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) 2003-2008 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(). */
33 /** If no timeout is specified then use a 60 second select timeout */
34 #define SELECT_TIMEOUT 60
37 extern size_t bwlimit_writemax;
38 extern int io_timeout;
42 extern int am_generator;
43 extern int inc_recurse;
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;
57 extern int filesfrom_convert;
58 extern iconv_t ic_send, ic_recv;
61 int csum_length = SHORT_SUM_LENGTH; /* initial value */
63 int ignore_timeout = 0;
66 int check_for_io_err = 0;
68 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
69 int kluge_around_eof = 0;
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;
82 static int iobuf_f_out = -1;
83 static char *iobuf_out;
84 static int iobuf_out_cnt;
86 int flist_forward_from = -1;
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;
94 static int write_batch_monitor_in = -1;
95 static int write_batch_monitor_out = -1;
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;
102 static xbuf iconv_buf = EMPTY_XBUF;
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;
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 */
117 #define REMOTE_OPTION_ERROR "rsync: on remote machine: -"
118 #define REMOTE_OPTION_ERROR2 ": unknown option"
120 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
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);
127 struct flist_ndx_item {
128 struct flist_ndx_item *next;
132 struct flist_ndx_list {
133 struct flist_ndx_item *head, *tail;
136 static struct flist_ndx_list redo_list, hlink_list;
138 struct msg_list_item {
139 struct msg_list_item *next;
145 struct msg_list_item *head, *tail;
148 static struct msg_list msg_queue;
150 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
152 struct flist_ndx_item *item;
154 if (!(item = new(struct flist_ndx_item)))
155 out_of_memory("flist_ndx_push");
159 lp->tail->next = item;
165 static int flist_ndx_pop(struct flist_ndx_list *lp)
167 struct flist_ndx_item *next;
174 next = lp->head->next;
183 static void got_flist_entry_status(enum festatus status, const char *buf)
185 int ndx = IVAL(buf, 0);
186 struct file_list *flist = flist_for_ndx(ndx);
188 assert(flist != NULL);
190 if (remove_source_files) {
192 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
196 flist->in_progress--;
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++;
213 flist_ndx_push(&redo_list, ndx);
220 static void check_timeout(void)
224 if (!io_timeout || ignore_timeout)
228 last_io_in = time(NULL);
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));
239 exit_cleanup(RERR_TIMEOUT);
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)
251 void set_io_timeout(int secs)
255 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
256 select_timeout = SELECT_TIMEOUT;
258 select_timeout = io_timeout;
260 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
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
267 void set_msg_fd_in(int fd)
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)
277 set_nonblocking(msg_fd_out);
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)
283 struct msg_list_item *m;
284 int sz = len + 4 + sizeof m[0] - 1;
286 if (!(m = (struct msg_list_item *)new_array(char, sz)))
287 out_of_memory("msg_list_add");
289 m->convert = convert;
290 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
291 memcpy(m->buf + 4, buf, len);
299 static void msg_flush(void)
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--;
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--;
329 static void check_for_d_option_error(const char *msg)
331 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
336 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
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)
344 for ( ; *msg != ':'; msg++) {
347 else if (*msg == 'e')
349 else if (strchr(rsync263_opts, *msg) == NULL)
355 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
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)
367 struct file_list *flist;
371 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
372 * to this routine from writefd_unbuffered(). */
375 defer_forwarding_messages++;
380 len = tag & 0xFFFFFF;
381 tag = (tag >> 24) - MPLEX_BASE;
383 check_for_io_err = 0;
387 if (len < 0 || len > 1 || !am_generator) {
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);
395 readfd(fd, buf, len);
396 stats.total_read = read_varlong(fd, 3);
401 if (len != 4 || !am_generator)
404 got_flist_entry_status(FES_REDO, buf);
407 if (len != 4 || !am_generator || !inc_recurse)
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));
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);
428 if (len != 0 || !am_generator || !inc_recurse)
435 readfd(fd, buf, len);
436 io_error |= IVAL(buf, 0);
439 if (len >= (int)sizeof buf || !am_generator)
441 readfd(fd, buf, len);
442 send_msg(MSG_DELETED, buf, len, 1);
445 if (len != 4 || !am_generator)
448 got_flist_entry_status(FES_SUCCESS, buf);
451 if (len != 4 || !am_generator)
454 got_flist_entry_status(FES_NO_SEND, buf);
456 case MSG_ERROR_SOCKET:
461 if (tag == MSG_ERROR_SOCKET)
462 io_end_multiplex_out();
474 rwrite((enum logcode)tag, buf, n, !am_generator);
479 rprintf(FERROR, "unknown message %d:%d [%s]\n",
480 tag, len, who_am_i());
481 exit_cleanup(RERR_STREAMIO);
486 if (!--defer_forwarding_messages && !no_flush)
490 /* This is used by the generator to limit how many file transfers can
491 * be active at once when --remove-source-files is specified. Without
492 * this, sender-side deletions were mostly happening at the end. */
493 void increment_active_files(int ndx, int itemizing, enum logcode code)
495 /* TODO: tune these limits? */
496 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
497 check_for_finished_files(itemizing, code, 0);
499 io_flush(NORMAL_FLUSH);
505 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
508 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
509 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
511 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
515 /* We need to convert buf before doing anything else so that we
516 * can include the (converted) byte length in the message header. */
517 if (convert && ic_send != (iconv_t)-1) {
520 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
521 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
523 iconvbufs(ic_send, &inbuf, &outbuf,
524 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
526 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
527 exit_cleanup(RERR_UNSUPPORTED);
530 n = len = outbuf.len;
533 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
534 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
536 memcpy(buffer + 4, buf, n);
538 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
540 defer_forwarding_keep = 1; /* defer_forwarding_messages++ on return */
541 writefd_unbuffered(fd, buffer, n+4);
542 defer_forwarding_keep = 0;
545 writefd_unbuffered(fd, buf+n, len-n);
547 if (!--defer_forwarding_messages && !no_flush)
551 int send_msg(enum msgcode code, const char *buf, int len, int convert)
553 if (msg_fd_out < 0) {
554 if (!defer_forwarding_messages)
555 return io_multiplex_write(code, buf, len, convert);
556 if (!io_multiplexing_out)
558 msg_list_add(&msg_queue, code, buf, len, convert);
561 if (flist_forward_from >= 0)
562 msg_list_add(&msg_queue, code, buf, len, convert);
564 mplex_write(msg_fd_out, code, buf, len, convert);
568 void send_msg_int(enum msgcode code, int num)
571 SIVAL(numbuf, 0, num);
572 send_msg(code, numbuf, 4, 0);
575 void wait_for_receiver(void)
578 io_flush(NORMAL_FLUSH);
583 int get_redo_num(void)
585 return flist_ndx_pop(&redo_list);
588 int get_hlink_num(void)
590 return flist_ndx_pop(&hlink_list);
594 * When we're the receiver and we have a local --files-from list of names
595 * that needs to be sent over the socket to the sender, we have to do two
596 * things at the same time: send the sender a list of what files we're
597 * processing and read the incoming file+info list from the sender. We do
598 * this by augmenting the read_timeout() function to copy this data. It
599 * uses ff_buf to read a block of data from f_in (when it is ready, since
600 * it might be a pipe) and then blast it out f_out (when it is ready to
601 * receive more data).
603 void io_set_filesfrom_fds(int f_in, int f_out)
605 io_filesfrom_f_in = f_in;
606 io_filesfrom_f_out = f_out;
607 alloc_xbuf(&ff_buf, 2048);
610 alloc_xbuf(&iconv_buf, 1024);
614 /* It's almost always an error to get an EOF when we're trying to read from the
615 * network, because the protocol is (for the most part) self-terminating.
617 * There is one case for the receiver when it is at the end of the transfer
618 * (hanging around reading any keep-alive packets that might come its way): if
619 * the sender dies before the generator's kill-signal comes through, we can end
620 * up here needing to loop until the kill-signal arrives. In this situation,
621 * kluge_around_eof will be < 0.
623 * There is another case for older protocol versions (< 24) where the module
624 * listing was not terminated, so we must ignore an EOF error in that case and
625 * exit. In this situation, kluge_around_eof will be > 0. */
626 static void whine_about_eof(int fd)
628 if (kluge_around_eof && fd == sock_f_in) {
630 if (kluge_around_eof > 0)
632 /* If we're still here after 10 seconds, exit with an error. */
633 for (i = 10*1000/20; i--; )
637 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
638 "(%s bytes received so far) [%s]\n",
639 big_num(stats.total_read, 0), who_am_i());
641 exit_cleanup(RERR_STREAMIO);
645 * Read from a socket with I/O timeout. return the number of bytes
646 * read. If no bytes can be read then exit, never return a number <= 0.
648 * TODO: If the remote shell connection fails, then current versions
649 * actually report an "unexpected EOF" error here. Since it's a
650 * fairly common mistake to try to use rsh when ssh is required, we
651 * should trap that: if we fail to read any data at all, we should
652 * give a better explanation. We can tell whether the connection has
653 * started by looking e.g. at whether the remote version is known yet.
655 static int read_timeout(int fd, char *buf, size_t len)
659 io_flush(FULL_FLUSH);
662 /* until we manage to read *something* */
671 if (io_filesfrom_f_out >= 0) {
673 if (ff_buf.len == 0) {
674 if (io_filesfrom_f_in >= 0) {
675 FD_SET(io_filesfrom_f_in, &r_fds);
676 new_fd = io_filesfrom_f_in;
678 io_filesfrom_f_out = -1;
682 FD_SET(io_filesfrom_f_out, &w_fds);
683 new_fd = io_filesfrom_f_out;
689 tv.tv_sec = select_timeout;
694 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
697 if (errno == EBADF) {
698 defer_forwarding_messages = 0;
699 exit_cleanup(RERR_SOCKETIO);
705 if (io_filesfrom_f_out >= 0) {
707 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
708 int l = write(io_filesfrom_f_out,
709 ff_buf.buf + ff_buf.pos,
712 if (!(ff_buf.len -= l))
716 } else if (errno != EINTR) {
717 /* XXX should we complain? */
718 io_filesfrom_f_out = -1;
721 } else if (io_filesfrom_f_in >= 0) {
722 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
724 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
726 xbuf *ibuf = &ff_buf;
728 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
730 if (l == 0 || errno != EINTR) {
731 /* Send end-of-file marker */
732 memcpy(ff_buf.buf, "\0\0", 2);
733 ff_buf.len = ff_lastchar? 2 : 1;
735 io_filesfrom_f_in = -1;
739 if (filesfrom_convert) {
742 iconvbufs(ic_send, &iconv_buf, &ff_buf,
743 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
748 char *s = ff_buf.buf + l;
749 /* Transform CR and/or LF into '\0' */
750 while (s-- > ff_buf.buf) {
751 if (*s == '\n' || *s == '\r')
756 /* Last buf ended with a '\0', so don't
757 * let this buf start with one. */
758 while (l && ff_buf.buf[ff_buf.pos] == '\0')
764 char *f = ff_buf.buf + ff_buf.pos;
767 /* Eliminate any multi-'\0' runs. */
769 if (!(*t++ = *f++)) {
770 while (f != eob && !*f)
782 if (!FD_ISSET(fd, &r_fds))
785 n = read(fd, buf, len);
789 whine_about_eof(fd); /* Doesn't return. */
790 if (errno == EINTR || errno == EWOULDBLOCK
794 /* Don't write errors on a dead socket. */
795 if (fd == sock_f_in) {
796 io_end_multiplex_out();
797 rsyserr(FERROR_SOCKET, errno, "read error");
799 rsyserr(FERROR, errno, "read error");
800 exit_cleanup(RERR_STREAMIO);
807 if (fd == sock_f_in && io_timeout)
808 last_io_in = time(NULL);
814 /* Read a line into the "buf" buffer. */
815 int read_line(int fd, char *buf, size_t bufsiz, int flags)
821 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
822 realloc_xbuf(&iconv_buf, bufsiz + 1024);
827 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
831 eob = s + bufsiz - 1;
833 cnt = read(fd, &ch, 1);
834 if (cnt < 0 && (errno == EWOULDBLOCK
835 || errno == EINTR || errno == EAGAIN)) {
842 tv.tv_sec = select_timeout;
844 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
846 /*if (FD_ISSET(fd, &e_fds))
847 rprintf(FINFO, "select exception on fd %d\n", fd); */
852 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
853 /* Skip empty lines if dumping comments. */
854 if (flags & RL_DUMP_COMMENTS && s == buf)
863 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
867 if (flags & RL_CONVERT) {
869 INIT_XBUF(outbuf, buf, 0, bufsiz);
871 iconv_buf.len = s - iconv_buf.buf;
872 iconvbufs(ic_recv, &iconv_buf, &outbuf,
873 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
874 outbuf.buf[outbuf.len] = '\0';
882 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
883 char ***argv_p, int *argc_p, char **request_p)
885 int maxargs = MAX_ARGS;
889 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
892 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
895 if (!(argv = new_array(char *, maxargs)))
896 out_of_memory("read_args");
897 if (mod_name && !protect_args)
898 argv[argc++] = "rsyncd";
901 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
904 if (argc == maxargs-1) {
906 if (!(argv = realloc_array(argv, char *, maxargs)))
907 out_of_memory("read_args");
912 *request_p = strdup(buf);
916 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
918 glob_expand(buf, &argv, &argc, &maxargs);
920 if (!(p = strdup(buf)))
921 out_of_memory("read_args");
923 if (*p == '.' && p[1] == '\0')
929 glob_expand(NULL, NULL, NULL, NULL);
935 int io_start_buffering_out(int f_out)
938 assert(f_out == iobuf_f_out);
941 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
942 out_of_memory("io_start_buffering_out");
948 int io_start_buffering_in(int f_in)
951 assert(f_in == iobuf_f_in);
954 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
955 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
956 out_of_memory("io_start_buffering_in");
961 void io_end_buffering_in(void)
968 iobuf_in_remaining = 0;
972 void io_end_buffering_out(void)
976 io_flush(FULL_FLUSH);
982 void maybe_flush_socket(int important)
984 if (iobuf_out && iobuf_out_cnt
985 && (important || time(NULL) - last_io_out >= 5))
986 io_flush(NORMAL_FLUSH);
989 void maybe_send_keepalive(void)
991 if (time(NULL) - last_io_out >= allowed_lull) {
992 if (!iobuf_out || !iobuf_out_cnt) {
993 if (protocol_version < 29)
994 return; /* there's nothing we can do */
995 if (protocol_version >= 30)
996 send_msg(MSG_NOOP, "", 0, 0);
998 write_int(sock_f_out, cur_flist->used);
999 write_shortint(sock_f_out, ITEM_IS_NEW);
1003 io_flush(NORMAL_FLUSH);
1007 void start_flist_forward(int f_in)
1009 assert(iobuf_out != NULL);
1010 assert(iobuf_f_out == msg_fd_out);
1011 flist_forward_from = f_in;
1014 void stop_flist_forward()
1016 flist_forward_from = -1;
1017 io_flush(FULL_FLUSH);
1021 * Continue trying to read len bytes - don't return until len has been
1024 static void read_loop(int fd, char *buf, size_t len)
1027 int n = read_timeout(fd, buf, len);
1035 * Read from the file descriptor handling multiplexing - return number
1038 * Never returns <= 0.
1040 static int readfd_unbuffered(int fd, char *buf, size_t len)
1044 char line[BIGPATHBUFLEN];
1046 if (!iobuf_in || fd != iobuf_f_in)
1047 return read_timeout(fd, buf, len);
1049 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1050 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1055 if (iobuf_in_remaining) {
1056 len = MIN(len, iobuf_in_remaining);
1057 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1058 iobuf_in_ndx += len;
1059 iobuf_in_remaining -= len;
1064 read_loop(fd, line, 4);
1065 tag = IVAL(line, 0);
1067 msg_bytes = tag & 0xFFFFFF;
1068 tag = (tag >> 24) - MPLEX_BASE;
1070 check_for_io_err = 0;
1074 if (msg_bytes > iobuf_in_siz) {
1075 if (!(iobuf_in = realloc_array(iobuf_in, char,
1077 out_of_memory("readfd_unbuffered");
1078 iobuf_in_siz = msg_bytes;
1080 read_loop(fd, iobuf_in, msg_bytes);
1081 iobuf_in_remaining = msg_bytes;
1086 maybe_send_keepalive();
1091 read_loop(fd, line, msg_bytes);
1092 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1093 io_error |= IVAL(line, 0);
1096 if (msg_bytes >= sizeof line)
1099 if (ic_recv != (iconv_t)-1) {
1104 INIT_CONST_XBUF(outbuf, line);
1105 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1108 inbuf.len = msg_bytes > sizeof ibuf
1109 ? sizeof ibuf : msg_bytes;
1110 read_loop(fd, inbuf.buf, inbuf.len);
1111 if (!(msg_bytes -= inbuf.len)
1112 && !ibuf[inbuf.len-1])
1113 inbuf.len--, add_null = 1;
1114 if (iconvbufs(ic_send, &inbuf, &outbuf,
1115 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1119 if (outbuf.len == outbuf.size)
1121 outbuf.buf[outbuf.len++] = '\0';
1123 msg_bytes = outbuf.len;
1126 read_loop(fd, line, msg_bytes);
1127 /* A directory name was sent with the trailing null */
1128 if (msg_bytes > 0 && !line[msg_bytes-1])
1129 log_delete(line, S_IFDIR);
1131 line[msg_bytes] = '\0';
1132 log_delete(line, S_IFREG);
1136 if (msg_bytes != 4) {
1138 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1139 tag, (long)msg_bytes, who_am_i());
1140 exit_cleanup(RERR_STREAMIO);
1142 read_loop(fd, line, msg_bytes);
1143 successful_send(IVAL(line, 0));
1148 read_loop(fd, line, msg_bytes);
1149 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1153 case MSG_ERROR_XFER:
1155 if (msg_bytes >= sizeof line) {
1158 "multiplexing overflow %d:%ld [%s]\n",
1159 tag, (long)msg_bytes, who_am_i());
1160 exit_cleanup(RERR_STREAMIO);
1162 read_loop(fd, line, msg_bytes);
1163 rwrite((enum logcode)tag, line, msg_bytes, 1);
1164 if (first_message) {
1165 if (list_only && !am_sender && tag == 1) {
1166 line[msg_bytes] = '\0';
1167 check_for_d_option_error(line);
1173 rprintf(FERROR, "unexpected tag %d [%s]\n",
1175 exit_cleanup(RERR_STREAMIO);
1179 if (iobuf_in_remaining == 0)
1180 io_flush(NORMAL_FLUSH);
1185 /* Do a buffered read from fd. Don't return until all N bytes have
1186 * been read. If all N can't be read then exit with an error. */
1187 static void readfd(int fd, char *buffer, size_t N)
1193 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1197 if (fd == write_batch_monitor_in) {
1198 if ((size_t)write(batch_fd, buffer, total) != total)
1199 exit_cleanup(RERR_FILEIO);
1202 if (fd == flist_forward_from)
1203 writefd(iobuf_f_out, buffer, total);
1205 if (fd == sock_f_in)
1206 stats.total_read += total;
1209 unsigned short read_shortint(int f)
1213 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1216 int32 read_int(int f)
1223 #if SIZEOF_INT32 > 4
1224 if (num & (int32)0x80000000)
1225 num |= ~(int32)0xffffffff;
1230 int32 read_varint(int f)
1240 readfd(f, (char*)&ch, 1);
1241 extra = int_byte_extra[ch / 4];
1243 uchar bit = ((uchar)1<<(8-extra));
1244 if (extra >= (int)sizeof u.b) {
1245 rprintf(FERROR, "Overflow in read_varint()\n");
1246 exit_cleanup(RERR_STREAMIO);
1248 readfd(f, u.b, extra);
1249 u.b[extra] = ch & (bit-1);
1252 #if CAREFUL_ALIGNMENT
1255 #if SIZEOF_INT32 > 4
1256 if (u.x & (int32)0x80000000)
1257 u.x |= ~(int32)0xffffffff;
1262 int64 read_varlong(int f, uchar min_bytes)
1271 #if SIZEOF_INT64 < 8
1276 readfd(f, b2, min_bytes);
1277 memcpy(u.b, b2+1, min_bytes-1);
1278 extra = int_byte_extra[CVAL(b2, 0) / 4];
1280 uchar bit = ((uchar)1<<(8-extra));
1281 if (min_bytes + extra > (int)sizeof u.b) {
1282 rprintf(FERROR, "Overflow in read_varlong()\n");
1283 exit_cleanup(RERR_STREAMIO);
1285 readfd(f, u.b + min_bytes - 1, extra);
1286 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1287 #if SIZEOF_INT64 < 8
1288 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1289 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1290 exit_cleanup(RERR_UNSUPPORTED);
1294 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1295 #if SIZEOF_INT64 < 8
1297 #elif CAREFUL_ALIGNMENT
1298 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1303 int64 read_longint(int f)
1305 #if SIZEOF_INT64 >= 8
1308 int32 num = read_int(f);
1310 if (num != (int32)0xffffffff)
1313 #if SIZEOF_INT64 < 8
1314 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1315 exit_cleanup(RERR_UNSUPPORTED);
1318 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1322 void read_buf(int f, char *buf, size_t len)
1327 void read_sbuf(int f, char *buf, size_t len)
1329 readfd(f, buf, len);
1333 uchar read_byte(int f)
1336 readfd(f, (char *)&c, 1);
1340 int read_vstring(int f, char *buf, int bufsize)
1342 int len = read_byte(f);
1345 len = (len & ~0x80) * 0x100 + read_byte(f);
1347 if (len >= bufsize) {
1348 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1354 readfd(f, buf, len);
1359 /* Populate a sum_struct with values from the socket. This is
1360 * called by both the sender and the receiver. */
1361 void read_sum_head(int f, struct sum_struct *sum)
1363 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1364 sum->count = read_int(f);
1365 if (sum->count < 0) {
1366 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1367 (long)sum->count, who_am_i());
1368 exit_cleanup(RERR_PROTOCOL);
1370 sum->blength = read_int(f);
1371 if (sum->blength < 0 || sum->blength > max_blength) {
1372 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1373 (long)sum->blength, who_am_i());
1374 exit_cleanup(RERR_PROTOCOL);
1376 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1377 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1378 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1379 sum->s2length, who_am_i());
1380 exit_cleanup(RERR_PROTOCOL);
1382 sum->remainder = read_int(f);
1383 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1384 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1385 (long)sum->remainder, who_am_i());
1386 exit_cleanup(RERR_PROTOCOL);
1390 /* Send the values from a sum_struct over the socket. Set sum to
1391 * NULL if there are no checksums to send. This is called by both
1392 * the generator and the sender. */
1393 void write_sum_head(int f, struct sum_struct *sum)
1395 static struct sum_struct null_sum;
1400 write_int(f, sum->count);
1401 write_int(f, sum->blength);
1402 if (protocol_version >= 27)
1403 write_int(f, sum->s2length);
1404 write_int(f, sum->remainder);
1408 * Sleep after writing to limit I/O bandwidth usage.
1410 * @todo Rather than sleeping after each write, it might be better to
1411 * use some kind of averaging. The current algorithm seems to always
1412 * use a bit less bandwidth than specified, because it doesn't make up
1413 * for slow periods. But arguably this is a feature. In addition, we
1414 * ought to take the time used to write the data into account.
1416 * During some phases of big transfers (file FOO is uptodate) this is
1417 * called with a small bytes_written every time. As the kernel has to
1418 * round small waits up to guarantee that we actually wait at least the
1419 * requested number of microseconds, this can become grossly inaccurate.
1420 * We therefore keep track of the bytes we've written over time and only
1421 * sleep when the accumulated delay is at least 1 tenth of a second.
1423 static void sleep_for_bwlimit(int bytes_written)
1425 static struct timeval prior_tv;
1426 static long total_written = 0;
1427 struct timeval tv, start_tv;
1428 long elapsed_usec, sleep_usec;
1430 #define ONE_SEC 1000000L /* # of microseconds in a second */
1432 if (!bwlimit_writemax)
1435 total_written += bytes_written;
1437 gettimeofday(&start_tv, NULL);
1438 if (prior_tv.tv_sec) {
1439 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1440 + (start_tv.tv_usec - prior_tv.tv_usec);
1441 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1442 if (total_written < 0)
1446 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1447 if (sleep_usec < ONE_SEC / 10) {
1448 prior_tv = start_tv;
1452 tv.tv_sec = sleep_usec / ONE_SEC;
1453 tv.tv_usec = sleep_usec % ONE_SEC;
1454 select(0, NULL, NULL, NULL, &tv);
1456 gettimeofday(&prior_tv, NULL);
1457 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1458 + (prior_tv.tv_usec - start_tv.tv_usec);
1459 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1462 /* Write len bytes to the file descriptor fd, looping as necessary to get
1463 * the job done and also (in certain circumstances) reading any data on
1464 * msg_fd_in to avoid deadlock.
1466 * This function underlies the multiplexing system. The body of the
1467 * application never calls this function directly. */
1468 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1470 size_t n, total = 0;
1471 fd_set w_fds, r_fds, e_fds;
1472 int maxfd, count, cnt, using_r_fds;
1477 defer_forwarding_messages++, defer_inc++;
1479 while (total < len) {
1486 if (msg_fd_in >= 0) {
1488 FD_SET(msg_fd_in, &r_fds);
1489 if (msg_fd_in > maxfd)
1495 tv.tv_sec = select_timeout;
1499 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1500 &w_fds, &e_fds, &tv);
1503 if (count < 0 && errno == EBADF)
1504 exit_cleanup(RERR_SOCKETIO);
1509 /*if (FD_ISSET(fd, &e_fds))
1510 rprintf(FINFO, "select exception on fd %d\n", fd); */
1512 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1515 if (!FD_ISSET(fd, &w_fds))
1519 if (bwlimit_writemax && n > bwlimit_writemax)
1520 n = bwlimit_writemax;
1521 cnt = write(fd, buf + total, n);
1527 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1533 /* Don't try to write errors back across the stream. */
1534 if (fd == sock_f_out)
1535 io_end_multiplex_out();
1536 /* Don't try to write errors down a failing msg pipe. */
1537 if (am_server && fd == msg_fd_out)
1538 exit_cleanup(RERR_STREAMIO);
1539 rsyserr(FERROR, errno,
1540 "writefd_unbuffered failed to write %ld bytes [%s]",
1541 (long)len, who_am_i());
1542 /* If the other side is sending us error messages, try
1543 * to grab any messages they sent before they died. */
1544 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1548 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1550 exit_cleanup(RERR_STREAMIO);
1554 defer_forwarding_messages++, defer_inc++;
1556 if (fd == sock_f_out) {
1557 if (io_timeout || am_generator)
1558 last_io_out = time(NULL);
1559 sleep_for_bwlimit(cnt);
1564 defer_inc -= defer_forwarding_keep;
1565 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1569 void io_flush(int flush_it_all)
1571 if (!iobuf_out_cnt || no_flush)
1574 if (io_multiplexing_out)
1575 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1577 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1580 if (flush_it_all && !defer_forwarding_messages)
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 if ((size_t)write(batch_fd, buf, len) != len)
1591 exit_cleanup(RERR_FILEIO);
1594 if (!iobuf_out || fd != iobuf_f_out) {
1595 writefd_unbuffered(fd, buf, len);
1600 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1602 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1608 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1609 io_flush(NORMAL_FLUSH);
1613 void write_shortint(int f, unsigned short x)
1617 b[1] = (char)(x >> 8);
1621 void write_int(int f, int32 x)
1628 void write_varint(int f, int32 x)
1636 while (cnt > 1 && b[cnt] == 0)
1638 bit = ((uchar)1<<(7-cnt+1));
1639 if (CVAL(b, cnt) >= bit) {
1643 *b = b[cnt] | ~(bit*2-1);
1650 void write_varlong(int f, int64 x, uchar min_bytes)
1657 #if SIZEOF_INT64 >= 8
1658 SIVAL(b, 5, x >> 32);
1660 if (x <= 0x7FFFFFFF && x >= 0)
1661 memset(b + 5, 0, 4);
1663 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1664 exit_cleanup(RERR_UNSUPPORTED);
1668 while (cnt > min_bytes && b[cnt] == 0)
1670 bit = ((uchar)1<<(7-cnt+min_bytes));
1671 if (CVAL(b, cnt) >= bit) {
1674 } else if (cnt > min_bytes)
1675 *b = b[cnt] | ~(bit*2-1);
1683 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1684 * 64-bit types on this platform.
1686 void write_longint(int f, int64 x)
1688 char b[12], * const s = b+4;
1691 if (x <= 0x7FFFFFFF && x >= 0) {
1696 #if SIZEOF_INT64 < 8
1697 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1698 exit_cleanup(RERR_UNSUPPORTED);
1701 SIVAL(s, 4, x >> 32);
1706 void write_buf(int f, const char *buf, size_t len)
1711 /** Write a string to the connection */
1712 void write_sbuf(int f, const char *buf)
1714 writefd(f, buf, strlen(buf));
1717 void write_byte(int f, uchar c)
1719 writefd(f, (char *)&c, 1);
1722 void write_vstring(int f, const char *str, int len)
1724 uchar lenbuf[3], *lb = lenbuf;
1729 "attempting to send over-long vstring (%d > %d)\n",
1731 exit_cleanup(RERR_PROTOCOL);
1733 *lb++ = len / 0x100 + 0x80;
1737 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1739 writefd(f, str, len);
1742 /* Send a file-list index using a byte-reduction method. */
1743 void write_ndx(int f, int32 ndx)
1745 static int32 prev_positive = -1, prev_negative = 1;
1746 int32 diff, cnt = 0;
1749 if (protocol_version < 30 || read_batch) {
1754 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1755 * negative nums as a positive after sending a leading 0xFF. */
1757 diff = ndx - prev_positive;
1758 prev_positive = ndx;
1759 } else if (ndx == NDX_DONE) {
1764 b[cnt++] = (char)0xFF;
1766 diff = ndx - prev_negative;
1767 prev_negative = ndx;
1770 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1771 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1772 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1773 if (diff < 0xFE && diff > 0)
1774 b[cnt++] = (char)diff;
1775 else if (diff < 0 || diff > 0x7FFF) {
1776 b[cnt++] = (char)0xFE;
1777 b[cnt++] = (char)((ndx >> 24) | 0x80);
1778 b[cnt++] = (char)ndx;
1779 b[cnt++] = (char)(ndx >> 8);
1780 b[cnt++] = (char)(ndx >> 16);
1782 b[cnt++] = (char)0xFE;
1783 b[cnt++] = (char)(diff >> 8);
1784 b[cnt++] = (char)diff;
1789 /* Receive a file-list index using a byte-reduction method. */
1790 int32 read_ndx(int f)
1792 static int32 prev_positive = -1, prev_negative = 1;
1793 int32 *prev_ptr, num;
1796 if (protocol_version < 30)
1800 if (CVAL(b, 0) == 0xFF) {
1802 prev_ptr = &prev_negative;
1803 } else if (CVAL(b, 0) == 0)
1806 prev_ptr = &prev_positive;
1807 if (CVAL(b, 0) == 0xFE) {
1809 if (CVAL(b, 0) & 0x80) {
1810 b[3] = CVAL(b, 0) & ~0x80;
1815 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1817 num = UVAL(b, 0) + *prev_ptr;
1819 if (prev_ptr == &prev_negative)
1824 /* Read a line of up to bufsiz-1 characters into buf. Strips
1825 * the (required) trailing newline and all carriage returns.
1826 * Returns 1 for success; 0 for I/O error or truncation. */
1827 int read_line_old(int f, char *buf, size_t bufsiz)
1829 bufsiz--; /* leave room for the null */
1830 while (bufsiz > 0) {
1832 read_buf(f, buf, 1);
1837 if (buf[0] != '\r') {
1846 void io_printf(int fd, const char *format, ...)
1849 char buf[BIGPATHBUFLEN];
1852 va_start(ap, format);
1853 len = vsnprintf(buf, sizeof buf, format, ap);
1857 exit_cleanup(RERR_STREAMIO);
1859 if (len > (int)sizeof buf) {
1860 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1861 exit_cleanup(RERR_STREAMIO);
1864 write_sbuf(fd, buf);
1867 /** Setup for multiplexing a MSG_* stream with the data stream. */
1868 void io_start_multiplex_out(void)
1870 io_flush(NORMAL_FLUSH);
1871 io_start_buffering_out(sock_f_out);
1872 io_multiplexing_out = 1;
1875 /** Setup for multiplexing a MSG_* stream with the data stream. */
1876 void io_start_multiplex_in(void)
1878 io_flush(NORMAL_FLUSH);
1879 io_start_buffering_in(sock_f_in);
1880 io_multiplexing_in = 1;
1883 /** Write an message to the multiplexed data stream. */
1884 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1886 if (!io_multiplexing_out)
1888 io_flush(NORMAL_FLUSH);
1889 stats.total_written += (len+4);
1890 mplex_write(sock_f_out, code, buf, len, convert);
1894 void io_end_multiplex_in(void)
1896 io_multiplexing_in = 0;
1897 io_end_buffering_in();
1900 /** Stop output multiplexing. */
1901 void io_end_multiplex_out(void)
1903 io_multiplexing_out = 0;
1904 io_end_buffering_out();
1907 void start_write_batch(int fd)
1909 /* Some communication has already taken place, but we don't
1910 * enable batch writing until here so that we can write a
1911 * canonical record of the communication even though the
1912 * actual communication so far depends on whether a daemon
1914 write_int(batch_fd, protocol_version);
1915 if (protocol_version >= 30)
1916 write_byte(batch_fd, inc_recurse);
1917 write_int(batch_fd, checksum_seed);
1920 write_batch_monitor_out = fd;
1922 write_batch_monitor_in = fd;
1925 void stop_write_batch(void)
1927 write_batch_monitor_out = -1;
1928 write_batch_monitor_in = -1;