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(). Both lists are
186 * protected by the vnet_sysinit_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 sx vnet_sysinit_sxlock;
195 #define VNET_SYSINIT_WLOCK() sx_xlock(&vnet_sysinit_sxlock);
196 #define VNET_SYSINIT_WUNLOCK() sx_xunlock(&vnet_sysinit_sxlock);
197 #define VNET_SYSINIT_RLOCK() sx_slock(&vnet_sysinit_sxlock);
198 #define VNET_SYSINIT_RUNLOCK() sx_sunlock(&vnet_sysinit_sxlock);
200 struct vnet_data_free {
203 TAILQ_ENTRY(vnet_data_free) vnd_link;
206 MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free", "VNET resource accounting");
207 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
208 TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
209 static struct sx vnet_data_free_lock;
212 * Allocate a virtual network stack.
219 vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
220 vnet->vnet_magic_n = VNET_MAGIC_N;
223 * Allocate storage for virtualized global variables and copy in
224 * initial values form our 'master' copy.
226 vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
227 memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
230 * All use of vnet-specific data will immediately subtract VNET_START
231 * from the base memory pointer, so pre-calculate that now to avoid
234 vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
236 /* Initialize / attach vnet module instances. */
237 CURVNET_SET_QUIET(vnet);
242 LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
249 * Destroy a virtual network stack.
252 vnet_destroy(struct vnet *vnet)
254 struct ifnet *ifp, *nifp;
256 KASSERT(vnet->vnet_sockcnt == 0,
257 ("%s: vnet still has sockets", __func__));
260 LIST_REMOVE(vnet, vnet_le);
263 CURVNET_SET_QUIET(vnet);
265 /* Return all inherited interfaces to their parent vnets. */
266 TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
267 if (ifp->if_home_vnet != ifp->if_vnet)
268 if_vmove(ifp, ifp->if_home_vnet);
275 * Release storage for the virtual network stack instance.
277 free(vnet->vnet_data_mem, M_VNET_DATA);
278 vnet->vnet_data_mem = NULL;
279 vnet->vnet_data_base = 0;
280 vnet->vnet_magic_n = 0xdeadbeef;
285 * Boot time initialization and allocation of virtual network stacks.
288 vnet_init_prelink(void *arg)
291 rw_init(&vnet_rwlock, "vnet_rwlock");
292 sx_init(&vnet_sxlock, "vnet_sxlock");
293 sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
294 LIST_INIT(&vnet_head);
296 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
297 vnet_init_prelink, NULL);
300 vnet0_init(void *arg)
303 /* Warn people before take off - in case we crash early. */
304 printf("WARNING: VIMAGE (virtualized network stack) is a highly "
305 "experimental feature.\n");
308 * We MUST clear curvnet in vi_init_done() before going SMP,
309 * otherwise CURVNET_SET() macros would scream about unnecessary
310 * curvnet recursions.
312 curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
314 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
317 vnet_init_done(void *unused)
323 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, vnet_init_done,
327 * Once on boot, initialize the modspace freelist to entirely cover modspace.
330 vnet_data_startup(void *dummy __unused)
332 struct vnet_data_free *df;
334 df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
335 df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
336 df->vnd_len = VNET_MODSIZE;
337 TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
338 sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
340 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, 0);
343 * When a module is loaded and requires storage for a virtualized global
344 * variable, allocate space from the modspace free list. This interface
345 * should be used only by the kernel linker.
348 vnet_data_alloc(int size)
350 struct vnet_data_free *df;
354 size = roundup2(size, sizeof(void *));
355 sx_xlock(&vnet_data_free_lock);
356 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
357 if (df->vnd_len < size)
359 if (df->vnd_len == size) {
360 s = (void *)df->vnd_start;
361 TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
362 free(df, M_VNET_DATA_FREE);
365 s = (void *)df->vnd_start;
367 df->vnd_start = df->vnd_start + size;
370 sx_xunlock(&vnet_data_free_lock);
376 * Free space for a virtualized global variable on module unload.
379 vnet_data_free(void *start_arg, int size)
381 struct vnet_data_free *df;
382 struct vnet_data_free *dn;
386 size = roundup2(size, sizeof(void *));
387 start = (uintptr_t)start_arg;
390 * Free a region of space and merge it with as many neighbors as
391 * possible. Keeping the list sorted simplifies this operation.
393 sx_xlock(&vnet_data_free_lock);
394 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
395 if (df->vnd_start > end)
398 * If we expand at the end of an entry we may have to merge
399 * it with the one following it as well.
401 if (df->vnd_start + df->vnd_len == start) {
403 dn = TAILQ_NEXT(df, vnd_link);
404 if (df->vnd_start + df->vnd_len == dn->vnd_start) {
405 df->vnd_len += dn->vnd_len;
406 TAILQ_REMOVE(&vnet_data_free_head, dn,
408 free(dn, M_VNET_DATA_FREE);
410 sx_xunlock(&vnet_data_free_lock);
413 if (df->vnd_start == end) {
414 df->vnd_start = start;
416 sx_xunlock(&vnet_data_free_lock);
420 dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
421 dn->vnd_start = start;
424 TAILQ_INSERT_BEFORE(df, dn, vnd_link);
426 TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
427 sx_xunlock(&vnet_data_free_lock);
431 * When a new virtualized global variable has been allocated, propagate its
432 * initial value to each already-allocated virtual network stack instance.
435 vnet_data_copy(void *start, int size)
440 LIST_FOREACH(vnet, &vnet_head, vnet_le)
441 memcpy((void *)((uintptr_t)vnet->vnet_data_base +
442 (uintptr_t)start), start, size);
447 * Variants on sysctl_handle_foo that know how to handle virtualized global
448 * variables: if 'arg1' is a pointer, then we transform it to the local vnet
452 vnet_sysctl_handle_int(SYSCTL_HANDLER_ARGS)
456 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
457 return (sysctl_handle_int(oidp, arg1, arg2, req));
461 vnet_sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
465 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
466 return (sysctl_handle_opaque(oidp, arg1, arg2, req));
470 vnet_sysctl_handle_string(SYSCTL_HANDLER_ARGS)
474 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
475 return (sysctl_handle_string(oidp, arg1, arg2, req));
479 vnet_sysctl_handle_uint(SYSCTL_HANDLER_ARGS)
483 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
484 return (sysctl_handle_int(oidp, arg1, arg2, req));
488 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
489 * and VNET_SYSUNINIT().
492 vnet_register_sysinit(void *arg)
494 struct vnet_sysinit *vs, *vs2;
498 KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
500 /* Add the constructor to the global list of vnet constructors. */
501 VNET_SYSINIT_WLOCK();
502 TAILQ_FOREACH(vs2, &vnet_constructors, link) {
503 if (vs2->subsystem > vs->subsystem)
505 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
509 TAILQ_INSERT_BEFORE(vs2, vs, link);
511 TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
514 * Invoke the constructor on all the existing vnets when it is
518 CURVNET_SET_QUIET(vnet);
522 VNET_SYSINIT_WUNLOCK();
526 vnet_deregister_sysinit(void *arg)
528 struct vnet_sysinit *vs;
532 /* Remove the constructor from the global list of vnet constructors. */
533 VNET_SYSINIT_WLOCK();
534 TAILQ_REMOVE(&vnet_constructors, vs, link);
535 VNET_SYSINIT_WUNLOCK();
539 vnet_register_sysuninit(void *arg)
541 struct vnet_sysinit *vs, *vs2;
545 /* Add the destructor to the global list of vnet destructors. */
546 VNET_SYSINIT_WLOCK();
547 TAILQ_FOREACH(vs2, &vnet_destructors, link) {
548 if (vs2->subsystem > vs->subsystem)
550 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
554 TAILQ_INSERT_BEFORE(vs2, vs, link);
556 TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
557 VNET_SYSINIT_WUNLOCK();
561 vnet_deregister_sysuninit(void *arg)
563 struct vnet_sysinit *vs;
569 * Invoke the destructor on all the existing vnets when it is
572 VNET_SYSINIT_WLOCK();
574 CURVNET_SET_QUIET(vnet);
579 /* Remove the destructor from the global list of vnet destructors. */
580 TAILQ_REMOVE(&vnet_destructors, vs, link);
581 VNET_SYSINIT_WUNLOCK();
585 * Invoke all registered vnet constructors on the current vnet. Used during
586 * vnet construction. The caller is responsible for ensuring the new vnet is
587 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
592 struct vnet_sysinit *vs;
594 VNET_SYSINIT_RLOCK();
595 TAILQ_FOREACH(vs, &vnet_constructors, link) {
598 VNET_SYSINIT_RUNLOCK();
602 * Invoke all registered vnet destructors on the current vnet. Used during
603 * vnet destruction. The caller is responsible for ensuring the dying vnet
604 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
609 struct vnet_sysinit *vs;
611 VNET_SYSINIT_RLOCK();
612 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
616 VNET_SYSINIT_RUNLOCK();
620 DB_SHOW_COMMAND(vnets, db_show_vnets)
622 VNET_ITERATOR_DECL(vnet_iter);
624 VNET_FOREACH(vnet_iter) {
625 db_printf("vnet = %p\n", vnet_iter);
626 db_printf(" vnet_magic_n = 0x%x (%s, orig 0x%x)\n",
627 vnet_iter->vnet_magic_n,
628 (vnet_iter->vnet_magic_n == VNET_MAGIC_N) ?
629 "ok" : "mismatch", VNET_MAGIC_N);
630 db_printf(" vnet_ifcnt = %u\n", vnet_iter->vnet_ifcnt);
631 db_printf(" vnet_sockcnt = %u\n", vnet_iter->vnet_sockcnt);
632 db_printf(" vnet_data_mem = %p\n", vnet_iter->vnet_data_mem);
633 db_printf(" vnet_data_base = 0x%jx\n",
634 (uintmax_t)vnet_iter->vnet_data_base);