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