#ifdef calls to hard-link functions.
[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: %s\n",
242                         human_num(stats.flist_size));
243                 if (stats.flist_buildtime) {
244                         rprintf(FINFO,
245                                 "File list generation time: %.3f seconds\n",
246                                 (double)stats.flist_buildtime / 1000);
247                         rprintf(FINFO,
248                                 "File list transfer time: %.3f seconds\n",
249                                 (double)stats.flist_xfertime / 1000);
250                 }
251                 rprintf(FINFO,"Total bytes sent: %s\n",
252                         human_num(total_written));
253                 rprintf(FINFO,"Total bytes received: %s\n",
254                         human_num(total_read));
255         }
256
257         if (verbose || do_stats) {
258                 rprintf(FCLIENT, "\n");
259                 rprintf(FINFO,
260                         "sent %s bytes  received %s bytes  %s bytes/sec\n",
261                         human_num(total_written), human_num(total_read),
262                         human_dnum((total_written + total_read)/(0.5 + (endtime - starttime)), 2));
263                 rprintf(FINFO, "total size is %s  speedup is %.2f\n",
264                         human_num(stats.total_size),
265                         (double)stats.total_size / (total_written+total_read));
266         }
267
268         fflush(stdout);
269         fflush(stderr);
270 }
271
272
273 /**
274  * If our C library can get malloc statistics, then show them to FINFO
275  **/
276 static void show_malloc_stats(void)
277 {
278 #ifdef HAVE_MALLINFO
279         struct mallinfo mi;
280
281         mi = mallinfo();
282
283         rprintf(FCLIENT, "\n");
284         rprintf(FINFO, RSYNC_NAME "[%d] (%s%s%s) heap statistics:\n",
285                 getpid(), am_server ? "server " : "",
286                 am_daemon ? "daemon " : "", who_am_i());
287         rprintf(FINFO, "  arena:     %10ld   (bytes from sbrk)\n",
288                 (long)mi.arena);
289         rprintf(FINFO, "  ordblks:   %10ld   (chunks not in use)\n",
290                 (long)mi.ordblks);
291         rprintf(FINFO, "  smblks:    %10ld\n",
292                 (long)mi.smblks);
293         rprintf(FINFO, "  hblks:     %10ld   (chunks from mmap)\n",
294                 (long)mi.hblks);
295         rprintf(FINFO, "  hblkhd:    %10ld   (bytes from mmap)\n",
296                 (long)mi.hblkhd);
297         rprintf(FINFO, "  allmem:    %10ld   (bytes from sbrk + mmap)\n",
298                 (long)mi.arena + mi.hblkhd);
299         rprintf(FINFO, "  usmblks:   %10ld\n",
300                 (long)mi.usmblks);
301         rprintf(FINFO, "  fsmblks:   %10ld\n",
302                 (long)mi.fsmblks);
303         rprintf(FINFO, "  uordblks:  %10ld   (bytes used)\n",
304                 (long)mi.uordblks);
305         rprintf(FINFO, "  fordblks:  %10ld   (bytes free)\n",
306                 (long)mi.fordblks);
307         rprintf(FINFO, "  keepcost:  %10ld   (bytes in releasable chunk)\n",
308                 (long)mi.keepcost);
309 #endif /* HAVE_MALLINFO */
310 }
311
312
313 /* Start the remote shell.   cmd may be NULL to use the default. */
314 static pid_t do_cmd(char *cmd, char *machine, char *user, char *path,
315                     int *f_in, int *f_out)
316 {
317         int i, argc = 0;
318         char *args[MAX_ARGS];
319         pid_t ret;
320         char *dir = NULL;
321         int dash_l_set = 0;
322
323         if (!read_batch && !local_server) {
324                 char *t, *f, in_quote = '\0';
325                 char *rsh_env = getenv(RSYNC_RSH_ENV);
326                 if (!cmd)
327                         cmd = rsh_env;
328                 if (!cmd)
329                         cmd = RSYNC_RSH;
330                 cmd = strdup(cmd);
331                 if (!cmd)
332                         goto oom;
333
334                 for (t = f = cmd; *f; f++) {
335                         if (*f == ' ')
336                                 continue;
337                         /* Comparison leaves rooms for server_options(). */
338                         if (argc >= MAX_ARGS - MAX_SERVER_ARGS) {
339                                 rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
340                                 exit_cleanup(RERR_SYNTAX);
341                         }
342                         args[argc++] = t;
343                         while (*f != ' ' || in_quote) {
344                                 if (!*f) {
345                                         if (in_quote) {
346                                                 rprintf(FERROR,
347                                                     "Missing trailing-%c in remote-shell command.\n",
348                                                     in_quote);
349                                                 exit_cleanup(RERR_SYNTAX);
350                                         }
351                                         f--;
352                                         break;
353                                 }
354                                 if (*f == '\'' || *f == '"') {
355                                         if (!in_quote) {
356                                                 in_quote = *f++;
357                                                 continue;
358                                         }
359                                         if (*f == in_quote && *++f != in_quote) {
360                                                 in_quote = '\0';
361                                                 continue;
362                                         }
363                                 }
364                                 *t++ = *f++;
365                         }
366                         *t++ = '\0';
367                 }
368
369                 /* check to see if we've already been given '-l user' in
370                  * the remote-shell command */
371                 for (i = 0; i < argc-1; i++) {
372                         if (!strcmp(args[i], "-l") && args[i+1][0] != '-')
373                                 dash_l_set = 1;
374                 }
375
376 #ifdef HAVE_REMSH
377                 /* remsh (on HPUX) takes the arguments the other way around */
378                 args[argc++] = machine;
379                 if (user && !(daemon_over_rsh && dash_l_set)) {
380                         args[argc++] = "-l";
381                         args[argc++] = user;
382                 }
383 #else
384                 if (user && !(daemon_over_rsh && dash_l_set)) {
385                         args[argc++] = "-l";
386                         args[argc++] = user;
387                 }
388                 args[argc++] = machine;
389 #endif
390
391                 args[argc++] = rsync_path;
392
393                 if (blocking_io < 0) {
394                         char *cp;
395                         if ((cp = strrchr(cmd, '/')) != NULL)
396                                 cp++;
397                         else
398                                 cp = cmd;
399                         if (strcmp(cp, "rsh") == 0 || strcmp(cp, "remsh") == 0)
400                                 blocking_io = 1;
401                 }
402
403                 server_options(args,&argc);
404
405                 if (argc >= MAX_ARGS - 2) {
406                         rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n");
407                         exit_cleanup(RERR_SYNTAX);
408                 }
409         }
410
411         args[argc++] = ".";
412
413         if (!daemon_over_rsh && path && *path)
414                 args[argc++] = path;
415
416         args[argc] = NULL;
417
418         if (verbose > 3) {
419                 for (i = 0; i < argc; i++)
420                         rprintf(FCLIENT, "cmd[%d]=%s ", i, args[i]);
421                 rprintf(FCLIENT, "\n");
422         }
423
424         if (read_batch) {
425                 int from_gen_pipe[2];
426                 if (fd_pair(from_gen_pipe) < 0) {
427                         rsyserr(FERROR, errno, "pipe");
428                         exit_cleanup(RERR_IPC);
429                 }
430                 batch_gen_fd = from_gen_pipe[0];
431                 *f_out = from_gen_pipe[1];
432                 *f_in = batch_fd;
433                 ret = -1; /* no child pid */
434         } else if (local_server) {
435                 /* If the user didn't request --[no-]whole-file, force
436                  * it on, but only if we're not batch processing. */
437                 if (whole_file < 0 && !write_batch)
438                         whole_file = 1;
439                 ret = local_child(argc, args, f_in, f_out, child_main);
440         } else
441                 ret = piped_child(args,f_in,f_out);
442
443         if (dir)
444                 free(dir);
445
446         return ret;
447
448   oom:
449         out_of_memory("do_cmd");
450         return 0; /* not reached */
451 }
452
453 /* The receiving side operates in one of two modes:
454  *
455  * 1. it receives any number of files into a destination directory,
456  * placing them according to their names in the file-list.
457  *
458  * 2. it receives a single file and saves it using the name in the
459  * destination path instead of its file-list name.  This requires a
460  * "local name" for writing out the destination file.
461  *
462  * So, our task is to figure out what mode/local-name we need.
463  * For mode 1, we change into the destination directory and return NULL.
464  * For mode 2, we change into the directory containing the destination
465  * file (if we aren't already there) and return the local-name. */
466 static char *get_local_name(struct file_list *flist, char *dest_path)
467 {
468         STRUCT_STAT st;
469         int statret;
470         char *cp;
471
472         if (verbose > 2) {
473                 rprintf(FINFO, "get_local_name count=%d %s\n",
474                         flist->count, NS(dest_path));
475         }
476
477         if (!dest_path || list_only)
478                 return NULL;
479
480         /* See what currently exists at the destination. */
481         if ((statret = do_stat(dest_path, &st)) == 0) {
482                 /* If the destination is a dir, enter it and use mode 1. */
483                 if (S_ISDIR(st.st_mode)) {
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 (flist->count > 1) {
492                         rprintf(FERROR,
493                                 "ERROR: destination must be a directory when"
494                                 " copying more than 1 file\n");
495                         exit_cleanup(RERR_FILESELECT);
496                 }
497                 /* Caution: flist->count could be 0! */
498                 if (flist->count == 1 && S_ISDIR(flist->files[0]->mode)) {
499                         rprintf(FERROR,
500                                 "ERROR: cannot overwrite non-directory"
501                                 " with a directory\n");
502                         exit_cleanup(RERR_FILESELECT);
503                 }
504         } else if (errno != ENOENT) {
505                 /* If we don't know what's at the destination, fail. */
506                 rsyserr(FERROR, errno, "ERROR: cannot stat destination %s",
507                         full_fname(dest_path));
508                 exit_cleanup(RERR_FILESELECT);
509         }
510
511         cp = strrchr(dest_path, '/');
512
513         /* If we need a destination directory because the transfer is not
514          * of a single non-directory or the user has requested one via a
515          * destination path ending in a slash, create one and use mode 1. */
516         if (flist->count > 1 || (cp && !cp[1])) {
517                 /* Lop off the final slash (if any). */
518                 if (cp && !cp[1])
519                         *cp = '\0';
520
521                 if (statret == 0) {
522                         rprintf(FERROR,
523                             "ERROR: destination path is not a directory\n");
524                         exit_cleanup(RERR_SYNTAX);
525                 }
526
527                 if (mkdir_defmode(dest_path) != 0) {
528                         rsyserr(FERROR, errno, "mkdir %s failed",
529                                 full_fname(dest_path));
530                         exit_cleanup(RERR_FILEIO);
531                 }
532
533                 new_root_dir = 1;
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 (!push_dir(dest_path, dry_run > 1)) {
544                         rsyserr(FERROR, errno, "push_dir#2 %s failed",
545                                 full_fname(dest_path));
546                         exit_cleanup(RERR_FILESELECT);
547                 }
548
549                 return NULL;
550         }
551
552         /* Otherwise, we are writing a single file, possibly on top of an
553          * existing non-directory.  Change to the item's parent directory
554          * (if it has a path component), return the basename of the
555          * destination file as the local name, and use mode 2. */
556         if (!cp)
557                 return dest_path;
558
559         if (cp == dest_path)
560                 dest_path = "/";
561
562         *cp = '\0';
563         if (!push_dir(dest_path, 0)) {
564                 rsyserr(FERROR, errno, "push_dir#3 %s failed",
565                         full_fname(dest_path));
566                 exit_cleanup(RERR_FILESELECT);
567         }
568         *cp = '/';
569
570         return cp + 1;
571 }
572
573 /* Call this if the destination dir (which is assumed to be in curr_dir)
574  * does not yet exist and we can't create it due to being in dry-run
575  * mode.  We'll fix dirs that can be relative to the non-existent dir. */
576 static void fix_basis_dirs(void)
577 {
578         char **dir, *new, *slash;
579         int len;
580
581         if (dry_run <= 1)
582                 return;
583
584         slash = strrchr(curr_dir, '/');
585
586         for (dir = basis_dir; *dir; dir++) {
587                 if (**dir == '/')
588                         continue;
589                 len = curr_dir_len + 1 + strlen(*dir) + 1;
590                 if (!(new = new_array(char, len)))
591                         out_of_memory("fix_basis_dirs");
592                 if (slash && strncmp(*dir, "../", 3) == 0) {
593                     /* We want to remove only one leading "../" prefix for
594                      * the directory we couldn't create in dry-run mode:
595                      * this ensures that any other ".." references get
596                      * evaluated the same as they would for a live copy. */
597                     *slash = '\0';
598                     pathjoin(new, len, curr_dir, *dir + 3);
599                     *slash = '/';
600                 } else
601                     pathjoin(new, len, curr_dir, *dir);
602                 *dir = new;
603         }
604 }
605
606 /* This is only called by the sender. */
607 static void read_final_goodbye(int f_in, int f_out)
608 {
609         int i;
610
611         if (protocol_version < 29)
612                 i = read_int(f_in);
613         else {
614                 while ((i = read_int(f_in)) == the_file_list->count
615                     && read_shortint(f_in) == ITEM_IS_NEW) {
616                         /* Forward the keep-alive (no-op) to the receiver. */
617                         write_int(f_out, the_file_list->count);
618                         write_shortint(f_out, ITEM_IS_NEW);
619                 }
620         }
621
622         if (i != NDX_DONE) {
623                 rprintf(FERROR, "Invalid packet at end of run (%d) [%s]\n",
624                         i, who_am_i());
625                 exit_cleanup(RERR_PROTOCOL);
626         }
627 }
628
629
630 static void do_server_sender(int f_in, int f_out, int argc, char *argv[])
631 {
632         struct file_list *flist;
633         char *dir = argv[0];
634
635         if (verbose > 2) {
636                 rprintf(FINFO, "server_sender starting pid=%ld\n",
637                         (long)getpid());
638         }
639
640         if (am_daemon && lp_write_only(module_id)) {
641                 rprintf(FERROR, "ERROR: module is write only\n");
642                 exit_cleanup(RERR_SYNTAX);
643                 return;
644         }
645         if (am_daemon && lp_read_only(module_id) && remove_source_files) {
646                 rprintf(FERROR,
647                     "ERROR: --remove-%s-files cannot be used with a read-only module\n",
648                     remove_source_files == 1 ? "source" : "sent");
649                 exit_cleanup(RERR_SYNTAX);
650                 return;
651         }
652
653         if (!relative_paths) {
654                 if (!push_dir(dir, 0)) {
655                         rsyserr(FERROR, errno, "push_dir#3 %s failed",
656                                 full_fname(dir));
657                         exit_cleanup(RERR_FILESELECT);
658                 }
659         }
660         argc--;
661         argv++;
662
663         if (argc == 0 && (recurse || list_only)) {
664                 argc = 1;
665                 argv--;
666                 argv[0] = ".";
667         }
668
669         flist = send_file_list(f_out,argc,argv);
670         if (!flist || flist->count == 0) {
671                 exit_cleanup(0);
672         }
673         the_file_list = flist;
674
675         io_start_buffering_in();
676         io_start_buffering_out();
677
678         send_files(flist,f_out,f_in);
679         io_flush(FULL_FLUSH);
680         handle_stats(f_out);
681         if (protocol_version >= 24)
682                 read_final_goodbye(f_in, f_out);
683         io_flush(FULL_FLUSH);
684         exit_cleanup(0);
685 }
686
687
688 static int do_recv(int f_in,int f_out,struct file_list *flist,char *local_name)
689 {
690         int pid;
691         int exit_code = 0;
692         int error_pipe[2];
693
694         /* The receiving side mustn't obey this, or an existing symlink that
695          * points to an identical file won't be replaced by the referent. */
696         copy_links = copy_dirlinks = 0;
697
698 #ifdef SUPPORT_HARD_LINKS
699         if (preserve_hard_links)
700                 init_hard_links();
701 #endif
702
703         if (fd_pair(error_pipe) < 0) {
704                 rsyserr(FERROR, errno, "pipe failed in do_recv");
705                 exit_cleanup(RERR_IPC);
706         }
707
708         io_flush(NORMAL_FLUSH);
709
710         if ((pid = do_fork()) == -1) {
711                 rsyserr(FERROR, errno, "fork failed in do_recv");
712                 exit_cleanup(RERR_IPC);
713         }
714
715         if (pid == 0) {
716                 close(error_pipe[0]);
717                 if (f_in != f_out)
718                         close(f_out);
719
720                 /* we can't let two processes write to the socket at one time */
721                 close_multiplexing_out();
722
723                 /* set place to send errors */
724                 set_msg_fd_out(error_pipe[1]);
725
726                 recv_files(f_in, flist, local_name);
727                 io_flush(FULL_FLUSH);
728                 handle_stats(f_in);
729
730                 send_msg(MSG_DONE, "", 0);
731                 io_flush(FULL_FLUSH);
732
733                 /* Handle any keep-alive packets from the post-processing work
734                  * that the generator does. */
735                 if (protocol_version >= 29) {
736                         kluge_around_eof = -1;
737
738                         /* This should only get stopped via a USR2 signal. */
739                         while (read_int(f_in) == flist->count
740                             && read_shortint(f_in) == ITEM_IS_NEW) {}
741
742                         rprintf(FERROR, "Invalid packet at end of run [%s]\n",
743                                 who_am_i());
744                         exit_cleanup(RERR_PROTOCOL);
745                 }
746
747                 /* Finally, we go to sleep until our parent kills us with a
748                  * USR2 signal.  We sleep for a short time, as on some OSes
749                  * a signal won't interrupt a sleep! */
750                 while (1)
751                         msleep(20);
752         }
753
754         am_generator = 1;
755         close_multiplexing_in();
756         if (write_batch && !am_server)
757                 stop_write_batch();
758
759         close(error_pipe[1]);
760         if (f_in != f_out)
761                 close(f_in);
762
763         io_start_buffering_out();
764
765         set_msg_fd_in(error_pipe[0]);
766
767         generate_files(f_out, flist, local_name);
768
769         handle_stats(-1);
770         io_flush(FULL_FLUSH);
771         if (protocol_version >= 24) {
772                 /* send a final goodbye message */
773                 write_int(f_out, NDX_DONE);
774         }
775         io_flush(FULL_FLUSH);
776
777         set_msg_fd_in(-1);
778         kill(pid, SIGUSR2);
779         wait_process_with_flush(pid, &exit_code);
780         return exit_code;
781 }
782
783
784 static void do_server_recv(int f_in, int f_out, int argc,char *argv[])
785 {
786         int exit_code;
787         struct file_list *flist;
788         char *local_name = NULL;
789         char *dir = NULL;
790         int save_verbose = verbose;
791
792         if (filesfrom_fd >= 0) {
793                 /* We can't mix messages with files-from data on the socket,
794                  * so temporarily turn off verbose messages. */
795                 verbose = 0;
796         }
797
798         if (verbose > 2) {
799                 rprintf(FINFO, "server_recv(%d) starting pid=%ld\n",
800                         argc, (long)getpid());
801         }
802
803         if (am_daemon && lp_read_only(module_id)) {
804                 rprintf(FERROR,"ERROR: module is read only\n");
805                 exit_cleanup(RERR_SYNTAX);
806                 return;
807         }
808
809         if (argc > 0) {
810                 dir = argv[0];
811                 argc--;
812                 argv++;
813                 if (!am_daemon && !push_dir(dir, 0)) {
814                         rsyserr(FERROR, errno, "push_dir#4 %s failed",
815                                 full_fname(dir));
816                         exit_cleanup(RERR_FILESELECT);
817                 }
818         }
819
820         io_start_buffering_in();
821         recv_filter_list(f_in);
822
823         if (filesfrom_fd >= 0) {
824                 /* We need to send the files-from names to the sender at the
825                  * same time that we receive the file-list from them, so we
826                  * need the IO routines to automatically write out the names
827                  * onto our f_out socket as we read the file-list.  This
828                  * avoids both deadlock and extra delays/buffers. */
829                 io_set_filesfrom_fds(filesfrom_fd, f_out);
830                 filesfrom_fd = -1;
831         }
832
833         flist = recv_file_list(f_in);
834         verbose = save_verbose;
835         if (!flist) {
836                 rprintf(FERROR,"server_recv: recv_file_list error\n");
837                 exit_cleanup(RERR_FILESELECT);
838         }
839         the_file_list = flist;
840
841         if (argc > 0)
842                 local_name = get_local_name(flist,argv[0]);
843
844         /* Now that we know what our destination directory turned out to be,
845          * we can sanitize the --link-/copy-/compare-dest args correctly. */
846         if (sanitize_paths) {
847                 char **dir;
848                 for (dir = basis_dir; *dir; dir++) {
849                         *dir = sanitize_path(NULL, *dir, NULL, curr_dir_depth, NULL);
850                 }
851                 if (partial_dir) {
852                         partial_dir = sanitize_path(NULL, partial_dir, NULL, curr_dir_depth, NULL);
853                 }
854         }
855         fix_basis_dirs();
856
857         if (server_filter_list.head) {
858                 char **dir;
859                 struct filter_list_struct *elp = &server_filter_list;
860
861                 for (dir = basis_dir; *dir; dir++) {
862                         if (check_filter(elp, *dir, 1) < 0)
863                                 goto options_rejected;
864                 }
865                 if (partial_dir && *partial_dir == '/'
866                  && check_filter(elp, partial_dir, 1) < 0) {
867                     options_rejected:
868                         rprintf(FERROR,
869                                 "Your options have been rejected by the server.\n");
870                         exit_cleanup(RERR_SYNTAX);
871                 }
872         }
873
874         exit_code = do_recv(f_in,f_out,flist,local_name);
875         exit_cleanup(exit_code);
876 }
877
878
879 int child_main(int argc, char *argv[])
880 {
881         start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv);
882         return 0;
883 }
884
885
886 void start_server(int f_in, int f_out, int argc, char *argv[])
887 {
888         set_nonblocking(f_in);
889         set_nonblocking(f_out);
890
891         io_set_sock_fds(f_in, f_out);
892         setup_protocol(f_out, f_in);
893 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
894         setup_iconv();
895 #endif
896
897         if (protocol_version >= 23)
898                 io_start_multiplex_out();
899
900         if (am_sender) {
901                 keep_dirlinks = 0; /* Must be disabled on the sender. */
902                 if (need_messages_from_generator)
903                         io_start_multiplex_in();
904
905                 recv_filter_list(f_in);
906                 do_server_sender(f_in, f_out, argc, argv);
907         } else {
908                 do_server_recv(f_in, f_out, argc, argv);
909         }
910         exit_cleanup(0);
911 }
912
913
914 /*
915  * This is called once the connection has been negotiated.  It is used
916  * for rsyncd, remote-shell, and local connections.
917  */
918 int client_run(int f_in, int f_out, pid_t pid, int argc, char *argv[])
919 {
920         struct file_list *flist = NULL;
921         int exit_code = 0, exit_code2 = 0;
922         char *local_name = NULL;
923
924         cleanup_child_pid = pid;
925         if (!read_batch) {
926                 set_nonblocking(f_in);
927                 set_nonblocking(f_out);
928         }
929
930         io_set_sock_fds(f_in, f_out);
931         setup_protocol(f_out,f_in);
932 #if defined HAVE_ICONV_OPEN && defined HAVE_ICONV_H
933         setup_iconv();
934 #endif
935
936         if (protocol_version >= 23 && !read_batch)
937                 io_start_multiplex_in();
938
939         /* We set our stderr file handle to blocking because ssh might have
940          * set it to non-blocking.  This can be particularly troublesome if
941          * stderr is a clone of stdout, because ssh would have set our stdout
942          * to non-blocking at the same time (which can easily cause us to lose
943          * output from our print statements).  This kluge shouldn't cause ssh
944          * any problems for how we use it.  Note also that we delayed setting
945          * this until after the above protocol setup so that we know for sure
946          * that ssh is done twiddling its file descriptors.  */
947         set_blocking(STDERR_FILENO);
948
949         if (am_sender) {
950                 keep_dirlinks = 0; /* Must be disabled on the sender. */
951                 io_start_buffering_out();
952                 if (!filesfrom_host)
953                         set_msg_fd_in(f_in);
954                 send_filter_list(f_out);
955                 if (filesfrom_host)
956                         filesfrom_fd = f_in;
957
958                 if (write_batch && !am_server)
959                         start_write_batch(f_out);
960                 flist = send_file_list(f_out, argc, argv);
961                 set_msg_fd_in(-1);
962                 if (verbose > 3)
963                         rprintf(FINFO,"file list sent\n");
964                 the_file_list = flist;
965
966                 io_flush(NORMAL_FLUSH);
967                 send_files(flist,f_out,f_in);
968                 io_flush(FULL_FLUSH);
969                 handle_stats(-1);
970                 if (protocol_version >= 24)
971                         read_final_goodbye(f_in, f_out);
972                 if (pid != -1) {
973                         if (verbose > 3)
974                                 rprintf(FINFO,"client_run waiting on %d\n", (int) pid);
975                         io_flush(FULL_FLUSH);
976                         wait_process_with_flush(pid, &exit_code);
977                 }
978                 output_summary();
979                 io_flush(FULL_FLUSH);
980                 exit_cleanup(exit_code);
981         }
982
983         if (need_messages_from_generator && !read_batch)
984                 io_start_multiplex_out();
985
986         if (argc == 0)
987                 list_only |= 1;
988
989         send_filter_list(read_batch ? -1 : f_out);
990
991         if (filesfrom_fd >= 0) {
992                 io_set_filesfrom_fds(filesfrom_fd, f_out);
993                 filesfrom_fd = -1;
994         }
995
996         if (write_batch && !am_server)
997                 start_write_batch(f_in);
998         flist = recv_file_list(f_in);
999         the_file_list = flist;
1000
1001         if (flist && flist->count > 0) {
1002                 local_name = get_local_name(flist, argv[0]);
1003
1004                 fix_basis_dirs();
1005
1006                 exit_code2 = do_recv(f_in, f_out, flist, local_name);
1007         } else {
1008                 handle_stats(-1);
1009                 output_summary();
1010         }
1011
1012         if (pid != -1) {
1013                 if (verbose > 3)
1014                         rprintf(FINFO,"client_run2 waiting on %d\n", (int) pid);
1015                 io_flush(FULL_FLUSH);
1016                 wait_process_with_flush(pid, &exit_code);
1017         }
1018
1019         return MAX(exit_code, exit_code2);
1020 }
1021
1022 static int copy_argv (char *argv[])
1023 {
1024         int i;
1025
1026         for (i = 0; argv[i]; i++) {
1027                 if (!(argv[i] = strdup(argv[i]))) {
1028                         rprintf (FERROR, "out of memory at %s(%d)\n",
1029                                  __FILE__, __LINE__);
1030                         return RERR_MALLOC;
1031                 }
1032         }
1033
1034         return 0;
1035 }
1036
1037
1038 /**
1039  * Start a client for either type of remote connection.  Work out
1040  * whether the arguments request a remote shell or rsyncd connection,
1041  * and call the appropriate connection function, then run_client.
1042  *
1043  * Calls either start_socket_client (for sockets) or do_cmd and
1044  * client_run (for ssh).
1045  **/
1046 static int start_client(int argc, char *argv[])
1047 {
1048         char *p;
1049         char *shell_machine = NULL;
1050         char *shell_path = NULL;
1051         char *shell_user = NULL;
1052         int ret;
1053         pid_t pid;
1054         int f_in,f_out;
1055         int rc;
1056
1057         /* Don't clobber argv[] so that ps(1) can still show the right
1058          * command line. */
1059         if ((rc = copy_argv(argv)))
1060                 return rc;
1061
1062         if (!read_batch) { /* for read_batch, NO source is specified */
1063                 shell_path = check_for_hostspec(argv[0], &shell_machine, &rsync_port);
1064                 if (shell_path) { /* source is remote */
1065                         char *dummy1;
1066                         int dummy2;
1067                         if (--argc
1068                          && check_for_hostspec(argv[argc], &dummy1, &dummy2)) {
1069                                 rprintf(FERROR,
1070                                         "The source and destination cannot both be remote.\n");
1071                                 exit_cleanup(RERR_SYNTAX);
1072                         }
1073                         argv++;
1074                         if (filesfrom_host && *filesfrom_host
1075                             && strcmp(filesfrom_host, shell_machine) != 0) {
1076                                 rprintf(FERROR,
1077                                         "--files-from hostname is not the same as the transfer hostname\n");
1078                                 exit_cleanup(RERR_SYNTAX);
1079                         }
1080                         if (rsync_port) {
1081                                 if (!shell_cmd) {
1082                                         return start_socket_client(shell_machine,
1083                                                                    shell_path,
1084                                                                    argc, argv);
1085                                 }
1086                                 daemon_over_rsh = 1;
1087                         }
1088
1089                         am_sender = 0;
1090                 } else { /* source is local, check dest arg */
1091                         am_sender = 1;
1092
1093                         if (argc > 1)
1094                                 p = argv[--argc];
1095                         else {
1096                                 p = ".";
1097                                 list_only = 1;
1098                         }
1099
1100                         shell_path = check_for_hostspec(p, &shell_machine, &rsync_port);
1101                         if (shell_path && filesfrom_host && *filesfrom_host
1102                             && strcmp(filesfrom_host, shell_machine) != 0) {
1103                                 rprintf(FERROR,
1104                                         "--files-from hostname is not the same as the transfer hostname\n");
1105                                 exit_cleanup(RERR_SYNTAX);
1106                         }
1107                         if (!shell_path) { /* no hostspec found, so src & dest are local */
1108                                 local_server = 1;
1109                                 if (filesfrom_host) {
1110                                         rprintf(FERROR,
1111                                                 "--files-from cannot be remote when the transfer is local\n");
1112                                         exit_cleanup(RERR_SYNTAX);
1113                                 }
1114                                 shell_machine = NULL;
1115                                 shell_path = p;
1116                         } else if (rsync_port) {
1117                                 if (!shell_cmd) {
1118                                         return start_socket_client(shell_machine,
1119                                                                    shell_path,
1120                                                                    argc, argv);
1121                                 }
1122                                 daemon_over_rsh = 1;
1123                         }
1124                 }
1125         } else {  /* read_batch */
1126                 local_server = 1;
1127                 shell_path = argv[argc-1];
1128                 if (check_for_hostspec(shell_path, &shell_machine, &rsync_port)) {
1129                         rprintf(FERROR, "remote destination is not allowed with --read-batch\n");
1130                         exit_cleanup(RERR_SYNTAX);
1131                 }
1132         }
1133
1134         if (shell_machine) {
1135                 p = strrchr(shell_machine,'@');
1136                 if (p) {
1137                         *p = 0;
1138                         shell_user = shell_machine;
1139                         shell_machine = p+1;
1140                 }
1141         }
1142
1143         if (verbose > 3) {
1144                 rprintf(FINFO,"cmd=%s machine=%s user=%s path=%s\n",
1145                         shell_cmd ? shell_cmd : "",
1146                         shell_machine ? shell_machine : "",
1147                         shell_user ? shell_user : "",
1148                         shell_path ? shell_path : "");
1149         }
1150
1151         /* for remote source, only single dest arg can remain ... */
1152         if (!am_sender && argc > 1) {
1153                 usage(FERROR);
1154                 exit_cleanup(RERR_SYNTAX);
1155         }
1156
1157         /* ... or no dest at all */
1158         if (!am_sender && argc == 0)
1159                 list_only |= 1;
1160
1161         pid = do_cmd(shell_cmd,shell_machine,shell_user,shell_path,
1162                      &f_in,&f_out);
1163
1164         /* if we're running an rsync server on the remote host over a
1165          * remote shell command, we need to do the RSYNCD protocol first */
1166         if (daemon_over_rsh) {
1167                 int tmpret;
1168                 tmpret = start_inband_exchange(shell_user, shell_path,
1169                                                f_in, f_out, argc);
1170                 if (tmpret < 0)
1171                         return tmpret;
1172         }
1173
1174         ret = client_run(f_in, f_out, pid, argc, argv);
1175
1176         fflush(stdout);
1177         fflush(stderr);
1178
1179         return ret;
1180 }
1181
1182
1183 static RETSIGTYPE sigusr1_handler(UNUSED(int val))
1184 {
1185         exit_cleanup(RERR_SIGNAL1);
1186 }
1187
1188 static RETSIGTYPE sigusr2_handler(UNUSED(int val))
1189 {
1190         if (!am_server)
1191                 output_summary();
1192         close_all();
1193         if (log_got_error)
1194                 _exit(RERR_PARTIAL);
1195         _exit(0);
1196 }
1197
1198 RETSIGTYPE remember_children(UNUSED(int val))
1199 {
1200 #ifdef WNOHANG
1201         int cnt, status;
1202         pid_t pid;
1203         /* An empty waitpid() loop was put here by Tridge and we could never
1204          * get him to explain why he put it in, so rather than taking it
1205          * out we're instead saving the child exit statuses for later use.
1206          * The waitpid() loop presumably eliminates all possibility of leaving
1207          * zombie children, maybe that's why he did it. */
1208         while ((pid = waitpid(-1, &status, WNOHANG)) > 0) {
1209                 /* save the child's exit status */
1210                 for (cnt = 0; cnt < MAXCHILDPROCS; cnt++) {
1211                         if (pid_stat_table[cnt].pid == 0) {
1212                                 pid_stat_table[cnt].pid = pid;
1213                                 pid_stat_table[cnt].status = status;
1214                                 break;
1215                         }
1216                 }
1217         }
1218 #endif
1219 #ifndef HAVE_SIGACTION
1220         signal(SIGCHLD, remember_children);
1221 #endif
1222 }
1223
1224
1225 /**
1226  * This routine catches signals and tries to send them to gdb.
1227  *
1228  * Because it's called from inside a signal handler it ought not to
1229  * use too many library routines.
1230  *
1231  * @todo Perhaps use "screen -X" instead/as well, to help people
1232  * debugging without easy access to X.  Perhaps use an environment
1233  * variable, or just call a script?
1234  *
1235  * @todo The /proc/ magic probably only works on Linux (and
1236  * Solaris?)  Can we be more portable?
1237  **/
1238 #ifdef MAINTAINER_MODE
1239 const char *get_panic_action(void)
1240 {
1241         const char *cmd_fmt = getenv("RSYNC_PANIC_ACTION");
1242
1243         if (cmd_fmt)
1244                 return cmd_fmt;
1245         else
1246                 return "xterm -display :0 -T Panic -n Panic "
1247                         "-e gdb /proc/%d/exe %d";
1248 }
1249
1250
1251 /**
1252  * Handle a fatal signal by launching a debugger, controlled by $RSYNC_PANIC_ACTION.
1253  *
1254  * This signal handler is only installed if we were configured with
1255  * --enable-maintainer-mode.  Perhaps it should always be on and we
1256  * should just look at the environment variable, but I'm a bit leery
1257  * of a signal sending us into a busy loop.
1258  **/
1259 static RETSIGTYPE rsync_panic_handler(UNUSED(int whatsig))
1260 {
1261         char cmd_buf[300];
1262         int ret;
1263
1264         snprintf(cmd_buf, sizeof cmd_buf, get_panic_action(),
1265                  getpid(), getpid());
1266
1267         /* Unless we failed to execute gdb, we allow the process to
1268          * continue.  I'm not sure if that's right. */
1269         ret = system(cmd_buf);
1270         if (ret)
1271                 _exit(ret);
1272 }
1273 #endif
1274
1275
1276 int main(int argc,char *argv[])
1277 {
1278         int ret;
1279         int orig_argc = argc;
1280         char **orig_argv = argv;
1281 #ifdef HAVE_SIGACTION
1282 # ifdef HAVE_SIGPROCMASK
1283         sigset_t sigmask;
1284
1285         sigemptyset(&sigmask);
1286 # endif
1287         sigact.sa_flags = SA_NOCLDSTOP;
1288 #endif
1289         SIGACTMASK(SIGUSR1, sigusr1_handler);
1290         SIGACTMASK(SIGUSR2, sigusr2_handler);
1291         SIGACTMASK(SIGCHLD, remember_children);
1292 #ifdef MAINTAINER_MODE
1293         SIGACTMASK(SIGSEGV, rsync_panic_handler);
1294         SIGACTMASK(SIGFPE, rsync_panic_handler);
1295         SIGACTMASK(SIGABRT, rsync_panic_handler);
1296         SIGACTMASK(SIGBUS, rsync_panic_handler);
1297 #endif
1298
1299         starttime = time(NULL);
1300         am_root = (MY_UID() == 0);
1301
1302         memset(&stats, 0, sizeof(stats));
1303
1304         if (argc < 2) {
1305                 usage(FERROR);
1306                 exit_cleanup(RERR_SYNTAX);
1307         }
1308
1309         /* we set a 0 umask so that correct file permissions can be
1310          * carried across */
1311         orig_umask = umask(0);
1312
1313 #if defined CONFIG_LOCALE && defined HAVE_SETLOCALE
1314         setlocale(LC_CTYPE, "");
1315 #endif
1316
1317         if (!parse_arguments(&argc, (const char ***) &argv, 1)) {
1318                 /* FIXME: We ought to call the same error-handling
1319                  * code here, rather than relying on getopt. */
1320                 option_error();
1321                 exit_cleanup(RERR_SYNTAX);
1322         }
1323
1324         SIGACTMASK(SIGINT, sig_int);
1325         SIGACTMASK(SIGHUP, sig_int);
1326         SIGACTMASK(SIGTERM, sig_int);
1327 #if defined HAVE_SIGACTION && HAVE_SIGPROCMASK
1328         sigprocmask(SIG_UNBLOCK, &sigmask, NULL);
1329 #endif
1330
1331         /* Ignore SIGPIPE; we consistently check error codes and will
1332          * see the EPIPE. */
1333         SIGACTION(SIGPIPE, SIG_IGN);
1334 #ifdef SIGXFSZ
1335         SIGACTION(SIGXFSZ, SIG_IGN);
1336 #endif
1337
1338         /* Initialize push_dir here because on some old systems getcwd
1339          * (implemented by forking "pwd" and reading its output) doesn't
1340          * work when there are other child processes.  Also, on all systems
1341          * that implement getcwd that way "pwd" can't be found after chroot. */
1342         push_dir(NULL, 0);
1343
1344         init_flist();
1345
1346         if ((write_batch || read_batch) && !am_server) {
1347                 if (write_batch)
1348                         write_batch_shell_file(orig_argc, orig_argv, argc);
1349
1350                 if (read_batch && strcmp(batch_name, "-") == 0)
1351                         batch_fd = STDIN_FILENO;
1352                 else {
1353                         batch_fd = do_open(batch_name,
1354                                    write_batch ? O_WRONLY | O_CREAT | O_TRUNC
1355                                    : O_RDONLY, S_IRUSR | S_IWUSR);
1356                 }
1357                 if (batch_fd < 0) {
1358                         rsyserr(FERROR, errno, "Batch file %s open error",
1359                                 full_fname(batch_name));
1360                         exit_cleanup(RERR_FILEIO);
1361                 }
1362                 if (read_batch)
1363                         read_stream_flags(batch_fd);
1364         }
1365         if (write_batch < 0)
1366                 dry_run = 1;
1367
1368         if (am_daemon && !am_server)
1369                 return daemon_main();
1370
1371         if (argc < 1) {
1372                 usage(FERROR);
1373                 exit_cleanup(RERR_SYNTAX);
1374         }
1375
1376         if (am_server) {
1377                 set_nonblocking(STDIN_FILENO);
1378                 set_nonblocking(STDOUT_FILENO);
1379                 if (am_daemon)
1380                         return start_daemon(STDIN_FILENO, STDOUT_FILENO);
1381                 start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv);
1382         }
1383
1384         ret = start_client(argc, argv);
1385         if (ret == -1)
1386                 exit_cleanup(RERR_STARTCLIENT);
1387         else
1388                 exit_cleanup(ret);
1389
1390         return ret;
1391 }