1: Jason M. Felice wrote:
2:
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
6: file's name.
7:
8: To use this patch, run these commands for a successful build:
9:
10: patch -p1 <patches/link-by-hash.diff
11: ./prepare-source
12: ./configure
13: make
14:
15: based-on: e94bad1c156fc3910f24e2b3b71a81b0b0bdeb70
16: diff --git a/Makefile.in b/Makefile.in
17: --- a/Makefile.in
18: +++ b/Makefile.in
19: @@ -44,7 +44,7 @@ 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: usage.o fileio.o batch.o clientname.o chmod.o acls.o xattrs.o
23: -OBJS3=progress.o pipe.o @ASM@
24: +OBJS3=progress.o pipe.o hashlink.o @ASM@
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
29: --- a/checksum.c
30: +++ b/checksum.c
31: @@ -40,6 +40,8 @@ extern int whole_file;
32: extern int checksum_seed;
33: extern int protocol_version;
34: extern int proper_seed_order;
35: +extern char *link_by_hash_dir;
36: +extern char link_by_hash_extra_sum[MAX_DIGEST_LEN];
37: extern const char *checksum_choice;
38:
39: struct name_num_obj valid_checksums = {
40: @@ -444,7 +446,7 @@ static union {
41: MD4_CTX m4;
42: #endif
43: MD5_CTX m5;
44: -} ctx;
45: +} ctx, ctx2;
46: #ifdef SUPPORT_XXHASH
47: static XXH64_state_t* xxh64_state;
48: #endif
49: @@ -483,6 +485,8 @@ void sum_init(int csum_type, int seed)
50: #endif
51: case CSUM_MD5:
52: MD5_Init(&ctx.m5);
53: + if (link_by_hash_dir)
54: + MD5_Init(&ctx2.m5);
55: break;
56: case CSUM_MD4:
57: #ifdef USE_OPENSSL
58: @@ -533,6 +537,8 @@ void sum_update(const char *p, int32 len)
59: #endif
60: case CSUM_MD5:
61: MD5_Update(&ctx.m5, (uchar *)p, len);
62: + if (link_by_hash_dir)
63: + MD5_Update(&ctx2.m5, (uchar *)p, len);
64: break;
65: case CSUM_MD4:
66: #ifdef USE_OPENSSL
67: @@ -598,6 +604,8 @@ int sum_end(char *sum)
68: #endif
69: case CSUM_MD5:
70: MD5_Final((uchar *)sum, &ctx.m5);
71: + if (link_by_hash_dir)
72: + MD5_Final((uchar *)link_by_hash_extra_sum, &ctx2.m5);
73: break;
74: case CSUM_MD4:
75: #ifdef USE_OPENSSL
76: diff --git a/clientserver.c b/clientserver.c
77: --- a/clientserver.c
78: +++ b/clientserver.c
79: @@ -52,6 +52,7 @@ extern int logfile_format_has_i;
80: extern int logfile_format_has_o_or_i;
81: extern char *bind_address;
82: extern char *config_file;
83: +extern char *link_by_hash_dir;
84: extern char *logfile_format;
85: extern char *files_from;
86: extern char *tmpdir;
87: @@ -665,6 +666,9 @@ static int rsync_module(int f_in, int f_out, int i, const char *addr, const char
88: return -1;
89: }
90:
91: + if (*lp_link_by_hash_dir(i))
92: + link_by_hash_dir = lp_link_by_hash_dir(i);
93: +
94: if (am_daemon > 0) {
95: rprintf(FLOG, "rsync allowed access on module %s from %s (%s)\n",
96: name, host, addr);
97: diff --git a/daemon-parm.txt b/daemon-parm.txt
98: --- a/daemon-parm.txt
99: +++ b/daemon-parm.txt
100: @@ -29,6 +29,7 @@ STRING hosts_deny NULL
101: STRING include NULL
102: STRING include_from NULL
103: STRING incoming_chmod NULL
104: +STRING link_by_hash_dir NULL
105: STRING lock_file DEFAULT_LOCK_FILE
106: STRING log_file NULL
107: STRING log_format "%o %h [%a] %m (%u) %f %l"
108: diff --git a/hashlink.c b/hashlink.c
109: new file mode 100644
110: --- /dev/null
111: +++ b/hashlink.c
112: @@ -0,0 +1,92 @@
113: +/*
114: + Copyright (C) Cronosys, LLC 2004
115: +
116: + This program is free software; you can redistribute it and/or modify
117: + it under the terms of the GNU General Public License as published by
118: + the Free Software Foundation; either version 2 of the License, or
119: + (at your option) any later version.
120: +
121: + This program is distributed in the hope that it will be useful,
122: + but WITHOUT ANY WARRANTY; without even the implied warranty of
123: + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
124: + GNU General Public License for more details.
125: +
126: + You should have received a copy of the GNU General Public License
127: + along with this program; if not, write to the Free Software
128: + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
129: +*/
130: +
131: +/* This file contains code used by the --link-by-hash option. */
132: +
133: +#include "rsync.h"
134: +#include "inums.h"
135: +
136: +extern int protocol_version;
137: +extern char *link_by_hash_dir;
138: +extern char sender_file_sum[MAX_DIGEST_LEN];
139: +
140: +char link_by_hash_extra_sum[MAX_DIGEST_LEN]; /* Only used when md4 sums are in the transfer */
141: +
142: +#ifdef HAVE_LINK
143: +
144: +/* This function is always called after a file is received, so the
145: + * sender_file_sum buffer has whatever the last checksum was for the
146: + * transferred file. */
147: +void link_by_hash(const char *fname, const char *fnametmp, struct file_struct *file)
148: +{
149: + STRUCT_STAT st;
150: + char *hashname, *last_slash, *num_str;
151: + const char *hex;
152: + int num = 0;
153: +
154: + /* We don't bother to hard-link 0-length files. */
155: + if (F_LENGTH(file) == 0)
156: + return;
157: +
158: + hex = sum_as_hex(5, protocol_version >= 30 ? sender_file_sum : link_by_hash_extra_sum, 0);
159: + if (asprintf(&hashname, "%s/%.3s/%.3s/%.3s/%s.%s.000000",
160: + link_by_hash_dir, hex, hex+3, hex+6, hex+9, big_num(F_LENGTH(file))) < 0)
161: + {
162: + out_of_memory("make_hash_name");
163: + }
164: +
165: + last_slash = strrchr(hashname, '/');
166: + num_str = strrchr(last_slash, '.') + 1;
167: +
168: + while (1) {
169: + if (num >= 999999) { /* Surely we'll never reach this... */
170: + if (DEBUG_GTE(HASHLINK, 1))
171: + rprintf(FINFO, "link-by-hash: giving up after \"%s\".\n", hashname);
172: + goto cleanup;
173: + }
174: + if (num > 0 && DEBUG_GTE(HASHLINK, 1))
175: + rprintf(FINFO, "link-by-hash: max link count exceeded, starting new file \"%s\".\n", hashname);
176: +
177: + snprintf(num_str, 7, "%d", num++);
178: + if (do_stat(hashname, &st) < 0)
179: + break;
180: +
181: + if (do_link(hashname, fnametmp) < 0) {
182: + if (errno == EMLINK)
183: + continue;
184: + rsyserr(FERROR, errno, "link \"%s\" -> \"%s\"", hashname, full_fname(fname));
185: + } else {
186: + if (DEBUG_GTE(HASHLINK, 2))
187: + rprintf(FINFO, "link-by-hash (existing): \"%s\" -> %s\n", hashname, full_fname(fname));
188: + robust_rename(fnametmp, fname, NULL, 0644);
189: + }
190: +
191: + goto cleanup;
192: + }
193: +
194: + if (DEBUG_GTE(HASHLINK, 2))
195: + rprintf(FINFO, "link-by-hash (new): %s -> \"%s\"\n", full_fname(fname), hashname);
196: +
197: + if (do_link(fname, hashname) < 0
198: + && (errno != ENOENT || make_path(hashname, MKP_DROP_NAME) < 0 || do_link(fname, hashname) < 0))
199: + rsyserr(FERROR, errno, "link \"%s\" -> \"%s\"", full_fname(fname), hashname);
200: +
201: + cleanup:
202: + free(hashname);
203: +}
204: +#endif
205: diff --git a/options.c b/options.c
206: --- a/options.c
207: +++ b/options.c
208: @@ -164,6 +164,7 @@ char *backup_suffix = NULL;
209: char *tmpdir = NULL;
210: char *partial_dir = NULL;
211: char *basis_dir[MAX_BASIS_DIRS+1];
212: +char *link_by_hash_dir = NULL;
213: char *config_file = NULL;
214: char *shell_cmd = NULL;
215: char *logfile_name = NULL;
216: @@ -221,7 +222,7 @@ static const char *debug_verbosity[] = {
217: /*2*/ "BIND,CMD,CONNECT,DEL,DELTASUM,DUP,FILTER,FLIST,ICONV",
218: /*3*/ "ACL,BACKUP,CONNECT2,DELTASUM2,DEL2,EXIT,FILTER2,FLIST2,FUZZY,GENR,OWN,RECV,SEND,TIME",
219: /*4*/ "CMD2,DELTASUM3,DEL3,EXIT2,FLIST3,ICONV2,OWN2,PROTO,TIME2",
220: - /*5*/ "CHDIR,DELTASUM4,FLIST4,FUZZY2,HASH,HLINK",
221: + /*5*/ "CHDIR,DELTASUM4,FLIST4,FUZZY2,HASH,HASHLINK,HLINK",
222: };
223:
224: #define MAX_VERBOSITY ((int)(sizeof debug_verbosity / sizeof debug_verbosity[0]) - 1)
225: @@ -291,6 +292,7 @@ static struct output_struct debug_words[COUNT_DEBUG+1] = {
226: DEBUG_WORD(FUZZY, W_REC, "Debug fuzzy scoring (levels 1-2)"),
227: DEBUG_WORD(GENR, W_REC, "Debug generator functions"),
228: DEBUG_WORD(HASH, W_SND|W_REC, "Debug hashtable code"),
229: + DEBUG_WORD(HASHLINK, W_REC, "Debug hashlink code (levels 1-2)"),
230: DEBUG_WORD(HLINK, W_SND|W_REC, "Debug hard-link actions (levels 1-3)"),
231: DEBUG_WORD(ICONV, W_CLI|W_SRV, "Debug iconv character conversions (levels 1-2)"),
232: DEBUG_WORD(IO, W_CLI|W_SRV, "Debug I/O routines (levels 1-4)"),
233: @@ -573,7 +575,7 @@ enum {OPT_SERVER = 1000, OPT_DAEMON, OPT_SENDER, OPT_EXCLUDE, OPT_EXCLUDE_FROM,
234: OPT_INCLUDE, OPT_INCLUDE_FROM, OPT_MODIFY_WINDOW, OPT_MIN_SIZE, OPT_CHMOD,
235: OPT_READ_BATCH, OPT_WRITE_BATCH, OPT_ONLY_WRITE_BATCH, OPT_MAX_SIZE,
236: OPT_NO_D, OPT_APPEND, OPT_NO_ICONV, OPT_INFO, OPT_DEBUG, OPT_BLOCK_SIZE,
237: - OPT_USERMAP, OPT_GROUPMAP, OPT_CHOWN, OPT_BWLIMIT, OPT_STDERR,
238: + OPT_USERMAP, OPT_GROUPMAP, OPT_CHOWN, OPT_BWLIMIT, OPT_STDERR, OPT_LINK_BY_HASH,
239: OPT_OLD_COMPRESS, OPT_NEW_COMPRESS, OPT_NO_COMPRESS,
240: OPT_STOP_AFTER, OPT_STOP_AT,
241: OPT_REFUSED_BASE = 9000};
242: @@ -733,6 +735,7 @@ static struct poptOption long_options[] = {
243: {"compare-dest", 0, POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
244: {"copy-dest", 0, POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
245: {"link-dest", 0, POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
246: + {"link-by-hash", 0, POPT_ARG_STRING, 0, OPT_LINK_BY_HASH, 0, 0},
247: {"fuzzy", 'y', POPT_ARG_NONE, 0, 'y', 0, 0 },
248: {"no-fuzzy", 0, POPT_ARG_VAL, &fuzzy_basis, 0, 0, 0 },
249: {"no-y", 0, POPT_ARG_VAL, &fuzzy_basis, 0, 0, 0 },
250: @@ -972,6 +975,9 @@ static void set_refuse_options(void)
251: ref = cp + 1;
252: }
253:
254: + if (*lp_link_by_hash_dir(module_id))
255: + parse_one_refuse_match(0, "link-by-hash", list_end);
256: +
257: if (am_daemon) {
258: #ifdef ICONV_OPTION
259: if (!*lp_charset(module_id))
260: @@ -1834,6 +1840,20 @@ int parse_arguments(int *argc_p, const char ***argv_p)
261: return 0;
262: #endif
263:
264: + case OPT_LINK_BY_HASH:
265: +#ifdef HAVE_LINK
266: + arg = poptGetOptArg(pc);
267: + if (sanitize_paths)
268: + arg = sanitize_path(NULL, arg, NULL, 0, SP_DEFAULT);
269: + link_by_hash_dir = (char *)arg;
270: + break;
271: +#else
272: + snprintf(err_buf, sizeof err_buf,
273: + "hard links are not supported on this %s\n",
274: + am_server ? "server" : "client");
275: + return 0;
276: +#endif
277: +
278: case OPT_STOP_AFTER: {
279: long val;
280: arg = poptGetOptArg(pc);
281: @@ -2186,6 +2206,8 @@ int parse_arguments(int *argc_p, const char ***argv_p)
282: tmpdir = sanitize_path(NULL, tmpdir, NULL, 0, SP_DEFAULT);
283: if (backup_dir)
284: backup_dir = sanitize_path(NULL, backup_dir, NULL, 0, SP_DEFAULT);
285: + if (link_by_hash_dir)
286: + link_by_hash_dir = sanitize_path(NULL, link_by_hash_dir, NULL, 0, SP_DEFAULT);
287: }
288: if (daemon_filter_list.head && !am_sender) {
289: filter_rule_list *elp = &daemon_filter_list;
290: @@ -2870,6 +2892,12 @@ void server_options(char **args, int *argc_p)
291: } else if (inplace)
292: args[ac++] = "--inplace";
293:
294: + if (link_by_hash_dir && am_sender) {
295: + args[ac++] = "--link-by-hash";
296: + args[ac++] = link_by_hash_dir;
297: + link_by_hash_dir = NULL; /* optimize sending-side checksums */
298: + }
299: +
300: if (files_from && (!am_sender || filesfrom_host)) {
301: if (filesfrom_host) {
302: args[ac++] = "--files-from";
303: diff --git a/rsync.1.md b/rsync.1.md
304: --- a/rsync.1.md
305: +++ b/rsync.1.md
306: @@ -424,6 +424,7 @@ detailed description below for a complete description.
307: --compare-dest=DIR also compare destination files relative to DIR
308: --copy-dest=DIR ... and include copies of unchanged files
309: --link-dest=DIR hardlink to files in DIR when unchanged
310: +--link-by-hash=DIR create hardlinks by hash into DIR
311: --compress, -z compress file data during the transfer
312: --compress-choice=STR choose the compression algorithm (aka --zc)
313: --compress-level=NUM explicitly set compression level (aka --zl)
314: @@ -2331,6 +2332,50 @@ your home directory (remove the '=' for that).
315: specified (or implied by `-a`). You can work-around this bug by avoiding
316: the `-o` option when sending to an old rsync.
317:
318: +0. `--link-by-hash=DIR`
319: +
320: + This option hard links the destination files into `DIR`, a link farm
321: + arranged by MD5 file hash. The result is that the system will only store
322: + (usually) one copy of the unique contents of each file, regardless of the
323: + file's name (it will use extra files if the links overflow the available
324: + maximum).
325: +
326: + This patch does not take into account file permissions, extended
327: + attributes, or ACLs when linking things together, so you should only use
328: + this if you don't care about preserving those extra file attributes (or if
329: + they are always the same for identical files).
330: +
331: + The DIR is relative to the destination directory, so either specify a full
332: + path to the hash hierarchy, or specify a relative path that puts the links
333: + outside the destination (e.g. "../links").
334: +
335: + Keep in mind that the hierarchy is never pruned, so if you need to reclaim
336: + space, you should remove any files that have just one link (since they are
337: + not linked into any destination dirs anymore):
338: +
339: + > find $DIR -links 1 -delete
340: +
341: + The link farm's directory hierarchy is determined by the file's (32-char)
342: + MD5 hash and the file-length. The hash is split up into directory shards.
343: + For example, if a file is 54321 bytes long, it could be stored like this:
344: +
345: + > $DIR/123/456/789/01234567890123456789012.54321.0
346: +
347: + Note that the directory layout in this patch was modified for version
348: + 3.1.0, so anyone using an older version of this patch should move their
349: + existing link hierarchy out of the way and then use the newer rsync to copy
350: + the saved hierarchy into its new layout. Assuming that no files have
351: + overflowed their link limits, this would work:
352: +
353: + > mv $DIR $DIR.old
354: + > rsync -aiv --link-by-hash=$DIR $DIR.old/ $DIR.tmp/
355: + > rm -rf $DIR.tmp
356: + > rm -rf $DIR.old
357: +
358: + If some of your files are at their link limit, you'd be better of using a
359: + script to calculate the md5 sum of each file in the hierarchy and move it
360: + to its new location.
361: +
362: 0. `--compress`, `-z`
363:
364: With this option, rsync compresses the file data as it is sent to the
365: diff --git a/rsync.c b/rsync.c
366: --- a/rsync.c
367: +++ b/rsync.c
368: @@ -50,6 +50,7 @@ extern int flist_eof;
369: extern int file_old_total;
370: extern int keep_dirlinks;
371: extern int make_backups;
372: +extern char *link_by_hash_dir;
373: extern int sanitize_paths;
374: extern struct file_list *cur_flist, *first_flist, *dir_flist;
375: extern struct chmod_mode_struct *daemon_chmod_modes;
376: @@ -748,6 +749,10 @@ int finish_transfer(const char *fname, const char *fnametmp,
377: }
378: if (ret == 0) {
379: /* The file was moved into place (not copied), so it's done. */
380: +#ifdef HAVE_LINK
381: + if (link_by_hash_dir)
382: + link_by_hash(fname, fnametmp, file);
383: +#endif
384: return 1;
385: }
386: /* The file was copied, so tweak the perms of the copied file. If it
387: diff --git a/rsync.h b/rsync.h
388: --- a/rsync.h
389: +++ b/rsync.h
390: @@ -1428,7 +1428,8 @@ extern short info_levels[], debug_levels[];
391: #define DEBUG_FUZZY (DEBUG_FLIST+1)
392: #define DEBUG_GENR (DEBUG_FUZZY+1)
393: #define DEBUG_HASH (DEBUG_GENR+1)
394: -#define DEBUG_HLINK (DEBUG_HASH+1)
395: +#define DEBUG_HASHLINK (DEBUG_HASH+1)
396: +#define DEBUG_HLINK (DEBUG_HASHLINK+1)
397: #define DEBUG_ICONV (DEBUG_HLINK+1)
398: #define DEBUG_IO (DEBUG_ICONV+1)
399: #define DEBUG_NSTR (DEBUG_IO+1)
400: diff --git a/rsyncd.conf.5.md b/rsyncd.conf.5.md
401: --- a/rsyncd.conf.5.md
402: +++ b/rsyncd.conf.5.md
403: @@ -354,6 +354,23 @@ the values of parameters. See the GLOBAL PARAMETERS section for more details.
404: is 0, which means no limit. A negative value disables the module. See
405: also the "lock file" parameter.
406:
407: +0. `link by hash dir`
408: +
409: + When the "link by hash dir" parameter is set to a non-empty string,
410: + received files will be hard linked into **DIR**, a link farm arranged by
411: + MD5 file hash. See the `--link-by-hash` option for a full explanation.
412: +
413: + The **DIR** must be accessible inside any chroot restrictions for the
414: + module, but can exist outside the transfer location if there is an
415: + inside-the-chroot path to the module (see "use chroot"). Note that a
416: + user-specified option does not allow this outside-the-transfer-area
417: + placement.
418: +
419: + If this parameter is set, it will disable the `--link-by-hash` command-line
420: + option for copies into the module.
421: +
422: +The default is for this parameter to be unset.
423: +
424: 0. `log file`
425:
426: When the "log file" parameter is set to a non-empty string, the rsync
427: diff -Nurp a/rsync.1 b/rsync.1
428: --- a/rsync.1
429: +++ b/rsync.1
430: @@ -500,6 +500,7 @@ detailed description below for a complet
431: --compare-dest=DIR also compare destination files relative to DIR
432: --copy-dest=DIR ... and include copies of unchanged files
433: --link-dest=DIR hardlink to files in DIR when unchanged
434: +--link-by-hash=DIR create hardlinks by hash into DIR
435: --compress, -z compress file data during the transfer
436: --compress-choice=STR choose the compression algorithm (aka --zc)
437: --compress-level=NUM explicitly set compression level (aka --zl)
438: @@ -2372,6 +2373,60 @@ Note that rsync versions prior to 2.6.1
439: \fB\-\-link-dest\fP from working properly for a non-super-user when \fB\-o\fP was
440: specified (or implied by \fB\-a\fP). You can work-around this bug by avoiding
441: the \fB\-o\fP option when sending to an old rsync.
442: +.IP "\fB\-\-link-by-hash=DIR\fP"
443: +This option hard links the destination files into \fBDIR\fP, a link farm
444: +arranged by MD5 file hash. The result is that the system will only store
445: +(usually) one copy of the unique contents of each file, regardless of the
446: +file's name (it will use extra files if the links overflow the available
447: +maximum).
448: +.IP
449: +This patch does not take into account file permissions, extended
450: +attributes, or ACLs when linking things together, so you should only use
451: +this if you don't care about preserving those extra file attributes (or if
452: +they are always the same for identical files).
453: +.IP
454: +The DIR is relative to the destination directory, so either specify a full
455: +path to the hash hierarchy, or specify a relative path that puts the links
456: +outside the destination (e.g. "../links").
457: +.IP
458: +Keep in mind that the hierarchy is never pruned, so if you need to reclaim
459: +space, you should remove any files that have just one link (since they are
460: +not linked into any destination dirs anymore):
461: +.RS 4
462: +.IP
463: +.nf
464: +find $DIR -links 1 -delete
465: +.fi
466: +.RE
467: +.IP
468: +The link farm's directory hierarchy is determined by the file's (32-char)
469: +MD5 hash and the file-length. The hash is split up into directory shards.
470: +For example, if a file is 54321 bytes long, it could be stored like this:
471: +.RS 4
472: +.IP
473: +.nf
474: +$DIR/123/456/789/01234567890123456789012.54321.0
475: +.fi
476: +.RE
477: +.IP
478: +Note that the directory layout in this patch was modified for version
479: +3.1.0, so anyone using an older version of this patch should move their
480: +existing link hierarchy out of the way and then use the newer rsync to copy
481: +the saved hierarchy into its new layout. Assuming that no files have
482: +overflowed their link limits, this would work:
483: +.RS 4
484: +.IP
485: +.nf
486: +mv $DIR $DIR.old
487: +rsync -aiv --link-by-hash=$DIR $DIR.old/ $DIR.tmp/
488: +rm -rf $DIR.tmp
489: +rm -rf $DIR.old
490: +.fi
491: +.RE
492: +.IP
493: +If some of your files are at their link limit, you'd be better of using a
494: +script to calculate the md5 sum of each file in the hierarchy and move it
495: +to its new location.
496: .IP "\fB\-\-compress\fP, \fB\-z\fP"
497: With this option, rsync compresses the file data as it is sent to the
498: destination machine, which reduces the amount of data being transmitted\ \-\-
499: diff -Nurp a/rsync.1.html b/rsync.1.html
500: --- a/rsync.1.html
501: +++ b/rsync.1.html
502: @@ -415,6 +415,7 @@ detailed description below for a complet
503: --compare-dest=DIR also compare destination files relative to DIR
504: --copy-dest=DIR ... and include copies of unchanged files
505: --link-dest=DIR hardlink to files in DIR when unchanged
506: +--link-by-hash=DIR create hardlinks by hash into DIR
507: --compress, -z compress file data during the transfer
508: --compress-choice=STR choose the compression algorithm (aka --zc)
509: --compress-level=NUM explicitly set compression level (aka --zl)
510: @@ -2210,6 +2211,50 @@ specified (or implied by <code>-a</code>
511: the <code>-o</code> option when sending to an old rsync.</p>
512: </dd>
513:
514: +<dt><code>--link-by-hash=DIR</code></dt><dd>
515: +<p>This option hard links the destination files into <code>DIR</code>, a link farm
516: +arranged by MD5 file hash. The result is that the system will only store
517: +(usually) one copy of the unique contents of each file, regardless of the
518: +file's name (it will use extra files if the links overflow the available
519: +maximum).</p>
520: +<p>This patch does not take into account file permissions, extended
521: +attributes, or ACLs when linking things together, so you should only use
522: +this if you don't care about preserving those extra file attributes (or if
523: +they are always the same for identical files).</p>
524: +<p>The DIR is relative to the destination directory, so either specify a full
525: +path to the hash hierarchy, or specify a relative path that puts the links
526: +outside the destination (e.g. "../links").</p>
527: +<p>Keep in mind that the hierarchy is never pruned, so if you need to reclaim
528: +space, you should remove any files that have just one link (since they are
529: +not linked into any destination dirs anymore):</p>
530: +<blockquote>
531: +<pre><code>find $DIR -links 1 -delete
532: +</code></pre>
533: +</blockquote>
534: +<p>The link farm's directory hierarchy is determined by the file's (32-char)
535: +MD5 hash and the file-length. The hash is split up into directory shards.
536: +For example, if a file is 54321 bytes long, it could be stored like this:</p>
537: +<blockquote>
538: +<pre><code>$DIR/123/456/789/01234567890123456789012.54321.0
539: +</code></pre>
540: +</blockquote>
541: +<p>Note that the directory layout in this patch was modified for version
542: +3.1.0, so anyone using an older version of this patch should move their
543: +existing link hierarchy out of the way and then use the newer rsync to copy
544: +the saved hierarchy into its new layout. Assuming that no files have
545: +overflowed their link limits, this would work:</p>
546: +<blockquote>
547: +<pre><code>mv $DIR $DIR.old
548: +rsync -aiv --link-by-hash=$DIR $DIR.old/ $DIR.tmp/
549: +rm -rf $DIR.tmp
550: +rm -rf $DIR.old
551: +</code></pre>
552: +</blockquote>
553: +<p>If some of your files are at their link limit, you'd be better of using a
554: +script to calculate the md5 sum of each file in the hierarchy and move it
555: +to its new location.</p>
556: +</dd>
557: +
558: <dt><code>--compress</code>, <code>-z</code></dt><dd>
559: <p>With this option, rsync compresses the file data as it is sent to the
560: destination machine, which reduces the amount of data being transmitted -⁠-⁠
561: diff -Nurp a/rsyncd.conf.5 b/rsyncd.conf.5
562: --- a/rsyncd.conf.5
563: +++ b/rsyncd.conf.5
564: @@ -335,6 +335,22 @@ connections you will allow. Any clients
565: been reached will receive a message telling them to try later. The default
566: is 0, which means no limit. A negative value disables the module. See
567: also the "lock file" parameter.
568: +.IP "\fBlink\ by\ hash\ dir\fP"
569: +When the "link by hash dir" parameter is set to a non-empty string,
570: +received files will be hard linked into \fBDIR\fP, a link farm arranged by
571: +MD5 file hash. See the \fB\-\-link-by-hash\fP option for a full explanation.
572: +.IP
573: +The \fBDIR\fP must be accessible inside any chroot restrictions for the
574: +module, but can exist outside the transfer location if there is an
575: +inside-the-chroot path to the module (see "use chroot"). Note that a
576: +user-specified option does not allow this outside-the-transfer-area
577: +placement.
578: +.IP
579: +If this parameter is set, it will disable the \fB\-\-link-by-hash\fP command-line
580: +option for copies into the module.
581: +.P
582: +The default is for this parameter to be unset.
583: +.P
584: .IP "\fBlog\ file\fP"
585: When the "log file" parameter is set to a non-empty string, the rsync
586: daemon will log messages to the indicated file rather than using syslog.
587: diff -Nurp a/rsyncd.conf.5.html b/rsyncd.conf.5.html
588: --- a/rsyncd.conf.5.html
589: +++ b/rsyncd.conf.5.html
590: @@ -342,6 +342,22 @@ is 0, which means no limit. A negative
591: also the "lock file" parameter.</p>
592: </dd>
593:
594: +<dt><code>link by hash dir</code></dt><dd>
595: +<p>When the "link by hash dir" parameter is set to a non-empty string,
596: +received files will be hard linked into <strong>DIR</strong>, a link farm arranged by
597: +MD5 file hash. See the <code>--link-by-hash</code> option for a full explanation.</p>
598: +<p>The <strong>DIR</strong> must be accessible inside any chroot restrictions for the
599: +module, but can exist outside the transfer location if there is an
600: +inside-the-chroot path to the module (see "use chroot"). Note that a
601: +user-specified option does not allow this outside-the-transfer-area
602: +placement.</p>
603: +<p>If this parameter is set, it will disable the <code>--link-by-hash</code> command-line
604: +option for copies into the module.</p>
605: +</dd>
606: +</dl>
607: +<p>The default is for this parameter to be unset.</p>
608: +<dl>
609: +
610: <dt><code>log file</code></dt><dd>
611: <p>When the "log file" parameter is set to a non-empty string, the rsync
612: daemon will log messages to the indicated file rather than using syslog.
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to link-by-hash.diff CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [ELWIX - Embedded LightWeight unIX -] / embedaddon / rsync / patches