storage/reader.go

// Copyright 2016 Google LLC
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//      http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

package storage

import (
	"context"
	"errors"
	"fmt"
	"hash/crc32"
	"io"
	"io/ioutil"
	"net/http"
	"net/url"
	"reflect"
	"strconv"
	"strings"
	"time"

	"cloud.google.com/go/internal/trace"
	"google.golang.org/api/googleapi"
)

var crc32cTable = crc32.MakeTable(crc32.Castagnoli)

// ReaderObjectAttrs are attributes about the object being read. These are populated
// during the New call. This struct only holds a subset of object attributes: to
// get the full set of attributes, use ObjectHandle.Attrs.
//
// Each field is read-only.
type ReaderObjectAttrs struct {
	// Size is the length of the object's content.
	Size int64

	// StartOffset is the byte offset within the object
	// from which reading begins.
	// This value is only non-zero for range requests.
	StartOffset int64

	// ContentType is the MIME type of the object's content.
	ContentType string

	// ContentEncoding is the encoding of the object's content.
	ContentEncoding string

	// CacheControl specifies whether and for how long browser and Internet
	// caches are allowed to cache your objects.
	CacheControl string

	// LastModified is the time that the object was last modified.
	LastModified time.Time

	// Generation is the generation number of the object's content.
	Generation int64

	// Metageneration is the version of the metadata for this object at
	// this generation. This field is used for preconditions and for
	// detecting changes in metadata. A metageneration number is only
	// meaningful in the context of a particular generation of a
	// particular object.
	Metageneration int64
}

// NewReader creates a new Reader to read the contents of the
// object.
// ErrObjectNotExist will be returned if the object is not found.
//
// The caller must call Close on the returned Reader when done reading.
func (o *ObjectHandle) NewReader(ctx context.Context) (*Reader, error) {
	return o.NewRangeReader(ctx, 0, -1)
}

// NewRangeReader reads part of an object, reading at most length bytes
// starting at the given offset. If length is negative, the object is read
// until the end. If offset is negative, the object is read abs(offset) bytes
// from the end, and length must also be negative to indicate all remaining
// bytes will be read.
//
// If the object's metadata property "Content-Encoding" is set to "gzip" or satisfies
// decompressive transcoding per https://cloud.google.com/storage/docs/transcoding
// that file will be served back whole, regardless of the requested range as
// Google Cloud Storage dictates.
func (o *ObjectHandle) NewRangeReader(ctx context.Context, offset, length int64) (r *Reader, err error) {
	ctx = trace.StartSpan(ctx, "cloud.google.com/go/storage.Object.NewRangeReader")
	defer func() { trace.EndSpan(ctx, err) }()

	if err := o.validate(); err != nil {
		return nil, err
	}
	if offset < 0 && length >= 0 {
		return nil, fmt.Errorf("storage: invalid offset %d < 0 requires negative length", offset)
	}
	if o.conds != nil {
		if err := o.conds.validate("NewRangeReader"); err != nil {
			return nil, err
		}
	}
	u := &url.URL{
		Scheme: o.c.scheme,
		Host:   o.c.readHost,
		Path:   fmt.Sprintf("/%s/%s", o.bucket, o.object),
	}
	verb := "GET"
	if length == 0 {
		verb = "HEAD"
	}
	req, err := http.NewRequest(verb, u.String(), nil)
	if err != nil {
		return nil, err
	}
	req = req.WithContext(ctx)
	if o.userProject != "" {
		req.Header.Set("X-Goog-User-Project", o.userProject)
	}
	if o.readCompressed {
		req.Header.Set("Accept-Encoding", "gzip")
	}
	if err := setEncryptionHeaders(req.Header, o.encryptionKey, false); err != nil {
		return nil, err
	}

	gen := o.gen

	// Define a function that initiates a Read with offset and length, assuming we
	// have already read seen bytes.
	reopen := func(seen int64) (*http.Response, error) {
		start := offset + seen
		if length < 0 && start < 0 {
			req.Header.Set("Range", fmt.Sprintf("bytes=%d", start))
		} else if length < 0 && start > 0 {
			req.Header.Set("Range", fmt.Sprintf("bytes=%d-", start))
		} else if length > 0 {
			// The end character isn't affected by how many bytes we've seen.
			req.Header.Set("Range", fmt.Sprintf("bytes=%d-%d", start, offset+length-1))
		}
		// We wait to assign conditions here because the generation number can change in between reopen() runs.
		req.URL.RawQuery = conditionsQuery(gen, o.conds)
		var res *http.Response
		err = runWithRetry(ctx, func() error {
			res, err = o.c.hc.Do(req)
			if err != nil {
				return err
			}
			if res.StatusCode == http.StatusNotFound {
				res.Body.Close()
				return ErrObjectNotExist
			}
			if res.StatusCode < 200 || res.StatusCode > 299 {
				body, _ := ioutil.ReadAll(res.Body)
				res.Body.Close()
				return &googleapi.Error{
					Code:   res.StatusCode,
					Header: res.Header,
					Body:   string(body),
				}
			}

			partialContentNotSatisfied :=
				!decompressiveTranscoding(res) &&
					start > 0 && length != 0 &&
					res.StatusCode != http.StatusPartialContent

			if partialContentNotSatisfied {
				res.Body.Close()
				return errors.New("storage: partial request not satisfied")
			}

			// With "Content-Encoding": "gzip" aka decompressive transcoding, GCS serves
			// back the whole file regardless of the range count passed in as per:
			//      https://cloud.google.com/storage/docs/transcoding#range,
			// thus we have to manually move the body forward by seen bytes.
			if decompressiveTranscoding(res) && seen > 0 {
				_, _ = io.CopyN(ioutil.Discard, res.Body, seen)
			}

			// If a generation hasn't been specified, and this is the first response we get, let's record the
			// generation. In future requests we'll use this generation as a precondition to avoid data races.
			if gen < 0 && res.Header.Get("X-Goog-Generation") != "" {
				gen64, err := strconv.ParseInt(res.Header.Get("X-Goog-Generation"), 10, 64)
				if err != nil {
					return err
				}
				gen = gen64
			}
			return nil
		})
		if err != nil {
			return nil, err
		}
		return res, nil
	}

	res, err := reopen(0)
	if err != nil {
		return nil, err
	}
	var (
		size        int64 // total size of object, even if a range was requested.
		checkCRC    bool
		crc         uint32
		startOffset int64 // non-zero if range request.
	)
	if res.StatusCode == http.StatusPartialContent {
		cr := strings.TrimSpace(res.Header.Get("Content-Range"))
		if !strings.HasPrefix(cr, "bytes ") || !strings.Contains(cr, "/") {
			return nil, fmt.Errorf("storage: invalid Content-Range %q", cr)
		}
		size, err = strconv.ParseInt(cr[strings.LastIndex(cr, "/")+1:], 10, 64)
		if err != nil {
			return nil, fmt.Errorf("storage: invalid Content-Range %q", cr)
		}

		dashIndex := strings.Index(cr, "-")
		if dashIndex >= 0 {
			startOffset, err = strconv.ParseInt(cr[len("bytes="):dashIndex], 10, 64)
			if err != nil {
				return nil, fmt.Errorf("storage: invalid Content-Range %q: %v", cr, err)
			}
		}
	} else {
		size = res.ContentLength
		// Check the CRC iff all of the following hold:
		// - We asked for content (length != 0).
		// - We got all the content (status != PartialContent).
		// - The server sent a CRC header.
		// - The Go http stack did not uncompress the file.
		// - We were not served compressed data that was uncompressed on download.
		// The problem with the last two cases is that the CRC will not match -- GCS
		// computes it on the compressed contents, but we compute it on the
		// uncompressed contents.
		if length != 0 && !res.Uncompressed && !uncompressedByServer(res) {
			crc, checkCRC = parseCRC32c(res)
		}
	}

	remain := res.ContentLength
	body := res.Body
	if length == 0 {
		remain = 0
		body.Close()
		body = emptyBody
	}
	var metaGen int64
	if res.Header.Get("X-Goog-Metageneration") != "" {
		metaGen, err = strconv.ParseInt(res.Header.Get("X-Goog-Metageneration"), 10, 64)
		if err != nil {
			return nil, err
		}
	}

	var lm time.Time
	if res.Header.Get("Last-Modified") != "" {
		lm, err = http.ParseTime(res.Header.Get("Last-Modified"))
		if err != nil {
			return nil, err
		}
	}

	attrs := ReaderObjectAttrs{
		Size:            size,
		ContentType:     res.Header.Get("Content-Type"),
		ContentEncoding: res.Header.Get("Content-Encoding"),
		CacheControl:    res.Header.Get("Cache-Control"),
		LastModified:    lm,
		StartOffset:     startOffset,
		Generation:      gen,
		Metageneration:  metaGen,
	}
	return &Reader{
		Attrs:    attrs,
		body:     body,
		size:     size,
		remain:   remain,
		wantCRC:  crc,
		checkCRC: checkCRC,
		reopen:   reopen,
	}, nil
}

// decompressiveTranscoding returns true if the request was served decompressed
// and different than its original storage form. This happens when the "Content-Encoding"
// header is "gzip".
// See:
//  * https://cloud.google.com/storage/docs/transcoding#transcoding_and_gzip
//  * https://github.com/googleapis/google-cloud-go/issues/1800
func decompressiveTranscoding(res *http.Response) bool {
	// Decompressive Transcoding.
	return res.Header.Get("Content-Encoding") == "gzip" ||
		res.Header.Get("X-Goog-Stored-Content-Encoding") == "gzip"
}

func uncompressedByServer(res *http.Response) bool {
	// If the data is stored as gzip but is not encoded as gzip, then it
	// was uncompressed by the server.
	return res.Header.Get("X-Goog-Stored-Content-Encoding") == "gzip" &&
		res.Header.Get("Content-Encoding") != "gzip"
}

func parseCRC32c(res *http.Response) (uint32, bool) {
	const prefix = "crc32c="
	for _, spec := range res.Header["X-Goog-Hash"] {
		if strings.HasPrefix(spec, prefix) {
			c, err := decodeUint32(spec[len(prefix):])
			if err == nil {
				return c, true
			}
		}
	}
	return 0, false
}

var emptyBody = ioutil.NopCloser(strings.NewReader(""))

// Reader reads a Cloud Storage object.
// It implements io.Reader.
//
// Typically, a Reader computes the CRC of the downloaded content and compares it to
// the stored CRC, returning an error from Read if there is a mismatch. This integrity check
// is skipped if transcoding occurs. See https://cloud.google.com/storage/docs/transcoding.
type Reader struct {
	Attrs              ReaderObjectAttrs
	body               io.ReadCloser
	seen, remain, size int64
	checkCRC           bool   // should we check the CRC?
	wantCRC            uint32 // the CRC32c value the server sent in the header
	gotCRC             uint32 // running crc
	reopen             func(seen int64) (*http.Response, error)
}

// Close closes the Reader. It must be called when done reading.
func (r *Reader) Close() error {
	return r.body.Close()
}

func (r *Reader) Read(p []byte) (int, error) {
	n, err := r.readWithRetry(p)
	if r.remain != -1 {
		r.remain -= int64(n)
	}
	if r.checkCRC {
		r.gotCRC = crc32.Update(r.gotCRC, crc32cTable, p[:n])
		// Check CRC here. It would be natural to check it in Close, but
		// everybody defers Close on the assumption that it doesn't return
		// anything worth looking at.
		if err == io.EOF {
			if r.gotCRC != r.wantCRC {
				return n, fmt.Errorf("storage: bad CRC on read: got %d, want %d",
					r.gotCRC, r.wantCRC)
			}
		}
	}
	return n, err
}

func (r *Reader) readWithRetry(p []byte) (int, error) {
	n := 0
	for len(p[n:]) > 0 {
		m, err := r.body.Read(p[n:])
		n += m
		r.seen += int64(m)
		if !shouldRetryRead(err) {
			return n, err
		}
		// Read failed, but we will try again. Send a ranged read request that takes
		// into account the number of bytes we've already seen.
		res, err := r.reopen(r.seen)
		if err != nil {
			// reopen already retries
			return n, err
		}
		r.body.Close()
		r.body = res.Body
	}
	return n, nil
}

func shouldRetryRead(err error) bool {
	if err == nil {
		return false
	}
	return strings.HasSuffix(err.Error(), "INTERNAL_ERROR") && strings.Contains(reflect.TypeOf(err).String(), "http2")
}

// Size returns the size of the object in bytes.
// The returned value is always the same and is not affected by
// calls to Read or Close.
//
// Deprecated: use Reader.Attrs.Size.
func (r *Reader) Size() int64 {
	return r.Attrs.Size
}

// Remain returns the number of bytes left to read, or -1 if unknown.
func (r *Reader) Remain() int64 {
	return r.remain
}

// ContentType returns the content type of the object.
//
// Deprecated: use Reader.Attrs.ContentType.
func (r *Reader) ContentType() string {
	return r.Attrs.ContentType
}

// ContentEncoding returns the content encoding of the object.
//
// Deprecated: use Reader.Attrs.ContentEncoding.
func (r *Reader) ContentEncoding() string {
	return r.Attrs.ContentEncoding
}

// CacheControl returns the cache control of the object.
//
// Deprecated: use Reader.Attrs.CacheControl.
func (r *Reader) CacheControl() string {
	return r.Attrs.CacheControl
}

// LastModified returns the value of the Last-Modified header.
//
// Deprecated: use Reader.Attrs.LastModified.
func (r *Reader) LastModified() (time.Time, error) {
	return r.Attrs.LastModified, nil
}