- Call push_dir() with its new boolean arg, including the spot in
[rsync.git] / main.c
1 /*
2  * The startup routines, including main(), for rsync.
3  *
4  * Copyright (C) 1996-2001 Andrew Tridgell <tridge@samba.org>
5  * Copyright (C) 1996 Paul Mackerras
6  * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7  * Copyright (C) 2003, 2004, 2005, 2006 Wayne Davison
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write to the Free Software Foundation, Inc.,
21  * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
22  */
23
24 #include "rsync.h"
25 #if defined CONFIG_LOCALE && defined HAVE_LOCALE_H
26 #include <locale.h>
27 #endif
28
29 extern int verbose;
30 extern int dry_run;
31 extern int list_only;
32 extern int am_root;
33 extern int am_server;
34 extern int am_sender;
35 extern int am_generator;
36 extern int am_daemon;
37 extern int blocking_io;
38 extern int remove_source_files;
39 extern int daemon_over_rsh;
40 extern int need_messages_from_generator;
41 extern int kluge_around_eof;
42 extern int do_stats;
43 extern int log_got_error;
44 extern int module_id;
45 extern int copy_links;
46 extern int copy_dirlinks;
47 extern int keep_dirlinks;
48 extern int preserve_hard_links;
49 extern int protocol_version;
50 extern int recurse;
51 extern int relative_paths;
52 extern int sanitize_paths;
53 extern int curr_dir_depth;
54 extern int curr_dir_len;
55 extern int module_id;
56 extern int rsync_port;
57 extern int whole_file;
58 extern int read_batch;
59 extern int write_batch;
60 extern int batch_fd;
61 extern int batch_gen_fd;
62 extern int filesfrom_fd;
63 extern pid_t cleanup_child_pid;
64 extern struct stats stats;
65 extern char *filesfrom_host;
66 extern char *partial_dir;
67 extern char *basis_dir[];
68 extern char *rsync_path;
69 extern char *shell_cmd;
70 extern char *batch_name;
71 extern char curr_dir[MAXPATHLEN];
72 extern struct filter_list_struct server_filter_list;
73
74 int local_server = 0;
75 mode_t orig_umask = 0;
76 struct file_list *the_file_list;
77
78 /* There's probably never more than at most 2 outstanding child processes,
79  * but set it higher, just in case. */
80 #define MAXCHILDPROCS 7
81
82 #ifdef HAVE_SIGACTION
83 # ifdef HAVE_SIGPROCMASK
84 #  define SIGACTMASK(n,h) SIGACTION(n,h), sigaddset(&sigmask,(n))
85 # else
86 #  define SIGACTMASK(n,h) SIGACTION(n,h)
87 # endif
88 static struct sigaction sigact;
89 #endif
90
91 struct pid_status {
92         pid_t pid;
93         int status;
94 } pid_stat_table[MAXCHILDPROCS];
95
96 static time_t starttime, endtime;
97 static int64 total_read, total_written;
98
99 static void show_malloc_stats(void);
100
101 /* Works like waitpid(), but if we already harvested the child pid in our
102  * remember_children(), we succeed instead of returning an error. */
103 pid_t wait_process(pid_t pid, int *status_ptr, int flags)
104 {
105         pid_t waited_pid = waitpid(pid, status_ptr, flags);
106
107         if (waited_pid == -1 && errno == ECHILD) {
108                 /* Status of requested child no longer available:  check to
109                  * see if it was processed by remember_children(). */
110                 int cnt;
111                 for (cnt = 0; cnt < MAXCHILDPROCS; cnt++) {
112                         if (pid == pid_stat_table[cnt].pid) {
113                                 *status_ptr = pid_stat_table[cnt].status;
114                                 pid_stat_table[cnt].pid = 0;
115                                 return pid;
116                         }
117                 }
118         }
119
120         return waited_pid;
121 }
122
123 /* Wait for a process to exit, calling io_flush while waiting. */
124 static void wait_process_with_flush(pid_t pid, int *exit_code_ptr)
125 {
126         pid_t waited_pid;
127         int status;
128
129         while ((waited_pid = wait_process(pid, &status, WNOHANG)) == 0) {
130                 msleep(20);
131                 io_flush(FULL_FLUSH);
132         }
133
134         /* TODO: If the child exited on a signal, then log an
135          * appropriate error message.  Perhaps we should also accept a
136          * message describing the purpose of the child.  Also indicate
137          * this to the caller so that they know something went wrong. */
138         if (waited_pid < 0) {
139                 rsyserr(FERROR, errno, "waitpid");
140                 *exit_code_ptr = RERR_WAITCHILD;
141         } else if (!WIFEXITED(status)) {
142 #ifdef WCOREDUMP
143                 if (WCOREDUMP(status))
144                         *exit_code_ptr = RERR_CRASHED;
145                 else
146 #endif
147                 if (WIFSIGNALED(status))
148                         *exit_code_ptr = RERR_TERMINATED;
149                 else
150                         *exit_code_ptr = RERR_WAITCHILD;
151         } else
152                 *exit_code_ptr = WEXITSTATUS(status);
153 }
154
155 /* This function gets called from all 3 processes.  We want the client side
156  * to actually output the text, but the sender is the only process that has
157  * all the stats we need.  So, if we're a client sender, we do the report.
158  * If we're a server sender, we write the stats on the supplied fd.  If
159  * we're the client receiver we read the stats from the supplied fd and do
160  * the report.  All processes might also generate a set of debug stats, if
161  * the verbose level is high enough (this is the only thing that the
162  * generator process and the server receiver ever do here). */
163 static void handle_stats(int f)
164 {
165         endtime = time(NULL);
166
167         /* Cache two stats because the read/write code can change it. */
168         total_read = stats.total_read;
169         total_written = stats.total_written;
170
171         if (do_stats && verbose > 1) {
172                 /* These come out from every process */
173                 show_malloc_stats();
174                 show_flist_stats();
175         }
176
177         if (am_generator)
178                 return;
179
180         if (am_daemon) {
181                 if (f == -1 || !am_sender)
182                         return;
183         }
184
185         if (am_server) {
186                 if (am_sender) {
187                         write_longint(f, total_read);
188                         write_longint(f, total_written);
189                         write_longint(f, stats.total_size);
190                         if (protocol_version >= 29) {
191                                 write_longint(f, stats.flist_buildtime);
192                                 write_longint(f, stats.flist_xfertime);
193                         }
194                 }
195                 return;
196         }
197
198         /* this is the client */
199
200         if (f < 0 && !am_sender) /* e.g. when we got an empty file list. */
201                 ;
202         else if (!am_sender) {
203                 /* Read the first two in opposite order because the meaning of
204                  * read/write swaps when switching from sender to receiver. */
205                 total_written = read_longint(f);
206                 total_read = read_longint(f);
207                 stats.total_size = read_longint(f);
208                 if (protocol_version >= 29) {
209                         stats.flist_buildtime = read_longint(f);
210                         stats.flist_xfertime = read_longint(f);
211                 }
212         } else if (write_batch) {
213                 /* The --read-batch process is going to be a client
214                  * receiver, so we need to give it the stats. */
215                 write_longint(batch_fd, total_read);
216                 write_longint(batch_fd, total_written);
217                 write_longint(batch_fd, stats.total_size);
218                 if (protocol_version >= 29) {
219                         write_longint(batch_fd, stats.flist_buildtime);
220                         write_longint(batch_fd, stats.flist_xfertime);
221                 }
222         }
223 }
224
225 static void output_summary(void)
226 {
227         if (do_stats) {
228                 rprintf(FCLIENT, "\n");
229                 rprintf(FINFO,"Number of files: %d\n", stats.num_files);
230                 rprintf(FINFO,"Number of files transferred: %d\n",
231                         stats.num_transferred_files);
232                 rprintf(FINFO,"Total file size: %s bytes\n",
233                         human_num(stats.total_size));
234                 rprintf(FINFO,"Total transferred file size: %s bytes\n",
235                         human_num(stats.total_transferred_size));
236                 rprintf(FINFO,"Literal data: %s bytes\n",
237                         human_num(stats.literal_data));
238                 rprintf(FINFO,"Matched data: %s bytes\n",
239                         human_num(stats.matched_data));
240                 rprintf(FINFO,"File list size: %d\n", stats.flist_size);
241                 if (stats.flist_buildtime) {
242                         rprintf(FINFO,
243                                 "File list generation time: %.3f seconds\n",
244                                 (double)stats.flist_buildtime / 1000);
245                         rprintf(FINFO,
246                                 "File list transfer time: %.3f seconds\n",
247                                 (double)stats.flist_xfertime / 1000);
248                 }
249                 rprintf(FINFO,"Total bytes sent: %s\n",
250                         human_num(total_written));
251                 rprintf(FINFO,"Total bytes received: %s\n",
252                         human_num(total_read));
253         }
254
255         if (verbose || do_stats) {
256                 rprintf(FCLIENT, "\n");
257                 rprintf(FINFO,
258                         "sent %s bytes  received %s bytes  %s bytes/sec\n",
259                         human_num(total_written), human_num(total_read),
260                         human_dnum((total_written + total_read)/(0.5 + (endtime - starttime)), 2));
261                 rprintf(FINFO, "total size is %s  speedup is %.2f\n",
262                         human_num(stats.total_size),
263                         (double)stats.total_size / (total_written+total_read));
264         }
265
266         fflush(stdout);
267         fflush(stderr);
268 }
269
270
271 /**
272  * If our C library can get malloc statistics, then show them to FINFO
273  **/
274 static void show_malloc_stats(void)
275 {
276 #ifdef HAVE_MALLINFO
277         struct mallinfo mi;
278
279         mi = mallinfo();
280
281         rprintf(FCLIENT, "\n");
282         rprintf(FINFO, RSYNC_NAME "[%d] (%s%s%s) heap statistics:\n",
283                 getpid(), am_server ? "server " : "",
284                 am_daemon ? "daemon " : "", who_am_i());
285         rprintf(FINFO, "  arena:     %10ld   (bytes from sbrk)\n",
286                 (long)mi.arena);
287         rprintf(FINFO, "  ordblks:   %10ld   (chunks not in use)\n",
288                 (long)mi.ordblks);
289         rprintf(FINFO, "  smblks:    %10ld\n",
290                 (long)mi.smblks);
291         rprintf(FINFO, "  hblks:     %10ld   (chunks from mmap)\n",
292                 (long)mi.hblks);
293         rprintf(FINFO, "  hblkhd:    %10ld   (bytes from mmap)\n",
294                 (long)mi.hblkhd);
295         rprintf(FINFO, "  allmem:    %10ld   (bytes from sbrk + mmap)\n",
296                 (long)mi.arena + mi.hblkhd);
297         rprintf(FINFO, "  usmblks:   %10ld\n",
298                 (long)mi.usmblks);
299         rprintf(FINFO, "  fsmblks:   %10ld\n",
300                 (long)mi.fsmblks);
301         rprintf(FINFO, "  uordblks:  %10ld   (bytes used)\n",
302                 (long)mi.uordblks);
303         rprintf(FINFO, "  fordblks:  %10ld   (bytes free)\n",
304                 (long)mi.fordblks);
305         rprintf(FINFO, "  keepcost:  %10ld   (bytes in releasable chunk)\n",
306                 (long)mi.keepcost);
307 #endif /* HAVE_MALLINFO */
308 }
309
310
311 /* Start the remote shell.   cmd may be NULL to use the default. */
312 static pid_t do_cmd(char *cmd, char *machine, char *user, char *path,
313                     int *f_in, int *f_out)
314 {
315         int i, argc = 0;
316         char *args[MAX_ARGS];
317         pid_t ret;
318         char *dir = NULL;
319         int dash_l_set = 0;
320
321         if (!read_batch && !local_server) {
322                 char *t, *f, in_quote = '\0';
323                 char *rsh_env = getenv(RSYNC_RSH_ENV);
324                 if (!cmd)
325                         cmd = rsh_env;
326                 if (!cmd)
327                         cmd = RSYNC_RSH;
328                 cmd = strdup(cmd);
329                 if (!cmd)
330                         goto oom;
331
332                 for (t = f = cmd; *f; f++) {
333                         if (*f == ' ')
334                                 continue;
335                         /* Comparison leaves rooms for server_options(). */
336                         if (argc >= MAX_ARGS - MAX_SERVER_ARGS) {
337                                 rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
338                                 exit_cleanup(RERR_SYNTAX);
339                         }
340                         args[argc++] = t;
341                         while (*f != ' ' || in_quote) {
342                                 if (!*f) {
343                                         if (in_quote) {
344                                                 rprintf(FERROR,
345                                                     "Missing trailing-%c in remote-shell command.\n",
346                                                     in_quote);
347                                                 exit_cleanup(RERR_SYNTAX);
348                                         }
349                                         f--;
350                                         break;
351                                 }
352                                 if (*f == '\'' || *f == '"') {
353                                         if (!in_quote) {
354                                                 in_quote = *f++;
355                                                 continue;
356                                         }
357                                         if (*f == in_quote && *++f != in_quote) {
358                                                 in_quote = '\0';
359                                                 continue;
360                                         }
361                                 }
362                                 *t++ = *f++;
363                         }
364                         *t++ = '\0';
365                 }
366
367                 /* check to see if we've already been given '-l user' in
368                  * the remote-shell command */
369                 for (i = 0; i < argc-1; i++) {
370                         if (!strcmp(args[i], "-l") && args[i+1][0] != '-')
371                                 dash_l_set = 1;
372                 }
373
374 #ifdef HAVE_REMSH
375                 /* remsh (on HPUX) takes the arguments the other way around */
376                 args[argc++] = machine;
377                 if (user && !(daemon_over_rsh && dash_l_set)) {
378                         args[argc++] = "-l";
379                         args[argc++] = user;
380                 }
381 #else
382                 if (user && !(daemon_over_rsh && dash_l_set)) {
383                         args[argc++] = "-l";
384                         args[argc++] = user;
385                 }
386                 args[argc++] = machine;
387 #endif
388
389                 args[argc++] = rsync_path;
390
391                 if (blocking_io < 0) {
392                         char *cp;
393                         if ((cp = strrchr(cmd, '/')) != NULL)
394                                 cp++;
395                         else
396                                 cp = cmd;
397                         if (strcmp(cp, "rsh") == 0 || strcmp(cp, "remsh") == 0)
398                                 blocking_io = 1;
399                 }
400
401                 server_options(args,&argc);
402
403                 if (argc >= MAX_ARGS - 2) {
404                         rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
405                         exit_cleanup(RERR_SYNTAX);
406                 }
407         }
408
409         args[argc++] = ".";
410
411         if (!daemon_over_rsh && path && *path)
412                 args[argc++] = path;
413
414         args[argc] = NULL;
415
416         if (verbose > 3) {
417                 for (i = 0; i < argc; i++)
418                         rprintf(FCLIENT, "cmd[%d]=%s ", i, args[i]);
419                 rprintf(FCLIENT, "\n");
420         }
421
422         if (read_batch) {
423                 int from_gen_pipe[2];
424                 if (fd_pair(from_gen_pipe) < 0) {
425                         rsyserr(FERROR, errno, "pipe");
426                         exit_cleanup(RERR_IPC);
427                 }
428                 batch_gen_fd = from_gen_pipe[0];
429                 *f_out = from_gen_pipe[1];
430                 *f_in = batch_fd;
431                 ret = -1; /* no child pid */
432         } else if (local_server) {
433                 /* If the user didn't request --[no-]whole-file, force
434                  * it on, but only if we're not batch processing. */
435                 if (whole_file < 0 && !write_batch)
436                         whole_file = 1;
437                 ret = local_child(argc, args, f_in, f_out, child_main);
438         } else
439                 ret = piped_child(args,f_in,f_out);
440
441         if (dir)
442                 free(dir);
443
444         return ret;
445
446   oom:
447         out_of_memory("do_cmd");
448         return 0; /* not reached */
449 }
450
451 /* The receiving side operates in one of two modes:
452  *
453  * 1. it receives any number of files into a destination directory,
454  * placing them according to their names in the file-list.
455  *
456  * 2. it receives a single file and saves it using the name in the
457  * destination path instead of its file-list name.  This requires a
458  * "local name" for writing out the destination file.
459  *
460  * So, our task is to figure out what mode/local-name we need.
461  * For mode 1, we change into the destination directory and return NULL.
462  * For mode 2, we change into the directory containing the destination
463  * file (if we aren't already there) and return the local-name. */
464 static char *get_local_name(struct file_list *flist, char *dest_path)
465 {
466         STRUCT_STAT st;
467         int statret;
468         char *cp;
469
470         if (verbose > 2) {
471                 rprintf(FINFO, "get_local_name count=%d %s\n",
472                         flist->count, NS(dest_path));
473         }
474
475         if (!dest_path || list_only)
476                 return NULL;
477
478         /* See what currently exists at the destination. */
479         if ((statret = safe_stat(dest_path, &st)) == 0) {
480                 /* If the destination is a dir, enter it and use mode 1. */
481                 if (S_ISDIR(st.st_mode)) {
482                         if (sanitize_paths)
483                                 die_on_unsafe_path(dest_path, 0);
484                         if (!push_dir(dest_path, 0)) {
485                                 rsyserr(FERROR, errno, "push_dir#1 %s failed",
486                                         full_fname(dest_path));
487                                 exit_cleanup(RERR_FILESELECT);
488                         }
489                         return NULL;
490                 }
491                 if (sanitize_paths && S_ISLNK(st.st_mode))
492                         die_on_unsafe_path(dest_path, 0);
493                 if (flist->count > 1) {
494                         rprintf(FERROR,
495                                 "ERROR: destination must be a directory when"
496                                 " copying more than 1 file\n");
497                         exit_cleanup(RERR_FILESELECT);
498                 }
499                 /* Caution: flist->count could be 0! */
500                 if (flist->count == 1 && S_ISDIR(flist->files[0]->mode)) {
501                         rprintf(FERROR,
502                                 "ERROR: cannot overwrite non-directory"
503                                 " with a directory\n");
504                         exit_cleanup(RERR_FILESELECT);
505                 }
506         } else if (errno != ENOENT) {
507                 /* If we don't know what's at the destination, fail. */
508                 rsyserr(FERROR, errno, "ERROR: cannot stat destination %s",
509                         full_fname(dest_path));
510                 exit_cleanup(RERR_FILESELECT);
511         }
512
513         cp = strrchr(dest_path, '/');
514
515         /* If we need a destination directory because the transfer is not
516          * of a single non-directory or the user has requested one via a
517          * destination path ending in a slash, create one and use mode 1. */
518         if (flist->count > 1 || (cp && !cp[1])) {
519                 /* Lop off the final slash (if any). */
520                 if (cp && !cp[1])
521                         *cp = '\0';
522
523                 if (statret == 0) {
524                         rprintf(FERROR,
525                             "ERROR: destination path is not a directory\n");
526                         exit_cleanup(RERR_SYNTAX);
527                 }
528
529                 if (mkdir_defmode(dest_path) != 0) {
530                         rsyserr(FERROR, errno, "mkdir %s failed",
531                                 full_fname(dest_path));
532                         exit_cleanup(RERR_FILEIO);
533                 }
534
535                 if (verbose)
536                         rprintf(FINFO, "created directory %s\n", dest_path);
537
538                 if (dry_run) {
539                         /* Indicate that dest dir doesn't really exist. */
540                         dry_run++;
541                 }
542
543                 if (sanitize_paths)
544                         die_on_unsafe_path(dest_path, 0);
545                 if (!push_dir(dest_path, dry_run > 1)) {
546                         rsyserr(FERROR, errno, "push_dir#2 %s failed",
547                                 full_fname(dest_path));
548                         exit_cleanup(RERR_FILESELECT);
549                 }
550
551                 return NULL;
552         }
553
554         /* Otherwise, we are writing a single file, possibly on top of an
555          * existing non-directory.  Change to the item's parent directory
556          * (if it has a path component), return the basename of the
557          * destination file as the local name, and use mode 2. */
558         if (!cp)
559                 return dest_path;
560
561         if (cp == dest_path)
562                 dest_path = "/";
563
564         *cp = '\0';
565         if (sanitize_paths)
566                 die_on_unsafe_path(dest_path, 0);
567         if (!push_dir(dest_path, 0)) {
568                 rsyserr(FERROR, errno, "push_dir#3 %s failed",
569                         full_fname(dest_path));
570                 exit_cleanup(RERR_FILESELECT);
571         }
572         *cp = '/';
573
574         return cp + 1;
575 }
576
577 /* Call this if the destination dir (which is assumed to be in curr_dir)
578  * does not yet exist and we can't create it due to being in dry-run
579  * mode.  We'll fix dirs that can be relative to the non-existent dir. */
580 static void fix_basis_dirs(void)
581 {
582         char **dir, *new;
583         int len;
584
585         for (dir = basis_dir; *dir; dir++) {
586                 if (**dir == '/')
587                         continue;
588                 len = curr_dir_len + 1 + strlen(*dir) + 1;
589                 if (!(new = new_array(char, len)))
590                         out_of_memory("fix_basis_dirs");
591                 pathjoin(new, len, curr_dir, *dir);
592                 clean_fname(new, 1);
593                 *dir = new;
594         }
595 }
596
597 /* This is only called by the sender. */
598 static void read_final_goodbye(int f_in, int f_out)
599 {
600         int i;
601
602         if (protocol_version < 29)
603                 i = read_int(f_in);
604         else {
605                 while ((i = read_int(f_in)) == the_file_list->count
606                     && read_shortint(f_in) == ITEM_IS_NEW) {
607                         /* Forward the keep-alive (no-op) to the receiver. */
608                         write_int(f_out, the_file_list->count);
609                         write_shortint(f_out, ITEM_IS_NEW);
610                 }
611         }
612
613         if (i != -1) {
614                 rprintf(FERROR, "Invalid packet at end of run (%d) [%s]\n",
615                         i, who_am_i());
616                 exit_cleanup(RERR_PROTOCOL);
617         }
618 }
619
620
621 static void do_server_sender(int f_in, int f_out, int argc, char *argv[])
622 {
623         struct file_list *flist;
624         char *dir = argv[0];
625
626         if (verbose > 2) {
627                 rprintf(FINFO, "server_sender starting pid=%ld\n",
628                         (long)getpid());
629         }
630
631         if (am_daemon && lp_write_only(module_id)) {
632                 rprintf(FERROR, "ERROR: module is write only\n");
633                 exit_cleanup(RERR_SYNTAX);
634                 return;
635         }
636         if (am_daemon && lp_read_only(module_id) && remove_source_files) {
637                 rprintf(FERROR,
638                     "ERROR: --remove-%s-files cannot be used with a read-only module\n",
639                     remove_source_files == 1 ? "source" : "sent");
640                 exit_cleanup(RERR_SYNTAX);
641                 return;
642         }
643
644         if (!relative_paths) {
645                 if (sanitize_paths)
646                         die_on_unsafe_path(dir, 0);
647                 if (!push_dir(dir, 0)) {
648                         rsyserr(FERROR, errno, "push_dir#3 %s failed",
649                                 full_fname(dir));
650                         exit_cleanup(RERR_FILESELECT);
651                 }
652         }
653         argc--;
654         argv++;
655
656         if (argc == 0 && (recurse || list_only)) {
657                 argc = 1;
658                 argv--;
659                 argv[0] = ".";
660         }
661
662         flist = send_file_list(f_out,argc,argv);
663         if (!flist || flist->count == 0) {
664                 exit_cleanup(0);
665         }
666         the_file_list = flist;
667
668         io_start_buffering_in();
669         io_start_buffering_out();
670
671         send_files(flist,f_out,f_in);
672         io_flush(FULL_FLUSH);
673         handle_stats(f_out);
674         if (protocol_version >= 24)
675                 read_final_goodbye(f_in, f_out);
676         io_flush(FULL_FLUSH);
677         exit_cleanup(0);
678 }
679
680
681 static int do_recv(int f_in,int f_out,struct file_list *flist,char *local_name)
682 {
683         int pid;
684         int exit_code = 0;
685         int error_pipe[2];
686
687         /* The receiving side mustn't obey this, or an existing symlink that
688          * points to an identical file won't be replaced by the referent. */
689         copy_links = copy_dirlinks = 0;
690
691         if (preserve_hard_links)
692                 init_hard_links();
693
694         if (fd_pair(error_pipe) < 0) {
695                 rsyserr(FERROR, errno, "pipe failed in do_recv");
696                 exit_cleanup(RERR_IPC);
697         }
698
699         io_flush(NORMAL_FLUSH);
700
701         if ((pid = do_fork()) == -1) {
702                 rsyserr(FERROR, errno, "fork failed in do_recv");
703                 exit_cleanup(RERR_IPC);
704         }
705
706         if (pid == 0) {
707                 close(error_pipe[0]);
708                 if (f_in != f_out)
709                         close(f_out);
710
711                 /* we can't let two processes write to the socket at one time */
712                 close_multiplexing_out();
713
714                 /* set place to send errors */
715                 set_msg_fd_out(error_pipe[1]);
716
717                 recv_files(f_in, flist, local_name);
718                 io_flush(FULL_FLUSH);
719                 handle_stats(f_in);
720
721                 send_msg(MSG_DONE, "", 0);
722                 io_flush(FULL_FLUSH);
723
724                 /* Handle any keep-alive packets from the post-processing work
725                  * that the generator does. */
726                 if (protocol_version >= 29) {
727                         kluge_around_eof = -1;
728
729                         /* This should only get stopped via a USR2 signal. */
730                         while (read_int(f_in) == flist->count
731                             && read_shortint(f_in) == ITEM_IS_NEW) {}
732
733                         rprintf(FERROR, "Invalid packet at end of run [%s]\n",
734                                 who_am_i());
735                         exit_cleanup(RERR_PROTOCOL);
736                 }
737
738                 /* Finally, we go to sleep until our parent kills us with a
739                  * USR2 signal.  We sleep for a short time, as on some OSes
740                  * a signal won't interrupt a sleep! */
741                 while (1)
742                         msleep(20);
743         }
744
745         am_generator = 1;
746         close_multiplexing_in();
747         if (write_batch && !am_server)
748                 stop_write_batch();
749
750         close(error_pipe[1]);
751         if (f_in != f_out)
752                 close(f_in);
753
754         io_start_buffering_out();
755
756         set_msg_fd_in(error_pipe[0]);
757
758         generate_files(f_out, flist, local_name);
759
760         handle_stats(-1);
761         io_flush(FULL_FLUSH);
762         if (protocol_version >= 24) {
763                 /* send a final goodbye message */
764                 write_int(f_out, -1);
765         }
766         io_flush(FULL_FLUSH);
767
768         set_msg_fd_in(-1);
769         kill(pid, SIGUSR2);
770         wait_process_with_flush(pid, &exit_code);
771         return exit_code;
772 }
773
774
775 static void do_server_recv(int f_in, int f_out, int argc,char *argv[])
776 {
777         int exit_code;
778         struct file_list *flist;
779         char *local_name = NULL;
780         char *dir = NULL;
781         int save_verbose = verbose;
782
783         if (filesfrom_fd >= 0) {
784                 /* We can't mix messages with files-from data on the socket,
785                  * so temporarily turn off verbose messages. */
786                 verbose = 0;
787         }
788
789         if (verbose > 2) {
790                 rprintf(FINFO, "server_recv(%d) starting pid=%ld\n",
791                         argc, (long)getpid());
792         }
793
794         if (am_daemon && lp_read_only(module_id)) {
795                 rprintf(FERROR,"ERROR: module is read only\n");
796                 exit_cleanup(RERR_SYNTAX);
797                 return;
798         }
799
800         if (argc > 0) {
801                 dir = argv[0];
802                 argc--;
803                 argv++;
804                 if (!am_daemon && !push_dir(dir, 0)) {
805                         rsyserr(FERROR, errno, "push_dir#4 %s failed",
806                                 full_fname(dir));
807                         exit_cleanup(RERR_FILESELECT);
808                 }
809         }
810
811         io_start_buffering_in();
812         recv_filter_list(f_in);
813
814         if (filesfrom_fd >= 0) {
815                 /* We need to send the files-from names to the sender at the
816                  * same time that we receive the file-list from them, so we
817                  * need the IO routines to automatically write out the names
818                  * onto our f_out socket as we read the file-list.  This
819                  * avoids both deadlock and extra delays/buffers. */
820                 io_set_filesfrom_fds(filesfrom_fd, f_out);
821                 filesfrom_fd = -1;
822         }
823
824         flist = recv_file_list(f_in);
825         verbose = save_verbose;
826         if (!flist) {
827                 rprintf(FERROR,"server_recv: recv_file_list error\n");
828                 exit_cleanup(RERR_FILESELECT);
829         }
830         the_file_list = flist;
831
832         if (argc > 0)
833                 local_name = get_local_name(flist,argv[0]);
834
835         /* Now that we know what our destination directory turned out to be,
836          * we can sanitize the --link-/copy-/compare-dest args correctly. */
837         if (sanitize_paths) {
838                 char **dir;
839                 for (dir = basis_dir; *dir; dir++) {
840                         *dir = sanitize_path(NULL, *dir, NULL, curr_dir_depth, NULL);
841                         die_on_unsafe_path(*dir, 0);
842                 }
843                 if (partial_dir) {
844                         partial_dir = sanitize_path(NULL, partial_dir, NULL, curr_dir_depth, NULL);
845                         /* A relative path gets this checked at every dir change. */
846                         if (*partial_dir == '/')
847                                 die_on_unsafe_path(partial_dir, 0);
848                 }
849         }
850         if (dry_run > 1)
851                 fix_basis_dirs();
852
853         if (server_filter_list.head) {
854                 char **dir;
855                 struct filter_list_struct *elp = &server_filter_list;
856
857                 for (dir = basis_dir; *dir; dir++) {
858                         if (check_filter(elp, *dir, 1) < 0)
859                                 goto options_rejected;
860                 }
861                 if (partial_dir && *partial_dir == '/'
862                  && check_filter(elp, partial_dir, 1) < 0) {
863                     options_rejected:
864                         rprintf(FERROR,
865                                 "Your options have been rejected by the server.\n");
866                         exit_cleanup(RERR_SYNTAX);
867                 }
868         }
869
870         exit_code = do_recv(f_in,f_out,flist,local_name);
871         exit_cleanup(exit_code);
872 }
873
874
875 int child_main(int argc, char *argv[])
876 {
877         start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv);
878         return 0;
879 }
880
881
882 void start_server(int f_in, int f_out, int argc, char *argv[])
883 {
884         set_nonblocking(f_in);
885         set_nonblocking(f_out);
886
887         io_set_sock_fds(f_in, f_out);
888         setup_protocol(f_out, f_in);
889 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
890         setup_iconv();
891 #endif
892
893         if (protocol_version >= 23)
894                 io_start_multiplex_out();
895
896         if (am_sender) {
897                 keep_dirlinks = 0; /* Must be disabled on the sender. */
898                 if (need_messages_from_generator)
899                         io_start_multiplex_in();
900
901                 recv_filter_list(f_in);
902                 do_server_sender(f_in, f_out, argc, argv);
903         } else {
904                 do_server_recv(f_in, f_out, argc, argv);
905         }
906         exit_cleanup(0);
907 }
908
909
910 /*
911  * This is called once the connection has been negotiated.  It is used
912  * for rsyncd, remote-shell, and local connections.
913  */
914 int client_run(int f_in, int f_out, pid_t pid, int argc, char *argv[])
915 {
916         struct file_list *flist = NULL;
917         int exit_code = 0, exit_code2 = 0;
918         char *local_name = NULL;
919
920         cleanup_child_pid = pid;
921         if (!read_batch) {
922                 set_nonblocking(f_in);
923                 set_nonblocking(f_out);
924         }
925
926         io_set_sock_fds(f_in, f_out);
927         setup_protocol(f_out,f_in);
928 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
929         setup_iconv();
930 #endif
931
932         if (protocol_version >= 23 && !read_batch)
933                 io_start_multiplex_in();
934
935         /* We set our stderr file handle to blocking because ssh might have
936          * set it to non-blocking.  This can be particularly troublesome if
937          * stderr is a clone of stdout, because ssh would have set our stdout
938          * to non-blocking at the same time (which can easily cause us to lose
939          * output from our print statements).  This kluge shouldn't cause ssh
940          * any problems for how we use it.  Note also that we delayed setting
941          * this until after the above protocol setup so that we know for sure
942          * that ssh is done twiddling its file descriptors.  */
943         set_blocking(STDERR_FILENO);
944
945         if (am_sender) {
946                 keep_dirlinks = 0; /* Must be disabled on the sender. */
947                 io_start_buffering_out();
948                 if (!filesfrom_host)
949                         set_msg_fd_in(f_in);
950                 send_filter_list(f_out);
951                 if (filesfrom_host)
952                         filesfrom_fd = f_in;
953
954                 if (write_batch && !am_server)
955                         start_write_batch(f_out);
956                 flist = send_file_list(f_out, argc, argv);
957                 set_msg_fd_in(-1);
958                 if (verbose > 3)
959                         rprintf(FINFO,"file list sent\n");
960                 the_file_list = flist;
961
962                 io_flush(NORMAL_FLUSH);
963                 send_files(flist,f_out,f_in);
964                 io_flush(FULL_FLUSH);
965                 handle_stats(-1);
966                 if (protocol_version >= 24)
967                         read_final_goodbye(f_in, f_out);
968                 if (pid != -1) {
969                         if (verbose > 3)
970                                 rprintf(FINFO,"client_run waiting on %d\n", (int) pid);
971                         io_flush(FULL_FLUSH);
972                         wait_process_with_flush(pid, &exit_code);
973                 }
974                 output_summary();
975                 io_flush(FULL_FLUSH);
976                 exit_cleanup(exit_code);
977         }
978
979         if (need_messages_from_generator && !read_batch)
980                 io_start_multiplex_out();
981
982         if (argc == 0)
983                 list_only |= 1;
984
985         send_filter_list(read_batch ? -1 : f_out);
986
987         if (filesfrom_fd >= 0) {
988                 io_set_filesfrom_fds(filesfrom_fd, f_out);
989                 filesfrom_fd = -1;
990         }
991
992         if (write_batch && !am_server)
993                 start_write_batch(f_in);
994         flist = recv_file_list(f_in);
995         the_file_list = flist;
996
997         if (flist && flist->count > 0) {
998                 local_name = get_local_name(flist, argv[0]);
999
1000                 if (dry_run > 1)
1001                         fix_basis_dirs();
1002
1003                 exit_code2 = do_recv(f_in, f_out, flist, local_name);
1004         } else {
1005                 handle_stats(-1);
1006                 output_summary();
1007         }
1008
1009         if (pid != -1) {
1010                 if (verbose > 3)
1011                         rprintf(FINFO,"client_run2 waiting on %d\n", (int) pid);
1012                 io_flush(FULL_FLUSH);
1013                 wait_process_with_flush(pid, &exit_code);
1014         }
1015
1016         return MAX(exit_code, exit_code2);
1017 }
1018
1019 static int copy_argv (char *argv[])
1020 {
1021         int i;
1022
1023         for (i = 0; argv[i]; i++) {
1024                 if (!(argv[i] = strdup(argv[i]))) {
1025                         rprintf (FERROR, "out of memory at %s(%d)\n",
1026                                  __FILE__, __LINE__);
1027                         return RERR_MALLOC;
1028                 }
1029         }
1030
1031         return 0;
1032 }
1033
1034
1035 /**
1036  * Start a client for either type of remote connection.  Work out
1037  * whether the arguments request a remote shell or rsyncd connection,
1038  * and call the appropriate connection function, then run_client.
1039  *
1040  * Calls either start_socket_client (for sockets) or do_cmd and
1041  * client_run (for ssh).
1042  **/
1043 static int start_client(int argc, char *argv[])
1044 {
1045         char *p;
1046         char *shell_machine = NULL;
1047         char *shell_path = NULL;
1048         char *shell_user = NULL;
1049         int ret;
1050         pid_t pid;
1051         int f_in,f_out;
1052         int rc;
1053
1054         /* Don't clobber argv[] so that ps(1) can still show the right
1055          * command line. */
1056         if ((rc = copy_argv(argv)))
1057                 return rc;
1058
1059         if (!read_batch) { /* for read_batch, NO source is specified */
1060                 shell_path = check_for_hostspec(argv[0], &shell_machine, &rsync_port);
1061                 if (shell_path) { /* source is remote */
1062                         char *dummy1;
1063                         int dummy2;
1064                         if (--argc
1065                          && check_for_hostspec(argv[argc], &dummy1, &dummy2)) {
1066                                 rprintf(FERROR,
1067                                         "The source and destination cannot both be remote.\n");
1068                                 exit_cleanup(RERR_SYNTAX);
1069                         }
1070                         argv++;
1071                         if (filesfrom_host && *filesfrom_host
1072                             && strcmp(filesfrom_host, shell_machine) != 0) {
1073                                 rprintf(FERROR,
1074                                         "--files-from hostname is not the same as the transfer hostname\n");
1075                                 exit_cleanup(RERR_SYNTAX);
1076                         }
1077                         if (rsync_port) {
1078                                 if (!shell_cmd) {
1079                                         return start_socket_client(shell_machine,
1080                                                                    shell_path,
1081                                                                    argc, argv);
1082                                 }
1083                                 daemon_over_rsh = 1;
1084                         }
1085
1086                         am_sender = 0;
1087                 } else { /* source is local, check dest arg */
1088                         am_sender = 1;
1089
1090                         if (argc > 1)
1091                                 p = argv[--argc];
1092                         else {
1093                                 p = ".";
1094                                 list_only = 1;
1095                         }
1096
1097                         shell_path = check_for_hostspec(p, &shell_machine, &rsync_port);
1098                         if (shell_path && filesfrom_host && *filesfrom_host
1099                             && strcmp(filesfrom_host, shell_machine) != 0) {
1100                                 rprintf(FERROR,
1101                                         "--files-from hostname is not the same as the transfer hostname\n");
1102                                 exit_cleanup(RERR_SYNTAX);
1103                         }
1104                         if (!shell_path) { /* no hostspec found, so src & dest are local */
1105                                 local_server = 1;
1106                                 if (filesfrom_host) {
1107                                         rprintf(FERROR,
1108                                                 "--files-from cannot be remote when the transfer is local\n");
1109                                         exit_cleanup(RERR_SYNTAX);
1110                                 }
1111                                 shell_machine = NULL;
1112                                 shell_path = p;
1113                         } else if (rsync_port) {
1114                                 if (!shell_cmd) {
1115                                         return start_socket_client(shell_machine,
1116                                                                    shell_path,
1117                                                                    argc, argv);
1118                                 }
1119                                 daemon_over_rsh = 1;
1120                         }
1121                 }
1122         } else {  /* read_batch */
1123                 local_server = 1;
1124                 shell_path = argv[argc-1];
1125                 if (check_for_hostspec(shell_path, &shell_machine, &rsync_port)) {
1126                         rprintf(FERROR, "remote destination is not allowed with --read-batch\n");
1127                         exit_cleanup(RERR_SYNTAX);
1128                 }
1129         }
1130
1131         if (shell_machine) {
1132                 p = strrchr(shell_machine,'@');
1133                 if (p) {
1134                         *p = 0;
1135                         shell_user = shell_machine;
1136                         shell_machine = p+1;
1137                 }
1138         }
1139
1140         if (verbose > 3) {
1141                 rprintf(FINFO,"cmd=%s machine=%s user=%s path=%s\n",
1142                         shell_cmd ? shell_cmd : "",
1143                         shell_machine ? shell_machine : "",
1144                         shell_user ? shell_user : "",
1145                         shell_path ? shell_path : "");
1146         }
1147
1148         /* for remote source, only single dest arg can remain ... */
1149         if (!am_sender && argc > 1) {
1150                 usage(FERROR);
1151                 exit_cleanup(RERR_SYNTAX);
1152         }
1153
1154         /* ... or no dest at all */
1155         if (!am_sender && argc == 0)
1156                 list_only |= 1;
1157
1158         pid = do_cmd(shell_cmd,shell_machine,shell_user,shell_path,
1159                      &f_in,&f_out);
1160
1161         /* if we're running an rsync server on the remote host over a
1162          * remote shell command, we need to do the RSYNCD protocol first */
1163         if (daemon_over_rsh) {
1164                 int tmpret;
1165                 tmpret = start_inband_exchange(shell_user, shell_path,
1166                                                f_in, f_out, argc);
1167                 if (tmpret < 0)
1168                         return tmpret;
1169         }
1170
1171         ret = client_run(f_in, f_out, pid, argc, argv);
1172
1173         fflush(stdout);
1174         fflush(stderr);
1175
1176         return ret;
1177 }
1178
1179
1180 static RETSIGTYPE sigusr1_handler(UNUSED(int val))
1181 {
1182         exit_cleanup(RERR_SIGNAL1);
1183 }
1184
1185 static RETSIGTYPE sigusr2_handler(UNUSED(int val))
1186 {
1187         if (!am_server)
1188                 output_summary();
1189         close_all();
1190         if (log_got_error)
1191                 _exit(RERR_PARTIAL);
1192         _exit(0);
1193 }
1194
1195 RETSIGTYPE remember_children(UNUSED(int val))
1196 {
1197 #ifdef WNOHANG
1198         int cnt, status;
1199         pid_t pid;
1200         /* An empty waitpid() loop was put here by Tridge and we could never
1201          * get him to explain why he put it in, so rather than taking it
1202          * out we're instead saving the child exit statuses for later use.
1203          * The waitpid() loop presumably eliminates all possibility of leaving
1204          * zombie children, maybe that's why he did it. */
1205         while ((pid = waitpid(-1, &status, WNOHANG)) > 0) {
1206                 /* save the child's exit status */
1207                 for (cnt = 0; cnt < MAXCHILDPROCS; cnt++) {
1208                         if (pid_stat_table[cnt].pid == 0) {
1209                                 pid_stat_table[cnt].pid = pid;
1210                                 pid_stat_table[cnt].status = status;
1211                                 break;
1212                         }
1213                 }
1214         }
1215 #endif
1216 #ifndef HAVE_SIGACTION
1217         signal(SIGCHLD, remember_children);
1218 #endif
1219 }
1220
1221
1222 /**
1223  * This routine catches signals and tries to send them to gdb.
1224  *
1225  * Because it's called from inside a signal handler it ought not to
1226  * use too many library routines.
1227  *
1228  * @todo Perhaps use "screen -X" instead/as well, to help people
1229  * debugging without easy access to X.  Perhaps use an environment
1230  * variable, or just call a script?
1231  *
1232  * @todo The /proc/ magic probably only works on Linux (and
1233  * Solaris?)  Can we be more portable?
1234  **/
1235 #ifdef MAINTAINER_MODE
1236 const char *get_panic_action(void)
1237 {
1238         const char *cmd_fmt = getenv("RSYNC_PANIC_ACTION");
1239
1240         if (cmd_fmt)
1241                 return cmd_fmt;
1242         else
1243                 return "xterm -display :0 -T Panic -n Panic "
1244                         "-e gdb /proc/%d/exe %d";
1245 }
1246
1247
1248 /**
1249  * Handle a fatal signal by launching a debugger, controlled by $RSYNC_PANIC_ACTION.
1250  *
1251  * This signal handler is only installed if we were configured with
1252  * --enable-maintainer-mode.  Perhaps it should always be on and we
1253  * should just look at the environment variable, but I'm a bit leery
1254  * of a signal sending us into a busy loop.
1255  **/
1256 static RETSIGTYPE rsync_panic_handler(UNUSED(int whatsig))
1257 {
1258         char cmd_buf[300];
1259         int ret;
1260
1261         sprintf(cmd_buf, get_panic_action(),
1262                 getpid(), getpid());
1263
1264         /* Unless we failed to execute gdb, we allow the process to
1265          * continue.  I'm not sure if that's right. */
1266         ret = system(cmd_buf);
1267         if (ret)
1268                 _exit(ret);
1269 }
1270 #endif
1271
1272
1273 int main(int argc,char *argv[])
1274 {
1275         int ret;
1276         int orig_argc = argc;
1277         char **orig_argv = argv;
1278 #ifdef HAVE_SIGACTION
1279 # ifdef HAVE_SIGPROCMASK
1280         sigset_t sigmask;
1281
1282         sigemptyset(&sigmask);
1283 # endif
1284         sigact.sa_flags = SA_NOCLDSTOP;
1285 #endif
1286         SIGACTMASK(SIGUSR1, sigusr1_handler);
1287         SIGACTMASK(SIGUSR2, sigusr2_handler);
1288         SIGACTMASK(SIGCHLD, remember_children);
1289 #ifdef MAINTAINER_MODE
1290         SIGACTMASK(SIGSEGV, rsync_panic_handler);
1291         SIGACTMASK(SIGFPE, rsync_panic_handler);
1292         SIGACTMASK(SIGABRT, rsync_panic_handler);
1293         SIGACTMASK(SIGBUS, rsync_panic_handler);
1294 #endif
1295
1296         starttime = time(NULL);
1297         am_root = (MY_UID() == 0);
1298
1299         memset(&stats, 0, sizeof(stats));
1300
1301         if (argc < 2) {
1302                 usage(FERROR);
1303                 exit_cleanup(RERR_SYNTAX);
1304         }
1305
1306         /* we set a 0 umask so that correct file permissions can be
1307          * carried across */
1308         orig_umask = umask(0);
1309
1310 #if defined CONFIG_LOCALE && defined HAVE_SETLOCALE
1311         setlocale(LC_CTYPE, "");
1312 #endif
1313
1314         if (!parse_arguments(&argc, (const char ***) &argv, 1)) {
1315                 /* FIXME: We ought to call the same error-handling
1316                  * code here, rather than relying on getopt. */
1317                 option_error();
1318                 exit_cleanup(RERR_SYNTAX);
1319         }
1320
1321         SIGACTMASK(SIGINT, sig_int);
1322         SIGACTMASK(SIGHUP, sig_int);
1323         SIGACTMASK(SIGTERM, sig_int);
1324 #if defined HAVE_SIGACTION && HAVE_SIGPROCMASK
1325         sigprocmask(SIG_UNBLOCK, &sigmask, NULL);
1326 #endif
1327
1328         /* Ignore SIGPIPE; we consistently check error codes and will
1329          * see the EPIPE. */
1330         SIGACTION(SIGPIPE, SIG_IGN);
1331 #ifdef SIGXFSZ
1332         SIGACTION(SIGXFSZ, SIG_IGN);
1333 #endif
1334
1335         /* Initialize push_dir here because on some old systems getcwd
1336          * (implemented by forking "pwd" and reading its output) doesn't
1337          * work when there are other child processes.  Also, on all systems
1338          * that implement getcwd that way "pwd" can't be found after chroot. */
1339         push_dir(NULL, 0);
1340
1341         init_flist();
1342
1343         if ((write_batch || read_batch) && !am_server) {
1344                 if (write_batch)
1345                         write_batch_shell_file(orig_argc, orig_argv, argc);
1346
1347                 if (read_batch && strcmp(batch_name, "-") == 0)
1348                         batch_fd = STDIN_FILENO;
1349                 else {
1350                         batch_fd = do_open(batch_name,
1351                                    write_batch ? O_WRONLY | O_CREAT | O_TRUNC
1352                                    : O_RDONLY, S_IRUSR | S_IWUSR);
1353                 }
1354                 if (batch_fd < 0) {
1355                         rsyserr(FERROR, errno, "Batch file %s open error",
1356                                 full_fname(batch_name));
1357                         exit_cleanup(RERR_FILEIO);
1358                 }
1359                 if (read_batch)
1360                         read_stream_flags(batch_fd);
1361         }
1362         if (write_batch < 0)
1363                 dry_run = 1;
1364
1365         if (am_daemon && !am_server)
1366                 return daemon_main();
1367
1368         if (argc < 1) {
1369                 usage(FERROR);
1370                 exit_cleanup(RERR_SYNTAX);
1371         }
1372
1373         if (am_server) {
1374                 set_nonblocking(STDIN_FILENO);
1375                 set_nonblocking(STDOUT_FILENO);
1376                 if (am_daemon)
1377                         return start_daemon(STDIN_FILENO, STDOUT_FILENO);
1378                 start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv);
1379         }
1380
1381         ret = start_client(argc, argv);
1382         if (ret == -1)
1383                 exit_cleanup(RERR_STARTCLIENT);
1384         else
1385                 exit_cleanup(ret);
1386
1387         return ret;
1388 }