X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=index%2Frecindex.c;h=2fe16b8cdeb9e872cf4248a03f1ad2ba211feabf;hb=5ed7bb5a4a7e46271cc161304b114eae54b53674;hp=b8ccae7d80494a4708a0f5521fb0411cec49c953;hpb=f3e73ab63fbc960d863d9c14bab3b9e79c400ffa;p=idzebra-moved-to-github.git diff --git a/index/recindex.c b/index/recindex.c index b8ccae7..2fe16b8 100644 --- a/index/recindex.c +++ b/index/recindex.c @@ -1,287 +1,591 @@ /* - * 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 1.2 1995-11-15 19:13:08 adam + * 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. + * + * 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 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" - -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" - -char *rec_strdup (const char *s) + +#include "recindxp.h" + +#if HAVE_BZLIB_H +#include +#endif +static void rec_write_head (Records p) { - char *p; + int r; - if (!s) - return NULL; - p = malloc (strlen(s)+1); - if (!p) + assert (p); + assert (p->index_BFile); + + r = bf_write (p->index_BFile, 0, 0, sizeof(p->head), &p->head); + if (r) { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); + logf (LOG_FATAL|LOG_ERRNO, "write head of %s", p->index_fname); exit (1); } - strcpy (p, s); - return p; } -static void rec_write_head (Records p) +static void rec_tmp_expand (Records p, int size) +{ + 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[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, + int ignoreError) { int r; + int pos = (sysno-1)*itemsize; - assert (p); - assert (p->index_fd != -1); - if (lseek (p->index_fd, (off_t) 0, SEEK_SET) == -1) + r = bf_read (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); + if (r != 1 && !ignoreError) { - logf (LOG_FATAL|LOG_ERRNO, "lseek to 0 in %s", p->index_fname); + logf (LOG_FATAL|LOG_ERRNO, "read in %s at pos %ld", + p->index_fname, (long) pos); exit (1); } - r = write (p->index_fd, &p->head, sizeof(p->head)); - switch (r) + return r; +} + +static void write_indx (Records p, int sysno, void *buf, int itemsize) +{ + int pos = (sysno-1)*itemsize; + + bf_write (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); +} + +static void rec_release_blocks (Records p, int sysno) +{ + struct record_index_entry entry; + int freeblock; + int block_and_ref[2]; + int dst_type; + int first = 1; + + logf (LOG_LOG, "release_blocks for sysno=%d", sysno); + if (read_indx (p, sysno, &entry, sizeof(entry), 1) != 1) + return ; + + freeblock = entry.next; + assert (freeblock > 0); + dst_type = freeblock & 7; + assert (dst_type < REC_BLOCK_TYPES); + freeblock = freeblock / 8; + while (freeblock) { - 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); + 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 (first) + { + block_and_ref[1]--; + if (block_and_ref[1]) + { + 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 = block_and_ref[0]; + + p->head.block_used[dst_type]--; } + p->head.total_bytes -= entry.size; } -Records rec_open (int rw) +static void rec_delete_single (Records p, Record rec) +{ + struct record_index_entry entry; + + rec_release_blocks (p, rec->sysno); + + 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_tmp_buf (Records p, int size, int *sysnos); + +static void rec_write_single (Records p, Record rec) { - Records p; - int r; - if (!(p = malloc (sizeof(*p)))) + int sysnos[2]; + int i, size = 0; + char *cptr; + + logf (LOG_LOG, " rec_write_single !!!!!!!!!!!!!!!!!!!!!!!!!!!!"); + for (i = 0; i < REC_NO_INFO; i++) + if (!rec->info[i]) + size += sizeof(*rec->size); + else + size += sizeof(*rec->size) + rec->size[i]; + + rec_tmp_expand (p, size); + + cptr = p->tmp_buf + sizeof(int); /* a hack! */ + for (i = 0; i < REC_NO_INFO; i++) { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); + 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]; + } } + sysnos[0] = rec->sysno; + sysnos[1] = -1; + rec_write_tmp_buf (p, size, sysnos); +} + +static void rec_write_tmp_buf (Records p, int size, int *sysnos) +{ + struct record_index_entry entry; + int no_written = 0; + char *cptr = p->tmp_buf; + int block_prev = -1, block_free; + int dst_type = 0; + int i; + + for (i = 1; i= p->head.block_move[i]) + dst_type = i; + while (no_written < size) + { + block_free = p->head.block_free[dst_type]; + if (block_free) + { + if (bf_read (p->data_BFile[dst_type], + 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", + p->data_fname[dst_type], block_free); + exit (1); + } + } + else + block_free = p->head.block_last[dst_type]++; + if (block_prev == -1) + { + 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 + { + memcpy (cptr, &block_free, sizeof(int)); + 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); + p->head.block_used[dst_type]++; + } + assert (block_prev != -1); + block_free = 0; + memcpy (cptr, &block_free, sizeof(int)); + bf_write (p->data_BFile[dst_type], block_prev, 0, + 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 p; + int i, r; + + p = (Records) 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) + p->tmp_size = 1024; + 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); exit (1); } - r = read (p->index_fd, &p->head, sizeof(p->head)); + r = bf_read (p->index_BFile, 0, 0, 0, p->tmp_buf); 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; + p->head.total_bytes = 0; + for (i = 0; ihead.block_free[i] = 0; + p->head.block_last[i] = 1; + p->head.block_used[i] = 0; + } + p->head.block_size[0] = 128; + p->head.block_move[0] = 0; + 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; + } if (rw) rec_write_head (p); break; - case sizeof(p->head): + case 1: + 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); 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) + for (i = 0; idata_fname); - exit (1); + char str[80]; + 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; } - p->cache_max = 100; - p->cache_cur = 0; - if (!(p->record_cache = malloc (sizeof(*p->record_cache)*p->cache_max))) + for (i = 0; idata_BFile[i] = bf_open (bfs, p->data_fname[i], + p->head.block_size[i], + rw))) + { + logf (LOG_FATAL|LOG_ERRNO, "bf_open %s", p->data_fname[i]); + exit (1); + } } + p->cache_max = 400; + p->cache_cur = 0; + p->record_cache = (struct record_cache_entry *) + xmalloc (sizeof(*p->record_cache)*p->cache_max); return p; } -static void read_indx (Records p, int sysno, void *buf, int itemsize) +static void rec_encode_unsigned (unsigned n, unsigned char *buf, int *len) { - 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) + (*len) = 0; + while (n > 127) { - 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); + buf[*len] = 128 + (n & 127); + n = n >> 7; + (*len)++; } + buf[*len] = n; + (*len)++; } -static void rec_write_single (Records p, Record rec) +static void rec_decode_unsigned(unsigned *np, unsigned char *buf, int *len) { - 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++; - else - size += strlen(rec->info[i])+1; - - entry.u.used.offset = p->head.data_size; - entry.u.used.size = size; - p->head.data_size += size; - p->head.data_used += size; + unsigned n = 0; + unsigned w = 1; + (*len) = 0; - 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)) + while (buf[*len] > 127) { - 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); + n += w*(buf[*len] & 127); + w = w << 7; + (*len)++; } - if (lseek (p->data_fd, entry.u.used.offset, SEEK_SET) == -1) + 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; idata_fname, entry.u.used.offset); - exit (1); + if (*out_offset + (int) rec->size[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]; + } } - if (p->tmp_size < entry.u.used.size) +} + +static void rec_write_multiple (Records p, int saveCount) +{ + int i; + int ref_count = 0; + 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++) { - free (p->tmp_buf); - p->tmp_size = entry.u.used.size + 16384; - if (!(p->tmp_buf = malloc (p->tmp_size))) + struct record_cache_entry *e = p->record_cache + i; + switch (e->flag) { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); + case recordFlagNew: + 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_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; + default: + break; } } - cptr = p->tmp_buf; - for (i = 0; i < REC_NO_INFO; i++) - if (!rec->info[i]) - *cptr++ = '\0'; - else - { - strcpy (cptr, rec->info[i]); - cptr += strlen(rec->info[i]) + 1; - } - for (got = 0; got < entry.u.used.size; got += r) + *sysnop = -1; + if (ref_count) { - 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); - } - got += r; + int csize = out_offset + (out_offset >> 6) + 620; + + rec_tmp_expand (p, csize); +#if HAVE_BZLIB_H + i = bzBuffToBuffCompress (p->tmp_buf+2*sizeof(int), &csize, + out_buf, out_offset, 9, 0, 30); + if (i != BZ_OK) + { + logf (LOG_FATAL, "bzBuffToCompress error code=%d", i); + exit (1); + } +#else + memcpy (p->tmp_buf + 2*sizeof(int), out_buf, out_offset); + csize = out_offset; +#endif + memcpy (p->tmp_buf + sizeof(int), &ref_count, sizeof(ref_count)); + /* -------- compression */ + rec_write_tmp_buf (p, csize + sizeof(int), sysnos); } + xfree (out_buf); + xfree (sysnos); } -static void rec_cache_flush (Records p) +static void rec_cache_flush (Records p, int saveCount) { - int i; - for (i = 0; icache_cur; i++) + 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; - if (e->dirty) + switch (e->flag) + { + case recordFlagNop: + break; + case recordFlagNew: rec_write_single (p, e->rec); - rec_rm (e->rec); + break; + case recordFlagWrite: + rec_update_single (p, e->rec); + break; + case recordFlagDelete: + rec_delete_single (p, e->rec); + break; + } + rec_rm (&e->rec); } - p->cache_cur = 0; + 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, int dirty) +static Record *rec_cache_lookup (Records p, int sysno, + enum recordCacheFlag flag) { int i; for (i = 0; icache_cur; i++) @@ -289,108 +593,161 @@ static Record *rec_cache_lookup (Records p, int sysno, int dirty) struct record_cache_entry *e = p->record_cache + i; if (e->rec->sysno == sysno) { - if (dirty) - e->dirty = 1; + if (flag != recordFlagNop && e->flag == recordFlagNop) + e->flag = flag; return &e->rec; } } return NULL; } -static void rec_cache_insert (Records p, Record rec, int dirty) +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 > 256000) + rec_cache_flush (p, 1); + } assert (p->cache_cur < p->cache_max); e = p->record_cache + (p->cache_cur)++; - e->dirty = 1; + e->flag = flag; e->rec = rec_cp (rec); } -void rec_close (Records *p) +void rec_close (Records *pp) { - assert (*p); + Records p = *pp; + int i; - rec_cache_flush (*p); - free ((*p)->record_cache); + assert (p); - if ((*p)->index_fd != -1) - close ((*p)->index_fd); + rec_cache_flush (p, 0); + xfree (p->record_cache); - if ((*p)->data_fd != -1) - close ((*p)->data_fd); + if (p->rw) + rec_write_head (p); - free ((*p)->tmp_buf); + if (p->index_BFile) + bf_close (p->index_BFile); - free (*p); - *p = NULL; + for (i = 0; idata_BFile[i]) + bf_close (p->data_BFile[i]); + xfree (p->data_fname[i]); + } + xfree (p->tmp_buf); + xfree (p); + *pp = NULL; } + Record rec_get (Records p, int sysno) { - int i; + int i, in_size; Record rec, *recp; struct record_index_entry entry; - int r, got; - char *nptr; + int freeblock, dst_type; + char *nptr, *cptr; + char *in_buf = 0; assert (sysno > 0); assert (p); - if ((recp = rec_cache_lookup (p, sysno, 0))) + if ((recp = rec_cache_lookup (p, sysno, recordFlagNop))) return rec_cp (*recp); - read_indx (p, sysno, &entry, sizeof(entry)); + if (!read_indx (p, sysno, &entry, sizeof(entry), 1)) + return NULL; /* record is not there! */ + + if (!entry.size) + return NULL; /* record is deleted */ + + dst_type = entry.next & 7; + assert (dst_type < REC_BLOCK_TYPES); + freeblock = entry.next / 8; + + assert (freeblock > 0); - if (!(rec = malloc (sizeof(*rec)))) - { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); - } - if (lseek (p->data_fd, entry.u.used.offset, SEEK_SET) == -1) + rec = (Record) xmalloc (sizeof(*rec)); + rec_tmp_expand (p, entry.size); + + cptr = p->tmp_buf; + bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + memcpy (&freeblock, cptr, sizeof(freeblock)); + + while (freeblock) { - logf (LOG_FATAL|LOG_ERRNO, "lseek in %s to pos %ld", - p->data_fname, entry.u.used.offset); - exit (1); + int 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); + memcpy (&freeblock, cptr, sizeof(freeblock)); + memcpy (cptr, &tmp, sizeof(tmp)); } - if (p->tmp_size < entry.u.used.size) + + rec->sysno = sysno; +#if HAVE_BZLIB_H + in_size = entry.size * 30+100; + in_buf = (char *) xmalloc (in_size); + i = bzBuffToBuffDecompress (in_buf, &in_size, p->tmp_buf+2*sizeof(int), + entry.size-sizeof(int), 0, 4); + if (i != BZ_OK) { - free (p->tmp_buf); - p->tmp_size = entry.u.used.size + 16384; - if (!(p->tmp_buf = malloc (p->tmp_size))) - { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); - } + logf (LOG_FATAL, "bzBuffToDecompress error code=%d", i); + exit (1); } - for (got = 0; got < entry.u.used.size; got += r) +#else + in_buf = p->tmp_buf + 2*sizeof(int); + in_size = entry.size - sizeof(int); +#endif + nptr = in_buf; /* skip ref count */ + while (nptr < in_buf + in_size) { - 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); - } - got += r; + 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] = (char *) xmalloc (rec->size[i]); + memcpy (rec->info[i], nptr, rec->size[i]); + nptr += rec->size[i]; + } + else + rec->info[i] = NULL; + } + if (this_sysno == sysno) + break; } - rec->sysno = sysno; - - nptr = p->tmp_buf; - for (i = 0; i < REC_NO_INFO; i++) - if (*nptr) - { - rec->info[i] = rec_strdup (nptr); - nptr += strlen(nptr)+1; - } - else - { - nptr++; - rec->info[i] = NULL; - } - rec_cache_insert (p, rec, 0); + xfree (in_buf); + rec_cache_insert (p, rec, recordFlagNop); return rec; } @@ -400,48 +757,73 @@ Record rec_new (Records p) Record rec; assert (p); - if (!(rec = malloc (sizeof(*rec)))) - { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); - } - if (p->head.index_free == 0) + rec = (Record) xmalloc (sizeof(*rec)); + if (1 || 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)); + 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; for (i = 0; i < REC_NO_INFO; i++) + { rec->info[i] = NULL; - rec_cache_insert (p, rec, 1); + rec->size[i] = 0; + } + rec_cache_insert (p, rec, recordFlagNew); return rec; } -void rec_put (Records p, Record rec) +void rec_del (Records p, Record *recpp) { Record *recp; - if ((recp = rec_cache_lookup (p, rec->sysno, 1))) + (p->head.no_records)--; + if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagDelete))) { - rec_rm (*recp); - *recp = rec_cp (rec); + rec_rm (recp); + *recp = *recpp; } else - rec_cache_insert (p, rec, 1); + { + rec_cache_insert (p, *recpp, recordFlagDelete); + rec_rm (recpp); + } + *recpp = NULL; } -void rec_rm (Record rec) +void rec_put (Records p, Record *recpp) +{ + Record *recp; + + if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagWrite))) + { + rec_rm (recp); + *recp = *recpp; + } + else + { + rec_cache_insert (p, *recpp, recordFlagWrite); + rec_rm (recpp); + } + *recpp = NULL; +} + +void rec_rm (Record *recpp) { int i; + + if (!*recpp) + return ; for (i = 0; i < REC_NO_INFO; i++) - free (rec->info[i]); - free (rec); + xfree ((*recpp)->info[i]); + xfree (*recpp); + *recpp = NULL; } Record rec_cp (Record rec) @@ -449,13 +831,36 @@ Record rec_cp (Record rec) Record n; int i; - if (!(n = malloc (sizeof(*n)))) - { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); - exit (1); - } + n = (Record) xmalloc (sizeof(*n)); n->sysno = rec->sysno; for (i = 0; i < REC_NO_INFO; i++) - n->info[i] = rec_strdup (rec->info[i]); + if (!rec->info[i]) + { + n->info[i] = NULL; + n->size[i] = 0; + } + else + { + n->size[i] = rec->size[i]; + n->info[i] = (char *) xmalloc (rec->size[i]); + memcpy (n->info[i], rec->info[i], rec->size[i]); + } return n; } + + +char *rec_strdup (const char *s, size_t *len) +{ + char *p; + + if (!s) + { + *len = 0; + return NULL; + } + *len = strlen(s)+1; + p = (char *) xmalloc (*len); + strcpy (p, s); + return p; +} +