Patchwork D3242: wireproto: define and implement HTTP handshake to upgrade protocol

login
register
mail settings
Submitter phabricator
Date April 11, 2018, 5:02 p.m.
Message ID <562b2a529ccf43e80a086c8f864cc8e6@localhost.localdomain>
Download mbox | patch
Permalink /patch/30728/
State Not Applicable
Headers show

Comments

phabricator - April 11, 2018, 5:02 p.m.
This revision was automatically updated to reflect the committed changes.
Closed by commit rHG734515aca84d: wireproto: define and implement HTTP handshake to upgrade protocol (authored by indygreg, committed by ).

REPOSITORY
  rHG Mercurial

CHANGES SINCE LAST UPDATE
  https://phab.mercurial-scm.org/D3242?vs=7960&id=8003

REVISION DETAIL
  https://phab.mercurial-scm.org/D3242

AFFECTED FILES
  mercurial/debugcommands.py
  mercurial/help/internals/wireprotocol.txt
  mercurial/wireprotoserver.py
  mercurial/wireprotov2server.py
  tests/test-wireproto-command-capabilities.t

CHANGE DETAILS




To: indygreg, #hg-reviewers, durin42
Cc: mercurial-devel

Patch

diff --git a/tests/test-wireproto-command-capabilities.t b/tests/test-wireproto-command-capabilities.t
--- a/tests/test-wireproto-command-capabilities.t
+++ b/tests/test-wireproto-command-capabilities.t
@@ -1,11 +1,201 @@ 
   $ . $TESTDIR/wireprotohelpers.sh
 
   $ hg init server
-  $ enablehttpv2 server
+  $ hg -R server serve -p $HGPORT -d --pid-file hg.pid -E error.log
+  $ cat hg.pid > $DAEMON_PIDS
+
+A normal capabilities request is serviced for version 1
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >     user-agent: test
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 Script output follows\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-0.1\r\n
+  s>     Content-Length: 458\r\n
+  s>     \r\n
+  s>     batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+
+A proper request without the API server enabled returns the legacy response
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: foo
+  >    x-hgproto-1: cbor
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgproto-1: cbor\r\n
+  s>     x-hgupgrade-1: foo\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 Script output follows\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-0.1\r\n
+  s>     Content-Length: 458\r\n
+  s>     \r\n
+  s>     batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+
+Restart with just API server enabled. This enables serving the new format.
+
+  $ killdaemons.py
+  $ cat error.log
+
+  $ cat >> server/.hg/hgrc << EOF
+  > [experimental]
+  > web.apiserver = true
+  > EOF
+
   $ hg -R server serve -p $HGPORT -d --pid-file hg.pid -E error.log
   $ cat hg.pid > $DAEMON_PIDS
 
-capabilities request returns an array of capability strings
+X-HgUpgrade-<N> without CBOR advertisement uses legacy response
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: foo bar
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgupgrade-1: foo bar\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 Script output follows\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-0.1\r\n
+  s>     Content-Length: 458\r\n
+  s>     \r\n
+  s>     batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+
+X-HgUpgrade-<N> without known serialization in X-HgProto-<N> uses legacy response
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: foo bar
+  >    x-hgproto-1: some value
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgproto-1: some value\r\n
+  s>     x-hgupgrade-1: foo bar\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 Script output follows\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-0.1\r\n
+  s>     Content-Length: 458\r\n
+  s>     \r\n
+  s>     batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+
+X-HgUpgrade-<N> + X-HgProto-<N> headers trigger new response format
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: foo bar
+  >    x-hgproto-1: cbor
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgproto-1: cbor\r\n
+  s>     x-hgupgrade-1: foo bar\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 OK\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-cbor\r\n
+  s>     Content-Length: 496\r\n
+  s>     \r\n
+  s>     \xa3Dapis\xa0GapibaseDapi/Nv1capabilitiesY\x01\xcabatch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+  cbor> {b'apibase': b'api/', b'apis': {}, b'v1capabilities': b'batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash'}
+
+Restart server to enable HTTPv2
+
+  $ killdaemons.py
+  $ enablehttpv2 server
+  $ hg -R server serve -p $HGPORT -d --pid-file hg.pid -E error.log
+
+Only requested API services are returned
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: foo bar
+  >    x-hgproto-1: cbor
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgproto-1: cbor\r\n
+  s>     x-hgupgrade-1: foo bar\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 OK\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-cbor\r\n
+  s>     Content-Length: 496\r\n
+  s>     \r\n
+  s>     \xa3Dapis\xa0GapibaseDapi/Nv1capabilitiesY\x01\xcabatch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+  cbor> {b'apibase': b'api/', b'apis': {}, b'v1capabilities': b'batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash'}
+
+Request for HTTPv2 service returns information about it
+
+  $ sendhttpraw << EOF
+  > httprequest GET ?cmd=capabilities
+  >    user-agent: test
+  >    x-hgupgrade-1: exp-http-v2-0001 foo bar
+  >    x-hgproto-1: cbor
+  > EOF
+  using raw connection to peer
+  s>     GET /?cmd=capabilities HTTP/1.1\r\n
+  s>     Accept-Encoding: identity\r\n
+  s>     user-agent: test\r\n
+  s>     x-hgproto-1: cbor\r\n
+  s>     x-hgupgrade-1: exp-http-v2-0001 foo bar\r\n
+  s>     host: $LOCALIP:$HGPORT\r\n (glob)
+  s>     \r\n
+  s> makefile('rb', None)
+  s>     HTTP/1.1 200 OK\r\n
+  s>     Server: testing stub value\r\n
+  s>     Date: $HTTP_DATE$\r\n
+  s>     Content-Type: application/mercurial-cbor\r\n
+  s>     Content-Length: 879\r\n
+  s>     \r\n
+  s>     \xa3Dapis\xa1Pexp-http-v2-0001\xa2Hcommands\xa7Eheads\xa2Dargs\xa1Jpubliconly\xf4Kpermissions\x81DpullEknown\xa2Dargs\xa1Enodes\x81HdeadbeefKpermissions\x81DpullFlookup\xa2Dargs\xa1CkeyCfooKpermissions\x81DpullGpushkey\xa2Dargs\xa4CkeyCkeyCnewCnewColdColdInamespaceBnsKpermissions\x81DpushHlistkeys\xa2Dargs\xa1InamespaceBnsKpermissions\x81DpullIbranchmap\xa2Dargs\xa0Kpermissions\x81DpullLcapabilities\xa2Dargs\xa0Kpermissions\x81DpullKcompression\x82\xa1DnameDzstd\xa1DnameDzlibGapibaseDapi/Nv1capabilitiesY\x01\xcabatch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash
+  cbor> {b'apibase': b'api/', b'apis': {b'exp-http-v2-0001': {b'commands': {b'branchmap': {b'args': {}, b'permissions': [b'pull']}, b'capabilities': {b'args': {}, b'permissions': [b'pull']}, b'heads': {b'args': {b'publiconly': False}, b'permissions': [b'pull']}, b'known': {b'args': {b'nodes': [b'deadbeef']}, b'permissions': [b'pull']}, b'listkeys': {b'args': {b'namespace': b'ns'}, b'permissions': [b'pull']}, b'lookup': {b'args': {b'key': b'foo'}, b'permissions': [b'pull']}, b'pushkey': {b'args': {b'key': b'key', b'namespace': b'ns', b'new': b'new', b'old': b'old'}, b'permissions': [b'push']}}, b'compression': [{b'name': b'zstd'}, {b'name': b'zlib'}]}}, b'v1capabilities': b'batch branchmap $USUAL_BUNDLE2_CAPS_SERVER$ changegroupsubset compression=$BUNDLE2_COMPRESSIONS$ getbundle httpheader=1024 httpmediatype=0.1rx,0.1tx,0.2tx known lookup pushkey streamreqs=generaldelta,revlogv1 unbundle=HG10GZ,HG10BZ,HG10UN unbundlehash'}
+
+capabilities command returns expected info
 
   $ sendhttpv2peer << EOF
   > command capabilities
diff --git a/mercurial/wireprotov2server.py b/mercurial/wireprotov2server.py
--- a/mercurial/wireprotov2server.py
+++ b/mercurial/wireprotov2server.py
@@ -365,6 +365,11 @@ 
     def checkperm(self, perm):
         raise NotImplementedError
 
+def httpv2apidescriptor(req, repo):
+    proto = httpv2protocolhandler(req, repo.ui)
+
+    return _capabilitiesv2(repo, proto)
+
 def _capabilitiesv2(repo, proto):
     """Obtain the set of capabilities for version 2 transports.
 
diff --git a/mercurial/wireprotoserver.py b/mercurial/wireprotoserver.py
--- a/mercurial/wireprotoserver.py
+++ b/mercurial/wireprotoserver.py
@@ -12,6 +12,9 @@ 
 import threading
 
 from .i18n import _
+from .thirdparty import (
+    cbor,
+)
 from .thirdparty.zope import (
     interface as zi,
 )
@@ -230,6 +233,18 @@ 
 
     return True
 
+def _availableapis(repo):
+    apis = set()
+
+    # Registered APIs are made available via config options of the name of
+    # the protocol.
+    for k, v in API_HANDLERS.items():
+        section, option = v['config']
+        if repo.ui.configbool(section, option):
+            apis.add(k)
+
+    return apis
+
 def handlewsgiapirequest(rctx, req, res, checkperm):
     """Handle requests to /api/*."""
     assert req.dispatchparts[0] == b'api'
@@ -247,13 +262,7 @@ 
     # The URL space is /api/<protocol>/*. The structure of URLs under varies
     # by <protocol>.
 
-    # Registered APIs are made available via config options of the name of
-    # the protocol.
-    availableapis = set()
-    for k, v in API_HANDLERS.items():
-        section, option = v['config']
-        if repo.ui.configbool(section, option):
-            availableapis.add(k)
+    availableapis = _availableapis(repo)
 
     # Requests to /api/ list available APIs.
     if req.dispatchparts == [b'api']:
@@ -287,10 +296,21 @@ 
                                    req.dispatchparts[2:])
 
 # Maps API name to metadata so custom API can be registered.
+# Keys are:
+#
+# config
+#    Config option that controls whether service is enabled.
+# handler
+#    Callable receiving (rctx, req, res, checkperm, urlparts) that is called
+#    when a request to this API is received.
+# apidescriptor
+#    Callable receiving (req, repo) that is called to obtain an API
+#    descriptor for this service. The response must be serializable to CBOR.
 API_HANDLERS = {
     wireprotov2server.HTTPV2: {
         'config': ('experimental', 'web.api.http-v2'),
         'handler': wireprotov2server.handlehttpv2request,
+        'apidescriptor': wireprotov2server.httpv2apidescriptor,
     },
 }
 
@@ -327,6 +347,54 @@ 
     opts = {'level': ui.configint('server', 'zliblevel')}
     return HGTYPE, util.compengines['zlib'], opts
 
+def processcapabilitieshandshake(repo, req, res, proto):
+    """Called during a ?cmd=capabilities request.
+
+    If the client is advertising support for a newer protocol, we send
+    a CBOR response with information about available services. If no
+    advertised services are available, we don't handle the request.
+    """
+    # Fall back to old behavior unless the API server is enabled.
+    if not repo.ui.configbool('experimental', 'web.apiserver'):
+        return False
+
+    clientapis = decodevaluefromheaders(req, b'X-HgUpgrade')
+    protocaps = decodevaluefromheaders(req, b'X-HgProto')
+    if not clientapis or not protocaps:
+        return False
+
+    # We currently only support CBOR responses.
+    protocaps = set(protocaps.split(' '))
+    if b'cbor' not in protocaps:
+        return False
+
+    descriptors = {}
+
+    for api in sorted(set(clientapis.split()) & _availableapis(repo)):
+        handler = API_HANDLERS[api]
+
+        descriptorfn = handler.get('apidescriptor')
+        if not descriptorfn:
+            continue
+
+        descriptors[api] = descriptorfn(req, repo)
+
+    v1caps = wireproto.dispatch(repo, proto, 'capabilities')
+    assert isinstance(v1caps, wireprototypes.bytesresponse)
+
+    m = {
+        # TODO allow this to be configurable.
+        'apibase': 'api/',
+        'apis': descriptors,
+        'v1capabilities': v1caps.data,
+    }
+
+    res.status = b'200 OK'
+    res.headers[b'Content-Type'] = b'application/mercurial-cbor'
+    res.setbodybytes(cbor.dumps(m, canonical=True))
+
+    return True
+
 def _callhttp(repo, req, res, proto, cmd):
     # Avoid cycle involving hg module.
     from .hgweb import common as hgwebcommon
@@ -363,6 +431,12 @@ 
 
     proto.checkperm(wireproto.commands[cmd].permission)
 
+    # Possibly handle a modern client wanting to switch protocols.
+    if (cmd == 'capabilities' and
+        processcapabilitieshandshake(repo, req, res, proto)):
+
+        return
+
     rsp = wireproto.dispatch(repo, proto, cmd)
 
     if isinstance(rsp, bytes):
diff --git a/mercurial/help/internals/wireprotocol.txt b/mercurial/help/internals/wireprotocol.txt
--- a/mercurial/help/internals/wireprotocol.txt
+++ b/mercurial/help/internals/wireprotocol.txt
@@ -42,8 +42,44 @@ 
 The client sends a ``capabilities`` command request (``?cmd=capabilities``)
 as soon as HTTP requests may be issued.
 
-The server responds with a capabilities string, which the client parses to
-learn about the server's abilities.
+By default, the server responds with a version 1 capabilities string, which
+the client parses to learn about the server's abilities. The ``Content-Type``
+for this response is ``application/mercurial-0.1`` or
+``application/mercurial-0.2`` depending on whether the client advertised
+support for version ``0.2`` in its request. (Clients aren't supposed to
+advertise support for ``0.2`` until the capabilities response indicates
+the server's support for that media type. However, a client could
+conceivably cache this metadata and issue the capabilities request in such
+a way to elicit an ``application/mercurial-0.2`` response.)
+
+Clients wishing to switch to a newer API service may send an
+``X-HgUpgrade-<X>`` header containing a space-delimited list of API service
+names the client is capable of speaking. The request MUST also include an
+``X-HgProto-<X>`` header advertising a known serialization format for the
+response. ``cbor`` is currently the only defined serialization format.
+
+If the request contains these headers, the response ``Content-Type`` MAY
+be for a different media type. e.g. ``application/mercurial-cbor`` if the
+client advertises support for CBOR.
+
+The response MUST be deserializable to a map with the following keys:
+
+apibase
+   URL path to API services, relative to the repository root. e.g. ``api/``.
+
+apis
+   A map of API service names to API descriptors. An API descriptor contains
+   more details about that API. In the case of the HTTP Version 2 Transport,
+   it will be the normal response to a ``capabilities`` command.
+
+   Only the services advertised by the client that are also available on
+   the server are advertised.
+
+v1capabilities
+   The capabilities string that would be returned by a version 1 response.
+
+The client can then inspect the server-advertised APIs and decide which
+API to use, including continuing to use the HTTP Version 1 Transport.
 
 HTTP Version 1 Transport
 ------------------------
@@ -123,6 +159,9 @@ 
 The content of the HTTP response body typically holds text describing the
 error.
 
+The ``application/mercurial-cbor`` media type indicates a CBOR payload
+and should be interpreted as identical to ``application/cbor``.
+
 Behavior of media types is further described in the ``Content Negotiation``
 section below.
 
@@ -1252,6 +1291,12 @@ 
    Indicates the client supports receiving ``application/mercurial-0.2``
    responses.
 
+cbor
+   Indicates the client supports receiving ``application/mercurial-cbor``
+   responses.
+
+   (Only intended to be used with version 2 transports.)
+
 comp
    Indicates compression formats the client can decode. Value is a list of
    comma delimited strings identifying compression formats ordered from
diff --git a/mercurial/debugcommands.py b/mercurial/debugcommands.py
--- a/mercurial/debugcommands.py
+++ b/mercurial/debugcommands.py
@@ -33,6 +33,9 @@ 
     nullrev,
     short,
 )
+from .thirdparty import (
+    cbor,
+)
 from . import (
     bundle2,
     changegroup,
@@ -3045,9 +3048,14 @@ 
             req.get_method = lambda: method
 
             try:
-                opener.open(req).read()
+                res = opener.open(req)
+                body = res.read()
             except util.urlerr.urlerror as e:
                 e.read()
+                continue
+
+            if res.headers.get('Content-Type') == 'application/mercurial-cbor':
+                ui.write(_('cbor> %s\n') % stringutil.pprint(cbor.loads(body)))
 
         elif action == 'close':
             peer.close()