| 1 | /* |
| 2 | * brin_xlog.c |
| 3 | * XLog replay routines for BRIN indexes |
| 4 | * |
| 5 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 6 | * Portions Copyright (c) 1994, Regents of the University of California |
| 7 | * |
| 8 | * IDENTIFICATION |
| 9 | * src/backend/access/brin/brin_xlog.c |
| 10 | */ |
| 11 | #include "postgres.h" |
| 12 | |
| 13 | #include "access/brin_page.h" |
| 14 | #include "access/brin_pageops.h" |
| 15 | #include "access/brin_xlog.h" |
| 16 | #include "access/bufmask.h" |
| 17 | #include "access/xlogutils.h" |
| 18 | |
| 19 | |
| 20 | /* |
| 21 | * xlog replay routines |
| 22 | */ |
| 23 | static void |
| 24 | brin_xlog_createidx(XLogReaderState *record) |
| 25 | { |
| 26 | XLogRecPtr lsn = record->EndRecPtr; |
| 27 | xl_brin_createidx *xlrec = (xl_brin_createidx *) XLogRecGetData(record); |
| 28 | Buffer buf; |
| 29 | Page page; |
| 30 | |
| 31 | /* create the index' metapage */ |
| 32 | buf = XLogInitBufferForRedo(record, 0); |
| 33 | Assert(BufferIsValid(buf)); |
| 34 | page = (Page) BufferGetPage(buf); |
| 35 | brin_metapage_init(page, xlrec->pagesPerRange, xlrec->version); |
| 36 | PageSetLSN(page, lsn); |
| 37 | MarkBufferDirty(buf); |
| 38 | UnlockReleaseBuffer(buf); |
| 39 | } |
| 40 | |
| 41 | /* |
| 42 | * Common part of an insert or update. Inserts the new tuple and updates the |
| 43 | * revmap. |
| 44 | */ |
| 45 | static void |
| 46 | brin_xlog_insert_update(XLogReaderState *record, |
| 47 | xl_brin_insert *xlrec) |
| 48 | { |
| 49 | XLogRecPtr lsn = record->EndRecPtr; |
| 50 | Buffer buffer; |
| 51 | BlockNumber regpgno; |
| 52 | Page page; |
| 53 | XLogRedoAction action; |
| 54 | |
| 55 | /* |
| 56 | * If we inserted the first and only tuple on the page, re-initialize the |
| 57 | * page from scratch. |
| 58 | */ |
| 59 | if (XLogRecGetInfo(record) & XLOG_BRIN_INIT_PAGE) |
| 60 | { |
| 61 | buffer = XLogInitBufferForRedo(record, 0); |
| 62 | page = BufferGetPage(buffer); |
| 63 | brin_page_init(page, BRIN_PAGETYPE_REGULAR); |
| 64 | action = BLK_NEEDS_REDO; |
| 65 | } |
| 66 | else |
| 67 | { |
| 68 | action = XLogReadBufferForRedo(record, 0, &buffer); |
| 69 | } |
| 70 | |
| 71 | /* need this page's blkno to store in revmap */ |
| 72 | regpgno = BufferGetBlockNumber(buffer); |
| 73 | |
| 74 | /* insert the index item into the page */ |
| 75 | if (action == BLK_NEEDS_REDO) |
| 76 | { |
| 77 | OffsetNumber offnum; |
| 78 | BrinTuple *tuple; |
| 79 | Size tuplen; |
| 80 | |
| 81 | tuple = (BrinTuple *) XLogRecGetBlockData(record, 0, &tuplen); |
| 82 | |
| 83 | Assert(tuple->bt_blkno == xlrec->heapBlk); |
| 84 | |
| 85 | page = (Page) BufferGetPage(buffer); |
| 86 | offnum = xlrec->offnum; |
| 87 | if (PageGetMaxOffsetNumber(page) + 1 < offnum) |
| 88 | elog(PANIC, "brin_xlog_insert_update: invalid max offset number" ); |
| 89 | |
| 90 | offnum = PageAddItem(page, (Item) tuple, tuplen, offnum, true, false); |
| 91 | if (offnum == InvalidOffsetNumber) |
| 92 | elog(PANIC, "brin_xlog_insert_update: failed to add tuple" ); |
| 93 | |
| 94 | PageSetLSN(page, lsn); |
| 95 | MarkBufferDirty(buffer); |
| 96 | } |
| 97 | if (BufferIsValid(buffer)) |
| 98 | UnlockReleaseBuffer(buffer); |
| 99 | |
| 100 | /* update the revmap */ |
| 101 | action = XLogReadBufferForRedo(record, 1, &buffer); |
| 102 | if (action == BLK_NEEDS_REDO) |
| 103 | { |
| 104 | ItemPointerData tid; |
| 105 | |
| 106 | ItemPointerSet(&tid, regpgno, xlrec->offnum); |
| 107 | page = (Page) BufferGetPage(buffer); |
| 108 | |
| 109 | brinSetHeapBlockItemptr(buffer, xlrec->pagesPerRange, xlrec->heapBlk, |
| 110 | tid); |
| 111 | PageSetLSN(page, lsn); |
| 112 | MarkBufferDirty(buffer); |
| 113 | } |
| 114 | if (BufferIsValid(buffer)) |
| 115 | UnlockReleaseBuffer(buffer); |
| 116 | |
| 117 | /* XXX no FSM updates here ... */ |
| 118 | } |
| 119 | |
| 120 | /* |
| 121 | * replay a BRIN index insertion |
| 122 | */ |
| 123 | static void |
| 124 | brin_xlog_insert(XLogReaderState *record) |
| 125 | { |
| 126 | xl_brin_insert *xlrec = (xl_brin_insert *) XLogRecGetData(record); |
| 127 | |
| 128 | brin_xlog_insert_update(record, xlrec); |
| 129 | } |
| 130 | |
| 131 | /* |
| 132 | * replay a BRIN index update |
| 133 | */ |
| 134 | static void |
| 135 | brin_xlog_update(XLogReaderState *record) |
| 136 | { |
| 137 | XLogRecPtr lsn = record->EndRecPtr; |
| 138 | xl_brin_update *xlrec = (xl_brin_update *) XLogRecGetData(record); |
| 139 | Buffer buffer; |
| 140 | XLogRedoAction action; |
| 141 | |
| 142 | /* First remove the old tuple */ |
| 143 | action = XLogReadBufferForRedo(record, 2, &buffer); |
| 144 | if (action == BLK_NEEDS_REDO) |
| 145 | { |
| 146 | Page page; |
| 147 | OffsetNumber offnum; |
| 148 | |
| 149 | page = (Page) BufferGetPage(buffer); |
| 150 | |
| 151 | offnum = xlrec->oldOffnum; |
| 152 | |
| 153 | PageIndexTupleDeleteNoCompact(page, offnum); |
| 154 | |
| 155 | PageSetLSN(page, lsn); |
| 156 | MarkBufferDirty(buffer); |
| 157 | } |
| 158 | |
| 159 | /* Then insert the new tuple and update revmap, like in an insertion. */ |
| 160 | brin_xlog_insert_update(record, &xlrec->insert); |
| 161 | |
| 162 | if (BufferIsValid(buffer)) |
| 163 | UnlockReleaseBuffer(buffer); |
| 164 | } |
| 165 | |
| 166 | /* |
| 167 | * Update a tuple on a single page. |
| 168 | */ |
| 169 | static void |
| 170 | brin_xlog_samepage_update(XLogReaderState *record) |
| 171 | { |
| 172 | XLogRecPtr lsn = record->EndRecPtr; |
| 173 | xl_brin_samepage_update *xlrec; |
| 174 | Buffer buffer; |
| 175 | XLogRedoAction action; |
| 176 | |
| 177 | xlrec = (xl_brin_samepage_update *) XLogRecGetData(record); |
| 178 | action = XLogReadBufferForRedo(record, 0, &buffer); |
| 179 | if (action == BLK_NEEDS_REDO) |
| 180 | { |
| 181 | Size tuplen; |
| 182 | BrinTuple *brintuple; |
| 183 | Page page; |
| 184 | OffsetNumber offnum; |
| 185 | |
| 186 | brintuple = (BrinTuple *) XLogRecGetBlockData(record, 0, &tuplen); |
| 187 | |
| 188 | page = (Page) BufferGetPage(buffer); |
| 189 | |
| 190 | offnum = xlrec->offnum; |
| 191 | |
| 192 | if (!PageIndexTupleOverwrite(page, offnum, (Item) brintuple, tuplen)) |
| 193 | elog(PANIC, "brin_xlog_samepage_update: failed to replace tuple" ); |
| 194 | |
| 195 | PageSetLSN(page, lsn); |
| 196 | MarkBufferDirty(buffer); |
| 197 | } |
| 198 | if (BufferIsValid(buffer)) |
| 199 | UnlockReleaseBuffer(buffer); |
| 200 | |
| 201 | /* XXX no FSM updates here ... */ |
| 202 | } |
| 203 | |
| 204 | /* |
| 205 | * Replay a revmap page extension |
| 206 | */ |
| 207 | static void |
| 208 | brin_xlog_revmap_extend(XLogReaderState *record) |
| 209 | { |
| 210 | XLogRecPtr lsn = record->EndRecPtr; |
| 211 | xl_brin_revmap_extend *xlrec; |
| 212 | Buffer metabuf; |
| 213 | Buffer buf; |
| 214 | Page page; |
| 215 | BlockNumber targetBlk; |
| 216 | XLogRedoAction action; |
| 217 | |
| 218 | xlrec = (xl_brin_revmap_extend *) XLogRecGetData(record); |
| 219 | |
| 220 | XLogRecGetBlockTag(record, 1, NULL, NULL, &targetBlk); |
| 221 | Assert(xlrec->targetBlk == targetBlk); |
| 222 | |
| 223 | /* Update the metapage */ |
| 224 | action = XLogReadBufferForRedo(record, 0, &metabuf); |
| 225 | if (action == BLK_NEEDS_REDO) |
| 226 | { |
| 227 | Page metapg; |
| 228 | BrinMetaPageData *metadata; |
| 229 | |
| 230 | metapg = BufferGetPage(metabuf); |
| 231 | metadata = (BrinMetaPageData *) PageGetContents(metapg); |
| 232 | |
| 233 | Assert(metadata->lastRevmapPage == xlrec->targetBlk - 1); |
| 234 | metadata->lastRevmapPage = xlrec->targetBlk; |
| 235 | |
| 236 | PageSetLSN(metapg, lsn); |
| 237 | |
| 238 | /* |
| 239 | * Set pd_lower just past the end of the metadata. This is essential, |
| 240 | * because without doing so, metadata will be lost if xlog.c |
| 241 | * compresses the page. (We must do this here because pre-v11 |
| 242 | * versions of PG did not set the metapage's pd_lower correctly, so a |
| 243 | * pg_upgraded index might contain the wrong value.) |
| 244 | */ |
| 245 | ((PageHeader) metapg)->pd_lower = |
| 246 | ((char *) metadata + sizeof(BrinMetaPageData)) - (char *) metapg; |
| 247 | |
| 248 | MarkBufferDirty(metabuf); |
| 249 | } |
| 250 | |
| 251 | /* |
| 252 | * Re-init the target block as a revmap page. There's never a full- page |
| 253 | * image here. |
| 254 | */ |
| 255 | |
| 256 | buf = XLogInitBufferForRedo(record, 1); |
| 257 | page = (Page) BufferGetPage(buf); |
| 258 | brin_page_init(page, BRIN_PAGETYPE_REVMAP); |
| 259 | |
| 260 | PageSetLSN(page, lsn); |
| 261 | MarkBufferDirty(buf); |
| 262 | |
| 263 | UnlockReleaseBuffer(buf); |
| 264 | if (BufferIsValid(metabuf)) |
| 265 | UnlockReleaseBuffer(metabuf); |
| 266 | } |
| 267 | |
| 268 | static void |
| 269 | brin_xlog_desummarize_page(XLogReaderState *record) |
| 270 | { |
| 271 | XLogRecPtr lsn = record->EndRecPtr; |
| 272 | xl_brin_desummarize *xlrec; |
| 273 | Buffer buffer; |
| 274 | XLogRedoAction action; |
| 275 | |
| 276 | xlrec = (xl_brin_desummarize *) XLogRecGetData(record); |
| 277 | |
| 278 | /* Update the revmap */ |
| 279 | action = XLogReadBufferForRedo(record, 0, &buffer); |
| 280 | if (action == BLK_NEEDS_REDO) |
| 281 | { |
| 282 | ItemPointerData iptr; |
| 283 | |
| 284 | ItemPointerSetInvalid(&iptr); |
| 285 | brinSetHeapBlockItemptr(buffer, xlrec->pagesPerRange, xlrec->heapBlk, iptr); |
| 286 | |
| 287 | PageSetLSN(BufferGetPage(buffer), lsn); |
| 288 | MarkBufferDirty(buffer); |
| 289 | } |
| 290 | if (BufferIsValid(buffer)) |
| 291 | UnlockReleaseBuffer(buffer); |
| 292 | |
| 293 | /* remove the leftover entry from the regular page */ |
| 294 | action = XLogReadBufferForRedo(record, 1, &buffer); |
| 295 | if (action == BLK_NEEDS_REDO) |
| 296 | { |
| 297 | Page regPg = BufferGetPage(buffer); |
| 298 | |
| 299 | PageIndexTupleDeleteNoCompact(regPg, xlrec->regOffset); |
| 300 | |
| 301 | PageSetLSN(regPg, lsn); |
| 302 | MarkBufferDirty(buffer); |
| 303 | } |
| 304 | if (BufferIsValid(buffer)) |
| 305 | UnlockReleaseBuffer(buffer); |
| 306 | } |
| 307 | |
| 308 | void |
| 309 | brin_redo(XLogReaderState *record) |
| 310 | { |
| 311 | uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK; |
| 312 | |
| 313 | switch (info & XLOG_BRIN_OPMASK) |
| 314 | { |
| 315 | case XLOG_BRIN_CREATE_INDEX: |
| 316 | brin_xlog_createidx(record); |
| 317 | break; |
| 318 | case XLOG_BRIN_INSERT: |
| 319 | brin_xlog_insert(record); |
| 320 | break; |
| 321 | case XLOG_BRIN_UPDATE: |
| 322 | brin_xlog_update(record); |
| 323 | break; |
| 324 | case XLOG_BRIN_SAMEPAGE_UPDATE: |
| 325 | brin_xlog_samepage_update(record); |
| 326 | break; |
| 327 | case XLOG_BRIN_REVMAP_EXTEND: |
| 328 | brin_xlog_revmap_extend(record); |
| 329 | break; |
| 330 | case XLOG_BRIN_DESUMMARIZE: |
| 331 | brin_xlog_desummarize_page(record); |
| 332 | break; |
| 333 | default: |
| 334 | elog(PANIC, "brin_redo: unknown op code %u" , info); |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Mask a BRIN page before doing consistency checks. |
| 340 | */ |
| 341 | void |
| 342 | brin_mask(char *pagedata, BlockNumber blkno) |
| 343 | { |
| 344 | Page page = (Page) pagedata; |
| 345 | PageHeader pagehdr = (PageHeader) page; |
| 346 | |
| 347 | mask_page_lsn_and_checksum(page); |
| 348 | |
| 349 | mask_page_hint_bits(page); |
| 350 | |
| 351 | /* |
| 352 | * Regular brin pages contain unused space which needs to be masked. |
| 353 | * Similarly for meta pages, but mask it only if pd_lower appears to have |
| 354 | * been set correctly. |
| 355 | */ |
| 356 | if (BRIN_IS_REGULAR_PAGE(page) || |
| 357 | (BRIN_IS_META_PAGE(page) && pagehdr->pd_lower > SizeOfPageHeaderData)) |
| 358 | { |
| 359 | mask_unused_space(page); |
| 360 | } |
| 361 | } |
| 362 | |