1 This builds on the sender optimization feature of the checksum-reading
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/checksum-reading.diff
8 patch -p1 <patches/checksum-updating.diff
9 ./configure (optional if already run)
12 diff --git a/clientserver.c b/clientserver.c
15 @@ -38,6 +38,7 @@ extern int sanitize_paths;
16 extern int filesfrom_fd;
17 extern int remote_protocol;
18 extern int protocol_version;
19 +extern int checksum_updating;
20 extern int io_timeout;
22 extern int write_batch;
23 @@ -704,6 +705,8 @@ static int rsync_module(int f_in, int f_out, int i, char *addr, char *host)
24 else if (am_root < 0) /* Treat --fake-super from client as --super. */
27 + checksum_updating = lp_checksum_updating(i);
29 if (filesfrom_fd == 0)
32 diff --git a/flist.c b/flist.c
43 @@ -58,6 +59,7 @@ extern int implied_dirs;
44 extern int file_extra_cnt;
45 extern int ignore_perishable;
46 extern int non_perishable_cnt;
47 +extern int checksum_updating;
48 extern int prune_empty_dirs;
49 extern int copy_links;
50 extern int copy_unsafe_links;
51 @@ -83,6 +85,9 @@ extern iconv_t ic_send, ic_recv;
53 #define PTR_SIZE (sizeof (struct file_struct *))
55 +#define FLAG_SUM_MISSING (1<<1) /* F_SUM() data is undefined */
56 +#define FLAG_SUM_KEEP (1<<2) /* keep entry when rewriting */
60 dev_t filesystem_dev; /* used to implement -x */
61 @@ -121,6 +126,9 @@ static char tmp_sum[MAX_DIGEST_LEN];
62 static char empty_sum[MAX_DIGEST_LEN];
63 static int flist_count_offset; /* for --delete --progress */
64 static int dir_count = 0;
65 +static int checksum_matches = 0;
66 +static int checksum_updates = 0;
67 +static int regular_skipped = 0;
68 static struct file_list *checksum_flist = NULL;
70 static void flist_sort_and_clean(struct file_list *flist, int strip_root);
71 @@ -316,7 +324,8 @@ static void flist_done_allocating(struct file_list *flist)
73 /* The len count is the length of the basename + 1 for the null. */
74 static int add_checksum(const char *dirname, const char *basename, int len,
75 - OFF_T file_length, time_t mtime, const char *sum)
76 + OFF_T file_length, time_t mtime, int32 ctime, int32 inode,
77 + const char *sum, const char *alt_sum, int flags)
79 struct file_struct *file;
80 int alloc_len, extra_len;
81 @@ -327,13 +336,14 @@ static int add_checksum(const char *dirname, const char *basename, int len,
85 - extra_len = (file_extra_cnt + (file_length > 0xFFFFFFFFu) + SUM_EXTRA_CNT)
86 + /* "2" is for a 32-bit ctime num and an 32-bit inode num. */
87 + extra_len = (file_extra_cnt + (file_length > 0xFFFFFFFFu) + SUM_EXTRA_CNT + 2)
89 #if EXTRA_ROUNDING > 0
90 if (extra_len & (EXTRA_ROUNDING * EXTRA_LEN))
91 extra_len = (extra_len | (EXTRA_ROUNDING * EXTRA_LEN)) + EXTRA_LEN;
93 - alloc_len = FILE_STRUCT_LEN + extra_len + len;
94 + alloc_len = FILE_STRUCT_LEN + extra_len + len + checksum_len*2 + 1;
95 bp = pool_alloc(checksum_flist->file_pool, alloc_len, "add_checksum");
97 memset(bp, 0, extra_len + FILE_STRUCT_LEN);
98 @@ -342,7 +352,14 @@ static int add_checksum(const char *dirname, const char *basename, int len,
99 bp += FILE_STRUCT_LEN;
101 memcpy(bp, basename, len);
103 + strlcpy(bp+len, alt_sum, checksum_len*2 + 1);
105 + memset(bp+len, '=', checksum_len*2);
106 + bp[len+checksum_len*2] = '\0';
109 + file->flags = flags;
110 file->mode = S_IFREG;
111 file->modtime = mtime;
112 file->len32 = (uint32)file_length;
113 @@ -353,6 +370,8 @@ static int add_checksum(const char *dirname, const char *basename, int len,
114 file->dirname = dirname;
116 memcpy(bp, sum, checksum_len);
117 + F_CTIME(file) = ctime;
118 + F_INODE(file) = inode;
120 flist_expand(checksum_flist, 1);
121 checksum_flist->files[checksum_flist->used++] = file;
122 @@ -362,17 +381,104 @@ static int add_checksum(const char *dirname, const char *basename, int len,
126 +static void write_checksums(const char *next_dirname, int whole_dir)
128 + static const char *dirname_save;
129 + char fbuf[MAXPATHLEN];
130 + const char *dirname;
131 + int used, new_entries, counts_match, no_skipped;
135 + dirname = dirname_save;
136 + dirname_save = next_dirname;
141 + used = checksum_flist->used;
142 + new_entries = checksum_updates != 0;
143 + counts_match = used == checksum_matches;
144 + no_skipped = whole_dir && regular_skipped == 0;
146 + clean_flist(checksum_flist, 0);
148 + checksum_flist->used = 0;
149 + checksum_matches = 0;
150 + checksum_updates = 0;
151 + regular_skipped = 0;
157 + if (pathjoin(fbuf, sizeof fbuf, dirname, ".rsyncsums") >= sizeof fbuf)
160 + strlcpy(fbuf, ".rsyncsums", sizeof fbuf);
162 + if (checksum_flist->high - checksum_flist->low < 0 && no_skipped) {
167 + if (!new_entries && (counts_match || !whole_dir))
170 + if (!(out_fp = fopen(fbuf, "w")))
174 + for (i = checksum_flist->low; i <= checksum_flist->high; i++) {
175 + struct file_struct *file = checksum_flist->sorted[i];
176 + const char *cp = F_SUM(file);
177 + const char *end = cp + checksum_len;
178 + const char *alt_sum = file->basename + strlen(file->basename) + 1;
179 + int32 ctime, inode;
180 + if (whole_dir && !(file->flags & FLAG_SUM_KEEP))
182 + ctime = F_CTIME(file);
183 + inode = F_INODE(file);
184 + if (protocol_version >= 30)
185 + fprintf(out_fp, "%s ", alt_sum);
186 + if (file->flags & FLAG_SUM_MISSING) {
189 + fprintf(out_fp, "==");
190 + } while (++cp != end);
193 + fprintf(out_fp, "%02x", (int)CVAL(cp, 0));
194 + } while (++cp != end);
196 + if (protocol_version < 30)
197 + fprintf(out_fp, " %s", alt_sum);
198 + if (*alt_sum == '=')
200 + fprintf(out_fp, " %10.0f %10.0f %10lu %10lu %s\n",
201 + (double)F_LENGTH(file), (double)file->modtime,
202 + (long)ctime, (long)inode, file->basename);
208 /* The direname value must remain unchanged during the lifespan of the
209 * created checksum_flist object because we use it directly. */
210 static void read_checksums(const char *dirname)
212 char line[MAXPATHLEN+1024], fbuf[MAXPATHLEN], sum[MAX_DIGEST_LEN];
213 + const char *alt_sum = NULL;
217 + int32 ctime, inode;
218 + int len, dlen, i, flags;
222 + if (checksum_updating)
223 + write_checksums(dirname, 0);
225 if (checksum_flist) {
226 /* Reset the pool memory and empty the file-list array. */
227 pool_free_old(checksum_flist->file_pool,
228 @@ -383,6 +489,9 @@ static void read_checksums(const char *dirname)
230 checksum_flist->low = 0;
231 checksum_flist->high = -1;
232 + checksum_matches = 0;
233 + checksum_updates = 0;
234 + regular_skipped = 0;
238 @@ -401,7 +510,7 @@ static void read_checksums(const char *dirname)
239 while (fgets(line, sizeof line, fp)) {
241 if (protocol_version >= 30) {
242 - char *alt_sum = cp;
245 while (*++cp == '=') {}
247 @@ -412,7 +521,14 @@ static void read_checksums(const char *dirname)
252 + for (i = 0; i < checksum_len*2; i++, cp++) {
258 + memset(sum, 0, checksum_len);
259 + flags = FLAG_SUM_MISSING;
261 for (i = 0; i < checksum_len*2; i++, cp++) {
263 @@ -430,13 +546,14 @@ static void read_checksums(const char *dirname)
271 while (*++cp == ' ') {}
273 if (protocol_version < 30) {
274 - char *alt_sum = cp;
277 while (*++cp == '=') {}
279 @@ -460,16 +577,16 @@ static void read_checksums(const char *dirname)
281 while (*++cp == ' ') {}
283 - /* Ignore ctime. */
287 + ctime = ctime * 10 + *cp++ - '0';
290 while (*++cp == ' ') {}
292 - /* Ignore inode. */
296 + inode = inode * 10 + *cp++ - '0';
299 while (*++cp == ' ') {}
300 @@ -486,8 +603,13 @@ static void read_checksums(const char *dirname)
303 strlcpy(fbuf+dlen, cp, sizeof fbuf - dlen);
304 + if (is_excluded(fbuf, 0, ALL_FILTERS)) {
305 + flags |= FLAG_SUM_KEEP;
306 + checksum_matches++;
309 - add_checksum(dirname, cp, len, file_length, mtime, sum);
310 + add_checksum(dirname, cp, len, file_length, mtime, ctime, inode,
311 + sum, alt_sum, flags);
315 @@ -1272,6 +1394,8 @@ struct file_struct *make_file(const char *fname, struct file_list *flist,
316 if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level)) {
317 if (ignore_perishable)
318 non_perishable_cnt++;
319 + if (S_ISREG(st.st_mode))
324 @@ -1402,13 +1526,36 @@ struct file_struct *make_file(const char *fname, struct file_list *flist,
326 if (flist && (j = flist_find(checksum_flist, file)) >= 0) {
327 struct file_struct *fp = checksum_flist->sorted[j];
328 + int32 ctime = F_CTIME(fp);
329 + int32 inode = F_INODE(fp);
330 if (F_LENGTH(fp) == st.st_size
331 - && fp->modtime == st.st_mtime)
332 - memcpy(tmp_sum, F_SUM(fp), MAX_DIGEST_LEN);
334 - file_checksum(thisname, tmp_sum, st.st_size);
336 + && fp->modtime == st.st_mtime
337 + && ctime == (int32)st.st_ctime
338 + && inode == (int32)st.st_ino) {
339 + if (fp->flags & FLAG_SUM_MISSING) {
340 + fp->flags &= ~FLAG_SUM_MISSING;
341 + checksum_updates++;
342 + file_checksum(thisname, tmp_sum, st.st_size);
343 + memcpy(F_SUM(fp), tmp_sum, MAX_DIGEST_LEN);
345 + checksum_matches++;
346 + memcpy(tmp_sum, F_SUM(fp), MAX_DIGEST_LEN);
348 + fp->flags |= FLAG_SUM_KEEP;
351 + goto compute_new_checksum;
354 + compute_new_checksum:
355 file_checksum(thisname, tmp_sum, st.st_size);
356 + if (checksum_updating && flist) {
357 + checksum_updates +=
358 + add_checksum(file->dirname, basename, basename_len,
359 + st.st_size, st.st_mtime, st.st_ctime,
360 + st.st_ino, tmp_sum, NULL, FLAG_SUM_KEEP);
365 /* This code is only used by the receiver when it is building
366 @@ -1703,6 +1850,9 @@ static void send_directory(int f, struct file_list *flist, char *fbuf, int len,
370 + if (checksum_updating && always_checksum && am_sender && f >= 0)
371 + write_checksums(NULL, 1);
373 if (f >= 0 && recurse && !divert_dirs) {
374 int i, end = flist->used - 1;
375 /* send_if_directory() bumps flist->used, so use "end". */
376 @@ -2265,7 +2415,7 @@ struct file_list *send_file_list(int f, int argc, char *argv[])
379 if (checksum_updating && always_checksum && flist_eof)
380 - read_checksums(NULL);
381 + read_checksums(NULL); /* writes any last updates */
385 diff --git a/loadparm.c b/loadparm.c
388 @@ -153,6 +153,7 @@ typedef struct
392 + BOOL checksum_updating;
395 BOOL ignore_nonreadable;
396 @@ -202,6 +203,7 @@ static service sDefault =
397 /* syslog_facility; */ LOG_DAEMON,
400 + /* checksum_updating; */ False,
401 /* fake_super; */ False,
402 /* ignore_errors; */ False,
403 /* ignore_nonreadable; */ False,
404 @@ -301,6 +303,7 @@ static struct parm_struct parm_table[] =
405 {"socket options", P_STRING, P_GLOBAL,&Globals.socket_options, NULL,0},
407 {"auth users", P_STRING, P_LOCAL, &sDefault.auth_users, NULL,0},
408 + {"checksum updating", P_BOOL, P_LOCAL, &sDefault.checksum_updating, NULL,0},
409 {"comment", P_STRING, P_LOCAL, &sDefault.comment, NULL,0},
410 {"dont compress", P_STRING, P_LOCAL, &sDefault.dont_compress, NULL,0},
411 {"exclude from", P_STRING, P_LOCAL, &sDefault.exclude_from, NULL,0},
412 @@ -421,6 +424,7 @@ FN_LOCAL_INTEGER(lp_max_connections, max_connections)
413 FN_LOCAL_INTEGER(lp_max_verbosity, max_verbosity)
414 FN_LOCAL_INTEGER(lp_timeout, timeout)
416 +FN_LOCAL_BOOL(lp_checksum_updating, checksum_updating)
417 FN_LOCAL_BOOL(lp_fake_super, fake_super)
418 FN_LOCAL_BOOL(lp_ignore_errors, ignore_errors)
419 FN_LOCAL_BOOL(lp_ignore_nonreadable, ignore_nonreadable)
420 diff --git a/options.c b/options.c
423 @@ -112,6 +112,7 @@ size_t bwlimit_writemax = 0;
424 int ignore_existing = 0;
425 int ignore_non_existing = 0;
426 int need_messages_from_generator = 0;
427 +int checksum_updating = 0;
428 int max_delete = INT_MIN;
431 @@ -312,6 +313,7 @@ void usage(enum logcode F)
432 rprintf(F," -q, --quiet suppress non-error messages\n");
433 rprintf(F," --no-motd suppress daemon-mode MOTD (see manpage caveat)\n");
434 rprintf(F," -c, --checksum skip based on checksum, not mod-time & size\n");
435 + rprintf(F," --checksum-updating sender updates .rsyncsums files\n");
436 rprintf(F," -a, --archive archive mode; equals -rlptgoD (no -H,-A,-X)\n");
437 rprintf(F," --no-OPTION turn off an implied OPTION (e.g. --no-D)\n");
438 rprintf(F," -r, --recursive recurse into directories\n");
439 @@ -560,6 +562,7 @@ static struct poptOption long_options[] = {
440 {"checksum", 'c', POPT_ARG_VAL, &always_checksum, 1, 0, 0 },
441 {"no-checksum", 0, POPT_ARG_VAL, &always_checksum, 0, 0, 0 },
442 {"no-c", 0, POPT_ARG_VAL, &always_checksum, 0, 0, 0 },
443 + {"checksum-updating",0, POPT_ARG_NONE, &checksum_updating, 0, 0, 0 },
444 {"block-size", 'B', POPT_ARG_LONG, &block_size, 0, 0, 0 },
445 {"compare-dest", 0, POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
446 {"copy-dest", 0, POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
447 @@ -1965,7 +1968,9 @@ void server_options(char **args, int *argc_p)
448 args[ac++] = basis_dir[i];
452 + } else if (checksum_updating)
453 + args[ac++] = "--checksum-updating";
458 diff --git a/rsync.h b/rsync.h
461 @@ -678,6 +678,10 @@ extern int xattrs_ndx;
462 #define F_SUM(f) ((char*)OPT_EXTRA(f, LEN64_BUMP(f) + HLINK_BUMP(f) \
463 + SUM_EXTRA_CNT - 1))
465 +/* These are only valid on an entry read from a checksum file. */
466 +#define F_CTIME(f) OPT_EXTRA(f, LEN64_BUMP(f) + SUM_EXTRA_CNT)->num
467 +#define F_INODE(f) OPT_EXTRA(f, LEN64_BUMP(f) + SUM_EXTRA_CNT + 1)->num
469 /* Some utility defines: */
470 #define F_IS_ACTIVE(f) (f)->basename[0]
471 #define F_IS_HLINKED(f) ((f)->flags & FLAG_HLINKED)
472 diff --git a/rsync.yo b/rsync.yo
475 @@ -317,6 +317,7 @@ to the detailed description below for a complete description. verb(
476 -q, --quiet suppress non-error messages
477 --no-motd suppress daemon-mode MOTD (see caveat)
478 -c, --checksum skip based on checksum, not mod-time & size
479 + --checksum-updating sender updates .rsyncsums files
480 -a, --archive archive mode; equals -rlptgoD (no -H,-A,-X)
481 --no-OPTION turn off an implied OPTION (e.g. --no-D)
482 -r, --recursive recurse into directories
483 @@ -516,9 +517,9 @@ uses a "quick check" that (by default) checks if each file's size and time
484 of last modification match between the sender and receiver. This option
485 changes this to compare a 128-bit MD4 checksum for each file that has a
486 matching size. Generating the checksums means that both sides will expend
487 -a lot of disk I/O reading all the data in the files in the transfer (and
488 -this is prior to any reading that will be done to transfer changed files),
489 -so this can slow things down significantly.
490 +a lot of disk I/O reading the data in all the files in the transfer, so
491 +this can slow things down significantly (and this is prior to any reading
492 +that will be done to transfer the files that have changed).
494 The sending side generates its checksums while it is doing the file-system
495 scan that builds the list of the available files. The receiver generates
496 @@ -526,12 +527,42 @@ its checksums when it is scanning for changed files, and will checksum any
497 file that has the same size as the corresponding sender's file: files with
498 either a changed size or a changed checksum are selected for transfer.
500 +Starting with version 3.0.0, the sending side will look for a checksum
501 +summary file and use a pre-generated checksum that it reads out of the file
502 +(as long as it matches the file's size and modified time). This allows a
503 +server to support the --checksum option to clients without having to
504 +recompute the checksums for each client. See the bf(--checksum-updating)
505 +option for a way to have rsync create/update these checksum files.
507 Note that rsync always verifies that each em(transferred) file was
508 correctly reconstructed on the receiving side by checking a whole-file
509 checksum that is generated when as the file is transferred, but that
510 automatic after-the-transfer verification has nothing to do with this
511 option's before-the-transfer "Does this file need to be updated?" check.
513 +dit(bf(--checksum-updating)) This option tells the sending side to create
514 +and/or update per-directory checksum files that are used by the
515 +bf(--checksum) option. The file that is updated is named .rsyncsums. If
516 +pre-transfer checksums are not being computed, this option has no effect.
518 +The checksum files stores the computed checksum, last-known size,
519 +modification time, and name for each file in the current directory. If a
520 +later transfer finds that a file matches its prior size and modification
521 +time, the checksum is assumed to still be correct. Otherwise it is
522 +recomputed and udpated in the file.
524 +To avoid transferring the system's checksum files, you can use an exclude
525 +(e.g. bf(--exclude=.rsyncsums)). To make this easier to type, you can use
526 +a popt alias. For instance, adding the following line in your ~/.popt file
527 +defines a bf(-cc) option that enables checksum updating and excludes the
530 +verb( rsync alias --cc --checksum-updating --exclude=.rsyncsums)
532 +An rsync daemon does not allow the client to control this setting, so see
533 +the "checksum updating" daemon config option for information on how to make
534 +a daemon maintain these checksum files.
536 dit(bf(-a, --archive)) This is equivalent to bf(-rlptgoD). It is a quick
537 way of saying you want recursion and want to preserve almost
538 everything (with -H being a notable omission).
539 diff --git a/rsyncd.conf.yo b/rsyncd.conf.yo
542 @@ -237,6 +237,20 @@ locking on this file to ensure that the max connections limit is not
543 exceeded for the modules sharing the lock file.
544 The default is tt(/var/run/rsyncd.lock).
546 +dit(bf(checksum updating)) This option tells rsync to update/create the
547 +checksum information in the per-directory checksum files when users copy
548 +files using the bf(--checksum) option. Any file that has changed since it
549 +was last checksummed (or is not mentioned) has its data updated in the
552 +Note that this updating will occur even if the module is listed as being
553 +read-only. If you want to hide these files (and you will almost always
554 +want to do), add ".rsyncsums" to the module's exclude setting.
556 +Note also that the client's command-line option, bf(--checksum-updating),
557 +has no effect on a daemon. A daemon will only update/create checksum files
558 +if this config option is true.
560 dit(bf(read only)) The "read only" option determines whether clients
561 will be able to upload files or not. If "read only" is true then any
562 attempted uploads will fail. If "read only" is false then uploads will