X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=index%2Frecindex.c;h=72e8259ce00f3ae3c4f97096a3857b8d89dc880e;hb=5adb31268b057741850de38eafd42baf8dc947ea;hp=17e6734e45f9c7a4db9315a25973b48a2e94607c;hpb=ef696645cc3b7e0f4027008d1dc589c0f0f90c1f;p=idzebra-moved-to-github.git diff --git a/index/recindex.c b/index/recindex.c index 17e6734..72e8259 100644 --- a/index/recindex.c +++ b/index/recindex.c @@ -4,7 +4,34 @@ * Sebastian Hammer, Adam Dickmeiss * * $Log: recindex.c,v $ - * Revision 1.23 1999-05-26 07:49:13 adam + * Revision 1.31 2001-02-26 22:14:59 adam + * Updated for BZIP2 1.0.X. Configure script doesn't enable 64 bit LFS + * on broken glibc on Redhat 7.0. + * + * Revision 1.30 2000/07/13 10:14:20 heikki + * Removed compiler warnings when making zebra + * + * Revision 1.29 2000/04/05 09:49:35 adam + * On Unix, zebra/z'mbol uses automake. + * + * Revision 1.28 1999/12/08 22:44:45 adam + * Zebra/Z'mbol dependencies added. + * + * 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 @@ -97,12 +124,32 @@ * 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 "recindxp.h" +#if HAVE_BZLIB_H +#include +#endif static void rec_write_head (Records p) { int r; @@ -118,13 +165,13 @@ 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 + 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_size = size + p->head.block_size[REC_BLOCK_TYPES-1]*2 + 2048; p->tmp_buf = (char *) xmalloc (p->tmp_size); } } @@ -155,12 +202,14 @@ 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 ; - p->head.total_bytes -= entry.size; + freeblock = entry.next; assert (freeblock > 0); dst_type = freeblock & 7; @@ -168,22 +217,43 @@ static void rec_release_blocks (Records p, int sysno) 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) @@ -198,40 +268,18 @@ 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, 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]; @@ -253,7 +301,11 @@ 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 { @@ -272,18 +324,14 @@ 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 (BFiles bfs, int rw) +Records rec_open (BFiles bfs, int rw, int compression_method) { Records p; int i, r; + int version; p = (Records) xmalloc (sizeof(*p)); + p->compression_method = compression_method; p->rw = rw; p->tmp_size = 1024; p->tmp_buf = (char *) xmalloc (p->tmp_size); @@ -299,6 +347,7 @@ 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; @@ -314,7 +363,7 @@ 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); @@ -323,9 +372,16 @@ 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); + 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; icache_max = 10; + p->cache_max = 400; p->cache_cur = 0; 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_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_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; - 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(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)); @@ -417,7 +634,7 @@ 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); @@ -434,6 +651,7 @@ void rec_close (Records *pp) assert (p); + zebra_mutex_destroy (&p->mutex); rec_cache_flush (p, 0); xfree (p->record_cache); @@ -454,14 +672,19 @@ void rec_close (Records *pp) *pp = NULL; } - -Record rec_get (Records p, int sysno) +static Record rec_get_int (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; +#if HAVE_BZLIB_H + int bz_size; +#endif + char compression_method; assert (sysno > 0); assert (p); @@ -469,7 +692,7 @@ 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) @@ -481,11 +704,12 @@ Record rec_get (Records p, int sysno) assert (freeblock > 0); - rec = (Record) 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) @@ -495,31 +719,112 @@ 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 (&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) { - memcpy (&rec->size[i], nptr, sizeof(*rec->size)); - nptr += sizeof(*rec->size); - if (rec->size[i]) - { - rec->info[i] = (char *) 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); + 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; } -Record rec_new (Records p) +Record rec_get (Records p, int 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; Record rec; @@ -536,6 +841,14 @@ Record rec_new (Records p) sysno = p->head.index_free; p->head.index_free = entry.next; } +#if ZMBOL +#else + if (sysno > 100000) + { + logf (LOG_FATAL, "100,000 record limit reached"); + exit (1); + } +#endif (p->head.no_records)++; rec->sysno = sysno; for (i = 0; i < REC_NO_INFO; i++) @@ -547,10 +860,21 @@ 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))) { @@ -562,6 +886,7 @@ void rec_del (Records p, Record *recpp) rec_cache_insert (p, *recpp, recordFlagDelete); rec_rm (recpp); } + zebra_mutex_unlock (&p->mutex); *recpp = NULL; } @@ -569,6 +894,7 @@ 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); @@ -579,6 +905,7 @@ void rec_put (Records p, Record *recpp) rec_cache_insert (p, *recpp, recordFlagWrite); rec_rm (recpp); } + zebra_mutex_unlock (&p->mutex); *recpp = NULL; }