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