Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nbtree.h
4 : * header file for postgres btree access method implementation.
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
8 : * Portions Copyright (c) 1994, Regents of the University of California
9 : *
10 : * src/include/access/nbtree.h
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 : #ifndef NBTREE_H
15 : #define NBTREE_H
16 :
17 : #include "access/amapi.h"
18 : #include "access/itup.h"
19 : #include "access/sdir.h"
20 : #include "access/tableam.h"
21 : #include "access/xlogreader.h"
22 : #include "catalog/pg_am_d.h"
23 : #include "catalog/pg_index.h"
24 : #include "lib/stringinfo.h"
25 : #include "storage/bufmgr.h"
26 : #include "storage/shm_toc.h"
27 :
28 : /* There's room for a 16-bit vacuum cycle ID in BTPageOpaqueData */
29 : typedef uint16 BTCycleId;
30 :
31 : /*
32 : * BTPageOpaqueData -- At the end of every page, we store a pointer
33 : * to both siblings in the tree. This is used to do forward/backward
34 : * index scans. The next-page link is also critical for recovery when
35 : * a search has navigated to the wrong page due to concurrent page splits
36 : * or deletions; see src/backend/access/nbtree/README for more info.
37 : *
38 : * In addition, we store the page's btree level (counting upwards from
39 : * zero at a leaf page) as well as some flag bits indicating the page type
40 : * and status. If the page is deleted, a BTDeletedPageData struct is stored
41 : * in the page's tuple area, while a standard BTPageOpaqueData struct is
42 : * stored in the page special area.
43 : *
44 : * We also store a "vacuum cycle ID". When a page is split while VACUUM is
45 : * processing the index, a nonzero value associated with the VACUUM run is
46 : * stored into both halves of the split page. (If VACUUM is not running,
47 : * both pages receive zero cycleids.) This allows VACUUM to detect whether
48 : * a page was split since it started, with a small probability of false match
49 : * if the page was last split some exact multiple of MAX_BT_CYCLE_ID VACUUMs
50 : * ago. Also, during a split, the BTP_SPLIT_END flag is cleared in the left
51 : * (original) page, and set in the right page, but only if the next page
52 : * to its right has a different cycleid.
53 : *
54 : * NOTE: the BTP_LEAF flag bit is redundant since level==0 could be tested
55 : * instead.
56 : *
57 : * NOTE: the btpo_level field used to be a union type in order to allow
58 : * deleted pages to store a 32-bit safexid in the same field. We now store
59 : * 64-bit/full safexid values using BTDeletedPageData instead.
60 : */
61 :
62 : typedef struct BTPageOpaqueData
63 : {
64 : BlockNumber btpo_prev; /* left sibling, or P_NONE if leftmost */
65 : BlockNumber btpo_next; /* right sibling, or P_NONE if rightmost */
66 : uint32 btpo_level; /* tree level --- zero for leaf pages */
67 : uint16 btpo_flags; /* flag bits, see below */
68 : BTCycleId btpo_cycleid; /* vacuum cycle ID of latest split */
69 : } BTPageOpaqueData;
70 :
71 : typedef BTPageOpaqueData *BTPageOpaque;
72 :
73 : #define BTPageGetOpaque(page) ((BTPageOpaque) PageGetSpecialPointer(page))
74 :
75 : /* Bits defined in btpo_flags */
76 : #define BTP_LEAF (1 << 0) /* leaf page, i.e. not internal page */
77 : #define BTP_ROOT (1 << 1) /* root page (has no parent) */
78 : #define BTP_DELETED (1 << 2) /* page has been deleted from tree */
79 : #define BTP_META (1 << 3) /* meta-page */
80 : #define BTP_HALF_DEAD (1 << 4) /* empty, but still in tree */
81 : #define BTP_SPLIT_END (1 << 5) /* rightmost page of split group */
82 : #define BTP_HAS_GARBAGE (1 << 6) /* page has LP_DEAD tuples (deprecated) */
83 : #define BTP_INCOMPLETE_SPLIT (1 << 7) /* right sibling's downlink is missing */
84 : #define BTP_HAS_FULLXID (1 << 8) /* contains BTDeletedPageData */
85 :
86 : /*
87 : * The max allowed value of a cycle ID is a bit less than 64K. This is
88 : * for convenience of pg_filedump and similar utilities: we want to use
89 : * the last 2 bytes of special space as an index type indicator, and
90 : * restricting cycle ID lets btree use that space for vacuum cycle IDs
91 : * while still allowing index type to be identified.
92 : */
93 : #define MAX_BT_CYCLE_ID 0xFF7F
94 :
95 :
96 : /*
97 : * The Meta page is always the first page in the btree index.
98 : * Its primary purpose is to point to the location of the btree root page.
99 : * We also point to the "fast" root, which is the current effective root;
100 : * see README for discussion.
101 : */
102 :
103 : typedef struct BTMetaPageData
104 : {
105 : uint32 btm_magic; /* should contain BTREE_MAGIC */
106 : uint32 btm_version; /* nbtree version (always <= BTREE_VERSION) */
107 : BlockNumber btm_root; /* current root location */
108 : uint32 btm_level; /* tree level of the root page */
109 : BlockNumber btm_fastroot; /* current "fast" root location */
110 : uint32 btm_fastlevel; /* tree level of the "fast" root page */
111 : /* remaining fields only valid when btm_version >= BTREE_NOVAC_VERSION */
112 :
113 : /* number of deleted, non-recyclable pages during last cleanup */
114 : uint32 btm_last_cleanup_num_delpages;
115 : /* number of heap tuples during last cleanup (deprecated) */
116 : float8 btm_last_cleanup_num_heap_tuples;
117 :
118 : bool btm_allequalimage; /* are all columns "equalimage"? */
119 : } BTMetaPageData;
120 :
121 : #define BTPageGetMeta(p) \
122 : ((BTMetaPageData *) PageGetContents(p))
123 :
124 : /*
125 : * The current Btree version is 4. That's what you'll get when you create
126 : * a new index.
127 : *
128 : * Btree version 3 was used in PostgreSQL v11. It is mostly the same as
129 : * version 4, but heap TIDs were not part of the keyspace. Index tuples
130 : * with duplicate keys could be stored in any order. We continue to
131 : * support reading and writing Btree versions 2 and 3, so that they don't
132 : * need to be immediately re-indexed at pg_upgrade. In order to get the
133 : * new heapkeyspace semantics, however, a REINDEX is needed.
134 : *
135 : * Deduplication is safe to use when the btm_allequalimage field is set to
136 : * true. It's safe to read the btm_allequalimage field on version 3, but
137 : * only version 4 indexes make use of deduplication. Even version 4
138 : * indexes created on PostgreSQL v12 will need a REINDEX to make use of
139 : * deduplication, though, since there is no other way to set
140 : * btm_allequalimage to true (pg_upgrade hasn't been taught to set the
141 : * metapage field).
142 : *
143 : * Btree version 2 is mostly the same as version 3. There are two new
144 : * fields in the metapage that were introduced in version 3. A version 2
145 : * metapage will be automatically upgraded to version 3 on the first
146 : * insert to it. INCLUDE indexes cannot use version 2.
147 : */
148 : #define BTREE_METAPAGE 0 /* first page is meta */
149 : #define BTREE_MAGIC 0x053162 /* magic number in metapage */
150 : #define BTREE_VERSION 4 /* current version number */
151 : #define BTREE_MIN_VERSION 2 /* minimum supported version */
152 : #define BTREE_NOVAC_VERSION 3 /* version with all meta fields set */
153 :
154 : /*
155 : * Maximum size of a btree index entry, including its tuple header.
156 : *
157 : * We actually need to be able to fit three items on every page,
158 : * so restrict any one item to 1/3 the per-page available space.
159 : *
160 : * There are rare cases where _bt_truncate() will need to enlarge
161 : * a heap index tuple to make space for a tiebreaker heap TID
162 : * attribute, which we account for here.
163 : */
164 : #define BTMaxItemSize \
165 : (MAXALIGN_DOWN((BLCKSZ - \
166 : MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \
167 : MAXALIGN(sizeof(BTPageOpaqueData))) / 3) - \
168 : MAXALIGN(sizeof(ItemPointerData)))
169 : #define BTMaxItemSizeNoHeapTid \
170 : MAXALIGN_DOWN((BLCKSZ - \
171 : MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \
172 : MAXALIGN(sizeof(BTPageOpaqueData))) / 3)
173 :
174 : /*
175 : * MaxTIDsPerBTreePage is an upper bound on the number of heap TIDs tuples
176 : * that may be stored on a btree leaf page. It is used to size the
177 : * per-page temporary buffers.
178 : *
179 : * Note: we don't bother considering per-tuple overheads here to keep
180 : * things simple (value is based on how many elements a single array of
181 : * heap TIDs must have to fill the space between the page header and
182 : * special area). The value is slightly higher (i.e. more conservative)
183 : * than necessary as a result, which is considered acceptable.
184 : */
185 : #define MaxTIDsPerBTreePage \
186 : (int) ((BLCKSZ - SizeOfPageHeaderData - sizeof(BTPageOpaqueData)) / \
187 : sizeof(ItemPointerData))
188 :
189 : /*
190 : * The leaf-page fillfactor defaults to 90% but is user-adjustable.
191 : * For pages above the leaf level, we use a fixed 70% fillfactor.
192 : * The fillfactor is applied during index build and when splitting
193 : * a rightmost page; when splitting non-rightmost pages we try to
194 : * divide the data equally. When splitting a page that's entirely
195 : * filled with a single value (duplicates), the effective leaf-page
196 : * fillfactor is 96%, regardless of whether the page is a rightmost
197 : * page.
198 : */
199 : #define BTREE_MIN_FILLFACTOR 10
200 : #define BTREE_DEFAULT_FILLFACTOR 90
201 : #define BTREE_NONLEAF_FILLFACTOR 70
202 : #define BTREE_SINGLEVAL_FILLFACTOR 96
203 :
204 : /*
205 : * In general, the btree code tries to localize its knowledge about
206 : * page layout to a couple of routines. However, we need a special
207 : * value to indicate "no page number" in those places where we expect
208 : * page numbers. We can use zero for this because we never need to
209 : * make a pointer to the metadata page.
210 : */
211 :
212 : #define P_NONE 0
213 :
214 : /*
215 : * Macros to test whether a page is leftmost or rightmost on its tree level,
216 : * as well as other state info kept in the opaque data.
217 : */
218 : #define P_LEFTMOST(opaque) ((opaque)->btpo_prev == P_NONE)
219 : #define P_RIGHTMOST(opaque) ((opaque)->btpo_next == P_NONE)
220 : #define P_ISLEAF(opaque) (((opaque)->btpo_flags & BTP_LEAF) != 0)
221 : #define P_ISROOT(opaque) (((opaque)->btpo_flags & BTP_ROOT) != 0)
222 : #define P_ISDELETED(opaque) (((opaque)->btpo_flags & BTP_DELETED) != 0)
223 : #define P_ISMETA(opaque) (((opaque)->btpo_flags & BTP_META) != 0)
224 : #define P_ISHALFDEAD(opaque) (((opaque)->btpo_flags & BTP_HALF_DEAD) != 0)
225 : #define P_IGNORE(opaque) (((opaque)->btpo_flags & (BTP_DELETED|BTP_HALF_DEAD)) != 0)
226 : #define P_HAS_GARBAGE(opaque) (((opaque)->btpo_flags & BTP_HAS_GARBAGE) != 0)
227 : #define P_INCOMPLETE_SPLIT(opaque) (((opaque)->btpo_flags & BTP_INCOMPLETE_SPLIT) != 0)
228 : #define P_HAS_FULLXID(opaque) (((opaque)->btpo_flags & BTP_HAS_FULLXID) != 0)
229 :
230 : /*
231 : * BTDeletedPageData is the page contents of a deleted page
232 : */
233 : typedef struct BTDeletedPageData
234 : {
235 : FullTransactionId safexid; /* See BTPageIsRecyclable() */
236 : } BTDeletedPageData;
237 :
238 : static inline void
239 6690 : BTPageSetDeleted(Page page, FullTransactionId safexid)
240 : {
241 : BTPageOpaque opaque;
242 : PageHeader header;
243 : BTDeletedPageData *contents;
244 :
245 6690 : opaque = BTPageGetOpaque(page);
246 6690 : header = ((PageHeader) page);
247 :
248 6690 : opaque->btpo_flags &= ~BTP_HALF_DEAD;
249 6690 : opaque->btpo_flags |= BTP_DELETED | BTP_HAS_FULLXID;
250 6690 : header->pd_lower = MAXALIGN(SizeOfPageHeaderData) +
251 : sizeof(BTDeletedPageData);
252 6690 : header->pd_upper = header->pd_special;
253 :
254 : /* Set safexid in deleted page */
255 6690 : contents = ((BTDeletedPageData *) PageGetContents(page));
256 6690 : contents->safexid = safexid;
257 6690 : }
258 :
259 : static inline FullTransactionId
260 552 : BTPageGetDeleteXid(Page page)
261 : {
262 : BTPageOpaque opaque;
263 : BTDeletedPageData *contents;
264 :
265 : /* We only expect to be called with a deleted page */
266 : Assert(!PageIsNew(page));
267 552 : opaque = BTPageGetOpaque(page);
268 : Assert(P_ISDELETED(opaque));
269 :
270 : /* pg_upgrade'd deleted page -- must be safe to recycle now */
271 552 : if (!P_HAS_FULLXID(opaque))
272 0 : return FirstNormalFullTransactionId;
273 :
274 : /* Get safexid from deleted page */
275 552 : contents = ((BTDeletedPageData *) PageGetContents(page));
276 552 : return contents->safexid;
277 : }
278 :
279 : /*
280 : * Is an existing page recyclable?
281 : *
282 : * This exists to centralize the policy on which deleted pages are now safe to
283 : * re-use. However, _bt_pendingfsm_finalize() duplicates some of the same
284 : * logic because it doesn't work directly with pages -- keep the two in sync.
285 : *
286 : * Note: PageIsNew() pages are always safe to recycle, but we can't deal with
287 : * them here (caller is responsible for that case themselves). Caller might
288 : * well need special handling for new pages anyway.
289 : */
290 : static inline bool
291 21768 : BTPageIsRecyclable(Page page, Relation heaprel)
292 : {
293 : BTPageOpaque opaque;
294 :
295 : Assert(!PageIsNew(page));
296 : Assert(heaprel != NULL);
297 :
298 : /* Recycling okay iff page is deleted and safexid is old enough */
299 21768 : opaque = BTPageGetOpaque(page);
300 21768 : if (P_ISDELETED(opaque))
301 : {
302 446 : FullTransactionId safexid = BTPageGetDeleteXid(page);
303 :
304 : /*
305 : * The page was deleted, but when? If it was just deleted, a scan
306 : * might have seen the downlink to it, and will read the page later.
307 : * As long as that can happen, we must keep the deleted page around as
308 : * a tombstone.
309 : *
310 : * For that check if the deletion XID could still be visible to
311 : * anyone. If not, then no scan that's still in progress could have
312 : * seen its downlink, and we can recycle it.
313 : */
314 446 : return GlobalVisCheckRemovableFullXid(heaprel, safexid);
315 : }
316 :
317 21322 : return false;
318 : }
319 :
320 : /*
321 : * BTVacState and BTPendingFSM are private nbtree.c state used during VACUUM.
322 : * They are exported for use by page deletion related code in nbtpage.c.
323 : */
324 : typedef struct BTPendingFSM
325 : {
326 : BlockNumber target; /* Page deleted by current VACUUM */
327 : FullTransactionId safexid; /* Page's BTDeletedPageData.safexid */
328 : } BTPendingFSM;
329 :
330 : typedef struct BTVacState
331 : {
332 : IndexVacuumInfo *info;
333 : IndexBulkDeleteResult *stats;
334 : IndexBulkDeleteCallback callback;
335 : void *callback_state;
336 : BTCycleId cycleid;
337 : MemoryContext pagedelcontext;
338 :
339 : /*
340 : * _bt_pendingfsm_finalize() state
341 : */
342 : int bufsize; /* pendingpages space (in # elements) */
343 : int maxbufsize; /* max bufsize that respects work_mem */
344 : BTPendingFSM *pendingpages; /* One entry per newly deleted page */
345 : int npendingpages; /* current # valid pendingpages */
346 : } BTVacState;
347 :
348 : /*
349 : * Lehman and Yao's algorithm requires a ``high key'' on every non-rightmost
350 : * page. The high key is not a tuple that is used to visit the heap. It is
351 : * a pivot tuple (see "Notes on B-Tree tuple format" below for definition).
352 : * The high key on a page is required to be greater than or equal to any
353 : * other key that appears on the page. If we find ourselves trying to
354 : * insert a key that is strictly > high key, we know we need to move right
355 : * (this should only happen if the page was split since we examined the
356 : * parent page).
357 : *
358 : * Our insertion algorithm guarantees that we can use the initial least key
359 : * on our right sibling as the high key. Once a page is created, its high
360 : * key changes only if the page is split.
361 : *
362 : * On a non-rightmost page, the high key lives in item 1 and data items
363 : * start in item 2. Rightmost pages have no high key, so we store data
364 : * items beginning in item 1.
365 : */
366 :
367 : #define P_HIKEY ((OffsetNumber) 1)
368 : #define P_FIRSTKEY ((OffsetNumber) 2)
369 : #define P_FIRSTDATAKEY(opaque) (P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY)
370 :
371 : /*
372 : * Notes on B-Tree tuple format, and key and non-key attributes:
373 : *
374 : * INCLUDE B-Tree indexes have non-key attributes. These are extra
375 : * attributes that may be returned by index-only scans, but do not influence
376 : * the order of items in the index (formally, non-key attributes are not
377 : * considered to be part of the key space). Non-key attributes are only
378 : * present in leaf index tuples whose item pointers actually point to heap
379 : * tuples (non-pivot tuples). _bt_check_natts() enforces the rules
380 : * described here.
381 : *
382 : * Non-pivot tuple format (plain/non-posting variant):
383 : *
384 : * t_tid | t_info | key values | INCLUDE columns, if any
385 : *
386 : * t_tid points to the heap TID, which is a tiebreaker key column as of
387 : * BTREE_VERSION 4.
388 : *
389 : * Non-pivot tuples complement pivot tuples, which only have key columns.
390 : * The sole purpose of pivot tuples is to represent how the key space is
391 : * separated. In general, any B-Tree index that has more than one level
392 : * (i.e. any index that does not just consist of a metapage and a single
393 : * leaf root page) must have some number of pivot tuples, since pivot
394 : * tuples are used for traversing the tree. Suffix truncation can omit
395 : * trailing key columns when a new pivot is formed, which makes minus
396 : * infinity their logical value. Since BTREE_VERSION 4 indexes treat heap
397 : * TID as a trailing key column that ensures that all index tuples are
398 : * physically unique, it is necessary to represent heap TID as a trailing
399 : * key column in pivot tuples, though very often this can be truncated
400 : * away, just like any other key column. (Actually, the heap TID is
401 : * omitted rather than truncated, since its representation is different to
402 : * the non-pivot representation.)
403 : *
404 : * Pivot tuple format:
405 : *
406 : * t_tid | t_info | key values | [heap TID]
407 : *
408 : * We store the number of columns present inside pivot tuples by abusing
409 : * their t_tid offset field, since pivot tuples never need to store a real
410 : * offset (pivot tuples generally store a downlink in t_tid, though). The
411 : * offset field only stores the number of columns/attributes when the
412 : * INDEX_ALT_TID_MASK bit is set, which doesn't count the trailing heap
413 : * TID column sometimes stored in pivot tuples -- that's represented by
414 : * the presence of BT_PIVOT_HEAP_TID_ATTR. The INDEX_ALT_TID_MASK bit in
415 : * t_info is always set on BTREE_VERSION 4 pivot tuples, since
416 : * BTreeTupleIsPivot() must work reliably on heapkeyspace versions.
417 : *
418 : * In version 2 or version 3 (!heapkeyspace) indexes, INDEX_ALT_TID_MASK
419 : * might not be set in pivot tuples. BTreeTupleIsPivot() won't work
420 : * reliably as a result. The number of columns stored is implicitly the
421 : * same as the number of columns in the index, just like any non-pivot
422 : * tuple. (The number of columns stored should not vary, since suffix
423 : * truncation of key columns is unsafe within any !heapkeyspace index.)
424 : *
425 : * The 12 least significant bits from t_tid's offset number are used to
426 : * represent the number of key columns within a pivot tuple. This leaves 4
427 : * status bits (BT_STATUS_OFFSET_MASK bits), which are shared by all tuples
428 : * that have the INDEX_ALT_TID_MASK bit set (set in t_info) to store basic
429 : * tuple metadata. BTreeTupleIsPivot() and BTreeTupleIsPosting() use the
430 : * BT_STATUS_OFFSET_MASK bits.
431 : *
432 : * Sometimes non-pivot tuples also use a representation that repurposes
433 : * t_tid to store metadata rather than a TID. PostgreSQL v13 introduced a
434 : * new non-pivot tuple format to support deduplication: posting list
435 : * tuples. Deduplication merges together multiple equal non-pivot tuples
436 : * into a logically equivalent, space efficient representation. A posting
437 : * list is an array of ItemPointerData elements. Non-pivot tuples are
438 : * merged together to form posting list tuples lazily, at the point where
439 : * we'd otherwise have to split a leaf page.
440 : *
441 : * Posting tuple format (alternative non-pivot tuple representation):
442 : *
443 : * t_tid | t_info | key values | posting list (TID array)
444 : *
445 : * Posting list tuples are recognized as such by having the
446 : * INDEX_ALT_TID_MASK status bit set in t_info and the BT_IS_POSTING status
447 : * bit set in t_tid's offset number. These flags redefine the content of
448 : * the posting tuple's t_tid to store the location of the posting list
449 : * (instead of a block number), as well as the total number of heap TIDs
450 : * present in the tuple (instead of a real offset number).
451 : *
452 : * The 12 least significant bits from t_tid's offset number are used to
453 : * represent the number of heap TIDs present in the tuple, leaving 4 status
454 : * bits (the BT_STATUS_OFFSET_MASK bits). Like any non-pivot tuple, the
455 : * number of columns stored is always implicitly the total number in the
456 : * index (in practice there can never be non-key columns stored, since
457 : * deduplication is not supported with INCLUDE indexes).
458 : */
459 : #define INDEX_ALT_TID_MASK INDEX_AM_RESERVED_BIT
460 :
461 : /* Item pointer offset bit masks */
462 : #define BT_OFFSET_MASK 0x0FFF
463 : #define BT_STATUS_OFFSET_MASK 0xF000
464 : /* BT_STATUS_OFFSET_MASK status bits */
465 : #define BT_PIVOT_HEAP_TID_ATTR 0x1000
466 : #define BT_IS_POSTING 0x2000
467 :
468 : /*
469 : * Mask allocated for number of keys in index tuple must be able to fit
470 : * maximum possible number of index attributes
471 : */
472 : StaticAssertDecl(BT_OFFSET_MASK >= INDEX_MAX_KEYS,
473 : "BT_OFFSET_MASK can't fit INDEX_MAX_KEYS");
474 :
475 : /*
476 : * Note: BTreeTupleIsPivot() can have false negatives (but not false
477 : * positives) when used with !heapkeyspace indexes
478 : */
479 : static inline bool
480 303081526 : BTreeTupleIsPivot(IndexTuple itup)
481 : {
482 303081526 : if ((itup->t_info & INDEX_ALT_TID_MASK) == 0)
483 220416040 : return false;
484 : /* absence of BT_IS_POSTING in offset number indicates pivot tuple */
485 82665486 : if ((ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) & BT_IS_POSTING) != 0)
486 3839376 : return false;
487 :
488 78826110 : return true;
489 : }
490 :
491 : static inline bool
492 138066544 : BTreeTupleIsPosting(IndexTuple itup)
493 : {
494 138066544 : if ((itup->t_info & INDEX_ALT_TID_MASK) == 0)
495 126464088 : return false;
496 : /* presence of BT_IS_POSTING in offset number indicates posting tuple */
497 11602456 : if ((ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) & BT_IS_POSTING) == 0)
498 301050 : return false;
499 :
500 11301406 : return true;
501 : }
502 :
503 : static inline void
504 446430 : BTreeTupleSetPosting(IndexTuple itup, uint16 nhtids, int postingoffset)
505 : {
506 : Assert(nhtids > 1);
507 : Assert((nhtids & BT_STATUS_OFFSET_MASK) == 0);
508 : Assert((size_t) postingoffset == MAXALIGN(postingoffset));
509 : Assert(postingoffset < INDEX_SIZE_MASK);
510 : Assert(!BTreeTupleIsPivot(itup));
511 :
512 446430 : itup->t_info |= INDEX_ALT_TID_MASK;
513 446430 : ItemPointerSetOffsetNumber(&itup->t_tid, (nhtids | BT_IS_POSTING));
514 446430 : ItemPointerSetBlockNumber(&itup->t_tid, postingoffset);
515 446430 : }
516 :
517 : static inline uint16
518 13445420 : BTreeTupleGetNPosting(IndexTuple posting)
519 : {
520 : OffsetNumber existing;
521 :
522 : Assert(BTreeTupleIsPosting(posting));
523 :
524 13445420 : existing = ItemPointerGetOffsetNumberNoCheck(&posting->t_tid);
525 13445420 : return (existing & BT_OFFSET_MASK);
526 : }
527 :
528 : static inline uint32
529 17804256 : BTreeTupleGetPostingOffset(IndexTuple posting)
530 : {
531 : Assert(BTreeTupleIsPosting(posting));
532 :
533 17804256 : return ItemPointerGetBlockNumberNoCheck(&posting->t_tid);
534 : }
535 :
536 : static inline ItemPointer
537 15458396 : BTreeTupleGetPosting(IndexTuple posting)
538 : {
539 30916792 : return (ItemPointer) ((char *) posting +
540 15458396 : BTreeTupleGetPostingOffset(posting));
541 : }
542 :
543 : static inline ItemPointer
544 12432176 : BTreeTupleGetPostingN(IndexTuple posting, int n)
545 : {
546 12432176 : return BTreeTupleGetPosting(posting) + n;
547 : }
548 :
549 : /*
550 : * Get/set downlink block number in pivot tuple.
551 : *
552 : * Note: Cannot assert that tuple is a pivot tuple. If we did so then
553 : * !heapkeyspace indexes would exhibit false positive assertion failures.
554 : */
555 : static inline BlockNumber
556 17506688 : BTreeTupleGetDownLink(IndexTuple pivot)
557 : {
558 17506688 : return ItemPointerGetBlockNumberNoCheck(&pivot->t_tid);
559 : }
560 :
561 : static inline void
562 71336 : BTreeTupleSetDownLink(IndexTuple pivot, BlockNumber blkno)
563 : {
564 71336 : ItemPointerSetBlockNumber(&pivot->t_tid, blkno);
565 71336 : }
566 :
567 : /*
568 : * Get number of attributes within tuple.
569 : *
570 : * Note that this does not include an implicit tiebreaker heap TID
571 : * attribute, if any. Note also that the number of key attributes must be
572 : * explicitly represented in all heapkeyspace pivot tuples.
573 : *
574 : * Note: This is defined as a macro rather than an inline function to
575 : * avoid including rel.h.
576 : */
577 : #define BTreeTupleGetNAtts(itup, rel) \
578 : ( \
579 : (BTreeTupleIsPivot(itup)) ? \
580 : ( \
581 : ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_OFFSET_MASK \
582 : ) \
583 : : \
584 : IndexRelationGetNumberOfAttributes(rel) \
585 : )
586 :
587 : /*
588 : * Set number of key attributes in tuple.
589 : *
590 : * The heap TID tiebreaker attribute bit may also be set here, indicating that
591 : * a heap TID value will be stored at the end of the tuple (i.e. using the
592 : * special pivot tuple representation).
593 : */
594 : static inline void
595 83622 : BTreeTupleSetNAtts(IndexTuple itup, uint16 nkeyatts, bool heaptid)
596 : {
597 : Assert(nkeyatts <= INDEX_MAX_KEYS);
598 : Assert((nkeyatts & BT_STATUS_OFFSET_MASK) == 0);
599 : Assert(!heaptid || nkeyatts > 0);
600 : Assert(!BTreeTupleIsPivot(itup) || nkeyatts == 0);
601 :
602 83622 : itup->t_info |= INDEX_ALT_TID_MASK;
603 :
604 83622 : if (heaptid)
605 1096 : nkeyatts |= BT_PIVOT_HEAP_TID_ATTR;
606 :
607 : /* BT_IS_POSTING bit is deliberately unset here */
608 83622 : ItemPointerSetOffsetNumber(&itup->t_tid, nkeyatts);
609 : Assert(BTreeTupleIsPivot(itup));
610 83622 : }
611 :
612 : /*
613 : * Get/set leaf page's "top parent" link from its high key. Used during page
614 : * deletion.
615 : *
616 : * Note: Cannot assert that tuple is a pivot tuple. If we did so then
617 : * !heapkeyspace indexes would exhibit false positive assertion failures.
618 : */
619 : static inline BlockNumber
620 5490 : BTreeTupleGetTopParent(IndexTuple leafhikey)
621 : {
622 5490 : return ItemPointerGetBlockNumberNoCheck(&leafhikey->t_tid);
623 : }
624 :
625 : static inline void
626 6692 : BTreeTupleSetTopParent(IndexTuple leafhikey, BlockNumber blkno)
627 : {
628 6692 : ItemPointerSetBlockNumber(&leafhikey->t_tid, blkno);
629 6692 : BTreeTupleSetNAtts(leafhikey, 0, false);
630 6692 : }
631 :
632 : /*
633 : * Get tiebreaker heap TID attribute, if any.
634 : *
635 : * This returns the first/lowest heap TID in the case of a posting list tuple.
636 : */
637 : static inline ItemPointer
638 33511464 : BTreeTupleGetHeapTID(IndexTuple itup)
639 : {
640 33511464 : if (BTreeTupleIsPivot(itup))
641 : {
642 : /* Pivot tuple heap TID representation? */
643 373084 : if ((ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) &
644 : BT_PIVOT_HEAP_TID_ATTR) != 0)
645 277948 : return (ItemPointer) ((char *) itup + IndexTupleSize(itup) -
646 : sizeof(ItemPointerData));
647 :
648 : /* Heap TID attribute was truncated */
649 95136 : return NULL;
650 : }
651 33138380 : else if (BTreeTupleIsPosting(itup))
652 526758 : return BTreeTupleGetPosting(itup);
653 :
654 32611622 : return &itup->t_tid;
655 : }
656 :
657 : /*
658 : * Get maximum heap TID attribute, which could be the only TID in the case of
659 : * a non-pivot tuple that does not have a posting list.
660 : *
661 : * Works with non-pivot tuples only.
662 : */
663 : static inline ItemPointer
664 233824 : BTreeTupleGetMaxHeapTID(IndexTuple itup)
665 : {
666 : Assert(!BTreeTupleIsPivot(itup));
667 :
668 233824 : if (BTreeTupleIsPosting(itup))
669 : {
670 233302 : uint16 nposting = BTreeTupleGetNPosting(itup);
671 :
672 233302 : return BTreeTupleGetPostingN(itup, nposting - 1);
673 : }
674 :
675 522 : return &itup->t_tid;
676 : }
677 :
678 : /*
679 : * Operator strategy numbers for B-tree have been moved to access/stratnum.h,
680 : * because many places need to use them in ScanKeyInit() calls.
681 : *
682 : * The strategy numbers are chosen so that we can commute them by
683 : * subtraction, thus:
684 : */
685 : #define BTCommuteStrategyNumber(strat) (BTMaxStrategyNumber + 1 - (strat))
686 :
687 : /*
688 : * When a new operator class is declared, we require that the user
689 : * supply us with an amproc procedure (BTORDER_PROC) for determining
690 : * whether, for two keys a and b, a < b, a = b, or a > b. This routine
691 : * must return < 0, 0, > 0, respectively, in these three cases.
692 : *
693 : * To facilitate accelerated sorting, an operator class may choose to
694 : * offer a second procedure (BTSORTSUPPORT_PROC). For full details, see
695 : * src/include/utils/sortsupport.h.
696 : *
697 : * To support window frames defined by "RANGE offset PRECEDING/FOLLOWING",
698 : * an operator class may choose to offer a third amproc procedure
699 : * (BTINRANGE_PROC), independently of whether it offers sortsupport.
700 : * For full details, see doc/src/sgml/btree.sgml.
701 : *
702 : * To facilitate B-Tree deduplication, an operator class may choose to
703 : * offer a forth amproc procedure (BTEQUALIMAGE_PROC). For full details,
704 : * see doc/src/sgml/btree.sgml.
705 : *
706 : * An operator class may choose to offer a fifth amproc procedure
707 : * (BTOPTIONS_PROC). These procedures define a set of user-visible
708 : * parameters that can be used to control operator class behavior. None of
709 : * the built-in B-Tree operator classes currently register an "options" proc.
710 : */
711 :
712 : #define BTORDER_PROC 1
713 : #define BTSORTSUPPORT_PROC 2
714 : #define BTINRANGE_PROC 3
715 : #define BTEQUALIMAGE_PROC 4
716 : #define BTOPTIONS_PROC 5
717 : #define BTNProcs 5
718 :
719 : /*
720 : * We need to be able to tell the difference between read and write
721 : * requests for pages, in order to do locking correctly.
722 : */
723 :
724 : #define BT_READ BUFFER_LOCK_SHARE
725 : #define BT_WRITE BUFFER_LOCK_EXCLUSIVE
726 :
727 : /*
728 : * BTStackData -- As we descend a tree, we push the location of pivot
729 : * tuples whose downlink we are about to follow onto a private stack. If
730 : * we split a leaf, we use this stack to walk back up the tree and insert
731 : * data into its parent page at the correct location. We also have to
732 : * recursively insert into the grandparent page if and when the parent page
733 : * splits. Our private stack can become stale due to concurrent page
734 : * splits and page deletions, but it should never give us an irredeemably
735 : * bad picture.
736 : */
737 : typedef struct BTStackData
738 : {
739 : BlockNumber bts_blkno;
740 : OffsetNumber bts_offset;
741 : struct BTStackData *bts_parent;
742 : } BTStackData;
743 :
744 : typedef BTStackData *BTStack;
745 :
746 : /*
747 : * BTScanInsertData is the btree-private state needed to find an initial
748 : * position for an indexscan, or to insert new tuples -- an "insertion
749 : * scankey" (not to be confused with a search scankey). It's used to descend
750 : * a B-Tree using _bt_search.
751 : *
752 : * heapkeyspace indicates if we expect all keys in the index to be physically
753 : * unique because heap TID is used as a tiebreaker attribute, and if index may
754 : * have truncated key attributes in pivot tuples. This is actually a property
755 : * of the index relation itself (not an indexscan). heapkeyspace indexes are
756 : * indexes whose version is >= version 4. It's convenient to keep this close
757 : * by, rather than accessing the metapage repeatedly.
758 : *
759 : * allequalimage is set to indicate that deduplication is safe for the index.
760 : * This is also a property of the index relation rather than an indexscan.
761 : *
762 : * anynullkeys indicates if any of the keys had NULL value when scankey was
763 : * built from index tuple (note that already-truncated tuple key attributes
764 : * set NULL as a placeholder key value, which also affects value of
765 : * anynullkeys). This is a convenience for unique index non-pivot tuple
766 : * insertion, which usually temporarily unsets scantid, but shouldn't iff
767 : * anynullkeys is true. Value generally matches non-pivot tuple's HasNulls
768 : * bit, but may not when inserting into an INCLUDE index (tuple header value
769 : * is affected by the NULL-ness of both key and non-key attributes).
770 : *
771 : * See comments in _bt_first for an explanation of the nextkey and backward
772 : * fields.
773 : *
774 : * scantid is the heap TID that is used as a final tiebreaker attribute. It
775 : * is set to NULL when index scan doesn't need to find a position for a
776 : * specific physical tuple. Must be set when inserting new tuples into
777 : * heapkeyspace indexes, since every tuple in the tree unambiguously belongs
778 : * in one exact position (it's never set with !heapkeyspace indexes, though).
779 : * Despite the representational difference, nbtree search code considers
780 : * scantid to be just another insertion scankey attribute.
781 : *
782 : * scankeys is an array of scan key entries for attributes that are compared
783 : * before scantid (user-visible attributes). keysz is the size of the array.
784 : * During insertion, there must be a scan key for every attribute, but when
785 : * starting a regular index scan some can be omitted. The array is used as a
786 : * flexible array member, though it's sized in a way that makes it possible to
787 : * use stack allocations. See nbtree/README for full details.
788 : */
789 : typedef struct BTScanInsertData
790 : {
791 : bool heapkeyspace;
792 : bool allequalimage;
793 : bool anynullkeys;
794 : bool nextkey;
795 : bool backward; /* backward index scan? */
796 : ItemPointer scantid; /* tiebreaker for scankeys */
797 : int keysz; /* Size of scankeys array */
798 : ScanKeyData scankeys[INDEX_MAX_KEYS]; /* Must appear last */
799 : } BTScanInsertData;
800 :
801 : typedef BTScanInsertData *BTScanInsert;
802 :
803 : /*
804 : * BTInsertStateData is a working area used during insertion.
805 : *
806 : * This is filled in after descending the tree to the first leaf page the new
807 : * tuple might belong on. Tracks the current position while performing
808 : * uniqueness check, before we have determined which exact page to insert
809 : * to.
810 : *
811 : * (This should be private to nbtinsert.c, but it's also used by
812 : * _bt_binsrch_insert)
813 : */
814 : typedef struct BTInsertStateData
815 : {
816 : IndexTuple itup; /* Item we're inserting */
817 : Size itemsz; /* Size of itup -- should be MAXALIGN()'d */
818 : BTScanInsert itup_key; /* Insertion scankey */
819 :
820 : /* Buffer containing leaf page we're likely to insert itup on */
821 : Buffer buf;
822 :
823 : /*
824 : * Cache of bounds within the current buffer. Only used for insertions
825 : * where _bt_check_unique is called. See _bt_binsrch_insert and
826 : * _bt_findinsertloc for details.
827 : */
828 : bool bounds_valid;
829 : OffsetNumber low;
830 : OffsetNumber stricthigh;
831 :
832 : /*
833 : * if _bt_binsrch_insert found the location inside existing posting list,
834 : * save the position inside the list. -1 sentinel value indicates overlap
835 : * with an existing posting list tuple that has its LP_DEAD bit set.
836 : */
837 : int postingoff;
838 : } BTInsertStateData;
839 :
840 : typedef BTInsertStateData *BTInsertState;
841 :
842 : /*
843 : * State used to representing an individual pending tuple during
844 : * deduplication.
845 : */
846 : typedef struct BTDedupInterval
847 : {
848 : OffsetNumber baseoff;
849 : uint16 nitems;
850 : } BTDedupInterval;
851 :
852 : /*
853 : * BTDedupStateData is a working area used during deduplication.
854 : *
855 : * The status info fields track the state of a whole-page deduplication pass.
856 : * State about the current pending posting list is also tracked.
857 : *
858 : * A pending posting list is comprised of a contiguous group of equal items
859 : * from the page, starting from page offset number 'baseoff'. This is the
860 : * offset number of the "base" tuple for new posting list. 'nitems' is the
861 : * current total number of existing items from the page that will be merged to
862 : * make a new posting list tuple, including the base tuple item. (Existing
863 : * items may themselves be posting list tuples, or regular non-pivot tuples.)
864 : *
865 : * The total size of the existing tuples to be freed when pending posting list
866 : * is processed gets tracked by 'phystupsize'. This information allows
867 : * deduplication to calculate the space saving for each new posting list
868 : * tuple, and for the entire pass over the page as a whole.
869 : */
870 : typedef struct BTDedupStateData
871 : {
872 : /* Deduplication status info for entire pass over page */
873 : bool deduplicate; /* Still deduplicating page? */
874 : int nmaxitems; /* Number of max-sized tuples so far */
875 : Size maxpostingsize; /* Limit on size of final tuple */
876 :
877 : /* Metadata about base tuple of current pending posting list */
878 : IndexTuple base; /* Use to form new posting list */
879 : OffsetNumber baseoff; /* page offset of base */
880 : Size basetupsize; /* base size without original posting list */
881 :
882 : /* Other metadata about pending posting list */
883 : ItemPointer htids; /* Heap TIDs in pending posting list */
884 : int nhtids; /* Number of heap TIDs in htids array */
885 : int nitems; /* Number of existing tuples/line pointers */
886 : Size phystupsize; /* Includes line pointer overhead */
887 :
888 : /*
889 : * Array of tuples to go on new version of the page. Contains one entry
890 : * for each group of consecutive items. Note that existing tuples that
891 : * will not become posting list tuples do not appear in the array (they
892 : * are implicitly unchanged by deduplication pass).
893 : */
894 : int nintervals; /* current number of intervals in array */
895 : BTDedupInterval intervals[MaxIndexTuplesPerPage];
896 : } BTDedupStateData;
897 :
898 : typedef BTDedupStateData *BTDedupState;
899 :
900 : /*
901 : * BTVacuumPostingData is state that represents how to VACUUM (or delete) a
902 : * posting list tuple when some (though not all) of its TIDs are to be
903 : * deleted.
904 : *
905 : * Convention is that itup field is the original posting list tuple on input,
906 : * and palloc()'d final tuple used to overwrite existing tuple on output.
907 : */
908 : typedef struct BTVacuumPostingData
909 : {
910 : /* Tuple that will be/was updated */
911 : IndexTuple itup;
912 : OffsetNumber updatedoffset;
913 :
914 : /* State needed to describe final itup in WAL */
915 : uint16 ndeletedtids;
916 : uint16 deletetids[FLEXIBLE_ARRAY_MEMBER];
917 : } BTVacuumPostingData;
918 :
919 : typedef BTVacuumPostingData *BTVacuumPosting;
920 :
921 : /*
922 : * BTScanOpaqueData is the btree-private state needed for an indexscan.
923 : * This consists of preprocessed scan keys (see _bt_preprocess_keys() for
924 : * details of the preprocessing), information about the current location
925 : * of the scan, and information about the marked location, if any. (We use
926 : * BTScanPosData to represent the data needed for each of current and marked
927 : * locations.) In addition we can remember some known-killed index entries
928 : * that must be marked before we can move off the current page.
929 : *
930 : * Index scans work a page at a time: we pin and read-lock the page, identify
931 : * all the matching items on the page and save them in BTScanPosData, then
932 : * release the read-lock while returning the items to the caller for
933 : * processing. This approach minimizes lock/unlock traffic. We must always
934 : * drop the lock to make it okay for caller to process the returned items.
935 : * Whether or not we can also release the pin during this window will vary.
936 : * We drop the pin eagerly (when safe) to avoid blocking progress by VACUUM
937 : * (see nbtree/README section about making concurrent TID recycling safe).
938 : * We'll always release both the lock and the pin on the current page before
939 : * moving on to its sibling page.
940 : *
941 : * If we are doing an index-only scan, we save the entire IndexTuple for each
942 : * matched item, otherwise only its heap TID and offset. The IndexTuples go
943 : * into a separate workspace array; each BTScanPosItem stores its tuple's
944 : * offset within that array. Posting list tuples store a "base" tuple once,
945 : * allowing the same key to be returned for each TID in the posting list
946 : * tuple.
947 : */
948 :
949 : typedef struct BTScanPosItem /* what we remember about each match */
950 : {
951 : ItemPointerData heapTid; /* TID of referenced heap item */
952 : OffsetNumber indexOffset; /* index item's location within page */
953 : LocationIndex tupleOffset; /* IndexTuple's offset in workspace, if any */
954 : } BTScanPosItem;
955 :
956 : typedef struct BTScanPosData
957 : {
958 : Buffer buf; /* currPage buf (invalid means unpinned) */
959 :
960 : /* page details as of the saved position's call to _bt_readpage */
961 : BlockNumber currPage; /* page referenced by items array */
962 : BlockNumber prevPage; /* currPage's left link */
963 : BlockNumber nextPage; /* currPage's right link */
964 : XLogRecPtr lsn; /* currPage's LSN */
965 :
966 : /* scan direction for the saved position's call to _bt_readpage */
967 : ScanDirection dir;
968 :
969 : /*
970 : * If we are doing an index-only scan, nextTupleOffset is the first free
971 : * location in the associated tuple storage workspace.
972 : */
973 : int nextTupleOffset;
974 :
975 : /*
976 : * moreLeft and moreRight track whether we think there may be matching
977 : * index entries to the left and right of the current page, respectively.
978 : */
979 : bool moreLeft;
980 : bool moreRight;
981 :
982 : /*
983 : * The items array is always ordered in index order (ie, increasing
984 : * indexoffset). When scanning backwards it is convenient to fill the
985 : * array back-to-front, so we start at the last slot and fill downwards.
986 : * Hence we need both a first-valid-entry and a last-valid-entry counter.
987 : * itemIndex is a cursor showing which entry was last returned to caller.
988 : */
989 : int firstItem; /* first valid index in items[] */
990 : int lastItem; /* last valid index in items[] */
991 : int itemIndex; /* current index in items[] */
992 :
993 : BTScanPosItem items[MaxTIDsPerBTreePage]; /* MUST BE LAST */
994 : } BTScanPosData;
995 :
996 : typedef BTScanPosData *BTScanPos;
997 :
998 : #define BTScanPosIsPinned(scanpos) \
999 : ( \
1000 : AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
1001 : !BufferIsValid((scanpos).buf)), \
1002 : BufferIsValid((scanpos).buf) \
1003 : )
1004 : #define BTScanPosUnpin(scanpos) \
1005 : do { \
1006 : ReleaseBuffer((scanpos).buf); \
1007 : (scanpos).buf = InvalidBuffer; \
1008 : } while (0)
1009 : #define BTScanPosUnpinIfPinned(scanpos) \
1010 : do { \
1011 : if (BTScanPosIsPinned(scanpos)) \
1012 : BTScanPosUnpin(scanpos); \
1013 : } while (0)
1014 :
1015 : #define BTScanPosIsValid(scanpos) \
1016 : ( \
1017 : AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
1018 : !BufferIsValid((scanpos).buf)), \
1019 : BlockNumberIsValid((scanpos).currPage) \
1020 : )
1021 : #define BTScanPosInvalidate(scanpos) \
1022 : do { \
1023 : (scanpos).buf = InvalidBuffer; \
1024 : (scanpos).currPage = InvalidBlockNumber; \
1025 : } while (0)
1026 :
1027 : /* We need one of these for each equality-type SK_SEARCHARRAY scan key */
1028 : typedef struct BTArrayKeyInfo
1029 : {
1030 : int scan_key; /* index of associated key in keyData */
1031 : int cur_elem; /* index of current element in elem_values */
1032 : int num_elems; /* number of elems in current array value */
1033 : Datum *elem_values; /* array of num_elems Datums */
1034 : } BTArrayKeyInfo;
1035 :
1036 : typedef struct BTScanOpaqueData
1037 : {
1038 : /* these fields are set by _bt_preprocess_keys(): */
1039 : bool qual_ok; /* false if qual can never be satisfied */
1040 : int numberOfKeys; /* number of preprocessed scan keys */
1041 : ScanKey keyData; /* array of preprocessed scan keys */
1042 :
1043 : /* workspace for SK_SEARCHARRAY support */
1044 : int numArrayKeys; /* number of equality-type array keys */
1045 : bool needPrimScan; /* New prim scan to continue in current dir? */
1046 : bool scanBehind; /* Check scan not still behind on next page? */
1047 : bool oppositeDirCheck; /* scanBehind opposite-scan-dir check? */
1048 : BTArrayKeyInfo *arrayKeys; /* info about each equality-type array key */
1049 : FmgrInfo *orderProcs; /* ORDER procs for required equality keys */
1050 : MemoryContext arrayContext; /* scan-lifespan context for array data */
1051 :
1052 : /* info about killed items if any (killedItems is NULL if never used) */
1053 : int *killedItems; /* currPos.items indexes of killed items */
1054 : int numKilled; /* number of currently stored items */
1055 :
1056 : /*
1057 : * If we are doing an index-only scan, these are the tuple storage
1058 : * workspaces for the currPos and markPos respectively. Each is of size
1059 : * BLCKSZ, so it can hold as much as a full page's worth of tuples.
1060 : */
1061 : char *currTuples; /* tuple storage for currPos */
1062 : char *markTuples; /* tuple storage for markPos */
1063 :
1064 : /*
1065 : * If the marked position is on the same page as current position, we
1066 : * don't use markPos, but just keep the marked itemIndex in markItemIndex
1067 : * (all the rest of currPos is valid for the mark position). Hence, to
1068 : * determine if there is a mark, first look at markItemIndex, then at
1069 : * markPos.
1070 : */
1071 : int markItemIndex; /* itemIndex, or -1 if not valid */
1072 :
1073 : /* keep these last in struct for efficiency */
1074 : BTScanPosData currPos; /* current position data */
1075 : BTScanPosData markPos; /* marked position, if any */
1076 : } BTScanOpaqueData;
1077 :
1078 : typedef BTScanOpaqueData *BTScanOpaque;
1079 :
1080 : /*
1081 : * _bt_readpage state used across _bt_checkkeys calls for a page
1082 : */
1083 : typedef struct BTReadPageState
1084 : {
1085 : /* Input parameters, set by _bt_readpage for _bt_checkkeys */
1086 : OffsetNumber minoff; /* Lowest non-pivot tuple's offset */
1087 : OffsetNumber maxoff; /* Highest non-pivot tuple's offset */
1088 : IndexTuple finaltup; /* Needed by scans with array keys */
1089 : Page page; /* Page being read */
1090 : bool firstpage; /* page is first for primitive scan? */
1091 :
1092 : /* Per-tuple input parameters, set by _bt_readpage for _bt_checkkeys */
1093 : OffsetNumber offnum; /* current tuple's page offset number */
1094 :
1095 : /* Output parameters, set by _bt_checkkeys for _bt_readpage */
1096 : OffsetNumber skip; /* Array keys "look ahead" skip offnum */
1097 : bool continuescan; /* Terminate ongoing (primitive) index scan? */
1098 :
1099 : /*
1100 : * Input and output parameters, set and unset by both _bt_readpage and
1101 : * _bt_checkkeys to manage precheck optimizations
1102 : */
1103 : bool prechecked; /* precheck set continuescan to 'true'? */
1104 : bool firstmatch; /* at least one match so far? */
1105 :
1106 : /*
1107 : * Private _bt_checkkeys state used to manage "look ahead" optimization
1108 : * (only used during scans with array keys)
1109 : */
1110 : int16 rechecks;
1111 : int16 targetdistance;
1112 :
1113 : } BTReadPageState;
1114 :
1115 : /*
1116 : * We use some private sk_flags bits in preprocessed scan keys. We're allowed
1117 : * to use bits 16-31 (see skey.h). The uppermost bits are copied from the
1118 : * index's indoption[] array entry for the index attribute.
1119 : */
1120 : #define SK_BT_REQFWD 0x00010000 /* required to continue forward scan */
1121 : #define SK_BT_REQBKWD 0x00020000 /* required to continue backward scan */
1122 : #define SK_BT_INDOPTION_SHIFT 24 /* must clear the above bits */
1123 : #define SK_BT_DESC (INDOPTION_DESC << SK_BT_INDOPTION_SHIFT)
1124 : #define SK_BT_NULLS_FIRST (INDOPTION_NULLS_FIRST << SK_BT_INDOPTION_SHIFT)
1125 :
1126 : typedef struct BTOptions
1127 : {
1128 : int32 varlena_header_; /* varlena header (do not touch directly!) */
1129 : int fillfactor; /* page fill factor in percent (0..100) */
1130 : float8 vacuum_cleanup_index_scale_factor; /* deprecated */
1131 : bool deduplicate_items; /* Try to deduplicate items? */
1132 : } BTOptions;
1133 :
1134 : #define BTGetFillFactor(relation) \
1135 : (AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
1136 : relation->rd_rel->relam == BTREE_AM_OID), \
1137 : (relation)->rd_options ? \
1138 : ((BTOptions *) (relation)->rd_options)->fillfactor : \
1139 : BTREE_DEFAULT_FILLFACTOR)
1140 : #define BTGetTargetPageFreeSpace(relation) \
1141 : (BLCKSZ * (100 - BTGetFillFactor(relation)) / 100)
1142 : #define BTGetDeduplicateItems(relation) \
1143 : (AssertMacro(relation->rd_rel->relkind == RELKIND_INDEX && \
1144 : relation->rd_rel->relam == BTREE_AM_OID), \
1145 : ((relation)->rd_options ? \
1146 : ((BTOptions *) (relation)->rd_options)->deduplicate_items : true))
1147 :
1148 : /*
1149 : * Constant definition for progress reporting. Phase numbers must match
1150 : * btbuildphasename.
1151 : */
1152 : /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 (see progress.h) */
1153 : #define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN 2
1154 : #define PROGRESS_BTREE_PHASE_PERFORMSORT_1 3
1155 : #define PROGRESS_BTREE_PHASE_PERFORMSORT_2 4
1156 : #define PROGRESS_BTREE_PHASE_LEAF_LOAD 5
1157 :
1158 : /*
1159 : * external entry points for btree, in nbtree.c
1160 : */
1161 : extern void btbuildempty(Relation index);
1162 : extern bool btinsert(Relation rel, Datum *values, bool *isnull,
1163 : ItemPointer ht_ctid, Relation heapRel,
1164 : IndexUniqueCheck checkUnique,
1165 : bool indexUnchanged,
1166 : struct IndexInfo *indexInfo);
1167 : extern IndexScanDesc btbeginscan(Relation rel, int nkeys, int norderbys);
1168 : extern Size btestimateparallelscan(int nkeys, int norderbys);
1169 : extern void btinitparallelscan(void *target);
1170 : extern bool btgettuple(IndexScanDesc scan, ScanDirection dir);
1171 : extern int64 btgetbitmap(IndexScanDesc scan, TIDBitmap *tbm);
1172 : extern void btrescan(IndexScanDesc scan, ScanKey scankey, int nscankeys,
1173 : ScanKey orderbys, int norderbys);
1174 : extern void btparallelrescan(IndexScanDesc scan);
1175 : extern void btendscan(IndexScanDesc scan);
1176 : extern void btmarkpos(IndexScanDesc scan);
1177 : extern void btrestrpos(IndexScanDesc scan);
1178 : extern IndexBulkDeleteResult *btbulkdelete(IndexVacuumInfo *info,
1179 : IndexBulkDeleteResult *stats,
1180 : IndexBulkDeleteCallback callback,
1181 : void *callback_state);
1182 : extern IndexBulkDeleteResult *btvacuumcleanup(IndexVacuumInfo *info,
1183 : IndexBulkDeleteResult *stats);
1184 : extern bool btcanreturn(Relation index, int attno);
1185 : extern int btgettreeheight(Relation rel);
1186 :
1187 : extern CompareType bttranslatestrategy(StrategyNumber strategy, Oid opfamily);
1188 : extern StrategyNumber bttranslatecmptype(CompareType cmptype, Oid opfamily);
1189 :
1190 : /*
1191 : * prototypes for internal functions in nbtree.c
1192 : */
1193 : extern bool _bt_parallel_seize(IndexScanDesc scan, BlockNumber *next_scan_page,
1194 : BlockNumber *last_curr_page, bool first);
1195 : extern void _bt_parallel_release(IndexScanDesc scan,
1196 : BlockNumber next_scan_page,
1197 : BlockNumber curr_page);
1198 : extern void _bt_parallel_done(IndexScanDesc scan);
1199 : extern void _bt_parallel_primscan_schedule(IndexScanDesc scan,
1200 : BlockNumber curr_page);
1201 :
1202 : /*
1203 : * prototypes for functions in nbtdedup.c
1204 : */
1205 : extern void _bt_dedup_pass(Relation rel, Buffer buf, IndexTuple newitem,
1206 : Size newitemsz, bool bottomupdedup);
1207 : extern bool _bt_bottomupdel_pass(Relation rel, Buffer buf, Relation heapRel,
1208 : Size newitemsz);
1209 : extern void _bt_dedup_start_pending(BTDedupState state, IndexTuple base,
1210 : OffsetNumber baseoff);
1211 : extern bool _bt_dedup_save_htid(BTDedupState state, IndexTuple itup);
1212 : extern Size _bt_dedup_finish_pending(Page newpage, BTDedupState state);
1213 : extern IndexTuple _bt_form_posting(IndexTuple base, ItemPointer htids,
1214 : int nhtids);
1215 : extern void _bt_update_posting(BTVacuumPosting vacposting);
1216 : extern IndexTuple _bt_swap_posting(IndexTuple newitem, IndexTuple oposting,
1217 : int postingoff);
1218 :
1219 : /*
1220 : * prototypes for functions in nbtinsert.c
1221 : */
1222 : extern bool _bt_doinsert(Relation rel, IndexTuple itup,
1223 : IndexUniqueCheck checkUnique, bool indexUnchanged,
1224 : Relation heapRel);
1225 : extern void _bt_finish_split(Relation rel, Relation heaprel, Buffer lbuf,
1226 : BTStack stack);
1227 : extern Buffer _bt_getstackbuf(Relation rel, Relation heaprel, BTStack stack,
1228 : BlockNumber child);
1229 :
1230 : /*
1231 : * prototypes for functions in nbtsplitloc.c
1232 : */
1233 : extern OffsetNumber _bt_findsplitloc(Relation rel, Page origpage,
1234 : OffsetNumber newitemoff, Size newitemsz, IndexTuple newitem,
1235 : bool *newitemonleft);
1236 :
1237 : /*
1238 : * prototypes for functions in nbtpage.c
1239 : */
1240 : extern void _bt_initmetapage(Page page, BlockNumber rootbknum, uint32 level,
1241 : bool allequalimage);
1242 : extern bool _bt_vacuum_needs_cleanup(Relation rel);
1243 : extern void _bt_set_cleanup_info(Relation rel, BlockNumber num_delpages);
1244 : extern void _bt_upgrademetapage(Page page);
1245 : extern Buffer _bt_getroot(Relation rel, Relation heaprel, int access);
1246 : extern Buffer _bt_gettrueroot(Relation rel);
1247 : extern int _bt_getrootheight(Relation rel);
1248 : extern void _bt_metaversion(Relation rel, bool *heapkeyspace,
1249 : bool *allequalimage);
1250 : extern void _bt_checkpage(Relation rel, Buffer buf);
1251 : extern Buffer _bt_getbuf(Relation rel, BlockNumber blkno, int access);
1252 : extern Buffer _bt_allocbuf(Relation rel, Relation heaprel);
1253 : extern Buffer _bt_relandgetbuf(Relation rel, Buffer obuf,
1254 : BlockNumber blkno, int access);
1255 : extern void _bt_relbuf(Relation rel, Buffer buf);
1256 : extern void _bt_lockbuf(Relation rel, Buffer buf, int access);
1257 : extern void _bt_unlockbuf(Relation rel, Buffer buf);
1258 : extern bool _bt_conditionallockbuf(Relation rel, Buffer buf);
1259 : extern void _bt_upgradelockbufcleanup(Relation rel, Buffer buf);
1260 : extern void _bt_pageinit(Page page, Size size);
1261 : extern void _bt_delitems_vacuum(Relation rel, Buffer buf,
1262 : OffsetNumber *deletable, int ndeletable,
1263 : BTVacuumPosting *updatable, int nupdatable);
1264 : extern void _bt_delitems_delete_check(Relation rel, Buffer buf,
1265 : Relation heapRel,
1266 : TM_IndexDeleteOp *delstate);
1267 : extern void _bt_pagedel(Relation rel, Buffer leafbuf, BTVacState *vstate);
1268 : extern void _bt_pendingfsm_init(Relation rel, BTVacState *vstate,
1269 : bool cleanuponly);
1270 : extern void _bt_pendingfsm_finalize(Relation rel, BTVacState *vstate);
1271 :
1272 : /*
1273 : * prototypes for functions in nbtpreprocesskeys.c
1274 : */
1275 : extern void _bt_preprocess_keys(IndexScanDesc scan);
1276 :
1277 : /*
1278 : * prototypes for functions in nbtsearch.c
1279 : */
1280 : extern BTStack _bt_search(Relation rel, Relation heaprel, BTScanInsert key,
1281 : Buffer *bufP, int access);
1282 : extern OffsetNumber _bt_binsrch_insert(Relation rel, BTInsertState insertstate);
1283 : extern int32 _bt_compare(Relation rel, BTScanInsert key, Page page, OffsetNumber offnum);
1284 : extern bool _bt_first(IndexScanDesc scan, ScanDirection dir);
1285 : extern bool _bt_next(IndexScanDesc scan, ScanDirection dir);
1286 : extern Buffer _bt_get_endpoint(Relation rel, uint32 level, bool rightmost);
1287 :
1288 : /*
1289 : * prototypes for functions in nbtutils.c
1290 : */
1291 : extern BTScanInsert _bt_mkscankey(Relation rel, IndexTuple itup);
1292 : extern void _bt_freestack(BTStack stack);
1293 : extern bool _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir);
1294 : extern int _bt_binsrch_array_skey(FmgrInfo *orderproc,
1295 : bool cur_elem_trig, ScanDirection dir,
1296 : Datum tupdatum, bool tupnull,
1297 : BTArrayKeyInfo *array, ScanKey cur,
1298 : int32 *set_elem_result);
1299 : extern void _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir);
1300 : extern bool _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys,
1301 : IndexTuple tuple, int tupnatts);
1302 : extern bool _bt_scanbehind_checkkeys(IndexScanDesc scan, ScanDirection dir,
1303 : IndexTuple finaltup);
1304 : extern void _bt_killitems(IndexScanDesc scan);
1305 : extern BTCycleId _bt_vacuum_cycleid(Relation rel);
1306 : extern BTCycleId _bt_start_vacuum(Relation rel);
1307 : extern void _bt_end_vacuum(Relation rel);
1308 : extern void _bt_end_vacuum_callback(int code, Datum arg);
1309 : extern Size BTreeShmemSize(void);
1310 : extern void BTreeShmemInit(void);
1311 : extern bytea *btoptions(Datum reloptions, bool validate);
1312 : extern bool btproperty(Oid index_oid, int attno,
1313 : IndexAMProperty prop, const char *propname,
1314 : bool *res, bool *isnull);
1315 : extern char *btbuildphasename(int64 phasenum);
1316 : extern IndexTuple _bt_truncate(Relation rel, IndexTuple lastleft,
1317 : IndexTuple firstright, BTScanInsert itup_key);
1318 : extern int _bt_keep_natts_fast(Relation rel, IndexTuple lastleft,
1319 : IndexTuple firstright);
1320 : extern bool _bt_check_natts(Relation rel, bool heapkeyspace, Page page,
1321 : OffsetNumber offnum);
1322 : extern void _bt_check_third_page(Relation rel, Relation heap,
1323 : bool needheaptidspace, Page page, IndexTuple newtup);
1324 : extern bool _bt_allequalimage(Relation rel, bool debugmessage);
1325 :
1326 : /*
1327 : * prototypes for functions in nbtvalidate.c
1328 : */
1329 : extern bool btvalidate(Oid opclassoid);
1330 : extern void btadjustmembers(Oid opfamilyoid,
1331 : Oid opclassoid,
1332 : List *operators,
1333 : List *functions);
1334 :
1335 : /*
1336 : * prototypes for functions in nbtsort.c
1337 : */
1338 : extern IndexBuildResult *btbuild(Relation heap, Relation index,
1339 : struct IndexInfo *indexInfo);
1340 : extern void _bt_parallel_build_main(dsm_segment *seg, shm_toc *toc);
1341 :
1342 : #endif /* NBTREE_H */
|
