3 This patch adds the --link-by-hash=DIR option, which hard links received files
4 in a link farm arranged by MD4 or MD5 file hash. The result is that the system
5 will only store one copy of the unique contents of each file, regardless of the
8 To use this patch, run these commands for a successful build:
10 patch -p1 <patches/link-by-hash.diff
15 based-on: 0488a14b9930bf91719ac0f1d1c0c8770ca10646
16 diff --git a/Makefile.in b/Makefile.in
19 @@ -37,7 +37,7 @@ OBJS1=flist.o rsync.o generator.o receiver.o cleanup.o sender.o exclude.o \
20 util.o util2.o main.o checksum.o match.o syscall.o log.o backup.o delete.o
21 OBJS2=options.o io.o compat.o hlink.o token.o uidlist.o socket.o hashtable.o \
22 fileio.o batch.o clientname.o chmod.o acls.o xattrs.o
23 -OBJS3=progress.o pipe.o
24 +OBJS3=progress.o pipe.o hashlink.o
25 DAEMON_OBJ = params.o loadparm.o clientserver.o access.o connection.o authenticate.o
26 popt_OBJS=popt/findme.o popt/popt.o popt/poptconfig.o \
27 popt/popthelp.o popt/poptparse.o
28 diff --git a/checksum.c b/checksum.c
35 +extern int checksum_len;
36 extern int checksum_seed;
37 extern int protocol_version;
38 +extern char *link_by_hash_dir;
39 +extern char link_by_hash_extra_sum[MAX_DIGEST_LEN];
42 a simple 32 bit checksum that can be upadted from either end
43 @@ -151,7 +154,7 @@ void file_checksum(char *fname, char *sum, OFF_T size)
46 static int32 sumresidue;
47 -static md_context md;
48 +static md_context md, md2;
50 void sum_init(int seed)
52 @@ -164,6 +167,8 @@ void sum_init(int seed)
56 + if (link_by_hash_dir)
61 @@ -182,6 +187,9 @@ void sum_update(const char *p, int32 len)
65 + if (link_by_hash_dir)
66 + md5_update(&md2, (uchar *)p, len);
68 if (len + sumresidue < CSUM_CHUNK) {
69 memcpy(md.buffer + sumresidue, p, len);
71 @@ -214,6 +222,9 @@ int sum_end(char *sum)
72 return MD5_DIGEST_LEN;
75 + if (link_by_hash_dir)
76 + md5_result(&md2, (uchar *)link_by_hash_extra_sum);
78 if (sumresidue || protocol_version >= 27)
79 mdfour_update(&md, (uchar *)md.buffer, sumresidue);
81 diff --git a/clientserver.c b/clientserver.c
84 @@ -50,6 +50,7 @@ extern int logfile_format_has_i;
85 extern int logfile_format_has_o_or_i;
86 extern char *bind_address;
87 extern char *config_file;
88 +extern char *link_by_hash_dir;
89 extern char *logfile_format;
90 extern char *files_from;
92 @@ -548,6 +549,9 @@ static int rsync_module(int f_in, int f_out, int i, const char *addr, const char
96 + if (*lp_link_by_hash_dir(i))
97 + link_by_hash_dir = lp_link_by_hash_dir(i);
99 if (am_daemon && am_server) {
100 rprintf(FLOG, "rsync allowed access on module %s from %s (%s)\n",
102 diff --git a/compat.c b/compat.c
105 @@ -55,6 +55,7 @@ extern char *partial_dir;
106 extern char *dest_option;
107 extern char *files_from;
108 extern char *filesfrom_host;
109 +extern char *link_by_hash_dir;
110 extern filter_rule_list filter_list;
111 extern int need_unsorted_flist;
113 diff --git a/hashlink.c b/hashlink.c
119 + Copyright (C) Cronosys, LLC 2004
121 + This program is free software; you can redistribute it and/or modify
122 + it under the terms of the GNU General Public License as published by
123 + the Free Software Foundation; either version 2 of the License, or
124 + (at your option) any later version.
126 + This program is distributed in the hope that it will be useful,
127 + but WITHOUT ANY WARRANTY; without even the implied warranty of
128 + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
129 + GNU General Public License for more details.
131 + You should have received a copy of the GNU General Public License
132 + along with this program; if not, write to the Free Software
133 + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
136 +/* This file contains code used by the --link-by-hash option. */
141 +extern int protocol_version;
142 +extern char *link_by_hash_dir;
143 +extern char sender_file_sum[MAX_DIGEST_LEN];
145 +char link_by_hash_extra_sum[MAX_DIGEST_LEN]; /* Only used when md4 sums are in the transfer */
149 +/* This function is always called after a file is received, so the
150 + * sender_file_sum buffer has whatever the last checksum was for the
151 + * transferred file. */
152 +void link_by_hash(const char *fname, const char *fnametmp, struct file_struct *file)
155 + char *hashname, *last_slash, *num_str;
159 + /* We don't bother to hard-link 0-length files. */
160 + if (F_LENGTH(file) == 0)
163 + hex = sum_as_hex(protocol_version >= 30 ? sender_file_sum : link_by_hash_extra_sum);
164 + if (asprintf(&hashname, "%s/%.3s/%.3s/%.3s/%s.%s.000000",
165 + link_by_hash_dir, hex, hex+3, hex+6, hex+9, big_num(F_LENGTH(file))) < 0)
167 + out_of_memory("make_hash_name");
170 + last_slash = strrchr(hashname, '/');
171 + num_str = strrchr(last_slash, '.') + 1;
174 + if (num >= 999999) { /* Surely we'll never reach this... */
175 + if (DEBUG_GTE(HASHLINK, 1))
176 + rprintf(FINFO, "link-by-hash: giving up after \"%s\".\n", hashname);
179 + if (num > 0 && DEBUG_GTE(HASHLINK, 1))
180 + rprintf(FINFO, "link-by-hash: max link count exceeded, starting new file \"%s\".\n", hashname);
182 + snprintf(num_str, 7, "%d", num++);
183 + if (do_stat(hashname, &st) < 0)
186 + if (do_link(hashname, fnametmp) < 0) {
187 + if (errno == EMLINK)
189 + rsyserr(FERROR, errno, "link \"%s\" -> \"%s\"", hashname, full_fname(fname));
191 + if (DEBUG_GTE(HASHLINK, 2))
192 + rprintf(FINFO, "link-by-hash (existing): \"%s\" -> %s\n", hashname, full_fname(fname));
193 + robust_rename(fnametmp, fname, NULL, 0644);
199 + if (DEBUG_GTE(HASHLINK, 2))
200 + rprintf(FINFO, "link-by-hash (new): %s -> \"%s\"\n", full_fname(fname), hashname);
202 + if (do_link(fname, hashname) < 0
203 + && (errno != ENOENT || make_path(hashname, MKP_DROP_NAME) < 0 || do_link(fname, hashname) < 0))
204 + rsyserr(FERROR, errno, "link \"%s\" -> \"%s\"", full_fname(fname), hashname);
210 diff --git a/loadparm.c b/loadparm.c
213 @@ -119,6 +119,7 @@ typedef struct {
216 char *incoming_chmod;
217 + char *link_by_hash_dir;
221 @@ -195,6 +196,7 @@ static const all_vars Defaults = {
223 /* include_from; */ NULL,
224 /* incoming_chmod; */ NULL,
225 + /* link_by_hash_dir; */ NULL,
226 /* lock_file; */ DEFAULT_LOCK_FILE,
227 /* log_file; */ NULL,
228 /* log_format; */ "%o %h [%a] %m (%u) %f %l",
229 @@ -336,6 +338,7 @@ static struct parm_struct parm_table[] =
230 {"include from", P_STRING, P_LOCAL, &Vars.l.include_from, NULL,0},
231 {"include", P_STRING, P_LOCAL, &Vars.l.include, NULL,0},
232 {"incoming chmod", P_STRING, P_LOCAL, &Vars.l.incoming_chmod, NULL,0},
233 + {"link by hash dir", P_STRING, P_LOCAL, &Vars.l.link_by_hash_dir, NULL,0},
234 {"list", P_BOOL, P_LOCAL, &Vars.l.list, NULL,0},
235 {"lock file", P_STRING, P_LOCAL, &Vars.l.lock_file, NULL,0},
236 {"log file", P_STRING, P_LOCAL, &Vars.l.log_file, NULL,0},
237 @@ -464,6 +467,7 @@ FN_LOCAL_STRING(lp_hosts_deny, hosts_deny)
238 FN_LOCAL_STRING(lp_include, include)
239 FN_LOCAL_STRING(lp_include_from, include_from)
240 FN_LOCAL_STRING(lp_incoming_chmod, incoming_chmod)
241 +FN_LOCAL_STRING(lp_link_by_hash_dir, link_by_hash_dir)
242 FN_LOCAL_STRING(lp_lock_file, lock_file)
243 FN_LOCAL_STRING(lp_log_file, log_file)
244 FN_LOCAL_STRING(lp_log_format, log_format)
245 diff --git a/options.c b/options.c
248 @@ -158,6 +158,7 @@ char *backup_suffix = NULL;
250 char *partial_dir = NULL;
251 char *basis_dir[MAX_BASIS_DIRS+1];
252 +char *link_by_hash_dir = NULL;
253 char *config_file = NULL;
254 char *shell_cmd = NULL;
255 char *logfile_name = NULL;
256 @@ -207,7 +208,7 @@ static const char *debug_verbosity[] = {
257 /*2*/ "BIND,CMD,CONNECT,DEL,DELTASUM,DUP,FILTER,FLIST,ICONV",
258 /*3*/ "ACL,BACKUP,CONNECT2,DELTASUM2,DEL2,EXIT,FILTER2,FLIST2,FUZZY,GENR,OWN,RECV,SEND,TIME",
259 /*4*/ "CMD2,DELTASUM3,DEL3,EXIT2,FLIST3,ICONV2,OWN2,PROTO,TIME2",
260 - /*5*/ "CHDIR,DELTASUM4,FLIST4,FUZZY2,HASH,HLINK",
261 + /*5*/ "CHDIR,DELTASUM4,FLIST4,FUZZY2,HASH,HASHLINK,HLINK",
264 #define MAX_VERBOSITY ((int)(sizeof debug_verbosity / sizeof debug_verbosity[0]) - 1)
265 @@ -277,6 +278,7 @@ static struct output_struct debug_words[COUNT_DEBUG+1] = {
266 DEBUG_WORD(FUZZY, W_REC, "Debug fuzzy scoring (levels 1-2)"),
267 DEBUG_WORD(GENR, W_REC, "Debug generator functions"),
268 DEBUG_WORD(HASH, W_SND|W_REC, "Debug hashtable code"),
269 + DEBUG_WORD(HASHLINK, W_REC, "Debug hashlink code (levels 1-2)"),
270 DEBUG_WORD(HLINK, W_SND|W_REC, "Debug hard-link actions (levels 1-3)"),
271 DEBUG_WORD(ICONV, W_CLI|W_SRV, "Debug iconv character conversions (levels 1-2)"),
272 DEBUG_WORD(IO, W_CLI|W_SRV, "Debug I/O routines (levels 1-4)"),
273 @@ -761,6 +763,7 @@ void usage(enum logcode F)
274 rprintf(F," --compare-dest=DIR also compare destination files relative to DIR\n");
275 rprintf(F," --copy-dest=DIR ... and include copies of unchanged files\n");
276 rprintf(F," --link-dest=DIR hardlink to files in DIR when unchanged\n");
277 + rprintf(F," --link-by-hash=DIR create hardlinks by hash into DIR\n");
278 rprintf(F," -z, --compress compress file data during the transfer\n");
279 rprintf(F," --compress-level=NUM explicitly set compression level\n");
280 rprintf(F," --skip-compress=LIST skip compressing files with a suffix in LIST\n");
281 @@ -817,7 +820,7 @@ enum {OPT_VERSION = 1000, OPT_DAEMON, OPT_SENDER, OPT_EXCLUDE, OPT_EXCLUDE_FROM,
282 OPT_FILTER, OPT_COMPARE_DEST, OPT_COPY_DEST, OPT_LINK_DEST, OPT_HELP,
283 OPT_INCLUDE, OPT_INCLUDE_FROM, OPT_MODIFY_WINDOW, OPT_MIN_SIZE, OPT_CHMOD,
284 OPT_READ_BATCH, OPT_WRITE_BATCH, OPT_ONLY_WRITE_BATCH, OPT_MAX_SIZE,
285 - OPT_NO_D, OPT_APPEND, OPT_NO_ICONV, OPT_INFO, OPT_DEBUG,
286 + OPT_NO_D, OPT_APPEND, OPT_NO_ICONV, OPT_INFO, OPT_DEBUG, OPT_LINK_BY_HASH,
287 OPT_USERMAP, OPT_GROUPMAP, OPT_CHOWN, OPT_BWLIMIT,
288 OPT_SERVER, OPT_REFUSED_BASE = 9000};
290 @@ -961,6 +964,7 @@ static struct poptOption long_options[] = {
291 {"compare-dest", 0, POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
292 {"copy-dest", 0, POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
293 {"link-dest", 0, POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
294 + {"link-by-hash", 0, POPT_ARG_STRING, 0, OPT_LINK_BY_HASH, 0, 0},
295 {"fuzzy", 'y', POPT_ARG_NONE, 0, 'y', 0, 0 },
296 {"no-fuzzy", 0, POPT_ARG_VAL, &fuzzy_basis, 0, 0, 0 },
297 {"no-y", 0, POPT_ARG_VAL, &fuzzy_basis, 0, 0, 0 },
298 @@ -1308,6 +1312,9 @@ int parse_arguments(int *argc_p, const char ***argv_p)
299 iconv_opt = strdup(arg);
302 + if (*lp_link_by_hash_dir(module_id))
303 + set_refuse_options("link-by-hash");
305 /* TODO: Call poptReadDefaultConfig; handle errors. */
307 /* The context leaks in case of an error, but if there's a
308 @@ -1794,6 +1801,21 @@ int parse_arguments(int *argc_p, const char ***argv_p)
312 + case OPT_LINK_BY_HASH:
314 + arg = poptGetOptArg(pc);
315 + if (sanitize_paths)
316 + arg = sanitize_path(NULL, arg, NULL, 0, SP_DEFAULT);
317 + link_by_hash_dir = (char *)arg;
320 + snprintf(err_buf, sizeof err_buf,
321 + "hard links are not supported on this %s\n",
322 + am_server ? "server" : "client");
323 + rprintf(FERROR, "ERROR: %s", err_buf);
328 /* A large opt value means that set_refuse_options()
329 * turned this option off. */
330 @@ -2069,6 +2091,8 @@ int parse_arguments(int *argc_p, const char ***argv_p)
331 tmpdir = sanitize_path(NULL, tmpdir, NULL, 0, SP_DEFAULT);
333 backup_dir = sanitize_path(NULL, backup_dir, NULL, 0, SP_DEFAULT);
334 + if (link_by_hash_dir)
335 + link_by_hash_dir = sanitize_path(NULL, link_by_hash_dir, NULL, 0, SP_DEFAULT);
337 if (daemon_filter_list.head && !am_sender) {
338 filter_rule_list *elp = &daemon_filter_list;
339 @@ -2714,6 +2738,12 @@ void server_options(char **args, int *argc_p)
341 args[ac++] = "--inplace";
343 + if (link_by_hash_dir && am_sender) {
344 + args[ac++] = "--link-by-hash";
345 + args[ac++] = link_by_hash_dir;
346 + link_by_hash_dir = NULL; /* optimize sending-side checksums */
349 if (files_from && (!am_sender || filesfrom_host)) {
350 if (filesfrom_host) {
351 args[ac++] = "--files-from";
352 diff --git a/rsync.c b/rsync.c
355 @@ -49,6 +49,7 @@ extern int flist_eof;
356 extern int file_old_total;
357 extern int keep_dirlinks;
358 extern int make_backups;
359 +extern char *link_by_hash_dir;
360 extern struct file_list *cur_flist, *first_flist, *dir_flist;
361 extern struct chmod_mode_struct *daemon_chmod_modes;
363 @@ -679,6 +680,10 @@ int finish_transfer(const char *fname, const char *fnametmp,
366 /* The file was moved into place (not copied), so it's done. */
368 + if (link_by_hash_dir)
369 + link_by_hash(fname, fnametmp, file);
373 /* The file was copied, so tweak the perms of the copied file. If it
374 diff --git a/rsync.h b/rsync.h
377 @@ -1263,7 +1263,8 @@ extern short info_levels[], debug_levels[];
378 #define DEBUG_FUZZY (DEBUG_FLIST+1)
379 #define DEBUG_GENR (DEBUG_FUZZY+1)
380 #define DEBUG_HASH (DEBUG_GENR+1)
381 -#define DEBUG_HLINK (DEBUG_HASH+1)
382 +#define DEBUG_HASHLINK (DEBUG_HASH+1)
383 +#define DEBUG_HLINK (DEBUG_HASHLINK+1)
384 #define DEBUG_ICONV (DEBUG_HLINK+1)
385 #define DEBUG_IO (DEBUG_ICONV+1)
386 #define DEBUG_OWN (DEBUG_IO+1)
387 diff --git a/rsync.yo b/rsync.yo
390 @@ -416,6 +416,7 @@ to the detailed description below for a complete description. verb(
391 --compare-dest=DIR also compare received files relative to DIR
392 --copy-dest=DIR ... and include copies of unchanged files
393 --link-dest=DIR hardlink to files in DIR when unchanged
394 + --link-by-hash=DIR create hardlinks by hash into DIR
395 -z, --compress compress file data during the transfer
396 --compress-level=NUM explicitly set compression level
397 --skip-compress=LIST skip compressing files with suffix in LIST
398 @@ -1849,6 +1850,48 @@ bf(--link-dest) from working properly for a non-super-user when bf(-o) was
399 specified (or implied by bf(-a)). You can work-around this bug by avoiding
400 the bf(-o) option when sending to an old rsync.
402 +dit(bf(--link-by-hash=DIR)) This option hard links the destination files into
403 +em(DIR), a link farm arranged by MD5 file hash. The result is that the system
404 +will only store (usually) one copy of the unique contents of each file,
405 +regardless of the file's name (it will use extra files if the links overflow
406 +the available maximum).
408 +This patch does not take into account file permissions, extended attributes,
409 +or ACLs when linking things together, so you should only use this if you
410 +don't care about preserving those extra file attributes (or if they are
411 +always the same for identical files).
413 +The DIR is relative to the destination directory, so either specify a full
414 +path to the hash hierarchy, or specify a relative path that puts the links
415 +outside the destination (e.g. "../links").
417 +Keep in mind that the hierarchy is never pruned, so if you need to reclaim
418 +space, you should remove any files that have just one link (since they are not
419 +linked into any destination dirs anymore):
421 + find $DIR -links 1 -delete
423 +The link farm's directory hierarchy is determined by the file's (32-char) MD5
424 +hash and the file-length. The hash is split up into directory shards. For
425 +example, if a file is 54321 bytes long, it could be stored like this:
427 + $DIR/123/456/789/01234567890123456789012.54321.0
429 +Note that the directory layout in this patch was modified for version 3.1.0,
430 +so anyone using an older version of this patch should move their existing
431 +link hierarchy out of the way and then use the newer rsync to copy the saved
432 +hierarchy into its new layout. Assuming that no files have overflowed their
433 +link limits, this would work:
436 + rsync -aiv --link-by-hash=$DIR $DIR.old/ $DIR.tmp/
440 +If some of your files are at their link limit, you'd be better of using a
441 +script to calculate the md5 sum of each file in the hierarchy and move it
442 +to its new location.
444 dit(bf(-z, --compress)) With this option, rsync compresses the file data
445 as it is sent to the destination machine, which reduces the amount of data
446 being transmitted -- something that is useful over a slow connection.
447 diff --git a/rsyncd.conf.yo b/rsyncd.conf.yo
450 @@ -283,6 +283,21 @@ message telling them to try later. The default is 0, which means no limit.
451 A negative value disables the module.
452 See also the "lock file" parameter.
454 +dit(bf(link by hash dir)) When the "link by hash dir" parameter is set to a
455 +non-empty string, received files will be hard linked into em(DIR), a link farm
456 +arranged by MD5 file hash. See the bf(--link-by-hash) option for a full
459 +The em(DIR) must be accessible inside any chroot restrictions for the module,
460 +but can exist outside the transfer location if there is an inside-the-chroot
461 +path to the module (see "use chroot"). Note that a user-specified option does
462 +not allow this outside-the-transfer-area placement.
464 +If this parameter is set, it will disable the bf(--link-by-hash) command-line
465 +option for copies into the module.
467 +The default is for this parameter to be unset.
469 dit(bf(log file)) When the "log file" parameter is set to a non-empty
470 string, the rsync daemon will log messages to the indicated file rather
471 than using syslog. This is particularly useful on systems (such as AIX)