X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=index%2Frecindex.c;h=81e390bce12fdb7352f1ce59d0d4d535c8dc062c;hb=c7700ec8d0d767670b6a3f3a860ba21d9dce8c28;hp=90efa6349b8667c1e068f83f82195d3c7bffd65f;hpb=e80772c51b80b8e5c9c3cfb037c988308a4014b9;p=idzebra-moved-to-github.git diff --git a/index/recindex.c b/index/recindex.c index 90efa63..81e390b 100644 --- a/index/recindex.c +++ b/index/recindex.c @@ -1,10 +1,82 @@ /* - * Copyright (C) 1994-1995, Index Data I/S + * Copyright (C) 1994-1999, Index Data * All rights reserved. * Sebastian Hammer, Adam Dickmeiss * * $Log: recindex.c,v $ - * Revision 1.9 1995-11-30 08:34:33 adam + * Revision 1.27 1999-10-29 10:02:33 adam + * Fixed decompression buffer overflow. + * + * Revision 1.26 1999/07/06 13:34:57 adam + * Fixed bug (introduced by previous commit). + * + * Revision 1.25 1999/07/06 12:28:04 adam + * Updated record index structure. Format includes version ID. Compression + * algorithm ID is stored for each record block. + * + * Revision 1.24 1999/06/25 13:48:02 adam + * Updated MSVC project files. + * Added BZIP2 record compression (not very well tested). + * + * Revision 1.23 1999/05/26 07:49:13 adam + * C++ compilation. + * + * Revision 1.22 1999/02/18 12:49:34 adam + * Changed file naming scheme for register files as well as record + * store/index files. + * + * Revision 1.21 1999/02/02 14:51:03 adam + * Updated WIN32 code specific sections. Changed header. + * + * 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. + * + * Revision 1.9 1995/11/30 08:34:33 adam * Started work on commit facility. * Changed a few malloc/free to xmalloc/xfree. * @@ -39,78 +111,32 @@ * Started work on better record management system. * */ + + +/* + * Format of first block + * next (4 bytes) + * ref_count (4 bytes) + * block (504 bytes) + * + * Format of subsequent blocks + * next (4 bytes) + * block (508 bytes) + * + * Format of each record + * sysno + * (length, data) - pairs + * length = 0 if same as previous + */ #include #include #include -#include -#include -#include -#include - -#include "recindex.h" - -#define USE_BF 1 - -#if USE_BF -#include - -#define REC_BLOCK_TYPES 2 -#define REC_HEAD_MAGIC "recindx" - -struct records_info { - int rw; - - char *index_fname; - BFile index_BFile; - - - char *data_fname[REC_BLOCK_TYPES]; - BFile data_BFile[REC_BLOCK_TYPES]; - - char *tmp_buf; - int tmp_size; - - struct record_cache_entry *record_cache; - int cache_size; - int cache_cur; - int cache_max; - - struct records_head { - char magic[8]; - int block_size[REC_BLOCK_TYPES]; - int block_free[REC_BLOCK_TYPES]; - int block_last[REC_BLOCK_TYPES]; - int block_used[REC_BLOCK_TYPES]; - int block_move[REC_BLOCK_TYPES]; - - int index_last; - int index_free; - int no_records; - - } head; -}; - -enum recordCacheFlag { recordFlagNop, recordFlagWrite, recordFlagNew, - recordFlagDelete }; - -struct record_cache_entry { - Record rec; - enum recordCacheFlag flag; -}; - -struct record_index_entry { - union { - struct { - int next; - int size; - } used; - struct { - int next; - } free; - } u; -}; +#include "recindxp.h" +#if HAVE_BZLIB_H +#include +#endif static void rec_write_head (Records p) { int r; @@ -126,14 +152,14 @@ static void rec_write_head (Records p) } } -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 + 256 || - p->tmp_size < p->head.block_size[dst_type]*2) + if (p->tmp_size < size + 2048 || + 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 + 256; - p->tmp_buf = xmalloc (p->tmp_size); + p->tmp_size = size + p->head.block_size[REC_BLOCK_TYPES-1]*2 + 2048; + p->tmp_buf = (char *) xmalloc (p->tmp_size); } } @@ -148,7 +174,6 @@ static int read_indx (Records p, int sysno, void *buf, int itemsize, { logf (LOG_FATAL|LOG_ERRNO, "read in %s at pos %ld", p->index_fname, (long) pos); - abort (); exit (1); } return r; @@ -164,34 +189,58 @@ static void write_indx (Records p, int sysno, void *buf, int itemsize) static void rec_release_blocks (Records p, int sysno) { struct record_index_entry entry; - int freeblock, freenext; + int freeblock; + char block_and_ref[sizeof(short) + sizeof(int)]; int dst_type; + int first = 1; if (read_indx (p, sysno, &entry, sizeof(entry), 1) != 1) return ; - freeblock = entry.u.used.next; + + freeblock = entry.next; assert (freeblock > 0); dst_type = 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, + sizeof(block_and_ref), block_and_ref) != 1) { logf (LOG_FATAL|LOG_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(int), sizeof(ref)); + --ref; + memcpy (block_and_ref + sizeof(int), &ref, sizeof(ref)); + if (ref) + { + if (bf_write (p->data_BFile[dst_type], freeblock, 0, + sizeof(block_and_ref), block_and_ref)) + { + logf (LOG_FATAL|LOG_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"); exit (1); } p->head.block_free[dst_type] = freeblock; - freeblock = freenext; + memcpy (&freeblock, block_and_ref, sizeof(int)); + p->head.block_used[dst_type]--; } + p->head.total_bytes -= entry.size; } static void rec_delete_single (Records p, Record rec) @@ -200,45 +249,24 @@ static void rec_delete_single (Records p, Record rec) rec_release_blocks (p, rec->sysno); - entry.u.free.next = p->head.index_free; + entry.next = p->head.index_free; + entry.size = 0; p->head.index_free = rec->sysno; 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, int *sysnos) { - int i, size = 0; - char *cptr; - int dst_type = 0; + struct record_index_entry entry; int no_written = 0; + char *cptr = p->tmp_buf; 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 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]; @@ -250,16 +278,21 @@ static void rec_write_single (Records p, Record rec) { logf (LOG_FATAL|LOG_ERRNO, "read in %s at free block %d", p->data_fname[dst_type], block_free); + exit (1); } } else block_free = p->head.block_last[dst_type]++; if (block_prev == -1) { - entry.u.used.next = block_free*8 + dst_type; - entry.u.used.size = size; - - write_indx (p, rec->sysno, &entry, sizeof(entry)); + entry.next = block_free*8 + dst_type; + entry.size = size; + p->head.total_bytes += size; + while (*sysnos > 0) + { + write_indx (p, *sysnos, &entry, sizeof(entry)); + sysnos++; + } } else { @@ -278,23 +311,19 @@ static void rec_write_single (Records p, Record rec) sizeof(int) + (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 (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 (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, 128, rw); if (p->index_BFile == NULL) { logf (LOG_FATAL|LOG_ERRNO, "open %s", p->index_fname); @@ -305,9 +334,11 @@ Records rec_open (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; + p->head.total_bytes = 0; for (i = 0; ihead.block_free[i] = 0; @@ -319,7 +350,7 @@ Records rec_open (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); @@ -328,22 +359,29 @@ Records rec_open (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); + logf (LOG_FATAL, "file %s has bad format", p->index_fname); exit (1); } + version = atoi (p->head.version); + if (version != REC_VERSION) + { + logf (LOG_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 (p->data_fname[i], + if (!(p->data_BFile[i] = bf_open (bfs, p->data_fname[i], p->head.block_size[i], rw))) { @@ -351,35 +389,197 @@ Records rec_open (int rw) 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); return p; } -static void rec_cache_flush (Records p) +static void rec_encode_unsigned (unsigned n, unsigned char *buf, int *len) +{ + (*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_cache_flush_block1 (Records p, Record rec, Record last_rec, + char **out_buf, int *out_size, + int *out_offset) { int i; - for (i = 0; icache_cur; 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_unsigned (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); + int *sysnos = (int *) xmalloc (sizeof(*sysnos) * (p->cache_cur + 1)); + int *sysnop = sysnos; + + 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); } - p->cache_cur = 0; + + *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); + i = bzBuffToBuffCompress (p->tmp_buf+sizeof(int)+sizeof(short)+ + sizeof(char), + &csize, out_buf, out_offset, 1, 0, 30); + if (i != BZ_OK) + { + logf (LOG_WARN, "bzBuffToBuffCompress error code=%d", i); + csize = 0; + } + logf (LOG_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(int) + sizeof(short) + sizeof(char), + out_buf, out_offset); + csize = out_offset; + compression_method = REC_COMPRESS_NONE; + } + memcpy (p->tmp_buf + sizeof(int), &ref_count, sizeof(ref_count)); + memcpy (p->tmp_buf + sizeof(int)+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, @@ -404,7 +604,20 @@ static void rec_cache_insert (Records p, Record rec, enum recordCacheFlag flag) struct record_cache_entry *e; if (p->cache_cur == p->cache_max) - rec_cache_flush (p); + rec_cache_flush (p, 1); + else if (p->cache_cur > 0) + { + int i, j; + int used = 0; + for (i = 0; icache_cur; i++) + { + Record r = (p->record_cache + i)->rec; + for (j = 0; jsize[j]; + } + if (used > 90000) + rec_cache_flush (p, 1); + } assert (p->cache_cur < p->cache_max); e = p->record_cache + (p->cache_cur)++; @@ -419,7 +632,7 @@ void rec_close (Records *pp) assert (p); - rec_cache_flush (p); + rec_cache_flush (p, 0); xfree (p->record_cache); if (p->rw) @@ -442,11 +655,15 @@ void rec_close (Records *pp) Record rec_get (Records p, int sysno) { - int i; + int i, in_size, r; Record rec, *recp; struct record_index_entry entry; int freeblock, dst_type; char *nptr, *cptr; + char *in_buf = 0; + char *bz_buf = 0; + int bz_size; + char compression_method; assert (sysno > 0); assert (p); @@ -454,19 +671,24 @@ Record rec_get (Records p, int sysno) if ((recp = rec_cache_lookup (p, sysno, recordFlagNop))) return rec_cp (*recp); - read_indx (p, sysno, &entry, sizeof(entry), 0); + if (read_indx (p, sysno, &entry, sizeof(entry), 1) < 1) + return NULL; /* record is not there! */ - dst_type = entry.u.used.next & 7; + if (!entry.size) + return NULL; /* record is deleted */ + + dst_type = entry.next & 7; assert (dst_type < REC_BLOCK_TYPES); - freeblock = entry.u.used.next / 8; + freeblock = entry.next / 8; assert (freeblock > 0); - rec = xmalloc (sizeof(*rec)); - rec_tmp_expand (p, entry.u.used.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) @@ -476,26 +698,92 @@ Record rec_get (Records p, int sysno) 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 (&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; - } + memcpy (&compression_method, p->tmp_buf + sizeof(int) + sizeof(short), + sizeof(compression_method)); + in_buf = p->tmp_buf + sizeof(int) + sizeof(short) + sizeof(char); + in_size = entry.size - sizeof(short) - sizeof(char); + switch (compression_method) + { + case REC_COMPRESS_BZIP2: +#if HAVE_BZLIB_H + bz_size = entry.size * 20 + 100; + while (1) + { + bz_buf = (char *) xmalloc (bz_size); + i = bzBuffToBuffDecompress (bz_buf, &bz_size, in_buf, in_size, 0, 0); + logf (LOG_LOG, "decompress %5d %5d", in_size, bz_size); + if (i == BZ_OK) + break; + logf (LOG_LOG, "failed"); + xfree (bz_buf); + bz_size *= 2; + } + in_buf = bz_buf; + in_size = bz_size; +#else + logf (LOG_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) + { + int this_sysno; + int len; + rec_decode_unsigned (&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]); + memcpy (np, rec->info[i], rec->size[i]); + 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; } @@ -506,8 +794,8 @@ Record rec_new (Records p) Record rec; assert (p); - rec = xmalloc (sizeof(*rec)); - if (p->head.index_free == 0) + rec = (Record) xmalloc (sizeof(*rec)); + if (1 || p->head.index_free == 0) sysno = (p->head.index_last)++; else { @@ -515,7 +803,7 @@ Record rec_new (Records p) read_indx (p, p->head.index_free, &entry, sizeof(entry), 0); sysno = p->head.index_free; - p->head.index_free = entry.u.free.next; + p->head.index_free = entry.next; } (p->head.no_records)++; rec->sysno = sysno; @@ -532,6 +820,7 @@ void rec_del (Records p, Record *recpp) { Record *recp; + (p->head.no_records)--; if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagDelete))) { rec_rm (recp); @@ -565,428 +854,9 @@ void rec_put (Records p, Record *recpp) void rec_rm (Record *recpp) { int i; - for (i = 0; i < REC_NO_INFO; i++) - xfree ((*recpp)->info[i]); - xfree (*recpp); - *recpp = NULL; -} - -Record rec_cp (Record rec) -{ - Record n; - int i; - - n = xmalloc (sizeof(*n)); - n->sysno = rec->sysno; - for (i = 0; i < REC_NO_INFO; i++) - if (!rec->info[i]) - { - n->info[i] = NULL; - n->size[i] = 0; - } - else - { - n->size[i] = rec->size[i]; - n->info[i] = xmalloc (rec->size[i]); - memcpy (n->info[i], rec->info[i], rec->size[i]); - } - return n; -} - -/* no BF --------------------------------------------------- */ -#else - -struct records_info { - int rw; - int index_fd; - char *index_fname; - int data_fd; - char *data_fname; - struct records_head { - char magic[8]; - int no_records; - int index_free; - int index_last; - int data_size; - int data_slack; - int data_used; - } head; - char *tmp_buf; - int tmp_size; - int cache_size; - int cache_cur; - int cache_max; - struct record_cache_entry *record_cache; -}; - -struct record_cache_entry { - Record rec; - int dirty; -}; - -struct record_index_entry { - union { - struct { - int offset; - int size; - } used; - struct { - int next; - } free; - } u; -}; - -#define REC_HEAD_MAGIC "rechead" - -static void rec_write_head (Records p) -{ - int r; - - assert (p); - assert (p->index_fd != -1); - if (lseek (p->index_fd, (off_t) 0, SEEK_SET) == -1) - { - logf (LOG_FATAL|LOG_ERRNO, "lseek to 0 in %s", p->index_fname); - exit (1); - } - r = write (p->index_fd, &p->head, sizeof(p->head)); - switch (r) - { - case -1: - logf (LOG_FATAL|LOG_ERRNO, "write head of %s", p->index_fname); - exit (1); - case sizeof(p->head): - break; - default: - logf (LOG_FATAL, "write head of %s. wrote %d", p->index_fname, r); - exit (1); - } -} - -Records rec_open (int rw) -{ - Records p; - int r; - - p = xmalloc (sizeof(*p)); - p->rw = rw; - p->tmp_buf = NULL; - p->tmp_size = 0; - p->data_fname = "recdata"; - p->data_fd = -1; - p->index_fname = "recindex"; - p->index_fd = open (p->index_fname, - rw ? (O_RDWR|O_CREAT) : O_RDONLY, 0666); - if (p->index_fd == -1) - { - logf (LOG_FATAL|LOG_ERRNO, "open %s", p->index_fname); - exit (1); - } - r = read (p->index_fd, &p->head, sizeof(p->head)); - switch (r) - { - case -1: - logf (LOG_FATAL|LOG_ERRNO, "read %s", p->index_fname); - exit (1); - case 0: - memcpy (p->head.magic, REC_HEAD_MAGIC, sizeof(p->head.magic)); - p->head.index_free = 0; - p->head.index_last = 1; - p->head.no_records = 0; - p->head.data_size = 0; - p->head.data_slack = 0; - p->head.data_used = 0; - if (rw) - rec_write_head (p); - break; - case 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); - exit (1); - } - break; - default: - logf (LOG_FATAL, "read head of %s. expected %d. got %d", - p->index_fname, sizeof(p->head), r); - exit (1); - } - p->data_fd = open (p->data_fname, - rw ? (O_RDWR|O_CREAT) : O_RDONLY, 0666); - if (p->data_fd == -1) - { - logf (LOG_FATAL|LOG_ERRNO, "open %s", p->data_fname); - exit (1); - } - p->cache_max = 10; - p->cache_cur = 0; - p->record_cache = xmalloc (sizeof(*p->record_cache)*p->cache_max)); - return p; -} - -static void read_indx (Records p, int sysno, void *buf, int itemsize) -{ - int r; - off_t pos = (sysno-1)*itemsize + sizeof(p->head); - - if (lseek (p->index_fd, pos, SEEK_SET) == (pos) -1) - { - logf (LOG_FATAL|LOG_ERRNO, "seek in %s to pos %ld", - p->index_fname, (long) pos); - exit (1); - } - r = read (p->index_fd, buf, itemsize); - if (r != itemsize) - { - if (r == -1) - logf (LOG_FATAL|LOG_ERRNO, "read in %s at pos %ld", - p->index_fname, (long) pos); - else - logf (LOG_FATAL, "read in %s at pos %ld", - p->index_fname, (long) pos); - exit (1); - } -} - -static void rec_write_single (Records p, Record rec) -{ - struct record_index_entry entry; - int r, i, size = 0, got; - char *cptr; - off_t pos = (rec->sysno-1)*sizeof(entry) + sizeof(p->head); - - for (i = 0; i < REC_NO_INFO; i++) - if (!rec->info[i]) - size += sizeof(*rec->size); - else - size += sizeof(*rec->size) + rec->size[i]; - - entry.u.used.offset = p->head.data_size; - entry.u.used.size = size; - p->head.data_size += size; - p->head.data_used += size; - - if (lseek (p->index_fd, pos, SEEK_SET) == (pos) -1) - { - logf (LOG_FATAL|LOG_ERRNO, "seek in %s to pos %ld", - p->index_fname, (long) pos); - exit (1); - } - r = write (p->index_fd, &entry, sizeof(entry)); - if (r != sizeof(entry)) - { - if (r == -1) - logf (LOG_FATAL|LOG_ERRNO, "write of %s at pos %ld", - p->index_fname, (long) pos); - else - logf (LOG_FATAL, "write of %s at pos %ld", - p->index_fname, (long) pos); - exit (1); - } - if (lseek (p->data_fd, entry.u.used.offset, SEEK_SET) == -1) - { - logf (LOG_FATAL|LOG_ERRNO, "lseek in %s to pos %ld", - p->data_fname, entry.u.used.offset); - exit (1); - } - if (p->tmp_size < entry.u.used.size) - { - xfree (p->tmp_buf); - p->tmp_size = entry.u.used.size + 16384; - p->tmp_buf = xmalloc (p->tmp_size)); - } - cptr = p->tmp_buf; - 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]; - } - } - for (got = 0; got < entry.u.used.size; got += r) - { - r = write (p->data_fd, p->tmp_buf + got, entry.u.used.size - got); - if (r <= 0) - { - logf (LOG_FATAL|LOG_ERRNO, "write of %s", p->data_fname); - exit (1); - } - } -} - -static void rec_cache_flush (Records p) -{ - int i; - for (i = 0; icache_cur; i++) - { - struct record_cache_entry *e = p->record_cache + i; - if (e->dirty) - rec_write_single (p, e->rec); - rec_rm (&e->rec); - } - p->cache_cur = 0; -} - -static Record *rec_cache_lookup (Records p, int sysno, int dirty) -{ - int i; - for (i = 0; icache_cur; i++) - { - struct record_cache_entry *e = p->record_cache + i; - if (e->rec->sysno == sysno) - { - if (dirty) - e->dirty = 1; - return &e->rec; - } - } - return NULL; -} - -static void rec_cache_insert (Records p, Record rec, int dirty) -{ - struct record_cache_entry *e; - - if (p->cache_cur == p->cache_max) - rec_cache_flush (p); - assert (p->cache_cur < p->cache_max); - - e = p->record_cache + (p->cache_cur)++; - e->dirty = dirty; - e->rec = rec_cp (rec); -} - -void rec_close (Records *p) -{ - assert (*p); - - rec_cache_flush (*p); - xfree ((*p)->record_cache); - - if ((*p)->rw) - rec_write_head (*p); - - if ((*p)->index_fd != -1) - close ((*p)->index_fd); - - if ((*p)->data_fd != -1) - close ((*p)->data_fd); - - xfree ((*p)->tmp_buf); - - xfree (*p); - *p = NULL; -} - -Record rec_get (Records p, int sysno) -{ - int i; - Record rec, *recp; - struct record_index_entry entry; - int r, got; - char *nptr; - - assert (sysno > 0); - assert (p); - - if ((recp = rec_cache_lookup (p, sysno, 0))) - return rec_cp (*recp); - read_indx (p, sysno, &entry, sizeof(entry)); - - rec = xmalloc (sizeof(*rec)); - if (lseek (p->data_fd, entry.u.used.offset, SEEK_SET) == -1) - { - logf (LOG_FATAL|LOG_ERRNO, "lseek in %s to pos %ld", - p->data_fname, entry.u.used.offset); - exit (1); - } - if (p->tmp_size < entry.u.used.size) - { - xfree (p->tmp_buf); - p->tmp_size = entry.u.used.size + 16384; - p->tmp_buf = xmalloc (p->tmp_size)); - } - for (got = 0; got < entry.u.used.size; got += r) - { - r = read (p->data_fd, p->tmp_buf + got, entry.u.used.size - got); - if (r <= 0) - { - logf (LOG_FATAL|LOG_ERRNO, "read of %s", p->data_fname); - exit (1); - } - } - rec->sysno = sysno; - - nptr = p->tmp_buf; - for (i = 0; i < REC_NO_INFO; i++) - { - 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; - } - rec_cache_insert (p, rec, 0); - return rec; -} - -Record rec_new (Records p) -{ - int sysno, i; - Record rec; - - assert (p); - rec = xmalloc (sizeof(*rec)); - if (p->head.index_free == 0) - sysno = (p->head.index_last)++; - else - { - struct record_index_entry entry; - - read_indx (p, p->head.index_free, &entry, sizeof(entry)); - sysno = p->head.index_free; - p->head.index_free = entry.u.free.next; - } - (p->head.no_records)++; - rec->sysno = sysno; - for (i = 0; i < REC_NO_INFO; i++) - { - rec->info[i] = NULL; - rec->size[i] = 0; - } - rec_cache_insert (p, rec, 1); - return rec; -} - -void rec_put (Records p, Record *recpp) -{ - Record *recp; - - if ((recp = rec_cache_lookup (p, (*recpp)->sysno, 1))) - { - rec_rm (recp); - *recp = *recpp; - } - else - { - rec_cache_insert (p, *recpp, 1); - rec_rm (recpp); - } - *recpp = NULL; -} - -void rec_rm (Record *recpp) -{ - int i; + if (!*recpp) + return ; for (i = 0; i < REC_NO_INFO; i++) xfree ((*recpp)->info[i]); xfree (*recpp); @@ -998,7 +868,7 @@ 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]) @@ -1009,19 +879,12 @@ 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; } -void rec_del (Records p, Record *recpp) -{ - assert (0); -} - - -#endif char *rec_strdup (const char *s, size_t *len) { @@ -1033,7 +896,7 @@ 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; }