nbd: Don't use *_to_cpup() functions
[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 "qemu/osdep.h"
20
21 #include "qemu.h"
22 #include "qemu-common.h"
23 #include "bsd-mman.h"
24
25 //#define DEBUG_MMAP
26
27 #if defined(CONFIG_USE_NPTL)
28 pthread_mutex_t mmap_mutex;
29 static int __thread mmap_lock_count;
30
31 void mmap_lock(void)
32 {
33 if (mmap_lock_count++ == 0) {
34 pthread_mutex_lock(&mmap_mutex);
35 }
36 }
37
38 void mmap_unlock(void)
39 {
40 if (--mmap_lock_count == 0) {
41 pthread_mutex_unlock(&mmap_mutex);
42 }
43 }
44
45 /* Grab lock to make sure things are in a consistent state after fork(). */
46 void mmap_fork_start(void)
47 {
48 if (mmap_lock_count)
49 abort();
50 pthread_mutex_lock(&mmap_mutex);
51 }
52
53 void mmap_fork_end(int child)
54 {
55 if (child)
56 pthread_mutex_init(&mmap_mutex, NULL);
57 else
58 pthread_mutex_unlock(&mmap_mutex);
59 }
60 #else
61 /* We aren't threadsafe to start with, so no need to worry about locking. */
62 void mmap_lock(void)
63 {
64 }
65
66 void mmap_unlock(void)
67 {
68 }
69 #endif
70
71 /* NOTE: all the constants are the HOST ones, but addresses are target. */
72 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
73 {
74 abi_ulong end, host_start, host_end, addr;
75 int prot1, ret;
76
77 #ifdef DEBUG_MMAP
78 printf("mprotect: start=0x" TARGET_FMT_lx
79 " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
80 prot & PROT_READ ? 'r' : '-',
81 prot & PROT_WRITE ? 'w' : '-',
82 prot & PROT_EXEC ? 'x' : '-');
83 #endif
84
85 if ((start & ~TARGET_PAGE_MASK) != 0)
86 return -EINVAL;
87 len = TARGET_PAGE_ALIGN(len);
88 end = start + len;
89 if (end < start)
90 return -EINVAL;
91 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
92 if (len == 0)
93 return 0;
94
95 mmap_lock();
96 host_start = start & qemu_host_page_mask;
97 host_end = HOST_PAGE_ALIGN(end);
98 if (start > host_start) {
99 /* handle host page containing start */
100 prot1 = prot;
101 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
102 prot1 |= page_get_flags(addr);
103 }
104 if (host_end == host_start + qemu_host_page_size) {
105 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
106 prot1 |= page_get_flags(addr);
107 }
108 end = host_end;
109 }
110 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
111 if (ret != 0)
112 goto error;
113 host_start += qemu_host_page_size;
114 }
115 if (end < host_end) {
116 prot1 = prot;
117 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
118 prot1 |= page_get_flags(addr);
119 }
120 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
121 prot1 & PAGE_BITS);
122 if (ret != 0)
123 goto error;
124 host_end -= qemu_host_page_size;
125 }
126
127 /* handle the pages in the middle */
128 if (host_start < host_end) {
129 ret = mprotect(g2h(host_start), host_end - host_start, prot);
130 if (ret != 0)
131 goto error;
132 }
133 page_set_flags(start, start + len, prot | PAGE_VALID);
134 mmap_unlock();
135 return 0;
136 error:
137 mmap_unlock();
138 return ret;
139 }
140
141 /* map an incomplete host page */
142 static int mmap_frag(abi_ulong real_start,
143 abi_ulong start, abi_ulong end,
144 int prot, int flags, int fd, abi_ulong offset)
145 {
146 abi_ulong real_end, addr;
147 void *host_start;
148 int prot1, prot_new;
149
150 real_end = real_start + qemu_host_page_size;
151 host_start = g2h(real_start);
152
153 /* get the protection of the target pages outside the mapping */
154 prot1 = 0;
155 for(addr = real_start; addr < real_end; addr++) {
156 if (addr < start || addr >= end)
157 prot1 |= page_get_flags(addr);
158 }
159
160 if (prot1 == 0) {
161 /* no page was there, so we allocate one */
162 void *p = mmap(host_start, qemu_host_page_size, prot,
163 flags | MAP_ANON, -1, 0);
164 if (p == MAP_FAILED)
165 return -1;
166 prot1 = prot;
167 }
168 prot1 &= PAGE_BITS;
169
170 prot_new = prot | prot1;
171 if (!(flags & MAP_ANON)) {
172 /* msync() won't work here, so we return an error if write is
173 possible while it is a shared mapping */
174 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
175 (prot & PROT_WRITE))
176 return -1;
177
178 /* adjust protection to be able to read */
179 if (!(prot1 & PROT_WRITE))
180 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
181
182 /* read the corresponding file data */
183 pread(fd, g2h(start), end - start, offset);
184
185 /* put final protection */
186 if (prot_new != (prot1 | PROT_WRITE))
187 mprotect(host_start, qemu_host_page_size, prot_new);
188 } else {
189 /* just update the protection */
190 if (prot_new != prot1) {
191 mprotect(host_start, qemu_host_page_size, prot_new);
192 }
193 }
194 return 0;
195 }
196
197 #if defined(__CYGWIN__)
198 /* Cygwin doesn't have a whole lot of address space. */
199 static abi_ulong mmap_next_start = 0x18000000;
200 #else
201 static abi_ulong mmap_next_start = 0x40000000;
202 #endif
203
204 unsigned long last_brk;
205
206 /* find a free memory area of size 'size'. The search starts at
207 'start'. If 'start' == 0, then a default start address is used.
208 Return -1 if error.
209 */
210 /* page_init() marks pages used by the host as reserved to be sure not
211 to use them. */
212 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
213 {
214 abi_ulong addr, addr1, addr_start;
215 int prot;
216 unsigned long new_brk;
217
218 new_brk = (unsigned long)sbrk(0);
219 if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
220 /* This is a hack to catch the host allocating memory with brk().
221 If it uses mmap then we loose.
222 FIXME: We really want to avoid the host allocating memory in
223 the first place, and maybe leave some slack to avoid switching
224 to mmap. */
225 page_set_flags(last_brk & TARGET_PAGE_MASK,
226 TARGET_PAGE_ALIGN(new_brk),
227 PAGE_RESERVED);
228 }
229 last_brk = new_brk;
230
231 size = HOST_PAGE_ALIGN(size);
232 start = start & qemu_host_page_mask;
233 addr = start;
234 if (addr == 0)
235 addr = mmap_next_start;
236 addr_start = addr;
237 for(;;) {
238 prot = 0;
239 for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
240 prot |= page_get_flags(addr1);
241 }
242 if (prot == 0)
243 break;
244 addr += qemu_host_page_size;
245 /* we found nothing */
246 if (addr == addr_start)
247 return (abi_ulong)-1;
248 }
249 if (start == 0)
250 mmap_next_start = addr + size;
251 return addr;
252 }
253
254 /* NOTE: all the constants are the HOST ones */
255 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
256 int flags, int fd, abi_ulong offset)
257 {
258 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
259 unsigned long host_start;
260
261 mmap_lock();
262 #ifdef DEBUG_MMAP
263 {
264 printf("mmap: start=0x" TARGET_FMT_lx
265 " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
266 start, len,
267 prot & PROT_READ ? 'r' : '-',
268 prot & PROT_WRITE ? 'w' : '-',
269 prot & PROT_EXEC ? 'x' : '-');
270 if (flags & MAP_FIXED)
271 printf("MAP_FIXED ");
272 if (flags & MAP_ANON)
273 printf("MAP_ANON ");
274 switch(flags & TARGET_BSD_MAP_FLAGMASK) {
275 case MAP_PRIVATE:
276 printf("MAP_PRIVATE ");
277 break;
278 case MAP_SHARED:
279 printf("MAP_SHARED ");
280 break;
281 default:
282 printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK);
283 break;
284 }
285 printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
286 }
287 #endif
288
289 if (offset & ~TARGET_PAGE_MASK) {
290 errno = EINVAL;
291 goto fail;
292 }
293
294 len = TARGET_PAGE_ALIGN(len);
295 if (len == 0)
296 goto the_end;
297 real_start = start & qemu_host_page_mask;
298
299 if (!(flags & MAP_FIXED)) {
300 abi_ulong mmap_start;
301 void *p;
302 host_offset = offset & qemu_host_page_mask;
303 host_len = len + offset - host_offset;
304 host_len = HOST_PAGE_ALIGN(host_len);
305 mmap_start = mmap_find_vma(real_start, host_len);
306 if (mmap_start == (abi_ulong)-1) {
307 errno = ENOMEM;
308 goto fail;
309 }
310 /* Note: we prefer to control the mapping address. It is
311 especially important if qemu_host_page_size >
312 qemu_real_host_page_size */
313 p = mmap(g2h(mmap_start),
314 host_len, prot, flags | MAP_FIXED, fd, host_offset);
315 if (p == MAP_FAILED)
316 goto fail;
317 /* update start so that it points to the file position at 'offset' */
318 host_start = (unsigned long)p;
319 if (!(flags & MAP_ANON))
320 host_start += offset - host_offset;
321 start = h2g(host_start);
322 } else {
323 int flg;
324 target_ulong addr;
325
326 if (start & ~TARGET_PAGE_MASK) {
327 errno = EINVAL;
328 goto fail;
329 }
330 end = start + len;
331 real_end = HOST_PAGE_ALIGN(end);
332
333 for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
334 flg = page_get_flags(addr);
335 if (flg & PAGE_RESERVED) {
336 errno = ENXIO;
337 goto fail;
338 }
339 }
340
341 /* worst case: we cannot map the file because the offset is not
342 aligned, so we read it */
343 if (!(flags & MAP_ANON) &&
344 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
345 /* msync() won't work here, so we return an error if write is
346 possible while it is a shared mapping */
347 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
348 (prot & PROT_WRITE)) {
349 errno = EINVAL;
350 goto fail;
351 }
352 retaddr = target_mmap(start, len, prot | PROT_WRITE,
353 MAP_FIXED | MAP_PRIVATE | MAP_ANON,
354 -1, 0);
355 if (retaddr == -1)
356 goto fail;
357 pread(fd, g2h(start), len, offset);
358 if (!(prot & PROT_WRITE)) {
359 ret = target_mprotect(start, len, prot);
360 if (ret != 0) {
361 start = ret;
362 goto the_end;
363 }
364 }
365 goto the_end;
366 }
367
368 /* handle the start of the mapping */
369 if (start > real_start) {
370 if (real_end == real_start + qemu_host_page_size) {
371 /* one single host page */
372 ret = mmap_frag(real_start, start, end,
373 prot, flags, fd, offset);
374 if (ret == -1)
375 goto fail;
376 goto the_end1;
377 }
378 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
379 prot, flags, fd, offset);
380 if (ret == -1)
381 goto fail;
382 real_start += qemu_host_page_size;
383 }
384 /* handle the end of the mapping */
385 if (end < real_end) {
386 ret = mmap_frag(real_end - qemu_host_page_size,
387 real_end - qemu_host_page_size, real_end,
388 prot, flags, fd,
389 offset + real_end - qemu_host_page_size - start);
390 if (ret == -1)
391 goto fail;
392 real_end -= qemu_host_page_size;
393 }
394
395 /* map the middle (easier) */
396 if (real_start < real_end) {
397 void *p;
398 unsigned long offset1;
399 if (flags & MAP_ANON)
400 offset1 = 0;
401 else
402 offset1 = offset + real_start - start;
403 p = mmap(g2h(real_start), real_end - real_start,
404 prot, flags, fd, offset1);
405 if (p == MAP_FAILED)
406 goto fail;
407 }
408 }
409 the_end1:
410 page_set_flags(start, start + len, prot | PAGE_VALID);
411 the_end:
412 #ifdef DEBUG_MMAP
413 printf("ret=0x" TARGET_FMT_lx "\n", start);
414 page_dump(stdout);
415 printf("\n");
416 #endif
417 mmap_unlock();
418 return start;
419 fail:
420 mmap_unlock();
421 return -1;
422 }
423
424 int target_munmap(abi_ulong start, abi_ulong len)
425 {
426 abi_ulong end, real_start, real_end, addr;
427 int prot, ret;
428
429 #ifdef DEBUG_MMAP
430 printf("munmap: start=0x%lx len=0x%lx\n", start, len);
431 #endif
432 if (start & ~TARGET_PAGE_MASK)
433 return -EINVAL;
434 len = TARGET_PAGE_ALIGN(len);
435 if (len == 0)
436 return -EINVAL;
437 mmap_lock();
438 end = start + len;
439 real_start = start & qemu_host_page_mask;
440 real_end = HOST_PAGE_ALIGN(end);
441
442 if (start > real_start) {
443 /* handle host page containing start */
444 prot = 0;
445 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
446 prot |= page_get_flags(addr);
447 }
448 if (real_end == real_start + qemu_host_page_size) {
449 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
450 prot |= page_get_flags(addr);
451 }
452 end = real_end;
453 }
454 if (prot != 0)
455 real_start += qemu_host_page_size;
456 }
457 if (end < real_end) {
458 prot = 0;
459 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
460 prot |= page_get_flags(addr);
461 }
462 if (prot != 0)
463 real_end -= qemu_host_page_size;
464 }
465
466 ret = 0;
467 /* unmap what we can */
468 if (real_start < real_end) {
469 ret = munmap(g2h(real_start), real_end - real_start);
470 }
471
472 if (ret == 0)
473 page_set_flags(start, start + len, 0);
474 mmap_unlock();
475 return ret;
476 }
477
478 int target_msync(abi_ulong start, abi_ulong len, int flags)
479 {
480 abi_ulong end;
481
482 if (start & ~TARGET_PAGE_MASK)
483 return -EINVAL;
484 len = TARGET_PAGE_ALIGN(len);
485 end = start + len;
486 if (end < start)
487 return -EINVAL;
488 if (end == start)
489 return 0;
490
491 start &= qemu_host_page_mask;
492 return msync(g2h(start), end - start, flags);
493 }