- When --dirs is specified w/o --recursive, avoid marking solo dirs
[rsync/rsync.git] / flist.c
1 /*
2    Copyright (C) Andrew Tridgell 1996
3    Copyright (C) Paul Mackerras 1996
4    Copyright (C) 2001, 2002 by Martin Pool <mbp@samba.org>
5
6    This program is free software; you can redistribute it and/or modify
7    it under the terms of the GNU General Public License as published by
8    the Free Software Foundation; either version 2 of the License, or
9    (at your option) any later version.
10
11    This program is distributed in the hope that it will be useful,
12    but WITHOUT ANY WARRANTY; without even the implied warranty of
13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14    GNU General Public License for more details.
15
16    You should have received a copy of the GNU General Public License
17    along with this program; if not, write to the Free Software
18    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
21 /** @file flist.c
22  * Generate and receive file lists
23  *
24  * @sa http://lists.samba.org/pipermail/rsync/2000-June/002351.html
25  *
26  **/
27
28 #include "rsync.h"
29
30 extern int verbose;
31 extern int dry_run;
32 extern int list_only;
33 extern int am_root;
34 extern int am_server;
35 extern int am_daemon;
36 extern int am_sender;
37 extern int do_progress;
38 extern int always_checksum;
39 extern int module_id;
40 extern int ignore_errors;
41 extern int numeric_ids;
42 extern int recurse;
43 extern int xfer_dirs;
44 extern int filesfrom_fd;
45 extern int one_file_system;
46 extern int keep_dirlinks;
47 extern int preserve_links;
48 extern int preserve_hard_links;
49 extern int preserve_perms;
50 extern int preserve_devices;
51 extern int preserve_uid;
52 extern int preserve_gid;
53 extern int relative_paths;
54 extern int implied_dirs;
55 extern int copy_links;
56 extern int copy_unsafe_links;
57 extern int protocol_version;
58 extern int sanitize_paths;
59 extern int orig_umask;
60 extern struct stats stats;
61 extern struct file_list *the_file_list;
62
63 extern char curr_dir[MAXPATHLEN];
64
65 extern struct filter_list_struct filter_list;
66 extern struct filter_list_struct server_filter_list;
67
68 int io_error;
69 dev_t filesystem_dev; /* used to implement -x */
70
71 static char empty_sum[MD4_SUM_LENGTH];
72 static int flist_count_offset;
73 static unsigned int file_struct_len;
74 static struct file_list *sorting_flist;
75
76 static void clean_flist(struct file_list *flist, int strip_root, int no_dups);
77 static void output_flist(struct file_list *flist);
78
79 void init_flist(void)
80 {
81         struct file_struct f;
82
83         /* Figure out how big the file_struct is without trailing padding */
84         file_struct_len = offsetof(struct file_struct, flags) + sizeof f.flags;
85 }
86
87
88 static int show_filelist_p(void)
89 {
90         return verbose && xfer_dirs && !am_server;
91 }
92
93 static void start_filelist_progress(char *kind)
94 {
95         rprintf(FINFO, "%s ... ", kind);
96         if (verbose > 1 || do_progress)
97                 rprintf(FINFO, "\n");
98         rflush(FINFO);
99 }
100
101
102 static void emit_filelist_progress(int count)
103 {
104         rprintf(FINFO, " %d files...\r", count);
105 }
106
107
108 static void maybe_emit_filelist_progress(int count)
109 {
110         if (do_progress && show_filelist_p() && (count % 100) == 0)
111                 emit_filelist_progress(count);
112 }
113
114
115 static void finish_filelist_progress(const struct file_list *flist)
116 {
117         if (do_progress) {
118                 /* This overwrites the progress line */
119                 rprintf(FINFO, "%d file%sto consider\n",
120                         flist->count, flist->count == 1 ? " " : "s ");
121         } else
122                 rprintf(FINFO, "done\n");
123 }
124
125 void show_flist_stats(void)
126 {
127         /* Nothing yet */
128 }
129
130
131 static void list_file_entry(struct file_struct *f)
132 {
133         char perms[11];
134
135         if (!f->basename) {
136                 /* this can happen if duplicate names were removed */
137                 return;
138         }
139
140         permstring(perms, f->mode);
141
142 #ifdef SUPPORT_LINKS
143         if (preserve_links && S_ISLNK(f->mode)) {
144                 rprintf(FINFO, "%s %11.0f %s %s -> %s\n",
145                         perms,
146                         (double)f->length, timestring(f->modtime),
147                         safe_fname(f_name(f)), safe_fname(f->u.link));
148         } else
149 #endif
150         {
151                 rprintf(FINFO, "%s %11.0f %s %s\n",
152                         perms,
153                         (double)f->length, timestring(f->modtime),
154                         safe_fname(f_name(f)));
155         }
156 }
157
158
159 /**
160  * Stat either a symlink or its referent, depending on the settings of
161  * copy_links, copy_unsafe_links, etc.
162  *
163  * @retval -1 on error
164  *
165  * @retval 0 for success
166  *
167  * @post If @p path is a symlink, then @p linkbuf (of size @c
168  * MAXPATHLEN) contains the symlink target.
169  *
170  * @post @p buffer contains information about the link or the
171  * referrent as appropriate, if they exist.
172  **/
173 static int readlink_stat(const char *path, STRUCT_STAT *buffer, char *linkbuf)
174 {
175 #ifdef SUPPORT_LINKS
176         if (copy_links)
177                 return do_stat(path, buffer);
178         if (link_stat(path, buffer, 0) < 0)
179                 return -1;
180         if (S_ISLNK(buffer->st_mode)) {
181                 int l = readlink((char *)path, linkbuf, MAXPATHLEN - 1);
182                 if (l == -1)
183                         return -1;
184                 linkbuf[l] = 0;
185                 if (copy_unsafe_links && unsafe_symlink(linkbuf, path)) {
186                         if (verbose > 1) {
187                                 rprintf(FINFO,"copying unsafe symlink \"%s\" -> \"%s\"\n",
188                                         safe_fname(path), safe_fname(linkbuf));
189                         }
190                         return do_stat(path, buffer);
191                 }
192         }
193         return 0;
194 #else
195         return do_stat(path, buffer);
196 #endif
197 }
198
199 int link_stat(const char *path, STRUCT_STAT *buffer, int follow_dirlinks)
200 {
201 #ifdef SUPPORT_LINKS
202         if (copy_links)
203                 return do_stat(path, buffer);
204         if (do_lstat(path, buffer) < 0)
205                 return -1;
206         if (follow_dirlinks && S_ISLNK(buffer->st_mode)) {
207                 STRUCT_STAT st;
208                 if (do_stat(path, &st) == 0 && S_ISDIR(st.st_mode))
209                         *buffer = st;
210         }
211         return 0;
212 #else
213         return do_stat(path, buffer);
214 #endif
215 }
216
217 /* This function is used to check if a file should be included/excluded
218  * from the list of files based on its name and type etc.  The value of
219  * filter_level is set to either SERVER_FILTERS or ALL_FILTERS. */
220 static int is_excluded(char *fname, int is_dir, int filter_level)
221 {
222 #if 0 /* This currently never happens, so avoid a useless compare. */
223         if (filter_level == NO_FILTERS)
224                 return 0;
225 #endif
226         if (fname) {
227                 /* never exclude '.', even if somebody does --exclude '*' */
228                 if (fname[0] == '.' && !fname[1])
229                         return 0;
230                 /* Handle the -R version of the '.' dir. */
231                 if (fname[0] == '/') {
232                         int len = strlen(fname);
233                         if (fname[len-1] == '.' && fname[len-2] == '/')
234                                 return 0;
235                 }
236         }
237         if (server_filter_list.head
238             && check_filter(&server_filter_list, fname, is_dir) < 0)
239                 return 1;
240         if (filter_level != ALL_FILTERS)
241                 return 0;
242         if (filter_list.head
243             && check_filter(&filter_list, fname, is_dir) < 0)
244                 return 1;
245         return 0;
246 }
247
248 static int to_wire_mode(mode_t mode)
249 {
250 #ifdef SUPPORT_LINKS
251         if (S_ISLNK(mode) && (_S_IFLNK != 0120000))
252                 return (mode & ~(_S_IFMT)) | 0120000;
253 #endif
254         return (int)mode;
255 }
256
257 static mode_t from_wire_mode(int mode)
258 {
259         if ((mode & (_S_IFMT)) == 0120000 && (_S_IFLNK != 0120000))
260                 return (mode & ~(_S_IFMT)) | _S_IFLNK;
261         return (mode_t)mode;
262 }
263
264
265 static void send_directory(int f, struct file_list *flist,
266                            char *fbuf, int len);
267
268 static char *flist_dir;
269 static int flist_dir_len;
270
271
272 /**
273  * Make sure @p flist is big enough to hold at least @p flist->count
274  * entries.
275  **/
276 void flist_expand(struct file_list *flist)
277 {
278         struct file_struct **new_ptr;
279
280         if (flist->count < flist->malloced)
281                 return;
282
283         if (flist->malloced < FLIST_START)
284                 flist->malloced = FLIST_START;
285         else if (flist->malloced >= FLIST_LINEAR)
286                 flist->malloced += FLIST_LINEAR;
287         else
288                 flist->malloced *= 2;
289
290         /*
291          * In case count jumped or we are starting the list
292          * with a known size just set it.
293          */
294         if (flist->malloced < flist->count)
295                 flist->malloced = flist->count;
296
297         new_ptr = realloc_array(flist->files, struct file_struct *,
298                                 flist->malloced);
299
300         if (verbose >= 2 && flist->malloced != FLIST_START) {
301                 rprintf(FINFO, "[%s] expand file_list to %.0f bytes, did%s move\n",
302                     who_am_i(),
303                     (double)sizeof flist->files[0] * flist->malloced,
304                     (new_ptr == flist->files) ? " not" : "");
305         }
306
307         flist->files = new_ptr;
308
309         if (!flist->files)
310                 out_of_memory("flist_expand");
311 }
312
313 void send_file_entry(struct file_struct *file, int f, unsigned short base_flags)
314 {
315         unsigned short flags;
316         static time_t modtime;
317         static mode_t mode;
318         static int64 dev;
319         static dev_t rdev;
320         static uint32 rdev_major;
321         static uid_t uid;
322         static gid_t gid;
323         static char lastname[MAXPATHLEN];
324         char fname[MAXPATHLEN];
325         int l1, l2;
326
327         if (f < 0)
328                 return;
329
330         if (!file) {
331                 write_byte(f, 0);
332                 modtime = 0, mode = 0;
333                 dev = 0, rdev = makedev(0, 0);
334                 rdev_major = 0;
335                 uid = 0, gid = 0;
336                 *lastname = '\0';
337                 return;
338         }
339
340         io_write_phase = "send_file_entry";
341
342         f_name_to(file, fname);
343
344         flags = base_flags;
345
346         if (file->mode == mode)
347                 flags |= XMIT_SAME_MODE;
348         else
349                 mode = file->mode;
350         if (preserve_devices) {
351                 if (protocol_version < 28) {
352                         if (IS_DEVICE(mode)) {
353                                 if (file->u.rdev == rdev)
354                                         flags |= XMIT_SAME_RDEV_pre28;
355                                 else
356                                         rdev = file->u.rdev;
357                         } else
358                                 rdev = makedev(0, 0);
359                 } else if (IS_DEVICE(mode)) {
360                         rdev = file->u.rdev;
361                         if ((uint32)major(rdev) == rdev_major)
362                                 flags |= XMIT_SAME_RDEV_MAJOR;
363                         else
364                                 rdev_major = major(rdev);
365                         if ((uint32)minor(rdev) <= 0xFFu)
366                                 flags |= XMIT_RDEV_MINOR_IS_SMALL;
367                 }
368         }
369         if (file->uid == uid)
370                 flags |= XMIT_SAME_UID;
371         else
372                 uid = file->uid;
373         if (file->gid == gid)
374                 flags |= XMIT_SAME_GID;
375         else
376                 gid = file->gid;
377         if (file->modtime == modtime)
378                 flags |= XMIT_SAME_TIME;
379         else
380                 modtime = file->modtime;
381
382 #ifdef SUPPORT_HARD_LINKS
383         if (file->link_u.idev) {
384                 if (file->F_DEV == dev) {
385                         if (protocol_version >= 28)
386                                 flags |= XMIT_SAME_DEV;
387                 } else
388                         dev = file->F_DEV;
389                 flags |= XMIT_HAS_IDEV_DATA;
390         }
391 #endif
392
393         for (l1 = 0;
394             lastname[l1] && (fname[l1] == lastname[l1]) && (l1 < 255);
395             l1++) {}
396         l2 = strlen(fname+l1);
397
398         if (l1 > 0)
399                 flags |= XMIT_SAME_NAME;
400         if (l2 > 255)
401                 flags |= XMIT_LONG_NAME;
402
403         /* We must make sure we don't send a zero flag byte or the
404          * other end will terminate the flist transfer.  Note that
405          * the use of XMIT_TOP_DIR on a non-dir has no meaning, so
406          * it's harmless way to add a bit to the first flag byte. */
407         if (protocol_version >= 28) {
408                 if (!flags && !S_ISDIR(mode))
409                         flags |= XMIT_TOP_DIR;
410                 if ((flags & 0xFF00) || !flags) {
411                         flags |= XMIT_EXTENDED_FLAGS;
412                         write_byte(f, flags);
413                         write_byte(f, flags >> 8);
414                 } else
415                         write_byte(f, flags);
416         } else {
417                 if (!(flags & 0xFF))
418                         flags |= S_ISDIR(mode) ? XMIT_LONG_NAME : XMIT_TOP_DIR;
419                 write_byte(f, flags);
420         }
421         if (flags & XMIT_SAME_NAME)
422                 write_byte(f, l1);
423         if (flags & XMIT_LONG_NAME)
424                 write_int(f, l2);
425         else
426                 write_byte(f, l2);
427         write_buf(f, fname + l1, l2);
428
429         write_longint(f, file->length);
430         if (!(flags & XMIT_SAME_TIME))
431                 write_int(f, modtime);
432         if (!(flags & XMIT_SAME_MODE))
433                 write_int(f, to_wire_mode(mode));
434         if (preserve_uid && !(flags & XMIT_SAME_UID)) {
435                 if (!numeric_ids)
436                         add_uid(uid);
437                 write_int(f, uid);
438         }
439         if (preserve_gid && !(flags & XMIT_SAME_GID)) {
440                 if (!numeric_ids)
441                         add_gid(gid);
442                 write_int(f, gid);
443         }
444         if (preserve_devices && IS_DEVICE(mode)) {
445                 if (protocol_version < 28) {
446                         if (!(flags & XMIT_SAME_RDEV_pre28))
447                                 write_int(f, (int)rdev);
448                 } else {
449                         if (!(flags & XMIT_SAME_RDEV_MAJOR))
450                                 write_int(f, major(rdev));
451                         if (flags & XMIT_RDEV_MINOR_IS_SMALL)
452                                 write_byte(f, minor(rdev));
453                         else
454                                 write_int(f, minor(rdev));
455                 }
456         }
457
458 #ifdef SUPPORT_LINKS
459         if (preserve_links && S_ISLNK(mode)) {
460                 int len = strlen(file->u.link);
461                 write_int(f, len);
462                 write_buf(f, file->u.link, len);
463         }
464 #endif
465
466 #ifdef SUPPORT_HARD_LINKS
467         if (flags & XMIT_HAS_IDEV_DATA) {
468                 if (protocol_version < 26) {
469                         /* 32-bit dev_t and ino_t */
470                         write_int(f, dev);
471                         write_int(f, file->F_INODE);
472                 } else {
473                         /* 64-bit dev_t and ino_t */
474                         if (!(flags & XMIT_SAME_DEV))
475                                 write_longint(f, dev);
476                         write_longint(f, file->F_INODE);
477                 }
478         }
479 #endif
480
481         if (always_checksum && (S_ISREG(mode) || protocol_version < 28)) {
482                 char *sum;
483                 int slen = protocol_version < 21 ? 2 : MD4_SUM_LENGTH;
484                 if (S_ISREG(mode))
485                         sum = file->u.sum;
486                 else {
487                         /* Prior to 28, we sent a useless set of nulls. */
488                         sum = empty_sum;
489                 }
490                 write_buf(f, sum, slen);
491         }
492
493         strlcpy(lastname, fname, MAXPATHLEN);
494
495         io_write_phase = "unknown";
496 }
497
498
499
500 static struct file_struct *receive_file_entry(struct file_list *flist,
501                                               unsigned short flags, int f)
502 {
503         static time_t modtime;
504         static mode_t mode;
505         static int64 dev;
506         static dev_t rdev;
507         static uint32 rdev_major;
508         static uid_t uid;
509         static gid_t gid;
510         static char lastname[MAXPATHLEN], *lastdir;
511         static int lastdir_depth, lastdir_len = -1;
512         static unsigned int del_hier_name_len = 0;
513         static int in_del_hier = 0;
514         char thisname[MAXPATHLEN];
515         unsigned int l1 = 0, l2 = 0;
516         int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
517         OFF_T file_length;
518         char *basename, *dirname, *bp;
519         struct file_struct *file;
520
521         if (!flist) {
522                 modtime = 0, mode = 0;
523                 dev = 0, rdev = makedev(0, 0);
524                 rdev_major = 0;
525                 uid = 0, gid = 0;
526                 *lastname = '\0';
527                 lastdir_len = -1;
528                 in_del_hier = 0;
529                 return NULL;
530         }
531
532         if (flags & XMIT_SAME_NAME)
533                 l1 = read_byte(f);
534
535         if (flags & XMIT_LONG_NAME)
536                 l2 = read_int(f);
537         else
538                 l2 = read_byte(f);
539
540         if (l2 >= MAXPATHLEN - l1) {
541                 rprintf(FERROR,
542                         "overflow: flags=0x%x l1=%d l2=%d lastname=%s\n",
543                         flags, l1, l2, safe_fname(lastname));
544                 overflow_exit("receive_file_entry");
545         }
546
547         strlcpy(thisname, lastname, l1 + 1);
548         read_sbuf(f, &thisname[l1], l2);
549         thisname[l1 + l2] = 0;
550
551         strlcpy(lastname, thisname, MAXPATHLEN);
552
553         clean_fname(thisname, 0);
554
555         if (sanitize_paths)
556                 sanitize_path(thisname, thisname, "", 0);
557
558         if ((basename = strrchr(thisname, '/')) != NULL) {
559                 dirname_len = ++basename - thisname; /* counts future '\0' */
560                 if (lastdir_len == dirname_len - 1
561                     && strncmp(thisname, lastdir, lastdir_len) == 0) {
562                         dirname = lastdir;
563                         dirname_len = 0; /* indicates no copy is needed */
564                 } else
565                         dirname = thisname;
566         } else {
567                 basename = thisname;
568                 dirname = NULL;
569                 dirname_len = 0;
570         }
571         basename_len = strlen(basename) + 1; /* count the '\0' */
572
573         file_length = read_longint(f);
574         if (!(flags & XMIT_SAME_TIME))
575                 modtime = (time_t)read_int(f);
576         if (!(flags & XMIT_SAME_MODE))
577                 mode = from_wire_mode(read_int(f));
578
579         if (preserve_uid && !(flags & XMIT_SAME_UID))
580                 uid = (uid_t)read_int(f);
581         if (preserve_gid && !(flags & XMIT_SAME_GID))
582                 gid = (gid_t)read_int(f);
583
584         if (preserve_devices) {
585                 if (protocol_version < 28) {
586                         if (IS_DEVICE(mode)) {
587                                 if (!(flags & XMIT_SAME_RDEV_pre28))
588                                         rdev = (dev_t)read_int(f);
589                         } else
590                                 rdev = makedev(0, 0);
591                 } else if (IS_DEVICE(mode)) {
592                         uint32 rdev_minor;
593                         if (!(flags & XMIT_SAME_RDEV_MAJOR))
594                                 rdev_major = read_int(f);
595                         if (flags & XMIT_RDEV_MINOR_IS_SMALL)
596                                 rdev_minor = read_byte(f);
597                         else
598                                 rdev_minor = read_int(f);
599                         rdev = makedev(rdev_major, rdev_minor);
600                 }
601         }
602
603 #ifdef SUPPORT_LINKS
604         if (preserve_links && S_ISLNK(mode)) {
605                 linkname_len = read_int(f) + 1; /* count the '\0' */
606                 if (linkname_len <= 0 || linkname_len > MAXPATHLEN) {
607                         rprintf(FERROR, "overflow: linkname_len=%d\n",
608                                 linkname_len - 1);
609                         overflow_exit("receive_file_entry");
610                 }
611         }
612         else
613 #endif
614                 linkname_len = 0;
615
616         sum_len = always_checksum && S_ISREG(mode) ? MD4_SUM_LENGTH : 0;
617
618         alloc_len = file_struct_len + dirname_len + basename_len
619                   + linkname_len + sum_len;
620         bp = pool_alloc(flist->file_pool, alloc_len, "receive_file_entry");
621
622         file = (struct file_struct *)bp;
623         memset(bp, 0, file_struct_len);
624         bp += file_struct_len;
625
626         file->flags = 0;
627         file->modtime = modtime;
628         file->length = file_length;
629         file->mode = mode;
630         file->uid = uid;
631         file->gid = gid;
632
633         if (dirname_len) {
634                 file->dirname = lastdir = bp;
635                 lastdir_len = dirname_len - 1;
636                 memcpy(bp, dirname, dirname_len - 1);
637                 bp += dirname_len;
638                 bp[-1] = '\0';
639                 lastdir_depth = count_dir_elements(lastdir);
640                 file->dir.depth = lastdir_depth + 1;
641         } else if (dirname) {
642                 file->dirname = dirname; /* we're reusing lastname */
643                 file->dir.depth = lastdir_depth + 1;
644         } else
645                 file->dir.depth = 1;
646
647         if (S_ISDIR(mode)) {
648                 if (basename_len == 1+1 && *basename == '.') /* +1 for '\0' */
649                         file->dir.depth--;
650                 if (flags & XMIT_TOP_DIR) {
651                         in_del_hier = recurse;
652                         del_hier_name_len = file->dir.depth == 0 ? 0 : l1 + l2;
653                         if (relative_paths && del_hier_name_len > 2
654                             && basename_len == 1+1 && *basename == '.')
655                                 del_hier_name_len -= 2;
656                         file->flags |= FLAG_TOP_DIR | FLAG_DEL_HERE;
657                 } else if (in_del_hier) {
658                         if (!relative_paths || !del_hier_name_len
659                          || (l1 >= del_hier_name_len
660                           && thisname[del_hier_name_len] == '/'))
661                                 file->flags |= FLAG_DEL_HERE;
662                         else
663                                 in_del_hier = 0;
664                 }
665         }
666
667         file->basename = bp;
668         memcpy(bp, basename, basename_len);
669         bp += basename_len;
670
671         if (preserve_devices && IS_DEVICE(mode))
672                 file->u.rdev = rdev;
673
674 #ifdef SUPPORT_LINKS
675         if (linkname_len) {
676                 file->u.link = bp;
677                 read_sbuf(f, bp, linkname_len - 1);
678                 if (sanitize_paths)
679                         sanitize_path(bp, bp, "", lastdir_depth);
680                 bp += linkname_len;
681         }
682 #endif
683
684 #ifdef SUPPORT_HARD_LINKS
685         if (preserve_hard_links && protocol_version < 28 && S_ISREG(mode))
686                 flags |= XMIT_HAS_IDEV_DATA;
687         if (flags & XMIT_HAS_IDEV_DATA) {
688                 int64 inode;
689                 if (protocol_version < 26) {
690                         dev = read_int(f);
691                         inode = read_int(f);
692                 } else {
693                         if (!(flags & XMIT_SAME_DEV))
694                                 dev = read_longint(f);
695                         inode = read_longint(f);
696                 }
697                 if (flist->hlink_pool) {
698                         file->link_u.idev = pool_talloc(flist->hlink_pool,
699                             struct idev, 1, "inode_table");
700                         file->F_INODE = inode;
701                         file->F_DEV = dev;
702                 }
703         }
704 #endif
705
706         if (always_checksum && (sum_len || protocol_version < 28)) {
707                 char *sum;
708                 int slen = protocol_version < 21 ? 2 : MD4_SUM_LENGTH;
709                 if (sum_len) {
710                         file->u.sum = sum = bp;
711                         /*bp += sum_len;*/
712                 } else {
713                         /* Prior to 28, we get a useless set of nulls. */
714                         sum = empty_sum;
715                 }
716                 read_buf(f, sum, slen);
717         }
718
719         if (!preserve_perms) {
720                 /* set an appropriate set of permissions based on original
721                  * permissions and umask. This emulates what GNU cp does */
722                 file->mode &= ~orig_umask;
723         }
724
725         return file;
726 }
727
728
729 /**
730  * Create a file_struct for a named file by reading its stat()
731  * information and performing extensive checks against global
732  * options.
733  *
734  * @return the new file, or NULL if there was an error or this file
735  * should be excluded.
736  *
737  * @todo There is a small optimization opportunity here to avoid
738  * stat()ing the file in some circumstances, which has a certain cost.
739  * We are called immediately after doing readdir(), and so we may
740  * already know the d_type of the file.  We could for example avoid
741  * statting directories if we're not recursing, but this is not a very
742  * important case.  Some systems may not have d_type.
743  **/
744 struct file_struct *make_file(char *fname, struct file_list *flist,
745                               int filter_level)
746 {
747         static char *lastdir;
748         static int lastdir_len = -1;
749         struct file_struct *file;
750         STRUCT_STAT st;
751         char sum[SUM_LENGTH];
752         char thisname[MAXPATHLEN];
753         char linkname[MAXPATHLEN];
754         int alloc_len, basename_len, dirname_len, linkname_len, sum_len;
755         char *basename, *dirname, *bp;
756         unsigned short flags = 0;
757
758         if (!flist || !flist->count)    /* Ignore lastdir when invalid. */
759                 lastdir_len = -1;
760
761         if (strlcpy(thisname, fname, sizeof thisname)
762             >= sizeof thisname - flist_dir_len) {
763                 rprintf(FINFO, "skipping overly long name: %s\n",
764                         safe_fname(fname));
765                 return NULL;
766         }
767         clean_fname(thisname, 0);
768         if (sanitize_paths)
769                 sanitize_path(thisname, thisname, "", 0);
770
771         memset(sum, 0, SUM_LENGTH);
772
773         if (readlink_stat(thisname, &st, linkname) != 0) {
774                 int save_errno = errno;
775                 /* See if file is excluded before reporting an error. */
776                 if (filter_level != NO_FILTERS
777                     && is_excluded(thisname, 0, filter_level))
778                         return NULL;
779                 if (save_errno == ENOENT) {
780 #ifdef SUPPORT_LINKS
781                         /* Avoid "vanished" error if symlink points nowhere. */
782                         if (copy_links && do_lstat(thisname, &st) == 0
783                             && S_ISLNK(st.st_mode)) {
784                                 io_error |= IOERR_GENERAL;
785                                 rprintf(FERROR, "symlink has no referent: %s\n",
786                                         full_fname(thisname));
787                         } else
788 #endif
789                         {
790                                 enum logcode c = am_daemon && protocol_version < 28
791                                     ? FERROR : FINFO;
792                                 io_error |= IOERR_VANISHED;
793                                 rprintf(c, "file has vanished: %s\n",
794                                         full_fname(thisname));
795                         }
796                 } else {
797                         io_error |= IOERR_GENERAL;
798                         rsyserr(FERROR, save_errno, "readlink %s failed",
799                                 full_fname(thisname));
800                 }
801                 return NULL;
802         }
803
804         /* backup.c calls us with filter_level set to NO_FILTERS. */
805         if (filter_level == NO_FILTERS)
806                 goto skip_filters;
807
808         if (S_ISDIR(st.st_mode) && !xfer_dirs) {
809                 rprintf(FINFO, "skipping directory %s\n", safe_fname(thisname));
810                 return NULL;
811         }
812
813         /* We only care about directories because we need to avoid recursing
814          * into a mount-point directory, not to avoid copying a symlinked
815          * file if -L (or similar) was specified. */
816         if (one_file_system && st.st_dev != filesystem_dev
817             && S_ISDIR(st.st_mode))
818                 flags |= FLAG_MOUNT_POINT;
819
820         if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level))
821                 return NULL;
822
823         if (lp_ignore_nonreadable(module_id)) {
824 #ifdef SUPPORT_LINKS
825                 if (!S_ISLNK(st.st_mode))
826 #endif
827                         if (access(thisname, R_OK) != 0)
828                                 return NULL;
829         }
830
831 skip_filters:
832
833         if (verbose > 2) {
834                 rprintf(FINFO, "[%s] make_file(%s,*,%d)\n",
835                         who_am_i(), safe_fname(thisname), filter_level);
836         }
837
838         if ((basename = strrchr(thisname, '/')) != NULL) {
839                 dirname_len = ++basename - thisname; /* counts future '\0' */
840                 if (lastdir_len == dirname_len - 1
841                     && strncmp(thisname, lastdir, lastdir_len) == 0) {
842                         dirname = lastdir;
843                         dirname_len = 0; /* indicates no copy is needed */
844                 } else
845                         dirname = thisname;
846         } else {
847                 basename = thisname;
848                 dirname = NULL;
849                 dirname_len = 0;
850         }
851         basename_len = strlen(basename) + 1; /* count the '\0' */
852
853 #ifdef SUPPORT_LINKS
854         linkname_len = S_ISLNK(st.st_mode) ? strlen(linkname) + 1 : 0;
855 #else
856         linkname_len = 0;
857 #endif
858
859         sum_len = always_checksum && S_ISREG(st.st_mode) ? MD4_SUM_LENGTH : 0;
860
861         alloc_len = file_struct_len + dirname_len + basename_len
862             + linkname_len + sum_len;
863         if (flist) {
864                 bp = pool_alloc(flist->file_pool, alloc_len,
865                     "receive_file_entry");
866         } else {
867                 if (!(bp = new_array(char, alloc_len)))
868                         out_of_memory("receive_file_entry");
869         }
870
871         file = (struct file_struct *)bp;
872         memset(bp, 0, file_struct_len);
873         bp += file_struct_len;
874
875         file->flags = flags;
876         file->modtime = st.st_mtime;
877         file->length = st.st_size;
878         file->mode = st.st_mode;
879         file->uid = st.st_uid;
880         file->gid = st.st_gid;
881
882 #ifdef SUPPORT_HARD_LINKS
883         if (flist && flist->hlink_pool) {
884                 if (protocol_version < 28) {
885                         if (S_ISREG(st.st_mode))
886                                 file->link_u.idev = pool_talloc(
887                                     flist->hlink_pool, struct idev, 1,
888                                     "inode_table");
889                 } else {
890                         if (!S_ISDIR(st.st_mode) && st.st_nlink > 1)
891                                 file->link_u.idev = pool_talloc(
892                                     flist->hlink_pool, struct idev, 1,
893                                     "inode_table");
894                 }
895         }
896         if (file->link_u.idev) {
897                 file->F_DEV = st.st_dev;
898                 file->F_INODE = st.st_ino;
899         }
900 #endif
901
902         if (dirname_len) {
903                 file->dirname = lastdir = bp;
904                 lastdir_len = dirname_len - 1;
905                 memcpy(bp, dirname, dirname_len - 1);
906                 bp += dirname_len;
907                 bp[-1] = '\0';
908         } else if (dirname)
909                 file->dirname = dirname;
910
911         file->basename = bp;
912         memcpy(bp, basename, basename_len);
913         bp += basename_len;
914
915 #ifdef HAVE_STRUCT_STAT_ST_RDEV
916         if (preserve_devices && IS_DEVICE(st.st_mode))
917                 file->u.rdev = st.st_rdev;
918 #endif
919
920 #ifdef SUPPORT_LINKS
921         if (linkname_len) {
922                 file->u.link = bp;
923                 memcpy(bp, linkname, linkname_len);
924                 bp += linkname_len;
925         }
926 #endif
927
928         if (sum_len) {
929                 file->u.sum = bp;
930                 file_checksum(thisname, bp, st.st_size);
931                 /*bp += sum_len;*/
932         }
933
934         file->dir.root = flist_dir;
935
936         /* This code is only used by the receiver when it is building
937          * a list of files for a delete pass. */
938         if (keep_dirlinks && linkname_len && flist) {
939                 STRUCT_STAT st2;
940                 int save_mode = file->mode;
941                 file->mode = S_IFDIR; /* find a directory w/our name */
942                 if (flist_find(the_file_list, file) >= 0
943                     && do_stat(thisname, &st2) == 0 && S_ISDIR(st2.st_mode)) {
944                         file->modtime = st2.st_mtime;
945                         file->length = st2.st_size;
946                         file->mode = st2.st_mode;
947                         file->uid = st2.st_uid;
948                         file->gid = st2.st_gid;
949                         file->u.link = NULL;
950                 } else
951                         file->mode = save_mode;
952         }
953
954         if (S_ISREG(st.st_mode) || S_ISLNK(st.st_mode))
955                 stats.total_size += st.st_size;
956
957         return file;
958 }
959
960
961 static struct file_struct *send_file_name(int f, struct file_list *flist,
962                                           char *fname, unsigned short base_flags)
963 {
964         struct file_struct *file;
965
966         file = make_file(fname, flist, f == -2 ? SERVER_FILTERS : ALL_FILTERS);
967         if (!file)
968                 return NULL;
969
970         maybe_emit_filelist_progress(flist->count + flist_count_offset);
971
972         flist_expand(flist);
973
974         if (file->basename[0]) {
975                 flist->files[flist->count++] = file;
976                 send_file_entry(file, f, base_flags);
977         }
978         return file;
979 }
980
981 static void send_if_directory(int f, struct file_list *flist,
982                               struct file_struct *file)
983 {
984         char fbuf[MAXPATHLEN];
985
986         if (S_ISDIR(file->mode)
987             && !(file->flags & FLAG_MOUNT_POINT) && f_name_to(file, fbuf)) {
988                 void *save_filters;
989                 unsigned int len = strlen(fbuf);
990                 if (len > 1 && fbuf[len-1] == '/')
991                         fbuf[--len] = '\0';
992                 if (len >= MAXPATHLEN - 1) {
993                         io_error |= IOERR_GENERAL;
994                         rprintf(FERROR, "skipping long-named directory: %s\n",
995                                 full_fname(fbuf));
996                         return;
997                 }
998                 save_filters = push_local_filters(fbuf, len);
999                 send_directory(f, flist, fbuf, len);
1000                 pop_local_filters(save_filters);
1001         }
1002 }
1003
1004
1005 /* This function is normally called by the sender, but the receiving side also
1006  * calls it from get_dirlist() with f set to -1 so that we just construct the
1007  * file list in memory without sending it over the wire.  Also, get_dirlist()
1008  * might call this with f set to -2, which also indicates that local filter
1009  * rules should be ignored. */
1010 static void send_directory(int f, struct file_list *flist,
1011                            char *fbuf, int len)
1012 {
1013         struct dirent *di;
1014         unsigned remainder;
1015         char *p;
1016         DIR *d;
1017         int start = flist->count;
1018
1019         if (!(d = opendir(fbuf))) {
1020                 io_error |= IOERR_GENERAL;
1021                 rsyserr(FERROR, errno, "opendir %s failed", full_fname(fbuf));
1022                 return;
1023         }
1024
1025         p = fbuf + len;
1026         if (len != 1 || *fbuf != '/')
1027                 *p++ = '/';
1028         *p = '\0';
1029         remainder = MAXPATHLEN - (p - fbuf);
1030
1031         for (errno = 0, di = readdir(d); di; errno = 0, di = readdir(d)) {
1032                 char *dname = d_name(di);
1033                 if (dname[0] == '.' && (dname[1] == '\0'
1034                     || (dname[1] == '.' && dname[2] == '\0')))
1035                         continue;
1036                 if (strlcpy(p, dname, remainder) < remainder)
1037                         send_file_name(f, flist, fbuf, 0);
1038                 else {
1039                         io_error |= IOERR_GENERAL;
1040                         rprintf(FINFO,
1041                                 "cannot send long-named file %s\n",
1042                                 full_fname(fbuf));
1043                 }
1044         }
1045
1046         fbuf[len] = '\0';
1047
1048         if (errno) {
1049                 io_error |= IOERR_GENERAL;
1050                 rsyserr(FERROR, errno, "readdir(%s)", full_fname(fbuf));
1051         }
1052
1053         closedir(d);
1054
1055         if (recurse) {
1056                 int i, end = flist->count - 1;
1057                 for (i = start; i <= end; i++)
1058                         send_if_directory(f, flist, flist->files[i]);
1059         }
1060 }
1061
1062
1063 struct file_list *send_file_list(int f, int argc, char *argv[])
1064 {
1065         int l;
1066         STRUCT_STAT st;
1067         char *p, *dir, olddir[sizeof curr_dir];
1068         char lastpath[MAXPATHLEN] = "";
1069         struct file_list *flist;
1070         struct timeval start_tv, end_tv;
1071         int64 start_write;
1072         int use_ff_fd = 0;
1073
1074         if (show_filelist_p())
1075                 start_filelist_progress("building file list");
1076
1077         start_write = stats.total_written;
1078         gettimeofday(&start_tv, NULL);
1079
1080         flist = flist_new(WITH_HLINK, "send_file_list");
1081
1082         io_start_buffering_out();
1083         if (filesfrom_fd >= 0) {
1084                 if (argv[0] && !push_dir(argv[0])) {
1085                         rsyserr(FERROR, errno, "push_dir %s failed",
1086                                 full_fname(argv[0]));
1087                         exit_cleanup(RERR_FILESELECT);
1088                 }
1089                 use_ff_fd = 1;
1090         }
1091
1092         while (1) {
1093                 char fname2[MAXPATHLEN];
1094                 char *fname = fname2;
1095                 int is_dot_dir;
1096
1097                 if (use_ff_fd) {
1098                         if (read_filesfrom_line(filesfrom_fd, fname) == 0)
1099                                 break;
1100                         sanitize_path(fname, fname, "", 0);
1101                 } else {
1102                         if (argc-- == 0)
1103                                 break;
1104                         strlcpy(fname, *argv++, MAXPATHLEN);
1105                         if (sanitize_paths)
1106                                 sanitize_path(fname, fname, "", 0);
1107                 }
1108
1109                 l = strlen(fname);
1110                 if (!l || fname[l - 1] == '/') {
1111                         if (l == 2 && fname[0] == '.') {
1112                                 /* Turn "./" into just "." rather than "./." */
1113                                 fname[1] = '\0';
1114                         } else {
1115                                 if (l + 1 >= MAXPATHLEN)
1116                                         overflow_exit("send_file_list");
1117                                 fname[l++] = '.';
1118                                 fname[l] = '\0';
1119                         }
1120                         is_dot_dir = 1;
1121                 } else if (l > 1 && fname[l-1] == '.' && fname[l-2] == '.'
1122                     && (l == 2 || fname[l-3] == '/')) {
1123                         if (l + 2 >= MAXPATHLEN)
1124                                 overflow_exit("send_file_list");
1125                         fname[l++] = '/';
1126                         fname[l++] = '.';
1127                         fname[l] = '\0';
1128                         is_dot_dir = 1;
1129                 } else {
1130                         is_dot_dir = fname[l-1] == '.'
1131                                    && (l == 1 || fname[l-2] == '/');
1132                 }
1133
1134                 if (link_stat(fname, &st, keep_dirlinks) != 0) {
1135                         io_error |= IOERR_GENERAL;
1136                         rsyserr(FERROR, errno, "link_stat %s failed",
1137                                 full_fname(fname));
1138                         continue;
1139                 }
1140
1141                 if (S_ISDIR(st.st_mode) && !xfer_dirs) {
1142                         rprintf(FINFO, "skipping directory %s\n",
1143                                 safe_fname(fname));
1144                         continue;
1145                 }
1146
1147                 dir = NULL;
1148                 olddir[0] = '\0';
1149
1150                 if (!relative_paths) {
1151                         p = strrchr(fname, '/');
1152                         if (p) {
1153                                 *p = '\0';
1154                                 if (p == fname)
1155                                         dir = "/";
1156                                 else
1157                                         dir = fname;
1158                                 fname = p + 1;
1159                         }
1160                 } else if ((p = strstr(fname, "/./")) != NULL) {
1161                         *p = '\0';
1162                         if (p == fname)
1163                                 dir = "/";
1164                         else
1165                                 dir = fname;
1166                         fname = p + 3;
1167                 }
1168
1169                 if (!*fname)
1170                         fname = ".";
1171
1172                 if (dir && *dir) {
1173                         static char *lastdir;
1174                         static int lastdir_len;
1175
1176                         strlcpy(olddir, curr_dir, sizeof olddir);
1177
1178                         if (!push_dir(dir)) {
1179                                 io_error |= IOERR_GENERAL;
1180                                 rsyserr(FERROR, errno, "push_dir %s failed",
1181                                         full_fname(dir));
1182                                 continue;
1183                         }
1184
1185                         if (lastdir && strcmp(lastdir, dir) == 0) {
1186                                 flist_dir = lastdir;
1187                                 flist_dir_len = lastdir_len;
1188                         } else {
1189                                 flist_dir = lastdir = strdup(dir);
1190                                 flist_dir_len = lastdir_len = strlen(dir);
1191                         }
1192                 }
1193
1194                 if (implied_dirs && (p=strrchr(fname,'/')) && p != fname) {
1195                         /* Send the implied directories at the start of the
1196                          * source spec, so we get their permissions right. */
1197                         char *lp = lastpath, *fn = fname, *slash = fname;
1198                         *p = '\0';
1199                         /* Skip any initial directories in our path that we
1200                          * have in common with lastpath. */
1201                         while (*fn && *lp == *fn) {
1202                                 if (*fn == '/')
1203                                         slash = fn;
1204                                 lp++, fn++;
1205                         }
1206                         *p = '/';
1207                         if (fn != p || (*lp && *lp != '/')) {
1208                                 int save_copy_links = copy_links;
1209                                 int save_xfer_dirs = xfer_dirs;
1210                                 copy_links = copy_unsafe_links;
1211                                 xfer_dirs = 1;
1212                                 while ((slash = strchr(slash+1, '/')) != 0) {
1213                                         *slash = '\0';
1214                                         send_file_name(f, flist, fname, 0);
1215                                         *slash = '/';
1216                                 }
1217                                 copy_links = save_copy_links;
1218                                 xfer_dirs = save_xfer_dirs;
1219                                 *p = '\0';
1220                                 strlcpy(lastpath, fname, sizeof lastpath);
1221                                 *p = '/';
1222                         }
1223                 }
1224
1225                 if (one_file_system)
1226                         filesystem_dev = st.st_dev;
1227
1228                 if (recurse || (xfer_dirs && is_dot_dir)) {
1229                         struct file_struct *file;
1230                         if ((file = send_file_name(f, flist, fname, XMIT_TOP_DIR)))
1231                                 send_if_directory(f, flist, file);
1232                 } else
1233                         send_file_name(f, flist, fname, 0);
1234
1235                 if (olddir[0]) {
1236                         flist_dir = NULL;
1237                         flist_dir_len = 0;
1238                         if (!pop_dir(olddir)) {
1239                                 rsyserr(FERROR, errno, "pop_dir %s failed",
1240                                         full_fname(olddir));
1241                                 exit_cleanup(RERR_FILESELECT);
1242                         }
1243                 }
1244         }
1245
1246         gettimeofday(&end_tv, NULL);
1247         stats.flist_buildtime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1248                               + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1249         if (stats.flist_buildtime == 0)
1250                 stats.flist_buildtime = 1;
1251         start_tv = end_tv;
1252
1253         send_file_entry(NULL, f, 0);
1254
1255         if (show_filelist_p())
1256                 finish_filelist_progress(flist);
1257
1258         gettimeofday(&end_tv, NULL);
1259         stats.flist_xfertime = (int64)(end_tv.tv_sec - start_tv.tv_sec) * 1000
1260                              + (end_tv.tv_usec - start_tv.tv_usec) / 1000;
1261
1262         if (flist->hlink_pool) {
1263                 pool_destroy(flist->hlink_pool);
1264                 flist->hlink_pool = NULL;
1265         }
1266
1267         /* Sort the list without removing any duplicates.  This allows the
1268          * receiving side to ask for any name they like, which gives us the
1269          * flexibility to change the way we unduplicate names in the future
1270          * without causing a compatibility problem with older versions. */
1271         clean_flist(flist, 0, 0);
1272
1273         /* Now send the uid/gid list. This was introduced in
1274          * protocol version 15 */
1275         send_uid_list(f);
1276
1277         /* send the io_error flag */
1278         write_int(f, lp_ignore_errors(module_id) ? 0 : io_error);
1279
1280         io_end_buffering();
1281         stats.flist_size = stats.total_written - start_write;
1282         stats.num_files = flist->count;
1283
1284         if (verbose > 3)
1285                 output_flist(flist);
1286
1287         if (verbose > 2)
1288                 rprintf(FINFO, "send_file_list done\n");
1289
1290         return flist;
1291 }
1292
1293
1294 struct file_list *recv_file_list(int f)
1295 {
1296         struct file_list *flist;
1297         unsigned short flags;
1298         int64 start_read;
1299
1300         if (show_filelist_p())
1301                 start_filelist_progress("receiving file list");
1302
1303         start_read = stats.total_read;
1304
1305         flist = flist_new(WITH_HLINK, "recv_file_list");
1306
1307         flist->count = 0;
1308         flist->malloced = 1000;
1309         flist->files = new_array(struct file_struct *, flist->malloced);
1310         if (!flist->files)
1311                 goto oom;
1312
1313
1314         while ((flags = read_byte(f)) != 0) {
1315                 struct file_struct *file;
1316
1317                 flist_expand(flist);
1318
1319                 if (protocol_version >= 28 && (flags & XMIT_EXTENDED_FLAGS))
1320                         flags |= read_byte(f) << 8;
1321                 file = receive_file_entry(flist, flags, f);
1322
1323                 if (S_ISREG(file->mode))
1324                         stats.total_size += file->length;
1325
1326                 flist->files[flist->count++] = file;
1327
1328                 maybe_emit_filelist_progress(flist->count);
1329
1330                 if (verbose > 2) {
1331                         rprintf(FINFO, "recv_file_name(%s)\n",
1332                                 safe_fname(f_name(file)));
1333                 }
1334         }
1335         receive_file_entry(NULL, 0, 0); /* Signal that we're done. */
1336
1337         if (verbose > 2)
1338                 rprintf(FINFO, "received %d names\n", flist->count);
1339
1340         if (show_filelist_p())
1341                 finish_filelist_progress(flist);
1342
1343         clean_flist(flist, relative_paths, 1);
1344
1345         if (f >= 0) {
1346                 /* Now send the uid/gid list. This was introduced in
1347                  * protocol version 15 */
1348                 recv_uid_list(f, flist);
1349
1350                 /* Recv the io_error flag */
1351                 if (lp_ignore_errors(module_id) || ignore_errors)
1352                         read_int(f);
1353                 else
1354                         io_error |= read_int(f);
1355         }
1356
1357         if (verbose > 3)
1358                 output_flist(flist);
1359
1360         if (list_only) {
1361                 int i;
1362                 for (i = 0; i < flist->count; i++)
1363                         list_file_entry(flist->files[i]);
1364         }
1365
1366         if (verbose > 2)
1367                 rprintf(FINFO, "recv_file_list done\n");
1368
1369         stats.flist_size = stats.total_read - start_read;
1370         stats.num_files = flist->count;
1371
1372         return flist;
1373
1374 oom:
1375         out_of_memory("recv_file_list");
1376         return NULL;            /* not reached */
1377 }
1378
1379
1380 static int file_compare(struct file_struct **file1, struct file_struct **file2)
1381 {
1382         return f_name_cmp(*file1, *file2);
1383 }
1384
1385
1386 /* Search for an identically-named item in the file list.  Note that the
1387  * items must agree in their directory-ness, or no match is returned. */
1388 int flist_find(struct file_list *flist, struct file_struct *f)
1389 {
1390         int low = flist->low, high = flist->high;
1391         int ret, mid, mid_up;
1392
1393         while (low <= high) {
1394                 mid = (low + high) / 2;
1395                 for (mid_up = mid; !flist->files[mid_up]->basename; mid_up++) {}
1396                 if (mid_up <= high)
1397                         ret = f_name_cmp(flist->files[mid_up], f);
1398                 else
1399                         ret = 1;
1400                 if (ret == 0) {
1401                         if (protocol_version < 29
1402                             && S_ISDIR(flist->files[mid_up]->mode)
1403                             != S_ISDIR(f->mode))
1404                                 return -1;
1405                         return mid_up;
1406                 }
1407                 if (ret > 0)
1408                         high = mid - 1;
1409                 else
1410                         low = mid_up + 1;
1411         }
1412         return -1;
1413 }
1414
1415
1416 /*
1417  * Free up any resources a file_struct has allocated
1418  * and clear the file.
1419  */
1420 void clear_file(int i, struct file_list *flist)
1421 {
1422         if (flist->hlink_pool && flist->files[i]->link_u.idev)
1423                 pool_free(flist->hlink_pool, 0, flist->files[i]->link_u.idev);
1424         memset(flist->files[i], 0, file_struct_len);
1425 }
1426
1427
1428 /*
1429  * allocate a new file list
1430  */
1431 struct file_list *flist_new(int with_hlink, char *msg)
1432 {
1433         struct file_list *flist;
1434
1435         flist = new(struct file_list);
1436         if (!flist)
1437                 out_of_memory(msg);
1438
1439         memset(flist, 0, sizeof (struct file_list));
1440
1441         if (!(flist->file_pool = pool_create(FILE_EXTENT, 0,
1442             out_of_memory, POOL_INTERN)))
1443                 out_of_memory(msg);
1444
1445 #ifdef SUPPORT_HARD_LINKS
1446         if (with_hlink && preserve_hard_links) {
1447                 if (!(flist->hlink_pool = pool_create(HLINK_EXTENT,
1448                     sizeof (struct idev), out_of_memory, POOL_INTERN)))
1449                         out_of_memory(msg);
1450         }
1451 #endif
1452
1453         return flist;
1454 }
1455
1456 /*
1457  * free up all elements in a flist
1458  */
1459 void flist_free(struct file_list *flist)
1460 {
1461         pool_destroy(flist->file_pool);
1462         pool_destroy(flist->hlink_pool);
1463         free(flist->files);
1464         free(flist);
1465 }
1466
1467
1468 /*
1469  * This routine ensures we don't have any duplicate names in our file list.
1470  * duplicate names can cause corruption because of the pipelining
1471  */
1472 static void clean_flist(struct file_list *flist, int strip_root, int no_dups)
1473 {
1474         int i, prev_i = 0;
1475
1476         if (!flist)
1477                 return;
1478         if (flist->count == 0) {
1479                 flist->high = -1;
1480                 return;
1481         }
1482
1483         sorting_flist = flist;
1484         qsort(flist->files, flist->count,
1485             sizeof flist->files[0], (int (*)())file_compare);
1486         sorting_flist = NULL;
1487
1488         for (i = no_dups? 0 : flist->count; i < flist->count; i++) {
1489                 if (flist->files[i]->basename) {
1490                         prev_i = i;
1491                         break;
1492                 }
1493         }
1494         flist->low = prev_i;
1495         while (++i < flist->count) {
1496                 int j;
1497                 struct file_struct *file = flist->files[i];
1498
1499                 if (!file->basename)
1500                         continue;
1501                 if (f_name_cmp(file, flist->files[prev_i]) == 0)
1502                         j = prev_i;
1503                 else if (protocol_version >= 29 && S_ISDIR(file->mode)) {
1504                         int save_mode = file->mode;
1505                         /* Make sure that this directory doesn't duplicate a
1506                          * non-directory earlier in the list. */
1507                         flist->high = prev_i;
1508                         file->mode = S_IFREG;
1509                         j = flist_find(flist, file);
1510                         file->mode = save_mode;
1511                 } else
1512                         j = -1;
1513                 if (j >= 0) {
1514                         struct file_struct *fp = flist->files[j];
1515                         int keep, drop;
1516                         /* If one is a dir and the other is not, we want to
1517                          * keep the dir because it might have contents in the
1518                          * list. */
1519                         if (S_ISDIR(file->mode) != S_ISDIR(fp->mode)) {
1520                                 if (S_ISDIR(file->mode))
1521                                         keep = i, drop = j;
1522                                 else
1523                                         keep = j, drop = i;
1524                         } else
1525                                 keep = j, drop = i;
1526                         if (verbose > 1 && !am_server) {
1527                                 rprintf(FINFO,
1528                                         "removing duplicate name %s from file list (%d)\n",
1529                                         safe_fname(f_name(file)), drop);
1530                         }
1531                         /* Make sure that if we unduplicate '.', that we don't
1532                          * lose track of a user-specified top directory. */
1533                         flist->files[keep]->flags |= flist->files[drop]->flags
1534                                                    & (FLAG_TOP_DIR|FLAG_DEL_HERE);
1535
1536                         clear_file(drop, flist);
1537
1538                         if (keep == i) {
1539                                 if (flist->low == drop) {
1540                                         for (j = drop + 1;
1541                                              j < i && !flist->files[j]->basename;
1542                                              j++) {}
1543                                         flist->low = j;
1544                                 }
1545                                 prev_i = i;
1546                         }
1547                 } else
1548                         prev_i = i;
1549         }
1550         flist->high = no_dups ? prev_i : flist->count - 1;
1551
1552         if (strip_root) {
1553                 /* We need to strip off the leading slashes for relative
1554                  * paths, but this must be done _after_ the sorting phase. */
1555                 for (i = flist->low; i <= flist->high; i++) {
1556                         struct file_struct *file = flist->files[i];
1557
1558                         if (!file->dirname)
1559                                 continue;
1560                         if (*file->dirname == '/') {
1561                                 char *s = file->dirname + 1;
1562                                 while (*s == '/') s++;
1563                                 memmove(file->dirname, s, strlen(s) + 1);
1564                         }
1565
1566                         if (!*file->dirname)
1567                                 file->dirname = NULL;
1568                 }
1569         }
1570 }
1571
1572
1573 static void output_flist(struct file_list *flist)
1574 {
1575         char uidbuf[16], gidbuf[16], depthbuf[16];
1576         struct file_struct *file;
1577         const char *who = who_am_i();
1578         int i;
1579
1580         for (i = 0; i < flist->count; i++) {
1581                 file = flist->files[i];
1582                 if ((am_root || am_sender) && preserve_uid)
1583                         sprintf(uidbuf, " uid=%ld", (long)file->uid);
1584                 else
1585                         *uidbuf = '\0';
1586                 if (preserve_gid && file->gid != GID_NONE)
1587                         sprintf(gidbuf, " gid=%ld", (long)file->gid);
1588                 else
1589                         *gidbuf = '\0';
1590                 if (!am_sender)
1591                         sprintf(depthbuf, "%d", file->dir.depth);
1592                 rprintf(FINFO, "[%s] i=%d %s %s%s%s%s mode=0%o len=%.0f%s%s flags=%x\n",
1593                         who, i, am_sender ? NS(file->dir.root) : depthbuf,
1594                         file->dirname ? safe_fname(file->dirname) : "",
1595                         file->dirname ? "/" : "", NS(file->basename),
1596                         S_ISDIR(file->mode) ? "/" : "", (int)file->mode,
1597                         (double)file->length, uidbuf, gidbuf, file->flags);
1598         }
1599 }
1600
1601
1602 enum fnc_state { s_DIR, s_SLASH, s_BASE, s_TRAILING };
1603 enum fnc_type { t_PATH, t_ITEM };
1604
1605 /* Compare the names of two file_struct entities, similar to how strcmp()
1606  * would do if it were operating on the joined strings.
1607  *
1608  * Some differences beginning with protocol_version 29: (1) directory names
1609  * are compared with an assumed trailing slash so that they compare in a
1610  * way that would cause them to sort immediately prior to any content they
1611  * may have; (2) a directory of any name compares after a non-directory of
1612  * any name at the same depth; (3) a directory with name "." compares prior
1613  * to anything else.  These changes mean that a directory and a non-dir
1614  * with the same name will not compare as equal (protocol_version >= 29).
1615  *
1616  * The dirname component can be an empty string, but the basename component
1617  * cannot (and never is in the current codebase).  The basename component
1618  * may be NULL (for a removed item), in which case it is considered to be
1619  * after any existing item. */
1620 int f_name_cmp(struct file_struct *f1, struct file_struct *f2)
1621 {
1622         int dif;
1623         const uchar *c1, *c2;
1624         enum fnc_state state1, state2;
1625         enum fnc_type type1, type2;
1626         enum fnc_type t_path = protocol_version >= 29 ? t_PATH : t_ITEM;
1627
1628         if (!f1 || !f1->basename) {
1629                 if (!f2 || !f2->basename)
1630                         return 0;
1631                 return -1;
1632         }
1633         if (!f2 || !f2->basename)
1634                 return 1;
1635
1636         c1 = (uchar*)f1->dirname;
1637         c2 = (uchar*)f2->dirname;
1638         if (c1 == c2)
1639                 c1 = c2 = NULL;
1640         if (!c1) {
1641                 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1642                 c1 = (uchar*)f1->basename;
1643                 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1644                         type1 = t_ITEM;
1645                         state1 = s_TRAILING;
1646                         c1 = (uchar*)"";
1647                 } else
1648                         state1 = s_BASE;
1649         } else if (!*c1) {
1650                 type1 = t_path;
1651                 state1 = s_SLASH;
1652                 c1 = (uchar*)"/";
1653         } else {
1654                 type1 = t_path;
1655                 state1 = s_DIR;
1656         }
1657         if (!c2) {
1658                 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1659                 c2 = (uchar*)f2->basename;
1660                 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1661                         type2 = t_ITEM;
1662                         state2 = s_TRAILING;
1663                         c2 = (uchar*)"";
1664                 } else
1665                         state2 = s_BASE;
1666         } else if (!*c2) {
1667                 type2 = t_path;
1668                 state2 = s_SLASH;
1669                 c2 = (uchar*)"/";
1670         } else {
1671                 type2 = t_path;
1672                 state2 = s_DIR;
1673         }
1674
1675         if (type1 != type2)
1676                 return type1 == t_PATH ? 1 : -1;
1677
1678         while (1) {
1679                 if ((dif = (int)*c1++ - (int)*c2++) != 0)
1680                         break;
1681                 if (!*c1) {
1682                         switch (state1) {
1683                         case s_DIR:
1684                                 state1 = s_SLASH;
1685                                 c1 = (uchar*)"/";
1686                                 break;
1687                         case s_SLASH:
1688                                 type1 = S_ISDIR(f1->mode) ? t_path : t_ITEM;
1689                                 c1 = (uchar*)f1->basename;
1690                                 if (type1 == t_PATH && *c1 == '.' && !c1[1]) {
1691                                         type1 = t_ITEM;
1692                                         state1 = s_TRAILING;
1693                                         c1 = (uchar*)"";
1694                                 } else
1695                                         state1 = s_BASE;
1696                                 break;
1697                         case s_BASE:
1698                                 state1 = s_TRAILING;
1699                                 if (type1 == t_PATH) {
1700                                         c1 = (uchar*)"/";
1701                                         break;
1702                                 }
1703                                 /* FALL THROUGH */
1704                         case s_TRAILING:
1705                                 type1 = t_ITEM;
1706                                 break;
1707                         }
1708                         if (*c2 && type1 != type2)
1709                                 return type1 == t_PATH ? 1 : -1;
1710                 }
1711                 if (!*c2) {
1712                         switch (state2) {
1713                         case s_DIR:
1714                                 state2 = s_SLASH;
1715                                 c2 = (uchar*)"/";
1716                                 break;
1717                         case s_SLASH:
1718                                 type2 = S_ISDIR(f2->mode) ? t_path : t_ITEM;
1719                                 c2 = (uchar*)f2->basename;
1720                                 if (type2 == t_PATH && *c2 == '.' && !c2[1]) {
1721                                         type2 = t_ITEM;
1722                                         state2 = s_TRAILING;
1723                                         c2 = (uchar*)"";
1724                                 } else
1725                                         state2 = s_BASE;
1726                                 break;
1727                         case s_BASE:
1728                                 state2 = s_TRAILING;
1729                                 if (type2 == t_PATH) {
1730                                         c2 = (uchar*)"/";
1731                                         break;
1732                                 }
1733                                 /* FALL THROUGH */
1734                         case s_TRAILING:
1735                                 if (!*c1)
1736                                         return 0;
1737                                 type2 = t_ITEM;
1738                                 break;
1739                         }
1740                         if (type1 != type2)
1741                                 return type1 == t_PATH ? 1 : -1;
1742                 }
1743         }
1744
1745         return dif;
1746 }
1747
1748
1749 /* Return a copy of the full filename of a flist entry, using the indicated
1750  * buffer.  No size-checking is done because we checked the size when creating
1751  * the file_struct entry.
1752  */
1753 char *f_name_to(struct file_struct *f, char *fbuf)
1754 {
1755         if (!f || !f->basename)
1756                 return NULL;
1757
1758         if (f->dirname) {
1759                 int len = strlen(f->dirname);
1760                 memcpy(fbuf, f->dirname, len);
1761                 fbuf[len] = '/';
1762                 strcpy(fbuf + len + 1, f->basename);
1763         } else
1764                 strcpy(fbuf, f->basename);
1765         return fbuf;
1766 }
1767
1768
1769 /* Like f_name_to(), but we rotate through 5 static buffers of our own. */
1770 char *f_name(struct file_struct *f)
1771 {
1772         static char names[5][MAXPATHLEN];
1773         static unsigned int n;
1774
1775         n = (n + 1) % (sizeof names / sizeof names[0]);
1776
1777         return f_name_to(f, names[n]);
1778 }
1779
1780
1781 /* Do a non-recursive scan of the named directory, possibly ignoring all
1782  * exclude rules except for the daemon's.  If "dlen" is >=0, it is the length
1783  * of the dirname string, and also indicates that "dirname" is a MAXPATHLEN
1784  * buffer (the functions we call will append names onto the end, but the old
1785  * dir value will be restored on exit). */
1786 struct file_list *get_dirlist(char *dirname, int dlen,
1787                               int ignore_filter_rules)
1788 {
1789         struct file_list *dirlist;
1790         char dirbuf[MAXPATHLEN];
1791         int save_recurse = recurse;
1792
1793         if (dlen < 0) {
1794                 dlen = strlcpy(dirbuf, dirname, MAXPATHLEN);
1795                 if (dlen >= MAXPATHLEN)
1796                         return NULL;
1797                 dirname = dirbuf;
1798         }
1799
1800         dirlist = flist_new(WITHOUT_HLINK, "get_dirlist");
1801
1802         recurse = 0;
1803         send_directory(ignore_filter_rules ? -2 : -1, dirlist, dirname, dlen);
1804         recurse = save_recurse;
1805         if (do_progress)
1806                 flist_count_offset += dirlist->count;
1807
1808         clean_flist(dirlist, 0, 0);
1809
1810         if (verbose > 3)
1811                 output_flist(dirlist);
1812
1813         return dirlist;
1814 }