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>
13 * Copyright (c) 2007 Paul Mundt <lethal@linux-sh.org>
16 #include <linux/module.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
33 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
41 unsigned long max_mapnr;
42 unsigned long num_physpages;
43 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
44 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
45 int sysctl_overcommit_ratio = 50; /* default is 50% */
46 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
47 int heap_stack_gap = 0;
49 EXPORT_SYMBOL(mem_map);
50 EXPORT_SYMBOL(num_physpages);
52 /* list of shareable VMAs */
53 struct rb_root nommu_vma_tree = RB_ROOT;
54 DECLARE_RWSEM(nommu_vma_sem);
56 struct vm_operations_struct generic_file_vm_ops = {
60 * Handle all mappings that got truncated by a "truncate()"
63 * NOTE! We have to be ready to update the memory sharing
64 * between the file and the memory map for a potential last
65 * incomplete page. Ugly, but necessary.
67 int vmtruncate(struct inode *inode, loff_t offset)
69 struct address_space *mapping = inode->i_mapping;
72 if (inode->i_size < offset)
74 i_size_write(inode, offset);
76 truncate_inode_pages(mapping, offset);
80 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
81 if (limit != RLIM_INFINITY && offset > limit)
83 if (offset > inode->i_sb->s_maxbytes)
85 i_size_write(inode, offset);
88 if (inode->i_op->truncate)
89 inode->i_op->truncate(inode);
92 send_sig(SIGXFSZ, current, 0);
97 EXPORT_SYMBOL(vmtruncate);
100 * Return the total memory allocated for this pointer, not
101 * just what the caller asked for.
103 * Doesn't have to be accurate, i.e. may have races.
105 unsigned int kobjsize(const void *objp)
110 * If the object we have should not have ksize performed on it,
113 if (!objp || !virt_addr_valid(objp))
116 page = virt_to_head_page(objp);
119 * If the allocator sets PageSlab, we know the pointer came from
126 * The ksize() function is only guaranteed to work for pointers
127 * returned by kmalloc(). So handle arbitrary pointers here.
129 return PAGE_SIZE << compound_order(page);
132 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
133 unsigned long start, int len, int flags,
134 struct page **pages, struct vm_area_struct **vmas)
136 struct vm_area_struct *vma;
137 unsigned long vm_flags;
139 int write = !!(flags & GUP_FLAGS_WRITE);
140 int force = !!(flags & GUP_FLAGS_FORCE);
141 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
143 /* calculate required read or write permissions.
144 * - if 'force' is set, we only require the "MAY" flags.
146 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
147 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
149 for (i = 0; i < len; i++) {
150 vma = find_vma(mm, start);
152 goto finish_or_fault;
154 /* protect what we can, including chardevs */
155 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
156 (!ignore && !(vm_flags & vma->vm_flags)))
157 goto finish_or_fault;
160 pages[i] = virt_to_page(start);
162 page_cache_get(pages[i]);
172 return i ? : -EFAULT;
177 * get a list of pages in an address range belonging to the specified process
178 * and indicate the VMA that covers each page
179 * - this is potentially dodgy as we may end incrementing the page count of a
180 * slab page or a secondary page from a compound page
181 * - don't permit access to VMAs that don't support it, such as I/O mappings
183 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
184 unsigned long start, int len, int write, int force,
185 struct page **pages, struct vm_area_struct **vmas)
190 flags |= GUP_FLAGS_WRITE;
192 flags |= GUP_FLAGS_FORCE;
194 return __get_user_pages(tsk, mm,
198 EXPORT_SYMBOL(get_user_pages);
200 DEFINE_RWLOCK(vmlist_lock);
201 struct vm_struct *vmlist;
203 void vfree(const void *addr)
207 EXPORT_SYMBOL(vfree);
209 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
212 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
213 * returns only a logical address.
215 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
217 EXPORT_SYMBOL(__vmalloc);
219 void *vmalloc_user(unsigned long size)
223 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
226 struct vm_area_struct *vma;
228 down_write(¤t->mm->mmap_sem);
229 vma = find_vma(current->mm, (unsigned long)ret);
231 vma->vm_flags |= VM_USERMAP;
232 up_write(¤t->mm->mmap_sem);
237 EXPORT_SYMBOL(vmalloc_user);
239 struct page *vmalloc_to_page(const void *addr)
241 return virt_to_page(addr);
243 EXPORT_SYMBOL(vmalloc_to_page);
245 unsigned long vmalloc_to_pfn(const void *addr)
247 return page_to_pfn(virt_to_page(addr));
249 EXPORT_SYMBOL(vmalloc_to_pfn);
251 long vread(char *buf, char *addr, unsigned long count)
253 memcpy(buf, addr, count);
257 long vwrite(char *buf, char *addr, unsigned long count)
259 /* Don't allow overflow */
260 if ((unsigned long) addr + count < count)
261 count = -(unsigned long) addr;
263 memcpy(addr, buf, count);
268 * vmalloc - allocate virtually continguos memory
270 * @size: allocation size
272 * Allocate enough pages to cover @size from the page level
273 * allocator and map them into continguos kernel virtual space.
275 * For tight control over page level allocator and protection flags
276 * use __vmalloc() instead.
278 void *vmalloc(unsigned long size)
280 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
282 EXPORT_SYMBOL(vmalloc);
284 void *vmalloc_node(unsigned long size, int node)
286 return vmalloc(size);
288 EXPORT_SYMBOL(vmalloc_node);
290 #ifndef PAGE_KERNEL_EXEC
291 # define PAGE_KERNEL_EXEC PAGE_KERNEL
295 * vmalloc_exec - allocate virtually contiguous, executable memory
296 * @size: allocation size
298 * Kernel-internal function to allocate enough pages to cover @size
299 * the page level allocator and map them into contiguous and
300 * executable kernel virtual space.
302 * For tight control over page level allocator and protection flags
303 * use __vmalloc() instead.
306 void *vmalloc_exec(unsigned long size)
308 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
312 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
313 * @size: allocation size
315 * Allocate enough 32bit PA addressable pages to cover @size from the
316 * page level allocator and map them into continguos kernel virtual space.
318 void *vmalloc_32(unsigned long size)
320 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
322 EXPORT_SYMBOL(vmalloc_32);
325 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
326 * @size: allocation size
328 * The resulting memory area is 32bit addressable and zeroed so it can be
329 * mapped to userspace without leaking data.
331 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
332 * remap_vmalloc_range() are permissible.
334 void *vmalloc_32_user(unsigned long size)
337 * We'll have to sort out the ZONE_DMA bits for 64-bit,
338 * but for now this can simply use vmalloc_user() directly.
340 return vmalloc_user(size);
342 EXPORT_SYMBOL(vmalloc_32_user);
344 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
351 void vunmap(const void *addr)
355 EXPORT_SYMBOL(vunmap);
358 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
361 void __attribute__((weak)) vmalloc_sync_all(void)
365 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
370 EXPORT_SYMBOL(vm_insert_page);
373 * sys_brk() for the most part doesn't need the global kernel
374 * lock, except when an application is doing something nasty
375 * like trying to un-brk an area that has already been mapped
376 * to a regular file. in this case, the unmapping will need
377 * to invoke file system routines that need the global lock.
379 asmlinkage unsigned long sys_brk(unsigned long brk)
381 struct mm_struct *mm = current->mm;
383 if (brk < mm->start_brk || brk > mm->context.end_brk)
390 * Always allow shrinking brk
392 if (brk <= mm->brk) {
398 * Ok, looks good - let it rip.
400 return mm->brk = brk;
404 static void show_process_blocks(void)
406 struct vm_list_struct *vml;
408 printk("Process blocks %d:", current->pid);
410 for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) {
411 printk(" %p: %p", vml, vml->vma);
413 printk(" (%d @%lx #%d)",
414 kobjsize((void *) vml->vma->vm_start),
416 atomic_read(&vml->vma->vm_usage));
417 printk(vml->next ? " ->" : ".\n");
423 * add a VMA into a process's mm_struct in the appropriate place in the list
424 * - should be called with mm->mmap_sem held writelocked
426 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
428 struct vm_list_struct **ppv;
430 for (ppv = ¤t->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
431 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
439 * look up the first VMA in which addr resides, NULL if none
440 * - should be called with mm->mmap_sem at least held readlocked
442 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
444 struct vm_list_struct *loop, *vml;
446 /* search the vm_start ordered list */
448 for (loop = mm->context.vmlist; loop; loop = loop->next) {
449 if (loop->vma->vm_start > addr)
454 if (vml && vml->vma->vm_end > addr)
459 EXPORT_SYMBOL(find_vma);
463 * - we don't extend stack VMAs under NOMMU conditions
465 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
467 return find_vma(mm, addr);
470 int expand_stack(struct vm_area_struct *vma, unsigned long address)
476 * look up the first VMA exactly that exactly matches addr
477 * - should be called with mm->mmap_sem at least held readlocked
479 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
482 struct vm_list_struct *vml;
484 /* search the vm_start ordered list */
485 for (vml = mm->context.vmlist; vml; vml = vml->next) {
486 if (vml->vma->vm_start == addr)
488 if (vml->vma->vm_start > addr)
496 * find a VMA in the global tree
498 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
500 struct vm_area_struct *vma;
501 struct rb_node *n = nommu_vma_tree.rb_node;
504 vma = rb_entry(n, struct vm_area_struct, vm_rb);
506 if (start < vma->vm_start)
508 else if (start > vma->vm_start)
518 * add a VMA in the global tree
520 static void add_nommu_vma(struct vm_area_struct *vma)
522 struct vm_area_struct *pvma;
523 struct address_space *mapping;
524 struct rb_node **p = &nommu_vma_tree.rb_node;
525 struct rb_node *parent = NULL;
527 /* add the VMA to the mapping */
529 mapping = vma->vm_file->f_mapping;
531 flush_dcache_mmap_lock(mapping);
532 vma_prio_tree_insert(vma, &mapping->i_mmap);
533 flush_dcache_mmap_unlock(mapping);
536 /* add the VMA to the master list */
539 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
541 if (vma->vm_start < pvma->vm_start) {
544 else if (vma->vm_start > pvma->vm_start) {
548 /* mappings are at the same address - this can only
549 * happen for shared-mem chardevs and shared file
550 * mappings backed by ramfs/tmpfs */
551 BUG_ON(!(pvma->vm_flags & VM_SHARED));
562 rb_link_node(&vma->vm_rb, parent, p);
563 rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
567 * delete a VMA from the global list
569 static void delete_nommu_vma(struct vm_area_struct *vma)
571 struct address_space *mapping;
573 /* remove the VMA from the mapping */
575 mapping = vma->vm_file->f_mapping;
577 flush_dcache_mmap_lock(mapping);
578 vma_prio_tree_remove(vma, &mapping->i_mmap);
579 flush_dcache_mmap_unlock(mapping);
582 /* remove from the master list */
583 rb_erase(&vma->vm_rb, &nommu_vma_tree);
587 * determine whether a mapping should be permitted and, if so, what sort of
588 * mapping we're capable of supporting
590 static int validate_mmap_request(struct file *file,
596 unsigned long *_capabilities)
598 unsigned long capabilities;
599 unsigned long reqprot = prot;
602 /* do the simple checks first */
603 if (flags & MAP_FIXED || addr) {
605 "%d: Can't do fixed-address/overlay mmap of RAM\n",
610 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
611 (flags & MAP_TYPE) != MAP_SHARED)
617 /* Careful about overflows.. */
618 len = PAGE_ALIGN(len);
619 if (!len || len > TASK_SIZE)
622 /* offset overflow? */
623 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
627 /* validate file mapping requests */
628 struct address_space *mapping;
630 /* files must support mmap */
631 if (!file->f_op || !file->f_op->mmap)
634 /* work out if what we've got could possibly be shared
635 * - we support chardevs that provide their own "memory"
636 * - we support files/blockdevs that are memory backed
638 mapping = file->f_mapping;
640 mapping = file->f_path.dentry->d_inode->i_mapping;
643 if (mapping && mapping->backing_dev_info)
644 capabilities = mapping->backing_dev_info->capabilities;
647 /* no explicit capabilities set, so assume some
649 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
652 capabilities = BDI_CAP_MAP_COPY;
667 /* eliminate any capabilities that we can't support on this
669 if (!file->f_op->get_unmapped_area)
670 capabilities &= ~BDI_CAP_MAP_DIRECT;
671 if (!file->f_op->read)
672 capabilities &= ~BDI_CAP_MAP_COPY;
674 if (flags & MAP_SHARED) {
675 /* do checks for writing, appending and locking */
676 if ((prot & PROT_WRITE) &&
677 !(file->f_mode & FMODE_WRITE))
680 if (IS_APPEND(file->f_path.dentry->d_inode) &&
681 (file->f_mode & FMODE_WRITE))
684 if (locks_verify_locked(file->f_path.dentry->d_inode))
687 if (!(capabilities & BDI_CAP_MAP_DIRECT))
690 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
691 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
692 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
694 printk("MAP_SHARED not completely supported on !MMU\n");
698 /* we mustn't privatise shared mappings */
699 capabilities &= ~BDI_CAP_MAP_COPY;
702 /* we're going to read the file into private memory we
704 if (!(capabilities & BDI_CAP_MAP_COPY))
707 /* we don't permit a private writable mapping to be
708 * shared with the backing device */
709 if (prot & PROT_WRITE)
710 capabilities &= ~BDI_CAP_MAP_DIRECT;
713 /* handle executable mappings and implied executable
715 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
716 if (prot & PROT_EXEC)
719 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
720 /* handle implication of PROT_EXEC by PROT_READ */
721 if (current->personality & READ_IMPLIES_EXEC) {
722 if (capabilities & BDI_CAP_EXEC_MAP)
726 else if ((prot & PROT_READ) &&
727 (prot & PROT_EXEC) &&
728 !(capabilities & BDI_CAP_EXEC_MAP)
730 /* backing file is not executable, try to copy */
731 capabilities &= ~BDI_CAP_MAP_DIRECT;
735 /* anonymous mappings are always memory backed and can be
738 capabilities = BDI_CAP_MAP_COPY;
740 /* handle PROT_EXEC implication by PROT_READ */
741 if ((prot & PROT_READ) &&
742 (current->personality & READ_IMPLIES_EXEC))
746 /* allow the security API to have its say */
747 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
752 *_capabilities = capabilities;
757 * we've determined that we can make the mapping, now translate what we
758 * now know into VMA flags
760 static unsigned long determine_vm_flags(struct file *file,
763 unsigned long capabilities)
765 unsigned long vm_flags;
767 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
768 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
769 /* vm_flags |= mm->def_flags; */
771 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
772 /* attempt to share read-only copies of mapped file chunks */
773 if (file && !(prot & PROT_WRITE))
774 vm_flags |= VM_MAYSHARE;
777 /* overlay a shareable mapping on the backing device or inode
778 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
780 if (flags & MAP_SHARED)
781 vm_flags |= VM_MAYSHARE | VM_SHARED;
782 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
783 vm_flags |= VM_MAYSHARE;
786 /* refuse to let anyone share private mappings with this process if
787 * it's being traced - otherwise breakpoints set in it may interfere
788 * with another untraced process
790 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
791 vm_flags &= ~VM_MAYSHARE;
797 * set up a shared mapping on a file
799 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
803 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
807 /* getting an ENOSYS error indicates that direct mmap isn't
808 * possible (as opposed to tried but failed) so we'll fall
809 * through to making a private copy of the data and mapping
815 * set up a private mapping or an anonymous shared mapping
817 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
822 /* invoke the file's mapping function so that it can keep track of
823 * shared mappings on devices or memory
824 * - VM_MAYSHARE will be set if it may attempt to share
827 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
828 if (ret != -ENOSYS) {
829 /* shouldn't return success if we're not sharing */
830 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
831 return ret; /* success or a real error */
834 /* getting an ENOSYS error indicates that direct mmap isn't
835 * possible (as opposed to tried but failed) so we'll try to
836 * make a private copy of the data and map that instead */
839 /* allocate some memory to hold the mapping
840 * - note that this may not return a page-aligned address if the object
841 * we're allocating is smaller than a page
843 base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
847 vma->vm_start = (unsigned long) base;
848 vma->vm_end = vma->vm_start + len;
849 vma->vm_flags |= VM_MAPPED_COPY;
852 if (len + WARN_ON_SLACK <= kobjsize(result))
853 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
854 len, current->pid, kobjsize(result) - len);
858 /* read the contents of a file into the copy */
862 fpos = vma->vm_pgoff;
867 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
873 /* clear the last little bit */
875 memset(base + ret, 0, len - ret);
878 /* if it's an anonymous mapping, then just clear it */
879 memset(base, 0, len);
890 printk("Allocation of length %lu from process %d failed\n",
897 * handle mapping creation for uClinux
899 unsigned long do_mmap_pgoff(struct file *file,
906 struct vm_list_struct *vml = NULL;
907 struct vm_area_struct *vma = NULL;
909 unsigned long capabilities, vm_flags;
913 if (!(flags & MAP_FIXED))
914 addr = round_hint_to_min(addr);
916 /* decide whether we should attempt the mapping, and if so what sort of
918 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
923 /* we've determined that we can make the mapping, now translate what we
924 * now know into VMA flags */
925 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
927 /* we're going to need to record the mapping if it works */
928 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
930 goto error_getting_vml;
932 down_write(&nommu_vma_sem);
934 /* if we want to share, we need to check for VMAs created by other
935 * mmap() calls that overlap with our proposed mapping
936 * - we can only share with an exact match on most regular files
937 * - shared mappings on character devices and memory backed files are
938 * permitted to overlap inexactly as far as we are concerned for in
939 * these cases, sharing is handled in the driver or filesystem rather
942 if (vm_flags & VM_MAYSHARE) {
943 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
944 unsigned long vmpglen;
946 /* suppress VMA sharing for shared regions */
947 if (vm_flags & VM_SHARED &&
948 capabilities & BDI_CAP_MAP_DIRECT)
949 goto dont_share_VMAs;
951 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
952 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
954 if (!(vma->vm_flags & VM_MAYSHARE))
957 /* search for overlapping mappings on the same file */
958 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
961 if (vma->vm_pgoff >= pgoff + pglen)
964 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
965 vmpglen >>= PAGE_SHIFT;
966 if (pgoff >= vma->vm_pgoff + vmpglen)
969 /* handle inexactly overlapping matches between mappings */
970 if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
971 if (!(capabilities & BDI_CAP_MAP_DIRECT))
972 goto sharing_violation;
976 /* we've found a VMA we can share */
977 atomic_inc(&vma->vm_usage);
980 result = (void *) vma->vm_start;
987 /* obtain the address at which to make a shared mapping
988 * - this is the hook for quasi-memory character devices to
989 * tell us the location of a shared mapping
991 if (file && file->f_op->get_unmapped_area) {
992 addr = file->f_op->get_unmapped_area(file, addr, len,
994 if (IS_ERR((void *) addr)) {
996 if (ret != (unsigned long) -ENOSYS)
999 /* the driver refused to tell us where to site
1000 * the mapping so we'll have to attempt to copy
1002 ret = (unsigned long) -ENODEV;
1003 if (!(capabilities & BDI_CAP_MAP_COPY))
1006 capabilities &= ~BDI_CAP_MAP_DIRECT;
1011 /* we're going to need a VMA struct as well */
1012 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
1014 goto error_getting_vma;
1016 INIT_LIST_HEAD(&vma->anon_vma_node);
1017 atomic_set(&vma->vm_usage, 1);
1020 if (vm_flags & VM_EXECUTABLE) {
1021 added_exe_file_vma(current->mm);
1022 vma->vm_mm = current->mm;
1025 vma->vm_file = file;
1026 vma->vm_flags = vm_flags;
1027 vma->vm_start = addr;
1028 vma->vm_end = addr + len;
1029 vma->vm_pgoff = pgoff;
1033 /* set up the mapping */
1034 if (file && vma->vm_flags & VM_SHARED)
1035 ret = do_mmap_shared_file(vma, len);
1037 ret = do_mmap_private(vma, len);
1041 /* okay... we have a mapping; now we have to register it */
1042 result = (void *) vma->vm_start;
1044 current->mm->total_vm += len >> PAGE_SHIFT;
1049 add_vma_to_mm(current->mm, vml);
1051 up_write(&nommu_vma_sem);
1053 if (prot & PROT_EXEC)
1054 flush_icache_range((unsigned long) result,
1055 (unsigned long) result + len);
1058 printk("do_mmap:\n");
1059 show_process_blocks();
1062 return (unsigned long) result;
1065 up_write(&nommu_vma_sem);
1070 if (vma->vm_flags & VM_EXECUTABLE)
1071 removed_exe_file_vma(vma->vm_mm);
1078 up_write(&nommu_vma_sem);
1079 printk("Attempt to share mismatched mappings\n");
1084 up_write(&nommu_vma_sem);
1086 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1092 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1097 EXPORT_SYMBOL(do_mmap_pgoff);
1100 * handle mapping disposal for uClinux
1102 static void put_vma(struct mm_struct *mm, struct vm_area_struct *vma)
1105 down_write(&nommu_vma_sem);
1107 if (atomic_dec_and_test(&vma->vm_usage)) {
1108 delete_nommu_vma(vma);
1110 if (vma->vm_ops && vma->vm_ops->close)
1111 vma->vm_ops->close(vma);
1113 /* IO memory and memory shared directly out of the pagecache from
1114 * ramfs/tmpfs mustn't be released here */
1115 if (vma->vm_flags & VM_MAPPED_COPY)
1116 kfree((void *) vma->vm_start);
1120 if (vma->vm_flags & VM_EXECUTABLE)
1121 removed_exe_file_vma(mm);
1126 up_write(&nommu_vma_sem);
1132 * - under NOMMU conditions the parameters must match exactly to the mapping to
1135 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1137 struct vm_list_struct *vml, **parent;
1138 unsigned long end = addr + len;
1141 printk("do_munmap:\n");
1144 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1145 if ((*parent)->vma->vm_start > addr)
1147 if ((*parent)->vma->vm_start == addr &&
1148 ((len == 0) || ((*parent)->vma->vm_end == end)))
1152 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1153 current->pid, current->comm, (void *) addr);
1159 put_vma(mm, vml->vma);
1161 *parent = vml->next;
1164 update_hiwater_vm(mm);
1165 mm->total_vm -= len >> PAGE_SHIFT;
1168 show_process_blocks();
1173 EXPORT_SYMBOL(do_munmap);
1175 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1178 struct mm_struct *mm = current->mm;
1180 down_write(&mm->mmap_sem);
1181 ret = do_munmap(mm, addr, len);
1182 up_write(&mm->mmap_sem);
1187 * Release all mappings
1189 void exit_mmap(struct mm_struct * mm)
1191 struct vm_list_struct *tmp;
1195 printk("Exit_mmap:\n");
1200 while ((tmp = mm->context.vmlist)) {
1201 mm->context.vmlist = tmp->next;
1202 put_vma(mm, tmp->vma);
1207 show_process_blocks();
1212 unsigned long do_brk(unsigned long addr, unsigned long len)
1218 * expand (or shrink) an existing mapping, potentially moving it at the same
1219 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1221 * under NOMMU conditions, we only permit changing a mapping's size, and only
1222 * as long as it stays within the hole allocated by the kmalloc() call in
1223 * do_mmap_pgoff() and the block is not shareable
1225 * MREMAP_FIXED is not supported under NOMMU conditions
1227 unsigned long do_mremap(unsigned long addr,
1228 unsigned long old_len, unsigned long new_len,
1229 unsigned long flags, unsigned long new_addr)
1231 struct vm_area_struct *vma;
1233 /* insanity checks first */
1235 return (unsigned long) -EINVAL;
1237 if (flags & MREMAP_FIXED && new_addr != addr)
1238 return (unsigned long) -EINVAL;
1240 vma = find_vma_exact(current->mm, addr);
1242 return (unsigned long) -EINVAL;
1244 if (vma->vm_end != vma->vm_start + old_len)
1245 return (unsigned long) -EFAULT;
1247 if (vma->vm_flags & VM_MAYSHARE)
1248 return (unsigned long) -EPERM;
1250 if (new_len > kobjsize((void *) addr))
1251 return (unsigned long) -ENOMEM;
1253 /* all checks complete - do it */
1254 vma->vm_end = vma->vm_start + new_len;
1256 return vma->vm_start;
1258 EXPORT_SYMBOL(do_mremap);
1260 asmlinkage unsigned long sys_mremap(unsigned long addr,
1261 unsigned long old_len, unsigned long new_len,
1262 unsigned long flags, unsigned long new_addr)
1266 down_write(¤t->mm->mmap_sem);
1267 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1268 up_write(¤t->mm->mmap_sem);
1272 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1273 unsigned int foll_flags)
1278 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1279 unsigned long to, unsigned long size, pgprot_t prot)
1281 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1284 EXPORT_SYMBOL(remap_pfn_range);
1286 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1287 unsigned long pgoff)
1289 unsigned int size = vma->vm_end - vma->vm_start;
1291 if (!(vma->vm_flags & VM_USERMAP))
1294 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1295 vma->vm_end = vma->vm_start + size;
1299 EXPORT_SYMBOL(remap_vmalloc_range);
1301 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1305 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1306 unsigned long len, unsigned long pgoff, unsigned long flags)
1311 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1315 void unmap_mapping_range(struct address_space *mapping,
1316 loff_t const holebegin, loff_t const holelen,
1320 EXPORT_SYMBOL(unmap_mapping_range);
1323 * ask for an unmapped area at which to create a mapping on a file
1325 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1326 unsigned long len, unsigned long pgoff,
1327 unsigned long flags)
1329 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1330 unsigned long, unsigned long);
1332 get_area = current->mm->get_unmapped_area;
1333 if (file && file->f_op && file->f_op->get_unmapped_area)
1334 get_area = file->f_op->get_unmapped_area;
1339 return get_area(file, addr, len, pgoff, flags);
1341 EXPORT_SYMBOL(get_unmapped_area);
1344 * Check that a process has enough memory to allocate a new virtual
1345 * mapping. 0 means there is enough memory for the allocation to
1346 * succeed and -ENOMEM implies there is not.
1348 * We currently support three overcommit policies, which are set via the
1349 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1351 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1352 * Additional code 2002 Jul 20 by Robert Love.
1354 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1356 * Note this is a helper function intended to be used by LSMs which
1357 * wish to use this logic.
1359 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1361 unsigned long free, allowed;
1363 vm_acct_memory(pages);
1366 * Sometimes we want to use more memory than we have
1368 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1371 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1374 free = global_page_state(NR_FILE_PAGES);
1375 free += nr_swap_pages;
1378 * Any slabs which are created with the
1379 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1380 * which are reclaimable, under pressure. The dentry
1381 * cache and most inode caches should fall into this
1383 free += global_page_state(NR_SLAB_RECLAIMABLE);
1386 * Leave the last 3% for root
1395 * nr_free_pages() is very expensive on large systems,
1396 * only call if we're about to fail.
1398 n = nr_free_pages();
1401 * Leave reserved pages. The pages are not for anonymous pages.
1403 if (n <= totalreserve_pages)
1406 n -= totalreserve_pages;
1409 * Leave the last 3% for root
1421 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1423 * Leave the last 3% for root
1426 allowed -= allowed / 32;
1427 allowed += total_swap_pages;
1429 /* Don't let a single process grow too big:
1430 leave 3% of the size of this process for other processes */
1432 allowed -= mm->total_vm / 32;
1435 * cast `allowed' as a signed long because vm_committed_space
1436 * sometimes has a negative value
1438 if (atomic_long_read(&vm_committed_space) < (long)allowed)
1441 vm_unacct_memory(pages);
1446 int in_gate_area_no_task(unsigned long addr)
1451 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1456 EXPORT_SYMBOL(filemap_fault);
1459 * Access another process' address space.
1460 * - source/target buffer must be kernel space
1462 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1464 struct vm_area_struct *vma;
1465 struct mm_struct *mm;
1467 if (addr + len < addr)
1470 mm = get_task_mm(tsk);
1474 down_read(&mm->mmap_sem);
1476 /* the access must start within one of the target process's mappings */
1477 vma = find_vma(mm, addr);
1479 /* don't overrun this mapping */
1480 if (addr + len >= vma->vm_end)
1481 len = vma->vm_end - addr;
1483 /* only read or write mappings where it is permitted */
1484 if (write && vma->vm_flags & VM_MAYWRITE)
1485 len -= copy_to_user((void *) addr, buf, len);
1486 else if (!write && vma->vm_flags & VM_MAYREAD)
1487 len -= copy_from_user(buf, (void *) addr, len);
1494 up_read(&mm->mmap_sem);