2 * Copyright (C) 1994-2002, Index Data
4 * Sebastian Hammer, Adam Dickmeiss
6 * $Id: mfile.c,v 1.48 2002-07-25 13:45:49 adam Exp $
11 * TODO: The size estimates in init may not be accurate due to
12 * only partially written final blocks.
15 #include <sys/types.h>
29 #include <zebra-lock.h>
33 static int scan_areadef(MFile_area ma, const char *ad, const char *base)
36 * If no definition is given, use current directory, unlimited.
38 char dirname[FILENAME_MAX+1];
39 mf_dir **dp = &ma->dirs, *dir = *dp;
46 int i = 0, fact = 1, multi;
49 while (*ad == ' ' || *ad == '\t')
53 if (!yaz_is_abspath(ad) && base)
55 strcpy (dirname, base);
61 if (*ad == ':' && strchr ("+-0123456789", ad[1]))
70 logf (LOG_WARN, "Missing colon after path: %s", ad0);
75 logf (LOG_WARN, "Empty path: %s", ad0);
78 while (*ad == ' ' || *ad == '\t')
88 if (*ad < '0' || *ad > '9')
90 logf (LOG_FATAL, "Missing size after path: %s", ad0);
94 while (*ad >= '0' && *ad <= '9')
95 size = size*10 + (*ad++ - '0');
98 case 'B': case 'b': multi = 1; break;
99 case 'K': case 'k': multi = 1024; break;
100 case 'M': case 'm': multi = 1048576; break;
101 case 'G': case 'g': multi = 1073741824; break;
103 logf (LOG_FATAL, "Missing unit: %s", ad0);
106 logf (LOG_FATAL, "Illegal unit: %c in %s", *ad, ad0);
110 *dp = dir = (mf_dir *) xmalloc(sizeof(mf_dir));
112 strcpy(dir->name, dirname);
113 dir->max_bytes = dir->avail_bytes = fact * size * multi;
119 static int file_position(MFile mf, int pos, int offset)
121 int off = 0, c = mf->cur_file, ps;
123 if ((c > 0 && pos <= mf->files[c-1].top) ||
124 (c < mf->no_files -1 && pos > mf->files[c].top))
127 while (c + 1 < mf->no_files && mf->files[c].top < pos)
129 off += mf->files[c].blocks;
132 assert(c < mf->no_files);
135 off = c ? (mf->files[c-1].top + 1) : 0;
136 if (mf->files[c].fd < 0)
138 if ((mf->files[c].fd = open(mf->files[c].path,
140 (O_BINARY|O_RDWR|O_CREAT) :
141 (O_BINARY|O_RDONLY), 0666)) < 0)
143 if (!mf->wr && errno == ENOENT && off == 0)
145 logf (LOG_WARN|LOG_ERRNO, "Failed to open %s", mf->files[c].path);
149 if (lseek(mf->files[c].fd, (ps = pos - off) * mf->blocksize + offset,
152 logf (LOG_WARN|LOG_ERRNO, "Failed to seek in %s", mf->files[c].path);
159 static int cmp_part_file(const void *p1, const void *p2)
161 return ((part_file *)p1)->number - ((part_file *)p2)->number;
165 * Create a new area, cotaining metafiles in directories.
166 * Find the part-files in each directory, and inventory the existing metafiles.
168 MFile_area mf_init(const char *name, const char *spec, const char *base)
170 MFile_area ma = (MFile_area) xmalloc(sizeof(*ma));
173 part_file *part_f = 0;
177 char metaname[FILENAME_MAX+1], tmpnam[FILENAME_MAX+1];
179 logf (LOG_DEBUG, "mf_init(%s)", name);
180 strcpy(ma->name, name);
183 if (scan_areadef(ma, spec, base) < 0)
185 logf (LOG_WARN, "Failed to access description of '%s'", name);
188 /* look at each directory */
189 for (dirp = ma->dirs; dirp; dirp = dirp->next)
191 if (!(dd = opendir(dirp->name)))
193 logf (LOG_WARN|LOG_ERRNO, "Failed to open directory %s",
197 /* look at each file */
198 while ((dent = readdir(dd)))
200 int len = strlen(dent->d_name);
201 const char *cp = strrchr (dent->d_name, '-');
202 if (strchr (".-", *dent->d_name))
204 if (len < 5 || !cp || strcmp (dent->d_name + len - 3, ".mf"))
207 memcpy (metaname, dent->d_name, cp - dent->d_name);
208 metaname[ cp - dent->d_name] = '\0';
210 for (meta_f = ma->mfiles; meta_f; meta_f = meta_f->next)
213 if (!strcmp(meta_f->name, metaname))
215 part_f = &meta_f->files[meta_f->no_files++];
222 meta_f = (meta_file *) xmalloc(sizeof(*meta_f));
223 zebra_mutex_init (&meta_f->mutex);
225 meta_f->next = ma->mfiles;
227 meta_f->cur_file = -1;
229 strcpy(meta_f->name, metaname);
230 part_f = &meta_f->files[0];
231 meta_f->no_files = 1;
233 part_f->number = number;
236 sprintf(tmpnam, "%s/%s", dirp->name, dent->d_name);
237 part_f->path = xstrdup(tmpnam);
239 if ((fd = open(part_f->path, O_BINARY|O_RDONLY)) < 0)
241 logf (LOG_FATAL|LOG_ERRNO, "Failed to access %s",
245 if ((part_f->bytes = lseek(fd, 0, SEEK_END)) < 0)
247 logf (LOG_FATAL|LOG_ERRNO, "Failed to seek in %s",
255 if (dirp->max_bytes >= 0)
256 dirp->avail_bytes -= part_f->bytes;
260 for (meta_f = ma->mfiles; meta_f; meta_f = meta_f->next)
262 logf (LOG_DEBUG, "mf_init: %s consists of %d part(s)", meta_f->name,
264 qsort(meta_f->files, meta_f->no_files, sizeof(part_file),
270 void mf_destroy(MFile_area ma)
288 meta_file *m = meta_f;
290 for (i = 0; i<m->no_files; i++)
292 xfree (m->files[i].path);
294 zebra_mutex_destroy (&meta_f->mutex);
295 meta_f = meta_f->next;
301 void mf_reset(MFile_area ma)
311 meta_file *m = meta_f;
314 for (i = 0; i<m->no_files; i++)
316 unlink (m->files[i].path);
317 xfree (m->files[i].path);
319 meta_f = meta_f->next;
327 * If !ma, Use MF_DEFAULT_AREA.
329 MFile mf_open(MFile_area ma, const char *name, int block_size, int wflag)
333 char tmp[FILENAME_MAX+1];
336 logf(LOG_DEBUG, "mf_open(%s bs=%d, %s)", name, block_size,
337 wflag ? "RW" : "RDONLY");
339 for (mnew = ma->mfiles; mnew; mnew = mnew->next)
340 if (!strcmp(name, mnew->name))
349 mnew = (meta_file *) xmalloc(sizeof(*mnew));
350 strcpy(mnew->name, name);
351 /* allocate one, empty file */
352 zebra_mutex_init (&mnew->mutex);
354 mnew->files[0].bytes = 0;
355 mnew->files[0].blocks = 0;
356 mnew->files[0].top = -1;
357 mnew->files[0].number = 0;
358 mnew->files[0].fd = -1;
359 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
360 for (dp = ma->dirs; dp && dp->max_bytes >= 0 && dp->avail_bytes <
361 mnew->min_bytes_creat; dp = dp->next);
364 logf (LOG_FATAL, "Insufficient space for new mfile.");
367 mnew->files[0].dir = dp;
368 sprintf(tmp, "%s/%s-%d.mf", dp->name, mnew->name, 0);
369 mnew->files[0].path = xstrdup(tmp);
371 mnew->next = ma->mfiles;
376 for (i = 0; i < mnew->no_files; i++)
378 if (mnew->files[i].bytes % block_size)
379 mnew->files[i].bytes += block_size - mnew->files[i].bytes %
381 mnew->files[i].blocks = mnew->files[i].bytes / block_size;
385 mnew->blocksize = block_size;
386 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
391 for (i = 0; i < mnew->no_files; i++)
393 mnew->files[i].blocks = mnew->files[i].bytes / mnew->blocksize;
394 if (i == mnew->no_files - 1)
395 mnew->files[i].top = -1;
398 i ? (mnew->files[i-1].top + mnew->files[i].blocks)
399 : (mnew->files[i].blocks - 1);
407 int mf_close(MFile mf)
411 logf (LOG_DEBUG, "mf_close(%s)", mf->name);
413 for (i = 0; i < mf->no_files; i++)
414 if (mf->files[i].fd >= 0)
417 fsync(mf->files[i].fd);
419 close(mf->files[i].fd);
420 mf->files[i].fd = -1;
427 * Read one block from a metafile. Interface mirrors bfile.
429 int mf_read(MFile mf, int no, int offset, int nbytes, void *buf)
433 zebra_mutex_lock (&mf->mutex);
434 if ((rd = file_position(mf, no, offset)) < 0)
438 zebra_mutex_unlock (&mf->mutex);
443 yaz_log (LOG_FATAL, "mf_read %s internal error", mf->name);
447 toread = nbytes ? nbytes : mf->blocksize;
448 if ((rd = read(mf->files[mf->cur_file].fd, buf, toread)) < 0)
450 logf (LOG_FATAL|LOG_ERRNO, "mf_read: Read failed (%s)",
451 mf->files[mf->cur_file].path);
454 zebra_mutex_unlock (&mf->mutex);
464 int mf_write(MFile mf, int no, int offset, int nbytes, const void *buf)
466 int ps, nblocks, towrite;
468 char tmp[FILENAME_MAX+1];
469 unsigned char dummych = '\xff';
471 zebra_mutex_lock (&mf->mutex);
472 if ((ps = file_position(mf, no, offset)) < 0)
474 yaz_log (LOG_FATAL, "mf_write %s internal error (1)", mf->name);
477 /* file needs to grow */
478 while (ps >= mf->files[mf->cur_file].blocks)
480 off_t needed = (ps - mf->files[mf->cur_file].blocks + 1) *
482 /* file overflow - allocate new file */
483 if (mf->files[mf->cur_file].dir->max_bytes >= 0 &&
484 needed > mf->files[mf->cur_file].dir->avail_bytes)
487 if ((nblocks = mf->files[mf->cur_file].dir->avail_bytes /
490 logf (LOG_DEBUG, "Capping off file %s at pos %d",
491 mf->files[mf->cur_file].path, nblocks);
492 if ((ps = file_position(mf,
493 (mf->cur_file ? mf->files[mf->cur_file-1].top : 0) +
494 mf->files[mf->cur_file].blocks + nblocks - 1, 0)) < 0)
496 yaz_log (LOG_FATAL, "mf_write %s internal error (2)",
500 logf (LOG_DEBUG, "ps = %d", ps);
501 if (write(mf->files[mf->cur_file].fd, &dummych, 1) < 1)
503 logf (LOG_ERRNO|LOG_FATAL, "mf_write %s internal error (3)",
507 mf->files[mf->cur_file].blocks += nblocks;
508 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
509 mf->files[mf->cur_file].dir->avail_bytes -= nblocks *
513 logf (LOG_DEBUG, "Creating new file.");
514 for (dp = mf->ma->dirs; dp && dp->max_bytes >= 0 &&
515 dp->avail_bytes < needed; dp = dp->next);
518 logf (LOG_FATAL, "Cannot allocate more space for %s",
522 mf->files[mf->cur_file].top = (mf->cur_file ?
523 mf->files[mf->cur_file-1].top : -1) +
524 mf->files[mf->cur_file].blocks;
525 mf->files[++(mf->cur_file)].top = -1;
526 mf->files[mf->cur_file].dir = dp;
527 mf->files[mf->cur_file].number =
528 mf->files[mf->cur_file-1].number + 1;
529 mf->files[mf->cur_file].blocks =
530 mf->files[mf->cur_file].bytes = 0;
531 mf->files[mf->cur_file].fd = -1;
532 sprintf(tmp, "%s/%s-%d.mf", dp->name, mf->name,
533 mf->files[mf->cur_file].number);
534 mf->files[mf->cur_file].path = xstrdup(tmp);
536 /* open new file and position at beginning */
537 if ((ps = file_position(mf, no, offset)) < 0)
539 yaz_log (LOG_FATAL, "mf_write %s internal error (4)",
546 nblocks = ps - mf->files[mf->cur_file].blocks + 1;
547 mf->files[mf->cur_file].blocks += nblocks;
548 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
549 if (mf->files[mf->cur_file].dir->max_bytes >= 0)
550 mf->files[mf->cur_file].dir->avail_bytes -=
551 nblocks * mf->blocksize;
554 towrite = nbytes ? nbytes : mf->blocksize;
555 if (write(mf->files[mf->cur_file].fd, buf, towrite) < towrite)
557 logf (LOG_FATAL|LOG_ERRNO, "Write failed for file %s part %d",
558 mf->name, mf->cur_file);
561 zebra_mutex_unlock (&mf->mutex);
566 * Destroy a metafile, unlinking component files. File must be open.
568 int mf_unlink(MFile mf)
572 for (i = 0; i < mf->no_files; i++)
573 unlink (mf->files[i].path);
578 * Unlink the file by name, rather than MFile-handle. File should be closed.
580 int mf_unlink_name(MFile_area ma, const char *name)