xref: /freebsd/sys/net/vnet.h (revision 89856f7e)
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 	u_int			 vnet_state;	/* SI_SUB_* */
74 	void			*vnet_data_mem;
75 	uintptr_t		 vnet_data_base;
76 };
77 #define	VNET_MAGIC_N	0x3e0d8f29
78 
79 /*
80  * These two virtual network stack allocator definitions are also required
81  * for libkvm so that it can evaluate virtualized global variables.
82  */
83 #define	VNET_SETNAME		"set_vnet"
84 #define	VNET_SYMPREFIX		"vnet_entry_"
85 #endif
86 
87 #ifdef _KERNEL
88 
89 #define	VNET_PCPUSTAT_DECLARE(type, name)	\
90     VNET_DECLARE(counter_u64_t, name[sizeof(type) / sizeof(uint64_t)])
91 
92 #define	VNET_PCPUSTAT_DEFINE(type, name)	\
93     VNET_DEFINE(counter_u64_t, name[sizeof(type) / sizeof(uint64_t)])
94 
95 #define	VNET_PCPUSTAT_ALLOC(name, wait)	\
96     COUNTER_ARRAY_ALLOC(VNET(name), \
97 	sizeof(VNET(name)) / sizeof(counter_u64_t), (wait))
98 
99 #define	VNET_PCPUSTAT_FREE(name)	\
100     COUNTER_ARRAY_FREE(VNET(name), sizeof(VNET(name)) / sizeof(counter_u64_t))
101 
102 #define	VNET_PCPUSTAT_ADD(type, name, f, v)	\
103     counter_u64_add(VNET(name)[offsetof(type, f) / sizeof(uint64_t)], (v))
104 
105 #define	VNET_PCPUSTAT_FETCH(type, name, f)	\
106     counter_u64_fetch(VNET(name)[offsetof(type, f) / sizeof(uint64_t)])
107 
108 #define	VNET_PCPUSTAT_SYSINIT(name)	\
109 static void				\
110 vnet_##name##_init(const void *unused)	\
111 {					\
112 	VNET_PCPUSTAT_ALLOC(name, M_WAITOK);	\
113 }					\
114 VNET_SYSINIT(vnet_ ## name ## _init, SI_SUB_INIT_IF,			\
115     SI_ORDER_FIRST, vnet_ ## name ## _init, NULL)
116 
117 #define	VNET_PCPUSTAT_SYSUNINIT(name)					\
118 static void								\
119 vnet_##name##_uninit(const void *unused)				\
120 {									\
121 	VNET_PCPUSTAT_FREE(name);					\
122 }									\
123 VNET_SYSUNINIT(vnet_ ## name ## _uninit, SI_SUB_INIT_IF,		\
124     SI_ORDER_FIRST, vnet_ ## name ## _uninit, NULL)
125 
126 #ifdef SYSCTL_OID
127 #define	SYSCTL_VNET_PCPUSTAT(parent, nbr, name, type, array, desc)	\
128 static int								\
129 array##_sysctl(SYSCTL_HANDLER_ARGS)					\
130 {									\
131 	type s;								\
132 	CTASSERT((sizeof(type) / sizeof(uint64_t)) ==			\
133 	    (sizeof(VNET(array)) / sizeof(counter_u64_t)));		\
134 	COUNTER_ARRAY_COPY(VNET(array), &s, sizeof(type) / sizeof(uint64_t));\
135 	if (req->newptr)						\
136 		COUNTER_ARRAY_ZERO(VNET(array),				\
137 		    sizeof(type) / sizeof(uint64_t));			\
138 	return (SYSCTL_OUT(req, &s, sizeof(type)));			\
139 }									\
140 SYSCTL_PROC(parent, nbr, name, CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_RW, \
141     NULL, 0, array ## _sysctl, "I", desc)
142 #endif /* SYSCTL_OID */
143 
144 #ifdef VIMAGE
145 #include <sys/lock.h>
146 #include <sys/proc.h>			/* for struct thread */
147 #include <sys/rwlock.h>
148 #include <sys/sx.h>
149 
150 /*
151  * Location of the kernel's 'set_vnet' linker set.
152  */
153 extern uintptr_t	*__start_set_vnet;
154 __GLOBL(__start_set_vnet);
155 extern uintptr_t	*__stop_set_vnet;
156 __GLOBL(__stop_set_vnet);
157 
158 #define	VNET_START	(uintptr_t)&__start_set_vnet
159 #define	VNET_STOP	(uintptr_t)&__stop_set_vnet
160 
161 /*
162  * Functions to allocate and destroy virtual network stacks.
163  */
164 struct vnet *vnet_alloc(void);
165 void	vnet_destroy(struct vnet *vnet);
166 
167 /*
168  * The current virtual network stack -- we may wish to move this to struct
169  * pcpu in the future.
170  */
171 #define	curvnet	curthread->td_vnet
172 
173 /*
174  * Various macros -- get and set the current network stack, but also
175  * assertions.
176  */
177 #if defined(INVARIANTS) || defined(VNET_DEBUG)
178 #define	VNET_ASSERT(exp, msg)	do {					\
179 	if (!(exp))							\
180 		panic msg;						\
181 } while (0)
182 #else
183 #define	VNET_ASSERT(exp, msg)	do {					\
184 } while (0)
185 #endif
186 
187 #ifdef VNET_DEBUG
188 void vnet_log_recursion(struct vnet *, const char *, int);
189 
190 #define	CURVNET_SET_QUIET(arg)						\
191 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
192 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
193 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
194 	struct vnet *saved_vnet = curvnet;				\
195 	const char *saved_vnet_lpush = curthread->td_vnet_lpush;	\
196 	curvnet = arg;							\
197 	curthread->td_vnet_lpush = __func__;
198 
199 #define	CURVNET_SET_VERBOSE(arg)					\
200 	CURVNET_SET_QUIET(arg)						\
201 	if (saved_vnet)							\
202 		vnet_log_recursion(saved_vnet, saved_vnet_lpush, __LINE__);
203 
204 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
205 
206 #define	CURVNET_RESTORE()						\
207 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
208 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
209 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
210 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
211 	curvnet = saved_vnet;						\
212 	curthread->td_vnet_lpush = saved_vnet_lpush;
213 #else /* !VNET_DEBUG */
214 
215 #define	CURVNET_SET_QUIET(arg)						\
216 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
217 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
218 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
219 	struct vnet *saved_vnet = curvnet;				\
220 	curvnet = arg;
221 
222 #define	CURVNET_SET_VERBOSE(arg)					\
223 	CURVNET_SET_QUIET(arg)
224 
225 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
226 
227 #define	CURVNET_RESTORE()						\
228 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
229 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
230 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
231 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
232 	curvnet = saved_vnet;
233 #endif /* VNET_DEBUG */
234 
235 extern struct vnet *vnet0;
236 #define	IS_DEFAULT_VNET(arg)	((arg) == vnet0)
237 
238 #define	CRED_TO_VNET(cr)	(cr)->cr_prison->pr_vnet
239 #define	TD_TO_VNET(td)		CRED_TO_VNET((td)->td_ucred)
240 #define	P_TO_VNET(p)		CRED_TO_VNET((p)->p_ucred)
241 
242 /*
243  * Global linked list of all virtual network stacks, along with read locks to
244  * access it.  If a caller may sleep while accessing the list, it must use
245  * the sleepable lock macros.
246  */
247 LIST_HEAD(vnet_list_head, vnet);
248 extern struct vnet_list_head vnet_head;
249 extern struct rwlock vnet_rwlock;
250 extern struct sx vnet_sxlock;
251 
252 #define	VNET_LIST_RLOCK()		sx_slock(&vnet_sxlock)
253 #define	VNET_LIST_RLOCK_NOSLEEP()	rw_rlock(&vnet_rwlock)
254 #define	VNET_LIST_RUNLOCK()		sx_sunlock(&vnet_sxlock)
255 #define	VNET_LIST_RUNLOCK_NOSLEEP()	rw_runlock(&vnet_rwlock)
256 
257 /*
258  * Iteration macros to walk the global list of virtual network stacks.
259  */
260 #define	VNET_ITERATOR_DECL(arg)	struct vnet *arg
261 #define	VNET_FOREACH(arg)	LIST_FOREACH((arg), &vnet_head, vnet_le)
262 
263 /*
264  * Virtual network stack memory allocator, which allows global variables to
265  * be automatically instantiated for each network stack instance.
266  */
267 #define	VNET_NAME(n)		vnet_entry_##n
268 #define	VNET_DECLARE(t, n)	extern t VNET_NAME(n)
269 #define	VNET_DEFINE(t, n)	t VNET_NAME(n) __section(VNET_SETNAME) __used
270 #define	_VNET_PTR(b, n)		(__typeof(VNET_NAME(n))*)		\
271 				    ((b) + (uintptr_t)&VNET_NAME(n))
272 
273 #define	_VNET(b, n)		(*_VNET_PTR(b, n))
274 
275 /*
276  * Virtualized global variable accessor macros.
277  */
278 #define	VNET_VNET_PTR(vnet, n)		_VNET_PTR((vnet)->vnet_data_base, n)
279 #define	VNET_VNET(vnet, n)		(*VNET_VNET_PTR((vnet), n))
280 
281 #define	VNET_PTR(n)		VNET_VNET_PTR(curvnet, n)
282 #define	VNET(n)			VNET_VNET(curvnet, n)
283 
284 /*
285  * Virtual network stack allocator interfaces from the kernel linker.
286  */
287 void	*vnet_data_alloc(int size);
288 void	 vnet_data_copy(void *start, int size);
289 void	 vnet_data_free(void *start_arg, int size);
290 
291 /*
292  * Virtual sysinit mechanism, allowing network stack components to declare
293  * startup and shutdown methods to be run when virtual network stack
294  * instances are created and destroyed.
295  */
296 #include <sys/kernel.h>
297 
298 /*
299  * SYSINIT/SYSUNINIT variants that provide per-vnet constructors and
300  * destructors.
301  */
302 struct vnet_sysinit {
303 	enum sysinit_sub_id	subsystem;
304 	enum sysinit_elem_order	order;
305 	sysinit_cfunc_t		func;
306 	const void		*arg;
307 	TAILQ_ENTRY(vnet_sysinit) link;
308 };
309 
310 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
311 	static struct vnet_sysinit ident ## _vnet_init = {		\
312 		subsystem,						\
313 		order,							\
314 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
315 		(arg)							\
316 	};								\
317 	SYSINIT(vnet_init_ ## ident, subsystem, order,			\
318 	    vnet_register_sysinit, &ident ## _vnet_init);		\
319 	SYSUNINIT(vnet_init_ ## ident, subsystem, order,		\
320 	    vnet_deregister_sysinit, &ident ## _vnet_init)
321 
322 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
323 	static struct vnet_sysinit ident ## _vnet_uninit = {		\
324 		subsystem,						\
325 		order,							\
326 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
327 		(arg)							\
328 	};								\
329 	SYSINIT(vnet_uninit_ ## ident, subsystem, order,		\
330 	    vnet_register_sysuninit, &ident ## _vnet_uninit);		\
331 	SYSUNINIT(vnet_uninit_ ## ident, subsystem, order,		\
332 	    vnet_deregister_sysuninit, &ident ## _vnet_uninit)
333 
334 /*
335  * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
336  */
337 void	 vnet_sysinit(void);
338 void	 vnet_sysuninit(void);
339 
340 /*
341  * Interfaces for managing per-vnet constructors and destructors.
342  */
343 void	vnet_register_sysinit(void *arg);
344 void	vnet_register_sysuninit(void *arg);
345 void	vnet_deregister_sysinit(void *arg);
346 void	vnet_deregister_sysuninit(void *arg);
347 
348 /*
349  * EVENTHANDLER(9) extensions.
350  */
351 #include <sys/eventhandler.h>
352 
353 void	vnet_global_eventhandler_iterator_func(void *, ...);
354 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
355 do {									\
356 	if (IS_DEFAULT_VNET(curvnet)) {					\
357 		(tag) = vimage_eventhandler_register(NULL, #name, func,	\
358 		    arg, priority,					\
359 		    vnet_global_eventhandler_iterator_func);		\
360 	}								\
361 } while(0)
362 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
363 do {									\
364 	if (IS_DEFAULT_VNET(curvnet)) {					\
365 		vimage_eventhandler_register(NULL, #name, func,		\
366 		    arg, priority,					\
367 		    vnet_global_eventhandler_iterator_func);		\
368 	}								\
369 } while(0)
370 
371 #else /* !VIMAGE */
372 
373 /*
374  * Various virtual network stack macros compile to no-ops without VIMAGE.
375  */
376 #define	curvnet			NULL
377 
378 #define	VNET_ASSERT(exp, msg)
379 #define	CURVNET_SET(arg)
380 #define	CURVNET_SET_QUIET(arg)
381 #define	CURVNET_RESTORE()
382 
383 #define	VNET_LIST_RLOCK()
384 #define	VNET_LIST_RLOCK_NOSLEEP()
385 #define	VNET_LIST_RUNLOCK()
386 #define	VNET_LIST_RUNLOCK_NOSLEEP()
387 #define	VNET_ITERATOR_DECL(arg)
388 #define	VNET_FOREACH(arg)
389 
390 #define	IS_DEFAULT_VNET(arg)	1
391 #define	CRED_TO_VNET(cr)	NULL
392 #define	TD_TO_VNET(td)		NULL
393 #define	P_TO_VNET(p)		NULL
394 
395 /*
396  * Versions of the VNET macros that compile to normal global variables and
397  * standard sysctl definitions.
398  */
399 #define	VNET_NAME(n)		n
400 #define	VNET_DECLARE(t, n)	extern t n
401 #define	VNET_DEFINE(t, n)	t n
402 #define	_VNET_PTR(b, n)		&VNET_NAME(n)
403 
404 /*
405  * Virtualized global variable accessor macros.
406  */
407 #define	VNET_VNET_PTR(vnet, n)		(&(n))
408 #define	VNET_VNET(vnet, n)		(n)
409 
410 #define	VNET_PTR(n)		(&(n))
411 #define	VNET(n)			(n)
412 
413 /*
414  * When VIMAGE isn't compiled into the kernel, VNET_SYSINIT/VNET_SYSUNINIT
415  * map into normal sysinits, which have the same ordering properties.
416  */
417 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
418 	SYSINIT(ident, subsystem, order, func, arg)
419 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
420 	SYSUNINIT(ident, subsystem, order, func, arg)
421 
422 /*
423  * Without VIMAGE revert to the default implementation.
424  */
425 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
426 	(tag) = eventhandler_register(NULL, #name, func, arg, priority)
427 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
428 	eventhandler_register(NULL, #name, func, arg, priority)
429 #endif /* VIMAGE */
430 #endif /* _KERNEL */
431 
432 #endif /* !_NET_VNET_H_ */
433