Version:  2.0.40 2.2.26 2.4.37 3.5 3.6 3.7 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

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

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