syz-manager: more consistently check disabled syscalls

We have program "validity" check duplicated 4 times
(initially it was just "does it deserialize?").
Then we added program length and disabled syscall.
But some of the sites have only a subset of checks.
Factor out program checking procedure into a separate function
and use it at all sites.

3 files changed, 36 insertions, 39 deletions

diff --git a/syz-manager/hub.go b/syz-manager/hub.go
index 51937b536..9028f0233 100644
--- a/syz-manager/hub.go
+++ b/syz-manager/hub.go
@@ -21,7 +21,7 @@ func (mgr *Manager) hubSyncLoop() {
 		cfg:           mgr.cfg,
 		target:        mgr.target,
 		stats:         mgr.stats,
-		enabledCalls:  mgr.checkResult.EnabledCalls[mgr.cfg.Sandbox],
+		enabledCalls:  mgr.targetEnabledSyscalls,
 		leak:          mgr.checkResult.Features[host.FeatureLeak].Enabled,
 		fresh:         mgr.fresh,
 		hubReproQueue: mgr.hubReproQueue,
@@ -37,7 +37,7 @@ type HubConnector struct {
 	cfg            *mgrconfig.Config
 	target         *prog.Target
 	stats          *Stats
-	enabledCalls   []int
+	enabledCalls   map[*prog.Syscall]bool
 	leak           bool
 	fresh          bool
 	hubCorpus      map[hash.Sig]bool
@@ -80,8 +80,8 @@ func (hc *HubConnector) connect(corpus [][]byte) (*rpctype.RPCClient, error) {
 		Manager: hc.cfg.Name,
 		Fresh:   hc.fresh,
 	}
-	for _, id := range hc.enabledCalls {
-		a.Calls = append(a.Calls, hc.target.Syscalls[id].Name)
+	for call := range hc.enabledCalls {
+		a.Calls = append(a.Calls, call.Name)
 	}
 	hubCorpus := make(map[hash.Sig]bool)
 	for _, inp := range corpus {
@@ -171,8 +171,10 @@ func (hc *HubConnector) processProgs(progs [][]byte) int {
 	dropped := 0
 	candidates := make([][]byte, 0, len(progs))
 	for _, inp := range progs {
-		p, err := hc.target.Deserialize(inp, prog.NonStrict)
-		if err != nil || len(p.Calls) > prog.MaxCalls {
+		bad, disabled := checkProgram(hc.target, hc.enabledCalls, inp)
+		if bad || disabled {
+			log.Logf(0, "rejecting program from hub (bad=%v, disabled=%v):\n%s",
+				bad, disabled, inp)
 			dropped++
 			continue
 		}
@@ -185,7 +187,10 @@ func (hc *HubConnector) processProgs(progs [][]byte) int {
 func (hc *HubConnector) processRepros(repros [][]byte) int {
 	dropped := 0
 	for _, repro := range repros {
-		if _, err := hc.target.Deserialize(repro, prog.NonStrict); err != nil {
+		bad, disabled := checkProgram(hc.target, hc.enabledCalls, repro)
+		if bad || disabled {
+			log.Logf(0, "rejecting repro from hub (bad=%v, disabled=%v):\n%s",
+				bad, disabled, repro)
 			dropped++
 			continue
 		}
diff --git a/syz-manager/manager.go b/syz-manager/manager.go
index ab1179f95..3e44d24e2 100644
--- a/syz-manager/manager.go
+++ b/syz-manager/manager.go
@@ -470,27 +470,14 @@ func (mgr *Manager) loadCorpus() {
 		fallthrough
 	case currentDBVersion:
 	}
-	broken, tooLong := 0, 0
+	broken := 0
 	for key, rec := range mgr.corpusDB.Records {
-		p, err := mgr.target.Deserialize(rec.Val, prog.NonStrict)
-		if err != nil {
+		bad, disabled := checkProgram(mgr.target, mgr.targetEnabledSyscalls, rec.Val)
+		if bad {
 			mgr.corpusDB.Delete(key)
 			broken++
 			continue
 		}
-		if len(p.Calls) > prog.MaxCalls {
-			mgr.corpusDB.Delete(key)
-			tooLong++
-			continue
-		}
-
-		disabled := false
-		for _, c := range p.Calls {
-			if !mgr.targetEnabledSyscalls[c.Meta] {
-				disabled = true
-				break
-			}
-		}
 		if disabled {
 			// This program contains a disabled syscall.
 			// We won't execute it, but remember its hash so
@@ -505,8 +492,8 @@ func (mgr *Manager) loadCorpus() {
 		})
 	}
 	mgr.fresh = len(mgr.corpusDB.Records) == 0
-	log.Logf(0, "%-24v: %v (deleted %v broken, %v too long)",
-		"corpus", len(mgr.candidates), broken, tooLong)
+	log.Logf(0, "%-24v: %v (deleted %v broken)",
+		"corpus", len(mgr.candidates), broken)
 
 	// Now this is ugly.
 	// We duplicate all inputs in the corpus and shuffle the second part.
@@ -525,6 +512,22 @@ func (mgr *Manager) loadCorpus() {
 	mgr.phase = phaseLoadedCorpus
 }
 
+func checkProgram(target *prog.Target, enabled map[*prog.Syscall]bool, data []byte) (bad, disabled bool) {
+	p, err := target.Deserialize(data, prog.NonStrict)
+	if err != nil {
+		return true, true
+	}
+	if len(p.Calls) > prog.MaxCalls {
+		return true, true
+	}
+	for _, c := range p.Calls {
+		if !enabled[c.Meta] {
+			return false, true
+		}
+	}
+	return false, false
+}
+
 func (mgr *Manager) runInstance(index int) (*Crash, error) {
 	mgr.checkUsedFiles()
 	inst, err := mgr.vmPool.Create(index)
diff --git a/syz-manager/rpc.go b/syz-manager/rpc.go
index d734b8da9..6cf979a38 100644
--- a/syz-manager/rpc.go
+++ b/syz-manager/rpc.go
@@ -211,22 +211,11 @@ func (serv *RPCServer) NewInput(a *rpctype.NewInputArgs, r *int) error {
 	inputSignal := a.Signal.Deserialize()
 	log.Logf(4, "new input from %v for syscall %v (signal=%v, cover=%v)",
 		a.Name, a.Call, inputSignal.Len(), len(a.Cover))
-	p, err := serv.target.Deserialize(a.RPCInput.Prog, prog.NonStrict)
-	if err != nil {
-		// This should not happen, but we see such cases episodically (probably corrupted VM memory).
-		log.Logf(0, "failed to deserialize program from fuzzer: %v\n%s", err, a.RPCInput.Prog)
-		return nil
-	}
-	if len(p.Calls) > prog.MaxCalls {
-		log.Logf(0, "rejecting too long program from fuzzer: %v calls\n%s", len(p.Calls), a.RPCInput.Prog)
+	bad, disabled := checkProgram(serv.target, serv.targetEnabledSyscalls, a.RPCInput.Prog)
+	if bad || disabled {
+		log.Logf(0, "rejecting program from fuzzer (bad=%v, disabled=%v):\n%s", bad, disabled, a.RPCInput.Prog)
 		return nil
 	}
-	for _, call := range p.Calls {
-		if !serv.targetEnabledSyscalls[call.Meta] {
-			log.Logf(0, "rejecting program with disabled call %v:\n%s", call.Meta.Name, a.RPCInput.Prog)
-			return nil
-		}
-	}
 	serv.mu.Lock()
 	defer serv.mu.Unlock()