2 * Copyright (c) 2004-2009 University of Zagreb
3 * Copyright (c) 2006-2009 FreeBSD Foundation
6 * This software was developed by the University of Zagreb and the
7 * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
10 * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
11 * Copyright (c) 2009 Robert N. M. Watson
12 * All rights reserved.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
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.
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
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/kernel.h>
44 #include <sys/systm.h>
45 #include <sys/sysctl.h>
46 #include <sys/linker_set.h>
48 #include <sys/malloc.h>
50 #include <sys/socket.h>
52 #include <sys/sysctl.h>
59 #include <net/if_var.h>
63 * This file implements core functions for virtual network stacks:
65 * - Virtual network stack management functions.
67 * - Virtual network stack memory allocator, which virtualizes global
68 * variables in the network stack
70 * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
71 * to register startup/shutdown events to be run for each virtual network
75 MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
78 * The virtual network stack list has two read-write locks, one sleepable and
79 * the other not, so that the list can be stablized and walked in a variety
80 * of network stack contexts. Both must be acquired exclusively to modify
81 * the list, but a read lock of either lock is sufficient to walk the list.
83 struct rwlock vnet_rwlock;
84 struct sx vnet_sxlock;
86 #define VNET_LIST_WLOCK() do { \
87 sx_xlock(&vnet_sxlock); \
88 rw_wlock(&vnet_rwlock); \
91 #define VNET_LIST_WUNLOCK() do { \
92 rw_wunlock(&vnet_rwlock); \
93 sx_xunlock(&vnet_sxlock); \
96 struct vnet_list_head vnet_head;
100 * The virtual network stack allocator provides storage for virtualized
101 * global variables. These variables are defined/declared using the
102 * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet'
103 * linker set. The details of the implementation are somewhat subtle, but
104 * allow the majority of most network subsystems to maintain
105 * virtualization-agnostic.
107 * The virtual network stack allocator handles variables in the base kernel
108 * vs. modules in similar but different ways. In both cases, virtualized
109 * global variables are marked as such by being declared to be part of the
110 * vnet linker set. These "master" copies of global variables serve two
113 * (1) They contain static initialization or "default" values for global
114 * variables which will be propagated to each virtual network stack
115 * instance when created. As with normal global variables, they default
118 * (2) They act as unique global names by which the variable can be referred
119 * to, regardless of network stack instance. The single global symbol
120 * will be used to calculate the location of a per-virtual instance
121 * variable at run-time.
123 * Each virtual network stack instance has a complete copy of each
124 * virtualized global variable, stored in a malloc'd block of memory
125 * referred to by vnet->vnet_data_mem. Critical to the design is that each
126 * per-instance memory block is laid out identically to the master block so
127 * that the offset of each global variable is the same across all blocks. To
128 * optimize run-time access, a precalculated 'base' address,
129 * vnet->vnet_data_base, is stored in each vnet, and is the amount that can
130 * be added to the address of a 'master' instance of a variable to get to the
133 * Virtualized global variables are handled in a similar manner, but as each
134 * module has its own 'set_vnet' linker set, and we want to keep all
135 * virtualized globals togther, we reserve space in the kernel's linker set
136 * for potential module variables using a per-vnet character array,
137 * 'modspace'. The virtual network stack allocator maintains a free list to
138 * track what space in the array is free (all, initially) and as modules are
139 * linked, allocates portions of the space to specific globals. The kernel
140 * module linker queries the virtual network stack allocator and will
141 * bind references of the global to the location during linking. It also
142 * calls into the virtual network stack allocator, once the memory is
143 * initialized, in order to propagate the new static initializations to all
144 * existing virtual network stack instances so that the soon-to-be executing
145 * module will find every network stack instance with proper default values.
149 * Location of the kernel's 'set_vnet' linker set.
151 extern uintptr_t *__start_set_vnet;
152 extern uintptr_t *__stop_set_vnet;
154 #define VNET_START (uintptr_t)&__start_set_vnet
155 #define VNET_STOP (uintptr_t)&__stop_set_vnet
158 * Number of bytes of data in the 'set_vnet' linker set, and hence the total
159 * size of all kernel virtualized global variables, and the malloc(9) type
160 * that will be used to allocate it.
162 #define VNET_BYTES (VNET_STOP - VNET_START)
164 MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
167 * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of
168 * global variables across all loaded modules. As this actually sizes an
169 * array declared as a virtualized global variable in the kernel itself, and
170 * we want the virtualized global variable space to be page-sized, we may
171 * have more space than that in practice.
173 #define VNET_MODMIN 8192
174 #define VNET_SIZE roundup2(VNET_BYTES, PAGE_SIZE)
175 #define VNET_MODSIZE (VNET_SIZE - (VNET_BYTES - VNET_MODMIN))
178 * Space to store virtualized global variables from loadable kernel modules,
179 * and the free list to manage it.
181 static VNET_DEFINE(char, modspace[VNET_MODMIN]);
184 * Global lists of subsystem constructor and destructors for vnets. They are
185 * registered via VNET_SYSINIT() and VNET_SYSUNINIT(). The lists are
186 * protected by the vnet_sxlock global lock.
188 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
189 TAILQ_HEAD_INITIALIZER(vnet_constructors);
190 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
191 TAILQ_HEAD_INITIALIZER(vnet_destructors);
193 struct vnet_data_free {
196 TAILQ_ENTRY(vnet_data_free) vnd_link;
199 MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free", "VNET resource accounting");
200 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
201 TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
202 static struct sx vnet_data_free_lock;
205 * Allocate a virtual network stack.
212 vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
213 vnet->vnet_magic_n = VNET_MAGIC_N;
216 * Allocate storage for virtualized global variables and copy in
217 * initial values form our 'master' copy.
219 vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
220 memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
223 * All use of vnet-specific data will immediately subtract VNET_START
224 * from the base memory pointer, so pre-calculate that now to avoid
227 vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
229 /* Initialize / attach vnet module instances. */
230 CURVNET_SET_QUIET(vnet);
232 sx_xlock(&vnet_sxlock);
236 rw_wlock(&vnet_rwlock);
237 LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
244 * Destroy a virtual network stack.
247 vnet_destroy(struct vnet *vnet)
249 struct ifnet *ifp, *nifp;
251 KASSERT(vnet->vnet_sockcnt == 0,
252 ("%s: vnet still has sockets", __func__));
255 LIST_REMOVE(vnet, vnet_le);
256 rw_wunlock(&vnet_rwlock);
258 CURVNET_SET_QUIET(vnet);
260 /* Return all inherited interfaces to their parent vnets. */
261 TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
262 if (ifp->if_home_vnet != ifp->if_vnet)
263 if_vmove(ifp, ifp->if_home_vnet);
267 sx_xunlock(&vnet_sxlock);
272 * Release storage for the virtual network stack instance.
274 free(vnet->vnet_data_mem, M_VNET_DATA);
275 vnet->vnet_data_mem = NULL;
276 vnet->vnet_data_base = 0;
277 vnet->vnet_magic_n = 0xdeadbeef;
282 * Boot time initialization and allocation of virtual network stacks.
285 vnet_init_prelink(void *arg)
288 rw_init(&vnet_rwlock, "vnet_rwlock");
289 sx_init(&vnet_sxlock, "vnet_sxlock");
290 LIST_INIT(&vnet_head);
292 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
293 vnet_init_prelink, NULL);
296 vnet0_init(void *arg)
299 /* Warn people before take off - in case we crash early. */
300 printf("WARNING: VIMAGE (virtualized network stack) is a highly "
301 "experimental feature.\n");
304 * We MUST clear curvnet in vi_init_done() before going SMP,
305 * otherwise CURVNET_SET() macros would scream about unnecessary
306 * curvnet recursions.
308 curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
310 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
313 vnet_init_done(void *unused)
319 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, vnet_init_done,
323 * Once on boot, initialize the modspace freelist to entirely cover modspace.
326 vnet_data_startup(void *dummy __unused)
328 struct vnet_data_free *df;
330 df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
331 df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
332 df->vnd_len = VNET_MODSIZE;
333 TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
334 sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
336 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, 0);
339 * When a module is loaded and requires storage for a virtualized global
340 * variable, allocate space from the modspace free list. This interface
341 * should be used only by the kernel linker.
344 vnet_data_alloc(int size)
346 struct vnet_data_free *df;
350 size = roundup2(size, sizeof(void *));
351 sx_xlock(&vnet_data_free_lock);
352 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
353 if (df->vnd_len < size)
355 if (df->vnd_len == size) {
356 s = (void *)df->vnd_start;
357 TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
358 free(df, M_VNET_DATA_FREE);
361 s = (void *)df->vnd_start;
363 df->vnd_start = df->vnd_start + size;
366 sx_xunlock(&vnet_data_free_lock);
372 * Free space for a virtualized global variable on module unload.
375 vnet_data_free(void *start_arg, int size)
377 struct vnet_data_free *df;
378 struct vnet_data_free *dn;
382 size = roundup2(size, sizeof(void *));
383 start = (uintptr_t)start_arg;
386 * Free a region of space and merge it with as many neighbors as
387 * possible. Keeping the list sorted simplifies this operation.
389 sx_xlock(&vnet_data_free_lock);
390 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
391 if (df->vnd_start > end)
394 * If we expand at the end of an entry we may have to merge
395 * it with the one following it as well.
397 if (df->vnd_start + df->vnd_len == start) {
399 dn = TAILQ_NEXT(df, vnd_link);
400 if (df->vnd_start + df->vnd_len == dn->vnd_start) {
401 df->vnd_len += dn->vnd_len;
402 TAILQ_REMOVE(&vnet_data_free_head, dn,
404 free(dn, M_VNET_DATA_FREE);
406 sx_xunlock(&vnet_data_free_lock);
409 if (df->vnd_start == end) {
410 df->vnd_start = start;
412 sx_xunlock(&vnet_data_free_lock);
416 dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
417 dn->vnd_start = start;
420 TAILQ_INSERT_BEFORE(df, dn, vnd_link);
422 TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
423 sx_xunlock(&vnet_data_free_lock);
427 * When a new virtualized global variable has been allocated, propagate its
428 * initial value to each already-allocated virtual network stack instance.
431 vnet_data_copy(void *start, int size)
436 LIST_FOREACH(vnet, &vnet_head, vnet_le)
437 memcpy((void *)((uintptr_t)vnet->vnet_data_base +
438 (uintptr_t)start), start, size);
443 * Variants on sysctl_handle_foo that know how to handle virtualized global
444 * variables: if 'arg1' is a pointer, then we transform it to the local vnet
448 vnet_sysctl_handle_int(SYSCTL_HANDLER_ARGS)
452 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
453 return (sysctl_handle_int(oidp, arg1, arg2, req));
457 vnet_sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
461 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
462 return (sysctl_handle_opaque(oidp, arg1, arg2, req));
466 vnet_sysctl_handle_string(SYSCTL_HANDLER_ARGS)
470 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
471 return (sysctl_handle_string(oidp, arg1, arg2, req));
475 vnet_sysctl_handle_uint(SYSCTL_HANDLER_ARGS)
479 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
480 return (sysctl_handle_int(oidp, arg1, arg2, req));
484 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
485 * and VNET_SYSUNINIT().
488 vnet_register_sysinit(void *arg)
490 struct vnet_sysinit *vs, *vs2;
494 KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
496 /* Add the constructor to the global list of vnet constructors. */
497 sx_xlock(&vnet_sxlock);
498 TAILQ_FOREACH(vs2, &vnet_constructors, link) {
499 if (vs2->subsystem > vs->subsystem)
501 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
505 TAILQ_INSERT_BEFORE(vs2, vs, link);
507 TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
510 * Invoke the constructor on all the existing vnets when it is
514 CURVNET_SET_QUIET(vnet);
518 sx_xunlock(&vnet_sxlock);
522 vnet_deregister_sysinit(void *arg)
524 struct vnet_sysinit *vs;
528 /* Remove the constructor from the global list of vnet constructors. */
529 sx_xlock(&vnet_sxlock);
530 TAILQ_REMOVE(&vnet_constructors, vs, link);
531 sx_xunlock(&vnet_sxlock);
535 vnet_register_sysuninit(void *arg)
537 struct vnet_sysinit *vs, *vs2;
541 /* Add the destructor to the global list of vnet destructors. */
542 sx_xlock(&vnet_sxlock);
543 TAILQ_FOREACH(vs2, &vnet_destructors, link) {
544 if (vs2->subsystem > vs->subsystem)
546 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
550 TAILQ_INSERT_BEFORE(vs2, vs, link);
552 TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
553 sx_xunlock(&vnet_sxlock);
557 vnet_deregister_sysuninit(void *arg)
559 struct vnet_sysinit *vs;
565 * Invoke the destructor on all the existing vnets when it is
568 sx_xlock(&vnet_sxlock);
570 CURVNET_SET_QUIET(vnet);
575 /* Remove the destructor from the global list of vnet destructors. */
576 TAILQ_REMOVE(&vnet_destructors, vs, link);
577 sx_xunlock(&vnet_sxlock);
581 * Invoke all registered vnet constructors on the current vnet. Used during
582 * vnet construction. The caller is responsible for ensuring the new vnet is
583 * the current vnet and that the vnet_sxlock lock is locked.
588 struct vnet_sysinit *vs;
590 sx_assert(&vnet_sxlock, SA_LOCKED);
591 TAILQ_FOREACH(vs, &vnet_constructors, link) {
597 * Invoke all registered vnet destructors on the current vnet. Used during
598 * vnet destruction. The caller is responsible for ensuring the dying vnet
599 * is the current vnet and that the vnet_sxlock lock is locked.
604 struct vnet_sysinit *vs;
606 sx_assert(&vnet_sxlock, SA_LOCKED);
607 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
614 DB_SHOW_COMMAND(vnets, db_show_vnets)
616 VNET_ITERATOR_DECL(vnet_iter);
618 #if SIZE_MAX == UINT32_MAX /* 32-bit arch */
619 db_printf(" vnet ifs socks\n");
620 #else /* 64-bit arch, most probaly... */
621 db_printf(" vnet ifs socks\n");
623 VNET_FOREACH(vnet_iter) {
624 db_printf("%p %3d %5d\n", vnet_iter, vnet_iter->vnet_ifcnt,
625 vnet_iter->vnet_sockcnt);