]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/kern/kern_sharedpage.c
Fix bhyve SVM guest escape.
[FreeBSD/FreeBSD.git] / sys / kern / kern_sharedpage.c
1 /*-
2  * Copyright (c) 2010, 2012 Konstantin Belousov <kib@FreeBSD.org>
3  * Copyright (c) 2015 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Konstantin Belousov
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33
34 #include "opt_compat.h"
35 #include "opt_vm.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/rwlock.h>
43 #include <sys/sysent.h>
44 #include <sys/sysctl.h>
45 #include <sys/vdso.h>
46
47 #include <vm/vm.h>
48 #include <vm/vm_param.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_kern.h>
52 #include <vm/vm_map.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_pager.h>
56
57 static struct sx shared_page_alloc_sx;
58 static vm_object_t shared_page_obj;
59 static int shared_page_free;
60 char *shared_page_mapping;
61
62 void
63 shared_page_write(int base, int size, const void *data)
64 {
65
66         bcopy(data, shared_page_mapping + base, size);
67 }
68
69 static int
70 shared_page_alloc_locked(int size, int align)
71 {
72         int res;
73
74         res = roundup(shared_page_free, align);
75         if (res + size >= IDX_TO_OFF(shared_page_obj->size))
76                 res = -1;
77         else
78                 shared_page_free = res + size;
79         return (res);
80 }
81
82 int
83 shared_page_alloc(int size, int align)
84 {
85         int res;
86
87         sx_xlock(&shared_page_alloc_sx);
88         res = shared_page_alloc_locked(size, align);
89         sx_xunlock(&shared_page_alloc_sx);
90         return (res);
91 }
92
93 int
94 shared_page_fill(int size, int align, const void *data)
95 {
96         int res;
97
98         sx_xlock(&shared_page_alloc_sx);
99         res = shared_page_alloc_locked(size, align);
100         if (res != -1)
101                 shared_page_write(res, size, data);
102         sx_xunlock(&shared_page_alloc_sx);
103         return (res);
104 }
105
106 static void
107 shared_page_init(void *dummy __unused)
108 {
109         vm_page_t m;
110         vm_offset_t addr;
111
112         sx_init(&shared_page_alloc_sx, "shpsx");
113         shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
114             VM_PROT_DEFAULT, 0, NULL);
115         VM_OBJECT_WLOCK(shared_page_obj);
116         m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_NOBUSY | VM_ALLOC_ZERO);
117         m->valid = VM_PAGE_BITS_ALL;
118         VM_OBJECT_WUNLOCK(shared_page_obj);
119         addr = kva_alloc(PAGE_SIZE);
120         pmap_qenter(addr, &m, 1);
121         shared_page_mapping = (char *)addr;
122 }
123
124 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init,
125     NULL);
126
127 /*
128  * Push the timehands update to the shared page.
129  *
130  * The lockless update scheme is similar to the one used to update the
131  * in-kernel timehands, see sys/kern/kern_tc.c:tc_windup() (which
132  * calls us after the timehands are updated).
133  */
134 static void
135 timehands_update(struct vdso_sv_tk *svtk)
136 {
137         struct vdso_timehands th;
138         struct vdso_timekeep *tk;
139         uint32_t enabled, idx;
140
141         enabled = tc_fill_vdso_timehands(&th);
142         th.th_gen = 0;
143         idx = svtk->sv_timekeep_curr;
144         if (++idx >= VDSO_TH_NUM)
145                 idx = 0;
146         svtk->sv_timekeep_curr = idx;
147         if (++svtk->sv_timekeep_gen == 0)
148                 svtk->sv_timekeep_gen = 1;
149
150         tk = (struct vdso_timekeep *)(shared_page_mapping +
151             svtk->sv_timekeep_off);
152         tk->tk_th[idx].th_gen = 0;
153         atomic_thread_fence_rel();
154         if (enabled)
155                 tk->tk_th[idx] = th;
156         atomic_store_rel_32(&tk->tk_th[idx].th_gen, svtk->sv_timekeep_gen);
157         atomic_store_rel_32(&tk->tk_current, idx);
158
159         /*
160          * The ordering of the assignment to tk_enabled relative to
161          * the update of the vdso_timehands is not important.
162          */
163         tk->tk_enabled = enabled;
164 }
165
166 #ifdef COMPAT_FREEBSD32
167 static void
168 timehands_update32(struct vdso_sv_tk *svtk)
169 {
170         struct vdso_timehands32 th;
171         struct vdso_timekeep32 *tk;
172         uint32_t enabled, idx;
173
174         enabled = tc_fill_vdso_timehands32(&th);
175         th.th_gen = 0;
176         idx = svtk->sv_timekeep_curr;
177         if (++idx >= VDSO_TH_NUM)
178                 idx = 0;
179         svtk->sv_timekeep_curr = idx;
180         if (++svtk->sv_timekeep_gen == 0)
181                 svtk->sv_timekeep_gen = 1;
182
183         tk = (struct vdso_timekeep32 *)(shared_page_mapping +
184             svtk->sv_timekeep_off);
185         tk->tk_th[idx].th_gen = 0;
186         atomic_thread_fence_rel();
187         if (enabled)
188                 tk->tk_th[idx] = th;
189         atomic_store_rel_32(&tk->tk_th[idx].th_gen, svtk->sv_timekeep_gen);
190         atomic_store_rel_32(&tk->tk_current, idx);
191         tk->tk_enabled = enabled;
192 }
193 #endif
194
195 /*
196  * This is hackish, but easiest way to avoid creating list structures
197  * that needs to be iterated over from the hardclock interrupt
198  * context.
199  */
200 static struct vdso_sv_tk *host_svtk;
201 #ifdef COMPAT_FREEBSD32
202 static struct vdso_sv_tk *compat32_svtk;
203 #endif
204
205 void
206 timekeep_push_vdso(void)
207 {
208
209         if (host_svtk != NULL)
210                 timehands_update(host_svtk);
211 #ifdef COMPAT_FREEBSD32
212         if (compat32_svtk != NULL)
213                 timehands_update32(compat32_svtk);
214 #endif
215 }
216
217 struct vdso_sv_tk *
218 alloc_sv_tk(void)
219 {
220         struct vdso_sv_tk *svtk;
221         int tk_base;
222         uint32_t tk_ver;
223
224         tk_ver = VDSO_TK_VER_CURR;
225         svtk = malloc(sizeof(struct vdso_sv_tk), M_TEMP, M_WAITOK | M_ZERO);
226         tk_base = shared_page_alloc(sizeof(struct vdso_timekeep) +
227             sizeof(struct vdso_timehands) * VDSO_TH_NUM, 16);
228         KASSERT(tk_base != -1, ("tk_base -1 for native"));
229         shared_page_write(tk_base + offsetof(struct vdso_timekeep, tk_ver),
230             sizeof(uint32_t), &tk_ver);
231         svtk->sv_timekeep_off = tk_base;
232         timekeep_push_vdso();
233         return (svtk);
234 }
235
236 #ifdef COMPAT_FREEBSD32
237 struct vdso_sv_tk *
238 alloc_sv_tk_compat32(void)
239 {
240         struct vdso_sv_tk *svtk;
241         int tk_base;
242         uint32_t tk_ver;
243
244         svtk = malloc(sizeof(struct vdso_sv_tk), M_TEMP, M_WAITOK | M_ZERO);
245         tk_ver = VDSO_TK_VER_CURR;
246         tk_base = shared_page_alloc(sizeof(struct vdso_timekeep32) +
247             sizeof(struct vdso_timehands32) * VDSO_TH_NUM, 16);
248         KASSERT(tk_base != -1, ("tk_base -1 for 32bit"));
249         shared_page_write(tk_base + offsetof(struct vdso_timekeep32,
250             tk_ver), sizeof(uint32_t), &tk_ver);
251         svtk->sv_timekeep_off = tk_base;
252         timekeep_push_vdso();
253         return (svtk);
254 }
255 #endif
256
257 void
258 exec_sysvec_init(void *param)
259 {
260         struct sysentvec *sv;
261
262         sv = (struct sysentvec *)param;
263         if ((sv->sv_flags & SV_SHP) == 0)
264                 return;
265         sv->sv_shared_page_obj = shared_page_obj;
266         sv->sv_sigcode_base = sv->sv_shared_page_base +
267             shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode);
268         if ((sv->sv_flags & SV_ABI_MASK) != SV_ABI_FREEBSD)
269                 return;
270         if ((sv->sv_flags & SV_TIMEKEEP) != 0) {
271 #ifdef COMPAT_FREEBSD32
272                 if ((sv->sv_flags & SV_ILP32) != 0) {
273                         KASSERT(compat32_svtk == NULL,
274                             ("Compat32 already registered"));
275                         compat32_svtk = alloc_sv_tk_compat32();
276                         sv->sv_timekeep_base = sv->sv_shared_page_base +
277                             compat32_svtk->sv_timekeep_off;
278                 } else {
279 #endif
280                         KASSERT(host_svtk == NULL, ("Host already registered"));
281                         host_svtk = alloc_sv_tk();
282                         sv->sv_timekeep_base = sv->sv_shared_page_base +
283                             host_svtk->sv_timekeep_off;
284 #ifdef COMPAT_FREEBSD32
285                 }
286 #endif
287         }
288 }