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;
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);
184 assert(ndx >= flist->ndx_start);
186 if (remove_source_files) {
188 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
192 flist->in_progress--;
196 if (remove_source_files)
197 send_msg(MSG_SUCCESS, buf, 4, 0);
198 if (preserve_hard_links) {
199 struct file_struct *file = flist->files[ndx - flist->ndx_start];
200 if (F_IS_HLINKED(file))
201 flist_ndx_push(&hlink_list, ndx);
207 flist_ndx_push(&redo_list, ndx);
214 static void check_timeout(void)
218 if (!io_timeout || ignore_timeout)
222 last_io_in = time(NULL);
228 if (t - last_io_in >= io_timeout) {
229 if (!am_server && !am_daemon) {
230 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
231 (int)(t-last_io_in));
233 exit_cleanup(RERR_TIMEOUT);
237 /* Note the fds used for the main socket (which might really be a pipe
238 * for a local transfer, but we can ignore that). */
239 void io_set_sock_fds(int f_in, int f_out)
245 void set_io_timeout(int secs)
249 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
250 select_timeout = SELECT_TIMEOUT;
252 select_timeout = io_timeout;
254 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
257 /* Setup the fd used to receive MSG_* messages. Only needed during the
258 * early stages of being a local sender (up through the sending of the
259 * file list) or when we're the generator (to fetch the messages from
261 void set_msg_fd_in(int fd)
266 /* Setup the fd used to send our MSG_* messages. Only needed when
267 * we're the receiver (to send our messages to the generator). */
268 void set_msg_fd_out(int fd)
271 set_nonblocking(msg_fd_out);
274 /* Add a message to the pending MSG_* list. */
275 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
277 struct msg_list_item *m;
278 int sz = len + 4 + sizeof m[0] - 1;
280 if (!(m = (struct msg_list_item *)new_array(char, sz)))
281 out_of_memory("msg_list_add");
283 m->convert = convert;
284 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
285 memcpy(m->buf + 4, buf, len);
293 static void msg_flush(void)
296 while (msg_queue.head && io_multiplexing_out) {
297 struct msg_list_item *m = msg_queue.head;
298 int len = IVAL(m->buf, 0) & 0xFFFFFF;
299 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
300 if (!(msg_queue.head = m->next))
301 msg_queue.tail = NULL;
302 stats.total_written += len + 4;
303 defer_forwarding_messages++;
304 mplex_write(sock_f_out, tag, m->buf + 4, len, m->convert);
305 defer_forwarding_messages--;
309 while (msg_queue.head) {
310 struct msg_list_item *m = msg_queue.head;
311 int len = IVAL(m->buf, 0) & 0xFFFFFF;
312 int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
313 if (!(msg_queue.head = m->next))
314 msg_queue.tail = NULL;
315 defer_forwarding_messages++;
316 mplex_write(msg_fd_out, tag, m->buf + 4, len, m->convert);
317 defer_forwarding_messages--;
323 /* Read a message from the MSG_* fd and handle it. This is called either
324 * during the early stages of being a local sender (up through the sending
325 * of the file list) or when we're the generator (to fetch the messages
326 * from the receiver). */
327 static void read_msg_fd(void)
331 struct file_list *flist;
335 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
336 * to this routine from writefd_unbuffered(). */
339 defer_forwarding_messages++;
344 len = tag & 0xFFFFFF;
345 tag = (tag >> 24) - MPLEX_BASE;
349 if (len < 0 || len > 1 || !am_generator) {
351 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
352 tag, len, who_am_i(),
353 inc_recurse ? "/inc" : "");
354 exit_cleanup(RERR_STREAMIO);
357 readfd(fd, buf, len);
358 stats.total_read = read_varlong(fd, 3);
363 if (len != 4 || !am_generator)
366 got_flist_entry_status(FES_REDO, buf);
369 if (len != 4 || !am_generator || !inc_recurse)
372 /* Read extra file list from receiver. */
373 assert(iobuf_in != NULL);
374 assert(iobuf_f_in == fd);
376 rprintf(FINFO, "[%s] receiving flist for dir %d\n",
377 who_am_i(), IVAL(buf,0));
379 flist = recv_file_list(fd);
380 flist->parent_ndx = IVAL(buf,0);
381 #ifdef SUPPORT_HARD_LINKS
382 if (preserve_hard_links)
383 match_hard_links(flist);
387 if (len != 0 || !am_generator || !inc_recurse)
392 if (len >= (int)sizeof buf || !am_generator)
394 readfd(fd, buf, len);
395 send_msg(MSG_DELETED, buf, len, 1);
398 if (len != 4 || !am_generator)
401 got_flist_entry_status(FES_SUCCESS, buf);
404 if (len != 4 || !am_generator)
407 got_flist_entry_status(FES_NO_SEND, buf);
413 if (tag == MSG_SOCKERR)
414 io_end_multiplex_out();
424 rwrite((enum logcode)tag, buf, n, !am_generator);
429 rprintf(FERROR, "unknown message %d:%d [%s]\n",
430 tag, len, who_am_i());
431 exit_cleanup(RERR_STREAMIO);
436 if (!--defer_forwarding_messages)
440 /* This is used by the generator to limit how many file transfers can
441 * be active at once when --remove-source-files is specified. Without
442 * this, sender-side deletions were mostly happening at the end. */
443 void increment_active_files(int ndx, int itemizing, enum logcode code)
445 /* TODO: tune these limits? */
446 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
447 check_for_finished_files(itemizing, code, 0);
449 io_flush(NORMAL_FLUSH);
455 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
458 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
459 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
461 char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
464 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
467 if (convert && ic_send == (iconv_t)-1)
471 if (convert || n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
474 memcpy(buffer + 4, buf, n);
476 writefd_unbuffered(fd, buffer, n+4);
485 INIT_CONST_XBUF(outbuf, buffer);
486 INIT_XBUF(inbuf, (char*)buf, len, -1);
488 defer_forwarding_messages++;
490 iconvbufs(ic_send, &inbuf, &outbuf,
491 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
492 writefd_unbuffered(fd, outbuf.buf, outbuf.len);
494 if (!--defer_forwarding_messages)
499 defer_forwarding_messages++;
500 writefd_unbuffered(fd, buf, len);
501 if (!--defer_forwarding_messages)
506 int send_msg(enum msgcode code, const char *buf, int len, int convert)
508 if (msg_fd_out < 0) {
509 if (!defer_forwarding_messages)
510 return io_multiplex_write(code, buf, len, convert);
511 if (!io_multiplexing_out)
513 msg_list_add(&msg_queue, code, buf, len, convert);
516 if (flist_forward_from >= 0)
517 msg_list_add(&msg_queue, code, buf, len, convert);
519 mplex_write(msg_fd_out, code, buf, len, convert);
523 void send_msg_int(enum msgcode code, int num)
526 SIVAL(numbuf, 0, num);
527 send_msg(code, numbuf, 4, 0);
530 void wait_for_receiver(void)
533 io_flush(NORMAL_FLUSH);
538 int get_redo_num(void)
540 return flist_ndx_pop(&redo_list);
543 int get_hlink_num(void)
545 return flist_ndx_pop(&hlink_list);
549 * When we're the receiver and we have a local --files-from list of names
550 * that needs to be sent over the socket to the sender, we have to do two
551 * things at the same time: send the sender a list of what files we're
552 * processing and read the incoming file+info list from the sender. We do
553 * this by augmenting the read_timeout() function to copy this data. It
554 * uses ff_buf to read a block of data from f_in (when it is ready, since
555 * it might be a pipe) and then blast it out f_out (when it is ready to
556 * receive more data).
558 void io_set_filesfrom_fds(int f_in, int f_out)
560 io_filesfrom_f_in = f_in;
561 io_filesfrom_f_out = f_out;
562 alloc_xbuf(&ff_buf, 2048);
565 alloc_xbuf(&iconv_buf, 1024);
569 /* It's almost always an error to get an EOF when we're trying to read from the
570 * network, because the protocol is (for the most part) self-terminating.
572 * There is one case for the receiver when it is at the end of the transfer
573 * (hanging around reading any keep-alive packets that might come its way): if
574 * the sender dies before the generator's kill-signal comes through, we can end
575 * up here needing to loop until the kill-signal arrives. In this situation,
576 * kluge_around_eof will be < 0.
578 * There is another case for older protocol versions (< 24) where the module
579 * listing was not terminated, so we must ignore an EOF error in that case and
580 * exit. In this situation, kluge_around_eof will be > 0. */
581 static void whine_about_eof(int fd)
583 if (kluge_around_eof && fd == sock_f_in) {
585 if (kluge_around_eof > 0)
587 /* If we're still here after 10 seconds, exit with an error. */
588 for (i = 10*1000/20; i--; )
592 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
593 "(%.0f bytes received so far) [%s]\n",
594 (double)stats.total_read, who_am_i());
596 exit_cleanup(RERR_STREAMIO);
600 * Read from a socket with I/O timeout. return the number of bytes
601 * read. If no bytes can be read then exit, never return a number <= 0.
603 * TODO: If the remote shell connection fails, then current versions
604 * actually report an "unexpected EOF" error here. Since it's a
605 * fairly common mistake to try to use rsh when ssh is required, we
606 * should trap that: if we fail to read any data at all, we should
607 * give a better explanation. We can tell whether the connection has
608 * started by looking e.g. at whether the remote version is known yet.
610 static int read_timeout(int fd, char *buf, size_t len)
614 io_flush(FULL_FLUSH);
617 /* until we manage to read *something* */
626 if (io_filesfrom_f_out >= 0) {
628 if (ff_buf.len == 0) {
629 if (io_filesfrom_f_in >= 0) {
630 FD_SET(io_filesfrom_f_in, &r_fds);
631 new_fd = io_filesfrom_f_in;
633 io_filesfrom_f_out = -1;
637 FD_SET(io_filesfrom_f_out, &w_fds);
638 new_fd = io_filesfrom_f_out;
644 tv.tv_sec = select_timeout;
649 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
652 if (errno == EBADF) {
653 defer_forwarding_messages = 0;
654 exit_cleanup(RERR_SOCKETIO);
660 if (io_filesfrom_f_out >= 0) {
662 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
663 int l = write(io_filesfrom_f_out,
664 ff_buf.buf + ff_buf.pos,
667 if (!(ff_buf.len -= l))
671 } else if (errno != EINTR) {
672 /* XXX should we complain? */
673 io_filesfrom_f_out = -1;
676 } else if (io_filesfrom_f_in >= 0) {
677 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
679 xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
681 xbuf *ibuf = &ff_buf;
683 int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
685 if (l == 0 || errno != EINTR) {
686 /* Send end-of-file marker */
687 memcpy(ff_buf.buf, "\0\0", 2);
688 ff_buf.len = ff_lastchar? 2 : 1;
690 io_filesfrom_f_in = -1;
694 if (filesfrom_convert) {
697 iconvbufs(ic_send, &iconv_buf, &ff_buf,
698 ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
703 char *s = ff_buf.buf + l;
704 /* Transform CR and/or LF into '\0' */
705 while (s-- > ff_buf.buf) {
706 if (*s == '\n' || *s == '\r')
711 /* Last buf ended with a '\0', so don't
712 * let this buf start with one. */
713 while (l && ff_buf.buf[ff_buf.pos] == '\0')
719 char *f = ff_buf.buf + ff_buf.pos;
722 /* Eliminate any multi-'\0' runs. */
724 if (!(*t++ = *f++)) {
725 while (f != eob && !*f)
737 if (!FD_ISSET(fd, &r_fds))
740 n = read(fd, buf, len);
744 whine_about_eof(fd); /* Doesn't return. */
745 if (errno == EINTR || errno == EWOULDBLOCK
749 /* Don't write errors on a dead socket. */
750 if (fd == sock_f_in) {
751 io_end_multiplex_out();
752 rsyserr(FSOCKERR, errno, "read error");
754 rsyserr(FERROR, errno, "read error");
755 exit_cleanup(RERR_STREAMIO);
762 if (fd == sock_f_in && io_timeout)
763 last_io_in = time(NULL);
769 /* Read a line into the "buf" buffer. */
770 int read_line(int fd, char *buf, size_t bufsiz, int flags)
776 if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
777 realloc_xbuf(&iconv_buf, bufsiz + 1024);
782 s = flags & RL_CONVERT ? iconv_buf.buf : buf;
786 eob = s + bufsiz - 1;
788 cnt = read(fd, &ch, 1);
789 if (cnt < 0 && (errno == EWOULDBLOCK
790 || errno == EINTR || errno == EAGAIN)) {
797 tv.tv_sec = select_timeout;
799 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
801 /*if (FD_ISSET(fd, &e_fds))
802 rprintf(FINFO, "select exception on fd %d\n", fd); */
807 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
808 /* Skip empty lines if dumping comments. */
809 if (flags & RL_DUMP_COMMENTS && s == buf)
818 if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
822 if (flags & RL_CONVERT) {
824 INIT_XBUF(outbuf, buf, 0, bufsiz);
826 iconv_buf.len = s - iconv_buf.buf;
827 iconvbufs(ic_recv, &iconv_buf, &outbuf,
828 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
829 outbuf.buf[outbuf.len] = '\0';
837 int read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
838 char ***argv_p, int *argc_p, char **request_p)
840 int maxargs = MAX_ARGS;
844 int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
847 rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
850 if (!(argv = new_array(char *, maxargs)))
851 out_of_memory("read_args");
853 argv[argc++] = "rsyncd";
856 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
859 if (argc == maxargs) {
861 if (!(argv = realloc_array(argv, char *, maxargs)))
862 out_of_memory("read_args");
867 *request_p = strdup(buf);
871 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
873 glob_expand(buf, &argv, &argc, &maxargs);
875 if (!(p = strdup(buf)))
876 out_of_memory("read_args");
878 if (*p == '.' && p[1] == '\0')
886 return dot_pos ? dot_pos : argc;
889 int io_start_buffering_out(int f_out)
892 assert(f_out == iobuf_f_out);
895 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
896 out_of_memory("io_start_buffering_out");
902 int io_start_buffering_in(int f_in)
905 assert(f_in == iobuf_f_in);
908 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
909 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
910 out_of_memory("io_start_buffering_in");
915 void io_end_buffering_in(void)
922 iobuf_in_remaining = 0;
926 void io_end_buffering_out(void)
930 io_flush(FULL_FLUSH);
936 void maybe_flush_socket(int important)
938 if (iobuf_out && iobuf_out_cnt
939 && (important || time(NULL) - last_io_out >= 5))
940 io_flush(NORMAL_FLUSH);
943 void maybe_send_keepalive(void)
945 if (time(NULL) - last_io_out >= allowed_lull) {
946 if (!iobuf_out || !iobuf_out_cnt) {
947 if (protocol_version < 29)
948 return; /* there's nothing we can do */
949 if (protocol_version >= 30)
950 send_msg(MSG_NOOP, "", 0, 0);
952 write_int(sock_f_out, cur_flist->used);
953 write_shortint(sock_f_out, ITEM_IS_NEW);
957 io_flush(NORMAL_FLUSH);
961 void start_flist_forward(int f_in)
963 assert(iobuf_out != NULL);
964 assert(iobuf_f_out == msg_fd_out);
965 flist_forward_from = f_in;
968 void stop_flist_forward()
970 flist_forward_from = -1;
971 io_flush(FULL_FLUSH);
975 * Continue trying to read len bytes - don't return until len has been
978 static void read_loop(int fd, char *buf, size_t len)
981 int n = read_timeout(fd, buf, len);
989 * Read from the file descriptor handling multiplexing - return number
992 * Never returns <= 0.
994 static int readfd_unbuffered(int fd, char *buf, size_t len)
998 char line[BIGPATHBUFLEN];
1000 if (!iobuf_in || fd != iobuf_f_in)
1001 return read_timeout(fd, buf, len);
1003 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1004 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1009 if (iobuf_in_remaining) {
1010 len = MIN(len, iobuf_in_remaining);
1011 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1012 iobuf_in_ndx += len;
1013 iobuf_in_remaining -= len;
1018 read_loop(fd, line, 4);
1019 tag = IVAL(line, 0);
1021 msg_bytes = tag & 0xFFFFFF;
1022 tag = (tag >> 24) - MPLEX_BASE;
1026 if (msg_bytes > iobuf_in_siz) {
1027 if (!(iobuf_in = realloc_array(iobuf_in, char,
1029 out_of_memory("readfd_unbuffered");
1030 iobuf_in_siz = msg_bytes;
1032 read_loop(fd, iobuf_in, msg_bytes);
1033 iobuf_in_remaining = msg_bytes;
1038 maybe_send_keepalive();
1043 read_loop(fd, line, msg_bytes);
1044 io_error |= IVAL(line, 0);
1047 if (msg_bytes >= sizeof line)
1050 if (ic_recv != (iconv_t)-1) {
1056 INIT_CONST_XBUF(outbuf, line);
1060 inbuf.len = msg_bytes > sizeof ibuf
1061 ? sizeof ibuf : msg_bytes;
1062 read_loop(fd, inbuf.buf, inbuf.len);
1063 if (!(msg_bytes -= inbuf.len)
1064 && !ibuf[inbuf.len-1])
1065 inbuf.len--, add_null = 1;
1066 if (iconvbufs(ic_send, &inbuf, &outbuf,
1067 ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1072 if (outbuf.len == outbuf.size)
1074 outbuf.buf[outbuf.len++] = '\0';
1076 msg_bytes = outbuf.len;
1079 read_loop(fd, line, msg_bytes);
1080 /* A directory name was sent with the trailing null */
1081 if (msg_bytes > 0 && !line[msg_bytes-1])
1082 log_delete(line, S_IFDIR);
1084 line[msg_bytes] = '\0';
1085 log_delete(line, S_IFREG);
1089 if (msg_bytes != 4) {
1091 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1092 tag, (long)msg_bytes, who_am_i());
1093 exit_cleanup(RERR_STREAMIO);
1095 read_loop(fd, line, msg_bytes);
1096 successful_send(IVAL(line, 0));
1101 read_loop(fd, line, msg_bytes);
1102 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1106 if (msg_bytes >= sizeof line) {
1109 "multiplexing overflow %d:%ld [%s]\n",
1110 tag, (long)msg_bytes, who_am_i());
1111 exit_cleanup(RERR_STREAMIO);
1113 read_loop(fd, line, msg_bytes);
1114 rwrite((enum logcode)tag, line, msg_bytes, 1);
1117 rprintf(FERROR, "unexpected tag %d [%s]\n",
1119 exit_cleanup(RERR_STREAMIO);
1123 if (iobuf_in_remaining == 0)
1124 io_flush(NORMAL_FLUSH);
1129 /* Do a buffered read from fd. Don't return until all N bytes have
1130 * been read. If all N can't be read then exit with an error. */
1131 static void readfd(int fd, char *buffer, size_t N)
1137 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1141 if (fd == write_batch_monitor_in) {
1142 if ((size_t)write(batch_fd, buffer, total) != total)
1143 exit_cleanup(RERR_FILEIO);
1146 if (fd == flist_forward_from)
1147 writefd(iobuf_f_out, buffer, total);
1149 if (fd == sock_f_in)
1150 stats.total_read += total;
1153 unsigned short read_shortint(int f)
1157 return (UVAL(b, 1) << 8) + UVAL(b, 0);
1160 int32 read_int(int f)
1167 #if SIZEOF_INT32 > 4
1168 if (num & (int32)0x80000000)
1169 num |= ~(int32)0xffffffff;
1174 int32 read_varint(int f)
1184 readfd(f, (char*)&ch, 1);
1185 extra = int_byte_extra[ch / 4];
1187 uchar bit = ((uchar)1<<(8-extra));
1188 if (extra >= (int)sizeof u.b) {
1189 rprintf(FERROR, "Overflow in read_varint()\n");
1190 exit_cleanup(RERR_STREAMIO);
1192 readfd(f, u.b, extra);
1193 u.b[extra] = ch & (bit-1);
1196 #if CAREFUL_ALIGNMENT
1199 #if SIZEOF_INT32 > 4
1200 if (u.x & (int32)0x80000000)
1201 u.x |= ~(int32)0xffffffff;
1206 int64 read_varlong(int f, uchar min_bytes)
1215 #if SIZEOF_INT64 < 8
1220 readfd(f, b2, min_bytes);
1221 memcpy(u.b, b2+1, min_bytes-1);
1222 extra = int_byte_extra[CVAL(b2, 0) / 4];
1224 uchar bit = ((uchar)1<<(8-extra));
1225 if (min_bytes + extra > (int)sizeof u.b) {
1226 rprintf(FERROR, "Overflow in read_varlong()\n");
1227 exit_cleanup(RERR_STREAMIO);
1229 readfd(f, u.b + min_bytes - 1, extra);
1230 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1231 #if SIZEOF_INT64 < 8
1232 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1233 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1234 exit_cleanup(RERR_UNSUPPORTED);
1238 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1239 #if SIZEOF_INT64 < 8
1241 #elif CAREFUL_ALIGNMENT
1242 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1247 int64 read_longint(int f)
1249 #if SIZEOF_INT64 >= 8
1252 int32 num = read_int(f);
1254 if (num != (int32)0xffffffff)
1257 #if SIZEOF_INT64 < 8
1258 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1259 exit_cleanup(RERR_UNSUPPORTED);
1262 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1266 void read_buf(int f, char *buf, size_t len)
1271 void read_sbuf(int f, char *buf, size_t len)
1273 readfd(f, buf, len);
1277 uchar read_byte(int f)
1280 readfd(f, (char *)&c, 1);
1284 int read_vstring(int f, char *buf, int bufsize)
1286 int len = read_byte(f);
1289 len = (len & ~0x80) * 0x100 + read_byte(f);
1291 if (len >= bufsize) {
1292 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1298 readfd(f, buf, len);
1303 /* Populate a sum_struct with values from the socket. This is
1304 * called by both the sender and the receiver. */
1305 void read_sum_head(int f, struct sum_struct *sum)
1307 sum->count = read_int(f);
1308 if (sum->count < 0) {
1309 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1310 (long)sum->count, who_am_i());
1311 exit_cleanup(RERR_PROTOCOL);
1313 sum->blength = read_int(f);
1314 if (sum->blength < 0 || sum->blength > MAX_BLOCK_SIZE) {
1315 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1316 (long)sum->blength, who_am_i());
1317 exit_cleanup(RERR_PROTOCOL);
1319 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1320 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1321 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1322 sum->s2length, who_am_i());
1323 exit_cleanup(RERR_PROTOCOL);
1325 sum->remainder = read_int(f);
1326 if (sum->remainder < 0 || sum->remainder > sum->blength) {
1327 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1328 (long)sum->remainder, who_am_i());
1329 exit_cleanup(RERR_PROTOCOL);
1333 /* Send the values from a sum_struct over the socket. Set sum to
1334 * NULL if there are no checksums to send. This is called by both
1335 * the generator and the sender. */
1336 void write_sum_head(int f, struct sum_struct *sum)
1338 static struct sum_struct null_sum;
1343 write_int(f, sum->count);
1344 write_int(f, sum->blength);
1345 if (protocol_version >= 27)
1346 write_int(f, sum->s2length);
1347 write_int(f, sum->remainder);
1351 * Sleep after writing to limit I/O bandwidth usage.
1353 * @todo Rather than sleeping after each write, it might be better to
1354 * use some kind of averaging. The current algorithm seems to always
1355 * use a bit less bandwidth than specified, because it doesn't make up
1356 * for slow periods. But arguably this is a feature. In addition, we
1357 * ought to take the time used to write the data into account.
1359 * During some phases of big transfers (file FOO is uptodate) this is
1360 * called with a small bytes_written every time. As the kernel has to
1361 * round small waits up to guarantee that we actually wait at least the
1362 * requested number of microseconds, this can become grossly inaccurate.
1363 * We therefore keep track of the bytes we've written over time and only
1364 * sleep when the accumulated delay is at least 1 tenth of a second.
1366 static void sleep_for_bwlimit(int bytes_written)
1368 static struct timeval prior_tv;
1369 static long total_written = 0;
1370 struct timeval tv, start_tv;
1371 long elapsed_usec, sleep_usec;
1373 #define ONE_SEC 1000000L /* # of microseconds in a second */
1375 if (!bwlimit_writemax)
1378 total_written += bytes_written;
1380 gettimeofday(&start_tv, NULL);
1381 if (prior_tv.tv_sec) {
1382 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1383 + (start_tv.tv_usec - prior_tv.tv_usec);
1384 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1385 if (total_written < 0)
1389 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1390 if (sleep_usec < ONE_SEC / 10) {
1391 prior_tv = start_tv;
1395 tv.tv_sec = sleep_usec / ONE_SEC;
1396 tv.tv_usec = sleep_usec % ONE_SEC;
1397 select(0, NULL, NULL, NULL, &tv);
1399 gettimeofday(&prior_tv, NULL);
1400 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1401 + (prior_tv.tv_usec - start_tv.tv_usec);
1402 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1405 /* Write len bytes to the file descriptor fd, looping as necessary to get
1406 * the job done and also (in certain circumstances) reading any data on
1407 * msg_fd_in to avoid deadlock.
1409 * This function underlies the multiplexing system. The body of the
1410 * application never calls this function directly. */
1411 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1413 size_t n, total = 0;
1414 fd_set w_fds, r_fds, e_fds;
1415 int maxfd, count, cnt, using_r_fds;
1420 defer_forwarding_messages++, defer_inc++;
1422 while (total < len) {
1429 if (msg_fd_in >= 0) {
1431 FD_SET(msg_fd_in, &r_fds);
1432 if (msg_fd_in > maxfd)
1438 tv.tv_sec = select_timeout;
1442 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1443 &w_fds, &e_fds, &tv);
1446 if (count < 0 && errno == EBADF)
1447 exit_cleanup(RERR_SOCKETIO);
1452 /*if (FD_ISSET(fd, &e_fds))
1453 rprintf(FINFO, "select exception on fd %d\n", fd); */
1455 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1458 if (!FD_ISSET(fd, &w_fds))
1462 if (bwlimit_writemax && n > bwlimit_writemax)
1463 n = bwlimit_writemax;
1464 cnt = write(fd, buf + total, n);
1470 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1476 /* Don't try to write errors back across the stream. */
1477 if (fd == sock_f_out)
1478 io_end_multiplex_out();
1479 /* Don't try to write errors down a failing msg pipe. */
1480 if (am_server && fd == msg_fd_out)
1481 exit_cleanup(RERR_STREAMIO);
1482 rsyserr(FERROR, errno,
1483 "writefd_unbuffered failed to write %ld bytes [%s]",
1484 (long)len, who_am_i());
1485 /* If the other side is sending us error messages, try
1486 * to grab any messages they sent before they died. */
1487 while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1491 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1493 exit_cleanup(RERR_STREAMIO);
1497 defer_forwarding_messages++, defer_inc++;
1499 if (fd == sock_f_out) {
1500 if (io_timeout || am_generator)
1501 last_io_out = time(NULL);
1502 sleep_for_bwlimit(cnt);
1507 if (!(defer_forwarding_messages -= defer_inc))
1511 void io_flush(int flush_it_all)
1513 if (!iobuf_out_cnt || no_flush)
1516 if (io_multiplexing_out)
1517 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1519 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1522 if (flush_it_all && !defer_forwarding_messages)
1526 static void writefd(int fd, const char *buf, size_t len)
1528 if (fd == sock_f_out)
1529 stats.total_written += len;
1531 if (fd == write_batch_monitor_out) {
1532 if ((size_t)write(batch_fd, buf, len) != len)
1533 exit_cleanup(RERR_FILEIO);
1536 if (!iobuf_out || fd != iobuf_f_out) {
1537 writefd_unbuffered(fd, buf, len);
1542 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1544 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1550 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1551 io_flush(NORMAL_FLUSH);
1555 void write_shortint(int f, unsigned short x)
1559 b[1] = (char)(x >> 8);
1563 void write_int(int f, int32 x)
1570 void write_varint(int f, int32 x)
1578 while (cnt > 1 && b[cnt] == 0)
1580 bit = ((uchar)1<<(7-cnt+1));
1581 if (CVAL(b, cnt) >= bit) {
1585 *b = b[cnt] | ~(bit*2-1);
1592 void write_varlong(int f, int64 x, uchar min_bytes)
1599 #if SIZEOF_INT64 >= 8
1600 SIVAL(b, 5, x >> 32);
1602 if (x <= 0x7FFFFFFF && x >= 0)
1603 memset(b + 5, 0, 4);
1605 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1606 exit_cleanup(RERR_UNSUPPORTED);
1610 while (cnt > min_bytes && b[cnt] == 0)
1612 bit = ((uchar)1<<(7-cnt+min_bytes));
1613 if (CVAL(b, cnt) >= bit) {
1616 } else if (cnt > min_bytes)
1617 *b = b[cnt] | ~(bit*2-1);
1625 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1626 * 64-bit types on this platform.
1628 void write_longint(int f, int64 x)
1630 char b[12], * const s = b+4;
1633 if (x <= 0x7FFFFFFF && x >= 0) {
1638 #if SIZEOF_INT64 < 8
1639 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1640 exit_cleanup(RERR_UNSUPPORTED);
1643 SIVAL(s, 4, x >> 32);
1648 void write_buf(int f, const char *buf, size_t len)
1653 /** Write a string to the connection */
1654 void write_sbuf(int f, const char *buf)
1656 writefd(f, buf, strlen(buf));
1659 void write_byte(int f, uchar c)
1661 writefd(f, (char *)&c, 1);
1664 void write_vstring(int f, const char *str, int len)
1666 uchar lenbuf[3], *lb = lenbuf;
1671 "attempting to send over-long vstring (%d > %d)\n",
1673 exit_cleanup(RERR_PROTOCOL);
1675 *lb++ = len / 0x100 + 0x80;
1679 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1681 writefd(f, str, len);
1684 /* Send a file-list index using a byte-reduction method. */
1685 void write_ndx(int f, int32 ndx)
1687 static int32 prev_positive = -1, prev_negative = 1;
1688 int32 diff, cnt = 0;
1691 if (protocol_version < 30 || read_batch) {
1696 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1697 * negative nums as a positive after sending a leading 0xFF. */
1699 diff = ndx - prev_positive;
1700 prev_positive = ndx;
1701 } else if (ndx == NDX_DONE) {
1706 b[cnt++] = (char)0xFF;
1708 diff = ndx - prev_negative;
1709 prev_negative = ndx;
1712 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1713 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1714 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1715 if (diff < 0xFE && diff > 0)
1716 b[cnt++] = (char)diff;
1717 else if (diff < 0 || diff > 0x7FFF) {
1718 b[cnt++] = (char)0xFE;
1719 b[cnt++] = (char)((ndx >> 24) | 0x80);
1720 b[cnt++] = (char)ndx;
1721 b[cnt++] = (char)(ndx >> 8);
1722 b[cnt++] = (char)(ndx >> 16);
1724 b[cnt++] = (char)0xFE;
1725 b[cnt++] = (char)(diff >> 8);
1726 b[cnt++] = (char)diff;
1731 /* Receive a file-list index using a byte-reduction method. */
1732 int32 read_ndx(int f)
1734 static int32 prev_positive = -1, prev_negative = 1;
1735 int32 *prev_ptr, num;
1738 if (protocol_version < 30)
1742 if (CVAL(b, 0) == 0xFF) {
1744 prev_ptr = &prev_negative;
1745 } else if (CVAL(b, 0) == 0)
1748 prev_ptr = &prev_positive;
1749 if (CVAL(b, 0) == 0xFE) {
1751 if (CVAL(b, 0) & 0x80) {
1752 b[3] = CVAL(b, 0) & ~0x80;
1757 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1759 num = UVAL(b, 0) + *prev_ptr;
1761 if (prev_ptr == &prev_negative)
1766 /* Read a line of up to bufsiz-1 characters into buf. Strips
1767 * the (required) trailing newline and all carriage returns.
1768 * Returns 1 for success; 0 for I/O error or truncation. */
1769 int read_line_old(int f, char *buf, size_t bufsiz)
1771 bufsiz--; /* leave room for the null */
1772 while (bufsiz > 0) {
1774 read_buf(f, buf, 1);
1779 if (buf[0] != '\r') {
1788 void io_printf(int fd, const char *format, ...)
1791 char buf[BIGPATHBUFLEN];
1794 va_start(ap, format);
1795 len = vsnprintf(buf, sizeof buf, format, ap);
1799 exit_cleanup(RERR_STREAMIO);
1801 if (len > (int)sizeof buf) {
1802 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1803 exit_cleanup(RERR_STREAMIO);
1806 write_sbuf(fd, buf);
1809 /** Setup for multiplexing a MSG_* stream with the data stream. */
1810 void io_start_multiplex_out(void)
1812 io_flush(NORMAL_FLUSH);
1813 io_start_buffering_out(sock_f_out);
1814 io_multiplexing_out = 1;
1817 /** Setup for multiplexing a MSG_* stream with the data stream. */
1818 void io_start_multiplex_in(void)
1820 io_flush(NORMAL_FLUSH);
1821 io_start_buffering_in(sock_f_in);
1822 io_multiplexing_in = 1;
1825 /** Write an message to the multiplexed data stream. */
1826 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1828 if (!io_multiplexing_out)
1830 io_flush(NORMAL_FLUSH);
1831 stats.total_written += (len+4);
1832 mplex_write(sock_f_out, code, buf, len, convert);
1836 void io_end_multiplex_in(void)
1838 io_multiplexing_in = 0;
1839 io_end_buffering_in();
1842 /** Stop output multiplexing. */
1843 void io_end_multiplex_out(void)
1845 io_multiplexing_out = 0;
1846 io_end_buffering_out();
1849 void start_write_batch(int fd)
1851 /* Some communication has already taken place, but we don't
1852 * enable batch writing until here so that we can write a
1853 * canonical record of the communication even though the
1854 * actual communication so far depends on whether a daemon
1856 write_int(batch_fd, protocol_version);
1857 write_int(batch_fd, checksum_seed);
1860 write_batch_monitor_out = fd;
1862 write_batch_monitor_in = fd;
1865 void stop_write_batch(void)
1867 write_batch_monitor_out = -1;
1868 write_batch_monitor_in = -1;