4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
7 * See Documentation/nommu-mmap.txt
9 * Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30 #include <linux/vs_base.h>
31 #include <linux/vs_memory.h>
33 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
39 unsigned long max_mapnr;
40 unsigned long num_physpages;
41 unsigned long askedalloc, realalloc;
42 atomic_t vm_committed_space = ATOMIC_INIT(0);
43 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
44 int sysctl_overcommit_ratio = 50; /* default is 50% */
45 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
46 int heap_stack_gap = 0;
48 EXPORT_SYMBOL(mem_map);
49 EXPORT_SYMBOL(__vm_enough_memory);
51 /* list of shareable VMAs */
52 struct rb_root nommu_vma_tree = RB_ROOT;
53 DECLARE_RWSEM(nommu_vma_sem);
55 struct vm_operations_struct generic_file_vm_ops = {
59 EXPORT_SYMBOL(vmalloc_to_page);
60 EXPORT_SYMBOL(vmalloc_32);
62 EXPORT_SYMBOL(vunmap);
65 * Handle all mappings that got truncated by a "truncate()"
68 * NOTE! We have to be ready to update the memory sharing
69 * between the file and the memory map for a potential last
70 * incomplete page. Ugly, but necessary.
72 int vmtruncate(struct inode *inode, loff_t offset)
74 struct address_space *mapping = inode->i_mapping;
77 if (inode->i_size < offset)
79 i_size_write(inode, offset);
81 truncate_inode_pages(mapping, offset);
85 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
86 if (limit != RLIM_INFINITY && offset > limit)
88 if (offset > inode->i_sb->s_maxbytes)
90 i_size_write(inode, offset);
93 if (inode->i_op && inode->i_op->truncate)
94 inode->i_op->truncate(inode);
97 send_sig(SIGXFSZ, current, 0);
102 EXPORT_SYMBOL(vmtruncate);
105 * Return the total memory allocated for this pointer, not
106 * just what the caller asked for.
108 * Doesn't have to be accurate, i.e. may have races.
110 unsigned int kobjsize(const void *objp)
114 if (!objp || !((page = virt_to_page(objp))))
120 BUG_ON(page->index < 0);
121 BUG_ON(page->index >= MAX_ORDER);
123 return (PAGE_SIZE << page->index);
127 * get a list of pages in an address range belonging to the specified process
128 * and indicate the VMA that covers each page
129 * - this is potentially dodgy as we may end incrementing the page count of a
130 * slab page or a secondary page from a compound page
131 * - don't permit access to VMAs that don't support it, such as I/O mappings
133 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
134 unsigned long start, int len, int write, int force,
135 struct page **pages, struct vm_area_struct **vmas)
137 struct vm_area_struct *vma;
138 unsigned long vm_flags;
141 /* calculate required read or write permissions.
142 * - if 'force' is set, we only require the "MAY" flags.
144 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
145 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
147 for (i = 0; i < len; i++) {
148 vma = find_vma(mm, start);
150 goto finish_or_fault;
152 /* protect what we can, including chardevs */
153 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
154 !(vm_flags & vma->vm_flags))
155 goto finish_or_fault;
158 pages[i] = virt_to_page(start);
160 page_cache_get(pages[i]);
170 return i ? : -EFAULT;
173 EXPORT_SYMBOL(get_user_pages);
175 DEFINE_RWLOCK(vmlist_lock);
176 struct vm_struct *vmlist;
178 void vfree(void *addr)
183 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
186 * kmalloc doesn't like __GFP_HIGHMEM for some reason
188 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
191 struct page * vmalloc_to_page(void *addr)
193 return virt_to_page(addr);
196 unsigned long vmalloc_to_pfn(void *addr)
198 return page_to_pfn(virt_to_page(addr));
202 long vread(char *buf, char *addr, unsigned long count)
204 memcpy(buf, addr, count);
208 long vwrite(char *buf, char *addr, unsigned long count)
210 /* Don't allow overflow */
211 if ((unsigned long) addr + count < count)
212 count = -(unsigned long) addr;
214 memcpy(addr, buf, count);
219 * vmalloc - allocate virtually continguos memory
221 * @size: allocation size
223 * Allocate enough pages to cover @size from the page level
224 * allocator and map them into continguos kernel virtual space.
226 * For tight control over page level allocator and protection flags
227 * use __vmalloc() instead.
229 void *vmalloc(unsigned long size)
231 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
233 EXPORT_SYMBOL(vmalloc);
235 void *vmalloc_node(unsigned long size, int node)
237 return vmalloc(size);
239 EXPORT_SYMBOL(vmalloc_node);
242 * vmalloc_32 - allocate virtually continguos memory (32bit addressable)
244 * @size: allocation size
246 * Allocate enough 32bit PA addressable pages to cover @size from the
247 * page level allocator and map them into continguos kernel virtual space.
249 void *vmalloc_32(unsigned long size)
251 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
254 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
260 void vunmap(void *addr)
266 * sys_brk() for the most part doesn't need the global kernel
267 * lock, except when an application is doing something nasty
268 * like trying to un-brk an area that has already been mapped
269 * to a regular file. in this case, the unmapping will need
270 * to invoke file system routines that need the global lock.
272 asmlinkage unsigned long sys_brk(unsigned long brk)
274 struct mm_struct *mm = current->mm;
276 if (brk < mm->start_brk || brk > mm->context.end_brk)
283 * Always allow shrinking brk
285 if (brk <= mm->brk) {
291 * Ok, looks good - let it rip.
293 return mm->brk = brk;
297 static void show_process_blocks(void)
299 struct vm_list_struct *vml;
301 printk("Process blocks %d:", current->pid);
303 for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) {
304 printk(" %p: %p", vml, vml->vma);
306 printk(" (%d @%lx #%d)",
307 kobjsize((void *) vml->vma->vm_start),
309 atomic_read(&vml->vma->vm_usage));
310 printk(vml->next ? " ->" : ".\n");
316 * add a VMA into a process's mm_struct in the appropriate place in the list
317 * - should be called with mm->mmap_sem held writelocked
319 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
321 struct vm_list_struct **ppv;
323 for (ppv = ¤t->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
324 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
332 * look up the first VMA in which addr resides, NULL if none
333 * - should be called with mm->mmap_sem at least held readlocked
335 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
337 struct vm_list_struct *loop, *vml;
339 /* search the vm_start ordered list */
341 for (loop = mm->context.vmlist; loop; loop = loop->next) {
342 if (loop->vma->vm_start > addr)
347 if (vml && vml->vma->vm_end > addr)
352 EXPORT_SYMBOL(find_vma);
356 * - we don't extend stack VMAs under NOMMU conditions
358 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
360 return find_vma(mm, addr);
364 * look up the first VMA exactly that exactly matches addr
365 * - should be called with mm->mmap_sem at least held readlocked
367 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
370 struct vm_list_struct *vml;
372 /* search the vm_start ordered list */
373 for (vml = mm->context.vmlist; vml; vml = vml->next) {
374 if (vml->vma->vm_start == addr)
376 if (vml->vma->vm_start > addr)
384 * find a VMA in the global tree
386 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
388 struct vm_area_struct *vma;
389 struct rb_node *n = nommu_vma_tree.rb_node;
392 vma = rb_entry(n, struct vm_area_struct, vm_rb);
394 if (start < vma->vm_start)
396 else if (start > vma->vm_start)
406 * add a VMA in the global tree
408 static void add_nommu_vma(struct vm_area_struct *vma)
410 struct vm_area_struct *pvma;
411 struct address_space *mapping;
412 struct rb_node **p = &nommu_vma_tree.rb_node;
413 struct rb_node *parent = NULL;
415 /* add the VMA to the mapping */
417 mapping = vma->vm_file->f_mapping;
419 flush_dcache_mmap_lock(mapping);
420 vma_prio_tree_insert(vma, &mapping->i_mmap);
421 flush_dcache_mmap_unlock(mapping);
424 /* add the VMA to the master list */
427 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
429 if (vma->vm_start < pvma->vm_start) {
432 else if (vma->vm_start > pvma->vm_start) {
436 /* mappings are at the same address - this can only
437 * happen for shared-mem chardevs and shared file
438 * mappings backed by ramfs/tmpfs */
439 BUG_ON(!(pvma->vm_flags & VM_SHARED));
450 rb_link_node(&vma->vm_rb, parent, p);
451 rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
455 * delete a VMA from the global list
457 static void delete_nommu_vma(struct vm_area_struct *vma)
459 struct address_space *mapping;
461 /* remove the VMA from the mapping */
463 mapping = vma->vm_file->f_mapping;
465 flush_dcache_mmap_lock(mapping);
466 vma_prio_tree_remove(vma, &mapping->i_mmap);
467 flush_dcache_mmap_unlock(mapping);
470 /* remove from the master list */
471 rb_erase(&vma->vm_rb, &nommu_vma_tree);
475 * determine whether a mapping should be permitted and, if so, what sort of
476 * mapping we're capable of supporting
478 static int validate_mmap_request(struct file *file,
484 unsigned long *_capabilities)
486 unsigned long capabilities;
487 unsigned long reqprot = prot;
490 /* do the simple checks first */
491 if (flags & MAP_FIXED || addr) {
493 "%d: Can't do fixed-address/overlay mmap of RAM\n",
498 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
499 (flags & MAP_TYPE) != MAP_SHARED)
505 /* Careful about overflows.. */
506 len = PAGE_ALIGN(len);
507 if (!len || len > TASK_SIZE)
510 /* offset overflow? */
511 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
515 /* validate file mapping requests */
516 struct address_space *mapping;
518 /* files must support mmap */
519 if (!file->f_op || !file->f_op->mmap)
522 /* work out if what we've got could possibly be shared
523 * - we support chardevs that provide their own "memory"
524 * - we support files/blockdevs that are memory backed
526 mapping = file->f_mapping;
528 mapping = file->f_path.dentry->d_inode->i_mapping;
531 if (mapping && mapping->backing_dev_info)
532 capabilities = mapping->backing_dev_info->capabilities;
535 /* no explicit capabilities set, so assume some
537 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
540 capabilities = BDI_CAP_MAP_COPY;
555 /* eliminate any capabilities that we can't support on this
557 if (!file->f_op->get_unmapped_area)
558 capabilities &= ~BDI_CAP_MAP_DIRECT;
559 if (!file->f_op->read)
560 capabilities &= ~BDI_CAP_MAP_COPY;
562 if (flags & MAP_SHARED) {
563 /* do checks for writing, appending and locking */
564 if ((prot & PROT_WRITE) &&
565 !(file->f_mode & FMODE_WRITE))
568 if (IS_APPEND(file->f_path.dentry->d_inode) &&
569 (file->f_mode & FMODE_WRITE))
572 if (locks_verify_locked(file->f_path.dentry->d_inode))
575 if (!(capabilities & BDI_CAP_MAP_DIRECT))
578 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
579 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
580 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
582 printk("MAP_SHARED not completely supported on !MMU\n");
586 /* we mustn't privatise shared mappings */
587 capabilities &= ~BDI_CAP_MAP_COPY;
590 /* we're going to read the file into private memory we
592 if (!(capabilities & BDI_CAP_MAP_COPY))
595 /* we don't permit a private writable mapping to be
596 * shared with the backing device */
597 if (prot & PROT_WRITE)
598 capabilities &= ~BDI_CAP_MAP_DIRECT;
601 /* handle executable mappings and implied executable
603 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
604 if (prot & PROT_EXEC)
607 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
608 /* handle implication of PROT_EXEC by PROT_READ */
609 if (current->personality & READ_IMPLIES_EXEC) {
610 if (capabilities & BDI_CAP_EXEC_MAP)
614 else if ((prot & PROT_READ) &&
615 (prot & PROT_EXEC) &&
616 !(capabilities & BDI_CAP_EXEC_MAP)
618 /* backing file is not executable, try to copy */
619 capabilities &= ~BDI_CAP_MAP_DIRECT;
623 /* anonymous mappings are always memory backed and can be
626 capabilities = BDI_CAP_MAP_COPY;
628 /* handle PROT_EXEC implication by PROT_READ */
629 if ((prot & PROT_READ) &&
630 (current->personality & READ_IMPLIES_EXEC))
634 /* allow the security API to have its say */
635 ret = security_file_mmap(file, reqprot, prot, flags);
640 *_capabilities = capabilities;
645 * we've determined that we can make the mapping, now translate what we
646 * now know into VMA flags
648 static unsigned long determine_vm_flags(struct file *file,
651 unsigned long capabilities)
653 unsigned long vm_flags;
655 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
656 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
657 /* vm_flags |= mm->def_flags; */
659 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
660 /* attempt to share read-only copies of mapped file chunks */
661 if (file && !(prot & PROT_WRITE))
662 vm_flags |= VM_MAYSHARE;
665 /* overlay a shareable mapping on the backing device or inode
666 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
668 if (flags & MAP_SHARED)
669 vm_flags |= VM_MAYSHARE | VM_SHARED;
670 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
671 vm_flags |= VM_MAYSHARE;
674 /* refuse to let anyone share private mappings with this process if
675 * it's being traced - otherwise breakpoints set in it may interfere
676 * with another untraced process
678 if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
679 vm_flags &= ~VM_MAYSHARE;
685 * set up a shared mapping on a file
687 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
691 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
695 /* getting an ENOSYS error indicates that direct mmap isn't
696 * possible (as opposed to tried but failed) so we'll fall
697 * through to making a private copy of the data and mapping
703 * set up a private mapping or an anonymous shared mapping
705 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
710 /* invoke the file's mapping function so that it can keep track of
711 * shared mappings on devices or memory
712 * - VM_MAYSHARE will be set if it may attempt to share
715 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
716 if (ret != -ENOSYS) {
717 /* shouldn't return success if we're not sharing */
718 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
719 return ret; /* success or a real error */
722 /* getting an ENOSYS error indicates that direct mmap isn't
723 * possible (as opposed to tried but failed) so we'll try to
724 * make a private copy of the data and map that instead */
727 /* allocate some memory to hold the mapping
728 * - note that this may not return a page-aligned address if the object
729 * we're allocating is smaller than a page
731 base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
735 vma->vm_start = (unsigned long) base;
736 vma->vm_end = vma->vm_start + len;
737 vma->vm_flags |= VM_MAPPED_COPY;
740 if (len + WARN_ON_SLACK <= kobjsize(result))
741 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
742 len, current->pid, kobjsize(result) - len);
746 /* read the contents of a file into the copy */
750 fpos = vma->vm_pgoff;
755 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
761 /* clear the last little bit */
763 memset(base + ret, 0, len - ret);
766 /* if it's an anonymous mapping, then just clear it */
767 memset(base, 0, len);
778 printk("Allocation of length %lu from process %d failed\n",
785 * handle mapping creation for uClinux
787 unsigned long do_mmap_pgoff(struct file *file,
794 struct vm_list_struct *vml = NULL;
795 struct vm_area_struct *vma = NULL;
797 unsigned long capabilities, vm_flags;
801 /* decide whether we should attempt the mapping, and if so what sort of
803 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
808 /* we've determined that we can make the mapping, now translate what we
809 * now know into VMA flags */
810 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
812 /* we're going to need to record the mapping if it works */
813 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
815 goto error_getting_vml;
817 down_write(&nommu_vma_sem);
819 /* if we want to share, we need to check for VMAs created by other
820 * mmap() calls that overlap with our proposed mapping
821 * - we can only share with an exact match on most regular files
822 * - shared mappings on character devices and memory backed files are
823 * permitted to overlap inexactly as far as we are concerned for in
824 * these cases, sharing is handled in the driver or filesystem rather
827 if (vm_flags & VM_MAYSHARE) {
828 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
829 unsigned long vmpglen;
831 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
832 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
834 if (!(vma->vm_flags & VM_MAYSHARE))
837 /* search for overlapping mappings on the same file */
838 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
841 if (vma->vm_pgoff >= pgoff + pglen)
844 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
845 vmpglen >>= PAGE_SHIFT;
846 if (pgoff >= vma->vm_pgoff + vmpglen)
849 /* handle inexactly overlapping matches between mappings */
850 if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
851 if (!(capabilities & BDI_CAP_MAP_DIRECT))
852 goto sharing_violation;
856 /* we've found a VMA we can share */
857 atomic_inc(&vma->vm_usage);
860 result = (void *) vma->vm_start;
866 /* obtain the address at which to make a shared mapping
867 * - this is the hook for quasi-memory character devices to
868 * tell us the location of a shared mapping
870 if (file && file->f_op->get_unmapped_area) {
871 addr = file->f_op->get_unmapped_area(file, addr, len,
873 if (IS_ERR((void *) addr)) {
875 if (ret != (unsigned long) -ENOSYS)
878 /* the driver refused to tell us where to site
879 * the mapping so we'll have to attempt to copy
881 ret = (unsigned long) -ENODEV;
882 if (!(capabilities & BDI_CAP_MAP_COPY))
885 capabilities &= ~BDI_CAP_MAP_DIRECT;
890 /* we're going to need a VMA struct as well */
891 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
893 goto error_getting_vma;
895 INIT_LIST_HEAD(&vma->anon_vma_node);
896 atomic_set(&vma->vm_usage, 1);
900 vma->vm_flags = vm_flags;
901 vma->vm_start = addr;
902 vma->vm_end = addr + len;
903 vma->vm_pgoff = pgoff;
907 /* set up the mapping */
908 if (file && vma->vm_flags & VM_SHARED)
909 ret = do_mmap_shared_file(vma, len);
911 ret = do_mmap_private(vma, len);
915 /* okay... we have a mapping; now we have to register it */
916 result = (void *) vma->vm_start;
918 if (vma->vm_flags & VM_MAPPED_COPY) {
919 realalloc += kobjsize(result);
923 realalloc += kobjsize(vma);
924 askedalloc += sizeof(*vma);
926 vx_vmpages_add(current->mm, len >> PAGE_SHIFT);
931 realalloc += kobjsize(vml);
932 askedalloc += sizeof(*vml);
934 add_vma_to_mm(current->mm, vml);
936 up_write(&nommu_vma_sem);
938 if (prot & PROT_EXEC)
939 flush_icache_range((unsigned long) result,
940 (unsigned long) result + len);
943 printk("do_mmap:\n");
944 show_process_blocks();
947 return (unsigned long) result;
950 up_write(&nommu_vma_sem);
960 up_write(&nommu_vma_sem);
961 printk("Attempt to share mismatched mappings\n");
966 up_write(&nommu_vma_sem);
968 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
974 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
981 * handle mapping disposal for uClinux
983 static void put_vma(struct vm_area_struct *vma)
986 down_write(&nommu_vma_sem);
988 if (atomic_dec_and_test(&vma->vm_usage)) {
989 delete_nommu_vma(vma);
991 if (vma->vm_ops && vma->vm_ops->close)
992 vma->vm_ops->close(vma);
994 /* IO memory and memory shared directly out of the pagecache from
995 * ramfs/tmpfs mustn't be released here */
996 if (vma->vm_flags & VM_MAPPED_COPY) {
997 realalloc -= kobjsize((void *) vma->vm_start);
998 askedalloc -= vma->vm_end - vma->vm_start;
999 kfree((void *) vma->vm_start);
1002 realalloc -= kobjsize(vma);
1003 askedalloc -= sizeof(*vma);
1010 up_write(&nommu_vma_sem);
1016 * - under NOMMU conditions the parameters must match exactly to the mapping to
1019 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1021 struct vm_list_struct *vml, **parent;
1022 unsigned long end = addr + len;
1025 printk("do_munmap:\n");
1028 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1029 if ((*parent)->vma->vm_start > addr)
1031 if ((*parent)->vma->vm_start == addr &&
1032 ((len == 0) || ((*parent)->vma->vm_end == end)))
1036 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1037 current->pid, current->comm, (void *) addr);
1045 *parent = vml->next;
1046 realalloc -= kobjsize(vml);
1047 askedalloc -= sizeof(*vml);
1050 update_hiwater_vm(mm);
1051 vx_vmpages_sub(mm, len >> PAGE_SHIFT);
1054 show_process_blocks();
1060 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1063 struct mm_struct *mm = current->mm;
1065 down_write(&mm->mmap_sem);
1066 ret = do_munmap(mm, addr, len);
1067 up_write(&mm->mmap_sem);
1072 * Release all mappings
1074 void exit_mmap(struct mm_struct * mm)
1076 struct vm_list_struct *tmp;
1080 printk("Exit_mmap:\n");
1083 vx_vmpages_sub(mm, mm->total_vm);
1085 while ((tmp = mm->context.vmlist)) {
1086 mm->context.vmlist = tmp->next;
1089 realalloc -= kobjsize(tmp);
1090 askedalloc -= sizeof(*tmp);
1095 show_process_blocks();
1100 unsigned long do_brk(unsigned long addr, unsigned long len)
1106 * expand (or shrink) an existing mapping, potentially moving it at the same
1107 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1109 * under NOMMU conditions, we only permit changing a mapping's size, and only
1110 * as long as it stays within the hole allocated by the kmalloc() call in
1111 * do_mmap_pgoff() and the block is not shareable
1113 * MREMAP_FIXED is not supported under NOMMU conditions
1115 unsigned long do_mremap(unsigned long addr,
1116 unsigned long old_len, unsigned long new_len,
1117 unsigned long flags, unsigned long new_addr)
1119 struct vm_area_struct *vma;
1121 /* insanity checks first */
1123 return (unsigned long) -EINVAL;
1125 if (flags & MREMAP_FIXED && new_addr != addr)
1126 return (unsigned long) -EINVAL;
1128 vma = find_vma_exact(current->mm, addr);
1130 return (unsigned long) -EINVAL;
1132 if (vma->vm_end != vma->vm_start + old_len)
1133 return (unsigned long) -EFAULT;
1135 if (vma->vm_flags & VM_MAYSHARE)
1136 return (unsigned long) -EPERM;
1138 if (new_len > kobjsize((void *) addr))
1139 return (unsigned long) -ENOMEM;
1141 /* all checks complete - do it */
1142 vma->vm_end = vma->vm_start + new_len;
1144 askedalloc -= old_len;
1145 askedalloc += new_len;
1147 return vma->vm_start;
1150 asmlinkage unsigned long sys_mremap(unsigned long addr,
1151 unsigned long old_len, unsigned long new_len,
1152 unsigned long flags, unsigned long new_addr)
1156 down_write(¤t->mm->mmap_sem);
1157 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1158 up_write(¤t->mm->mmap_sem);
1162 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1163 unsigned int foll_flags)
1168 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1169 unsigned long to, unsigned long size, pgprot_t prot)
1171 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1174 EXPORT_SYMBOL(remap_pfn_range);
1176 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1180 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1181 unsigned long len, unsigned long pgoff, unsigned long flags)
1186 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1190 void unmap_mapping_range(struct address_space *mapping,
1191 loff_t const holebegin, loff_t const holelen,
1195 EXPORT_SYMBOL(unmap_mapping_range);
1198 * Check that a process has enough memory to allocate a new virtual
1199 * mapping. 0 means there is enough memory for the allocation to
1200 * succeed and -ENOMEM implies there is not.
1202 * We currently support three overcommit policies, which are set via the
1203 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1205 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1206 * Additional code 2002 Jul 20 by Robert Love.
1208 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1210 * Note this is a helper function intended to be used by LSMs which
1211 * wish to use this logic.
1213 int __vm_enough_memory(long pages, int cap_sys_admin)
1215 unsigned long free, allowed;
1217 vm_acct_memory(pages);
1220 * Sometimes we want to use more memory than we have
1222 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1225 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1228 free = global_page_state(NR_FILE_PAGES);
1229 free += nr_swap_pages;
1232 * Any slabs which are created with the
1233 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1234 * which are reclaimable, under pressure. The dentry
1235 * cache and most inode caches should fall into this
1237 free += global_page_state(NR_SLAB_RECLAIMABLE);
1240 * Leave the last 3% for root
1249 * nr_free_pages() is very expensive on large systems,
1250 * only call if we're about to fail.
1252 n = nr_free_pages();
1255 * Leave reserved pages. The pages are not for anonymous pages.
1257 if (n <= totalreserve_pages)
1260 n -= totalreserve_pages;
1263 * Leave the last 3% for root
1275 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1277 * Leave the last 3% for root
1280 allowed -= allowed / 32;
1281 allowed += total_swap_pages;
1283 /* Don't let a single process grow too big:
1284 leave 3% of the size of this process for other processes */
1285 allowed -= current->mm->total_vm / 32;
1288 * cast `allowed' as a signed long because vm_committed_space
1289 * sometimes has a negative value
1291 if (atomic_read(&vm_committed_space) < (long)allowed)
1294 vm_unacct_memory(pages);
1299 int in_gate_area_no_task(unsigned long addr)
1304 struct page *filemap_nopage(struct vm_area_struct *area,
1305 unsigned long address, int *type)
1312 * Access another process' address space.
1313 * - source/target buffer must be kernel space
1315 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1317 struct vm_area_struct *vma;
1318 struct mm_struct *mm;
1320 if (addr + len < addr)
1323 mm = get_task_mm(tsk);
1327 down_read(&mm->mmap_sem);
1329 /* the access must start within one of the target process's mappings */
1330 vma = find_vma(mm, addr);
1332 /* don't overrun this mapping */
1333 if (addr + len >= vma->vm_end)
1334 len = vma->vm_end - addr;
1336 /* only read or write mappings where it is permitted */
1337 if (write && vma->vm_flags & VM_MAYWRITE)
1338 len -= copy_to_user((void *) addr, buf, len);
1339 else if (!write && vma->vm_flags & VM_MAYREAD)
1340 len -= copy_from_user(buf, (void *) addr, len);
1347 up_read(&mm->mmap_sem);