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) 2003-2007 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;
48 extern int read_batch;
49 extern int csum_length;
50 extern int protect_args;
51 extern int checksum_seed;
52 extern int protocol_version;
53 extern int remove_source_files;
54 extern int preserve_hard_links;
55 extern struct stats stats;
56 extern struct file_list *cur_flist;
58 extern int filesfrom_convert;
59 extern iconv_t ic_send, ic_recv;
62 const char phase_unknown[] = "unknown";
63 int ignore_timeout = 0;
67 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
68 int kluge_around_eof = 0;
75 static int iobuf_f_in = -1;
76 static char *iobuf_in;
77 static size_t iobuf_in_siz;
78 static size_t iobuf_in_ndx;
79 static size_t iobuf_in_remaining;
81 static int iobuf_f_out = -1;
82 static char *iobuf_out;
83 static int iobuf_out_cnt;
85 int flist_forward_from = -1;
87 static int io_multiplexing_out;
88 static int io_multiplexing_in;
89 static time_t last_io_in;
90 static time_t last_io_out;
93 static int write_batch_monitor_in = -1;
94 static int write_batch_monitor_out = -1;
96 static int io_filesfrom_f_in = -1;
97 static int io_filesfrom_f_out = -1;
98 static xbuf ff_buf = EMPTY_XBUF;
99 static char ff_lastchar;
101 static xbuf iconv_buf = EMPTY_XBUF;
103 static int defer_forwarding_messages = 0, defer_forwarding_keep = 0;
104 static int select_timeout = SELECT_TIMEOUT;
105 static int active_filecnt = 0;
106 static OFF_T active_bytecnt = 0;
108 static char int_byte_extra[64] = {
109 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
110 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
111 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
112 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
115 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
117 static void readfd(int fd, char *buffer, size_t N);
118 static void writefd(int fd, const char *buf, size_t len);
119 static void writefd_unbuffered(int fd, const char *buf, size_t len);
120 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
122 struct flist_ndx_item {
123 struct flist_ndx_item *next;
127 struct flist_ndx_list {
128 struct flist_ndx_item *head, *tail;
131 static struct flist_ndx_list redo_list, hlink_list;
133 struct msg_list_item {
134 struct msg_list_item *next;
140 struct msg_list_item *head, *tail;
143 static struct msg_list msg_queue;
145 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
147 struct flist_ndx_item *item;
149 if (!(item = new(struct flist_ndx_item)))
150 out_of_memory("flist_ndx_push");
154 lp->tail->next = item;
160 static int flist_ndx_pop(struct flist_ndx_list *lp)
162 struct flist_ndx_item *next;
169 next = lp->head->next;
178 static void got_flist_entry_status(enum festatus status, const char *buf)
180 int ndx = IVAL(buf, 0);
181 struct file_list *flist = flist_for_ndx(ndx);
183 assert(flist != NULL);
185 if (remove_source_files) {
187 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
191 flist->in_progress--;
195 if (remove_source_files)
196 send_msg(MSG_SUCCESS, buf, 4, 0);
197 if (preserve_hard_links) {
198 struct file_struct *file = flist->files[ndx - flist->ndx_start];
199 if (F_IS_HLINKED(file)) {
200 flist_ndx_push(&hlink_list, ndx);
201 flist->in_progress++;
208 flist_ndx_push(&redo_list, ndx);
215 static void check_timeout(void)
219 if (!io_timeout || ignore_timeout)
223 last_io_in = time(NULL);
229 if (t - last_io_in >= io_timeout) {
230 if (!am_server && !am_daemon) {
231 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
232 (int)(t-last_io_in));
234 exit_cleanup(RERR_TIMEOUT);
238 /* Note the fds used for the main socket (which might really be a pipe
239 * for a local transfer, but we can ignore that). */
240 void io_set_sock_fds(int f_in, int f_out)
246 void set_io_timeout(int secs)
250 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
251 select_timeout = SELECT_TIMEOUT;
253 select_timeout = io_timeout;
255 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
258 /* Setup the fd used to receive MSG_* messages. Only needed during the
259 * early stages of being a local sender (up through the sending of the
260 * file list) or when we're the generator (to fetch the messages from
262 void set_msg_fd_in(int fd)
267 /* Setup the fd used to send our MSG_* messages. Only needed when
268 * we're the receiver (to send our messages to the generator). */
269 void set_msg_fd_out(int fd)
272 set_nonblocking(msg_fd_out);
275 /* Add a message to the pending MSG_* list. */
276 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
278 struct msg_list_item *m;
279 int sz = len + 4 + sizeof m[0] - 1;
281 if (!(m = (struct msg_list_item *)new_array(char, sz)))
282 out_of_memory("msg_list_add");
284 m->convert = convert;
285 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
286 memcpy(m->buf + 4, buf, len);
294 static void msg_flush(void)
297 while (msg_queue.head && io_multiplexing_out) {
298 struct msg_list_item *m = msg_queue.head;
299 int len = IVAL(m->buf, 0) & 0xFFFFFF;
300 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
301 if (!(msg_queue.head = m->next))
302 msg_queue.tail = NULL;
303 stats.total_written += len + 4;
304 defer_forwarding_messages++;
305 mplex_write(sock_f_out, tag, m->buf + 4, len, m->convert);
306 defer_forwarding_messages--;
310 while (msg_queue.head) {
311 struct msg_list_item *m = msg_queue.head;
312 int len = IVAL(m->buf, 0) & 0xFFFFFF;
313 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
314 if (!(msg_queue.head = m->next))
315 msg_queue.tail = NULL;
316 defer_forwarding_messages++;
317 mplex_write(msg_fd_out, tag, m->buf + 4, len, m->convert);
318 defer_forwarding_messages--;
324 /* Read a message from the MSG_* fd and handle it. This is called either
325 * during the early stages of being a local sender (up through the sending
326 * of the file list) or when we're the generator (to fetch the messages
327 * from the receiver). */
328 static void read_msg_fd(void)
332 struct file_list *flist;
336 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
337 * to this routine from writefd_unbuffered(). */
340 defer_forwarding_messages++;
345 len = tag & 0xFFFFFF;
346 tag = (tag >> 24) - MPLEX_BASE;
350 if (len < 0 || len > 1 || !am_generator) {
352 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
353 tag, len, who_am_i(),
354 inc_recurse ? "/inc" : "");
355 exit_cleanup(RERR_STREAMIO);
358 readfd(fd, buf, len);
359 stats.total_read = read_varlong(fd, 3);
364 if (len != 4 || !am_generator)
367 got_flist_entry_status(FES_REDO, buf);
370 if (len != 4 || !am_generator || !inc_recurse)
373 /* Read extra file list from receiver. */
374 assert(iobuf_in != NULL);
375 assert(iobuf_f_in == fd);
377 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
378 who_am_i(), IVAL(buf,0));
380 flist = recv_file_list(fd);
381 flist->parent_ndx = IVAL(buf,0);
382 #ifdef SUPPORT_HARD_LINKS
383 if (preserve_hard_links)
384 match_hard_links(flist);
388 if (len != 0 || !am_generator || !inc_recurse)
395 readfd(fd, buf, len);
396 io_error |= IVAL(buf, 0);
399 if (len >= (int)sizeof buf || !am_generator)
401 readfd(fd, buf, len);
402 send_msg(MSG_DELETED, buf, len, 1);
405 if (len != 4 || !am_generator)
408 got_flist_entry_status(FES_SUCCESS, buf);
411 if (len != 4 || !am_generator)
414 got_flist_entry_status(FES_NO_SEND, buf);
420 if (tag == MSG_SOCKERR)
421 io_end_multiplex_out();
431 rwrite((enum logcode)tag, buf, n, !am_generator);
436 rprintf(FERROR, "unknown message %d:%d [%s]\n",
437 tag, len, who_am_i());
438 exit_cleanup(RERR_STREAMIO);
443 if (!--defer_forwarding_messages && !no_flush)
447 /* This is used by the generator to limit how many file transfers can
448 * be active at once when --remove-source-files is specified. Without
449 * this, sender-side deletions were mostly happening at the end. */
450 void increment_active_files(int ndx, int itemizing, enum logcode code)
452 /* TODO: tune these limits? */
453 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
454 check_for_finished_files(itemizing, code, 0);
456 io_flush(NORMAL_FLUSH);
462 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
465 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
466 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
468 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
472 /* We need to convert buf before doing anything else so that we
473 * can include the (converted) byte length in the message header. */
474 if (convert && ic_send != (iconv_t)-1) {
477 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
478 INIT_XBUF(inbuf, (char*)buf, len, -1);
480 iconvbufs(ic_send, &inbuf, &outbuf,
481 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
483 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
484 exit_cleanup(RERR_UNSUPPORTED);
487 n = len = outbuf.len;
490 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
491 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
493 memcpy(buffer + 4, buf, n);
495 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
497 defer_forwarding_keep = 1; /* defer_forwarding_messages++ on return */
498 writefd_unbuffered(fd, buffer, n+4);
499 defer_forwarding_keep = 0;
502 writefd_unbuffered(fd, buf+n, len-n);
504 if (!--defer_forwarding_messages && !no_flush)
508 int send_msg(enum msgcode code, const char *buf, int len, int convert)
510 if (msg_fd_out < 0) {
511 if (!defer_forwarding_messages)
512 return io_multiplex_write(code, buf, len, convert);
513 if (!io_multiplexing_out)
515 msg_list_add(&msg_queue, code, buf, len, convert);
518 if (flist_forward_from >= 0)
519 msg_list_add(&msg_queue, code, buf, len, convert);
521 mplex_write(msg_fd_out, code, buf, len, convert);
525 void send_msg_int(enum msgcode code, int num)
528 SIVAL(numbuf, 0, num);
529 send_msg(code, numbuf, 4, 0);
532 void wait_for_receiver(void)
535 io_flush(NORMAL_FLUSH);
540 int get_redo_num(void)
542 return flist_ndx_pop(&redo_list);
545 int get_hlink_num(void)
547 return flist_ndx_pop(&hlink_list);
551 * When we're the receiver and we have a local --files-from list of names
552 * that needs to be sent over the socket to the sender, we have to do two
553 * things at the same time: send the sender a list of what files we're
554 * processing and read the incoming file+info list from the sender. We do
555 * this by augmenting the read_timeout() function to copy this data. It
556 * uses ff_buf to read a block of data from f_in (when it is ready, since
557 * it might be a pipe) and then blast it out f_out (when it is ready to
558 * receive more data).
560 void io_set_filesfrom_fds(int f_in, int f_out)
562 io_filesfrom_f_in = f_in;
563 io_filesfrom_f_out = f_out;
564 alloc_xbuf(&ff_buf, 2048);
567 alloc_xbuf(&iconv_buf, 1024);
571 /* It's almost always an error to get an EOF when we're trying to read from the
572 * network, because the protocol is (for the most part) self-terminating.
574 * There is one case for the receiver when it is at the end of the transfer
575 * (hanging around reading any keep-alive packets that might come its way): if
576 * the sender dies before the generator's kill-signal comes through, we can end
577 * up here needing to loop until the kill-signal arrives. In this situation,
578 * kluge_around_eof will be < 0.
580 * There is another case for older protocol versions (< 24) where the module
581 * listing was not terminated, so we must ignore an EOF error in that case and
582 * exit. In this situation, kluge_around_eof will be > 0. */
583 static void whine_about_eof(int fd)
585 if (kluge_around_eof && fd == sock_f_in) {
587 if (kluge_around_eof > 0)
589 /* If we're still here after 10 seconds, exit with an error. */
590 for (i = 10*1000/20; i--; )
594 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
595 "(%.0f bytes received so far) [%s]\n",
596 (double)stats.total_read, who_am_i());
598 exit_cleanup(RERR_STREAMIO);
602 * Read from a socket with I/O timeout. return the number of bytes
603 * read. If no bytes can be read then exit, never return a number <= 0.
605 * TODO: If the remote shell connection fails, then current versions
606 * actually report an "unexpected EOF" error here. Since it's a
607 * fairly common mistake to try to use rsh when ssh is required, we
608 * should trap that: if we fail to read any data at all, we should
609 * give a better explanation. We can tell whether the connection has
610 * started by looking e.g. at whether the remote version is known yet.
612 static int read_timeout(int fd, char *buf, size_t len)
616 io_flush(FULL_FLUSH);
619 /* until we manage to read *something* */
628 if (io_filesfrom_f_out >= 0) {
630 if (ff_buf.len == 0) {
631 if (io_filesfrom_f_in >= 0) {
632 FD_SET(io_filesfrom_f_in, &r_fds);
633 new_fd = io_filesfrom_f_in;
635 io_filesfrom_f_out = -1;
639 FD_SET(io_filesfrom_f_out, &w_fds);
640 new_fd = io_filesfrom_f_out;
646 tv.tv_sec = select_timeout;
651 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
654 if (errno == EBADF) {
655 defer_forwarding_messages = 0;
656 exit_cleanup(RERR_SOCKETIO);
662 if (io_filesfrom_f_out >= 0) {
664 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
665 int l = write(io_filesfrom_f_out,
666 ff_buf.buf + ff_buf.pos,
669 if (!(ff_buf.len -= l))
673 } else if (errno != EINTR) {
674 /* XXX should we complain? */
675 io_filesfrom_f_out = -1;
678 } else if (io_filesfrom_f_in >= 0) {
679 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
681 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
683 xbuf *ibuf = &ff_buf;
685 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
687 if (l == 0 || errno != EINTR) {
688 /* Send end-of-file marker */
689 memcpy(ff_buf.buf, "\0\0", 2);
690 ff_buf.len = ff_lastchar? 2 : 1;
692 io_filesfrom_f_in = -1;
696 if (filesfrom_convert) {
699 iconvbufs(ic_send, &iconv_buf, &ff_buf,
700 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
705 char *s = ff_buf.buf + l;
706 /* Transform CR and/or LF into '\0' */
707 while (s-- > ff_buf.buf) {
708 if (*s == '\n' || *s == '\r')
713 /* Last buf ended with a '\0', so don't
714 * let this buf start with one. */
715 while (l && ff_buf.buf[ff_buf.pos] == '\0')
721 char *f = ff_buf.buf + ff_buf.pos;
724 /* Eliminate any multi-'\0' runs. */
726 if (!(*t++ = *f++)) {
727 while (f != eob && !*f)
739 if (!FD_ISSET(fd, &r_fds))
742 n = read(fd, buf, len);
746 whine_about_eof(fd); /* Doesn't return. */
747 if (errno == EINTR || errno == EWOULDBLOCK
751 /* Don't write errors on a dead socket. */
752 if (fd == sock_f_in) {
753 io_end_multiplex_out();
754 rsyserr(FSOCKERR, errno, "read error");
756 rsyserr(FERROR, errno, "read error");
757 exit_cleanup(RERR_STREAMIO);
764 if (fd == sock_f_in && io_timeout)
765 last_io_in = time(NULL);
771 /* Read a line into the "buf" buffer. */
772 int read_line(int fd, char *buf, size_t bufsiz, int flags)
778 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
779 realloc_xbuf(&iconv_buf, bufsiz + 1024);
784 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
788 eob = s + bufsiz - 1;
790 cnt = read(fd, &ch, 1);
791 if (cnt < 0 && (errno == EWOULDBLOCK
792 || errno == EINTR || errno == EAGAIN)) {
799 tv.tv_sec = select_timeout;
801 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
803 /*if (FD_ISSET(fd, &e_fds))
804 rprintf(FINFO, "select exception on fd %d\n", fd); */
809 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
810 /* Skip empty lines if dumping comments. */
811 if (flags & RL_DUMP_COMMENTS && s == buf)
820 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
824 if (flags & RL_CONVERT) {
826 INIT_XBUF(outbuf, buf, 0, bufsiz);
828 iconv_buf.len = s - iconv_buf.buf;
829 iconvbufs(ic_recv, &iconv_buf, &outbuf,
830 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
831 outbuf.buf[outbuf.len] = '\0';
839 int read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
840 char ***argv_p, int *argc_p, char **request_p)
842 int maxargs = MAX_ARGS;
846 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
849 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
852 if (!(argv = new_array(char *, maxargs)))
853 out_of_memory("read_args");
855 argv[argc++] = "rsyncd";
858 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
861 if (argc == maxargs) {
863 if (!(argv = realloc_array(argv, char *, maxargs)))
864 out_of_memory("read_args");
869 *request_p = strdup(buf);
873 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
875 glob_expand(buf, &argv, &argc, &maxargs);
877 if (!(p = strdup(buf)))
878 out_of_memory("read_args");
880 if (*p == '.' && p[1] == '\0')
888 return dot_pos ? dot_pos : argc;
891 int io_start_buffering_out(int f_out)
894 assert(f_out == iobuf_f_out);
897 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
898 out_of_memory("io_start_buffering_out");
904 int io_start_buffering_in(int f_in)
907 assert(f_in == iobuf_f_in);
910 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
911 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
912 out_of_memory("io_start_buffering_in");
917 void io_end_buffering_in(void)
924 iobuf_in_remaining = 0;
928 void io_end_buffering_out(void)
932 io_flush(FULL_FLUSH);
938 void maybe_flush_socket(int important)
940 if (iobuf_out && iobuf_out_cnt
941 && (important || time(NULL) - last_io_out >= 5))
942 io_flush(NORMAL_FLUSH);
945 void maybe_send_keepalive(void)
947 if (time(NULL) - last_io_out >= allowed_lull) {
948 if (!iobuf_out || !iobuf_out_cnt) {
949 if (protocol_version < 29)
950 return; /* there's nothing we can do */
951 if (protocol_version >= 30)
952 send_msg(MSG_NOOP, "", 0, 0);
954 write_int(sock_f_out, cur_flist->used);
955 write_shortint(sock_f_out, ITEM_IS_NEW);
959 io_flush(NORMAL_FLUSH);
963 void start_flist_forward(int f_in)
965 assert(iobuf_out != NULL);
966 assert(iobuf_f_out == msg_fd_out);
967 flist_forward_from = f_in;
970 void stop_flist_forward()
972 flist_forward_from = -1;
973 io_flush(FULL_FLUSH);
977 * Continue trying to read len bytes - don't return until len has been
980 static void read_loop(int fd, char *buf, size_t len)
983 int n = read_timeout(fd, buf, len);
991 * Read from the file descriptor handling multiplexing - return number
994 * Never returns <= 0.
996 static int readfd_unbuffered(int fd, char *buf, size_t len)
1000 char line[BIGPATHBUFLEN];
1002 if (!iobuf_in || fd != iobuf_f_in)
1003 return read_timeout(fd, buf, len);
1005 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1006 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1011 if (iobuf_in_remaining) {
1012 len = MIN(len, iobuf_in_remaining);
1013 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1014 iobuf_in_ndx += len;
1015 iobuf_in_remaining -= len;
1020 read_loop(fd, line, 4);
1021 tag = IVAL(line, 0);
1023 msg_bytes = tag & 0xFFFFFF;
1024 tag = (tag >> 24) - MPLEX_BASE;
1028 if (msg_bytes > iobuf_in_siz) {
1029 if (!(iobuf_in = realloc_array(iobuf_in, char,
1031 out_of_memory("readfd_unbuffered");
1032 iobuf_in_siz = msg_bytes;
1034 read_loop(fd, iobuf_in, msg_bytes);
1035 iobuf_in_remaining = msg_bytes;
1040 maybe_send_keepalive();
1045 read_loop(fd, line, msg_bytes);
1046 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1047 io_error |= IVAL(line, 0);
1050 if (msg_bytes >= sizeof line)
1053 if (ic_recv != (iconv_t)-1) {
1059 INIT_CONST_XBUF(outbuf, line);
1060 INIT_XBUF(inbuf, ibuf, 0, -1);
1063 inbuf.len = msg_bytes > sizeof ibuf
1064 ? sizeof ibuf : msg_bytes;
1065 read_loop(fd, inbuf.buf, inbuf.len);
1066 if (!(msg_bytes -= inbuf.len)
1067 && !ibuf[inbuf.len-1])
1068 inbuf.len--, add_null = 1;
1069 if (iconvbufs(ic_send, &inbuf, &outbuf,
1070 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1075 if (outbuf.len == outbuf.size)
1077 outbuf.buf[outbuf.len++] = '\0';
1079 msg_bytes = outbuf.len;
1082 read_loop(fd, line, msg_bytes);
1083 /* A directory name was sent with the trailing null */
1084 if (msg_bytes > 0 && !line[msg_bytes-1])
1085 log_delete(line, S_IFDIR);
1087 line[msg_bytes] = '\0';
1088 log_delete(line, S_IFREG);
1092 if (msg_bytes != 4) {
1094 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1095 tag, (long)msg_bytes, who_am_i());
1096 exit_cleanup(RERR_STREAMIO);
1098 read_loop(fd, line, msg_bytes);
1099 successful_send(IVAL(line, 0));
1104 read_loop(fd, line, msg_bytes);
1105 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1109 if (msg_bytes >= sizeof line) {
1112 "multiplexing overflow %d:%ld [%s]\n",
1113 tag, (long)msg_bytes, who_am_i());
1114 exit_cleanup(RERR_STREAMIO);
1116 read_loop(fd, line, msg_bytes);
1117 rwrite((enum logcode)tag, line, msg_bytes, 1);
1120 rprintf(FERROR, "unexpected tag %d [%s]\n",
1122 exit_cleanup(RERR_STREAMIO);
1126 if (iobuf_in_remaining == 0)
1127 io_flush(NORMAL_FLUSH);
1132 /* Do a buffered read from fd. Don't return until all N bytes have
1133 * been read. If all N can't be read then exit with an error. */
1134 static void readfd(int fd, char *buffer, size_t N)
1140 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1144 if (fd == write_batch_monitor_in) {
1145 if ((size_t)write(batch_fd, buffer, total) != total)
1146 exit_cleanup(RERR_FILEIO);
1149 if (fd == flist_forward_from)
1150 writefd(iobuf_f_out, buffer, total);
1152 if (fd == sock_f_in)
1153 stats.total_read += total;
1156 unsigned short read_shortint(int f)
1160 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1163 int32 read_int(int f)
1170 #if SIZEOF_INT32 > 4
1171 if (num & (int32)0x80000000)
1172 num |= ~(int32)0xffffffff;
1177 int32 read_varint(int f)
1187 readfd(f, (char*)&ch, 1);
1188 extra = int_byte_extra[ch / 4];
1190 uchar bit = ((uchar)1<<(8-extra));
1191 if (extra >= (int)sizeof u.b) {
1192 rprintf(FERROR, "Overflow in read_varint()\n");
1193 exit_cleanup(RERR_STREAMIO);
1195 readfd(f, u.b, extra);
1196 u.b[extra] = ch & (bit-1);
1199 #if CAREFUL_ALIGNMENT
1202 #if SIZEOF_INT32 > 4
1203 if (u.x & (int32)0x80000000)
1204 u.x |= ~(int32)0xffffffff;
1209 int64 read_varlong(int f, uchar min_bytes)
1218 #if SIZEOF_INT64 < 8
1223 readfd(f, b2, min_bytes);
1224 memcpy(u.b, b2+1, min_bytes-1);
1225 extra = int_byte_extra[CVAL(b2, 0) / 4];
1227 uchar bit = ((uchar)1<<(8-extra));
1228 if (min_bytes + extra > (int)sizeof u.b) {
1229 rprintf(FERROR, "Overflow in read_varlong()\n");
1230 exit_cleanup(RERR_STREAMIO);
1232 readfd(f, u.b + min_bytes - 1, extra);
1233 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1234 #if SIZEOF_INT64 < 8
1235 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1236 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1237 exit_cleanup(RERR_UNSUPPORTED);
1241 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1242 #if SIZEOF_INT64 < 8
1244 #elif CAREFUL_ALIGNMENT
1245 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1250 int64 read_longint(int f)
1252 #if SIZEOF_INT64 >= 8
1255 int32 num = read_int(f);
1257 if (num != (int32)0xffffffff)
1260 #if SIZEOF_INT64 < 8
1261 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1262 exit_cleanup(RERR_UNSUPPORTED);
1265 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1269 void read_buf(int f, char *buf, size_t len)
1274 void read_sbuf(int f, char *buf, size_t len)
1276 readfd(f, buf, len);
1280 uchar read_byte(int f)
1283 readfd(f, (char *)&c, 1);
1287 int read_vstring(int f, char *buf, int bufsize)
1289 int len = read_byte(f);
1292 len = (len & ~0x80) * 0x100 + read_byte(f);
1294 if (len >= bufsize) {
1295 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1301 readfd(f, buf, len);
1306 /* Populate a sum_struct with values from the socket. This is
1307 * called by both the sender and the receiver. */
1308 void read_sum_head(int f, struct sum_struct *sum)
1310 sum->count = read_int(f);
1311 if (sum->count < 0) {
1312 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1313 (long)sum->count, who_am_i());
1314 exit_cleanup(RERR_PROTOCOL);
1316 sum->blength = read_int(f);
1317 if (sum->blength < 0 || sum->blength > MAX_BLOCK_SIZE) {
1318 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1319 (long)sum->blength, who_am_i());
1320 exit_cleanup(RERR_PROTOCOL);
1322 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1323 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1324 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1325 sum->s2length, who_am_i());
1326 exit_cleanup(RERR_PROTOCOL);
1328 sum->remainder = read_int(f);
1329 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1330 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1331 (long)sum->remainder, who_am_i());
1332 exit_cleanup(RERR_PROTOCOL);
1336 /* Send the values from a sum_struct over the socket. Set sum to
1337 * NULL if there are no checksums to send. This is called by both
1338 * the generator and the sender. */
1339 void write_sum_head(int f, struct sum_struct *sum)
1341 static struct sum_struct null_sum;
1346 write_int(f, sum->count);
1347 write_int(f, sum->blength);
1348 if (protocol_version >= 27)
1349 write_int(f, sum->s2length);
1350 write_int(f, sum->remainder);
1354 * Sleep after writing to limit I/O bandwidth usage.
1356 * @todo Rather than sleeping after each write, it might be better to
1357 * use some kind of averaging. The current algorithm seems to always
1358 * use a bit less bandwidth than specified, because it doesn't make up
1359 * for slow periods. But arguably this is a feature. In addition, we
1360 * ought to take the time used to write the data into account.
1362 * During some phases of big transfers (file FOO is uptodate) this is
1363 * called with a small bytes_written every time. As the kernel has to
1364 * round small waits up to guarantee that we actually wait at least the
1365 * requested number of microseconds, this can become grossly inaccurate.
1366 * We therefore keep track of the bytes we've written over time and only
1367 * sleep when the accumulated delay is at least 1 tenth of a second.
1369 static void sleep_for_bwlimit(int bytes_written)
1371 static struct timeval prior_tv;
1372 static long total_written = 0;
1373 struct timeval tv, start_tv;
1374 long elapsed_usec, sleep_usec;
1376 #define ONE_SEC 1000000L /* # of microseconds in a second */
1378 if (!bwlimit_writemax)
1381 total_written += bytes_written;
1383 gettimeofday(&start_tv, NULL);
1384 if (prior_tv.tv_sec) {
1385 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1386 + (start_tv.tv_usec - prior_tv.tv_usec);
1387 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1388 if (total_written < 0)
1392 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1393 if (sleep_usec < ONE_SEC / 10) {
1394 prior_tv = start_tv;
1398 tv.tv_sec = sleep_usec / ONE_SEC;
1399 tv.tv_usec = sleep_usec % ONE_SEC;
1400 select(0, NULL, NULL, NULL, &tv);
1402 gettimeofday(&prior_tv, NULL);
1403 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1404 + (prior_tv.tv_usec - start_tv.tv_usec);
1405 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1408 /* Write len bytes to the file descriptor fd, looping as necessary to get
1409 * the job done and also (in certain circumstances) reading any data on
1410 * msg_fd_in to avoid deadlock.
1412 * This function underlies the multiplexing system. The body of the
1413 * application never calls this function directly. */
1414 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1416 size_t n, total = 0;
1417 fd_set w_fds, r_fds, e_fds;
1418 int maxfd, count, cnt, using_r_fds;
1423 defer_forwarding_messages++, defer_inc++;
1425 while (total < len) {
1432 if (msg_fd_in >= 0) {
1434 FD_SET(msg_fd_in, &r_fds);
1435 if (msg_fd_in > maxfd)
1441 tv.tv_sec = select_timeout;
1445 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1446 &w_fds, &e_fds, &tv);
1449 if (count < 0 && errno == EBADF)
1450 exit_cleanup(RERR_SOCKETIO);
1455 /*if (FD_ISSET(fd, &e_fds))
1456 rprintf(FINFO, "select exception on fd %d\n", fd); */
1458 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1461 if (!FD_ISSET(fd, &w_fds))
1465 if (bwlimit_writemax && n > bwlimit_writemax)
1466 n = bwlimit_writemax;
1467 cnt = write(fd, buf + total, n);
1473 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1479 /* Don't try to write errors back across the stream. */
1480 if (fd == sock_f_out)
1481 io_end_multiplex_out();
1482 /* Don't try to write errors down a failing msg pipe. */
1483 if (am_server && fd == msg_fd_out)
1484 exit_cleanup(RERR_STREAMIO);
1485 rsyserr(FERROR, errno,
1486 "writefd_unbuffered failed to write %ld bytes [%s]",
1487 (long)len, who_am_i());
1488 /* If the other side is sending us error messages, try
1489 * to grab any messages they sent before they died. */
1490 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1494 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1496 exit_cleanup(RERR_STREAMIO);
1500 defer_forwarding_messages++, defer_inc++;
1502 if (fd == sock_f_out) {
1503 if (io_timeout || am_generator)
1504 last_io_out = time(NULL);
1505 sleep_for_bwlimit(cnt);
1510 defer_inc -= defer_forwarding_keep;
1511 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1515 void io_flush(int flush_it_all)
1517 if (!iobuf_out_cnt || no_flush)
1520 if (io_multiplexing_out)
1521 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1523 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1526 if (flush_it_all && !defer_forwarding_messages)
1530 static void writefd(int fd, const char *buf, size_t len)
1532 if (fd == sock_f_out)
1533 stats.total_written += len;
1535 if (fd == write_batch_monitor_out) {
1536 if ((size_t)write(batch_fd, buf, len) != len)
1537 exit_cleanup(RERR_FILEIO);
1540 if (!iobuf_out || fd != iobuf_f_out) {
1541 writefd_unbuffered(fd, buf, len);
1546 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1548 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1554 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1555 io_flush(NORMAL_FLUSH);
1559 void write_shortint(int f, unsigned short x)
1563 b[1] = (char)(x >> 8);
1567 void write_int(int f, int32 x)
1574 void write_varint(int f, int32 x)
1582 while (cnt > 1 && b[cnt] == 0)
1584 bit = ((uchar)1<<(7-cnt+1));
1585 if (CVAL(b, cnt) >= bit) {
1589 *b = b[cnt] | ~(bit*2-1);
1596 void write_varlong(int f, int64 x, uchar min_bytes)
1603 #if SIZEOF_INT64 >= 8
1604 SIVAL(b, 5, x >> 32);
1606 if (x <= 0x7FFFFFFF && x >= 0)
1607 memset(b + 5, 0, 4);
1609 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1610 exit_cleanup(RERR_UNSUPPORTED);
1614 while (cnt > min_bytes && b[cnt] == 0)
1616 bit = ((uchar)1<<(7-cnt+min_bytes));
1617 if (CVAL(b, cnt) >= bit) {
1620 } else if (cnt > min_bytes)
1621 *b = b[cnt] | ~(bit*2-1);
1629 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1630 * 64-bit types on this platform.
1632 void write_longint(int f, int64 x)
1634 char b[12], * const s = b+4;
1637 if (x <= 0x7FFFFFFF && x >= 0) {
1642 #if SIZEOF_INT64 < 8
1643 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1644 exit_cleanup(RERR_UNSUPPORTED);
1647 SIVAL(s, 4, x >> 32);
1652 void write_buf(int f, const char *buf, size_t len)
1657 /** Write a string to the connection */
1658 void write_sbuf(int f, const char *buf)
1660 writefd(f, buf, strlen(buf));
1663 void write_byte(int f, uchar c)
1665 writefd(f, (char *)&c, 1);
1668 void write_vstring(int f, const char *str, int len)
1670 uchar lenbuf[3], *lb = lenbuf;
1675 "attempting to send over-long vstring (%d > %d)\n",
1677 exit_cleanup(RERR_PROTOCOL);
1679 *lb++ = len / 0x100 + 0x80;
1683 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1685 writefd(f, str, len);
1688 /* Send a file-list index using a byte-reduction method. */
1689 void write_ndx(int f, int32 ndx)
1691 static int32 prev_positive = -1, prev_negative = 1;
1692 int32 diff, cnt = 0;
1695 if (protocol_version < 30 || read_batch) {
1700 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1701 * negative nums as a positive after sending a leading 0xFF. */
1703 diff = ndx - prev_positive;
1704 prev_positive = ndx;
1705 } else if (ndx == NDX_DONE) {
1710 b[cnt++] = (char)0xFF;
1712 diff = ndx - prev_negative;
1713 prev_negative = ndx;
1716 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1717 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1718 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1719 if (diff < 0xFE && diff > 0)
1720 b[cnt++] = (char)diff;
1721 else if (diff < 0 || diff > 0x7FFF) {
1722 b[cnt++] = (char)0xFE;
1723 b[cnt++] = (char)((ndx >> 24) | 0x80);
1724 b[cnt++] = (char)ndx;
1725 b[cnt++] = (char)(ndx >> 8);
1726 b[cnt++] = (char)(ndx >> 16);
1728 b[cnt++] = (char)0xFE;
1729 b[cnt++] = (char)(diff >> 8);
1730 b[cnt++] = (char)diff;
1735 /* Receive a file-list index using a byte-reduction method. */
1736 int32 read_ndx(int f)
1738 static int32 prev_positive = -1, prev_negative = 1;
1739 int32 *prev_ptr, num;
1742 if (protocol_version < 30)
1746 if (CVAL(b, 0) == 0xFF) {
1748 prev_ptr = &prev_negative;
1749 } else if (CVAL(b, 0) == 0)
1752 prev_ptr = &prev_positive;
1753 if (CVAL(b, 0) == 0xFE) {
1755 if (CVAL(b, 0) & 0x80) {
1756 b[3] = CVAL(b, 0) & ~0x80;
1761 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1763 num = UVAL(b, 0) + *prev_ptr;
1765 if (prev_ptr == &prev_negative)
1770 /* Read a line of up to bufsiz-1 characters into buf. Strips
1771 * the (required) trailing newline and all carriage returns.
1772 * Returns 1 for success; 0 for I/O error or truncation. */
1773 int read_line_old(int f, char *buf, size_t bufsiz)
1775 bufsiz--; /* leave room for the null */
1776 while (bufsiz > 0) {
1778 read_buf(f, buf, 1);
1783 if (buf[0] != '\r') {
1792 void io_printf(int fd, const char *format, ...)
1795 char buf[BIGPATHBUFLEN];
1798 va_start(ap, format);
1799 len = vsnprintf(buf, sizeof buf, format, ap);
1803 exit_cleanup(RERR_STREAMIO);
1805 if (len > (int)sizeof buf) {
1806 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1807 exit_cleanup(RERR_STREAMIO);
1810 write_sbuf(fd, buf);
1813 /** Setup for multiplexing a MSG_* stream with the data stream. */
1814 void io_start_multiplex_out(void)
1816 io_flush(NORMAL_FLUSH);
1817 io_start_buffering_out(sock_f_out);
1818 io_multiplexing_out = 1;
1821 /** Setup for multiplexing a MSG_* stream with the data stream. */
1822 void io_start_multiplex_in(void)
1824 io_flush(NORMAL_FLUSH);
1825 io_start_buffering_in(sock_f_in);
1826 io_multiplexing_in = 1;
1829 /** Write an message to the multiplexed data stream. */
1830 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1832 if (!io_multiplexing_out)
1834 io_flush(NORMAL_FLUSH);
1835 stats.total_written += (len+4);
1836 mplex_write(sock_f_out, code, buf, len, convert);
1840 void io_end_multiplex_in(void)
1842 io_multiplexing_in = 0;
1843 io_end_buffering_in();
1846 /** Stop output multiplexing. */
1847 void io_end_multiplex_out(void)
1849 io_multiplexing_out = 0;
1850 io_end_buffering_out();
1853 void start_write_batch(int fd)
1855 /* Some communication has already taken place, but we don't
1856 * enable batch writing until here so that we can write a
1857 * canonical record of the communication even though the
1858 * actual communication so far depends on whether a daemon
1860 write_int(batch_fd, protocol_version);
1861 if (protocol_version >= 30)
1862 write_byte(batch_fd, inc_recurse);
1863 write_int(batch_fd, checksum_seed);
1866 write_batch_monitor_out = fd;
1868 write_batch_monitor_in = fd;
1871 void stop_write_batch(void)
1873 write_batch_monitor_out = -1;
1874 write_batch_monitor_in = -1;