1 /* -*- c-file-style: "linux" -*-
3 * Copyright (C) 1996-2001 by Andrew Tridgell
4 * Copyright (C) Paul Mackerras 1996
5 * Copyright (C) 2001, 2002 by Martin Pool <mbp@samba.org>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 * Socket and pipe I/O utilities used in rsync.
27 * rsync provides its own multiplexing system, which is used to send
28 * stderr and stdout over a single socket. We need this because
29 * stdout normally carries the binary data stream, and stderr all our
32 * For historical reasons this is off during the start of the
33 * connection, but it's switched on quite early using
34 * io_start_multiplex_out() and io_start_multiplex_in().
39 /** If no timeout is specified then use a 60 second select timeout */
40 #define SELECT_TIMEOUT 60
42 static int io_multiplexing_out;
43 static int io_multiplexing_in;
44 static int multiplex_in_fd = -1;
45 static int multiplex_out_fd = -1;
46 static time_t last_io;
50 extern size_t bwlimit_writemax;
52 extern int io_timeout;
57 extern char *remote_filesfrom_file;
58 extern struct stats stats;
60 const char phase_unknown[] = "unknown";
61 int select_timeout = SELECT_TIMEOUT;
64 * The connection might be dropped at some point; perhaps because the
65 * remote instance crashed. Just giving the offset on the stream is
66 * not very helpful. So instead we try to make io_phase_name point to
69 * For buffered/multiplexed I/O these names will be somewhat
70 * approximate; perhaps for ease of support we would rather make the
71 * buffer always flush when a single application-level I/O finishes.
73 * @todo Perhaps we want some simple stack functionality, but there's
74 * no need to overdo it.
76 const char *io_write_phase = phase_unknown;
77 const char *io_read_phase = phase_unknown;
79 /** Ignore EOF errors while reading a module listing if the remote
80 version is 24 or less. */
81 int kludge_around_eof = False;
86 static int io_filesfrom_f_in = -1;
87 static int io_filesfrom_f_out = -1;
88 static char io_filesfrom_buf[2048];
89 static char *io_filesfrom_bp;
90 static char io_filesfrom_lastchar;
91 static int io_filesfrom_buflen;
93 static void read_loop(int fd, char *buf, size_t len);
96 struct redo_list *next;
100 static struct redo_list *redo_list_head;
101 static struct redo_list *redo_list_tail;
104 struct msg_list *next;
109 static struct msg_list *msg_list_head;
110 static struct msg_list *msg_list_tail;
112 static void redo_list_add(int num)
114 struct redo_list *rl;
116 if (!(rl = new(struct redo_list)))
117 exit_cleanup(RERR_MALLOC);
121 redo_list_tail->next = rl;
127 static void check_timeout(void)
135 last_io = time(NULL);
141 if (last_io && io_timeout && (t-last_io) >= io_timeout) {
142 if (!am_server && !am_daemon) {
143 rprintf(FERROR, "io timeout after %d seconds - exiting\n",
146 exit_cleanup(RERR_TIMEOUT);
150 /** Setup the fd used to receive MSG_* messages. Only needed when
151 * we're the generator because the sender and receiver both use the
152 * multiplexed I/O setup. */
153 void set_msg_fd_in(int fd)
158 /** Setup the fd used to send our MSG_* messages. Only needed when
159 * we're the receiver because the generator and the sender both use
160 * the multiplexed I/O setup. */
161 void set_msg_fd_out(int fd)
164 set_nonblocking(msg_fd_out);
167 /* Add a message to the pending MSG_* list. */
168 static void msg_list_add(int code, char *buf, int len)
172 if (!(ml = new(struct msg_list)))
173 exit_cleanup(RERR_MALLOC);
175 if (!(ml->buf = new_array(char, len+4)))
176 exit_cleanup(RERR_MALLOC);
177 SIVAL(ml->buf, 0, ((code+MPLEX_BASE)<<24) | len);
178 memcpy(ml->buf+4, buf, len);
181 msg_list_tail->next = ml;
187 void send_msg(enum msgcode code, char *buf, int len)
189 msg_list_add(code, buf, len);
190 msg_list_push(NORMAL_FLUSH);
193 /** Read a message from the MSG_* fd and dispatch it. This is only
194 * called by the generator. */
195 static void read_msg_fd(void)
202 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
203 * to this routine from read_timeout() and writefd_unbuffered(). */
206 read_loop(fd, buf, 4);
209 len = tag & 0xFFFFFF;
210 tag = (tag >> 24) - MPLEX_BASE;
215 rprintf(FERROR, "invalid message %d:%d\n", tag, len);
216 exit_cleanup(RERR_STREAMIO);
222 rprintf(FERROR, "invalid message %d:%d\n", tag, len);
223 exit_cleanup(RERR_STREAMIO);
225 read_loop(fd, buf, 4);
226 redo_list_add(IVAL(buf,0));
235 read_loop(fd, buf, n);
236 rwrite((enum logcode)tag, buf, n);
241 rprintf(FERROR, "unknown message %d:%d\n", tag, len);
242 exit_cleanup(RERR_STREAMIO);
248 /* Try to push messages off the list onto the wire. If we leave with more
249 * to do, return 0. On error, return -1. If everything flushed, return 1.
250 * This is only active in the receiver. */
251 int msg_list_push(int flush_it_all)
253 static int written = 0;
260 while (msg_list_head) {
261 struct msg_list *ml = msg_list_head;
262 int n = write(msg_fd_out, ml->buf + written, ml->len - written);
266 if (errno != EWOULDBLOCK && errno != EAGAIN)
271 FD_SET(msg_fd_out, &fds);
272 tv.tv_sec = select_timeout;
274 if (!select(msg_fd_out+1, NULL, &fds, NULL, &tv))
276 } else if ((written += n) == ml->len) {
278 msg_list_head = ml->next;
280 msg_list_tail = NULL;
288 int get_redo_num(void)
290 struct redo_list *next;
293 while (!redo_list_head)
296 num = redo_list_head->num;
297 next = redo_list_head->next;
298 free(redo_list_head);
299 redo_list_head = next;
301 redo_list_tail = NULL;
307 * When we're the receiver and we have a local --files-from list of names
308 * that needs to be sent over the socket to the sender, we have to do two
309 * things at the same time: send the sender a list of what files we're
310 * processing and read the incoming file+info list from the sender. We do
311 * this by augmenting the read_timeout() function to copy this data. It
312 * uses the io_filesfrom_buf to read a block of data from f_in (when it is
313 * ready, since it might be a pipe) and then blast it out f_out (when it
314 * is ready to receive more data).
316 void io_set_filesfrom_fds(int f_in, int f_out)
318 io_filesfrom_f_in = f_in;
319 io_filesfrom_f_out = f_out;
320 io_filesfrom_bp = io_filesfrom_buf;
321 io_filesfrom_lastchar = '\0';
322 io_filesfrom_buflen = 0;
326 * It's almost always an error to get an EOF when we're trying to read
327 * from the network, because the protocol is self-terminating.
329 * However, there is one unfortunate cases where it is not, which is
330 * rsync <2.4.6 sending a list of modules on a server, since the list
331 * is terminated by closing the socket. So, for the section of the
332 * program where that is a problem (start_socket_client),
333 * kludge_around_eof is True and we just exit.
335 static void whine_about_eof(void)
337 if (kludge_around_eof)
340 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
341 "(%.0f bytes read so far)\n",
342 (double)stats.total_read);
344 exit_cleanup(RERR_STREAMIO);
348 static void die_from_readerr(int err)
350 /* this prevents us trying to write errors on a dead socket */
351 io_multiplexing_close();
353 rsyserr(FERROR, err, "read error");
354 exit_cleanup(RERR_STREAMIO);
359 * Read from a socket with I/O timeout. return the number of bytes
360 * read. If no bytes can be read then exit, never return a number <= 0.
362 * TODO: If the remote shell connection fails, then current versions
363 * actually report an "unexpected EOF" error here. Since it's a
364 * fairly common mistake to try to use rsh when ssh is required, we
365 * should trap that: if we fail to read any data at all, we should
366 * give a better explanation. We can tell whether the connection has
367 * started by looking e.g. at whether the remote version is known yet.
369 static int read_timeout(int fd, char *buf, size_t len)
373 io_flush(NORMAL_FLUSH);
376 /* until we manage to read *something* */
385 if (msg_fd_in >= 0) {
386 FD_SET(msg_fd_in, &r_fds);
387 if (msg_fd_in > maxfd)
389 } else if (msg_list_head) {
390 FD_SET(msg_fd_out, &w_fds);
391 if (msg_fd_out > maxfd)
394 if (io_filesfrom_f_out >= 0) {
396 if (io_filesfrom_buflen == 0) {
397 if (io_filesfrom_f_in >= 0) {
398 FD_SET(io_filesfrom_f_in, &r_fds);
399 new_fd = io_filesfrom_f_in;
401 io_filesfrom_f_out = -1;
405 FD_SET(io_filesfrom_f_out, &w_fds);
406 new_fd = io_filesfrom_f_out;
412 tv.tv_sec = select_timeout;
417 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
421 exit_cleanup(RERR_SOCKETIO);
426 if (msg_fd_in >= 0 && FD_ISSET(msg_fd_in, &r_fds))
428 else if (msg_list_head && FD_ISSET(msg_fd_out, &w_fds))
429 msg_list_push(NORMAL_FLUSH);
431 if (io_filesfrom_f_out >= 0) {
432 if (io_filesfrom_buflen) {
433 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
434 int l = write(io_filesfrom_f_out,
436 io_filesfrom_buflen);
438 if (!(io_filesfrom_buflen -= l))
439 io_filesfrom_bp = io_filesfrom_buf;
441 io_filesfrom_bp += l;
443 /* XXX should we complain? */
444 io_filesfrom_f_out = -1;
447 } else if (io_filesfrom_f_in >= 0) {
448 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
449 int l = read(io_filesfrom_f_in,
451 sizeof io_filesfrom_buf);
453 /* Send end-of-file marker */
454 io_filesfrom_buf[0] = '\0';
455 io_filesfrom_buf[1] = '\0';
456 io_filesfrom_buflen = io_filesfrom_lastchar? 2 : 1;
457 io_filesfrom_f_in = -1;
460 char *s = io_filesfrom_buf + l;
461 /* Transform CR and/or LF into '\0' */
462 while (s-- > io_filesfrom_buf) {
463 if (*s == '\n' || *s == '\r')
467 if (!io_filesfrom_lastchar) {
468 /* Last buf ended with a '\0', so don't
469 * let this buf start with one. */
470 while (l && !*io_filesfrom_bp)
471 io_filesfrom_bp++, l--;
474 io_filesfrom_bp = io_filesfrom_buf;
476 char *f = io_filesfrom_bp;
479 /* Eliminate any multi-'\0' runs. */
481 if (!(*t++ = *f++)) {
482 while (f != eob && !*f)
486 io_filesfrom_lastchar = f[-1];
488 io_filesfrom_buflen = l;
494 if (!FD_ISSET(fd, &r_fds))
497 n = read(fd, buf, len);
501 whine_about_eof(); /* Doesn't return. */
502 if (errno == EINTR || errno == EWOULDBLOCK
505 die_from_readerr(errno); /* Doesn't return. */
512 last_io = time(NULL);
519 * Read a line into the "fname" buffer (which must be at least MAXPATHLEN
522 int read_filesfrom_line(int fd, char *fname)
524 char ch, *s, *eob = fname + MAXPATHLEN - 1;
526 int reading_remotely = remote_filesfrom_file != NULL;
527 int nulls = eol_nulls || reading_remotely;
532 cnt = read(fd, &ch, 1);
533 if (cnt < 0 && (errno == EWOULDBLOCK
534 || errno == EINTR || errno == EAGAIN)) {
539 tv.tv_sec = select_timeout;
541 if (!select(fd+1, &fds, NULL, NULL, &tv))
547 if (nulls? !ch : (ch == '\r' || ch == '\n')) {
548 /* Skip empty lines if reading locally. */
549 if (!reading_remotely && s == fname)
559 if (*fname == '#' || *fname == ';')
567 * Continue trying to read len bytes - don't return until len has been
570 static void read_loop(int fd, char *buf, size_t len)
573 int n = read_timeout(fd, buf, len);
582 * Read from the file descriptor handling multiplexing - return number
585 * Never returns <= 0.
587 static int readfd_unbuffered(int fd, char *buf, size_t len)
589 static size_t remaining;
593 static size_t bufferIdx = 0;
594 static size_t bufferSz;
596 if (fd != multiplex_in_fd)
597 return read_timeout(fd, buf, len);
599 if (!io_multiplexing_in && remaining == 0) {
601 bufferSz = 2 * IO_BUFFER_SIZE;
602 buffer = new_array(char, bufferSz);
604 out_of_memory("readfd_unbuffered");
606 remaining = read_timeout(fd, buffer, bufferSz);
612 len = MIN(len, remaining);
613 memcpy(buf, buffer + bufferIdx, len);
620 read_loop(fd, line, 4);
623 remaining = tag & 0xFFFFFF;
624 tag = (tag >> 24) - MPLEX_BASE;
628 if (!buffer || remaining > bufferSz) {
629 buffer = realloc_array(buffer, char, remaining);
631 out_of_memory("readfd_unbuffered");
632 bufferSz = remaining;
634 read_loop(fd, buffer, remaining);
639 if (remaining >= sizeof line) {
640 rprintf(FERROR, "multiplexing overflow %d:%ld\n\n",
641 tag, (long)remaining);
642 exit_cleanup(RERR_STREAMIO);
644 read_loop(fd, line, remaining);
645 rwrite((enum logcode)tag, line, remaining);
649 rprintf(FERROR, "unexpected tag %d\n", tag);
650 exit_cleanup(RERR_STREAMIO);
655 io_flush(NORMAL_FLUSH);
663 * Do a buffered read from @p fd. Don't return until all @p n bytes
664 * have been read. If all @p n can't be read then exit with an
667 static void readfd(int fd, char *buffer, size_t N)
673 ret = readfd_unbuffered(fd, buffer + total, N-total);
677 stats.total_read += total;
681 int32 read_int(int f)
688 if (ret == (int32)0xffffffff)
693 int64 read_longint(int f)
699 if ((int32)ret != (int32)0xffffffff)
703 rprintf(FERROR,"Integer overflow - attempted 64 bit offset\n");
704 exit_cleanup(RERR_UNSUPPORTED);
707 ret = IVAL(b,0) | (((int64)IVAL(b,4))<<32);
713 void read_buf(int f,char *buf,size_t len)
718 void read_sbuf(int f,char *buf,size_t len)
724 unsigned char read_byte(int f)
727 read_buf(f, (char *)&c, 1);
733 * Sleep after writing to limit I/O bandwidth usage.
735 * @todo Rather than sleeping after each write, it might be better to
736 * use some kind of averaging. The current algorithm seems to always
737 * use a bit less bandwidth than specified, because it doesn't make up
738 * for slow periods. But arguably this is a feature. In addition, we
739 * ought to take the time used to write the data into account.
741 * During some phases of big transfers (file FOO is uptodate) this is
742 * called with a small bytes_written every time. As the kernel has to
743 * round small waits up to guarantee that we actually wait at least the
744 * requested number of microseconds, this can become grossly inaccurate.
745 * We therefore keep track of the bytes we've written over time and only
746 * sleep when the accumulated delay is at least 1 tenth of a second.
748 static void sleep_for_bwlimit(int bytes_written)
750 static struct timeval prior_tv;
751 static long total_written = 0;
752 struct timeval tv, start_tv;
753 long elapsed_usec, sleep_usec;
755 #define ONE_SEC 1000000L /* # of microseconds in a second */
760 total_written += bytes_written;
762 gettimeofday(&start_tv, NULL);
763 if (prior_tv.tv_sec) {
764 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
765 + (start_tv.tv_usec - prior_tv.tv_usec);
766 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
767 if (total_written < 0)
771 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
772 if (sleep_usec < ONE_SEC / 10) {
777 tv.tv_sec = sleep_usec / ONE_SEC;
778 tv.tv_usec = sleep_usec % ONE_SEC;
779 select(0, NULL, NULL, NULL, &tv);
781 gettimeofday(&prior_tv, NULL);
782 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
783 + (prior_tv.tv_usec - start_tv.tv_usec);
784 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
789 * Write len bytes to the file descriptor @p fd.
791 * This function underlies the multiplexing system. The body of the
792 * application never calls this function directly.
794 static void writefd_unbuffered(int fd,char *buf,size_t len)
798 int maxfd, count, ret;
801 if (fd == msg_fd_out) {
802 rprintf(FERROR, "Internal error: wrong write used in receiver.\n");
803 exit_cleanup(RERR_PROTOCOL);
808 while (total < len) {
813 if (msg_fd_in >= 0) {
815 FD_SET(msg_fd_in,&r_fds);
816 if (msg_fd_in > maxfd)
820 tv.tv_sec = select_timeout;
824 count = select(maxfd + 1, msg_fd_in >= 0 ? &r_fds : NULL,
828 if (count < 0 && errno == EBADF)
829 exit_cleanup(RERR_SOCKETIO);
834 if (msg_fd_in >= 0 && FD_ISSET(msg_fd_in, &r_fds))
837 if (!FD_ISSET(fd, &w_fds))
841 if (bwlimit && n > bwlimit_writemax)
842 n = bwlimit_writemax;
843 ret = write(fd, buf + total, n);
849 if (errno == EWOULDBLOCK || errno == EAGAIN) {
855 /* Don't try to write errors back across the stream. */
856 io_multiplexing_close();
857 rsyserr(FERROR, errno,
858 "writefd_unbuffered failed to write %ld bytes: phase \"%s\"",
859 (long)len, io_write_phase);
860 exit_cleanup(RERR_STREAMIO);
863 sleep_for_bwlimit(ret);
868 last_io = time(NULL);
875 static char *io_buffer;
876 static int io_buffer_count;
878 void io_start_buffering_out(int fd)
882 multiplex_out_fd = fd;
883 io_buffer = new_array(char, IO_BUFFER_SIZE);
885 out_of_memory("writefd");
889 void io_start_buffering_in(int fd)
891 multiplex_in_fd = fd;
895 * Write an message to a multiplexed stream. If this fails then rsync
898 static void mplex_write(int fd, enum msgcode code, char *buf, size_t len)
903 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
905 if (n > sizeof buffer - 4)
906 n = sizeof buffer - 4;
908 memcpy(&buffer[4], buf, n);
909 writefd_unbuffered(fd, buffer, n+4);
915 writefd_unbuffered(fd, buf, len);
919 void io_flush(int flush_it_all)
921 int fd = multiplex_out_fd;
923 msg_list_push(flush_it_all);
925 if (!io_buffer_count || no_flush)
928 if (io_multiplexing_out)
929 mplex_write(fd, MSG_DATA, io_buffer, io_buffer_count);
931 writefd_unbuffered(fd, io_buffer, io_buffer_count);
936 void io_end_buffering(void)
938 io_flush(NORMAL_FLUSH);
939 if (!io_multiplexing_out) {
945 static void writefd(int fd,char *buf,size_t len)
947 stats.total_written += len;
949 if (fd == msg_fd_out) {
950 rprintf(FERROR, "Internal error: wrong write used in receiver.\n");
951 exit_cleanup(RERR_PROTOCOL);
954 if (!io_buffer || fd != multiplex_out_fd) {
955 writefd_unbuffered(fd, buf, len);
960 int n = MIN((int)len, IO_BUFFER_SIZE-io_buffer_count);
962 memcpy(io_buffer+io_buffer_count, buf, n);
965 io_buffer_count += n;
968 if (io_buffer_count == IO_BUFFER_SIZE)
969 io_flush(NORMAL_FLUSH);
974 void write_int(int f,int32 x)
982 void write_int_named(int f, int32 x, const char *phase)
984 io_write_phase = phase;
986 io_write_phase = phase_unknown;
991 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
992 * 64-bit types on this platform.
994 void write_longint(int f, int64 x)
998 if (x <= 0x7FFFFFFF) {
999 write_int(f, (int)x);
1004 rprintf(FERROR,"Integer overflow - attempted 64 bit offset\n");
1005 exit_cleanup(RERR_UNSUPPORTED);
1007 write_int(f, (int32)0xFFFFFFFF);
1008 SIVAL(b,0,(x&0xFFFFFFFF));
1009 SIVAL(b,4,((x>>32)&0xFFFFFFFF));
1015 void write_buf(int f,char *buf,size_t len)
1020 /** Write a string to the connection */
1021 static void write_sbuf(int f,char *buf)
1023 write_buf(f, buf, strlen(buf));
1027 void write_byte(int f,unsigned char c)
1029 write_buf(f,(char *)&c,1);
1035 * Read a line of up to @p maxlen characters into @p buf (not counting
1036 * the trailing null). Strips the (required) trailing newline and all
1039 * @return 1 for success; 0 for I/O error or truncation.
1041 int read_line(int f, char *buf, size_t maxlen)
1045 read_buf(f, buf, 1);
1050 if (buf[0] != '\r') {
1060 void io_printf(int fd, const char *format, ...)
1066 va_start(ap, format);
1067 len = vsnprintf(buf, sizeof buf, format, ap);
1071 exit_cleanup(RERR_STREAMIO);
1073 write_sbuf(fd, buf);
1077 /** Setup for multiplexing a MSG_* stream with the data stream. */
1078 void io_start_multiplex_out(int fd)
1080 multiplex_out_fd = fd;
1081 io_flush(NORMAL_FLUSH);
1082 io_start_buffering_out(fd);
1083 io_multiplexing_out = 1;
1086 /** Setup for multiplexing a MSG_* stream with the data stream. */
1087 void io_start_multiplex_in(int fd)
1089 multiplex_in_fd = fd;
1090 io_flush(NORMAL_FLUSH);
1091 io_multiplexing_in = 1;
1094 /** Write an message to the multiplexed data stream. */
1095 int io_multiplex_write(enum msgcode code, char *buf, size_t len)
1097 if (!io_multiplexing_out)
1100 io_flush(NORMAL_FLUSH);
1101 stats.total_written += (len+4);
1102 mplex_write(multiplex_out_fd, code, buf, len);
1106 /** Stop output multiplexing. */
1107 void io_multiplexing_close(void)
1109 io_multiplexing_out = 0;