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
70struct 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)	\
113static void				\
114vnet_##name##_init(const void *unused)	\
115{					\
116	VNET_PCPUSTAT_ALLOC(name, M_WAITOK);	\
117}					\
118VNET_SYSINIT(vnet_ ## name ## _init, SI_SUB_INIT_IF,			\
119    SI_ORDER_FIRST, vnet_ ## name ## _init, NULL)
120
121#define	VNET_PCPUSTAT_SYSUNINIT(name)					\
122static void								\
123vnet_##name##_uninit(const void *unused)				\
124{									\
125	VNET_PCPUSTAT_FREE(name);					\
126}									\
127VNET_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)	\
132static int								\
133array##_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}									\
144SYSCTL_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 */
157extern uintptr_t	*__start_set_vnet;
158__GLOBL(__start_set_vnet);
159extern 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 */
168struct vnet *vnet_alloc(void);
169void	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
192void 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
239extern 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 */
251LIST_HEAD(vnet_list_head, vnet);
252extern struct vnet_list_head vnet_head;
253extern struct rwlock vnet_rwlock;
254extern 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 */
304void	*vnet_data_alloc(int size);
305void	 vnet_data_copy(void *start, int size);
306void	 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 */
319struct 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	CTASSERT((subsystem) > SI_SUB_VNET &&				\
329	    (subsystem) <= SI_SUB_VNET_DONE);				\
330	static struct vnet_sysinit ident ## _vnet_init = {		\
331		subsystem,						\
332		order,							\
333		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
334		(arg)							\
335	};								\
336	SYSINIT(vnet_init_ ## ident, subsystem, order,			\
337	    vnet_register_sysinit, &ident ## _vnet_init);		\
338	SYSUNINIT(vnet_init_ ## ident, subsystem, order,		\
339	    vnet_deregister_sysinit, &ident ## _vnet_init)
340
341#define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
342	CTASSERT((subsystem) > SI_SUB_VNET &&				\
343	    (subsystem) <= SI_SUB_VNET_DONE);				\
344	static struct vnet_sysinit ident ## _vnet_uninit = {		\
345		subsystem,						\
346		order,							\
347		(sysinit_cfunc_t)(sysinit_nfunc_t)func,			\
348		(arg)							\
349	};								\
350	SYSINIT(vnet_uninit_ ## ident, subsystem, order,		\
351	    vnet_register_sysuninit, &ident ## _vnet_uninit);		\
352	SYSUNINIT(vnet_uninit_ ## ident, subsystem, order,		\
353	    vnet_deregister_sysuninit, &ident ## _vnet_uninit)
354
355/*
356 * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
357 */
358void	 vnet_sysinit(void);
359void	 vnet_sysuninit(void);
360
361/*
362 * Interfaces for managing per-vnet constructors and destructors.
363 */
364void	vnet_register_sysinit(void *arg);
365void	vnet_register_sysuninit(void *arg);
366void	vnet_deregister_sysinit(void *arg);
367void	vnet_deregister_sysuninit(void *arg);
368
369/*
370 * EVENTHANDLER(9) extensions.
371 */
372#include <sys/eventhandler.h>
373
374void	vnet_global_eventhandler_iterator_func(void *, ...);
375#define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
376do {									\
377	if (IS_DEFAULT_VNET(curvnet)) {					\
378		(tag) = vimage_eventhandler_register(NULL, #name, func,	\
379		    arg, priority,					\
380		    vnet_global_eventhandler_iterator_func);		\
381	}								\
382} while(0)
383#define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
384do {									\
385	if (IS_DEFAULT_VNET(curvnet)) {					\
386		vimage_eventhandler_register(NULL, #name, func,		\
387		    arg, priority,					\
388		    vnet_global_eventhandler_iterator_func);		\
389	}								\
390} while(0)
391
392#else /* !VIMAGE */
393
394/*
395 * Various virtual network stack macros compile to no-ops without VIMAGE.
396 */
397#define	curvnet			NULL
398
399#define	VNET_ASSERT(exp, msg)
400#define	CURVNET_SET(arg)
401#define	CURVNET_SET_QUIET(arg)
402#define	CURVNET_RESTORE()
403
404#define	VNET_LIST_RLOCK()
405#define	VNET_LIST_RLOCK_NOSLEEP()
406#define	VNET_LIST_RUNLOCK()
407#define	VNET_LIST_RUNLOCK_NOSLEEP()
408#define	VNET_ITERATOR_DECL(arg)
409#define	VNET_FOREACH(arg)
410
411#define	IS_DEFAULT_VNET(arg)	1
412#define	CRED_TO_VNET(cr)	NULL
413#define	TD_TO_VNET(td)		NULL
414#define	P_TO_VNET(p)		NULL
415
416/*
417 * Versions of the VNET macros that compile to normal global variables and
418 * standard sysctl definitions.
419 */
420#define	VNET_NAME(n)		n
421#define	VNET_DECLARE(t, n)	extern t n
422#define	VNET_DEFINE(t, n)	struct _hack; t n
423#define	VNET_DEFINE_STATIC(t, n)	static t n
424#define	_VNET_PTR(b, n)		&VNET_NAME(n)
425
426/*
427 * Virtualized global variable accessor macros.
428 */
429#define	VNET_VNET_PTR(vnet, n)		(&(n))
430#define	VNET_VNET(vnet, n)		(n)
431
432#define	VNET_PTR(n)		(&(n))
433#define	VNET(n)			(n)
434
435/*
436 * When VIMAGE isn't compiled into the kernel, VNET_SYSINIT/VNET_SYSUNINIT
437 * map into normal sysinits, which have the same ordering properties.
438 */
439#define	VNET_SYSINIT(ident, subsystem, order, func, arg)		\
440	SYSINIT(ident, subsystem, order, func, arg)
441#define	VNET_SYSUNINIT(ident, subsystem, order, func, arg)		\
442	SYSUNINIT(ident, subsystem, order, func, arg)
443
444/*
445 * Without VIMAGE revert to the default implementation.
446 */
447#define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
448	(tag) = eventhandler_register(NULL, #name, func, arg, priority)
449#define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority)	\
450	eventhandler_register(NULL, #name, func, arg, priority)
451#endif /* VIMAGE */
452#endif /* _KERNEL */
453
454#endif /* !_NET_VNET_H_ */
455