[release-branch.go1.20] mime/multipart: limit memory/inode consumption of ReadForm

Reader.ReadForm is documented as storing "up to maxMemory bytes + 10MB"
in memory. Parsed forms can consume substantially more memory than
this limit, since ReadForm does not account for map entry overhead
and MIME headers.

In addition, while the amount of disk memory consumed by ReadForm can
be constrained by limiting the size of the parsed input, ReadForm will
create one temporary file per form part stored on disk, potentially
consuming a large number of inodes.

Update ReadForm's memory accounting to include part names,
MIME headers, and map entry overhead.

Update ReadForm to store all on-disk file parts in a single
temporary file.

Files returned by FileHeader.Open are documented as having a concrete
type of *os.File when a file is stored on disk. The change to use a
single temporary file for all parts means that this is no longer the
case when a form contains more than a single file part stored on disk.

The previous behavior of storing each file part in a separate disk
file may be reenabled with GODEBUG=multipartfiles=distinct.

Update Reader.NextPart and Reader.NextRawPart to set a 10MiB cap
on the size of MIME headers.

Thanks to Jakob Ackermann (@das7pad) for reporting this issue.

Updates #58006
Fixes #58363
Fixes CVE-2022-41725

Change-Id: Ibd780a6c4c83ac8bcfd3cbe344f042e9940f2eab
Reviewed-on: https://team-review.git.corp.google.com/c/golang/go-private/+/1714276
Reviewed-by: Julie Qiu <julieqiu@google.com>
TryBot-Result: Security TryBots <security-trybots@go-security-trybots.iam.gserviceaccount.com>
Reviewed-by: Roland Shoemaker <bracewell@google.com>
Run-TryBot: Damien Neil <dneil@google.com>
(cherry picked from commit 7d0da0029bfbe3228cc5216ced8c7b3184eb517d)
Reviewed-on: https://team-review.git.corp.google.com/c/golang/go-private/+/1728950
Reviewed-by: Damien Neil <dneil@google.com>
Run-TryBot: Roland Shoemaker <bracewell@google.com>
Reviewed-by: Tatiana Bradley <tatianabradley@google.com>
Reviewed-on: https://go-review.googlesource.com/c/go/+/468120
Auto-Submit: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>

6 files changed, 297 insertions, 38 deletions

diff --git a/src/mime/multipart/formdata.go b/src/mime/multipart/formdata.go
index fca5f9e15f..41bc886d16 100644
--- a/src/mime/multipart/formdata.go
+++ b/src/mime/multipart/formdata.go
@@ -7,6 +7,7 @@ package multipart
 import (
 	"bytes"
 	"errors"
+	"internal/godebug"
 	"io"
 	"math"
 	"net/textproto"
@@ -31,25 +32,61 @@ func (r *Reader) ReadForm(maxMemory int64) (*Form, error) {
 	return r.readForm(maxMemory)
 }
 
+var multipartFiles = godebug.New("multipartfiles")
+
 func (r *Reader) readForm(maxMemory int64) (_ *Form, err error) {
 	form := &Form{make(map[string][]string), make(map[string][]*FileHeader)}
+	var (
+		file    *os.File
+		fileOff int64
+	)
+	numDiskFiles := 0
+	combineFiles := multipartFiles.Value() != "distinct"
 	defer func() {
+		if file != nil {
+			if cerr := file.Close(); err == nil {
+				err = cerr
+			}
+		}
+		if combineFiles && numDiskFiles > 1 {
+			for _, fhs := range form.File {
+				for _, fh := range fhs {
+					fh.tmpshared = true
+				}
+			}
+		}
 		if err != nil {
 			form.RemoveAll()
+			if file != nil {
+				os.Remove(file.Name())
+			}
 		}
 	}()
 
-	// Reserve an additional 10 MB for non-file parts.
-	maxValueBytes := maxMemory + int64(10<<20)
-	if maxValueBytes <= 0 {
+	// maxFileMemoryBytes is the maximum bytes of file data we will store in memory.
+	// Data past this limit is written to disk.
+	// This limit strictly applies to content, not metadata (filenames, MIME headers, etc.),
+	// since metadata is always stored in memory, not disk.
+	//
+	// maxMemoryBytes is the maximum bytes we will store in memory, including file content,
+	// non-file part values, metdata, and map entry overhead.
+	//
+	// We reserve an additional 10 MB in maxMemoryBytes for non-file data.
+	//
+	// The relationship between these parameters, as well as the overly-large and
+	// unconfigurable 10 MB added on to maxMemory, is unfortunate but difficult to change
+	// within the constraints of the API as documented.
+	maxFileMemoryBytes := maxMemory
+	maxMemoryBytes := maxMemory + int64(10<<20)
+	if maxMemoryBytes <= 0 {
 		if maxMemory < 0 {
-			maxValueBytes = 0
+			maxMemoryBytes = 0
 		} else {
-			maxValueBytes = math.MaxInt64
+			maxMemoryBytes = math.MaxInt64
 		}
 	}
 	for {
-		p, err := r.NextPart()
+		p, err := r.nextPart(false, maxMemoryBytes)
 		if err == io.EOF {
 			break
 		}
@@ -63,16 +100,27 @@ func (r *Reader) readForm(maxMemory int64) (_ *Form, err error) {
 		}
 		filename := p.FileName()
 
+		// Multiple values for the same key (one map entry, longer slice) are cheaper
+		// than the same number of values for different keys (many map entries), but
+		// using a consistent per-value cost for overhead is simpler.
+		maxMemoryBytes -= int64(len(name))
+		maxMemoryBytes -= 100 // map overhead
+		if maxMemoryBytes < 0 {
+			// We can't actually take this path, since nextPart would already have
+			// rejected the MIME headers for being too large. Check anyway.
+			return nil, ErrMessageTooLarge
+		}
+
 		var b bytes.Buffer
 
 		if filename == "" {
 			// value, store as string in memory
-			n, err := io.CopyN(&b, p, maxValueBytes+1)
+			n, err := io.CopyN(&b, p, maxMemoryBytes+1)
 			if err != nil && err != io.EOF {
 				return nil, err
 			}
-			maxValueBytes -= n
-			if maxValueBytes < 0 {
+			maxMemoryBytes -= n
+			if maxMemoryBytes < 0 {
 				return nil, ErrMessageTooLarge
 			}
 			form.Value[name] = append(form.Value[name], b.String())
@@ -80,35 +128,45 @@ func (r *Reader) readForm(maxMemory int64) (_ *Form, err error) {
 		}
 
 		// file, store in memory or on disk
+		maxMemoryBytes -= mimeHeaderSize(p.Header)
+		if maxMemoryBytes < 0 {
+			return nil, ErrMessageTooLarge
+		}
 		fh := &FileHeader{
 			Filename: filename,
 			Header:   p.Header,
 		}
-		n, err := io.CopyN(&b, p, maxMemory+1)
+		n, err := io.CopyN(&b, p, maxFileMemoryBytes+1)
 		if err != nil && err != io.EOF {
 			return nil, err
 		}
-		if n > maxMemory {
-			// too big, write to disk and flush buffer
-			file, err := os.CreateTemp("", "multipart-")
-			if err != nil {
-				return nil, err
+		if n > maxFileMemoryBytes {
+			if file == nil {
+				file, err = os.CreateTemp(r.tempDir, "multipart-")
+				if err != nil {
+					return nil, err
+				}
 			}
+			numDiskFiles++
 			size, err := io.Copy(file, io.MultiReader(&b, p))
-			if cerr := file.Close(); err == nil {
-				err = cerr
-			}
 			if err != nil {
-				os.Remove(file.Name())
 				return nil, err
 			}
 			fh.tmpfile = file.Name()
 			fh.Size = size
+			fh.tmpoff = fileOff
+			fileOff += size
+			if !combineFiles {
+				if err := file.Close(); err != nil {
+					return nil, err
+				}
+				file = nil
+			}
 		} else {
 			fh.content = b.Bytes()
 			fh.Size = int64(len(fh.content))
-			maxMemory -= n
-			maxValueBytes -= n
+			maxFileMemoryBytes -= n
+			maxMemoryBytes -= n
 		}
 		form.File[name] = append(form.File[name], fh)
 	}
@@ -116,6 +174,17 @@ func (r *Reader) readForm(maxMemory int64) (_ *Form, err error) {
 	return form, nil
 }
 
+func mimeHeaderSize(h textproto.MIMEHeader) (size int64) {
+	for k, vs := range h {
+		size += int64(len(k))
+		size += 100 // map entry overhead
+		for _, v := range vs {
+			size += int64(len(v))
+		}
+	}
+	return size
+}
+
 // Form is a parsed multipart form.
 // Its File parts are stored either in memory or on disk,
 // and are accessible via the *FileHeader's Open method.
@@ -133,7 +202,7 @@ func (f *Form) RemoveAll() error {
 		for _, fh := range fhs {
 			if fh.tmpfile != "" {
 				e := os.Remove(fh.tmpfile)
-				if e != nil && err == nil {
+				if e != nil && !errors.Is(e, os.ErrNotExist) && err == nil {
 					err = e
 				}
 			}
@@ -148,15 +217,25 @@ type FileHeader struct {
 	Header   textproto.MIMEHeader
 	Size     int64
 
-	content []byte
-	tmpfile string
+	content   []byte
+	tmpfile   string
+	tmpoff    int64
+	tmpshared bool
 }
 
 // Open opens and returns the FileHeader's associated File.
 func (fh *FileHeader) Open() (File, error) {
 	if b := fh.content; b != nil {
 		r := io.NewSectionReader(bytes.NewReader(b), 0, int64(len(b)))
-		return sectionReadCloser{r}, nil
+		return sectionReadCloser{r, nil}, nil
+	}
+	if fh.tmpshared {
+		f, err := os.Open(fh.tmpfile)
+		if err != nil {
+			return nil, err
+		}
+		r := io.NewSectionReader(f, fh.tmpoff, fh.Size)
+		return sectionReadCloser{r, f}, nil
 	}
 	return os.Open(fh.tmpfile)
 }
@@ -175,8 +254,12 @@ type File interface {
 
 type sectionReadCloser struct {
 	*io.SectionReader
+	io.Closer
 }
 
 func (rc sectionReadCloser) Close() error {
+	if rc.Closer != nil {
+		return rc.Closer.Close()
+	}
 	return nil
 }
diff --git a/src/mime/multipart/formdata_test.go b/src/mime/multipart/formdata_test.go
index 8a4eabcee0..8a862be717 100644
--- a/src/mime/multipart/formdata_test.go
+++ b/src/mime/multipart/formdata_test.go
@@ -5,8 +5,11 @@
 package multipart
 
 import (
+	"bytes"
+	"fmt"
 	"io"
 	"math"
+	"net/textproto"
 	"os"
 	"strings"
 	"testing"
@@ -207,8 +210,8 @@ Content-Disposition: form-data; name="largetext"
 		maxMemory int64
 		err       error
 	}{
-		{"smaller", 50, nil},
-		{"exact-fit", 25, nil},
+		{"smaller", 50 + int64(len("largetext")) + 100, nil},
+		{"exact-fit", 25 + int64(len("largetext")) + 100, nil},
 		{"too-large", 0, ErrMessageTooLarge},
 	}
 	for _, tc := range testCases {
@@ -223,7 +226,7 @@ Content-Disposition: form-data; name="largetext"
 				defer f.RemoveAll()
 			}
 			if tc.err != err {
-				t.Fatalf("ReadForm error - got: %v; expected: %v", tc.err, err)
+				t.Fatalf("ReadForm error - got: %v; expected: %v", err, tc.err)
 			}
 			if err == nil {
 				if g := f.Value["largetext"][0]; g != largeTextValue {
@@ -233,3 +236,135 @@ Content-Disposition: form-data; name="largetext"
 		})
 	}
 }
+
+// TestReadForm_MetadataTooLarge verifies that we account for the size of field names,
+// MIME headers, and map entry overhead while limiting the memory consumption of parsed forms.
+func TestReadForm_MetadataTooLarge(t *testing.T) {
+	for _, test := range []struct {
+		name string
+		f    func(*Writer)
+	}{{
+		name: "large name",
+		f: func(fw *Writer) {
+			name := strings.Repeat("a", 10<<20)
+			w, _ := fw.CreateFormField(name)
+			w.Write([]byte("value"))
+		},
+	}, {
+		name: "large MIME header",
+		f: func(fw *Writer) {
+			h := make(textproto.MIMEHeader)
+			h.Set("Content-Disposition", `form-data; name="a"`)
+			h.Set("X-Foo", strings.Repeat("a", 10<<20))
+			w, _ := fw.CreatePart(h)
+			w.Write([]byte("value"))
+		},
+	}, {
+		name: "many parts",
+		f: func(fw *Writer) {
+			for i := 0; i < 110000; i++ {
+				w, _ := fw.CreateFormField("f")
+				w.Write([]byte("v"))
+			}
+		},
+	}} {
+		t.Run(test.name, func(t *testing.T) {
+			var buf bytes.Buffer
+			fw := NewWriter(&buf)
+			test.f(fw)
+			if err := fw.Close(); err != nil {
+				t.Fatal(err)
+			}
+			fr := NewReader(&buf, fw.Boundary())
+			_, err := fr.ReadForm(0)
+			if err != ErrMessageTooLarge {
+				t.Errorf("fr.ReadForm() = %v, want ErrMessageTooLarge", err)
+			}
+		})
+	}
+}
+
+// TestReadForm_ManyFiles_Combined tests that a multipart form containing many files only
+// results in a single on-disk file.
+func TestReadForm_ManyFiles_Combined(t *testing.T) {
+	const distinct = false
+	testReadFormManyFiles(t, distinct)
+}
+
+// TestReadForm_ManyFiles_Distinct tests that setting GODEBUG=multipartfiles=distinct
+// results in every file in a multipart form being placed in a distinct on-disk file.
+func TestReadForm_ManyFiles_Distinct(t *testing.T) {
+	t.Setenv("GODEBUG", "multipartfiles=distinct")
+	const distinct = true
+	testReadFormManyFiles(t, distinct)
+}
+
+func testReadFormManyFiles(t *testing.T, distinct bool) {
+	var buf bytes.Buffer
+	fw := NewWriter(&buf)
+	const numFiles = 10
+	for i := 0; i < numFiles; i++ {
+		name := fmt.Sprint(i)
+		w, err := fw.CreateFormFile(name, name)
+		if err != nil {
+			t.Fatal(err)
+		}
+		w.Write([]byte(name))
+	}
+	if err := fw.Close(); err != nil {
+		t.Fatal(err)
+	}
+	fr := NewReader(&buf, fw.Boundary())
+	fr.tempDir = t.TempDir()
+	form, err := fr.ReadForm(0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for i := 0; i < numFiles; i++ {
+		name := fmt.Sprint(i)
+		if got := len(form.File[name]); got != 1 {
+			t.Fatalf("form.File[%q] has %v entries, want 1", name, got)
+		}
+		fh := form.File[name][0]
+		file, err := fh.Open()
+		if err != nil {
+			t.Fatalf("form.File[%q].Open() = %v", name, err)
+		}
+		if distinct {
+			if _, ok := file.(*os.File); !ok {
+				t.Fatalf("form.File[%q].Open: %T, want *os.File", name, file)
+			}
+		}
+		got, err := io.ReadAll(file)
+		file.Close()
+		if string(got) != name || err != nil {
+			t.Fatalf("read form.File[%q]: %q, %v; want %q, nil", name, string(got), err, name)
+		}
+	}
+	dir, err := os.Open(fr.tempDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer dir.Close()
+	names, err := dir.Readdirnames(0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	wantNames := 1
+	if distinct {
+		wantNames = numFiles
+	}
+	if len(names) != wantNames {
+		t.Fatalf("temp dir contains %v files; want 1", len(names))
+	}
+	if err := form.RemoveAll(); err != nil {
+		t.Fatalf("form.RemoveAll() = %v", err)
+	}
+	names, err = dir.Readdirnames(0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(names) != 0 {
+		t.Fatalf("temp dir contains %v files; want 0", len(names))
+	}
+}
diff --git a/src/mime/multipart/multipart.go b/src/mime/multipart/multipart.go
index b3a904f0af..86ea926346 100644
--- a/src/mime/multipart/multipart.go
+++ b/src/mime/multipart/multipart.go
@@ -128,12 +128,12 @@ func (r *stickyErrorReader) Read(p []byte) (n int, _ error) {
 	return n, r.err
 }
 
-func newPart(mr *Reader, rawPart bool) (*Part, error) {
+func newPart(mr *Reader, rawPart bool, maxMIMEHeaderSize int64) (*Part, error) {
 	bp := &Part{
 		Header: make(map[string][]string),
 		mr:     mr,
 	}
-	if err := bp.populateHeaders(); err != nil {
+	if err := bp.populateHeaders(maxMIMEHeaderSize); err != nil {
 		return nil, err
 	}
 	bp.r = partReader{bp}
@@ -149,12 +149,16 @@ func newPart(mr *Reader, rawPart bool) (*Part, error) {
 	return bp, nil
 }
 
-func (p *Part) populateHeaders() error {
+func (p *Part) populateHeaders(maxMIMEHeaderSize int64) error {
 	r := textproto.NewReader(p.mr.bufReader)
-	header, err := r.ReadMIMEHeader()
+	header, err := readMIMEHeader(r, maxMIMEHeaderSize)
 	if err == nil {
 		p.Header = header
 	}
+	// TODO: Add a distinguishable error to net/textproto.
+	if err != nil && err.Error() == "message too large" {
+		err = ErrMessageTooLarge
+	}
 	return err
 }
 
@@ -311,6 +315,7 @@ func (p *Part) Close() error {
 // isn't supported.
 type Reader struct {
 	bufReader *bufio.Reader
+	tempDir   string // used in tests
 
 	currentPart *Part
 	partsRead   int
@@ -321,6 +326,10 @@ type Reader struct {
 	dashBoundary     []byte // "--boundary"
 }
 
+// maxMIMEHeaderSize is the maximum size of a MIME header we will parse,
+// including header keys, values, and map overhead.
+const maxMIMEHeaderSize = 10 << 20
+
 // NextPart returns the next part in the multipart or an error.
 // When there are no more parts, the error io.EOF is returned.
 //
@@ -328,7 +337,7 @@ type Reader struct {
 // has a value of "quoted-printable", that header is instead
 // hidden and the body is transparently decoded during Read calls.
 func (r *Reader) NextPart() (*Part, error) {
-	return r.nextPart(false)
+	return r.nextPart(false, maxMIMEHeaderSize)
 }
 
 // NextRawPart returns the next part in the multipart or an error.
@@ -337,10 +346,10 @@ func (r *Reader) NextPart() (*Part, error) {
 // Unlike NextPart, it does not have special handling for
 // "Content-Transfer-Encoding: quoted-printable".
 func (r *Reader) NextRawPart() (*Part, error) {
-	return r.nextPart(true)
+	return r.nextPart(true, maxMIMEHeaderSize)
 }
 
-func (r *Reader) nextPart(rawPart bool) (*Part, error) {
+func (r *Reader) nextPart(rawPart bool, maxMIMEHeaderSize int64) (*Part, error) {
 	if r.currentPart != nil {
 		r.currentPart.Close()
 	}
@@ -365,7 +374,7 @@ func (r *Reader) nextPart(rawPart bool) (*Part, error) {
 
 		if r.isBoundaryDelimiterLine(line) {
 			r.partsRead++
-			bp, err := newPart(r, rawPart)
+			bp, err := newPart(r, rawPart, maxMIMEHeaderSize)
 			if err != nil {
 				return nil, err
 			}
diff --git a/src/mime/multipart/readmimeheader.go b/src/mime/multipart/readmimeheader.go
new file mode 100644
index 0000000000..6836928c9e
--- /dev/null
+++ b/src/mime/multipart/readmimeheader.go
@@ -0,0 +1,14 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+package multipart
+
+import (
+	"net/textproto"
+	_ "unsafe" // for go:linkname
+)
+
+// readMIMEHeader is defined in package net/textproto.
+//
+//go:linkname readMIMEHeader net/textproto.readMIMEHeader
+func readMIMEHeader(r *textproto.Reader, lim int64) (textproto.MIMEHeader, error)
diff --git a/src/net/http/request_test.go b/src/net/http/request_test.go
index 686a8699fb..23e49d6b8e 100644
--- a/src/net/http/request_test.go
+++ b/src/net/http/request_test.go
@@ -1097,7 +1097,7 @@ func testMissingFile(t *testing.T, req *Request) {
 		t.Errorf("FormFile file = %v, want nil", f)
 	}
 	if fh != nil {
-		t.Errorf("FormFile file header = %q, want nil", fh)
+		t.Errorf("FormFile file header = %v, want nil", fh)
 	}
 	if err != ErrMissingFile {
 		t.Errorf("FormFile err = %q, want ErrMissingFile", err)
diff --git a/src/net/textproto/reader.go b/src/net/textproto/reader.go
index 4e4999b3c9..8e800088c1 100644
--- a/src/net/textproto/reader.go
+++ b/src/net/textproto/reader.go
@@ -7,8 +7,10 @@ package textproto
 import (
 	"bufio"
 	"bytes"
+	"errors"
 	"fmt"
 	"io"
+	"math"
 	"strconv"
 	"strings"
 	"sync"
@@ -477,6 +479,12 @@ var colon = []byte(":")
 //		"Long-Key": {"Even Longer Value"},
 //	}
 func (r *Reader) ReadMIMEHeader() (MIMEHeader, error) {
+	return readMIMEHeader(r, math.MaxInt64)
+}
+
+// readMIMEHeader is a version of ReadMIMEHeader which takes a limit on the header size.
+// It is called by the mime/multipart package.
+func readMIMEHeader(r *Reader, lim int64) (MIMEHeader, error) {
 	// Avoid lots of small slice allocations later by allocating one
 	// large one ahead of time which we'll cut up into smaller
 	// slices. If this isn't big enough later, we allocate small ones.
@@ -526,9 +534,19 @@ func (r *Reader) ReadMIMEHeader() (MIMEHeader, error) {
 		}
 
 		// Skip initial spaces in value.
-		value := strings.TrimLeft(string(v), " \t")
+		value := string(bytes.TrimLeft(v, " \t"))
 
 		vv := m[key]
+		if vv == nil {
+			lim -= int64(len(key))
+			lim -= 100 // map entry overhead
+		}
+		lim -= int64(len(value))
+		if lim < 0 {
+			// TODO: This should be a distinguishable error (ErrMessageTooLarge)
+			// to allow mime/multipart to detect it.
+			return m, errors.New("message too large")
+		}
 		if vv == nil && len(strs) > 0 {
 			// More than likely this will be a single-element key.
 			// Most headers aren't multi-valued.