]>
Commit | Line | Data |
---|---|---|
54936004 FB |
1 | /* |
2 | * mmap support for qemu | |
5fafdf24 | 3 | * |
54936004 FB |
4 | * Copyright (c) 2003 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/>. |
54936004 | 18 | */ |
d39594e9 | 19 | #include "qemu/osdep.h" |
11d96056 | 20 | #include "trace.h" |
10d0d505 | 21 | #include "exec/log.h" |
54936004 | 22 | #include "qemu.h" |
3b249d26 | 23 | #include "user-internals.h" |
5423e6d3 | 24 | #include "user-mmap.h" |
8655b4c7 | 25 | #include "target_mman.h" |
54936004 | 26 | |
1e6eec8b | 27 | static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER; |
dfd3f85c | 28 | static __thread int mmap_lock_count; |
c8a706fe PB |
29 | |
30 | void mmap_lock(void) | |
31 | { | |
32 | if (mmap_lock_count++ == 0) { | |
33 | pthread_mutex_lock(&mmap_mutex); | |
34 | } | |
35 | } | |
36 | ||
37 | void mmap_unlock(void) | |
38 | { | |
39 | if (--mmap_lock_count == 0) { | |
40 | pthread_mutex_unlock(&mmap_mutex); | |
41 | } | |
42 | } | |
d5975363 | 43 | |
301e40ed AB |
44 | bool have_mmap_lock(void) |
45 | { | |
46 | return mmap_lock_count > 0 ? true : false; | |
47 | } | |
48 | ||
d5975363 PB |
49 | /* Grab lock to make sure things are in a consistent state after fork(). */ |
50 | void mmap_fork_start(void) | |
51 | { | |
52 | if (mmap_lock_count) | |
53 | abort(); | |
54 | pthread_mutex_lock(&mmap_mutex); | |
55 | } | |
56 | ||
57 | void mmap_fork_end(int child) | |
58 | { | |
2b730f79 | 59 | if (child) { |
d5975363 | 60 | pthread_mutex_init(&mmap_mutex, NULL); |
2b730f79 | 61 | } else { |
d5975363 | 62 | pthread_mutex_unlock(&mmap_mutex); |
2b730f79 | 63 | } |
d5975363 | 64 | } |
c8a706fe | 65 | |
9dba3ca5 RH |
66 | /* |
67 | * Validate target prot bitmask. | |
68 | * Return the prot bitmask for the host in *HOST_PROT. | |
69 | * Return 0 if the target prot bitmask is invalid, otherwise | |
70 | * the internal qemu page_flags (which will include PAGE_VALID). | |
71 | */ | |
0dd55812 | 72 | static int validate_prot_to_pageflags(int prot) |
9dba3ca5 RH |
73 | { |
74 | int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM; | |
75 | int page_flags = (prot & PAGE_BITS) | PAGE_VALID; | |
76 | ||
be5d6f48 | 77 | #ifdef TARGET_AARCH64 |
d109b46d | 78 | { |
be5d6f48 | 79 | ARMCPU *cpu = ARM_CPU(thread_cpu); |
d109b46d RH |
80 | |
81 | /* | |
82 | * The PROT_BTI bit is only accepted if the cpu supports the feature. | |
83 | * Since this is the unusual case, don't bother checking unless | |
84 | * the bit has been requested. If set and valid, record the bit | |
85 | * within QEMU's page_flags. | |
86 | */ | |
87 | if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) { | |
be5d6f48 RH |
88 | valid |= TARGET_PROT_BTI; |
89 | page_flags |= PAGE_BTI; | |
90 | } | |
d109b46d RH |
91 | /* Similarly for the PROT_MTE bit. */ |
92 | if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) { | |
93 | valid |= TARGET_PROT_MTE; | |
94 | page_flags |= PAGE_MTE; | |
95 | } | |
be5d6f48 | 96 | } |
4c184e70 HD |
97 | #elif defined(TARGET_HPPA) |
98 | valid |= PROT_GROWSDOWN | PROT_GROWSUP; | |
be5d6f48 RH |
99 | #endif |
100 | ||
9dba3ca5 RH |
101 | return prot & ~valid ? 0 : page_flags; |
102 | } | |
103 | ||
0dd55812 RH |
104 | /* |
105 | * For the host, we need not pass anything except read/write/exec. | |
106 | * While PROT_SEM is allowed by all hosts, it is also ignored, so | |
107 | * don't bother transforming guest bit to host bit. Any other | |
108 | * target-specific prot bits will not be understood by the host | |
109 | * and will need to be encoded into page_flags for qemu emulation. | |
110 | * | |
111 | * Pages that are executable by the guest will never be executed | |
112 | * by the host, but the host will need to be able to read them. | |
113 | */ | |
114 | static int target_to_host_prot(int prot) | |
115 | { | |
116 | return (prot & (PROT_READ | PROT_WRITE)) | | |
117 | (prot & PROT_EXEC ? PROT_READ : 0); | |
118 | } | |
119 | ||
53a5960a | 120 | /* NOTE: all the constants are the HOST ones, but addresses are target. */ |
9dba3ca5 | 121 | int target_mprotect(abi_ulong start, abi_ulong len, int target_prot) |
54936004 | 122 | { |
7bdc1acc RH |
123 | abi_ulong starts[3]; |
124 | abi_ulong lens[3]; | |
125 | int prots[3]; | |
126 | abi_ulong host_start, host_last, last; | |
127 | int prot1, ret, page_flags, nranges; | |
54936004 | 128 | |
9dba3ca5 | 129 | trace_target_mprotect(start, len, target_prot); |
54936004 | 130 | |
9dba3ca5 | 131 | if ((start & ~TARGET_PAGE_MASK) != 0) { |
78cf3390 | 132 | return -TARGET_EINVAL; |
9dba3ca5 | 133 | } |
0dd55812 | 134 | page_flags = validate_prot_to_pageflags(target_prot); |
9dba3ca5 RH |
135 | if (!page_flags) { |
136 | return -TARGET_EINVAL; | |
137 | } | |
7bdc1acc RH |
138 | if (len == 0) { |
139 | return 0; | |
140 | } | |
54936004 | 141 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 142 | if (!guest_range_valid_untagged(start, len)) { |
78cf3390 | 143 | return -TARGET_ENOMEM; |
ebf9a363 | 144 | } |
3b46e624 | 145 | |
7bdc1acc | 146 | last = start + len - 1; |
83fb7adf | 147 | host_start = start & qemu_host_page_mask; |
7bdc1acc RH |
148 | host_last = HOST_PAGE_ALIGN(last) - 1; |
149 | nranges = 0; | |
150 | ||
151 | mmap_lock(); | |
152 | ||
153 | if (host_last - host_start < qemu_host_page_size) { | |
154 | /* Single host page contains all guest pages: sum the prot. */ | |
0dd55812 | 155 | prot1 = target_prot; |
7bdc1acc RH |
156 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { |
157 | prot1 |= page_get_flags(a); | |
54936004 | 158 | } |
7bdc1acc RH |
159 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { |
160 | prot1 |= page_get_flags(a + 1); | |
d418c81e | 161 | } |
7bdc1acc RH |
162 | starts[nranges] = host_start; |
163 | lens[nranges] = qemu_host_page_size; | |
164 | prots[nranges] = prot1; | |
165 | nranges++; | |
166 | } else { | |
167 | if (host_start < start) { | |
168 | /* Host page contains more than one guest page: sum the prot. */ | |
169 | prot1 = target_prot; | |
170 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { | |
171 | prot1 |= page_get_flags(a); | |
172 | } | |
173 | /* If the resulting sum differs, create a new range. */ | |
174 | if (prot1 != target_prot) { | |
175 | starts[nranges] = host_start; | |
176 | lens[nranges] = qemu_host_page_size; | |
177 | prots[nranges] = prot1; | |
178 | nranges++; | |
179 | host_start += qemu_host_page_size; | |
180 | } | |
9dba3ca5 | 181 | } |
7bdc1acc RH |
182 | |
183 | if (last < host_last) { | |
184 | /* Host page contains more than one guest page: sum the prot. */ | |
185 | prot1 = target_prot; | |
186 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { | |
187 | prot1 |= page_get_flags(a + 1); | |
188 | } | |
189 | /* If the resulting sum differs, create a new range. */ | |
190 | if (prot1 != target_prot) { | |
191 | host_last -= qemu_host_page_size; | |
192 | starts[nranges] = host_last + 1; | |
193 | lens[nranges] = qemu_host_page_size; | |
194 | prots[nranges] = prot1; | |
195 | nranges++; | |
196 | } | |
54936004 | 197 | } |
7bdc1acc RH |
198 | |
199 | /* Create a range for the middle, if any remains. */ | |
200 | if (host_start < host_last) { | |
201 | starts[nranges] = host_start; | |
202 | lens[nranges] = host_last - host_start + 1; | |
203 | prots[nranges] = target_prot; | |
204 | nranges++; | |
9dba3ca5 | 205 | } |
54936004 | 206 | } |
3b46e624 | 207 | |
7bdc1acc RH |
208 | for (int i = 0; i < nranges; ++i) { |
209 | ret = mprotect(g2h_untagged(starts[i]), lens[i], | |
210 | target_to_host_prot(prots[i])); | |
9dba3ca5 | 211 | if (ret != 0) { |
c8a706fe | 212 | goto error; |
9dba3ca5 | 213 | } |
54936004 | 214 | } |
aa98e2d8 | 215 | |
7bdc1acc | 216 | page_set_flags(start, last, page_flags); |
aa98e2d8 IL |
217 | ret = 0; |
218 | ||
7bdc1acc | 219 | error: |
c8a706fe PB |
220 | mmap_unlock(); |
221 | return ret; | |
54936004 FB |
222 | } |
223 | ||
224 | /* map an incomplete host page */ | |
99982beb RH |
225 | static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last, |
226 | int prot, int flags, int fd, off_t offset) | |
54936004 | 227 | { |
99982beb | 228 | abi_ulong real_last; |
53a5960a | 229 | void *host_start; |
99982beb RH |
230 | int prot_old, prot_new; |
231 | int host_prot_old, host_prot_new; | |
54936004 | 232 | |
99982beb RH |
233 | if (!(flags & MAP_ANONYMOUS) |
234 | && (flags & MAP_TYPE) == MAP_SHARED | |
235 | && (prot & PROT_WRITE)) { | |
236 | /* | |
237 | * msync() won't work with the partial page, so we return an | |
238 | * error if write is possible while it is a shared mapping. | |
239 | */ | |
240 | errno = EINVAL; | |
241 | return false; | |
242 | } | |
243 | ||
244 | real_last = real_start + qemu_host_page_size - 1; | |
3e8f1628 | 245 | host_start = g2h_untagged(real_start); |
54936004 | 246 | |
99982beb RH |
247 | /* Get the protection of the target pages outside the mapping. */ |
248 | prot_old = 0; | |
249 | for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
250 | prot_old |= page_get_flags(a); | |
251 | } | |
252 | for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) { | |
253 | prot_old |= page_get_flags(a); | |
54936004 | 254 | } |
3b46e624 | 255 | |
99982beb RH |
256 | if (prot_old == 0) { |
257 | /* | |
258 | * Since !(prot_old & PAGE_VALID), there were no guest pages | |
259 | * outside of the fragment we need to map. Allocate a new host | |
260 | * page to cover, discarding whatever else may have been present. | |
261 | */ | |
0dd55812 RH |
262 | void *p = mmap(host_start, qemu_host_page_size, |
263 | target_to_host_prot(prot), | |
80210bcd | 264 | flags | MAP_ANONYMOUS, -1, 0); |
2b730f79 | 265 | if (p == MAP_FAILED) { |
99982beb | 266 | return false; |
2b730f79 | 267 | } |
99982beb | 268 | prot_old = prot; |
54936004 | 269 | } |
99982beb | 270 | prot_new = prot | prot_old; |
54936004 | 271 | |
99982beb RH |
272 | host_prot_old = target_to_host_prot(prot_old); |
273 | host_prot_new = target_to_host_prot(prot_new); | |
3b46e624 | 274 | |
99982beb RH |
275 | /* Adjust protection to be able to write. */ |
276 | if (!(host_prot_old & PROT_WRITE)) { | |
277 | host_prot_old |= PROT_WRITE; | |
278 | mprotect(host_start, qemu_host_page_size, host_prot_old); | |
279 | } | |
3b46e624 | 280 | |
99982beb RH |
281 | /* Read or zero the new guest pages. */ |
282 | if (flags & MAP_ANONYMOUS) { | |
283 | memset(g2h_untagged(start), 0, last - start + 1); | |
54936004 | 284 | } else { |
99982beb RH |
285 | if (pread(fd, g2h_untagged(start), last - start + 1, offset) == -1) { |
286 | return false; | |
e6deac9c | 287 | } |
54936004 | 288 | } |
99982beb RH |
289 | |
290 | /* Put final protection */ | |
291 | if (host_prot_new != host_prot_old) { | |
292 | mprotect(host_start, qemu_host_page_size, host_prot_new); | |
293 | } | |
294 | return true; | |
54936004 FB |
295 | } |
296 | ||
14f24e14 | 297 | #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64 |
aab613fb LY |
298 | #ifdef TARGET_AARCH64 |
299 | # define TASK_UNMAPPED_BASE 0x5500000000 | |
300 | #else | |
14f24e14 | 301 | # define TASK_UNMAPPED_BASE (1ul << 38) |
aab613fb | 302 | #endif |
a03e2d42 | 303 | #else |
9c9b5d7b HD |
304 | #ifdef TARGET_HPPA |
305 | # define TASK_UNMAPPED_BASE 0xfa000000 | |
306 | #else | |
14f24e14 | 307 | # define TASK_UNMAPPED_BASE 0x40000000 |
a03e2d42 | 308 | #endif |
9c9b5d7b | 309 | #endif |
59e9d91c | 310 | abi_ulong mmap_next_start = TASK_UNMAPPED_BASE; |
a03e2d42 | 311 | |
0776590d PB |
312 | unsigned long last_brk; |
313 | ||
2b730f79 RH |
314 | /* |
315 | * Subroutine of mmap_find_vma, used when we have pre-allocated | |
316 | * a chunk of guest address space. | |
317 | */ | |
30ab9ef2 RH |
318 | static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, |
319 | abi_ulong align) | |
68a1c816 | 320 | { |
4c13048e | 321 | target_ulong ret; |
68a1c816 | 322 | |
4c13048e RH |
323 | ret = page_find_range_empty(start, reserved_va, size, align); |
324 | if (ret == -1 && start > mmap_min_addr) { | |
325 | /* Restart at the beginning of the address space. */ | |
326 | ret = page_find_range_empty(mmap_min_addr, start - 1, size, align); | |
68a1c816 PB |
327 | } |
328 | ||
4c13048e | 329 | return ret; |
68a1c816 PB |
330 | } |
331 | ||
fe3b4152 KS |
332 | /* |
333 | * Find and reserve a free memory area of size 'size'. The search | |
334 | * starts at 'start'. | |
335 | * It must be called with mmap_lock() held. | |
336 | * Return -1 if error. | |
337 | */ | |
30ab9ef2 | 338 | abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align) |
a03e2d42 | 339 | { |
14f24e14 | 340 | void *ptr, *prev; |
fe3b4152 | 341 | abi_ulong addr; |
14f24e14 | 342 | int wrapped, repeat; |
fe3b4152 | 343 | |
443b7505 RH |
344 | align = MAX(align, qemu_host_page_size); |
345 | ||
fe3b4152 | 346 | /* If 'start' == 0, then a default start address is used. */ |
14f24e14 | 347 | if (start == 0) { |
fe3b4152 | 348 | start = mmap_next_start; |
14f24e14 RH |
349 | } else { |
350 | start &= qemu_host_page_mask; | |
351 | } | |
30ab9ef2 | 352 | start = ROUND_UP(start, align); |
14f24e14 RH |
353 | |
354 | size = HOST_PAGE_ALIGN(size); | |
fe3b4152 | 355 | |
b76f21a7 | 356 | if (reserved_va) { |
30ab9ef2 | 357 | return mmap_find_vma_reserved(start, size, align); |
68a1c816 PB |
358 | } |
359 | ||
a03e2d42 | 360 | addr = start; |
14f24e14 RH |
361 | wrapped = repeat = 0; |
362 | prev = 0; | |
fe3b4152 | 363 | |
14f24e14 | 364 | for (;; prev = ptr) { |
fe3b4152 KS |
365 | /* |
366 | * Reserve needed memory area to avoid a race. | |
367 | * It should be discarded using: | |
368 | * - mmap() with MAP_FIXED flag | |
369 | * - mremap() with MREMAP_FIXED flag | |
370 | * - shmat() with SHM_REMAP flag | |
371 | */ | |
3e8f1628 | 372 | ptr = mmap(g2h_untagged(addr), size, PROT_NONE, |
2b730f79 | 373 | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0); |
fe3b4152 KS |
374 | |
375 | /* ENOMEM, if host address space has no memory */ | |
14f24e14 | 376 | if (ptr == MAP_FAILED) { |
fe3b4152 | 377 | return (abi_ulong)-1; |
14f24e14 RH |
378 | } |
379 | ||
2b730f79 RH |
380 | /* |
381 | * Count the number of sequential returns of the same address. | |
382 | * This is used to modify the search algorithm below. | |
383 | */ | |
14f24e14 RH |
384 | repeat = (ptr == prev ? repeat + 1 : 0); |
385 | ||
386 | if (h2g_valid(ptr + size - 1)) { | |
387 | addr = h2g(ptr); | |
fe3b4152 | 388 | |
30ab9ef2 | 389 | if ((addr & (align - 1)) == 0) { |
14f24e14 RH |
390 | /* Success. */ |
391 | if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) { | |
392 | mmap_next_start = addr + size; | |
393 | } | |
394 | return addr; | |
395 | } | |
fe3b4152 | 396 | |
14f24e14 RH |
397 | /* The address is not properly aligned for the target. */ |
398 | switch (repeat) { | |
399 | case 0: | |
2b730f79 RH |
400 | /* |
401 | * Assume the result that the kernel gave us is the | |
402 | * first with enough free space, so start again at the | |
403 | * next higher target page. | |
404 | */ | |
30ab9ef2 | 405 | addr = ROUND_UP(addr, align); |
14f24e14 RH |
406 | break; |
407 | case 1: | |
2b730f79 RH |
408 | /* |
409 | * Sometimes the kernel decides to perform the allocation | |
410 | * at the top end of memory instead. | |
411 | */ | |
30ab9ef2 | 412 | addr &= -align; |
14f24e14 RH |
413 | break; |
414 | case 2: | |
415 | /* Start over at low memory. */ | |
416 | addr = 0; | |
417 | break; | |
418 | default: | |
419 | /* Fail. This unaligned block must the last. */ | |
420 | addr = -1; | |
421 | break; | |
422 | } | |
423 | } else { | |
2b730f79 RH |
424 | /* |
425 | * Since the result the kernel gave didn't fit, start | |
426 | * again at low memory. If any repetition, fail. | |
427 | */ | |
14f24e14 RH |
428 | addr = (repeat ? -1 : 0); |
429 | } | |
430 | ||
431 | /* Unmap and try again. */ | |
fe3b4152 | 432 | munmap(ptr, size); |
fe3b4152 | 433 | |
14f24e14 | 434 | /* ENOMEM if we checked the whole of the target address space. */ |
d0b3e4f5 | 435 | if (addr == (abi_ulong)-1) { |
a03e2d42 | 436 | return (abi_ulong)-1; |
14f24e14 RH |
437 | } else if (addr == 0) { |
438 | if (wrapped) { | |
439 | return (abi_ulong)-1; | |
440 | } | |
441 | wrapped = 1; | |
2b730f79 RH |
442 | /* |
443 | * Don't actually use 0 when wrapping, instead indicate | |
444 | * that we'd truly like an allocation in low memory. | |
445 | */ | |
14f24e14 RH |
446 | addr = (mmap_min_addr > TARGET_PAGE_SIZE |
447 | ? TARGET_PAGE_ALIGN(mmap_min_addr) | |
448 | : TARGET_PAGE_SIZE); | |
449 | } else if (wrapped && addr >= start) { | |
450 | return (abi_ulong)-1; | |
451 | } | |
a03e2d42 | 452 | } |
a03e2d42 FB |
453 | } |
454 | ||
54936004 | 455 | /* NOTE: all the constants are the HOST ones */ |
9dba3ca5 | 456 | abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot, |
55baec0f | 457 | int flags, int fd, off_t offset) |
54936004 | 458 | { |
f9cd8f5e RH |
459 | abi_ulong ret, last, real_start, real_last, retaddr, host_len; |
460 | abi_ulong passthrough_start = -1, passthrough_last = 0; | |
0dd55812 | 461 | int page_flags; |
55baec0f | 462 | off_t host_offset; |
54936004 | 463 | |
c8a706fe | 464 | mmap_lock(); |
9dba3ca5 | 465 | trace_target_mmap(start, len, target_prot, flags, fd, offset); |
54936004 | 466 | |
38138fab | 467 | if (!len) { |
e89f07d3 | 468 | errno = EINVAL; |
c8a706fe | 469 | goto fail; |
e89f07d3 | 470 | } |
54936004 | 471 | |
0dd55812 | 472 | page_flags = validate_prot_to_pageflags(target_prot); |
9dba3ca5 RH |
473 | if (!page_flags) { |
474 | errno = EINVAL; | |
475 | goto fail; | |
476 | } | |
477 | ||
38138fab | 478 | /* Also check for overflows... */ |
54936004 | 479 | len = TARGET_PAGE_ALIGN(len); |
38138fab AB |
480 | if (!len) { |
481 | errno = ENOMEM; | |
482 | goto fail; | |
483 | } | |
484 | ||
485 | if (offset & ~TARGET_PAGE_MASK) { | |
486 | errno = EINVAL; | |
487 | goto fail; | |
488 | } | |
489 | ||
228168cb RH |
490 | /* |
491 | * If we're mapping shared memory, ensure we generate code for parallel | |
492 | * execution and flush old translations. This will work up to the level | |
493 | * supported by the host -- anything that requires EXCP_ATOMIC will not | |
494 | * be atomic with respect to an external process. | |
495 | */ | |
496 | if (flags & MAP_SHARED) { | |
497 | CPUState *cpu = thread_cpu; | |
498 | if (!(cpu->tcg_cflags & CF_PARALLEL)) { | |
499 | cpu->tcg_cflags |= CF_PARALLEL; | |
500 | tb_flush(cpu); | |
501 | } | |
502 | } | |
503 | ||
53a5960a | 504 | real_start = start & qemu_host_page_mask; |
a5e7ee46 RH |
505 | host_offset = offset & qemu_host_page_mask; |
506 | ||
2b730f79 RH |
507 | /* |
508 | * If the user is asking for the kernel to find a location, do that | |
509 | * before we truncate the length for mapping files below. | |
510 | */ | |
03798605 | 511 | if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
a5e7ee46 RH |
512 | host_len = len + offset - host_offset; |
513 | host_len = HOST_PAGE_ALIGN(host_len); | |
30ab9ef2 | 514 | start = mmap_find_vma(real_start, host_len, TARGET_PAGE_SIZE); |
a5e7ee46 RH |
515 | if (start == (abi_ulong)-1) { |
516 | errno = ENOMEM; | |
517 | goto fail; | |
518 | } | |
519 | } | |
54936004 | 520 | |
2b730f79 RH |
521 | /* |
522 | * When mapping files into a memory area larger than the file, accesses | |
523 | * to pages beyond the file size will cause a SIGBUS. | |
524 | * | |
525 | * For example, if mmaping a file of 100 bytes on a host with 4K pages | |
526 | * emulating a target with 8K pages, the target expects to be able to | |
527 | * access the first 8K. But the host will trap us on any access beyond | |
528 | * 4K. | |
529 | * | |
530 | * When emulating a target with a larger page-size than the hosts, we | |
531 | * may need to truncate file maps at EOF and add extra anonymous pages | |
532 | * up to the targets page boundary. | |
533 | */ | |
8e3b0cbb | 534 | if ((qemu_real_host_page_size() < qemu_host_page_size) && |
35f2fd04 MAL |
535 | !(flags & MAP_ANONYMOUS)) { |
536 | struct stat sb; | |
54c5a2ae | 537 | |
2b730f79 RH |
538 | if (fstat(fd, &sb) == -1) { |
539 | goto fail; | |
540 | } | |
54c5a2ae | 541 | |
2b730f79 RH |
542 | /* Are we trying to create a map beyond EOF?. */ |
543 | if (offset + len > sb.st_size) { | |
544 | /* | |
545 | * If so, truncate the file map at eof aligned with | |
546 | * the hosts real pagesize. Additional anonymous maps | |
547 | * will be created beyond EOF. | |
548 | */ | |
549 | len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset); | |
550 | } | |
54c5a2ae EI |
551 | } |
552 | ||
03798605 | 553 | if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
55baec0f | 554 | uintptr_t host_start; |
0dd55812 | 555 | int host_prot; |
a03e2d42 | 556 | void *p; |
a5e7ee46 | 557 | |
a03e2d42 FB |
558 | host_len = len + offset - host_offset; |
559 | host_len = HOST_PAGE_ALIGN(host_len); | |
0dd55812 | 560 | host_prot = target_to_host_prot(target_prot); |
a5e7ee46 | 561 | |
2b730f79 RH |
562 | /* |
563 | * Note: we prefer to control the mapping address. It is | |
564 | * especially important if qemu_host_page_size > | |
565 | * qemu_real_host_page_size. | |
566 | */ | |
3e8f1628 | 567 | p = mmap(g2h_untagged(start), host_len, host_prot, |
a5e7ee46 | 568 | flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0); |
9dba3ca5 | 569 | if (p == MAP_FAILED) { |
c8a706fe | 570 | goto fail; |
9dba3ca5 | 571 | } |
a03e2d42 | 572 | /* update start so that it points to the file position at 'offset' */ |
55baec0f | 573 | host_start = (uintptr_t)p; |
54c5a2ae | 574 | if (!(flags & MAP_ANONYMOUS)) { |
3e8f1628 | 575 | p = mmap(g2h_untagged(start), len, host_prot, |
54c5a2ae | 576 | flags | MAP_FIXED, fd, host_offset); |
8384274e | 577 | if (p == MAP_FAILED) { |
3e8f1628 | 578 | munmap(g2h_untagged(start), host_len); |
8384274e JB |
579 | goto fail; |
580 | } | |
a03e2d42 | 581 | host_start += offset - host_offset; |
54c5a2ae | 582 | } |
a03e2d42 | 583 | start = h2g(host_start); |
f9cd8f5e | 584 | last = start + len - 1; |
f93b7695 | 585 | passthrough_start = start; |
f9cd8f5e | 586 | passthrough_last = last; |
a03e2d42 FB |
587 | } else { |
588 | if (start & ~TARGET_PAGE_MASK) { | |
e89f07d3 | 589 | errno = EINVAL; |
c8a706fe | 590 | goto fail; |
e89f07d3 | 591 | } |
f9cd8f5e RH |
592 | last = start + len - 1; |
593 | real_last = HOST_PAGE_ALIGN(last) - 1; | |
7ab240ad | 594 | |
7d37435b PB |
595 | /* |
596 | * Test if requested memory area fits target address space | |
597 | * It can fail only on 64-bit host with 32-bit target. | |
598 | * On any other target/host host mmap() handles this error correctly. | |
599 | */ | |
f9cd8f5e | 600 | if (last < start || !guest_range_valid_untagged(start, len)) { |
ebf9a363 | 601 | errno = ENOMEM; |
45bc1f52 AJ |
602 | goto fail; |
603 | } | |
604 | ||
03798605 RH |
605 | /* Validate that the chosen range is empty. */ |
606 | if ((flags & MAP_FIXED_NOREPLACE) | |
f9cd8f5e | 607 | && !page_check_range_empty(start, last)) { |
03798605 RH |
608 | errno = EEXIST; |
609 | goto fail; | |
610 | } | |
611 | ||
2b730f79 RH |
612 | /* |
613 | * worst case: we cannot map the file because the offset is not | |
614 | * aligned, so we read it | |
615 | */ | |
a03e2d42 FB |
616 | if (!(flags & MAP_ANONYMOUS) && |
617 | (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) { | |
2b730f79 RH |
618 | /* |
619 | * msync() won't work here, so we return an error if write is | |
620 | * possible while it is a shared mapping | |
621 | */ | |
0dd55812 RH |
622 | if ((flags & MAP_TYPE) == MAP_SHARED |
623 | && (target_prot & PROT_WRITE)) { | |
a03e2d42 | 624 | errno = EINVAL; |
c8a706fe | 625 | goto fail; |
a03e2d42 | 626 | } |
9dba3ca5 | 627 | retaddr = target_mmap(start, len, target_prot | PROT_WRITE, |
03798605 RH |
628 | (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) |
629 | | MAP_PRIVATE | MAP_ANONYMOUS, | |
a03e2d42 | 630 | -1, 0); |
2b730f79 | 631 | if (retaddr == -1) { |
c8a706fe | 632 | goto fail; |
2b730f79 RH |
633 | } |
634 | if (pread(fd, g2h_untagged(start), len, offset) == -1) { | |
fb7e378c | 635 | goto fail; |
2b730f79 | 636 | } |
0dd55812 | 637 | if (!(target_prot & PROT_WRITE)) { |
9dba3ca5 | 638 | ret = target_mprotect(start, len, target_prot); |
86abac06 | 639 | assert(ret == 0); |
a03e2d42 FB |
640 | } |
641 | goto the_end; | |
54936004 | 642 | } |
2b730f79 | 643 | |
a03e2d42 FB |
644 | /* handle the start of the mapping */ |
645 | if (start > real_start) { | |
f9cd8f5e | 646 | if (real_last == real_start + qemu_host_page_size - 1) { |
a03e2d42 | 647 | /* one single host page */ |
f9cd8f5e | 648 | if (!mmap_frag(real_start, start, last, |
99982beb | 649 | target_prot, flags, fd, offset)) { |
c8a706fe | 650 | goto fail; |
2b730f79 | 651 | } |
a03e2d42 FB |
652 | goto the_end1; |
653 | } | |
99982beb RH |
654 | if (!mmap_frag(real_start, start, |
655 | real_start + qemu_host_page_size - 1, | |
656 | target_prot, flags, fd, offset)) { | |
c8a706fe | 657 | goto fail; |
2b730f79 | 658 | } |
a03e2d42 FB |
659 | real_start += qemu_host_page_size; |
660 | } | |
661 | /* handle the end of the mapping */ | |
f9cd8f5e RH |
662 | if (last < real_last) { |
663 | abi_ulong real_page = real_last - qemu_host_page_size + 1; | |
664 | if (!mmap_frag(real_page, real_page, last, | |
99982beb | 665 | target_prot, flags, fd, |
f9cd8f5e | 666 | offset + real_page - start)) { |
c8a706fe | 667 | goto fail; |
2b730f79 | 668 | } |
f9cd8f5e | 669 | real_last -= qemu_host_page_size; |
54936004 | 670 | } |
3b46e624 | 671 | |
a03e2d42 | 672 | /* map the middle (easier) */ |
f9cd8f5e | 673 | if (real_start < real_last) { |
a03e2d42 | 674 | void *p; |
55baec0f RH |
675 | off_t offset1; |
676 | ||
2b730f79 | 677 | if (flags & MAP_ANONYMOUS) { |
a03e2d42 | 678 | offset1 = 0; |
2b730f79 | 679 | } else { |
a03e2d42 | 680 | offset1 = offset + real_start - start; |
2b730f79 | 681 | } |
f9cd8f5e | 682 | p = mmap(g2h_untagged(real_start), real_last - real_start + 1, |
0dd55812 | 683 | target_to_host_prot(target_prot), flags, fd, offset1); |
2b730f79 | 684 | if (p == MAP_FAILED) { |
c8a706fe | 685 | goto fail; |
2b730f79 | 686 | } |
f93b7695 | 687 | passthrough_start = real_start; |
f9cd8f5e | 688 | passthrough_last = real_last; |
a03e2d42 | 689 | } |
54936004 FB |
690 | } |
691 | the_end1: | |
26bab757 RH |
692 | if (flags & MAP_ANONYMOUS) { |
693 | page_flags |= PAGE_ANON; | |
694 | } | |
d9c58585 | 695 | page_flags |= PAGE_RESET; |
f9cd8f5e RH |
696 | if (passthrough_start > passthrough_last) { |
697 | page_set_flags(start, last, page_flags); | |
f93b7695 IL |
698 | } else { |
699 | if (start < passthrough_start) { | |
49840a4a | 700 | page_set_flags(start, passthrough_start - 1, page_flags); |
f93b7695 | 701 | } |
f9cd8f5e | 702 | page_set_flags(passthrough_start, passthrough_last, |
f93b7695 | 703 | page_flags | PAGE_PASSTHROUGH); |
f9cd8f5e RH |
704 | if (passthrough_last < last) { |
705 | page_set_flags(passthrough_last + 1, last, page_flags); | |
f93b7695 IL |
706 | } |
707 | } | |
54936004 | 708 | the_end: |
d0e165ae | 709 | trace_target_mmap_complete(start); |
10d0d505 | 710 | if (qemu_loglevel_mask(CPU_LOG_PAGE)) { |
93756fdc RH |
711 | FILE *f = qemu_log_trylock(); |
712 | if (f) { | |
713 | fprintf(f, "page layout changed following mmap\n"); | |
714 | page_dump(f); | |
715 | qemu_log_unlock(f); | |
716 | } | |
10d0d505 | 717 | } |
c8a706fe | 718 | mmap_unlock(); |
54936004 | 719 | return start; |
c8a706fe PB |
720 | fail: |
721 | mmap_unlock(); | |
722 | return -1; | |
54936004 FB |
723 | } |
724 | ||
558a4411 | 725 | static void mmap_reserve_or_unmap(abi_ulong start, abi_ulong len) |
68a1c816 PB |
726 | { |
727 | abi_ulong real_start; | |
260561d8 RH |
728 | abi_ulong real_last; |
729 | abi_ulong real_len; | |
730 | abi_ulong last; | |
731 | abi_ulong a; | |
558a4411 | 732 | void *host_start; |
68a1c816 PB |
733 | int prot; |
734 | ||
260561d8 | 735 | last = start + len - 1; |
68a1c816 | 736 | real_start = start & qemu_host_page_mask; |
260561d8 RH |
737 | real_last = HOST_PAGE_ALIGN(last) - 1; |
738 | ||
739 | /* | |
740 | * If guest pages remain on the first or last host pages, | |
741 | * adjust the deallocation to retain those guest pages. | |
742 | * The single page special case is required for the last page, | |
743 | * lest real_start overflow to zero. | |
744 | */ | |
745 | if (real_last - real_start < qemu_host_page_size) { | |
68a1c816 | 746 | prot = 0; |
260561d8 RH |
747 | for (a = real_start; a < start; a += TARGET_PAGE_SIZE) { |
748 | prot |= page_get_flags(a); | |
68a1c816 | 749 | } |
260561d8 RH |
750 | for (a = last; a < real_last; a += TARGET_PAGE_SIZE) { |
751 | prot |= page_get_flags(a + 1); | |
752 | } | |
753 | if (prot != 0) { | |
754 | return; | |
755 | } | |
756 | } else { | |
757 | for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
758 | prot |= page_get_flags(a); | |
68a1c816 | 759 | } |
2b730f79 | 760 | if (prot != 0) { |
68a1c816 | 761 | real_start += qemu_host_page_size; |
2b730f79 | 762 | } |
260561d8 RH |
763 | |
764 | for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) { | |
765 | prot |= page_get_flags(a + 1); | |
68a1c816 | 766 | } |
2b730f79 | 767 | if (prot != 0) { |
260561d8 RH |
768 | real_last -= qemu_host_page_size; |
769 | } | |
770 | ||
771 | if (real_last < real_start) { | |
772 | return; | |
2b730f79 | 773 | } |
68a1c816 | 774 | } |
260561d8 RH |
775 | |
776 | real_len = real_last - real_start + 1; | |
777 | host_start = g2h_untagged(real_start); | |
778 | ||
558a4411 RH |
779 | if (reserved_va) { |
780 | void *ptr = mmap(host_start, real_len, PROT_NONE, | |
781 | MAP_FIXED | MAP_ANONYMOUS | |
782 | | MAP_PRIVATE | MAP_NORESERVE, -1, 0); | |
783 | assert(ptr == host_start); | |
784 | } else { | |
785 | int ret = munmap(host_start, real_len); | |
786 | assert(ret == 0); | |
787 | } | |
68a1c816 PB |
788 | } |
789 | ||
992f48a0 | 790 | int target_munmap(abi_ulong start, abi_ulong len) |
54936004 | 791 | { |
b7b18d26 AB |
792 | trace_target_munmap(start, len); |
793 | ||
2b730f79 | 794 | if (start & ~TARGET_PAGE_MASK) { |
78cf3390 | 795 | return -TARGET_EINVAL; |
2b730f79 | 796 | } |
54936004 | 797 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 798 | if (len == 0 || !guest_range_valid_untagged(start, len)) { |
78cf3390 | 799 | return -TARGET_EINVAL; |
ebf9a363 MF |
800 | } |
801 | ||
c8a706fe | 802 | mmap_lock(); |
d7b0c5d0 RH |
803 | mmap_reserve_or_unmap(start, len); |
804 | page_set_flags(start, start + len - 1, 0); | |
c8a706fe | 805 | mmap_unlock(); |
d7b0c5d0 RH |
806 | |
807 | return 0; | |
54936004 FB |
808 | } |
809 | ||
992f48a0 BS |
810 | abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, |
811 | abi_ulong new_size, unsigned long flags, | |
812 | abi_ulong new_addr) | |
54936004 FB |
813 | { |
814 | int prot; | |
f19412a2 | 815 | void *host_addr; |
54936004 | 816 | |
46b12f46 | 817 | if (!guest_range_valid_untagged(old_addr, old_size) || |
ebf9a363 | 818 | ((flags & MREMAP_FIXED) && |
46b12f46 | 819 | !guest_range_valid_untagged(new_addr, new_size)) || |
ccc5ccc1 | 820 | ((flags & MREMAP_MAYMOVE) == 0 && |
46b12f46 | 821 | !guest_range_valid_untagged(old_addr, new_size))) { |
ebf9a363 MF |
822 | errno = ENOMEM; |
823 | return -1; | |
824 | } | |
825 | ||
c8a706fe | 826 | mmap_lock(); |
f19412a2 | 827 | |
68a1c816 | 828 | if (flags & MREMAP_FIXED) { |
3e8f1628 RH |
829 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
830 | flags, g2h_untagged(new_addr)); | |
68a1c816 | 831 | |
b76f21a7 | 832 | if (reserved_va && host_addr != MAP_FAILED) { |
2b730f79 RH |
833 | /* |
834 | * If new and old addresses overlap then the above mremap will | |
835 | * already have failed with EINVAL. | |
836 | */ | |
558a4411 | 837 | mmap_reserve_or_unmap(old_addr, old_size); |
68a1c816 PB |
838 | } |
839 | } else if (flags & MREMAP_MAYMOVE) { | |
f19412a2 AJ |
840 | abi_ulong mmap_start; |
841 | ||
30ab9ef2 | 842 | mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE); |
f19412a2 AJ |
843 | |
844 | if (mmap_start == -1) { | |
845 | errno = ENOMEM; | |
846 | host_addr = MAP_FAILED; | |
68a1c816 | 847 | } else { |
3e8f1628 RH |
848 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
849 | flags | MREMAP_FIXED, | |
850 | g2h_untagged(mmap_start)); | |
b76f21a7 | 851 | if (reserved_va) { |
558a4411 | 852 | mmap_reserve_or_unmap(old_addr, old_size); |
c65ffe6d | 853 | } |
68a1c816 | 854 | } |
3af72a4d | 855 | } else { |
68a1c816 | 856 | int prot = 0; |
b76f21a7 | 857 | if (reserved_va && old_size < new_size) { |
68a1c816 PB |
858 | abi_ulong addr; |
859 | for (addr = old_addr + old_size; | |
860 | addr < old_addr + new_size; | |
861 | addr++) { | |
862 | prot |= page_get_flags(addr); | |
863 | } | |
864 | } | |
865 | if (prot == 0) { | |
3e8f1628 RH |
866 | host_addr = mremap(g2h_untagged(old_addr), |
867 | old_size, new_size, flags); | |
56d19084 TK |
868 | |
869 | if (host_addr != MAP_FAILED) { | |
870 | /* Check if address fits target address space */ | |
46b12f46 | 871 | if (!guest_range_valid_untagged(h2g(host_addr), new_size)) { |
56d19084 | 872 | /* Revert mremap() changes */ |
3e8f1628 RH |
873 | host_addr = mremap(g2h_untagged(old_addr), |
874 | new_size, old_size, flags); | |
56d19084 TK |
875 | errno = ENOMEM; |
876 | host_addr = MAP_FAILED; | |
877 | } else if (reserved_va && old_size > new_size) { | |
558a4411 RH |
878 | mmap_reserve_or_unmap(old_addr + old_size, |
879 | old_size - new_size); | |
56d19084 | 880 | } |
68a1c816 PB |
881 | } |
882 | } else { | |
883 | errno = ENOMEM; | |
884 | host_addr = MAP_FAILED; | |
885 | } | |
f19412a2 AJ |
886 | } |
887 | ||
888 | if (host_addr == MAP_FAILED) { | |
c8a706fe PB |
889 | new_addr = -1; |
890 | } else { | |
891 | new_addr = h2g(host_addr); | |
892 | prot = page_get_flags(old_addr); | |
49840a4a RH |
893 | page_set_flags(old_addr, old_addr + old_size - 1, 0); |
894 | page_set_flags(new_addr, new_addr + new_size - 1, | |
d9c58585 | 895 | prot | PAGE_VALID | PAGE_RESET); |
c8a706fe PB |
896 | } |
897 | mmap_unlock(); | |
54936004 FB |
898 | return new_addr; |
899 | } | |
892a4f6a | 900 | |
892a4f6a IL |
901 | abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice) |
902 | { | |
903 | abi_ulong len, end; | |
904 | int ret = 0; | |
905 | ||
906 | if (start & ~TARGET_PAGE_MASK) { | |
907 | return -TARGET_EINVAL; | |
908 | } | |
909 | len = TARGET_PAGE_ALIGN(len_in); | |
910 | ||
911 | if (len_in && !len) { | |
912 | return -TARGET_EINVAL; | |
913 | } | |
914 | ||
915 | end = start + len; | |
916 | if (end < start) { | |
917 | return -TARGET_EINVAL; | |
918 | } | |
919 | ||
920 | if (end == start) { | |
921 | return 0; | |
922 | } | |
923 | ||
924 | if (!guest_range_valid_untagged(start, len)) { | |
925 | return -TARGET_EINVAL; | |
926 | } | |
927 | ||
4530deb1 HD |
928 | /* Translate for some architectures which have different MADV_xxx values */ |
929 | switch (advice) { | |
930 | case TARGET_MADV_DONTNEED: /* alpha */ | |
931 | advice = MADV_DONTNEED; | |
932 | break; | |
933 | case TARGET_MADV_WIPEONFORK: /* parisc */ | |
934 | advice = MADV_WIPEONFORK; | |
935 | break; | |
936 | case TARGET_MADV_KEEPONFORK: /* parisc */ | |
937 | advice = MADV_KEEPONFORK; | |
938 | break; | |
939 | /* we do not care about the other MADV_xxx values yet */ | |
940 | } | |
941 | ||
892a4f6a | 942 | /* |
4530deb1 HD |
943 | * Most advice values are hints, so ignoring and returning success is ok. |
944 | * | |
945 | * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and | |
946 | * MADV_KEEPONFORK are not hints and need to be emulated. | |
892a4f6a | 947 | * |
4530deb1 HD |
948 | * A straight passthrough for those may not be safe because qemu sometimes |
949 | * turns private file-backed mappings into anonymous mappings. | |
ecb796db RH |
950 | * If all guest pages have PAGE_PASSTHROUGH set, mappings have the |
951 | * same semantics for the host as for the guest. | |
892a4f6a | 952 | * |
4530deb1 HD |
953 | * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and |
954 | * return failure if not. | |
955 | * | |
956 | * MADV_DONTNEED is passed through as well, if possible. | |
957 | * If passthrough isn't possible, we nevertheless (wrongly!) return | |
958 | * success, which is broken but some userspace programs fail to work | |
959 | * otherwise. Completely implementing such emulation is quite complicated | |
960 | * though. | |
892a4f6a IL |
961 | */ |
962 | mmap_lock(); | |
4530deb1 HD |
963 | switch (advice) { |
964 | case MADV_WIPEONFORK: | |
965 | case MADV_KEEPONFORK: | |
966 | ret = -EINVAL; | |
967 | /* fall through */ | |
968 | case MADV_DONTNEED: | |
ecb796db | 969 | if (page_check_range(start, len, PAGE_PASSTHROUGH)) { |
4530deb1 HD |
970 | ret = get_errno(madvise(g2h_untagged(start), len, advice)); |
971 | if ((advice == MADV_DONTNEED) && (ret == 0)) { | |
10310cbd | 972 | page_reset_target_data(start, start + len - 1); |
4530deb1 | 973 | } |
dbbf8975 | 974 | } |
892a4f6a IL |
975 | } |
976 | mmap_unlock(); | |
977 | ||
978 | return ret; | |
979 | } |