xref: /openbmc/qemu/linux-user/mmap.c (revision 2fda0e77)
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 "exec/page-protection.h"
24 #include "qemu.h"
25 #include "user-internals.h"
26 #include "user-mmap.h"
27 #include "target_mman.h"
28 #include "qemu/interval-tree.h"
29 
30 #ifdef TARGET_ARM
31 #include "target/arm/cpu-features.h"
32 #endif
33 
34 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
35 static __thread int mmap_lock_count;
36 
37 void mmap_lock(void)
38 {
39     if (mmap_lock_count++ == 0) {
40         pthread_mutex_lock(&mmap_mutex);
41     }
42 }
43 
44 void mmap_unlock(void)
45 {
46     assert(mmap_lock_count > 0);
47     if (--mmap_lock_count == 0) {
48         pthread_mutex_unlock(&mmap_mutex);
49     }
50 }
51 
52 bool have_mmap_lock(void)
53 {
54     return mmap_lock_count > 0 ? true : false;
55 }
56 
57 /* Grab lock to make sure things are in a consistent state after fork().  */
58 void mmap_fork_start(void)
59 {
60     if (mmap_lock_count)
61         abort();
62     pthread_mutex_lock(&mmap_mutex);
63 }
64 
65 void mmap_fork_end(int child)
66 {
67     if (child) {
68         pthread_mutex_init(&mmap_mutex, NULL);
69     } else {
70         pthread_mutex_unlock(&mmap_mutex);
71     }
72 }
73 
74 /* Protected by mmap_lock. */
75 static IntervalTreeRoot shm_regions;
76 
77 static void shm_region_add(abi_ptr start, abi_ptr last)
78 {
79     IntervalTreeNode *i = g_new0(IntervalTreeNode, 1);
80 
81     i->start = start;
82     i->last = last;
83     interval_tree_insert(i, &shm_regions);
84 }
85 
86 static abi_ptr shm_region_find(abi_ptr start)
87 {
88     IntervalTreeNode *i;
89 
90     for (i = interval_tree_iter_first(&shm_regions, start, start); i;
91          i = interval_tree_iter_next(i, start, start)) {
92         if (i->start == start) {
93             return i->last;
94         }
95     }
96     return 0;
97 }
98 
99 static void shm_region_rm_complete(abi_ptr start, abi_ptr last)
100 {
101     IntervalTreeNode *i, *n;
102 
103     for (i = interval_tree_iter_first(&shm_regions, start, last); i; i = n) {
104         n = interval_tree_iter_next(i, start, last);
105         if (i->start >= start && i->last <= last) {
106             interval_tree_remove(i, &shm_regions);
107             g_free(i);
108         }
109     }
110 }
111 
112 /*
113  * Validate target prot bitmask.
114  * Return the prot bitmask for the host in *HOST_PROT.
115  * Return 0 if the target prot bitmask is invalid, otherwise
116  * the internal qemu page_flags (which will include PAGE_VALID).
117  */
118 static int validate_prot_to_pageflags(int prot)
119 {
120     int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM;
121     int page_flags = (prot & PAGE_RWX) | PAGE_VALID;
122 
123 #ifdef TARGET_AARCH64
124     {
125         ARMCPU *cpu = ARM_CPU(thread_cpu);
126 
127         /*
128          * The PROT_BTI bit is only accepted if the cpu supports the feature.
129          * Since this is the unusual case, don't bother checking unless
130          * the bit has been requested.  If set and valid, record the bit
131          * within QEMU's page_flags.
132          */
133         if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) {
134             valid |= TARGET_PROT_BTI;
135             page_flags |= PAGE_BTI;
136         }
137         /* Similarly for the PROT_MTE bit. */
138         if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) {
139             valid |= TARGET_PROT_MTE;
140             page_flags |= PAGE_MTE;
141         }
142     }
143 #elif defined(TARGET_HPPA)
144     valid |= PROT_GROWSDOWN | PROT_GROWSUP;
145 #endif
146 
147     return prot & ~valid ? 0 : page_flags;
148 }
149 
150 /*
151  * For the host, we need not pass anything except read/write/exec.
152  * While PROT_SEM is allowed by all hosts, it is also ignored, so
153  * don't bother transforming guest bit to host bit.  Any other
154  * target-specific prot bits will not be understood by the host
155  * and will need to be encoded into page_flags for qemu emulation.
156  *
157  * Pages that are executable by the guest will never be executed
158  * by the host, but the host will need to be able to read them.
159  */
160 static int target_to_host_prot(int prot)
161 {
162     return (prot & (PROT_READ | PROT_WRITE)) |
163            (prot & PROT_EXEC ? PROT_READ : 0);
164 }
165 
166 /* NOTE: all the constants are the HOST ones, but addresses are target. */
167 int target_mprotect(abi_ulong start, abi_ulong len, int target_prot)
168 {
169     int host_page_size = qemu_real_host_page_size();
170     abi_ulong starts[3];
171     abi_ulong lens[3];
172     int prots[3];
173     abi_ulong host_start, host_last, last;
174     int prot1, ret, page_flags, nranges;
175 
176     trace_target_mprotect(start, len, target_prot);
177 
178     if ((start & ~TARGET_PAGE_MASK) != 0) {
179         return -TARGET_EINVAL;
180     }
181     page_flags = validate_prot_to_pageflags(target_prot);
182     if (!page_flags) {
183         return -TARGET_EINVAL;
184     }
185     if (len == 0) {
186         return 0;
187     }
188     len = TARGET_PAGE_ALIGN(len);
189     if (!guest_range_valid_untagged(start, len)) {
190         return -TARGET_ENOMEM;
191     }
192 
193     last = start + len - 1;
194     host_start = start & -host_page_size;
195     host_last = ROUND_UP(last, host_page_size) - 1;
196     nranges = 0;
197 
198     mmap_lock();
199 
200     if (host_last - host_start < host_page_size) {
201         /* Single host page contains all guest pages: sum the prot. */
202         prot1 = target_prot;
203         for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
204             prot1 |= page_get_flags(a);
205         }
206         for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
207             prot1 |= page_get_flags(a + 1);
208         }
209         starts[nranges] = host_start;
210         lens[nranges] = host_page_size;
211         prots[nranges] = prot1;
212         nranges++;
213     } else {
214         if (host_start < start) {
215             /* Host page contains more than one guest page: sum the prot. */
216             prot1 = target_prot;
217             for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
218                 prot1 |= page_get_flags(a);
219             }
220             /* If the resulting sum differs, create a new range. */
221             if (prot1 != target_prot) {
222                 starts[nranges] = host_start;
223                 lens[nranges] = host_page_size;
224                 prots[nranges] = prot1;
225                 nranges++;
226                 host_start += host_page_size;
227             }
228         }
229 
230         if (last < host_last) {
231             /* Host page contains more than one guest page: sum the prot. */
232             prot1 = target_prot;
233             for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
234                 prot1 |= page_get_flags(a + 1);
235             }
236             /* If the resulting sum differs, create a new range. */
237             if (prot1 != target_prot) {
238                 host_last -= host_page_size;
239                 starts[nranges] = host_last + 1;
240                 lens[nranges] = host_page_size;
241                 prots[nranges] = prot1;
242                 nranges++;
243             }
244         }
245 
246         /* Create a range for the middle, if any remains. */
247         if (host_start < host_last) {
248             starts[nranges] = host_start;
249             lens[nranges] = host_last - host_start + 1;
250             prots[nranges] = target_prot;
251             nranges++;
252         }
253     }
254 
255     for (int i = 0; i < nranges; ++i) {
256         ret = mprotect(g2h_untagged(starts[i]), lens[i],
257                        target_to_host_prot(prots[i]));
258         if (ret != 0) {
259             goto error;
260         }
261     }
262 
263     page_set_flags(start, last, page_flags);
264     ret = 0;
265 
266  error:
267     mmap_unlock();
268     return ret;
269 }
270 
271 /*
272  * Perform munmap on behalf of the target, with host parameters.
273  * If reserved_va, we must replace the memory reservation.
274  */
275 static int do_munmap(void *addr, size_t len)
276 {
277     if (reserved_va) {
278         void *ptr = mmap(addr, len, PROT_NONE,
279                          MAP_FIXED | MAP_ANONYMOUS
280                          | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
281         return ptr == addr ? 0 : -1;
282     }
283     return munmap(addr, len);
284 }
285 
286 /*
287  * Map an incomplete host page.
288  *
289  * Here be dragons.  This case will not work if there is an existing
290  * overlapping host page, which is file mapped, and for which the mapping
291  * is beyond the end of the file.  In that case, we will see SIGBUS when
292  * trying to write a portion of this page.
293  *
294  * FIXME: Work around this with a temporary signal handler and longjmp.
295  */
296 static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last,
297                       int prot, int flags, int fd, off_t offset)
298 {
299     int host_page_size = qemu_real_host_page_size();
300     abi_ulong real_last;
301     void *host_start;
302     int prot_old, prot_new;
303     int host_prot_old, host_prot_new;
304 
305     if (!(flags & MAP_ANONYMOUS)
306         && (flags & MAP_TYPE) == MAP_SHARED
307         && (prot & PROT_WRITE)) {
308         /*
309          * msync() won't work with the partial page, so we return an
310          * error if write is possible while it is a shared mapping.
311          */
312         errno = EINVAL;
313         return false;
314     }
315 
316     real_last = real_start + host_page_size - 1;
317     host_start = g2h_untagged(real_start);
318 
319     /* Get the protection of the target pages outside the mapping. */
320     prot_old = 0;
321     for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) {
322         prot_old |= page_get_flags(a);
323     }
324     for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) {
325         prot_old |= page_get_flags(a);
326     }
327 
328     if (prot_old == 0) {
329         /*
330          * Since !(prot_old & PAGE_VALID), there were no guest pages
331          * outside of the fragment we need to map.  Allocate a new host
332          * page to cover, discarding whatever else may have been present.
333          */
334         void *p = mmap(host_start, host_page_size,
335                        target_to_host_prot(prot),
336                        flags | MAP_ANONYMOUS, -1, 0);
337         if (p != host_start) {
338             if (p != MAP_FAILED) {
339                 do_munmap(p, host_page_size);
340                 errno = EEXIST;
341             }
342             return false;
343         }
344         prot_old = prot;
345     }
346     prot_new = prot | prot_old;
347 
348     host_prot_old = target_to_host_prot(prot_old);
349     host_prot_new = target_to_host_prot(prot_new);
350 
351     /* Adjust protection to be able to write. */
352     if (!(host_prot_old & PROT_WRITE)) {
353         host_prot_old |= PROT_WRITE;
354         mprotect(host_start, host_page_size, host_prot_old);
355     }
356 
357     /* Read or zero the new guest pages. */
358     if (flags & MAP_ANONYMOUS) {
359         memset(g2h_untagged(start), 0, last - start + 1);
360     } else {
361         if (pread(fd, g2h_untagged(start), last - start + 1, offset) == -1) {
362             return false;
363         }
364     }
365 
366     /* Put final protection */
367     if (host_prot_new != host_prot_old) {
368         mprotect(host_start, host_page_size, host_prot_new);
369     }
370     return true;
371 }
372 
373 abi_ulong task_unmapped_base;
374 abi_ulong elf_et_dyn_base;
375 abi_ulong mmap_next_start;
376 
377 /*
378  * Subroutine of mmap_find_vma, used when we have pre-allocated
379  * a chunk of guest address space.
380  */
381 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size,
382                                         abi_ulong align)
383 {
384     target_ulong ret;
385 
386     ret = page_find_range_empty(start, reserved_va, size, align);
387     if (ret == -1 && start > mmap_min_addr) {
388         /* Restart at the beginning of the address space. */
389         ret = page_find_range_empty(mmap_min_addr, start - 1, size, align);
390     }
391 
392     return ret;
393 }
394 
395 /*
396  * Find and reserve a free memory area of size 'size'. The search
397  * starts at 'start'.
398  * It must be called with mmap_lock() held.
399  * Return -1 if error.
400  */
401 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align)
402 {
403     int host_page_size = qemu_real_host_page_size();
404     void *ptr, *prev;
405     abi_ulong addr;
406     int wrapped, repeat;
407 
408     align = MAX(align, host_page_size);
409 
410     /* If 'start' == 0, then a default start address is used. */
411     if (start == 0) {
412         start = mmap_next_start;
413     } else {
414         start &= -host_page_size;
415     }
416     start = ROUND_UP(start, align);
417     size = ROUND_UP(size, host_page_size);
418 
419     if (reserved_va) {
420         return mmap_find_vma_reserved(start, size, align);
421     }
422 
423     addr = start;
424     wrapped = repeat = 0;
425     prev = 0;
426 
427     for (;; prev = ptr) {
428         /*
429          * Reserve needed memory area to avoid a race.
430          * It should be discarded using:
431          *  - mmap() with MAP_FIXED flag
432          *  - mremap() with MREMAP_FIXED flag
433          *  - shmat() with SHM_REMAP flag
434          */
435         ptr = mmap(g2h_untagged(addr), size, PROT_NONE,
436                    MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
437 
438         /* ENOMEM, if host address space has no memory */
439         if (ptr == MAP_FAILED) {
440             return (abi_ulong)-1;
441         }
442 
443         /*
444          * Count the number of sequential returns of the same address.
445          * This is used to modify the search algorithm below.
446          */
447         repeat = (ptr == prev ? repeat + 1 : 0);
448 
449         if (h2g_valid(ptr + size - 1)) {
450             addr = h2g(ptr);
451 
452             if ((addr & (align - 1)) == 0) {
453                 /* Success.  */
454                 if (start == mmap_next_start && addr >= task_unmapped_base) {
455                     mmap_next_start = addr + size;
456                 }
457                 return addr;
458             }
459 
460             /* The address is not properly aligned for the target.  */
461             switch (repeat) {
462             case 0:
463                 /*
464                  * Assume the result that the kernel gave us is the
465                  * first with enough free space, so start again at the
466                  * next higher target page.
467                  */
468                 addr = ROUND_UP(addr, align);
469                 break;
470             case 1:
471                 /*
472                  * Sometimes the kernel decides to perform the allocation
473                  * at the top end of memory instead.
474                  */
475                 addr &= -align;
476                 break;
477             case 2:
478                 /* Start over at low memory.  */
479                 addr = 0;
480                 break;
481             default:
482                 /* Fail.  This unaligned block must the last.  */
483                 addr = -1;
484                 break;
485             }
486         } else {
487             /*
488              * Since the result the kernel gave didn't fit, start
489              * again at low memory.  If any repetition, fail.
490              */
491             addr = (repeat ? -1 : 0);
492         }
493 
494         /* Unmap and try again.  */
495         munmap(ptr, size);
496 
497         /* ENOMEM if we checked the whole of the target address space.  */
498         if (addr == (abi_ulong)-1) {
499             return (abi_ulong)-1;
500         } else if (addr == 0) {
501             if (wrapped) {
502                 return (abi_ulong)-1;
503             }
504             wrapped = 1;
505             /*
506              * Don't actually use 0 when wrapping, instead indicate
507              * that we'd truly like an allocation in low memory.
508              */
509             addr = (mmap_min_addr > TARGET_PAGE_SIZE
510                      ? TARGET_PAGE_ALIGN(mmap_min_addr)
511                      : TARGET_PAGE_SIZE);
512         } else if (wrapped && addr >= start) {
513             return (abi_ulong)-1;
514         }
515     }
516 }
517 
518 /*
519  * Record a successful mmap within the user-exec interval tree.
520  */
521 static abi_long mmap_end(abi_ulong start, abi_ulong last,
522                          abi_ulong passthrough_start,
523                          abi_ulong passthrough_last,
524                          int flags, int page_flags)
525 {
526     if (flags & MAP_ANONYMOUS) {
527         page_flags |= PAGE_ANON;
528     }
529     page_flags |= PAGE_RESET;
530     if (passthrough_start > passthrough_last) {
531         page_set_flags(start, last, page_flags);
532     } else {
533         if (start < passthrough_start) {
534             page_set_flags(start, passthrough_start - 1, page_flags);
535         }
536         page_set_flags(passthrough_start, passthrough_last,
537                        page_flags | PAGE_PASSTHROUGH);
538         if (passthrough_last < last) {
539             page_set_flags(passthrough_last + 1, last, page_flags);
540         }
541     }
542     shm_region_rm_complete(start, last);
543     trace_target_mmap_complete(start);
544     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
545         FILE *f = qemu_log_trylock();
546         if (f) {
547             fprintf(f, "page layout changed following mmap\n");
548             page_dump(f);
549             qemu_log_unlock(f);
550         }
551     }
552     return start;
553 }
554 
555 /*
556  * Special case host page size == target page size,
557  * where there are no edge conditions.
558  */
559 static abi_long mmap_h_eq_g(abi_ulong start, abi_ulong len,
560                             int host_prot, int flags, int page_flags,
561                             int fd, off_t offset)
562 {
563     void *p, *want_p = g2h_untagged(start);
564     abi_ulong last;
565 
566     p = mmap(want_p, len, host_prot, flags, fd, offset);
567     if (p == MAP_FAILED) {
568         return -1;
569     }
570     /* If the host kernel does not support MAP_FIXED_NOREPLACE, emulate. */
571     if ((flags & MAP_FIXED_NOREPLACE) && p != want_p) {
572         do_munmap(p, len);
573         errno = EEXIST;
574         return -1;
575     }
576 
577     start = h2g(p);
578     last = start + len - 1;
579     return mmap_end(start, last, start, last, flags, page_flags);
580 }
581 
582 /*
583  * Special case host page size < target page size.
584  *
585  * The two special cases are increased guest alignment, and mapping
586  * past the end of a file.
587  *
588  * When mapping files into a memory area larger than the file,
589  * accesses to pages beyond the file size will cause a SIGBUS.
590  *
591  * For example, if mmaping a file of 100 bytes on a host with 4K
592  * pages emulating a target with 8K pages, the target expects to
593  * be able to access the first 8K. But the host will trap us on
594  * any access beyond 4K.
595  *
596  * When emulating a target with a larger page-size than the hosts,
597  * we may need to truncate file maps at EOF and add extra anonymous
598  * pages up to the targets page boundary.
599  *
600  * This workaround only works for files that do not change.
601  * If the file is later extended (e.g. ftruncate), the SIGBUS
602  * vanishes and the proper behaviour is that changes within the
603  * anon page should be reflected in the file.
604  *
605  * However, this case is rather common with executable images,
606  * so the workaround is important for even trivial tests, whereas
607  * the mmap of of a file being extended is less common.
608  */
609 static abi_long mmap_h_lt_g(abi_ulong start, abi_ulong len, int host_prot,
610                             int mmap_flags, int page_flags, int fd,
611                             off_t offset, int host_page_size)
612 {
613     void *p, *want_p = g2h_untagged(start);
614     off_t fileend_adj = 0;
615     int flags = mmap_flags;
616     abi_ulong last, pass_last;
617 
618     if (!(flags & MAP_ANONYMOUS)) {
619         struct stat sb;
620 
621         if (fstat(fd, &sb) == -1) {
622             return -1;
623         }
624         if (offset >= sb.st_size) {
625             /*
626              * The entire map is beyond the end of the file.
627              * Transform it to an anonymous mapping.
628              */
629             flags |= MAP_ANONYMOUS;
630             fd = -1;
631             offset = 0;
632         } else if (offset + len > sb.st_size) {
633             /*
634              * A portion of the map is beyond the end of the file.
635              * Truncate the file portion of the allocation.
636              */
637             fileend_adj = offset + len - sb.st_size;
638         }
639     }
640 
641     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
642         if (fileend_adj) {
643             p = mmap(want_p, len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
644         } else {
645             p = mmap(want_p, len, host_prot, flags, fd, offset);
646         }
647         if (p != want_p) {
648             if (p != MAP_FAILED) {
649                 /* Host does not support MAP_FIXED_NOREPLACE: emulate. */
650                 do_munmap(p, len);
651                 errno = EEXIST;
652             }
653             return -1;
654         }
655 
656         if (fileend_adj) {
657             void *t = mmap(p, len - fileend_adj, host_prot,
658                            (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED,
659                            fd, offset);
660 
661             if (t == MAP_FAILED) {
662                 int save_errno = errno;
663 
664                 /*
665                  * We failed a map over the top of the successful anonymous
666                  * mapping above. The only failure mode is running out of VMAs,
667                  * and there's nothing that we can do to detect that earlier.
668                  * If we have replaced an existing mapping with MAP_FIXED,
669                  * then we cannot properly recover.  It's a coin toss whether
670                  * it would be better to exit or continue here.
671                  */
672                 if (!(flags & MAP_FIXED_NOREPLACE) &&
673                     !page_check_range_empty(start, start + len - 1)) {
674                     qemu_log("QEMU target_mmap late failure: %s",
675                              strerror(save_errno));
676                 }
677 
678                 do_munmap(want_p, len);
679                 errno = save_errno;
680                 return -1;
681             }
682         }
683     } else {
684         size_t host_len, part_len;
685 
686         /*
687          * Take care to align the host memory.  Perform a larger anonymous
688          * allocation and extract the aligned portion.  Remap the file on
689          * top of that.
690          */
691         host_len = len + TARGET_PAGE_SIZE - host_page_size;
692         p = mmap(want_p, host_len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
693         if (p == MAP_FAILED) {
694             return -1;
695         }
696 
697         part_len = (uintptr_t)p & (TARGET_PAGE_SIZE - 1);
698         if (part_len) {
699             part_len = TARGET_PAGE_SIZE - part_len;
700             do_munmap(p, part_len);
701             p += part_len;
702             host_len -= part_len;
703         }
704         if (len < host_len) {
705             do_munmap(p + len, host_len - len);
706         }
707 
708         if (!(flags & MAP_ANONYMOUS)) {
709             void *t = mmap(p, len - fileend_adj, host_prot,
710                            flags | MAP_FIXED, fd, offset);
711 
712             if (t == MAP_FAILED) {
713                 int save_errno = errno;
714                 do_munmap(p, len);
715                 errno = save_errno;
716                 return -1;
717             }
718         }
719 
720         start = h2g(p);
721     }
722 
723     last = start + len - 1;
724     if (fileend_adj) {
725         pass_last = ROUND_UP(last - fileend_adj, host_page_size) - 1;
726     } else {
727         pass_last = last;
728     }
729     return mmap_end(start, last, start, pass_last, mmap_flags, page_flags);
730 }
731 
732 /*
733  * Special case host page size > target page size.
734  *
735  * The two special cases are address and file offsets that are valid
736  * for the guest that cannot be directly represented by the host.
737  */
738 static abi_long mmap_h_gt_g(abi_ulong start, abi_ulong len,
739                             int target_prot, int host_prot,
740                             int flags, int page_flags, int fd,
741                             off_t offset, int host_page_size)
742 {
743     void *p, *want_p = g2h_untagged(start);
744     off_t host_offset = offset & -host_page_size;
745     abi_ulong last, real_start, real_last;
746     bool misaligned_offset = false;
747     size_t host_len;
748 
749     if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
750         /*
751          * Adjust the offset to something representable on the host.
752          */
753         host_len = len + offset - host_offset;
754         p = mmap(want_p, host_len, host_prot, flags, fd, host_offset);
755         if (p == MAP_FAILED) {
756             return -1;
757         }
758 
759         /* Update start to the file position at offset. */
760         p += offset - host_offset;
761 
762         start = h2g(p);
763         last = start + len - 1;
764         return mmap_end(start, last, start, last, flags, page_flags);
765     }
766 
767     if (!(flags & MAP_ANONYMOUS)) {
768         misaligned_offset = (start ^ offset) & (host_page_size - 1);
769 
770         /*
771          * The fallback for misalignment is a private mapping + read.
772          * This carries none of semantics required of MAP_SHARED.
773          */
774         if (misaligned_offset && (flags & MAP_TYPE) != MAP_PRIVATE) {
775             errno = EINVAL;
776             return -1;
777         }
778     }
779 
780     last = start + len - 1;
781     real_start = start & -host_page_size;
782     real_last = ROUND_UP(last, host_page_size) - 1;
783 
784     /*
785      * Handle the start and end of the mapping.
786      */
787     if (real_start < start) {
788         abi_ulong real_page_last = real_start + host_page_size - 1;
789         if (last <= real_page_last) {
790             /* Entire allocation a subset of one host page. */
791             if (!mmap_frag(real_start, start, last, target_prot,
792                            flags, fd, offset)) {
793                 return -1;
794             }
795             return mmap_end(start, last, -1, 0, flags, page_flags);
796         }
797 
798         if (!mmap_frag(real_start, start, real_page_last, target_prot,
799                        flags, fd, offset)) {
800             return -1;
801         }
802         real_start = real_page_last + 1;
803     }
804 
805     if (last < real_last) {
806         abi_ulong real_page_start = real_last - host_page_size + 1;
807         if (!mmap_frag(real_page_start, real_page_start, last,
808                        target_prot, flags, fd,
809                        offset + real_page_start - start)) {
810             return -1;
811         }
812         real_last = real_page_start - 1;
813     }
814 
815     if (real_start > real_last) {
816         return mmap_end(start, last, -1, 0, flags, page_flags);
817     }
818 
819     /*
820      * Handle the middle of the mapping.
821      */
822 
823     host_len = real_last - real_start + 1;
824     want_p += real_start - start;
825 
826     if (flags & MAP_ANONYMOUS) {
827         p = mmap(want_p, host_len, host_prot, flags, -1, 0);
828     } else if (!misaligned_offset) {
829         p = mmap(want_p, host_len, host_prot, flags, fd,
830                  offset + real_start - start);
831     } else {
832         p = mmap(want_p, host_len, host_prot | PROT_WRITE,
833                  flags | MAP_ANONYMOUS, -1, 0);
834     }
835     if (p != want_p) {
836         if (p != MAP_FAILED) {
837             do_munmap(p, host_len);
838             errno = EEXIST;
839         }
840         return -1;
841     }
842 
843     if (misaligned_offset) {
844         /* TODO: The read could be short. */
845         if (pread(fd, p, host_len, offset + real_start - start) != host_len) {
846             do_munmap(p, host_len);
847             return -1;
848         }
849         if (!(host_prot & PROT_WRITE)) {
850             mprotect(p, host_len, host_prot);
851         }
852     }
853 
854     return mmap_end(start, last, -1, 0, flags, page_flags);
855 }
856 
857 static abi_long target_mmap__locked(abi_ulong start, abi_ulong len,
858                                     int target_prot, int flags, int page_flags,
859                                     int fd, off_t offset)
860 {
861     int host_page_size = qemu_real_host_page_size();
862     int host_prot;
863 
864     /*
865      * For reserved_va, we are in full control of the allocation.
866      * Find a suitable hole and convert to MAP_FIXED.
867      */
868     if (reserved_va) {
869         if (flags & MAP_FIXED_NOREPLACE) {
870             /* Validate that the chosen range is empty. */
871             if (!page_check_range_empty(start, start + len - 1)) {
872                 errno = EEXIST;
873                 return -1;
874             }
875             flags = (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED;
876         } else if (!(flags & MAP_FIXED)) {
877             abi_ulong real_start = start & -host_page_size;
878             off_t host_offset = offset & -host_page_size;
879             size_t real_len = len + offset - host_offset;
880             abi_ulong align = MAX(host_page_size, TARGET_PAGE_SIZE);
881 
882             start = mmap_find_vma(real_start, real_len, align);
883             if (start == (abi_ulong)-1) {
884                 errno = ENOMEM;
885                 return -1;
886             }
887             start += offset - host_offset;
888             flags |= MAP_FIXED;
889         }
890     }
891 
892     host_prot = target_to_host_prot(target_prot);
893 
894     if (host_page_size == TARGET_PAGE_SIZE) {
895         return mmap_h_eq_g(start, len, host_prot, flags,
896                            page_flags, fd, offset);
897     } else if (host_page_size < TARGET_PAGE_SIZE) {
898         return mmap_h_lt_g(start, len, host_prot, flags,
899                            page_flags, fd, offset, host_page_size);
900     } else {
901         return mmap_h_gt_g(start, len, target_prot, host_prot, flags,
902                            page_flags, fd, offset, host_page_size);
903     }
904 }
905 
906 /* NOTE: all the constants are the HOST ones */
907 abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot,
908                      int flags, int fd, off_t offset)
909 {
910     abi_long ret;
911     int page_flags;
912 
913     trace_target_mmap(start, len, target_prot, flags, fd, offset);
914 
915     if (!len) {
916         errno = EINVAL;
917         return -1;
918     }
919 
920     page_flags = validate_prot_to_pageflags(target_prot);
921     if (!page_flags) {
922         errno = EINVAL;
923         return -1;
924     }
925 
926     /* Also check for overflows... */
927     len = TARGET_PAGE_ALIGN(len);
928     if (!len || len != (size_t)len) {
929         errno = ENOMEM;
930         return -1;
931     }
932 
933     if (offset & ~TARGET_PAGE_MASK) {
934         errno = EINVAL;
935         return -1;
936     }
937     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
938         if (start & ~TARGET_PAGE_MASK) {
939             errno = EINVAL;
940             return -1;
941         }
942         if (!guest_range_valid_untagged(start, len)) {
943             errno = ENOMEM;
944             return -1;
945         }
946     }
947 
948     mmap_lock();
949 
950     ret = target_mmap__locked(start, len, target_prot, flags,
951                               page_flags, fd, offset);
952 
953     mmap_unlock();
954 
955     /*
956      * If we're mapping shared memory, ensure we generate code for parallel
957      * execution and flush old translations.  This will work up to the level
958      * supported by the host -- anything that requires EXCP_ATOMIC will not
959      * be atomic with respect to an external process.
960      */
961     if (ret != -1 && (flags & MAP_TYPE) != MAP_PRIVATE) {
962         CPUState *cpu = thread_cpu;
963         if (!tcg_cflags_has(cpu, CF_PARALLEL)) {
964             tcg_cflags_set(cpu, CF_PARALLEL);
965             tb_flush(cpu);
966         }
967     }
968 
969     return ret;
970 }
971 
972 static int mmap_reserve_or_unmap(abi_ulong start, abi_ulong len)
973 {
974     int host_page_size = qemu_real_host_page_size();
975     abi_ulong real_start;
976     abi_ulong real_last;
977     abi_ulong real_len;
978     abi_ulong last;
979     abi_ulong a;
980     void *host_start;
981     int prot;
982 
983     last = start + len - 1;
984     real_start = start & -host_page_size;
985     real_last = ROUND_UP(last, host_page_size) - 1;
986 
987     /*
988      * If guest pages remain on the first or last host pages,
989      * adjust the deallocation to retain those guest pages.
990      * The single page special case is required for the last page,
991      * lest real_start overflow to zero.
992      */
993     if (real_last - real_start < host_page_size) {
994         prot = 0;
995         for (a = real_start; a < start; a += TARGET_PAGE_SIZE) {
996             prot |= page_get_flags(a);
997         }
998         for (a = last; a < real_last; a += TARGET_PAGE_SIZE) {
999             prot |= page_get_flags(a + 1);
1000         }
1001         if (prot != 0) {
1002             return 0;
1003         }
1004     } else {
1005         for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) {
1006             prot |= page_get_flags(a);
1007         }
1008         if (prot != 0) {
1009             real_start += host_page_size;
1010         }
1011 
1012         for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) {
1013             prot |= page_get_flags(a + 1);
1014         }
1015         if (prot != 0) {
1016             real_last -= host_page_size;
1017         }
1018 
1019         if (real_last < real_start) {
1020             return 0;
1021         }
1022     }
1023 
1024     real_len = real_last - real_start + 1;
1025     host_start = g2h_untagged(real_start);
1026 
1027     return do_munmap(host_start, real_len);
1028 }
1029 
1030 int target_munmap(abi_ulong start, abi_ulong len)
1031 {
1032     int ret;
1033 
1034     trace_target_munmap(start, len);
1035 
1036     if (start & ~TARGET_PAGE_MASK) {
1037         errno = EINVAL;
1038         return -1;
1039     }
1040     len = TARGET_PAGE_ALIGN(len);
1041     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1042         errno = EINVAL;
1043         return -1;
1044     }
1045 
1046     mmap_lock();
1047     ret = mmap_reserve_or_unmap(start, len);
1048     if (likely(ret == 0)) {
1049         page_set_flags(start, start + len - 1, 0);
1050         shm_region_rm_complete(start, start + len - 1);
1051     }
1052     mmap_unlock();
1053 
1054     return ret;
1055 }
1056 
1057 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
1058                        abi_ulong new_size, unsigned long flags,
1059                        abi_ulong new_addr)
1060 {
1061     int prot;
1062     void *host_addr;
1063 
1064     if (!guest_range_valid_untagged(old_addr, old_size) ||
1065         ((flags & MREMAP_FIXED) &&
1066          !guest_range_valid_untagged(new_addr, new_size)) ||
1067         ((flags & MREMAP_MAYMOVE) == 0 &&
1068          !guest_range_valid_untagged(old_addr, new_size))) {
1069         errno = ENOMEM;
1070         return -1;
1071     }
1072 
1073     mmap_lock();
1074 
1075     if (flags & MREMAP_FIXED) {
1076         host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1077                            flags, g2h_untagged(new_addr));
1078 
1079         if (reserved_va && host_addr != MAP_FAILED) {
1080             /*
1081              * If new and old addresses overlap then the above mremap will
1082              * already have failed with EINVAL.
1083              */
1084             mmap_reserve_or_unmap(old_addr, old_size);
1085         }
1086     } else if (flags & MREMAP_MAYMOVE) {
1087         abi_ulong mmap_start;
1088 
1089         mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE);
1090 
1091         if (mmap_start == -1) {
1092             errno = ENOMEM;
1093             host_addr = MAP_FAILED;
1094         } else {
1095             host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1096                                flags | MREMAP_FIXED,
1097                                g2h_untagged(mmap_start));
1098             if (reserved_va) {
1099                 mmap_reserve_or_unmap(old_addr, old_size);
1100             }
1101         }
1102     } else {
1103         int page_flags = 0;
1104         if (reserved_va && old_size < new_size) {
1105             abi_ulong addr;
1106             for (addr = old_addr + old_size;
1107                  addr < old_addr + new_size;
1108                  addr++) {
1109                 page_flags |= page_get_flags(addr);
1110             }
1111         }
1112         if (page_flags == 0) {
1113             host_addr = mremap(g2h_untagged(old_addr),
1114                                old_size, new_size, flags);
1115 
1116             if (host_addr != MAP_FAILED) {
1117                 /* Check if address fits target address space */
1118                 if (!guest_range_valid_untagged(h2g(host_addr), new_size)) {
1119                     /* Revert mremap() changes */
1120                     host_addr = mremap(g2h_untagged(old_addr),
1121                                        new_size, old_size, flags);
1122                     errno = ENOMEM;
1123                     host_addr = MAP_FAILED;
1124                 } else if (reserved_va && old_size > new_size) {
1125                     mmap_reserve_or_unmap(old_addr + old_size,
1126                                           old_size - new_size);
1127                 }
1128             }
1129         } else {
1130             errno = ENOMEM;
1131             host_addr = MAP_FAILED;
1132         }
1133     }
1134 
1135     if (host_addr == MAP_FAILED) {
1136         new_addr = -1;
1137     } else {
1138         new_addr = h2g(host_addr);
1139         prot = page_get_flags(old_addr);
1140         page_set_flags(old_addr, old_addr + old_size - 1, 0);
1141         shm_region_rm_complete(old_addr, old_addr + old_size - 1);
1142         page_set_flags(new_addr, new_addr + new_size - 1,
1143                        prot | PAGE_VALID | PAGE_RESET);
1144         shm_region_rm_complete(new_addr, new_addr + new_size - 1);
1145     }
1146     mmap_unlock();
1147     return new_addr;
1148 }
1149 
1150 abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice)
1151 {
1152     abi_ulong len;
1153     int ret = 0;
1154 
1155     if (start & ~TARGET_PAGE_MASK) {
1156         return -TARGET_EINVAL;
1157     }
1158     if (len_in == 0) {
1159         return 0;
1160     }
1161     len = TARGET_PAGE_ALIGN(len_in);
1162     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1163         return -TARGET_EINVAL;
1164     }
1165 
1166     /* Translate for some architectures which have different MADV_xxx values */
1167     switch (advice) {
1168     case TARGET_MADV_DONTNEED:      /* alpha */
1169         advice = MADV_DONTNEED;
1170         break;
1171     case TARGET_MADV_WIPEONFORK:    /* parisc */
1172         advice = MADV_WIPEONFORK;
1173         break;
1174     case TARGET_MADV_KEEPONFORK:    /* parisc */
1175         advice = MADV_KEEPONFORK;
1176         break;
1177     /* we do not care about the other MADV_xxx values yet */
1178     }
1179 
1180     /*
1181      * Most advice values are hints, so ignoring and returning success is ok.
1182      *
1183      * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and
1184      * MADV_KEEPONFORK are not hints and need to be emulated.
1185      *
1186      * A straight passthrough for those may not be safe because qemu sometimes
1187      * turns private file-backed mappings into anonymous mappings.
1188      * If all guest pages have PAGE_PASSTHROUGH set, mappings have the
1189      * same semantics for the host as for the guest.
1190      *
1191      * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and
1192      * return failure if not.
1193      *
1194      * MADV_DONTNEED is passed through as well, if possible.
1195      * If passthrough isn't possible, we nevertheless (wrongly!) return
1196      * success, which is broken but some userspace programs fail to work
1197      * otherwise. Completely implementing such emulation is quite complicated
1198      * though.
1199      */
1200     mmap_lock();
1201     switch (advice) {
1202     case MADV_WIPEONFORK:
1203     case MADV_KEEPONFORK:
1204         ret = -EINVAL;
1205         /* fall through */
1206     case MADV_DONTNEED:
1207         if (page_check_range(start, len, PAGE_PASSTHROUGH)) {
1208             ret = get_errno(madvise(g2h_untagged(start), len, advice));
1209             if ((advice == MADV_DONTNEED) && (ret == 0)) {
1210                 page_reset_target_data(start, start + len - 1);
1211             }
1212         }
1213     }
1214     mmap_unlock();
1215 
1216     return ret;
1217 }
1218 
1219 #ifndef TARGET_FORCE_SHMLBA
1220 /*
1221  * For most architectures, SHMLBA is the same as the page size;
1222  * some architectures have larger values, in which case they should
1223  * define TARGET_FORCE_SHMLBA and provide a target_shmlba() function.
1224  * This corresponds to the kernel arch code defining __ARCH_FORCE_SHMLBA
1225  * and defining its own value for SHMLBA.
1226  *
1227  * The kernel also permits SHMLBA to be set by the architecture to a
1228  * value larger than the page size without setting __ARCH_FORCE_SHMLBA;
1229  * this means that addresses are rounded to the large size if
1230  * SHM_RND is set but addresses not aligned to that size are not rejected
1231  * as long as they are at least page-aligned. Since the only architecture
1232  * which uses this is ia64 this code doesn't provide for that oddity.
1233  */
1234 static inline abi_ulong target_shmlba(CPUArchState *cpu_env)
1235 {
1236     return TARGET_PAGE_SIZE;
1237 }
1238 #endif
1239 
1240 #if defined(__arm__) || defined(__mips__) || defined(__sparc__)
1241 #define HOST_FORCE_SHMLBA 1
1242 #else
1243 #define HOST_FORCE_SHMLBA 0
1244 #endif
1245 
1246 abi_ulong target_shmat(CPUArchState *cpu_env, int shmid,
1247                        abi_ulong shmaddr, int shmflg)
1248 {
1249     CPUState *cpu = env_cpu(cpu_env);
1250     struct shmid_ds shm_info;
1251     int ret;
1252     int h_pagesize;
1253     int t_shmlba, h_shmlba, m_shmlba;
1254     size_t t_len, h_len, m_len;
1255 
1256     /* shmat pointers are always untagged */
1257 
1258     /*
1259      * Because we can't use host shmat() unless the address is sufficiently
1260      * aligned for the host, we'll need to check both.
1261      * TODO: Could be fixed with softmmu.
1262      */
1263     t_shmlba = target_shmlba(cpu_env);
1264     h_pagesize = qemu_real_host_page_size();
1265     h_shmlba = (HOST_FORCE_SHMLBA ? SHMLBA : h_pagesize);
1266     m_shmlba = MAX(t_shmlba, h_shmlba);
1267 
1268     if (shmaddr) {
1269         if (shmaddr & (m_shmlba - 1)) {
1270             if (shmflg & SHM_RND) {
1271                 /*
1272                  * The guest is allowing the kernel to round the address.
1273                  * Assume that the guest is ok with us rounding to the
1274                  * host required alignment too.  Anyway if we don't, we'll
1275                  * get an error from the kernel.
1276                  */
1277                 shmaddr &= ~(m_shmlba - 1);
1278                 if (shmaddr == 0 && (shmflg & SHM_REMAP)) {
1279                     return -TARGET_EINVAL;
1280                 }
1281             } else {
1282                 int require = TARGET_PAGE_SIZE;
1283 #ifdef TARGET_FORCE_SHMLBA
1284                 require = t_shmlba;
1285 #endif
1286                 /*
1287                  * Include host required alignment, as otherwise we cannot
1288                  * use host shmat at all.
1289                  */
1290                 require = MAX(require, h_shmlba);
1291                 if (shmaddr & (require - 1)) {
1292                     return -TARGET_EINVAL;
1293                 }
1294             }
1295         }
1296     } else {
1297         if (shmflg & SHM_REMAP) {
1298             return -TARGET_EINVAL;
1299         }
1300     }
1301     /* All rounding now manually concluded. */
1302     shmflg &= ~SHM_RND;
1303 
1304     /* Find out the length of the shared memory segment. */
1305     ret = get_errno(shmctl(shmid, IPC_STAT, &shm_info));
1306     if (is_error(ret)) {
1307         /* can't get length, bail out */
1308         return ret;
1309     }
1310     t_len = TARGET_PAGE_ALIGN(shm_info.shm_segsz);
1311     h_len = ROUND_UP(shm_info.shm_segsz, h_pagesize);
1312     m_len = MAX(t_len, h_len);
1313 
1314     if (!guest_range_valid_untagged(shmaddr, m_len)) {
1315         return -TARGET_EINVAL;
1316     }
1317 
1318     WITH_MMAP_LOCK_GUARD() {
1319         bool mapped = false;
1320         void *want, *test;
1321         abi_ulong last;
1322 
1323         if (!shmaddr) {
1324             shmaddr = mmap_find_vma(0, m_len, m_shmlba);
1325             if (shmaddr == -1) {
1326                 return -TARGET_ENOMEM;
1327             }
1328             mapped = !reserved_va;
1329         } else if (shmflg & SHM_REMAP) {
1330             /*
1331              * If host page size > target page size, the host shmat may map
1332              * more memory than the guest expects.  Reject a mapping that
1333              * would replace memory in the unexpected gap.
1334              * TODO: Could be fixed with softmmu.
1335              */
1336             if (t_len < h_len &&
1337                 !page_check_range_empty(shmaddr + t_len,
1338                                         shmaddr + h_len - 1)) {
1339                 return -TARGET_EINVAL;
1340             }
1341         } else {
1342             if (!page_check_range_empty(shmaddr, shmaddr + m_len - 1)) {
1343                 return -TARGET_EINVAL;
1344             }
1345         }
1346 
1347         /* All placement is now complete. */
1348         want = (void *)g2h_untagged(shmaddr);
1349 
1350         /*
1351          * Map anonymous pages across the entire range, then remap with
1352          * the shared memory.  This is required for a number of corner
1353          * cases for which host and guest page sizes differ.
1354          */
1355         if (h_len != t_len) {
1356             int mmap_p = PROT_READ | (shmflg & SHM_RDONLY ? 0 : PROT_WRITE);
1357             int mmap_f = MAP_PRIVATE | MAP_ANONYMOUS
1358                        | (reserved_va || mapped || (shmflg & SHM_REMAP)
1359                           ? MAP_FIXED : MAP_FIXED_NOREPLACE);
1360 
1361             test = mmap(want, m_len, mmap_p, mmap_f, -1, 0);
1362             if (unlikely(test != want)) {
1363                 /* shmat returns EINVAL not EEXIST like mmap. */
1364                 ret = (test == MAP_FAILED && errno != EEXIST
1365                        ? get_errno(-1) : -TARGET_EINVAL);
1366                 if (mapped) {
1367                     do_munmap(want, m_len);
1368                 }
1369                 return ret;
1370             }
1371             mapped = true;
1372         }
1373 
1374         if (reserved_va || mapped) {
1375             shmflg |= SHM_REMAP;
1376         }
1377         test = shmat(shmid, want, shmflg);
1378         if (test == MAP_FAILED) {
1379             ret = get_errno(-1);
1380             if (mapped) {
1381                 do_munmap(want, m_len);
1382             }
1383             return ret;
1384         }
1385         assert(test == want);
1386 
1387         last = shmaddr + m_len - 1;
1388         page_set_flags(shmaddr, last,
1389                        PAGE_VALID | PAGE_RESET | PAGE_READ |
1390                        (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE) |
1391                        (shmflg & SHM_EXEC ? PAGE_EXEC : 0));
1392 
1393         shm_region_rm_complete(shmaddr, last);
1394         shm_region_add(shmaddr, last);
1395     }
1396 
1397     /*
1398      * We're mapping shared memory, so ensure we generate code for parallel
1399      * execution and flush old translations.  This will work up to the level
1400      * supported by the host -- anything that requires EXCP_ATOMIC will not
1401      * be atomic with respect to an external process.
1402      */
1403     if (!tcg_cflags_has(cpu, CF_PARALLEL)) {
1404         tcg_cflags_set(cpu, CF_PARALLEL);
1405         tb_flush(cpu);
1406     }
1407 
1408     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
1409         FILE *f = qemu_log_trylock();
1410         if (f) {
1411             fprintf(f, "page layout changed following shmat\n");
1412             page_dump(f);
1413             qemu_log_unlock(f);
1414         }
1415     }
1416     return shmaddr;
1417 }
1418 
1419 abi_long target_shmdt(abi_ulong shmaddr)
1420 {
1421     abi_long rv;
1422 
1423     /* shmdt pointers are always untagged */
1424 
1425     WITH_MMAP_LOCK_GUARD() {
1426         abi_ulong last = shm_region_find(shmaddr);
1427         if (last == 0) {
1428             return -TARGET_EINVAL;
1429         }
1430 
1431         rv = get_errno(shmdt(g2h_untagged(shmaddr)));
1432         if (rv == 0) {
1433             abi_ulong size = last - shmaddr + 1;
1434 
1435             page_set_flags(shmaddr, last, 0);
1436             shm_region_rm_complete(shmaddr, last);
1437             mmap_reserve_or_unmap(shmaddr, size);
1438         }
1439     }
1440     return rv;
1441 }
1442