master
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
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
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 <machine/param.h> /* for CACHE_LINE_SIZE */
67#include <sys/queue.h>
68
69struct vnet {
70 LIST_ENTRY(vnet) vnet_le; /* all vnets list */
71 u_int vnet_magic_n;
72 u_int vnet_ifcnt;
73 u_int vnet_sockcnt;
74 u_int vnet_state; /* SI_SUB_* */
75 void *vnet_data_mem;
76 uintptr_t vnet_data_base;
77 bool vnet_shutdown; /* Shutdown in progress. */
78} __aligned(CACHE_LINE_SIZE);
79#define VNET_MAGIC_N 0x5e4a6f28
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, \
145 CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_NEEDGIANT, \
146 NULL, 0, array ## _sysctl, "I", desc)
147#endif /* SYSCTL_OID */
148
149#ifdef VIMAGE
150#include <sys/lock.h>
151#include <sys/proc.h> /* for struct thread */
152#include <sys/rwlock.h>
153#include <sys/sx.h>
154
155/*
156 * Location of the kernel's 'set_vnet' linker set.
157 */
158extern uintptr_t *__start_set_vnet;
159__GLOBL(__start_set_vnet);
160extern uintptr_t *__stop_set_vnet;
161__GLOBL(__stop_set_vnet);
162
163#define VNET_START (uintptr_t)&__start_set_vnet
164#define VNET_STOP (uintptr_t)&__stop_set_vnet
165
166/*
167 * Functions to allocate and destroy virtual network stacks.
168 */
169struct vnet *vnet_alloc(void);
170void vnet_destroy(struct vnet *vnet);
171
172/*
173 * The current virtual network stack -- we may wish to move this to struct
174 * pcpu in the future.
175 */
176#define curvnet curthread->td_vnet
177
178/*
179 * Various macros -- get and set the current network stack, but also
180 * assertions.
181 */
182#if defined(INVARIANTS) || defined(VNET_DEBUG)
183#define VNET_ASSERT(exp, msg) do { \
184 if (!(exp)) \
185 panic msg; \
186} while (0)
187#else
188#define VNET_ASSERT(exp, msg) do { \
189} while (0)
190#endif
191
192#ifdef VNET_DEBUG
193void vnet_log_recursion(struct vnet *, const char *, int);
194
195#define CURVNET_SET_QUIET(arg) \
196 VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
197 ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p", \
198 __FILE__, __LINE__, __func__, curvnet, (arg))); \
199 struct vnet *saved_vnet = curvnet; \
200 const char *saved_vnet_lpush = curthread->td_vnet_lpush; \
201 curvnet = arg; \
202 curthread->td_vnet_lpush = __func__;
203
204#define CURVNET_SET_VERBOSE(arg) \
205 CURVNET_SET_QUIET(arg) \
206 if (saved_vnet) \
207 vnet_log_recursion(saved_vnet, saved_vnet_lpush, __LINE__);
208
209#define CURVNET_SET(arg) CURVNET_SET_VERBOSE(arg)
210
211#define CURVNET_RESTORE() \
212 VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL || \
213 saved_vnet->vnet_magic_n == VNET_MAGIC_N), \
214 ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p", \
215 __FILE__, __LINE__, __func__, curvnet, saved_vnet)); \
216 curvnet = saved_vnet; \
217 curthread->td_vnet_lpush = saved_vnet_lpush;
218#else /* !VNET_DEBUG */
219
220#define CURVNET_SET_QUIET(arg) \
221 VNET_ASSERT((arg) != NULL && (arg)->vnet_magic_n == VNET_MAGIC_N, \
222 ("CURVNET_SET at %s:%d %s() curvnet=%p vnet=%p", \
223 __FILE__, __LINE__, __func__, curvnet, (arg))); \
224 struct vnet *saved_vnet = curvnet; \
225 curvnet = arg;
226
227#define CURVNET_SET_VERBOSE(arg) \
228 CURVNET_SET_QUIET(arg)
229
230#define CURVNET_SET(arg) CURVNET_SET_VERBOSE(arg)
231
232#define CURVNET_RESTORE() \
233 VNET_ASSERT(curvnet != NULL && (saved_vnet == NULL || \
234 saved_vnet->vnet_magic_n == VNET_MAGIC_N), \
235 ("CURVNET_RESTORE at %s:%d %s() curvnet=%p saved_vnet=%p", \
236 __FILE__, __LINE__, __func__, curvnet, saved_vnet)); \
237 curvnet = saved_vnet;
238#endif /* VNET_DEBUG */
239
240#define CURVNET_ASSERT_SET() \
241 VNET_ASSERT(curvnet != NULL, ("vnet is not set at %s:%d %s()", \
242 __FILE__, __LINE__, __func__))
243
244extern struct vnet *vnet0;
245#define IS_DEFAULT_VNET(arg) ((arg) == vnet0)
246
247#define CRED_TO_VNET(cr) (cr)->cr_prison->pr_vnet
248#define TD_TO_VNET(td) CRED_TO_VNET((td)->td_ucred)
249#define P_TO_VNET(p) CRED_TO_VNET((p)->p_ucred)
250
251/*
252 * Global linked list of all virtual network stacks, along with read locks to
253 * access it. If a caller may sleep while accessing the list, it must use
254 * the sleepable lock macros.
255 */
256LIST_HEAD(vnet_list_head, vnet);
257extern struct vnet_list_head vnet_head;
258extern struct rwlock vnet_rwlock;
259extern struct sx vnet_sxlock;
260
261#define VNET_LIST_RLOCK() sx_slock(&vnet_sxlock)
262#define VNET_LIST_RLOCK_NOSLEEP() rw_rlock(&vnet_rwlock)
263#define VNET_LIST_RUNLOCK() sx_sunlock(&vnet_sxlock)
264#define VNET_LIST_RUNLOCK_NOSLEEP() rw_runlock(&vnet_rwlock)
265
266/*
267 * Iteration macros to walk the global list of virtual network stacks.
268 */
269#define VNET_ITERATOR_DECL(arg) struct vnet *arg
270#define VNET_FOREACH(arg) LIST_FOREACH((arg), &vnet_head, vnet_le)
271
272/*
273 * Virtual network stack memory allocator, which allows global variables to
274 * be automatically instantiated for each network stack instance.
275 */
276#define VNET_NAME(n) vnet_entry_##n
277#define VNET_DECLARE(t, n) extern t VNET_NAME(n)
278/* struct _hack is to stop this from being used with static data */
279#define VNET_DEFINE(t, n) \
280 struct _hack; t VNET_NAME(n) __section(VNET_SETNAME) __used
281#if defined(KLD_MODULE) && (defined(__aarch64__) || defined(__riscv) \
282 || defined(__powerpc64__) || defined(__i386__))
283/*
284 * As with DPCPU_DEFINE_STATIC we are unable to mark this data as static
285 * in modules on some architectures.
286 */
287#define VNET_DEFINE_STATIC(t, n) \
288 t VNET_NAME(n) __section(VNET_SETNAME) __used
289#else
290#define VNET_DEFINE_STATIC(t, n) \
291 static t VNET_NAME(n) __section(VNET_SETNAME) __used
292#endif
293#define _VNET_PTR(b, n) (__typeof(VNET_NAME(n))*) \
294 ((b) + (uintptr_t)&VNET_NAME(n))
295
296#define _VNET(b, n) (*_VNET_PTR(b, n))
297
298/*
299 * Virtualized global variable accessor macros.
300 */
301#define VNET_VNET_PTR(vnet, n) _VNET_PTR((vnet)->vnet_data_base, n)
302#define VNET_VNET(vnet, n) (*VNET_VNET_PTR((vnet), n))
303
304#define VNET_PTR(n) VNET_VNET_PTR(curvnet, n)
305#define VNET(n) VNET_VNET(curvnet, n)
306
307/*
308 * Virtual network stack allocator interfaces from the kernel linker.
309 */
310void *vnet_data_alloc(int size);
311void vnet_data_copy(void *start, int size);
312void vnet_data_free(void *start_arg, int size);
313
314/*
315 * Virtual sysinit mechanism, allowing network stack components to declare
316 * startup and shutdown methods to be run when virtual network stack
317 * instances are created and destroyed.
318 */
319#include <sys/kernel.h>
320
321/*
322 * SYSINIT/SYSUNINIT variants that provide per-vnet constructors and
323 * destructors.
324 */
325struct vnet_sysinit {
326 enum sysinit_sub_id subsystem;
327 enum sysinit_elem_order order;
328 sysinit_cfunc_t func;
329 const void *arg;
330 TAILQ_ENTRY(vnet_sysinit) link;
331};
332
333#define VNET_SYSINIT(ident, subsystem, order, func, arg) \
334 CTASSERT((subsystem) > SI_SUB_VNET && \
335 (subsystem) <= SI_SUB_VNET_DONE); \
336 static struct vnet_sysinit ident ## _vnet_init = { \
337 subsystem, \
338 order, \
339 (sysinit_cfunc_t)(sysinit_nfunc_t)func, \
340 (arg) \
341 }; \
342 SYSINIT(vnet_init_ ## ident, subsystem, order, \
343 vnet_register_sysinit, &ident ## _vnet_init); \
344 SYSUNINIT(vnet_init_ ## ident, subsystem, order, \
345 vnet_deregister_sysinit, &ident ## _vnet_init)
346
347#define VNET_SYSUNINIT(ident, subsystem, order, func, arg) \
348 CTASSERT((subsystem) > SI_SUB_VNET && \
349 (subsystem) <= SI_SUB_VNET_DONE); \
350 static struct vnet_sysinit ident ## _vnet_uninit = { \
351 subsystem, \
352 order, \
353 (sysinit_cfunc_t)(sysinit_nfunc_t)func, \
354 (arg) \
355 }; \
356 SYSINIT(vnet_uninit_ ## ident, subsystem, order, \
357 vnet_register_sysuninit, &ident ## _vnet_uninit); \
358 SYSUNINIT(vnet_uninit_ ## ident, subsystem, order, \
359 vnet_deregister_sysuninit, &ident ## _vnet_uninit)
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#define CURVNET_ASSERT_SET() \
404
405#define VNET_LIST_RLOCK()
406#define VNET_LIST_RLOCK_NOSLEEP()
407#define VNET_LIST_RUNLOCK()
408#define VNET_LIST_RUNLOCK_NOSLEEP()
409#define VNET_ITERATOR_DECL(arg)
410#define VNET_FOREACH(arg) for (int _vn = 0; _vn == 0; _vn++)
411
412#define IS_DEFAULT_VNET(arg) 1
413#define CRED_TO_VNET(cr) NULL
414#define TD_TO_VNET(td) NULL
415#define P_TO_VNET(p) NULL
416
417/*
418 * Versions of the VNET macros that compile to normal global variables and
419 * standard sysctl definitions.
420 */
421#define VNET_NAME(n) n
422#define VNET_DECLARE(t, n) extern t n
423#define VNET_DEFINE(t, n) struct _hack; t n
424#define VNET_DEFINE_STATIC(t, n) static t n
425#define _VNET_PTR(b, n) &VNET_NAME(n)
426
427/*
428 * Virtualized global variable accessor macros.
429 */
430#define VNET_VNET_PTR(vnet, n) (&(n))
431#define VNET_VNET(vnet, n) (n)
432
433#define VNET_PTR(n) (&(n))
434#define VNET(n) (n)
435
436/*
437 * When VIMAGE isn't compiled into the kernel, VNET_SYSINIT/VNET_SYSUNINIT
438 * map into normal sysinits, which have the same ordering properties.
439 */
440#define VNET_SYSINIT(ident, subsystem, order, func, arg) \
441 SYSINIT(ident, subsystem, order, func, arg)
442#define VNET_SYSUNINIT(ident, subsystem, order, func, arg) \
443 SYSUNINIT(ident, subsystem, order, func, arg)
444
445/*
446 * Without VIMAGE revert to the default implementation.
447 */
448#define VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG(tag, name, func, arg, priority) \
449 (tag) = eventhandler_register(NULL, #name, func, arg, priority)
450#define VNET_GLOBAL_EVENTHANDLER_REGISTER(name, func, arg, priority) \
451 eventhandler_register(NULL, #name, func, arg, priority)
452#endif /* VIMAGE */
453#endif /* _KERNEL */
454
455#endif /* !_NET_VNET_H_ */