2 * Copyright (c) 2004 Tim J. Robbins
3 * Copyright (c) 2002 Doug Rabson
4 * Copyright (c) 2000 Marcel Moolenaar
5 * Copyright (c) 1994-1995 Søren Schmidt
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer
13 * in this position and unchanged.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. The name of the author may not be used to endorse or promote products
18 * derived from this software without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/fcntl.h>
43 #include <sys/resourcevar.h>
44 #include <sys/rwlock.h>
45 #include <sys/syscallsubr.h>
46 #include <sys/sysent.h>
47 #include <sys/sysproto.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_object.h>
54 #include <compat/linux/linux_emul.h>
55 #include <compat/linux/linux_mmap.h>
56 #include <compat/linux/linux_persona.h>
57 #include <compat/linux/linux_util.h>
59 #define STACK_SIZE (2 * 1024 * 1024)
60 #define GUARD_SIZE (4 * PAGE_SIZE)
62 #if defined(__amd64__)
63 static void linux_fixup_prot(struct thread *td, int *prot);
67 linux_mmap_check_fp(struct file *fp, int flags, int prot, int maxprot)
70 /* Linux mmap() just fails for O_WRONLY files */
71 if ((fp->f_flag & FREAD) == 0)
78 linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
79 int flags, int fd, off_t pos)
81 struct mmap_req mr, mr_fixed;
82 struct proc *p = td->td_proc;
83 struct vmspace *vms = td->td_proc->p_vmspace;
86 LINUX_CTR6(mmap2, "0x%lx, %ld, %ld, 0x%08lx, %ld, 0x%lx",
87 addr, len, prot, flags, fd, pos);
94 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE
96 if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE)))
99 if (flags & LINUX_MAP_SHARED)
100 bsd_flags |= MAP_SHARED;
101 if (flags & LINUX_MAP_PRIVATE)
102 bsd_flags |= MAP_PRIVATE;
103 if (flags & LINUX_MAP_FIXED)
104 bsd_flags |= MAP_FIXED;
105 if (flags & LINUX_MAP_ANON) {
106 /* Enforce pos to be on page boundary, then ignore. */
107 if ((pos & PAGE_MASK) != 0)
110 bsd_flags |= MAP_ANON;
112 bsd_flags |= MAP_NOSYNC;
113 if (flags & LINUX_MAP_GROWSDOWN)
114 bsd_flags |= MAP_STACK;
116 #if defined(__amd64__)
118 * According to the Linux mmap(2) man page, "MAP_32BIT flag
119 * is ignored when MAP_FIXED is set."
121 if ((flags & LINUX_MAP_32BIT) && (flags & LINUX_MAP_FIXED) == 0)
122 bsd_flags |= MAP_32BIT;
125 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC
126 * on Linux/i386 if the binary requires executable stack.
127 * We do this only for IA32 emulation as on native i386 this is does not
128 * make sense without PAE.
130 * XXX. Linux checks that the file system is not mounted with noexec.
132 linux_fixup_prot(td, &prot);
135 /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */
136 fd = (bsd_flags & MAP_ANON) ? -1 : fd;
137 if (flags & LINUX_MAP_GROWSDOWN) {
139 * The Linux MAP_GROWSDOWN option does not limit auto
140 * growth of the region. Linux mmap with this option
141 * takes as addr the initial BOS, and as len, the initial
142 * region size. It can then grow down from addr without
143 * limit. However, Linux threads has an implicit internal
144 * limit to stack size of STACK_SIZE. Its just not
145 * enforced explicitly in Linux. But, here we impose
146 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack
147 * region, since we can do this with our mmap.
149 * Our mmap with MAP_STACK takes addr as the maximum
150 * downsize limit on BOS, and as len the max size of
151 * the region. It then maps the top SGROWSIZ bytes,
152 * and auto grows the region down, up to the limit
155 * If we don't use the MAP_STACK option, the effect
156 * of this code is to allocate a stack region of a
157 * fixed size of (STACK_SIZE - GUARD_SIZE).
160 if ((caddr_t)addr + len > vms->vm_maxsaddr) {
162 * Some Linux apps will attempt to mmap
163 * thread stacks near the top of their
164 * address space. If their TOS is greater
165 * than vm_maxsaddr, vm_map_growstack()
166 * will confuse the thread stack with the
167 * process stack and deliver a SEGV if they
168 * attempt to grow the thread stack past their
169 * current stacksize rlimit. To avoid this,
170 * adjust vm_maxsaddr upwards to reflect
171 * the current stacksize rlimit rather
172 * than the maximum possible stacksize.
173 * It would be better to adjust the
174 * mmap'ed region, but some apps do not check
175 * mmap's return value.
178 vms->vm_maxsaddr = (char *)round_page(vms->vm_stacktop) -
179 lim_cur_proc(p, RLIMIT_STACK);
184 * This gives us our maximum stack size and a new BOS.
185 * If we're using VM_STACK, then mmap will just map
186 * the top SGROWSIZ bytes, and let the stack grow down
187 * to the limit at BOS. If we're not using VM_STACK
188 * we map the full stack, since we don't have a way
191 if (len <= STACK_SIZE - GUARD_SIZE) {
192 addr = addr - (STACK_SIZE - GUARD_SIZE - len);
193 len = STACK_SIZE - GUARD_SIZE;
198 * FreeBSD is free to ignore the address hint if MAP_FIXED wasn't
199 * passed. However, some Linux applications, like the ART runtime,
200 * depend on the hint. If the MAP_FIXED wasn't passed, but the
201 * address is not zero, try with MAP_FIXED and MAP_EXCL first,
202 * and fall back to the normal behaviour if that fails.
204 mr = (struct mmap_req) {
208 .mr_flags = bsd_flags,
211 .mr_check_fp_fn = linux_mmap_check_fp,
213 if (addr != 0 && (bsd_flags & MAP_FIXED) == 0 &&
214 (bsd_flags & MAP_EXCL) == 0) {
216 mr_fixed.mr_flags |= MAP_FIXED | MAP_EXCL;
217 error = kern_mmap(td, &mr_fixed);
222 error = kern_mmap(td, &mr);
224 LINUX_CTR2(mmap2, "return: %d (%p)", error, td->td_retval[0]);
230 linux_mprotect_common(struct thread *td, uintptr_t addr, size_t len, int prot)
233 /* XXX Ignore PROT_GROWSDOWN and PROT_GROWSUP for now. */
234 prot &= ~(LINUX_PROT_GROWSDOWN | LINUX_PROT_GROWSUP);
235 if ((prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
238 #if defined(__amd64__)
239 linux_fixup_prot(td, &prot);
241 return (kern_mprotect(td, addr, len, prot));
245 * Implement Linux madvise(MADV_DONTNEED), which has unusual semantics: for
246 * anonymous memory, pages in the range are immediately discarded.
249 linux_madvise_dontneed(struct thread *td, vm_offset_t start, vm_offset_t end)
252 vm_map_entry_t entry;
253 vm_object_t backing_object, object;
254 vm_offset_t estart, eend;
255 vm_pindex_t pstart, pend;
258 map = &td->td_proc->p_vmspace->vm_map;
260 if (!vm_map_range_valid(map, start, end))
262 start = trunc_page(start);
263 end = round_page(end);
266 vm_map_lock_read(map);
267 if (!vm_map_lookup_entry(map, start, &entry))
268 entry = vm_map_entry_succ(entry);
269 for (; entry->start < end; entry = vm_map_entry_succ(entry)) {
270 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
273 if (entry->wired_count != 0) {
278 object = entry->object.vm_object;
281 if ((object->flags & (OBJ_UNMANAGED | OBJ_FICTITIOUS)) != 0)
284 pstart = OFF_TO_IDX(entry->offset);
285 if (start > entry->start) {
286 pstart += atop(start - entry->start);
289 estart = entry->start;
291 pend = OFF_TO_IDX(entry->offset) +
292 atop(entry->end - entry->start);
293 if (entry->end > end) {
294 pend -= atop(entry->end - end);
300 if ((object->flags & (OBJ_ANON | OBJ_ONEMAPPING)) ==
301 (OBJ_ANON | OBJ_ONEMAPPING)) {
303 * Singly-mapped anonymous memory is discarded. This
304 * does not match Linux's semantics when the object
305 * belongs to a shadow chain of length > 1, since
306 * subsequent faults may retrieve pages from an
307 * intermediate anonymous object. However, handling
308 * this case correctly introduces a fair bit of
311 VM_OBJECT_WLOCK(object);
312 if ((object->flags & OBJ_ONEMAPPING) != 0) {
313 vm_object_collapse(object);
314 vm_object_page_remove(object, pstart, pend, 0);
315 backing_object = object->backing_object;
316 if (backing_object != NULL &&
317 (backing_object->flags & OBJ_ANON) != 0)
319 "possibly incorrect MADV_DONTNEED");
320 VM_OBJECT_WUNLOCK(object);
323 VM_OBJECT_WUNLOCK(object);
327 * Handle shared mappings. Remove them outright instead of
328 * calling pmap_advise(), for consistency with Linux.
330 pmap_remove(map->pmap, estart, eend);
331 vm_object_madvise(object, pstart, pend, MADV_DONTNEED);
333 vm_map_unlock_read(map);
339 linux_madvise_common(struct thread *td, uintptr_t addr, size_t len, int behav)
343 case LINUX_MADV_NORMAL:
344 return (kern_madvise(td, addr, len, MADV_NORMAL));
345 case LINUX_MADV_RANDOM:
346 return (kern_madvise(td, addr, len, MADV_RANDOM));
347 case LINUX_MADV_SEQUENTIAL:
348 return (kern_madvise(td, addr, len, MADV_SEQUENTIAL));
349 case LINUX_MADV_WILLNEED:
350 return (kern_madvise(td, addr, len, MADV_WILLNEED));
351 case LINUX_MADV_DONTNEED:
352 return (linux_madvise_dontneed(td, addr, addr + len));
353 case LINUX_MADV_FREE:
354 return (kern_madvise(td, addr, len, MADV_FREE));
355 case LINUX_MADV_REMOVE:
356 linux_msg(curthread, "unsupported madvise MADV_REMOVE");
358 case LINUX_MADV_DONTFORK:
359 return (kern_minherit(td, addr, len, INHERIT_NONE));
360 case LINUX_MADV_DOFORK:
361 return (kern_minherit(td, addr, len, INHERIT_COPY));
362 case LINUX_MADV_MERGEABLE:
363 linux_msg(curthread, "unsupported madvise MADV_MERGEABLE");
365 case LINUX_MADV_UNMERGEABLE:
366 /* We don't merge anyway. */
368 case LINUX_MADV_HUGEPAGE:
369 /* Ignored; on FreeBSD huge pages are always on. */
371 case LINUX_MADV_NOHUGEPAGE:
374 * Don't warn - Firefox uses it a lot, and in real Linux it's
375 * an optional feature.
377 linux_msg(curthread, "unsupported madvise MADV_NOHUGEPAGE");
380 case LINUX_MADV_DONTDUMP:
381 return (kern_madvise(td, addr, len, MADV_NOCORE));
382 case LINUX_MADV_DODUMP:
383 return (kern_madvise(td, addr, len, MADV_CORE));
384 case LINUX_MADV_WIPEONFORK:
385 return (kern_minherit(td, addr, len, INHERIT_ZERO));
386 case LINUX_MADV_KEEPONFORK:
387 return (kern_minherit(td, addr, len, INHERIT_COPY));
388 case LINUX_MADV_HWPOISON:
389 linux_msg(curthread, "unsupported madvise MADV_HWPOISON");
391 case LINUX_MADV_SOFT_OFFLINE:
392 linux_msg(curthread, "unsupported madvise MADV_SOFT_OFFLINE");
396 * -1 is sometimes used as a dummy value to detect simplistic
397 * madvise(2) stub implementations. This safeguard is used by
398 * BoringSSL, for example, before assuming MADV_WIPEONFORK is
399 * safe to use. Don't produce an "unsupported" error message
400 * for this special dummy value, which is unlikely to be used
401 * by any new advisory behavior feature.
405 linux_msg(curthread, "unsupported madvise behav %d", behav);
410 #if defined(__amd64__)
412 linux_fixup_prot(struct thread *td, int *prot)
414 struct linux_pemuldata *pem;
416 if (SV_PROC_FLAG(td->td_proc, SV_ILP32) && *prot & PROT_READ) {
417 pem = pem_find(td->td_proc);
418 if (pem->persona & LINUX_READ_IMPLIES_EXEC)