Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5

Linux/mm/memory.c

  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/ksm.h>
 49 #include <linux/rmap.h>
 50 #include <linux/export.h>
 51 #include <linux/delayacct.h>
 52 #include <linux/init.h>
 53 #include <linux/pfn_t.h>
 54 #include <linux/writeback.h>
 55 #include <linux/memcontrol.h>
 56 #include <linux/mmu_notifier.h>
 57 #include <linux/kallsyms.h>
 58 #include <linux/swapops.h>
 59 #include <linux/elf.h>
 60 #include <linux/gfp.h>
 61 #include <linux/migrate.h>
 62 #include <linux/string.h>
 63 #include <linux/dma-debug.h>
 64 #include <linux/debugfs.h>
 65 #include <linux/userfaultfd_k.h>
 66 
 67 #include <asm/io.h>
 68 #include <asm/pgalloc.h>
 69 #include <asm/uaccess.h>
 70 #include <asm/tlb.h>
 71 #include <asm/tlbflush.h>
 72 #include <asm/pgtable.h>
 73 
 74 #include "internal.h"
 75 
 76 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 77 #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
 78 #endif
 79 
 80 #ifndef CONFIG_NEED_MULTIPLE_NODES
 81 /* use the per-pgdat data instead for discontigmem - mbligh */
 82 unsigned long max_mapnr;
 83 struct page *mem_map;
 84 
 85 EXPORT_SYMBOL(max_mapnr);
 86 EXPORT_SYMBOL(mem_map);
 87 #endif
 88 
 89 /*
 90  * A number of key systems in x86 including ioremap() rely on the assumption
 91  * that high_memory defines the upper bound on direct map memory, then end
 92  * of ZONE_NORMAL.  Under CONFIG_DISCONTIG this means that max_low_pfn and
 93  * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
 94  * and ZONE_HIGHMEM.
 95  */
 96 void * high_memory;
 97 
 98 EXPORT_SYMBOL(high_memory);
 99 
100 /*
101  * Randomize the address space (stacks, mmaps, brk, etc.).
102  *
103  * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
104  *   as ancient (libc5 based) binaries can segfault. )
105  */
106 int randomize_va_space __read_mostly =
107 #ifdef CONFIG_COMPAT_BRK
108                                         1;
109 #else
110                                         2;
111 #endif
112 
113 static int __init disable_randmaps(char *s)
114 {
115         randomize_va_space = 0;
116         return 1;
117 }
118 __setup("norandmaps", disable_randmaps);
119 
120 unsigned long zero_pfn __read_mostly;
121 unsigned long highest_memmap_pfn __read_mostly;
122 
123 EXPORT_SYMBOL(zero_pfn);
124 
125 /*
126  * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
127  */
128 static int __init init_zero_pfn(void)
129 {
130         zero_pfn = page_to_pfn(ZERO_PAGE(0));
131         return 0;
132 }
133 core_initcall(init_zero_pfn);
134 
135 
136 #if defined(SPLIT_RSS_COUNTING)
137 
138 void sync_mm_rss(struct mm_struct *mm)
139 {
140         int i;
141 
142         for (i = 0; i < NR_MM_COUNTERS; i++) {
143                 if (current->rss_stat.count[i]) {
144                         add_mm_counter(mm, i, current->rss_stat.count[i]);
145                         current->rss_stat.count[i] = 0;
146                 }
147         }
148         current->rss_stat.events = 0;
149 }
150 
151 static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
152 {
153         struct task_struct *task = current;
154 
155         if (likely(task->mm == mm))
156                 task->rss_stat.count[member] += val;
157         else
158                 add_mm_counter(mm, member, val);
159 }
160 #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
161 #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
162 
163 /* sync counter once per 64 page faults */
164 #define TASK_RSS_EVENTS_THRESH  (64)
165 static void check_sync_rss_stat(struct task_struct *task)
166 {
167         if (unlikely(task != current))
168                 return;
169         if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
170                 sync_mm_rss(task->mm);
171 }
172 #else /* SPLIT_RSS_COUNTING */
173 
174 #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
175 #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
176 
177 static void check_sync_rss_stat(struct task_struct *task)
178 {
179 }
180 
181 #endif /* SPLIT_RSS_COUNTING */
182 
183 #ifdef HAVE_GENERIC_MMU_GATHER
184 
185 static bool tlb_next_batch(struct mmu_gather *tlb)
186 {
187         struct mmu_gather_batch *batch;
188 
189         batch = tlb->active;
190         if (batch->next) {
191                 tlb->active = batch->next;
192                 return true;
193         }
194 
195         if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
196                 return false;
197 
198         batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
199         if (!batch)
200                 return false;
201 
202         tlb->batch_count++;
203         batch->next = NULL;
204         batch->nr   = 0;
205         batch->max  = MAX_GATHER_BATCH;
206 
207         tlb->active->next = batch;
208         tlb->active = batch;
209 
210         return true;
211 }
212 
213 /* tlb_gather_mmu
214  *      Called to initialize an (on-stack) mmu_gather structure for page-table
215  *      tear-down from @mm. The @fullmm argument is used when @mm is without
216  *      users and we're going to destroy the full address space (exit/execve).
217  */
218 void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
219 {
220         tlb->mm = mm;
221 
222         /* Is it from 0 to ~0? */
223         tlb->fullmm     = !(start | (end+1));
224         tlb->need_flush_all = 0;
225         tlb->local.next = NULL;
226         tlb->local.nr   = 0;
227         tlb->local.max  = ARRAY_SIZE(tlb->__pages);
228         tlb->active     = &tlb->local;
229         tlb->batch_count = 0;
230 
231 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
232         tlb->batch = NULL;
233 #endif
234 
235         __tlb_reset_range(tlb);
236 }
237 
238 static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
239 {
240         if (!tlb->end)
241                 return;
242 
243         tlb_flush(tlb);
244         mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
245 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
246         tlb_table_flush(tlb);
247 #endif
248         __tlb_reset_range(tlb);
249 }
250 
251 static void tlb_flush_mmu_free(struct mmu_gather *tlb)
252 {
253         struct mmu_gather_batch *batch;
254 
255         for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
256                 free_pages_and_swap_cache(batch->pages, batch->nr);
257                 batch->nr = 0;
258         }
259         tlb->active = &tlb->local;
260 }
261 
262 void tlb_flush_mmu(struct mmu_gather *tlb)
263 {
264         tlb_flush_mmu_tlbonly(tlb);
265         tlb_flush_mmu_free(tlb);
266 }
267 
268 /* tlb_finish_mmu
269  *      Called at the end of the shootdown operation to free up any resources
270  *      that were required.
271  */
272 void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
273 {
274         struct mmu_gather_batch *batch, *next;
275 
276         tlb_flush_mmu(tlb);
277 
278         /* keep the page table cache within bounds */
279         check_pgt_cache();
280 
281         for (batch = tlb->local.next; batch; batch = next) {
282                 next = batch->next;
283                 free_pages((unsigned long)batch, 0);
284         }
285         tlb->local.next = NULL;
286 }
287 
288 /* __tlb_remove_page
289  *      Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
290  *      handling the additional races in SMP caused by other CPUs caching valid
291  *      mappings in their TLBs. Returns the number of free page slots left.
292  *      When out of page slots we must call tlb_flush_mmu().
293  */
294 int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
295 {
296         struct mmu_gather_batch *batch;
297 
298         VM_BUG_ON(!tlb->end);
299 
300         batch = tlb->active;
301         batch->pages[batch->nr++] = page;
302         if (batch->nr == batch->max) {
303                 if (!tlb_next_batch(tlb))
304                         return 0;
305                 batch = tlb->active;
306         }
307         VM_BUG_ON_PAGE(batch->nr > batch->max, page);
308 
309         return batch->max - batch->nr;
310 }
311 
312 #endif /* HAVE_GENERIC_MMU_GATHER */
313 
314 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
315 
316 /*
317  * See the comment near struct mmu_table_batch.
318  */
319 
320 static void tlb_remove_table_smp_sync(void *arg)
321 {
322         /* Simply deliver the interrupt */
323 }
324 
325 static void tlb_remove_table_one(void *table)
326 {
327         /*
328          * This isn't an RCU grace period and hence the page-tables cannot be
329          * assumed to be actually RCU-freed.
330          *
331          * It is however sufficient for software page-table walkers that rely on
332          * IRQ disabling. See the comment near struct mmu_table_batch.
333          */
334         smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
335         __tlb_remove_table(table);
336 }
337 
338 static void tlb_remove_table_rcu(struct rcu_head *head)
339 {
340         struct mmu_table_batch *batch;
341         int i;
342 
343         batch = container_of(head, struct mmu_table_batch, rcu);
344 
345         for (i = 0; i < batch->nr; i++)
346                 __tlb_remove_table(batch->tables[i]);
347 
348         free_page((unsigned long)batch);
349 }
350 
351 void tlb_table_flush(struct mmu_gather *tlb)
352 {
353         struct mmu_table_batch **batch = &tlb->batch;
354 
355         if (*batch) {
356                 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
357                 *batch = NULL;
358         }
359 }
360 
361 void tlb_remove_table(struct mmu_gather *tlb, void *table)
362 {
363         struct mmu_table_batch **batch = &tlb->batch;
364 
365         /*
366          * When there's less then two users of this mm there cannot be a
367          * concurrent page-table walk.
368          */
369         if (atomic_read(&tlb->mm->mm_users) < 2) {
370                 __tlb_remove_table(table);
371                 return;
372         }
373 
374         if (*batch == NULL) {
375                 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
376                 if (*batch == NULL) {
377                         tlb_remove_table_one(table);
378                         return;
379                 }
380                 (*batch)->nr = 0;
381         }
382         (*batch)->tables[(*batch)->nr++] = table;
383         if ((*batch)->nr == MAX_TABLE_BATCH)
384                 tlb_table_flush(tlb);
385 }
386 
387 #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
388 
389 /*
390  * Note: this doesn't free the actual pages themselves. That
391  * has been handled earlier when unmapping all the memory regions.
392  */
393 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
394                            unsigned long addr)
395 {
396         pgtable_t token = pmd_pgtable(*pmd);
397         pmd_clear(pmd);
398         pte_free_tlb(tlb, token, addr);
399         atomic_long_dec(&tlb->mm->nr_ptes);
400 }
401 
402 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
403                                 unsigned long addr, unsigned long end,
404                                 unsigned long floor, unsigned long ceiling)
405 {
406         pmd_t *pmd;
407         unsigned long next;
408         unsigned long start;
409 
410         start = addr;
411         pmd = pmd_offset(pud, addr);
412         do {
413                 next = pmd_addr_end(addr, end);
414                 if (pmd_none_or_clear_bad(pmd))
415                         continue;
416                 free_pte_range(tlb, pmd, addr);
417         } while (pmd++, addr = next, addr != end);
418 
419         start &= PUD_MASK;
420         if (start < floor)
421                 return;
422         if (ceiling) {
423                 ceiling &= PUD_MASK;
424                 if (!ceiling)
425                         return;
426         }
427         if (end - 1 > ceiling - 1)
428                 return;
429 
430         pmd = pmd_offset(pud, start);
431         pud_clear(pud);
432         pmd_free_tlb(tlb, pmd, start);
433         mm_dec_nr_pmds(tlb->mm);
434 }
435 
436 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
437                                 unsigned long addr, unsigned long end,
438                                 unsigned long floor, unsigned long ceiling)
439 {
440         pud_t *pud;
441         unsigned long next;
442         unsigned long start;
443 
444         start = addr;
445         pud = pud_offset(pgd, addr);
446         do {
447                 next = pud_addr_end(addr, end);
448                 if (pud_none_or_clear_bad(pud))
449                         continue;
450                 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
451         } while (pud++, addr = next, addr != end);
452 
453         start &= PGDIR_MASK;
454         if (start < floor)
455                 return;
456         if (ceiling) {
457                 ceiling &= PGDIR_MASK;
458                 if (!ceiling)
459                         return;
460         }
461         if (end - 1 > ceiling - 1)
462                 return;
463 
464         pud = pud_offset(pgd, start);
465         pgd_clear(pgd);
466         pud_free_tlb(tlb, pud, start);
467 }
468 
469 /*
470  * This function frees user-level page tables of a process.
471  */
472 void free_pgd_range(struct mmu_gather *tlb,
473                         unsigned long addr, unsigned long end,
474                         unsigned long floor, unsigned long ceiling)
475 {
476         pgd_t *pgd;
477         unsigned long next;
478 
479         /*
480          * The next few lines have given us lots of grief...
481          *
482          * Why are we testing PMD* at this top level?  Because often
483          * there will be no work to do at all, and we'd prefer not to
484          * go all the way down to the bottom just to discover that.
485          *
486          * Why all these "- 1"s?  Because 0 represents both the bottom
487          * of the address space and the top of it (using -1 for the
488          * top wouldn't help much: the masks would do the wrong thing).
489          * The rule is that addr 0 and floor 0 refer to the bottom of
490          * the address space, but end 0 and ceiling 0 refer to the top
491          * Comparisons need to use "end - 1" and "ceiling - 1" (though
492          * that end 0 case should be mythical).
493          *
494          * Wherever addr is brought up or ceiling brought down, we must
495          * be careful to reject "the opposite 0" before it confuses the
496          * subsequent tests.  But what about where end is brought down
497          * by PMD_SIZE below? no, end can't go down to 0 there.
498          *
499          * Whereas we round start (addr) and ceiling down, by different
500          * masks at different levels, in order to test whether a table
501          * now has no other vmas using it, so can be freed, we don't
502          * bother to round floor or end up - the tests don't need that.
503          */
504 
505         addr &= PMD_MASK;
506         if (addr < floor) {
507                 addr += PMD_SIZE;
508                 if (!addr)
509                         return;
510         }
511         if (ceiling) {
512                 ceiling &= PMD_MASK;
513                 if (!ceiling)
514                         return;
515         }
516         if (end - 1 > ceiling - 1)
517                 end -= PMD_SIZE;
518         if (addr > end - 1)
519                 return;
520 
521         pgd = pgd_offset(tlb->mm, addr);
522         do {
523                 next = pgd_addr_end(addr, end);
524                 if (pgd_none_or_clear_bad(pgd))
525                         continue;
526                 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
527         } while (pgd++, addr = next, addr != end);
528 }
529 
530 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
531                 unsigned long floor, unsigned long ceiling)
532 {
533         while (vma) {
534                 struct vm_area_struct *next = vma->vm_next;
535                 unsigned long addr = vma->vm_start;
536 
537                 /*
538                  * Hide vma from rmap and truncate_pagecache before freeing
539                  * pgtables
540                  */
541                 unlink_anon_vmas(vma);
542                 unlink_file_vma(vma);
543 
544                 if (is_vm_hugetlb_page(vma)) {
545                         hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
546                                 floor, next? next->vm_start: ceiling);
547                 } else {
548                         /*
549                          * Optimization: gather nearby vmas into one call down
550                          */
551                         while (next && next->vm_start <= vma->vm_end + PMD_SIZE
552                                && !is_vm_hugetlb_page(next)) {
553                                 vma = next;
554                                 next = vma->vm_next;
555                                 unlink_anon_vmas(vma);
556                                 unlink_file_vma(vma);
557                         }
558                         free_pgd_range(tlb, addr, vma->vm_end,
559                                 floor, next? next->vm_start: ceiling);
560                 }
561                 vma = next;
562         }
563 }
564 
565 int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
566                 pmd_t *pmd, unsigned long address)
567 {
568         spinlock_t *ptl;
569         pgtable_t new = pte_alloc_one(mm, address);
570         if (!new)
571                 return -ENOMEM;
572 
573         /*
574          * Ensure all pte setup (eg. pte page lock and page clearing) are
575          * visible before the pte is made visible to other CPUs by being
576          * put into page tables.
577          *
578          * The other side of the story is the pointer chasing in the page
579          * table walking code (when walking the page table without locking;
580          * ie. most of the time). Fortunately, these data accesses consist
581          * of a chain of data-dependent loads, meaning most CPUs (alpha
582          * being the notable exception) will already guarantee loads are
583          * seen in-order. See the alpha page table accessors for the
584          * smp_read_barrier_depends() barriers in page table walking code.
585          */
586         smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
587 
588         ptl = pmd_lock(mm, pmd);
589         if (likely(pmd_none(*pmd))) {   /* Has another populated it ? */
590                 atomic_long_inc(&mm->nr_ptes);
591                 pmd_populate(mm, pmd, new);
592                 new = NULL;
593         }
594         spin_unlock(ptl);
595         if (new)
596                 pte_free(mm, new);
597         return 0;
598 }
599 
600 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
601 {
602         pte_t *new = pte_alloc_one_kernel(&init_mm, address);
603         if (!new)
604                 return -ENOMEM;
605 
606         smp_wmb(); /* See comment in __pte_alloc */
607 
608         spin_lock(&init_mm.page_table_lock);
609         if (likely(pmd_none(*pmd))) {   /* Has another populated it ? */
610                 pmd_populate_kernel(&init_mm, pmd, new);
611                 new = NULL;
612         }
613         spin_unlock(&init_mm.page_table_lock);
614         if (new)
615                 pte_free_kernel(&init_mm, new);
616         return 0;
617 }
618 
619 static inline void init_rss_vec(int *rss)
620 {
621         memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
622 }
623 
624 static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
625 {
626         int i;
627 
628         if (current->mm == mm)
629                 sync_mm_rss(mm);
630         for (i = 0; i < NR_MM_COUNTERS; i++)
631                 if (rss[i])
632                         add_mm_counter(mm, i, rss[i]);
633 }
634 
635 /*
636  * This function is called to print an error when a bad pte
637  * is found. For example, we might have a PFN-mapped pte in
638  * a region that doesn't allow it.
639  *
640  * The calling function must still handle the error.
641  */
642 static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
643                           pte_t pte, struct page *page)
644 {
645         pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
646         pud_t *pud = pud_offset(pgd, addr);
647         pmd_t *pmd = pmd_offset(pud, addr);
648         struct address_space *mapping;
649         pgoff_t index;
650         static unsigned long resume;
651         static unsigned long nr_shown;
652         static unsigned long nr_unshown;
653 
654         /*
655          * Allow a burst of 60 reports, then keep quiet for that minute;
656          * or allow a steady drip of one report per second.
657          */
658         if (nr_shown == 60) {
659                 if (time_before(jiffies, resume)) {
660                         nr_unshown++;
661                         return;
662                 }
663                 if (nr_unshown) {
664                         printk(KERN_ALERT
665                                 "BUG: Bad page map: %lu messages suppressed\n",
666                                 nr_unshown);
667                         nr_unshown = 0;
668                 }
669                 nr_shown = 0;
670         }
671         if (nr_shown++ == 0)
672                 resume = jiffies + 60 * HZ;
673 
674         mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
675         index = linear_page_index(vma, addr);
676 
677         printk(KERN_ALERT
678                 "BUG: Bad page map in process %s  pte:%08llx pmd:%08llx\n",
679                 current->comm,
680                 (long long)pte_val(pte), (long long)pmd_val(*pmd));
681         if (page)
682                 dump_page(page, "bad pte");
683         printk(KERN_ALERT
684                 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
685                 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
686         /*
687          * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
688          */
689         pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
690                  vma->vm_file,
691                  vma->vm_ops ? vma->vm_ops->fault : NULL,
692                  vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
693                  mapping ? mapping->a_ops->readpage : NULL);
694         dump_stack();
695         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
696 }
697 
698 /*
699  * vm_normal_page -- This function gets the "struct page" associated with a pte.
700  *
701  * "Special" mappings do not wish to be associated with a "struct page" (either
702  * it doesn't exist, or it exists but they don't want to touch it). In this
703  * case, NULL is returned here. "Normal" mappings do have a struct page.
704  *
705  * There are 2 broad cases. Firstly, an architecture may define a pte_special()
706  * pte bit, in which case this function is trivial. Secondly, an architecture
707  * may not have a spare pte bit, which requires a more complicated scheme,
708  * described below.
709  *
710  * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
711  * special mapping (even if there are underlying and valid "struct pages").
712  * COWed pages of a VM_PFNMAP are always normal.
713  *
714  * The way we recognize COWed pages within VM_PFNMAP mappings is through the
715  * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
716  * set, and the vm_pgoff will point to the first PFN mapped: thus every special
717  * mapping will always honor the rule
718  *
719  *      pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
720  *
721  * And for normal mappings this is false.
722  *
723  * This restricts such mappings to be a linear translation from virtual address
724  * to pfn. To get around this restriction, we allow arbitrary mappings so long
725  * as the vma is not a COW mapping; in that case, we know that all ptes are
726  * special (because none can have been COWed).
727  *
728  *
729  * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
730  *
731  * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
732  * page" backing, however the difference is that _all_ pages with a struct
733  * page (that is, those where pfn_valid is true) are refcounted and considered
734  * normal pages by the VM. The disadvantage is that pages are refcounted
735  * (which can be slower and simply not an option for some PFNMAP users). The
736  * advantage is that we don't have to follow the strict linearity rule of
737  * PFNMAP mappings in order to support COWable mappings.
738  *
739  */
740 #ifdef __HAVE_ARCH_PTE_SPECIAL
741 # define HAVE_PTE_SPECIAL 1
742 #else
743 # define HAVE_PTE_SPECIAL 0
744 #endif
745 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
746                                 pte_t pte)
747 {
748         unsigned long pfn = pte_pfn(pte);
749 
750         if (HAVE_PTE_SPECIAL) {
751                 if (likely(!pte_special(pte)))
752                         goto check_pfn;
753                 if (vma->vm_ops && vma->vm_ops->find_special_page)
754                         return vma->vm_ops->find_special_page(vma, addr);
755                 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
756                         return NULL;
757                 if (!is_zero_pfn(pfn))
758                         print_bad_pte(vma, addr, pte, NULL);
759                 return NULL;
760         }
761 
762         /* !HAVE_PTE_SPECIAL case follows: */
763 
764         if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
765                 if (vma->vm_flags & VM_MIXEDMAP) {
766                         if (!pfn_valid(pfn))
767                                 return NULL;
768                         goto out;
769                 } else {
770                         unsigned long off;
771                         off = (addr - vma->vm_start) >> PAGE_SHIFT;
772                         if (pfn == vma->vm_pgoff + off)
773                                 return NULL;
774                         if (!is_cow_mapping(vma->vm_flags))
775                                 return NULL;
776                 }
777         }
778 
779         if (is_zero_pfn(pfn))
780                 return NULL;
781 check_pfn:
782         if (unlikely(pfn > highest_memmap_pfn)) {
783                 print_bad_pte(vma, addr, pte, NULL);
784                 return NULL;
785         }
786 
787         /*
788          * NOTE! We still have PageReserved() pages in the page tables.
789          * eg. VDSO mappings can cause them to exist.
790          */
791 out:
792         return pfn_to_page(pfn);
793 }
794 
795 /*
796  * copy one vm_area from one task to the other. Assumes the page tables
797  * already present in the new task to be cleared in the whole range
798  * covered by this vma.
799  */
800 
801 static inline unsigned long
802 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
803                 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
804                 unsigned long addr, int *rss)
805 {
806         unsigned long vm_flags = vma->vm_flags;
807         pte_t pte = *src_pte;
808         struct page *page;
809 
810         /* pte contains position in swap or file, so copy. */
811         if (unlikely(!pte_present(pte))) {
812                 swp_entry_t entry = pte_to_swp_entry(pte);
813 
814                 if (likely(!non_swap_entry(entry))) {
815                         if (swap_duplicate(entry) < 0)
816                                 return entry.val;
817 
818                         /* make sure dst_mm is on swapoff's mmlist. */
819                         if (unlikely(list_empty(&dst_mm->mmlist))) {
820                                 spin_lock(&mmlist_lock);
821                                 if (list_empty(&dst_mm->mmlist))
822                                         list_add(&dst_mm->mmlist,
823                                                         &src_mm->mmlist);
824                                 spin_unlock(&mmlist_lock);
825                         }
826                         rss[MM_SWAPENTS]++;
827                 } else if (is_migration_entry(entry)) {
828                         page = migration_entry_to_page(entry);
829 
830                         rss[mm_counter(page)]++;
831 
832                         if (is_write_migration_entry(entry) &&
833                                         is_cow_mapping(vm_flags)) {
834                                 /*
835                                  * COW mappings require pages in both
836                                  * parent and child to be set to read.
837                                  */
838                                 make_migration_entry_read(&entry);
839                                 pte = swp_entry_to_pte(entry);
840                                 if (pte_swp_soft_dirty(*src_pte))
841                                         pte = pte_swp_mksoft_dirty(pte);
842                                 set_pte_at(src_mm, addr, src_pte, pte);
843                         }
844                 }
845                 goto out_set_pte;
846         }
847 
848         /*
849          * If it's a COW mapping, write protect it both
850          * in the parent and the child
851          */
852         if (is_cow_mapping(vm_flags)) {
853                 ptep_set_wrprotect(src_mm, addr, src_pte);
854                 pte = pte_wrprotect(pte);
855         }
856 
857         /*
858          * If it's a shared mapping, mark it clean in
859          * the child
860          */
861         if (vm_flags & VM_SHARED)
862                 pte = pte_mkclean(pte);
863         pte = pte_mkold(pte);
864 
865         page = vm_normal_page(vma, addr, pte);
866         if (page) {
867                 get_page(page);
868                 page_dup_rmap(page, false);
869                 rss[mm_counter(page)]++;
870         }
871 
872 out_set_pte:
873         set_pte_at(dst_mm, addr, dst_pte, pte);
874         return 0;
875 }
876 
877 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
878                    pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
879                    unsigned long addr, unsigned long end)
880 {
881         pte_t *orig_src_pte, *orig_dst_pte;
882         pte_t *src_pte, *dst_pte;
883         spinlock_t *src_ptl, *dst_ptl;
884         int progress = 0;
885         int rss[NR_MM_COUNTERS];
886         swp_entry_t entry = (swp_entry_t){0};
887 
888 again:
889         init_rss_vec(rss);
890 
891         dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
892         if (!dst_pte)
893                 return -ENOMEM;
894         src_pte = pte_offset_map(src_pmd, addr);
895         src_ptl = pte_lockptr(src_mm, src_pmd);
896         spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
897         orig_src_pte = src_pte;
898         orig_dst_pte = dst_pte;
899         arch_enter_lazy_mmu_mode();
900 
901         do {
902                 /*
903                  * We are holding two locks at this point - either of them
904                  * could generate latencies in another task on another CPU.
905                  */
906                 if (progress >= 32) {
907                         progress = 0;
908                         if (need_resched() ||
909                             spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
910                                 break;
911                 }
912                 if (pte_none(*src_pte)) {
913                         progress++;
914                         continue;
915                 }
916                 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
917                                                         vma, addr, rss);
918                 if (entry.val)
919                         break;
920                 progress += 8;
921         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
922 
923         arch_leave_lazy_mmu_mode();
924         spin_unlock(src_ptl);
925         pte_unmap(orig_src_pte);
926         add_mm_rss_vec(dst_mm, rss);
927         pte_unmap_unlock(orig_dst_pte, dst_ptl);
928         cond_resched();
929 
930         if (entry.val) {
931                 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
932                         return -ENOMEM;
933                 progress = 0;
934         }
935         if (addr != end)
936                 goto again;
937         return 0;
938 }
939 
940 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
941                 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
942                 unsigned long addr, unsigned long end)
943 {
944         pmd_t *src_pmd, *dst_pmd;
945         unsigned long next;
946 
947         dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
948         if (!dst_pmd)
949                 return -ENOMEM;
950         src_pmd = pmd_offset(src_pud, addr);
951         do {
952                 next = pmd_addr_end(addr, end);
953                 if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) {
954                         int err;
955                         VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
956                         err = copy_huge_pmd(dst_mm, src_mm,
957                                             dst_pmd, src_pmd, addr, vma);
958                         if (err == -ENOMEM)
959                                 return -ENOMEM;
960                         if (!err)
961                                 continue;
962                         /* fall through */
963                 }
964                 if (pmd_none_or_clear_bad(src_pmd))
965                         continue;
966                 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
967                                                 vma, addr, next))
968                         return -ENOMEM;
969         } while (dst_pmd++, src_pmd++, addr = next, addr != end);
970         return 0;
971 }
972 
973 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
974                 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
975                 unsigned long addr, unsigned long end)
976 {
977         pud_t *src_pud, *dst_pud;
978         unsigned long next;
979 
980         dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
981         if (!dst_pud)
982                 return -ENOMEM;
983         src_pud = pud_offset(src_pgd, addr);
984         do {
985                 next = pud_addr_end(addr, end);
986                 if (pud_none_or_clear_bad(src_pud))
987                         continue;
988                 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
989                                                 vma, addr, next))
990                         return -ENOMEM;
991         } while (dst_pud++, src_pud++, addr = next, addr != end);
992         return 0;
993 }
994 
995 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
996                 struct vm_area_struct *vma)
997 {
998         pgd_t *src_pgd, *dst_pgd;
999         unsigned long next;
1000         unsigned long addr = vma->vm_start;
1001         unsigned long end = vma->vm_end;
1002         unsigned long mmun_start;       /* For mmu_notifiers */
1003         unsigned long mmun_end;         /* For mmu_notifiers */
1004         bool is_cow;
1005         int ret;
1006 
1007         /*
1008          * Don't copy ptes where a page fault will fill them correctly.
1009          * Fork becomes much lighter when there are big shared or private
1010          * readonly mappings. The tradeoff is that copy_page_range is more
1011          * efficient than faulting.
1012          */
1013         if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
1014                         !vma->anon_vma)
1015                 return 0;
1016 
1017         if (is_vm_hugetlb_page(vma))
1018                 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1019 
1020         if (unlikely(vma->vm_flags & VM_PFNMAP)) {
1021                 /*
1022                  * We do not free on error cases below as remove_vma
1023                  * gets called on error from higher level routine
1024                  */
1025                 ret = track_pfn_copy(vma);
1026                 if (ret)
1027                         return ret;
1028         }
1029 
1030         /*
1031          * We need to invalidate the secondary MMU mappings only when
1032          * there could be a permission downgrade on the ptes of the
1033          * parent mm. And a permission downgrade will only happen if
1034          * is_cow_mapping() returns true.
1035          */
1036         is_cow = is_cow_mapping(vma->vm_flags);
1037         mmun_start = addr;
1038         mmun_end   = end;
1039         if (is_cow)
1040                 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1041                                                     mmun_end);
1042 
1043         ret = 0;
1044         dst_pgd = pgd_offset(dst_mm, addr);
1045         src_pgd = pgd_offset(src_mm, addr);
1046         do {
1047                 next = pgd_addr_end(addr, end);
1048                 if (pgd_none_or_clear_bad(src_pgd))
1049                         continue;
1050                 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1051                                             vma, addr, next))) {
1052                         ret = -ENOMEM;
1053                         break;
1054                 }
1055         } while (dst_pgd++, src_pgd++, addr = next, addr != end);
1056 
1057         if (is_cow)
1058                 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
1059         return ret;
1060 }
1061 
1062 static unsigned long zap_pte_range(struct mmu_gather *tlb,
1063                                 struct vm_area_struct *vma, pmd_t *pmd,
1064                                 unsigned long addr, unsigned long end,
1065                                 struct zap_details *details)
1066 {
1067         struct mm_struct *mm = tlb->mm;
1068         int force_flush = 0;
1069         int rss[NR_MM_COUNTERS];
1070         spinlock_t *ptl;
1071         pte_t *start_pte;
1072         pte_t *pte;
1073         swp_entry_t entry;
1074 
1075 again:
1076         init_rss_vec(rss);
1077         start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1078         pte = start_pte;
1079         arch_enter_lazy_mmu_mode();
1080         do {
1081                 pte_t ptent = *pte;
1082                 if (pte_none(ptent)) {
1083                         continue;
1084                 }
1085 
1086                 if (pte_present(ptent)) {
1087                         struct page *page;
1088 
1089                         page = vm_normal_page(vma, addr, ptent);
1090                         if (unlikely(details) && page) {
1091                                 /*
1092                                  * unmap_shared_mapping_pages() wants to
1093                                  * invalidate cache without truncating:
1094                                  * unmap shared but keep private pages.
1095                                  */
1096                                 if (details->check_mapping &&
1097                                     details->check_mapping != page->mapping)
1098                                         continue;
1099                         }
1100                         ptent = ptep_get_and_clear_full(mm, addr, pte,
1101                                                         tlb->fullmm);
1102                         tlb_remove_tlb_entry(tlb, pte, addr);
1103                         if (unlikely(!page))
1104                                 continue;
1105 
1106                         if (!PageAnon(page)) {
1107                                 if (pte_dirty(ptent)) {
1108                                         force_flush = 1;
1109                                         set_page_dirty(page);
1110                                 }
1111                                 if (pte_young(ptent) &&
1112                                     likely(!(vma->vm_flags & VM_SEQ_READ)))
1113                                         mark_page_accessed(page);
1114                         }
1115                         rss[mm_counter(page)]--;
1116                         page_remove_rmap(page, false);
1117                         if (unlikely(page_mapcount(page) < 0))
1118                                 print_bad_pte(vma, addr, ptent, page);
1119                         if (unlikely(!__tlb_remove_page(tlb, page))) {
1120                                 force_flush = 1;
1121                                 addr += PAGE_SIZE;
1122                                 break;
1123                         }
1124                         continue;
1125                 }
1126                 /* If details->check_mapping, we leave swap entries. */
1127                 if (unlikely(details))
1128                         continue;
1129 
1130                 entry = pte_to_swp_entry(ptent);
1131                 if (!non_swap_entry(entry))
1132                         rss[MM_SWAPENTS]--;
1133                 else if (is_migration_entry(entry)) {
1134                         struct page *page;
1135 
1136                         page = migration_entry_to_page(entry);
1137                         rss[mm_counter(page)]--;
1138                 }
1139                 if (unlikely(!free_swap_and_cache(entry)))
1140                         print_bad_pte(vma, addr, ptent, NULL);
1141                 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1142         } while (pte++, addr += PAGE_SIZE, addr != end);
1143 
1144         add_mm_rss_vec(mm, rss);
1145         arch_leave_lazy_mmu_mode();
1146 
1147         /* Do the actual TLB flush before dropping ptl */
1148         if (force_flush)
1149                 tlb_flush_mmu_tlbonly(tlb);
1150         pte_unmap_unlock(start_pte, ptl);
1151 
1152         /*
1153          * If we forced a TLB flush (either due to running out of
1154          * batch buffers or because we needed to flush dirty TLB
1155          * entries before releasing the ptl), free the batched
1156          * memory too. Restart if we didn't do everything.
1157          */
1158         if (force_flush) {
1159                 force_flush = 0;
1160                 tlb_flush_mmu_free(tlb);
1161 
1162                 if (addr != end)
1163                         goto again;
1164         }
1165 
1166         return addr;
1167 }
1168 
1169 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
1170                                 struct vm_area_struct *vma, pud_t *pud,
1171                                 unsigned long addr, unsigned long end,
1172                                 struct zap_details *details)
1173 {
1174         pmd_t *pmd;
1175         unsigned long next;
1176 
1177         pmd = pmd_offset(pud, addr);
1178         do {
1179                 next = pmd_addr_end(addr, end);
1180                 if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
1181                         if (next - addr != HPAGE_PMD_SIZE) {
1182 #ifdef CONFIG_DEBUG_VM
1183                                 if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {
1184                                         pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1185                                                 __func__, addr, end,
1186                                                 vma->vm_start,
1187                                                 vma->vm_end);
1188                                         BUG();
1189                                 }
1190 #endif
1191                                 split_huge_pmd(vma, pmd, addr);
1192                         } else if (zap_huge_pmd(tlb, vma, pmd, addr))
1193                                 goto next;
1194                         /* fall through */
1195                 }
1196                 /*
1197                  * Here there can be other concurrent MADV_DONTNEED or
1198                  * trans huge page faults running, and if the pmd is
1199                  * none or trans huge it can change under us. This is
1200                  * because MADV_DONTNEED holds the mmap_sem in read
1201                  * mode.
1202                  */
1203                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1204                         goto next;
1205                 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
1206 next:
1207                 cond_resched();
1208         } while (pmd++, addr = next, addr != end);
1209 
1210         return addr;
1211 }
1212 
1213 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
1214                                 struct vm_area_struct *vma, pgd_t *pgd,
1215                                 unsigned long addr, unsigned long end,
1216                                 struct zap_details *details)
1217 {
1218         pud_t *pud;
1219         unsigned long next;
1220 
1221         pud = pud_offset(pgd, addr);
1222         do {
1223                 next = pud_addr_end(addr, end);
1224                 if (pud_none_or_clear_bad(pud))
1225                         continue;
1226                 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1227         } while (pud++, addr = next, addr != end);
1228 
1229         return addr;
1230 }
1231 
1232 static void unmap_page_range(struct mmu_gather *tlb,
1233                              struct vm_area_struct *vma,
1234                              unsigned long addr, unsigned long end,
1235                              struct zap_details *details)
1236 {
1237         pgd_t *pgd;
1238         unsigned long next;
1239 
1240         if (details && !details->check_mapping)
1241                 details = NULL;
1242 
1243         BUG_ON(addr >= end);
1244         tlb_start_vma(tlb, vma);
1245         pgd = pgd_offset(vma->vm_mm, addr);
1246         do {
1247                 next = pgd_addr_end(addr, end);
1248                 if (pgd_none_or_clear_bad(pgd))
1249                         continue;
1250                 next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1251         } while (pgd++, addr = next, addr != end);
1252         tlb_end_vma(tlb, vma);
1253 }
1254 
1255 
1256 static void unmap_single_vma(struct mmu_gather *tlb,
1257                 struct vm_area_struct *vma, unsigned long start_addr,
1258                 unsigned long end_addr,
1259                 struct zap_details *details)
1260 {
1261         unsigned long start = max(vma->vm_start, start_addr);
1262         unsigned long end;
1263 
1264         if (start >= vma->vm_end)
1265                 return;
1266         end = min(vma->vm_end, end_addr);
1267         if (end <= vma->vm_start)
1268                 return;
1269 
1270         if (vma->vm_file)
1271                 uprobe_munmap(vma, start, end);
1272 
1273         if (unlikely(vma->vm_flags & VM_PFNMAP))
1274                 untrack_pfn(vma, 0, 0);
1275 
1276         if (start != end) {
1277                 if (unlikely(is_vm_hugetlb_page(vma))) {
1278                         /*
1279                          * It is undesirable to test vma->vm_file as it
1280                          * should be non-null for valid hugetlb area.
1281                          * However, vm_file will be NULL in the error
1282                          * cleanup path of mmap_region. When
1283                          * hugetlbfs ->mmap method fails,
1284                          * mmap_region() nullifies vma->vm_file
1285                          * before calling this function to clean up.
1286                          * Since no pte has actually been setup, it is
1287                          * safe to do nothing in this case.
1288                          */
1289                         if (vma->vm_file) {
1290                                 i_mmap_lock_write(vma->vm_file->f_mapping);
1291                                 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
1292                                 i_mmap_unlock_write(vma->vm_file->f_mapping);
1293                         }
1294                 } else
1295                         unmap_page_range(tlb, vma, start, end, details);
1296         }
1297 }
1298 
1299 /**
1300  * unmap_vmas - unmap a range of memory covered by a list of vma's
1301  * @tlb: address of the caller's struct mmu_gather
1302  * @vma: the starting vma
1303  * @start_addr: virtual address at which to start unmapping
1304  * @end_addr: virtual address at which to end unmapping
1305  *
1306  * Unmap all pages in the vma list.
1307  *
1308  * Only addresses between `start' and `end' will be unmapped.
1309  *
1310  * The VMA list must be sorted in ascending virtual address order.
1311  *
1312  * unmap_vmas() assumes that the caller will flush the whole unmapped address
1313  * range after unmap_vmas() returns.  So the only responsibility here is to
1314  * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1315  * drops the lock and schedules.
1316  */
1317 void unmap_vmas(struct mmu_gather *tlb,
1318                 struct vm_area_struct *vma, unsigned long start_addr,
1319                 unsigned long end_addr)
1320 {
1321         struct mm_struct *mm = vma->vm_mm;
1322 
1323         mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
1324         for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
1325                 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
1326         mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
1327 }
1328 
1329 /**
1330  * zap_page_range - remove user pages in a given range
1331  * @vma: vm_area_struct holding the applicable pages
1332  * @start: starting address of pages to zap
1333  * @size: number of bytes to zap
1334  * @details: details of shared cache invalidation
1335  *
1336  * Caller must protect the VMA list
1337  */
1338 void zap_page_range(struct vm_area_struct *vma, unsigned long start,
1339                 unsigned long size, struct zap_details *details)
1340 {
1341         struct mm_struct *mm = vma->vm_mm;
1342         struct mmu_gather tlb;
1343         unsigned long end = start + size;
1344 
1345         lru_add_drain();
1346         tlb_gather_mmu(&tlb, mm, start, end);
1347         update_hiwater_rss(mm);
1348         mmu_notifier_invalidate_range_start(mm, start, end);
1349         for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
1350                 unmap_single_vma(&tlb, vma, start, end, details);
1351         mmu_notifier_invalidate_range_end(mm, start, end);
1352         tlb_finish_mmu(&tlb, start, end);
1353 }
1354 
1355 /**
1356  * zap_page_range_single - remove user pages in a given range
1357  * @vma: vm_area_struct holding the applicable pages
1358  * @address: starting address of pages to zap
1359  * @size: number of bytes to zap
1360  * @details: details of shared cache invalidation
1361  *
1362  * The range must fit into one VMA.
1363  */
1364 static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1365                 unsigned long size, struct zap_details *details)
1366 {
1367         struct mm_struct *mm = vma->vm_mm;
1368         struct mmu_gather tlb;
1369         unsigned long end = address + size;
1370 
1371         lru_add_drain();
1372         tlb_gather_mmu(&tlb, mm, address, end);
1373         update_hiwater_rss(mm);
1374         mmu_notifier_invalidate_range_start(mm, address, end);
1375         unmap_single_vma(&tlb, vma, address, end, details);
1376         mmu_notifier_invalidate_range_end(mm, address, end);
1377         tlb_finish_mmu(&tlb, address, end);
1378 }
1379 
1380 /**
1381  * zap_vma_ptes - remove ptes mapping the vma
1382  * @vma: vm_area_struct holding ptes to be zapped
1383  * @address: starting address of pages to zap
1384  * @size: number of bytes to zap
1385  *
1386  * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1387  *
1388  * The entire address range must be fully contained within the vma.
1389  *
1390  * Returns 0 if successful.
1391  */
1392 int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1393                 unsigned long size)
1394 {
1395         if (address < vma->vm_start || address + size > vma->vm_end ||
1396                         !(vma->vm_flags & VM_PFNMAP))
1397                 return -1;
1398         zap_page_range_single(vma, address, size, NULL);
1399         return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(zap_vma_ptes);
1402 
1403 pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
1404                         spinlock_t **ptl)
1405 {
1406         pgd_t * pgd = pgd_offset(mm, addr);
1407         pud_t * pud = pud_alloc(mm, pgd, addr);
1408         if (pud) {
1409                 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1410                 if (pmd) {
1411                         VM_BUG_ON(pmd_trans_huge(*pmd));
1412                         return pte_alloc_map_lock(mm, pmd, addr, ptl);
1413                 }
1414         }
1415         return NULL;
1416 }
1417 
1418 /*
1419  * This is the old fallback for page remapping.
1420  *
1421  * For historical reasons, it only allows reserved pages. Only
1422  * old drivers should use this, and they needed to mark their
1423  * pages reserved for the old functions anyway.
1424  */
1425 static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1426                         struct page *page, pgprot_t prot)
1427 {
1428         struct mm_struct *mm = vma->vm_mm;
1429         int retval;
1430         pte_t *pte;
1431         spinlock_t *ptl;
1432 
1433         retval = -EINVAL;
1434         if (PageAnon(page))
1435                 goto out;
1436         retval = -ENOMEM;
1437         flush_dcache_page(page);
1438         pte = get_locked_pte(mm, addr, &ptl);
1439         if (!pte)
1440                 goto out;
1441         retval = -EBUSY;
1442         if (!pte_none(*pte))
1443                 goto out_unlock;
1444 
1445         /* Ok, finally just insert the thing.. */
1446         get_page(page);
1447         inc_mm_counter_fast(mm, mm_counter_file(page));
1448         page_add_file_rmap(page);
1449         set_pte_at(mm, addr, pte, mk_pte(page, prot));
1450 
1451         retval = 0;
1452         pte_unmap_unlock(pte, ptl);
1453         return retval;
1454 out_unlock:
1455         pte_unmap_unlock(pte, ptl);
1456 out:
1457         return retval;
1458 }
1459 
1460 /**
1461  * vm_insert_page - insert single page into user vma
1462  * @vma: user vma to map to
1463  * @addr: target user address of this page
1464  * @page: source kernel page
1465  *
1466  * This allows drivers to insert individual pages they've allocated
1467  * into a user vma.
1468  *
1469  * The page has to be a nice clean _individual_ kernel allocation.
1470  * If you allocate a compound page, you need to have marked it as
1471  * such (__GFP_COMP), or manually just split the page up yourself
1472  * (see split_page()).
1473  *
1474  * NOTE! Traditionally this was done with "remap_pfn_range()" which
1475  * took an arbitrary page protection parameter. This doesn't allow
1476  * that. Your vma protection will have to be set up correctly, which
1477  * means that if you want a shared writable mapping, you'd better
1478  * ask for a shared writable mapping!
1479  *
1480  * The page does not need to be reserved.
1481  *
1482  * Usually this function is called from f_op->mmap() handler
1483  * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1484  * Caller must set VM_MIXEDMAP on vma if it wants to call this
1485  * function from other places, for example from page-fault handler.
1486  */
1487 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1488                         struct page *page)
1489 {
1490         if (addr < vma->vm_start || addr >= vma->vm_end)
1491                 return -EFAULT;
1492         if (!page_count(page))
1493                 return -EINVAL;
1494         if (!(vma->vm_flags & VM_MIXEDMAP)) {
1495                 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1496                 BUG_ON(vma->vm_flags & VM_PFNMAP);
1497                 vma->vm_flags |= VM_MIXEDMAP;
1498         }
1499         return insert_page(vma, addr, page, vma->vm_page_prot);
1500 }
1501 EXPORT_SYMBOL(vm_insert_page);
1502 
1503 static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1504                         pfn_t pfn, pgprot_t prot)
1505 {
1506         struct mm_struct *mm = vma->vm_mm;
1507         int retval;
1508         pte_t *pte, entry;
1509         spinlock_t *ptl;
1510 
1511         retval = -ENOMEM;
1512         pte = get_locked_pte(mm, addr, &ptl);
1513         if (!pte)
1514                 goto out;
1515         retval = -EBUSY;
1516         if (!pte_none(*pte))
1517                 goto out_unlock;
1518 
1519         /* Ok, finally just insert the thing.. */
1520         if (pfn_t_devmap(pfn))
1521                 entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
1522         else
1523                 entry = pte_mkspecial(pfn_t_pte(pfn, prot));
1524         set_pte_at(mm, addr, pte, entry);
1525         update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
1526 
1527         retval = 0;
1528 out_unlock:
1529         pte_unmap_unlock(pte, ptl);
1530 out:
1531         return retval;
1532 }
1533 
1534 /**
1535  * vm_insert_pfn - insert single pfn into user vma
1536  * @vma: user vma to map to
1537  * @addr: target user address of this page
1538  * @pfn: source kernel pfn
1539  *
1540  * Similar to vm_insert_page, this allows drivers to insert individual pages
1541  * they've allocated into a user vma. Same comments apply.
1542  *
1543  * This function should only be called from a vm_ops->fault handler, and
1544  * in that case the handler should return NULL.
1545  *
1546  * vma cannot be a COW mapping.
1547  *
1548  * As this is called only for pages that do not currently exist, we
1549  * do not need to flush old virtual caches or the TLB.
1550  */
1551 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1552                         unsigned long pfn)
1553 {
1554         int ret;
1555         pgprot_t pgprot = vma->vm_page_prot;
1556         /*
1557          * Technically, architectures with pte_special can avoid all these
1558          * restrictions (same for remap_pfn_range).  However we would like
1559          * consistency in testing and feature parity among all, so we should
1560          * try to keep these invariants in place for everybody.
1561          */
1562         BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1563         BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1564                                                 (VM_PFNMAP|VM_MIXEDMAP));
1565         BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1566         BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
1567 
1568         if (addr < vma->vm_start || addr >= vma->vm_end)
1569                 return -EFAULT;
1570         if (track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)))
1571                 return -EINVAL;
1572 
1573         ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot);
1574 
1575         return ret;
1576 }
1577 EXPORT_SYMBOL(vm_insert_pfn);
1578 
1579 int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1580                         pfn_t pfn)
1581 {
1582         BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1583 
1584         if (addr < vma->vm_start || addr >= vma->vm_end)
1585                 return -EFAULT;
1586 
1587         /*
1588          * If we don't have pte special, then we have to use the pfn_valid()
1589          * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1590          * refcount the page if pfn_valid is true (hence insert_page rather
1591          * than insert_pfn).  If a zero_pfn were inserted into a VM_MIXEDMAP
1592          * without pte special, it would there be refcounted as a normal page.
1593          */
1594         if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
1595                 struct page *page;
1596 
1597                 /*
1598                  * At this point we are committed to insert_page()
1599                  * regardless of whether the caller specified flags that
1600                  * result in pfn_t_has_page() == false.
1601                  */
1602                 page = pfn_to_page(pfn_t_to_pfn(pfn));
1603                 return insert_page(vma, addr, page, vma->vm_page_prot);
1604         }
1605         return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1606 }
1607 EXPORT_SYMBOL(vm_insert_mixed);
1608 
1609 /*
1610  * maps a range of physical memory into the requested pages. the old
1611  * mappings are removed. any references to nonexistent pages results
1612  * in null mappings (currently treated as "copy-on-access")
1613  */
1614 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1615                         unsigned long addr, unsigned long end,
1616                         unsigned long pfn, pgprot_t prot)
1617 {
1618         pte_t *pte;
1619         spinlock_t *ptl;
1620 
1621         pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1622         if (!pte)
1623                 return -ENOMEM;
1624         arch_enter_lazy_mmu_mode();
1625         do {
1626                 BUG_ON(!pte_none(*pte));
1627                 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
1628                 pfn++;
1629         } while (pte++, addr += PAGE_SIZE, addr != end);
1630         arch_leave_lazy_mmu_mode();
1631         pte_unmap_unlock(pte - 1, ptl);
1632         return 0;
1633 }
1634 
1635 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1636                         unsigned long addr, unsigned long end,
1637                         unsigned long pfn, pgprot_t prot)
1638 {
1639         pmd_t *pmd;
1640         unsigned long next;
1641 
1642         pfn -= addr >> PAGE_SHIFT;
1643         pmd = pmd_alloc(mm, pud, addr);
1644         if (!pmd)
1645                 return -ENOMEM;
1646         VM_BUG_ON(pmd_trans_huge(*pmd));
1647         do {
1648                 next = pmd_addr_end(addr, end);
1649                 if (remap_pte_range(mm, pmd, addr, next,
1650                                 pfn + (addr >> PAGE_SHIFT), prot))
1651                         return -ENOMEM;
1652         } while (pmd++, addr = next, addr != end);
1653         return 0;
1654 }
1655 
1656 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1657                         unsigned long addr, unsigned long end,
1658                         unsigned long pfn, pgprot_t prot)
1659 {
1660         pud_t *pud;
1661         unsigned long next;
1662 
1663         pfn -= addr >> PAGE_SHIFT;
1664         pud = pud_alloc(mm, pgd, addr);
1665         if (!pud)
1666                 return -ENOMEM;
1667         do {
1668                 next = pud_addr_end(addr, end);
1669                 if (remap_pmd_range(mm, pud, addr, next,
1670                                 pfn + (addr >> PAGE_SHIFT), prot))
1671                         return -ENOMEM;
1672         } while (pud++, addr = next, addr != end);
1673         return 0;
1674 }
1675 
1676 /**
1677  * remap_pfn_range - remap kernel memory to userspace
1678  * @vma: user vma to map to
1679  * @addr: target user address to start at
1680  * @pfn: physical address of kernel memory
1681  * @size: size of map area
1682  * @prot: page protection flags for this mapping
1683  *
1684  *  Note: this is only safe if the mm semaphore is held when called.
1685  */
1686 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1687                     unsigned long pfn, unsigned long size, pgprot_t prot)
1688 {
1689         pgd_t *pgd;
1690         unsigned long next;
1691         unsigned long end = addr + PAGE_ALIGN(size);
1692         struct mm_struct *mm = vma->vm_mm;
1693         int err;
1694 
1695         /*
1696          * Physically remapped pages are special. Tell the
1697          * rest of the world about it:
1698          *   VM_IO tells people not to look at these pages
1699          *      (accesses can have side effects).
1700          *   VM_PFNMAP tells the core MM that the base pages are just
1701          *      raw PFN mappings, and do not have a "struct page" associated
1702          *      with them.
1703          *   VM_DONTEXPAND
1704          *      Disable vma merging and expanding with mremap().
1705          *   VM_DONTDUMP
1706          *      Omit vma from core dump, even when VM_IO turned off.
1707          *
1708          * There's a horrible special case to handle copy-on-write
1709          * behaviour that some programs depend on. We mark the "original"
1710          * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1711          * See vm_normal_page() for details.
1712          */
1713         if (is_cow_mapping(vma->vm_flags)) {
1714                 if (addr != vma->vm_start || end != vma->vm_end)
1715                         return -EINVAL;
1716                 vma->vm_pgoff = pfn;
1717         }
1718 
1719         err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
1720         if (err)
1721                 return -EINVAL;
1722 
1723         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1724 
1725         BUG_ON(addr >= end);
1726         pfn -= addr >> PAGE_SHIFT;
1727         pgd = pgd_offset(mm, addr);
1728         flush_cache_range(vma, addr, end);
1729         do {
1730                 next = pgd_addr_end(addr, end);
1731                 err = remap_pud_range(mm, pgd, addr, next,
1732                                 pfn + (addr >> PAGE_SHIFT), prot);
1733                 if (err)
1734                         break;
1735         } while (pgd++, addr = next, addr != end);
1736 
1737         if (err)
1738                 untrack_pfn(vma, pfn, PAGE_ALIGN(size));
1739 
1740         return err;
1741 }
1742 EXPORT_SYMBOL(remap_pfn_range);
1743 
1744 /**
1745  * vm_iomap_memory - remap memory to userspace
1746  * @vma: user vma to map to
1747  * @start: start of area
1748  * @len: size of area
1749  *
1750  * This is a simplified io_remap_pfn_range() for common driver use. The
1751  * driver just needs to give us the physical memory range to be mapped,
1752  * we'll figure out the rest from the vma information.
1753  *
1754  * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1755  * whatever write-combining details or similar.
1756  */
1757 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1758 {
1759         unsigned long vm_len, pfn, pages;
1760 
1761         /* Check that the physical memory area passed in looks valid */
1762         if (start + len < start)
1763                 return -EINVAL;
1764         /*
1765          * You *really* shouldn't map things that aren't page-aligned,
1766          * but we've historically allowed it because IO memory might
1767          * just have smaller alignment.
1768          */
1769         len += start & ~PAGE_MASK;
1770         pfn = start >> PAGE_SHIFT;
1771         pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
1772         if (pfn + pages < pfn)
1773                 return -EINVAL;
1774 
1775         /* We start the mapping 'vm_pgoff' pages into the area */
1776         if (vma->vm_pgoff > pages)
1777                 return -EINVAL;
1778         pfn += vma->vm_pgoff;
1779         pages -= vma->vm_pgoff;
1780 
1781         /* Can we fit all of the mapping? */
1782         vm_len = vma->vm_end - vma->vm_start;
1783         if (vm_len >> PAGE_SHIFT > pages)
1784                 return -EINVAL;
1785 
1786         /* Ok, let it rip */
1787         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1788 }
1789 EXPORT_SYMBOL(vm_iomap_memory);
1790 
1791 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1792                                      unsigned long addr, unsigned long end,
1793                                      pte_fn_t fn, void *data)
1794 {
1795         pte_t *pte;
1796         int err;
1797         pgtable_t token;
1798         spinlock_t *uninitialized_var(ptl);
1799 
1800         pte = (mm == &init_mm) ?
1801                 pte_alloc_kernel(pmd, addr) :
1802                 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1803         if (!pte)
1804                 return -ENOMEM;
1805 
1806         BUG_ON(pmd_huge(*pmd));
1807 
1808         arch_enter_lazy_mmu_mode();
1809 
1810         token = pmd_pgtable(*pmd);
1811 
1812         do {
1813                 err = fn(pte++, token, addr, data);
1814                 if (err)
1815                         break;
1816         } while (addr += PAGE_SIZE, addr != end);
1817 
1818         arch_leave_lazy_mmu_mode();
1819 
1820         if (mm != &init_mm)
1821                 pte_unmap_unlock(pte-1, ptl);
1822         return err;
1823 }
1824 
1825 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1826                                      unsigned long addr, unsigned long end,
1827                                      pte_fn_t fn, void *data)
1828 {
1829         pmd_t *pmd;
1830         unsigned long next;
1831         int err;
1832 
1833         BUG_ON(pud_huge(*pud));
1834 
1835         pmd = pmd_alloc(mm, pud, addr);
1836         if (!pmd)
1837                 return -ENOMEM;
1838         do {
1839                 next = pmd_addr_end(addr, end);
1840                 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1841                 if (err)
1842                         break;
1843         } while (pmd++, addr = next, addr != end);
1844         return err;
1845 }
1846 
1847 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1848                                      unsigned long addr, unsigned long end,
1849                                      pte_fn_t fn, void *data)
1850 {
1851         pud_t *pud;
1852         unsigned long next;
1853         int err;
1854 
1855         pud = pud_alloc(mm, pgd, addr);
1856         if (!pud)
1857                 return -ENOMEM;
1858         do {
1859                 next = pud_addr_end(addr, end);
1860                 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1861                 if (err)
1862                         break;
1863         } while (pud++, addr = next, addr != end);
1864         return err;
1865 }
1866 
1867 /*
1868  * Scan a region of virtual memory, filling in page tables as necessary
1869  * and calling a provided function on each leaf page table.
1870  */
1871 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1872                         unsigned long size, pte_fn_t fn, void *data)
1873 {
1874         pgd_t *pgd;
1875         unsigned long next;
1876         unsigned long end = addr + size;
1877         int err;
1878 
1879         BUG_ON(addr >= end);
1880         pgd = pgd_offset(mm, addr);
1881         do {
1882                 next = pgd_addr_end(addr, end);
1883                 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1884                 if (err)
1885                         break;
1886         } while (pgd++, addr = next, addr != end);
1887 
1888         return err;
1889 }
1890 EXPORT_SYMBOL_GPL(apply_to_page_range);
1891 
1892 /*
1893  * handle_pte_fault chooses page fault handler according to an entry which was
1894  * read non-atomically.  Before making any commitment, on those architectures
1895  * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
1896  * parts, do_swap_page must check under lock before unmapping the pte and
1897  * proceeding (but do_wp_page is only called after already making such a check;
1898  * and do_anonymous_page can safely check later on).
1899  */
1900 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1901                                 pte_t *page_table, pte_t orig_pte)
1902 {
1903         int same = 1;
1904 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1905         if (sizeof(pte_t) > sizeof(unsigned long)) {
1906                 spinlock_t *ptl = pte_lockptr(mm, pmd);
1907                 spin_lock(ptl);
1908                 same = pte_same(*page_table, orig_pte);
1909                 spin_unlock(ptl);
1910         }
1911 #endif
1912         pte_unmap(page_table);
1913         return same;
1914 }
1915 
1916 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1917 {
1918         debug_dma_assert_idle(src);
1919 
1920         /*
1921          * If the source page was a PFN mapping, we don't have
1922          * a "struct page" for it. We do a best-effort copy by
1923          * just copying from the original user address. If that
1924          * fails, we just zero-fill it. Live with it.
1925          */
1926         if (unlikely(!src)) {
1927                 void *kaddr = kmap_atomic(dst);
1928                 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1929 
1930                 /*
1931                  * This really shouldn't fail, because the page is there
1932                  * in the page tables. But it might just be unreadable,
1933                  * in which case we just give up and fill the result with
1934                  * zeroes.
1935                  */
1936                 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1937                         clear_page(kaddr);
1938                 kunmap_atomic(kaddr);
1939                 flush_dcache_page(dst);
1940         } else
1941                 copy_user_highpage(dst, src, va, vma);
1942 }
1943 
1944 static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
1945 {
1946         struct file *vm_file = vma->vm_file;
1947 
1948         if (vm_file)
1949                 return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
1950 
1951         /*
1952          * Special mappings (e.g. VDSO) do not have any file so fake
1953          * a default GFP_KERNEL for them.
1954          */
1955         return GFP_KERNEL;
1956 }
1957 
1958 /*
1959  * Notify the address space that the page is about to become writable so that
1960  * it can prohibit this or wait for the page to get into an appropriate state.
1961  *
1962  * We do this without the lock held, so that it can sleep if it needs to.
1963  */
1964 static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page,
1965                unsigned long address)
1966 {
1967         struct vm_fault vmf;
1968         int ret;
1969 
1970         vmf.virtual_address = (void __user *)(address & PAGE_MASK);
1971         vmf.pgoff = page->index;
1972         vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
1973         vmf.gfp_mask = __get_fault_gfp_mask(vma);
1974         vmf.page = page;
1975         vmf.cow_page = NULL;
1976 
1977         ret = vma->vm_ops->page_mkwrite(vma, &vmf);
1978         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
1979                 return ret;
1980         if (unlikely(!(ret & VM_FAULT_LOCKED))) {
1981                 lock_page(page);
1982                 if (!page->mapping) {
1983                         unlock_page(page);
1984                         return 0; /* retry */
1985                 }
1986                 ret |= VM_FAULT_LOCKED;
1987         } else
1988                 VM_BUG_ON_PAGE(!PageLocked(page), page);
1989         return ret;
1990 }
1991 
1992 /*
1993  * Handle write page faults for pages that can be reused in the current vma
1994  *
1995  * This can happen either due to the mapping being with the VM_SHARED flag,
1996  * or due to us being the last reference standing to the page. In either
1997  * case, all we need to do here is to mark the page as writable and update
1998  * any related book-keeping.
1999  */
2000 static inline int wp_page_reuse(struct mm_struct *mm,
2001                         struct vm_area_struct *vma, unsigned long address,
2002                         pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
2003                         struct page *page, int page_mkwrite,
2004                         int dirty_shared)
2005         __releases(ptl)
2006 {
2007         pte_t entry;
2008         /*
2009          * Clear the pages cpupid information as the existing
2010          * information potentially belongs to a now completely
2011          * unrelated process.
2012          */
2013         if (page)
2014                 page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2015 
2016         flush_cache_page(vma, address, pte_pfn(orig_pte));
2017         entry = pte_mkyoung(orig_pte);
2018         entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2019         if (ptep_set_access_flags(vma, address, page_table, entry, 1))
2020                 update_mmu_cache(vma, address, page_table);
2021         pte_unmap_unlock(page_table, ptl);
2022 
2023         if (dirty_shared) {
2024                 struct address_space *mapping;
2025                 int dirtied;
2026 
2027                 if (!page_mkwrite)
2028                         lock_page(page);
2029 
2030                 dirtied = set_page_dirty(page);
2031                 VM_BUG_ON_PAGE(PageAnon(page), page);
2032                 mapping = page->mapping;
2033                 unlock_page(page);
2034                 page_cache_release(page);
2035 
2036                 if ((dirtied || page_mkwrite) && mapping) {
2037                         /*
2038                          * Some device drivers do not set page.mapping
2039                          * but still dirty their pages
2040                          */
2041                         balance_dirty_pages_ratelimited(mapping);
2042                 }
2043 
2044                 if (!page_mkwrite)
2045                         file_update_time(vma->vm_file);
2046         }
2047 
2048         return VM_FAULT_WRITE;
2049 }
2050 
2051 /*
2052  * Handle the case of a page which we actually need to copy to a new page.
2053  *
2054  * Called with mmap_sem locked and the old page referenced, but
2055  * without the ptl held.
2056  *
2057  * High level logic flow:
2058  *
2059  * - Allocate a page, copy the content of the old page to the new one.
2060  * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2061  * - Take the PTL. If the pte changed, bail out and release the allocated page
2062  * - If the pte is still the way we remember it, update the page table and all
2063  *   relevant references. This includes dropping the reference the page-table
2064  *   held to the old page, as well as updating the rmap.
2065  * - In any case, unlock the PTL and drop the reference we took to the old page.
2066  */
2067 static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma,
2068                         unsigned long address, pte_t *page_table, pmd_t *pmd,
2069                         pte_t orig_pte, struct page *old_page)
2070 {
2071         struct page *new_page = NULL;
2072         spinlock_t *ptl = NULL;
2073         pte_t entry;
2074         int page_copied = 0;
2075         const unsigned long mmun_start = address & PAGE_MASK;   /* For mmu_notifiers */
2076         const unsigned long mmun_end = mmun_start + PAGE_SIZE;  /* For mmu_notifiers */
2077         struct mem_cgroup *memcg;
2078 
2079         if (unlikely(anon_vma_prepare(vma)))
2080                 goto oom;
2081 
2082         if (is_zero_pfn(pte_pfn(orig_pte))) {
2083                 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2084                 if (!new_page)
2085                         goto oom;
2086         } else {
2087                 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2088                 if (!new_page)
2089                         goto oom;
2090                 cow_user_page(new_page, old_page, address, vma);
2091         }
2092 
2093         if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false))
2094                 goto oom_free_new;
2095 
2096         __SetPageUptodate(new_page);
2097 
2098         mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2099 
2100         /*
2101          * Re-check the pte - we dropped the lock
2102          */
2103         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2104         if (likely(pte_same(*page_table, orig_pte))) {
2105                 if (old_page) {
2106                         if (!PageAnon(old_page)) {
2107                                 dec_mm_counter_fast(mm,
2108                                                 mm_counter_file(old_page));
2109                                 inc_mm_counter_fast(mm, MM_ANONPAGES);
2110                         }
2111                 } else {
2112                         inc_mm_counter_fast(mm, MM_ANONPAGES);
2113                 }
2114                 flush_cache_page(vma, address, pte_pfn(orig_pte));
2115                 entry = mk_pte(new_page, vma->vm_page_prot);
2116                 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2117                 /*
2118                  * Clear the pte entry and flush it first, before updating the
2119                  * pte with the new entry. This will avoid a race condition
2120                  * seen in the presence of one thread doing SMC and another
2121                  * thread doing COW.
2122                  */
2123                 ptep_clear_flush_notify(vma, address, page_table);
2124                 page_add_new_anon_rmap(new_page, vma, address, false);
2125                 mem_cgroup_commit_charge(new_page, memcg, false, false);
2126                 lru_cache_add_active_or_unevictable(new_page, vma);
2127                 /*
2128                  * We call the notify macro here because, when using secondary
2129                  * mmu page tables (such as kvm shadow page tables), we want the
2130                  * new page to be mapped directly into the secondary page table.
2131                  */
2132                 set_pte_at_notify(mm, address, page_table, entry);
2133                 update_mmu_cache(vma, address, page_table);
2134                 if (old_page) {
2135                         /*
2136                          * Only after switching the pte to the new page may
2137                          * we remove the mapcount here. Otherwise another
2138                          * process may come and find the rmap count decremented
2139                          * before the pte is switched to the new page, and
2140                          * "reuse" the old page writing into it while our pte
2141                          * here still points into it and can be read by other
2142                          * threads.
2143                          *
2144                          * The critical issue is to order this
2145                          * page_remove_rmap with the ptp_clear_flush above.
2146                          * Those stores are ordered by (if nothing else,)
2147                          * the barrier present in the atomic_add_negative
2148                          * in page_remove_rmap.
2149                          *
2150                          * Then the TLB flush in ptep_clear_flush ensures that
2151                          * no process can access the old page before the
2152                          * decremented mapcount is visible. And the old page
2153                          * cannot be reused until after the decremented
2154                          * mapcount is visible. So transitively, TLBs to
2155                          * old page will be flushed before it can be reused.
2156                          */
2157                         page_remove_rmap(old_page, false);
2158                 }
2159 
2160                 /* Free the old page.. */
2161                 new_page = old_page;
2162                 page_copied = 1;
2163         } else {
2164                 mem_cgroup_cancel_charge(new_page, memcg, false);
2165         }
2166 
2167         if (new_page)
2168                 page_cache_release(new_page);
2169 
2170         pte_unmap_unlock(page_table, ptl);
2171         mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2172         if (old_page) {
2173                 /*
2174                  * Don't let another task, with possibly unlocked vma,
2175                  * keep the mlocked page.
2176                  */
2177                 if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2178                         lock_page(old_page);    /* LRU manipulation */
2179                         if (PageMlocked(old_page))
2180                                 munlock_vma_page(old_page);
2181                         unlock_page(old_page);
2182                 }
2183                 page_cache_release(old_page);
2184         }
2185         return page_copied ? VM_FAULT_WRITE : 0;
2186 oom_free_new:
2187         page_cache_release(new_page);
2188 oom:
2189         if (old_page)
2190                 page_cache_release(old_page);
2191         return VM_FAULT_OOM;
2192 }
2193 
2194 /*
2195  * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2196  * mapping
2197  */
2198 static int wp_pfn_shared(struct mm_struct *mm,
2199                         struct vm_area_struct *vma, unsigned long address,
2200                         pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
2201                         pmd_t *pmd)
2202 {
2203         if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
2204                 struct vm_fault vmf = {
2205                         .page = NULL,
2206                         .pgoff = linear_page_index(vma, address),
2207                         .virtual_address = (void __user *)(address & PAGE_MASK),
2208                         .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE,
2209                 };
2210                 int ret;
2211 
2212                 pte_unmap_unlock(page_table, ptl);
2213                 ret = vma->vm_ops->pfn_mkwrite(vma, &vmf);
2214                 if (ret & VM_FAULT_ERROR)
2215                         return ret;
2216                 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2217                 /*
2218                  * We might have raced with another page fault while we
2219                  * released the pte_offset_map_lock.
2220                  */
2221                 if (!pte_same(*page_table, orig_pte)) {
2222                         pte_unmap_unlock(page_table, ptl);
2223                         return 0;
2224                 }
2225         }
2226         return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte,
2227                              NULL, 0, 0);
2228 }
2229 
2230 static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma,
2231                           unsigned long address, pte_t *page_table,
2232                           pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte,
2233                           struct page *old_page)
2234         __releases(ptl)
2235 {
2236         int page_mkwrite = 0;
2237 
2238         page_cache_get(old_page);
2239 
2240         if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2241                 int tmp;
2242 
2243                 pte_unmap_unlock(page_table, ptl);
2244                 tmp = do_page_mkwrite(vma, old_page, address);
2245                 if (unlikely(!tmp || (tmp &
2246                                       (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2247                         page_cache_release(old_page);
2248                         return tmp;
2249                 }
2250                 /*
2251                  * Since we dropped the lock we need to revalidate
2252                  * the PTE as someone else may have changed it.  If
2253                  * they did, we just return, as we can count on the
2254                  * MMU to tell us if they didn't also make it writable.
2255                  */
2256                 page_table = pte_offset_map_lock(mm, pmd, address,
2257                                                  &ptl);
2258                 if (!pte_same(*page_table, orig_pte)) {
2259                         unlock_page(old_page);
2260                         pte_unmap_unlock(page_table, ptl);
2261                         page_cache_release(old_page);
2262                         return 0;
2263                 }
2264                 page_mkwrite = 1;
2265         }
2266 
2267         return wp_page_reuse(mm, vma, address, page_table, ptl,
2268                              orig_pte, old_page, page_mkwrite, 1);
2269 }
2270 
2271 /*
2272  * This routine handles present pages, when users try to write
2273  * to a shared page. It is done by copying the page to a new address
2274  * and decrementing the shared-page counter for the old page.
2275  *
2276  * Note that this routine assumes that the protection checks have been
2277  * done by the caller (the low-level page fault routine in most cases).
2278  * Thus we can safely just mark it writable once we've done any necessary
2279  * COW.
2280  *
2281  * We also mark the page dirty at this point even though the page will
2282  * change only once the write actually happens. This avoids a few races,
2283  * and potentially makes it more efficient.
2284  *
2285  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2286  * but allow concurrent faults), with pte both mapped and locked.
2287  * We return with mmap_sem still held, but pte unmapped and unlocked.
2288  */
2289 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2290                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2291                 spinlock_t *ptl, pte_t orig_pte)
2292         __releases(ptl)
2293 {
2294         struct page *old_page;
2295 
2296         old_page = vm_normal_page(vma, address, orig_pte);
2297         if (!old_page) {
2298                 /*
2299                  * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2300                  * VM_PFNMAP VMA.
2301                  *
2302                  * We should not cow pages in a shared writeable mapping.
2303                  * Just mark the pages writable and/or call ops->pfn_mkwrite.
2304                  */
2305                 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2306                                      (VM_WRITE|VM_SHARED))
2307                         return wp_pfn_shared(mm, vma, address, page_table, ptl,
2308                                              orig_pte, pmd);
2309 
2310                 pte_unmap_unlock(page_table, ptl);
2311                 return wp_page_copy(mm, vma, address, page_table, pmd,
2312                                     orig_pte, old_page);
2313         }
2314 
2315         /*
2316          * Take out anonymous pages first, anonymous shared vmas are
2317          * not dirty accountable.
2318          */
2319         if (PageAnon(old_page) && !PageKsm(old_page)) {
2320                 if (!trylock_page(old_page)) {
2321                         page_cache_get(old_page);
2322                         pte_unmap_unlock(page_table, ptl);
2323                         lock_page(old_page);
2324                         page_table = pte_offset_map_lock(mm, pmd, address,
2325                                                          &ptl);
2326                         if (!pte_same(*page_table, orig_pte)) {
2327                                 unlock_page(old_page);
2328                                 pte_unmap_unlock(page_table, ptl);
2329                                 page_cache_release(old_page);
2330                                 return 0;
2331                         }
2332                         page_cache_release(old_page);
2333                 }
2334                 if (reuse_swap_page(old_page)) {
2335                         /*
2336                          * The page is all ours.  Move it to our anon_vma so
2337                          * the rmap code will not search our parent or siblings.
2338                          * Protected against the rmap code by the page lock.
2339                          */
2340                         page_move_anon_rmap(old_page, vma, address);
2341                         unlock_page(old_page);
2342                         return wp_page_reuse(mm, vma, address, page_table, ptl,
2343                                              orig_pte, old_page, 0, 0);
2344                 }
2345                 unlock_page(old_page);
2346         } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2347                                         (VM_WRITE|VM_SHARED))) {
2348                 return wp_page_shared(mm, vma, address, page_table, pmd,
2349                                       ptl, orig_pte, old_page);
2350         }
2351 
2352         /*
2353          * Ok, we need to copy. Oh, well..
2354          */
2355         page_cache_get(old_page);
2356 
2357         pte_unmap_unlock(page_table, ptl);
2358         return wp_page_copy(mm, vma, address, page_table, pmd,
2359                             orig_pte, old_page);
2360 }
2361 
2362 static void unmap_mapping_range_vma(struct vm_area_struct *vma,
2363                 unsigned long start_addr, unsigned long end_addr,
2364                 struct zap_details *details)
2365 {
2366         zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
2367 }
2368 
2369 static inline void unmap_mapping_range_tree(struct rb_root *root,
2370                                             struct zap_details *details)
2371 {
2372         struct vm_area_struct *vma;
2373         pgoff_t vba, vea, zba, zea;
2374 
2375         vma_interval_tree_foreach(vma, root,
2376                         details->first_index, details->last_index) {
2377 
2378                 vba = vma->vm_pgoff;
2379                 vea = vba + vma_pages(vma) - 1;
2380                 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2381                 zba = details->first_index;
2382                 if (zba < vba)
2383                         zba = vba;
2384                 zea = details->last_index;
2385                 if (zea > vea)
2386                         zea = vea;
2387 
2388                 unmap_mapping_range_vma(vma,
2389                         ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2390                         ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2391                                 details);
2392         }
2393 }
2394 
2395 /**
2396  * unmap_mapping_range - unmap the portion of all mmaps in the specified
2397  * address_space corresponding to the specified page range in the underlying
2398  * file.
2399  *
2400  * @mapping: the address space containing mmaps to be unmapped.
2401  * @holebegin: byte in first page to unmap, relative to the start of
2402  * the underlying file.  This will be rounded down to a PAGE_SIZE
2403  * boundary.  Note that this is different from truncate_pagecache(), which
2404  * must keep the partial page.  In contrast, we must get rid of
2405  * partial pages.
2406  * @holelen: size of prospective hole in bytes.  This will be rounded
2407  * up to a PAGE_SIZE boundary.  A holelen of zero truncates to the
2408  * end of the file.
2409  * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2410  * but 0 when invalidating pagecache, don't throw away private data.
2411  */
2412 void unmap_mapping_range(struct address_space *mapping,
2413                 loff_t const holebegin, loff_t const holelen, int even_cows)
2414 {
2415         struct zap_details details;
2416         pgoff_t hba = holebegin >> PAGE_SHIFT;
2417         pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2418 
2419         /* Check for overflow. */
2420         if (sizeof(holelen) > sizeof(hlen)) {
2421                 long long holeend =
2422                         (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2423                 if (holeend & ~(long long)ULONG_MAX)
2424                         hlen = ULONG_MAX - hba + 1;
2425         }
2426 
2427         details.check_mapping = even_cows? NULL: mapping;
2428         details.first_index = hba;
2429         details.last_index = hba + hlen - 1;
2430         if (details.last_index < details.first_index)
2431                 details.last_index = ULONG_MAX;
2432 
2433 
2434         /* DAX uses i_mmap_lock to serialise file truncate vs page fault */
2435         i_mmap_lock_write(mapping);
2436         if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
2437                 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2438         i_mmap_unlock_write(mapping);
2439 }
2440 EXPORT_SYMBOL(unmap_mapping_range);
2441 
2442 /*
2443  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2444  * but allow concurrent faults), and pte mapped but not yet locked.
2445  * We return with pte unmapped and unlocked.
2446  *
2447  * We return with the mmap_sem locked or unlocked in the same cases
2448  * as does filemap_fault().
2449  */
2450 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2451                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2452                 unsigned int flags, pte_t orig_pte)
2453 {
2454         spinlock_t *ptl;
2455         struct page *page, *swapcache;
2456         struct mem_cgroup *memcg;
2457         swp_entry_t entry;
2458         pte_t pte;
2459         int locked;
2460         int exclusive = 0;
2461         int ret = 0;
2462 
2463         if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2464                 goto out;
2465 
2466         entry = pte_to_swp_entry(orig_pte);
2467         if (unlikely(non_swap_entry(entry))) {
2468                 if (is_migration_entry(entry)) {
2469                         migration_entry_wait(mm, pmd, address);
2470                 } else if (is_hwpoison_entry(entry)) {
2471                         ret = VM_FAULT_HWPOISON;
2472                 } else {
2473                         print_bad_pte(vma, address, orig_pte, NULL);
2474                         ret = VM_FAULT_SIGBUS;
2475                 }
2476                 goto out;
2477         }
2478         delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2479         page = lookup_swap_cache(entry);
2480         if (!page) {
2481                 page = swapin_readahead(entry,
2482                                         GFP_HIGHUSER_MOVABLE, vma, address);
2483                 if (!page) {
2484                         /*
2485                          * Back out if somebody else faulted in this pte
2486                          * while we released the pte lock.
2487                          */
2488                         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2489                         if (likely(pte_same(*page_table, orig_pte)))
2490                                 ret = VM_FAULT_OOM;
2491                         delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2492                         goto unlock;
2493                 }
2494 
2495                 /* Had to read the page from swap area: Major fault */
2496                 ret = VM_FAULT_MAJOR;
2497                 count_vm_event(PGMAJFAULT);
2498                 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
2499         } else if (PageHWPoison(page)) {
2500                 /*
2501                  * hwpoisoned dirty swapcache pages are kept for killing
2502                  * owner processes (which may be unknown at hwpoison time)
2503                  */
2504                 ret = VM_FAULT_HWPOISON;
2505                 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2506                 swapcache = page;
2507                 goto out_release;
2508         }
2509 
2510         swapcache = page;
2511         locked = lock_page_or_retry(page, mm, flags);
2512 
2513         delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2514         if (!locked) {
2515                 ret |= VM_FAULT_RETRY;
2516                 goto out_release;
2517         }
2518 
2519         /*
2520          * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2521          * release the swapcache from under us.  The page pin, and pte_same
2522          * test below, are not enough to exclude that.  Even if it is still
2523          * swapcache, we need to check that the page's swap has not changed.
2524          */
2525         if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
2526                 goto out_page;
2527 
2528         page = ksm_might_need_to_copy(page, vma, address);
2529         if (unlikely(!page)) {
2530                 ret = VM_FAULT_OOM;
2531                 page = swapcache;
2532                 goto out_page;
2533         }
2534 
2535         if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) {
2536                 ret = VM_FAULT_OOM;
2537                 goto out_page;
2538         }
2539 
2540         /*
2541          * Back out if somebody else already faulted in this pte.
2542          */
2543         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2544         if (unlikely(!pte_same(*page_table, orig_pte)))
2545                 goto out_nomap;
2546 
2547         if (unlikely(!PageUptodate(page))) {
2548                 ret = VM_FAULT_SIGBUS;
2549                 goto out_nomap;
2550         }
2551 
2552         /*
2553          * The page isn't present yet, go ahead with the fault.
2554          *
2555          * Be careful about the sequence of operations here.
2556          * To get its accounting right, reuse_swap_page() must be called
2557          * while the page is counted on swap but not yet in mapcount i.e.
2558          * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2559          * must be called after the swap_free(), or it will never succeed.
2560          */
2561 
2562         inc_mm_counter_fast(mm, MM_ANONPAGES);
2563         dec_mm_counter_fast(mm, MM_SWAPENTS);
2564         pte = mk_pte(page, vma->vm_page_prot);
2565         if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
2566                 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2567                 flags &= ~FAULT_FLAG_WRITE;
2568                 ret |= VM_FAULT_WRITE;
2569                 exclusive = RMAP_EXCLUSIVE;
2570         }
2571         flush_icache_page(vma, page);
2572         if (pte_swp_soft_dirty(orig_pte))
2573                 pte = pte_mksoft_dirty(pte);
2574         set_pte_at(mm, address, page_table, pte);
2575         if (page == swapcache) {
2576                 do_page_add_anon_rmap(page, vma, address, exclusive);
2577                 mem_cgroup_commit_charge(page, memcg, true, false);
2578         } else { /* ksm created a completely new copy */
2579                 page_add_new_anon_rmap(page, vma, address, false);
2580                 mem_cgroup_commit_charge(page, memcg, false, false);
2581                 lru_cache_add_active_or_unevictable(page, vma);
2582         }
2583 
2584         swap_free(entry);
2585         if (mem_cgroup_swap_full(page) ||
2586             (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
2587                 try_to_free_swap(page);
2588         unlock_page(page);
2589         if (page != swapcache) {
2590                 /*
2591                  * Hold the lock to avoid the swap entry to be reused
2592                  * until we take the PT lock for the pte_same() check
2593                  * (to avoid false positives from pte_same). For
2594                  * further safety release the lock after the swap_free
2595                  * so that the swap count won't change under a
2596                  * parallel locked swapcache.
2597                  */
2598                 unlock_page(swapcache);
2599                 page_cache_release(swapcache);
2600         }
2601 
2602         if (flags & FAULT_FLAG_WRITE) {
2603                 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2604                 if (ret & VM_FAULT_ERROR)
2605                         ret &= VM_FAULT_ERROR;
2606                 goto out;
2607         }
2608 
2609         /* No need to invalidate - it was non-present before */
2610         update_mmu_cache(vma, address, page_table);
2611 unlock:
2612         pte_unmap_unlock(page_table, ptl);
2613 out:
2614         return ret;
2615 out_nomap:
2616         mem_cgroup_cancel_charge(page, memcg, false);
2617         pte_unmap_unlock(page_table, ptl);
2618 out_page:
2619         unlock_page(page);
2620 out_release:
2621         page_cache_release(page);
2622         if (page != swapcache) {
2623                 unlock_page(swapcache);
2624                 page_cache_release(swapcache);
2625         }
2626         return ret;
2627 }
2628 
2629 /*
2630  * This is like a special single-page "expand_{down|up}wards()",
2631  * except we must first make sure that 'address{-|+}PAGE_SIZE'
2632  * doesn't hit another vma.
2633  */
2634 static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
2635 {
2636         address &= PAGE_MASK;
2637         if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
2638                 struct vm_area_struct *prev = vma->vm_prev;
2639 
2640                 /*
2641                  * Is there a mapping abutting this one below?
2642                  *
2643                  * That's only ok if it's the same stack mapping
2644                  * that has gotten split..
2645                  */
2646                 if (prev && prev->vm_end == address)
2647                         return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
2648 
2649                 return expand_downwards(vma, address - PAGE_SIZE);
2650         }
2651         if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
2652                 struct vm_area_struct *next = vma->vm_next;
2653 
2654                 /* As VM_GROWSDOWN but s/below/above/ */
2655                 if (next && next->vm_start == address + PAGE_SIZE)
2656                         return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
2657 
2658                 return expand_upwards(vma, address + PAGE_SIZE);
2659         }
2660         return 0;
2661 }
2662 
2663 /*
2664  * We enter with non-exclusive mmap_sem (to exclude vma changes,
2665  * but allow concurrent faults), and pte mapped but not yet locked.
2666  * We return with mmap_sem still held, but pte unmapped and unlocked.
2667  */
2668 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2669                 unsigned long address, pte_t *page_table, pmd_t *pmd,
2670                 unsigned int flags)
2671 {
2672         struct mem_cgroup *memcg;
2673         struct page *page;
2674         spinlock_t *ptl;
2675         pte_t entry;
2676 
2677         pte_unmap(page_table);
2678 
2679         /* File mapping without ->vm_ops ? */
2680         if (vma->vm_flags & VM_SHARED)
2681                 return VM_FAULT_SIGBUS;
2682 
2683         /* Check if we need to add a guard page to the stack */
2684         if (check_stack_guard_page(vma, address) < 0)
2685                 return VM_FAULT_SIGSEGV;
2686 
2687         /* Use the zero-page for reads */
2688         if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) {
2689                 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2690                                                 vma->vm_page_prot));
2691                 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2692                 if (!pte_none(*page_table))
2693                         goto unlock;
2694                 /* Deliver the page fault to userland, check inside PT lock */
2695                 if (userfaultfd_missing(vma)) {
2696                         pte_unmap_unlock(page_table, ptl);
2697                         return handle_userfault(vma, address, flags,
2698                                                 VM_UFFD_MISSING);
2699                 }
2700                 goto setpte;
2701         }
2702 
2703         /* Allocate our own private page. */
2704         if (unlikely(anon_vma_prepare(vma)))
2705                 goto oom;
2706         page = alloc_zeroed_user_highpage_movable(vma, address);
2707         if (!page)
2708                 goto oom;
2709 
2710         if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false))
2711                 goto oom_free_page;
2712 
2713         /*
2714          * The memory barrier inside __SetPageUptodate makes sure that
2715          * preceeding stores to the page contents become visible before
2716          * the set_pte_at() write.
2717          */
2718         __SetPageUptodate(page);
2719 
2720         entry = mk_pte(page, vma->vm_page_prot);
2721         if (vma->vm_flags & VM_WRITE)
2722                 entry = pte_mkwrite(pte_mkdirty(entry));
2723 
2724         page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2725         if (!pte_none(*page_table))
2726                 goto release;
2727 
2728         /* Deliver the page fault to userland, check inside PT lock */
2729         if (userfaultfd_missing(vma)) {
2730                 pte_unmap_unlock(page_table, ptl);
2731                 mem_cgroup_cancel_charge(page, memcg, false);
2732                 page_cache_release(page);
2733                 return handle_userfault(vma, address, flags,
2734                                         VM_UFFD_MISSING);
2735         }
2736 
2737         inc_mm_counter_fast(mm, MM_ANONPAGES);
2738         page_add_new_anon_rmap(page, vma, address, false);
2739         mem_cgroup_commit_charge(page, memcg, false, false);
2740         lru_cache_add_active_or_unevictable(page, vma);
2741 setpte:
2742         set_pte_at(mm, address, page_table, entry);
2743 
2744         /* No need to invalidate - it was non-present before */
2745         update_mmu_cache(vma, address, page_table);
2746 unlock:
2747         pte_unmap_unlock(page_table, ptl);
2748         return 0;
2749 release:
2750         mem_cgroup_cancel_charge(page, memcg, false);
2751         page_cache_release(page);
2752         goto unlock;
2753 oom_free_page:
2754         page_cache_release(page);
2755 oom:
2756         return VM_FAULT_OOM;
2757 }
2758 
2759 /*
2760  * The mmap_sem must have been held on entry, and may have been
2761  * released depending on flags and vma->vm_ops->fault() return value.
2762  * See filemap_fault() and __lock_page_retry().
2763  */
2764 static int __do_fault(struct vm_area_struct *vma, unsigned long address,
2765                         pgoff_t pgoff, unsigned int flags,
2766                         struct page *cow_page, struct page **page)
2767 {
2768         struct vm_fault vmf;
2769         int ret;
2770 
2771         vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2772         vmf.pgoff = pgoff;
2773         vmf.flags = flags;
2774         vmf.page = NULL;
2775         vmf.gfp_mask = __get_fault_gfp_mask(vma);
2776         vmf.cow_page = cow_page;
2777 
2778         ret = vma->vm_ops->fault(vma, &vmf);
2779         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2780                 return ret;
2781         if (!vmf.page)
2782                 goto out;
2783 
2784         if (unlikely(PageHWPoison(vmf.page))) {
2785                 if (ret & VM_FAULT_LOCKED)
2786                         unlock_page(vmf.page);
2787                 page_cache_release(vmf.page);
2788                 return VM_FAULT_HWPOISON;
2789         }
2790 
2791         if (unlikely(!(ret & VM_FAULT_LOCKED)))
2792                 lock_page(vmf.page);
2793         else
2794                 VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
2795 
2796  out:
2797         *page = vmf.page;
2798         return ret;
2799 }
2800 
2801 /**
2802  * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2803  *
2804  * @vma: virtual memory area
2805  * @address: user virtual address
2806  * @page: page to map
2807  * @pte: pointer to target page table entry
2808  * @write: true, if new entry is writable
2809  * @anon: true, if it's anonymous page
2810  *
2811  * Caller must hold page table lock relevant for @pte.
2812  *
2813  * Target users are page handler itself and implementations of
2814  * vm_ops->map_pages.
2815  */
2816 void do_set_pte(struct vm_area_struct *vma, unsigned long address,
2817                 struct page *page, pte_t *pte, bool write, bool anon)
2818 {
2819         pte_t entry;
2820 
2821         flush_icache_page(vma, page);
2822         entry = mk_pte(page, vma->vm_page_prot);
2823         if (write)
2824                 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2825         if (anon) {
2826                 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2827                 page_add_new_anon_rmap(page, vma, address, false);
2828         } else {
2829                 inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
2830                 page_add_file_rmap(page);
2831         }
2832         set_pte_at(vma->vm_mm, address, pte, entry);
2833 
2834         /* no need to invalidate: a not-present page won't be cached */
2835         update_mmu_cache(vma, address, pte);
2836 }
2837 
2838 static unsigned long fault_around_bytes __read_mostly =
2839         rounddown_pow_of_two(65536);
2840 
2841 #ifdef CONFIG_DEBUG_FS
2842 static int fault_around_bytes_get(void *data, u64 *val)
2843 {
2844         *val = fault_around_bytes;
2845         return 0;
2846 }
2847 
2848 /*
2849  * fault_around_pages() and fault_around_mask() expects fault_around_bytes
2850  * rounded down to nearest page order. It's what do_fault_around() expects to
2851  * see.
2852  */
2853 static int fault_around_bytes_set(void *data, u64 val)
2854 {
2855         if (val / PAGE_SIZE > PTRS_PER_PTE)
2856                 return -EINVAL;
2857         if (val > PAGE_SIZE)
2858                 fault_around_bytes = rounddown_pow_of_two(val);
2859         else
2860                 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
2861         return 0;
2862 }
2863 DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops,
2864                 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
2865 
2866 static int __init fault_around_debugfs(void)
2867 {
2868         void *ret;
2869 
2870         ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL,
2871                         &fault_around_bytes_fops);
2872         if (!ret)
2873                 pr_warn("Failed to create fault_around_bytes in debugfs");
2874         return 0;
2875 }
2876 late_initcall(fault_around_debugfs);
2877 #endif
2878 
2879 /*
2880  * do_fault_around() tries to map few pages around the fault address. The hope
2881  * is that the pages will be needed soon and this will lower the number of
2882  * faults to handle.
2883  *
2884  * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2885  * not ready to be mapped: not up-to-date, locked, etc.
2886  *
2887  * This function is called with the page table lock taken. In the split ptlock
2888  * case the page table lock only protects only those entries which belong to
2889  * the page table corresponding to the fault address.
2890  *
2891  * This function doesn't cross the VMA boundaries, in order to call map_pages()
2892  * only once.
2893  *
2894  * fault_around_pages() defines how many pages we'll try to map.
2895  * do_fault_around() expects it to return a power of two less than or equal to
2896  * PTRS_PER_PTE.
2897  *
2898  * The virtual address of the area that we map is naturally aligned to the
2899  * fault_around_pages() value (and therefore to page order).  This way it's
2900  * easier to guarantee that we don't cross page table boundaries.
2901  */
2902 static void do_fault_around(struct vm_area_struct *vma, unsigned long address,
2903                 pte_t *pte, pgoff_t pgoff, unsigned int flags)
2904 {
2905         unsigned long start_addr, nr_pages, mask;
2906         pgoff_t max_pgoff;
2907         struct vm_fault vmf;
2908         int off;
2909 
2910         nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
2911         mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
2912 
2913         start_addr = max(address & mask, vma->vm_start);
2914         off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
2915         pte -= off;
2916         pgoff -= off;
2917 
2918         /*
2919          *  max_pgoff is either end of page table or end of vma
2920          *  or fault_around_pages() from pgoff, depending what is nearest.
2921          */
2922         max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
2923                 PTRS_PER_PTE - 1;
2924         max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1,
2925                         pgoff + nr_pages - 1);
2926 
2927         /* Check if it makes any sense to call ->map_pages */
2928         while (!pte_none(*pte)) {
2929                 if (++pgoff > max_pgoff)
2930                         return;
2931                 start_addr += PAGE_SIZE;
2932                 if (start_addr >= vma->vm_end)
2933                         return;
2934                 pte++;
2935         }
2936 
2937         vmf.virtual_address = (void __user *) start_addr;
2938         vmf.pte = pte;
2939         vmf.pgoff = pgoff;
2940         vmf.max_pgoff = max_pgoff;
2941         vmf.flags = flags;
2942         vmf.gfp_mask = __get_fault_gfp_mask(vma);
2943         vma->vm_ops->map_pages(vma, &vmf);
2944 }
2945 
2946 static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2947                 unsigned long address, pmd_t *pmd,
2948                 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2949 {
2950         struct page *fault_page;
2951         spinlock_t *ptl;
2952         pte_t *pte;
2953         int ret = 0;
2954 
2955         /*
2956          * Let's call ->map_pages() first and use ->fault() as fallback
2957          * if page by the offset is not ready to be mapped (cold cache or
2958          * something).
2959          */
2960         if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
2961                 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2962                 do_fault_around(vma, address, pte, pgoff, flags);
2963                 if (!pte_same(*pte, orig_pte))
2964                         goto unlock_out;
2965                 pte_unmap_unlock(pte, ptl);
2966         }
2967 
2968         ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
2969         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2970                 return ret;
2971 
2972         pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2973         if (unlikely(!pte_same(*pte, orig_pte))) {
2974                 pte_unmap_unlock(pte, ptl);
2975                 unlock_page(fault_page);
2976                 page_cache_release(fault_page);
2977                 return ret;
2978         }
2979         do_set_pte(vma, address, fault_page, pte, false, false);
2980         unlock_page(fault_page);
2981 unlock_out:
2982         pte_unmap_unlock(pte, ptl);
2983         return ret;
2984 }
2985 
2986 static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2987                 unsigned long address, pmd_t *pmd,
2988                 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2989 {
2990         struct page *fault_page, *new_page;
2991         struct mem_cgroup *memcg;
2992         spinlock_t *ptl;
2993         pte_t *pte;
2994         int ret;
2995 
2996         if (unlikely(anon_vma_prepare(vma)))
2997                 return VM_FAULT_OOM;
2998 
2999         new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
3000         if (!new_page)
3001                 return VM_FAULT_OOM;
3002 
3003         if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) {
3004                 page_cache_release(new_page);
3005                 return VM_FAULT_OOM;
3006         }
3007 
3008         ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page);
3009         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3010                 goto uncharge_out;
3011 
3012         if (fault_page)
3013                 copy_user_highpage(new_page, fault_page, address, vma);
3014         __SetPageUptodate(new_page);
3015 
3016         pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3017         if (unlikely(!pte_same(*pte, orig_pte))) {
3018                 pte_unmap_unlock(pte, ptl);
3019                 if (fault_page) {
3020                         unlock_page(fault_page);
3021                         page_cache_release(fault_page);
3022                 } else {
3023                         /*
3024                          * The fault handler has no page to lock, so it holds
3025                          * i_mmap_lock for read to protect against truncate.
3026                          */
3027                         i_mmap_unlock_read(vma->vm_file->f_mapping);
3028                 }
3029                 goto uncharge_out;
3030         }
3031         do_set_pte(vma, address, new_page, pte, true, true);
3032         mem_cgroup_commit_charge(new_page, memcg, false, false);
3033         lru_cache_add_active_or_unevictable(new_page, vma);
3034         pte_unmap_unlock(pte, ptl);
3035         if (fault_page) {
3036                 unlock_page(fault_page);
3037                 page_cache_release(fault_page);
3038         } else {
3039                 /*
3040                  * The fault handler has no page to lock, so it holds
3041                  * i_mmap_lock for read to protect against truncate.
3042                  */
3043                 i_mmap_unlock_read(vma->vm_file->f_mapping);
3044         }
3045         return ret;
3046 uncharge_out:
3047         mem_cgroup_cancel_charge(new_page, memcg, false);
3048         page_cache_release(new_page);
3049         return ret;
3050 }
3051 
3052 static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3053                 unsigned long address, pmd_t *pmd,
3054                 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
3055 {
3056         struct page *fault_page;
3057         struct address_space *mapping;
3058         spinlock_t *ptl;
3059         pte_t *pte;
3060         int dirtied = 0;
3061         int ret, tmp;
3062 
3063         ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
3064         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3065                 return ret;
3066 
3067         /*
3068          * Check if the backing address space wants to know that the page is
3069          * about to become writable
3070          */
3071         if (vma->vm_ops->page_mkwrite) {
3072                 unlock_page(fault_page);
3073                 tmp = do_page_mkwrite(vma, fault_page, address);
3074                 if (unlikely(!tmp ||
3075                                 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
3076                         page_cache_release(fault_page);
3077                         return tmp;
3078                 }
3079         }
3080 
3081         pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3082         if (unlikely(!pte_same(*pte, orig_pte))) {
3083                 pte_unmap_unlock(pte, ptl);
3084                 unlock_page(fault_page);
3085                 page_cache_release(fault_page);
3086                 return ret;
3087         }
3088         do_set_pte(vma, address, fault_page, pte, true, false);
3089         pte_unmap_unlock(pte, ptl);
3090 
3091         if (set_page_dirty(fault_page))
3092                 dirtied = 1;
3093         /*
3094          * Take a local copy of the address_space - page.mapping may be zeroed
3095          * by truncate after unlock_page().   The address_space itself remains
3096          * pinned by vma->vm_file's reference.  We rely on unlock_page()'s
3097          * release semantics to prevent the compiler from undoing this copying.
3098          */
3099         mapping = page_rmapping(fault_page);
3100         unlock_page(fault_page);
3101         if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
3102                 /*
3103                  * Some device drivers do not set page.mapping but still
3104                  * dirty their pages
3105                  */
3106                 balance_dirty_pages_ratelimited(mapping);
3107         }
3108 
3109         if (!vma->vm_ops->page_mkwrite)
3110                 file_update_time(vma->vm_file);
3111 
3112         return ret;
3113 }
3114 
3115 /*
3116  * We enter with non-exclusive mmap_sem (to exclude vma changes,
3117  * but allow concurrent faults).
3118  * The mmap_sem may have been released depending on flags and our
3119  * return value.  See filemap_fault() and __lock_page_or_retry().
3120  */
3121 static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3122                 unsigned long address, pte_t *page_table, pmd_t *pmd,
3123                 unsigned int flags, pte_t orig_pte)
3124 {
3125         pgoff_t pgoff = (((address & PAGE_MASK)
3126                         - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3127 
3128         pte_unmap(page_table);
3129         /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3130         if (!vma->vm_ops->fault)
3131                 return VM_FAULT_SIGBUS;
3132         if (!(flags & FAULT_FLAG_WRITE))
3133                 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3134                                 orig_pte);
3135         if (!(vma->vm_flags & VM_SHARED))
3136                 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3137                                 orig_pte);
3138         return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
3139 }
3140 
3141 static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
3142                                 unsigned long addr, int page_nid,
3143                                 int *flags)
3144 {
3145         get_page(page);
3146 
3147         count_vm_numa_event(NUMA_HINT_FAULTS);
3148         if (page_nid == numa_node_id()) {
3149                 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
3150                 *flags |= TNF_FAULT_LOCAL;
3151         }
3152 
3153         return mpol_misplaced(page, vma, addr);
3154 }
3155 
3156 static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
3157                    unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3158 {
3159         struct page *page = NULL;
3160         spinlock_t *ptl;
3161         int page_nid = -1;
3162         int last_cpupid;
3163         int target_nid;
3164         bool migrated = false;
3165         bool was_writable = pte_write(pte);
3166         int flags = 0;
3167 
3168         /* A PROT_NONE fault should not end up here */
3169         BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));
3170 
3171         /*
3172         * The "pte" at this point cannot be used safely without
3173         * validation through pte_unmap_same(). It's of NUMA type but
3174         * the pfn may be screwed if the read is non atomic.
3175         *
3176         * We can safely just do a "set_pte_at()", because the old
3177         * page table entry is not accessible, so there would be no
3178         * concurrent hardware modifications to the PTE.
3179         */
3180         ptl = pte_lockptr(mm, pmd);
3181         spin_lock(ptl);
3182         if (unlikely(!pte_same(*ptep, pte))) {
3183                 pte_unmap_unlock(ptep, ptl);
3184                 goto out;
3185         }
3186 
3187         /* Make it present again */
3188         pte = pte_modify(pte, vma->vm_page_prot);
3189         pte = pte_mkyoung(pte);
3190         if (was_writable)
3191                 pte = pte_mkwrite(pte);
3192         set_pte_at(mm, addr, ptep, pte);
3193         update_mmu_cache(vma, addr, ptep);
3194 
3195         page = vm_normal_page(vma, addr, pte);
3196         if (!page) {
3197                 pte_unmap_unlock(ptep, ptl);
3198                 return 0;
3199         }
3200 
3201         /* TODO: handle PTE-mapped THP */
3202         if (PageCompound(page)) {
3203                 pte_unmap_unlock(ptep, ptl);
3204                 return 0;
3205         }
3206 
3207         /*
3208          * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3209          * much anyway since they can be in shared cache state. This misses
3210          * the case where a mapping is writable but the process never writes
3211          * to it but pte_write gets cleared during protection updates and
3212          * pte_dirty has unpredictable behaviour between PTE scan updates,
3213          * background writeback, dirty balancing and application behaviour.
3214          */
3215         if (!(vma->vm_flags & VM_WRITE))
3216                 flags |= TNF_NO_GROUP;
3217 
3218         /*
3219          * Flag if the page is shared between multiple address spaces. This
3220          * is later used when determining whether to group tasks together
3221          */
3222         if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3223                 flags |= TNF_SHARED;
3224 
3225         last_cpupid = page_cpupid_last(page);
3226         page_nid = page_to_nid(page);
3227         target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
3228         pte_unmap_unlock(ptep, ptl);
3229         if (target_nid == -1) {
3230                 put_page(page);
3231                 goto out;
3232         }
3233 
3234         /* Migrate to the requested node */
3235         migrated = migrate_misplaced_page(page, vma, target_nid);
3236         if (migrated) {
3237                 page_nid = target_nid;
3238                 flags |= TNF_MIGRATED;
3239         } else
3240                 flags |= TNF_MIGRATE_FAIL;
3241 
3242 out:
3243         if (page_nid != -1)
3244                 task_numa_fault(last_cpupid, page_nid, 1, flags);
3245         return 0;
3246 }
3247 
3248 static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3249                         unsigned long address, pmd_t *pmd, unsigned int flags)
3250 {
3251         if (vma_is_anonymous(vma))
3252                 return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags);
3253         if (vma->vm_ops->pmd_fault)
3254                 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3255         return VM_FAULT_FALLBACK;
3256 }
3257 
3258 static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3259                         unsigned long address, pmd_t *pmd, pmd_t orig_pmd,
3260                         unsigned int flags)
3261 {
3262         if (vma_is_anonymous(vma))
3263                 return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd);
3264         if (vma->vm_ops->pmd_fault)
3265                 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3266         return VM_FAULT_FALLBACK;
3267 }
3268 
3269 /*
3270  * These routines also need to handle stuff like marking pages dirty
3271  * and/or accessed for architectures that don't do it in hardware (most
3272  * RISC architectures).  The early dirtying is also good on the i386.
3273  *
3274  * There is also a hook called "update_mmu_cache()" that architectures
3275  * with external mmu caches can use to update those (ie the Sparc or
3276  * PowerPC hashed page tables that act as extended TLBs).
3277  *
3278  * We enter with non-exclusive mmap_sem (to exclude vma changes,
3279  * but allow concurrent faults), and pte mapped but not yet locked.
3280  * We return with pte unmapped and unlocked.
3281  *
3282  * The mmap_sem may have been released depending on flags and our
3283  * return value.  See filemap_fault() and __lock_page_or_retry().
3284  */
3285 static int handle_pte_fault(struct mm_struct *mm,
3286                      struct vm_area_struct *vma, unsigned long address,
3287                      pte_t *pte, pmd_t *pmd, unsigned int flags)
3288 {
3289         pte_t entry;
3290         spinlock_t *ptl;
3291 
3292         /*
3293          * some architectures can have larger ptes than wordsize,
3294          * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y,
3295          * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses.
3296          * The code below just needs a consistent view for the ifs and
3297          * we later double check anyway with the ptl lock held. So here
3298          * a barrier will do.
3299          */
3300         entry = *pte;
3301         barrier();
3302         if (!pte_present(entry)) {
3303                 if (pte_none(entry)) {
3304                         if (vma_is_anonymous(vma))
3305                                 return do_anonymous_page(mm, vma, address,
3306                                                          pte, pmd, flags);
3307                         else
3308                                 return do_fault(mm, vma, address, pte, pmd,
3309                                                 flags, entry);
3310                 }
3311                 return do_swap_page(mm, vma, address,
3312                                         pte, pmd, flags, entry);
3313         }
3314 
3315         if (pte_protnone(entry))
3316                 return do_numa_page(mm, vma, address, entry, pte, pmd);
3317 
3318         ptl = pte_lockptr(mm, pmd);
3319         spin_lock(ptl);
3320         if (unlikely(!pte_same(*pte, entry)))
3321                 goto unlock;
3322         if (flags & FAULT_FLAG_WRITE) {
3323                 if (!pte_write(entry))
3324                         return do_wp_page(mm, vma, address,
3325                                         pte, pmd, ptl, entry);
3326                 entry = pte_mkdirty(entry);
3327         }
3328         entry = pte_mkyoung(entry);
3329         if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
3330                 update_mmu_cache(vma, address, pte);
3331         } else {
3332                 /*
3333                  * This is needed only for protection faults but the arch code
3334                  * is not yet telling us if this is a protection fault or not.
3335                  * This still avoids useless tlb flushes for .text page faults
3336                  * with threads.
3337                  */
3338                 if (flags & FAULT_FLAG_WRITE)
3339                         flush_tlb_fix_spurious_fault(vma, address);
3340         }
3341 unlock:
3342         pte_unmap_unlock(pte, ptl);
3343         return 0;
3344 }
3345 
3346 /*
3347  * By the time we get here, we already hold the mm semaphore
3348  *
3349  * The mmap_sem may have been released depending on flags and our
3350  * return value.  See filemap_fault() and __lock_page_or_retry().
3351  */
3352 static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3353                              unsigned long address, unsigned int flags)
3354 {
3355         pgd_t *pgd;
3356         pud_t *pud;
3357         pmd_t *pmd;
3358         pte_t *pte;
3359 
3360         if (unlikely(is_vm_hugetlb_page(vma)))
3361                 return hugetlb_fault(mm, vma, address, flags);
3362 
3363         pgd = pgd_offset(mm, address);
3364         pud = pud_alloc(mm, pgd, address);
3365         if (!pud)
3366                 return VM_FAULT_OOM;
3367         pmd = pmd_alloc(mm, pud, address);
3368         if (!pmd)
3369                 return VM_FAULT_OOM;
3370         if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
3371                 int ret = create_huge_pmd(mm, vma, address, pmd, flags);
3372                 if (!(ret & VM_FAULT_FALLBACK))
3373                         return ret;
3374         } else {
3375                 pmd_t orig_pmd = *pmd;
3376                 int ret;
3377 
3378                 barrier();
3379                 if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
3380                         unsigned int dirty = flags & FAULT_FLAG_WRITE;
3381 
3382                         if (pmd_protnone(orig_pmd))
3383                                 return do_huge_pmd_numa_page(mm, vma, address,
3384                                                              orig_pmd, pmd);
3385 
3386                         if (dirty && !pmd_write(orig_pmd)) {
3387                                 ret = wp_huge_pmd(mm, vma, address, pmd,
3388                                                         orig_pmd, flags);
3389                                 if (!(ret & VM_FAULT_FALLBACK))
3390                                         return ret;
3391                         } else {
3392                                 huge_pmd_set_accessed(mm, vma, address, pmd,
3393                                                       orig_pmd, dirty);
3394                                 return 0;
3395                         }
3396                 }
3397         }
3398 
3399         /*
3400          * Use __pte_alloc instead of pte_alloc_map, because we can't
3401          * run pte_offset_map on the pmd, if an huge pmd could
3402          * materialize from under us from a different thread.
3403          */
3404         if (unlikely(pmd_none(*pmd)) &&
3405             unlikely(__pte_alloc(mm, vma, pmd, address)))
3406                 return VM_FAULT_OOM;
3407         /*
3408          * If a huge pmd materialized under us just retry later.  Use
3409          * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd
3410          * didn't become pmd_trans_huge under us and then back to pmd_none, as
3411          * a result of MADV_DONTNEED running immediately after a huge pmd fault
3412          * in a different thread of this mm, in turn leading to a misleading
3413          * pmd_trans_huge() retval.  All we have to ensure is that it is a
3414          * regular pmd that we can walk with pte_offset_map() and we can do that
3415          * through an atomic read in C, which is what pmd_trans_unstable()
3416          * provides.
3417          */
3418         if (unlikely(pmd_trans_unstable(pmd) || pmd_devmap(*pmd)))
3419                 return 0;
3420         /*
3421          * A regular pmd is established and it can't morph into a huge pmd
3422          * from under us anymore at this point because we hold the mmap_sem
3423          * read mode and khugepaged takes it in write mode. So now it's
3424          * safe to run pte_offset_map().
3425          */
3426         pte = pte_offset_map(pmd, address);
3427 
3428         return handle_pte_fault(mm, vma, address, pte, pmd, flags);
3429 }
3430 
3431 /*
3432  * By the time we get here, we already hold the mm semaphore
3433  *
3434  * The mmap_sem may have been released depending on flags and our
3435  * return value.  See filemap_fault() and __lock_page_or_retry().
3436  */
3437 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3438                     unsigned long address, unsigned int flags)
3439 {
3440         int ret;
3441 
3442         __set_current_state(TASK_RUNNING);
3443 
3444         count_vm_event(PGFAULT);
3445         mem_cgroup_count_vm_event(mm, PGFAULT);
3446 
3447         /* do counter updates before entering really critical section. */
3448         check_sync_rss_stat(current);
3449 
3450         /*
3451          * Enable the memcg OOM handling for faults triggered in user
3452          * space.  Kernel faults are handled more gracefully.
3453          */
3454         if (flags & FAULT_FLAG_USER)
3455                 mem_cgroup_oom_enable();
3456 
3457         ret = __handle_mm_fault(mm, vma, address, flags);
3458 
3459         if (flags & FAULT_FLAG_USER) {
3460                 mem_cgroup_oom_disable();
3461                 /*
3462                  * The task may have entered a memcg OOM situation but
3463                  * if the allocation error was handled gracefully (no
3464                  * VM_FAULT_OOM), there is no need to kill anything.
3465                  * Just clean up the OOM state peacefully.
3466                  */
3467                 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3468                         mem_cgroup_oom_synchronize(false);
3469         }
3470 
3471         return ret;
3472 }
3473 EXPORT_SYMBOL_GPL(handle_mm_fault);
3474 
3475 #ifndef __PAGETABLE_PUD_FOLDED
3476 /*
3477  * Allocate page upper directory.
3478  * We've already handled the fast-path in-line.
3479  */
3480 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
3481 {
3482         pud_t *new = pud_alloc_one(mm, address);
3483         if (!new)
3484                 return -ENOMEM;
3485 
3486         smp_wmb(); /* See comment in __pte_alloc */
3487 
3488         spin_lock(&mm->page_table_lock);
3489         if (pgd_present(*pgd))          /* Another has populated it */
3490                 pud_free(mm, new);
3491         else
3492                 pgd_populate(mm, pgd, new);
3493         spin_unlock(&mm->page_table_lock);
3494         return 0;
3495 }
3496 #endif /* __PAGETABLE_PUD_FOLDED */
3497 
3498 #ifndef __PAGETABLE_PMD_FOLDED
3499 /*
3500  * Allocate page middle directory.
3501  * We've already handled the fast-path in-line.
3502  */
3503 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
3504 {
3505         pmd_t *new = pmd_alloc_one(mm, address);
3506         if (!new)
3507                 return -ENOMEM;
3508 
3509         smp_wmb(); /* See comment in __pte_alloc */
3510 
3511         spin_lock(&mm->page_table_lock);
3512 #ifndef __ARCH_HAS_4LEVEL_HACK
3513         if (!pud_present(*pud)) {
3514                 mm_inc_nr_pmds(mm);
3515                 pud_populate(mm, pud, new);
3516         } else  /* Another has populated it */
3517                 pmd_free(mm, new);
3518 #else
3519         if (!pgd_present(*pud)) {
3520                 mm_inc_nr_pmds(mm);
3521                 pgd_populate(mm, pud, new);
3522         } else /* Another has populated it */
3523                 pmd_free(mm, new);
3524 #endif /* __ARCH_HAS_4LEVEL_HACK */
3525         spin_unlock(&mm->page_table_lock);
3526         return 0;
3527 }
3528 #endif /* __PAGETABLE_PMD_FOLDED */
3529 
3530 static int __follow_pte(struct mm_struct *mm, unsigned long address,
3531                 pte_t **ptepp, spinlock_t **ptlp)
3532 {
3533         pgd_t *pgd;
3534         pud_t *pud;
3535         pmd_t *pmd;
3536         pte_t *ptep;
3537 
3538         pgd = pgd_offset(mm, address);
3539         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3540                 goto out;
3541 
3542         pud = pud_offset(pgd, address);
3543         if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3544                 goto out;
3545 
3546         pmd = pmd_offset(pud, address);
3547         VM_BUG_ON(pmd_trans_huge(*pmd));
3548         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3549                 goto out;
3550 
3551         /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3552         if (pmd_huge(*pmd))
3553                 goto out;
3554 
3555         ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3556         if (!ptep)
3557                 goto out;
3558         if (!pte_present(*ptep))
3559                 goto unlock;
3560         *ptepp = ptep;
3561         return 0;
3562 unlock:
3563         pte_unmap_unlock(ptep, *ptlp);
3564 out:
3565         return -EINVAL;
3566 }
3567 
3568 static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3569                              pte_t **ptepp, spinlock_t **ptlp)
3570 {
3571         int res;
3572 
3573         /* (void) is needed to make gcc happy */
3574         (void) __cond_lock(*ptlp,
3575                            !(res = __follow_pte(mm, address, ptepp, ptlp)));
3576         return res;
3577 }
3578 
3579 /**
3580  * follow_pfn - look up PFN at a user virtual address
3581  * @vma: memory mapping
3582  * @address: user virtual address
3583  * @pfn: location to store found PFN
3584  *
3585  * Only IO mappings and raw PFN mappings are allowed.
3586  *
3587  * Returns zero and the pfn at @pfn on success, -ve otherwise.
3588  */
3589 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3590         unsigned long *pfn)
3591 {
3592         int ret = -EINVAL;
3593         spinlock_t *ptl;
3594         pte_t *ptep;
3595 
3596         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3597                 return ret;
3598 
3599         ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3600         if (ret)
3601                 return ret;
3602         *pfn = pte_pfn(*ptep);
3603         pte_unmap_unlock(ptep, ptl);
3604         return 0;
3605 }
3606 EXPORT_SYMBOL(follow_pfn);
3607 
3608 #ifdef CONFIG_HAVE_IOREMAP_PROT
3609 int follow_phys(struct vm_area_struct *vma,
3610                 unsigned long address, unsigned int flags,
3611                 unsigned long *prot, resource_size_t *phys)
3612 {
3613         int ret = -EINVAL;
3614         pte_t *ptep, pte;
3615         spinlock_t *ptl;
3616 
3617         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3618                 goto out;
3619 
3620         if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
3621                 goto out;
3622         pte = *ptep;
3623 
3624         if ((flags & FOLL_WRITE) && !pte_write(pte))
3625                 goto unlock;
3626 
3627         *prot = pgprot_val(pte_pgprot(pte));
3628         *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
3629 
3630         ret = 0;
3631 unlock:
3632         pte_unmap_unlock(ptep, ptl);
3633 out:
3634         return ret;
3635 }
3636 
3637 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3638                         void *buf, int len, int write)
3639 {
3640         resource_size_t phys_addr;
3641         unsigned long prot = 0;
3642         void __iomem *maddr;
3643         int offset = addr & (PAGE_SIZE-1);
3644 
3645         if (follow_phys(vma, addr, write, &prot, &phys_addr))
3646                 return -EINVAL;
3647 
3648         maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
3649         if (write)
3650                 memcpy_toio(maddr + offset, buf, len);
3651         else
3652                 memcpy_fromio(buf, maddr + offset, len);
3653         iounmap(maddr);
3654 
3655         return len;
3656 }
3657 EXPORT_SYMBOL_GPL(generic_access_phys);
3658 #endif
3659 
3660 /*
3661  * Access another process' address space as given in mm.  If non-NULL, use the
3662  * given task for page fault accounting.
3663  */
3664 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3665                 unsigned long addr, void *buf, int len, int write)
3666 {
3667         struct vm_area_struct *vma;
3668         void *old_buf = buf;
3669 
3670         down_read(&mm->mmap_sem);
3671         /* ignore errors, just check how much was successfully transferred */
3672         while (len) {
3673                 int bytes, ret, offset;
3674                 void *maddr;
3675                 struct page *page = NULL;
3676 
3677                 ret = get_user_pages(tsk, mm, addr, 1,
3678                                 write, 1, &page, &vma);
3679                 if (ret <= 0) {
3680 #ifndef CONFIG_HAVE_IOREMAP_PROT
3681                         break;
3682 #else
3683                         /*
3684                          * Check if this is a VM_IO | VM_PFNMAP VMA, which
3685                          * we can access using slightly different code.
3686                          */
3687                         vma = find_vma(mm, addr);
3688                         if (!vma || vma->vm_start > addr)
3689                                 break;
3690                         if (vma->vm_ops && vma->vm_ops->access)
3691                                 ret = vma->vm_ops->access(vma, addr, buf,
3692                                                           len, write);
3693                         if (ret <= 0)
3694                                 break;
3695                         bytes = ret;
3696 #endif
3697                 } else {
3698                         bytes = len;
3699                         offset = addr & (PAGE_SIZE-1);
3700                         if (bytes > PAGE_SIZE-offset)
3701                                 bytes = PAGE_SIZE-offset;
3702 
3703                         maddr = kmap(page);
3704                         if (write) {
3705                                 copy_to_user_page(vma, page, addr,
3706                                                   maddr + offset, buf, bytes);
3707                                 set_page_dirty_lock(page);
3708                         } else {
3709                                 copy_from_user_page(vma, page, addr,
3710                                                     buf, maddr + offset, bytes);
3711                         }
3712                         kunmap(page);
3713                         page_cache_release(page);
3714                 }
3715                 len -= bytes;
3716                 buf += bytes;
3717                 addr += bytes;
3718         }
3719         up_read(&mm->mmap_sem);
3720 
3721         return buf - old_buf;
3722 }
3723 
3724 /**
3725  * access_remote_vm - access another process' address space
3726  * @mm:         the mm_struct of the target address space
3727  * @addr:       start address to access
3728  * @buf:        source or destination buffer
3729  * @len:        number of bytes to transfer
3730  * @write:      whether the access is a write
3731  *
3732  * The caller must hold a reference on @mm.
3733  */
3734 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3735                 void *buf, int len, int write)
3736 {
3737         return __access_remote_vm(NULL, mm, addr, buf, len, write);
3738 }
3739 
3740 /*
3741  * Access another process' address space.
3742  * Source/target buffer must be kernel space,
3743  * Do not walk the page table directly, use get_user_pages
3744  */
3745 int access_process_vm(struct task_struct *tsk, unsigned long addr,
3746                 void *buf, int len, int write)
3747 {
3748         struct mm_struct *mm;
3749         int ret;
3750 
3751         mm = get_task_mm(tsk);
3752         if (!mm)
3753                 return 0;
3754 
3755         ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3756         mmput(mm);
3757 
3758         return ret;
3759 }
3760 
3761 /*
3762  * Print the name of a VMA.
3763  */
3764 void print_vma_addr(char *prefix, unsigned long ip)
3765 {
3766         struct mm_struct *mm = current->mm;
3767         struct vm_area_struct *vma;
3768 
3769         /*
3770          * Do not print if we are in atomic
3771          * contexts (in exception stacks, etc.):
3772          */
3773         if (preempt_count())
3774                 return;
3775 
3776         down_read(&mm->mmap_sem);
3777         vma = find_vma(mm, ip);
3778         if (vma && vma->vm_file) {
3779                 struct file *f = vma->vm_file;
3780                 char *buf = (char *)__get_free_page(GFP_KERNEL);
3781                 if (buf) {
3782                         char *p;
3783 
3784                         p = file_path(f, buf, PAGE_SIZE);
3785                         if (IS_ERR(p))
3786                                 p = "?";
3787                         printk("%s%s[%lx+%lx]", prefix, kbasename(p),
3788                                         vma->vm_start,
3789                                         vma->vm_end - vma->vm_start);
3790                         free_page((unsigned long)buf);
3791                 }
3792         }
3793         up_read(&mm->mmap_sem);
3794 }
3795 
3796 #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
3797 void __might_fault(const char *file, int line)
3798 {
3799         /*
3800          * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3801          * holding the mmap_sem, this is safe because kernel memory doesn't
3802          * get paged out, therefore we'll never actually fault, and the
3803          * below annotations will generate false positives.
3804          */
3805         if (segment_eq(get_fs(), KERNEL_DS))
3806                 return;
3807         if (pagefault_disabled())
3808                 return;
3809         __might_sleep(file, line, 0);
3810 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
3811         if (current->mm)
3812                 might_lock_read(&current->mm->mmap_sem);
3813 #endif
3814 }
3815 EXPORT_SYMBOL(__might_fault);
3816 #endif
3817 
3818 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3819 static void clear_gigantic_page(struct page *page,
3820                                 unsigned long addr,
3821                                 unsigned int pages_per_huge_page)
3822 {
3823         int i;
3824         struct page *p = page;
3825 
3826         might_sleep();
3827         for (i = 0; i < pages_per_huge_page;
3828              i++, p = mem_map_next(p, page, i)) {
3829                 cond_resched();
3830                 clear_user_highpage(p, addr + i * PAGE_SIZE);
3831         }
3832 }
3833 void clear_huge_page(struct page *page,
3834                      unsigned long addr, unsigned int pages_per_huge_page)
3835 {
3836         int i;
3837 
3838         if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3839                 clear_gigantic_page(page, addr, pages_per_huge_page);
3840                 return;
3841         }
3842 
3843         might_sleep();
3844         for (i = 0; i < pages_per_huge_page; i++) {
3845                 cond_resched();
3846                 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3847         }
3848 }
3849 
3850 static void copy_user_gigantic_page(struct page *dst, struct page *src,
3851                                     unsigned long addr,
3852                                     struct vm_area_struct *vma,
3853                                     unsigned int pages_per_huge_page)
3854 {
3855         int i;
3856         struct page *dst_base = dst;
3857         struct page *src_base = src;
3858 
3859         for (i = 0; i < pages_per_huge_page; ) {
3860                 cond_resched();
3861                 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3862 
3863                 i++;
3864                 dst = mem_map_next(dst, dst_base, i);
3865                 src = mem_map_next(src, src_base, i);
3866         }
3867 }
3868 
3869 void copy_user_huge_page(struct page *dst, struct page *src,
3870                          unsigned long addr, struct vm_area_struct *vma,
3871                          unsigned int pages_per_huge_page)
3872 {
3873         int i;
3874 
3875         if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3876                 copy_user_gigantic_page(dst, src, addr, vma,
3877                                         pages_per_huge_page);
3878                 return;
3879         }
3880 
3881         might_sleep();
3882         for (i = 0; i < pages_per_huge_page; i++) {
3883                 cond_resched();
3884                 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
3885         }
3886 }
3887 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
3888 
3889 #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
3890 
3891 static struct kmem_cache *page_ptl_cachep;
3892 
3893 void __init ptlock_cache_init(void)
3894 {
3895         page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
3896                         SLAB_PANIC, NULL);
3897 }
3898 
3899 bool ptlock_alloc(struct page *page)
3900 {
3901         spinlock_t *ptl;
3902 
3903         ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
3904         if (!ptl)
3905                 return false;
3906         page->ptl = ptl;
3907         return true;
3908 }
3909 
3910 void ptlock_free(struct page *page)
3911 {
3912         kmem_cache_free(page_ptl_cachep, page->ptl);
3913 }
3914 #endif
3915 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us