1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
// Copyright (c) The Diem Core Contributors
// SPDX-License-Identifier: Apache-2.0
// Analysis pass which analyzes how to injects global invariants into the bytecode.
use crate::{
function_target::{FunctionData, FunctionTarget},
function_target_pipeline::{
FunctionTargetProcessor, FunctionTargetsHolder, FunctionVariant, VerificationFlavor,
},
stackless_bytecode::{BorrowNode, Bytecode, Operation},
usage_analysis,
verification_analysis::{is_invariant_suspendable, InvariantAnalysisData},
};
use move_binary_format::file_format::CodeOffset;
use move_model::{
ast::ConditionKind,
model::{FunctionEnv, GlobalEnv, GlobalId, QualifiedInstId, StructId},
ty::{Type, TypeDisplayContext, TypeInstantiationDerivation, TypeUnificationAdapter, Variance},
};
use std::{
collections::{BTreeMap, BTreeSet},
fmt,
};
/// A named struct for holding the information on how an invariant is relevant to a bytecode.
#[derive(Default, Clone)]
pub struct PerBytecodeRelevance {
/// for each `inst_fun` (instantiation of function type parameters) in the key set, the
/// associated value is a set of `inst_inv` (instantiation of invariant type parameters) that
/// are applicable to the concrete function instance `F<inst_fun>`.
pub insts: BTreeMap<Vec<Type>, BTreeSet<Vec<Type>>>,
}
/// A named struct for holding the information on how invariants are relevant to a function.
#[derive(Clone)]
pub struct PerFunctionRelevance {
/// Invariants that needs to be assumed at function entrypoint
/// - Key: global invariants that needs to be assumed before the first instruction,
/// - Value: the instantiation information per each related invariant.
pub entrypoint_assumptions: BTreeMap<GlobalId, PerBytecodeRelevance>,
/// For each bytecode at given code offset, the associated value is a map of
/// - Key: global invariants that needs to be asserted after the bytecode instruction and
/// - Value: the instantiation information per each related invariant.
pub per_bytecode_assertions: BTreeMap<CodeOffset, BTreeMap<GlobalId, PerBytecodeRelevance>>,
}
/// Get verification information for this function.
pub fn get_info<'env>(target: &FunctionTarget<'env>) -> &'env PerFunctionRelevance {
target
.get_annotations()
.get::<PerFunctionRelevance>()
.expect("Global invariant analysis not performed")
}
// The function target processor
pub struct GlobalInvariantAnalysisProcessor {}
impl GlobalInvariantAnalysisProcessor {
pub fn new() -> Box<Self> {
Box::new(Self {})
}
}
impl FunctionTargetProcessor for GlobalInvariantAnalysisProcessor {
fn process(
&self,
targets: &mut FunctionTargetsHolder,
fun_env: &FunctionEnv<'_>,
mut data: FunctionData,
) -> FunctionData {
if fun_env.is_native() || fun_env.is_intrinsic() {
// Nothing to do
return data;
}
if !data.variant.is_verified() {
// Only need to instrument if this is a verification variant
return data;
}
// Analyze invariants
let target = FunctionTarget::new(fun_env, &data);
let analysis_result = PerFunctionRelevance::analyze(&target, targets);
data.annotations.set(analysis_result);
// This is an analysis pass, nothing gets changed
data
}
fn name(&self) -> String {
"global_invariant_analysis".to_string()
}
fn dump_result(
&self,
f: &mut fmt::Formatter<'_>,
env: &GlobalEnv,
targets: &FunctionTargetsHolder,
) -> fmt::Result {
// utils
let type_display_ctxt = TypeDisplayContext::WithEnv {
env,
type_param_names: None,
};
let display_type_slice = |tys: &[Type]| -> String {
let content = tys
.iter()
.map(|t| t.display(&type_display_ctxt).to_string())
.collect::<Vec<_>>()
.join(", ");
format!("<{}>", content)
};
let make_indent = |indent: usize| " ".repeat(indent);
let display_inv_relevance = |f: &mut fmt::Formatter,
invs: &BTreeMap<GlobalId, PerBytecodeRelevance>,
header: &str,
assert_or_assume: &str|
-> fmt::Result {
let mut indent = 1;
// oneliner for empty invs
if invs.is_empty() {
return writeln!(f, "{}{} {{}}", make_indent(indent), header);
}
writeln!(f, "{}{} {{", make_indent(indent), header)?;
indent += 1;
for (inv_id, inv_rel) in invs {
writeln!(
f,
"{}{} {} = [",
make_indent(indent),
assert_or_assume,
inv_id
)?;
indent += 1;
for (rel_inst, inv_insts) in &inv_rel.insts {
writeln!(
f,
"{}{} -> [",
make_indent(indent),
display_type_slice(rel_inst)
)?;
indent += 1;
for inv_inst in inv_insts {
writeln!(f, "{}{}", make_indent(indent), display_type_slice(inv_inst))?;
}
indent -= 1;
writeln!(f, "{}]", make_indent(indent))?;
}
indent -= 1;
writeln!(f, "{}]", make_indent(indent))?;
}
indent -= 1;
writeln!(f, "{}}}", make_indent(indent))
};
writeln!(
f,
"\n********* Result of global invariant instrumentation *********\n"
)?;
for (fun_id, fun_variant) in targets.get_funs_and_variants() {
if !matches!(
fun_variant,
FunctionVariant::Verification(VerificationFlavor::Regular)
) {
// the analysis results are available in the regular verification variant
continue;
}
let fenv = env.get_function(fun_id);
let target = targets.get_target(&fenv, &fun_variant);
let result = target
.get_annotations()
.get::<PerFunctionRelevance>()
.expect("Analysis not performed");
writeln!(f, "{}: [", fenv.get_full_name_str())?;
// display entrypoint assumptions
display_inv_relevance(f, &result.entrypoint_assumptions, "entrypoint", "assume")?;
// display per-bytecode assertions
for (code_offset, code_invs) in &result.per_bytecode_assertions {
let bc = target.data.code.get(*code_offset as usize).unwrap();
let header = format!("{}: {}", code_offset, bc.display(&target, &BTreeMap::new()));
display_inv_relevance(f, code_invs, &header, "assert")?;
}
writeln!(f, "]")?;
}
writeln!(f, "\n********* Global invariants by ID *********\n")?;
let mut all_invs = BTreeSet::new();
for menv in env.get_modules() {
all_invs.extend(env.get_global_invariants_by_module(menv.get_id()));
}
for inv_id in all_invs {
let inv = env.get_global_invariant(inv_id).unwrap();
let inv_src = env.get_source(&inv.loc).unwrap_or("<unknown invariant>");
writeln!(f, "{} => {}", inv_id, inv_src)?;
}
writeln!(f)
}
}
/// This impl block is about the analysis pass
impl PerFunctionRelevance {
/// Collect and build the relevance analysis information for this function target.
fn analyze(target: &FunctionTarget, targets: &FunctionTargetsHolder) -> Self {
use BorrowNode::*;
use Bytecode::*;
use Operation::*;
// collect information
let fid = target.func_env.get_qualified_id();
let env = target.global_env();
let inv_analysis = env
.get_extension::<InvariantAnalysisData>()
.expect("Verification analysis not performed");
let check_suspendable_inv_on_return =
inv_analysis.fun_set_with_inv_check_on_exit.contains(&fid);
let inv_applicability = inv_analysis
.fun_to_inv_map
.get(&fid)
.expect("Invariant applicability not available");
let mem_analysis = usage_analysis::get_memory_usage(target);
let fun_type_params = target.get_type_parameters();
let fun_type_params_arity = fun_type_params.len();
// collect invariant applicability and instantiation information for entrypoint assumptions
//
// NOTE: why do we use the `InvariantRelevance::accessed` set instead of other sets?
//
// - The reason we choose `accessed` over `direct_accessed` is that sometimes we need to
// assume invariants that are applicable to callees only and not applicable to the caller.
// The reason is that if we inline a callee function, proving properties about the inlined
// function might require assumptions about the memories it touches (e.g., proving the
// `borrow_global_mut<R>(addr)` does not abort with the invariant that resource `R` must
// exist under account `addr` after operation has started).
//
// It does not hurt (in terms of soundness or completeness of the proofs) to assume extra
// invariants in the `accessed` set even when these assumptions are not actually used in
// proofs of any asserts. We might re-consider this when performance (due to too many
// assumptions added to the proof system) becomes an issue.
//
// - The reason we choose `direct_accessed` over `direct_modified` is that we may need
// assumptions from global invariants to prove properties in the code.
//
// For example, we may have an `invariant exists<A>(0x1) ==> exists<B>(0x1);` while in
// the code, we have `if (exists<A>(0x1)) { borrow_global<B>(0x1); }`. With the global
// invariant, we know that the `borrow_global` won't abort. But we won't be able to prove
// this property without the global invariant.
let entrypoint_invariants: BTreeSet<_> = inv_applicability
.accessed
.iter()
.filter_map(|&inv_id| {
let inv = env.get_global_invariant(inv_id).unwrap();
// update invariants should not be assumed at function entrypoint.
matches!(inv.kind, ConditionKind::GlobalInvariant(..)).then(|| inv_id)
})
.collect();
let entrypoint_assumptions = Self::calculate_invariant_relevance(
env,
mem_analysis.accessed.all.iter(),
&entrypoint_invariants,
fun_type_params_arity,
);
// if this function defers invariant checking on return, filter out invariants that are
// suspended in body.
//
// NOTE: why do we use the `InvariantRelevance::direct_modified` set instead of other sets?
//
// First, be reminded that in the rest of this function, we aim to find which invariants
// should be *asserted* at each bytecode instruction. Therefore, if a bytecode instruction
// only reads some memory but never modifies one, we don't need to assert the invariant.
// This rules out the `direct_accessed` and `accessed` sets.
//
// Second, similar to the reason why we choose `direct_accessed` set over `accessed` for
// invariants that constitute entrypoint assumptions, we choose `direct_modified` over
// `modified` is that we don't want to assert invariants that are applicable to callees
// only and not applicable to the caller. The reason is still: if a suspendable invariant is
// delegated to the caller, that invariant will appear in the `direct_modified` set on the
// caller side.
let (inv_related_return, inv_related_normal): (BTreeSet<_>, BTreeSet<_>) =
if check_suspendable_inv_on_return {
inv_applicability
.direct_modified
.iter()
.cloned()
.partition(|inv_id| is_invariant_suspendable(env, *inv_id))
} else {
(BTreeSet::new(), inv_applicability.direct_modified.clone())
};
// collect invariant applicability and instantiation information per bytecode, i.e.,
// - which invariants should be instrumented after each instruction and
// - per each invariant applicable, how to instantiate them.
let mut per_bytecode_assertions = BTreeMap::new();
let mut mem_related_on_return = BTreeSet::new();
for (code_offset, bc) in target.data.code.iter().enumerate() {
let code_offset = code_offset as CodeOffset;
// collect memory modified in operations
let mem_related = match bc {
Call(_, _, oper, _, _) => match oper {
Function(mid, fid, inst) | OpaqueCallEnd(mid, fid, inst) => {
let callee_fid = mid.qualified(*fid);
// shortcut the call if the callee does not delegate invariant checking.
//
// NOTE: in this case, memories modified by the callee are NOT back
// propagated to the caller in the `verification_analysis.rs`, which means,
// the `InvariantRelevance::direct_modified` set for the caller does NOT
// necessarily cover invariants that are related to the callee.
if !inv_analysis.fun_set_with_no_inv_check.contains(&callee_fid) {
continue;
}
let callee_env = env.get_function(callee_fid);
let callee_target =
targets.get_target(&callee_env, &FunctionVariant::Baseline);
let callee_usage = usage_analysis::get_memory_usage(&callee_target);
// NOTE: it is important to include *ALL* memories modified by the callee
// instead of just the direct ones --- if a function `F` delegates
// suspendable invariant checking to its caller, all the functions that `F`
// calls will not check suspendable invariants anymore.
callee_usage.modified.get_all_inst(inst)
}
MoveTo(mid, sid, inst) | MoveFrom(mid, sid, inst) => {
let mem = mid.qualified_inst(*sid, inst.to_owned());
std::iter::once(mem).collect()
}
WriteBack(GlobalRoot(mem), _) => std::iter::once(mem.clone()).collect(),
// shortcut other operations
_ => continue,
},
Ret(..) if check_suspendable_inv_on_return => {
std::mem::take(&mut mem_related_on_return)
}
// shortcut other bytecodes
_ => continue,
};
// mark whether we are processing a return instruction
let is_return = matches!(bc, Ret(..));
// select the related invariants based on whether this bytecode instruction is a return
let inv_related = if is_return {
&inv_related_return
} else {
&inv_related_normal
};
// collect instantiation information
let relevance = Self::calculate_invariant_relevance(
env,
mem_related.iter(),
inv_related,
fun_type_params_arity,
);
per_bytecode_assertions.insert(code_offset, relevance);
// save the related memories for return point if the function defers that
if check_suspendable_inv_on_return && !is_return {
mem_related_on_return.extend(mem_related);
}
}
// sanity check: the deferred memory is indeed consumed by a return instruction, UNLESS
// the deferred memory do not touch anything that is checked in any suspendable invariant.
if !mem_related_on_return.is_empty() {
let mut deferred_invs = vec![];
'error_check: for inv_id in inv_related_return {
let inv = env.get_global_invariant(inv_id).unwrap();
for inv_mem in &inv.mem_usage {
let inv_ty = inv_mem.to_type();
for rel_mem in &mem_related_on_return {
let rel_ty = rel_mem.to_type();
let adapter =
TypeUnificationAdapter::new_pair(&rel_ty, &inv_ty, true, true);
if adapter.unify(Variance::Allow, false).is_none() {
deferred_invs.push(inv_id);
continue 'error_check;
}
}
}
}
if !deferred_invs.is_empty() {
env.error(
&target.get_loc(),
&format!(
"Function `{}` defers the checking of {} suspendable invariants to the \
return point, but the function never returns",
target.func_env.get_full_name_str(),
deferred_invs.len(),
),
);
}
}
// wrap and return the analysis result
Self {
entrypoint_assumptions,
per_bytecode_assertions,
}
}
/// Given a set of memories, calculate the global invariants that are related to this memory
/// set and for each related global invariant, derive how to instantiate the invariant to make
/// it relevant.
fn calculate_invariant_relevance<'a>(
env: &GlobalEnv,
mem_related: impl Iterator<Item = &'a QualifiedInstId<StructId>>,
inv_related: &BTreeSet<GlobalId>,
fun_type_params_arity: usize,
) -> BTreeMap<GlobalId, PerBytecodeRelevance> {
let mut result = BTreeMap::new();
for rel_mem in mem_related {
let rel_ty = rel_mem.to_type();
for inv_id in inv_related {
let inv = env.get_global_invariant(*inv_id).unwrap();
let inv_type_params = match &inv.kind {
ConditionKind::GlobalInvariant(params) => params,
ConditionKind::GlobalInvariantUpdate(params) => params,
_ => unreachable!(
"A global invariant must have a condition kind of either \
`GlobalInvariant` or `GlobalInvariantUpdate`"
),
};
let inv_type_params_arity = inv_type_params.len();
for inv_mem in &inv.mem_usage {
let inv_ty = inv_mem.to_type();
// make sure these two types unify before trying to instantiate them
let adapter = TypeUnificationAdapter::new_pair(&rel_ty, &inv_ty, true, true);
if adapter.unify(Variance::Allow, false).is_none() {
continue;
}
// instantiate the bytecode first
//
// NOTE: in fact, in this phase, we don't intend to instantiation the function
// nor do we want to collect information on how this function (or this bytecode)
// needs to be instantiated. All we care is how the invariant should be
// instantiated in order to be instrumented at this code point, with a generic
// function and generic code.
//
// But unfortunately, based on how the type unification logic is written now,
// this two-step instantiation is needed in order to find all possible
// instantiations of the invariant. I won't deny that there might be a way to
// collect invariant instantiation combinations without instantiating the
// function type parameters, but I haven't iron out one so far.
let rel_insts = TypeInstantiationDerivation::progressive_instantiation(
std::iter::once(&rel_ty),
std::iter::once(&inv_ty),
true,
true,
true,
false,
fun_type_params_arity,
true,
false,
);
// for each instantiation of the bytecode, instantiate the invariants
for rel_inst in rel_insts {
let inst_rel_ty = rel_ty.instantiate(&rel_inst);
let inv_insts = TypeInstantiationDerivation::progressive_instantiation(
std::iter::once(&inst_rel_ty),
std::iter::once(&inv_ty),
false,
true,
false,
true,
inv_type_params_arity,
false,
true,
);
// record the relevance information
result
.entry(*inv_id)
.or_insert_with(PerBytecodeRelevance::default)
.insts
.entry(rel_inst)
.or_insert_with(BTreeSet::new)
.extend(inv_insts);
}
}
}
}
result
}
}