Add a new mode 2 to the LDC that allows to use the memory as a source for code

CHANGELOG.md

@@ -8,6 +8,7 @@ and this project adheres to [Semantic Versioning](http://semver.org/).
 ## [Unreleased]
 
 ### Added
+- [#849](https://github.com/FuelLabs/fuel-vm/pull/849): Add a new mode `2` to the LDC that allows to use the memory as a source for code.
 - [#848](https://github.com/FuelLabs/fuel-vm/pull/848): Allow usage of the blob opcode `BSIZ`, `BLDD`, and `LDC` with mode `1` in the predicates.
 - [#838](https://github.com/FuelLabs/fuel-vm/pull/838): Implemented `AsRef<[u8]>` and `TryFrom<&[u8]>` for DA compression types: ScriptCode, PredicateCode, RegistryKey.
 - [#820](https://github.com/FuelLabs/fuel-vm/pull/820): Add fuzzing in CI with ClusterFuzzLite.

fuel-asm/src/lib.rs

@@ -182,8 +182,8 @@ impl_instructions! {
     0x30 CSIZ csiz [dst: RegId contract_id_addr: RegId]
     "Get current block proposer's address."
     0x31 CB cb [dst: RegId]
-    "Load a contract's code as executable."
-    0x32 LDC ldc [contract_id_addr: RegId offset: RegId len: RegId mode: Imm06]
+    "Load code as executable either from contract, blob, or memory."
+    0x32 LDC ldc [src_addr: RegId offset: RegId len: RegId mode: Imm06]
     "Log an event."
     0x33 LOG log [a: RegId b: RegId c: RegId d: RegId]
     "Log data."

fuel-vm/src/interpreter/blockchain.rs

@@ -106,7 +106,7 @@ where
     /// ```
     pub(crate) fn load_contract_code(
         &mut self,
-        id_addr: Word,
+        addr: Word,
         offset: Word,
         length_unpadded: Word,
         mode: Imm06,
@@ -152,8 +152,9 @@ where
         };
 
         match mode.to_u8() {
-            0 => input.load_contract_code(id_addr, offset, length_unpadded),
-            1 => input.load_blob_code(id_addr, offset, length_unpadded),
+            0 => input.load_contract_code(addr, offset, length_unpadded),
+            1 => input.load_blob_code(addr, offset, length_unpadded),
+            2 => input.load_memory_code(addr, offset, length_unpadded),
             _ => Err(PanicReason::InvalidImmediateValue.into()),
         }
     }
@@ -585,7 +586,6 @@ where
     /// contract_code = contracts[contract_id]
     /// mem[$ssp, $rC] = contract_code[$rB, $rC]
     /// ```
-    /// Returns the total length of the contract code that was loaded from storage.
     pub(crate) fn load_contract_code(
         mut self,
         contract_id_addr: Word,
@@ -684,11 +684,10 @@ where
     /// Loads blob ID pointed by `a`, and then for that blob,
     /// copies `c` bytes from it starting from offset `b` into the stack.
     /// ```txt
-    /// contract_id = mem[$rA, 32]
-    /// contract_code = contracts[contract_id]
-    /// mem[$ssp, $rC] = contract_code[$rB, $rC]
+    /// blob_id = mem[$rA, 32]
+    /// blob_code = blobs[blob_id]
+    /// mem[$ssp, $rC] = blob_code[$rB, $rC]
     /// ```
-    /// Returns the total length of the contract code that was loaded from storage.
     pub(crate) fn load_blob_code(
         mut self,
         blob_id_addr: Word,
@@ -778,6 +777,89 @@ where
 
         Ok(())
     }
+
+    /// Copies `c` bytes from starting the memory `a` and offset `b` into the
+    /// stack.
+    ///
+    /// ```txt
+    /// mem[$ssp, $rC] = memory[$rA + $rB, $rC]
+    /// ```
+    pub(crate) fn load_memory_code(
+        mut self,
+        input_src_addr: Word,
+        input_offset: Word,
+        length_unpadded: Word,
+    ) -> IoResult<(), S::DataError>
+    where
+        S: InterpreterStorage,
+    {
+        let ssp = *self.ssp;
+        let sp = *self.sp;
+        let dst = ssp;
+
+        if ssp != sp {
+            return Err(PanicReason::ExpectedUnallocatedStack.into())
+        }
+
+        if length_unpadded == 0 {
+            inc_pc(self.pc)?;
+            return Ok(())
+        }
+
+        let current_contract = current_contract(self.context, self.fp, self.memory)?;
+
+        let length = bytes::padded_len_word(length_unpadded).unwrap_or(Word::MAX);
+
+        // Fetch the storage blob
+        let profiler = ProfileGas {
+            pc: self.pc.as_ref(),
+            is: self.is,
+            current_contract,
+            profiler: self.profiler,
+        };
+        let charge_len = length;
+        dependent_gas_charge_without_base(
+            self.cgas,
+            self.ggas,
+            profiler,
+            self.gas_cost,
+            charge_len,
+        )?;
+
+        let new_sp = ssp.saturating_add(length);
+        self.memory.grow_stack(new_sp)?;
+
+        // Set up ownership registers for the copy using old ssp
+        let owner = OwnershipRegisters::only_allow_stack_write(new_sp, ssp, *self.hp);
+        let src = input_src_addr.saturating_add(input_offset);
+
+        // Copy the code.
+        self.memory.memcopy(dst, src, length, owner)?;
+
+        // Mark stack space as allocated
+        *self.sp = new_sp;
+        *self.ssp = new_sp;
+
+        // Update frame code size, if we have a stack frame (i.e. fp > 0)
+        if self.context.is_internal() {
+            let code_size_ptr =
+                (*self.fp).saturating_add(CallFrame::code_size_offset() as Word);
+            let old_code_size =
+                Word::from_be_bytes(self.memory.read_bytes(code_size_ptr)?);
+            let old_code_size = padded_len_word(old_code_size)
+                .expect("Code size cannot overflow with padding");
+            let new_code_size = old_code_size
+                .checked_add(length as Word)
+                .ok_or(PanicReason::MemoryOverflow)?;
+
+            self.memory
+                .write_bytes_noownerchecks(code_size_ptr, new_code_size.to_be_bytes())?;
+        }
+
+        inc_pc(self.pc)?;
+
+        Ok(())
+    }
 }
 
 struct BurnCtx<'vm, S> {

fuel-vm/src/interpreter/memory.rs

@@ -326,23 +326,74 @@ impl MemoryInstance {
         Ok(())
     }
 
-    /// Copies the memory from `src` to `dst`.
+    /// Copies the memory from `src` to `dst` verifying ownership.
     #[inline]
     #[track_caller]
-    pub fn memcopy_noownerchecks<A: ToAddr, B: ToAddr, C: ToAddr>(
+    pub fn memcopy(
         &mut self,
-        dst: A,
-        src: B,
-        len: C,
+        dst: Word,
+        src: Word,
+        length: Word,
+        owner: OwnershipRegisters,
     ) -> Result<(), PanicReason> {
-        // TODO: Optimize
+        let dst_range = self.verify(dst, length)?;
+        let src_range = self.verify(src, length)?;
+
+        if dst_range.start() <= src_range.start() && src_range.start() < dst_range.end()
+            || src_range.start() <= dst_range.start()
+                && dst_range.start() < src_range.end()
+            || dst_range.start() < src_range.end() && src_range.end() <= dst_range.end()
+            || src_range.start() < dst_range.end() && dst_range.end() <= src_range.end()
+        {
+            return Err(PanicReason::MemoryWriteOverlap)
+        }
 
-        let src = src.to_addr()?;
-        let dst = dst.to_addr()?;
-        let len = len.to_addr()?;
+        owner.verify_ownership(&dst_range)?;
+
+        if src_range.end() <= self.stack.len() {
+            if dst_range.end() <= self.stack.len() {
+                self.stack
+                    .copy_within(src_range.usizes(), dst_range.start());
+            } else if dst_range.start() >= self.heap_offset() {
+                #[allow(clippy::arithmetic_side_effects)]
+                // Safety: subtractions are checked above
+                let dst_start = dst_range.start() - self.heap_offset();
+                #[allow(clippy::arithmetic_side_effects)]
+                // Safety: subtractions are checked above
+                let dst_end = dst_range.end() - self.heap_offset();
+
+                let src_array = &self.stack[src_range.usizes()];
+                let dst_array = &mut self.heap[dst_start..dst_end];
+                dst_array.copy_from_slice(src_array);
+            } else {
+                unreachable!("Range was verified to be valid")
+            }
+        } else if src_range.start() >= self.heap_offset() {
+            #[allow(clippy::arithmetic_side_effects)]
+            // Safety: subtractions are checked above
+            let src_start = src_range.start() - self.heap_offset();
+            #[allow(clippy::arithmetic_side_effects)]
+            // Safety: subtractions are checked above
+            let src_end = src_range.end() - self.heap_offset();
+
+            if dst_range.end() <= self.stack.len() {
+                let src_array = &self.heap[src_start..src_end];
+
+                let dst_array = &mut self.stack[dst_range.usizes()];
+                dst_array.copy_from_slice(src_array);
+            } else if dst_range.start() >= self.heap_offset() {
+                #[allow(clippy::arithmetic_side_effects)]
+                // Safety: subtractions are checked above
+                let dst_start = dst_range.start() - self.heap_offset();
+
+                self.heap.copy_within(src_start..src_end, dst_start);
+            } else {
+                unreachable!("Range was verified to be valid")
+            }
+        } else {
+            unreachable!("Range was verified to be valid")
+        }
 
-        let tmp = self.read(src, len)?.to_vec();
-        self.write_noownerchecks(dst, len)?.copy_from_slice(&tmp);
         Ok(())
     }
 
@@ -857,23 +908,11 @@ pub(crate) fn memcopy(
     memory: &mut MemoryInstance,
     owner: OwnershipRegisters,
     pc: RegMut<PC>,
-    a: Word,
-    b: Word,
-    c: Word,
+    dst: Word,
+    src: Word,
+    length: Word,
 ) -> SimpleResult<()> {
-    let dst_range = memory.verify(a, c)?;
-    let src_range = memory.verify(b, c)?;
-
-    if dst_range.start() <= src_range.start() && src_range.start() < dst_range.end()
-        || src_range.start() <= dst_range.start() && dst_range.start() < src_range.end()
-        || dst_range.start() < src_range.end() && src_range.end() <= dst_range.end()
-        || src_range.start() < dst_range.end() && dst_range.end() <= src_range.end()
-    {
-        return Err(PanicReason::MemoryWriteOverlap.into())
-    }
-
-    owner.verify_ownership(&dst_range)?;
-    memory.memcopy_noownerchecks(a, b, c)?;
+    memory.memcopy(dst, src, length, owner)?;
 
     Ok(inc_pc(pc)?)
 }

fuel-vm/src/predicate.rs

@@ -321,7 +321,7 @@ mod tests {
                 Ok(()),
             ),
             (
-                // PC exceeding predicate bounds
+                // Use `LDC` with mode `1` to load the blob into the predicate.
                 vec![
                     // Allocate 32 byte on the heap.
                     op::movi(0x10, 32),
@@ -345,6 +345,41 @@ mod tests {
                 CORRECT_GAS,
                 Ok(()),
             ),
+            (
+                // Use `LDC` with mode `2` to load the part of the predicate from the
+                // transaction.
+                vec![
+                    // Skip the return opcodes. One of two opcodes is a good opcode that
+                    // returns `0x1`. This opcode is our source for the `LDC`
+                    // opcode. We will copy return good opcode to the end
+                    // of the `ssp` via `LDC`. And jump there to
+                    // return `true` from the predicate.
+                    op::jmpf(ZERO, 2),
+                    // Bad return opcode that we want to skip.
+                    op::ret(0x0),
+                    // Good return opcode that we want to use for the `LDC`.
+                    op::ret(0x1),
+                    // This will be our zeroed blob id
+                    op::move_(0x10, IS),
+                    // We don't need to copy `jmpf` and bad `ret` opcodes via `LDC`.
+                    op::addi(0x10, 0x10, 2 * Instruction::SIZE as u16),
+                    // Store start of the code
+                    op::move_(0x12, SSP),
+                    // Subtract the start of the code from the end of the code
+                    op::sub(0x12, 0x12, IS),
+                    // Divide the code by the instruction size to get the number of
+                    // instructions
+                    op::divi(0x12, 0x12, Instruction::SIZE as u16),
+                    // We want to load only on good `ret` opcode.
+                    op::movi(0x11, Instruction::SIZE as u32),
+                    // Load the code from the memory address `0x10` with the `0x11` size
+                    op::ldc(0x10, ZERO, 0x11, 2),
+                    // Jump to a new code location
+                    op::jmp(0x12),
+                ],
+                CORRECT_GAS,
+                Ok(()),
+            ),
         ];
 
         assert_inputs_are_validated_for_predicates(inputs, good_blob)
@@ -399,7 +434,7 @@ mod tests {
                 )),
             ),
             (
-                // PC exceeding predicate bounds
+                // Use `LDC` with mode `1` to load the blob into the predicate.
                 vec![
                     // Allocate 32 byte on the heap.
                     op::movi(0x10, 32),
@@ -425,6 +460,43 @@ mod tests {
                     PanicReason::PredicateReturnedNonOne,
                 )),
             ),
+            (
+                // Use `LDC` with mode `2` to load the part of the predicate from the
+                // transaction.
+                vec![
+                    // Skip the return opcodes. One of two opcodes is a bad opcode that
+                    // returns `0x0`. This opcode is our source for the `LDC`
+                    // opcode. We will copy return bad opcode to the end
+                    // of the `ssp` via `LDC`. And jump there to
+                    // return `false` from the predicate adn fail.
+                    op::jmpf(ZERO, 2),
+                    // Good return opcode that we want to skip.
+                    op::ret(0x1),
+                    // Bad return opcode that we want to use for the `LDC`.
+                    op::ret(0x0),
+                    // This will be our zeroed blob id
+                    op::move_(0x10, IS),
+                    // We don't need to copy `jmpf` and bad `ret` opcodes via `LDC`.
+                    op::addi(0x10, 0x10, 2 * Instruction::SIZE as u16),
+                    // Store start of the code
+                    op::move_(0x12, SSP),
+                    // Subtract the start of the code from the end of the code
+                    op::sub(0x12, 0x12, IS),
+                    // Divide the code by the instruction size to get the number of
+                    // instructions
+                    op::divi(0x12, 0x12, Instruction::SIZE as u16),
+                    // We want to load only on bad `ret` opcode.
+                    op::movi(0x11, Instruction::SIZE as u32),
+                    // Load the code from the memory address `0x10` with the `0x11` size
+                    op::ldc(0x10, ZERO, 0x11, 2),
+                    // Jump to a new code location
+                    op::jmp(0x12),
+                ],
+                CORRECT_GAS,
+                Err(PredicateVerificationFailed::Panic(
+                    PanicReason::PredicateReturnedNonOne,
+                )),
+            ),
         ];
 
         assert_inputs_are_validated_for_predicates(inputs, bad_blob)