xref: /openbmc/qemu/bsd-user/mmap.c (revision 795c40b8)
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 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
28 static int __thread mmap_lock_count;
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 }
43 
44 bool have_mmap_lock(void)
45 {
46     return mmap_lock_count > 0 ? true : false;
47 }
48 
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 {
59     if (child)
60         pthread_mutex_init(&mmap_mutex, NULL);
61     else
62         pthread_mutex_unlock(&mmap_mutex);
63 }
64 
65 /* NOTE: all the constants are the HOST ones, but addresses are target. */
66 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
67 {
68     abi_ulong end, host_start, host_end, addr;
69     int prot1, ret;
70 
71 #ifdef DEBUG_MMAP
72     printf("mprotect: start=0x" TARGET_FMT_lx
73            " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
74            prot & PROT_READ ? 'r' : '-',
75            prot & PROT_WRITE ? 'w' : '-',
76            prot & PROT_EXEC ? 'x' : '-');
77 #endif
78 
79     if ((start & ~TARGET_PAGE_MASK) != 0)
80         return -EINVAL;
81     len = TARGET_PAGE_ALIGN(len);
82     end = start + len;
83     if (end < start)
84         return -EINVAL;
85     prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
86     if (len == 0)
87         return 0;
88 
89     mmap_lock();
90     host_start = start & qemu_host_page_mask;
91     host_end = HOST_PAGE_ALIGN(end);
92     if (start > host_start) {
93         /* handle host page containing start */
94         prot1 = prot;
95         for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
96             prot1 |= page_get_flags(addr);
97         }
98         if (host_end == host_start + qemu_host_page_size) {
99             for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
100                 prot1 |= page_get_flags(addr);
101             }
102             end = host_end;
103         }
104         ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
105         if (ret != 0)
106             goto error;
107         host_start += qemu_host_page_size;
108     }
109     if (end < host_end) {
110         prot1 = prot;
111         for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
112             prot1 |= page_get_flags(addr);
113         }
114         ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
115                        prot1 & PAGE_BITS);
116         if (ret != 0)
117             goto error;
118         host_end -= qemu_host_page_size;
119     }
120 
121     /* handle the pages in the middle */
122     if (host_start < host_end) {
123         ret = mprotect(g2h(host_start), host_end - host_start, prot);
124         if (ret != 0)
125             goto error;
126     }
127     page_set_flags(start, start + len, prot | PAGE_VALID);
128     mmap_unlock();
129     return 0;
130 error:
131     mmap_unlock();
132     return ret;
133 }
134 
135 /* map an incomplete host page */
136 static int mmap_frag(abi_ulong real_start,
137                      abi_ulong start, abi_ulong end,
138                      int prot, int flags, int fd, abi_ulong offset)
139 {
140     abi_ulong real_end, addr;
141     void *host_start;
142     int prot1, prot_new;
143 
144     real_end = real_start + qemu_host_page_size;
145     host_start = g2h(real_start);
146 
147     /* get the protection of the target pages outside the mapping */
148     prot1 = 0;
149     for(addr = real_start; addr < real_end; addr++) {
150         if (addr < start || addr >= end)
151             prot1 |= page_get_flags(addr);
152     }
153 
154     if (prot1 == 0) {
155         /* no page was there, so we allocate one */
156         void *p = mmap(host_start, qemu_host_page_size, prot,
157                        flags | MAP_ANON, -1, 0);
158         if (p == MAP_FAILED)
159             return -1;
160         prot1 = prot;
161     }
162     prot1 &= PAGE_BITS;
163 
164     prot_new = prot | prot1;
165     if (!(flags & MAP_ANON)) {
166         /* msync() won't work here, so we return an error if write is
167            possible while it is a shared mapping */
168         if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
169             (prot & PROT_WRITE))
170             return -1;
171 
172         /* adjust protection to be able to read */
173         if (!(prot1 & PROT_WRITE))
174             mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
175 
176         /* read the corresponding file data */
177         pread(fd, g2h(start), end - start, offset);
178 
179         /* put final protection */
180         if (prot_new != (prot1 | PROT_WRITE))
181             mprotect(host_start, qemu_host_page_size, prot_new);
182     } else {
183         /* just update the protection */
184         if (prot_new != prot1) {
185             mprotect(host_start, qemu_host_page_size, prot_new);
186         }
187     }
188     return 0;
189 }
190 
191 static abi_ulong mmap_next_start = 0x40000000;
192 
193 unsigned long last_brk;
194 
195 /* find a free memory area of size 'size'. The search starts at
196    'start'. If 'start' == 0, then a default start address is used.
197    Return -1 if error.
198 */
199 /* page_init() marks pages used by the host as reserved to be sure not
200    to use them. */
201 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
202 {
203     abi_ulong addr, addr1, addr_start;
204     int prot;
205     unsigned long new_brk;
206 
207     new_brk = (unsigned long)sbrk(0);
208     if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
209         /* This is a hack to catch the host allocating memory with brk().
210            If it uses mmap then we loose.
211            FIXME: We really want to avoid the host allocating memory in
212            the first place, and maybe leave some slack to avoid switching
213            to mmap.  */
214         page_set_flags(last_brk & TARGET_PAGE_MASK,
215                        TARGET_PAGE_ALIGN(new_brk),
216                        PAGE_RESERVED);
217     }
218     last_brk = new_brk;
219 
220     size = HOST_PAGE_ALIGN(size);
221     start = start & qemu_host_page_mask;
222     addr = start;
223     if (addr == 0)
224         addr = mmap_next_start;
225     addr_start = addr;
226     for(;;) {
227         prot = 0;
228         for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
229             prot |= page_get_flags(addr1);
230         }
231         if (prot == 0)
232             break;
233         addr += qemu_host_page_size;
234         /* we found nothing */
235         if (addr == addr_start)
236             return (abi_ulong)-1;
237     }
238     if (start == 0)
239         mmap_next_start = addr + size;
240     return addr;
241 }
242 
243 /* NOTE: all the constants are the HOST ones */
244 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
245                      int flags, int fd, abi_ulong offset)
246 {
247     abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
248     unsigned long host_start;
249 
250     mmap_lock();
251 #ifdef DEBUG_MMAP
252     {
253         printf("mmap: start=0x" TARGET_FMT_lx
254                " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
255                start, len,
256                prot & PROT_READ ? 'r' : '-',
257                prot & PROT_WRITE ? 'w' : '-',
258                prot & PROT_EXEC ? 'x' : '-');
259         if (flags & MAP_FIXED)
260             printf("MAP_FIXED ");
261         if (flags & MAP_ANON)
262             printf("MAP_ANON ");
263         switch(flags & TARGET_BSD_MAP_FLAGMASK) {
264         case MAP_PRIVATE:
265             printf("MAP_PRIVATE ");
266             break;
267         case MAP_SHARED:
268             printf("MAP_SHARED ");
269             break;
270         default:
271             printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK);
272             break;
273         }
274         printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
275     }
276 #endif
277 
278     if (offset & ~TARGET_PAGE_MASK) {
279         errno = EINVAL;
280         goto fail;
281     }
282 
283     len = TARGET_PAGE_ALIGN(len);
284     if (len == 0)
285         goto the_end;
286     real_start = start & qemu_host_page_mask;
287 
288     if (!(flags & MAP_FIXED)) {
289         abi_ulong mmap_start;
290         void *p;
291         host_offset = offset & qemu_host_page_mask;
292         host_len = len + offset - host_offset;
293         host_len = HOST_PAGE_ALIGN(host_len);
294         mmap_start = mmap_find_vma(real_start, host_len);
295         if (mmap_start == (abi_ulong)-1) {
296             errno = ENOMEM;
297             goto fail;
298         }
299         /* Note: we prefer to control the mapping address. It is
300            especially important if qemu_host_page_size >
301            qemu_real_host_page_size */
302         p = mmap(g2h(mmap_start),
303                  host_len, prot, flags | MAP_FIXED, fd, host_offset);
304         if (p == MAP_FAILED)
305             goto fail;
306         /* update start so that it points to the file position at 'offset' */
307         host_start = (unsigned long)p;
308         if (!(flags & MAP_ANON))
309             host_start += offset - host_offset;
310         start = h2g(host_start);
311     } else {
312         int flg;
313         target_ulong addr;
314 
315         if (start & ~TARGET_PAGE_MASK) {
316             errno = EINVAL;
317             goto fail;
318         }
319         end = start + len;
320         real_end = HOST_PAGE_ALIGN(end);
321 
322         for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
323             flg = page_get_flags(addr);
324             if (flg & PAGE_RESERVED) {
325                 errno = ENXIO;
326                 goto fail;
327             }
328         }
329 
330         /* worst case: we cannot map the file because the offset is not
331            aligned, so we read it */
332         if (!(flags & MAP_ANON) &&
333             (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
334             /* msync() won't work here, so we return an error if write is
335                possible while it is a shared mapping */
336             if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
337                 (prot & PROT_WRITE)) {
338                 errno = EINVAL;
339                 goto fail;
340             }
341             retaddr = target_mmap(start, len, prot | PROT_WRITE,
342                                   MAP_FIXED | MAP_PRIVATE | MAP_ANON,
343                                   -1, 0);
344             if (retaddr == -1)
345                 goto fail;
346             pread(fd, g2h(start), len, offset);
347             if (!(prot & PROT_WRITE)) {
348                 ret = target_mprotect(start, len, prot);
349                 if (ret != 0) {
350                     start = ret;
351                     goto the_end;
352                 }
353             }
354             goto the_end;
355         }
356 
357         /* handle the start of the mapping */
358         if (start > real_start) {
359             if (real_end == real_start + qemu_host_page_size) {
360                 /* one single host page */
361                 ret = mmap_frag(real_start, start, end,
362                                 prot, flags, fd, offset);
363                 if (ret == -1)
364                     goto fail;
365                 goto the_end1;
366             }
367             ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
368                             prot, flags, fd, offset);
369             if (ret == -1)
370                 goto fail;
371             real_start += qemu_host_page_size;
372         }
373         /* handle the end of the mapping */
374         if (end < real_end) {
375             ret = mmap_frag(real_end - qemu_host_page_size,
376                             real_end - qemu_host_page_size, real_end,
377                             prot, flags, fd,
378                             offset + real_end - qemu_host_page_size - start);
379             if (ret == -1)
380                 goto fail;
381             real_end -= qemu_host_page_size;
382         }
383 
384         /* map the middle (easier) */
385         if (real_start < real_end) {
386             void *p;
387             unsigned long offset1;
388             if (flags & MAP_ANON)
389                 offset1 = 0;
390             else
391                 offset1 = offset + real_start - start;
392             p = mmap(g2h(real_start), real_end - real_start,
393                      prot, flags, fd, offset1);
394             if (p == MAP_FAILED)
395                 goto fail;
396         }
397     }
398  the_end1:
399     page_set_flags(start, start + len, prot | PAGE_VALID);
400  the_end:
401 #ifdef DEBUG_MMAP
402     printf("ret=0x" TARGET_FMT_lx "\n", start);
403     page_dump(stdout);
404     printf("\n");
405 #endif
406     mmap_unlock();
407     return start;
408 fail:
409     mmap_unlock();
410     return -1;
411 }
412 
413 int target_munmap(abi_ulong start, abi_ulong len)
414 {
415     abi_ulong end, real_start, real_end, addr;
416     int prot, ret;
417 
418 #ifdef DEBUG_MMAP
419     printf("munmap: start=0x%lx len=0x%lx\n", start, len);
420 #endif
421     if (start & ~TARGET_PAGE_MASK)
422         return -EINVAL;
423     len = TARGET_PAGE_ALIGN(len);
424     if (len == 0)
425         return -EINVAL;
426     mmap_lock();
427     end = start + len;
428     real_start = start & qemu_host_page_mask;
429     real_end = HOST_PAGE_ALIGN(end);
430 
431     if (start > real_start) {
432         /* handle host page containing start */
433         prot = 0;
434         for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
435             prot |= page_get_flags(addr);
436         }
437         if (real_end == real_start + qemu_host_page_size) {
438             for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
439                 prot |= page_get_flags(addr);
440             }
441             end = real_end;
442         }
443         if (prot != 0)
444             real_start += qemu_host_page_size;
445     }
446     if (end < real_end) {
447         prot = 0;
448         for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
449             prot |= page_get_flags(addr);
450         }
451         if (prot != 0)
452             real_end -= qemu_host_page_size;
453     }
454 
455     ret = 0;
456     /* unmap what we can */
457     if (real_start < real_end) {
458         ret = munmap(g2h(real_start), real_end - real_start);
459     }
460 
461     if (ret == 0)
462         page_set_flags(start, start + len, 0);
463     mmap_unlock();
464     return ret;
465 }
466 
467 int target_msync(abi_ulong start, abi_ulong len, int flags)
468 {
469     abi_ulong end;
470 
471     if (start & ~TARGET_PAGE_MASK)
472         return -EINVAL;
473     len = TARGET_PAGE_ALIGN(len);
474     end = start + len;
475     if (end < start)
476         return -EINVAL;
477     if (end == start)
478         return 0;
479 
480     start &= qemu_host_page_mask;
481     return msync(g2h(start), end - start, flags);
482 }
483