Mercurial > public > mercurial-scm > hg
view mercurial/wireprotoserver.py @ 37288:9bfcbe4f4745
wireproto: add streams to frame-based protocol
Previously, the frame-based protocol was just a series of frames,
with each frame associated with a request ID.
In order to scale the protocol, we'll want to enable the use of
compression. While it is possible to enable compression at the
socket/pipe level, this has its disadvantages. The big one is it
undermines the point of frames being standalone, atomic units that
can be read and written: if you add compression above the framing
protocol, you are back to having a stream-based protocol as opposed
to something frame-based.
So in order to preserve frames, compression needs to occur at
the frame payload level.
Compressing each frame's payload individually will limit compression
ratios because the window size of the compressor will be limited
by the max frame size, which is 32-64kb as currently defined. It
will also add CPU overhead, as it is more efficient for compressors
to operate on fewer, larger blocks of data than more, smaller blocks.
So compressing each frame independently is out.
This means we need to compress each frame's payload as if it is part
of a larger stream.
The simplest approach is to have 1 stream per connection. This
could certainly work. However, it has disadvantages (documented below).
We could also have 1 stream per RPC/command invocation. (This is the
model HTTP/2 goes with.) This also has disadvantages.
The main disadvantage to one global stream is that it has the very
real potential to create CPU bottlenecks doing compression. Networks
are only getting faster and the performance of single CPU cores has
been relatively flat. Newer compression formats like zstandard offer
better CPU cycle efficiency than predecessors like zlib. But it still
all too common to saturate your CPU with compression overhead long
before you saturate the network pipe.
The main disadvantage with streams per request is that you can't
reap the benefits of the compression context for multiple requests.
For example, if you send 1000 RPC requests (or HTTP/2 requests for
that matter), the response to each would have its own compression
context. The overall size of the raw responses would be larger because
compression contexts wouldn't be able to reference data from another
request or response.
The approach for streams as implemented in this commit is to support
N streams per connection and for streams to potentially span requests
and responses. As explained by the added internals docs, this
facilitates servers and clients delegating independent streams and
compression to independent threads / CPU cores. This helps alleviate
the CPU bottleneck of compression. This design also allows compression
contexts to be reused across requests/responses. This can result in
improved compression ratios and less overhead for compressors and
decompressors having to build new contexts.
Another feature that was defined was the ability for individual frames
within a stream to declare whether that individual frame's payload
uses the content encoding (read: compression) defined by the stream.
The idea here is that some servers may serve data from a combination
of caches and dynamic resolution. Data coming from caches may be
pre-compressed. We want to facilitate servers being able to essentially
stream bytes from caches to the wire with minimal overhead. Being
able to mix and match with frames are compressed within a stream
enables these types of advanced server functionality.
This commit defines the new streams mechanism. Basic code for
supporting streams in frames has been added. But that code is
seriously lacking and doesn't fully conform to the defined protocol.
For example, we don't close any streams. And support for content
encoding within streams is not yet implemented. The change was
rather invasive and I didn't think it would be reasonable to implement
the entire feature in a single commit.
For the record, I would have loved to reuse an existing multiplexing
protocol to build the new wire protocol on top of. However, I couldn't
find a protocol that offers the performance and scaling characteristics
that I desired. Namely, it should support multiple compression
contexts to facilitate scaling out to multiple CPU cores and
compression contexts should be able to live longer than single RPC
requests. HTTP/2 *almost* fits the bill. But the semantics of HTTP
message exchange state that streams can only live for a single
request-response. We /could/ tunnel on top of HTTP/2 streams and
frames with HEADER and DATA frames. But there's no guarantee that
HTTP/2 libraries and proxies would allow us to use HTTP/2 streams
and frames without the HTTP message exchange semantics defined in
RFC 7540 Section 8. Other RPC protocols like gRPC tunnel are built
on top of HTTP/2 and thus preserve its semantics of stream per
RPC invocation. Even QUIC does this. We could attempt to invent a
higher-level stream that spans HTTP/2 streams. But this would be
violating HTTP/2 because there is no guarantee that HTTP/2 streams
are routed to the same server. The best we can do - which is what
this protocol does - is shoehorn all request and response data into
a single HTTP message and create streams within. At that point, we've
defined a Content-Type in HTTP parlance. It just so happens our
media type can also work as a standalone, stream-based protocol,
without leaning on HTTP or similar protocol.
Differential Revision: https://phab.mercurial-scm.org/D2907
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Mon, 26 Mar 2018 11:00:16 -0700 |
parents | 3ed344546d9e |
children | 5fadc63ac99f |
line wrap: on
line source
# Copyright 21 May 2005 - (c) 2005 Jake Edge <jake@edge2.net> # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> # # 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, wireprotoframing, wireprototypes, ) from .utils import ( procutil, ) 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' FRAMINGTYPE = b'application/mercurial-exp-framing-0002' HTTPV2 = wireprototypes.HTTPV2 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, req, ui, checkperm): 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 = self._req.qsparams.asdictoflists() 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._req.urlscheme, urlreq.quote(self._req.remotehost or ''), urlreq.quote(self._req.remoteuser or '')) def addcapabilities(self, repo, caps): caps.append(b'batch') 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, 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. ``req`` is a ``parsedrequest`` instance. ``res`` is a ``wsgiresponse`` instance. 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.qsparams: return False cmd = req.qsparams['cmd'] # 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 # 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.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(req, repo.ui, lambda perm: checkperm(rctx, req, 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: _callhttp(repo, req, res, proto, cmd) except hgwebcommon.ErrorResponse as e: 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 def handlewsgiapirequest(rctx, req, res, checkperm): """Handle requests to /api/*.""" assert req.dispatchparts[0] == b'api' repo = rctx.repo # This whole URL space is experimental for now. But we want to # reserve the URL space. So, 404 all URLs if the feature isn't enabled. if not repo.ui.configbool('experimental', 'web.apiserver'): res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('Experimental API server endpoint not enabled')) return # 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) # Requests to /api/ list available APIs. if req.dispatchparts == [b'api']: res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' lines = [_('APIs can be accessed at /api/<name>, where <name> can be ' 'one of the following:\n')] if availableapis: lines.extend(sorted(availableapis)) else: lines.append(_('(no available APIs)\n')) res.setbodybytes(b'\n'.join(lines)) return proto = req.dispatchparts[1] if proto not in API_HANDLERS: res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('Unknown API: %s\nKnown APIs: %s') % ( proto, b', '.join(sorted(availableapis)))) return if proto not in availableapis: res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('API %s not enabled\n') % proto) return API_HANDLERS[proto]['handler'](rctx, req, res, checkperm, req.dispatchparts[2:]) def _handlehttpv2request(rctx, req, res, checkperm, urlparts): from .hgweb import common as hgwebcommon # URL space looks like: <permissions>/<command>, where <permission> can # be ``ro`` or ``rw`` to signal read-only or read-write, respectively. # Root URL does nothing meaningful... yet. if not urlparts: res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('HTTP version 2 API handler')) return if len(urlparts) == 1: res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('do not know how to process %s\n') % req.dispatchpath) return permission, command = urlparts[0:2] if permission not in (b'ro', b'rw'): res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('unknown permission: %s') % permission) return if req.method != 'POST': res.status = b'405 Method Not Allowed' res.headers[b'Allow'] = b'POST' res.setbodybytes(_('commands require POST requests')) return # At some point we'll want to use our own API instead of recycling the # behavior of version 1 of the wire protocol... # TODO return reasonable responses - not responses that overload the # HTTP status line message for error reporting. try: checkperm(rctx, req, 'pull' if permission == b'ro' else 'push') except hgwebcommon.ErrorResponse as e: res.status = hgwebcommon.statusmessage(e.code, pycompat.bytestr(e)) for k, v in e.headers: res.headers[k] = v res.setbodybytes('permission denied') return # We have a special endpoint to reflect the request back at the client. if command == b'debugreflect': _processhttpv2reflectrequest(rctx.repo.ui, rctx.repo, req, res) return # Extra commands that we handle that aren't really wire protocol # commands. Think extra hard before making this hackery available to # extension. extracommands = {'multirequest'} if command not in wireproto.commands and command not in extracommands: res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('unknown wire protocol command: %s\n') % command) return repo = rctx.repo ui = repo.ui proto = httpv2protocolhandler(req, ui) if (not wireproto.commands.commandavailable(command, proto) and command not in extracommands): res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('invalid wire protocol command: %s') % command) return # TODO consider cases where proxies may add additional Accept headers. if req.headers.get(b'Accept') != FRAMINGTYPE: res.status = b'406 Not Acceptable' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('client MUST specify Accept header with value: %s\n') % FRAMINGTYPE) return if req.headers.get(b'Content-Type') != FRAMINGTYPE: res.status = b'415 Unsupported Media Type' # TODO we should send a response with appropriate media type, # since client does Accept it. res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('client MUST send Content-Type header with ' 'value: %s\n') % FRAMINGTYPE) return _processhttpv2request(ui, repo, req, res, permission, command, proto) def _processhttpv2reflectrequest(ui, repo, req, res): """Reads unified frame protocol request and dumps out state to client. This special endpoint can be used to help debug the wire protocol. Instead of routing the request through the normal dispatch mechanism, we instead read all frames, decode them, and feed them into our state tracker. We then dump the log of all that activity back out to the client. """ import json # Reflection APIs have a history of being abused, accidentally disclosing # sensitive data, etc. So we have a config knob. if not ui.configbool('experimental', 'web.api.debugreflect'): res.status = b'404 Not Found' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('debugreflect service not available')) return # We assume we have a unified framing protocol request body. reactor = wireprotoframing.serverreactor() states = [] while True: frame = wireprotoframing.readframe(req.bodyfh) if not frame: states.append(b'received: <no frame>') break states.append(b'received: %d %d %d %s' % (frame.typeid, frame.flags, frame.requestid, frame.payload)) action, meta = reactor.onframerecv(frame) states.append(json.dumps((action, meta), sort_keys=True, separators=(', ', ': '))) action, meta = reactor.oninputeof() meta['action'] = action states.append(json.dumps(meta, sort_keys=True, separators=(', ',': '))) res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(b'\n'.join(states)) def _processhttpv2request(ui, repo, req, res, authedperm, reqcommand, proto): """Post-validation handler for HTTPv2 requests. Called when the HTTP request contains unified frame-based protocol frames for evaluation. """ # TODO Some HTTP clients are full duplex and can receive data before # the entire request is transmitted. Figure out a way to indicate support # for that so we can opt into full duplex mode. reactor = wireprotoframing.serverreactor(deferoutput=True) seencommand = False while True: frame = wireprotoframing.readframe(req.bodyfh) if not frame: break action, meta = reactor.onframerecv(frame) if action == 'wantframe': # Need more data before we can do anything. continue elif action == 'runcommand': sentoutput = _httpv2runcommand(ui, repo, req, res, authedperm, reqcommand, reactor, meta, issubsequent=seencommand) if sentoutput: return seencommand = True elif action == 'error': # TODO define proper error mechanism. res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(meta['message'] + b'\n') return else: raise error.ProgrammingError( 'unhandled action from frame processor: %s' % action) action, meta = reactor.oninputeof() if action == 'sendframes': # We assume we haven't started sending the response yet. If we're # wrong, the response type will raise an exception. res.status = b'200 OK' res.headers[b'Content-Type'] = FRAMINGTYPE res.setbodygen(meta['framegen']) elif action == 'noop': pass else: raise error.ProgrammingError('unhandled action from frame processor: %s' % action) def _httpv2runcommand(ui, repo, req, res, authedperm, reqcommand, reactor, command, issubsequent): """Dispatch a wire protocol command made from HTTPv2 requests. The authenticated permission (``authedperm``) along with the original command from the URL (``reqcommand``) are passed in. """ # We already validated that the session has permissions to perform the # actions in ``authedperm``. In the unified frame protocol, the canonical # command to run is expressed in a frame. However, the URL also requested # to run a specific command. We need to be careful that the command we # run doesn't have permissions requirements greater than what was granted # by ``authedperm``. # # Our rule for this is we only allow one command per HTTP request and # that command must match the command in the URL. However, we make # an exception for the ``multirequest`` URL. This URL is allowed to # execute multiple commands. We double check permissions of each command # as it is invoked to ensure there is no privilege escalation. # TODO consider allowing multiple commands to regular command URLs # iff each command is the same. proto = httpv2protocolhandler(req, ui, args=command['args']) if reqcommand == b'multirequest': if not wireproto.commands.commandavailable(command['command'], proto): # TODO proper error mechanism res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('wire protocol command not available: %s') % command['command']) return True # TODO don't use assert here, since it may be elided by -O. assert authedperm in (b'ro', b'rw') wirecommand = wireproto.commands[command['command']] assert wirecommand.permission in ('push', 'pull') if authedperm == b'ro' and wirecommand.permission != 'pull': # TODO proper error mechanism res.status = b'403 Forbidden' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('insufficient permissions to execute ' 'command: %s') % command['command']) return True # TODO should we also call checkperm() here? Maybe not if we're going # to overhaul that API. The granted scope from the URL check should # be good enough. else: # Don't allow multiple commands outside of ``multirequest`` URL. if issubsequent: # TODO proper error mechanism res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('multiple commands cannot be issued to this ' 'URL')) return True if reqcommand != command['command']: # TODO define proper error mechanism res.status = b'200 OK' res.headers[b'Content-Type'] = b'text/plain' res.setbodybytes(_('command in frame must match command in URL')) return True rsp = wireproto.dispatch(repo, proto, command['command']) res.status = b'200 OK' res.headers[b'Content-Type'] = FRAMINGTYPE stream = wireprotoframing.stream(2) if isinstance(rsp, wireprototypes.bytesresponse): action, meta = reactor.onbytesresponseready(stream, command['requestid'], rsp.data) else: action, meta = reactor.onapplicationerror( _('unhandled response type from wire proto command')) if action == 'sendframes': res.setbodygen(meta['framegen']) return True elif action == 'noop': return False else: raise error.ProgrammingError('unhandled event from reactor: %s' % action) # Maps API name to metadata so custom API can be registered. API_HANDLERS = { HTTPV2: { 'config': ('experimental', 'web.api.http-v2'), 'handler': _handlehttpv2request, }, } class httpv2protocolhandler(wireprototypes.baseprotocolhandler): def __init__(self, req, ui, args=None): self._req = req self._ui = ui self._args = args @property def name(self): return HTTPV2 def getargs(self, args): data = {} for k in args.split(): if k == '*': raise NotImplementedError('do not support * args') else: data[k] = self._args[k] return [data[k] for k in args.split()] def forwardpayload(self, fp): raise NotImplementedError @contextlib.contextmanager def mayberedirectstdio(self): raise NotImplementedError def client(self): raise NotImplementedError def addcapabilities(self, repo, caps): return caps def checkperm(self, perm): raise NotImplementedError 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, 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): 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): setresponse(HTTP_OK, HGTYPE, bodybytes=rsp) elif isinstance(rsp, wireprototypes.bytesresponse): setresponse(HTTP_OK, HGTYPE, bodybytes=rsp.data) elif isinstance(rsp, wireprototypes.streamreslegacy): 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) setresponse(HTTP_OK, mediatype, bodygen=gen) elif isinstance(rsp, wireprototypes.pushres): rsp = '%d\n%s' % (rsp.res, rsp.output) setresponse(HTTP_OK, HGTYPE, bodybytes=rsp) elif isinstance(rsp, wireprototypes.pusherr): rsp = '0\n%s\n' % rsp.res res.drain = True setresponse(HTTP_OK, HGTYPE, bodybytes=rsp) elif isinstance(rsp, wireprototypes.ooberror): 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): caps.append(b'batch') 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 procutil.setbinary(self._fin) procutil.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)