gotosocial/internal/transport/finger.go

233 lines
8 KiB
Go
Raw Normal View History

// GoToSocial
// Copyright (C) GoToSocial Authors admin@gotosocial.org
// SPDX-License-Identifier: AGPL-3.0-or-later
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package transport
import (
"context"
"encoding/xml"
"fmt"
"io"
"net/http"
"net/url"
apimodel "github.com/superseriousbusiness/gotosocial/internal/api/model"
apiutil "github.com/superseriousbusiness/gotosocial/internal/api/util"
"github.com/superseriousbusiness/gotosocial/internal/gtserror"
)
// webfingerURLFor returns the URL to try a webfinger request against, as
// well as if the URL was retrieved from cache. When the URL is retrieved
// from cache we don't have to try and do host-meta discovery
func (t *transport) webfingerURLFor(targetDomain string) (string, bool) {
url := "https://" + targetDomain + "/.well-known/webfinger"
wc := t.controller.state.Caches.GTS.Webfinger
// We're doing the manual locking/unlocking here to be able to
// safely call Cache.Get instead of Get, as the latter updates the
// item expiry which we don't want to do here
wc.Lock()
item, ok := wc.Cache.Get(targetDomain)
wc.Unlock()
if ok {
url = item.Value
}
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
return url, ok
}
func prepWebfingerReq(ctx context.Context, loc, domain, username string) (*http.Request, error) {
req, err := http.NewRequestWithContext(ctx, http.MethodGet, loc, nil)
if err != nil {
return nil, err
}
value := url.QueryEscape("acct:" + username + "@" + domain)
req.URL.RawQuery = "resource=" + value
// Prefer application/jrd+json, fall back to application/json.
// See https://www.rfc-editor.org/rfc/rfc7033#section-10.2.
//
// Some implementations don't handle multiple accept headers properly,
// including Gin itself. So concat the accept header with a comma
// instead which seems to work reliably
req.Header.Add("Accept", string(apiutil.AppJRDJSON)+","+string(apiutil.AppJSON))
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
req.Header.Set("Host", req.URL.Host)
return req, nil
}
func (t *transport) Finger(ctx context.Context, targetUsername string, targetDomain string) ([]byte, error) {
// Generate new GET request
url, cached := t.webfingerURLFor(targetDomain)
req, err := prepWebfingerReq(ctx, url, targetDomain, targetUsername)
if err != nil {
return nil, err
}
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
// Perform the HTTP request
rsp, err := t.GET(req)
if err != nil {
return nil, err
}
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
defer rsp.Body.Close()
// Check if the request succeeded so we can bail out early or if we explicitly
// got a "this resource is gone" response which will happen when a user has
// deleted the account
if rsp.StatusCode == http.StatusOK || rsp.StatusCode == http.StatusGone {
if cached {
// If we got a response we consider successful on a cached URL, i.e one set
// by us later on when a host-meta based webfinger request succeeded, set it
// again here to renew the TTL
t.controller.state.Caches.GTS.Webfinger.Set(targetDomain, url)
}
if rsp.StatusCode == http.StatusGone {
return nil, fmt.Errorf("account has been deleted/is gone")
}
// Ensure that the incoming request content-type is expected.
if ct := rsp.Header.Get("Content-Type"); !apiutil.JSONJRDContentType(ct) {
err := gtserror.Newf("non webfinger type response: %s", ct)
return nil, gtserror.SetMalformed(err)
}
return io.ReadAll(rsp.Body)
}
// From here on out, we're handling different failure scenarios and
// deciding whether we should do a host-meta based fallback or not
// Response status codes >= 500 are returned as errors by the wrapped HTTP client.
//
// if (rsp.StatusCode >= 500 && rsp.StatusCode < 600) || cached {
// In case we got a 5xx, bail out irrespective of if the value
// was cached or not. The target may be broken or be signalling
// us to back-off.
//
// If it's any error but the URL was cached, bail out too
// return nil, gtserror.NewResponseError(rsp)
// }
// So far we've failed to get a successful response from the expected
// webfinger endpoint. Lets try and discover the webfinger endpoint
// through /.well-known/host-meta
host, err := t.webfingerFromHostMeta(ctx, targetDomain)
if err != nil {
return nil, fmt.Errorf("failed to discover webfinger URL fallback for: %s through host-meta: %w", targetDomain, err)
}
// Check if the original and host-meta URL are the same. If they
// are there's no sense in us trying the request again as it just
// failed
if host == url {
return nil, fmt.Errorf("webfinger discovery on %s returned endpoint we already tried: %s", targetDomain, host)
}
// Now that we have a different URL for the webfinger
// endpoint, try the request against that endpoint instead
req, err = prepWebfingerReq(ctx, host, targetDomain, targetUsername)
if err != nil {
return nil, err
}
// Perform the HTTP request
rsp, err = t.GET(req)
if err != nil {
return nil, err
}
defer rsp.Body.Close()
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
if rsp.StatusCode != http.StatusOK {
// A HTTP 410 indicates we got a response to our webfinger query, but the resource
// we asked for is gone. This means the endpoint itself is valid and we should
// cache it for future queries to the same domain
if rsp.StatusCode == http.StatusGone {
t.controller.state.Caches.GTS.Webfinger.Set(targetDomain, host)
return nil, fmt.Errorf("account has been deleted/is gone")
}
// We've reached the end of the line here, both the original request
// and our attempt to resolve it through the fallback have failed
return nil, gtserror.NewFromResponse(rsp)
}
[security] transport.Controller{} and transport.Transport{} security and performance improvements (#564) * cache transports in controller by privkey-generated pubkey, add retry logic to transport requests Signed-off-by: kim <grufwub@gmail.com> * update code comments, defer mutex unlocks Signed-off-by: kim <grufwub@gmail.com> * add count to 'performing request' log message Signed-off-by: kim <grufwub@gmail.com> * reduce repeated conversions of same url.URL object Signed-off-by: kim <grufwub@gmail.com> * move worker.Worker to concurrency subpackage, add WorkQueue type, limit transport http client use by WorkQueue Signed-off-by: kim <grufwub@gmail.com> * fix security advisories regarding max outgoing conns, max rsp body size - implemented by a new httpclient.Client{} that wraps an underlying client with a queue to limit connections, and limit reader wrapping a response body with a configured maximum size - update pub.HttpClient args passed around to be this new httpclient.Client{} Signed-off-by: kim <grufwub@gmail.com> * add httpclient tests, move ip validation to separate package + change mechanism Signed-off-by: kim <grufwub@gmail.com> * fix merge conflicts Signed-off-by: kim <grufwub@gmail.com> * use singular mutex in transport rather than separate signer mus Signed-off-by: kim <grufwub@gmail.com> * improved useragent string Signed-off-by: kim <grufwub@gmail.com> * add note regarding missing test Signed-off-by: kim <grufwub@gmail.com> * remove useragent field from transport (instead store in controller) Signed-off-by: kim <grufwub@gmail.com> * shutup linter Signed-off-by: kim <grufwub@gmail.com> * reset other signing headers on each loop iteration Signed-off-by: kim <grufwub@gmail.com> * respect request ctx during retry-backoff sleep period Signed-off-by: kim <grufwub@gmail.com> * use external pkg with docs explaining performance "hack" Signed-off-by: kim <grufwub@gmail.com> * use http package constants instead of string method literals Signed-off-by: kim <grufwub@gmail.com> * add license file headers Signed-off-by: kim <grufwub@gmail.com> * update code comment to match new func names Signed-off-by: kim <grufwub@gmail.com> * updates to user-agent string Signed-off-by: kim <grufwub@gmail.com> * update signed testrig models to fit with new transport logic (instead uses separate signer now) Signed-off-by: kim <grufwub@gmail.com> * fuck you linter Signed-off-by: kim <grufwub@gmail.com>
2022-05-15 12:16:43 +03:00
// Set the URL in cache here, since host-meta told us this should be the
// valid one, it's different from the default and our request to it did
// not fail in any manner
t.controller.state.Caches.GTS.Webfinger.Set(targetDomain, host)
return io.ReadAll(rsp.Body)
}
func (t *transport) webfingerFromHostMeta(ctx context.Context, targetDomain string) (string, error) {
// Build the request for the host-meta endpoint
hmurl := "https://" + targetDomain + "/.well-known/host-meta"
req, err := http.NewRequestWithContext(ctx, http.MethodGet, hmurl, nil)
if err != nil {
return "", err
}
// We're doing XML
req.Header.Add("Accept", string(apiutil.AppXML))
req.Header.Add("Accept", "application/xrd+xml")
req.Header.Set("Host", req.URL.Host)
// Perform the HTTP request
rsp, err := t.GET(req)
if err != nil {
return "", err
}
defer rsp.Body.Close()
// Doesn't look like host-meta is working for this instance
if rsp.StatusCode != http.StatusOK {
return "", fmt.Errorf("GET request for %s failed: %s", req.URL.String(), rsp.Status)
}
// Ensure that the incoming request content-type is expected.
if ct := rsp.Header.Get("Content-Type"); !apiutil.XMLXRDContentType(ct) {
err := gtserror.Newf("non host-meta type response: %s", ct)
return "", gtserror.SetMalformed(err)
}
e := xml.NewDecoder(rsp.Body)
var hm apimodel.HostMeta
if err := e.Decode(&hm); err != nil {
// We got something, but it's not a host-meta document we understand
return "", fmt.Errorf("failed to decode host-meta response for %s at %s: %w", targetDomain, req.URL.String(), err)
}
for _, link := range hm.Link {
// Based on what we currently understand, there should not be more than one
// of these with Rel="lrdd" in a host-meta document
if link.Rel == "lrdd" {
u, err := url.Parse(link.Template)
if err != nil {
return "", fmt.Errorf("lrdd link is not a valid url: %w", err)
}
// Get rid of the query template, we only want the scheme://host/path part
u.RawQuery = ""
urlStr := u.String()
return urlStr, nil
}
}
return "", fmt.Errorf("no webfinger URL found")
}