Added logic to the receiving side to ensure that the --delete-during
[rsync/rsync.git] / io.c
1 /*
2  * Socket and pipe I/O utilities used in rsync.
3  *
4  * Copyright (C) 1996-2001 Andrew Tridgell
5  * Copyright (C) 1996 Paul Mackerras
6  * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7  * Copyright (C) 2003-2008 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 3 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, visit the http://fsf.org website.
21  */
22
23 /* Rsync provides its own multiplexing system, which is used to send
24  * stderr and stdout over a single socket.
25  *
26  * For historical reasons this is off during the start of the
27  * connection, but it's switched on quite early using
28  * io_start_multiplex_out() and io_start_multiplex_in(). */
29
30 #include "rsync.h"
31 #include "ifuncs.h"
32
33 /** If no timeout is specified then use a 60 second select timeout */
34 #define SELECT_TIMEOUT 60
35
36 extern int bwlimit;
37 extern size_t bwlimit_writemax;
38 extern int io_timeout;
39 extern int am_server;
40 extern int am_daemon;
41 extern int am_sender;
42 extern int am_generator;
43 extern int inc_recurse;
44 extern int io_error;
45 extern int eol_nulls;
46 extern int flist_eof;
47 extern int list_only;
48 extern int read_batch;
49 extern int protect_args;
50 extern int checksum_seed;
51 extern int protocol_version;
52 extern int remove_source_files;
53 extern int preserve_hard_links;
54 extern struct stats stats;
55 extern struct file_list *cur_flist;
56 #ifdef ICONV_OPTION
57 extern int filesfrom_convert;
58 extern iconv_t ic_send, ic_recv;
59 #endif
60
61 int csum_length = SHORT_SUM_LENGTH; /* initial value */
62 int allowed_lull = 0;
63 int ignore_timeout = 0;
64 int batch_fd = -1;
65 int msgdone_cnt = 0;
66 int check_for_io_err = 0;
67
68 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
69 int kluge_around_eof = 0;
70
71 int msg_fd_in = -1;
72 int msg_fd_out = -1;
73 int sock_f_in = -1;
74 int sock_f_out = -1;
75
76 static int iobuf_f_in = -1;
77 static char *iobuf_in;
78 static size_t iobuf_in_siz;
79 static size_t iobuf_in_ndx;
80 static size_t iobuf_in_remaining;
81
82 static int iobuf_f_out = -1;
83 static char *iobuf_out;
84 static int iobuf_out_cnt;
85
86 int flist_forward_from = -1;
87
88 static int io_multiplexing_out;
89 static int io_multiplexing_in;
90 static time_t last_io_in;
91 static time_t last_io_out;
92 static int no_flush;
93
94 static int write_batch_monitor_in = -1;
95 static int write_batch_monitor_out = -1;
96
97 static int io_filesfrom_f_in = -1;
98 static int io_filesfrom_f_out = -1;
99 static xbuf ff_buf = EMPTY_XBUF;
100 static char ff_lastchar;
101 #ifdef ICONV_OPTION
102 static xbuf iconv_buf = EMPTY_XBUF;
103 #endif
104 static int defer_forwarding_messages = 0, defer_forwarding_keep = 0;
105 static int select_timeout = SELECT_TIMEOUT;
106 static int active_filecnt = 0;
107 static OFF_T active_bytecnt = 0;
108 static int first_message = 1;
109
110 static char int_byte_extra[64] = {
111         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
112         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
113         1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
114         2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
115 };
116
117 #define REMOTE_OPTION_ERROR "rsync: on remote machine: -"
118 #define REMOTE_OPTION_ERROR2 ": unknown option"
119
120 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
121
122 static void readfd(int fd, char *buffer, size_t N);
123 static void writefd(int fd, const char *buf, size_t len);
124 static void writefd_unbuffered(int fd, const char *buf, size_t len);
125 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
126
127 struct flist_ndx_item {
128         struct flist_ndx_item *next;
129         int ndx;
130 };
131
132 struct flist_ndx_list {
133         struct flist_ndx_item *head, *tail;
134 };
135
136 static struct flist_ndx_list redo_list, hlink_list;
137
138 struct msg_list_item {
139         struct msg_list_item *next;
140         char convert;
141         char buf[1];
142 };
143
144 struct msg_list {
145         struct msg_list_item *head, *tail;
146 };
147
148 static struct msg_list msg_queue;
149
150 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
151 {
152         struct flist_ndx_item *item;
153
154         if (!(item = new(struct flist_ndx_item)))
155                 out_of_memory("flist_ndx_push");
156         item->next = NULL;
157         item->ndx = ndx;
158         if (lp->tail)
159                 lp->tail->next = item;
160         else
161                 lp->head = item;
162         lp->tail = item;
163 }
164
165 static int flist_ndx_pop(struct flist_ndx_list *lp)
166 {
167         struct flist_ndx_item *next;
168         int ndx;
169
170         if (!lp->head)
171                 return -1;
172
173         ndx = lp->head->ndx;
174         next = lp->head->next;
175         free(lp->head);
176         lp->head = next;
177         if (!next)
178                 lp->tail = NULL;
179
180         return ndx;
181 }
182
183 static void got_flist_entry_status(enum festatus status, const char *buf)
184 {
185         int ndx = IVAL(buf, 0);
186         struct file_list *flist = flist_for_ndx(ndx);
187
188         assert(flist != NULL);
189
190         if (remove_source_files) {
191                 active_filecnt--;
192                 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
193         }
194
195         if (inc_recurse)
196                 flist->in_progress--;
197
198         switch (status) {
199         case FES_SUCCESS:
200                 if (remove_source_files)
201                         send_msg(MSG_SUCCESS, buf, 4, 0);
202                 if (preserve_hard_links) {
203                         struct file_struct *file = flist->files[ndx - flist->ndx_start];
204                         if (F_IS_HLINKED(file)) {
205                                 flist_ndx_push(&hlink_list, ndx);
206                                 flist->in_progress++;
207                         }
208                 }
209                 break;
210         case FES_REDO:
211                 if (inc_recurse)
212                         flist->to_redo++;
213                 flist_ndx_push(&redo_list, ndx);
214                 break;
215         case FES_NO_SEND:
216                 break;
217         }
218 }
219
220 static void check_timeout(void)
221 {
222         time_t t;
223
224         if (!io_timeout || ignore_timeout)
225                 return;
226
227         if (!last_io_in) {
228                 last_io_in = time(NULL);
229                 return;
230         }
231
232         t = time(NULL);
233
234         if (t - last_io_in >= io_timeout) {
235                 if (!am_server && !am_daemon) {
236                         rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
237                                 (int)(t-last_io_in));
238                 }
239                 exit_cleanup(RERR_TIMEOUT);
240         }
241 }
242
243 /* Note the fds used for the main socket (which might really be a pipe
244  * for a local transfer, but we can ignore that). */
245 void io_set_sock_fds(int f_in, int f_out)
246 {
247         sock_f_in = f_in;
248         sock_f_out = f_out;
249 }
250
251 void set_io_timeout(int secs)
252 {
253         io_timeout = secs;
254
255         if (!io_timeout || io_timeout > SELECT_TIMEOUT)
256                 select_timeout = SELECT_TIMEOUT;
257         else
258                 select_timeout = io_timeout;
259
260         allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
261 }
262
263 /* Setup the fd used to receive MSG_* messages.  Only needed during the
264  * early stages of being a local sender (up through the sending of the
265  * file list) or when we're the generator (to fetch the messages from
266  * the receiver). */
267 void set_msg_fd_in(int fd)
268 {
269         msg_fd_in = fd;
270 }
271
272 /* Setup the fd used to send our MSG_* messages.  Only needed when
273  * we're the receiver (to send our messages to the generator). */
274 void set_msg_fd_out(int fd)
275 {
276         msg_fd_out = fd;
277         set_nonblocking(msg_fd_out);
278 }
279
280 /* Add a message to the pending MSG_* list. */
281 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
282 {
283         struct msg_list_item *m;
284         int sz = len + 4 + sizeof m[0] - 1;
285
286         if (!(m = (struct msg_list_item *)new_array(char, sz)))
287                 out_of_memory("msg_list_add");
288         m->next = NULL;
289         m->convert = convert;
290         SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
291         memcpy(m->buf + 4, buf, len);
292         if (lst->tail)
293                 lst->tail->next = m;
294         else
295                 lst->head = m;
296         lst->tail = m;
297 }
298
299 static void msg_flush(void)
300 {
301         if (am_generator) {
302                 while (msg_queue.head && io_multiplexing_out) {
303                         struct msg_list_item *m = msg_queue.head;
304                         int len = IVAL(m->buf, 0) & 0xFFFFFF;
305                         int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
306                         if (!(msg_queue.head = m->next))
307                                 msg_queue.tail = NULL;
308                         stats.total_written += len + 4;
309                         defer_forwarding_messages++;
310                         mplex_write(sock_f_out, tag, m->buf + 4, len, m->convert);
311                         defer_forwarding_messages--;
312                         free(m);
313                 }
314         } else {
315                 while (msg_queue.head) {
316                         struct msg_list_item *m = msg_queue.head;
317                         int len = IVAL(m->buf, 0) & 0xFFFFFF;
318                         int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
319                         if (!(msg_queue.head = m->next))
320                                 msg_queue.tail = NULL;
321                         defer_forwarding_messages++;
322                         mplex_write(msg_fd_out, tag, m->buf + 4, len, m->convert);
323                         defer_forwarding_messages--;
324                         free(m);
325                 }
326         }
327 }
328
329 static void check_for_d_option_error(const char *msg)
330 {
331         static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
332         char *colon;
333         int saw_d = 0;
334
335         if (*msg != 'r'
336          || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
337                 return;
338
339         msg += sizeof REMOTE_OPTION_ERROR - 1;
340         if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
341          || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
342                 return;
343
344         for ( ; *msg != ':'; msg++) {
345                 if (*msg == 'd')
346                         saw_d = 1;
347                 else if (*msg == 'e')
348                         break;
349                 else if (strchr(rsync263_opts, *msg) == NULL)
350                         return;
351         }
352
353         if (saw_d) {
354                 rprintf(FWARNING,
355                     "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
356         }
357 }
358
359 /* Read a message from the MSG_* fd and handle it.  This is called either
360  * during the early stages of being a local sender (up through the sending
361  * of the file list) or when we're the generator (to fetch the messages
362  * from the receiver). */
363 static void read_msg_fd(void)
364 {
365         char buf[2048];
366         size_t n;
367         struct file_list *flist;
368         int fd = msg_fd_in;
369         int tag, len;
370
371         /* Temporarily disable msg_fd_in.  This is needed to avoid looping back
372          * to this routine from writefd_unbuffered(). */
373         no_flush++;
374         msg_fd_in = -1;
375         defer_forwarding_messages++;
376
377         readfd(fd, buf, 4);
378         tag = IVAL(buf, 0);
379
380         len = tag & 0xFFFFFF;
381         tag = (tag >> 24) - MPLEX_BASE;
382
383         check_for_io_err = 0;
384
385         switch (tag) {
386         case MSG_DONE:
387                 if (len < 0 || len > 1 || !am_generator) {
388                   invalid_msg:
389                         rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
390                                 tag, len, who_am_i(),
391                                 inc_recurse ? "/inc" : "");
392                         exit_cleanup(RERR_STREAMIO);
393                 }
394                 if (len) {
395                         readfd(fd, buf, len);
396                         stats.total_read = read_varlong(fd, 3);
397                 }
398                 msgdone_cnt++;
399                 break;
400         case MSG_REDO:
401                 if (len != 4 || !am_generator)
402                         goto invalid_msg;
403                 readfd(fd, buf, 4);
404                 got_flist_entry_status(FES_REDO, buf);
405                 break;
406         case MSG_FLIST:
407                 if (len != 4 || !am_generator || !inc_recurse)
408                         goto invalid_msg;
409                 readfd(fd, buf, 4);
410                 /* Read extra file list from receiver. */
411                 assert(iobuf_in != NULL);
412                 assert(iobuf_f_in == fd);
413                 if (DEBUG_GTE(FLIST, 2)) {
414                         rprintf(FINFO, "[%s] receiving flist for dir %d\n",
415                                 who_am_i(), IVAL(buf,0));
416                 }
417                 flist = recv_file_list(fd);
418                 flist->parent_ndx = IVAL(buf,0);
419                 /* If the sender is going to send us an MSG_IO_ERROR value, it
420                  * will always be the very next message following MSG_FLIST. */
421                 check_for_io_err = 1;
422 #ifdef SUPPORT_HARD_LINKS
423                 if (preserve_hard_links)
424                         match_hard_links(flist);
425 #endif
426                 break;
427         case MSG_FLIST_EOF:
428                 if (len != 0 || !am_generator || !inc_recurse)
429                         goto invalid_msg;
430                 flist_eof = 1;
431                 break;
432         case MSG_IO_ERROR:
433                 if (len != 4)
434                         goto invalid_msg;
435                 readfd(fd, buf, len);
436                 io_error |= IVAL(buf, 0);
437                 break;
438         case MSG_DELETED:
439                 if (len >= (int)sizeof buf || !am_generator)
440                         goto invalid_msg;
441                 readfd(fd, buf, len);
442                 send_msg(MSG_DELETED, buf, len, 1);
443                 break;
444         case MSG_SUCCESS:
445                 if (len != 4 || !am_generator)
446                         goto invalid_msg;
447                 readfd(fd, buf, 4);
448                 got_flist_entry_status(FES_SUCCESS, buf);
449                 break;
450         case MSG_NO_SEND:
451                 if (len != 4 || !am_generator)
452                         goto invalid_msg;
453                 readfd(fd, buf, 4);
454                 got_flist_entry_status(FES_NO_SEND, buf);
455                 break;
456         case MSG_ERROR_SOCKET:
457         case MSG_CLIENT:
458                 if (!am_generator)
459                         goto invalid_msg;
460                 if (tag == MSG_ERROR_SOCKET)
461                         io_end_multiplex_out();
462                 /* FALL THROUGH */
463         case MSG_INFO:
464         case MSG_ERROR:
465         case MSG_ERROR_XFER:
466         case MSG_WARNING:
467         case MSG_LOG:
468                 while (len) {
469                         n = len;
470                         if (n >= sizeof buf)
471                                 n = sizeof buf - 1;
472                         readfd(fd, buf, n);
473                         rwrite((enum logcode)tag, buf, n, !am_generator);
474                         len -= n;
475                 }
476                 break;
477         default:
478                 rprintf(FERROR, "unknown message %d:%d [%s]\n",
479                         tag, len, who_am_i());
480                 exit_cleanup(RERR_STREAMIO);
481         }
482
483         no_flush--;
484         msg_fd_in = fd;
485         if (!--defer_forwarding_messages && !no_flush)
486                 msg_flush();
487 }
488
489 /* This is used by the generator to limit how many file transfers can
490  * be active at once when --remove-source-files is specified.  Without
491  * this, sender-side deletions were mostly happening at the end. */
492 void increment_active_files(int ndx, int itemizing, enum logcode code)
493 {
494         /* TODO: tune these limits? */
495         while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
496                 check_for_finished_files(itemizing, code, 0);
497                 if (iobuf_out_cnt)
498                         io_flush(NORMAL_FLUSH);
499                 else
500                         read_msg_fd();
501         }
502
503         active_filecnt++;
504         active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
505 }
506
507 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
508 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
509 {
510         char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
511         size_t n = len;
512
513 #ifdef ICONV_OPTION
514         /* We need to convert buf before doing anything else so that we
515          * can include the (converted) byte length in the message header. */
516         if (convert && ic_send != (iconv_t)-1) {
517                 xbuf outbuf, inbuf;
518
519                 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
520                 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
521
522                 iconvbufs(ic_send, &inbuf, &outbuf,
523                           ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
524                 if (inbuf.len > 0) {
525                         rprintf(FERROR, "overflowed conversion buffer in mplex_write");
526                         exit_cleanup(RERR_UNSUPPORTED);
527                 }
528
529                 n = len = outbuf.len;
530         } else
531 #endif
532         if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
533                 n = 0;    /* We'd rather do 2 writes than too much memcpy(). */
534         else
535                 memcpy(buffer + 4, buf, n);
536
537         SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
538
539         defer_forwarding_keep = 1; /* defer_forwarding_messages++ on return */
540         writefd_unbuffered(fd, buffer, n+4);
541         defer_forwarding_keep = 0;
542
543         if (len > n)
544                 writefd_unbuffered(fd, buf+n, len-n);
545
546         if (!--defer_forwarding_messages && !no_flush)
547                 msg_flush();
548 }
549
550 int send_msg(enum msgcode code, const char *buf, int len, int convert)
551 {
552         if (msg_fd_out < 0) {
553                 if (!defer_forwarding_messages)
554                         return io_multiplex_write(code, buf, len, convert);
555                 if (!io_multiplexing_out)
556                         return 0;
557                 msg_list_add(&msg_queue, code, buf, len, convert);
558                 return 1;
559         }
560         if (flist_forward_from >= 0)
561                 msg_list_add(&msg_queue, code, buf, len, convert);
562         else
563                 mplex_write(msg_fd_out, code, buf, len, convert);
564         return 1;
565 }
566
567 void send_msg_int(enum msgcode code, int num)
568 {
569         char numbuf[4];
570         SIVAL(numbuf, 0, num);
571         send_msg(code, numbuf, 4, 0);
572 }
573
574 void wait_for_receiver(void)
575 {
576         if (iobuf_out_cnt)
577                 io_flush(NORMAL_FLUSH);
578         else
579                 read_msg_fd();
580 }
581
582 int get_redo_num(void)
583 {
584         return flist_ndx_pop(&redo_list);
585 }
586
587 int get_hlink_num(void)
588 {
589         return flist_ndx_pop(&hlink_list);
590 }
591
592 /**
593  * When we're the receiver and we have a local --files-from list of names
594  * that needs to be sent over the socket to the sender, we have to do two
595  * things at the same time: send the sender a list of what files we're
596  * processing and read the incoming file+info list from the sender.  We do
597  * this by augmenting the read_timeout() function to copy this data.  It
598  * uses ff_buf to read a block of data from f_in (when it is ready, since
599  * it might be a pipe) and then blast it out f_out (when it is ready to
600  * receive more data).
601  */
602 void io_set_filesfrom_fds(int f_in, int f_out)
603 {
604         io_filesfrom_f_in = f_in;
605         io_filesfrom_f_out = f_out;
606         alloc_xbuf(&ff_buf, 2048);
607 #ifdef ICONV_OPTION
608         if (protect_args)
609                 alloc_xbuf(&iconv_buf, 1024);
610 #endif
611 }
612
613 /* It's almost always an error to get an EOF when we're trying to read from the
614  * network, because the protocol is (for the most part) self-terminating.
615  *
616  * There is one case for the receiver when it is at the end of the transfer
617  * (hanging around reading any keep-alive packets that might come its way): if
618  * the sender dies before the generator's kill-signal comes through, we can end
619  * up here needing to loop until the kill-signal arrives.  In this situation,
620  * kluge_around_eof will be < 0.
621  *
622  * There is another case for older protocol versions (< 24) where the module
623  * listing was not terminated, so we must ignore an EOF error in that case and
624  * exit.  In this situation, kluge_around_eof will be > 0. */
625 static void whine_about_eof(int fd)
626 {
627         if (kluge_around_eof && fd == sock_f_in) {
628                 int i;
629                 if (kluge_around_eof > 0)
630                         exit_cleanup(0);
631                 /* If we're still here after 10 seconds, exit with an error. */
632                 for (i = 10*1000/20; i--; )
633                         msleep(20);
634         }
635
636         rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
637                 "(%s bytes received so far) [%s]\n",
638                 big_num(stats.total_read, 0), who_am_i());
639
640         exit_cleanup(RERR_STREAMIO);
641 }
642
643 /**
644  * Read from a socket with I/O timeout. return the number of bytes
645  * read. If no bytes can be read then exit, never return a number <= 0.
646  *
647  * TODO: If the remote shell connection fails, then current versions
648  * actually report an "unexpected EOF" error here.  Since it's a
649  * fairly common mistake to try to use rsh when ssh is required, we
650  * should trap that: if we fail to read any data at all, we should
651  * give a better explanation.  We can tell whether the connection has
652  * started by looking e.g. at whether the remote version is known yet.
653  */
654 static int read_timeout(int fd, char *buf, size_t len)
655 {
656         int n, cnt = 0;
657
658         io_flush(FULL_FLUSH);
659
660         while (cnt == 0) {
661                 /* until we manage to read *something* */
662                 fd_set r_fds, w_fds;
663                 struct timeval tv;
664                 int maxfd = fd;
665                 int count;
666
667                 FD_ZERO(&r_fds);
668                 FD_ZERO(&w_fds);
669                 FD_SET(fd, &r_fds);
670                 if (io_filesfrom_f_out >= 0) {
671                         int new_fd;
672                         if (ff_buf.len == 0) {
673                                 if (io_filesfrom_f_in >= 0) {
674                                         FD_SET(io_filesfrom_f_in, &r_fds);
675                                         new_fd = io_filesfrom_f_in;
676                                 } else {
677                                         io_filesfrom_f_out = -1;
678                                         new_fd = -1;
679                                 }
680                         } else {
681                                 FD_SET(io_filesfrom_f_out, &w_fds);
682                                 new_fd = io_filesfrom_f_out;
683                         }
684                         if (new_fd > maxfd)
685                                 maxfd = new_fd;
686                 }
687
688                 tv.tv_sec = select_timeout;
689                 tv.tv_usec = 0;
690
691                 errno = 0;
692
693                 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
694
695                 if (count <= 0) {
696                         if (errno == EBADF) {
697                                 defer_forwarding_messages = 0;
698                                 exit_cleanup(RERR_SOCKETIO);
699                         }
700                         check_timeout();
701                         continue;
702                 }
703
704                 if (io_filesfrom_f_out >= 0) {
705                         if (ff_buf.len) {
706                                 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
707                                         int l = write(io_filesfrom_f_out,
708                                                       ff_buf.buf + ff_buf.pos,
709                                                       ff_buf.len);
710                                         if (l > 0) {
711                                                 if (!(ff_buf.len -= l))
712                                                         ff_buf.pos = 0;
713                                                 else
714                                                         ff_buf.pos += l;
715                                         } else if (errno != EINTR) {
716                                                 /* XXX should we complain? */
717                                                 io_filesfrom_f_out = -1;
718                                         }
719                                 }
720                         } else if (io_filesfrom_f_in >= 0) {
721                                 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
722 #ifdef ICONV_OPTION
723                                         xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
724 #else
725                                         xbuf *ibuf = &ff_buf;
726 #endif
727                                         int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
728                                         if (l <= 0) {
729                                                 if (l == 0 || errno != EINTR) {
730                                                         /* Send end-of-file marker */
731                                                         memcpy(ff_buf.buf, "\0\0", 2);
732                                                         ff_buf.len = ff_lastchar? 2 : 1;
733                                                         ff_buf.pos = 0;
734                                                         io_filesfrom_f_in = -1;
735                                                 }
736                                         } else {
737 #ifdef ICONV_OPTION
738                                                 if (filesfrom_convert) {
739                                                         iconv_buf.pos = 0;
740                                                         iconv_buf.len = l;
741                                                         iconvbufs(ic_send, &iconv_buf, &ff_buf,
742                                                             ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
743                                                         l = ff_buf.len;
744                                                 }
745 #endif
746                                                 if (!eol_nulls) {
747                                                         char *s = ff_buf.buf + l;
748                                                         /* Transform CR and/or LF into '\0' */
749                                                         while (s-- > ff_buf.buf) {
750                                                                 if (*s == '\n' || *s == '\r')
751                                                                         *s = '\0';
752                                                         }
753                                                 }
754                                                 if (!ff_lastchar) {
755                                                         /* Last buf ended with a '\0', so don't
756                                                          * let this buf start with one. */
757                                                         while (l && ff_buf.buf[ff_buf.pos] == '\0')
758                                                                 ff_buf.pos++, l--;
759                                                 }
760                                                 if (!l)
761                                                         ff_buf.pos = 0;
762                                                 else {
763                                                         char *f = ff_buf.buf + ff_buf.pos;
764                                                         char *t = f;
765                                                         char *eob = f + l;
766                                                         /* Eliminate any multi-'\0' runs. */
767                                                         while (f != eob) {
768                                                                 if (!(*t++ = *f++)) {
769                                                                         while (f != eob && !*f)
770                                                                                 f++, l--;
771                                                                 }
772                                                         }
773                                                         ff_lastchar = f[-1];
774                                                 }
775                                                 ff_buf.len = l;
776                                         }
777                                 }
778                         }
779                 }
780
781                 if (!FD_ISSET(fd, &r_fds))
782                         continue;
783
784                 n = read(fd, buf, len);
785
786                 if (n <= 0) {
787                         if (n == 0)
788                                 whine_about_eof(fd); /* Doesn't return. */
789                         if (errno == EINTR || errno == EWOULDBLOCK
790                             || errno == EAGAIN)
791                                 continue;
792
793                         /* Don't write errors on a dead socket. */
794                         if (fd == sock_f_in) {
795                                 io_end_multiplex_out();
796                                 rsyserr(FERROR_SOCKET, errno, "read error");
797                         } else
798                                 rsyserr(FERROR, errno, "read error");
799                         exit_cleanup(RERR_STREAMIO);
800                 }
801
802                 buf += n;
803                 len -= n;
804                 cnt += n;
805
806                 if (fd == sock_f_in && io_timeout)
807                         last_io_in = time(NULL);
808         }
809
810         return cnt;
811 }
812
813 /* Read a line into the "buf" buffer. */
814 int read_line(int fd, char *buf, size_t bufsiz, int flags)
815 {
816         char ch, *s, *eob;
817         int cnt;
818
819 #ifdef ICONV_OPTION
820         if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
821                 realloc_xbuf(&iconv_buf, bufsiz + 1024);
822 #endif
823
824   start:
825 #ifdef ICONV_OPTION
826         s = flags & RL_CONVERT ? iconv_buf.buf : buf;
827 #else
828         s = buf;
829 #endif
830         eob = s + bufsiz - 1;
831         while (1) {
832                 cnt = read(fd, &ch, 1);
833                 if (cnt < 0 && (errno == EWOULDBLOCK
834                   || errno == EINTR || errno == EAGAIN)) {
835                         struct timeval tv;
836                         fd_set r_fds, e_fds;
837                         FD_ZERO(&r_fds);
838                         FD_SET(fd, &r_fds);
839                         FD_ZERO(&e_fds);
840                         FD_SET(fd, &e_fds);
841                         tv.tv_sec = select_timeout;
842                         tv.tv_usec = 0;
843                         if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
844                                 check_timeout();
845                         /*if (FD_ISSET(fd, &e_fds))
846                                 rprintf(FINFO, "select exception on fd %d\n", fd); */
847                         continue;
848                 }
849                 if (cnt != 1)
850                         break;
851                 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
852                         /* Skip empty lines if dumping comments. */
853                         if (flags & RL_DUMP_COMMENTS && s == buf)
854                                 continue;
855                         break;
856                 }
857                 if (s < eob)
858                         *s++ = ch;
859         }
860         *s = '\0';
861
862         if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
863                 goto start;
864
865 #ifdef ICONV_OPTION
866         if (flags & RL_CONVERT) {
867                 xbuf outbuf;
868                 INIT_XBUF(outbuf, buf, 0, bufsiz);
869                 iconv_buf.pos = 0;
870                 iconv_buf.len = s - iconv_buf.buf;
871                 iconvbufs(ic_recv, &iconv_buf, &outbuf,
872                           ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
873                 outbuf.buf[outbuf.len] = '\0';
874                 return outbuf.len;
875         }
876 #endif
877
878         return s - buf;
879 }
880
881 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
882                char ***argv_p, int *argc_p, char **request_p)
883 {
884         int maxargs = MAX_ARGS;
885         int dot_pos = 0;
886         int argc = 0;
887         char **argv, *p;
888         int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
889
890 #ifdef ICONV_OPTION
891         rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
892 #endif
893
894         if (!(argv = new_array(char *, maxargs)))
895                 out_of_memory("read_args");
896         if (mod_name && !protect_args)
897                 argv[argc++] = "rsyncd";
898
899         while (1) {
900                 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
901                         break;
902
903                 if (argc == maxargs-1) {
904                         maxargs += MAX_ARGS;
905                         if (!(argv = realloc_array(argv, char *, maxargs)))
906                                 out_of_memory("read_args");
907                 }
908
909                 if (dot_pos) {
910                         if (request_p) {
911                                 *request_p = strdup(buf);
912                                 request_p = NULL;
913                         }
914                         if (mod_name)
915                                 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
916                         else
917                                 glob_expand(buf, &argv, &argc, &maxargs);
918                 } else {
919                         if (!(p = strdup(buf)))
920                                 out_of_memory("read_args");
921                         argv[argc++] = p;
922                         if (*p == '.' && p[1] == '\0')
923                                 dot_pos = argc;
924                 }
925         }
926         argv[argc] = NULL;
927
928         glob_expand(NULL, NULL, NULL, NULL);
929
930         *argc_p = argc;
931         *argv_p = argv;
932 }
933
934 int io_start_buffering_out(int f_out)
935 {
936         if (iobuf_out) {
937                 assert(f_out == iobuf_f_out);
938                 return 0;
939         }
940         if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
941                 out_of_memory("io_start_buffering_out");
942         iobuf_out_cnt = 0;
943         iobuf_f_out = f_out;
944         return 1;
945 }
946
947 int io_start_buffering_in(int f_in)
948 {
949         if (iobuf_in) {
950                 assert(f_in == iobuf_f_in);
951                 return 0;
952         }
953         iobuf_in_siz = 2 * IO_BUFFER_SIZE;
954         if (!(iobuf_in = new_array(char, iobuf_in_siz)))
955                 out_of_memory("io_start_buffering_in");
956         iobuf_f_in = f_in;
957         return 1;
958 }
959
960 void io_end_buffering_in(void)
961 {
962         if (!iobuf_in)
963                 return;
964         free(iobuf_in);
965         iobuf_in = NULL;
966         iobuf_in_ndx = 0;
967         iobuf_in_remaining = 0;
968         iobuf_f_in = -1;
969 }
970
971 void io_end_buffering_out(void)
972 {
973         if (!iobuf_out)
974                 return;
975         io_flush(FULL_FLUSH);
976         free(iobuf_out);
977         iobuf_out = NULL;
978         iobuf_f_out = -1;
979 }
980
981 void maybe_flush_socket(int important)
982 {
983         if (iobuf_out && iobuf_out_cnt
984          && (important || time(NULL) - last_io_out >= 5))
985                 io_flush(NORMAL_FLUSH);
986 }
987
988 void maybe_send_keepalive(void)
989 {
990         if (time(NULL) - last_io_out >= allowed_lull) {
991                 if (!iobuf_out || !iobuf_out_cnt) {
992                         if (protocol_version < 29)
993                                 return; /* there's nothing we can do */
994                         if (protocol_version >= 30)
995                                 send_msg(MSG_NOOP, "", 0, 0);
996                         else {
997                                 write_int(sock_f_out, cur_flist->used);
998                                 write_shortint(sock_f_out, ITEM_IS_NEW);
999                         }
1000                 }
1001                 if (iobuf_out)
1002                         io_flush(NORMAL_FLUSH);
1003         }
1004 }
1005
1006 void start_flist_forward(int f_in)
1007 {
1008         assert(iobuf_out != NULL);
1009         assert(iobuf_f_out == msg_fd_out);
1010         flist_forward_from = f_in;
1011 }
1012
1013 void stop_flist_forward()
1014 {
1015         flist_forward_from = -1;
1016         io_flush(FULL_FLUSH);
1017 }
1018
1019 /**
1020  * Continue trying to read len bytes - don't return until len has been
1021  * read.
1022  **/
1023 static void read_loop(int fd, char *buf, size_t len)
1024 {
1025         while (len) {
1026                 int n = read_timeout(fd, buf, len);
1027
1028                 buf += n;
1029                 len -= n;
1030         }
1031 }
1032
1033 /**
1034  * Read from the file descriptor handling multiplexing - return number
1035  * of bytes read.
1036  *
1037  * Never returns <= 0.
1038  */
1039 static int readfd_unbuffered(int fd, char *buf, size_t len)
1040 {
1041         size_t msg_bytes;
1042         int tag, cnt = 0;
1043         char line[BIGPATHBUFLEN];
1044
1045         if (!iobuf_in || fd != iobuf_f_in)
1046                 return read_timeout(fd, buf, len);
1047
1048         if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1049                 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1050                 iobuf_in_ndx = 0;
1051         }
1052
1053         while (cnt == 0) {
1054                 if (iobuf_in_remaining) {
1055                         len = MIN(len, iobuf_in_remaining);
1056                         memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1057                         iobuf_in_ndx += len;
1058                         iobuf_in_remaining -= len;
1059                         cnt = len;
1060                         break;
1061                 }
1062
1063                 read_loop(fd, line, 4);
1064                 tag = IVAL(line, 0);
1065
1066                 msg_bytes = tag & 0xFFFFFF;
1067                 tag = (tag >> 24) - MPLEX_BASE;
1068
1069                 check_for_io_err = 0;
1070
1071                 switch (tag) {
1072                 case MSG_DATA:
1073                         if (msg_bytes > iobuf_in_siz) {
1074                                 if (!(iobuf_in = realloc_array(iobuf_in, char,
1075                                                                msg_bytes)))
1076                                         out_of_memory("readfd_unbuffered");
1077                                 iobuf_in_siz = msg_bytes;
1078                         }
1079                         read_loop(fd, iobuf_in, msg_bytes);
1080                         iobuf_in_remaining = msg_bytes;
1081                         iobuf_in_ndx = 0;
1082                         break;
1083                 case MSG_NOOP:
1084                         if (am_sender)
1085                                 maybe_send_keepalive();
1086                         break;
1087                 case MSG_IO_ERROR:
1088                         if (msg_bytes != 4)
1089                                 goto invalid_msg;
1090                         read_loop(fd, line, msg_bytes);
1091                         send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1092                         io_error |= IVAL(line, 0);
1093                         break;
1094                 case MSG_DELETED:
1095                         if (msg_bytes >= sizeof line)
1096                                 goto overflow;
1097 #ifdef ICONV_OPTION
1098                         if (ic_recv != (iconv_t)-1) {
1099                                 xbuf outbuf, inbuf;
1100                                 char ibuf[512];
1101                                 int add_null = 0;
1102
1103                                 INIT_CONST_XBUF(outbuf, line);
1104                                 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1105
1106                                 while (msg_bytes) {
1107                                         inbuf.len = msg_bytes > sizeof ibuf
1108                                                   ? sizeof ibuf : msg_bytes;
1109                                         read_loop(fd, inbuf.buf, inbuf.len);
1110                                         if (!(msg_bytes -= inbuf.len)
1111                                          && !ibuf[inbuf.len-1])
1112                                                 inbuf.len--, add_null = 1;
1113                                         if (iconvbufs(ic_send, &inbuf, &outbuf,
1114                                             ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1115                                                 goto overflow;
1116                                 }
1117                                 if (add_null) {
1118                                         if (outbuf.len == outbuf.size)
1119                                                 goto overflow;
1120                                         outbuf.buf[outbuf.len++] = '\0';
1121                                 }
1122                                 msg_bytes = outbuf.len;
1123                         } else
1124 #endif
1125                                 read_loop(fd, line, msg_bytes);
1126                         /* A directory name was sent with the trailing null */
1127                         if (msg_bytes > 0 && !line[msg_bytes-1])
1128                                 log_delete(line, S_IFDIR);
1129                         else {
1130                                 line[msg_bytes] = '\0';
1131                                 log_delete(line, S_IFREG);
1132                         }
1133                         break;
1134                 case MSG_SUCCESS:
1135                         if (msg_bytes != 4) {
1136                           invalid_msg:
1137                                 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1138                                         tag, (long)msg_bytes, who_am_i());
1139                                 exit_cleanup(RERR_STREAMIO);
1140                         }
1141                         read_loop(fd, line, msg_bytes);
1142                         successful_send(IVAL(line, 0));
1143                         break;
1144                 case MSG_NO_SEND:
1145                         if (msg_bytes != 4)
1146                                 goto invalid_msg;
1147                         read_loop(fd, line, msg_bytes);
1148                         send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1149                         break;
1150                 case MSG_INFO:
1151                 case MSG_ERROR:
1152                 case MSG_ERROR_XFER:
1153                 case MSG_WARNING:
1154                         if (msg_bytes >= sizeof line) {
1155                             overflow:
1156                                 rprintf(FERROR,
1157                                         "multiplexing overflow %d:%ld [%s]\n",
1158                                         tag, (long)msg_bytes, who_am_i());
1159                                 exit_cleanup(RERR_STREAMIO);
1160                         }
1161                         read_loop(fd, line, msg_bytes);
1162                         rwrite((enum logcode)tag, line, msg_bytes, 1);
1163                         if (first_message) {
1164                                 if (list_only && !am_sender && tag == 1) {
1165                                         line[msg_bytes] = '\0';
1166                                         check_for_d_option_error(line);
1167                                 }
1168                                 first_message = 0;
1169                         }
1170                         break;
1171                 default:
1172                         rprintf(FERROR, "unexpected tag %d [%s]\n",
1173                                 tag, who_am_i());
1174                         exit_cleanup(RERR_STREAMIO);
1175                 }
1176         }
1177
1178         if (iobuf_in_remaining == 0)
1179                 io_flush(NORMAL_FLUSH);
1180
1181         return cnt;
1182 }
1183
1184 /* Do a buffered read from fd.  Don't return until all N bytes have
1185  * been read.  If all N can't be read then exit with an error. */
1186 static void readfd(int fd, char *buffer, size_t N)
1187 {
1188         int  cnt;
1189         size_t total = 0;
1190
1191         while (total < N) {
1192                 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1193                 total += cnt;
1194         }
1195
1196         if (fd == write_batch_monitor_in) {
1197                 if ((size_t)write(batch_fd, buffer, total) != total)
1198                         exit_cleanup(RERR_FILEIO);
1199         }
1200
1201         if (fd == flist_forward_from)
1202                 writefd(iobuf_f_out, buffer, total);
1203
1204         if (fd == sock_f_in)
1205                 stats.total_read += total;
1206 }
1207
1208 unsigned short read_shortint(int f)
1209 {
1210         char b[2];
1211         readfd(f, b, 2);
1212         return (UVAL(b, 1) << 8) + UVAL(b, 0);
1213 }
1214
1215 int32 read_int(int f)
1216 {
1217         char b[4];
1218         int32 num;
1219
1220         readfd(f, b, 4);
1221         num = IVAL(b, 0);
1222 #if SIZEOF_INT32 > 4
1223         if (num & (int32)0x80000000)
1224                 num |= ~(int32)0xffffffff;
1225 #endif
1226         return num;
1227 }
1228
1229 int32 read_varint(int f)
1230 {
1231         union {
1232             char b[5];
1233             int32 x;
1234         } u;
1235         uchar ch;
1236         int extra;
1237
1238         u.x = 0;
1239         readfd(f, (char*)&ch, 1);
1240         extra = int_byte_extra[ch / 4];
1241         if (extra) {
1242                 uchar bit = ((uchar)1<<(8-extra));
1243                 if (extra >= (int)sizeof u.b) {
1244                         rprintf(FERROR, "Overflow in read_varint()\n");
1245                         exit_cleanup(RERR_STREAMIO);
1246                 }
1247                 readfd(f, u.b, extra);
1248                 u.b[extra] = ch & (bit-1);
1249         } else
1250                 u.b[0] = ch;
1251 #if CAREFUL_ALIGNMENT
1252         u.x = IVAL(u.b,0);
1253 #endif
1254 #if SIZEOF_INT32 > 4
1255         if (u.x & (int32)0x80000000)
1256                 u.x |= ~(int32)0xffffffff;
1257 #endif
1258         return u.x;
1259 }
1260
1261 int64 read_varlong(int f, uchar min_bytes)
1262 {
1263         union {
1264             char b[9];
1265             int64 x;
1266         } u;
1267         char b2[8];
1268         int extra;
1269
1270 #if SIZEOF_INT64 < 8
1271         memset(u.b, 0, 8);
1272 #else
1273         u.x = 0;
1274 #endif
1275         readfd(f, b2, min_bytes);
1276         memcpy(u.b, b2+1, min_bytes-1);
1277         extra = int_byte_extra[CVAL(b2, 0) / 4];
1278         if (extra) {
1279                 uchar bit = ((uchar)1<<(8-extra));
1280                 if (min_bytes + extra > (int)sizeof u.b) {
1281                         rprintf(FERROR, "Overflow in read_varlong()\n");
1282                         exit_cleanup(RERR_STREAMIO);
1283                 }
1284                 readfd(f, u.b + min_bytes - 1, extra);
1285                 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1286 #if SIZEOF_INT64 < 8
1287                 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1288                         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1289                         exit_cleanup(RERR_UNSUPPORTED);
1290                 }
1291 #endif
1292         } else
1293                 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1294 #if SIZEOF_INT64 < 8
1295         u.x = IVAL(u.b,0);
1296 #elif CAREFUL_ALIGNMENT
1297         u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1298 #endif
1299         return u.x;
1300 }
1301
1302 int64 read_longint(int f)
1303 {
1304 #if SIZEOF_INT64 >= 8
1305         char b[9];
1306 #endif
1307         int32 num = read_int(f);
1308
1309         if (num != (int32)0xffffffff)
1310                 return num;
1311
1312 #if SIZEOF_INT64 < 8
1313         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1314         exit_cleanup(RERR_UNSUPPORTED);
1315 #else
1316         readfd(f, b, 8);
1317         return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1318 #endif
1319 }
1320
1321 void read_buf(int f, char *buf, size_t len)
1322 {
1323         readfd(f,buf,len);
1324 }
1325
1326 void read_sbuf(int f, char *buf, size_t len)
1327 {
1328         readfd(f, buf, len);
1329         buf[len] = '\0';
1330 }
1331
1332 uchar read_byte(int f)
1333 {
1334         uchar c;
1335         readfd(f, (char *)&c, 1);
1336         return c;
1337 }
1338
1339 int read_vstring(int f, char *buf, int bufsize)
1340 {
1341         int len = read_byte(f);
1342
1343         if (len & 0x80)
1344                 len = (len & ~0x80) * 0x100 + read_byte(f);
1345
1346         if (len >= bufsize) {
1347                 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1348                         len, bufsize - 1);
1349                 return -1;
1350         }
1351
1352         if (len)
1353                 readfd(f, buf, len);
1354         buf[len] = '\0';
1355         return len;
1356 }
1357
1358 /* Populate a sum_struct with values from the socket.  This is
1359  * called by both the sender and the receiver. */
1360 void read_sum_head(int f, struct sum_struct *sum)
1361 {
1362         int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1363         sum->count = read_int(f);
1364         if (sum->count < 0) {
1365                 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1366                         (long)sum->count, who_am_i());
1367                 exit_cleanup(RERR_PROTOCOL);
1368         }
1369         sum->blength = read_int(f);
1370         if (sum->blength < 0 || sum->blength > max_blength) {
1371                 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1372                         (long)sum->blength, who_am_i());
1373                 exit_cleanup(RERR_PROTOCOL);
1374         }
1375         sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1376         if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1377                 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1378                         sum->s2length, who_am_i());
1379                 exit_cleanup(RERR_PROTOCOL);
1380         }
1381         sum->remainder = read_int(f);
1382         if (sum->remainder < 0 || sum->remainder > sum->blength) {
1383                 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1384                         (long)sum->remainder, who_am_i());
1385                 exit_cleanup(RERR_PROTOCOL);
1386         }
1387 }
1388
1389 /* Send the values from a sum_struct over the socket.  Set sum to
1390  * NULL if there are no checksums to send.  This is called by both
1391  * the generator and the sender. */
1392 void write_sum_head(int f, struct sum_struct *sum)
1393 {
1394         static struct sum_struct null_sum;
1395
1396         if (sum == NULL)
1397                 sum = &null_sum;
1398
1399         write_int(f, sum->count);
1400         write_int(f, sum->blength);
1401         if (protocol_version >= 27)
1402                 write_int(f, sum->s2length);
1403         write_int(f, sum->remainder);
1404 }
1405
1406 /**
1407  * Sleep after writing to limit I/O bandwidth usage.
1408  *
1409  * @todo Rather than sleeping after each write, it might be better to
1410  * use some kind of averaging.  The current algorithm seems to always
1411  * use a bit less bandwidth than specified, because it doesn't make up
1412  * for slow periods.  But arguably this is a feature.  In addition, we
1413  * ought to take the time used to write the data into account.
1414  *
1415  * During some phases of big transfers (file FOO is uptodate) this is
1416  * called with a small bytes_written every time.  As the kernel has to
1417  * round small waits up to guarantee that we actually wait at least the
1418  * requested number of microseconds, this can become grossly inaccurate.
1419  * We therefore keep track of the bytes we've written over time and only
1420  * sleep when the accumulated delay is at least 1 tenth of a second.
1421  **/
1422 static void sleep_for_bwlimit(int bytes_written)
1423 {
1424         static struct timeval prior_tv;
1425         static long total_written = 0;
1426         struct timeval tv, start_tv;
1427         long elapsed_usec, sleep_usec;
1428
1429 #define ONE_SEC 1000000L /* # of microseconds in a second */
1430
1431         if (!bwlimit_writemax)
1432                 return;
1433
1434         total_written += bytes_written;
1435
1436         gettimeofday(&start_tv, NULL);
1437         if (prior_tv.tv_sec) {
1438                 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1439                              + (start_tv.tv_usec - prior_tv.tv_usec);
1440                 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1441                 if (total_written < 0)
1442                         total_written = 0;
1443         }
1444
1445         sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1446         if (sleep_usec < ONE_SEC / 10) {
1447                 prior_tv = start_tv;
1448                 return;
1449         }
1450
1451         tv.tv_sec  = sleep_usec / ONE_SEC;
1452         tv.tv_usec = sleep_usec % ONE_SEC;
1453         select(0, NULL, NULL, NULL, &tv);
1454
1455         gettimeofday(&prior_tv, NULL);
1456         elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1457                      + (prior_tv.tv_usec - start_tv.tv_usec);
1458         total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1459 }
1460
1461 /* Write len bytes to the file descriptor fd, looping as necessary to get
1462  * the job done and also (in certain circumstances) reading any data on
1463  * msg_fd_in to avoid deadlock.
1464  *
1465  * This function underlies the multiplexing system.  The body of the
1466  * application never calls this function directly. */
1467 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1468 {
1469         size_t n, total = 0;
1470         fd_set w_fds, r_fds, e_fds;
1471         int maxfd, count, cnt, using_r_fds;
1472         int defer_inc = 0;
1473         struct timeval tv;
1474
1475         if (no_flush++)
1476                 defer_forwarding_messages++, defer_inc++;
1477
1478         while (total < len) {
1479                 FD_ZERO(&w_fds);
1480                 FD_SET(fd, &w_fds);
1481                 FD_ZERO(&e_fds);
1482                 FD_SET(fd, &e_fds);
1483                 maxfd = fd;
1484
1485                 if (msg_fd_in >= 0) {
1486                         FD_ZERO(&r_fds);
1487                         FD_SET(msg_fd_in, &r_fds);
1488                         if (msg_fd_in > maxfd)
1489                                 maxfd = msg_fd_in;
1490                         using_r_fds = 1;
1491                 } else
1492                         using_r_fds = 0;
1493
1494                 tv.tv_sec = select_timeout;
1495                 tv.tv_usec = 0;
1496
1497                 errno = 0;
1498                 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1499                                &w_fds, &e_fds, &tv);
1500
1501                 if (count <= 0) {
1502                         if (count < 0 && errno == EBADF)
1503                                 exit_cleanup(RERR_SOCKETIO);
1504                         check_timeout();
1505                         continue;
1506                 }
1507
1508                 /*if (FD_ISSET(fd, &e_fds))
1509                         rprintf(FINFO, "select exception on fd %d\n", fd); */
1510
1511                 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1512                         read_msg_fd();
1513
1514                 if (!FD_ISSET(fd, &w_fds))
1515                         continue;
1516
1517                 n = len - total;
1518                 if (bwlimit_writemax && n > bwlimit_writemax)
1519                         n = bwlimit_writemax;
1520                 cnt = write(fd, buf + total, n);
1521
1522                 if (cnt <= 0) {
1523                         if (cnt < 0) {
1524                                 if (errno == EINTR)
1525                                         continue;
1526                                 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1527                                         msleep(1);
1528                                         continue;
1529                                 }
1530                         }
1531
1532                         /* Don't try to write errors back across the stream. */
1533                         if (fd == sock_f_out)
1534                                 io_end_multiplex_out();
1535                         /* Don't try to write errors down a failing msg pipe. */
1536                         if (am_server && fd == msg_fd_out)
1537                                 exit_cleanup(RERR_STREAMIO);
1538                         rsyserr(FERROR, errno,
1539                                 "writefd_unbuffered failed to write %ld bytes [%s]",
1540                                 (long)len, who_am_i());
1541                         /* If the other side is sending us error messages, try
1542                          * to grab any messages they sent before they died. */
1543                         while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1544                                 char buf[1024];
1545                                 set_io_timeout(30);
1546                                 ignore_timeout = 0;
1547                                 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1548                         }
1549                         exit_cleanup(RERR_STREAMIO);
1550                 }
1551
1552                 total += cnt;
1553                 defer_forwarding_messages++, defer_inc++;
1554
1555                 if (fd == sock_f_out) {
1556                         if (io_timeout || am_generator)
1557                                 last_io_out = time(NULL);
1558                         sleep_for_bwlimit(cnt);
1559                 }
1560         }
1561
1562         no_flush--;
1563         defer_inc -= defer_forwarding_keep;
1564         if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1565                 msg_flush();
1566 }
1567
1568 void io_flush(int flush_it_all)
1569 {
1570         if (!iobuf_out_cnt || no_flush)
1571                 return;
1572
1573         if (io_multiplexing_out)
1574                 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1575         else
1576                 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1577         iobuf_out_cnt = 0;
1578
1579         if (flush_it_all && !defer_forwarding_messages)
1580                 msg_flush();
1581 }
1582
1583 static void writefd(int fd, const char *buf, size_t len)
1584 {
1585         if (fd == sock_f_out)
1586                 stats.total_written += len;
1587
1588         if (fd == write_batch_monitor_out) {
1589                 if ((size_t)write(batch_fd, buf, len) != len)
1590                         exit_cleanup(RERR_FILEIO);
1591         }
1592
1593         if (!iobuf_out || fd != iobuf_f_out) {
1594                 writefd_unbuffered(fd, buf, len);
1595                 return;
1596         }
1597
1598         while (len) {
1599                 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1600                 if (n > 0) {
1601                         memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1602                         buf += n;
1603                         len -= n;
1604                         iobuf_out_cnt += n;
1605                 }
1606
1607                 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1608                         io_flush(NORMAL_FLUSH);
1609         }
1610 }
1611
1612 void write_shortint(int f, unsigned short x)
1613 {
1614         char b[2];
1615         b[0] = (char)x;
1616         b[1] = (char)(x >> 8);
1617         writefd(f, b, 2);
1618 }
1619
1620 void write_int(int f, int32 x)
1621 {
1622         char b[4];
1623         SIVAL(b, 0, x);
1624         writefd(f, b, 4);
1625 }
1626
1627 void write_varint(int f, int32 x)
1628 {
1629         char b[5];
1630         uchar bit;
1631         int cnt = 4;
1632
1633         SIVAL(b, 1, x);
1634
1635         while (cnt > 1 && b[cnt] == 0)
1636                 cnt--;
1637         bit = ((uchar)1<<(7-cnt+1));
1638         if (CVAL(b, cnt) >= bit) {
1639                 cnt++;
1640                 *b = ~(bit-1);
1641         } else if (cnt > 1)
1642                 *b = b[cnt] | ~(bit*2-1);
1643         else
1644                 *b = b[cnt];
1645
1646         writefd(f, b, cnt);
1647 }
1648
1649 void write_varlong(int f, int64 x, uchar min_bytes)
1650 {
1651         char b[9];
1652         uchar bit;
1653         int cnt = 8;
1654
1655         SIVAL(b, 1, x);
1656 #if SIZEOF_INT64 >= 8
1657         SIVAL(b, 5, x >> 32);
1658 #else
1659         if (x <= 0x7FFFFFFF && x >= 0)
1660                 memset(b + 5, 0, 4);
1661         else {
1662                 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1663                 exit_cleanup(RERR_UNSUPPORTED);
1664         }
1665 #endif
1666
1667         while (cnt > min_bytes && b[cnt] == 0)
1668                 cnt--;
1669         bit = ((uchar)1<<(7-cnt+min_bytes));
1670         if (CVAL(b, cnt) >= bit) {
1671                 cnt++;
1672                 *b = ~(bit-1);
1673         } else if (cnt > min_bytes)
1674                 *b = b[cnt] | ~(bit*2-1);
1675         else
1676                 *b = b[cnt];
1677
1678         writefd(f, b, cnt);
1679 }
1680
1681 /*
1682  * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1683  * 64-bit types on this platform.
1684  */
1685 void write_longint(int f, int64 x)
1686 {
1687         char b[12], * const s = b+4;
1688
1689         SIVAL(s, 0, x);
1690         if (x <= 0x7FFFFFFF && x >= 0) {
1691                 writefd(f, s, 4);
1692                 return;
1693         }
1694
1695 #if SIZEOF_INT64 < 8
1696         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1697         exit_cleanup(RERR_UNSUPPORTED);
1698 #else
1699         memset(b, 0xFF, 4);
1700         SIVAL(s, 4, x >> 32);
1701         writefd(f, b, 12);
1702 #endif
1703 }
1704
1705 void write_buf(int f, const char *buf, size_t len)
1706 {
1707         writefd(f,buf,len);
1708 }
1709
1710 /** Write a string to the connection */
1711 void write_sbuf(int f, const char *buf)
1712 {
1713         writefd(f, buf, strlen(buf));
1714 }
1715
1716 void write_byte(int f, uchar c)
1717 {
1718         writefd(f, (char *)&c, 1);
1719 }
1720
1721 void write_vstring(int f, const char *str, int len)
1722 {
1723         uchar lenbuf[3], *lb = lenbuf;
1724
1725         if (len > 0x7F) {
1726                 if (len > 0x7FFF) {
1727                         rprintf(FERROR,
1728                                 "attempting to send over-long vstring (%d > %d)\n",
1729                                 len, 0x7FFF);
1730                         exit_cleanup(RERR_PROTOCOL);
1731                 }
1732                 *lb++ = len / 0x100 + 0x80;
1733         }
1734         *lb = len;
1735
1736         writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1737         if (len)
1738                 writefd(f, str, len);
1739 }
1740
1741 /* Send a file-list index using a byte-reduction method. */
1742 void write_ndx(int f, int32 ndx)
1743 {
1744         static int32 prev_positive = -1, prev_negative = 1;
1745         int32 diff, cnt = 0;
1746         char b[6];
1747
1748         if (protocol_version < 30 || read_batch) {
1749                 write_int(f, ndx);
1750                 return;
1751         }
1752
1753         /* Send NDX_DONE as a single-byte 0 with no side effects.  Send
1754          * negative nums as a positive after sending a leading 0xFF. */
1755         if (ndx >= 0) {
1756                 diff = ndx - prev_positive;
1757                 prev_positive = ndx;
1758         } else if (ndx == NDX_DONE) {
1759                 *b = 0;
1760                 writefd(f, b, 1);
1761                 return;
1762         } else {
1763                 b[cnt++] = (char)0xFF;
1764                 ndx = -ndx;
1765                 diff = ndx - prev_negative;
1766                 prev_negative = ndx;
1767         }
1768
1769         /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1770          * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1771          * & all 4 bytes of the (non-negative) num with the high-bit set. */
1772         if (diff < 0xFE && diff > 0)
1773                 b[cnt++] = (char)diff;
1774         else if (diff < 0 || diff > 0x7FFF) {
1775                 b[cnt++] = (char)0xFE;
1776                 b[cnt++] = (char)((ndx >> 24) | 0x80);
1777                 b[cnt++] = (char)ndx;
1778                 b[cnt++] = (char)(ndx >> 8);
1779                 b[cnt++] = (char)(ndx >> 16);
1780         } else {
1781                 b[cnt++] = (char)0xFE;
1782                 b[cnt++] = (char)(diff >> 8);
1783                 b[cnt++] = (char)diff;
1784         }
1785         writefd(f, b, cnt);
1786 }
1787
1788 /* Receive a file-list index using a byte-reduction method. */
1789 int32 read_ndx(int f)
1790 {
1791         static int32 prev_positive = -1, prev_negative = 1;
1792         int32 *prev_ptr, num;
1793         char b[4];
1794
1795         if (protocol_version < 30)
1796                 return read_int(f);
1797
1798         readfd(f, b, 1);
1799         if (CVAL(b, 0) == 0xFF) {
1800                 readfd(f, b, 1);
1801                 prev_ptr = &prev_negative;
1802         } else if (CVAL(b, 0) == 0)
1803                 return NDX_DONE;
1804         else
1805                 prev_ptr = &prev_positive;
1806         if (CVAL(b, 0) == 0xFE) {
1807                 readfd(f, b, 2);
1808                 if (CVAL(b, 0) & 0x80) {
1809                         b[3] = CVAL(b, 0) & ~0x80;
1810                         b[0] = b[1];
1811                         readfd(f, b+1, 2);
1812                         num = IVAL(b, 0);
1813                 } else
1814                         num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1815         } else
1816                 num = UVAL(b, 0) + *prev_ptr;
1817         *prev_ptr = num;
1818         if (prev_ptr == &prev_negative)
1819                 num = -num;
1820         return num;
1821 }
1822
1823 /* Read a line of up to bufsiz-1 characters into buf.  Strips
1824  * the (required) trailing newline and all carriage returns.
1825  * Returns 1 for success; 0 for I/O error or truncation. */
1826 int read_line_old(int f, char *buf, size_t bufsiz)
1827 {
1828         bufsiz--; /* leave room for the null */
1829         while (bufsiz > 0) {
1830                 buf[0] = 0;
1831                 read_buf(f, buf, 1);
1832                 if (buf[0] == 0)
1833                         return 0;
1834                 if (buf[0] == '\n')
1835                         break;
1836                 if (buf[0] != '\r') {
1837                         buf++;
1838                         bufsiz--;
1839                 }
1840         }
1841         *buf = '\0';
1842         return bufsiz > 0;
1843 }
1844
1845 void io_printf(int fd, const char *format, ...)
1846 {
1847         va_list ap;
1848         char buf[BIGPATHBUFLEN];
1849         int len;
1850
1851         va_start(ap, format);
1852         len = vsnprintf(buf, sizeof buf, format, ap);
1853         va_end(ap);
1854
1855         if (len < 0)
1856                 exit_cleanup(RERR_STREAMIO);
1857
1858         if (len > (int)sizeof buf) {
1859                 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1860                 exit_cleanup(RERR_STREAMIO);
1861         }
1862
1863         write_sbuf(fd, buf);
1864 }
1865
1866 /** Setup for multiplexing a MSG_* stream with the data stream. */
1867 void io_start_multiplex_out(void)
1868 {
1869         io_flush(NORMAL_FLUSH);
1870         io_start_buffering_out(sock_f_out);
1871         io_multiplexing_out = 1;
1872 }
1873
1874 /** Setup for multiplexing a MSG_* stream with the data stream. */
1875 void io_start_multiplex_in(void)
1876 {
1877         io_flush(NORMAL_FLUSH);
1878         io_start_buffering_in(sock_f_in);
1879         io_multiplexing_in = 1;
1880 }
1881
1882 /** Write an message to the multiplexed data stream. */
1883 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1884 {
1885         if (!io_multiplexing_out)
1886                 return 0;
1887         io_flush(NORMAL_FLUSH);
1888         stats.total_written += (len+4);
1889         mplex_write(sock_f_out, code, buf, len, convert);
1890         return 1;
1891 }
1892
1893 void io_end_multiplex_in(void)
1894 {
1895         io_multiplexing_in = 0;
1896         io_end_buffering_in();
1897 }
1898
1899 /** Stop output multiplexing. */
1900 void io_end_multiplex_out(void)
1901 {
1902         io_multiplexing_out = 0;
1903         io_end_buffering_out();
1904 }
1905
1906 void start_write_batch(int fd)
1907 {
1908         /* Some communication has already taken place, but we don't
1909          * enable batch writing until here so that we can write a
1910          * canonical record of the communication even though the
1911          * actual communication so far depends on whether a daemon
1912          * is involved. */
1913         write_int(batch_fd, protocol_version);
1914         if (protocol_version >= 30)
1915                 write_byte(batch_fd, inc_recurse);
1916         write_int(batch_fd, checksum_seed);
1917
1918         if (am_sender)
1919                 write_batch_monitor_out = fd;
1920         else
1921                 write_batch_monitor_in = fd;
1922 }
1923
1924 void stop_write_batch(void)
1925 {
1926         write_batch_monitor_out = -1;
1927         write_batch_monitor_in = -1;
1928 }