Prevent premature hsplit() calls, and only trigger REHASH after hsplit()

Triggering a hsplit due to long chain length allows an attacker
to create a carefully chosen set of keys which can cause the hash
to use 2 * (2**32) * sizeof(void *) bytes ram. AKA a DOS via memory
exhaustion. Doing so also takes non trivial time.

Eliminating this check, and only inspecting chain length after a
normal hsplit() (triggered when keys>buckets) prevents the attack
entirely, and makes such attacks relatively benign.

(cherry picked from commit f2a571dae7d70f7e3b59022834d8003ecd2df884)
(which was itself cherry picked (with changes) from commit f1220d61455253b170e81427c9d0357831ca0fac)

3 files changed, 39 insertions, 25 deletions

diff --git a/ext/Hash-Util-FieldHash/t/10_hash.t b/ext/Hash-Util-FieldHash/t/10_hash.t
index 29c2f4dafa..c266b6a45b 100644
--- a/ext/Hash-Util-FieldHash/t/10_hash.t
+++ b/ext/Hash-Util-FieldHash/t/10_hash.t
@@ -46,15 +46,29 @@ use constant START     => "a";
 
 # some initial hash data
 fieldhash my %h2;
-%h2 = map {$_ => 1} 'a'..'cc';
+my $counter= "a";
+$h2{$counter++}++ while $counter ne 'cd';
 
 ok (!Internals::HvREHASH(%h2), 
     "starting with pre-populated non-pathological hash (rehash flag if off)");
 
 my @keys = get_keys(\%h2);
+my $buckets= buckets(\%h2);
 $h2{$_}++ for @keys;
+$h2{$counter++}++ while buckets(\%h2) == $buckets; # force a split
 ok (Internals::HvREHASH(%h2), 
-    scalar(@keys) . " colliding into the same bucket keys are triggering rehash");
+    scalar(@keys) . " colliding into the same bucket keys are triggering rehash after split");
+
+# returns the number of buckets in a hash
+sub buckets {
+    my $hr = shift;
+    my $keys_buckets= scalar(%$hr);
+    if ($keys_buckets=~m!/([0-9]+)\z!) {
+        return 0+$1;
+    } else {
+        return 8;
+    }
+}
 
 sub get_keys {
     my $hr = shift;
diff --git a/hv.c b/hv.c
index 5d0f2f8d4a..6bfedd5f76 100644
--- a/hv.c
+++ b/hv.c
@@ -35,7 +35,8 @@ holds the key and hash value.
 #define PERL_HASH_INTERNAL_ACCESS
 #include "perl.h"
 
-#define HV_MAX_LENGTH_BEFORE_SPLIT 14
+#define HV_MAX_LENGTH_BEFORE_REHASH 14
+#define SHOULD_DO_HSPLIT(xhv) ((xhv)->xhv_keys > (xhv)->xhv_max) /* HvTOTALKEYS(hv) > HvMAX(hv) */
 
 static const char S_strtab_error[]
     = "Cannot modify shared string table in hv_%s";
@@ -818,23 +819,8 @@ Perl_hv_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen,
 	xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
 	if (!counter) {				/* initial entry? */
 	    xhv->xhv_fill++; /* HvFILL(hv)++ */
-	} else if (xhv->xhv_keys > (IV)xhv->xhv_max) {
+	} else if ( SHOULD_DO_HSPLIT(xhv) ) {
 	    hsplit(hv);
-	} else if(!HvREHASH(hv)) {
-	    U32 n_links = 1;
-
-	    while ((counter = HeNEXT(counter)))
-		n_links++;
-
-	    if (n_links > HV_MAX_LENGTH_BEFORE_SPLIT) {
-		/* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit
-		   bucket splits on a rehashed hash, as we're not going to
-		   split it again, and if someone is lucky (evil) enough to
-		   get all the keys in one list they could exhaust our memory
-		   as we repeatedly double the number of buckets on every
-		   entry. Linear search feels a less worse thing to do.  */
-		hsplit(hv);
-	    }
 	}
     }
 
@@ -1180,7 +1166,7 @@ S_hsplit(pTHX_ HV *hv)
 
 
     /* Pick your policy for "hashing isn't working" here:  */
-    if (longest_chain <= HV_MAX_LENGTH_BEFORE_SPLIT /* split worked?  */
+    if (longest_chain <= HV_MAX_LENGTH_BEFORE_REHASH /* split worked?  */
 	|| HvREHASH(hv)) {
 	return;
     }
@@ -2506,8 +2492,8 @@ S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags)
 	xhv->xhv_keys++; /* HvTOTALKEYS(hv)++ */
 	if (!next) {			/* initial entry? */
 	    xhv->xhv_fill++; /* HvFILL(hv)++ */
-	} else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
-		hsplit(PL_strtab);
+	} else if ( SHOULD_DO_HSPLIT(xhv) ) {
+            hsplit(PL_strtab);
 	}
     }
 
diff --git a/t/op/hash.t b/t/op/hash.t
index 9bde518732..45eb782b65 100644
--- a/t/op/hash.t
+++ b/t/op/hash.t
@@ -39,22 +39,36 @@ use constant THRESHOLD => 14;
 use constant START     => "a";
 
 # some initial hash data
-my %h2 = map {$_ => 1} 'a'..'cc';
+my %h2;
+my $counter= "a";
+$h2{$counter++}++ while $counter ne 'cd';
 
 ok (!Internals::HvREHASH(%h2), 
     "starting with pre-populated non-pathological hash (rehash flag if off)");
 
 my @keys = get_keys(\%h2);
+my $buckets= buckets(\%h2);
 $h2{$_}++ for @keys;
+$h2{$counter++}++ while buckets(\%h2) == $buckets; # force a split
 ok (Internals::HvREHASH(%h2), 
-    scalar(@keys) . " colliding into the same bucket keys are triggering rehash");
+    scalar(@keys) . " colliding into the same bucket keys are triggering rehash after split");
+
+# returns the number of buckets in a hash
+sub buckets {
+    my $hr = shift;
+    my $keys_buckets= scalar(%$hr);
+    if ($keys_buckets=~m!/([0-9]+)\z!) {
+        return 0+$1;
+    } else {
+        return 8;
+    }
+}
 
 sub get_keys {
     my $hr = shift;
 
     # the minimum of bits required to mount the attack on a hash
     my $min_bits = log(THRESHOLD)/log(2);
-
     # if the hash has already been populated with a significant amount
     # of entries the number of mask bits can be higher
     my $keys = scalar keys %$hr;