hgweb: create dedicated type for WSGI responses

We have refactored the request side of WSGI processing into a dedicated
type. Now let's do the same thing for the response side.

We invent a ``wsgiresponse`` type. It takes an instance of a
request (for consulation) and the WSGI application's "start_response"
handler.

The type basically allows setting the HTTP status line, response
headers, and the response body.

The WSGI application calls sendresponse() to start sending output.
Output is emitted as a generator to be fed through the WSGI application.
According to PEP 3333, this is the preferred way for output to be
transmitted. (Our legacy ``wsgirequest`` exposed a write() to send
data. We do not wish to support this API because it isn't recommended
by PEP 3333.)

The wire protocol code has been ported to use the new API.

Differential Revision: https://phab.mercurial-scm.org/D2775

mercurial/hgweb/hgweb_mod.py

@@ -1,438 +1,440
 # hgweb/hgweb_mod.py - Web interface for a repository.
 #
 # Copyright 21 May 2005 - (c) 2005 Jake Edge <[e-mail]>
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import contextlib
 import os
 
 from .common import (
     ErrorResponse,
     HTTP_BAD_REQUEST,
     HTTP_NOT_FOUND,
     HTTP_NOT_MODIFIED,
     HTTP_OK,
     HTTP_SERVER_ERROR,
     caching,
     cspvalues,
     permhooks,
 )
 
 from .. import (
     encoding,
     error,
     formatter,
     hg,
     hook,
     profiling,
     pycompat,
     repoview,
     templatefilters,
     templater,
     ui as uimod,
     util,
     wireprotoserver,
 )
 
 from . import (
     request as requestmod,
     webcommands,
     webutil,
     wsgicgi,
 )
 
 archivespecs = util.sortdict((
     ('zip', ('application/zip', 'zip', '.zip', None)),
     ('gz', ('application/x-gzip', 'tgz', '.tar.gz', None)),
     ('bz2', ('application/x-bzip2', 'tbz2', '.tar.bz2', None)),
 ))
 
 def getstyle(req, configfn, templatepath):
     fromreq = req.form.get('style', [None])[0]
     styles = (
         fromreq,
         configfn('web', 'style'),
         'paper',
     )
     return styles, templater.stylemap(styles, templatepath)
 
 def makebreadcrumb(url, prefix=''):
     '''Return a 'URL breadcrumb' list
 
     A 'URL breadcrumb' is a list of URL-name pairs,
     corresponding to each of the path items on a URL.
     This can be used to create path navigation entries.
     '''
     if url.endswith('/'):
         url = url[:-1]
     if prefix:
         url = '/' + prefix + url
     relpath = url
     if relpath.startswith('/'):
         relpath = relpath[1:]
 
     breadcrumb = []
     urlel = url
     pathitems = [''] + relpath.split('/')
     for pathel in reversed(pathitems):
         if not pathel or not urlel:
             break
         breadcrumb.append({'url': urlel, 'name': pathel})
         urlel = os.path.dirname(urlel)
     return reversed(breadcrumb)
 
 class requestcontext(object):
     """Holds state/context for an individual request.
 
     Servers can be multi-threaded. Holding state on the WSGI application
     is prone to race conditions. Instances of this class exist to hold
     mutable and race-free state for requests.
     """
     def __init__(self, app, repo):
         self.repo = repo
         self.reponame = app.reponame
 
         self.archivespecs = archivespecs
 
         self.maxchanges = self.configint('web', 'maxchanges')
         self.stripecount = self.configint('web', 'stripes')
         self.maxshortchanges = self.configint('web', 'maxshortchanges')
         self.maxfiles = self.configint('web', 'maxfiles')
         self.allowpull = self.configbool('web', 'allow-pull')
 
         # we use untrusted=False to prevent a repo owner from using
         # web.templates in .hg/hgrc to get access to any file readable
         # by the user running the CGI script
         self.templatepath = self.config('web', 'templates', untrusted=False)
 
         # This object is more expensive to build than simple config values.
         # It is shared across requests. The app will replace the object
         # if it is updated. Since this is a reference and nothing should
         # modify the underlying object, it should be constant for the lifetime
         # of the request.
         self.websubtable = app.websubtable
 
         self.csp, self.nonce = cspvalues(self.repo.ui)
 
     # Trust the settings from the .hg/hgrc files by default.
     def config(self, section, name, default=uimod._unset, untrusted=True):
         return self.repo.ui.config(section, name, default,
                                    untrusted=untrusted)
 
     def configbool(self, section, name, default=uimod._unset, untrusted=True):
         return self.repo.ui.configbool(section, name, default,
                                        untrusted=untrusted)
 
     def configint(self, section, name, default=uimod._unset, untrusted=True):
         return self.repo.ui.configint(section, name, default,
                                       untrusted=untrusted)
 
     def configlist(self, section, name, default=uimod._unset, untrusted=True):
         return self.repo.ui.configlist(section, name, default,
                                        untrusted=untrusted)
 
     def archivelist(self, nodeid):
         allowed = self.configlist('web', 'allow_archive')
         for typ, spec in self.archivespecs.iteritems():
             if typ in allowed or self.configbool('web', 'allow%s' % typ):
                 yield {'type': typ, 'extension': spec[2], 'node': nodeid}
 
     def templater(self, wsgireq, req):
         # determine scheme, port and server name
         # this is needed to create absolute urls
         logourl = self.config('web', 'logourl')
         logoimg = self.config('web', 'logoimg')
         staticurl = (self.config('web', 'staticurl')
                      or req.apppath + '/static/')
         if not staticurl.endswith('/'):
             staticurl += '/'
 
         # some functions for the templater
 
         def motd(**map):
             yield self.config('web', 'motd')
 
         # figure out which style to use
 
         vars = {}
         styles, (style, mapfile) = getstyle(wsgireq, self.config,
                                             self.templatepath)
         if style == styles[0]:
             vars['style'] = style
 
         sessionvars = webutil.sessionvars(vars, '?')
 
         if not self.reponame:
             self.reponame = (self.config('web', 'name', '')
                              or wsgireq.env.get('REPO_NAME')
                              or req.apppath or self.repo.root)
 
         def websubfilter(text):
             return templatefilters.websub(text, self.websubtable)
 
         # create the templater
         # TODO: export all keywords: defaults = templatekw.keywords.copy()
         defaults = {
             'url': req.apppath + '/',
             'logourl': logourl,
             'logoimg': logoimg,
             'staticurl': staticurl,
             'urlbase': req.advertisedbaseurl,
             'repo': self.reponame,
             'encoding': encoding.encoding,
             'motd': motd,
             'sessionvars': sessionvars,
             'pathdef': makebreadcrumb(req.apppath),
             'style': style,
             'nonce': self.nonce,
         }
         tres = formatter.templateresources(self.repo.ui, self.repo)
         tmpl = templater.templater.frommapfile(mapfile,
                                                filters={'websub': websubfilter},
                                                defaults=defaults,
                                                resources=tres)
         return tmpl
 
 
 class hgweb(object):
     """HTTP server for individual repositories.
 
     Instances of this class serve HTTP responses for a particular
     repository.
 
     Instances are typically used as WSGI applications.
 
     Some servers are multi-threaded. On these servers, there may
     be multiple active threads inside __call__.
     """
     def __init__(self, repo, name=None, baseui=None):
         if isinstance(repo, str):
             if baseui:
                 u = baseui.copy()
             else:
                 u = uimod.ui.load()
             r = hg.repository(u, repo)
         else:
             # we trust caller to give us a private copy
             r = repo
 
         r.ui.setconfig('ui', 'report_untrusted', 'off', 'hgweb')
         r.baseui.setconfig('ui', 'report_untrusted', 'off', 'hgweb')
         r.ui.setconfig('ui', 'nontty', 'true', 'hgweb')
         r.baseui.setconfig('ui', 'nontty', 'true', 'hgweb')
         # resolve file patterns relative to repo root
         r.ui.setconfig('ui', 'forcecwd', r.root, 'hgweb')
         r.baseui.setconfig('ui', 'forcecwd', r.root, 'hgweb')
         # displaying bundling progress bar while serving feel wrong and may
         # break some wsgi implementation.
         r.ui.setconfig('progress', 'disable', 'true', 'hgweb')
         r.baseui.setconfig('progress', 'disable', 'true', 'hgweb')
         self._repos = [hg.cachedlocalrepo(self._webifyrepo(r))]
         self._lastrepo = self._repos[0]
         hook.redirect(True)
         self.reponame = name
 
     def _webifyrepo(self, repo):
         repo = getwebview(repo)
         self.websubtable = webutil.getwebsubs(repo)
         return repo
 
     @contextlib.contextmanager
     def _obtainrepo(self):
         """Obtain a repo unique to the caller.
 
         Internally we maintain a stack of cachedlocalrepo instances
         to be handed out. If one is available, we pop it and return it,
         ensuring it is up to date in the process. If one is not available,
         we clone the most recently used repo instance and return it.
 
         It is currently possible for the stack to grow without bounds
         if the server allows infinite threads. However, servers should
         have a thread limit, thus establishing our limit.
         """
         if self._repos:
             cached = self._repos.pop()
             r, created = cached.fetch()
         else:
             cached = self._lastrepo.copy()
             r, created = cached.fetch()
         if created:
             r = self._webifyrepo(r)
 
         self._lastrepo = cached
         self.mtime = cached.mtime
         try:
             yield r
         finally:
             self._repos.append(cached)
 
     def run(self):
         """Start a server from CGI environment.
 
         Modern servers should be using WSGI and should avoid this
         method, if possible.
         """
         if not encoding.environ.get('GATEWAY_INTERFACE',
                                     '').startswith("CGI/1."):
             raise RuntimeError("This function is only intended to be "
                                "called while running as a CGI script.")
         wsgicgi.launch(self)
 
     def __call__(self, env, respond):
         """Run the WSGI application.
 
         This may be called by multiple threads.
         """
         req = requestmod.wsgirequest(env, respond)
         return self.run_wsgi(req)
 
     def run_wsgi(self, wsgireq):
         """Internal method to run the WSGI application.
 
         This is typically only called by Mercurial. External consumers
         should be using instances of this class as the WSGI application.
         """
         with self._obtainrepo() as repo:
             profile = repo.ui.configbool('profiling', 'enabled')
             with profiling.profile(repo.ui, enabled=profile):
                 for r in self._runwsgi(wsgireq, repo):
                     yield r
 
     def _runwsgi(self, wsgireq, repo):
         req = wsgireq.req
+        res = wsgireq.res
         rctx = requestcontext(self, repo)
 
         # This state is global across all threads.
         encoding.encoding = rctx.config('web', 'encoding')
         rctx.repo.ui.environ = wsgireq.env
 
         if rctx.csp:
             # hgwebdir may have added CSP header. Since we generate our own,
             # replace it.
             wsgireq.headers = [h for h in wsgireq.headers
                                if h[0] != 'Content-Security-Policy']
             wsgireq.headers.append(('Content-Security-Policy', rctx.csp))
+            res.headers['Content-Security-Policy'] = rctx.csp
 
-        handled, res = wireprotoserver.handlewsgirequest(
-            rctx, wsgireq, req, self.check_perm)
+        handled = wireprotoserver.handlewsgirequest(
+            rctx, wsgireq, req, res, self.check_perm)
         if handled:
-            return res
+            return res.sendresponse()
 
         if req.havepathinfo:
             query = req.dispatchpath
         else:
             query = req.querystring.partition('&')[0].partition(';')[0]
 
         # translate user-visible url structure to internal structure
 
         args = query.split('/', 2)
         if 'cmd' not in wsgireq.form and args and args[0]:
             cmd = args.pop(0)
             style = cmd.rfind('-')
             if style != -1:
                 wsgireq.form['style'] = [cmd[:style]]
                 cmd = cmd[style + 1:]
 
             # avoid accepting e.g. style parameter as command
             if util.safehasattr(webcommands, cmd):
                 wsgireq.form['cmd'] = [cmd]
 
             if cmd == 'static':
                 wsgireq.form['file'] = ['/'.join(args)]
             else:
                 if args and args[0]:
                     node = args.pop(0).replace('%2F', '/')
                     wsgireq.form['node'] = [node]
                 if args:
                     wsgireq.form['file'] = args
 
             ua = req.headers.get('User-Agent', '')
             if cmd == 'rev' and 'mercurial' in ua:
                 wsgireq.form['style'] = ['raw']
 
             if cmd == 'archive':
                 fn = wsgireq.form['node'][0]
                 for type_, spec in rctx.archivespecs.iteritems():
                     ext = spec[2]
                     if fn.endswith(ext):
                         wsgireq.form['node'] = [fn[:-len(ext)]]
                         wsgireq.form['type'] = [type_]
         else:
             cmd = wsgireq.form.get('cmd', [''])[0]
 
         # process the web interface request
 
         try:
             tmpl = rctx.templater(wsgireq, req)
             ctype = tmpl('mimetype', encoding=encoding.encoding)
             ctype = templater.stringify(ctype)
 
             # check read permissions non-static content
             if cmd != 'static':
                 self.check_perm(rctx, wsgireq, None)
 
             if cmd == '':
                 wsgireq.form['cmd'] = [tmpl.cache['default']]
                 cmd = wsgireq.form['cmd'][0]
 
             # Don't enable caching if using a CSP nonce because then it wouldn't
             # be a nonce.
             if rctx.configbool('web', 'cache') and not rctx.nonce:
                 caching(self, wsgireq) # sets ETag header or raises NOT_MODIFIED
             if cmd not in webcommands.__all__:
                 msg = 'no such method: %s' % cmd
                 raise ErrorResponse(HTTP_BAD_REQUEST, msg)
             elif cmd == 'file' and 'raw' in wsgireq.form.get('style', []):
                 rctx.ctype = ctype
                 content = webcommands.rawfile(rctx, wsgireq, tmpl)
             else:
                 content = getattr(webcommands, cmd)(rctx, wsgireq, tmpl)
                 wsgireq.respond(HTTP_OK, ctype)
 
             return content
 
         except (error.LookupError, error.RepoLookupError) as err:
             wsgireq.respond(HTTP_NOT_FOUND, ctype)
             msg = pycompat.bytestr(err)
             if (util.safehasattr(err, 'name') and
                 not isinstance(err,  error.ManifestLookupError)):
                 msg = 'revision not found: %s' % err.name
             return tmpl('error', error=msg)
         except (error.RepoError, error.RevlogError) as inst:
             wsgireq.respond(HTTP_SERVER_ERROR, ctype)
             return tmpl('error', error=pycompat.bytestr(inst))
         except ErrorResponse as inst:
             wsgireq.respond(inst, ctype)
             if inst.code == HTTP_NOT_MODIFIED:
                 # Not allowed to return a body on a 304
                 return ['']
             return tmpl('error', error=pycompat.bytestr(inst))
 
     def check_perm(self, rctx, req, op):
         for permhook in permhooks:
             permhook(rctx, req, op)
 
 def getwebview(repo):
     """The 'web.view' config controls changeset filter to hgweb. Possible
     values are ``served``, ``visible`` and ``all``. Default is ``served``.
     The ``served`` filter only shows changesets that can be pulled from the
     hgweb instance.  The``visible`` filter includes secret changesets but
     still excludes "hidden" one.
 
     See the repoview module for details.
 
     The option has been around undocumented since Mercurial 2.5, but no
     user ever asked about it. So we better keep it undocumented for now."""
     # experimental config: web.view
     viewconfig = repo.ui.config('web', 'view', untrusted=True)
     if viewconfig == 'all':
         return repo.unfiltered()
     elif viewconfig in repoview.filtertable:
         return repo.filtered(viewconfig)
     else:
         return repo.filtered('served')

mercurial/hgweb/request.py

@@ -1,339 +1,463
 # hgweb/request.py - An http request from either CGI or the standalone server.
 #
 # Copyright 21 May 2005 - (c) 2005 Jake Edge <[e-mail]>
 # Copyright 2005, 2006 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import errno
 import socket
 import wsgiref.headers as wsgiheaders
 #import wsgiref.validate
 
 from .common import (
     ErrorResponse,
     HTTP_NOT_MODIFIED,
     statusmessage,
 )
 
 from ..thirdparty import (
     attr,
 )
 from .. import (
+    error,
     pycompat,
     util,
 )
 
 @attr.s(frozen=True)
 class parsedrequest(object):
     """Represents a parsed WSGI request.
 
     Contains both parsed parameters as well as a handle on the input stream.
     """
 
     # Request method.
     method = attr.ib()
     # Full URL for this request.
     url = attr.ib()
     # URL without any path components. Just <proto>://<host><port>.
     baseurl = attr.ib()
     # Advertised URL. Like ``url`` and ``baseurl`` but uses SERVER_NAME instead
     # of HTTP: Host header for hostname. This is likely what clients used.
     advertisedurl = attr.ib()
     advertisedbaseurl = attr.ib()
     # WSGI application path.
     apppath = attr.ib()
     # List of path parts to be used for dispatch.
     dispatchparts = attr.ib()
     # URL path component (no query string) used for dispatch.
     dispatchpath = attr.ib()
     # Whether there is a path component to this request. This can be true
     # when ``dispatchpath`` is empty due to REPO_NAME muckery.
     havepathinfo = attr.ib()
     # Raw query string (part after "?" in URL).
     querystring = attr.ib()
     # List of 2-tuples of query string arguments.
     querystringlist = attr.ib()
     # Dict of query string arguments. Values are lists with at least 1 item.
     querystringdict = attr.ib()
     # wsgiref.headers.Headers instance. Operates like a dict with case
     # insensitive keys.
     headers = attr.ib()
     # Request body input stream.
     bodyfh = attr.ib()
 
 def parserequestfromenv(env, bodyfh):
     """Parse URL components from environment variables.
 
     WSGI defines request attributes via environment variables. This function
     parses the environment variables into a data structure.
     """
     # PEP-0333 defines the WSGI spec and is a useful reference for this code.
 
     # We first validate that the incoming object conforms with the WSGI spec.
     # We only want to be dealing with spec-conforming WSGI implementations.
     # TODO enable this once we fix internal violations.
     #wsgiref.validate.check_environ(env)
 
     # PEP-0333 states that environment keys and values are native strings
     # (bytes on Python 2 and str on Python 3). The code points for the Unicode
     # strings on Python 3 must be between \00000-\000FF. We deal with bytes
     # in Mercurial, so mass convert string keys and values to bytes.
     if pycompat.ispy3:
         env = {k.encode('latin-1'): v for k, v in env.iteritems()}
         env = {k: v.encode('latin-1') if isinstance(v, str) else v
                for k, v in env.iteritems()}
 
     # https://www.python.org/dev/peps/pep-0333/#environ-variables defines
     # the environment variables.
     # https://www.python.org/dev/peps/pep-0333/#url-reconstruction defines
     # how URLs are reconstructed.
     fullurl = env['wsgi.url_scheme'] + '://'
     advertisedfullurl = fullurl
 
     def addport(s):
         if env['wsgi.url_scheme'] == 'https':
             if env['SERVER_PORT'] != '443':
                 s += ':' + env['SERVER_PORT']
         else:
             if env['SERVER_PORT'] != '80':
                 s += ':' + env['SERVER_PORT']
 
         return s
 
     if env.get('HTTP_HOST'):
         fullurl += env['HTTP_HOST']
     else:
         fullurl += env['SERVER_NAME']
         fullurl = addport(fullurl)
 
     advertisedfullurl += env['SERVER_NAME']
     advertisedfullurl = addport(advertisedfullurl)
 
     baseurl = fullurl
     advertisedbaseurl = advertisedfullurl
 
     fullurl += util.urlreq.quote(env.get('SCRIPT_NAME', ''))
     advertisedfullurl += util.urlreq.quote(env.get('SCRIPT_NAME', ''))
     fullurl += util.urlreq.quote(env.get('PATH_INFO', ''))
     advertisedfullurl += util.urlreq.quote(env.get('PATH_INFO', ''))
 
     if env.get('QUERY_STRING'):
         fullurl += '?' + env['QUERY_STRING']
         advertisedfullurl += '?' + env['QUERY_STRING']
 
     # When dispatching requests, we look at the URL components (PATH_INFO
     # and QUERY_STRING) after the application root (SCRIPT_NAME). But hgwebdir
     # has the concept of "virtual" repositories. This is defined via REPO_NAME.
     # If REPO_NAME is defined, we append it to SCRIPT_NAME to form a new app
     # root. We also exclude its path components from PATH_INFO when resolving
     # the dispatch path.
 
     apppath = env['SCRIPT_NAME']
 
     if env.get('REPO_NAME'):
         if not apppath.endswith('/'):
             apppath += '/'
 
         apppath += env.get('REPO_NAME')
 
     if 'PATH_INFO' in env:
         dispatchparts = env['PATH_INFO'].strip('/').split('/')
 
         # Strip out repo parts.
         repoparts = env.get('REPO_NAME', '').split('/')
         if dispatchparts[:len(repoparts)] == repoparts:
             dispatchparts = dispatchparts[len(repoparts):]
     else:
         dispatchparts = []
 
     dispatchpath = '/'.join(dispatchparts)
 
     querystring = env.get('QUERY_STRING', '')
 
     # We store as a list so we have ordering information. We also store as
     # a dict to facilitate fast lookup.
     querystringlist = util.urlreq.parseqsl(querystring, keep_blank_values=True)
 
     querystringdict = {}
     for k, v in querystringlist:
         if k in querystringdict:
             querystringdict[k].append(v)
         else:
             querystringdict[k] = [v]
 
     # HTTP_* keys contain HTTP request headers. The Headers structure should
     # perform case normalization for us. We just rewrite underscore to dash
     # so keys match what likely went over the wire.
     headers = []
     for k, v in env.iteritems():
         if k.startswith('HTTP_'):
             headers.append((k[len('HTTP_'):].replace('_', '-'), v))
 
     headers = wsgiheaders.Headers(headers)
 
     # This is kind of a lie because the HTTP header wasn't explicitly
     # sent. But for all intents and purposes it should be OK to lie about
     # this, since a consumer will either either value to determine how many
     # bytes are available to read.
     if 'CONTENT_LENGTH' in env and 'HTTP_CONTENT_LENGTH' not in env:
         headers['Content-Length'] = env['CONTENT_LENGTH']
 
     # TODO do this once we remove wsgirequest.inp, otherwise we could have
     # multiple readers from the underlying input stream.
     #bodyfh = env['wsgi.input']
     #if 'Content-Length' in headers:
     #    bodyfh = util.cappedreader(bodyfh, int(headers['Content-Length']))
 
     return parsedrequest(method=env['REQUEST_METHOD'],
                          url=fullurl, baseurl=baseurl,
                          advertisedurl=advertisedfullurl,
                          advertisedbaseurl=advertisedbaseurl,
                          apppath=apppath,
                          dispatchparts=dispatchparts, dispatchpath=dispatchpath,
                          havepathinfo='PATH_INFO' in env,
                          querystring=querystring,
                          querystringlist=querystringlist,
                          querystringdict=querystringdict,
                          headers=headers,
                          bodyfh=bodyfh)
 
+class wsgiresponse(object):
+    """Represents a response to a WSGI request.
+
+    A response consists of a status line, headers, and a body.
+
+    Consumers must populate the ``status`` and ``headers`` fields and
+    make a call to a ``setbody*()`` method before the response can be
+    issued.
+
+    When it is time to start sending the response over the wire,
+    ``sendresponse()`` is called. It handles emitting the header portion
+    of the response message. It then yields chunks of body data to be
+    written to the peer. Typically, the WSGI application itself calls
+    and returns the value from ``sendresponse()``.
+    """
+
+    def __init__(self, req, startresponse):
+        """Create an empty response tied to a specific request.
+
+        ``req`` is a ``parsedrequest``. ``startresponse`` is the
+        ``start_response`` function passed to the WSGI application.
+        """
+        self._req = req
+        self._startresponse = startresponse
+
+        self.status = None
+        self.headers = wsgiheaders.Headers([])
+
+        self._bodybytes = None
+        self._bodygen = None
+        self._started = False
+
+    def setbodybytes(self, b):
+        """Define the response body as static bytes."""
+        if self._bodybytes is not None or self._bodygen is not None:
+            raise error.ProgrammingError('cannot define body multiple times')
+
+        self._bodybytes = b
+        self.headers['Content-Length'] = '%d' % len(b)
+
+    def setbodygen(self, gen):
+        """Define the response body as a generator of bytes."""
+        if self._bodybytes is not None or self._bodygen is not None:
+            raise error.ProgrammingError('cannot define body multiple times')
+
+        self._bodygen = gen
+
+    def sendresponse(self):
+        """Send the generated response to the client.
+
+        Before this is called, ``status`` must be set and one of
+        ``setbodybytes()`` or ``setbodygen()`` must be called.
+
+        Calling this method multiple times is not allowed.
+        """
+        if self._started:
+            raise error.ProgrammingError('sendresponse() called multiple times')
+
+        self._started = True
+
+        if not self.status:
+            raise error.ProgrammingError('status line not defined')
+
+        if self._bodybytes is None and self._bodygen is None:
+            raise error.ProgrammingError('response body not defined')
+
+        # Various HTTP clients (notably httplib) won't read the HTTP response
+        # until the HTTP request has been sent in full. If servers (us) send a
+        # response before the HTTP request has been fully sent, the connection
+        # may deadlock because neither end is reading.
+        #
+        # We work around this by "draining" the request data before
+        # sending any response in some conditions.
+        drain = False
+        close = False
+
+        # If the client sent Expect: 100-continue, we assume it is smart enough
+        # to deal with the server sending a response before reading the request.
+        # (httplib doesn't do this.)
+        if self._req.headers.get('Expect', '').lower() == '100-continue':
+            pass
+        # Only tend to request methods that have bodies. Strictly speaking,
+        # we should sniff for a body. But this is fine for our existing
+        # WSGI applications.
+        elif self._req.method not in ('POST', 'PUT'):
+            pass
+        else:
+            # If we don't know how much data to read, there's no guarantee
+            # that we can drain the request responsibly. The WSGI
+            # specification only says that servers *should* ensure the
+            # input stream doesn't overrun the actual request. So there's
+            # no guarantee that reading until EOF won't corrupt the stream
+            # state.
+            if not isinstance(self._req.bodyfh, util.cappedreader):
+                close = True
+            else:
+                # We /could/ only drain certain HTTP response codes. But 200 and
+                # non-200 wire protocol responses both require draining. Since
+                # we have a capped reader in place for all situations where we
+                # drain, it is safe to read from that stream. We'll either do
+                # a drain or no-op if we're already at EOF.
+                drain = True
+
+        if close:
+            self.headers['Connection'] = 'Close'
+
+        if drain:
+            assert isinstance(self._req.bodyfh, util.cappedreader)
+            while True:
+                chunk = self._req.bodyfh.read(32768)
+                if not chunk:
+                    break
+
+        self._startresponse(pycompat.sysstr(self.status), self.headers.items())
+        if self._bodybytes:
+            yield self._bodybytes
+        elif self._bodygen:
+            for chunk in self._bodygen:
+                yield chunk
+        else:
+            error.ProgrammingError('do not know how to send body')
+
 class wsgirequest(object):
     """Higher-level API for a WSGI request.
 
     WSGI applications are invoked with 2 arguments. They are used to
     instantiate instances of this class, which provides higher-level APIs
     for obtaining request parameters, writing HTTP output, etc.
     """
     def __init__(self, wsgienv, start_response):
         version = wsgienv[r'wsgi.version']
         if (version < (1, 0)) or (version >= (2, 0)):
             raise RuntimeError("Unknown and unsupported WSGI version %d.%d"
                                % version)
 
         inp = wsgienv[r'wsgi.input']
 
         if r'HTTP_CONTENT_LENGTH' in wsgienv:
             inp = util.cappedreader(inp, int(wsgienv[r'HTTP_CONTENT_LENGTH']))
         elif r'CONTENT_LENGTH' in wsgienv:
             inp = util.cappedreader(inp, int(wsgienv[r'CONTENT_LENGTH']))
 
         self.err = wsgienv[r'wsgi.errors']
         self.threaded = wsgienv[r'wsgi.multithread']
         self.multiprocess = wsgienv[r'wsgi.multiprocess']
         self.run_once = wsgienv[r'wsgi.run_once']
         self.env = wsgienv
         self.req = parserequestfromenv(wsgienv, inp)
         self.form = self.req.querystringdict
+        self.res = wsgiresponse(self.req, start_response)
         self._start_response = start_response
         self.server_write = None
         self.headers = []
 
     def respond(self, status, type, filename=None, body=None):
         if not isinstance(type, str):
             type = pycompat.sysstr(type)
         if self._start_response is not None:
             self.headers.append((r'Content-Type', type))
             if filename:
                 filename = (filename.rpartition('/')[-1]
                             .replace('\\', '\\\\').replace('"', '\\"'))
                 self.headers.append(('Content-Disposition',
                                      'inline; filename="%s"' % filename))
             if body is not None:
                 self.headers.append((r'Content-Length', str(len(body))))
 
             for k, v in self.headers:
                 if not isinstance(v, str):
                     raise TypeError('header value must be string: %r' % (v,))
 
             if isinstance(status, ErrorResponse):
                 self.headers.extend(status.headers)
                 if status.code == HTTP_NOT_MODIFIED:
                     # RFC 2616 Section 10.3.5: 304 Not Modified has cases where
                     # it MUST NOT include any headers other than these and no
                     # body
                     self.headers = [(k, v) for (k, v) in self.headers if
                                     k in ('Date', 'ETag', 'Expires',
                                           'Cache-Control', 'Vary')]
                 status = statusmessage(status.code, pycompat.bytestr(status))
             elif status == 200:
                 status = '200 Script output follows'
             elif isinstance(status, int):
                 status = statusmessage(status)
 
             # Various HTTP clients (notably httplib) won't read the HTTP
             # response until the HTTP request has been sent in full. If servers
             # (us) send a response before the HTTP request has been fully sent,
             # the connection may deadlock because neither end is reading.
             #
             # We work around this by "draining" the request data before
             # sending any response in some conditions.
             drain = False
             close = False
 
             # If the client sent Expect: 100-continue, we assume it is smart
             # enough to deal with the server sending a response before reading
             # the request. (httplib doesn't do this.)
             if self.env.get(r'HTTP_EXPECT', r'').lower() == r'100-continue':
                 pass
             # Only tend to request methods that have bodies. Strictly speaking,
             # we should sniff for a body. But this is fine for our existing
             # WSGI applications.
             elif self.env[r'REQUEST_METHOD'] not in (r'POST', r'PUT'):
                 pass
             else:
                 # If we don't know how much data to read, there's no guarantee
                 # that we can drain the request responsibly. The WSGI
                 # specification only says that servers *should* ensure the
                 # input stream doesn't overrun the actual request. So there's
                 # no guarantee that reading until EOF won't corrupt the stream
                 # state.
                 if not isinstance(self.req.bodyfh, util.cappedreader):
                     close = True
                 else:
                     # We /could/ only drain certain HTTP response codes. But 200
                     # and non-200 wire protocol responses both require draining.
                     # Since we have a capped reader in place for all situations
                     # where we drain, it is safe to read from that stream. We'll
                     # either do a drain or no-op if we're already at EOF.
                     drain = True
 
             if close:
                 self.headers.append((r'Connection', r'Close'))
 
             if drain:
                 assert isinstance(self.req.bodyfh, util.cappedreader)
                 while True:
                     chunk = self.req.bodyfh.read(32768)
                     if not chunk:
                         break
 
             self.server_write = self._start_response(
                 pycompat.sysstr(status), self.headers)
             self._start_response = None
             self.headers = []
         if body is not None:
             self.write(body)
             self.server_write = None
 
     def write(self, thing):
         if thing:
             try:
                 self.server_write(thing)
             except socket.error as inst:
                 if inst[0] != errno.ECONNRESET:
                     raise
 
     def flush(self):
         return None
 
 def wsgiapplication(app_maker):
     '''For compatibility with old CGI scripts. A plain hgweb() or hgwebdir()
     can and should now be used as a WSGI application.'''
     application = app_maker()
     def run_wsgi(env, respond):
         return application(env, respond)
     return run_wsgi

mercurial/wireprotoserver.py

@@ -1,649 +1,655
 # Copyright 21 May 2005 - (c) 2005 Jake Edge <[e-mail]>
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import contextlib
 import struct
 import sys
 import threading
 
 from .i18n import _
 from . import (
     encoding,
     error,
     hook,
     pycompat,
     util,
     wireproto,
     wireprototypes,
 )
 
 stringio = util.stringio
 
 urlerr = util.urlerr
 urlreq = util.urlreq
 
 HTTP_OK = 200
 
 HGTYPE = 'application/mercurial-0.1'
 HGTYPE2 = 'application/mercurial-0.2'
 HGERRTYPE = 'application/hg-error'
 
 SSHV1 = wireprototypes.SSHV1
 SSHV2 = wireprototypes.SSHV2
 
 def decodevaluefromheaders(req, headerprefix):
     """Decode a long value from multiple HTTP request headers.
 
     Returns the value as a bytes, not a str.
     """
     chunks = []
     i = 1
     while True:
         v = req.headers.get(b'%s-%d' % (headerprefix, i))
         if v is None:
             break
         chunks.append(pycompat.bytesurl(v))
         i += 1
 
     return ''.join(chunks)
 
 class httpv1protocolhandler(wireprototypes.baseprotocolhandler):
     def __init__(self, wsgireq, req, ui, checkperm):
         self._wsgireq = wsgireq
         self._req = req
         self._ui = ui
         self._checkperm = checkperm
 
     @property
     def name(self):
         return 'http-v1'
 
     def getargs(self, args):
         knownargs = self._args()
         data = {}
         keys = args.split()
         for k in keys:
             if k == '*':
                 star = {}
                 for key in knownargs.keys():
                     if key != 'cmd' and key not in keys:
                         star[key] = knownargs[key][0]
                 data['*'] = star
             else:
                 data[k] = knownargs[k][0]
         return [data[k] for k in keys]
 
     def _args(self):
         args = util.rapply(pycompat.bytesurl, self._wsgireq.form.copy())
         postlen = int(self._req.headers.get(b'X-HgArgs-Post', 0))
         if postlen:
             args.update(urlreq.parseqs(
                 self._req.bodyfh.read(postlen), keep_blank_values=True))
             return args
 
         argvalue = decodevaluefromheaders(self._req, b'X-HgArg')
         args.update(urlreq.parseqs(argvalue, keep_blank_values=True))
         return args
 
     def forwardpayload(self, fp):
         # Existing clients *always* send Content-Length.
         length = int(self._req.headers[b'Content-Length'])
 
         # If httppostargs is used, we need to read Content-Length
         # minus the amount that was consumed by args.
         length -= int(self._req.headers.get(b'X-HgArgs-Post', 0))
         for s in util.filechunkiter(self._req.bodyfh, limit=length):
             fp.write(s)
 
     @contextlib.contextmanager
     def mayberedirectstdio(self):
         oldout = self._ui.fout
         olderr = self._ui.ferr
 
         out = util.stringio()
 
         try:
             self._ui.fout = out
             self._ui.ferr = out
             yield out
         finally:
             self._ui.fout = oldout
             self._ui.ferr = olderr
 
     def client(self):
         return 'remote:%s:%s:%s' % (
             self._wsgireq.env.get('wsgi.url_scheme') or 'http',
             urlreq.quote(self._wsgireq.env.get('REMOTE_HOST', '')),
             urlreq.quote(self._wsgireq.env.get('REMOTE_USER', '')))
 
     def addcapabilities(self, repo, caps):
         caps.append('httpheader=%d' %
                     repo.ui.configint('server', 'maxhttpheaderlen'))
         if repo.ui.configbool('experimental', 'httppostargs'):
             caps.append('httppostargs')
 
         # FUTURE advertise 0.2rx once support is implemented
         # FUTURE advertise minrx and mintx after consulting config option
         caps.append('httpmediatype=0.1rx,0.1tx,0.2tx')
 
         compengines = wireproto.supportedcompengines(repo.ui, util.SERVERROLE)
         if compengines:
             comptypes = ','.join(urlreq.quote(e.wireprotosupport().name)
                                  for e in compengines)
             caps.append('compression=%s' % comptypes)
 
         return caps
 
     def checkperm(self, perm):
         return self._checkperm(perm)
 
 # This method exists mostly so that extensions like remotefilelog can
 # disable a kludgey legacy method only over http. As of early 2018,
 # there are no other known users, so with any luck we can discard this
 # hook if remotefilelog becomes a first-party extension.
 def iscmd(cmd):
     return cmd in wireproto.commands
 
-def handlewsgirequest(rctx, wsgireq, req, checkperm):
+def handlewsgirequest(rctx, wsgireq, req, res, checkperm):
     """Possibly process a wire protocol request.
 
     If the current request is a wire protocol request, the request is
     processed by this function.
 
     ``wsgireq`` is a ``wsgirequest`` instance.
     ``req`` is a ``parsedrequest`` instance.
+    ``res`` is a ``wsgiresponse`` instance.
 
-    Returns a 2-tuple of (bool, response) where the 1st element indicates
-    whether the request was handled and the 2nd element is a return
-    value for a WSGI application (often a generator of bytes).
+    Returns a bool indicating if the request was serviced. If set, the caller
+    should stop processing the request, as a response has already been issued.
     """
     # Avoid cycle involving hg module.
     from .hgweb import common as hgwebcommon
 
     repo = rctx.repo
 
     # HTTP version 1 wire protocol requests are denoted by a "cmd" query
     # string parameter. If it isn't present, this isn't a wire protocol
     # request.
     if 'cmd' not in req.querystringdict:
-        return False, None
+        return False
 
     cmd = req.querystringdict['cmd'][0]
 
     # The "cmd" request parameter is used by both the wire protocol and hgweb.
     # While not all wire protocol commands are available for all transports,
     # if we see a "cmd" value that resembles a known wire protocol command, we
     # route it to a protocol handler. This is better than routing possible
     # wire protocol requests to hgweb because it prevents hgweb from using
     # known wire protocol commands and it is less confusing for machine
     # clients.
     if not iscmd(cmd):
-        return False, None
+        return False
 
     # The "cmd" query string argument is only valid on the root path of the
     # repo. e.g. ``/?cmd=foo``, ``/repo?cmd=foo``. URL paths within the repo
     # like ``/blah?cmd=foo`` are not allowed. So don't recognize the request
     # in this case. We send an HTTP 404 for backwards compatibility reasons.
     if req.dispatchpath:
-        res = _handlehttperror(
-            hgwebcommon.ErrorResponse(hgwebcommon.HTTP_NOT_FOUND), wsgireq,
-            req)
-
-        return True, res
+        res.status = hgwebcommon.statusmessage(404)
+        res.headers['Content-Type'] = HGTYPE
+        # TODO This is not a good response to issue for this request. This
+        # is mostly for BC for now.
+        res.setbodybytes('0\n%s\n' % b'Not Found')
+        return True
 
     proto = httpv1protocolhandler(wsgireq, req, repo.ui,
                                   lambda perm: checkperm(rctx, wsgireq, perm))
 
     # The permissions checker should be the only thing that can raise an
     # ErrorResponse. It is kind of a layer violation to catch an hgweb
     # exception here. So consider refactoring into a exception type that
     # is associated with the wire protocol.
     try:
-        res = _callhttp(repo, wsgireq, req, proto, cmd)
+        _callhttp(repo, wsgireq, req, res, proto, cmd)
     except hgwebcommon.ErrorResponse as e:
-        res = _handlehttperror(e, wsgireq, req)
+        for k, v in e.headers:
+            res.headers[k] = v
+        res.status = hgwebcommon.statusmessage(e.code, pycompat.bytestr(e))
+        # TODO This response body assumes the failed command was
+        # "unbundle." That assumption is not always valid.
+        res.setbodybytes('0\n%s\n' % pycompat.bytestr(e))
 
-    return True, res
+    return True
 
 def _httpresponsetype(ui, req, prefer_uncompressed):
     """Determine the appropriate response type and compression settings.
 
     Returns a tuple of (mediatype, compengine, engineopts).
     """
     # Determine the response media type and compression engine based
     # on the request parameters.
     protocaps = decodevaluefromheaders(req, 'X-HgProto').split(' ')
 
     if '0.2' in protocaps:
         # All clients are expected to support uncompressed data.
         if prefer_uncompressed:
             return HGTYPE2, util._noopengine(), {}
 
         # Default as defined by wire protocol spec.
         compformats = ['zlib', 'none']
         for cap in protocaps:
             if cap.startswith('comp='):
                 compformats = cap[5:].split(',')
                 break
 
         # Now find an agreed upon compression format.
         for engine in wireproto.supportedcompengines(ui, util.SERVERROLE):
             if engine.wireprotosupport().name in compformats:
                 opts = {}
                 level = ui.configint('server', '%slevel' % engine.name())
                 if level is not None:
                     opts['level'] = level
 
                 return HGTYPE2, engine, opts
 
         # No mutually supported compression format. Fall back to the
         # legacy protocol.
 
     # Don't allow untrusted settings because disabling compression or
     # setting a very high compression level could lead to flooding
     # the server's network or CPU.
     opts = {'level': ui.configint('server', 'zliblevel')}
     return HGTYPE, util.compengines['zlib'], opts
 
-def _callhttp(repo, wsgireq, req, proto, cmd):
+def _callhttp(repo, wsgireq, req, res, proto, cmd):
+    # Avoid cycle involving hg module.
+    from .hgweb import common as hgwebcommon
+
     def genversion2(gen, engine, engineopts):
         # application/mercurial-0.2 always sends a payload header
         # identifying the compression engine.
         name = engine.wireprotosupport().name
         assert 0 < len(name) < 256
         yield struct.pack('B', len(name))
         yield name
 
         for chunk in gen:
             yield chunk
 
+    def setresponse(code, contenttype, bodybytes=None, bodygen=None):
+        if code == HTTP_OK:
+            res.status = '200 Script output follows'
+        else:
+            res.status = hgwebcommon.statusmessage(code)
+
+        res.headers['Content-Type'] = contenttype
+
+        if bodybytes is not None:
+            res.setbodybytes(bodybytes)
+        if bodygen is not None:
+            res.setbodygen(bodygen)
+
     if not wireproto.commands.commandavailable(cmd, proto):
-        wsgireq.respond(HTTP_OK, HGERRTYPE,
-                        body=_('requested wire protocol command is not '
-                               'available over HTTP'))
-        return []
+        setresponse(HTTP_OK, HGERRTYPE,
+                    _('requested wire protocol command is not available over '
+                      'HTTP'))
+        return
 
     proto.checkperm(wireproto.commands[cmd].permission)
 
     rsp = wireproto.dispatch(repo, proto, cmd)
 
     if isinstance(rsp, bytes):
-        wsgireq.respond(HTTP_OK, HGTYPE, body=rsp)
-        return []
+        setresponse(HTTP_OK, HGTYPE, bodybytes=rsp)
     elif isinstance(rsp, wireprototypes.bytesresponse):
-        wsgireq.respond(HTTP_OK, HGTYPE, body=rsp.data)
-        return []
+        setresponse(HTTP_OK, HGTYPE, bodybytes=rsp.data)
     elif isinstance(rsp, wireprototypes.streamreslegacy):
-        gen = rsp.gen
-        wsgireq.respond(HTTP_OK, HGTYPE)
-        return gen
+        setresponse(HTTP_OK, HGTYPE, bodygen=rsp.gen)
     elif isinstance(rsp, wireprototypes.streamres):
         gen = rsp.gen
 
         # This code for compression should not be streamres specific. It
         # is here because we only compress streamres at the moment.
         mediatype, engine, engineopts = _httpresponsetype(
             repo.ui, req, rsp.prefer_uncompressed)
         gen = engine.compressstream(gen, engineopts)
 
         if mediatype == HGTYPE2:
             gen = genversion2(gen, engine, engineopts)
 
-        wsgireq.respond(HTTP_OK, mediatype)
-        return gen
+        setresponse(HTTP_OK, mediatype, bodygen=gen)
     elif isinstance(rsp, wireprototypes.pushres):
         rsp = '%d\n%s' % (rsp.res, rsp.output)
-        wsgireq.respond(HTTP_OK, HGTYPE, body=rsp)
-        return []
+        setresponse(HTTP_OK, HGTYPE, bodybytes=rsp)
     elif isinstance(rsp, wireprototypes.pusherr):
         rsp = '0\n%s\n' % rsp.res
-        wsgireq.respond(HTTP_OK, HGTYPE, body=rsp)
-        return []
+        res.drain = True
+        setresponse(HTTP_OK, HGTYPE, bodybytes=rsp)
     elif isinstance(rsp, wireprototypes.ooberror):
-        rsp = rsp.message
-        wsgireq.respond(HTTP_OK, HGERRTYPE, body=rsp)
-        return []
-    raise error.ProgrammingError('hgweb.protocol internal failure', rsp)
-
-def _handlehttperror(e, wsgireq, req):
-    """Called when an ErrorResponse is raised during HTTP request processing."""
-
-    # TODO This response body assumes the failed command was
-    # "unbundle." That assumption is not always valid.
-    wsgireq.respond(e, HGTYPE, body='0\n%s\n' % pycompat.bytestr(e))
-
-    return ''
+        setresponse(HTTP_OK, HGERRTYPE, bodybytes=rsp.message)
+    else:
+        raise error.ProgrammingError('hgweb.protocol internal failure', rsp)
 
 def _sshv1respondbytes(fout, value):
     """Send a bytes response for protocol version 1."""
     fout.write('%d\n' % len(value))
     fout.write(value)
     fout.flush()
 
 def _sshv1respondstream(fout, source):
     write = fout.write
     for chunk in source.gen:
         write(chunk)
     fout.flush()
 
 def _sshv1respondooberror(fout, ferr, rsp):
     ferr.write(b'%s\n-\n' % rsp)
     ferr.flush()
     fout.write(b'\n')
     fout.flush()
 
 class sshv1protocolhandler(wireprototypes.baseprotocolhandler):
     """Handler for requests services via version 1 of SSH protocol."""
     def __init__(self, ui, fin, fout):
         self._ui = ui
         self._fin = fin
         self._fout = fout
 
     @property
     def name(self):
         return wireprototypes.SSHV1
 
     def getargs(self, args):
         data = {}
         keys = args.split()
         for n in xrange(len(keys)):
             argline = self._fin.readline()[:-1]
             arg, l = argline.split()
             if arg not in keys:
                 raise error.Abort(_("unexpected parameter %r") % arg)
             if arg == '*':
                 star = {}
                 for k in xrange(int(l)):
                     argline = self._fin.readline()[:-1]
                     arg, l = argline.split()
                     val = self._fin.read(int(l))
                     star[arg] = val
                 data['*'] = star
             else:
                 val = self._fin.read(int(l))
                 data[arg] = val
         return [data[k] for k in keys]
 
     def forwardpayload(self, fpout):
         # We initially send an empty response. This tells the client it is
         # OK to start sending data. If a client sees any other response, it
         # interprets it as an error.
         _sshv1respondbytes(self._fout, b'')
 
         # The file is in the form:
         #
         # <chunk size>\n<chunk>
         # ...
         # 0\n
         count = int(self._fin.readline())
         while count:
             fpout.write(self._fin.read(count))
             count = int(self._fin.readline())
 
     @contextlib.contextmanager
     def mayberedirectstdio(self):
         yield None
 
     def client(self):
         client = encoding.environ.get('SSH_CLIENT', '').split(' ', 1)[0]
         return 'remote:ssh:' + client
 
     def addcapabilities(self, repo, caps):
         return caps
 
     def checkperm(self, perm):
         pass
 
 class sshv2protocolhandler(sshv1protocolhandler):
     """Protocol handler for version 2 of the SSH protocol."""
 
     @property
     def name(self):
         return wireprototypes.SSHV2
 
 def _runsshserver(ui, repo, fin, fout, ev):
     # This function operates like a state machine of sorts. The following
     # states are defined:
     #
     # protov1-serving
     #    Server is in protocol version 1 serving mode. Commands arrive on
     #    new lines. These commands are processed in this state, one command
     #    after the other.
     #
     # protov2-serving
     #    Server is in protocol version 2 serving mode.
     #
     # upgrade-initial
     #    The server is going to process an upgrade request.
     #
     # upgrade-v2-filter-legacy-handshake
     #    The protocol is being upgraded to version 2. The server is expecting
     #    the legacy handshake from version 1.
     #
     # upgrade-v2-finish
     #    The upgrade to version 2 of the protocol is imminent.
     #
     # shutdown
     #    The server is shutting down, possibly in reaction to a client event.
     #
     # And here are their transitions:
     #
     # protov1-serving -> shutdown
     #    When server receives an empty request or encounters another
     #    error.
     #
     # protov1-serving -> upgrade-initial
     #    An upgrade request line was seen.
     #
     # upgrade-initial -> upgrade-v2-filter-legacy-handshake
     #    Upgrade to version 2 in progress. Server is expecting to
     #    process a legacy handshake.
     #
     # upgrade-v2-filter-legacy-handshake -> shutdown
     #    Client did not fulfill upgrade handshake requirements.
     #
     # upgrade-v2-filter-legacy-handshake -> upgrade-v2-finish
     #    Client fulfilled version 2 upgrade requirements. Finishing that
     #    upgrade.
     #
     # upgrade-v2-finish -> protov2-serving
     #    Protocol upgrade to version 2 complete. Server can now speak protocol
     #    version 2.
     #
     # protov2-serving -> protov1-serving
     #    Ths happens by default since protocol version 2 is the same as
     #    version 1 except for the handshake.
 
     state = 'protov1-serving'
     proto = sshv1protocolhandler(ui, fin, fout)
     protoswitched = False
 
     while not ev.is_set():
         if state == 'protov1-serving':
             # Commands are issued on new lines.
             request = fin.readline()[:-1]
 
             # Empty lines signal to terminate the connection.
             if not request:
                 state = 'shutdown'
                 continue
 
             # It looks like a protocol upgrade request. Transition state to
             # handle it.
             if request.startswith(b'upgrade '):
                 if protoswitched:
                     _sshv1respondooberror(fout, ui.ferr,
                                           b'cannot upgrade protocols multiple '
                                           b'times')
                     state = 'shutdown'
                     continue
 
                 state = 'upgrade-initial'
                 continue
 
             available = wireproto.commands.commandavailable(request, proto)
 
             # This command isn't available. Send an empty response and go
             # back to waiting for a new command.
             if not available:
                 _sshv1respondbytes(fout, b'')
                 continue
 
             rsp = wireproto.dispatch(repo, proto, request)
 
             if isinstance(rsp, bytes):
                 _sshv1respondbytes(fout, rsp)
             elif isinstance(rsp, wireprototypes.bytesresponse):
                 _sshv1respondbytes(fout, rsp.data)
             elif isinstance(rsp, wireprototypes.streamres):
                 _sshv1respondstream(fout, rsp)
             elif isinstance(rsp, wireprototypes.streamreslegacy):
                 _sshv1respondstream(fout, rsp)
             elif isinstance(rsp, wireprototypes.pushres):
                 _sshv1respondbytes(fout, b'')
                 _sshv1respondbytes(fout, b'%d' % rsp.res)
             elif isinstance(rsp, wireprototypes.pusherr):
                 _sshv1respondbytes(fout, rsp.res)
             elif isinstance(rsp, wireprototypes.ooberror):
                 _sshv1respondooberror(fout, ui.ferr, rsp.message)
             else:
                 raise error.ProgrammingError('unhandled response type from '
                                              'wire protocol command: %s' % rsp)
 
         # For now, protocol version 2 serving just goes back to version 1.
         elif state == 'protov2-serving':
             state = 'protov1-serving'
             continue
 
         elif state == 'upgrade-initial':
             # We should never transition into this state if we've switched
             # protocols.
             assert not protoswitched
             assert proto.name == wireprototypes.SSHV1
 
             # Expected: upgrade <token> <capabilities>
             # If we get something else, the request is malformed. It could be
             # from a future client that has altered the upgrade line content.
             # We treat this as an unknown command.
             try:
                 token, caps = request.split(b' ')[1:]
             except ValueError:
                 _sshv1respondbytes(fout, b'')
                 state = 'protov1-serving'
                 continue
 
             # Send empty response if we don't support upgrading protocols.
             if not ui.configbool('experimental', 'sshserver.support-v2'):
                 _sshv1respondbytes(fout, b'')
                 state = 'protov1-serving'
                 continue
 
             try:
                 caps = urlreq.parseqs(caps)
             except ValueError:
                 _sshv1respondbytes(fout, b'')
                 state = 'protov1-serving'
                 continue
 
             # We don't see an upgrade request to protocol version 2. Ignore
             # the upgrade request.
             wantedprotos = caps.get(b'proto', [b''])[0]
             if SSHV2 not in wantedprotos:
                 _sshv1respondbytes(fout, b'')
                 state = 'protov1-serving'
                 continue
 
             # It looks like we can honor this upgrade request to protocol 2.
             # Filter the rest of the handshake protocol request lines.
             state = 'upgrade-v2-filter-legacy-handshake'
             continue
 
         elif state == 'upgrade-v2-filter-legacy-handshake':
             # Client should have sent legacy handshake after an ``upgrade``
             # request. Expected lines:
             #
             #    hello
             #    between
             #    pairs 81
             #    0000...-0000...
 
             ok = True
             for line in (b'hello', b'between', b'pairs 81'):
                 request = fin.readline()[:-1]
 
                 if request != line:
                     _sshv1respondooberror(fout, ui.ferr,
                                           b'malformed handshake protocol: '
                                           b'missing %s' % line)
                     ok = False
                     state = 'shutdown'
                     break
 
             if not ok:
                 continue
 
             request = fin.read(81)
             if request != b'%s-%s' % (b'0' * 40, b'0' * 40):
                 _sshv1respondooberror(fout, ui.ferr,
                                       b'malformed handshake protocol: '
                                       b'missing between argument value')
                 state = 'shutdown'
                 continue
 
             state = 'upgrade-v2-finish'
             continue
 
         elif state == 'upgrade-v2-finish':
             # Send the upgrade response.
             fout.write(b'upgraded %s %s\n' % (token, SSHV2))
             servercaps = wireproto.capabilities(repo, proto)
             rsp = b'capabilities: %s' % servercaps.data
             fout.write(b'%d\n%s\n' % (len(rsp), rsp))
             fout.flush()
 
             proto = sshv2protocolhandler(ui, fin, fout)
             protoswitched = True
 
             state = 'protov2-serving'
             continue
 
         elif state == 'shutdown':
             break
 
         else:
             raise error.ProgrammingError('unhandled ssh server state: %s' %
                                          state)
 
 class sshserver(object):
     def __init__(self, ui, repo, logfh=None):
         self._ui = ui
         self._repo = repo
         self._fin = ui.fin
         self._fout = ui.fout
 
         # Log write I/O to stdout and stderr if configured.
         if logfh:
             self._fout = util.makeloggingfileobject(
                 logfh, self._fout, 'o', logdata=True)
             ui.ferr = util.makeloggingfileobject(
                 logfh, ui.ferr, 'e', logdata=True)
 
         hook.redirect(True)
         ui.fout = repo.ui.fout = ui.ferr
 
         # Prevent insertion/deletion of CRs
         util.setbinary(self._fin)
         util.setbinary(self._fout)
 
     def serve_forever(self):
         self.serveuntil(threading.Event())
         sys.exit(0)
 
     def serveuntil(self, ev):
         """Serve until a threading.Event is set."""
         _runsshserver(self._ui, self._repo, self._fin, self._fout, ev)