void operator()() {
if (work.empty())
return;
int specimen_id = work.front().specimen_id;
// Database connections don't survive over fork() according to SqLite and PostgreSQL documentation, so open it again
SqlDatabase::TransactionPtr tx = SqlDatabase::Connection::create(databaseUrl)->transaction();
OutputGroups ogroups; // do not load from database (that might take a very long time)
if (opt.verbosity>=LACONIC) {
if (opt.verbosity>=EFFUSIVE)
std::cerr <<argv0 <<": " <<std::string(100, '#') <<"\n";
std::cerr <<argv0 <<": processing binary specimen \"" <<files.name(specimen_id) <<"\"\n";
}
// Parse the specimen
SgProject *project = files.load_ast(tx, specimen_id);
if (!project)
project = open_specimen(tx, files, specimen_id, argv0);
if (!project) {
std::cerr <<argv0 <<": problems loading specimen\n";
exit(1);
}
// Get list of specimen functions and initialize the instruction cache
std::vector<SgAsmFunction*> all_functions = SageInterface::querySubTree<SgAsmFunction>(project);
IdFunctionMap functions = existing_functions(tx, files, all_functions);
FunctionIdMap function_ids;
AddressIdMap entry2id; // maps function entry address to function ID
for (IdFunctionMap::iterator fi=functions.begin(); fi!=functions.end(); ++fi) {
function_ids[fi->second] = fi->first;
entry2id[fi->second->get_entry_va()] = fi->first;
}
InstructionProvidor insns = InstructionProvidor(all_functions);
// Split the work list into chunks, each containing testsPerChunk except the last, which may contain fewer.
static const size_t testsPerChunk = 25;
size_t nChunks = (work.size() + testsPerChunk - 1) / testsPerChunk;
std::vector<SomeTests> jobs;
for (size_t i=0; i<nChunks; ++i) {
size_t beginWorkIdx = i * testsPerChunk;
size_t endWorkIdx = std::min((i+1)*testsPerChunk, work.size());
Work partWork(work.begin()+beginWorkIdx, work.begin()+endWorkIdx);
jobs.push_back(SomeTests(partWork, databaseUrl, functions, function_ids, &insns, cmd_id, &entry2id));
}
// Run the parts in parallel using the maximum parallelism specified on the command-line. We must commit our
// transaction before forking, otherwise the children won't see the rows we've added to various tables.
tx->commit();
tx.reset();
size_t nfailed = runInParallel(jobs, opt.nprocs);
if (nfailed!=0) {
std::cerr <<"SpecimenProcessor: " <<StringUtility::plural(nfailed, "jobs") <<" failed\n";
exit(1);
}
}
int
main(int argc, char *argv[])
{
// Parse command-line
opt.nprocs = nProcessors();
int argno = parse_commandline(argc, argv);
if (argno+1!=argc)
usage(1);
std::string databaseUrl = argv[argno++];
SqlDatabase::TransactionPtr tx = SqlDatabase::Connection::create(databaseUrl)->transaction();
int64_t cmd_id = start_command(tx, argc, argv, "running tests");
// Load worklist
MultiWork work;
load_sorted_work(work/*out*/);
if (work.empty())
return 0;
// Load information about files. The transaction is not saved anywhere.
FilesTable files(tx);
// We must commit our transaction before we fork, otherwise the child processes won't be able to see the rows we've
// inserted. Specifically, the row in the semantic_history table that says who we are. Destroy the smart pointer so that
// the connection is even closed.
tx->commit();
tx.reset();
// Process work items for each specimen sequentially
BOOST_FOREACH (const Work &workForSpecimen, work)
if (forkAndWait(SpecimenProcessor(workForSpecimen, files, databaseUrl, cmd_id)))
exit(1);
// Indicate that this command is finished
tx = SqlDatabase::Connection::create(databaseUrl)->transaction();
finish_command(tx, cmd_id, "ran tests");
tx->commit();
return 0;
}
void operator()() {
// Database connections don't survive over fork() according to SqLite and PostgreSQL documentation, so open it again
SqlDatabase::TransactionPtr tx = SqlDatabase::Connection::create(databaseUrl)->transaction();
// Use zero for the number of tests ran so that this child process doesn't try to update the semantic_history table.
// If two or more processes try to change the same row (which they will if there's a non-zero number of tests) then
// they will deadlock with each other.
static const size_t NO_TESTS_RAN = 0;
NameSet builtin_function_names;
add_builtin_functions(builtin_function_names/*out*/);
InputGroup igroup;
WorkItem prevWorkItem;
SgAsmInterpretation *prev_interp = NULL;
MemoryMap ro_map;
Disassembler::AddressSet whitelist_exports; // dynamic functions that should be called
PointerDetectors pointers;
InsnCoverage insn_coverage;
DynamicCallGraph dynamic_cg;
Tracer tracer;
ConsumedInputs consumed_inputs;
FuncAnalyses funcinfo;
OutputGroups ogroups; // do not load from database (that might take a very long time)
time_t last_checkpoint = time(NULL);
for (size_t workIdx=0; workIdx<work.size(); ++workIdx) {
WorkItem &workItem = work[workIdx];
// Load the input group from the database if necessary.
if (workItem.igroup_id!=prevWorkItem.igroup_id) {
if (!igroup.load(tx, workItem.igroup_id)) {
std::cerr <<argv0 <<": input group " <<workItem.igroup_id <<" is empty or does not exist\n";
exit(1);
}
}
// Find the function to test
IdFunctionMap::iterator func_found = functions.find(workItem.func_id);
assert(func_found!=functions.end());
SgAsmFunction *func = func_found->second;
if (opt.verbosity>=LACONIC) {
if (opt.verbosity>=EFFUSIVE)
std::cerr <<argv0 <<": " <<std::string(100, '=') <<"\n";
std::cerr <<argv0 <<": processing function " <<function_to_str(func, function_ids) <<"\n";
}
SgAsmInterpretation *interp = SageInterface::getEnclosingNode<SgAsmInterpretation>(func);
assert(interp!=NULL);
// Do per-interpretation stuff
if (interp!=prev_interp) {
prev_interp = interp;
assert(interp->get_map()!=NULL);
ro_map = *interp->get_map();
ro_map.require(MemoryMap::READABLE).prohibit(MemoryMap::WRITABLE).keep();
Disassembler::AddressSet whitelist_imports = get_import_addresses(interp, builtin_function_names);
whitelist_exports.clear(); // imports are addresses of import table slots; exports are functions
overmap_dynlink_addresses(interp, *insns, opt.params.follow_calls, &ro_map, GOTPLT_VALUE,
whitelist_imports, whitelist_exports/*out*/);
if (opt.verbosity>=EFFUSIVE) {
std::cerr <<argv0 <<": memory map for SgAsmInterpretation:\n";
interp->get_map()->dump(std::cerr, argv0+": ");
}
}
// Run the test
assert(insns!=NULL);
assert(entry2id!=NULL);
std::cerr <<"process " <<getpid() <<" about to run test " <<workIdx <<"/" <<work.size() <<" " <<workItem <<"\n";
runOneTest(tx, workItem, pointers, func, function_ids, insn_coverage, dynamic_cg, tracer, consumed_inputs,
interp, whitelist_exports, cmd_id, igroup, funcinfo, *insns, &ro_map, *entry2id, ogroups);
++ntests_ran;
// Checkpoint
if (opt.checkpoint>0 && time(NULL)-last_checkpoint > opt.checkpoint) {
if (!opt.dry_run)
tx = checkpoint(tx, ogroups, tracer, insn_coverage, dynamic_cg, consumed_inputs, NULL, NO_TESTS_RAN,
cmd_id);
last_checkpoint = time(NULL);
}
prevWorkItem = workItem;
}
std::cerr <<"process " <<getpid() <<" is done testing; now finishing up...\n";
if (!tx->is_terminated()) {
SqlDatabase::StatementPtr stmt = tx->statement("insert into semantic_funcpartials"
" (func_id, ncalls, nretused, ntests, nvoids) values"
" (?, ?, ?, ?, ?)");
for (FuncAnalyses::iterator fi=funcinfo.begin(); fi!=funcinfo.end(); ++fi) {
stmt->bind(0, fi->first);
stmt->bind(1, fi->second.ncalls);
stmt->bind(2, fi->second.nretused);
stmt->bind(3, fi->second.ntests);
stmt->bind(4, fi->second.nvoids);
stmt->execute();
}
}
// Cleanup
if (!tx->is_terminated() && !opt.dry_run) {
std::cerr <<"process " <<getpid() <<" is doing the final checkpoint\n";
checkpoint(tx, ogroups, tracer, insn_coverage, dynamic_cg, consumed_inputs, NULL, NO_TESTS_RAN, cmd_id);
}
tx.reset();
std::cerr <<"process " <<getpid() <<" finished\n";
}