githaven/modules/git/batch_reader.go
zeripath 0dcc74a8a7
Prevent dangling cat-file calls (goroutine alternative) (#19454)
If an `os/exec.Command` is passed non `*os.File` as an input/output, go
will create `os.Pipe`s and wait for their closure in `cmd.Wait()`.  If
the code following this is responsible for closing `io.Pipe`s or other
handlers then on process death from context cancellation the `Wait` can
hang.

There are two possible solutions:

1. use `os.Pipe` as the input/output as `cmd.Wait` does not wait for these.
2. create a goroutine waiting on the context cancellation that will close the inputs.

This PR provides the second option - which is a simpler change that can
be more easily backported.

Closes #19448

Signed-off-by: Andrew Thornton <art27@cantab.net>
2022-04-22 17:20:04 +02:00

349 lines
9.2 KiB
Go

// Copyright 2020 The Gitea Authors. All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.
package git
import (
"bufio"
"bytes"
"context"
"fmt"
"io"
"math"
"runtime"
"strconv"
"strings"
"code.gitea.io/gitea/modules/log"
"github.com/djherbis/buffer"
"github.com/djherbis/nio/v3"
)
// WriteCloserError wraps an io.WriteCloser with an additional CloseWithError function
type WriteCloserError interface {
io.WriteCloser
CloseWithError(err error) error
}
// EnsureValidGitRepository runs git rev-parse in the repository path - thus ensuring that the repository is a valid repository.
// Run before opening git cat-file.
// This is needed otherwise the git cat-file will hang for invalid repositories.
func EnsureValidGitRepository(ctx context.Context, repoPath string) error {
stderr := strings.Builder{}
err := NewCommand(ctx, "rev-parse").
SetDescription(fmt.Sprintf("%s rev-parse [repo_path: %s]", GitExecutable, repoPath)).
Run(&RunOpts{
Dir: repoPath,
Stderr: &stderr,
})
if err != nil {
return ConcatenateError(err, (&stderr).String())
}
return nil
}
// CatFileBatchCheck opens git cat-file --batch-check in the provided repo and returns a stdin pipe, a stdout reader and cancel function
func CatFileBatchCheck(ctx context.Context, repoPath string) (WriteCloserError, *bufio.Reader, func()) {
batchStdinReader, batchStdinWriter := io.Pipe()
batchStdoutReader, batchStdoutWriter := io.Pipe()
ctx, ctxCancel := context.WithCancel(ctx)
closed := make(chan struct{})
cancel := func() {
ctxCancel()
_ = batchStdoutReader.Close()
_ = batchStdinWriter.Close()
<-closed
}
// Ensure cancel is called as soon as the provided context is cancelled
go func() {
<-ctx.Done()
cancel()
}()
_, filename, line, _ := runtime.Caller(2)
filename = strings.TrimPrefix(filename, callerPrefix)
go func() {
stderr := strings.Builder{}
err := NewCommand(ctx, "cat-file", "--batch-check").
SetDescription(fmt.Sprintf("%s cat-file --batch-check [repo_path: %s] (%s:%d)", GitExecutable, repoPath, filename, line)).
Run(&RunOpts{
Dir: repoPath,
Stdin: batchStdinReader,
Stdout: batchStdoutWriter,
Stderr: &stderr,
})
if err != nil {
_ = batchStdoutWriter.CloseWithError(ConcatenateError(err, (&stderr).String()))
_ = batchStdinReader.CloseWithError(ConcatenateError(err, (&stderr).String()))
} else {
_ = batchStdoutWriter.Close()
_ = batchStdinReader.Close()
}
close(closed)
}()
// For simplicities sake we'll use a buffered reader to read from the cat-file --batch-check
batchReader := bufio.NewReader(batchStdoutReader)
return batchStdinWriter, batchReader, cancel
}
// CatFileBatch opens git cat-file --batch in the provided repo and returns a stdin pipe, a stdout reader and cancel function
func CatFileBatch(ctx context.Context, repoPath string) (WriteCloserError, *bufio.Reader, func()) {
// We often want to feed the commits in order into cat-file --batch, followed by their trees and sub trees as necessary.
// so let's create a batch stdin and stdout
batchStdinReader, batchStdinWriter := io.Pipe()
batchStdoutReader, batchStdoutWriter := nio.Pipe(buffer.New(32 * 1024))
ctx, ctxCancel := context.WithCancel(ctx)
closed := make(chan struct{})
cancel := func() {
ctxCancel()
_ = batchStdinWriter.Close()
_ = batchStdoutReader.Close()
<-closed
}
// Ensure cancel is called as soon as the provided context is cancelled
go func() {
<-ctx.Done()
cancel()
}()
_, filename, line, _ := runtime.Caller(2)
filename = strings.TrimPrefix(filename, callerPrefix)
go func() {
stderr := strings.Builder{}
err := NewCommand(ctx, "cat-file", "--batch").
SetDescription(fmt.Sprintf("%s cat-file --batch [repo_path: %s] (%s:%d)", GitExecutable, repoPath, filename, line)).
Run(&RunOpts{
Dir: repoPath,
Stdin: batchStdinReader,
Stdout: batchStdoutWriter,
Stderr: &stderr,
})
if err != nil {
_ = batchStdoutWriter.CloseWithError(ConcatenateError(err, (&stderr).String()))
_ = batchStdinReader.CloseWithError(ConcatenateError(err, (&stderr).String()))
} else {
_ = batchStdoutWriter.Close()
_ = batchStdinReader.Close()
}
close(closed)
}()
// For simplicities sake we'll us a buffered reader to read from the cat-file --batch
batchReader := bufio.NewReaderSize(batchStdoutReader, 32*1024)
return batchStdinWriter, batchReader, cancel
}
// ReadBatchLine reads the header line from cat-file --batch
// We expect:
// <sha> SP <type> SP <size> LF
// sha is a 40byte not 20byte here
func ReadBatchLine(rd *bufio.Reader) (sha []byte, typ string, size int64, err error) {
typ, err = rd.ReadString('\n')
if err != nil {
return
}
if len(typ) == 1 {
typ, err = rd.ReadString('\n')
if err != nil {
return
}
}
idx := strings.IndexByte(typ, ' ')
if idx < 0 {
log.Debug("missing space typ: %s", typ)
err = ErrNotExist{ID: string(sha)}
return
}
sha = []byte(typ[:idx])
typ = typ[idx+1:]
idx = strings.IndexByte(typ, ' ')
if idx < 0 {
err = ErrNotExist{ID: string(sha)}
return
}
sizeStr := typ[idx+1 : len(typ)-1]
typ = typ[:idx]
size, err = strconv.ParseInt(sizeStr, 10, 64)
return
}
// ReadTagObjectID reads a tag object ID hash from a cat-file --batch stream, throwing away the rest of the stream.
func ReadTagObjectID(rd *bufio.Reader, size int64) (string, error) {
id := ""
var n int64
headerLoop:
for {
line, err := rd.ReadBytes('\n')
if err != nil {
return "", err
}
n += int64(len(line))
idx := bytes.Index(line, []byte{' '})
if idx < 0 {
continue
}
if string(line[:idx]) == "object" {
id = string(line[idx+1 : len(line)-1])
break headerLoop
}
}
// Discard the rest of the tag
discard := size - n + 1
for discard > math.MaxInt32 {
_, err := rd.Discard(math.MaxInt32)
if err != nil {
return id, err
}
discard -= math.MaxInt32
}
_, err := rd.Discard(int(discard))
return id, err
}
// ReadTreeID reads a tree ID from a cat-file --batch stream, throwing away the rest of the stream.
func ReadTreeID(rd *bufio.Reader, size int64) (string, error) {
id := ""
var n int64
headerLoop:
for {
line, err := rd.ReadBytes('\n')
if err != nil {
return "", err
}
n += int64(len(line))
idx := bytes.Index(line, []byte{' '})
if idx < 0 {
continue
}
if string(line[:idx]) == "tree" {
id = string(line[idx+1 : len(line)-1])
break headerLoop
}
}
// Discard the rest of the commit
discard := size - n + 1
for discard > math.MaxInt32 {
_, err := rd.Discard(math.MaxInt32)
if err != nil {
return id, err
}
discard -= math.MaxInt32
}
_, err := rd.Discard(int(discard))
return id, err
}
// git tree files are a list:
// <mode-in-ascii> SP <fname> NUL <20-byte SHA>
//
// Unfortunately this 20-byte notation is somewhat in conflict to all other git tools
// Therefore we need some method to convert these 20-byte SHAs to a 40-byte SHA
// constant hextable to help quickly convert between 20byte and 40byte hashes
const hextable = "0123456789abcdef"
// To40ByteSHA converts a 20-byte SHA into a 40-byte sha. Input and output can be the
// same 40 byte slice to support in place conversion without allocations.
// This is at least 100x quicker that hex.EncodeToString
// NB This requires that out is a 40-byte slice
func To40ByteSHA(sha, out []byte) []byte {
for i := 19; i >= 0; i-- {
v := sha[i]
vhi, vlo := v>>4, v&0x0f
shi, slo := hextable[vhi], hextable[vlo]
out[i*2], out[i*2+1] = shi, slo
}
return out
}
// ParseTreeLine reads an entry from a tree in a cat-file --batch stream
// This carefully avoids allocations - except where fnameBuf is too small.
// It is recommended therefore to pass in an fnameBuf large enough to avoid almost all allocations
//
// Each line is composed of:
// <mode-in-ascii-dropping-initial-zeros> SP <fname> NUL <20-byte SHA>
//
// We don't attempt to convert the 20-byte SHA to 40-byte SHA to save a lot of time
func ParseTreeLine(rd *bufio.Reader, modeBuf, fnameBuf, shaBuf []byte) (mode, fname, sha []byte, n int, err error) {
var readBytes []byte
// Read the Mode & fname
readBytes, err = rd.ReadSlice('\x00')
if err != nil {
return
}
idx := bytes.IndexByte(readBytes, ' ')
if idx < 0 {
log.Debug("missing space in readBytes ParseTreeLine: %s", readBytes)
err = &ErrNotExist{}
return
}
n += idx + 1
copy(modeBuf, readBytes[:idx])
if len(modeBuf) >= idx {
modeBuf = modeBuf[:idx]
} else {
modeBuf = append(modeBuf, readBytes[len(modeBuf):idx]...)
}
mode = modeBuf
readBytes = readBytes[idx+1:]
// Deal with the fname
copy(fnameBuf, readBytes)
if len(fnameBuf) > len(readBytes) {
fnameBuf = fnameBuf[:len(readBytes)]
} else {
fnameBuf = append(fnameBuf, readBytes[len(fnameBuf):]...)
}
for err == bufio.ErrBufferFull {
readBytes, err = rd.ReadSlice('\x00')
fnameBuf = append(fnameBuf, readBytes...)
}
n += len(fnameBuf)
if err != nil {
return
}
fnameBuf = fnameBuf[:len(fnameBuf)-1]
fname = fnameBuf
// Deal with the 20-byte SHA
idx = 0
for idx < 20 {
read := 0
read, err = rd.Read(shaBuf[idx:20])
n += read
if err != nil {
return
}
idx += read
}
sha = shaBuf
return
}
var callerPrefix string
func init() {
_, filename, _, _ := runtime.Caller(0)
callerPrefix = strings.TrimSuffix(filename, "modules/git/batch_reader.go")
}