X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/3ecc1fc9faccc10684b52d90314d7c2176316cbe..2f83fcd45b4b23db2bb5bb4afbe1e863ebd77ec6:/services/keep-web/s3.go
diff --git a/services/keep-web/s3.go b/services/keep-web/s3.go
index 97201d2922..3e60f3006d 100644
--- a/services/keep-web/s3.go
+++ b/services/keep-web/s3.go
@@ -2,17 +2,21 @@
//
// SPDX-License-Identifier: AGPL-3.0
-package main
+package keepweb
import (
"crypto/hmac"
"crypto/sha256"
+ "encoding/base64"
+ "encoding/json"
"encoding/xml"
"errors"
"fmt"
"hash"
"io"
+ "mime"
"net/http"
+ "net/textproto"
"net/url"
"os"
"path/filepath"
@@ -24,7 +28,6 @@ import (
"git.arvados.org/arvados.git/sdk/go/arvados"
"git.arvados.org/arvados.git/sdk/go/ctxlog"
- "github.com/AdRoll/goamz/s3"
)
const (
@@ -33,6 +36,63 @@ const (
s3MaxClockSkew = 5 * time.Minute
)
+type commonPrefix struct {
+ Prefix string
+}
+
+type listV1Resp struct {
+ XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
+ Name string
+ Prefix string
+ Delimiter string
+ Marker string
+ MaxKeys int
+ IsTruncated bool
+ Contents []s3Key
+ // If we use a []string here, xml marshals an empty tag when
+ // CommonPrefixes is nil, which confuses some clients. Fix by
+ // using this nested struct instead.
+ CommonPrefixes []commonPrefix
+ // Similarly, we need omitempty here, because an empty
+ // tag confuses some clients (e.g.,
+ // github.com/aws/aws-sdk-net never terminates its
+ // paging loop).
+ NextMarker string `xml:"NextMarker,omitempty"`
+ // ListObjectsV2 has a KeyCount response field.
+ KeyCount int
+}
+
+type listV2Resp struct {
+ XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
+ IsTruncated bool
+ Contents []s3Key
+ Name string
+ Prefix string
+ Delimiter string
+ MaxKeys int
+ CommonPrefixes []commonPrefix
+ EncodingType string `xml:",omitempty"`
+ KeyCount int
+ ContinuationToken string `xml:",omitempty"`
+ NextContinuationToken string `xml:",omitempty"`
+ StartAfter string `xml:",omitempty"`
+}
+
+type s3Key struct {
+ Key string
+ LastModified string
+ Size int64
+ // The following fields are not populated, but are here in
+ // case clients rely on the keys being present in xml
+ // responses.
+ ETag string
+ StorageClass string
+ Owner struct {
+ ID string
+ DisplayName string
+ }
+}
+
func hmacstring(msg string, key []byte) []byte {
h := hmac.New(sha256.New, key)
io.WriteString(h, msg)
@@ -75,6 +135,8 @@ func s3querystring(u *url.URL) string {
return strings.Join(keys, "&")
}
+var reMultipleSlashChars = regexp.MustCompile(`//+`)
+
func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string, error) {
timefmt, timestr := "20060102T150405Z", r.Header.Get("X-Amz-Date")
if timestr == "" {
@@ -97,11 +159,39 @@ func s3stringToSign(alg, scope, signedHeaders string, r *http.Request) (string,
}
}
- canonicalRequest := fmt.Sprintf("%s\n%s\n%s\n%s\n%s\n%s", r.Method, r.URL.EscapedPath(), s3querystring(r.URL), canonicalHeaders, signedHeaders, r.Header.Get("X-Amz-Content-Sha256"))
+ normalizedPath := normalizePath(r.URL.Path)
+ ctxlog.FromContext(r.Context()).Debugf("normalizedPath %q", normalizedPath)
+ canonicalRequest := fmt.Sprintf("%s\n%s\n%s\n%s\n%s\n%s", r.Method, normalizedPath, s3querystring(r.URL), canonicalHeaders, signedHeaders, r.Header.Get("X-Amz-Content-Sha256"))
ctxlog.FromContext(r.Context()).Debugf("s3stringToSign: canonicalRequest %s", canonicalRequest)
return fmt.Sprintf("%s\n%s\n%s\n%s", alg, r.Header.Get("X-Amz-Date"), scope, hashdigest(sha256.New(), canonicalRequest)), nil
}
+func normalizePath(s string) string {
+ // (url.URL).EscapedPath() would be incorrect here. AWS
+ // documentation specifies the URL path should be normalized
+ // according to RFC 3986, i.e., unescaping ALPHA / DIGIT / "-"
+ // / "." / "_" / "~". The implication is that everything other
+ // than those chars (and "/") _must_ be percent-encoded --
+ // even chars like ";" and "," that are not normally
+ // percent-encoded in paths.
+ out := ""
+ for _, c := range []byte(reMultipleSlashChars.ReplaceAllString(s, "/")) {
+ if (c >= 'a' && c <= 'z') ||
+ (c >= 'A' && c <= 'Z') ||
+ (c >= '0' && c <= '9') ||
+ c == '-' ||
+ c == '.' ||
+ c == '_' ||
+ c == '~' ||
+ c == '/' {
+ out += string(c)
+ } else {
+ out += fmt.Sprintf("%%%02X", c)
+ }
+ }
+ return out
+}
+
func s3signature(secretKey, scope, signedHeaders, stringToSign string) (string, error) {
// scope is {datestamp}/{region}/{service}/aws4_request
drs := strings.Split(scope, "/")
@@ -153,8 +243,8 @@ func (h *handler) checks3signature(r *http.Request) (string, error) {
}
client := (&arvados.Client{
- APIHost: h.Config.cluster.Services.Controller.ExternalURL.Host,
- Insecure: h.Config.cluster.TLS.Insecure,
+ APIHost: h.Cluster.Services.Controller.ExternalURL.Host,
+ Insecure: h.Cluster.TLS.Insecure,
}).WithRequestID(r.Header.Get("X-Request-Id"))
var aca arvados.APIClientAuthorization
var secret string
@@ -162,7 +252,7 @@ func (h *handler) checks3signature(r *http.Request) (string, error) {
if len(key) == 27 && key[5:12] == "-gj3su-" {
// Access key is the UUID of an Arvados token, secret
// key is the secret part.
- ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+h.Config.cluster.SystemRootToken)
+ ctx := arvados.ContextWithAuthorization(r.Context(), "Bearer "+h.Cluster.SystemRootToken)
err = client.RequestAndDecodeContext(ctx, &aca, "GET", "arvados/v1/api_client_authorizations/"+key, nil, nil)
secret = aca.APIToken
} else {
@@ -216,6 +306,8 @@ var UnauthorizedAccess = "UnauthorizedAccess"
var InvalidRequest = "InvalidRequest"
var SignatureDoesNotMatch = "SignatureDoesNotMatch"
+var reRawQueryIndicatesAPI = regexp.MustCompile(`^[a-z]+(&|$)`)
+
// serveS3 handles r and returns true if r is a request from an S3
// client, otherwise it returns false.
func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
@@ -238,20 +330,26 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
return false
}
- _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), token)
+ fs, sess, tokenUser, err := h.Cache.GetSession(token)
if err != nil {
- s3ErrorResponse(w, InternalError, "Pool failed: "+h.clientPool.Err().Error(), r.URL.Path, http.StatusInternalServerError)
+ s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
return true
}
- defer release()
-
- fs := client.SiteFileSystem(kc)
- fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
+ defer sess.Release()
+ readfs := fs
+ if writeMethod[r.Method] {
+ // Create a FileSystem for this request, to avoid
+ // exposing incomplete write operations to concurrent
+ // requests.
+ client := sess.client.WithRequestID(r.Header.Get("X-Request-Id"))
+ fs = client.SiteFileSystem(sess.keepclient)
+ fs.ForwardSlashNameSubstitution(h.Cluster.Collections.ForwardSlashNameSubstitution)
+ }
var objectNameGiven bool
var bucketName string
fspath := "/by_id"
- if id := parseCollectionIDFromDNSName(r.Host); id != "" {
+ if id := arvados.CollectionIDFromDNSName(r.Host); id != "" {
fspath += "/" + id
bucketName = id
objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 0
@@ -259,7 +357,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
bucketName = strings.SplitN(strings.TrimPrefix(r.URL.Path, "/"), "/", 2)[0]
objectNameGiven = strings.Count(strings.TrimSuffix(r.URL.Path, "/"), "/") > 1
}
- fspath += r.URL.Path
+ fspath += reMultipleSlashChars.ReplaceAllString(r.URL.Path, "/")
switch {
case r.Method == http.MethodGet && !objectNameGiven:
@@ -269,16 +367,36 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
w.Header().Set("Content-Type", "application/xml")
io.WriteString(w, xml.Header)
fmt.Fprintln(w, ``)
+ } else if _, ok = r.URL.Query()["location"]; ok {
+ // GetBucketLocation
+ w.Header().Set("Content-Type", "application/xml")
+ io.WriteString(w, xml.Header)
+ fmt.Fprintln(w, ``+
+ h.Cluster.ClusterID+
+ ``)
+ } else if reRawQueryIndicatesAPI.MatchString(r.URL.RawQuery) {
+ // GetBucketWebsite ("GET /bucketid/?website"), GetBucketTagging, etc.
+ s3ErrorResponse(w, InvalidRequest, "API not supported", r.URL.Path+"?"+r.URL.RawQuery, http.StatusBadRequest)
} else {
// ListObjects
h.s3list(bucketName, w, r, fs)
}
return true
case r.Method == http.MethodGet || r.Method == http.MethodHead:
+ if reRawQueryIndicatesAPI.MatchString(r.URL.RawQuery) {
+ // GetObjectRetention ("GET /bucketid/objectid?retention&versionID=..."), etc.
+ s3ErrorResponse(w, InvalidRequest, "API not supported", r.URL.Path+"?"+r.URL.RawQuery, http.StatusBadRequest)
+ return true
+ }
fi, err := fs.Stat(fspath)
if r.Method == "HEAD" && !objectNameGiven {
// HeadBucket
if err == nil && fi.IsDir() {
+ err = setFileInfoHeaders(w.Header(), fs, fspath)
+ if err != nil {
+ s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
+ return true
+ }
w.WriteHeader(http.StatusOK)
} else if os.IsNotExist(err) {
s3ErrorResponse(w, NoSuchBucket, "The specified bucket does not exist.", r.URL.Path, http.StatusNotFound)
@@ -287,7 +405,12 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
}
return true
}
- if err == nil && fi.IsDir() && objectNameGiven && strings.HasSuffix(fspath, "/") && h.Config.cluster.Collections.S3FolderObjects {
+ if err == nil && fi.IsDir() && objectNameGiven && strings.HasSuffix(fspath, "/") && h.Cluster.Collections.S3FolderObjects {
+ err = setFileInfoHeaders(w.Header(), fs, fspath)
+ if err != nil {
+ s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
+ return true
+ }
w.Header().Set("Content-Type", "application/x-directory")
w.WriteHeader(http.StatusOK)
return true
@@ -298,19 +421,36 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
s3ErrorResponse(w, NoSuchKey, "The specified key does not exist.", r.URL.Path, http.StatusNotFound)
return true
}
+
+ if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) {
+ http.Error(w, "Not permitted", http.StatusForbidden)
+ return true
+ }
+ h.logUploadOrDownload(r, sess.arvadosclient, fs, fspath, nil, tokenUser)
+
// shallow copy r, and change URL path
r := *r
r.URL.Path = fspath
+ err = setFileInfoHeaders(w.Header(), fs, fspath)
+ if err != nil {
+ s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusBadGateway)
+ return true
+ }
http.FileServer(fs).ServeHTTP(w, &r)
return true
case r.Method == http.MethodPut:
+ if reRawQueryIndicatesAPI.MatchString(r.URL.RawQuery) {
+ // PutObjectAcl ("PUT /bucketid/objectid?acl&versionID=..."), etc.
+ s3ErrorResponse(w, InvalidRequest, "API not supported", r.URL.Path+"?"+r.URL.RawQuery, http.StatusBadRequest)
+ return true
+ }
if !objectNameGiven {
s3ErrorResponse(w, InvalidArgument, "Missing object name in PUT request.", r.URL.Path, http.StatusBadRequest)
return true
}
var objectIsDir bool
if strings.HasSuffix(fspath, "/") {
- if !h.Config.cluster.Collections.S3FolderObjects {
+ if !h.Cluster.Collections.S3FolderObjects {
s3ErrorResponse(w, InvalidArgument, "invalid object name: trailing slash", r.URL.Path, http.StatusBadRequest)
return true
}
@@ -352,7 +492,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
return true
}
err = fs.Mkdir(dir, 0755)
- if err == arvados.ErrInvalidArgument {
+ if errors.Is(err, arvados.ErrInvalidArgument) || errors.Is(err, arvados.ErrInvalidOperation) {
// Cannot create a directory
// here.
err = fmt.Errorf("mkdir %q failed: %w", dir, err)
@@ -376,6 +516,13 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
return true
}
defer f.Close()
+
+ if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) {
+ http.Error(w, "Not permitted", http.StatusForbidden)
+ return true
+ }
+ h.logUploadOrDownload(r, sess.arvadosclient, fs, fspath, nil, tokenUser)
+
_, err = io.Copy(f, r.Body)
if err != nil {
err = fmt.Errorf("write to %q failed: %w", r.URL.Path, err)
@@ -389,7 +536,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
return true
}
}
- err = fs.Sync()
+ err = h.syncCollection(fs, readfs, fspath)
if err != nil {
err = fmt.Errorf("sync failed: %w", err)
s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
@@ -398,6 +545,11 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
w.WriteHeader(http.StatusOK)
return true
case r.Method == http.MethodDelete:
+ if reRawQueryIndicatesAPI.MatchString(r.URL.RawQuery) {
+ // DeleteObjectTagging ("DELETE /bucketid/objectid?tagging&versionID=..."), etc.
+ s3ErrorResponse(w, InvalidRequest, "API not supported", r.URL.Path+"?"+r.URL.RawQuery, http.StatusBadRequest)
+ return true
+ }
if !objectNameGiven || r.URL.Path == "/" {
s3ErrorResponse(w, InvalidArgument, "missing object name in DELETE request", r.URL.Path, http.StatusBadRequest)
return true
@@ -436,7 +588,7 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
s3ErrorResponse(w, InvalidArgument, err.Error(), r.URL.Path, http.StatusBadRequest)
return true
}
- err = fs.Sync()
+ err = h.syncCollection(fs, readfs, fspath)
if err != nil {
err = fmt.Errorf("sync failed: %w", err)
s3ErrorResponse(w, InternalError, err.Error(), r.URL.Path, http.StatusInternalServerError)
@@ -446,11 +598,92 @@ func (h *handler) serveS3(w http.ResponseWriter, r *http.Request) bool {
return true
default:
s3ErrorResponse(w, InvalidRequest, "method not allowed", r.URL.Path, http.StatusMethodNotAllowed)
-
return true
}
}
+// Save modifications to the indicated collection in srcfs, then (if
+// successful) ensure they are also reflected in dstfs.
+func (h *handler) syncCollection(srcfs, dstfs arvados.CustomFileSystem, path string) error {
+ coll, _ := h.determineCollection(srcfs, path)
+ if coll == nil || coll.UUID == "" {
+ return errors.New("could not determine collection to sync")
+ }
+ d, err := srcfs.OpenFile("by_id/"+coll.UUID, os.O_RDWR, 0777)
+ if err != nil {
+ return err
+ }
+ defer d.Close()
+ err = d.Sync()
+ if err != nil {
+ return err
+ }
+ snap, err := d.Snapshot()
+ if err != nil {
+ return err
+ }
+ dstd, err := dstfs.OpenFile("by_id/"+coll.UUID, os.O_RDWR, 0777)
+ if err != nil {
+ return err
+ }
+ defer dstd.Close()
+ return dstd.Splice(snap)
+}
+
+func setFileInfoHeaders(header http.Header, fs arvados.CustomFileSystem, path string) error {
+ maybeEncode := func(s string) string {
+ for _, c := range s {
+ if c > '\u007f' || c < ' ' {
+ return mime.BEncoding.Encode("UTF-8", s)
+ }
+ }
+ return s
+ }
+ path = strings.TrimSuffix(path, "/")
+ var props map[string]interface{}
+ for {
+ fi, err := fs.Stat(path)
+ if err != nil {
+ return err
+ }
+ switch src := fi.Sys().(type) {
+ case *arvados.Collection:
+ props = src.Properties
+ case *arvados.Group:
+ props = src.Properties
+ default:
+ if err, ok := src.(error); ok {
+ return err
+ }
+ // Try parent
+ cut := strings.LastIndexByte(path, '/')
+ if cut < 0 {
+ return nil
+ }
+ path = path[:cut]
+ continue
+ }
+ break
+ }
+ for k, v := range props {
+ if !validMIMEHeaderKey(k) {
+ continue
+ }
+ k = "x-amz-meta-" + k
+ if s, ok := v.(string); ok {
+ header.Set(k, maybeEncode(s))
+ } else if j, err := json.Marshal(v); err == nil {
+ header.Set(k, maybeEncode(string(j)))
+ }
+ }
+ return nil
+}
+
+func validMIMEHeaderKey(k string) bool {
+ check := "z-" + k
+ return check != textproto.CanonicalMIMEHeaderKey(check)
+}
+
// Call fn on the given path (directory) and its contents, in
// lexicographic order.
//
@@ -509,19 +742,61 @@ var errDone = errors.New("done")
func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request, fs arvados.CustomFileSystem) {
var params struct {
- delimiter string
- marker string
- maxKeys int
- prefix string
+ v2 bool
+ delimiter string
+ maxKeys int
+ prefix string
+ marker string // decoded continuationToken (v2) or provided by client (v1)
+ startAfter string // v2
+ continuationToken string // v2
+ encodingTypeURL bool // v2
}
params.delimiter = r.FormValue("delimiter")
- params.marker = r.FormValue("marker")
if mk, _ := strconv.ParseInt(r.FormValue("max-keys"), 10, 64); mk > 0 && mk < s3MaxKeys {
params.maxKeys = int(mk)
} else {
params.maxKeys = s3MaxKeys
}
params.prefix = r.FormValue("prefix")
+ switch r.FormValue("list-type") {
+ case "":
+ case "2":
+ params.v2 = true
+ default:
+ http.Error(w, "invalid list-type parameter", http.StatusBadRequest)
+ return
+ }
+ if params.v2 {
+ params.continuationToken = r.FormValue("continuation-token")
+ marker, err := base64.StdEncoding.DecodeString(params.continuationToken)
+ if err != nil {
+ http.Error(w, "invalid continuation token", http.StatusBadRequest)
+ return
+ }
+ // marker and start-after perform the same function,
+ // but we keep them separate so we can repeat them
+ // back to the client in the response.
+ params.marker = string(marker)
+ params.startAfter = r.FormValue("start-after")
+ switch r.FormValue("encoding-type") {
+ case "":
+ case "url":
+ params.encodingTypeURL = true
+ default:
+ http.Error(w, "invalid encoding-type parameter", http.StatusBadRequest)
+ return
+ }
+ } else {
+ // marker is functionally equivalent to start-after.
+ params.marker = r.FormValue("marker")
+ }
+
+ // startAfter is params.marker or params.startAfter, whichever
+ // comes last.
+ startAfter := params.startAfter
+ if startAfter < params.marker {
+ startAfter = params.marker
+ }
bucketdir := "by_id/" + bucket
// walkpath is the directory (relative to bucketdir) we need
@@ -538,34 +813,23 @@ func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request,
walkpath = ""
}
- type commonPrefix struct {
- Prefix string
- }
- type listResp struct {
- XMLName string `xml:"http://s3.amazonaws.com/doc/2006-03-01/ ListBucketResult"`
- s3.ListResp
- // s3.ListResp marshals an empty tag when
- // CommonPrefixes is nil, which confuses some clients.
- // Fix by using this nested struct instead.
- CommonPrefixes []commonPrefix
- // Similarly, we need omitempty here, because an empty
- // tag confuses some clients (e.g.,
- // github.com/aws/aws-sdk-net never terminates its
- // paging loop).
- NextMarker string `xml:"NextMarker,omitempty"`
- // ListObjectsV2 has a KeyCount response field.
- KeyCount int
- }
- resp := listResp{
- ListResp: s3.ListResp{
- Name: bucket,
- Prefix: params.prefix,
- Delimiter: params.delimiter,
- Marker: params.marker,
- MaxKeys: params.maxKeys,
- },
+ resp := listV2Resp{
+ Name: bucket,
+ Prefix: params.prefix,
+ Delimiter: params.delimiter,
+ MaxKeys: params.maxKeys,
+ ContinuationToken: r.FormValue("continuation-token"),
+ StartAfter: params.startAfter,
}
+
+ // nextMarker will be the last path we add to either
+ // resp.Contents or commonPrefixes. It will be included in
+ // the response as NextMarker or NextContinuationToken if
+ // needed.
+ nextMarker := ""
+
commonPrefixes := map[string]bool{}
+ full := false
err := walkFS(fs, strings.TrimSuffix(bucketdir+"/"+walkpath, "/"), true, func(path string, fi os.FileInfo) error {
if path == bucketdir {
return nil
@@ -576,38 +840,31 @@ func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request,
path += "/"
filesize = 0
}
- if len(path) <= len(params.prefix) {
- if path > params.prefix[:len(path)] {
- // with prefix "foobar", walking "fooz" means we're done
- return errDone
- }
- if path < params.prefix[:len(path)] {
- // with prefix "foobar", walking "foobag" is pointless
- return filepath.SkipDir
- }
- if fi.IsDir() && !strings.HasPrefix(params.prefix+"/", path) {
- // with prefix "foo/bar", walking "fo"
- // is pointless (but walking "foo" or
- // "foo/bar" is necessary)
- return filepath.SkipDir
- }
- if len(path) < len(params.prefix) {
- // can't skip anything, and this entry
- // isn't in the results, so just
- // continue descent
- return nil
- }
- } else {
- if path[:len(params.prefix)] > params.prefix {
- // with prefix "foobar", nothing we
- // see after "foozzz" is relevant
- return errDone
- }
- }
- if path < params.marker || path < params.prefix {
+ if strings.HasPrefix(params.prefix, path) && params.prefix != path {
+ // Descend into subtree until we reach desired prefix
return nil
+ } else if path < params.prefix {
+ // Not an ancestor or descendant of desired
+ // prefix, therefore none of its descendants
+ // can be either -- skip
+ return filepath.SkipDir
+ } else if path > params.prefix && !strings.HasPrefix(path, params.prefix) {
+ // We must have traversed everything under
+ // desired prefix
+ return errDone
+ } else if path == startAfter {
+ // Skip startAfter itself, just descend into
+ // subtree
+ return nil
+ } else if strings.HasPrefix(startAfter, path) {
+ // Descend into subtree in case it contains
+ // something after startAfter
+ return nil
+ } else if path < startAfter {
+ // Skip ahead until we reach startAfter
+ return filepath.SkipDir
}
- if fi.IsDir() && !h.Config.cluster.Collections.S3FolderObjects {
+ if fi.IsDir() && !h.Cluster.Collections.S3FolderObjects {
// Note we don't add anything to
// commonPrefixes here even if delimiter is
// "/". We descend into the directory, and
@@ -622,28 +879,42 @@ func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request,
// "z", when we hit "foobar/baz", we
// add "/baz" to commonPrefixes and
// stop descending.
- commonPrefixes[path[:len(params.prefix)+idx+1]] = true
- return filepath.SkipDir
+ prefix := path[:len(params.prefix)+idx+1]
+ if prefix == startAfter {
+ return nil
+ } else if prefix < startAfter && !strings.HasPrefix(startAfter, prefix) {
+ return nil
+ } else if full {
+ resp.IsTruncated = true
+ return errDone
+ } else {
+ commonPrefixes[prefix] = true
+ nextMarker = prefix
+ full = len(resp.Contents)+len(commonPrefixes) >= params.maxKeys
+ return filepath.SkipDir
+ }
}
}
- if len(resp.Contents)+len(commonPrefixes) >= params.maxKeys {
+ if full {
resp.IsTruncated = true
- if params.delimiter != "" {
- resp.NextMarker = path
- }
return errDone
}
- resp.Contents = append(resp.Contents, s3.Key{
+ resp.Contents = append(resp.Contents, s3Key{
Key: path,
LastModified: fi.ModTime().UTC().Format("2006-01-02T15:04:05.999") + "Z",
Size: filesize,
})
+ nextMarker = path
+ full = len(resp.Contents)+len(commonPrefixes) >= params.maxKeys
return nil
})
if err != nil && err != errDone {
http.Error(w, err.Error(), http.StatusInternalServerError)
return
}
+ if params.delimiter == "" && !params.v2 || !resp.IsTruncated {
+ nextMarker = ""
+ }
if params.delimiter != "" {
resp.CommonPrefixes = make([]commonPrefix, 0, len(commonPrefixes))
for prefix := range commonPrefixes {
@@ -652,9 +923,64 @@ func (h *handler) s3list(bucket string, w http.ResponseWriter, r *http.Request,
sort.Slice(resp.CommonPrefixes, func(i, j int) bool { return resp.CommonPrefixes[i].Prefix < resp.CommonPrefixes[j].Prefix })
}
resp.KeyCount = len(resp.Contents)
+ var respV1orV2 interface{}
+
+ if params.encodingTypeURL {
+ // https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html
+ // "If you specify the encoding-type request
+ // parameter, Amazon S3 includes this element in the
+ // response, and returns encoded key name values in
+ // the following response elements:
+ //
+ // Delimiter, Prefix, Key, and StartAfter.
+ //
+ // Type: String
+ //
+ // Valid Values: url"
+ //
+ // This is somewhat vague but in practice it appears
+ // to mean x-www-form-urlencoded as in RFC1866 8.2.1
+ // para 1 (encode space as "+") rather than straight
+ // percent-encoding as in RFC1738 2.2. Presumably,
+ // the intent is to allow the client to decode XML and
+ // then paste the strings directly into another URI
+ // query or POST form like "https://host/path?foo=" +
+ // foo + "&bar=" + bar.
+ resp.EncodingType = "url"
+ resp.Delimiter = url.QueryEscape(resp.Delimiter)
+ resp.Prefix = url.QueryEscape(resp.Prefix)
+ resp.StartAfter = url.QueryEscape(resp.StartAfter)
+ for i, ent := range resp.Contents {
+ ent.Key = url.QueryEscape(ent.Key)
+ resp.Contents[i] = ent
+ }
+ for i, ent := range resp.CommonPrefixes {
+ ent.Prefix = url.QueryEscape(ent.Prefix)
+ resp.CommonPrefixes[i] = ent
+ }
+ }
+
+ if params.v2 {
+ resp.NextContinuationToken = base64.StdEncoding.EncodeToString([]byte(nextMarker))
+ respV1orV2 = resp
+ } else {
+ respV1orV2 = listV1Resp{
+ CommonPrefixes: resp.CommonPrefixes,
+ NextMarker: nextMarker,
+ KeyCount: resp.KeyCount,
+ IsTruncated: resp.IsTruncated,
+ Name: bucket,
+ Prefix: params.prefix,
+ Delimiter: params.delimiter,
+ Marker: params.marker,
+ MaxKeys: params.maxKeys,
+ Contents: resp.Contents,
+ }
+ }
+
w.Header().Set("Content-Type", "application/xml")
io.WriteString(w, xml.Header)
- if err := xml.NewEncoder(w).Encode(resp); err != nil {
+ if err := xml.NewEncoder(w).Encode(respV1orV2); err != nil {
ctxlog.FromContext(r.Context()).WithError(err).Error("error writing xml response")
}
}