2 * Socket and pipe I/O utilities used in rsync.
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) 2003-2009 Wayne Davison
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.
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.
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.
23 /* Rsync provides its own multiplexing system, which is used to send
24 * stderr and stdout over a single socket.
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(). */
34 /** If no timeout is specified then use a 60 second select timeout */
35 #define SELECT_TIMEOUT 60
38 extern size_t bwlimit_writemax;
39 extern int io_timeout;
43 extern int am_generator;
44 extern int inc_recurse;
48 extern int file_total;
49 extern int file_old_total;
51 extern int read_batch;
52 extern int protect_args;
53 extern int checksum_seed;
54 extern int protocol_version;
55 extern int remove_source_files;
56 extern int preserve_hard_links;
57 extern struct stats stats;
58 extern struct file_list *cur_flist;
60 extern int filesfrom_convert;
61 extern iconv_t ic_send, ic_recv;
64 int csum_length = SHORT_SUM_LENGTH; /* initial value */
66 int ignore_timeout = 0;
70 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
71 int kluge_around_eof = 0;
78 static int iobuf_f_in = -1;
79 static char *iobuf_in;
80 static size_t iobuf_in_siz;
81 static size_t iobuf_in_ndx;
82 static size_t iobuf_in_remaining;
84 static int iobuf_f_out = -1;
85 static char *iobuf_out;
86 static int iobuf_out_cnt;
88 int flist_forward_from = -1;
90 static int io_multiplexing_out;
91 static int io_multiplexing_in;
92 static time_t last_io_in;
93 static time_t last_io_out;
96 static int write_batch_monitor_in = -1;
97 static int write_batch_monitor_out = -1;
99 static int io_filesfrom_f_in = -1;
100 static int io_filesfrom_f_out = -1;
101 static xbuf ff_buf = EMPTY_XBUF;
102 static char ff_lastchar;
104 static xbuf iconv_buf = EMPTY_XBUF;
106 static int defer_forwarding_messages = 0, keep_defer_forwarding = 0;
107 static int select_timeout = SELECT_TIMEOUT;
108 static int active_filecnt = 0;
109 static OFF_T active_bytecnt = 0;
110 static int first_message = 1;
112 static char int_byte_extra[64] = {
113 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
114 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
115 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
116 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
119 #define REMOTE_OPTION_ERROR "rsync: on remote machine: -"
120 #define REMOTE_OPTION_ERROR2 ": unknown option"
122 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
124 static void readfd(int fd, char *buffer, size_t N);
125 static void writefd(int fd, const char *buf, size_t len);
126 static void writefd_unbuffered(int fd, const char *buf, size_t len);
127 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
129 static flist_ndx_list redo_list, hlink_list;
131 struct msg_list_item {
132 struct msg_list_item *next;
138 struct msg_list_item *head, *tail;
141 static struct msg_list msg_queue;
143 static void got_flist_entry_status(enum festatus status, const char *buf)
145 int ndx = IVAL(buf, 0);
146 struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
148 if (remove_source_files) {
150 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
154 flist->in_progress--;
158 if (remove_source_files)
159 send_msg(MSG_SUCCESS, buf, 4, 0);
160 if (preserve_hard_links) {
161 struct file_struct *file = flist->files[ndx - flist->ndx_start];
162 if (F_IS_HLINKED(file)) {
163 flist_ndx_push(&hlink_list, ndx);
164 flist->in_progress++;
171 flist->in_progress++;
176 flist_ndx_push(&redo_list, ndx);
183 static void check_timeout(void)
187 if (!io_timeout || ignore_timeout)
191 last_io_in = time(NULL);
197 if (t - last_io_in >= io_timeout) {
198 if (!am_server && !am_daemon) {
199 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
200 (int)(t-last_io_in));
202 exit_cleanup(RERR_TIMEOUT);
206 /* Note the fds used for the main socket (which might really be a pipe
207 * for a local transfer, but we can ignore that). */
208 void io_set_sock_fds(int f_in, int f_out)
214 void set_io_timeout(int secs)
218 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
219 select_timeout = SELECT_TIMEOUT;
221 select_timeout = io_timeout;
223 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
226 /* Setup the fd used to receive MSG_* messages. Only needed during the
227 * early stages of being a local sender (up through the sending of the
228 * file list) or when we're the generator (to fetch the messages from
230 void set_msg_fd_in(int fd)
235 /* Setup the fd used to send our MSG_* messages. Only needed when
236 * we're the receiver (to send our messages to the generator). */
237 void set_msg_fd_out(int fd)
240 set_nonblocking(msg_fd_out);
243 /* Add a message to the pending MSG_* list. */
244 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
246 struct msg_list_item *m;
247 int sz = len + 4 + sizeof m[0] - 1;
249 if (!(m = (struct msg_list_item *)new_array(char, sz)))
250 out_of_memory("msg_list_add");
252 m->convert = convert;
253 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
254 memcpy(m->buf + 4, buf, len);
262 static inline int flush_a_msg(int fd)
264 struct msg_list_item *m = msg_queue.head;
265 int len = IVAL(m->buf, 0) & 0xFFFFFF;
266 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
268 if (!(msg_queue.head = m->next))
269 msg_queue.tail = NULL;
271 defer_forwarding_messages++;
272 mplex_write(fd, tag, m->buf + 4, len, m->convert);
273 defer_forwarding_messages--;
280 static void msg_flush(void)
283 while (msg_queue.head && io_multiplexing_out)
284 stats.total_written += flush_a_msg(sock_f_out) + 4;
286 while (msg_queue.head)
287 (void)flush_a_msg(msg_fd_out);
291 static void check_for_d_option_error(const char *msg)
293 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
298 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
301 msg += sizeof REMOTE_OPTION_ERROR - 1;
302 if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
303 || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
306 for ( ; *msg != ':'; msg++) {
309 else if (*msg == 'e')
311 else if (strchr(rsync263_opts, *msg) == NULL)
317 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
321 /* Read a message from the MSG_* fd and handle it. This is called either
322 * during the early stages of being a local sender (up through the sending
323 * of the file list) or when we're the generator (to fetch the messages
324 * from the receiver). */
325 static void read_msg_fd(void)
329 struct file_list *flist;
333 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
334 * to this routine from writefd_unbuffered(). */
337 defer_forwarding_messages++;
342 len = tag & 0xFFFFFF;
343 tag = (tag >> 24) - MPLEX_BASE;
347 if (len < 0 || len > 1 || !am_generator) {
349 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
350 tag, len, who_am_i(),
351 inc_recurse ? "/inc" : "");
352 exit_cleanup(RERR_STREAMIO);
355 readfd(fd, buf, len);
356 stats.total_read = read_varlong(fd, 3);
361 if (len != 4 || !am_generator)
364 got_flist_entry_status(FES_REDO, buf);
367 if (len != 4 || !am_generator || !inc_recurse)
370 /* Read extra file list from receiver. */
371 assert(iobuf_in != NULL);
372 assert(iobuf_f_in == fd);
373 if (DEBUG_GTE(FLIST, 2)) {
374 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
375 who_am_i(), IVAL(buf,0));
377 flist = recv_file_list(fd);
378 flist->parent_ndx = IVAL(buf,0);
379 #ifdef SUPPORT_HARD_LINKS
380 if (preserve_hard_links)
381 match_hard_links(flist);
385 if (len != 0 || !am_generator || !inc_recurse)
392 readfd(fd, buf, len);
393 io_error |= IVAL(buf, 0);
396 if (len >= (int)sizeof buf || !am_generator)
398 readfd(fd, buf, len);
399 send_msg(MSG_DELETED, buf, len, 1);
402 if (len != 4 || !am_generator)
405 got_flist_entry_status(FES_SUCCESS, buf);
408 if (len != 4 || !am_generator)
411 got_flist_entry_status(FES_NO_SEND, buf);
413 case MSG_ERROR_SOCKET:
418 if (tag == MSG_ERROR_SOCKET)
419 io_end_multiplex_out();
431 rwrite((enum logcode)tag, buf, n, !am_generator);
436 rprintf(FERROR, "unknown message %d:%d [%s]\n",
437 tag, len, who_am_i());
438 exit_cleanup(RERR_STREAMIO);
443 if (!--defer_forwarding_messages && !no_flush)
447 /* This is used by the generator to limit how many file transfers can
448 * be active at once when --remove-source-files is specified. Without
449 * this, sender-side deletions were mostly happening at the end. */
450 void increment_active_files(int ndx, int itemizing, enum logcode code)
453 /* TODO: tune these limits? */
454 int limit = active_bytecnt >= 128*1024 ? 10 : 50;
455 if (active_filecnt < limit)
457 check_for_finished_files(itemizing, code, 0);
458 if (active_filecnt < limit)
461 io_flush(NORMAL_FLUSH);
467 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
470 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
471 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
473 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
477 /* We need to convert buf before doing anything else so that we
478 * can include the (converted) byte length in the message header. */
479 if (convert && ic_send != (iconv_t)-1) {
482 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
483 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
485 iconvbufs(ic_send, &inbuf, &outbuf,
486 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
488 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
489 exit_cleanup(RERR_UNSUPPORTED);
492 n = len = outbuf.len;
495 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
496 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
498 memcpy(buffer + 4, buf, n);
500 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
502 keep_defer_forwarding++; /* defer_forwarding_messages++ on return */
503 writefd_unbuffered(fd, buffer, n+4);
504 keep_defer_forwarding--;
507 writefd_unbuffered(fd, buf+n, len-n);
509 if (!--defer_forwarding_messages && !no_flush)
513 int send_msg(enum msgcode code, const char *buf, int len, int convert)
515 if (msg_fd_out < 0) {
516 if (!defer_forwarding_messages)
517 return io_multiplex_write(code, buf, len, convert);
518 if (!io_multiplexing_out)
520 msg_list_add(&msg_queue, code, buf, len, convert);
523 if (flist_forward_from >= 0)
524 msg_list_add(&msg_queue, code, buf, len, convert);
526 mplex_write(msg_fd_out, code, buf, len, convert);
530 void send_msg_int(enum msgcode code, int num)
533 SIVAL(numbuf, 0, num);
534 send_msg(code, numbuf, 4, 0);
537 void wait_for_receiver(void)
539 if (io_flush(FULL_FLUSH))
544 int get_redo_num(void)
546 return flist_ndx_pop(&redo_list);
549 int get_hlink_num(void)
551 return flist_ndx_pop(&hlink_list);
555 * When we're the receiver and we have a local --files-from list of names
556 * that needs to be sent over the socket to the sender, we have to do two
557 * things at the same time: send the sender a list of what files we're
558 * processing and read the incoming file+info list from the sender. We do
559 * this by augmenting the read_timeout() function to copy this data. It
560 * uses ff_buf to read a block of data from f_in (when it is ready, since
561 * it might be a pipe) and then blast it out f_out (when it is ready to
562 * receive more data).
564 void io_set_filesfrom_fds(int f_in, int f_out)
566 io_filesfrom_f_in = f_in;
567 io_filesfrom_f_out = f_out;
568 alloc_xbuf(&ff_buf, 2048);
571 alloc_xbuf(&iconv_buf, 1024);
575 /* It's almost always an error to get an EOF when we're trying to read from the
576 * network, because the protocol is (for the most part) self-terminating.
578 * There is one case for the receiver when it is at the end of the transfer
579 * (hanging around reading any keep-alive packets that might come its way): if
580 * the sender dies before the generator's kill-signal comes through, we can end
581 * up here needing to loop until the kill-signal arrives. In this situation,
582 * kluge_around_eof will be < 0.
584 * There is another case for older protocol versions (< 24) where the module
585 * listing was not terminated, so we must ignore an EOF error in that case and
586 * exit. In this situation, kluge_around_eof will be > 0. */
587 static void whine_about_eof(int fd)
589 if (kluge_around_eof && fd == sock_f_in) {
591 if (kluge_around_eof > 0)
593 /* If we're still here after 10 seconds, exit with an error. */
594 for (i = 10*1000/20; i--; )
598 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
599 "(%s bytes received so far) [%s]\n",
600 big_num(stats.total_read), who_am_i());
602 exit_cleanup(RERR_STREAMIO);
606 * Read from a socket with I/O timeout. return the number of bytes
607 * read. If no bytes can be read then exit, never return a number <= 0.
609 * TODO: If the remote shell connection fails, then current versions
610 * actually report an "unexpected EOF" error here. Since it's a
611 * fairly common mistake to try to use rsh when ssh is required, we
612 * should trap that: if we fail to read any data at all, we should
613 * give a better explanation. We can tell whether the connection has
614 * started by looking e.g. at whether the remote version is known yet.
616 static int read_timeout(int fd, char *buf, size_t len)
620 io_flush(FULL_FLUSH);
623 /* until we manage to read *something* */
632 if (io_filesfrom_f_out >= 0) {
634 if (ff_buf.len == 0) {
635 if (io_filesfrom_f_in >= 0) {
636 FD_SET(io_filesfrom_f_in, &r_fds);
637 new_fd = io_filesfrom_f_in;
639 io_filesfrom_f_out = -1;
643 FD_SET(io_filesfrom_f_out, &w_fds);
644 new_fd = io_filesfrom_f_out;
650 if (am_sender && inc_recurse && !flist_eof && !defer_forwarding_messages && !cnt
651 && file_total - file_old_total < MAX_FILECNT_LOOKAHEAD
652 && file_total - file_old_total >= MIN_FILECNT_LOOKAHEAD)
655 tv.tv_sec = select_timeout;
660 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
663 if (errno == EBADF) {
664 defer_forwarding_messages = 0;
665 exit_cleanup(RERR_SOCKETIO);
667 if (am_sender && tv.tv_sec == 0)
668 send_extra_file_list(sock_f_out, -1);
674 if (io_filesfrom_f_out >= 0) {
676 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
677 int l = write(io_filesfrom_f_out,
678 ff_buf.buf + ff_buf.pos,
681 if (!(ff_buf.len -= l))
685 } else if (errno != EINTR) {
686 /* XXX should we complain? */
687 io_filesfrom_f_out = -1;
690 } else if (io_filesfrom_f_in >= 0) {
691 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
693 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
695 xbuf *ibuf = &ff_buf;
697 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
699 if (l == 0 || errno != EINTR) {
700 /* Send end-of-file marker */
701 memcpy(ff_buf.buf, "\0\0", 2);
702 ff_buf.len = ff_lastchar? 2 : 1;
704 io_filesfrom_f_in = -1;
708 if (filesfrom_convert) {
711 iconvbufs(ic_send, &iconv_buf, &ff_buf,
712 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
717 char *s = ff_buf.buf + l;
718 /* Transform CR and/or LF into '\0' */
719 while (s-- > ff_buf.buf) {
720 if (*s == '\n' || *s == '\r')
725 /* Last buf ended with a '\0', so don't
726 * let this buf start with one. */
727 while (l && ff_buf.buf[ff_buf.pos] == '\0')
733 char *f = ff_buf.buf + ff_buf.pos;
736 /* Eliminate any multi-'\0' runs. */
738 if (!(*t++ = *f++)) {
739 while (f != eob && !*f)
751 if (!FD_ISSET(fd, &r_fds))
754 n = read(fd, buf, len);
758 whine_about_eof(fd); /* Doesn't return. */
759 if (errno == EINTR || errno == EWOULDBLOCK
763 /* Don't write errors on a dead socket. */
764 if (fd == sock_f_in) {
765 io_end_multiplex_out();
766 rsyserr(FERROR_SOCKET, errno, "read error");
768 rsyserr(FERROR, errno, "read error");
769 exit_cleanup(RERR_STREAMIO);
776 if (fd == sock_f_in && io_timeout)
777 last_io_in = time(NULL);
783 /* Read a line into the "buf" buffer. */
784 int read_line(int fd, char *buf, size_t bufsiz, int flags)
790 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
791 realloc_xbuf(&iconv_buf, bufsiz + 1024);
796 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
800 eob = s + bufsiz - 1;
802 cnt = read(fd, &ch, 1);
803 if (cnt < 0 && (errno == EWOULDBLOCK
804 || errno == EINTR || errno == EAGAIN)) {
811 tv.tv_sec = select_timeout;
813 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
815 /*if (FD_ISSET(fd, &e_fds))
816 rprintf(FINFO, "select exception on fd %d\n", fd); */
821 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
822 /* Skip empty lines if dumping comments. */
823 if (flags & RL_DUMP_COMMENTS && s == buf)
832 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
836 if (flags & RL_CONVERT) {
838 INIT_XBUF(outbuf, buf, 0, bufsiz);
840 iconv_buf.len = s - iconv_buf.buf;
841 iconvbufs(ic_recv, &iconv_buf, &outbuf,
842 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
843 outbuf.buf[outbuf.len] = '\0';
851 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
852 char ***argv_p, int *argc_p, char **request_p)
854 int maxargs = MAX_ARGS;
858 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
861 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
864 if (!(argv = new_array(char *, maxargs)))
865 out_of_memory("read_args");
866 if (mod_name && !protect_args)
867 argv[argc++] = "rsyncd";
870 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
873 if (argc == maxargs-1) {
875 if (!(argv = realloc_array(argv, char *, maxargs)))
876 out_of_memory("read_args");
881 *request_p = strdup(buf);
885 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
887 glob_expand(buf, &argv, &argc, &maxargs);
889 if (!(p = strdup(buf)))
890 out_of_memory("read_args");
892 if (*p == '.' && p[1] == '\0')
898 glob_expand(NULL, NULL, NULL, NULL);
904 int io_start_buffering_out(int f_out)
907 assert(f_out == iobuf_f_out);
910 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
911 out_of_memory("io_start_buffering_out");
917 int io_start_buffering_in(int f_in)
920 assert(f_in == iobuf_f_in);
923 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
924 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
925 out_of_memory("io_start_buffering_in");
930 void io_end_buffering_in(void)
937 iobuf_in_remaining = 0;
941 void io_end_buffering_out(void)
945 io_flush(FULL_FLUSH);
951 void maybe_flush_socket(int important)
953 if (iobuf_out && iobuf_out_cnt
954 && (important || time(NULL) - last_io_out >= 5))
955 io_flush(NORMAL_FLUSH);
958 void maybe_send_keepalive(void)
960 if (time(NULL) - last_io_out >= allowed_lull) {
961 if (!iobuf_out || !iobuf_out_cnt) {
962 if (protocol_version < 29)
963 return; /* there's nothing we can do */
964 if (protocol_version >= 30)
965 send_msg(MSG_NOOP, "", 0, 0);
967 write_int(sock_f_out, cur_flist->used);
968 write_shortint(sock_f_out, ITEM_IS_NEW);
972 io_flush(NORMAL_FLUSH);
976 void start_flist_forward(int f_in)
978 assert(iobuf_out != NULL);
979 assert(iobuf_f_out == msg_fd_out);
980 flist_forward_from = f_in;
983 void stop_flist_forward()
985 flist_forward_from = -1;
986 io_flush(FULL_FLUSH);
990 * Continue trying to read len bytes - don't return until len has been
993 static void read_loop(int fd, char *buf, size_t len)
996 int n = read_timeout(fd, buf, len);
1004 * Read from the file descriptor handling multiplexing - return number
1007 * Never returns <= 0.
1009 static int readfd_unbuffered(int fd, char *buf, size_t len)
1013 char line[BIGPATHBUFLEN];
1015 if (!iobuf_in || fd != iobuf_f_in)
1016 return read_timeout(fd, buf, len);
1018 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1019 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1024 if (iobuf_in_remaining) {
1025 len = MIN(len, iobuf_in_remaining);
1026 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1027 iobuf_in_ndx += len;
1028 iobuf_in_remaining -= len;
1033 read_loop(fd, line, 4);
1034 tag = IVAL(line, 0);
1036 msg_bytes = tag & 0xFFFFFF;
1037 tag = (tag >> 24) - MPLEX_BASE;
1041 if (msg_bytes > iobuf_in_siz) {
1042 if (!(iobuf_in = realloc_array(iobuf_in, char,
1044 out_of_memory("readfd_unbuffered");
1045 iobuf_in_siz = msg_bytes;
1047 read_loop(fd, iobuf_in, msg_bytes);
1048 iobuf_in_remaining = msg_bytes;
1053 maybe_send_keepalive();
1058 read_loop(fd, line, msg_bytes);
1059 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1060 io_error |= IVAL(line, 0);
1066 if (am_sender && am_server)
1067 write_del_stats(sock_f_out);
1070 if (msg_bytes >= sizeof line)
1073 if (ic_recv != (iconv_t)-1) {
1078 INIT_CONST_XBUF(outbuf, line);
1079 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1082 inbuf.len = msg_bytes > sizeof ibuf
1083 ? sizeof ibuf : msg_bytes;
1084 read_loop(fd, inbuf.buf, inbuf.len);
1085 if (!(msg_bytes -= inbuf.len)
1086 && !ibuf[inbuf.len-1])
1087 inbuf.len--, add_null = 1;
1088 if (iconvbufs(ic_send, &inbuf, &outbuf,
1089 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1093 if (outbuf.len == outbuf.size)
1095 outbuf.buf[outbuf.len++] = '\0';
1097 msg_bytes = outbuf.len;
1100 read_loop(fd, line, msg_bytes);
1101 /* A directory name was sent with the trailing null */
1102 if (msg_bytes > 0 && !line[msg_bytes-1])
1103 log_delete(line, S_IFDIR);
1105 line[msg_bytes] = '\0';
1106 log_delete(line, S_IFREG);
1110 if (msg_bytes != 4) {
1112 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1113 tag, (long)msg_bytes, who_am_i());
1114 exit_cleanup(RERR_STREAMIO);
1116 read_loop(fd, line, msg_bytes);
1117 successful_send(IVAL(line, 0));
1122 read_loop(fd, line, msg_bytes);
1123 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1127 case MSG_ERROR_XFER:
1129 if (msg_bytes >= sizeof line) {
1132 "multiplexing overflow %d:%ld [%s]\n",
1133 tag, (long)msg_bytes, who_am_i());
1134 exit_cleanup(RERR_STREAMIO);
1136 read_loop(fd, line, msg_bytes);
1137 rwrite((enum logcode)tag, line, msg_bytes, 1);
1138 if (first_message) {
1139 if (list_only && !am_sender && tag == 1) {
1140 line[msg_bytes] = '\0';
1141 check_for_d_option_error(line);
1147 rprintf(FERROR, "unexpected tag %d [%s]\n",
1149 exit_cleanup(RERR_STREAMIO);
1153 if (iobuf_in_remaining == 0)
1154 io_flush(NORMAL_FLUSH);
1159 /* Do a buffered read from fd. Don't return until all N bytes have
1160 * been read. If all N can't be read then exit with an error. */
1161 static void readfd(int fd, char *buffer, size_t N)
1167 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1171 if (fd == write_batch_monitor_in) {
1172 if ((size_t)write(batch_fd, buffer, total) != total)
1173 exit_cleanup(RERR_FILEIO);
1176 if (fd == flist_forward_from)
1177 writefd(iobuf_f_out, buffer, total);
1179 if (fd == sock_f_in)
1180 stats.total_read += total;
1183 unsigned short read_shortint(int f)
1187 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1190 int32 read_int(int f)
1197 #if SIZEOF_INT32 > 4
1198 if (num & (int32)0x80000000)
1199 num |= ~(int32)0xffffffff;
1204 int32 read_varint(int f)
1214 readfd(f, (char*)&ch, 1);
1215 extra = int_byte_extra[ch / 4];
1217 uchar bit = ((uchar)1<<(8-extra));
1218 if (extra >= (int)sizeof u.b) {
1219 rprintf(FERROR, "Overflow in read_varint()\n");
1220 exit_cleanup(RERR_STREAMIO);
1222 readfd(f, u.b, extra);
1223 u.b[extra] = ch & (bit-1);
1226 #if CAREFUL_ALIGNMENT
1229 #if SIZEOF_INT32 > 4
1230 if (u.x & (int32)0x80000000)
1231 u.x |= ~(int32)0xffffffff;
1236 int64 read_varlong(int f, uchar min_bytes)
1245 #if SIZEOF_INT64 < 8
1250 readfd(f, b2, min_bytes);
1251 memcpy(u.b, b2+1, min_bytes-1);
1252 extra = int_byte_extra[CVAL(b2, 0) / 4];
1254 uchar bit = ((uchar)1<<(8-extra));
1255 if (min_bytes + extra > (int)sizeof u.b) {
1256 rprintf(FERROR, "Overflow in read_varlong()\n");
1257 exit_cleanup(RERR_STREAMIO);
1259 readfd(f, u.b + min_bytes - 1, extra);
1260 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1261 #if SIZEOF_INT64 < 8
1262 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1263 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1264 exit_cleanup(RERR_UNSUPPORTED);
1268 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1269 #if SIZEOF_INT64 < 8
1271 #elif CAREFUL_ALIGNMENT
1272 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1277 int64 read_longint(int f)
1279 #if SIZEOF_INT64 >= 8
1282 int32 num = read_int(f);
1284 if (num != (int32)0xffffffff)
1287 #if SIZEOF_INT64 < 8
1288 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1289 exit_cleanup(RERR_UNSUPPORTED);
1292 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1296 void read_buf(int f, char *buf, size_t len)
1301 void read_sbuf(int f, char *buf, size_t len)
1303 readfd(f, buf, len);
1307 uchar read_byte(int f)
1310 readfd(f, (char *)&c, 1);
1314 int read_vstring(int f, char *buf, int bufsize)
1316 int len = read_byte(f);
1319 len = (len & ~0x80) * 0x100 + read_byte(f);
1321 if (len >= bufsize) {
1322 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1328 readfd(f, buf, len);
1333 /* Populate a sum_struct with values from the socket. This is
1334 * called by both the sender and the receiver. */
1335 void read_sum_head(int f, struct sum_struct *sum)
1337 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1338 sum->count = read_int(f);
1339 if (sum->count < 0) {
1340 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1341 (long)sum->count, who_am_i());
1342 exit_cleanup(RERR_PROTOCOL);
1344 sum->blength = read_int(f);
1345 if (sum->blength < 0 || sum->blength > max_blength) {
1346 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1347 (long)sum->blength, who_am_i());
1348 exit_cleanup(RERR_PROTOCOL);
1350 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1351 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1352 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1353 sum->s2length, who_am_i());
1354 exit_cleanup(RERR_PROTOCOL);
1356 sum->remainder = read_int(f);
1357 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1358 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1359 (long)sum->remainder, who_am_i());
1360 exit_cleanup(RERR_PROTOCOL);
1364 /* Send the values from a sum_struct over the socket. Set sum to
1365 * NULL if there are no checksums to send. This is called by both
1366 * the generator and the sender. */
1367 void write_sum_head(int f, struct sum_struct *sum)
1369 static struct sum_struct null_sum;
1374 write_int(f, sum->count);
1375 write_int(f, sum->blength);
1376 if (protocol_version >= 27)
1377 write_int(f, sum->s2length);
1378 write_int(f, sum->remainder);
1382 * Sleep after writing to limit I/O bandwidth usage.
1384 * @todo Rather than sleeping after each write, it might be better to
1385 * use some kind of averaging. The current algorithm seems to always
1386 * use a bit less bandwidth than specified, because it doesn't make up
1387 * for slow periods. But arguably this is a feature. In addition, we
1388 * ought to take the time used to write the data into account.
1390 * During some phases of big transfers (file FOO is uptodate) this is
1391 * called with a small bytes_written every time. As the kernel has to
1392 * round small waits up to guarantee that we actually wait at least the
1393 * requested number of microseconds, this can become grossly inaccurate.
1394 * We therefore keep track of the bytes we've written over time and only
1395 * sleep when the accumulated delay is at least 1 tenth of a second.
1397 static void sleep_for_bwlimit(int bytes_written)
1399 static struct timeval prior_tv;
1400 static long total_written = 0;
1401 struct timeval tv, start_tv;
1402 long elapsed_usec, sleep_usec;
1404 #define ONE_SEC 1000000L /* # of microseconds in a second */
1406 if (!bwlimit_writemax)
1409 total_written += bytes_written;
1411 gettimeofday(&start_tv, NULL);
1412 if (prior_tv.tv_sec) {
1413 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1414 + (start_tv.tv_usec - prior_tv.tv_usec);
1415 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1416 if (total_written < 0)
1420 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1421 if (sleep_usec < ONE_SEC / 10) {
1422 prior_tv = start_tv;
1426 tv.tv_sec = sleep_usec / ONE_SEC;
1427 tv.tv_usec = sleep_usec % ONE_SEC;
1428 select(0, NULL, NULL, NULL, &tv);
1430 gettimeofday(&prior_tv, NULL);
1431 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1432 + (prior_tv.tv_usec - start_tv.tv_usec);
1433 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1436 static const char *what_fd_is(int fd)
1438 static char buf[20];
1440 if (fd == sock_f_out)
1442 else if (fd == msg_fd_out)
1443 return "message fd";
1444 else if (fd == batch_fd)
1445 return "batch file";
1447 snprintf(buf, sizeof buf, "fd %d", fd);
1452 /* Write len bytes to the file descriptor fd, looping as necessary to get
1453 * the job done and also (in certain circumstances) reading any data on
1454 * msg_fd_in to avoid deadlock.
1456 * This function underlies the multiplexing system. The body of the
1457 * application never calls this function directly. */
1458 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1460 size_t n, total = 0;
1461 fd_set w_fds, r_fds, e_fds;
1462 int maxfd, count, cnt, using_r_fds;
1467 defer_forwarding_messages++, defer_inc++;
1469 while (total < len) {
1476 if (msg_fd_in >= 0) {
1478 FD_SET(msg_fd_in, &r_fds);
1479 if (msg_fd_in > maxfd)
1485 tv.tv_sec = select_timeout;
1489 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1490 &w_fds, &e_fds, &tv);
1493 if (count < 0 && errno == EBADF)
1494 exit_cleanup(RERR_SOCKETIO);
1499 /*if (FD_ISSET(fd, &e_fds))
1500 rprintf(FINFO, "select exception on fd %d\n", fd); */
1502 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1505 if (!FD_ISSET(fd, &w_fds))
1509 if (bwlimit_writemax && n > bwlimit_writemax)
1510 n = bwlimit_writemax;
1511 cnt = write(fd, buf + total, n);
1517 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1523 /* Don't try to write errors back across the stream. */
1524 if (fd == sock_f_out)
1525 io_end_multiplex_out();
1526 /* Don't try to write errors down a failing msg pipe. */
1527 if (am_server && fd == msg_fd_out)
1528 exit_cleanup(RERR_STREAMIO);
1529 rsyserr(FERROR, errno,
1530 "writefd_unbuffered failed to write %ld bytes to %s [%s]",
1531 (long)len, what_fd_is(fd), who_am_i());
1532 /* If the other side is sending us error messages, try
1533 * to grab any messages they sent before they died. */
1534 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1538 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1540 exit_cleanup(RERR_STREAMIO);
1544 defer_forwarding_messages++, defer_inc++;
1546 if (fd == sock_f_out) {
1547 if (io_timeout || am_generator)
1548 last_io_out = time(NULL);
1549 sleep_for_bwlimit(cnt);
1554 if (keep_defer_forwarding)
1556 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1560 int io_flush(int flush_it_all)
1562 int flushed_something = 0;
1567 if (iobuf_out_cnt) {
1568 if (io_multiplexing_out)
1569 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1571 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1573 flushed_something = 1;
1576 if (flush_it_all && !defer_forwarding_messages && msg_queue.head) {
1578 flushed_something = 1;
1581 return flushed_something;
1584 static void writefd(int fd, const char *buf, size_t len)
1586 if (fd == sock_f_out)
1587 stats.total_written += len;
1589 if (fd == write_batch_monitor_out)
1590 writefd_unbuffered(batch_fd, buf, len);
1592 if (!iobuf_out || fd != iobuf_f_out) {
1593 writefd_unbuffered(fd, buf, len);
1598 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1600 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1606 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1607 io_flush(NORMAL_FLUSH);
1611 void write_shortint(int f, unsigned short x)
1615 b[1] = (char)(x >> 8);
1619 void write_int(int f, int32 x)
1626 void write_varint(int f, int32 x)
1634 while (cnt > 1 && b[cnt] == 0)
1636 bit = ((uchar)1<<(7-cnt+1));
1637 if (CVAL(b, cnt) >= bit) {
1641 *b = b[cnt] | ~(bit*2-1);
1648 void write_varlong(int f, int64 x, uchar min_bytes)
1655 #if SIZEOF_INT64 >= 8
1656 SIVAL(b, 5, x >> 32);
1658 if (x <= 0x7FFFFFFF && x >= 0)
1659 memset(b + 5, 0, 4);
1661 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1662 exit_cleanup(RERR_UNSUPPORTED);
1666 while (cnt > min_bytes && b[cnt] == 0)
1668 bit = ((uchar)1<<(7-cnt+min_bytes));
1669 if (CVAL(b, cnt) >= bit) {
1672 } else if (cnt > min_bytes)
1673 *b = b[cnt] | ~(bit*2-1);
1681 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1682 * 64-bit types on this platform.
1684 void write_longint(int f, int64 x)
1686 char b[12], * const s = b+4;
1689 if (x <= 0x7FFFFFFF && x >= 0) {
1694 #if SIZEOF_INT64 < 8
1695 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1696 exit_cleanup(RERR_UNSUPPORTED);
1699 SIVAL(s, 4, x >> 32);
1704 void write_buf(int f, const char *buf, size_t len)
1709 /** Write a string to the connection */
1710 void write_sbuf(int f, const char *buf)
1712 writefd(f, buf, strlen(buf));
1715 void write_byte(int f, uchar c)
1717 writefd(f, (char *)&c, 1);
1720 void write_vstring(int f, const char *str, int len)
1722 uchar lenbuf[3], *lb = lenbuf;
1727 "attempting to send over-long vstring (%d > %d)\n",
1729 exit_cleanup(RERR_PROTOCOL);
1731 *lb++ = len / 0x100 + 0x80;
1735 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1737 writefd(f, str, len);
1740 /* Send a file-list index using a byte-reduction method. */
1741 void write_ndx(int f, int32 ndx)
1743 static int32 prev_positive = -1, prev_negative = 1;
1744 int32 diff, cnt = 0;
1747 if (protocol_version < 30 || read_batch) {
1752 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1753 * negative nums as a positive after sending a leading 0xFF. */
1755 diff = ndx - prev_positive;
1756 prev_positive = ndx;
1757 } else if (ndx == NDX_DONE) {
1762 b[cnt++] = (char)0xFF;
1764 diff = ndx - prev_negative;
1765 prev_negative = ndx;
1768 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1769 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1770 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1771 if (diff < 0xFE && diff > 0)
1772 b[cnt++] = (char)diff;
1773 else if (diff < 0 || diff > 0x7FFF) {
1774 b[cnt++] = (char)0xFE;
1775 b[cnt++] = (char)((ndx >> 24) | 0x80);
1776 b[cnt++] = (char)ndx;
1777 b[cnt++] = (char)(ndx >> 8);
1778 b[cnt++] = (char)(ndx >> 16);
1780 b[cnt++] = (char)0xFE;
1781 b[cnt++] = (char)(diff >> 8);
1782 b[cnt++] = (char)diff;
1787 /* Receive a file-list index using a byte-reduction method. */
1788 int32 read_ndx(int f)
1790 static int32 prev_positive = -1, prev_negative = 1;
1791 int32 *prev_ptr, num;
1794 if (protocol_version < 30)
1798 if (CVAL(b, 0) == 0xFF) {
1800 prev_ptr = &prev_negative;
1801 } else if (CVAL(b, 0) == 0)
1804 prev_ptr = &prev_positive;
1805 if (CVAL(b, 0) == 0xFE) {
1807 if (CVAL(b, 0) & 0x80) {
1808 b[3] = CVAL(b, 0) & ~0x80;
1813 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1815 num = UVAL(b, 0) + *prev_ptr;
1817 if (prev_ptr == &prev_negative)
1822 /* Read a line of up to bufsiz-1 characters into buf. Strips
1823 * the (required) trailing newline and all carriage returns.
1824 * Returns 1 for success; 0 for I/O error or truncation. */
1825 int read_line_old(int f, char *buf, size_t bufsiz)
1827 bufsiz--; /* leave room for the null */
1828 while (bufsiz > 0) {
1830 read_buf(f, buf, 1);
1835 if (buf[0] != '\r') {
1844 void io_printf(int fd, const char *format, ...)
1847 char buf[BIGPATHBUFLEN];
1850 va_start(ap, format);
1851 len = vsnprintf(buf, sizeof buf, format, ap);
1855 exit_cleanup(RERR_STREAMIO);
1857 if (len > (int)sizeof buf) {
1858 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1859 exit_cleanup(RERR_STREAMIO);
1862 write_sbuf(fd, buf);
1865 /** Setup for multiplexing a MSG_* stream with the data stream. */
1866 void io_start_multiplex_out(void)
1868 io_flush(NORMAL_FLUSH);
1869 io_start_buffering_out(sock_f_out);
1870 io_multiplexing_out = 1;
1873 /** Setup for multiplexing a MSG_* stream with the data stream. */
1874 void io_start_multiplex_in(void)
1876 io_flush(NORMAL_FLUSH);
1877 io_start_buffering_in(sock_f_in);
1878 io_multiplexing_in = 1;
1881 /** Write an message to the multiplexed data stream. */
1882 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1884 if (!io_multiplexing_out)
1886 io_flush(NORMAL_FLUSH);
1887 stats.total_written += (len+4);
1888 mplex_write(sock_f_out, code, buf, len, convert);
1892 void io_end_multiplex_in(void)
1894 io_multiplexing_in = 0;
1895 io_end_buffering_in();
1898 /** Stop output multiplexing. */
1899 void io_end_multiplex_out(void)
1901 io_multiplexing_out = 0;
1902 io_end_buffering_out();
1905 void start_write_batch(int fd)
1907 /* Some communication has already taken place, but we don't
1908 * enable batch writing until here so that we can write a
1909 * canonical record of the communication even though the
1910 * actual communication so far depends on whether a daemon
1912 write_int(batch_fd, protocol_version);
1913 if (protocol_version >= 30)
1914 write_byte(batch_fd, inc_recurse);
1915 write_int(batch_fd, checksum_seed);
1918 write_batch_monitor_out = fd;
1920 write_batch_monitor_in = fd;
1923 void stop_write_batch(void)
1925 write_batch_monitor_out = -1;
1926 write_batch_monitor_in = -1;