Seed Dict and Set

This changes Dict and Set to have a compilation depedent seed.
The seed is not exposed to userland in anyway.
This gets a much more DOS resistant Dict and Set with no cost.

crates/compiler/builtins/bitcode/src/main.zig

@@ -188,6 +188,7 @@ comptime {
     exportUtilsFn(utils.isUnique, "is_unique");
     exportUtilsFn(utils.decrefCheckNullC, "decref_check_null");
     exportUtilsFn(utils.allocateWithRefcountC, "allocate_with_refcount");
+    exportUtilsFn(utils.dictPseudoSeed, "dict_pseudo_seed");
 
     @export(panic_utils.panic, .{ .name = "roc_builtins.utils." ++ "panic", .linkage = .Weak });
 

crates/compiler/builtins/bitcode/src/utils.zig

@@ -436,3 +436,12 @@ test "increfC, static data" {
     increfRcPtrC(ptr_to_refcount, 2);
     try std.testing.expectEqual(mock_rc, REFCOUNT_MAX_ISIZE);
 }
+
+// This returns a compilation dependent pseudo random seed for dictionaries.
+// The seed is the address of this function.
+// This avoids all roc Dicts using a known seed and being trivial to DOS.
+// Still not as secure as true random, but a lot better.
+// This value must not change between calls unless Dict is changed to store the seed on creation.
+pub fn dictPseudoSeed() callconv(.C) u64 {
+    return @intCast(u64, @ptrToInt(dictPseudoSeed));
+}

crates/compiler/builtins/roc/Dict.roc

@@ -286,7 +286,7 @@ walkUntil = \@Dict { data }, initialState, transform ->
 get : Dict k v, k -> Result v [KeyNotFound] | k has Hash & Eq
 get = \@Dict { metadata, dataIndices, data }, key ->
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -314,7 +314,7 @@ get = \@Dict { metadata, dataIndices, data }, key ->
 contains : Dict k v, k -> Bool | k has Hash & Eq
 contains = \@Dict { metadata, dataIndices, data }, key ->
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -339,7 +339,7 @@ contains = \@Dict { metadata, dataIndices, data }, key ->
 insert : Dict k v, k, v -> Dict k v | k has Hash & Eq
 insert = \@Dict { metadata, dataIndices, data, size }, key, value ->
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -386,7 +386,7 @@ remove : Dict k v, k -> Dict k v | k has Hash & Eq
 remove = \@Dict { metadata, dataIndices, data, size }, key ->
     # TODO: change this from swap remove to tombstone and test is performance is still good.
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -569,7 +569,7 @@ swapAndUpdateDataIndex : Dict k v, Nat, Nat -> Dict k v | k has Hash & Eq
 swapAndUpdateDataIndex = \@Dict { metadata, dataIndices, data, size }, removedIndex, lastIndex ->
     (T key _) = listGetUnsafe data lastIndex
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -711,7 +711,7 @@ rehashHelper = \dict, oldMetadata, oldDataIndices, oldData, index ->
 insertForRehash : Dict k v, k, Nat -> Dict k v | k has Hash & Eq
 insertForRehash = \@Dict { metadata, dataIndices, data, size }, key, dataIndex ->
     hashKey =
-        createLowLevelHasher {}
+        createLowLevelHasher PseudoRandSeed
         |> Hash.hash key
         |> complete
     h1Key = h1 hashKey
@@ -993,8 +993,16 @@ LowLevelHasher := { originalSeed : U64, state : U64 } has [
 # TODO hide behind an InternalList.roc module
 listGetUnsafe : List a, Nat -> a
 
-createLowLevelHasher : { seed ? U64 } -> LowLevelHasher
-createLowLevelHasher = \{ seed ? 0x526F_6352_616E_643F } ->
+# Returns a application specific pseudo random seed for Dict.
+# This avoids trivial DOS attacks.
+pseudoSeed : {} -> U64
+
+createLowLevelHasher : [PseudoRandSeed, WithSeed U64] -> LowLevelHasher
+createLowLevelHasher = \seedOpt ->
+    seed =
+        when seedOpt is
+            PseudoRandSeed -> pseudoSeed {}
+            WithSeed s -> s
     @LowLevelHasher { originalSeed: seed, state: seed }
 
 combineState : LowLevelHasher, { a : U64, b : U64, seed : U64, length : U64 } -> LowLevelHasher
@@ -1188,12 +1196,14 @@ wyr3 = \list, index, k ->
 
     Num.bitwiseOr a p3
 
+testSeed = WithSeed 0x526F_6352_616E_643F
+
 # TODO: would be great to have table driven expects for this.
 # Would also be great to have some sort of property based hasher
 # where we can compare `addU*` functions to the `addBytes` function.
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes []
         |> complete
 
@@ -1201,7 +1211,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0x42]
         |> complete
 
@@ -1209,7 +1219,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU8 0x42
         |> complete
 
@@ -1217,7 +1227,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0xFF, 0xFF]
         |> complete
 
@@ -1225,7 +1235,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU16 0xFFFF
         |> complete
 
@@ -1233,7 +1243,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0x36, 0xA7]
         |> complete
 
@@ -1241,7 +1251,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU16 0xA736
         |> complete
 
@@ -1249,7 +1259,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0x00, 0x00, 0x00, 0x00]
         |> complete
 
@@ -1257,7 +1267,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU32 0x0000_0000
         |> complete
 
@@ -1265,7 +1275,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0xA9, 0x2F, 0xEE, 0x21]
         |> complete
 
@@ -1273,7 +1283,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU32 0x21EE_2FA9
         |> complete
 
@@ -1281,7 +1291,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes [0x5D, 0x66, 0xB1, 0x8F, 0x68, 0x44, 0xC7, 0x03, 0xE1, 0xDD, 0x23, 0x34, 0xBB, 0x9A, 0x42, 0xA7]
         |> complete
 
@@ -1289,7 +1299,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addU128 0xA742_9ABB_3423_DDE1_03C7_4468_8FB1_665D
         |> complete
 
@@ -1297,7 +1307,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashStrBytes "abcdefghijklmnopqrstuvwxyz"
         |> complete
 
@@ -1305,7 +1315,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashStrBytes "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789"
         |> complete
 
@@ -1313,7 +1323,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashStrBytes "1234567890123456789012345678901234567890123456789012345678901234567890"
         |> complete
 
@@ -1321,7 +1331,7 @@ expect
 
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> addBytes (List.repeat 0x77 100)
         |> complete
 
@@ -1331,7 +1341,7 @@ expect
 # Apparently it won't pick the default integer.
 expect
     hash =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashUnordered [8u8, 82u8, 3u8, 8u8, 24u8] List.walk
         |> complete
 
@@ -1339,12 +1349,12 @@ expect
 
 expect
     hash1 =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashUnordered ([0u8, 1u8, 2u8, 3u8, 4u8]) List.walk
         |> complete
 
     hash2 =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashUnordered [4u8, 3u8, 2u8, 1u8, 0u8] List.walk
         |> complete
 
@@ -1352,12 +1362,12 @@ expect
 
 expect
     hash1 =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashUnordered [0u8, 1u8, 2u8, 3u8, 4u8] List.walk
         |> complete
 
     hash2 =
-        createLowLevelHasher {}
+        createLowLevelHasher testSeed
         |> Hash.hashUnordered [4u8, 3u8, 2u8, 1u8, 0u8, 0u8] List.walk
         |> complete
 

crates/compiler/builtins/src/bitcode.rs

@@ -390,6 +390,7 @@ pub const UTILS_INCREF_DATA_PTR: &str = "roc_builtins.utils.incref_data_ptr";
 pub const UTILS_DECREF_DATA_PTR: &str = "roc_builtins.utils.decref_data_ptr";
 pub const UTILS_IS_UNIQUE: &str = "roc_builtins.utils.is_unique";
 pub const UTILS_DECREF_CHECK_NULL: &str = "roc_builtins.utils.decref_check_null";
+pub const UTILS_DICT_PSEUDO_SEED: &str = "roc_builtins.utils.dict_pseudo_seed";
 
 pub const UTILS_EXPECT_FAILED_START_SHARED_BUFFER: &str =
     "roc_builtins.utils.expect_failed_start_shared_buffer";

crates/compiler/can/src/builtins.rs

@@ -217,6 +217,7 @@ map_symbol_to_lowlevel_and_arity! {
     BoxExpr; BOX_BOX_FUNCTION; 1,
     UnboxExpr; BOX_UNBOX; 1,
     Unreachable; LIST_UNREACHABLE; 1,
+    DictPseudoSeed; DICT_PSEUDO_SEED; 1,
 }
 
 /// Some builtins cannot be constructed in code gen alone, and need to be defined

crates/compiler/gen_dev/src/lib.rs

@@ -1548,6 +1548,13 @@ trait Backend<'a> {
                 arg_layouts,
                 ret_layout,
             ),
+            LowLevel::DictPseudoSeed => self.build_fn_call(
+                sym,
+                bitcode::UTILS_DICT_PSEUDO_SEED.to_string(),
+                args,
+                arg_layouts,
+                ret_layout,
+            ),
             LowLevel::NumToStr => {
                 let arg_layout = arg_layouts[0];
                 let intrinsic = match self.interner().get(arg_layout).repr {

crates/compiler/gen_llvm/src/llvm/lowlevel.rs

@@ -1310,6 +1310,11 @@ pub(crate) fn run_low_level<'a, 'ctx>(
                 ptr.into()
             }
         },
+        DictPseudoSeed => {
+            // Dict.pseudoSeed : {} -> u64
+
+            call_bitcode_fn(env, &[], bitcode::UTILS_DICT_PSEUDO_SEED)
+        }
     }
 }
 

crates/compiler/gen_wasm/src/low_level.rs

@@ -1982,6 +1982,7 @@ impl<'a> LowLevelCall<'a> {
                 },
                 StoredValue::StackMemory { .. } => { /* do nothing */ }
             },
+            DictPseudoSeed => self.load_args_and_call_zig(backend, bitcode::UTILS_DICT_PSEUDO_SEED),
         }
     }
 

crates/compiler/module/src/low_level.rs

@@ -126,6 +126,7 @@ pub enum LowLevel {
     BoxExpr,
     UnboxExpr,
     Unreachable,
+    DictPseudoSeed,
 }
 
 macro_rules! higher_order {
@@ -345,4 +346,5 @@ map_symbol_to_lowlevel! {
     Or <= BOOL_OR,
     Not <= BOOL_NOT,
     Unreachable <= LIST_UNREACHABLE,
+    DictPseudoSeed <= DICT_PSEUDO_SEED,
 }

crates/compiler/module/src/symbol.rs

@@ -1463,6 +1463,7 @@ define_builtins! {
         21 DICT_UPDATE: "update"
 
         22 DICT_LIST_GET_UNSAFE: "listGetUnsafe"
+        23 DICT_PSEUDO_SEED: "pseudoSeed"
     }
     9 SET: "Set" => {
         0 SET_SET: "Set" exposed_type=true // the Set.Set type alias

crates/compiler/mono/src/borrow.rs

@@ -947,6 +947,7 @@ pub fn lowlevel_borrow_signature(arena: &Bump, op: LowLevel) -> &[Ownership] {
     // - other refcounted arguments are Borrowed
     match op {
         Unreachable => arena.alloc_slice_copy(&[irrelevant]),
+        DictPseudoSeed => arena.alloc_slice_copy(&[irrelevant]),
         ListLen | StrIsEmpty | StrToScalars | StrCountGraphemes | StrGraphemes
         | StrCountUtf8Bytes | StrGetCapacity | ListGetCapacity => {
             arena.alloc_slice_copy(&[borrowed])