xref: /freebsd/sys/net/vnet.h (revision 123af6ec70016f5556da5972d4d63c7d175c06d3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2006-2009 University of Zagreb
5  * Copyright (c) 2006-2009 FreeBSD Foundation
6  * All rights reserved.
7  *
8  * This software was developed by the University of Zagreb and the
9  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
10  * FreeBSD Foundation.
11  *
12  * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
13  * Copyright (c) 2009 Robert N. M. Watson
14  * All rights reserved.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  * $FreeBSD$
38  */
39 
40 /*-
41  * This header file defines several sets of interfaces supporting virtualized
42  * network stacks:
43  *
44  * - Definition of 'struct vnet' and functions and macros to allocate/free/
45  *   manipulate it.
46  *
47  * - A virtual network stack memory allocator, which provides support for
48  *   virtualized global variables via a special linker set, set_vnet.
49  *
50  * - Virtualized sysinits/sysuninits, which allow constructors and
51  *   destructors to be run for each network stack subsystem as virtual
52  *   instances are created and destroyed.
53  *
54  * If VIMAGE isn't compiled into the kernel, virtualized global variables
55  * compile to normal global variables, and virtualized sysinits to regular
56  * sysinits.
57  */
58 
59 #ifndef _NET_VNET_H_
60 #define	_NET_VNET_H_
61 
62 /*
63  * struct vnet describes a virtualized network stack, and is primarily a
64  * pointer to storage for virtualized global variables.  Expose to userspace
65  * as required for libkvm.
66  */
67 #if defined(_KERNEL) || defined(_WANT_VNET)
68 #include <sys/queue.h>
69 
70 struct vnet {
71 	LIST_ENTRY(vnet)	 vnet_le;	/* all vnets list */
72 	u_int			 vnet_magic_n;
73 	u_int			 vnet_ifcnt;
74 	u_int			 vnet_sockcnt;
75 	u_int			 vnet_state;	/* SI_SUB_* */
76 	void			*vnet_data_mem;
77 	uintptr_t		 vnet_data_base;
78 };
79 #define	VNET_MAGIC_N	0x3e0d8f29
80 
81 /*
82  * These two virtual network stack allocator definitions are also required
83  * for libkvm so that it can evaluate virtualized global variables.
84  */
85 #define	VNET_SETNAME		"set_vnet"
86 #define	VNET_SYMPREFIX		"vnet_entry_"
87 #endif
88 
89 #ifdef _KERNEL
90 
91 #define	VNET_PCPUSTAT_DECLARE(type, name)	\
92     VNET_DECLARE(counter_u64_t, name[sizeof(type) / sizeof(uint64_t)])
93 
94 #define	VNET_PCPUSTAT_DEFINE(type, name)	\
95     VNET_DEFINE(counter_u64_t, name[sizeof(type) / sizeof(uint64_t)])
96 #define	VNET_PCPUSTAT_DEFINE_STATIC(type, name)	\
97     VNET_DEFINE_STATIC(counter_u64_t, name[sizeof(type) / sizeof(uint64_t)])
98 
99 #define	VNET_PCPUSTAT_ALLOC(name, wait)	\
100     COUNTER_ARRAY_ALLOC(VNET(name), \
101 	sizeof(VNET(name)) / sizeof(counter_u64_t), (wait))
102 
103 #define	VNET_PCPUSTAT_FREE(name)	\
104     COUNTER_ARRAY_FREE(VNET(name), sizeof(VNET(name)) / sizeof(counter_u64_t))
105 
106 #define	VNET_PCPUSTAT_ADD(type, name, f, v)	\
107     counter_u64_add(VNET(name)[offsetof(type, f) / sizeof(uint64_t)], (v))
108 
109 #define	VNET_PCPUSTAT_FETCH(type, name, f)	\
110     counter_u64_fetch(VNET(name)[offsetof(type, f) / sizeof(uint64_t)])
111 
112 #define	VNET_PCPUSTAT_SYSINIT(name)	\
113 static void				\
114 vnet_##name##_init(const void *unused)	\
115 {					\
116 	VNET_PCPUSTAT_ALLOC(name, M_WAITOK);	\
117 }					\
118 VNET_SYSINIT(vnet_ ## name ## _init, SI_SUB_INIT_IF,			\
119     SI_ORDER_FIRST, vnet_ ## name ## _init, NULL)
120 
121 #define	VNET_PCPUSTAT_SYSUNINIT(name)					\
122 static void								\
123 vnet_##name##_uninit(const void *unused)				\
124 {									\
125 	VNET_PCPUSTAT_FREE(name);					\
126 }									\
127 VNET_SYSUNINIT(vnet_ ## name ## _uninit, SI_SUB_INIT_IF,		\
128     SI_ORDER_FIRST, vnet_ ## name ## _uninit, NULL)
129 
130 #ifdef SYSCTL_OID
131 #define	SYSCTL_VNET_PCPUSTAT(parent, nbr, name, type, array, desc)	\
132 static int								\
133 array##_sysctl(SYSCTL_HANDLER_ARGS)					\
134 {									\
135 	type s;								\
136 	CTASSERT((sizeof(type) / sizeof(uint64_t)) ==			\
137 	    (sizeof(VNET(array)) / sizeof(counter_u64_t)));		\
138 	COUNTER_ARRAY_COPY(VNET(array), &s, sizeof(type) / sizeof(uint64_t));\
139 	if (req->newptr)						\
140 		COUNTER_ARRAY_ZERO(VNET(array),				\
141 		    sizeof(type) / sizeof(uint64_t));			\
142 	return (SYSCTL_OUT(req, &s, sizeof(type)));			\
143 }									\
144 SYSCTL_PROC(parent, nbr, name, CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_RW, \
145     NULL, 0, array ## _sysctl, "I", desc)
146 #endif /* SYSCTL_OID */
147 
148 #ifdef VIMAGE
149 #include <sys/lock.h>
150 #include <sys/proc.h>			/* for struct thread */
151 #include <sys/rwlock.h>
152 #include <sys/sx.h>
153 
154 /*
155  * Location of the kernel's 'set_vnet' linker set.
156  */
157 extern uintptr_t	*__start_set_vnet;
158 __GLOBL(__start_set_vnet);
159 extern uintptr_t	*__stop_set_vnet;
160 __GLOBL(__stop_set_vnet);
161 
162 #define	VNET_START	(uintptr_t)&__start_set_vnet
163 #define	VNET_STOP	(uintptr_t)&__stop_set_vnet
164 
165 /*
166  * Functions to allocate and destroy virtual network stacks.
167  */
168 struct vnet *vnet_alloc(void);
169 void	vnet_destroy(struct vnet *vnet);
170 
171 /*
172  * The current virtual network stack -- we may wish to move this to struct
173  * pcpu in the future.
174  */
175 #define	curvnet	curthread->td_vnet
176 
177 /*
178  * Various macros -- get and set the current network stack, but also
179  * assertions.
180  */
181 #if defined(INVARIANTS) || defined(VNET_DEBUG)
182 #define	VNET_ASSERT(exp, msg)	do {					\
183 	if (!(exp))							\
184 		panic msg;						\
185 } while (0)
186 #else
187 #define	VNET_ASSERT(exp, msg)	do {					\
188 } while (0)
189 #endif
190 
191 #ifdef VNET_DEBUG
192 void vnet_log_recursion(struct vnet *, const char *, int);
193 
194 #define	CURVNET_SET_QUIET(arg)						\
195 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
196 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
197 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
198 	struct vnet *saved_vnet = curvnet;				\
199 	const char *saved_vnet_lpush = curthread->td_vnet_lpush;	\
200 	curvnet = arg;							\
201 	curthread->td_vnet_lpush = __func__;
202 
203 #define	CURVNET_SET_VERBOSE(arg)					\
204 	CURVNET_SET_QUIET(arg)						\
205 	if (saved_vnet)							\
206 		vnet_log_recursion(saved_vnet, saved_vnet_lpush, __LINE__);
207 
208 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
209 
210 #define	CURVNET_RESTORE()						\
211 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
212 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
213 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
214 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
215 	curvnet = saved_vnet;						\
216 	curthread->td_vnet_lpush = saved_vnet_lpush;
217 #else /* !VNET_DEBUG */
218 
219 #define	CURVNET_SET_QUIET(arg)						\
220 	VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
221 	    ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p",		\
222 	    __FILE__, __LINE__, __func__, curvnet, (arg)));		\
223 	struct vnet *saved_vnet = curvnet;				\
224 	curvnet = arg;
225 
226 #define	CURVNET_SET_VERBOSE(arg)					\
227 	CURVNET_SET_QUIET(arg)
228 
229 #define	CURVNET_SET(arg)	CURVNET_SET_VERBOSE(arg)
230 
231 #define	CURVNET_RESTORE()						\
232 	VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL ||		\
233 	    saved_vnet->vnet_magic_n == VNET_MAGIC_N),			\
234 	    ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p",	\
235 	    __FILE__, __LINE__, __func__, curvnet, saved_vnet));	\
236 	curvnet = saved_vnet;
237 #endif /* VNET_DEBUG */
238 
239 extern struct vnet *vnet0;
240 #define	IS_DEFAULT_VNET(arg)	((arg) == vnet0)
241 
242 #define	CRED_TO_VNET(cr)	(cr)->cr_prison->pr_vnet
243 #define	TD_TO_VNET(td)		CRED_TO_VNET((td)->td_ucred)
244 #define	P_TO_VNET(p)		CRED_TO_VNET((p)->p_ucred)
245 
246 /*
247  * Global linked list of all virtual network stacks, along with read locks to
248  * access it.  If a caller may sleep while accessing the list, it must use
249  * the sleepable lock macros.
250  */
251 LIST_HEAD(vnet_list_head, vnet);
252 extern struct vnet_list_head vnet_head;
253 extern struct rwlock vnet_rwlock;
254 extern struct sx vnet_sxlock;
255 
256 #define	VNET_LIST_RLOCK()		sx_slock(&vnet_sxlock)
257 #define	VNET_LIST_RLOCK_NOSLEEP()	rw_rlock(&vnet_rwlock)
258 #define	VNET_LIST_RUNLOCK()		sx_sunlock(&vnet_sxlock)
259 #define	VNET_LIST_RUNLOCK_NOSLEEP()	rw_runlock(&vnet_rwlock)
260 
261 /*
262  * Iteration macros to walk the global list of virtual network stacks.
263  */
264 #define	VNET_ITERATOR_DECL(arg)	struct vnet *arg
265 #define	VNET_FOREACH(arg)	LIST_FOREACH((arg), &vnet_head, vnet_le)
266 
267 /*
268  * Virtual network stack memory allocator, which allows global variables to
269  * be automatically instantiated for each network stack instance.
270  */
271 #define	VNET_NAME(n)		vnet_entry_##n
272 #define	VNET_DECLARE(t, n)	extern t VNET_NAME(n)
273 /* struct _hack is to stop this from being used with static data */
274 #define	VNET_DEFINE(t, n)	\
275     struct _hack; t VNET_NAME(n) __section(VNET_SETNAME) __used
276 #if defined(KLD_MODULE) && (defined(__aarch64__) || defined(__riscv))
277 /*
278  * As with DPCPU_DEFINE_STATIC we are unable to mark this data as static
279  * in modules on some architectures.
280  */
281 #define	VNET_DEFINE_STATIC(t, n) \
282     t VNET_NAME(n) __section(VNET_SETNAME) __used
283 #else
284 #define	VNET_DEFINE_STATIC(t, n) \
285     static t VNET_NAME(n) __section(VNET_SETNAME) __used
286 #endif
287 #define	_VNET_PTR(b, n)		(__typeof(VNET_NAME(n))*)		\
288 				    ((b) + (uintptr_t)&VNET_NAME(n))
289 
290 #define	_VNET(b, n)		(*_VNET_PTR(b, n))
291 
292 /*
293  * Virtualized global variable accessor macros.
294  */
295 #define	VNET_VNET_PTR(vnet, n)		_VNET_PTR((vnet)->vnet_data_base, n)
296 #define	VNET_VNET(vnet, n)		(*VNET_VNET_PTR((vnet), n))
297 
298 #define	VNET_PTR(n)		VNET_VNET_PTR(curvnet, n)
299 #define	VNET(n)			VNET_VNET(curvnet, n)
300 
301 /*
302  * Virtual network stack allocator interfaces from the kernel linker.
303  */
304 void	*vnet_data_alloc(int size);
305 void	 vnet_data_copy(void *start, int size);
306 void	 vnet_data_free(void *start_arg, int size);
307 
308 /*
309  * Virtual sysinit mechanism, allowing network stack components to declare
310  * startup and shutdown methods to be run when virtual network stack
311  * instances are created and destroyed.
312  */
313 #include <sys/kernel.h>
314 
315 /*
316  * SYSINIT/SYSUNINIT variants that provide per-vnet constructors and
317  * destructors.
318  */
319 struct vnet_sysinit {
320 	enum sysinit_sub_id	subsystem;
321 	enum sysinit_elem_order	order;
322 	sysinit_cfunc_t		func;
323 	const void		*arg;
324 	TAILQ_ENTRY(vnet_sysinit) link;
325 };
326 
327 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
328 	static struct vnet_sysinit ident ## _vnet_init = {		\
329 		subsystem,						\
330 		order,							\
331 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
332 		(arg)							\
333 	};								\
334 	SYSINIT(vnet_init_ ## ident, subsystem, order,			\
335 	    vnet_register_sysinit, &ident ## _vnet_init);		\
336 	SYSUNINIT(vnet_init_ ## ident, subsystem, order,		\
337 	    vnet_deregister_sysinit, &ident ## _vnet_init)
338 
339 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
340 	static struct vnet_sysinit ident ## _vnet_uninit = {		\
341 		subsystem,						\
342 		order,							\
343 		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
344 		(arg)							\
345 	};								\
346 	SYSINIT(vnet_uninit_ ## ident, subsystem, order,		\
347 	    vnet_register_sysuninit, &ident ## _vnet_uninit);		\
348 	SYSUNINIT(vnet_uninit_ ## ident, subsystem, order,		\
349 	    vnet_deregister_sysuninit, &ident ## _vnet_uninit)
350 
351 /*
352  * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
353  */
354 void	 vnet_sysinit(void);
355 void	 vnet_sysuninit(void);
356 
357 /*
358  * Interfaces for managing per-vnet constructors and destructors.
359  */
360 void	vnet_register_sysinit(void *arg);
361 void	vnet_register_sysuninit(void *arg);
362 void	vnet_deregister_sysinit(void *arg);
363 void	vnet_deregister_sysuninit(void *arg);
364 
365 /*
366  * EVENTHANDLER(9) extensions.
367  */
368 #include <sys/eventhandler.h>
369 
370 void	vnet_global_eventhandler_iterator_func(void *, ...);
371 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
372 do {									\
373 	if (IS_DEFAULT_VNET(curvnet)) {					\
374 		(tag) = vimage_eventhandler_register(NULL, #name, func,	\
375 		    arg, priority,					\
376 		    vnet_global_eventhandler_iterator_func);		\
377 	}								\
378 } while(0)
379 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
380 do {									\
381 	if (IS_DEFAULT_VNET(curvnet)) {					\
382 		vimage_eventhandler_register(NULL, #name, func,		\
383 		    arg, priority,					\
384 		    vnet_global_eventhandler_iterator_func);		\
385 	}								\
386 } while(0)
387 
388 #else /* !VIMAGE */
389 
390 /*
391  * Various virtual network stack macros compile to no-ops without VIMAGE.
392  */
393 #define	curvnet			NULL
394 
395 #define	VNET_ASSERT(exp, msg)
396 #define	CURVNET_SET(arg)
397 #define	CURVNET_SET_QUIET(arg)
398 #define	CURVNET_RESTORE()
399 
400 #define	VNET_LIST_RLOCK()
401 #define	VNET_LIST_RLOCK_NOSLEEP()
402 #define	VNET_LIST_RUNLOCK()
403 #define	VNET_LIST_RUNLOCK_NOSLEEP()
404 #define	VNET_ITERATOR_DECL(arg)
405 #define	VNET_FOREACH(arg)
406 
407 #define	IS_DEFAULT_VNET(arg)	1
408 #define	CRED_TO_VNET(cr)	NULL
409 #define	TD_TO_VNET(td)		NULL
410 #define	P_TO_VNET(p)		NULL
411 
412 /*
413  * Versions of the VNET macros that compile to normal global variables and
414  * standard sysctl definitions.
415  */
416 #define	VNET_NAME(n)		n
417 #define	VNET_DECLARE(t, n)	extern t n
418 #define	VNET_DEFINE(t, n)	struct _hack; t n
419 #define	VNET_DEFINE_STATIC(t, n)	static t n
420 #define	_VNET_PTR(b, n)		&VNET_NAME(n)
421 
422 /*
423  * Virtualized global variable accessor macros.
424  */
425 #define	VNET_VNET_PTR(vnet, n)		(&(n))
426 #define	VNET_VNET(vnet, n)		(n)
427 
428 #define	VNET_PTR(n)		(&(n))
429 #define	VNET(n)			(n)
430 
431 /*
432  * When VIMAGE isn't compiled into the kernel, VNET_SYSINIT/VNET_SYSUNINIT
433  * map into normal sysinits, which have the same ordering properties.
434  */
435 #define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
436 	SYSINIT(ident, subsystem, order, func, arg)
437 #define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
438 	SYSUNINIT(ident, subsystem, order, func, arg)
439 
440 /*
441  * Without VIMAGE revert to the default implementation.
442  */
443 #define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
444 	(tag) = eventhandler_register(NULL, #name, func, arg, priority)
445 #define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
446 	eventhandler_register(NULL, #name, func, arg, priority)
447 #endif /* VIMAGE */
448 #endif /* _KERNEL */
449 
450 #endif /* !_NET_VNET_H_ */
451