xref: /freebsd/sys/kern/kern_sharedpage.c (revision 141b44e0afc2912a0a1c7bbb4116921ead1d58d4)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2010, 2012 Konstantin Belousov <kib@FreeBSD.org>
5  * Copyright (c) 2015 The FreeBSD Foundation
6  * All rights reserved.
7  *
8  * Portions of this software were developed by Konstantin Belousov
9  * under sponsorship from the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/rwlock.h>
42 #include <sys/stddef.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 #ifdef RANDOM_FENESTRASX
63 static struct vdso_fxrng_generation *fxrng_shpage_mapping;
64 
65 static bool fxrng_enabled = true;
66 SYSCTL_BOOL(_debug, OID_AUTO, fxrng_vdso_enable, CTLFLAG_RWTUN, &fxrng_enabled,
67     0, "Enable FXRNG VDSO");
68 #endif
69 
70 void
shared_page_write(int base,int size,const void * data)71 shared_page_write(int base, int size, const void *data)
72 {
73 
74 	bcopy(data, shared_page_mapping + base, size);
75 }
76 
77 static int
shared_page_alloc_locked(int size,int align)78 shared_page_alloc_locked(int size, int align)
79 {
80 	int res;
81 
82 	res = roundup(shared_page_free, align);
83 	if (res + size >= IDX_TO_OFF(shared_page_obj->size))
84 		res = -1;
85 	else
86 		shared_page_free = res + size;
87 	return (res);
88 }
89 
90 int
shared_page_alloc(int size,int align)91 shared_page_alloc(int size, int align)
92 {
93 	int res;
94 
95 	sx_xlock(&shared_page_alloc_sx);
96 	res = shared_page_alloc_locked(size, align);
97 	sx_xunlock(&shared_page_alloc_sx);
98 	return (res);
99 }
100 
101 int
shared_page_fill(int size,int align,const void * data)102 shared_page_fill(int size, int align, const void *data)
103 {
104 	int res;
105 
106 	sx_xlock(&shared_page_alloc_sx);
107 	res = shared_page_alloc_locked(size, align);
108 	if (res != -1)
109 		shared_page_write(res, size, data);
110 	sx_xunlock(&shared_page_alloc_sx);
111 	return (res);
112 }
113 
114 static void
shared_page_init(void * dummy __unused)115 shared_page_init(void *dummy __unused)
116 {
117 	vm_page_t m;
118 	vm_offset_t addr;
119 
120 	sx_init(&shared_page_alloc_sx, "shpsx");
121 	shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
122 	    VM_PROT_DEFAULT, 0, NULL);
123 	VM_OBJECT_WLOCK(shared_page_obj);
124 	m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_ZERO);
125 	VM_OBJECT_WUNLOCK(shared_page_obj);
126 	vm_page_valid(m);
127 	vm_page_xunbusy(m);
128 	addr = kva_alloc(PAGE_SIZE);
129 	pmap_qenter(addr, &m, 1);
130 	shared_page_mapping = (char *)addr;
131 }
132 
133 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, shared_page_init, NULL);
134 
135 /*
136  * Push the timehands update to the shared page.
137  *
138  * The lockless update scheme is similar to the one used to update the
139  * in-kernel timehands, see sys/kern/kern_tc.c:tc_windup() (which
140  * calls us after the timehands are updated).
141  */
142 static void
timehands_update(struct vdso_sv_tk * svtk)143 timehands_update(struct vdso_sv_tk *svtk)
144 {
145 	struct vdso_timehands th;
146 	struct vdso_timekeep *tk;
147 	uint32_t enabled, idx;
148 
149 	enabled = tc_fill_vdso_timehands(&th);
150 	th.th_gen = 0;
151 	idx = svtk->sv_timekeep_curr;
152 	if (++idx >= VDSO_TH_NUM)
153 		idx = 0;
154 	svtk->sv_timekeep_curr = idx;
155 	if (++svtk->sv_timekeep_gen == 0)
156 		svtk->sv_timekeep_gen = 1;
157 
158 	tk = (struct vdso_timekeep *)(shared_page_mapping +
159 	    svtk->sv_timekeep_off);
160 	tk->tk_th[idx].th_gen = 0;
161 	atomic_thread_fence_rel();
162 	if (enabled)
163 		tk->tk_th[idx] = th;
164 	atomic_store_rel_32(&tk->tk_th[idx].th_gen, svtk->sv_timekeep_gen);
165 	atomic_store_rel_32(&tk->tk_current, idx);
166 
167 	/*
168 	 * The ordering of the assignment to tk_enabled relative to
169 	 * the update of the vdso_timehands is not important.
170 	 */
171 	tk->tk_enabled = enabled;
172 }
173 
174 #ifdef COMPAT_FREEBSD32
175 static void
timehands_update32(struct vdso_sv_tk * svtk)176 timehands_update32(struct vdso_sv_tk *svtk)
177 {
178 	struct vdso_timehands32 th;
179 	struct vdso_timekeep32 *tk;
180 	uint32_t enabled, idx;
181 
182 	enabled = tc_fill_vdso_timehands32(&th);
183 	th.th_gen = 0;
184 	idx = svtk->sv_timekeep_curr;
185 	if (++idx >= VDSO_TH_NUM)
186 		idx = 0;
187 	svtk->sv_timekeep_curr = idx;
188 	if (++svtk->sv_timekeep_gen == 0)
189 		svtk->sv_timekeep_gen = 1;
190 
191 	tk = (struct vdso_timekeep32 *)(shared_page_mapping +
192 	    svtk->sv_timekeep_off);
193 	tk->tk_th[idx].th_gen = 0;
194 	atomic_thread_fence_rel();
195 	if (enabled)
196 		tk->tk_th[idx] = th;
197 	atomic_store_rel_32(&tk->tk_th[idx].th_gen, svtk->sv_timekeep_gen);
198 	atomic_store_rel_32(&tk->tk_current, idx);
199 	tk->tk_enabled = enabled;
200 }
201 #endif
202 
203 /*
204  * This is hackish, but easiest way to avoid creating list structures
205  * that needs to be iterated over from the hardclock interrupt
206  * context.
207  */
208 static struct vdso_sv_tk *host_svtk;
209 #ifdef COMPAT_FREEBSD32
210 static struct vdso_sv_tk *compat32_svtk;
211 #endif
212 
213 void
timekeep_push_vdso(void)214 timekeep_push_vdso(void)
215 {
216 
217 	if (host_svtk != NULL)
218 		timehands_update(host_svtk);
219 #ifdef COMPAT_FREEBSD32
220 	if (compat32_svtk != NULL)
221 		timehands_update32(compat32_svtk);
222 #endif
223 }
224 
225 struct vdso_sv_tk *
alloc_sv_tk(void)226 alloc_sv_tk(void)
227 {
228 	struct vdso_sv_tk *svtk;
229 	int tk_base;
230 	uint32_t tk_ver;
231 
232 	tk_ver = VDSO_TK_VER_CURR;
233 	svtk = malloc(sizeof(struct vdso_sv_tk), M_TEMP, M_WAITOK | M_ZERO);
234 	tk_base = shared_page_alloc(sizeof(struct vdso_timekeep) +
235 	    sizeof(struct vdso_timehands) * VDSO_TH_NUM, 16);
236 	KASSERT(tk_base != -1, ("tk_base -1 for native"));
237 	shared_page_write(tk_base + offsetof(struct vdso_timekeep, tk_ver),
238 	    sizeof(uint32_t), &tk_ver);
239 	svtk->sv_timekeep_off = tk_base;
240 	timekeep_push_vdso();
241 	return (svtk);
242 }
243 
244 #ifdef COMPAT_FREEBSD32
245 struct vdso_sv_tk *
alloc_sv_tk_compat32(void)246 alloc_sv_tk_compat32(void)
247 {
248 	struct vdso_sv_tk *svtk;
249 	int tk_base;
250 	uint32_t tk_ver;
251 
252 	svtk = malloc(sizeof(struct vdso_sv_tk), M_TEMP, M_WAITOK | M_ZERO);
253 	tk_ver = VDSO_TK_VER_CURR;
254 	tk_base = shared_page_alloc(sizeof(struct vdso_timekeep32) +
255 	    sizeof(struct vdso_timehands32) * VDSO_TH_NUM, 16);
256 	KASSERT(tk_base != -1, ("tk_base -1 for 32bit"));
257 	shared_page_write(tk_base + offsetof(struct vdso_timekeep32,
258 	    tk_ver), sizeof(uint32_t), &tk_ver);
259 	svtk->sv_timekeep_off = tk_base;
260 	timekeep_push_vdso();
261 	return (svtk);
262 }
263 #endif
264 
265 #ifdef RANDOM_FENESTRASX
266 void
fxrng_push_seed_generation(uint64_t gen)267 fxrng_push_seed_generation(uint64_t gen)
268 {
269 	if (fxrng_shpage_mapping == NULL || !fxrng_enabled)
270 		return;
271 	KASSERT(gen < INT32_MAX,
272 	    ("fxrng seed version shouldn't roll over a 32-bit counter "
273 	     "for approximately 456,000 years"));
274 	atomic_store_rel_32(&fxrng_shpage_mapping->fx_generation32,
275 	    (uint32_t)gen);
276 }
277 
278 static void
alloc_sv_fxrng_generation(void)279 alloc_sv_fxrng_generation(void)
280 {
281 	int base;
282 
283 	/*
284 	 * Allocate a full cache line for the fxrng root generation (64-bit
285 	 * counter, or truncated 32-bit counter on ILP32 userspace).  It is
286 	 * important that the line is not shared with frequently dirtied data,
287 	 * and the shared page allocator lacks a __read_mostly mechanism.
288 	 * However, PAGE_SIZE is typically large relative to the amount of
289 	 * stuff we've got in it so far, so maybe the possible waste isn't an
290 	 * issue.
291 	 */
292 	base = shared_page_alloc(CACHE_LINE_SIZE, CACHE_LINE_SIZE);
293 	KASSERT(base != -1, ("%s: base allocation failed", __func__));
294 	fxrng_shpage_mapping = (void *)(shared_page_mapping + base);
295 	*fxrng_shpage_mapping = (struct vdso_fxrng_generation) {
296 		.fx_vdso_version = VDSO_FXRNG_VER_CURR,
297 	};
298 }
299 #endif /* RANDOM_FENESTRASX */
300 
301 void
exec_sysvec_init(void * param)302 exec_sysvec_init(void *param)
303 {
304 	struct sysentvec *sv;
305 	u_int flags;
306 	int res;
307 
308 	sv = param;
309 	flags = sv->sv_flags;
310 	if ((flags & SV_SHP) == 0)
311 		return;
312 	MPASS(sv->sv_shared_page_obj == NULL);
313 	MPASS(sv->sv_shared_page_base != 0);
314 
315 	sv->sv_shared_page_obj = shared_page_obj;
316 	if ((flags & SV_ABI_MASK) == SV_ABI_FREEBSD) {
317 		if ((flags & SV_DSO_SIG) != 0) {
318 			res = shared_page_fill((uintptr_t)sv->sv_szsigcode,
319 			    16, sv->sv_sigcode);
320 			if (res == -1)
321 				panic("copying vdso to shared page");
322 			sv->sv_vdso_offset = res;
323 			sv->sv_sigcode_offset = res + sv->sv_sigcodeoff;
324 		} else {
325 			res = shared_page_fill(*(sv->sv_szsigcode),
326 			    16, sv->sv_sigcode);
327 			if (res == -1)
328 				panic("copying sigtramp to shared page");
329 			sv->sv_sigcode_offset = res;
330 		}
331 	}
332 	if ((flags & SV_TIMEKEEP) != 0) {
333 #ifdef COMPAT_FREEBSD32
334 		if ((flags & SV_ILP32) != 0) {
335 			if ((flags & SV_ABI_MASK) == SV_ABI_FREEBSD) {
336 				KASSERT(compat32_svtk == NULL,
337 				    ("Compat32 already registered"));
338 				compat32_svtk = alloc_sv_tk_compat32();
339 			} else {
340 				KASSERT(compat32_svtk != NULL,
341 				    ("Compat32 not registered"));
342 			}
343 			sv->sv_timekeep_offset = compat32_svtk->sv_timekeep_off;
344 		} else {
345 #endif
346 			if ((flags & SV_ABI_MASK) == SV_ABI_FREEBSD) {
347 				KASSERT(host_svtk == NULL,
348 				    ("Host already registered"));
349 				host_svtk = alloc_sv_tk();
350 			} else {
351 				KASSERT(host_svtk != NULL,
352 				    ("Host not registered"));
353 			}
354 			sv->sv_timekeep_offset = host_svtk->sv_timekeep_off;
355 #ifdef COMPAT_FREEBSD32
356 		}
357 #endif
358 	}
359 #ifdef RANDOM_FENESTRASX
360 	if ((flags & (SV_ABI_MASK | SV_RNG_SEED_VER)) ==
361 	    (SV_ABI_FREEBSD | SV_RNG_SEED_VER)) {
362 		/*
363 		 * Only allocate a single VDSO entry for multiple sysentvecs,
364 		 * i.e., native and COMPAT32.
365 		 */
366 		if (fxrng_shpage_mapping == NULL)
367 			alloc_sv_fxrng_generation();
368 		sv->sv_fxrng_gen_offset =
369 		    (char *)fxrng_shpage_mapping - shared_page_mapping;
370 	}
371 #endif
372 }
373 
374 void
exec_sysvec_init_secondary(struct sysentvec * sv,struct sysentvec * sv2)375 exec_sysvec_init_secondary(struct sysentvec *sv, struct sysentvec *sv2)
376 {
377 	MPASS((sv2->sv_flags & SV_ABI_MASK) == (sv->sv_flags & SV_ABI_MASK));
378 	MPASS((sv2->sv_flags & SV_TIMEKEEP) == (sv->sv_flags & SV_TIMEKEEP));
379 	MPASS((sv2->sv_flags & SV_SHP) != 0 && (sv->sv_flags & SV_SHP) != 0);
380 	MPASS((sv2->sv_flags & SV_DSO_SIG) == (sv->sv_flags & SV_DSO_SIG));
381 	MPASS((sv2->sv_flags & SV_RNG_SEED_VER) ==
382 	    (sv->sv_flags & SV_RNG_SEED_VER));
383 
384 	sv2->sv_shared_page_obj = sv->sv_shared_page_obj;
385 	sv2->sv_sigcode_offset = sv->sv_sigcode_offset;
386 	sv2->sv_vdso_offset = sv->sv_vdso_offset;
387 	if ((sv2->sv_flags & SV_ABI_MASK) != SV_ABI_FREEBSD)
388 		return;
389 	sv2->sv_timekeep_offset = sv->sv_timekeep_offset;
390 	sv2->sv_fxrng_gen_offset = sv->sv_fxrng_gen_offset;
391 }
392