Fix hash_update_hash_key() to handle same-bucket case correctly.
authorTom Lane <[email protected]>
Tue, 15 Jan 2013 02:57:15 +0000 (21:57 -0500)
committerTom Lane <[email protected]>
Tue, 15 Jan 2013 02:57:15 +0000 (21:57 -0500)
Original coding would corrupt the hashtable if the item being updated was
at the end of its bucket chain and the new hash key hashed to that same
bucket.  Diagnosis and fix by Heikki Linnakangas.

src/backend/utils/hash/dynahash.c

index 1d473218a7761bc051354ddcfd0c09059d2e3507..5454befe15219ceb22e8902e3177d51eb0984272 100644 (file)
@@ -1022,6 +1022,7 @@ hash_update_hash_key(HTAB *hashp,
        uint32          newhashvalue;
        Size            keysize;
        uint32          bucket;
+       uint32          newbucket;
        long            segment_num;
        long            segment_ndx;
        HASHSEGMENT segp;
@@ -1078,10 +1079,10 @@ hash_update_hash_key(HTAB *hashp,
         */
        newhashvalue = hashp->hash(newKeyPtr, hashp->keysize);
 
-       bucket = calc_bucket(hctl, newhashvalue);
+       newbucket = calc_bucket(hctl, newhashvalue);
 
-       segment_num = bucket >> hashp->sshift;
-       segment_ndx = MOD(bucket, hashp->ssize);
+       segment_num = newbucket >> hashp->sshift;
+       segment_ndx = MOD(newbucket, hashp->ssize);
 
        segp = hashp->dir[segment_num];
 
@@ -1115,12 +1116,22 @@ hash_update_hash_key(HTAB *hashp,
 
        currBucket = existingElement;
 
-       /* OK to remove record from old hash bucket's chain. */
-       *oldPrevPtr = currBucket->link;
+       /*
+        * If old and new hash values belong to the same bucket, we need not
+        * change any chain links, and indeed should not since this simplistic
+        * update will corrupt the list if currBucket is the last element.  (We
+        * cannot fall out earlier, however, since we need to scan the bucket to
+        * check for duplicate keys.)
+        */
+       if (bucket != newbucket)
+       {
+               /* OK to remove record from old hash bucket's chain. */
+               *oldPrevPtr = currBucket->link;
 
-       /* link into new hashbucket chain */
-       *prevBucketPtr = currBucket;
-       currBucket->link = NULL;
+               /* link into new hashbucket chain */
+               *prevBucketPtr = currBucket;
+               currBucket->link = NULL;
+       }
 
        /* copy new key into record */
        currBucket->hashvalue = newhashvalue;