X-Git-Url: http://git.indexdata.com/?p=idzebra-moved-to-github.git;a=blobdiff_plain;f=index%2Frecindex.c;h=793898b794e3ff4f5abfafd74a1db69a46f14724;hp=73ef6339d94b3e9eadfc238d6a9d99472b23998a;hb=6c9fcd3b5d3108702fa1ffc92dab4ab6060f9a19;hpb=b8fb0cb30ffc1fb35a34608a27245dd091566bb9 diff --git a/index/recindex.c b/index/recindex.c index 73ef633..793898b 100644 --- a/index/recindex.c +++ b/index/recindex.c @@ -1,91 +1,41 @@ +/* $Id: recindex.c,v 1.41 2005-01-15 19:38:26 adam Exp $ + Copyright (C) 1995-2005 + Index Data ApS + +This file is part of the Zebra server. + +Zebra is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 2, or (at your option) any later +version. + +Zebra is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +You should have received a copy of the GNU General Public License +along with Zebra; see the file LICENSE.zebra. If not, write to the +Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA +02111-1307, USA. +*/ + +#define RIDX_CHUNK 128 + /* - * Copyright (C) 1994-1998, Index Data I/S - * All rights reserved. - * Sebastian Hammer, Adam Dickmeiss - * - * $Log: recindex.c,v $ - * Revision 1.20 1998-01-12 15:04:08 adam - * The test option (-s) only uses read-lock (and not write lock). - * - * Revision 1.19 1997/09/17 12:19:16 adam - * Zebra version corresponds to YAZ version 1.4. - * Changed Zebra server so that it doesn't depend on global common_resource. - * - * Revision 1.18 1997/07/15 16:28:42 adam - * Bug fix: storeData didn't work with files with multiple records. - * Bug fix: fixed memory management with records; not really well - * thought through. - * - * Revision 1.17 1997/02/12 20:39:46 adam - * Implemented options -f that limits the log to the first - * records. - * Changed some log messages also. - * - * Revision 1.16 1996/06/04 10:19:00 adam - * Minor changes - removed include of ctype.h. - * - * Revision 1.15 1996/05/13 14:23:06 adam - * Work on compaction of set/use bytes in dictionary. - * - * Revision 1.14 1996/02/01 20:48:15 adam - * The total size of records are always checked in rec_cache_insert to - * reduce memory usage. - * - * Revision 1.13 1995/12/11 09:12:49 adam - * The rec_get function returns NULL if record doesn't exist - will - * happen in the server if the result set records have been deleted since - * the creation of the set (i.e. the search). - * The server saves a result temporarily if it is 'volatile', i.e. the - * set is register dependent. - * - * Revision 1.12 1995/12/07 17:38:47 adam - * Work locking mechanisms for concurrent updates/commit. - * - * Revision 1.11 1995/12/06 13:58:26 adam - * Improved flushing of records - all flushes except the last one - * don't write the last accessed. Also flush takes place if record - * info occupy more than about 256k. - * - * Revision 1.10 1995/12/06 12:41:24 adam - * New command 'stat' for the index program. - * Filenames can be read from stdin by specifying '-'. - * Bug fix/enhancement of the transformation from terms to regular - * expressons in the search engine. + * Format of first block + * next (8 bytes) + * ref_count (2 bytes) + * block (500 bytes) * - * Revision 1.9 1995/11/30 08:34:33 adam - * Started work on commit facility. - * Changed a few malloc/free to xmalloc/xfree. - * - * Revision 1.8 1995/11/28 14:26:21 adam - * Bug fix: recordId with constant wasn't right. - * Bug fix: recordId dictionary entry wasn't deleted when needed. - * - * Revision 1.7 1995/11/28 09:09:43 adam - * Zebra config renamed. - * Use setting 'recordId' to identify record now. - * Bug fix in recindex.c: rec_release_blocks was invokeded even - * though the blocks were already released. - * File traversal properly deletes records when needed. - * - * Revision 1.6 1995/11/25 10:24:06 adam - * More record fields - they are enumerated now. - * New options: flagStoreData flagStoreKey. - * - * Revision 1.5 1995/11/22 17:19:18 adam - * Record management uses the bfile system. - * - * Revision 1.4 1995/11/20 16:59:46 adam - * New update method: the 'old' keys are saved for each records. - * - * Revision 1.3 1995/11/16 15:34:55 adam - * Uses new record management system in both indexer and server. - * - * Revision 1.2 1995/11/15 19:13:08 adam - * Work on record management. - * - * Revision 1.1 1995/11/15 14:46:20 adam - * Started work on better record management system. + * Format of subsequent blocks + * next (8 bytes) + * block (502 bytes) * + * Format of each record + * sysno + * (length, data) - pairs + * length = 0 if same as previous */ #include #include @@ -93,6 +43,9 @@ #include "recindxp.h" +#if HAVE_BZLIB_H +#include +#endif static void rec_write_head (Records p) { int r; @@ -103,77 +56,118 @@ static void rec_write_head (Records p) r = bf_write (p->index_BFile, 0, 0, sizeof(p->head), &p->head); if (r) { - logf (LOG_FATAL|LOG_ERRNO, "write head of %s", p->index_fname); + yaz_log (YLOG_FATAL|YLOG_ERRNO, "write head of %s", p->index_fname); exit (1); } } -static void rec_tmp_expand (Records p, int size, int dst_type) +static void rec_tmp_expand (Records p, int size) { if (p->tmp_size < size + 2048 || - p->tmp_size < p->head.block_size[dst_type]*2) + p->tmp_size < p->head.block_size[REC_BLOCK_TYPES-1]*2) { xfree (p->tmp_buf); - p->tmp_size = size + p->head.block_size[dst_type]*2 + 2048; - p->tmp_buf = xmalloc (p->tmp_size); + p->tmp_size = size + (int) + (p->head.block_size[REC_BLOCK_TYPES-1])*2 + 2048; + p->tmp_buf = (char *) xmalloc (p->tmp_size); } } -static int read_indx (Records p, int sysno, void *buf, int itemsize, +static int read_indx (Records p, SYSNO sysno, void *buf, int itemsize, int ignoreError) { int r; - int pos = (sysno-1)*itemsize; + zint pos = (sysno-1)*itemsize; + int off = (int) (pos%RIDX_CHUNK); + int sz1 = RIDX_CHUNK - off; /* sz1 is size of buffer to read.. */ + + if (sz1 > itemsize) + sz1 = itemsize; /* no more than itemsize bytes */ - r = bf_read (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); + r = bf_read (p->index_BFile, 1+pos/RIDX_CHUNK, off, sz1, buf); + if (r == 1 && sz1 < itemsize) /* boundary? - must read second part */ + r = bf_read (p->index_BFile, 2+pos/RIDX_CHUNK, 0, itemsize - sz1, + (char*) buf + sz1); if (r != 1 && !ignoreError) { - logf (LOG_FATAL|LOG_ERRNO, "read in %s at pos %ld", + yaz_log (YLOG_FATAL|YLOG_ERRNO, "read in %s at pos %ld", p->index_fname, (long) pos); exit (1); } return r; } -static void write_indx (Records p, int sysno, void *buf, int itemsize) +static void write_indx (Records p, SYSNO sysno, void *buf, int itemsize) { - int pos = (sysno-1)*itemsize; + zint pos = (sysno-1)*itemsize; + int off = (int) (pos%RIDX_CHUNK); + int sz1 = RIDX_CHUNK - off; /* sz1 is size of buffer to read.. */ - bf_write (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); + if (sz1 > itemsize) + sz1 = itemsize; /* no more than itemsize bytes */ + + bf_write(p->index_BFile, 1+pos/RIDX_CHUNK, off, sz1, buf); + if (sz1 < itemsize) /* boundary? must write second part */ + bf_write(p->index_BFile, 2+pos/RIDX_CHUNK, 0, itemsize - sz1, + (char*) buf + sz1); } -static void rec_release_blocks (Records p, int sysno) +static void rec_release_blocks (Records p, SYSNO sysno) { struct record_index_entry entry; - int freeblock, freenext; + zint freeblock; + char block_and_ref[sizeof(zint) + sizeof(short)]; int dst_type; + int first = 1; if (read_indx (p, sysno, &entry, sizeof(entry), 1) != 1) return ; - p->head.total_bytes -= entry.size; + freeblock = entry.next; assert (freeblock > 0); - dst_type = freeblock & 7; + dst_type = (int) (freeblock & 7); assert (dst_type < REC_BLOCK_TYPES); freeblock = freeblock / 8; while (freeblock) { - if (bf_read (p->data_BFile[dst_type], freeblock, 0, sizeof(freenext), - &freenext) != 1) + if (bf_read (p->data_BFile[dst_type], freeblock, 0, + first ? sizeof(block_and_ref) : sizeof(zint), + block_and_ref) != 1) { - logf (LOG_FATAL|LOG_ERRNO, "read in rec_del_single"); + yaz_log (YLOG_FATAL|YLOG_ERRNO, "read in rec_del_single"); exit (1); } - if (bf_write (p->data_BFile[dst_type], freeblock, 0, sizeof(freenext), + if (first) + { + short ref; + memcpy (&ref, block_and_ref + sizeof(freeblock), sizeof(ref)); + --ref; + memcpy (block_and_ref + sizeof(freeblock), &ref, sizeof(ref)); + if (ref) + { + if (bf_write (p->data_BFile[dst_type], freeblock, 0, + sizeof(block_and_ref), block_and_ref)) + { + yaz_log (YLOG_FATAL|YLOG_ERRNO, "write in rec_del_single"); + exit (1); + } + return; + } + first = 0; + } + + if (bf_write (p->data_BFile[dst_type], freeblock, 0, sizeof(freeblock), &p->head.block_free[dst_type])) { - logf (LOG_FATAL|LOG_ERRNO, "write in rec_del_single"); + yaz_log (YLOG_FATAL|YLOG_ERRNO, "write in rec_del_single"); exit (1); } p->head.block_free[dst_type] = freeblock; - freeblock = freenext; + memcpy (&freeblock, block_and_ref, sizeof(freeblock)); + p->head.block_used[dst_type]--; } + p->head.total_bytes -= entry.size; } static void rec_delete_single (Records p, Record rec) @@ -188,40 +182,18 @@ static void rec_delete_single (Records p, Record rec) write_indx (p, rec->sysno, &entry, sizeof(entry)); } - -static void rec_write_single (Records p, Record rec) +static void rec_write_tmp_buf (Records p, int size, SYSNO *sysnos) { - int i, size = 0; - char *cptr; - int dst_type = 0; - int no_written = 0; - int block_prev = -1, block_free; struct record_index_entry entry; - - for (i = 0; i < REC_NO_INFO; i++) - if (!rec->info[i]) - size += sizeof(*rec->size); - else - size += sizeof(*rec->size) + rec->size[i]; + int no_written = 0; + char *cptr = p->tmp_buf; + zint block_prev = -1, block_free; + int dst_type = 0; + int i; for (i = 1; i= p->head.block_move[i]) dst_type = i; - - rec_tmp_expand (p, size, dst_type); - - cptr = p->tmp_buf + sizeof(int); /* a hack! */ - for (i = 0; i < REC_NO_INFO; i++) - { - memcpy (cptr, &rec->size[i], sizeof(*rec->size)); - cptr += sizeof(*rec->size); - if (rec->info[i]) - { - memcpy (cptr, rec->info[i], rec->size[i]); - cptr += rec->size[i]; - } - } - cptr = p->tmp_buf; while (no_written < size) { block_free = p->head.block_free[dst_type]; @@ -231,7 +203,7 @@ static void rec_write_single (Records p, Record rec) block_free, 0, sizeof(*p->head.block_free), &p->head.block_free[dst_type]) != 1) { - logf (LOG_FATAL|LOG_ERRNO, "read in %s at free block %d", + yaz_log (YLOG_FATAL|YLOG_ERRNO, "read in %s at free block " ZINT_FORMAT, p->data_fname[dst_type], block_free); exit (1); } @@ -243,45 +215,45 @@ static void rec_write_single (Records p, Record rec) entry.next = block_free*8 + dst_type; entry.size = size; p->head.total_bytes += size; - write_indx (p, rec->sysno, &entry, sizeof(entry)); + while (*sysnos > 0) + { + write_indx (p, *sysnos, &entry, sizeof(entry)); + sysnos++; + } } else { - memcpy (cptr, &block_free, sizeof(int)); + memcpy (cptr, &block_free, sizeof(block_free)); bf_write (p->data_BFile[dst_type], block_prev, 0, 0, cptr); cptr = p->tmp_buf + no_written; } block_prev = block_free; - no_written += p->head.block_size[dst_type] - sizeof(int); + no_written += (int)(p->head.block_size[dst_type]) - sizeof(zint); p->head.block_used[dst_type]++; } assert (block_prev != -1); block_free = 0; - memcpy (cptr, &block_free, sizeof(int)); + memcpy (cptr, &block_free, sizeof(block_free)); bf_write (p->data_BFile[dst_type], block_prev, 0, - sizeof(int) + (p->tmp_buf+size) - cptr, cptr); + sizeof(block_free) + (p->tmp_buf+size) - cptr, cptr); } -static void rec_update_single (Records p, Record rec) -{ - rec_release_blocks (p, rec->sysno); - rec_write_single (p, rec); -} - -Records rec_open (BFiles bfs, int rw) +Records rec_open (BFiles bfs, int rw, int compression_method) { Records p; int i, r; + int version; - p = xmalloc (sizeof(*p)); + p = (Records) xmalloc (sizeof(*p)); + p->compression_method = compression_method; p->rw = rw; p->tmp_size = 1024; - p->tmp_buf = xmalloc (p->tmp_size); - p->index_fname = "recindex"; - p->index_BFile = bf_open (bfs, p->index_fname, 128, rw); + p->tmp_buf = (char *) xmalloc (p->tmp_size); + p->index_fname = "reci"; + p->index_BFile = bf_open (bfs, p->index_fname, RIDX_CHUNK, rw); if (p->index_BFile == NULL) { - logf (LOG_FATAL|LOG_ERRNO, "open %s", p->index_fname); + yaz_log (YLOG_FATAL|YLOG_ERRNO, "open %s", p->index_fname); exit (1); } r = bf_read (p->index_BFile, 0, 0, 0, p->tmp_buf); @@ -289,6 +261,7 @@ Records rec_open (BFiles bfs, int rw) { case 0: memcpy (p->head.magic, REC_HEAD_MAGIC, sizeof(p->head.magic)); + sprintf (p->head.version, "%3d", REC_VERSION); p->head.index_free = 0; p->head.index_last = 1; p->head.no_records = 0; @@ -304,7 +277,7 @@ Records rec_open (BFiles bfs, int rw) for (i = 1; ihead.block_size[i] = p->head.block_size[i-1] * 4; - p->head.block_move[i] = p->head.block_size[i] * 3; + p->head.block_move[i] = p->head.block_size[i] * 24; } if (rw) rec_write_head (p); @@ -313,67 +286,266 @@ Records rec_open (BFiles bfs, int rw) memcpy (&p->head, p->tmp_buf, sizeof(p->head)); if (memcmp (p->head.magic, REC_HEAD_MAGIC, sizeof(p->head.magic))) { - logf (LOG_FATAL, "read %s. bad header", p->index_fname); + yaz_log (YLOG_FATAL, "file %s has bad format", p->index_fname); exit (1); } + version = atoi (p->head.version); + if (version != REC_VERSION) + { + yaz_log (YLOG_FATAL, "file %s is version %d, but version" + " %d is required", p->index_fname, version, REC_VERSION); + exit (1); + } break; } for (i = 0; idata_fname[i] = xmalloc (strlen(str)+1); + sprintf (str, "recd%c", i + 'A'); + p->data_fname[i] = (char *) xmalloc (strlen(str)+1); strcpy (p->data_fname[i], str); p->data_BFile[i] = NULL; } for (i = 0; idata_BFile[i] = bf_open (bfs, p->data_fname[i], - p->head.block_size[i], + (int) (p->head.block_size[i]), rw))) { - logf (LOG_FATAL|LOG_ERRNO, "bf_open %s", p->data_fname[i]); + yaz_log (YLOG_FATAL|YLOG_ERRNO, "bf_open %s", p->data_fname[i]); exit (1); } } - p->cache_max = 10; + p->cache_max = 400; p->cache_cur = 0; - p->record_cache = xmalloc (sizeof(*p->record_cache)*p->cache_max); + p->record_cache = (struct record_cache_entry *) + xmalloc (sizeof(*p->record_cache)*p->cache_max); + zebra_mutex_init (&p->mutex); return p; } -static void rec_cache_flush (Records p, int saveCount) +static void rec_encode_unsigned (unsigned n, unsigned char *buf, int *len) { - int i, j; + (*len) = 0; + while (n > 127) + { + buf[*len] = 128 + (n & 127); + n = n >> 7; + (*len)++; + } + buf[*len] = n; + (*len)++; +} + +static void rec_decode_unsigned(unsigned *np, unsigned char *buf, int *len) +{ + unsigned n = 0; + unsigned w = 1; + (*len) = 0; + + while (buf[*len] > 127) + { + n += w*(buf[*len] & 127); + w = w << 7; + (*len)++; + } + n += w * buf[*len]; + (*len)++; + *np = n; +} + +static void rec_encode_zint (zint n, unsigned char *buf, int *len) +{ + (*len) = 0; + while (n > 127) + { + buf[*len] = (unsigned) (128 + (n & 127)); + n = n >> 7; + (*len)++; + } + buf[*len] = (unsigned) n; + (*len)++; +} + +static void rec_decode_zint(zint *np, unsigned char *buf, int *len) +{ + zint n = 0; + zint w = 1; + (*len) = 0; + + while (buf[*len] > 127) + { + n += w*(buf[*len] & 127); + w = w << 7; + (*len)++; + } + n += w * buf[*len]; + (*len)++; + *np = n; +} + +static void rec_cache_flush_block1 (Records p, Record rec, Record last_rec, + char **out_buf, int *out_size, + int *out_offset) +{ + int i; + int len; + + for (i = 0; isize[i] + 20 > *out_size) + { + int new_size = *out_offset + rec->size[i] + 65536; + char *np = (char *) xmalloc (new_size); + if (*out_offset) + memcpy (np, *out_buf, *out_offset); + xfree (*out_buf); + *out_size = new_size; + *out_buf = np; + } + if (i == 0) + { + rec_encode_zint (rec->sysno, *out_buf + *out_offset, &len); + (*out_offset) += len; + } + if (rec->size[i] == 0) + { + rec_encode_unsigned (1, *out_buf + *out_offset, &len); + (*out_offset) += len; + } + else if (last_rec && rec->size[i] == last_rec->size[i] && + !memcmp (rec->info[i], last_rec->info[i], rec->size[i])) + { + rec_encode_unsigned (0, *out_buf + *out_offset, &len); + (*out_offset) += len; + } + else + { + rec_encode_unsigned (rec->size[i]+1, *out_buf + *out_offset, &len); + (*out_offset) += len; + memcpy (*out_buf + *out_offset, rec->info[i], rec->size[i]); + (*out_offset) += rec->size[i]; + } + } +} + +static void rec_write_multiple (Records p, int saveCount) +{ + int i; + short ref_count = 0; + char compression_method; + Record last_rec = 0; + int out_size = 1000; + int out_offset = 0; + char *out_buf = (char *) xmalloc (out_size); + SYSNO *sysnos = (SYSNO *) xmalloc (sizeof(*sysnos) * (p->cache_cur + 1)); + SYSNO *sysnop = sysnos; - if (saveCount >= p->cache_cur) - saveCount = 0; for (i = 0; icache_cur - saveCount; i++) { struct record_cache_entry *e = p->record_cache + i; switch (e->flag) { - case recordFlagNop: - break; case recordFlagNew: - rec_write_single (p, e->rec); + rec_cache_flush_block1 (p, e->rec, last_rec, &out_buf, + &out_size, &out_offset); + *sysnop++ = e->rec->sysno; + ref_count++; + e->flag = recordFlagNop; + last_rec = e->rec; break; case recordFlagWrite: - rec_update_single (p, e->rec); + rec_release_blocks (p, e->rec->sysno); + rec_cache_flush_block1 (p, e->rec, last_rec, &out_buf, + &out_size, &out_offset); + *sysnop++ = e->rec->sysno; + ref_count++; + e->flag = recordFlagNop; + last_rec = e->rec; break; case recordFlagDelete: rec_delete_single (p, e->rec); + e->flag = recordFlagNop; break; + default: + break; } - rec_rm (&e->rec); } + + *sysnop = -1; + if (ref_count) + { + int csize = 0; /* indicate compression "not performed yet" */ + compression_method = p->compression_method; + switch (compression_method) + { + case REC_COMPRESS_BZIP2: +#if HAVE_BZLIB_H + csize = out_offset + (out_offset >> 6) + 620; + rec_tmp_expand (p, csize); +#ifdef BZ_CONFIG_ERROR + i = BZ2_bzBuffToBuffCompress +#else + i = bzBuffToBuffCompress +#endif + (p->tmp_buf+sizeof(zint)+sizeof(short)+ + sizeof(char), + &csize, out_buf, out_offset, 1, 0, 30); + if (i != BZ_OK) + { + yaz_log (YLOG_WARN, "bzBuffToBuffCompress error code=%d", i); + csize = 0; + } + yaz_log (YLOG_LOG, "compress %4d %5d %5d", ref_count, out_offset, + csize); +#endif + break; + case REC_COMPRESS_NONE: + break; + } + if (!csize) + { + /* either no compression or compression not supported ... */ + csize = out_offset; + rec_tmp_expand (p, csize); + memcpy (p->tmp_buf + sizeof(zint) + sizeof(short) + sizeof(char), + out_buf, out_offset); + csize = out_offset; + compression_method = REC_COMPRESS_NONE; + } + memcpy (p->tmp_buf + sizeof(zint), &ref_count, sizeof(ref_count)); + memcpy (p->tmp_buf + sizeof(zint)+sizeof(short), + &compression_method, sizeof(compression_method)); + + /* -------- compression */ + rec_write_tmp_buf (p, csize + sizeof(short) + sizeof(char), sysnos); + } + xfree (out_buf); + xfree (sysnos); +} + +static void rec_cache_flush (Records p, int saveCount) +{ + int i, j; + + if (saveCount >= p->cache_cur) + saveCount = 0; + + rec_write_multiple (p, saveCount); + + for (i = 0; icache_cur - saveCount; i++) + { + struct record_cache_entry *e = p->record_cache + i; + rec_rm (&e->rec); + } + /* i still being used ... */ for (j = 0; jrecord_cache+j, p->record_cache+i, sizeof(*p->record_cache)); p->cache_cur = saveCount; } -static Record *rec_cache_lookup (Records p, int sysno, +static Record *rec_cache_lookup (Records p, SYSNO sysno, enum recordCacheFlag flag) { int i; @@ -406,7 +578,7 @@ static void rec_cache_insert (Records p, Record rec, enum recordCacheFlag flag) for (j = 0; jsize[j]; } - if (used > 256000) + if (used > 90000) rec_cache_flush (p, 1); } assert (p->cache_cur < p->cache_max); @@ -423,6 +595,7 @@ void rec_close (Records *pp) assert (p); + zebra_mutex_destroy (&p->mutex); rec_cache_flush (p, 0); xfree (p->record_cache); @@ -443,14 +616,20 @@ void rec_close (Records *pp) *pp = NULL; } - -Record rec_get (Records p, int sysno) +static Record rec_get_int (Records p, SYSNO sysno) { - int i; + int i, in_size, r; Record rec, *recp; struct record_index_entry entry; - int freeblock, dst_type; + zint freeblock; + int dst_type; char *nptr, *cptr; + char *in_buf = 0; + char *bz_buf = 0; +#if HAVE_BZLIB_H + int bz_size; +#endif + char compression_method; assert (sysno > 0); assert (p); @@ -458,63 +637,147 @@ Record rec_get (Records p, int sysno) if ((recp = rec_cache_lookup (p, sysno, recordFlagNop))) return rec_cp (*recp); - if (!read_indx (p, sysno, &entry, sizeof(entry), 1)) + if (read_indx (p, sysno, &entry, sizeof(entry), 1) < 1) return NULL; /* record is not there! */ if (!entry.size) return NULL; /* record is deleted */ - dst_type = entry.next & 7; + dst_type = (int) (entry.next & 7); assert (dst_type < REC_BLOCK_TYPES); freeblock = entry.next / 8; assert (freeblock > 0); - rec = xmalloc (sizeof(*rec)); - rec_tmp_expand (p, entry.size, dst_type); + rec_tmp_expand (p, entry.size); cptr = p->tmp_buf; - bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + r = bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + if (r < 0) + return 0; memcpy (&freeblock, cptr, sizeof(freeblock)); while (freeblock) { - int tmp; + zint tmp; cptr += p->head.block_size[dst_type] - sizeof(freeblock); memcpy (&tmp, cptr, sizeof(tmp)); - bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + r = bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + if (r < 0) + return 0; memcpy (&freeblock, cptr, sizeof(freeblock)); memcpy (cptr, &tmp, sizeof(tmp)); } + rec = (Record) xmalloc (sizeof(*rec)); rec->sysno = sysno; - nptr = p->tmp_buf + sizeof(freeblock); - for (i = 0; i < REC_NO_INFO; i++) + memcpy (&compression_method, p->tmp_buf + sizeof(zint) + sizeof(short), + sizeof(compression_method)); + in_buf = p->tmp_buf + sizeof(zint) + sizeof(short) + sizeof(char); + in_size = entry.size - sizeof(short) - sizeof(char); + switch (compression_method) { - memcpy (&rec->size[i], nptr, sizeof(*rec->size)); - nptr += sizeof(*rec->size); - if (rec->size[i]) - { - rec->info[i] = xmalloc (rec->size[i]); - memcpy (rec->info[i], nptr, rec->size[i]); - nptr += rec->size[i]; - } - else - rec->info[i] = NULL; + case REC_COMPRESS_BZIP2: +#if HAVE_BZLIB_H + bz_size = entry.size * 20 + 100; + while (1) + { + bz_buf = (char *) xmalloc (bz_size); +#ifdef BZ_CONFIG_ERROR + i = BZ2_bzBuffToBuffDecompress +#else + i = bzBuffToBuffDecompress +#endif + (bz_buf, &bz_size, in_buf, in_size, 0, 0); + yaz_log (YLOG_LOG, "decompress %5d %5d", in_size, bz_size); + if (i == BZ_OK) + break; + yaz_log (YLOG_LOG, "failed"); + xfree (bz_buf); + bz_size *= 2; + } + in_buf = bz_buf; + in_size = bz_size; +#else + yaz_log (YLOG_FATAL, "cannot decompress record(s) in BZIP2 format"); + exit (1); +#endif + break; + case REC_COMPRESS_NONE: + break; } + for (i = 0; iinfo[i] = 0; + + nptr = in_buf; /* skip ref count */ + while (nptr < in_buf + in_size) + { + zint this_sysno; + int len; + rec_decode_zint (&this_sysno, nptr, &len); + nptr += len; + + for (i = 0; i < REC_NO_INFO; i++) + { + int this_size; + rec_decode_unsigned (&this_size, nptr, &len); + nptr += len; + + if (this_size == 0) + continue; + rec->size[i] = this_size-1; + + if (rec->size[i]) + { + rec->info[i] = nptr; + nptr += rec->size[i]; + } + else + rec->info[i] = NULL; + } + if (this_sysno == sysno) + break; + } + for (i = 0; iinfo[i] && rec->size[i]) + { + char *np = xmalloc (rec->size[i]+1); + memcpy (np, rec->info[i], rec->size[i]); + np[rec->size[i]] = '\0'; + rec->info[i] = np; + } + else + { + assert (rec->info[i] == 0); + assert (rec->size[i] == 0); + } + } + xfree (bz_buf); rec_cache_insert (p, rec, recordFlagNop); return rec; } -Record rec_new (Records p) +Record rec_get (Records p, SYSNO sysno) +{ + Record rec; + zebra_mutex_lock (&p->mutex); + + rec = rec_get_int (p, sysno); + zebra_mutex_unlock (&p->mutex); + return rec; +} + +static Record rec_new_int (Records p) { - int sysno, i; + int i; + SYSNO sysno; Record rec; assert (p); - rec = xmalloc (sizeof(*rec)); + rec = (Record) xmalloc (sizeof(*rec)); if (1 || p->head.index_free == 0) sysno = (p->head.index_last)++; else @@ -536,10 +799,21 @@ Record rec_new (Records p) return rec; } +Record rec_new (Records p) +{ + Record rec; + zebra_mutex_lock (&p->mutex); + + rec = rec_new_int (p); + zebra_mutex_unlock (&p->mutex); + return rec; +} + void rec_del (Records p, Record *recpp) { Record *recp; + zebra_mutex_lock (&p->mutex); (p->head.no_records)--; if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagDelete))) { @@ -551,6 +825,7 @@ void rec_del (Records p, Record *recpp) rec_cache_insert (p, *recpp, recordFlagDelete); rec_rm (recpp); } + zebra_mutex_unlock (&p->mutex); *recpp = NULL; } @@ -558,6 +833,7 @@ void rec_put (Records p, Record *recpp) { Record *recp; + zebra_mutex_lock (&p->mutex); if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagWrite))) { rec_rm (recp); @@ -568,6 +844,7 @@ void rec_put (Records p, Record *recpp) rec_cache_insert (p, *recpp, recordFlagWrite); rec_rm (recpp); } + zebra_mutex_unlock (&p->mutex); *recpp = NULL; } @@ -588,7 +865,7 @@ Record rec_cp (Record rec) Record n; int i; - n = xmalloc (sizeof(*n)); + n = (Record) xmalloc (sizeof(*n)); n->sysno = rec->sysno; for (i = 0; i < REC_NO_INFO; i++) if (!rec->info[i]) @@ -599,7 +876,7 @@ Record rec_cp (Record rec) else { n->size[i] = rec->size[i]; - n->info[i] = xmalloc (rec->size[i]); + n->info[i] = (char *) xmalloc (rec->size[i]); memcpy (n->info[i], rec->info[i], rec->size[i]); } return n; @@ -616,7 +893,7 @@ char *rec_strdup (const char *s, size_t *len) return NULL; } *len = strlen(s)+1; - p = xmalloc (*len); + p = (char *) xmalloc (*len); strcpy (p, s); return p; }