xref: /openbmc/qemu/linux-user/mmap.c (revision 3bf5c57a11827d9fa706524d57ee3e5af68a429e)
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 "exec/mmap-lock.h"
25 #include "qemu.h"
26 #include "user/page-protection.h"
27 #include "user-internals.h"
28 #include "user-mmap.h"
29 #include "target_mman.h"
30 #include "qemu/interval-tree.h"
31 
32 #ifdef TARGET_ARM
33 #include "target/arm/cpu-features.h"
34 #endif
35 
36 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
37 static __thread int mmap_lock_count;
38 
39 void mmap_lock(void)
40 {
41     if (mmap_lock_count++ == 0) {
42         pthread_mutex_lock(&mmap_mutex);
43     }
44 }
45 
46 void mmap_unlock(void)
47 {
48     assert(mmap_lock_count > 0);
49     if (--mmap_lock_count == 0) {
50         pthread_mutex_unlock(&mmap_mutex);
51     }
52 }
53 
54 bool have_mmap_lock(void)
55 {
56     return mmap_lock_count > 0 ? true : false;
57 }
58 
59 /* Grab lock to make sure things are in a consistent state after fork().  */
60 void mmap_fork_start(void)
61 {
62     if (mmap_lock_count)
63         abort();
64     pthread_mutex_lock(&mmap_mutex);
65 }
66 
67 void mmap_fork_end(int child)
68 {
69     if (child) {
70         pthread_mutex_init(&mmap_mutex, NULL);
71     } else {
72         pthread_mutex_unlock(&mmap_mutex);
73     }
74 }
75 
76 /* Protected by mmap_lock. */
77 static IntervalTreeRoot shm_regions;
78 
79 static void shm_region_add(abi_ptr start, abi_ptr last)
80 {
81     IntervalTreeNode *i = g_new0(IntervalTreeNode, 1);
82 
83     i->start = start;
84     i->last = last;
85     interval_tree_insert(i, &shm_regions);
86 }
87 
88 static abi_ptr shm_region_find(abi_ptr start)
89 {
90     IntervalTreeNode *i;
91 
92     for (i = interval_tree_iter_first(&shm_regions, start, start); i;
93          i = interval_tree_iter_next(i, start, start)) {
94         if (i->start == start) {
95             return i->last;
96         }
97     }
98     return 0;
99 }
100 
101 static void shm_region_rm_complete(abi_ptr start, abi_ptr last)
102 {
103     IntervalTreeNode *i, *n;
104 
105     for (i = interval_tree_iter_first(&shm_regions, start, last); i; i = n) {
106         n = interval_tree_iter_next(i, start, last);
107         if (i->start >= start && i->last <= last) {
108             interval_tree_remove(i, &shm_regions);
109             g_free(i);
110         }
111     }
112 }
113 
114 /*
115  * Validate target prot bitmask.
116  * Return the prot bitmask for the host in *HOST_PROT.
117  * Return 0 if the target prot bitmask is invalid, otherwise
118  * the internal qemu page_flags (which will include PAGE_VALID).
119  */
120 static int validate_prot_to_pageflags(int prot)
121 {
122     int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM;
123     int page_flags = (prot & PAGE_RWX) | PAGE_VALID;
124 
125 #ifdef TARGET_AARCH64
126     {
127         ARMCPU *cpu = ARM_CPU(thread_cpu);
128 
129         /*
130          * The PROT_BTI bit is only accepted if the cpu supports the feature.
131          * Since this is the unusual case, don't bother checking unless
132          * the bit has been requested.  If set and valid, record the bit
133          * within QEMU's page_flags.
134          */
135         if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) {
136             valid |= TARGET_PROT_BTI;
137             page_flags |= PAGE_BTI;
138         }
139         /* Similarly for the PROT_MTE bit. */
140         if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) {
141             valid |= TARGET_PROT_MTE;
142             page_flags |= PAGE_MTE;
143         }
144     }
145 #elif defined(TARGET_HPPA)
146     valid |= PROT_GROWSDOWN | PROT_GROWSUP;
147 #endif
148 
149     return prot & ~valid ? 0 : page_flags;
150 }
151 
152 /*
153  * For the host, we need not pass anything except read/write/exec.
154  * While PROT_SEM is allowed by all hosts, it is also ignored, so
155  * don't bother transforming guest bit to host bit.  Any other
156  * target-specific prot bits will not be understood by the host
157  * and will need to be encoded into page_flags for qemu emulation.
158  *
159  * Pages that are executable by the guest will never be executed
160  * by the host, but the host will need to be able to read them.
161  */
162 static int target_to_host_prot(int prot)
163 {
164     return (prot & (PROT_READ | PROT_WRITE)) |
165            (prot & PROT_EXEC ? PROT_READ : 0);
166 }
167 
168 /* Target bits to be cleared by mprotect if not present in target_prot. */
169 #ifdef TARGET_AARCH64
170 #define TARGET_PAGE_NOTSTICKY  PAGE_BTI
171 #else
172 #define TARGET_PAGE_NOTSTICKY  0
173 #endif
174 
175 /* NOTE: all the constants are the HOST ones, but addresses are target. */
176 int target_mprotect(abi_ulong start, abi_ulong len, int target_prot)
177 {
178     int host_page_size = qemu_real_host_page_size();
179     abi_ulong starts[3];
180     abi_ulong lens[3];
181     int prots[3];
182     abi_ulong host_start, host_last, last;
183     int prot1, ret, page_flags, nranges;
184 
185     trace_target_mprotect(start, len, target_prot);
186 
187     if ((start & ~TARGET_PAGE_MASK) != 0) {
188         return -TARGET_EINVAL;
189     }
190     page_flags = validate_prot_to_pageflags(target_prot);
191     if (!page_flags) {
192         return -TARGET_EINVAL;
193     }
194     if (len == 0) {
195         return 0;
196     }
197     len = TARGET_PAGE_ALIGN(len);
198     if (!guest_range_valid_untagged(start, len)) {
199         return -TARGET_ENOMEM;
200     }
201 
202     last = start + len - 1;
203     host_start = start & -host_page_size;
204     host_last = ROUND_UP(last, host_page_size) - 1;
205     nranges = 0;
206 
207     mmap_lock();
208 
209     if (host_last - host_start < host_page_size) {
210         /* Single host page contains all guest pages: sum the prot. */
211         prot1 = target_prot;
212         for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
213             prot1 |= page_get_flags(a);
214         }
215         for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
216             prot1 |= page_get_flags(a + 1);
217         }
218         starts[nranges] = host_start;
219         lens[nranges] = host_page_size;
220         prots[nranges] = prot1;
221         nranges++;
222     } else {
223         if (host_start < start) {
224             /* Host page contains more than one guest page: sum the prot. */
225             prot1 = target_prot;
226             for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) {
227                 prot1 |= page_get_flags(a);
228             }
229             /* If the resulting sum differs, create a new range. */
230             if (prot1 != target_prot) {
231                 starts[nranges] = host_start;
232                 lens[nranges] = host_page_size;
233                 prots[nranges] = prot1;
234                 nranges++;
235                 host_start += host_page_size;
236             }
237         }
238 
239         if (last < host_last) {
240             /* Host page contains more than one guest page: sum the prot. */
241             prot1 = target_prot;
242             for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) {
243                 prot1 |= page_get_flags(a + 1);
244             }
245             /* If the resulting sum differs, create a new range. */
246             if (prot1 != target_prot) {
247                 host_last -= host_page_size;
248                 starts[nranges] = host_last + 1;
249                 lens[nranges] = host_page_size;
250                 prots[nranges] = prot1;
251                 nranges++;
252             }
253         }
254 
255         /* Create a range for the middle, if any remains. */
256         if (host_start < host_last) {
257             starts[nranges] = host_start;
258             lens[nranges] = host_last - host_start + 1;
259             prots[nranges] = target_prot;
260             nranges++;
261         }
262     }
263 
264     for (int i = 0; i < nranges; ++i) {
265         ret = mprotect(g2h_untagged(starts[i]), lens[i],
266                        target_to_host_prot(prots[i]));
267         if (ret != 0) {
268             goto error;
269         }
270     }
271 
272     page_set_flags(start, last, page_flags, PAGE_RWX | TARGET_PAGE_NOTSTICKY);
273     ret = 0;
274 
275  error:
276     mmap_unlock();
277     return ret;
278 }
279 
280 /*
281  * Perform munmap on behalf of the target, with host parameters.
282  * If reserved_va, we must replace the memory reservation.
283  */
284 static int do_munmap(void *addr, size_t len)
285 {
286     if (reserved_va) {
287         void *ptr = mmap(addr, len, PROT_NONE,
288                          MAP_FIXED | MAP_ANONYMOUS
289                          | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
290         return ptr == addr ? 0 : -1;
291     }
292     return munmap(addr, len);
293 }
294 
295 /*
296  * Perform a pread on behalf of target_mmap.  We can reach EOF, we can be
297  * interrupted by signals, and in general there's no good error return path.
298  * If @zero, zero the rest of the block at EOF.
299  * Return true on success.
300  */
301 static bool mmap_pread(int fd, void *p, size_t len, off_t offset, bool zero)
302 {
303     while (1) {
304         ssize_t r = pread(fd, p, len, offset);
305 
306         if (likely(r == len)) {
307             /* Complete */
308             return true;
309         }
310         if (r == 0) {
311             /* EOF */
312             if (zero) {
313                 memset(p, 0, len);
314             }
315             return true;
316         }
317         if (r > 0) {
318             /* Short read */
319             p += r;
320             len -= r;
321             offset += r;
322         } else if (errno != EINTR) {
323             /* Error */
324             return false;
325         }
326     }
327 }
328 
329 /*
330  * Map an incomplete host page.
331  *
332  * Here be dragons.  This case will not work if there is an existing
333  * overlapping host page, which is file mapped, and for which the mapping
334  * is beyond the end of the file.  In that case, we will see SIGBUS when
335  * trying to write a portion of this page.
336  *
337  * FIXME: Work around this with a temporary signal handler and longjmp.
338  */
339 static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last,
340                       int prot, int flags, int fd, off_t offset)
341 {
342     int host_page_size = qemu_real_host_page_size();
343     abi_ulong real_last;
344     void *host_start;
345     int prot_old, prot_new;
346     int host_prot_old, host_prot_new;
347 
348     if (!(flags & MAP_ANONYMOUS)
349         && (flags & MAP_TYPE) == MAP_SHARED
350         && (prot & PROT_WRITE)) {
351         /*
352          * msync() won't work with the partial page, so we return an
353          * error if write is possible while it is a shared mapping.
354          */
355         errno = EINVAL;
356         return false;
357     }
358 
359     real_last = real_start + host_page_size - 1;
360     host_start = g2h_untagged(real_start);
361 
362     /* Get the protection of the target pages outside the mapping. */
363     prot_old = 0;
364     for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) {
365         prot_old |= page_get_flags(a);
366     }
367     for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) {
368         prot_old |= page_get_flags(a);
369     }
370 
371     if (prot_old == 0) {
372         /*
373          * Since !(prot_old & PAGE_VALID), there were no guest pages
374          * outside of the fragment we need to map.  Allocate a new host
375          * page to cover, discarding whatever else may have been present.
376          */
377         void *p = mmap(host_start, host_page_size,
378                        target_to_host_prot(prot),
379                        flags | MAP_ANONYMOUS, -1, 0);
380         if (p != host_start) {
381             if (p != MAP_FAILED) {
382                 do_munmap(p, host_page_size);
383                 errno = EEXIST;
384             }
385             return false;
386         }
387         prot_old = prot;
388     }
389     prot_new = prot | prot_old;
390 
391     host_prot_old = target_to_host_prot(prot_old);
392     host_prot_new = target_to_host_prot(prot_new);
393 
394     /* Adjust protection to be able to write. */
395     if (!(host_prot_old & PROT_WRITE)) {
396         host_prot_old |= PROT_WRITE;
397         mprotect(host_start, host_page_size, host_prot_old);
398     }
399 
400     /* Read or zero the new guest pages. */
401     if (flags & MAP_ANONYMOUS) {
402         memset(g2h_untagged(start), 0, last - start + 1);
403     } else if (!mmap_pread(fd, g2h_untagged(start), last - start + 1,
404                            offset, true)) {
405         return false;
406     }
407 
408     /* Put final protection */
409     if (host_prot_new != host_prot_old) {
410         mprotect(host_start, host_page_size, host_prot_new);
411     }
412     return true;
413 }
414 
415 abi_ulong task_unmapped_base;
416 abi_ulong elf_et_dyn_base;
417 abi_ulong mmap_next_start;
418 
419 /*
420  * Subroutine of mmap_find_vma, used when we have pre-allocated
421  * a chunk of guest address space.
422  */
423 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size,
424                                         abi_ulong align)
425 {
426     target_ulong ret;
427 
428     ret = page_find_range_empty(start, reserved_va, size, align);
429     if (ret == -1 && start > mmap_min_addr) {
430         /* Restart at the beginning of the address space. */
431         ret = page_find_range_empty(mmap_min_addr, start - 1, size, align);
432     }
433 
434     return ret;
435 }
436 
437 /*
438  * Find and reserve a free memory area of size 'size'. The search
439  * starts at 'start'.
440  * It must be called with mmap_lock() held.
441  * Return -1 if error.
442  */
443 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align)
444 {
445     int host_page_size = qemu_real_host_page_size();
446     void *ptr, *prev;
447     abi_ulong addr;
448     int wrapped, repeat;
449 
450     align = MAX(align, host_page_size);
451 
452     /* If 'start' == 0, then a default start address is used. */
453     if (start == 0) {
454         start = mmap_next_start;
455     } else {
456         start &= -host_page_size;
457     }
458     start = ROUND_UP(start, align);
459     size = ROUND_UP(size, host_page_size);
460 
461     if (reserved_va) {
462         return mmap_find_vma_reserved(start, size, align);
463     }
464 
465     addr = start;
466     wrapped = repeat = 0;
467     prev = 0;
468 
469     for (;; prev = ptr) {
470         /*
471          * Reserve needed memory area to avoid a race.
472          * It should be discarded using:
473          *  - mmap() with MAP_FIXED flag
474          *  - mremap() with MREMAP_FIXED flag
475          *  - shmat() with SHM_REMAP flag
476          */
477         ptr = mmap(g2h_untagged(addr), size, PROT_NONE,
478                    MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0);
479 
480         /* ENOMEM, if host address space has no memory */
481         if (ptr == MAP_FAILED) {
482             return (abi_ulong)-1;
483         }
484 
485         /*
486          * Count the number of sequential returns of the same address.
487          * This is used to modify the search algorithm below.
488          */
489         repeat = (ptr == prev ? repeat + 1 : 0);
490 
491         if (h2g_valid(ptr + size - 1)) {
492             addr = h2g(ptr);
493 
494             if ((addr & (align - 1)) == 0) {
495                 /* Success.  */
496                 if (start == mmap_next_start && addr >= task_unmapped_base) {
497                     mmap_next_start = addr + size;
498                 }
499                 return addr;
500             }
501 
502             /* The address is not properly aligned for the target.  */
503             switch (repeat) {
504             case 0:
505                 /*
506                  * Assume the result that the kernel gave us is the
507                  * first with enough free space, so start again at the
508                  * next higher target page.
509                  */
510                 addr = ROUND_UP(addr, align);
511                 break;
512             case 1:
513                 /*
514                  * Sometimes the kernel decides to perform the allocation
515                  * at the top end of memory instead.
516                  */
517                 addr &= -align;
518                 break;
519             case 2:
520                 /* Start over at low memory.  */
521                 addr = 0;
522                 break;
523             default:
524                 /* Fail.  This unaligned block must the last.  */
525                 addr = -1;
526                 break;
527             }
528         } else {
529             /*
530              * Since the result the kernel gave didn't fit, start
531              * again at low memory.  If any repetition, fail.
532              */
533             addr = (repeat ? -1 : 0);
534         }
535 
536         /* Unmap and try again.  */
537         munmap(ptr, size);
538 
539         /* ENOMEM if we checked the whole of the target address space.  */
540         if (addr == (abi_ulong)-1) {
541             return (abi_ulong)-1;
542         } else if (addr == 0) {
543             if (wrapped) {
544                 return (abi_ulong)-1;
545             }
546             wrapped = 1;
547             /*
548              * Don't actually use 0 when wrapping, instead indicate
549              * that we'd truly like an allocation in low memory.
550              */
551             addr = (mmap_min_addr > TARGET_PAGE_SIZE
552                      ? TARGET_PAGE_ALIGN(mmap_min_addr)
553                      : TARGET_PAGE_SIZE);
554         } else if (wrapped && addr >= start) {
555             return (abi_ulong)-1;
556         }
557     }
558 }
559 
560 /*
561  * Record a successful mmap within the user-exec interval tree.
562  */
563 static abi_long mmap_end(abi_ulong start, abi_ulong last,
564                          abi_ulong passthrough_start,
565                          abi_ulong passthrough_last,
566                          int flags, int page_flags)
567 {
568     if (flags & MAP_ANONYMOUS) {
569         page_flags |= PAGE_ANON;
570     }
571     if (passthrough_start > passthrough_last) {
572         page_set_flags(start, last, page_flags, PAGE_VALID);
573     } else {
574         if (start < passthrough_start) {
575             page_set_flags(start, passthrough_start - 1,
576                            page_flags, PAGE_VALID);
577         }
578         page_set_flags(passthrough_start, passthrough_last,
579                        page_flags | PAGE_PASSTHROUGH, PAGE_VALID);
580         if (passthrough_last < last) {
581             page_set_flags(passthrough_last + 1, last, page_flags, PAGE_VALID);
582         }
583     }
584     shm_region_rm_complete(start, last);
585     trace_target_mmap_complete(start);
586     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
587         FILE *f = qemu_log_trylock();
588         if (f) {
589             fprintf(f, "page layout changed following mmap\n");
590             page_dump(f);
591             qemu_log_unlock(f);
592         }
593     }
594     return start;
595 }
596 
597 /*
598  * Special case host page size == target page size,
599  * where there are no edge conditions.
600  */
601 static abi_long mmap_h_eq_g(abi_ulong start, abi_ulong len,
602                             int host_prot, int flags, int page_flags,
603                             int fd, off_t offset)
604 {
605     void *p, *want_p = NULL;
606     abi_ulong last;
607 
608     if (start || (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
609         want_p = g2h_untagged(start);
610     }
611 
612     p = mmap(want_p, len, host_prot, flags, fd, offset);
613     if (p == MAP_FAILED) {
614         return -1;
615     }
616     /* If the host kernel does not support MAP_FIXED_NOREPLACE, emulate. */
617     if ((flags & MAP_FIXED_NOREPLACE) && p != want_p) {
618         do_munmap(p, len);
619         errno = EEXIST;
620         return -1;
621     }
622 
623     start = h2g(p);
624     last = start + len - 1;
625     return mmap_end(start, last, start, last, flags, page_flags);
626 }
627 
628 /*
629  * Special case host page size < target page size.
630  *
631  * The two special cases are increased guest alignment, and mapping
632  * past the end of a file.
633  *
634  * When mapping files into a memory area larger than the file,
635  * accesses to pages beyond the file size will cause a SIGBUS.
636  *
637  * For example, if mmaping a file of 100 bytes on a host with 4K
638  * pages emulating a target with 8K pages, the target expects to
639  * be able to access the first 8K. But the host will trap us on
640  * any access beyond 4K.
641  *
642  * When emulating a target with a larger page-size than the hosts,
643  * we may need to truncate file maps at EOF and add extra anonymous
644  * pages up to the targets page boundary.
645  *
646  * This workaround only works for files that do not change.
647  * If the file is later extended (e.g. ftruncate), the SIGBUS
648  * vanishes and the proper behaviour is that changes within the
649  * anon page should be reflected in the file.
650  *
651  * However, this case is rather common with executable images,
652  * so the workaround is important for even trivial tests, whereas
653  * the mmap of a file being extended is less common.
654  */
655 static abi_long mmap_h_lt_g(abi_ulong start, abi_ulong len, int host_prot,
656                             int mmap_flags, int page_flags, int fd,
657                             off_t offset, int host_page_size)
658 {
659     void *p, *want_p = NULL;
660     off_t fileend_adj = 0;
661     int flags = mmap_flags;
662     abi_ulong last, pass_last;
663 
664     if (start || (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
665         want_p = g2h_untagged(start);
666     }
667 
668     if (!(flags & MAP_ANONYMOUS)) {
669         struct stat sb;
670 
671         if (fstat(fd, &sb) == -1) {
672             return -1;
673         }
674         if (offset >= sb.st_size) {
675             /*
676              * The entire map is beyond the end of the file.
677              * Transform it to an anonymous mapping.
678              */
679             flags |= MAP_ANONYMOUS;
680             fd = -1;
681             offset = 0;
682         } else if (offset + len > sb.st_size) {
683             /*
684              * A portion of the map is beyond the end of the file.
685              * Truncate the file portion of the allocation.
686              */
687             fileend_adj = offset + len - sb.st_size;
688         }
689     }
690 
691     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
692         if (fileend_adj) {
693             p = mmap(want_p, len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
694         } else {
695             p = mmap(want_p, len, host_prot, flags, fd, offset);
696         }
697         if (p != want_p) {
698             if (p != MAP_FAILED) {
699                 /* Host does not support MAP_FIXED_NOREPLACE: emulate. */
700                 do_munmap(p, len);
701                 errno = EEXIST;
702             }
703             return -1;
704         }
705 
706         if (fileend_adj) {
707             void *t = mmap(p, len - fileend_adj, host_prot,
708                            (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED,
709                            fd, offset);
710 
711             if (t == MAP_FAILED) {
712                 int save_errno = errno;
713 
714                 /*
715                  * We failed a map over the top of the successful anonymous
716                  * mapping above. The only failure mode is running out of VMAs,
717                  * and there's nothing that we can do to detect that earlier.
718                  * If we have replaced an existing mapping with MAP_FIXED,
719                  * then we cannot properly recover.  It's a coin toss whether
720                  * it would be better to exit or continue here.
721                  */
722                 if (!(flags & MAP_FIXED_NOREPLACE) &&
723                     !page_check_range_empty(start, start + len - 1)) {
724                     qemu_log("QEMU target_mmap late failure: %s",
725                              strerror(save_errno));
726                 }
727 
728                 do_munmap(want_p, len);
729                 errno = save_errno;
730                 return -1;
731             }
732         }
733     } else {
734         size_t host_len, part_len;
735 
736         /*
737          * Take care to align the host memory.  Perform a larger anonymous
738          * allocation and extract the aligned portion.  Remap the file on
739          * top of that.
740          */
741         host_len = len + TARGET_PAGE_SIZE - host_page_size;
742         p = mmap(want_p, host_len, host_prot, flags | MAP_ANONYMOUS, -1, 0);
743         if (p == MAP_FAILED) {
744             return -1;
745         }
746 
747         part_len = (uintptr_t)p & (TARGET_PAGE_SIZE - 1);
748         if (part_len) {
749             part_len = TARGET_PAGE_SIZE - part_len;
750             do_munmap(p, part_len);
751             p += part_len;
752             host_len -= part_len;
753         }
754         if (len < host_len) {
755             do_munmap(p + len, host_len - len);
756         }
757 
758         if (!(flags & MAP_ANONYMOUS)) {
759             void *t = mmap(p, len - fileend_adj, host_prot,
760                            flags | MAP_FIXED, fd, offset);
761 
762             if (t == MAP_FAILED) {
763                 int save_errno = errno;
764                 do_munmap(p, len);
765                 errno = save_errno;
766                 return -1;
767             }
768         }
769 
770         start = h2g(p);
771     }
772 
773     last = start + len - 1;
774     if (fileend_adj) {
775         pass_last = ROUND_UP(last - fileend_adj, host_page_size) - 1;
776     } else {
777         pass_last = last;
778     }
779     return mmap_end(start, last, start, pass_last, mmap_flags, page_flags);
780 }
781 
782 /*
783  * Special case host page size > target page size.
784  *
785  * The two special cases are address and file offsets that are valid
786  * for the guest that cannot be directly represented by the host.
787  */
788 static abi_long mmap_h_gt_g(abi_ulong start, abi_ulong len,
789                             int target_prot, int host_prot,
790                             int flags, int page_flags, int fd,
791                             off_t offset, int host_page_size)
792 {
793     void *p, *want_p = NULL;
794     off_t host_offset = offset & -host_page_size;
795     abi_ulong last, real_start, real_last;
796     bool misaligned_offset = false;
797     size_t host_len;
798 
799     if (start || (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
800         want_p = g2h_untagged(start);
801     }
802 
803     if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) {
804         /*
805          * Adjust the offset to something representable on the host.
806          */
807         host_len = len + offset - host_offset;
808         p = mmap(want_p, host_len, host_prot, flags, fd, host_offset);
809         if (p == MAP_FAILED) {
810             return -1;
811         }
812 
813         /* Update start to the file position at offset. */
814         p += offset - host_offset;
815 
816         start = h2g(p);
817         last = start + len - 1;
818         return mmap_end(start, last, start, last, flags, page_flags);
819     }
820 
821     if (!(flags & MAP_ANONYMOUS)) {
822         misaligned_offset = (start ^ offset) & (host_page_size - 1);
823 
824         /*
825          * The fallback for misalignment is a private mapping + read.
826          * This carries none of semantics required of MAP_SHARED.
827          */
828         if (misaligned_offset && (flags & MAP_TYPE) != MAP_PRIVATE) {
829             errno = EINVAL;
830             return -1;
831         }
832     }
833 
834     last = start + len - 1;
835     real_start = start & -host_page_size;
836     real_last = ROUND_UP(last, host_page_size) - 1;
837 
838     /*
839      * Handle the start and end of the mapping.
840      */
841     if (real_start < start) {
842         abi_ulong real_page_last = real_start + host_page_size - 1;
843         if (last <= real_page_last) {
844             /* Entire allocation a subset of one host page. */
845             if (!mmap_frag(real_start, start, last, target_prot,
846                            flags, fd, offset)) {
847                 return -1;
848             }
849             return mmap_end(start, last, -1, 0, flags, page_flags);
850         }
851 
852         if (!mmap_frag(real_start, start, real_page_last, target_prot,
853                        flags, fd, offset)) {
854             return -1;
855         }
856         real_start = real_page_last + 1;
857     }
858 
859     if (last < real_last) {
860         abi_ulong real_page_start = real_last - host_page_size + 1;
861         if (!mmap_frag(real_page_start, real_page_start, last,
862                        target_prot, flags, fd,
863                        offset + real_page_start - start)) {
864             return -1;
865         }
866         real_last = real_page_start - 1;
867     }
868 
869     if (real_start > real_last) {
870         return mmap_end(start, last, -1, 0, flags, page_flags);
871     }
872 
873     /*
874      * Handle the middle of the mapping.
875      */
876 
877     host_len = real_last - real_start + 1;
878     want_p += real_start - start;
879 
880     if (flags & MAP_ANONYMOUS) {
881         p = mmap(want_p, host_len, host_prot, flags, -1, 0);
882     } else if (!misaligned_offset) {
883         p = mmap(want_p, host_len, host_prot, flags, fd,
884                  offset + real_start - start);
885     } else {
886         p = mmap(want_p, host_len, host_prot | PROT_WRITE,
887                  flags | MAP_ANONYMOUS, -1, 0);
888     }
889     if (p != want_p) {
890         if (p != MAP_FAILED) {
891             do_munmap(p, host_len);
892             errno = EEXIST;
893         }
894         return -1;
895     }
896 
897     if (misaligned_offset) {
898         if (!mmap_pread(fd, p, host_len, offset + real_start - start, false)) {
899             do_munmap(p, host_len);
900             return -1;
901         }
902         if (!(host_prot & PROT_WRITE)) {
903             mprotect(p, host_len, host_prot);
904         }
905     }
906 
907     return mmap_end(start, last, -1, 0, flags, page_flags);
908 }
909 
910 static abi_long target_mmap__locked(abi_ulong start, abi_ulong len,
911                                     int target_prot, int flags, int page_flags,
912                                     int fd, off_t offset)
913 {
914     int host_page_size = qemu_real_host_page_size();
915     int host_prot;
916 
917     /*
918      * For reserved_va, we are in full control of the allocation.
919      * Find a suitable hole and convert to MAP_FIXED.
920      */
921     if (reserved_va) {
922         if (flags & MAP_FIXED_NOREPLACE) {
923             /* Validate that the chosen range is empty. */
924             if (!page_check_range_empty(start, start + len - 1)) {
925                 errno = EEXIST;
926                 return -1;
927             }
928             flags = (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED;
929         } else if (!(flags & MAP_FIXED)) {
930             abi_ulong real_start = start & -host_page_size;
931             off_t host_offset = offset & -host_page_size;
932             size_t real_len = len + offset - host_offset;
933             abi_ulong align = MAX(host_page_size, TARGET_PAGE_SIZE);
934 
935             start = mmap_find_vma(real_start, real_len, align);
936             if (start == (abi_ulong)-1) {
937                 errno = ENOMEM;
938                 return -1;
939             }
940             start += offset - host_offset;
941             flags |= MAP_FIXED;
942         }
943     }
944 
945     host_prot = target_to_host_prot(target_prot);
946 
947     if (host_page_size == TARGET_PAGE_SIZE) {
948         return mmap_h_eq_g(start, len, host_prot, flags,
949                            page_flags, fd, offset);
950     } else if (host_page_size < TARGET_PAGE_SIZE) {
951         return mmap_h_lt_g(start, len, host_prot, flags,
952                            page_flags, fd, offset, host_page_size);
953     } else {
954         return mmap_h_gt_g(start, len, target_prot, host_prot, flags,
955                            page_flags, fd, offset, host_page_size);
956     }
957 }
958 
959 /* NOTE: all the constants are the HOST ones */
960 abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot,
961                      int flags, int fd, off_t offset)
962 {
963     abi_long ret;
964     int page_flags;
965 
966     trace_target_mmap(start, len, target_prot, flags, fd, offset);
967 
968     if (!len) {
969         errno = EINVAL;
970         return -1;
971     }
972 
973     page_flags = validate_prot_to_pageflags(target_prot);
974     if (!page_flags) {
975         errno = EINVAL;
976         return -1;
977     }
978 
979     /* Also check for overflows... */
980     len = TARGET_PAGE_ALIGN(len);
981     if (!len || len != (size_t)len) {
982         errno = ENOMEM;
983         return -1;
984     }
985 
986     if (offset & ~TARGET_PAGE_MASK) {
987         errno = EINVAL;
988         return -1;
989     }
990     if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) {
991         if (start & ~TARGET_PAGE_MASK) {
992             errno = EINVAL;
993             return -1;
994         }
995         if (!guest_range_valid_untagged(start, len)) {
996             errno = ENOMEM;
997             return -1;
998         }
999     }
1000 
1001     mmap_lock();
1002 
1003     ret = target_mmap__locked(start, len, target_prot, flags,
1004                               page_flags, fd, offset);
1005 
1006     mmap_unlock();
1007 
1008     /*
1009      * If we're mapping shared memory, ensure we generate code for parallel
1010      * execution and flush old translations.  This will work up to the level
1011      * supported by the host -- anything that requires EXCP_ATOMIC will not
1012      * be atomic with respect to an external process.
1013      */
1014     if (ret != -1 && (flags & MAP_TYPE) != MAP_PRIVATE) {
1015         begin_parallel_context(thread_cpu);
1016     }
1017 
1018     return ret;
1019 }
1020 
1021 static int mmap_reserve_or_unmap(abi_ulong start, abi_ulong len)
1022 {
1023     int host_page_size = qemu_real_host_page_size();
1024     abi_ulong real_start;
1025     abi_ulong real_last;
1026     abi_ulong real_len;
1027     abi_ulong last;
1028     abi_ulong a;
1029     void *host_start;
1030     int prot;
1031 
1032     last = start + len - 1;
1033     real_start = start & -host_page_size;
1034     real_last = ROUND_UP(last, host_page_size) - 1;
1035 
1036     /*
1037      * If guest pages remain on the first or last host pages,
1038      * adjust the deallocation to retain those guest pages.
1039      * The single page special case is required for the last page,
1040      * lest real_start overflow to zero.
1041      */
1042     if (real_last - real_start < host_page_size) {
1043         prot = 0;
1044         for (a = real_start; a < start; a += TARGET_PAGE_SIZE) {
1045             prot |= page_get_flags(a);
1046         }
1047         for (a = last; a < real_last; a += TARGET_PAGE_SIZE) {
1048             prot |= page_get_flags(a + 1);
1049         }
1050         if (prot != 0) {
1051             return 0;
1052         }
1053     } else {
1054         for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) {
1055             prot |= page_get_flags(a);
1056         }
1057         if (prot != 0) {
1058             real_start += host_page_size;
1059         }
1060 
1061         for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) {
1062             prot |= page_get_flags(a + 1);
1063         }
1064         if (prot != 0) {
1065             real_last -= host_page_size;
1066         }
1067 
1068         if (real_last < real_start) {
1069             return 0;
1070         }
1071     }
1072 
1073     real_len = real_last - real_start + 1;
1074     host_start = g2h_untagged(real_start);
1075 
1076     return do_munmap(host_start, real_len);
1077 }
1078 
1079 int target_munmap(abi_ulong start, abi_ulong len)
1080 {
1081     int ret;
1082 
1083     trace_target_munmap(start, len);
1084 
1085     if (start & ~TARGET_PAGE_MASK) {
1086         errno = EINVAL;
1087         return -1;
1088     }
1089     len = TARGET_PAGE_ALIGN(len);
1090     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1091         errno = EINVAL;
1092         return -1;
1093     }
1094 
1095     mmap_lock();
1096     ret = mmap_reserve_or_unmap(start, len);
1097     if (likely(ret == 0)) {
1098         page_set_flags(start, start + len - 1, 0, PAGE_VALID);
1099         shm_region_rm_complete(start, start + len - 1);
1100     }
1101     mmap_unlock();
1102 
1103     return ret;
1104 }
1105 
1106 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
1107                        abi_ulong new_size, unsigned long flags,
1108                        abi_ulong new_addr)
1109 {
1110     int prot;
1111     void *host_addr;
1112 
1113     if (!guest_range_valid_untagged(old_addr, old_size) ||
1114         ((flags & MREMAP_FIXED) &&
1115          !guest_range_valid_untagged(new_addr, new_size)) ||
1116         ((flags & MREMAP_MAYMOVE) == 0 &&
1117          !guest_range_valid_untagged(old_addr, new_size))) {
1118         errno = ENOMEM;
1119         return -1;
1120     }
1121 
1122     mmap_lock();
1123 
1124     if (flags & MREMAP_FIXED) {
1125         host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1126                            flags, g2h_untagged(new_addr));
1127 
1128         if (reserved_va && host_addr != MAP_FAILED) {
1129             /*
1130              * If new and old addresses overlap then the above mremap will
1131              * already have failed with EINVAL.
1132              */
1133             mmap_reserve_or_unmap(old_addr, old_size);
1134         }
1135     } else if (flags & MREMAP_MAYMOVE) {
1136         abi_ulong mmap_start;
1137 
1138         mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE);
1139 
1140         if (mmap_start == -1) {
1141             errno = ENOMEM;
1142             host_addr = MAP_FAILED;
1143         } else {
1144             host_addr = mremap(g2h_untagged(old_addr), old_size, new_size,
1145                                flags | MREMAP_FIXED,
1146                                g2h_untagged(mmap_start));
1147             if (reserved_va) {
1148                 mmap_reserve_or_unmap(old_addr, old_size);
1149             }
1150         }
1151     } else {
1152         int page_flags = 0;
1153         if (reserved_va && old_size < new_size) {
1154             abi_ulong addr;
1155             for (addr = old_addr + old_size;
1156                  addr < old_addr + new_size;
1157                  addr++) {
1158                 page_flags |= page_get_flags(addr);
1159             }
1160         }
1161         if (page_flags == 0) {
1162             host_addr = mremap(g2h_untagged(old_addr),
1163                                old_size, new_size, flags);
1164 
1165             if (host_addr != MAP_FAILED) {
1166                 /* Check if address fits target address space */
1167                 if (!guest_range_valid_untagged(h2g(host_addr), new_size)) {
1168                     /* Revert mremap() changes */
1169                     host_addr = mremap(g2h_untagged(old_addr),
1170                                        new_size, old_size, flags);
1171                     errno = ENOMEM;
1172                     host_addr = MAP_FAILED;
1173                 } else if (reserved_va && old_size > new_size) {
1174                     mmap_reserve_or_unmap(old_addr + old_size,
1175                                           old_size - new_size);
1176                 }
1177             }
1178         } else {
1179             errno = ENOMEM;
1180             host_addr = MAP_FAILED;
1181         }
1182     }
1183 
1184     if (host_addr == MAP_FAILED) {
1185         new_addr = -1;
1186     } else {
1187         new_addr = h2g(host_addr);
1188         prot = page_get_flags(old_addr);
1189         page_set_flags(old_addr, old_addr + old_size - 1, 0, PAGE_VALID);
1190         shm_region_rm_complete(old_addr, old_addr + old_size - 1);
1191         page_set_flags(new_addr, new_addr + new_size - 1,
1192                        prot | PAGE_VALID, PAGE_VALID);
1193         shm_region_rm_complete(new_addr, new_addr + new_size - 1);
1194     }
1195     mmap_unlock();
1196     return new_addr;
1197 }
1198 
1199 abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice)
1200 {
1201     abi_ulong len;
1202     int ret = 0;
1203 
1204     if (start & ~TARGET_PAGE_MASK) {
1205         return -TARGET_EINVAL;
1206     }
1207     if (len_in == 0) {
1208         return 0;
1209     }
1210     len = TARGET_PAGE_ALIGN(len_in);
1211     if (len == 0 || !guest_range_valid_untagged(start, len)) {
1212         return -TARGET_EINVAL;
1213     }
1214 
1215     /* Translate for some architectures which have different MADV_xxx values */
1216     switch (advice) {
1217     case TARGET_MADV_DONTNEED:      /* alpha */
1218         advice = MADV_DONTNEED;
1219         break;
1220     case TARGET_MADV_WIPEONFORK:    /* parisc */
1221         advice = MADV_WIPEONFORK;
1222         break;
1223     case TARGET_MADV_KEEPONFORK:    /* parisc */
1224         advice = MADV_KEEPONFORK;
1225         break;
1226     /* we do not care about the other MADV_xxx values yet */
1227     }
1228 
1229     /*
1230      * Most advice values are hints, so ignoring and returning success is ok.
1231      *
1232      * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and
1233      * MADV_KEEPONFORK are not hints and need to be emulated.
1234      *
1235      * A straight passthrough for those may not be safe because qemu sometimes
1236      * turns private file-backed mappings into anonymous mappings.
1237      * If all guest pages have PAGE_PASSTHROUGH set, mappings have the
1238      * same semantics for the host as for the guest.
1239      *
1240      * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and
1241      * return failure if not.
1242      *
1243      * MADV_DONTNEED is passed through as well, if possible.
1244      * If passthrough isn't possible, we nevertheless (wrongly!) return
1245      * success, which is broken but some userspace programs fail to work
1246      * otherwise. Completely implementing such emulation is quite complicated
1247      * though.
1248      */
1249     mmap_lock();
1250     switch (advice) {
1251     case MADV_DONTDUMP:
1252         page_set_flags(start, start + len - 1, PAGE_DONTDUMP, 0);
1253         break;
1254     case MADV_DODUMP:
1255         page_set_flags(start, start + len - 1, 0, PAGE_DONTDUMP);
1256         break;
1257     case MADV_WIPEONFORK:
1258     case MADV_KEEPONFORK:
1259         ret = -EINVAL;
1260         /* fall through */
1261     case MADV_DONTNEED:
1262         if (page_check_range(start, len, PAGE_PASSTHROUGH)) {
1263             ret = get_errno(madvise(g2h_untagged(start), len, advice));
1264             if ((advice == MADV_DONTNEED) && (ret == 0)) {
1265                 page_reset_target_data(start, start + len - 1);
1266             }
1267         }
1268     }
1269     mmap_unlock();
1270 
1271     return ret;
1272 }
1273 
1274 #ifndef TARGET_FORCE_SHMLBA
1275 /*
1276  * For most architectures, SHMLBA is the same as the page size;
1277  * some architectures have larger values, in which case they should
1278  * define TARGET_FORCE_SHMLBA and provide a target_shmlba() function.
1279  * This corresponds to the kernel arch code defining __ARCH_FORCE_SHMLBA
1280  * and defining its own value for SHMLBA.
1281  *
1282  * The kernel also permits SHMLBA to be set by the architecture to a
1283  * value larger than the page size without setting __ARCH_FORCE_SHMLBA;
1284  * this means that addresses are rounded to the large size if
1285  * SHM_RND is set but addresses not aligned to that size are not rejected
1286  * as long as they are at least page-aligned. Since the only architecture
1287  * which uses this is ia64 this code doesn't provide for that oddity.
1288  */
1289 static inline abi_ulong target_shmlba(CPUArchState *cpu_env)
1290 {
1291     return TARGET_PAGE_SIZE;
1292 }
1293 #endif
1294 
1295 #if defined(__arm__) || defined(__mips__) || defined(__sparc__)
1296 #define HOST_FORCE_SHMLBA 1
1297 #else
1298 #define HOST_FORCE_SHMLBA 0
1299 #endif
1300 
1301 abi_ulong target_shmat(CPUArchState *cpu_env, int shmid,
1302                        abi_ulong shmaddr, int shmflg)
1303 {
1304     CPUState *cpu = env_cpu(cpu_env);
1305     struct shmid_ds shm_info;
1306     int ret;
1307     int h_pagesize;
1308     int t_shmlba, h_shmlba, m_shmlba;
1309     size_t t_len, h_len, m_len;
1310 
1311     /* shmat pointers are always untagged */
1312 
1313     /*
1314      * Because we can't use host shmat() unless the address is sufficiently
1315      * aligned for the host, we'll need to check both.
1316      * TODO: Could be fixed with softmmu.
1317      */
1318     t_shmlba = target_shmlba(cpu_env);
1319     h_pagesize = qemu_real_host_page_size();
1320     h_shmlba = (HOST_FORCE_SHMLBA ? SHMLBA : h_pagesize);
1321     m_shmlba = MAX(t_shmlba, h_shmlba);
1322 
1323     if (shmaddr) {
1324         if (shmaddr & (m_shmlba - 1)) {
1325             if (shmflg & SHM_RND) {
1326                 /*
1327                  * The guest is allowing the kernel to round the address.
1328                  * Assume that the guest is ok with us rounding to the
1329                  * host required alignment too.  Anyway if we don't, we'll
1330                  * get an error from the kernel.
1331                  */
1332                 shmaddr &= ~(m_shmlba - 1);
1333                 if (shmaddr == 0 && (shmflg & SHM_REMAP)) {
1334                     return -TARGET_EINVAL;
1335                 }
1336             } else {
1337                 int require = TARGET_PAGE_SIZE;
1338 #ifdef TARGET_FORCE_SHMLBA
1339                 require = t_shmlba;
1340 #endif
1341                 /*
1342                  * Include host required alignment, as otherwise we cannot
1343                  * use host shmat at all.
1344                  */
1345                 require = MAX(require, h_shmlba);
1346                 if (shmaddr & (require - 1)) {
1347                     return -TARGET_EINVAL;
1348                 }
1349             }
1350         }
1351     } else {
1352         if (shmflg & SHM_REMAP) {
1353             return -TARGET_EINVAL;
1354         }
1355     }
1356     /* All rounding now manually concluded. */
1357     shmflg &= ~SHM_RND;
1358 
1359     /* Find out the length of the shared memory segment. */
1360     ret = get_errno(shmctl(shmid, IPC_STAT, &shm_info));
1361     if (is_error(ret)) {
1362         /* can't get length, bail out */
1363         return ret;
1364     }
1365     t_len = TARGET_PAGE_ALIGN(shm_info.shm_segsz);
1366     h_len = ROUND_UP(shm_info.shm_segsz, h_pagesize);
1367     m_len = MAX(t_len, h_len);
1368 
1369     if (!guest_range_valid_untagged(shmaddr, m_len)) {
1370         return -TARGET_EINVAL;
1371     }
1372 
1373     WITH_MMAP_LOCK_GUARD() {
1374         bool mapped = false;
1375         void *want, *test;
1376         abi_ulong last;
1377 
1378         if (!shmaddr) {
1379             shmaddr = mmap_find_vma(0, m_len, m_shmlba);
1380             if (shmaddr == -1) {
1381                 return -TARGET_ENOMEM;
1382             }
1383             mapped = !reserved_va;
1384         } else if (shmflg & SHM_REMAP) {
1385             /*
1386              * If host page size > target page size, the host shmat may map
1387              * more memory than the guest expects.  Reject a mapping that
1388              * would replace memory in the unexpected gap.
1389              * TODO: Could be fixed with softmmu.
1390              */
1391             if (t_len < h_len &&
1392                 !page_check_range_empty(shmaddr + t_len,
1393                                         shmaddr + h_len - 1)) {
1394                 return -TARGET_EINVAL;
1395             }
1396         } else {
1397             if (!page_check_range_empty(shmaddr, shmaddr + m_len - 1)) {
1398                 return -TARGET_EINVAL;
1399             }
1400         }
1401 
1402         /* All placement is now complete. */
1403         want = (void *)g2h_untagged(shmaddr);
1404 
1405         /*
1406          * Map anonymous pages across the entire range, then remap with
1407          * the shared memory.  This is required for a number of corner
1408          * cases for which host and guest page sizes differ.
1409          */
1410         if (h_len != t_len) {
1411             int mmap_p = PROT_READ | (shmflg & SHM_RDONLY ? 0 : PROT_WRITE);
1412             int mmap_f = MAP_PRIVATE | MAP_ANONYMOUS
1413                        | (reserved_va || mapped || (shmflg & SHM_REMAP)
1414                           ? MAP_FIXED : MAP_FIXED_NOREPLACE);
1415 
1416             test = mmap(want, m_len, mmap_p, mmap_f, -1, 0);
1417             if (unlikely(test != want)) {
1418                 /* shmat returns EINVAL not EEXIST like mmap. */
1419                 ret = (test == MAP_FAILED && errno != EEXIST
1420                        ? get_errno(-1) : -TARGET_EINVAL);
1421                 if (mapped) {
1422                     do_munmap(want, m_len);
1423                 }
1424                 return ret;
1425             }
1426             mapped = true;
1427         }
1428 
1429         if (reserved_va || mapped) {
1430             shmflg |= SHM_REMAP;
1431         }
1432         test = shmat(shmid, want, shmflg);
1433         if (test == MAP_FAILED) {
1434             ret = get_errno(-1);
1435             if (mapped) {
1436                 do_munmap(want, m_len);
1437             }
1438             return ret;
1439         }
1440         assert(test == want);
1441 
1442         last = shmaddr + m_len - 1;
1443         page_set_flags(shmaddr, last,
1444                        PAGE_VALID | PAGE_READ |
1445                        (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE) |
1446                        (shmflg & SHM_EXEC ? PAGE_EXEC : 0),
1447                        PAGE_VALID);
1448 
1449         shm_region_rm_complete(shmaddr, last);
1450         shm_region_add(shmaddr, last);
1451     }
1452 
1453     /*
1454      * We're mapping shared memory, so ensure we generate code for parallel
1455      * execution and flush old translations.  This will work up to the level
1456      * supported by the host -- anything that requires EXCP_ATOMIC will not
1457      * be atomic with respect to an external process.
1458      */
1459     begin_parallel_context(cpu);
1460 
1461     if (qemu_loglevel_mask(CPU_LOG_PAGE)) {
1462         FILE *f = qemu_log_trylock();
1463         if (f) {
1464             fprintf(f, "page layout changed following shmat\n");
1465             page_dump(f);
1466             qemu_log_unlock(f);
1467         }
1468     }
1469     return shmaddr;
1470 }
1471 
1472 abi_long target_shmdt(abi_ulong shmaddr)
1473 {
1474     abi_long rv;
1475 
1476     /* shmdt pointers are always untagged */
1477 
1478     WITH_MMAP_LOCK_GUARD() {
1479         abi_ulong last = shm_region_find(shmaddr);
1480         if (last == 0) {
1481             return -TARGET_EINVAL;
1482         }
1483 
1484         rv = get_errno(shmdt(g2h_untagged(shmaddr)));
1485         if (rv == 0) {
1486             abi_ulong size = last - shmaddr + 1;
1487 
1488             page_set_flags(shmaddr, last, 0, PAGE_VALID);
1489             shm_region_rm_complete(shmaddr, last);
1490             mmap_reserve_or_unmap(shmaddr, size);
1491         }
1492     }
1493     return rv;
1494 }
1495