Add lock_page_killable
[linux-2.6/mini2440.git] / mm / nommu.c
blobb989cb928a7cf29ec5ad66ff519f80633bfe8129
1 /*
2 * linux/mm/nommu.c
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>
15 #include <linux/module.h>
16 #include <linux/mm.h>
17 #include <linux/mman.h>
18 #include <linux/swap.h>
19 #include <linux/file.h>
20 #include <linux/highmem.h>
21 #include <linux/pagemap.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ptrace.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mount.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
32 #include <asm/uaccess.h>
33 #include <asm/tlb.h>
34 #include <asm/tlbflush.h>
36 void *high_memory;
37 struct page *mem_map;
38 unsigned long max_mapnr;
39 unsigned long num_physpages;
40 unsigned long askedalloc, realalloc;
41 atomic_t vm_committed_space = ATOMIC_INIT(0);
42 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
43 int sysctl_overcommit_ratio = 50; /* default is 50% */
44 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
45 int heap_stack_gap = 0;
47 EXPORT_SYMBOL(mem_map);
48 EXPORT_SYMBOL(num_physpages);
50 /* list of shareable VMAs */
51 struct rb_root nommu_vma_tree = RB_ROOT;
52 DECLARE_RWSEM(nommu_vma_sem);
54 struct vm_operations_struct generic_file_vm_ops = {
58 * Handle all mappings that got truncated by a "truncate()"
59 * system call.
61 * NOTE! We have to be ready to update the memory sharing
62 * between the file and the memory map for a potential last
63 * incomplete page. Ugly, but necessary.
65 int vmtruncate(struct inode *inode, loff_t offset)
67 struct address_space *mapping = inode->i_mapping;
68 unsigned long limit;
70 if (inode->i_size < offset)
71 goto do_expand;
72 i_size_write(inode, offset);
74 truncate_inode_pages(mapping, offset);
75 goto out_truncate;
77 do_expand:
78 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
79 if (limit != RLIM_INFINITY && offset > limit)
80 goto out_sig;
81 if (offset > inode->i_sb->s_maxbytes)
82 goto out;
83 i_size_write(inode, offset);
85 out_truncate:
86 if (inode->i_op && inode->i_op->truncate)
87 inode->i_op->truncate(inode);
88 return 0;
89 out_sig:
90 send_sig(SIGXFSZ, current, 0);
91 out:
92 return -EFBIG;
95 EXPORT_SYMBOL(vmtruncate);
98 * Return the total memory allocated for this pointer, not
99 * just what the caller asked for.
101 * Doesn't have to be accurate, i.e. may have races.
103 unsigned int kobjsize(const void *objp)
105 struct page *page;
107 if (!objp || !((page = virt_to_page(objp))))
108 return 0;
110 if (PageSlab(page))
111 return ksize(objp);
113 BUG_ON(page->index < 0);
114 BUG_ON(page->index >= MAX_ORDER);
116 return (PAGE_SIZE << page->index);
120 * get a list of pages in an address range belonging to the specified process
121 * and indicate the VMA that covers each page
122 * - this is potentially dodgy as we may end incrementing the page count of a
123 * slab page or a secondary page from a compound page
124 * - don't permit access to VMAs that don't support it, such as I/O mappings
126 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
127 unsigned long start, int len, int write, int force,
128 struct page **pages, struct vm_area_struct **vmas)
130 struct vm_area_struct *vma;
131 unsigned long vm_flags;
132 int i;
134 /* calculate required read or write permissions.
135 * - if 'force' is set, we only require the "MAY" flags.
137 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
138 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
140 for (i = 0; i < len; i++) {
141 vma = find_vma(mm, start);
142 if (!vma)
143 goto finish_or_fault;
145 /* protect what we can, including chardevs */
146 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
147 !(vm_flags & vma->vm_flags))
148 goto finish_or_fault;
150 if (pages) {
151 pages[i] = virt_to_page(start);
152 if (pages[i])
153 page_cache_get(pages[i]);
155 if (vmas)
156 vmas[i] = vma;
157 start += PAGE_SIZE;
160 return i;
162 finish_or_fault:
163 return i ? : -EFAULT;
165 EXPORT_SYMBOL(get_user_pages);
167 DEFINE_RWLOCK(vmlist_lock);
168 struct vm_struct *vmlist;
170 void vfree(void *addr)
172 kfree(addr);
174 EXPORT_SYMBOL(vfree);
176 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
179 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
180 * returns only a logical address.
182 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
184 EXPORT_SYMBOL(__vmalloc);
186 struct page * vmalloc_to_page(void *addr)
188 return virt_to_page(addr);
190 EXPORT_SYMBOL(vmalloc_to_page);
192 unsigned long vmalloc_to_pfn(void *addr)
194 return page_to_pfn(virt_to_page(addr));
196 EXPORT_SYMBOL(vmalloc_to_pfn);
198 long vread(char *buf, char *addr, unsigned long count)
200 memcpy(buf, addr, count);
201 return count;
204 long vwrite(char *buf, char *addr, unsigned long count)
206 /* Don't allow overflow */
207 if ((unsigned long) addr + count < count)
208 count = -(unsigned long) addr;
210 memcpy(addr, buf, count);
211 return(count);
215 * vmalloc - allocate virtually continguos memory
217 * @size: allocation size
219 * Allocate enough pages to cover @size from the page level
220 * allocator and map them into continguos kernel virtual space.
222 * For tight control over page level allocator and protection flags
223 * use __vmalloc() instead.
225 void *vmalloc(unsigned long size)
227 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
229 EXPORT_SYMBOL(vmalloc);
231 void *vmalloc_node(unsigned long size, int node)
233 return vmalloc(size);
235 EXPORT_SYMBOL(vmalloc_node);
238 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
239 * @size: allocation size
241 * Allocate enough 32bit PA addressable pages to cover @size from the
242 * page level allocator and map them into continguos kernel virtual space.
244 void *vmalloc_32(unsigned long size)
246 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
248 EXPORT_SYMBOL(vmalloc_32);
251 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
252 * @size: allocation size
254 * The resulting memory area is 32bit addressable and zeroed so it can be
255 * mapped to userspace without leaking data.
257 void *vmalloc_32_user(unsigned long size)
259 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
261 EXPORT_SYMBOL(vmalloc_32_user);
263 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
265 BUG();
266 return NULL;
268 EXPORT_SYMBOL(vmap);
270 void vunmap(void *addr)
272 BUG();
274 EXPORT_SYMBOL(vunmap);
277 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
278 * have one.
280 void __attribute__((weak)) vmalloc_sync_all(void)
284 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
285 struct page *page)
287 return -EINVAL;
289 EXPORT_SYMBOL(vm_insert_page);
292 * sys_brk() for the most part doesn't need the global kernel
293 * lock, except when an application is doing something nasty
294 * like trying to un-brk an area that has already been mapped
295 * to a regular file. in this case, the unmapping will need
296 * to invoke file system routines that need the global lock.
298 asmlinkage unsigned long sys_brk(unsigned long brk)
300 struct mm_struct *mm = current->mm;
302 if (brk < mm->start_brk || brk > mm->context.end_brk)
303 return mm->brk;
305 if (mm->brk == brk)
306 return mm->brk;
309 * Always allow shrinking brk
311 if (brk <= mm->brk) {
312 mm->brk = brk;
313 return brk;
317 * Ok, looks good - let it rip.
319 return mm->brk = brk;
322 #ifdef DEBUG
323 static void show_process_blocks(void)
325 struct vm_list_struct *vml;
327 printk("Process blocks %d:", current->pid);
329 for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
330 printk(" %p: %p", vml, vml->vma);
331 if (vml->vma)
332 printk(" (%d @%lx #%d)",
333 kobjsize((void *) vml->vma->vm_start),
334 vml->vma->vm_start,
335 atomic_read(&vml->vma->vm_usage));
336 printk(vml->next ? " ->" : ".\n");
339 #endif /* DEBUG */
342 * add a VMA into a process's mm_struct in the appropriate place in the list
343 * - should be called with mm->mmap_sem held writelocked
345 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
347 struct vm_list_struct **ppv;
349 for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
350 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
351 break;
353 vml->next = *ppv;
354 *ppv = vml;
358 * look up the first VMA in which addr resides, NULL if none
359 * - should be called with mm->mmap_sem at least held readlocked
361 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
363 struct vm_list_struct *loop, *vml;
365 /* search the vm_start ordered list */
366 vml = NULL;
367 for (loop = mm->context.vmlist; loop; loop = loop->next) {
368 if (loop->vma->vm_start > addr)
369 break;
370 vml = loop;
373 if (vml && vml->vma->vm_end > addr)
374 return vml->vma;
376 return NULL;
378 EXPORT_SYMBOL(find_vma);
381 * find a VMA
382 * - we don't extend stack VMAs under NOMMU conditions
384 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
386 return find_vma(mm, addr);
389 int expand_stack(struct vm_area_struct *vma, unsigned long address)
391 return -ENOMEM;
395 * look up the first VMA exactly that exactly matches addr
396 * - should be called with mm->mmap_sem at least held readlocked
398 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
399 unsigned long addr)
401 struct vm_list_struct *vml;
403 /* search the vm_start ordered list */
404 for (vml = mm->context.vmlist; vml; vml = vml->next) {
405 if (vml->vma->vm_start == addr)
406 return vml->vma;
407 if (vml->vma->vm_start > addr)
408 break;
411 return NULL;
415 * find a VMA in the global tree
417 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
419 struct vm_area_struct *vma;
420 struct rb_node *n = nommu_vma_tree.rb_node;
422 while (n) {
423 vma = rb_entry(n, struct vm_area_struct, vm_rb);
425 if (start < vma->vm_start)
426 n = n->rb_left;
427 else if (start > vma->vm_start)
428 n = n->rb_right;
429 else
430 return vma;
433 return NULL;
437 * add a VMA in the global tree
439 static void add_nommu_vma(struct vm_area_struct *vma)
441 struct vm_area_struct *pvma;
442 struct address_space *mapping;
443 struct rb_node **p = &nommu_vma_tree.rb_node;
444 struct rb_node *parent = NULL;
446 /* add the VMA to the mapping */
447 if (vma->vm_file) {
448 mapping = vma->vm_file->f_mapping;
450 flush_dcache_mmap_lock(mapping);
451 vma_prio_tree_insert(vma, &mapping->i_mmap);
452 flush_dcache_mmap_unlock(mapping);
455 /* add the VMA to the master list */
456 while (*p) {
457 parent = *p;
458 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
460 if (vma->vm_start < pvma->vm_start) {
461 p = &(*p)->rb_left;
463 else if (vma->vm_start > pvma->vm_start) {
464 p = &(*p)->rb_right;
466 else {
467 /* mappings are at the same address - this can only
468 * happen for shared-mem chardevs and shared file
469 * mappings backed by ramfs/tmpfs */
470 BUG_ON(!(pvma->vm_flags & VM_SHARED));
472 if (vma < pvma)
473 p = &(*p)->rb_left;
474 else if (vma > pvma)
475 p = &(*p)->rb_right;
476 else
477 BUG();
481 rb_link_node(&vma->vm_rb, parent, p);
482 rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
486 * delete a VMA from the global list
488 static void delete_nommu_vma(struct vm_area_struct *vma)
490 struct address_space *mapping;
492 /* remove the VMA from the mapping */
493 if (vma->vm_file) {
494 mapping = vma->vm_file->f_mapping;
496 flush_dcache_mmap_lock(mapping);
497 vma_prio_tree_remove(vma, &mapping->i_mmap);
498 flush_dcache_mmap_unlock(mapping);
501 /* remove from the master list */
502 rb_erase(&vma->vm_rb, &nommu_vma_tree);
506 * determine whether a mapping should be permitted and, if so, what sort of
507 * mapping we're capable of supporting
509 static int validate_mmap_request(struct file *file,
510 unsigned long addr,
511 unsigned long len,
512 unsigned long prot,
513 unsigned long flags,
514 unsigned long pgoff,
515 unsigned long *_capabilities)
517 unsigned long capabilities;
518 unsigned long reqprot = prot;
519 int ret;
521 /* do the simple checks first */
522 if (flags & MAP_FIXED || addr) {
523 printk(KERN_DEBUG
524 "%d: Can't do fixed-address/overlay mmap of RAM\n",
525 current->pid);
526 return -EINVAL;
529 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
530 (flags & MAP_TYPE) != MAP_SHARED)
531 return -EINVAL;
533 if (!len)
534 return -EINVAL;
536 /* Careful about overflows.. */
537 len = PAGE_ALIGN(len);
538 if (!len || len > TASK_SIZE)
539 return -ENOMEM;
541 /* offset overflow? */
542 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
543 return -EOVERFLOW;
545 if (file) {
546 /* validate file mapping requests */
547 struct address_space *mapping;
549 /* files must support mmap */
550 if (!file->f_op || !file->f_op->mmap)
551 return -ENODEV;
553 /* work out if what we've got could possibly be shared
554 * - we support chardevs that provide their own "memory"
555 * - we support files/blockdevs that are memory backed
557 mapping = file->f_mapping;
558 if (!mapping)
559 mapping = file->f_path.dentry->d_inode->i_mapping;
561 capabilities = 0;
562 if (mapping && mapping->backing_dev_info)
563 capabilities = mapping->backing_dev_info->capabilities;
565 if (!capabilities) {
566 /* no explicit capabilities set, so assume some
567 * defaults */
568 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
569 case S_IFREG:
570 case S_IFBLK:
571 capabilities = BDI_CAP_MAP_COPY;
572 break;
574 case S_IFCHR:
575 capabilities =
576 BDI_CAP_MAP_DIRECT |
577 BDI_CAP_READ_MAP |
578 BDI_CAP_WRITE_MAP;
579 break;
581 default:
582 return -EINVAL;
586 /* eliminate any capabilities that we can't support on this
587 * device */
588 if (!file->f_op->get_unmapped_area)
589 capabilities &= ~BDI_CAP_MAP_DIRECT;
590 if (!file->f_op->read)
591 capabilities &= ~BDI_CAP_MAP_COPY;
593 if (flags & MAP_SHARED) {
594 /* do checks for writing, appending and locking */
595 if ((prot & PROT_WRITE) &&
596 !(file->f_mode & FMODE_WRITE))
597 return -EACCES;
599 if (IS_APPEND(file->f_path.dentry->d_inode) &&
600 (file->f_mode & FMODE_WRITE))
601 return -EACCES;
603 if (locks_verify_locked(file->f_path.dentry->d_inode))
604 return -EAGAIN;
606 if (!(capabilities & BDI_CAP_MAP_DIRECT))
607 return -ENODEV;
609 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
610 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
611 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
613 printk("MAP_SHARED not completely supported on !MMU\n");
614 return -EINVAL;
617 /* we mustn't privatise shared mappings */
618 capabilities &= ~BDI_CAP_MAP_COPY;
620 else {
621 /* we're going to read the file into private memory we
622 * allocate */
623 if (!(capabilities & BDI_CAP_MAP_COPY))
624 return -ENODEV;
626 /* we don't permit a private writable mapping to be
627 * shared with the backing device */
628 if (prot & PROT_WRITE)
629 capabilities &= ~BDI_CAP_MAP_DIRECT;
632 /* handle executable mappings and implied executable
633 * mappings */
634 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
635 if (prot & PROT_EXEC)
636 return -EPERM;
638 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
639 /* handle implication of PROT_EXEC by PROT_READ */
640 if (current->personality & READ_IMPLIES_EXEC) {
641 if (capabilities & BDI_CAP_EXEC_MAP)
642 prot |= PROT_EXEC;
645 else if ((prot & PROT_READ) &&
646 (prot & PROT_EXEC) &&
647 !(capabilities & BDI_CAP_EXEC_MAP)
649 /* backing file is not executable, try to copy */
650 capabilities &= ~BDI_CAP_MAP_DIRECT;
653 else {
654 /* anonymous mappings are always memory backed and can be
655 * privately mapped
657 capabilities = BDI_CAP_MAP_COPY;
659 /* handle PROT_EXEC implication by PROT_READ */
660 if ((prot & PROT_READ) &&
661 (current->personality & READ_IMPLIES_EXEC))
662 prot |= PROT_EXEC;
665 /* allow the security API to have its say */
666 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
667 if (ret < 0)
668 return ret;
670 /* looks okay */
671 *_capabilities = capabilities;
672 return 0;
676 * we've determined that we can make the mapping, now translate what we
677 * now know into VMA flags
679 static unsigned long determine_vm_flags(struct file *file,
680 unsigned long prot,
681 unsigned long flags,
682 unsigned long capabilities)
684 unsigned long vm_flags;
686 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
687 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
688 /* vm_flags |= mm->def_flags; */
690 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
691 /* attempt to share read-only copies of mapped file chunks */
692 if (file && !(prot & PROT_WRITE))
693 vm_flags |= VM_MAYSHARE;
695 else {
696 /* overlay a shareable mapping on the backing device or inode
697 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
698 * romfs/cramfs */
699 if (flags & MAP_SHARED)
700 vm_flags |= VM_MAYSHARE | VM_SHARED;
701 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
702 vm_flags |= VM_MAYSHARE;
705 /* refuse to let anyone share private mappings with this process if
706 * it's being traced - otherwise breakpoints set in it may interfere
707 * with another untraced process
709 if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
710 vm_flags &= ~VM_MAYSHARE;
712 return vm_flags;
716 * set up a shared mapping on a file
718 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
720 int ret;
722 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
723 if (ret != -ENOSYS)
724 return ret;
726 /* getting an ENOSYS error indicates that direct mmap isn't
727 * possible (as opposed to tried but failed) so we'll fall
728 * through to making a private copy of the data and mapping
729 * that if we can */
730 return -ENODEV;
734 * set up a private mapping or an anonymous shared mapping
736 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
738 void *base;
739 int ret;
741 /* invoke the file's mapping function so that it can keep track of
742 * shared mappings on devices or memory
743 * - VM_MAYSHARE will be set if it may attempt to share
745 if (vma->vm_file) {
746 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
747 if (ret != -ENOSYS) {
748 /* shouldn't return success if we're not sharing */
749 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
750 return ret; /* success or a real error */
753 /* getting an ENOSYS error indicates that direct mmap isn't
754 * possible (as opposed to tried but failed) so we'll try to
755 * make a private copy of the data and map that instead */
758 /* allocate some memory to hold the mapping
759 * - note that this may not return a page-aligned address if the object
760 * we're allocating is smaller than a page
762 base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
763 if (!base)
764 goto enomem;
766 vma->vm_start = (unsigned long) base;
767 vma->vm_end = vma->vm_start + len;
768 vma->vm_flags |= VM_MAPPED_COPY;
770 #ifdef WARN_ON_SLACK
771 if (len + WARN_ON_SLACK <= kobjsize(result))
772 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
773 len, current->pid, kobjsize(result) - len);
774 #endif
776 if (vma->vm_file) {
777 /* read the contents of a file into the copy */
778 mm_segment_t old_fs;
779 loff_t fpos;
781 fpos = vma->vm_pgoff;
782 fpos <<= PAGE_SHIFT;
784 old_fs = get_fs();
785 set_fs(KERNEL_DS);
786 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
787 set_fs(old_fs);
789 if (ret < 0)
790 goto error_free;
792 /* clear the last little bit */
793 if (ret < len)
794 memset(base + ret, 0, len - ret);
796 } else {
797 /* if it's an anonymous mapping, then just clear it */
798 memset(base, 0, len);
801 return 0;
803 error_free:
804 kfree(base);
805 vma->vm_start = 0;
806 return ret;
808 enomem:
809 printk("Allocation of length %lu from process %d failed\n",
810 len, current->pid);
811 show_free_areas();
812 return -ENOMEM;
816 * handle mapping creation for uClinux
818 unsigned long do_mmap_pgoff(struct file *file,
819 unsigned long addr,
820 unsigned long len,
821 unsigned long prot,
822 unsigned long flags,
823 unsigned long pgoff)
825 struct vm_list_struct *vml = NULL;
826 struct vm_area_struct *vma = NULL;
827 struct rb_node *rb;
828 unsigned long capabilities, vm_flags;
829 void *result;
830 int ret;
832 if (!(flags & MAP_FIXED))
833 addr = round_hint_to_min(addr);
835 /* decide whether we should attempt the mapping, and if so what sort of
836 * mapping */
837 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
838 &capabilities);
839 if (ret < 0)
840 return ret;
842 /* we've determined that we can make the mapping, now translate what we
843 * now know into VMA flags */
844 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
846 /* we're going to need to record the mapping if it works */
847 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
848 if (!vml)
849 goto error_getting_vml;
851 down_write(&nommu_vma_sem);
853 /* if we want to share, we need to check for VMAs created by other
854 * mmap() calls that overlap with our proposed mapping
855 * - we can only share with an exact match on most regular files
856 * - shared mappings on character devices and memory backed files are
857 * permitted to overlap inexactly as far as we are concerned for in
858 * these cases, sharing is handled in the driver or filesystem rather
859 * than here
861 if (vm_flags & VM_MAYSHARE) {
862 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
863 unsigned long vmpglen;
865 /* suppress VMA sharing for shared regions */
866 if (vm_flags & VM_SHARED &&
867 capabilities & BDI_CAP_MAP_DIRECT)
868 goto dont_share_VMAs;
870 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
871 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
873 if (!(vma->vm_flags & VM_MAYSHARE))
874 continue;
876 /* search for overlapping mappings on the same file */
877 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
878 continue;
880 if (vma->vm_pgoff >= pgoff + pglen)
881 continue;
883 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
884 vmpglen >>= PAGE_SHIFT;
885 if (pgoff >= vma->vm_pgoff + vmpglen)
886 continue;
888 /* handle inexactly overlapping matches between mappings */
889 if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
890 if (!(capabilities & BDI_CAP_MAP_DIRECT))
891 goto sharing_violation;
892 continue;
895 /* we've found a VMA we can share */
896 atomic_inc(&vma->vm_usage);
898 vml->vma = vma;
899 result = (void *) vma->vm_start;
900 goto shared;
903 dont_share_VMAs:
904 vma = NULL;
906 /* obtain the address at which to make a shared mapping
907 * - this is the hook for quasi-memory character devices to
908 * tell us the location of a shared mapping
910 if (file && file->f_op->get_unmapped_area) {
911 addr = file->f_op->get_unmapped_area(file, addr, len,
912 pgoff, flags);
913 if (IS_ERR((void *) addr)) {
914 ret = addr;
915 if (ret != (unsigned long) -ENOSYS)
916 goto error;
918 /* the driver refused to tell us where to site
919 * the mapping so we'll have to attempt to copy
920 * it */
921 ret = (unsigned long) -ENODEV;
922 if (!(capabilities & BDI_CAP_MAP_COPY))
923 goto error;
925 capabilities &= ~BDI_CAP_MAP_DIRECT;
930 /* we're going to need a VMA struct as well */
931 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
932 if (!vma)
933 goto error_getting_vma;
935 INIT_LIST_HEAD(&vma->anon_vma_node);
936 atomic_set(&vma->vm_usage, 1);
937 if (file)
938 get_file(file);
939 vma->vm_file = file;
940 vma->vm_flags = vm_flags;
941 vma->vm_start = addr;
942 vma->vm_end = addr + len;
943 vma->vm_pgoff = pgoff;
945 vml->vma = vma;
947 /* set up the mapping */
948 if (file && vma->vm_flags & VM_SHARED)
949 ret = do_mmap_shared_file(vma, len);
950 else
951 ret = do_mmap_private(vma, len);
952 if (ret < 0)
953 goto error;
955 /* okay... we have a mapping; now we have to register it */
956 result = (void *) vma->vm_start;
958 if (vma->vm_flags & VM_MAPPED_COPY) {
959 realalloc += kobjsize(result);
960 askedalloc += len;
963 realalloc += kobjsize(vma);
964 askedalloc += sizeof(*vma);
966 current->mm->total_vm += len >> PAGE_SHIFT;
968 add_nommu_vma(vma);
970 shared:
971 realalloc += kobjsize(vml);
972 askedalloc += sizeof(*vml);
974 add_vma_to_mm(current->mm, vml);
976 up_write(&nommu_vma_sem);
978 if (prot & PROT_EXEC)
979 flush_icache_range((unsigned long) result,
980 (unsigned long) result + len);
982 #ifdef DEBUG
983 printk("do_mmap:\n");
984 show_process_blocks();
985 #endif
987 return (unsigned long) result;
989 error:
990 up_write(&nommu_vma_sem);
991 kfree(vml);
992 if (vma) {
993 if (vma->vm_file)
994 fput(vma->vm_file);
995 kfree(vma);
997 return ret;
999 sharing_violation:
1000 up_write(&nommu_vma_sem);
1001 printk("Attempt to share mismatched mappings\n");
1002 kfree(vml);
1003 return -EINVAL;
1005 error_getting_vma:
1006 up_write(&nommu_vma_sem);
1007 kfree(vml);
1008 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1009 len, current->pid);
1010 show_free_areas();
1011 return -ENOMEM;
1013 error_getting_vml:
1014 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1015 len, current->pid);
1016 show_free_areas();
1017 return -ENOMEM;
1019 EXPORT_SYMBOL(do_mmap_pgoff);
1022 * handle mapping disposal for uClinux
1024 static void put_vma(struct vm_area_struct *vma)
1026 if (vma) {
1027 down_write(&nommu_vma_sem);
1029 if (atomic_dec_and_test(&vma->vm_usage)) {
1030 delete_nommu_vma(vma);
1032 if (vma->vm_ops && vma->vm_ops->close)
1033 vma->vm_ops->close(vma);
1035 /* IO memory and memory shared directly out of the pagecache from
1036 * ramfs/tmpfs mustn't be released here */
1037 if (vma->vm_flags & VM_MAPPED_COPY) {
1038 realalloc -= kobjsize((void *) vma->vm_start);
1039 askedalloc -= vma->vm_end - vma->vm_start;
1040 kfree((void *) vma->vm_start);
1043 realalloc -= kobjsize(vma);
1044 askedalloc -= sizeof(*vma);
1046 if (vma->vm_file)
1047 fput(vma->vm_file);
1048 kfree(vma);
1051 up_write(&nommu_vma_sem);
1056 * release a mapping
1057 * - under NOMMU conditions the parameters must match exactly to the mapping to
1058 * be removed
1060 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1062 struct vm_list_struct *vml, **parent;
1063 unsigned long end = addr + len;
1065 #ifdef DEBUG
1066 printk("do_munmap:\n");
1067 #endif
1069 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1070 if ((*parent)->vma->vm_start > addr)
1071 break;
1072 if ((*parent)->vma->vm_start == addr &&
1073 ((len == 0) || ((*parent)->vma->vm_end == end)))
1074 goto found;
1077 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1078 current->pid, current->comm, (void *) addr);
1079 return -EINVAL;
1081 found:
1082 vml = *parent;
1084 put_vma(vml->vma);
1086 *parent = vml->next;
1087 realalloc -= kobjsize(vml);
1088 askedalloc -= sizeof(*vml);
1089 kfree(vml);
1091 update_hiwater_vm(mm);
1092 mm->total_vm -= len >> PAGE_SHIFT;
1094 #ifdef DEBUG
1095 show_process_blocks();
1096 #endif
1098 return 0;
1100 EXPORT_SYMBOL(do_munmap);
1102 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1104 int ret;
1105 struct mm_struct *mm = current->mm;
1107 down_write(&mm->mmap_sem);
1108 ret = do_munmap(mm, addr, len);
1109 up_write(&mm->mmap_sem);
1110 return ret;
1114 * Release all mappings
1116 void exit_mmap(struct mm_struct * mm)
1118 struct vm_list_struct *tmp;
1120 if (mm) {
1121 #ifdef DEBUG
1122 printk("Exit_mmap:\n");
1123 #endif
1125 mm->total_vm = 0;
1127 while ((tmp = mm->context.vmlist)) {
1128 mm->context.vmlist = tmp->next;
1129 put_vma(tmp->vma);
1131 realalloc -= kobjsize(tmp);
1132 askedalloc -= sizeof(*tmp);
1133 kfree(tmp);
1136 #ifdef DEBUG
1137 show_process_blocks();
1138 #endif
1142 unsigned long do_brk(unsigned long addr, unsigned long len)
1144 return -ENOMEM;
1148 * expand (or shrink) an existing mapping, potentially moving it at the same
1149 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1151 * under NOMMU conditions, we only permit changing a mapping's size, and only
1152 * as long as it stays within the hole allocated by the kmalloc() call in
1153 * do_mmap_pgoff() and the block is not shareable
1155 * MREMAP_FIXED is not supported under NOMMU conditions
1157 unsigned long do_mremap(unsigned long addr,
1158 unsigned long old_len, unsigned long new_len,
1159 unsigned long flags, unsigned long new_addr)
1161 struct vm_area_struct *vma;
1163 /* insanity checks first */
1164 if (new_len == 0)
1165 return (unsigned long) -EINVAL;
1167 if (flags & MREMAP_FIXED && new_addr != addr)
1168 return (unsigned long) -EINVAL;
1170 vma = find_vma_exact(current->mm, addr);
1171 if (!vma)
1172 return (unsigned long) -EINVAL;
1174 if (vma->vm_end != vma->vm_start + old_len)
1175 return (unsigned long) -EFAULT;
1177 if (vma->vm_flags & VM_MAYSHARE)
1178 return (unsigned long) -EPERM;
1180 if (new_len > kobjsize((void *) addr))
1181 return (unsigned long) -ENOMEM;
1183 /* all checks complete - do it */
1184 vma->vm_end = vma->vm_start + new_len;
1186 askedalloc -= old_len;
1187 askedalloc += new_len;
1189 return vma->vm_start;
1191 EXPORT_SYMBOL(do_mremap);
1193 asmlinkage unsigned long sys_mremap(unsigned long addr,
1194 unsigned long old_len, unsigned long new_len,
1195 unsigned long flags, unsigned long new_addr)
1197 unsigned long ret;
1199 down_write(&current->mm->mmap_sem);
1200 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1201 up_write(&current->mm->mmap_sem);
1202 return ret;
1205 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1206 unsigned int foll_flags)
1208 return NULL;
1211 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1212 unsigned long to, unsigned long size, pgprot_t prot)
1214 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1215 return 0;
1217 EXPORT_SYMBOL(remap_pfn_range);
1219 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1223 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1224 unsigned long len, unsigned long pgoff, unsigned long flags)
1226 return -ENOMEM;
1229 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1233 void unmap_mapping_range(struct address_space *mapping,
1234 loff_t const holebegin, loff_t const holelen,
1235 int even_cows)
1238 EXPORT_SYMBOL(unmap_mapping_range);
1241 * ask for an unmapped area at which to create a mapping on a file
1243 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1244 unsigned long len, unsigned long pgoff,
1245 unsigned long flags)
1247 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1248 unsigned long, unsigned long);
1250 get_area = current->mm->get_unmapped_area;
1251 if (file && file->f_op && file->f_op->get_unmapped_area)
1252 get_area = file->f_op->get_unmapped_area;
1254 if (!get_area)
1255 return -ENOSYS;
1257 return get_area(file, addr, len, pgoff, flags);
1259 EXPORT_SYMBOL(get_unmapped_area);
1262 * Check that a process has enough memory to allocate a new virtual
1263 * mapping. 0 means there is enough memory for the allocation to
1264 * succeed and -ENOMEM implies there is not.
1266 * We currently support three overcommit policies, which are set via the
1267 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1269 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1270 * Additional code 2002 Jul 20 by Robert Love.
1272 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1274 * Note this is a helper function intended to be used by LSMs which
1275 * wish to use this logic.
1277 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1279 unsigned long free, allowed;
1281 vm_acct_memory(pages);
1284 * Sometimes we want to use more memory than we have
1286 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1287 return 0;
1289 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1290 unsigned long n;
1292 free = global_page_state(NR_FILE_PAGES);
1293 free += nr_swap_pages;
1296 * Any slabs which are created with the
1297 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1298 * which are reclaimable, under pressure. The dentry
1299 * cache and most inode caches should fall into this
1301 free += global_page_state(NR_SLAB_RECLAIMABLE);
1304 * Leave the last 3% for root
1306 if (!cap_sys_admin)
1307 free -= free / 32;
1309 if (free > pages)
1310 return 0;
1313 * nr_free_pages() is very expensive on large systems,
1314 * only call if we're about to fail.
1316 n = nr_free_pages();
1319 * Leave reserved pages. The pages are not for anonymous pages.
1321 if (n <= totalreserve_pages)
1322 goto error;
1323 else
1324 n -= totalreserve_pages;
1327 * Leave the last 3% for root
1329 if (!cap_sys_admin)
1330 n -= n / 32;
1331 free += n;
1333 if (free > pages)
1334 return 0;
1336 goto error;
1339 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1341 * Leave the last 3% for root
1343 if (!cap_sys_admin)
1344 allowed -= allowed / 32;
1345 allowed += total_swap_pages;
1347 /* Don't let a single process grow too big:
1348 leave 3% of the size of this process for other processes */
1349 allowed -= current->mm->total_vm / 32;
1352 * cast `allowed' as a signed long because vm_committed_space
1353 * sometimes has a negative value
1355 if (atomic_read(&vm_committed_space) < (long)allowed)
1356 return 0;
1357 error:
1358 vm_unacct_memory(pages);
1360 return -ENOMEM;
1363 int in_gate_area_no_task(unsigned long addr)
1365 return 0;
1368 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1370 BUG();
1371 return 0;
1373 EXPORT_SYMBOL(filemap_fault);
1376 * Access another process' address space.
1377 * - source/target buffer must be kernel space
1379 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1381 struct vm_area_struct *vma;
1382 struct mm_struct *mm;
1384 if (addr + len < addr)
1385 return 0;
1387 mm = get_task_mm(tsk);
1388 if (!mm)
1389 return 0;
1391 down_read(&mm->mmap_sem);
1393 /* the access must start within one of the target process's mappings */
1394 vma = find_vma(mm, addr);
1395 if (vma) {
1396 /* don't overrun this mapping */
1397 if (addr + len >= vma->vm_end)
1398 len = vma->vm_end - addr;
1400 /* only read or write mappings where it is permitted */
1401 if (write && vma->vm_flags & VM_MAYWRITE)
1402 len -= copy_to_user((void *) addr, buf, len);
1403 else if (!write && vma->vm_flags & VM_MAYREAD)
1404 len -= copy_from_user(buf, (void *) addr, len);
1405 else
1406 len = 0;
1407 } else {
1408 len = 0;
1411 up_read(&mm->mmap_sem);
1412 mmput(mm);
1413 return len;