1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 * Copyright (c) 2015 Joyent, Inc.
25 */
26
27 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
28 /* All Rights Reserved */
29
30
31 #include <sys/types.h>
32 #include <sys/bitmap.h>
33 #include <sys/sysmacros.h>
34 #include <sys/kmem.h>
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/user.h>
38 #include <sys/unistd.h>
39 #include <sys/errno.h>
40 #include <sys/proc.h>
41 #include <sys/mman.h>
42 #include <sys/tuneable.h>
43 #include <sys/cmn_err.h>
44 #include <sys/cred.h>
45 #include <sys/vmsystm.h>
46 #include <sys/debug.h>
47 #include <sys/policy.h>
48
49 #include <vm/as.h>
50 #include <vm/seg.h>
51
52 static uint_t mem_getpgszc(size_t);
53
54 /*
55 * Memory control operations
56 */
57 int
58 memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask)
59 {
60 struct as *as = ttoproc(curthread)->p_as;
61 struct proc *p = ttoproc(curthread);
62 size_t pgsz;
63 uint_t szc, oszc, pgcmd;
64 int error = 0;
65 faultcode_t fc;
66 uintptr_t iarg;
67 STRUCT_DECL(memcntl_mha, mha);
68
69 if (mask)
70 return (set_errno(EINVAL));
71 if ((cmd == MC_LOCKAS) || (cmd == MC_UNLOCKAS)) {
72 if ((addr != 0) || (len != 0)) {
73 return (set_errno(EINVAL));
74 }
75 } else if (cmd != MC_HAT_ADVISE) {
76 if (((uintptr_t)addr & PAGEOFFSET) != 0 || len == 0) {
77 return (set_errno(EINVAL));
78 }
79 /*
80 * We're only concerned with the address range
81 * here, not the protections. The protections
82 * are only used as a "filter" in this code,
83 * they aren't set or modified here.
84 */
85 if (valid_usr_range(addr, len, 0, as,
86 as->a_userlimit) != RANGE_OKAY) {
87 return (set_errno(ENOMEM));
88 }
89 }
90
91 if (cmd == MC_HAT_ADVISE) {
92 if (attr != 0 || mask != 0) {
93 return (set_errno(EINVAL));
94 }
95
96 } else {
97 if ((VALID_ATTR & attr) != attr) {
98 return (set_errno(EINVAL));
99 }
100 if ((attr & SHARED) && (attr & PRIVATE)) {
101 return (set_errno(EINVAL));
102 }
103 if (((cmd == MC_LOCKAS) || (cmd == MC_LOCK) ||
104 (cmd == MC_UNLOCKAS) || (cmd == MC_UNLOCK)) &&
105 (error = secpolicy_lock_memory(CRED())) != 0)
106 return (set_errno(error));
107 }
108 if (attr) {
109 attr |= PROT_USER;
110 }
111
112 switch (cmd) {
113 case MC_SYNC:
114 /*
115 * MS_SYNC used to be defined to be zero but is now non-zero.
116 * For binary compatibility we still accept zero
117 * (the absence of MS_ASYNC) to mean the same thing.
118 * Binary compatibility is not an issue for MS_INVALCURPROC.
119 */
120 iarg = (uintptr_t)arg;
121 if ((iarg & ~MS_INVALIDATE) == 0)
122 iarg |= MS_SYNC;
123
124 if (((iarg &
125 ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE|MS_INVALCURPROC)) != 0) ||
126 ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC)) ||
127 ((iarg & (MS_INVALIDATE|MS_INVALCURPROC)) ==
128 (MS_INVALIDATE|MS_INVALCURPROC))) {
129 error = set_errno(EINVAL);
130 } else {
131 error = as_ctl(as, addr, len, cmd, attr, iarg, NULL, 0);
132 if (error) {
133 (void) set_errno(error);
134 }
135 }
136 return (error);
137 case MC_LOCKAS:
138 if ((uintptr_t)arg & ~(MCL_FUTURE|MCL_CURRENT) ||
139 (uintptr_t)arg == 0) {
140 return (set_errno(EINVAL));
141 }
142 break;
143 case MC_LOCK:
144 case MC_UNLOCKAS:
145 case MC_UNLOCK:
146 break;
147 case MC_HAT_ADVISE:
148 /*
149 * Set prefered page size.
150 */
151 STRUCT_INIT(mha, get_udatamodel());
152 if (copyin(arg, STRUCT_BUF(mha), STRUCT_SIZE(mha))) {
153 return (set_errno(EFAULT));
154 }
155
156 pgcmd = STRUCT_FGET(mha, mha_cmd);
157
158 /*
159 * Currently only MHA_MAPSIZE_VA, MHA_MAPSIZE_STACK
160 * and MHA_MAPSIZE_BSSBRK are supported. Only one
161 * command may be specified at a time.
162 */
163 if ((~(MHA_MAPSIZE_VA|MHA_MAPSIZE_STACK|MHA_MAPSIZE_BSSBRK) &
164 pgcmd) || pgcmd == 0 || !ISP2(pgcmd) ||
165 STRUCT_FGET(mha, mha_flags))
166 return (set_errno(EINVAL));
167
168 pgsz = STRUCT_FGET(mha, mha_pagesize);
169
170 /*
171 * call platform specific map_pgsz() routine to get the
172 * optimal pgsz if pgsz is 0.
173 *
174 * For stack and heap operations addr and len must be zero.
175 */
176 if ((pgcmd & (MHA_MAPSIZE_BSSBRK|MHA_MAPSIZE_STACK)) != 0) {
177 if (addr != NULL || len != 0) {
178 return (set_errno(EINVAL));
179 }
180
181 /*
182 * Disable autompss for this process unless pgsz == 0,
183 * which means the system should pick. In the
184 * pgsz == 0 case, leave the SAUTOLPG setting alone, as
185 * we don't want to enable it when someone has
186 * disabled automatic large page selection for the
187 * whole system.
188 */
189 mutex_enter(&p->p_lock);
190 if (pgsz != 0) {
191 p->p_flag &= ~SAUTOLPG;
192 }
193 mutex_exit(&p->p_lock);
194
195 as_rangelock(as);
196
197 if (pgsz == 0) {
198 int type;
199
200 if (pgcmd == MHA_MAPSIZE_BSSBRK)
201 type = MAPPGSZ_HEAP;
202 else
203 type = MAPPGSZ_STK;
204
205 pgsz = map_pgsz(type, p, 0, 0, 1);
206 }
207 } else {
208 /*
209 * addr and len must be valid for range specified.
210 */
211 if (valid_usr_range(addr, len, 0, as,
212 as->a_userlimit) != RANGE_OKAY) {
213 return (set_errno(ENOMEM));
214 }
215 /*
216 * Note that we don't disable automatic large page
217 * selection for anon segments based on use of
218 * memcntl().
219 */
220 if (pgsz == 0) {
221 error = as_set_default_lpsize(as, addr, len);
222 if (error) {
223 (void) set_errno(error);
224 }
225 return (error);
226 }
227
228 /*
229 * addr and len must be prefered page size aligned
230 */
231 if (!IS_P2ALIGNED(addr, pgsz) ||
232 !IS_P2ALIGNED(len, pgsz)) {
233 return (set_errno(EINVAL));
234 }
235 }
236
237 szc = mem_getpgszc(pgsz);
238 if (szc == (uint_t)-1) {
239 if ((pgcmd & (MHA_MAPSIZE_BSSBRK|MHA_MAPSIZE_STACK))
240 != 0) {
241 as_rangeunlock(as);
242 }
243 return (set_errno(EINVAL));
244 }
245
246 /*
247 * For stack and heap operations we first need to pad
248 * out existing range (create new mappings) to the new
249 * prefered page size boundary. Also the start of the
250 * .bss for the heap or user's stack base may not be on
251 * the new prefered page size boundary. For these cases
252 * we align the base of the request on the new prefered
253 * page size.
254 */
255 if (pgcmd & MHA_MAPSIZE_BSSBRK) {
256 if (szc == p->p_brkpageszc) {
257 as_rangeunlock(as);
258 return (0);
259 }
260 if (szc > p->p_brkpageszc) {
261 error = brk_internal(p->p_brkbase
262 + p->p_brksize, szc);
263 if (error) {
264 as_rangeunlock(as);
265 return (set_errno(error));
266 }
267 }
268 /*
269 * It is possible for brk_internal to silently fail to
270 * promote the heap size, so don't panic or ASSERT.
271 */
272 if (!IS_P2ALIGNED(p->p_brkbase + p->p_brksize, pgsz)) {
273 as_rangeunlock(as);
274 return (set_errno(ENOMEM));
275 }
276 oszc = p->p_brkpageszc;
277 p->p_brkpageszc = szc;
278
279 addr = (caddr_t)P2ROUNDUP((uintptr_t)p->p_bssbase,
280 pgsz);
281 len = (p->p_brkbase + p->p_brksize) - addr;
282 ASSERT(IS_P2ALIGNED(len, pgsz));
283 /*
284 * Perhaps no existing pages to promote.
285 */
286 if (len == 0) {
287 as_rangeunlock(as);
288 return (0);
289 }
290 }
291 /*
292 * The code below, as does grow.c, assumes stacks always grow
293 * downward.
294 */
295 if (pgcmd & MHA_MAPSIZE_STACK) {
296 if (szc == p->p_stkpageszc) {
297 as_rangeunlock(as);
298 return (0);
299 }
300
301 if (szc > p->p_stkpageszc) {
302 error = grow_internal(p->p_usrstack -
303 p->p_stksize, szc);
304 if (error) {
305 as_rangeunlock(as);
306 return (set_errno(error));
307 }
308 }
309 /*
310 * It is possible for grow_internal to silently fail to
311 * promote the stack size, so don't panic or ASSERT.
312 */
313 if (!IS_P2ALIGNED(p->p_usrstack - p->p_stksize, pgsz)) {
314 as_rangeunlock(as);
315 return (set_errno(ENOMEM));
316 }
317 oszc = p->p_stkpageszc;
318 p->p_stkpageszc = szc;
319
320 addr = p->p_usrstack - p->p_stksize;
321 len = P2ALIGN(p->p_stksize, pgsz);
322
323 /*
324 * Perhaps nothing to promote.
325 */
326 if (len == 0 || addr >= p->p_usrstack ||
327 (addr + len) < addr) {
328 as_rangeunlock(as);
329 return (0);
330 }
331 }
332 ASSERT(IS_P2ALIGNED(addr, pgsz));
333 ASSERT(IS_P2ALIGNED(len, pgsz));
334 error = as_setpagesize(as, addr, len, szc, B_TRUE);
335
336 /*
337 * On stack or heap failures restore original
338 * pg size code.
339 */
340 if (error) {
341 if ((pgcmd & MHA_MAPSIZE_BSSBRK) != 0) {
342 p->p_brkpageszc = oszc;
343 }
344 if ((pgcmd & MHA_MAPSIZE_STACK) != 0) {
345 p->p_stkpageszc = oszc;
346 }
347 (void) set_errno(error);
348 }
349 if ((pgcmd & (MHA_MAPSIZE_BSSBRK|MHA_MAPSIZE_STACK)) != 0) {
350 as_rangeunlock(as);
351 }
352 return (error);
353 case MC_ADVISE:
354 if ((uintptr_t)arg == MADV_FREE ||
355 (uintptr_t)arg == MADV_PURGE) {
356 len &= PAGEMASK;
357 }
358 switch ((uintptr_t)arg) {
359 case MADV_WILLNEED:
360 fc = as_faulta(as, addr, len);
361 if (fc) {
362 if (FC_CODE(fc) == FC_OBJERR)
363 error = set_errno(FC_ERRNO(fc));
364 else if (FC_CODE(fc) == FC_NOMAP)
365 error = set_errno(ENOMEM);
366 else
367 error = set_errno(EINVAL);
368 return (error);
369 }
370 break;
371
372 case MADV_DONTNEED:
373 /*
374 * For now, don't need is turned into an as_ctl(MC_SYNC)
375 * operation flagged for async invalidate.
376 */
377 error = as_ctl(as, addr, len, MC_SYNC, attr,
378 MS_ASYNC | MS_INVALIDATE, NULL, 0);
379 if (error)
380 (void) set_errno(error);
381 return (error);
382
383 default:
384 error = as_ctl(as, addr, len, cmd, attr,
385 (uintptr_t)arg, NULL, 0);
386 if (error)
387 (void) set_errno(error);
388 return (error);
389 }
390 break;
391 case MC_INHERIT_ZERO:
392 if (arg != 0 || attr != 0 || mask != 0)
393 return (set_errno(EINVAL));
394 break;
395 default:
396 return (set_errno(EINVAL));
397 }
398
399 error = as_ctl(as, addr, len, cmd, attr, (uintptr_t)arg, NULL, 0);
400
401 if (error)
402 (void) set_errno(error);
403 return (error);
404 }
405
406 /*
407 * Return page size code for page size passed in. If
408 * matching page size not found or supported, return -1.
409 */
410 static uint_t
411 mem_getpgszc(size_t pgsz) {
412 return ((uint_t)page_szc_user_filtered(pgsz));
413 }