 @@ -310,6 +311,7 @@ void usage(enum logcode F)
428 rprintf(F," -q, --quiet suppress non-error messages\n");
429 rprintf(F," --no-motd suppress daemon-mode MOTD (see manpage caveat)\n");
430 rprintf(F," -c, --checksum skip based on checksum, not mod-time & size\n");
431 + rprintf(F," --checksum-updating sender updates .rsyncsums files\n");
432 rprintf(F," -a, --archive archive mode; equals -rlptgoD (no -H,-A,-X)\n");
433 rprintf(F," --no-OPTION turn off an implied OPTION (e.g. --no-D)\n");
434 rprintf(F," -r, --recursive recurse into directories\n");
435 @@ -557,6 +559,7 @@ static struct poptOption long_options[]
436 {"checksum", 'c', POPT_ARG_VAL, &always_checksum, 1, 0, 0 },
437 {"no-checksum", 0, POPT_ARG_VAL, &always_checksum, 0, 0, 0 },
438 {"no-c", 0, POPT_ARG_VAL, &always_checksum, 0, 0, 0 },
439 + {"checksum-updating",0, POPT_ARG_NONE, &checksum_updating, 0, 0, 0 },
440 {"block-size", 'B', POPT_ARG_LONG, &block_size, 0, 0, 0 },
441 {"compare-dest", 0, POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
442 {"copy-dest", 0, POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
443 @@ -1954,7 +1957,9 @@ void server_options(char **args, int *ar
444 args[ac++] = basis_dir[i];
448 + } else if (checksum_updating)
449 + args[ac++] = "--checksum-updating";
456 @@ -662,6 +662,10 @@ extern int xattrs_ndx;
457 #define F_SUM(f) ((char*)OPT_EXTRA(f, LEN64_BUMP(f) + HLINK_BUMP(f) \
458 + SUM_EXTRA_CNT - 1))
460 +/* These are only valid on an entry read from a checksum file. */
461 +#define F_CTIME(f) OPT_EXTRA(f, LEN64_BUMP(f) + SUM_EXTRA_CNT)->num
462 +#define F_INODE(f) OPT_EXTRA(f, LEN64_BUMP(f) + SUM_EXTRA_CNT + 1)->num
464 /* Some utility defines: */
465 #define F_IS_ACTIVE(f) (f)->basename[0]
466 #define F_IS_HLINKED(f) ((f)->flags & FLAG_HLINKED)
469 @@ -317,6 +317,7 @@ to the detailed description below for a
470 -q, --quiet suppress non-error messages
471 --no-motd suppress daemon-mode MOTD (see caveat)
472 -c, --checksum skip based on checksum, not mod-time & size
473 + --checksum-updating sender updates .rsyncsums files
474 -a, --archive archive mode; equals -rlptgoD (no -H,-A,-X)
475 --no-OPTION turn off an implied OPTION (e.g. --no-D)
476 -r, --recursive recurse into directories
477 @@ -515,9 +516,9 @@ uses a "quick check" that (by default) c
478 of last modification match between the sender and receiver. This option
479 changes this to compare a 128-bit MD4 checksum for each file that has a
480 matching size. Generating the checksums means that both sides will expend
481 -a lot of disk I/O reading all the data in the files in the transfer (and
482 -this is prior to any reading that will be done to transfer changed files),
483 -so this can slow things down significantly.
484 +a lot of disk I/O reading the data in all the files in the transfer, so
485 +this can slow things down significantly (and this is prior to any reading
486 +that will be done to transfer the files that have changed).
488 The sending side generates its checksums while it is doing the file-system
489 scan that builds the list of the available files. The receiver generates
490 @@ -525,12 +526,42 @@ its checksums when it is scanning for ch
491 file that has the same size as the corresponding sender's file: files with
492 either a changed size or a changed checksum are selected for transfer.
494 +Starting with version 3.0.0, the sending side will look for a checksum
495 +summary file and use a pre-generated checksum that it reads out of the file
496 +(as long as it matches the file's size and modified time). This allows a
497 +server to support the --checksum option to clients without having to
498 +recompute the checksums for each client. See the bf(--checksum-updating)
499 +option for a way to have rsync create/update these checksum files.
501 Note that rsync always verifies that each em(transferred) file was
502 correctly reconstructed on the receiving side by checking a whole-file
503 checksum that is generated when as the file is transferred, but that
504 automatic after-the-transfer verification has nothing to do with this
505 option's before-the-transfer "Does this file need to be updated?" check.
507 +dit(bf(--checksum-updating)) This option tells the sending side to create
508 +and/or update per-directory checksum files that are used by the
509 +bf(--checksum) option. The file that is updated is named .rsyncsums. If
510 +pre-transfer checksums are not being computed, this option has no effect.
512 +The checksum files stores the computed checksum, last-known size,
513 +modification time, and name for each file in the current directory. If a
514 +later transfer finds that a file matches its prior size and modification
515 +time, the checksum is assumed to still be correct. Otherwise it is
516 +recomputed and udpated in the file.
518 +To avoid transferring the system's checksum files, you can use an exclude
519 +(e.g. bf(--exclude=.rsyncsums)). To make this easier to type, you can use
520 +a popt alias. For instance, adding the following line in your ~/.popt file
521 +defines a bf(-cc) option that enables checksum updating and excludes the
524 +verb( rsync alias --cc --checksum-updating --exclude=.rsyncsums)
526 +An rsync daemon does not allow the client to control this setting, so see
527 +the "checksum updating" daemon config option for information on how to make
528 +a daemon maintain these checksum files.
530 dit(bf(-a, --archive)) This is equivalent to bf(-rlptgoD). It is a quick
531 way of saying you want recursion and want to preserve almost
532 everything (with -H being a notable omission).
533 --- old/rsyncd.conf.yo
534 +++ new/rsyncd.conf.yo
535 @@ -200,6 +200,20 @@ locking on this file to ensure that the
536 exceeded for the modules sharing the lock file.
537 The default is tt(/var/run/rsyncd.lock).
539 +dit(bf(checksum updating)) This option tells rsync to update/create the
540 +checksum information in the per-directory checksum files when users copy
541 +files using the bf(--checksum) option. Any file that has changed since it
542 +was last checksummed (or is not mentioned) has its data updated in the
545 +Note that this updating will occur even if the module is listed as being
546 +read-only. If you want to hide these files (and you will almost always
547 +want to do), add ".rsyncsums" to the module's exclude setting.
549 +Note also that the client's command-line option, bf(--checksum-updating),
550 +has no effect on a daemon. A daemon will only update/create checksum files
551 +if this config option is true.
553 dit(bf(read only)) The "read only" option determines whether clients
554 will be able to upload files or not. If "read only" is true then any
555 attempted uploads will fail. If "read only" is false then uploads will