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) 2003-2009 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_receiver;
44 extern int am_generator;
45 extern int inc_recurse;
50 extern int read_batch;
51 extern int csum_length;
52 extern int protect_args;
53 extern int checksum_seed;
54 extern int protocol_version;
55 extern int remove_source_files;
56 extern int preserve_hard_links;
57 extern struct stats stats;
58 extern struct file_list *cur_flist;
60 extern int filesfrom_convert;
61 extern iconv_t ic_send, ic_recv;
64 const char phase_unknown[] = "unknown";
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 check_timeout(void)
126 /* On the receiving side, the generator is now handling timeouts, so
127 * the receiver ignores them. Note that the am_receiver flag is not
128 * set until the receiver forks from the generator, so timeouts will be
129 * based on receiving data on the receiving side until that event. */
130 if (!io_timeout || am_receiver)
138 chk = MAX(last_io_out, last_io_in);
139 if (t - chk >= io_timeout) {
140 if (am_server || am_daemon)
141 exit_cleanup(RERR_TIMEOUT);
142 rprintf(FERROR, "[%s] io timeout after %d seconds -- exiting\n",
143 who_am_i(), (int)(t-chk));
144 exit_cleanup(RERR_TIMEOUT);
148 static void readfd(int fd, char *buffer, size_t N);
149 static void writefd(int fd, const char *buf, size_t len);
150 static void writefd_unbuffered(int fd, const char *buf, size_t len);
151 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
153 static flist_ndx_list redo_list, hlink_list;
155 struct msg_list_item {
156 struct msg_list_item *next;
162 struct msg_list_item *head, *tail;
165 static struct msg_list msg_queue;
167 static void got_flist_entry_status(enum festatus status, const char *buf)
169 int ndx = IVAL(buf, 0);
170 struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
172 if (remove_source_files) {
174 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
178 flist->in_progress--;
182 if (remove_source_files)
183 send_msg(MSG_SUCCESS, buf, 4, 0);
184 if (preserve_hard_links) {
185 struct file_struct *file = flist->files[ndx - flist->ndx_start];
186 if (F_IS_HLINKED(file)) {
187 flist_ndx_push(&hlink_list, ndx);
188 flist->in_progress++;
195 flist->in_progress++;
200 flist_ndx_push(&redo_list, ndx);
207 /* Note the fds used for the main socket (which might really be a pipe
208 * for a local transfer, but we can ignore that). */
209 void io_set_sock_fds(int f_in, int f_out)
215 void set_io_timeout(int secs)
219 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
220 select_timeout = SELECT_TIMEOUT;
222 select_timeout = io_timeout;
224 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
227 /* Setup the fd used to receive MSG_* messages. Only needed during the
228 * early stages of being a local sender (up through the sending of the
229 * file list) or when we're the generator (to fetch the messages from
231 void set_msg_fd_in(int fd)
236 /* Setup the fd used to send our MSG_* messages. Only needed when
237 * we're the receiver (to send our messages to the generator). */
238 void set_msg_fd_out(int fd)
241 set_nonblocking(msg_fd_out);
244 /* Add a message to the pending MSG_* list. */
245 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
247 struct msg_list_item *m;
248 int sz = len + 4 + sizeof m[0] - 1;
250 if (!(m = (struct msg_list_item *)new_array(char, sz)))
251 out_of_memory("msg_list_add");
253 m->convert = convert;
254 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
255 memcpy(m->buf + 4, buf, len);
263 static inline int flush_a_msg(int fd)
265 struct msg_list_item *m = msg_queue.head;
266 int len = IVAL(m->buf, 0) & 0xFFFFFF;
267 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
269 if (!(msg_queue.head = m->next))
270 msg_queue.tail = NULL;
272 defer_forwarding_messages++;
273 mplex_write(fd, tag, m->buf + 4, len, m->convert);
274 defer_forwarding_messages--;
281 static void msg_flush(void)
284 while (msg_queue.head && io_multiplexing_out)
285 stats.total_written += flush_a_msg(sock_f_out) + 4;
287 while (msg_queue.head)
288 (void)flush_a_msg(msg_fd_out);
292 static void check_for_d_option_error(const char *msg)
294 static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
299 || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
302 msg += sizeof REMOTE_OPTION_ERROR - 1;
303 if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
304 || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
307 for ( ; *msg != ':'; msg++) {
310 else if (*msg == 'e')
312 else if (strchr(rsync263_opts, *msg) == NULL)
318 "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
322 /* Read a message from the MSG_* fd and handle it. This is called either
323 * during the early stages of being a local sender (up through the sending
324 * of the file list) or when we're the generator (to fetch the messages
325 * from the receiver). */
326 static void read_msg_fd(void)
330 struct file_list *flist;
334 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
335 * to this routine from writefd_unbuffered(). */
338 defer_forwarding_messages++;
343 len = tag & 0xFFFFFF;
344 tag = (tag >> 24) - MPLEX_BASE;
348 if (len < 0 || len > 1 || !am_generator) {
350 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
351 tag, len, who_am_i(),
352 inc_recurse ? "/inc" : "");
353 exit_cleanup(RERR_STREAMIO);
356 readfd(fd, buf, len);
357 stats.total_read = read_varlong(fd, 3);
362 if (len != 4 || !am_generator)
365 got_flist_entry_status(FES_REDO, buf);
368 if (len != 4 || !am_generator || !inc_recurse)
371 /* Read extra file list from receiver. */
372 assert(iobuf_in != NULL);
373 assert(iobuf_f_in == fd);
375 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
376 who_am_i(), IVAL(buf,0));
378 flist = recv_file_list(fd);
379 flist->parent_ndx = IVAL(buf,0);
380 #ifdef SUPPORT_HARD_LINKS
381 if (preserve_hard_links)
382 match_hard_links(flist);
386 if (len != 0 || !am_generator || !inc_recurse)
393 readfd(fd, buf, len);
394 io_error |= IVAL(buf, 0);
397 if (len >= (int)sizeof buf || !am_generator)
399 readfd(fd, buf, len);
400 send_msg(MSG_DELETED, buf, len, 1);
403 if (len != 4 || !am_generator)
406 got_flist_entry_status(FES_SUCCESS, buf);
409 if (len != 4 || !am_generator)
412 got_flist_entry_status(FES_NO_SEND, buf);
414 case MSG_ERROR_SOCKET:
419 if (tag == MSG_ERROR_SOCKET)
420 io_end_multiplex_out();
432 rwrite((enum logcode)tag, buf, n, !am_generator);
437 rprintf(FERROR, "unknown message %d:%d [%s]\n",
438 tag, len, who_am_i());
439 exit_cleanup(RERR_STREAMIO);
444 if (!--defer_forwarding_messages && !no_flush)
448 /* This is used by the generator to limit how many file transfers can
449 * be active at once when --remove-source-files is specified. Without
450 * this, sender-side deletions were mostly happening at the end. */
451 void increment_active_files(int ndx, int itemizing, enum logcode code)
454 /* TODO: tune these limits? */
455 int limit = active_bytecnt >= 128*1024 ? 10 : 50;
456 if (active_filecnt < limit)
458 check_for_finished_files(itemizing, code, 0);
459 if (active_filecnt < limit)
462 io_flush(NORMAL_FLUSH);
468 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
471 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
472 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
474 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
478 /* We need to convert buf before doing anything else so that we
479 * can include the (converted) byte length in the message header. */
480 if (convert && ic_send != (iconv_t)-1) {
483 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
484 INIT_XBUF(inbuf, (char*)buf, len, -1);
486 iconvbufs(ic_send, &inbuf, &outbuf,
487 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
489 rprintf(FERROR, "overflowed conversion buffer in mplex_write");
490 exit_cleanup(RERR_UNSUPPORTED);
493 n = len = outbuf.len;
496 if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
497 n = 0; /* We'd rather do 2 writes than too much memcpy(). */
499 memcpy(buffer + 4, buf, n);
501 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
503 keep_defer_forwarding++; /* defer_forwarding_messages++ on return */
504 writefd_unbuffered(fd, buffer, n+4);
505 keep_defer_forwarding--;
508 writefd_unbuffered(fd, buf+n, len-n);
510 if (!--defer_forwarding_messages && !no_flush)
514 int send_msg(enum msgcode code, const char *buf, int len, int convert)
516 if (msg_fd_out < 0) {
517 if (!defer_forwarding_messages)
518 return io_multiplex_write(code, buf, len, convert);
519 if (!io_multiplexing_out)
521 msg_list_add(&msg_queue, code, buf, len, convert);
524 if (flist_forward_from >= 0)
525 msg_list_add(&msg_queue, code, buf, len, convert);
527 mplex_write(msg_fd_out, code, buf, len, convert);
531 void send_msg_int(enum msgcode code, int num)
534 SIVAL(numbuf, 0, num);
535 send_msg(code, numbuf, 4, 0);
538 void wait_for_receiver(void)
540 if (io_flush(NORMAL_FLUSH))
545 int get_redo_num(void)
547 return flist_ndx_pop(&redo_list);
550 int get_hlink_num(void)
552 return flist_ndx_pop(&hlink_list);
556 * When we're the receiver and we have a local --files-from list of names
557 * that needs to be sent over the socket to the sender, we have to do two
558 * things at the same time: send the sender a list of what files we're
559 * processing and read the incoming file+info list from the sender. We do
560 * this by augmenting the read_timeout() function to copy this data. It
561 * uses ff_buf to read a block of data from f_in (when it is ready, since
562 * it might be a pipe) and then blast it out f_out (when it is ready to
563 * receive more data).
565 void io_set_filesfrom_fds(int f_in, int f_out)
567 io_filesfrom_f_in = f_in;
568 io_filesfrom_f_out = f_out;
569 alloc_xbuf(&ff_buf, 2048);
572 alloc_xbuf(&iconv_buf, 1024);
576 /* It's almost always an error to get an EOF when we're trying to read from the
577 * network, because the protocol is (for the most part) self-terminating.
579 * There is one case for the receiver when it is at the end of the transfer
580 * (hanging around reading any keep-alive packets that might come its way): if
581 * the sender dies before the generator's kill-signal comes through, we can end
582 * up here needing to loop until the kill-signal arrives. In this situation,
583 * kluge_around_eof will be < 0.
585 * There is another case for older protocol versions (< 24) where the module
586 * listing was not terminated, so we must ignore an EOF error in that case and
587 * exit. In this situation, kluge_around_eof will be > 0. */
588 static void whine_about_eof(int fd)
590 if (kluge_around_eof && fd == sock_f_in) {
592 if (kluge_around_eof > 0)
594 /* If we're still here after 10 seconds, exit with an error. */
595 for (i = 10*1000/20; i--; )
599 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
600 "(%.0f bytes received so far) [%s]\n",
601 (double)stats.total_read, who_am_i());
603 exit_cleanup(RERR_STREAMIO);
607 * Read from a socket with I/O timeout. return the number of bytes
608 * read. If no bytes can be read then exit, never return a number <= 0.
610 * TODO: If the remote shell connection fails, then current versions
611 * actually report an "unexpected EOF" error here. Since it's a
612 * fairly common mistake to try to use rsh when ssh is required, we
613 * should trap that: if we fail to read any data at all, we should
614 * give a better explanation. We can tell whether the connection has
615 * started by looking e.g. at whether the remote version is known yet.
617 static int read_timeout(int fd, char *buf, size_t len)
621 io_flush(FULL_FLUSH);
624 /* until we manage to read *something* */
633 if (io_filesfrom_f_out >= 0) {
635 if (ff_buf.len == 0) {
636 if (io_filesfrom_f_in >= 0) {
637 FD_SET(io_filesfrom_f_in, &r_fds);
638 new_fd = io_filesfrom_f_in;
640 io_filesfrom_f_out = -1;
644 FD_SET(io_filesfrom_f_out, &w_fds);
645 new_fd = io_filesfrom_f_out;
651 tv.tv_sec = select_timeout;
656 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
659 if (errno == EBADF) {
660 defer_forwarding_messages = 0;
661 exit_cleanup(RERR_SOCKETIO);
667 if (io_filesfrom_f_out >= 0) {
669 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
670 int l = write(io_filesfrom_f_out,
671 ff_buf.buf + ff_buf.pos,
674 if (!(ff_buf.len -= l))
678 } else if (errno != EINTR) {
679 /* XXX should we complain? */
680 io_filesfrom_f_out = -1;
683 } else if (io_filesfrom_f_in >= 0) {
684 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
686 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
688 xbuf *ibuf = &ff_buf;
690 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
692 if (l == 0 || errno != EINTR) {
693 /* Send end-of-file marker */
694 memcpy(ff_buf.buf, "\0\0", 2);
695 ff_buf.len = ff_lastchar? 2 : 1;
697 io_filesfrom_f_in = -1;
701 if (filesfrom_convert) {
704 iconvbufs(ic_send, &iconv_buf, &ff_buf,
705 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
710 char *s = ff_buf.buf + l;
711 /* Transform CR and/or LF into '\0' */
712 while (s-- > ff_buf.buf) {
713 if (*s == '\n' || *s == '\r')
718 /* Last buf ended with a '\0', so don't
719 * let this buf start with one. */
720 while (l && ff_buf.buf[ff_buf.pos] == '\0')
726 char *f = ff_buf.buf + ff_buf.pos;
729 /* Eliminate any multi-'\0' runs. */
731 if (!(*t++ = *f++)) {
732 while (f != eob && !*f)
744 if (!FD_ISSET(fd, &r_fds))
747 n = read(fd, buf, len);
751 whine_about_eof(fd); /* Doesn't return. */
752 if (errno == EINTR || errno == EWOULDBLOCK
756 /* Don't write errors on a dead socket. */
757 if (fd == sock_f_in) {
758 io_end_multiplex_out();
759 rsyserr(FERROR_SOCKET, errno, "read error");
761 rsyserr(FERROR, errno, "read error");
762 exit_cleanup(RERR_STREAMIO);
769 if (fd == sock_f_in && io_timeout)
770 last_io_in = time(NULL);
776 /* Read a line into the "buf" buffer. */
777 int read_line(int fd, char *buf, size_t bufsiz, int flags)
783 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
784 realloc_xbuf(&iconv_buf, bufsiz + 1024);
789 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
793 eob = s + bufsiz - 1;
795 cnt = read(fd, &ch, 1);
796 if (cnt < 0 && (errno == EWOULDBLOCK
797 || errno == EINTR || errno == EAGAIN)) {
804 tv.tv_sec = select_timeout;
806 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
808 /*if (FD_ISSET(fd, &e_fds))
809 rprintf(FINFO, "select exception on fd %d\n", fd); */
814 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
815 /* Skip empty lines if dumping comments. */
816 if (flags & RL_DUMP_COMMENTS && s == buf)
825 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
829 if (flags & RL_CONVERT) {
831 INIT_XBUF(outbuf, buf, 0, bufsiz);
833 iconv_buf.len = s - iconv_buf.buf;
834 iconvbufs(ic_recv, &iconv_buf, &outbuf,
835 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
836 outbuf.buf[outbuf.len] = '\0';
844 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
845 char ***argv_p, int *argc_p, char **request_p)
847 int maxargs = MAX_ARGS;
851 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
854 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
857 if (!(argv = new_array(char *, maxargs)))
858 out_of_memory("read_args");
859 if (mod_name && !protect_args)
860 argv[argc++] = "rsyncd";
863 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
866 if (argc == maxargs-1) {
868 if (!(argv = realloc_array(argv, char *, maxargs)))
869 out_of_memory("read_args");
874 *request_p = strdup(buf);
878 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
880 glob_expand(buf, &argv, &argc, &maxargs);
882 if (!(p = strdup(buf)))
883 out_of_memory("read_args");
885 if (*p == '.' && p[1] == '\0')
891 glob_expand(NULL, NULL, NULL, NULL);
897 int io_start_buffering_out(int f_out)
900 assert(f_out == iobuf_f_out);
903 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
904 out_of_memory("io_start_buffering_out");
910 int io_start_buffering_in(int f_in)
913 assert(f_in == iobuf_f_in);
916 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
917 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
918 out_of_memory("io_start_buffering_in");
923 void io_end_buffering_in(void)
930 iobuf_in_remaining = 0;
934 void io_end_buffering_out(void)
938 io_flush(FULL_FLUSH);
944 void maybe_flush_socket(int important)
946 if (iobuf_out && iobuf_out_cnt
947 && (important || time(NULL) - last_io_out >= 5))
948 io_flush(NORMAL_FLUSH);
951 /* Older rsync versions used to send either a MSG_NOOP (protocol 30) or a
952 * raw-data-based keep-alive (protocol 29), both of which implied forwarding of
953 * the message through the sender. Since the new timeout method does not need
954 * any forwarding, we just send an empty MSG_DATA message, which works with all
955 * rsync versions. This avoids any message forwarding, and leaves the raw-data
956 * stream alone (since we can never be quite sure if that stream is in the
957 * right state for a keep-alive message). */
958 void maybe_send_keepalive(void)
960 if (time(NULL) - last_io_out >= allowed_lull) {
961 if (!iobuf_out || !iobuf_out_cnt)
962 send_msg(MSG_DATA, "", 0, 0);
964 io_flush(NORMAL_FLUSH);
968 void start_flist_forward(int f_in)
970 assert(iobuf_out != NULL);
971 assert(iobuf_f_out == msg_fd_out);
972 flist_forward_from = f_in;
975 void stop_flist_forward()
977 flist_forward_from = -1;
978 io_flush(FULL_FLUSH);
982 * Continue trying to read len bytes - don't return until len has been
985 static void read_loop(int fd, char *buf, size_t len)
988 int n = read_timeout(fd, buf, len);
996 * Read from the file descriptor handling multiplexing - return number
999 * Never returns <= 0.
1001 static int readfd_unbuffered(int fd, char *buf, size_t len)
1005 char line[BIGPATHBUFLEN];
1007 if (!iobuf_in || fd != iobuf_f_in)
1008 return read_timeout(fd, buf, len);
1010 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1011 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1016 if (iobuf_in_remaining) {
1017 len = MIN(len, iobuf_in_remaining);
1018 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1019 iobuf_in_ndx += len;
1020 iobuf_in_remaining -= len;
1025 read_loop(fd, line, 4);
1026 tag = IVAL(line, 0);
1028 msg_bytes = tag & 0xFFFFFF;
1029 tag = (tag >> 24) - MPLEX_BASE;
1033 if (msg_bytes > iobuf_in_siz) {
1034 if (!(iobuf_in = realloc_array(iobuf_in, char,
1036 out_of_memory("readfd_unbuffered");
1037 iobuf_in_siz = msg_bytes;
1039 read_loop(fd, iobuf_in, msg_bytes);
1040 iobuf_in_remaining = msg_bytes;
1044 /* Support protocol-30 keep-alive method. */
1048 maybe_send_keepalive();
1053 read_loop(fd, line, msg_bytes);
1054 send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1055 io_error |= IVAL(line, 0);
1058 if (msg_bytes >= sizeof line)
1061 if (ic_recv != (iconv_t)-1) {
1067 INIT_CONST_XBUF(outbuf, line);
1068 INIT_XBUF(inbuf, ibuf, 0, -1);
1071 inbuf.len = msg_bytes > sizeof ibuf
1072 ? sizeof ibuf : msg_bytes;
1073 read_loop(fd, inbuf.buf, inbuf.len);
1074 if (!(msg_bytes -= inbuf.len)
1075 && !ibuf[inbuf.len-1])
1076 inbuf.len--, add_null = 1;
1077 if (iconvbufs(ic_send, &inbuf, &outbuf,
1078 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1083 if (outbuf.len == outbuf.size)
1085 outbuf.buf[outbuf.len++] = '\0';
1087 msg_bytes = outbuf.len;
1090 read_loop(fd, line, msg_bytes);
1091 /* A directory name was sent with the trailing null */
1092 if (msg_bytes > 0 && !line[msg_bytes-1])
1093 log_delete(line, S_IFDIR);
1095 line[msg_bytes] = '\0';
1096 log_delete(line, S_IFREG);
1100 if (msg_bytes != 4) {
1102 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1103 tag, (long)msg_bytes, who_am_i());
1104 exit_cleanup(RERR_STREAMIO);
1106 read_loop(fd, line, msg_bytes);
1107 successful_send(IVAL(line, 0));
1112 read_loop(fd, line, msg_bytes);
1113 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1117 case MSG_ERROR_XFER:
1119 if (msg_bytes >= sizeof line) {
1122 "multiplexing overflow %d:%ld [%s]\n",
1123 tag, (long)msg_bytes, who_am_i());
1124 exit_cleanup(RERR_STREAMIO);
1126 read_loop(fd, line, msg_bytes);
1127 rwrite((enum logcode)tag, line, msg_bytes, 1);
1128 if (first_message) {
1129 if (list_only && !am_sender && tag == 1) {
1130 line[msg_bytes] = '\0';
1131 check_for_d_option_error(line);
1137 rprintf(FERROR, "unexpected tag %d [%s]\n",
1139 exit_cleanup(RERR_STREAMIO);
1143 if (iobuf_in_remaining == 0)
1144 io_flush(NORMAL_FLUSH);
1149 /* Do a buffered read from fd. Don't return until all N bytes have
1150 * been read. If all N can't be read then exit with an error. */
1151 static void readfd(int fd, char *buffer, size_t N)
1157 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1161 if (fd == write_batch_monitor_in) {
1162 if ((size_t)write(batch_fd, buffer, total) != total)
1163 exit_cleanup(RERR_FILEIO);
1166 if (fd == flist_forward_from)
1167 writefd(iobuf_f_out, buffer, total);
1169 if (fd == sock_f_in)
1170 stats.total_read += total;
1173 unsigned short read_shortint(int f)
1177 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1180 int32 read_int(int f)
1187 #if SIZEOF_INT32 > 4
1188 if (num & (int32)0x80000000)
1189 num |= ~(int32)0xffffffff;
1194 int32 read_varint(int f)
1204 readfd(f, (char*)&ch, 1);
1205 extra = int_byte_extra[ch / 4];
1207 uchar bit = ((uchar)1<<(8-extra));
1208 if (extra >= (int)sizeof u.b) {
1209 rprintf(FERROR, "Overflow in read_varint()\n");
1210 exit_cleanup(RERR_STREAMIO);
1212 readfd(f, u.b, extra);
1213 u.b[extra] = ch & (bit-1);
1216 #if CAREFUL_ALIGNMENT
1219 #if SIZEOF_INT32 > 4
1220 if (u.x & (int32)0x80000000)
1221 u.x |= ~(int32)0xffffffff;
1226 int64 read_varlong(int f, uchar min_bytes)
1235 #if SIZEOF_INT64 < 8
1240 readfd(f, b2, min_bytes);
1241 memcpy(u.b, b2+1, min_bytes-1);
1242 extra = int_byte_extra[CVAL(b2, 0) / 4];
1244 uchar bit = ((uchar)1<<(8-extra));
1245 if (min_bytes + extra > (int)sizeof u.b) {
1246 rprintf(FERROR, "Overflow in read_varlong()\n");
1247 exit_cleanup(RERR_STREAMIO);
1249 readfd(f, u.b + min_bytes - 1, extra);
1250 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1251 #if SIZEOF_INT64 < 8
1252 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1253 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1254 exit_cleanup(RERR_UNSUPPORTED);
1258 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1259 #if SIZEOF_INT64 < 8
1261 #elif CAREFUL_ALIGNMENT
1262 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1267 int64 read_longint(int f)
1269 #if SIZEOF_INT64 >= 8
1272 int32 num = read_int(f);
1274 if (num != (int32)0xffffffff)
1277 #if SIZEOF_INT64 < 8
1278 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1279 exit_cleanup(RERR_UNSUPPORTED);
1282 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1286 void read_buf(int f, char *buf, size_t len)
1291 void read_sbuf(int f, char *buf, size_t len)
1293 readfd(f, buf, len);
1297 uchar read_byte(int f)
1300 readfd(f, (char *)&c, 1);
1304 int read_vstring(int f, char *buf, int bufsize)
1306 int len = read_byte(f);
1309 len = (len & ~0x80) * 0x100 + read_byte(f);
1311 if (len >= bufsize) {
1312 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1318 readfd(f, buf, len);
1323 /* Populate a sum_struct with values from the socket. This is
1324 * called by both the sender and the receiver. */
1325 void read_sum_head(int f, struct sum_struct *sum)
1327 int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1328 sum->count = read_int(f);
1329 if (sum->count < 0) {
1330 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1331 (long)sum->count, who_am_i());
1332 exit_cleanup(RERR_PROTOCOL);
1334 sum->blength = read_int(f);
1335 if (sum->blength < 0 || sum->blength > max_blength) {
1336 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1337 (long)sum->blength, who_am_i());
1338 exit_cleanup(RERR_PROTOCOL);
1340 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1341 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1342 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1343 sum->s2length, who_am_i());
1344 exit_cleanup(RERR_PROTOCOL);
1346 sum->remainder = read_int(f);
1347 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1348 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1349 (long)sum->remainder, who_am_i());
1350 exit_cleanup(RERR_PROTOCOL);
1354 /* Send the values from a sum_struct over the socket. Set sum to
1355 * NULL if there are no checksums to send. This is called by both
1356 * the generator and the sender. */
1357 void write_sum_head(int f, struct sum_struct *sum)
1359 static struct sum_struct null_sum;
1364 write_int(f, sum->count);
1365 write_int(f, sum->blength);
1366 if (protocol_version >= 27)
1367 write_int(f, sum->s2length);
1368 write_int(f, sum->remainder);
1372 * Sleep after writing to limit I/O bandwidth usage.
1374 * @todo Rather than sleeping after each write, it might be better to
1375 * use some kind of averaging. The current algorithm seems to always
1376 * use a bit less bandwidth than specified, because it doesn't make up
1377 * for slow periods. But arguably this is a feature. In addition, we
1378 * ought to take the time used to write the data into account.
1380 * During some phases of big transfers (file FOO is uptodate) this is
1381 * called with a small bytes_written every time. As the kernel has to
1382 * round small waits up to guarantee that we actually wait at least the
1383 * requested number of microseconds, this can become grossly inaccurate.
1384 * We therefore keep track of the bytes we've written over time and only
1385 * sleep when the accumulated delay is at least 1 tenth of a second.
1387 static void sleep_for_bwlimit(int bytes_written)
1389 static struct timeval prior_tv;
1390 static long total_written = 0;
1391 struct timeval tv, start_tv;
1392 long elapsed_usec, sleep_usec;
1394 #define ONE_SEC 1000000L /* # of microseconds in a second */
1396 if (!bwlimit_writemax)
1399 total_written += bytes_written;
1401 gettimeofday(&start_tv, NULL);
1402 if (prior_tv.tv_sec) {
1403 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1404 + (start_tv.tv_usec - prior_tv.tv_usec);
1405 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1406 if (total_written < 0)
1410 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1411 if (sleep_usec < ONE_SEC / 10) {
1412 prior_tv = start_tv;
1416 tv.tv_sec = sleep_usec / ONE_SEC;
1417 tv.tv_usec = sleep_usec % ONE_SEC;
1418 select(0, NULL, NULL, NULL, &tv);
1420 gettimeofday(&prior_tv, NULL);
1421 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1422 + (prior_tv.tv_usec - start_tv.tv_usec);
1423 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1426 static const char *what_fd_is(int fd)
1428 static char buf[20];
1430 if (fd == sock_f_out)
1432 else if (fd == msg_fd_out)
1433 return "message fd";
1434 else if (fd == batch_fd)
1435 return "batch file";
1437 snprintf(buf, sizeof buf, "fd %d", fd);
1442 /* Write len bytes to the file descriptor fd, looping as necessary to get
1443 * the job done and also (in certain circumstances) reading any data on
1444 * msg_fd_in to avoid deadlock.
1446 * This function underlies the multiplexing system. The body of the
1447 * application never calls this function directly. */
1448 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1450 size_t n, total = 0;
1451 fd_set w_fds, r_fds, e_fds;
1452 int maxfd, count, cnt, using_r_fds;
1457 defer_forwarding_messages++, defer_inc++;
1459 while (total < len) {
1466 if (msg_fd_in >= 0) {
1468 FD_SET(msg_fd_in, &r_fds);
1469 if (msg_fd_in > maxfd)
1475 tv.tv_sec = select_timeout;
1479 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1480 &w_fds, &e_fds, &tv);
1483 if (count < 0 && errno == EBADF)
1484 exit_cleanup(RERR_SOCKETIO);
1489 /*if (FD_ISSET(fd, &e_fds))
1490 rprintf(FINFO, "select exception on fd %d\n", fd); */
1492 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1495 if (!FD_ISSET(fd, &w_fds))
1499 if (bwlimit_writemax && n > bwlimit_writemax)
1500 n = bwlimit_writemax;
1501 cnt = write(fd, buf + total, n);
1507 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1513 /* Don't try to write errors back across the stream. */
1514 if (fd == sock_f_out)
1515 io_end_multiplex_out();
1516 /* Don't try to write errors down a failing msg pipe. */
1517 if (am_server && fd == msg_fd_out)
1518 exit_cleanup(RERR_STREAMIO);
1519 rsyserr(FERROR, errno,
1520 "writefd_unbuffered failed to write %ld bytes to %s [%s]",
1521 (long)len, what_fd_is(fd), who_am_i());
1522 /* If the other side is sending us error messages, try
1523 * to grab any messages they sent before they died. */
1524 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1527 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1529 exit_cleanup(RERR_STREAMIO);
1533 defer_forwarding_messages++, defer_inc++;
1535 if (fd == sock_f_out) {
1536 if (io_timeout || am_generator)
1537 last_io_out = time(NULL);
1538 sleep_for_bwlimit(cnt);
1543 if (keep_defer_forwarding)
1545 if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1549 int io_flush(int flush_it_all)
1551 int flushed_something = 0;
1556 if (iobuf_out_cnt) {
1557 if (io_multiplexing_out)
1558 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1560 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1562 flushed_something = 1;
1565 if (flush_it_all && !defer_forwarding_messages && msg_queue.head) {
1567 flushed_something = 1;
1570 return flushed_something;
1573 static void writefd(int fd, const char *buf, size_t len)
1575 if (fd == sock_f_out)
1576 stats.total_written += len;
1578 if (fd == write_batch_monitor_out)
1579 writefd_unbuffered(batch_fd, buf, len);
1581 if (!iobuf_out || fd != iobuf_f_out) {
1582 writefd_unbuffered(fd, buf, len);
1587 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1589 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1595 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1596 io_flush(NORMAL_FLUSH);
1600 void write_shortint(int f, unsigned short x)
1604 b[1] = (char)(x >> 8);
1608 void write_int(int f, int32 x)
1615 void write_varint(int f, int32 x)
1623 while (cnt > 1 && b[cnt] == 0)
1625 bit = ((uchar)1<<(7-cnt+1));
1626 if (CVAL(b, cnt) >= bit) {
1630 *b = b[cnt] | ~(bit*2-1);
1637 void write_varlong(int f, int64 x, uchar min_bytes)
1644 #if SIZEOF_INT64 >= 8
1645 SIVAL(b, 5, x >> 32);
1647 if (x <= 0x7FFFFFFF && x >= 0)
1648 memset(b + 5, 0, 4);
1650 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1651 exit_cleanup(RERR_UNSUPPORTED);
1655 while (cnt > min_bytes && b[cnt] == 0)
1657 bit = ((uchar)1<<(7-cnt+min_bytes));
1658 if (CVAL(b, cnt) >= bit) {
1661 } else if (cnt > min_bytes)
1662 *b = b[cnt] | ~(bit*2-1);
1670 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1671 * 64-bit types on this platform.
1673 void write_longint(int f, int64 x)
1675 char b[12], * const s = b+4;
1678 if (x <= 0x7FFFFFFF && x >= 0) {
1683 #if SIZEOF_INT64 < 8
1684 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1685 exit_cleanup(RERR_UNSUPPORTED);
1688 SIVAL(s, 4, x >> 32);
1693 void write_buf(int f, const char *buf, size_t len)
1698 /** Write a string to the connection */
1699 void write_sbuf(int f, const char *buf)
1701 writefd(f, buf, strlen(buf));
1704 void write_byte(int f, uchar c)
1706 writefd(f, (char *)&c, 1);
1709 void write_vstring(int f, const char *str, int len)
1711 uchar lenbuf[3], *lb = lenbuf;
1716 "attempting to send over-long vstring (%d > %d)\n",
1718 exit_cleanup(RERR_PROTOCOL);
1720 *lb++ = len / 0x100 + 0x80;
1724 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1726 writefd(f, str, len);
1729 /* Send a file-list index using a byte-reduction method. */
1730 void write_ndx(int f, int32 ndx)
1732 static int32 prev_positive = -1, prev_negative = 1;
1733 int32 diff, cnt = 0;
1736 if (protocol_version < 30 || read_batch) {
1741 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1742 * negative nums as a positive after sending a leading 0xFF. */
1744 diff = ndx - prev_positive;
1745 prev_positive = ndx;
1746 } else if (ndx == NDX_DONE) {
1751 b[cnt++] = (char)0xFF;
1753 diff = ndx - prev_negative;
1754 prev_negative = ndx;
1757 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1758 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1759 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1760 if (diff < 0xFE && diff > 0)
1761 b[cnt++] = (char)diff;
1762 else if (diff < 0 || diff > 0x7FFF) {
1763 b[cnt++] = (char)0xFE;
1764 b[cnt++] = (char)((ndx >> 24) | 0x80);
1765 b[cnt++] = (char)ndx;
1766 b[cnt++] = (char)(ndx >> 8);
1767 b[cnt++] = (char)(ndx >> 16);
1769 b[cnt++] = (char)0xFE;
1770 b[cnt++] = (char)(diff >> 8);
1771 b[cnt++] = (char)diff;
1776 /* Receive a file-list index using a byte-reduction method. */
1777 int32 read_ndx(int f)
1779 static int32 prev_positive = -1, prev_negative = 1;
1780 int32 *prev_ptr, num;
1783 if (protocol_version < 30)
1787 if (CVAL(b, 0) == 0xFF) {
1789 prev_ptr = &prev_negative;
1790 } else if (CVAL(b, 0) == 0)
1793 prev_ptr = &prev_positive;
1794 if (CVAL(b, 0) == 0xFE) {
1796 if (CVAL(b, 0) & 0x80) {
1797 b[3] = CVAL(b, 0) & ~0x80;
1802 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1804 num = UVAL(b, 0) + *prev_ptr;
1806 if (prev_ptr == &prev_negative)
1811 /* Read a line of up to bufsiz-1 characters into buf. Strips
1812 * the (required) trailing newline and all carriage returns.
1813 * Returns 1 for success; 0 for I/O error or truncation. */
1814 int read_line_old(int f, char *buf, size_t bufsiz)
1816 bufsiz--; /* leave room for the null */
1817 while (bufsiz > 0) {
1819 read_buf(f, buf, 1);
1824 if (buf[0] != '\r') {
1833 void io_printf(int fd, const char *format, ...)
1836 char buf[BIGPATHBUFLEN];
1839 va_start(ap, format);
1840 len = vsnprintf(buf, sizeof buf, format, ap);
1844 exit_cleanup(RERR_STREAMIO);
1846 if (len > (int)sizeof buf) {
1847 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1848 exit_cleanup(RERR_STREAMIO);
1851 write_sbuf(fd, buf);
1854 /** Setup for multiplexing a MSG_* stream with the data stream. */
1855 void io_start_multiplex_out(void)
1857 io_flush(NORMAL_FLUSH);
1858 io_start_buffering_out(sock_f_out);
1859 io_multiplexing_out = 1;
1862 /** Setup for multiplexing a MSG_* stream with the data stream. */
1863 void io_start_multiplex_in(void)
1865 io_flush(NORMAL_FLUSH);
1866 io_start_buffering_in(sock_f_in);
1867 io_multiplexing_in = 1;
1870 /** Write an message to the multiplexed data stream. */
1871 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1873 if (!io_multiplexing_out)
1875 io_flush(NORMAL_FLUSH);
1876 stats.total_written += (len+4);
1877 mplex_write(sock_f_out, code, buf, len, convert);
1881 void io_end_multiplex_in(void)
1883 io_multiplexing_in = 0;
1884 io_end_buffering_in();
1887 /** Stop output multiplexing. */
1888 void io_end_multiplex_out(void)
1890 io_multiplexing_out = 0;
1891 io_end_buffering_out();
1894 void start_write_batch(int fd)
1896 /* Some communication has already taken place, but we don't
1897 * enable batch writing until here so that we can write a
1898 * canonical record of the communication even though the
1899 * actual communication so far depends on whether a daemon
1901 write_int(batch_fd, protocol_version);
1902 if (protocol_version >= 30)
1903 write_byte(batch_fd, inc_recurse);
1904 write_int(batch_fd, checksum_seed);
1907 write_batch_monitor_out = fd;
1909 write_batch_monitor_in = fd;
1912 void stop_write_batch(void)
1914 write_batch_monitor_out = -1;
1915 write_batch_monitor_in = -1;