xref: /openbmc/linux/mm/memory.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *  linux/mm/memory.c
3  *
4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
5  */
6 
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
10  */
11 
12 /*
13  * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14  * pages started 02.12.91, seems to work. - Linus.
15  *
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.
19  *
20  * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21  */
22 
23 /*
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.
29  */
30 
31 /*
32  * 05.04.94  -  Multi-page memory management added for v1.1.
33  * 		Idea by Alex Bligh (alex@cconcepts.co.uk)
34  *
35  * 16.07.99  -  Support of BIGMEM added by Gerhard Wichert, Siemens AG
36  *		(Gerhard.Wichert@pdb.siemens.de)
37  *
38  * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39  */
40 
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/delayacct.h>
51 #include <linux/init.h>
52 #include <linux/writeback.h>
53 
54 #include <asm/pgalloc.h>
55 #include <asm/uaccess.h>
56 #include <asm/tlb.h>
57 #include <asm/tlbflush.h>
58 #include <asm/pgtable.h>
59 
60 #include <linux/swapops.h>
61 #include <linux/elf.h>
62 
63 #ifndef CONFIG_NEED_MULTIPLE_NODES
64 /* use the per-pgdat data instead for discontigmem - mbligh */
65 unsigned long max_mapnr;
66 struct page *mem_map;
67 
68 EXPORT_SYMBOL(max_mapnr);
69 EXPORT_SYMBOL(mem_map);
70 #endif
71 
72 unsigned long num_physpages;
73 /*
74  * A number of key systems in x86 including ioremap() rely on the assumption
75  * that high_memory defines the upper bound on direct map memory, then end
76  * of ZONE_NORMAL.  Under CONFIG_DISCONTIG this means that max_low_pfn and
77  * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
78  * and ZONE_HIGHMEM.
79  */
80 void * high_memory;
81 unsigned long vmalloc_earlyreserve;
82 
83 EXPORT_SYMBOL(num_physpages);
84 EXPORT_SYMBOL(high_memory);
85 EXPORT_SYMBOL(vmalloc_earlyreserve);
86 
87 int randomize_va_space __read_mostly = 1;
88 
89 static int __init disable_randmaps(char *s)
90 {
91 	randomize_va_space = 0;
92 	return 1;
93 }
94 __setup("norandmaps", disable_randmaps);
95 
96 
97 /*
98  * If a p?d_bad entry is found while walking page tables, report
99  * the error, before resetting entry to p?d_none.  Usually (but
100  * very seldom) called out from the p?d_none_or_clear_bad macros.
101  */
102 
103 void pgd_clear_bad(pgd_t *pgd)
104 {
105 	pgd_ERROR(*pgd);
106 	pgd_clear(pgd);
107 }
108 
109 void pud_clear_bad(pud_t *pud)
110 {
111 	pud_ERROR(*pud);
112 	pud_clear(pud);
113 }
114 
115 void pmd_clear_bad(pmd_t *pmd)
116 {
117 	pmd_ERROR(*pmd);
118 	pmd_clear(pmd);
119 }
120 
121 /*
122  * Note: this doesn't free the actual pages themselves. That
123  * has been handled earlier when unmapping all the memory regions.
124  */
125 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
126 {
127 	struct page *page = pmd_page(*pmd);
128 	pmd_clear(pmd);
129 	pte_lock_deinit(page);
130 	pte_free_tlb(tlb, page);
131 	dec_zone_page_state(page, NR_PAGETABLE);
132 	tlb->mm->nr_ptes--;
133 }
134 
135 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
136 				unsigned long addr, unsigned long end,
137 				unsigned long floor, unsigned long ceiling)
138 {
139 	pmd_t *pmd;
140 	unsigned long next;
141 	unsigned long start;
142 
143 	start = addr;
144 	pmd = pmd_offset(pud, addr);
145 	do {
146 		next = pmd_addr_end(addr, end);
147 		if (pmd_none_or_clear_bad(pmd))
148 			continue;
149 		free_pte_range(tlb, pmd);
150 	} while (pmd++, addr = next, addr != end);
151 
152 	start &= PUD_MASK;
153 	if (start < floor)
154 		return;
155 	if (ceiling) {
156 		ceiling &= PUD_MASK;
157 		if (!ceiling)
158 			return;
159 	}
160 	if (end - 1 > ceiling - 1)
161 		return;
162 
163 	pmd = pmd_offset(pud, start);
164 	pud_clear(pud);
165 	pmd_free_tlb(tlb, pmd);
166 }
167 
168 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
169 				unsigned long addr, unsigned long end,
170 				unsigned long floor, unsigned long ceiling)
171 {
172 	pud_t *pud;
173 	unsigned long next;
174 	unsigned long start;
175 
176 	start = addr;
177 	pud = pud_offset(pgd, addr);
178 	do {
179 		next = pud_addr_end(addr, end);
180 		if (pud_none_or_clear_bad(pud))
181 			continue;
182 		free_pmd_range(tlb, pud, addr, next, floor, ceiling);
183 	} while (pud++, addr = next, addr != end);
184 
185 	start &= PGDIR_MASK;
186 	if (start < floor)
187 		return;
188 	if (ceiling) {
189 		ceiling &= PGDIR_MASK;
190 		if (!ceiling)
191 			return;
192 	}
193 	if (end - 1 > ceiling - 1)
194 		return;
195 
196 	pud = pud_offset(pgd, start);
197 	pgd_clear(pgd);
198 	pud_free_tlb(tlb, pud);
199 }
200 
201 /*
202  * This function frees user-level page tables of a process.
203  *
204  * Must be called with pagetable lock held.
205  */
206 void free_pgd_range(struct mmu_gather **tlb,
207 			unsigned long addr, unsigned long end,
208 			unsigned long floor, unsigned long ceiling)
209 {
210 	pgd_t *pgd;
211 	unsigned long next;
212 	unsigned long start;
213 
214 	/*
215 	 * The next few lines have given us lots of grief...
216 	 *
217 	 * Why are we testing PMD* at this top level?  Because often
218 	 * there will be no work to do at all, and we'd prefer not to
219 	 * go all the way down to the bottom just to discover that.
220 	 *
221 	 * Why all these "- 1"s?  Because 0 represents both the bottom
222 	 * of the address space and the top of it (using -1 for the
223 	 * top wouldn't help much: the masks would do the wrong thing).
224 	 * The rule is that addr 0 and floor 0 refer to the bottom of
225 	 * the address space, but end 0 and ceiling 0 refer to the top
226 	 * Comparisons need to use "end - 1" and "ceiling - 1" (though
227 	 * that end 0 case should be mythical).
228 	 *
229 	 * Wherever addr is brought up or ceiling brought down, we must
230 	 * be careful to reject "the opposite 0" before it confuses the
231 	 * subsequent tests.  But what about where end is brought down
232 	 * by PMD_SIZE below? no, end can't go down to 0 there.
233 	 *
234 	 * Whereas we round start (addr) and ceiling down, by different
235 	 * masks at different levels, in order to test whether a table
236 	 * now has no other vmas using it, so can be freed, we don't
237 	 * bother to round floor or end up - the tests don't need that.
238 	 */
239 
240 	addr &= PMD_MASK;
241 	if (addr < floor) {
242 		addr += PMD_SIZE;
243 		if (!addr)
244 			return;
245 	}
246 	if (ceiling) {
247 		ceiling &= PMD_MASK;
248 		if (!ceiling)
249 			return;
250 	}
251 	if (end - 1 > ceiling - 1)
252 		end -= PMD_SIZE;
253 	if (addr > end - 1)
254 		return;
255 
256 	start = addr;
257 	pgd = pgd_offset((*tlb)->mm, addr);
258 	do {
259 		next = pgd_addr_end(addr, end);
260 		if (pgd_none_or_clear_bad(pgd))
261 			continue;
262 		free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
263 	} while (pgd++, addr = next, addr != end);
264 
265 	if (!(*tlb)->fullmm)
266 		flush_tlb_pgtables((*tlb)->mm, start, end);
267 }
268 
269 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
270 		unsigned long floor, unsigned long ceiling)
271 {
272 	while (vma) {
273 		struct vm_area_struct *next = vma->vm_next;
274 		unsigned long addr = vma->vm_start;
275 
276 		/*
277 		 * Hide vma from rmap and vmtruncate before freeing pgtables
278 		 */
279 		anon_vma_unlink(vma);
280 		unlink_file_vma(vma);
281 
282 		if (is_vm_hugetlb_page(vma)) {
283 			hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
284 				floor, next? next->vm_start: ceiling);
285 		} else {
286 			/*
287 			 * Optimization: gather nearby vmas into one call down
288 			 */
289 			while (next && next->vm_start <= vma->vm_end + PMD_SIZE
290 			       && !is_vm_hugetlb_page(next)) {
291 				vma = next;
292 				next = vma->vm_next;
293 				anon_vma_unlink(vma);
294 				unlink_file_vma(vma);
295 			}
296 			free_pgd_range(tlb, addr, vma->vm_end,
297 				floor, next? next->vm_start: ceiling);
298 		}
299 		vma = next;
300 	}
301 }
302 
303 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
304 {
305 	struct page *new = pte_alloc_one(mm, address);
306 	if (!new)
307 		return -ENOMEM;
308 
309 	pte_lock_init(new);
310 	spin_lock(&mm->page_table_lock);
311 	if (pmd_present(*pmd)) {	/* Another has populated it */
312 		pte_lock_deinit(new);
313 		pte_free(new);
314 	} else {
315 		mm->nr_ptes++;
316 		inc_zone_page_state(new, NR_PAGETABLE);
317 		pmd_populate(mm, pmd, new);
318 	}
319 	spin_unlock(&mm->page_table_lock);
320 	return 0;
321 }
322 
323 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
324 {
325 	pte_t *new = pte_alloc_one_kernel(&init_mm, address);
326 	if (!new)
327 		return -ENOMEM;
328 
329 	spin_lock(&init_mm.page_table_lock);
330 	if (pmd_present(*pmd))		/* Another has populated it */
331 		pte_free_kernel(new);
332 	else
333 		pmd_populate_kernel(&init_mm, pmd, new);
334 	spin_unlock(&init_mm.page_table_lock);
335 	return 0;
336 }
337 
338 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
339 {
340 	if (file_rss)
341 		add_mm_counter(mm, file_rss, file_rss);
342 	if (anon_rss)
343 		add_mm_counter(mm, anon_rss, anon_rss);
344 }
345 
346 /*
347  * This function is called to print an error when a bad pte
348  * is found. For example, we might have a PFN-mapped pte in
349  * a region that doesn't allow it.
350  *
351  * The calling function must still handle the error.
352  */
353 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
354 {
355 	printk(KERN_ERR "Bad pte = %08llx, process = %s, "
356 			"vm_flags = %lx, vaddr = %lx\n",
357 		(long long)pte_val(pte),
358 		(vma->vm_mm == current->mm ? current->comm : "???"),
359 		vma->vm_flags, vaddr);
360 	dump_stack();
361 }
362 
363 static inline int is_cow_mapping(unsigned int flags)
364 {
365 	return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
366 }
367 
368 /*
369  * This function gets the "struct page" associated with a pte.
370  *
371  * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
372  * will have each page table entry just pointing to a raw page frame
373  * number, and as far as the VM layer is concerned, those do not have
374  * pages associated with them - even if the PFN might point to memory
375  * that otherwise is perfectly fine and has a "struct page".
376  *
377  * The way we recognize those mappings is through the rules set up
378  * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
379  * and the vm_pgoff will point to the first PFN mapped: thus every
380  * page that is a raw mapping will always honor the rule
381  *
382  *	pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
383  *
384  * and if that isn't true, the page has been COW'ed (in which case it
385  * _does_ have a "struct page" associated with it even if it is in a
386  * VM_PFNMAP range).
387  */
388 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
389 {
390 	unsigned long pfn = pte_pfn(pte);
391 
392 	if (unlikely(vma->vm_flags & VM_PFNMAP)) {
393 		unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
394 		if (pfn == vma->vm_pgoff + off)
395 			return NULL;
396 		if (!is_cow_mapping(vma->vm_flags))
397 			return NULL;
398 	}
399 
400 	/*
401 	 * Add some anal sanity checks for now. Eventually,
402 	 * we should just do "return pfn_to_page(pfn)", but
403 	 * in the meantime we check that we get a valid pfn,
404 	 * and that the resulting page looks ok.
405 	 */
406 	if (unlikely(!pfn_valid(pfn))) {
407 		print_bad_pte(vma, pte, addr);
408 		return NULL;
409 	}
410 
411 	/*
412 	 * NOTE! We still have PageReserved() pages in the page
413 	 * tables.
414 	 *
415 	 * The PAGE_ZERO() pages and various VDSO mappings can
416 	 * cause them to exist.
417 	 */
418 	return pfn_to_page(pfn);
419 }
420 
421 /*
422  * copy one vm_area from one task to the other. Assumes the page tables
423  * already present in the new task to be cleared in the whole range
424  * covered by this vma.
425  */
426 
427 static inline void
428 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
429 		pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
430 		unsigned long addr, int *rss)
431 {
432 	unsigned long vm_flags = vma->vm_flags;
433 	pte_t pte = *src_pte;
434 	struct page *page;
435 
436 	/* pte contains position in swap or file, so copy. */
437 	if (unlikely(!pte_present(pte))) {
438 		if (!pte_file(pte)) {
439 			swp_entry_t entry = pte_to_swp_entry(pte);
440 
441 			swap_duplicate(entry);
442 			/* make sure dst_mm is on swapoff's mmlist. */
443 			if (unlikely(list_empty(&dst_mm->mmlist))) {
444 				spin_lock(&mmlist_lock);
445 				if (list_empty(&dst_mm->mmlist))
446 					list_add(&dst_mm->mmlist,
447 						 &src_mm->mmlist);
448 				spin_unlock(&mmlist_lock);
449 			}
450 			if (is_write_migration_entry(entry) &&
451 					is_cow_mapping(vm_flags)) {
452 				/*
453 				 * COW mappings require pages in both parent
454 				 * and child to be set to read.
455 				 */
456 				make_migration_entry_read(&entry);
457 				pte = swp_entry_to_pte(entry);
458 				set_pte_at(src_mm, addr, src_pte, pte);
459 			}
460 		}
461 		goto out_set_pte;
462 	}
463 
464 	/*
465 	 * If it's a COW mapping, write protect it both
466 	 * in the parent and the child
467 	 */
468 	if (is_cow_mapping(vm_flags)) {
469 		ptep_set_wrprotect(src_mm, addr, src_pte);
470 		pte = pte_wrprotect(pte);
471 	}
472 
473 	/*
474 	 * If it's a shared mapping, mark it clean in
475 	 * the child
476 	 */
477 	if (vm_flags & VM_SHARED)
478 		pte = pte_mkclean(pte);
479 	pte = pte_mkold(pte);
480 
481 	page = vm_normal_page(vma, addr, pte);
482 	if (page) {
483 		get_page(page);
484 		page_dup_rmap(page);
485 		rss[!!PageAnon(page)]++;
486 	}
487 
488 out_set_pte:
489 	set_pte_at(dst_mm, addr, dst_pte, pte);
490 }
491 
492 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
493 		pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
494 		unsigned long addr, unsigned long end)
495 {
496 	pte_t *src_pte, *dst_pte;
497 	spinlock_t *src_ptl, *dst_ptl;
498 	int progress = 0;
499 	int rss[2];
500 
501 again:
502 	rss[1] = rss[0] = 0;
503 	dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
504 	if (!dst_pte)
505 		return -ENOMEM;
506 	src_pte = pte_offset_map_nested(src_pmd, addr);
507 	src_ptl = pte_lockptr(src_mm, src_pmd);
508 	spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
509 	arch_enter_lazy_mmu_mode();
510 
511 	do {
512 		/*
513 		 * We are holding two locks at this point - either of them
514 		 * could generate latencies in another task on another CPU.
515 		 */
516 		if (progress >= 32) {
517 			progress = 0;
518 			if (need_resched() ||
519 			    need_lockbreak(src_ptl) ||
520 			    need_lockbreak(dst_ptl))
521 				break;
522 		}
523 		if (pte_none(*src_pte)) {
524 			progress++;
525 			continue;
526 		}
527 		copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
528 		progress += 8;
529 	} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
530 
531 	arch_leave_lazy_mmu_mode();
532 	spin_unlock(src_ptl);
533 	pte_unmap_nested(src_pte - 1);
534 	add_mm_rss(dst_mm, rss[0], rss[1]);
535 	pte_unmap_unlock(dst_pte - 1, dst_ptl);
536 	cond_resched();
537 	if (addr != end)
538 		goto again;
539 	return 0;
540 }
541 
542 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
543 		pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
544 		unsigned long addr, unsigned long end)
545 {
546 	pmd_t *src_pmd, *dst_pmd;
547 	unsigned long next;
548 
549 	dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
550 	if (!dst_pmd)
551 		return -ENOMEM;
552 	src_pmd = pmd_offset(src_pud, addr);
553 	do {
554 		next = pmd_addr_end(addr, end);
555 		if (pmd_none_or_clear_bad(src_pmd))
556 			continue;
557 		if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
558 						vma, addr, next))
559 			return -ENOMEM;
560 	} while (dst_pmd++, src_pmd++, addr = next, addr != end);
561 	return 0;
562 }
563 
564 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
565 		pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
566 		unsigned long addr, unsigned long end)
567 {
568 	pud_t *src_pud, *dst_pud;
569 	unsigned long next;
570 
571 	dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
572 	if (!dst_pud)
573 		return -ENOMEM;
574 	src_pud = pud_offset(src_pgd, addr);
575 	do {
576 		next = pud_addr_end(addr, end);
577 		if (pud_none_or_clear_bad(src_pud))
578 			continue;
579 		if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
580 						vma, addr, next))
581 			return -ENOMEM;
582 	} while (dst_pud++, src_pud++, addr = next, addr != end);
583 	return 0;
584 }
585 
586 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
587 		struct vm_area_struct *vma)
588 {
589 	pgd_t *src_pgd, *dst_pgd;
590 	unsigned long next;
591 	unsigned long addr = vma->vm_start;
592 	unsigned long end = vma->vm_end;
593 
594 	/*
595 	 * Don't copy ptes where a page fault will fill them correctly.
596 	 * Fork becomes much lighter when there are big shared or private
597 	 * readonly mappings. The tradeoff is that copy_page_range is more
598 	 * efficient than faulting.
599 	 */
600 	if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
601 		if (!vma->anon_vma)
602 			return 0;
603 	}
604 
605 	if (is_vm_hugetlb_page(vma))
606 		return copy_hugetlb_page_range(dst_mm, src_mm, vma);
607 
608 	dst_pgd = pgd_offset(dst_mm, addr);
609 	src_pgd = pgd_offset(src_mm, addr);
610 	do {
611 		next = pgd_addr_end(addr, end);
612 		if (pgd_none_or_clear_bad(src_pgd))
613 			continue;
614 		if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
615 						vma, addr, next))
616 			return -ENOMEM;
617 	} while (dst_pgd++, src_pgd++, addr = next, addr != end);
618 	return 0;
619 }
620 
621 static unsigned long zap_pte_range(struct mmu_gather *tlb,
622 				struct vm_area_struct *vma, pmd_t *pmd,
623 				unsigned long addr, unsigned long end,
624 				long *zap_work, struct zap_details *details)
625 {
626 	struct mm_struct *mm = tlb->mm;
627 	pte_t *pte;
628 	spinlock_t *ptl;
629 	int file_rss = 0;
630 	int anon_rss = 0;
631 
632 	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
633 	arch_enter_lazy_mmu_mode();
634 	do {
635 		pte_t ptent = *pte;
636 		if (pte_none(ptent)) {
637 			(*zap_work)--;
638 			continue;
639 		}
640 
641 		(*zap_work) -= PAGE_SIZE;
642 
643 		if (pte_present(ptent)) {
644 			struct page *page;
645 
646 			page = vm_normal_page(vma, addr, ptent);
647 			if (unlikely(details) && page) {
648 				/*
649 				 * unmap_shared_mapping_pages() wants to
650 				 * invalidate cache without truncating:
651 				 * unmap shared but keep private pages.
652 				 */
653 				if (details->check_mapping &&
654 				    details->check_mapping != page->mapping)
655 					continue;
656 				/*
657 				 * Each page->index must be checked when
658 				 * invalidating or truncating nonlinear.
659 				 */
660 				if (details->nonlinear_vma &&
661 				    (page->index < details->first_index ||
662 				     page->index > details->last_index))
663 					continue;
664 			}
665 			ptent = ptep_get_and_clear_full(mm, addr, pte,
666 							tlb->fullmm);
667 			tlb_remove_tlb_entry(tlb, pte, addr);
668 			if (unlikely(!page))
669 				continue;
670 			if (unlikely(details) && details->nonlinear_vma
671 			    && linear_page_index(details->nonlinear_vma,
672 						addr) != page->index)
673 				set_pte_at(mm, addr, pte,
674 					   pgoff_to_pte(page->index));
675 			if (PageAnon(page))
676 				anon_rss--;
677 			else {
678 				if (pte_dirty(ptent))
679 					set_page_dirty(page);
680 				if (pte_young(ptent))
681 					SetPageReferenced(page);
682 				file_rss--;
683 			}
684 			page_remove_rmap(page, vma);
685 			tlb_remove_page(tlb, page);
686 			continue;
687 		}
688 		/*
689 		 * If details->check_mapping, we leave swap entries;
690 		 * if details->nonlinear_vma, we leave file entries.
691 		 */
692 		if (unlikely(details))
693 			continue;
694 		if (!pte_file(ptent))
695 			free_swap_and_cache(pte_to_swp_entry(ptent));
696 		pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
697 	} while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
698 
699 	add_mm_rss(mm, file_rss, anon_rss);
700 	arch_leave_lazy_mmu_mode();
701 	pte_unmap_unlock(pte - 1, ptl);
702 
703 	return addr;
704 }
705 
706 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
707 				struct vm_area_struct *vma, pud_t *pud,
708 				unsigned long addr, unsigned long end,
709 				long *zap_work, struct zap_details *details)
710 {
711 	pmd_t *pmd;
712 	unsigned long next;
713 
714 	pmd = pmd_offset(pud, addr);
715 	do {
716 		next = pmd_addr_end(addr, end);
717 		if (pmd_none_or_clear_bad(pmd)) {
718 			(*zap_work)--;
719 			continue;
720 		}
721 		next = zap_pte_range(tlb, vma, pmd, addr, next,
722 						zap_work, details);
723 	} while (pmd++, addr = next, (addr != end && *zap_work > 0));
724 
725 	return addr;
726 }
727 
728 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
729 				struct vm_area_struct *vma, pgd_t *pgd,
730 				unsigned long addr, unsigned long end,
731 				long *zap_work, struct zap_details *details)
732 {
733 	pud_t *pud;
734 	unsigned long next;
735 
736 	pud = pud_offset(pgd, addr);
737 	do {
738 		next = pud_addr_end(addr, end);
739 		if (pud_none_or_clear_bad(pud)) {
740 			(*zap_work)--;
741 			continue;
742 		}
743 		next = zap_pmd_range(tlb, vma, pud, addr, next,
744 						zap_work, details);
745 	} while (pud++, addr = next, (addr != end && *zap_work > 0));
746 
747 	return addr;
748 }
749 
750 static unsigned long unmap_page_range(struct mmu_gather *tlb,
751 				struct vm_area_struct *vma,
752 				unsigned long addr, unsigned long end,
753 				long *zap_work, struct zap_details *details)
754 {
755 	pgd_t *pgd;
756 	unsigned long next;
757 
758 	if (details && !details->check_mapping && !details->nonlinear_vma)
759 		details = NULL;
760 
761 	BUG_ON(addr >= end);
762 	tlb_start_vma(tlb, vma);
763 	pgd = pgd_offset(vma->vm_mm, addr);
764 	do {
765 		next = pgd_addr_end(addr, end);
766 		if (pgd_none_or_clear_bad(pgd)) {
767 			(*zap_work)--;
768 			continue;
769 		}
770 		next = zap_pud_range(tlb, vma, pgd, addr, next,
771 						zap_work, details);
772 	} while (pgd++, addr = next, (addr != end && *zap_work > 0));
773 	tlb_end_vma(tlb, vma);
774 
775 	return addr;
776 }
777 
778 #ifdef CONFIG_PREEMPT
779 # define ZAP_BLOCK_SIZE	(8 * PAGE_SIZE)
780 #else
781 /* No preempt: go for improved straight-line efficiency */
782 # define ZAP_BLOCK_SIZE	(1024 * PAGE_SIZE)
783 #endif
784 
785 /**
786  * unmap_vmas - unmap a range of memory covered by a list of vma's
787  * @tlbp: address of the caller's struct mmu_gather
788  * @vma: the starting vma
789  * @start_addr: virtual address at which to start unmapping
790  * @end_addr: virtual address at which to end unmapping
791  * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
792  * @details: details of nonlinear truncation or shared cache invalidation
793  *
794  * Returns the end address of the unmapping (restart addr if interrupted).
795  *
796  * Unmap all pages in the vma list.
797  *
798  * We aim to not hold locks for too long (for scheduling latency reasons).
799  * So zap pages in ZAP_BLOCK_SIZE bytecounts.  This means we need to
800  * return the ending mmu_gather to the caller.
801  *
802  * Only addresses between `start' and `end' will be unmapped.
803  *
804  * The VMA list must be sorted in ascending virtual address order.
805  *
806  * unmap_vmas() assumes that the caller will flush the whole unmapped address
807  * range after unmap_vmas() returns.  So the only responsibility here is to
808  * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
809  * drops the lock and schedules.
810  */
811 unsigned long unmap_vmas(struct mmu_gather **tlbp,
812 		struct vm_area_struct *vma, unsigned long start_addr,
813 		unsigned long end_addr, unsigned long *nr_accounted,
814 		struct zap_details *details)
815 {
816 	long zap_work = ZAP_BLOCK_SIZE;
817 	unsigned long tlb_start = 0;	/* For tlb_finish_mmu */
818 	int tlb_start_valid = 0;
819 	unsigned long start = start_addr;
820 	spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
821 	int fullmm = (*tlbp)->fullmm;
822 
823 	for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
824 		unsigned long end;
825 
826 		start = max(vma->vm_start, start_addr);
827 		if (start >= vma->vm_end)
828 			continue;
829 		end = min(vma->vm_end, end_addr);
830 		if (end <= vma->vm_start)
831 			continue;
832 
833 		if (vma->vm_flags & VM_ACCOUNT)
834 			*nr_accounted += (end - start) >> PAGE_SHIFT;
835 
836 		while (start != end) {
837 			if (!tlb_start_valid) {
838 				tlb_start = start;
839 				tlb_start_valid = 1;
840 			}
841 
842 			if (unlikely(is_vm_hugetlb_page(vma))) {
843 				unmap_hugepage_range(vma, start, end);
844 				zap_work -= (end - start) /
845 						(HPAGE_SIZE / PAGE_SIZE);
846 				start = end;
847 			} else
848 				start = unmap_page_range(*tlbp, vma,
849 						start, end, &zap_work, details);
850 
851 			if (zap_work > 0) {
852 				BUG_ON(start != end);
853 				break;
854 			}
855 
856 			tlb_finish_mmu(*tlbp, tlb_start, start);
857 
858 			if (need_resched() ||
859 				(i_mmap_lock && need_lockbreak(i_mmap_lock))) {
860 				if (i_mmap_lock) {
861 					*tlbp = NULL;
862 					goto out;
863 				}
864 				cond_resched();
865 			}
866 
867 			*tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
868 			tlb_start_valid = 0;
869 			zap_work = ZAP_BLOCK_SIZE;
870 		}
871 	}
872 out:
873 	return start;	/* which is now the end (or restart) address */
874 }
875 
876 /**
877  * zap_page_range - remove user pages in a given range
878  * @vma: vm_area_struct holding the applicable pages
879  * @address: starting address of pages to zap
880  * @size: number of bytes to zap
881  * @details: details of nonlinear truncation or shared cache invalidation
882  */
883 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
884 		unsigned long size, struct zap_details *details)
885 {
886 	struct mm_struct *mm = vma->vm_mm;
887 	struct mmu_gather *tlb;
888 	unsigned long end = address + size;
889 	unsigned long nr_accounted = 0;
890 
891 	lru_add_drain();
892 	tlb = tlb_gather_mmu(mm, 0);
893 	update_hiwater_rss(mm);
894 	end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
895 	if (tlb)
896 		tlb_finish_mmu(tlb, address, end);
897 	return end;
898 }
899 
900 /*
901  * Do a quick page-table lookup for a single page.
902  */
903 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
904 			unsigned int flags)
905 {
906 	pgd_t *pgd;
907 	pud_t *pud;
908 	pmd_t *pmd;
909 	pte_t *ptep, pte;
910 	spinlock_t *ptl;
911 	struct page *page;
912 	struct mm_struct *mm = vma->vm_mm;
913 
914 	page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
915 	if (!IS_ERR(page)) {
916 		BUG_ON(flags & FOLL_GET);
917 		goto out;
918 	}
919 
920 	page = NULL;
921 	pgd = pgd_offset(mm, address);
922 	if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
923 		goto no_page_table;
924 
925 	pud = pud_offset(pgd, address);
926 	if (pud_none(*pud) || unlikely(pud_bad(*pud)))
927 		goto no_page_table;
928 
929 	pmd = pmd_offset(pud, address);
930 	if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
931 		goto no_page_table;
932 
933 	if (pmd_huge(*pmd)) {
934 		BUG_ON(flags & FOLL_GET);
935 		page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
936 		goto out;
937 	}
938 
939 	ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
940 	if (!ptep)
941 		goto out;
942 
943 	pte = *ptep;
944 	if (!pte_present(pte))
945 		goto unlock;
946 	if ((flags & FOLL_WRITE) && !pte_write(pte))
947 		goto unlock;
948 	page = vm_normal_page(vma, address, pte);
949 	if (unlikely(!page))
950 		goto unlock;
951 
952 	if (flags & FOLL_GET)
953 		get_page(page);
954 	if (flags & FOLL_TOUCH) {
955 		if ((flags & FOLL_WRITE) &&
956 		    !pte_dirty(pte) && !PageDirty(page))
957 			set_page_dirty(page);
958 		mark_page_accessed(page);
959 	}
960 unlock:
961 	pte_unmap_unlock(ptep, ptl);
962 out:
963 	return page;
964 
965 no_page_table:
966 	/*
967 	 * When core dumping an enormous anonymous area that nobody
968 	 * has touched so far, we don't want to allocate page tables.
969 	 */
970 	if (flags & FOLL_ANON) {
971 		page = ZERO_PAGE(address);
972 		if (flags & FOLL_GET)
973 			get_page(page);
974 		BUG_ON(flags & FOLL_WRITE);
975 	}
976 	return page;
977 }
978 
979 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
980 		unsigned long start, int len, int write, int force,
981 		struct page **pages, struct vm_area_struct **vmas)
982 {
983 	int i;
984 	unsigned int vm_flags;
985 
986 	/*
987 	 * Require read or write permissions.
988 	 * If 'force' is set, we only require the "MAY" flags.
989 	 */
990 	vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
991 	vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
992 	i = 0;
993 
994 	do {
995 		struct vm_area_struct *vma;
996 		unsigned int foll_flags;
997 
998 		vma = find_extend_vma(mm, start);
999 		if (!vma && in_gate_area(tsk, start)) {
1000 			unsigned long pg = start & PAGE_MASK;
1001 			struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1002 			pgd_t *pgd;
1003 			pud_t *pud;
1004 			pmd_t *pmd;
1005 			pte_t *pte;
1006 			if (write) /* user gate pages are read-only */
1007 				return i ? : -EFAULT;
1008 			if (pg > TASK_SIZE)
1009 				pgd = pgd_offset_k(pg);
1010 			else
1011 				pgd = pgd_offset_gate(mm, pg);
1012 			BUG_ON(pgd_none(*pgd));
1013 			pud = pud_offset(pgd, pg);
1014 			BUG_ON(pud_none(*pud));
1015 			pmd = pmd_offset(pud, pg);
1016 			if (pmd_none(*pmd))
1017 				return i ? : -EFAULT;
1018 			pte = pte_offset_map(pmd, pg);
1019 			if (pte_none(*pte)) {
1020 				pte_unmap(pte);
1021 				return i ? : -EFAULT;
1022 			}
1023 			if (pages) {
1024 				struct page *page = vm_normal_page(gate_vma, start, *pte);
1025 				pages[i] = page;
1026 				if (page)
1027 					get_page(page);
1028 			}
1029 			pte_unmap(pte);
1030 			if (vmas)
1031 				vmas[i] = gate_vma;
1032 			i++;
1033 			start += PAGE_SIZE;
1034 			len--;
1035 			continue;
1036 		}
1037 
1038 		if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1039 				|| !(vm_flags & vma->vm_flags))
1040 			return i ? : -EFAULT;
1041 
1042 		if (is_vm_hugetlb_page(vma)) {
1043 			i = follow_hugetlb_page(mm, vma, pages, vmas,
1044 						&start, &len, i);
1045 			continue;
1046 		}
1047 
1048 		foll_flags = FOLL_TOUCH;
1049 		if (pages)
1050 			foll_flags |= FOLL_GET;
1051 		if (!write && !(vma->vm_flags & VM_LOCKED) &&
1052 		    (!vma->vm_ops || !vma->vm_ops->nopage))
1053 			foll_flags |= FOLL_ANON;
1054 
1055 		do {
1056 			struct page *page;
1057 
1058 			if (write)
1059 				foll_flags |= FOLL_WRITE;
1060 
1061 			cond_resched();
1062 			while (!(page = follow_page(vma, start, foll_flags))) {
1063 				int ret;
1064 				ret = __handle_mm_fault(mm, vma, start,
1065 						foll_flags & FOLL_WRITE);
1066 				/*
1067 				 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1068 				 * broken COW when necessary, even if maybe_mkwrite
1069 				 * decided not to set pte_write. We can thus safely do
1070 				 * subsequent page lookups as if they were reads.
1071 				 */
1072 				if (ret & VM_FAULT_WRITE)
1073 					foll_flags &= ~FOLL_WRITE;
1074 
1075 				switch (ret & ~VM_FAULT_WRITE) {
1076 				case VM_FAULT_MINOR:
1077 					tsk->min_flt++;
1078 					break;
1079 				case VM_FAULT_MAJOR:
1080 					tsk->maj_flt++;
1081 					break;
1082 				case VM_FAULT_SIGBUS:
1083 					return i ? i : -EFAULT;
1084 				case VM_FAULT_OOM:
1085 					return i ? i : -ENOMEM;
1086 				default:
1087 					BUG();
1088 				}
1089 				cond_resched();
1090 			}
1091 			if (pages) {
1092 				pages[i] = page;
1093 
1094 				flush_anon_page(vma, page, start);
1095 				flush_dcache_page(page);
1096 			}
1097 			if (vmas)
1098 				vmas[i] = vma;
1099 			i++;
1100 			start += PAGE_SIZE;
1101 			len--;
1102 		} while (len && start < vma->vm_end);
1103 	} while (len);
1104 	return i;
1105 }
1106 EXPORT_SYMBOL(get_user_pages);
1107 
1108 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1109 			unsigned long addr, unsigned long end, pgprot_t prot)
1110 {
1111 	pte_t *pte;
1112 	spinlock_t *ptl;
1113 	int err = 0;
1114 
1115 	pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1116 	if (!pte)
1117 		return -EAGAIN;
1118 	arch_enter_lazy_mmu_mode();
1119 	do {
1120 		struct page *page = ZERO_PAGE(addr);
1121 		pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1122 
1123 		if (unlikely(!pte_none(*pte))) {
1124 			err = -EEXIST;
1125 			pte++;
1126 			break;
1127 		}
1128 		page_cache_get(page);
1129 		page_add_file_rmap(page);
1130 		inc_mm_counter(mm, file_rss);
1131 		set_pte_at(mm, addr, pte, zero_pte);
1132 	} while (pte++, addr += PAGE_SIZE, addr != end);
1133 	arch_leave_lazy_mmu_mode();
1134 	pte_unmap_unlock(pte - 1, ptl);
1135 	return err;
1136 }
1137 
1138 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1139 			unsigned long addr, unsigned long end, pgprot_t prot)
1140 {
1141 	pmd_t *pmd;
1142 	unsigned long next;
1143 	int err;
1144 
1145 	pmd = pmd_alloc(mm, pud, addr);
1146 	if (!pmd)
1147 		return -EAGAIN;
1148 	do {
1149 		next = pmd_addr_end(addr, end);
1150 		err = zeromap_pte_range(mm, pmd, addr, next, prot);
1151 		if (err)
1152 			break;
1153 	} while (pmd++, addr = next, addr != end);
1154 	return err;
1155 }
1156 
1157 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1158 			unsigned long addr, unsigned long end, pgprot_t prot)
1159 {
1160 	pud_t *pud;
1161 	unsigned long next;
1162 	int err;
1163 
1164 	pud = pud_alloc(mm, pgd, addr);
1165 	if (!pud)
1166 		return -EAGAIN;
1167 	do {
1168 		next = pud_addr_end(addr, end);
1169 		err = zeromap_pmd_range(mm, pud, addr, next, prot);
1170 		if (err)
1171 			break;
1172 	} while (pud++, addr = next, addr != end);
1173 	return err;
1174 }
1175 
1176 int zeromap_page_range(struct vm_area_struct *vma,
1177 			unsigned long addr, unsigned long size, pgprot_t prot)
1178 {
1179 	pgd_t *pgd;
1180 	unsigned long next;
1181 	unsigned long end = addr + size;
1182 	struct mm_struct *mm = vma->vm_mm;
1183 	int err;
1184 
1185 	BUG_ON(addr >= end);
1186 	pgd = pgd_offset(mm, addr);
1187 	flush_cache_range(vma, addr, end);
1188 	do {
1189 		next = pgd_addr_end(addr, end);
1190 		err = zeromap_pud_range(mm, pgd, addr, next, prot);
1191 		if (err)
1192 			break;
1193 	} while (pgd++, addr = next, addr != end);
1194 	return err;
1195 }
1196 
1197 pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
1198 {
1199 	pgd_t * pgd = pgd_offset(mm, addr);
1200 	pud_t * pud = pud_alloc(mm, pgd, addr);
1201 	if (pud) {
1202 		pmd_t * pmd = pmd_alloc(mm, pud, addr);
1203 		if (pmd)
1204 			return pte_alloc_map_lock(mm, pmd, addr, ptl);
1205 	}
1206 	return NULL;
1207 }
1208 
1209 /*
1210  * This is the old fallback for page remapping.
1211  *
1212  * For historical reasons, it only allows reserved pages. Only
1213  * old drivers should use this, and they needed to mark their
1214  * pages reserved for the old functions anyway.
1215  */
1216 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1217 {
1218 	int retval;
1219 	pte_t *pte;
1220 	spinlock_t *ptl;
1221 
1222 	retval = -EINVAL;
1223 	if (PageAnon(page))
1224 		goto out;
1225 	retval = -ENOMEM;
1226 	flush_dcache_page(page);
1227 	pte = get_locked_pte(mm, addr, &ptl);
1228 	if (!pte)
1229 		goto out;
1230 	retval = -EBUSY;
1231 	if (!pte_none(*pte))
1232 		goto out_unlock;
1233 
1234 	/* Ok, finally just insert the thing.. */
1235 	get_page(page);
1236 	inc_mm_counter(mm, file_rss);
1237 	page_add_file_rmap(page);
1238 	set_pte_at(mm, addr, pte, mk_pte(page, prot));
1239 
1240 	retval = 0;
1241 out_unlock:
1242 	pte_unmap_unlock(pte, ptl);
1243 out:
1244 	return retval;
1245 }
1246 
1247 /**
1248  * vm_insert_page - insert single page into user vma
1249  * @vma: user vma to map to
1250  * @addr: target user address of this page
1251  * @page: source kernel page
1252  *
1253  * This allows drivers to insert individual pages they've allocated
1254  * into a user vma.
1255  *
1256  * The page has to be a nice clean _individual_ kernel allocation.
1257  * If you allocate a compound page, you need to have marked it as
1258  * such (__GFP_COMP), or manually just split the page up yourself
1259  * (see split_page()).
1260  *
1261  * NOTE! Traditionally this was done with "remap_pfn_range()" which
1262  * took an arbitrary page protection parameter. This doesn't allow
1263  * that. Your vma protection will have to be set up correctly, which
1264  * means that if you want a shared writable mapping, you'd better
1265  * ask for a shared writable mapping!
1266  *
1267  * The page does not need to be reserved.
1268  */
1269 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1270 {
1271 	if (addr < vma->vm_start || addr >= vma->vm_end)
1272 		return -EFAULT;
1273 	if (!page_count(page))
1274 		return -EINVAL;
1275 	vma->vm_flags |= VM_INSERTPAGE;
1276 	return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1277 }
1278 EXPORT_SYMBOL(vm_insert_page);
1279 
1280 /**
1281  * vm_insert_pfn - insert single pfn into user vma
1282  * @vma: user vma to map to
1283  * @addr: target user address of this page
1284  * @pfn: source kernel pfn
1285  *
1286  * Similar to vm_inert_page, this allows drivers to insert individual pages
1287  * they've allocated into a user vma. Same comments apply.
1288  *
1289  * This function should only be called from a vm_ops->fault handler, and
1290  * in that case the handler should return NULL.
1291  */
1292 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1293 		unsigned long pfn)
1294 {
1295 	struct mm_struct *mm = vma->vm_mm;
1296 	int retval;
1297 	pte_t *pte, entry;
1298 	spinlock_t *ptl;
1299 
1300 	BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1301 	BUG_ON(is_cow_mapping(vma->vm_flags));
1302 
1303 	retval = -ENOMEM;
1304 	pte = get_locked_pte(mm, addr, &ptl);
1305 	if (!pte)
1306 		goto out;
1307 	retval = -EBUSY;
1308 	if (!pte_none(*pte))
1309 		goto out_unlock;
1310 
1311 	/* Ok, finally just insert the thing.. */
1312 	entry = pfn_pte(pfn, vma->vm_page_prot);
1313 	set_pte_at(mm, addr, pte, entry);
1314 	update_mmu_cache(vma, addr, entry);
1315 
1316 	retval = 0;
1317 out_unlock:
1318 	pte_unmap_unlock(pte, ptl);
1319 
1320 out:
1321 	return retval;
1322 }
1323 EXPORT_SYMBOL(vm_insert_pfn);
1324 
1325 /*
1326  * maps a range of physical memory into the requested pages. the old
1327  * mappings are removed. any references to nonexistent pages results
1328  * in null mappings (currently treated as "copy-on-access")
1329  */
1330 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1331 			unsigned long addr, unsigned long end,
1332 			unsigned long pfn, pgprot_t prot)
1333 {
1334 	pte_t *pte;
1335 	spinlock_t *ptl;
1336 
1337 	pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1338 	if (!pte)
1339 		return -ENOMEM;
1340 	arch_enter_lazy_mmu_mode();
1341 	do {
1342 		BUG_ON(!pte_none(*pte));
1343 		set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1344 		pfn++;
1345 	} while (pte++, addr += PAGE_SIZE, addr != end);
1346 	arch_leave_lazy_mmu_mode();
1347 	pte_unmap_unlock(pte - 1, ptl);
1348 	return 0;
1349 }
1350 
1351 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1352 			unsigned long addr, unsigned long end,
1353 			unsigned long pfn, pgprot_t prot)
1354 {
1355 	pmd_t *pmd;
1356 	unsigned long next;
1357 
1358 	pfn -= addr >> PAGE_SHIFT;
1359 	pmd = pmd_alloc(mm, pud, addr);
1360 	if (!pmd)
1361 		return -ENOMEM;
1362 	do {
1363 		next = pmd_addr_end(addr, end);
1364 		if (remap_pte_range(mm, pmd, addr, next,
1365 				pfn + (addr >> PAGE_SHIFT), prot))
1366 			return -ENOMEM;
1367 	} while (pmd++, addr = next, addr != end);
1368 	return 0;
1369 }
1370 
1371 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1372 			unsigned long addr, unsigned long end,
1373 			unsigned long pfn, pgprot_t prot)
1374 {
1375 	pud_t *pud;
1376 	unsigned long next;
1377 
1378 	pfn -= addr >> PAGE_SHIFT;
1379 	pud = pud_alloc(mm, pgd, addr);
1380 	if (!pud)
1381 		return -ENOMEM;
1382 	do {
1383 		next = pud_addr_end(addr, end);
1384 		if (remap_pmd_range(mm, pud, addr, next,
1385 				pfn + (addr >> PAGE_SHIFT), prot))
1386 			return -ENOMEM;
1387 	} while (pud++, addr = next, addr != end);
1388 	return 0;
1389 }
1390 
1391 /**
1392  * remap_pfn_range - remap kernel memory to userspace
1393  * @vma: user vma to map to
1394  * @addr: target user address to start at
1395  * @pfn: physical address of kernel memory
1396  * @size: size of map area
1397  * @prot: page protection flags for this mapping
1398  *
1399  *  Note: this is only safe if the mm semaphore is held when called.
1400  */
1401 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1402 		    unsigned long pfn, unsigned long size, pgprot_t prot)
1403 {
1404 	pgd_t *pgd;
1405 	unsigned long next;
1406 	unsigned long end = addr + PAGE_ALIGN(size);
1407 	struct mm_struct *mm = vma->vm_mm;
1408 	int err;
1409 
1410 	/*
1411 	 * Physically remapped pages are special. Tell the
1412 	 * rest of the world about it:
1413 	 *   VM_IO tells people not to look at these pages
1414 	 *	(accesses can have side effects).
1415 	 *   VM_RESERVED is specified all over the place, because
1416 	 *	in 2.4 it kept swapout's vma scan off this vma; but
1417 	 *	in 2.6 the LRU scan won't even find its pages, so this
1418 	 *	flag means no more than count its pages in reserved_vm,
1419 	 * 	and omit it from core dump, even when VM_IO turned off.
1420 	 *   VM_PFNMAP tells the core MM that the base pages are just
1421 	 *	raw PFN mappings, and do not have a "struct page" associated
1422 	 *	with them.
1423 	 *
1424 	 * There's a horrible special case to handle copy-on-write
1425 	 * behaviour that some programs depend on. We mark the "original"
1426 	 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1427 	 */
1428 	if (is_cow_mapping(vma->vm_flags)) {
1429 		if (addr != vma->vm_start || end != vma->vm_end)
1430 			return -EINVAL;
1431 		vma->vm_pgoff = pfn;
1432 	}
1433 
1434 	vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1435 
1436 	BUG_ON(addr >= end);
1437 	pfn -= addr >> PAGE_SHIFT;
1438 	pgd = pgd_offset(mm, addr);
1439 	flush_cache_range(vma, addr, end);
1440 	do {
1441 		next = pgd_addr_end(addr, end);
1442 		err = remap_pud_range(mm, pgd, addr, next,
1443 				pfn + (addr >> PAGE_SHIFT), prot);
1444 		if (err)
1445 			break;
1446 	} while (pgd++, addr = next, addr != end);
1447 	return err;
1448 }
1449 EXPORT_SYMBOL(remap_pfn_range);
1450 
1451 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1452 				     unsigned long addr, unsigned long end,
1453 				     pte_fn_t fn, void *data)
1454 {
1455 	pte_t *pte;
1456 	int err;
1457 	struct page *pmd_page;
1458 	spinlock_t *uninitialized_var(ptl);
1459 
1460 	pte = (mm == &init_mm) ?
1461 		pte_alloc_kernel(pmd, addr) :
1462 		pte_alloc_map_lock(mm, pmd, addr, &ptl);
1463 	if (!pte)
1464 		return -ENOMEM;
1465 
1466 	BUG_ON(pmd_huge(*pmd));
1467 
1468 	pmd_page = pmd_page(*pmd);
1469 
1470 	do {
1471 		err = fn(pte, pmd_page, addr, data);
1472 		if (err)
1473 			break;
1474 	} while (pte++, addr += PAGE_SIZE, addr != end);
1475 
1476 	if (mm != &init_mm)
1477 		pte_unmap_unlock(pte-1, ptl);
1478 	return err;
1479 }
1480 
1481 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1482 				     unsigned long addr, unsigned long end,
1483 				     pte_fn_t fn, void *data)
1484 {
1485 	pmd_t *pmd;
1486 	unsigned long next;
1487 	int err;
1488 
1489 	pmd = pmd_alloc(mm, pud, addr);
1490 	if (!pmd)
1491 		return -ENOMEM;
1492 	do {
1493 		next = pmd_addr_end(addr, end);
1494 		err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1495 		if (err)
1496 			break;
1497 	} while (pmd++, addr = next, addr != end);
1498 	return err;
1499 }
1500 
1501 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1502 				     unsigned long addr, unsigned long end,
1503 				     pte_fn_t fn, void *data)
1504 {
1505 	pud_t *pud;
1506 	unsigned long next;
1507 	int err;
1508 
1509 	pud = pud_alloc(mm, pgd, addr);
1510 	if (!pud)
1511 		return -ENOMEM;
1512 	do {
1513 		next = pud_addr_end(addr, end);
1514 		err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1515 		if (err)
1516 			break;
1517 	} while (pud++, addr = next, addr != end);
1518 	return err;
1519 }
1520 
1521 /*
1522  * Scan a region of virtual memory, filling in page tables as necessary
1523  * and calling a provided function on each leaf page table.
1524  */
1525 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1526 			unsigned long size, pte_fn_t fn, void *data)
1527 {
1528 	pgd_t *pgd;
1529 	unsigned long next;
1530 	unsigned long end = addr + size;
1531 	int err;
1532 
1533 	BUG_ON(addr >= end);
1534 	pgd = pgd_offset(mm, addr);
1535 	do {
1536 		next = pgd_addr_end(addr, end);
1537 		err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1538 		if (err)
1539 			break;
1540 	} while (pgd++, addr = next, addr != end);
1541 	return err;
1542 }
1543 EXPORT_SYMBOL_GPL(apply_to_page_range);
1544 
1545 /*
1546  * handle_pte_fault chooses page fault handler according to an entry
1547  * which was read non-atomically.  Before making any commitment, on
1548  * those architectures or configurations (e.g. i386 with PAE) which
1549  * might give a mix of unmatched parts, do_swap_page and do_file_page
1550  * must check under lock before unmapping the pte and proceeding
1551  * (but do_wp_page is only called after already making such a check;
1552  * and do_anonymous_page and do_no_page can safely check later on).
1553  */
1554 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1555 				pte_t *page_table, pte_t orig_pte)
1556 {
1557 	int same = 1;
1558 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1559 	if (sizeof(pte_t) > sizeof(unsigned long)) {
1560 		spinlock_t *ptl = pte_lockptr(mm, pmd);
1561 		spin_lock(ptl);
1562 		same = pte_same(*page_table, orig_pte);
1563 		spin_unlock(ptl);
1564 	}
1565 #endif
1566 	pte_unmap(page_table);
1567 	return same;
1568 }
1569 
1570 /*
1571  * Do pte_mkwrite, but only if the vma says VM_WRITE.  We do this when
1572  * servicing faults for write access.  In the normal case, do always want
1573  * pte_mkwrite.  But get_user_pages can cause write faults for mappings
1574  * that do not have writing enabled, when used by access_process_vm.
1575  */
1576 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1577 {
1578 	if (likely(vma->vm_flags & VM_WRITE))
1579 		pte = pte_mkwrite(pte);
1580 	return pte;
1581 }
1582 
1583 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1584 {
1585 	/*
1586 	 * If the source page was a PFN mapping, we don't have
1587 	 * a "struct page" for it. We do a best-effort copy by
1588 	 * just copying from the original user address. If that
1589 	 * fails, we just zero-fill it. Live with it.
1590 	 */
1591 	if (unlikely(!src)) {
1592 		void *kaddr = kmap_atomic(dst, KM_USER0);
1593 		void __user *uaddr = (void __user *)(va & PAGE_MASK);
1594 
1595 		/*
1596 		 * This really shouldn't fail, because the page is there
1597 		 * in the page tables. But it might just be unreadable,
1598 		 * in which case we just give up and fill the result with
1599 		 * zeroes.
1600 		 */
1601 		if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1602 			memset(kaddr, 0, PAGE_SIZE);
1603 		kunmap_atomic(kaddr, KM_USER0);
1604 		flush_dcache_page(dst);
1605 		return;
1606 
1607 	}
1608 	copy_user_highpage(dst, src, va, vma);
1609 }
1610 
1611 /*
1612  * This routine handles present pages, when users try to write
1613  * to a shared page. It is done by copying the page to a new address
1614  * and decrementing the shared-page counter for the old page.
1615  *
1616  * Note that this routine assumes that the protection checks have been
1617  * done by the caller (the low-level page fault routine in most cases).
1618  * Thus we can safely just mark it writable once we've done any necessary
1619  * COW.
1620  *
1621  * We also mark the page dirty at this point even though the page will
1622  * change only once the write actually happens. This avoids a few races,
1623  * and potentially makes it more efficient.
1624  *
1625  * We enter with non-exclusive mmap_sem (to exclude vma changes,
1626  * but allow concurrent faults), with pte both mapped and locked.
1627  * We return with mmap_sem still held, but pte unmapped and unlocked.
1628  */
1629 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1630 		unsigned long address, pte_t *page_table, pmd_t *pmd,
1631 		spinlock_t *ptl, pte_t orig_pte)
1632 {
1633 	struct page *old_page, *new_page;
1634 	pte_t entry;
1635 	int reuse = 0, ret = VM_FAULT_MINOR;
1636 	struct page *dirty_page = NULL;
1637 
1638 	old_page = vm_normal_page(vma, address, orig_pte);
1639 	if (!old_page)
1640 		goto gotten;
1641 
1642 	/*
1643 	 * Take out anonymous pages first, anonymous shared vmas are
1644 	 * not dirty accountable.
1645 	 */
1646 	if (PageAnon(old_page)) {
1647 		if (!TestSetPageLocked(old_page)) {
1648 			reuse = can_share_swap_page(old_page);
1649 			unlock_page(old_page);
1650 		}
1651 	} else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1652 					(VM_WRITE|VM_SHARED))) {
1653 		/*
1654 		 * Only catch write-faults on shared writable pages,
1655 		 * read-only shared pages can get COWed by
1656 		 * get_user_pages(.write=1, .force=1).
1657 		 */
1658 		if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1659 			/*
1660 			 * Notify the address space that the page is about to
1661 			 * become writable so that it can prohibit this or wait
1662 			 * for the page to get into an appropriate state.
1663 			 *
1664 			 * We do this without the lock held, so that it can
1665 			 * sleep if it needs to.
1666 			 */
1667 			page_cache_get(old_page);
1668 			pte_unmap_unlock(page_table, ptl);
1669 
1670 			if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1671 				goto unwritable_page;
1672 
1673 			/*
1674 			 * Since we dropped the lock we need to revalidate
1675 			 * the PTE as someone else may have changed it.  If
1676 			 * they did, we just return, as we can count on the
1677 			 * MMU to tell us if they didn't also make it writable.
1678 			 */
1679 			page_table = pte_offset_map_lock(mm, pmd, address,
1680 							 &ptl);
1681 			page_cache_release(old_page);
1682 			if (!pte_same(*page_table, orig_pte))
1683 				goto unlock;
1684 		}
1685 		dirty_page = old_page;
1686 		get_page(dirty_page);
1687 		reuse = 1;
1688 	}
1689 
1690 	if (reuse) {
1691 		flush_cache_page(vma, address, pte_pfn(orig_pte));
1692 		entry = pte_mkyoung(orig_pte);
1693 		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1694 		ptep_set_access_flags(vma, address, page_table, entry, 1);
1695 		update_mmu_cache(vma, address, entry);
1696 		lazy_mmu_prot_update(entry);
1697 		ret |= VM_FAULT_WRITE;
1698 		goto unlock;
1699 	}
1700 
1701 	/*
1702 	 * Ok, we need to copy. Oh, well..
1703 	 */
1704 	page_cache_get(old_page);
1705 gotten:
1706 	pte_unmap_unlock(page_table, ptl);
1707 
1708 	if (unlikely(anon_vma_prepare(vma)))
1709 		goto oom;
1710 	if (old_page == ZERO_PAGE(address)) {
1711 		new_page = alloc_zeroed_user_highpage(vma, address);
1712 		if (!new_page)
1713 			goto oom;
1714 	} else {
1715 		new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1716 		if (!new_page)
1717 			goto oom;
1718 		cow_user_page(new_page, old_page, address, vma);
1719 	}
1720 
1721 	/*
1722 	 * Re-check the pte - we dropped the lock
1723 	 */
1724 	page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1725 	if (likely(pte_same(*page_table, orig_pte))) {
1726 		if (old_page) {
1727 			page_remove_rmap(old_page, vma);
1728 			if (!PageAnon(old_page)) {
1729 				dec_mm_counter(mm, file_rss);
1730 				inc_mm_counter(mm, anon_rss);
1731 			}
1732 		} else
1733 			inc_mm_counter(mm, anon_rss);
1734 		flush_cache_page(vma, address, pte_pfn(orig_pte));
1735 		entry = mk_pte(new_page, vma->vm_page_prot);
1736 		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1737 		lazy_mmu_prot_update(entry);
1738 		/*
1739 		 * Clear the pte entry and flush it first, before updating the
1740 		 * pte with the new entry. This will avoid a race condition
1741 		 * seen in the presence of one thread doing SMC and another
1742 		 * thread doing COW.
1743 		 */
1744 		ptep_clear_flush(vma, address, page_table);
1745 		set_pte_at(mm, address, page_table, entry);
1746 		update_mmu_cache(vma, address, entry);
1747 		lru_cache_add_active(new_page);
1748 		page_add_new_anon_rmap(new_page, vma, address);
1749 
1750 		/* Free the old page.. */
1751 		new_page = old_page;
1752 		ret |= VM_FAULT_WRITE;
1753 	}
1754 	if (new_page)
1755 		page_cache_release(new_page);
1756 	if (old_page)
1757 		page_cache_release(old_page);
1758 unlock:
1759 	pte_unmap_unlock(page_table, ptl);
1760 	if (dirty_page) {
1761 		set_page_dirty_balance(dirty_page);
1762 		put_page(dirty_page);
1763 	}
1764 	return ret;
1765 oom:
1766 	if (old_page)
1767 		page_cache_release(old_page);
1768 	return VM_FAULT_OOM;
1769 
1770 unwritable_page:
1771 	page_cache_release(old_page);
1772 	return VM_FAULT_SIGBUS;
1773 }
1774 
1775 /*
1776  * Helper functions for unmap_mapping_range().
1777  *
1778  * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1779  *
1780  * We have to restart searching the prio_tree whenever we drop the lock,
1781  * since the iterator is only valid while the lock is held, and anyway
1782  * a later vma might be split and reinserted earlier while lock dropped.
1783  *
1784  * The list of nonlinear vmas could be handled more efficiently, using
1785  * a placeholder, but handle it in the same way until a need is shown.
1786  * It is important to search the prio_tree before nonlinear list: a vma
1787  * may become nonlinear and be shifted from prio_tree to nonlinear list
1788  * while the lock is dropped; but never shifted from list to prio_tree.
1789  *
1790  * In order to make forward progress despite restarting the search,
1791  * vm_truncate_count is used to mark a vma as now dealt with, so we can
1792  * quickly skip it next time around.  Since the prio_tree search only
1793  * shows us those vmas affected by unmapping the range in question, we
1794  * can't efficiently keep all vmas in step with mapping->truncate_count:
1795  * so instead reset them all whenever it wraps back to 0 (then go to 1).
1796  * mapping->truncate_count and vma->vm_truncate_count are protected by
1797  * i_mmap_lock.
1798  *
1799  * In order to make forward progress despite repeatedly restarting some
1800  * large vma, note the restart_addr from unmap_vmas when it breaks out:
1801  * and restart from that address when we reach that vma again.  It might
1802  * have been split or merged, shrunk or extended, but never shifted: so
1803  * restart_addr remains valid so long as it remains in the vma's range.
1804  * unmap_mapping_range forces truncate_count to leap over page-aligned
1805  * values so we can save vma's restart_addr in its truncate_count field.
1806  */
1807 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1808 
1809 static void reset_vma_truncate_counts(struct address_space *mapping)
1810 {
1811 	struct vm_area_struct *vma;
1812 	struct prio_tree_iter iter;
1813 
1814 	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1815 		vma->vm_truncate_count = 0;
1816 	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1817 		vma->vm_truncate_count = 0;
1818 }
1819 
1820 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1821 		unsigned long start_addr, unsigned long end_addr,
1822 		struct zap_details *details)
1823 {
1824 	unsigned long restart_addr;
1825 	int need_break;
1826 
1827 again:
1828 	restart_addr = vma->vm_truncate_count;
1829 	if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1830 		start_addr = restart_addr;
1831 		if (start_addr >= end_addr) {
1832 			/* Top of vma has been split off since last time */
1833 			vma->vm_truncate_count = details->truncate_count;
1834 			return 0;
1835 		}
1836 	}
1837 
1838 	restart_addr = zap_page_range(vma, start_addr,
1839 					end_addr - start_addr, details);
1840 	need_break = need_resched() ||
1841 			need_lockbreak(details->i_mmap_lock);
1842 
1843 	if (restart_addr >= end_addr) {
1844 		/* We have now completed this vma: mark it so */
1845 		vma->vm_truncate_count = details->truncate_count;
1846 		if (!need_break)
1847 			return 0;
1848 	} else {
1849 		/* Note restart_addr in vma's truncate_count field */
1850 		vma->vm_truncate_count = restart_addr;
1851 		if (!need_break)
1852 			goto again;
1853 	}
1854 
1855 	spin_unlock(details->i_mmap_lock);
1856 	cond_resched();
1857 	spin_lock(details->i_mmap_lock);
1858 	return -EINTR;
1859 }
1860 
1861 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1862 					    struct zap_details *details)
1863 {
1864 	struct vm_area_struct *vma;
1865 	struct prio_tree_iter iter;
1866 	pgoff_t vba, vea, zba, zea;
1867 
1868 restart:
1869 	vma_prio_tree_foreach(vma, &iter, root,
1870 			details->first_index, details->last_index) {
1871 		/* Skip quickly over those we have already dealt with */
1872 		if (vma->vm_truncate_count == details->truncate_count)
1873 			continue;
1874 
1875 		vba = vma->vm_pgoff;
1876 		vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1877 		/* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1878 		zba = details->first_index;
1879 		if (zba < vba)
1880 			zba = vba;
1881 		zea = details->last_index;
1882 		if (zea > vea)
1883 			zea = vea;
1884 
1885 		if (unmap_mapping_range_vma(vma,
1886 			((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1887 			((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1888 				details) < 0)
1889 			goto restart;
1890 	}
1891 }
1892 
1893 static inline void unmap_mapping_range_list(struct list_head *head,
1894 					    struct zap_details *details)
1895 {
1896 	struct vm_area_struct *vma;
1897 
1898 	/*
1899 	 * In nonlinear VMAs there is no correspondence between virtual address
1900 	 * offset and file offset.  So we must perform an exhaustive search
1901 	 * across *all* the pages in each nonlinear VMA, not just the pages
1902 	 * whose virtual address lies outside the file truncation point.
1903 	 */
1904 restart:
1905 	list_for_each_entry(vma, head, shared.vm_set.list) {
1906 		/* Skip quickly over those we have already dealt with */
1907 		if (vma->vm_truncate_count == details->truncate_count)
1908 			continue;
1909 		details->nonlinear_vma = vma;
1910 		if (unmap_mapping_range_vma(vma, vma->vm_start,
1911 					vma->vm_end, details) < 0)
1912 			goto restart;
1913 	}
1914 }
1915 
1916 /**
1917  * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
1918  * @mapping: the address space containing mmaps to be unmapped.
1919  * @holebegin: byte in first page to unmap, relative to the start of
1920  * the underlying file.  This will be rounded down to a PAGE_SIZE
1921  * boundary.  Note that this is different from vmtruncate(), which
1922  * must keep the partial page.  In contrast, we must get rid of
1923  * partial pages.
1924  * @holelen: size of prospective hole in bytes.  This will be rounded
1925  * up to a PAGE_SIZE boundary.  A holelen of zero truncates to the
1926  * end of the file.
1927  * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1928  * but 0 when invalidating pagecache, don't throw away private data.
1929  */
1930 void unmap_mapping_range(struct address_space *mapping,
1931 		loff_t const holebegin, loff_t const holelen, int even_cows)
1932 {
1933 	struct zap_details details;
1934 	pgoff_t hba = holebegin >> PAGE_SHIFT;
1935 	pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1936 
1937 	/* Check for overflow. */
1938 	if (sizeof(holelen) > sizeof(hlen)) {
1939 		long long holeend =
1940 			(holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1941 		if (holeend & ~(long long)ULONG_MAX)
1942 			hlen = ULONG_MAX - hba + 1;
1943 	}
1944 
1945 	details.check_mapping = even_cows? NULL: mapping;
1946 	details.nonlinear_vma = NULL;
1947 	details.first_index = hba;
1948 	details.last_index = hba + hlen - 1;
1949 	if (details.last_index < details.first_index)
1950 		details.last_index = ULONG_MAX;
1951 	details.i_mmap_lock = &mapping->i_mmap_lock;
1952 
1953 	spin_lock(&mapping->i_mmap_lock);
1954 
1955 	/* serialize i_size write against truncate_count write */
1956 	smp_wmb();
1957 	/* Protect against page faults, and endless unmapping loops */
1958 	mapping->truncate_count++;
1959 	/*
1960 	 * For archs where spin_lock has inclusive semantics like ia64
1961 	 * this smp_mb() will prevent to read pagetable contents
1962 	 * before the truncate_count increment is visible to
1963 	 * other cpus.
1964 	 */
1965 	smp_mb();
1966 	if (unlikely(is_restart_addr(mapping->truncate_count))) {
1967 		if (mapping->truncate_count == 0)
1968 			reset_vma_truncate_counts(mapping);
1969 		mapping->truncate_count++;
1970 	}
1971 	details.truncate_count = mapping->truncate_count;
1972 
1973 	if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1974 		unmap_mapping_range_tree(&mapping->i_mmap, &details);
1975 	if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1976 		unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1977 	spin_unlock(&mapping->i_mmap_lock);
1978 }
1979 EXPORT_SYMBOL(unmap_mapping_range);
1980 
1981 /**
1982  * vmtruncate - unmap mappings "freed" by truncate() syscall
1983  * @inode: inode of the file used
1984  * @offset: file offset to start truncating
1985  *
1986  * NOTE! We have to be ready to update the memory sharing
1987  * between the file and the memory map for a potential last
1988  * incomplete page.  Ugly, but necessary.
1989  */
1990 int vmtruncate(struct inode * inode, loff_t offset)
1991 {
1992 	struct address_space *mapping = inode->i_mapping;
1993 	unsigned long limit;
1994 
1995 	if (inode->i_size < offset)
1996 		goto do_expand;
1997 	/*
1998 	 * truncation of in-use swapfiles is disallowed - it would cause
1999 	 * subsequent swapout to scribble on the now-freed blocks.
2000 	 */
2001 	if (IS_SWAPFILE(inode))
2002 		goto out_busy;
2003 	i_size_write(inode, offset);
2004 	unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2005 	truncate_inode_pages(mapping, offset);
2006 	goto out_truncate;
2007 
2008 do_expand:
2009 	limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2010 	if (limit != RLIM_INFINITY && offset > limit)
2011 		goto out_sig;
2012 	if (offset > inode->i_sb->s_maxbytes)
2013 		goto out_big;
2014 	i_size_write(inode, offset);
2015 
2016 out_truncate:
2017 	if (inode->i_op && inode->i_op->truncate)
2018 		inode->i_op->truncate(inode);
2019 	return 0;
2020 out_sig:
2021 	send_sig(SIGXFSZ, current, 0);
2022 out_big:
2023 	return -EFBIG;
2024 out_busy:
2025 	return -ETXTBSY;
2026 }
2027 EXPORT_SYMBOL(vmtruncate);
2028 
2029 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2030 {
2031 	struct address_space *mapping = inode->i_mapping;
2032 
2033 	/*
2034 	 * If the underlying filesystem is not going to provide
2035 	 * a way to truncate a range of blocks (punch a hole) -
2036 	 * we should return failure right now.
2037 	 */
2038 	if (!inode->i_op || !inode->i_op->truncate_range)
2039 		return -ENOSYS;
2040 
2041 	mutex_lock(&inode->i_mutex);
2042 	down_write(&inode->i_alloc_sem);
2043 	unmap_mapping_range(mapping, offset, (end - offset), 1);
2044 	truncate_inode_pages_range(mapping, offset, end);
2045 	inode->i_op->truncate_range(inode, offset, end);
2046 	up_write(&inode->i_alloc_sem);
2047 	mutex_unlock(&inode->i_mutex);
2048 
2049 	return 0;
2050 }
2051 
2052 /**
2053  * swapin_readahead - swap in pages in hope we need them soon
2054  * @entry: swap entry of this memory
2055  * @addr: address to start
2056  * @vma: user vma this addresses belong to
2057  *
2058  * Primitive swap readahead code. We simply read an aligned block of
2059  * (1 << page_cluster) entries in the swap area. This method is chosen
2060  * because it doesn't cost us any seek time.  We also make sure to queue
2061  * the 'original' request together with the readahead ones...
2062  *
2063  * This has been extended to use the NUMA policies from the mm triggering
2064  * the readahead.
2065  *
2066  * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
2067  */
2068 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
2069 {
2070 #ifdef CONFIG_NUMA
2071 	struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2072 #endif
2073 	int i, num;
2074 	struct page *new_page;
2075 	unsigned long offset;
2076 
2077 	/*
2078 	 * Get the number of handles we should do readahead io to.
2079 	 */
2080 	num = valid_swaphandles(entry, &offset);
2081 	for (i = 0; i < num; offset++, i++) {
2082 		/* Ok, do the async read-ahead now */
2083 		new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2084 							   offset), vma, addr);
2085 		if (!new_page)
2086 			break;
2087 		page_cache_release(new_page);
2088 #ifdef CONFIG_NUMA
2089 		/*
2090 		 * Find the next applicable VMA for the NUMA policy.
2091 		 */
2092 		addr += PAGE_SIZE;
2093 		if (addr == 0)
2094 			vma = NULL;
2095 		if (vma) {
2096 			if (addr >= vma->vm_end) {
2097 				vma = next_vma;
2098 				next_vma = vma ? vma->vm_next : NULL;
2099 			}
2100 			if (vma && addr < vma->vm_start)
2101 				vma = NULL;
2102 		} else {
2103 			if (next_vma && addr >= next_vma->vm_start) {
2104 				vma = next_vma;
2105 				next_vma = vma->vm_next;
2106 			}
2107 		}
2108 #endif
2109 	}
2110 	lru_add_drain();	/* Push any new pages onto the LRU now */
2111 }
2112 
2113 /*
2114  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2115  * but allow concurrent faults), and pte mapped but not yet locked.
2116  * We return with mmap_sem still held, but pte unmapped and unlocked.
2117  */
2118 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2119 		unsigned long address, pte_t *page_table, pmd_t *pmd,
2120 		int write_access, pte_t orig_pte)
2121 {
2122 	spinlock_t *ptl;
2123 	struct page *page;
2124 	swp_entry_t entry;
2125 	pte_t pte;
2126 	int ret = VM_FAULT_MINOR;
2127 
2128 	if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2129 		goto out;
2130 
2131 	entry = pte_to_swp_entry(orig_pte);
2132 	if (is_migration_entry(entry)) {
2133 		migration_entry_wait(mm, pmd, address);
2134 		goto out;
2135 	}
2136 	delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2137 	page = lookup_swap_cache(entry);
2138 	if (!page) {
2139 		grab_swap_token(); /* Contend for token _before_ read-in */
2140  		swapin_readahead(entry, address, vma);
2141  		page = read_swap_cache_async(entry, vma, address);
2142 		if (!page) {
2143 			/*
2144 			 * Back out if somebody else faulted in this pte
2145 			 * while we released the pte lock.
2146 			 */
2147 			page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2148 			if (likely(pte_same(*page_table, orig_pte)))
2149 				ret = VM_FAULT_OOM;
2150 			delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2151 			goto unlock;
2152 		}
2153 
2154 		/* Had to read the page from swap area: Major fault */
2155 		ret = VM_FAULT_MAJOR;
2156 		count_vm_event(PGMAJFAULT);
2157 	}
2158 
2159 	delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2160 	mark_page_accessed(page);
2161 	lock_page(page);
2162 
2163 	/*
2164 	 * Back out if somebody else already faulted in this pte.
2165 	 */
2166 	page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2167 	if (unlikely(!pte_same(*page_table, orig_pte)))
2168 		goto out_nomap;
2169 
2170 	if (unlikely(!PageUptodate(page))) {
2171 		ret = VM_FAULT_SIGBUS;
2172 		goto out_nomap;
2173 	}
2174 
2175 	/* The page isn't present yet, go ahead with the fault. */
2176 
2177 	inc_mm_counter(mm, anon_rss);
2178 	pte = mk_pte(page, vma->vm_page_prot);
2179 	if (write_access && can_share_swap_page(page)) {
2180 		pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2181 		write_access = 0;
2182 	}
2183 
2184 	flush_icache_page(vma, page);
2185 	set_pte_at(mm, address, page_table, pte);
2186 	page_add_anon_rmap(page, vma, address);
2187 
2188 	swap_free(entry);
2189 	if (vm_swap_full())
2190 		remove_exclusive_swap_page(page);
2191 	unlock_page(page);
2192 
2193 	if (write_access) {
2194 		if (do_wp_page(mm, vma, address,
2195 				page_table, pmd, ptl, pte) == VM_FAULT_OOM)
2196 			ret = VM_FAULT_OOM;
2197 		goto out;
2198 	}
2199 
2200 	/* No need to invalidate - it was non-present before */
2201 	update_mmu_cache(vma, address, pte);
2202 	lazy_mmu_prot_update(pte);
2203 unlock:
2204 	pte_unmap_unlock(page_table, ptl);
2205 out:
2206 	return ret;
2207 out_nomap:
2208 	pte_unmap_unlock(page_table, ptl);
2209 	unlock_page(page);
2210 	page_cache_release(page);
2211 	return ret;
2212 }
2213 
2214 /*
2215  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2216  * but allow concurrent faults), and pte mapped but not yet locked.
2217  * We return with mmap_sem still held, but pte unmapped and unlocked.
2218  */
2219 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2220 		unsigned long address, pte_t *page_table, pmd_t *pmd,
2221 		int write_access)
2222 {
2223 	struct page *page;
2224 	spinlock_t *ptl;
2225 	pte_t entry;
2226 
2227 	if (write_access) {
2228 		/* Allocate our own private page. */
2229 		pte_unmap(page_table);
2230 
2231 		if (unlikely(anon_vma_prepare(vma)))
2232 			goto oom;
2233 		page = alloc_zeroed_user_highpage(vma, address);
2234 		if (!page)
2235 			goto oom;
2236 
2237 		entry = mk_pte(page, vma->vm_page_prot);
2238 		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2239 
2240 		page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2241 		if (!pte_none(*page_table))
2242 			goto release;
2243 		inc_mm_counter(mm, anon_rss);
2244 		lru_cache_add_active(page);
2245 		page_add_new_anon_rmap(page, vma, address);
2246 	} else {
2247 		/* Map the ZERO_PAGE - vm_page_prot is readonly */
2248 		page = ZERO_PAGE(address);
2249 		page_cache_get(page);
2250 		entry = mk_pte(page, vma->vm_page_prot);
2251 
2252 		ptl = pte_lockptr(mm, pmd);
2253 		spin_lock(ptl);
2254 		if (!pte_none(*page_table))
2255 			goto release;
2256 		inc_mm_counter(mm, file_rss);
2257 		page_add_file_rmap(page);
2258 	}
2259 
2260 	set_pte_at(mm, address, page_table, entry);
2261 
2262 	/* No need to invalidate - it was non-present before */
2263 	update_mmu_cache(vma, address, entry);
2264 	lazy_mmu_prot_update(entry);
2265 unlock:
2266 	pte_unmap_unlock(page_table, ptl);
2267 	return VM_FAULT_MINOR;
2268 release:
2269 	page_cache_release(page);
2270 	goto unlock;
2271 oom:
2272 	return VM_FAULT_OOM;
2273 }
2274 
2275 /*
2276  * do_no_page() tries to create a new page mapping. It aggressively
2277  * tries to share with existing pages, but makes a separate copy if
2278  * the "write_access" parameter is true in order to avoid the next
2279  * page fault.
2280  *
2281  * As this is called only for pages that do not currently exist, we
2282  * do not need to flush old virtual caches or the TLB.
2283  *
2284  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2285  * but allow concurrent faults), and pte mapped but not yet locked.
2286  * We return with mmap_sem still held, but pte unmapped and unlocked.
2287  */
2288 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2289 		unsigned long address, pte_t *page_table, pmd_t *pmd,
2290 		int write_access)
2291 {
2292 	spinlock_t *ptl;
2293 	struct page *new_page;
2294 	struct address_space *mapping = NULL;
2295 	pte_t entry;
2296 	unsigned int sequence = 0;
2297 	int ret = VM_FAULT_MINOR;
2298 	int anon = 0;
2299 	struct page *dirty_page = NULL;
2300 
2301 	pte_unmap(page_table);
2302 	BUG_ON(vma->vm_flags & VM_PFNMAP);
2303 
2304 	if (vma->vm_file) {
2305 		mapping = vma->vm_file->f_mapping;
2306 		sequence = mapping->truncate_count;
2307 		smp_rmb(); /* serializes i_size against truncate_count */
2308 	}
2309 retry:
2310 	new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2311 	/*
2312 	 * No smp_rmb is needed here as long as there's a full
2313 	 * spin_lock/unlock sequence inside the ->nopage callback
2314 	 * (for the pagecache lookup) that acts as an implicit
2315 	 * smp_mb() and prevents the i_size read to happen
2316 	 * after the next truncate_count read.
2317 	 */
2318 
2319 	/* no page was available -- either SIGBUS, OOM or REFAULT */
2320 	if (unlikely(new_page == NOPAGE_SIGBUS))
2321 		return VM_FAULT_SIGBUS;
2322 	else if (unlikely(new_page == NOPAGE_OOM))
2323 		return VM_FAULT_OOM;
2324 	else if (unlikely(new_page == NOPAGE_REFAULT))
2325 		return VM_FAULT_MINOR;
2326 
2327 	/*
2328 	 * Should we do an early C-O-W break?
2329 	 */
2330 	if (write_access) {
2331 		if (!(vma->vm_flags & VM_SHARED)) {
2332 			struct page *page;
2333 
2334 			if (unlikely(anon_vma_prepare(vma)))
2335 				goto oom;
2336 			page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2337 			if (!page)
2338 				goto oom;
2339 			copy_user_highpage(page, new_page, address, vma);
2340 			page_cache_release(new_page);
2341 			new_page = page;
2342 			anon = 1;
2343 
2344 		} else {
2345 			/* if the page will be shareable, see if the backing
2346 			 * address space wants to know that the page is about
2347 			 * to become writable */
2348 			if (vma->vm_ops->page_mkwrite &&
2349 			    vma->vm_ops->page_mkwrite(vma, new_page) < 0
2350 			    ) {
2351 				page_cache_release(new_page);
2352 				return VM_FAULT_SIGBUS;
2353 			}
2354 		}
2355 	}
2356 
2357 	page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2358 	/*
2359 	 * For a file-backed vma, someone could have truncated or otherwise
2360 	 * invalidated this page.  If unmap_mapping_range got called,
2361 	 * retry getting the page.
2362 	 */
2363 	if (mapping && unlikely(sequence != mapping->truncate_count)) {
2364 		pte_unmap_unlock(page_table, ptl);
2365 		page_cache_release(new_page);
2366 		cond_resched();
2367 		sequence = mapping->truncate_count;
2368 		smp_rmb();
2369 		goto retry;
2370 	}
2371 
2372 	/*
2373 	 * This silly early PAGE_DIRTY setting removes a race
2374 	 * due to the bad i386 page protection. But it's valid
2375 	 * for other architectures too.
2376 	 *
2377 	 * Note that if write_access is true, we either now have
2378 	 * an exclusive copy of the page, or this is a shared mapping,
2379 	 * so we can make it writable and dirty to avoid having to
2380 	 * handle that later.
2381 	 */
2382 	/* Only go through if we didn't race with anybody else... */
2383 	if (pte_none(*page_table)) {
2384 		flush_icache_page(vma, new_page);
2385 		entry = mk_pte(new_page, vma->vm_page_prot);
2386 		if (write_access)
2387 			entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2388 		set_pte_at(mm, address, page_table, entry);
2389 		if (anon) {
2390 			inc_mm_counter(mm, anon_rss);
2391 			lru_cache_add_active(new_page);
2392 			page_add_new_anon_rmap(new_page, vma, address);
2393 		} else {
2394 			inc_mm_counter(mm, file_rss);
2395 			page_add_file_rmap(new_page);
2396 			if (write_access) {
2397 				dirty_page = new_page;
2398 				get_page(dirty_page);
2399 			}
2400 		}
2401 	} else {
2402 		/* One of our sibling threads was faster, back out. */
2403 		page_cache_release(new_page);
2404 		goto unlock;
2405 	}
2406 
2407 	/* no need to invalidate: a not-present page shouldn't be cached */
2408 	update_mmu_cache(vma, address, entry);
2409 	lazy_mmu_prot_update(entry);
2410 unlock:
2411 	pte_unmap_unlock(page_table, ptl);
2412 	if (dirty_page) {
2413 		set_page_dirty_balance(dirty_page);
2414 		put_page(dirty_page);
2415 	}
2416 	return ret;
2417 oom:
2418 	page_cache_release(new_page);
2419 	return VM_FAULT_OOM;
2420 }
2421 
2422 /*
2423  * do_no_pfn() tries to create a new page mapping for a page without
2424  * a struct_page backing it
2425  *
2426  * As this is called only for pages that do not currently exist, we
2427  * do not need to flush old virtual caches or the TLB.
2428  *
2429  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2430  * but allow concurrent faults), and pte mapped but not yet locked.
2431  * We return with mmap_sem still held, but pte unmapped and unlocked.
2432  *
2433  * It is expected that the ->nopfn handler always returns the same pfn
2434  * for a given virtual mapping.
2435  *
2436  * Mark this `noinline' to prevent it from bloating the main pagefault code.
2437  */
2438 static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2439 		     unsigned long address, pte_t *page_table, pmd_t *pmd,
2440 		     int write_access)
2441 {
2442 	spinlock_t *ptl;
2443 	pte_t entry;
2444 	unsigned long pfn;
2445 	int ret = VM_FAULT_MINOR;
2446 
2447 	pte_unmap(page_table);
2448 	BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2449 	BUG_ON(is_cow_mapping(vma->vm_flags));
2450 
2451 	pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2452 	if (unlikely(pfn == NOPFN_OOM))
2453 		return VM_FAULT_OOM;
2454 	else if (unlikely(pfn == NOPFN_SIGBUS))
2455 		return VM_FAULT_SIGBUS;
2456 	else if (unlikely(pfn == NOPFN_REFAULT))
2457 		return VM_FAULT_MINOR;
2458 
2459 	page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2460 
2461 	/* Only go through if we didn't race with anybody else... */
2462 	if (pte_none(*page_table)) {
2463 		entry = pfn_pte(pfn, vma->vm_page_prot);
2464 		if (write_access)
2465 			entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2466 		set_pte_at(mm, address, page_table, entry);
2467 	}
2468 	pte_unmap_unlock(page_table, ptl);
2469 	return ret;
2470 }
2471 
2472 /*
2473  * Fault of a previously existing named mapping. Repopulate the pte
2474  * from the encoded file_pte if possible. This enables swappable
2475  * nonlinear vmas.
2476  *
2477  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2478  * but allow concurrent faults), and pte mapped but not yet locked.
2479  * We return with mmap_sem still held, but pte unmapped and unlocked.
2480  */
2481 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2482 		unsigned long address, pte_t *page_table, pmd_t *pmd,
2483 		int write_access, pte_t orig_pte)
2484 {
2485 	pgoff_t pgoff;
2486 	int err;
2487 
2488 	if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2489 		return VM_FAULT_MINOR;
2490 
2491 	if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2492 		/*
2493 		 * Page table corrupted: show pte and kill process.
2494 		 */
2495 		print_bad_pte(vma, orig_pte, address);
2496 		return VM_FAULT_OOM;
2497 	}
2498 	/* We can then assume vm->vm_ops && vma->vm_ops->populate */
2499 
2500 	pgoff = pte_to_pgoff(orig_pte);
2501 	err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2502 					vma->vm_page_prot, pgoff, 0);
2503 	if (err == -ENOMEM)
2504 		return VM_FAULT_OOM;
2505 	if (err)
2506 		return VM_FAULT_SIGBUS;
2507 	return VM_FAULT_MAJOR;
2508 }
2509 
2510 /*
2511  * These routines also need to handle stuff like marking pages dirty
2512  * and/or accessed for architectures that don't do it in hardware (most
2513  * RISC architectures).  The early dirtying is also good on the i386.
2514  *
2515  * There is also a hook called "update_mmu_cache()" that architectures
2516  * with external mmu caches can use to update those (ie the Sparc or
2517  * PowerPC hashed page tables that act as extended TLBs).
2518  *
2519  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2520  * but allow concurrent faults), and pte mapped but not yet locked.
2521  * We return with mmap_sem still held, but pte unmapped and unlocked.
2522  */
2523 static inline int handle_pte_fault(struct mm_struct *mm,
2524 		struct vm_area_struct *vma, unsigned long address,
2525 		pte_t *pte, pmd_t *pmd, int write_access)
2526 {
2527 	pte_t entry;
2528 	pte_t old_entry;
2529 	spinlock_t *ptl;
2530 
2531 	old_entry = entry = *pte;
2532 	if (!pte_present(entry)) {
2533 		if (pte_none(entry)) {
2534 			if (vma->vm_ops) {
2535 				if (vma->vm_ops->nopage)
2536 					return do_no_page(mm, vma, address,
2537 							  pte, pmd,
2538 							  write_access);
2539 				if (unlikely(vma->vm_ops->nopfn))
2540 					return do_no_pfn(mm, vma, address, pte,
2541 							 pmd, write_access);
2542 			}
2543 			return do_anonymous_page(mm, vma, address,
2544 						 pte, pmd, write_access);
2545 		}
2546 		if (pte_file(entry))
2547 			return do_file_page(mm, vma, address,
2548 					pte, pmd, write_access, entry);
2549 		return do_swap_page(mm, vma, address,
2550 					pte, pmd, write_access, entry);
2551 	}
2552 
2553 	ptl = pte_lockptr(mm, pmd);
2554 	spin_lock(ptl);
2555 	if (unlikely(!pte_same(*pte, entry)))
2556 		goto unlock;
2557 	if (write_access) {
2558 		if (!pte_write(entry))
2559 			return do_wp_page(mm, vma, address,
2560 					pte, pmd, ptl, entry);
2561 		entry = pte_mkdirty(entry);
2562 	}
2563 	entry = pte_mkyoung(entry);
2564 	if (!pte_same(old_entry, entry)) {
2565 		ptep_set_access_flags(vma, address, pte, entry, write_access);
2566 		update_mmu_cache(vma, address, entry);
2567 		lazy_mmu_prot_update(entry);
2568 	} else {
2569 		/*
2570 		 * This is needed only for protection faults but the arch code
2571 		 * is not yet telling us if this is a protection fault or not.
2572 		 * This still avoids useless tlb flushes for .text page faults
2573 		 * with threads.
2574 		 */
2575 		if (write_access)
2576 			flush_tlb_page(vma, address);
2577 	}
2578 unlock:
2579 	pte_unmap_unlock(pte, ptl);
2580 	return VM_FAULT_MINOR;
2581 }
2582 
2583 /*
2584  * By the time we get here, we already hold the mm semaphore
2585  */
2586 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2587 		unsigned long address, int write_access)
2588 {
2589 	pgd_t *pgd;
2590 	pud_t *pud;
2591 	pmd_t *pmd;
2592 	pte_t *pte;
2593 
2594 	__set_current_state(TASK_RUNNING);
2595 
2596 	count_vm_event(PGFAULT);
2597 
2598 	if (unlikely(is_vm_hugetlb_page(vma)))
2599 		return hugetlb_fault(mm, vma, address, write_access);
2600 
2601 	pgd = pgd_offset(mm, address);
2602 	pud = pud_alloc(mm, pgd, address);
2603 	if (!pud)
2604 		return VM_FAULT_OOM;
2605 	pmd = pmd_alloc(mm, pud, address);
2606 	if (!pmd)
2607 		return VM_FAULT_OOM;
2608 	pte = pte_alloc_map(mm, pmd, address);
2609 	if (!pte)
2610 		return VM_FAULT_OOM;
2611 
2612 	return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2613 }
2614 
2615 EXPORT_SYMBOL_GPL(__handle_mm_fault);
2616 
2617 #ifndef __PAGETABLE_PUD_FOLDED
2618 /*
2619  * Allocate page upper directory.
2620  * We've already handled the fast-path in-line.
2621  */
2622 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2623 {
2624 	pud_t *new = pud_alloc_one(mm, address);
2625 	if (!new)
2626 		return -ENOMEM;
2627 
2628 	spin_lock(&mm->page_table_lock);
2629 	if (pgd_present(*pgd))		/* Another has populated it */
2630 		pud_free(new);
2631 	else
2632 		pgd_populate(mm, pgd, new);
2633 	spin_unlock(&mm->page_table_lock);
2634 	return 0;
2635 }
2636 #endif /* __PAGETABLE_PUD_FOLDED */
2637 
2638 #ifndef __PAGETABLE_PMD_FOLDED
2639 /*
2640  * Allocate page middle directory.
2641  * We've already handled the fast-path in-line.
2642  */
2643 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2644 {
2645 	pmd_t *new = pmd_alloc_one(mm, address);
2646 	if (!new)
2647 		return -ENOMEM;
2648 
2649 	spin_lock(&mm->page_table_lock);
2650 #ifndef __ARCH_HAS_4LEVEL_HACK
2651 	if (pud_present(*pud))		/* Another has populated it */
2652 		pmd_free(new);
2653 	else
2654 		pud_populate(mm, pud, new);
2655 #else
2656 	if (pgd_present(*pud))		/* Another has populated it */
2657 		pmd_free(new);
2658 	else
2659 		pgd_populate(mm, pud, new);
2660 #endif /* __ARCH_HAS_4LEVEL_HACK */
2661 	spin_unlock(&mm->page_table_lock);
2662 	return 0;
2663 }
2664 #endif /* __PAGETABLE_PMD_FOLDED */
2665 
2666 int make_pages_present(unsigned long addr, unsigned long end)
2667 {
2668 	int ret, len, write;
2669 	struct vm_area_struct * vma;
2670 
2671 	vma = find_vma(current->mm, addr);
2672 	if (!vma)
2673 		return -1;
2674 	write = (vma->vm_flags & VM_WRITE) != 0;
2675 	BUG_ON(addr >= end);
2676 	BUG_ON(end > vma->vm_end);
2677 	len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2678 	ret = get_user_pages(current, current->mm, addr,
2679 			len, write, 0, NULL, NULL);
2680 	if (ret < 0)
2681 		return ret;
2682 	return ret == len ? 0 : -1;
2683 }
2684 
2685 /*
2686  * Map a vmalloc()-space virtual address to the physical page.
2687  */
2688 struct page * vmalloc_to_page(void * vmalloc_addr)
2689 {
2690 	unsigned long addr = (unsigned long) vmalloc_addr;
2691 	struct page *page = NULL;
2692 	pgd_t *pgd = pgd_offset_k(addr);
2693 	pud_t *pud;
2694 	pmd_t *pmd;
2695 	pte_t *ptep, pte;
2696 
2697 	if (!pgd_none(*pgd)) {
2698 		pud = pud_offset(pgd, addr);
2699 		if (!pud_none(*pud)) {
2700 			pmd = pmd_offset(pud, addr);
2701 			if (!pmd_none(*pmd)) {
2702 				ptep = pte_offset_map(pmd, addr);
2703 				pte = *ptep;
2704 				if (pte_present(pte))
2705 					page = pte_page(pte);
2706 				pte_unmap(ptep);
2707 			}
2708 		}
2709 	}
2710 	return page;
2711 }
2712 
2713 EXPORT_SYMBOL(vmalloc_to_page);
2714 
2715 /*
2716  * Map a vmalloc()-space virtual address to the physical page frame number.
2717  */
2718 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2719 {
2720 	return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2721 }
2722 
2723 EXPORT_SYMBOL(vmalloc_to_pfn);
2724 
2725 #if !defined(__HAVE_ARCH_GATE_AREA)
2726 
2727 #if defined(AT_SYSINFO_EHDR)
2728 static struct vm_area_struct gate_vma;
2729 
2730 static int __init gate_vma_init(void)
2731 {
2732 	gate_vma.vm_mm = NULL;
2733 	gate_vma.vm_start = FIXADDR_USER_START;
2734 	gate_vma.vm_end = FIXADDR_USER_END;
2735 	gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2736 	gate_vma.vm_page_prot = __P101;
2737 	/*
2738 	 * Make sure the vDSO gets into every core dump.
2739 	 * Dumping its contents makes post-mortem fully interpretable later
2740 	 * without matching up the same kernel and hardware config to see
2741 	 * what PC values meant.
2742 	 */
2743 	gate_vma.vm_flags |= VM_ALWAYSDUMP;
2744 	return 0;
2745 }
2746 __initcall(gate_vma_init);
2747 #endif
2748 
2749 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2750 {
2751 #ifdef AT_SYSINFO_EHDR
2752 	return &gate_vma;
2753 #else
2754 	return NULL;
2755 #endif
2756 }
2757 
2758 int in_gate_area_no_task(unsigned long addr)
2759 {
2760 #ifdef AT_SYSINFO_EHDR
2761 	if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2762 		return 1;
2763 #endif
2764 	return 0;
2765 }
2766 
2767 #endif	/* __HAVE_ARCH_GATE_AREA */
2768 
2769 /*
2770  * Access another process' address space.
2771  * Source/target buffer must be kernel space,
2772  * Do not walk the page table directly, use get_user_pages
2773  */
2774 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2775 {
2776 	struct mm_struct *mm;
2777 	struct vm_area_struct *vma;
2778 	struct page *page;
2779 	void *old_buf = buf;
2780 
2781 	mm = get_task_mm(tsk);
2782 	if (!mm)
2783 		return 0;
2784 
2785 	down_read(&mm->mmap_sem);
2786 	/* ignore errors, just check how much was sucessfully transfered */
2787 	while (len) {
2788 		int bytes, ret, offset;
2789 		void *maddr;
2790 
2791 		ret = get_user_pages(tsk, mm, addr, 1,
2792 				write, 1, &page, &vma);
2793 		if (ret <= 0)
2794 			break;
2795 
2796 		bytes = len;
2797 		offset = addr & (PAGE_SIZE-1);
2798 		if (bytes > PAGE_SIZE-offset)
2799 			bytes = PAGE_SIZE-offset;
2800 
2801 		maddr = kmap(page);
2802 		if (write) {
2803 			copy_to_user_page(vma, page, addr,
2804 					  maddr + offset, buf, bytes);
2805 			set_page_dirty_lock(page);
2806 		} else {
2807 			copy_from_user_page(vma, page, addr,
2808 					    buf, maddr + offset, bytes);
2809 		}
2810 		kunmap(page);
2811 		page_cache_release(page);
2812 		len -= bytes;
2813 		buf += bytes;
2814 		addr += bytes;
2815 	}
2816 	up_read(&mm->mmap_sem);
2817 	mmput(mm);
2818 
2819 	return buf - old_buf;
2820 }
2821