A helper file for cleanup.c.
[rsync/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 int new_root_dir = 0;
76 mode_t orig_umask = 0;
77 struct file_list *the_file_list;
78
79 /* There's probably never more than at most 2 outstanding child processes,
80  * but set it higher, just in case. */
81 #define MAXCHILDPROCS 7
82
83 #ifdef HAVE_SIGACTION
84 # ifdef HAVE_SIGPROCMASK
85 #  define SIGACTMASK(n,h) SIGACTION(n,h), sigaddset(&sigmask,(n))
86 # else
87 #  define SIGACTMASK(n,h) SIGACTION(n,h)
88 # endif
89 static struct sigaction sigact;
90 #endif
91
92 struct pid_status {
93         pid_t pid;
94         int status;
95 } pid_stat_table[MAXCHILDPROCS];
96
97 static time_t starttime, endtime;
98 static int64 total_read, total_written;
99
100 static void show_malloc_stats(void);
101
102 /* Works like waitpid(), but if we already harvested the child pid in our
103  * remember_children(), we succeed instead of returning an error. */
104 pid_t wait_process(pid_t pid, int *status_ptr, int flags)
105 {
106         pid_t waited_pid = waitpid(pid, status_ptr, flags);
107
108         if (waited_pid == -1 && errno == ECHILD) {
109                 /* Status of requested child no longer available:  check to
110                  * see if it was processed by remember_children(). */
111                 int cnt;
112                 for (cnt = 0; cnt < MAXCHILDPROCS; cnt++) {
113                         if (pid == pid_stat_table[cnt].pid) {
114                                 *status_ptr = pid_stat_table[cnt].status;
115                                 pid_stat_table[cnt].pid = 0;
116                                 return pid;
117                         }
118                 }
119         }
120
121         return waited_pid;
122 }
123
124 /* Wait for a process to exit, calling io_flush while waiting. */
125 static void wait_process_with_flush(pid_t pid, int *exit_code_ptr)
126 {
127         pid_t waited_pid;
128         int status;
129
130         while ((waited_pid = wait_process(pid, &status, WNOHANG)) == 0) {
131                 msleep(20);
132                 io_flush(FULL_FLUSH);
133         }
134
135         /* TODO: If the child exited on a signal, then log an
136          * appropriate error message.  Perhaps we should also accept a
137          * message describing the purpose of the child.  Also indicate
138          * this to the caller so that they know something went wrong. */
139         if (waited_pid < 0) {
140                 rsyserr(FERROR, errno, "waitpid");
141                 *exit_code_ptr = RERR_WAITCHILD;
142         } else if (!WIFEXITED(status)) {
143 #ifdef WCOREDUMP
144                 if (WCOREDUMP(status))
145                         *exit_code_ptr = RERR_CRASHED;
146                 else
147 #endif
148                 if (WIFSIGNALED(status))
149                         *exit_code_ptr = RERR_TERMINATED;
150                 else
151                         *exit_code_ptr = RERR_WAITCHILD;
152         } else
153                 *exit_code_ptr = WEXITSTATUS(status);
154 }
155
156 /* This function gets called from all 3 processes.  We want the client side
157  * to actually output the text, but the sender is the only process that has
158  * all the stats we need.  So, if we're a client sender, we do the report.
159  * If we're a server sender, we write the stats on the supplied fd.  If
160  * we're the client receiver we read the stats from the supplied fd and do
161  * the report.  All processes might also generate a set of debug stats, if
162  * the verbose level is high enough (this is the only thing that the
163  * generator process and the server receiver ever do here). */
164 static void handle_stats(int f)
165 {
166         endtime = time(NULL);
167
168         /* Cache two stats because the read/write code can change it. */
169         total_read = stats.total_read;
170         total_written = stats.total_written;
171
172         if (do_stats && verbose > 1) {
173                 /* These come out from every process */
174                 show_malloc_stats();
175                 show_flist_stats();
176         }
177
178         if (am_generator)
179                 return;
180
181         if (am_daemon) {
182                 if (f == -1 || !am_sender)
183                         return;
184         }
185
186         if (am_server) {
187                 if (am_sender) {
188                         write_longint(f, total_read);
189                         write_longint(f, total_written);
190                         write_longint(f, stats.total_size);
191                         if (protocol_version >= 29) {
192                                 write_longint(f, stats.flist_buildtime);
193                                 write_longint(f, stats.flist_xfertime);
194                         }
195                 }
196                 return;
197         }
198
199         /* this is the client */
200
201         if (f < 0 && !am_sender) /* e.g. when we got an empty file list. */
202                 ;
203         else if (!am_sender) {
204                 /* Read the first two in opposite order because the meaning of
205                  * read/write swaps when switching from sender to receiver. */
206                 total_written = read_longint(f);
207                 total_read = read_longint(f);
208                 stats.total_size = read_longint(f);
209                 if (protocol_version >= 29) {
210                         stats.flist_buildtime = read_longint(f);
211                         stats.flist_xfertime = read_longint(f);
212                 }
213         } else if (write_batch) {
214                 /* The --read-batch process is going to be a client
215                  * receiver, so we need to give it the stats. */
216                 write_longint(batch_fd, total_read);
217                 write_longint(batch_fd, total_written);
218                 write_longint(batch_fd, stats.total_size);
219                 if (protocol_version >= 29) {
220                         write_longint(batch_fd, stats.flist_buildtime);
221                         write_longint(batch_fd, stats.flist_xfertime);
222                 }
223         }
224 }
225
226 static void output_summary(void)
227 {
228         if (do_stats) {
229                 rprintf(FCLIENT, "\n");
230                 rprintf(FINFO,"Number of files: %d\n", stats.num_files);
231                 rprintf(FINFO,"Number of files transferred: %d\n",
232                         stats.num_transferred_files);
233                 rprintf(FINFO,"Total file size: %s bytes\n",
234                         human_num(stats.total_size));
235                 rprintf(FINFO,"Total transferred file size: %s bytes\n",
236                         human_num(stats.total_transferred_size));
237                 rprintf(FINFO,"Literal data: %s bytes\n",
238                         human_num(stats.literal_data));
239                 rprintf(FINFO,"Matched data: %s bytes\n",
240                         human_num(stats.matched_data));
241                 rprintf(FINFO,"File list size: %d\n", stats.flist_size);
242                 if (stats.flist_buildtime) {
243                         rprintf(FINFO,
244                                 "File list generation time: %.3f seconds\n",
245                                 (double)stats.flist_buildtime / 1000);
246                         rprintf(FINFO,
247                                 "File list transfer time: %.3f seconds\n",
248                                 (double)stats.flist_xfertime / 1000);
249                 }
250                 rprintf(FINFO,"Total bytes sent: %s\n",
251                         human_num(total_written));
252                 rprintf(FINFO,"Total bytes received: %s\n",
253                         human_num(total_read));
254         }
255
256         if (verbose || do_stats) {
257                 rprintf(FCLIENT, "\n");
258                 rprintf(FINFO,
259                         "sent %s bytes  received %s bytes  %s bytes/sec\n",
260                         human_num(total_written), human_num(total_read),
261                         human_dnum((total_written + total_read)/(0.5 + (endtime - starttime)), 2));
262                 rprintf(FINFO, "total size is %s  speedup is %.2f\n",
263                         human_num(stats.total_size),
264                         (double)stats.total_size / (total_written+total_read));
265         }
266
267         fflush(stdout);
268         fflush(stderr);
269 }
270
271
272 /**
273  * If our C library can get malloc statistics, then show them to FINFO
274  **/
275 static void show_malloc_stats(void)
276 {
277 #ifdef HAVE_MALLINFO
278         struct mallinfo mi;
279
280         mi = mallinfo();
281
282         rprintf(FCLIENT, "\n");
283         rprintf(FINFO, RSYNC_NAME "[%d] (%s%s%s) heap statistics:\n",
284                 getpid(), am_server ? "server " : "",
285                 am_daemon ? "daemon " : "", who_am_i());
286         rprintf(FINFO, "  arena:     %10ld   (bytes from sbrk)\n",
287                 (long)mi.arena);
288         rprintf(FINFO, "  ordblks:   %10ld   (chunks not in use)\n",
289                 (long)mi.ordblks);
290         rprintf(FINFO, "  smblks:    %10ld\n",
291                 (long)mi.smblks);
292         rprintf(FINFO, "  hblks:     %10ld   (chunks from mmap)\n",
293                 (long)mi.hblks);
294         rprintf(FINFO, "  hblkhd:    %10ld   (bytes from mmap)\n",
295                 (long)mi.hblkhd);
296         rprintf(FINFO, "  allmem:    %10ld   (bytes from sbrk + mmap)\n",
297                 (long)mi.arena + mi.hblkhd);
298         rprintf(FINFO, "  usmblks:   %10ld\n",
299                 (long)mi.usmblks);
300         rprintf(FINFO, "  fsmblks:   %10ld\n",
301                 (long)mi.fsmblks);
302         rprintf(FINFO, "  uordblks:  %10ld   (bytes used)\n",
303                 (long)mi.uordblks);
304         rprintf(FINFO, "  fordblks:  %10ld   (bytes free)\n",
305                 (long)mi.fordblks);
306         rprintf(FINFO, "  keepcost:  %10ld   (bytes in releasable chunk)\n",
307                 (long)mi.keepcost);
308 #endif /* HAVE_MALLINFO */
309 }
310
311
312 /* Start the remote shell.   cmd may be NULL to use the default. */
313 static pid_t do_cmd(char *cmd, char *machine, char *user, char *path,
314                     int *f_in, int *f_out)
315 {
316         int i, argc = 0;
317         char *args[MAX_ARGS];
318         pid_t ret;
319         char *dir = NULL;
320         int dash_l_set = 0;
321
322         if (!read_batch && !local_server) {
323                 char *t, *f, in_quote = '\0';
324                 char *rsh_env = getenv(RSYNC_RSH_ENV);
325                 if (!cmd)
326                         cmd = rsh_env;
327                 if (!cmd)
328                         cmd = RSYNC_RSH;
329                 cmd = strdup(cmd);
330                 if (!cmd)
331                         goto oom;
332
333                 for (t = f = cmd; *f; f++) {
334                         if (*f == ' ')
335                                 continue;
336                         /* Comparison leaves rooms for server_options(). */
337                         if (argc >= MAX_ARGS - MAX_SERVER_ARGS) {
338                                 rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
339                                 exit_cleanup(RERR_SYNTAX);
340                         }
341                         args[argc++] = t;
342                         while (*f != ' ' || in_quote) {
343                                 if (!*f) {
344                                         if (in_quote) {
345                                                 rprintf(FERROR,
346                                                     "Missing trailing-%c in remote-shell command.\n",
347                                                     in_quote);
348                                                 exit_cleanup(RERR_SYNTAX);
349                                         }
350                                         f--;
351                                         break;
352                                 }
353                                 if (*f == '\'' || *f == '"') {
354                                         if (!in_quote) {
355                                                 in_quote = *f++;
356                                                 continue;
357                                         }
358                                         if (*f == in_quote && *++f != in_quote) {
359                                                 in_quote = '\0';
360                                                 continue;
361                                         }
362                                 }
363                                 *t++ = *f++;
364                         }
365                         *t++ = '\0';
366                 }
367
368                 /* check to see if we've already been given '-l user' in
369                  * the remote-shell command */
370                 for (i = 0; i < argc-1; i++) {
371                         if (!strcmp(args[i], "-l") && args[i+1][0] != '-')
372                                 dash_l_set = 1;
373                 }
374
375 #ifdef HAVE_REMSH
376                 /* remsh (on HPUX) takes the arguments the other way around */
377                 args[argc++] = machine;
378                 if (user && !(daemon_over_rsh && dash_l_set)) {
379                         args[argc++] = "-l";
380                         args[argc++] = user;
381                 }
382 #else
383                 if (user && !(daemon_over_rsh && dash_l_set)) {
384                         args[argc++] = "-l";
385                         args[argc++] = user;
386                 }
387                 args[argc++] = machine;
388 #endif
389
390                 args[argc++] = rsync_path;
391
392                 if (blocking_io < 0) {
393                         char *cp;
394                         if ((cp = strrchr(cmd, '/')) != NULL)
395                                 cp++;
396                         else
397                                 cp = cmd;
398                         if (strcmp(cp, "rsh") == 0 || strcmp(cp, "remsh") == 0)
399                                 blocking_io = 1;
400                 }
401
402                 server_options(args,&argc);
403
404                 if (argc >= MAX_ARGS - 2) {
405                         rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
406                         exit_cleanup(RERR_SYNTAX);
407                 }
408         }
409
410         args[argc++] = ".";
411
412         if (!daemon_over_rsh && path && *path)
413                 args[argc++] = path;
414
415         args[argc] = NULL;
416
417         if (verbose > 3) {
418                 for (i = 0; i < argc; i++)
419                         rprintf(FCLIENT, "cmd[%d]=%s ", i, args[i]);
420                 rprintf(FCLIENT, "\n");
421         }
422
423         if (read_batch) {
424                 int from_gen_pipe[2];
425                 if (fd_pair(from_gen_pipe) < 0) {
426                         rsyserr(FERROR, errno, "pipe");
427                         exit_cleanup(RERR_IPC);
428                 }
429                 batch_gen_fd = from_gen_pipe[0];
430                 *f_out = from_gen_pipe[1];
431                 *f_in = batch_fd;
432                 ret = -1; /* no child pid */
433         } else if (local_server) {
434                 /* If the user didn't request --[no-]whole-file, force
435                  * it on, but only if we're not batch processing. */
436                 if (whole_file < 0 && !write_batch)
437                         whole_file = 1;
438                 ret = local_child(argc, args, f_in, f_out, child_main);
439         } else
440                 ret = piped_child(args,f_in,f_out);
441
442         if (dir)
443                 free(dir);
444
445         return ret;
446
447   oom:
448         out_of_memory("do_cmd");
449         return 0; /* not reached */
450 }
451
452 /* The receiving side operates in one of two modes:
453  *
454  * 1. it receives any number of files into a destination directory,
455  * placing them according to their names in the file-list.
456  *
457  * 2. it receives a single file and saves it using the name in the
458  * destination path instead of its file-list name.  This requires a
459  * "local name" for writing out the destination file.
460  *
461  * So, our task is to figure out what mode/local-name we need.
462  * For mode 1, we change into the destination directory and return NULL.
463  * For mode 2, we change into the directory containing the destination
464  * file (if we aren't already there) and return the local-name. */
465 static char *get_local_name(struct file_list *flist, char *dest_path)
466 {
467         STRUCT_STAT st;
468         int statret;
469         char *cp;
470
471         if (verbose > 2) {
472                 rprintf(FINFO, "get_local_name count=%d %s\n",
473                         flist->count, NS(dest_path));
474         }
475
476         if (!dest_path || list_only)
477                 return NULL;
478
479         /* See what currently exists at the destination. */
480         if ((statret = do_stat(dest_path, &st)) == 0) {
481                 /* If the destination is a dir, enter it and use mode 1. */
482                 if (S_ISDIR(st.st_mode)) {
483                         if (!push_dir(dest_path, 0)) {
484                                 rsyserr(FERROR, errno, "push_dir#1 %s failed",
485                                         full_fname(dest_path));
486                                 exit_cleanup(RERR_FILESELECT);
487                         }
488                         return NULL;
489                 }
490                 if (flist->count > 1) {
491                         rprintf(FERROR,
492                                 "ERROR: destination must be a directory when"
493                                 " copying more than 1 file\n");
494                         exit_cleanup(RERR_FILESELECT);
495                 }
496                 /* Caution: flist->count could be 0! */
497                 if (flist->count == 1 && S_ISDIR(flist->files[0]->mode)) {
498                         rprintf(FERROR,
499                                 "ERROR: cannot overwrite non-directory"
500                                 " with a directory\n");
501                         exit_cleanup(RERR_FILESELECT);
502                 }
503         } else if (errno != ENOENT) {
504                 /* If we don't know what's at the destination, fail. */
505                 rsyserr(FERROR, errno, "ERROR: cannot stat destination %s",
506                         full_fname(dest_path));
507                 exit_cleanup(RERR_FILESELECT);
508         }
509
510         cp = strrchr(dest_path, '/');
511
512         /* If we need a destination directory because the transfer is not
513          * of a single non-directory or the user has requested one via a
514          * destination path ending in a slash, create one and use mode 1. */
515         if (flist->count > 1 || (cp && !cp[1])) {
516                 /* Lop off the final slash (if any). */
517                 if (cp && !cp[1])
518                         *cp = '\0';
519
520                 if (statret == 0) {
521                         rprintf(FERROR,
522                             "ERROR: destination path is not a directory\n");
523                         exit_cleanup(RERR_SYNTAX);
524                 }
525
526                 if (mkdir_defmode(dest_path) != 0) {
527                         rsyserr(FERROR, errno, "mkdir %s failed",
528                                 full_fname(dest_path));
529                         exit_cleanup(RERR_FILEIO);
530                 }
531
532                 new_root_dir = 1;
533
534                 if (verbose)
535                         rprintf(FINFO, "created directory %s\n", dest_path);
536
537                 if (dry_run) {
538                         /* Indicate that dest dir doesn't really exist. */
539                         dry_run++;
540                 }
541
542                 if (!push_dir(dest_path, dry_run > 1)) {
543                         rsyserr(FERROR, errno, "push_dir#2 %s failed",
544                                 full_fname(dest_path));
545                         exit_cleanup(RERR_FILESELECT);
546                 }
547
548                 return NULL;
549         }
550
551         /* Otherwise, we are writing a single file, possibly on top of an
552          * existing non-directory.  Change to the item's parent directory
553          * (if it has a path component), return the basename of the
554          * destination file as the local name, and use mode 2. */
555         if (!cp)
556                 return dest_path;
557
558         if (cp == dest_path)
559                 dest_path = "/";
560
561         *cp = '\0';
562         if (!push_dir(dest_path, 0)) {
563                 rsyserr(FERROR, errno, "push_dir#3 %s failed",
564                         full_fname(dest_path));
565                 exit_cleanup(RERR_FILESELECT);
566         }
567         *cp = '/';
568
569         return cp + 1;
570 }
571
572 /* Call this if the destination dir (which is assumed to be in curr_dir)
573  * does not yet exist and we can't create it due to being in dry-run
574  * mode.  We'll fix dirs that can be relative to the non-existent dir. */
575 static void fix_basis_dirs(void)
576 {
577         char **dir, *new, *slash;
578         int len;
579
580         if (dry_run <= 1)
581                 return;
582
583         slash = strrchr(curr_dir, '/');
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                 if (slash && strncmp(*dir, "../", 3) == 0) {
592                     /* We want to remove only one leading "../" prefix for
593                      * the directory we couldn't create in dry-run mode:
594                      * this ensures that any other ".." references get
595                      * evaluated the same as they would for a live copy. */
596                     *slash = '\0';
597                     pathjoin(new, len, curr_dir, *dir + 3);
598                     *slash = '/';
599                 } else
600                     pathjoin(new, len, curr_dir, *dir);
601                 *dir = new;
602         }
603 }
604
605 /* This is only called by the sender. */
606 static void read_final_goodbye(int f_in, int f_out)
607 {
608         int i;
609
610         if (protocol_version < 29)
611                 i = read_int(f_in);
612         else {
613                 while ((i = read_int(f_in)) == the_file_list->count
614                     && read_shortint(f_in) == ITEM_IS_NEW) {
615                         /* Forward the keep-alive (no-op) to the receiver. */
616                         write_int(f_out, the_file_list->count);
617                         write_shortint(f_out, ITEM_IS_NEW);
618                 }
619         }
620
621         if (i != -1) {
622                 rprintf(FERROR, "Invalid packet at end of run (%d) [%s]\n",
623                         i, who_am_i());
624                 exit_cleanup(RERR_PROTOCOL);
625         }
626 }
627
628
629 static void do_server_sender(int f_in, int f_out, int argc, char *argv[])
630 {
631         struct file_list *flist;
632         char *dir = argv[0];
633
634         if (verbose > 2) {
635                 rprintf(FINFO, "server_sender starting pid=%ld\n",
636                         (long)getpid());
637         }
638
639         if (am_daemon && lp_write_only(module_id)) {
640                 rprintf(FERROR, "ERROR: module is write only\n");
641                 exit_cleanup(RERR_SYNTAX);
642                 return;
643         }
644         if (am_daemon && lp_read_only(module_id) && remove_source_files) {
645                 rprintf(FERROR,
646                     "ERROR: --remove-%s-files cannot be used with a read-only module\n",
647                     remove_source_files == 1 ? "source" : "sent");
648                 exit_cleanup(RERR_SYNTAX);
649                 return;
650         }
651
652         if (!relative_paths) {
653                 if (!push_dir(dir, 0)) {
654                         rsyserr(FERROR, errno, "push_dir#3 %s failed",
655                                 full_fname(dir));
656                         exit_cleanup(RERR_FILESELECT);
657                 }
658         }
659         argc--;
660         argv++;
661
662         if (argc == 0 && (recurse || list_only)) {
663                 argc = 1;
664                 argv--;
665                 argv[0] = ".";
666         }
667
668         flist = send_file_list(f_out,argc,argv);
669         if (!flist || flist->count == 0) {
670                 exit_cleanup(0);
671         }
672         the_file_list = flist;
673
674         io_start_buffering_in();
675         io_start_buffering_out();
676
677         send_files(flist,f_out,f_in);
678         io_flush(FULL_FLUSH);
679         handle_stats(f_out);
680         if (protocol_version >= 24)
681                 read_final_goodbye(f_in, f_out);
682         io_flush(FULL_FLUSH);
683         exit_cleanup(0);
684 }
685
686
687 static int do_recv(int f_in,int f_out,struct file_list *flist,char *local_name)
688 {
689         int pid;
690         int exit_code = 0;
691         int error_pipe[2];
692
693         /* The receiving side mustn't obey this, or an existing symlink that
694          * points to an identical file won't be replaced by the referent. */
695         copy_links = copy_dirlinks = 0;
696
697         if (preserve_hard_links)
698                 init_hard_links();
699
700         if (fd_pair(error_pipe) < 0) {
701                 rsyserr(FERROR, errno, "pipe failed in do_recv");
702                 exit_cleanup(RERR_IPC);
703         }
704
705         io_flush(NORMAL_FLUSH);
706
707         if ((pid = do_fork()) == -1) {
708                 rsyserr(FERROR, errno, "fork failed in do_recv");
709                 exit_cleanup(RERR_IPC);
710         }
711
712         if (pid == 0) {
713                 close(error_pipe[0]);
714                 if (f_in != f_out)
715                         close(f_out);
716
717                 /* we can't let two processes write to the socket at one time */
718                 close_multiplexing_out();
719
720                 /* set place to send errors */
721                 set_msg_fd_out(error_pipe[1]);
722
723                 recv_files(f_in, flist, local_name);
724                 io_flush(FULL_FLUSH);
725                 handle_stats(f_in);
726
727                 send_msg(MSG_DONE, "", 0);
728                 io_flush(FULL_FLUSH);
729
730                 /* Handle any keep-alive packets from the post-processing work
731                  * that the generator does. */
732                 if (protocol_version >= 29) {
733                         kluge_around_eof = -1;
734
735                         /* This should only get stopped via a USR2 signal. */
736                         while (read_int(f_in) == flist->count
737                             && read_shortint(f_in) == ITEM_IS_NEW) {}
738
739                         rprintf(FERROR, "Invalid packet at end of run [%s]\n",
740                                 who_am_i());
741                         exit_cleanup(RERR_PROTOCOL);
742                 }
743
744                 /* Finally, we go to sleep until our parent kills us with a
745                  * USR2 signal.  We sleep for a short time, as on some OSes
746                  * a signal won't interrupt a sleep! */
747                 while (1)
748                         msleep(20);
749         }
750
751         am_generator = 1;
752         close_multiplexing_in();
753         if (write_batch && !am_server)
754                 stop_write_batch();
755
756         close(error_pipe[1]);
757         if (f_in != f_out)
758                 close(f_in);
759
760         io_start_buffering_out();
761
762         set_msg_fd_in(error_pipe[0]);
763
764         generate_files(f_out, flist, local_name);
765
766         handle_stats(-1);
767         io_flush(FULL_FLUSH);
768         if (protocol_version >= 24) {
769                 /* send a final goodbye message */
770                 write_int(f_out, -1);
771         }
772         io_flush(FULL_FLUSH);
773
774         set_msg_fd_in(-1);
775         kill(pid, SIGUSR2);
776         wait_process_with_flush(pid, &exit_code);
777         return exit_code;
778 }
779
780
781 static void do_server_recv(int f_in, int f_out, int argc,char *argv[])
782 {
783         int exit_code;
784         struct file_list *flist;
785         char *local_name = NULL;
786         char *dir = NULL;
787         int save_verbose = verbose;
788
789         if (filesfrom_fd >= 0) {
790                 /* We can't mix messages with files-from data on the socket,
791                  * so temporarily turn off verbose messages. */
792                 verbose = 0;
793         }
794
795         if (verbose > 2) {
796                 rprintf(FINFO, "server_recv(%d) starting pid=%ld\n",
797                         argc, (long)getpid());
798         }
799
800         if (am_daemon && lp_read_only(module_id)) {
801                 rprintf(FERROR,"ERROR: module is read only\n");
802                 exit_cleanup(RERR_SYNTAX);
803                 return;
804         }
805
806         if (argc > 0) {
807                 dir = argv[0];
808                 argc--;
809                 argv++;
810                 if (!am_daemon && !push_dir(dir, 0)) {
811                         rsyserr(FERROR, errno, "push_dir#4 %s failed",
812                                 full_fname(dir));
813                         exit_cleanup(RERR_FILESELECT);
814                 }
815         }
816
817         io_start_buffering_in();
818         recv_filter_list(f_in);
819
820         if (filesfrom_fd >= 0) {
821                 /* We need to send the files-from names to the sender at the
822                  * same time that we receive the file-list from them, so we
823                  * need the IO routines to automatically write out the names
824                  * onto our f_out socket as we read the file-list.  This
825                  * avoids both deadlock and extra delays/buffers. */
826                 io_set_filesfrom_fds(filesfrom_fd, f_out);
827                 filesfrom_fd = -1;
828         }
829
830         flist = recv_file_list(f_in);
831         verbose = save_verbose;
832         if (!flist) {
833                 rprintf(FERROR,"server_recv: recv_file_list error\n");
834                 exit_cleanup(RERR_FILESELECT);
835         }
836         the_file_list = flist;
837
838         if (argc > 0)
839                 local_name = get_local_name(flist,argv[0]);
840
841         /* Now that we know what our destination directory turned out to be,
842          * we can sanitize the --link-/copy-/compare-dest args correctly. */
843         if (sanitize_paths) {
844                 char **dir;
845                 for (dir = basis_dir; *dir; dir++) {
846                         *dir = sanitize_path(NULL, *dir, NULL, curr_dir_depth, NULL);
847                 }
848                 if (partial_dir) {
849                         partial_dir = sanitize_path(NULL, partial_dir, NULL, curr_dir_depth, NULL);
850                 }
851         }
852         fix_basis_dirs();
853
854         if (server_filter_list.head) {
855                 char **dir;
856                 struct filter_list_struct *elp = &server_filter_list;
857
858                 for (dir = basis_dir; *dir; dir++) {
859                         if (check_filter(elp, *dir, 1) < 0)
860                                 goto options_rejected;
861                 }
862                 if (partial_dir && *partial_dir == '/'
863                  && check_filter(elp, partial_dir, 1) < 0) {
864                     options_rejected:
865                         rprintf(FERROR,
866                                 "Your options have been rejected by the server.\n");
867                         exit_cleanup(RERR_SYNTAX);
868                 }
869         }
870
871         exit_code = do_recv(f_in,f_out,flist,local_name);
872         exit_cleanup(exit_code);
873 }
874
875
876 int child_main(int argc, char *argv[])
877 {
878         start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv);
879         return 0;
880 }
881
882
883 void start_server(int f_in, int f_out, int argc, char *argv[])
884 {
885         set_nonblocking(f_in);
886         set_nonblocking(f_out);
887
888         io_set_sock_fds(f_in, f_out);
889         setup_protocol(f_out, f_in);
890 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
891         setup_iconv();
892 #endif
893
894         if (protocol_version >= 23)
895                 io_start_multiplex_out();
896
897         if (am_sender) {
898                 keep_dirlinks = 0; /* Must be disabled on the sender. */
899                 if (need_messages_from_generator)
900                         io_start_multiplex_in();
901
902                 recv_filter_list(f_in);
903                 do_server_sender(f_in, f_out, argc, argv);
904         } else {
905                 do_server_recv(f_in, f_out, argc, argv);
906         }
907         exit_cleanup(0);
908 }
909
910
911 /*
912  * This is called once the connection has been negotiated.  It is used
913  * for rsyncd, remote-shell, and local connections.
914  */
915 int client_run(int f_in, int f_out, pid_t pid, int argc, char *argv[])
916 {
917         struct file_list *flist = NULL;
918         int exit_code = 0, exit_code2 = 0;
919         char *local_name = NULL;
920
921         cleanup_child_pid = pid;
922         if (!read_batch) {
923                 set_nonblocking(f_in);
924                 set_nonblocking(f_out);
925         }
926
927         io_set_sock_fds(f_in, f_out);
928         setup_protocol(f_out,f_in);
929 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
930         setup_iconv();
931 #endif
932
933         if (protocol_version >= 23 && !read_batch)
934                 io_start_multiplex_in();
935
936         /* We set our stderr file handle to blocking because ssh might have
937          * set it to non-blocking.  This can be particularly troublesome if
938          * stderr is a clone of stdout, because ssh would have set our stdout
939          * to non-blocking at the same time (which can easily cause us to lose
940          * output from our print statements).  This kluge shouldn't cause ssh
941          * any problems for how we use it.  Note also that we delayed setting
942          * this until after the above protocol setup so that we know for sure
943          * that ssh is done twiddling its file descriptors.  */
944         set_blocking(STDERR_FILENO);
945
946         if (am_sender) {
947                 keep_dirlinks = 0; /* Must be disabled on the sender. */
948                 io_start_buffering_out();
949                 if (!filesfrom_host)
950                         set_msg_fd_in(f_in);
951                 send_filter_list(f_out);
952                 if (filesfrom_host)
953                         filesfrom_fd = f_in;
954
955                 if (write_batch && !am_server)
956                         start_write_batch(f_out);
957                 flist = send_file_list(f_out, argc, argv);
958                 set_msg_fd_in(-1);
959                 if (verbose > 3)
960                         rprintf(FINFO,"file list sent\n");
961                 the_file_list = flist;
962
963                 io_flush(NORMAL_FLUSH);
964                 send_files(flist,f_out,f_in);
965                 io_flush(FULL_FLUSH);
966                 handle_stats(-1);
967                 if (protocol_version >= 24)
968                         read_final_goodbye(f_in, f_out);
969                 if (pid != -1) {
970                         if (verbose > 3)
971                                 rprintf(FINFO,"client_run waiting on %d\n", (int) pid);
972                         io_flush(FULL_FLUSH);
973                         wait_process_with_flush(pid, &exit_code);
974                 }
975                 output_summary();
976                 io_flush(FULL_FLUSH);
977                 exit_cleanup(exit_code);
978         }
979
980         if (need_messages_from_generator && !read_batch)
981                 io_start_multiplex_out();
982
983         if (argc == 0)
984                 list_only |= 1;
985
986         send_filter_list(read_batch ? -1 : f_out);
987
988         if (filesfrom_fd >= 0) {
989                 io_set_filesfrom_fds(filesfrom_fd, f_out);
990                 filesfrom_fd = -1;
991         }
992
993         if (write_batch && !am_server)
994                 start_write_batch(f_in);
995         flist = recv_file_list(f_in);
996         the_file_list = flist;
997
998         if (flist && flist->count > 0) {
999                 local_name = get_local_name(flist, argv[0]);
1000
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         snprintf(cmd_buf, sizeof 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 }