Got rid of the NVAL*() defines.
[rsync/rsync.git] / io.c
CommitLineData
0f78b815
WD
1/*
2 * Socket and pipe I/O utilities used in rsync.
d62bcc17 3 *
0f78b815
WD
4 * Copyright (C) 1996-2001 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
ba2133d6 7 * Copyright (C) 2003-2007 Wayne Davison
d62bcc17 8 *
880da007 9 * This program is free software; you can redistribute it and/or modify
ba2133d6
WD
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
d62bcc17 12 *
880da007
MP
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
d62bcc17 17 *
e7c67065
WD
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
880da007 21 */
720b47f2 22
0f78b815
WD
23/* Rsync provides its own multiplexing system, which is used to send
24 * stderr and stdout over a single socket.
87ee2481
MP
25 *
26 * For historical reasons this is off during the start of the
27 * connection, but it's switched on quite early using
0f78b815 28 * io_start_multiplex_out() and io_start_multiplex_in(). */
720b47f2 29
720b47f2
AT
30#include "rsync.h"
31
880da007 32/** If no timeout is specified then use a 60 second select timeout */
8cd9fd4e
AT
33#define SELECT_TIMEOUT 60
34
7a55d06e 35extern int bwlimit;
71e58630 36extern size_t bwlimit_writemax;
6ba9279f 37extern int io_timeout;
cdf236aa 38extern int allowed_lull;
d17e1dd2
WD
39extern int am_server;
40extern int am_daemon;
41extern int am_sender;
98f8c9a5 42extern int am_generator;
3ea6e0e7 43extern int inc_recurse;
8ef246e0 44extern int io_error;
e626b29e 45extern int eol_nulls;
8ef246e0 46extern int flist_eof;
3b0a30eb 47extern int read_batch;
188fed95 48extern int csum_length;
b9f592fb
WD
49extern int checksum_seed;
50extern int protocol_version;
47c11975 51extern int remove_source_files;
cdf236aa 52extern int preserve_hard_links;
d19320fd 53extern char *filesfrom_host;
a800434a 54extern struct stats stats;
8ef246e0 55extern struct file_list *cur_flist, *first_flist;
720b47f2 56
a86179f4 57const char phase_unknown[] = "unknown";
9e2409ab 58int ignore_timeout = 0;
b9f592fb 59int batch_fd = -1;
04c722d5 60int msgdone_cnt = 0;
805edf9d 61
cb2e1f18 62/* Ignore an EOF error if non-zero. See whine_about_eof(). */
574c2500 63int kluge_around_eof = 0;
7a55d06e 64
d17e1dd2
WD
65int msg_fd_in = -1;
66int msg_fd_out = -1;
cdf236aa
WD
67int sock_f_in = -1;
68int sock_f_out = -1;
7a55d06e 69
8ef246e0
WD
70static int iobuf_f_in = -1;
71static char *iobuf_in;
72static size_t iobuf_in_siz;
73static size_t iobuf_in_ndx;
74static size_t iobuf_in_remaining;
75
76static int iobuf_f_out = -1;
77static char *iobuf_out;
78static int iobuf_out_cnt;
79
80int flist_forward_from = -1;
81
1f75bb10
WD
82static int io_multiplexing_out;
83static int io_multiplexing_in;
3b0a30eb
WD
84static time_t last_io_in;
85static time_t last_io_out;
1f75bb10
WD
86static int no_flush;
87
b9f592fb
WD
88static int write_batch_monitor_in = -1;
89static int write_batch_monitor_out = -1;
90
56014c8c
WD
91static int io_filesfrom_f_in = -1;
92static int io_filesfrom_f_out = -1;
93static char io_filesfrom_buf[2048];
94static char *io_filesfrom_bp;
95static char io_filesfrom_lastchar;
96static int io_filesfrom_buflen;
954bbed8 97static int defer_forwarding_messages = 0;
3b0a30eb 98static int select_timeout = SELECT_TIMEOUT;
d6081c82
WD
99static int active_filecnt = 0;
100static OFF_T active_bytecnt = 0;
720b47f2 101
351e23ad
WD
102static char int_byte_extra[64] = {
103 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
104 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
105 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
106 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
482f48cc
WD
107};
108
dde5b772 109static void readfd(int fd, char *buffer, size_t N);
8ef246e0
WD
110static void writefd(int fd, const char *buf, size_t len);
111static void writefd_unbuffered(int fd, const char *buf, size_t len);
112static void decrement_active_files(int ndx);
113static void decrement_flist_in_progress(int ndx, int redo);
ff41a59f 114
c6816b94
WD
115struct flist_ndx_item {
116 struct flist_ndx_item *next;
117 int ndx;
d17e1dd2
WD
118};
119
c6816b94
WD
120struct flist_ndx_list {
121 struct flist_ndx_item *head, *tail;
122};
123
cdf236aa 124static struct flist_ndx_list redo_list, hlink_list;
d17e1dd2 125
08c88178
WD
126struct msg_list_item {
127 struct msg_list_item *next;
d17e1dd2 128 int len;
954bbed8 129 char buf[1];
d17e1dd2
WD
130};
131
08c88178
WD
132struct msg_list {
133 struct msg_list_item *head, *tail;
134};
135
3be97bf9 136static struct msg_list msg2sndr;
d17e1dd2 137
c6816b94 138static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
d17e1dd2 139{
c6816b94
WD
140 struct flist_ndx_item *item;
141
142 if (!(item = new(struct flist_ndx_item)))
143 out_of_memory("flist_ndx_push");
144 item->next = NULL;
145 item->ndx = ndx;
146 if (lp->tail)
147 lp->tail->next = item;
d17e1dd2 148 else
c6816b94
WD
149 lp->head = item;
150 lp->tail = item;
151}
152
153static int flist_ndx_pop(struct flist_ndx_list *lp)
154{
155 struct flist_ndx_item *next;
156 int ndx;
157
158 if (!lp->head)
159 return -1;
160
161 ndx = lp->head->ndx;
162 next = lp->head->next;
163 free(lp->head);
164 lp->head = next;
165 if (!next)
166 lp->tail = NULL;
167
168 return ndx;
d17e1dd2
WD
169}
170
8d9dc9f9
AT
171static void check_timeout(void)
172{
173 time_t t;
90ba34e2 174
9e2409ab 175 if (!io_timeout || ignore_timeout)
d17e1dd2 176 return;
8d9dc9f9 177
3b0a30eb
WD
178 if (!last_io_in) {
179 last_io_in = time(NULL);
8d9dc9f9
AT
180 return;
181 }
182
183 t = time(NULL);
184
3b0a30eb 185 if (t - last_io_in >= io_timeout) {
0adb99b9 186 if (!am_server && !am_daemon) {
4ccfd96c 187 rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
3b0a30eb 188 (int)(t-last_io_in));
0adb99b9 189 }
65417579 190 exit_cleanup(RERR_TIMEOUT);
8d9dc9f9
AT
191 }
192}
193
1f75bb10
WD
194/* Note the fds used for the main socket (which might really be a pipe
195 * for a local transfer, but we can ignore that). */
196void io_set_sock_fds(int f_in, int f_out)
197{
198 sock_f_in = f_in;
199 sock_f_out = f_out;
200}
201
3b0a30eb
WD
202void set_io_timeout(int secs)
203{
204 io_timeout = secs;
205
206 if (!io_timeout || io_timeout > SELECT_TIMEOUT)
207 select_timeout = SELECT_TIMEOUT;
208 else
209 select_timeout = io_timeout;
210
211 allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
212}
213
98f8c9a5
WD
214/* Setup the fd used to receive MSG_* messages. Only needed during the
215 * early stages of being a local sender (up through the sending of the
216 * file list) or when we're the generator (to fetch the messages from
217 * the receiver). */
d17e1dd2
WD
218void set_msg_fd_in(int fd)
219{
220 msg_fd_in = fd;
221}
222
98f8c9a5
WD
223/* Setup the fd used to send our MSG_* messages. Only needed when
224 * we're the receiver (to send our messages to the generator). */
d17e1dd2
WD
225void set_msg_fd_out(int fd)
226{
227 msg_fd_out = fd;
228 set_nonblocking(msg_fd_out);
229}
230
231/* Add a message to the pending MSG_* list. */
4a19c3b2 232static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len)
554e0a8d 233{
954bbed8
WD
234 struct msg_list_item *m;
235 int sz = len + 4 + sizeof m[0] - 1;
d17e1dd2 236
954bbed8 237 if (!(m = (struct msg_list_item *)new_array(char, sz)))
c6816b94 238 out_of_memory("msg_list_add");
954bbed8
WD
239 m->next = NULL;
240 m->len = len + 4;
241 SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
242 memcpy(m->buf + 4, buf, len);
243 if (lst->tail)
244 lst->tail->next = m;
d17e1dd2 245 else
954bbed8
WD
246 lst->head = m;
247 lst->tail = m;
554e0a8d
AT
248}
249
3be97bf9
WD
250static void msg2sndr_flush(void)
251{
252 while (msg2sndr.head && io_multiplexing_out) {
253 struct msg_list_item *m = msg2sndr.head;
254 if (!(msg2sndr.head = m->next))
255 msg2sndr.tail = NULL;
256 stats.total_written += m->len;
257 defer_forwarding_messages = 1;
258 writefd_unbuffered(sock_f_out, m->buf, m->len);
259 defer_forwarding_messages = 0;
260 free(m);
261 }
262}
263
98f8c9a5
WD
264/* Read a message from the MSG_* fd and handle it. This is called either
265 * during the early stages of being a local sender (up through the sending
266 * of the file list) or when we're the generator (to fetch the messages
267 * from the receiver). */
d17e1dd2 268static void read_msg_fd(void)
554e0a8d 269{
f9c6b3e7 270 char buf[2048];
06ce139f 271 size_t n;
8ef246e0 272 struct file_list *flist;
d17e1dd2 273 int fd = msg_fd_in;
ff41a59f
AT
274 int tag, len;
275
00bdf899 276 /* Temporarily disable msg_fd_in. This is needed to avoid looping back
af436313 277 * to this routine from writefd_unbuffered(). */
3be97bf9 278 no_flush++;
d17e1dd2 279 msg_fd_in = -1;
be21e29c 280 defer_forwarding_messages++;
554e0a8d 281
dde5b772 282 readfd(fd, buf, 4);
ff41a59f
AT
283 tag = IVAL(buf, 0);
284
285 len = tag & 0xFFFFFF;
d17e1dd2 286 tag = (tag >> 24) - MPLEX_BASE;
ff41a59f 287
d17e1dd2
WD
288 switch (tag) {
289 case MSG_DONE:
cc7b86bf 290 if (len < 0 || len > 1 || !am_generator) {
8ef246e0
WD
291 invalid_msg:
292 rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
293 tag, len, who_am_i(),
3ea6e0e7 294 inc_recurse ? "/inc" : "");
d17e1dd2 295 exit_cleanup(RERR_STREAMIO);
13c7bcbb 296 }
cc7b86bf
WD
297 if (len) {
298 readfd(fd, buf, len);
351e23ad 299 stats.total_read = read_varlong(fd, 3);
cc7b86bf 300 }
04c722d5 301 msgdone_cnt++;
d17e1dd2
WD
302 break;
303 case MSG_REDO:
8ef246e0
WD
304 if (len != 4 || !am_generator)
305 goto invalid_msg;
dde5b772 306 readfd(fd, buf, 4);
47c11975 307 if (remove_source_files)
d6081c82 308 decrement_active_files(IVAL(buf,0));
c6816b94 309 flist_ndx_push(&redo_list, IVAL(buf,0));
3ea6e0e7 310 if (inc_recurse)
8ef246e0
WD
311 decrement_flist_in_progress(IVAL(buf,0), 1);
312 break;
313 case MSG_FLIST:
3ea6e0e7 314 if (len != 4 || !am_generator || !inc_recurse)
8ef246e0 315 goto invalid_msg;
dde5b772 316 readfd(fd, buf, 4);
8ef246e0
WD
317 /* Read extra file list from receiver. */
318 assert(iobuf_in != NULL);
319 assert(iobuf_f_in == fd);
320 flist = recv_file_list(fd);
321 flist->parent_ndx = IVAL(buf,0);
322 break;
323 case MSG_FLIST_EOF:
3ea6e0e7 324 if (len != 0 || !am_generator || !inc_recurse)
8ef246e0
WD
325 goto invalid_msg;
326 flist_eof = 1;
d17e1dd2 327 break;
0d67e00a 328 case MSG_DELETED:
8ef246e0
WD
329 if (len >= (int)sizeof buf || !am_generator)
330 goto invalid_msg;
dde5b772 331 readfd(fd, buf, len);
12ccc73a 332 send_msg(MSG_DELETED, buf, len);
0d67e00a 333 break;
9981c27e 334 case MSG_SUCCESS:
8ef246e0
WD
335 if (len != 4 || !am_generator)
336 goto invalid_msg;
dde5b772 337 readfd(fd, buf, len);
47c11975 338 if (remove_source_files) {
d6081c82 339 decrement_active_files(IVAL(buf,0));
12ccc73a 340 send_msg(MSG_SUCCESS, buf, len);
d6081c82 341 }
cdf236aa
WD
342 if (preserve_hard_links)
343 flist_ndx_push(&hlink_list, IVAL(buf,0));
3ea6e0e7 344 if (inc_recurse)
8ef246e0
WD
345 decrement_flist_in_progress(IVAL(buf,0), 0);
346 break;
347 case MSG_NO_SEND:
348 if (len != 4 || !am_generator)
349 goto invalid_msg;
dde5b772 350 readfd(fd, buf, len);
3ea6e0e7 351 if (inc_recurse)
8ef246e0 352 decrement_flist_in_progress(IVAL(buf,0), 0);
9981c27e 353 break;
72f2d1b3 354 case MSG_SOCKERR:
64119c79 355 case MSG_CLIENT:
8ef246e0
WD
356 if (!am_generator)
357 goto invalid_msg;
64119c79 358 if (tag == MSG_SOCKERR)
8ef246e0 359 io_end_multiplex_out();
72f2d1b3 360 /* FALL THROUGH */
d17e1dd2
WD
361 case MSG_INFO:
362 case MSG_ERROR:
363 case MSG_LOG:
364 while (len) {
365 n = len;
366 if (n >= sizeof buf)
367 n = sizeof buf - 1;
dde5b772 368 readfd(fd, buf, n);
54b0dfa0 369 rwrite((enum logcode)tag, buf, n);
d17e1dd2
WD
370 len -= n;
371 }
372 break;
373 default:
f8b9da1a
WD
374 rprintf(FERROR, "unknown message %d:%d [%s]\n",
375 tag, len, who_am_i());
d17e1dd2
WD
376 exit_cleanup(RERR_STREAMIO);
377 }
378
3be97bf9 379 no_flush--;
d17e1dd2 380 msg_fd_in = fd;
3be97bf9
WD
381 if (!--defer_forwarding_messages)
382 msg2sndr_flush();
d17e1dd2
WD
383}
384
d6081c82 385/* This is used by the generator to limit how many file transfers can
47c11975 386 * be active at once when --remove-source-files is specified. Without
d6081c82
WD
387 * this, sender-side deletions were mostly happening at the end. */
388void increment_active_files(int ndx, int itemizing, enum logcode code)
389{
390 /* TODO: tune these limits? */
5b986297 391 while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
04c722d5
WD
392 check_for_finished_files(itemizing, code, 0);
393 if (iobuf_out_cnt)
394 io_flush(NORMAL_FLUSH);
395 else
396 read_msg_fd();
d6081c82
WD
397 }
398
399 active_filecnt++;
121bfb2b 400 active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
d6081c82
WD
401}
402
8ef246e0 403static void decrement_active_files(int ndx)
d6081c82 404{
8ef246e0
WD
405 struct file_list *flist = flist_for_ndx(ndx);
406 assert(flist != NULL);
d6081c82 407 active_filecnt--;
8ef246e0
WD
408 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
409}
410
411static void decrement_flist_in_progress(int ndx, int redo)
412{
413 struct file_list *flist = cur_flist ? cur_flist : first_flist;
414
415 while (ndx < flist->ndx_start) {
416 if (flist == first_flist) {
417 invalid_ndx:
418 rprintf(FERROR,
419 "Invalid file index: %d (%d - %d) [%s]\n",
420 ndx, first_flist->ndx_start,
421 first_flist->prev->ndx_start + first_flist->prev->count - 1,
422 who_am_i());
423 exit_cleanup(RERR_PROTOCOL);
424 }
425 flist = flist->prev;
426 }
427 while (ndx >= flist->ndx_start + flist->count) {
428 if (!(flist = flist->next))
429 goto invalid_ndx;
430 }
431
432 flist->in_progress--;
433 if (redo)
434 flist->to_redo++;
d6081c82
WD
435}
436
3be97bf9
WD
437/* Write an message to a multiplexed stream. If this fails, rsync exits. */
438static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len)
d17e1dd2 439{
3be97bf9
WD
440 char buffer[1024];
441 size_t n = len;
d17e1dd2 442
3be97bf9
WD
443 SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
444
445 if (n > sizeof buffer - 4)
446 n = 0;
447 else
448 memcpy(buffer + 4, buf, n);
449
450 writefd_unbuffered(fd, buffer, n+4);
451
452 len -= n;
453 buf += n;
454
455 if (len) {
456 defer_forwarding_messages++;
457 writefd_unbuffered(fd, buf, len);
458 if (!--defer_forwarding_messages)
459 msg2sndr_flush();
8ef246e0 460 }
d17e1dd2
WD
461}
462
4a19c3b2 463int send_msg(enum msgcode code, const char *buf, int len)
37f35d89
WD
464{
465 if (msg_fd_out < 0) {
12ccc73a
WD
466 if (!defer_forwarding_messages)
467 return io_multiplex_write(code, buf, len);
468 if (!io_multiplexing_out)
469 return 0;
470 msg_list_add(&msg2sndr, code, buf, len);
471 return 1;
37f35d89 472 }
3be97bf9 473 mplex_write(msg_fd_out, code, buf, len);
12ccc73a 474 return 1;
37f35d89
WD
475}
476
155d9206
WD
477void send_msg_int(enum msgcode code, int num)
478{
479 char numbuf[4];
480 SIVAL(numbuf, 0, num);
481 send_msg(code, numbuf, 4);
482}
483
8ef246e0 484void wait_for_receiver(void)
d17e1dd2 485{
2a5df862
WD
486 if (iobuf_out_cnt)
487 io_flush(NORMAL_FLUSH);
488 else
489 read_msg_fd();
8ef246e0 490}
554e0a8d 491
8ef246e0
WD
492int get_redo_num(void)
493{
c6816b94 494 return flist_ndx_pop(&redo_list);
554e0a8d
AT
495}
496
cdf236aa
WD
497int get_hlink_num(void)
498{
499 return flist_ndx_pop(&hlink_list);
500}
501
56014c8c
WD
502/**
503 * When we're the receiver and we have a local --files-from list of names
504 * that needs to be sent over the socket to the sender, we have to do two
505 * things at the same time: send the sender a list of what files we're
506 * processing and read the incoming file+info list from the sender. We do
507 * this by augmenting the read_timeout() function to copy this data. It
508 * uses the io_filesfrom_buf to read a block of data from f_in (when it is
509 * ready, since it might be a pipe) and then blast it out f_out (when it
510 * is ready to receive more data).
511 */
512void io_set_filesfrom_fds(int f_in, int f_out)
513{
514 io_filesfrom_f_in = f_in;
515 io_filesfrom_f_out = f_out;
516 io_filesfrom_bp = io_filesfrom_buf;
517 io_filesfrom_lastchar = '\0';
518 io_filesfrom_buflen = 0;
519}
720b47f2 520
cb2e1f18
WD
521/* It's almost always an error to get an EOF when we're trying to read from the
522 * network, because the protocol is (for the most part) self-terminating.
880da007 523 *
cb2e1f18
WD
524 * There is one case for the receiver when it is at the end of the transfer
525 * (hanging around reading any keep-alive packets that might come its way): if
526 * the sender dies before the generator's kill-signal comes through, we can end
527 * up here needing to loop until the kill-signal arrives. In this situation,
528 * kluge_around_eof will be < 0.
529 *
530 * There is another case for older protocol versions (< 24) where the module
531 * listing was not terminated, so we must ignore an EOF error in that case and
532 * exit. In this situation, kluge_around_eof will be > 0. */
1f75bb10 533static void whine_about_eof(int fd)
7a55d06e 534{
574c2500 535 if (kluge_around_eof && fd == sock_f_in) {
55bb7fff 536 int i;
574c2500
WD
537 if (kluge_around_eof > 0)
538 exit_cleanup(0);
55bb7fff
WD
539 /* If we're still here after 10 seconds, exit with an error. */
540 for (i = 10*1000/20; i--; )
574c2500
WD
541 msleep(20);
542 }
3151cbae 543
00bdf899 544 rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
dca68b0a
WD
545 "(%.0f bytes received so far) [%s]\n",
546 (double)stats.total_read, who_am_i());
00bdf899
WD
547
548 exit_cleanup(RERR_STREAMIO);
7a55d06e 549}
720b47f2 550
880da007 551/**
6ed6d7f5 552 * Read from a socket with I/O timeout. return the number of bytes
c3563c46
MP
553 * read. If no bytes can be read then exit, never return a number <= 0.
554 *
8886f8d0
MP
555 * TODO: If the remote shell connection fails, then current versions
556 * actually report an "unexpected EOF" error here. Since it's a
557 * fairly common mistake to try to use rsh when ssh is required, we
558 * should trap that: if we fail to read any data at all, we should
559 * give a better explanation. We can tell whether the connection has
560 * started by looking e.g. at whether the remote version is known yet.
c3563c46 561 */
3151cbae 562static int read_timeout(int fd, char *buf, size_t len)
8d9dc9f9 563{
d8aeda1e 564 int n, cnt = 0;
4c36ddbe 565
dde5b772 566 io_flush(FULL_FLUSH);
ea2111d1 567
d8aeda1e 568 while (cnt == 0) {
7a55d06e 569 /* until we manage to read *something* */
56014c8c 570 fd_set r_fds, w_fds;
4c36ddbe 571 struct timeval tv;
1ea087a7 572 int maxfd = fd;
a57873b7 573 int count;
4c36ddbe 574
56014c8c 575 FD_ZERO(&r_fds);
a7026ba9 576 FD_ZERO(&w_fds);
56014c8c 577 FD_SET(fd, &r_fds);
3309507d 578 if (io_filesfrom_f_out >= 0) {
56014c8c
WD
579 int new_fd;
580 if (io_filesfrom_buflen == 0) {
3309507d 581 if (io_filesfrom_f_in >= 0) {
56014c8c
WD
582 FD_SET(io_filesfrom_f_in, &r_fds);
583 new_fd = io_filesfrom_f_in;
584 } else {
585 io_filesfrom_f_out = -1;
586 new_fd = -1;
587 }
588 } else {
56014c8c
WD
589 FD_SET(io_filesfrom_f_out, &w_fds);
590 new_fd = io_filesfrom_f_out;
591 }
1ea087a7
WD
592 if (new_fd > maxfd)
593 maxfd = new_fd;
554e0a8d
AT
594 }
595
e626b29e 596 tv.tv_sec = select_timeout;
4c36ddbe
AT
597 tv.tv_usec = 0;
598
554e0a8d
AT
599 errno = 0;
600
a7026ba9 601 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
a57873b7 602
a57873b7 603 if (count <= 0) {
2a5df862
WD
604 if (errno == EBADF) {
605 defer_forwarding_messages = 0;
554e0a8d 606 exit_cleanup(RERR_SOCKETIO);
2a5df862 607 }
bd717af8 608 check_timeout();
4c36ddbe
AT
609 continue;
610 }
611
3309507d 612 if (io_filesfrom_f_out >= 0) {
56014c8c
WD
613 if (io_filesfrom_buflen) {
614 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
615 int l = write(io_filesfrom_f_out,
616 io_filesfrom_bp,
617 io_filesfrom_buflen);
618 if (l > 0) {
619 if (!(io_filesfrom_buflen -= l))
620 io_filesfrom_bp = io_filesfrom_buf;
621 else
622 io_filesfrom_bp += l;
623 } else {
624 /* XXX should we complain? */
625 io_filesfrom_f_out = -1;
626 }
627 }
3309507d 628 } else if (io_filesfrom_f_in >= 0) {
56014c8c
WD
629 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
630 int l = read(io_filesfrom_f_in,
631 io_filesfrom_buf,
632 sizeof io_filesfrom_buf);
633 if (l <= 0) {
634 /* Send end-of-file marker */
635 io_filesfrom_buf[0] = '\0';
636 io_filesfrom_buf[1] = '\0';
637 io_filesfrom_buflen = io_filesfrom_lastchar? 2 : 1;
638 io_filesfrom_f_in = -1;
639 } else {
56014c8c
WD
640 if (!eol_nulls) {
641 char *s = io_filesfrom_buf + l;
642 /* Transform CR and/or LF into '\0' */
643 while (s-- > io_filesfrom_buf) {
644 if (*s == '\n' || *s == '\r')
645 *s = '\0';
646 }
647 }
648 if (!io_filesfrom_lastchar) {
649 /* Last buf ended with a '\0', so don't
650 * let this buf start with one. */
651 while (l && !*io_filesfrom_bp)
652 io_filesfrom_bp++, l--;
653 }
654 if (!l)
655 io_filesfrom_bp = io_filesfrom_buf;
656 else {
657 char *f = io_filesfrom_bp;
658 char *t = f;
659 char *eob = f + l;
660 /* Eliminate any multi-'\0' runs. */
661 while (f != eob) {
662 if (!(*t++ = *f++)) {
663 while (f != eob && !*f)
664 f++, l--;
665 }
666 }
667 io_filesfrom_lastchar = f[-1];
668 }
669 io_filesfrom_buflen = l;
670 }
671 }
672 }
673 }
674
f89b9368
WD
675 if (!FD_ISSET(fd, &r_fds))
676 continue;
554e0a8d 677
4c36ddbe
AT
678 n = read(fd, buf, len);
679
1ea087a7
WD
680 if (n <= 0) {
681 if (n == 0)
1f75bb10 682 whine_about_eof(fd); /* Doesn't return. */
d62bcc17
WD
683 if (errno == EINTR || errno == EWOULDBLOCK
684 || errno == EAGAIN)
7a55d06e 685 continue;
1f75bb10
WD
686
687 /* Don't write errors on a dead socket. */
72f2d1b3 688 if (fd == sock_f_in) {
8ef246e0 689 io_end_multiplex_out();
72f2d1b3
WD
690 rsyserr(FSOCKERR, errno, "read error");
691 } else
692 rsyserr(FERROR, errno, "read error");
1f75bb10 693 exit_cleanup(RERR_STREAMIO);
8d9dc9f9 694 }
00bdf899
WD
695
696 buf += n;
697 len -= n;
d8aeda1e 698 cnt += n;
1f75bb10 699
3b0a30eb
WD
700 if (fd == sock_f_in && io_timeout)
701 last_io_in = time(NULL);
4c36ddbe 702 }
8d9dc9f9 703
d8aeda1e 704 return cnt;
4c36ddbe 705}
8d9dc9f9 706
56014c8c
WD
707/**
708 * Read a line into the "fname" buffer (which must be at least MAXPATHLEN
709 * characters long).
710 */
711int read_filesfrom_line(int fd, char *fname)
712{
713 char ch, *s, *eob = fname + MAXPATHLEN - 1;
714 int cnt;
d19320fd 715 int reading_remotely = filesfrom_host != NULL;
56014c8c
WD
716 int nulls = eol_nulls || reading_remotely;
717
718 start:
719 s = fname;
720 while (1) {
721 cnt = read(fd, &ch, 1);
722 if (cnt < 0 && (errno == EWOULDBLOCK
723 || errno == EINTR || errno == EAGAIN)) {
724 struct timeval tv;
6c850772
WD
725 fd_set r_fds, e_fds;
726 FD_ZERO(&r_fds);
727 FD_SET(fd, &r_fds);
728 FD_ZERO(&e_fds);
729 FD_SET(fd, &e_fds);
e626b29e 730 tv.tv_sec = select_timeout;
56014c8c 731 tv.tv_usec = 0;
6c850772 732 if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
56014c8c 733 check_timeout();
6c850772
WD
734 if (FD_ISSET(fd, &e_fds)) {
735 rsyserr(FINFO, errno,
736 "select exception on fd %d", fd);
737 }
56014c8c
WD
738 continue;
739 }
740 if (cnt != 1)
741 break;
742 if (nulls? !ch : (ch == '\r' || ch == '\n')) {
743 /* Skip empty lines if reading locally. */
744 if (!reading_remotely && s == fname)
745 continue;
746 break;
747 }
748 if (s < eob)
749 *s++ = ch;
750 }
751 *s = '\0';
7a55d06e 752
6b45fcf1
WD
753 /* Dump comments. */
754 if (*fname == '#' || *fname == ';')
56014c8c
WD
755 goto start;
756
757 return s - fname;
758}
7a55d06e 759
8ef246e0 760int io_start_buffering_out(int f_out)
1f75bb10 761{
8ef246e0
WD
762 if (iobuf_out) {
763 assert(f_out == iobuf_f_out);
764 return 0;
765 }
1f75bb10
WD
766 if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
767 out_of_memory("io_start_buffering_out");
768 iobuf_out_cnt = 0;
8ef246e0
WD
769 iobuf_f_out = f_out;
770 return 1;
1f75bb10
WD
771}
772
8ef246e0 773int io_start_buffering_in(int f_in)
1f75bb10 774{
8ef246e0
WD
775 if (iobuf_in) {
776 assert(f_in == iobuf_f_in);
777 return 0;
778 }
1f75bb10
WD
779 iobuf_in_siz = 2 * IO_BUFFER_SIZE;
780 if (!(iobuf_in = new_array(char, iobuf_in_siz)))
781 out_of_memory("io_start_buffering_in");
8ef246e0
WD
782 iobuf_f_in = f_in;
783 return 1;
1f75bb10
WD
784}
785
8ef246e0 786void io_end_buffering_in(void)
1f75bb10 787{
8ef246e0
WD
788 if (!iobuf_in)
789 return;
790 free(iobuf_in);
791 iobuf_in = NULL;
792 iobuf_in_ndx = 0;
793 iobuf_in_remaining = 0;
794 iobuf_f_in = -1;
795}
796
797void io_end_buffering_out(void)
798{
799 if (!iobuf_out)
800 return;
801 io_flush(FULL_FLUSH);
802 free(iobuf_out);
803 iobuf_out = NULL;
804 iobuf_f_out = -1;
1f75bb10
WD
805}
806
be91bd81 807void maybe_flush_socket(int important)
626bec8e 808{
be91bd81
WD
809 if (iobuf_out && iobuf_out_cnt
810 && (important || time(NULL) - last_io_out >= 5))
626bec8e
WD
811 io_flush(NORMAL_FLUSH);
812}
813
cdf236aa 814void maybe_send_keepalive(void)
9e2409ab 815{
3b0a30eb 816 if (time(NULL) - last_io_out >= allowed_lull) {
9e2409ab 817 if (!iobuf_out || !iobuf_out_cnt) {
3221f451
WD
818 if (protocol_version < 29)
819 return; /* there's nothing we can do */
8ef246e0
WD
820 if (protocol_version >= 30)
821 send_msg(MSG_NOOP, "", 0);
822 else {
823 write_int(sock_f_out, cur_flist->count);
824 write_shortint(sock_f_out, ITEM_IS_NEW);
825 }
9e2409ab
WD
826 }
827 if (iobuf_out)
828 io_flush(NORMAL_FLUSH);
829 }
830}
831
8ef246e0
WD
832void start_flist_forward(int f_in)
833{
834 assert(iobuf_out != NULL);
835 assert(iobuf_f_out == msg_fd_out);
836 flist_forward_from = f_in;
837}
838
839void stop_flist_forward()
840{
8ef246e0 841 flist_forward_from = -1;
dde5b772 842 io_flush(FULL_FLUSH);
8ef246e0
WD
843}
844
880da007
MP
845/**
846 * Continue trying to read len bytes - don't return until len has been
847 * read.
848 **/
3151cbae 849static void read_loop(int fd, char *buf, size_t len)
4c36ddbe
AT
850{
851 while (len) {
852 int n = read_timeout(fd, buf, len);
853
854 buf += n;
855 len -= n;
8d9dc9f9
AT
856 }
857}
858
7a55d06e
MP
859/**
860 * Read from the file descriptor handling multiplexing - return number
861 * of bytes read.
d62bcc17
WD
862 *
863 * Never returns <= 0.
7a55d06e 864 */
c399d22a 865static int readfd_unbuffered(int fd, char *buf, size_t len)
8d9dc9f9 866{
f2a4853c 867 size_t msg_bytes;
d8aeda1e 868 int tag, cnt = 0;
33544bf4 869 char line[BIGPATHBUFLEN];
8d9dc9f9 870
8ef246e0 871 if (!iobuf_in || fd != iobuf_f_in)
4c36ddbe 872 return read_timeout(fd, buf, len);
8d9dc9f9 873
8ef246e0
WD
874 if (!io_multiplexing_in && iobuf_in_remaining == 0) {
875 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1f75bb10 876 iobuf_in_ndx = 0;
76c21947
WD
877 }
878
d8aeda1e 879 while (cnt == 0) {
8ef246e0
WD
880 if (iobuf_in_remaining) {
881 len = MIN(len, iobuf_in_remaining);
1f75bb10
WD
882 memcpy(buf, iobuf_in + iobuf_in_ndx, len);
883 iobuf_in_ndx += len;
8ef246e0 884 iobuf_in_remaining -= len;
d8aeda1e 885 cnt = len;
76c21947 886 break;
8d9dc9f9
AT
887 }
888
909ce14f 889 read_loop(fd, line, 4);
ff41a59f 890 tag = IVAL(line, 0);
679e7657 891
f2a4853c 892 msg_bytes = tag & 0xFFFFFF;
d17e1dd2 893 tag = (tag >> 24) - MPLEX_BASE;
8d9dc9f9 894
d17e1dd2
WD
895 switch (tag) {
896 case MSG_DATA:
f2a4853c 897 if (msg_bytes > iobuf_in_siz) {
1f75bb10 898 if (!(iobuf_in = realloc_array(iobuf_in, char,
f2a4853c 899 msg_bytes)))
c399d22a 900 out_of_memory("readfd_unbuffered");
f2a4853c 901 iobuf_in_siz = msg_bytes;
76c21947 902 }
f2a4853c 903 read_loop(fd, iobuf_in, msg_bytes);
8ef246e0 904 iobuf_in_remaining = msg_bytes;
1f75bb10 905 iobuf_in_ndx = 0;
d17e1dd2 906 break;
8ef246e0
WD
907 case MSG_NOOP:
908 if (am_sender)
909 maybe_send_keepalive();
910 break;
911 case MSG_IO_ERROR:
912 if (msg_bytes != 4)
913 goto invalid_msg;
914 read_loop(fd, line, msg_bytes);
915 io_error |= IVAL(line, 0);
916 break;
0d67e00a 917 case MSG_DELETED:
f2a4853c
WD
918 if (msg_bytes >= sizeof line)
919 goto overflow;
920 read_loop(fd, line, msg_bytes);
0d67e00a 921 /* A directory name was sent with the trailing null */
f2a4853c 922 if (msg_bytes > 0 && !line[msg_bytes-1])
0d67e00a 923 log_delete(line, S_IFDIR);
12bda6f7
WD
924 else {
925 line[msg_bytes] = '\0';
0d67e00a 926 log_delete(line, S_IFREG);
12bda6f7 927 }
0d67e00a 928 break;
9981c27e 929 case MSG_SUCCESS:
f2a4853c 930 if (msg_bytes != 4) {
8ef246e0 931 invalid_msg:
cdf236aa 932 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
f2a4853c 933 tag, (long)msg_bytes, who_am_i());
9981c27e
WD
934 exit_cleanup(RERR_STREAMIO);
935 }
f2a4853c 936 read_loop(fd, line, msg_bytes);
9981c27e 937 successful_send(IVAL(line, 0));
9981c27e 938 break;
8ef246e0
WD
939 case MSG_NO_SEND:
940 if (msg_bytes != 4)
941 goto invalid_msg;
942 read_loop(fd, line, msg_bytes);
943 send_msg_int(MSG_NO_SEND, IVAL(line, 0));
944 break;
d17e1dd2
WD
945 case MSG_INFO:
946 case MSG_ERROR:
f2a4853c
WD
947 if (msg_bytes >= sizeof line) {
948 overflow:
bd9fca47 949 rprintf(FERROR,
cdf236aa 950 "multiplexing overflow %d:%ld [%s]\n",
f2a4853c 951 tag, (long)msg_bytes, who_am_i());
d17e1dd2
WD
952 exit_cleanup(RERR_STREAMIO);
953 }
f2a4853c
WD
954 read_loop(fd, line, msg_bytes);
955 rwrite((enum logcode)tag, line, msg_bytes);
d17e1dd2
WD
956 break;
957 default:
cdf236aa
WD
958 rprintf(FERROR, "unexpected tag %d [%s]\n",
959 tag, who_am_i());
65417579 960 exit_cleanup(RERR_STREAMIO);
8d9dc9f9 961 }
8d9dc9f9
AT
962 }
963
8ef246e0 964 if (iobuf_in_remaining == 0)
d17e1dd2 965 io_flush(NORMAL_FLUSH);
76c21947 966
d8aeda1e 967 return cnt;
8d9dc9f9
AT
968}
969
dde5b772
WD
970/* Do a buffered read from fd. Don't return until all N bytes have
971 * been read. If all N can't be read then exit with an error. */
3151cbae 972static void readfd(int fd, char *buffer, size_t N)
720b47f2 973{
d8aeda1e 974 int cnt;
d62bcc17 975 size_t total = 0;
3151cbae 976
6ba9279f 977 while (total < N) {
d8aeda1e
WD
978 cnt = readfd_unbuffered(fd, buffer + total, N-total);
979 total += cnt;
7f28dbee 980 }
1b7c47cb 981
b9f592fb
WD
982 if (fd == write_batch_monitor_in) {
983 if ((size_t)write(batch_fd, buffer, total) != total)
984 exit_cleanup(RERR_FILEIO);
985 }
1f75bb10 986
8ef246e0
WD
987 if (fd == flist_forward_from)
988 writefd(iobuf_f_out, buffer, total);
989
1f75bb10
WD
990 if (fd == sock_f_in)
991 stats.total_read += total;
720b47f2
AT
992}
993
3a993aa4 994unsigned short read_shortint(int f)
9361f839 995{
482f48cc
WD
996 char b[2];
997 readfd(f, b, 2);
998 return (UVAL(b, 1) << 8) + UVAL(b, 0);
9361f839
WD
999}
1000
b7922338 1001int32 read_int(int f)
720b47f2 1002{
4c36ddbe 1003 char b[4];
d8aeda1e 1004 int32 num;
d730b113 1005
482f48cc
WD
1006 readfd(f, b, 4);
1007 num = IVAL(b, 0);
1008#if SIZEOF_INT32 > 4
1009 if (num & (int32)0x80000000)
1010 num |= ~(int32)0xffffffff;
1011#endif
d8aeda1e 1012 return num;
720b47f2
AT
1013}
1014
351e23ad 1015int32 read_varint(int f)
3a6a366f 1016{
351e23ad
WD
1017 union {
1018 char b[5];
1019 int32 x;
1020 } u;
1021 uchar ch;
1022 int extra;
1023
1024 u.x = 0;
1025 readfd(f, (char*)&ch, 1);
1026 extra = int_byte_extra[ch / 4];
1027 if (extra) {
1028 uchar bit = ((uchar)1<<(8-extra));
1029 if (extra >= (int)sizeof u.b) {
1030 rprintf(FERROR, "Overflow in read_varint()\n");
1031 exit_cleanup(RERR_STREAMIO);
1032 }
1033 readfd(f, u.b, extra);
1034 u.b[extra] = ch & (bit-1);
1035 } else
1036 u.b[0] = ch;
1037#if CAREFUL_ALIGNMENT
1038 u.x = IVAL(u.b,0);
1039#endif
1040#if SIZEOF_INT32 > 4
1041 if (u.x & (int32)0x80000000)
1042 u.x |= ~(int32)0xffffffff;
1043#endif
1044 return u.x;
1045}
71c46176 1046
351e23ad
WD
1047int64 read_varlong(int f, uchar min_bytes)
1048{
1049 union {
1050 char b[9];
1051 int64 x;
1052 } u;
1053 char b2[8];
1054 int extra;
71c46176 1055
031fa9ad 1056#if SIZEOF_INT64 < 8
351e23ad 1057 memset(u.b, 0, 8);
031fa9ad 1058#else
351e23ad 1059 u.x = 0;
031fa9ad 1060#endif
351e23ad
WD
1061 readfd(f, b2, min_bytes);
1062 memcpy(u.b, b2+1, min_bytes-1);
1063 extra = int_byte_extra[CVAL(b2, 0) / 4];
1064 if (extra) {
1065 uchar bit = ((uchar)1<<(8-extra));
1066 if (min_bytes + extra > (int)sizeof u.b) {
1067 rprintf(FERROR, "Overflow in read_varlong()\n");
1068 exit_cleanup(RERR_STREAMIO);
1069 }
1070 readfd(f, u.b + min_bytes - 1, extra);
1071 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
4ea4acf1 1072#if SIZEOF_INT64 < 8
351e23ad 1073 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
4ea4acf1
WD
1074 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1075 exit_cleanup(RERR_UNSUPPORTED);
1076 }
1077#endif
351e23ad
WD
1078 } else
1079 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1080#if SIZEOF_INT64 < 8
1081 u.x = IVAL(u.b,0);
1082#elif CAREFUL_ALIGNMENT
1083 u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1084#endif
1085 return u.x;
1086}
1087
1088int64 read_longint(int f)
1089{
4ea4acf1 1090#if SIZEOF_INT64 >= 8
351e23ad 1091 char b[9];
4ea4acf1 1092#endif
351e23ad 1093 int32 num = read_int(f);
71c46176 1094
351e23ad
WD
1095 if (num != (int32)0xffffffff)
1096 return num;
1097
1098#if SIZEOF_INT64 < 8
1099 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1100 exit_cleanup(RERR_UNSUPPORTED);
1101#else
1102 readfd(f, b, 8);
1103 return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1104#endif
3a6a366f
AT
1105}
1106
4a19c3b2 1107void read_buf(int f, char *buf, size_t len)
720b47f2 1108{
4c36ddbe 1109 readfd(f,buf,len);
720b47f2
AT
1110}
1111
4a19c3b2 1112void read_sbuf(int f, char *buf, size_t len)
575f2fca 1113{
93095cbe 1114 readfd(f, buf, len);
af436313 1115 buf[len] = '\0';
575f2fca
AT
1116}
1117
9361f839 1118uchar read_byte(int f)
182dca5c 1119{
9361f839 1120 uchar c;
93095cbe 1121 readfd(f, (char *)&c, 1);
4c36ddbe 1122 return c;
182dca5c 1123}
720b47f2 1124
46e99b09
WD
1125int read_vstring(int f, char *buf, int bufsize)
1126{
1127 int len = read_byte(f);
1128
1129 if (len & 0x80)
1130 len = (len & ~0x80) * 0x100 + read_byte(f);
1131
1132 if (len >= bufsize) {
1133 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1134 len, bufsize - 1);
9a6ed83f 1135 return -1;
46e99b09
WD
1136 }
1137
1138 if (len)
1139 readfd(f, buf, len);
1140 buf[len] = '\0';
1141 return len;
1142}
1143
188fed95
WD
1144/* Populate a sum_struct with values from the socket. This is
1145 * called by both the sender and the receiver. */
1146void read_sum_head(int f, struct sum_struct *sum)
1147{
1148 sum->count = read_int(f);
c638caa6
WD
1149 if (sum->count < 0) {
1150 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1151 (long)sum->count, who_am_i());
1152 exit_cleanup(RERR_PROTOCOL);
1153 }
188fed95
WD
1154 sum->blength = read_int(f);
1155 if (sum->blength < 0 || sum->blength > MAX_BLOCK_SIZE) {
cdf236aa
WD
1156 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1157 (long)sum->blength, who_am_i());
188fed95
WD
1158 exit_cleanup(RERR_PROTOCOL);
1159 }
1160 sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
a0456b9c 1161 if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
cdf236aa
WD
1162 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1163 sum->s2length, who_am_i());
188fed95
WD
1164 exit_cleanup(RERR_PROTOCOL);
1165 }
1166 sum->remainder = read_int(f);
1167 if (sum->remainder < 0 || sum->remainder > sum->blength) {
cdf236aa
WD
1168 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1169 (long)sum->remainder, who_am_i());
188fed95
WD
1170 exit_cleanup(RERR_PROTOCOL);
1171 }
1172}
1173
c207d7ec
WD
1174/* Send the values from a sum_struct over the socket. Set sum to
1175 * NULL if there are no checksums to send. This is called by both
1176 * the generator and the sender. */
1177void write_sum_head(int f, struct sum_struct *sum)
1178{
1179 static struct sum_struct null_sum;
1180
1181 if (sum == NULL)
1182 sum = &null_sum;
1183
1184 write_int(f, sum->count);
1185 write_int(f, sum->blength);
1186 if (protocol_version >= 27)
1187 write_int(f, sum->s2length);
1188 write_int(f, sum->remainder);
1189}
1190
08571358
MP
1191/**
1192 * Sleep after writing to limit I/O bandwidth usage.
1193 *
1194 * @todo Rather than sleeping after each write, it might be better to
1195 * use some kind of averaging. The current algorithm seems to always
1196 * use a bit less bandwidth than specified, because it doesn't make up
1197 * for slow periods. But arguably this is a feature. In addition, we
1198 * ought to take the time used to write the data into account.
71e58630
WD
1199 *
1200 * During some phases of big transfers (file FOO is uptodate) this is
1201 * called with a small bytes_written every time. As the kernel has to
1202 * round small waits up to guarantee that we actually wait at least the
1203 * requested number of microseconds, this can become grossly inaccurate.
1204 * We therefore keep track of the bytes we've written over time and only
1205 * sleep when the accumulated delay is at least 1 tenth of a second.
08571358
MP
1206 **/
1207static void sleep_for_bwlimit(int bytes_written)
1208{
71e58630 1209 static struct timeval prior_tv;
0f78b815 1210 static long total_written = 0;
71e58630
WD
1211 struct timeval tv, start_tv;
1212 long elapsed_usec, sleep_usec;
1213
1214#define ONE_SEC 1000000L /* # of microseconds in a second */
08571358 1215
9a0cfff5 1216 if (!bwlimit_writemax)
08571358 1217 return;
e681e820 1218
0f78b815 1219 total_written += bytes_written;
71e58630
WD
1220
1221 gettimeofday(&start_tv, NULL);
1222 if (prior_tv.tv_sec) {
1223 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1224 + (start_tv.tv_usec - prior_tv.tv_usec);
1225 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1226 if (total_written < 0)
1227 total_written = 0;
1228 }
3151cbae 1229
71e58630
WD
1230 sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1231 if (sleep_usec < ONE_SEC / 10) {
1232 prior_tv = start_tv;
1233 return;
1234 }
08571358 1235
71e58630
WD
1236 tv.tv_sec = sleep_usec / ONE_SEC;
1237 tv.tv_usec = sleep_usec % ONE_SEC;
98b332ed 1238 select(0, NULL, NULL, NULL, &tv);
71e58630
WD
1239
1240 gettimeofday(&prior_tv, NULL);
1241 elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1242 + (prior_tv.tv_usec - start_tv.tv_usec);
1243 total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
08571358
MP
1244}
1245
1f75bb10 1246/* Write len bytes to the file descriptor fd, looping as necessary to get
0fb2fc4a 1247 * the job done and also (in certain circumstances) reading any data on
af436313 1248 * msg_fd_in to avoid deadlock.
880da007
MP
1249 *
1250 * This function underlies the multiplexing system. The body of the
1f75bb10 1251 * application never calls this function directly. */
4a19c3b2 1252static void writefd_unbuffered(int fd, const char *buf, size_t len)
720b47f2 1253{
1ea087a7 1254 size_t n, total = 0;
6c850772 1255 fd_set w_fds, r_fds, e_fds;
d8aeda1e 1256 int maxfd, count, cnt, using_r_fds;
7d51b837 1257 int defer_save = defer_forwarding_messages;
8d9dc9f9 1258 struct timeval tv;
720b47f2 1259
3be97bf9
WD
1260 if (no_flush++)
1261 defer_forwarding_messages = 1;
e44f9a12 1262
4c36ddbe 1263 while (total < len) {
8d9dc9f9 1264 FD_ZERO(&w_fds);
6c850772
WD
1265 FD_SET(fd, &w_fds);
1266 FD_ZERO(&e_fds);
1267 FD_SET(fd, &e_fds);
1ea087a7 1268 maxfd = fd;
4c36ddbe 1269
954bbed8 1270 if (msg_fd_in >= 0) {
56014c8c 1271 FD_ZERO(&r_fds);
6c850772 1272 FD_SET(msg_fd_in, &r_fds);
1ea087a7
WD
1273 if (msg_fd_in > maxfd)
1274 maxfd = msg_fd_in;
3eeac9bc
WD
1275 using_r_fds = 1;
1276 } else
1277 using_r_fds = 0;
8d9dc9f9 1278
e626b29e 1279 tv.tv_sec = select_timeout;
8d9dc9f9 1280 tv.tv_usec = 0;
4c36ddbe 1281
554e0a8d 1282 errno = 0;
3eeac9bc 1283 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
6c850772 1284 &w_fds, &e_fds, &tv);
4c36ddbe
AT
1285
1286 if (count <= 0) {
1ea087a7 1287 if (count < 0 && errno == EBADF)
554e0a8d 1288 exit_cleanup(RERR_SOCKETIO);
bd717af8 1289 check_timeout();
8d9dc9f9
AT
1290 continue;
1291 }
4c36ddbe 1292
6c850772
WD
1293 if (FD_ISSET(fd, &e_fds)) {
1294 rsyserr(FINFO, errno,
1295 "select exception on fd %d", fd);
1296 }
1297
3eeac9bc 1298 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
d17e1dd2 1299 read_msg_fd();
554e0a8d 1300
9a6ed83f 1301 if (!FD_ISSET(fd, &w_fds))
1ea087a7 1302 continue;
4c36ddbe 1303
1ea087a7 1304 n = len - total;
9a0cfff5 1305 if (bwlimit_writemax && n > bwlimit_writemax)
1ea087a7 1306 n = bwlimit_writemax;
d8aeda1e 1307 cnt = write(fd, buf + total, n);
1ea087a7 1308
d8aeda1e
WD
1309 if (cnt <= 0) {
1310 if (cnt < 0) {
3309507d
WD
1311 if (errno == EINTR)
1312 continue;
1313 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1314 msleep(1);
1315 continue;
1316 }
f0359dd0
AT
1317 }
1318
1ea087a7 1319 /* Don't try to write errors back across the stream. */
1f75bb10 1320 if (fd == sock_f_out)
8ef246e0 1321 io_end_multiplex_out();
2a5df862
WD
1322 /* Don't try to write errors down a failing msg pipe. */
1323 if (am_server && fd == msg_fd_out)
1324 exit_cleanup(RERR_STREAMIO);
1ea087a7 1325 rsyserr(FERROR, errno,
b88c2e8f
WD
1326 "writefd_unbuffered failed to write %ld bytes [%s]",
1327 (long)len, who_am_i());
d1b31da7
WD
1328 /* If the other side is sending us error messages, try
1329 * to grab any messages they sent before they died. */
7f459268 1330 while (fd == sock_f_out && io_multiplexing_in) {
3b0a30eb 1331 set_io_timeout(30);
9e2409ab 1332 ignore_timeout = 0;
d1b31da7
WD
1333 readfd_unbuffered(sock_f_in, io_filesfrom_buf,
1334 sizeof io_filesfrom_buf);
1335 }
1ea087a7
WD
1336 exit_cleanup(RERR_STREAMIO);
1337 }
4c36ddbe 1338
d8aeda1e 1339 total += cnt;
954bbed8 1340 defer_forwarding_messages = 1;
a800434a 1341
1f75bb10 1342 if (fd == sock_f_out) {
626bec8e 1343 if (io_timeout || am_generator)
3b0a30eb 1344 last_io_out = time(NULL);
d8aeda1e 1345 sleep_for_bwlimit(cnt);
1f75bb10 1346 }
4c36ddbe 1347 }
e44f9a12
AT
1348
1349 no_flush--;
3be97bf9 1350 if (!(defer_forwarding_messages = defer_save))
7d51b837 1351 msg2sndr_flush();
d6dead6b
AT
1352}
1353
dde5b772 1354void io_flush(int flush_it_all)
d6dead6b 1355{
3be97bf9 1356 if (flush_it_all && !defer_forwarding_messages)
dde5b772 1357 msg2sndr_flush();
90ba34e2 1358
1f75bb10 1359 if (!iobuf_out_cnt || no_flush)
d17e1dd2 1360 return;
8d9dc9f9 1361
d17e1dd2 1362 if (io_multiplexing_out)
3be97bf9 1363 mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt);
d17e1dd2 1364 else
8ef246e0 1365 writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1f75bb10 1366 iobuf_out_cnt = 0;
8d9dc9f9
AT
1367}
1368
4a19c3b2 1369static void writefd(int fd, const char *buf, size_t len)
d6dead6b 1370{
1f75bb10
WD
1371 if (fd == sock_f_out)
1372 stats.total_written += len;
1373
b9f592fb
WD
1374 if (fd == write_batch_monitor_out) {
1375 if ((size_t)write(batch_fd, buf, len) != len)
1376 exit_cleanup(RERR_FILEIO);
1377 }
1378
8ef246e0 1379 if (!iobuf_out || fd != iobuf_f_out) {
4c36ddbe
AT
1380 writefd_unbuffered(fd, buf, len);
1381 return;
1382 }
d6dead6b
AT
1383
1384 while (len) {
1f75bb10 1385 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
d6dead6b 1386 if (n > 0) {
1f75bb10 1387 memcpy(iobuf_out+iobuf_out_cnt, buf, n);
d6dead6b
AT
1388 buf += n;
1389 len -= n;
1f75bb10 1390 iobuf_out_cnt += n;
d6dead6b 1391 }
3151cbae 1392
1f75bb10 1393 if (iobuf_out_cnt == IO_BUFFER_SIZE)
d17e1dd2 1394 io_flush(NORMAL_FLUSH);
d6dead6b 1395 }
d6dead6b 1396}
720b47f2 1397
3a993aa4 1398void write_shortint(int f, unsigned short x)
9361f839 1399{
3a993aa4
WD
1400 char b[2];
1401 b[0] = (char)x;
1402 b[1] = (char)(x >> 8);
1403 writefd(f, b, 2);
9361f839
WD
1404}
1405
351e23ad
WD
1406void write_int(int f, int32 x)
1407{
1408 char b[4];
1409 SIVAL(b, 0, x);
1410 writefd(f, b, 4);
1411}
1412
f31514ad 1413void write_varint(int f, int32 x)
1c3344a1
WD
1414{
1415 char b[5];
351e23ad
WD
1416 uchar bit;
1417 int cnt = 4;
1418
1419 SIVAL(b, 1, x);
1420
1421 while (cnt > 1 && b[cnt] == 0)
1422 cnt--;
1423 bit = ((uchar)1<<(7-cnt+1));
1424 if (CVAL(b, cnt) >= bit) {
1425 cnt++;
1426 *b = ~(bit-1);
1427 } else if (cnt > 1)
1428 *b = b[cnt] | ~(bit*2-1);
1429 else
1430 *b = b[cnt];
1431
1432 writefd(f, b, cnt);
1c3344a1
WD
1433}
1434
351e23ad 1435void write_varlong(int f, int64 x, uchar min_bytes)
720b47f2 1436{
351e23ad
WD
1437 char b[9];
1438 uchar bit;
1439 int cnt = 8;
1440
1441 SIVAL(b, 1, x);
1442#if SIZEOF_INT64 >= 8
1443 SIVAL(b, 5, x >> 32);
1444#else
1445 if (x <= 0x7FFFFFFF && x >= 0)
1446 memset(b + 5, 0, 4);
1447 else {
1448 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1449 exit_cleanup(RERR_UNSUPPORTED);
1450 }
1451#endif
1452
1453 while (cnt > min_bytes && b[cnt] == 0)
1454 cnt--;
1455 bit = ((uchar)1<<(7-cnt+min_bytes));
1456 if (CVAL(b, cnt) >= bit) {
1457 cnt++;
1458 *b = ~(bit-1);
1459 } else if (cnt > min_bytes)
1460 *b = b[cnt] | ~(bit*2-1);
1461 else
1462 *b = b[cnt];
1463
1464 writefd(f, b, cnt);
720b47f2
AT
1465}
1466
7a24c346
MP
1467/*
1468 * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1469 * 64-bit types on this platform.
1470 */
71c46176 1471void write_longint(int f, int64 x)
3a6a366f 1472{
351e23ad 1473 char b[12], * const s = b+4;
3a6a366f 1474
351e23ad
WD
1475 SIVAL(s, 0, x);
1476 if (x <= 0x7FFFFFFF && x >= 0) {
482f48cc 1477 writefd(f, s, 4);
351e23ad
WD
1478 return;
1479 }
3a6a366f 1480
4ea4acf1 1481#if SIZEOF_INT64 < 8
351e23ad
WD
1482 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1483 exit_cleanup(RERR_UNSUPPORTED);
4ea4acf1 1484#else
351e23ad
WD
1485 memset(b, 0xFF, 4);
1486 SIVAL(s, 4, x >> 32);
1487 writefd(f, b, 12);
4ea4acf1 1488#endif
3a6a366f
AT
1489}
1490
4a19c3b2 1491void write_buf(int f, const char *buf, size_t len)
720b47f2 1492{
4c36ddbe 1493 writefd(f,buf,len);
720b47f2
AT
1494}
1495
880da007 1496/** Write a string to the connection */
4a19c3b2 1497void write_sbuf(int f, const char *buf)
f0fca04e 1498{
93095cbe 1499 writefd(f, buf, strlen(buf));
f0fca04e
AT
1500}
1501
9361f839 1502void write_byte(int f, uchar c)
182dca5c 1503{
93095cbe 1504 writefd(f, (char *)&c, 1);
182dca5c
AT
1505}
1506
4a19c3b2 1507void write_vstring(int f, const char *str, int len)
46e99b09
WD
1508{
1509 uchar lenbuf[3], *lb = lenbuf;
1510
1511 if (len > 0x7F) {
1512 if (len > 0x7FFF) {
1513 rprintf(FERROR,
1514 "attempting to send over-long vstring (%d > %d)\n",
1515 len, 0x7FFF);
1516 exit_cleanup(RERR_PROTOCOL);
1517 }
1518 *lb++ = len / 0x100 + 0x80;
1519 }
1520 *lb = len;
1521
1522 writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1523 if (len)
1524 writefd(f, str, len);
1525}
1526
8a65e0ce
WD
1527/* Send a file-list index using a byte-reduction method. */
1528void write_ndx(int f, int32 ndx)
1529{
1530 static int32 prev_positive = -1, prev_negative = 1;
1531 int32 diff, cnt = 0;
1532 char b[6];
1533
1534 if (protocol_version < 30 || read_batch) {
1535 write_int(f, ndx);
1536 return;
1537 }
1538
1539 /* Send NDX_DONE as a single-byte 0 with no side effects. Send
1540 * negative nums as a positive after sending a leading 0xFF. */
1541 if (ndx >= 0) {
1542 diff = ndx - prev_positive;
1543 prev_positive = ndx;
1544 } else if (ndx == NDX_DONE) {
1545 *b = 0;
1546 writefd(f, b, 1);
1547 return;
1548 } else {
1549 b[cnt++] = (char)0xFF;
1550 ndx = -ndx;
1551 diff = ndx - prev_negative;
1552 prev_negative = ndx;
1553 }
1554
1555 /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1556 * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1557 * & all 4 bytes of the (non-negative) num with the high-bit set. */
1558 if (diff < 0xFE && diff > 0)
1559 b[cnt++] = (char)diff;
1560 else if (diff < 0 || diff > 0x7FFF) {
1561 b[cnt++] = (char)0xFE;
1562 b[cnt++] = (char)((ndx >> 24) | 0x80);
8a65e0ce 1563 b[cnt++] = (char)ndx;
351e23ad
WD
1564 b[cnt++] = (char)(ndx >> 8);
1565 b[cnt++] = (char)(ndx >> 16);
8a65e0ce
WD
1566 } else {
1567 b[cnt++] = (char)0xFE;
1568 b[cnt++] = (char)(diff >> 8);
1569 b[cnt++] = (char)diff;
1570 }
1571 writefd(f, b, cnt);
1572}
1573
1574/* Receive a file-list index using a byte-reduction method. */
1575int32 read_ndx(int f)
1576{
1577 static int32 prev_positive = -1, prev_negative = 1;
1578 int32 *prev_ptr, num;
1579 char b[4];
1580
1581 if (protocol_version < 30)
1582 return read_int(f);
1583
1584 readfd(f, b, 1);
1585 if (CVAL(b, 0) == 0xFF) {
1586 readfd(f, b, 1);
1587 prev_ptr = &prev_negative;
1588 } else if (CVAL(b, 0) == 0)
1589 return NDX_DONE;
1590 else
1591 prev_ptr = &prev_positive;
1592 if (CVAL(b, 0) == 0xFE) {
1593 readfd(f, b, 2);
1594 if (CVAL(b, 0) & 0x80) {
351e23ad
WD
1595 b[3] = CVAL(b, 0) & ~0x80;
1596 b[0] = b[1];
1597 readfd(f, b+1, 2);
1598 num = IVAL(b, 0);
8a65e0ce 1599 } else
351e23ad 1600 num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
8a65e0ce 1601 } else
351e23ad 1602 num = UVAL(b, 0) + *prev_ptr;
8a65e0ce
WD
1603 *prev_ptr = num;
1604 if (prev_ptr == &prev_negative)
1605 num = -num;
1606 return num;
1607}
1608
914cc65c 1609/**
6ed6d7f5
WD
1610 * Read a line of up to @p maxlen characters into @p buf (not counting
1611 * the trailing null). Strips the (required) trailing newline and all
1612 * carriage returns.
914cc65c 1613 *
6ed6d7f5 1614 * @return 1 for success; 0 for I/O error or truncation.
914cc65c 1615 **/
9dd891bb 1616int read_line(int f, char *buf, size_t maxlen)
f0fca04e
AT
1617{
1618 while (maxlen) {
528bfcd7 1619 buf[0] = 0;
f0fca04e 1620 read_buf(f, buf, 1);
914cc65c
MP
1621 if (buf[0] == 0)
1622 return 0;
6ed6d7f5 1623 if (buf[0] == '\n')
f0fca04e 1624 break;
f0fca04e
AT
1625 if (buf[0] != '\r') {
1626 buf++;
1627 maxlen--;
1628 }
1629 }
6ed6d7f5
WD
1630 *buf = '\0';
1631 return maxlen > 0;
f0fca04e
AT
1632}
1633
f0fca04e
AT
1634void io_printf(int fd, const char *format, ...)
1635{
d62bcc17 1636 va_list ap;
33544bf4 1637 char buf[BIGPATHBUFLEN];
f0fca04e 1638 int len;
3151cbae 1639
f0fca04e 1640 va_start(ap, format);
3151cbae 1641 len = vsnprintf(buf, sizeof buf, format, ap);
f0fca04e
AT
1642 va_end(ap);
1643
f89b9368
WD
1644 if (len < 0)
1645 exit_cleanup(RERR_STREAMIO);
f0fca04e 1646
33544bf4
WD
1647 if (len > (int)sizeof buf) {
1648 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1649 exit_cleanup(RERR_STREAMIO);
1650 }
1651
f0fca04e
AT
1652 write_sbuf(fd, buf);
1653}
8d9dc9f9 1654
d17e1dd2 1655/** Setup for multiplexing a MSG_* stream with the data stream. */
1f75bb10 1656void io_start_multiplex_out(void)
8d9dc9f9 1657{
d17e1dd2 1658 io_flush(NORMAL_FLUSH);
8ef246e0 1659 io_start_buffering_out(sock_f_out);
8d9dc9f9
AT
1660 io_multiplexing_out = 1;
1661}
1662
d17e1dd2 1663/** Setup for multiplexing a MSG_* stream with the data stream. */
1f75bb10 1664void io_start_multiplex_in(void)
8d9dc9f9 1665{
d17e1dd2 1666 io_flush(NORMAL_FLUSH);
8ef246e0 1667 io_start_buffering_in(sock_f_in);
8d9dc9f9
AT
1668 io_multiplexing_in = 1;
1669}
1670
d17e1dd2 1671/** Write an message to the multiplexed data stream. */
4a19c3b2 1672int io_multiplex_write(enum msgcode code, const char *buf, size_t len)
8d9dc9f9 1673{
f89b9368
WD
1674 if (!io_multiplexing_out)
1675 return 0;
d17e1dd2 1676 io_flush(NORMAL_FLUSH);
1b7c47cb 1677 stats.total_written += (len+4);
3be97bf9 1678 mplex_write(sock_f_out, code, buf, len);
8d9dc9f9
AT
1679 return 1;
1680}
1681
8ef246e0 1682void io_end_multiplex_in(void)
7f459268
WD
1683{
1684 io_multiplexing_in = 0;
8ef246e0 1685 io_end_buffering_in();
7f459268
WD
1686}
1687
d17e1dd2 1688/** Stop output multiplexing. */
8ef246e0 1689void io_end_multiplex_out(void)
554e0a8d
AT
1690{
1691 io_multiplexing_out = 0;
8ef246e0 1692 io_end_buffering_out();
554e0a8d
AT
1693}
1694
b9f592fb
WD
1695void start_write_batch(int fd)
1696{
1697 /* Some communication has already taken place, but we don't
1698 * enable batch writing until here so that we can write a
1699 * canonical record of the communication even though the
1700 * actual communication so far depends on whether a daemon
1701 * is involved. */
1702 write_int(batch_fd, protocol_version);
1703 write_int(batch_fd, checksum_seed);
b9f592fb
WD
1704
1705 if (am_sender)
1706 write_batch_monitor_out = fd;
1707 else
1708 write_batch_monitor_in = fd;
1709}
1710
1711void stop_write_batch(void)
1712{
1713 write_batch_monitor_out = -1;
1714 write_batch_monitor_in = -1;
1715}