[PATCH] powerpc: Make the vDSO functions set error code (#2)
[linux-2.6/sactl.git] / mm / memory.c
blob2998cfc12f5bc2c2a60d5ad6a724e953e7765e1b
1 /*
2 * linux/mm/memory.c
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
7 /*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
42 #include <linux/mm.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
54 #include <asm/tlb.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr;
64 struct page *mem_map;
66 EXPORT_SYMBOL(max_mapnr);
67 EXPORT_SYMBOL(mem_map);
68 #endif
70 unsigned long num_physpages;
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
76 * and ZONE_HIGHMEM.
78 void * high_memory;
79 unsigned long vmalloc_earlyreserve;
81 EXPORT_SYMBOL(num_physpages);
82 EXPORT_SYMBOL(high_memory);
83 EXPORT_SYMBOL(vmalloc_earlyreserve);
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
91 void pgd_clear_bad(pgd_t *pgd)
93 pgd_ERROR(*pgd);
94 pgd_clear(pgd);
97 void pud_clear_bad(pud_t *pud)
99 pud_ERROR(*pud);
100 pud_clear(pud);
103 void pmd_clear_bad(pmd_t *pmd)
105 pmd_ERROR(*pmd);
106 pmd_clear(pmd);
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
113 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
115 struct page *page = pmd_page(*pmd);
116 pmd_clear(pmd);
117 pte_lock_deinit(page);
118 pte_free_tlb(tlb, page);
119 dec_page_state(nr_page_table_pages);
120 tlb->mm->nr_ptes--;
123 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
124 unsigned long addr, unsigned long end,
125 unsigned long floor, unsigned long ceiling)
127 pmd_t *pmd;
128 unsigned long next;
129 unsigned long start;
131 start = addr;
132 pmd = pmd_offset(pud, addr);
133 do {
134 next = pmd_addr_end(addr, end);
135 if (pmd_none_or_clear_bad(pmd))
136 continue;
137 free_pte_range(tlb, pmd);
138 } while (pmd++, addr = next, addr != end);
140 start &= PUD_MASK;
141 if (start < floor)
142 return;
143 if (ceiling) {
144 ceiling &= PUD_MASK;
145 if (!ceiling)
146 return;
148 if (end - 1 > ceiling - 1)
149 return;
151 pmd = pmd_offset(pud, start);
152 pud_clear(pud);
153 pmd_free_tlb(tlb, pmd);
156 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
157 unsigned long addr, unsigned long end,
158 unsigned long floor, unsigned long ceiling)
160 pud_t *pud;
161 unsigned long next;
162 unsigned long start;
164 start = addr;
165 pud = pud_offset(pgd, addr);
166 do {
167 next = pud_addr_end(addr, end);
168 if (pud_none_or_clear_bad(pud))
169 continue;
170 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
171 } while (pud++, addr = next, addr != end);
173 start &= PGDIR_MASK;
174 if (start < floor)
175 return;
176 if (ceiling) {
177 ceiling &= PGDIR_MASK;
178 if (!ceiling)
179 return;
181 if (end - 1 > ceiling - 1)
182 return;
184 pud = pud_offset(pgd, start);
185 pgd_clear(pgd);
186 pud_free_tlb(tlb, pud);
190 * This function frees user-level page tables of a process.
192 * Must be called with pagetable lock held.
194 void free_pgd_range(struct mmu_gather **tlb,
195 unsigned long addr, unsigned long end,
196 unsigned long floor, unsigned long ceiling)
198 pgd_t *pgd;
199 unsigned long next;
200 unsigned long start;
203 * The next few lines have given us lots of grief...
205 * Why are we testing PMD* at this top level? Because often
206 * there will be no work to do at all, and we'd prefer not to
207 * go all the way down to the bottom just to discover that.
209 * Why all these "- 1"s? Because 0 represents both the bottom
210 * of the address space and the top of it (using -1 for the
211 * top wouldn't help much: the masks would do the wrong thing).
212 * The rule is that addr 0 and floor 0 refer to the bottom of
213 * the address space, but end 0 and ceiling 0 refer to the top
214 * Comparisons need to use "end - 1" and "ceiling - 1" (though
215 * that end 0 case should be mythical).
217 * Wherever addr is brought up or ceiling brought down, we must
218 * be careful to reject "the opposite 0" before it confuses the
219 * subsequent tests. But what about where end is brought down
220 * by PMD_SIZE below? no, end can't go down to 0 there.
222 * Whereas we round start (addr) and ceiling down, by different
223 * masks at different levels, in order to test whether a table
224 * now has no other vmas using it, so can be freed, we don't
225 * bother to round floor or end up - the tests don't need that.
228 addr &= PMD_MASK;
229 if (addr < floor) {
230 addr += PMD_SIZE;
231 if (!addr)
232 return;
234 if (ceiling) {
235 ceiling &= PMD_MASK;
236 if (!ceiling)
237 return;
239 if (end - 1 > ceiling - 1)
240 end -= PMD_SIZE;
241 if (addr > end - 1)
242 return;
244 start = addr;
245 pgd = pgd_offset((*tlb)->mm, addr);
246 do {
247 next = pgd_addr_end(addr, end);
248 if (pgd_none_or_clear_bad(pgd))
249 continue;
250 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
251 } while (pgd++, addr = next, addr != end);
253 if (!(*tlb)->fullmm)
254 flush_tlb_pgtables((*tlb)->mm, start, end);
257 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
258 unsigned long floor, unsigned long ceiling)
260 while (vma) {
261 struct vm_area_struct *next = vma->vm_next;
262 unsigned long addr = vma->vm_start;
265 * Hide vma from rmap and vmtruncate before freeing pgtables
267 anon_vma_unlink(vma);
268 unlink_file_vma(vma);
270 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
271 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
272 floor, next? next->vm_start: ceiling);
273 } else {
275 * Optimization: gather nearby vmas into one call down
277 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
278 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
279 HPAGE_SIZE)) {
280 vma = next;
281 next = vma->vm_next;
282 anon_vma_unlink(vma);
283 unlink_file_vma(vma);
285 free_pgd_range(tlb, addr, vma->vm_end,
286 floor, next? next->vm_start: ceiling);
288 vma = next;
292 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
294 struct page *new = pte_alloc_one(mm, address);
295 if (!new)
296 return -ENOMEM;
298 pte_lock_init(new);
299 spin_lock(&mm->page_table_lock);
300 if (pmd_present(*pmd)) { /* Another has populated it */
301 pte_lock_deinit(new);
302 pte_free(new);
303 } else {
304 mm->nr_ptes++;
305 inc_page_state(nr_page_table_pages);
306 pmd_populate(mm, pmd, new);
308 spin_unlock(&mm->page_table_lock);
309 return 0;
312 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
314 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
315 if (!new)
316 return -ENOMEM;
318 spin_lock(&init_mm.page_table_lock);
319 if (pmd_present(*pmd)) /* Another has populated it */
320 pte_free_kernel(new);
321 else
322 pmd_populate_kernel(&init_mm, pmd, new);
323 spin_unlock(&init_mm.page_table_lock);
324 return 0;
327 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
329 if (file_rss)
330 add_mm_counter(mm, file_rss, file_rss);
331 if (anon_rss)
332 add_mm_counter(mm, anon_rss, anon_rss);
336 * This function is called to print an error when a pte in a
337 * !VM_RESERVED region is found pointing to an invalid pfn (which
338 * is an error.
340 * The calling function must still handle the error.
342 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
344 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
345 "vm_flags = %lx, vaddr = %lx\n",
346 (long long)pte_val(pte),
347 (vma->vm_mm == current->mm ? current->comm : "???"),
348 vma->vm_flags, vaddr);
349 dump_stack();
353 * copy one vm_area from one task to the other. Assumes the page tables
354 * already present in the new task to be cleared in the whole range
355 * covered by this vma.
358 static inline void
359 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
360 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
361 unsigned long addr, int *rss)
363 unsigned long vm_flags = vma->vm_flags;
364 pte_t pte = *src_pte;
365 struct page *page;
366 unsigned long pfn;
368 /* pte contains position in swap or file, so copy. */
369 if (unlikely(!pte_present(pte))) {
370 if (!pte_file(pte)) {
371 swap_duplicate(pte_to_swp_entry(pte));
372 /* make sure dst_mm is on swapoff's mmlist. */
373 if (unlikely(list_empty(&dst_mm->mmlist))) {
374 spin_lock(&mmlist_lock);
375 if (list_empty(&dst_mm->mmlist))
376 list_add(&dst_mm->mmlist,
377 &src_mm->mmlist);
378 spin_unlock(&mmlist_lock);
381 goto out_set_pte;
384 /* If the region is VM_RESERVED, the mapping is not
385 * mapped via rmap - duplicate the pte as is.
387 if (vm_flags & VM_RESERVED)
388 goto out_set_pte;
390 pfn = pte_pfn(pte);
391 /* If the pte points outside of valid memory but
392 * the region is not VM_RESERVED, we have a problem.
394 if (unlikely(!pfn_valid(pfn))) {
395 print_bad_pte(vma, pte, addr);
396 goto out_set_pte; /* try to do something sane */
399 page = pfn_to_page(pfn);
402 * If it's a COW mapping, write protect it both
403 * in the parent and the child
405 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
406 ptep_set_wrprotect(src_mm, addr, src_pte);
407 pte = *src_pte;
411 * If it's a shared mapping, mark it clean in
412 * the child
414 if (vm_flags & VM_SHARED)
415 pte = pte_mkclean(pte);
416 pte = pte_mkold(pte);
417 get_page(page);
418 page_dup_rmap(page);
419 rss[!!PageAnon(page)]++;
421 out_set_pte:
422 set_pte_at(dst_mm, addr, dst_pte, pte);
425 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
426 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
427 unsigned long addr, unsigned long end)
429 pte_t *src_pte, *dst_pte;
430 spinlock_t *src_ptl, *dst_ptl;
431 int progress = 0;
432 int rss[2];
434 again:
435 rss[1] = rss[0] = 0;
436 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
437 if (!dst_pte)
438 return -ENOMEM;
439 src_pte = pte_offset_map_nested(src_pmd, addr);
440 src_ptl = pte_lockptr(src_mm, src_pmd);
441 spin_lock(src_ptl);
443 do {
445 * We are holding two locks at this point - either of them
446 * could generate latencies in another task on another CPU.
448 if (progress >= 32) {
449 progress = 0;
450 if (need_resched() ||
451 need_lockbreak(src_ptl) ||
452 need_lockbreak(dst_ptl))
453 break;
455 if (pte_none(*src_pte)) {
456 progress++;
457 continue;
459 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
460 progress += 8;
461 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
463 spin_unlock(src_ptl);
464 pte_unmap_nested(src_pte - 1);
465 add_mm_rss(dst_mm, rss[0], rss[1]);
466 pte_unmap_unlock(dst_pte - 1, dst_ptl);
467 cond_resched();
468 if (addr != end)
469 goto again;
470 return 0;
473 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
474 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
475 unsigned long addr, unsigned long end)
477 pmd_t *src_pmd, *dst_pmd;
478 unsigned long next;
480 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
481 if (!dst_pmd)
482 return -ENOMEM;
483 src_pmd = pmd_offset(src_pud, addr);
484 do {
485 next = pmd_addr_end(addr, end);
486 if (pmd_none_or_clear_bad(src_pmd))
487 continue;
488 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
489 vma, addr, next))
490 return -ENOMEM;
491 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
492 return 0;
495 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
496 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
497 unsigned long addr, unsigned long end)
499 pud_t *src_pud, *dst_pud;
500 unsigned long next;
502 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
503 if (!dst_pud)
504 return -ENOMEM;
505 src_pud = pud_offset(src_pgd, addr);
506 do {
507 next = pud_addr_end(addr, end);
508 if (pud_none_or_clear_bad(src_pud))
509 continue;
510 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
511 vma, addr, next))
512 return -ENOMEM;
513 } while (dst_pud++, src_pud++, addr = next, addr != end);
514 return 0;
517 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
518 struct vm_area_struct *vma)
520 pgd_t *src_pgd, *dst_pgd;
521 unsigned long next;
522 unsigned long addr = vma->vm_start;
523 unsigned long end = vma->vm_end;
526 * Don't copy ptes where a page fault will fill them correctly.
527 * Fork becomes much lighter when there are big shared or private
528 * readonly mappings. The tradeoff is that copy_page_range is more
529 * efficient than faulting.
531 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_RESERVED))) {
532 if (!vma->anon_vma)
533 return 0;
536 if (is_vm_hugetlb_page(vma))
537 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
539 dst_pgd = pgd_offset(dst_mm, addr);
540 src_pgd = pgd_offset(src_mm, addr);
541 do {
542 next = pgd_addr_end(addr, end);
543 if (pgd_none_or_clear_bad(src_pgd))
544 continue;
545 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
546 vma, addr, next))
547 return -ENOMEM;
548 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
549 return 0;
552 static unsigned long zap_pte_range(struct mmu_gather *tlb,
553 struct vm_area_struct *vma, pmd_t *pmd,
554 unsigned long addr, unsigned long end,
555 long *zap_work, struct zap_details *details)
557 struct mm_struct *mm = tlb->mm;
558 pte_t *pte;
559 spinlock_t *ptl;
560 int file_rss = 0;
561 int anon_rss = 0;
563 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
564 do {
565 pte_t ptent = *pte;
566 if (pte_none(ptent)) {
567 (*zap_work)--;
568 continue;
570 if (pte_present(ptent)) {
571 struct page *page = NULL;
573 (*zap_work) -= PAGE_SIZE;
575 if (!(vma->vm_flags & VM_RESERVED)) {
576 unsigned long pfn = pte_pfn(ptent);
577 if (unlikely(!pfn_valid(pfn)))
578 print_bad_pte(vma, ptent, addr);
579 else
580 page = pfn_to_page(pfn);
582 if (unlikely(details) && page) {
584 * unmap_shared_mapping_pages() wants to
585 * invalidate cache without truncating:
586 * unmap shared but keep private pages.
588 if (details->check_mapping &&
589 details->check_mapping != page->mapping)
590 continue;
592 * Each page->index must be checked when
593 * invalidating or truncating nonlinear.
595 if (details->nonlinear_vma &&
596 (page->index < details->first_index ||
597 page->index > details->last_index))
598 continue;
600 ptent = ptep_get_and_clear_full(mm, addr, pte,
601 tlb->fullmm);
602 tlb_remove_tlb_entry(tlb, pte, addr);
603 if (unlikely(!page))
604 continue;
605 if (unlikely(details) && details->nonlinear_vma
606 && linear_page_index(details->nonlinear_vma,
607 addr) != page->index)
608 set_pte_at(mm, addr, pte,
609 pgoff_to_pte(page->index));
610 if (PageAnon(page))
611 anon_rss--;
612 else {
613 if (pte_dirty(ptent))
614 set_page_dirty(page);
615 if (pte_young(ptent))
616 mark_page_accessed(page);
617 file_rss--;
619 page_remove_rmap(page);
620 tlb_remove_page(tlb, page);
621 continue;
624 * If details->check_mapping, we leave swap entries;
625 * if details->nonlinear_vma, we leave file entries.
627 if (unlikely(details))
628 continue;
629 if (!pte_file(ptent))
630 free_swap_and_cache(pte_to_swp_entry(ptent));
631 pte_clear_full(mm, addr, pte, tlb->fullmm);
632 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
634 add_mm_rss(mm, file_rss, anon_rss);
635 pte_unmap_unlock(pte - 1, ptl);
637 return addr;
640 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
641 struct vm_area_struct *vma, pud_t *pud,
642 unsigned long addr, unsigned long end,
643 long *zap_work, struct zap_details *details)
645 pmd_t *pmd;
646 unsigned long next;
648 pmd = pmd_offset(pud, addr);
649 do {
650 next = pmd_addr_end(addr, end);
651 if (pmd_none_or_clear_bad(pmd)) {
652 (*zap_work)--;
653 continue;
655 next = zap_pte_range(tlb, vma, pmd, addr, next,
656 zap_work, details);
657 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
659 return addr;
662 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
663 struct vm_area_struct *vma, pgd_t *pgd,
664 unsigned long addr, unsigned long end,
665 long *zap_work, struct zap_details *details)
667 pud_t *pud;
668 unsigned long next;
670 pud = pud_offset(pgd, addr);
671 do {
672 next = pud_addr_end(addr, end);
673 if (pud_none_or_clear_bad(pud)) {
674 (*zap_work)--;
675 continue;
677 next = zap_pmd_range(tlb, vma, pud, addr, next,
678 zap_work, details);
679 } while (pud++, addr = next, (addr != end && *zap_work > 0));
681 return addr;
684 static unsigned long unmap_page_range(struct mmu_gather *tlb,
685 struct vm_area_struct *vma,
686 unsigned long addr, unsigned long end,
687 long *zap_work, struct zap_details *details)
689 pgd_t *pgd;
690 unsigned long next;
692 if (details && !details->check_mapping && !details->nonlinear_vma)
693 details = NULL;
695 BUG_ON(addr >= end);
696 tlb_start_vma(tlb, vma);
697 pgd = pgd_offset(vma->vm_mm, addr);
698 do {
699 next = pgd_addr_end(addr, end);
700 if (pgd_none_or_clear_bad(pgd)) {
701 (*zap_work)--;
702 continue;
704 next = zap_pud_range(tlb, vma, pgd, addr, next,
705 zap_work, details);
706 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
707 tlb_end_vma(tlb, vma);
709 return addr;
712 #ifdef CONFIG_PREEMPT
713 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
714 #else
715 /* No preempt: go for improved straight-line efficiency */
716 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
717 #endif
720 * unmap_vmas - unmap a range of memory covered by a list of vma's
721 * @tlbp: address of the caller's struct mmu_gather
722 * @vma: the starting vma
723 * @start_addr: virtual address at which to start unmapping
724 * @end_addr: virtual address at which to end unmapping
725 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
726 * @details: details of nonlinear truncation or shared cache invalidation
728 * Returns the end address of the unmapping (restart addr if interrupted).
730 * Unmap all pages in the vma list.
732 * We aim to not hold locks for too long (for scheduling latency reasons).
733 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
734 * return the ending mmu_gather to the caller.
736 * Only addresses between `start' and `end' will be unmapped.
738 * The VMA list must be sorted in ascending virtual address order.
740 * unmap_vmas() assumes that the caller will flush the whole unmapped address
741 * range after unmap_vmas() returns. So the only responsibility here is to
742 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
743 * drops the lock and schedules.
745 unsigned long unmap_vmas(struct mmu_gather **tlbp,
746 struct vm_area_struct *vma, unsigned long start_addr,
747 unsigned long end_addr, unsigned long *nr_accounted,
748 struct zap_details *details)
750 long zap_work = ZAP_BLOCK_SIZE;
751 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
752 int tlb_start_valid = 0;
753 unsigned long start = start_addr;
754 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
755 int fullmm = (*tlbp)->fullmm;
757 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
758 unsigned long end;
760 start = max(vma->vm_start, start_addr);
761 if (start >= vma->vm_end)
762 continue;
763 end = min(vma->vm_end, end_addr);
764 if (end <= vma->vm_start)
765 continue;
767 if (vma->vm_flags & VM_ACCOUNT)
768 *nr_accounted += (end - start) >> PAGE_SHIFT;
770 while (start != end) {
771 if (!tlb_start_valid) {
772 tlb_start = start;
773 tlb_start_valid = 1;
776 if (unlikely(is_vm_hugetlb_page(vma))) {
777 unmap_hugepage_range(vma, start, end);
778 zap_work -= (end - start) /
779 (HPAGE_SIZE / PAGE_SIZE);
780 start = end;
781 } else
782 start = unmap_page_range(*tlbp, vma,
783 start, end, &zap_work, details);
785 if (zap_work > 0) {
786 BUG_ON(start != end);
787 break;
790 tlb_finish_mmu(*tlbp, tlb_start, start);
792 if (need_resched() ||
793 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
794 if (i_mmap_lock) {
795 *tlbp = NULL;
796 goto out;
798 cond_resched();
801 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
802 tlb_start_valid = 0;
803 zap_work = ZAP_BLOCK_SIZE;
806 out:
807 return start; /* which is now the end (or restart) address */
811 * zap_page_range - remove user pages in a given range
812 * @vma: vm_area_struct holding the applicable pages
813 * @address: starting address of pages to zap
814 * @size: number of bytes to zap
815 * @details: details of nonlinear truncation or shared cache invalidation
817 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
818 unsigned long size, struct zap_details *details)
820 struct mm_struct *mm = vma->vm_mm;
821 struct mmu_gather *tlb;
822 unsigned long end = address + size;
823 unsigned long nr_accounted = 0;
825 lru_add_drain();
826 tlb = tlb_gather_mmu(mm, 0);
827 update_hiwater_rss(mm);
828 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
829 if (tlb)
830 tlb_finish_mmu(tlb, address, end);
831 return end;
835 * Do a quick page-table lookup for a single page.
837 struct page *follow_page(struct mm_struct *mm, unsigned long address,
838 unsigned int flags)
840 pgd_t *pgd;
841 pud_t *pud;
842 pmd_t *pmd;
843 pte_t *ptep, pte;
844 spinlock_t *ptl;
845 unsigned long pfn;
846 struct page *page;
848 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
849 if (!IS_ERR(page)) {
850 BUG_ON(flags & FOLL_GET);
851 goto out;
854 page = NULL;
855 pgd = pgd_offset(mm, address);
856 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
857 goto no_page_table;
859 pud = pud_offset(pgd, address);
860 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
861 goto no_page_table;
863 pmd = pmd_offset(pud, address);
864 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
865 goto no_page_table;
867 if (pmd_huge(*pmd)) {
868 BUG_ON(flags & FOLL_GET);
869 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
870 goto out;
873 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
874 if (!ptep)
875 goto out;
877 pte = *ptep;
878 if (!pte_present(pte))
879 goto unlock;
880 if ((flags & FOLL_WRITE) && !pte_write(pte))
881 goto unlock;
882 pfn = pte_pfn(pte);
883 if (!pfn_valid(pfn))
884 goto unlock;
886 page = pfn_to_page(pfn);
887 if (flags & FOLL_GET)
888 get_page(page);
889 if (flags & FOLL_TOUCH) {
890 if ((flags & FOLL_WRITE) &&
891 !pte_dirty(pte) && !PageDirty(page))
892 set_page_dirty(page);
893 mark_page_accessed(page);
895 unlock:
896 pte_unmap_unlock(ptep, ptl);
897 out:
898 return page;
900 no_page_table:
902 * When core dumping an enormous anonymous area that nobody
903 * has touched so far, we don't want to allocate page tables.
905 if (flags & FOLL_ANON) {
906 page = ZERO_PAGE(address);
907 if (flags & FOLL_GET)
908 get_page(page);
909 BUG_ON(flags & FOLL_WRITE);
911 return page;
914 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
915 unsigned long start, int len, int write, int force,
916 struct page **pages, struct vm_area_struct **vmas)
918 int i;
919 unsigned int vm_flags;
922 * Require read or write permissions.
923 * If 'force' is set, we only require the "MAY" flags.
925 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
926 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
927 i = 0;
929 do {
930 struct vm_area_struct *vma;
931 unsigned int foll_flags;
933 vma = find_extend_vma(mm, start);
934 if (!vma && in_gate_area(tsk, start)) {
935 unsigned long pg = start & PAGE_MASK;
936 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
937 pgd_t *pgd;
938 pud_t *pud;
939 pmd_t *pmd;
940 pte_t *pte;
941 if (write) /* user gate pages are read-only */
942 return i ? : -EFAULT;
943 if (pg > TASK_SIZE)
944 pgd = pgd_offset_k(pg);
945 else
946 pgd = pgd_offset_gate(mm, pg);
947 BUG_ON(pgd_none(*pgd));
948 pud = pud_offset(pgd, pg);
949 BUG_ON(pud_none(*pud));
950 pmd = pmd_offset(pud, pg);
951 if (pmd_none(*pmd))
952 return i ? : -EFAULT;
953 pte = pte_offset_map(pmd, pg);
954 if (pte_none(*pte)) {
955 pte_unmap(pte);
956 return i ? : -EFAULT;
958 if (pages) {
959 pages[i] = pte_page(*pte);
960 get_page(pages[i]);
962 pte_unmap(pte);
963 if (vmas)
964 vmas[i] = gate_vma;
965 i++;
966 start += PAGE_SIZE;
967 len--;
968 continue;
971 if (!vma || (vma->vm_flags & (VM_IO | VM_RESERVED))
972 || !(vm_flags & vma->vm_flags))
973 return i ? : -EFAULT;
975 if (is_vm_hugetlb_page(vma)) {
976 i = follow_hugetlb_page(mm, vma, pages, vmas,
977 &start, &len, i);
978 continue;
981 foll_flags = FOLL_TOUCH;
982 if (pages)
983 foll_flags |= FOLL_GET;
984 if (!write && !(vma->vm_flags & VM_LOCKED) &&
985 (!vma->vm_ops || !vma->vm_ops->nopage))
986 foll_flags |= FOLL_ANON;
988 do {
989 struct page *page;
991 if (write)
992 foll_flags |= FOLL_WRITE;
994 cond_resched();
995 while (!(page = follow_page(mm, start, foll_flags))) {
996 int ret;
997 ret = __handle_mm_fault(mm, vma, start,
998 foll_flags & FOLL_WRITE);
1000 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1001 * broken COW when necessary, even if maybe_mkwrite
1002 * decided not to set pte_write. We can thus safely do
1003 * subsequent page lookups as if they were reads.
1005 if (ret & VM_FAULT_WRITE)
1006 foll_flags &= ~FOLL_WRITE;
1008 switch (ret & ~VM_FAULT_WRITE) {
1009 case VM_FAULT_MINOR:
1010 tsk->min_flt++;
1011 break;
1012 case VM_FAULT_MAJOR:
1013 tsk->maj_flt++;
1014 break;
1015 case VM_FAULT_SIGBUS:
1016 return i ? i : -EFAULT;
1017 case VM_FAULT_OOM:
1018 return i ? i : -ENOMEM;
1019 default:
1020 BUG();
1023 if (pages) {
1024 pages[i] = page;
1025 flush_dcache_page(page);
1027 if (vmas)
1028 vmas[i] = vma;
1029 i++;
1030 start += PAGE_SIZE;
1031 len--;
1032 } while (len && start < vma->vm_end);
1033 } while (len);
1034 return i;
1036 EXPORT_SYMBOL(get_user_pages);
1038 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1039 unsigned long addr, unsigned long end, pgprot_t prot)
1041 pte_t *pte;
1042 spinlock_t *ptl;
1044 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1045 if (!pte)
1046 return -ENOMEM;
1047 do {
1048 struct page *page = ZERO_PAGE(addr);
1049 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1050 page_cache_get(page);
1051 page_add_file_rmap(page);
1052 inc_mm_counter(mm, file_rss);
1053 BUG_ON(!pte_none(*pte));
1054 set_pte_at(mm, addr, pte, zero_pte);
1055 } while (pte++, addr += PAGE_SIZE, addr != end);
1056 pte_unmap_unlock(pte - 1, ptl);
1057 return 0;
1060 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1061 unsigned long addr, unsigned long end, pgprot_t prot)
1063 pmd_t *pmd;
1064 unsigned long next;
1066 pmd = pmd_alloc(mm, pud, addr);
1067 if (!pmd)
1068 return -ENOMEM;
1069 do {
1070 next = pmd_addr_end(addr, end);
1071 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1072 return -ENOMEM;
1073 } while (pmd++, addr = next, addr != end);
1074 return 0;
1077 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1078 unsigned long addr, unsigned long end, pgprot_t prot)
1080 pud_t *pud;
1081 unsigned long next;
1083 pud = pud_alloc(mm, pgd, addr);
1084 if (!pud)
1085 return -ENOMEM;
1086 do {
1087 next = pud_addr_end(addr, end);
1088 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1089 return -ENOMEM;
1090 } while (pud++, addr = next, addr != end);
1091 return 0;
1094 int zeromap_page_range(struct vm_area_struct *vma,
1095 unsigned long addr, unsigned long size, pgprot_t prot)
1097 pgd_t *pgd;
1098 unsigned long next;
1099 unsigned long end = addr + size;
1100 struct mm_struct *mm = vma->vm_mm;
1101 int err;
1103 BUG_ON(addr >= end);
1104 pgd = pgd_offset(mm, addr);
1105 flush_cache_range(vma, addr, end);
1106 do {
1107 next = pgd_addr_end(addr, end);
1108 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1109 if (err)
1110 break;
1111 } while (pgd++, addr = next, addr != end);
1112 return err;
1116 * maps a range of physical memory into the requested pages. the old
1117 * mappings are removed. any references to nonexistent pages results
1118 * in null mappings (currently treated as "copy-on-access")
1120 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1121 unsigned long addr, unsigned long end,
1122 unsigned long pfn, pgprot_t prot)
1124 pte_t *pte;
1125 spinlock_t *ptl;
1127 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1128 if (!pte)
1129 return -ENOMEM;
1130 do {
1131 BUG_ON(!pte_none(*pte));
1132 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1133 pfn++;
1134 } while (pte++, addr += PAGE_SIZE, addr != end);
1135 pte_unmap_unlock(pte - 1, ptl);
1136 return 0;
1139 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1140 unsigned long addr, unsigned long end,
1141 unsigned long pfn, pgprot_t prot)
1143 pmd_t *pmd;
1144 unsigned long next;
1146 pfn -= addr >> PAGE_SHIFT;
1147 pmd = pmd_alloc(mm, pud, addr);
1148 if (!pmd)
1149 return -ENOMEM;
1150 do {
1151 next = pmd_addr_end(addr, end);
1152 if (remap_pte_range(mm, pmd, addr, next,
1153 pfn + (addr >> PAGE_SHIFT), prot))
1154 return -ENOMEM;
1155 } while (pmd++, addr = next, addr != end);
1156 return 0;
1159 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1160 unsigned long addr, unsigned long end,
1161 unsigned long pfn, pgprot_t prot)
1163 pud_t *pud;
1164 unsigned long next;
1166 pfn -= addr >> PAGE_SHIFT;
1167 pud = pud_alloc(mm, pgd, addr);
1168 if (!pud)
1169 return -ENOMEM;
1170 do {
1171 next = pud_addr_end(addr, end);
1172 if (remap_pmd_range(mm, pud, addr, next,
1173 pfn + (addr >> PAGE_SHIFT), prot))
1174 return -ENOMEM;
1175 } while (pud++, addr = next, addr != end);
1176 return 0;
1179 /* Note: this is only safe if the mm semaphore is held when called. */
1180 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1181 unsigned long pfn, unsigned long size, pgprot_t prot)
1183 pgd_t *pgd;
1184 unsigned long next;
1185 unsigned long end = addr + PAGE_ALIGN(size);
1186 struct mm_struct *mm = vma->vm_mm;
1187 int err;
1190 * Physically remapped pages are special. Tell the
1191 * rest of the world about it:
1192 * VM_IO tells people not to look at these pages
1193 * (accesses can have side effects).
1194 * VM_RESERVED tells the core MM not to "manage" these pages
1195 * (e.g. refcount, mapcount, try to swap them out).
1197 vma->vm_flags |= VM_IO | VM_RESERVED;
1199 BUG_ON(addr >= end);
1200 pfn -= addr >> PAGE_SHIFT;
1201 pgd = pgd_offset(mm, addr);
1202 flush_cache_range(vma, addr, end);
1203 do {
1204 next = pgd_addr_end(addr, end);
1205 err = remap_pud_range(mm, pgd, addr, next,
1206 pfn + (addr >> PAGE_SHIFT), prot);
1207 if (err)
1208 break;
1209 } while (pgd++, addr = next, addr != end);
1210 return err;
1212 EXPORT_SYMBOL(remap_pfn_range);
1215 * handle_pte_fault chooses page fault handler according to an entry
1216 * which was read non-atomically. Before making any commitment, on
1217 * those architectures or configurations (e.g. i386 with PAE) which
1218 * might give a mix of unmatched parts, do_swap_page and do_file_page
1219 * must check under lock before unmapping the pte and proceeding
1220 * (but do_wp_page is only called after already making such a check;
1221 * and do_anonymous_page and do_no_page can safely check later on).
1223 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1224 pte_t *page_table, pte_t orig_pte)
1226 int same = 1;
1227 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1228 if (sizeof(pte_t) > sizeof(unsigned long)) {
1229 spinlock_t *ptl = pte_lockptr(mm, pmd);
1230 spin_lock(ptl);
1231 same = pte_same(*page_table, orig_pte);
1232 spin_unlock(ptl);
1234 #endif
1235 pte_unmap(page_table);
1236 return same;
1240 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1241 * servicing faults for write access. In the normal case, do always want
1242 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1243 * that do not have writing enabled, when used by access_process_vm.
1245 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1247 if (likely(vma->vm_flags & VM_WRITE))
1248 pte = pte_mkwrite(pte);
1249 return pte;
1253 * This routine handles present pages, when users try to write
1254 * to a shared page. It is done by copying the page to a new address
1255 * and decrementing the shared-page counter for the old page.
1257 * Note that this routine assumes that the protection checks have been
1258 * done by the caller (the low-level page fault routine in most cases).
1259 * Thus we can safely just mark it writable once we've done any necessary
1260 * COW.
1262 * We also mark the page dirty at this point even though the page will
1263 * change only once the write actually happens. This avoids a few races,
1264 * and potentially makes it more efficient.
1266 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1267 * but allow concurrent faults), with pte both mapped and locked.
1268 * We return with mmap_sem still held, but pte unmapped and unlocked.
1270 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1271 unsigned long address, pte_t *page_table, pmd_t *pmd,
1272 spinlock_t *ptl, pte_t orig_pte)
1274 struct page *old_page, *new_page;
1275 unsigned long pfn = pte_pfn(orig_pte);
1276 pte_t entry;
1277 int ret = VM_FAULT_MINOR;
1279 BUG_ON(vma->vm_flags & VM_RESERVED);
1281 if (unlikely(!pfn_valid(pfn))) {
1283 * Page table corrupted: show pte and kill process.
1285 print_bad_pte(vma, orig_pte, address);
1286 ret = VM_FAULT_OOM;
1287 goto unlock;
1289 old_page = pfn_to_page(pfn);
1291 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1292 int reuse = can_share_swap_page(old_page);
1293 unlock_page(old_page);
1294 if (reuse) {
1295 flush_cache_page(vma, address, pfn);
1296 entry = pte_mkyoung(orig_pte);
1297 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1298 ptep_set_access_flags(vma, address, page_table, entry, 1);
1299 update_mmu_cache(vma, address, entry);
1300 lazy_mmu_prot_update(entry);
1301 ret |= VM_FAULT_WRITE;
1302 goto unlock;
1307 * Ok, we need to copy. Oh, well..
1309 page_cache_get(old_page);
1310 pte_unmap_unlock(page_table, ptl);
1312 if (unlikely(anon_vma_prepare(vma)))
1313 goto oom;
1314 if (old_page == ZERO_PAGE(address)) {
1315 new_page = alloc_zeroed_user_highpage(vma, address);
1316 if (!new_page)
1317 goto oom;
1318 } else {
1319 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1320 if (!new_page)
1321 goto oom;
1322 copy_user_highpage(new_page, old_page, address);
1326 * Re-check the pte - we dropped the lock
1328 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1329 if (likely(pte_same(*page_table, orig_pte))) {
1330 page_remove_rmap(old_page);
1331 if (!PageAnon(old_page)) {
1332 inc_mm_counter(mm, anon_rss);
1333 dec_mm_counter(mm, file_rss);
1335 flush_cache_page(vma, address, pfn);
1336 entry = mk_pte(new_page, vma->vm_page_prot);
1337 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1338 ptep_establish(vma, address, page_table, entry);
1339 update_mmu_cache(vma, address, entry);
1340 lazy_mmu_prot_update(entry);
1341 lru_cache_add_active(new_page);
1342 page_add_anon_rmap(new_page, vma, address);
1344 /* Free the old page.. */
1345 new_page = old_page;
1346 ret |= VM_FAULT_WRITE;
1348 page_cache_release(new_page);
1349 page_cache_release(old_page);
1350 unlock:
1351 pte_unmap_unlock(page_table, ptl);
1352 return ret;
1353 oom:
1354 page_cache_release(old_page);
1355 return VM_FAULT_OOM;
1359 * Helper functions for unmap_mapping_range().
1361 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1363 * We have to restart searching the prio_tree whenever we drop the lock,
1364 * since the iterator is only valid while the lock is held, and anyway
1365 * a later vma might be split and reinserted earlier while lock dropped.
1367 * The list of nonlinear vmas could be handled more efficiently, using
1368 * a placeholder, but handle it in the same way until a need is shown.
1369 * It is important to search the prio_tree before nonlinear list: a vma
1370 * may become nonlinear and be shifted from prio_tree to nonlinear list
1371 * while the lock is dropped; but never shifted from list to prio_tree.
1373 * In order to make forward progress despite restarting the search,
1374 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1375 * quickly skip it next time around. Since the prio_tree search only
1376 * shows us those vmas affected by unmapping the range in question, we
1377 * can't efficiently keep all vmas in step with mapping->truncate_count:
1378 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1379 * mapping->truncate_count and vma->vm_truncate_count are protected by
1380 * i_mmap_lock.
1382 * In order to make forward progress despite repeatedly restarting some
1383 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1384 * and restart from that address when we reach that vma again. It might
1385 * have been split or merged, shrunk or extended, but never shifted: so
1386 * restart_addr remains valid so long as it remains in the vma's range.
1387 * unmap_mapping_range forces truncate_count to leap over page-aligned
1388 * values so we can save vma's restart_addr in its truncate_count field.
1390 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1392 static void reset_vma_truncate_counts(struct address_space *mapping)
1394 struct vm_area_struct *vma;
1395 struct prio_tree_iter iter;
1397 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1398 vma->vm_truncate_count = 0;
1399 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1400 vma->vm_truncate_count = 0;
1403 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1404 unsigned long start_addr, unsigned long end_addr,
1405 struct zap_details *details)
1407 unsigned long restart_addr;
1408 int need_break;
1410 again:
1411 restart_addr = vma->vm_truncate_count;
1412 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1413 start_addr = restart_addr;
1414 if (start_addr >= end_addr) {
1415 /* Top of vma has been split off since last time */
1416 vma->vm_truncate_count = details->truncate_count;
1417 return 0;
1421 restart_addr = zap_page_range(vma, start_addr,
1422 end_addr - start_addr, details);
1423 need_break = need_resched() ||
1424 need_lockbreak(details->i_mmap_lock);
1426 if (restart_addr >= end_addr) {
1427 /* We have now completed this vma: mark it so */
1428 vma->vm_truncate_count = details->truncate_count;
1429 if (!need_break)
1430 return 0;
1431 } else {
1432 /* Note restart_addr in vma's truncate_count field */
1433 vma->vm_truncate_count = restart_addr;
1434 if (!need_break)
1435 goto again;
1438 spin_unlock(details->i_mmap_lock);
1439 cond_resched();
1440 spin_lock(details->i_mmap_lock);
1441 return -EINTR;
1444 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1445 struct zap_details *details)
1447 struct vm_area_struct *vma;
1448 struct prio_tree_iter iter;
1449 pgoff_t vba, vea, zba, zea;
1451 restart:
1452 vma_prio_tree_foreach(vma, &iter, root,
1453 details->first_index, details->last_index) {
1454 /* Skip quickly over those we have already dealt with */
1455 if (vma->vm_truncate_count == details->truncate_count)
1456 continue;
1458 vba = vma->vm_pgoff;
1459 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1460 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1461 zba = details->first_index;
1462 if (zba < vba)
1463 zba = vba;
1464 zea = details->last_index;
1465 if (zea > vea)
1466 zea = vea;
1468 if (unmap_mapping_range_vma(vma,
1469 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1470 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1471 details) < 0)
1472 goto restart;
1476 static inline void unmap_mapping_range_list(struct list_head *head,
1477 struct zap_details *details)
1479 struct vm_area_struct *vma;
1482 * In nonlinear VMAs there is no correspondence between virtual address
1483 * offset and file offset. So we must perform an exhaustive search
1484 * across *all* the pages in each nonlinear VMA, not just the pages
1485 * whose virtual address lies outside the file truncation point.
1487 restart:
1488 list_for_each_entry(vma, head, shared.vm_set.list) {
1489 /* Skip quickly over those we have already dealt with */
1490 if (vma->vm_truncate_count == details->truncate_count)
1491 continue;
1492 details->nonlinear_vma = vma;
1493 if (unmap_mapping_range_vma(vma, vma->vm_start,
1494 vma->vm_end, details) < 0)
1495 goto restart;
1500 * unmap_mapping_range - unmap the portion of all mmaps
1501 * in the specified address_space corresponding to the specified
1502 * page range in the underlying file.
1503 * @mapping: the address space containing mmaps to be unmapped.
1504 * @holebegin: byte in first page to unmap, relative to the start of
1505 * the underlying file. This will be rounded down to a PAGE_SIZE
1506 * boundary. Note that this is different from vmtruncate(), which
1507 * must keep the partial page. In contrast, we must get rid of
1508 * partial pages.
1509 * @holelen: size of prospective hole in bytes. This will be rounded
1510 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1511 * end of the file.
1512 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1513 * but 0 when invalidating pagecache, don't throw away private data.
1515 void unmap_mapping_range(struct address_space *mapping,
1516 loff_t const holebegin, loff_t const holelen, int even_cows)
1518 struct zap_details details;
1519 pgoff_t hba = holebegin >> PAGE_SHIFT;
1520 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1522 /* Check for overflow. */
1523 if (sizeof(holelen) > sizeof(hlen)) {
1524 long long holeend =
1525 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1526 if (holeend & ~(long long)ULONG_MAX)
1527 hlen = ULONG_MAX - hba + 1;
1530 details.check_mapping = even_cows? NULL: mapping;
1531 details.nonlinear_vma = NULL;
1532 details.first_index = hba;
1533 details.last_index = hba + hlen - 1;
1534 if (details.last_index < details.first_index)
1535 details.last_index = ULONG_MAX;
1536 details.i_mmap_lock = &mapping->i_mmap_lock;
1538 spin_lock(&mapping->i_mmap_lock);
1540 /* serialize i_size write against truncate_count write */
1541 smp_wmb();
1542 /* Protect against page faults, and endless unmapping loops */
1543 mapping->truncate_count++;
1545 * For archs where spin_lock has inclusive semantics like ia64
1546 * this smp_mb() will prevent to read pagetable contents
1547 * before the truncate_count increment is visible to
1548 * other cpus.
1550 smp_mb();
1551 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1552 if (mapping->truncate_count == 0)
1553 reset_vma_truncate_counts(mapping);
1554 mapping->truncate_count++;
1556 details.truncate_count = mapping->truncate_count;
1558 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1559 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1560 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1561 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1562 spin_unlock(&mapping->i_mmap_lock);
1564 EXPORT_SYMBOL(unmap_mapping_range);
1567 * Handle all mappings that got truncated by a "truncate()"
1568 * system call.
1570 * NOTE! We have to be ready to update the memory sharing
1571 * between the file and the memory map for a potential last
1572 * incomplete page. Ugly, but necessary.
1574 int vmtruncate(struct inode * inode, loff_t offset)
1576 struct address_space *mapping = inode->i_mapping;
1577 unsigned long limit;
1579 if (inode->i_size < offset)
1580 goto do_expand;
1582 * truncation of in-use swapfiles is disallowed - it would cause
1583 * subsequent swapout to scribble on the now-freed blocks.
1585 if (IS_SWAPFILE(inode))
1586 goto out_busy;
1587 i_size_write(inode, offset);
1588 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1589 truncate_inode_pages(mapping, offset);
1590 goto out_truncate;
1592 do_expand:
1593 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1594 if (limit != RLIM_INFINITY && offset > limit)
1595 goto out_sig;
1596 if (offset > inode->i_sb->s_maxbytes)
1597 goto out_big;
1598 i_size_write(inode, offset);
1600 out_truncate:
1601 if (inode->i_op && inode->i_op->truncate)
1602 inode->i_op->truncate(inode);
1603 return 0;
1604 out_sig:
1605 send_sig(SIGXFSZ, current, 0);
1606 out_big:
1607 return -EFBIG;
1608 out_busy:
1609 return -ETXTBSY;
1612 EXPORT_SYMBOL(vmtruncate);
1615 * Primitive swap readahead code. We simply read an aligned block of
1616 * (1 << page_cluster) entries in the swap area. This method is chosen
1617 * because it doesn't cost us any seek time. We also make sure to queue
1618 * the 'original' request together with the readahead ones...
1620 * This has been extended to use the NUMA policies from the mm triggering
1621 * the readahead.
1623 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1625 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1627 #ifdef CONFIG_NUMA
1628 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1629 #endif
1630 int i, num;
1631 struct page *new_page;
1632 unsigned long offset;
1635 * Get the number of handles we should do readahead io to.
1637 num = valid_swaphandles(entry, &offset);
1638 for (i = 0; i < num; offset++, i++) {
1639 /* Ok, do the async read-ahead now */
1640 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1641 offset), vma, addr);
1642 if (!new_page)
1643 break;
1644 page_cache_release(new_page);
1645 #ifdef CONFIG_NUMA
1647 * Find the next applicable VMA for the NUMA policy.
1649 addr += PAGE_SIZE;
1650 if (addr == 0)
1651 vma = NULL;
1652 if (vma) {
1653 if (addr >= vma->vm_end) {
1654 vma = next_vma;
1655 next_vma = vma ? vma->vm_next : NULL;
1657 if (vma && addr < vma->vm_start)
1658 vma = NULL;
1659 } else {
1660 if (next_vma && addr >= next_vma->vm_start) {
1661 vma = next_vma;
1662 next_vma = vma->vm_next;
1665 #endif
1667 lru_add_drain(); /* Push any new pages onto the LRU now */
1671 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1672 * but allow concurrent faults), and pte mapped but not yet locked.
1673 * We return with mmap_sem still held, but pte unmapped and unlocked.
1675 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1676 unsigned long address, pte_t *page_table, pmd_t *pmd,
1677 int write_access, pte_t orig_pte)
1679 spinlock_t *ptl;
1680 struct page *page;
1681 swp_entry_t entry;
1682 pte_t pte;
1683 int ret = VM_FAULT_MINOR;
1685 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
1686 goto out;
1688 entry = pte_to_swp_entry(orig_pte);
1689 page = lookup_swap_cache(entry);
1690 if (!page) {
1691 swapin_readahead(entry, address, vma);
1692 page = read_swap_cache_async(entry, vma, address);
1693 if (!page) {
1695 * Back out if somebody else faulted in this pte
1696 * while we released the pte lock.
1698 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1699 if (likely(pte_same(*page_table, orig_pte)))
1700 ret = VM_FAULT_OOM;
1701 goto unlock;
1704 /* Had to read the page from swap area: Major fault */
1705 ret = VM_FAULT_MAJOR;
1706 inc_page_state(pgmajfault);
1707 grab_swap_token();
1710 mark_page_accessed(page);
1711 lock_page(page);
1714 * Back out if somebody else already faulted in this pte.
1716 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1717 if (unlikely(!pte_same(*page_table, orig_pte)))
1718 goto out_nomap;
1720 if (unlikely(!PageUptodate(page))) {
1721 ret = VM_FAULT_SIGBUS;
1722 goto out_nomap;
1725 /* The page isn't present yet, go ahead with the fault. */
1727 inc_mm_counter(mm, anon_rss);
1728 pte = mk_pte(page, vma->vm_page_prot);
1729 if (write_access && can_share_swap_page(page)) {
1730 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1731 write_access = 0;
1734 flush_icache_page(vma, page);
1735 set_pte_at(mm, address, page_table, pte);
1736 page_add_anon_rmap(page, vma, address);
1738 swap_free(entry);
1739 if (vm_swap_full())
1740 remove_exclusive_swap_page(page);
1741 unlock_page(page);
1743 if (write_access) {
1744 if (do_wp_page(mm, vma, address,
1745 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
1746 ret = VM_FAULT_OOM;
1747 goto out;
1750 /* No need to invalidate - it was non-present before */
1751 update_mmu_cache(vma, address, pte);
1752 lazy_mmu_prot_update(pte);
1753 unlock:
1754 pte_unmap_unlock(page_table, ptl);
1755 out:
1756 return ret;
1757 out_nomap:
1758 pte_unmap_unlock(page_table, ptl);
1759 unlock_page(page);
1760 page_cache_release(page);
1761 return ret;
1765 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1766 * but allow concurrent faults), and pte mapped but not yet locked.
1767 * We return with mmap_sem still held, but pte unmapped and unlocked.
1769 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1770 unsigned long address, pte_t *page_table, pmd_t *pmd,
1771 int write_access)
1773 struct page *page;
1774 spinlock_t *ptl;
1775 pte_t entry;
1777 if (write_access) {
1778 /* Allocate our own private page. */
1779 pte_unmap(page_table);
1781 if (unlikely(anon_vma_prepare(vma)))
1782 goto oom;
1783 page = alloc_zeroed_user_highpage(vma, address);
1784 if (!page)
1785 goto oom;
1787 entry = mk_pte(page, vma->vm_page_prot);
1788 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1790 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1791 if (!pte_none(*page_table))
1792 goto release;
1793 inc_mm_counter(mm, anon_rss);
1794 lru_cache_add_active(page);
1795 SetPageReferenced(page);
1796 page_add_anon_rmap(page, vma, address);
1797 } else {
1798 /* Map the ZERO_PAGE - vm_page_prot is readonly */
1799 page = ZERO_PAGE(address);
1800 page_cache_get(page);
1801 entry = mk_pte(page, vma->vm_page_prot);
1803 ptl = pte_lockptr(mm, pmd);
1804 spin_lock(ptl);
1805 if (!pte_none(*page_table))
1806 goto release;
1807 inc_mm_counter(mm, file_rss);
1808 page_add_file_rmap(page);
1811 set_pte_at(mm, address, page_table, entry);
1813 /* No need to invalidate - it was non-present before */
1814 update_mmu_cache(vma, address, entry);
1815 lazy_mmu_prot_update(entry);
1816 unlock:
1817 pte_unmap_unlock(page_table, ptl);
1818 return VM_FAULT_MINOR;
1819 release:
1820 page_cache_release(page);
1821 goto unlock;
1822 oom:
1823 return VM_FAULT_OOM;
1827 * do_no_page() tries to create a new page mapping. It aggressively
1828 * tries to share with existing pages, but makes a separate copy if
1829 * the "write_access" parameter is true in order to avoid the next
1830 * page fault.
1832 * As this is called only for pages that do not currently exist, we
1833 * do not need to flush old virtual caches or the TLB.
1835 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1836 * but allow concurrent faults), and pte mapped but not yet locked.
1837 * We return with mmap_sem still held, but pte unmapped and unlocked.
1839 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1840 unsigned long address, pte_t *page_table, pmd_t *pmd,
1841 int write_access)
1843 spinlock_t *ptl;
1844 struct page *new_page;
1845 struct address_space *mapping = NULL;
1846 pte_t entry;
1847 unsigned int sequence = 0;
1848 int ret = VM_FAULT_MINOR;
1849 int anon = 0;
1851 pte_unmap(page_table);
1853 if (vma->vm_file) {
1854 mapping = vma->vm_file->f_mapping;
1855 sequence = mapping->truncate_count;
1856 smp_rmb(); /* serializes i_size against truncate_count */
1858 retry:
1859 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1861 * No smp_rmb is needed here as long as there's a full
1862 * spin_lock/unlock sequence inside the ->nopage callback
1863 * (for the pagecache lookup) that acts as an implicit
1864 * smp_mb() and prevents the i_size read to happen
1865 * after the next truncate_count read.
1868 /* no page was available -- either SIGBUS or OOM */
1869 if (new_page == NOPAGE_SIGBUS)
1870 return VM_FAULT_SIGBUS;
1871 if (new_page == NOPAGE_OOM)
1872 return VM_FAULT_OOM;
1875 * Should we do an early C-O-W break?
1877 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1878 struct page *page;
1880 if (unlikely(anon_vma_prepare(vma)))
1881 goto oom;
1882 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1883 if (!page)
1884 goto oom;
1885 copy_user_highpage(page, new_page, address);
1886 page_cache_release(new_page);
1887 new_page = page;
1888 anon = 1;
1891 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1893 * For a file-backed vma, someone could have truncated or otherwise
1894 * invalidated this page. If unmap_mapping_range got called,
1895 * retry getting the page.
1897 if (mapping && unlikely(sequence != mapping->truncate_count)) {
1898 pte_unmap_unlock(page_table, ptl);
1899 page_cache_release(new_page);
1900 cond_resched();
1901 sequence = mapping->truncate_count;
1902 smp_rmb();
1903 goto retry;
1907 * This silly early PAGE_DIRTY setting removes a race
1908 * due to the bad i386 page protection. But it's valid
1909 * for other architectures too.
1911 * Note that if write_access is true, we either now have
1912 * an exclusive copy of the page, or this is a shared mapping,
1913 * so we can make it writable and dirty to avoid having to
1914 * handle that later.
1916 /* Only go through if we didn't race with anybody else... */
1917 if (pte_none(*page_table)) {
1918 flush_icache_page(vma, new_page);
1919 entry = mk_pte(new_page, vma->vm_page_prot);
1920 if (write_access)
1921 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1922 set_pte_at(mm, address, page_table, entry);
1923 if (anon) {
1924 inc_mm_counter(mm, anon_rss);
1925 lru_cache_add_active(new_page);
1926 page_add_anon_rmap(new_page, vma, address);
1927 } else if (!(vma->vm_flags & VM_RESERVED)) {
1928 inc_mm_counter(mm, file_rss);
1929 page_add_file_rmap(new_page);
1931 } else {
1932 /* One of our sibling threads was faster, back out. */
1933 page_cache_release(new_page);
1934 goto unlock;
1937 /* no need to invalidate: a not-present page shouldn't be cached */
1938 update_mmu_cache(vma, address, entry);
1939 lazy_mmu_prot_update(entry);
1940 unlock:
1941 pte_unmap_unlock(page_table, ptl);
1942 return ret;
1943 oom:
1944 page_cache_release(new_page);
1945 return VM_FAULT_OOM;
1949 * Fault of a previously existing named mapping. Repopulate the pte
1950 * from the encoded file_pte if possible. This enables swappable
1951 * nonlinear vmas.
1953 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1954 * but allow concurrent faults), and pte mapped but not yet locked.
1955 * We return with mmap_sem still held, but pte unmapped and unlocked.
1957 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
1958 unsigned long address, pte_t *page_table, pmd_t *pmd,
1959 int write_access, pte_t orig_pte)
1961 pgoff_t pgoff;
1962 int err;
1964 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
1965 return VM_FAULT_MINOR;
1967 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
1969 * Page table corrupted: show pte and kill process.
1971 print_bad_pte(vma, orig_pte, address);
1972 return VM_FAULT_OOM;
1974 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1976 pgoff = pte_to_pgoff(orig_pte);
1977 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
1978 vma->vm_page_prot, pgoff, 0);
1979 if (err == -ENOMEM)
1980 return VM_FAULT_OOM;
1981 if (err)
1982 return VM_FAULT_SIGBUS;
1983 return VM_FAULT_MAJOR;
1987 * These routines also need to handle stuff like marking pages dirty
1988 * and/or accessed for architectures that don't do it in hardware (most
1989 * RISC architectures). The early dirtying is also good on the i386.
1991 * There is also a hook called "update_mmu_cache()" that architectures
1992 * with external mmu caches can use to update those (ie the Sparc or
1993 * PowerPC hashed page tables that act as extended TLBs).
1995 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1996 * but allow concurrent faults), and pte mapped but not yet locked.
1997 * We return with mmap_sem still held, but pte unmapped and unlocked.
1999 static inline int handle_pte_fault(struct mm_struct *mm,
2000 struct vm_area_struct *vma, unsigned long address,
2001 pte_t *pte, pmd_t *pmd, int write_access)
2003 pte_t entry;
2004 pte_t old_entry;
2005 spinlock_t *ptl;
2007 old_entry = entry = *pte;
2008 if (!pte_present(entry)) {
2009 if (pte_none(entry)) {
2010 if (!vma->vm_ops || !vma->vm_ops->nopage)
2011 return do_anonymous_page(mm, vma, address,
2012 pte, pmd, write_access);
2013 return do_no_page(mm, vma, address,
2014 pte, pmd, write_access);
2016 if (pte_file(entry))
2017 return do_file_page(mm, vma, address,
2018 pte, pmd, write_access, entry);
2019 return do_swap_page(mm, vma, address,
2020 pte, pmd, write_access, entry);
2023 ptl = pte_lockptr(mm, pmd);
2024 spin_lock(ptl);
2025 if (unlikely(!pte_same(*pte, entry)))
2026 goto unlock;
2027 if (write_access) {
2028 if (!pte_write(entry))
2029 return do_wp_page(mm, vma, address,
2030 pte, pmd, ptl, entry);
2031 entry = pte_mkdirty(entry);
2033 entry = pte_mkyoung(entry);
2034 if (!pte_same(old_entry, entry)) {
2035 ptep_set_access_flags(vma, address, pte, entry, write_access);
2036 update_mmu_cache(vma, address, entry);
2037 lazy_mmu_prot_update(entry);
2038 } else {
2040 * This is needed only for protection faults but the arch code
2041 * is not yet telling us if this is a protection fault or not.
2042 * This still avoids useless tlb flushes for .text page faults
2043 * with threads.
2045 if (write_access)
2046 flush_tlb_page(vma, address);
2048 unlock:
2049 pte_unmap_unlock(pte, ptl);
2050 return VM_FAULT_MINOR;
2054 * By the time we get here, we already hold the mm semaphore
2056 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2057 unsigned long address, int write_access)
2059 pgd_t *pgd;
2060 pud_t *pud;
2061 pmd_t *pmd;
2062 pte_t *pte;
2064 __set_current_state(TASK_RUNNING);
2066 inc_page_state(pgfault);
2068 if (unlikely(is_vm_hugetlb_page(vma)))
2069 return hugetlb_fault(mm, vma, address, write_access);
2071 pgd = pgd_offset(mm, address);
2072 pud = pud_alloc(mm, pgd, address);
2073 if (!pud)
2074 return VM_FAULT_OOM;
2075 pmd = pmd_alloc(mm, pud, address);
2076 if (!pmd)
2077 return VM_FAULT_OOM;
2078 pte = pte_alloc_map(mm, pmd, address);
2079 if (!pte)
2080 return VM_FAULT_OOM;
2082 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2085 #ifndef __PAGETABLE_PUD_FOLDED
2087 * Allocate page upper directory.
2088 * We've already handled the fast-path in-line.
2090 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2092 pud_t *new = pud_alloc_one(mm, address);
2093 if (!new)
2094 return -ENOMEM;
2096 spin_lock(&mm->page_table_lock);
2097 if (pgd_present(*pgd)) /* Another has populated it */
2098 pud_free(new);
2099 else
2100 pgd_populate(mm, pgd, new);
2101 spin_unlock(&mm->page_table_lock);
2102 return 0;
2104 #endif /* __PAGETABLE_PUD_FOLDED */
2106 #ifndef __PAGETABLE_PMD_FOLDED
2108 * Allocate page middle directory.
2109 * We've already handled the fast-path in-line.
2111 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2113 pmd_t *new = pmd_alloc_one(mm, address);
2114 if (!new)
2115 return -ENOMEM;
2117 spin_lock(&mm->page_table_lock);
2118 #ifndef __ARCH_HAS_4LEVEL_HACK
2119 if (pud_present(*pud)) /* Another has populated it */
2120 pmd_free(new);
2121 else
2122 pud_populate(mm, pud, new);
2123 #else
2124 if (pgd_present(*pud)) /* Another has populated it */
2125 pmd_free(new);
2126 else
2127 pgd_populate(mm, pud, new);
2128 #endif /* __ARCH_HAS_4LEVEL_HACK */
2129 spin_unlock(&mm->page_table_lock);
2130 return 0;
2132 #endif /* __PAGETABLE_PMD_FOLDED */
2134 int make_pages_present(unsigned long addr, unsigned long end)
2136 int ret, len, write;
2137 struct vm_area_struct * vma;
2139 vma = find_vma(current->mm, addr);
2140 if (!vma)
2141 return -1;
2142 write = (vma->vm_flags & VM_WRITE) != 0;
2143 if (addr >= end)
2144 BUG();
2145 if (end > vma->vm_end)
2146 BUG();
2147 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2148 ret = get_user_pages(current, current->mm, addr,
2149 len, write, 0, NULL, NULL);
2150 if (ret < 0)
2151 return ret;
2152 return ret == len ? 0 : -1;
2156 * Map a vmalloc()-space virtual address to the physical page.
2158 struct page * vmalloc_to_page(void * vmalloc_addr)
2160 unsigned long addr = (unsigned long) vmalloc_addr;
2161 struct page *page = NULL;
2162 pgd_t *pgd = pgd_offset_k(addr);
2163 pud_t *pud;
2164 pmd_t *pmd;
2165 pte_t *ptep, pte;
2167 if (!pgd_none(*pgd)) {
2168 pud = pud_offset(pgd, addr);
2169 if (!pud_none(*pud)) {
2170 pmd = pmd_offset(pud, addr);
2171 if (!pmd_none(*pmd)) {
2172 ptep = pte_offset_map(pmd, addr);
2173 pte = *ptep;
2174 if (pte_present(pte))
2175 page = pte_page(pte);
2176 pte_unmap(ptep);
2180 return page;
2183 EXPORT_SYMBOL(vmalloc_to_page);
2186 * Map a vmalloc()-space virtual address to the physical page frame number.
2188 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2190 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2193 EXPORT_SYMBOL(vmalloc_to_pfn);
2195 #if !defined(__HAVE_ARCH_GATE_AREA)
2197 #if defined(AT_SYSINFO_EHDR)
2198 static struct vm_area_struct gate_vma;
2200 static int __init gate_vma_init(void)
2202 gate_vma.vm_mm = NULL;
2203 gate_vma.vm_start = FIXADDR_USER_START;
2204 gate_vma.vm_end = FIXADDR_USER_END;
2205 gate_vma.vm_page_prot = PAGE_READONLY;
2206 gate_vma.vm_flags = VM_RESERVED;
2207 return 0;
2209 __initcall(gate_vma_init);
2210 #endif
2212 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2214 #ifdef AT_SYSINFO_EHDR
2215 return &gate_vma;
2216 #else
2217 return NULL;
2218 #endif
2221 int in_gate_area_no_task(unsigned long addr)
2223 #ifdef AT_SYSINFO_EHDR
2224 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2225 return 1;
2226 #endif
2227 return 0;
2230 #endif /* __HAVE_ARCH_GATE_AREA */