1 /*- 2 * Copyright (c) 1990 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Mike Olson. 7 * 8 * %sccs.include.redist.c% 9 */ 10 11 #if defined(LIBC_SCCS) && !defined(lint) 12 static char sccsid[] = "@(#)bt_put.c 5.6 (Berkeley) 12/11/91"; 13 #endif /* LIBC_SCCS and not lint */ 14 15 #include <sys/types.h> 16 #include <errno.h> 17 #include <db.h> 18 #include <stdio.h> 19 #include <stdlib.h> 20 #include <string.h> 21 #include "btree.h" 22 23 static EPG *bt_fast __P((BTREE *, const DBT *, const DBT *, int *)); 24 25 /* 26 * __BT_PUT -- Add a btree item to the tree. 27 * 28 * Parameters: 29 * dbp: pointer to access method 30 * key: key 31 * data: data 32 * flag: R_NOOVERWRITE 33 * 34 * Returns: 35 * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the 36 * tree and R_NOOVERWRITE specified. 37 */ 38 int 39 __bt_put(dbp, key, data, flags) 40 const DB *dbp; 41 const DBT *key, *data; 42 u_int flags; 43 { 44 BTREE *t; 45 DBT tkey, tdata; 46 EPG *e; 47 PAGE *h; 48 index_t index, nxtindex; 49 pgno_t pg; 50 size_t nbytes; 51 int dflags, exact, status; 52 char *dest, db[NOVFLSIZE], kb[NOVFLSIZE]; 53 54 switch (flags) { 55 case R_CURSOR: 56 if (ISSET(t, BTF_DELCRSR)) { 57 errno = EINVAL; 58 return (RET_ERROR); 59 } 60 break; 61 case 0: 62 case R_NOOVERWRITE: 63 break; 64 default: 65 errno = EINVAL; 66 return (RET_ERROR); 67 } 68 69 t = dbp->internal; 70 if (ISSET(t, BTF_RDONLY)) { 71 errno = EPERM; 72 return (RET_ERROR); 73 } 74 75 /* 76 * If the key/data won't fit on a page, store it on indirect pages. 77 * Only store the key on the overflow page if it's too big after the 78 * data is on an overflow page. 79 * 80 * XXX 81 * If the insert fails later on, these pages aren't recovered. 82 */ 83 dflags = 0; 84 if (key->size + data->size > t->bt_ovflsize) { 85 if (key->size > t->bt_ovflsize) { 86 storekey: if (__ovfl_put(t, key, &pg) == RET_ERROR) 87 return (RET_ERROR); 88 tkey.data = kb; 89 tkey.size = NOVFLSIZE; 90 *(pgno_t *)kb = pg; 91 *(size_t *)(kb + sizeof(pgno_t)) = key->size; 92 dflags |= P_BIGKEY; 93 key = &tkey; 94 } 95 if (key->size + data->size > t->bt_ovflsize) { 96 if (__ovfl_put(t, data, &pg) == RET_ERROR) 97 return (RET_ERROR); 98 tdata.data = db; 99 tdata.size = NOVFLSIZE; 100 *(pgno_t *)db = pg; 101 *(size_t *)(db + sizeof(pgno_t)) = data->size; 102 dflags |= P_BIGDATA; 103 data = &tdata; 104 } 105 if (key->size + data->size > t->bt_ovflsize) 106 goto storekey; 107 } 108 109 /* Replace the cursor. */ 110 if (flags == R_CURSOR) { 111 if ((h = mpool_get(t->bt_mp, t->bt_bcursor.pgno, 0)) == NULL) 112 return (RET_ERROR); 113 index = t->bt_bcursor.index; 114 goto delete; 115 } 116 117 /* 118 * Find the key to delete, or, the location at which to insert. Bt_fast 119 * and __bt_search pin the returned page. 120 */ 121 if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL) 122 if ((e = __bt_search(t, key, &exact)) == NULL) 123 return (RET_ERROR); 124 h = e->page; 125 index = e->index; 126 127 /* 128 * Add the specified key/data pair to the tree. If an identical key 129 * is already in the tree, and R_NOOVERWRITE is set, an error is 130 * returned. If R_NOOVERWRITE is not set, the key is either added (if 131 * duplicates are permitted) or an error is returned. 132 * 133 * Pages are split as required. 134 */ 135 switch (flags) { 136 case R_NOOVERWRITE: 137 if (!exact) 138 break; 139 /* 140 * One special case is if the cursor references the record and 141 * it's been flagged for deletion. Then, we delete the record, 142 * leaving the cursor there -- this means that the inserted 143 * record will not be seen in a cursor scan. 144 */ 145 if (ISSET(t, BTF_DELCRSR) && t->bt_bcursor.pgno == h->pgno && 146 t->bt_bcursor.index == index) { 147 UNSET(t, BTF_DELCRSR); 148 goto delete; 149 } 150 BT_CLR(t); 151 mpool_put(t->bt_mp, h, 0); 152 return (RET_SPECIAL); 153 default: 154 if (!exact || NOTSET(t, BTF_NODUPS)) 155 break; 156 delete: if (__bt_dleaf(t, h, index) == RET_ERROR) { 157 BT_CLR(t); 158 mpool_put(t->bt_mp, h, 0); 159 return (RET_ERROR); 160 } 161 break; 162 } 163 164 /* 165 * If not enough room, or the user has put a ceiling on the number of 166 * keys permitted in the page, split the page. The split code will 167 * insert the key and data and unpin the current page. If inserting 168 * into the offset array, shift the pointers up. 169 */ 170 nbytes = NBLEAFDBT(key->size, data->size); 171 if (h->upper - h->lower < nbytes + sizeof(index_t)) { 172 status = __bt_split(t, h, key, data, dflags, nbytes, index); 173 if (status == RET_SUCCESS) 174 SET(t, BTF_MODIFIED); 175 return (status); 176 } 177 178 if (index < (nxtindex = NEXTINDEX(h))) 179 bcopy(h->linp + index, h->linp + index + 1, 180 (nxtindex - index) * sizeof(index_t)); 181 h->lower += sizeof(index_t); 182 183 h->linp[index] = h->upper -= nbytes; 184 dest = (char *)h + h->upper; 185 WR_BLEAF(dest, key, data, dflags); 186 187 if (t->bt_order == NOT) 188 if (h->nextpg == P_INVALID) { 189 if (index == NEXTINDEX(h) - 1) { 190 t->bt_order = FORWARD; 191 t->bt_last.index = index; 192 t->bt_last.pgno = h->pgno; 193 } 194 } else if (h->prevpg == P_INVALID) { 195 if (index == 0) { 196 t->bt_order = BACK; 197 t->bt_last.index = 0; 198 t->bt_last.pgno = h->pgno; 199 } 200 } 201 202 mpool_put(t->bt_mp, h, MPOOL_DIRTY); 203 BT_CLR(t); 204 SET(t, BTF_MODIFIED); 205 return (RET_SUCCESS); 206 } 207 208 #ifdef STATISTICS 209 u_long bt_cache_hit, bt_cache_miss; 210 #endif 211 212 /* 213 * BT_FAST -- Do a quick check for sorted data. 214 * 215 * Parameters: 216 * t: tree 217 * key: key to insert 218 * 219 * Returns: 220 * EPG for new record or NULL if not found. 221 */ 222 static EPG * 223 bt_fast(t, key, data, exactp) 224 BTREE *t; 225 const DBT *key, *data; 226 int *exactp; 227 { 228 EPG e; 229 PAGE *h; 230 size_t nbytes; 231 int cmp; 232 233 if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) { 234 t->bt_order = NOT; 235 return (NULL); 236 } 237 e.page = h; 238 e.index = t->bt_last.index; 239 240 /* 241 * If won't fit in this page or have too many keys in this page, have 242 * to search to get split stack. 243 */ 244 nbytes = NBLEAFDBT(key->size, data->size); 245 if (h->upper - h->lower < nbytes + sizeof(index_t)) 246 goto miss; 247 248 if (t->bt_order == FORWARD) { 249 if (e.page->nextpg != P_INVALID) 250 goto miss; 251 if (e.index != NEXTINDEX(h) - 1) 252 goto miss; 253 if ((cmp = __bt_cmp(t, key, &e)) < 0) 254 goto miss; 255 t->bt_last.index = cmp ? ++e.index : e.index; 256 } else { 257 if (e.page->prevpg != P_INVALID) 258 goto miss; 259 if (e.index != 0) 260 goto miss; 261 if ((cmp = __bt_cmp(t, key, &e)) > 0) 262 goto miss; 263 t->bt_last.index = 0; 264 } 265 *exactp = cmp == 0; 266 #ifdef STATISTICS 267 ++bt_cache_hit; 268 #endif 269 return (&e); 270 271 miss: 272 #ifdef STATISTICS 273 ++bt_cache_miss; 274 #endif 275 t->bt_order = NOT; 276 mpool_put(t->bt_mp, h, 0); 277 return (NULL); 278 } 279