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;
39 extern int allowed_lull;
43 extern int am_generator;
44 extern int inc_recurse;
49 extern int read_batch;
50 extern int csum_length;
51 extern int protect_args;
52 extern int checksum_seed;
53 extern int protocol_version;
54 extern int remove_source_files;
55 extern int preserve_hard_links;
56 extern struct stats stats;
57 extern struct file_list *cur_flist;
59 extern int filesfrom_convert;
60 extern iconv_t ic_send, ic_recv;
63 const char phase_unknown[] = "unknown";
64 int ignore_timeout = 0;
67 int check_for_io_err = 0;
69 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
70 int kluge_around_eof = 0;
77 static int iobuf_f_in = -1;
78 static char *iobuf_in;
79 static size_t iobuf_in_siz;
80 static size_t iobuf_in_ndx;
81 static size_t iobuf_in_remaining;
83 static int iobuf_f_out = -1;
84 static char *iobuf_out;
85 static int iobuf_out_cnt;
87 int flist_forward_from = -1;
89 static int io_multiplexing_out;
90 static int io_multiplexing_in;
91 static time_t last_io_in;
92 static time_t last_io_out;
95 static int write_batch_monitor_in = -1;
96 static int write_batch_monitor_out = -1;
98 static int io_filesfrom_f_in = -1;
99 static int io_filesfrom_f_out = -1;
100 static xbuf ff_buf = EMPTY_XBUF;
101 static char ff_lastchar;
103 static xbuf iconv_buf = EMPTY_XBUF;
105 static int defer_forwarding_messages = 0, defer_forwarding_keep = 0;
106 static int select_timeout = SELECT_TIMEOUT;
107 static int active_filecnt = 0;
108 static OFF_T active_bytecnt = 0;
109 static int first_message = 1;
111 static char int_byte_extra[64] = {
112 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
113 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
114 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
115 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
118 #define REMOTE_OPTION_ERROR "rsync: on remote machine: -"
119 #define REMOTE_OPTION_ERROR2 ": unknown option"
121 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
123 static void readfd(int fd, char *buffer, size_t N);
124 static void writefd(int fd, const char *buf, size_t len);
125 static void writefd_unbuffered(int fd, const char *buf, size_t len);
126 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
128 struct flist_ndx_item {
129 struct flist_ndx_item *next;
133 struct flist_ndx_list {
134 struct flist_ndx_item *head, *tail;
137 static struct flist_ndx_list redo_list, hlink_list;
139 struct msg_list_item {
140 struct msg_list_item *next;
146 struct msg_list_item *head, *tail;
149 static struct msg_list msg_queue;
151 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
153 struct flist_ndx_item *item;
155 if (!(item = new(struct flist_ndx_item)))
156 out_of_memory("flist_ndx_push");
160 lp->tail->next = item;
166 static int flist_ndx_pop(struct flist_ndx_list *lp)
168 struct flist_ndx_item *next;
175 next = lp->head->next;
184 static void got_flist_entry_status(enum festatus status, const char *buf)
186 int ndx = IVAL(buf, 0);
187 struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
189 if (remove_source_files) {
191 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
195 flist->in_progress--;
199 if (remove_source_files)
200 send_msg(MSG_SUCCESS, buf, 4, 0);
201 if (preserve_hard_links) {
202 struct file_struct *file = flist->files[ndx - flist->ndx_start];
203 if (F_IS_HLINKED(file)) {
204 flist_ndx_push(&hlink_list, ndx);
205 flist->in_progress++;
212 flist_ndx_push(&redo_list, ndx);
219 static void check_timeout(void)
223 if (!io_timeout || ignore_timeout)
227 last_io_in = time(NULL);
233 if (t - last_io_in >= io_timeout) {
234 if (!am_server && !am_daemon) {
235 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
236 (int)(t-last_io_in));
238 exit_cleanup(RERR_TIMEOUT);
242 /* Note the fds used for the main socket (which might really be a pipe
243 * for a local transfer, but we can ignore that). */
244 void io_set_sock_fds(int f_in, int f_out)
250 void set_io_timeout(int secs)
254 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
255 select_timeout = SELECT_TIMEOUT;
257 select_timeout = io_timeout;
259 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
262 /* Setup the fd used to receive MSG_* messages. Only needed during the
263 * early stages of being a local sender (up through the sending of the
264 * file list) or when we're the generator (to fetch the messages from
266 void set_msg_fd_in(int fd)
271 /* Setup the fd used to send our MSG_* messages. Only needed when
272 * we're the receiver (to send our messages to the generator). */
273 void set_msg_fd_out(int fd)
276 set_nonblocking(msg_fd_out);
279 /* Add a message to the pending MSG_* list. */
280 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
282 struct msg_list_item *m;
283 int sz = len + 4 + sizeof m[0] - 1;
285 if (!(m = (struct msg_list_item *)new_array(char, sz)))
286 out_of_memory("msg_list_add");
288 m->convert = convert;
289 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
290 memcpy(m->buf + 4, buf, len);
298 static void msg_flush(void)
301 while (msg_queue.head && io_multiplexing_out) {
302 struct msg_list_item *m = msg_queue.head;
303 int len = IVAL(m->buf, 0) & 0xFFFFFF;
304 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
305 if (!(msg_queue.head = m->next))
306 msg_queue.tail = NULL;
307 stats.total_written += len + 4;
308 defer_forwarding_messages++;
309 mplex_write(sock_f_out, tag, m->buf + 4, len, m->convert);
310 defer_forwarding_messages--;
314 while (msg_queue.head) {
315 struct msg_list_item *m = msg_queue.head;
316 int len = IVAL(m->buf, 0) & 0xFFFFFF;
317 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
318 if (!(msg_queue.head = m->next))
319 msg_queue.tail = NULL;
320 defer_forwarding_messages++;
321 mplex_write(msg_fd_out, tag, m->buf + 4, len, m->convert);
322 defer_forwarding_messages--;
328 static void check_for_d_option_error(const char *msg)
330 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
335 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
338 msg += sizeof REMOTE_OPTION_ERROR - 1;
339 if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
340 || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
343 for ( ; *msg != ':'; msg++) {
346 else if (*msg == 'e')
348 else if (strchr(rsync263_opts, *msg) == NULL)
354 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
358 /* Read a message from the MSG_* fd and handle it. This is called either
359 * during the early stages of being a local sender (up through the sending
360 * of the file list) or when we're the generator (to fetch the messages
361 * from the receiver). */
362 static void read_msg_fd(void)
366 struct file_list *flist;
370 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
371 * to this routine from writefd_unbuffered(). */
374 defer_forwarding_messages++;
379 len = tag & 0xFFFFFF;
380 tag = (tag >> 24) - MPLEX_BASE;
382 check_for_io_err = 0;
386 if (len < 0 || len > 1 || !am_generator) {
388 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
389 tag, len, who_am_i(),
390 inc_recurse ? "/inc" : "");
391 exit_cleanup(RERR_STREAMIO);
394 readfd(fd, buf, len);
395 stats.total_read = read_varlong(fd, 3);
400 if (len != 4 || !am_generator)
403 got_flist_entry_status(FES_REDO, buf);
406 if (len != 4 || !am_generator || !inc_recurse)
409 /* Read extra file list from receiver. */
410 assert(iobuf_in != NULL);
411 assert(iobuf_f_in == fd);
413 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
414 who_am_i(), IVAL(buf,0));
416 flist = recv_file_list(fd);
417 flist->parent_ndx = IVAL(buf,0);
418 /* If the sender is going to send us an MSG_IO_ERROR value, it
419 * will always be the very next message following MSG_FLIST. */
420 check_for_io_err = 1;
421 #ifdef SUPPORT_HARD_LINKS
422 if (preserve_hard_links)
423 match_hard_links(flist);
427 if (len != 0 || !am_generator || !inc_recurse)
434 readfd(fd, buf, len);
435 io_error |= IVAL(buf, 0);
438 if (len >= (int)sizeof buf || !am_generator)
440 readfd(fd, buf, len);
441 send_msg(MSG_DELETED, buf, len, 1);
444 if (len != 4 || !am_generator)
447 got_flist_entry_status(FES_SUCCESS, buf);
450 if (len != 4 || !am_generator)
453 got_flist_entry_status(FES_NO_SEND, buf);
455 case MSG_ERROR_SOCKET:
460 if (tag == MSG_ERROR_SOCKET)
461 io_end_multiplex_out();
473 rwrite((enum logcode)tag, buf, n, !am_generator);
478 rprintf(FERROR, "unknown message %d:%d [%s]\n",
479 tag, len, who_am_i());
480 exit_cleanup(RERR_STREAMIO);
485 if (!--defer_forwarding_messages && !no_flush)
489 /* This is used by the generator to limit how many file transfers can
490 * be active at once when --remove-source-files is specified. Without
491 * this, sender-side deletions were mostly happening at the end. */
492 void increment_active_files(int ndx, int itemizing, enum logcode code)
494 /* TODO: tune these limits? */
495 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
496 check_for_finished_files(itemizing, code, 0);
498 io_flush(NORMAL_FLUSH);
504 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
507 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
508 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
510 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
514 /* We need to convert buf before doing anything else so that we
515 * can include the (converted) byte length in the message header. */
516 if (convert && ic_send != (iconv_t)-1) {
519 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
520 INIT_XBUF(inbuf, (char*)buf, len, -1);
522 iconvbufs(ic_send, &inbuf, &outbuf,
523 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
525 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
526 exit_cleanup(RERR_UNSUPPORTED);
529 n = len = outbuf.len;
532 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
533 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
535 memcpy(buffer + 4, buf, n);
537 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
539 defer_forwarding_keep = 1; /* defer_forwarding_messages++ on return */
540 writefd_unbuffered(fd, buffer, n+4);
541 defer_forwarding_keep = 0;
544 writefd_unbuffered(fd, buf+n, len-n);
546 if (!--defer_forwarding_messages && !no_flush)
550 int send_msg(enum msgcode code, const char *buf, int len, int convert)
552 if (msg_fd_out < 0) {
553 if (!defer_forwarding_messages)
554 return io_multiplex_write(code, buf, len, convert);
555 if (!io_multiplexing_out)
557 msg_list_add(&msg_queue, code, buf, len, convert);
560 if (flist_forward_from >= 0)
561 msg_list_add(&msg_queue, code, buf, len, convert);
563 mplex_write(msg_fd_out, code, buf, len, convert);
567 void send_msg_int(enum msgcode code, int num)
570 SIVAL(numbuf, 0, num);
571 send_msg(code, numbuf, 4, 0);
574 void wait_for_receiver(void)
577 io_flush(NORMAL_FLUSH);
582 int get_redo_num(void)
584 return flist_ndx_pop(&redo_list);
587 int get_hlink_num(void)
589 return flist_ndx_pop(&hlink_list);
593 * When we're the receiver and we have a local --files-from list of names
594 * that needs to be sent over the socket to the sender, we have to do two
595 * things at the same time: send the sender a list of what files we're
596 * processing and read the incoming file+info list from the sender. We do
597 * this by augmenting the read_timeout() function to copy this data. It
598 * uses ff_buf to read a block of data from f_in (when it is ready, since
599 * it might be a pipe) and then blast it out f_out (when it is ready to
600 * receive more data).
602 void io_set_filesfrom_fds(int f_in, int f_out)
604 io_filesfrom_f_in = f_in;
605 io_filesfrom_f_out = f_out;
606 alloc_xbuf(&ff_buf, 2048);
609 alloc_xbuf(&iconv_buf, 1024);
613 /* It's almost always an error to get an EOF when we're trying to read from the
614 * network, because the protocol is (for the most part) self-terminating.
616 * There is one case for the receiver when it is at the end of the transfer
617 * (hanging around reading any keep-alive packets that might come its way): if
618 * the sender dies before the generator's kill-signal comes through, we can end
619 * up here needing to loop until the kill-signal arrives. In this situation,
620 * kluge_around_eof will be < 0.
622 * There is another case for older protocol versions (< 24) where the module
623 * listing was not terminated, so we must ignore an EOF error in that case and
624 * exit. In this situation, kluge_around_eof will be > 0. */
625 static void whine_about_eof(int fd)
627 if (kluge_around_eof && fd == sock_f_in) {
629 if (kluge_around_eof > 0)
631 /* If we're still here after 10 seconds, exit with an error. */
632 for (i = 10*1000/20; i--; )
636 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
637 "(%.0f bytes received so far) [%s]\n",
638 (double)stats.total_read, who_am_i());
640 exit_cleanup(RERR_STREAMIO);
644 * Read from a socket with I/O timeout. return the number of bytes
645 * read. If no bytes can be read then exit, never return a number <= 0.
647 * TODO: If the remote shell connection fails, then current versions
648 * actually report an "unexpected EOF" error here. Since it's a
649 * fairly common mistake to try to use rsh when ssh is required, we
650 * should trap that: if we fail to read any data at all, we should
651 * give a better explanation. We can tell whether the connection has
652 * started by looking e.g. at whether the remote version is known yet.
654 static int read_timeout(int fd, char *buf, size_t len)
658 io_flush(FULL_FLUSH);
661 /* until we manage to read *something* */
670 if (io_filesfrom_f_out >= 0) {
672 if (ff_buf.len == 0) {
673 if (io_filesfrom_f_in >= 0) {
674 FD_SET(io_filesfrom_f_in, &r_fds);
675 new_fd = io_filesfrom_f_in;
677 io_filesfrom_f_out = -1;
681 FD_SET(io_filesfrom_f_out, &w_fds);
682 new_fd = io_filesfrom_f_out;
688 tv.tv_sec = select_timeout;
693 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
696 if (errno == EBADF) {
697 defer_forwarding_messages = 0;
698 exit_cleanup(RERR_SOCKETIO);
704 if (io_filesfrom_f_out >= 0) {
706 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
707 int l = write(io_filesfrom_f_out,
708 ff_buf.buf + ff_buf.pos,
711 if (!(ff_buf.len -= l))
715 } else if (errno != EINTR) {
716 /* XXX should we complain? */
717 io_filesfrom_f_out = -1;
720 } else if (io_filesfrom_f_in >= 0) {
721 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
723 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
725 xbuf *ibuf = &ff_buf;
727 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
729 if (l == 0 || errno != EINTR) {
730 /* Send end-of-file marker */
731 memcpy(ff_buf.buf, "\0\0", 2);
732 ff_buf.len = ff_lastchar? 2 : 1;
734 io_filesfrom_f_in = -1;
738 if (filesfrom_convert) {
741 iconvbufs(ic_send, &iconv_buf, &ff_buf,
742 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
747 char *s = ff_buf.buf + l;
748 /* Transform CR and/or LF into '\0' */
749 while (s-- > ff_buf.buf) {
750 if (*s == '\n' || *s == '\r')
755 /* Last buf ended with a '\0', so don't
756 * let this buf start with one. */
757 while (l && ff_buf.buf[ff_buf.pos] == '\0')
763 char *f = ff_buf.buf + ff_buf.pos;
766 /* Eliminate any multi-'\0' runs. */
768 if (!(*t++ = *f++)) {
769 while (f != eob && !*f)
781 if (!FD_ISSET(fd, &r_fds))
784 n = read(fd, buf, len);
788 whine_about_eof(fd); /* Doesn't return. */
789 if (errno == EINTR || errno == EWOULDBLOCK
793 /* Don't write errors on a dead socket. */
794 if (fd == sock_f_in) {
795 io_end_multiplex_out();
796 rsyserr(FERROR_SOCKET, errno, "read error");
798 rsyserr(FERROR, errno, "read error");
799 exit_cleanup(RERR_STREAMIO);
806 if (fd == sock_f_in && io_timeout)
807 last_io_in = time(NULL);
813 /* Read a line into the "buf" buffer. */
814 int read_line(int fd, char *buf, size_t bufsiz, int flags)
820 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
821 realloc_xbuf(&iconv_buf, bufsiz + 1024);
826 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
830 eob = s + bufsiz - 1;
832 cnt = read(fd, &ch, 1);
833 if (cnt < 0 && (errno == EWOULDBLOCK
834 || errno == EINTR || errno == EAGAIN)) {
841 tv.tv_sec = select_timeout;
843 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
845 /*if (FD_ISSET(fd, &e_fds))
846 rprintf(FINFO, "select exception on fd %d\n", fd); */
851 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
852 /* Skip empty lines if dumping comments. */
853 if (flags & RL_DUMP_COMMENTS && s == buf)
862 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
866 if (flags & RL_CONVERT) {
868 INIT_XBUF(outbuf, buf, 0, bufsiz);
870 iconv_buf.len = s - iconv_buf.buf;
871 iconvbufs(ic_recv, &iconv_buf, &outbuf,
872 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
873 outbuf.buf[outbuf.len] = '\0';
881 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
882 char ***argv_p, int *argc_p, char **request_p)
884 int maxargs = MAX_ARGS;
888 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
891 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
894 if (!(argv = new_array(char *, maxargs)))
895 out_of_memory("read_args");
896 if (mod_name && !protect_args)
897 argv[argc++] = "rsyncd";
900 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
903 if (argc == maxargs-1) {
905 if (!(argv = realloc_array(argv, char *, maxargs)))
906 out_of_memory("read_args");
911 *request_p = strdup(buf);
915 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
917 glob_expand(buf, &argv, &argc, &maxargs);
919 if (!(p = strdup(buf)))
920 out_of_memory("read_args");
922 if (*p == '.' && p[1] == '\0')
928 glob_expand(NULL, NULL, NULL, NULL);
934 int io_start_buffering_out(int f_out)
937 assert(f_out == iobuf_f_out);
940 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
941 out_of_memory("io_start_buffering_out");
947 int io_start_buffering_in(int f_in)
950 assert(f_in == iobuf_f_in);
953 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
954 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
955 out_of_memory("io_start_buffering_in");
960 void io_end_buffering_in(void)
967 iobuf_in_remaining = 0;
971 void io_end_buffering_out(void)
975 io_flush(FULL_FLUSH);
981 void maybe_flush_socket(int important)
983 if (iobuf_out && iobuf_out_cnt
984 && (important || time(NULL) - last_io_out >= 5))
985 io_flush(NORMAL_FLUSH);
988 void maybe_send_keepalive(void)
990 if (time(NULL) - last_io_out >= allowed_lull) {
991 if (!iobuf_out || !iobuf_out_cnt) {
992 if (protocol_version < 29)
993 return; /* there's nothing we can do */
994 if (protocol_version >= 30)
995 send_msg(MSG_NOOP, "", 0, 0);
997 write_int(sock_f_out, cur_flist->used);
998 write_shortint(sock_f_out, ITEM_IS_NEW);
1002 io_flush(NORMAL_FLUSH);
1006 void start_flist_forward(int f_in)
1008 assert(iobuf_out != NULL);
1009 assert(iobuf_f_out == msg_fd_out);
1010 flist_forward_from = f_in;
1013 void stop_flist_forward()
1015 flist_forward_from = -1;
1016 io_flush(FULL_FLUSH);
1020 * Continue trying to read len bytes - don't return until len has been
1023 static void read_loop(int fd, char *buf, size_t len)
1026 int n = read_timeout(fd, buf, len);
1034 * Read from the file descriptor handling multiplexing - return number
1037 * Never returns <= 0.
1039 static int readfd_unbuffered(int fd, char *buf, size_t len)
1043 char line[BIGPATHBUFLEN];
1045 if (!iobuf_in || fd != iobuf_f_in)
1046 return read_timeout(fd, buf, len);
1048 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1049 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1054 if (iobuf_in_remaining) {
1055 len = MIN(len, iobuf_in_remaining);
1056 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1057 iobuf_in_ndx += len;
1058 iobuf_in_remaining -= len;
1063 read_loop(fd, line, 4);
1064 tag = IVAL(line, 0);
1066 msg_bytes = tag & 0xFFFFFF;
1067 tag = (tag >> 24) - MPLEX_BASE;
1069 check_for_io_err = 0;
1073 if (msg_bytes > iobuf_in_siz) {
1074 if (!(iobuf_in = realloc_array(iobuf_in, char,
1076 out_of_memory("readfd_unbuffered");
1077 iobuf_in_siz = msg_bytes;
1079 read_loop(fd, iobuf_in, msg_bytes);
1080 iobuf_in_remaining = msg_bytes;
1085 maybe_send_keepalive();
1090 read_loop(fd, line, msg_bytes);
1091 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1092 io_error |= IVAL(line, 0);
1095 if (msg_bytes >= sizeof line)
1098 if (ic_recv != (iconv_t)-1) {
1104 INIT_CONST_XBUF(outbuf, line);
1105 INIT_XBUF(inbuf, ibuf, 0, -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)
1120 if (outbuf.len == outbuf.size)
1122 outbuf.buf[outbuf.len++] = '\0';
1124 msg_bytes = outbuf.len;
1127 read_loop(fd, line, msg_bytes);
1128 /* A directory name was sent with the trailing null */
1129 if (msg_bytes > 0 && !line[msg_bytes-1])
1130 log_delete(line, S_IFDIR);
1132 line[msg_bytes] = '\0';
1133 log_delete(line, S_IFREG);
1137 if (msg_bytes != 4) {
1139 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1140 tag, (long)msg_bytes, who_am_i());
1141 exit_cleanup(RERR_STREAMIO);
1143 read_loop(fd, line, msg_bytes);
1144 successful_send(IVAL(line, 0));
1149 read_loop(fd, line, msg_bytes);
1150 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1154 case MSG_ERROR_XFER:
1156 if (msg_bytes >= sizeof line) {
1159 "multiplexing overflow %d:%ld [%s]\n",
1160 tag, (long)msg_bytes, who_am_i());
1161 exit_cleanup(RERR_STREAMIO);
1163 read_loop(fd, line, msg_bytes);
1164 rwrite((enum logcode)tag, line, msg_bytes, 1);
1165 if (first_message) {
1166 if (list_only && !am_sender && tag == 1) {
1167 line[msg_bytes] = '\0';
1168 check_for_d_option_error(line);
1174 rprintf(FERROR, "unexpected tag %d [%s]\n",
1176 exit_cleanup(RERR_STREAMIO);
1180 if (iobuf_in_remaining == 0)
1181 io_flush(NORMAL_FLUSH);
1186 /* Do a buffered read from fd. Don't return until all N bytes have
1187 * been read. If all N can't be read then exit with an error. */
1188 static void readfd(int fd, char *buffer, size_t N)
1194 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1198 if (fd == write_batch_monitor_in) {
1199 if ((size_t)write(batch_fd, buffer, total) != total)
1200 exit_cleanup(RERR_FILEIO);
1203 if (fd == flist_forward_from)
1204 writefd(iobuf_f_out, buffer, total);
1206 if (fd == sock_f_in)
1207 stats.total_read += total;
1210 unsigned short read_shortint(int f)
1214 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1217 int32 read_int(int f)
1224 #if SIZEOF_INT32 > 4
1225 if (num & (int32)0x80000000)
1226 num |= ~(int32)0xffffffff;
1231 int32 read_varint(int f)
1241 readfd(f, (char*)&ch, 1);
1242 extra = int_byte_extra[ch / 4];
1244 uchar bit = ((uchar)1<<(8-extra));
1245 if (extra >= (int)sizeof u.b) {
1246 rprintf(FERROR, "Overflow in read_varint()\n");
1247 exit_cleanup(RERR_STREAMIO);
1249 readfd(f, u.b, extra);
1250 u.b[extra] = ch & (bit-1);
1253 #if CAREFUL_ALIGNMENT
1256 #if SIZEOF_INT32 > 4
1257 if (u.x & (int32)0x80000000)
1258 u.x |= ~(int32)0xffffffff;
1263 int64 read_varlong(int f, uchar min_bytes)
1272 #if SIZEOF_INT64 < 8
1277 readfd(f, b2, min_bytes);
1278 memcpy(u.b, b2+1, min_bytes-1);
1279 extra = int_byte_extra[CVAL(b2, 0) / 4];
1281 uchar bit = ((uchar)1<<(8-extra));
1282 if (min_bytes + extra > (int)sizeof u.b) {
1283 rprintf(FERROR, "Overflow in read_varlong()\n");
1284 exit_cleanup(RERR_STREAMIO);
1286 readfd(f, u.b + min_bytes - 1, extra);
1287 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1288 #if SIZEOF_INT64 < 8
1289 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1290 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1291 exit_cleanup(RERR_UNSUPPORTED);
1295 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1296 #if SIZEOF_INT64 < 8
1298 #elif CAREFUL_ALIGNMENT
1299 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1304 int64 read_longint(int f)
1306 #if SIZEOF_INT64 >= 8
1309 int32 num = read_int(f);
1311 if (num != (int32)0xffffffff)
1314 #if SIZEOF_INT64 < 8
1315 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1316 exit_cleanup(RERR_UNSUPPORTED);
1319 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1323 void read_buf(int f, char *buf, size_t len)
1328 void read_sbuf(int f, char *buf, size_t len)
1330 readfd(f, buf, len);
1334 uchar read_byte(int f)
1337 readfd(f, (char *)&c, 1);
1341 int read_vstring(int f, char *buf, int bufsize)
1343 int len = read_byte(f);
1346 len = (len & ~0x80) * 0x100 + read_byte(f);
1348 if (len >= bufsize) {
1349 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1355 readfd(f, buf, len);
1360 /* Populate a sum_struct with values from the socket. This is
1361 * called by both the sender and the receiver. */
1362 void read_sum_head(int f, struct sum_struct *sum)
1364 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1365 sum->count = read_int(f);
1366 if (sum->count < 0) {
1367 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1368 (long)sum->count, who_am_i());
1369 exit_cleanup(RERR_PROTOCOL);
1371 sum->blength = read_int(f);
1372 if (sum->blength < 0 || sum->blength > max_blength) {
1373 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1374 (long)sum->blength, who_am_i());
1375 exit_cleanup(RERR_PROTOCOL);
1377 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1378 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1379 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1380 sum->s2length, who_am_i());
1381 exit_cleanup(RERR_PROTOCOL);
1383 sum->remainder = read_int(f);
1384 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1385 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1386 (long)sum->remainder, who_am_i());
1387 exit_cleanup(RERR_PROTOCOL);
1391 /* Send the values from a sum_struct over the socket. Set sum to
1392 * NULL if there are no checksums to send. This is called by both
1393 * the generator and the sender. */
1394 void write_sum_head(int f, struct sum_struct *sum)
1396 static struct sum_struct null_sum;
1401 write_int(f, sum->count);
1402 write_int(f, sum->blength);
1403 if (protocol_version >= 27)
1404 write_int(f, sum->s2length);
1405 write_int(f, sum->remainder);
1409 * Sleep after writing to limit I/O bandwidth usage.
1411 * @todo Rather than sleeping after each write, it might be better to
1412 * use some kind of averaging. The current algorithm seems to always
1413 * use a bit less bandwidth than specified, because it doesn't make up
1414 * for slow periods. But arguably this is a feature. In addition, we
1415 * ought to take the time used to write the data into account.
1417 * During some phases of big transfers (file FOO is uptodate) this is
1418 * called with a small bytes_written every time. As the kernel has to
1419 * round small waits up to guarantee that we actually wait at least the
1420 * requested number of microseconds, this can become grossly inaccurate.
1421 * We therefore keep track of the bytes we've written over time and only
1422 * sleep when the accumulated delay is at least 1 tenth of a second.
1424 static void sleep_for_bwlimit(int bytes_written)
1426 static struct timeval prior_tv;
1427 static long total_written = 0;
1428 struct timeval tv, start_tv;
1429 long elapsed_usec, sleep_usec;
1431 #define ONE_SEC 1000000L /* # of microseconds in a second */
1433 if (!bwlimit_writemax)
1436 total_written += bytes_written;
1438 gettimeofday(&start_tv, NULL);
1439 if (prior_tv.tv_sec) {
1440 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1441 + (start_tv.tv_usec - prior_tv.tv_usec);
1442 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1443 if (total_written < 0)
1447 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1448 if (sleep_usec < ONE_SEC / 10) {
1449 prior_tv = start_tv;
1453 tv.tv_sec = sleep_usec / ONE_SEC;
1454 tv.tv_usec = sleep_usec % ONE_SEC;
1455 select(0, NULL, NULL, NULL, &tv);
1457 gettimeofday(&prior_tv, NULL);
1458 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1459 + (prior_tv.tv_usec - start_tv.tv_usec);
1460 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1463 /* Write len bytes to the file descriptor fd, looping as necessary to get
1464 * the job done and also (in certain circumstances) reading any data on
1465 * msg_fd_in to avoid deadlock.
1467 * This function underlies the multiplexing system. The body of the
1468 * application never calls this function directly. */
1469 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1471 size_t n, total = 0;
1472 fd_set w_fds, r_fds, e_fds;
1473 int maxfd, count, cnt, using_r_fds;
1478 defer_forwarding_messages++, defer_inc++;
1480 while (total < len) {
1487 if (msg_fd_in >= 0) {
1489 FD_SET(msg_fd_in, &r_fds);
1490 if (msg_fd_in > maxfd)
1496 tv.tv_sec = select_timeout;
1500 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1501 &w_fds, &e_fds, &tv);
1504 if (count < 0 && errno == EBADF)
1505 exit_cleanup(RERR_SOCKETIO);
1510 /*if (FD_ISSET(fd, &e_fds))
1511 rprintf(FINFO, "select exception on fd %d\n", fd); */
1513 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1516 if (!FD_ISSET(fd, &w_fds))
1520 if (bwlimit_writemax && n > bwlimit_writemax)
1521 n = bwlimit_writemax;
1522 cnt = write(fd, buf + total, n);
1528 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1534 /* Don't try to write errors back across the stream. */
1535 if (fd == sock_f_out)
1536 io_end_multiplex_out();
1537 /* Don't try to write errors down a failing msg pipe. */
1538 if (am_server && fd == msg_fd_out)
1539 exit_cleanup(RERR_STREAMIO);
1540 rsyserr(FERROR, errno,
1541 "writefd_unbuffered failed to write %ld bytes [%s]",
1542 (long)len, who_am_i());
1543 /* If the other side is sending us error messages, try
1544 * to grab any messages they sent before they died. */
1545 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1549 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1551 exit_cleanup(RERR_STREAMIO);
1555 defer_forwarding_messages++, defer_inc++;
1557 if (fd == sock_f_out) {
1558 if (io_timeout || am_generator)
1559 last_io_out = time(NULL);
1560 sleep_for_bwlimit(cnt);
1565 defer_inc -= defer_forwarding_keep;
1566 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1570 void io_flush(int flush_it_all)
1572 if (!iobuf_out_cnt || no_flush)
1575 if (io_multiplexing_out)
1576 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1578 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1581 if (flush_it_all && !defer_forwarding_messages)
1585 static void writefd(int fd, const char *buf, size_t len)
1587 if (fd == sock_f_out)
1588 stats.total_written += len;
1590 if (fd == write_batch_monitor_out) {
1591 if ((size_t)write(batch_fd, buf, len) != len)
1592 exit_cleanup(RERR_FILEIO);
1595 if (!iobuf_out || fd != iobuf_f_out) {
1596 writefd_unbuffered(fd, buf, len);
1601 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1603 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1609 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1610 io_flush(NORMAL_FLUSH);
1614 void write_shortint(int f, unsigned short x)
1618 b[1] = (char)(x >> 8);
1622 void write_int(int f, int32 x)
1629 void write_varint(int f, int32 x)
1637 while (cnt > 1 && b[cnt] == 0)
1639 bit = ((uchar)1<<(7-cnt+1));
1640 if (CVAL(b, cnt) >= bit) {
1644 *b = b[cnt] | ~(bit*2-1);
1651 void write_varlong(int f, int64 x, uchar min_bytes)
1658 #if SIZEOF_INT64 >= 8
1659 SIVAL(b, 5, x >> 32);
1661 if (x <= 0x7FFFFFFF && x >= 0)
1662 memset(b + 5, 0, 4);
1664 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1665 exit_cleanup(RERR_UNSUPPORTED);
1669 while (cnt > min_bytes && b[cnt] == 0)
1671 bit = ((uchar)1<<(7-cnt+min_bytes));
1672 if (CVAL(b, cnt) >= bit) {
1675 } else if (cnt > min_bytes)
1676 *b = b[cnt] | ~(bit*2-1);
1684 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1685 * 64-bit types on this platform.
1687 void write_longint(int f, int64 x)
1689 char b[12], * const s = b+4;
1692 if (x <= 0x7FFFFFFF && x >= 0) {
1697 #if SIZEOF_INT64 < 8
1698 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1699 exit_cleanup(RERR_UNSUPPORTED);
1702 SIVAL(s, 4, x >> 32);
1707 void write_buf(int f, const char *buf, size_t len)
1712 /** Write a string to the connection */
1713 void write_sbuf(int f, const char *buf)
1715 writefd(f, buf, strlen(buf));
1718 void write_byte(int f, uchar c)
1720 writefd(f, (char *)&c, 1);
1723 void write_vstring(int f, const char *str, int len)
1725 uchar lenbuf[3], *lb = lenbuf;
1730 "attempting to send over-long vstring (%d > %d)\n",
1732 exit_cleanup(RERR_PROTOCOL);
1734 *lb++ = len / 0x100 + 0x80;
1738 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1740 writefd(f, str, len);
1743 /* Send a file-list index using a byte-reduction method. */
1744 void write_ndx(int f, int32 ndx)
1746 static int32 prev_positive = -1, prev_negative = 1;
1747 int32 diff, cnt = 0;
1750 if (protocol_version < 30 || read_batch) {
1755 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1756 * negative nums as a positive after sending a leading 0xFF. */
1758 diff = ndx - prev_positive;
1759 prev_positive = ndx;
1760 } else if (ndx == NDX_DONE) {
1765 b[cnt++] = (char)0xFF;
1767 diff = ndx - prev_negative;
1768 prev_negative = ndx;
1771 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1772 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1773 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1774 if (diff < 0xFE && diff > 0)
1775 b[cnt++] = (char)diff;
1776 else if (diff < 0 || diff > 0x7FFF) {
1777 b[cnt++] = (char)0xFE;
1778 b[cnt++] = (char)((ndx >> 24) | 0x80);
1779 b[cnt++] = (char)ndx;
1780 b[cnt++] = (char)(ndx >> 8);
1781 b[cnt++] = (char)(ndx >> 16);
1783 b[cnt++] = (char)0xFE;
1784 b[cnt++] = (char)(diff >> 8);
1785 b[cnt++] = (char)diff;
1790 /* Receive a file-list index using a byte-reduction method. */
1791 int32 read_ndx(int f)
1793 static int32 prev_positive = -1, prev_negative = 1;
1794 int32 *prev_ptr, num;
1797 if (protocol_version < 30)
1801 if (CVAL(b, 0) == 0xFF) {
1803 prev_ptr = &prev_negative;
1804 } else if (CVAL(b, 0) == 0)
1807 prev_ptr = &prev_positive;
1808 if (CVAL(b, 0) == 0xFE) {
1810 if (CVAL(b, 0) & 0x80) {
1811 b[3] = CVAL(b, 0) & ~0x80;
1816 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1818 num = UVAL(b, 0) + *prev_ptr;
1820 if (prev_ptr == &prev_negative)
1825 /* Read a line of up to bufsiz-1 characters into buf. Strips
1826 * the (required) trailing newline and all carriage returns.
1827 * Returns 1 for success; 0 for I/O error or truncation. */
1828 int read_line_old(int f, char *buf, size_t bufsiz)
1830 bufsiz--; /* leave room for the null */
1831 while (bufsiz > 0) {
1833 read_buf(f, buf, 1);
1838 if (buf[0] != '\r') {
1847 void io_printf(int fd, const char *format, ...)
1850 char buf[BIGPATHBUFLEN];
1853 va_start(ap, format);
1854 len = vsnprintf(buf, sizeof buf, format, ap);
1858 exit_cleanup(RERR_STREAMIO);
1860 if (len > (int)sizeof buf) {
1861 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1862 exit_cleanup(RERR_STREAMIO);
1865 write_sbuf(fd, buf);
1868 /** Setup for multiplexing a MSG_* stream with the data stream. */
1869 void io_start_multiplex_out(void)
1871 io_flush(NORMAL_FLUSH);
1872 io_start_buffering_out(sock_f_out);
1873 io_multiplexing_out = 1;
1876 /** Setup for multiplexing a MSG_* stream with the data stream. */
1877 void io_start_multiplex_in(void)
1879 io_flush(NORMAL_FLUSH);
1880 io_start_buffering_in(sock_f_in);
1881 io_multiplexing_in = 1;
1884 /** Write an message to the multiplexed data stream. */
1885 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1887 if (!io_multiplexing_out)
1889 io_flush(NORMAL_FLUSH);
1890 stats.total_written += (len+4);
1891 mplex_write(sock_f_out, code, buf, len, convert);
1895 void io_end_multiplex_in(void)
1897 io_multiplexing_in = 0;
1898 io_end_buffering_in();
1901 /** Stop output multiplexing. */
1902 void io_end_multiplex_out(void)
1904 io_multiplexing_out = 0;
1905 io_end_buffering_out();
1908 void start_write_batch(int fd)
1910 /* Some communication has already taken place, but we don't
1911 * enable batch writing until here so that we can write a
1912 * canonical record of the communication even though the
1913 * actual communication so far depends on whether a daemon
1915 write_int(batch_fd, protocol_version);
1916 if (protocol_version >= 30)
1917 write_byte(batch_fd, inc_recurse);
1918 write_int(batch_fd, checksum_seed);
1921 write_batch_monitor_out = fd;
1923 write_batch_monitor_in = fd;
1926 void stop_write_batch(void)
1928 write_batch_monitor_out = -1;
1929 write_batch_monitor_in = -1;