pkg/email/lore: demand patches in patch series emails

On LKML, users sometimes reply to a patch series while keeping the
original title. At the same time, sometimes new series are sent as a
reply to the previous versions. It all confuses the lore archive
parsing logic.

When reading the archive, remember if the patch were present. If not,
mark the series as corrupted and thus prevent is appearance in
syz-cluster.

Author: a-nogikh

pkg/email/lore/parse.go

@@ -93,6 +93,11 @@ func PatchSeries(emails []*Email) []*Series {
 				// The cover email is not of interest.
 				continue
 			}
+			if !email.HasPatch {
+				// Sometimes users reply to the series keeping the original subject.
+				// Ignore such messages.
+				continue
+			}
 			if hasSeq[seq] {
 				// It's weird if that really happens, but let's skip for now.
 				continue

pkg/email/lore/parse_test.go

@@ -13,6 +13,7 @@ import (
 	"github.com/google/syzkaller/dashboard/dashapi"
 	"github.com/google/syzkaller/pkg/email"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestThreadsCollection(t *testing.T) {
@@ -381,6 +382,17 @@ func TestDiscussionType(t *testing.T) {
 	}
 }
 
+const dummyPatch = `diff --git a/kernel/kcov.c b/kernel/kcov.c
+index 85e5546cd791..949ea4574412 100644
+--- a/kernel/kcov.c
++++ b/kernel/kcov.c
+@@ -127,7 +127,6 @@ void kcov_task_exit(struct task_struct *t)
+ 	if (kcov == NULL)
+ 		return;
+-	spin_lock(&kcov->lock);
+ 	if (WARN_ON(kcov->t != t)) {
+`
+
 func TestParseSeries(t *testing.T) {
 	messages := []string{
 		// A simple patch series.
@@ -390,8 +402,7 @@ Message-ID: <First>
 From: UserA <a@user.com>
 Content-Type: text/plain
 
-
-Some text`,
+` + dummyPatch,
 		// A series with a cover.
 		`Date: Sun, 7 May 2017 19:55:00 -0700
 Subject: [PATCH net v2 00/02] A longer series
@@ -409,8 +420,7 @@ To: UserA <a@user.com>, UserB <b@user.com>
 Content-Type: text/plain
 In-Reply-To: <Second>
 
-
-Patch 1/2`,
+` + dummyPatch,
 		`Date: Sun, 7 May 2017 19:56:00 -0700
 Subject: [PATCH net v2 02/02] Second patch
 Message-ID: <Second-2>
@@ -419,16 +429,24 @@ To: UserA <a@user.com>, UserB <b@user.com>
 Content-Type: text/plain
 In-Reply-To: <Second>
 
-
-Patch 2/2`,
-		// Missing patches.
+` + dummyPatch,
+		// Some missing patches.
 		`Date: Sun, 7 May 2017 19:57:00 -0700
 Subject: [PATCH 01/03] Series
 Message-ID: <Third>
 From: Someone <a@b.com>
 Content-Type: text/plain
 
-Bug report`,
+` + dummyPatch,
+		// Reply with a patch subject.
+		`Date: Sun, 7 May 2017 19:57:00 -0700
+Subject: [PATCH] Series
+Message-ID: <Fourth>
+From: Someone <a@b.com>
+Content-Type: text/plain
+In-Reply-To: <Something>
+
+No patch, just text`,
 	}
 
 	var emails []*Email
@@ -441,7 +459,7 @@ Bug report`,
 	}
 
 	series := PatchSeries(emails)
-	assert.Len(t, series, 3)
+	assert.Len(t, series, 4)
 
 	expectPerID := map[string]*Series{
 		"<First>": {
@@ -480,6 +498,12 @@ Bug report`,
 				},
 			},
 		},
+		"<Fourth>": {
+			Subject:   "Series",
+			Version:   1,
+			Corrupted: "the subject mentions 1 patches, 0 are found",
+			Patches:   nil,
+		},
 	}
 	for _, s := range series {
 		expect := expectPerID[s.MessageID]
@@ -491,7 +515,7 @@ Bug report`,
 			assert.Equal(t, expect.Corrupted, s.Corrupted, "corrupted differs")
 			assert.Equal(t, expect.Subject, s.Subject, "subject differs")
 			assert.Equal(t, expect.Version, s.Version, "version differs")
-			assert.Len(t, s.Patches, len(expect.Patches), "patch count differs")
+			require.Len(t, s.Patches, len(expect.Patches), "patch count differs")
 			for i, expectPatch := range expect.Patches {
 				got := s.Patches[i]
 				assert.Equal(t, expectPatch.Seq, got.Seq, "seq differs")

pkg/email/lore/read.go

@@ -54,7 +54,7 @@ func emailFromRaw(body []byte, emails, domains []string) (*Email, error) {
 	if err != nil {
 		return nil, err
 	}
-	ret := &Email{Email: msg}
+	ret := &Email{Email: msg, HasPatch: msg.Patch != ""}
 	// Keep memory consumption low.
 	ret.Body = ""
 	ret.Patch = ""