#include "postgres.h"
#include "miscadmin.h"
#include "pg_trace.h"
#include "pgstat.h"
#include "storage/lmgr.h"
#include "storage/proc.h"
#include "utils/memutils.h"

Include dependency graph for deadlock.c:

Data Structures
struct	EDGE

struct	WAIT_ORDER

struct	DEADLOCK_INFO

Functions
static bool	DeadLockCheckRecurse (PGPROC *proc)

static int	TestConfiguration (PGPROC *startProc)

static bool	FindLockCycle (PGPROC checkProc, EDGE softEdges, int *nSoftEdges)

static bool	FindLockCycleRecurse (PGPROC checkProc, int depth, EDGE softEdges, int *nSoftEdges)

static bool	FindLockCycleRecurseMember (PGPROC checkProc, PGPROC checkProcLeader, int depth, EDGE softEdges, int nSoftEdges)

static bool	ExpandConstraints (EDGE *constraints, int nConstraints)

static bool	TopoSort (LOCK lock, EDGE constraints, int nConstraints, PGPROC **ordering)

void	InitDeadLockChecking (void)

DeadLockState	DeadLockCheck (PGPROC *proc)

PGPROC *	GetBlockingAutoVacuumPgproc (void)

void	DeadLockReport (void)

void	RememberSimpleDeadLock (PGPROC proc1, LOCKMODE lockmode, LOCK lock, PGPROC *proc2)

Variables
static PGPROC **	visitedProcs

static int	nVisitedProcs

static PGPROC **	topoProcs

static int *	beforeConstraints

static int *	afterConstraints

static WAIT_ORDER *	waitOrders

static int	nWaitOrders

static PGPROC **	waitOrderProcs

static EDGE *	curConstraints

static int	nCurConstraints

static int	maxCurConstraints

static EDGE *	possibleConstraints

static int	nPossibleConstraints

static int	maxPossibleConstraints

static DEADLOCK_INFO *	deadlockDetails

static int	nDeadlockDetails

static PGPROC *	blocking_autovacuum_proc = NULL

Function Documentation

◆ DeadLockCheck()

DeadLockState DeadLockCheck ( PGPROC * proc )

Definition at line 217 of file deadlock.c.

 {
     /* Initialize to "no constraints" */
     nCurConstraints = 0;
     nPossibleConstraints = 0;
     nWaitOrders = 0;
  
     /* Initialize to not blocked by an autovacuum worker */
     blocking_autovacuum_proc = NULL;
  
     /* Search for deadlocks and possible fixes */
     if (DeadLockCheckRecurse(proc))
     {
         /*
          * Call FindLockCycle one more time, to record the correct
          * deadlockDetails[] for the basic state with no rearrangements.
          */
         int         nSoftEdges;
  
         TRACE_POSTGRESQL_DEADLOCK_FOUND();
  
         nWaitOrders = 0;
         if (!FindLockCycle(proc, possibleConstraints, &nSoftEdges))
             elog(FATAL, "deadlock seems to have disappeared");
  
         return DS_HARD_DEADLOCK;    /* cannot find a non-deadlocked state */
     }
  
     /* Apply any needed rearrangements of wait queues */
     for (int i = 0; i < nWaitOrders; i++)
     {
         LOCK       *lock = waitOrders[i].lock;
         PGPROC    **procs = waitOrders[i].procs;
         int         nProcs = waitOrders[i].nProcs;
         dclist_head *waitQueue = &lock->waitProcs;
  
         Assert(nProcs == dclist_count(waitQueue));
  
 #ifdef DEBUG_DEADLOCK
         PrintLockQueue(lock, "DeadLockCheck:");
 #endif
  
         /* Reset the queue and re-add procs in the desired order */
         dclist_init(waitQueue);
         for (int j = 0; j < nProcs; j++)
             dclist_push_tail(waitQueue, &procs[j]->links);
  
 #ifdef DEBUG_DEADLOCK
         PrintLockQueue(lock, "rearranged to:");
 #endif
  
         /* See if any waiters for the lock can be woken up now */
         ProcLockWakeup(GetLocksMethodTable(lock), lock);
     }
  
     /* Return code tells caller if we had to escape a deadlock or not */
     if (nWaitOrders > 0)
         return DS_SOFT_DEADLOCK;
     else if (blocking_autovacuum_proc != NULL)
         return DS_BLOCKED_BY_AUTOVACUUM;
     else
         return DS_NO_DEADLOCK;
 }

References Assert(), blocking_autovacuum_proc, dclist_count(), dclist_init(), dclist_push_tail(), DeadLockCheckRecurse(), DS_BLOCKED_BY_AUTOVACUUM, DS_HARD_DEADLOCK, DS_NO_DEADLOCK, DS_SOFT_DEADLOCK, elog(), FATAL, FindLockCycle(), GetLocksMethodTable(), i, j, links, WAIT_ORDER::lock, nCurConstraints, nPossibleConstraints, WAIT_ORDER::nProcs, nWaitOrders, possibleConstraints, ProcLockWakeup(), WAIT_ORDER::procs, waitOrders, and LOCK::waitProcs.

Referenced by CheckDeadLock().

◆ DeadLockCheckRecurse()

static bool DeadLockCheckRecurse ( PGPROC * proc )

static

Definition at line 309 of file deadlock.c.

 {
     int         nEdges;
     int         oldPossibleConstraints;
     bool        savedList;
     int         i;
  
     nEdges = TestConfiguration(proc);
     if (nEdges < 0)
         return true;            /* hard deadlock --- no solution */
     if (nEdges == 0)
         return false;           /* good configuration found */
     if (nCurConstraints >= maxCurConstraints)
         return true;            /* out of room for active constraints? */
     oldPossibleConstraints = nPossibleConstraints;
     if (nPossibleConstraints + nEdges + MaxBackends <= maxPossibleConstraints)
     {
         /* We can save the edge list in possibleConstraints[] */
         nPossibleConstraints += nEdges;
         savedList = true;
     }
     else
     {
         /* Not room; will need to regenerate the edges on-the-fly */
         savedList = false;
     }
  
     /*
      * Try each available soft edge as an addition to the configuration.
      */
     for (i = 0; i < nEdges; i++)
     {
         if (!savedList && i > 0)
         {
             /* Regenerate the list of possible added constraints */
             if (nEdges != TestConfiguration(proc))
                 elog(FATAL, "inconsistent results during deadlock check");
         }
         curConstraints[nCurConstraints] =
             possibleConstraints[oldPossibleConstraints + i];
         nCurConstraints++;
         if (!DeadLockCheckRecurse(proc))
             return false;       /* found a valid solution! */
         /* give up on that added constraint, try again */
         nCurConstraints--;
     }
     nPossibleConstraints = oldPossibleConstraints;
     return true;                /* no solution found */
 }

References curConstraints, elog(), FATAL, i, MaxBackends, maxCurConstraints, maxPossibleConstraints, nCurConstraints, nPossibleConstraints, possibleConstraints, and TestConfiguration().

Referenced by DeadLockCheck().

◆ DeadLockReport()

void DeadLockReport ( void )

Definition at line 1073 of file deadlock.c.

 {
     StringInfoData clientbuf;   /* errdetail for client */
     StringInfoData logbuf;      /* errdetail for server log */
     StringInfoData locktagbuf;
     int         i;
  
     initStringInfo(&clientbuf);
     initStringInfo(&logbuf);
     initStringInfo(&locktagbuf);
  
     /* Generate the "waits for" lines sent to the client */
     for (i = 0; i < nDeadlockDetails; i++)
     {
         DEADLOCK_INFO *info = &deadlockDetails[i];
         int         nextpid;
  
         /* The last proc waits for the first one... */
         if (i < nDeadlockDetails - 1)
             nextpid = info[1].pid;
         else
             nextpid = deadlockDetails[0].pid;
  
         /* reset locktagbuf to hold next object description */
         resetStringInfo(&locktagbuf);
  
         DescribeLockTag(&locktagbuf, &info->locktag);
  
         if (i > 0)
             appendStringInfoChar(&clientbuf, '\n');
  
         appendStringInfo(&clientbuf,
                          _("Process %d waits for %s on %s; blocked by process %d."),
                          info->pid,
                          GetLockmodeName(info->locktag.locktag_lockmethodid,
                                          info->lockmode),
                          locktagbuf.data,
                          nextpid);
     }
  
     /* Duplicate all the above for the server ... */
     appendBinaryStringInfo(&logbuf, clientbuf.data, clientbuf.len);
  
     /* ... and add info about query strings */
     for (i = 0; i < nDeadlockDetails; i++)
     {
         DEADLOCK_INFO *info = &deadlockDetails[i];
  
         appendStringInfoChar(&logbuf, '\n');
  
         appendStringInfo(&logbuf,
                          _("Process %d: %s"),
                          info->pid,
                          pgstat_get_backend_current_activity(info->pid, false));
     }
  
     pgstat_report_deadlock();
  
     ereport(ERROR,
             (errcode(ERRCODE_T_R_DEADLOCK_DETECTED),
              errmsg("deadlock detected"),
              errdetail_internal("%s", clientbuf.data),
              errdetail_log("%s", logbuf.data),
              errhint("See server log for query details.")));
 }

References _, appendBinaryStringInfo(), appendStringInfo(), appendStringInfoChar(), StringInfoData::data, deadlockDetails, DescribeLockTag(), ereport, errcode(), ERRCODE_T_R_DEADLOCK_DETECTED, errdetail_internal(), errdetail_log(), errhint(), errmsg(), ERROR, GetLockmodeName(), i, initStringInfo(), StringInfoData::len, DEADLOCK_INFO::lockmode, DEADLOCK_INFO::locktag, LOCKTAG::locktag_lockmethodid, nDeadlockDetails, pgstat_get_backend_current_activity(), pgstat_report_deadlock(), DEADLOCK_INFO::pid, and resetStringInfo().

Referenced by WaitOnLock().

◆ ExpandConstraints()

static bool ExpandConstraints	(	EDGE *	constraints,
		int	nConstraints
	)

static

Definition at line 788 of file deadlock.c.

 {
     int         nWaitOrderProcs = 0;
     int         i,
                 j;
  
     nWaitOrders = 0;
  
     /*
      * Scan constraint list backwards.  This is because the last-added
      * constraint is the only one that could fail, and so we want to test it
      * for inconsistency first.
      */
     for (i = nConstraints; --i >= 0;)
     {
         LOCK       *lock = constraints[i].lock;
  
         /* Did we already make a list for this lock? */
         for (j = nWaitOrders; --j >= 0;)
         {
             if (waitOrders[j].lock == lock)
                 break;
         }
         if (j >= 0)
             continue;
         /* No, so allocate a new list */
         waitOrders[nWaitOrders].lock = lock;
         waitOrders[nWaitOrders].procs = waitOrderProcs + nWaitOrderProcs;
         waitOrders[nWaitOrders].nProcs = dclist_count(&lock->waitProcs);
         nWaitOrderProcs += dclist_count(&lock->waitProcs);
         Assert(nWaitOrderProcs <= MaxBackends);
  
         /*
          * Do the topo sort.  TopoSort need not examine constraints after this
          * one, since they must be for different locks.
          */
         if (!TopoSort(lock, constraints, i + 1,
                       waitOrders[nWaitOrders].procs))
             return false;
         nWaitOrders++;
     }
     return true;
 }

References Assert(), dclist_count(), i, j, EDGE::lock, WAIT_ORDER::lock, MaxBackends, WAIT_ORDER::nProcs, nWaitOrders, WAIT_ORDER::procs, TopoSort(), waitOrderProcs, waitOrders, and LOCK::waitProcs.

Referenced by TestConfiguration().

◆ FindLockCycle()

static bool FindLockCycle	(	PGPROC *	checkProc,
		EDGE *	softEdges,
		int *	nSoftEdges
	)

static

Definition at line 443 of file deadlock.c.

 {
     nVisitedProcs = 0;
     nDeadlockDetails = 0;
     *nSoftEdges = 0;
     return FindLockCycleRecurse(checkProc, 0, softEdges, nSoftEdges);
 }

References FindLockCycleRecurse(), nDeadlockDetails, and nVisitedProcs.

Referenced by DeadLockCheck(), and TestConfiguration().

◆ FindLockCycleRecurse()

static bool FindLockCycleRecurse	(	PGPROC *	checkProc,
		int	depth,
		EDGE *	softEdges,
		int *	nSoftEdges
	)

static

Definition at line 454 of file deadlock.c.

 {
     int         i;
     dlist_iter  iter;
  
     /*
      * If this process is a lock group member, check the leader instead. (Note
      * that we might be the leader, in which case this is a no-op.)
      */
     if (checkProc->lockGroupLeader != NULL)
         checkProc = checkProc->lockGroupLeader;
  
     /*
      * Have we already seen this proc?
      */
     for (i = 0; i < nVisitedProcs; i++)
     {
         if (visitedProcs[i] == checkProc)
         {
             /* If we return to starting point, we have a deadlock cycle */
             if (i == 0)
             {
                 /*
                  * record total length of cycle --- outer levels will now fill
                  * deadlockDetails[]
                  */
                 Assert(depth <= MaxBackends);
                 nDeadlockDetails = depth;
  
                 return true;
             }
  
             /*
              * Otherwise, we have a cycle but it does not include the start
              * point, so say "no deadlock".
              */
             return false;
         }
     }
     /* Mark proc as seen */
     Assert(nVisitedProcs < MaxBackends);
     visitedProcs[nVisitedProcs++] = checkProc;
  
     /*
      * If the process is waiting, there is an outgoing waits-for edge to each
      * process that blocks it.
      */
     if (checkProc->links.next != NULL && checkProc->waitLock != NULL &&
         FindLockCycleRecurseMember(checkProc, checkProc, depth, softEdges,
                                    nSoftEdges))
         return true;
  
     /*
      * If the process is not waiting, there could still be outgoing waits-for
      * edges if it is part of a lock group, because other members of the lock
      * group might be waiting even though this process is not.  (Given lock
      * groups {A1, A2} and {B1, B2}, if A1 waits for B1 and B2 waits for A2,
      * that is a deadlock even neither of B1 and A2 are waiting for anything.)
      */
     dlist_foreach(iter, &checkProc->lockGroupMembers)
     {
         PGPROC     *memberProc;
  
         memberProc = dlist_container(PGPROC, lockGroupLink, iter.cur);
  
         if (memberProc->links.next != NULL && memberProc->waitLock != NULL &&
             memberProc != checkProc &&
             FindLockCycleRecurseMember(memberProc, checkProc, depth, softEdges,
                                        nSoftEdges))
             return true;
     }
  
     return false;
 }

References Assert(), dlist_iter::cur, dlist_container, dlist_foreach, FindLockCycleRecurseMember(), i, PGPROC::links, PGPROC::lockGroupLeader, PGPROC::lockGroupMembers, MaxBackends, nDeadlockDetails, dlist_node::next, nVisitedProcs, visitedProcs, and PGPROC::waitLock.

Referenced by FindLockCycle(), and FindLockCycleRecurseMember().

◆ FindLockCycleRecurseMember()

static bool FindLockCycleRecurseMember	(	PGPROC *	checkProc,
		PGPROC *	checkProcLeader,
		int	depth,
		EDGE *	softEdges,
		int *	nSoftEdges
	)

static

Definition at line 533 of file deadlock.c.

 {
     PGPROC     *proc;
     LOCK       *lock = checkProc->waitLock;
     dlist_iter  proclock_iter;
     LockMethod  lockMethodTable;
     int         conflictMask;
     int         i;
     int         numLockModes,
                 lm;
  
     /*
      * The relation extension or page lock can never participate in actual
      * deadlock cycle.  See Asserts in LockAcquireExtended.  So, there is no
      * advantage in checking wait edges from them.
      */
     if (LOCK_LOCKTAG(*lock) == LOCKTAG_RELATION_EXTEND ||
         (LOCK_LOCKTAG(*lock) == LOCKTAG_PAGE))
         return false;
  
     lockMethodTable = GetLocksMethodTable(lock);
     numLockModes = lockMethodTable->numLockModes;
     conflictMask = lockMethodTable->conflictTab[checkProc->waitLockMode];
  
     /*
      * Scan for procs that already hold conflicting locks.  These are "hard"
      * edges in the waits-for graph.
      */
     dlist_foreach(proclock_iter, &lock->procLocks)
     {
         PROCLOCK   *proclock = dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
         PGPROC     *leader;
  
         proc = proclock->tag.myProc;
         leader = proc->lockGroupLeader == NULL ? proc : proc->lockGroupLeader;
  
         /* A proc never blocks itself or any other lock group member */
         if (leader != checkProcLeader)
         {
             for (lm = 1; lm <= numLockModes; lm++)
             {
                 if ((proclock->holdMask & LOCKBIT_ON(lm)) &&
                     (conflictMask & LOCKBIT_ON(lm)))
                 {
                     /* This proc hard-blocks checkProc */
                     if (FindLockCycleRecurse(proc, depth + 1,
                                              softEdges, nSoftEdges))
                     {
                         /* fill deadlockDetails[] */
                         DEADLOCK_INFO *info = &deadlockDetails[depth];
  
                         info->locktag = lock->tag;
                         info->lockmode = checkProc->waitLockMode;
                         info->pid = checkProc->pid;
  
                         return true;
                     }
  
                     /*
                      * No deadlock here, but see if this proc is an autovacuum
                      * that is directly hard-blocking our own proc.  If so,
                      * report it so that the caller can send a cancel signal
                      * to it, if appropriate.  If there's more than one such
                      * proc, it's indeterminate which one will be reported.
                      *
                      * We don't touch autovacuums that are indirectly blocking
                      * us; it's up to the direct blockee to take action.  This
                      * rule simplifies understanding the behavior and ensures
                      * that an autovacuum won't be canceled with less than
                      * deadlock_timeout grace period.
                      *
                      * Note we read statusFlags without any locking.  This is
                      * OK only for checking the PROC_IS_AUTOVACUUM flag,
                      * because that flag is set at process start and never
                      * reset.  There is logic elsewhere to avoid canceling an
                      * autovacuum that is working to prevent XID wraparound
                      * problems (which needs to read a different statusFlags
                      * bit), but we don't do that here to avoid grabbing
                      * ProcArrayLock.
                      */
                     if (checkProc == MyProc &&
                         proc->statusFlags & PROC_IS_AUTOVACUUM)
                         blocking_autovacuum_proc = proc;
  
                     /* We're done looking at this proclock */
                     break;
                 }
             }
         }
     }
  
     /*
      * Scan for procs that are ahead of this one in the lock's wait queue.
      * Those that have conflicting requests soft-block this one.  This must be
      * done after the hard-block search, since if another proc both hard- and
      * soft-blocks this one, we want to call it a hard edge.
      *
      * If there is a proposed re-ordering of the lock's wait order, use that
      * rather than the current wait order.
      */
     for (i = 0; i < nWaitOrders; i++)
     {
         if (waitOrders[i].lock == lock)
             break;
     }
  
     if (i < nWaitOrders)
     {
         /* Use the given hypothetical wait queue order */
         PGPROC    **procs = waitOrders[i].procs;
         int         queue_size = waitOrders[i].nProcs;
  
         for (i = 0; i < queue_size; i++)
         {
             PGPROC     *leader;
  
             proc = procs[i];
             leader = proc->lockGroupLeader == NULL ? proc :
                 proc->lockGroupLeader;
  
             /*
              * TopoSort will always return an ordering with group members
              * adjacent to each other in the wait queue (see comments
              * therein). So, as soon as we reach a process in the same lock
              * group as checkProc, we know we've found all the conflicts that
              * precede any member of the lock group lead by checkProcLeader.
              */
             if (leader == checkProcLeader)
                 break;
  
             /* Is there a conflict with this guy's request? */
             if ((LOCKBIT_ON(proc->waitLockMode) & conflictMask) != 0)
             {
                 /* This proc soft-blocks checkProc */
                 if (FindLockCycleRecurse(proc, depth + 1,
                                          softEdges, nSoftEdges))
                 {
                     /* fill deadlockDetails[] */
                     DEADLOCK_INFO *info = &deadlockDetails[depth];
  
                     info->locktag = lock->tag;
                     info->lockmode = checkProc->waitLockMode;
                     info->pid = checkProc->pid;
  
                     /*
                      * Add this edge to the list of soft edges in the cycle
                      */
                     Assert(*nSoftEdges < MaxBackends);
                     softEdges[*nSoftEdges].waiter = checkProcLeader;
                     softEdges[*nSoftEdges].blocker = leader;
                     softEdges[*nSoftEdges].lock = lock;
                     (*nSoftEdges)++;
                     return true;
                 }
             }
         }
     }
     else
     {
         PGPROC     *lastGroupMember = NULL;
         dlist_iter  proc_iter;
         dclist_head *waitQueue;
  
         /* Use the true lock wait queue order */
         waitQueue = &lock->waitProcs;
  
         /*
          * Find the last member of the lock group that is present in the wait
          * queue.  Anything after this is not a soft lock conflict. If group
          * locking is not in use, then we know immediately which process we're
          * looking for, but otherwise we've got to search the wait queue to
          * find the last process actually present.
          */
         if (checkProc->lockGroupLeader == NULL)
             lastGroupMember = checkProc;
         else
         {
             dclist_foreach(proc_iter, waitQueue)
             {
                 proc = dlist_container(PGPROC, links, proc_iter.cur);
  
                 if (proc->lockGroupLeader == checkProcLeader)
                     lastGroupMember = proc;
             }
             Assert(lastGroupMember != NULL);
         }
  
         /*
          * OK, now rescan (or scan) the queue to identify the soft conflicts.
          */
         dclist_foreach(proc_iter, waitQueue)
         {
             PGPROC     *leader;
  
             proc = dlist_container(PGPROC, links, proc_iter.cur);
  
             leader = proc->lockGroupLeader == NULL ? proc :
                 proc->lockGroupLeader;
  
             /* Done when we reach the target proc */
             if (proc == lastGroupMember)
                 break;
  
             /* Is there a conflict with this guy's request? */
             if ((LOCKBIT_ON(proc->waitLockMode) & conflictMask) != 0 &&
                 leader != checkProcLeader)
             {
                 /* This proc soft-blocks checkProc */
                 if (FindLockCycleRecurse(proc, depth + 1,
                                          softEdges, nSoftEdges))
                 {
                     /* fill deadlockDetails[] */
                     DEADLOCK_INFO *info = &deadlockDetails[depth];
  
                     info->locktag = lock->tag;
                     info->lockmode = checkProc->waitLockMode;
                     info->pid = checkProc->pid;
  
                     /*
                      * Add this edge to the list of soft edges in the cycle
                      */
                     Assert(*nSoftEdges < MaxBackends);
                     softEdges[*nSoftEdges].waiter = checkProcLeader;
                     softEdges[*nSoftEdges].blocker = leader;
                     softEdges[*nSoftEdges].lock = lock;
                     (*nSoftEdges)++;
                     return true;
                 }
             }
         }
     }
  
     /*
      * No conflict detected here.
      */
     return false;
 }

References Assert(), EDGE::blocker, blocking_autovacuum_proc, LockMethodData::conflictTab, dlist_iter::cur, dclist_foreach, deadlockDetails, dlist_container, dlist_foreach, FindLockCycleRecurse(), GetLocksMethodTable(), PROCLOCK::holdMask, i, links, EDGE::lock, LOCK_LOCKTAG, LOCKBIT_ON, PGPROC::lockGroupLeader, DEADLOCK_INFO::lockmode, DEADLOCK_INFO::locktag, LOCKTAG_PAGE, LOCKTAG_RELATION_EXTEND, MaxBackends, MyProc, PROCLOCKTAG::myProc, WAIT_ORDER::nProcs, LockMethodData::numLockModes, nWaitOrders, DEADLOCK_INFO::pid, PGPROC::pid, PROC_IS_AUTOVACUUM, LOCK::procLocks, WAIT_ORDER::procs, PGPROC::statusFlags, LOCK::tag, PROCLOCK::tag, EDGE::waiter, PGPROC::waitLock, PGPROC::waitLockMode, waitOrders, and LOCK::waitProcs.

Referenced by FindLockCycleRecurse().

◆ GetBlockingAutoVacuumPgproc()

PGPROC* GetBlockingAutoVacuumPgproc ( void )

Definition at line 287 of file deadlock.c.

 {
     PGPROC     *ptr;
  
     ptr = blocking_autovacuum_proc;
     blocking_autovacuum_proc = NULL;
  
     return ptr;
 }

References blocking_autovacuum_proc.

Referenced by ProcSleep().

◆ InitDeadLockChecking()

void InitDeadLockChecking ( void )

Definition at line 143 of file deadlock.c.

 {
     MemoryContext oldcxt;
  
     /* Make sure allocations are permanent */
     oldcxt = MemoryContextSwitchTo(TopMemoryContext);
  
     /*
      * FindLockCycle needs at most MaxBackends entries in visitedProcs[] and
      * deadlockDetails[].
      */
     visitedProcs = (PGPROC **) palloc(MaxBackends * sizeof(PGPROC *));
     deadlockDetails = (DEADLOCK_INFO *) palloc(MaxBackends * sizeof(DEADLOCK_INFO));
  
     /*
      * TopoSort needs to consider at most MaxBackends wait-queue entries, and
      * it needn't run concurrently with FindLockCycle.
      */
     topoProcs = visitedProcs;   /* re-use this space */
     beforeConstraints = (int *) palloc(MaxBackends * sizeof(int));
     afterConstraints = (int *) palloc(MaxBackends * sizeof(int));
  
     /*
      * We need to consider rearranging at most MaxBackends/2 wait queues
      * (since it takes at least two waiters in a queue to create a soft edge),
      * and the expanded form of the wait queues can't involve more than
      * MaxBackends total waiters.
      */
     waitOrders = (WAIT_ORDER *)
         palloc((MaxBackends / 2) * sizeof(WAIT_ORDER));
     waitOrderProcs = (PGPROC **) palloc(MaxBackends * sizeof(PGPROC *));
  
     /*
      * Allow at most MaxBackends distinct constraints in a configuration. (Is
      * this enough?  In practice it seems it should be, but I don't quite see
      * how to prove it.  If we run out, we might fail to find a workable wait
      * queue rearrangement even though one exists.)  NOTE that this number
      * limits the maximum recursion depth of DeadLockCheckRecurse. Making it
      * really big might potentially allow a stack-overflow problem.
      */
     maxCurConstraints = MaxBackends;
     curConstraints = (EDGE *) palloc(maxCurConstraints * sizeof(EDGE));
  
     /*
      * Allow up to 3*MaxBackends constraints to be saved without having to
      * re-run TestConfiguration.  (This is probably more than enough, but we
      * can survive if we run low on space by doing excess runs of
      * TestConfiguration to re-compute constraint lists each time needed.) The
      * last MaxBackends entries in possibleConstraints[] are reserved as
      * output workspace for FindLockCycle.
      */
     maxPossibleConstraints = MaxBackends * 4;
     possibleConstraints =
         (EDGE *) palloc(maxPossibleConstraints * sizeof(EDGE));
  
     MemoryContextSwitchTo(oldcxt);
 }

References afterConstraints, beforeConstraints, curConstraints, deadlockDetails, MaxBackends, maxCurConstraints, maxPossibleConstraints, MemoryContextSwitchTo(), palloc(), possibleConstraints, TopMemoryContext, topoProcs, visitedProcs, waitOrderProcs, and waitOrders.

Referenced by InitProcess().

◆ RememberSimpleDeadLock()

void RememberSimpleDeadLock	(	PGPROC *	proc1,
		LOCKMODE	lockmode,
		LOCK *	lock,
		PGPROC *	proc2
	)

Definition at line 1145 of file deadlock.c.

 {
     DEADLOCK_INFO *info = &deadlockDetails[0];
  
     info->locktag = lock->tag;
     info->lockmode = lockmode;
     info->pid = proc1->pid;
     info++;
     info->locktag = proc2->waitLock->tag;
     info->lockmode = proc2->waitLockMode;
     info->pid = proc2->pid;
     nDeadlockDetails = 2;
 }

References deadlockDetails, DEADLOCK_INFO::lockmode, DEADLOCK_INFO::locktag, nDeadlockDetails, DEADLOCK_INFO::pid, PGPROC::pid, LOCK::tag, PGPROC::waitLock, and PGPROC::waitLockMode.

Referenced by ProcSleep().

◆ TestConfiguration()

static int TestConfiguration ( PGPROC * startProc )

static

Definition at line 375 of file deadlock.c.

 {
     int         softFound = 0;
     EDGE       *softEdges = possibleConstraints + nPossibleConstraints;
     int         nSoftEdges;
     int         i;
  
     /*
      * Make sure we have room for FindLockCycle's output.
      */
     if (nPossibleConstraints + MaxBackends > maxPossibleConstraints)
         return -1;
  
     /*
      * Expand current constraint set into wait orderings.  Fail if the
      * constraint set is not self-consistent.
      */
     if (!ExpandConstraints(curConstraints, nCurConstraints))
         return -1;
  
     /*
      * Check for cycles involving startProc or any of the procs mentioned in
      * constraints.  We check startProc last because if it has a soft cycle
      * still to be dealt with, we want to deal with that first.
      */
     for (i = 0; i < nCurConstraints; i++)
     {
         if (FindLockCycle(curConstraints[i].waiter, softEdges, &nSoftEdges))
         {
             if (nSoftEdges == 0)
                 return -1;      /* hard deadlock detected */
             softFound = nSoftEdges;
         }
         if (FindLockCycle(curConstraints[i].blocker, softEdges, &nSoftEdges))
         {
             if (nSoftEdges == 0)
                 return -1;      /* hard deadlock detected */
             softFound = nSoftEdges;
         }
     }
     if (FindLockCycle(startProc, softEdges, &nSoftEdges))
     {
         if (nSoftEdges == 0)
             return -1;          /* hard deadlock detected */
         softFound = nSoftEdges;
     }
     return softFound;
 }

References curConstraints, ExpandConstraints(), FindLockCycle(), i, MaxBackends, maxPossibleConstraints, nCurConstraints, nPossibleConstraints, and possibleConstraints.

Referenced by DeadLockCheckRecurse().

◆ TopoSort()

static bool TopoSort	(	LOCK *	lock,
		EDGE *	constraints,
		int	nConstraints,
		PGPROC **	ordering
	)

static

Definition at line 860 of file deadlock.c.

 {
     dclist_head *waitQueue = &lock->waitProcs;
     int         queue_size = dclist_count(waitQueue);
     PGPROC     *proc;
     int         i,
                 j,
                 jj,
                 k,
                 kk,
                 last;
     dlist_iter  proc_iter;
  
     /* First, fill topoProcs[] array with the procs in their current order */
     i = 0;
     dclist_foreach(proc_iter, waitQueue)
     {
         proc = dlist_container(PGPROC, links, proc_iter.cur);
         topoProcs[i++] = proc;
     }
     Assert(i == queue_size);
  
     /*
      * Scan the constraints, and for each proc in the array, generate a count
      * of the number of constraints that say it must be before something else,
      * plus a list of the constraints that say it must be after something
      * else. The count for the j'th proc is stored in beforeConstraints[j],
      * and the head of its list in afterConstraints[j].  Each constraint
      * stores its list link in constraints[i].link (note any constraint will
      * be in just one list). The array index for the before-proc of the i'th
      * constraint is remembered in constraints[i].pred.
      *
      * Note that it's not necessarily the case that every constraint affects
      * this particular wait queue.  Prior to group locking, a process could be
      * waiting for at most one lock.  But a lock group can be waiting for
      * zero, one, or multiple locks.  Since topoProcs[] is an array of the
      * processes actually waiting, while constraints[] is an array of group
      * leaders, we've got to scan through topoProcs[] for each constraint,
      * checking whether both a waiter and a blocker for that group are
      * present.  If so, the constraint is relevant to this wait queue; if not,
      * it isn't.
      */
     MemSet(beforeConstraints, 0, queue_size * sizeof(int));
     MemSet(afterConstraints, 0, queue_size * sizeof(int));
     for (i = 0; i < nConstraints; i++)
     {
         /*
          * Find a representative process that is on the lock queue and part of
          * the waiting lock group.  This may or may not be the leader, which
          * may or may not be waiting at all.  If there are any other processes
          * in the same lock group on the queue, set their number of
          * beforeConstraints to -1 to indicate that they should be emitted
          * with their groupmates rather than considered separately.
          *
          * In this loop and the similar one just below, it's critical that we
          * consistently select the same representative member of any one lock
          * group, so that all the constraints are associated with the same
          * proc, and the -1's are only associated with not-representative
          * members.  We select the last one in the topoProcs array.
          */
         proc = constraints[i].waiter;
         Assert(proc != NULL);
         jj = -1;
         for (j = queue_size; --j >= 0;)
         {
             PGPROC     *waiter = topoProcs[j];
  
             if (waiter == proc || waiter->lockGroupLeader == proc)
             {
                 Assert(waiter->waitLock == lock);
                 if (jj == -1)
                     jj = j;
                 else
                 {
                     Assert(beforeConstraints[j] <= 0);
                     beforeConstraints[j] = -1;
                 }
             }
         }
  
         /* If no matching waiter, constraint is not relevant to this lock. */
         if (jj < 0)
             continue;
  
         /*
          * Similarly, find a representative process that is on the lock queue
          * and waiting for the blocking lock group.  Again, this could be the
          * leader but does not need to be.
          */
         proc = constraints[i].blocker;
         Assert(proc != NULL);
         kk = -1;
         for (k = queue_size; --k >= 0;)
         {
             PGPROC     *blocker = topoProcs[k];
  
             if (blocker == proc || blocker->lockGroupLeader == proc)
             {
                 Assert(blocker->waitLock == lock);
                 if (kk == -1)
                     kk = k;
                 else
                 {
                     Assert(beforeConstraints[k] <= 0);
                     beforeConstraints[k] = -1;
                 }
             }
         }
  
         /* If no matching blocker, constraint is not relevant to this lock. */
         if (kk < 0)
             continue;
  
         Assert(beforeConstraints[jj] >= 0);
         beforeConstraints[jj]++;    /* waiter must come before */
         /* add this constraint to list of after-constraints for blocker */
         constraints[i].pred = jj;
         constraints[i].link = afterConstraints[kk];
         afterConstraints[kk] = i + 1;
     }
  
     /*--------------------
      * Now scan the topoProcs array backwards.  At each step, output the
      * last proc that has no remaining before-constraints plus any other
      * members of the same lock group; then decrease the beforeConstraints
      * count of each of the procs it was constrained against.
      * i = index of ordering[] entry we want to output this time
      * j = search index for topoProcs[]
      * k = temp for scanning constraint list for proc j
      * last = last non-null index in topoProcs (avoid redundant searches)
      *--------------------
      */
     last = queue_size - 1;
     for (i = queue_size - 1; i >= 0;)
     {
         int         c;
         int         nmatches = 0;
  
         /* Find next candidate to output */
         while (topoProcs[last] == NULL)
             last--;
         for (j = last; j >= 0; j--)
         {
             if (topoProcs[j] != NULL && beforeConstraints[j] == 0)
                 break;
         }
  
         /* If no available candidate, topological sort fails */
         if (j < 0)
             return false;
  
         /*
          * Output everything in the lock group.  There's no point in
          * outputting an ordering where members of the same lock group are not
          * consecutive on the wait queue: if some other waiter is between two
          * requests that belong to the same group, then either it conflicts
          * with both of them and is certainly not a solution; or it conflicts
          * with at most one of them and is thus isomorphic to an ordering
          * where the group members are consecutive.
          */
         proc = topoProcs[j];
         if (proc->lockGroupLeader != NULL)
             proc = proc->lockGroupLeader;
         Assert(proc != NULL);
         for (c = 0; c <= last; ++c)
         {
             if (topoProcs[c] == proc || (topoProcs[c] != NULL &&
                                          topoProcs[c]->lockGroupLeader == proc))
             {
                 ordering[i - nmatches] = topoProcs[c];
                 topoProcs[c] = NULL;
                 ++nmatches;
             }
         }
         Assert(nmatches > 0);
         i -= nmatches;
  
         /* Update beforeConstraints counts of its predecessors */
         for (k = afterConstraints[j]; k > 0; k = constraints[k - 1].link)
             beforeConstraints[constraints[k - 1].pred]--;
     }
  
     /* Done */
     return true;
 }