2 * Socket and pipe I/O utilities used in rsync.
4 * Copyright (C) 1996-2001 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7 * Copyright (C) 2003-2008 Wayne Davison
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 3 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, visit the http://fsf.org website.
23 /* Rsync provides its own multiplexing system, which is used to send
24 * stderr and stdout over a single socket.
26 * For historical reasons this is off during the start of the
27 * connection, but it's switched on quite early using
28 * io_start_multiplex_out() and io_start_multiplex_in(). */
33 /** If no timeout is specified then use a 60 second select timeout */
34 #define SELECT_TIMEOUT 60
37 extern size_t bwlimit_writemax;
38 extern int io_timeout;
42 extern int am_generator;
43 extern int inc_recurse;
48 extern int read_batch;
49 extern int protect_args;
50 extern int checksum_seed;
51 extern int protocol_version;
52 extern int remove_source_files;
53 extern int preserve_hard_links;
54 extern struct stats stats;
55 extern struct file_list *cur_flist;
57 extern int filesfrom_convert;
58 extern iconv_t ic_send, ic_recv;
61 int csum_length = SHORT_SUM_LENGTH; /* initial value */
63 int ignore_timeout = 0;
66 int check_for_io_err = 0;
68 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
69 int kluge_around_eof = 0;
76 static int iobuf_f_in = -1;
77 static char *iobuf_in;
78 static size_t iobuf_in_siz;
79 static size_t iobuf_in_ndx;
80 static size_t iobuf_in_remaining;
82 static int iobuf_f_out = -1;
83 static char *iobuf_out;
84 static int iobuf_out_cnt;
86 int flist_forward_from = -1;
88 static int io_multiplexing_out;
89 static int io_multiplexing_in;
90 static time_t last_io_in;
91 static time_t last_io_out;
94 static int write_batch_monitor_in = -1;
95 static int write_batch_monitor_out = -1;
97 static int io_filesfrom_f_in = -1;
98 static int io_filesfrom_f_out = -1;
99 static xbuf ff_buf = EMPTY_XBUF;
100 static char ff_lastchar;
102 static xbuf iconv_buf = EMPTY_XBUF;
104 static int defer_forwarding_messages = 0, 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 struct flist_ndx_item {
128 struct flist_ndx_item *next;
132 struct flist_ndx_list {
133 struct flist_ndx_item *head, *tail;
136 static struct flist_ndx_list redo_list, hlink_list;
138 struct msg_list_item {
139 struct msg_list_item *next;
145 struct msg_list_item *head, *tail;
148 static struct msg_list msg_queue;
150 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
152 struct flist_ndx_item *item;
154 if (!(item = new(struct flist_ndx_item)))
155 out_of_memory("flist_ndx_push");
159 lp->tail->next = item;
165 static int flist_ndx_pop(struct flist_ndx_list *lp)
167 struct flist_ndx_item *next;
174 next = lp->head->next;
183 static void got_flist_entry_status(enum festatus status, const char *buf)
185 int ndx = IVAL(buf, 0);
186 struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
188 if (remove_source_files) {
190 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
194 flist->in_progress--;
198 if (remove_source_files)
199 send_msg(MSG_SUCCESS, buf, 4, 0);
200 if (preserve_hard_links) {
201 struct file_struct *file = flist->files[ndx - flist->ndx_start];
202 if (F_IS_HLINKED(file)) {
203 flist_ndx_push(&hlink_list, ndx);
204 flist->in_progress++;
211 flist_ndx_push(&redo_list, ndx);
218 static void check_timeout(void)
222 if (!io_timeout || ignore_timeout)
226 last_io_in = time(NULL);
232 if (t - last_io_in >= io_timeout) {
233 if (!am_server && !am_daemon) {
234 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
235 (int)(t-last_io_in));
237 exit_cleanup(RERR_TIMEOUT);
241 /* Note the fds used for the main socket (which might really be a pipe
242 * for a local transfer, but we can ignore that). */
243 void io_set_sock_fds(int f_in, int f_out)
249 void set_io_timeout(int secs)
253 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
254 select_timeout = SELECT_TIMEOUT;
256 select_timeout = io_timeout;
258 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
261 /* Setup the fd used to receive MSG_* messages. Only needed during the
262 * early stages of being a local sender (up through the sending of the
263 * file list) or when we're the generator (to fetch the messages from
265 void set_msg_fd_in(int fd)
270 /* Setup the fd used to send our MSG_* messages. Only needed when
271 * we're the receiver (to send our messages to the generator). */
272 void set_msg_fd_out(int fd)
275 set_nonblocking(msg_fd_out);
278 /* Add a message to the pending MSG_* list. */
279 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
281 struct msg_list_item *m;
282 int sz = len + 4 + sizeof m[0] - 1;
284 if (!(m = (struct msg_list_item *)new_array(char, sz)))
285 out_of_memory("msg_list_add");
287 m->convert = convert;
288 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
289 memcpy(m->buf + 4, buf, len);
297 static inline int flush_a_msg(int fd)
299 struct msg_list_item *m = msg_queue.head;
300 int len = IVAL(m->buf, 0) & 0xFFFFFF;
301 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
303 if (!(msg_queue.head = m->next))
304 msg_queue.tail = NULL;
306 defer_forwarding_messages++;
307 mplex_write(fd, tag, m->buf + 4, len, m->convert);
308 defer_forwarding_messages--;
315 static void msg_flush(void)
318 while (msg_queue.head && io_multiplexing_out)
319 stats.total_written += flush_a_msg(sock_f_out) + 4;
321 while (msg_queue.head)
322 (void)flush_a_msg(msg_fd_out);
326 static void check_for_d_option_error(const char *msg)
328 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
333 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
336 msg += sizeof REMOTE_OPTION_ERROR - 1;
337 if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
338 || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
341 for ( ; *msg != ':'; msg++) {
344 else if (*msg == 'e')
346 else if (strchr(rsync263_opts, *msg) == NULL)
352 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
356 /* Read a message from the MSG_* fd and handle it. This is called either
357 * during the early stages of being a local sender (up through the sending
358 * of the file list) or when we're the generator (to fetch the messages
359 * from the receiver). */
360 static void read_msg_fd(void)
364 struct file_list *flist;
368 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
369 * to this routine from writefd_unbuffered(). */
372 defer_forwarding_messages++;
377 len = tag & 0xFFFFFF;
378 tag = (tag >> 24) - MPLEX_BASE;
380 check_for_io_err = 0;
384 if (len < 0 || len > 1 || !am_generator) {
386 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
387 tag, len, who_am_i(),
388 inc_recurse ? "/inc" : "");
389 exit_cleanup(RERR_STREAMIO);
392 readfd(fd, buf, len);
393 stats.total_read = read_varlong(fd, 3);
398 if (len != 4 || !am_generator)
401 got_flist_entry_status(FES_REDO, buf);
404 if (len != 4 || !am_generator || !inc_recurse)
407 /* Read extra file list from receiver. */
408 assert(iobuf_in != NULL);
409 assert(iobuf_f_in == fd);
410 if (DEBUG_GTE(FLIST, 2)) {
411 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
412 who_am_i(), IVAL(buf,0));
414 flist = recv_file_list(fd);
415 flist->parent_ndx = IVAL(buf,0);
416 /* If the sender is going to send us an MSG_IO_ERROR value, it
417 * will always be the very next message following MSG_FLIST. */
418 check_for_io_err = 1;
419 #ifdef SUPPORT_HARD_LINKS
420 if (preserve_hard_links)
421 match_hard_links(flist);
425 if (len != 0 || !am_generator || !inc_recurse)
432 readfd(fd, buf, len);
433 io_error |= IVAL(buf, 0);
436 if (len >= (int)sizeof buf || !am_generator)
438 readfd(fd, buf, len);
439 send_msg(MSG_DELETED, buf, len, 1);
442 if (len != 4 || !am_generator)
445 got_flist_entry_status(FES_SUCCESS, buf);
448 if (len != 4 || !am_generator)
451 got_flist_entry_status(FES_NO_SEND, buf);
453 case MSG_ERROR_SOCKET:
458 if (tag == MSG_ERROR_SOCKET)
459 io_end_multiplex_out();
471 rwrite((enum logcode)tag, buf, n, !am_generator);
476 rprintf(FERROR, "unknown message %d:%d [%s]\n",
477 tag, len, who_am_i());
478 exit_cleanup(RERR_STREAMIO);
483 if (!--defer_forwarding_messages && !no_flush)
487 /* This is used by the generator to limit how many file transfers can
488 * be active at once when --remove-source-files is specified. Without
489 * this, sender-side deletions were mostly happening at the end. */
490 void increment_active_files(int ndx, int itemizing, enum logcode code)
492 /* TODO: tune these limits? */
493 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
494 check_for_finished_files(itemizing, code, 0);
496 io_flush(NORMAL_FLUSH);
502 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
505 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
506 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
508 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
512 /* We need to convert buf before doing anything else so that we
513 * can include the (converted) byte length in the message header. */
514 if (convert && ic_send != (iconv_t)-1) {
517 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
518 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
520 iconvbufs(ic_send, &inbuf, &outbuf,
521 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
523 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
524 exit_cleanup(RERR_UNSUPPORTED);
527 n = len = outbuf.len;
530 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
531 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
533 memcpy(buffer + 4, buf, n);
535 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
537 keep_defer_forwarding++; /* defer_forwarding_messages++ on return */
538 writefd_unbuffered(fd, buffer, n+4);
539 keep_defer_forwarding--;
542 writefd_unbuffered(fd, buf+n, len-n);
544 if (!--defer_forwarding_messages && !no_flush)
548 int send_msg(enum msgcode code, const char *buf, int len, int convert)
550 if (msg_fd_out < 0) {
551 if (!defer_forwarding_messages)
552 return io_multiplex_write(code, buf, len, convert);
553 if (!io_multiplexing_out)
555 msg_list_add(&msg_queue, code, buf, len, convert);
558 if (flist_forward_from >= 0)
559 msg_list_add(&msg_queue, code, buf, len, convert);
561 mplex_write(msg_fd_out, code, buf, len, convert);
565 void send_msg_int(enum msgcode code, int num)
568 SIVAL(numbuf, 0, num);
569 send_msg(code, numbuf, 4, 0);
572 void wait_for_receiver(void)
574 if (io_flush(FULL_FLUSH))
579 int get_redo_num(void)
581 return flist_ndx_pop(&redo_list);
584 int get_hlink_num(void)
586 return flist_ndx_pop(&hlink_list);
590 * When we're the receiver and we have a local --files-from list of names
591 * that needs to be sent over the socket to the sender, we have to do two
592 * things at the same time: send the sender a list of what files we're
593 * processing and read the incoming file+info list from the sender. We do
594 * this by augmenting the read_timeout() function to copy this data. It
595 * uses ff_buf to read a block of data from f_in (when it is ready, since
596 * it might be a pipe) and then blast it out f_out (when it is ready to
597 * receive more data).
599 void io_set_filesfrom_fds(int f_in, int f_out)
601 io_filesfrom_f_in = f_in;
602 io_filesfrom_f_out = f_out;
603 alloc_xbuf(&ff_buf, 2048);
606 alloc_xbuf(&iconv_buf, 1024);
610 /* It's almost always an error to get an EOF when we're trying to read from the
611 * network, because the protocol is (for the most part) self-terminating.
613 * There is one case for the receiver when it is at the end of the transfer
614 * (hanging around reading any keep-alive packets that might come its way): if
615 * the sender dies before the generator's kill-signal comes through, we can end
616 * up here needing to loop until the kill-signal arrives. In this situation,
617 * kluge_around_eof will be < 0.
619 * There is another case for older protocol versions (< 24) where the module
620 * listing was not terminated, so we must ignore an EOF error in that case and
621 * exit. In this situation, kluge_around_eof will be > 0. */
622 static void whine_about_eof(int fd)
624 if (kluge_around_eof && fd == sock_f_in) {
626 if (kluge_around_eof > 0)
628 /* If we're still here after 10 seconds, exit with an error. */
629 for (i = 10*1000/20; i--; )
633 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
634 "(%s bytes received so far) [%s]\n",
635 big_num(stats.total_read, 0), who_am_i());
637 exit_cleanup(RERR_STREAMIO);
641 * Read from a socket with I/O timeout. return the number of bytes
642 * read. If no bytes can be read then exit, never return a number <= 0.
644 * TODO: If the remote shell connection fails, then current versions
645 * actually report an "unexpected EOF" error here. Since it's a
646 * fairly common mistake to try to use rsh when ssh is required, we
647 * should trap that: if we fail to read any data at all, we should
648 * give a better explanation. We can tell whether the connection has
649 * started by looking e.g. at whether the remote version is known yet.
651 static int read_timeout(int fd, char *buf, size_t len)
655 io_flush(FULL_FLUSH);
658 /* until we manage to read *something* */
667 if (io_filesfrom_f_out >= 0) {
669 if (ff_buf.len == 0) {
670 if (io_filesfrom_f_in >= 0) {
671 FD_SET(io_filesfrom_f_in, &r_fds);
672 new_fd = io_filesfrom_f_in;
674 io_filesfrom_f_out = -1;
678 FD_SET(io_filesfrom_f_out, &w_fds);
679 new_fd = io_filesfrom_f_out;
685 tv.tv_sec = select_timeout;
690 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
693 if (errno == EBADF) {
694 defer_forwarding_messages = 0;
695 exit_cleanup(RERR_SOCKETIO);
701 if (io_filesfrom_f_out >= 0) {
703 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
704 int l = write(io_filesfrom_f_out,
705 ff_buf.buf + ff_buf.pos,
708 if (!(ff_buf.len -= l))
712 } else if (errno != EINTR) {
713 /* XXX should we complain? */
714 io_filesfrom_f_out = -1;
717 } else if (io_filesfrom_f_in >= 0) {
718 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
720 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
722 xbuf *ibuf = &ff_buf;
724 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
726 if (l == 0 || errno != EINTR) {
727 /* Send end-of-file marker */
728 memcpy(ff_buf.buf, "\0\0", 2);
729 ff_buf.len = ff_lastchar? 2 : 1;
731 io_filesfrom_f_in = -1;
735 if (filesfrom_convert) {
738 iconvbufs(ic_send, &iconv_buf, &ff_buf,
739 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
744 char *s = ff_buf.buf + l;
745 /* Transform CR and/or LF into '\0' */
746 while (s-- > ff_buf.buf) {
747 if (*s == '\n' || *s == '\r')
752 /* Last buf ended with a '\0', so don't
753 * let this buf start with one. */
754 while (l && ff_buf.buf[ff_buf.pos] == '\0')
760 char *f = ff_buf.buf + ff_buf.pos;
763 /* Eliminate any multi-'\0' runs. */
765 if (!(*t++ = *f++)) {
766 while (f != eob && !*f)
778 if (!FD_ISSET(fd, &r_fds))
781 n = read(fd, buf, len);
785 whine_about_eof(fd); /* Doesn't return. */
786 if (errno == EINTR || errno == EWOULDBLOCK
790 /* Don't write errors on a dead socket. */
791 if (fd == sock_f_in) {
792 io_end_multiplex_out();
793 rsyserr(FERROR_SOCKET, errno, "read error");
795 rsyserr(FERROR, errno, "read error");
796 exit_cleanup(RERR_STREAMIO);
803 if (fd == sock_f_in && io_timeout)
804 last_io_in = time(NULL);
810 /* Read a line into the "buf" buffer. */
811 int read_line(int fd, char *buf, size_t bufsiz, int flags)
817 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
818 realloc_xbuf(&iconv_buf, bufsiz + 1024);
823 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
827 eob = s + bufsiz - 1;
829 cnt = read(fd, &ch, 1);
830 if (cnt < 0 && (errno == EWOULDBLOCK
831 || errno == EINTR || errno == EAGAIN)) {
838 tv.tv_sec = select_timeout;
840 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
842 /*if (FD_ISSET(fd, &e_fds))
843 rprintf(FINFO, "select exception on fd %d\n", fd); */
848 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
849 /* Skip empty lines if dumping comments. */
850 if (flags & RL_DUMP_COMMENTS && s == buf)
859 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
863 if (flags & RL_CONVERT) {
865 INIT_XBUF(outbuf, buf, 0, bufsiz);
867 iconv_buf.len = s - iconv_buf.buf;
868 iconvbufs(ic_recv, &iconv_buf, &outbuf,
869 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
870 outbuf.buf[outbuf.len] = '\0';
878 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
879 char ***argv_p, int *argc_p, char **request_p)
881 int maxargs = MAX_ARGS;
885 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
888 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
891 if (!(argv = new_array(char *, maxargs)))
892 out_of_memory("read_args");
893 if (mod_name && !protect_args)
894 argv[argc++] = "rsyncd";
897 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
900 if (argc == maxargs-1) {
902 if (!(argv = realloc_array(argv, char *, maxargs)))
903 out_of_memory("read_args");
908 *request_p = strdup(buf);
912 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
914 glob_expand(buf, &argv, &argc, &maxargs);
916 if (!(p = strdup(buf)))
917 out_of_memory("read_args");
919 if (*p == '.' && p[1] == '\0')
925 glob_expand(NULL, NULL, NULL, NULL);
931 int io_start_buffering_out(int f_out)
934 assert(f_out == iobuf_f_out);
937 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
938 out_of_memory("io_start_buffering_out");
944 int io_start_buffering_in(int f_in)
947 assert(f_in == iobuf_f_in);
950 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
951 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
952 out_of_memory("io_start_buffering_in");
957 void io_end_buffering_in(void)
964 iobuf_in_remaining = 0;
968 void io_end_buffering_out(void)
972 io_flush(FULL_FLUSH);
978 void maybe_flush_socket(int important)
980 if (iobuf_out && iobuf_out_cnt
981 && (important || time(NULL) - last_io_out >= 5))
982 io_flush(NORMAL_FLUSH);
985 void maybe_send_keepalive(void)
987 if (time(NULL) - last_io_out >= allowed_lull) {
988 if (!iobuf_out || !iobuf_out_cnt) {
989 if (protocol_version < 29)
990 return; /* there's nothing we can do */
991 if (protocol_version >= 30)
992 send_msg(MSG_NOOP, "", 0, 0);
994 write_int(sock_f_out, cur_flist->used);
995 write_shortint(sock_f_out, ITEM_IS_NEW);
999 io_flush(NORMAL_FLUSH);
1003 void start_flist_forward(int f_in)
1005 assert(iobuf_out != NULL);
1006 assert(iobuf_f_out == msg_fd_out);
1007 flist_forward_from = f_in;
1010 void stop_flist_forward()
1012 flist_forward_from = -1;
1013 io_flush(FULL_FLUSH);
1017 * Continue trying to read len bytes - don't return until len has been
1020 static void read_loop(int fd, char *buf, size_t len)
1023 int n = read_timeout(fd, buf, len);
1031 * Read from the file descriptor handling multiplexing - return number
1034 * Never returns <= 0.
1036 static int readfd_unbuffered(int fd, char *buf, size_t len)
1040 char line[BIGPATHBUFLEN];
1042 if (!iobuf_in || fd != iobuf_f_in)
1043 return read_timeout(fd, buf, len);
1045 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1046 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1051 if (iobuf_in_remaining) {
1052 len = MIN(len, iobuf_in_remaining);
1053 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1054 iobuf_in_ndx += len;
1055 iobuf_in_remaining -= len;
1060 read_loop(fd, line, 4);
1061 tag = IVAL(line, 0);
1063 msg_bytes = tag & 0xFFFFFF;
1064 tag = (tag >> 24) - MPLEX_BASE;
1066 check_for_io_err = 0;
1070 if (msg_bytes > iobuf_in_siz) {
1071 if (!(iobuf_in = realloc_array(iobuf_in, char,
1073 out_of_memory("readfd_unbuffered");
1074 iobuf_in_siz = msg_bytes;
1076 read_loop(fd, iobuf_in, msg_bytes);
1077 iobuf_in_remaining = msg_bytes;
1082 maybe_send_keepalive();
1087 read_loop(fd, line, msg_bytes);
1088 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1089 io_error |= IVAL(line, 0);
1092 if (msg_bytes >= sizeof line)
1095 if (ic_recv != (iconv_t)-1) {
1100 INIT_CONST_XBUF(outbuf, line);
1101 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1104 inbuf.len = msg_bytes > sizeof ibuf
1105 ? sizeof ibuf : msg_bytes;
1106 read_loop(fd, inbuf.buf, inbuf.len);
1107 if (!(msg_bytes -= inbuf.len)
1108 && !ibuf[inbuf.len-1])
1109 inbuf.len--, add_null = 1;
1110 if (iconvbufs(ic_send, &inbuf, &outbuf,
1111 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1115 if (outbuf.len == outbuf.size)
1117 outbuf.buf[outbuf.len++] = '\0';
1119 msg_bytes = outbuf.len;
1122 read_loop(fd, line, msg_bytes);
1123 /* A directory name was sent with the trailing null */
1124 if (msg_bytes > 0 && !line[msg_bytes-1])
1125 log_delete(line, S_IFDIR);
1127 line[msg_bytes] = '\0';
1128 log_delete(line, S_IFREG);
1132 if (msg_bytes != 4) {
1134 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1135 tag, (long)msg_bytes, who_am_i());
1136 exit_cleanup(RERR_STREAMIO);
1138 read_loop(fd, line, msg_bytes);
1139 successful_send(IVAL(line, 0));
1144 read_loop(fd, line, msg_bytes);
1145 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1149 case MSG_ERROR_XFER:
1151 if (msg_bytes >= sizeof line) {
1154 "multiplexing overflow %d:%ld [%s]\n",
1155 tag, (long)msg_bytes, who_am_i());
1156 exit_cleanup(RERR_STREAMIO);
1158 read_loop(fd, line, msg_bytes);
1159 rwrite((enum logcode)tag, line, msg_bytes, 1);
1160 if (first_message) {
1161 if (list_only && !am_sender && tag == 1) {
1162 line[msg_bytes] = '\0';
1163 check_for_d_option_error(line);
1169 rprintf(FERROR, "unexpected tag %d [%s]\n",
1171 exit_cleanup(RERR_STREAMIO);
1175 if (iobuf_in_remaining == 0)
1176 io_flush(NORMAL_FLUSH);
1181 /* Do a buffered read from fd. Don't return until all N bytes have
1182 * been read. If all N can't be read then exit with an error. */
1183 static void readfd(int fd, char *buffer, size_t N)
1189 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1193 if (fd == write_batch_monitor_in) {
1194 if ((size_t)write(batch_fd, buffer, total) != total)
1195 exit_cleanup(RERR_FILEIO);
1198 if (fd == flist_forward_from)
1199 writefd(iobuf_f_out, buffer, total);
1201 if (fd == sock_f_in)
1202 stats.total_read += total;
1205 unsigned short read_shortint(int f)
1209 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1212 int32 read_int(int f)
1219 #if SIZEOF_INT32 > 4
1220 if (num & (int32)0x80000000)
1221 num |= ~(int32)0xffffffff;
1226 int32 read_varint(int f)
1236 readfd(f, (char*)&ch, 1);
1237 extra = int_byte_extra[ch / 4];
1239 uchar bit = ((uchar)1<<(8-extra));
1240 if (extra >= (int)sizeof u.b) {
1241 rprintf(FERROR, "Overflow in read_varint()\n");
1242 exit_cleanup(RERR_STREAMIO);
1244 readfd(f, u.b, extra);
1245 u.b[extra] = ch & (bit-1);
1248 #if CAREFUL_ALIGNMENT
1251 #if SIZEOF_INT32 > 4
1252 if (u.x & (int32)0x80000000)
1253 u.x |= ~(int32)0xffffffff;
1258 int64 read_varlong(int f, uchar min_bytes)
1267 #if SIZEOF_INT64 < 8
1272 readfd(f, b2, min_bytes);
1273 memcpy(u.b, b2+1, min_bytes-1);
1274 extra = int_byte_extra[CVAL(b2, 0) / 4];
1276 uchar bit = ((uchar)1<<(8-extra));
1277 if (min_bytes + extra > (int)sizeof u.b) {
1278 rprintf(FERROR, "Overflow in read_varlong()\n");
1279 exit_cleanup(RERR_STREAMIO);
1281 readfd(f, u.b + min_bytes - 1, extra);
1282 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1283 #if SIZEOF_INT64 < 8
1284 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1285 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1286 exit_cleanup(RERR_UNSUPPORTED);
1290 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1291 #if SIZEOF_INT64 < 8
1293 #elif CAREFUL_ALIGNMENT
1294 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1299 int64 read_longint(int f)
1301 #if SIZEOF_INT64 >= 8
1304 int32 num = read_int(f);
1306 if (num != (int32)0xffffffff)
1309 #if SIZEOF_INT64 < 8
1310 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1311 exit_cleanup(RERR_UNSUPPORTED);
1314 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1318 void read_buf(int f, char *buf, size_t len)
1323 void read_sbuf(int f, char *buf, size_t len)
1325 readfd(f, buf, len);
1329 uchar read_byte(int f)
1332 readfd(f, (char *)&c, 1);
1336 int read_vstring(int f, char *buf, int bufsize)
1338 int len = read_byte(f);
1341 len = (len & ~0x80) * 0x100 + read_byte(f);
1343 if (len >= bufsize) {
1344 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1350 readfd(f, buf, len);
1355 /* Populate a sum_struct with values from the socket. This is
1356 * called by both the sender and the receiver. */
1357 void read_sum_head(int f, struct sum_struct *sum)
1359 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1360 sum->count = read_int(f);
1361 if (sum->count < 0) {
1362 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1363 (long)sum->count, who_am_i());
1364 exit_cleanup(RERR_PROTOCOL);
1366 sum->blength = read_int(f);
1367 if (sum->blength < 0 || sum->blength > max_blength) {
1368 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1369 (long)sum->blength, who_am_i());
1370 exit_cleanup(RERR_PROTOCOL);
1372 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1373 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1374 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1375 sum->s2length, who_am_i());
1376 exit_cleanup(RERR_PROTOCOL);
1378 sum->remainder = read_int(f);
1379 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1380 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1381 (long)sum->remainder, who_am_i());
1382 exit_cleanup(RERR_PROTOCOL);
1386 /* Send the values from a sum_struct over the socket. Set sum to
1387 * NULL if there are no checksums to send. This is called by both
1388 * the generator and the sender. */
1389 void write_sum_head(int f, struct sum_struct *sum)
1391 static struct sum_struct null_sum;
1396 write_int(f, sum->count);
1397 write_int(f, sum->blength);
1398 if (protocol_version >= 27)
1399 write_int(f, sum->s2length);
1400 write_int(f, sum->remainder);
1404 * Sleep after writing to limit I/O bandwidth usage.
1406 * @todo Rather than sleeping after each write, it might be better to
1407 * use some kind of averaging. The current algorithm seems to always
1408 * use a bit less bandwidth than specified, because it doesn't make up
1409 * for slow periods. But arguably this is a feature. In addition, we
1410 * ought to take the time used to write the data into account.
1412 * During some phases of big transfers (file FOO is uptodate) this is
1413 * called with a small bytes_written every time. As the kernel has to
1414 * round small waits up to guarantee that we actually wait at least the
1415 * requested number of microseconds, this can become grossly inaccurate.
1416 * We therefore keep track of the bytes we've written over time and only
1417 * sleep when the accumulated delay is at least 1 tenth of a second.
1419 static void sleep_for_bwlimit(int bytes_written)
1421 static struct timeval prior_tv;
1422 static long total_written = 0;
1423 struct timeval tv, start_tv;
1424 long elapsed_usec, sleep_usec;
1426 #define ONE_SEC 1000000L /* # of microseconds in a second */
1428 if (!bwlimit_writemax)
1431 total_written += bytes_written;
1433 gettimeofday(&start_tv, NULL);
1434 if (prior_tv.tv_sec) {
1435 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1436 + (start_tv.tv_usec - prior_tv.tv_usec);
1437 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1438 if (total_written < 0)
1442 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1443 if (sleep_usec < ONE_SEC / 10) {
1444 prior_tv = start_tv;
1448 tv.tv_sec = sleep_usec / ONE_SEC;
1449 tv.tv_usec = sleep_usec % ONE_SEC;
1450 select(0, NULL, NULL, NULL, &tv);
1452 gettimeofday(&prior_tv, NULL);
1453 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1454 + (prior_tv.tv_usec - start_tv.tv_usec);
1455 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1458 /* Write len bytes to the file descriptor fd, looping as necessary to get
1459 * the job done and also (in certain circumstances) reading any data on
1460 * msg_fd_in to avoid deadlock.
1462 * This function underlies the multiplexing system. The body of the
1463 * application never calls this function directly. */
1464 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1466 size_t n, total = 0;
1467 fd_set w_fds, r_fds, e_fds;
1468 int maxfd, count, cnt, using_r_fds;
1473 defer_forwarding_messages++, defer_inc++;
1475 while (total < len) {
1482 if (msg_fd_in >= 0) {
1484 FD_SET(msg_fd_in, &r_fds);
1485 if (msg_fd_in > maxfd)
1491 tv.tv_sec = select_timeout;
1495 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1496 &w_fds, &e_fds, &tv);
1499 if (count < 0 && errno == EBADF)
1500 exit_cleanup(RERR_SOCKETIO);
1505 /*if (FD_ISSET(fd, &e_fds))
1506 rprintf(FINFO, "select exception on fd %d\n", fd); */
1508 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1511 if (!FD_ISSET(fd, &w_fds))
1515 if (bwlimit_writemax && n > bwlimit_writemax)
1516 n = bwlimit_writemax;
1517 cnt = write(fd, buf + total, n);
1523 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1529 /* Don't try to write errors back across the stream. */
1530 if (fd == sock_f_out)
1531 io_end_multiplex_out();
1532 /* Don't try to write errors down a failing msg pipe. */
1533 if (am_server && fd == msg_fd_out)
1534 exit_cleanup(RERR_STREAMIO);
1535 rsyserr(FERROR, errno,
1536 "writefd_unbuffered failed to write %ld bytes [%s]",
1537 (long)len, who_am_i());
1538 /* If the other side is sending us error messages, try
1539 * to grab any messages they sent before they died. */
1540 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1544 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1546 exit_cleanup(RERR_STREAMIO);
1550 defer_forwarding_messages++, defer_inc++;
1552 if (fd == sock_f_out) {
1553 if (io_timeout || am_generator)
1554 last_io_out = time(NULL);
1555 sleep_for_bwlimit(cnt);
1560 if (keep_defer_forwarding)
1562 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1566 int io_flush(int flush_it_all)
1568 int flushed_something = 0;
1573 if (iobuf_out_cnt) {
1574 if (io_multiplexing_out)
1575 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1577 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1579 flushed_something = 1;
1582 if (flush_it_all && !defer_forwarding_messages && msg_queue.head) {
1584 flushed_something = 1;
1587 return flushed_something;
1590 static void writefd(int fd, const char *buf, size_t len)
1592 if (fd == sock_f_out)
1593 stats.total_written += len;
1595 if (fd == write_batch_monitor_out) {
1596 if ((size_t)write(batch_fd, buf, len) != len)
1597 exit_cleanup(RERR_FILEIO);
1600 if (!iobuf_out || fd != iobuf_f_out) {
1601 writefd_unbuffered(fd, buf, len);
1606 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1608 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1614 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1615 io_flush(NORMAL_FLUSH);
1619 void write_shortint(int f, unsigned short x)
1623 b[1] = (char)(x >> 8);
1627 void write_int(int f, int32 x)
1634 void write_varint(int f, int32 x)
1642 while (cnt > 1 && b[cnt] == 0)
1644 bit = ((uchar)1<<(7-cnt+1));
1645 if (CVAL(b, cnt) >= bit) {
1649 *b = b[cnt] | ~(bit*2-1);
1656 void write_varlong(int f, int64 x, uchar min_bytes)
1663 #if SIZEOF_INT64 >= 8
1664 SIVAL(b, 5, x >> 32);
1666 if (x <= 0x7FFFFFFF && x >= 0)
1667 memset(b + 5, 0, 4);
1669 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1670 exit_cleanup(RERR_UNSUPPORTED);
1674 while (cnt > min_bytes && b[cnt] == 0)
1676 bit = ((uchar)1<<(7-cnt+min_bytes));
1677 if (CVAL(b, cnt) >= bit) {
1680 } else if (cnt > min_bytes)
1681 *b = b[cnt] | ~(bit*2-1);
1689 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1690 * 64-bit types on this platform.
1692 void write_longint(int f, int64 x)
1694 char b[12], * const s = b+4;
1697 if (x <= 0x7FFFFFFF && x >= 0) {
1702 #if SIZEOF_INT64 < 8
1703 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1704 exit_cleanup(RERR_UNSUPPORTED);
1707 SIVAL(s, 4, x >> 32);
1712 void write_buf(int f, const char *buf, size_t len)
1717 /** Write a string to the connection */
1718 void write_sbuf(int f, const char *buf)
1720 writefd(f, buf, strlen(buf));
1723 void write_byte(int f, uchar c)
1725 writefd(f, (char *)&c, 1);
1728 void write_vstring(int f, const char *str, int len)
1730 uchar lenbuf[3], *lb = lenbuf;
1735 "attempting to send over-long vstring (%d > %d)\n",
1737 exit_cleanup(RERR_PROTOCOL);
1739 *lb++ = len / 0x100 + 0x80;
1743 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1745 writefd(f, str, len);
1748 /* Send a file-list index using a byte-reduction method. */
1749 void write_ndx(int f, int32 ndx)
1751 static int32 prev_positive = -1, prev_negative = 1;
1752 int32 diff, cnt = 0;
1755 if (protocol_version < 30 || read_batch) {
1760 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1761 * negative nums as a positive after sending a leading 0xFF. */
1763 diff = ndx - prev_positive;
1764 prev_positive = ndx;
1765 } else if (ndx == NDX_DONE) {
1770 b[cnt++] = (char)0xFF;
1772 diff = ndx - prev_negative;
1773 prev_negative = ndx;
1776 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1777 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1778 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1779 if (diff < 0xFE && diff > 0)
1780 b[cnt++] = (char)diff;
1781 else if (diff < 0 || diff > 0x7FFF) {
1782 b[cnt++] = (char)0xFE;
1783 b[cnt++] = (char)((ndx >> 24) | 0x80);
1784 b[cnt++] = (char)ndx;
1785 b[cnt++] = (char)(ndx >> 8);
1786 b[cnt++] = (char)(ndx >> 16);
1788 b[cnt++] = (char)0xFE;
1789 b[cnt++] = (char)(diff >> 8);
1790 b[cnt++] = (char)diff;
1795 /* Receive a file-list index using a byte-reduction method. */
1796 int32 read_ndx(int f)
1798 static int32 prev_positive = -1, prev_negative = 1;
1799 int32 *prev_ptr, num;
1802 if (protocol_version < 30)
1806 if (CVAL(b, 0) == 0xFF) {
1808 prev_ptr = &prev_negative;
1809 } else if (CVAL(b, 0) == 0)
1812 prev_ptr = &prev_positive;
1813 if (CVAL(b, 0) == 0xFE) {
1815 if (CVAL(b, 0) & 0x80) {
1816 b[3] = CVAL(b, 0) & ~0x80;
1821 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1823 num = UVAL(b, 0) + *prev_ptr;
1825 if (prev_ptr == &prev_negative)
1830 /* Read a line of up to bufsiz-1 characters into buf. Strips
1831 * the (required) trailing newline and all carriage returns.
1832 * Returns 1 for success; 0 for I/O error or truncation. */
1833 int read_line_old(int f, char *buf, size_t bufsiz)
1835 bufsiz--; /* leave room for the null */
1836 while (bufsiz > 0) {
1838 read_buf(f, buf, 1);
1843 if (buf[0] != '\r') {
1852 void io_printf(int fd, const char *format, ...)
1855 char buf[BIGPATHBUFLEN];
1858 va_start(ap, format);
1859 len = vsnprintf(buf, sizeof buf, format, ap);
1863 exit_cleanup(RERR_STREAMIO);
1865 if (len > (int)sizeof buf) {
1866 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1867 exit_cleanup(RERR_STREAMIO);
1870 write_sbuf(fd, buf);
1873 /** Setup for multiplexing a MSG_* stream with the data stream. */
1874 void io_start_multiplex_out(void)
1876 io_flush(NORMAL_FLUSH);
1877 io_start_buffering_out(sock_f_out);
1878 io_multiplexing_out = 1;
1881 /** Setup for multiplexing a MSG_* stream with the data stream. */
1882 void io_start_multiplex_in(void)
1884 io_flush(NORMAL_FLUSH);
1885 io_start_buffering_in(sock_f_in);
1886 io_multiplexing_in = 1;
1889 /** Write an message to the multiplexed data stream. */
1890 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1892 if (!io_multiplexing_out)
1894 io_flush(NORMAL_FLUSH);
1895 stats.total_written += (len+4);
1896 mplex_write(sock_f_out, code, buf, len, convert);
1900 void io_end_multiplex_in(void)
1902 io_multiplexing_in = 0;
1903 io_end_buffering_in();
1906 /** Stop output multiplexing. */
1907 void io_end_multiplex_out(void)
1909 io_multiplexing_out = 0;
1910 io_end_buffering_out();
1913 void start_write_batch(int fd)
1915 /* Some communication has already taken place, but we don't
1916 * enable batch writing until here so that we can write a
1917 * canonical record of the communication even though the
1918 * actual communication so far depends on whether a daemon
1920 write_int(batch_fd, protocol_version);
1921 if (protocol_version >= 30)
1922 write_byte(batch_fd, inc_recurse);
1923 write_int(batch_fd, checksum_seed);
1926 write_batch_monitor_out = fd;
1928 write_batch_monitor_in = fd;
1931 void stop_write_batch(void)
1933 write_batch_monitor_out = -1;
1934 write_batch_monitor_in = -1;