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(). */
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;
49 extern int read_batch;
50 extern int protect_args;
51 extern int checksum_seed;
52 extern int protocol_version;
53 extern int remove_source_files;
54 extern int preserve_hard_links;
55 extern struct stats stats;
56 extern struct file_list *cur_flist;
58 extern int filesfrom_convert;
59 extern iconv_t ic_send, ic_recv;
62 int csum_length = SHORT_SUM_LENGTH; /* initial value */
64 int ignore_timeout = 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, keep_defer_forwarding = 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 static flist_ndx_list redo_list, hlink_list;
129 struct msg_list_item {
130 struct msg_list_item *next;
136 struct msg_list_item *head, *tail;
139 static struct msg_list msg_queue;
141 static void got_flist_entry_status(enum festatus status, const char *buf)
143 int ndx = IVAL(buf, 0);
144 struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
146 if (remove_source_files) {
148 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
152 flist->in_progress--;
156 if (remove_source_files)
157 send_msg(MSG_SUCCESS, buf, 4, 0);
158 if (preserve_hard_links) {
159 struct file_struct *file = flist->files[ndx - flist->ndx_start];
160 if (F_IS_HLINKED(file)) {
161 flist_ndx_push(&hlink_list, ndx);
162 flist->in_progress++;
169 flist->in_progress++;
174 flist_ndx_push(&redo_list, ndx);
181 static void check_timeout(void)
185 if (!io_timeout || ignore_timeout)
189 last_io_in = time(NULL);
195 if (t - last_io_in >= io_timeout) {
196 if (!am_server && !am_daemon) {
197 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
198 (int)(t-last_io_in));
200 exit_cleanup(RERR_TIMEOUT);
204 /* Note the fds used for the main socket (which might really be a pipe
205 * for a local transfer, but we can ignore that). */
206 void io_set_sock_fds(int f_in, int f_out)
212 void set_io_timeout(int secs)
216 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
217 select_timeout = SELECT_TIMEOUT;
219 select_timeout = io_timeout;
221 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
224 /* Setup the fd used to receive MSG_* messages. Only needed during the
225 * early stages of being a local sender (up through the sending of the
226 * file list) or when we're the generator (to fetch the messages from
228 void set_msg_fd_in(int fd)
233 /* Setup the fd used to send our MSG_* messages. Only needed when
234 * we're the receiver (to send our messages to the generator). */
235 void set_msg_fd_out(int fd)
238 set_nonblocking(msg_fd_out);
241 /* Add a message to the pending MSG_* list. */
242 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
244 struct msg_list_item *m;
245 int sz = len + 4 + sizeof m[0] - 1;
247 if (!(m = (struct msg_list_item *)new_array(char, sz)))
248 out_of_memory("msg_list_add");
250 m->convert = convert;
251 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
252 memcpy(m->buf + 4, buf, len);
260 static inline int flush_a_msg(int fd)
262 struct msg_list_item *m = msg_queue.head;
263 int len = IVAL(m->buf, 0) & 0xFFFFFF;
264 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
266 if (!(msg_queue.head = m->next))
267 msg_queue.tail = NULL;
269 defer_forwarding_messages++;
270 mplex_write(fd, tag, m->buf + 4, len, m->convert);
271 defer_forwarding_messages--;
278 static void msg_flush(void)
281 while (msg_queue.head && io_multiplexing_out)
282 stats.total_written += flush_a_msg(sock_f_out) + 4;
284 while (msg_queue.head)
285 (void)flush_a_msg(msg_fd_out);
289 static void check_for_d_option_error(const char *msg)
291 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
296 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
299 msg += sizeof REMOTE_OPTION_ERROR - 1;
300 if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
301 || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
304 for ( ; *msg != ':'; msg++) {
307 else if (*msg == 'e')
309 else if (strchr(rsync263_opts, *msg) == NULL)
315 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
319 /* Read a message from the MSG_* fd and handle it. This is called either
320 * during the early stages of being a local sender (up through the sending
321 * of the file list) or when we're the generator (to fetch the messages
322 * from the receiver). */
323 static void read_msg_fd(void)
327 struct file_list *flist;
331 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
332 * to this routine from writefd_unbuffered(). */
335 defer_forwarding_messages++;
340 len = tag & 0xFFFFFF;
341 tag = (tag >> 24) - MPLEX_BASE;
345 if (len < 0 || len > 1 || !am_generator) {
347 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
348 tag, len, who_am_i(),
349 inc_recurse ? "/inc" : "");
350 exit_cleanup(RERR_STREAMIO);
353 readfd(fd, buf, len);
354 stats.total_read = read_varlong(fd, 3);
359 if (len != 4 || !am_generator)
362 got_flist_entry_status(FES_REDO, buf);
365 if (len != 4 || !am_generator || !inc_recurse)
368 /* Read extra file list from receiver. */
369 assert(iobuf_in != NULL);
370 assert(iobuf_f_in == fd);
371 if (DEBUG_GTE(FLIST, 2)) {
372 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
373 who_am_i(), IVAL(buf,0));
375 flist = recv_file_list(fd);
376 flist->parent_ndx = IVAL(buf,0);
377 #ifdef SUPPORT_HARD_LINKS
378 if (preserve_hard_links)
379 match_hard_links(flist);
383 if (len != 0 || !am_generator || !inc_recurse)
390 readfd(fd, buf, len);
391 io_error |= IVAL(buf, 0);
394 if (len >= (int)sizeof buf || !am_generator)
396 readfd(fd, buf, len);
397 send_msg(MSG_DELETED, buf, len, 1);
400 if (len != 4 || !am_generator)
403 got_flist_entry_status(FES_SUCCESS, buf);
406 if (len != 4 || !am_generator)
409 got_flist_entry_status(FES_NO_SEND, buf);
411 case MSG_ERROR_SOCKET:
416 if (tag == MSG_ERROR_SOCKET)
417 io_end_multiplex_out();
429 rwrite((enum logcode)tag, buf, n, !am_generator);
434 rprintf(FERROR, "unknown message %d:%d [%s]\n",
435 tag, len, who_am_i());
436 exit_cleanup(RERR_STREAMIO);
441 if (!--defer_forwarding_messages && !no_flush)
445 /* This is used by the generator to limit how many file transfers can
446 * be active at once when --remove-source-files is specified. Without
447 * this, sender-side deletions were mostly happening at the end. */
448 void increment_active_files(int ndx, int itemizing, enum logcode code)
450 /* TODO: tune these limits? */
451 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
452 check_for_finished_files(itemizing, code, 0);
454 io_flush(NORMAL_FLUSH);
460 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
463 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
464 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
466 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
470 /* We need to convert buf before doing anything else so that we
471 * can include the (converted) byte length in the message header. */
472 if (convert && ic_send != (iconv_t)-1) {
475 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
476 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
478 iconvbufs(ic_send, &inbuf, &outbuf,
479 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
481 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
482 exit_cleanup(RERR_UNSUPPORTED);
485 n = len = outbuf.len;
488 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
489 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
491 memcpy(buffer + 4, buf, n);
493 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
495 keep_defer_forwarding++; /* defer_forwarding_messages++ on return */
496 writefd_unbuffered(fd, buffer, n+4);
497 keep_defer_forwarding--;
500 writefd_unbuffered(fd, buf+n, len-n);
502 if (!--defer_forwarding_messages && !no_flush)
506 int send_msg(enum msgcode code, const char *buf, int len, int convert)
508 if (msg_fd_out < 0) {
509 if (!defer_forwarding_messages)
510 return io_multiplex_write(code, buf, len, convert);
511 if (!io_multiplexing_out)
513 msg_list_add(&msg_queue, code, buf, len, convert);
516 if (flist_forward_from >= 0)
517 msg_list_add(&msg_queue, code, buf, len, convert);
519 mplex_write(msg_fd_out, code, buf, len, convert);
523 void send_msg_int(enum msgcode code, int num)
526 SIVAL(numbuf, 0, num);
527 send_msg(code, numbuf, 4, 0);
530 void wait_for_receiver(void)
532 if (io_flush(FULL_FLUSH))
537 int get_redo_num(void)
539 return flist_ndx_pop(&redo_list);
542 int get_hlink_num(void)
544 return flist_ndx_pop(&hlink_list);
548 * When we're the receiver and we have a local --files-from list of names
549 * that needs to be sent over the socket to the sender, we have to do two
550 * things at the same time: send the sender a list of what files we're
551 * processing and read the incoming file+info list from the sender. We do
552 * this by augmenting the read_timeout() function to copy this data. It
553 * uses ff_buf to read a block of data from f_in (when it is ready, since
554 * it might be a pipe) and then blast it out f_out (when it is ready to
555 * receive more data).
557 void io_set_filesfrom_fds(int f_in, int f_out)
559 io_filesfrom_f_in = f_in;
560 io_filesfrom_f_out = f_out;
561 alloc_xbuf(&ff_buf, 2048);
564 alloc_xbuf(&iconv_buf, 1024);
568 /* It's almost always an error to get an EOF when we're trying to read from the
569 * network, because the protocol is (for the most part) self-terminating.
571 * There is one case for the receiver when it is at the end of the transfer
572 * (hanging around reading any keep-alive packets that might come its way): if
573 * the sender dies before the generator's kill-signal comes through, we can end
574 * up here needing to loop until the kill-signal arrives. In this situation,
575 * kluge_around_eof will be < 0.
577 * There is another case for older protocol versions (< 24) where the module
578 * listing was not terminated, so we must ignore an EOF error in that case and
579 * exit. In this situation, kluge_around_eof will be > 0. */
580 static void whine_about_eof(int fd)
582 if (kluge_around_eof && fd == sock_f_in) {
584 if (kluge_around_eof > 0)
586 /* If we're still here after 10 seconds, exit with an error. */
587 for (i = 10*1000/20; i--; )
591 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
592 "(%s bytes received so far) [%s]\n",
593 big_num(stats.total_read), who_am_i());
595 exit_cleanup(RERR_STREAMIO);
599 * Read from a socket with I/O timeout. return the number of bytes
600 * read. If no bytes can be read then exit, never return a number <= 0.
602 * TODO: If the remote shell connection fails, then current versions
603 * actually report an "unexpected EOF" error here. Since it's a
604 * fairly common mistake to try to use rsh when ssh is required, we
605 * should trap that: if we fail to read any data at all, we should
606 * give a better explanation. We can tell whether the connection has
607 * started by looking e.g. at whether the remote version is known yet.
609 static int read_timeout(int fd, char *buf, size_t len)
613 io_flush(FULL_FLUSH);
616 /* until we manage to read *something* */
625 if (io_filesfrom_f_out >= 0) {
627 if (ff_buf.len == 0) {
628 if (io_filesfrom_f_in >= 0) {
629 FD_SET(io_filesfrom_f_in, &r_fds);
630 new_fd = io_filesfrom_f_in;
632 io_filesfrom_f_out = -1;
636 FD_SET(io_filesfrom_f_out, &w_fds);
637 new_fd = io_filesfrom_f_out;
643 tv.tv_sec = select_timeout;
648 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
651 if (errno == EBADF) {
652 defer_forwarding_messages = 0;
653 exit_cleanup(RERR_SOCKETIO);
659 if (io_filesfrom_f_out >= 0) {
661 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
662 int l = write(io_filesfrom_f_out,
663 ff_buf.buf + ff_buf.pos,
666 if (!(ff_buf.len -= l))
670 } else if (errno != EINTR) {
671 /* XXX should we complain? */
672 io_filesfrom_f_out = -1;
675 } else if (io_filesfrom_f_in >= 0) {
676 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
678 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
680 xbuf *ibuf = &ff_buf;
682 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
684 if (l == 0 || errno != EINTR) {
685 /* Send end-of-file marker */
686 memcpy(ff_buf.buf, "\0\0", 2);
687 ff_buf.len = ff_lastchar? 2 : 1;
689 io_filesfrom_f_in = -1;
693 if (filesfrom_convert) {
696 iconvbufs(ic_send, &iconv_buf, &ff_buf,
697 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
702 char *s = ff_buf.buf + l;
703 /* Transform CR and/or LF into '\0' */
704 while (s-- > ff_buf.buf) {
705 if (*s == '\n' || *s == '\r')
710 /* Last buf ended with a '\0', so don't
711 * let this buf start with one. */
712 while (l && ff_buf.buf[ff_buf.pos] == '\0')
718 char *f = ff_buf.buf + ff_buf.pos;
721 /* Eliminate any multi-'\0' runs. */
723 if (!(*t++ = *f++)) {
724 while (f != eob && !*f)
736 if (!FD_ISSET(fd, &r_fds))
739 n = read(fd, buf, len);
743 whine_about_eof(fd); /* Doesn't return. */
744 if (errno == EINTR || errno == EWOULDBLOCK
748 /* Don't write errors on a dead socket. */
749 if (fd == sock_f_in) {
750 io_end_multiplex_out();
751 rsyserr(FERROR_SOCKET, errno, "read error");
753 rsyserr(FERROR, errno, "read error");
754 exit_cleanup(RERR_STREAMIO);
761 if (fd == sock_f_in && io_timeout)
762 last_io_in = time(NULL);
768 /* Read a line into the "buf" buffer. */
769 int read_line(int fd, char *buf, size_t bufsiz, int flags)
775 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
776 realloc_xbuf(&iconv_buf, bufsiz + 1024);
781 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
785 eob = s + bufsiz - 1;
787 cnt = read(fd, &ch, 1);
788 if (cnt < 0 && (errno == EWOULDBLOCK
789 || errno == EINTR || errno == EAGAIN)) {
796 tv.tv_sec = select_timeout;
798 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
800 /*if (FD_ISSET(fd, &e_fds))
801 rprintf(FINFO, "select exception on fd %d\n", fd); */
806 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
807 /* Skip empty lines if dumping comments. */
808 if (flags & RL_DUMP_COMMENTS && s == buf)
817 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
821 if (flags & RL_CONVERT) {
823 INIT_XBUF(outbuf, buf, 0, bufsiz);
825 iconv_buf.len = s - iconv_buf.buf;
826 iconvbufs(ic_recv, &iconv_buf, &outbuf,
827 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
828 outbuf.buf[outbuf.len] = '\0';
836 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
837 char ***argv_p, int *argc_p, char **request_p)
839 int maxargs = MAX_ARGS;
843 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
846 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
849 if (!(argv = new_array(char *, maxargs)))
850 out_of_memory("read_args");
851 if (mod_name && !protect_args)
852 argv[argc++] = "rsyncd";
855 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
858 if (argc == maxargs-1) {
860 if (!(argv = realloc_array(argv, char *, maxargs)))
861 out_of_memory("read_args");
866 *request_p = strdup(buf);
870 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
872 glob_expand(buf, &argv, &argc, &maxargs);
874 if (!(p = strdup(buf)))
875 out_of_memory("read_args");
877 if (*p == '.' && p[1] == '\0')
883 glob_expand(NULL, NULL, NULL, NULL);
889 int io_start_buffering_out(int f_out)
892 assert(f_out == iobuf_f_out);
895 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
896 out_of_memory("io_start_buffering_out");
902 int io_start_buffering_in(int f_in)
905 assert(f_in == iobuf_f_in);
908 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
909 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
910 out_of_memory("io_start_buffering_in");
915 void io_end_buffering_in(void)
922 iobuf_in_remaining = 0;
926 void io_end_buffering_out(void)
930 io_flush(FULL_FLUSH);
936 void maybe_flush_socket(int important)
938 if (iobuf_out && iobuf_out_cnt
939 && (important || time(NULL) - last_io_out >= 5))
940 io_flush(NORMAL_FLUSH);
943 void maybe_send_keepalive(void)
945 if (time(NULL) - last_io_out >= allowed_lull) {
946 if (!iobuf_out || !iobuf_out_cnt) {
947 if (protocol_version < 29)
948 return; /* there's nothing we can do */
949 if (protocol_version >= 30)
950 send_msg(MSG_NOOP, "", 0, 0);
952 write_int(sock_f_out, cur_flist->used);
953 write_shortint(sock_f_out, ITEM_IS_NEW);
957 io_flush(NORMAL_FLUSH);
961 void start_flist_forward(int f_in)
963 assert(iobuf_out != NULL);
964 assert(iobuf_f_out == msg_fd_out);
965 flist_forward_from = f_in;
968 void stop_flist_forward()
970 flist_forward_from = -1;
971 io_flush(FULL_FLUSH);
975 * Continue trying to read len bytes - don't return until len has been
978 static void read_loop(int fd, char *buf, size_t len)
981 int n = read_timeout(fd, buf, len);
989 * Read from the file descriptor handling multiplexing - return number
992 * Never returns <= 0.
994 static int readfd_unbuffered(int fd, char *buf, size_t len)
998 char line[BIGPATHBUFLEN];
1000 if (!iobuf_in || fd != iobuf_f_in)
1001 return read_timeout(fd, buf, len);
1003 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1004 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1009 if (iobuf_in_remaining) {
1010 len = MIN(len, iobuf_in_remaining);
1011 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1012 iobuf_in_ndx += len;
1013 iobuf_in_remaining -= len;
1018 read_loop(fd, line, 4);
1019 tag = IVAL(line, 0);
1021 msg_bytes = tag & 0xFFFFFF;
1022 tag = (tag >> 24) - MPLEX_BASE;
1026 if (msg_bytes > iobuf_in_siz) {
1027 if (!(iobuf_in = realloc_array(iobuf_in, char,
1029 out_of_memory("readfd_unbuffered");
1030 iobuf_in_siz = msg_bytes;
1032 read_loop(fd, iobuf_in, msg_bytes);
1033 iobuf_in_remaining = msg_bytes;
1038 maybe_send_keepalive();
1043 read_loop(fd, line, msg_bytes);
1044 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1045 io_error |= IVAL(line, 0);
1048 if (msg_bytes >= sizeof line)
1051 if (ic_recv != (iconv_t)-1) {
1056 INIT_CONST_XBUF(outbuf, line);
1057 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1060 inbuf.len = msg_bytes > sizeof ibuf
1061 ? sizeof ibuf : msg_bytes;
1062 read_loop(fd, inbuf.buf, inbuf.len);
1063 if (!(msg_bytes -= inbuf.len)
1064 && !ibuf[inbuf.len-1])
1065 inbuf.len--, add_null = 1;
1066 if (iconvbufs(ic_send, &inbuf, &outbuf,
1067 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1071 if (outbuf.len == outbuf.size)
1073 outbuf.buf[outbuf.len++] = '\0';
1075 msg_bytes = outbuf.len;
1078 read_loop(fd, line, msg_bytes);
1079 /* A directory name was sent with the trailing null */
1080 if (msg_bytes > 0 && !line[msg_bytes-1])
1081 log_delete(line, S_IFDIR);
1083 line[msg_bytes] = '\0';
1084 log_delete(line, S_IFREG);
1088 if (msg_bytes != 4) {
1090 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1091 tag, (long)msg_bytes, who_am_i());
1092 exit_cleanup(RERR_STREAMIO);
1094 read_loop(fd, line, msg_bytes);
1095 successful_send(IVAL(line, 0));
1100 read_loop(fd, line, msg_bytes);
1101 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1105 case MSG_ERROR_XFER:
1107 if (msg_bytes >= sizeof line) {
1110 "multiplexing overflow %d:%ld [%s]\n",
1111 tag, (long)msg_bytes, who_am_i());
1112 exit_cleanup(RERR_STREAMIO);
1114 read_loop(fd, line, msg_bytes);
1115 rwrite((enum logcode)tag, line, msg_bytes, 1);
1116 if (first_message) {
1117 if (list_only && !am_sender && tag == 1) {
1118 line[msg_bytes] = '\0';
1119 check_for_d_option_error(line);
1125 rprintf(FERROR, "unexpected tag %d [%s]\n",
1127 exit_cleanup(RERR_STREAMIO);
1131 if (iobuf_in_remaining == 0)
1132 io_flush(NORMAL_FLUSH);
1137 /* Do a buffered read from fd. Don't return until all N bytes have
1138 * been read. If all N can't be read then exit with an error. */
1139 static void readfd(int fd, char *buffer, size_t N)
1145 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1149 if (fd == write_batch_monitor_in) {
1150 if ((size_t)write(batch_fd, buffer, total) != total)
1151 exit_cleanup(RERR_FILEIO);
1154 if (fd == flist_forward_from)
1155 writefd(iobuf_f_out, buffer, total);
1157 if (fd == sock_f_in)
1158 stats.total_read += total;
1161 unsigned short read_shortint(int f)
1165 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1168 int32 read_int(int f)
1175 #if SIZEOF_INT32 > 4
1176 if (num & (int32)0x80000000)
1177 num |= ~(int32)0xffffffff;
1182 int32 read_varint(int f)
1192 readfd(f, (char*)&ch, 1);
1193 extra = int_byte_extra[ch / 4];
1195 uchar bit = ((uchar)1<<(8-extra));
1196 if (extra >= (int)sizeof u.b) {
1197 rprintf(FERROR, "Overflow in read_varint()\n");
1198 exit_cleanup(RERR_STREAMIO);
1200 readfd(f, u.b, extra);
1201 u.b[extra] = ch & (bit-1);
1204 #if CAREFUL_ALIGNMENT
1207 #if SIZEOF_INT32 > 4
1208 if (u.x & (int32)0x80000000)
1209 u.x |= ~(int32)0xffffffff;
1214 int64 read_varlong(int f, uchar min_bytes)
1223 #if SIZEOF_INT64 < 8
1228 readfd(f, b2, min_bytes);
1229 memcpy(u.b, b2+1, min_bytes-1);
1230 extra = int_byte_extra[CVAL(b2, 0) / 4];
1232 uchar bit = ((uchar)1<<(8-extra));
1233 if (min_bytes + extra > (int)sizeof u.b) {
1234 rprintf(FERROR, "Overflow in read_varlong()\n");
1235 exit_cleanup(RERR_STREAMIO);
1237 readfd(f, u.b + min_bytes - 1, extra);
1238 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1239 #if SIZEOF_INT64 < 8
1240 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1241 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1242 exit_cleanup(RERR_UNSUPPORTED);
1246 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1247 #if SIZEOF_INT64 < 8
1249 #elif CAREFUL_ALIGNMENT
1250 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1255 int64 read_longint(int f)
1257 #if SIZEOF_INT64 >= 8
1260 int32 num = read_int(f);
1262 if (num != (int32)0xffffffff)
1265 #if SIZEOF_INT64 < 8
1266 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1267 exit_cleanup(RERR_UNSUPPORTED);
1270 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1274 void read_buf(int f, char *buf, size_t len)
1279 void read_sbuf(int f, char *buf, size_t len)
1281 readfd(f, buf, len);
1285 uchar read_byte(int f)
1288 readfd(f, (char *)&c, 1);
1292 int read_vstring(int f, char *buf, int bufsize)
1294 int len = read_byte(f);
1297 len = (len & ~0x80) * 0x100 + read_byte(f);
1299 if (len >= bufsize) {
1300 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1306 readfd(f, buf, len);
1311 /* Populate a sum_struct with values from the socket. This is
1312 * called by both the sender and the receiver. */
1313 void read_sum_head(int f, struct sum_struct *sum)
1315 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1316 sum->count = read_int(f);
1317 if (sum->count < 0) {
1318 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1319 (long)sum->count, who_am_i());
1320 exit_cleanup(RERR_PROTOCOL);
1322 sum->blength = read_int(f);
1323 if (sum->blength < 0 || sum->blength > max_blength) {
1324 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1325 (long)sum->blength, who_am_i());
1326 exit_cleanup(RERR_PROTOCOL);
1328 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1329 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1330 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1331 sum->s2length, who_am_i());
1332 exit_cleanup(RERR_PROTOCOL);
1334 sum->remainder = read_int(f);
1335 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1336 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1337 (long)sum->remainder, who_am_i());
1338 exit_cleanup(RERR_PROTOCOL);
1342 /* Send the values from a sum_struct over the socket. Set sum to
1343 * NULL if there are no checksums to send. This is called by both
1344 * the generator and the sender. */
1345 void write_sum_head(int f, struct sum_struct *sum)
1347 static struct sum_struct null_sum;
1352 write_int(f, sum->count);
1353 write_int(f, sum->blength);
1354 if (protocol_version >= 27)
1355 write_int(f, sum->s2length);
1356 write_int(f, sum->remainder);
1360 * Sleep after writing to limit I/O bandwidth usage.
1362 * @todo Rather than sleeping after each write, it might be better to
1363 * use some kind of averaging. The current algorithm seems to always
1364 * use a bit less bandwidth than specified, because it doesn't make up
1365 * for slow periods. But arguably this is a feature. In addition, we
1366 * ought to take the time used to write the data into account.
1368 * During some phases of big transfers (file FOO is uptodate) this is
1369 * called with a small bytes_written every time. As the kernel has to
1370 * round small waits up to guarantee that we actually wait at least the
1371 * requested number of microseconds, this can become grossly inaccurate.
1372 * We therefore keep track of the bytes we've written over time and only
1373 * sleep when the accumulated delay is at least 1 tenth of a second.
1375 static void sleep_for_bwlimit(int bytes_written)
1377 static struct timeval prior_tv;
1378 static long total_written = 0;
1379 struct timeval tv, start_tv;
1380 long elapsed_usec, sleep_usec;
1382 #define ONE_SEC 1000000L /* # of microseconds in a second */
1384 if (!bwlimit_writemax)
1387 total_written += bytes_written;
1389 gettimeofday(&start_tv, NULL);
1390 if (prior_tv.tv_sec) {
1391 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1392 + (start_tv.tv_usec - prior_tv.tv_usec);
1393 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1394 if (total_written < 0)
1398 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1399 if (sleep_usec < ONE_SEC / 10) {
1400 prior_tv = start_tv;
1404 tv.tv_sec = sleep_usec / ONE_SEC;
1405 tv.tv_usec = sleep_usec % ONE_SEC;
1406 select(0, NULL, NULL, NULL, &tv);
1408 gettimeofday(&prior_tv, NULL);
1409 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1410 + (prior_tv.tv_usec - start_tv.tv_usec);
1411 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1414 /* Write len bytes to the file descriptor fd, looping as necessary to get
1415 * the job done and also (in certain circumstances) reading any data on
1416 * msg_fd_in to avoid deadlock.
1418 * This function underlies the multiplexing system. The body of the
1419 * application never calls this function directly. */
1420 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1422 size_t n, total = 0;
1423 fd_set w_fds, r_fds, e_fds;
1424 int maxfd, count, cnt, using_r_fds;
1429 defer_forwarding_messages++, defer_inc++;
1431 while (total < len) {
1438 if (msg_fd_in >= 0) {
1440 FD_SET(msg_fd_in, &r_fds);
1441 if (msg_fd_in > maxfd)
1447 tv.tv_sec = select_timeout;
1451 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1452 &w_fds, &e_fds, &tv);
1455 if (count < 0 && errno == EBADF)
1456 exit_cleanup(RERR_SOCKETIO);
1461 /*if (FD_ISSET(fd, &e_fds))
1462 rprintf(FINFO, "select exception on fd %d\n", fd); */
1464 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1467 if (!FD_ISSET(fd, &w_fds))
1471 if (bwlimit_writemax && n > bwlimit_writemax)
1472 n = bwlimit_writemax;
1473 cnt = write(fd, buf + total, n);
1479 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1485 /* Don't try to write errors back across the stream. */
1486 if (fd == sock_f_out)
1487 io_end_multiplex_out();
1488 /* Don't try to write errors down a failing msg pipe. */
1489 if (am_server && fd == msg_fd_out)
1490 exit_cleanup(RERR_STREAMIO);
1491 rsyserr(FERROR, errno,
1492 "writefd_unbuffered failed to write %ld bytes [%s]",
1493 (long)len, who_am_i());
1494 /* If the other side is sending us error messages, try
1495 * to grab any messages they sent before they died. */
1496 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1500 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1502 exit_cleanup(RERR_STREAMIO);
1506 defer_forwarding_messages++, defer_inc++;
1508 if (fd == sock_f_out) {
1509 if (io_timeout || am_generator)
1510 last_io_out = time(NULL);
1511 sleep_for_bwlimit(cnt);
1516 if (keep_defer_forwarding)
1518 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1522 int io_flush(int flush_it_all)
1524 int flushed_something = 0;
1529 if (iobuf_out_cnt) {
1530 if (io_multiplexing_out)
1531 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1533 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1535 flushed_something = 1;
1538 if (flush_it_all && !defer_forwarding_messages && msg_queue.head) {
1540 flushed_something = 1;
1543 return flushed_something;
1546 static void writefd(int fd, const char *buf, size_t len)
1548 if (fd == sock_f_out)
1549 stats.total_written += len;
1551 if (fd == write_batch_monitor_out) {
1552 if ((size_t)write(batch_fd, buf, len) != len)
1553 exit_cleanup(RERR_FILEIO);
1556 if (!iobuf_out || fd != iobuf_f_out) {
1557 writefd_unbuffered(fd, buf, len);
1562 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1564 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1570 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1571 io_flush(NORMAL_FLUSH);
1575 void write_shortint(int f, unsigned short x)
1579 b[1] = (char)(x >> 8);
1583 void write_int(int f, int32 x)
1590 void write_varint(int f, int32 x)
1598 while (cnt > 1 && b[cnt] == 0)
1600 bit = ((uchar)1<<(7-cnt+1));
1601 if (CVAL(b, cnt) >= bit) {
1605 *b = b[cnt] | ~(bit*2-1);
1612 void write_varlong(int f, int64 x, uchar min_bytes)
1619 #if SIZEOF_INT64 >= 8
1620 SIVAL(b, 5, x >> 32);
1622 if (x <= 0x7FFFFFFF && x >= 0)
1623 memset(b + 5, 0, 4);
1625 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1626 exit_cleanup(RERR_UNSUPPORTED);
1630 while (cnt > min_bytes && b[cnt] == 0)
1632 bit = ((uchar)1<<(7-cnt+min_bytes));
1633 if (CVAL(b, cnt) >= bit) {
1636 } else if (cnt > min_bytes)
1637 *b = b[cnt] | ~(bit*2-1);
1645 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1646 * 64-bit types on this platform.
1648 void write_longint(int f, int64 x)
1650 char b[12], * const s = b+4;
1653 if (x <= 0x7FFFFFFF && x >= 0) {
1658 #if SIZEOF_INT64 < 8
1659 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1660 exit_cleanup(RERR_UNSUPPORTED);
1663 SIVAL(s, 4, x >> 32);
1668 void write_buf(int f, const char *buf, size_t len)
1673 /** Write a string to the connection */
1674 void write_sbuf(int f, const char *buf)
1676 writefd(f, buf, strlen(buf));
1679 void write_byte(int f, uchar c)
1681 writefd(f, (char *)&c, 1);
1684 void write_vstring(int f, const char *str, int len)
1686 uchar lenbuf[3], *lb = lenbuf;
1691 "attempting to send over-long vstring (%d > %d)\n",
1693 exit_cleanup(RERR_PROTOCOL);
1695 *lb++ = len / 0x100 + 0x80;
1699 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1701 writefd(f, str, len);
1704 /* Send a file-list index using a byte-reduction method. */
1705 void write_ndx(int f, int32 ndx)
1707 static int32 prev_positive = -1, prev_negative = 1;
1708 int32 diff, cnt = 0;
1711 if (protocol_version < 30 || read_batch) {
1716 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1717 * negative nums as a positive after sending a leading 0xFF. */
1719 diff = ndx - prev_positive;
1720 prev_positive = ndx;
1721 } else if (ndx == NDX_DONE) {
1726 b[cnt++] = (char)0xFF;
1728 diff = ndx - prev_negative;
1729 prev_negative = ndx;
1732 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1733 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1734 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1735 if (diff < 0xFE && diff > 0)
1736 b[cnt++] = (char)diff;
1737 else if (diff < 0 || diff > 0x7FFF) {
1738 b[cnt++] = (char)0xFE;
1739 b[cnt++] = (char)((ndx >> 24) | 0x80);
1740 b[cnt++] = (char)ndx;
1741 b[cnt++] = (char)(ndx >> 8);
1742 b[cnt++] = (char)(ndx >> 16);
1744 b[cnt++] = (char)0xFE;
1745 b[cnt++] = (char)(diff >> 8);
1746 b[cnt++] = (char)diff;
1751 /* Receive a file-list index using a byte-reduction method. */
1752 int32 read_ndx(int f)
1754 static int32 prev_positive = -1, prev_negative = 1;
1755 int32 *prev_ptr, num;
1758 if (protocol_version < 30)
1762 if (CVAL(b, 0) == 0xFF) {
1764 prev_ptr = &prev_negative;
1765 } else if (CVAL(b, 0) == 0)
1768 prev_ptr = &prev_positive;
1769 if (CVAL(b, 0) == 0xFE) {
1771 if (CVAL(b, 0) & 0x80) {
1772 b[3] = CVAL(b, 0) & ~0x80;
1777 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1779 num = UVAL(b, 0) + *prev_ptr;
1781 if (prev_ptr == &prev_negative)
1786 /* Read a line of up to bufsiz-1 characters into buf. Strips
1787 * the (required) trailing newline and all carriage returns.
1788 * Returns 1 for success; 0 for I/O error or truncation. */
1789 int read_line_old(int f, char *buf, size_t bufsiz)
1791 bufsiz--; /* leave room for the null */
1792 while (bufsiz > 0) {
1794 read_buf(f, buf, 1);
1799 if (buf[0] != '\r') {
1808 void io_printf(int fd, const char *format, ...)
1811 char buf[BIGPATHBUFLEN];
1814 va_start(ap, format);
1815 len = vsnprintf(buf, sizeof buf, format, ap);
1819 exit_cleanup(RERR_STREAMIO);
1821 if (len > (int)sizeof buf) {
1822 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1823 exit_cleanup(RERR_STREAMIO);
1826 write_sbuf(fd, buf);
1829 /** Setup for multiplexing a MSG_* stream with the data stream. */
1830 void io_start_multiplex_out(void)
1832 io_flush(NORMAL_FLUSH);
1833 io_start_buffering_out(sock_f_out);
1834 io_multiplexing_out = 1;
1837 /** Setup for multiplexing a MSG_* stream with the data stream. */
1838 void io_start_multiplex_in(void)
1840 io_flush(NORMAL_FLUSH);
1841 io_start_buffering_in(sock_f_in);
1842 io_multiplexing_in = 1;
1845 /** Write an message to the multiplexed data stream. */
1846 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1848 if (!io_multiplexing_out)
1850 io_flush(NORMAL_FLUSH);
1851 stats.total_written += (len+4);
1852 mplex_write(sock_f_out, code, buf, len, convert);
1856 void io_end_multiplex_in(void)
1858 io_multiplexing_in = 0;
1859 io_end_buffering_in();
1862 /** Stop output multiplexing. */
1863 void io_end_multiplex_out(void)
1865 io_multiplexing_out = 0;
1866 io_end_buffering_out();
1869 void start_write_batch(int fd)
1871 /* Some communication has already taken place, but we don't
1872 * enable batch writing until here so that we can write a
1873 * canonical record of the communication even though the
1874 * actual communication so far depends on whether a daemon
1876 write_int(batch_fd, protocol_version);
1877 if (protocol_version >= 30)
1878 write_byte(batch_fd, inc_recurse);
1879 write_int(batch_fd, checksum_seed);
1882 write_batch_monitor_out = fd;
1884 write_batch_monitor_in = fd;
1887 void stop_write_batch(void)
1889 write_batch_monitor_out = -1;
1890 write_batch_monitor_in = -1;