1
0
mirror of https://git.tartarus.org/simon/putty.git synced 2025-01-09 17:38:00 +00:00
putty-source/network.h
Simon Tatham ca70b1285d Allow creating FdSocket/HandleSocket before the fds/handles.
Previously, a setup function returning one of these socket types (such
as platform_new_connection) had to do all its setup synchronously,
because if it was going to call make_fd_socket or make_handle_socket,
it had to have the actual fds or HANDLEs ready-made. If some kind of
asynchronous operation were needed before those fds become available,
there would be no way the function could achieve it, except by
becoming a whole extra permanent Socket wrapper layer.

Now there is, because you can make an FdSocket when you don't yet have
the fds, or a HandleSocket without the HANDLEs. Instead, you provide
an instance of the new trait 'DeferredSocketOpener', which is
responsible for setting in motion whatever asynchronous setup
procedure it needs, and when that finishes, calling back to
setup_fd_socket / setup_handle_socket to provide the missing pieces.

In the meantime, the FdSocket or HandleSocket will sit there inertly,
buffering any data the client might eagerly hand it via sk_write(),
and waiting for its setup to finish. When it does finish, buffered
data will be released.

In FdSocket, this is easy enough, because we were doing our own
buffering anyway - we called the uxsel system to find out when the fds
were readable/writable, and then wrote to them from our own bufchain.
So more or less all I had to do was make the try_send function do
nothing if the setup phase wasn't finished yet.

In HandleSocket, on the other hand, we're passing all our data to the
underlying handle-io.c system, and making _that_ deferrable in the
same way would be much more painful, because that's the place where
the scary threads live. So instead I've arranged it by replacing the
whole vtable, so that a deferred HandleSocket and a normal
HandleSocket are effectively separate trait implementations that can
share their state structure. And in fact that state struct itself now
contains a big anonymous union, containing one branch to go with each
vtable.

Nothing yet uses this system, but the next commit will do so.
2021-12-22 15:45:41 +00:00

433 lines
17 KiB
C

/*
* Networking abstraction in PuTTY.
*
* The way this works is: a back end can choose to open any number
* of sockets - including zero, which might be necessary in some.
* It can register a bunch of callbacks (most notably for when
* data is received) for each socket, and it can call the networking
* abstraction to send data without having to worry about blocking.
* The stuff behind the abstraction takes care of selects and
* nonblocking writes and all that sort of painful gubbins.
*/
#ifndef PUTTY_NETWORK_H
#define PUTTY_NETWORK_H
#include "defs.h"
typedef struct SocketVtable SocketVtable;
typedef struct PlugVtable PlugVtable;
struct Socket {
const struct SocketVtable *vt;
};
struct SocketVtable {
Plug *(*plug) (Socket *s, Plug *p);
/* use a different plug (return the old one) */
/* if p is NULL, it doesn't change the plug */
/* but it does return the one it's using */
void (*close) (Socket *s);
size_t (*write) (Socket *s, const void *data, size_t len);
size_t (*write_oob) (Socket *s, const void *data, size_t len);
void (*write_eof) (Socket *s);
void (*set_frozen) (Socket *s, bool is_frozen);
/* ignored by tcp, but vital for ssl */
const char *(*socket_error) (Socket *s);
SocketPeerInfo *(*peer_info) (Socket *s);
};
typedef union { void *p; int i; } accept_ctx_t;
typedef Socket *(*accept_fn_t)(accept_ctx_t ctx, Plug *plug);
struct Plug {
const struct PlugVtable *vt;
};
typedef enum PlugLogType {
PLUGLOG_CONNECT_TRYING,
PLUGLOG_CONNECT_FAILED,
PLUGLOG_CONNECT_SUCCESS,
PLUGLOG_PROXY_MSG,
} PlugLogType;
typedef enum PlugCloseType {
PLUGCLOSE_NORMAL,
PLUGCLOSE_ERROR,
PLUGCLOSE_BROKEN_PIPE,
PLUGCLOSE_USER_ABORT,
} PlugCloseType;
struct PlugVtable {
/*
* Passes the client progress reports on the process of setting
* up the connection.
*
* - PLUGLOG_CONNECT_TRYING means we are about to try to connect
* to address `addr' (error_msg and error_code are ignored)
*
* - PLUGLOG_CONNECT_FAILED means we have failed to connect to
* address `addr' (error_msg and error_code are supplied). This
* is not a fatal error - we may well have other candidate
* addresses to fall back to. When it _is_ fatal, the closing()
* function will be called.
*
* - PLUGLOG_CONNECT_SUCCESS means we have succeeded in making a
* connection. `addr' gives the address we connected to, if
* available. (But sometimes, in cases of complicated proxy
* setups, it might not be available, so receivers of this log
* event should be prepared to deal with addr==NULL.)
*
* - PLUGLOG_PROXY_MSG means that error_msg contains a line of
* logging information from whatever the connection is being
* proxied through. This will typically be a wodge of
* standard-error output from a local proxy command, so the
* receiver should probably prefix it to indicate this.
*
* Note that sometimes log messages may be sent even to Socket
* types that don't involve making an outgoing connection, e.g.
* because the same core implementation (such as Windows handle
* sockets) is shared between listening and connecting sockets. So
* all Plugs must implement this method, even if only to ignore
* the logged events.
*/
void (*log)(Plug *p, PlugLogType type, SockAddr *addr, int port,
const char *error_msg, int error_code);
/*
* Notifies the Plug that the socket is closing, and something
* about why.
*
* - PLUGCLOSE_NORMAL means an ordinary non-error closure. In
* this case, error_msg should be ignored (and hopefully
* callers will have passed NULL).
*
* - PLUGCLOSE_ERROR indicates that an OS error occurred, and
* 'error_msg' contains a string describing it, for use in
* diagnostics. (Ownership of the string is not transferred.)
* This error class covers anything other than the special
* case below:
*
* - PLUGCLOSE_BROKEN_PIPE behaves like PLUGCLOSE_ERROR (in
* particular, there's still an error message provided), but
* distinguishes the particular error condition signalled by
* EPIPE / ERROR_BROKEN_PIPE, which ssh/sharing.c needs to
* recognise and handle specially in one situation.
*
* - PLUGCLOSE_USER_ABORT means that the close has happened as a
* result of some kind of deliberate user action (e.g. hitting
* ^C at a password prompt presented by a proxy socket setup
* phase). This can be used to suppress interactive error
* messages sent to the user (such as dialog boxes), on the
* grounds that the user already knows. However, 'error_msg'
* will still contain some appropriate text, so that
* non-interactive error reporting (e.g. event logs) can still
* record why the connection terminated.
*/
void (*closing)(Plug *p, PlugCloseType type, const char *error_msg);
/*
* Provides incoming socket data to the Plug. Three cases:
*
* - urgent==0. `data' points to `len' bytes of perfectly
* ordinary data.
*
* - urgent==1. `data' points to `len' bytes of data,
* which were read from before an Urgent pointer.
*
* - urgent==2. `data' points to `len' bytes of data,
* the first of which was the one at the Urgent mark.
*/
void (*receive) (Plug *p, int urgent, const char *data, size_t len);
/*
* Called when the pending send backlog on a socket is cleared or
* partially cleared. The new backlog size is passed in the
* `bufsize' parameter.
*/
void (*sent) (Plug *p, size_t bufsize);
/*
* Only called on listener-type sockets, and is passed a
* constructor function+context that will create a fresh Socket
* describing the connection. It returns nonzero if it doesn't
* want the connection for some reason, or 0 on success.
*/
int (*accepting)(Plug *p, accept_fn_t constructor, accept_ctx_t ctx);
};
/* Proxy indirection layer.
*
* Calling new_connection transfers ownership of 'addr': the proxy
* layer is now responsible for freeing it, and the caller shouldn't
* assume it exists any more.
*
* If calling this from a backend with a Seat, you can also give it a
* pointer to the backend's Interactor trait. In that situation, it
* might replace the backend's seat with a temporary seat of its own,
* and give the real Seat to an Interactor somewhere in the proxy
* system so that it can ask for passwords (and, in the case of SSH
* proxying, other prompts like host key checks). If that happens,
* then the resulting 'temp seat' is the backend's property, and it
* will have to remember to free it when cleaning up, or after
* flushing it back into the real seat when the network connection
* attempt completes.
*
* You can free your TempSeat and resume using the real Seat when one
* of two things happens: either your Plug's closing() method is
* called (indicating failure to connect), or its log() method is
* called with PLUGLOG_CONNECT_SUCCESS. In the latter case, you'll
* probably want to flush the TempSeat's contents into the real Seat,
* of course.
*/
Socket *new_connection(SockAddr *addr, const char *hostname,
int port, bool privport,
bool oobinline, bool nodelay, bool keepalive,
Plug *plug, Conf *conf, Interactor *interactor);
Socket *new_listener(const char *srcaddr, int port, Plug *plug,
bool local_host_only, Conf *conf, int addressfamily);
SockAddr *name_lookup(const char *host, int port, char **canonicalname,
Conf *conf, int addressfamily, LogContext *logctx,
const char *lookup_reason_for_logging);
/* platform-dependent callback from new_connection() */
/* (same caveat about addr as new_connection()) */
Socket *platform_new_connection(SockAddr *addr, const char *hostname,
int port, bool privport,
bool oobinline, bool nodelay, bool keepalive,
Plug *plug, Conf *conf, Interactor *itr);
/* callback for SSH jump-host proxying */
Socket *sshproxy_new_connection(SockAddr *addr, const char *hostname,
int port, bool privport,
bool oobinline, bool nodelay, bool keepalive,
Plug *plug, Conf *conf, Interactor *itr);
/* socket functions */
void sk_init(void); /* called once at program startup */
void sk_cleanup(void); /* called just before program exit */
SockAddr *sk_namelookup(const char *host, char **canonicalname, int address_family);
SockAddr *sk_nonamelookup(const char *host);
void sk_getaddr(SockAddr *addr, char *buf, int buflen);
bool sk_addr_needs_port(SockAddr *addr);
bool sk_hostname_is_local(const char *name);
bool sk_address_is_local(SockAddr *addr);
bool sk_address_is_special_local(SockAddr *addr);
int sk_addrtype(SockAddr *addr);
void sk_addrcopy(SockAddr *addr, char *buf);
void sk_addr_free(SockAddr *addr);
/* sk_addr_dup generates another SockAddr which contains the same data
* as the original one and can be freed independently. May not actually
* physically _duplicate_ it: incrementing a reference count so that
* one more free is required before it disappears is an acceptable
* implementation. */
SockAddr *sk_addr_dup(SockAddr *addr);
/* NB, control of 'addr' is passed via sk_new, which takes responsibility
* for freeing it, as for new_connection() */
Socket *sk_new(SockAddr *addr, int port, bool privport, bool oobinline,
bool nodelay, bool keepalive, Plug *p);
Socket *sk_newlistener(const char *srcaddr, int port, Plug *plug,
bool local_host_only, int address_family);
static inline Plug *sk_plug(Socket *s, Plug *p)
{ return s->vt->plug(s, p); }
static inline void sk_close(Socket *s)
{ s->vt->close(s); }
static inline size_t sk_write(Socket *s, const void *data, size_t len)
{ return s->vt->write(s, data, len); }
static inline size_t sk_write_oob(Socket *s, const void *data, size_t len)
{ return s->vt->write_oob(s, data, len); }
static inline void sk_write_eof(Socket *s)
{ s->vt->write_eof(s); }
static inline void plug_log(
Plug *p, int type, SockAddr *addr, int port, const char *msg, int code)
{ p->vt->log(p, type, addr, port, msg, code); }
static inline void plug_closing(Plug *p, PlugCloseType type, const char *msg)
{ p->vt->closing(p, type, msg); }
static inline void plug_closing_normal(Plug *p)
{ p->vt->closing(p, PLUGCLOSE_NORMAL, NULL); }
static inline void plug_closing_error(Plug *p, const char *msg)
{ p->vt->closing(p, PLUGCLOSE_ERROR, msg); }
static inline void plug_closing_user_abort(Plug *p)
{ p->vt->closing(p, PLUGCLOSE_USER_ABORT, "User aborted connection setup"); }
static inline void plug_receive(Plug *p, int urg, const char *data, size_t len)
{ p->vt->receive(p, urg, data, len); }
static inline void plug_sent (Plug *p, size_t bufsize)
{ p->vt->sent(p, bufsize); }
static inline int plug_accepting(Plug *p, accept_fn_t cons, accept_ctx_t ctx)
{ return p->vt->accepting(p, cons, ctx); }
/*
* Special error values are returned from sk_namelookup and sk_new
* if there's a problem. These functions extract an error message,
* or return NULL if there's no problem.
*/
const char *sk_addr_error(SockAddr *addr);
static inline const char *sk_socket_error(Socket *s)
{ return s->vt->socket_error(s); }
/*
* Set the `frozen' flag on a socket. A frozen socket is one in
* which all READABLE notifications are ignored, so that data is
* not accepted from the peer until the socket is unfrozen. This
* exists for two purposes:
*
* - Port forwarding: when a local listening port receives a
* connection, we do not want to receive data from the new
* socket until we have somewhere to send it. Hence, we freeze
* the socket until its associated SSH channel is ready; then we
* unfreeze it and pending data is delivered.
*
* - Socket buffering: if an SSH channel (or the whole connection)
* backs up or presents a zero window, we must freeze the
* associated local socket in order to avoid unbounded buffer
* growth.
*/
static inline void sk_set_frozen(Socket *s, bool is_frozen)
{ s->vt->set_frozen(s, is_frozen); }
/*
* Return a structure giving some information about the other end of
* the socket. May be NULL, if nothing is available at all. If it is
* not NULL, then it is dynamically allocated, and should be freed by
* a call to sk_free_peer_info(). See below for the definition.
*/
static inline SocketPeerInfo *sk_peer_info(Socket *s)
{ return s->vt->peer_info(s); }
/*
* The structure returned from sk_peer_info, and a function to free
* one (in misc.c).
*/
struct SocketPeerInfo {
int addressfamily;
/*
* Text form of the IPv4 or IPv6 address of the other end of the
* socket, if available, in the standard text representation.
*/
const char *addr_text;
/*
* Binary form of the same address. Filled in if and only if
* addr_text is not NULL. You can tell which branch of the union
* is used by examining 'addressfamily'.
*/
union {
unsigned char ipv6[16];
unsigned char ipv4[4];
} addr_bin;
/*
* Remote port number, or -1 if not available.
*/
int port;
/*
* Free-form text suitable for putting in log messages. For IP
* sockets, repeats the address and port information from above.
* But it can be completely different, e.g. for Unix-domain
* sockets it gives information about the uid, gid and pid of the
* connecting process.
*/
const char *log_text;
};
void sk_free_peer_info(SocketPeerInfo *pi);
/*
* Simple wrapper on getservbyname(), needed by ssh.c. Returns the
* port number, in host byte order (suitable for printf and so on).
* Returns 0 on failure. Any platform not supporting getservbyname
* can just return 0 - this function is not required to handle
* numeric port specifications.
*/
int net_service_lookup(char *service);
/*
* Look up the local hostname; return value needs freeing.
* May return NULL.
*/
char *get_hostname(void);
/*
* Trivial socket implementation which just stores an error. Found in
* errsock.c.
*
* The consume_string variant takes an already-formatted dynamically
* allocated string, and takes over ownership of that string.
*/
Socket *new_error_socket_fmt(Plug *plug, const char *fmt, ...)
PRINTF_LIKE(2, 3);
Socket *new_error_socket_consume_string(Plug *plug, char *errmsg);
/*
* Trivial plug that does absolutely nothing. Found in nullplug.c.
*/
extern Plug *const nullplug;
/*
* Some trivial no-op plug functions, also in nullplug.c; exposed here
* so that other Plug implementations can use them too.
*
* In particular, nullplug_log is useful to Plugs that don't need to
* worry about logging.
*/
void nullplug_log(Plug *plug, PlugLogType type, SockAddr *addr,
int port, const char *err_msg, int err_code);
void nullplug_closing(Plug *plug, PlugCloseType type, const char *error_msg);
void nullplug_receive(Plug *plug, int urgent, const char *data, size_t len);
void nullplug_sent(Plug *plug, size_t bufsize);
/* ----------------------------------------------------------------------
* Functions defined outside the network code, which have to be
* declared in this header file rather than the main putty.h because
* they use types defined here.
*/
void backend_socket_log(Seat *seat, LogContext *logctx,
PlugLogType type, SockAddr *addr, int port,
const char *error_msg, int error_code, Conf *conf,
bool session_started);
typedef struct ProxyStderrBuf {
char buf[8192];
size_t size;
} ProxyStderrBuf;
void psb_init(ProxyStderrBuf *psb);
void log_proxy_stderr(
Plug *plug, ProxyStderrBuf *psb, const void *vdata, size_t len);
/* ----------------------------------------------------------------------
* The DeferredSocketOpener trait. This is a thing that some Socket
* implementations may choose to own if they need to delay actually
* setting up the underlying connection. For example, sockets used in
* local-proxy handling (Unix FdSocket / Windows HandleSocket) might
* need to do this if they have to prompt the user interactively for
* parts of the command they'll run.
*
* Mostly, a DeferredSocketOpener implementation will keep to itself,
* arrange its own callbacks in order to do whatever setup it needs,
* and when it's ready, call back to its parent Socket via some
* implementation-specific API of its own. So the shared API here
* requires almost nothing: the only thing we need is a free function,
* so that if the owner of a Socket of this kind needs to close it
* before the deferred connection process is finished, the Socket can
* also clean up the DeferredSocketOpener dangling off it.
*/
struct DeferredSocketOpener {
const DeferredSocketOpenerVtable *vt;
};
struct DeferredSocketOpenerVtable {
void (*free)(DeferredSocketOpener *);
};
static inline void deferred_socket_opener_free(DeferredSocketOpener *dso)
{ dso->vt->free(dso); }
#endif