granicus.if.org Git - postgresql/blob - src/backend/replication/logical/reorderbuffer.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * reorderbuffer.c
   4  *        PostgreSQL logical replay/reorder buffer management
   5  *
   6  *
   7  * Copyright (c) 2012-2018, PostgreSQL Global Development Group
   8  *
   9  *
  10  * IDENTIFICATION
  11  *        src/backend/replication/reorderbuffer.c
  12  *
  13  * NOTES
  14  *        This module gets handed individual pieces of transactions in the order
  15  *        they are written to the WAL and is responsible to reassemble them into
  16  *        toplevel transaction sized pieces. When a transaction is completely
  17  *        reassembled - signalled by reading the transaction commit record - it
  18  *        will then call the output plugin (c.f. ReorderBufferCommit()) with the
  19  *        individual changes. The output plugins rely on snapshots built by
  20  *        snapbuild.c which hands them to us.
  21  *
  22  *        Transactions and subtransactions/savepoints in postgres are not
  23  *        immediately linked to each other from outside the performing
  24  *        backend. Only at commit/abort (or special xact_assignment records) they
  25  *        are linked together. Which means that we will have to splice together a
  26  *        toplevel transaction from its subtransactions. To do that efficiently we
  27  *        build a binary heap indexed by the smallest current lsn of the individual
  28  *        subtransactions' changestreams. As the individual streams are inherently
  29  *        ordered by LSN - since that is where we build them from - the transaction
  30  *        can easily be reassembled by always using the subtransaction with the
  31  *        smallest current LSN from the heap.
  32  *
  33  *        In order to cope with large transactions - which can be several times as
  34  *        big as the available memory - this module supports spooling the contents
  35  *        of a large transactions to disk. When the transaction is replayed the
  36  *        contents of individual (sub-)transactions will be read from disk in
  37  *        chunks.
  38  *
  39  *        This module also has to deal with reassembling toast records from the
  40  *        individual chunks stored in WAL. When a new (or initial) version of a
  41  *        tuple is stored in WAL it will always be preceded by the toast chunks
  42  *        emitted for the columns stored out of line. Within a single toplevel
  43  *        transaction there will be no other data carrying records between a row's
  44  *        toast chunks and the row data itself. See ReorderBufferToast* for
  45  *        details.
  46  *
  47  *        ReorderBuffer uses two special memory context types - SlabContext for
  48  *        allocations of fixed-length structures (changes and transactions), and
  49  *        GenerationContext for the variable-length transaction data (allocated
  50  *        and freed in groups with similar lifespan).
  51  *
  52  * -------------------------------------------------------------------------
  53  */
  54 #include "postgres.h"
  55
  56 #include <unistd.h>
  57 #include <sys/stat.h>
  58
  59 #include "access/rewriteheap.h"
  60 #include "access/transam.h"
  61 #include "access/tuptoaster.h"
  62 #include "access/xact.h"
  63 #include "access/xlog_internal.h"
  64 #include "catalog/catalog.h"
  65 #include "lib/binaryheap.h"
  66 #include "miscadmin.h"
  67 #include "pgstat.h"
  68 #include "replication/logical.h"
  69 #include "replication/reorderbuffer.h"
  70 #include "replication/slot.h"
  71 #include "replication/snapbuild.h"      /* just for SnapBuildSnapDecRefcount */
  72 #include "storage/bufmgr.h"
  73 #include "storage/fd.h"
  74 #include "storage/sinval.h"
  75 #include "utils/builtins.h"
  76 #include "utils/combocid.h"
  77 #include "utils/memdebug.h"
  78 #include "utils/memutils.h"
  79 #include "utils/rel.h"
  80 #include "utils/relfilenodemap.h"
  81 #include "utils/tqual.h"
  82
  83
  84 /* entry for a hash table we use to map from xid to our transaction state */
  85 typedef struct ReorderBufferTXNByIdEnt
  86 {
  87         TransactionId xid;
  88         ReorderBufferTXN *txn;
  89 } ReorderBufferTXNByIdEnt;
  90
  91 /* data structures for (relfilenode, ctid) => (cmin, cmax) mapping */
  92 typedef struct ReorderBufferTupleCidKey
  93 {
  94         RelFileNode relnode;
  95         ItemPointerData tid;
  96 } ReorderBufferTupleCidKey;
  97
  98 typedef struct ReorderBufferTupleCidEnt
  99 {
 100         ReorderBufferTupleCidKey key;
 101         CommandId       cmin;
 102         CommandId       cmax;
 103         CommandId       combocid;               /* just for debugging */
 104 } ReorderBufferTupleCidEnt;
 105
 106 /* k-way in-order change iteration support structures */
 107 typedef struct ReorderBufferIterTXNEntry
 108 {
 109         XLogRecPtr      lsn;
 110         ReorderBufferChange *change;
 111         ReorderBufferTXN *txn;
 112         int                     fd;
 113         XLogSegNo       segno;
 114 } ReorderBufferIterTXNEntry;
 115
 116 typedef struct ReorderBufferIterTXNState
 117 {
 118         binaryheap *heap;
 119         Size            nr_txns;
 120         dlist_head      old_change;
 121         ReorderBufferIterTXNEntry entries[FLEXIBLE_ARRAY_MEMBER];
 122 } ReorderBufferIterTXNState;
 123
 124 /* toast datastructures */
 125 typedef struct ReorderBufferToastEnt
 126 {
 127         Oid                     chunk_id;               /* toast_table.chunk_id */
 128         int32           last_chunk_seq; /* toast_table.chunk_seq of the last chunk we
 129                                                                  * have seen */
 130         Size            num_chunks;             /* number of chunks we've already seen */
 131         Size            size;                   /* combined size of chunks seen */
 132         dlist_head      chunks;                 /* linked list of chunks */
 133         struct varlena *reconstructed;  /* reconstructed varlena now pointed to in
 134                                                                          * main tup */
 135 } ReorderBufferToastEnt;
 136
 137 /* Disk serialization support datastructures */
 138 typedef struct ReorderBufferDiskChange
 139 {
 140         Size            size;
 141         ReorderBufferChange change;
 142         /* data follows */
 143 } ReorderBufferDiskChange;
 144
 145 /*
 146  * Maximum number of changes kept in memory, per transaction. After that,
 147  * changes are spooled to disk.
 148  *
 149  * The current value should be sufficient to decode the entire transaction
 150  * without hitting disk in OLTP workloads, while starting to spool to disk in
 151  * other workloads reasonably fast.
 152  *
 153  * At some point in the future it probably makes sense to have a more elaborate
 154  * resource management here, but it's not entirely clear what that would look
 155  * like.
 156  */
 157 static const Size max_changes_in_memory = 4096;
 158
 159 /* ---------------------------------------
 160  * primary reorderbuffer support routines
 161  * ---------------------------------------
 162  */
 163 static ReorderBufferTXN *ReorderBufferGetTXN(ReorderBuffer *rb);
 164 static void ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
 165 static ReorderBufferTXN *ReorderBufferTXNByXid(ReorderBuffer *rb,
 166                                           TransactionId xid, bool create, bool *is_new,
 167                                           XLogRecPtr lsn, bool create_as_top);
 168
 169 static void AssertTXNLsnOrder(ReorderBuffer *rb);
 170
 171 /* ---------------------------------------
 172  * support functions for lsn-order iterating over the ->changes of a
 173  * transaction and its subtransactions
 174  *
 175  * used for iteration over the k-way heap merge of a transaction and its
 176  * subtransactions
 177  * ---------------------------------------
 178  */
 179 static ReorderBufferIterTXNState *ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn);
 180 static ReorderBufferChange *ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state);
 181 static void ReorderBufferIterTXNFinish(ReorderBuffer *rb,
 182                                                    ReorderBufferIterTXNState *state);
 183 static void ReorderBufferExecuteInvalidations(ReorderBuffer *rb, ReorderBufferTXN *txn);
 184
 185 /*
 186  * ---------------------------------------
 187  * Disk serialization support functions
 188  * ---------------------------------------
 189  */
 190 static void ReorderBufferCheckSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
 191 static void ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
 192 static void ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
 193                                                          int fd, ReorderBufferChange *change);
 194 static Size ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
 195                                                         int *fd, XLogSegNo *segno);
 196 static void ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
 197                                                    char *change);
 198 static void ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn);
 199
 200 static void ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap);
 201 static Snapshot ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
 202                                           ReorderBufferTXN *txn, CommandId cid);
 203
 204 /* ---------------------------------------
 205  * toast reassembly support
 206  * ---------------------------------------
 207  */
 208 static void ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn);
 209 static void ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn);
 210 static void ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
 211                                                   Relation relation, ReorderBufferChange *change);
 212 static void ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
 213                                                           Relation relation, ReorderBufferChange *change);
 214
 215
 216 /*
 217  * Allocate a new ReorderBuffer
 218  */
 219 ReorderBuffer *
 220 ReorderBufferAllocate(void)
 221 {
 222         ReorderBuffer *buffer;
 223         HASHCTL         hash_ctl;
 224         MemoryContext new_ctx;
 225
 226         /* allocate memory in own context, to have better accountability */
 227         new_ctx = AllocSetContextCreate(CurrentMemoryContext,
 228                                                                         "ReorderBuffer",
 229                                                                         ALLOCSET_DEFAULT_SIZES);
 230
 231         buffer =
 232                 (ReorderBuffer *) MemoryContextAlloc(new_ctx, sizeof(ReorderBuffer));
 233
 234         memset(&hash_ctl, 0, sizeof(hash_ctl));
 235
 236         buffer->context = new_ctx;
 237
 238         buffer->change_context = SlabContextCreate(new_ctx,
 239                                                                                            "Change",
 240                                                                                            0,
 241                                                                                            SLAB_DEFAULT_BLOCK_SIZE,
 242                                                                                            sizeof(ReorderBufferChange));
 243
 244         buffer->txn_context = SlabContextCreate(new_ctx,
 245                                                                                         "TXN",
 246                                                                                         0,
 247                                                                                         SLAB_DEFAULT_BLOCK_SIZE,
 248                                                                                         sizeof(ReorderBufferTXN));
 249
 250         buffer->tup_context = GenerationContextCreate(new_ctx,
 251                                                                                                   "Tuples",
 252                                                                                                   0,
 253                                                                                                   SLAB_LARGE_BLOCK_SIZE);
 254
 255         hash_ctl.keysize = sizeof(TransactionId);
 256         hash_ctl.entrysize = sizeof(ReorderBufferTXNByIdEnt);
 257         hash_ctl.hcxt = buffer->context;
 258
 259         buffer->by_txn = hash_create("ReorderBufferByXid", 1000, &hash_ctl,
 260                                                                  HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
 261
 262         buffer->by_txn_last_xid = InvalidTransactionId;
 263         buffer->by_txn_last_txn = NULL;
 264
 265         buffer->outbuf = NULL;
 266         buffer->outbufsize = 0;
 267
 268         buffer->current_restart_decoding_lsn = InvalidXLogRecPtr;
 269
 270         dlist_init(&buffer->toplevel_by_lsn);
 271
 272         return buffer;
 273 }
 274
 275 /*
 276  * Free a ReorderBuffer
 277  */
 278 void
 279 ReorderBufferFree(ReorderBuffer *rb)
 280 {
 281         MemoryContext context = rb->context;
 282
 283         /*
 284          * We free separately allocated data by entirely scrapping reorderbuffer's
 285          * memory context.
 286          */
 287         MemoryContextDelete(context);
 288 }
 289
 290 /*
 291  * Get an unused, possibly preallocated, ReorderBufferTXN.
 292  */
 293 static ReorderBufferTXN *
 294 ReorderBufferGetTXN(ReorderBuffer *rb)
 295 {
 296         ReorderBufferTXN *txn;
 297
 298         txn = (ReorderBufferTXN *)
 299                 MemoryContextAlloc(rb->txn_context, sizeof(ReorderBufferTXN));
 300
 301         memset(txn, 0, sizeof(ReorderBufferTXN));
 302
 303         dlist_init(&txn->changes);
 304         dlist_init(&txn->tuplecids);
 305         dlist_init(&txn->subtxns);
 306
 307         return txn;
 308 }
 309
 310 /*
 311  * Free a ReorderBufferTXN.
 312  *
 313  * Deallocation might be delayed for efficiency purposes, for details check
 314  * the comments above max_cached_changes's definition.
 315  */
 316 static void
 317 ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
 318 {
 319         /* clean the lookup cache if we were cached (quite likely) */
 320         if (rb->by_txn_last_xid == txn->xid)
 321         {
 322                 rb->by_txn_last_xid = InvalidTransactionId;
 323                 rb->by_txn_last_txn = NULL;
 324         }
 325
 326         /* free data that's contained */
 327
 328         if (txn->tuplecid_hash != NULL)
 329         {
 330                 hash_destroy(txn->tuplecid_hash);
 331                 txn->tuplecid_hash = NULL;
 332         }
 333
 334         if (txn->invalidations)
 335         {
 336                 pfree(txn->invalidations);
 337                 txn->invalidations = NULL;
 338         }
 339
 340         pfree(txn);
 341 }
 342
 343 /*
 344  * Get an unused, possibly preallocated, ReorderBufferChange.
 345  */
 346 ReorderBufferChange *
 347 ReorderBufferGetChange(ReorderBuffer *rb)
 348 {
 349         ReorderBufferChange *change;
 350
 351         change = (ReorderBufferChange *)
 352                 MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));
 353
 354         memset(change, 0, sizeof(ReorderBufferChange));
 355         return change;
 356 }
 357
 358 /*
 359  * Free an ReorderBufferChange.
 360  *
 361  * Deallocation might be delayed for efficiency purposes, for details check
 362  * the comments above max_cached_changes's definition.
 363  */
 364 void
 365 ReorderBufferReturnChange(ReorderBuffer *rb, ReorderBufferChange *change)
 366 {
 367         /* free contained data */
 368         switch (change->action)
 369         {
 370                 case REORDER_BUFFER_CHANGE_INSERT:
 371                 case REORDER_BUFFER_CHANGE_UPDATE:
 372                 case REORDER_BUFFER_CHANGE_DELETE:
 373                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
 374                         if (change->data.tp.newtuple)
 375                         {
 376                                 ReorderBufferReturnTupleBuf(rb, change->data.tp.newtuple);
 377                                 change->data.tp.newtuple = NULL;
 378                         }
 379
 380                         if (change->data.tp.oldtuple)
 381                         {
 382                                 ReorderBufferReturnTupleBuf(rb, change->data.tp.oldtuple);
 383                                 change->data.tp.oldtuple = NULL;
 384                         }
 385                         break;
 386                 case REORDER_BUFFER_CHANGE_MESSAGE:
 387                         if (change->data.msg.prefix != NULL)
 388                                 pfree(change->data.msg.prefix);
 389                         change->data.msg.prefix = NULL;
 390                         if (change->data.msg.message != NULL)
 391                                 pfree(change->data.msg.message);
 392                         change->data.msg.message = NULL;
 393                         break;
 394                 case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
 395                         if (change->data.snapshot)
 396                         {
 397                                 ReorderBufferFreeSnap(rb, change->data.snapshot);
 398                                 change->data.snapshot = NULL;
 399                         }
 400                         break;
 401                         /* no data in addition to the struct itself */
 402                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
 403                 case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
 404                 case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
 405                         break;
 406         }
 407
 408         pfree(change);
 409 }
 410
 411 /*
 412  * Get an unused, possibly preallocated, ReorderBufferTupleBuf fitting at
 413  * least a tuple of size tuple_len (excluding header overhead).
 414  */
 415 ReorderBufferTupleBuf *
 416 ReorderBufferGetTupleBuf(ReorderBuffer *rb, Size tuple_len)
 417 {
 418         ReorderBufferTupleBuf *tuple;
 419         Size            alloc_len;
 420
 421         alloc_len = tuple_len + SizeofHeapTupleHeader;
 422
 423         tuple = (ReorderBufferTupleBuf *)
 424                 MemoryContextAlloc(rb->tup_context,
 425                                                    sizeof(ReorderBufferTupleBuf) +
 426                                                    MAXIMUM_ALIGNOF + alloc_len);
 427         tuple->alloc_tuple_size = alloc_len;
 428         tuple->tuple.t_data = ReorderBufferTupleBufData(tuple);
 429
 430         return tuple;
 431 }
 432
 433 /*
 434  * Free an ReorderBufferTupleBuf.
 435  *
 436  * Deallocation might be delayed for efficiency purposes, for details check
 437  * the comments above max_cached_changes's definition.
 438  */
 439 void
 440 ReorderBufferReturnTupleBuf(ReorderBuffer *rb, ReorderBufferTupleBuf *tuple)
 441 {
 442         pfree(tuple);
 443 }
 444
 445 /*
 446  * Return the ReorderBufferTXN from the given buffer, specified by Xid.
 447  * If create is true, and a transaction doesn't already exist, create it
 448  * (with the given LSN, and as top transaction if that's specified);
 449  * when this happens, is_new is set to true.
 450  */
 451 static ReorderBufferTXN *
 452 ReorderBufferTXNByXid(ReorderBuffer *rb, TransactionId xid, bool create,
 453                                           bool *is_new, XLogRecPtr lsn, bool create_as_top)
 454 {
 455         ReorderBufferTXN *txn;
 456         ReorderBufferTXNByIdEnt *ent;
 457         bool            found;
 458
 459         Assert(TransactionIdIsValid(xid));
 460         Assert(!create || lsn != InvalidXLogRecPtr);
 461
 462         /*
 463          * Check the one-entry lookup cache first
 464          */
 465         if (TransactionIdIsValid(rb->by_txn_last_xid) &&
 466                 rb->by_txn_last_xid == xid)
 467         {
 468                 txn = rb->by_txn_last_txn;
 469
 470                 if (txn != NULL)
 471                 {
 472                         /* found it, and it's valid */
 473                         if (is_new)
 474                                 *is_new = false;
 475                         return txn;
 476                 }
 477
 478                 /*
 479                  * cached as non-existent, and asked not to create? Then nothing else
 480                  * to do.
 481                  */
 482                 if (!create)
 483                         return NULL;
 484                 /* otherwise fall through to create it */
 485         }
 486
 487         /*
 488          * If the cache wasn't hit or it yielded an "does-not-exist" and we want
 489          * to create an entry.
 490          */
 491
 492         /* search the lookup table */
 493         ent = (ReorderBufferTXNByIdEnt *)
 494                 hash_search(rb->by_txn,
 495                                         (void *) &xid,
 496                                         create ? HASH_ENTER : HASH_FIND,
 497                                         &found);
 498         if (found)
 499                 txn = ent->txn;
 500         else if (create)
 501         {
 502                 /* initialize the new entry, if creation was requested */
 503                 Assert(ent != NULL);
 504
 505                 ent->txn = ReorderBufferGetTXN(rb);
 506                 ent->txn->xid = xid;
 507                 txn = ent->txn;
 508                 txn->first_lsn = lsn;
 509                 txn->restart_decoding_lsn = rb->current_restart_decoding_lsn;
 510
 511                 if (create_as_top)
 512                 {
 513                         dlist_push_tail(&rb->toplevel_by_lsn, &txn->node);
 514                         AssertTXNLsnOrder(rb);
 515                 }
 516         }
 517         else
 518                 txn = NULL;                             /* not found and not asked to create */
 519
 520         /* update cache */
 521         rb->by_txn_last_xid = xid;
 522         rb->by_txn_last_txn = txn;
 523
 524         if (is_new)
 525                 *is_new = !found;
 526
 527         Assert(!create || txn != NULL);
 528         return txn;
 529 }
 530
 531 /*
 532  * Queue a change into a transaction so it can be replayed upon commit.
 533  */
 534 void
 535 ReorderBufferQueueChange(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
 536                                                  ReorderBufferChange *change)
 537 {
 538         ReorderBufferTXN *txn;
 539
 540         txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
 541
 542         change->lsn = lsn;
 543         Assert(InvalidXLogRecPtr != lsn);
 544         dlist_push_tail(&txn->changes, &change->node);
 545         txn->nentries++;
 546         txn->nentries_mem++;
 547
 548         ReorderBufferCheckSerializeTXN(rb, txn);
 549 }
 550
 551 /*
 552  * Queue message into a transaction so it can be processed upon commit.
 553  */
 554 void
 555 ReorderBufferQueueMessage(ReorderBuffer *rb, TransactionId xid,
 556                                                   Snapshot snapshot, XLogRecPtr lsn,
 557                                                   bool transactional, const char *prefix,
 558                                                   Size message_size, const char *message)
 559 {
 560         if (transactional)
 561         {
 562                 MemoryContext oldcontext;
 563                 ReorderBufferChange *change;
 564
 565                 Assert(xid != InvalidTransactionId);
 566
 567                 oldcontext = MemoryContextSwitchTo(rb->context);
 568
 569                 change = ReorderBufferGetChange(rb);
 570                 change->action = REORDER_BUFFER_CHANGE_MESSAGE;
 571                 change->data.msg.prefix = pstrdup(prefix);
 572                 change->data.msg.message_size = message_size;
 573                 change->data.msg.message = palloc(message_size);
 574                 memcpy(change->data.msg.message, message, message_size);
 575
 576                 ReorderBufferQueueChange(rb, xid, lsn, change);
 577
 578                 MemoryContextSwitchTo(oldcontext);
 579         }
 580         else
 581         {
 582                 ReorderBufferTXN *txn = NULL;
 583                 volatile Snapshot snapshot_now = snapshot;
 584
 585                 if (xid != InvalidTransactionId)
 586                         txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
 587
 588                 /* setup snapshot to allow catalog access */
 589                 SetupHistoricSnapshot(snapshot_now, NULL);
 590                 PG_TRY();
 591                 {
 592                         rb->message(rb, txn, lsn, false, prefix, message_size, message);
 593
 594                         TeardownHistoricSnapshot(false);
 595                 }
 596                 PG_CATCH();
 597                 {
 598                         TeardownHistoricSnapshot(true);
 599                         PG_RE_THROW();
 600                 }
 601                 PG_END_TRY();
 602         }
 603 }
 604
 605
 606 static void
 607 AssertTXNLsnOrder(ReorderBuffer *rb)
 608 {
 609 #ifdef USE_ASSERT_CHECKING
 610         dlist_iter      iter;
 611         XLogRecPtr      prev_first_lsn = InvalidXLogRecPtr;
 612
 613         dlist_foreach(iter, &rb->toplevel_by_lsn)
 614         {
 615                 ReorderBufferTXN *cur_txn;
 616
 617                 cur_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
 618                 Assert(cur_txn->first_lsn != InvalidXLogRecPtr);
 619
 620                 if (cur_txn->end_lsn != InvalidXLogRecPtr)
 621                         Assert(cur_txn->first_lsn <= cur_txn->end_lsn);
 622
 623                 if (prev_first_lsn != InvalidXLogRecPtr)
 624                         Assert(prev_first_lsn < cur_txn->first_lsn);
 625
 626                 Assert(!cur_txn->is_known_as_subxact);
 627                 prev_first_lsn = cur_txn->first_lsn;
 628         }
 629 #endif
 630 }
 631
 632 ReorderBufferTXN *
 633 ReorderBufferGetOldestTXN(ReorderBuffer *rb)
 634 {
 635         ReorderBufferTXN *txn;
 636
 637         if (dlist_is_empty(&rb->toplevel_by_lsn))
 638                 return NULL;
 639
 640         AssertTXNLsnOrder(rb);
 641
 642         txn = dlist_head_element(ReorderBufferTXN, node, &rb->toplevel_by_lsn);
 643
 644         Assert(!txn->is_known_as_subxact);
 645         Assert(txn->first_lsn != InvalidXLogRecPtr);
 646         return txn;
 647 }
 648
 649 void
 650 ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr)
 651 {
 652         rb->current_restart_decoding_lsn = ptr;
 653 }
 654
 655 void
 656 ReorderBufferAssignChild(ReorderBuffer *rb, TransactionId xid,
 657                                                  TransactionId subxid, XLogRecPtr lsn)
 658 {
 659         ReorderBufferTXN *txn;
 660         ReorderBufferTXN *subtxn;
 661         bool            new_top;
 662         bool            new_sub;
 663
 664         txn = ReorderBufferTXNByXid(rb, xid, true, &new_top, lsn, true);
 665         subtxn = ReorderBufferTXNByXid(rb, subxid, true, &new_sub, lsn, false);
 666
 667         if (new_sub)
 668         {
 669                 /*
 670                  * we assign subtransactions to top level transaction even if we don't
 671                  * have data for it yet, assignment records frequently reference xids
 672                  * that have not yet produced any records. Knowing those aren't top
 673                  * level xids allows us to make processing cheaper in some places.
 674                  */
 675                 dlist_push_tail(&txn->subtxns, &subtxn->node);
 676                 txn->nsubtxns++;
 677         }
 678         else if (!subtxn->is_known_as_subxact)
 679         {
 680                 subtxn->is_known_as_subxact = true;
 681                 Assert(subtxn->nsubtxns == 0);
 682
 683                 /* remove from lsn order list of top-level transactions */
 684                 dlist_delete(&subtxn->node);
 685
 686                 /* add to toplevel transaction */
 687                 dlist_push_tail(&txn->subtxns, &subtxn->node);
 688                 txn->nsubtxns++;
 689         }
 690         else if (new_top)
 691         {
 692                 elog(ERROR, "existing subxact assigned to unknown toplevel xact");
 693         }
 694 }
 695
 696 /*
 697  * Associate a subtransaction with its toplevel transaction at commit
 698  * time. There may be no further changes added after this.
 699  */
 700 void
 701 ReorderBufferCommitChild(ReorderBuffer *rb, TransactionId xid,
 702                                                  TransactionId subxid, XLogRecPtr commit_lsn,
 703                                                  XLogRecPtr end_lsn)
 704 {
 705         ReorderBufferTXN *txn;
 706         ReorderBufferTXN *subtxn;
 707
 708         subtxn = ReorderBufferTXNByXid(rb, subxid, false, NULL,
 709                                                                    InvalidXLogRecPtr, false);
 710
 711         /*
 712          * No need to do anything if that subtxn didn't contain any changes
 713          */
 714         if (!subtxn)
 715                 return;
 716
 717         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, commit_lsn, true);
 718
 719         if (txn == NULL)
 720                 elog(ERROR, "subxact logged without previous toplevel record");
 721
 722         /*
 723          * Pass our base snapshot to the parent transaction if it doesn't have
 724          * one, or ours is older. That can happen if there are no changes in the
 725          * toplevel transaction but in one of the child transactions. This allows
 726          * the parent to simply use its base snapshot initially.
 727          */
 728         if (subtxn->base_snapshot != NULL &&
 729                 (txn->base_snapshot == NULL ||
 730                  txn->base_snapshot_lsn > subtxn->base_snapshot_lsn))
 731         {
 732                 txn->base_snapshot = subtxn->base_snapshot;
 733                 txn->base_snapshot_lsn = subtxn->base_snapshot_lsn;
 734                 subtxn->base_snapshot = NULL;
 735                 subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
 736         }
 737
 738         subtxn->final_lsn = commit_lsn;
 739         subtxn->end_lsn = end_lsn;
 740
 741         if (!subtxn->is_known_as_subxact)
 742         {
 743                 subtxn->is_known_as_subxact = true;
 744                 Assert(subtxn->nsubtxns == 0);
 745
 746                 /* remove from lsn order list of top-level transactions */
 747                 dlist_delete(&subtxn->node);
 748
 749                 /* add to subtransaction list */
 750                 dlist_push_tail(&txn->subtxns, &subtxn->node);
 751                 txn->nsubtxns++;
 752         }
 753 }
 754
 755
 756 /*
 757  * Support for efficiently iterating over a transaction's and its
 758  * subtransactions' changes.
 759  *
 760  * We do by doing a k-way merge between transactions/subtransactions. For that
 761  * we model the current heads of the different transactions as a binary heap
 762  * so we easily know which (sub-)transaction has the change with the smallest
 763  * lsn next.
 764  *
 765  * We assume the changes in individual transactions are already sorted by LSN.
 766  */
 767
 768 /*
 769  * Binary heap comparison function.
 770  */
 771 static int
 772 ReorderBufferIterCompare(Datum a, Datum b, void *arg)
 773 {
 774         ReorderBufferIterTXNState *state = (ReorderBufferIterTXNState *) arg;
 775         XLogRecPtr      pos_a = state->entries[DatumGetInt32(a)].lsn;
 776         XLogRecPtr      pos_b = state->entries[DatumGetInt32(b)].lsn;
 777
 778         if (pos_a < pos_b)
 779                 return 1;
 780         else if (pos_a == pos_b)
 781                 return 0;
 782         return -1;
 783 }
 784
 785 /*
 786  * Allocate & initialize an iterator which iterates in lsn order over a
 787  * transaction and all its subtransactions.
 788  */
 789 static ReorderBufferIterTXNState *
 790 ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn)
 791 {
 792         Size            nr_txns = 0;
 793         ReorderBufferIterTXNState *state;
 794         dlist_iter      cur_txn_i;
 795         int32           off;
 796
 797         /*
 798          * Calculate the size of our heap: one element for every transaction that
 799          * contains changes.  (Besides the transactions already in the reorder
 800          * buffer, we count the one we were directly passed.)
 801          */
 802         if (txn->nentries > 0)
 803                 nr_txns++;
 804
 805         dlist_foreach(cur_txn_i, &txn->subtxns)
 806         {
 807                 ReorderBufferTXN *cur_txn;
 808
 809                 cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
 810
 811                 if (cur_txn->nentries > 0)
 812                         nr_txns++;
 813         }
 814
 815         /*
 816          * TODO: Consider adding fastpath for the rather common nr_txns=1 case, no
 817          * need to allocate/build a heap then.
 818          */
 819
 820         /* allocate iteration state */
 821         state = (ReorderBufferIterTXNState *)
 822                 MemoryContextAllocZero(rb->context,
 823                                                            sizeof(ReorderBufferIterTXNState) +
 824                                                            sizeof(ReorderBufferIterTXNEntry) * nr_txns);
 825
 826         state->nr_txns = nr_txns;
 827         dlist_init(&state->old_change);
 828
 829         for (off = 0; off < state->nr_txns; off++)
 830         {
 831                 state->entries[off].fd = -1;
 832                 state->entries[off].segno = 0;
 833         }
 834
 835         /* allocate heap */
 836         state->heap = binaryheap_allocate(state->nr_txns,
 837                                                                           ReorderBufferIterCompare,
 838                                                                           state);
 839
 840         /*
 841          * Now insert items into the binary heap, in an unordered fashion.  (We
 842          * will run a heap assembly step at the end; this is more efficient.)
 843          */
 844
 845         off = 0;
 846
 847         /* add toplevel transaction if it contains changes */
 848         if (txn->nentries > 0)
 849         {
 850                 ReorderBufferChange *cur_change;
 851
 852                 if (txn->serialized)
 853                 {
 854                         /* serialize remaining changes */
 855                         ReorderBufferSerializeTXN(rb, txn);
 856                         ReorderBufferRestoreChanges(rb, txn, &state->entries[off].fd,
 857                                                                                 &state->entries[off].segno);
 858                 }
 859
 860                 cur_change = dlist_head_element(ReorderBufferChange, node,
 861                                                                                 &txn->changes);
 862
 863                 state->entries[off].lsn = cur_change->lsn;
 864                 state->entries[off].change = cur_change;
 865                 state->entries[off].txn = txn;
 866
 867                 binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
 868         }
 869
 870         /* add subtransactions if they contain changes */
 871         dlist_foreach(cur_txn_i, &txn->subtxns)
 872         {
 873                 ReorderBufferTXN *cur_txn;
 874
 875                 cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
 876
 877                 if (cur_txn->nentries > 0)
 878                 {
 879                         ReorderBufferChange *cur_change;
 880
 881                         if (cur_txn->serialized)
 882                         {
 883                                 /* serialize remaining changes */
 884                                 ReorderBufferSerializeTXN(rb, cur_txn);
 885                                 ReorderBufferRestoreChanges(rb, cur_txn,
 886                                                                                         &state->entries[off].fd,
 887                                                                                         &state->entries[off].segno);
 888                         }
 889                         cur_change = dlist_head_element(ReorderBufferChange, node,
 890                                                                                         &cur_txn->changes);
 891
 892                         state->entries[off].lsn = cur_change->lsn;
 893                         state->entries[off].change = cur_change;
 894                         state->entries[off].txn = cur_txn;
 895
 896                         binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
 897                 }
 898         }
 899
 900         /* assemble a valid binary heap */
 901         binaryheap_build(state->heap);
 902
 903         return state;
 904 }
 905
 906 /*
 907  * Return the next change when iterating over a transaction and its
 908  * subtransactions.
 909  *
 910  * Returns NULL when no further changes exist.
 911  */
 912 static ReorderBufferChange *
 913 ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state)
 914 {
 915         ReorderBufferChange *change;
 916         ReorderBufferIterTXNEntry *entry;
 917         int32           off;
 918
 919         /* nothing there anymore */
 920         if (state->heap->bh_size == 0)
 921                 return NULL;
 922
 923         off = DatumGetInt32(binaryheap_first(state->heap));
 924         entry = &state->entries[off];
 925
 926         /* free memory we might have "leaked" in the previous *Next call */
 927         if (!dlist_is_empty(&state->old_change))
 928         {
 929                 change = dlist_container(ReorderBufferChange, node,
 930                                                                  dlist_pop_head_node(&state->old_change));
 931                 ReorderBufferReturnChange(rb, change);
 932                 Assert(dlist_is_empty(&state->old_change));
 933         }
 934
 935         change = entry->change;
 936
 937         /*
 938          * update heap with information about which transaction has the next
 939          * relevant change in LSN order
 940          */
 941
 942         /* there are in-memory changes */
 943         if (dlist_has_next(&entry->txn->changes, &entry->change->node))
 944         {
 945                 dlist_node *next = dlist_next_node(&entry->txn->changes, &change->node);
 946                 ReorderBufferChange *next_change =
 947                 dlist_container(ReorderBufferChange, node, next);
 948
 949                 /* txn stays the same */
 950                 state->entries[off].lsn = next_change->lsn;
 951                 state->entries[off].change = next_change;
 952
 953                 binaryheap_replace_first(state->heap, Int32GetDatum(off));
 954                 return change;
 955         }
 956
 957         /* try to load changes from disk */
 958         if (entry->txn->nentries != entry->txn->nentries_mem)
 959         {
 960                 /*
 961                  * Ugly: restoring changes will reuse *Change records, thus delete the
 962                  * current one from the per-tx list and only free in the next call.
 963                  */
 964                 dlist_delete(&change->node);
 965                 dlist_push_tail(&state->old_change, &change->node);
 966
 967                 if (ReorderBufferRestoreChanges(rb, entry->txn, &entry->fd,
 968                                                                                 &state->entries[off].segno))
 969                 {
 970                         /* successfully restored changes from disk */
 971                         ReorderBufferChange *next_change =
 972                         dlist_head_element(ReorderBufferChange, node,
 973                                                            &entry->txn->changes);
 974
 975                         elog(DEBUG2, "restored %u/%u changes from disk",
 976                                  (uint32) entry->txn->nentries_mem,
 977                                  (uint32) entry->txn->nentries);
 978
 979                         Assert(entry->txn->nentries_mem);
 980                         /* txn stays the same */
 981                         state->entries[off].lsn = next_change->lsn;
 982                         state->entries[off].change = next_change;
 983                         binaryheap_replace_first(state->heap, Int32GetDatum(off));
 984
 985                         return change;
 986                 }
 987         }
 988
 989         /* ok, no changes there anymore, remove */
 990         binaryheap_remove_first(state->heap);
 991
 992         return change;
 993 }
 994
 995 /*
 996  * Deallocate the iterator
 997  */
 998 static void
 999 ReorderBufferIterTXNFinish(ReorderBuffer *rb,
1000                                                    ReorderBufferIterTXNState *state)
1001 {
1002         int32           off;
1003
1004         for (off = 0; off < state->nr_txns; off++)
1005         {
1006                 if (state->entries[off].fd != -1)
1007                         CloseTransientFile(state->entries[off].fd);
1008         }
1009
1010         /* free memory we might have "leaked" in the last *Next call */
1011         if (!dlist_is_empty(&state->old_change))
1012         {
1013                 ReorderBufferChange *change;
1014
1015                 change = dlist_container(ReorderBufferChange, node,
1016                                                                  dlist_pop_head_node(&state->old_change));
1017                 ReorderBufferReturnChange(rb, change);
1018                 Assert(dlist_is_empty(&state->old_change));
1019         }
1020
1021         binaryheap_free(state->heap);
1022         pfree(state);
1023 }
1024
1025 /*
1026  * Cleanup the contents of a transaction, usually after the transaction
1027  * committed or aborted.
1028  */
1029 static void
1030 ReorderBufferCleanupTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1031 {
1032         bool            found;
1033         dlist_mutable_iter iter;
1034
1035         /* cleanup subtransactions & their changes */
1036         dlist_foreach_modify(iter, &txn->subtxns)
1037         {
1038                 ReorderBufferTXN *subtxn;
1039
1040                 subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1041
1042                 /*
1043                  * Subtransactions are always associated to the toplevel TXN, even if
1044                  * they originally were happening inside another subtxn, so we won't
1045                  * ever recurse more than one level deep here.
1046                  */
1047                 Assert(subtxn->is_known_as_subxact);
1048                 Assert(subtxn->nsubtxns == 0);
1049
1050                 ReorderBufferCleanupTXN(rb, subtxn);
1051         }
1052
1053         /* cleanup changes in the toplevel txn */
1054         dlist_foreach_modify(iter, &txn->changes)
1055         {
1056                 ReorderBufferChange *change;
1057
1058                 change = dlist_container(ReorderBufferChange, node, iter.cur);
1059
1060                 ReorderBufferReturnChange(rb, change);
1061         }
1062
1063         /*
1064          * Cleanup the tuplecids we stored for decoding catalog snapshot access.
1065          * They are always stored in the toplevel transaction.
1066          */
1067         dlist_foreach_modify(iter, &txn->tuplecids)
1068         {
1069                 ReorderBufferChange *change;
1070
1071                 change = dlist_container(ReorderBufferChange, node, iter.cur);
1072                 Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1073                 ReorderBufferReturnChange(rb, change);
1074         }
1075
1076         if (txn->base_snapshot != NULL)
1077         {
1078                 SnapBuildSnapDecRefcount(txn->base_snapshot);
1079                 txn->base_snapshot = NULL;
1080                 txn->base_snapshot_lsn = InvalidXLogRecPtr;
1081         }
1082
1083         /*
1084          * Remove TXN from its containing list.
1085          *
1086          * Note: if txn->is_known_as_subxact, we are deleting the TXN from its
1087          * parent's list of known subxacts; this leaves the parent's nsubxacts
1088          * count too high, but we don't care.  Otherwise, we are deleting the TXN
1089          * from the LSN-ordered list of toplevel TXNs.
1090          */
1091         dlist_delete(&txn->node);
1092
1093         /* now remove reference from buffer */
1094         hash_search(rb->by_txn,
1095                                 (void *) &txn->xid,
1096                                 HASH_REMOVE,
1097                                 &found);
1098         Assert(found);
1099
1100         /* remove entries spilled to disk */
1101         if (txn->serialized)
1102                 ReorderBufferRestoreCleanup(rb, txn);
1103
1104         /* deallocate */
1105         ReorderBufferReturnTXN(rb, txn);
1106 }
1107
1108 /*
1109  * Build a hash with a (relfilenode, ctid) -> (cmin, cmax) mapping for use by
1110  * tqual.c's HeapTupleSatisfiesHistoricMVCC.
1111  */
1112 static void
1113 ReorderBufferBuildTupleCidHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
1114 {
1115         dlist_iter      iter;
1116         HASHCTL         hash_ctl;
1117
1118         if (!txn->has_catalog_changes || dlist_is_empty(&txn->tuplecids))
1119                 return;
1120
1121         memset(&hash_ctl, 0, sizeof(hash_ctl));
1122
1123         hash_ctl.keysize = sizeof(ReorderBufferTupleCidKey);
1124         hash_ctl.entrysize = sizeof(ReorderBufferTupleCidEnt);
1125         hash_ctl.hcxt = rb->context;
1126
1127         /*
1128          * create the hash with the exact number of to-be-stored tuplecids from
1129          * the start
1130          */
1131         txn->tuplecid_hash =
1132                 hash_create("ReorderBufferTupleCid", txn->ntuplecids, &hash_ctl,
1133                                         HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1134
1135         dlist_foreach(iter, &txn->tuplecids)
1136         {
1137                 ReorderBufferTupleCidKey key;
1138                 ReorderBufferTupleCidEnt *ent;
1139                 bool            found;
1140                 ReorderBufferChange *change;
1141
1142                 change = dlist_container(ReorderBufferChange, node, iter.cur);
1143
1144                 Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1145
1146                 /* be careful about padding */
1147                 memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
1148
1149                 key.relnode = change->data.tuplecid.node;
1150
1151                 ItemPointerCopy(&change->data.tuplecid.tid,
1152                                                 &key.tid);
1153
1154                 ent = (ReorderBufferTupleCidEnt *)
1155                         hash_search(txn->tuplecid_hash,
1156                                                 (void *) &key,
1157                                                 HASH_ENTER | HASH_FIND,
1158                                                 &found);
1159                 if (!found)
1160                 {
1161                         ent->cmin = change->data.tuplecid.cmin;
1162                         ent->cmax = change->data.tuplecid.cmax;
1163                         ent->combocid = change->data.tuplecid.combocid;
1164                 }
1165                 else
1166                 {
1167                         Assert(ent->cmin == change->data.tuplecid.cmin);
1168                         Assert(ent->cmax == InvalidCommandId ||
1169                                    ent->cmax == change->data.tuplecid.cmax);
1170
1171                         /*
1172                          * if the tuple got valid in this transaction and now got deleted
1173                          * we already have a valid cmin stored. The cmax will be
1174                          * InvalidCommandId though.
1175                          */
1176                         ent->cmax = change->data.tuplecid.cmax;
1177                 }
1178         }
1179 }
1180
1181 /*
1182  * Copy a provided snapshot so we can modify it privately. This is needed so
1183  * that catalog modifying transactions can look into intermediate catalog
1184  * states.
1185  */
1186 static Snapshot
1187 ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
1188                                           ReorderBufferTXN *txn, CommandId cid)
1189 {
1190         Snapshot        snap;
1191         dlist_iter      iter;
1192         int                     i = 0;
1193         Size            size;
1194
1195         size = sizeof(SnapshotData) +
1196                 sizeof(TransactionId) * orig_snap->xcnt +
1197                 sizeof(TransactionId) * (txn->nsubtxns + 1);
1198
1199         snap = MemoryContextAllocZero(rb->context, size);
1200         memcpy(snap, orig_snap, sizeof(SnapshotData));
1201
1202         snap->copied = true;
1203         snap->active_count = 1;         /* mark as active so nobody frees it */
1204         snap->regd_count = 0;
1205         snap->xip = (TransactionId *) (snap + 1);
1206
1207         memcpy(snap->xip, orig_snap->xip, sizeof(TransactionId) * snap->xcnt);
1208
1209         /*
1210          * snap->subxip contains all txids that belong to our transaction which we
1211          * need to check via cmin/cmax. That's why we store the toplevel
1212          * transaction in there as well.
1213          */
1214         snap->subxip = snap->xip + snap->xcnt;
1215         snap->subxip[i++] = txn->xid;
1216
1217         /*
1218          * nsubxcnt isn't decreased when subtransactions abort, so count manually.
1219          * Since it's an upper boundary it is safe to use it for the allocation
1220          * above.
1221          */
1222         snap->subxcnt = 1;
1223
1224         dlist_foreach(iter, &txn->subtxns)
1225         {
1226                 ReorderBufferTXN *sub_txn;
1227
1228                 sub_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
1229                 snap->subxip[i++] = sub_txn->xid;
1230                 snap->subxcnt++;
1231         }
1232
1233         /* sort so we can bsearch() later */
1234         qsort(snap->subxip, snap->subxcnt, sizeof(TransactionId), xidComparator);
1235
1236         /* store the specified current CommandId */
1237         snap->curcid = cid;
1238
1239         return snap;
1240 }
1241
1242 /*
1243  * Free a previously ReorderBufferCopySnap'ed snapshot
1244  */
1245 static void
1246 ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap)
1247 {
1248         if (snap->copied)
1249                 pfree(snap);
1250         else
1251                 SnapBuildSnapDecRefcount(snap);
1252 }
1253
1254 /*
1255  * Perform the replay of a transaction and it's non-aborted subtransactions.
1256  *
1257  * Subtransactions previously have to be processed by
1258  * ReorderBufferCommitChild(), even if previously assigned to the toplevel
1259  * transaction with ReorderBufferAssignChild.
1260  *
1261  * We currently can only decode a transaction's contents in when their commit
1262  * record is read because that's currently the only place where we know about
1263  * cache invalidations. Thus, once a toplevel commit is read, we iterate over
1264  * the top and subtransactions (using a k-way merge) and replay the changes in
1265  * lsn order.
1266  */
1267 void
1268 ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
1269                                         XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
1270                                         TimestampTz commit_time,
1271                                         RepOriginId origin_id, XLogRecPtr origin_lsn)
1272 {
1273         ReorderBufferTXN *txn;
1274         volatile Snapshot snapshot_now;
1275         volatile CommandId command_id = FirstCommandId;
1276         bool            using_subtxn;
1277         ReorderBufferIterTXNState *volatile iterstate = NULL;
1278
1279         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
1280                                                                 false);
1281
1282         /* unknown transaction, nothing to replay */
1283         if (txn == NULL)
1284                 return;
1285
1286         txn->final_lsn = commit_lsn;
1287         txn->end_lsn = end_lsn;
1288         txn->commit_time = commit_time;
1289         txn->origin_id = origin_id;
1290         txn->origin_lsn = origin_lsn;
1291
1292         /*
1293          * If this transaction didn't have any real changes in our database, it's
1294          * OK not to have a snapshot. Note that ReorderBufferCommitChild will have
1295          * transferred its snapshot to this transaction if it had one and the
1296          * toplevel tx didn't.
1297          */
1298         if (txn->base_snapshot == NULL)
1299         {
1300                 Assert(txn->ninvalidations == 0);
1301                 ReorderBufferCleanupTXN(rb, txn);
1302                 return;
1303         }
1304
1305         snapshot_now = txn->base_snapshot;
1306
1307         /* build data to be able to lookup the CommandIds of catalog tuples */
1308         ReorderBufferBuildTupleCidHash(rb, txn);
1309
1310         /* setup the initial snapshot */
1311         SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
1312
1313         /*
1314          * Decoding needs access to syscaches et al., which in turn use
1315          * heavyweight locks and such. Thus we need to have enough state around to
1316          * keep track of those.  The easiest way is to simply use a transaction
1317          * internally.  That also allows us to easily enforce that nothing writes
1318          * to the database by checking for xid assignments.
1319          *
1320          * When we're called via the SQL SRF there's already a transaction
1321          * started, so start an explicit subtransaction there.
1322          */
1323         using_subtxn = IsTransactionOrTransactionBlock();
1324
1325         PG_TRY();
1326         {
1327                 ReorderBufferChange *change;
1328                 ReorderBufferChange *specinsert = NULL;
1329
1330                 if (using_subtxn)
1331                         BeginInternalSubTransaction("replay");
1332                 else
1333                         StartTransactionCommand();
1334
1335                 rb->begin(rb, txn);
1336
1337                 iterstate = ReorderBufferIterTXNInit(rb, txn);
1338                 while ((change = ReorderBufferIterTXNNext(rb, iterstate)) != NULL)
1339                 {
1340                         Relation        relation = NULL;
1341                         Oid                     reloid;
1342
1343                         switch (change->action)
1344                         {
1345                                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
1346
1347                                         /*
1348                                          * Confirmation for speculative insertion arrived. Simply
1349                                          * use as a normal record. It'll be cleaned up at the end
1350                                          * of INSERT processing.
1351                                          */
1352                                         Assert(specinsert->data.tp.oldtuple == NULL);
1353                                         change = specinsert;
1354                                         change->action = REORDER_BUFFER_CHANGE_INSERT;
1355
1356                                         /* intentionally fall through */
1357                                 case REORDER_BUFFER_CHANGE_INSERT:
1358                                 case REORDER_BUFFER_CHANGE_UPDATE:
1359                                 case REORDER_BUFFER_CHANGE_DELETE:
1360                                         Assert(snapshot_now);
1361
1362                                         reloid = RelidByRelfilenode(change->data.tp.relnode.spcNode,
1363                                                                                                 change->data.tp.relnode.relNode);
1364
1365                                         /*
1366                                          * Catalog tuple without data, emitted while catalog was
1367                                          * in the process of being rewritten.
1368                                          */
1369                                         if (reloid == InvalidOid &&
1370                                                 change->data.tp.newtuple == NULL &&
1371                                                 change->data.tp.oldtuple == NULL)
1372                                                 goto change_done;
1373                                         else if (reloid == InvalidOid)
1374                                                 elog(ERROR, "could not map filenode \"%s\" to relation OID",
1375                                                          relpathperm(change->data.tp.relnode,
1376                                                                                  MAIN_FORKNUM));
1377
1378                                         relation = RelationIdGetRelation(reloid);
1379
1380                                         if (relation == NULL)
1381                                                 elog(ERROR, "could not open relation with OID %u (for filenode \"%s\")",
1382                                                          reloid,
1383                                                          relpathperm(change->data.tp.relnode,
1384                                                                                  MAIN_FORKNUM));
1385
1386                                         if (!RelationIsLogicallyLogged(relation))
1387                                                 goto change_done;
1388
1389                                         /*
1390                                          * For now ignore sequence changes entirely. Most of the
1391                                          * time they don't log changes using records we
1392                                          * understand, so it doesn't make sense to handle the few
1393                                          * cases we do.
1394                                          */
1395                                         if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
1396                                                 goto change_done;
1397
1398                                         /* user-triggered change */
1399                                         if (!IsToastRelation(relation))
1400                                         {
1401                                                 ReorderBufferToastReplace(rb, txn, relation, change);
1402                                                 rb->apply_change(rb, txn, relation, change);
1403
1404                                                 /*
1405                                                  * Only clear reassembled toast chunks if we're sure
1406                                                  * they're not required anymore. The creator of the
1407                                                  * tuple tells us.
1408                                                  */
1409                                                 if (change->data.tp.clear_toast_afterwards)
1410                                                         ReorderBufferToastReset(rb, txn);
1411                                         }
1412                                         /* we're not interested in toast deletions */
1413                                         else if (change->action == REORDER_BUFFER_CHANGE_INSERT)
1414                                         {
1415                                                 /*
1416                                                  * Need to reassemble the full toasted Datum in
1417                                                  * memory, to ensure the chunks don't get reused till
1418                                                  * we're done remove it from the list of this
1419                                                  * transaction's changes. Otherwise it will get
1420                                                  * freed/reused while restoring spooled data from
1421                                                  * disk.
1422                                                  */
1423                                                 dlist_delete(&change->node);
1424                                                 ReorderBufferToastAppendChunk(rb, txn, relation,
1425                                                                                                           change);
1426                                         }
1427
1428                         change_done:
1429
1430                                         /*
1431                                          * Either speculative insertion was confirmed, or it was
1432                                          * unsuccessful and the record isn't needed anymore.
1433                                          */
1434                                         if (specinsert != NULL)
1435                                         {
1436                                                 ReorderBufferReturnChange(rb, specinsert);
1437                                                 specinsert = NULL;
1438                                         }
1439
1440                                         if (relation != NULL)
1441                                         {
1442                                                 RelationClose(relation);
1443                                                 relation = NULL;
1444                                         }
1445                                         break;
1446
1447                                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
1448
1449                                         /*
1450                                          * Speculative insertions are dealt with by delaying the
1451                                          * processing of the insert until the confirmation record
1452                                          * arrives. For that we simply unlink the record from the
1453                                          * chain, so it does not get freed/reused while restoring
1454                                          * spooled data from disk.
1455                                          *
1456                                          * This is safe in the face of concurrent catalog changes
1457                                          * because the relevant relation can't be changed between
1458                                          * speculative insertion and confirmation due to
1459                                          * CheckTableNotInUse() and locking.
1460                                          */
1461
1462                                         /* clear out a pending (and thus failed) speculation */
1463                                         if (specinsert != NULL)
1464                                         {
1465                                                 ReorderBufferReturnChange(rb, specinsert);
1466                                                 specinsert = NULL;
1467                                         }
1468
1469                                         /* and memorize the pending insertion */
1470                                         dlist_delete(&change->node);
1471                                         specinsert = change;
1472                                         break;
1473
1474                                 case REORDER_BUFFER_CHANGE_MESSAGE:
1475                                         rb->message(rb, txn, change->lsn, true,
1476                                                                 change->data.msg.prefix,
1477                                                                 change->data.msg.message_size,
1478                                                                 change->data.msg.message);
1479                                         break;
1480
1481                                 case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
1482                                         /* get rid of the old */
1483                                         TeardownHistoricSnapshot(false);
1484
1485                                         if (snapshot_now->copied)
1486                                         {
1487                                                 ReorderBufferFreeSnap(rb, snapshot_now);
1488                                                 snapshot_now =
1489                                                         ReorderBufferCopySnap(rb, change->data.snapshot,
1490                                                                                                   txn, command_id);
1491                                         }
1492
1493                                         /*
1494                                          * Restored from disk, need to be careful not to double
1495                                          * free. We could introduce refcounting for that, but for
1496                                          * now this seems infrequent enough not to care.
1497                                          */
1498                                         else if (change->data.snapshot->copied)
1499                                         {
1500                                                 snapshot_now =
1501                                                         ReorderBufferCopySnap(rb, change->data.snapshot,
1502                                                                                                   txn, command_id);
1503                                         }
1504                                         else
1505                                         {
1506                                                 snapshot_now = change->data.snapshot;
1507                                         }
1508
1509
1510                                         /* and continue with the new one */
1511                                         SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
1512                                         break;
1513
1514                                 case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
1515                                         Assert(change->data.command_id != InvalidCommandId);
1516
1517                                         if (command_id < change->data.command_id)
1518                                         {
1519                                                 command_id = change->data.command_id;
1520
1521                                                 if (!snapshot_now->copied)
1522                                                 {
1523                                                         /* we don't use the global one anymore */
1524                                                         snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
1525                                                                                                                                  txn, command_id);
1526                                                 }
1527
1528                                                 snapshot_now->curcid = command_id;
1529
1530                                                 TeardownHistoricSnapshot(false);
1531                                                 SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
1532
1533                                                 /*
1534                                                  * Every time the CommandId is incremented, we could
1535                                                  * see new catalog contents, so execute all
1536                                                  * invalidations.
1537                                                  */
1538                                                 ReorderBufferExecuteInvalidations(rb, txn);
1539                                         }
1540
1541                                         break;
1542
1543                                 case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
1544                                         elog(ERROR, "tuplecid value in changequeue");
1545                                         break;
1546                         }
1547                 }
1548
1549                 /*
1550                  * There's a speculative insertion remaining, just clean in up, it
1551                  * can't have been successful, otherwise we'd gotten a confirmation
1552                  * record.
1553                  */
1554                 if (specinsert)
1555                 {
1556                         ReorderBufferReturnChange(rb, specinsert);
1557                         specinsert = NULL;
1558                 }
1559
1560                 /* clean up the iterator */
1561                 ReorderBufferIterTXNFinish(rb, iterstate);
1562                 iterstate = NULL;
1563
1564                 /* call commit callback */
1565                 rb->commit(rb, txn, commit_lsn);
1566
1567                 /* this is just a sanity check against bad output plugin behaviour */
1568                 if (GetCurrentTransactionIdIfAny() != InvalidTransactionId)
1569                         elog(ERROR, "output plugin used XID %u",
1570                                  GetCurrentTransactionId());
1571
1572                 /* cleanup */
1573                 TeardownHistoricSnapshot(false);
1574
1575                 /*
1576                  * Aborting the current (sub-)transaction as a whole has the right
1577                  * semantics. We want all locks acquired in here to be released, not
1578                  * reassigned to the parent and we do not want any database access
1579                  * have persistent effects.
1580                  */
1581                 AbortCurrentTransaction();
1582
1583                 /* make sure there's no cache pollution */
1584                 ReorderBufferExecuteInvalidations(rb, txn);
1585
1586                 if (using_subtxn)
1587                         RollbackAndReleaseCurrentSubTransaction();
1588
1589                 if (snapshot_now->copied)
1590                         ReorderBufferFreeSnap(rb, snapshot_now);
1591
1592                 /* remove potential on-disk data, and deallocate */
1593                 ReorderBufferCleanupTXN(rb, txn);
1594         }
1595         PG_CATCH();
1596         {
1597                 /* TODO: Encapsulate cleanup from the PG_TRY and PG_CATCH blocks */
1598                 if (iterstate)
1599                         ReorderBufferIterTXNFinish(rb, iterstate);
1600
1601                 TeardownHistoricSnapshot(true);
1602
1603                 /*
1604                  * Force cache invalidation to happen outside of a valid transaction
1605                  * to prevent catalog access as we just caught an error.
1606                  */
1607                 AbortCurrentTransaction();
1608
1609                 /* make sure there's no cache pollution */
1610                 ReorderBufferExecuteInvalidations(rb, txn);
1611
1612                 if (using_subtxn)
1613                         RollbackAndReleaseCurrentSubTransaction();
1614
1615                 if (snapshot_now->copied)
1616                         ReorderBufferFreeSnap(rb, snapshot_now);
1617
1618                 /* remove potential on-disk data, and deallocate */
1619                 ReorderBufferCleanupTXN(rb, txn);
1620
1621                 PG_RE_THROW();
1622         }
1623         PG_END_TRY();
1624 }
1625
1626 /*
1627  * Abort a transaction that possibly has previous changes. Needs to be first
1628  * called for subtransactions and then for the toplevel xid.
1629  *
1630  * NB: Transactions handled here have to have actively aborted (i.e. have
1631  * produced an abort record). Implicitly aborted transactions are handled via
1632  * ReorderBufferAbortOld(); transactions we're just not interested in, but
1633  * which have committed are handled in ReorderBufferForget().
1634  *
1635  * This function purges this transaction and its contents from memory and
1636  * disk.
1637  */
1638 void
1639 ReorderBufferAbort(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
1640 {
1641         ReorderBufferTXN *txn;
1642
1643         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
1644                                                                 false);
1645
1646         /* unknown, nothing to remove */
1647         if (txn == NULL)
1648                 return;
1649
1650         /* cosmetic... */
1651         txn->final_lsn = lsn;
1652
1653         /* remove potential on-disk data, and deallocate */
1654         ReorderBufferCleanupTXN(rb, txn);
1655 }
1656
1657 /*
1658  * Abort all transactions that aren't actually running anymore because the
1659  * server restarted.
1660  *
1661  * NB: These really have to be transactions that have aborted due to a server
1662  * crash/immediate restart, as we don't deal with invalidations here.
1663  */
1664 void
1665 ReorderBufferAbortOld(ReorderBuffer *rb, TransactionId oldestRunningXid)
1666 {
1667         dlist_mutable_iter it;
1668
1669         /*
1670          * Iterate through all (potential) toplevel TXNs and abort all that are
1671          * older than what possibly can be running. Once we've found the first
1672          * that is alive we stop, there might be some that acquired an xid earlier
1673          * but started writing later, but it's unlikely and they will be cleaned
1674          * up in a later call to this function.
1675          */
1676         dlist_foreach_modify(it, &rb->toplevel_by_lsn)
1677         {
1678                 ReorderBufferTXN *txn;
1679
1680                 txn = dlist_container(ReorderBufferTXN, node, it.cur);
1681
1682                 if (TransactionIdPrecedes(txn->xid, oldestRunningXid))
1683                 {
1684                         /*
1685                          * We set final_lsn on a transaction when we decode its commit or
1686                          * abort record, but we never see those records for crashed
1687                          * transactions.  To ensure cleanup of these transactions, set
1688                          * final_lsn to that of their last change; this causes
1689                          * ReorderBufferRestoreCleanup to do the right thing.
1690                          */
1691                         if (txn->serialized && txn->final_lsn == 0)
1692                         {
1693                                 ReorderBufferChange *last =
1694                                         dlist_tail_element(ReorderBufferChange, node, &txn->changes);
1695
1696                                 txn->final_lsn = last->lsn;
1697                         }
1698
1699                         elog(DEBUG2, "aborting old transaction %u", txn->xid);
1700
1701                         /* remove potential on-disk data, and deallocate this tx */
1702                         ReorderBufferCleanupTXN(rb, txn);
1703                 }
1704                 else
1705                         return;
1706         }
1707 }
1708
1709 /*
1710  * Forget the contents of a transaction if we aren't interested in it's
1711  * contents. Needs to be first called for subtransactions and then for the
1712  * toplevel xid.
1713  *
1714  * This is significantly different to ReorderBufferAbort() because
1715  * transactions that have committed need to be treated differently from aborted
1716  * ones since they may have modified the catalog.
1717  *
1718  * Note that this is only allowed to be called in the moment a transaction
1719  * commit has just been read, not earlier; otherwise later records referring
1720  * to this xid might re-create the transaction incompletely.
1721  */
1722 void
1723 ReorderBufferForget(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
1724 {
1725         ReorderBufferTXN *txn;
1726
1727         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
1728                                                                 false);
1729
1730         /* unknown, nothing to forget */
1731         if (txn == NULL)
1732                 return;
1733
1734         /* cosmetic... */
1735         txn->final_lsn = lsn;
1736
1737         /*
1738          * Process cache invalidation messages if there are any. Even if we're not
1739          * interested in the transaction's contents, it could have manipulated the
1740          * catalog and we need to update the caches according to that.
1741          */
1742         if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
1743                 ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
1744                                                                                    txn->invalidations);
1745         else
1746                 Assert(txn->ninvalidations == 0);
1747
1748         /* remove potential on-disk data, and deallocate */
1749         ReorderBufferCleanupTXN(rb, txn);
1750 }
1751
1752 /*
1753  * Execute invalidations happening outside the context of a decoded
1754  * transaction. That currently happens either for xid-less commits
1755  * (c.f. RecordTransactionCommit()) or for invalidations in uninteresting
1756  * transactions (via ReorderBufferForget()).
1757  */
1758 void
1759 ReorderBufferImmediateInvalidation(ReorderBuffer *rb, uint32 ninvalidations,
1760                                                                    SharedInvalidationMessage *invalidations)
1761 {
1762         bool            use_subtxn = IsTransactionOrTransactionBlock();
1763         int                     i;
1764
1765         if (use_subtxn)
1766                 BeginInternalSubTransaction("replay");
1767
1768         /*
1769          * Force invalidations to happen outside of a valid transaction - that way
1770          * entries will just be marked as invalid without accessing the catalog.
1771          * That's advantageous because we don't need to setup the full state
1772          * necessary for catalog access.
1773          */
1774         if (use_subtxn)
1775                 AbortCurrentTransaction();
1776
1777         for (i = 0; i < ninvalidations; i++)
1778                 LocalExecuteInvalidationMessage(&invalidations[i]);
1779
1780         if (use_subtxn)
1781                 RollbackAndReleaseCurrentSubTransaction();
1782 }
1783
1784 /*
1785  * Tell reorderbuffer about an xid seen in the WAL stream. Has to be called at
1786  * least once for every xid in XLogRecord->xl_xid (other places in records
1787  * may, but do not have to be passed through here).
1788  *
1789  * Reorderbuffer keeps some datastructures about transactions in LSN order,
1790  * for efficiency. To do that it has to know about when transactions are seen
1791  * first in the WAL. As many types of records are not actually interesting for
1792  * logical decoding, they do not necessarily pass though here.
1793  */
1794 void
1795 ReorderBufferProcessXid(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
1796 {
1797         /* many records won't have an xid assigned, centralize check here */
1798         if (xid != InvalidTransactionId)
1799                 ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
1800 }
1801
1802 /*
1803  * Add a new snapshot to this transaction that may only used after lsn 'lsn'
1804  * because the previous snapshot doesn't describe the catalog correctly for
1805  * following rows.
1806  */
1807 void
1808 ReorderBufferAddSnapshot(ReorderBuffer *rb, TransactionId xid,
1809                                                  XLogRecPtr lsn, Snapshot snap)
1810 {
1811         ReorderBufferChange *change = ReorderBufferGetChange(rb);
1812
1813         change->data.snapshot = snap;
1814         change->action = REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT;
1815
1816         ReorderBufferQueueChange(rb, xid, lsn, change);
1817 }
1818
1819 /*
1820  * Setup the base snapshot of a transaction. The base snapshot is the snapshot
1821  * that is used to decode all changes until either this transaction modifies
1822  * the catalog or another catalog modifying transaction commits.
1823  *
1824  * Needs to be called before any changes are added with
1825  * ReorderBufferQueueChange().
1826  */
1827 void
1828 ReorderBufferSetBaseSnapshot(ReorderBuffer *rb, TransactionId xid,
1829                                                          XLogRecPtr lsn, Snapshot snap)
1830 {
1831         ReorderBufferTXN *txn;
1832         bool            is_new;
1833
1834         txn = ReorderBufferTXNByXid(rb, xid, true, &is_new, lsn, true);
1835         Assert(txn->base_snapshot == NULL);
1836         Assert(snap != NULL);
1837
1838         txn->base_snapshot = snap;
1839         txn->base_snapshot_lsn = lsn;
1840 }
1841
1842 /*
1843  * Access the catalog with this CommandId at this point in the changestream.
1844  *
1845  * May only be called for command ids > 1
1846  */
1847 void
1848 ReorderBufferAddNewCommandId(ReorderBuffer *rb, TransactionId xid,
1849                                                          XLogRecPtr lsn, CommandId cid)
1850 {
1851         ReorderBufferChange *change = ReorderBufferGetChange(rb);
1852
1853         change->data.command_id = cid;
1854         change->action = REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID;
1855
1856         ReorderBufferQueueChange(rb, xid, lsn, change);
1857 }
1858
1859
1860 /*
1861  * Add new (relfilenode, tid) -> (cmin, cmax) mappings.
1862  */
1863 void
1864 ReorderBufferAddNewTupleCids(ReorderBuffer *rb, TransactionId xid,
1865                                                          XLogRecPtr lsn, RelFileNode node,
1866                                                          ItemPointerData tid, CommandId cmin,
1867                                                          CommandId cmax, CommandId combocid)
1868 {
1869         ReorderBufferChange *change = ReorderBufferGetChange(rb);
1870         ReorderBufferTXN *txn;
1871
1872         txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
1873
1874         change->data.tuplecid.node = node;
1875         change->data.tuplecid.tid = tid;
1876         change->data.tuplecid.cmin = cmin;
1877         change->data.tuplecid.cmax = cmax;
1878         change->data.tuplecid.combocid = combocid;
1879         change->lsn = lsn;
1880         change->action = REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID;
1881
1882         dlist_push_tail(&txn->tuplecids, &change->node);
1883         txn->ntuplecids++;
1884 }
1885
1886 /*
1887  * Setup the invalidation of the toplevel transaction.
1888  *
1889  * This needs to be done before ReorderBufferCommit is called!
1890  */
1891 void
1892 ReorderBufferAddInvalidations(ReorderBuffer *rb, TransactionId xid,
1893                                                           XLogRecPtr lsn, Size nmsgs,
1894                                                           SharedInvalidationMessage *msgs)
1895 {
1896         ReorderBufferTXN *txn;
1897
1898         txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
1899
1900         if (txn->ninvalidations != 0)
1901                 elog(ERROR, "only ever add one set of invalidations");
1902
1903         Assert(nmsgs > 0);
1904
1905         txn->ninvalidations = nmsgs;
1906         txn->invalidations = (SharedInvalidationMessage *)
1907                 MemoryContextAlloc(rb->context,
1908                                                    sizeof(SharedInvalidationMessage) * nmsgs);
1909         memcpy(txn->invalidations, msgs,
1910                    sizeof(SharedInvalidationMessage) * nmsgs);
1911 }
1912
1913 /*
1914  * Apply all invalidations we know. Possibly we only need parts at this point
1915  * in the changestream but we don't know which those are.
1916  */
1917 static void
1918 ReorderBufferExecuteInvalidations(ReorderBuffer *rb, ReorderBufferTXN *txn)
1919 {
1920         int                     i;
1921
1922         for (i = 0; i < txn->ninvalidations; i++)
1923                 LocalExecuteInvalidationMessage(&txn->invalidations[i]);
1924 }
1925
1926 /*
1927  * Mark a transaction as containing catalog changes
1928  */
1929 void
1930 ReorderBufferXidSetCatalogChanges(ReorderBuffer *rb, TransactionId xid,
1931                                                                   XLogRecPtr lsn)
1932 {
1933         ReorderBufferTXN *txn;
1934
1935         txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
1936
1937         txn->has_catalog_changes = true;
1938 }
1939
1940 /*
1941  * Query whether a transaction is already *known* to contain catalog
1942  * changes. This can be wrong until directly before the commit!
1943  */
1944 bool
1945 ReorderBufferXidHasCatalogChanges(ReorderBuffer *rb, TransactionId xid)
1946 {
1947         ReorderBufferTXN *txn;
1948
1949         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
1950                                                                 false);
1951         if (txn == NULL)
1952                 return false;
1953
1954         return txn->has_catalog_changes;
1955 }
1956
1957 /*
1958  * Have we already added the first snapshot?
1959  */
1960 bool
1961 ReorderBufferXidHasBaseSnapshot(ReorderBuffer *rb, TransactionId xid)
1962 {
1963         ReorderBufferTXN *txn;
1964
1965         txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
1966                                                                 false);
1967
1968         /* transaction isn't known yet, ergo no snapshot */
1969         if (txn == NULL)
1970                 return false;
1971
1972         /*
1973          * TODO: It would be a nice improvement if we would check the toplevel
1974          * transaction in subtransactions, but we'd need to keep track of a bit
1975          * more state.
1976          */
1977         return txn->base_snapshot != NULL;
1978 }
1979
1980
1981 /*
1982  * ---------------------------------------
1983  * Disk serialization support
1984  * ---------------------------------------
1985  */
1986
1987 /*
1988  * Ensure the IO buffer is >= sz.
1989  */
1990 static void
1991 ReorderBufferSerializeReserve(ReorderBuffer *rb, Size sz)
1992 {
1993         if (!rb->outbufsize)
1994         {
1995                 rb->outbuf = MemoryContextAlloc(rb->context, sz);
1996                 rb->outbufsize = sz;
1997         }
1998         else if (rb->outbufsize < sz)
1999         {
2000                 rb->outbuf = repalloc(rb->outbuf, sz);
2001                 rb->outbufsize = sz;
2002         }
2003 }
2004
2005 /*
2006  * Check whether the transaction tx should spill its data to disk.
2007  */
2008 static void
2009 ReorderBufferCheckSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
2010 {
2011         /*
2012          * TODO: improve accounting so we cheaply can take subtransactions into
2013          * account here.
2014          */
2015         if (txn->nentries_mem >= max_changes_in_memory)
2016         {
2017                 ReorderBufferSerializeTXN(rb, txn);
2018                 Assert(txn->nentries_mem == 0);
2019         }
2020 }
2021
2022 /*
2023  * Spill data of a large transaction (and its subtransactions) to disk.
2024  */
2025 static void
2026 ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
2027 {
2028         dlist_iter      subtxn_i;
2029         dlist_mutable_iter change_i;
2030         int                     fd = -1;
2031         XLogSegNo       curOpenSegNo = 0;
2032         Size            spilled = 0;
2033         char            path[MAXPGPATH];
2034
2035         elog(DEBUG2, "spill %u changes in XID %u to disk",
2036                  (uint32) txn->nentries_mem, txn->xid);
2037
2038         /* do the same to all child TXs */
2039         dlist_foreach(subtxn_i, &txn->subtxns)
2040         {
2041                 ReorderBufferTXN *subtxn;
2042
2043                 subtxn = dlist_container(ReorderBufferTXN, node, subtxn_i.cur);
2044                 ReorderBufferSerializeTXN(rb, subtxn);
2045         }
2046
2047         /* serialize changestream */
2048         dlist_foreach_modify(change_i, &txn->changes)
2049         {
2050                 ReorderBufferChange *change;
2051
2052                 change = dlist_container(ReorderBufferChange, node, change_i.cur);
2053
2054                 /*
2055                  * store in segment in which it belongs by start lsn, don't split over
2056                  * multiple segments tho
2057                  */
2058                 if (fd == -1 ||
2059                         !XLByteInSeg(change->lsn, curOpenSegNo, wal_segment_size))
2060                 {
2061                         XLogRecPtr      recptr;
2062
2063                         if (fd != -1)
2064                                 CloseTransientFile(fd);
2065
2066                         XLByteToSeg(change->lsn, curOpenSegNo, wal_segment_size);
2067                         XLogSegNoOffsetToRecPtr(curOpenSegNo, 0, recptr, wal_segment_size);
2068
2069                         /*
2070                          * No need to care about TLIs here, only used during a single run,
2071                          * so each LSN only maps to a specific WAL record.
2072                          */
2073                         sprintf(path, "pg_replslot/%s/xid-%u-lsn-%X-%X.snap",
2074                                         NameStr(MyReplicationSlot->data.name), txn->xid,
2075                                         (uint32) (recptr >> 32), (uint32) recptr);
2076
2077                         /* open segment, create it if necessary */
2078                         fd = OpenTransientFile(path,
2079                                                                    O_CREAT | O_WRONLY | O_APPEND | PG_BINARY);
2080
2081                         if (fd < 0)
2082                                 ereport(ERROR,
2083                                                 (errcode_for_file_access(),
2084                                                  errmsg("could not open file \"%s\": %m",
2085                                                                 path)));
2086                 }
2087
2088                 ReorderBufferSerializeChange(rb, txn, fd, change);
2089                 dlist_delete(&change->node);
2090                 ReorderBufferReturnChange(rb, change);
2091
2092                 spilled++;
2093         }
2094
2095         Assert(spilled == txn->nentries_mem);
2096         Assert(dlist_is_empty(&txn->changes));
2097         txn->nentries_mem = 0;
2098         txn->serialized = true;
2099
2100         if (fd != -1)
2101                 CloseTransientFile(fd);
2102 }
2103
2104 /*
2105  * Serialize individual change to disk.
2106  */
2107 static void
2108 ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
2109                                                          int fd, ReorderBufferChange *change)
2110 {
2111         ReorderBufferDiskChange *ondisk;
2112         Size            sz = sizeof(ReorderBufferDiskChange);
2113
2114         ReorderBufferSerializeReserve(rb, sz);
2115
2116         ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2117         memcpy(&ondisk->change, change, sizeof(ReorderBufferChange));
2118
2119         switch (change->action)
2120         {
2121                         /* fall through these, they're all similar enough */
2122                 case REORDER_BUFFER_CHANGE_INSERT:
2123                 case REORDER_BUFFER_CHANGE_UPDATE:
2124                 case REORDER_BUFFER_CHANGE_DELETE:
2125                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2126                         {
2127                                 char       *data;
2128                                 ReorderBufferTupleBuf *oldtup,
2129                                                    *newtup;
2130                                 Size            oldlen = 0;
2131                                 Size            newlen = 0;
2132
2133                                 oldtup = change->data.tp.oldtuple;
2134                                 newtup = change->data.tp.newtuple;
2135
2136                                 if (oldtup)
2137                                 {
2138                                         sz += sizeof(HeapTupleData);
2139                                         oldlen = oldtup->tuple.t_len;
2140                                         sz += oldlen;
2141                                 }
2142
2143                                 if (newtup)
2144                                 {
2145                                         sz += sizeof(HeapTupleData);
2146                                         newlen = newtup->tuple.t_len;
2147                                         sz += newlen;
2148                                 }
2149
2150                                 /* make sure we have enough space */
2151                                 ReorderBufferSerializeReserve(rb, sz);
2152
2153                                 data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
2154                                 /* might have been reallocated above */
2155                                 ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2156
2157                                 if (oldlen)
2158                                 {
2159                                         memcpy(data, &oldtup->tuple, sizeof(HeapTupleData));
2160                                         data += sizeof(HeapTupleData);
2161
2162                                         memcpy(data, oldtup->tuple.t_data, oldlen);
2163                                         data += oldlen;
2164                                 }
2165
2166                                 if (newlen)
2167                                 {
2168                                         memcpy(data, &newtup->tuple, sizeof(HeapTupleData));
2169                                         data += sizeof(HeapTupleData);
2170
2171                                         memcpy(data, newtup->tuple.t_data, newlen);
2172                                         data += newlen;
2173                                 }
2174                                 break;
2175                         }
2176                 case REORDER_BUFFER_CHANGE_MESSAGE:
2177                         {
2178                                 char       *data;
2179                                 Size            prefix_size = strlen(change->data.msg.prefix) + 1;
2180
2181                                 sz += prefix_size + change->data.msg.message_size +
2182                                         sizeof(Size) + sizeof(Size);
2183                                 ReorderBufferSerializeReserve(rb, sz);
2184
2185                                 data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
2186
2187                                 /* might have been reallocated above */
2188                                 ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2189
2190                                 /* write the prefix including the size */
2191                                 memcpy(data, &prefix_size, sizeof(Size));
2192                                 data += sizeof(Size);
2193                                 memcpy(data, change->data.msg.prefix,
2194                                            prefix_size);
2195                                 data += prefix_size;
2196
2197                                 /* write the message including the size */
2198                                 memcpy(data, &change->data.msg.message_size, sizeof(Size));
2199                                 data += sizeof(Size);
2200                                 memcpy(data, change->data.msg.message,
2201                                            change->data.msg.message_size);
2202                                 data += change->data.msg.message_size;
2203
2204                                 break;
2205                         }
2206                 case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2207                         {
2208                                 Snapshot        snap;
2209                                 char       *data;
2210
2211                                 snap = change->data.snapshot;
2212
2213                                 sz += sizeof(SnapshotData) +
2214                                         sizeof(TransactionId) * snap->xcnt +
2215                                         sizeof(TransactionId) * snap->subxcnt
2216                                         ;
2217
2218                                 /* make sure we have enough space */
2219                                 ReorderBufferSerializeReserve(rb, sz);
2220                                 data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
2221                                 /* might have been reallocated above */
2222                                 ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2223
2224                                 memcpy(data, snap, sizeof(SnapshotData));
2225                                 data += sizeof(SnapshotData);
2226
2227                                 if (snap->xcnt)
2228                                 {
2229                                         memcpy(data, snap->xip,
2230                                                    sizeof(TransactionId) * snap->xcnt);
2231                                         data += sizeof(TransactionId) * snap->xcnt;
2232                                 }
2233
2234                                 if (snap->subxcnt)
2235                                 {
2236                                         memcpy(data, snap->subxip,
2237                                                    sizeof(TransactionId) * snap->subxcnt);
2238                                         data += sizeof(TransactionId) * snap->subxcnt;
2239                                 }
2240                                 break;
2241                         }
2242                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2243                 case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2244                 case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2245                         /* ReorderBufferChange contains everything important */
2246                         break;
2247         }
2248
2249         ondisk->size = sz;
2250
2251         pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_WRITE);
2252         if (write(fd, rb->outbuf, ondisk->size) != ondisk->size)
2253         {
2254                 int                     save_errno = errno;
2255
2256                 CloseTransientFile(fd);
2257                 errno = save_errno;
2258                 ereport(ERROR,
2259                                 (errcode_for_file_access(),
2260                                  errmsg("could not write to data file for XID %u: %m",
2261                                                 txn->xid)));
2262         }
2263         pgstat_report_wait_end();
2264
2265         Assert(ondisk->change.action == change->action);
2266 }
2267
2268 /*
2269  * Restore a number of changes spilled to disk back into memory.
2270  */
2271 static Size
2272 ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
2273                                                         int *fd, XLogSegNo *segno)
2274 {
2275         Size            restored = 0;
2276         XLogSegNo       last_segno;
2277         dlist_mutable_iter cleanup_iter;
2278
2279         Assert(txn->first_lsn != InvalidXLogRecPtr);
2280         Assert(txn->final_lsn != InvalidXLogRecPtr);
2281
2282         /* free current entries, so we have memory for more */
2283         dlist_foreach_modify(cleanup_iter, &txn->changes)
2284         {
2285                 ReorderBufferChange *cleanup =
2286                 dlist_container(ReorderBufferChange, node, cleanup_iter.cur);
2287
2288                 dlist_delete(&cleanup->node);
2289                 ReorderBufferReturnChange(rb, cleanup);
2290         }
2291         txn->nentries_mem = 0;
2292         Assert(dlist_is_empty(&txn->changes));
2293
2294         XLByteToSeg(txn->final_lsn, last_segno, wal_segment_size);
2295
2296         while (restored < max_changes_in_memory && *segno <= last_segno)
2297         {
2298                 int                     readBytes;
2299                 ReorderBufferDiskChange *ondisk;
2300
2301                 if (*fd == -1)
2302                 {
2303                         XLogRecPtr      recptr;
2304                         char            path[MAXPGPATH];
2305
2306                         /* first time in */
2307                         if (*segno == 0)
2308                         {
2309                                 XLByteToSeg(txn->first_lsn, *segno, wal_segment_size);
2310                         }
2311
2312                         Assert(*segno != 0 || dlist_is_empty(&txn->changes));
2313                         XLogSegNoOffsetToRecPtr(*segno, 0, recptr, wal_segment_size);
2314
2315                         /*
2316                          * No need to care about TLIs here, only used during a single run,
2317                          * so each LSN only maps to a specific WAL record.
2318                          */
2319                         sprintf(path, "pg_replslot/%s/xid-%u-lsn-%X-%X.snap",
2320                                         NameStr(MyReplicationSlot->data.name), txn->xid,
2321                                         (uint32) (recptr >> 32), (uint32) recptr);
2322
2323                         *fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
2324                         if (*fd < 0 && errno == ENOENT)
2325                         {
2326                                 *fd = -1;
2327                                 (*segno)++;
2328                                 continue;
2329                         }
2330                         else if (*fd < 0)
2331                                 ereport(ERROR,
2332                                                 (errcode_for_file_access(),
2333                                                  errmsg("could not open file \"%s\": %m",
2334                                                                 path)));
2335
2336                 }
2337
2338                 /*
2339                  * Read the statically sized part of a change which has information
2340                  * about the total size. If we couldn't read a record, we're at the
2341                  * end of this file.
2342                  */
2343                 ReorderBufferSerializeReserve(rb, sizeof(ReorderBufferDiskChange));
2344                 pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_READ);
2345                 readBytes = read(*fd, rb->outbuf, sizeof(ReorderBufferDiskChange));
2346                 pgstat_report_wait_end();
2347
2348                 /* eof */
2349                 if (readBytes == 0)
2350                 {
2351                         CloseTransientFile(*fd);
2352                         *fd = -1;
2353                         (*segno)++;
2354                         continue;
2355                 }
2356                 else if (readBytes < 0)
2357                         ereport(ERROR,
2358                                         (errcode_for_file_access(),
2359                                          errmsg("could not read from reorderbuffer spill file: %m")));
2360                 else if (readBytes != sizeof(ReorderBufferDiskChange))
2361                         ereport(ERROR,
2362                                         (errcode_for_file_access(),
2363                                          errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
2364                                                         readBytes,
2365                                                         (uint32) sizeof(ReorderBufferDiskChange))));
2366
2367                 ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2368
2369                 ReorderBufferSerializeReserve(rb,
2370                                                                           sizeof(ReorderBufferDiskChange) + ondisk->size);
2371                 ondisk = (ReorderBufferDiskChange *) rb->outbuf;
2372
2373                 pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_READ);
2374                 readBytes = read(*fd, rb->outbuf + sizeof(ReorderBufferDiskChange),
2375                                                  ondisk->size - sizeof(ReorderBufferDiskChange));
2376                 pgstat_report_wait_end();
2377
2378                 if (readBytes < 0)
2379                         ereport(ERROR,
2380                                         (errcode_for_file_access(),
2381                                          errmsg("could not read from reorderbuffer spill file: %m")));
2382                 else if (readBytes != ondisk->size - sizeof(ReorderBufferDiskChange))
2383                         ereport(ERROR,
2384                                         (errcode_for_file_access(),
2385                                          errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
2386                                                         readBytes,
2387                                                         (uint32) (ondisk->size - sizeof(ReorderBufferDiskChange)))));
2388
2389                 /*
2390                  * ok, read a full change from disk, now restore it into proper
2391                  * in-memory format
2392                  */
2393                 ReorderBufferRestoreChange(rb, txn, rb->outbuf);
2394                 restored++;
2395         }
2396
2397         return restored;
2398 }
2399
2400 /*
2401  * Convert change from its on-disk format to in-memory format and queue it onto
2402  * the TXN's ->changes list.
2403  *
2404  * Note: although "data" is declared char*, at entry it points to a
2405  * maxalign'd buffer, making it safe in most of this function to assume
2406  * that the pointed-to data is suitably aligned for direct access.
2407  */
2408 static void
2409 ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
2410                                                    char *data)
2411 {
2412         ReorderBufferDiskChange *ondisk;
2413         ReorderBufferChange *change;
2414
2415         ondisk = (ReorderBufferDiskChange *) data;
2416
2417         change = ReorderBufferGetChange(rb);
2418
2419         /* copy static part */
2420         memcpy(change, &ondisk->change, sizeof(ReorderBufferChange));
2421
2422         data += sizeof(ReorderBufferDiskChange);
2423
2424         /* restore individual stuff */
2425         switch (change->action)
2426         {
2427                         /* fall through these, they're all similar enough */
2428                 case REORDER_BUFFER_CHANGE_INSERT:
2429                 case REORDER_BUFFER_CHANGE_UPDATE:
2430                 case REORDER_BUFFER_CHANGE_DELETE:
2431                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2432                         if (change->data.tp.oldtuple)
2433                         {
2434                                 uint32          tuplelen = ((HeapTuple) data)->t_len;
2435
2436                                 change->data.tp.oldtuple =
2437                                         ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
2438
2439                                 /* restore ->tuple */
2440                                 memcpy(&change->data.tp.oldtuple->tuple, data,
2441                                            sizeof(HeapTupleData));
2442                                 data += sizeof(HeapTupleData);
2443
2444                                 /* reset t_data pointer into the new tuplebuf */
2445                                 change->data.tp.oldtuple->tuple.t_data =
2446                                         ReorderBufferTupleBufData(change->data.tp.oldtuple);
2447
2448                                 /* restore tuple data itself */
2449                                 memcpy(change->data.tp.oldtuple->tuple.t_data, data, tuplelen);
2450                                 data += tuplelen;
2451                         }
2452
2453                         if (change->data.tp.newtuple)
2454                         {
2455                                 /* here, data might not be suitably aligned! */
2456                                 uint32          tuplelen;
2457
2458                                 memcpy(&tuplelen, data + offsetof(HeapTupleData, t_len),
2459                                            sizeof(uint32));
2460
2461                                 change->data.tp.newtuple =
2462                                         ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
2463
2464                                 /* restore ->tuple */
2465                                 memcpy(&change->data.tp.newtuple->tuple, data,
2466                                            sizeof(HeapTupleData));
2467                                 data += sizeof(HeapTupleData);
2468
2469                                 /* reset t_data pointer into the new tuplebuf */
2470                                 change->data.tp.newtuple->tuple.t_data =
2471                                         ReorderBufferTupleBufData(change->data.tp.newtuple);
2472
2473                                 /* restore tuple data itself */
2474                                 memcpy(change->data.tp.newtuple->tuple.t_data, data, tuplelen);
2475                                 data += tuplelen;
2476                         }
2477
2478                         break;
2479                 case REORDER_BUFFER_CHANGE_MESSAGE:
2480                         {
2481                                 Size            prefix_size;
2482
2483                                 /* read prefix */
2484                                 memcpy(&prefix_size, data, sizeof(Size));
2485                                 data += sizeof(Size);
2486                                 change->data.msg.prefix = MemoryContextAlloc(rb->context,
2487                                                                                                                          prefix_size);
2488                                 memcpy(change->data.msg.prefix, data, prefix_size);
2489                                 Assert(change->data.msg.prefix[prefix_size - 1] == '\0');
2490                                 data += prefix_size;
2491
2492                                 /* read the message */
2493                                 memcpy(&change->data.msg.message_size, data, sizeof(Size));
2494                                 data += sizeof(Size);
2495                                 change->data.msg.message = MemoryContextAlloc(rb->context,
2496                                                                                                                           change->data.msg.message_size);
2497                                 memcpy(change->data.msg.message, data,
2498                                            change->data.msg.message_size);
2499                                 data += change->data.msg.message_size;
2500
2501                                 break;
2502                         }
2503                 case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2504                         {
2505                                 Snapshot        oldsnap;
2506                                 Snapshot        newsnap;
2507                                 Size            size;
2508
2509                                 oldsnap = (Snapshot) data;
2510
2511                                 size = sizeof(SnapshotData) +
2512                                         sizeof(TransactionId) * oldsnap->xcnt +
2513                                         sizeof(TransactionId) * (oldsnap->subxcnt + 0);
2514
2515                                 change->data.snapshot = MemoryContextAllocZero(rb->context, size);
2516
2517                                 newsnap = change->data.snapshot;
2518
2519                                 memcpy(newsnap, data, size);
2520                                 newsnap->xip = (TransactionId *)
2521                                         (((char *) newsnap) + sizeof(SnapshotData));
2522                                 newsnap->subxip = newsnap->xip + newsnap->xcnt;
2523                                 newsnap->copied = true;
2524                                 break;
2525                         }
2526                         /* the base struct contains all the data, easy peasy */
2527                 case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2528                 case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2529                 case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2530                         break;
2531         }
2532
2533         dlist_push_tail(&txn->changes, &change->node);
2534         txn->nentries_mem++;
2535 }
2536
2537 /*
2538  * Remove all on-disk stored for the passed in transaction.
2539  */
2540 static void
2541 ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn)
2542 {
2543         XLogSegNo       first;
2544         XLogSegNo       cur;
2545         XLogSegNo       last;
2546
2547         Assert(txn->first_lsn != InvalidXLogRecPtr);
2548         Assert(txn->final_lsn != InvalidXLogRecPtr);
2549
2550         XLByteToSeg(txn->first_lsn, first, wal_segment_size);
2551         XLByteToSeg(txn->final_lsn, last, wal_segment_size);
2552
2553         /* iterate over all possible filenames, and delete them */
2554         for (cur = first; cur <= last; cur++)
2555         {
2556                 char            path[MAXPGPATH];
2557                 XLogRecPtr      recptr;
2558
2559                 XLogSegNoOffsetToRecPtr(cur, 0, recptr, wal_segment_size);
2560
2561                 sprintf(path, "pg_replslot/%s/xid-%u-lsn-%X-%X.snap",
2562                                 NameStr(MyReplicationSlot->data.name), txn->xid,
2563                                 (uint32) (recptr >> 32), (uint32) recptr);
2564                 if (unlink(path) != 0 && errno != ENOENT)
2565                         ereport(ERROR,
2566                                         (errcode_for_file_access(),
2567                                          errmsg("could not remove file \"%s\": %m", path)));
2568         }
2569 }
2570
2571 /*
2572  * Delete all data spilled to disk after we've restarted/crashed. It will be
2573  * recreated when the respective slots are reused.
2574  */
2575 void
2576 StartupReorderBuffer(void)
2577 {
2578         DIR                *logical_dir;
2579         struct dirent *logical_de;
2580
2581         DIR                *spill_dir;
2582         struct dirent *spill_de;
2583
2584         logical_dir = AllocateDir("pg_replslot");
2585         while ((logical_de = ReadDir(logical_dir, "pg_replslot")) != NULL)
2586         {
2587                 struct stat statbuf;
2588                 char            path[MAXPGPATH * 2 + 12];
2589
2590                 if (strcmp(logical_de->d_name, ".") == 0 ||
2591                         strcmp(logical_de->d_name, "..") == 0)
2592                         continue;
2593
2594                 /* if it cannot be a slot, skip the directory */
2595                 if (!ReplicationSlotValidateName(logical_de->d_name, DEBUG2))
2596                         continue;
2597
2598                 /*
2599                  * ok, has to be a surviving logical slot, iterate and delete
2600                  * everything starting with xid-*
2601                  */
2602                 sprintf(path, "pg_replslot/%s", logical_de->d_name);
2603
2604                 /* we're only creating directories here, skip if it's not our's */
2605                 if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
2606                         continue;
2607
2608                 spill_dir = AllocateDir(path);
2609                 while ((spill_de = ReadDir(spill_dir, path)) != NULL)
2610                 {
2611                         if (strcmp(spill_de->d_name, ".") == 0 ||
2612                                 strcmp(spill_de->d_name, "..") == 0)
2613                                 continue;
2614
2615                         /* only look at names that can be ours */
2616                         if (strncmp(spill_de->d_name, "xid", 3) == 0)
2617                         {
2618                                 sprintf(path, "pg_replslot/%s/%s", logical_de->d_name,
2619                                                 spill_de->d_name);
2620
2621                                 if (unlink(path) != 0)
2622                                         ereport(PANIC,
2623                                                         (errcode_for_file_access(),
2624                                                          errmsg("could not remove file \"%s\": %m",
2625                                                                         path)));
2626                         }
2627                 }
2628                 FreeDir(spill_dir);
2629         }
2630         FreeDir(logical_dir);
2631 }
2632
2633 /* ---------------------------------------
2634  * toast reassembly support
2635  * ---------------------------------------
2636  */
2637
2638 /*
2639  * Initialize per tuple toast reconstruction support.
2640  */
2641 static void
2642 ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
2643 {
2644         HASHCTL         hash_ctl;
2645
2646         Assert(txn->toast_hash == NULL);
2647
2648         memset(&hash_ctl, 0, sizeof(hash_ctl));
2649         hash_ctl.keysize = sizeof(Oid);
2650         hash_ctl.entrysize = sizeof(ReorderBufferToastEnt);
2651         hash_ctl.hcxt = rb->context;
2652         txn->toast_hash = hash_create("ReorderBufferToastHash", 5, &hash_ctl,
2653                                                                   HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
2654 }
2655
2656 /*
2657  * Per toast-chunk handling for toast reconstruction
2658  *
2659  * Appends a toast chunk so we can reconstruct it when the tuple "owning" the
2660  * toasted Datum comes along.
2661  */
2662 static void
2663 ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
2664                                                           Relation relation, ReorderBufferChange *change)
2665 {
2666         ReorderBufferToastEnt *ent;
2667         ReorderBufferTupleBuf *newtup;
2668         bool            found;
2669         int32           chunksize;
2670         bool            isnull;
2671         Pointer         chunk;
2672         TupleDesc       desc = RelationGetDescr(relation);
2673         Oid                     chunk_id;
2674         int32           chunk_seq;
2675
2676         if (txn->toast_hash == NULL)
2677                 ReorderBufferToastInitHash(rb, txn);
2678
2679         Assert(IsToastRelation(relation));
2680
2681         newtup = change->data.tp.newtuple;
2682         chunk_id = DatumGetObjectId(fastgetattr(&newtup->tuple, 1, desc, &isnull));
2683         Assert(!isnull);
2684         chunk_seq = DatumGetInt32(fastgetattr(&newtup->tuple, 2, desc, &isnull));
2685         Assert(!isnull);
2686
2687         ent = (ReorderBufferToastEnt *)
2688                 hash_search(txn->toast_hash,
2689                                         (void *) &chunk_id,
2690                                         HASH_ENTER,
2691                                         &found);
2692
2693         if (!found)
2694         {
2695                 Assert(ent->chunk_id == chunk_id);
2696                 ent->num_chunks = 0;
2697                 ent->last_chunk_seq = 0;
2698                 ent->size = 0;
2699                 ent->reconstructed = NULL;
2700                 dlist_init(&ent->chunks);
2701
2702                 if (chunk_seq != 0)
2703                         elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq 0",
2704                                  chunk_seq, chunk_id);
2705         }
2706         else if (found && chunk_seq != ent->last_chunk_seq + 1)
2707                 elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq %d",
2708                          chunk_seq, chunk_id, ent->last_chunk_seq + 1);
2709
2710         chunk = DatumGetPointer(fastgetattr(&newtup->tuple, 3, desc, &isnull));
2711         Assert(!isnull);
2712
2713         /* calculate size so we can allocate the right size at once later */
2714         if (!VARATT_IS_EXTENDED(chunk))
2715                 chunksize = VARSIZE(chunk) - VARHDRSZ;
2716         else if (VARATT_IS_SHORT(chunk))
2717                 /* could happen due to heap_form_tuple doing its thing */
2718                 chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
2719         else
2720                 elog(ERROR, "unexpected type of toast chunk");
2721
2722         ent->size += chunksize;
2723         ent->last_chunk_seq = chunk_seq;
2724         ent->num_chunks++;
2725         dlist_push_tail(&ent->chunks, &change->node);
2726 }
2727
2728 /*
2729  * Rejigger change->newtuple to point to in-memory toast tuples instead to
2730  * on-disk toast tuples that may not longer exist (think DROP TABLE or VACUUM).
2731  *
2732  * We cannot replace unchanged toast tuples though, so those will still point
2733  * to on-disk toast data.
2734  */
2735 static void
2736 ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
2737                                                   Relation relation, ReorderBufferChange *change)
2738 {
2739         TupleDesc       desc;
2740         int                     natt;
2741         Datum      *attrs;
2742         bool       *isnull;
2743         bool       *free;
2744         HeapTuple       tmphtup;
2745         Relation        toast_rel;
2746         TupleDesc       toast_desc;
2747         MemoryContext oldcontext;
2748         ReorderBufferTupleBuf *newtup;
2749
2750         /* no toast tuples changed */
2751         if (txn->toast_hash == NULL)
2752                 return;
2753
2754         oldcontext = MemoryContextSwitchTo(rb->context);
2755
2756         /* we should only have toast tuples in an INSERT or UPDATE */
2757         Assert(change->data.tp.newtuple);
2758
2759         desc = RelationGetDescr(relation);
2760
2761         toast_rel = RelationIdGetRelation(relation->rd_rel->reltoastrelid);
2762         toast_desc = RelationGetDescr(toast_rel);
2763
2764         /* should we allocate from stack instead? */
2765         attrs = palloc0(sizeof(Datum) * desc->natts);
2766         isnull = palloc0(sizeof(bool) * desc->natts);
2767         free = palloc0(sizeof(bool) * desc->natts);
2768
2769         newtup = change->data.tp.newtuple;
2770
2771         heap_deform_tuple(&newtup->tuple, desc, attrs, isnull);
2772
2773         for (natt = 0; natt < desc->natts; natt++)
2774         {
2775                 Form_pg_attribute attr = TupleDescAttr(desc, natt);
2776                 ReorderBufferToastEnt *ent;
2777                 struct varlena *varlena;
2778
2779                 /* va_rawsize is the size of the original datum -- including header */
2780                 struct varatt_external toast_pointer;
2781                 struct varatt_indirect redirect_pointer;
2782                 struct varlena *new_datum = NULL;
2783                 struct varlena *reconstructed;
2784                 dlist_iter      it;
2785                 Size            data_done = 0;
2786
2787                 /* system columns aren't toasted */
2788                 if (attr->attnum < 0)
2789                         continue;
2790
2791                 if (attr->attisdropped)
2792                         continue;
2793
2794                 /* not a varlena datatype */
2795                 if (attr->attlen != -1)
2796                         continue;
2797
2798                 /* no data */
2799                 if (isnull[natt])
2800                         continue;
2801
2802                 /* ok, we know we have a toast datum */
2803                 varlena = (struct varlena *) DatumGetPointer(attrs[natt]);
2804
2805                 /* no need to do anything if the tuple isn't external */
2806                 if (!VARATT_IS_EXTERNAL(varlena))
2807                         continue;
2808
2809                 VARATT_EXTERNAL_GET_POINTER(toast_pointer, varlena);
2810
2811                 /*
2812                  * Check whether the toast tuple changed, replace if so.
2813                  */
2814                 ent = (ReorderBufferToastEnt *)
2815                         hash_search(txn->toast_hash,
2816                                                 (void *) &toast_pointer.va_valueid,
2817                                                 HASH_FIND,
2818                                                 NULL);
2819                 if (ent == NULL)
2820                         continue;
2821
2822                 new_datum =
2823                         (struct varlena *) palloc0(INDIRECT_POINTER_SIZE);
2824
2825                 free[natt] = true;
2826
2827                 reconstructed = palloc0(toast_pointer.va_rawsize);
2828
2829                 ent->reconstructed = reconstructed;
2830
2831                 /* stitch toast tuple back together from its parts */
2832                 dlist_foreach(it, &ent->chunks)
2833                 {
2834                         bool            isnull;
2835                         ReorderBufferChange *cchange;
2836                         ReorderBufferTupleBuf *ctup;
2837                         Pointer         chunk;
2838
2839                         cchange = dlist_container(ReorderBufferChange, node, it.cur);
2840                         ctup = cchange->data.tp.newtuple;
2841                         chunk = DatumGetPointer(
2842                                                                         fastgetattr(&ctup->tuple, 3, toast_desc, &isnull));
2843
2844                         Assert(!isnull);
2845                         Assert(!VARATT_IS_EXTERNAL(chunk));
2846                         Assert(!VARATT_IS_SHORT(chunk));
2847
2848                         memcpy(VARDATA(reconstructed) + data_done,
2849                                    VARDATA(chunk),
2850                                    VARSIZE(chunk) - VARHDRSZ);
2851                         data_done += VARSIZE(chunk) - VARHDRSZ;
2852                 }
2853                 Assert(data_done == toast_pointer.va_extsize);
2854
2855                 /* make sure its marked as compressed or not */
2856                 if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
2857                         SET_VARSIZE_COMPRESSED(reconstructed, data_done + VARHDRSZ);
2858                 else
2859                         SET_VARSIZE(reconstructed, data_done + VARHDRSZ);
2860
2861                 memset(&redirect_pointer, 0, sizeof(redirect_pointer));
2862                 redirect_pointer.pointer = reconstructed;
2863
2864                 SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
2865                 memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
2866                            sizeof(redirect_pointer));
2867
2868                 attrs[natt] = PointerGetDatum(new_datum);
2869         }
2870
2871         /*
2872          * Build tuple in separate memory & copy tuple back into the tuplebuf
2873          * passed to the output plugin. We can't directly heap_fill_tuple() into
2874          * the tuplebuf because attrs[] will point back into the current content.
2875          */
2876         tmphtup = heap_form_tuple(desc, attrs, isnull);
2877         Assert(newtup->tuple.t_len <= MaxHeapTupleSize);
2878         Assert(ReorderBufferTupleBufData(newtup) == newtup->tuple.t_data);
2879
2880         memcpy(newtup->tuple.t_data, tmphtup->t_data, tmphtup->t_len);
2881         newtup->tuple.t_len = tmphtup->t_len;
2882
2883         /*
2884          * free resources we won't further need, more persistent stuff will be
2885          * free'd in ReorderBufferToastReset().
2886          */
2887         RelationClose(toast_rel);
2888         pfree(tmphtup);
2889         for (natt = 0; natt < desc->natts; natt++)
2890         {
2891                 if (free[natt])
2892                         pfree(DatumGetPointer(attrs[natt]));
2893         }
2894         pfree(attrs);
2895         pfree(free);
2896         pfree(isnull);
2897
2898         MemoryContextSwitchTo(oldcontext);
2899 }
2900
2901 /*
2902  * Free all resources allocated for toast reconstruction.
2903  */
2904 static void
2905 ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn)
2906 {
2907         HASH_SEQ_STATUS hstat;
2908         ReorderBufferToastEnt *ent;
2909
2910         if (txn->toast_hash == NULL)
2911                 return;
2912
2913         /* sequentially walk over the hash and free everything */
2914         hash_seq_init(&hstat, txn->toast_hash);
2915         while ((ent = (ReorderBufferToastEnt *) hash_seq_search(&hstat)) != NULL)
2916         {
2917                 dlist_mutable_iter it;
2918
2919                 if (ent->reconstructed != NULL)
2920                         pfree(ent->reconstructed);
2921
2922                 dlist_foreach_modify(it, &ent->chunks)
2923                 {
2924                         ReorderBufferChange *change =
2925                         dlist_container(ReorderBufferChange, node, it.cur);
2926
2927                         dlist_delete(&change->node);
2928                         ReorderBufferReturnChange(rb, change);
2929                 }
2930         }
2931
2932         hash_destroy(txn->toast_hash);
2933         txn->toast_hash = NULL;
2934 }
2935
2936
2937 /* ---------------------------------------
2938  * Visibility support for logical decoding
2939  *
2940  *
2941  * Lookup actual cmin/cmax values when using decoding snapshot. We can't
2942  * always rely on stored cmin/cmax values because of two scenarios:
2943  *
2944  * * A tuple got changed multiple times during a single transaction and thus
2945  *       has got a combocid. Combocid's are only valid for the duration of a
2946  *       single transaction.
2947  * * A tuple with a cmin but no cmax (and thus no combocid) got
2948  *       deleted/updated in another transaction than the one which created it
2949  *       which we are looking at right now. As only one of cmin, cmax or combocid
2950  *       is actually stored in the heap we don't have access to the value we
2951  *       need anymore.
2952  *
2953  * To resolve those problems we have a per-transaction hash of (cmin,
2954  * cmax) tuples keyed by (relfilenode, ctid) which contains the actual
2955  * (cmin, cmax) values. That also takes care of combocids by simply
2956  * not caring about them at all. As we have the real cmin/cmax values
2957  * combocids aren't interesting.
2958  *
2959  * As we only care about catalog tuples here the overhead of this
2960  * hashtable should be acceptable.
2961  *
2962  * Heap rewrites complicate this a bit, check rewriteheap.c for
2963  * details.
2964  * -------------------------------------------------------------------------
2965  */
2966
2967 /* struct for qsort()ing mapping files by lsn somewhat efficiently */
2968 typedef struct RewriteMappingFile
2969 {
2970         XLogRecPtr      lsn;
2971         char            fname[MAXPGPATH];
2972 } RewriteMappingFile;
2973
2974 #if NOT_USED
2975 static void
2976 DisplayMapping(HTAB *tuplecid_data)
2977 {
2978         HASH_SEQ_STATUS hstat;
2979         ReorderBufferTupleCidEnt *ent;
2980
2981         hash_seq_init(&hstat, tuplecid_data);
2982         while ((ent = (ReorderBufferTupleCidEnt *) hash_seq_search(&hstat)) != NULL)
2983         {
2984                 elog(DEBUG3, "mapping: node: %u/%u/%u tid: %u/%u cmin: %u, cmax: %u",
2985                          ent->key.relnode.dbNode,
2986                          ent->key.relnode.spcNode,
2987                          ent->key.relnode.relNode,
2988                          ItemPointerGetBlockNumber(&ent->key.tid),
2989                          ItemPointerGetOffsetNumber(&ent->key.tid),
2990                          ent->cmin,
2991                          ent->cmax
2992                         );
2993         }
2994 }
2995 #endif
2996
2997 /*
2998  * Apply a single mapping file to tuplecid_data.
2999  *
3000  * The mapping file has to have been verified to be a) committed b) for our
3001  * transaction c) applied in LSN order.
3002  */
3003 static void
3004 ApplyLogicalMappingFile(HTAB *tuplecid_data, Oid relid, const char *fname)
3005 {
3006         char            path[MAXPGPATH];
3007         int                     fd;
3008         int                     readBytes;
3009         LogicalRewriteMappingData map;
3010
3011         sprintf(path, "pg_logical/mappings/%s", fname);
3012         fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3013         if (fd < 0)
3014                 ereport(ERROR,
3015                                 (errcode_for_file_access(),
3016                                  errmsg("could not open file \"%s\": %m", path)));
3017
3018         while (true)
3019         {
3020                 ReorderBufferTupleCidKey key;
3021                 ReorderBufferTupleCidEnt *ent;
3022                 ReorderBufferTupleCidEnt *new_ent;
3023                 bool            found;
3024
3025                 /* be careful about padding */
3026                 memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
3027
3028                 /* read all mappings till the end of the file */
3029                 pgstat_report_wait_start(WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ);
3030                 readBytes = read(fd, &map, sizeof(LogicalRewriteMappingData));
3031                 pgstat_report_wait_end();
3032
3033                 if (readBytes < 0)
3034                         ereport(ERROR,
3035                                         (errcode_for_file_access(),
3036                                          errmsg("could not read file \"%s\": %m",
3037                                                         path)));
3038                 else if (readBytes == 0)        /* EOF */
3039                         break;
3040                 else if (readBytes != sizeof(LogicalRewriteMappingData))
3041                         ereport(ERROR,
3042                                         (errcode_for_file_access(),
3043                                          errmsg("could not read from file \"%s\": read %d instead of %d bytes",
3044                                                         path, readBytes,
3045                                                         (int32) sizeof(LogicalRewriteMappingData))));
3046
3047                 key.relnode = map.old_node;
3048                 ItemPointerCopy(&map.old_tid,
3049                                                 &key.tid);
3050
3051
3052                 ent = (ReorderBufferTupleCidEnt *)
3053                         hash_search(tuplecid_data,
3054                                                 (void *) &key,
3055                                                 HASH_FIND,
3056                                                 NULL);
3057
3058                 /* no existing mapping, no need to update */
3059                 if (!ent)
3060                         continue;
3061
3062                 key.relnode = map.new_node;
3063                 ItemPointerCopy(&map.new_tid,
3064                                                 &key.tid);
3065
3066                 new_ent = (ReorderBufferTupleCidEnt *)
3067                         hash_search(tuplecid_data,
3068                                                 (void *) &key,
3069                                                 HASH_ENTER,
3070                                                 &found);
3071
3072                 if (found)
3073                 {
3074                         /*
3075                          * Make sure the existing mapping makes sense. We sometime update
3076                          * old records that did not yet have a cmax (e.g. pg_class' own
3077                          * entry while rewriting it) during rewrites, so allow that.
3078                          */
3079                         Assert(ent->cmin == InvalidCommandId || ent->cmin == new_ent->cmin);
3080                         Assert(ent->cmax == InvalidCommandId || ent->cmax == new_ent->cmax);
3081                 }
3082                 else
3083                 {
3084                         /* update mapping */
3085                         new_ent->cmin = ent->cmin;
3086                         new_ent->cmax = ent->cmax;
3087                         new_ent->combocid = ent->combocid;
3088                 }
3089         }
3090 }
3091
3092
3093 /*
3094  * Check whether the TransactionOId 'xid' is in the pre-sorted array 'xip'.
3095  */
3096 static bool
3097 TransactionIdInArray(TransactionId xid, TransactionId *xip, Size num)
3098 {
3099         return bsearch(&xid, xip, num,
3100                                    sizeof(TransactionId), xidComparator) != NULL;
3101 }
3102
3103 /*
3104  * qsort() comparator for sorting RewriteMappingFiles in LSN order.
3105  */
3106 static int
3107 file_sort_by_lsn(const void *a_p, const void *b_p)
3108 {
3109         RewriteMappingFile *a = *(RewriteMappingFile **) a_p;
3110         RewriteMappingFile *b = *(RewriteMappingFile **) b_p;
3111
3112         if (a->lsn < b->lsn)
3113                 return -1;
3114         else if (a->lsn > b->lsn)
3115                 return 1;
3116         return 0;
3117 }
3118
3119 /*
3120  * Apply any existing logical remapping files if there are any targeted at our
3121  * transaction for relid.
3122  */
3123 static void
3124 UpdateLogicalMappings(HTAB *tuplecid_data, Oid relid, Snapshot snapshot)
3125 {
3126         DIR                *mapping_dir;
3127         struct dirent *mapping_de;
3128         List       *files = NIL;
3129         ListCell   *file;
3130         RewriteMappingFile **files_a;
3131         size_t          off;
3132         Oid                     dboid = IsSharedRelation(relid) ? InvalidOid : MyDatabaseId;
3133
3134         mapping_dir = AllocateDir("pg_logical/mappings");
3135         while ((mapping_de = ReadDir(mapping_dir, "pg_logical/mappings")) != NULL)
3136         {
3137                 Oid                     f_dboid;
3138                 Oid                     f_relid;
3139                 TransactionId f_mapped_xid;
3140                 TransactionId f_create_xid;
3141                 XLogRecPtr      f_lsn;
3142                 uint32          f_hi,
3143                                         f_lo;
3144                 RewriteMappingFile *f;
3145
3146                 if (strcmp(mapping_de->d_name, ".") == 0 ||
3147                         strcmp(mapping_de->d_name, "..") == 0)
3148                         continue;
3149
3150                 /* Ignore files that aren't ours */
3151                 if (strncmp(mapping_de->d_name, "map-", 4) != 0)
3152                         continue;
3153
3154                 if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
3155                                    &f_dboid, &f_relid, &f_hi, &f_lo,
3156                                    &f_mapped_xid, &f_create_xid) != 6)
3157                         elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
3158
3159                 f_lsn = ((uint64) f_hi) << 32 | f_lo;
3160
3161                 /* mapping for another database */
3162                 if (f_dboid != dboid)
3163                         continue;
3164
3165                 /* mapping for another relation */
3166                 if (f_relid != relid)
3167                         continue;
3168
3169                 /* did the creating transaction abort? */
3170                 if (!TransactionIdDidCommit(f_create_xid))
3171                         continue;
3172
3173                 /* not for our transaction */
3174                 if (!TransactionIdInArray(f_mapped_xid, snapshot->subxip, snapshot->subxcnt))
3175                         continue;
3176
3177                 /* ok, relevant, queue for apply */
3178                 f = palloc(sizeof(RewriteMappingFile));
3179                 f->lsn = f_lsn;
3180                 strcpy(f->fname, mapping_de->d_name);
3181                 files = lappend(files, f);
3182         }
3183         FreeDir(mapping_dir);
3184
3185         /* build array we can easily sort */
3186         files_a = palloc(list_length(files) * sizeof(RewriteMappingFile *));
3187         off = 0;
3188         foreach(file, files)
3189         {
3190                 files_a[off++] = lfirst(file);
3191         }
3192
3193         /* sort files so we apply them in LSN order */
3194         qsort(files_a, list_length(files), sizeof(RewriteMappingFile *),
3195                   file_sort_by_lsn);
3196
3197         for (off = 0; off < list_length(files); off++)
3198         {
3199                 RewriteMappingFile *f = files_a[off];
3200
3201                 elog(DEBUG1, "applying mapping: \"%s\" in %u", f->fname,
3202                          snapshot->subxip[0]);
3203                 ApplyLogicalMappingFile(tuplecid_data, relid, f->fname);
3204                 pfree(f);
3205         }
3206 }
3207
3208 /*
3209  * Lookup cmin/cmax of a tuple, during logical decoding where we can't rely on
3210  * combocids.
3211  */
3212 bool
3213 ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
3214                                                           Snapshot snapshot,
3215                                                           HeapTuple htup, Buffer buffer,
3216                                                           CommandId *cmin, CommandId *cmax)
3217 {
3218         ReorderBufferTupleCidKey key;
3219         ReorderBufferTupleCidEnt *ent;
3220         ForkNumber      forkno;
3221         BlockNumber blockno;
3222         bool            updated_mapping = false;
3223
3224         /* be careful about padding */
3225         memset(&key, 0, sizeof(key));
3226
3227         Assert(!BufferIsLocal(buffer));
3228
3229         /*
3230          * get relfilenode from the buffer, no convenient way to access it other
3231          * than that.
3232          */
3233         BufferGetTag(buffer, &key.relnode, &forkno, &blockno);
3234
3235         /* tuples can only be in the main fork */
3236         Assert(forkno == MAIN_FORKNUM);
3237         Assert(blockno == ItemPointerGetBlockNumber(&htup->t_self));
3238
3239         ItemPointerCopy(&htup->t_self,
3240                                         &key.tid);
3241
3242 restart:
3243         ent = (ReorderBufferTupleCidEnt *)
3244                 hash_search(tuplecid_data,
3245                                         (void *) &key,
3246                                         HASH_FIND,
3247                                         NULL);
3248
3249         /*
3250          * failed to find a mapping, check whether the table was rewritten and
3251          * apply mapping if so, but only do that once - there can be no new
3252          * mappings while we are in here since we have to hold a lock on the
3253          * relation.
3254          */
3255         if (ent == NULL && !updated_mapping)
3256         {
3257                 UpdateLogicalMappings(tuplecid_data, htup->t_tableOid, snapshot);
3258                 /* now check but don't update for a mapping again */
3259                 updated_mapping = true;
3260                 goto restart;
3261         }
3262         else if (ent == NULL)
3263                 return false;
3264
3265         if (cmin)
3266                 *cmin = ent->cmin;
3267         if (cmax)
3268                 *cmax = ent->cmax;
3269         return true;
3270 }