Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0-only
2 : : /*
3 : : * linux/mm/memory.c
4 : : *
5 : : * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 : : */
7 : :
8 : : /*
9 : : * demand-loading started 01.12.91 - seems it is high on the list of
10 : : * things wanted, and it should be easy to implement. - Linus
11 : : */
12 : :
13 : : /*
14 : : * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
15 : : * pages started 02.12.91, seems to work. - Linus.
16 : : *
17 : : * Tested sharing by executing about 30 /bin/sh: under the old kernel it
18 : : * would have taken more than the 6M I have free, but it worked well as
19 : : * far as I could see.
20 : : *
21 : : * Also corrected some "invalidate()"s - I wasn't doing enough of them.
22 : : */
23 : :
24 : : /*
25 : : * Real VM (paging to/from disk) started 18.12.91. Much more work and
26 : : * thought has to go into this. Oh, well..
27 : : * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
28 : : * Found it. Everything seems to work now.
29 : : * 20.12.91 - Ok, making the swap-device changeable like the root.
30 : : */
31 : :
32 : : /*
33 : : * 05.04.94 - Multi-page memory management added for v1.1.
34 : : * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 : : *
36 : : * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
37 : : * (Gerhard.Wichert@pdb.siemens.de)
38 : : *
39 : : * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
40 : : */
41 : :
42 : : #include <linux/kernel_stat.h>
43 : : #include <linux/mm.h>
44 : : #include <linux/sched/mm.h>
45 : : #include <linux/sched/coredump.h>
46 : : #include <linux/sched/numa_balancing.h>
47 : : #include <linux/sched/task.h>
48 : : #include <linux/hugetlb.h>
49 : : #include <linux/mman.h>
50 : : #include <linux/swap.h>
51 : : #include <linux/highmem.h>
52 : : #include <linux/pagemap.h>
53 : : #include <linux/memremap.h>
54 : : #include <linux/ksm.h>
55 : : #include <linux/rmap.h>
56 : : #include <linux/export.h>
57 : : #include <linux/delayacct.h>
58 : : #include <linux/init.h>
59 : : #include <linux/pfn_t.h>
60 : : #include <linux/writeback.h>
61 : : #include <linux/memcontrol.h>
62 : : #include <linux/mmu_notifier.h>
63 : : #include <linux/swapops.h>
64 : : #include <linux/elf.h>
65 : : #include <linux/gfp.h>
66 : : #include <linux/migrate.h>
67 : : #include <linux/string.h>
68 : : #include <linux/dma-debug.h>
69 : : #include <linux/debugfs.h>
70 : : #include <linux/userfaultfd_k.h>
71 : : #include <linux/dax.h>
72 : : #include <linux/oom.h>
73 : : #include <linux/numa.h>
74 : :
75 : : #include <asm/io.h>
76 : : #include <asm/mmu_context.h>
77 : : #include <asm/pgalloc.h>
78 : : #include <linux/uaccess.h>
79 : : #include <asm/tlb.h>
80 : : #include <asm/tlbflush.h>
81 : : #include <asm/pgtable.h>
82 : :
83 : : #include "internal.h"
84 : :
85 : : #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST)
86 : : #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
87 : : #endif
88 : :
89 : : #ifndef CONFIG_NEED_MULTIPLE_NODES
90 : : /* use the per-pgdat data instead for discontigmem - mbligh */
91 : : unsigned long max_mapnr;
92 : : EXPORT_SYMBOL(max_mapnr);
93 : :
94 : : struct page *mem_map;
95 : : EXPORT_SYMBOL(mem_map);
96 : : #endif
97 : :
98 : : /*
99 : : * A number of key systems in x86 including ioremap() rely on the assumption
100 : : * that high_memory defines the upper bound on direct map memory, then end
101 : : * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
102 : : * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
103 : : * and ZONE_HIGHMEM.
104 : : */
105 : : void *high_memory;
106 : : EXPORT_SYMBOL(high_memory);
107 : :
108 : : /*
109 : : * Randomize the address space (stacks, mmaps, brk, etc.).
110 : : *
111 : : * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
112 : : * as ancient (libc5 based) binaries can segfault. )
113 : : */
114 : : int randomize_va_space __read_mostly =
115 : : #ifdef CONFIG_COMPAT_BRK
116 : : 1;
117 : : #else
118 : : 2;
119 : : #endif
120 : :
121 : 0 : static int __init disable_randmaps(char *s)
122 : : {
123 : 0 : randomize_va_space = 0;
124 : 0 : return 1;
125 : : }
126 : : __setup("norandmaps", disable_randmaps);
127 : :
128 : : unsigned long zero_pfn __read_mostly;
129 : : EXPORT_SYMBOL(zero_pfn);
130 : :
131 : : unsigned long highest_memmap_pfn __read_mostly;
132 : :
133 : : /*
134 : : * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
135 : : */
136 : 207 : static int __init init_zero_pfn(void)
137 : : {
138 : 207 : zero_pfn = page_to_pfn(ZERO_PAGE(0));
139 : 207 : return 0;
140 : : }
141 : : core_initcall(init_zero_pfn);
142 : :
143 : :
144 : : #if defined(SPLIT_RSS_COUNTING)
145 : :
146 : 3597760 : void sync_mm_rss(struct mm_struct *mm)
147 : : {
148 : : int i;
149 : :
150 [ + + ]: 17935044 : for (i = 0; i < NR_MM_COUNTERS; i++) {
151 [ + + ]: 14345559 : if (current->rss_stat.count[i]) {
152 : : add_mm_counter(mm, i, current->rss_stat.count[i]);
153 : 4950398 : current->rss_stat.count[i] = 0;
154 : : }
155 : : }
156 : 3589485 : current->rss_stat.events = 0;
157 : 3589485 : }
158 : :
159 : 186917850 : static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
160 : : {
161 : 186917850 : struct task_struct *task = current;
162 : :
163 [ + + ]: 186917850 : if (likely(task->mm == mm))
164 : 186683018 : task->rss_stat.count[member] += val;
165 : : else
166 : : add_mm_counter(mm, member, val);
167 : 186917983 : }
168 : : #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
169 : : #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
170 : :
171 : : /* sync counter once per 64 page faults */
172 : : #define TASK_RSS_EVENTS_THRESH (64)
173 : 46581363 : static void check_sync_rss_stat(struct task_struct *task)
174 : : {
175 [ + - ]: 46581363 : if (unlikely(task != current))
176 : 46581363 : return;
177 [ + + ]: 46586739 : if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
178 : 295275 : sync_mm_rss(task->mm);
179 : : }
180 : : #else /* SPLIT_RSS_COUNTING */
181 : :
182 : : #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
183 : : #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
184 : :
185 : : static void check_sync_rss_stat(struct task_struct *task)
186 : : {
187 : : }
188 : :
189 : : #endif /* SPLIT_RSS_COUNTING */
190 : :
191 : : /*
192 : : * Note: this doesn't free the actual pages themselves. That
193 : : * has been handled earlier when unmapping all the memory regions.
194 : : */
195 : 2998909 : static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
196 : : unsigned long addr)
197 : : {
198 : 2998909 : pgtable_t token = pmd_pgtable(*pmd);
199 : 2998909 : pmd_clear(pmd);
200 : 2998803 : pte_free_tlb(tlb, token, addr);
201 : 2998823 : mm_dec_nr_ptes(tlb->mm);
202 : 2998859 : }
203 : :
204 : 3634980 : static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
205 : : unsigned long addr, unsigned long end,
206 : : unsigned long floor, unsigned long ceiling)
207 : : {
208 : : pmd_t *pmd;
209 : : unsigned long next;
210 : : unsigned long start;
211 : :
212 : : start = addr;
213 : : pmd = pmd_offset(pud, addr);
214 : : do {
215 : : next = pmd_addr_end(addr, end);
216 [ + + ]: 3634819 : if (pmd_none_or_clear_bad(pmd))
217 : 636005 : continue;
218 : 2998814 : free_pte_range(tlb, pmd, addr);
219 : : } while (pmd++, addr = next, addr != end);
220 : :
221 : 3634956 : start &= PUD_MASK;
222 [ + - ]: 3634956 : if (start < floor)
223 : : return;
224 [ + + ]: 3634956 : if (ceiling) {
225 : 2851379 : ceiling &= PUD_MASK;
226 [ + + ]: 2851379 : if (!ceiling)
227 : : return;
228 : : }
229 [ + - ]: 3634571 : if (end - 1 > ceiling - 1)
230 : : return;
231 : :
232 : : pmd = pmd_offset(pud, start);
233 : : pud_clear(pud);
234 : 3634571 : pmd_free_tlb(tlb, pmd, start);
235 : : mm_dec_nr_pmds(tlb->mm);
236 : : }
237 : :
238 : 3634856 : static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
239 : : unsigned long addr, unsigned long end,
240 : : unsigned long floor, unsigned long ceiling)
241 : : {
242 : : pud_t *pud;
243 : : unsigned long next;
244 : : unsigned long start;
245 : :
246 : : start = addr;
247 : : pud = pud_offset(p4d, addr);
248 : : do {
249 : : next = pud_addr_end(addr, end);
250 : : if (pud_none_or_clear_bad(pud))
251 : : continue;
252 : 3634856 : free_pmd_range(tlb, pud, addr, next, floor, ceiling);
253 : : } while (pud++, addr = next, addr != end);
254 : :
255 : 3634788 : start &= P4D_MASK;
256 [ + + ]: 3634788 : if (start < floor)
257 : : return;
258 [ + + ]: 3634644 : if (ceiling) {
259 : 2851356 : ceiling &= P4D_MASK;
260 [ + + ]: 2851356 : if (!ceiling)
261 : : return;
262 : : }
263 [ + - ]: 3634672 : if (end - 1 > ceiling - 1)
264 : : return;
265 : :
266 : : pud = pud_offset(p4d, start);
267 : : p4d_clear(p4d);
268 : 3634672 : pud_free_tlb(tlb, pud, start);
269 : : mm_dec_nr_puds(tlb->mm);
270 : : }
271 : :
272 : : static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd,
273 : : unsigned long addr, unsigned long end,
274 : : unsigned long floor, unsigned long ceiling)
275 : : {
276 : : p4d_t *p4d;
277 : : unsigned long next;
278 : : unsigned long start;
279 : :
280 : : start = addr;
281 : : p4d = p4d_offset(pgd, addr);
282 : : do {
283 : : next = p4d_addr_end(addr, end);
284 : : if (p4d_none_or_clear_bad(p4d))
285 : : continue;
286 : 3634888 : free_pud_range(tlb, p4d, addr, next, floor, ceiling);
287 : : } while (p4d++, addr = next, addr != end);
288 : :
289 : : start &= PGDIR_MASK;
290 : : if (start < floor)
291 : : return;
292 : : if (ceiling) {
293 : : ceiling &= PGDIR_MASK;
294 : : if (!ceiling)
295 : : return;
296 : : }
297 : : if (end - 1 > ceiling - 1)
298 : : return;
299 : :
300 : : p4d = p4d_offset(pgd, start);
301 : : pgd_clear(pgd);
302 : : p4d_free_tlb(tlb, p4d, start);
303 : : }
304 : :
305 : : /*
306 : : * This function frees user-level page tables of a process.
307 : : */
308 : 5024504 : void free_pgd_range(struct mmu_gather *tlb,
309 : : unsigned long addr, unsigned long end,
310 : : unsigned long floor, unsigned long ceiling)
311 : : {
312 : : pgd_t *pgd;
313 : : unsigned long next;
314 : :
315 : : /*
316 : : * The next few lines have given us lots of grief...
317 : : *
318 : : * Why are we testing PMD* at this top level? Because often
319 : : * there will be no work to do at all, and we'd prefer not to
320 : : * go all the way down to the bottom just to discover that.
321 : : *
322 : : * Why all these "- 1"s? Because 0 represents both the bottom
323 : : * of the address space and the top of it (using -1 for the
324 : : * top wouldn't help much: the masks would do the wrong thing).
325 : : * The rule is that addr 0 and floor 0 refer to the bottom of
326 : : * the address space, but end 0 and ceiling 0 refer to the top
327 : : * Comparisons need to use "end - 1" and "ceiling - 1" (though
328 : : * that end 0 case should be mythical).
329 : : *
330 : : * Wherever addr is brought up or ceiling brought down, we must
331 : : * be careful to reject "the opposite 0" before it confuses the
332 : : * subsequent tests. But what about where end is brought down
333 : : * by PMD_SIZE below? no, end can't go down to 0 there.
334 : : *
335 : : * Whereas we round start (addr) and ceiling down, by different
336 : : * masks at different levels, in order to test whether a table
337 : : * now has no other vmas using it, so can be freed, we don't
338 : : * bother to round floor or end up - the tests don't need that.
339 : : */
340 : :
341 : 5024504 : addr &= PMD_MASK;
342 [ + + ]: 5024504 : if (addr < floor) {
343 : 3066066 : addr += PMD_SIZE;
344 [ + + ]: 3066066 : if (!addr)
345 : : return;
346 : : }
347 [ + + ]: 5021652 : if (ceiling) {
348 : 4295075 : ceiling &= PMD_MASK;
349 [ + - ]: 4295075 : if (!ceiling)
350 : : return;
351 : : }
352 [ + + ]: 5024980 : if (end - 1 > ceiling - 1)
353 : 2363019 : end -= PMD_SIZE;
354 [ + + ]: 5024980 : if (addr > end - 1)
355 : : return;
356 : : /*
357 : : * We add page table cache pages with PAGE_SIZE,
358 : : * (see pte_free_tlb()), flush the tlb if we need
359 : : */
360 : : tlb_change_page_size(tlb, PAGE_SIZE);
361 : 1891895 : pgd = pgd_offset(tlb->mm, addr);
362 : : do {
363 [ + + ]: 3634888 : next = pgd_addr_end(addr, end);
364 : : if (pgd_none_or_clear_bad(pgd))
365 : : continue;
366 : : free_p4d_range(tlb, pgd, addr, next, floor, ceiling);
367 [ + + ]: 3634857 : } while (pgd++, addr = next, addr != end);
368 : : }
369 : :
370 : 3152414 : void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
371 : : unsigned long floor, unsigned long ceiling)
372 : : {
373 [ + + ]: 11091979 : while (vma) {
374 : 4785495 : struct vm_area_struct *next = vma->vm_next;
375 : 4785495 : unsigned long addr = vma->vm_start;
376 : :
377 : : /*
378 : : * Hide vma from rmap and truncate_pagecache before freeing
379 : : * pgtables
380 : : */
381 : 4785495 : unlink_anon_vmas(vma);
382 : 4789628 : unlink_file_vma(vma);
383 : :
384 : : if (is_vm_hugetlb_page(vma)) {
385 : : hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
386 : : floor, next ? next->vm_start : ceiling);
387 : : } else {
388 : : /*
389 : : * Optimization: gather nearby vmas into one call down
390 : : */
391 [ + + + + ]: 24851869 : while (next && next->vm_start <= vma->vm_end + PMD_SIZE
392 : : && !is_vm_hugetlb_page(next)) {
393 : : vma = next;
394 : 20061022 : next = vma->vm_next;
395 : 20061022 : unlink_anon_vmas(vma);
396 : 20060932 : unlink_file_vma(vma);
397 : : }
398 [ + + ]: 4790847 : free_pgd_range(tlb, addr, vma->vm_end,
399 : : floor, next ? next->vm_start : ceiling);
400 : : }
401 : 4787151 : vma = next;
402 : : }
403 : 3154070 : }
404 : :
405 : 2316513 : int __pte_alloc(struct mm_struct *mm, pmd_t *pmd)
406 : : {
407 : : spinlock_t *ptl;
408 : 2316513 : pgtable_t new = pte_alloc_one(mm);
409 [ + + ]: 2316432 : if (!new)
410 : : return -ENOMEM;
411 : :
412 : : /*
413 : : * Ensure all pte setup (eg. pte page lock and page clearing) are
414 : : * visible before the pte is made visible to other CPUs by being
415 : : * put into page tables.
416 : : *
417 : : * The other side of the story is the pointer chasing in the page
418 : : * table walking code (when walking the page table without locking;
419 : : * ie. most of the time). Fortunately, these data accesses consist
420 : : * of a chain of data-dependent loads, meaning most CPUs (alpha
421 : : * being the notable exception) will already guarantee loads are
422 : : * seen in-order. See the alpha page table accessors for the
423 : : * smp_read_barrier_depends() barriers in page table walking code.
424 : : */
425 : 2316813 : smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
426 : :
427 : : ptl = pmd_lock(mm, pmd);
428 [ + + ]: 2317115 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
429 : : mm_inc_nr_ptes(mm);
430 : : pmd_populate(mm, pmd, new);
431 : : new = NULL;
432 : : }
433 : : spin_unlock(ptl);
434 [ + + ]: 2316698 : if (new)
435 : 3 : pte_free(mm, new);
436 : : return 0;
437 : : }
438 : :
439 : 11192 : int __pte_alloc_kernel(pmd_t *pmd)
440 : : {
441 : 11192 : pte_t *new = pte_alloc_one_kernel(&init_mm);
442 [ + - ]: 11191 : if (!new)
443 : : return -ENOMEM;
444 : :
445 : 11191 : smp_wmb(); /* See comment in __pte_alloc */
446 : :
447 : : spin_lock(&init_mm.page_table_lock);
448 [ + + ]: 11192 : if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
449 : : pmd_populate_kernel(&init_mm, pmd, new);
450 : : new = NULL;
451 : : }
452 : : spin_unlock(&init_mm.page_table_lock);
453 [ + + ]: 11192 : if (new)
454 : : pte_free_kernel(&init_mm, new);
455 : : return 0;
456 : : }
457 : :
458 : : static inline void init_rss_vec(int *rss)
459 : : {
460 : 32907732 : memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
461 : : }
462 : :
463 : 32911081 : static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
464 : : {
465 : : int i;
466 : :
467 [ + + ]: 32911081 : if (current->mm == mm)
468 : 2527851 : sync_mm_rss(mm);
469 [ + + ]: 131608621 : for (i = 0; i < NR_MM_COUNTERS; i++)
470 [ + + ]: 131622734 : if (rss[i])
471 : : add_mm_counter(mm, i, rss[i]);
472 : 32908033 : }
473 : :
474 : : /*
475 : : * This function is called to print an error when a bad pte
476 : : * is found. For example, we might have a PFN-mapped pte in
477 : : * a region that doesn't allow it.
478 : : *
479 : : * The calling function must still handle the error.
480 : : */
481 : 0 : static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
482 : : pte_t pte, struct page *page)
483 : : {
484 : 0 : pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
485 : : p4d_t *p4d = p4d_offset(pgd, addr);
486 : : pud_t *pud = pud_offset(p4d, addr);
487 : : pmd_t *pmd = pmd_offset(pud, addr);
488 : : struct address_space *mapping;
489 : : pgoff_t index;
490 : : static unsigned long resume;
491 : : static unsigned long nr_shown;
492 : : static unsigned long nr_unshown;
493 : :
494 : : /*
495 : : * Allow a burst of 60 reports, then keep quiet for that minute;
496 : : * or allow a steady drip of one report per second.
497 : : */
498 [ # # ]: 0 : if (nr_shown == 60) {
499 [ # # ]: 0 : if (time_before(jiffies, resume)) {
500 : 0 : nr_unshown++;
501 : 0 : return;
502 : : }
503 [ # # ]: 0 : if (nr_unshown) {
504 : 0 : pr_alert("BUG: Bad page map: %lu messages suppressed\n",
505 : : nr_unshown);
506 : 0 : nr_unshown = 0;
507 : : }
508 : 0 : nr_shown = 0;
509 : : }
510 [ # # ]: 0 : if (nr_shown++ == 0)
511 : 0 : resume = jiffies + 60 * HZ;
512 : :
513 [ # # ]: 0 : mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
514 : : index = linear_page_index(vma, addr);
515 : :
516 : 0 : pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
517 : : current->comm,
518 : : (long long)pte_val(pte), (long long)pmd_val(*pmd));
519 [ # # ]: 0 : if (page)
520 : 0 : dump_page(page, "bad pte");
521 : 0 : pr_alert("addr:%px vm_flags:%08lx anon_vma:%px mapping:%px index:%lx\n",
522 : : (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
523 [ # # # # : 0 : pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n",
# # ]
524 : : vma->vm_file,
525 : : vma->vm_ops ? vma->vm_ops->fault : NULL,
526 : : vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
527 : : mapping ? mapping->a_ops->readpage : NULL);
528 : 0 : dump_stack();
529 : 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
530 : : }
531 : :
532 : : /*
533 : : * vm_normal_page -- This function gets the "struct page" associated with a pte.
534 : : *
535 : : * "Special" mappings do not wish to be associated with a "struct page" (either
536 : : * it doesn't exist, or it exists but they don't want to touch it). In this
537 : : * case, NULL is returned here. "Normal" mappings do have a struct page.
538 : : *
539 : : * There are 2 broad cases. Firstly, an architecture may define a pte_special()
540 : : * pte bit, in which case this function is trivial. Secondly, an architecture
541 : : * may not have a spare pte bit, which requires a more complicated scheme,
542 : : * described below.
543 : : *
544 : : * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
545 : : * special mapping (even if there are underlying and valid "struct pages").
546 : : * COWed pages of a VM_PFNMAP are always normal.
547 : : *
548 : : * The way we recognize COWed pages within VM_PFNMAP mappings is through the
549 : : * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
550 : : * set, and the vm_pgoff will point to the first PFN mapped: thus every special
551 : : * mapping will always honor the rule
552 : : *
553 : : * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
554 : : *
555 : : * And for normal mappings this is false.
556 : : *
557 : : * This restricts such mappings to be a linear translation from virtual address
558 : : * to pfn. To get around this restriction, we allow arbitrary mappings so long
559 : : * as the vma is not a COW mapping; in that case, we know that all ptes are
560 : : * special (because none can have been COWed).
561 : : *
562 : : *
563 : : * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
564 : : *
565 : : * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
566 : : * page" backing, however the difference is that _all_ pages with a struct
567 : : * page (that is, those where pfn_valid is true) are refcounted and considered
568 : : * normal pages by the VM. The disadvantage is that pages are refcounted
569 : : * (which can be slower and simply not an option for some PFNMAP users). The
570 : : * advantage is that we don't have to follow the strict linearity rule of
571 : : * PFNMAP mappings in order to support COWable mappings.
572 : : *
573 : : */
574 : 222591641 : struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
575 : : pte_t pte)
576 : : {
577 : 222591641 : unsigned long pfn = pte_pfn(pte);
578 : :
579 : : if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) {
580 : : if (likely(!pte_special(pte)))
581 : : goto check_pfn;
582 : : if (vma->vm_ops && vma->vm_ops->find_special_page)
583 : : return vma->vm_ops->find_special_page(vma, addr);
584 : : if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
585 : : return NULL;
586 : : if (is_zero_pfn(pfn))
587 : : return NULL;
588 : : if (pte_devmap(pte))
589 : : return NULL;
590 : :
591 : : print_bad_pte(vma, addr, pte, NULL);
592 : : return NULL;
593 : : }
594 : :
595 : : /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */
596 : :
597 [ + + ]: 222591641 : if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
598 [ - + ]: 2095875 : if (vma->vm_flags & VM_MIXEDMAP) {
599 [ # # ]: 0 : if (!pfn_valid(pfn))
600 : : return NULL;
601 : : goto out;
602 : : } else {
603 : : unsigned long off;
604 : 2095875 : off = (addr - vma->vm_start) >> PAGE_SHIFT;
605 [ + - ]: 2095875 : if (pfn == vma->vm_pgoff + off)
606 : : return NULL;
607 [ + + ]: 2095875 : if (!is_cow_mapping(vma->vm_flags))
608 : : return NULL;
609 : : }
610 : : }
611 : :
612 [ + + ]: 220500278 : if (is_zero_pfn(pfn))
613 : : return NULL;
614 : :
615 : : check_pfn:
616 [ - + ]: 219510731 : if (unlikely(pfn > highest_memmap_pfn)) {
617 : 0 : print_bad_pte(vma, addr, pte, NULL);
618 : 0 : return NULL;
619 : : }
620 : :
621 : : /*
622 : : * NOTE! We still have PageReserved() pages in the page tables.
623 : : * eg. VDSO mappings can cause them to exist.
624 : : */
625 : : out:
626 : 219519950 : return pfn_to_page(pfn);
627 : : }
628 : :
629 : : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
630 : : struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
631 : : pmd_t pmd)
632 : : {
633 : : unsigned long pfn = pmd_pfn(pmd);
634 : :
635 : : /*
636 : : * There is no pmd_special() but there may be special pmds, e.g.
637 : : * in a direct-access (dax) mapping, so let's just replicate the
638 : : * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here.
639 : : */
640 : : if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
641 : : if (vma->vm_flags & VM_MIXEDMAP) {
642 : : if (!pfn_valid(pfn))
643 : : return NULL;
644 : : goto out;
645 : : } else {
646 : : unsigned long off;
647 : : off = (addr - vma->vm_start) >> PAGE_SHIFT;
648 : : if (pfn == vma->vm_pgoff + off)
649 : : return NULL;
650 : : if (!is_cow_mapping(vma->vm_flags))
651 : : return NULL;
652 : : }
653 : : }
654 : :
655 : : if (pmd_devmap(pmd))
656 : : return NULL;
657 : : if (is_zero_pfn(pfn))
658 : : return NULL;
659 : : if (unlikely(pfn > highest_memmap_pfn))
660 : : return NULL;
661 : :
662 : : /*
663 : : * NOTE! We still have PageReserved() pages in the page tables.
664 : : * eg. VDSO mappings can cause them to exist.
665 : : */
666 : : out:
667 : : return pfn_to_page(pfn);
668 : : }
669 : : #endif
670 : :
671 : : /*
672 : : * copy one vm_area from one task to the other. Assumes the page tables
673 : : * already present in the new task to be cleared in the whole range
674 : : * covered by this vma.
675 : : */
676 : :
677 : : static inline unsigned long
678 : 22777650 : copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
679 : : pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
680 : : unsigned long addr, int *rss)
681 : : {
682 : 22777650 : unsigned long vm_flags = vma->vm_flags;
683 : 22777650 : pte_t pte = *src_pte;
684 : : struct page *page;
685 : :
686 : : /* pte contains position in swap or file, so copy. */
687 [ - + ]: 22777650 : if (unlikely(!pte_present(pte))) {
688 : : swp_entry_t entry = pte_to_swp_entry(pte);
689 : :
690 [ # # ]: 0 : if (likely(!non_swap_entry(entry))) {
691 [ # # ]: 0 : if (swap_duplicate(entry) < 0)
692 : : return entry.val;
693 : :
694 : : /* make sure dst_mm is on swapoff's mmlist. */
695 [ # # ]: 0 : if (unlikely(list_empty(&dst_mm->mmlist))) {
696 : : spin_lock(&mmlist_lock);
697 [ # # ]: 0 : if (list_empty(&dst_mm->mmlist))
698 : 0 : list_add(&dst_mm->mmlist,
699 : : &src_mm->mmlist);
700 : : spin_unlock(&mmlist_lock);
701 : : }
702 : 0 : rss[MM_SWAPENTS]++;
703 [ # # ]: 0 : } else if (is_migration_entry(entry)) {
704 : 0 : page = migration_entry_to_page(entry);
705 : :
706 : 0 : rss[mm_counter(page)]++;
707 : :
708 [ # # # # ]: 0 : if (is_write_migration_entry(entry) &&
709 : : is_cow_mapping(vm_flags)) {
710 : : /*
711 : : * COW mappings require pages in both
712 : : * parent and child to be set to read.
713 : : */
714 : : make_migration_entry_read(&entry);
715 : : pte = swp_entry_to_pte(entry);
716 : : if (pte_swp_soft_dirty(*src_pte))
717 : : pte = pte_swp_mksoft_dirty(pte);
718 : 0 : set_pte_at(src_mm, addr, src_pte, pte);
719 : : }
720 : : } else if (is_device_private_entry(entry)) {
721 : : page = device_private_entry_to_page(entry);
722 : :
723 : : /*
724 : : * Update rss count even for unaddressable pages, as
725 : : * they should treated just like normal pages in this
726 : : * respect.
727 : : *
728 : : * We will likely want to have some new rss counters
729 : : * for unaddressable pages, at some point. But for now
730 : : * keep things as they are.
731 : : */
732 : : get_page(page);
733 : : rss[mm_counter(page)]++;
734 : : page_dup_rmap(page, false);
735 : :
736 : : /*
737 : : * We do not preserve soft-dirty information, because so
738 : : * far, checkpoint/restore is the only feature that
739 : : * requires that. And checkpoint/restore does not work
740 : : * when a device driver is involved (you cannot easily
741 : : * save and restore device driver state).
742 : : */
743 : : if (is_write_device_private_entry(entry) &&
744 : : is_cow_mapping(vm_flags)) {
745 : : make_device_private_entry_read(&entry);
746 : : pte = swp_entry_to_pte(entry);
747 : : set_pte_at(src_mm, addr, src_pte, pte);
748 : : }
749 : : }
750 : : goto out_set_pte;
751 : : }
752 : :
753 : : /*
754 : : * If it's a COW mapping, write protect it both
755 : : * in the parent and the child
756 : : */
757 [ + + + + ]: 22777650 : if (is_cow_mapping(vm_flags) && pte_write(pte)) {
758 : : ptep_set_wrprotect(src_mm, addr, src_pte);
759 : : pte = pte_wrprotect(pte);
760 : : }
761 : :
762 : : /*
763 : : * If it's a shared mapping, mark it clean in
764 : : * the child
765 : : */
766 [ + + ]: 22777629 : if (vm_flags & VM_SHARED)
767 : : pte = pte_mkclean(pte);
768 : : pte = pte_mkold(pte);
769 : :
770 : 22777629 : page = vm_normal_page(vma, addr, pte);
771 [ + + ]: 22777843 : if (page) {
772 : 21816632 : get_page(page);
773 : 21817330 : page_dup_rmap(page, false);
774 : 21817485 : rss[mm_counter(page)]++;
775 : : } else if (pte_devmap(pte)) {
776 : : page = pte_page(pte);
777 : : }
778 : :
779 : : out_set_pte:
780 : 22778404 : set_pte_at(dst_mm, addr, dst_pte, pte);
781 : 22777578 : return 0;
782 : : }
783 : :
784 : 7753982 : static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
785 : : pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
786 : : unsigned long addr, unsigned long end)
787 : : {
788 : : pte_t *orig_src_pte, *orig_dst_pte;
789 : : pte_t *src_pte, *dst_pte;
790 : : spinlock_t *src_ptl, *dst_ptl;
791 : : int progress = 0;
792 : : int rss[NR_MM_COUNTERS];
793 : : swp_entry_t entry = (swp_entry_t){0};
794 : :
795 : : again:
796 : : init_rss_vec(rss);
797 : :
798 [ + + + + ]: 23268910 : dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
799 [ + - ]: 7756260 : if (!dst_pte)
800 : : return -ENOMEM;
801 : 15512520 : src_pte = pte_offset_map(src_pmd, addr);
802 : : src_ptl = pte_lockptr(src_mm, src_pmd);
803 : 7756260 : spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
804 : : orig_src_pte = src_pte;
805 : : orig_dst_pte = dst_pte;
806 : : arch_enter_lazy_mmu_mode();
807 : :
808 : : do {
809 : : /*
810 : : * We are holding two locks at this point - either of them
811 : : * could generate latencies in another task on another CPU.
812 : : */
813 [ + + ]: 61135948 : if (progress >= 32) {
814 : : progress = 0;
815 [ + + ]: 4180900 : if (need_resched() ||
816 : : spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
817 : : break;
818 : : }
819 [ + + ]: 61133659 : if (pte_none(*src_pte)) {
820 : 38356381 : progress++;
821 : 38356381 : continue;
822 : : }
823 : 22777278 : entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
824 : : vma, addr, rss);
825 [ + + ]: 22777212 : if (entry.val)
826 : : break;
827 : 22777776 : progress += 8;
828 [ + + ]: 61134157 : } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
829 : :
830 : : arch_leave_lazy_mmu_mode();
831 : : spin_unlock(src_ptl);
832 : : pte_unmap(orig_src_pte);
833 : 7756308 : add_mm_rss_vec(dst_mm, rss);
834 : : pte_unmap_unlock(orig_dst_pte, dst_ptl);
835 : 7756320 : cond_resched();
836 : :
837 [ - + ]: 7756359 : if (entry.val) {
838 [ # # ]: 0 : if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
839 : : return -ENOMEM;
840 : : progress = 0;
841 : : }
842 [ + + ]: 7756248 : if (addr != end)
843 : : goto again;
844 : : return 0;
845 : : }
846 : :
847 : 7900139 : static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
848 : : pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
849 : : unsigned long addr, unsigned long end)
850 : : {
851 : : pmd_t *src_pmd, *dst_pmd;
852 : : unsigned long next;
853 : :
854 : : dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
855 [ + + ]: 7900139 : if (!dst_pmd)
856 : : return -ENOMEM;
857 : : src_pmd = pmd_offset(src_pud, addr);
858 : : do {
859 : : next = pmd_addr_end(addr, end);
860 : : if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd)
861 : : || pmd_devmap(*src_pmd)) {
862 : : int err;
863 : : VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma);
864 : : err = copy_huge_pmd(dst_mm, src_mm,
865 : : dst_pmd, src_pmd, addr, vma);
866 : : if (err == -ENOMEM)
867 : : return -ENOMEM;
868 : : if (!err)
869 : : continue;
870 : : /* fall through */
871 : : }
872 [ + + ]: 7900178 : if (pmd_none_or_clear_bad(src_pmd))
873 : 146240 : continue;
874 [ + + ]: 7753938 : if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
875 : : vma, addr, next))
876 : : return -ENOMEM;
877 : : } while (dst_pmd++, src_pmd++, addr = next, addr != end);
878 : : return 0;
879 : : }
880 : :
881 : : static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
882 : : p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma,
883 : : unsigned long addr, unsigned long end)
884 : : {
885 : : pud_t *src_pud, *dst_pud;
886 : : unsigned long next;
887 : :
888 : : dst_pud = pud_alloc(dst_mm, dst_p4d, addr);
889 [ + + ]: 7900259 : if (!dst_pud)
890 : : return -ENOMEM;
891 : : src_pud = pud_offset(src_p4d, addr);
892 : : do {
893 : : next = pud_addr_end(addr, end);
894 : : if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) {
895 : : int err;
896 : :
897 : : VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma);
898 : : err = copy_huge_pud(dst_mm, src_mm,
899 : : dst_pud, src_pud, addr, vma);
900 : : if (err == -ENOMEM)
901 : : return -ENOMEM;
902 : : if (!err)
903 : : continue;
904 : : /* fall through */
905 : : }
906 : : if (pud_none_or_clear_bad(src_pud))
907 : : continue;
908 [ + + ]: 7900197 : if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
909 : : vma, addr, next))
910 : : return -ENOMEM;
911 : : } while (dst_pud++, src_pud++, addr = next, addr != end);
912 : : return 0;
913 : : }
914 : :
915 : 7900314 : static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
916 : : pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
917 : : unsigned long addr, unsigned long end)
918 : : {
919 : : p4d_t *src_p4d, *dst_p4d;
920 : : unsigned long next;
921 : :
922 : : dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr);
923 [ + + ]: 7900314 : if (!dst_p4d)
924 : : return -ENOMEM;
925 : : src_p4d = p4d_offset(src_pgd, addr);
926 : : do {
927 : : next = p4d_addr_end(addr, end);
928 : : if (p4d_none_or_clear_bad(src_p4d))
929 : : continue;
930 [ + + ]: 7900357 : if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d,
931 : : vma, addr, next))
932 : : return -ENOMEM;
933 : : } while (dst_p4d++, src_p4d++, addr = next, addr != end);
934 : : return 0;
935 : : }
936 : :
937 : 14829402 : int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
938 : : struct vm_area_struct *vma)
939 : : {
940 : : pgd_t *src_pgd, *dst_pgd;
941 : : unsigned long next;
942 : 14829402 : unsigned long addr = vma->vm_start;
943 : 14829402 : unsigned long end = vma->vm_end;
944 : : struct mmu_notifier_range range;
945 : : bool is_cow;
946 : : int ret;
947 : :
948 : : /*
949 : : * Don't copy ptes where a page fault will fill them correctly.
950 : : * Fork becomes much lighter when there are big shared or private
951 : : * readonly mappings. The tradeoff is that copy_page_range is more
952 : : * efficient than faulting.
953 : : */
954 [ + + + + ]: 29658398 : if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
955 : 14828996 : !vma->anon_vma)
956 : : return 0;
957 : :
958 : : if (is_vm_hugetlb_page(vma))
959 : : return copy_hugetlb_page_range(dst_mm, src_mm, vma);
960 : :
961 : : if (unlikely(vma->vm_flags & VM_PFNMAP)) {
962 : : /*
963 : : * We do not free on error cases below as remove_vma
964 : : * gets called on error from higher level routine
965 : : */
966 : : ret = track_pfn_copy(vma);
967 : : if (ret)
968 : : return ret;
969 : : }
970 : :
971 : : /*
972 : : * We need to invalidate the secondary MMU mappings only when
973 : : * there could be a permission downgrade on the ptes of the
974 : : * parent mm. And a permission downgrade will only happen if
975 : : * is_cow_mapping() returns true.
976 : : */
977 : : is_cow = is_cow_mapping(vma->vm_flags);
978 : :
979 : : if (is_cow) {
980 : : mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE,
981 : : 0, vma, src_mm, addr, end);
982 : : mmu_notifier_invalidate_range_start(&range);
983 : : }
984 : :
985 : : ret = 0;
986 : 7674505 : dst_pgd = pgd_offset(dst_mm, addr);
987 : 7674505 : src_pgd = pgd_offset(src_mm, addr);
988 : : do {
989 [ + + ]: 7900351 : next = pgd_addr_end(addr, end);
990 : : if (pgd_none_or_clear_bad(src_pgd))
991 : : continue;
992 [ + + ]: 7900351 : if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd,
993 : : vma, addr, next))) {
994 : : ret = -ENOMEM;
995 : : break;
996 : : }
997 [ + + ]: 7900169 : } while (dst_pgd++, src_pgd++, addr = next, addr != end);
998 : :
999 : : if (is_cow)
1000 : : mmu_notifier_invalidate_range_end(&range);
1001 : 7674348 : return ret;
1002 : : }
1003 : :
1004 : 25127918 : static unsigned long zap_pte_range(struct mmu_gather *tlb,
1005 : : struct vm_area_struct *vma, pmd_t *pmd,
1006 : : unsigned long addr, unsigned long end,
1007 : : struct zap_details *details)
1008 : : {
1009 : 25127918 : struct mm_struct *mm = tlb->mm;
1010 : : int force_flush = 0;
1011 : : int rss[NR_MM_COUNTERS];
1012 : : spinlock_t *ptl;
1013 : : pte_t *start_pte;
1014 : : pte_t *pte;
1015 : : swp_entry_t entry;
1016 : :
1017 : : tlb_change_page_size(tlb, PAGE_SIZE);
1018 : : again:
1019 : : init_rss_vec(rss);
1020 : 25151497 : start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1021 : : pte = start_pte;
1022 : : flush_tlb_batched_pending(mm);
1023 : : arch_enter_lazy_mmu_mode();
1024 : : do {
1025 : 725816707 : pte_t ptent = *pte;
1026 [ + + ]: 725816707 : if (pte_none(ptent))
1027 : 538447225 : continue;
1028 : :
1029 [ + + ]: 187395037 : if (need_resched())
1030 : : break;
1031 : :
1032 [ + - ]: 187367199 : if (pte_present(ptent)) {
1033 : : struct page *page;
1034 : :
1035 : 187367199 : page = vm_normal_page(vma, addr, ptent);
1036 [ - + # # ]: 187364647 : if (unlikely(details) && page) {
1037 : : /*
1038 : : * unmap_shared_mapping_pages() wants to
1039 : : * invalidate cache without truncating:
1040 : : * unmap shared but keep private pages.
1041 : : */
1042 [ # # # # ]: 0 : if (details->check_mapping &&
1043 : 0 : details->check_mapping != page_rmapping(page))
1044 : 0 : continue;
1045 : : }
1046 : : ptent = ptep_get_and_clear_full(mm, addr, pte,
1047 : : tlb->fullmm);
1048 : 187247303 : tlb_remove_tlb_entry(tlb, pte, addr);
1049 [ + + ]: 187247303 : if (unlikely(!page))
1050 : 1611075 : continue;
1051 : :
1052 [ + + ]: 185729066 : if (!PageAnon(page)) {
1053 [ + + ]: 152193163 : if (pte_dirty(ptent)) {
1054 : : force_flush = 1;
1055 : 839726 : set_page_dirty(page);
1056 : : }
1057 [ + + + + ]: 303979004 : if (pte_young(ptent) &&
1058 : 151790327 : likely(!(vma->vm_flags & VM_SEQ_READ)))
1059 : 151790928 : mark_page_accessed(page);
1060 : : }
1061 : 185715814 : rss[mm_counter(page)]--;
1062 : 185745139 : page_remove_rmap(page, false);
1063 [ - + ]: 185721548 : if (unlikely(page_mapcount(page) < 0))
1064 : 0 : print_bad_pte(vma, addr, ptent, page);
1065 [ - + ]: 185763088 : if (unlikely(__tlb_remove_page(tlb, page))) {
1066 : : force_flush = 1;
1067 : 0 : addr += PAGE_SIZE;
1068 : 0 : break;
1069 : : }
1070 : 185763088 : continue;
1071 : : }
1072 : :
1073 : : entry = pte_to_swp_entry(ptent);
1074 : : if (non_swap_entry(entry) && is_device_private_entry(entry)) {
1075 : : struct page *page = device_private_entry_to_page(entry);
1076 : :
1077 : : if (unlikely(details && details->check_mapping)) {
1078 : : /*
1079 : : * unmap_shared_mapping_pages() wants to
1080 : : * invalidate cache without truncating:
1081 : : * unmap shared but keep private pages.
1082 : : */
1083 : : if (details->check_mapping !=
1084 : : page_rmapping(page))
1085 : : continue;
1086 : : }
1087 : :
1088 : : pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1089 : : rss[mm_counter(page)]--;
1090 : : page_remove_rmap(page, false);
1091 : : put_page(page);
1092 : : continue;
1093 : : }
1094 : :
1095 : : /* If details->check_mapping, we leave swap entries. */
1096 [ # # ]: 0 : if (unlikely(details))
1097 : 0 : continue;
1098 : :
1099 [ # # ]: 0 : if (!non_swap_entry(entry))
1100 : 0 : rss[MM_SWAPENTS]--;
1101 [ # # ]: 0 : else if (is_migration_entry(entry)) {
1102 : : struct page *page;
1103 : :
1104 : 0 : page = migration_entry_to_page(entry);
1105 : 0 : rss[mm_counter(page)]--;
1106 : : }
1107 [ # # ]: 0 : if (unlikely(!free_swap_and_cache(entry)))
1108 : 0 : print_bad_pte(vma, addr, ptent, NULL);
1109 : : pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1110 [ + + ]: 725335053 : } while (pte++, addr += PAGE_SIZE, addr != end);
1111 : :
1112 : 25161393 : add_mm_rss_vec(mm, rss);
1113 : : arch_leave_lazy_mmu_mode();
1114 : :
1115 : : /* Do the actual TLB flush before dropping ptl */
1116 [ + + ]: 25154444 : if (force_flush)
1117 : 2367 : tlb_flush_mmu_tlbonly(tlb);
1118 : : pte_unmap_unlock(start_pte, ptl);
1119 : :
1120 : : /*
1121 : : * If we forced a TLB flush (either due to running out of
1122 : : * batch buffers or because we needed to flush dirty TLB
1123 : : * entries before releasing the ptl), free the batched
1124 : : * memory too. Restart if we didn't do everything.
1125 : : */
1126 [ + + ]: 25153911 : if (force_flush) {
1127 : : force_flush = 0;
1128 : 2367 : tlb_flush_mmu(tlb);
1129 : : }
1130 : :
1131 [ + + ]: 25152004 : if (addr != end) {
1132 : 23578 : cond_resched();
1133 : 23579 : goto again;
1134 : : }
1135 : :
1136 : 25128426 : return addr;
1137 : : }
1138 : :
1139 : 26502410 : static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
1140 : : struct vm_area_struct *vma, pud_t *pud,
1141 : : unsigned long addr, unsigned long end,
1142 : : struct zap_details *details)
1143 : : {
1144 : : pmd_t *pmd;
1145 : : unsigned long next;
1146 : :
1147 : : pmd = pmd_offset(pud, addr);
1148 : : do {
1149 : : next = pmd_addr_end(addr, end);
1150 : : if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
1151 : : if (next - addr != HPAGE_PMD_SIZE)
1152 : : __split_huge_pmd(vma, pmd, addr, false, NULL);
1153 : : else if (zap_huge_pmd(tlb, vma, pmd, addr))
1154 : : goto next;
1155 : : /* fall through */
1156 : : }
1157 : : /*
1158 : : * Here there can be other concurrent MADV_DONTNEED or
1159 : : * trans huge page faults running, and if the pmd is
1160 : : * none or trans huge it can change under us. This is
1161 : : * because MADV_DONTNEED holds the mmap_sem in read
1162 : : * mode.
1163 : : */
1164 [ + + ]: 26502410 : if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1165 : : goto next;
1166 : 25128944 : next = zap_pte_range(tlb, vma, pmd, addr, next, details);
1167 : : next:
1168 : 26498426 : cond_resched();
1169 [ - + ]: 26498074 : } while (pmd++, addr = next, addr != end);
1170 : :
1171 : 26498074 : return addr;
1172 : : }
1173 : :
1174 : 26500665 : static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
1175 : : struct vm_area_struct *vma, p4d_t *p4d,
1176 : : unsigned long addr, unsigned long end,
1177 : : struct zap_details *details)
1178 : : {
1179 : : pud_t *pud;
1180 : : unsigned long next;
1181 : :
1182 : : pud = pud_offset(p4d, addr);
1183 : : do {
1184 : : next = pud_addr_end(addr, end);
1185 : : if (pud_trans_huge(*pud) || pud_devmap(*pud)) {
1186 : : if (next - addr != HPAGE_PUD_SIZE) {
1187 : : VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma);
1188 : : split_huge_pud(vma, pud, addr);
1189 : : } else if (zap_huge_pud(tlb, vma, pud, addr))
1190 : : goto next;
1191 : : /* fall through */
1192 : : }
1193 : : if (pud_none_or_clear_bad(pud))
1194 : : continue;
1195 : 26500665 : next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1196 : : next:
1197 : 26500608 : cond_resched();
1198 [ - + ]: 26500477 : } while (pud++, addr = next, addr != end);
1199 : :
1200 : 26500477 : return addr;
1201 : : }
1202 : :
1203 : : static inline unsigned long zap_p4d_range(struct mmu_gather *tlb,
1204 : : struct vm_area_struct *vma, pgd_t *pgd,
1205 : : unsigned long addr, unsigned long end,
1206 : : struct zap_details *details)
1207 : : {
1208 : : p4d_t *p4d;
1209 : : unsigned long next;
1210 : :
1211 : : p4d = p4d_offset(pgd, addr);
1212 : : do {
1213 : : next = p4d_addr_end(addr, end);
1214 : : if (p4d_none_or_clear_bad(p4d))
1215 : : continue;
1216 : 26502381 : next = zap_pud_range(tlb, vma, p4d, addr, next, details);
1217 [ + + ]: 26499367 : } while (p4d++, addr = next, addr != end);
1218 : :
1219 : 26498129 : return addr;
1220 : : }
1221 : :
1222 : 24857059 : void unmap_page_range(struct mmu_gather *tlb,
1223 : : struct vm_area_struct *vma,
1224 : : unsigned long addr, unsigned long end,
1225 : : struct zap_details *details)
1226 : : {
1227 : : pgd_t *pgd;
1228 : : unsigned long next;
1229 : :
1230 [ - + ]: 24857059 : BUG_ON(addr >= end);
1231 : 24857059 : tlb_start_vma(tlb, vma);
1232 : 24864171 : pgd = pgd_offset(vma->vm_mm, addr);
1233 : : do {
1234 [ + + ]: 26501143 : next = pgd_addr_end(addr, end);
1235 : : if (pgd_none_or_clear_bad(pgd))
1236 : : continue;
1237 : : next = zap_p4d_range(tlb, vma, pgd, addr, next, details);
1238 [ + + ]: 26498129 : } while (pgd++, addr = next, addr != end);
1239 : : tlb_end_vma(tlb, vma);
1240 : 24860154 : }
1241 : :
1242 : :
1243 : 24862250 : static void unmap_single_vma(struct mmu_gather *tlb,
1244 : : struct vm_area_struct *vma, unsigned long start_addr,
1245 : : unsigned long end_addr,
1246 : : struct zap_details *details)
1247 : : {
1248 : 24862250 : unsigned long start = max(vma->vm_start, start_addr);
1249 : : unsigned long end;
1250 : :
1251 [ + + ]: 24862250 : if (start >= vma->vm_end)
1252 : : return;
1253 : 24862018 : end = min(vma->vm_end, end_addr);
1254 [ + + ]: 24862018 : if (end <= vma->vm_start)
1255 : : return;
1256 : :
1257 : : if (vma->vm_file)
1258 : : uprobe_munmap(vma, start, end);
1259 : :
1260 : : if (unlikely(vma->vm_flags & VM_PFNMAP))
1261 : : untrack_pfn(vma, 0, 0);
1262 : :
1263 [ + + ]: 24862136 : if (start != end) {
1264 : : if (unlikely(is_vm_hugetlb_page(vma))) {
1265 : : /*
1266 : : * It is undesirable to test vma->vm_file as it
1267 : : * should be non-null for valid hugetlb area.
1268 : : * However, vm_file will be NULL in the error
1269 : : * cleanup path of mmap_region. When
1270 : : * hugetlbfs ->mmap method fails,
1271 : : * mmap_region() nullifies vma->vm_file
1272 : : * before calling this function to clean up.
1273 : : * Since no pte has actually been setup, it is
1274 : : * safe to do nothing in this case.
1275 : : */
1276 : : if (vma->vm_file) {
1277 : : i_mmap_lock_write(vma->vm_file->f_mapping);
1278 : : __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
1279 : : i_mmap_unlock_write(vma->vm_file->f_mapping);
1280 : : }
1281 : : } else
1282 : 24861497 : unmap_page_range(tlb, vma, start, end, details);
1283 : : }
1284 : : }
1285 : :
1286 : : /**
1287 : : * unmap_vmas - unmap a range of memory covered by a list of vma's
1288 : : * @tlb: address of the caller's struct mmu_gather
1289 : : * @vma: the starting vma
1290 : : * @start_addr: virtual address at which to start unmapping
1291 : : * @end_addr: virtual address at which to end unmapping
1292 : : *
1293 : : * Unmap all pages in the vma list.
1294 : : *
1295 : : * Only addresses between `start' and `end' will be unmapped.
1296 : : *
1297 : : * The VMA list must be sorted in ascending virtual address order.
1298 : : *
1299 : : * unmap_vmas() assumes that the caller will flush the whole unmapped address
1300 : : * range after unmap_vmas() returns. So the only responsibility here is to
1301 : : * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1302 : : * drops the lock and schedules.
1303 : : */
1304 : 3153509 : void unmap_vmas(struct mmu_gather *tlb,
1305 : : struct vm_area_struct *vma, unsigned long start_addr,
1306 : : unsigned long end_addr)
1307 : : {
1308 : : struct mmu_notifier_range range;
1309 : :
1310 : : mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm,
1311 : : start_addr, end_addr);
1312 : : mmu_notifier_invalidate_range_start(&range);
1313 [ + + + + ]: 24848413 : for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
1314 : 24848983 : unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
1315 : : mmu_notifier_invalidate_range_end(&range);
1316 : 3152939 : }
1317 : :
1318 : : /**
1319 : : * zap_page_range - remove user pages in a given range
1320 : : * @vma: vm_area_struct holding the applicable pages
1321 : : * @start: starting address of pages to zap
1322 : : * @size: number of bytes to zap
1323 : : *
1324 : : * Caller must protect the VMA list
1325 : : */
1326 : 9396 : void zap_page_range(struct vm_area_struct *vma, unsigned long start,
1327 : : unsigned long size)
1328 : : {
1329 : : struct mmu_notifier_range range;
1330 : : struct mmu_gather tlb;
1331 : :
1332 : 9396 : lru_add_drain();
1333 : 9396 : mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm,
1334 : : start, start + size);
1335 : 9396 : tlb_gather_mmu(&tlb, vma->vm_mm, start, range.end);
1336 : 9393 : update_hiwater_rss(vma->vm_mm);
1337 : : mmu_notifier_invalidate_range_start(&range);
1338 [ + + + + ]: 9396 : for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next)
1339 : 9395 : unmap_single_vma(&tlb, vma, start, range.end, NULL);
1340 : : mmu_notifier_invalidate_range_end(&range);
1341 : 9394 : tlb_finish_mmu(&tlb, start, range.end);
1342 : 9396 : }
1343 : :
1344 : : /**
1345 : : * zap_page_range_single - remove user pages in a given range
1346 : : * @vma: vm_area_struct holding the applicable pages
1347 : : * @address: starting address of pages to zap
1348 : : * @size: number of bytes to zap
1349 : : * @details: details of shared cache invalidation
1350 : : *
1351 : : * The range must fit into one VMA.
1352 : : */
1353 : 0 : static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1354 : : unsigned long size, struct zap_details *details)
1355 : : {
1356 : : struct mmu_notifier_range range;
1357 : : struct mmu_gather tlb;
1358 : :
1359 : 0 : lru_add_drain();
1360 : 0 : mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm,
1361 : : address, address + size);
1362 : 0 : tlb_gather_mmu(&tlb, vma->vm_mm, address, range.end);
1363 : 0 : update_hiwater_rss(vma->vm_mm);
1364 : : mmu_notifier_invalidate_range_start(&range);
1365 : 0 : unmap_single_vma(&tlb, vma, address, range.end, details);
1366 : : mmu_notifier_invalidate_range_end(&range);
1367 : 0 : tlb_finish_mmu(&tlb, address, range.end);
1368 : 0 : }
1369 : :
1370 : : /**
1371 : : * zap_vma_ptes - remove ptes mapping the vma
1372 : : * @vma: vm_area_struct holding ptes to be zapped
1373 : : * @address: starting address of pages to zap
1374 : : * @size: number of bytes to zap
1375 : : *
1376 : : * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1377 : : *
1378 : : * The entire address range must be fully contained within the vma.
1379 : : *
1380 : : */
1381 : 0 : void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1382 : : unsigned long size)
1383 : : {
1384 [ # # # # : 0 : if (address < vma->vm_start || address + size > vma->vm_end ||
# # ]
1385 : 0 : !(vma->vm_flags & VM_PFNMAP))
1386 : 0 : return;
1387 : :
1388 : 0 : zap_page_range_single(vma, address, size, NULL);
1389 : : }
1390 : : EXPORT_SYMBOL_GPL(zap_vma_ptes);
1391 : :
1392 : 828 : pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
1393 : : spinlock_t **ptl)
1394 : : {
1395 : : pgd_t *pgd;
1396 : : p4d_t *p4d;
1397 : : pud_t *pud;
1398 : : pmd_t *pmd;
1399 : :
1400 : 828 : pgd = pgd_offset(mm, addr);
1401 : : p4d = p4d_alloc(mm, pgd, addr);
1402 [ + - ]: 828 : if (!p4d)
1403 : : return NULL;
1404 : : pud = pud_alloc(mm, p4d, addr);
1405 [ + - ]: 828 : if (!pud)
1406 : : return NULL;
1407 : : pmd = pmd_alloc(mm, pud, addr);
1408 [ + - ]: 828 : if (!pmd)
1409 : : return NULL;
1410 : :
1411 : : VM_BUG_ON(pmd_trans_huge(*pmd));
1412 [ - + # # ]: 2484 : return pte_alloc_map_lock(mm, pmd, addr, ptl);
1413 : : }
1414 : :
1415 : : /*
1416 : : * This is the old fallback for page remapping.
1417 : : *
1418 : : * For historical reasons, it only allows reserved pages. Only
1419 : : * old drivers should use this, and they needed to mark their
1420 : : * pages reserved for the old functions anyway.
1421 : : */
1422 : 0 : static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1423 : : struct page *page, pgprot_t prot)
1424 : : {
1425 : 0 : struct mm_struct *mm = vma->vm_mm;
1426 : : int retval;
1427 : : pte_t *pte;
1428 : : spinlock_t *ptl;
1429 : :
1430 : : retval = -EINVAL;
1431 [ # # # # : 0 : if (PageAnon(page) || PageSlab(page) || page_has_type(page))
# # ]
1432 : : goto out;
1433 : : retval = -ENOMEM;
1434 : 0 : flush_dcache_page(page);
1435 : : pte = get_locked_pte(mm, addr, &ptl);
1436 [ # # ]: 0 : if (!pte)
1437 : : goto out;
1438 : : retval = -EBUSY;
1439 [ # # ]: 0 : if (!pte_none(*pte))
1440 : : goto out_unlock;
1441 : :
1442 : : /* Ok, finally just insert the thing.. */
1443 : 0 : get_page(page);
1444 : 0 : inc_mm_counter_fast(mm, mm_counter_file(page));
1445 : 0 : page_add_file_rmap(page, false);
1446 : 0 : set_pte_at(mm, addr, pte, mk_pte(page, prot));
1447 : :
1448 : : retval = 0;
1449 : : out_unlock:
1450 : 0 : pte_unmap_unlock(pte, ptl);
1451 : : out:
1452 : 0 : return retval;
1453 : : }
1454 : :
1455 : : /**
1456 : : * vm_insert_page - insert single page into user vma
1457 : : * @vma: user vma to map to
1458 : : * @addr: target user address of this page
1459 : : * @page: source kernel page
1460 : : *
1461 : : * This allows drivers to insert individual pages they've allocated
1462 : : * into a user vma.
1463 : : *
1464 : : * The page has to be a nice clean _individual_ kernel allocation.
1465 : : * If you allocate a compound page, you need to have marked it as
1466 : : * such (__GFP_COMP), or manually just split the page up yourself
1467 : : * (see split_page()).
1468 : : *
1469 : : * NOTE! Traditionally this was done with "remap_pfn_range()" which
1470 : : * took an arbitrary page protection parameter. This doesn't allow
1471 : : * that. Your vma protection will have to be set up correctly, which
1472 : : * means that if you want a shared writable mapping, you'd better
1473 : : * ask for a shared writable mapping!
1474 : : *
1475 : : * The page does not need to be reserved.
1476 : : *
1477 : : * Usually this function is called from f_op->mmap() handler
1478 : : * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1479 : : * Caller must set VM_MIXEDMAP on vma if it wants to call this
1480 : : * function from other places, for example from page-fault handler.
1481 : : *
1482 : : * Return: %0 on success, negative error code otherwise.
1483 : : */
1484 : 0 : int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1485 : : struct page *page)
1486 : : {
1487 [ # # # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
1488 : : return -EFAULT;
1489 [ # # ]: 0 : if (!page_count(page))
1490 : : return -EINVAL;
1491 [ # # ]: 0 : if (!(vma->vm_flags & VM_MIXEDMAP)) {
1492 [ # # ]: 0 : BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1493 [ # # ]: 0 : BUG_ON(vma->vm_flags & VM_PFNMAP);
1494 : 0 : vma->vm_flags |= VM_MIXEDMAP;
1495 : : }
1496 : 0 : return insert_page(vma, addr, page, vma->vm_page_prot);
1497 : : }
1498 : : EXPORT_SYMBOL(vm_insert_page);
1499 : :
1500 : : /*
1501 : : * __vm_map_pages - maps range of kernel pages into user vma
1502 : : * @vma: user vma to map to
1503 : : * @pages: pointer to array of source kernel pages
1504 : : * @num: number of pages in page array
1505 : : * @offset: user's requested vm_pgoff
1506 : : *
1507 : : * This allows drivers to map range of kernel pages into a user vma.
1508 : : *
1509 : : * Return: 0 on success and error code otherwise.
1510 : : */
1511 : 0 : static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages,
1512 : : unsigned long num, unsigned long offset)
1513 : : {
1514 : : unsigned long count = vma_pages(vma);
1515 : : unsigned long uaddr = vma->vm_start;
1516 : : int ret, i;
1517 : :
1518 : : /* Fail if the user requested offset is beyond the end of the object */
1519 [ # # ]: 0 : if (offset >= num)
1520 : : return -ENXIO;
1521 : :
1522 : : /* Fail if the user requested size exceeds available object size */
1523 [ # # ]: 0 : if (count > num - offset)
1524 : : return -ENXIO;
1525 : :
1526 [ # # ]: 0 : for (i = 0; i < count; i++) {
1527 : 0 : ret = vm_insert_page(vma, uaddr, pages[offset + i]);
1528 [ # # ]: 0 : if (ret < 0)
1529 : 0 : return ret;
1530 : 0 : uaddr += PAGE_SIZE;
1531 : : }
1532 : :
1533 : : return 0;
1534 : : }
1535 : :
1536 : : /**
1537 : : * vm_map_pages - maps range of kernel pages starts with non zero offset
1538 : : * @vma: user vma to map to
1539 : : * @pages: pointer to array of source kernel pages
1540 : : * @num: number of pages in page array
1541 : : *
1542 : : * Maps an object consisting of @num pages, catering for the user's
1543 : : * requested vm_pgoff
1544 : : *
1545 : : * If we fail to insert any page into the vma, the function will return
1546 : : * immediately leaving any previously inserted pages present. Callers
1547 : : * from the mmap handler may immediately return the error as their caller
1548 : : * will destroy the vma, removing any successfully inserted pages. Other
1549 : : * callers should make their own arrangements for calling unmap_region().
1550 : : *
1551 : : * Context: Process context. Called by mmap handlers.
1552 : : * Return: 0 on success and error code otherwise.
1553 : : */
1554 : 0 : int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
1555 : : unsigned long num)
1556 : : {
1557 : 0 : return __vm_map_pages(vma, pages, num, vma->vm_pgoff);
1558 : : }
1559 : : EXPORT_SYMBOL(vm_map_pages);
1560 : :
1561 : : /**
1562 : : * vm_map_pages_zero - map range of kernel pages starts with zero offset
1563 : : * @vma: user vma to map to
1564 : : * @pages: pointer to array of source kernel pages
1565 : : * @num: number of pages in page array
1566 : : *
1567 : : * Similar to vm_map_pages(), except that it explicitly sets the offset
1568 : : * to 0. This function is intended for the drivers that did not consider
1569 : : * vm_pgoff.
1570 : : *
1571 : : * Context: Process context. Called by mmap handlers.
1572 : : * Return: 0 on success and error code otherwise.
1573 : : */
1574 : 0 : int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
1575 : : unsigned long num)
1576 : : {
1577 : 0 : return __vm_map_pages(vma, pages, num, 0);
1578 : : }
1579 : : EXPORT_SYMBOL(vm_map_pages_zero);
1580 : :
1581 : 0 : static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1582 : : pfn_t pfn, pgprot_t prot, bool mkwrite)
1583 : : {
1584 : 0 : struct mm_struct *mm = vma->vm_mm;
1585 : : pte_t *pte, entry;
1586 : : spinlock_t *ptl;
1587 : :
1588 : : pte = get_locked_pte(mm, addr, &ptl);
1589 [ # # ]: 0 : if (!pte)
1590 : : return VM_FAULT_OOM;
1591 [ # # ]: 0 : if (!pte_none(*pte)) {
1592 [ # # ]: 0 : if (mkwrite) {
1593 : : /*
1594 : : * For read faults on private mappings the PFN passed
1595 : : * in may not match the PFN we have mapped if the
1596 : : * mapped PFN is a writeable COW page. In the mkwrite
1597 : : * case we are creating a writable PTE for a shared
1598 : : * mapping and we expect the PFNs to match. If they
1599 : : * don't match, we are likely racing with block
1600 : : * allocation and mapping invalidation so just skip the
1601 : : * update.
1602 : : */
1603 [ # # ]: 0 : if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) {
1604 [ # # # # ]: 0 : WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte)));
1605 : : goto out_unlock;
1606 : : }
1607 : : entry = pte_mkyoung(*pte);
1608 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1609 : 0 : if (ptep_set_access_flags(vma, addr, pte, entry, 1))
1610 : : update_mmu_cache(vma, addr, pte);
1611 : : }
1612 : : goto out_unlock;
1613 : : }
1614 : :
1615 : : /* Ok, finally just insert the thing.. */
1616 : : if (pfn_t_devmap(pfn))
1617 : : entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
1618 : : else
1619 : : entry = pte_mkspecial(pfn_t_pte(pfn, prot));
1620 : :
1621 [ # # ]: 0 : if (mkwrite) {
1622 : : entry = pte_mkyoung(entry);
1623 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1624 : : }
1625 : :
1626 : 0 : set_pte_at(mm, addr, pte, entry);
1627 : : update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
1628 : :
1629 : : out_unlock:
1630 : 0 : pte_unmap_unlock(pte, ptl);
1631 : 0 : return VM_FAULT_NOPAGE;
1632 : : }
1633 : :
1634 : : /**
1635 : : * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot
1636 : : * @vma: user vma to map to
1637 : : * @addr: target user address of this page
1638 : : * @pfn: source kernel pfn
1639 : : * @pgprot: pgprot flags for the inserted page
1640 : : *
1641 : : * This is exactly like vmf_insert_pfn(), except that it allows drivers to
1642 : : * to override pgprot on a per-page basis.
1643 : : *
1644 : : * This only makes sense for IO mappings, and it makes no sense for
1645 : : * COW mappings. In general, using multiple vmas is preferable;
1646 : : * vmf_insert_pfn_prot should only be used if using multiple VMAs is
1647 : : * impractical.
1648 : : *
1649 : : * Context: Process context. May allocate using %GFP_KERNEL.
1650 : : * Return: vm_fault_t value.
1651 : : */
1652 : 0 : vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
1653 : : unsigned long pfn, pgprot_t pgprot)
1654 : : {
1655 : : /*
1656 : : * Technically, architectures with pte_special can avoid all these
1657 : : * restrictions (same for remap_pfn_range). However we would like
1658 : : * consistency in testing and feature parity among all, so we should
1659 : : * try to keep these invariants in place for everybody.
1660 : : */
1661 [ # # ]: 0 : BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1662 [ # # ]: 0 : BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1663 : : (VM_PFNMAP|VM_MIXEDMAP));
1664 [ # # # # ]: 0 : BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1665 [ # # # # ]: 0 : BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
1666 : :
1667 [ # # # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
1668 : : return VM_FAULT_SIGBUS;
1669 : :
1670 : : if (!pfn_modify_allowed(pfn, pgprot))
1671 : : return VM_FAULT_SIGBUS;
1672 : :
1673 : : track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV));
1674 : :
1675 : 0 : return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot,
1676 : : false);
1677 : : }
1678 : : EXPORT_SYMBOL(vmf_insert_pfn_prot);
1679 : :
1680 : : /**
1681 : : * vmf_insert_pfn - insert single pfn into user vma
1682 : : * @vma: user vma to map to
1683 : : * @addr: target user address of this page
1684 : : * @pfn: source kernel pfn
1685 : : *
1686 : : * Similar to vm_insert_page, this allows drivers to insert individual pages
1687 : : * they've allocated into a user vma. Same comments apply.
1688 : : *
1689 : : * This function should only be called from a vm_ops->fault handler, and
1690 : : * in that case the handler should return the result of this function.
1691 : : *
1692 : : * vma cannot be a COW mapping.
1693 : : *
1694 : : * As this is called only for pages that do not currently exist, we
1695 : : * do not need to flush old virtual caches or the TLB.
1696 : : *
1697 : : * Context: Process context. May allocate using %GFP_KERNEL.
1698 : : * Return: vm_fault_t value.
1699 : : */
1700 : 0 : vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1701 : : unsigned long pfn)
1702 : : {
1703 : 0 : return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
1704 : : }
1705 : : EXPORT_SYMBOL(vmf_insert_pfn);
1706 : :
1707 : : static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn)
1708 : : {
1709 : : /* these checks mirror the abort conditions in vm_normal_page */
1710 [ # # ]: 0 : if (vma->vm_flags & VM_MIXEDMAP)
1711 : : return true;
1712 : : if (pfn_t_devmap(pfn))
1713 : : return true;
1714 : : if (pfn_t_special(pfn))
1715 : : return true;
1716 [ # # ]: 0 : if (is_zero_pfn(pfn_t_to_pfn(pfn)))
1717 : : return true;
1718 : : return false;
1719 : : }
1720 : :
1721 : 0 : static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma,
1722 : : unsigned long addr, pfn_t pfn, bool mkwrite)
1723 : : {
1724 : 0 : pgprot_t pgprot = vma->vm_page_prot;
1725 : : int err;
1726 : :
1727 [ # # ]: 0 : BUG_ON(!vm_mixed_ok(vma, pfn));
1728 : :
1729 [ # # # # ]: 0 : if (addr < vma->vm_start || addr >= vma->vm_end)
1730 : : return VM_FAULT_SIGBUS;
1731 : :
1732 : : track_pfn_insert(vma, &pgprot, pfn);
1733 : :
1734 : : if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot))
1735 : : return VM_FAULT_SIGBUS;
1736 : :
1737 : : /*
1738 : : * If we don't have pte special, then we have to use the pfn_valid()
1739 : : * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1740 : : * refcount the page if pfn_valid is true (hence insert_page rather
1741 : : * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1742 : : * without pte special, it would there be refcounted as a normal page.
1743 : : */
1744 : : if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) &&
1745 [ # # ]: 0 : !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
1746 : : struct page *page;
1747 : :
1748 : : /*
1749 : : * At this point we are committed to insert_page()
1750 : : * regardless of whether the caller specified flags that
1751 : : * result in pfn_t_has_page() == false.
1752 : : */
1753 : 0 : page = pfn_to_page(pfn_t_to_pfn(pfn));
1754 : 0 : err = insert_page(vma, addr, page, pgprot);
1755 : : } else {
1756 : 0 : return insert_pfn(vma, addr, pfn, pgprot, mkwrite);
1757 : : }
1758 : :
1759 [ # # ]: 0 : if (err == -ENOMEM)
1760 : : return VM_FAULT_OOM;
1761 [ # # ]: 0 : if (err < 0 && err != -EBUSY)
1762 : : return VM_FAULT_SIGBUS;
1763 : :
1764 : 0 : return VM_FAULT_NOPAGE;
1765 : : }
1766 : :
1767 : 0 : vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1768 : : pfn_t pfn)
1769 : : {
1770 : 0 : return __vm_insert_mixed(vma, addr, pfn, false);
1771 : : }
1772 : : EXPORT_SYMBOL(vmf_insert_mixed);
1773 : :
1774 : : /*
1775 : : * If the insertion of PTE failed because someone else already added a
1776 : : * different entry in the mean time, we treat that as success as we assume
1777 : : * the same entry was actually inserted.
1778 : : */
1779 : 0 : vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
1780 : : unsigned long addr, pfn_t pfn)
1781 : : {
1782 : 0 : return __vm_insert_mixed(vma, addr, pfn, true);
1783 : : }
1784 : : EXPORT_SYMBOL(vmf_insert_mixed_mkwrite);
1785 : :
1786 : : /*
1787 : : * maps a range of physical memory into the requested pages. the old
1788 : : * mappings are removed. any references to nonexistent pages results
1789 : : * in null mappings (currently treated as "copy-on-access")
1790 : : */
1791 : 2044 : static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1792 : : unsigned long addr, unsigned long end,
1793 : : unsigned long pfn, pgprot_t prot)
1794 : : {
1795 : : pte_t *pte;
1796 : : spinlock_t *ptl;
1797 : : int err = 0;
1798 : :
1799 [ + + + - ]: 6132 : pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1800 [ + - ]: 2044 : if (!pte)
1801 : : return -ENOMEM;
1802 : : arch_enter_lazy_mmu_mode();
1803 : : do {
1804 [ - + ]: 838350 : BUG_ON(!pte_none(*pte));
1805 : : if (!pfn_modify_allowed(pfn, prot)) {
1806 : : err = -EACCES;
1807 : : break;
1808 : : }
1809 : 838350 : set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
1810 : 838350 : pfn++;
1811 [ + + ]: 838350 : } while (pte++, addr += PAGE_SIZE, addr != end);
1812 : : arch_leave_lazy_mmu_mode();
1813 : : pte_unmap_unlock(pte - 1, ptl);
1814 : 2044 : return err;
1815 : : }
1816 : :
1817 : 2044 : static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1818 : : unsigned long addr, unsigned long end,
1819 : : unsigned long pfn, pgprot_t prot)
1820 : : {
1821 : : pmd_t *pmd;
1822 : : unsigned long next;
1823 : : int err;
1824 : :
1825 : 2044 : pfn -= addr >> PAGE_SHIFT;
1826 : : pmd = pmd_alloc(mm, pud, addr);
1827 [ + - ]: 2044 : if (!pmd)
1828 : : return -ENOMEM;
1829 : : VM_BUG_ON(pmd_trans_huge(*pmd));
1830 : : do {
1831 : : next = pmd_addr_end(addr, end);
1832 : 2044 : err = remap_pte_range(mm, pmd, addr, next,
1833 : 2044 : pfn + (addr >> PAGE_SHIFT), prot);
1834 [ - + ]: 2044 : if (err)
1835 : 0 : return err;
1836 : : } while (pmd++, addr = next, addr != end);
1837 : : return 0;
1838 : : }
1839 : :
1840 : 2044 : static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d,
1841 : : unsigned long addr, unsigned long end,
1842 : : unsigned long pfn, pgprot_t prot)
1843 : : {
1844 : : pud_t *pud;
1845 : : unsigned long next;
1846 : : int err;
1847 : :
1848 : 2044 : pfn -= addr >> PAGE_SHIFT;
1849 : : pud = pud_alloc(mm, p4d, addr);
1850 [ + - ]: 2044 : if (!pud)
1851 : : return -ENOMEM;
1852 : : do {
1853 : : next = pud_addr_end(addr, end);
1854 : 2044 : err = remap_pmd_range(mm, pud, addr, next,
1855 : 2044 : pfn + (addr >> PAGE_SHIFT), prot);
1856 [ - + ]: 2044 : if (err)
1857 : 0 : return err;
1858 : : } while (pud++, addr = next, addr != end);
1859 : : return 0;
1860 : : }
1861 : :
1862 : 2044 : static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd,
1863 : : unsigned long addr, unsigned long end,
1864 : : unsigned long pfn, pgprot_t prot)
1865 : : {
1866 : : p4d_t *p4d;
1867 : : unsigned long next;
1868 : : int err;
1869 : :
1870 : 2044 : pfn -= addr >> PAGE_SHIFT;
1871 : : p4d = p4d_alloc(mm, pgd, addr);
1872 [ + - ]: 2044 : if (!p4d)
1873 : : return -ENOMEM;
1874 : : do {
1875 : : next = p4d_addr_end(addr, end);
1876 : 2044 : err = remap_pud_range(mm, p4d, addr, next,
1877 : 2044 : pfn + (addr >> PAGE_SHIFT), prot);
1878 [ - + ]: 2044 : if (err)
1879 : 0 : return err;
1880 : : } while (p4d++, addr = next, addr != end);
1881 : : return 0;
1882 : : }
1883 : :
1884 : : /**
1885 : : * remap_pfn_range - remap kernel memory to userspace
1886 : : * @vma: user vma to map to
1887 : : * @addr: target user address to start at
1888 : : * @pfn: physical address of kernel memory
1889 : : * @size: size of map area
1890 : : * @prot: page protection flags for this mapping
1891 : : *
1892 : : * Note: this is only safe if the mm semaphore is held when called.
1893 : : *
1894 : : * Return: %0 on success, negative error code otherwise.
1895 : : */
1896 : 414 : int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1897 : : unsigned long pfn, unsigned long size, pgprot_t prot)
1898 : : {
1899 : : pgd_t *pgd;
1900 : : unsigned long next;
1901 : 414 : unsigned long end = addr + PAGE_ALIGN(size);
1902 : 414 : struct mm_struct *mm = vma->vm_mm;
1903 : : unsigned long remap_pfn = pfn;
1904 : : int err;
1905 : :
1906 : : /*
1907 : : * Physically remapped pages are special. Tell the
1908 : : * rest of the world about it:
1909 : : * VM_IO tells people not to look at these pages
1910 : : * (accesses can have side effects).
1911 : : * VM_PFNMAP tells the core MM that the base pages are just
1912 : : * raw PFN mappings, and do not have a "struct page" associated
1913 : : * with them.
1914 : : * VM_DONTEXPAND
1915 : : * Disable vma merging and expanding with mremap().
1916 : : * VM_DONTDUMP
1917 : : * Omit vma from core dump, even when VM_IO turned off.
1918 : : *
1919 : : * There's a horrible special case to handle copy-on-write
1920 : : * behaviour that some programs depend on. We mark the "original"
1921 : : * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1922 : : * See vm_normal_page() for details.
1923 : : */
1924 [ - + ]: 828 : if (is_cow_mapping(vma->vm_flags)) {
1925 [ # # # # ]: 0 : if (addr != vma->vm_start || end != vma->vm_end)
1926 : : return -EINVAL;
1927 : 0 : vma->vm_pgoff = pfn;
1928 : : }
1929 : :
1930 : : err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size));
1931 : : if (err)
1932 : : return -EINVAL;
1933 : :
1934 : 414 : vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1935 : :
1936 [ - + ]: 414 : BUG_ON(addr >= end);
1937 : 414 : pfn -= addr >> PAGE_SHIFT;
1938 : 414 : pgd = pgd_offset(mm, addr);
1939 : 414 : flush_cache_range(vma, addr, end);
1940 : : do {
1941 [ + + ]: 2044 : next = pgd_addr_end(addr, end);
1942 : 2044 : err = remap_p4d_range(mm, pgd, addr, next,
1943 : 2044 : pfn + (addr >> PAGE_SHIFT), prot);
1944 [ + - ]: 2044 : if (err)
1945 : : break;
1946 [ + + ]: 2044 : } while (pgd++, addr = next, addr != end);
1947 : :
1948 : : if (err)
1949 : : untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size));
1950 : :
1951 : 414 : return err;
1952 : : }
1953 : : EXPORT_SYMBOL(remap_pfn_range);
1954 : :
1955 : : /**
1956 : : * vm_iomap_memory - remap memory to userspace
1957 : : * @vma: user vma to map to
1958 : : * @start: start of area
1959 : : * @len: size of area
1960 : : *
1961 : : * This is a simplified io_remap_pfn_range() for common driver use. The
1962 : : * driver just needs to give us the physical memory range to be mapped,
1963 : : * we'll figure out the rest from the vma information.
1964 : : *
1965 : : * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1966 : : * whatever write-combining details or similar.
1967 : : *
1968 : : * Return: %0 on success, negative error code otherwise.
1969 : : */
1970 : 414 : int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1971 : : {
1972 : : unsigned long vm_len, pfn, pages;
1973 : :
1974 : : /* Check that the physical memory area passed in looks valid */
1975 [ + - ]: 414 : if (start + len < start)
1976 : : return -EINVAL;
1977 : : /*
1978 : : * You *really* shouldn't map things that aren't page-aligned,
1979 : : * but we've historically allowed it because IO memory might
1980 : : * just have smaller alignment.
1981 : : */
1982 : 414 : len += start & ~PAGE_MASK;
1983 : 414 : pfn = start >> PAGE_SHIFT;
1984 : 414 : pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
1985 [ + - ]: 414 : if (pfn + pages < pfn)
1986 : : return -EINVAL;
1987 : :
1988 : : /* We start the mapping 'vm_pgoff' pages into the area */
1989 [ + - ]: 414 : if (vma->vm_pgoff > pages)
1990 : : return -EINVAL;
1991 : 414 : pfn += vma->vm_pgoff;
1992 : 414 : pages -= vma->vm_pgoff;
1993 : :
1994 : : /* Can we fit all of the mapping? */
1995 : 414 : vm_len = vma->vm_end - vma->vm_start;
1996 [ + - ]: 414 : if (vm_len >> PAGE_SHIFT > pages)
1997 : : return -EINVAL;
1998 : :
1999 : : /* Ok, let it rip */
2000 : 414 : return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
2001 : : }
2002 : : EXPORT_SYMBOL(vm_iomap_memory);
2003 : :
2004 : 95407 : static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2005 : : unsigned long addr, unsigned long end,
2006 : : pte_fn_t fn, void *data)
2007 : : {
2008 : : pte_t *pte;
2009 : : int err;
2010 : : spinlock_t *uninitialized_var(ptl);
2011 : :
2012 : : pte = (mm == &init_mm) ?
2013 [ + - - + : 190814 : pte_alloc_kernel(pmd, addr) :
# # ]
2014 [ # # # # ]: 0 : pte_alloc_map_lock(mm, pmd, addr, &ptl);
2015 [ + - ]: 95407 : if (!pte)
2016 : : return -ENOMEM;
2017 : :
2018 : : BUG_ON(pmd_huge(*pmd));
2019 : :
2020 : : arch_enter_lazy_mmu_mode();
2021 : :
2022 : : do {
2023 : 1656209 : err = fn(pte++, addr, data);
2024 [ + - ]: 1656208 : if (err)
2025 : : break;
2026 [ + + ]: 1656209 : } while (addr += PAGE_SIZE, addr != end);
2027 : :
2028 : : arch_leave_lazy_mmu_mode();
2029 : :
2030 [ - + ]: 95407 : if (mm != &init_mm)
2031 : : pte_unmap_unlock(pte-1, ptl);
2032 : 95407 : return err;
2033 : : }
2034 : :
2035 : : static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2036 : : unsigned long addr, unsigned long end,
2037 : : pte_fn_t fn, void *data)
2038 : : {
2039 : : pmd_t *pmd;
2040 : : unsigned long next;
2041 : : int err;
2042 : :
2043 : : BUG_ON(pud_huge(*pud));
2044 : :
2045 : : pmd = pmd_alloc(mm, pud, addr);
2046 [ + - ]: 95407 : if (!pmd)
2047 : : return -ENOMEM;
2048 : : do {
2049 : : next = pmd_addr_end(addr, end);
2050 : 95408 : err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2051 : : if (err)
2052 : : break;
2053 : : } while (pmd++, addr = next, addr != end);
2054 : : return err;
2055 : : }
2056 : :
2057 : : static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d,
2058 : : unsigned long addr, unsigned long end,
2059 : : pte_fn_t fn, void *data)
2060 : : {
2061 : : pud_t *pud;
2062 : : unsigned long next;
2063 : : int err;
2064 : :
2065 : : pud = pud_alloc(mm, p4d, addr);
2066 [ + - ]: 95407 : if (!pud)
2067 : : return -ENOMEM;
2068 : : do {
2069 : : next = pud_addr_end(addr, end);
2070 : : err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2071 : : if (err)
2072 : : break;
2073 : : } while (pud++, addr = next, addr != end);
2074 : : return err;
2075 : : }
2076 : :
2077 : 95408 : static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2078 : : unsigned long addr, unsigned long end,
2079 : : pte_fn_t fn, void *data)
2080 : : {
2081 : : p4d_t *p4d;
2082 : : unsigned long next;
2083 : : int err;
2084 : :
2085 : : p4d = p4d_alloc(mm, pgd, addr);
2086 [ + + ]: 95408 : if (!p4d)
2087 : : return -ENOMEM;
2088 : : do {
2089 : : next = p4d_addr_end(addr, end);
2090 : : err = apply_to_pud_range(mm, p4d, addr, next, fn, data);
2091 : : if (err)
2092 : : break;
2093 : : } while (p4d++, addr = next, addr != end);
2094 : 95407 : return err;
2095 : : }
2096 : :
2097 : : /*
2098 : : * Scan a region of virtual memory, filling in page tables as necessary
2099 : : * and calling a provided function on each leaf page table.
2100 : : */
2101 : 93363 : int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2102 : : unsigned long size, pte_fn_t fn, void *data)
2103 : : {
2104 : : pgd_t *pgd;
2105 : : unsigned long next;
2106 : 93363 : unsigned long end = addr + size;
2107 : : int err;
2108 : :
2109 [ - + + - ]: 93363 : if (WARN_ON(addr >= end))
2110 : : return -EINVAL;
2111 : :
2112 : 93362 : pgd = pgd_offset(mm, addr);
2113 : : do {
2114 [ + + ]: 95407 : next = pgd_addr_end(addr, end);
2115 : 95407 : err = apply_to_p4d_range(mm, pgd, addr, next, fn, data);
2116 [ + - ]: 95407 : if (err)
2117 : : break;
2118 [ + + ]: 95407 : } while (pgd++, addr = next, addr != end);
2119 : :
2120 : 93362 : return err;
2121 : : }
2122 : : EXPORT_SYMBOL_GPL(apply_to_page_range);
2123 : :
2124 : : /*
2125 : : * handle_pte_fault chooses page fault handler according to an entry which was
2126 : : * read non-atomically. Before making any commitment, on those architectures
2127 : : * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
2128 : : * parts, do_swap_page must check under lock before unmapping the pte and
2129 : : * proceeding (but do_wp_page is only called after already making such a check;
2130 : : * and do_anonymous_page can safely check later on).
2131 : : */
2132 : : static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
2133 : : pte_t *page_table, pte_t orig_pte)
2134 : : {
2135 : : int same = 1;
2136 : : #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2137 : : if (sizeof(pte_t) > sizeof(unsigned long)) {
2138 : : spinlock_t *ptl = pte_lockptr(mm, pmd);
2139 : : spin_lock(ptl);
2140 : : same = pte_same(*page_table, orig_pte);
2141 : : spin_unlock(ptl);
2142 : : }
2143 : : #endif
2144 : : pte_unmap(page_table);
2145 : : return same;
2146 : : }
2147 : :
2148 : 6610763 : static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
2149 : : {
2150 : : debug_dma_assert_idle(src);
2151 : :
2152 : : /*
2153 : : * If the source page was a PFN mapping, we don't have
2154 : : * a "struct page" for it. We do a best-effort copy by
2155 : : * just copying from the original user address. If that
2156 : : * fails, we just zero-fill it. Live with it.
2157 : : */
2158 [ - + ]: 6610763 : if (unlikely(!src)) {
2159 : 0 : void *kaddr = kmap_atomic(dst);
2160 : 0 : void __user *uaddr = (void __user *)(va & PAGE_MASK);
2161 : :
2162 : : /*
2163 : : * This really shouldn't fail, because the page is there
2164 : : * in the page tables. But it might just be unreadable,
2165 : : * in which case we just give up and fill the result with
2166 : : * zeroes.
2167 : : */
2168 [ # # ]: 0 : if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
2169 : 0 : clear_page(kaddr);
2170 : : kunmap_atomic(kaddr);
2171 : 0 : flush_dcache_page(dst);
2172 : : } else
2173 : 6610763 : copy_user_highpage(dst, src, va, vma);
2174 : 6615909 : }
2175 : :
2176 : : static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
2177 : : {
2178 : 46599698 : struct file *vm_file = vma->vm_file;
2179 : :
2180 [ + + ]: 46599698 : if (vm_file)
2181 : 51887636 : return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
2182 : :
2183 : : /*
2184 : : * Special mappings (e.g. VDSO) do not have any file so fake
2185 : : * a default GFP_KERNEL for them.
2186 : : */
2187 : : return GFP_KERNEL;
2188 : : }
2189 : :
2190 : : /*
2191 : : * Notify the address space that the page is about to become writable so that
2192 : : * it can prohibit this or wait for the page to get into an appropriate state.
2193 : : *
2194 : : * We do this without the lock held, so that it can sleep if it needs to.
2195 : : */
2196 : 1090 : static vm_fault_t do_page_mkwrite(struct vm_fault *vmf)
2197 : : {
2198 : : vm_fault_t ret;
2199 : 1090 : struct page *page = vmf->page;
2200 : 1090 : unsigned int old_flags = vmf->flags;
2201 : :
2202 : 1090 : vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2203 : :
2204 [ + - + - ]: 2180 : if (vmf->vma->vm_file &&
2205 : 1090 : IS_SWAPFILE(vmf->vma->vm_file->f_mapping->host))
2206 : : return VM_FAULT_SIGBUS;
2207 : :
2208 : 1090 : ret = vmf->vma->vm_ops->page_mkwrite(vmf);
2209 : : /* Restore original flags so that caller is not surprised */
2210 : 1090 : vmf->flags = old_flags;
2211 [ + - ]: 1090 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2212 : : return ret;
2213 [ - + ]: 1090 : if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2214 : 0 : lock_page(page);
2215 [ # # ]: 0 : if (!page->mapping) {
2216 : 0 : unlock_page(page);
2217 : 0 : return 0; /* retry */
2218 : : }
2219 : 0 : ret |= VM_FAULT_LOCKED;
2220 : : } else
2221 : : VM_BUG_ON_PAGE(!PageLocked(page), page);
2222 : 1090 : return ret;
2223 : : }
2224 : :
2225 : : /*
2226 : : * Handle dirtying of a page in shared file mapping on a write fault.
2227 : : *
2228 : : * The function expects the page to be locked and unlocks it.
2229 : : */
2230 : 884831 : static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf)
2231 : : {
2232 : 884831 : struct vm_area_struct *vma = vmf->vma;
2233 : : struct address_space *mapping;
2234 : 884831 : struct page *page = vmf->page;
2235 : : bool dirtied;
2236 [ + - + + ]: 884831 : bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite;
2237 : :
2238 : 884831 : dirtied = set_page_dirty(page);
2239 : : VM_BUG_ON_PAGE(PageAnon(page), page);
2240 : : /*
2241 : : * Take a local copy of the address_space - page.mapping may be zeroed
2242 : : * by truncate after unlock_page(). The address_space itself remains
2243 : : * pinned by vma->vm_file's reference. We rely on unlock_page()'s
2244 : : * release semantics to prevent the compiler from undoing this copying.
2245 : : */
2246 : 884831 : mapping = page_rmapping(page);
2247 : 884831 : unlock_page(page);
2248 : :
2249 [ + + ]: 884831 : if (!page_mkwrite)
2250 : 883741 : file_update_time(vma->vm_file);
2251 : :
2252 : : /*
2253 : : * Throttle page dirtying rate down to writeback speed.
2254 : : *
2255 : : * mapping may be NULL here because some device drivers do not
2256 : : * set page.mapping but still dirty their pages
2257 : : *
2258 : : * Drop the mmap_sem before waiting on IO, if we can. The file
2259 : : * is pinning the mapping, as per above.
2260 : : */
2261 [ + + + - ]: 884831 : if ((dirtied || page_mkwrite) && mapping) {
2262 : : struct file *fpin;
2263 : :
2264 : 846685 : fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2265 : 846685 : balance_dirty_pages_ratelimited(mapping);
2266 [ + - ]: 846685 : if (fpin) {
2267 : 846685 : fput(fpin);
2268 : 846685 : return VM_FAULT_RETRY;
2269 : : }
2270 : : }
2271 : :
2272 : : return 0;
2273 : : }
2274 : :
2275 : : /*
2276 : : * Handle write page faults for pages that can be reused in the current vma
2277 : : *
2278 : : * This can happen either due to the mapping being with the VM_SHARED flag,
2279 : : * or due to us being the last reference standing to the page. In either
2280 : : * case, all we need to do here is to mark the page as writable and update
2281 : : * any related book-keeping.
2282 : : */
2283 : 4494227 : static inline void wp_page_reuse(struct vm_fault *vmf)
2284 : : __releases(vmf->ptl)
2285 : : {
2286 : 4494227 : struct vm_area_struct *vma = vmf->vma;
2287 : : struct page *page = vmf->page;
2288 : : pte_t entry;
2289 : : /*
2290 : : * Clear the pages cpupid information as the existing
2291 : : * information potentially belongs to a now completely
2292 : : * unrelated process.
2293 : : */
2294 : : if (page)
2295 : : page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2296 : :
2297 : 4494227 : flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
2298 : 4494048 : entry = pte_mkyoung(vmf->orig_pte);
2299 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2300 : 4494048 : if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
2301 : : update_mmu_cache(vma, vmf->address, vmf->pte);
2302 : 4494359 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2303 : 4495019 : }
2304 : :
2305 : : /*
2306 : : * Handle the case of a page which we actually need to copy to a new page.
2307 : : *
2308 : : * Called with mmap_sem locked and the old page referenced, but
2309 : : * without the ptl held.
2310 : : *
2311 : : * High level logic flow:
2312 : : *
2313 : : * - Allocate a page, copy the content of the old page to the new one.
2314 : : * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2315 : : * - Take the PTL. If the pte changed, bail out and release the allocated page
2316 : : * - If the pte is still the way we remember it, update the page table and all
2317 : : * relevant references. This includes dropping the reference the page-table
2318 : : * held to the old page, as well as updating the rmap.
2319 : : * - In any case, unlock the PTL and drop the reference we took to the old page.
2320 : : */
2321 : 7128001 : static vm_fault_t wp_page_copy(struct vm_fault *vmf)
2322 : : {
2323 : 7128001 : struct vm_area_struct *vma = vmf->vma;
2324 : 7128001 : struct mm_struct *mm = vma->vm_mm;
2325 : 7128001 : struct page *old_page = vmf->page;
2326 : : struct page *new_page = NULL;
2327 : : pte_t entry;
2328 : : int page_copied = 0;
2329 : : struct mem_cgroup *memcg;
2330 : : struct mmu_notifier_range range;
2331 : :
2332 [ + - ]: 7120182 : if (unlikely(anon_vma_prepare(vma)))
2333 : : goto oom;
2334 : :
2335 [ + + ]: 14247742 : if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
2336 : 514051 : new_page = alloc_zeroed_user_highpage_movable(vma,
2337 : : vmf->address);
2338 [ + + ]: 514232 : if (!new_page)
2339 : : goto oom;
2340 : : } else {
2341 : : new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
2342 : : vmf->address);
2343 [ + + ]: 6616544 : if (!new_page)
2344 : : goto oom;
2345 : 6612855 : cow_user_page(new_page, old_page, vmf->address, vma);
2346 : : }
2347 : :
2348 [ + - ]: 7130786 : if (mem_cgroup_try_charge_delay(new_page, mm, GFP_KERNEL, &memcg, false))
2349 : : goto oom_free_new;
2350 : :
2351 : : __SetPageUptodate(new_page);
2352 : :
2353 : 7123512 : mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm,
2354 : : vmf->address & PAGE_MASK,
2355 : : (vmf->address & PAGE_MASK) + PAGE_SIZE);
2356 : : mmu_notifier_invalidate_range_start(&range);
2357 : :
2358 : : /*
2359 : : * Re-check the pte - we dropped the lock
2360 : : */
2361 : 21375365 : vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
2362 [ + + ]: 14256682 : if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
2363 [ + + ]: 7128297 : if (old_page) {
2364 [ + + ]: 6610907 : if (!PageAnon(old_page)) {
2365 : 3201542 : dec_mm_counter_fast(mm,
2366 : : mm_counter_file(old_page));
2367 : 3203558 : inc_mm_counter_fast(mm, MM_ANONPAGES);
2368 : : }
2369 : : } else {
2370 : 514138 : inc_mm_counter_fast(mm, MM_ANONPAGES);
2371 : : }
2372 : 7128128 : flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
2373 : 7123661 : entry = mk_pte(new_page, vma->vm_page_prot);
2374 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2375 : : /*
2376 : : * Clear the pte entry and flush it first, before updating the
2377 : : * pte with the new entry. This will avoid a race condition
2378 : : * seen in the presence of one thread doing SMC and another
2379 : : * thread doing COW.
2380 : : */
2381 : 7123661 : ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
2382 : 7123087 : page_add_new_anon_rmap(new_page, vma, vmf->address, false);
2383 : 7127738 : mem_cgroup_commit_charge(new_page, memcg, false, false);
2384 : 7127783 : lru_cache_add_active_or_unevictable(new_page, vma);
2385 : : /*
2386 : : * We call the notify macro here because, when using secondary
2387 : : * mmu page tables (such as kvm shadow page tables), we want the
2388 : : * new page to be mapped directly into the secondary page table.
2389 : : */
2390 : 7127452 : set_pte_at_notify(mm, vmf->address, vmf->pte, entry);
2391 : : update_mmu_cache(vma, vmf->address, vmf->pte);
2392 [ + + ]: 7118256 : if (old_page) {
2393 : : /*
2394 : : * Only after switching the pte to the new page may
2395 : : * we remove the mapcount here. Otherwise another
2396 : : * process may come and find the rmap count decremented
2397 : : * before the pte is switched to the new page, and
2398 : : * "reuse" the old page writing into it while our pte
2399 : : * here still points into it and can be read by other
2400 : : * threads.
2401 : : *
2402 : : * The critical issue is to order this
2403 : : * page_remove_rmap with the ptp_clear_flush above.
2404 : : * Those stores are ordered by (if nothing else,)
2405 : : * the barrier present in the atomic_add_negative
2406 : : * in page_remove_rmap.
2407 : : *
2408 : : * Then the TLB flush in ptep_clear_flush ensures that
2409 : : * no process can access the old page before the
2410 : : * decremented mapcount is visible. And the old page
2411 : : * cannot be reused until after the decremented
2412 : : * mapcount is visible. So transitively, TLBs to
2413 : : * old page will be flushed before it can be reused.
2414 : : */
2415 : 6610907 : page_remove_rmap(old_page, false);
2416 : : }
2417 : :
2418 : : /* Free the old page.. */
2419 : : new_page = old_page;
2420 : : page_copied = 1;
2421 : : } else {
2422 : 44 : mem_cgroup_cancel_charge(new_page, memcg, false);
2423 : : }
2424 : :
2425 [ + + ]: 7129878 : if (new_page)
2426 : 6615675 : put_page(new_page);
2427 : :
2428 : 7130482 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2429 : : /*
2430 : : * No need to double call mmu_notifier->invalidate_range() callback as
2431 : : * the above ptep_clear_flush_notify() did already call it.
2432 : : */
2433 : : mmu_notifier_invalidate_range_only_end(&range);
2434 [ + + ]: 7123094 : if (old_page) {
2435 : : /*
2436 : : * Don't let another task, with possibly unlocked vma,
2437 : : * keep the mlocked page.
2438 : : */
2439 [ + + - + ]: 6615728 : if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2440 : 0 : lock_page(old_page); /* LRU manipulation */
2441 [ # # ]: 0 : if (PageMlocked(old_page))
2442 : 0 : munlock_vma_page(old_page);
2443 : 0 : unlock_page(old_page);
2444 : : }
2445 : 6615728 : put_page(old_page);
2446 : : }
2447 [ + + ]: 7128897 : return page_copied ? VM_FAULT_WRITE : 0;
2448 : : oom_free_new:
2449 : 0 : put_page(new_page);
2450 : : oom:
2451 [ # # ]: 0 : if (old_page)
2452 : 0 : put_page(old_page);
2453 : : return VM_FAULT_OOM;
2454 : : }
2455 : :
2456 : : /**
2457 : : * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE
2458 : : * writeable once the page is prepared
2459 : : *
2460 : : * @vmf: structure describing the fault
2461 : : *
2462 : : * This function handles all that is needed to finish a write page fault in a
2463 : : * shared mapping due to PTE being read-only once the mapped page is prepared.
2464 : : * It handles locking of PTE and modifying it.
2465 : : *
2466 : : * The function expects the page to be locked or other protection against
2467 : : * concurrent faults / writeback (such as DAX radix tree locks).
2468 : : *
2469 : : * Return: %VM_FAULT_WRITE on success, %0 when PTE got changed before
2470 : : * we acquired PTE lock.
2471 : : */
2472 : 100 : vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf)
2473 : : {
2474 [ - + # # ]: 100 : WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
2475 : 300 : vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address,
2476 : : &vmf->ptl);
2477 : : /*
2478 : : * We might have raced with another page fault while we released the
2479 : : * pte_offset_map_lock.
2480 : : */
2481 [ - + ]: 100 : if (!pte_same(*vmf->pte, vmf->orig_pte)) {
2482 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2483 : 0 : return VM_FAULT_NOPAGE;
2484 : : }
2485 : 100 : wp_page_reuse(vmf);
2486 : 100 : return 0;
2487 : : }
2488 : :
2489 : : /*
2490 : : * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2491 : : * mapping
2492 : : */
2493 : 0 : static vm_fault_t wp_pfn_shared(struct vm_fault *vmf)
2494 : : {
2495 : 0 : struct vm_area_struct *vma = vmf->vma;
2496 : :
2497 [ # # # # ]: 0 : if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
2498 : : vm_fault_t ret;
2499 : :
2500 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2501 : 0 : vmf->flags |= FAULT_FLAG_MKWRITE;
2502 : 0 : ret = vma->vm_ops->pfn_mkwrite(vmf);
2503 [ # # ]: 0 : if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))
2504 : : return ret;
2505 : 0 : return finish_mkwrite_fault(vmf);
2506 : : }
2507 : 0 : wp_page_reuse(vmf);
2508 : 0 : return VM_FAULT_WRITE;
2509 : : }
2510 : :
2511 : 100 : static vm_fault_t wp_page_shared(struct vm_fault *vmf)
2512 : : __releases(vmf->ptl)
2513 : : {
2514 : 100 : struct vm_area_struct *vma = vmf->vma;
2515 : : vm_fault_t ret = VM_FAULT_WRITE;
2516 : :
2517 : 100 : get_page(vmf->page);
2518 : :
2519 [ + - + - ]: 100 : if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2520 : : vm_fault_t tmp;
2521 : :
2522 : 100 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2523 : 100 : tmp = do_page_mkwrite(vmf);
2524 [ + - - + ]: 100 : if (unlikely(!tmp || (tmp &
2525 : : (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2526 : 0 : put_page(vmf->page);
2527 : 0 : return tmp;
2528 : : }
2529 : 100 : tmp = finish_mkwrite_fault(vmf);
2530 [ - + ]: 100 : if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2531 : 0 : unlock_page(vmf->page);
2532 : 0 : put_page(vmf->page);
2533 : 0 : return tmp;
2534 : : }
2535 : : } else {
2536 : 0 : wp_page_reuse(vmf);
2537 : 0 : lock_page(vmf->page);
2538 : : }
2539 : 100 : ret |= fault_dirty_shared_page(vmf);
2540 : 100 : put_page(vmf->page);
2541 : :
2542 : 100 : return ret;
2543 : : }
2544 : :
2545 : : /*
2546 : : * This routine handles present pages, when users try to write
2547 : : * to a shared page. It is done by copying the page to a new address
2548 : : * and decrementing the shared-page counter for the old page.
2549 : : *
2550 : : * Note that this routine assumes that the protection checks have been
2551 : : * done by the caller (the low-level page fault routine in most cases).
2552 : : * Thus we can safely just mark it writable once we've done any necessary
2553 : : * COW.
2554 : : *
2555 : : * We also mark the page dirty at this point even though the page will
2556 : : * change only once the write actually happens. This avoids a few races,
2557 : : * and potentially makes it more efficient.
2558 : : *
2559 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
2560 : : * but allow concurrent faults), with pte both mapped and locked.
2561 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
2562 : : */
2563 : 11622225 : static vm_fault_t do_wp_page(struct vm_fault *vmf)
2564 : : __releases(vmf->ptl)
2565 : : {
2566 : 11622225 : struct vm_area_struct *vma = vmf->vma;
2567 : :
2568 : 11622225 : vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
2569 [ + + ]: 11616561 : if (!vmf->page) {
2570 : : /*
2571 : : * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2572 : : * VM_PFNMAP VMA.
2573 : : *
2574 : : * We should not cow pages in a shared writeable mapping.
2575 : : * Just mark the pages writable and/or call ops->pfn_mkwrite.
2576 : : */
2577 [ - + ]: 514105 : if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2578 : : (VM_WRITE|VM_SHARED))
2579 : 0 : return wp_pfn_shared(vmf);
2580 : :
2581 : 514105 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2582 : 514171 : return wp_page_copy(vmf);
2583 : : }
2584 : :
2585 : : /*
2586 : : * Take out anonymous pages first, anonymous shared vmas are
2587 : : * not dirty accountable.
2588 : : */
2589 [ + + ]: 11105688 : if (PageAnon(vmf->page)) {
2590 : : int total_map_swapcount;
2591 : : if (PageKsm(vmf->page) && (PageSwapCache(vmf->page) ||
2592 : : page_count(vmf->page) != 1))
2593 : : goto copy;
2594 [ + + ]: 7902689 : if (!trylock_page(vmf->page)) {
2595 : 1149 : get_page(vmf->page);
2596 : 1153 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2597 : 1144 : lock_page(vmf->page);
2598 : 3462 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2599 : : vmf->address, &vmf->ptl);
2600 [ - + ]: 1154 : if (!pte_same(*vmf->pte, vmf->orig_pte)) {
2601 : 0 : unlock_page(vmf->page);
2602 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2603 : 0 : put_page(vmf->page);
2604 : 0 : return 0;
2605 : : }
2606 : 1154 : put_page(vmf->page);
2607 : : }
2608 : : if (PageKsm(vmf->page)) {
2609 : : bool reused = reuse_ksm_page(vmf->page, vmf->vma,
2610 : : vmf->address);
2611 : : unlock_page(vmf->page);
2612 : : if (!reused)
2613 : : goto copy;
2614 : : wp_page_reuse(vmf);
2615 : : return VM_FAULT_WRITE;
2616 : : }
2617 [ + + ]: 7903565 : if (reuse_swap_page(vmf->page, &total_map_swapcount)) {
2618 [ + + ]: 4495005 : if (total_map_swapcount == 1) {
2619 : : /*
2620 : : * The page is all ours. Move it to
2621 : : * our anon_vma so the rmap code will
2622 : : * not search our parent or siblings.
2623 : : * Protected against the rmap code by
2624 : : * the page lock.
2625 : : */
2626 : 4494063 : page_move_anon_rmap(vmf->page, vma);
2627 : : }
2628 : 4495559 : unlock_page(vmf->page);
2629 : 4494272 : wp_page_reuse(vmf);
2630 : 4494947 : return VM_FAULT_WRITE;
2631 : : }
2632 : 3410649 : unlock_page(vmf->page);
2633 [ + + ]: 3202999 : } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2634 : : (VM_WRITE|VM_SHARED))) {
2635 : 100 : return wp_page_shared(vmf);
2636 : : }
2637 : : copy:
2638 : : /*
2639 : : * Ok, we need to copy. Oh, well..
2640 : : */
2641 : 6613453 : get_page(vmf->page);
2642 : :
2643 : 6614846 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2644 : 6616854 : return wp_page_copy(vmf);
2645 : : }
2646 : :
2647 : : static void unmap_mapping_range_vma(struct vm_area_struct *vma,
2648 : : unsigned long start_addr, unsigned long end_addr,
2649 : : struct zap_details *details)
2650 : : {
2651 : 0 : zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
2652 : : }
2653 : :
2654 : 0 : static inline void unmap_mapping_range_tree(struct rb_root_cached *root,
2655 : : struct zap_details *details)
2656 : : {
2657 : : struct vm_area_struct *vma;
2658 : : pgoff_t vba, vea, zba, zea;
2659 : :
2660 [ # # ]: 0 : vma_interval_tree_foreach(vma, root,
2661 : : details->first_index, details->last_index) {
2662 : :
2663 : 0 : vba = vma->vm_pgoff;
2664 : 0 : vea = vba + vma_pages(vma) - 1;
2665 : 0 : zba = details->first_index;
2666 [ # # ]: 0 : if (zba < vba)
2667 : : zba = vba;
2668 : 0 : zea = details->last_index;
2669 [ # # ]: 0 : if (zea > vea)
2670 : : zea = vea;
2671 : :
2672 : 0 : unmap_mapping_range_vma(vma,
2673 : 0 : ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2674 : 0 : ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2675 : : details);
2676 : : }
2677 : 0 : }
2678 : :
2679 : : /**
2680 : : * unmap_mapping_pages() - Unmap pages from processes.
2681 : : * @mapping: The address space containing pages to be unmapped.
2682 : : * @start: Index of first page to be unmapped.
2683 : : * @nr: Number of pages to be unmapped. 0 to unmap to end of file.
2684 : : * @even_cows: Whether to unmap even private COWed pages.
2685 : : *
2686 : : * Unmap the pages in this address space from any userspace process which
2687 : : * has them mmaped. Generally, you want to remove COWed pages as well when
2688 : : * a file is being truncated, but not when invalidating pages from the page
2689 : : * cache.
2690 : : */
2691 : 2294 : void unmap_mapping_pages(struct address_space *mapping, pgoff_t start,
2692 : : pgoff_t nr, bool even_cows)
2693 : : {
2694 : 2294 : struct zap_details details = { };
2695 : :
2696 [ - + ]: 2294 : details.check_mapping = even_cows ? NULL : mapping;
2697 : 2294 : details.first_index = start;
2698 : 2294 : details.last_index = start + nr - 1;
2699 [ - + ]: 2294 : if (details.last_index < details.first_index)
2700 : 0 : details.last_index = ULONG_MAX;
2701 : :
2702 : : i_mmap_lock_write(mapping);
2703 [ - + ]: 2294 : if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)))
2704 : 0 : unmap_mapping_range_tree(&mapping->i_mmap, &details);
2705 : : i_mmap_unlock_write(mapping);
2706 : 2294 : }
2707 : :
2708 : : /**
2709 : : * unmap_mapping_range - unmap the portion of all mmaps in the specified
2710 : : * address_space corresponding to the specified byte range in the underlying
2711 : : * file.
2712 : : *
2713 : : * @mapping: the address space containing mmaps to be unmapped.
2714 : : * @holebegin: byte in first page to unmap, relative to the start of
2715 : : * the underlying file. This will be rounded down to a PAGE_SIZE
2716 : : * boundary. Note that this is different from truncate_pagecache(), which
2717 : : * must keep the partial page. In contrast, we must get rid of
2718 : : * partial pages.
2719 : : * @holelen: size of prospective hole in bytes. This will be rounded
2720 : : * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2721 : : * end of the file.
2722 : : * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2723 : : * but 0 when invalidating pagecache, don't throw away private data.
2724 : : */
2725 : 2294 : void unmap_mapping_range(struct address_space *mapping,
2726 : : loff_t const holebegin, loff_t const holelen, int even_cows)
2727 : : {
2728 : 2294 : pgoff_t hba = holebegin >> PAGE_SHIFT;
2729 : 2294 : pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2730 : :
2731 : : /* Check for overflow. */
2732 : : if (sizeof(holelen) > sizeof(hlen)) {
2733 : 2294 : long long holeend =
2734 : 2294 : (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2735 [ - + ]: 2294 : if (holeend & ~(long long)ULONG_MAX)
2736 : 0 : hlen = ULONG_MAX - hba + 1;
2737 : : }
2738 : :
2739 : 2294 : unmap_mapping_pages(mapping, hba, hlen, even_cows);
2740 : 2294 : }
2741 : : EXPORT_SYMBOL(unmap_mapping_range);
2742 : :
2743 : : /*
2744 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
2745 : : * but allow concurrent faults), and pte mapped but not yet locked.
2746 : : * We return with pte unmapped and unlocked.
2747 : : *
2748 : : * We return with the mmap_sem locked or unlocked in the same cases
2749 : : * as does filemap_fault().
2750 : : */
2751 : 0 : vm_fault_t do_swap_page(struct vm_fault *vmf)
2752 : : {
2753 : 0 : struct vm_area_struct *vma = vmf->vma;
2754 : : struct page *page = NULL, *swapcache;
2755 : : struct mem_cgroup *memcg;
2756 : : swp_entry_t entry;
2757 : : pte_t pte;
2758 : : int locked;
2759 : : int exclusive = 0;
2760 : : vm_fault_t ret = 0;
2761 : :
2762 : : if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte))
2763 : : goto out;
2764 : :
2765 : 0 : entry = pte_to_swp_entry(vmf->orig_pte);
2766 [ # # ]: 0 : if (unlikely(non_swap_entry(entry))) {
2767 [ # # ]: 0 : if (is_migration_entry(entry)) {
2768 : 0 : migration_entry_wait(vma->vm_mm, vmf->pmd,
2769 : : vmf->address);
2770 : : } else if (is_device_private_entry(entry)) {
2771 : : vmf->page = device_private_entry_to_page(entry);
2772 : : ret = vmf->page->pgmap->ops->migrate_to_ram(vmf);
2773 : : } else if (is_hwpoison_entry(entry)) {
2774 : : ret = VM_FAULT_HWPOISON;
2775 : : } else {
2776 : 0 : print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
2777 : : ret = VM_FAULT_SIGBUS;
2778 : : }
2779 : : goto out;
2780 : : }
2781 : :
2782 : :
2783 : : delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2784 : 0 : page = lookup_swap_cache(entry, vma, vmf->address);
2785 : : swapcache = page;
2786 : :
2787 [ # # ]: 0 : if (!page) {
2788 : 0 : struct swap_info_struct *si = swp_swap_info(entry);
2789 : :
2790 [ # # # # ]: 0 : if (si->flags & SWP_SYNCHRONOUS_IO &&
2791 : 0 : __swap_count(entry) == 1) {
2792 : : /* skip swapcache */
2793 : : page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
2794 : : vmf->address);
2795 [ # # ]: 0 : if (page) {
2796 : : __SetPageLocked(page);
2797 : : __SetPageSwapBacked(page);
2798 : 0 : set_page_private(page, entry.val);
2799 : 0 : lru_cache_add_anon(page);
2800 : 0 : swap_readpage(page, true);
2801 : : }
2802 : : } else {
2803 : 0 : page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
2804 : : vmf);
2805 : : swapcache = page;
2806 : : }
2807 : :
2808 [ # # ]: 0 : if (!page) {
2809 : : /*
2810 : : * Back out if somebody else faulted in this pte
2811 : : * while we released the pte lock.
2812 : : */
2813 : 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2814 : : vmf->address, &vmf->ptl);
2815 [ # # ]: 0 : if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
2816 : : ret = VM_FAULT_OOM;
2817 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2818 : : goto unlock;
2819 : : }
2820 : :
2821 : : /* Had to read the page from swap area: Major fault */
2822 : : ret = VM_FAULT_MAJOR;
2823 : : count_vm_event(PGMAJFAULT);
2824 : 0 : count_memcg_event_mm(vma->vm_mm, PGMAJFAULT);
2825 : : } else if (PageHWPoison(page)) {
2826 : : /*
2827 : : * hwpoisoned dirty swapcache pages are kept for killing
2828 : : * owner processes (which may be unknown at hwpoison time)
2829 : : */
2830 : : ret = VM_FAULT_HWPOISON;
2831 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2832 : : goto out_release;
2833 : : }
2834 : :
2835 : 0 : locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags);
2836 : :
2837 : : delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2838 [ # # ]: 0 : if (!locked) {
2839 : 0 : ret |= VM_FAULT_RETRY;
2840 : 0 : goto out_release;
2841 : : }
2842 : :
2843 : : /*
2844 : : * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2845 : : * release the swapcache from under us. The page pin, and pte_same
2846 : : * test below, are not enough to exclude that. Even if it is still
2847 : : * swapcache, we need to check that the page's swap has not changed.
2848 : : */
2849 [ # # # # ]: 0 : if (unlikely((!PageSwapCache(page) ||
2850 [ # # ]: 0 : page_private(page) != entry.val)) && swapcache)
2851 : : goto out_page;
2852 : :
2853 : : page = ksm_might_need_to_copy(page, vma, vmf->address);
2854 [ # # ]: 0 : if (unlikely(!page)) {
2855 : : ret = VM_FAULT_OOM;
2856 : : page = swapcache;
2857 : : goto out_page;
2858 : : }
2859 : :
2860 [ # # ]: 0 : if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL,
2861 : : &memcg, false)) {
2862 : : ret = VM_FAULT_OOM;
2863 : : goto out_page;
2864 : : }
2865 : :
2866 : : /*
2867 : : * Back out if somebody else already faulted in this pte.
2868 : : */
2869 : 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
2870 : : &vmf->ptl);
2871 [ # # ]: 0 : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
2872 : : goto out_nomap;
2873 : :
2874 [ # # ]: 0 : if (unlikely(!PageUptodate(page))) {
2875 : : ret = VM_FAULT_SIGBUS;
2876 : : goto out_nomap;
2877 : : }
2878 : :
2879 : : /*
2880 : : * The page isn't present yet, go ahead with the fault.
2881 : : *
2882 : : * Be careful about the sequence of operations here.
2883 : : * To get its accounting right, reuse_swap_page() must be called
2884 : : * while the page is counted on swap but not yet in mapcount i.e.
2885 : : * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2886 : : * must be called after the swap_free(), or it will never succeed.
2887 : : */
2888 : :
2889 : 0 : inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2890 : 0 : dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS);
2891 : 0 : pte = mk_pte(page, vma->vm_page_prot);
2892 [ # # # # ]: 0 : if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) {
2893 : : pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2894 : 0 : vmf->flags &= ~FAULT_FLAG_WRITE;
2895 : 0 : ret |= VM_FAULT_WRITE;
2896 : : exclusive = RMAP_EXCLUSIVE;
2897 : : }
2898 : : flush_icache_page(vma, page);
2899 : : if (pte_swp_soft_dirty(vmf->orig_pte))
2900 : : pte = pte_mksoft_dirty(pte);
2901 : 0 : set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
2902 : : arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte);
2903 : 0 : vmf->orig_pte = pte;
2904 : :
2905 : : /* ksm created a completely new copy */
2906 [ # # ]: 0 : if (unlikely(page != swapcache && swapcache)) {
2907 : 0 : page_add_new_anon_rmap(page, vma, vmf->address, false);
2908 : 0 : mem_cgroup_commit_charge(page, memcg, false, false);
2909 : 0 : lru_cache_add_active_or_unevictable(page, vma);
2910 : : } else {
2911 : 0 : do_page_add_anon_rmap(page, vma, vmf->address, exclusive);
2912 : 0 : mem_cgroup_commit_charge(page, memcg, true, false);
2913 : 0 : activate_page(page);
2914 : : }
2915 : :
2916 : 0 : swap_free(entry);
2917 [ # # # # ]: 0 : if (mem_cgroup_swap_full(page) ||
2918 [ # # ]: 0 : (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
2919 : 0 : try_to_free_swap(page);
2920 : 0 : unlock_page(page);
2921 [ # # ]: 0 : if (page != swapcache && swapcache) {
2922 : : /*
2923 : : * Hold the lock to avoid the swap entry to be reused
2924 : : * until we take the PT lock for the pte_same() check
2925 : : * (to avoid false positives from pte_same). For
2926 : : * further safety release the lock after the swap_free
2927 : : * so that the swap count won't change under a
2928 : : * parallel locked swapcache.
2929 : : */
2930 : 0 : unlock_page(swapcache);
2931 : 0 : put_page(swapcache);
2932 : : }
2933 : :
2934 [ # # ]: 0 : if (vmf->flags & FAULT_FLAG_WRITE) {
2935 : 0 : ret |= do_wp_page(vmf);
2936 [ # # ]: 0 : if (ret & VM_FAULT_ERROR)
2937 : : ret &= VM_FAULT_ERROR;
2938 : : goto out;
2939 : : }
2940 : :
2941 : : /* No need to invalidate - it was non-present before */
2942 : : update_mmu_cache(vma, vmf->address, vmf->pte);
2943 : : unlock:
2944 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2945 : : out:
2946 : 0 : return ret;
2947 : : out_nomap:
2948 : 0 : mem_cgroup_cancel_charge(page, memcg, false);
2949 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
2950 : : out_page:
2951 : 0 : unlock_page(page);
2952 : : out_release:
2953 : 0 : put_page(page);
2954 [ # # ]: 0 : if (page != swapcache && swapcache) {
2955 : 0 : unlock_page(swapcache);
2956 : 0 : put_page(swapcache);
2957 : : }
2958 : 0 : return ret;
2959 : : }
2960 : :
2961 : : /*
2962 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
2963 : : * but allow concurrent faults), and pte mapped but not yet locked.
2964 : : * We return with mmap_sem still held, but pte unmapped and unlocked.
2965 : : */
2966 : 10488202 : static vm_fault_t do_anonymous_page(struct vm_fault *vmf)
2967 : : {
2968 : 10488202 : struct vm_area_struct *vma = vmf->vma;
2969 : : struct mem_cgroup *memcg;
2970 : : struct page *page;
2971 : : vm_fault_t ret = 0;
2972 : : pte_t entry;
2973 : :
2974 : : /* File mapping without ->vm_ops ? */
2975 [ + + ]: 10488202 : if (vma->vm_flags & VM_SHARED)
2976 : : return VM_FAULT_SIGBUS;
2977 : :
2978 : : /*
2979 : : * Use pte_alloc() instead of pte_alloc_map(). We can't run
2980 : : * pte_offset_map() on pmds where a huge pmd might be created
2981 : : * from a different thread.
2982 : : *
2983 : : * pte_alloc_map() is safe to use under down_write(mmap_sem) or when
2984 : : * parallel threads are excluded by other means.
2985 : : *
2986 : : * Here we only have down_read(mmap_sem).
2987 : : */
2988 [ + + + - ]: 10487306 : if (pte_alloc(vma->vm_mm, vmf->pmd))
2989 : : return VM_FAULT_OOM;
2990 : :
2991 : : /* See the comment in pte_alloc_one_map() */
2992 : : if (unlikely(pmd_trans_unstable(vmf->pmd)))
2993 : : return 0;
2994 : :
2995 : : /* Use the zero-page for reads */
2996 [ + + ]: 10488734 : if (!(vmf->flags & FAULT_FLAG_WRITE) &&
2997 : : !mm_forbids_zeropage(vma->vm_mm)) {
2998 : 746625 : entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
2999 : : vma->vm_page_prot));
3000 : 2239976 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
3001 : : vmf->address, &vmf->ptl);
3002 [ + + ]: 746726 : if (!pte_none(*vmf->pte))
3003 : : goto unlock;
3004 : 746674 : ret = check_stable_address_space(vma->vm_mm);
3005 [ + + ]: 746674 : if (ret)
3006 : : goto unlock;
3007 : : /* Deliver the page fault to userland, check inside PT lock */
3008 : : if (userfaultfd_missing(vma)) {
3009 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3010 : : return handle_userfault(vmf, VM_UFFD_MISSING);
3011 : : }
3012 : : goto setpte;
3013 : : }
3014 : :
3015 : : /* Allocate our own private page. */
3016 [ + - ]: 9742059 : if (unlikely(anon_vma_prepare(vma)))
3017 : : goto oom;
3018 : 9742059 : page = alloc_zeroed_user_highpage_movable(vma, vmf->address);
3019 [ + + ]: 9741620 : if (!page)
3020 : : goto oom;
3021 : :
3022 [ + - ]: 9739703 : if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, &memcg,
3023 : : false))
3024 : : goto oom_free_page;
3025 : :
3026 : : /*
3027 : : * The memory barrier inside __SetPageUptodate makes sure that
3028 : : * preceeding stores to the page contents become visible before
3029 : : * the set_pte_at() write.
3030 : : */
3031 : : __SetPageUptodate(page);
3032 : :
3033 : 9738305 : entry = mk_pte(page, vma->vm_page_prot);
3034 [ + + ]: 9738305 : if (vma->vm_flags & VM_WRITE)
3035 : : entry = pte_mkwrite(pte_mkdirty(entry));
3036 : :
3037 : 29218463 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3038 : : &vmf->ptl);
3039 [ + + ]: 9741853 : if (!pte_none(*vmf->pte))
3040 : : goto release;
3041 : :
3042 : 9739953 : ret = check_stable_address_space(vma->vm_mm);
3043 [ + + ]: 9739953 : if (ret)
3044 : : goto release;
3045 : :
3046 : : /* Deliver the page fault to userland, check inside PT lock */
3047 : : if (userfaultfd_missing(vma)) {
3048 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3049 : : mem_cgroup_cancel_charge(page, memcg, false);
3050 : : put_page(page);
3051 : : return handle_userfault(vmf, VM_UFFD_MISSING);
3052 : : }
3053 : :
3054 : 9740074 : inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
3055 : 9740639 : page_add_new_anon_rmap(page, vma, vmf->address, false);
3056 : 9738626 : mem_cgroup_commit_charge(page, memcg, false, false);
3057 : 9740045 : lru_cache_add_active_or_unevictable(page, vma);
3058 : : setpte:
3059 : 10486415 : set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
3060 : :
3061 : : /* No need to invalidate - it was non-present before */
3062 : : update_mmu_cache(vma, vmf->address, vmf->pte);
3063 : : unlock:
3064 : 10484467 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3065 : 10488097 : return ret;
3066 : : release:
3067 : 1779 : mem_cgroup_cancel_charge(page, memcg, false);
3068 : 9 : put_page(page);
3069 : 9 : goto unlock;
3070 : : oom_free_page:
3071 : 0 : put_page(page);
3072 : : oom:
3073 : : return VM_FAULT_OOM;
3074 : : }
3075 : :
3076 : : /*
3077 : : * The mmap_sem must have been held on entry, and may have been
3078 : : * released depending on flags and vma->vm_ops->fault() return value.
3079 : : * See filemap_fault() and __lock_page_retry().
3080 : : */
3081 : 4085388 : static vm_fault_t __do_fault(struct vm_fault *vmf)
3082 : : {
3083 : 4085388 : struct vm_area_struct *vma = vmf->vma;
3084 : : vm_fault_t ret;
3085 : :
3086 : : /*
3087 : : * Preallocate pte before we take page_lock because this might lead to
3088 : : * deadlocks for memcg reclaim which waits for pages under writeback:
3089 : : * lock_page(A)
3090 : : * SetPageWriteback(A)
3091 : : * unlock_page(A)
3092 : : * lock_page(B)
3093 : : * lock_page(B)
3094 : : * pte_alloc_pne
3095 : : * shrink_page_list
3096 : : * wait_on_page_writeback(A)
3097 : : * SetPageWriteback(B)
3098 : : * unlock_page(B)
3099 : : * # flush A, B to clear the writeback
3100 : : */
3101 [ + + + + ]: 4085388 : if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) {
3102 : 646692 : vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm);
3103 [ + - ]: 646774 : if (!vmf->prealloc_pte)
3104 : : return VM_FAULT_OOM;
3105 : 646794 : smp_wmb(); /* See comment in __pte_alloc() */
3106 : : }
3107 : :
3108 : 4085534 : ret = vma->vm_ops->fault(vmf);
3109 [ + + ]: 4094525 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY |
3110 : : VM_FAULT_DONE_COW)))
3111 : : return ret;
3112 : :
3113 : : if (unlikely(PageHWPoison(vmf->page))) {
3114 : : if (ret & VM_FAULT_LOCKED)
3115 : : unlock_page(vmf->page);
3116 : : put_page(vmf->page);
3117 : : vmf->page = NULL;
3118 : : return VM_FAULT_HWPOISON;
3119 : : }
3120 : :
3121 [ + + ]: 3739986 : if (unlikely(!(ret & VM_FAULT_LOCKED)))
3122 : 267305 : lock_page(vmf->page);
3123 : : else
3124 : : VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page);
3125 : :
3126 : 3736649 : return ret;
3127 : : }
3128 : :
3129 : : /*
3130 : : * The ordering of these checks is important for pmds with _PAGE_DEVMAP set.
3131 : : * If we check pmd_trans_unstable() first we will trip the bad_pmd() check
3132 : : * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly
3133 : : * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
3134 : : */
3135 : : static int pmd_devmap_trans_unstable(pmd_t *pmd)
3136 : : {
3137 : : return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
3138 : : }
3139 : :
3140 : 18400748 : static vm_fault_t pte_alloc_one_map(struct vm_fault *vmf)
3141 : : {
3142 : 18400748 : struct vm_area_struct *vma = vmf->vma;
3143 : :
3144 [ + + ]: 18400748 : if (!pmd_none(*vmf->pmd))
3145 : : goto map_pte;
3146 [ + - ]: 1274437 : if (vmf->prealloc_pte) {
3147 : 2548894 : vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3148 [ - + ]: 1274457 : if (unlikely(!pmd_none(*vmf->pmd))) {
3149 : : spin_unlock(vmf->ptl);
3150 : : goto map_pte;
3151 : : }
3152 : :
3153 : 1274457 : mm_inc_nr_ptes(vma->vm_mm);
3154 : 1274454 : pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
3155 : 1274358 : spin_unlock(vmf->ptl);
3156 : 1274320 : vmf->prealloc_pte = NULL;
3157 [ # # # # : 0 : } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) {
# # ]
3158 : : return VM_FAULT_OOM;
3159 : : }
3160 : : map_pte:
3161 : : /*
3162 : : * If a huge pmd materialized under us just retry later. Use
3163 : : * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of
3164 : : * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge
3165 : : * under us and then back to pmd_none, as a result of MADV_DONTNEED
3166 : : * running immediately after a huge pmd fault in a different thread of
3167 : : * this mm, in turn leading to a misleading pmd_trans_huge() retval.
3168 : : * All we have to ensure is that it is a regular pmd that we can walk
3169 : : * with pte_offset_map() and we can do that through an atomic read in
3170 : : * C, which is what pmd_trans_unstable() provides.
3171 : : */
3172 : : if (pmd_devmap_trans_unstable(vmf->pmd))
3173 : : return VM_FAULT_NOPAGE;
3174 : :
3175 : : /*
3176 : : * At this point we know that our vmf->pmd points to a page of ptes
3177 : : * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge()
3178 : : * for the duration of the fault. If a racing MADV_DONTNEED runs and
3179 : : * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still
3180 : : * be valid and we will re-check to make sure the vmf->pte isn't
3181 : : * pte_none() under vmf->ptl protection when we return to
3182 : : * alloc_set_pte().
3183 : : */
3184 : 55213429 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3185 : : &vmf->ptl);
3186 : 18412167 : return 0;
3187 : : }
3188 : :
3189 : : #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3190 : : static void deposit_prealloc_pte(struct vm_fault *vmf)
3191 : : {
3192 : : struct vm_area_struct *vma = vmf->vma;
3193 : :
3194 : : pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
3195 : : /*
3196 : : * We are going to consume the prealloc table,
3197 : : * count that as nr_ptes.
3198 : : */
3199 : : mm_inc_nr_ptes(vma->vm_mm);
3200 : : vmf->prealloc_pte = NULL;
3201 : : }
3202 : :
3203 : : static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
3204 : : {
3205 : : struct vm_area_struct *vma = vmf->vma;
3206 : : bool write = vmf->flags & FAULT_FLAG_WRITE;
3207 : : unsigned long haddr = vmf->address & HPAGE_PMD_MASK;
3208 : : pmd_t entry;
3209 : : int i;
3210 : : vm_fault_t ret;
3211 : :
3212 : : if (!transhuge_vma_suitable(vma, haddr))
3213 : : return VM_FAULT_FALLBACK;
3214 : :
3215 : : ret = VM_FAULT_FALLBACK;
3216 : : page = compound_head(page);
3217 : :
3218 : : /*
3219 : : * Archs like ppc64 need additonal space to store information
3220 : : * related to pte entry. Use the preallocated table for that.
3221 : : */
3222 : : if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) {
3223 : : vmf->prealloc_pte = pte_alloc_one(vma->vm_mm);
3224 : : if (!vmf->prealloc_pte)
3225 : : return VM_FAULT_OOM;
3226 : : smp_wmb(); /* See comment in __pte_alloc() */
3227 : : }
3228 : :
3229 : : vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3230 : : if (unlikely(!pmd_none(*vmf->pmd)))
3231 : : goto out;
3232 : :
3233 : : for (i = 0; i < HPAGE_PMD_NR; i++)
3234 : : flush_icache_page(vma, page + i);
3235 : :
3236 : : entry = mk_huge_pmd(page, vma->vm_page_prot);
3237 : : if (write)
3238 : : entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
3239 : :
3240 : : add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR);
3241 : : page_add_file_rmap(page, true);
3242 : : /*
3243 : : * deposit and withdraw with pmd lock held
3244 : : */
3245 : : if (arch_needs_pgtable_deposit())
3246 : : deposit_prealloc_pte(vmf);
3247 : :
3248 : : set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
3249 : :
3250 : : update_mmu_cache_pmd(vma, haddr, vmf->pmd);
3251 : :
3252 : : /* fault is handled */
3253 : : ret = 0;
3254 : : count_vm_event(THP_FILE_MAPPED);
3255 : : out:
3256 : : spin_unlock(vmf->ptl);
3257 : : return ret;
3258 : : }
3259 : : #else
3260 : : static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
3261 : : {
3262 : : BUILD_BUG();
3263 : : return 0;
3264 : : }
3265 : : #endif
3266 : :
3267 : : /**
3268 : : * alloc_set_pte - setup new PTE entry for given page and add reverse page
3269 : : * mapping. If needed, the fucntion allocates page table or use pre-allocated.
3270 : : *
3271 : : * @vmf: fault environment
3272 : : * @memcg: memcg to charge page (only for private mappings)
3273 : : * @page: page to map
3274 : : *
3275 : : * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on
3276 : : * return.
3277 : : *
3278 : : * Target users are page handler itself and implementations of
3279 : : * vm_ops->map_pages.
3280 : : *
3281 : : * Return: %0 on success, %VM_FAULT_ code in case of error.
3282 : : */
3283 : 183730018 : vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
3284 : : struct page *page)
3285 : : {
3286 : 183730018 : struct vm_area_struct *vma = vmf->vma;
3287 : 183730018 : bool write = vmf->flags & FAULT_FLAG_WRITE;
3288 : : pte_t entry;
3289 : : vm_fault_t ret;
3290 : :
3291 : : if (pmd_none(*vmf->pmd) && PageTransCompound(page) &&
3292 : : IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
3293 : : /* THP on COW? */
3294 : : VM_BUG_ON_PAGE(memcg, page);
3295 : :
3296 : : ret = do_set_pmd(vmf, page);
3297 : : if (ret != VM_FAULT_FALLBACK)
3298 : : return ret;
3299 : : }
3300 : :
3301 [ + + ]: 183730018 : if (!vmf->pte) {
3302 : 18391394 : ret = pte_alloc_one_map(vmf);
3303 [ + + ]: 18409631 : if (ret)
3304 : : return ret;
3305 : : }
3306 : :
3307 : : /* Re-check under ptl */
3308 [ + + ]: 183729671 : if (unlikely(!pte_none(*vmf->pte)))
3309 : : return VM_FAULT_NOPAGE;
3310 : :
3311 : : flush_icache_page(vma, page);
3312 : 170263133 : entry = mk_pte(page, vma->vm_page_prot);
3313 [ + + ]: 170263133 : if (write)
3314 : : entry = maybe_mkwrite(pte_mkdirty(entry), vma);
3315 : : /* copy-on-write page */
3316 [ + + + + ]: 170263133 : if (write && !(vma->vm_flags & VM_SHARED)) {
3317 : 1722303 : inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
3318 : 1730222 : page_add_new_anon_rmap(page, vma, vmf->address, false);
3319 : 1731416 : mem_cgroup_commit_charge(page, memcg, false, false);
3320 : 1727040 : lru_cache_add_active_or_unevictable(page, vma);
3321 : : } else {
3322 : 168540830 : inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
3323 : 168541231 : page_add_file_rmap(page, false);
3324 : : }
3325 : 170280505 : set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
3326 : :
3327 : : /* no need to invalidate: a not-present page won't be cached */
3328 : : update_mmu_cache(vma, vmf->address, vmf->pte);
3329 : :
3330 : 170260015 : return 0;
3331 : : }
3332 : :
3333 : :
3334 : : /**
3335 : : * finish_fault - finish page fault once we have prepared the page to fault
3336 : : *
3337 : : * @vmf: structure describing the fault
3338 : : *
3339 : : * This function handles all that is needed to finish a page fault once the
3340 : : * page to fault in is prepared. It handles locking of PTEs, inserts PTE for
3341 : : * given page, adds reverse page mapping, handles memcg charges and LRU
3342 : : * addition.
3343 : : *
3344 : : * The function expects the page to be locked and on success it consumes a
3345 : : * reference of a page being mapped (for the PTE which maps it).
3346 : : *
3347 : : * Return: %0 on success, %VM_FAULT_ code in case of error.
3348 : : */
3349 : 3725092 : vm_fault_t finish_fault(struct vm_fault *vmf)
3350 : : {
3351 : : struct page *page;
3352 : : vm_fault_t ret = 0;
3353 : :
3354 : : /* Did we COW the page? */
3355 [ + + + + ]: 6342425 : if ((vmf->flags & FAULT_FLAG_WRITE) &&
3356 : 2617333 : !(vmf->vma->vm_flags & VM_SHARED))
3357 : 1722087 : page = vmf->cow_page;
3358 : : else
3359 : 2003005 : page = vmf->page;
3360 : :
3361 : : /*
3362 : : * check even for read faults because we might have lost our CoWed
3363 : : * page
3364 : : */
3365 [ + + ]: 3725092 : if (!(vmf->vma->vm_flags & VM_SHARED))
3366 : 2001772 : ret = check_stable_address_space(vmf->vma->vm_mm);
3367 [ + - ]: 3725092 : if (!ret)
3368 : 3735623 : ret = alloc_set_pte(vmf, vmf->memcg, page);
3369 [ + - ]: 3732771 : if (vmf->pte)
3370 : 3732771 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3371 : 3738828 : return ret;
3372 : : }
3373 : :
3374 : : static unsigned long fault_around_bytes __read_mostly =
3375 : : rounddown_pow_of_two(65536);
3376 : :
3377 : : #ifdef CONFIG_DEBUG_FS
3378 : 0 : static int fault_around_bytes_get(void *data, u64 *val)
3379 : : {
3380 : 0 : *val = fault_around_bytes;
3381 : 0 : return 0;
3382 : : }
3383 : :
3384 : : /*
3385 : : * fault_around_bytes must be rounded down to the nearest page order as it's
3386 : : * what do_fault_around() expects to see.
3387 : : */
3388 : 0 : static int fault_around_bytes_set(void *data, u64 val)
3389 : : {
3390 [ # # ]: 0 : if (val / PAGE_SIZE > PTRS_PER_PTE)
3391 : : return -EINVAL;
3392 [ # # ]: 0 : if (val > PAGE_SIZE)
3393 [ # # # # : 0 : fault_around_bytes = rounddown_pow_of_two(val);
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # ]
3394 : : else
3395 : 0 : fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
3396 : : return 0;
3397 : : }
3398 : 0 : DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops,
3399 : : fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
3400 : :
3401 : 207 : static int __init fault_around_debugfs(void)
3402 : : {
3403 : 207 : debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL,
3404 : : &fault_around_bytes_fops);
3405 : 207 : return 0;
3406 : : }
3407 : : late_initcall(fault_around_debugfs);
3408 : : #endif
3409 : :
3410 : : /*
3411 : : * do_fault_around() tries to map few pages around the fault address. The hope
3412 : : * is that the pages will be needed soon and this will lower the number of
3413 : : * faults to handle.
3414 : : *
3415 : : * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
3416 : : * not ready to be mapped: not up-to-date, locked, etc.
3417 : : *
3418 : : * This function is called with the page table lock taken. In the split ptlock
3419 : : * case the page table lock only protects only those entries which belong to
3420 : : * the page table corresponding to the fault address.
3421 : : *
3422 : : * This function doesn't cross the VMA boundaries, in order to call map_pages()
3423 : : * only once.
3424 : : *
3425 : : * fault_around_bytes defines how many bytes we'll try to map.
3426 : : * do_fault_around() expects it to be set to a power of two less than or equal
3427 : : * to PTRS_PER_PTE.
3428 : : *
3429 : : * The virtual address of the area that we map is naturally aligned to
3430 : : * fault_around_bytes rounded down to the machine page size
3431 : : * (and therefore to page order). This way it's easier to guarantee
3432 : : * that we don't cross page table boundaries.
3433 : : */
3434 : 14743370 : static vm_fault_t do_fault_around(struct vm_fault *vmf)
3435 : : {
3436 : 14743370 : unsigned long address = vmf->address, nr_pages, mask;
3437 : 14743370 : pgoff_t start_pgoff = vmf->pgoff;
3438 : : pgoff_t end_pgoff;
3439 : : int off;
3440 : : vm_fault_t ret = 0;
3441 : :
3442 : 14743370 : nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
3443 : 14743370 : mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
3444 : :
3445 : 14743370 : vmf->address = max(address & mask, vmf->vma->vm_start);
3446 : 14743370 : off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
3447 : 14743370 : start_pgoff -= off;
3448 : :
3449 : : /*
3450 : : * end_pgoff is either the end of the page table, the end of
3451 : : * the vma or nr_pages from start_pgoff, depending what is nearest.
3452 : : */
3453 : 14743370 : end_pgoff = start_pgoff -
3454 : 14743370 : ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
3455 : : PTRS_PER_PTE - 1;
3456 : 14743370 : end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1,
3457 : : start_pgoff + nr_pages - 1);
3458 : :
3459 [ + + ]: 14743370 : if (pmd_none(*vmf->pmd)) {
3460 : 662691 : vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm);
3461 [ + - ]: 662615 : if (!vmf->prealloc_pte)
3462 : : goto out;
3463 : 662766 : smp_wmb(); /* See comment in __pte_alloc() */
3464 : : }
3465 : :
3466 : 14743253 : vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff);
3467 : :
3468 : : /* Huge page is mapped? Page fault is solved */
3469 : : if (pmd_trans_huge(*vmf->pmd)) {
3470 : : ret = VM_FAULT_NOPAGE;
3471 : : goto out;
3472 : : }
3473 : :
3474 : : /* ->map_pages() haven't done anything useful. Cold page cache? */
3475 [ + + ]: 14759426 : if (!vmf->pte)
3476 : : goto out;
3477 : :
3478 : : /* check if the page fault is solved */
3479 : 14675210 : vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT);
3480 [ + + ]: 14675210 : if (!pte_none(*vmf->pte))
3481 : : ret = VM_FAULT_NOPAGE;
3482 : 14675210 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3483 : : out:
3484 : 14762470 : vmf->address = address;
3485 : 14762470 : vmf->pte = NULL;
3486 : 14762470 : return ret;
3487 : : }
3488 : :
3489 : 15869305 : static vm_fault_t do_read_fault(struct vm_fault *vmf)
3490 : : {
3491 : 15869305 : struct vm_area_struct *vma = vmf->vma;
3492 : : vm_fault_t ret = 0;
3493 : :
3494 : : /*
3495 : : * Let's call ->map_pages() first and use ->fault() as fallback
3496 : : * if page by the offset is not ready to be mapped (cold cache or
3497 : : * something).
3498 : : */
3499 [ + + + + ]: 15869305 : if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
3500 : 14744434 : ret = do_fault_around(vmf);
3501 [ + + ]: 14759631 : if (ret)
3502 : : return ret;
3503 : : }
3504 : :
3505 : 1457622 : ret = __do_fault(vmf);
3506 [ + + ]: 1445743 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3507 : : return ret;
3508 : :
3509 : 1118296 : ret |= finish_fault(vmf);
3510 : 1118301 : unlock_page(vmf->page);
3511 [ - + ]: 1118298 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3512 : 0 : put_page(vmf->page);
3513 : 1118300 : return ret;
3514 : : }
3515 : :
3516 : 1760527 : static vm_fault_t do_cow_fault(struct vm_fault *vmf)
3517 : : {
3518 : 1760527 : struct vm_area_struct *vma = vmf->vma;
3519 : : vm_fault_t ret;
3520 : :
3521 [ + - ]: 1764456 : if (unlikely(anon_vma_prepare(vma)))
3522 : : return VM_FAULT_OOM;
3523 : :
3524 : 1760882 : vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address);
3525 [ + + ]: 1760882 : if (!vmf->cow_page)
3526 : : return VM_FAULT_OOM;
3527 : :
3528 [ - + ]: 1757517 : if (mem_cgroup_try_charge_delay(vmf->cow_page, vma->vm_mm, GFP_KERNEL,
3529 : : &vmf->memcg, false)) {
3530 : 0 : put_page(vmf->cow_page);
3531 : 0 : return VM_FAULT_OOM;
3532 : : }
3533 : :
3534 : 1756158 : ret = __do_fault(vmf);
3535 [ + + ]: 1755798 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3536 : : goto uncharge_out;
3537 [ + + ]: 1727849 : if (ret & VM_FAULT_DONE_COW)
3538 : : return ret;
3539 : :
3540 : 1718687 : copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma);
3541 : 1736873 : __SetPageUptodate(vmf->cow_page);
3542 : :
3543 : 1731340 : ret |= finish_fault(vmf);
3544 : 1729398 : unlock_page(vmf->page);
3545 : 1735081 : put_page(vmf->page);
3546 [ + + ]: 1737448 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3547 : : goto uncharge_out;
3548 : : return ret;
3549 : : uncharge_out:
3550 : 34340 : mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false);
3551 : 26683 : put_page(vmf->cow_page);
3552 : 26684 : return ret;
3553 : : }
3554 : :
3555 : 884731 : static vm_fault_t do_shared_fault(struct vm_fault *vmf)
3556 : : {
3557 : 884731 : struct vm_area_struct *vma = vmf->vma;
3558 : : vm_fault_t ret, tmp;
3559 : :
3560 : 884731 : ret = __do_fault(vmf);
3561 [ + - ]: 884731 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3562 : : return ret;
3563 : :
3564 : : /*
3565 : : * Check if the backing address space wants to know that the page is
3566 : : * about to become writable
3567 : : */
3568 [ + + ]: 884731 : if (vma->vm_ops->page_mkwrite) {
3569 : 990 : unlock_page(vmf->page);
3570 : 990 : tmp = do_page_mkwrite(vmf);
3571 [ + - - + ]: 990 : if (unlikely(!tmp ||
3572 : : (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
3573 : 0 : put_page(vmf->page);
3574 : 0 : return tmp;
3575 : : }
3576 : : }
3577 : :
3578 : 884731 : ret |= finish_fault(vmf);
3579 [ - + ]: 884731 : if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
3580 : : VM_FAULT_RETRY))) {
3581 : 0 : unlock_page(vmf->page);
3582 : 0 : put_page(vmf->page);
3583 : 0 : return ret;
3584 : : }
3585 : :
3586 : 884731 : ret |= fault_dirty_shared_page(vmf);
3587 : 884731 : return ret;
3588 : : }
3589 : :
3590 : : /*
3591 : : * We enter with non-exclusive mmap_sem (to exclude vma changes,
3592 : : * but allow concurrent faults).
3593 : : * The mmap_sem may have been released depending on flags and our
3594 : : * return value. See filemap_fault() and __lock_page_or_retry().
3595 : : * If mmap_sem is released, vma may become invalid (for example
3596 : : * by other thread calling munmap()).
3597 : : */
3598 : 18503567 : static vm_fault_t do_fault(struct vm_fault *vmf)
3599 : : {
3600 : 18503567 : struct vm_area_struct *vma = vmf->vma;
3601 : 18503567 : struct mm_struct *vm_mm = vma->vm_mm;
3602 : : vm_fault_t ret;
3603 : :
3604 : : /*
3605 : : * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND
3606 : : */
3607 [ - + ]: 18503567 : if (!vma->vm_ops->fault) {
3608 : : /*
3609 : : * If we find a migration pmd entry or a none pmd entry, which
3610 : : * should never happen, return SIGBUS
3611 : : */
3612 [ # # ]: 0 : if (unlikely(!pmd_present(*vmf->pmd)))
3613 : : ret = VM_FAULT_SIGBUS;
3614 : : else {
3615 : 0 : vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm,
3616 : : vmf->pmd,
3617 : : vmf->address,
3618 : : &vmf->ptl);
3619 : : /*
3620 : : * Make sure this is not a temporary clearing of pte
3621 : : * by holding ptl and checking again. A R/M/W update
3622 : : * of pte involves: take ptl, clearing the pte so that
3623 : : * we don't have concurrent modification by hardware
3624 : : * followed by an update.
3625 : : */
3626 [ # # ]: 0 : if (unlikely(pte_none(*vmf->pte)))
3627 : : ret = VM_FAULT_SIGBUS;
3628 : : else
3629 : : ret = VM_FAULT_NOPAGE;
3630 : :
3631 : 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3632 : : }
3633 [ + + ]: 18503567 : } else if (!(vmf->flags & FAULT_FLAG_WRITE))
3634 : 15867377 : ret = do_read_fault(vmf);
3635 [ + + ]: 2636190 : else if (!(vma->vm_flags & VM_SHARED))
3636 : 1751459 : ret = do_cow_fault(vmf);
3637 : : else
3638 : 884731 : ret = do_shared_fault(vmf);
3639 : :
3640 : : /* preallocated pagetable is unused: free it */
3641 [ + + ]: 18519108 : if (vmf->prealloc_pte) {
3642 : 35538 : pte_free(vm_mm, vmf->prealloc_pte);
3643 : 35536 : vmf->prealloc_pte = NULL;
3644 : : }
3645 : 18519106 : return ret;
3646 : : }
3647 : :
3648 : : static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
3649 : : unsigned long addr, int page_nid,
3650 : : int *flags)
3651 : : {
3652 : : get_page(page);
3653 : :
3654 : : count_vm_numa_event(NUMA_HINT_FAULTS);
3655 : : if (page_nid == numa_node_id()) {
3656 : : count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
3657 : : *flags |= TNF_FAULT_LOCAL;
3658 : : }
3659 : :
3660 : : return mpol_misplaced(page, vma, addr);
3661 : : }
3662 : :
3663 : : static vm_fault_t do_numa_page(struct vm_fault *vmf)
3664 : : {
3665 : : struct vm_area_struct *vma = vmf->vma;
3666 : : struct page *page = NULL;
3667 : : int page_nid = NUMA_NO_NODE;
3668 : : int last_cpupid;
3669 : : int target_nid;
3670 : : bool migrated = false;
3671 : : pte_t pte, old_pte;
3672 : : bool was_writable = pte_savedwrite(vmf->orig_pte);
3673 : : int flags = 0;
3674 : :
3675 : : /*
3676 : : * The "pte" at this point cannot be used safely without
3677 : : * validation through pte_unmap_same(). It's of NUMA type but
3678 : : * the pfn may be screwed if the read is non atomic.
3679 : : */
3680 : : vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd);
3681 : : spin_lock(vmf->ptl);
3682 : : if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
3683 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3684 : : goto out;
3685 : : }
3686 : :
3687 : : /*
3688 : : * Make it present again, Depending on how arch implementes non
3689 : : * accessible ptes, some can allow access by kernel mode.
3690 : : */
3691 : : old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte);
3692 : : pte = pte_modify(old_pte, vma->vm_page_prot);
3693 : : pte = pte_mkyoung(pte);
3694 : : if (was_writable)
3695 : : pte = pte_mkwrite(pte);
3696 : : ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte);
3697 : : update_mmu_cache(vma, vmf->address, vmf->pte);
3698 : :
3699 : : page = vm_normal_page(vma, vmf->address, pte);
3700 : : if (!page) {
3701 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3702 : : return 0;
3703 : : }
3704 : :
3705 : : /* TODO: handle PTE-mapped THP */
3706 : : if (PageCompound(page)) {
3707 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3708 : : return 0;
3709 : : }
3710 : :
3711 : : /*
3712 : : * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3713 : : * much anyway since they can be in shared cache state. This misses
3714 : : * the case where a mapping is writable but the process never writes
3715 : : * to it but pte_write gets cleared during protection updates and
3716 : : * pte_dirty has unpredictable behaviour between PTE scan updates,
3717 : : * background writeback, dirty balancing and application behaviour.
3718 : : */
3719 : : if (!pte_write(pte))
3720 : : flags |= TNF_NO_GROUP;
3721 : :
3722 : : /*
3723 : : * Flag if the page is shared between multiple address spaces. This
3724 : : * is later used when determining whether to group tasks together
3725 : : */
3726 : : if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3727 : : flags |= TNF_SHARED;
3728 : :
3729 : : last_cpupid = page_cpupid_last(page);
3730 : : page_nid = page_to_nid(page);
3731 : : target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid,
3732 : : &flags);
3733 : : pte_unmap_unlock(vmf->pte, vmf->ptl);
3734 : : if (target_nid == NUMA_NO_NODE) {
3735 : : put_page(page);
3736 : : goto out;
3737 : : }
3738 : :
3739 : : /* Migrate to the requested node */
3740 : : migrated = migrate_misplaced_page(page, vma, target_nid);
3741 : : if (migrated) {
3742 : : page_nid = target_nid;
3743 : : flags |= TNF_MIGRATED;
3744 : : } else
3745 : : flags |= TNF_MIGRATE_FAIL;
3746 : :
3747 : : out:
3748 : : if (page_nid != NUMA_NO_NODE)
3749 : : task_numa_fault(last_cpupid, page_nid, 1, flags);
3750 : : return 0;
3751 : : }
3752 : :
3753 : : static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf)
3754 : : {
3755 : : if (vma_is_anonymous(vmf->vma))
3756 : : return do_huge_pmd_anonymous_page(vmf);
3757 : : if (vmf->vma->vm_ops->huge_fault)
3758 : : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
3759 : : return VM_FAULT_FALLBACK;
3760 : : }
3761 : :
3762 : : /* `inline' is required to avoid gcc 4.1.2 build error */
3763 : : static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd)
3764 : : {
3765 : : if (vma_is_anonymous(vmf->vma))
3766 : : return do_huge_pmd_wp_page(vmf, orig_pmd);
3767 : : if (vmf->vma->vm_ops->huge_fault)
3768 : : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
3769 : :
3770 : : /* COW handled on pte level: split pmd */
3771 : : VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma);
3772 : : __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
3773 : :
3774 : : return VM_FAULT_FALLBACK;
3775 : : }
3776 : :
3777 : : static inline bool vma_is_accessible(struct vm_area_struct *vma)
3778 : : {
3779 : : return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE);
3780 : : }
3781 : :
3782 : : static vm_fault_t create_huge_pud(struct vm_fault *vmf)
3783 : : {
3784 : : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3785 : : /* No support for anonymous transparent PUD pages yet */
3786 : : if (vma_is_anonymous(vmf->vma))
3787 : : return VM_FAULT_FALLBACK;
3788 : : if (vmf->vma->vm_ops->huge_fault)
3789 : : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
3790 : : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3791 : : return VM_FAULT_FALLBACK;
3792 : : }
3793 : :
3794 : : static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud)
3795 : : {
3796 : : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3797 : : /* No support for anonymous transparent PUD pages yet */
3798 : : if (vma_is_anonymous(vmf->vma))
3799 : : return VM_FAULT_FALLBACK;
3800 : : if (vmf->vma->vm_ops->huge_fault)
3801 : : return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
3802 : : #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3803 : : return VM_FAULT_FALLBACK;
3804 : : }
3805 : :
3806 : : /*
3807 : : * These routines also need to handle stuff like marking pages dirty
3808 : : * and/or accessed for architectures that don't do it in hardware (most
3809 : : * RISC architectures). The early dirtying is also good on the i386.
3810 : : *
3811 : : * There is also a hook called "update_mmu_cache()" that architectures
3812 : : * with external mmu caches can use to update those (ie the Sparc or
3813 : : * PowerPC hashed page tables that act as extended TLBs).
3814 : : *
3815 : : * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow
3816 : : * concurrent faults).
3817 : : *
3818 : : * The mmap_sem may have been released depending on flags and our return value.
3819 : : * See filemap_fault() and __lock_page_or_retry().
3820 : : */
3821 : 46560862 : static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
3822 : : {
3823 : : pte_t entry;
3824 : :
3825 [ + + ]: 46560862 : if (unlikely(pmd_none(*vmf->pmd))) {
3826 : : /*
3827 : : * Leave __pte_alloc() until later: because vm_ops->fault may
3828 : : * want to allocate huge page, and if we expose page table
3829 : : * for an instant, it will be difficult to retract from
3830 : : * concurrent faults and from rmap lookups.
3831 : : */
3832 : 1816311 : vmf->pte = NULL;
3833 : : } else {
3834 : : /* See comment in pte_alloc_one_map() */
3835 : : if (pmd_devmap_trans_unstable(vmf->pmd))
3836 : : return 0;
3837 : : /*
3838 : : * A regular pmd is established and it can't morph into a huge
3839 : : * pmd from under us anymore at this point because we hold the
3840 : : * mmap_sem read mode and khugepaged takes it in write mode.
3841 : : * So now it's safe to run pte_offset_map().
3842 : : */
3843 : 44744551 : vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
3844 : 44744551 : vmf->orig_pte = *vmf->pte;
3845 : :
3846 : : /*
3847 : : * some architectures can have larger ptes than wordsize,
3848 : : * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and
3849 : : * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic
3850 : : * accesses. The code below just needs a consistent view
3851 : : * for the ifs and we later double check anyway with the
3852 : : * ptl lock held. So here a barrier will do.
3853 : : */
3854 : 44744551 : barrier();
3855 [ + + ]: 44791295 : if (pte_none(vmf->orig_pte)) {
3856 : : pte_unmap(vmf->pte);
3857 : 27176607 : vmf->pte = NULL;
3858 : : }
3859 : : }
3860 : :
3861 [ + + ]: 46607606 : if (!vmf->pte) {
3862 [ + + ]: 58011436 : if (vma_is_anonymous(vmf->vma))
3863 : 10486689 : return do_anonymous_page(vmf);
3864 : : else
3865 : 18519029 : return do_fault(vmf);
3866 : : }
3867 : :
3868 [ - + ]: 17601888 : if (!pte_present(vmf->orig_pte))
3869 : 0 : return do_swap_page(vmf);
3870 : :
3871 : : if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
3872 : : return do_numa_page(vmf);
3873 : :
3874 : 35203776 : vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
3875 : : spin_lock(vmf->ptl);
3876 : 17600313 : entry = vmf->orig_pte;
3877 [ + + ]: 35200626 : if (unlikely(!pte_same(*vmf->pte, entry)))
3878 : : goto unlock;
3879 [ + + ]: 17600049 : if (vmf->flags & FAULT_FLAG_WRITE) {
3880 [ + + ]: 13311365 : if (!pte_write(entry))
3881 : 11625943 : return do_wp_page(vmf);
3882 : : entry = pte_mkdirty(entry);
3883 : : }
3884 : : entry = pte_mkyoung(entry);
3885 [ + + ]: 5974106 : if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry,
3886 : : vmf->flags & FAULT_FLAG_WRITE)) {
3887 : : update_mmu_cache(vmf->vma, vmf->address, vmf->pte);
3888 : : } else {
3889 : : /*
3890 : : * This is needed only for protection faults but the arch code
3891 : : * is not yet telling us if this is a protection fault or not.
3892 : : * This still avoids useless tlb flushes for .text page faults
3893 : : * with threads.
3894 : : */
3895 [ + + ]: 846889 : if (vmf->flags & FAULT_FLAG_WRITE)
3896 : 846865 : flush_tlb_fix_spurious_fault(vmf->vma, vmf->address);
3897 : : }
3898 : : unlock:
3899 : 5975827 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3900 : 5975715 : return 0;
3901 : : }
3902 : :
3903 : : /*
3904 : : * By the time we get here, we already hold the mm semaphore
3905 : : *
3906 : : * The mmap_sem may have been released depending on flags and our
3907 : : * return value. See filemap_fault() and __lock_page_or_retry().
3908 : : */
3909 : 46599698 : static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma,
3910 : : unsigned long address, unsigned int flags)
3911 : : {
3912 : 186398792 : struct vm_fault vmf = {
3913 : : .vma = vma,
3914 : 46599698 : .address = address & PAGE_MASK,
3915 : : .flags = flags,
3916 : : .pgoff = linear_page_index(vma, address),
3917 : : .gfp_mask = __get_fault_gfp_mask(vma),
3918 : : };
3919 : : unsigned int dirty = flags & FAULT_FLAG_WRITE;
3920 : 46599698 : struct mm_struct *mm = vma->vm_mm;
3921 : : pgd_t *pgd;
3922 : : p4d_t *p4d;
3923 : : vm_fault_t ret;
3924 : :
3925 : 46599698 : pgd = pgd_offset(mm, address);
3926 : : p4d = p4d_alloc(mm, pgd, address);
3927 [ + + ]: 46599698 : if (!p4d)
3928 : : return VM_FAULT_OOM;
3929 : :
3930 : 46578714 : vmf.pud = pud_alloc(mm, p4d, address);
3931 [ + + ]: 46578714 : if (!vmf.pud)
3932 : : return VM_FAULT_OOM;
3933 : : if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) {
3934 : : ret = create_huge_pud(&vmf);
3935 : : if (!(ret & VM_FAULT_FALLBACK))
3936 : : return ret;
3937 : : } else {
3938 : : pud_t orig_pud = *vmf.pud;
3939 : :
3940 : 46579107 : barrier();
3941 : : if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
3942 : :
3943 : : /* NUMA case for anonymous PUDs would go here */
3944 : :
3945 : : if (dirty && !pud_write(orig_pud)) {
3946 : : ret = wp_huge_pud(&vmf, orig_pud);
3947 : : if (!(ret & VM_FAULT_FALLBACK))
3948 : : return ret;
3949 : : } else {
3950 : : huge_pud_set_accessed(&vmf, orig_pud);
3951 : : return 0;
3952 : : }
3953 : : }
3954 : : }
3955 : :
3956 : 46592160 : vmf.pmd = pmd_alloc(mm, vmf.pud, address);
3957 [ + + ]: 46592160 : if (!vmf.pmd)
3958 : : return VM_FAULT_OOM;
3959 : : if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) {
3960 : : ret = create_huge_pmd(&vmf);
3961 : : if (!(ret & VM_FAULT_FALLBACK))
3962 : : return ret;
3963 : : } else {
3964 : : pmd_t orig_pmd = *vmf.pmd;
3965 : :
3966 : 46578294 : barrier();
3967 : : if (unlikely(is_swap_pmd(orig_pmd))) {
3968 : : VM_BUG_ON(thp_migration_supported() &&
3969 : : !is_pmd_migration_entry(orig_pmd));
3970 : : if (is_pmd_migration_entry(orig_pmd))
3971 : : pmd_migration_entry_wait(mm, vmf.pmd);
3972 : : return 0;
3973 : : }
3974 : : if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
3975 : : if (pmd_protnone(orig_pmd) && vma_is_accessible(vma))
3976 : : return do_huge_pmd_numa_page(&vmf, orig_pmd);
3977 : :
3978 : : if (dirty && !pmd_write(orig_pmd)) {
3979 : : ret = wp_huge_pmd(&vmf, orig_pmd);
3980 : : if (!(ret & VM_FAULT_FALLBACK))
3981 : : return ret;
3982 : : } else {
3983 : : huge_pmd_set_accessed(&vmf, orig_pmd);
3984 : : return 0;
3985 : : }
3986 : : }
3987 : : }
3988 : :
3989 : 46597404 : return handle_pte_fault(&vmf);
3990 : : }
3991 : :
3992 : : /*
3993 : : * By the time we get here, we already hold the mm semaphore
3994 : : *
3995 : : * The mmap_sem may have been released depending on flags and our
3996 : : * return value. See filemap_fault() and __lock_page_or_retry().
3997 : : */
3998 : 46599429 : vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
3999 : : unsigned int flags)
4000 : : {
4001 : : vm_fault_t ret;
4002 : :
4003 : 46599429 : __set_current_state(TASK_RUNNING);
4004 : :
4005 : : count_vm_event(PGFAULT);
4006 : 46589296 : count_memcg_event_mm(vma->vm_mm, PGFAULT);
4007 : :
4008 : : /* do counter updates before entering really critical section. */
4009 : 46567354 : check_sync_rss_stat(current);
4010 : :
4011 : : if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
4012 : : flags & FAULT_FLAG_INSTRUCTION,
4013 : : flags & FAULT_FLAG_REMOTE))
4014 : : return VM_FAULT_SIGSEGV;
4015 : :
4016 : : /*
4017 : : * Enable the memcg OOM handling for faults triggered in user
4018 : : * space. Kernel faults are handled more gracefully.
4019 : : */
4020 [ + + ]: 46610676 : if (flags & FAULT_FLAG_USER)
4021 : 44201648 : mem_cgroup_enter_user_fault();
4022 : :
4023 : : if (unlikely(is_vm_hugetlb_page(vma)))
4024 : : ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
4025 : : else
4026 : 46622350 : ret = __handle_mm_fault(vma, address, flags);
4027 : :
4028 [ + + ]: 46605973 : if (flags & FAULT_FLAG_USER) {
4029 : 44195341 : mem_cgroup_exit_user_fault();
4030 : : /*
4031 : : * The task may have entered a memcg OOM situation but
4032 : : * if the allocation error was handled gracefully (no
4033 : : * VM_FAULT_OOM), there is no need to kill anything.
4034 : : * Just clean up the OOM state peacefully.
4035 : : */
4036 [ - + # # ]: 88423264 : if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
4037 : 0 : mem_cgroup_oom_synchronize(false);
4038 : : }
4039 : :
4040 : : return ret;
4041 : : }
4042 : : EXPORT_SYMBOL_GPL(handle_mm_fault);
4043 : :
4044 : : #ifndef __PAGETABLE_P4D_FOLDED
4045 : : /*
4046 : : * Allocate p4d page table.
4047 : : * We've already handled the fast-path in-line.
4048 : : */
4049 : : int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
4050 : : {
4051 : : p4d_t *new = p4d_alloc_one(mm, address);
4052 : : if (!new)
4053 : : return -ENOMEM;
4054 : :
4055 : : smp_wmb(); /* See comment in __pte_alloc */
4056 : :
4057 : : spin_lock(&mm->page_table_lock);
4058 : : if (pgd_present(*pgd)) /* Another has populated it */
4059 : : p4d_free(mm, new);
4060 : : else
4061 : : pgd_populate(mm, pgd, new);
4062 : : spin_unlock(&mm->page_table_lock);
4063 : : return 0;
4064 : : }
4065 : : #endif /* __PAGETABLE_P4D_FOLDED */
4066 : :
4067 : : #ifndef __PAGETABLE_PUD_FOLDED
4068 : : /*
4069 : : * Allocate page upper directory.
4070 : : * We've already handled the fast-path in-line.
4071 : : */
4072 : : int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address)
4073 : : {
4074 : : pud_t *new = pud_alloc_one(mm, address);
4075 : : if (!new)
4076 : : return -ENOMEM;
4077 : :
4078 : : smp_wmb(); /* See comment in __pte_alloc */
4079 : :
4080 : : spin_lock(&mm->page_table_lock);
4081 : : #ifndef __ARCH_HAS_5LEVEL_HACK
4082 : : if (!p4d_present(*p4d)) {
4083 : : mm_inc_nr_puds(mm);
4084 : : p4d_populate(mm, p4d, new);
4085 : : } else /* Another has populated it */
4086 : : pud_free(mm, new);
4087 : : #else
4088 : : if (!pgd_present(*p4d)) {
4089 : : mm_inc_nr_puds(mm);
4090 : : pgd_populate(mm, p4d, new);
4091 : : } else /* Another has populated it */
4092 : : pud_free(mm, new);
4093 : : #endif /* __ARCH_HAS_5LEVEL_HACK */
4094 : : spin_unlock(&mm->page_table_lock);
4095 : : return 0;
4096 : : }
4097 : : #endif /* __PAGETABLE_PUD_FOLDED */
4098 : :
4099 : : #ifndef __PAGETABLE_PMD_FOLDED
4100 : : /*
4101 : : * Allocate page middle directory.
4102 : : * We've already handled the fast-path in-line.
4103 : : */
4104 : : int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
4105 : : {
4106 : : spinlock_t *ptl;
4107 : : pmd_t *new = pmd_alloc_one(mm, address);
4108 : : if (!new)
4109 : : return -ENOMEM;
4110 : :
4111 : : smp_wmb(); /* See comment in __pte_alloc */
4112 : :
4113 : : ptl = pud_lock(mm, pud);
4114 : : #ifndef __ARCH_HAS_4LEVEL_HACK
4115 : : if (!pud_present(*pud)) {
4116 : : mm_inc_nr_pmds(mm);
4117 : : pud_populate(mm, pud, new);
4118 : : } else /* Another has populated it */
4119 : : pmd_free(mm, new);
4120 : : #else
4121 : : if (!pgd_present(*pud)) {
4122 : : mm_inc_nr_pmds(mm);
4123 : : pgd_populate(mm, pud, new);
4124 : : } else /* Another has populated it */
4125 : : pmd_free(mm, new);
4126 : : #endif /* __ARCH_HAS_4LEVEL_HACK */
4127 : : spin_unlock(ptl);
4128 : : return 0;
4129 : : }
4130 : : #endif /* __PAGETABLE_PMD_FOLDED */
4131 : :
4132 : 0 : static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address,
4133 : : struct mmu_notifier_range *range,
4134 : : pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
4135 : : {
4136 : : pgd_t *pgd;
4137 : : p4d_t *p4d;
4138 : : pud_t *pud;
4139 : : pmd_t *pmd;
4140 : : pte_t *ptep;
4141 : :
4142 : 0 : pgd = pgd_offset(mm, address);
4143 : : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
4144 : : goto out;
4145 : :
4146 : : p4d = p4d_offset(pgd, address);
4147 : : if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d)))
4148 : : goto out;
4149 : :
4150 : : pud = pud_offset(p4d, address);
4151 : : if (pud_none(*pud) || unlikely(pud_bad(*pud)))
4152 : : goto out;
4153 : :
4154 : : pmd = pmd_offset(pud, address);
4155 : : VM_BUG_ON(pmd_trans_huge(*pmd));
4156 : :
4157 : : if (pmd_huge(*pmd)) {
4158 : : if (!pmdpp)
4159 : : goto out;
4160 : :
4161 : : if (range) {
4162 : : mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0,
4163 : : NULL, mm, address & PMD_MASK,
4164 : : (address & PMD_MASK) + PMD_SIZE);
4165 : : mmu_notifier_invalidate_range_start(range);
4166 : : }
4167 : : *ptlp = pmd_lock(mm, pmd);
4168 : : if (pmd_huge(*pmd)) {
4169 : : *pmdpp = pmd;
4170 : : return 0;
4171 : : }
4172 : : spin_unlock(*ptlp);
4173 : : if (range)
4174 : : mmu_notifier_invalidate_range_end(range);
4175 : : }
4176 : :
4177 [ # # # # ]: 0 : if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
4178 : : goto out;
4179 : :
4180 [ # # ]: 0 : if (range) {
4181 : 0 : mmu_notifier_range_init(range, MMU_NOTIFY_CLEAR, 0, NULL, mm,
4182 : : address & PAGE_MASK,
4183 : : (address & PAGE_MASK) + PAGE_SIZE);
4184 : : mmu_notifier_invalidate_range_start(range);
4185 : : }
4186 : 0 : ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
4187 [ # # ]: 0 : if (!pte_present(*ptep))
4188 : : goto unlock;
4189 : 0 : *ptepp = ptep;
4190 : 0 : return 0;
4191 : : unlock:
4192 : 0 : pte_unmap_unlock(ptep, *ptlp);
4193 : : if (range)
4194 : : mmu_notifier_invalidate_range_end(range);
4195 : : out:
4196 : : return -EINVAL;
4197 : : }
4198 : :
4199 : : static inline int follow_pte(struct mm_struct *mm, unsigned long address,
4200 : : pte_t **ptepp, spinlock_t **ptlp)
4201 : : {
4202 : : int res;
4203 : :
4204 : : /* (void) is needed to make gcc happy */
4205 : 0 : (void) __cond_lock(*ptlp,
4206 : : !(res = __follow_pte_pmd(mm, address, NULL,
4207 : : ptepp, NULL, ptlp)));
4208 : : return res;
4209 : : }
4210 : :
4211 : 0 : int follow_pte_pmd(struct mm_struct *mm, unsigned long address,
4212 : : struct mmu_notifier_range *range,
4213 : : pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
4214 : : {
4215 : : int res;
4216 : :
4217 : : /* (void) is needed to make gcc happy */
4218 : 0 : (void) __cond_lock(*ptlp,
4219 : : !(res = __follow_pte_pmd(mm, address, range,
4220 : : ptepp, pmdpp, ptlp)));
4221 : 0 : return res;
4222 : : }
4223 : : EXPORT_SYMBOL(follow_pte_pmd);
4224 : :
4225 : : /**
4226 : : * follow_pfn - look up PFN at a user virtual address
4227 : : * @vma: memory mapping
4228 : : * @address: user virtual address
4229 : : * @pfn: location to store found PFN
4230 : : *
4231 : : * Only IO mappings and raw PFN mappings are allowed.
4232 : : *
4233 : : * Return: zero and the pfn at @pfn on success, -ve otherwise.
4234 : : */
4235 : 0 : int follow_pfn(struct vm_area_struct *vma, unsigned long address,
4236 : : unsigned long *pfn)
4237 : : {
4238 : : int ret = -EINVAL;
4239 : : spinlock_t *ptl;
4240 : : pte_t *ptep;
4241 : :
4242 [ # # ]: 0 : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4243 : : return ret;
4244 : :
4245 : 0 : ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
4246 [ # # ]: 0 : if (ret)
4247 : : return ret;
4248 : 0 : *pfn = pte_pfn(*ptep);
4249 : 0 : pte_unmap_unlock(ptep, ptl);
4250 : 0 : return 0;
4251 : : }
4252 : : EXPORT_SYMBOL(follow_pfn);
4253 : :
4254 : : #ifdef CONFIG_HAVE_IOREMAP_PROT
4255 : : int follow_phys(struct vm_area_struct *vma,
4256 : : unsigned long address, unsigned int flags,
4257 : : unsigned long *prot, resource_size_t *phys)
4258 : : {
4259 : : int ret = -EINVAL;
4260 : : pte_t *ptep, pte;
4261 : : spinlock_t *ptl;
4262 : :
4263 : : if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4264 : : goto out;
4265 : :
4266 : : if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
4267 : : goto out;
4268 : : pte = *ptep;
4269 : :
4270 : : if ((flags & FOLL_WRITE) && !pte_write(pte))
4271 : : goto unlock;
4272 : :
4273 : : *prot = pgprot_val(pte_pgprot(pte));
4274 : : *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
4275 : :
4276 : : ret = 0;
4277 : : unlock:
4278 : : pte_unmap_unlock(ptep, ptl);
4279 : : out:
4280 : : return ret;
4281 : : }
4282 : :
4283 : : int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
4284 : : void *buf, int len, int write)
4285 : : {
4286 : : resource_size_t phys_addr;
4287 : : unsigned long prot = 0;
4288 : : void __iomem *maddr;
4289 : : int offset = addr & (PAGE_SIZE-1);
4290 : :
4291 : : if (follow_phys(vma, addr, write, &prot, &phys_addr))
4292 : : return -EINVAL;
4293 : :
4294 : : maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
4295 : : if (!maddr)
4296 : : return -ENOMEM;
4297 : :
4298 : : if (write)
4299 : : memcpy_toio(maddr + offset, buf, len);
4300 : : else
4301 : : memcpy_fromio(buf, maddr + offset, len);
4302 : : iounmap(maddr);
4303 : :
4304 : : return len;
4305 : : }
4306 : : EXPORT_SYMBOL_GPL(generic_access_phys);
4307 : : #endif
4308 : :
4309 : : /*
4310 : : * Access another process' address space as given in mm. If non-NULL, use the
4311 : : * given task for page fault accounting.
4312 : : */
4313 : 71538 : int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
4314 : : unsigned long addr, void *buf, int len, unsigned int gup_flags)
4315 : : {
4316 : : struct vm_area_struct *vma;
4317 : : void *old_buf = buf;
4318 : 71538 : int write = gup_flags & FOLL_WRITE;
4319 : :
4320 [ + + ]: 71538 : if (down_read_killable(&mm->mmap_sem))
4321 : : return 0;
4322 : :
4323 : : /* ignore errors, just check how much was successfully transferred */
4324 [ + + ]: 143098 : while (len) {
4325 : : int bytes, ret, offset;
4326 : : void *maddr;
4327 : 72837 : struct page *page = NULL;
4328 : :
4329 : 72837 : ret = get_user_pages_remote(tsk, mm, addr, 1,
4330 : : gup_flags, &page, &vma, NULL);
4331 [ + + ]: 72846 : if (ret <= 0) {
4332 : : #ifndef CONFIG_HAVE_IOREMAP_PROT
4333 : : break;
4334 : : #else
4335 : : /*
4336 : : * Check if this is a VM_IO | VM_PFNMAP VMA, which
4337 : : * we can access using slightly different code.
4338 : : */
4339 : : vma = find_vma(mm, addr);
4340 : : if (!vma || vma->vm_start > addr)
4341 : : break;
4342 : : if (vma->vm_ops && vma->vm_ops->access)
4343 : : ret = vma->vm_ops->access(vma, addr, buf,
4344 : : len, write);
4345 : : if (ret <= 0)
4346 : : break;
4347 : : bytes = ret;
4348 : : #endif
4349 : : } else {
4350 : : bytes = len;
4351 : 71542 : offset = addr & (PAGE_SIZE-1);
4352 [ + + ]: 71542 : if (bytes > PAGE_SIZE-offset)
4353 : 1304 : bytes = PAGE_SIZE-offset;
4354 : :
4355 : 71542 : maddr = kmap(page);
4356 [ - + ]: 71546 : if (write) {
4357 : 0 : copy_to_user_page(vma, page, addr,
4358 : : maddr + offset, buf, bytes);
4359 : 0 : set_page_dirty_lock(page);
4360 : : } else {
4361 : 71546 : copy_from_user_page(vma, page, addr,
4362 : : buf, maddr + offset, bytes);
4363 : : }
4364 : : kunmap(page);
4365 : 71525 : put_page(page);
4366 : : }
4367 : 71538 : len -= bytes;
4368 : 71538 : buf += bytes;
4369 : 71538 : addr += bytes;
4370 : : }
4371 : 71565 : up_read(&mm->mmap_sem);
4372 : :
4373 : 71564 : return buf - old_buf;
4374 : : }
4375 : :
4376 : : /**
4377 : : * access_remote_vm - access another process' address space
4378 : : * @mm: the mm_struct of the target address space
4379 : : * @addr: start address to access
4380 : : * @buf: source or destination buffer
4381 : : * @len: number of bytes to transfer
4382 : : * @gup_flags: flags modifying lookup behaviour
4383 : : *
4384 : : * The caller must hold a reference on @mm.
4385 : : *
4386 : : * Return: number of bytes copied from source to destination.
4387 : : */
4388 : 71564 : int access_remote_vm(struct mm_struct *mm, unsigned long addr,
4389 : : void *buf, int len, unsigned int gup_flags)
4390 : : {
4391 : 71564 : return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
4392 : : }
4393 : :
4394 : : /*
4395 : : * Access another process' address space.
4396 : : * Source/target buffer must be kernel space,
4397 : : * Do not walk the page table directly, use get_user_pages
4398 : : */
4399 : 0 : int access_process_vm(struct task_struct *tsk, unsigned long addr,
4400 : : void *buf, int len, unsigned int gup_flags)
4401 : : {
4402 : : struct mm_struct *mm;
4403 : : int ret;
4404 : :
4405 : 0 : mm = get_task_mm(tsk);
4406 [ # # ]: 0 : if (!mm)
4407 : : return 0;
4408 : :
4409 : 0 : ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
4410 : :
4411 : 0 : mmput(mm);
4412 : :
4413 : 0 : return ret;
4414 : : }
4415 : : EXPORT_SYMBOL_GPL(access_process_vm);
4416 : :
4417 : : /*
4418 : : * Print the name of a VMA.
4419 : : */
4420 : 0 : void print_vma_addr(char *prefix, unsigned long ip)
4421 : : {
4422 : 0 : struct mm_struct *mm = current->mm;
4423 : : struct vm_area_struct *vma;
4424 : :
4425 : : /*
4426 : : * we might be running from an atomic context so we cannot sleep
4427 : : */
4428 [ # # ]: 0 : if (!down_read_trylock(&mm->mmap_sem))
4429 : 0 : return;
4430 : :
4431 : 0 : vma = find_vma(mm, ip);
4432 [ # # # # ]: 0 : if (vma && vma->vm_file) {
4433 : : struct file *f = vma->vm_file;
4434 : 0 : char *buf = (char *)__get_free_page(GFP_NOWAIT);
4435 [ # # ]: 0 : if (buf) {
4436 : : char *p;
4437 : :
4438 : 0 : p = file_path(f, buf, PAGE_SIZE);
4439 [ # # ]: 0 : if (IS_ERR(p))
4440 : : p = "?";
4441 : 0 : printk("%s%s[%lx+%lx]", prefix, kbasename(p),
4442 : : vma->vm_start,
4443 : 0 : vma->vm_end - vma->vm_start);
4444 : 0 : free_page((unsigned long)buf);
4445 : : }
4446 : : }
4447 : 0 : up_read(&mm->mmap_sem);
4448 : : }
4449 : :
4450 : : #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
4451 : : void __might_fault(const char *file, int line)
4452 : : {
4453 : : /*
4454 : : * Some code (nfs/sunrpc) uses socket ops on kernel memory while
4455 : : * holding the mmap_sem, this is safe because kernel memory doesn't
4456 : : * get paged out, therefore we'll never actually fault, and the
4457 : : * below annotations will generate false positives.
4458 : : */
4459 : : if (uaccess_kernel())
4460 : : return;
4461 : : if (pagefault_disabled())
4462 : : return;
4463 : : __might_sleep(file, line, 0);
4464 : : #if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
4465 : : if (current->mm)
4466 : : might_lock_read(¤t->mm->mmap_sem);
4467 : : #endif
4468 : : }
4469 : : EXPORT_SYMBOL(__might_fault);
4470 : : #endif
4471 : :
4472 : : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4473 : : /*
4474 : : * Process all subpages of the specified huge page with the specified
4475 : : * operation. The target subpage will be processed last to keep its
4476 : : * cache lines hot.
4477 : : */
4478 : : static inline void process_huge_page(
4479 : : unsigned long addr_hint, unsigned int pages_per_huge_page,
4480 : : void (*process_subpage)(unsigned long addr, int idx, void *arg),
4481 : : void *arg)
4482 : : {
4483 : : int i, n, base, l;
4484 : : unsigned long addr = addr_hint &
4485 : : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
4486 : :
4487 : : /* Process target subpage last to keep its cache lines hot */
4488 : : might_sleep();
4489 : : n = (addr_hint - addr) / PAGE_SIZE;
4490 : : if (2 * n <= pages_per_huge_page) {
4491 : : /* If target subpage in first half of huge page */
4492 : : base = 0;
4493 : : l = n;
4494 : : /* Process subpages at the end of huge page */
4495 : : for (i = pages_per_huge_page - 1; i >= 2 * n; i--) {
4496 : : cond_resched();
4497 : : process_subpage(addr + i * PAGE_SIZE, i, arg);
4498 : : }
4499 : : } else {
4500 : : /* If target subpage in second half of huge page */
4501 : : base = pages_per_huge_page - 2 * (pages_per_huge_page - n);
4502 : : l = pages_per_huge_page - n;
4503 : : /* Process subpages at the begin of huge page */
4504 : : for (i = 0; i < base; i++) {
4505 : : cond_resched();
4506 : : process_subpage(addr + i * PAGE_SIZE, i, arg);
4507 : : }
4508 : : }
4509 : : /*
4510 : : * Process remaining subpages in left-right-left-right pattern
4511 : : * towards the target subpage
4512 : : */
4513 : : for (i = 0; i < l; i++) {
4514 : : int left_idx = base + i;
4515 : : int right_idx = base + 2 * l - 1 - i;
4516 : :
4517 : : cond_resched();
4518 : : process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg);
4519 : : cond_resched();
4520 : : process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg);
4521 : : }
4522 : : }
4523 : :
4524 : : static void clear_gigantic_page(struct page *page,
4525 : : unsigned long addr,
4526 : : unsigned int pages_per_huge_page)
4527 : : {
4528 : : int i;
4529 : : struct page *p = page;
4530 : :
4531 : : might_sleep();
4532 : : for (i = 0; i < pages_per_huge_page;
4533 : : i++, p = mem_map_next(p, page, i)) {
4534 : : cond_resched();
4535 : : clear_user_highpage(p, addr + i * PAGE_SIZE);
4536 : : }
4537 : : }
4538 : :
4539 : : static void clear_subpage(unsigned long addr, int idx, void *arg)
4540 : : {
4541 : : struct page *page = arg;
4542 : :
4543 : : clear_user_highpage(page + idx, addr);
4544 : : }
4545 : :
4546 : : void clear_huge_page(struct page *page,
4547 : : unsigned long addr_hint, unsigned int pages_per_huge_page)
4548 : : {
4549 : : unsigned long addr = addr_hint &
4550 : : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
4551 : :
4552 : : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4553 : : clear_gigantic_page(page, addr, pages_per_huge_page);
4554 : : return;
4555 : : }
4556 : :
4557 : : process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page);
4558 : : }
4559 : :
4560 : : static void copy_user_gigantic_page(struct page *dst, struct page *src,
4561 : : unsigned long addr,
4562 : : struct vm_area_struct *vma,
4563 : : unsigned int pages_per_huge_page)
4564 : : {
4565 : : int i;
4566 : : struct page *dst_base = dst;
4567 : : struct page *src_base = src;
4568 : :
4569 : : for (i = 0; i < pages_per_huge_page; ) {
4570 : : cond_resched();
4571 : : copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
4572 : :
4573 : : i++;
4574 : : dst = mem_map_next(dst, dst_base, i);
4575 : : src = mem_map_next(src, src_base, i);
4576 : : }
4577 : : }
4578 : :
4579 : : struct copy_subpage_arg {
4580 : : struct page *dst;
4581 : : struct page *src;
4582 : : struct vm_area_struct *vma;
4583 : : };
4584 : :
4585 : : static void copy_subpage(unsigned long addr, int idx, void *arg)
4586 : : {
4587 : : struct copy_subpage_arg *copy_arg = arg;
4588 : :
4589 : : copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx,
4590 : : addr, copy_arg->vma);
4591 : : }
4592 : :
4593 : : void copy_user_huge_page(struct page *dst, struct page *src,
4594 : : unsigned long addr_hint, struct vm_area_struct *vma,
4595 : : unsigned int pages_per_huge_page)
4596 : : {
4597 : : unsigned long addr = addr_hint &
4598 : : ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
4599 : : struct copy_subpage_arg arg = {
4600 : : .dst = dst,
4601 : : .src = src,
4602 : : .vma = vma,
4603 : : };
4604 : :
4605 : : if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4606 : : copy_user_gigantic_page(dst, src, addr, vma,
4607 : : pages_per_huge_page);
4608 : : return;
4609 : : }
4610 : :
4611 : : process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg);
4612 : : }
4613 : :
4614 : : long copy_huge_page_from_user(struct page *dst_page,
4615 : : const void __user *usr_src,
4616 : : unsigned int pages_per_huge_page,
4617 : : bool allow_pagefault)
4618 : : {
4619 : : void *src = (void *)usr_src;
4620 : : void *page_kaddr;
4621 : : unsigned long i, rc = 0;
4622 : : unsigned long ret_val = pages_per_huge_page * PAGE_SIZE;
4623 : :
4624 : : for (i = 0; i < pages_per_huge_page; i++) {
4625 : : if (allow_pagefault)
4626 : : page_kaddr = kmap(dst_page + i);
4627 : : else
4628 : : page_kaddr = kmap_atomic(dst_page + i);
4629 : : rc = copy_from_user(page_kaddr,
4630 : : (const void __user *)(src + i * PAGE_SIZE),
4631 : : PAGE_SIZE);
4632 : : if (allow_pagefault)
4633 : : kunmap(dst_page + i);
4634 : : else
4635 : : kunmap_atomic(page_kaddr);
4636 : :
4637 : : ret_val -= (PAGE_SIZE - rc);
4638 : : if (rc)
4639 : : break;
4640 : :
4641 : : cond_resched();
4642 : : }
4643 : : return ret_val;
4644 : : }
4645 : : #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
4646 : :
4647 : : #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
4648 : :
4649 : : static struct kmem_cache *page_ptl_cachep;
4650 : :
4651 : : void __init ptlock_cache_init(void)
4652 : : {
4653 : : page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
4654 : : SLAB_PANIC, NULL);
4655 : : }
4656 : :
4657 : : bool ptlock_alloc(struct page *page)
4658 : : {
4659 : : spinlock_t *ptl;
4660 : :
4661 : : ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
4662 : : if (!ptl)
4663 : : return false;
4664 : : page->ptl = ptl;
4665 : : return true;
4666 : : }
4667 : :
4668 : : void ptlock_free(struct page *page)
4669 : : {
4670 : : kmem_cache_free(page_ptl_cachep, page->ptl);
4671 : : }
4672 : : #endif
|
