xref: /qemu/linux-user/mmap.c (revision d884e272)
1 /*
2  *  mmap support for qemu
3  *
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
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 #include "qemu/osdep.h"
20 #include <sys/shm.h>
21 #include "trace.h"
22 #include "exec/log.h"
23 #include "qemu.h"
24 #include "user-internals.h"
25 #include "user-mmap.h"
26 #include "target_mman.h"
27 #include "qemu/interval-tree.h"
28 
29 #ifdef TARGET_ARM
30 #include "target/arm/cpu-features.h"
31 #endif
32 
33 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
34 static __thread int mmap_lock_count;
35 
36 void mmap_lock(void)
37 {
38     if (mmap_lock_count++ == 0) {
39         pthread_mutex_lock(&mmap_mutex);
40     }
41 }
42 
43 void mmap_unlock(void)
44 {
45     assert(mmap_lock_count > 0);
46     if (--mmap_lock_count == 0) {
47         pthread_mutex_unlock(&mmap_mutex);
48     }
49 }
50 
51 bool have_mmap_lock(void)
52 {
53     return mmap_lock_count > 0 ? true : false;
54 }
55 
56 /* Grab lock to make sure things are in a consistent state after fork().  */
57 void mmap_fork_start(void)
58 {
59     if (mmap_lock_count)
60         abort();
61     pthread_mutex_lock(&mmap_mutex);
62 }
63 
64 void mmap_fork_end(int child)
65 {
66     if (child) {
67         pthread_mutex_init(&mmap_mutex, NULL);
68     } else {
69         pthread_mutex_unlock(&mmap_mutex);
70     }
71 }
72 
73 /* Protected by mmap_lock. */
74 static IntervalTreeRoot shm_regions;
75 
76 static void shm_region_add(abi_ptr start, abi_ptr last)
77 {
78     IntervalTreeNode *i = g_new0(IntervalTreeNode, 1);
79 
80     i->start = start;
81     i->last = last;
82     interval_tree_insert(i, &shm_regions);
83 }
84 
85 static abi_ptr shm_region_find(abi_ptr start)
86 {
87     IntervalTreeNode *i;
88 
89     for (i = interval_tree_iter_first(&shm_regions, start, start); i;
90          i = interval_tree_iter_next(i, start, start)) {
91         if (i->start == start) {
92             return i->last;
93         }
94     }
95     return 0;
96 }
97 
98 static void shm_region_rm_complete(abi_ptr start, abi_ptr last)
99 {
100     IntervalTreeNode *i, *n;
101 
102     for (i = interval_tree_iter_first(&shm_regions, start, last); i; i = n) {
103         n = interval_tree_iter_next(i, start, last);
104         if (i->start >= start && i->last <= last) {
105             interval_tree_remove(i, &shm_regions);
106             g_free(i);
107         }
108     }
109 }
110 
111 /*
112  * Validate target prot bitmask.
113  * Return the prot bitmask for the host in *HOST_PROT.
114  * Return 0 if the target prot bitmask is invalid, otherwise
115  * the internal qemu page_flags (which will include PAGE_VALID).
116  */
117 static int validate_prot_to_pageflags(int prot)
118 {
119     int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM;
120     int page_flags = (prot & PAGE_BITS) | PAGE_VALID;
121 
122 #ifdef TARGET_AARCH64
123     {
124         ARMCPU *cpu = ARM_CPU(thread_cpu);
125 
126         /*
127          * The PROT_BTI bit is only accepted if the cpu supports the feature.
128          * Since this is the unusual case, don't bother checking unless
129          * the bit has been requested.  If set and valid, record the bit
130          * within QEMU's page_flags.
131          */
132         if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) {
133             valid |= TARGET_PROT_BTI;
134             page_flags |= PAGE_BTI;
135         }
136         /* Similarly for the PROT_MTE bit. */
137         if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) {
138             valid |= TARGET_PROT_MTE;
139             page_flags |= PAGE_MTE;
140         }
141     }
142 #elif defined(TARGET_HPPA)
143     valid |= PROT_GROWSDOWN | PROT_GROWSUP;
144 #endif
145 
146     return prot & ~valid ? 0 : page_flags;
147 }
148 
149 /*
150  * For the host, we need not pass anything except read/write/exec.
151  * While PROT_SEM is allowed by all hosts, it is also ignored, so
152  * don't bother transforming guest bit to host bit.  Any other
153  * target-specific prot bits will not be understood by the host
154  * and will need to be encoded into page_flags for qemu emulation.
155  *
156  * Pages that are executable by the guest will never be executed
157  * by the host, but the host will need to be able to read them.
158  */
159 static int target_to_host_prot(int prot)
160 {
161     return (prot & (PROT_READ | PROT_WRITE)) |
162            (prot & PROT_EXEC ? PROT_READ : 0);
163 }
164 
165 /* NOTE: all the constants are the HOST ones, but addresses are target. */
166 int target_mprotect(abi_ulong start, abi_ulong len, int target_prot)
167 {
168     int host_page_size = qemu_real_host_page_size();
169     abi_ulong starts[3];
170     abi_ulong lens[3];
171     int prots[3];
172     abi_ulong host_start, host_last, last;
173     int prot1, ret, page_flags, nranges;
174 
175     trace_target_mprotect(start, len, target_prot);
176 
177     if ((start & ~TARGET_PAGE_MASK) != 0) {
178         return -TARGET_EINVAL;
179     }
180     page_flags = validate_prot_to_pageflags(target_prot);
181     if (!page_flags) {
182         return -TARGET_EINVAL;
183     }
184     if (len == 0) {
185         return 0;
186     }
187     len = TARGET_PAGE_ALIGN(len);
188     if (!guest_range_valid_untagged(start, len)) {
189         return -TARGET_ENOMEM;
190     }
191 
192     last = start + len - 1;
193     host_start = start & -host_page_size;
194     host_last = ROUND_UP(last, host_page_size) - 1;
195     nranges = 0;
196 
197     mmap_lock();
198 
199     if (host_last - host_start < host_page_size) {
200         /* Single host page contains all guest pages: sum the prot. */
201         prot1 = target_prot;
202         for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
203             prot1 |= page_get_flags(a);
204         }
205         for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
206             prot1 |= page_get_flags(a + 1);
207         }
208         starts[nranges] = host_start;
209         lens[nranges] = host_page_size;
210         prots[nranges] = prot1;
211         nranges++;
212     } else {
213         if (host_start < start) {
214             /* Host page contains more than one guest page: sum the prot. */
215             prot1 = target_prot;
216             for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
217                 prot1 |= page_get_flags(a);
218             }
219             /* If the resulting sum differs, create a new range. */
220             if (prot1 != target_prot) {
221                 starts[nranges] = host_start;
222                 lens[nranges] = host_page_size;
223                 prots[nranges] = prot1;
224                 nranges++;
225                 host_start += host_page_size;
226             }
227         }
228 
229         if (last < host_last) {
230             /* Host page contains more than one guest page: sum the prot. */
231             prot1 = target_prot;
232             for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
233                 prot1 |= page_get_flags(a + 1);
234             }
235             /* If the resulting sum differs, create a new range. */
236             if (prot1 != target_prot) {
237                 host_last -= host_page_size;
238                 starts[nranges] = host_last + 1;
239                 lens[nranges] = host_page_size;
240                 prots[nranges] = prot1;
241                 nranges++;
242             }
243         }
244 
245         /* Create a range for the middle, if any remains. */
246         if (host_start < host_last) {
247             starts[nranges] = host_start;
248             lens[nranges] = host_last - host_start + 1;
249             prots[nranges] = target_prot;
250             nranges++;
251         }
252     }
253 
254     for (int i = 0; i < nranges; ++i) {
255         ret = mprotect(g2h_untagged(starts[i]), lens[i],
256                        target_to_host_prot(prots[i]));
257         if (ret != 0) {
258             goto error;
259         }
260     }
261 
262     page_set_flags(start, last, page_flags);
263     ret = 0;
264 
265  error:
266     mmap_unlock();
267     return ret;
268 }
269 
270 /*
271  * Perform munmap on behalf of the target, with host parameters.
272  * If reserved_va, we must replace the memory reservation.
273  */
274 static int do_munmap(void *addr, size_t len)
275 {
276     if (reserved_va) {
277         void *ptr = mmap(addr, len, PROT_NONE,
278                          MAP_FIXED | MAP_ANONYMOUS
279                          | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
280         return ptr == addr ? 0 : -1;
281     }
282     return munmap(addr, len);
283 }
284 
285 /*
286  * Map an incomplete host page.
287  *
288  * Here be dragons.  This case will not work if there is an existing
289  * overlapping host page, which is file mapped, and for which the mapping
290  * is beyond the end of the file.  In that case, we will see SIGBUS when
291  * trying to write a portion of this page.
292  *
293  * FIXME: Work around this with a temporary signal handler and longjmp.
294  */
295 static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last,
296                       int prot, int flags, int fd, off_t offset)
297 {
298     int host_page_size = qemu_real_host_page_size();
299     abi_ulong real_last;
300     void *host_start;
301     int prot_old, prot_new;
302     int host_prot_old, host_prot_new;
303 
304     if (!(flags & MAP_ANONYMOUS)
305         && (flags & MAP_TYPE) == MAP_SHARED
306         && (prot & PROT_WRITE)) {
307         /*
308          * msync() won't work with the partial page, so we return an
309          * error if write is possible while it is a shared mapping.
310          */
311         errno = EINVAL;
312         return false;
313     }
314 
315     real_last = real_start + host_page_size - 1;
316     host_start = g2h_untagged(real_start);
317 
318     /* Get the protection of the target pages outside the mapping. */
319     prot_old = 0;
320     for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) {
321         prot_old |= page_get_flags(a);
322     }
323     for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) {
324         prot_old |= page_get_flags(a);
325     }
326 
327     if (prot_old == 0) {
328         /*
329          * Since !(prot_old & PAGE_VALID), there were no guest pages
330          * outside of the fragment we need to map.  Allocate a new host
331          * page to cover, discarding whatever else may have been present.
332          */
333         void *p = mmap(host_start, host_page_size,
334                        target_to_host_prot(prot),
335                        flags | MAP_ANONYMOUS, -1, 0);
336         if (p != host_start) {
337             if (p != MAP_FAILED) {
338                 do_munmap(p, host_page_size);
339                 errno = EEXIST;
340             }
341             return false;
342         }
343         prot_old = prot;
344     }
345     prot_new = prot | prot_old;
346 
347     host_prot_old = target_to_host_prot(prot_old);
348     host_prot_new = target_to_host_prot(prot_new);
349 
350     /* Adjust protection to be able to write. */
351     if (!(host_prot_old & PROT_WRITE)) {
352         host_prot_old |= PROT_WRITE;
353         mprotect(host_start, host_page_size, host_prot_old);
354     }
355 
356     /* Read or zero the new guest pages. */
357     if (flags & MAP_ANONYMOUS) {
358         memset(g2h_untagged(start), 0, last - start + 1);
359     } else {
360         if (pread(fd, g2h_untagged(start), last - start + 1, offset) == -1) {
361             return false;
362         }
363     }
364 
365     /* Put final protection */
366     if (host_prot_new != host_prot_old) {
367         mprotect(host_start, host_page_size, host_prot_new);
368     }
369     return true;
370 }
371 
372 abi_ulong task_unmapped_base;
373 abi_ulong elf_et_dyn_base;
374 abi_ulong mmap_next_start;
375 
376 /*
377  * Subroutine of mmap_find_vma, used when we have pre-allocated
378  * a chunk of guest address space.
379  */
380 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size,
381                                         abi_ulong align)
382 {
383     target_ulong ret;
384 
385     ret = page_find_range_empty(start, reserved_va, size, align);
386     if (ret == -1 && start > mmap_min_addr) {
387         /* Restart at the beginning of the address space. */
388         ret = page_find_range_empty(mmap_min_addr, start - 1, size, align);
389     }
390 
391     return ret;
392 }
393 
394 /*
395  * Find and reserve a free memory area of size 'size'. The search
396  * starts at 'start'.
397  * It must be called with mmap_lock() held.
398  * Return -1 if error.
399  */
400 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align)
401 {
402     int host_page_size = qemu_real_host_page_size();
403     void *ptr, *prev;
404     abi_ulong addr;
405     int wrapped, repeat;
406 
407     align = MAX(align, host_page_size);
408 
409     /* If 'start' == 0, then a default start address is used. */
410     if (start == 0) {
411         start = mmap_next_start;
412     } else {
413         start &= -host_page_size;
414     }
415     start = ROUND_UP(start, align);
416     size = ROUND_UP(size, host_page_size);
417 
418     if (reserved_va) {
419         return mmap_find_vma_reserved(start, size, align);
420     }
421 
422     addr = start;
423     wrapped = repeat = 0;
424     prev = 0;
425 
426     for (;; prev = ptr) {
427         /*
428          * Reserve needed memory area to avoid a race.
429          * It should be discarded using:
430          *  - mmap() with MAP_FIXED flag
431          *  - mremap() with MREMAP_FIXED flag
432          *  - shmat() with SHM_REMAP flag
433          */
434         ptr = mmap(g2h_untagged(addr), size, PROT_NONE,
435                    MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
436 
437         /* ENOMEM, if host address space has no memory */
438         if (ptr == MAP_FAILED) {
439             return (abi_ulong)-1;
440         }
441 
442         /*
443          * Count the number of sequential returns of the same address.
444          * This is used to modify the search algorithm below.
445          */
446         repeat = (ptr == prev ? repeat + 1 : 0);
447 
448         if (h2g_valid(ptr + size - 1)) {
449             addr = h2g(ptr);
450 
451             if ((addr & (align - 1)) == 0) {
452                 /* Success.  */
453                 if (start == mmap_next_start && addr >= task_unmapped_base) {
454                     mmap_next_start = addr + size;
455                 }
456                 return addr;
457             }
458 
459             /* The address is not properly aligned for the target.  */
460             switch (repeat) {
461             case 0:
462                 /*
463                  * Assume the result that the kernel gave us is the
464                  * first with enough free space, so start again at the
465                  * next higher target page.
466                  */
467                 addr = ROUND_UP(addr, align);
468                 break;
469             case 1:
470                 /*
471                  * Sometimes the kernel decides to perform the allocation
472                  * at the top end of memory instead.
473                  */
474                 addr &= -align;
475                 break;
476             case 2:
477                 /* Start over at low memory.  */
478                 addr = 0;
479                 break;
480             default:
481                 /* Fail.  This unaligned block must the last.  */
482                 addr = -1;
483                 break;
484             }
485         } else {
486             /*
487              * Since the result the kernel gave didn't fit, start
488              * again at low memory.  If any repetition, fail.
489              */
490             addr = (repeat ? -1 : 0);
491         }
492 
493         /* Unmap and try again.  */
494         munmap(ptr, size);
495 
496         /* ENOMEM if we checked the whole of the target address space.  */
497         if (addr == (abi_ulong)-1) {
498             return (abi_ulong)-1;
499         } else if (addr == 0) {
500             if (wrapped) {
501                 return (abi_ulong)-1;
502             }
503             wrapped = 1;
504             /*
505              * Don't actually use 0 when wrapping, instead indicate
506              * that we'd truly like an allocation in low memory.
507              */
508             addr = (mmap_min_addr > TARGET_PAGE_SIZE
509                      ? TARGET_PAGE_ALIGN(mmap_min_addr)
510                      : TARGET_PAGE_SIZE);
511         } else if (wrapped && addr >= start) {
512             return (abi_ulong)-1;
513         }
514     }
515 }
516 
517 /*
518  * Record a successful mmap within the user-exec interval tree.
519  */
520 static abi_long mmap_end(abi_ulong start, abi_ulong last,
521                          abi_ulong passthrough_start,
522                          abi_ulong passthrough_last,
523                          int flags, int page_flags)
524 {
525     if (flags & MAP_ANONYMOUS) {
526         page_flags |= PAGE_ANON;
527     }
528     page_flags |= PAGE_RESET;
529     if (passthrough_start > passthrough_last) {
530         page_set_flags(start, last, page_flags);
531     } else {
532         if (start < passthrough_start) {
533             page_set_flags(start, passthrough_start - 1, page_flags);
534         }
535         page_set_flags(passthrough_start, passthrough_last,
536                        page_flags | PAGE_PASSTHROUGH);
537         if (passthrough_last < last) {
538             page_set_flags(passthrough_last + 1, last, page_flags);
539         }
540     }
541     shm_region_rm_complete(start, last);
542     trace_target_mmap_complete(start);
543     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
544         FILE *f = qemu_log_trylock();
545         if (f) {
546             fprintf(f, "page layout changed following mmap\n");
547             page_dump(f);
548             qemu_log_unlock(f);
549         }
550     }
551     return start;
552 }
553 
554 /*
555  * Special case host page size == target page size,
556  * where there are no edge conditions.
557  */
558 static abi_long mmap_h_eq_g(abi_ulong start, abi_ulong len,
559                             int host_prot, int flags, int page_flags,
560                             int fd, off_t offset)
561 {
562     void *p, *want_p = g2h_untagged(start);
563     abi_ulong last;
564 
565     p = mmap(want_p, len, host_prot, flags, fd, offset);
566     if (p == MAP_FAILED) {
567         return -1;
568     }
569     /* If the host kernel does not support MAP_FIXED_NOREPLACE, emulate. */
570     if ((flags & MAP_FIXED_NOREPLACE) && p != want_p) {
571         do_munmap(p, len);
572         errno = EEXIST;
573         return -1;
574     }
575 
576     start = h2g(p);
577     last = start + len - 1;
578     return mmap_end(start, last, start, last, flags, page_flags);
579 }
580 
581 /*
582  * Special case host page size < target page size.
583  *
584  * The two special cases are increased guest alignment, and mapping
585  * past the end of a file.
586  *
587  * When mapping files into a memory area larger than the file,
588  * accesses to pages beyond the file size will cause a SIGBUS.
589  *
590  * For example, if mmaping a file of 100 bytes on a host with 4K
591  * pages emulating a target with 8K pages, the target expects to
592  * be able to access the first 8K. But the host will trap us on
593  * any access beyond 4K.
594  *
595  * When emulating a target with a larger page-size than the hosts,
596  * we may need to truncate file maps at EOF and add extra anonymous
597  * pages up to the targets page boundary.
598  *
599  * This workaround only works for files that do not change.
600  * If the file is later extended (e.g. ftruncate), the SIGBUS
601  * vanishes and the proper behaviour is that changes within the
602  * anon page should be reflected in the file.
603  *
604  * However, this case is rather common with executable images,
605  * so the workaround is important for even trivial tests, whereas
606  * the mmap of of a file being extended is less common.
607  */
608 static abi_long mmap_h_lt_g(abi_ulong start, abi_ulong len, int host_prot,
609                             int mmap_flags, int page_flags, int fd,
610                             off_t offset, int host_page_size)
611 {
612     void *p, *want_p = g2h_untagged(start);
613     off_t fileend_adj = 0;
614     int flags = mmap_flags;
615     abi_ulong last, pass_last;
616 
617     if (!(flags & MAP_ANONYMOUS)) {
618         struct stat sb;
619 
620         if (fstat(fd, &sb) == -1) {
621             return -1;
622         }
623         if (offset >= sb.st_size) {
624             /*
625              * The entire map is beyond the end of the file.
626              * Transform it to an anonymous mapping.
627              */
628             flags |= MAP_ANONYMOUS;
629             fd = -1;
630             offset = 0;
631         } else if (offset + len > sb.st_size) {
632             /*
633              * A portion of the map is beyond the end of the file.
634              * Truncate the file portion of the allocation.
635              */
636             fileend_adj = offset + len - sb.st_size;
637         }
638     }
639 
640     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
641         if (fileend_adj) {
642             p = mmap(want_p, len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
643         } else {
644             p = mmap(want_p, len, host_prot, flags, fd, offset);
645         }
646         if (p != want_p) {
647             if (p != MAP_FAILED) {
648                 /* Host does not support MAP_FIXED_NOREPLACE: emulate. */
649                 do_munmap(p, len);
650                 errno = EEXIST;
651             }
652             return -1;
653         }
654 
655         if (fileend_adj) {
656             void *t = mmap(p, len - fileend_adj, host_prot,
657                            (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED,
658                            fd, offset);
659 
660             if (t == MAP_FAILED) {
661                 int save_errno = errno;
662 
663                 /*
664                  * We failed a map over the top of the successful anonymous
665                  * mapping above. The only failure mode is running out of VMAs,
666                  * and there's nothing that we can do to detect that earlier.
667                  * If we have replaced an existing mapping with MAP_FIXED,
668                  * then we cannot properly recover.  It's a coin toss whether
669                  * it would be better to exit or continue here.
670                  */
671                 if (!(flags & MAP_FIXED_NOREPLACE) &&
672                     !page_check_range_empty(start, start + len - 1)) {
673                     qemu_log("QEMU target_mmap late failure: %s",
674                              strerror(save_errno));
675                 }
676 
677                 do_munmap(want_p, len);
678                 errno = save_errno;
679                 return -1;
680             }
681         }
682     } else {
683         size_t host_len, part_len;
684 
685         /*
686          * Take care to align the host memory.  Perform a larger anonymous
687          * allocation and extract the aligned portion.  Remap the file on
688          * top of that.
689          */
690         host_len = len + TARGET_PAGE_SIZE - host_page_size;
691         p = mmap(want_p, host_len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
692         if (p == MAP_FAILED) {
693             return -1;
694         }
695 
696         part_len = (uintptr_t)p & (TARGET_PAGE_SIZE - 1);
697         if (part_len) {
698             part_len = TARGET_PAGE_SIZE - part_len;
699             do_munmap(p, part_len);
700             p += part_len;
701             host_len -= part_len;
702         }
703         if (len < host_len) {
704             do_munmap(p + len, host_len - len);
705         }
706 
707         if (!(flags & MAP_ANONYMOUS)) {
708             void *t = mmap(p, len - fileend_adj, host_prot,
709                            flags | MAP_FIXED, fd, offset);
710 
711             if (t == MAP_FAILED) {
712                 int save_errno = errno;
713                 do_munmap(p, len);
714                 errno = save_errno;
715                 return -1;
716             }
717         }
718 
719         start = h2g(p);
720     }
721 
722     last = start + len - 1;
723     if (fileend_adj) {
724         pass_last = ROUND_UP(last - fileend_adj, host_page_size) - 1;
725     } else {
726         pass_last = last;
727     }
728     return mmap_end(start, last, start, pass_last, mmap_flags, page_flags);
729 }
730 
731 /*
732  * Special case host page size > target page size.
733  *
734  * The two special cases are address and file offsets that are valid
735  * for the guest that cannot be directly represented by the host.
736  */
737 static abi_long mmap_h_gt_g(abi_ulong start, abi_ulong len,
738                             int target_prot, int host_prot,
739                             int flags, int page_flags, int fd,
740                             off_t offset, int host_page_size)
741 {
742     void *p, *want_p = g2h_untagged(start);
743     off_t host_offset = offset & -host_page_size;
744     abi_ulong last, real_start, real_last;
745     bool misaligned_offset = false;
746     size_t host_len;
747 
748     if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
749         /*
750          * Adjust the offset to something representable on the host.
751          */
752         host_len = len + offset - host_offset;
753         p = mmap(want_p, host_len, host_prot, flags, fd, host_offset);
754         if (p == MAP_FAILED) {
755             return -1;
756         }
757 
758         /* Update start to the file position at offset. */
759         p += offset - host_offset;
760 
761         start = h2g(p);
762         last = start + len - 1;
763         return mmap_end(start, last, start, last, flags, page_flags);
764     }
765 
766     if (!(flags & MAP_ANONYMOUS)) {
767         misaligned_offset = (start ^ offset) & (host_page_size - 1);
768 
769         /*
770          * The fallback for misalignment is a private mapping + read.
771          * This carries none of semantics required of MAP_SHARED.
772          */
773         if (misaligned_offset && (flags & MAP_TYPE) != MAP_PRIVATE) {
774             errno = EINVAL;
775             return -1;
776         }
777     }
778 
779     last = start + len - 1;
780     real_start = start & -host_page_size;
781     real_last = ROUND_UP(last, host_page_size) - 1;
782 
783     /*
784      * Handle the start and end of the mapping.
785      */
786     if (real_start < start) {
787         abi_ulong real_page_last = real_start + host_page_size - 1;
788         if (last <= real_page_last) {
789             /* Entire allocation a subset of one host page. */
790             if (!mmap_frag(real_start, start, last, target_prot,
791                            flags, fd, offset)) {
792                 return -1;
793             }
794             return mmap_end(start, last, -1, 0, flags, page_flags);
795         }
796 
797         if (!mmap_frag(real_start, start, real_page_last, target_prot,
798                        flags, fd, offset)) {
799             return -1;
800         }
801         real_start = real_page_last + 1;
802     }
803 
804     if (last < real_last) {
805         abi_ulong real_page_start = real_last - host_page_size + 1;
806         if (!mmap_frag(real_page_start, real_page_start, last,
807                        target_prot, flags, fd,
808                        offset + real_page_start - start)) {
809             return -1;
810         }
811         real_last = real_page_start - 1;
812     }
813 
814     if (real_start > real_last) {
815         return mmap_end(start, last, -1, 0, flags, page_flags);
816     }
817 
818     /*
819      * Handle the middle of the mapping.
820      */
821 
822     host_len = real_last - real_start + 1;
823     want_p += real_start - start;
824 
825     if (flags & MAP_ANONYMOUS) {
826         p = mmap(want_p, host_len, host_prot, flags, -1, 0);
827     } else if (!misaligned_offset) {
828         p = mmap(want_p, host_len, host_prot, flags, fd,
829                  offset + real_start - start);
830     } else {
831         p = mmap(want_p, host_len, host_prot | PROT_WRITE,
832                  flags | MAP_ANONYMOUS, -1, 0);
833     }
834     if (p != want_p) {
835         if (p != MAP_FAILED) {
836             do_munmap(p, host_len);
837             errno = EEXIST;
838         }
839         return -1;
840     }
841 
842     if (misaligned_offset) {
843         /* TODO: The read could be short. */
844         if (pread(fd, p, host_len, offset + real_start - start) != host_len) {
845             do_munmap(p, host_len);
846             return -1;
847         }
848         if (!(host_prot & PROT_WRITE)) {
849             mprotect(p, host_len, host_prot);
850         }
851     }
852 
853     return mmap_end(start, last, -1, 0, flags, page_flags);
854 }
855 
856 static abi_long target_mmap__locked(abi_ulong start, abi_ulong len,
857                                     int target_prot, int flags, int page_flags,
858                                     int fd, off_t offset)
859 {
860     int host_page_size = qemu_real_host_page_size();
861     int host_prot;
862 
863     /*
864      * For reserved_va, we are in full control of the allocation.
865      * Find a suitable hole and convert to MAP_FIXED.
866      */
867     if (reserved_va) {
868         if (flags & MAP_FIXED_NOREPLACE) {
869             /* Validate that the chosen range is empty. */
870             if (!page_check_range_empty(start, start + len - 1)) {
871                 errno = EEXIST;
872                 return -1;
873             }
874             flags = (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED;
875         } else if (!(flags & MAP_FIXED)) {
876             abi_ulong real_start = start & -host_page_size;
877             off_t host_offset = offset & -host_page_size;
878             size_t real_len = len + offset - host_offset;
879             abi_ulong align = MAX(host_page_size, TARGET_PAGE_SIZE);
880 
881             start = mmap_find_vma(real_start, real_len, align);
882             if (start == (abi_ulong)-1) {
883                 errno = ENOMEM;
884                 return -1;
885             }
886             start += offset - host_offset;
887             flags |= MAP_FIXED;
888         }
889     }
890 
891     host_prot = target_to_host_prot(target_prot);
892 
893     if (host_page_size == TARGET_PAGE_SIZE) {
894         return mmap_h_eq_g(start, len, host_prot, flags,
895                            page_flags, fd, offset);
896     } else if (host_page_size < TARGET_PAGE_SIZE) {
897         return mmap_h_lt_g(start, len, host_prot, flags,
898                            page_flags, fd, offset, host_page_size);
899     } else {
900         return mmap_h_gt_g(start, len, target_prot, host_prot, flags,
901                            page_flags, fd, offset, host_page_size);
902     }
903 }
904 
905 /* NOTE: all the constants are the HOST ones */
906 abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot,
907                      int flags, int fd, off_t offset)
908 {
909     abi_long ret;
910     int page_flags;
911 
912     trace_target_mmap(start, len, target_prot, flags, fd, offset);
913 
914     if (!len) {
915         errno = EINVAL;
916         return -1;
917     }
918 
919     page_flags = validate_prot_to_pageflags(target_prot);
920     if (!page_flags) {
921         errno = EINVAL;
922         return -1;
923     }
924 
925     /* Also check for overflows... */
926     len = TARGET_PAGE_ALIGN(len);
927     if (!len || len != (size_t)len) {
928         errno = ENOMEM;
929         return -1;
930     }
931 
932     if (offset & ~TARGET_PAGE_MASK) {
933         errno = EINVAL;
934         return -1;
935     }
936     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
937         if (start & ~TARGET_PAGE_MASK) {
938             errno = EINVAL;
939             return -1;
940         }
941         if (!guest_range_valid_untagged(start, len)) {
942             errno = ENOMEM;
943             return -1;
944         }
945     }
946 
947     mmap_lock();
948 
949     ret = target_mmap__locked(start, len, target_prot, flags,
950                               page_flags, fd, offset);
951 
952     mmap_unlock();
953 
954     /*
955      * If we're mapping shared memory, ensure we generate code for parallel
956      * execution and flush old translations.  This will work up to the level
957      * supported by the host -- anything that requires EXCP_ATOMIC will not
958      * be atomic with respect to an external process.
959      */
960     if (ret != -1 && (flags & MAP_TYPE) != MAP_PRIVATE) {
961         CPUState *cpu = thread_cpu;
962         if (!(cpu->tcg_cflags & CF_PARALLEL)) {
963             cpu->tcg_cflags |= CF_PARALLEL;
964             tb_flush(cpu);
965         }
966     }
967 
968     return ret;
969 }
970 
971 static int mmap_reserve_or_unmap(abi_ulong start, abi_ulong len)
972 {
973     int host_page_size = qemu_real_host_page_size();
974     abi_ulong real_start;
975     abi_ulong real_last;
976     abi_ulong real_len;
977     abi_ulong last;
978     abi_ulong a;
979     void *host_start;
980     int prot;
981 
982     last = start + len - 1;
983     real_start = start & -host_page_size;
984     real_last = ROUND_UP(last, host_page_size) - 1;
985 
986     /*
987      * If guest pages remain on the first or last host pages,
988      * adjust the deallocation to retain those guest pages.
989      * The single page special case is required for the last page,
990      * lest real_start overflow to zero.
991      */
992     if (real_last - real_start < host_page_size) {
993         prot = 0;
994         for (a = real_start; a < start; a += TARGET_PAGE_SIZE) {
995             prot |= page_get_flags(a);
996         }
997         for (a = last; a < real_last; a += TARGET_PAGE_SIZE) {
998             prot |= page_get_flags(a + 1);
999         }
1000         if (prot != 0) {
1001             return 0;
1002         }
1003     } else {
1004         for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) {
1005             prot |= page_get_flags(a);
1006         }
1007         if (prot != 0) {
1008             real_start += host_page_size;
1009         }
1010 
1011         for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) {
1012             prot |= page_get_flags(a + 1);
1013         }
1014         if (prot != 0) {
1015             real_last -= host_page_size;
1016         }
1017 
1018         if (real_last < real_start) {
1019             return 0;
1020         }
1021     }
1022 
1023     real_len = real_last - real_start + 1;
1024     host_start = g2h_untagged(real_start);
1025 
1026     return do_munmap(host_start, real_len);
1027 }
1028 
1029 int target_munmap(abi_ulong start, abi_ulong len)
1030 {
1031     int ret;
1032 
1033     trace_target_munmap(start, len);
1034 
1035     if (start & ~TARGET_PAGE_MASK) {
1036         errno = EINVAL;
1037         return -1;
1038     }
1039     len = TARGET_PAGE_ALIGN(len);
1040     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1041         errno = EINVAL;
1042         return -1;
1043     }
1044 
1045     mmap_lock();
1046     ret = mmap_reserve_or_unmap(start, len);
1047     if (likely(ret == 0)) {
1048         page_set_flags(start, start + len - 1, 0);
1049         shm_region_rm_complete(start, start + len - 1);
1050     }
1051     mmap_unlock();
1052 
1053     return ret;
1054 }
1055 
1056 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
1057                        abi_ulong new_size, unsigned long flags,
1058                        abi_ulong new_addr)
1059 {
1060     int prot;
1061     void *host_addr;
1062 
1063     if (!guest_range_valid_untagged(old_addr, old_size) ||
1064         ((flags & MREMAP_FIXED) &&
1065          !guest_range_valid_untagged(new_addr, new_size)) ||
1066         ((flags & MREMAP_MAYMOVE) == 0 &&
1067          !guest_range_valid_untagged(old_addr, new_size))) {
1068         errno = ENOMEM;
1069         return -1;
1070     }
1071 
1072     mmap_lock();
1073 
1074     if (flags & MREMAP_FIXED) {
1075         host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1076                            flags, g2h_untagged(new_addr));
1077 
1078         if (reserved_va && host_addr != MAP_FAILED) {
1079             /*
1080              * If new and old addresses overlap then the above mremap will
1081              * already have failed with EINVAL.
1082              */
1083             mmap_reserve_or_unmap(old_addr, old_size);
1084         }
1085     } else if (flags & MREMAP_MAYMOVE) {
1086         abi_ulong mmap_start;
1087 
1088         mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE);
1089 
1090         if (mmap_start == -1) {
1091             errno = ENOMEM;
1092             host_addr = MAP_FAILED;
1093         } else {
1094             host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1095                                flags | MREMAP_FIXED,
1096                                g2h_untagged(mmap_start));
1097             if (reserved_va) {
1098                 mmap_reserve_or_unmap(old_addr, old_size);
1099             }
1100         }
1101     } else {
1102         int page_flags = 0;
1103         if (reserved_va && old_size < new_size) {
1104             abi_ulong addr;
1105             for (addr = old_addr + old_size;
1106                  addr < old_addr + new_size;
1107                  addr++) {
1108                 page_flags |= page_get_flags(addr);
1109             }
1110         }
1111         if (page_flags == 0) {
1112             host_addr = mremap(g2h_untagged(old_addr),
1113                                old_size, new_size, flags);
1114 
1115             if (host_addr != MAP_FAILED) {
1116                 /* Check if address fits target address space */
1117                 if (!guest_range_valid_untagged(h2g(host_addr), new_size)) {
1118                     /* Revert mremap() changes */
1119                     host_addr = mremap(g2h_untagged(old_addr),
1120                                        new_size, old_size, flags);
1121                     errno = ENOMEM;
1122                     host_addr = MAP_FAILED;
1123                 } else if (reserved_va && old_size > new_size) {
1124                     mmap_reserve_or_unmap(old_addr + old_size,
1125                                           old_size - new_size);
1126                 }
1127             }
1128         } else {
1129             errno = ENOMEM;
1130             host_addr = MAP_FAILED;
1131         }
1132     }
1133 
1134     if (host_addr == MAP_FAILED) {
1135         new_addr = -1;
1136     } else {
1137         new_addr = h2g(host_addr);
1138         prot = page_get_flags(old_addr);
1139         page_set_flags(old_addr, old_addr + old_size - 1, 0);
1140         shm_region_rm_complete(old_addr, old_addr + old_size - 1);
1141         page_set_flags(new_addr, new_addr + new_size - 1,
1142                        prot | PAGE_VALID | PAGE_RESET);
1143         shm_region_rm_complete(new_addr, new_addr + new_size - 1);
1144     }
1145     mmap_unlock();
1146     return new_addr;
1147 }
1148 
1149 abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice)
1150 {
1151     abi_ulong len;
1152     int ret = 0;
1153 
1154     if (start & ~TARGET_PAGE_MASK) {
1155         return -TARGET_EINVAL;
1156     }
1157     if (len_in == 0) {
1158         return 0;
1159     }
1160     len = TARGET_PAGE_ALIGN(len_in);
1161     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1162         return -TARGET_EINVAL;
1163     }
1164 
1165     /* Translate for some architectures which have different MADV_xxx values */
1166     switch (advice) {
1167     case TARGET_MADV_DONTNEED:      /* alpha */
1168         advice = MADV_DONTNEED;
1169         break;
1170     case TARGET_MADV_WIPEONFORK:    /* parisc */
1171         advice = MADV_WIPEONFORK;
1172         break;
1173     case TARGET_MADV_KEEPONFORK:    /* parisc */
1174         advice = MADV_KEEPONFORK;
1175         break;
1176     /* we do not care about the other MADV_xxx values yet */
1177     }
1178 
1179     /*
1180      * Most advice values are hints, so ignoring and returning success is ok.
1181      *
1182      * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and
1183      * MADV_KEEPONFORK are not hints and need to be emulated.
1184      *
1185      * A straight passthrough for those may not be safe because qemu sometimes
1186      * turns private file-backed mappings into anonymous mappings.
1187      * If all guest pages have PAGE_PASSTHROUGH set, mappings have the
1188      * same semantics for the host as for the guest.
1189      *
1190      * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and
1191      * return failure if not.
1192      *
1193      * MADV_DONTNEED is passed through as well, if possible.
1194      * If passthrough isn't possible, we nevertheless (wrongly!) return
1195      * success, which is broken but some userspace programs fail to work
1196      * otherwise. Completely implementing such emulation is quite complicated
1197      * though.
1198      */
1199     mmap_lock();
1200     switch (advice) {
1201     case MADV_WIPEONFORK:
1202     case MADV_KEEPONFORK:
1203         ret = -EINVAL;
1204         /* fall through */
1205     case MADV_DONTNEED:
1206         if (page_check_range(start, len, PAGE_PASSTHROUGH)) {
1207             ret = get_errno(madvise(g2h_untagged(start), len, advice));
1208             if ((advice == MADV_DONTNEED) && (ret == 0)) {
1209                 page_reset_target_data(start, start + len - 1);
1210             }
1211         }
1212     }
1213     mmap_unlock();
1214 
1215     return ret;
1216 }
1217 
1218 #ifndef TARGET_FORCE_SHMLBA
1219 /*
1220  * For most architectures, SHMLBA is the same as the page size;
1221  * some architectures have larger values, in which case they should
1222  * define TARGET_FORCE_SHMLBA and provide a target_shmlba() function.
1223  * This corresponds to the kernel arch code defining __ARCH_FORCE_SHMLBA
1224  * and defining its own value for SHMLBA.
1225  *
1226  * The kernel also permits SHMLBA to be set by the architecture to a
1227  * value larger than the page size without setting __ARCH_FORCE_SHMLBA;
1228  * this means that addresses are rounded to the large size if
1229  * SHM_RND is set but addresses not aligned to that size are not rejected
1230  * as long as they are at least page-aligned. Since the only architecture
1231  * which uses this is ia64 this code doesn't provide for that oddity.
1232  */
1233 static inline abi_ulong target_shmlba(CPUArchState *cpu_env)
1234 {
1235     return TARGET_PAGE_SIZE;
1236 }
1237 #endif
1238 
1239 #if defined(__arm__) || defined(__mips__) || defined(__sparc__)
1240 #define HOST_FORCE_SHMLBA 1
1241 #else
1242 #define HOST_FORCE_SHMLBA 0
1243 #endif
1244 
1245 abi_ulong target_shmat(CPUArchState *cpu_env, int shmid,
1246                        abi_ulong shmaddr, int shmflg)
1247 {
1248     CPUState *cpu = env_cpu(cpu_env);
1249     struct shmid_ds shm_info;
1250     int ret;
1251     int h_pagesize;
1252     int t_shmlba, h_shmlba, m_shmlba;
1253     size_t t_len, h_len, m_len;
1254 
1255     /* shmat pointers are always untagged */
1256 
1257     /*
1258      * Because we can't use host shmat() unless the address is sufficiently
1259      * aligned for the host, we'll need to check both.
1260      * TODO: Could be fixed with softmmu.
1261      */
1262     t_shmlba = target_shmlba(cpu_env);
1263     h_pagesize = qemu_real_host_page_size();
1264     h_shmlba = (HOST_FORCE_SHMLBA ? SHMLBA : h_pagesize);
1265     m_shmlba = MAX(t_shmlba, h_shmlba);
1266 
1267     if (shmaddr) {
1268         if (shmaddr & (m_shmlba - 1)) {
1269             if (shmflg & SHM_RND) {
1270                 /*
1271                  * The guest is allowing the kernel to round the address.
1272                  * Assume that the guest is ok with us rounding to the
1273                  * host required alignment too.  Anyway if we don't, we'll
1274                  * get an error from the kernel.
1275                  */
1276                 shmaddr &= ~(m_shmlba - 1);
1277                 if (shmaddr == 0 && (shmflg & SHM_REMAP)) {
1278                     return -TARGET_EINVAL;
1279                 }
1280             } else {
1281                 int require = TARGET_PAGE_SIZE;
1282 #ifdef TARGET_FORCE_SHMLBA
1283                 require = t_shmlba;
1284 #endif
1285                 /*
1286                  * Include host required alignment, as otherwise we cannot
1287                  * use host shmat at all.
1288                  */
1289                 require = MAX(require, h_shmlba);
1290                 if (shmaddr & (require - 1)) {
1291                     return -TARGET_EINVAL;
1292                 }
1293             }
1294         }
1295     } else {
1296         if (shmflg & SHM_REMAP) {
1297             return -TARGET_EINVAL;
1298         }
1299     }
1300     /* All rounding now manually concluded. */
1301     shmflg &= ~SHM_RND;
1302 
1303     /* Find out the length of the shared memory segment. */
1304     ret = get_errno(shmctl(shmid, IPC_STAT, &shm_info));
1305     if (is_error(ret)) {
1306         /* can't get length, bail out */
1307         return ret;
1308     }
1309     t_len = TARGET_PAGE_ALIGN(shm_info.shm_segsz);
1310     h_len = ROUND_UP(shm_info.shm_segsz, h_pagesize);
1311     m_len = MAX(t_len, h_len);
1312 
1313     if (!guest_range_valid_untagged(shmaddr, m_len)) {
1314         return -TARGET_EINVAL;
1315     }
1316 
1317     WITH_MMAP_LOCK_GUARD() {
1318         bool mapped = false;
1319         void *want, *test;
1320         abi_ulong last;
1321 
1322         if (!shmaddr) {
1323             shmaddr = mmap_find_vma(0, m_len, m_shmlba);
1324             if (shmaddr == -1) {
1325                 return -TARGET_ENOMEM;
1326             }
1327             mapped = !reserved_va;
1328         } else if (shmflg & SHM_REMAP) {
1329             /*
1330              * If host page size > target page size, the host shmat may map
1331              * more memory than the guest expects.  Reject a mapping that
1332              * would replace memory in the unexpected gap.
1333              * TODO: Could be fixed with softmmu.
1334              */
1335             if (t_len < h_len &&
1336                 !page_check_range_empty(shmaddr + t_len,
1337                                         shmaddr + h_len - 1)) {
1338                 return -TARGET_EINVAL;
1339             }
1340         } else {
1341             if (!page_check_range_empty(shmaddr, shmaddr + m_len - 1)) {
1342                 return -TARGET_EINVAL;
1343             }
1344         }
1345 
1346         /* All placement is now complete. */
1347         want = (void *)g2h_untagged(shmaddr);
1348 
1349         /*
1350          * Map anonymous pages across the entire range, then remap with
1351          * the shared memory.  This is required for a number of corner
1352          * cases for which host and guest page sizes differ.
1353          */
1354         if (h_len != t_len) {
1355             int mmap_p = PROT_READ | (shmflg & SHM_RDONLY ? 0 : PROT_WRITE);
1356             int mmap_f = MAP_PRIVATE | MAP_ANONYMOUS
1357                        | (reserved_va || (shmflg & SHM_REMAP)
1358                           ? MAP_FIXED : MAP_FIXED_NOREPLACE);
1359 
1360             test = mmap(want, m_len, mmap_p, mmap_f, -1, 0);
1361             if (unlikely(test != want)) {
1362                 /* shmat returns EINVAL not EEXIST like mmap. */
1363                 ret = (test == MAP_FAILED && errno != EEXIST
1364                        ? get_errno(-1) : -TARGET_EINVAL);
1365                 if (mapped) {
1366                     do_munmap(want, m_len);
1367                 }
1368                 return ret;
1369             }
1370             mapped = true;
1371         }
1372 
1373         if (reserved_va || mapped) {
1374             shmflg |= SHM_REMAP;
1375         }
1376         test = shmat(shmid, want, shmflg);
1377         if (test == MAP_FAILED) {
1378             ret = get_errno(-1);
1379             if (mapped) {
1380                 do_munmap(want, m_len);
1381             }
1382             return ret;
1383         }
1384         assert(test == want);
1385 
1386         last = shmaddr + m_len - 1;
1387         page_set_flags(shmaddr, last,
1388                        PAGE_VALID | PAGE_RESET | PAGE_READ |
1389                        (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE) |
1390                        (shmflg & SHM_EXEC ? PAGE_EXEC : 0));
1391 
1392         shm_region_rm_complete(shmaddr, last);
1393         shm_region_add(shmaddr, last);
1394     }
1395 
1396     /*
1397      * We're mapping shared memory, so ensure we generate code for parallel
1398      * execution and flush old translations.  This will work up to the level
1399      * supported by the host -- anything that requires EXCP_ATOMIC will not
1400      * be atomic with respect to an external process.
1401      */
1402     if (!(cpu->tcg_cflags & CF_PARALLEL)) {
1403         cpu->tcg_cflags |= CF_PARALLEL;
1404         tb_flush(cpu);
1405     }
1406 
1407     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
1408         FILE *f = qemu_log_trylock();
1409         if (f) {
1410             fprintf(f, "page layout changed following shmat\n");
1411             page_dump(f);
1412             qemu_log_unlock(f);
1413         }
1414     }
1415     return shmaddr;
1416 }
1417 
1418 abi_long target_shmdt(abi_ulong shmaddr)
1419 {
1420     abi_long rv;
1421 
1422     /* shmdt pointers are always untagged */
1423 
1424     WITH_MMAP_LOCK_GUARD() {
1425         abi_ulong last = shm_region_find(shmaddr);
1426         if (last == 0) {
1427             return -TARGET_EINVAL;
1428         }
1429 
1430         rv = get_errno(shmdt(g2h_untagged(shmaddr)));
1431         if (rv == 0) {
1432             abi_ulong size = last - shmaddr + 1;
1433 
1434             page_set_flags(shmaddr, last, 0);
1435             shm_region_rm_complete(shmaddr, last);
1436             mmap_reserve_or_unmap(shmaddr, size);
1437         }
1438     }
1439     return rv;
1440 }
1441