xref: /qemu/bsd-user/mmap.c (revision 530e7615ce3c01882e582c84dc6304ab98a3d5c5)
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, write to the Free Software
18  *  Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
19  *  MA 02110-1301, USA.
20  */
21 #include <stdlib.h>
22 #include <stdio.h>
23 #include <stdarg.h>
24 #include <string.h>
25 #include <unistd.h>
26 #include <errno.h>
27 #include <sys/mman.h>
28 
29 #include "qemu.h"
30 #include "qemu-common.h"
31 #include "bsd-mman.h"
32 
33 //#define DEBUG_MMAP
34 
35 #if defined(USE_NPTL)
36 pthread_mutex_t mmap_mutex;
37 static int __thread mmap_lock_count;
38 
39 void mmap_lock(void)
40 {
41     if (mmap_lock_count++ == 0) {
42         pthread_mutex_lock(&mmap_mutex);
43     }
44 }
45 
46 void mmap_unlock(void)
47 {
48     if (--mmap_lock_count == 0) {
49         pthread_mutex_unlock(&mmap_mutex);
50     }
51 }
52 
53 /* Grab lock to make sure things are in a consistent state after fork().  */
54 void mmap_fork_start(void)
55 {
56     if (mmap_lock_count)
57         abort();
58     pthread_mutex_lock(&mmap_mutex);
59 }
60 
61 void mmap_fork_end(int child)
62 {
63     if (child)
64         pthread_mutex_init(&mmap_mutex, NULL);
65     else
66         pthread_mutex_unlock(&mmap_mutex);
67 }
68 #else
69 /* We aren't threadsafe to start with, so no need to worry about locking.  */
70 void mmap_lock(void)
71 {
72 }
73 
74 void mmap_unlock(void)
75 {
76 }
77 #endif
78 
79 void *qemu_vmalloc(size_t size)
80 {
81     void *p;
82     unsigned long addr;
83     mmap_lock();
84     /* Use map and mark the pages as used.  */
85     p = mmap(NULL, size, PROT_READ | PROT_WRITE,
86              MAP_PRIVATE | MAP_ANON, -1, 0);
87 
88     addr = (unsigned long)p;
89     if (addr == (target_ulong) addr) {
90         /* Allocated region overlaps guest address space.
91            This may recurse.  */
92         page_set_flags(addr & TARGET_PAGE_MASK, TARGET_PAGE_ALIGN(addr + size),
93                        PAGE_RESERVED);
94     }
95 
96     mmap_unlock();
97     return p;
98 }
99 
100 void *qemu_malloc(size_t size)
101 {
102     char * p;
103     size += 16;
104     p = qemu_vmalloc(size);
105     *(size_t *)p = size;
106     return p + 16;
107 }
108 
109 /* We use map, which is always zero initialized.  */
110 void * qemu_mallocz(size_t size)
111 {
112     return qemu_malloc(size);
113 }
114 
115 void qemu_free(void *ptr)
116 {
117     /* FIXME: We should unmark the reserved pages here.  However this gets
118        complicated when one target page spans multiple host pages, so we
119        don't bother.  */
120     size_t *p;
121     p = (size_t *)((char *)ptr - 16);
122     munmap(p, *p);
123 }
124 
125 /* NOTE: all the constants are the HOST ones, but addresses are target. */
126 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
127 {
128     abi_ulong end, host_start, host_end, addr;
129     int prot1, ret;
130 
131 #ifdef DEBUG_MMAP
132     printf("mprotect: start=0x" TARGET_FMT_lx
133            " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
134            prot & PROT_READ ? 'r' : '-',
135            prot & PROT_WRITE ? 'w' : '-',
136            prot & PROT_EXEC ? 'x' : '-');
137 #endif
138 
139     if ((start & ~TARGET_PAGE_MASK) != 0)
140         return -EINVAL;
141     len = TARGET_PAGE_ALIGN(len);
142     end = start + len;
143     if (end < start)
144         return -EINVAL;
145     prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
146     if (len == 0)
147         return 0;
148 
149     mmap_lock();
150     host_start = start & qemu_host_page_mask;
151     host_end = HOST_PAGE_ALIGN(end);
152     if (start > host_start) {
153         /* handle host page containing start */
154         prot1 = prot;
155         for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
156             prot1 |= page_get_flags(addr);
157         }
158         if (host_end == host_start + qemu_host_page_size) {
159             for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
160                 prot1 |= page_get_flags(addr);
161             }
162             end = host_end;
163         }
164         ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
165         if (ret != 0)
166             goto error;
167         host_start += qemu_host_page_size;
168     }
169     if (end < host_end) {
170         prot1 = prot;
171         for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
172             prot1 |= page_get_flags(addr);
173         }
174         ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
175                        prot1 & PAGE_BITS);
176         if (ret != 0)
177             goto error;
178         host_end -= qemu_host_page_size;
179     }
180 
181     /* handle the pages in the middle */
182     if (host_start < host_end) {
183         ret = mprotect(g2h(host_start), host_end - host_start, prot);
184         if (ret != 0)
185             goto error;
186     }
187     page_set_flags(start, start + len, prot | PAGE_VALID);
188     mmap_unlock();
189     return 0;
190 error:
191     mmap_unlock();
192     return ret;
193 }
194 
195 /* map an incomplete host page */
196 static int mmap_frag(abi_ulong real_start,
197                      abi_ulong start, abi_ulong end,
198                      int prot, int flags, int fd, abi_ulong offset)
199 {
200     abi_ulong real_end, addr;
201     void *host_start;
202     int prot1, prot_new;
203 
204     real_end = real_start + qemu_host_page_size;
205     host_start = g2h(real_start);
206 
207     /* get the protection of the target pages outside the mapping */
208     prot1 = 0;
209     for(addr = real_start; addr < real_end; addr++) {
210         if (addr < start || addr >= end)
211             prot1 |= page_get_flags(addr);
212     }
213 
214     if (prot1 == 0) {
215         /* no page was there, so we allocate one */
216         void *p = mmap(host_start, qemu_host_page_size, prot,
217                        flags | MAP_ANON, -1, 0);
218         if (p == MAP_FAILED)
219             return -1;
220         prot1 = prot;
221     }
222     prot1 &= PAGE_BITS;
223 
224     prot_new = prot | prot1;
225     if (!(flags & MAP_ANON)) {
226         /* msync() won't work here, so we return an error if write is
227            possible while it is a shared mapping */
228         if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
229             (prot & PROT_WRITE))
230             return -EINVAL;
231 
232         /* adjust protection to be able to read */
233         if (!(prot1 & PROT_WRITE))
234             mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
235 
236         /* read the corresponding file data */
237         pread(fd, g2h(start), end - start, offset);
238 
239         /* put final protection */
240         if (prot_new != (prot1 | PROT_WRITE))
241             mprotect(host_start, qemu_host_page_size, prot_new);
242     } else {
243         /* just update the protection */
244         if (prot_new != prot1) {
245             mprotect(host_start, qemu_host_page_size, prot_new);
246         }
247     }
248     return 0;
249 }
250 
251 #if defined(__CYGWIN__)
252 /* Cygwin doesn't have a whole lot of address space.  */
253 static abi_ulong mmap_next_start = 0x18000000;
254 #else
255 static abi_ulong mmap_next_start = 0x40000000;
256 #endif
257 
258 unsigned long last_brk;
259 
260 /* find a free memory area of size 'size'. The search starts at
261    'start'. If 'start' == 0, then a default start address is used.
262    Return -1 if error.
263 */
264 /* page_init() marks pages used by the host as reserved to be sure not
265    to use them. */
266 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
267 {
268     abi_ulong addr, addr1, addr_start;
269     int prot;
270     unsigned long new_brk;
271 
272     new_brk = (unsigned long)sbrk(0);
273     if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
274         /* This is a hack to catch the host allocating memory with brk().
275            If it uses mmap then we loose.
276            FIXME: We really want to avoid the host allocating memory in
277            the first place, and maybe leave some slack to avoid switching
278            to mmap.  */
279         page_set_flags(last_brk & TARGET_PAGE_MASK,
280                        TARGET_PAGE_ALIGN(new_brk),
281                        PAGE_RESERVED);
282     }
283     last_brk = new_brk;
284 
285     size = HOST_PAGE_ALIGN(size);
286     start = start & qemu_host_page_mask;
287     addr = start;
288     if (addr == 0)
289         addr = mmap_next_start;
290     addr_start = addr;
291     for(;;) {
292         prot = 0;
293         for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
294             prot |= page_get_flags(addr1);
295         }
296         if (prot == 0)
297             break;
298         addr += qemu_host_page_size;
299         /* we found nothing */
300         if (addr == addr_start)
301             return (abi_ulong)-1;
302     }
303     if (start == 0)
304         mmap_next_start = addr + size;
305     return addr;
306 }
307 
308 /* NOTE: all the constants are the HOST ones */
309 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
310                      int flags, int fd, abi_ulong offset)
311 {
312     abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
313     unsigned long host_start;
314 
315     mmap_lock();
316 #ifdef DEBUG_MMAP
317     {
318         printf("mmap: start=0x" TARGET_FMT_lx
319                " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
320                start, len,
321                prot & PROT_READ ? 'r' : '-',
322                prot & PROT_WRITE ? 'w' : '-',
323                prot & PROT_EXEC ? 'x' : '-');
324         if (flags & MAP_FIXED)
325             printf("MAP_FIXED ");
326         if (flags & MAP_ANON)
327             printf("MAP_ANON ");
328         switch(flags & TARGET_BSD_MAP_FLAGMASK) {
329         case MAP_PRIVATE:
330             printf("MAP_PRIVATE ");
331             break;
332         case MAP_SHARED:
333             printf("MAP_SHARED ");
334             break;
335         default:
336             printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK);
337             break;
338         }
339         printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
340     }
341 #endif
342 
343     if (offset & ~TARGET_PAGE_MASK) {
344         errno = EINVAL;
345         goto fail;
346     }
347 
348     len = TARGET_PAGE_ALIGN(len);
349     if (len == 0)
350         goto the_end;
351     real_start = start & qemu_host_page_mask;
352 
353     if (!(flags & MAP_FIXED)) {
354         abi_ulong mmap_start;
355         void *p;
356         host_offset = offset & qemu_host_page_mask;
357         host_len = len + offset - host_offset;
358         host_len = HOST_PAGE_ALIGN(host_len);
359         mmap_start = mmap_find_vma(real_start, host_len);
360         if (mmap_start == (abi_ulong)-1) {
361             errno = ENOMEM;
362             goto fail;
363         }
364         /* Note: we prefer to control the mapping address. It is
365            especially important if qemu_host_page_size >
366            qemu_real_host_page_size */
367         p = mmap(g2h(mmap_start),
368                  host_len, prot, flags | MAP_FIXED, fd, host_offset);
369         if (p == MAP_FAILED)
370             goto fail;
371         /* update start so that it points to the file position at 'offset' */
372         host_start = (unsigned long)p;
373         if (!(flags & MAP_ANON))
374             host_start += offset - host_offset;
375         start = h2g(host_start);
376     } else {
377         int flg;
378         target_ulong addr;
379 
380         if (start & ~TARGET_PAGE_MASK) {
381             errno = EINVAL;
382             goto fail;
383         }
384         end = start + len;
385         real_end = HOST_PAGE_ALIGN(end);
386 
387         for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
388             flg = page_get_flags(addr);
389             if (flg & PAGE_RESERVED) {
390                 errno = ENXIO;
391                 goto fail;
392             }
393         }
394 
395         /* worst case: we cannot map the file because the offset is not
396            aligned, so we read it */
397         if (!(flags & MAP_ANON) &&
398             (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
399             /* msync() won't work here, so we return an error if write is
400                possible while it is a shared mapping */
401             if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
402                 (prot & PROT_WRITE)) {
403                 errno = EINVAL;
404                 goto fail;
405             }
406             retaddr = target_mmap(start, len, prot | PROT_WRITE,
407                                   MAP_FIXED | MAP_PRIVATE | MAP_ANON,
408                                   -1, 0);
409             if (retaddr == -1)
410                 goto fail;
411             pread(fd, g2h(start), len, offset);
412             if (!(prot & PROT_WRITE)) {
413                 ret = target_mprotect(start, len, prot);
414                 if (ret != 0) {
415                     start = ret;
416                     goto the_end;
417                 }
418             }
419             goto the_end;
420         }
421 
422         /* handle the start of the mapping */
423         if (start > real_start) {
424             if (real_end == real_start + qemu_host_page_size) {
425                 /* one single host page */
426                 ret = mmap_frag(real_start, start, end,
427                                 prot, flags, fd, offset);
428                 if (ret == -1)
429                     goto fail;
430                 goto the_end1;
431             }
432             ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
433                             prot, flags, fd, offset);
434             if (ret == -1)
435                 goto fail;
436             real_start += qemu_host_page_size;
437         }
438         /* handle the end of the mapping */
439         if (end < real_end) {
440             ret = mmap_frag(real_end - qemu_host_page_size,
441                             real_end - qemu_host_page_size, real_end,
442                             prot, flags, fd,
443                             offset + real_end - qemu_host_page_size - start);
444             if (ret == -1)
445                 goto fail;
446             real_end -= qemu_host_page_size;
447         }
448 
449         /* map the middle (easier) */
450         if (real_start < real_end) {
451             void *p;
452             unsigned long offset1;
453             if (flags & MAP_ANON)
454                 offset1 = 0;
455             else
456                 offset1 = offset + real_start - start;
457             p = mmap(g2h(real_start), real_end - real_start,
458                      prot, flags, fd, offset1);
459             if (p == MAP_FAILED)
460                 goto fail;
461         }
462     }
463  the_end1:
464     page_set_flags(start, start + len, prot | PAGE_VALID);
465  the_end:
466 #ifdef DEBUG_MMAP
467     printf("ret=0x" TARGET_FMT_lx "\n", start);
468     page_dump(stdout);
469     printf("\n");
470 #endif
471     mmap_unlock();
472     return start;
473 fail:
474     mmap_unlock();
475     return -1;
476 }
477 
478 int target_munmap(abi_ulong start, abi_ulong len)
479 {
480     abi_ulong end, real_start, real_end, addr;
481     int prot, ret;
482 
483 #ifdef DEBUG_MMAP
484     printf("munmap: start=0x%lx len=0x%lx\n", start, len);
485 #endif
486     if (start & ~TARGET_PAGE_MASK)
487         return -EINVAL;
488     len = TARGET_PAGE_ALIGN(len);
489     if (len == 0)
490         return -EINVAL;
491     mmap_lock();
492     end = start + len;
493     real_start = start & qemu_host_page_mask;
494     real_end = HOST_PAGE_ALIGN(end);
495 
496     if (start > real_start) {
497         /* handle host page containing start */
498         prot = 0;
499         for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
500             prot |= page_get_flags(addr);
501         }
502         if (real_end == real_start + qemu_host_page_size) {
503             for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
504                 prot |= page_get_flags(addr);
505             }
506             end = real_end;
507         }
508         if (prot != 0)
509             real_start += qemu_host_page_size;
510     }
511     if (end < real_end) {
512         prot = 0;
513         for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
514             prot |= page_get_flags(addr);
515         }
516         if (prot != 0)
517             real_end -= qemu_host_page_size;
518     }
519 
520     ret = 0;
521     /* unmap what we can */
522     if (real_start < real_end) {
523         ret = munmap(g2h(real_start), real_end - real_start);
524     }
525 
526     if (ret == 0)
527         page_set_flags(start, start + len, 0);
528     mmap_unlock();
529     return ret;
530 }
531 
532 int target_msync(abi_ulong start, abi_ulong len, int flags)
533 {
534     abi_ulong end;
535 
536     if (start & ~TARGET_PAGE_MASK)
537         return -EINVAL;
538     len = TARGET_PAGE_ALIGN(len);
539     end = start + len;
540     if (end < start)
541         return -EINVAL;
542     if (end == start)
543         return 0;
544 
545     start &= qemu_host_page_mask;
546     return msync(g2h(start), end - start, flags);
547 }
548