fix(cpe): bound CPE candidate generation for pathological package metadata (#5356)

* fix(cpe): cap sub-selection prefixes to bound candidate cross-product

a name with thousands of `-`/`_` separators produced one vendor and product candidate per segment, and the vendor x product cross-product made CPE generation roughly cubic in time and memory (8KB name: ~250s, 4.8GB RSS). `generateSubSelections` now stops after 10 prefixes. The full name is still a candidate, so real package names are unaffected.

Signed-off-by: Alex Goodman <wagoodman@users.noreply.github.com>

* fix(cpe): bound candidate count and field length before the cross-product

the prefix cap only covered one way to inflate the vendor x product cross-product in `FromPackageAttributes`. Java groupIDs add a vendor (and, with no artifactID, a product) per dot segment, and .NET executables add one of each per file, so a 1000-segment `Bundle-SymbolicName` still produced ~1M CPEs and several GB.

- vendor and product candidates are each capped at 64 (shortest first, so truncation is deterministic)
- names, versions, and candidates over 512 bytes are skipped, since every byte is paid once per CPE in key building and regex validation

real packages produce around a dozen candidates with names well under 100 bytes, so output for real inputs is unchanged. Also corrects the earlier comment: prefixes only become vendor candidates and the cost was quadratic, not cubic.

Signed-off-by: Alex Goodman <wagoodman@users.noreply.github.com>

---------

Signed-off-by: Alex Goodman <wagoodman@users.noreply.github.com>

syft/pkg/cataloger/internal/cpegenerate/generate.go

@@ -3,10 +3,12 @@ package cpegenerate
 import (
 	"bufio"
 	"bytes"
+	"cmp"
 	_ "embed"
 	"encoding/json"
 	"fmt"
 	"regexp"
+	"slices"
 	"sort"
 	"strings"
 	"sync"
@@ -145,8 +147,14 @@ func FromPackageAttributes(p pkg.Package) []cpe.CPE {
 		return nil
 	}
 
-	vendors := candidateVendors(p)
-	products := candidateProducts(p)
+	// real CPE fields are short, and every byte of a field is paid once per generated CPE (key building and regex
+	// validation), so a hostile name or version is not worth generating candidates for at all.
+	if len(p.Name) > maxCandidateFieldLength || len(p.Version) > maxCandidateFieldLength {
+		return nil
+	}
+
+	vendors := boundCandidates(candidateVendors(p))
+	products := boundCandidates(candidateProducts(p))
 	targetSWs := candidateTargetSw(p)
 	if len(products) == 0 {
 		return nil
@@ -183,6 +191,32 @@ func FromPackageAttributes(p pkg.Package) []cpe.CPE {
 	return result
 }
 
+// maxCandidateFieldLength bounds the name, version, and each vendor and product candidate that reaches the
+// cross-product in FromPackageAttributes. Real values are well under this.
+const maxCandidateFieldLength = 512
+
+// maxCandidates bounds how many vendor and product candidates each feed the cross-product in FromPackageAttributes,
+// so a single package yields at most maxCandidates² CPEs. Some sources add a candidate per input element with no bound
+// of their own (e.g. one vendor per java groupID segment, one product per .NET executable), and real packages produce
+// a dozen or so candidates at most.
+const maxCandidates = 64
+
+// boundCandidates drops candidates longer than maxCandidateFieldLength and keeps at most maxCandidates of the rest.
+// Candidates are ordered shortest first (then lexically) so that truncation is deterministic and keeps the short,
+// generic values that tend to be real vendors and products.
+func boundCandidates(candidates []string) []string {
+	candidates = slices.DeleteFunc(candidates, func(c string) bool {
+		return len(c) > maxCandidateFieldLength
+	})
+	if len(candidates) <= maxCandidates {
+		return candidates
+	}
+	slices.SortFunc(candidates, func(a, b string) int {
+		return cmp.Or(cmp.Compare(len(a), len(b)), strings.Compare(a, b))
+	})
+	return candidates[:maxCandidates]
+}
+
 func candidateTargetSw(p pkg.Package) []string {
 	switch p.Type {
 	case pkg.WordpressPluginPkg:
@@ -359,13 +393,19 @@ func addAllSubSelections(fields fieldCandidateSet) {
 	}
 }
 
+// maxSubSelections bounds how many prefixes generateSubSelections returns. Each prefix becomes a vendor candidate, so
+// without a bound a name with N separators yields O(N) vendors of O(N) length each (quadratic in time and memory).
+// Real package names rarely exceed a handful of segments.
+const maxSubSelections = 10
+
 // generateSubSelections attempts to split a field by hyphens and underscores and return a list of sensible sub-selections
-// that can be used as product or vendor candidates. E.g. jenkins-ci-tools -> [jenkins-ci-tools, jenkins-ci, jenkins].
+// that can be used as product or vendor candidates. E.g. jenkins-ci-tools -> [jenkins, jenkins-ci, jenkins-ci-tools].
+// At most maxSubSelections prefixes are returned, shortest first.
 func generateSubSelections(field string) (results []string) {
 	scanner := bufio.NewScanner(strings.NewReader(field))
 	scanner.Split(scanByHyphenOrUnderscore)
 	var lastToken uint8
-	for scanner.Scan() {
+	for len(results) < maxSubSelections && scanner.Scan() {
 		rawCandidate := scanner.Text()
 		if len(rawCandidate) == 0 {
 			break

syft/pkg/cataloger/internal/cpegenerate/generate_test.go

@@ -1121,6 +1121,11 @@ func Test_generateSubSelections(t *testing.T) {
 			field:    "_",
 			expected: nil,
 		},
+		{
+			// capped at maxSubSelections, keeping the shortest prefixes (the full name is not returned)
+			field:    "a_b_c_d_e_f_g_h_i_j_k_l",
+			expected: []string{"a", "a_b", "a_b_c", "a_b_c_d", "a_b_c_d_e", "a_b_c_d_e_f", "a_b_c_d_e_f_g", "a_b_c_d_e_f_g_h", "a_b_c_d_e_f_g_h_i", "a_b_c_d_e_f_g_h_i_j"},
+		},
 	}
 	for _, test := range tests {
 		t.Run(test.field, func(t *testing.T) {
@@ -1265,3 +1270,59 @@ func TestAddBinaryPackageDigitVariations(t *testing.T) {
 		})
 	}
 }
+
+func TestFromPackageAttributes_boundsSubSelectionBlowup(t *testing.T) {
+	// a name with many separators used to produce a vendor candidate per segment, each O(name length), which was
+	// quadratic in time and memory.
+	name := strings.Repeat("a_", 200) + "a"
+	p := pkg.Package{Name: name, Version: "1.0", Type: pkg.PythonPkg, Language: pkg.Python}
+
+	cpes := FromPackageAttributes(p)
+	assert.NotEmpty(t, cpes)
+	assert.LessOrEqual(t, len(cpes), 1000)
+}
+
+func TestFromPackageAttributes_boundsCandidateCrossProduct(t *testing.T) {
+	// every groupID segment becomes a vendor, and with no artifactID every segment past the second also becomes a
+	// product, so the vendor x product cross-product is quadratic in segment count without a bound.
+	var segments []string
+	for i := range 1000 {
+		segments = append(segments, fmt.Sprintf("s%d", i))
+	}
+	p := pkg.Package{
+		Name:     "thing",
+		Version:  "1.0",
+		Type:     pkg.JavaPkg,
+		Language: pkg.Java,
+		Metadata: pkg.JavaArchive{
+			Manifest: &pkg.JavaManifest{
+				Main: pkg.KeyValues{{Key: "Bundle-SymbolicName", Value: "com.x." + strings.Join(segments, ".")}},
+			},
+		},
+	}
+
+	cpes := FromPackageAttributes(p)
+	assert.NotEmpty(t, cpes)
+	assert.LessOrEqual(t, len(cpes), maxCandidates*maxCandidates)
+}
+
+func TestFromPackageAttributes_skipsOversizedFields(t *testing.T) {
+	long := strings.Repeat("a", maxCandidateFieldLength+1)
+
+	assert.Empty(t, FromPackageAttributes(pkg.Package{Name: long, Version: "1.0", Type: pkg.PythonPkg}))
+	assert.Empty(t, FromPackageAttributes(pkg.Package{Name: "thing", Version: long, Type: pkg.PythonPkg}))
+	assert.NotEmpty(t, FromPackageAttributes(pkg.Package{Name: "thing", Version: "1.0", Type: pkg.PythonPkg}))
+}
+
+func Test_boundCandidates(t *testing.T) {
+	long := strings.Repeat("a", maxCandidateFieldLength+1)
+	assert.Equal(t, []string{"b", "a"}, boundCandidates([]string{"b", long, "a"}), "under the cap order is untouched")
+
+	var many []string
+	for i := range maxCandidates + 10 {
+		many = append(many, strings.Repeat("x", maxCandidates+10-i))
+	}
+	got := boundCandidates(many)
+	assert.Len(t, got, maxCandidates)
+	assert.Equal(t, "x", got[0], "shortest candidates are kept")
+}