]>
Commit | Line | Data |
---|---|---|
84778508 BS |
1 | /* |
2 | * mmap support for qemu | |
3 | * | |
4 | * Copyright (c) 2003 - 2008 Fabrice Bellard | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
8167ee88 | 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
84778508 BS |
18 | */ |
19 | #include <stdlib.h> | |
20 | #include <stdio.h> | |
21 | #include <stdarg.h> | |
22 | #include <string.h> | |
23 | #include <unistd.h> | |
24 | #include <errno.h> | |
25 | #include <sys/mman.h> | |
26 | ||
27 | #include "qemu.h" | |
28 | #include "qemu-common.h" | |
6c173b3c | 29 | #include "bsd-mman.h" |
84778508 BS |
30 | |
31 | //#define DEBUG_MMAP | |
32 | ||
2f7bb878 | 33 | #if defined(CONFIG_USE_NPTL) |
84778508 BS |
34 | pthread_mutex_t mmap_mutex; |
35 | static int __thread mmap_lock_count; | |
36 | ||
37 | void mmap_lock(void) | |
38 | { | |
39 | if (mmap_lock_count++ == 0) { | |
40 | pthread_mutex_lock(&mmap_mutex); | |
41 | } | |
42 | } | |
43 | ||
44 | void mmap_unlock(void) | |
45 | { | |
46 | if (--mmap_lock_count == 0) { | |
47 | pthread_mutex_unlock(&mmap_mutex); | |
48 | } | |
49 | } | |
50 | ||
51 | /* Grab lock to make sure things are in a consistent state after fork(). */ | |
52 | void mmap_fork_start(void) | |
53 | { | |
54 | if (mmap_lock_count) | |
55 | abort(); | |
56 | pthread_mutex_lock(&mmap_mutex); | |
57 | } | |
58 | ||
59 | void mmap_fork_end(int child) | |
60 | { | |
61 | if (child) | |
62 | pthread_mutex_init(&mmap_mutex, NULL); | |
63 | else | |
64 | pthread_mutex_unlock(&mmap_mutex); | |
65 | } | |
66 | #else | |
67 | /* We aren't threadsafe to start with, so no need to worry about locking. */ | |
68 | void mmap_lock(void) | |
69 | { | |
70 | } | |
71 | ||
72 | void mmap_unlock(void) | |
73 | { | |
74 | } | |
75 | #endif | |
76 | ||
77 | void *qemu_vmalloc(size_t size) | |
78 | { | |
79 | void *p; | |
80 | unsigned long addr; | |
81 | mmap_lock(); | |
82 | /* Use map and mark the pages as used. */ | |
83 | p = mmap(NULL, size, PROT_READ | PROT_WRITE, | |
84 | MAP_PRIVATE | MAP_ANON, -1, 0); | |
85 | ||
86 | addr = (unsigned long)p; | |
87 | if (addr == (target_ulong) addr) { | |
88 | /* Allocated region overlaps guest address space. | |
89 | This may recurse. */ | |
90 | page_set_flags(addr & TARGET_PAGE_MASK, TARGET_PAGE_ALIGN(addr + size), | |
91 | PAGE_RESERVED); | |
92 | } | |
93 | ||
94 | mmap_unlock(); | |
95 | return p; | |
96 | } | |
97 | ||
98 | void *qemu_malloc(size_t size) | |
99 | { | |
100 | char * p; | |
101 | size += 16; | |
102 | p = qemu_vmalloc(size); | |
103 | *(size_t *)p = size; | |
104 | return p + 16; | |
105 | } | |
106 | ||
107 | /* We use map, which is always zero initialized. */ | |
108 | void * qemu_mallocz(size_t size) | |
109 | { | |
110 | return qemu_malloc(size); | |
111 | } | |
112 | ||
113 | void qemu_free(void *ptr) | |
114 | { | |
115 | /* FIXME: We should unmark the reserved pages here. However this gets | |
116 | complicated when one target page spans multiple host pages, so we | |
117 | don't bother. */ | |
118 | size_t *p; | |
119 | p = (size_t *)((char *)ptr - 16); | |
120 | munmap(p, *p); | |
121 | } | |
122 | ||
004c9ef4 BS |
123 | void *qemu_realloc(void *ptr, size_t size) |
124 | { | |
125 | size_t old_size, copy; | |
126 | void *new_ptr; | |
127 | ||
baa8c602 | 128 | if (!ptr) |
129 | return qemu_malloc(size); | |
004c9ef4 BS |
130 | old_size = *(size_t *)((char *)ptr - 16); |
131 | copy = old_size < size ? old_size : size; | |
132 | new_ptr = qemu_malloc(size); | |
133 | memcpy(new_ptr, ptr, copy); | |
134 | qemu_free(ptr); | |
135 | return new_ptr; | |
136 | } | |
137 | ||
84778508 BS |
138 | /* NOTE: all the constants are the HOST ones, but addresses are target. */ |
139 | int target_mprotect(abi_ulong start, abi_ulong len, int prot) | |
140 | { | |
141 | abi_ulong end, host_start, host_end, addr; | |
142 | int prot1, ret; | |
143 | ||
144 | #ifdef DEBUG_MMAP | |
145 | printf("mprotect: start=0x" TARGET_FMT_lx | |
146 | " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len, | |
147 | prot & PROT_READ ? 'r' : '-', | |
148 | prot & PROT_WRITE ? 'w' : '-', | |
149 | prot & PROT_EXEC ? 'x' : '-'); | |
150 | #endif | |
151 | ||
152 | if ((start & ~TARGET_PAGE_MASK) != 0) | |
153 | return -EINVAL; | |
154 | len = TARGET_PAGE_ALIGN(len); | |
155 | end = start + len; | |
156 | if (end < start) | |
157 | return -EINVAL; | |
158 | prot &= PROT_READ | PROT_WRITE | PROT_EXEC; | |
159 | if (len == 0) | |
160 | return 0; | |
161 | ||
162 | mmap_lock(); | |
163 | host_start = start & qemu_host_page_mask; | |
164 | host_end = HOST_PAGE_ALIGN(end); | |
165 | if (start > host_start) { | |
166 | /* handle host page containing start */ | |
167 | prot1 = prot; | |
168 | for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) { | |
169 | prot1 |= page_get_flags(addr); | |
170 | } | |
171 | if (host_end == host_start + qemu_host_page_size) { | |
172 | for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) { | |
173 | prot1 |= page_get_flags(addr); | |
174 | } | |
175 | end = host_end; | |
176 | } | |
177 | ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS); | |
178 | if (ret != 0) | |
179 | goto error; | |
180 | host_start += qemu_host_page_size; | |
181 | } | |
182 | if (end < host_end) { | |
183 | prot1 = prot; | |
184 | for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) { | |
185 | prot1 |= page_get_flags(addr); | |
186 | } | |
187 | ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size, | |
188 | prot1 & PAGE_BITS); | |
189 | if (ret != 0) | |
190 | goto error; | |
191 | host_end -= qemu_host_page_size; | |
192 | } | |
193 | ||
194 | /* handle the pages in the middle */ | |
195 | if (host_start < host_end) { | |
196 | ret = mprotect(g2h(host_start), host_end - host_start, prot); | |
197 | if (ret != 0) | |
198 | goto error; | |
199 | } | |
200 | page_set_flags(start, start + len, prot | PAGE_VALID); | |
201 | mmap_unlock(); | |
202 | return 0; | |
203 | error: | |
204 | mmap_unlock(); | |
205 | return ret; | |
206 | } | |
207 | ||
208 | /* map an incomplete host page */ | |
209 | static int mmap_frag(abi_ulong real_start, | |
210 | abi_ulong start, abi_ulong end, | |
211 | int prot, int flags, int fd, abi_ulong offset) | |
212 | { | |
213 | abi_ulong real_end, addr; | |
214 | void *host_start; | |
215 | int prot1, prot_new; | |
216 | ||
217 | real_end = real_start + qemu_host_page_size; | |
218 | host_start = g2h(real_start); | |
219 | ||
220 | /* get the protection of the target pages outside the mapping */ | |
221 | prot1 = 0; | |
222 | for(addr = real_start; addr < real_end; addr++) { | |
223 | if (addr < start || addr >= end) | |
224 | prot1 |= page_get_flags(addr); | |
225 | } | |
226 | ||
227 | if (prot1 == 0) { | |
228 | /* no page was there, so we allocate one */ | |
229 | void *p = mmap(host_start, qemu_host_page_size, prot, | |
230 | flags | MAP_ANON, -1, 0); | |
231 | if (p == MAP_FAILED) | |
232 | return -1; | |
233 | prot1 = prot; | |
234 | } | |
235 | prot1 &= PAGE_BITS; | |
236 | ||
237 | prot_new = prot | prot1; | |
238 | if (!(flags & MAP_ANON)) { | |
239 | /* msync() won't work here, so we return an error if write is | |
240 | possible while it is a shared mapping */ | |
6c173b3c | 241 | if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && |
84778508 | 242 | (prot & PROT_WRITE)) |
059bca46 | 243 | return -1; |
84778508 BS |
244 | |
245 | /* adjust protection to be able to read */ | |
246 | if (!(prot1 & PROT_WRITE)) | |
247 | mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE); | |
248 | ||
249 | /* read the corresponding file data */ | |
250 | pread(fd, g2h(start), end - start, offset); | |
251 | ||
252 | /* put final protection */ | |
253 | if (prot_new != (prot1 | PROT_WRITE)) | |
254 | mprotect(host_start, qemu_host_page_size, prot_new); | |
255 | } else { | |
256 | /* just update the protection */ | |
257 | if (prot_new != prot1) { | |
258 | mprotect(host_start, qemu_host_page_size, prot_new); | |
259 | } | |
260 | } | |
261 | return 0; | |
262 | } | |
263 | ||
264 | #if defined(__CYGWIN__) | |
265 | /* Cygwin doesn't have a whole lot of address space. */ | |
266 | static abi_ulong mmap_next_start = 0x18000000; | |
267 | #else | |
268 | static abi_ulong mmap_next_start = 0x40000000; | |
269 | #endif | |
270 | ||
271 | unsigned long last_brk; | |
272 | ||
273 | /* find a free memory area of size 'size'. The search starts at | |
274 | 'start'. If 'start' == 0, then a default start address is used. | |
275 | Return -1 if error. | |
276 | */ | |
277 | /* page_init() marks pages used by the host as reserved to be sure not | |
278 | to use them. */ | |
279 | static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size) | |
280 | { | |
281 | abi_ulong addr, addr1, addr_start; | |
282 | int prot; | |
283 | unsigned long new_brk; | |
284 | ||
285 | new_brk = (unsigned long)sbrk(0); | |
286 | if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) { | |
287 | /* This is a hack to catch the host allocating memory with brk(). | |
288 | If it uses mmap then we loose. | |
289 | FIXME: We really want to avoid the host allocating memory in | |
290 | the first place, and maybe leave some slack to avoid switching | |
291 | to mmap. */ | |
292 | page_set_flags(last_brk & TARGET_PAGE_MASK, | |
293 | TARGET_PAGE_ALIGN(new_brk), | |
294 | PAGE_RESERVED); | |
295 | } | |
296 | last_brk = new_brk; | |
297 | ||
298 | size = HOST_PAGE_ALIGN(size); | |
299 | start = start & qemu_host_page_mask; | |
300 | addr = start; | |
301 | if (addr == 0) | |
302 | addr = mmap_next_start; | |
303 | addr_start = addr; | |
304 | for(;;) { | |
305 | prot = 0; | |
306 | for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) { | |
307 | prot |= page_get_flags(addr1); | |
308 | } | |
309 | if (prot == 0) | |
310 | break; | |
311 | addr += qemu_host_page_size; | |
312 | /* we found nothing */ | |
313 | if (addr == addr_start) | |
314 | return (abi_ulong)-1; | |
315 | } | |
316 | if (start == 0) | |
317 | mmap_next_start = addr + size; | |
318 | return addr; | |
319 | } | |
320 | ||
321 | /* NOTE: all the constants are the HOST ones */ | |
322 | abi_long target_mmap(abi_ulong start, abi_ulong len, int prot, | |
323 | int flags, int fd, abi_ulong offset) | |
324 | { | |
325 | abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len; | |
326 | unsigned long host_start; | |
327 | ||
328 | mmap_lock(); | |
329 | #ifdef DEBUG_MMAP | |
330 | { | |
331 | printf("mmap: start=0x" TARGET_FMT_lx | |
332 | " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=", | |
333 | start, len, | |
334 | prot & PROT_READ ? 'r' : '-', | |
335 | prot & PROT_WRITE ? 'w' : '-', | |
336 | prot & PROT_EXEC ? 'x' : '-'); | |
337 | if (flags & MAP_FIXED) | |
338 | printf("MAP_FIXED "); | |
339 | if (flags & MAP_ANON) | |
340 | printf("MAP_ANON "); | |
6c173b3c | 341 | switch(flags & TARGET_BSD_MAP_FLAGMASK) { |
84778508 BS |
342 | case MAP_PRIVATE: |
343 | printf("MAP_PRIVATE "); | |
344 | break; | |
345 | case MAP_SHARED: | |
346 | printf("MAP_SHARED "); | |
347 | break; | |
348 | default: | |
6c173b3c | 349 | printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK); |
84778508 BS |
350 | break; |
351 | } | |
352 | printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset); | |
353 | } | |
354 | #endif | |
355 | ||
356 | if (offset & ~TARGET_PAGE_MASK) { | |
357 | errno = EINVAL; | |
358 | goto fail; | |
359 | } | |
360 | ||
361 | len = TARGET_PAGE_ALIGN(len); | |
362 | if (len == 0) | |
363 | goto the_end; | |
364 | real_start = start & qemu_host_page_mask; | |
365 | ||
366 | if (!(flags & MAP_FIXED)) { | |
367 | abi_ulong mmap_start; | |
368 | void *p; | |
369 | host_offset = offset & qemu_host_page_mask; | |
370 | host_len = len + offset - host_offset; | |
371 | host_len = HOST_PAGE_ALIGN(host_len); | |
372 | mmap_start = mmap_find_vma(real_start, host_len); | |
373 | if (mmap_start == (abi_ulong)-1) { | |
374 | errno = ENOMEM; | |
375 | goto fail; | |
376 | } | |
377 | /* Note: we prefer to control the mapping address. It is | |
378 | especially important if qemu_host_page_size > | |
379 | qemu_real_host_page_size */ | |
380 | p = mmap(g2h(mmap_start), | |
381 | host_len, prot, flags | MAP_FIXED, fd, host_offset); | |
382 | if (p == MAP_FAILED) | |
383 | goto fail; | |
384 | /* update start so that it points to the file position at 'offset' */ | |
385 | host_start = (unsigned long)p; | |
386 | if (!(flags & MAP_ANON)) | |
387 | host_start += offset - host_offset; | |
388 | start = h2g(host_start); | |
389 | } else { | |
390 | int flg; | |
391 | target_ulong addr; | |
392 | ||
393 | if (start & ~TARGET_PAGE_MASK) { | |
394 | errno = EINVAL; | |
395 | goto fail; | |
396 | } | |
397 | end = start + len; | |
398 | real_end = HOST_PAGE_ALIGN(end); | |
399 | ||
400 | for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) { | |
401 | flg = page_get_flags(addr); | |
402 | if (flg & PAGE_RESERVED) { | |
403 | errno = ENXIO; | |
404 | goto fail; | |
405 | } | |
406 | } | |
407 | ||
408 | /* worst case: we cannot map the file because the offset is not | |
409 | aligned, so we read it */ | |
410 | if (!(flags & MAP_ANON) && | |
411 | (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) { | |
412 | /* msync() won't work here, so we return an error if write is | |
413 | possible while it is a shared mapping */ | |
6c173b3c | 414 | if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && |
84778508 BS |
415 | (prot & PROT_WRITE)) { |
416 | errno = EINVAL; | |
417 | goto fail; | |
418 | } | |
419 | retaddr = target_mmap(start, len, prot | PROT_WRITE, | |
420 | MAP_FIXED | MAP_PRIVATE | MAP_ANON, | |
421 | -1, 0); | |
422 | if (retaddr == -1) | |
423 | goto fail; | |
424 | pread(fd, g2h(start), len, offset); | |
425 | if (!(prot & PROT_WRITE)) { | |
426 | ret = target_mprotect(start, len, prot); | |
427 | if (ret != 0) { | |
428 | start = ret; | |
429 | goto the_end; | |
430 | } | |
431 | } | |
432 | goto the_end; | |
433 | } | |
434 | ||
435 | /* handle the start of the mapping */ | |
436 | if (start > real_start) { | |
437 | if (real_end == real_start + qemu_host_page_size) { | |
438 | /* one single host page */ | |
439 | ret = mmap_frag(real_start, start, end, | |
440 | prot, flags, fd, offset); | |
441 | if (ret == -1) | |
442 | goto fail; | |
443 | goto the_end1; | |
444 | } | |
445 | ret = mmap_frag(real_start, start, real_start + qemu_host_page_size, | |
446 | prot, flags, fd, offset); | |
447 | if (ret == -1) | |
448 | goto fail; | |
449 | real_start += qemu_host_page_size; | |
450 | } | |
451 | /* handle the end of the mapping */ | |
452 | if (end < real_end) { | |
453 | ret = mmap_frag(real_end - qemu_host_page_size, | |
454 | real_end - qemu_host_page_size, real_end, | |
455 | prot, flags, fd, | |
456 | offset + real_end - qemu_host_page_size - start); | |
457 | if (ret == -1) | |
458 | goto fail; | |
459 | real_end -= qemu_host_page_size; | |
460 | } | |
461 | ||
462 | /* map the middle (easier) */ | |
463 | if (real_start < real_end) { | |
464 | void *p; | |
465 | unsigned long offset1; | |
466 | if (flags & MAP_ANON) | |
467 | offset1 = 0; | |
468 | else | |
469 | offset1 = offset + real_start - start; | |
470 | p = mmap(g2h(real_start), real_end - real_start, | |
471 | prot, flags, fd, offset1); | |
472 | if (p == MAP_FAILED) | |
473 | goto fail; | |
474 | } | |
475 | } | |
476 | the_end1: | |
477 | page_set_flags(start, start + len, prot | PAGE_VALID); | |
478 | the_end: | |
479 | #ifdef DEBUG_MMAP | |
480 | printf("ret=0x" TARGET_FMT_lx "\n", start); | |
481 | page_dump(stdout); | |
482 | printf("\n"); | |
483 | #endif | |
484 | mmap_unlock(); | |
485 | return start; | |
486 | fail: | |
487 | mmap_unlock(); | |
488 | return -1; | |
489 | } | |
490 | ||
491 | int target_munmap(abi_ulong start, abi_ulong len) | |
492 | { | |
493 | abi_ulong end, real_start, real_end, addr; | |
494 | int prot, ret; | |
495 | ||
496 | #ifdef DEBUG_MMAP | |
497 | printf("munmap: start=0x%lx len=0x%lx\n", start, len); | |
498 | #endif | |
499 | if (start & ~TARGET_PAGE_MASK) | |
500 | return -EINVAL; | |
501 | len = TARGET_PAGE_ALIGN(len); | |
502 | if (len == 0) | |
503 | return -EINVAL; | |
504 | mmap_lock(); | |
505 | end = start + len; | |
506 | real_start = start & qemu_host_page_mask; | |
507 | real_end = HOST_PAGE_ALIGN(end); | |
508 | ||
509 | if (start > real_start) { | |
510 | /* handle host page containing start */ | |
511 | prot = 0; | |
512 | for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) { | |
513 | prot |= page_get_flags(addr); | |
514 | } | |
515 | if (real_end == real_start + qemu_host_page_size) { | |
516 | for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { | |
517 | prot |= page_get_flags(addr); | |
518 | } | |
519 | end = real_end; | |
520 | } | |
521 | if (prot != 0) | |
522 | real_start += qemu_host_page_size; | |
523 | } | |
524 | if (end < real_end) { | |
525 | prot = 0; | |
526 | for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { | |
527 | prot |= page_get_flags(addr); | |
528 | } | |
529 | if (prot != 0) | |
530 | real_end -= qemu_host_page_size; | |
531 | } | |
532 | ||
533 | ret = 0; | |
534 | /* unmap what we can */ | |
535 | if (real_start < real_end) { | |
536 | ret = munmap(g2h(real_start), real_end - real_start); | |
537 | } | |
538 | ||
539 | if (ret == 0) | |
540 | page_set_flags(start, start + len, 0); | |
541 | mmap_unlock(); | |
542 | return ret; | |
543 | } | |
544 | ||
545 | int target_msync(abi_ulong start, abi_ulong len, int flags) | |
546 | { | |
547 | abi_ulong end; | |
548 | ||
549 | if (start & ~TARGET_PAGE_MASK) | |
550 | return -EINVAL; | |
551 | len = TARGET_PAGE_ALIGN(len); | |
552 | end = start + len; | |
553 | if (end < start) | |
554 | return -EINVAL; | |
555 | if (end == start) | |
556 | return 0; | |
557 | ||
558 | start &= qemu_host_page_mask; | |
559 | return msync(g2h(start), end - start, flags); | |
560 | } |