Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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 #include <linux/memcontrol.h>
 54 
 55 #include <asm/pgalloc.h>
 56 #include <asm/uaccess.h>
 57 #include <asm/tlb.h>
 58 #include <asm/tlbflush.h>
 59 #include <asm/pgtable.h>
 60 
 61 #include <linux/swapops.h>
 62 #include <linux/elf.h>
 63 
 64 #ifndef CONFIG_NEED_MULTIPLE_NODES
 65 /* use the per-pgdat data instead for discontigmem - mbligh */
 66 unsigned long max_mapnr;
 67 struct page *mem_map;
 68 
 69 EXPORT_SYMBOL(max_mapnr);
 70 EXPORT_SYMBOL(mem_map);
 71 #endif
 72 
 73 unsigned long num_physpages;
 74 /*
 75  * A number of key systems in x86 including ioremap() rely on the assumption
 76  * that high_memory defines the upper bound on direct map memory, then end
 77  * of ZONE_NORMAL.  Under CONFIG_DISCONTIG this means that max_low_pfn and
 78  * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
 79  * and ZONE_HIGHMEM.
 80  */
 81 void * high_memory;
 82 
 83 EXPORT_SYMBOL(num_physpages);
 84 EXPORT_SYMBOL(high_memory);
 85 
 86 /*
 87  * Randomize the address space (stacks, mmaps, brk, etc.).
 88  *
 89  * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
 90  *   as ancient (libc5 based) binaries can segfault. )
 91  */
 92 int randomize_va_space __read_mostly =
 93 #ifdef CONFIG_COMPAT_BRK
 94                                         1;
 95 #else
 96                                         2;
 97 #endif
 98 
 99 static int __init disable_randmaps(char *s)
100 {
101         randomize_va_space = 0;
102         return 1;
103 }
104 __setup("norandmaps", disable_randmaps);
105 
106 
107 /*
108  * If a p?d_bad entry is found while walking page tables, report
109  * the error, before resetting entry to p?d_none.  Usually (but
110  * very seldom) called out from the p?d_none_or_clear_bad macros.
111  */
112 
113 void pgd_clear_bad(pgd_t *pgd)
114 {
115         pgd_ERROR(*pgd);
116         pgd_clear(pgd);
117 }
118 
119 void pud_clear_bad(pud_t *pud)
120 {
121         pud_ERROR(*pud);
122         pud_clear(pud);
123 }
124 
125 void pmd_clear_bad(pmd_t *pmd)
126 {
127         pmd_ERROR(*pmd);
128         pmd_clear(pmd);
129 }
130 
131 /*
132  * Note: this doesn't free the actual pages themselves. That
133  * has been handled earlier when unmapping all the memory regions.
134  */
135 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
136 {
137         pgtable_t token = pmd_pgtable(*pmd);
138         pmd_clear(pmd);
139         pte_free_tlb(tlb, token);
140         tlb->mm->nr_ptes--;
141 }
142 
143 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
144                                 unsigned long addr, unsigned long end,
145                                 unsigned long floor, unsigned long ceiling)
146 {
147         pmd_t *pmd;
148         unsigned long next;
149         unsigned long start;
150 
151         start = addr;
152         pmd = pmd_offset(pud, addr);
153         do {
154                 next = pmd_addr_end(addr, end);
155                 if (pmd_none_or_clear_bad(pmd))
156                         continue;
157                 free_pte_range(tlb, pmd);
158         } while (pmd++, addr = next, addr != end);
159 
160         start &= PUD_MASK;
161         if (start < floor)
162                 return;
163         if (ceiling) {
164                 ceiling &= PUD_MASK;
165                 if (!ceiling)
166                         return;
167         }
168         if (end - 1 > ceiling - 1)
169                 return;
170 
171         pmd = pmd_offset(pud, start);
172         pud_clear(pud);
173         pmd_free_tlb(tlb, pmd);
174 }
175 
176 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
177                                 unsigned long addr, unsigned long end,
178                                 unsigned long floor, unsigned long ceiling)
179 {
180         pud_t *pud;
181         unsigned long next;
182         unsigned long start;
183 
184         start = addr;
185         pud = pud_offset(pgd, addr);
186         do {
187                 next = pud_addr_end(addr, end);
188                 if (pud_none_or_clear_bad(pud))
189                         continue;
190                 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
191         } while (pud++, addr = next, addr != end);
192 
193         start &= PGDIR_MASK;
194         if (start < floor)
195                 return;
196         if (ceiling) {
197                 ceiling &= PGDIR_MASK;
198                 if (!ceiling)
199                         return;
200         }
201         if (end - 1 > ceiling - 1)
202                 return;
203 
204         pud = pud_offset(pgd, start);
205         pgd_clear(pgd);
206         pud_free_tlb(tlb, pud);
207 }
208 
209 /*
210  * This function frees user-level page tables of a process.
211  *
212  * Must be called with pagetable lock held.
213  */
214 void free_pgd_range(struct mmu_gather **tlb,
215                         unsigned long addr, unsigned long end,
216                         unsigned long floor, unsigned long ceiling)
217 {
218         pgd_t *pgd;
219         unsigned long next;
220         unsigned long start;
221 
222         /*
223          * The next few lines have given us lots of grief...
224          *
225          * Why are we testing PMD* at this top level?  Because often
226          * there will be no work to do at all, and we'd prefer not to
227          * go all the way down to the bottom just to discover that.
228          *
229          * Why all these "- 1"s?  Because 0 represents both the bottom
230          * of the address space and the top of it (using -1 for the
231          * top wouldn't help much: the masks would do the wrong thing).
232          * The rule is that addr 0 and floor 0 refer to the bottom of
233          * the address space, but end 0 and ceiling 0 refer to the top
234          * Comparisons need to use "end - 1" and "ceiling - 1" (though
235          * that end 0 case should be mythical).
236          *
237          * Wherever addr is brought up or ceiling brought down, we must
238          * be careful to reject "the opposite 0" before it confuses the
239          * subsequent tests.  But what about where end is brought down
240          * by PMD_SIZE below? no, end can't go down to 0 there.
241          *
242          * Whereas we round start (addr) and ceiling down, by different
243          * masks at different levels, in order to test whether a table
244          * now has no other vmas using it, so can be freed, we don't
245          * bother to round floor or end up - the tests don't need that.
246          */
247 
248         addr &= PMD_MASK;
249         if (addr < floor) {
250                 addr += PMD_SIZE;
251                 if (!addr)
252                         return;
253         }
254         if (ceiling) {
255                 ceiling &= PMD_MASK;
256                 if (!ceiling)
257                         return;
258         }
259         if (end - 1 > ceiling - 1)
260                 end -= PMD_SIZE;
261         if (addr > end - 1)
262                 return;
263 
264         start = addr;
265         pgd = pgd_offset((*tlb)->mm, addr);
266         do {
267                 next = pgd_addr_end(addr, end);
268                 if (pgd_none_or_clear_bad(pgd))
269                         continue;
270                 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
271         } while (pgd++, addr = next, addr != end);
272 }
273 
274 #ifdef CONFIG_IA64
275 #define tlb_start_addr(tlb)     (tlb)->start_addr
276 #define tlb_end_addr(tlb)       (tlb)->end_addr
277 #else
278 #define tlb_start_addr(tlb)     0UL     /* only ia64 really uses it */
279 #define tlb_end_addr(tlb)       0UL     /* only ia64 really uses it */
280 #endif
281 
282 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
283                 unsigned long floor, unsigned long ceiling)
284 {
285 #ifdef CONFIG_PREEMPT
286         struct vm_area_struct *unlink = vma;
287         int fullmm = (*tlb)->fullmm;
288 
289         if (!vma)       /* Sometimes when exiting after an oops */
290                 return;
291 #ifndef CONFIG_PREEMPT_RT
292         if (vma->vm_next)
293 #endif
294                 tlb_finish_mmu(*tlb, tlb_start_addr(*tlb), tlb_end_addr(*tlb));
295         /*
296          * Hide vma from rmap and vmtruncate before freeeing pgtables,
297          * with preemption enabled, except when unmapping just one area.
298          */
299         while (unlink) {
300                 anon_vma_unlink(unlink);
301                 unlink_file_vma(unlink);
302                 unlink = unlink->vm_next;
303         }
304 #ifndef CONFIG_PREEMPT_RT
305         if (vma->vm_next)
306 #endif
307                 *tlb = tlb_gather_mmu(vma->vm_mm, fullmm);
308 #endif
309         while (vma) {
310                 struct vm_area_struct *next = vma->vm_next;
311                 unsigned long addr = vma->vm_start;
312 
313 #ifndef CONFIG_PREEMPT
314                 /*
315                  * Hide vma from rmap and vmtruncate before freeing pgtables
316                  */
317                 anon_vma_unlink(vma);
318                 unlink_file_vma(vma);
319 #endif
320 
321                 if (is_vm_hugetlb_page(vma)) {
322                         hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
323                                 floor, next? next->vm_start: ceiling);
324                 } else {
325                         /*
326                          * Optimization: gather nearby vmas into one call down
327                          */
328                         while (next && next->vm_start <= vma->vm_end + PMD_SIZE
329                                && !is_vm_hugetlb_page(next)) {
330                                 vma = next;
331                                 next = vma->vm_next;
332 #ifndef CONFIG_PREEMPT
333                                 anon_vma_unlink(vma);
334                                 unlink_file_vma(vma);
335 #endif
336                         }
337                         free_pgd_range(tlb, addr, vma->vm_end,
338                                 floor, next? next->vm_start: ceiling);
339                 }
340                 vma = next;
341         }
342 }
343 
344 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
345 {
346         pgtable_t new = pte_alloc_one(mm, address);
347         if (!new)
348                 return -ENOMEM;
349 
350         spin_lock(&mm->page_table_lock);
351         if (!pmd_present(*pmd)) {       /* Has another populated it ? */
352                 mm->nr_ptes++;
353                 pmd_populate(mm, pmd, new);
354                 new = NULL;
355         }
356         spin_unlock(&mm->page_table_lock);
357         if (new)
358                 pte_free(mm, new);
359         return 0;
360 }
361 
362 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
363 {
364         pte_t *new = pte_alloc_one_kernel(&init_mm, address);
365         if (!new)
366                 return -ENOMEM;
367 
368         spin_lock(&init_mm.page_table_lock);
369         if (!pmd_present(*pmd)) {       /* Has another populated it ? */
370                 pmd_populate_kernel(&init_mm, pmd, new);
371                 new = NULL;
372         }
373         spin_unlock(&init_mm.page_table_lock);
374         if (new)
375                 pte_free_kernel(&init_mm, new);
376         return 0;
377 }
378 
379 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
380 {
381         if (file_rss)
382                 add_mm_counter(mm, file_rss, file_rss);
383         if (anon_rss)
384                 add_mm_counter(mm, anon_rss, anon_rss);
385 }
386 
387 /*
388  * This function is called to print an error when a bad pte
389  * is found. For example, we might have a PFN-mapped pte in
390  * a region that doesn't allow it.
391  *
392  * The calling function must still handle the error.
393  */
394 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
395 {
396         printk(KERN_ERR "Bad pte = %08llx, process = %s, "
397                         "vm_flags = %lx, vaddr = %lx\n",
398                 (long long)pte_val(pte),
399                 (vma->vm_mm == current->mm ? current->comm : "???"),
400                 vma->vm_flags, vaddr);
401         dump_stack();
402 }
403 
404 static inline int is_cow_mapping(unsigned int flags)
405 {
406         return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
407 }
408 
409 /*
410  * This function gets the "struct page" associated with a pte.
411  *
412  * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
413  * will have each page table entry just pointing to a raw page frame
414  * number, and as far as the VM layer is concerned, those do not have
415  * pages associated with them - even if the PFN might point to memory
416  * that otherwise is perfectly fine and has a "struct page".
417  *
418  * The way we recognize those mappings is through the rules set up
419  * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
420  * and the vm_pgoff will point to the first PFN mapped: thus every
421  * page that is a raw mapping will always honor the rule
422  *
423  *      pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
424  *
425  * and if that isn't true, the page has been COW'ed (in which case it
426  * _does_ have a "struct page" associated with it even if it is in a
427  * VM_PFNMAP range).
428  */
429 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
430 {
431         unsigned long pfn = pte_pfn(pte);
432 
433         if (unlikely(vma->vm_flags & VM_PFNMAP)) {
434                 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
435                 if (pfn == vma->vm_pgoff + off)
436                         return NULL;
437                 if (!is_cow_mapping(vma->vm_flags))
438                         return NULL;
439         }
440 
441 #ifdef CONFIG_DEBUG_VM
442         /*
443          * Add some anal sanity checks for now. Eventually,
444          * we should just do "return pfn_to_page(pfn)", but
445          * in the meantime we check that we get a valid pfn,
446          * and that the resulting page looks ok.
447          */
448         if (unlikely(!pfn_valid(pfn))) {
449                 print_bad_pte(vma, pte, addr);
450                 return NULL;
451         }
452 #endif
453 
454         /*
455          * NOTE! We still have PageReserved() pages in the page 
456          * tables. 
457          *
458          * The PAGE_ZERO() pages and various VDSO mappings can
459          * cause them to exist.
460          */
461         return pfn_to_page(pfn);
462 }
463 
464 /*
465  * copy one vm_area from one task to the other. Assumes the page tables
466  * already present in the new task to be cleared in the whole range
467  * covered by this vma.
468  */
469 
470 static inline void
471 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
472                 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
473                 unsigned long addr, int *rss)
474 {
475         unsigned long vm_flags = vma->vm_flags;
476         pte_t pte = *src_pte;
477         struct page *page;
478 
479         /* pte contains position in swap or file, so copy. */
480         if (unlikely(!pte_present(pte))) {
481                 if (!pte_file(pte)) {
482                         swp_entry_t entry = pte_to_swp_entry(pte);
483 
484                         swap_duplicate(entry);
485                         /* make sure dst_mm is on swapoff's mmlist. */
486                         if (unlikely(list_empty(&dst_mm->mmlist))) {
487                                 spin_lock(&mmlist_lock);
488                                 if (list_empty(&dst_mm->mmlist))
489                                         list_add(&dst_mm->mmlist,
490                                                  &src_mm->mmlist);
491                                 spin_unlock(&mmlist_lock);
492                         }
493                         if (is_write_migration_entry(entry) &&
494                                         is_cow_mapping(vm_flags)) {
495                                 /*
496                                  * COW mappings require pages in both parent
497                                  * and child to be set to read.
498                                  */
499                                 make_migration_entry_read(&entry);
500                                 pte = swp_entry_to_pte(entry);
501                                 set_pte_at(src_mm, addr, src_pte, pte);
502                         }
503                 }
504                 goto out_set_pte;
505         }
506 
507         /*
508          * If it's a COW mapping, write protect it both
509          * in the parent and the child
510          */
511         if (is_cow_mapping(vm_flags)) {
512                 ptep_set_wrprotect(src_mm, addr, src_pte);
513                 pte = pte_wrprotect(pte);
514         }
515 
516         /*
517          * If it's a shared mapping, mark it clean in
518          * the child
519          */
520         if (vm_flags & VM_SHARED)
521                 pte = pte_mkclean(pte);
522         pte = pte_mkold(pte);
523 
524         page = vm_normal_page(vma, addr, pte);
525         if (page) {
526                 get_page(page);
527                 page_dup_rmap(page, vma, addr);
528                 rss[!!PageAnon(page)]++;
529         }
530 
531 out_set_pte:
532         set_pte_at(dst_mm, addr, dst_pte, pte);
533 }
534 
535 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
536                 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
537                 unsigned long addr, unsigned long end)
538 {
539         pte_t *src_pte, *dst_pte;
540         spinlock_t *src_ptl, *dst_ptl;
541         int progress = 0;
542         int rss[2];
543 
544 again:
545         rss[1] = rss[0] = 0;
546         dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
547         if (!dst_pte)
548                 return -ENOMEM;
549         src_pte = pte_offset_map_nested(src_pmd, addr);
550         src_ptl = pte_lockptr(src_mm, src_pmd);
551         spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
552         arch_enter_lazy_mmu_mode();
553 
554         do {
555                 /*
556                  * We are holding two locks at this point - either of them
557                  * could generate latencies in another task on another CPU.
558                  */
559                 if (progress >= 32) {
560                         progress = 0;
561                         if (need_resched() ||
562                             spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
563                                 break;
564                 }
565                 if (pte_none(*src_pte)) {
566                         progress++;
567                         continue;
568                 }
569                 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
570                 progress += 8;
571         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
572 
573         arch_leave_lazy_mmu_mode();
574         spin_unlock(src_ptl);
575         pte_unmap_nested(src_pte - 1);
576         add_mm_rss(dst_mm, rss[0], rss[1]);
577         pte_unmap_unlock(dst_pte - 1, dst_ptl);
578         cond_resched();
579         if (addr != end)
580                 goto again;
581         return 0;
582 }
583 
584 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
585                 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
586                 unsigned long addr, unsigned long end)
587 {
588         pmd_t *src_pmd, *dst_pmd;
589         unsigned long next;
590 
591         dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
592         if (!dst_pmd)
593                 return -ENOMEM;
594         src_pmd = pmd_offset(src_pud, addr);
595         do {
596                 next = pmd_addr_end(addr, end);
597                 if (pmd_none_or_clear_bad(src_pmd))
598                         continue;
599                 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
600                                                 vma, addr, next))
601                         return -ENOMEM;
602         } while (dst_pmd++, src_pmd++, addr = next, addr != end);
603         return 0;
604 }
605 
606 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
607                 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
608                 unsigned long addr, unsigned long end)
609 {
610         pud_t *src_pud, *dst_pud;
611         unsigned long next;
612 
613         dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
614         if (!dst_pud)
615                 return -ENOMEM;
616         src_pud = pud_offset(src_pgd, addr);
617         do {
618                 next = pud_addr_end(addr, end);
619                 if (pud_none_or_clear_bad(src_pud))
620                         continue;
621                 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
622                                                 vma, addr, next))
623                         return -ENOMEM;
624         } while (dst_pud++, src_pud++, addr = next, addr != end);
625         return 0;
626 }
627 
628 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
629                 struct vm_area_struct *vma)
630 {
631         pgd_t *src_pgd, *dst_pgd;
632         unsigned long next;
633         unsigned long addr = vma->vm_start;
634         unsigned long end = vma->vm_end;
635 
636         /*
637          * Don't copy ptes where a page fault will fill them correctly.
638          * Fork becomes much lighter when there are big shared or private
639          * readonly mappings. The tradeoff is that copy_page_range is more
640          * efficient than faulting.
641          */
642         if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
643                 if (!vma->anon_vma)
644                         return 0;
645         }
646 
647         if (is_vm_hugetlb_page(vma))
648                 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
649 
650         dst_pgd = pgd_offset(dst_mm, addr);
651         src_pgd = pgd_offset(src_mm, addr);
652         do {
653                 next = pgd_addr_end(addr, end);
654                 if (pgd_none_or_clear_bad(src_pgd))
655                         continue;
656                 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
657                                                 vma, addr, next))
658                         return -ENOMEM;
659         } while (dst_pgd++, src_pgd++, addr = next, addr != end);
660         return 0;
661 }
662 
663 static unsigned long zap_pte_range(struct mmu_gather *tlb,
664                                 struct vm_area_struct *vma, pmd_t *pmd,
665                                 unsigned long addr, unsigned long end,
666                                 long *zap_work, struct zap_details *details)
667 {
668         struct mm_struct *mm = tlb->mm;
669         pte_t *pte;
670         spinlock_t *ptl;
671         int file_rss = 0;
672         int anon_rss = 0;
673 
674         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
675         arch_enter_lazy_mmu_mode();
676         do {
677                 pte_t ptent = *pte;
678                 if (pte_none(ptent)) {
679                         (*zap_work)--;
680                         continue;
681                 }
682 
683                 (*zap_work) -= PAGE_SIZE;
684 
685                 if (pte_present(ptent)) {
686                         struct page *page;
687 
688                         page = vm_normal_page(vma, addr, ptent);
689                         if (unlikely(details) && page) {
690                                 /*
691                                  * unmap_shared_mapping_pages() wants to
692                                  * invalidate cache without truncating:
693                                  * unmap shared but keep private pages.
694                                  */
695                                 if (details->check_mapping &&
696                                     details->check_mapping != page->mapping)
697                                         continue;
698                                 /*
699                                  * Each page->index must be checked when
700                                  * invalidating or truncating nonlinear.
701                                  */
702                                 if (details->nonlinear_vma &&
703                                     (page->index < details->first_index ||
704                                      page->index > details->last_index))
705                                         continue;
706                         }
707                         ptent = ptep_get_and_clear_full(mm, addr, pte,
708                                                         tlb->fullmm);
709                         tlb_remove_tlb_entry(tlb, pte, addr);
710                         if (unlikely(!page))
711                                 continue;
712                         if (unlikely(details) && details->nonlinear_vma
713                             && linear_page_index(details->nonlinear_vma,
714                                                 addr) != page->index)
715                                 set_pte_at(mm, addr, pte,
716                                            pgoff_to_pte(page->index));
717                         if (PageAnon(page))
718                                 anon_rss--;
719                         else {
720                                 if (pte_dirty(ptent))
721                                         set_page_dirty(page);
722                                 if (pte_young(ptent))
723                                         SetPageReferenced(page);
724                                 file_rss--;
725                         }
726                         page_remove_rmap(page, vma);
727                         tlb_remove_page(tlb, page);
728                         continue;
729                 }
730                 /*
731                  * If details->check_mapping, we leave swap entries;
732                  * if details->nonlinear_vma, we leave file entries.
733                  */
734                 if (unlikely(details))
735                         continue;
736                 if (!pte_file(ptent))
737                         free_swap_and_cache(pte_to_swp_entry(ptent));
738                 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
739         } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
740 
741         add_mm_rss(mm, file_rss, anon_rss);
742         arch_leave_lazy_mmu_mode();
743         pte_unmap_unlock(pte - 1, ptl);
744 
745         return addr;
746 }
747 
748 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
749                                 struct vm_area_struct *vma, pud_t *pud,
750                                 unsigned long addr, unsigned long end,
751                                 long *zap_work, struct zap_details *details)
752 {
753         pmd_t *pmd;
754         unsigned long next;
755 
756         pmd = pmd_offset(pud, addr);
757         do {
758                 next = pmd_addr_end(addr, end);
759                 if (pmd_none_or_clear_bad(pmd)) {
760                         (*zap_work)--;
761                         continue;
762                 }
763                 next = zap_pte_range(tlb, vma, pmd, addr, next,
764                                                 zap_work, details);
765         } while (pmd++, addr = next, (addr != end && *zap_work > 0));
766 
767         return addr;
768 }
769 
770 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
771                                 struct vm_area_struct *vma, pgd_t *pgd,
772                                 unsigned long addr, unsigned long end,
773                                 long *zap_work, struct zap_details *details)
774 {
775         pud_t *pud;
776         unsigned long next;
777 
778         pud = pud_offset(pgd, addr);
779         do {
780                 next = pud_addr_end(addr, end);
781                 if (pud_none_or_clear_bad(pud)) {
782                         (*zap_work)--;
783                         continue;
784                 }
785                 next = zap_pmd_range(tlb, vma, pud, addr, next,
786                                                 zap_work, details);
787         } while (pud++, addr = next, (addr != end && *zap_work > 0));
788 
789         return addr;
790 }
791 
792 static unsigned long unmap_page_range(struct mmu_gather *tlb,
793                                 struct vm_area_struct *vma,
794                                 unsigned long addr, unsigned long end,
795                                 long *zap_work, struct zap_details *details)
796 {
797         pgd_t *pgd;
798         unsigned long next;
799 
800         if (details && !details->check_mapping && !details->nonlinear_vma)
801                 details = NULL;
802 
803         BUG_ON(addr >= end);
804         tlb_start_vma(tlb, vma);
805         pgd = pgd_offset(vma->vm_mm, addr);
806         do {
807                 next = pgd_addr_end(addr, end);
808                 if (pgd_none_or_clear_bad(pgd)) {
809                         (*zap_work)--;
810                         continue;
811                 }
812                 next = zap_pud_range(tlb, vma, pgd, addr, next,
813                                                 zap_work, details);
814         } while (pgd++, addr = next, (addr != end && *zap_work > 0));
815         tlb_end_vma(tlb, vma);
816 
817         return addr;
818 }
819 
820 #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_RT)
821 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
822 #else
823 /*
824  * No preempt: go for improved straight-line efficiency
825  * on PREEMPT_RT this is not a critical latency-path.
826  */
827 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
828 #endif
829 
830 /**
831  * unmap_vmas - unmap a range of memory covered by a list of vma's
832  * @tlbp: address of the caller's struct mmu_gather
833  * @vma: the starting vma
834  * @start_addr: virtual address at which to start unmapping
835  * @end_addr: virtual address at which to end unmapping
836  * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
837  * @details: details of nonlinear truncation or shared cache invalidation
838  *
839  * Returns the end address of the unmapping (restart addr if interrupted).
840  *
841  * Unmap all pages in the vma list.
842  *
843  * We aim to not hold locks for too long (for scheduling latency reasons).
844  * So zap pages in ZAP_BLOCK_SIZE bytecounts.  This means we need to
845  * return the ending mmu_gather to the caller.
846  *
847  * Only addresses between `start' and `end' will be unmapped.
848  *
849  * The VMA list must be sorted in ascending virtual address order.
850  *
851  * unmap_vmas() assumes that the caller will flush the whole unmapped address
852  * range after unmap_vmas() returns.  So the only responsibility here is to
853  * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
854  * drops the lock and schedules.
855  */
856 unsigned long unmap_vmas(struct mmu_gather **tlbp,
857                 struct vm_area_struct *vma, unsigned long start_addr,
858                 unsigned long end_addr, unsigned long *nr_accounted,
859                 struct zap_details *details)
860 {
861         long zap_work = ZAP_BLOCK_SIZE;
862         unsigned long tlb_start = 0;    /* For tlb_finish_mmu */
863         int tlb_start_valid = 0;
864         unsigned long start = start_addr;
865         spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
866         int fullmm = (*tlbp)->fullmm;
867 
868         for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
869                 unsigned long end;
870 
871                 start = max(vma->vm_start, start_addr);
872                 if (start >= vma->vm_end)
873                         continue;
874                 end = min(vma->vm_end, end_addr);
875                 if (end <= vma->vm_start)
876                         continue;
877 
878                 if (vma->vm_flags & VM_ACCOUNT)
879                         *nr_accounted += (end - start) >> PAGE_SHIFT;
880 
881                 while (start != end) {
882                         if (!tlb_start_valid) {
883                                 tlb_start = start;
884                                 tlb_start_valid = 1;
885                         }
886 
887                         if (unlikely(is_vm_hugetlb_page(vma))) {
888                                 unmap_hugepage_range(vma, start, end);
889                                 zap_work -= (end - start) /
890                                                 (HPAGE_SIZE / PAGE_SIZE);
891                                 start = end;
892                         } else
893                                 start = unmap_page_range(*tlbp, vma,
894                                                 start, end, &zap_work, details);
895 
896                         if (zap_work > 0) {
897                                 BUG_ON(start != end);
898                                 break;
899                         }
900 
901                         tlb_finish_mmu(*tlbp, tlb_start, start);
902 
903                         if (need_resched() ||
904                                 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
905                                 if (i_mmap_lock) {
906                                         *tlbp = NULL;
907                                         goto out;
908                                 }
909                                 cond_resched();
910                         }
911 
912                         *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
913                         tlb_start_valid = 0;
914                         zap_work = ZAP_BLOCK_SIZE;
915                 }
916         }
917 out:
918         return start;   /* which is now the end (or restart) address */
919 }
920 
921 /**
922  * zap_page_range - remove user pages in a given range
923  * @vma: vm_area_struct holding the applicable pages
924  * @address: starting address of pages to zap
925  * @size: number of bytes to zap
926  * @details: details of nonlinear truncation or shared cache invalidation
927  */
928 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
929                 unsigned long size, struct zap_details *details)
930 {
931         struct mm_struct *mm = vma->vm_mm;
932         struct mmu_gather *tlb;
933         unsigned long end = address + size;
934         unsigned long nr_accounted = 0;
935 
936         lru_add_drain();
937         tlb = tlb_gather_mmu(mm, 0);
938         update_hiwater_rss(mm);
939         end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
940         if (tlb)
941                 tlb_finish_mmu(tlb, address, end);
942         return end;
943 }
944 
945 /*
946  * Do a quick page-table lookup for a single page.
947  */
948 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
949                         unsigned int flags)
950 {
951         pgd_t *pgd;
952         pud_t *pud;
953         pmd_t *pmd;
954         pte_t *ptep, pte;
955         spinlock_t *ptl;
956         struct page *page;
957         struct mm_struct *mm = vma->vm_mm;
958 
959         page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
960         if (!IS_ERR(page)) {
961                 BUG_ON(flags & FOLL_GET);
962                 goto out;
963         }
964 
965         page = NULL;
966         pgd = pgd_offset(mm, address);
967         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
968                 goto no_page_table;
969 
970         pud = pud_offset(pgd, address);
971         if (pud_none(*pud) || unlikely(pud_bad(*pud)))
972                 goto no_page_table;
973         
974         pmd = pmd_offset(pud, address);
975         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
976                 goto no_page_table;
977 
978         if (pmd_huge(*pmd)) {
979                 BUG_ON(flags & FOLL_GET);
980                 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
981                 goto out;
982         }
983 
984         ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
985         if (!ptep)
986                 goto out;
987 
988         pte = *ptep;
989         if (!pte_present(pte))
990                 goto unlock;
991         if ((flags & FOLL_WRITE) && !pte_write(pte))
992                 goto unlock;
993         page = vm_normal_page(vma, address, pte);
994         if (unlikely(!page))
995                 goto unlock;
996 
997         if (flags & FOLL_GET)
998                 get_page(page);
999         if (flags & FOLL_TOUCH) {
1000                 if ((flags & FOLL_WRITE) &&
1001                     !pte_dirty(pte) && !PageDirty(page))
1002                         set_page_dirty(page);
1003                 mark_page_accessed(page);
1004         }
1005 unlock:
1006         pte_unmap_unlock(ptep, ptl);
1007 out:
1008         return page;
1009 
1010 no_page_table:
1011         /*
1012          * When core dumping an enormous anonymous area that nobody
1013          * has touched so far, we don't want to allocate page tables.
1014          */
1015         if (flags & FOLL_ANON) {
1016                 page = ZERO_PAGE(0);
1017                 if (flags & FOLL_GET)
1018                         get_page(page);
1019                 BUG_ON(flags & FOLL_WRITE);
1020         }
1021         return page;
1022 }
1023 
1024 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1025                 unsigned long start, int len, int write, int force,
1026                 struct page **pages, struct vm_area_struct **vmas)
1027 {
1028         int i;
1029         unsigned int vm_flags;
1030 
1031         if (len <= 0)
1032                 return 0;
1033         /* 
1034          * Require read or write permissions.
1035          * If 'force' is set, we only require the "MAY" flags.
1036          */
1037         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1038         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1039         i = 0;
1040 
1041         do {
1042                 struct vm_area_struct *vma;
1043                 unsigned int foll_flags;
1044 
1045                 vma = find_extend_vma(mm, start);
1046                 if (!vma && in_gate_area(tsk, start)) {
1047                         unsigned long pg = start & PAGE_MASK;
1048                         struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1049                         pgd_t *pgd;
1050                         pud_t *pud;
1051                         pmd_t *pmd;
1052                         pte_t *pte;
1053                         if (write) /* user gate pages are read-only */
1054                                 return i ? : -EFAULT;
1055                         if (pg > TASK_SIZE)
1056                                 pgd = pgd_offset_k(pg);
1057                         else
1058                                 pgd = pgd_offset_gate(mm, pg);
1059                         BUG_ON(pgd_none(*pgd));
1060                         pud = pud_offset(pgd, pg);
1061                         BUG_ON(pud_none(*pud));
1062                         pmd = pmd_offset(pud, pg);
1063                         if (pmd_none(*pmd))
1064                                 return i ? : -EFAULT;
1065                         pte = pte_offset_map(pmd, pg);
1066                         if (pte_none(*pte)) {
1067                                 pte_unmap(pte);
1068                                 return i ? : -EFAULT;
1069                         }
1070                         if (pages) {
1071                                 struct page *page = vm_normal_page(gate_vma, start, *pte);
1072                                 pages[i] = page;
1073                                 if (page)
1074                                         get_page(page);
1075                         }
1076                         pte_unmap(pte);
1077                         if (vmas)
1078                                 vmas[i] = gate_vma;
1079                         i++;
1080                         start += PAGE_SIZE;
1081                         len--;
1082                         continue;
1083                 }
1084 
1085                 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1086                                 || !(vm_flags & vma->vm_flags))
1087                         return i ? : -EFAULT;
1088 
1089                 if (is_vm_hugetlb_page(vma)) {
1090                         i = follow_hugetlb_page(mm, vma, pages, vmas,
1091                                                 &start, &len, i, write);
1092                         continue;
1093                 }
1094 
1095                 foll_flags = FOLL_TOUCH;
1096                 if (pages)
1097                         foll_flags |= FOLL_GET;
1098                 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1099                     (!vma->vm_ops || (!vma->vm_ops->nopage &&
1100                                         !vma->vm_ops->fault)))
1101                         foll_flags |= FOLL_ANON;
1102 
1103                 do {
1104                         struct page *page;
1105 
1106                         /*
1107                          * If tsk is ooming, cut off its access to large memory
1108                          * allocations. It has a pending SIGKILL, but it can't
1109                          * be processed until returning to user space.
1110                          */
1111                         if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1112                                 return -ENOMEM;
1113 
1114                         if (write)
1115                                 foll_flags |= FOLL_WRITE;
1116 
1117                         cond_resched();
1118                         while (!(page = follow_page(vma, start, foll_flags))) {
1119                                 int ret;
1120                                 ret = handle_mm_fault(mm, vma, start,
1121                                                 foll_flags & FOLL_WRITE);
1122                                 if (ret & VM_FAULT_ERROR) {
1123                                         if (ret & VM_FAULT_OOM)
1124                                                 return i ? i : -ENOMEM;
1125                                         else if (ret & VM_FAULT_SIGBUS)
1126                                                 return i ? i : -EFAULT;
1127                                         BUG();
1128                                 }
1129                                 if (ret & VM_FAULT_MAJOR)
1130                                         tsk->maj_flt++;
1131                                 else
1132                                         tsk->min_flt++;
1133 
1134                                 /*
1135                                  * The VM_FAULT_WRITE bit tells us that
1136                                  * do_wp_page has broken COW when necessary,
1137                                  * even if maybe_mkwrite decided not to set
1138                                  * pte_write. We can thus safely do subsequent
1139                                  * page lookups as if they were reads.
1140                                  */
1141                                 if (ret & VM_FAULT_WRITE)
1142                                         foll_flags &= ~FOLL_WRITE;
1143 
1144                                 cond_resched();
1145                         }
1146                         if (pages) {
1147                                 pages[i] = page;
1148 
1149                                 flush_anon_page(vma, page, start);
1150                                 flush_dcache_page(page);
1151                         }
1152                         if (vmas)
1153                                 vmas[i] = vma;
1154                         i++;
1155                         start += PAGE_SIZE;
1156                         len--;
1157                 } while (len && start < vma->vm_end);
1158         } while (len);
1159         return i;
1160 }
1161 EXPORT_SYMBOL(get_user_pages);
1162 
1163 pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1164                         spinlock_t **ptl)
1165 {
1166         pgd_t * pgd = pgd_offset(mm, addr);
1167         pud_t * pud = pud_alloc(mm, pgd, addr);
1168         if (pud) {
1169                 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1170                 if (pmd)
1171                         return pte_alloc_map_lock(mm, pmd, addr, ptl);
1172         }
1173         return NULL;
1174 }
1175 
1176 /*
1177  * This is the old fallback for page remapping.
1178  *
1179  * For historical reasons, it only allows reserved pages. Only
1180  * old drivers should use this, and they needed to mark their
1181  * pages reserved for the old functions anyway.
1182  */
1183 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1184 {
1185         int retval;
1186         pte_t *pte;
1187         spinlock_t *ptl;
1188 
1189         retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
1190         if (retval)
1191                 goto out;
1192 
1193         retval = -EINVAL;
1194         if (PageAnon(page))
1195                 goto out_uncharge;
1196         retval = -ENOMEM;
1197         flush_dcache_page(page);
1198         pte = get_locked_pte(mm, addr, &ptl);
1199         if (!pte)
1200                 goto out_uncharge;
1201         retval = -EBUSY;
1202         if (!pte_none(*pte))
1203                 goto out_unlock;
1204 
1205         /* Ok, finally just insert the thing.. */
1206         get_page(page);
1207         inc_mm_counter(mm, file_rss);
1208         page_add_file_rmap(page);
1209         set_pte_at(mm, addr, pte, mk_pte(page, prot));
1210 
1211         retval = 0;
1212         pte_unmap_unlock(pte, ptl);
1213         return retval;
1214 out_unlock:
1215         pte_unmap_unlock(pte, ptl);
1216 out_uncharge:
1217         mem_cgroup_uncharge_page(page);
1218 out:
1219         return retval;
1220 }
1221 
1222 /**
1223  * vm_insert_page - insert single page into user vma
1224  * @vma: user vma to map to
1225  * @addr: target user address of this page
1226  * @page: source kernel page
1227  *
1228  * This allows drivers to insert individual pages they've allocated
1229  * into a user vma.
1230  *
1231  * The page has to be a nice clean _individual_ kernel allocation.
1232  * If you allocate a compound page, you need to have marked it as
1233  * such (__GFP_COMP), or manually just split the page up yourself
1234  * (see split_page()).
1235  *
1236  * NOTE! Traditionally this was done with "remap_pfn_range()" which
1237  * took an arbitrary page protection parameter. This doesn't allow
1238  * that. Your vma protection will have to be set up correctly, which
1239  * means that if you want a shared writable mapping, you'd better
1240  * ask for a shared writable mapping!
1241  *
1242  * The page does not need to be reserved.
1243  */
1244 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1245 {
1246         if (addr < vma->vm_start || addr >= vma->vm_end)
1247                 return -EFAULT;
1248         if (!page_count(page))
1249                 return -EINVAL;
1250         vma->vm_flags |= VM_INSERTPAGE;
1251         return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1252 }
1253 EXPORT_SYMBOL(vm_insert_page);
1254 
1255 /**
1256  * vm_insert_pfn - insert single pfn into user vma
1257  * @vma: user vma to map to
1258  * @addr: target user address of this page
1259  * @pfn: source kernel pfn
1260  *
1261  * Similar to vm_inert_page, this allows drivers to insert individual pages
1262  * they've allocated into a user vma. Same comments apply.
1263  *
1264  * This function should only be called from a vm_ops->fault handler, and
1265  * in that case the handler should return NULL.
1266  */
1267 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1268                 unsigned long pfn)
1269 {
1270         struct mm_struct *mm = vma->vm_mm;
1271         int retval;
1272         pte_t *pte, entry;
1273         spinlock_t *ptl;
1274 
1275         BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1276         BUG_ON(is_cow_mapping(vma->vm_flags));
1277 
1278         retval = -ENOMEM;
1279         pte = get_locked_pte(mm, addr, &ptl);
1280         if (!pte)
1281                 goto out;
1282         retval = -EBUSY;
1283         if (!pte_none(*pte))
1284                 goto out_unlock;
1285 
1286         /* Ok, finally just insert the thing.. */
1287         entry = pfn_pte(pfn, vma->vm_page_prot);
1288         set_pte_at(mm, addr, pte, entry);
1289         update_mmu_cache(vma, addr, entry);
1290 
1291         retval = 0;
1292 out_unlock:
1293         pte_unmap_unlock(pte, ptl);
1294 
1295 out:
1296         return retval;
1297 }
1298 EXPORT_SYMBOL(vm_insert_pfn);
1299 
1300 /*
1301  * maps a range of physical memory into the requested pages. the old
1302  * mappings are removed. any references to nonexistent pages results
1303  * in null mappings (currently treated as "copy-on-access")
1304  */
1305 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1306                         unsigned long addr, unsigned long end,
1307                         unsigned long pfn, pgprot_t prot)
1308 {
1309         pte_t *pte;
1310         spinlock_t *ptl;
1311 
1312         pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1313         if (!pte)
1314                 return -ENOMEM;
1315         arch_enter_lazy_mmu_mode();
1316         do {
1317                 BUG_ON(!pte_none(*pte));
1318                 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1319                 pfn++;
1320         } while (pte++, addr += PAGE_SIZE, addr != end);
1321         arch_leave_lazy_mmu_mode();
1322         pte_unmap_unlock(pte - 1, ptl);
1323         return 0;
1324 }
1325 
1326 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1327                         unsigned long addr, unsigned long end,
1328                         unsigned long pfn, pgprot_t prot)
1329 {
1330         pmd_t *pmd;
1331         unsigned long next;
1332 
1333         pfn -= addr >> PAGE_SHIFT;
1334         pmd = pmd_alloc(mm, pud, addr);
1335         if (!pmd)
1336                 return -ENOMEM;
1337         do {
1338                 next = pmd_addr_end(addr, end);
1339                 if (remap_pte_range(mm, pmd, addr, next,
1340                                 pfn + (addr >> PAGE_SHIFT), prot))
1341                         return -ENOMEM;
1342         } while (pmd++, addr = next, addr != end);
1343         return 0;
1344 }
1345 
1346 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1347                         unsigned long addr, unsigned long end,
1348                         unsigned long pfn, pgprot_t prot)
1349 {
1350         pud_t *pud;
1351         unsigned long next;
1352 
1353         pfn -= addr >> PAGE_SHIFT;
1354         pud = pud_alloc(mm, pgd, addr);
1355         if (!pud)
1356                 return -ENOMEM;
1357         do {
1358                 next = pud_addr_end(addr, end);
1359                 if (remap_pmd_range(mm, pud, addr, next,
1360                                 pfn + (addr >> PAGE_SHIFT), prot))
1361                         return -ENOMEM;
1362         } while (pud++, addr = next, addr != end);
1363         return 0;
1364 }
1365 
1366 /**
1367  * remap_pfn_range - remap kernel memory to userspace
1368  * @vma: user vma to map to
1369  * @addr: target user address to start at
1370  * @pfn: physical address of kernel memory
1371  * @size: size of map area
1372  * @prot: page protection flags for this mapping
1373  *
1374  *  Note: this is only safe if the mm semaphore is held when called.
1375  */
1376 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1377                     unsigned long pfn, unsigned long size, pgprot_t prot)
1378 {
1379         pgd_t *pgd;
1380         unsigned long next;
1381         unsigned long end = addr + PAGE_ALIGN(size);
1382         struct mm_struct *mm = vma->vm_mm;
1383         int err;
1384 
1385         /*
1386          * Physically remapped pages are special. Tell the
1387          * rest of the world about it:
1388          *   VM_IO tells people not to look at these pages
1389          *      (accesses can have side effects).
1390          *   VM_RESERVED is specified all over the place, because
1391          *      in 2.4 it kept swapout's vma scan off this vma; but
1392          *      in 2.6 the LRU scan won't even find its pages, so this
1393          *      flag means no more than count its pages in reserved_vm,
1394          *      and omit it from core dump, even when VM_IO turned off.
1395          *   VM_PFNMAP tells the core MM that the base pages are just
1396          *      raw PFN mappings, and do not have a "struct page" associated
1397          *      with them.
1398          *
1399          * There's a horrible special case to handle copy-on-write
1400          * behaviour that some programs depend on. We mark the "original"
1401          * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1402          */
1403         if (is_cow_mapping(vma->vm_flags)) {
1404                 if (addr != vma->vm_start || end != vma->vm_end)
1405                         return -EINVAL;
1406                 vma->vm_pgoff = pfn;
1407         }
1408 
1409         vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1410 
1411         BUG_ON(addr >= end);
1412         pfn -= addr >> PAGE_SHIFT;
1413         pgd = pgd_offset(mm, addr);
1414         flush_cache_range(vma, addr, end);
1415         do {
1416                 next = pgd_addr_end(addr, end);
1417                 err = remap_pud_range(mm, pgd, addr, next,
1418                                 pfn + (addr >> PAGE_SHIFT), prot);
1419                 if (err)
1420                         break;
1421         } while (pgd++, addr = next, addr != end);
1422         return err;
1423 }
1424 EXPORT_SYMBOL(remap_pfn_range);
1425 
1426 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1427                                      unsigned long addr, unsigned long end,
1428                                      pte_fn_t fn, void *data)
1429 {
1430         pte_t *pte;
1431         int err;
1432         pgtable_t token;
1433         spinlock_t *uninitialized_var(ptl);
1434 
1435         pte = (mm == &init_mm) ?
1436                 pte_alloc_kernel(pmd, addr) :
1437                 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1438         if (!pte)
1439                 return -ENOMEM;
1440 
1441         BUG_ON(pmd_huge(*pmd));
1442 
1443         token = pmd_pgtable(*pmd);
1444 
1445         do {
1446                 err = fn(pte, token, addr, data);
1447                 if (err)
1448                         break;
1449         } while (pte++, addr += PAGE_SIZE, addr != end);
1450 
1451         if (mm != &init_mm)
1452                 pte_unmap_unlock(pte-1, ptl);
1453         return err;
1454 }
1455 
1456 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1457                                      unsigned long addr, unsigned long end,
1458                                      pte_fn_t fn, void *data)
1459 {
1460         pmd_t *pmd;
1461         unsigned long next;
1462         int err;
1463 
1464         pmd = pmd_alloc(mm, pud, addr);
1465         if (!pmd)
1466                 return -ENOMEM;
1467         do {
1468                 next = pmd_addr_end(addr, end);
1469                 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1470                 if (err)
1471                         break;
1472         } while (pmd++, addr = next, addr != end);
1473         return err;
1474 }
1475 
1476 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1477                                      unsigned long addr, unsigned long end,
1478                                      pte_fn_t fn, void *data)
1479 {
1480         pud_t *pud;
1481         unsigned long next;
1482         int err;
1483 
1484         pud = pud_alloc(mm, pgd, addr);
1485         if (!pud)
1486                 return -ENOMEM;
1487         do {
1488                 next = pud_addr_end(addr, end);
1489                 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1490                 if (err)
1491                         break;
1492         } while (pud++, addr = next, addr != end);
1493         return err;
1494 }
1495 
1496 /*
1497  * Scan a region of virtual memory, filling in page tables as necessary
1498  * and calling a provided function on each leaf page table.
1499  */
1500 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1501                         unsigned long size, pte_fn_t fn, void *data)
1502 {
1503         pgd_t *pgd;
1504         unsigned long next;
1505         unsigned long end = addr + size;
1506         int err;
1507 
1508         BUG_ON(addr >= end);
1509         pgd = pgd_offset(mm, addr);
1510         do {
1511                 next = pgd_addr_end(addr, end);
1512                 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1513                 if (err)
1514                         break;
1515         } while (pgd++, addr = next, addr != end);
1516         return err;
1517 }
1518 EXPORT_SYMBOL_GPL(apply_to_page_range);
1519 
1520 /*
1521  * handle_pte_fault chooses page fault handler according to an entry
1522  * which was read non-atomically.  Before making any commitment, on
1523  * those architectures or configurations (e.g. i386 with PAE) which
1524  * might give a mix of unmatched parts, do_swap_page and do_file_page
1525  * must check under lock before unmapping the pte and proceeding
1526  * (but do_wp_page is only called after already making such a check;
1527  * and do_anonymous_page and do_no_page can safely check later on).
1528  */
1529 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1530                                 pte_t *page_table, pte_t orig_pte)
1531 {
1532         int same = 1;
1533 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1534         if (sizeof(pte_t) > sizeof(unsigned long)) {
1535                 spinlock_t *ptl = pte_lockptr(mm, pmd);
1536                 spin_lock(ptl);
1537                 same = pte_same(*page_table, orig_pte);
1538                 spin_unlock(ptl);
1539         }
1540 #endif
1541         pte_unmap(page_table);
1542         return same;
1543 }
1544 
1545 /*
1546  * Do pte_mkwrite, but only if the vma says VM_WRITE.  We do this when
1547  * servicing faults for write access.  In the normal case, do always want
1548  * pte_mkwrite.  But get_user_pages can cause write faults for mappings
1549  * that do not have writing enabled, when used by access_process_vm.
1550  */
1551 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1552 {
1553         if (likely(vma->vm_flags & VM_WRITE))
1554                 pte = pte_mkwrite(pte);
1555         return pte;
1556 }
1557 
1558 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1559 {
1560         /*
1561          * If the source page was a PFN mapping, we don't have
1562          * a "struct page" for it. We do a best-effort copy by
1563          * just copying from the original user address. If that
1564          * fails, we just zero-fill it. Live with it.
1565          */
1566         if (unlikely(!src)) {
1567                 void *kaddr = kmap_atomic(dst, KM_USER0);
1568                 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1569 
1570                 /*
1571                  * This really shouldn't fail, because the page is there
1572                  * in the page tables. But it might just be unreadable,
1573                  * in which case we just give up and fill the result with
1574                  * zeroes.
1575                  */
1576                 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1577                         memset(kaddr, 0, PAGE_SIZE);
1578                 kunmap_atomic(kaddr, KM_USER0);
1579                 flush_dcache_page(dst);
1580         } else
1581                 copy_user_highpage(dst, src, va, vma);
1582 }
1583 
1584 /*
1585  * This routine handles present pages, when users try to write
1586  * to a shared page. It is done by copying the page to a new address
1587  * and decrementing the shared-page counter for the old page.
1588  *
1589  * Note that this routine assumes that the protection checks have been
1590  * done by the caller (the low-level page fault routine in most cases).
1591  * Thus we can safely just mark it writable once we've done any necessary
1592  * COW.
1593  *
1594  * We also mark the page dirty at this point even though the page will
1595  * change only once the write actually happens. This avoids a few races,
1596  * and potentially makes it more efficient.
1597  *
1598  * We enter with non-exclusive mmap_sem (to exclude vma changes,
1599  * but allow concurrent faults), with pte both mapped and locked.
1600  * We return with mmap_sem still held, but pte unmapped and unlocked.
1601  */
1602 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1603                 unsigned long address, pte_t *page_table, pmd_t *pmd,
1604                 spinlock_t *ptl, pte_t orig_pte)
1605 {
1606         struct page *old_page, *new_page;
1607         pte_t entry;
1608         int reuse = 0, ret = 0;
1609         int page_mkwrite = 0;
1610         struct page *dirty_page = NULL;
1611 
1612         old_page = vm_normal_page(vma, address, orig_pte);
1613         if (!old_page)
1614                 goto gotten;
1615 
1616         /*
1617          * Take out anonymous pages first, anonymous shared vmas are
1618          * not dirty accountable.
1619          */
1620         if (PageAnon(old_page)) {
1621                 if (!TestSetPageLocked(old_page)) {
1622                         reuse = can_share_swap_page(old_page);
1623                         unlock_page(old_page);
1624                 }
1625         } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1626                                         (VM_WRITE|VM_SHARED))) {
1627                 /*
1628                  * Only catch write-faults on shared writable pages,
1629                  * read-only shared pages can get COWed by
1630                  * get_user_pages(.write=1, .force=1).
1631                  */
1632                 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1633                         /*
1634                          * Notify the address space that the page is about to
1635                          * become writable so that it can prohibit this or wait
1636                          * for the page to get into an appropriate state.
1637                          *
1638                          * We do this without the lock held, so that it can
1639                          * sleep if it needs to.
1640                          */
1641                         page_cache_get(old_page);
1642                         pte_unmap_unlock(page_table, ptl);
1643 
1644                         if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1645                                 goto unwritable_page;
1646 
1647                         /*
1648                          * Since we dropped the lock we need to revalidate
1649                          * the PTE as someone else may have changed it.  If
1650                          * they did, we just return, as we can count on the
1651                          * MMU to tell us if they didn't also make it writable.
1652                          */
1653                         page_table = pte_offset_map_lock(mm, pmd, address,
1654                                                          &ptl);
1655                         page_cache_release(old_page);
1656                         if (!pte_same(*page_table, orig_pte))
1657                                 goto unlock;
1658 
1659                         page_mkwrite = 1;
1660                 }
1661                 dirty_page = old_page;
1662                 get_page(dirty_page);
1663                 reuse = 1;
1664         }
1665 
1666         if (reuse) {
1667                 flush_cache_page(vma, address, pte_pfn(orig_pte));
1668                 entry = pte_mkyoung(orig_pte);
1669                 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1670                 if (ptep_set_access_flags(vma, address, page_table, entry,1))
1671                         update_mmu_cache(vma, address, entry);
1672                 ret |= VM_FAULT_WRITE;
1673                 goto unlock;
1674         }
1675 
1676         /*
1677          * Ok, we need to copy. Oh, well..
1678          */
1679         page_cache_get(old_page);
1680 gotten:
1681         pte_unmap_unlock(page_table, ptl);
1682 
1683         if (unlikely(anon_vma_prepare(vma)))
1684                 goto oom;
1685         VM_BUG_ON(old_page == ZERO_PAGE(0));
1686         new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1687         if (!new_page)
1688                 goto oom;
1689         cow_user_page(new_page, old_page, address, vma);
1690         __SetPageUptodate(new_page);
1691 
1692         if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
1693                 goto oom_free_new;
1694 
1695         /*
1696          * Re-check the pte - we dropped the lock
1697          */
1698         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1699         if (likely(pte_same(*page_table, orig_pte))) {
1700                 if (old_page) {
1701                         page_remove_rmap(old_page, vma);
1702                         if (!PageAnon(old_page)) {
1703                                 dec_mm_counter(mm, file_rss);
1704                                 inc_mm_counter(mm, anon_rss);
1705                         }
1706                 } else
1707                         inc_mm_counter(mm, anon_rss);
1708                 flush_cache_page(vma, address, pte_pfn(orig_pte));
1709                 entry = mk_pte(new_page, vma->vm_page_prot);
1710                 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1711                 /*
1712                  * Clear the pte entry and flush it first, before updating the
1713                  * pte with the new entry. This will avoid a race condition
1714                  * seen in the presence of one thread doing SMC and another
1715                  * thread doing COW.
1716                  */
1717                 ptep_clear_flush(vma, address, page_table);
1718                 set_pte_at(mm, address, page_table, entry);
1719                 update_mmu_cache(vma, address, entry);
1720                 lru_cache_add_active(new_page);
1721                 page_add_new_anon_rmap(new_page, vma, address);
1722 
1723                 /* Free the old page.. */
1724                 new_page = old_page;
1725                 ret |= VM_FAULT_WRITE;
1726         } else
1727                 mem_cgroup_uncharge_page(new_page);
1728 
1729         if (new_page)
1730                 page_cache_release(new_page);
1731         if (old_page)
1732                 page_cache_release(old_page);
1733 unlock:
1734         pte_unmap_unlock(page_table, ptl);
1735         if (dirty_page) {
1736                 if (vma->vm_file)
1737                         file_update_time(vma->vm_file);
1738 
1739                 /*
1740                  * Yes, Virginia, this is actually required to prevent a race
1741                  * with clear_page_dirty_for_io() from clearing the page dirty
1742                  * bit after it clear all dirty ptes, but before a racing
1743                  * do_wp_page installs a dirty pte.
1744                  *
1745                  * do_no_page is protected similarly.
1746                  */
1747                 wait_on_page_locked(dirty_page);
1748                 set_page_dirty_balance(dirty_page, page_mkwrite);
1749                 put_page(dirty_page);
1750         }
1751         return ret;
1752 oom_free_new:
1753         page_cache_release(new_page);
1754 oom:
1755         if (old_page)
1756                 page_cache_release(old_page);
1757         return VM_FAULT_OOM;
1758 
1759 unwritable_page:
1760         page_cache_release(old_page);
1761         return VM_FAULT_SIGBUS;
1762 }
1763 
1764 /*
1765  * Helper functions for unmap_mapping_range().
1766  *
1767  * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1768  *
1769  * We have to restart searching the prio_tree whenever we drop the lock,
1770  * since the iterator is only valid while the lock is held, and anyway
1771  * a later vma might be split and reinserted earlier while lock dropped.
1772  *
1773  * The list of nonlinear vmas could be handled more efficiently, using
1774  * a placeholder, but handle it in the same way until a need is shown.
1775  * It is important to search the prio_tree before nonlinear list: a vma
1776  * may become nonlinear and be shifted from prio_tree to nonlinear list
1777  * while the lock is dropped; but never shifted from list to prio_tree.
1778  *
1779  * In order to make forward progress despite restarting the search,
1780  * vm_truncate_count is used to mark a vma as now dealt with, so we can
1781  * quickly skip it next time around.  Since the prio_tree search only
1782  * shows us those vmas affected by unmapping the range in question, we
1783  * can't efficiently keep all vmas in step with mapping->truncate_count:
1784  * so instead reset them all whenever it wraps back to 0 (then go to 1).
1785  * mapping->truncate_count and vma->vm_truncate_count are protected by
1786  * i_mmap_lock.
1787  *
1788  * In order to make forward progress despite repeatedly restarting some
1789  * large vma, note the restart_addr from unmap_vmas when it breaks out:
1790  * and restart from that address when we reach that vma again.  It might
1791  * have been split or merged, shrunk or extended, but never shifted: so
1792  * restart_addr remains valid so long as it remains in the vma's range.
1793  * unmap_mapping_range forces truncate_count to leap over page-aligned
1794  * values so we can save vma's restart_addr in its truncate_count field.
1795  */
1796 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1797 
1798 static void reset_vma_truncate_counts(struct address_space *mapping)
1799 {
1800         struct vm_area_struct *vma;
1801         struct prio_tree_iter iter;
1802 
1803         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1804                 vma->vm_truncate_count = 0;
1805         list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1806                 vma->vm_truncate_count = 0;
1807 }
1808 
1809 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1810                 unsigned long start_addr, unsigned long end_addr,
1811                 struct zap_details *details)
1812 {
1813         unsigned long restart_addr;
1814         int need_break;
1815 
1816         /*
1817          * files that support invalidating or truncating portions of the
1818          * file from under mmaped areas must have their ->fault function
1819          * return a locked page (and set VM_FAULT_LOCKED in the return).
1820          * This provides synchronisation against concurrent unmapping here.
1821          */
1822 
1823 again:
1824         restart_addr = vma->vm_truncate_count;
1825         if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1826                 start_addr = restart_addr;
1827                 if (start_addr >= end_addr) {
1828                         /* Top of vma has been split off since last time */
1829                         vma->vm_truncate_count = details->truncate_count;
1830                         return 0;
1831                 }
1832         }
1833 
1834         restart_addr = zap_page_range(vma, start_addr,
1835                                         end_addr - start_addr, details);
1836         need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
1837 
1838         if (restart_addr >= end_addr) {
1839                 /* We have now completed this vma: mark it so */
1840                 vma->vm_truncate_count = details->truncate_count;
1841                 if (!need_break)
1842                         return 0;
1843         } else {
1844                 /* Note restart_addr in vma's truncate_count field */
1845                 vma->vm_truncate_count = restart_addr;
1846                 if (!need_break)
1847                         goto again;
1848         }
1849 
1850         spin_unlock(details->i_mmap_lock);
1851         cond_resched();
1852         spin_lock(details->i_mmap_lock);
1853         return -EINTR;
1854 }
1855 
1856 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1857                                             struct zap_details *details)
1858 {
1859         struct vm_area_struct *vma;
1860         struct prio_tree_iter iter;
1861         pgoff_t vba, vea, zba, zea;
1862 
1863 restart:
1864         vma_prio_tree_foreach(vma, &iter, root,
1865                         details->first_index, details->last_index) {
1866                 /* Skip quickly over those we have already dealt with */
1867                 if (vma->vm_truncate_count == details->truncate_count)
1868                         continue;
1869 
1870                 vba = vma->vm_pgoff;
1871                 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1872                 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1873                 zba = details->first_index;
1874                 if (zba < vba)
1875                         zba = vba;
1876                 zea = details->last_index;
1877                 if (zea > vea)
1878                         zea = vea;
1879 
1880                 if (unmap_mapping_range_vma(vma,
1881                         ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1882                         ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1883                                 details) < 0)
1884                         goto restart;
1885         }
1886 }
1887 
1888 static inline void unmap_mapping_range_list(struct list_head *head,
1889                                             struct zap_details *details)
1890 {
1891         struct vm_area_struct *vma;
1892 
1893         /*
1894          * In nonlinear VMAs there is no correspondence between virtual address
1895          * offset and file offset.  So we must perform an exhaustive search
1896          * across *all* the pages in each nonlinear VMA, not just the pages
1897          * whose virtual address lies outside the file truncation point.
1898          */
1899 restart:
1900         list_for_each_entry(vma, head, shared.vm_set.list) {
1901                 /* Skip quickly over those we have already dealt with */
1902                 if (vma->vm_truncate_count == details->truncate_count)
1903                         continue;
1904                 details->nonlinear_vma = vma;
1905                 if (unmap_mapping_range_vma(vma, vma->vm_start,
1906                                         vma->vm_end, details) < 0)
1907                         goto restart;
1908         }
1909 }
1910 
1911 /**
1912  * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
1913  * @mapping: the address space containing mmaps to be unmapped.
1914  * @holebegin: byte in first page to unmap, relative to the start of
1915  * the underlying file.  This will be rounded down to a PAGE_SIZE
1916  * boundary.  Note that this is different from vmtruncate(), which
1917  * must keep the partial page.  In contrast, we must get rid of
1918  * partial pages.
1919  * @holelen: size of prospective hole in bytes.  This will be rounded
1920  * up to a PAGE_SIZE boundary.  A holelen of zero truncates to the
1921  * end of the file.
1922  * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1923  * but 0 when invalidating pagecache, don't throw away private data.
1924  */
1925 void unmap_mapping_range(struct address_space *mapping,
1926                 loff_t const holebegin, loff_t const holelen, int even_cows)
1927 {
1928         struct zap_details details;
1929         pgoff_t hba = holebegin >> PAGE_SHIFT;
1930         pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1931 
1932         /* Check for overflow. */
1933         if (sizeof(holelen) > sizeof(hlen)) {
1934                 long long holeend =
1935                         (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1936                 if (holeend & ~(long long)ULONG_MAX)
1937                         hlen = ULONG_MAX - hba + 1;
1938         }
1939 
1940         details.check_mapping = even_cows? NULL: mapping;
1941         details.nonlinear_vma = NULL;
1942         details.first_index = hba;
1943         details.last_index = hba + hlen - 1;
1944         if (details.last_index < details.first_index)
1945                 details.last_index = ULONG_MAX;
1946         details.i_mmap_lock = &mapping->i_mmap_lock;
1947 
1948         spin_lock(&mapping->i_mmap_lock);
1949 
1950         /* Protect against endless unmapping loops */
1951         mapping->truncate_count++;
1952         if (unlikely(is_restart_addr(mapping->truncate_count))) {
1953                 if (mapping->truncate_count == 0)
1954                         reset_vma_truncate_counts(mapping);
1955                 mapping->truncate_count++;
1956         }
1957         details.truncate_count = mapping->truncate_count;
1958 
1959         if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1960                 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1961         if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1962                 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1963         spin_unlock(&mapping->i_mmap_lock);
1964 }
1965 EXPORT_SYMBOL(unmap_mapping_range);
1966 
1967 /**
1968  * vmtruncate - unmap mappings "freed" by truncate() syscall
1969  * @inode: inode of the file used
1970  * @offset: file offset to start truncating
1971  *
1972  * NOTE! We have to be ready to update the memory sharing
1973  * between the file and the memory map for a potential last
1974  * incomplete page.  Ugly, but necessary.
1975  */
1976 int vmtruncate(struct inode * inode, loff_t offset)
1977 {
1978         if (inode->i_size < offset) {
1979                 unsigned long limit;
1980 
1981                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1982                 if (limit != RLIM_INFINITY && offset > limit)
1983                         goto out_sig;
1984                 if (offset > inode->i_sb->s_maxbytes)
1985                         goto out_big;
1986                 i_size_write(inode, offset);
1987         } else {
1988                 struct address_space *mapping = inode->i_mapping;
1989 
1990                 /*
1991                  * truncation of in-use swapfiles is disallowed - it would
1992                  * cause subsequent swapout to scribble on the now-freed
1993                  * blocks.
1994                  */
1995                 if (IS_SWAPFILE(inode))
1996                         return -ETXTBSY;
1997                 i_size_write(inode, offset);
1998 
1999                 /*
2000                  * unmap_mapping_range is called twice, first simply for
2001                  * efficiency so that truncate_inode_pages does fewer
2002                  * single-page unmaps.  However after this first call, and
2003                  * before truncate_inode_pages finishes, it is possible for
2004                  * private pages to be COWed, which remain after
2005                  * truncate_inode_pages finishes, hence the second
2006                  * unmap_mapping_range call must be made for correctness.
2007                  */
2008                 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2009                 truncate_inode_pages(mapping, offset);
2010                 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2011         }
2012 
2013         if (inode->i_op && inode->i_op->truncate)
2014                 inode->i_op->truncate(inode);
2015         return 0;
2016 
2017 out_sig:
2018         send_sig(SIGXFSZ, current, 0);
2019 out_big:
2020         return -EFBIG;
2021 }
2022 EXPORT_SYMBOL(vmtruncate);
2023 
2024 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2025 {
2026         struct address_space *mapping = inode->i_mapping;
2027 
2028         /*
2029          * If the underlying filesystem is not going to provide
2030          * a way to truncate a range of blocks (punch a hole) -
2031          * we should return failure right now.
2032          */
2033         if (!inode->i_op || !inode->i_op->truncate_range)
2034                 return -ENOSYS;
2035 
2036         mutex_lock(&inode->i_mutex);
2037         down_write(&inode->i_alloc_sem);
2038         unmap_mapping_range(mapping, offset, (end - offset), 1);
2039         truncate_inode_pages_range(mapping, offset, end);
2040         unmap_mapping_range(mapping, offset, (end - offset), 1);
2041         inode->i_op->truncate_range(inode, offset, end);
2042         up_write(&inode->i_alloc_sem);
2043         mutex_unlock(&inode->i_mutex);
2044 
2045         return 0;
2046 }
2047 
2048 /*
2049  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2050  * but allow concurrent faults), and pte mapped but not yet locked.
2051  * We return with mmap_sem still held, but pte unmapped and unlocked.
2052  */
2053 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2054                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2055                 int write_access, pte_t orig_pte)
2056 {
2057         spinlock_t *ptl;
2058         struct page *page;
2059         swp_entry_t entry;
2060         pte_t pte;
2061         int ret = 0;
2062 
2063         if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2064                 goto out;
2065 
2066         entry = pte_to_swp_entry(orig_pte);
2067         if (is_migration_entry(entry)) {
2068                 migration_entry_wait(mm, pmd, address);
2069                 goto out;
2070         }
2071         delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2072         page = lookup_swap_cache(entry);
2073         if (!page) {
2074                 grab_swap_token(); /* Contend for token _before_ read-in */
2075                 page = swapin_readahead(entry,
2076                                         GFP_HIGHUSER_MOVABLE, vma, address);
2077                 if (!page) {
2078                         /*
2079                          * Back out if somebody else faulted in this pte
2080                          * while we released the pte lock.
2081                          */
2082                         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2083                         if (likely(pte_same(*page_table, orig_pte)))
2084                                 ret = VM_FAULT_OOM;
2085                         delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2086                         goto unlock;
2087                 }
2088 
2089                 /* Had to read the page from swap area: Major fault */
2090                 ret = VM_FAULT_MAJOR;
2091                 count_vm_event(PGMAJFAULT);
2092         }
2093 
2094         if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2095                 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2096                 ret = VM_FAULT_OOM;
2097                 goto out;
2098         }
2099 
2100         mark_page_accessed(page);
2101         lock_page(page);
2102         delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2103 
2104         /*
2105          * Back out if somebody else already faulted in this pte.
2106          */
2107         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2108         if (unlikely(!pte_same(*page_table, orig_pte)))
2109                 goto out_nomap;
2110 
2111         if (unlikely(!PageUptodate(page))) {
2112                 ret = VM_FAULT_SIGBUS;
2113                 goto out_nomap;
2114         }
2115 
2116         /* The page isn't present yet, go ahead with the fault. */
2117 
2118         inc_mm_counter(mm, anon_rss);
2119         pte = mk_pte(page, vma->vm_page_prot);
2120         if (write_access && can_share_swap_page(page)) {
2121                 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2122                 write_access = 0;
2123         }
2124 
2125         flush_icache_page(vma, page);
2126         set_pte_at(mm, address, page_table, pte);
2127         page_add_anon_rmap(page, vma, address);
2128 
2129         swap_free(entry);
2130         if (vm_swap_full())
2131                 remove_exclusive_swap_page(page);
2132         unlock_page(page);
2133 
2134         if (write_access) {
2135                 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2136                 if (ret & VM_FAULT_ERROR)
2137                         ret &= VM_FAULT_ERROR;
2138                 goto out;
2139         }
2140 
2141         /* No need to invalidate - it was non-present before */
2142         update_mmu_cache(vma, address, pte);
2143 unlock:
2144         pte_unmap_unlock(page_table, ptl);
2145 out:
2146         return ret;
2147 out_nomap:
2148         mem_cgroup_uncharge_page(page);
2149         pte_unmap_unlock(page_table, ptl);
2150         unlock_page(page);
2151         page_cache_release(page);
2152         return ret;
2153 }
2154 
2155 /*
2156  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2157  * but allow concurrent faults), and pte mapped but not yet locked.
2158  * We return with mmap_sem still held, but pte unmapped and unlocked.
2159  */
2160 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2161                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2162                 int write_access)
2163 {
2164         struct page *page;
2165         spinlock_t *ptl;
2166         pte_t entry;
2167 
2168         /* Allocate our own private page. */
2169         pte_unmap(page_table);
2170 
2171         if (unlikely(anon_vma_prepare(vma)))
2172                 goto oom;
2173         page = alloc_zeroed_user_highpage_movable(vma, address);
2174         if (!page)
2175                 goto oom;
2176         __SetPageUptodate(page);
2177 
2178         if (mem_cgroup_charge(page, mm, GFP_KERNEL))
2179                 goto oom_free_page;
2180 
2181         entry = mk_pte(page, vma->vm_page_prot);
2182         entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2183 
2184         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2185         if (!pte_none(*page_table))
2186                 goto release;
2187         inc_mm_counter(mm, anon_rss);
2188         lru_cache_add_active(page);
2189         page_add_new_anon_rmap(page, vma, address);
2190         set_pte_at(mm, address, page_table, entry);
2191 
2192         /* No need to invalidate - it was non-present before */
2193         update_mmu_cache(vma, address, entry);
2194 unlock:
2195         pte_unmap_unlock(page_table, ptl);
2196         return 0;
2197 release:
2198         mem_cgroup_uncharge_page(page);
2199         page_cache_release(page);
2200         goto unlock;
2201 oom_free_page:
2202         page_cache_release(page);
2203 oom:
2204         return VM_FAULT_OOM;
2205 }
2206 
2207 /*
2208  * __do_fault() tries to create a new page mapping. It aggressively
2209  * tries to share with existing pages, but makes a separate copy if
2210  * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2211  * the next page fault.
2212  *
2213  * As this is called only for pages that do not currently exist, we
2214  * do not need to flush old virtual caches or the TLB.
2215  *
2216  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2217  * but allow concurrent faults), and pte neither mapped nor locked.
2218  * We return with mmap_sem still held, but pte unmapped and unlocked.
2219  */
2220 static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2221                 unsigned long address, pmd_t *pmd,
2222                 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2223 {
2224         pte_t *page_table;
2225         spinlock_t *ptl;
2226         struct page *page;
2227         pte_t entry;
2228         int anon = 0;
2229         struct page *dirty_page = NULL;
2230         struct vm_fault vmf;
2231         int ret;
2232         int page_mkwrite = 0;
2233 
2234         vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2235         vmf.pgoff = pgoff;
2236         vmf.flags = flags;
2237         vmf.page = NULL;
2238 
2239         BUG_ON(vma->vm_flags & VM_PFNMAP);
2240 
2241         if (likely(vma->vm_ops->fault)) {
2242                 ret = vma->vm_ops->fault(vma, &vmf);
2243                 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2244                         return ret;
2245         } else {
2246                 /* Legacy ->nopage path */
2247                 ret = 0;
2248                 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2249                 /* no page was available -- either SIGBUS or OOM */
2250                 if (unlikely(vmf.page == NOPAGE_SIGBUS))
2251                         return VM_FAULT_SIGBUS;
2252                 else if (unlikely(vmf.page == NOPAGE_OOM))
2253                         return VM_FAULT_OOM;
2254         }
2255 
2256         /*
2257          * For consistency in subsequent calls, make the faulted page always
2258          * locked.
2259          */
2260         if (unlikely(!(ret & VM_FAULT_LOCKED)))
2261                 lock_page(vmf.page);
2262         else
2263                 VM_BUG_ON(!PageLocked(vmf.page));
2264 
2265         /*
2266          * Should we do an early C-O-W break?
2267          */
2268         page = vmf.page;
2269         if (flags & FAULT_FLAG_WRITE) {
2270                 if (!(vma->vm_flags & VM_SHARED)) {
2271                         anon = 1;
2272                         if (unlikely(anon_vma_prepare(vma))) {
2273                                 ret = VM_FAULT_OOM;
2274                                 goto out;
2275                         }
2276                         page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2277                                                 vma, address);
2278                         if (!page) {
2279                                 ret = VM_FAULT_OOM;
2280                                 goto out;
2281                         }
2282                         copy_user_highpage(page, vmf.page, address, vma);
2283                         __SetPageUptodate(page);
2284                 } else {
2285                         /*
2286                          * If the page will be shareable, see if the backing
2287                          * address space wants to know that the page is about
2288                          * to become writable
2289                          */
2290                         if (vma->vm_ops->page_mkwrite) {
2291                                 unlock_page(page);
2292                                 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
2293                                         ret = VM_FAULT_SIGBUS;
2294                                         anon = 1; /* no anon but release vmf.page */
2295                                         goto out_unlocked;
2296                                 }
2297                                 lock_page(page);
2298                                 /*
2299                                  * XXX: this is not quite right (racy vs
2300                                  * invalidate) to unlock and relock the page
2301                                  * like this, however a better fix requires
2302                                  * reworking page_mkwrite locking API, which
2303                                  * is better done later.
2304                                  */
2305                                 if (!page->mapping) {
2306                                         ret = 0;
2307                                         anon = 1; /* no anon but release vmf.page */
2308                                         goto out;
2309                                 }
2310                                 page_mkwrite = 1;
2311                         }
2312                 }
2313 
2314         }
2315 
2316         if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2317                 ret = VM_FAULT_OOM;
2318                 goto out;
2319         }
2320 
2321         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2322 
2323         /*
2324          * This silly early PAGE_DIRTY setting removes a race
2325          * due to the bad i386 page protection. But it's valid
2326          * for other architectures too.
2327          *
2328          * Note that if write_access is true, we either now have
2329          * an exclusive copy of the page, or this is a shared mapping,
2330          * so we can make it writable and dirty to avoid having to
2331          * handle that later.
2332          */
2333         /* Only go through if we didn't race with anybody else... */
2334         if (likely(pte_same(*page_table, orig_pte))) {
2335                 flush_icache_page(vma, page);
2336                 entry = mk_pte(page, vma->vm_page_prot);
2337                 if (flags & FAULT_FLAG_WRITE)
2338                         entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2339                 set_pte_at(mm, address, page_table, entry);
2340                 if (anon) {
2341                         inc_mm_counter(mm, anon_rss);
2342                         lru_cache_add_active(page);
2343                         page_add_new_anon_rmap(page, vma, address);
2344                 } else {
2345                         inc_mm_counter(mm, file_rss);
2346                         page_add_file_rmap(page);
2347                         if (flags & FAULT_FLAG_WRITE) {
2348                                 dirty_page = page;
2349                                 get_page(dirty_page);
2350                         }
2351                 }
2352 
2353                 /* no need to invalidate: a not-present page won't be cached */
2354                 update_mmu_cache(vma, address, entry);
2355         } else {
2356                 mem_cgroup_uncharge_page(page);
2357                 if (anon)
2358                         page_cache_release(page);
2359                 else
2360                         anon = 1; /* no anon but release faulted_page */
2361         }
2362 
2363         pte_unmap_unlock(page_table, ptl);
2364 
2365 out:
2366         unlock_page(vmf.page);
2367 out_unlocked:
2368         if (anon)
2369                 page_cache_release(vmf.page);
2370         else if (dirty_page) {
2371                 if (vma->vm_file)
2372                         file_update_time(vma->vm_file);
2373 
2374                 set_page_dirty_balance(dirty_page, page_mkwrite);
2375                 put_page(dirty_page);
2376         }
2377 
2378         return ret;
2379 }
2380 
2381 static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2382                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2383                 int write_access, pte_t orig_pte)
2384 {
2385         pgoff_t pgoff = (((address & PAGE_MASK)
2386                         - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2387         unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2388 
2389         pte_unmap(page_table);
2390         return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2391 }
2392 
2393 
2394 /*
2395  * do_no_pfn() tries to create a new page mapping for a page without
2396  * a struct_page backing it
2397  *
2398  * As this is called only for pages that do not currently exist, we
2399  * do not need to flush old virtual caches or the TLB.
2400  *
2401  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2402  * but allow concurrent faults), and pte mapped but not yet locked.
2403  * We return with mmap_sem still held, but pte unmapped and unlocked.
2404  *
2405  * It is expected that the ->nopfn handler always returns the same pfn
2406  * for a given virtual mapping.
2407  *
2408  * Mark this `noinline' to prevent it from bloating the main pagefault code.
2409  */
2410 static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2411                      unsigned long address, pte_t *page_table, pmd_t *pmd,
2412                      int write_access)
2413 {
2414         spinlock_t *ptl;
2415         pte_t entry;
2416         unsigned long pfn;
2417 
2418         pte_unmap(page_table);
2419         BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2420         BUG_ON(is_cow_mapping(vma->vm_flags));
2421 
2422         pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2423         if (unlikely(pfn == NOPFN_OOM))
2424                 return VM_FAULT_OOM;
2425         else if (unlikely(pfn == NOPFN_SIGBUS))
2426                 return VM_FAULT_SIGBUS;
2427         else if (unlikely(pfn == NOPFN_REFAULT))
2428                 return 0;
2429 
2430         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2431 
2432         /* Only go through if we didn't race with anybody else... */
2433         if (pte_none(*page_table)) {
2434                 entry = pfn_pte(pfn, vma->vm_page_prot);
2435                 if (write_access)
2436                         entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2437                 set_pte_at(mm, address, page_table, entry);
2438         }
2439         pte_unmap_unlock(page_table, ptl);
2440         return 0;
2441 }
2442 
2443 /*
2444  * Fault of a previously existing named mapping. Repopulate the pte
2445  * from the encoded file_pte if possible. This enables swappable
2446  * nonlinear vmas.
2447  *
2448  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2449  * but allow concurrent faults), and pte mapped but not yet locked.
2450  * We return with mmap_sem still held, but pte unmapped and unlocked.
2451  */
2452 static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2453                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2454                 int write_access, pte_t orig_pte)
2455 {
2456         unsigned int flags = FAULT_FLAG_NONLINEAR |
2457                                 (write_access ? FAULT_FLAG_WRITE : 0);
2458         pgoff_t pgoff;
2459 
2460         if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2461                 return 0;
2462 
2463         if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2464                         !(vma->vm_flags & VM_CAN_NONLINEAR))) {
2465                 /*
2466                  * Page table corrupted: show pte and kill process.
2467                  */
2468                 print_bad_pte(vma, orig_pte, address);
2469                 return VM_FAULT_OOM;
2470         }
2471 
2472         pgoff = pte_to_pgoff(orig_pte);
2473         return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2474 }
2475 
2476 /*
2477  * These routines also need to handle stuff like marking pages dirty
2478  * and/or accessed for architectures that don't do it in hardware (most
2479  * RISC architectures).  The early dirtying is also good on the i386.
2480  *
2481  * There is also a hook called "update_mmu_cache()" that architectures
2482  * with external mmu caches can use to update those (ie the Sparc or
2483  * PowerPC hashed page tables that act as extended TLBs).
2484  *
2485  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2486  * but allow concurrent faults), and pte mapped but not yet locked.
2487  * We return with mmap_sem still held, but pte unmapped and unlocked.
2488  */
2489 static inline int handle_pte_fault(struct mm_struct *mm,
2490                 struct vm_area_struct *vma, unsigned long address,
2491                 pte_t *pte, pmd_t *pmd, int write_access)
2492 {
2493         pte_t entry;
2494         spinlock_t *ptl;
2495 
2496         entry = *pte;
2497         if (!pte_present(entry)) {
2498                 if (pte_none(entry)) {
2499                         if (vma->vm_ops) {
2500                                 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2501                                         return do_linear_fault(mm, vma, address,
2502                                                 pte, pmd, write_access, entry);
2503                                 if (unlikely(vma->vm_ops->nopfn))
2504                                         return do_no_pfn(mm, vma, address, pte,
2505                                                          pmd, write_access);
2506                         }
2507                         return do_anonymous_page(mm, vma, address,
2508                                                  pte, pmd, write_access);
2509                 }
2510                 if (pte_file(entry))
2511                         return do_nonlinear_fault(mm, vma, address,
2512                                         pte, pmd, write_access, entry);
2513                 return do_swap_page(mm, vma, address,
2514                                         pte, pmd, write_access, entry);
2515         }
2516 
2517         ptl = pte_lockptr(mm, pmd);
2518         spin_lock(ptl);
2519         if (unlikely(!pte_same(*pte, entry)))
2520                 goto unlock;
2521         if (write_access) {
2522                 if (!pte_write(entry))
2523                         return do_wp_page(mm, vma, address,
2524                                         pte, pmd, ptl, entry);
2525                 entry = pte_mkdirty(entry);
2526         }
2527         entry = pte_mkyoung(entry);
2528         if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
2529                 update_mmu_cache(vma, address, entry);
2530         } else {
2531                 /*
2532                  * This is needed only for protection faults but the arch code
2533                  * is not yet telling us if this is a protection fault or not.
2534                  * This still avoids useless tlb flushes for .text page faults
2535                  * with threads.
2536                  */
2537                 if (write_access)
2538                         flush_tlb_page(vma, address);
2539         }
2540 unlock:
2541         pte_unmap_unlock(pte, ptl);
2542         return 0;
2543 }
2544 
2545 void pagefault_disable(void)
2546 {
2547         current->pagefault_disabled++;
2548         /*
2549          * make sure to have issued the store before a pagefault
2550          * can hit.
2551          */
2552         barrier();
2553 }
2554 EXPORT_SYMBOL(pagefault_disable);
2555 
2556 void pagefault_enable(void)
2557 {
2558         /*
2559          * make sure to issue those last loads/stores before enabling
2560          * the pagefault handler again.
2561          */
2562         barrier();
2563         current->pagefault_disabled--;
2564 }
2565 EXPORT_SYMBOL(pagefault_enable);
2566 
2567 /*
2568  * By the time we get here, we already hold the mm semaphore
2569  */
2570 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2571                 unsigned long address, int write_access)
2572 {
2573         pgd_t *pgd;
2574         pud_t *pud;
2575         pmd_t *pmd;
2576         pte_t *pte;
2577 
2578         __set_current_state(TASK_RUNNING);
2579 
2580         count_vm_event(PGFAULT);
2581 
2582         if (unlikely(is_vm_hugetlb_page(vma)))
2583                 return hugetlb_fault(mm, vma, address, write_access);
2584 
2585         pgd = pgd_offset(mm, address);
2586         pud = pud_alloc(mm, pgd, address);
2587         if (!pud)
2588                 return VM_FAULT_OOM;
2589         pmd = pmd_alloc(mm, pud, address);
2590         if (!pmd)
2591                 return VM_FAULT_OOM;
2592         pte = pte_alloc_map(mm, pmd, address);
2593         if (!pte)
2594                 return VM_FAULT_OOM;
2595 
2596         return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2597 }
2598 
2599 #ifndef __PAGETABLE_PUD_FOLDED
2600 /*
2601  * Allocate page upper directory.
2602  * We've already handled the fast-path in-line.
2603  */
2604 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2605 {
2606         pud_t *new = pud_alloc_one(mm, address);
2607         if (!new)
2608                 return -ENOMEM;
2609 
2610         spin_lock(&mm->page_table_lock);
2611         if (pgd_present(*pgd))          /* Another has populated it */
2612                 pud_free(mm, new);
2613         else
2614                 pgd_populate(mm, pgd, new);
2615         spin_unlock(&mm->page_table_lock);
2616         return 0;
2617 }
2618 #endif /* __PAGETABLE_PUD_FOLDED */
2619 
2620 #ifndef __PAGETABLE_PMD_FOLDED
2621 /*
2622  * Allocate page middle directory.
2623  * We've already handled the fast-path in-line.
2624  */
2625 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2626 {
2627         pmd_t *new = pmd_alloc_one(mm, address);
2628         if (!new)
2629                 return -ENOMEM;
2630 
2631         spin_lock(&mm->page_table_lock);
2632 #ifndef __ARCH_HAS_4LEVEL_HACK
2633         if (pud_present(*pud))          /* Another has populated it */
2634                 pmd_free(mm, new);
2635         else
2636                 pud_populate(mm, pud, new);
2637 #else
2638         if (pgd_present(*pud))          /* Another has populated it */
2639                 pmd_free(mm, new);
2640         else
2641                 pgd_populate(mm, pud, new);
2642 #endif /* __ARCH_HAS_4LEVEL_HACK */
2643         spin_unlock(&mm->page_table_lock);
2644         return 0;
2645 }
2646 #endif /* __PAGETABLE_PMD_FOLDED */
2647 
2648 int make_pages_present(unsigned long addr, unsigned long end)
2649 {
2650         int ret, len, write;
2651         struct vm_area_struct * vma;
2652 
2653         vma = find_vma(current->mm, addr);
2654         if (!vma)
2655                 return -1;
2656         write = (vma->vm_flags & VM_WRITE) != 0;
2657         BUG_ON(addr >= end);
2658         BUG_ON(end > vma->vm_end);
2659         len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
2660         ret = get_user_pages(current, current->mm, addr,
2661                         len, write, 0, NULL, NULL);
2662         if (ret < 0)
2663                 return ret;
2664         return ret == len ? 0 : -1;
2665 }
2666 
2667 #if !defined(__HAVE_ARCH_GATE_AREA)
2668 
2669 #if defined(AT_SYSINFO_EHDR)
2670 static struct vm_area_struct gate_vma;
2671 
2672 static int __init gate_vma_init(void)
2673 {
2674         gate_vma.vm_mm = NULL;
2675         gate_vma.vm_start = FIXADDR_USER_START;
2676         gate_vma.vm_end = FIXADDR_USER_END;
2677         gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2678         gate_vma.vm_page_prot = __P101;
2679         /*
2680          * Make sure the vDSO gets into every core dump.
2681          * Dumping its contents makes post-mortem fully interpretable later
2682          * without matching up the same kernel and hardware config to see
2683          * what PC values meant.
2684          */
2685         gate_vma.vm_flags |= VM_ALWAYSDUMP;
2686         return 0;
2687 }
2688 __initcall(gate_vma_init);
2689 #endif
2690 
2691 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2692 {
2693 #ifdef AT_SYSINFO_EHDR
2694         return &gate_vma;
2695 #else
2696         return NULL;
2697 #endif
2698 }
2699 
2700 int in_gate_area_no_task(unsigned long addr)
2701 {
2702 #ifdef AT_SYSINFO_EHDR
2703         if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2704                 return 1;
2705 #endif
2706         return 0;
2707 }
2708 
2709 #endif  /* __HAVE_ARCH_GATE_AREA */
2710 
2711 /*
2712  * Access another process' address space.
2713  * Source/target buffer must be kernel space,
2714  * Do not walk the page table directly, use get_user_pages
2715  */
2716 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2717 {
2718         struct mm_struct *mm;
2719         struct vm_area_struct *vma;
2720         struct page *page;
2721         void *old_buf = buf;
2722 
2723         mm = get_task_mm(tsk);
2724         if (!mm)
2725                 return 0;
2726 
2727         down_read(&mm->mmap_sem);
2728         /* ignore errors, just check how much was successfully transferred */
2729         while (len) {
2730                 int bytes, ret, offset;
2731                 void *maddr;
2732 
2733                 ret = get_user_pages(tsk, mm, addr, 1,
2734                                 write, 1, &page, &vma);
2735                 if (ret <= 0)
2736                         break;
2737 
2738                 bytes = len;
2739                 offset = addr & (PAGE_SIZE-1);
2740                 if (bytes > PAGE_SIZE-offset)
2741                         bytes = PAGE_SIZE-offset;
2742 
2743                 maddr = kmap(page);
2744                 if (write) {
2745                         copy_to_user_page(vma, page, addr,
2746                                           maddr + offset, buf, bytes);
2747                         set_page_dirty_lock(page);
2748                 } else {
2749                         copy_from_user_page(vma, page, addr,
2750                                             buf, maddr + offset, bytes);
2751                 }
2752                 kunmap(page);
2753                 page_cache_release(page);
2754                 len -= bytes;
2755                 buf += bytes;
2756                 addr += bytes;
2757         }
2758         up_read(&mm->mmap_sem);
2759         mmput(mm);
2760 
2761         return buf - old_buf;
2762 }
2763 
2764 /*
2765  * Print the name of a VMA.
2766  */
2767 void print_vma_addr(char *prefix, unsigned long ip)
2768 {
2769         struct mm_struct *mm = current->mm;
2770         struct vm_area_struct *vma;
2771 
2772         /*
2773          * Do not print if we are in atomic
2774          * contexts (in exception stacks, etc.):
2775          */
2776         if (preempt_count())
2777                 return;
2778 
2779         down_read(&mm->mmap_sem);
2780         vma = find_vma(mm, ip);
2781         if (vma && vma->vm_file) {
2782                 struct file *f = vma->vm_file;
2783                 char *buf = (char *)__get_free_page(GFP_KERNEL);
2784                 if (buf) {
2785                         char *p, *s;
2786 
2787                         p = d_path(&f->f_path, buf, PAGE_SIZE);
2788                         if (IS_ERR(p))
2789                                 p = "?";
2790                         s = strrchr(p, '/');
2791                         if (s)
2792                                 p = s+1;
2793                         printk("%s%s[%lx+%lx]", prefix, p,
2794                                         vma->vm_start,
2795                                         vma->vm_end - vma->vm_start);
2796                         free_page((unsigned long)buf);
2797                 }
2798         }
2799         up_read(&current->mm->mmap_sem);
2800 }
2801 
  This page was automatically generated by the LXR engine.