14262: Make sure cancel() from proxy.Do() gets called
[arvados.git] / lib / controller / federation.go
1 // Copyright (C) The Arvados Authors. All rights reserved.
2 //
3 // SPDX-License-Identifier: AGPL-3.0
4
5 package controller
6
7 import (
8         "bytes"
9         "context"
10         "database/sql"
11         "encoding/json"
12         "fmt"
13         "io"
14         "io/ioutil"
15         "net/http"
16         "net/url"
17         "regexp"
18         "strings"
19
20         "git.curoverse.com/arvados.git/sdk/go/arvados"
21         "git.curoverse.com/arvados.git/sdk/go/auth"
22         "github.com/jmcvetta/randutil"
23 )
24
25 var pathPattern = `^/arvados/v1/%s(/([0-9a-z]{5})-%s-[0-9a-z]{15})?(.*)$`
26 var wfRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "workflows", "7fd4e"))
27 var containersRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "containers", "dz642"))
28 var containerRequestsRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "container_requests", "xvhdp"))
29 var collectionRe = regexp.MustCompile(fmt.Sprintf(pathPattern, "collections", "4zz18"))
30 var collectionByPDHRe = regexp.MustCompile(`^/arvados/v1/collections/([0-9a-fA-F]{32}\+[0-9]+)+$`)
31
32 func (h *Handler) remoteClusterRequest(remoteID string, req *http.Request) (*http.Response, context.CancelFunc, error) {
33         remote, ok := h.Cluster.RemoteClusters[remoteID]
34         if !ok {
35                 return nil, nil, HTTPError{fmt.Sprintf("no proxy available for cluster %v", remoteID), http.StatusNotFound}
36         }
37         scheme := remote.Scheme
38         if scheme == "" {
39                 scheme = "https"
40         }
41         saltedReq, err := h.saltAuthToken(req, remoteID)
42         if err != nil {
43                 return nil, nil, err
44         }
45         urlOut := &url.URL{
46                 Scheme:   scheme,
47                 Host:     remote.Host,
48                 Path:     saltedReq.URL.Path,
49                 RawPath:  saltedReq.URL.RawPath,
50                 RawQuery: saltedReq.URL.RawQuery,
51         }
52         client := h.secureClient
53         if remote.Insecure {
54                 client = h.insecureClient
55         }
56         return h.proxy.Do(saltedReq, urlOut, client)
57 }
58
59 // Buffer request body, parse form parameters in request, and then
60 // replace original body with the buffer so it can be re-read by
61 // downstream proxy steps.
62 func loadParamsFromForm(req *http.Request) error {
63         var postBody *bytes.Buffer
64         if req.Body != nil && req.Header.Get("Content-Type") == "application/x-www-form-urlencoded" {
65                 var cl int64
66                 if req.ContentLength > 0 {
67                         cl = req.ContentLength
68                 }
69                 postBody = bytes.NewBuffer(make([]byte, 0, cl))
70                 originalBody := req.Body
71                 defer originalBody.Close()
72                 req.Body = ioutil.NopCloser(io.TeeReader(req.Body, postBody))
73         }
74
75         err := req.ParseForm()
76         if err != nil {
77                 return err
78         }
79
80         if req.Body != nil && postBody != nil {
81                 req.Body = ioutil.NopCloser(postBody)
82         }
83         return nil
84 }
85
86 func (h *Handler) setupProxyRemoteCluster(next http.Handler) http.Handler {
87         mux := http.NewServeMux()
88
89         wfHandler := &genericFederatedRequestHandler{next, h, wfRe, nil}
90         containersHandler := &genericFederatedRequestHandler{next, h, containersRe, nil}
91         containerRequestsHandler := &genericFederatedRequestHandler{next, h, containerRequestsRe,
92                 []federatedRequestDelegate{remoteContainerRequestCreate}}
93
94         mux.Handle("/arvados/v1/workflows", wfHandler)
95         mux.Handle("/arvados/v1/workflows/", wfHandler)
96         mux.Handle("/arvados/v1/containers", containersHandler)
97         mux.Handle("/arvados/v1/containers/", containersHandler)
98         mux.Handle("/arvados/v1/container_requests", containerRequestsHandler)
99         mux.Handle("/arvados/v1/container_requests/", containerRequestsHandler)
100         mux.Handle("/arvados/v1/collections", next)
101         mux.Handle("/arvados/v1/collections/", &collectionFederatedRequestHandler{next, h})
102         mux.Handle("/", next)
103
104         return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
105                 parts := strings.Split(req.Header.Get("Authorization"), "/")
106                 alreadySalted := (len(parts) == 3 && parts[0] == "Bearer v2" && len(parts[2]) == 40)
107
108                 if alreadySalted ||
109                         strings.Index(req.Header.Get("Via"), "arvados-controller") != -1 {
110                         // The token is already salted, or this is a
111                         // request from another instance of
112                         // arvados-controller.  In either case, we
113                         // don't want to proxy this query, so just
114                         // continue down the instance handler stack.
115                         next.ServeHTTP(w, req)
116                         return
117                 }
118
119                 mux.ServeHTTP(w, req)
120         })
121
122         return mux
123 }
124
125 type CurrentUser struct {
126         Authorization arvados.APIClientAuthorization
127         UUID          string
128 }
129
130 // validateAPItoken extracts the token from the provided http request,
131 // checks it again api_client_authorizations table in the database,
132 // and fills in the token scope and user UUID.  Does not handle remote
133 // tokens unless they are already in the database and not expired.
134 func (h *Handler) validateAPItoken(req *http.Request, token string) (*CurrentUser, error) {
135         user := CurrentUser{Authorization: arvados.APIClientAuthorization{APIToken: token}}
136         db, err := h.db(req)
137         if err != nil {
138                 return nil, err
139         }
140
141         var uuid string
142         if strings.HasPrefix(token, "v2/") {
143                 sp := strings.Split(token, "/")
144                 uuid = sp[1]
145                 token = sp[2]
146         }
147         user.Authorization.APIToken = token
148         var scopes string
149         err = db.QueryRowContext(req.Context(), `SELECT api_client_authorizations.uuid, api_client_authorizations.scopes, users.uuid FROM api_client_authorizations JOIN users on api_client_authorizations.user_id=users.id WHERE api_token=$1 AND (expires_at IS NULL OR expires_at > current_timestamp) LIMIT 1`, token).Scan(&user.Authorization.UUID, &scopes, &user.UUID)
150         if err != nil {
151                 return nil, err
152         }
153         if uuid != "" && user.Authorization.UUID != uuid {
154                 return nil, fmt.Errorf("UUID embedded in v2 token did not match record")
155         }
156         err = json.Unmarshal([]byte(scopes), &user.Authorization.Scopes)
157         if err != nil {
158                 return nil, err
159         }
160         return &user, nil
161 }
162
163 func (h *Handler) createAPItoken(req *http.Request, userUUID string, scopes []string) (*arvados.APIClientAuthorization, error) {
164         db, err := h.db(req)
165         if err != nil {
166                 return nil, err
167         }
168         rd, err := randutil.String(15, "abcdefghijklmnopqrstuvwxyz0123456789")
169         if err != nil {
170                 return nil, err
171         }
172         uuid := fmt.Sprintf("%v-gj3su-%v", h.Cluster.ClusterID, rd)
173         token, err := randutil.String(50, "abcdefghijklmnopqrstuvwxyz0123456789")
174         if err != nil {
175                 return nil, err
176         }
177         if len(scopes) == 0 {
178                 scopes = append(scopes, "all")
179         }
180         scopesjson, err := json.Marshal(scopes)
181         if err != nil {
182                 return nil, err
183         }
184         _, err = db.ExecContext(req.Context(),
185                 `INSERT INTO api_client_authorizations
186 (uuid, api_token, expires_at, scopes,
187 user_id,
188 api_client_id, created_at, updated_at)
189 VALUES ($1, $2, CURRENT_TIMESTAMP + INTERVAL '2 weeks', $3,
190 (SELECT id FROM users WHERE users.uuid=$4 LIMIT 1),
191 0, CURRENT_TIMESTAMP, CURRENT_TIMESTAMP)`,
192                 uuid, token, string(scopesjson), userUUID)
193
194         if err != nil {
195                 return nil, err
196         }
197
198         return &arvados.APIClientAuthorization{
199                 UUID:      uuid,
200                 APIToken:  token,
201                 ExpiresAt: "",
202                 Scopes:    scopes}, nil
203 }
204
205 // Extract the auth token supplied in req, and replace it with a
206 // salted token for the remote cluster.
207 func (h *Handler) saltAuthToken(req *http.Request, remote string) (updatedReq *http.Request, err error) {
208         updatedReq = (&http.Request{
209                 Method:        req.Method,
210                 URL:           req.URL,
211                 Header:        req.Header,
212                 Body:          req.Body,
213                 ContentLength: req.ContentLength,
214                 Host:          req.Host,
215         }).WithContext(req.Context())
216
217         creds := auth.NewCredentials()
218         creds.LoadTokensFromHTTPRequest(updatedReq)
219         if len(creds.Tokens) == 0 && updatedReq.Header.Get("Content-Type") == "application/x-www-form-encoded" {
220                 // Override ParseForm's 10MiB limit by ensuring
221                 // req.Body is a *http.maxBytesReader.
222                 updatedReq.Body = http.MaxBytesReader(nil, updatedReq.Body, 1<<28) // 256MiB. TODO: use MaxRequestSize from discovery doc or config.
223                 if err := creds.LoadTokensFromHTTPRequestBody(updatedReq); err != nil {
224                         return nil, err
225                 }
226                 // Replace req.Body with a buffer that re-encodes the
227                 // form without api_token, in case we end up
228                 // forwarding the request.
229                 if updatedReq.PostForm != nil {
230                         updatedReq.PostForm.Del("api_token")
231                 }
232                 updatedReq.Body = ioutil.NopCloser(bytes.NewBufferString(updatedReq.PostForm.Encode()))
233         }
234         if len(creds.Tokens) == 0 {
235                 return updatedReq, nil
236         }
237
238         token, err := auth.SaltToken(creds.Tokens[0], remote)
239
240         if err == auth.ErrObsoleteToken {
241                 // If the token exists in our own database, salt it
242                 // for the remote. Otherwise, assume it was issued by
243                 // the remote, and pass it through unmodified.
244                 currentUser, err := h.validateAPItoken(req, creds.Tokens[0])
245                 if err == sql.ErrNoRows {
246                         // Not ours; pass through unmodified.
247                         token = creds.Tokens[0]
248                 } else if err != nil {
249                         return nil, err
250                 } else {
251                         // Found; make V2 version and salt it.
252                         token, err = auth.SaltToken(currentUser.Authorization.TokenV2(), remote)
253                         if err != nil {
254                                 return nil, err
255                         }
256                 }
257         } else if err != nil {
258                 return nil, err
259         }
260         updatedReq.Header = http.Header{}
261         for k, v := range req.Header {
262                 if k != "Authorization" {
263                         updatedReq.Header[k] = v
264                 }
265         }
266         updatedReq.Header.Set("Authorization", "Bearer "+token)
267
268         // Remove api_token=... from the the query string, in case we
269         // end up forwarding the request.
270         if values, err := url.ParseQuery(updatedReq.URL.RawQuery); err != nil {
271                 return nil, err
272         } else if _, ok := values["api_token"]; ok {
273                 delete(values, "api_token")
274                 updatedReq.URL = &url.URL{
275                         Scheme:   req.URL.Scheme,
276                         Host:     req.URL.Host,
277                         Path:     req.URL.Path,
278                         RawPath:  req.URL.RawPath,
279                         RawQuery: values.Encode(),
280                 }
281         }
282         return updatedReq, nil
283 }