keep the PID file locked for the lifetime of the process
[qemu.git] / bsd-user / mmap.c
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
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
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"
29 #include "bsd-mman.h"
30
31 //#define DEBUG_MMAP
32
33 #if defined(CONFIG_USE_NPTL)
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 mmap_lock();
81 /* Use map and mark the pages as used. */
82 p = mmap(NULL, size, PROT_READ | PROT_WRITE,
83 MAP_PRIVATE | MAP_ANON, -1, 0);
84
85 if (h2g_valid(p)) {
86 /* Allocated region overlaps guest address space.
87 This may recurse. */
88 abi_ulong addr = h2g(p);
89 page_set_flags(addr & TARGET_PAGE_MASK, TARGET_PAGE_ALIGN(addr + size),
90 PAGE_RESERVED);
91 }
92
93 mmap_unlock();
94 return p;
95 }
96
97 void *g_malloc(size_t size)
98 {
99 char * p;
100 size += 16;
101 p = qemu_vmalloc(size);
102 *(size_t *)p = size;
103 return p + 16;
104 }
105
106 /* We use map, which is always zero initialized. */
107 void * g_malloc0(size_t size)
108 {
109 return g_malloc(size);
110 }
111
112 void g_free(void *ptr)
113 {
114 /* FIXME: We should unmark the reserved pages here. However this gets
115 complicated when one target page spans multiple host pages, so we
116 don't bother. */
117 size_t *p;
118 p = (size_t *)((char *)ptr - 16);
119 munmap(p, *p);
120 }
121
122 void *g_realloc(void *ptr, size_t size)
123 {
124 size_t old_size, copy;
125 void *new_ptr;
126
127 if (!ptr)
128 return g_malloc(size);
129 old_size = *(size_t *)((char *)ptr - 16);
130 copy = old_size < size ? old_size : size;
131 new_ptr = g_malloc(size);
132 memcpy(new_ptr, ptr, copy);
133 g_free(ptr);
134 return new_ptr;
135 }
136
137 /* NOTE: all the constants are the HOST ones, but addresses are target. */
138 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
139 {
140 abi_ulong end, host_start, host_end, addr;
141 int prot1, ret;
142
143 #ifdef DEBUG_MMAP
144 printf("mprotect: start=0x" TARGET_FMT_lx
145 " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
146 prot & PROT_READ ? 'r' : '-',
147 prot & PROT_WRITE ? 'w' : '-',
148 prot & PROT_EXEC ? 'x' : '-');
149 #endif
150
151 if ((start & ~TARGET_PAGE_MASK) != 0)
152 return -EINVAL;
153 len = TARGET_PAGE_ALIGN(len);
154 end = start + len;
155 if (end < start)
156 return -EINVAL;
157 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
158 if (len == 0)
159 return 0;
160
161 mmap_lock();
162 host_start = start & qemu_host_page_mask;
163 host_end = HOST_PAGE_ALIGN(end);
164 if (start > host_start) {
165 /* handle host page containing start */
166 prot1 = prot;
167 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
168 prot1 |= page_get_flags(addr);
169 }
170 if (host_end == host_start + qemu_host_page_size) {
171 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
172 prot1 |= page_get_flags(addr);
173 }
174 end = host_end;
175 }
176 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
177 if (ret != 0)
178 goto error;
179 host_start += qemu_host_page_size;
180 }
181 if (end < host_end) {
182 prot1 = prot;
183 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
184 prot1 |= page_get_flags(addr);
185 }
186 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
187 prot1 & PAGE_BITS);
188 if (ret != 0)
189 goto error;
190 host_end -= qemu_host_page_size;
191 }
192
193 /* handle the pages in the middle */
194 if (host_start < host_end) {
195 ret = mprotect(g2h(host_start), host_end - host_start, prot);
196 if (ret != 0)
197 goto error;
198 }
199 page_set_flags(start, start + len, prot | PAGE_VALID);
200 mmap_unlock();
201 return 0;
202 error:
203 mmap_unlock();
204 return ret;
205 }
206
207 /* map an incomplete host page */
208 static int mmap_frag(abi_ulong real_start,
209 abi_ulong start, abi_ulong end,
210 int prot, int flags, int fd, abi_ulong offset)
211 {
212 abi_ulong real_end, addr;
213 void *host_start;
214 int prot1, prot_new;
215
216 real_end = real_start + qemu_host_page_size;
217 host_start = g2h(real_start);
218
219 /* get the protection of the target pages outside the mapping */
220 prot1 = 0;
221 for(addr = real_start; addr < real_end; addr++) {
222 if (addr < start || addr >= end)
223 prot1 |= page_get_flags(addr);
224 }
225
226 if (prot1 == 0) {
227 /* no page was there, so we allocate one */
228 void *p = mmap(host_start, qemu_host_page_size, prot,
229 flags | MAP_ANON, -1, 0);
230 if (p == MAP_FAILED)
231 return -1;
232 prot1 = prot;
233 }
234 prot1 &= PAGE_BITS;
235
236 prot_new = prot | prot1;
237 if (!(flags & MAP_ANON)) {
238 /* msync() won't work here, so we return an error if write is
239 possible while it is a shared mapping */
240 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
241 (prot & PROT_WRITE))
242 return -1;
243
244 /* adjust protection to be able to read */
245 if (!(prot1 & PROT_WRITE))
246 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
247
248 /* read the corresponding file data */
249 pread(fd, g2h(start), end - start, offset);
250
251 /* put final protection */
252 if (prot_new != (prot1 | PROT_WRITE))
253 mprotect(host_start, qemu_host_page_size, prot_new);
254 } else {
255 /* just update the protection */
256 if (prot_new != prot1) {
257 mprotect(host_start, qemu_host_page_size, prot_new);
258 }
259 }
260 return 0;
261 }
262
263 #if defined(__CYGWIN__)
264 /* Cygwin doesn't have a whole lot of address space. */
265 static abi_ulong mmap_next_start = 0x18000000;
266 #else
267 static abi_ulong mmap_next_start = 0x40000000;
268 #endif
269
270 unsigned long last_brk;
271
272 /* find a free memory area of size 'size'. The search starts at
273 'start'. If 'start' == 0, then a default start address is used.
274 Return -1 if error.
275 */
276 /* page_init() marks pages used by the host as reserved to be sure not
277 to use them. */
278 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
279 {
280 abi_ulong addr, addr1, addr_start;
281 int prot;
282 unsigned long new_brk;
283
284 new_brk = (unsigned long)sbrk(0);
285 if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
286 /* This is a hack to catch the host allocating memory with brk().
287 If it uses mmap then we loose.
288 FIXME: We really want to avoid the host allocating memory in
289 the first place, and maybe leave some slack to avoid switching
290 to mmap. */
291 page_set_flags(last_brk & TARGET_PAGE_MASK,
292 TARGET_PAGE_ALIGN(new_brk),
293 PAGE_RESERVED);
294 }
295 last_brk = new_brk;
296
297 size = HOST_PAGE_ALIGN(size);
298 start = start & qemu_host_page_mask;
299 addr = start;
300 if (addr == 0)
301 addr = mmap_next_start;
302 addr_start = addr;
303 for(;;) {
304 prot = 0;
305 for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
306 prot |= page_get_flags(addr1);
307 }
308 if (prot == 0)
309 break;
310 addr += qemu_host_page_size;
311 /* we found nothing */
312 if (addr == addr_start)
313 return (abi_ulong)-1;
314 }
315 if (start == 0)
316 mmap_next_start = addr + size;
317 return addr;
318 }
319
320 /* NOTE: all the constants are the HOST ones */
321 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
322 int flags, int fd, abi_ulong offset)
323 {
324 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
325 unsigned long host_start;
326
327 mmap_lock();
328 #ifdef DEBUG_MMAP
329 {
330 printf("mmap: start=0x" TARGET_FMT_lx
331 " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
332 start, len,
333 prot & PROT_READ ? 'r' : '-',
334 prot & PROT_WRITE ? 'w' : '-',
335 prot & PROT_EXEC ? 'x' : '-');
336 if (flags & MAP_FIXED)
337 printf("MAP_FIXED ");
338 if (flags & MAP_ANON)
339 printf("MAP_ANON ");
340 switch(flags & TARGET_BSD_MAP_FLAGMASK) {
341 case MAP_PRIVATE:
342 printf("MAP_PRIVATE ");
343 break;
344 case MAP_SHARED:
345 printf("MAP_SHARED ");
346 break;
347 default:
348 printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK);
349 break;
350 }
351 printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
352 }
353 #endif
354
355 if (offset & ~TARGET_PAGE_MASK) {
356 errno = EINVAL;
357 goto fail;
358 }
359
360 len = TARGET_PAGE_ALIGN(len);
361 if (len == 0)
362 goto the_end;
363 real_start = start & qemu_host_page_mask;
364
365 if (!(flags & MAP_FIXED)) {
366 abi_ulong mmap_start;
367 void *p;
368 host_offset = offset & qemu_host_page_mask;
369 host_len = len + offset - host_offset;
370 host_len = HOST_PAGE_ALIGN(host_len);
371 mmap_start = mmap_find_vma(real_start, host_len);
372 if (mmap_start == (abi_ulong)-1) {
373 errno = ENOMEM;
374 goto fail;
375 }
376 /* Note: we prefer to control the mapping address. It is
377 especially important if qemu_host_page_size >
378 qemu_real_host_page_size */
379 p = mmap(g2h(mmap_start),
380 host_len, prot, flags | MAP_FIXED, fd, host_offset);
381 if (p == MAP_FAILED)
382 goto fail;
383 /* update start so that it points to the file position at 'offset' */
384 host_start = (unsigned long)p;
385 if (!(flags & MAP_ANON))
386 host_start += offset - host_offset;
387 start = h2g(host_start);
388 } else {
389 int flg;
390 target_ulong addr;
391
392 if (start & ~TARGET_PAGE_MASK) {
393 errno = EINVAL;
394 goto fail;
395 }
396 end = start + len;
397 real_end = HOST_PAGE_ALIGN(end);
398
399 for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
400 flg = page_get_flags(addr);
401 if (flg & PAGE_RESERVED) {
402 errno = ENXIO;
403 goto fail;
404 }
405 }
406
407 /* worst case: we cannot map the file because the offset is not
408 aligned, so we read it */
409 if (!(flags & MAP_ANON) &&
410 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
411 /* msync() won't work here, so we return an error if write is
412 possible while it is a shared mapping */
413 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
414 (prot & PROT_WRITE)) {
415 errno = EINVAL;
416 goto fail;
417 }
418 retaddr = target_mmap(start, len, prot | PROT_WRITE,
419 MAP_FIXED | MAP_PRIVATE | MAP_ANON,
420 -1, 0);
421 if (retaddr == -1)
422 goto fail;
423 pread(fd, g2h(start), len, offset);
424 if (!(prot & PROT_WRITE)) {
425 ret = target_mprotect(start, len, prot);
426 if (ret != 0) {
427 start = ret;
428 goto the_end;
429 }
430 }
431 goto the_end;
432 }
433
434 /* handle the start of the mapping */
435 if (start > real_start) {
436 if (real_end == real_start + qemu_host_page_size) {
437 /* one single host page */
438 ret = mmap_frag(real_start, start, end,
439 prot, flags, fd, offset);
440 if (ret == -1)
441 goto fail;
442 goto the_end1;
443 }
444 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
445 prot, flags, fd, offset);
446 if (ret == -1)
447 goto fail;
448 real_start += qemu_host_page_size;
449 }
450 /* handle the end of the mapping */
451 if (end < real_end) {
452 ret = mmap_frag(real_end - qemu_host_page_size,
453 real_end - qemu_host_page_size, real_end,
454 prot, flags, fd,
455 offset + real_end - qemu_host_page_size - start);
456 if (ret == -1)
457 goto fail;
458 real_end -= qemu_host_page_size;
459 }
460
461 /* map the middle (easier) */
462 if (real_start < real_end) {
463 void *p;
464 unsigned long offset1;
465 if (flags & MAP_ANON)
466 offset1 = 0;
467 else
468 offset1 = offset + real_start - start;
469 p = mmap(g2h(real_start), real_end - real_start,
470 prot, flags, fd, offset1);
471 if (p == MAP_FAILED)
472 goto fail;
473 }
474 }
475 the_end1:
476 page_set_flags(start, start + len, prot | PAGE_VALID);
477 the_end:
478 #ifdef DEBUG_MMAP
479 printf("ret=0x" TARGET_FMT_lx "\n", start);
480 page_dump(stdout);
481 printf("\n");
482 #endif
483 mmap_unlock();
484 return start;
485 fail:
486 mmap_unlock();
487 return -1;
488 }
489
490 int target_munmap(abi_ulong start, abi_ulong len)
491 {
492 abi_ulong end, real_start, real_end, addr;
493 int prot, ret;
494
495 #ifdef DEBUG_MMAP
496 printf("munmap: start=0x%lx len=0x%lx\n", start, len);
497 #endif
498 if (start & ~TARGET_PAGE_MASK)
499 return -EINVAL;
500 len = TARGET_PAGE_ALIGN(len);
501 if (len == 0)
502 return -EINVAL;
503 mmap_lock();
504 end = start + len;
505 real_start = start & qemu_host_page_mask;
506 real_end = HOST_PAGE_ALIGN(end);
507
508 if (start > real_start) {
509 /* handle host page containing start */
510 prot = 0;
511 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
512 prot |= page_get_flags(addr);
513 }
514 if (real_end == real_start + qemu_host_page_size) {
515 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
516 prot |= page_get_flags(addr);
517 }
518 end = real_end;
519 }
520 if (prot != 0)
521 real_start += qemu_host_page_size;
522 }
523 if (end < real_end) {
524 prot = 0;
525 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
526 prot |= page_get_flags(addr);
527 }
528 if (prot != 0)
529 real_end -= qemu_host_page_size;
530 }
531
532 ret = 0;
533 /* unmap what we can */
534 if (real_start < real_end) {
535 ret = munmap(g2h(real_start), real_end - real_start);
536 }
537
538 if (ret == 0)
539 page_set_flags(start, start + len, 0);
540 mmap_unlock();
541 return ret;
542 }
543
544 int target_msync(abi_ulong start, abi_ulong len, int flags)
545 {
546 abi_ulong end;
547
548 if (start & ~TARGET_PAGE_MASK)
549 return -EINVAL;
550 len = TARGET_PAGE_ALIGN(len);
551 end = start + len;
552 if (end < start)
553 return -EINVAL;
554 if (end == start)
555 return 0;
556
557 start &= qemu_host_page_mask;
558 return msync(g2h(start), end - start, flags);
559 }