xref: /freebsd/sys/net/vnet.h (revision 641a6cfb86023499caafe26a4d821a0b885cf00b)
1 /*-
2  * Copyright (c) 2006-2009 University of Zagreb
3  * Copyright (c) 2006-2009 FreeBSD Foundation
4  * All rights reserved.
5  *
6  * This software was developed by the University of Zagreb and the
7  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
8  * FreeBSD Foundation.
9  *
10  * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
11  * Copyright (c) 2009 Robert N. M. Watson
12  * All rights reserved.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  * $FreeBSD$
36  */
37 
38 /*-
39  * This header file defines several sets of interfaces supporting virtualized
40  * network stacks:
41  *
42  * - Definition of 'struct vnet' and functions and macros to allocate/free/
43  *   manipulate it.
44  *
45  * - A virtual network stack memory allocator, which provides support for
46  *   virtualized global variables via a special linker set, set_vnet.
47  *
48  * - Virtualized sysinits/sysuninits, which allow constructors and
49  *   destructors to be run for each network stack subsystem as virtual
50  *   instances are created and destroyed.
51  *
52  * If VIMAGE isn't compiled into the kernel, virtualized global variables
53  * compile to normal global variables, and virtualized sysinits to regular
54  * sysinits.
55  */
56 
57 #ifndef _NET_VNET_H_
58 #define	_NET_VNET_H_
59 
60 /*
61  * struct vnet describes a virtualized network stack, and is primarily a
62  * pointer to storage for virtualized global variables.  Expose to userspace
63  * as required for libkvm.
64  */
65 #if defined(_KERNEL) || defined(_WANT_VNET)
66 #include <sys/queue.h>
67 
68 struct vnet {
69 	LIST_ENTRY(vnet)	 vnet_le;	/* all vnets list */
70 	u_int			 vnet_magic_n;
71 	u_int			 vnet_ifcnt;
72 	u_int			 vnet_sockcnt;
73 	void			*vnet_data_mem;
74 	uintptr_t		 vnet_data_base;
75 };
76 #define	VNET_MAGIC_N	0x3e0d8f29
77 
78 /*
79  * These two virtual network stack allocator definitions are also required
80  * for libkvm so that it can evaluate virtualized global variables.
81  */
82 #define	VNET_SETNAME		"set_vnet"
83 #define	VNET_SYMPREFIX		"vnet_entry_"
84 #endif
85 
86 #ifdef _KERNEL
87 
88 #ifdef VIMAGE
89 #include <sys/lock.h>
90 #include <sys/proc.h>			/* for struct thread */
91 #include <sys/rwlock.h>
92 #include <sys/sx.h>
93 
94 /*
95  * Location of the kernel's 'set_vnet' linker set.
96  */
97 extern uintptr_t	*__start_set_vnet;
98 __GLOBL(__start_set_vnet);
99 extern uintptr_t	*__stop_set_vnet;
100 __GLOBL(__stop_set_vnet);
101 
102 #define	VNET_START	(uintptr_t)&__start_set_vnet
103 #define	VNET_STOP	(uintptr_t)&__stop_set_vnet
104 
105 /*
106  * Functions to allocate and destroy virtual network stacks.
107  */
108 struct vnet *vnet_alloc(void);
109 void	vnet_destroy(struct vnet *vnet);
110 
111 /*
112  * The current virtual network stack -- we may wish to move this to struct
113  * pcpu in the future.
114  */
115 #define	curvnet	curthread->td_vnet
116 
117 /*
118  * Various macros -- get and set the current network stack, but also
119  * assertions.
120  */
121 #if defined(INVARIANTS) || defined(VNET_DEBUG)
122 #define	VNET_ASSERT(exp, msg)	do {					\
123 	if (!(exp))							\
124 		panic msg;						\
125 } while (0)
126 #else
127 #define	VNET_ASSERT(exp, msg)	do {					\
128 } while (0)
129 #endif
130 
131 #ifdef VNET_DEBUG
132 void vnet_log_recursion(struct vnet *, const char *, int);
133 
134 #define	CURVNET_SET_QUIET(arg)						\
135 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
136 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
137 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
138 	struct vnet *saved_vnet = curvnet;				\
139 	const char *saved_vnet_lpush = curthread->td_vnet_lpush;	\
140 	curvnet = arg;							\
141 	curthread->td_vnet_lpush = __func__;
142 
143 #define	CURVNET_SET_VERBOSE(arg)					\
144 	CURVNET_SET_QUIET(arg)						\
145 	if (saved_vnet)							\
146 		vnet_log_recursion(saved_vnet, saved_vnet_lpush, __LINE__);
147 
148 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
149 
150 #define	CURVNET_RESTORE()						\
151 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
152 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
153 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
154 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
155 	curvnet = saved_vnet;						\
156 	curthread->td_vnet_lpush = saved_vnet_lpush;
157 #else /* !VNET_DEBUG */
158 
159 #define	CURVNET_SET_QUIET(arg)						\
160 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
161 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
162 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
163 	struct vnet *saved_vnet = curvnet;				\
164 	curvnet = arg;
165 
166 #define	CURVNET_SET_VERBOSE(arg)					\
167 	CURVNET_SET_QUIET(arg)
168 
169 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
170 
171 #define	CURVNET_RESTORE()						\
172 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
173 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
174 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
175 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
176 	curvnet = saved_vnet;
177 #endif /* VNET_DEBUG */
178 
179 extern struct vnet *vnet0;
180 #define	IS_DEFAULT_VNET(arg)	((arg) == vnet0)
181 
182 #define	CRED_TO_VNET(cr)	(cr)->cr_prison->pr_vnet
183 #define	TD_TO_VNET(td)		CRED_TO_VNET((td)->td_ucred)
184 #define	P_TO_VNET(p)		CRED_TO_VNET((p)->p_ucred)
185 
186 /*
187  * Global linked list of all virtual network stacks, along with read locks to
188  * access it.  If a caller may sleep while accessing the list, it must use
189  * the sleepable lock macros.
190  */
191 LIST_HEAD(vnet_list_head, vnet);
192 extern struct vnet_list_head vnet_head;
193 extern struct rwlock vnet_rwlock;
194 extern struct sx vnet_sxlock;
195 
196 #define	VNET_LIST_RLOCK()		sx_slock(&vnet_sxlock)
197 #define	VNET_LIST_RLOCK_NOSLEEP()	rw_rlock(&vnet_rwlock)
198 #define	VNET_LIST_RUNLOCK()		sx_sunlock(&vnet_sxlock)
199 #define	VNET_LIST_RUNLOCK_NOSLEEP()	rw_runlock(&vnet_rwlock)
200 
201 /*
202  * Iteration macros to walk the global list of virtual network stacks.
203  */
204 #define	VNET_ITERATOR_DECL(arg)	struct vnet *arg
205 #define	VNET_FOREACH(arg)	LIST_FOREACH((arg), &vnet_head, vnet_le)
206 
207 /*
208  * Virtual network stack memory allocator, which allows global variables to
209  * be automatically instantiated for each network stack instance.
210  */
211 #define	VNET_NAME(n)		vnet_entry_##n
212 #define	VNET_DECLARE(t, n)	extern t VNET_NAME(n)
213 #define	VNET_DEFINE(t, n)	t VNET_NAME(n) __section(VNET_SETNAME) __used
214 #define	_VNET_PTR(b, n)		(__typeof(VNET_NAME(n))*)		\
215 				    ((b) + (uintptr_t)&VNET_NAME(n))
216 
217 #define	_VNET(b, n)		(*_VNET_PTR(b, n))
218 
219 /*
220  * Virtualized global variable accessor macros.
221  */
222 #define	VNET_VNET_PTR(vnet, n)		_VNET_PTR((vnet)->vnet_data_base, n)
223 #define	VNET_VNET(vnet, n)		(*VNET_VNET_PTR((vnet), n))
224 
225 #define	VNET_PTR(n)		VNET_VNET_PTR(curvnet, n)
226 #define	VNET(n)			VNET_VNET(curvnet, n)
227 
228 /*
229  * Virtual network stack allocator interfaces from the kernel linker.
230  */
231 void	*vnet_data_alloc(int size);
232 void	 vnet_data_copy(void *start, int size);
233 void	 vnet_data_free(void *start_arg, int size);
234 
235 /*
236  * Sysctl variants for vnet-virtualized global variables.  Include
237  * <sys/sysctl.h> to expose these definitions.
238  *
239  * Note: SYSCTL_PROC() handler functions will need to resolve pointer
240  * arguments themselves, if required.
241  */
242 #ifdef SYSCTL_OID
243 int	vnet_sysctl_handle_int(SYSCTL_HANDLER_ARGS);
244 int	vnet_sysctl_handle_opaque(SYSCTL_HANDLER_ARGS);
245 int	vnet_sysctl_handle_string(SYSCTL_HANDLER_ARGS);
246 int	vnet_sysctl_handle_uint(SYSCTL_HANDLER_ARGS);
247 
248 #define	SYSCTL_VNET_INT(parent, nbr, name, access, ptr, val, descr)	\
249 	SYSCTL_OID(parent, nbr, name,					\
250 	    CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_VNET|(access),		\
251 	    ptr, val, vnet_sysctl_handle_int, "I", descr)
252 #define	SYSCTL_VNET_PROC(parent, nbr, name, access, ptr, arg, handler,	\
253 	    fmt, descr)							\
254 	CTASSERT(((access) & CTLTYPE) != 0);				\
255 	SYSCTL_OID(parent, nbr, name, CTLFLAG_VNET|(access), ptr, arg, 	\
256 	    handler, fmt, descr)
257 #define	SYSCTL_VNET_OPAQUE(parent, nbr, name, access, ptr, len, fmt,    \
258 	    descr)							\
259 	SYSCTL_OID(parent, nbr, name,					\
260 	    CTLTYPE_OPAQUE|CTLFLAG_VNET|(access), ptr, len, 		\
261 	    vnet_sysctl_handle_opaque, fmt, descr)
262 #define	SYSCTL_VNET_STRING(parent, nbr, name, access, arg, len, descr)	\
263 	SYSCTL_OID(parent, nbr, name,					\
264 	    CTLTYPE_STRING|CTLFLAG_VNET|(access),			\
265 	    arg, len, vnet_sysctl_handle_string, "A", descr)
266 #define	SYSCTL_VNET_STRUCT(parent, nbr, name, access, ptr, type, descr)	\
267 	SYSCTL_OID(parent, nbr, name,					\
268 	    CTLTYPE_OPAQUE|CTLFLAG_VNET|(access), ptr,			\
269 	    sizeof(struct type), vnet_sysctl_handle_opaque, "S," #type,	\
270 	    descr)
271 #define	SYSCTL_VNET_UINT(parent, nbr, name, access, ptr, val, descr)	\
272 	SYSCTL_OID(parent, nbr, name,					\
273 	    CTLTYPE_UINT|CTLFLAG_MPSAFE|CTLFLAG_VNET|(access),		\
274 	    ptr, val, vnet_sysctl_handle_uint, "IU", descr)
275 #define	VNET_SYSCTL_ARG(req, arg1) do {					\
276 	if (arg1 != NULL)						\
277 		arg1 = (void *)(TD_TO_VNET((req)->td)->vnet_data_base +	\
278 		    (uintptr_t)(arg1));					\
279 } while (0)
280 #endif /* SYSCTL_OID */
281 
282 /*
283  * Virtual sysinit mechanism, allowing network stack components to declare
284  * startup and shutdown methods to be run when virtual network stack
285  * instances are created and destroyed.
286  */
287 #include <sys/kernel.h>
288 
289 /*
290  * SYSINIT/SYSUNINIT variants that provide per-vnet constructors and
291  * destructors.
292  */
293 struct vnet_sysinit {
294 	enum sysinit_sub_id	subsystem;
295 	enum sysinit_elem_order	order;
296 	sysinit_cfunc_t		func;
297 	const void		*arg;
298 	TAILQ_ENTRY(vnet_sysinit) link;
299 };
300 
301 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
302 	static struct vnet_sysinit ident ## _vnet_init = {		\
303 		subsystem,						\
304 		order,							\
305 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
306 		(arg)							\
307 	};								\
308 	SYSINIT(vnet_init_ ## ident, subsystem, order,			\
309 	    vnet_register_sysinit, &ident ## _vnet_init);		\
310 	SYSUNINIT(vnet_init_ ## ident, subsystem, order,		\
311 	    vnet_deregister_sysinit, &ident ## _vnet_init)
312 
313 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
314 	static struct vnet_sysinit ident ## _vnet_uninit = {		\
315 		subsystem,						\
316 		order,							\
317 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
318 		(arg)							\
319 	};								\
320 	SYSINIT(vnet_uninit_ ## ident, subsystem, order,		\
321 	    vnet_register_sysuninit, &ident ## _vnet_uninit);		\
322 	SYSUNINIT(vnet_uninit_ ## ident, subsystem, order,		\
323 	    vnet_deregister_sysuninit, &ident ## _vnet_uninit)
324 
325 /*
326  * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
327  */
328 void	 vnet_sysinit(void);
329 void	 vnet_sysuninit(void);
330 
331 /*
332  * Interfaces for managing per-vnet constructors and destructors.
333  */
334 void	vnet_register_sysinit(void *arg);
335 void	vnet_register_sysuninit(void *arg);
336 void	vnet_deregister_sysinit(void *arg);
337 void	vnet_deregister_sysuninit(void *arg);
338 
339 /*
340  * EVENTHANDLER(9) extensions.
341  */
342 #include <sys/eventhandler.h>
343 
344 void	vnet_global_eventhandler_iterator_func(void *, ...);
345 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
346 do {									\
347 	if (IS_DEFAULT_VNET(curvnet)) {					\
348 		(tag) = vimage_eventhandler_register(NULL, #name, func,	\
349 		    arg, priority,					\
350 		    vnet_global_eventhandler_iterator_func);		\
351 	}								\
352 } while(0)
353 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
354 do {									\
355 	if (IS_DEFAULT_VNET(curvnet)) {					\
356 		vimage_eventhandler_register(NULL, #name, func,		\
357 		    arg, priority,					\
358 		    vnet_global_eventhandler_iterator_func);		\
359 	}								\
360 } while(0)
361 
362 #else /* !VIMAGE */
363 
364 /*
365  * Various virtual network stack macros compile to no-ops without VIMAGE.
366  */
367 #define	curvnet			NULL
368 
369 #define	VNET_ASSERT(exp, msg)
370 #define	CURVNET_SET(arg)
371 #define	CURVNET_SET_QUIET(arg)
372 #define	CURVNET_RESTORE()
373 
374 #define	VNET_LIST_RLOCK()
375 #define	VNET_LIST_RLOCK_NOSLEEP()
376 #define	VNET_LIST_RUNLOCK()
377 #define	VNET_LIST_RUNLOCK_NOSLEEP()
378 #define	VNET_ITERATOR_DECL(arg)
379 #define	VNET_FOREACH(arg)
380 
381 #define	IS_DEFAULT_VNET(arg)	1
382 #define	CRED_TO_VNET(cr)	NULL
383 #define	TD_TO_VNET(td)		NULL
384 #define	P_TO_VNET(p)		NULL
385 
386 /*
387  * Versions of the VNET macros that compile to normal global variables and
388  * standard sysctl definitions.
389  */
390 #define	VNET_NAME(n)		n
391 #define	VNET_DECLARE(t, n)	extern t n
392 #define	VNET_DEFINE(t, n)	t n
393 #define	_VNET_PTR(b, n)		&VNET_NAME(n)
394 
395 /*
396  * Virtualized global variable accessor macros.
397  */
398 #define	VNET_VNET_PTR(vnet, n)		(&(n))
399 #define	VNET_VNET(vnet, n)		(n)
400 
401 #define	VNET_PTR(n)		(&(n))
402 #define	VNET(n)			(n)
403 
404 /*
405  * When VIMAGE isn't compiled into the kernel, virtaulized SYSCTLs simply
406  * become normal SYSCTLs.
407  */
408 #ifdef SYSCTL_OID
409 #define	SYSCTL_VNET_INT(parent, nbr, name, access, ptr, val, descr)	\
410 	SYSCTL_INT(parent, nbr, name, access, ptr, val, descr)
411 #define	SYSCTL_VNET_PROC(parent, nbr, name, access, ptr, arg, handler,	\
412 	    fmt, descr)							\
413 	SYSCTL_PROC(parent, nbr, name, access, ptr, arg, handler, fmt,	\
414 	    descr)
415 #define	SYSCTL_VNET_OPAQUE(parent, nbr, name, access, ptr, len, fmt,    \
416 	    descr)							\
417 	SYSCTL_OPAQUE(parent, nbr, name, access, ptr, len, fmt, descr)
418 #define	SYSCTL_VNET_STRING(parent, nbr, name, access, arg, len, descr)	\
419 	SYSCTL_STRING(parent, nbr, name, access, arg, len, descr)
420 #define	SYSCTL_VNET_STRUCT(parent, nbr, name, access, ptr, type, descr)	\
421 	SYSCTL_STRUCT(parent, nbr, name, access, ptr, type, descr)
422 #define	SYSCTL_VNET_UINT(parent, nbr, name, access, ptr, val, descr)	\
423 	SYSCTL_UINT(parent, nbr, name, access, ptr, val, descr)
424 #define	VNET_SYSCTL_ARG(req, arg1)
425 #endif /* SYSCTL_OID */
426 
427 /*
428  * When VIMAGE isn't compiled into the kernel, VNET_SYSINIT/VNET_SYSUNINIT
429  * map into normal sysinits, which have the same ordering properties.
430  */
431 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
432 	SYSINIT(ident, subsystem, order, func, arg)
433 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
434 	SYSUNINIT(ident, subsystem, order, func, arg)
435 
436 /*
437  * Without VIMAGE revert to the default implementation.
438  */
439 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
440 	(tag) = eventhandler_register(NULL, #name, func, arg, priority)
441 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
442 	eventhandler_register(NULL, #name, func, arg, priority)
443 #endif /* VIMAGE */
444 #endif /* _KERNEL */
445 
446 #endif /* !_NET_VNET_H_ */
447