2 * Copyright (C) 1994-1999, Index Data
4 * Sebastian Hammer, Adam Dickmeiss
7 * Revision 1.37 2000-03-15 15:00:30 adam
8 * First work on threaded version.
10 * Revision 1.36 1999/12/08 15:03:11 adam
11 * Implemented bf_reset.
13 * Revision 1.35 1999/10/14 14:33:50 adam
14 * Added truncation 5=106.
16 * Revision 1.34 1999/05/26 07:49:12 adam
19 * Revision 1.33 1999/05/12 13:08:06 adam
20 * First version of ISAMS.
22 * Revision 1.32 1999/04/28 14:53:07 adam
23 * Fixed stupid bug regarding split-files.
25 * Revision 1.31 1999/02/18 12:49:33 adam
26 * Changed file naming scheme for register files as well as record
29 * Revision 1.30 1999/02/02 14:50:02 adam
30 * Updated WIN32 code specific sections. Changed header.
32 * Revision 1.29 1998/05/27 14:28:34 adam
33 * Fixed bug in mf_write. 'Cap off' byte written at wrong offset.
35 * Revision 1.28 1998/05/20 10:00:35 adam
36 * Fixed register spec so that colon isn't treated as size separator
37 * unless followed by [0-9+-] in order to allow DOS drive specifications.
39 * Revision 1.27 1998/02/10 11:55:07 adam
42 * Revision 1.26 1997/10/27 14:25:38 adam
45 * Revision 1.25 1997/09/18 08:59:16 adam
46 * Extra generic handle for the character mapping routines.
48 * Revision 1.24 1997/09/17 12:19:06 adam
49 * Zebra version corresponds to YAZ version 1.4.
50 * Changed Zebra server so that it doesn't depend on global common_resource.
52 * Revision 1.23 1997/09/09 13:37:53 adam
53 * Partial port to WIN95/NT.
55 * Revision 1.22 1997/09/04 13:56:39 adam
56 * Added O_BINARY to open calls.
58 * Revision 1.21 1996/10/29 13:56:18 adam
59 * Include of zebrautl.h instead of alexutil.h.
61 * Revision 1.20 1996/05/14 12:10:16 quinn
64 * Revision 1.19 1996/05/01 07:16:30 quinn
67 * Revision 1.18 1996/04/09 06:47:30 adam
68 * Function scan_areadef doesn't use sscanf (%n fails on this Linux).
70 * Revision 1.17 1996/03/20 13:29:11 quinn
73 * Revision 1.16 1995/12/12 15:57:57 adam
74 * Implemented mf_unlink. cf_unlink uses mf_unlink.
76 * Revision 1.15 1995/12/08 16:21:14 adam
77 * Work on commit/update.
79 * Revision 1.14 1995/12/05 13:12:37 quinn
82 * Revision 1.13 1995/11/30 17:00:50 adam
83 * Several bug fixes. Commit system runs now.
85 * Revision 1.12 1995/11/24 17:26:11 quinn
86 * Mostly about making some ISAM stuff in the config file optional.
88 * Revision 1.11 1995/11/13 09:32:43 quinn
91 * Revision 1.10 1995/09/04 12:33:22 adam
92 * Various cleanup. YAZ util used instead.
94 * Revision 1.9 1994/11/04 14:26:39 quinn
97 * Revision 1.8 1994/10/05 16:56:42 quinn
100 * Revision 1.7 1994/09/19 14:12:37 quinn
103 * Revision 1.6 1994/09/14 13:10:15 quinn
104 * Corrected some bugs in the init-phase
106 * Revision 1.5 1994/09/12 08:01:51 quinn
109 * Revision 1.4 1994/09/01 14:51:07 quinn
110 * Allowed mf_write to write beyond eof+1.
112 * Revision 1.3 1994/08/24 09:37:17 quinn
113 * Changed reaction to read return values.
115 * Revision 1.2 1994/08/23 14:50:48 quinn
118 * Revision 1.1 1994/08/23 14:41:33 quinn
119 * First functional version.
125 * TODO: The size estimates in init may not be accurate due to
126 * only partially written final blocks.
129 #include <sys/types.h>
143 #include <zebra-lock.h>
144 #include <zebrautl.h>
147 static int scan_areadef(MFile_area ma, const char *name, const char *ad)
150 * If no definition is given, use current directory, unlimited.
152 char dirname[FILENAME_MAX+1];
153 mf_dir **dp = &ma->dirs, *dir = *dp;
159 const char *ad0 = ad;
160 int i = 0, fact = 1, multi;
163 while (*ad == ' ' || *ad == '\t')
169 if (*ad == ':' && strchr ("+-0123456789", ad[1]))
171 if (i < FILENAME_MAX)
178 logf (LOG_WARN, "Missing colon after path: %s", ad0);
183 logf (LOG_WARN, "Empty path: %s", ad0);
186 while (*ad == ' ' || *ad == '\t')
196 if (*ad < '0' || *ad > '9')
198 logf (LOG_FATAL, "Missing size after path: %s", ad0);
202 while (*ad >= '0' && *ad <= '9')
203 size = size*10 + (*ad++ - '0');
206 case 'B': case 'b': multi = 1; break;
207 case 'K': case 'k': multi = 1024; break;
208 case 'M': case 'm': multi = 1048576; break;
209 case 'G': case 'g': multi = 1073741824; break;
211 logf (LOG_FATAL, "Missing unit: %s", ad0);
214 logf (LOG_FATAL, "Illegal unit: %c in %s", *ad, ad0);
218 *dp = dir = (mf_dir *) xmalloc(sizeof(mf_dir));
220 strcpy(dir->name, dirname);
221 dir->max_bytes = dir->avail_bytes = fact * size * multi;
227 static int file_position(MFile mf, int pos, int offset)
229 int off = 0, c = mf->cur_file, ps;
231 if ((c > 0 && pos <= mf->files[c-1].top) ||
232 (c < mf->no_files -1 && pos > mf->files[c].top))
235 while (c + 1 < mf->no_files && mf->files[c].top < pos)
237 off += mf->files[c].blocks;
240 assert(c < mf->no_files);
243 off = c ? (mf->files[c-1].top + 1) : 0;
244 if (mf->files[c].fd < 0 && (mf->files[c].fd = open(mf->files[c].path,
245 mf->wr ? (O_BINARY|O_RDWR|O_CREAT) : (O_BINARY|O_RDONLY), 0666)) < 0)
247 if (!mf->wr && errno == ENOENT && off == 0)
249 logf (LOG_WARN|LOG_ERRNO, "Failed to open %s", mf->files[c].path);
252 if (lseek(mf->files[c].fd, (ps = pos - off) * mf->blocksize + offset,
255 logf (LOG_WARN|LOG_ERRNO, "Failed to seek in %s", mf->files[c].path);
262 static int cmp_part_file(const void *p1, const void *p2)
264 return ((part_file *)p1)->number - ((part_file *)p2)->number;
268 * Create a new area, cotaining metafiles in directories.
269 * Find the part-files in each directory, and inventory the existing metafiles.
271 MFile_area mf_init(const char *name, const char *spec)
273 MFile_area ma = (MFile_area) xmalloc(sizeof(*ma));
276 part_file *part_f = 0;
280 char metaname[FILENAME_MAX+1], tmpnam[FILENAME_MAX+1];
282 logf (LOG_DEBUG, "mf_init(%s)", name);
283 strcpy(ma->name, name);
286 if (scan_areadef(ma, name, spec) < 0)
288 logf (LOG_WARN, "Failed to access description of '%s'", name);
291 /* look at each directory */
292 for (dirp = ma->dirs; dirp; dirp = dirp->next)
294 if (!(dd = opendir(dirp->name)))
296 logf (LOG_WARN|LOG_ERRNO, "Failed to open directory %s",
300 /* look at each file */
301 while ((dent = readdir(dd)))
303 if (*dent->d_name == '.')
305 if (sscanf(dent->d_name, "%[^-]-%d.mf", metaname, &number) != 2)
307 logf (LOG_DEBUG, "bf: %s is not a part-file.", dent->d_name);
310 for (meta_f = ma->mfiles; meta_f; meta_f = meta_f->next)
313 if (!strcmp(meta_f->name, metaname))
315 part_f = &meta_f->files[meta_f->no_files++];
322 meta_f = (meta_file *) xmalloc(sizeof(*meta_f));
324 meta_f->next = ma->mfiles;
326 meta_f->cur_file = -1;
328 strcpy(meta_f->name, metaname);
329 part_f = &meta_f->files[0];
330 meta_f->no_files = 1;
332 part_f->number = number;
335 sprintf(tmpnam, "%s/%s", dirp->name, dent->d_name);
336 part_f->path = xstrdup(tmpnam);
338 if ((fd = open(part_f->path, O_BINARY|O_RDONLY)) < 0)
340 logf (LOG_FATAL|LOG_ERRNO, "Failed to access %s",
344 if ((part_f->bytes = lseek(fd, 0, SEEK_END)) < 0)
346 logf (LOG_FATAL|LOG_ERRNO, "Failed to seek in %s",
351 if (dirp->max_bytes >= 0)
352 dirp->avail_bytes -= part_f->bytes;
356 for (meta_f = ma->mfiles; meta_f; meta_f = meta_f->next)
358 logf (LOG_DEBUG, "mf_init: %s consists of %d part(s)", meta_f->name,
360 qsort(meta_f->files, meta_f->no_files, sizeof(part_file),
366 void mf_destroy(MFile_area ma)
384 meta_file *m = meta_f;
386 for (i = 0; i<m->no_files; i++)
388 xfree (m->files[i].path);
390 zebra_mutex_destroy (&meta_f->mutex);
391 meta_f = meta_f->next;
397 void mf_reset(MFile_area ma)
407 meta_file *m = meta_f;
410 for (i = 0; i<m->no_files; i++)
412 unlink (m->files[i].path);
413 xfree (m->files[i].path);
415 meta_f = meta_f->next;
423 * If !ma, Use MF_DEFAULT_AREA.
425 MFile mf_open(MFile_area ma, const char *name, int block_size, int wflag)
429 char tmp[FILENAME_MAX+1];
432 logf(LOG_DEBUG, "mf_open(%s bs=%d, %s)", name, block_size,
433 wflag ? "RW" : "RDONLY");
435 for (mnew = ma->mfiles; mnew; mnew = mnew->next)
436 if (!strcmp(name, mnew->name))
445 mnew = (meta_file *) xmalloc(sizeof(*mnew));
446 strcpy(mnew->name, name);
447 /* allocate one, empty file */
448 zebra_mutex_init (&mnew->mutex);
450 mnew->files[0].bytes = 0;
451 mnew->files[0].blocks = 0;
452 mnew->files[0].top = -1;
453 mnew->files[0].number = 0;
454 mnew->files[0].fd = -1;
455 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
456 for (dp = ma->dirs; dp && dp->max_bytes >= 0 && dp->avail_bytes <
457 mnew->min_bytes_creat; dp = dp->next);
460 logf (LOG_FATAL, "Insufficient space for new mfile.");
463 mnew->files[0].dir = dp;
464 sprintf(tmp, "%s/%s-%d.mf", dp->name, mnew->name, 0);
465 mnew->files[0].path = xstrdup(tmp);
467 mnew->next = ma->mfiles;
472 for (i = 0; i < mnew->no_files; i++)
474 if (mnew->files[i].bytes % block_size)
475 mnew->files[i].bytes += block_size - mnew->files[i].bytes %
477 mnew->files[i].blocks = mnew->files[i].bytes / block_size;
481 mnew->blocksize = block_size;
482 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
487 for (i = 0; i < mnew->no_files; i++)
489 mnew->files[i].blocks = mnew->files[i].bytes / mnew->blocksize;
490 if (i == mnew->no_files - 1)
491 mnew->files[i].top = -1;
494 i ? (mnew->files[i-1].top + mnew->files[i].blocks)
495 : (mnew->files[i].blocks - 1);
503 int mf_close(MFile mf)
507 logf (LOG_DEBUG, "mf_close(%s)", mf->name);
509 for (i = 0; i < mf->no_files; i++)
510 if (mf->files[i].fd >= 0)
512 close(mf->files[i].fd);
513 mf->files[i].fd = -1;
520 * Read one block from a metafile. Interface mirrors bfile.
522 int mf_read(MFile mf, int no, int offset, int nbytes, void *buf)
526 zebra_mutex_lock (&mf->mutex);
527 if ((rd = file_position(mf, no, offset)) < 0)
534 toread = nbytes ? nbytes : mf->blocksize;
535 if ((rd = read(mf->files[mf->cur_file].fd, buf, toread)) < 0)
537 logf (LOG_FATAL|LOG_ERRNO, "mf_read: Read failed (%s)",
538 mf->files[mf->cur_file].path);
541 zebra_mutex_unlock (&mf->mutex);
551 int mf_write(MFile mf, int no, int offset, int nbytes, const void *buf)
553 int ps, nblocks, towrite;
555 char tmp[FILENAME_MAX+1];
556 unsigned char dummych = '\xff';
558 zebra_mutex_lock (&mf->mutex);
559 if ((ps = file_position(mf, no, offset)) < 0)
561 /* file needs to grow */
562 while (ps >= mf->files[mf->cur_file].blocks)
564 /* file overflow - allocate new file */
565 if (mf->files[mf->cur_file].dir->max_bytes >= 0 &&
566 (ps - mf->files[mf->cur_file].blocks + 1) * mf->blocksize >
567 mf->files[mf->cur_file].dir->avail_bytes)
570 if ((nblocks = mf->files[mf->cur_file].dir->avail_bytes /
573 logf (LOG_DEBUG, "Capping off file %s at pos %d",
574 mf->files[mf->cur_file].path, nblocks);
575 if ((ps = file_position(mf,
576 (mf->cur_file ? mf->files[mf->cur_file-1].top : 0) +
577 mf->files[mf->cur_file].blocks + nblocks - 1, 0)) < 0)
579 logf (LOG_DEBUG, "ps = %d", ps);
580 if (write(mf->files[mf->cur_file].fd, &dummych, 1) < 1)
582 logf (LOG_ERRNO|LOG_FATAL, "write dummy");
585 mf->files[mf->cur_file].blocks += nblocks;
586 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
587 mf->files[mf->cur_file].dir->avail_bytes -= nblocks *
591 logf (LOG_DEBUG, "Creating new file.");
592 for (dp = mf->ma->dirs; dp && dp->max_bytes >= 0 &&
593 dp->avail_bytes < mf->min_bytes_creat; dp = dp->next);
596 logf (LOG_FATAL, "Cannot allocate more space for %s",
600 mf->files[mf->cur_file].top = (mf->cur_file ?
601 mf->files[mf->cur_file-1].top : -1) +
602 mf->files[mf->cur_file].blocks;
603 mf->files[++(mf->cur_file)].top = -1;
604 mf->files[mf->cur_file].dir = dp;
605 mf->files[mf->cur_file].number =
606 mf->files[mf->cur_file-1].number + 1;
607 mf->files[mf->cur_file].blocks =
608 mf->files[mf->cur_file].bytes = 0;
609 mf->files[mf->cur_file].fd = -1;
610 sprintf(tmp, "%s/%s-%d.mf", dp->name, mf->name,
611 mf->files[mf->cur_file].number);
612 mf->files[mf->cur_file].path = xstrdup(tmp);
614 /* open new file and position at beginning */
615 if ((ps = file_position(mf, no, offset)) < 0)
620 nblocks = ps - mf->files[mf->cur_file].blocks + 1;
621 mf->files[mf->cur_file].blocks += nblocks;
622 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
623 if (mf->files[mf->cur_file].dir->max_bytes >= 0)
624 mf->files[mf->cur_file].dir->avail_bytes -=
625 nblocks * mf->blocksize;
628 towrite = nbytes ? nbytes : mf->blocksize;
629 if (write(mf->files[mf->cur_file].fd, buf, towrite) < towrite)
631 logf (LOG_FATAL|LOG_ERRNO, "Write failed for file %s part %d",
632 mf->name, mf->cur_file);
635 zebra_mutex_unlock (&mf->mutex);
640 * Destroy a metafile, unlinking component files. File must be open.
642 int mf_unlink(MFile mf)
646 for (i = 0; i < mf->no_files; i++)
647 unlink (mf->files[i].path);
652 * Unlink the file by name, rather than MFile-handle. File should be closed.
654 int mf_unlink_name(MFile_area ma, const char *name)