Branch data Line data Source code
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 : :
64 : : #include <asm/io.h>
65 : : #include <asm/pgalloc.h>
66 : : #include <asm/uaccess.h>
67 : : #include <asm/tlb.h>
68 : : #include <asm/tlbflush.h>
69 : : #include <asm/pgtable.h>
70 : :
71 : : #include "internal.h"
72 : :
73 : : #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
74 : : #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
75 : : #endif
76 : :
77 : : #ifndef CONFIG_NEED_MULTIPLE_NODES
78 : : /* use the per-pgdat data instead for discontigmem - mbligh */
79 : : unsigned long max_mapnr;
80 : : struct page *mem_map;
81 : :
82 : : EXPORT_SYMBOL(max_mapnr);
83 : : EXPORT_SYMBOL(mem_map);
84 : : #endif
85 : :
86 : : /*
87 : : * A number of key systems in x86 including ioremap() rely on the assumption
88 : : * that high_memory defines the upper bound on direct map memory, then end
89 : : * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
90 : : * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
91 : : * and ZONE_HIGHMEM.
92 : : */
93 : : void * high_memory;
94 : :
95 : : EXPORT_SYMBOL(high_memory);
96 : :
97 : : /*
98 : : * Randomize the address space (stacks, mmaps, brk, etc.).
99 : : *
100 : : * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
101 : : * as ancient (libc5 based) binaries can segfault. )
102 : : */
103 : : int randomize_va_space __read_mostly =
104 : : #ifdef CONFIG_COMPAT_BRK
105 : : 1;
106 : : #else
107 : : 2;
108 : : #endif
109 : :
110 : 0 : static int __init disable_randmaps(char *s)
111 : : {
112 : 0 : randomize_va_space = 0;
113 : 0 : return 1;
114 : : }
115 : : __setup("norandmaps", disable_randmaps);
116 : :
117 : : unsigned long zero_pfn __read_mostly;
118 : : unsigned long highest_memmap_pfn __read_mostly;
119 : :
120 : : /*
121 : : * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
122 : : */
123 : 0 : static int __init init_zero_pfn(void)
124 : : {
125 : 0 : zero_pfn = page_to_pfn(ZERO_PAGE(0));
126 : 0 : return 0;
127 : : }
128 : : core_initcall(init_zero_pfn);
129 : :
130 : :
131 : : #if defined(SPLIT_RSS_COUNTING)
132 : :
133 : 0 : void sync_mm_rss(struct mm_struct *mm)
134 : : {
135 : : int i;
136 : :
137 [ + + ]: 11660387 : for (i = 0; i < NR_MM_COUNTERS; i++) {
138 [ + + ]: 8745254 : if (current->rss_stat.count[i]) {
139 : : add_mm_counter(mm, i, current->rss_stat.count[i]);
140 : 2812164 : current->rss_stat.count[i] = 0;
141 : : }
142 : : }
143 : 2915133 : current->rss_stat.events = 0;
144 : 2915133 : }
145 : :
146 : 0 : static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
147 : : {
148 : 39508180 : struct task_struct *task = current;
149 : :
150 [ + + ]: 39508180 : if (likely(task->mm == mm))
151 : 37522078 : task->rss_stat.count[member] += val;
152 : : else
153 : : add_mm_counter(mm, member, val);
154 : 39508181 : }
155 : : #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
156 : : #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
157 : :
158 : : /* sync counter once per 64 page faults */
159 : : #define TASK_RSS_EVENTS_THRESH (64)
160 : 0 : static void check_sync_rss_stat(struct task_struct *task)
161 : : {
162 [ + ]: 66926906 : if (unlikely(task != current))
163 : 0 : return;
164 [ + + ]: 66927355 : if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
165 : 248430 : sync_mm_rss(task->mm);
166 : : }
167 : : #else /* SPLIT_RSS_COUNTING */
168 : :
169 : : #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
170 : : #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
171 : :
172 : : static void check_sync_rss_stat(struct task_struct *task)
173 : : {
174 : : }
175 : :
176 : : #endif /* SPLIT_RSS_COUNTING */
177 : :
178 : : #ifdef HAVE_GENERIC_MMU_GATHER
179 : :
180 : : static int tlb_next_batch(struct mmu_gather *tlb)
181 : : {
182 : : struct mmu_gather_batch *batch;
183 : :
184 : : batch = tlb->active;
185 : : if (batch->next) {
186 : : tlb->active = batch->next;
187 : : return 1;
188 : : }
189 : :
190 : : if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
191 : : return 0;
192 : :
193 : : batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
194 : : if (!batch)
195 : : return 0;
196 : :
197 : : tlb->batch_count++;
198 : : batch->next = NULL;
199 : : batch->nr = 0;
200 : : batch->max = MAX_GATHER_BATCH;
201 : :
202 : : tlb->active->next = batch;
203 : : tlb->active = batch;
204 : :
205 : : return 1;
206 : : }
207 : :
208 : : /* tlb_gather_mmu
209 : : * Called to initialize an (on-stack) mmu_gather structure for page-table
210 : : * tear-down from @mm. The @fullmm argument is used when @mm is without
211 : : * users and we're going to destroy the full address space (exit/execve).
212 : : */
213 : : void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
214 : : {
215 : : tlb->mm = mm;
216 : :
217 : : /* Is it from 0 to ~0? */
218 : : tlb->fullmm = !(start | (end+1));
219 : : tlb->need_flush_all = 0;
220 : : tlb->start = start;
221 : : tlb->end = end;
222 : : tlb->need_flush = 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 : :
234 : : void tlb_flush_mmu(struct mmu_gather *tlb)
235 : : {
236 : : struct mmu_gather_batch *batch;
237 : :
238 : : if (!tlb->need_flush)
239 : : return;
240 : : tlb->need_flush = 0;
241 : : tlb_flush(tlb);
242 : : #ifdef CONFIG_HAVE_RCU_TABLE_FREE
243 : : tlb_table_flush(tlb);
244 : : #endif
245 : :
246 : : for (batch = &tlb->local; batch; batch = batch->next) {
247 : : free_pages_and_swap_cache(batch->pages, batch->nr);
248 : : batch->nr = 0;
249 : : }
250 : : tlb->active = &tlb->local;
251 : : }
252 : :
253 : : /* tlb_finish_mmu
254 : : * Called at the end of the shootdown operation to free up any resources
255 : : * that were required.
256 : : */
257 : : void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
258 : : {
259 : : struct mmu_gather_batch *batch, *next;
260 : :
261 : : tlb_flush_mmu(tlb);
262 : :
263 : : /* keep the page table cache within bounds */
264 : : check_pgt_cache();
265 : :
266 : : for (batch = tlb->local.next; batch; batch = next) {
267 : : next = batch->next;
268 : : free_pages((unsigned long)batch, 0);
269 : : }
270 : : tlb->local.next = NULL;
271 : : }
272 : :
273 : : /* __tlb_remove_page
274 : : * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
275 : : * handling the additional races in SMP caused by other CPUs caching valid
276 : : * mappings in their TLBs. Returns the number of free page slots left.
277 : : * When out of page slots we must call tlb_flush_mmu().
278 : : */
279 : : int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
280 : : {
281 : : struct mmu_gather_batch *batch;
282 : :
283 : : VM_BUG_ON(!tlb->need_flush);
284 : :
285 : : batch = tlb->active;
286 : : batch->pages[batch->nr++] = page;
287 : : if (batch->nr == batch->max) {
288 : : if (!tlb_next_batch(tlb))
289 : : return 0;
290 : : batch = tlb->active;
291 : : }
292 : : VM_BUG_ON_PAGE(batch->nr > batch->max, page);
293 : :
294 : : return batch->max - batch->nr;
295 : : }
296 : :
297 : : #endif /* HAVE_GENERIC_MMU_GATHER */
298 : :
299 : : #ifdef CONFIG_HAVE_RCU_TABLE_FREE
300 : :
301 : : /*
302 : : * See the comment near struct mmu_table_batch.
303 : : */
304 : :
305 : : static void tlb_remove_table_smp_sync(void *arg)
306 : : {
307 : : /* Simply deliver the interrupt */
308 : : }
309 : :
310 : : static void tlb_remove_table_one(void *table)
311 : : {
312 : : /*
313 : : * This isn't an RCU grace period and hence the page-tables cannot be
314 : : * assumed to be actually RCU-freed.
315 : : *
316 : : * It is however sufficient for software page-table walkers that rely on
317 : : * IRQ disabling. See the comment near struct mmu_table_batch.
318 : : */
319 : : smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
320 : : __tlb_remove_table(table);
321 : : }
322 : :
323 : : static void tlb_remove_table_rcu(struct rcu_head *head)
324 : : {
325 : : struct mmu_table_batch *batch;
326 : : int i;
327 : :
328 : : batch = container_of(head, struct mmu_table_batch, rcu);
329 : :
330 : : for (i = 0; i < batch->nr; i++)
331 : : __tlb_remove_table(batch->tables[i]);
332 : :
333 : : free_page((unsigned long)batch);
334 : : }
335 : :
336 : : void tlb_table_flush(struct mmu_gather *tlb)
337 : : {
338 : : struct mmu_table_batch **batch = &tlb->batch;
339 : :
340 : : if (*batch) {
341 : : call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
342 : : *batch = NULL;
343 : : }
344 : : }
345 : :
346 : : void tlb_remove_table(struct mmu_gather *tlb, void *table)
347 : : {
348 : : struct mmu_table_batch **batch = &tlb->batch;
349 : :
350 : : tlb->need_flush = 1;
351 : :
352 : : /*
353 : : * When there's less then two users of this mm there cannot be a
354 : : * concurrent page-table walk.
355 : : */
356 : : if (atomic_read(&tlb->mm->mm_users) < 2) {
357 : : __tlb_remove_table(table);
358 : : return;
359 : : }
360 : :
361 : : if (*batch == NULL) {
362 : : *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
363 : : if (*batch == NULL) {
364 : : tlb_remove_table_one(table);
365 : : return;
366 : : }
367 : : (*batch)->nr = 0;
368 : : }
369 : : (*batch)->tables[(*batch)->nr++] = table;
370 : : if ((*batch)->nr == MAX_TABLE_BATCH)
371 : : tlb_table_flush(tlb);
372 : : }
373 : :
374 : : #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
375 : :
376 : : /*
377 : : * Note: this doesn't free the actual pages themselves. That
378 : : * has been handled earlier when unmapping all the memory regions.
379 : : */
380 : 0 : static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
381 : : unsigned long addr)
382 : : {
383 : 2493531 : pgtable_t token = pmd_pgtable(*pmd);
384 : 2493531 : pmd_clear(pmd);
385 : : pte_free_tlb(tlb, token, addr);
386 : 9 : atomic_long_dec(&tlb->mm->nr_ptes);
387 : 2493544 : }
388 : :
389 : : static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
390 : : unsigned long addr, unsigned long end,
391 : : unsigned long floor, unsigned long ceiling)
392 : : {
393 : : pmd_t *pmd;
394 : : unsigned long next;
395 : : unsigned long start;
396 : :
397 : : start = addr;
398 : : pmd = pmd_offset(pud, addr);
399 : : do {
400 : : next = pmd_addr_end(addr, end);
401 [ + + ]: 2634616 : if (pmd_none_or_clear_bad(pmd))
402 : 141085 : continue;
403 : 2493531 : free_pte_range(tlb, pmd, addr);
404 : : } while (pmd++, addr = next, addr != end);
405 : :
406 : : start &= PUD_MASK;
407 : : if (start < floor)
408 : : return;
409 : : if (ceiling) {
410 : : ceiling &= PUD_MASK;
411 : : if (!ceiling)
412 : : return;
413 : : }
414 : : if (end - 1 > ceiling - 1)
415 : : return;
416 : :
417 : : pmd = pmd_offset(pud, start);
418 : : pud_clear(pud);
419 : : pmd_free_tlb(tlb, pmd, start);
420 : : }
421 : :
422 : : static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
423 : : unsigned long addr, unsigned long end,
424 : : unsigned long floor, unsigned long ceiling)
425 : : {
426 : : pud_t *pud;
427 : : unsigned long next;
428 : : unsigned long start;
429 : :
430 : : start = addr;
431 : : pud = pud_offset(pgd, addr);
432 : : do {
433 : : next = pud_addr_end(addr, end);
434 : : if (pud_none_or_clear_bad(pud))
435 : : continue;
436 : : free_pmd_range(tlb, pud, addr, next, floor, ceiling);
437 : : } while (pud++, addr = next, addr != end);
438 : :
439 : : start &= PGDIR_MASK;
440 : : if (start < floor)
441 : : return;
442 : : if (ceiling) {
443 : : ceiling &= PGDIR_MASK;
444 : : if (!ceiling)
445 : : return;
446 : : }
447 : : if (end - 1 > ceiling - 1)
448 : : return;
449 : :
450 : : pud = pud_offset(pgd, start);
451 : : pgd_clear(pgd);
452 : : pud_free_tlb(tlb, pud, start);
453 : : }
454 : :
455 : : /*
456 : : * This function frees user-level page tables of a process.
457 : : */
458 : 0 : void free_pgd_range(struct mmu_gather *tlb,
459 : : unsigned long addr, unsigned long end,
460 : : unsigned long floor, unsigned long ceiling)
461 : : {
462 : : pgd_t *pgd;
463 : : unsigned long next;
464 : :
465 : : /*
466 : : * The next few lines have given us lots of grief...
467 : : *
468 : : * Why are we testing PMD* at this top level? Because often
469 : : * there will be no work to do at all, and we'd prefer not to
470 : : * go all the way down to the bottom just to discover that.
471 : : *
472 : : * Why all these "- 1"s? Because 0 represents both the bottom
473 : : * of the address space and the top of it (using -1 for the
474 : : * top wouldn't help much: the masks would do the wrong thing).
475 : : * The rule is that addr 0 and floor 0 refer to the bottom of
476 : : * the address space, but end 0 and ceiling 0 refer to the top
477 : : * Comparisons need to use "end - 1" and "ceiling - 1" (though
478 : : * that end 0 case should be mythical).
479 : : *
480 : : * Wherever addr is brought up or ceiling brought down, we must
481 : : * be careful to reject "the opposite 0" before it confuses the
482 : : * subsequent tests. But what about where end is brought down
483 : : * by PMD_SIZE below? no, end can't go down to 0 there.
484 : : *
485 : : * Whereas we round start (addr) and ceiling down, by different
486 : : * masks at different levels, in order to test whether a table
487 : : * now has no other vmas using it, so can be freed, we don't
488 : : * bother to round floor or end up - the tests don't need that.
489 : : */
490 : :
491 : 3922834 : addr &= PMD_MASK;
492 [ + + ]: 3922834 : if (addr < floor) {
493 : 1509293 : addr += PMD_SIZE;
494 [ + ]: 1509293 : if (!addr)
495 : : return;
496 : : }
497 [ + + ]: 7845677 : if (ceiling) {
498 : 2767035 : ceiling &= PMD_MASK;
499 [ + + ]: 2767035 : if (!ceiling)
500 : : return;
501 : : }
502 [ + + ]: 3922838 : if (end - 1 > ceiling - 1)
503 : 362833 : end -= PMD_SIZE;
504 [ + + ]: 3922838 : if (addr > end - 1)
505 : : return;
506 : :
507 : 2425314 : pgd = pgd_offset(tlb->mm, addr);
508 : : do {
509 [ + + ]: 2634615 : next = pgd_addr_end(addr, end);
510 : : if (pgd_none_or_clear_bad(pgd))
511 : : continue;
512 : : free_pud_range(tlb, pgd, addr, next, floor, ceiling);
513 [ + + ]: 2634618 : } while (pgd++, addr = next, addr != end);
514 : : }
515 : :
516 : 0 : void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
517 : : unsigned long floor, unsigned long ceiling)
518 : : {
519 [ + + ]: 5432483 : while (vma) {
520 : 3895570 : struct vm_area_struct *next = vma->vm_next;
521 : 3895570 : unsigned long addr = vma->vm_start;
522 : :
523 : : /*
524 : : * Hide vma from rmap and truncate_pagecache before freeing
525 : : * pgtables
526 : : */
527 : 3895570 : unlink_anon_vmas(vma);
528 : 3895554 : unlink_file_vma(vma);
529 : :
530 : : if (is_vm_hugetlb_page(vma)) {
531 : : hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
532 : : floor, next? next->vm_start: ceiling);
533 : : } else {
534 : : /*
535 : : * Optimization: gather nearby vmas into one call down
536 : : */
537 [ + + ][ + + ]: 21842707 : while (next && next->vm_start <= vma->vm_end + PMD_SIZE
538 : : && !is_vm_hugetlb_page(next)) {
539 : : vma = next;
540 : 16410192 : next = vma->vm_next;
541 : 16410192 : unlink_anon_vmas(vma);
542 : 16410246 : unlink_file_vma(vma);
543 : : }
544 [ + + ]: 3895609 : free_pgd_range(tlb, addr, vma->vm_end,
545 : : floor, next? next->vm_start: ceiling);
546 : : }
547 : : vma = next;
548 : : }
549 : 1536913 : }
550 : :
551 : 0 : int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
552 : : pmd_t *pmd, unsigned long address)
553 : : {
554 : : spinlock_t *ptl;
555 : : pgtable_t new = pte_alloc_one(mm, address);
556 : : int wait_split_huge_page;
557 [ + ]: 3620771 : if (!new)
558 : : return -ENOMEM;
559 : :
560 : : /*
561 : : * Ensure all pte setup (eg. pte page lock and page clearing) are
562 : : * visible before the pte is made visible to other CPUs by being
563 : : * put into page tables.
564 : : *
565 : : * The other side of the story is the pointer chasing in the page
566 : : * table walking code (when walking the page table without locking;
567 : : * ie. most of the time). Fortunately, these data accesses consist
568 : : * of a chain of data-dependent loads, meaning most CPUs (alpha
569 : : * being the notable exception) will already guarantee loads are
570 : : * seen in-order. See the alpha page table accessors for the
571 : : * smp_read_barrier_depends() barriers in page table walking code.
572 : : */
573 : 3620775 : smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
574 : :
575 : : ptl = pmd_lock(mm, pmd);
576 : : wait_split_huge_page = 0;
577 [ + + ]: 3620778 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
578 : 3620766 : atomic_long_inc(&mm->nr_ptes);
579 : : pmd_populate(mm, pmd, new);
580 : : new = NULL;
581 : : } else if (unlikely(pmd_trans_splitting(*pmd)))
582 : : wait_split_huge_page = 1;
583 : : spin_unlock(ptl);
584 [ + + ]: 3620773 : if (new)
585 : : pte_free(mm, new);
586 : : if (wait_split_huge_page)
587 : : wait_split_huge_page(vma->anon_vma, pmd);
588 : : return 0;
589 : : }
590 : :
591 : 0 : int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
592 : : {
593 : : pte_t *new = pte_alloc_one_kernel(&init_mm, address);
594 [ + - ]: 46 : if (!new)
595 : : return -ENOMEM;
596 : :
597 : 46 : smp_wmb(); /* See comment in __pte_alloc */
598 : :
599 : : spin_lock(&init_mm.page_table_lock);
600 [ + - ]: 46 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
601 : : pmd_populate_kernel(&init_mm, pmd, new);
602 : : new = NULL;
603 : : } else
604 : : VM_BUG_ON(pmd_trans_splitting(*pmd));
605 : : spin_unlock(&init_mm.page_table_lock);
606 [ - + ]: 46 : if (new)
607 : : pte_free_kernel(&init_mm, new);
608 : : return 0;
609 : : }
610 : :
611 : : static inline void init_rss_vec(int *rss)
612 : : {
613 : 31783110 : memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
614 : : }
615 : :
616 : : static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
617 : : {
618 : : int i;
619 : :
620 [ + + - + ]: 31783333 : if (current->mm == mm)
621 : 31783333 : sync_mm_rss(mm);
622 [ + + ][ + + ]: 127132329 : for (i = 0; i < NR_MM_COUNTERS; i++)
623 [ + + ][ + + ]: 95348785 : if (rss[i])
624 : : add_mm_counter(mm, i, rss[i]);
625 : : }
626 : :
627 : : /*
628 : : * This function is called to print an error when a bad pte
629 : : * is found. For example, we might have a PFN-mapped pte in
630 : : * a region that doesn't allow it.
631 : : *
632 : : * The calling function must still handle the error.
633 : : */
634 : 0 : static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
635 : : pte_t pte, struct page *page)
636 : : {
637 : 0 : pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
638 : : pud_t *pud = pud_offset(pgd, addr);
639 : : pmd_t *pmd = pmd_offset(pud, addr);
640 : : struct address_space *mapping;
641 : : pgoff_t index;
642 : : static unsigned long resume;
643 : : static unsigned long nr_shown;
644 : : static unsigned long nr_unshown;
645 : :
646 : : /*
647 : : * Allow a burst of 60 reports, then keep quiet for that minute;
648 : : * or allow a steady drip of one report per second.
649 : : */
650 [ # # ]: 0 : if (nr_shown == 60) {
651 [ # # ]: 0 : if (time_before(jiffies, resume)) {
652 : 0 : nr_unshown++;
653 : 0 : return;
654 : : }
655 [ # # ]: 0 : if (nr_unshown) {
656 : 0 : printk(KERN_ALERT
657 : : "BUG: Bad page map: %lu messages suppressed\n",
658 : : nr_unshown);
659 : 0 : nr_unshown = 0;
660 : : }
661 : 0 : nr_shown = 0;
662 : : }
663 [ # # ]: 0 : if (nr_shown++ == 0)
664 : 0 : resume = jiffies + 60 * HZ;
665 : :
666 [ # # ]: 0 : mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
667 : : index = linear_page_index(vma, addr);
668 : :
669 : 0 : printk(KERN_ALERT
670 : : "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
671 : 0 : current->comm,
672 : 0 : (long long)pte_val(pte), (long long)pmd_val(*pmd));
673 [ # # ]: 0 : if (page)
674 : 0 : dump_page(page, "bad pte");
675 : 0 : printk(KERN_ALERT
676 : : "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
677 : : (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
678 : : /*
679 : : * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
680 : : */
681 [ # # ]: 0 : if (vma->vm_ops)
682 : 0 : printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n",
683 : : vma->vm_ops->fault);
684 [ # # ]: 0 : if (vma->vm_file)
685 : 0 : printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n",
686 : 0 : vma->vm_file->f_op->mmap);
687 : 0 : dump_stack();
688 : 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
689 : : }
690 : :
691 : : static inline bool is_cow_mapping(vm_flags_t flags)
692 : : {
693 : 23540969 : return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
694 : : }
695 : :
696 : : /*
697 : : * vm_normal_page -- This function gets the "struct page" associated with a pte.
698 : : *
699 : : * "Special" mappings do not wish to be associated with a "struct page" (either
700 : : * it doesn't exist, or it exists but they don't want to touch it). In this
701 : : * case, NULL is returned here. "Normal" mappings do have a struct page.
702 : : *
703 : : * There are 2 broad cases. Firstly, an architecture may define a pte_special()
704 : : * pte bit, in which case this function is trivial. Secondly, an architecture
705 : : * may not have a spare pte bit, which requires a more complicated scheme,
706 : : * described below.
707 : : *
708 : : * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
709 : : * special mapping (even if there are underlying and valid "struct pages").
710 : : * COWed pages of a VM_PFNMAP are always normal.
711 : : *
712 : : * The way we recognize COWed pages within VM_PFNMAP mappings is through the
713 : : * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
714 : : * set, and the vm_pgoff will point to the first PFN mapped: thus every special
715 : : * mapping will always honor the rule
716 : : *
717 : : * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
718 : : *
719 : : * And for normal mappings this is false.
720 : : *
721 : : * This restricts such mappings to be a linear translation from virtual address
722 : : * to pfn. To get around this restriction, we allow arbitrary mappings so long
723 : : * as the vma is not a COW mapping; in that case, we know that all ptes are
724 : : * special (because none can have been COWed).
725 : : *
726 : : *
727 : : * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
728 : : *
729 : : * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
730 : : * page" backing, however the difference is that _all_ pages with a struct
731 : : * page (that is, those where pfn_valid is true) are refcounted and considered
732 : : * normal pages by the VM. The disadvantage is that pages are refcounted
733 : : * (which can be slower and simply not an option for some PFNMAP users). The
734 : : * advantage is that we don't have to follow the strict linearity rule of
735 : : * PFNMAP mappings in order to support COWable mappings.
736 : : *
737 : : */
738 : : #ifdef __HAVE_ARCH_PTE_SPECIAL
739 : : # define HAVE_PTE_SPECIAL 1
740 : : #else
741 : : # define HAVE_PTE_SPECIAL 0
742 : : #endif
743 : 0 : struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
744 : : pte_t pte)
745 : : {
746 : 105154779 : unsigned long pfn = pte_pfn(pte);
747 : :
748 : : if (HAVE_PTE_SPECIAL) {
749 : : if (likely(!pte_special(pte)))
750 : : goto check_pfn;
751 : : if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
752 : : return NULL;
753 : : if (!is_zero_pfn(pfn))
754 : : print_bad_pte(vma, addr, pte, NULL);
755 : : return NULL;
756 : : }
757 : :
758 : : /* !HAVE_PTE_SPECIAL case follows: */
759 : :
760 [ - + ]: 105154779 : if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
761 [ # # ]: 0 : if (vma->vm_flags & VM_MIXEDMAP) {
762 [ # # ]: 0 : if (!pfn_valid(pfn))
763 : : return NULL;
764 : : goto out;
765 : : } else {
766 : : unsigned long off;
767 : 0 : off = (addr - vma->vm_start) >> PAGE_SHIFT;
768 [ # # ]: 0 : if (pfn == vma->vm_pgoff + off)
769 : : return NULL;
770 [ # # ]: 0 : if (!is_cow_mapping(vma->vm_flags))
771 : : return NULL;
772 : : }
773 : : }
774 : :
775 [ + + ]: 105155810 : if (is_zero_pfn(pfn))
776 : : return NULL;
777 : : check_pfn:
778 [ - + ]: 105067320 : if (unlikely(pfn > highest_memmap_pfn)) {
779 : 0 : print_bad_pte(vma, addr, pte, NULL);
780 : 0 : return NULL;
781 : : }
782 : :
783 : : /*
784 : : * NOTE! We still have PageReserved() pages in the page tables.
785 : : * eg. VDSO mappings can cause them to exist.
786 : : */
787 : : out:
788 : 105066317 : return pfn_to_page(pfn);
789 : : }
790 : :
791 : : /*
792 : : * copy one vm_area from one task to the other. Assumes the page tables
793 : : * already present in the new task to be cleared in the whole range
794 : : * covered by this vma.
795 : : */
796 : :
797 : : static inline unsigned long
798 : : copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
799 : : pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
800 : : unsigned long addr, int *rss)
801 : : {
802 : 23540969 : unsigned long vm_flags = vma->vm_flags;
803 : : pte_t pte = *src_pte;
804 : 23541060 : struct page *page;
805 : :
806 : : /* pte contains position in swap or file, so copy. */
807 [ - + ]: 23540969 : if (unlikely(!pte_present(pte))) {
808 [ # # ]: 0 : if (!pte_file(pte)) {
809 : : swp_entry_t entry = pte_to_swp_entry(pte);
810 : :
811 [ # # ]: 0 : if (swap_duplicate(entry) < 0)
812 : : return entry.val;
813 : :
814 : : /* make sure dst_mm is on swapoff's mmlist. */
815 [ # # ]: 0 : if (unlikely(list_empty(&dst_mm->mmlist))) {
816 : : spin_lock(&mmlist_lock);
817 [ # # ]: 0 : if (list_empty(&dst_mm->mmlist))
818 : 0 : list_add(&dst_mm->mmlist,
819 : : &src_mm->mmlist);
820 : : spin_unlock(&mmlist_lock);
821 : : }
822 [ # # ]: 0 : if (likely(!non_swap_entry(entry)))
823 : 0 : rss[MM_SWAPENTS]++;
824 [ # # ]: 0 : else if (is_migration_entry(entry)) {
825 : : page = migration_entry_to_page(entry);
826 : :
827 [ # # ]: 0 : if (PageAnon(page))
828 : 0 : rss[MM_ANONPAGES]++;
829 : : else
830 : 0 : rss[MM_FILEPAGES]++;
831 : :
832 [ # # ][ # # ]: 0 : if (is_write_migration_entry(entry) &&
833 : : is_cow_mapping(vm_flags)) {
834 : : /*
835 : : * COW mappings require pages in both
836 : : * parent and child to be set to read.
837 : : */
838 : : make_migration_entry_read(&entry);
839 : : pte = swp_entry_to_pte(entry);
840 : : if (pte_swp_soft_dirty(*src_pte))
841 : : pte = pte_swp_mksoft_dirty(pte);
842 : : set_pte_at(src_mm, addr, src_pte, pte);
843 : : }
844 : : }
845 : : }
846 : : goto out_set_pte;
847 : : }
848 : :
849 : : /*
850 : : * If it's a COW mapping, write protect it both
851 : : * in the parent and the child
852 : : */
853 [ + ]: 23540969 : if (is_cow_mapping(vm_flags)) {
854 : : ptep_set_wrprotect(src_mm, addr, src_pte);
855 : : pte = pte_wrprotect(pte);
856 : : }
857 : :
858 : : /*
859 : : * If it's a shared mapping, mark it clean in
860 : : * the child
861 : : */
862 [ - + ]: 23541080 : if (vm_flags & VM_SHARED)
863 : : pte = pte_mkclean(pte);
864 : : pte = pte_mkold(pte);
865 : :
866 : 23541080 : page = vm_normal_page(vma, addr, pte);
867 [ + ]: 23541014 : if (page) {
868 : : get_page(page);
869 : : page_dup_rmap(page);
870 [ + + ]: 23541060 : if (PageAnon(page))
871 : 23540867 : rss[MM_ANONPAGES]++;
872 : : else
873 : 193 : rss[MM_FILEPAGES]++;
874 : : }
875 : :
876 : : out_set_pte:
877 : : set_pte_at(dst_mm, addr, dst_pte, pte);
878 : : return 0;
879 : : }
880 : :
881 : 12237132 : int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
882 : 24475207 : pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
883 : : unsigned long addr, unsigned long end)
884 : : {
885 : : pte_t *orig_src_pte, *orig_dst_pte;
886 : : pte_t *src_pte, *dst_pte;
887 : : spinlock_t *src_ptl, *dst_ptl;
888 : : int progress = 0;
889 : : int rss[NR_MM_COUNTERS];
890 : : swp_entry_t entry = (swp_entry_t){0};
891 : :
892 : : again:
893 : : init_rss_vec(rss);
894 : :
895 [ + + ][ + + ]: 24475203 : dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
896 [ + - ]: 12237606 : if (!dst_pte)
897 : : return -ENOMEM;
898 : 12237606 : src_pte = pte_offset_map(src_pmd, addr);
899 : : src_ptl = pte_lockptr(src_mm, src_pmd);
900 : 12237611 : spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
901 : : orig_src_pte = src_pte;
902 : : orig_dst_pte = dst_pte;
903 : : arch_enter_lazy_mmu_mode();
904 : :
905 : : do {
906 : : /*
907 : : * We are holding two locks at this point - either of them
908 : : * could generate latencies in another task on another CPU.
909 : : */
910 [ + + ]: 120661043 : if (progress >= 32) {
911 : : progress = 0;
912 [ + + ]: 4028469 : if (need_resched() ||
913 : : spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
914 : : break;
915 : : }
916 [ + + ]: 120660587 : if (pte_none(*src_pte)) {
917 : 97119618 : progress++;
918 : 97119618 : continue;
919 : : }
920 : : entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
921 : : vma, addr, rss);
922 [ + - ]: 23541078 : if (entry.val)
923 : : break;
924 : 23541078 : progress += 8;
925 [ + + ]: 120660696 : } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
926 : :
927 : : arch_leave_lazy_mmu_mode();
928 : : spin_unlock(src_ptl);
929 : 12237609 : pte_unmap(orig_src_pte);
930 : : add_mm_rss_vec(dst_mm, rss);
931 : 12237586 : pte_unmap_unlock(orig_dst_pte, dst_ptl);
932 : 12237597 : cond_resched();
933 : :
934 [ - + ]: 12237593 : if (entry.val) {
935 [ # # ]: 0 : if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
936 : : return -ENOMEM;
937 : : progress = 0;
938 : : }
939 [ + + ]: 12237587 : if (addr != end)
940 : : goto again;
941 : : return 0;
942 : : }
943 : :
944 : : static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
945 : : pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
946 : : unsigned long addr, unsigned long end)
947 : : {
948 : : pmd_t *src_pmd, *dst_pmd;
949 : : unsigned long next;
950 : :
951 : : dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
952 [ + - ]: 12259612 : if (!dst_pmd)
953 : : return -ENOMEM;
954 : : src_pmd = pmd_offset(src_pud, addr);
955 : : do {
956 : : next = pmd_addr_end(addr, end);
957 : : if (pmd_trans_huge(*src_pmd)) {
958 : : int err;
959 : : VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
960 : : err = copy_huge_pmd(dst_mm, src_mm,
961 : : dst_pmd, src_pmd, addr, vma);
962 : : if (err == -ENOMEM)
963 : : return -ENOMEM;
964 : : if (!err)
965 : : continue;
966 : : /* fall through */
967 : : }
968 [ + + ]: 12259624 : if (pmd_none_or_clear_bad(src_pmd))
969 : 22493 : continue;
970 [ + ]: 12237131 : if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
971 : : vma, addr, next))
972 : : return -ENOMEM;
973 : : } while (dst_pmd++, src_pmd++, addr = next, addr != end);
974 : : return 0;
975 : : }
976 : :
977 : : static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
978 : : pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
979 : : unsigned long addr, unsigned long end)
980 : : {
981 : : pud_t *src_pud, *dst_pud;
982 : : unsigned long next;
983 : :
984 : : dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
985 [ + + ]: 12259625 : if (!dst_pud)
986 : : return -ENOMEM;
987 : : src_pud = pud_offset(src_pgd, addr);
988 : : do {
989 : : next = pud_addr_end(addr, end);
990 : : if (pud_none_or_clear_bad(src_pud))
991 : : continue;
992 [ + ]: 12259617 : if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
993 : : vma, addr, next))
994 : : return -ENOMEM;
995 : : } while (dst_pud++, src_pud++, addr = next, addr != end);
996 : : return 0;
997 : : }
998 : :
999 : 0 : int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1000 : : struct vm_area_struct *vma)
1001 : : {
1002 : : pgd_t *src_pgd, *dst_pgd;
1003 : : unsigned long next;
1004 : 18090220 : unsigned long addr = vma->vm_start;
1005 : 18090220 : unsigned long end = vma->vm_end;
1006 : : unsigned long mmun_start; /* For mmu_notifiers */
1007 : : unsigned long mmun_end; /* For mmu_notifiers */
1008 : : bool is_cow;
1009 : : int ret;
1010 : :
1011 : : /*
1012 : : * Don't copy ptes where a page fault will fill them correctly.
1013 : : * Fork becomes much lighter when there are big shared or private
1014 : : * readonly mappings. The tradeoff is that copy_page_range is more
1015 : : * efficient than faulting.
1016 : : */
1017 [ + ]: 18090220 : if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR |
1018 : : VM_PFNMAP | VM_MIXEDMAP))) {
1019 [ + ]: 18090250 : if (!vma->anon_vma)
1020 : : return 0;
1021 : : }
1022 : :
1023 : : if (is_vm_hugetlb_page(vma))
1024 : : return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1025 : :
1026 : : if (unlikely(vma->vm_flags & VM_PFNMAP)) {
1027 : : /*
1028 : : * We do not free on error cases below as remove_vma
1029 : : * gets called on error from higher level routine
1030 : : */
1031 : : ret = track_pfn_copy(vma);
1032 : : if (ret)
1033 : : return ret;
1034 : : }
1035 : :
1036 : : /*
1037 : : * We need to invalidate the secondary MMU mappings only when
1038 : : * there could be a permission downgrade on the ptes of the
1039 : : * parent mm. And a permission downgrade will only happen if
1040 : : * is_cow_mapping() returns true.
1041 : : */
1042 : : is_cow = is_cow_mapping(vma->vm_flags);
1043 : : mmun_start = addr;
1044 : : mmun_end = end;
1045 : : if (is_cow)
1046 : : mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1047 : : mmun_end);
1048 : :
1049 : : ret = 0;
1050 : 30286136 : dst_pgd = pgd_offset(dst_mm, addr);
1051 : 30286136 : src_pgd = pgd_offset(src_mm, addr);
1052 : : do {
1053 [ + + ]: 30349845 : next = pgd_addr_end(addr, end);
1054 : : if (pgd_none_or_clear_bad(src_pgd))
1055 : : continue;
1056 [ + + ]: 12259630 : if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1057 : : vma, addr, next))) {
1058 : : ret = -ENOMEM;
1059 : : break;
1060 : : }
1061 [ + + ]: 12259629 : } while (dst_pgd++, src_pgd++, addr = next, addr != end);
1062 : :
1063 : : if (is_cow)
1064 : : mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
1065 : 12195921 : return ret;
1066 : : }
1067 : :
1068 : 0 : static unsigned long zap_pte_range(struct mmu_gather *tlb,
1069 : 19545697 : struct vm_area_struct *vma, pmd_t *pmd,
1070 : : unsigned long addr, unsigned long end,
1071 : : struct zap_details *details)
1072 : : {
1073 : 19536043 : struct mm_struct *mm = tlb->mm;
1074 : : int force_flush = 0;
1075 : : int rss[NR_MM_COUNTERS];
1076 : : spinlock_t *ptl;
1077 : : pte_t *start_pte;
1078 : : pte_t *pte;
1079 : :
1080 : : again:
1081 : : init_rss_vec(rss);
1082 : 19545697 : start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1083 : : pte = start_pte;
1084 : : arch_enter_lazy_mmu_mode();
1085 : : do {
1086 : 468807125 : pte_t ptent = *pte;
1087 [ + + ]: 468807125 : if (pte_none(ptent)) {
1088 : 386579639 : continue;
1089 : : }
1090 : :
1091 [ + - ]: 82227486 : if (pte_present(ptent)) {
1092 : 62666634 : struct page *page;
1093 : :
1094 : 62691443 : page = vm_normal_page(vma, addr, ptent);
1095 [ - + ][ # # ]: 62693911 : if (unlikely(details) && page) {
1096 : : /*
1097 : : * unmap_shared_mapping_pages() wants to
1098 : : * invalidate cache without truncating:
1099 : : * unmap shared but keep private pages.
1100 : : */
1101 [ # # ][ # # ]: 0 : if (details->check_mapping &&
1102 : 0 : details->check_mapping != page->mapping)
1103 : 0 : continue;
1104 : : /*
1105 : : * Each page->index must be checked when
1106 : : * invalidating or truncating nonlinear.
1107 : : */
1108 [ # # ][ # # ]: 0 : if (details->nonlinear_vma &&
1109 [ # # ]: 0 : (page->index < details->first_index ||
1110 : 0 : page->index > details->last_index))
1111 : 0 : continue;
1112 : : }
1113 : : ptent = ptep_get_and_clear_full(mm, addr, pte,
1114 : : tlb->fullmm);
1115 : : tlb_remove_tlb_entry(tlb, pte, addr);
1116 [ + + ]: 62695809 : if (unlikely(!page))
1117 : 31162 : continue;
1118 [ - + ][ # # ]: 62664647 : if (unlikely(details) && details->nonlinear_vma
1119 [ # # ]: 0 : && linear_page_index(details->nonlinear_vma,
1120 : 0 : addr) != page->index) {
1121 : 0 : pte_t ptfile = pgoff_to_pte(page->index);
1122 : : if (pte_soft_dirty(ptent))
1123 : : pte_file_mksoft_dirty(ptfile);
1124 : : set_pte_at(mm, addr, pte, ptfile);
1125 : : }
1126 [ + + ]: 62666634 : if (PageAnon(page))
1127 : 32497519 : rss[MM_ANONPAGES]--;
1128 : : else {
1129 [ + + ]: 30169115 : if (pte_dirty(ptent))
1130 : 274584 : set_page_dirty(page);
1131 [ + + ][ + ]: 30168456 : if (pte_young(ptent) &&
1132 : 30166308 : likely(!(vma->vm_flags & VM_SEQ_READ)))
1133 : 30166450 : mark_page_accessed(page);
1134 : 30170659 : rss[MM_FILEPAGES]--;
1135 : : }
1136 : 62668178 : page_remove_rmap(page);
1137 [ - + ]: 62666517 : if (unlikely(page_mapcount(page) < 0))
1138 : 0 : print_bad_pte(vma, addr, ptent, page);
1139 : 62664285 : force_flush = !__tlb_remove_page(tlb, page);
1140 [ + + ]: 62664285 : if (force_flush)
1141 : : break;
1142 : 62654633 : continue;
1143 : : }
1144 : : /*
1145 : : * If details->check_mapping, we leave swap entries;
1146 : : * if details->nonlinear_vma, we leave file entries.
1147 : : */
1148 [ # # ]: 0 : if (unlikely(details))
1149 : 0 : continue;
1150 [ # # ]: 0 : if (pte_file(ptent)) {
1151 [ # # ]: 0 : if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
1152 : 0 : print_bad_pte(vma, addr, ptent, NULL);
1153 : : } else {
1154 : : swp_entry_t entry = pte_to_swp_entry(ptent);
1155 : :
1156 [ # # ]: 0 : if (!non_swap_entry(entry))
1157 : 0 : rss[MM_SWAPENTS]--;
1158 [ # # ]: 0 : else if (is_migration_entry(entry)) {
1159 : 0 : struct page *page;
1160 : :
1161 : : page = migration_entry_to_page(entry);
1162 : :
1163 [ # # ]: 0 : if (PageAnon(page))
1164 : 0 : rss[MM_ANONPAGES]--;
1165 : : else
1166 : 0 : rss[MM_FILEPAGES]--;
1167 : : }
1168 [ # # ]: 0 : if (unlikely(!free_swap_and_cache(entry)))
1169 : 0 : print_bad_pte(vma, addr, ptent, NULL);
1170 : : }
1171 : : pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1172 [ + + ]: 449226614 : } while (pte++, addr += PAGE_SIZE, addr != end);
1173 : :
1174 : : add_mm_rss_vec(mm, rss);
1175 : : arch_leave_lazy_mmu_mode();
1176 : 19545459 : pte_unmap_unlock(start_pte, ptl);
1177 : :
1178 : : /*
1179 : : * mmu_gather ran out of room to batch pages, we break out of
1180 : : * the PTE lock to avoid doing the potential expensive TLB invalidate
1181 : : * and page-free while holding it.
1182 : : */
1183 [ + + ]: 19545766 : if (force_flush) {
1184 : : unsigned long old_end;
1185 : :
1186 : : force_flush = 0;
1187 : :
1188 : : /*
1189 : : * Flush the TLB just for the previous segment,
1190 : : * then update the range to be the remaining
1191 : : * TLB range.
1192 : : */
1193 : 9474 : old_end = tlb->end;
1194 : 9474 : tlb->end = addr;
1195 : :
1196 : : tlb_flush_mmu(tlb);
1197 : :
1198 : 9474 : tlb->start = addr;
1199 : 9474 : tlb->end = old_end;
1200 : :
1201 [ + - ]: 9474 : if (addr != end)
1202 : : goto again;
1203 : : }
1204 : :
1205 : 19536292 : return addr;
1206 : : }
1207 : :
1208 : : static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
1209 : : struct vm_area_struct *vma, pud_t *pud,
1210 : : unsigned long addr, unsigned long end,
1211 : : struct zap_details *details)
1212 : : {
1213 : : pmd_t *pmd;
1214 : : unsigned long next;
1215 : :
1216 : : pmd = pmd_offset(pud, addr);
1217 : : do {
1218 : : next = pmd_addr_end(addr, end);
1219 : : if (pmd_trans_huge(*pmd)) {
1220 : : if (next - addr != HPAGE_PMD_SIZE) {
1221 : : #ifdef CONFIG_DEBUG_VM
1222 : : if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {
1223 : : pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1224 : : __func__, addr, end,
1225 : : vma->vm_start,
1226 : : vma->vm_end);
1227 : : BUG();
1228 : : }
1229 : : #endif
1230 : : split_huge_page_pmd(vma, addr, pmd);
1231 : : } else if (zap_huge_pmd(tlb, vma, pmd, addr))
1232 : : goto next;
1233 : : /* fall through */
1234 : : }
1235 : : /*
1236 : : * Here there can be other concurrent MADV_DONTNEED or
1237 : : * trans huge page faults running, and if the pmd is
1238 : : * none or trans huge it can change under us. This is
1239 : : * because MADV_DONTNEED holds the mmap_sem in read
1240 : : * mode.
1241 : : */
1242 [ + + ]: 20570718 : if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1243 : : goto next;
1244 : 19536001 : next = zap_pte_range(tlb, vma, pmd, addr, next, details);
1245 : : next:
1246 : 20571114 : cond_resched();
1247 [ + ]: 20571065 : } while (pmd++, addr = next, addr != end);
1248 : :
1249 : : return addr;
1250 : : }
1251 : :
1252 : : static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
1253 : : struct vm_area_struct *vma, pgd_t *pgd,
1254 : : unsigned long addr, unsigned long end,
1255 : : struct zap_details *details)
1256 : : {
1257 : : pud_t *pud;
1258 : : unsigned long next;
1259 : :
1260 : : pud = pud_offset(pgd, addr);
1261 : : do {
1262 : : next = pud_addr_end(addr, end);
1263 : : if (pud_none_or_clear_bad(pud))
1264 : : continue;
1265 : : next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1266 [ + ]: 20571098 : } while (pud++, addr = next, addr != end);
1267 : :
1268 : : return addr;
1269 : : }
1270 : :
1271 : 0 : static void unmap_page_range(struct mmu_gather *tlb,
1272 : : struct vm_area_struct *vma,
1273 : : unsigned long addr, unsigned long end,
1274 : : struct zap_details *details)
1275 : : {
1276 : : pgd_t *pgd;
1277 : : unsigned long next;
1278 : :
1279 [ + + ][ + + ]: 20311069 : if (details && !details->check_mapping && !details->nonlinear_vma)
[ + - ]
1280 : : details = NULL;
1281 : :
1282 [ - + ]: 20311069 : BUG_ON(addr >= end);
1283 : : mem_cgroup_uncharge_start();
1284 : : tlb_start_vma(tlb, vma);
1285 : 20310972 : pgd = pgd_offset(vma->vm_mm, addr);
1286 : : do {
1287 [ + + ]: 20570816 : next = pgd_addr_end(addr, end);
1288 : : if (pgd_none_or_clear_bad(pgd))
1289 : : continue;
1290 : : next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1291 [ + + ]: 20571135 : } while (pgd++, addr = next, addr != end);
1292 : : tlb_end_vma(tlb, vma);
1293 : : mem_cgroup_uncharge_end();
1294 : 20311249 : }
1295 : :
1296 : :
1297 : 0 : static void unmap_single_vma(struct mmu_gather *tlb,
1298 : : struct vm_area_struct *vma, unsigned long start_addr,
1299 : : unsigned long end_addr,
1300 : : struct zap_details *details)
1301 : : {
1302 : 20310966 : unsigned long start = max(vma->vm_start, start_addr);
1303 : : unsigned long end;
1304 : :
1305 [ + ]: 20310966 : if (start >= vma->vm_end)
1306 : : return;
1307 : 20311056 : end = min(vma->vm_end, end_addr);
1308 [ + + ]: 20311056 : if (end <= vma->vm_start)
1309 : : return;
1310 : :
1311 [ + + ]: 20310511 : if (vma->vm_file)
1312 : 13416195 : uprobe_munmap(vma, start, end);
1313 : :
1314 : : if (unlikely(vma->vm_flags & VM_PFNMAP))
1315 : : untrack_pfn(vma, 0, 0);
1316 : :
1317 [ + - ]: 20310893 : if (start != end) {
1318 : : if (unlikely(is_vm_hugetlb_page(vma))) {
1319 : : /*
1320 : : * It is undesirable to test vma->vm_file as it
1321 : : * should be non-null for valid hugetlb area.
1322 : : * However, vm_file will be NULL in the error
1323 : : * cleanup path of do_mmap_pgoff. When
1324 : : * hugetlbfs ->mmap method fails,
1325 : : * do_mmap_pgoff() nullifies vma->vm_file
1326 : : * before calling this function to clean up.
1327 : : * Since no pte has actually been setup, it is
1328 : : * safe to do nothing in this case.
1329 : : */
1330 : : if (vma->vm_file) {
1331 : : mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
1332 : : __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
1333 : : mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
1334 : : }
1335 : : } else
1336 : 20310893 : unmap_page_range(tlb, vma, start, end, details);
1337 : : }
1338 : : }
1339 : :
1340 : : /**
1341 : : * unmap_vmas - unmap a range of memory covered by a list of vma's
1342 : : * @tlb: address of the caller's struct mmu_gather
1343 : : * @vma: the starting vma
1344 : : * @start_addr: virtual address at which to start unmapping
1345 : : * @end_addr: virtual address at which to end unmapping
1346 : : *
1347 : : * Unmap all pages in the vma list.
1348 : : *
1349 : : * Only addresses between `start' and `end' will be unmapped.
1350 : : *
1351 : : * The VMA list must be sorted in ascending virtual address order.
1352 : : *
1353 : : * unmap_vmas() assumes that the caller will flush the whole unmapped address
1354 : : * range after unmap_vmas() returns. So the only responsibility here is to
1355 : : * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1356 : : * drops the lock and schedules.
1357 : : */
1358 : 1536898 : void unmap_vmas(struct mmu_gather *tlb,
1359 : : struct vm_area_struct *vma, unsigned long start_addr,
1360 : : unsigned long end_addr)
1361 : : {
1362 : : struct mm_struct *mm = vma->vm_mm;
1363 : :
1364 : : mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
1365 [ + + ][ + + ]: 21842350 : for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
1366 : 20305329 : unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
1367 : : mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
1368 : 123 : }
1369 : :
1370 : : /**
1371 : : * zap_page_range - remove user pages in a given range
1372 : : * @vma: vm_area_struct holding the applicable pages
1373 : : * @start: starting address of pages to zap
1374 : : * @size: number of bytes to zap
1375 : : * @details: details of nonlinear truncation or shared cache invalidation
1376 : : *
1377 : : * Caller must protect the VMA list
1378 : : */
1379 : 0 : void zap_page_range(struct vm_area_struct *vma, unsigned long start,
1380 : : unsigned long size, struct zap_details *details)
1381 : : {
1382 : 5519 : struct mm_struct *mm = vma->vm_mm;
1383 : : struct mmu_gather tlb;
1384 : 5519 : unsigned long end = start + size;
1385 : :
1386 : 5519 : lru_add_drain();
1387 : : tlb_gather_mmu(&tlb, mm, start, end);
1388 : : update_hiwater_rss(mm);
1389 : : mmu_notifier_invalidate_range_start(mm, start, end);
1390 [ + ][ + + ]: 11037 : for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
1391 : 5519 : unmap_single_vma(&tlb, vma, start, end, details);
1392 : : mmu_notifier_invalidate_range_end(mm, start, end);
1393 : : tlb_finish_mmu(&tlb, start, end);
1394 : 5517 : }
1395 : :
1396 : : /**
1397 : : * zap_page_range_single - remove user pages in a given range
1398 : : * @vma: vm_area_struct holding the applicable pages
1399 : : * @address: starting address of pages to zap
1400 : : * @size: number of bytes to zap
1401 : : * @details: details of nonlinear truncation or shared cache invalidation
1402 : : *
1403 : : * The range must fit into one VMA.
1404 : : */
1405 : 0 : static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1406 : : unsigned long size, struct zap_details *details)
1407 : : {
1408 : 5 : struct mm_struct *mm = vma->vm_mm;
1409 : : struct mmu_gather tlb;
1410 : 5 : unsigned long end = address + size;
1411 : :
1412 : 5 : lru_add_drain();
1413 : : tlb_gather_mmu(&tlb, mm, address, end);
1414 : : update_hiwater_rss(mm);
1415 : : mmu_notifier_invalidate_range_start(mm, address, end);
1416 : 5 : unmap_single_vma(&tlb, vma, address, end, details);
1417 : : mmu_notifier_invalidate_range_end(mm, address, end);
1418 : : tlb_finish_mmu(&tlb, address, end);
1419 : 5 : }
1420 : :
1421 : : /**
1422 : : * zap_vma_ptes - remove ptes mapping the vma
1423 : : * @vma: vm_area_struct holding ptes to be zapped
1424 : : * @address: starting address of pages to zap
1425 : : * @size: number of bytes to zap
1426 : : *
1427 : : * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1428 : : *
1429 : : * The entire address range must be fully contained within the vma.
1430 : : *
1431 : : * Returns 0 if successful.
1432 : : */
1433 : 0 : int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1434 : : unsigned long size)
1435 : : {
1436 [ # # ][ # # ]: 0 : if (address < vma->vm_start || address + size > vma->vm_end ||
[ # # ]
1437 : 0 : !(vma->vm_flags & VM_PFNMAP))
1438 : : return -1;
1439 : 0 : zap_page_range_single(vma, address, size, NULL);
1440 : 0 : return 0;
1441 : : }
1442 : : EXPORT_SYMBOL_GPL(zap_vma_ptes);
1443 : :
1444 : : /**
1445 : : * follow_page_mask - look up a page descriptor from a user-virtual address
1446 : : * @vma: vm_area_struct mapping @address
1447 : : * @address: virtual address to look up
1448 : : * @flags: flags modifying lookup behaviour
1449 : : * @page_mask: on output, *page_mask is set according to the size of the page
1450 : : *
1451 : : * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1452 : : *
1453 : : * Returns the mapped (struct page *), %NULL if no mapping exists, or
1454 : : * an error pointer if there is a mapping to something not represented
1455 : : * by a page descriptor (see also vm_normal_page()).
1456 : : */
1457 : 0 : struct page *follow_page_mask(struct vm_area_struct *vma,
1458 : : unsigned long address, unsigned int flags,
1459 : : unsigned int *page_mask)
1460 : : {
1461 : : pgd_t *pgd;
1462 : : pud_t *pud;
1463 : : pmd_t *pmd;
1464 : : pte_t *ptep, pte;
1465 : : spinlock_t *ptl;
1466 : : struct page *page;
1467 : 4604771 : struct mm_struct *mm = vma->vm_mm;
1468 : :
1469 : 4604771 : *page_mask = 0;
1470 : :
1471 : : page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1472 : : if (!IS_ERR(page)) {
1473 : : BUG_ON(flags & FOLL_GET);
1474 : : goto out;
1475 : : }
1476 : :
1477 : : page = NULL;
1478 : 4604771 : pgd = pgd_offset(mm, address);
1479 : : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
1480 : : goto no_page_table;
1481 : :
1482 : : pud = pud_offset(pgd, address);
1483 : : if (pud_none(*pud))
1484 : : goto no_page_table;
1485 : : if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
1486 : : if (flags & FOLL_GET)
1487 : : goto out;
1488 : : page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1489 : : goto out;
1490 : : }
1491 : : if (unlikely(pud_bad(*pud)))
1492 : : goto no_page_table;
1493 : :
1494 : : pmd = pmd_offset(pud, address);
1495 [ + + ]: 4604771 : if (pmd_none(*pmd))
1496 : : goto no_page_table;
1497 : : if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
1498 : : page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1499 : : if (flags & FOLL_GET) {
1500 : : /*
1501 : : * Refcount on tail pages are not well-defined and
1502 : : * shouldn't be taken. The caller should handle a NULL
1503 : : * return when trying to follow tail pages.
1504 : : */
1505 : : if (PageHead(page))
1506 : : get_page(page);
1507 : : else {
1508 : : page = NULL;
1509 : : goto out;
1510 : : }
1511 : : }
1512 : : goto out;
1513 : : }
1514 : : if ((flags & FOLL_NUMA) && pmd_numa(*pmd))
1515 : : goto no_page_table;
1516 : : if (pmd_trans_huge(*pmd)) {
1517 : : if (flags & FOLL_SPLIT) {
1518 : : split_huge_page_pmd(vma, address, pmd);
1519 : : goto split_fallthrough;
1520 : : }
1521 : : ptl = pmd_lock(mm, pmd);
1522 : : if (likely(pmd_trans_huge(*pmd))) {
1523 : : if (unlikely(pmd_trans_splitting(*pmd))) {
1524 : : spin_unlock(ptl);
1525 : : wait_split_huge_page(vma->anon_vma, pmd);
1526 : : } else {
1527 : : page = follow_trans_huge_pmd(vma, address,
1528 : : pmd, flags);
1529 : : spin_unlock(ptl);
1530 : : *page_mask = HPAGE_PMD_NR - 1;
1531 : : goto out;
1532 : : }
1533 : : } else
1534 : : spin_unlock(ptl);
1535 : : /* fall through */
1536 : : }
1537 : : split_fallthrough:
1538 [ + + ]: 9178568 : if (unlikely(pmd_bad(*pmd)))
1539 : : goto no_page_table;
1540 : :
1541 : 4573796 : ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
1542 : :
1543 : 4573799 : pte = *ptep;
1544 [ + + ]: 4573799 : if (!pte_present(pte)) {
1545 : : swp_entry_t entry;
1546 : : /*
1547 : : * KSM's break_ksm() relies upon recognizing a ksm page
1548 : : * even while it is being migrated, so for that case we
1549 : : * need migration_entry_wait().
1550 : : */
1551 [ - + ]: 2014715 : if (likely(!(flags & FOLL_MIGRATION)))
1552 : : goto no_page;
1553 [ # # ][ # # ]: 0 : if (pte_none(pte) || pte_file(pte))
1554 : : goto no_page;
1555 : : entry = pte_to_swp_entry(pte);
1556 [ # # ]: 0 : if (!is_migration_entry(entry))
1557 : : goto no_page;
1558 : 0 : pte_unmap_unlock(ptep, ptl);
1559 : 0 : migration_entry_wait(mm, pmd, address);
1560 : : goto split_fallthrough;
1561 : : }
1562 : : if ((flags & FOLL_NUMA) && pte_numa(pte))
1563 : : goto no_page;
1564 [ + + ][ + + ]: 2559084 : if ((flags & FOLL_WRITE) && !pte_write(pte))
1565 : : goto unlock;
1566 : :
1567 : 2559081 : page = vm_normal_page(vma, address, pte);
1568 [ + + ]: 2559081 : if (unlikely(!page)) {
1569 [ + - ][ + - ]: 2 : if ((flags & FOLL_DUMP) ||
1570 : 2 : !is_zero_pfn(pte_pfn(pte)))
1571 : : goto bad_page;
1572 : 2 : page = pte_page(pte);
1573 : : }
1574 : :
1575 [ + + ]: 2559081 : if (flags & FOLL_GET)
1576 : : get_page_foll(page);
1577 [ + + ]: 2559078 : if (flags & FOLL_TOUCH) {
1578 [ + + ][ - + ]: 2557625 : if ((flags & FOLL_WRITE) &&
1579 [ # # ]: 0 : !pte_dirty(pte) && !PageDirty(page))
1580 : 0 : set_page_dirty(page);
1581 : : /*
1582 : : * pte_mkyoung() would be more correct here, but atomic care
1583 : : * is needed to avoid losing the dirty bit: it is easier to use
1584 : : * mark_page_accessed().
1585 : : */
1586 : 2557625 : mark_page_accessed(page);
1587 : : }
1588 [ + + ][ + + ]: 2559082 : if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
1589 : : /*
1590 : : * The preliminary mapping check is mainly to avoid the
1591 : : * pointless overhead of lock_page on the ZERO_PAGE
1592 : : * which might bounce very badly if there is contention.
1593 : : *
1594 : : * If the page is already locked, we don't need to
1595 : : * handle it now - vmscan will handle it later if and
1596 : : * when it attempts to reclaim the page.
1597 : : */
1598 [ + - + - ]: 20808 : if (page->mapping && trylock_page(page)) {
1599 : 10404 : lru_add_drain(); /* push cached pages to LRU */
1600 : : /*
1601 : : * Because we lock page here, and migration is
1602 : : * blocked by the pte's page reference, and we
1603 : : * know the page is still mapped, we don't even
1604 : : * need to check for file-cache page truncation.
1605 : : */
1606 : 10404 : mlock_vma_page(page);
1607 : 10404 : unlock_page(page);
1608 : : }
1609 : : }
1610 : : unlock:
1611 : 2559085 : pte_unmap_unlock(ptep, ptl);
1612 : : out:
1613 : 2559085 : return page;
1614 : :
1615 : : bad_page:
1616 : 0 : pte_unmap_unlock(ptep, ptl);
1617 : 0 : return ERR_PTR(-EFAULT);
1618 : :
1619 : : no_page:
1620 : 2014715 : pte_unmap_unlock(ptep, ptl);
1621 [ + + ]: 2014715 : if (!pte_none(pte))
1622 : : return page;
1623 : :
1624 : : no_page_table:
1625 : : /*
1626 : : * When core dumping an enormous anonymous area that nobody
1627 : : * has touched so far, we don't want to allocate unnecessary pages or
1628 : : * page tables. Return error instead of NULL to skip handle_mm_fault,
1629 : : * then get_dump_page() will return NULL to leave a hole in the dump.
1630 : : * But we can only make this optimization where a hole would surely
1631 : : * be zero-filled if handle_mm_fault() actually did handle it.
1632 : : */
1633 [ + + ][ + + ]: 2045659 : if ((flags & FOLL_DUMP) &&
1634 [ + - ]: 109 : (!vma->vm_ops || !vma->vm_ops->fault))
1635 : : return ERR_PTR(-EFAULT);
1636 : 2044512 : return page;
1637 : : }
1638 : :
1639 : : static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)
1640 : : {
1641 [ + + ]: 58313 : return stack_guard_page_start(vma, addr) ||
1642 : : stack_guard_page_end(vma, addr+PAGE_SIZE);
1643 : : }
1644 : :
1645 : : /**
1646 : : * __get_user_pages() - pin user pages in memory
1647 : : * @tsk: task_struct of target task
1648 : : * @mm: mm_struct of target mm
1649 : : * @start: starting user address
1650 : : * @nr_pages: number of pages from start to pin
1651 : : * @gup_flags: flags modifying pin behaviour
1652 : : * @pages: array that receives pointers to the pages pinned.
1653 : : * Should be at least nr_pages long. Or NULL, if caller
1654 : : * only intends to ensure the pages are faulted in.
1655 : : * @vmas: array of pointers to vmas corresponding to each page.
1656 : : * Or NULL if the caller does not require them.
1657 : : * @nonblocking: whether waiting for disk IO or mmap_sem contention
1658 : : *
1659 : : * Returns number of pages pinned. This may be fewer than the number
1660 : : * requested. If nr_pages is 0 or negative, returns 0. If no pages
1661 : : * were pinned, returns -errno. Each page returned must be released
1662 : : * with a put_page() call when it is finished with. vmas will only
1663 : : * remain valid while mmap_sem is held.
1664 : : *
1665 : : * Must be called with mmap_sem held for read or write.
1666 : : *
1667 : : * __get_user_pages walks a process's page tables and takes a reference to
1668 : : * each struct page that each user address corresponds to at a given
1669 : : * instant. That is, it takes the page that would be accessed if a user
1670 : : * thread accesses the given user virtual address at that instant.
1671 : : *
1672 : : * This does not guarantee that the page exists in the user mappings when
1673 : : * __get_user_pages returns, and there may even be a completely different
1674 : : * page there in some cases (eg. if mmapped pagecache has been invalidated
1675 : : * and subsequently re faulted). However it does guarantee that the page
1676 : : * won't be freed completely. And mostly callers simply care that the page
1677 : : * contains data that was valid *at some point in time*. Typically, an IO
1678 : : * or similar operation cannot guarantee anything stronger anyway because
1679 : : * locks can't be held over the syscall boundary.
1680 : : *
1681 : : * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1682 : : * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1683 : : * appropriate) must be called after the page is finished with, and
1684 : : * before put_page is called.
1685 : : *
1686 : : * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
1687 : : * or mmap_sem contention, and if waiting is needed to pin all pages,
1688 : : * *@nonblocking will be set to 0.
1689 : : *
1690 : : * In most cases, get_user_pages or get_user_pages_fast should be used
1691 : : * instead of __get_user_pages. __get_user_pages should be used only if
1692 : : * you need some special @gup_flags.
1693 : : */
1694 : 0 : long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1695 : : unsigned long start, unsigned long nr_pages,
1696 : : unsigned int gup_flags, struct page **pages,
1697 : : struct vm_area_struct **vmas, int *nonblocking)
1698 : : {
1699 : : long i;
1700 : : unsigned long vm_flags;
1701 : : unsigned int page_mask;
1702 : :
1703 [ + + ]: 2410700 : if (!nr_pages)
1704 : : return 0;
1705 : :
1706 : : VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1707 : :
1708 : : /*
1709 : : * Require read or write permissions.
1710 : : * If FOLL_FORCE is set, we only require the "MAY" flags.
1711 : : */
1712 [ + + ]: 2410697 : vm_flags = (gup_flags & FOLL_WRITE) ?
1713 : : (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1714 [ + + ]: 2410697 : vm_flags &= (gup_flags & FOLL_FORCE) ?
1715 : : (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1716 : :
1717 : : /*
1718 : : * If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault
1719 : : * would be called on PROT_NONE ranges. We must never invoke
1720 : : * handle_mm_fault on PROT_NONE ranges or the NUMA hinting
1721 : : * page faults would unprotect the PROT_NONE ranges if
1722 : : * _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd
1723 : : * bitflag. So to avoid that, don't set FOLL_NUMA if
1724 : : * FOLL_FORCE is set.
1725 : : */
1726 [ + + ]: 2410697 : if (!(gup_flags & FOLL_FORCE))
1727 : 2410697 : gup_flags |= FOLL_NUMA;
1728 : :
1729 : : i = 0;
1730 : :
1731 : : do {
1732 : : struct vm_area_struct *vma;
1733 : :
1734 : 2410697 : vma = find_extend_vma(mm, start);
1735 [ + + ][ + + ]: 2410700 : if (!vma && in_gate_area(mm, start)) {
1736 : 18 : unsigned long pg = start & PAGE_MASK;
1737 : : pgd_t *pgd;
1738 : : pud_t *pud;
1739 : : pmd_t *pmd;
1740 : : pte_t *pte;
1741 : :
1742 : : /* user gate pages are read-only */
1743 [ - + ]: 18 : if (gup_flags & FOLL_WRITE)
1744 [ # # ]: 0 : return i ? : -EFAULT;
1745 [ + - ]: 18 : if (pg > TASK_SIZE)
1746 : 18 : pgd = pgd_offset_k(pg);
1747 : : else
1748 : 0 : pgd = pgd_offset_gate(mm, pg);
1749 : : BUG_ON(pgd_none(*pgd));
1750 : : pud = pud_offset(pgd, pg);
1751 : : BUG_ON(pud_none(*pud));
1752 : : pmd = pmd_offset(pud, pg);
1753 [ - + ]: 18 : if (pmd_none(*pmd))
1754 [ # # ]: 0 : return i ? : -EFAULT;
1755 : : VM_BUG_ON(pmd_trans_huge(*pmd));
1756 : 18 : pte = pte_offset_map(pmd, pg);
1757 [ - + ]: 18 : if (pte_none(*pte)) {
1758 : 0 : pte_unmap(pte);
1759 [ # # ]: 0 : return i ? : -EFAULT;
1760 : : }
1761 : 18 : vma = get_gate_vma(mm);
1762 [ + - ]: 18 : if (pages) {
1763 : : struct page *page;
1764 : :
1765 : 18 : page = vm_normal_page(vma, start, *pte);
1766 [ - + ]: 18 : if (!page) {
1767 [ # # ][ # # ]: 0 : if (!(gup_flags & FOLL_DUMP) &&
1768 : 0 : is_zero_pfn(pte_pfn(*pte)))
1769 : 0 : page = pte_page(*pte);
1770 : : else {
1771 : 0 : pte_unmap(pte);
1772 [ # # ]: 0 : return i ? : -EFAULT;
1773 : : }
1774 : : }
1775 : 18 : pages[i] = page;
1776 : : get_page(page);
1777 : : }
1778 : 18 : pte_unmap(pte);
1779 : 18 : page_mask = 0;
1780 : 18 : goto next_page;
1781 : : }
1782 : :
1783 [ + + ][ + + ]: 2410682 : if (!vma ||
1784 [ + + ]: 2410673 : (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1785 : 2410673 : !(vm_flags & vma->vm_flags))
1786 [ - + ]: 24 : return i ? : -EFAULT;
1787 : :
1788 : : if (is_vm_hugetlb_page(vma)) {
1789 : : i = follow_hugetlb_page(mm, vma, pages, vmas,
1790 : : &start, &nr_pages, i, gup_flags);
1791 : : continue;
1792 : : }
1793 : :
1794 : : do {
1795 : : struct page *page;
1796 : : unsigned int foll_flags = gup_flags;
1797 : : unsigned int page_increm;
1798 : :
1799 : : /*
1800 : : * If we have a pending SIGKILL, don't keep faulting
1801 : : * pages and potentially allocating memory.
1802 : : */
1803 [ - + ]: 2561197 : if (unlikely(fatal_signal_pending(current)))
1804 [ # # ]: 0 : return i ? i : -ERESTARTSYS;
1805 : :
1806 : 2561197 : cond_resched();
1807 [ + + ]: 4603602 : while (!(page = follow_page_mask(vma, start,
1808 : : foll_flags, &page_mask))) {
1809 : : int ret;
1810 : : unsigned int fault_flags = 0;
1811 : :
1812 : : /* For mlock, just skip the stack guard page. */
1813 [ + + ]: 2044454 : if (foll_flags & FOLL_MLOCK) {
1814 [ + ]: 58313 : if (stack_guard_page(vma, start))
1815 : : goto next_page;
1816 : : }
1817 [ + ]: 2044458 : if (foll_flags & FOLL_WRITE)
1818 : : fault_flags |= FAULT_FLAG_WRITE;
1819 [ + ]: 0 : if (nonblocking)
1820 : 58310 : fault_flags |= FAULT_FLAG_ALLOW_RETRY;
1821 [ - ]: 0 : if (foll_flags & FOLL_NOWAIT)
1822 : 0 : fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);
1823 : :
1824 : 0 : ret = handle_mm_fault(mm, vma, start,
1825 : : fault_flags);
1826 : :
1827 [ - + ]: 2044452 : if (ret & VM_FAULT_ERROR) {
1828 [ # # ]: 0 : if (ret & VM_FAULT_OOM)
1829 [ # # ]: 0 : return i ? i : -ENOMEM;
1830 [ # # ]: 0 : if (ret & (VM_FAULT_HWPOISON |
1831 : : VM_FAULT_HWPOISON_LARGE)) {
1832 [ # # ]: 0 : if (i)
1833 : : return i;
1834 [ # # ]: 0 : else if (gup_flags & FOLL_HWPOISON)
1835 : : return -EHWPOISON;
1836 : : else
1837 : 0 : return -EFAULT;
1838 : : }
1839 [ # # ]: 0 : if (ret & VM_FAULT_SIGBUS)
1840 [ # # ]: 0 : return i ? i : -EFAULT;
1841 : 0 : BUG();
1842 : : }
1843 : :
1844 [ + ]: 2044452 : if (tsk) {
1845 [ + + ]: 2044454 : if (ret & VM_FAULT_MAJOR)
1846 : 8 : tsk->maj_flt++;
1847 : : else
1848 : 2044446 : tsk->min_flt++;
1849 : : }
1850 : :
1851 [ + + ]: 2044452 : if (ret & VM_FAULT_RETRY) {
1852 [ + - ]: 2052 : if (nonblocking)
1853 : 2052 : *nonblocking = 0;
1854 : 2052 : return i;
1855 : : }
1856 : :
1857 : : /*
1858 : : * The VM_FAULT_WRITE bit tells us that
1859 : : * do_wp_page has broken COW when necessary,
1860 : : * even if maybe_mkwrite decided not to set
1861 : : * pte_write. We can thus safely do subsequent
1862 : : * page lookups as if they were reads. But only
1863 : : * do so when looping for pte_write is futile:
1864 : : * in some cases userspace may also be wanting
1865 : : * to write to the gotten user page, which a
1866 : : * read fault here might prevent (a readonly
1867 : : * page might get reCOWed by userspace write).
1868 : : */
1869 [ + + ][ - + ]: 2042400 : if ((ret & VM_FAULT_WRITE) &&
1870 : 2 : !(vma->vm_flags & VM_WRITE))
1871 : 0 : foll_flags &= ~FOLL_WRITE;
1872 : :
1873 : 2042400 : cond_resched();
1874 : : }
1875 [ + + ]: 2559148 : if (IS_ERR(page))
1876 [ + - ]: 2290 : return i ? i : PTR_ERR(page);
1877 [ + + ]: 2558003 : if (pages) {
1878 : 2498107 : pages[i] = page;
1879 : :
1880 : : flush_anon_page(vma, page, start);
1881 : 2498113 : flush_dcache_page(page);
1882 : 2498109 : page_mask = 0;
1883 : : }
1884 : : next_page:
1885 [ + + ]: 2558019 : if (vmas) {
1886 : 453 : vmas[i] = vma;
1887 : 453 : page_mask = 0;
1888 : : }
1889 : 2558019 : page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
1890 [ - + ]: 2558019 : if (page_increm > nr_pages)
1891 : : page_increm = nr_pages;
1892 : 2558019 : i += page_increm;
1893 : 2558019 : start += page_increm * PAGE_SIZE;
1894 : 2558019 : nr_pages -= page_increm;
1895 [ + + ][ + + ]: 2558019 : } while (nr_pages && start < vma->vm_end);
1896 [ - + ]: 2407480 : } while (nr_pages);
1897 : : return i;
1898 : : }
1899 : : EXPORT_SYMBOL(__get_user_pages);
1900 : :
1901 : : /*
1902 : : * fixup_user_fault() - manually resolve a user page fault
1903 : : * @tsk: the task_struct to use for page fault accounting, or
1904 : : * NULL if faults are not to be recorded.
1905 : : * @mm: mm_struct of target mm
1906 : : * @address: user address
1907 : : * @fault_flags:flags to pass down to handle_mm_fault()
1908 : : *
1909 : : * This is meant to be called in the specific scenario where for locking reasons
1910 : : * we try to access user memory in atomic context (within a pagefault_disable()
1911 : : * section), this returns -EFAULT, and we want to resolve the user fault before
1912 : : * trying again.
1913 : : *
1914 : : * Typically this is meant to be used by the futex code.
1915 : : *
1916 : : * The main difference with get_user_pages() is that this function will
1917 : : * unconditionally call handle_mm_fault() which will in turn perform all the
1918 : : * necessary SW fixup of the dirty and young bits in the PTE, while
1919 : : * handle_mm_fault() only guarantees to update these in the struct page.
1920 : : *
1921 : : * This is important for some architectures where those bits also gate the
1922 : : * access permission to the page because they are maintained in software. On
1923 : : * such architectures, gup() will not be enough to make a subsequent access
1924 : : * succeed.
1925 : : *
1926 : : * This should be called with the mm_sem held for read.
1927 : : */
1928 : 0 : int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
1929 : : unsigned long address, unsigned int fault_flags)
1930 : : {
1931 : : struct vm_area_struct *vma;
1932 : : int ret;
1933 : :
1934 : 0 : vma = find_extend_vma(mm, address);
1935 [ # # ][ # # ]: 0 : if (!vma || address < vma->vm_start)
1936 : : return -EFAULT;
1937 : :
1938 : 0 : ret = handle_mm_fault(mm, vma, address, fault_flags);
1939 [ # # ]: 0 : if (ret & VM_FAULT_ERROR) {
1940 [ # # ]: 0 : if (ret & VM_FAULT_OOM)
1941 : : return -ENOMEM;
1942 [ # # ]: 0 : if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
1943 : : return -EHWPOISON;
1944 [ # # ]: 0 : if (ret & VM_FAULT_SIGBUS)
1945 : : return -EFAULT;
1946 : 0 : BUG();
1947 : : }
1948 [ # # ]: 0 : if (tsk) {
1949 [ # # ]: 0 : if (ret & VM_FAULT_MAJOR)
1950 : 0 : tsk->maj_flt++;
1951 : : else
1952 : 0 : tsk->min_flt++;
1953 : : }
1954 : : return 0;
1955 : : }
1956 : :
1957 : : /*
1958 : : * get_user_pages() - pin user pages in memory
1959 : : * @tsk: the task_struct to use for page fault accounting, or
1960 : : * NULL if faults are not to be recorded.
1961 : : * @mm: mm_struct of target mm
1962 : : * @start: starting user address
1963 : : * @nr_pages: number of pages from start to pin
1964 : : * @write: whether pages will be written to by the caller
1965 : : * @force: whether to force write access even if user mapping is
1966 : : * readonly. This will result in the page being COWed even
1967 : : * in MAP_SHARED mappings. You do not want this.
1968 : : * @pages: array that receives pointers to the pages pinned.
1969 : : * Should be at least nr_pages long. Or NULL, if caller
1970 : : * only intends to ensure the pages are faulted in.
1971 : : * @vmas: array of pointers to vmas corresponding to each page.
1972 : : * Or NULL if the caller does not require them.
1973 : : *
1974 : : * Returns number of pages pinned. This may be fewer than the number
1975 : : * requested. If nr_pages is 0 or negative, returns 0. If no pages
1976 : : * were pinned, returns -errno. Each page returned must be released
1977 : : * with a put_page() call when it is finished with. vmas will only
1978 : : * remain valid while mmap_sem is held.
1979 : : *
1980 : : * Must be called with mmap_sem held for read or write.
1981 : : *
1982 : : * get_user_pages walks a process's page tables and takes a reference to
1983 : : * each struct page that each user address corresponds to at a given
1984 : : * instant. That is, it takes the page that would be accessed if a user
1985 : : * thread accesses the given user virtual address at that instant.
1986 : : *
1987 : : * This does not guarantee that the page exists in the user mappings when
1988 : : * get_user_pages returns, and there may even be a completely different
1989 : : * page there in some cases (eg. if mmapped pagecache has been invalidated
1990 : : * and subsequently re faulted). However it does guarantee that the page
1991 : : * won't be freed completely. And mostly callers simply care that the page
1992 : : * contains data that was valid *at some point in time*. Typically, an IO
1993 : : * or similar operation cannot guarantee anything stronger anyway because
1994 : : * locks can't be held over the syscall boundary.
1995 : : *
1996 : : * If write=0, the page must not be written to. If the page is written to,
1997 : : * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1998 : : * after the page is finished with, and before put_page is called.
1999 : : *
2000 : : * get_user_pages is typically used for fewer-copy IO operations, to get a
2001 : : * handle on the memory by some means other than accesses via the user virtual
2002 : : * addresses. The pages may be submitted for DMA to devices or accessed via
2003 : : * their kernel linear mapping (via the kmap APIs). Care should be taken to
2004 : : * use the correct cache flushing APIs.
2005 : : *
2006 : : * See also get_user_pages_fast, for performance critical applications.
2007 : : */
2008 : 0 : long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
2009 : : unsigned long start, unsigned long nr_pages, int write,
2010 : : int force, struct page **pages, struct vm_area_struct **vmas)
2011 : : {
2012 : : int flags = FOLL_TOUCH;
2013 : :
2014 [ + - ]: 2406668 : if (pages)
2015 : : flags |= FOLL_GET;
2016 [ + + ]: 2406668 : if (write)
2017 : 2191606 : flags |= FOLL_WRITE;
2018 [ # # ]: 2406668 : if (force)
2019 : 2041575 : flags |= FOLL_FORCE;
2020 : :
2021 : 0 : return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
2022 : : NULL);
2023 : : }
2024 : : EXPORT_SYMBOL(get_user_pages);
2025 : :
2026 : : /**
2027 : : * get_dump_page() - pin user page in memory while writing it to core dump
2028 : : * @addr: user address
2029 : : *
2030 : : * Returns struct page pointer of user page pinned for dump,
2031 : : * to be freed afterwards by page_cache_release() or put_page().
2032 : : *
2033 : : * Returns NULL on any kind of failure - a hole must then be inserted into
2034 : : * the corefile, to preserve alignment with its headers; and also returns
2035 : : * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
2036 : : * allowing a hole to be left in the corefile to save diskspace.
2037 : : *
2038 : : * Called without mmap_sem, but after all other threads have been killed.
2039 : : */
2040 : : #ifdef CONFIG_ELF_CORE
2041 : 0 : struct page *get_dump_page(unsigned long addr)
2042 : : {
2043 : : struct vm_area_struct *vma;
2044 : : struct page *page;
2045 : :
2046 [ + + ]: 1538 : if (__get_user_pages(current, current->mm, addr, 1,
2047 : : FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
2048 : : NULL) < 1)
2049 : : return NULL;
2050 : 393 : flush_cache_page(vma, addr, page_to_pfn(page));
2051 : 393 : return page;
2052 : : }
2053 : : #endif /* CONFIG_ELF_CORE */
2054 : :
2055 : 0 : pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
2056 : : spinlock_t **ptl)
2057 : : {
2058 : 1111 : pgd_t * pgd = pgd_offset(mm, addr);
2059 : : pud_t * pud = pud_alloc(mm, pgd, addr);
2060 [ + - ]: 1111 : if (pud) {
2061 : 1111 : pmd_t * pmd = pmd_alloc(mm, pud, addr);
2062 [ + - ]: 1111 : if (pmd) {
2063 : : VM_BUG_ON(pmd_trans_huge(*pmd));
2064 [ + + ][ + - ]: 2222 : return pte_alloc_map_lock(mm, pmd, addr, ptl);
2065 : : }
2066 : : }
2067 : : return NULL;
2068 : : }
2069 : :
2070 : : /*
2071 : : * This is the old fallback for page remapping.
2072 : : *
2073 : : * For historical reasons, it only allows reserved pages. Only
2074 : : * old drivers should use this, and they needed to mark their
2075 : : * pages reserved for the old functions anyway.
2076 : : */
2077 : 0 : static int insert_page(struct vm_area_struct *vma, unsigned long addr,
2078 : 0 : struct page *page, pgprot_t prot)
2079 : : {
2080 : 0 : struct mm_struct *mm = vma->vm_mm;
2081 : : int retval;
2082 : : pte_t *pte;
2083 : : spinlock_t *ptl;
2084 : :
2085 : : retval = -EINVAL;
2086 [ # # ]: 0 : if (PageAnon(page))
2087 : : goto out;
2088 : : retval = -ENOMEM;
2089 : 0 : flush_dcache_page(page);
2090 : : pte = get_locked_pte(mm, addr, &ptl);
2091 [ # # ]: 0 : if (!pte)
2092 : : goto out;
2093 : : retval = -EBUSY;
2094 [ # # ]: 0 : if (!pte_none(*pte))
2095 : : goto out_unlock;
2096 : :
2097 : : /* Ok, finally just insert the thing.. */
2098 : : get_page(page);
2099 : 0 : inc_mm_counter_fast(mm, MM_FILEPAGES);
2100 : 0 : page_add_file_rmap(page);
2101 : 0 : set_pte_at(mm, addr, pte, mk_pte(page, prot));
2102 : :
2103 : : retval = 0;
2104 : 0 : pte_unmap_unlock(pte, ptl);
2105 : : return retval;
2106 : : out_unlock:
2107 : 0 : pte_unmap_unlock(pte, ptl);
2108 : : out:
2109 : : return retval;
2110 : : }
2111 : :
2112 : : /**
2113 : : * vm_insert_page - insert single page into user vma
2114 : : * @vma: user vma to map to
2115 : : * @addr: target user address of this page
2116 : : * @page: source kernel page
2117 : : *
2118 : : * This allows drivers to insert individual pages they've allocated
2119 : : * into a user vma.
2120 : : *
2121 : : * The page has to be a nice clean _individual_ kernel allocation.
2122 : : * If you allocate a compound page, you need to have marked it as
2123 : : * such (__GFP_COMP), or manually just split the page up yourself
2124 : : * (see split_page()).
2125 : : *
2126 : : * NOTE! Traditionally this was done with "remap_pfn_range()" which
2127 : : * took an arbitrary page protection parameter. This doesn't allow
2128 : : * that. Your vma protection will have to be set up correctly, which
2129 : : * means that if you want a shared writable mapping, you'd better
2130 : : * ask for a shared writable mapping!
2131 : : *
2132 : : * The page does not need to be reserved.
2133 : : *
2134 : : * Usually this function is called from f_op->mmap() handler
2135 : : * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
2136 : : * Caller must set VM_MIXEDMAP on vma if it wants to call this
2137 : : * function from other places, for example from page-fault handler.
2138 : : */
2139 : 0 : int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
2140 : : struct page *page)
2141 : : {
2142 [ # # ][ # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
2143 : : return -EFAULT;
2144 [ # # ]: 0 : if (!page_count(page))
2145 : : return -EINVAL;
2146 [ # # ]: 0 : if (!(vma->vm_flags & VM_MIXEDMAP)) {
2147 [ # # ]: 0 : BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
2148 [ # # ]: 0 : BUG_ON(vma->vm_flags & VM_PFNMAP);
2149 : 0 : vma->vm_flags |= VM_MIXEDMAP;
2150 : : }
2151 : 0 : return insert_page(vma, addr, page, vma->vm_page_prot);
2152 : : }
2153 : : EXPORT_SYMBOL(vm_insert_page);
2154 : :
2155 : 0 : static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2156 : : unsigned long pfn, pgprot_t prot)
2157 : : {
2158 : 0 : struct mm_struct *mm = vma->vm_mm;
2159 : : int retval;
2160 : : pte_t *pte, entry;
2161 : : spinlock_t *ptl;
2162 : :
2163 : : retval = -ENOMEM;
2164 : : pte = get_locked_pte(mm, addr, &ptl);
2165 [ # # ]: 0 : if (!pte)
2166 : : goto out;
2167 : : retval = -EBUSY;
2168 [ # # ]: 0 : if (!pte_none(*pte))
2169 : : goto out_unlock;
2170 : :
2171 : : /* Ok, finally just insert the thing.. */
2172 : 0 : entry = pte_mkspecial(pfn_pte(pfn, prot));
2173 : : set_pte_at(mm, addr, pte, entry);
2174 : : update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
2175 : :
2176 : : retval = 0;
2177 : : out_unlock:
2178 : 0 : pte_unmap_unlock(pte, ptl);
2179 : : out:
2180 : 0 : return retval;
2181 : : }
2182 : :
2183 : : /**
2184 : : * vm_insert_pfn - insert single pfn into user vma
2185 : : * @vma: user vma to map to
2186 : : * @addr: target user address of this page
2187 : : * @pfn: source kernel pfn
2188 : : *
2189 : : * Similar to vm_insert_page, this allows drivers to insert individual pages
2190 : : * they've allocated into a user vma. Same comments apply.
2191 : : *
2192 : : * This function should only be called from a vm_ops->fault handler, and
2193 : : * in that case the handler should return NULL.
2194 : : *
2195 : : * vma cannot be a COW mapping.
2196 : : *
2197 : : * As this is called only for pages that do not currently exist, we
2198 : : * do not need to flush old virtual caches or the TLB.
2199 : : */
2200 : 0 : int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2201 : : unsigned long pfn)
2202 : : {
2203 : : int ret;
2204 : 0 : pgprot_t pgprot = vma->vm_page_prot;
2205 : : /*
2206 : : * Technically, architectures with pte_special can avoid all these
2207 : : * restrictions (same for remap_pfn_range). However we would like
2208 : : * consistency in testing and feature parity among all, so we should
2209 : : * try to keep these invariants in place for everybody.
2210 : : */
2211 [ # # ]: 0 : BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2212 [ # # ]: 0 : BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
2213 : : (VM_PFNMAP|VM_MIXEDMAP));
2214 [ # # ][ # # ]: 0 : BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
2215 [ # # ][ # # ]: 0 : BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2216 : :
2217 [ # # ][ # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
2218 : : return -EFAULT;
2219 : : if (track_pfn_insert(vma, &pgprot, pfn))
2220 : : return -EINVAL;
2221 : :
2222 : 0 : ret = insert_pfn(vma, addr, pfn, pgprot);
2223 : :
2224 : 0 : return ret;
2225 : : }
2226 : : EXPORT_SYMBOL(vm_insert_pfn);
2227 : :
2228 : 0 : int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
2229 : : unsigned long pfn)
2230 : : {
2231 [ # # ]: 0 : BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
2232 : :
2233 [ # # ][ # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
2234 : : return -EFAULT;
2235 : :
2236 : : /*
2237 : : * If we don't have pte special, then we have to use the pfn_valid()
2238 : : * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
2239 : : * refcount the page if pfn_valid is true (hence insert_page rather
2240 : : * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
2241 : : * without pte special, it would there be refcounted as a normal page.
2242 : : */
2243 [ # # ]: 0 : if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
2244 : : struct page *page;
2245 : :
2246 : 0 : page = pfn_to_page(pfn);
2247 : 0 : return insert_page(vma, addr, page, vma->vm_page_prot);
2248 : : }
2249 : 0 : return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
2250 : : }
2251 : : EXPORT_SYMBOL(vm_insert_mixed);
2252 : :
2253 : : /*
2254 : : * maps a range of physical memory into the requested pages. the old
2255 : : * mappings are removed. any references to nonexistent pages results
2256 : : * in null mappings (currently treated as "copy-on-access")
2257 : : */
2258 : 0 : static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
2259 : : unsigned long addr, unsigned long end,
2260 : : unsigned long pfn, pgprot_t prot)
2261 : : {
2262 : : pte_t *pte;
2263 : : spinlock_t *ptl;
2264 : :
2265 [ # # ][ # # ]: 0 : pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
2266 [ # # ]: 0 : if (!pte)
2267 : : return -ENOMEM;
2268 : : arch_enter_lazy_mmu_mode();
2269 : : do {
2270 [ # # ]: 0 : BUG_ON(!pte_none(*pte));
2271 : 0 : set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
2272 : 0 : pfn++;
2273 [ # # ]: 0 : } while (pte++, addr += PAGE_SIZE, addr != end);
2274 : : arch_leave_lazy_mmu_mode();
2275 : 0 : pte_unmap_unlock(pte - 1, ptl);
2276 : 0 : return 0;
2277 : : }
2278 : :
2279 : : static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
2280 : : unsigned long addr, unsigned long end,
2281 : : unsigned long pfn, pgprot_t prot)
2282 : : {
2283 : : pmd_t *pmd;
2284 : : unsigned long next;
2285 : :
2286 : : pfn -= addr >> PAGE_SHIFT;
2287 : : pmd = pmd_alloc(mm, pud, addr);
2288 [ # # ]: 0 : if (!pmd)
2289 : : return -ENOMEM;
2290 : : VM_BUG_ON(pmd_trans_huge(*pmd));
2291 : : do {
2292 : : next = pmd_addr_end(addr, end);
2293 [ # # ]: 0 : if (remap_pte_range(mm, pmd, addr, next,
2294 : : pfn + (addr >> PAGE_SHIFT), prot))
2295 : : return -ENOMEM;
2296 : : } while (pmd++, addr = next, addr != end);
2297 : : return 0;
2298 : : }
2299 : :
2300 : : static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
2301 : : unsigned long addr, unsigned long end,
2302 : : unsigned long pfn, pgprot_t prot)
2303 : : {
2304 : : pud_t *pud;
2305 : : unsigned long next;
2306 : :
2307 : 0 : pfn -= addr >> PAGE_SHIFT;
2308 : : pud = pud_alloc(mm, pgd, addr);
2309 [ # # ]: 0 : if (!pud)
2310 : : return -ENOMEM;
2311 : : do {
2312 : : next = pud_addr_end(addr, end);
2313 [ # # ]: 0 : if (remap_pmd_range(mm, pud, addr, next,
2314 : : pfn + (addr >> PAGE_SHIFT), prot))
2315 : : return -ENOMEM;
2316 : : } while (pud++, addr = next, addr != end);
2317 : : return 0;
2318 : : }
2319 : :
2320 : : /**
2321 : : * remap_pfn_range - remap kernel memory to userspace
2322 : : * @vma: user vma to map to
2323 : : * @addr: target user address to start at
2324 : : * @pfn: physical address of kernel memory
2325 : : * @size: size of map area
2326 : : * @prot: page protection flags for this mapping
2327 : : *
2328 : : * Note: this is only safe if the mm semaphore is held when called.
2329 : : */
2330 : 0 : int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
2331 : : unsigned long pfn, unsigned long size, pgprot_t prot)
2332 : : {
2333 : : pgd_t *pgd;
2334 : : unsigned long next;
2335 : 0 : unsigned long end = addr + PAGE_ALIGN(size);
2336 : 0 : struct mm_struct *mm = vma->vm_mm;
2337 : : int err;
2338 : :
2339 : : /*
2340 : : * Physically remapped pages are special. Tell the
2341 : : * rest of the world about it:
2342 : : * VM_IO tells people not to look at these pages
2343 : : * (accesses can have side effects).
2344 : : * VM_PFNMAP tells the core MM that the base pages are just
2345 : : * raw PFN mappings, and do not have a "struct page" associated
2346 : : * with them.
2347 : : * VM_DONTEXPAND
2348 : : * Disable vma merging and expanding with mremap().
2349 : : * VM_DONTDUMP
2350 : : * Omit vma from core dump, even when VM_IO turned off.
2351 : : *
2352 : : * There's a horrible special case to handle copy-on-write
2353 : : * behaviour that some programs depend on. We mark the "original"
2354 : : * un-COW'ed pages by matching them up with "vma->vm_pgoff".
2355 : : * See vm_normal_page() for details.
2356 : : */
2357 [ # # ]: 0 : if (is_cow_mapping(vma->vm_flags)) {
2358 [ # # ][ # # ]: 0 : if (addr != vma->vm_start || end != vma->vm_end)
2359 : : return -EINVAL;
2360 : 0 : vma->vm_pgoff = pfn;
2361 : : }
2362 : :
2363 : : err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
2364 : : if (err)
2365 : : return -EINVAL;
2366 : :
2367 : 0 : vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
2368 : :
2369 [ # # ]: 0 : BUG_ON(addr >= end);
2370 : 0 : pfn -= addr >> PAGE_SHIFT;
2371 : 0 : pgd = pgd_offset(mm, addr);
2372 : 0 : flush_cache_range(vma, addr, end);
2373 : : do {
2374 [ # # ]: 0 : next = pgd_addr_end(addr, end);
2375 : 0 : err = remap_pud_range(mm, pgd, addr, next,
2376 : 0 : pfn + (addr >> PAGE_SHIFT), prot);
2377 [ # # ]: 0 : if (err)
2378 : : break;
2379 [ # # ]: 0 : } while (pgd++, addr = next, addr != end);
2380 : :
2381 : : if (err)
2382 : : untrack_pfn(vma, pfn, PAGE_ALIGN(size));
2383 : :
2384 : : return err;
2385 : : }
2386 : : EXPORT_SYMBOL(remap_pfn_range);
2387 : :
2388 : : /**
2389 : : * vm_iomap_memory - remap memory to userspace
2390 : : * @vma: user vma to map to
2391 : : * @start: start of area
2392 : : * @len: size of area
2393 : : *
2394 : : * This is a simplified io_remap_pfn_range() for common driver use. The
2395 : : * driver just needs to give us the physical memory range to be mapped,
2396 : : * we'll figure out the rest from the vma information.
2397 : : *
2398 : : * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
2399 : : * whatever write-combining details or similar.
2400 : : */
2401 : 0 : int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
2402 : : {
2403 : : unsigned long vm_len, pfn, pages;
2404 : :
2405 : : /* Check that the physical memory area passed in looks valid */
2406 [ # # ]: 0 : if (start + len < start)
2407 : : return -EINVAL;
2408 : : /*
2409 : : * You *really* shouldn't map things that aren't page-aligned,
2410 : : * but we've historically allowed it because IO memory might
2411 : : * just have smaller alignment.
2412 : : */
2413 : 0 : len += start & ~PAGE_MASK;
2414 : 0 : pfn = start >> PAGE_SHIFT;
2415 : 0 : pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
2416 [ # # ]: 0 : if (pfn + pages < pfn)
2417 : : return -EINVAL;
2418 : :
2419 : : /* We start the mapping 'vm_pgoff' pages into the area */
2420 [ # # ]: 0 : if (vma->vm_pgoff > pages)
2421 : : return -EINVAL;
2422 : 0 : pfn += vma->vm_pgoff;
2423 : 0 : pages -= vma->vm_pgoff;
2424 : :
2425 : : /* Can we fit all of the mapping? */
2426 : 0 : vm_len = vma->vm_end - vma->vm_start;
2427 [ # # ]: 0 : if (vm_len >> PAGE_SHIFT > pages)
2428 : : return -EINVAL;
2429 : :
2430 : : /* Ok, let it rip */
2431 : 0 : return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
2432 : : }
2433 : : EXPORT_SYMBOL(vm_iomap_memory);
2434 : :
2435 : 0 : static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2436 : : unsigned long addr, unsigned long end,
2437 : : pte_fn_t fn, void *data)
2438 : : {
2439 : : pte_t *pte;
2440 : : int err;
2441 : : pgtable_t token;
2442 : : spinlock_t *uninitialized_var(ptl);
2443 : :
2444 : : pte = (mm == &init_mm) ?
2445 [ # # ][ # # ]: 0 : pte_alloc_kernel(pmd, addr) :
[ # # ]
2446 [ # # ][ # # ]: 0 : pte_alloc_map_lock(mm, pmd, addr, &ptl);
2447 [ # # ]: 0 : if (!pte)
2448 : : return -ENOMEM;
2449 : :
2450 : : BUG_ON(pmd_huge(*pmd));
2451 : :
2452 : : arch_enter_lazy_mmu_mode();
2453 : :
2454 : 0 : token = pmd_pgtable(*pmd);
2455 : :
2456 : : do {
2457 : 0 : err = fn(pte++, token, addr, data);
2458 [ # # ]: 0 : if (err)
2459 : : break;
2460 [ # # ]: 0 : } while (addr += PAGE_SIZE, addr != end);
2461 : :
2462 : : arch_leave_lazy_mmu_mode();
2463 : :
2464 [ # # ]: 0 : if (mm != &init_mm)
2465 : 0 : pte_unmap_unlock(pte-1, ptl);
2466 : 0 : return err;
2467 : : }
2468 : :
2469 : : static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2470 : : unsigned long addr, unsigned long end,
2471 : : pte_fn_t fn, void *data)
2472 : : {
2473 : : pmd_t *pmd;
2474 : : unsigned long next;
2475 : : int err;
2476 : :
2477 : : BUG_ON(pud_huge(*pud));
2478 : :
2479 : : pmd = pmd_alloc(mm, pud, addr);
2480 [ # # ]: 0 : if (!pmd)
2481 : : return -ENOMEM;
2482 : : do {
2483 : : next = pmd_addr_end(addr, end);
2484 : 0 : err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2485 : : if (err)
2486 : : break;
2487 : : } while (pmd++, addr = next, addr != end);
2488 : : return err;
2489 : : }
2490 : :
2491 : : static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
2492 : : unsigned long addr, unsigned long end,
2493 : : pte_fn_t fn, void *data)
2494 : : {
2495 : : pud_t *pud;
2496 : : unsigned long next;
2497 : : int err;
2498 : :
2499 : : pud = pud_alloc(mm, pgd, addr);
2500 [ # # ]: 0 : if (!pud)
2501 : : return -ENOMEM;
2502 : : do {
2503 : : next = pud_addr_end(addr, end);
2504 : : err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2505 : : if (err)
2506 : : break;
2507 : : } while (pud++, addr = next, addr != end);
2508 : : return err;
2509 : : }
2510 : :
2511 : : /*
2512 : : * Scan a region of virtual memory, filling in page tables as necessary
2513 : : * and calling a provided function on each leaf page table.
2514 : : */
2515 : 0 : int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2516 : : unsigned long size, pte_fn_t fn, void *data)
2517 : : {
2518 : : pgd_t *pgd;
2519 : : unsigned long next;
2520 : 0 : unsigned long end = addr + size;
2521 : : int err;
2522 : :
2523 [ # # ]: 0 : BUG_ON(addr >= end);
2524 : 0 : pgd = pgd_offset(mm, addr);
2525 : : do {
2526 [ # # ]: 0 : next = pgd_addr_end(addr, end);
2527 : : err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
2528 [ # # ]: 0 : if (err)
2529 : : break;
2530 [ # # ]: 0 : } while (pgd++, addr = next, addr != end);
2531 : :
2532 : 0 : return err;
2533 : : }
2534 : : EXPORT_SYMBOL_GPL(apply_to_page_range);
2535 : :
2536 : : /*
2537 : : * handle_pte_fault chooses page fault handler according to an entry
2538 : : * which was read non-atomically. Before making any commitment, on
2539 : : * those architectures or configurations (e.g. i386 with PAE) which
2540 : : * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
2541 : : * must check under lock before unmapping the pte and proceeding
2542 : : * (but do_wp_page is only called after already making such a check;
2543 : : * and do_anonymous_page can safely check later on).
2544 : : */
2545 : : static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
2546 : : pte_t *page_table, pte_t orig_pte)
2547 : : {
2548 : : int same = 1;
2549 : : #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2550 : : if (sizeof(pte_t) > sizeof(unsigned long)) {
2551 : : spinlock_t *ptl = pte_lockptr(mm, pmd);
2552 : : spin_lock(ptl);
2553 : : same = pte_same(*page_table, orig_pte);
2554 : : spin_unlock(ptl);
2555 : : }
2556 : : #endif
2557 : 32 : pte_unmap(page_table);
2558 : : return same;
2559 : : }
2560 : :
2561 : : static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
2562 : : {
2563 : : debug_dma_assert_idle(src);
2564 : :
2565 : : /*
2566 : : * If the source page was a PFN mapping, we don't have
2567 : : * a "struct page" for it. We do a best-effort copy by
2568 : : * just copying from the original user address. If that
2569 : : * fails, we just zero-fill it. Live with it.
2570 : : */
2571 [ - + ]: 9258152 : if (unlikely(!src)) {
2572 : 0 : void *kaddr = kmap_atomic(dst);
2573 : 0 : void __user *uaddr = (void __user *)(va & PAGE_MASK);
2574 : :
2575 : : /*
2576 : : * This really shouldn't fail, because the page is there
2577 : : * in the page tables. But it might just be unreadable,
2578 : : * in which case we just give up and fill the result with
2579 : : * zeroes.
2580 : : */
2581 [ # # ]: 0 : if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
2582 : 0 : clear_page(kaddr);
2583 : 0 : kunmap_atomic(kaddr);
2584 : 0 : flush_dcache_page(dst);
2585 : : } else
2586 : 9258152 : copy_user_highpage(dst, src, va, vma);
2587 : : }
2588 : :
2589 : : /*
2590 : : * This routine handles present pages, when users try to write
2591 : : * to a shared page. It is done by copying the page to a new address
2592 : : * and decrementing the shared-page counter for the old page.
2593 : : *
2594 : : * Note that this routine assumes that the protection checks have been
2595 : : * done by the caller (the low-level page fault routine in most cases).
2596 : : * Thus we can safely just mark it writable once we've done any necessary
2597 : : * COW.
2598 : : *
2599 : : * We also mark the page dirty at this point even though the page will
2600 : : * change only once the write actually happens. This avoids a few races,
2601 : : * and potentially makes it more efficient.
2602 : : *
2603 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
2604 : : * but allow concurrent faults), with pte both mapped and locked.
2605 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
2606 : : */
2607 : 0 : static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2608 : 9390479 : unsigned long address, pte_t *page_table, pmd_t *pmd,
2609 : : spinlock_t *ptl, pte_t orig_pte)
2610 : : __releases(ptl)
2611 : : {
2612 : 25538283 : struct page *old_page, *new_page = NULL;
2613 : : pte_t entry;
2614 : : int ret = 0;
2615 : : int page_mkwrite = 0;
2616 : : struct page *dirty_page = NULL;
2617 : : unsigned long mmun_start = 0; /* For mmu_notifiers */
2618 : : unsigned long mmun_end = 0; /* For mmu_notifiers */
2619 : :
2620 : 16337471 : old_page = vm_normal_page(vma, address, orig_pte);
2621 [ + + ]: 16337343 : if (!old_page) {
2622 : : /*
2623 : : * VM_MIXEDMAP !pfn_valid() case
2624 : : *
2625 : : * We should not cow pages in a shared writeable mapping.
2626 : : * Just mark the pages writable as we can't do any dirty
2627 : : * accounting on raw pfn maps.
2628 : : */
2629 [ - + ]: 57315 : if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2630 : : (VM_WRITE|VM_SHARED))
2631 : : goto reuse;
2632 : : goto gotten;
2633 : : }
2634 : :
2635 : : /*
2636 : : * Take out anonymous pages first, anonymous shared vmas are
2637 : : * not dirty accountable.
2638 : : */
2639 [ + + ]: 16280028 : if (PageAnon(old_page) && !PageKsm(old_page)) {
2640 [ + + ]: 16018189 : if (!trylock_page(old_page)) {
2641 : : page_cache_get(old_page);
2642 : 521 : pte_unmap_unlock(page_table, ptl);
2643 : : lock_page(old_page);
2644 : 521 : page_table = pte_offset_map_lock(mm, pmd, address,
2645 : : &ptl);
2646 [ - + ]: 521 : if (!pte_same(*page_table, orig_pte)) {
2647 : 0 : unlock_page(old_page);
2648 : 0 : goto unlock;
2649 : : }
2650 : 521 : page_cache_release(old_page);
2651 : : }
2652 [ + + ]: 16018189 : if (reuse_swap_page(old_page)) {
2653 : : /*
2654 : : * The page is all ours. Move it to our anon_vma so
2655 : : * the rmap code will not search our parent or siblings.
2656 : : * Protected against the rmap code by the page lock.
2657 : : */
2658 : 6947289 : page_move_anon_rmap(old_page, vma, address);
2659 : 6947289 : unlock_page(old_page);
2660 : 6947245 : goto reuse;
2661 : : }
2662 : 9071285 : unlock_page(old_page);
2663 [ + + ]: 261767 : } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2664 : : (VM_WRITE|VM_SHARED))) {
2665 : : /*
2666 : : * Only catch write-faults on shared writable pages,
2667 : : * read-only shared pages can get COWed by
2668 : : * get_user_pages(.write=1, .force=1).
2669 : : */
2670 [ + - ][ + - ]: 74717 : if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2671 : : struct vm_fault vmf;
2672 : : int tmp;
2673 : :
2674 : 74717 : vmf.virtual_address = (void __user *)(address &
2675 : : PAGE_MASK);
2676 : 74717 : vmf.pgoff = old_page->index;
2677 : 74717 : vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2678 : 74717 : vmf.page = old_page;
2679 : :
2680 : : /*
2681 : : * Notify the address space that the page is about to
2682 : : * become writable so that it can prohibit this or wait
2683 : : * for the page to get into an appropriate state.
2684 : : *
2685 : : * We do this without the lock held, so that it can
2686 : : * sleep if it needs to.
2687 : : */
2688 : : page_cache_get(old_page);
2689 : 74717 : pte_unmap_unlock(page_table, ptl);
2690 : :
2691 : 74717 : tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2692 [ + - ]: 74716 : if (unlikely(tmp &
2693 : : (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2694 : : ret = tmp;
2695 : 0 : goto unwritable_page;
2696 : : }
2697 [ - + ]: 74716 : if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2698 : : lock_page(old_page);
2699 [ # # ]: 0 : if (!old_page->mapping) {
2700 : : ret = 0; /* retry the fault */
2701 : 0 : unlock_page(old_page);
2702 : 0 : goto unwritable_page;
2703 : : }
2704 : : } else
2705 : : VM_BUG_ON_PAGE(!PageLocked(old_page), old_page);
2706 : :
2707 : : /*
2708 : : * Since we dropped the lock we need to revalidate
2709 : : * the PTE as someone else may have changed it. If
2710 : : * they did, we just return, as we can count on the
2711 : : * MMU to tell us if they didn't also make it writable.
2712 : : */
2713 : 74716 : page_table = pte_offset_map_lock(mm, pmd, address,
2714 : : &ptl);
2715 [ - + ]: 74717 : if (!pte_same(*page_table, orig_pte)) {
2716 : 0 : unlock_page(old_page);
2717 : 74717 : goto unlock;
2718 : : }
2719 : :
2720 : : page_mkwrite = 1;
2721 : : }
2722 : : dirty_page = old_page;
2723 : : get_page(dirty_page);
2724 : :
2725 : : reuse:
2726 : : /*
2727 : : * Clear the pages cpupid information as the existing
2728 : : * information potentially belongs to a now completely
2729 : : * unrelated process.
2730 : : */
2731 : : if (old_page)
2732 : : page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1);
2733 : :
2734 : 7021962 : flush_cache_page(vma, address, pte_pfn(orig_pte));
2735 : : entry = pte_mkyoung(orig_pte);
2736 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2737 : 7021957 : if (ptep_set_access_flags(vma, address, page_table, entry,1))
2738 : : update_mmu_cache(vma, address, page_table);
2739 : 7022028 : pte_unmap_unlock(page_table, ptl);
2740 : : ret |= VM_FAULT_WRITE;
2741 : :
2742 [ + + ]: 7022019 : if (!dirty_page)
2743 : : return ret;
2744 : :
2745 : : /*
2746 : : * Yes, Virginia, this is actually required to prevent a race
2747 : : * with clear_page_dirty_for_io() from clearing the page dirty
2748 : : * bit after it clear all dirty ptes, but before a racing
2749 : : * do_wp_page installs a dirty pte.
2750 : : *
2751 : : * __do_fault is protected similarly.
2752 : : */
2753 [ - + ]: 74717 : if (!page_mkwrite) {
2754 : : wait_on_page_locked(dirty_page);
2755 : 0 : set_page_dirty_balance(dirty_page, page_mkwrite);
2756 : : /* file_update_time outside page_lock */
2757 [ # # ]: 0 : if (vma->vm_file)
2758 : 0 : file_update_time(vma->vm_file);
2759 : : }
2760 : 74717 : put_page(dirty_page);
2761 [ + - ]: 74717 : if (page_mkwrite) {
2762 : 74717 : struct address_space *mapping = dirty_page->mapping;
2763 : :
2764 : 74717 : set_page_dirty(dirty_page);
2765 : 74717 : unlock_page(dirty_page);
2766 : 74717 : page_cache_release(dirty_page);
2767 [ + - ]: 74717 : if (mapping) {
2768 : : /*
2769 : : * Some device drivers do not set page.mapping
2770 : : * but still dirty their pages
2771 : : */
2772 : 74717 : balance_dirty_pages_ratelimited(mapping);
2773 : : }
2774 : : }
2775 : :
2776 : : return ret;
2777 : : }
2778 : :
2779 : : /*
2780 : : * Ok, we need to copy. Oh, well..
2781 : : */
2782 : : page_cache_get(old_page);
2783 : : gotten:
2784 : 9315721 : pte_unmap_unlock(page_table, ptl);
2785 : :
2786 [ + ]: 9315658 : if (unlikely(anon_vma_prepare(vma)))
2787 : : goto oom;
2788 : :
2789 [ + + ]: 9315677 : if (is_zero_pfn(pte_pfn(orig_pte))) {
2790 : : new_page = alloc_zeroed_user_highpage_movable(vma, address);
2791 [ + + ]: 57345 : if (!new_page)
2792 : : goto oom;
2793 : : } else {
2794 : : new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2795 [ + ]: 9258139 : if (!new_page)
2796 : : goto oom;
2797 : : cow_user_page(new_page, old_page, address, vma);
2798 : : }
2799 : : __SetPageUptodate(new_page);
2800 : :
2801 : : if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
2802 : : goto oom_free_new;
2803 : :
2804 : : mmun_start = address & PAGE_MASK;
2805 : : mmun_end = mmun_start + PAGE_SIZE;
2806 : : mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2807 : :
2808 : : /*
2809 : : * Re-check the pte - we dropped the lock
2810 : : */
2811 : 9315242 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2812 [ + + ]: 9315663 : if (likely(pte_same(*page_table, orig_pte))) {
2813 [ + + ]: 9315569 : if (old_page) {
2814 [ + + ]: 9258255 : if (!PageAnon(old_page)) {
2815 : 187052 : dec_mm_counter_fast(mm, MM_FILEPAGES);
2816 : 187052 : inc_mm_counter_fast(mm, MM_ANONPAGES);
2817 : : }
2818 : : } else
2819 : 57314 : inc_mm_counter_fast(mm, MM_ANONPAGES);
2820 : 9315568 : flush_cache_page(vma, address, pte_pfn(orig_pte));
2821 : 9315559 : entry = mk_pte(new_page, vma->vm_page_prot);
2822 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2823 : : /*
2824 : : * Clear the pte entry and flush it first, before updating the
2825 : : * pte with the new entry. This will avoid a race condition
2826 : : * seen in the presence of one thread doing SMC and another
2827 : : * thread doing COW.
2828 : : */
2829 : 9315559 : ptep_clear_flush(vma, address, page_table);
2830 : 9315738 : page_add_new_anon_rmap(new_page, vma, address);
2831 : : /*
2832 : : * We call the notify macro here because, when using secondary
2833 : : * mmu page tables (such as kvm shadow page tables), we want the
2834 : : * new page to be mapped directly into the secondary page table.
2835 : : */
2836 : : set_pte_at_notify(mm, address, page_table, entry);
2837 : : update_mmu_cache(vma, address, page_table);
2838 [ + + ]: 9315523 : if (old_page) {
2839 : : /*
2840 : : * Only after switching the pte to the new page may
2841 : : * we remove the mapcount here. Otherwise another
2842 : : * process may come and find the rmap count decremented
2843 : : * before the pte is switched to the new page, and
2844 : : * "reuse" the old page writing into it while our pte
2845 : : * here still points into it and can be read by other
2846 : : * threads.
2847 : : *
2848 : : * The critical issue is to order this
2849 : : * page_remove_rmap with the ptp_clear_flush above.
2850 : : * Those stores are ordered by (if nothing else,)
2851 : : * the barrier present in the atomic_add_negative
2852 : : * in page_remove_rmap.
2853 : : *
2854 : : * Then the TLB flush in ptep_clear_flush ensures that
2855 : : * no process can access the old page before the
2856 : : * decremented mapcount is visible. And the old page
2857 : : * cannot be reused until after the decremented
2858 : : * mapcount is visible. So transitively, TLBs to
2859 : : * old page will be flushed before it can be reused.
2860 : : */
2861 : 9315523 : page_remove_rmap(old_page);
2862 : : }
2863 : :
2864 : : /* Free the old page.. */
2865 : : new_page = old_page;
2866 : : ret |= VM_FAULT_WRITE;
2867 : : } else
2868 : : mem_cgroup_uncharge_page(new_page);
2869 : :
2870 [ + + ]: 9315630 : if (new_page)
2871 : 9258044 : page_cache_release(new_page);
2872 : : unlock:
2873 : 9315709 : pte_unmap_unlock(page_table, ptl);
2874 : : if (mmun_end > mmun_start)
2875 : : mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2876 [ + + ]: 9315601 : if (old_page) {
2877 : : /*
2878 : : * Don't let another task, with possibly unlocked vma,
2879 : : * keep the mlocked page.
2880 : : */
2881 [ + + ][ + + ]: 9258326 : if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
2882 : : lock_page(old_page); /* LRU manipulation */
2883 : 14 : munlock_vma_page(old_page);
2884 : 14 : unlock_page(old_page);
2885 : : }
2886 : 9258326 : page_cache_release(old_page);
2887 : : }
2888 : 9315665 : return ret;
2889 : : oom_free_new:
2890 : : page_cache_release(new_page);
2891 : : oom:
2892 [ # # ]: 0 : if (old_page)
2893 : 0 : page_cache_release(old_page);
2894 : : return VM_FAULT_OOM;
2895 : :
2896 : : unwritable_page:
2897 : 0 : page_cache_release(old_page);
2898 : 0 : return ret;
2899 : : }
2900 : :
2901 : 0 : static void unmap_mapping_range_vma(struct vm_area_struct *vma,
2902 : : unsigned long start_addr, unsigned long end_addr,
2903 : : struct zap_details *details)
2904 : : {
2905 : 5 : zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
2906 : 5 : }
2907 : :
2908 : : static inline void unmap_mapping_range_tree(struct rb_root *root,
2909 : : struct zap_details *details)
2910 : : {
2911 : 5 : struct vm_area_struct *vma;
2912 : : pgoff_t vba, vea, zba, zea;
2913 : :
2914 [ + + ]: 12 : vma_interval_tree_foreach(vma, root,
2915 : : details->first_index, details->last_index) {
2916 : :
2917 : 5 : vba = vma->vm_pgoff;
2918 : 5 : vea = vba + vma_pages(vma) - 1;
2919 : : /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2920 : 5 : zba = details->first_index;
2921 [ - + ]: 5 : if (zba < vba)
2922 : : zba = vba;
2923 : 5 : zea = details->last_index;
2924 [ + + ]: 5 : if (zea > vea)
2925 : : zea = vea;
2926 : :
2927 : 5 : unmap_mapping_range_vma(vma,
2928 : 5 : ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2929 : 5 : ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2930 : : details);
2931 : : }
2932 : : }
2933 : :
2934 : : static inline void unmap_mapping_range_list(struct list_head *head,
2935 : : struct zap_details *details)
2936 : : {
2937 : : struct vm_area_struct *vma;
2938 : :
2939 : : /*
2940 : : * In nonlinear VMAs there is no correspondence between virtual address
2941 : : * offset and file offset. So we must perform an exhaustive search
2942 : : * across *all* the pages in each nonlinear VMA, not just the pages
2943 : : * whose virtual address lies outside the file truncation point.
2944 : : */
2945 [ # # ]: 0 : list_for_each_entry(vma, head, shared.nonlinear) {
2946 : 0 : details->nonlinear_vma = vma;
2947 : 0 : unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);
2948 : : }
2949 : : }
2950 : :
2951 : : /**
2952 : : * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2953 : : * @mapping: the address space containing mmaps to be unmapped.
2954 : : * @holebegin: byte in first page to unmap, relative to the start of
2955 : : * the underlying file. This will be rounded down to a PAGE_SIZE
2956 : : * boundary. Note that this is different from truncate_pagecache(), which
2957 : : * must keep the partial page. In contrast, we must get rid of
2958 : : * partial pages.
2959 : : * @holelen: size of prospective hole in bytes. This will be rounded
2960 : : * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2961 : : * end of the file.
2962 : : * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2963 : : * but 0 when invalidating pagecache, don't throw away private data.
2964 : : */
2965 : 0 : void unmap_mapping_range(struct address_space *mapping,
2966 : : loff_t const holebegin, loff_t const holelen, int even_cows)
2967 : : {
2968 : : struct zap_details details;
2969 : 74007 : pgoff_t hba = holebegin >> PAGE_SHIFT;
2970 : 74007 : pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2971 : :
2972 : : /* Check for overflow. */
2973 : : if (sizeof(holelen) > sizeof(hlen)) {
2974 : 74007 : long long holeend =
2975 : 74007 : (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2976 [ - + ]: 74007 : if (holeend & ~(long long)ULONG_MAX)
2977 : 0 : hlen = ULONG_MAX - hba + 1;
2978 : : }
2979 : :
2980 [ + + ]: 74007 : details.check_mapping = even_cows? NULL: mapping;
2981 : 74007 : details.nonlinear_vma = NULL;
2982 : 74007 : details.first_index = hba;
2983 : 74007 : details.last_index = hba + hlen - 1;
2984 [ + + ]: 74007 : if (details.last_index < details.first_index)
2985 : 67512 : details.last_index = ULONG_MAX;
2986 : :
2987 : :
2988 : 74007 : mutex_lock(&mapping->i_mmap_mutex);
2989 [ + + ]: 74007 : if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
2990 : 7 : unmap_mapping_range_tree(&mapping->i_mmap, &details);
2991 [ - + ]: 74007 : if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2992 : : unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2993 : 74007 : mutex_unlock(&mapping->i_mmap_mutex);
2994 : 74007 : }
2995 : : EXPORT_SYMBOL(unmap_mapping_range);
2996 : :
2997 : : /*
2998 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
2999 : : * but allow concurrent faults), and pte mapped but not yet locked.
3000 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
3001 : : */
3002 : 0 : static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
3003 : 0 : unsigned long address, pte_t *page_table, pmd_t *pmd,
3004 : : unsigned int flags, pte_t orig_pte)
3005 : : {
3006 : : spinlock_t *ptl;
3007 : : struct page *page, *swapcache;
3008 : : swp_entry_t entry;
3009 : : pte_t pte;
3010 : : int locked;
3011 : : struct mem_cgroup *ptr;
3012 : : int exclusive = 0;
3013 : : int ret = 0;
3014 : :
3015 : : if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
3016 : : goto out;
3017 : :
3018 : : entry = pte_to_swp_entry(orig_pte);
3019 [ # # ]: 0 : if (unlikely(non_swap_entry(entry))) {
3020 [ # # ]: 0 : if (is_migration_entry(entry)) {
3021 : 0 : migration_entry_wait(mm, pmd, address);
3022 : : } else if (is_hwpoison_entry(entry)) {
3023 : : ret = VM_FAULT_HWPOISON;
3024 : : } else {
3025 : 0 : print_bad_pte(vma, address, orig_pte, NULL);
3026 : : ret = VM_FAULT_SIGBUS;
3027 : : }
3028 : : goto out;
3029 : : }
3030 : : delayacct_set_flag(DELAYACCT_PF_SWAPIN);
3031 : 0 : page = lookup_swap_cache(entry);
3032 [ # # ]: 0 : if (!page) {
3033 : 0 : page = swapin_readahead(entry,
3034 : : GFP_HIGHUSER_MOVABLE, vma, address);
3035 [ # # ]: 0 : if (!page) {
3036 : : /*
3037 : : * Back out if somebody else faulted in this pte
3038 : : * while we released the pte lock.
3039 : : */
3040 : 0 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3041 [ # # ]: 0 : if (likely(pte_same(*page_table, orig_pte)))
3042 : : ret = VM_FAULT_OOM;
3043 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
3044 : : goto unlock;
3045 : : }
3046 : :
3047 : : /* Had to read the page from swap area: Major fault */
3048 : : ret = VM_FAULT_MAJOR;
3049 : : count_vm_event(PGMAJFAULT);
3050 : : mem_cgroup_count_vm_event(mm, PGMAJFAULT);
3051 : : } else if (PageHWPoison(page)) {
3052 : : /*
3053 : : * hwpoisoned dirty swapcache pages are kept for killing
3054 : : * owner processes (which may be unknown at hwpoison time)
3055 : : */
3056 : : ret = VM_FAULT_HWPOISON;
3057 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
3058 : : swapcache = page;
3059 : : goto out_release;
3060 : : }
3061 : :
3062 : : swapcache = page;
3063 : : locked = lock_page_or_retry(page, mm, flags);
3064 : :
3065 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
3066 [ # # ]: 0 : if (!locked) {
3067 : 0 : ret |= VM_FAULT_RETRY;
3068 : 0 : goto out_release;
3069 : : }
3070 : :
3071 : : /*
3072 : : * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
3073 : : * release the swapcache from under us. The page pin, and pte_same
3074 : : * test below, are not enough to exclude that. Even if it is still
3075 : : * swapcache, we need to check that the page's swap has not changed.
3076 : : */
3077 [ # # ][ # # ]: 0 : if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
3078 : : goto out_page;
3079 : :
3080 : : page = ksm_might_need_to_copy(page, vma, address);
3081 [ # # ]: 0 : if (unlikely(!page)) {
3082 : : ret = VM_FAULT_OOM;
3083 : : page = swapcache;
3084 : : goto out_page;
3085 : : }
3086 : :
3087 : : if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
3088 : : ret = VM_FAULT_OOM;
3089 : : goto out_page;
3090 : : }
3091 : :
3092 : : /*
3093 : : * Back out if somebody else already faulted in this pte.
3094 : : */
3095 : 0 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3096 [ # # ]: 0 : if (unlikely(!pte_same(*page_table, orig_pte)))
3097 : : goto out_nomap;
3098 : :
3099 [ # # ]: 0 : if (unlikely(!PageUptodate(page))) {
3100 : : ret = VM_FAULT_SIGBUS;
3101 : : goto out_nomap;
3102 : : }
3103 : :
3104 : : /*
3105 : : * The page isn't present yet, go ahead with the fault.
3106 : : *
3107 : : * Be careful about the sequence of operations here.
3108 : : * To get its accounting right, reuse_swap_page() must be called
3109 : : * while the page is counted on swap but not yet in mapcount i.e.
3110 : : * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
3111 : : * must be called after the swap_free(), or it will never succeed.
3112 : : * Because delete_from_swap_page() may be called by reuse_swap_page(),
3113 : : * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
3114 : : * in page->private. In this case, a record in swap_cgroup is silently
3115 : : * discarded at swap_free().
3116 : : */
3117 : :
3118 : 0 : inc_mm_counter_fast(mm, MM_ANONPAGES);
3119 : 0 : dec_mm_counter_fast(mm, MM_SWAPENTS);
3120 : 0 : pte = mk_pte(page, vma->vm_page_prot);
3121 [ # # ][ # # ]: 0 : if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
3122 : : pte = maybe_mkwrite(pte_mkdirty(pte), vma);
3123 : 0 : flags &= ~FAULT_FLAG_WRITE;
3124 : 0 : ret |= VM_FAULT_WRITE;
3125 : : exclusive = 1;
3126 : : }
3127 : : flush_icache_page(vma, page);
3128 : : if (pte_swp_soft_dirty(orig_pte))
3129 : : pte = pte_mksoft_dirty(pte);
3130 : : set_pte_at(mm, address, page_table, pte);
3131 : : if (page == swapcache)
3132 : 0 : do_page_add_anon_rmap(page, vma, address, exclusive);
3133 : : else /* ksm created a completely new copy */
3134 : : page_add_new_anon_rmap(page, vma, address);
3135 : : /* It's better to call commit-charge after rmap is established */
3136 : : mem_cgroup_commit_charge_swapin(page, ptr);
3137 : :
3138 : 0 : swap_free(entry);
3139 [ # # ][ # # ]: 0 : if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
[ # # ]
3140 : 0 : try_to_free_swap(page);
3141 : 0 : unlock_page(page);
3142 : : if (page != swapcache) {
3143 : : /*
3144 : : * Hold the lock to avoid the swap entry to be reused
3145 : : * until we take the PT lock for the pte_same() check
3146 : : * (to avoid false positives from pte_same). For
3147 : : * further safety release the lock after the swap_free
3148 : : * so that the swap count won't change under a
3149 : : * parallel locked swapcache.
3150 : : */
3151 : : unlock_page(swapcache);
3152 : : page_cache_release(swapcache);
3153 : : }
3154 : :
3155 [ # # ]: 0 : if (flags & FAULT_FLAG_WRITE) {
3156 : 0 : ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
3157 [ # # ]: 0 : if (ret & VM_FAULT_ERROR)
3158 : : ret &= VM_FAULT_ERROR;
3159 : : goto out;
3160 : : }
3161 : :
3162 : : /* No need to invalidate - it was non-present before */
3163 : : update_mmu_cache(vma, address, page_table);
3164 : : unlock:
3165 : 0 : pte_unmap_unlock(page_table, ptl);
3166 : : out:
3167 : 0 : return ret;
3168 : : out_nomap:
3169 : : mem_cgroup_cancel_charge_swapin(ptr);
3170 : 0 : pte_unmap_unlock(page_table, ptl);
3171 : : out_page:
3172 : 0 : unlock_page(page);
3173 : : out_release:
3174 : 0 : page_cache_release(page);
3175 : : if (page != swapcache) {
3176 : : unlock_page(swapcache);
3177 : : page_cache_release(swapcache);
3178 : : }
3179 : 0 : return ret;
3180 : : }
3181 : :
3182 : : /*
3183 : : * This is like a special single-page "expand_{down|up}wards()",
3184 : : * except we must first make sure that 'address{-|+}PAGE_SIZE'
3185 : : * doesn't hit another vma.
3186 : : */
3187 : : static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
3188 : : {
3189 : 8645098 : address &= PAGE_MASK;
3190 [ + + ][ + + ]: 8645098 : if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
3191 : 1986191 : struct vm_area_struct *prev = vma->vm_prev;
3192 : :
3193 : : /*
3194 : : * Is there a mapping abutting this one below?
3195 : : *
3196 : : * That's only ok if it's the same stack mapping
3197 : : * that has gotten split..
3198 : : */
3199 [ + + ][ + + ]: 1986191 : if (prev && prev->vm_end == address)
3200 [ - + ]: 1 : return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
3201 : :
3202 : 1986190 : expand_downwards(vma, address - PAGE_SIZE);
3203 : : }
3204 : : if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
3205 : : struct vm_area_struct *next = vma->vm_next;
3206 : :
3207 : : /* As VM_GROWSDOWN but s/below/above/ */
3208 : : if (next && next->vm_start == address + PAGE_SIZE)
3209 : : return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
3210 : :
3211 : : expand_upwards(vma, address + PAGE_SIZE);
3212 : : }
3213 : : return 0;
3214 : : }
3215 : :
3216 : : /*
3217 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
3218 : : * but allow concurrent faults), and pte mapped but not yet locked.
3219 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
3220 : : */
3221 : 0 : static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
3222 : 8644094 : unsigned long address, pte_t *page_table, pmd_t *pmd,
3223 : : unsigned int flags)
3224 : : {
3225 : : struct page *page;
3226 : : spinlock_t *ptl;
3227 : : pte_t entry;
3228 : :
3229 : 8644982 : pte_unmap(page_table);
3230 : :
3231 : : /* Check if we need to add a guard page to the stack */
3232 [ + - ]: 8645041 : if (check_stack_guard_page(vma, address) < 0)
3233 : : return VM_FAULT_SIGBUS;
3234 : :
3235 : : /* Use the zero-page for reads */
3236 [ + + ]: 8645041 : if (!(flags & FAULT_FLAG_WRITE)) {
3237 : 88466 : entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
3238 : : vma->vm_page_prot));
3239 : 88466 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3240 [ + ]: 88466 : if (!pte_none(*page_table))
3241 : : goto unlock;
3242 : : goto setpte;
3243 : : }
3244 : :
3245 : : /* Allocate our own private page. */
3246 [ + + ]: 8556575 : if (unlikely(anon_vma_prepare(vma)))
3247 : : goto oom;
3248 : : page = alloc_zeroed_user_highpage_movable(vma, address);
3249 [ + + ]: 8556401 : if (!page)
3250 : : goto oom;
3251 : : /*
3252 : : * The memory barrier inside __SetPageUptodate makes sure that
3253 : : * preceeding stores to the page contents become visible before
3254 : : * the set_pte_at() write.
3255 : : */
3256 : : __SetPageUptodate(page);
3257 : :
3258 : : if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
3259 : : goto oom_free_page;
3260 : :
3261 : 8555628 : entry = mk_pte(page, vma->vm_page_prot);
3262 [ + ]: 8555628 : if (vma->vm_flags & VM_WRITE)
3263 : : entry = pte_mkwrite(pte_mkdirty(entry));
3264 : :
3265 : 8555628 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3266 [ + - ]: 8556183 : if (!pte_none(*page_table))
3267 : : goto release;
3268 : :
3269 : 8556183 : inc_mm_counter_fast(mm, MM_ANONPAGES);
3270 : 8556493 : page_add_new_anon_rmap(page, vma, address);
3271 : : setpte:
3272 : : set_pte_at(mm, address, page_table, entry);
3273 : :
3274 : : /* No need to invalidate - it was non-present before */
3275 : : update_mmu_cache(vma, address, page_table);
3276 : : unlock:
3277 : 8645010 : pte_unmap_unlock(page_table, ptl);
3278 : 8644992 : return 0;
3279 : : release:
3280 : : mem_cgroup_uncharge_page(page);
3281 : 0 : page_cache_release(page);
3282 : 0 : goto unlock;
3283 : : oom_free_page:
3284 : : page_cache_release(page);
3285 : : oom:
3286 : : return VM_FAULT_OOM;
3287 : : }
3288 : :
3289 : : /*
3290 : : * __do_fault() tries to create a new page mapping. It aggressively
3291 : : * tries to share with existing pages, but makes a separate copy if
3292 : : * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
3293 : : * the next page fault.
3294 : : *
3295 : : * As this is called only for pages that do not currently exist, we
3296 : : * do not need to flush old virtual caches or the TLB.
3297 : : *
3298 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
3299 : : * but allow concurrent faults), and pte neither mapped nor locked.
3300 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
3301 : : */
3302 : 0 : static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3303 : 30521452 : unsigned long address, pmd_t *pmd,
3304 : : pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
3305 : : {
3306 : : pte_t *page_table;
3307 : : spinlock_t *ptl;
3308 : : struct page *page;
3309 : : struct page *cow_page;
3310 : : pte_t entry;
3311 : : int anon = 0;
3312 : : struct page *dirty_page = NULL;
3313 : : struct vm_fault vmf;
3314 : : int ret;
3315 : : int page_mkwrite = 0;
3316 : :
3317 : : /*
3318 : : * If we do COW later, allocate page befor taking lock_page()
3319 : : * on the file cache page. This will reduce lock holding time.
3320 : : */
3321 [ + + ][ + + ]: 30537039 : if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {
3322 : :
3323 [ + - ]: 162447 : if (unlikely(anon_vma_prepare(vma)))
3324 : : return VM_FAULT_OOM;
3325 : :
3326 : : cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
3327 [ + ]: 162378 : if (!cow_page)
3328 : : return VM_FAULT_OOM;
3329 : :
3330 : : if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) {
3331 : : page_cache_release(cow_page);
3332 : : return VM_FAULT_OOM;
3333 : : }
3334 : : } else
3335 : : cow_page = NULL;
3336 : :
3337 : 30536972 : vmf.virtual_address = (void __user *)(address & PAGE_MASK);
3338 : 30536972 : vmf.pgoff = pgoff;
3339 : 30536972 : vmf.flags = flags;
3340 : 30536972 : vmf.page = NULL;
3341 : :
3342 : 30536972 : ret = vma->vm_ops->fault(vma, &vmf);
3343 [ + + ]: 30534513 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
3344 : : VM_FAULT_RETRY)))
3345 : : goto uncharge_out;
3346 : :
3347 : : if (unlikely(PageHWPoison(vmf.page))) {
3348 : : if (ret & VM_FAULT_LOCKED)
3349 : : unlock_page(vmf.page);
3350 : : ret = VM_FAULT_HWPOISON;
3351 : : page_cache_release(vmf.page);
3352 : : goto uncharge_out;
3353 : : }
3354 : :
3355 : : /*
3356 : : * For consistency in subsequent calls, make the faulted page always
3357 : : * locked.
3358 : : */
3359 [ + + ]: 30520912 : if (unlikely(!(ret & VM_FAULT_LOCKED)))
3360 : 24 : lock_page(vmf.page);
3361 : : else
3362 : : VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
3363 : :
3364 : : /*
3365 : : * Should we do an early C-O-W break?
3366 : : */
3367 : 30521312 : page = vmf.page;
3368 [ + + ]: 30521312 : if (flags & FAULT_FLAG_WRITE) {
3369 [ + + ]: 301930 : if (!(vma->vm_flags & VM_SHARED)) {
3370 : : page = cow_page;
3371 : : anon = 1;
3372 : 160881 : copy_user_highpage(page, vmf.page, address, vma);
3373 : : __SetPageUptodate(page);
3374 : : } else {
3375 : : /*
3376 : : * If the page will be shareable, see if the backing
3377 : : * address space wants to know that the page is about
3378 : : * to become writable
3379 : : */
3380 [ + + ]: 141049 : if (vma->vm_ops->page_mkwrite) {
3381 : : int tmp;
3382 : :
3383 : 130880 : unlock_page(page);
3384 : 130706 : vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
3385 : 130706 : tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
3386 [ + ]: 130873 : if (unlikely(tmp &
3387 : : (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
3388 : : ret = tmp;
3389 : : goto unwritable_page;
3390 : : }
3391 [ - + ]: 130878 : if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
3392 : : lock_page(page);
3393 [ # # ]: 0 : if (!page->mapping) {
3394 : : ret = 0; /* retry the fault */
3395 : 0 : unlock_page(page);
3396 : 0 : goto unwritable_page;
3397 : : }
3398 : : } else
3399 : : VM_BUG_ON_PAGE(!PageLocked(page), page);
3400 : : page_mkwrite = 1;
3401 : : }
3402 : : }
3403 : :
3404 : : }
3405 : :
3406 : 30521452 : page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3407 : :
3408 : : /*
3409 : : * This silly early PAGE_DIRTY setting removes a race
3410 : : * due to the bad i386 page protection. But it's valid
3411 : : * for other architectures too.
3412 : : *
3413 : : * Note that if FAULT_FLAG_WRITE is set, we either now have
3414 : : * an exclusive copy of the page, or this is a shared mapping,
3415 : : * so we can make it writable and dirty to avoid having to
3416 : : * handle that later.
3417 : : */
3418 : : /* Only go through if we didn't race with anybody else... */
3419 [ + + ]: 61059986 : if (likely(pte_same(*page_table, orig_pte))) {
3420 : : flush_icache_page(vma, page);
3421 : 30522944 : entry = mk_pte(page, vma->vm_page_prot);
3422 [ + + ]: 30522944 : if (flags & FAULT_FLAG_WRITE)
3423 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
3424 : : else if (pte_file(orig_pte) && pte_file_soft_dirty(orig_pte))
3425 : : pte_mksoft_dirty(entry);
3426 [ + + ]: 30522944 : if (anon) {
3427 : 161037 : inc_mm_counter_fast(mm, MM_ANONPAGES);
3428 : 161053 : page_add_new_anon_rmap(page, vma, address);
3429 : : } else {
3430 : 30361907 : inc_mm_counter_fast(mm, MM_FILEPAGES);
3431 : 30362849 : page_add_file_rmap(page);
3432 [ + + ]: 30361737 : if (flags & FAULT_FLAG_WRITE) {
3433 : : dirty_page = page;
3434 : : get_page(dirty_page);
3435 : : }
3436 : : }
3437 : : set_pte_at(mm, address, page_table, entry);
3438 : :
3439 : : /* no need to invalidate: a not-present page won't be cached */
3440 : : update_mmu_cache(vma, address, page_table);
3441 : : } else {
3442 : : if (cow_page)
3443 : : mem_cgroup_uncharge_page(cow_page);
3444 [ - + ]: 3 : if (anon)
3445 : 0 : page_cache_release(page);
3446 : : else
3447 : : anon = 1; /* no anon but release faulted_page */
3448 : : }
3449 : :
3450 : 30521807 : pte_unmap_unlock(page_table, ptl);
3451 : :
3452 [ + + ]: 30520671 : if (dirty_page) {
3453 : 141032 : struct address_space *mapping = page->mapping;
3454 : : int dirtied = 0;
3455 : :
3456 [ + + ]: 141032 : if (set_page_dirty(dirty_page))
3457 : : dirtied = 1;
3458 : 141062 : unlock_page(dirty_page);
3459 : 141059 : put_page(dirty_page);
3460 [ + + ][ + ]: 140933 : if ((dirtied || page_mkwrite) && mapping) {
3461 : : /*
3462 : : * Some device drivers do not set page.mapping but still
3463 : : * dirty their pages
3464 : : */
3465 : 140973 : balance_dirty_pages_ratelimited(mapping);
3466 : : }
3467 : :
3468 : : /* file_update_time outside page_lock */
3469 [ + + ][ + + ]: 142725 : if (vma->vm_file && !page_mkwrite)
3470 : 10179 : file_update_time(vma->vm_file);
3471 : : } else {
3472 : 30379639 : unlock_page(vmf.page);
3473 [ + + ]: 30373598 : if (anon)
3474 : 160888 : page_cache_release(vmf.page);
3475 : : }
3476 : :
3477 : 30516331 : return ret;
3478 : :
3479 : : unwritable_page:
3480 : 0 : page_cache_release(page);
3481 : 0 : return ret;
3482 : : uncharge_out:
3483 : : /* fs's fault handler get error */
3484 [ + + ]: 13601 : if (cow_page) {
3485 : : mem_cgroup_uncharge_page(cow_page);
3486 : 1360 : page_cache_release(cow_page);
3487 : : }
3488 : 13600 : return ret;
3489 : : }
3490 : :
3491 : 0 : static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3492 : : unsigned long address, pte_t *page_table, pmd_t *pmd,
3493 : : unsigned int flags, pte_t orig_pte)
3494 : : {
3495 : 61072324 : pgoff_t pgoff = (((address & PAGE_MASK)
3496 : 30536162 : - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3497 : :
3498 : 30536162 : pte_unmap(page_table);
3499 : 30536878 : return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
3500 : : }
3501 : :
3502 : : /*
3503 : : * Fault of a previously existing named mapping. Repopulate the pte
3504 : : * from the encoded file_pte if possible. This enables swappable
3505 : : * nonlinear vmas.
3506 : : *
3507 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
3508 : : * but allow concurrent faults), and pte mapped but not yet locked.
3509 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
3510 : : */
3511 : 0 : static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3512 : : unsigned long address, pte_t *page_table, pmd_t *pmd,
3513 : : unsigned int flags, pte_t orig_pte)
3514 : : {
3515 : : pgoff_t pgoff;
3516 : :
3517 : 32 : flags |= FAULT_FLAG_NONLINEAR;
3518 : :
3519 : : if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
3520 : : return 0;
3521 : :
3522 [ - + ]: 32 : if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
3523 : : /*
3524 : : * Page table corrupted: show pte and kill process.
3525 : : */
3526 : 0 : print_bad_pte(vma, address, orig_pte, NULL);
3527 : 0 : return VM_FAULT_SIGBUS;
3528 : : }
3529 : :
3530 : 32 : pgoff = pte_to_pgoff(orig_pte);
3531 : 32 : return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
3532 : : }
3533 : :
3534 : 0 : int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
3535 : : unsigned long addr, int page_nid,
3536 : : int *flags)
3537 : : {
3538 : : get_page(page);
3539 : :
3540 : : count_vm_numa_event(NUMA_HINT_FAULTS);
3541 [ # # ]: 0 : if (page_nid == numa_node_id()) {
3542 : : count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
3543 : 0 : *flags |= TNF_FAULT_LOCAL;
3544 : : }
3545 : :
3546 : 0 : return mpol_misplaced(page, vma, addr);
3547 : : }
3548 : :
3549 : 0 : int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
3550 : 0 : unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3551 : : {
3552 : : struct page *page = NULL;
3553 : : spinlock_t *ptl;
3554 : : int page_nid = -1;
3555 : : int last_cpupid;
3556 : : int target_nid;
3557 : : bool migrated = false;
3558 : 0 : int flags = 0;
3559 : :
3560 : : /*
3561 : : * The "pte" at this point cannot be used safely without
3562 : : * validation through pte_unmap_same(). It's of NUMA type but
3563 : : * the pfn may be screwed if the read is non atomic.
3564 : : *
3565 : : * ptep_modify_prot_start is not called as this is clearing
3566 : : * the _PAGE_NUMA bit and it is not really expected that there
3567 : : * would be concurrent hardware modifications to the PTE.
3568 : : */
3569 : : ptl = pte_lockptr(mm, pmd);
3570 : : spin_lock(ptl);
3571 [ # # ]: 0 : if (unlikely(!pte_same(*ptep, pte))) {
3572 : 0 : pte_unmap_unlock(ptep, ptl);
3573 : 0 : goto out;
3574 : : }
3575 : :
3576 : : pte = pte_mknonnuma(pte);
3577 : : set_pte_at(mm, addr, ptep, pte);
3578 : : update_mmu_cache(vma, addr, ptep);
3579 : :
3580 : 0 : page = vm_normal_page(vma, addr, pte);
3581 [ # # ]: 0 : if (!page) {
3582 : 0 : pte_unmap_unlock(ptep, ptl);
3583 : 0 : return 0;
3584 : : }
3585 [ # # ]: 0 : BUG_ON(is_zero_pfn(page_to_pfn(page)));
3586 : :
3587 : : /*
3588 : : * Avoid grouping on DSO/COW pages in specific and RO pages
3589 : : * in general, RO pages shouldn't hurt as much anyway since
3590 : : * they can be in shared cache state.
3591 : : */
3592 [ # # ]: 0 : if (!pte_write(pte))
3593 : 0 : flags |= TNF_NO_GROUP;
3594 : :
3595 : : /*
3596 : : * Flag if the page is shared between multiple address spaces. This
3597 : : * is later used when determining whether to group tasks together
3598 : : */
3599 [ # # ][ # # ]: 0 : if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3600 : 0 : flags |= TNF_SHARED;
3601 : :
3602 : : last_cpupid = page_cpupid_last(page);
3603 : : page_nid = page_to_nid(page);
3604 : 0 : target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
3605 : 0 : pte_unmap_unlock(ptep, ptl);
3606 [ # # ]: 0 : if (target_nid == -1) {
3607 : 0 : put_page(page);
3608 : 0 : goto out;
3609 : : }
3610 : :
3611 : : /* Migrate to the requested node */
3612 : : migrated = migrate_misplaced_page(page, vma, target_nid);
3613 : : if (migrated) {
3614 : : page_nid = target_nid;
3615 : : flags |= TNF_MIGRATED;
3616 : : }
3617 : :
3618 : : out:
3619 : : if (page_nid != -1)
3620 : : task_numa_fault(last_cpupid, page_nid, 1, flags);
3621 : : return 0;
3622 : : }
3623 : :
3624 : : /*
3625 : : * These routines also need to handle stuff like marking pages dirty
3626 : : * and/or accessed for architectures that don't do it in hardware (most
3627 : : * RISC architectures). The early dirtying is also good on the i386.
3628 : : *
3629 : : * There is also a hook called "update_mmu_cache()" that architectures
3630 : : * with external mmu caches can use to update those (ie the Sparc or
3631 : : * PowerPC hashed page tables that act as extended TLBs).
3632 : : *
3633 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
3634 : : * but allow concurrent faults), and pte mapped but not yet locked.
3635 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
3636 : : */
3637 : 0 : static int handle_pte_fault(struct mm_struct *mm,
3638 : : struct vm_area_struct *vma, unsigned long address,
3639 : 27750514 : pte_t *pte, pmd_t *pmd, unsigned int flags)
3640 : : {
3641 : : pte_t entry;
3642 : : spinlock_t *ptl;
3643 : :
3644 : 66931492 : entry = *pte;
3645 [ + + ]: 66931492 : if (!pte_present(entry)) {
3646 [ + + ]: 39180978 : if (pte_none(entry)) {
3647 [ + + ]: 39180946 : if (vma->vm_ops) {
3648 [ + + ]: 30536486 : if (likely(vma->vm_ops->fault))
3649 : 30536253 : return do_linear_fault(mm, vma, address,
3650 : : pte, pmd, flags, entry);
3651 : : }
3652 : 8644693 : return do_anonymous_page(mm, vma, address,
3653 : : pte, pmd, flags);
3654 : : }
3655 [ + - ]: 32 : if (pte_file(entry))
3656 : 32 : return do_nonlinear_fault(mm, vma, address,
3657 : : pte, pmd, flags, entry);
3658 : 0 : return do_swap_page(mm, vma, address,
3659 : : pte, pmd, flags, entry);
3660 : : }
3661 : :
3662 : : if (pte_numa(entry))
3663 : : return do_numa_page(mm, vma, address, entry, pte, pmd);
3664 : :
3665 : 27750514 : ptl = pte_lockptr(mm, pmd);
3666 : : spin_lock(ptl);
3667 [ + ]: 27750490 : if (unlikely(!pte_same(*pte, entry)))
3668 : : goto unlock;
3669 [ + + ]: 27750835 : if (flags & FAULT_FLAG_WRITE) {
3670 [ + + ]: 16492833 : if (!pte_write(entry))
3671 : 16337587 : return do_wp_page(mm, vma, address,
3672 : : pte, pmd, ptl, entry);
3673 : : entry = pte_mkdirty(entry);
3674 : : }
3675 : : entry = pte_mkyoung(entry);
3676 [ + + ]: 11413248 : if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
3677 : : update_mmu_cache(vma, address, pte);
3678 : : } else {
3679 : : /*
3680 : : * This is needed only for protection faults but the arch code
3681 : : * is not yet telling us if this is a protection fault or not.
3682 : : * This still avoids useless tlb flushes for .text page faults
3683 : : * with threads.
3684 : : */
3685 [ - + ]: 216 : if (flags & FAULT_FLAG_WRITE)
3686 : 0 : flush_tlb_fix_spurious_fault(vma, address);
3687 : : }
3688 : : unlock:
3689 : 11413970 : pte_unmap_unlock(pte, ptl);
3690 : 11413844 : return 0;
3691 : : }
3692 : :
3693 : : /*
3694 : : * By the time we get here, we already hold the mm semaphore
3695 : : */
3696 : 0 : static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3697 : : unsigned long address, unsigned int flags)
3698 : : {
3699 : : pgd_t *pgd;
3700 : : pud_t *pud;
3701 : : pmd_t *pmd;
3702 : : pte_t *pte;
3703 : :
3704 : : if (unlikely(is_vm_hugetlb_page(vma)))
3705 : : return hugetlb_fault(mm, vma, address, flags);
3706 : :
3707 : 66929549 : pgd = pgd_offset(mm, address);
3708 : : pud = pud_alloc(mm, pgd, address);
3709 [ + + ]: 66929549 : if (!pud)
3710 : : return VM_FAULT_OOM;
3711 : : pmd = pmd_alloc(mm, pud, address);
3712 [ + + ]: 66896142 : if (!pmd)
3713 : : return VM_FAULT_OOM;
3714 : : if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
3715 : : int ret = VM_FAULT_FALLBACK;
3716 : : if (!vma->vm_ops)
3717 : : ret = do_huge_pmd_anonymous_page(mm, vma, address,
3718 : : pmd, flags);
3719 : : if (!(ret & VM_FAULT_FALLBACK))
3720 : : return ret;
3721 : : } else {
3722 : : pmd_t orig_pmd = *pmd;
3723 : : int ret;
3724 : :
3725 : 66895889 : barrier();
3726 : : if (pmd_trans_huge(orig_pmd)) {
3727 : : unsigned int dirty = flags & FAULT_FLAG_WRITE;
3728 : :
3729 : : /*
3730 : : * If the pmd is splitting, return and retry the
3731 : : * the fault. Alternative: wait until the split
3732 : : * is done, and goto retry.
3733 : : */
3734 : : if (pmd_trans_splitting(orig_pmd))
3735 : : return 0;
3736 : :
3737 : : if (pmd_numa(orig_pmd))
3738 : : return do_huge_pmd_numa_page(mm, vma, address,
3739 : : orig_pmd, pmd);
3740 : :
3741 : : if (dirty && !pmd_write(orig_pmd)) {
3742 : : ret = do_huge_pmd_wp_page(mm, vma, address, pmd,
3743 : : orig_pmd);
3744 : : if (!(ret & VM_FAULT_FALLBACK))
3745 : : return ret;
3746 : : } else {
3747 : : huge_pmd_set_accessed(mm, vma, address, pmd,
3748 : : orig_pmd, dirty);
3749 : : return 0;
3750 : : }
3751 : : }
3752 : : }
3753 : :
3754 : : /* THP should already have been handled */
3755 : : BUG_ON(pmd_numa(*pmd));
3756 : :
3757 : : /*
3758 : : * Use __pte_alloc instead of pte_alloc_map, because we can't
3759 : : * run pte_offset_map on the pmd, if an huge pmd could
3760 : : * materialize from under us from a different thread.
3761 : : */
3762 [ + + + ]: 67078571 : if (unlikely(pmd_none(*pmd)) &&
3763 : 169097 : unlikely(__pte_alloc(mm, vma, pmd, address)))
3764 : : return VM_FAULT_OOM;
3765 : : /* if an huge pmd materialized from under us just retry later */
3766 : : if (unlikely(pmd_trans_huge(*pmd)))
3767 : : return 0;
3768 : : /*
3769 : : * A regular pmd is established and it can't morph into a huge pmd
3770 : : * from under us anymore at this point because we hold the mmap_sem
3771 : : * read mode and khugepaged takes it in write mode. So now it's
3772 : : * safe to run pte_offset_map().
3773 : : */
3774 : 66909476 : pte = pte_offset_map(pmd, address);
3775 : :
3776 : 66928057 : return handle_pte_fault(mm, vma, address, pte, pmd, flags);
3777 : : }
3778 : :
3779 : 0 : int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3780 : : unsigned long address, unsigned int flags)
3781 : : {
3782 : : int ret;
3783 : :
3784 : 66910987 : __set_current_state(TASK_RUNNING);
3785 : :
3786 : : count_vm_event(PGFAULT);
3787 : : mem_cgroup_count_vm_event(mm, PGFAULT);
3788 : :
3789 : : /* do counter updates before entering really critical section. */
3790 : 66927754 : check_sync_rss_stat(current);
3791 : :
3792 : : /*
3793 : : * Enable the memcg OOM handling for faults triggered in user
3794 : : * space. Kernel faults are handled more gracefully.
3795 : : */
3796 : : if (flags & FAULT_FLAG_USER)
3797 : : mem_cgroup_oom_enable();
3798 : :
3799 : 66929172 : ret = __handle_mm_fault(mm, vma, address, flags);
3800 : :
3801 : : if (flags & FAULT_FLAG_USER) {
3802 : : mem_cgroup_oom_disable();
3803 : : /*
3804 : : * The task may have entered a memcg OOM situation but
3805 : : * if the allocation error was handled gracefully (no
3806 : : * VM_FAULT_OOM), there is no need to kill anything.
3807 : : * Just clean up the OOM state peacefully.
3808 : : */
3809 : : if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3810 : : mem_cgroup_oom_synchronize(false);
3811 : : }
3812 : :
3813 : 66916082 : return ret;
3814 : : }
3815 : :
3816 : : #ifndef __PAGETABLE_PUD_FOLDED
3817 : : /*
3818 : : * Allocate page upper directory.
3819 : : * We've already handled the fast-path in-line.
3820 : : */
3821 : : int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
3822 : : {
3823 : : pud_t *new = pud_alloc_one(mm, address);
3824 : : if (!new)
3825 : : return -ENOMEM;
3826 : :
3827 : : smp_wmb(); /* See comment in __pte_alloc */
3828 : :
3829 : : spin_lock(&mm->page_table_lock);
3830 : : if (pgd_present(*pgd)) /* Another has populated it */
3831 : : pud_free(mm, new);
3832 : : else
3833 : : pgd_populate(mm, pgd, new);
3834 : : spin_unlock(&mm->page_table_lock);
3835 : : return 0;
3836 : : }
3837 : : #endif /* __PAGETABLE_PUD_FOLDED */
3838 : :
3839 : : #ifndef __PAGETABLE_PMD_FOLDED
3840 : : /*
3841 : : * Allocate page middle directory.
3842 : : * We've already handled the fast-path in-line.
3843 : : */
3844 : 0 : int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
3845 : : {
3846 : 0 : pmd_t *new = pmd_alloc_one(mm, address);
3847 : : if (!new)
3848 : : return -ENOMEM;
3849 : :
3850 : : smp_wmb(); /* See comment in __pte_alloc */
3851 : :
3852 : : spin_lock(&mm->page_table_lock);
3853 : : #ifndef __ARCH_HAS_4LEVEL_HACK
3854 : : if (pud_present(*pud)) /* Another has populated it */
3855 : : pmd_free(mm, new);
3856 : : else
3857 : : pud_populate(mm, pud, new);
3858 : : #else
3859 : : if (pgd_present(*pud)) /* Another has populated it */
3860 : : pmd_free(mm, new);
3861 : : else
3862 : : pgd_populate(mm, pud, new);
3863 : : #endif /* __ARCH_HAS_4LEVEL_HACK */
3864 : : spin_unlock(&mm->page_table_lock);
3865 : : return 0;
3866 : : }
3867 : : #endif /* __PAGETABLE_PMD_FOLDED */
3868 : :
3869 : : #if !defined(__HAVE_ARCH_GATE_AREA)
3870 : :
3871 : : #if defined(AT_SYSINFO_EHDR)
3872 : : static struct vm_area_struct gate_vma;
3873 : :
3874 : : static int __init gate_vma_init(void)
3875 : : {
3876 : : gate_vma.vm_mm = NULL;
3877 : : gate_vma.vm_start = FIXADDR_USER_START;
3878 : : gate_vma.vm_end = FIXADDR_USER_END;
3879 : : gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3880 : : gate_vma.vm_page_prot = __P101;
3881 : :
3882 : : return 0;
3883 : : }
3884 : : __initcall(gate_vma_init);
3885 : : #endif
3886 : :
3887 : : struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
3888 : : {
3889 : : #ifdef AT_SYSINFO_EHDR
3890 : : return &gate_vma;
3891 : : #else
3892 : : return NULL;
3893 : : #endif
3894 : : }
3895 : :
3896 : : int in_gate_area_no_mm(unsigned long addr)
3897 : : {
3898 : : #ifdef AT_SYSINFO_EHDR
3899 : : if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3900 : : return 1;
3901 : : #endif
3902 : : return 0;
3903 : : }
3904 : :
3905 : : #endif /* __HAVE_ARCH_GATE_AREA */
3906 : :
3907 : 0 : static int __follow_pte(struct mm_struct *mm, unsigned long address,
3908 : : pte_t **ptepp, spinlock_t **ptlp)
3909 : : {
3910 : : pgd_t *pgd;
3911 : : pud_t *pud;
3912 : : pmd_t *pmd;
3913 : : pte_t *ptep;
3914 : :
3915 : 0 : pgd = pgd_offset(mm, address);
3916 : : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3917 : : goto out;
3918 : :
3919 : : pud = pud_offset(pgd, address);
3920 : : if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3921 : : goto out;
3922 : :
3923 : : pmd = pmd_offset(pud, address);
3924 : : VM_BUG_ON(pmd_trans_huge(*pmd));
3925 [ # # ][ # # ]: 0 : if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3926 : : goto out;
3927 : :
3928 : : /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3929 : : if (pmd_huge(*pmd))
3930 : : goto out;
3931 : :
3932 : 0 : ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3933 [ # # ]: 0 : if (!ptep)
3934 : : goto out;
3935 [ # # ]: 0 : if (!pte_present(*ptep))
3936 : : goto unlock;
3937 : 0 : *ptepp = ptep;
3938 : : return 0;
3939 : : unlock:
3940 : 0 : pte_unmap_unlock(ptep, *ptlp);
3941 : : out:
3942 : : return -EINVAL;
3943 : : }
3944 : :
3945 : : static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3946 : : pte_t **ptepp, spinlock_t **ptlp)
3947 : : {
3948 : : int res;
3949 : :
3950 : : /* (void) is needed to make gcc happy */
3951 : 0 : (void) __cond_lock(*ptlp,
3952 : : !(res = __follow_pte(mm, address, ptepp, ptlp)));
3953 : : return res;
3954 : : }
3955 : :
3956 : : /**
3957 : : * follow_pfn - look up PFN at a user virtual address
3958 : : * @vma: memory mapping
3959 : : * @address: user virtual address
3960 : : * @pfn: location to store found PFN
3961 : : *
3962 : : * Only IO mappings and raw PFN mappings are allowed.
3963 : : *
3964 : : * Returns zero and the pfn at @pfn on success, -ve otherwise.
3965 : : */
3966 : 0 : int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3967 : : unsigned long *pfn)
3968 : : {
3969 : : int ret = -EINVAL;
3970 : : spinlock_t *ptl;
3971 : : pte_t *ptep;
3972 : :
3973 [ # # ]: 0 : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3974 : : return ret;
3975 : :
3976 : 0 : ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3977 [ # # ]: 0 : if (ret)
3978 : : return ret;
3979 : 0 : *pfn = pte_pfn(*ptep);
3980 : 0 : pte_unmap_unlock(ptep, ptl);
3981 : 0 : return 0;
3982 : : }
3983 : : EXPORT_SYMBOL(follow_pfn);
3984 : :
3985 : : #ifdef CONFIG_HAVE_IOREMAP_PROT
3986 : : int follow_phys(struct vm_area_struct *vma,
3987 : : unsigned long address, unsigned int flags,
3988 : : unsigned long *prot, resource_size_t *phys)
3989 : : {
3990 : : int ret = -EINVAL;
3991 : : pte_t *ptep, pte;
3992 : : spinlock_t *ptl;
3993 : :
3994 : : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3995 : : goto out;
3996 : :
3997 : : if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
3998 : : goto out;
3999 : : pte = *ptep;
4000 : :
4001 : : if ((flags & FOLL_WRITE) && !pte_write(pte))
4002 : : goto unlock;
4003 : :
4004 : : *prot = pgprot_val(pte_pgprot(pte));
4005 : : *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
4006 : :
4007 : : ret = 0;
4008 : : unlock:
4009 : : pte_unmap_unlock(ptep, ptl);
4010 : : out:
4011 : : return ret;
4012 : : }
4013 : :
4014 : : int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
4015 : : void *buf, int len, int write)
4016 : : {
4017 : : resource_size_t phys_addr;
4018 : : unsigned long prot = 0;
4019 : : void __iomem *maddr;
4020 : : int offset = addr & (PAGE_SIZE-1);
4021 : :
4022 : : if (follow_phys(vma, addr, write, &prot, &phys_addr))
4023 : : return -EINVAL;
4024 : :
4025 : : maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
4026 : : if (write)
4027 : : memcpy_toio(maddr + offset, buf, len);
4028 : : else
4029 : : memcpy_fromio(buf, maddr + offset, len);
4030 : : iounmap(maddr);
4031 : :
4032 : : return len;
4033 : : }
4034 : : EXPORT_SYMBOL_GPL(generic_access_phys);
4035 : : #endif
4036 : :
4037 : : /*
4038 : : * Access another process' address space as given in mm. If non-NULL, use the
4039 : : * given task for page fault accounting.
4040 : : */
4041 : 0 : static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
4042 : : unsigned long addr, void *buf, int len, int write)
4043 : : {
4044 : : struct vm_area_struct *vma;
4045 : : void *old_buf = buf;
4046 : :
4047 : 62 : down_read(&mm->mmap_sem);
4048 : : /* ignore errors, just check how much was successfully transferred */
4049 [ + + ]: 122 : while (len) {
4050 : : int bytes, ret, offset;
4051 : : void *maddr;
4052 : 62 : struct page *page = NULL;
4053 : :
4054 : 62 : ret = get_user_pages(tsk, mm, addr, 1,
4055 : : write, 1, &page, &vma);
4056 [ + + ]: 62 : if (ret <= 0) {
4057 : : /*
4058 : : * Check if this is a VM_IO | VM_PFNMAP VMA, which
4059 : : * we can access using slightly different code.
4060 : : */
4061 : : #ifdef CONFIG_HAVE_IOREMAP_PROT
4062 : : vma = find_vma(mm, addr);
4063 : : if (!vma || vma->vm_start > addr)
4064 : : break;
4065 : : if (vma->vm_ops && vma->vm_ops->access)
4066 : : ret = vma->vm_ops->access(vma, addr, buf,
4067 : : len, write);
4068 : : if (ret <= 0)
4069 : : #endif
4070 : 2 : break;
4071 : : bytes = ret;
4072 : : } else {
4073 : : bytes = len;
4074 : 60 : offset = addr & (PAGE_SIZE-1);
4075 [ - + ]: 60 : if (bytes > PAGE_SIZE-offset)
4076 : 0 : bytes = PAGE_SIZE-offset;
4077 : :
4078 : 60 : maddr = kmap(page);
4079 [ - + ]: 60 : if (write) {
4080 : 0 : copy_to_user_page(vma, page, addr,
4081 : : maddr + offset, buf, bytes);
4082 : 0 : set_page_dirty_lock(page);
4083 : : } else {
4084 : 60 : copy_from_user_page(vma, page, addr,
4085 : : buf, maddr + offset, bytes);
4086 : : }
4087 : 60 : kunmap(page);
4088 : 60 : page_cache_release(page);
4089 : : }
4090 : 60 : len -= bytes;
4091 : 60 : buf += bytes;
4092 : 60 : addr += bytes;
4093 : : }
4094 : 62 : up_read(&mm->mmap_sem);
4095 : :
4096 : 62 : return buf - old_buf;
4097 : : }
4098 : :
4099 : : /**
4100 : : * access_remote_vm - access another process' address space
4101 : : * @mm: the mm_struct of the target address space
4102 : : * @addr: start address to access
4103 : : * @buf: source or destination buffer
4104 : : * @len: number of bytes to transfer
4105 : : * @write: whether the access is a write
4106 : : *
4107 : : * The caller must hold a reference on @mm.
4108 : : */
4109 : 0 : int access_remote_vm(struct mm_struct *mm, unsigned long addr,
4110 : : void *buf, int len, int write)
4111 : : {
4112 : 6 : return __access_remote_vm(NULL, mm, addr, buf, len, write);
4113 : : }
4114 : :
4115 : : /*
4116 : : * Access another process' address space.
4117 : : * Source/target buffer must be kernel space,
4118 : : * Do not walk the page table directly, use get_user_pages
4119 : : */
4120 : 0 : int access_process_vm(struct task_struct *tsk, unsigned long addr,
4121 : : void *buf, int len, int write)
4122 : : {
4123 : : struct mm_struct *mm;
4124 : : int ret;
4125 : :
4126 : 56 : mm = get_task_mm(tsk);
4127 [ + - ]: 56 : if (!mm)
4128 : : return 0;
4129 : :
4130 : 56 : ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
4131 : 56 : mmput(mm);
4132 : :
4133 : 56 : return ret;
4134 : : }
4135 : :
4136 : : /*
4137 : : * Print the name of a VMA.
4138 : : */
4139 : 0 : void print_vma_addr(char *prefix, unsigned long ip)
4140 : : {
4141 : 0 : struct mm_struct *mm = current->mm;
4142 : : struct vm_area_struct *vma;
4143 : :
4144 : : /*
4145 : : * Do not print if we are in atomic
4146 : : * contexts (in exception stacks, etc.):
4147 : : */
4148 [ # # ]: 0 : if (preempt_count())
4149 : 0 : return;
4150 : :
4151 : 0 : down_read(&mm->mmap_sem);
4152 : 0 : vma = find_vma(mm, ip);
4153 [ # # ][ # # ]: 0 : if (vma && vma->vm_file) {
4154 : : struct file *f = vma->vm_file;
4155 : 0 : char *buf = (char *)__get_free_page(GFP_KERNEL);
4156 [ # # ]: 0 : if (buf) {
4157 : : char *p;
4158 : :
4159 : 0 : p = d_path(&f->f_path, buf, PAGE_SIZE);
4160 [ # # ]: 0 : if (IS_ERR(p))
4161 : : p = "?";
4162 : 0 : printk("%s%s[%lx+%lx]", prefix, kbasename(p),
4163 : : vma->vm_start,
4164 : 0 : vma->vm_end - vma->vm_start);
4165 : 0 : free_page((unsigned long)buf);
4166 : : }
4167 : : }
4168 : 0 : up_read(&mm->mmap_sem);
4169 : : }
4170 : :
4171 : : #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
4172 : : void might_fault(void)
4173 : : {
4174 : : /*
4175 : : * Some code (nfs/sunrpc) uses socket ops on kernel memory while
4176 : : * holding the mmap_sem, this is safe because kernel memory doesn't
4177 : : * get paged out, therefore we'll never actually fault, and the
4178 : : * below annotations will generate false positives.
4179 : : */
4180 : : if (segment_eq(get_fs(), KERNEL_DS))
4181 : : return;
4182 : :
4183 : : /*
4184 : : * it would be nicer only to annotate paths which are not under
4185 : : * pagefault_disable, however that requires a larger audit and
4186 : : * providing helpers like get_user_atomic.
4187 : : */
4188 : : if (in_atomic())
4189 : : return;
4190 : :
4191 : : __might_sleep(__FILE__, __LINE__, 0);
4192 : :
4193 : : if (current->mm)
4194 : : might_lock_read(¤t->mm->mmap_sem);
4195 : : }
4196 : : EXPORT_SYMBOL(might_fault);
4197 : : #endif
4198 : :
4199 : : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4200 : : static void clear_gigantic_page(struct page *page,
4201 : : unsigned long addr,
4202 : : unsigned int pages_per_huge_page)
4203 : : {
4204 : : int i;
4205 : : struct page *p = page;
4206 : :
4207 : : might_sleep();
4208 : : for (i = 0; i < pages_per_huge_page;
4209 : : i++, p = mem_map_next(p, page, i)) {
4210 : : cond_resched();
4211 : : clear_user_highpage(p, addr + i * PAGE_SIZE);
4212 : : }
4213 : : }
4214 : : void clear_huge_page(struct page *page,
4215 : : unsigned long addr, unsigned int pages_per_huge_page)
4216 : : {
4217 : : int i;
4218 : :
4219 : : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4220 : : clear_gigantic_page(page, addr, pages_per_huge_page);
4221 : : return;
4222 : : }
4223 : :
4224 : : might_sleep();
4225 : : for (i = 0; i < pages_per_huge_page; i++) {
4226 : : cond_resched();
4227 : : clear_user_highpage(page + i, addr + i * PAGE_SIZE);
4228 : : }
4229 : : }
4230 : :
4231 : : static void copy_user_gigantic_page(struct page *dst, struct page *src,
4232 : : unsigned long addr,
4233 : : struct vm_area_struct *vma,
4234 : : unsigned int pages_per_huge_page)
4235 : : {
4236 : : int i;
4237 : : struct page *dst_base = dst;
4238 : : struct page *src_base = src;
4239 : :
4240 : : for (i = 0; i < pages_per_huge_page; ) {
4241 : : cond_resched();
4242 : : copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
4243 : :
4244 : : i++;
4245 : : dst = mem_map_next(dst, dst_base, i);
4246 : : src = mem_map_next(src, src_base, i);
4247 : : }
4248 : : }
4249 : :
4250 : : void copy_user_huge_page(struct page *dst, struct page *src,
4251 : : unsigned long addr, struct vm_area_struct *vma,
4252 : : unsigned int pages_per_huge_page)
4253 : : {
4254 : : int i;
4255 : :
4256 : : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4257 : : copy_user_gigantic_page(dst, src, addr, vma,
4258 : : pages_per_huge_page);
4259 : : return;
4260 : : }
4261 : :
4262 : : might_sleep();
4263 : : for (i = 0; i < pages_per_huge_page; i++) {
4264 : : cond_resched();
4265 : : copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
4266 : : }
4267 : : }
4268 : : #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
4269 : :
4270 : : #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
4271 : :
4272 : : static struct kmem_cache *page_ptl_cachep;
4273 : :
4274 : : void __init ptlock_cache_init(void)
4275 : : {
4276 : : page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
4277 : : SLAB_PANIC, NULL);
4278 : : }
4279 : :
4280 : : bool ptlock_alloc(struct page *page)
4281 : : {
4282 : : spinlock_t *ptl;
4283 : :
4284 : : ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
4285 : : if (!ptl)
4286 : : return false;
4287 : : page->ptl = ptl;
4288 : : return true;
4289 : : }
4290 : :
4291 : : void ptlock_free(struct page *page)
4292 : : {
4293 : : kmem_cache_free(page_ptl_cachep, page->ptl);
4294 : : }
4295 : : #endif
|
