1752d86f1ccec2896fc4ba975b4f4d3cd1504d4e
[rsync/rsync.git] / io.c
1 /*
2  * Socket and pipe I/O utilities used in rsync.
3  *
4  * Copyright (C) 1996-2001 Andrew Tridgell
5  * Copyright (C) 1996 Paul Mackerras
6  * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7  * Copyright (C) 2003-2008 Wayne Davison
8  *
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.
13  *
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.
18  *
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.
21  */
22
23 /* Rsync provides its own multiplexing system, which is used to send
24  * stderr and stdout over a single socket.
25  *
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(). */
29
30 #include "rsync.h"
31 #include "ifuncs.h"
32
33 /** If no timeout is specified then use a 60 second select timeout */
34 #define SELECT_TIMEOUT 60
35
36 extern int bwlimit;
37 extern size_t bwlimit_writemax;
38 extern int io_timeout;
39 extern int am_server;
40 extern int am_daemon;
41 extern int am_sender;
42 extern int am_generator;
43 extern int inc_recurse;
44 extern int io_error;
45 extern int eol_nulls;
46 extern int flist_eof;
47 extern int list_only;
48 extern int read_batch;
49 extern int protect_args;
50 extern int checksum_seed;
51 extern int protocol_version;
52 extern int remove_source_files;
53 extern int preserve_hard_links;
54 extern struct stats stats;
55 extern struct file_list *cur_flist;
56 #ifdef ICONV_OPTION
57 extern int filesfrom_convert;
58 extern iconv_t ic_send, ic_recv;
59 #endif
60
61 int csum_length = SHORT_SUM_LENGTH; /* initial value */
62 int allowed_lull = 0;
63 int ignore_timeout = 0;
64 int batch_fd = -1;
65 int msgdone_cnt = 0;
66
67 /* Ignore an EOF error if non-zero. See whine_about_eof(). */
68 int kluge_around_eof = 0;
69
70 int msg_fd_in = -1;
71 int msg_fd_out = -1;
72 int sock_f_in = -1;
73 int sock_f_out = -1;
74
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;
80
81 static int iobuf_f_out = -1;
82 static char *iobuf_out;
83 static int iobuf_out_cnt;
84
85 int flist_forward_from = -1;
86
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;
91 static int no_flush;
92
93 static int write_batch_monitor_in = -1;
94 static int write_batch_monitor_out = -1;
95
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;
100 #ifdef ICONV_OPTION
101 static xbuf iconv_buf = EMPTY_XBUF;
102 #endif
103 static int defer_forwarding_messages = 0, keep_defer_forwarding = 0;
104 static int select_timeout = SELECT_TIMEOUT;
105 static int active_filecnt = 0;
106 static OFF_T active_bytecnt = 0;
107 static int first_message = 1;
108
109 static char int_byte_extra[64] = {
110         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (00 - 3F)/4 */
111         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* (40 - 7F)/4 */
112         1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* (80 - BF)/4 */
113         2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 5, 6, /* (C0 - FF)/4 */
114 };
115
116 #define REMOTE_OPTION_ERROR "rsync: on remote machine: -"
117 #define REMOTE_OPTION_ERROR2 ": unknown option"
118
119 enum festatus { FES_SUCCESS, FES_REDO, FES_NO_SEND };
120
121 static void readfd(int fd, char *buffer, size_t N);
122 static void writefd(int fd, const char *buf, size_t len);
123 static void writefd_unbuffered(int fd, const char *buf, size_t len);
124 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert);
125
126 struct flist_ndx_item {
127         struct flist_ndx_item *next;
128         int ndx;
129 };
130
131 struct flist_ndx_list {
132         struct flist_ndx_item *head, *tail;
133 };
134
135 static struct flist_ndx_list redo_list, hlink_list;
136
137 struct msg_list_item {
138         struct msg_list_item *next;
139         char convert;
140         char buf[1];
141 };
142
143 struct msg_list {
144         struct msg_list_item *head, *tail;
145 };
146
147 static struct msg_list msg_queue;
148
149 static void flist_ndx_push(struct flist_ndx_list *lp, int ndx)
150 {
151         struct flist_ndx_item *item;
152
153         if (!(item = new(struct flist_ndx_item)))
154                 out_of_memory("flist_ndx_push");
155         item->next = NULL;
156         item->ndx = ndx;
157         if (lp->tail)
158                 lp->tail->next = item;
159         else
160                 lp->head = item;
161         lp->tail = item;
162 }
163
164 static int flist_ndx_pop(struct flist_ndx_list *lp)
165 {
166         struct flist_ndx_item *next;
167         int ndx;
168
169         if (!lp->head)
170                 return -1;
171
172         ndx = lp->head->ndx;
173         next = lp->head->next;
174         free(lp->head);
175         lp->head = next;
176         if (!next)
177                 lp->tail = NULL;
178
179         return ndx;
180 }
181
182 static void got_flist_entry_status(enum festatus status, const char *buf)
183 {
184         int ndx = IVAL(buf, 0);
185         struct file_list *flist = flist_for_ndx(ndx, "got_flist_entry_status");
186
187         if (remove_source_files) {
188                 active_filecnt--;
189                 active_bytecnt -= F_LENGTH(flist->files[ndx - flist->ndx_start]);
190         }
191
192         if (inc_recurse)
193                 flist->in_progress--;
194
195         switch (status) {
196         case FES_SUCCESS:
197                 if (remove_source_files)
198                         send_msg(MSG_SUCCESS, buf, 4, 0);
199                 if (preserve_hard_links) {
200                         struct file_struct *file = flist->files[ndx - flist->ndx_start];
201                         if (F_IS_HLINKED(file)) {
202                                 flist_ndx_push(&hlink_list, ndx);
203                                 flist->in_progress++;
204                         }
205                 }
206                 break;
207         case FES_REDO:
208                 if (inc_recurse)
209                         flist->to_redo++;
210                 flist_ndx_push(&redo_list, ndx);
211                 break;
212         case FES_NO_SEND:
213                 break;
214         }
215 }
216
217 static void check_timeout(void)
218 {
219         time_t t;
220
221         if (!io_timeout || ignore_timeout)
222                 return;
223
224         if (!last_io_in) {
225                 last_io_in = time(NULL);
226                 return;
227         }
228
229         t = time(NULL);
230
231         if (t - last_io_in >= io_timeout) {
232                 if (!am_server && !am_daemon) {
233                         rprintf(FERROR, "io timeout after %d seconds -- exiting\n",
234                                 (int)(t-last_io_in));
235                 }
236                 exit_cleanup(RERR_TIMEOUT);
237         }
238 }
239
240 /* Note the fds used for the main socket (which might really be a pipe
241  * for a local transfer, but we can ignore that). */
242 void io_set_sock_fds(int f_in, int f_out)
243 {
244         sock_f_in = f_in;
245         sock_f_out = f_out;
246 }
247
248 void set_io_timeout(int secs)
249 {
250         io_timeout = secs;
251
252         if (!io_timeout || io_timeout > SELECT_TIMEOUT)
253                 select_timeout = SELECT_TIMEOUT;
254         else
255                 select_timeout = io_timeout;
256
257         allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2;
258 }
259
260 /* Setup the fd used to receive MSG_* messages.  Only needed during the
261  * early stages of being a local sender (up through the sending of the
262  * file list) or when we're the generator (to fetch the messages from
263  * the receiver). */
264 void set_msg_fd_in(int fd)
265 {
266         msg_fd_in = fd;
267 }
268
269 /* Setup the fd used to send our MSG_* messages.  Only needed when
270  * we're the receiver (to send our messages to the generator). */
271 void set_msg_fd_out(int fd)
272 {
273         msg_fd_out = fd;
274         set_nonblocking(msg_fd_out);
275 }
276
277 /* Add a message to the pending MSG_* list. */
278 static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len, int convert)
279 {
280         struct msg_list_item *m;
281         int sz = len + 4 + sizeof m[0] - 1;
282
283         if (!(m = (struct msg_list_item *)new_array(char, sz)))
284                 out_of_memory("msg_list_add");
285         m->next = NULL;
286         m->convert = convert;
287         SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len);
288         memcpy(m->buf + 4, buf, len);
289         if (lst->tail)
290                 lst->tail->next = m;
291         else
292                 lst->head = m;
293         lst->tail = m;
294 }
295
296 static inline int flush_a_msg(int fd)
297 {
298         struct msg_list_item *m = msg_queue.head;
299         int len = IVAL(m->buf, 0) & 0xFFFFFF;
300         int tag = *((uchar*)m->buf+3) - MPLEX_BASE;
301
302         if (!(msg_queue.head = m->next))
303                 msg_queue.tail = NULL;
304
305         defer_forwarding_messages++;
306         mplex_write(fd, tag, m->buf + 4, len, m->convert);
307         defer_forwarding_messages--;
308
309         free(m);
310
311         return len;
312 }
313
314 static void msg_flush(void)
315 {
316         if (am_generator) {
317                 while (msg_queue.head && io_multiplexing_out)
318                         stats.total_written += flush_a_msg(sock_f_out) + 4;
319         } else {
320                 while (msg_queue.head)
321                         (void)flush_a_msg(msg_fd_out);
322         }
323 }
324
325 static void check_for_d_option_error(const char *msg)
326 {
327         static char rsync263_opts[] = "BCDHIKLPRSTWabceghlnopqrtuvxz";
328         char *colon;
329         int saw_d = 0;
330
331         if (*msg != 'r'
332          || strncmp(msg, REMOTE_OPTION_ERROR, sizeof REMOTE_OPTION_ERROR - 1) != 0)
333                 return;
334
335         msg += sizeof REMOTE_OPTION_ERROR - 1;
336         if (*msg == '-' || (colon = strchr(msg, ':')) == NULL
337          || strncmp(colon, REMOTE_OPTION_ERROR2, sizeof REMOTE_OPTION_ERROR2 - 1) != 0)
338                 return;
339
340         for ( ; *msg != ':'; msg++) {
341                 if (*msg == 'd')
342                         saw_d = 1;
343                 else if (*msg == 'e')
344                         break;
345                 else if (strchr(rsync263_opts, *msg) == NULL)
346                         return;
347         }
348
349         if (saw_d) {
350                 rprintf(FWARNING,
351                     "*** Try using \"--old-d\" if remote rsync is <= 2.6.3 ***\n");
352         }
353 }
354
355 /* Read a message from the MSG_* fd and handle it.  This is called either
356  * during the early stages of being a local sender (up through the sending
357  * of the file list) or when we're the generator (to fetch the messages
358  * from the receiver). */
359 static void read_msg_fd(void)
360 {
361         char buf[2048];
362         size_t n;
363         struct file_list *flist;
364         int fd = msg_fd_in;
365         int tag, len;
366
367         /* Temporarily disable msg_fd_in.  This is needed to avoid looping back
368          * to this routine from writefd_unbuffered(). */
369         no_flush++;
370         msg_fd_in = -1;
371         defer_forwarding_messages++;
372
373         readfd(fd, buf, 4);
374         tag = IVAL(buf, 0);
375
376         len = tag & 0xFFFFFF;
377         tag = (tag >> 24) - MPLEX_BASE;
378
379         switch (tag) {
380         case MSG_DONE:
381                 if (len < 0 || len > 1 || !am_generator) {
382                   invalid_msg:
383                         rprintf(FERROR, "invalid message %d:%d [%s%s]\n",
384                                 tag, len, who_am_i(),
385                                 inc_recurse ? "/inc" : "");
386                         exit_cleanup(RERR_STREAMIO);
387                 }
388                 if (len) {
389                         readfd(fd, buf, len);
390                         stats.total_read = read_varlong(fd, 3);
391                 }
392                 msgdone_cnt++;
393                 break;
394         case MSG_REDO:
395                 if (len != 4 || !am_generator)
396                         goto invalid_msg;
397                 readfd(fd, buf, 4);
398                 got_flist_entry_status(FES_REDO, buf);
399                 break;
400         case MSG_FLIST:
401                 if (len != 4 || !am_generator || !inc_recurse)
402                         goto invalid_msg;
403                 readfd(fd, buf, 4);
404                 /* Read extra file list from receiver. */
405                 assert(iobuf_in != NULL);
406                 assert(iobuf_f_in == fd);
407                 if (DEBUG_GTE(FLIST, 2)) {
408                         rprintf(FINFO, "[%s] receiving flist for dir %d\n",
409                                 who_am_i(), IVAL(buf,0));
410                 }
411                 flist = recv_file_list(fd);
412                 flist->parent_ndx = IVAL(buf,0);
413 #ifdef SUPPORT_HARD_LINKS
414                 if (preserve_hard_links)
415                         match_hard_links(flist);
416 #endif
417                 break;
418         case MSG_FLIST_EOF:
419                 if (len != 0 || !am_generator || !inc_recurse)
420                         goto invalid_msg;
421                 flist_eof = 1;
422                 break;
423         case MSG_IO_ERROR:
424                 if (len != 4)
425                         goto invalid_msg;
426                 readfd(fd, buf, len);
427                 io_error |= IVAL(buf, 0);
428                 break;
429         case MSG_DELETED:
430                 if (len >= (int)sizeof buf || !am_generator)
431                         goto invalid_msg;
432                 readfd(fd, buf, len);
433                 send_msg(MSG_DELETED, buf, len, 1);
434                 break;
435         case MSG_SUCCESS:
436                 if (len != 4 || !am_generator)
437                         goto invalid_msg;
438                 readfd(fd, buf, 4);
439                 got_flist_entry_status(FES_SUCCESS, buf);
440                 break;
441         case MSG_NO_SEND:
442                 if (len != 4 || !am_generator)
443                         goto invalid_msg;
444                 readfd(fd, buf, 4);
445                 got_flist_entry_status(FES_NO_SEND, buf);
446                 break;
447         case MSG_ERROR_SOCKET:
448         case MSG_ERROR_UTF8:
449         case MSG_CLIENT:
450                 if (!am_generator)
451                         goto invalid_msg;
452                 if (tag == MSG_ERROR_SOCKET)
453                         io_end_multiplex_out();
454                 /* FALL THROUGH */
455         case MSG_INFO:
456         case MSG_ERROR:
457         case MSG_ERROR_XFER:
458         case MSG_WARNING:
459         case MSG_LOG:
460                 while (len) {
461                         n = len;
462                         if (n >= sizeof buf)
463                                 n = sizeof buf - 1;
464                         readfd(fd, buf, n);
465                         rwrite((enum logcode)tag, buf, n, !am_generator);
466                         len -= n;
467                 }
468                 break;
469         default:
470                 rprintf(FERROR, "unknown message %d:%d [%s]\n",
471                         tag, len, who_am_i());
472                 exit_cleanup(RERR_STREAMIO);
473         }
474
475         no_flush--;
476         msg_fd_in = fd;
477         if (!--defer_forwarding_messages && !no_flush)
478                 msg_flush();
479 }
480
481 /* This is used by the generator to limit how many file transfers can
482  * be active at once when --remove-source-files is specified.  Without
483  * this, sender-side deletions were mostly happening at the end. */
484 void increment_active_files(int ndx, int itemizing, enum logcode code)
485 {
486         /* TODO: tune these limits? */
487         while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) {
488                 check_for_finished_files(itemizing, code, 0);
489                 if (iobuf_out_cnt)
490                         io_flush(NORMAL_FLUSH);
491                 else
492                         read_msg_fd();
493         }
494
495         active_filecnt++;
496         active_bytecnt += F_LENGTH(cur_flist->files[ndx - cur_flist->ndx_start]);
497 }
498
499 /* Write an message to a multiplexed stream. If this fails, rsync exits. */
500 static void mplex_write(int fd, enum msgcode code, const char *buf, size_t len, int convert)
501 {
502         char buffer[BIGPATHBUFLEN]; /* Oversized for use by iconv code. */
503         size_t n = len;
504
505 #ifdef ICONV_OPTION
506         /* We need to convert buf before doing anything else so that we
507          * can include the (converted) byte length in the message header. */
508         if (convert && ic_send != (iconv_t)-1) {
509                 xbuf outbuf, inbuf;
510
511                 INIT_XBUF(outbuf, buffer + 4, 0, sizeof buffer - 4);
512                 INIT_XBUF(inbuf, (char*)buf, len, (size_t)-1);
513
514                 iconvbufs(ic_send, &inbuf, &outbuf,
515                           ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
516                 if (inbuf.len > 0) {
517                         rprintf(FERROR, "overflowed conversion buffer in mplex_write");
518                         exit_cleanup(RERR_UNSUPPORTED);
519                 }
520
521                 n = len = outbuf.len;
522         } else
523 #endif
524         if (n > 1024 - 4) /* BIGPATHBUFLEN can handle 1024 bytes */
525                 n = 0;    /* We'd rather do 2 writes than too much memcpy(). */
526         else
527                 memcpy(buffer + 4, buf, n);
528
529         SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len);
530
531         keep_defer_forwarding++; /* defer_forwarding_messages++ on return */
532         writefd_unbuffered(fd, buffer, n+4);
533         keep_defer_forwarding--;
534
535         if (len > n)
536                 writefd_unbuffered(fd, buf+n, len-n);
537
538         if (!--defer_forwarding_messages && !no_flush)
539                 msg_flush();
540 }
541
542 int send_msg(enum msgcode code, const char *buf, int len, int convert)
543 {
544         if (msg_fd_out < 0) {
545                 if (!defer_forwarding_messages)
546                         return io_multiplex_write(code, buf, len, convert);
547                 if (!io_multiplexing_out)
548                         return 0;
549                 msg_list_add(&msg_queue, code, buf, len, convert);
550                 return 1;
551         }
552         if (flist_forward_from >= 0)
553                 msg_list_add(&msg_queue, code, buf, len, convert);
554         else
555                 mplex_write(msg_fd_out, code, buf, len, convert);
556         return 1;
557 }
558
559 void send_msg_int(enum msgcode code, int num)
560 {
561         char numbuf[4];
562         SIVAL(numbuf, 0, num);
563         send_msg(code, numbuf, 4, 0);
564 }
565
566 void wait_for_receiver(void)
567 {
568         if (io_flush(FULL_FLUSH))
569                 return;
570         read_msg_fd();
571 }
572
573 int get_redo_num(void)
574 {
575         return flist_ndx_pop(&redo_list);
576 }
577
578 int get_hlink_num(void)
579 {
580         return flist_ndx_pop(&hlink_list);
581 }
582
583 /**
584  * When we're the receiver and we have a local --files-from list of names
585  * that needs to be sent over the socket to the sender, we have to do two
586  * things at the same time: send the sender a list of what files we're
587  * processing and read the incoming file+info list from the sender.  We do
588  * this by augmenting the read_timeout() function to copy this data.  It
589  * uses ff_buf to read a block of data from f_in (when it is ready, since
590  * it might be a pipe) and then blast it out f_out (when it is ready to
591  * receive more data).
592  */
593 void io_set_filesfrom_fds(int f_in, int f_out)
594 {
595         io_filesfrom_f_in = f_in;
596         io_filesfrom_f_out = f_out;
597         alloc_xbuf(&ff_buf, 2048);
598 #ifdef ICONV_OPTION
599         if (protect_args)
600                 alloc_xbuf(&iconv_buf, 1024);
601 #endif
602 }
603
604 /* It's almost always an error to get an EOF when we're trying to read from the
605  * network, because the protocol is (for the most part) self-terminating.
606  *
607  * There is one case for the receiver when it is at the end of the transfer
608  * (hanging around reading any keep-alive packets that might come its way): if
609  * the sender dies before the generator's kill-signal comes through, we can end
610  * up here needing to loop until the kill-signal arrives.  In this situation,
611  * kluge_around_eof will be < 0.
612  *
613  * There is another case for older protocol versions (< 24) where the module
614  * listing was not terminated, so we must ignore an EOF error in that case and
615  * exit.  In this situation, kluge_around_eof will be > 0. */
616 static void whine_about_eof(int fd)
617 {
618         if (kluge_around_eof && fd == sock_f_in) {
619                 int i;
620                 if (kluge_around_eof > 0)
621                         exit_cleanup(0);
622                 /* If we're still here after 10 seconds, exit with an error. */
623                 for (i = 10*1000/20; i--; )
624                         msleep(20);
625         }
626
627         rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed "
628                 "(%s bytes received so far) [%s]\n",
629                 big_num(stats.total_read, 0), who_am_i());
630
631         exit_cleanup(RERR_STREAMIO);
632 }
633
634 /**
635  * Read from a socket with I/O timeout. return the number of bytes
636  * read. If no bytes can be read then exit, never return a number <= 0.
637  *
638  * TODO: If the remote shell connection fails, then current versions
639  * actually report an "unexpected EOF" error here.  Since it's a
640  * fairly common mistake to try to use rsh when ssh is required, we
641  * should trap that: if we fail to read any data at all, we should
642  * give a better explanation.  We can tell whether the connection has
643  * started by looking e.g. at whether the remote version is known yet.
644  */
645 static int read_timeout(int fd, char *buf, size_t len)
646 {
647         int n, cnt = 0;
648
649         io_flush(FULL_FLUSH);
650
651         while (cnt == 0) {
652                 /* until we manage to read *something* */
653                 fd_set r_fds, w_fds;
654                 struct timeval tv;
655                 int maxfd = fd;
656                 int count;
657
658                 FD_ZERO(&r_fds);
659                 FD_ZERO(&w_fds);
660                 FD_SET(fd, &r_fds);
661                 if (io_filesfrom_f_out >= 0) {
662                         int new_fd;
663                         if (ff_buf.len == 0) {
664                                 if (io_filesfrom_f_in >= 0) {
665                                         FD_SET(io_filesfrom_f_in, &r_fds);
666                                         new_fd = io_filesfrom_f_in;
667                                 } else {
668                                         io_filesfrom_f_out = -1;
669                                         new_fd = -1;
670                                 }
671                         } else {
672                                 FD_SET(io_filesfrom_f_out, &w_fds);
673                                 new_fd = io_filesfrom_f_out;
674                         }
675                         if (new_fd > maxfd)
676                                 maxfd = new_fd;
677                 }
678
679                 tv.tv_sec = select_timeout;
680                 tv.tv_usec = 0;
681
682                 errno = 0;
683
684                 count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv);
685
686                 if (count <= 0) {
687                         if (errno == EBADF) {
688                                 defer_forwarding_messages = 0;
689                                 exit_cleanup(RERR_SOCKETIO);
690                         }
691                         check_timeout();
692                         continue;
693                 }
694
695                 if (io_filesfrom_f_out >= 0) {
696                         if (ff_buf.len) {
697                                 if (FD_ISSET(io_filesfrom_f_out, &w_fds)) {
698                                         int l = write(io_filesfrom_f_out,
699                                                       ff_buf.buf + ff_buf.pos,
700                                                       ff_buf.len);
701                                         if (l > 0) {
702                                                 if (!(ff_buf.len -= l))
703                                                         ff_buf.pos = 0;
704                                                 else
705                                                         ff_buf.pos += l;
706                                         } else if (errno != EINTR) {
707                                                 /* XXX should we complain? */
708                                                 io_filesfrom_f_out = -1;
709                                         }
710                                 }
711                         } else if (io_filesfrom_f_in >= 0) {
712                                 if (FD_ISSET(io_filesfrom_f_in, &r_fds)) {
713 #ifdef ICONV_OPTION
714                                         xbuf *ibuf = filesfrom_convert ? &iconv_buf : &ff_buf;
715 #else
716                                         xbuf *ibuf = &ff_buf;
717 #endif
718                                         int l = read(io_filesfrom_f_in, ibuf->buf, ibuf->size);
719                                         if (l <= 0) {
720                                                 if (l == 0 || errno != EINTR) {
721                                                         /* Send end-of-file marker */
722                                                         memcpy(ff_buf.buf, "\0\0", 2);
723                                                         ff_buf.len = ff_lastchar? 2 : 1;
724                                                         ff_buf.pos = 0;
725                                                         io_filesfrom_f_in = -1;
726                                                 }
727                                         } else {
728 #ifdef ICONV_OPTION
729                                                 if (filesfrom_convert) {
730                                                         iconv_buf.pos = 0;
731                                                         iconv_buf.len = l;
732                                                         iconvbufs(ic_send, &iconv_buf, &ff_buf,
733                                                             ICB_EXPAND_OUT|ICB_INCLUDE_BAD|ICB_INCLUDE_INCOMPLETE);
734                                                         l = ff_buf.len;
735                                                 }
736 #endif
737                                                 if (!eol_nulls) {
738                                                         char *s = ff_buf.buf + l;
739                                                         /* Transform CR and/or LF into '\0' */
740                                                         while (s-- > ff_buf.buf) {
741                                                                 if (*s == '\n' || *s == '\r')
742                                                                         *s = '\0';
743                                                         }
744                                                 }
745                                                 if (!ff_lastchar) {
746                                                         /* Last buf ended with a '\0', so don't
747                                                          * let this buf start with one. */
748                                                         while (l && ff_buf.buf[ff_buf.pos] == '\0')
749                                                                 ff_buf.pos++, l--;
750                                                 }
751                                                 if (!l)
752                                                         ff_buf.pos = 0;
753                                                 else {
754                                                         char *f = ff_buf.buf + ff_buf.pos;
755                                                         char *t = f;
756                                                         char *eob = f + l;
757                                                         /* Eliminate any multi-'\0' runs. */
758                                                         while (f != eob) {
759                                                                 if (!(*t++ = *f++)) {
760                                                                         while (f != eob && !*f)
761                                                                                 f++, l--;
762                                                                 }
763                                                         }
764                                                         ff_lastchar = f[-1];
765                                                 }
766                                                 ff_buf.len = l;
767                                         }
768                                 }
769                         }
770                 }
771
772                 if (!FD_ISSET(fd, &r_fds))
773                         continue;
774
775                 n = read(fd, buf, len);
776
777                 if (n <= 0) {
778                         if (n == 0)
779                                 whine_about_eof(fd); /* Doesn't return. */
780                         if (errno == EINTR || errno == EWOULDBLOCK
781                             || errno == EAGAIN)
782                                 continue;
783
784                         /* Don't write errors on a dead socket. */
785                         if (fd == sock_f_in) {
786                                 io_end_multiplex_out();
787                                 rsyserr(FERROR_SOCKET, errno, "read error");
788                         } else
789                                 rsyserr(FERROR, errno, "read error");
790                         exit_cleanup(RERR_STREAMIO);
791                 }
792
793                 buf += n;
794                 len -= n;
795                 cnt += n;
796
797                 if (fd == sock_f_in && io_timeout)
798                         last_io_in = time(NULL);
799         }
800
801         return cnt;
802 }
803
804 /* Read a line into the "buf" buffer. */
805 int read_line(int fd, char *buf, size_t bufsiz, int flags)
806 {
807         char ch, *s, *eob;
808         int cnt;
809
810 #ifdef ICONV_OPTION
811         if (flags & RL_CONVERT && iconv_buf.size < bufsiz)
812                 realloc_xbuf(&iconv_buf, bufsiz + 1024);
813 #endif
814
815   start:
816 #ifdef ICONV_OPTION
817         s = flags & RL_CONVERT ? iconv_buf.buf : buf;
818 #else
819         s = buf;
820 #endif
821         eob = s + bufsiz - 1;
822         while (1) {
823                 cnt = read(fd, &ch, 1);
824                 if (cnt < 0 && (errno == EWOULDBLOCK
825                   || errno == EINTR || errno == EAGAIN)) {
826                         struct timeval tv;
827                         fd_set r_fds, e_fds;
828                         FD_ZERO(&r_fds);
829                         FD_SET(fd, &r_fds);
830                         FD_ZERO(&e_fds);
831                         FD_SET(fd, &e_fds);
832                         tv.tv_sec = select_timeout;
833                         tv.tv_usec = 0;
834                         if (!select(fd+1, &r_fds, NULL, &e_fds, &tv))
835                                 check_timeout();
836                         /*if (FD_ISSET(fd, &e_fds))
837                                 rprintf(FINFO, "select exception on fd %d\n", fd); */
838                         continue;
839                 }
840                 if (cnt != 1)
841                         break;
842                 if (flags & RL_EOL_NULLS ? ch == '\0' : (ch == '\r' || ch == '\n')) {
843                         /* Skip empty lines if dumping comments. */
844                         if (flags & RL_DUMP_COMMENTS && s == buf)
845                                 continue;
846                         break;
847                 }
848                 if (s < eob)
849                         *s++ = ch;
850         }
851         *s = '\0';
852
853         if (flags & RL_DUMP_COMMENTS && (*buf == '#' || *buf == ';'))
854                 goto start;
855
856 #ifdef ICONV_OPTION
857         if (flags & RL_CONVERT) {
858                 xbuf outbuf;
859                 INIT_XBUF(outbuf, buf, 0, bufsiz);
860                 iconv_buf.pos = 0;
861                 iconv_buf.len = s - iconv_buf.buf;
862                 iconvbufs(ic_recv, &iconv_buf, &outbuf,
863                           ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE);
864                 outbuf.buf[outbuf.len] = '\0';
865                 return outbuf.len;
866         }
867 #endif
868
869         return s - buf;
870 }
871
872 void read_args(int f_in, char *mod_name, char *buf, size_t bufsiz, int rl_nulls,
873                char ***argv_p, int *argc_p, char **request_p)
874 {
875         int maxargs = MAX_ARGS;
876         int dot_pos = 0;
877         int argc = 0;
878         char **argv, *p;
879         int rl_flags = (rl_nulls ? RL_EOL_NULLS : 0);
880
881 #ifdef ICONV_OPTION
882         rl_flags |= (protect_args && ic_recv != (iconv_t)-1 ? RL_CONVERT : 0);
883 #endif
884
885         if (!(argv = new_array(char *, maxargs)))
886                 out_of_memory("read_args");
887         if (mod_name && !protect_args)
888                 argv[argc++] = "rsyncd";
889
890         while (1) {
891                 if (read_line(f_in, buf, bufsiz, rl_flags) == 0)
892                         break;
893
894                 if (argc == maxargs-1) {
895                         maxargs += MAX_ARGS;
896                         if (!(argv = realloc_array(argv, char *, maxargs)))
897                                 out_of_memory("read_args");
898                 }
899
900                 if (dot_pos) {
901                         if (request_p) {
902                                 *request_p = strdup(buf);
903                                 request_p = NULL;
904                         }
905                         if (mod_name)
906                                 glob_expand_module(mod_name, buf, &argv, &argc, &maxargs);
907                         else
908                                 glob_expand(buf, &argv, &argc, &maxargs);
909                 } else {
910                         if (!(p = strdup(buf)))
911                                 out_of_memory("read_args");
912                         argv[argc++] = p;
913                         if (*p == '.' && p[1] == '\0')
914                                 dot_pos = argc;
915                 }
916         }
917         argv[argc] = NULL;
918
919         glob_expand(NULL, NULL, NULL, NULL);
920
921         *argc_p = argc;
922         *argv_p = argv;
923 }
924
925 int io_start_buffering_out(int f_out)
926 {
927         if (iobuf_out) {
928                 assert(f_out == iobuf_f_out);
929                 return 0;
930         }
931         if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE)))
932                 out_of_memory("io_start_buffering_out");
933         iobuf_out_cnt = 0;
934         iobuf_f_out = f_out;
935         return 1;
936 }
937
938 int io_start_buffering_in(int f_in)
939 {
940         if (iobuf_in) {
941                 assert(f_in == iobuf_f_in);
942                 return 0;
943         }
944         iobuf_in_siz = 2 * IO_BUFFER_SIZE;
945         if (!(iobuf_in = new_array(char, iobuf_in_siz)))
946                 out_of_memory("io_start_buffering_in");
947         iobuf_f_in = f_in;
948         return 1;
949 }
950
951 void io_end_buffering_in(void)
952 {
953         if (!iobuf_in)
954                 return;
955         free(iobuf_in);
956         iobuf_in = NULL;
957         iobuf_in_ndx = 0;
958         iobuf_in_remaining = 0;
959         iobuf_f_in = -1;
960 }
961
962 void io_end_buffering_out(void)
963 {
964         if (!iobuf_out)
965                 return;
966         io_flush(FULL_FLUSH);
967         free(iobuf_out);
968         iobuf_out = NULL;
969         iobuf_f_out = -1;
970 }
971
972 void maybe_flush_socket(int important)
973 {
974         if (iobuf_out && iobuf_out_cnt
975          && (important || time(NULL) - last_io_out >= 5))
976                 io_flush(NORMAL_FLUSH);
977 }
978
979 void maybe_send_keepalive(void)
980 {
981         if (time(NULL) - last_io_out >= allowed_lull) {
982                 if (!iobuf_out || !iobuf_out_cnt) {
983                         if (protocol_version < 29)
984                                 return; /* there's nothing we can do */
985                         if (protocol_version >= 30)
986                                 send_msg(MSG_NOOP, "", 0, 0);
987                         else {
988                                 write_int(sock_f_out, cur_flist->used);
989                                 write_shortint(sock_f_out, ITEM_IS_NEW);
990                         }
991                 }
992                 if (iobuf_out)
993                         io_flush(NORMAL_FLUSH);
994         }
995 }
996
997 void start_flist_forward(int f_in)
998 {
999         assert(iobuf_out != NULL);
1000         assert(iobuf_f_out == msg_fd_out);
1001         flist_forward_from = f_in;
1002 }
1003
1004 void stop_flist_forward()
1005 {
1006         flist_forward_from = -1;
1007         io_flush(FULL_FLUSH);
1008 }
1009
1010 /**
1011  * Continue trying to read len bytes - don't return until len has been
1012  * read.
1013  **/
1014 static void read_loop(int fd, char *buf, size_t len)
1015 {
1016         while (len) {
1017                 int n = read_timeout(fd, buf, len);
1018
1019                 buf += n;
1020                 len -= n;
1021         }
1022 }
1023
1024 /**
1025  * Read from the file descriptor handling multiplexing - return number
1026  * of bytes read.
1027  *
1028  * Never returns <= 0.
1029  */
1030 static int readfd_unbuffered(int fd, char *buf, size_t len)
1031 {
1032         size_t msg_bytes;
1033         int tag, cnt = 0;
1034         char line[BIGPATHBUFLEN];
1035
1036         if (!iobuf_in || fd != iobuf_f_in)
1037                 return read_timeout(fd, buf, len);
1038
1039         if (!io_multiplexing_in && iobuf_in_remaining == 0) {
1040                 iobuf_in_remaining = read_timeout(fd, iobuf_in, iobuf_in_siz);
1041                 iobuf_in_ndx = 0;
1042         }
1043
1044         while (cnt == 0) {
1045                 if (iobuf_in_remaining) {
1046                         len = MIN(len, iobuf_in_remaining);
1047                         memcpy(buf, iobuf_in + iobuf_in_ndx, len);
1048                         iobuf_in_ndx += len;
1049                         iobuf_in_remaining -= len;
1050                         cnt = len;
1051                         break;
1052                 }
1053
1054                 read_loop(fd, line, 4);
1055                 tag = IVAL(line, 0);
1056
1057                 msg_bytes = tag & 0xFFFFFF;
1058                 tag = (tag >> 24) - MPLEX_BASE;
1059
1060                 switch (tag) {
1061                 case MSG_DATA:
1062                         if (msg_bytes > iobuf_in_siz) {
1063                                 if (!(iobuf_in = realloc_array(iobuf_in, char,
1064                                                                msg_bytes)))
1065                                         out_of_memory("readfd_unbuffered");
1066                                 iobuf_in_siz = msg_bytes;
1067                         }
1068                         read_loop(fd, iobuf_in, msg_bytes);
1069                         iobuf_in_remaining = msg_bytes;
1070                         iobuf_in_ndx = 0;
1071                         break;
1072                 case MSG_NOOP:
1073                         if (am_sender)
1074                                 maybe_send_keepalive();
1075                         break;
1076                 case MSG_IO_ERROR:
1077                         if (msg_bytes != 4)
1078                                 goto invalid_msg;
1079                         read_loop(fd, line, msg_bytes);
1080                         send_msg_int(MSG_IO_ERROR, IVAL(line, 0));
1081                         io_error |= IVAL(line, 0);
1082                         break;
1083                 case MSG_DELETED:
1084                         if (msg_bytes >= sizeof line)
1085                                 goto overflow;
1086 #ifdef ICONV_OPTION
1087                         if (ic_recv != (iconv_t)-1) {
1088                                 xbuf outbuf, inbuf;
1089                                 char ibuf[512];
1090                                 int add_null = 0;
1091
1092                                 INIT_CONST_XBUF(outbuf, line);
1093                                 INIT_XBUF(inbuf, ibuf, 0, (size_t)-1);
1094
1095                                 while (msg_bytes) {
1096                                         inbuf.len = msg_bytes > sizeof ibuf
1097                                                   ? sizeof ibuf : msg_bytes;
1098                                         read_loop(fd, inbuf.buf, inbuf.len);
1099                                         if (!(msg_bytes -= inbuf.len)
1100                                          && !ibuf[inbuf.len-1])
1101                                                 inbuf.len--, add_null = 1;
1102                                         if (iconvbufs(ic_send, &inbuf, &outbuf,
1103                                             ICB_INCLUDE_BAD | ICB_INCLUDE_INCOMPLETE) < 0)
1104                                                 goto overflow;
1105                                 }
1106                                 if (add_null) {
1107                                         if (outbuf.len == outbuf.size)
1108                                                 goto overflow;
1109                                         outbuf.buf[outbuf.len++] = '\0';
1110                                 }
1111                                 msg_bytes = outbuf.len;
1112                         } else
1113 #endif
1114                                 read_loop(fd, line, msg_bytes);
1115                         /* A directory name was sent with the trailing null */
1116                         if (msg_bytes > 0 && !line[msg_bytes-1])
1117                                 log_delete(line, S_IFDIR);
1118                         else {
1119                                 line[msg_bytes] = '\0';
1120                                 log_delete(line, S_IFREG);
1121                         }
1122                         break;
1123                 case MSG_SUCCESS:
1124                         if (msg_bytes != 4) {
1125                           invalid_msg:
1126                                 rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n",
1127                                         tag, (long)msg_bytes, who_am_i());
1128                                 exit_cleanup(RERR_STREAMIO);
1129                         }
1130                         read_loop(fd, line, msg_bytes);
1131                         successful_send(IVAL(line, 0));
1132                         break;
1133                 case MSG_NO_SEND:
1134                         if (msg_bytes != 4)
1135                                 goto invalid_msg;
1136                         read_loop(fd, line, msg_bytes);
1137                         send_msg_int(MSG_NO_SEND, IVAL(line, 0));
1138                         break;
1139                 case MSG_INFO:
1140                 case MSG_ERROR:
1141                 case MSG_ERROR_XFER:
1142                 case MSG_WARNING:
1143                         if (msg_bytes >= sizeof line) {
1144                             overflow:
1145                                 rprintf(FERROR,
1146                                         "multiplexing overflow %d:%ld [%s]\n",
1147                                         tag, (long)msg_bytes, who_am_i());
1148                                 exit_cleanup(RERR_STREAMIO);
1149                         }
1150                         read_loop(fd, line, msg_bytes);
1151                         rwrite((enum logcode)tag, line, msg_bytes, 1);
1152                         if (first_message) {
1153                                 if (list_only && !am_sender && tag == 1) {
1154                                         line[msg_bytes] = '\0';
1155                                         check_for_d_option_error(line);
1156                                 }
1157                                 first_message = 0;
1158                         }
1159                         break;
1160                 default:
1161                         rprintf(FERROR, "unexpected tag %d [%s]\n",
1162                                 tag, who_am_i());
1163                         exit_cleanup(RERR_STREAMIO);
1164                 }
1165         }
1166
1167         if (iobuf_in_remaining == 0)
1168                 io_flush(NORMAL_FLUSH);
1169
1170         return cnt;
1171 }
1172
1173 /* Do a buffered read from fd.  Don't return until all N bytes have
1174  * been read.  If all N can't be read then exit with an error. */
1175 static void readfd(int fd, char *buffer, size_t N)
1176 {
1177         int  cnt;
1178         size_t total = 0;
1179
1180         while (total < N) {
1181                 cnt = readfd_unbuffered(fd, buffer + total, N-total);
1182                 total += cnt;
1183         }
1184
1185         if (fd == write_batch_monitor_in) {
1186                 if ((size_t)write(batch_fd, buffer, total) != total)
1187                         exit_cleanup(RERR_FILEIO);
1188         }
1189
1190         if (fd == flist_forward_from)
1191                 writefd(iobuf_f_out, buffer, total);
1192
1193         if (fd == sock_f_in)
1194                 stats.total_read += total;
1195 }
1196
1197 unsigned short read_shortint(int f)
1198 {
1199         char b[2];
1200         readfd(f, b, 2);
1201         return (UVAL(b, 1) << 8) + UVAL(b, 0);
1202 }
1203
1204 int32 read_int(int f)
1205 {
1206         char b[4];
1207         int32 num;
1208
1209         readfd(f, b, 4);
1210         num = IVAL(b, 0);
1211 #if SIZEOF_INT32 > 4
1212         if (num & (int32)0x80000000)
1213                 num |= ~(int32)0xffffffff;
1214 #endif
1215         return num;
1216 }
1217
1218 int32 read_varint(int f)
1219 {
1220         union {
1221             char b[5];
1222             int32 x;
1223         } u;
1224         uchar ch;
1225         int extra;
1226
1227         u.x = 0;
1228         readfd(f, (char*)&ch, 1);
1229         extra = int_byte_extra[ch / 4];
1230         if (extra) {
1231                 uchar bit = ((uchar)1<<(8-extra));
1232                 if (extra >= (int)sizeof u.b) {
1233                         rprintf(FERROR, "Overflow in read_varint()\n");
1234                         exit_cleanup(RERR_STREAMIO);
1235                 }
1236                 readfd(f, u.b, extra);
1237                 u.b[extra] = ch & (bit-1);
1238         } else
1239                 u.b[0] = ch;
1240 #if CAREFUL_ALIGNMENT
1241         u.x = IVAL(u.b,0);
1242 #endif
1243 #if SIZEOF_INT32 > 4
1244         if (u.x & (int32)0x80000000)
1245                 u.x |= ~(int32)0xffffffff;
1246 #endif
1247         return u.x;
1248 }
1249
1250 int64 read_varlong(int f, uchar min_bytes)
1251 {
1252         union {
1253             char b[9];
1254             int64 x;
1255         } u;
1256         char b2[8];
1257         int extra;
1258
1259 #if SIZEOF_INT64 < 8
1260         memset(u.b, 0, 8);
1261 #else
1262         u.x = 0;
1263 #endif
1264         readfd(f, b2, min_bytes);
1265         memcpy(u.b, b2+1, min_bytes-1);
1266         extra = int_byte_extra[CVAL(b2, 0) / 4];
1267         if (extra) {
1268                 uchar bit = ((uchar)1<<(8-extra));
1269                 if (min_bytes + extra > (int)sizeof u.b) {
1270                         rprintf(FERROR, "Overflow in read_varlong()\n");
1271                         exit_cleanup(RERR_STREAMIO);
1272                 }
1273                 readfd(f, u.b + min_bytes - 1, extra);
1274                 u.b[min_bytes + extra - 1] = CVAL(b2, 0) & (bit-1);
1275 #if SIZEOF_INT64 < 8
1276                 if (min_bytes + extra > 5 || u.b[4] || CVAL(u.b,3) & 0x80) {
1277                         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1278                         exit_cleanup(RERR_UNSUPPORTED);
1279                 }
1280 #endif
1281         } else
1282                 u.b[min_bytes + extra - 1] = CVAL(b2, 0);
1283 #if SIZEOF_INT64 < 8
1284         u.x = IVAL(u.b,0);
1285 #elif CAREFUL_ALIGNMENT
1286         u.x = IVAL(u.b,0) | (((int64)IVAL(u.b,4))<<32);
1287 #endif
1288         return u.x;
1289 }
1290
1291 int64 read_longint(int f)
1292 {
1293 #if SIZEOF_INT64 >= 8
1294         char b[9];
1295 #endif
1296         int32 num = read_int(f);
1297
1298         if (num != (int32)0xffffffff)
1299                 return num;
1300
1301 #if SIZEOF_INT64 < 8
1302         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1303         exit_cleanup(RERR_UNSUPPORTED);
1304 #else
1305         readfd(f, b, 8);
1306         return IVAL(b,0) | (((int64)IVAL(b,4))<<32);
1307 #endif
1308 }
1309
1310 void read_buf(int f, char *buf, size_t len)
1311 {
1312         readfd(f,buf,len);
1313 }
1314
1315 void read_sbuf(int f, char *buf, size_t len)
1316 {
1317         readfd(f, buf, len);
1318         buf[len] = '\0';
1319 }
1320
1321 uchar read_byte(int f)
1322 {
1323         uchar c;
1324         readfd(f, (char *)&c, 1);
1325         return c;
1326 }
1327
1328 int read_vstring(int f, char *buf, int bufsize)
1329 {
1330         int len = read_byte(f);
1331
1332         if (len & 0x80)
1333                 len = (len & ~0x80) * 0x100 + read_byte(f);
1334
1335         if (len >= bufsize) {
1336                 rprintf(FERROR, "over-long vstring received (%d > %d)\n",
1337                         len, bufsize - 1);
1338                 return -1;
1339         }
1340
1341         if (len)
1342                 readfd(f, buf, len);
1343         buf[len] = '\0';
1344         return len;
1345 }
1346
1347 /* Populate a sum_struct with values from the socket.  This is
1348  * called by both the sender and the receiver. */
1349 void read_sum_head(int f, struct sum_struct *sum)
1350 {
1351         int32 max_blength = protocol_version < 30 ? OLD_MAX_BLOCK_SIZE : MAX_BLOCK_SIZE;
1352         sum->count = read_int(f);
1353         if (sum->count < 0) {
1354                 rprintf(FERROR, "Invalid checksum count %ld [%s]\n",
1355                         (long)sum->count, who_am_i());
1356                 exit_cleanup(RERR_PROTOCOL);
1357         }
1358         sum->blength = read_int(f);
1359         if (sum->blength < 0 || sum->blength > max_blength) {
1360                 rprintf(FERROR, "Invalid block length %ld [%s]\n",
1361                         (long)sum->blength, who_am_i());
1362                 exit_cleanup(RERR_PROTOCOL);
1363         }
1364         sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f);
1365         if (sum->s2length < 0 || sum->s2length > MAX_DIGEST_LEN) {
1366                 rprintf(FERROR, "Invalid checksum length %d [%s]\n",
1367                         sum->s2length, who_am_i());
1368                 exit_cleanup(RERR_PROTOCOL);
1369         }
1370         sum->remainder = read_int(f);
1371         if (sum->remainder < 0 || sum->remainder > sum->blength) {
1372                 rprintf(FERROR, "Invalid remainder length %ld [%s]\n",
1373                         (long)sum->remainder, who_am_i());
1374                 exit_cleanup(RERR_PROTOCOL);
1375         }
1376 }
1377
1378 /* Send the values from a sum_struct over the socket.  Set sum to
1379  * NULL if there are no checksums to send.  This is called by both
1380  * the generator and the sender. */
1381 void write_sum_head(int f, struct sum_struct *sum)
1382 {
1383         static struct sum_struct null_sum;
1384
1385         if (sum == NULL)
1386                 sum = &null_sum;
1387
1388         write_int(f, sum->count);
1389         write_int(f, sum->blength);
1390         if (protocol_version >= 27)
1391                 write_int(f, sum->s2length);
1392         write_int(f, sum->remainder);
1393 }
1394
1395 /**
1396  * Sleep after writing to limit I/O bandwidth usage.
1397  *
1398  * @todo Rather than sleeping after each write, it might be better to
1399  * use some kind of averaging.  The current algorithm seems to always
1400  * use a bit less bandwidth than specified, because it doesn't make up
1401  * for slow periods.  But arguably this is a feature.  In addition, we
1402  * ought to take the time used to write the data into account.
1403  *
1404  * During some phases of big transfers (file FOO is uptodate) this is
1405  * called with a small bytes_written every time.  As the kernel has to
1406  * round small waits up to guarantee that we actually wait at least the
1407  * requested number of microseconds, this can become grossly inaccurate.
1408  * We therefore keep track of the bytes we've written over time and only
1409  * sleep when the accumulated delay is at least 1 tenth of a second.
1410  **/
1411 static void sleep_for_bwlimit(int bytes_written)
1412 {
1413         static struct timeval prior_tv;
1414         static long total_written = 0;
1415         struct timeval tv, start_tv;
1416         long elapsed_usec, sleep_usec;
1417
1418 #define ONE_SEC 1000000L /* # of microseconds in a second */
1419
1420         if (!bwlimit_writemax)
1421                 return;
1422
1423         total_written += bytes_written;
1424
1425         gettimeofday(&start_tv, NULL);
1426         if (prior_tv.tv_sec) {
1427                 elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC
1428                              + (start_tv.tv_usec - prior_tv.tv_usec);
1429                 total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024);
1430                 if (total_written < 0)
1431                         total_written = 0;
1432         }
1433
1434         sleep_usec = total_written * (ONE_SEC/1024) / bwlimit;
1435         if (sleep_usec < ONE_SEC / 10) {
1436                 prior_tv = start_tv;
1437                 return;
1438         }
1439
1440         tv.tv_sec  = sleep_usec / ONE_SEC;
1441         tv.tv_usec = sleep_usec % ONE_SEC;
1442         select(0, NULL, NULL, NULL, &tv);
1443
1444         gettimeofday(&prior_tv, NULL);
1445         elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC
1446                      + (prior_tv.tv_usec - start_tv.tv_usec);
1447         total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024);
1448 }
1449
1450 /* Write len bytes to the file descriptor fd, looping as necessary to get
1451  * the job done and also (in certain circumstances) reading any data on
1452  * msg_fd_in to avoid deadlock.
1453  *
1454  * This function underlies the multiplexing system.  The body of the
1455  * application never calls this function directly. */
1456 static void writefd_unbuffered(int fd, const char *buf, size_t len)
1457 {
1458         size_t n, total = 0;
1459         fd_set w_fds, r_fds, e_fds;
1460         int maxfd, count, cnt, using_r_fds;
1461         int defer_inc = 0;
1462         struct timeval tv;
1463
1464         if (no_flush++)
1465                 defer_forwarding_messages++, defer_inc++;
1466
1467         while (total < len) {
1468                 FD_ZERO(&w_fds);
1469                 FD_SET(fd, &w_fds);
1470                 FD_ZERO(&e_fds);
1471                 FD_SET(fd, &e_fds);
1472                 maxfd = fd;
1473
1474                 if (msg_fd_in >= 0) {
1475                         FD_ZERO(&r_fds);
1476                         FD_SET(msg_fd_in, &r_fds);
1477                         if (msg_fd_in > maxfd)
1478                                 maxfd = msg_fd_in;
1479                         using_r_fds = 1;
1480                 } else
1481                         using_r_fds = 0;
1482
1483                 tv.tv_sec = select_timeout;
1484                 tv.tv_usec = 0;
1485
1486                 errno = 0;
1487                 count = select(maxfd + 1, using_r_fds ? &r_fds : NULL,
1488                                &w_fds, &e_fds, &tv);
1489
1490                 if (count <= 0) {
1491                         if (count < 0 && errno == EBADF)
1492                                 exit_cleanup(RERR_SOCKETIO);
1493                         check_timeout();
1494                         continue;
1495                 }
1496
1497                 /*if (FD_ISSET(fd, &e_fds))
1498                         rprintf(FINFO, "select exception on fd %d\n", fd); */
1499
1500                 if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds))
1501                         read_msg_fd();
1502
1503                 if (!FD_ISSET(fd, &w_fds))
1504                         continue;
1505
1506                 n = len - total;
1507                 if (bwlimit_writemax && n > bwlimit_writemax)
1508                         n = bwlimit_writemax;
1509                 cnt = write(fd, buf + total, n);
1510
1511                 if (cnt <= 0) {
1512                         if (cnt < 0) {
1513                                 if (errno == EINTR)
1514                                         continue;
1515                                 if (errno == EWOULDBLOCK || errno == EAGAIN) {
1516                                         msleep(1);
1517                                         continue;
1518                                 }
1519                         }
1520
1521                         /* Don't try to write errors back across the stream. */
1522                         if (fd == sock_f_out)
1523                                 io_end_multiplex_out();
1524                         /* Don't try to write errors down a failing msg pipe. */
1525                         if (am_server && fd == msg_fd_out)
1526                                 exit_cleanup(RERR_STREAMIO);
1527                         rsyserr(FERROR, errno,
1528                                 "writefd_unbuffered failed to write %ld bytes [%s]",
1529                                 (long)len, who_am_i());
1530                         /* If the other side is sending us error messages, try
1531                          * to grab any messages they sent before they died. */
1532                         while (!am_server && fd == sock_f_out && io_multiplexing_in) {
1533                                 char buf[1024];
1534                                 set_io_timeout(30);
1535                                 ignore_timeout = 0;
1536                                 readfd_unbuffered(sock_f_in, buf, sizeof buf);
1537                         }
1538                         exit_cleanup(RERR_STREAMIO);
1539                 }
1540
1541                 total += cnt;
1542                 defer_forwarding_messages++, defer_inc++;
1543
1544                 if (fd == sock_f_out) {
1545                         if (io_timeout || am_generator)
1546                                 last_io_out = time(NULL);
1547                         sleep_for_bwlimit(cnt);
1548                 }
1549         }
1550
1551         no_flush--;
1552         if (keep_defer_forwarding)
1553                 defer_inc--;
1554         if (!(defer_forwarding_messages -= defer_inc) && !no_flush)
1555                 msg_flush();
1556 }
1557
1558 int io_flush(int flush_it_all)
1559 {
1560         int flushed_something = 0;
1561
1562         if (no_flush)
1563                 return 0;
1564
1565         if (iobuf_out_cnt) {
1566                 if (io_multiplexing_out)
1567                         mplex_write(sock_f_out, MSG_DATA, iobuf_out, iobuf_out_cnt, 0);
1568                 else
1569                         writefd_unbuffered(iobuf_f_out, iobuf_out, iobuf_out_cnt);
1570                 iobuf_out_cnt = 0;
1571                 flushed_something = 1;
1572         }
1573
1574         if (flush_it_all && !defer_forwarding_messages && msg_queue.head) {
1575                 msg_flush();
1576                 flushed_something = 1;
1577         }
1578
1579         return flushed_something;
1580 }
1581
1582 static void writefd(int fd, const char *buf, size_t len)
1583 {
1584         if (fd == sock_f_out)
1585                 stats.total_written += len;
1586
1587         if (fd == write_batch_monitor_out) {
1588                 if ((size_t)write(batch_fd, buf, len) != len)
1589                         exit_cleanup(RERR_FILEIO);
1590         }
1591
1592         if (!iobuf_out || fd != iobuf_f_out) {
1593                 writefd_unbuffered(fd, buf, len);
1594                 return;
1595         }
1596
1597         while (len) {
1598                 int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt);
1599                 if (n > 0) {
1600                         memcpy(iobuf_out+iobuf_out_cnt, buf, n);
1601                         buf += n;
1602                         len -= n;
1603                         iobuf_out_cnt += n;
1604                 }
1605
1606                 if (iobuf_out_cnt == IO_BUFFER_SIZE)
1607                         io_flush(NORMAL_FLUSH);
1608         }
1609 }
1610
1611 void write_shortint(int f, unsigned short x)
1612 {
1613         char b[2];
1614         b[0] = (char)x;
1615         b[1] = (char)(x >> 8);
1616         writefd(f, b, 2);
1617 }
1618
1619 void write_int(int f, int32 x)
1620 {
1621         char b[4];
1622         SIVAL(b, 0, x);
1623         writefd(f, b, 4);
1624 }
1625
1626 void write_varint(int f, int32 x)
1627 {
1628         char b[5];
1629         uchar bit;
1630         int cnt = 4;
1631
1632         SIVAL(b, 1, x);
1633
1634         while (cnt > 1 && b[cnt] == 0)
1635                 cnt--;
1636         bit = ((uchar)1<<(7-cnt+1));
1637         if (CVAL(b, cnt) >= bit) {
1638                 cnt++;
1639                 *b = ~(bit-1);
1640         } else if (cnt > 1)
1641                 *b = b[cnt] | ~(bit*2-1);
1642         else
1643                 *b = b[cnt];
1644
1645         writefd(f, b, cnt);
1646 }
1647
1648 void write_varlong(int f, int64 x, uchar min_bytes)
1649 {
1650         char b[9];
1651         uchar bit;
1652         int cnt = 8;
1653
1654         SIVAL(b, 1, x);
1655 #if SIZEOF_INT64 >= 8
1656         SIVAL(b, 5, x >> 32);
1657 #else
1658         if (x <= 0x7FFFFFFF && x >= 0)
1659                 memset(b + 5, 0, 4);
1660         else {
1661                 rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1662                 exit_cleanup(RERR_UNSUPPORTED);
1663         }
1664 #endif
1665
1666         while (cnt > min_bytes && b[cnt] == 0)
1667                 cnt--;
1668         bit = ((uchar)1<<(7-cnt+min_bytes));
1669         if (CVAL(b, cnt) >= bit) {
1670                 cnt++;
1671                 *b = ~(bit-1);
1672         } else if (cnt > min_bytes)
1673                 *b = b[cnt] | ~(bit*2-1);
1674         else
1675                 *b = b[cnt];
1676
1677         writefd(f, b, cnt);
1678 }
1679
1680 /*
1681  * Note: int64 may actually be a 32-bit type if ./configure couldn't find any
1682  * 64-bit types on this platform.
1683  */
1684 void write_longint(int f, int64 x)
1685 {
1686         char b[12], * const s = b+4;
1687
1688         SIVAL(s, 0, x);
1689         if (x <= 0x7FFFFFFF && x >= 0) {
1690                 writefd(f, s, 4);
1691                 return;
1692         }
1693
1694 #if SIZEOF_INT64 < 8
1695         rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n");
1696         exit_cleanup(RERR_UNSUPPORTED);
1697 #else
1698         memset(b, 0xFF, 4);
1699         SIVAL(s, 4, x >> 32);
1700         writefd(f, b, 12);
1701 #endif
1702 }
1703
1704 void write_buf(int f, const char *buf, size_t len)
1705 {
1706         writefd(f,buf,len);
1707 }
1708
1709 /** Write a string to the connection */
1710 void write_sbuf(int f, const char *buf)
1711 {
1712         writefd(f, buf, strlen(buf));
1713 }
1714
1715 void write_byte(int f, uchar c)
1716 {
1717         writefd(f, (char *)&c, 1);
1718 }
1719
1720 void write_vstring(int f, const char *str, int len)
1721 {
1722         uchar lenbuf[3], *lb = lenbuf;
1723
1724         if (len > 0x7F) {
1725                 if (len > 0x7FFF) {
1726                         rprintf(FERROR,
1727                                 "attempting to send over-long vstring (%d > %d)\n",
1728                                 len, 0x7FFF);
1729                         exit_cleanup(RERR_PROTOCOL);
1730                 }
1731                 *lb++ = len / 0x100 + 0x80;
1732         }
1733         *lb = len;
1734
1735         writefd(f, (char*)lenbuf, lb - lenbuf + 1);
1736         if (len)
1737                 writefd(f, str, len);
1738 }
1739
1740 /* Send a file-list index using a byte-reduction method. */
1741 void write_ndx(int f, int32 ndx)
1742 {
1743         static int32 prev_positive = -1, prev_negative = 1;
1744         int32 diff, cnt = 0;
1745         char b[6];
1746
1747         if (protocol_version < 30 || read_batch) {
1748                 write_int(f, ndx);
1749                 return;
1750         }
1751
1752         /* Send NDX_DONE as a single-byte 0 with no side effects.  Send
1753          * negative nums as a positive after sending a leading 0xFF. */
1754         if (ndx >= 0) {
1755                 diff = ndx - prev_positive;
1756                 prev_positive = ndx;
1757         } else if (ndx == NDX_DONE) {
1758                 *b = 0;
1759                 writefd(f, b, 1);
1760                 return;
1761         } else {
1762                 b[cnt++] = (char)0xFF;
1763                 ndx = -ndx;
1764                 diff = ndx - prev_negative;
1765                 prev_negative = ndx;
1766         }
1767
1768         /* A diff of 1 - 253 is sent as a one-byte diff; a diff of 254 - 32767
1769          * or 0 is sent as a 0xFE + a two-byte diff; otherwise we send 0xFE
1770          * & all 4 bytes of the (non-negative) num with the high-bit set. */
1771         if (diff < 0xFE && diff > 0)
1772                 b[cnt++] = (char)diff;
1773         else if (diff < 0 || diff > 0x7FFF) {
1774                 b[cnt++] = (char)0xFE;
1775                 b[cnt++] = (char)((ndx >> 24) | 0x80);
1776                 b[cnt++] = (char)ndx;
1777                 b[cnt++] = (char)(ndx >> 8);
1778                 b[cnt++] = (char)(ndx >> 16);
1779         } else {
1780                 b[cnt++] = (char)0xFE;
1781                 b[cnt++] = (char)(diff >> 8);
1782                 b[cnt++] = (char)diff;
1783         }
1784         writefd(f, b, cnt);
1785 }
1786
1787 /* Receive a file-list index using a byte-reduction method. */
1788 int32 read_ndx(int f)
1789 {
1790         static int32 prev_positive = -1, prev_negative = 1;
1791         int32 *prev_ptr, num;
1792         char b[4];
1793
1794         if (protocol_version < 30)
1795                 return read_int(f);
1796
1797         readfd(f, b, 1);
1798         if (CVAL(b, 0) == 0xFF) {
1799                 readfd(f, b, 1);
1800                 prev_ptr = &prev_negative;
1801         } else if (CVAL(b, 0) == 0)
1802                 return NDX_DONE;
1803         else
1804                 prev_ptr = &prev_positive;
1805         if (CVAL(b, 0) == 0xFE) {
1806                 readfd(f, b, 2);
1807                 if (CVAL(b, 0) & 0x80) {
1808                         b[3] = CVAL(b, 0) & ~0x80;
1809                         b[0] = b[1];
1810                         readfd(f, b+1, 2);
1811                         num = IVAL(b, 0);
1812                 } else
1813                         num = (UVAL(b,0)<<8) + UVAL(b,1) + *prev_ptr;
1814         } else
1815                 num = UVAL(b, 0) + *prev_ptr;
1816         *prev_ptr = num;
1817         if (prev_ptr == &prev_negative)
1818                 num = -num;
1819         return num;
1820 }
1821
1822 /* Read a line of up to bufsiz-1 characters into buf.  Strips
1823  * the (required) trailing newline and all carriage returns.
1824  * Returns 1 for success; 0 for I/O error or truncation. */
1825 int read_line_old(int f, char *buf, size_t bufsiz)
1826 {
1827         bufsiz--; /* leave room for the null */
1828         while (bufsiz > 0) {
1829                 buf[0] = 0;
1830                 read_buf(f, buf, 1);
1831                 if (buf[0] == 0)
1832                         return 0;
1833                 if (buf[0] == '\n')
1834                         break;
1835                 if (buf[0] != '\r') {
1836                         buf++;
1837                         bufsiz--;
1838                 }
1839         }
1840         *buf = '\0';
1841         return bufsiz > 0;
1842 }
1843
1844 void io_printf(int fd, const char *format, ...)
1845 {
1846         va_list ap;
1847         char buf[BIGPATHBUFLEN];
1848         int len;
1849
1850         va_start(ap, format);
1851         len = vsnprintf(buf, sizeof buf, format, ap);
1852         va_end(ap);
1853
1854         if (len < 0)
1855                 exit_cleanup(RERR_STREAMIO);
1856
1857         if (len > (int)sizeof buf) {
1858                 rprintf(FERROR, "io_printf() was too long for the buffer.\n");
1859                 exit_cleanup(RERR_STREAMIO);
1860         }
1861
1862         write_sbuf(fd, buf);
1863 }
1864
1865 /** Setup for multiplexing a MSG_* stream with the data stream. */
1866 void io_start_multiplex_out(void)
1867 {
1868         io_flush(NORMAL_FLUSH);
1869         io_start_buffering_out(sock_f_out);
1870         io_multiplexing_out = 1;
1871 }
1872
1873 /** Setup for multiplexing a MSG_* stream with the data stream. */
1874 void io_start_multiplex_in(void)
1875 {
1876         io_flush(NORMAL_FLUSH);
1877         io_start_buffering_in(sock_f_in);
1878         io_multiplexing_in = 1;
1879 }
1880
1881 /** Write an message to the multiplexed data stream. */
1882 int io_multiplex_write(enum msgcode code, const char *buf, size_t len, int convert)
1883 {
1884         if (!io_multiplexing_out)
1885                 return 0;
1886         io_flush(NORMAL_FLUSH);
1887         stats.total_written += (len+4);
1888         mplex_write(sock_f_out, code, buf, len, convert);
1889         return 1;
1890 }
1891
1892 void io_end_multiplex_in(void)
1893 {
1894         io_multiplexing_in = 0;
1895         io_end_buffering_in();
1896 }
1897
1898 /** Stop output multiplexing. */
1899 void io_end_multiplex_out(void)
1900 {
1901         io_multiplexing_out = 0;
1902         io_end_buffering_out();
1903 }
1904
1905 void start_write_batch(int fd)
1906 {
1907         /* Some communication has already taken place, but we don't
1908          * enable batch writing until here so that we can write a
1909          * canonical record of the communication even though the
1910          * actual communication so far depends on whether a daemon
1911          * is involved. */
1912         write_int(batch_fd, protocol_version);
1913         if (protocol_version >= 30)
1914                 write_byte(batch_fd, inc_recurse);
1915         write_int(batch_fd, checksum_seed);
1916
1917         if (am_sender)
1918                 write_batch_monitor_out = fd;
1919         else
1920                 write_batch_monitor_in = fd;
1921 }
1922
1923 void stop_write_batch(void)
1924 {
1925         write_batch_monitor_out = -1;
1926         write_batch_monitor_in = -1;
1927 }