2 * Copyright (C) 1994-2002, Index Data
4 * Sebastian Hammer, Adam Dickmeiss
6 * $Id: mfile.c,v 1.46 2002-07-16 09:52:20 heikki 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",
253 if (dirp->max_bytes >= 0)
254 dirp->avail_bytes -= part_f->bytes;
258 for (meta_f = ma->mfiles; meta_f; meta_f = meta_f->next)
260 logf (LOG_DEBUG, "mf_init: %s consists of %d part(s)", meta_f->name,
262 qsort(meta_f->files, meta_f->no_files, sizeof(part_file),
268 void mf_destroy(MFile_area ma)
286 meta_file *m = meta_f;
288 for (i = 0; i<m->no_files; i++)
290 xfree (m->files[i].path);
292 zebra_mutex_destroy (&meta_f->mutex);
293 meta_f = meta_f->next;
299 void mf_reset(MFile_area ma)
309 meta_file *m = meta_f;
312 for (i = 0; i<m->no_files; i++)
314 unlink (m->files[i].path);
315 xfree (m->files[i].path);
317 meta_f = meta_f->next;
325 * If !ma, Use MF_DEFAULT_AREA.
327 MFile mf_open(MFile_area ma, const char *name, int block_size, int wflag)
331 char tmp[FILENAME_MAX+1];
334 logf(LOG_DEBUG, "mf_open(%s bs=%d, %s)", name, block_size,
335 wflag ? "RW" : "RDONLY");
337 for (mnew = ma->mfiles; mnew; mnew = mnew->next)
338 if (!strcmp(name, mnew->name))
347 mnew = (meta_file *) xmalloc(sizeof(*mnew));
348 strcpy(mnew->name, name);
349 /* allocate one, empty file */
350 zebra_mutex_init (&mnew->mutex);
352 mnew->files[0].bytes = 0;
353 mnew->files[0].blocks = 0;
354 mnew->files[0].top = -1;
355 mnew->files[0].number = 0;
356 mnew->files[0].fd = -1;
357 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
358 for (dp = ma->dirs; dp && dp->max_bytes >= 0 && dp->avail_bytes <
359 mnew->min_bytes_creat; dp = dp->next);
362 logf (LOG_FATAL, "Insufficient space for new mfile.");
365 mnew->files[0].dir = dp;
366 sprintf(tmp, "%s/%s-%d.mf", dp->name, mnew->name, 0);
367 mnew->files[0].path = xstrdup(tmp);
369 mnew->next = ma->mfiles;
374 for (i = 0; i < mnew->no_files; i++)
376 if (mnew->files[i].bytes % block_size)
377 mnew->files[i].bytes += block_size - mnew->files[i].bytes %
379 mnew->files[i].blocks = mnew->files[i].bytes / block_size;
383 mnew->blocksize = block_size;
384 mnew->min_bytes_creat = MF_MIN_BLOCKS_CREAT * block_size;
389 for (i = 0; i < mnew->no_files; i++)
391 mnew->files[i].blocks = mnew->files[i].bytes / mnew->blocksize;
392 if (i == mnew->no_files - 1)
393 mnew->files[i].top = -1;
396 i ? (mnew->files[i-1].top + mnew->files[i].blocks)
397 : (mnew->files[i].blocks - 1);
405 int mf_close(MFile mf)
409 logf (LOG_DEBUG, "mf_close(%s)", mf->name);
411 for (i = 0; i < mf->no_files; i++)
412 if (mf->files[i].fd >= 0)
414 fsync(mf->files[i].fd);
415 close(mf->files[i].fd);
416 mf->files[i].fd = -1;
423 * Read one block from a metafile. Interface mirrors bfile.
425 int mf_read(MFile mf, int no, int offset, int nbytes, void *buf)
429 zebra_mutex_lock (&mf->mutex);
430 if ((rd = file_position(mf, no, offset)) < 0)
434 zebra_mutex_unlock (&mf->mutex);
439 yaz_log (LOG_FATAL, "mf_read %s internal error", mf->name);
443 toread = nbytes ? nbytes : mf->blocksize;
444 if ((rd = read(mf->files[mf->cur_file].fd, buf, toread)) < 0)
446 logf (LOG_FATAL|LOG_ERRNO, "mf_read: Read failed (%s)",
447 mf->files[mf->cur_file].path);
450 zebra_mutex_unlock (&mf->mutex);
460 int mf_write(MFile mf, int no, int offset, int nbytes, const void *buf)
462 int ps, nblocks, towrite;
464 char tmp[FILENAME_MAX+1];
465 unsigned char dummych = '\xff';
467 zebra_mutex_lock (&mf->mutex);
468 if ((ps = file_position(mf, no, offset)) < 0)
470 yaz_log (LOG_FATAL, "mf_write %s internal error (1)", mf->name);
473 /* file needs to grow */
474 while (ps >= mf->files[mf->cur_file].blocks)
476 off_t needed = (ps - mf->files[mf->cur_file].blocks + 1) *
478 /* file overflow - allocate new file */
479 if (mf->files[mf->cur_file].dir->max_bytes >= 0 &&
480 needed > mf->files[mf->cur_file].dir->avail_bytes)
483 if ((nblocks = mf->files[mf->cur_file].dir->avail_bytes /
486 logf (LOG_DEBUG, "Capping off file %s at pos %d",
487 mf->files[mf->cur_file].path, nblocks);
488 if ((ps = file_position(mf,
489 (mf->cur_file ? mf->files[mf->cur_file-1].top : 0) +
490 mf->files[mf->cur_file].blocks + nblocks - 1, 0)) < 0)
492 yaz_log (LOG_FATAL, "mf_write %s internal error (2)",
496 logf (LOG_DEBUG, "ps = %d", ps);
497 if (write(mf->files[mf->cur_file].fd, &dummych, 1) < 1)
499 logf (LOG_ERRNO|LOG_FATAL, "mf_write %s internal error (3)",
503 mf->files[mf->cur_file].blocks += nblocks;
504 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
505 mf->files[mf->cur_file].dir->avail_bytes -= nblocks *
509 logf (LOG_DEBUG, "Creating new file.");
510 for (dp = mf->ma->dirs; dp && dp->max_bytes >= 0 &&
511 dp->avail_bytes < needed; dp = dp->next);
514 logf (LOG_FATAL, "Cannot allocate more space for %s",
518 mf->files[mf->cur_file].top = (mf->cur_file ?
519 mf->files[mf->cur_file-1].top : -1) +
520 mf->files[mf->cur_file].blocks;
521 mf->files[++(mf->cur_file)].top = -1;
522 mf->files[mf->cur_file].dir = dp;
523 mf->files[mf->cur_file].number =
524 mf->files[mf->cur_file-1].number + 1;
525 mf->files[mf->cur_file].blocks =
526 mf->files[mf->cur_file].bytes = 0;
527 mf->files[mf->cur_file].fd = -1;
528 sprintf(tmp, "%s/%s-%d.mf", dp->name, mf->name,
529 mf->files[mf->cur_file].number);
530 mf->files[mf->cur_file].path = xstrdup(tmp);
532 /* open new file and position at beginning */
533 if ((ps = file_position(mf, no, offset)) < 0)
535 yaz_log (LOG_FATAL, "mf_write %s internal error (4)",
542 nblocks = ps - mf->files[mf->cur_file].blocks + 1;
543 mf->files[mf->cur_file].blocks += nblocks;
544 mf->files[mf->cur_file].bytes += nblocks * mf->blocksize;
545 if (mf->files[mf->cur_file].dir->max_bytes >= 0)
546 mf->files[mf->cur_file].dir->avail_bytes -=
547 nblocks * mf->blocksize;
550 towrite = nbytes ? nbytes : mf->blocksize;
551 if (write(mf->files[mf->cur_file].fd, buf, towrite) < towrite)
553 logf (LOG_FATAL|LOG_ERRNO, "Write failed for file %s part %d",
554 mf->name, mf->cur_file);
557 zebra_mutex_unlock (&mf->mutex);
562 * Destroy a metafile, unlinking component files. File must be open.
564 int mf_unlink(MFile mf)
568 for (i = 0; i < mf->no_files; i++)
569 unlink (mf->files[i].path);
574 * Unlink the file by name, rather than MFile-handle. File should be closed.
576 int mf_unlink_name(MFile_area ma, const char *name)