1 /*
2 * Copyright (C) 1984-2012 Mark Nudelman
3 * Modified for use with illumos by Garrett D'Amore.
4 * Copyright 2014 Garrett D'Amore <garrett@damore.org>
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Less License, as specified in the README file.
8 *
9 * For more information, see the README file.
10 */
11
12 /*
13 * Low level character input from the input file.
14 * We use these special purpose routines which optimize moving
15 * both forward and backward from the current read pointer.
16 */
17
18 #include <sys/stat.h>
19
20 #include "less.h"
21
22 extern dev_t curr_dev;
23 extern ino_t curr_ino;
24 extern int less_is_more;
25
26 typedef off_t BLOCKNUM;
27
28 int ignore_eoi;
29
30 /*
31 * Pool of buffers holding the most recently used blocks of the input file.
32 * The buffer pool is kept as a doubly-linked circular list,
33 * in order from most- to least-recently used.
34 * The circular list is anchored by the file state "thisfile".
35 */
36 struct bufnode {
37 struct bufnode *next, *prev;
38 struct bufnode *hnext, *hprev;
39 };
40
41 #define LBUFSIZE 8192
42 struct buf {
43 struct bufnode node;
44 BLOCKNUM block;
45 unsigned int datasize;
46 unsigned char data[LBUFSIZE];
47 };
48 #define bufnode_buf(bn) ((struct buf *)bn)
49
50 /*
51 * The file state is maintained in a filestate structure.
52 * A pointer to the filestate is kept in the ifile structure.
53 */
54 #define BUFHASH_SIZE 64
55 struct filestate {
56 struct bufnode buflist;
57 struct bufnode hashtbl[BUFHASH_SIZE];
58 int file;
59 int flags;
60 off_t fpos;
61 int nbufs;
62 BLOCKNUM block;
63 unsigned int offset;
64 off_t fsize;
65 };
66
67 #define ch_bufhead thisfile->buflist.next
68 #define ch_buftail thisfile->buflist.prev
69 #define ch_nbufs thisfile->nbufs
70 #define ch_block thisfile->block
71 #define ch_offset thisfile->offset
72 #define ch_fpos thisfile->fpos
73 #define ch_fsize thisfile->fsize
74 #define ch_flags thisfile->flags
75 #define ch_file thisfile->file
76
77 #define END_OF_CHAIN (&thisfile->buflist)
78 #define END_OF_HCHAIN(h) (&thisfile->hashtbl[h])
79 #define BUFHASH(blk) ((blk) & (BUFHASH_SIZE-1))
80
81 /*
82 * Macros to manipulate the list of buffers in thisfile->buflist.
83 */
84 #define FOR_BUFS(bn) \
85 for ((bn) = ch_bufhead; (bn) != END_OF_CHAIN; (bn) = (bn)->next)
86
87 #define BUF_RM(bn) \
88 (bn)->next->prev = (bn)->prev; \
89 (bn)->prev->next = (bn)->next;
90
91 #define BUF_INS_HEAD(bn) \
92 (bn)->next = ch_bufhead; \
93 (bn)->prev = END_OF_CHAIN; \
94 ch_bufhead->prev = (bn); \
95 ch_bufhead = (bn);
96
97 #define BUF_INS_TAIL(bn) \
98 (bn)->next = END_OF_CHAIN; \
99 (bn)->prev = ch_buftail; \
100 ch_buftail->next = (bn); \
101 ch_buftail = (bn);
102
103 /*
104 * Macros to manipulate the list of buffers in thisfile->hashtbl[n].
105 */
106 #define FOR_BUFS_IN_CHAIN(h, bn) \
107 for ((bn) = thisfile->hashtbl[h].hnext; \
108 (bn) != END_OF_HCHAIN(h); (bn) = (bn)->hnext)
109
110 #define BUF_HASH_RM(bn) \
111 (bn)->hnext->hprev = (bn)->hprev; \
112 (bn)->hprev->hnext = (bn)->hnext;
113
114 #define BUF_HASH_INS(bn, h) \
115 (bn)->hnext = thisfile->hashtbl[h].hnext; \
116 (bn)->hprev = END_OF_HCHAIN(h); \
117 thisfile->hashtbl[h].hnext->hprev = (bn); \
118 thisfile->hashtbl[h].hnext = (bn);
119
120 static struct filestate *thisfile;
121 static int ch_ungotchar = -1;
122 static int maxbufs = -1;
123
124 extern int autobuf;
125 extern int secure;
126 extern int screen_trashed;
127 extern int follow_mode;
128 extern IFILE curr_ifile;
129 extern int logfile;
130 extern char *namelogfile;
131
132 static int ch_addbuf(void);
133
134
135 /*
136 * Get the character pointed to by the read pointer.
137 */
138 int
ch_get(void)139 ch_get(void)
140 {
141 struct buf *bp;
142 struct bufnode *bn;
143 int n;
144 int slept;
145 int h;
146 off_t pos;
147 off_t len;
148
149 if (thisfile == NULL)
150 return (EOI);
151
152 /*
153 * Quick check for the common case where
154 * the desired char is in the head buffer.
155 */
156 if (ch_bufhead != END_OF_CHAIN) {
157 bp = bufnode_buf(ch_bufhead);
158 if (ch_block == bp->block && ch_offset < bp->datasize)
159 return (bp->data[ch_offset]);
160 }
161
162 slept = FALSE;
163
164 /*
165 * Look for a buffer holding the desired block.
166 */
167 h = BUFHASH(ch_block);
168 FOR_BUFS_IN_CHAIN(h, bn) {
169 bp = bufnode_buf(bn);
170 if (bp->block == ch_block) {
171 if (ch_offset >= bp->datasize)
172 /*
173 * Need more data in this buffer.
174 */
175 break;
176 goto found;
177 }
178 }
179 if (bn == END_OF_HCHAIN(h)) {
180 /*
181 * Block is not in a buffer.
182 * Take the least recently used buffer
183 * and read the desired block into it.
184 * If the LRU buffer has data in it,
185 * then maybe allocate a new buffer.
186 */
187 if (ch_buftail == END_OF_CHAIN ||
188 bufnode_buf(ch_buftail)->block != -1) {
189 /*
190 * There is no empty buffer to use.
191 * Allocate a new buffer if:
192 * 1. We can't seek on this file and -b is not in
193 * effect; or
194 * 2. We haven't allocated the max buffers for this
195 * file yet.
196 */
197 if ((autobuf && !(ch_flags & CH_CANSEEK)) ||
198 (maxbufs < 0 || ch_nbufs < maxbufs))
199 if (ch_addbuf())
200 /*
201 * Allocation failed: turn off autobuf.
202 */
203 autobuf = OPT_OFF;
204 }
205 bn = ch_buftail;
206 bp = bufnode_buf(bn);
207 BUF_HASH_RM(bn); /* Remove from old hash chain. */
208 bp->block = ch_block;
209 bp->datasize = 0;
210 BUF_HASH_INS(bn, h); /* Insert into new hash chain. */
211 }
212
213 read_more:
214 pos = (ch_block * LBUFSIZE) + bp->datasize;
215 if ((len = ch_length()) != -1 && pos >= len)
216 /*
217 * At end of file.
218 */
219 return (EOI);
220
221 if (pos != ch_fpos) {
222 /*
223 * Not at the correct position: must seek.
224 * If input is a pipe, we're in trouble (can't seek on a pipe).
225 * Some data has been lost: just return "?".
226 */
227 if (!(ch_flags & CH_CANSEEK))
228 return ('?');
229 if (lseek(ch_file, (off_t)pos, SEEK_SET) == (off_t)-1) {
230 error("seek error", NULL);
231 clear_eol();
232 return (EOI);
233 }
234 ch_fpos = pos;
235 }
236
237 /*
238 * Read the block.
239 * If we read less than a full block, that's ok.
240 * We use partial block and pick up the rest next time.
241 */
242 if (ch_ungotchar != -1) {
243 bp->data[bp->datasize] = (unsigned char)ch_ungotchar;
244 n = 1;
245 ch_ungotchar = -1;
246 } else {
247 n = iread(ch_file, &bp->data[bp->datasize],
248 (unsigned int)(LBUFSIZE - bp->datasize));
249 }
250
251 if (n == READ_INTR)
252 return (EOI);
253 if (n < 0) {
254 error("read error", NULL);
255 clear_eol();
256 n = 0;
257 }
258
259 /*
260 * If we have a log file, write the new data to it.
261 */
262 if (!secure && logfile >= 0 && n > 0)
263 (void) write(logfile, (char *)&bp->data[bp->datasize], n);
264
265 ch_fpos += n;
266 bp->datasize += n;
267
268 /*
269 * If we have read to end of file, set ch_fsize to indicate
270 * the position of the end of file.
271 */
272 if (n == 0) {
273 ch_fsize = pos;
274 if (ignore_eoi) {
275 /*
276 * We are ignoring EOF.
277 * Wait a while, then try again.
278 */
279 if (!slept) {
280 PARG parg;
281 parg.p_string = wait_message();
282 ierror("%s", &parg);
283 }
284 sleep(1);
285 slept = TRUE;
286
287 if (follow_mode == FOLLOW_NAME) {
288 /*
289 * See whether the file's i-number has changed.
290 * If so, force the file to be closed and
291 * reopened.
292 */
293 struct stat st;
294 int r = stat(get_filename(curr_ifile), &st);
295 if (r == 0 && (st.st_ino != curr_ino ||
296 st.st_dev != curr_dev)) {
297 /*
298 * screen_trashed=2 causes
299 * make_display to reopen the file.
300 */
301 screen_trashed = 2;
302 return (EOI);
303 }
304 }
305 }
306 if (any_sigs())
307 return (EOI);
308 }
309
310 found:
311 if (ch_bufhead != bn) {
312 /*
313 * Move the buffer to the head of the buffer chain.
314 * This orders the buffer chain, most- to least-recently used.
315 */
316 BUF_RM(bn);
317 BUF_INS_HEAD(bn);
318
319 /*
320 * Move to head of hash chain too.
321 */
322 BUF_HASH_RM(bn);
323 BUF_HASH_INS(bn, h);
324 }
325
326 if (ch_offset >= bp->datasize)
327 /*
328 * After all that, we still don't have enough data.
329 * Go back and try again.
330 */
331 goto read_more;
332
333 return (bp->data[ch_offset]);
334 }
335
336 /*
337 * ch_ungetchar is a rather kludgy and limited way to push
338 * a single char onto an input file descriptor.
339 */
340 void
ch_ungetchar(int c)341 ch_ungetchar(int c)
342 {
343 if (c != -1 && ch_ungotchar != -1)
344 error("ch_ungetchar overrun", NULL);
345 ch_ungotchar = c;
346 }
347
348 /*
349 * Close the logfile.
350 * If we haven't read all of standard input into it, do that now.
351 */
352 void
end_logfile(void)353 end_logfile(void)
354 {
355 static int tried = FALSE;
356
357 if (logfile < 0)
358 return;
359 if (!tried && ch_fsize == -1) {
360 tried = TRUE;
361 ierror("Finishing logfile", NULL);
362 while (ch_forw_get() != EOI)
363 if (abort_sigs())
364 break;
365 }
366 close(logfile);
367 logfile = -1;
368 free(namelogfile);
369 namelogfile = NULL;
370 }
371
372 /*
373 * Start a log file AFTER less has already been running.
374 * Invoked from the - command; see toggle_option().
375 * Write all the existing buffered data to the log file.
376 */
377 void
sync_logfile(void)378 sync_logfile(void)
379 {
380 struct buf *bp;
381 struct bufnode *bn;
382 int warned = FALSE;
383 BLOCKNUM block;
384 BLOCKNUM nblocks;
385
386 nblocks = (ch_fpos + LBUFSIZE - 1) / LBUFSIZE;
387 for (block = 0; block < nblocks; block++) {
388 int wrote = FALSE;
389 FOR_BUFS(bn) {
390 bp = bufnode_buf(bn);
391 if (bp->block == block) {
392 (void) write(logfile, (char *)bp->data,
393 bp->datasize);
394 wrote = TRUE;
395 break;
396 }
397 }
398 if (!wrote && !warned) {
399 error("Warning: log file is incomplete", NULL);
400 warned = TRUE;
401 }
402 }
403 }
404
405 /*
406 * Determine if a specific block is currently in one of the buffers.
407 */
408 static int
buffered(BLOCKNUM block)409 buffered(BLOCKNUM block)
410 {
411 struct buf *bp;
412 struct bufnode *bn;
413 int h;
414
415 h = BUFHASH(block);
416 FOR_BUFS_IN_CHAIN(h, bn) {
417 bp = bufnode_buf(bn);
418 if (bp->block == block)
419 return (TRUE);
420 }
421 return (FALSE);
422 }
423
424 /*
425 * Seek to a specified position in the file.
426 * Return 0 if successful, non-zero if can't seek there.
427 */
428 int
ch_seek(off_t pos)429 ch_seek(off_t pos)
430 {
431 BLOCKNUM new_block;
432 off_t len;
433
434 if (thisfile == NULL)
435 return (0);
436
437 len = ch_length();
438 if (pos < ch_zero() || (len != -1 && pos > len))
439 return (1);
440
441 new_block = pos / LBUFSIZE;
442 if (!(ch_flags & CH_CANSEEK) && pos != ch_fpos &&
443 !buffered(new_block)) {
444 if (ch_fpos > pos)
445 return (1);
446 while (ch_fpos < pos) {
447 if (ch_forw_get() == EOI)
448 return (1);
449 if (abort_sigs())
450 return (1);
451 }
452 return (0);
453 }
454 /*
455 * Set read pointer.
456 */
457 ch_block = new_block;
458 ch_offset = pos % LBUFSIZE;
459 return (0);
460 }
461
462 /*
463 * Seek to the end of the file.
464 */
465 int
ch_end_seek(void)466 ch_end_seek(void)
467 {
468 off_t len;
469
470 if (thisfile == NULL)
471 return (0);
472
473 if (ch_flags & CH_CANSEEK)
474 ch_fsize = filesize(ch_file);
475
476 len = ch_length();
477 if (len != -1)
478 return (ch_seek(len));
479
480 /*
481 * Do it the slow way: read till end of data.
482 */
483 while (ch_forw_get() != EOI)
484 if (abort_sigs())
485 return (1);
486 return (0);
487 }
488
489 /*
490 * Seek to the beginning of the file, or as close to it as we can get.
491 * We may not be able to seek there if input is a pipe and the
492 * beginning of the pipe is no longer buffered.
493 */
494 int
ch_beg_seek(void)495 ch_beg_seek(void)
496 {
497 struct bufnode *bn;
498 struct bufnode *firstbn;
499
500 /*
501 * Try a plain ch_seek first.
502 */
503 if (ch_seek(ch_zero()) == 0)
504 return (0);
505
506 /*
507 * Can't get to position 0.
508 * Look thru the buffers for the one closest to position 0.
509 */
510 firstbn = ch_bufhead;
511 if (firstbn == END_OF_CHAIN)
512 return (1);
513 FOR_BUFS(bn) {
514 if (bufnode_buf(bn)->block < bufnode_buf(firstbn)->block)
515 firstbn = bn;
516 }
517 ch_block = bufnode_buf(firstbn)->block;
518 ch_offset = 0;
519 return (0);
520 }
521
522 /*
523 * Return the length of the file, if known.
524 */
525 off_t
ch_length(void)526 ch_length(void)
527 {
528 if (thisfile == NULL)
529 return (-1);
530 if (ignore_eoi)
531 return (-1);
532 if (ch_flags & CH_NODATA)
533 return (0);
534 return (ch_fsize);
535 }
536
537 /*
538 * Return the current position in the file.
539 */
540 off_t
ch_tell(void)541 ch_tell(void)
542 {
543 if (thisfile == NULL)
544 return (-1);
545 return ((ch_block * LBUFSIZE) + ch_offset);
546 }
547
548 /*
549 * Get the current char and post-increment the read pointer.
550 */
551 int
ch_forw_get(void)552 ch_forw_get(void)
553 {
554 int c;
555
556 if (thisfile == NULL)
557 return (EOI);
558 c = ch_get();
559 if (c == EOI)
560 return (EOI);
561 if (ch_offset < LBUFSIZE-1) {
562 ch_offset++;
563 } else {
564 ch_block ++;
565 ch_offset = 0;
566 }
567 return (c);
568 }
569
570 /*
571 * Pre-decrement the read pointer and get the new current char.
572 */
573 int
ch_back_get(void)574 ch_back_get(void)
575 {
576 if (thisfile == NULL)
577 return (EOI);
578 if (ch_offset > 0) {
579 ch_offset --;
580 } else {
581 if (ch_block <= 0)
582 return (EOI);
583 if (!(ch_flags & CH_CANSEEK) && !buffered(ch_block-1))
584 return (EOI);
585 ch_block--;
586 ch_offset = LBUFSIZE-1;
587 }
588 return (ch_get());
589 }
590
591 /*
592 * Set max amount of buffer space.
593 * bufspace is in units of 1024 bytes. -1 mean no limit.
594 */
595 void
ch_setbufspace(int bufspace)596 ch_setbufspace(int bufspace)
597 {
598 if (bufspace < 0) {
599 maxbufs = -1;
600 } else {
601 maxbufs = ((bufspace * 1024) + LBUFSIZE-1) / LBUFSIZE;
602 if (maxbufs < 1)
603 maxbufs = 1;
604 }
605 }
606
607 /*
608 * Flush (discard) any saved file state, including buffer contents.
609 */
610 void
ch_flush(void)611 ch_flush(void)
612 {
613 struct bufnode *bn;
614
615 if (thisfile == NULL)
616 return;
617
618 if (!(ch_flags & CH_CANSEEK)) {
619 /*
620 * If input is a pipe, we don't flush buffer contents,
621 * since the contents can't be recovered.
622 */
623 ch_fsize = -1;
624 return;
625 }
626
627 /*
628 * Initialize all the buffers.
629 */
630 FOR_BUFS(bn) {
631 bufnode_buf(bn)->block = -1;
632 }
633
634 /*
635 * Figure out the size of the file, if we can.
636 */
637 ch_fsize = filesize(ch_file);
638
639 /*
640 * Seek to a known position: the beginning of the file.
641 */
642 ch_fpos = 0;
643 ch_block = 0; /* ch_fpos / LBUFSIZE; */
644 ch_offset = 0; /* ch_fpos % LBUFSIZE; */
645
646 if (lseek(ch_file, (off_t)0, SEEK_SET) == (off_t)-1) {
647 /*
648 * Warning only; even if the seek fails for some reason,
649 * there's a good chance we're at the beginning anyway.
650 * {{ I think this is bogus reasoning. }}
651 */
652 error("seek error to 0", NULL);
653 }
654 }
655
656 /*
657 * Allocate a new buffer.
658 * The buffer is added to the tail of the buffer chain.
659 */
660 static int
ch_addbuf(void)661 ch_addbuf(void)
662 {
663 struct buf *bp;
664 struct bufnode *bn;
665
666 /*
667 * Allocate and initialize a new buffer and link it
668 * onto the tail of the buffer list.
669 */
670 bp = calloc(1, sizeof (struct buf));
671 if (bp == NULL)
672 return (1);
673 ch_nbufs++;
674 bp->block = -1;
675 bn = &bp->node;
676
677 BUF_INS_TAIL(bn);
678 BUF_HASH_INS(bn, 0);
679 return (0);
680 }
681
682 /*
683 *
684 */
685 static void
init_hashtbl(void)686 init_hashtbl(void)
687 {
688 int h;
689
690 for (h = 0; h < BUFHASH_SIZE; h++) {
691 thisfile->hashtbl[h].hnext = END_OF_HCHAIN(h);
692 thisfile->hashtbl[h].hprev = END_OF_HCHAIN(h);
693 }
694 }
695
696 /*
697 * Delete all buffers for this file.
698 */
699 static void
ch_delbufs(void)700 ch_delbufs(void)
701 {
702 struct bufnode *bn;
703
704 while (ch_bufhead != END_OF_CHAIN) {
705 bn = ch_bufhead;
706 BUF_RM(bn);
707 free(bufnode_buf(bn));
708 }
709 ch_nbufs = 0;
710 init_hashtbl();
711 }
712
713 /*
714 * Is it possible to seek on a file descriptor?
715 */
716 int
seekable(int f)717 seekable(int f)
718 {
719 return (lseek(f, (off_t)1, SEEK_SET) != (off_t)-1);
720 }
721
722 /*
723 * Force EOF to be at the current read position.
724 * This is used after an ignore_eof read, during which the EOF may change.
725 */
726 void
ch_set_eof(void)727 ch_set_eof(void)
728 {
729 ch_fsize = ch_fpos;
730 }
731
732
733 /*
734 * Initialize file state for a new file.
735 */
736 void
ch_init(int f,int flags)737 ch_init(int f, int flags)
738 {
739 /*
740 * See if we already have a filestate for this file.
741 */
742 thisfile = get_filestate(curr_ifile);
743 if (thisfile == NULL) {
744 /*
745 * Allocate and initialize a new filestate.
746 */
747 thisfile = calloc(1, sizeof (struct filestate));
748 thisfile->buflist.next = thisfile->buflist.prev = END_OF_CHAIN;
749 thisfile->nbufs = 0;
750 thisfile->flags = 0;
751 thisfile->fpos = 0;
752 thisfile->block = 0;
753 thisfile->offset = 0;
754 thisfile->file = -1;
755 thisfile->fsize = -1;
756 ch_flags = flags;
757 init_hashtbl();
758 /*
759 * Try to seek; set CH_CANSEEK if it works.
760 */
761 if ((flags & CH_CANSEEK) && !seekable(f))
762 ch_flags &= ~CH_CANSEEK;
763 set_filestate(curr_ifile, (void *) thisfile);
764 }
765 if (thisfile->file == -1)
766 thisfile->file = f;
767 ch_flush();
768 }
769
770 /*
771 * Close a filestate.
772 */
773 void
ch_close(void)774 ch_close(void)
775 {
776 int keepstate = FALSE;
777
778 if (thisfile == NULL)
779 return;
780
781 if (ch_flags & (CH_CANSEEK|CH_HELPFILE)) {
782 /*
783 * We can seek or re-open, so we don't need to keep buffers.
784 */
785 ch_delbufs();
786 } else {
787 keepstate = TRUE;
788 }
789 if (!(ch_flags & CH_KEEPOPEN)) {
790 /*
791 * We don't need to keep the file descriptor open
792 * (because we can re-open it.)
793 */
794 close(ch_file);
795 ch_file = -1;
796 } else {
797 keepstate = TRUE;
798 }
799 if (!keepstate) {
800 /*
801 * We don't even need to keep the filestate structure.
802 */
803 free(thisfile);
804 thisfile = NULL;
805 set_filestate(curr_ifile, NULL);
806 }
807 }
808
809 /*
810 * Return ch_flags for the current file.
811 */
812 int
ch_getflags(void)813 ch_getflags(void)
814 {
815 if (thisfile == NULL)
816 return (0);
817 return (ch_flags);
818 }
819